From: Stephen Rothwell Date: Thu, 5 Nov 2015 00:15:20 +0000 (+1100) Subject: Merge remote-tracking branch 'orangefs/for-next' X-Git-Tag: KARO-TXUL-2015-12-04~117 X-Git-Url: https://git.kernelconcepts.de/?p=karo-tx-linux.git;a=commitdiff_plain;h=e6b81abe2d9940cf210adb0799b6aabc1622d822;hp=d6f5480a5549f1073d13e505319cb270a2c06719 Merge remote-tracking branch 'orangefs/for-next' --- diff --git a/Documentation/ABI/stable/sysfs-fs-orangefs b/Documentation/ABI/stable/sysfs-fs-orangefs new file mode 100644 index 000000000000..affdb114bd33 --- /dev/null +++ b/Documentation/ABI/stable/sysfs-fs-orangefs @@ -0,0 +1,87 @@ +What: /sys/fs/orangefs/perf_counters/* +Date: Jun 2015 +Contact: Mike Marshall +Description: + Counters and settings for various caches. + Read only. + + +What: /sys/fs/orangefs/perf_counter_reset +Date: June 2015 +Contact: Mike Marshall +Description: + echo a 0 or a 1 into perf_counter_reset to + reset all the counters in + /sys/fs/orangefs/perf_counters + except ones with PINT_PERF_PRESERVE set. + + +What: /sys/fs/orangefs/perf_time_interval_secs +Date: Jun 2015 +Contact: Mike Marshall +Description: + Length of perf counter intervals in + seconds. + + +What: /sys/fs/orangefs/perf_history_size +Date: Jun 2015 +Contact: Mike Marshall +Description: + The perf_counters cache statistics have N, or + perf_history_size, samples. The default is + one. + + Every perf_time_interval_secs the (first) + samples are reset. + + If N is greater than one, the "current" set + of samples is reset, and the samples from the + other N-1 intervals remain available. + + +What: /sys/fs/orangefs/op_timeout_secs +Date: Jun 2015 +Contact: Mike Marshall +Description: + Service operation timeout in seconds. + + +What: /sys/fs/orangefs/slot_timeout_secs +Date: Jun 2015 +Contact: Mike Marshall +Description: + "Slot" timeout in seconds. A "slot" + is an indexed buffer in the shared + memory segment used for communication + between the kernel module and userspace. + Slots are requested and waited for, + the wait times out after slot_timeout_secs. + + +What: /sys/fs/orangefs/acache/* +Date: Jun 2015 +Contact: Mike Marshall +Description: + Attribute cache configurable settings. + + +What: /sys/fs/orangefs/ncache/* +Date: Jun 2015 +Contact: Mike Marshall +Description: + Name cache configurable settings. + + +What: /sys/fs/orangefs/capcache/* +Date: Jun 2015 +Contact: Mike Marshall +Description: + Capability cache configurable settings. + + +What: /sys/fs/orangefs/ccache/* +Date: Jun 2015 +Contact: Mike Marshall +Description: + Credential cache configurable settings. diff --git a/Documentation/filesystems/orangefs.txt b/Documentation/filesystems/orangefs.txt new file mode 100644 index 000000000000..ec9c8416427e --- /dev/null +++ b/Documentation/filesystems/orangefs.txt @@ -0,0 +1,137 @@ +ORANGEFS +======== + +OrangeFS is an LGPL userspace scale-out parallel storage system. It is ideal +for large storage problems faced by HPC, BigData, Streaming Video, +Genomics, Bioinformatics. + +Orangefs, originally called PVFS, was first developed in 1993 by +Walt Ligon and Eric Blumer as a parallel file system for Parallel +Virtual Machine (PVM) as part of a NASA grant to study the I/O patterns +of parallel programs. + +Orangefs features include: + + * Distributes file data among multiple file servers + * Supports simultaneous access by multiple clients + * Stores file data and metadata on servers using local file system + and access methods + * Userspace implementation is easy to install and maintain + * Direct MPI support + * Stateless + + +MAILING LIST +============ + +http://beowulf-underground.org/mailman/listinfo/pvfs2-users + + +DOCUMENTATION +============= + +http://www.orangefs.org/documentation/ + + +USERSPACE FILESYSTEM SOURCE +=========================== + +http://www.orangefs.org/download + +Orangefs versions prior to 2.9.3 would not be compatible with the +upstream version of the kernel client. + + +BUILDING THE USERSPACE FILESYSTEM ON A SINGLE SERVER +==================================================== + +When Orangefs is upstream, "--with-kernel" shouldn't be needed, but +until then the path to where the kernel with the Orangefs kernel client +patch was built is needed to ensure that pvfs2-client-core (the bridge +between kernel space and user space) will build properly. You can omit +--prefix if you don't care that things are sprinkled around in +/usr/local. + +./configure --prefix=/opt/ofs --with-kernel=/path/to/orangefs/kernel + +make + +make install + +Create an orangefs config file: +/opt/ofs/bin/pvfs2-genconfig /etc/pvfs2.conf + + for "Enter hostnames", use the hostname, don't let it default to + localhost. + +create a pvfs2tab file in /etc: +cat /etc/pvfs2tab +tcp://myhostname:3334/orangefs /mymountpoint pvfs2 defaults,noauto 0 0 + +create the mount point you specified in the tab file if needed: +mkdir /mymountpoint + +bootstrap the server: +/opt/ofs/sbin/pvfs2-server /etc/pvfs2.conf -f + +start the server: +/opt/osf/sbin/pvfs2-server /etc/pvfs2.conf + +Now the server is running. At this point you might like to +prove things are working with: + +/opt/osf/bin/pvfs2-ls /mymountpoint + +You might not want to enforce selinux, it doesn't seem to matter by +linux 3.11... + +If stuff seems to be working, turn on the client core: +/opt/osf/sbin/pvfs2-client -p /opt/osf/sbin/pvfs2-client-core + +Mount your filesystem. +mount -t pvfs2 tcp://myhostname:3334/orangefs /mymountpoint + + +OPTIONS +======= + +The following mount options are accepted: + + acl + Allow the use of Access Control Lists on files and directories. + + intr + Some operations between the kernel client and the user space + filesystem can be interruptible, such as changes in debug levels + and the setting of tunable parameters. + + local_lock + Enable posix locking from the perspective of "this" kernel. The + default file_operations lock action is to return ENOSYS. Posix + locking kicks in if the filesystem is mounted with -o local_lock. + Distributed locking is being worked on for the future. + + +DEBUGGING +========= + +If you want the debug (GOSSIP) statments in a particular +source file (inode.c for example) go to syslog: + + echo inode > /sys/kernel/debug/orangefs/kernel-debug + +No debugging (the default): + + echo none > /sys/kernel/debug/orangefs/kernel-debug + +Debugging from several source files: + + echo inode,dir > /sys/kernel/debug/orangefs/kernel-debug + +All debugging: + + echo all > /sys/kernel/debug/orangefs/kernel-debug + +Get a list of all debugging keywords: + + cat /sys/kernel/debug/orangefs/debug-help diff --git a/fs/Kconfig b/fs/Kconfig index da3f32f1a4e4..7d1bef1999af 100644 --- a/fs/Kconfig +++ b/fs/Kconfig @@ -190,6 +190,7 @@ menuconfig MISC_FILESYSTEMS if MISC_FILESYSTEMS +source "fs/orangefs/Kconfig" source "fs/adfs/Kconfig" source "fs/affs/Kconfig" source "fs/ecryptfs/Kconfig" diff --git a/fs/Makefile b/fs/Makefile index 79f522575cba..f0f2f00951f5 100644 --- a/fs/Makefile +++ b/fs/Makefile @@ -105,6 +105,7 @@ obj-$(CONFIG_AUTOFS4_FS) += autofs4/ obj-$(CONFIG_ADFS_FS) += adfs/ obj-$(CONFIG_FUSE_FS) += fuse/ obj-$(CONFIG_OVERLAY_FS) += overlayfs/ +obj-$(CONFIG_ORANGEFS_FS) += orangefs/ obj-$(CONFIG_UDF_FS) += udf/ obj-$(CONFIG_SUN_OPENPROMFS) += openpromfs/ obj-$(CONFIG_OMFS_FS) += omfs/ diff --git a/fs/orangefs/Kconfig b/fs/orangefs/Kconfig new file mode 100644 index 000000000000..1554c02489de --- /dev/null +++ b/fs/orangefs/Kconfig @@ -0,0 +1,6 @@ +config ORANGEFS_FS + tristate "ORANGEFS (Powered by PVFS) support" + select FS_POSIX_ACL + help + Orange is a parallel file system designed for use on high end + computing (HEC) systems. diff --git a/fs/orangefs/Makefile b/fs/orangefs/Makefile new file mode 100644 index 000000000000..828b36a6916d --- /dev/null +++ b/fs/orangefs/Makefile @@ -0,0 +1,10 @@ +# +# Makefile for the ORANGEFS filesystem. +# + +obj-$(CONFIG_ORANGEFS_FS) += orangefs.o + +orangefs-objs := acl.o file.o pvfs2-cache.o pvfs2-utils.o xattr.o dcache.o \ + inode.o pvfs2-sysfs.o pvfs2-mod.o super.o devpvfs2-req.o \ + namei.o symlink.o dir.o pvfs2-bufmap.o \ + pvfs2-debugfs.o waitqueue.o diff --git a/fs/orangefs/acl.c b/fs/orangefs/acl.c new file mode 100644 index 000000000000..e462b81a3ba1 --- /dev/null +++ b/fs/orangefs/acl.c @@ -0,0 +1,175 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" +#include +#include + +struct posix_acl *pvfs2_get_acl(struct inode *inode, int type) +{ + struct posix_acl *acl; + int ret; + char *key = NULL, *value = NULL; + + switch (type) { + case ACL_TYPE_ACCESS: + key = PVFS2_XATTR_NAME_ACL_ACCESS; + break; + case ACL_TYPE_DEFAULT: + key = PVFS2_XATTR_NAME_ACL_DEFAULT; + break; + default: + gossip_err("pvfs2_get_acl: bogus value of type %d\n", type); + return ERR_PTR(-EINVAL); + } + /* + * Rather than incurring a network call just to determine the exact + * length of the attribute, I just allocate a max length to save on + * the network call. Conceivably, we could pass NULL to + * pvfs2_inode_getxattr() to probe the length of the value, but + * I don't do that for now. + */ + value = kmalloc(PVFS_MAX_XATTR_VALUELEN, GFP_KERNEL); + if (value == NULL) + return ERR_PTR(-ENOMEM); + + gossip_debug(GOSSIP_ACL_DEBUG, + "inode %pU, key %s, type %d\n", + get_khandle_from_ino(inode), + key, + type); + ret = pvfs2_inode_getxattr(inode, + "", + key, + value, + PVFS_MAX_XATTR_VALUELEN); + /* if the key exists, convert it to an in-memory rep */ + if (ret > 0) { + acl = posix_acl_from_xattr(&init_user_ns, value, ret); + } else if (ret == -ENODATA || ret == -ENOSYS) { + acl = NULL; + } else { + gossip_err("inode %pU retrieving acl's failed with error %d\n", + get_khandle_from_ino(inode), + ret); + acl = ERR_PTR(ret); + } + /* kfree(NULL) is safe, so don't worry if value ever got used */ + kfree(value); + return acl; +} + +int pvfs2_set_acl(struct inode *inode, struct posix_acl *acl, int type) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + int error = 0; + void *value = NULL; + size_t size = 0; + const char *name = NULL; + + switch (type) { + case ACL_TYPE_ACCESS: + name = PVFS2_XATTR_NAME_ACL_ACCESS; + if (acl) { + umode_t mode = inode->i_mode; + /* + * can we represent this with the traditional file + * mode permission bits? + */ + error = posix_acl_equiv_mode(acl, &mode); + if (error < 0) { + gossip_err("%s: posix_acl_equiv_mode err: %d\n", + __func__, + error); + return error; + } + + if (inode->i_mode != mode) + SetModeFlag(pvfs2_inode); + inode->i_mode = mode; + mark_inode_dirty_sync(inode); + if (error == 0) + acl = NULL; + } + break; + case ACL_TYPE_DEFAULT: + name = PVFS2_XATTR_NAME_ACL_DEFAULT; + break; + default: + gossip_err("%s: invalid type %d!\n", __func__, type); + return -EINVAL; + } + + gossip_debug(GOSSIP_ACL_DEBUG, + "%s: inode %pU, key %s type %d\n", + __func__, get_khandle_from_ino(inode), + name, + type); + + if (acl) { + size = posix_acl_xattr_size(acl->a_count); + value = kmalloc(size, GFP_KERNEL); + if (!value) + return -ENOMEM; + + error = posix_acl_to_xattr(&init_user_ns, acl, value, size); + if (error < 0) + goto out; + } + + gossip_debug(GOSSIP_ACL_DEBUG, + "%s: name %s, value %p, size %zd, acl %p\n", + __func__, name, value, size, acl); + /* + * Go ahead and set the extended attribute now. NOTE: Suppose acl + * was NULL, then value will be NULL and size will be 0 and that + * will xlate to a removexattr. However, we don't want removexattr + * complain if attributes does not exist. + */ + error = pvfs2_inode_setxattr(inode, "", name, value, size, 0); + +out: + kfree(value); + if (!error) + set_cached_acl(inode, type, acl); + return error; +} + +int pvfs2_init_acl(struct inode *inode, struct inode *dir) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct posix_acl *default_acl, *acl; + umode_t mode = inode->i_mode; + int error = 0; + + ClearModeFlag(pvfs2_inode); + + error = posix_acl_create(dir, &mode, &default_acl, &acl); + if (error) + return error; + + if (default_acl) { + error = pvfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); + posix_acl_release(default_acl); + } + + if (acl) { + if (!error) + error = pvfs2_set_acl(inode, acl, ACL_TYPE_ACCESS); + posix_acl_release(acl); + } + + /* If mode of the inode was changed, then do a forcible ->setattr */ + if (mode != inode->i_mode) { + SetModeFlag(pvfs2_inode); + inode->i_mode = mode; + pvfs2_flush_inode(inode); + } + + return error; +} diff --git a/fs/orangefs/dcache.c b/fs/orangefs/dcache.c new file mode 100644 index 000000000000..9466b179bf24 --- /dev/null +++ b/fs/orangefs/dcache.c @@ -0,0 +1,142 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +/* + * Implementation of dentry (directory cache) functions. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" + +/* Returns 1 if dentry can still be trusted, else 0. */ +static int pvfs2_revalidate_lookup(struct dentry *dentry) +{ + struct dentry *parent_dentry = dget_parent(dentry); + struct inode *parent_inode = parent_dentry->d_inode; + struct pvfs2_inode_s *parent = PVFS2_I(parent_inode); + struct inode *inode = dentry->d_inode; + struct pvfs2_kernel_op_s *new_op; + int ret = 0; + int err = 0; + + gossip_debug(GOSSIP_DCACHE_DEBUG, "%s: attempting lookup.\n", __func__); + + new_op = op_alloc(PVFS2_VFS_OP_LOOKUP); + if (!new_op) + goto out_put_parent; + + new_op->upcall.req.lookup.sym_follow = PVFS2_LOOKUP_LINK_NO_FOLLOW; + new_op->upcall.req.lookup.parent_refn = parent->refn; + strncpy(new_op->upcall.req.lookup.d_name, + dentry->d_name.name, + PVFS2_NAME_LEN); + + gossip_debug(GOSSIP_DCACHE_DEBUG, + "%s:%s:%d interrupt flag [%d]\n", + __FILE__, + __func__, + __LINE__, + get_interruptible_flag(parent_inode)); + + err = service_operation(new_op, "pvfs2_lookup", + get_interruptible_flag(parent_inode)); + if (err) + goto out_drop; + + if (new_op->downcall.status != 0 || + !match_handle(new_op->downcall.resp.lookup.refn.khandle, inode)) { + gossip_debug(GOSSIP_DCACHE_DEBUG, + "%s:%s:%d " + "lookup failure |%s| or no match |%s|.\n", + __FILE__, + __func__, + __LINE__, + new_op->downcall.status ? "true" : "false", + match_handle(new_op->downcall.resp.lookup.refn.khandle, + inode) ? "false" : "true"); + gossip_debug(GOSSIP_DCACHE_DEBUG, + "%s:%s:%d revalidate failed\n", + __FILE__, __func__, __LINE__); + goto out_drop; + } + + ret = 1; +out_release_op: + op_release(new_op); +out_put_parent: + dput(parent_dentry); + return ret; +out_drop: + d_drop(dentry); + goto out_release_op; +} + +/* + * Verify that dentry is valid. + * + * Should return 1 if dentry can still be trusted, else 0 + */ +static int pvfs2_d_revalidate(struct dentry *dentry, unsigned int flags) +{ + struct inode *inode; + int ret = 0; + + if (flags & LOOKUP_RCU) + return -ECHILD; + + gossip_debug(GOSSIP_DCACHE_DEBUG, "%s: called on dentry %p.\n", + __func__, dentry); + + /* find inode from dentry */ + if (!dentry->d_inode) { + gossip_debug(GOSSIP_DCACHE_DEBUG, "%s: negative dentry.\n", + __func__); + goto invalid_exit; + } + + gossip_debug(GOSSIP_DCACHE_DEBUG, "%s: inode valid.\n", __func__); + inode = dentry->d_inode; + + /* + * first perform a lookup to make sure that the object not only + * exists, but is still in the expected place in the name space + */ + if (!is_root_handle(inode)) { + if (!pvfs2_revalidate_lookup(dentry)) + goto invalid_exit; + } else { + gossip_debug(GOSSIP_DCACHE_DEBUG, + "%s: root handle, lookup skipped.\n", + __func__); + } + + /* now perform getattr */ + gossip_debug(GOSSIP_DCACHE_DEBUG, + "%s: doing getattr: inode: %p, handle: %pU\n", + __func__, + inode, + get_khandle_from_ino(inode)); + ret = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); + gossip_debug(GOSSIP_DCACHE_DEBUG, + "%s: getattr %s (ret = %d), returning %s for dentry i_count=%d\n", + __func__, + (ret == 0 ? "succeeded" : "failed"), + ret, + (ret == 0 ? "valid" : "INVALID"), + atomic_read(&inode->i_count)); + if (ret != 0) + goto invalid_exit; + + /* dentry is valid! */ + return 1; + +invalid_exit: + return 0; +} + +const struct dentry_operations pvfs2_dentry_operations = { + .d_revalidate = pvfs2_d_revalidate, +}; diff --git a/fs/orangefs/devpvfs2-req.c b/fs/orangefs/devpvfs2-req.c new file mode 100644 index 000000000000..ede842f05b62 --- /dev/null +++ b/fs/orangefs/devpvfs2-req.c @@ -0,0 +1,1004 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * Changes by Acxiom Corporation to add protocol version to kernel + * communication, Copyright Acxiom Corporation, 2005. + * + * See COPYING in top-level directory. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-dev-proto.h" +#include "pvfs2-bufmap.h" + +#include +#include + +/* this file implements the /dev/pvfs2-req device node */ + +static int open_access_count; + +#define DUMP_DEVICE_ERROR() \ +do { \ + gossip_err("*****************************************************\n");\ + gossip_err("PVFS2 Device Error: You cannot open the device file "); \ + gossip_err("\n/dev/%s more than once. Please make sure that\nthere " \ + "are no ", PVFS2_REQDEVICE_NAME); \ + gossip_err("instances of a program using this device\ncurrently " \ + "running. (You must verify this!)\n"); \ + gossip_err("For example, you can use the lsof program as follows:\n");\ + gossip_err("'lsof | grep %s' (run this as root)\n", \ + PVFS2_REQDEVICE_NAME); \ + gossip_err(" open_access_count = %d\n", open_access_count); \ + gossip_err("*****************************************************\n");\ +} while (0) + +static int hash_func(__u64 tag, int table_size) +{ + return do_div(tag, (unsigned int)table_size); +} + +static void pvfs2_devreq_add_op(struct pvfs2_kernel_op_s *op) +{ + int index = hash_func(op->tag, hash_table_size); + + spin_lock(&htable_ops_in_progress_lock); + list_add_tail(&op->list, &htable_ops_in_progress[index]); + spin_unlock(&htable_ops_in_progress_lock); +} + +static struct pvfs2_kernel_op_s *pvfs2_devreq_remove_op(__u64 tag) +{ + struct pvfs2_kernel_op_s *op, *next; + int index; + + index = hash_func(tag, hash_table_size); + + spin_lock(&htable_ops_in_progress_lock); + list_for_each_entry_safe(op, + next, + &htable_ops_in_progress[index], + list) { + if (op->tag == tag) { + list_del(&op->list); + spin_unlock(&htable_ops_in_progress_lock); + return op; + } + } + + spin_unlock(&htable_ops_in_progress_lock); + return NULL; +} + +static int pvfs2_devreq_open(struct inode *inode, struct file *file) +{ + int ret = -EINVAL; + + if (!(file->f_flags & O_NONBLOCK)) { + gossip_err("pvfs2: device cannot be opened in blocking mode\n"); + goto out; + } + ret = -EACCES; + gossip_debug(GOSSIP_DEV_DEBUG, "pvfs2-client-core: opening device\n"); + mutex_lock(&devreq_mutex); + + if (open_access_count == 0) { + ret = generic_file_open(inode, file); + if (ret == 0) + open_access_count++; + } else { + DUMP_DEVICE_ERROR(); + } + mutex_unlock(&devreq_mutex); + +out: + + gossip_debug(GOSSIP_DEV_DEBUG, + "pvfs2-client-core: open device complete (ret = %d)\n", + ret); + return ret; +} + +static ssize_t pvfs2_devreq_read(struct file *file, + char __user *buf, + size_t count, loff_t *offset) +{ + int ret = 0; + ssize_t len = 0; + struct pvfs2_kernel_op_s *cur_op = NULL; + static __s32 magic = PVFS2_DEVREQ_MAGIC; + __s32 proto_ver = PVFS_KERNEL_PROTO_VERSION; + + if (!(file->f_flags & O_NONBLOCK)) { + /* We do not support blocking reads/opens any more */ + gossip_err("pvfs2: blocking reads are not supported! (pvfs2-client-core bug)\n"); + return -EINVAL; + } else { + struct pvfs2_kernel_op_s *op = NULL, *temp = NULL; + /* get next op (if any) from top of list */ + spin_lock(&pvfs2_request_list_lock); + list_for_each_entry_safe(op, temp, &pvfs2_request_list, list) { + __s32 fsid = fsid_of_op(op); + /* + * Check if this op's fsid is known and needs + * remounting + */ + if (fsid != PVFS_FS_ID_NULL && + fs_mount_pending(fsid) == 1) { + gossip_debug(GOSSIP_DEV_DEBUG, + "Skipping op tag %llu %s\n", + llu(op->tag), + get_opname_string(op)); + continue; + } else { + /* + * op does not belong to any particular fsid + * or already mounted.. let it through + */ + cur_op = op; + spin_lock(&cur_op->lock); + list_del(&cur_op->list); + cur_op->op_linger_tmp--; + /* + * if there is a trailer, re-add it to + * the request list. + */ + if (cur_op->op_linger == 2 && + cur_op->op_linger_tmp == 1) { + if (cur_op->upcall.trailer_size <= 0 || + cur_op->upcall.trailer_buf == NULL) + gossip_err("BUG:trailer_size is %ld and trailer buf is %p\n", (long)cur_op->upcall.trailer_size, cur_op->upcall.trailer_buf); + /* re-add it to the head of the list */ + list_add(&cur_op->list, + &pvfs2_request_list); + } + spin_unlock(&cur_op->lock); + break; + } + } + spin_unlock(&pvfs2_request_list_lock); + } + + if (cur_op) { + spin_lock(&cur_op->lock); + + gossip_debug(GOSSIP_DEV_DEBUG, + "client-core: reading op tag %llu %s\n", + llu(cur_op->tag), get_opname_string(cur_op)); + if (op_state_in_progress(cur_op) || op_state_serviced(cur_op)) { + if (cur_op->op_linger == 1) + gossip_err("WARNING: Current op already queued...skipping\n"); + } else if (cur_op->op_linger == 1 || + (cur_op->op_linger == 2 && + cur_op->op_linger_tmp == 0)) { + /* + * atomically move the operation to the + * htable_ops_in_progress + */ + set_op_state_inprogress(cur_op); + pvfs2_devreq_add_op(cur_op); + } + + spin_unlock(&cur_op->lock); + + /* 2 cases + * a) OPs with no trailers + * b) OPs with trailers, Stage 1 + * Either way push the upcall out + */ + if (cur_op->op_linger == 1 || + (cur_op->op_linger == 2 && cur_op->op_linger_tmp == 1)) { + len = MAX_ALIGNED_DEV_REQ_UPSIZE; + if ((size_t) len <= count) { + ret = copy_to_user(buf, + &proto_ver, + sizeof(__s32)); + if (ret == 0) { + ret = copy_to_user(buf + sizeof(__s32), + &magic, + sizeof(__s32)); + if (ret == 0) { + ret = copy_to_user(buf+2 * sizeof(__s32), + &cur_op->tag, + sizeof(__u64)); + if (ret == 0) { + ret = copy_to_user( + buf + + 2 * + sizeof(__s32) + + sizeof(__u64), + &cur_op->upcall, + sizeof(struct pvfs2_upcall_s)); + } + } + } + + if (ret) { + gossip_err("Failed to copy data to user space\n"); + len = -EFAULT; + } + } else { + gossip_err + ("Failed to copy data to user space\n"); + len = -EIO; + } + } + /* Stage 2: Push the trailer out */ + else if (cur_op->op_linger == 2 && cur_op->op_linger_tmp == 0) { + len = cur_op->upcall.trailer_size; + if ((size_t) len <= count) { + ret = copy_to_user(buf, + cur_op->upcall.trailer_buf, + len); + if (ret) { + gossip_err("Failed to copy trailer to user space\n"); + len = -EFAULT; + } + } else { + gossip_err("Read buffer for trailer is too small (%ld as opposed to %ld)\n", + (long)count, + (long)len); + len = -EIO; + } + } else { + gossip_err("cur_op: %p (op_linger %d), (op_linger_tmp %d), erroneous request list?\n", + cur_op, + cur_op->op_linger, + cur_op->op_linger_tmp); + len = 0; + } + } else if (file->f_flags & O_NONBLOCK) { + /* + * if in non-blocking mode, return EAGAIN since no requests are + * ready yet + */ + len = -EAGAIN; + } + return len; +} + +/* Function for writev() callers into the device */ +static ssize_t pvfs2_devreq_writev(struct file *file, + const struct iovec *iov, + size_t count, + loff_t *offset) +{ + struct pvfs2_kernel_op_s *op = NULL; + void *buffer = NULL; + void *ptr = NULL; + unsigned long i = 0; + static int max_downsize = MAX_ALIGNED_DEV_REQ_DOWNSIZE; + int ret = 0, num_remaining = max_downsize; + int notrailer_count = 4; /* num elements in iovec without trailer */ + int payload_size = 0; + __s32 magic = 0; + __s32 proto_ver = 0; + __u64 tag = 0; + ssize_t total_returned_size = 0; + + /* Either there is a trailer or there isn't */ + if (count != notrailer_count && count != (notrailer_count + 1)) { + gossip_err("Error: Number of iov vectors is (%zu) and notrailer count is %d\n", + count, + notrailer_count); + return -EPROTO; + } + buffer = dev_req_alloc(); + if (!buffer) + return -ENOMEM; + ptr = buffer; + + for (i = 0; i < notrailer_count; i++) { + if (iov[i].iov_len > num_remaining) { + gossip_err + ("writev error: Freeing buffer and returning\n"); + dev_req_release(buffer); + return -EMSGSIZE; + } + ret = copy_from_user(ptr, iov[i].iov_base, iov[i].iov_len); + if (ret) { + gossip_err("Failed to copy data from user space\n"); + dev_req_release(buffer); + return -EIO; + } + num_remaining -= iov[i].iov_len; + ptr += iov[i].iov_len; + payload_size += iov[i].iov_len; + } + total_returned_size = payload_size; + + /* these elements are currently 8 byte aligned (8 bytes for (version + + * magic) 8 bytes for tag). If you add another element, either + * make it 8 bytes big, or use get_unaligned when asigning. + */ + ptr = buffer; + proto_ver = *((__s32 *) ptr); + ptr += sizeof(__s32); + + magic = *((__s32 *) ptr); + ptr += sizeof(__s32); + + tag = *((__u64 *) ptr); + ptr += sizeof(__u64); + + if (magic != PVFS2_DEVREQ_MAGIC) { + gossip_err("Error: Device magic number does not match.\n"); + dev_req_release(buffer); + return -EPROTO; + } + + /* + * proto_ver = 20902 for 2.9.2 + */ + + op = pvfs2_devreq_remove_op(tag); + if (op) { + /* Increase ref count! */ + get_op(op); + /* cut off magic and tag from payload size */ + payload_size -= (2 * sizeof(__s32) + sizeof(__u64)); + if (payload_size <= sizeof(struct pvfs2_downcall_s)) + /* copy the passed in downcall into the op */ + memcpy(&op->downcall, + ptr, + sizeof(struct pvfs2_downcall_s)); + else + gossip_debug(GOSSIP_DEV_DEBUG, + "writev: Ignoring %d bytes\n", + payload_size); + + /* Do not allocate needlessly if client-core forgets + * to reset trailer size on op errors. + */ + if (op->downcall.status == 0 && op->downcall.trailer_size > 0) { + gossip_debug(GOSSIP_DEV_DEBUG, + "writev: trailer size %ld\n", + (unsigned long)op->downcall.trailer_size); + if (count != (notrailer_count + 1)) { + gossip_err("Error: trailer size (%ld) is non-zero, no trailer elements though? (%zu)\n", (unsigned long)op->downcall.trailer_size, count); + dev_req_release(buffer); + put_op(op); + return -EPROTO; + } + if (iov[notrailer_count].iov_len > + op->downcall.trailer_size) { + gossip_err("writev error: trailer size (%ld) != iov_len (%ld)\n", (unsigned long)op->downcall.trailer_size, (unsigned long)iov[notrailer_count].iov_len); + dev_req_release(buffer); + put_op(op); + return -EMSGSIZE; + } + /* Allocate a buffer large enough to hold the + * trailer bytes. + */ + op->downcall.trailer_buf = + vmalloc(op->downcall.trailer_size); + if (op->downcall.trailer_buf != NULL) { + gossip_debug(GOSSIP_DEV_DEBUG, "vmalloc: %p\n", + op->downcall.trailer_buf); + ret = copy_from_user(op->downcall.trailer_buf, + iov[notrailer_count]. + iov_base, + iov[notrailer_count]. + iov_len); + if (ret) { + gossip_err("Failed to copy trailer data from user space\n"); + dev_req_release(buffer); + gossip_debug(GOSSIP_DEV_DEBUG, + "vfree: %p\n", + op->downcall.trailer_buf); + vfree(op->downcall.trailer_buf); + op->downcall.trailer_buf = NULL; + put_op(op); + return -EIO; + } + } else { + /* Change downcall status */ + op->downcall.status = -ENOMEM; + gossip_err("writev: could not vmalloc for trailer!\n"); + } + } + + /* if this operation is an I/O operation and if it was + * initiated on behalf of a *synchronous* VFS I/O operation, + * only then we need to wait + * for all data to be copied before we can return to avoid + * buffer corruption and races that can pull the buffers + * out from under us. + * + * Essentially we're synchronizing with other parts of the + * vfs implicitly by not allowing the user space + * application reading/writing this device to return until + * the buffers are done being used. + */ + if ((op->upcall.type == PVFS2_VFS_OP_FILE_IO && + op->upcall.req.io.async_vfs_io == PVFS_VFS_SYNC_IO) || + op->upcall.type == PVFS2_VFS_OP_FILE_IOX) { + int timed_out = 0; + DECLARE_WAITQUEUE(wait_entry, current); + + /* tell the vfs op waiting on a waitqueue + * that this op is done + */ + spin_lock(&op->lock); + set_op_state_serviced(op); + spin_unlock(&op->lock); + + add_wait_queue_exclusive(&op->io_completion_waitq, + &wait_entry); + wake_up_interruptible(&op->waitq); + + while (1) { + set_current_state(TASK_INTERRUPTIBLE); + + spin_lock(&op->lock); + if (op->io_completed) { + spin_unlock(&op->lock); + break; + } + spin_unlock(&op->lock); + + if (!signal_pending(current)) { + int timeout = + MSECS_TO_JIFFIES(1000 * + op_timeout_secs); + if (!schedule_timeout(timeout)) { + gossip_debug(GOSSIP_DEV_DEBUG, "*** I/O wait time is up\n"); + timed_out = 1; + break; + } + continue; + } + + gossip_debug(GOSSIP_DEV_DEBUG, "*** signal on I/O wait -- aborting\n"); + break; + } + + set_current_state(TASK_RUNNING); + remove_wait_queue(&op->io_completion_waitq, + &wait_entry); + + /* NOTE: for I/O operations we handle releasing the op + * object except in the case of timeout. the reason we + * can't free the op in timeout cases is that the op + * service logic in the vfs retries operations using + * the same op ptr, thus it can't be freed. + */ + if (!timed_out) + op_release(op); + } else { + + /* + * tell the vfs op waiting on a waitqueue that + * this op is done + */ + spin_lock(&op->lock); + set_op_state_serviced(op); + spin_unlock(&op->lock); + /* + * for every other operation (i.e. non-I/O), we need to + * wake up the callers for downcall completion + * notification + */ + wake_up_interruptible(&op->waitq); + } + } else { + /* ignore downcalls that we're not interested in */ + gossip_debug(GOSSIP_DEV_DEBUG, + "WARNING: No one's waiting for tag %llu\n", + llu(tag)); + } + dev_req_release(buffer); + + return total_returned_size; +} + +static ssize_t pvfs2_devreq_write_iter(struct kiocb *iocb, + struct iov_iter *iter) +{ + return pvfs2_devreq_writev(iocb->ki_filp, + iter->iov, + iter->nr_segs, + &iocb->ki_pos); +} + +/* Returns whether any FS are still pending remounted */ +static int mark_all_pending_mounts(void) +{ + int unmounted = 1; + struct pvfs2_sb_info_s *pvfs2_sb = NULL; + + spin_lock(&pvfs2_superblocks_lock); + list_for_each_entry(pvfs2_sb, &pvfs2_superblocks, list) { + /* All of these file system require a remount */ + pvfs2_sb->mount_pending = 1; + unmounted = 0; + } + spin_unlock(&pvfs2_superblocks_lock); + return unmounted; +} + +/* + * Determine if a given file system needs to be remounted or not + * Returns -1 on error + * 0 if already mounted + * 1 if needs remount + */ +int fs_mount_pending(__s32 fsid) +{ + int mount_pending = -1; + struct pvfs2_sb_info_s *pvfs2_sb = NULL; + + spin_lock(&pvfs2_superblocks_lock); + list_for_each_entry(pvfs2_sb, &pvfs2_superblocks, list) { + if (pvfs2_sb->fs_id == fsid) { + mount_pending = pvfs2_sb->mount_pending; + break; + } + } + spin_unlock(&pvfs2_superblocks_lock); + return mount_pending; +} + +/* + * NOTE: gets called when the last reference to this device is dropped. + * Using the open_access_count variable, we enforce a reference count + * on this file so that it can be opened by only one process at a time. + * the devreq_mutex is used to make sure all i/o has completed + * before we call pvfs_bufmap_finalize, and similar such tricky + * situations + */ +static int pvfs2_devreq_release(struct inode *inode, struct file *file) +{ + int unmounted = 0; + + gossip_debug(GOSSIP_DEV_DEBUG, + "%s:pvfs2-client-core: exiting, closing device\n", + __func__); + + mutex_lock(&devreq_mutex); + pvfs_bufmap_finalize(); + + open_access_count--; + + unmounted = mark_all_pending_mounts(); + gossip_debug(GOSSIP_DEV_DEBUG, "PVFS2 Device Close: Filesystem(s) %s\n", + (unmounted ? "UNMOUNTED" : "MOUNTED")); + mutex_unlock(&devreq_mutex); + + /* + * Walk through the list of ops in the request list, mark them + * as purged and wake them up. + */ + purge_waiting_ops(); + /* + * Walk through the hash table of in progress operations; mark + * them as purged and wake them up + */ + purge_inprogress_ops(); + gossip_debug(GOSSIP_DEV_DEBUG, + "pvfs2-client-core: device close complete\n"); + return 0; +} + +int is_daemon_in_service(void) +{ + int in_service; + + /* + * What this function does is checks if client-core is alive + * based on the access count we maintain on the device. + */ + mutex_lock(&devreq_mutex); + in_service = open_access_count == 1 ? 0 : -EIO; + mutex_unlock(&devreq_mutex); + return in_service; +} + +static inline long check_ioctl_command(unsigned int command) +{ + /* Check for valid ioctl codes */ + if (_IOC_TYPE(command) != PVFS_DEV_MAGIC) { + gossip_err("device ioctl magic numbers don't match! Did you rebuild pvfs2-client-core/libpvfs2? [cmd %x, magic %x != %x]\n", + command, + _IOC_TYPE(command), + PVFS_DEV_MAGIC); + return -EINVAL; + } + /* and valid ioctl commands */ + if (_IOC_NR(command) >= PVFS_DEV_MAXNR || _IOC_NR(command) <= 0) { + gossip_err("Invalid ioctl command number [%d >= %d]\n", + _IOC_NR(command), PVFS_DEV_MAXNR); + return -ENOIOCTLCMD; + } + return 0; +} + +static long dispatch_ioctl_command(unsigned int command, unsigned long arg) +{ + static __s32 magic = PVFS2_DEVREQ_MAGIC; + static __s32 max_up_size = MAX_ALIGNED_DEV_REQ_UPSIZE; + static __s32 max_down_size = MAX_ALIGNED_DEV_REQ_DOWNSIZE; + struct PVFS_dev_map_desc user_desc; + int ret = 0; + struct dev_mask_info_s mask_info = { 0 }; + struct dev_mask2_info_s mask2_info = { 0, 0 }; + int upstream_kmod = 1; + struct list_head *tmp = NULL; + struct pvfs2_sb_info_s *pvfs2_sb = NULL; + + /* mtmoore: add locking here */ + + switch (command) { + case PVFS_DEV_GET_MAGIC: + return ((put_user(magic, (__s32 __user *) arg) == -EFAULT) ? + -EIO : + 0); + case PVFS_DEV_GET_MAX_UPSIZE: + return ((put_user(max_up_size, + (__s32 __user *) arg) == -EFAULT) ? + -EIO : + 0); + case PVFS_DEV_GET_MAX_DOWNSIZE: + return ((put_user(max_down_size, + (__s32 __user *) arg) == -EFAULT) ? + -EIO : + 0); + case PVFS_DEV_MAP: + ret = copy_from_user(&user_desc, + (struct PVFS_dev_map_desc __user *) + arg, + sizeof(struct PVFS_dev_map_desc)); + return ret ? -EIO : pvfs_bufmap_initialize(&user_desc); + case PVFS_DEV_REMOUNT_ALL: + gossip_debug(GOSSIP_DEV_DEBUG, + "pvfs2_devreq_ioctl: got PVFS_DEV_REMOUNT_ALL\n"); + + /* + * remount all mounted pvfs2 volumes to regain the lost + * dynamic mount tables (if any) -- NOTE: this is done + * without keeping the superblock list locked due to the + * upcall/downcall waiting. also, the request semaphore is + * used to ensure that no operations will be serviced until + * all of the remounts are serviced (to avoid ops between + * mounts to fail) + */ + ret = mutex_lock_interruptible(&request_mutex); + if (ret < 0) + return ret; + gossip_debug(GOSSIP_DEV_DEBUG, + "pvfs2_devreq_ioctl: priority remount in progress\n"); + list_for_each(tmp, &pvfs2_superblocks) { + pvfs2_sb = + list_entry(tmp, struct pvfs2_sb_info_s, list); + if (pvfs2_sb && (pvfs2_sb->sb)) { + gossip_debug(GOSSIP_DEV_DEBUG, + "Remounting SB %p\n", + pvfs2_sb); + + ret = pvfs2_remount(pvfs2_sb->sb); + if (ret) { + gossip_debug(GOSSIP_DEV_DEBUG, + "SB %p remount failed\n", + pvfs2_sb); + break; + } + } + } + gossip_debug(GOSSIP_DEV_DEBUG, + "pvfs2_devreq_ioctl: priority remount complete\n"); + mutex_unlock(&request_mutex); + return ret; + + case PVFS_DEV_UPSTREAM: + ret = copy_to_user((void __user *)arg, + &upstream_kmod, + sizeof(upstream_kmod)); + + if (ret != 0) + return -EIO; + else + return ret; + + case PVFS_DEV_CLIENT_MASK: + ret = copy_from_user(&mask2_info, + (void __user *)arg, + sizeof(struct dev_mask2_info_s)); + + if (ret != 0) + return -EIO; + + client_debug_mask.mask1 = mask2_info.mask1_value; + client_debug_mask.mask2 = mask2_info.mask2_value; + + pr_info("%s: client debug mask has been been received " + ":%llx: :%llx:\n", + __func__, + (unsigned long long)client_debug_mask.mask1, + (unsigned long long)client_debug_mask.mask2); + + return ret; + + case PVFS_DEV_CLIENT_STRING: + ret = copy_from_user(&client_debug_array_string, + (void __user *)arg, + PVFS2_MAX_DEBUG_STRING_LEN); + if (ret != 0) { + pr_info("%s: " + "PVFS_DEV_CLIENT_STRING: copy_from_user failed" + "\n", + __func__); + return -EIO; + } + + pr_info("%s: client debug array string has been been received." + "\n", + __func__); + + if (!help_string_initialized) { + + /* Free the "we don't know yet" default string... */ + kfree(debug_help_string); + + /* build a proper debug help string */ + if (orangefs_prepare_debugfs_help_string(0)) { + gossip_err("%s: " + "prepare_debugfs_help_string failed" + "\n", + __func__); + return -EIO; + } + + /* Replace the boilerplate boot-time debug-help file. */ + debugfs_remove(help_file_dentry); + + help_file_dentry = + debugfs_create_file( + ORANGEFS_KMOD_DEBUG_HELP_FILE, + 0444, + debug_dir, + debug_help_string, + &debug_help_fops); + + if (!help_file_dentry) { + gossip_err("%s: debugfs_create_file failed for" + " :%s:!\n", + __func__, + ORANGEFS_KMOD_DEBUG_HELP_FILE); + return -EIO; + } + } + + debug_mask_to_string(&client_debug_mask, 1); + + debugfs_remove(client_debug_dentry); + + pvfs2_client_debug_init(); + + help_string_initialized++; + + return ret; + + case PVFS_DEV_DEBUG: + ret = copy_from_user(&mask_info, + (void __user *)arg, + sizeof(mask_info)); + + if (ret != 0) + return -EIO; + + if (mask_info.mask_type == KERNEL_MASK) { + if ((mask_info.mask_value == 0) + && (kernel_mask_set_mod_init)) { + /* + * the kernel debug mask was set when the + * kernel module was loaded; don't override + * it if the client-core was started without + * a value for PVFS2_KMODMASK. + */ + return 0; + } + debug_mask_to_string(&mask_info.mask_value, + mask_info.mask_type); + gossip_debug_mask = mask_info.mask_value; + pr_info("PVFS: kernel debug mask has been modified to " + ":%s: :%llx:\n", + kernel_debug_string, + (unsigned long long)gossip_debug_mask); + } else if (mask_info.mask_type == CLIENT_MASK) { + debug_mask_to_string(&mask_info.mask_value, + mask_info.mask_type); + pr_info("PVFS: client debug mask has been modified to" + ":%s: :%llx:\n", + client_debug_string, + llu(mask_info.mask_value)); + } else { + gossip_lerr("Invalid mask type....\n"); + return -EINVAL; + } + + return ret; + + default: + return -ENOIOCTLCMD; + } + return -ENOIOCTLCMD; +} + +static long pvfs2_devreq_ioctl(struct file *file, + unsigned int command, unsigned long arg) +{ + long ret; + + /* Check for properly constructed commands */ + ret = check_ioctl_command(command); + if (ret < 0) + return (int)ret; + + return (int)dispatch_ioctl_command(command, arg); +} + +#ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ + +/* Compat structure for the PVFS_DEV_MAP ioctl */ +struct PVFS_dev_map_desc32 { + compat_uptr_t ptr; + __s32 total_size; + __s32 size; + __s32 count; +}; + +static unsigned long translate_dev_map26(unsigned long args, long *error) +{ + struct PVFS_dev_map_desc32 __user *p32 = (void __user *)args; + /* + * Depending on the architecture, allocate some space on the + * user-call-stack based on our expected layout. + */ + struct PVFS_dev_map_desc __user *p = + compat_alloc_user_space(sizeof(*p)); + compat_uptr_t addr; + + *error = 0; + /* get the ptr from the 32 bit user-space */ + if (get_user(addr, &p32->ptr)) + goto err; + /* try to put that into a 64-bit layout */ + if (put_user(compat_ptr(addr), &p->ptr)) + goto err; + /* copy the remaining fields */ + if (copy_in_user(&p->total_size, &p32->total_size, sizeof(__s32))) + goto err; + if (copy_in_user(&p->size, &p32->size, sizeof(__s32))) + goto err; + if (copy_in_user(&p->count, &p32->count, sizeof(__s32))) + goto err; + return (unsigned long)p; +err: + *error = -EFAULT; + return 0; +} + +/* + * 32 bit user-space apps' ioctl handlers when kernel modules + * is compiled as a 64 bit one + */ +static long pvfs2_devreq_compat_ioctl(struct file *filp, unsigned int cmd, + unsigned long args) +{ + long ret; + unsigned long arg = args; + + /* Check for properly constructed commands */ + ret = check_ioctl_command(cmd); + if (ret < 0) + return ret; + if (cmd == PVFS_DEV_MAP) { + /* + * convert the arguments to what we expect internally + * in kernel space + */ + arg = translate_dev_map26(args, &ret); + if (ret < 0) { + gossip_err("Could not translate dev map\n"); + return ret; + } + } + /* no other ioctl requires translation */ + return dispatch_ioctl_command(cmd, arg); +} + +#endif /* CONFIG_COMPAT is in .config */ + +/* + * The following two ioctl32 functions had been refactored into the above + * CONFIG_COMPAT ifdef, but that was an over simplification that was + * not noticed until we tried to compile on power pc... + */ +#if (defined(CONFIG_COMPAT) && !defined(HAVE_REGISTER_IOCTL32_CONVERSION)) || !defined(CONFIG_COMPAT) +static int pvfs2_ioctl32_init(void) +{ + return 0; +} + +static void pvfs2_ioctl32_cleanup(void) +{ + return; +} +#endif + +/* the assigned character device major number */ +static int pvfs2_dev_major; + +/* + * Initialize pvfs2 device specific state: + * Must be called at module load time only + */ +int pvfs2_dev_init(void) +{ + int ret; + + /* register the ioctl32 sub-system */ + ret = pvfs2_ioctl32_init(); + if (ret < 0) + return ret; + + /* register pvfs2-req device */ + pvfs2_dev_major = register_chrdev(0, + PVFS2_REQDEVICE_NAME, + &pvfs2_devreq_file_operations); + if (pvfs2_dev_major < 0) { + gossip_debug(GOSSIP_DEV_DEBUG, + "Failed to register /dev/%s (error %d)\n", + PVFS2_REQDEVICE_NAME, pvfs2_dev_major); + pvfs2_ioctl32_cleanup(); + return pvfs2_dev_major; + } + + gossip_debug(GOSSIP_DEV_DEBUG, + "*** /dev/%s character device registered ***\n", + PVFS2_REQDEVICE_NAME); + gossip_debug(GOSSIP_DEV_DEBUG, "'mknod /dev/%s c %d 0'.\n", + PVFS2_REQDEVICE_NAME, pvfs2_dev_major); + return 0; +} + +void pvfs2_dev_cleanup(void) +{ + unregister_chrdev(pvfs2_dev_major, PVFS2_REQDEVICE_NAME); + gossip_debug(GOSSIP_DEV_DEBUG, + "*** /dev/%s character device unregistered ***\n", + PVFS2_REQDEVICE_NAME); + /* unregister the ioctl32 sub-system */ + pvfs2_ioctl32_cleanup(); +} + +static unsigned int pvfs2_devreq_poll(struct file *file, + struct poll_table_struct *poll_table) +{ + int poll_revent_mask = 0; + + if (open_access_count == 1) { + poll_wait(file, &pvfs2_request_list_waitq, poll_table); + + spin_lock(&pvfs2_request_list_lock); + if (!list_empty(&pvfs2_request_list)) + poll_revent_mask |= POLL_IN; + spin_unlock(&pvfs2_request_list_lock); + } + return poll_revent_mask; +} + +const struct file_operations pvfs2_devreq_file_operations = { + .owner = THIS_MODULE, + .read = pvfs2_devreq_read, + .write_iter = pvfs2_devreq_write_iter, + .open = pvfs2_devreq_open, + .release = pvfs2_devreq_release, + .unlocked_ioctl = pvfs2_devreq_ioctl, + +#ifdef CONFIG_COMPAT /* CONFIG_COMPAT is in .config */ + .compat_ioctl = pvfs2_devreq_compat_ioctl, +#endif + .poll = pvfs2_devreq_poll +}; diff --git a/fs/orangefs/dir.c b/fs/orangefs/dir.c new file mode 100644 index 000000000000..daf497384501 --- /dev/null +++ b/fs/orangefs/dir.c @@ -0,0 +1,345 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" + +struct readdir_handle_s { + int buffer_index; + struct pvfs2_readdir_response_s readdir_response; + void *dents_buf; +}; + +/* + * decode routine needed by kmod to make sense of the shared page for readdirs. + */ +static long decode_dirents(char *ptr, struct pvfs2_readdir_response_s *readdir) +{ + int i; + struct pvfs2_readdir_response_s *rd = + (struct pvfs2_readdir_response_s *) ptr; + char *buf = ptr; + char **pptr = &buf; + + readdir->token = rd->token; + readdir->pvfs_dirent_outcount = rd->pvfs_dirent_outcount; + readdir->dirent_array = kmalloc(readdir->pvfs_dirent_outcount * + sizeof(*readdir->dirent_array), + GFP_KERNEL); + if (readdir->dirent_array == NULL) + return -ENOMEM; + *pptr += offsetof(struct pvfs2_readdir_response_s, dirent_array); + for (i = 0; i < readdir->pvfs_dirent_outcount; i++) { + dec_string(pptr, &readdir->dirent_array[i].d_name, + &readdir->dirent_array[i].d_length); + readdir->dirent_array[i].khandle = + *(struct pvfs2_khandle *) *pptr; + *pptr += 16; + } + return (unsigned long)*pptr - (unsigned long)ptr; +} + +static long readdir_handle_ctor(struct readdir_handle_s *rhandle, void *buf, + int buffer_index) +{ + long ret; + + if (buf == NULL) { + gossip_err + ("Invalid NULL buffer specified in readdir_handle_ctor\n"); + return -ENOMEM; + } + if (buffer_index < 0) { + gossip_err + ("Invalid buffer index specified in readdir_handle_ctor\n"); + return -EINVAL; + } + rhandle->buffer_index = buffer_index; + rhandle->dents_buf = buf; + ret = decode_dirents(buf, &rhandle->readdir_response); + if (ret < 0) { + gossip_err("Could not decode readdir from buffer %ld\n", ret); + rhandle->buffer_index = -1; + gossip_debug(GOSSIP_DIR_DEBUG, "vfree %p\n", buf); + vfree(buf); + rhandle->dents_buf = NULL; + } + return ret; +} + +static void readdir_handle_dtor(struct pvfs2_bufmap *bufmap, + struct readdir_handle_s *rhandle) +{ + if (rhandle == NULL) + return; + + /* kfree(NULL) is safe */ + kfree(rhandle->readdir_response.dirent_array); + rhandle->readdir_response.dirent_array = NULL; + + if (rhandle->buffer_index >= 0) { + readdir_index_put(bufmap, rhandle->buffer_index); + rhandle->buffer_index = -1; + } + if (rhandle->dents_buf) { + gossip_debug(GOSSIP_DIR_DEBUG, "vfree %p\n", + rhandle->dents_buf); + vfree(rhandle->dents_buf); + rhandle->dents_buf = NULL; + } +} + +/* + * Read directory entries from an instance of an open directory. + */ +static int pvfs2_readdir(struct file *file, struct dir_context *ctx) +{ + struct pvfs2_bufmap *bufmap = NULL; + int ret = 0; + int buffer_index; + /* + * ptoken supports Orangefs' distributed directory logic, added + * in 2.9.2. + */ + __u64 *ptoken = file->private_data; + __u64 pos = 0; + ino_t ino = 0; + struct dentry *dentry = file->f_path.dentry; + struct pvfs2_kernel_op_s *new_op = NULL; + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(dentry->d_inode); + int buffer_full = 0; + struct readdir_handle_s rhandle; + int i = 0; + int len = 0; + ino_t current_ino = 0; + char *current_entry = NULL; + long bytes_decoded; + + gossip_debug(GOSSIP_DIR_DEBUG, + "%s: ctx->pos:%lld, ptoken = %llu\n", + __func__, + lld(ctx->pos), + llu(*ptoken)); + + pos = (__u64) ctx->pos; + + /* are we done? */ + if (pos == PVFS_READDIR_END) { + gossip_debug(GOSSIP_DIR_DEBUG, + "Skipping to termination path\n"); + return 0; + } + + gossip_debug(GOSSIP_DIR_DEBUG, + "pvfs2_readdir called on %s (pos=%llu)\n", + dentry->d_name.name, llu(pos)); + + rhandle.buffer_index = -1; + rhandle.dents_buf = NULL; + memset(&rhandle.readdir_response, 0, sizeof(rhandle.readdir_response)); + + new_op = op_alloc(PVFS2_VFS_OP_READDIR); + if (!new_op) + return -ENOMEM; + + new_op->uses_shared_memory = 1; + new_op->upcall.req.readdir.refn = pvfs2_inode->refn; + new_op->upcall.req.readdir.max_dirent_count = MAX_DIRENT_COUNT_READDIR; + + gossip_debug(GOSSIP_DIR_DEBUG, + "%s: upcall.req.readdir.refn.khandle: %pU\n", + __func__, + &new_op->upcall.req.readdir.refn.khandle); + + new_op->upcall.req.readdir.token = *ptoken; + +get_new_buffer_index: + ret = readdir_index_get(&bufmap, &buffer_index); + if (ret < 0) { + gossip_lerr("pvfs2_readdir: readdir_index_get() failure (%d)\n", + ret); + goto out_free_op; + } + new_op->upcall.req.readdir.buf_index = buffer_index; + + ret = service_operation(new_op, + "pvfs2_readdir", + get_interruptible_flag(dentry->d_inode)); + + gossip_debug(GOSSIP_DIR_DEBUG, + "Readdir downcall status is %d. ret:%d\n", + new_op->downcall.status, + ret); + + if (ret == -EAGAIN && op_state_purged(new_op)) { + /* + * readdir shared memory aread has been wiped due to + * pvfs2-client-core restarting, so we must get a new + * index into the shared memory. + */ + gossip_debug(GOSSIP_DIR_DEBUG, + "%s: Getting new buffer_index for retry of readdir..\n", + __func__); + readdir_index_put(bufmap, buffer_index); + goto get_new_buffer_index; + } + + if (ret == -EIO && op_state_purged(new_op)) { + gossip_err("%s: Client is down. Aborting readdir call.\n", + __func__); + readdir_index_put(bufmap, buffer_index); + goto out_free_op; + } + + if (ret < 0 || new_op->downcall.status != 0) { + gossip_debug(GOSSIP_DIR_DEBUG, + "Readdir request failed. Status:%d\n", + new_op->downcall.status); + readdir_index_put(bufmap, buffer_index); + if (ret >= 0) + ret = new_op->downcall.status; + goto out_free_op; + } + + bytes_decoded = + readdir_handle_ctor(&rhandle, + new_op->downcall.trailer_buf, + buffer_index); + if (bytes_decoded < 0) { + gossip_err("pvfs2_readdir: Could not decode trailer buffer into a readdir response %d\n", + ret); + ret = bytes_decoded; + readdir_index_put(bufmap, buffer_index); + goto out_free_op; + } + + if (bytes_decoded != new_op->downcall.trailer_size) { + gossip_err("pvfs2_readdir: # bytes decoded (%ld) " + "!= trailer size (%ld)\n", + bytes_decoded, + (long)new_op->downcall.trailer_size); + ret = -EINVAL; + goto out_destroy_handle; + } + + /* + * pvfs2 doesn't actually store dot and dot-dot, but + * we need to have them represented. + */ + if (pos == 0) { + ino = get_ino_from_khandle(dentry->d_inode); + gossip_debug(GOSSIP_DIR_DEBUG, + "%s: calling dir_emit of \".\" with pos = %llu\n", + __func__, + llu(pos)); + ret = dir_emit(ctx, ".", 1, ino, DT_DIR); + pos += 1; + } + + if (pos == 1) { + ino = get_parent_ino_from_dentry(dentry); + gossip_debug(GOSSIP_DIR_DEBUG, + "%s: calling dir_emit of \"..\" with pos = %llu\n", + __func__, + llu(pos)); + ret = dir_emit(ctx, "..", 2, ino, DT_DIR); + pos += 1; + } + + /* + * we stored PVFS_ITERATE_NEXT in ctx->pos last time around + * to prevent "finding" dot and dot-dot on any iteration + * other than the first. + */ + if (ctx->pos == PVFS_ITERATE_NEXT) + ctx->pos = 0; + + for (i = ctx->pos; + i < rhandle.readdir_response.pvfs_dirent_outcount; + i++) { + len = rhandle.readdir_response.dirent_array[i].d_length; + current_entry = rhandle.readdir_response.dirent_array[i].d_name; + current_ino = pvfs2_khandle_to_ino( + &(rhandle.readdir_response.dirent_array[i].khandle)); + + gossip_debug(GOSSIP_DIR_DEBUG, + "calling dir_emit for %s with len %d" + ", ctx->pos %ld\n", + current_entry, + len, + (unsigned long)ctx->pos); + /* + * type is unknown. We don't return object type + * in the dirent_array. This leaves getdents + * clueless about type. + */ + ret = + dir_emit(ctx, current_entry, len, current_ino, DT_UNKNOWN); + if (!ret) + break; + ctx->pos++; + gossip_debug(GOSSIP_DIR_DEBUG, + "%s: ctx->pos:%lld\n", + __func__, + lld(ctx->pos)); + + } + + /* + * we ran all the way through the last batch, set up for + * getting another batch... + */ + if (ret) { + *ptoken = rhandle.readdir_response.token; + ctx->pos = PVFS_ITERATE_NEXT; + } + + /* + * Did we hit the end of the directory? + */ + if (rhandle.readdir_response.token == PVFS_READDIR_END && + !buffer_full) { + gossip_debug(GOSSIP_DIR_DEBUG, + "End of dir detected; setting ctx->pos to PVFS_READDIR_END.\n"); + ctx->pos = PVFS_READDIR_END; + } + +out_destroy_handle: + readdir_handle_dtor(bufmap, &rhandle); +out_free_op: + op_release(new_op); + gossip_debug(GOSSIP_DIR_DEBUG, "pvfs2_readdir returning %d\n", ret); + return ret; +} + +static int pvfs2_dir_open(struct inode *inode, struct file *file) +{ + __u64 *ptoken; + + file->private_data = kmalloc(sizeof(__u64), GFP_KERNEL); + if (!file->private_data) + return -ENOMEM; + + ptoken = file->private_data; + *ptoken = PVFS_READDIR_START; + return 0; +} + +static int pvfs2_dir_release(struct inode *inode, struct file *file) +{ + pvfs2_flush_inode(inode); + kfree(file->private_data); + return 0; +} + +/** PVFS2 implementation of VFS directory operations */ +const struct file_operations pvfs2_dir_operations = { + .read = generic_read_dir, + .iterate = pvfs2_readdir, + .open = pvfs2_dir_open, + .release = pvfs2_dir_release, +}; diff --git a/fs/orangefs/downcall.h b/fs/orangefs/downcall.h new file mode 100644 index 000000000000..f8bea46e7c6a --- /dev/null +++ b/fs/orangefs/downcall.h @@ -0,0 +1,138 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +/* + * Definitions of downcalls used in Linux kernel module. + */ + +#ifndef __DOWNCALL_H +#define __DOWNCALL_H + +/* + * Sanitized the device-client core interaction + * for clean 32-64 bit usage + */ +struct pvfs2_io_response { + __s64 amt_complete; +}; + +struct pvfs2_iox_response { + __s64 amt_complete; +}; + +struct pvfs2_lookup_response { + struct pvfs2_object_kref refn; +}; + +struct pvfs2_create_response { + struct pvfs2_object_kref refn; +}; + +struct pvfs2_symlink_response { + struct pvfs2_object_kref refn; +}; + +struct pvfs2_getattr_response { + struct PVFS_sys_attr_s attributes; + char link_target[PVFS2_NAME_LEN]; +}; + +struct pvfs2_mkdir_response { + struct pvfs2_object_kref refn; +}; + +/* + * duplication of some system interface structures so that I don't have + * to allocate extra memory + */ +struct pvfs2_dirent { + char *d_name; + int d_length; + struct pvfs2_khandle khandle; +}; + +struct pvfs2_statfs_response { + __s64 block_size; + __s64 blocks_total; + __s64 blocks_avail; + __s64 files_total; + __s64 files_avail; +}; + +struct pvfs2_fs_mount_response { + __s32 fs_id; + __s32 id; + struct pvfs2_khandle root_khandle; +}; + +/* the getxattr response is the attribute value */ +struct pvfs2_getxattr_response { + __s32 val_sz; + __s32 __pad1; + char val[PVFS_MAX_XATTR_VALUELEN]; +}; + +/* the listxattr response is an array of attribute names */ +struct pvfs2_listxattr_response { + __s32 returned_count; + __s32 __pad1; + __u64 token; + char key[PVFS_MAX_XATTR_LISTLEN * PVFS_MAX_XATTR_NAMELEN]; + __s32 keylen; + __s32 __pad2; + __s32 lengths[PVFS_MAX_XATTR_LISTLEN]; +}; + +struct pvfs2_param_response { + __s64 value; +}; + +#define PERF_COUNT_BUF_SIZE 4096 +struct pvfs2_perf_count_response { + char buffer[PERF_COUNT_BUF_SIZE]; +}; + +#define FS_KEY_BUF_SIZE 4096 +struct pvfs2_fs_key_response { + __s32 fs_keylen; + __s32 __pad1; + char fs_key[FS_KEY_BUF_SIZE]; +}; + +struct pvfs2_downcall_s { + __s32 type; + __s32 status; + /* currently trailer is used only by readdir */ + __s64 trailer_size; + char *trailer_buf; + + union { + struct pvfs2_io_response io; + struct pvfs2_iox_response iox; + struct pvfs2_lookup_response lookup; + struct pvfs2_create_response create; + struct pvfs2_symlink_response sym; + struct pvfs2_getattr_response getattr; + struct pvfs2_mkdir_response mkdir; + struct pvfs2_statfs_response statfs; + struct pvfs2_fs_mount_response fs_mount; + struct pvfs2_getxattr_response getxattr; + struct pvfs2_listxattr_response listxattr; + struct pvfs2_param_response param; + struct pvfs2_perf_count_response perf_count; + struct pvfs2_fs_key_response fs_key; + } resp; +}; + +struct pvfs2_readdir_response_s { + __u64 token; + __u64 directory_version; + __u32 __pad2; + __u32 pvfs_dirent_outcount; + struct pvfs2_dirent *dirent_array; +}; + +#endif /* __DOWNCALL_H */ diff --git a/fs/orangefs/file.c b/fs/orangefs/file.c new file mode 100644 index 000000000000..feb1764c2f80 --- /dev/null +++ b/fs/orangefs/file.c @@ -0,0 +1,1002 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +/* + * Linux VFS file operations. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" +#include +#include + +#define wake_up_daemon_for_return(op) \ +do { \ + spin_lock(&op->lock); \ + op->io_completed = 1; \ + spin_unlock(&op->lock); \ + wake_up_interruptible(&op->io_completion_waitq);\ +} while (0) + +/* + * Copy to client-core's address space from the buffers specified + * by the iovec upto total_size bytes. + * NOTE: the iovector can either contain addresses which + * can futher be kernel-space or user-space addresses. + * or it can pointers to struct page's + */ +static int precopy_buffers(struct pvfs2_bufmap *bufmap, + int buffer_index, + const struct iovec *vec, + unsigned long nr_segs, + size_t total_size) +{ + int ret = 0; + struct iov_iter iter; + + /* + * copy data from application/kernel by pulling it out + * of the iovec. + */ + + + if (total_size) { + iov_iter_init(&iter, WRITE, vec, nr_segs, total_size); + ret = pvfs_bufmap_copy_from_iovec(bufmap, + &iter, + buffer_index, + total_size); + if (ret < 0) + gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n", + __func__, + (long)ret); + } + + if (ret < 0) + gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n", + __func__, + (long)ret); + return ret; +} + +/* + * Copy from client-core's address space to the buffers specified + * by the iovec upto total_size bytes. + * NOTE: the iovector can either contain addresses which + * can futher be kernel-space or user-space addresses. + * or it can pointers to struct page's + */ +static int postcopy_buffers(struct pvfs2_bufmap *bufmap, + int buffer_index, + const struct iovec *vec, + int nr_segs, + size_t total_size) +{ + int ret = 0; + + struct iov_iter iter; + + /* + * copy data to application/kernel by pushing it out to + * the iovec. NOTE; target buffers can be addresses or + * struct page pointers. + */ + if (total_size) { + iov_iter_init(&iter, READ, vec, nr_segs, total_size); + ret = pvfs_bufmap_copy_to_iovec(bufmap, + &iter, + buffer_index); + if (ret < 0) + gossip_err("%s: Failed to copy-out buffers. Please make sure that the pvfs2-client is running (%ld)\n", + __func__, + (long)ret); + } + return ret; +} + +/* + * Post and wait for the I/O upcall to finish + */ +static ssize_t wait_for_direct_io(enum PVFS_io_type type, struct inode *inode, + loff_t *offset, struct iovec *vec, unsigned long nr_segs, + size_t total_size, loff_t readahead_size) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_khandle *handle = &pvfs2_inode->refn.khandle; + struct pvfs2_bufmap *bufmap = NULL; + struct pvfs2_kernel_op_s *new_op = NULL; + int buffer_index = -1; + ssize_t ret; + + new_op = op_alloc(PVFS2_VFS_OP_FILE_IO); + if (!new_op) { + ret = -ENOMEM; + goto out; + } + /* synchronous I/O */ + new_op->upcall.req.io.async_vfs_io = PVFS_VFS_SYNC_IO; + new_op->upcall.req.io.readahead_size = readahead_size; + new_op->upcall.req.io.io_type = type; + new_op->upcall.req.io.refn = pvfs2_inode->refn; + +populate_shared_memory: + /* get a shared buffer index */ + ret = pvfs_bufmap_get(&bufmap, &buffer_index); + if (ret < 0) { + gossip_debug(GOSSIP_FILE_DEBUG, + "%s: pvfs_bufmap_get failure (%ld)\n", + __func__, (long)ret); + goto out; + } + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): GET op %p -> buffer_index %d\n", + __func__, + handle, + new_op, + buffer_index); + + new_op->uses_shared_memory = 1; + new_op->upcall.req.io.buf_index = buffer_index; + new_op->upcall.req.io.count = total_size; + new_op->upcall.req.io.offset = *offset; + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): nr_segs %lu, offset: %llu total_size: %zd\n", + __func__, + handle, + nr_segs, + llu(*offset), + total_size); + /* + * Stage 1: copy the buffers into client-core's address space + * precopy_buffers only pertains to writes. + */ + if (type == PVFS_IO_WRITE) { + ret = precopy_buffers(bufmap, + buffer_index, + vec, + nr_segs, + total_size); + if (ret < 0) + goto out; + } + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): Calling post_io_request with tag (%llu)\n", + __func__, + handle, + llu(new_op->tag)); + + /* Stage 2: Service the I/O operation */ + ret = service_operation(new_op, + type == PVFS_IO_WRITE ? + "file_write" : + "file_read", + get_interruptible_flag(inode)); + + /* + * If service_operation() returns -EAGAIN #and# the operation was + * purged from pvfs2_request_list or htable_ops_in_progress, then + * we know that the client was restarted, causing the shared memory + * area to be wiped clean. To restart a write operation in this + * case, we must re-copy the data from the user's iovec to a NEW + * shared memory location. To restart a read operation, we must get + * a new shared memory location. + */ + if (ret == -EAGAIN && op_state_purged(new_op)) { + pvfs_bufmap_put(bufmap, buffer_index); + gossip_debug(GOSSIP_FILE_DEBUG, + "%s:going to repopulate_shared_memory.\n", + __func__); + goto populate_shared_memory; + } + + if (ret < 0) { + handle_io_error(); /* defined in pvfs2-kernel.h */ + /* + * don't write an error to syslog on signaled operation + * termination unless we've got debugging turned on, as + * this can happen regularly (i.e. ctrl-c) + */ + if (ret == -EINTR) + gossip_debug(GOSSIP_FILE_DEBUG, + "%s: returning error %ld\n", __func__, + (long)ret); + else + gossip_err("%s: error in %s handle %pU, returning %zd\n", + __func__, + type == PVFS_IO_READ ? + "read from" : "write to", + handle, ret); + goto out; + } + + /* + * Stage 3: Post copy buffers from client-core's address space + * postcopy_buffers only pertains to reads. + */ + if (type == PVFS_IO_READ) { + ret = postcopy_buffers(bufmap, + buffer_index, + vec, + nr_segs, + new_op->downcall.resp.io.amt_complete); + if (ret < 0) { + /* + * put error codes in downcall so that handle_io_error() + * preserves it properly + */ + new_op->downcall.status = ret; + handle_io_error(); + goto out; + } + } + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): Amount written as returned by the sys-io call:%d\n", + __func__, + handle, + (int)new_op->downcall.resp.io.amt_complete); + + ret = new_op->downcall.resp.io.amt_complete; + + /* + * tell the device file owner waiting on I/O that this read has + * completed and it can return now. in this exact case, on + * wakeup the daemon will free the op, so we *cannot* touch it + * after this. + */ + wake_up_daemon_for_return(new_op); + new_op = NULL; + +out: + if (buffer_index >= 0) { + pvfs_bufmap_put(bufmap, buffer_index); + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): PUT buffer_index %d\n", + __func__, handle, buffer_index); + buffer_index = -1; + } + if (new_op) { + op_release(new_op); + new_op = NULL; + } + return ret; +} + +/* + * The reason we need to do this is to be able to support readv and writev + * that are larger than (pvfs_bufmap_size_query()) Default is + * PVFS2_BUFMAP_DEFAULT_DESC_SIZE MB. What that means is that we will + * create a new io vec descriptor for those memory addresses that + * go beyond the limit. Return value for this routine is negative in case + * of errors and 0 in case of success. + * + * Further, the new_nr_segs pointer is updated to hold the new value + * of number of iovecs, the new_vec pointer is updated to hold the pointer + * to the new split iovec, and the size array is an array of integers holding + * the number of iovecs that straddle pvfs_bufmap_size_query(). + * The max_new_nr_segs value is computed by the caller and returned. + * (It will be (count of all iov_len/ block_size) + 1). + */ +static int split_iovecs(unsigned long max_new_nr_segs, /* IN */ + unsigned long nr_segs, /* IN */ + const struct iovec *original_iovec, /* IN */ + unsigned long *new_nr_segs, /* OUT */ + struct iovec **new_vec, /* OUT */ + unsigned long *seg_count, /* OUT */ + unsigned long **seg_array) /* OUT */ +{ + unsigned long seg; + unsigned long count = 0; + unsigned long begin_seg; + unsigned long tmpnew_nr_segs = 0; + struct iovec *new_iovec = NULL; + struct iovec *orig_iovec; + unsigned long *sizes = NULL; + unsigned long sizes_count = 0; + + if (nr_segs <= 0 || + original_iovec == NULL || + new_nr_segs == NULL || + new_vec == NULL || + seg_count == NULL || + seg_array == NULL || + max_new_nr_segs <= 0) { + gossip_err("Invalid parameters to split_iovecs\n"); + return -EINVAL; + } + *new_nr_segs = 0; + *new_vec = NULL; + *seg_count = 0; + *seg_array = NULL; + /* copy the passed in iovec descriptor to a temp structure */ + orig_iovec = kmalloc_array(nr_segs, + sizeof(*orig_iovec), + PVFS2_BUFMAP_GFP_FLAGS); + if (orig_iovec == NULL) { + gossip_err( + "split_iovecs: Could not allocate memory for %lu bytes!\n", + (unsigned long)(nr_segs * sizeof(*orig_iovec))); + return -ENOMEM; + } + new_iovec = kcalloc(max_new_nr_segs, + sizeof(*new_iovec), + PVFS2_BUFMAP_GFP_FLAGS); + if (new_iovec == NULL) { + kfree(orig_iovec); + gossip_err( + "split_iovecs: Could not allocate memory for %lu bytes!\n", + (unsigned long)(max_new_nr_segs * sizeof(*new_iovec))); + return -ENOMEM; + } + sizes = kcalloc(max_new_nr_segs, + sizeof(*sizes), + PVFS2_BUFMAP_GFP_FLAGS); + if (sizes == NULL) { + kfree(new_iovec); + kfree(orig_iovec); + gossip_err( + "split_iovecs: Could not allocate memory for %lu bytes!\n", + (unsigned long)(max_new_nr_segs * sizeof(*sizes))); + return -ENOMEM; + } + /* copy the passed in iovec to a temp structure */ + memcpy(orig_iovec, original_iovec, nr_segs * sizeof(*orig_iovec)); + begin_seg = 0; +repeat: + for (seg = begin_seg; seg < nr_segs; seg++) { + if (tmpnew_nr_segs >= max_new_nr_segs || + sizes_count >= max_new_nr_segs) { + kfree(sizes); + kfree(orig_iovec); + kfree(new_iovec); + gossip_err + ("split_iovecs: exceeded the index limit (%lu)\n", + tmpnew_nr_segs); + return -EINVAL; + } + if (count + orig_iovec[seg].iov_len < + pvfs_bufmap_size_query()) { + count += orig_iovec[seg].iov_len; + memcpy(&new_iovec[tmpnew_nr_segs], + &orig_iovec[seg], + sizeof(*new_iovec)); + tmpnew_nr_segs++; + sizes[sizes_count]++; + } else { + new_iovec[tmpnew_nr_segs].iov_base = + orig_iovec[seg].iov_base; + new_iovec[tmpnew_nr_segs].iov_len = + (pvfs_bufmap_size_query() - count); + tmpnew_nr_segs++; + sizes[sizes_count]++; + sizes_count++; + begin_seg = seg; + orig_iovec[seg].iov_base += + (pvfs_bufmap_size_query() - count); + orig_iovec[seg].iov_len -= + (pvfs_bufmap_size_query() - count); + count = 0; + break; + } + } + if (seg != nr_segs) + goto repeat; + else + sizes_count++; + + *new_nr_segs = tmpnew_nr_segs; + /* new_iovec is freed by the caller */ + *new_vec = new_iovec; + *seg_count = sizes_count; + /* seg_array is also freed by the caller */ + *seg_array = sizes; + kfree(orig_iovec); + return 0; +} + +static long bound_max_iovecs(const struct iovec *curr, unsigned long nr_segs, + ssize_t *total_count) +{ + unsigned long i; + long max_nr_iovecs; + ssize_t total; + ssize_t count; + + total = 0; + count = 0; + max_nr_iovecs = 0; + for (i = 0; i < nr_segs; i++) { + const struct iovec *iv = &curr[i]; + + count += iv->iov_len; + if (unlikely((ssize_t) (count | iv->iov_len) < 0)) + return -EINVAL; + if (total + iv->iov_len < pvfs_bufmap_size_query()) { + total += iv->iov_len; + max_nr_iovecs++; + } else { + total = + (total + iv->iov_len - pvfs_bufmap_size_query()); + max_nr_iovecs += (total / pvfs_bufmap_size_query() + 2); + } + } + *total_count = count; + return max_nr_iovecs; +} + +/* + * Common entry point for read/write/readv/writev + * This function will dispatch it to either the direct I/O + * or buffered I/O path depending on the mount options and/or + * augmented/extended metadata attached to the file. + * Note: File extended attributes override any mount options. + */ +static ssize_t do_readv_writev(enum PVFS_io_type type, struct file *file, + loff_t *offset, const struct iovec *iov, unsigned long nr_segs) +{ + struct inode *inode = file->f_mapping->host; + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_khandle *handle = &pvfs2_inode->refn.khandle; + ssize_t ret; + ssize_t total_count; + unsigned int to_free; + size_t count; + unsigned long seg; + unsigned long new_nr_segs; + unsigned long max_new_nr_segs; + unsigned long seg_count; + unsigned long *seg_array; + struct iovec *iovecptr; + struct iovec *ptr; + + total_count = 0; + ret = -EINVAL; + count = 0; + to_free = 0; + + /* Compute total and max number of segments after split */ + max_new_nr_segs = bound_max_iovecs(iov, nr_segs, &count); + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n", + __func__, + handle, + (int)count); + + if (type == PVFS_IO_WRITE) { + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): proceeding with offset : %llu, " + "size %d\n", + __func__, + handle, + llu(*offset), + (int)count); + } + + if (count == 0) { + ret = 0; + goto out; + } + + /* + * if the total size of data transfer requested is greater than + * the kernel-set blocksize of PVFS2, then we split the iovecs + * such that no iovec description straddles a block size limit + */ + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s: pvfs_bufmap_size:%d\n", + __func__, + pvfs_bufmap_size_query()); + + if (count > pvfs_bufmap_size_query()) { + /* + * Split up the given iovec description such that + * no iovec descriptor straddles over the block-size limitation. + * This makes us our job easier to stage the I/O. + * In addition, this function will also compute an array + * with seg_count entries that will store the number of + * segments that straddle the block-size boundaries. + */ + ret = split_iovecs(max_new_nr_segs, /* IN */ + nr_segs, /* IN */ + iov, /* IN */ + &new_nr_segs, /* OUT */ + &iovecptr, /* OUT */ + &seg_count, /* OUT */ + &seg_array); /* OUT */ + if (ret < 0) { + gossip_err("%s: Failed to split iovecs to satisfy larger than blocksize readv/writev request %zd\n", + __func__, + ret); + goto out; + } + gossip_debug(GOSSIP_FILE_DEBUG, + "%s: Splitting iovecs from %lu to %lu" + " [max_new %lu]\n", + __func__, + nr_segs, + new_nr_segs, + max_new_nr_segs); + /* We must free seg_array and iovecptr */ + to_free = 1; + } else { + new_nr_segs = nr_segs; + /* use the given iovec description */ + iovecptr = (struct iovec *)iov; + /* There is only 1 element in the seg_array */ + seg_count = 1; + /* and its value is the number of segments passed in */ + seg_array = &nr_segs; + /* We dont have to free up anything */ + to_free = 0; + } + ptr = iovecptr; + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU) %zd@%llu\n", + __func__, + handle, + count, + llu(*offset)); + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): new_nr_segs: %lu, seg_count: %lu\n", + __func__, + handle, + new_nr_segs, seg_count); + +/* PVFS2_KERNEL_DEBUG is a CFLAGS define. */ +#ifdef PVFS2_KERNEL_DEBUG + for (seg = 0; seg < new_nr_segs; seg++) + gossip_debug(GOSSIP_FILE_DEBUG, + "%s: %d) %p to %p [%d bytes]\n", + __func__, + (int)seg + 1, + iovecptr[seg].iov_base, + iovecptr[seg].iov_base + iovecptr[seg].iov_len, + (int)iovecptr[seg].iov_len); + for (seg = 0; seg < seg_count; seg++) + gossip_debug(GOSSIP_FILE_DEBUG, + "%s: %zd) %lu\n", + __func__, + seg + 1, + seg_array[seg]); +#endif + seg = 0; + while (total_count < count) { + size_t each_count; + size_t amt_complete; + + /* how much to transfer in this loop iteration */ + each_count = + (((count - total_count) > pvfs_bufmap_size_query()) ? + pvfs_bufmap_size_query() : + (count - total_count)); + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): size of each_count(%d)\n", + __func__, + handle, + (int)each_count); + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): BEFORE wait_for_io: offset is %d\n", + __func__, + handle, + (int)*offset); + + ret = wait_for_direct_io(type, inode, offset, ptr, + seg_array[seg], each_count, 0); + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): return from wait_for_io:%d\n", + __func__, + handle, + (int)ret); + + if (ret < 0) + goto out; + + /* advance the iovec pointer */ + ptr += seg_array[seg]; + seg++; + *offset += ret; + total_count += ret; + amt_complete = ret; + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): AFTER wait_for_io: offset is %d\n", + __func__, + handle, + (int)*offset); + + /* + * if we got a short I/O operations, + * fall out and return what we got so far + */ + if (amt_complete < each_count) + break; + } /*end while */ + + if (total_count > 0) + ret = total_count; +out: + if (to_free) { + kfree(iovecptr); + kfree(seg_array); + } + if (ret > 0) { + if (type == PVFS_IO_READ) { + file_accessed(file); + } else { + SetMtimeFlag(pvfs2_inode); + inode->i_mtime = CURRENT_TIME; + mark_inode_dirty_sync(inode); + } + } + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): Value(%d) returned.\n", + __func__, + handle, + (int)ret); + + return ret; +} + +/* + * Read data from a specified offset in a file (referenced by inode). + * Data may be placed either in a user or kernel buffer. + */ +ssize_t pvfs2_inode_read(struct inode *inode, + char __user *buf, + size_t count, + loff_t *offset, + loff_t readahead_size) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + size_t bufmap_size; + struct iovec vec; + ssize_t ret = -EINVAL; + + g_pvfs2_stats.reads++; + + vec.iov_base = buf; + vec.iov_len = count; + + bufmap_size = pvfs_bufmap_size_query(); + if (count > bufmap_size) { + gossip_debug(GOSSIP_FILE_DEBUG, + "%s: count is too large (%zd/%zd)!\n", + __func__, count, bufmap_size); + return -EINVAL; + } + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU) %zd@%llu\n", + __func__, + &pvfs2_inode->refn.khandle, + count, + llu(*offset)); + + ret = wait_for_direct_io(PVFS_IO_READ, inode, offset, &vec, 1, + count, readahead_size); + if (ret > 0) + *offset += ret; + + gossip_debug(GOSSIP_FILE_DEBUG, + "%s(%pU): Value(%zd) returned.\n", + __func__, + &pvfs2_inode->refn.khandle, + ret); + + return ret; +} + +static ssize_t pvfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) +{ + struct file *file = iocb->ki_filp; + loff_t pos = *(&iocb->ki_pos); + ssize_t rc = 0; + unsigned long nr_segs = iter->nr_segs; + + BUG_ON(iocb->private); + + gossip_debug(GOSSIP_FILE_DEBUG, "pvfs2_file_read_iter\n"); + + g_pvfs2_stats.reads++; + + rc = do_readv_writev(PVFS_IO_READ, + file, + &pos, + iter->iov, + nr_segs); + iocb->ki_pos = pos; + + return rc; +} + +static ssize_t pvfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *iter) +{ + struct file *file = iocb->ki_filp; + loff_t pos = *(&iocb->ki_pos); + unsigned long nr_segs = iter->nr_segs; + ssize_t rc; + + BUG_ON(iocb->private); + + gossip_debug(GOSSIP_FILE_DEBUG, "pvfs2_file_write_iter\n"); + + mutex_lock(&file->f_mapping->host->i_mutex); + + /* Make sure generic_write_checks sees an up to date inode size. */ + if (file->f_flags & O_APPEND) { + rc = pvfs2_inode_getattr(file->f_mapping->host, + PVFS_ATTR_SYS_SIZE); + if (rc) { + gossip_err("%s: pvfs2_inode_getattr failed, rc:%zd:.\n", + __func__, rc); + goto out; + } + } + + if (file->f_pos > i_size_read(file->f_mapping->host)) + pvfs2_i_size_write(file->f_mapping->host, file->f_pos); + + rc = generic_write_checks(iocb, iter); + + if (rc <= 0) { + gossip_err("%s: generic_write_checks failed, rc:%zd:.\n", + __func__, rc); + goto out; + } + + rc = do_readv_writev(PVFS_IO_WRITE, + file, + &pos, + iter->iov, + nr_segs); + if (rc < 0) { + gossip_err("%s: do_readv_writev failed, rc:%zd:.\n", + __func__, rc); + goto out; + } + + iocb->ki_pos = pos; + g_pvfs2_stats.writes++; + +out: + + mutex_unlock(&file->f_mapping->host->i_mutex); + return rc; +} + +/* + * Perform a miscellaneous operation on a file. + */ +static long pvfs2_ioctl(struct file *file, unsigned int cmd, unsigned long arg) +{ + int ret = -ENOTTY; + __u64 val = 0; + unsigned long uval; + + gossip_debug(GOSSIP_FILE_DEBUG, + "pvfs2_ioctl: called with cmd %d\n", + cmd); + + /* + * we understand some general ioctls on files, such as the immutable + * and append flags + */ + if (cmd == FS_IOC_GETFLAGS) { + val = 0; + ret = pvfs2_xattr_get_default(file->f_path.dentry, + "user.pvfs2.meta_hint", + &val, + sizeof(val), + 0); + if (ret < 0 && ret != -ENODATA) + return ret; + else if (ret == -ENODATA) + val = 0; + uval = val; + gossip_debug(GOSSIP_FILE_DEBUG, + "pvfs2_ioctl: FS_IOC_GETFLAGS: %llu\n", + (unsigned long long)uval); + return put_user(uval, (int __user *)arg); + } else if (cmd == FS_IOC_SETFLAGS) { + ret = 0; + if (get_user(uval, (int __user *)arg)) + return -EFAULT; + /* + * PVFS_MIRROR_FL is set internally when the mirroring mode + * is turned on for a file. The user is not allowed to turn + * on this bit, but the bit is present if the user first gets + * the flags and then updates the flags with some new + * settings. So, we ignore it in the following edit. bligon. + */ + if ((uval & ~PVFS_MIRROR_FL) & + (~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL))) { + gossip_err("pvfs2_ioctl: the FS_IOC_SETFLAGS only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n"); + return -EINVAL; + } + val = uval; + gossip_debug(GOSSIP_FILE_DEBUG, + "pvfs2_ioctl: FS_IOC_SETFLAGS: %llu\n", + (unsigned long long)val); + ret = pvfs2_xattr_set_default(file->f_path.dentry, + "user.pvfs2.meta_hint", + &val, + sizeof(val), + 0, + 0); + } + + return ret; +} + +/* + * Memory map a region of a file. + */ +static int pvfs2_file_mmap(struct file *file, struct vm_area_struct *vma) +{ + gossip_debug(GOSSIP_FILE_DEBUG, + "pvfs2_file_mmap: called on %s\n", + (file ? + (char *)file->f_path.dentry->d_name.name : + (char *)"Unknown")); + + /* set the sequential readahead hint */ + vma->vm_flags |= VM_SEQ_READ; + vma->vm_flags &= ~VM_RAND_READ; + + /* Use readonly mmap since we cannot support writable maps. */ + return generic_file_readonly_mmap(file, vma); +} + +#define mapping_nrpages(idata) ((idata)->nrpages) + +/* + * Called to notify the module that there are no more references to + * this file (i.e. no processes have it open). + * + * \note Not called when each file is closed. + */ +static int pvfs2_file_release(struct inode *inode, struct file *file) +{ + gossip_debug(GOSSIP_FILE_DEBUG, + "pvfs2_file_release: called on %s\n", + file->f_path.dentry->d_name.name); + + pvfs2_flush_inode(inode); + + /* + * remove all associated inode pages from the page cache and mmap + * readahead cache (if any); this forces an expensive refresh of + * data for the next caller of mmap (or 'get_block' accesses) + */ + if (file->f_path.dentry->d_inode && + file->f_path.dentry->d_inode->i_mapping && + mapping_nrpages(&file->f_path.dentry->d_inode->i_data)) + truncate_inode_pages(file->f_path.dentry->d_inode->i_mapping, + 0); + return 0; +} + +/* + * Push all data for a specific file onto permanent storage. + */ +static int pvfs2_fsync(struct file *file, + loff_t start, + loff_t end, + int datasync) +{ + int ret = -EINVAL; + struct pvfs2_inode_s *pvfs2_inode = + PVFS2_I(file->f_path.dentry->d_inode); + struct pvfs2_kernel_op_s *new_op = NULL; + + /* required call */ + filemap_write_and_wait_range(file->f_mapping, start, end); + + new_op = op_alloc(PVFS2_VFS_OP_FSYNC); + if (!new_op) + return -ENOMEM; + new_op->upcall.req.fsync.refn = pvfs2_inode->refn; + + ret = service_operation(new_op, + "pvfs2_fsync", + get_interruptible_flag(file->f_path.dentry->d_inode)); + + gossip_debug(GOSSIP_FILE_DEBUG, + "pvfs2_fsync got return value of %d\n", + ret); + + op_release(new_op); + + pvfs2_flush_inode(file->f_path.dentry->d_inode); + return ret; +} + +/* + * Change the file pointer position for an instance of an open file. + * + * \note If .llseek is overriden, we must acquire lock as described in + * Documentation/filesystems/Locking. + * + * Future upgrade could support SEEK_DATA and SEEK_HOLE but would + * require much changes to the FS + */ +static loff_t pvfs2_file_llseek(struct file *file, loff_t offset, int origin) +{ + int ret = -EINVAL; + struct inode *inode = file->f_path.dentry->d_inode; + + if (!inode) { + gossip_err("pvfs2_file_llseek: invalid inode (NULL)\n"); + return ret; + } + + if (origin == PVFS2_SEEK_END) { + /* + * revalidate the inode's file size. + * NOTE: We are only interested in file size here, + * so we set mask accordingly. + */ + ret = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_SIZE); + if (ret) { + gossip_debug(GOSSIP_FILE_DEBUG, + "%s:%s:%d calling make bad inode\n", + __FILE__, + __func__, + __LINE__); + pvfs2_make_bad_inode(inode); + return ret; + } + } + + gossip_debug(GOSSIP_FILE_DEBUG, + "pvfs2_file_llseek: offset is %ld | origin is %d" + " | inode size is %lu\n", + (long)offset, + origin, + (unsigned long)file->f_path.dentry->d_inode->i_size); + + return generic_file_llseek(file, offset, origin); +} + +/* + * Support local locks (locks that only this kernel knows about) + * if Orangefs was mounted -o local_lock. + */ +static int pvfs2_lock(struct file *filp, int cmd, struct file_lock *fl) +{ + int rc = -EINVAL; + + if (PVFS2_SB(filp->f_inode->i_sb)->flags & PVFS2_OPT_LOCAL_LOCK) { + if (cmd == F_GETLK) { + rc = 0; + posix_test_lock(filp, fl); + } else { + rc = posix_lock_file(filp, fl, NULL); + } + } + + return rc; +} + +/** PVFS2 implementation of VFS file operations */ +const struct file_operations pvfs2_file_operations = { + .llseek = pvfs2_file_llseek, + .read_iter = pvfs2_file_read_iter, + .write_iter = pvfs2_file_write_iter, + .lock = pvfs2_lock, + .unlocked_ioctl = pvfs2_ioctl, + .mmap = pvfs2_file_mmap, + .open = generic_file_open, + .release = pvfs2_file_release, + .fsync = pvfs2_fsync, +}; diff --git a/fs/orangefs/inode.c b/fs/orangefs/inode.c new file mode 100644 index 000000000000..4f7c45a44c1f --- /dev/null +++ b/fs/orangefs/inode.c @@ -0,0 +1,469 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +/* + * Linux VFS inode operations. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" + +static int read_one_page(struct page *page) +{ + void *page_data; + int ret; + int max_block; + ssize_t bytes_read = 0; + struct inode *inode = page->mapping->host; + const __u32 blocksize = PAGE_CACHE_SIZE; /* inode->i_blksize */ + const __u32 blockbits = PAGE_CACHE_SHIFT; /* inode->i_blkbits */ + + gossip_debug(GOSSIP_INODE_DEBUG, + "pvfs2_readpage called with page %p\n", + page); + page_data = pvfs2_kmap(page); + + max_block = ((inode->i_size / blocksize) + 1); + + if (page->index < max_block) { + loff_t blockptr_offset = (((loff_t) page->index) << blockbits); + + bytes_read = pvfs2_inode_read(inode, + (char __user *) page_data, + blocksize, + &blockptr_offset, + inode->i_size); + } + /* only zero remaining unread portions of the page data */ + if (bytes_read > 0) + memset(page_data + bytes_read, 0, blocksize - bytes_read); + else + memset(page_data, 0, blocksize); + /* takes care of potential aliasing */ + flush_dcache_page(page); + if (bytes_read < 0) { + ret = bytes_read; + SetPageError(page); + } else { + SetPageUptodate(page); + if (PageError(page)) + ClearPageError(page); + ret = 0; + } + pvfs2_kunmap(page); + /* unlock the page after the ->readpage() routine completes */ + unlock_page(page); + return ret; +} + +static int pvfs2_readpage(struct file *file, struct page *page) +{ + return read_one_page(page); +} + +static int pvfs2_readpages(struct file *file, + struct address_space *mapping, + struct list_head *pages, + unsigned nr_pages) +{ + int page_idx; + int ret; + + gossip_debug(GOSSIP_INODE_DEBUG, "pvfs2_readpages called\n"); + + for (page_idx = 0; page_idx < nr_pages; page_idx++) { + struct page *page; + + page = list_entry(pages->prev, struct page, lru); + list_del(&page->lru); + if (!add_to_page_cache(page, + mapping, + page->index, + GFP_KERNEL)) { + ret = read_one_page(page); + gossip_debug(GOSSIP_INODE_DEBUG, + "failure adding page to cache, read_one_page returned: %d\n", + ret); + } else { + page_cache_release(page); + } + } + BUG_ON(!list_empty(pages)); + return 0; +} + +static void pvfs2_invalidatepage(struct page *page, + unsigned int offset, + unsigned int length) +{ + gossip_debug(GOSSIP_INODE_DEBUG, + "pvfs2_invalidatepage called on page %p " + "(offset is %u)\n", + page, + offset); + + ClearPageUptodate(page); + ClearPageMappedToDisk(page); + return; + +} + +static int pvfs2_releasepage(struct page *page, gfp_t foo) +{ + gossip_debug(GOSSIP_INODE_DEBUG, + "pvfs2_releasepage called on page %p\n", + page); + return 0; +} + +/* + * Having a direct_IO entry point in the address_space_operations + * struct causes the kernel to allows us to use O_DIRECT on + * open. Nothing will ever call this thing, but in the future we + * will need to be able to use O_DIRECT on open in order to support + * AIO. Modeled after NFS, they do this too. + */ +/* + * static ssize_t pvfs2_direct_IO(int rw, + * struct kiocb *iocb, + * struct iov_iter *iter, + * loff_t offset) + *{ + * gossip_debug(GOSSIP_INODE_DEBUG, + * "pvfs2_direct_IO: %s\n", + * iocb->ki_filp->f_path.dentry->d_name.name); + * + * return -EINVAL; + *} + */ + +struct backing_dev_info pvfs2_backing_dev_info = { + .name = "pvfs2", + .ra_pages = 0, + .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, +}; + +/** PVFS2 implementation of address space operations */ +const struct address_space_operations pvfs2_address_operations = { + .readpage = pvfs2_readpage, + .readpages = pvfs2_readpages, + .invalidatepage = pvfs2_invalidatepage, + .releasepage = pvfs2_releasepage, +/* .direct_IO = pvfs2_direct_IO */ +}; + +static int pvfs2_setattr_size(struct inode *inode, struct iattr *iattr) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_kernel_op_s *new_op; + loff_t orig_size = i_size_read(inode); + int ret = -EINVAL; + + gossip_debug(GOSSIP_INODE_DEBUG, + "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n", + __func__, + get_khandle_from_ino(inode), + &pvfs2_inode->refn.khandle, + pvfs2_inode->refn.fs_id, + iattr->ia_size); + + truncate_setsize(inode, iattr->ia_size); + + new_op = op_alloc(PVFS2_VFS_OP_TRUNCATE); + if (!new_op) + return -ENOMEM; + + new_op->upcall.req.truncate.refn = pvfs2_inode->refn; + new_op->upcall.req.truncate.size = (__s64) iattr->ia_size; + + ret = service_operation(new_op, __func__, + get_interruptible_flag(inode)); + + /* + * the truncate has no downcall members to retrieve, but + * the status value tells us if it went through ok or not + */ + gossip_debug(GOSSIP_INODE_DEBUG, + "pvfs2: pvfs2_truncate got return value of %d\n", + ret); + + op_release(new_op); + + if (ret != 0) + return ret; + + /* + * Only change the c/mtime if we are changing the size or we are + * explicitly asked to change it. This handles the semantic difference + * between truncate() and ftruncate() as implemented in the VFS. + * + * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a + * special case where we need to update the times despite not having + * these flags set. For all other operations the VFS set these flags + * explicitly if it wants a timestamp update. + */ + if (orig_size != i_size_read(inode) && + !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { + iattr->ia_ctime = iattr->ia_mtime = + current_fs_time(inode->i_sb); + iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; + } + + return ret; +} + +/* + * Change attributes of an object referenced by dentry. + */ +int pvfs2_setattr(struct dentry *dentry, struct iattr *iattr) +{ + int ret = -EINVAL; + struct inode *inode = dentry->d_inode; + + gossip_debug(GOSSIP_INODE_DEBUG, + "pvfs2_setattr: called on %s\n", + dentry->d_name.name); + + ret = inode_change_ok(inode, iattr); + if (ret) + goto out; + + if ((iattr->ia_valid & ATTR_SIZE) && + iattr->ia_size != i_size_read(inode)) { + ret = pvfs2_setattr_size(inode, iattr); + if (ret) + goto out; + } + + setattr_copy(inode, iattr); + mark_inode_dirty(inode); + + ret = pvfs2_inode_setattr(inode, iattr); + gossip_debug(GOSSIP_INODE_DEBUG, + "pvfs2_setattr: inode_setattr returned %d\n", + ret); + + if (!ret && (iattr->ia_valid & ATTR_MODE)) + /* change mod on a file that has ACLs */ + ret = posix_acl_chmod(inode, inode->i_mode); + +out: + gossip_debug(GOSSIP_INODE_DEBUG, "pvfs2_setattr: returning %d\n", ret); + return ret; +} + +/* + * Obtain attributes of an object given a dentry + */ +int pvfs2_getattr(struct vfsmount *mnt, + struct dentry *dentry, + struct kstat *kstat) +{ + int ret = -ENOENT; + struct inode *inode = dentry->d_inode; + struct pvfs2_inode_s *pvfs2_inode = NULL; + + gossip_debug(GOSSIP_INODE_DEBUG, + "pvfs2_getattr: called on %s\n", + dentry->d_name.name); + + /* + * Similar to the above comment, a getattr also expects that all + * fields/attributes of the inode would be refreshed. So again, we + * dont have too much of a choice but refresh all the attributes. + */ + ret = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); + if (ret == 0) { + generic_fillattr(inode, kstat); + /* override block size reported to stat */ + pvfs2_inode = PVFS2_I(inode); + kstat->blksize = pvfs2_inode->blksize; + } else { + /* assume an I/O error and flag inode as bad */ + gossip_debug(GOSSIP_INODE_DEBUG, + "%s:%s:%d calling make bad inode\n", + __FILE__, + __func__, + __LINE__); + pvfs2_make_bad_inode(inode); + } + return ret; +} + +/* PVFS2 implementation of VFS inode operations for files */ +struct inode_operations pvfs2_file_inode_operations = { + .get_acl = pvfs2_get_acl, + .set_acl = pvfs2_set_acl, + .setattr = pvfs2_setattr, + .getattr = pvfs2_getattr, + .setxattr = generic_setxattr, + .getxattr = generic_getxattr, + .listxattr = pvfs2_listxattr, + .removexattr = generic_removexattr, +}; + +static int pvfs2_init_iops(struct inode *inode) +{ + inode->i_mapping->a_ops = &pvfs2_address_operations; + + switch (inode->i_mode & S_IFMT) { + case S_IFREG: + inode->i_op = &pvfs2_file_inode_operations; + inode->i_fop = &pvfs2_file_operations; + inode->i_blkbits = PAGE_CACHE_SHIFT; + break; + case S_IFLNK: + inode->i_op = &pvfs2_symlink_inode_operations; + break; + case S_IFDIR: + inode->i_op = &pvfs2_dir_inode_operations; + inode->i_fop = &pvfs2_dir_operations; + break; + default: + gossip_debug(GOSSIP_INODE_DEBUG, + "%s: unsupported mode\n", + __func__); + return -EINVAL; + } + + return 0; +} + +/* + * Given a PVFS2 object identifier (fsid, handle), convert it into a ino_t type + * that will be used as a hash-index from where the handle will + * be searched for in the VFS hash table of inodes. + */ +static inline ino_t pvfs2_handle_hash(struct pvfs2_object_kref *ref) +{ + if (!ref) + return 0; + return pvfs2_khandle_to_ino(&(ref->khandle)); +} + +/* + * Called to set up an inode from iget5_locked. + */ +static int pvfs2_set_inode(struct inode *inode, void *data) +{ + struct pvfs2_object_kref *ref = (struct pvfs2_object_kref *) data; + struct pvfs2_inode_s *pvfs2_inode = NULL; + + /* Make sure that we have sane parameters */ + if (!data || !inode) + return 0; + pvfs2_inode = PVFS2_I(inode); + if (!pvfs2_inode) + return 0; + pvfs2_inode->refn.fs_id = ref->fs_id; + pvfs2_inode->refn.khandle = ref->khandle; + return 0; +} + +/* + * Called to determine if handles match. + */ +static int pvfs2_test_inode(struct inode *inode, void *data) +{ + struct pvfs2_object_kref *ref = (struct pvfs2_object_kref *) data; + struct pvfs2_inode_s *pvfs2_inode = NULL; + + pvfs2_inode = PVFS2_I(inode); + return (!PVFS_khandle_cmp(&(pvfs2_inode->refn.khandle), &(ref->khandle)) + && pvfs2_inode->refn.fs_id == ref->fs_id); +} + +/* + * Front-end to lookup the inode-cache maintained by the VFS using the PVFS2 + * file handle. + * + * @sb: the file system super block instance. + * @ref: The PVFS2 object for which we are trying to locate an inode structure. + */ +struct inode *pvfs2_iget(struct super_block *sb, struct pvfs2_object_kref *ref) +{ + struct inode *inode = NULL; + unsigned long hash; + int error; + + hash = pvfs2_handle_hash(ref); + inode = iget5_locked(sb, hash, pvfs2_test_inode, pvfs2_set_inode, ref); + if (!inode || !(inode->i_state & I_NEW)) + return inode; + + error = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); + if (error) { + iget_failed(inode); + return ERR_PTR(error); + } + + inode->i_ino = hash; /* needed for stat etc */ + pvfs2_init_iops(inode); + unlock_new_inode(inode); + + gossip_debug(GOSSIP_INODE_DEBUG, + "iget handle %pU, fsid %d hash %ld i_ino %lu\n", + &ref->khandle, + ref->fs_id, + hash, + inode->i_ino); + + return inode; +} + +/* + * Allocate an inode for a newly created file and insert it into the inode hash. + */ +struct inode *pvfs2_new_inode(struct super_block *sb, struct inode *dir, + int mode, dev_t dev, struct pvfs2_object_kref *ref) +{ + unsigned long hash = pvfs2_handle_hash(ref); + struct inode *inode; + int error; + + gossip_debug(GOSSIP_INODE_DEBUG, + "pvfs2_get_custom_inode_common: called\n" + "(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n", + sb, + MAJOR(dev), + MINOR(dev), + mode); + + inode = new_inode(sb); + if (!inode) + return NULL; + + pvfs2_set_inode(inode, ref); + inode->i_ino = hash; /* needed for stat etc */ + + error = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); + if (error) + goto out_iput; + + pvfs2_init_iops(inode); + + inode->i_mode = mode; + inode->i_uid = current_fsuid(); + inode->i_gid = current_fsgid(); + inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; + inode->i_size = PAGE_CACHE_SIZE; + inode->i_rdev = dev; + + error = insert_inode_locked4(inode, hash, pvfs2_test_inode, ref); + if (error < 0) + goto out_iput; + + gossip_debug(GOSSIP_INODE_DEBUG, + "Initializing ACL's for inode %pU\n", + get_khandle_from_ino(inode)); + pvfs2_init_acl(inode, dir); + return inode; + +out_iput: + iput(inode); + return ERR_PTR(error); +} diff --git a/fs/orangefs/namei.c b/fs/orangefs/namei.c new file mode 100644 index 000000000000..05f6feadfd0d --- /dev/null +++ b/fs/orangefs/namei.c @@ -0,0 +1,469 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +/* + * Linux VFS namei operations. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" + +/* + * Get a newly allocated inode to go with a negative dentry. + */ +static int pvfs2_create(struct inode *dir, + struct dentry *dentry, + umode_t mode, + bool exclusive) +{ + struct pvfs2_inode_s *parent = PVFS2_I(dir); + struct pvfs2_kernel_op_s *new_op; + struct inode *inode; + int ret; + + gossip_debug(GOSSIP_NAME_DEBUG, "%s: called\n", __func__); + + new_op = op_alloc(PVFS2_VFS_OP_CREATE); + if (!new_op) + return -ENOMEM; + + new_op->upcall.req.create.parent_refn = parent->refn; + + fill_default_sys_attrs(new_op->upcall.req.create.attributes, + PVFS_TYPE_METAFILE, mode); + + strncpy(new_op->upcall.req.create.d_name, + dentry->d_name.name, PVFS2_NAME_LEN); + + ret = service_operation(new_op, __func__, get_interruptible_flag(dir)); + + gossip_debug(GOSSIP_NAME_DEBUG, + "Create Got PVFS2 handle %pU on fsid %d (ret=%d)\n", + &new_op->downcall.resp.create.refn.khandle, + new_op->downcall.resp.create.refn.fs_id, ret); + + if (ret < 0) { + gossip_debug(GOSSIP_NAME_DEBUG, + "%s: failed with error code %d\n", + __func__, ret); + goto out; + } + + inode = pvfs2_new_inode(dir->i_sb, dir, S_IFREG | mode, 0, + &new_op->downcall.resp.create.refn); + if (IS_ERR(inode)) { + gossip_err("*** Failed to allocate pvfs2 file inode\n"); + ret = PTR_ERR(inode); + goto out; + } + + gossip_debug(GOSSIP_NAME_DEBUG, + "Assigned file inode new number of %pU\n", + get_khandle_from_ino(inode)); + + d_instantiate(dentry, inode); + unlock_new_inode(inode); + + gossip_debug(GOSSIP_NAME_DEBUG, + "Inode (Regular File) %pU -> %s\n", + get_khandle_from_ino(inode), + dentry->d_name.name); + + SetMtimeFlag(parent); + dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb); + mark_inode_dirty_sync(dir); + ret = 0; +out: + op_release(new_op); + gossip_debug(GOSSIP_NAME_DEBUG, "%s: returning %d\n", __func__, ret); + return ret; +} + +/* + * Attempt to resolve an object name (dentry->d_name), parent handle, and + * fsid into a handle for the object. + */ +static struct dentry *pvfs2_lookup(struct inode *dir, struct dentry *dentry, + unsigned int flags) +{ + struct pvfs2_inode_s *parent = PVFS2_I(dir); + struct pvfs2_kernel_op_s *new_op; + struct inode *inode; + struct dentry *res; + int ret = -EINVAL; + + /* + * in theory we could skip a lookup here (if the intent is to + * create) in order to avoid a potentially failed lookup, but + * leaving it in can skip a valid lookup and try to create a file + * that already exists (e.g. the vfs already handles checking for + * -EEXIST on O_EXCL opens, which is broken if we skip this lookup + * in the create path) + */ + gossip_debug(GOSSIP_NAME_DEBUG, "%s called on %s\n", + __func__, dentry->d_name.name); + + if (dentry->d_name.len > (PVFS2_NAME_LEN - 1)) + return ERR_PTR(-ENAMETOOLONG); + + new_op = op_alloc(PVFS2_VFS_OP_LOOKUP); + if (!new_op) + return ERR_PTR(-ENOMEM); + + new_op->upcall.req.lookup.sym_follow = flags & LOOKUP_FOLLOW; + + gossip_debug(GOSSIP_NAME_DEBUG, "%s:%s:%d using parent %pU\n", + __FILE__, + __func__, + __LINE__, + &parent->refn.khandle); + new_op->upcall.req.lookup.parent_refn = parent->refn; + + strncpy(new_op->upcall.req.lookup.d_name, dentry->d_name.name, + PVFS2_NAME_LEN); + + gossip_debug(GOSSIP_NAME_DEBUG, + "%s: doing lookup on %s under %pU,%d (follow=%s)\n", + __func__, + new_op->upcall.req.lookup.d_name, + &new_op->upcall.req.lookup.parent_refn.khandle, + new_op->upcall.req.lookup.parent_refn.fs_id, + ((new_op->upcall.req.lookup.sym_follow == + PVFS2_LOOKUP_LINK_FOLLOW) ? "yes" : "no")); + + ret = service_operation(new_op, __func__, get_interruptible_flag(dir)); + + gossip_debug(GOSSIP_NAME_DEBUG, + "Lookup Got %pU, fsid %d (ret=%d)\n", + &new_op->downcall.resp.lookup.refn.khandle, + new_op->downcall.resp.lookup.refn.fs_id, + ret); + + if (ret < 0) { + if (ret == -ENOENT) { + /* + * if no inode was found, add a negative dentry to + * dcache anyway; if we don't, we don't hold expected + * lookup semantics and we most noticeably break + * during directory renames. + * + * however, if the operation failed or exited, do not + * add the dentry (e.g. in the case that a touch is + * issued on a file that already exists that was + * interrupted during this lookup -- no need to add + * another negative dentry for an existing file) + */ + + gossip_debug(GOSSIP_NAME_DEBUG, + "pvfs2_lookup: Adding *negative* dentry " + "%p for %s\n", + dentry, + dentry->d_name.name); + + d_add(dentry, NULL); + res = NULL; + goto out; + } + + /* must be a non-recoverable error */ + res = ERR_PTR(ret); + goto out; + } + + inode = pvfs2_iget(dir->i_sb, &new_op->downcall.resp.lookup.refn); + if (IS_ERR(inode)) { + gossip_debug(GOSSIP_NAME_DEBUG, + "error %ld from iget\n", PTR_ERR(inode)); + res = ERR_CAST(inode); + goto out; + } + + gossip_debug(GOSSIP_NAME_DEBUG, + "%s:%s:%d " + "Found good inode [%lu] with count [%d]\n", + __FILE__, + __func__, + __LINE__, + inode->i_ino, + (int)atomic_read(&inode->i_count)); + + /* update dentry/inode pair into dcache */ + res = d_splice_alias(inode, dentry); + + gossip_debug(GOSSIP_NAME_DEBUG, + "Lookup success (inode ct = %d)\n", + (int)atomic_read(&inode->i_count)); +out: + op_release(new_op); + return res; +} + +/* return 0 on success; non-zero otherwise */ +static int pvfs2_unlink(struct inode *dir, struct dentry *dentry) +{ + struct inode *inode = dentry->d_inode; + struct pvfs2_inode_s *parent = PVFS2_I(dir); + struct pvfs2_kernel_op_s *new_op; + int ret; + + gossip_debug(GOSSIP_NAME_DEBUG, + "%s: called on %s\n" + " (inode %pU): Parent is %pU | fs_id %d\n", + __func__, + dentry->d_name.name, + get_khandle_from_ino(inode), + &parent->refn.khandle, + parent->refn.fs_id); + + new_op = op_alloc(PVFS2_VFS_OP_REMOVE); + if (!new_op) + return -ENOMEM; + + new_op->upcall.req.remove.parent_refn = parent->refn; + strncpy(new_op->upcall.req.remove.d_name, dentry->d_name.name, + PVFS2_NAME_LEN); + + ret = service_operation(new_op, "pvfs2_unlink", + get_interruptible_flag(inode)); + + /* when request is serviced properly, free req op struct */ + op_release(new_op); + + if (!ret) { + drop_nlink(inode); + + SetMtimeFlag(parent); + dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb); + mark_inode_dirty_sync(dir); + } + return ret; +} + +/* + * PVFS2 does not support hard links. + */ +static int pvfs2_link(struct dentry *old_dentry, + struct inode *dir, + struct dentry *dentry) +{ + return -EPERM; +} + +/* + * PVFS2 does not support special files. + */ +static int pvfs2_mknod(struct inode *dir, + struct dentry *dentry, + umode_t mode, + dev_t rdev) +{ + return -EPERM; +} + +static int pvfs2_symlink(struct inode *dir, + struct dentry *dentry, + const char *symname) +{ + struct pvfs2_inode_s *parent = PVFS2_I(dir); + struct pvfs2_kernel_op_s *new_op; + struct inode *inode; + int mode = 755; + int ret; + + gossip_debug(GOSSIP_NAME_DEBUG, "%s: called\n", __func__); + + if (!symname) + return -EINVAL; + + new_op = op_alloc(PVFS2_VFS_OP_SYMLINK); + if (!new_op) + return -ENOMEM; + + new_op->upcall.req.sym.parent_refn = parent->refn; + + fill_default_sys_attrs(new_op->upcall.req.sym.attributes, + PVFS_TYPE_SYMLINK, + mode); + + strncpy(new_op->upcall.req.sym.entry_name, + dentry->d_name.name, + PVFS2_NAME_LEN); + strncpy(new_op->upcall.req.sym.target, symname, PVFS2_NAME_LEN); + + ret = service_operation(new_op, __func__, get_interruptible_flag(dir)); + + gossip_debug(GOSSIP_NAME_DEBUG, + "Symlink Got PVFS2 handle %pU on fsid %d (ret=%d)\n", + &new_op->downcall.resp.sym.refn.khandle, + new_op->downcall.resp.sym.refn.fs_id, ret); + + if (ret < 0) { + gossip_debug(GOSSIP_NAME_DEBUG, + "%s: failed with error code %d\n", + __func__, ret); + goto out; + } + + inode = pvfs2_new_inode(dir->i_sb, dir, S_IFLNK | mode, 0, + &new_op->downcall.resp.sym.refn); + if (IS_ERR(inode)) { + gossip_err + ("*** Failed to allocate pvfs2 symlink inode\n"); + ret = PTR_ERR(inode); + goto out; + } + + gossip_debug(GOSSIP_NAME_DEBUG, + "Assigned symlink inode new number of %pU\n", + get_khandle_from_ino(inode)); + + d_instantiate(dentry, inode); + unlock_new_inode(inode); + + gossip_debug(GOSSIP_NAME_DEBUG, + "Inode (Symlink) %pU -> %s\n", + get_khandle_from_ino(inode), + dentry->d_name.name); + + SetMtimeFlag(parent); + dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb); + mark_inode_dirty_sync(dir); + ret = 0; +out: + op_release(new_op); + return ret; +} + +static int pvfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) +{ + struct pvfs2_inode_s *parent = PVFS2_I(dir); + struct pvfs2_kernel_op_s *new_op; + struct inode *inode; + int ret; + + new_op = op_alloc(PVFS2_VFS_OP_MKDIR); + if (!new_op) + return -ENOMEM; + + new_op->upcall.req.mkdir.parent_refn = parent->refn; + + fill_default_sys_attrs(new_op->upcall.req.mkdir.attributes, + PVFS_TYPE_DIRECTORY, mode); + + strncpy(new_op->upcall.req.mkdir.d_name, + dentry->d_name.name, PVFS2_NAME_LEN); + + ret = service_operation(new_op, __func__, get_interruptible_flag(dir)); + + gossip_debug(GOSSIP_NAME_DEBUG, + "Mkdir Got PVFS2 handle %pU on fsid %d\n", + &new_op->downcall.resp.mkdir.refn.khandle, + new_op->downcall.resp.mkdir.refn.fs_id); + + if (ret < 0) { + gossip_debug(GOSSIP_NAME_DEBUG, + "%s: failed with error code %d\n", + __func__, ret); + goto out; + } + + inode = pvfs2_new_inode(dir->i_sb, dir, S_IFDIR | mode, 0, + &new_op->downcall.resp.mkdir.refn); + if (IS_ERR(inode)) { + gossip_err("*** Failed to allocate pvfs2 dir inode\n"); + ret = PTR_ERR(inode); + goto out; + } + + gossip_debug(GOSSIP_NAME_DEBUG, + "Assigned dir inode new number of %pU\n", + get_khandle_from_ino(inode)); + + d_instantiate(dentry, inode); + unlock_new_inode(inode); + + gossip_debug(GOSSIP_NAME_DEBUG, + "Inode (Directory) %pU -> %s\n", + get_khandle_from_ino(inode), + dentry->d_name.name); + + /* + * NOTE: we have no good way to keep nlink consistent for directories + * across clients; keep constant at 1. + */ + SetMtimeFlag(parent); + dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb); + mark_inode_dirty_sync(dir); +out: + op_release(new_op); + return ret; +} + +static int pvfs2_rename(struct inode *old_dir, + struct dentry *old_dentry, + struct inode *new_dir, + struct dentry *new_dentry) +{ + struct pvfs2_kernel_op_s *new_op; + int ret; + + gossip_debug(GOSSIP_NAME_DEBUG, + "pvfs2_rename: called (%s/%s => %s/%s) ct=%d\n", + old_dentry->d_parent->d_name.name, + old_dentry->d_name.name, + new_dentry->d_parent->d_name.name, + new_dentry->d_name.name, + d_count(new_dentry)); + + new_op = op_alloc(PVFS2_VFS_OP_RENAME); + if (!new_op) + return -EINVAL; + + new_op->upcall.req.rename.old_parent_refn = PVFS2_I(old_dir)->refn; + new_op->upcall.req.rename.new_parent_refn = PVFS2_I(new_dir)->refn; + + strncpy(new_op->upcall.req.rename.d_old_name, + old_dentry->d_name.name, + PVFS2_NAME_LEN); + strncpy(new_op->upcall.req.rename.d_new_name, + new_dentry->d_name.name, + PVFS2_NAME_LEN); + + ret = service_operation(new_op, + "pvfs2_rename", + get_interruptible_flag(old_dentry->d_inode)); + + gossip_debug(GOSSIP_NAME_DEBUG, + "pvfs2_rename: got downcall status %d\n", + ret); + + if (new_dentry->d_inode) + new_dentry->d_inode->i_ctime = CURRENT_TIME; + + op_release(new_op); + return ret; +} + +/* PVFS2 implementation of VFS inode operations for directories */ +struct inode_operations pvfs2_dir_inode_operations = { + .lookup = pvfs2_lookup, + .get_acl = pvfs2_get_acl, + .set_acl = pvfs2_set_acl, + .create = pvfs2_create, + .link = pvfs2_link, + .unlink = pvfs2_unlink, + .symlink = pvfs2_symlink, + .mkdir = pvfs2_mkdir, + .rmdir = pvfs2_unlink, + .mknod = pvfs2_mknod, + .rename = pvfs2_rename, + .setattr = pvfs2_setattr, + .getattr = pvfs2_getattr, + .setxattr = generic_setxattr, + .getxattr = generic_getxattr, + .removexattr = generic_removexattr, + .listxattr = pvfs2_listxattr, +}; diff --git a/fs/orangefs/protocol.h b/fs/orangefs/protocol.h new file mode 100644 index 000000000000..85f611fe0536 --- /dev/null +++ b/fs/orangefs/protocol.h @@ -0,0 +1,443 @@ +#include +#include +#include +#include + +extern struct client_debug_mask *cdm_array; +extern char *debug_help_string; +extern int help_string_initialized; +extern struct dentry *debug_dir; +extern struct dentry *help_file_dentry; +extern struct dentry *client_debug_dentry; +extern const struct file_operations debug_help_fops; +extern int client_all_index; +extern int client_verbose_index; +extern int cdm_element_count; +#define DEBUG_HELP_STRING_SIZE 4096 +#define HELP_STRING_UNINITIALIZED \ + "Client Debug Keywords are unknown until the first time\n" \ + "the client is started after boot.\n" +#define ORANGEFS_KMOD_DEBUG_HELP_FILE "debug-help" +#define ORANGEFS_KMOD_DEBUG_FILE "kernel-debug" +#define ORANGEFS_CLIENT_DEBUG_FILE "client-debug" +#define PVFS2_VERBOSE "verbose" +#define PVFS2_ALL "all" + +/* pvfs2-config.h ***********************************************************/ +#define PVFS2_VERSION_MAJOR 2 +#define PVFS2_VERSION_MINOR 9 +#define PVFS2_VERSION_SUB 0 + +/* khandle stuff ***********************************************************/ + +/* + * The 2.9 core will put 64 bit handles in here like this: + * 1234 0000 0000 5678 + * The 3.0 and beyond cores will put 128 bit handles in here like this: + * 1234 5678 90AB CDEF + * The kernel module will always use the first four bytes and + * the last four bytes as an inum. + */ +struct pvfs2_khandle { + unsigned char u[16]; +} __aligned(8); + +/* + * kernel version of an object ref. + */ +struct pvfs2_object_kref { + struct pvfs2_khandle khandle; + __s32 fs_id; + __s32 __pad1; +}; + +/* + * compare 2 khandles assumes little endian thus from large address to + * small address + */ +static inline int PVFS_khandle_cmp(const struct pvfs2_khandle *kh1, + const struct pvfs2_khandle *kh2) +{ + int i; + + for (i = 15; i >= 0; i--) { + if (kh1->u[i] > kh2->u[i]) + return 1; + if (kh1->u[i] < kh2->u[i]) + return -1; + } + + return 0; +} + +static inline void PVFS_khandle_to(const struct pvfs2_khandle *kh, + void *p, int size) +{ + + memset(p, 0, size); + memcpy(p, kh->u, 16); + +} + +static inline void PVFS_khandle_from(struct pvfs2_khandle *kh, + void *p, int size) +{ + memset(kh, 0, 16); + memcpy(kh->u, p, 16); + +} + +/* pvfs2-types.h ************************************************************/ +typedef __u32 PVFS_uid; +typedef __u32 PVFS_gid; +typedef __s32 PVFS_fs_id; +typedef __u32 PVFS_permissions; +typedef __u64 PVFS_time; +typedef __s64 PVFS_size; +typedef __u64 PVFS_flags; +typedef __u64 PVFS_ds_position; +typedef __s32 PVFS_error; +typedef __s64 PVFS_offset; + +#define PVFS2_SUPER_MAGIC 0x20030528 + +/* + * PVFS2 error codes are a signed 32-bit integer. Error codes are negative, but + * the sign is stripped before decoding. + */ + +/* Bit 31 is not used since it is the sign. */ + +/* + * Bit 30 specifies that this is a PVFS2 error. A PVFS2 error is either an + * encoded errno value or a PVFS2 protocol error. + */ +#define PVFS_ERROR_BIT (1 << 30) + +/* + * Bit 29 specifies that this is a PVFS2 protocol error and not an encoded + * errno value. + */ +#define PVFS_NON_ERRNO_ERROR_BIT (1 << 29) + +/* + * Bits 9, 8, and 7 specify the error class, which encodes the section of + * server code the error originated in for logging purposes. It is not used + * in the kernel except to be masked out. + */ +#define PVFS_ERROR_CLASS_BITS 0x380 + +/* Bits 6 - 0 are reserved for the actual error code. */ +#define PVFS_ERROR_NUMBER_BITS 0x7f + +/* Encoded errno values are decoded by PINT_errno_mapping in pvfs2-utils.c. */ + +/* Our own PVFS2 protocol error codes. */ +#define PVFS_ECANCEL (1|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) +#define PVFS_EDEVINIT (2|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) +#define PVFS_EDETAIL (3|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) +#define PVFS_EHOSTNTFD (4|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) +#define PVFS_EADDRNTFD (5|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) +#define PVFS_ENORECVR (6|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) +#define PVFS_ETRYAGAIN (7|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) +#define PVFS_ENOTPVFS (8|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) +#define PVFS_ESECURITY (9|PVFS_NON_ERRNO_ERROR_BIT|PVFS_ERROR_BIT) + +/* permission bits */ +#define PVFS_O_EXECUTE (1 << 0) +#define PVFS_O_WRITE (1 << 1) +#define PVFS_O_READ (1 << 2) +#define PVFS_G_EXECUTE (1 << 3) +#define PVFS_G_WRITE (1 << 4) +#define PVFS_G_READ (1 << 5) +#define PVFS_U_EXECUTE (1 << 6) +#define PVFS_U_WRITE (1 << 7) +#define PVFS_U_READ (1 << 8) +/* no PVFS_U_VTX (sticky bit) */ +#define PVFS_G_SGID (1 << 10) +#define PVFS_U_SUID (1 << 11) + +/* definition taken from stdint.h */ +#define INT32_MAX (2147483647) +#define PVFS_ITERATE_START (INT32_MAX - 1) +#define PVFS_ITERATE_END (INT32_MAX - 2) +#define PVFS_ITERATE_NEXT (INT32_MAX - 3) +#define PVFS_READDIR_START PVFS_ITERATE_START +#define PVFS_READDIR_END PVFS_ITERATE_END +#define PVFS_IMMUTABLE_FL FS_IMMUTABLE_FL +#define PVFS_APPEND_FL FS_APPEND_FL +#define PVFS_NOATIME_FL FS_NOATIME_FL +#define PVFS_MIRROR_FL 0x01000000ULL +#define PVFS_O_EXECUTE (1 << 0) +#define PVFS_FS_ID_NULL ((__s32)0) +#define PVFS_ATTR_SYS_UID (1 << 0) +#define PVFS_ATTR_SYS_GID (1 << 1) +#define PVFS_ATTR_SYS_PERM (1 << 2) +#define PVFS_ATTR_SYS_ATIME (1 << 3) +#define PVFS_ATTR_SYS_CTIME (1 << 4) +#define PVFS_ATTR_SYS_MTIME (1 << 5) +#define PVFS_ATTR_SYS_TYPE (1 << 6) +#define PVFS_ATTR_SYS_ATIME_SET (1 << 7) +#define PVFS_ATTR_SYS_MTIME_SET (1 << 8) +#define PVFS_ATTR_SYS_SIZE (1 << 20) +#define PVFS_ATTR_SYS_LNK_TARGET (1 << 24) +#define PVFS_ATTR_SYS_DFILE_COUNT (1 << 25) +#define PVFS_ATTR_SYS_DIRENT_COUNT (1 << 26) +#define PVFS_ATTR_SYS_BLKSIZE (1 << 28) +#define PVFS_ATTR_SYS_MIRROR_COPIES_COUNT (1 << 29) +#define PVFS_ATTR_SYS_COMMON_ALL \ + (PVFS_ATTR_SYS_UID | \ + PVFS_ATTR_SYS_GID | \ + PVFS_ATTR_SYS_PERM | \ + PVFS_ATTR_SYS_ATIME | \ + PVFS_ATTR_SYS_CTIME | \ + PVFS_ATTR_SYS_MTIME | \ + PVFS_ATTR_SYS_TYPE) + +#define PVFS_ATTR_SYS_ALL_SETABLE \ +(PVFS_ATTR_SYS_COMMON_ALL-PVFS_ATTR_SYS_TYPE) + +#define PVFS_ATTR_SYS_ALL_NOHINT \ + (PVFS_ATTR_SYS_COMMON_ALL | \ + PVFS_ATTR_SYS_SIZE | \ + PVFS_ATTR_SYS_LNK_TARGET | \ + PVFS_ATTR_SYS_DFILE_COUNT | \ + PVFS_ATTR_SYS_MIRROR_COPIES_COUNT | \ + PVFS_ATTR_SYS_DIRENT_COUNT | \ + PVFS_ATTR_SYS_BLKSIZE) +#define PVFS_XATTR_REPLACE 0x2 +#define PVFS_XATTR_CREATE 0x1 +#define PVFS_MAX_SERVER_ADDR_LEN 256 +#define PVFS_NAME_MAX 256 +/* + * max extended attribute name len as imposed by the VFS and exploited for the + * upcall request types. + * NOTE: Please retain them as multiples of 8 even if you wish to change them + * This is *NECESSARY* for supporting 32 bit user-space binaries on a 64-bit + * kernel. Due to implementation within DBPF, this really needs to be + * PVFS_NAME_MAX, which it was the same value as, but no reason to let it + * break if that changes in the future. + */ +#define PVFS_MAX_XATTR_NAMELEN PVFS_NAME_MAX /* Not the same as + * XATTR_NAME_MAX defined + * by + */ +#define PVFS_MAX_XATTR_VALUELEN 8192 /* Not the same as XATTR_SIZE_MAX + * defined by + */ +#define PVFS_MAX_XATTR_LISTLEN 16 /* Not the same as XATTR_LIST_MAX + * defined by + */ +/* + * PVFS I/O operation types, used in both system and server interfaces. + */ +enum PVFS_io_type { + PVFS_IO_READ = 1, + PVFS_IO_WRITE = 2 +}; + +/* + * If this enum is modified the server parameters related to the precreate pool + * batch and low threshold sizes may need to be modified to reflect this + * change. + */ +enum pvfs2_ds_type { + PVFS_TYPE_NONE = 0, + PVFS_TYPE_METAFILE = (1 << 0), + PVFS_TYPE_DATAFILE = (1 << 1), + PVFS_TYPE_DIRECTORY = (1 << 2), + PVFS_TYPE_SYMLINK = (1 << 3), + PVFS_TYPE_DIRDATA = (1 << 4), + PVFS_TYPE_INTERNAL = (1 << 5) /* for the server's private use */ +}; + +/* + * PVFS_certificate simply stores a buffer with the buffer size. + * The buffer can be converted to an OpenSSL X509 struct for use. + */ +struct PVFS_certificate { + __u32 buf_size; + unsigned char *buf; +}; + +/* + * A credential identifies a user and is signed by the client/user + * private key. + */ +struct PVFS_credential { + __u32 userid; /* user id */ + __u32 num_groups; /* length of group_array */ + __u32 *group_array; /* groups for which the user is a member */ + char *issuer; /* alias of the issuing server */ + __u64 timeout; /* seconds after epoch to time out */ + __u32 sig_size; /* length of the signature in bytes */ + unsigned char *signature; /* digital signature */ + struct PVFS_certificate certificate; /* user certificate buffer */ +}; +#define extra_size_PVFS_credential (PVFS_REQ_LIMIT_GROUPS * \ + sizeof(__u32) + \ + PVFS_REQ_LIMIT_ISSUER + \ + PVFS_REQ_LIMIT_SIGNATURE + \ + extra_size_PVFS_certificate) + +/* This structure is used by the VFS-client interaction alone */ +struct PVFS_keyval_pair { + char key[PVFS_MAX_XATTR_NAMELEN]; + __s32 key_sz; /* __s32 for portable, fixed-size structures */ + __s32 val_sz; + char val[PVFS_MAX_XATTR_VALUELEN]; +}; + +/* pvfs2-sysint.h ***********************************************************/ +/* Describes attributes for a file, directory, or symlink. */ +struct PVFS_sys_attr_s { + __u32 owner; + __u32 group; + __u32 perms; + __u64 atime; + __u64 mtime; + __u64 ctime; + __s64 size; + + /* NOTE: caller must free if valid */ + char *link_target; + + /* Changed to __s32 so that size of structure does not change */ + __s32 dfile_count; + + /* Changed to __s32 so that size of structure does not change */ + __s32 distr_dir_servers_initial; + + /* Changed to __s32 so that size of structure does not change */ + __s32 distr_dir_servers_max; + + /* Changed to __s32 so that size of structure does not change */ + __s32 distr_dir_split_size; + + __u32 mirror_copies_count; + + /* NOTE: caller must free if valid */ + char *dist_name; + + /* NOTE: caller must free if valid */ + char *dist_params; + + __s64 dirent_count; + enum pvfs2_ds_type objtype; + __u64 flags; + __u32 mask; + __s64 blksize; +}; + +#define PVFS2_LOOKUP_LINK_NO_FOLLOW 0 +#define PVFS2_LOOKUP_LINK_FOLLOW 1 + +/* pint-dev.h ***************************************************************/ + +/* parameter structure used in PVFS_DEV_DEBUG ioctl command */ +struct dev_mask_info_s { + enum { + KERNEL_MASK, + CLIENT_MASK, + } mask_type; + __u64 mask_value; +}; + +struct dev_mask2_info_s { + __u64 mask1_value; + __u64 mask2_value; +}; + +/* pvfs2-util.h *************************************************************/ +__s32 PVFS_util_translate_mode(int mode); + +/* pvfs2-debug.h ************************************************************/ +#include "pvfs2-debug.h" + +/* pvfs2-internal.h *********************************************************/ +#define llu(x) (unsigned long long)(x) +#define lld(x) (long long)(x) + +/* pint-dev-shared.h ********************************************************/ +#define PVFS_DEV_MAGIC 'k' + +#define PVFS2_READDIR_DEFAULT_DESC_COUNT 5 + +#define DEV_GET_MAGIC 0x1 +#define DEV_GET_MAX_UPSIZE 0x2 +#define DEV_GET_MAX_DOWNSIZE 0x3 +#define DEV_MAP 0x4 +#define DEV_REMOUNT_ALL 0x5 +#define DEV_DEBUG 0x6 +#define DEV_UPSTREAM 0x7 +#define DEV_CLIENT_MASK 0x8 +#define DEV_CLIENT_STRING 0x9 +#define DEV_MAX_NR 0xa + +/* supported ioctls, codes are with respect to user-space */ +enum { + PVFS_DEV_GET_MAGIC = _IOW(PVFS_DEV_MAGIC, DEV_GET_MAGIC, __s32), + PVFS_DEV_GET_MAX_UPSIZE = + _IOW(PVFS_DEV_MAGIC, DEV_GET_MAX_UPSIZE, __s32), + PVFS_DEV_GET_MAX_DOWNSIZE = + _IOW(PVFS_DEV_MAGIC, DEV_GET_MAX_DOWNSIZE, __s32), + PVFS_DEV_MAP = _IO(PVFS_DEV_MAGIC, DEV_MAP), + PVFS_DEV_REMOUNT_ALL = _IO(PVFS_DEV_MAGIC, DEV_REMOUNT_ALL), + PVFS_DEV_DEBUG = _IOR(PVFS_DEV_MAGIC, DEV_DEBUG, __s32), + PVFS_DEV_UPSTREAM = _IOW(PVFS_DEV_MAGIC, DEV_UPSTREAM, int), + PVFS_DEV_CLIENT_MASK = _IOW(PVFS_DEV_MAGIC, + DEV_CLIENT_MASK, + struct dev_mask2_info_s), + PVFS_DEV_CLIENT_STRING = _IOW(PVFS_DEV_MAGIC, + DEV_CLIENT_STRING, + char *), + PVFS_DEV_MAXNR = DEV_MAX_NR, +}; + +/* + * version number for use in communicating between kernel space and user + * space. Zero signifies the upstream version of the kernel module. + */ +#define PVFS_KERNEL_PROTO_VERSION 0 + +/* + * describes memory regions to map in the PVFS_DEV_MAP ioctl. + * NOTE: See devpvfs2-req.c for 32 bit compat structure. + * Since this structure has a variable-sized layout that is different + * on 32 and 64 bit platforms, we need to normalize to a 64 bit layout + * on such systems before servicing ioctl calls from user-space binaries + * that may be 32 bit! + */ +struct PVFS_dev_map_desc { + void *ptr; + __s32 total_size; + __s32 size; + __s32 count; +}; + +/* gossip.h *****************************************************************/ + +#ifdef GOSSIP_DISABLE_DEBUG +#define gossip_debug(mask, format, f...) do {} while (0) +#else +extern __u64 gossip_debug_mask; +extern struct client_debug_mask client_debug_mask; + +/* try to avoid function call overhead by checking masks in macro */ +#define gossip_debug(mask, format, f...) \ +do { \ + if (gossip_debug_mask & mask) \ + printk(format, ##f); \ +} while (0) +#endif /* GOSSIP_DISABLE_DEBUG */ + +/* do file and line number printouts w/ the GNU preprocessor */ +#define gossip_ldebug(mask, format, f...) \ + gossip_debug(mask, "%s: " format, __func__, ##f) + +#define gossip_err printk +#define gossip_lerr(format, f...) \ + gossip_err("%s line %d: " format, \ + __FILE__, \ + __LINE__, \ + ##f) diff --git a/fs/orangefs/pvfs2-bufmap.c b/fs/orangefs/pvfs2-bufmap.c new file mode 100644 index 000000000000..9d0392a3e824 --- /dev/null +++ b/fs/orangefs/pvfs2-bufmap.c @@ -0,0 +1,566 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" + +DECLARE_WAIT_QUEUE_HEAD(pvfs2_bufmap_init_waitq); + +static struct pvfs2_bufmap { + atomic_t refcnt; + + int desc_size; + int desc_shift; + int desc_count; + int total_size; + int page_count; + + struct page **page_array; + struct pvfs_bufmap_desc *desc_array; + + /* array to track usage of buffer descriptors */ + int *buffer_index_array; + spinlock_t buffer_index_lock; + + /* array to track usage of buffer descriptors for readdir */ + int readdir_index_array[PVFS2_READDIR_DEFAULT_DESC_COUNT]; + spinlock_t readdir_index_lock; +} *__pvfs2_bufmap; + +static DEFINE_SPINLOCK(pvfs2_bufmap_lock); + +static void +pvfs2_bufmap_unmap(struct pvfs2_bufmap *bufmap) +{ + int i; + + for (i = 0; i < bufmap->page_count; i++) + page_cache_release(bufmap->page_array[i]); +} + +static void +pvfs2_bufmap_free(struct pvfs2_bufmap *bufmap) +{ + kfree(bufmap->page_array); + kfree(bufmap->desc_array); + kfree(bufmap->buffer_index_array); + kfree(bufmap); +} + +struct pvfs2_bufmap *pvfs2_bufmap_ref(void) +{ + struct pvfs2_bufmap *bufmap = NULL; + + spin_lock(&pvfs2_bufmap_lock); + if (__pvfs2_bufmap) { + bufmap = __pvfs2_bufmap; + atomic_inc(&bufmap->refcnt); + } + spin_unlock(&pvfs2_bufmap_lock); + return bufmap; +} + +void pvfs2_bufmap_unref(struct pvfs2_bufmap *bufmap) +{ + if (atomic_dec_and_lock(&bufmap->refcnt, &pvfs2_bufmap_lock)) { + __pvfs2_bufmap = NULL; + spin_unlock(&pvfs2_bufmap_lock); + + pvfs2_bufmap_unmap(bufmap); + pvfs2_bufmap_free(bufmap); + } +} + +inline int pvfs_bufmap_size_query(void) +{ + struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref(); + int size = bufmap ? bufmap->desc_size : 0; + + pvfs2_bufmap_unref(bufmap); + return size; +} + +inline int pvfs_bufmap_shift_query(void) +{ + struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref(); + int shift = bufmap ? bufmap->desc_shift : 0; + + pvfs2_bufmap_unref(bufmap); + return shift; +} + +static DECLARE_WAIT_QUEUE_HEAD(bufmap_waitq); +static DECLARE_WAIT_QUEUE_HEAD(readdir_waitq); + +/* + * get_bufmap_init + * + * If bufmap_init is 1, then the shared memory system, including the + * buffer_index_array, is available. Otherwise, it is not. + * + * returns the value of bufmap_init + */ +int get_bufmap_init(void) +{ + return __pvfs2_bufmap ? 1 : 0; +} + + +static struct pvfs2_bufmap * +pvfs2_bufmap_alloc(struct PVFS_dev_map_desc *user_desc) +{ + struct pvfs2_bufmap *bufmap; + + bufmap = kzalloc(sizeof(*bufmap), GFP_KERNEL); + if (!bufmap) + goto out; + + atomic_set(&bufmap->refcnt, 1); + bufmap->total_size = user_desc->total_size; + bufmap->desc_count = user_desc->count; + bufmap->desc_size = user_desc->size; + bufmap->desc_shift = ilog2(bufmap->desc_size); + + spin_lock_init(&bufmap->buffer_index_lock); + bufmap->buffer_index_array = + kcalloc(bufmap->desc_count, sizeof(int), GFP_KERNEL); + if (!bufmap->buffer_index_array) { + gossip_err("pvfs2: could not allocate %d buffer indices\n", + bufmap->desc_count); + goto out_free_bufmap; + } + spin_lock_init(&bufmap->readdir_index_lock); + + bufmap->desc_array = + kcalloc(bufmap->desc_count, sizeof(struct pvfs_bufmap_desc), + GFP_KERNEL); + if (!bufmap->desc_array) { + gossip_err("pvfs2: could not allocate %d descriptors\n", + bufmap->desc_count); + goto out_free_index_array; + } + + bufmap->page_count = bufmap->total_size / PAGE_SIZE; + + /* allocate storage to track our page mappings */ + bufmap->page_array = + kcalloc(bufmap->page_count, sizeof(struct page *), GFP_KERNEL); + if (!bufmap->page_array) + goto out_free_desc_array; + + return bufmap; + +out_free_desc_array: + kfree(bufmap->desc_array); +out_free_index_array: + kfree(bufmap->buffer_index_array); +out_free_bufmap: + kfree(bufmap); +out: + return NULL; +} + +static int +pvfs2_bufmap_map(struct pvfs2_bufmap *bufmap, + struct PVFS_dev_map_desc *user_desc) +{ + int pages_per_desc = bufmap->desc_size / PAGE_SIZE; + int offset = 0, ret, i; + + /* map the pages */ + down_write(¤t->mm->mmap_sem); + ret = get_user_pages(current, + current->mm, + (unsigned long)user_desc->ptr, + bufmap->page_count, + 1, + 0, + bufmap->page_array, + NULL); + up_write(¤t->mm->mmap_sem); + + if (ret < 0) + return ret; + + if (ret != bufmap->page_count) { + gossip_err("pvfs2 error: asked for %d pages, only got %d.\n", + bufmap->page_count, ret); + + for (i = 0; i < ret; i++) { + SetPageError(bufmap->page_array[i]); + page_cache_release(bufmap->page_array[i]); + } + return -ENOMEM; + } + + /* + * ideally we want to get kernel space pointers for each page, but + * we can't kmap that many pages at once if highmem is being used. + * so instead, we just kmap/kunmap the page address each time the + * kaddr is needed. + */ + for (i = 0; i < bufmap->page_count; i++) + flush_dcache_page(bufmap->page_array[i]); + + /* build a list of available descriptors */ + for (offset = 0, i = 0; i < bufmap->desc_count; i++) { + bufmap->desc_array[i].page_array = &bufmap->page_array[offset]; + bufmap->desc_array[i].array_count = pages_per_desc; + bufmap->desc_array[i].uaddr = + (user_desc->ptr + (i * pages_per_desc * PAGE_SIZE)); + offset += pages_per_desc; + } + + return 0; +} + +/* + * pvfs_bufmap_initialize() + * + * initializes the mapped buffer interface + * + * returns 0 on success, -errno on failure + */ +int pvfs_bufmap_initialize(struct PVFS_dev_map_desc *user_desc) +{ + struct pvfs2_bufmap *bufmap; + int ret = -EINVAL; + + gossip_debug(GOSSIP_BUFMAP_DEBUG, + "pvfs_bufmap_initialize: called (ptr (" + "%p) sz (%d) cnt(%d).\n", + user_desc->ptr, + user_desc->size, + user_desc->count); + + /* + * sanity check alignment and size of buffer that caller wants to + * work with + */ + if (PAGE_ALIGN((unsigned long)user_desc->ptr) != + (unsigned long)user_desc->ptr) { + gossip_err("pvfs2 error: memory alignment (front). %p\n", + user_desc->ptr); + goto out; + } + + if (PAGE_ALIGN(((unsigned long)user_desc->ptr + user_desc->total_size)) + != (unsigned long)(user_desc->ptr + user_desc->total_size)) { + gossip_err("pvfs2 error: memory alignment (back).(%p + %d)\n", + user_desc->ptr, + user_desc->total_size); + goto out; + } + + if (user_desc->total_size != (user_desc->size * user_desc->count)) { + gossip_err("pvfs2 error: user provided an oddly sized buffer: (%d, %d, %d)\n", + user_desc->total_size, + user_desc->size, + user_desc->count); + goto out; + } + + if ((user_desc->size % PAGE_SIZE) != 0) { + gossip_err("pvfs2 error: bufmap size not page size divisible (%d).\n", + user_desc->size); + goto out; + } + + ret = -ENOMEM; + bufmap = pvfs2_bufmap_alloc(user_desc); + if (!bufmap) + goto out; + + ret = pvfs2_bufmap_map(bufmap, user_desc); + if (ret) + goto out_free_bufmap; + + + spin_lock(&pvfs2_bufmap_lock); + if (__pvfs2_bufmap) { + spin_unlock(&pvfs2_bufmap_lock); + gossip_err("pvfs2: error: bufmap already initialized.\n"); + ret = -EALREADY; + goto out_unmap_bufmap; + } + __pvfs2_bufmap = bufmap; + spin_unlock(&pvfs2_bufmap_lock); + + /* + * If there are operations in pvfs2_bufmap_init_waitq, wake them up. + * This scenario occurs when the client-core is restarted and I/O + * requests in the in-progress or waiting tables are restarted. I/O + * requests cannot be restarted until the shared memory system is + * completely re-initialized, so we put the I/O requests in this + * waitq until initialization has completed. NOTE: the I/O requests + * are also on a timer, so they don't wait forever just in case the + * client-core doesn't come back up. + */ + wake_up_interruptible(&pvfs2_bufmap_init_waitq); + + gossip_debug(GOSSIP_BUFMAP_DEBUG, + "pvfs_bufmap_initialize: exiting normally\n"); + return 0; + +out_unmap_bufmap: + pvfs2_bufmap_unmap(bufmap); +out_free_bufmap: + pvfs2_bufmap_free(bufmap); +out: + return ret; +} + +/* + * pvfs_bufmap_finalize() + * + * shuts down the mapped buffer interface and releases any resources + * associated with it + * + * no return value + */ +void pvfs_bufmap_finalize(void) +{ + gossip_debug(GOSSIP_BUFMAP_DEBUG, "pvfs2_bufmap_finalize: called\n"); + BUG_ON(!__pvfs2_bufmap); + pvfs2_bufmap_unref(__pvfs2_bufmap); + gossip_debug(GOSSIP_BUFMAP_DEBUG, + "pvfs2_bufmap_finalize: exiting normally\n"); +} + +struct slot_args { + int slot_count; + int *slot_array; + spinlock_t *slot_lock; + wait_queue_head_t *slot_wq; +}; + +static int wait_for_a_slot(struct slot_args *slargs, int *buffer_index) +{ + int ret = -1; + int i = 0; + DECLARE_WAITQUEUE(my_wait, current); + + + add_wait_queue_exclusive(slargs->slot_wq, &my_wait); + + while (1) { + set_current_state(TASK_INTERRUPTIBLE); + + /* + * check for available desc, slot_lock is the appropriate + * index_lock + */ + spin_lock(slargs->slot_lock); + for (i = 0; i < slargs->slot_count; i++) + if (slargs->slot_array[i] == 0) { + slargs->slot_array[i] = 1; + *buffer_index = i; + ret = 0; + break; + } + spin_unlock(slargs->slot_lock); + + /* if we acquired a buffer, then break out of while */ + if (ret == 0) + break; + + if (!signal_pending(current)) { + int timeout = + MSECS_TO_JIFFIES(1000 * slot_timeout_secs); + gossip_debug(GOSSIP_BUFMAP_DEBUG, + "[BUFMAP]: waiting %d " + "seconds for a slot\n", + slot_timeout_secs); + if (!schedule_timeout(timeout)) { + gossip_debug(GOSSIP_BUFMAP_DEBUG, + "*** wait_for_a_slot timed out\n"); + ret = -ETIMEDOUT; + break; + } + gossip_debug(GOSSIP_BUFMAP_DEBUG, + "[BUFMAP]: woken up by a slot becoming available.\n"); + continue; + } + + gossip_debug(GOSSIP_BUFMAP_DEBUG, "pvfs2: %s interrupted.\n", + __func__); + ret = -EINTR; + break; + } + + set_current_state(TASK_RUNNING); + remove_wait_queue(slargs->slot_wq, &my_wait); + return ret; +} + +static void put_back_slot(struct slot_args *slargs, int buffer_index) +{ + /* slot_lock is the appropriate index_lock */ + spin_lock(slargs->slot_lock); + if (buffer_index < 0 || buffer_index >= slargs->slot_count) { + spin_unlock(slargs->slot_lock); + return; + } + + /* put the desc back on the queue */ + slargs->slot_array[buffer_index] = 0; + spin_unlock(slargs->slot_lock); + + /* wake up anyone who may be sleeping on the queue */ + wake_up_interruptible(slargs->slot_wq); +} + +/* + * pvfs_bufmap_get() + * + * gets a free mapped buffer descriptor, will sleep until one becomes + * available if necessary + * + * returns 0 on success, -errno on failure + */ +int pvfs_bufmap_get(struct pvfs2_bufmap **mapp, int *buffer_index) +{ + struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref(); + struct slot_args slargs; + int ret; + + if (!bufmap) { + gossip_err("pvfs2: please confirm that pvfs2-client daemon is running.\n"); + return -EIO; + } + + slargs.slot_count = bufmap->desc_count; + slargs.slot_array = bufmap->buffer_index_array; + slargs.slot_lock = &bufmap->buffer_index_lock; + slargs.slot_wq = &bufmap_waitq; + ret = wait_for_a_slot(&slargs, buffer_index); + if (ret) + pvfs2_bufmap_unref(bufmap); + *mapp = bufmap; + return ret; +} + +/* + * pvfs_bufmap_put() + * + * returns a mapped buffer descriptor to the collection + * + * no return value + */ +void pvfs_bufmap_put(struct pvfs2_bufmap *bufmap, int buffer_index) +{ + struct slot_args slargs; + + slargs.slot_count = bufmap->desc_count; + slargs.slot_array = bufmap->buffer_index_array; + slargs.slot_lock = &bufmap->buffer_index_lock; + slargs.slot_wq = &bufmap_waitq; + put_back_slot(&slargs, buffer_index); + pvfs2_bufmap_unref(bufmap); +} + +/* + * readdir_index_get() + * + * gets a free descriptor, will sleep until one becomes + * available if necessary. + * Although the readdir buffers are not mapped into kernel space + * we could do that at a later point of time. Regardless, these + * indices are used by the client-core. + * + * returns 0 on success, -errno on failure + */ +int readdir_index_get(struct pvfs2_bufmap **mapp, int *buffer_index) +{ + struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref(); + struct slot_args slargs; + int ret; + + if (!bufmap) { + gossip_err("pvfs2: please confirm that pvfs2-client daemon is running.\n"); + return -EIO; + } + + slargs.slot_count = PVFS2_READDIR_DEFAULT_DESC_COUNT; + slargs.slot_array = bufmap->readdir_index_array; + slargs.slot_lock = &bufmap->readdir_index_lock; + slargs.slot_wq = &readdir_waitq; + ret = wait_for_a_slot(&slargs, buffer_index); + if (ret) + pvfs2_bufmap_unref(bufmap); + *mapp = bufmap; + return ret; +} + +void readdir_index_put(struct pvfs2_bufmap *bufmap, int buffer_index) +{ + struct slot_args slargs; + + slargs.slot_count = PVFS2_READDIR_DEFAULT_DESC_COUNT; + slargs.slot_array = bufmap->readdir_index_array; + slargs.slot_lock = &bufmap->readdir_index_lock; + slargs.slot_wq = &readdir_waitq; + put_back_slot(&slargs, buffer_index); + pvfs2_bufmap_unref(bufmap); +} + +int pvfs_bufmap_copy_from_iovec(struct pvfs2_bufmap *bufmap, + struct iov_iter *iter, + int buffer_index, + size_t size) +{ + struct pvfs_bufmap_desc *to; + struct page *page; + size_t copied; + int i; + + gossip_debug(GOSSIP_BUFMAP_DEBUG, + "%s: buffer_index:%d: size:%lu:\n", + __func__, buffer_index, size); + + to = &bufmap->desc_array[buffer_index]; + + for (i = 0; size; i++) { + page = to->page_array[i]; + copied = copy_page_from_iter(page, 0, PAGE_SIZE, iter); + size -= copied; + if ((copied == 0) && (size)) + break; + } + + return size ? -EFAULT : 0; + +} + +/* + * Iterate through the array of pages containing the bytes from + * a file being read. + * + */ +int pvfs_bufmap_copy_to_iovec(struct pvfs2_bufmap *bufmap, + struct iov_iter *iter, + int buffer_index) +{ + struct pvfs_bufmap_desc *from; + struct page *page; + int i; + size_t written; + + gossip_debug(GOSSIP_BUFMAP_DEBUG, + "%s: buffer_index:%d: iov_iter_count(iter):%lu:\n", + __func__, buffer_index, iov_iter_count(iter)); + + from = &bufmap->desc_array[buffer_index]; + + for (i = 0; iov_iter_count(iter); i++) { + page = from->page_array[i]; + written = copy_page_to_iter(page, 0, PAGE_SIZE, iter); + if ((written == 0) && (iov_iter_count(iter))) + break; + } + + return iov_iter_count(iter) ? -EFAULT : 0; +} diff --git a/fs/orangefs/pvfs2-bufmap.h b/fs/orangefs/pvfs2-bufmap.h new file mode 100644 index 000000000000..a0f84c045d73 --- /dev/null +++ b/fs/orangefs/pvfs2-bufmap.h @@ -0,0 +1,61 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +#ifndef __PVFS2_BUFMAP_H +#define __PVFS2_BUFMAP_H + +/* used to describe mapped buffers */ +struct pvfs_bufmap_desc { + void *uaddr; /* user space address pointer */ + struct page **page_array; /* array of mapped pages */ + int array_count; /* size of above arrays */ + struct list_head list_link; +}; + +struct pvfs2_bufmap; + +struct pvfs2_bufmap *pvfs2_bufmap_ref(void); +void pvfs2_bufmap_unref(struct pvfs2_bufmap *bufmap); + +/* + * pvfs_bufmap_size_query is now an inline function because buffer + * sizes are not hardcoded + */ +int pvfs_bufmap_size_query(void); + +int pvfs_bufmap_shift_query(void); + +int pvfs_bufmap_initialize(struct PVFS_dev_map_desc *user_desc); + +int get_bufmap_init(void); + +void pvfs_bufmap_finalize(void); + +int pvfs_bufmap_get(struct pvfs2_bufmap **mapp, int *buffer_index); + +void pvfs_bufmap_put(struct pvfs2_bufmap *bufmap, int buffer_index); + +int readdir_index_get(struct pvfs2_bufmap **mapp, int *buffer_index); + +void readdir_index_put(struct pvfs2_bufmap *bufmap, int buffer_index); + +int pvfs_bufmap_copy_from_iovec(struct pvfs2_bufmap *bufmap, + struct iov_iter *iter, + int buffer_index, + size_t size); + +int pvfs_bufmap_copy_to_iovec(struct pvfs2_bufmap *bufmap, + struct iov_iter *iter, + int buffer_index); + +size_t pvfs_bufmap_copy_to_user_task_iovec(struct task_struct *tsk, + struct iovec *iovec, + unsigned long nr_segs, + struct pvfs2_bufmap *bufmap, + int buffer_index, + size_t bytes_to_be_copied); + +#endif /* __PVFS2_BUFMAP_H */ diff --git a/fs/orangefs/pvfs2-cache.c b/fs/orangefs/pvfs2-cache.c new file mode 100644 index 000000000000..15251884ba4a --- /dev/null +++ b/fs/orangefs/pvfs2-cache.c @@ -0,0 +1,260 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" + +/* tags assigned to kernel upcall operations */ +static __u64 next_tag_value; +static DEFINE_SPINLOCK(next_tag_value_lock); + +/* the pvfs2 memory caches */ + +/* a cache for pvfs2 upcall/downcall operations */ +static struct kmem_cache *op_cache; + +/* a cache for device (/dev/pvfs2-req) communication */ +static struct kmem_cache *dev_req_cache; + +/* a cache for pvfs2_kiocb objects (i.e pvfs2 iocb structures ) */ +static struct kmem_cache *pvfs2_kiocb_cache; + +int op_cache_initialize(void) +{ + op_cache = kmem_cache_create("pvfs2_op_cache", + sizeof(struct pvfs2_kernel_op_s), + 0, + PVFS2_CACHE_CREATE_FLAGS, + NULL); + + if (!op_cache) { + gossip_err("Cannot create pvfs2_op_cache\n"); + return -ENOMEM; + } + + /* initialize our atomic tag counter */ + spin_lock(&next_tag_value_lock); + next_tag_value = 100; + spin_unlock(&next_tag_value_lock); + return 0; +} + +int op_cache_finalize(void) +{ + kmem_cache_destroy(op_cache); + return 0; +} + +char *get_opname_string(struct pvfs2_kernel_op_s *new_op) +{ + if (new_op) { + __s32 type = new_op->upcall.type; + + if (type == PVFS2_VFS_OP_FILE_IO) + return "OP_FILE_IO"; + else if (type == PVFS2_VFS_OP_LOOKUP) + return "OP_LOOKUP"; + else if (type == PVFS2_VFS_OP_CREATE) + return "OP_CREATE"; + else if (type == PVFS2_VFS_OP_GETATTR) + return "OP_GETATTR"; + else if (type == PVFS2_VFS_OP_REMOVE) + return "OP_REMOVE"; + else if (type == PVFS2_VFS_OP_MKDIR) + return "OP_MKDIR"; + else if (type == PVFS2_VFS_OP_READDIR) + return "OP_READDIR"; + else if (type == PVFS2_VFS_OP_READDIRPLUS) + return "OP_READDIRPLUS"; + else if (type == PVFS2_VFS_OP_SETATTR) + return "OP_SETATTR"; + else if (type == PVFS2_VFS_OP_SYMLINK) + return "OP_SYMLINK"; + else if (type == PVFS2_VFS_OP_RENAME) + return "OP_RENAME"; + else if (type == PVFS2_VFS_OP_STATFS) + return "OP_STATFS"; + else if (type == PVFS2_VFS_OP_TRUNCATE) + return "OP_TRUNCATE"; + else if (type == PVFS2_VFS_OP_MMAP_RA_FLUSH) + return "OP_MMAP_RA_FLUSH"; + else if (type == PVFS2_VFS_OP_FS_MOUNT) + return "OP_FS_MOUNT"; + else if (type == PVFS2_VFS_OP_FS_UMOUNT) + return "OP_FS_UMOUNT"; + else if (type == PVFS2_VFS_OP_GETXATTR) + return "OP_GETXATTR"; + else if (type == PVFS2_VFS_OP_SETXATTR) + return "OP_SETXATTR"; + else if (type == PVFS2_VFS_OP_LISTXATTR) + return "OP_LISTXATTR"; + else if (type == PVFS2_VFS_OP_REMOVEXATTR) + return "OP_REMOVEXATTR"; + else if (type == PVFS2_VFS_OP_PARAM) + return "OP_PARAM"; + else if (type == PVFS2_VFS_OP_PERF_COUNT) + return "OP_PERF_COUNT"; + else if (type == PVFS2_VFS_OP_CANCEL) + return "OP_CANCEL"; + else if (type == PVFS2_VFS_OP_FSYNC) + return "OP_FSYNC"; + else if (type == PVFS2_VFS_OP_FSKEY) + return "OP_FSKEY"; + else if (type == PVFS2_VFS_OP_FILE_IOX) + return "OP_FILE_IOX"; + } + return "OP_UNKNOWN?"; +} + +static struct pvfs2_kernel_op_s *op_alloc_common(__s32 op_linger, __s32 type) +{ + struct pvfs2_kernel_op_s *new_op = NULL; + + new_op = kmem_cache_alloc(op_cache, PVFS2_CACHE_ALLOC_FLAGS); + if (new_op) { + memset(new_op, 0, sizeof(struct pvfs2_kernel_op_s)); + + INIT_LIST_HEAD(&new_op->list); + spin_lock_init(&new_op->lock); + init_waitqueue_head(&new_op->waitq); + + init_waitqueue_head(&new_op->io_completion_waitq); + atomic_set(&new_op->aio_ref_count, 0); + + pvfs2_op_initialize(new_op); + + /* initialize the op specific tag and upcall credentials */ + spin_lock(&next_tag_value_lock); + new_op->tag = next_tag_value++; + if (next_tag_value == 0) + next_tag_value = 100; + spin_unlock(&next_tag_value_lock); + new_op->upcall.type = type; + new_op->attempts = 0; + gossip_debug(GOSSIP_CACHE_DEBUG, + "Alloced OP (%p: %llu %s)\n", + new_op, + llu(new_op->tag), + get_opname_string(new_op)); + + new_op->upcall.uid = from_kuid(current_user_ns(), + current_fsuid()); + + new_op->upcall.gid = from_kgid(current_user_ns(), + current_fsgid()); + + new_op->op_linger = new_op->op_linger_tmp = op_linger; + } else { + gossip_err("op_alloc: kmem_cache_alloc failed!\n"); + } + return new_op; +} + +struct pvfs2_kernel_op_s *op_alloc(__s32 type) +{ + return op_alloc_common(1, type); +} + +struct pvfs2_kernel_op_s *op_alloc_trailer(__s32 type) +{ + return op_alloc_common(2, type); +} + +void op_release(struct pvfs2_kernel_op_s *pvfs2_op) +{ + if (pvfs2_op) { + gossip_debug(GOSSIP_CACHE_DEBUG, + "Releasing OP (%p: %llu)\n", + pvfs2_op, + llu(pvfs2_op->tag)); + pvfs2_op_initialize(pvfs2_op); + kmem_cache_free(op_cache, pvfs2_op); + } else { + gossip_err("NULL pointer in op_release\n"); + } +} + +int dev_req_cache_initialize(void) +{ + dev_req_cache = kmem_cache_create("pvfs2_devreqcache", + MAX_ALIGNED_DEV_REQ_DOWNSIZE, + 0, + PVFS2_CACHE_CREATE_FLAGS, + NULL); + + if (!dev_req_cache) { + gossip_err("Cannot create pvfs2_dev_req_cache\n"); + return -ENOMEM; + } + return 0; +} + +int dev_req_cache_finalize(void) +{ + kmem_cache_destroy(dev_req_cache); + return 0; +} + +void *dev_req_alloc(void) +{ + void *buffer; + + buffer = kmem_cache_alloc(dev_req_cache, PVFS2_CACHE_ALLOC_FLAGS); + if (buffer == NULL) + gossip_err("Failed to allocate from dev_req_cache\n"); + else + memset(buffer, 0, sizeof(MAX_ALIGNED_DEV_REQ_DOWNSIZE)); + return buffer; +} + +void dev_req_release(void *buffer) +{ + if (buffer) + kmem_cache_free(dev_req_cache, buffer); + else + gossip_err("NULL pointer passed to dev_req_release\n"); +} + +int kiocb_cache_initialize(void) +{ + pvfs2_kiocb_cache = kmem_cache_create("pvfs2_kiocbcache", + sizeof(struct pvfs2_kiocb_s), + 0, + PVFS2_CACHE_CREATE_FLAGS, + NULL); + + if (!pvfs2_kiocb_cache) { + gossip_err("Cannot create pvfs2_kiocb_cache!\n"); + return -ENOMEM; + } + return 0; +} + +int kiocb_cache_finalize(void) +{ + kmem_cache_destroy(pvfs2_kiocb_cache); + return 0; +} + +struct pvfs2_kiocb_s *kiocb_alloc(void) +{ + struct pvfs2_kiocb_s *x = NULL; + + x = kmem_cache_alloc(pvfs2_kiocb_cache, PVFS2_CACHE_ALLOC_FLAGS); + if (x == NULL) + gossip_err("kiocb_alloc: kmem_cache_alloc failed!\n"); + else + memset(x, 0, sizeof(struct pvfs2_kiocb_s)); + return x; +} + +void kiocb_release(struct pvfs2_kiocb_s *x) +{ + if (x) + kmem_cache_free(pvfs2_kiocb_cache, x); + else + gossip_err("kiocb_release: kmem_cache_free NULL pointer!\n"); +} diff --git a/fs/orangefs/pvfs2-debug.h b/fs/orangefs/pvfs2-debug.h new file mode 100644 index 000000000000..fd71d6c84cf6 --- /dev/null +++ b/fs/orangefs/pvfs2-debug.h @@ -0,0 +1,292 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +/* This file just defines debugging masks to be used with the gossip + * logging utility. All debugging masks for PVFS2 are kept here to make + * sure we don't have collisions. + */ + +#ifndef __PVFS2_DEBUG_H +#define __PVFS2_DEBUG_H + +#ifdef __KERNEL__ +#include +#else +#include +#endif + +#define GOSSIP_NO_DEBUG (__u64)0 +#define GOSSIP_BMI_DEBUG_TCP ((__u64)1 << 0) +#define GOSSIP_BMI_DEBUG_CONTROL ((__u64)1 << 1) +#define GOSSIP_BMI_DEBUG_OFFSETS ((__u64)1 << 2) +#define GOSSIP_BMI_DEBUG_GM ((__u64)1 << 3) +#define GOSSIP_JOB_DEBUG ((__u64)1 << 4) +#define GOSSIP_SERVER_DEBUG ((__u64)1 << 5) +#define GOSSIP_STO_DEBUG_CTRL ((__u64)1 << 6) +#define GOSSIP_STO_DEBUG_DEFAULT ((__u64)1 << 7) +#define GOSSIP_FLOW_DEBUG ((__u64)1 << 8) +#define GOSSIP_BMI_DEBUG_GM_MEM ((__u64)1 << 9) +#define GOSSIP_REQUEST_DEBUG ((__u64)1 << 10) +#define GOSSIP_FLOW_PROTO_DEBUG ((__u64)1 << 11) +#define GOSSIP_NCACHE_DEBUG ((__u64)1 << 12) +#define GOSSIP_CLIENT_DEBUG ((__u64)1 << 13) +#define GOSSIP_REQ_SCHED_DEBUG ((__u64)1 << 14) +#define GOSSIP_ACACHE_DEBUG ((__u64)1 << 15) +#define GOSSIP_TROVE_DEBUG ((__u64)1 << 16) +#define GOSSIP_TROVE_OP_DEBUG ((__u64)1 << 17) +#define GOSSIP_DIST_DEBUG ((__u64)1 << 18) +#define GOSSIP_BMI_DEBUG_IB ((__u64)1 << 19) +#define GOSSIP_DBPF_ATTRCACHE_DEBUG ((__u64)1 << 20) +#define GOSSIP_MMAP_RCACHE_DEBUG ((__u64)1 << 21) +#define GOSSIP_LOOKUP_DEBUG ((__u64)1 << 22) +#define GOSSIP_REMOVE_DEBUG ((__u64)1 << 23) +#define GOSSIP_GETATTR_DEBUG ((__u64)1 << 24) +#define GOSSIP_READDIR_DEBUG ((__u64)1 << 25) +#define GOSSIP_IO_DEBUG ((__u64)1 << 26) +#define GOSSIP_DBPF_OPEN_CACHE_DEBUG ((__u64)1 << 27) +#define GOSSIP_PERMISSIONS_DEBUG ((__u64)1 << 28) +#define GOSSIP_CANCEL_DEBUG ((__u64)1 << 29) +#define GOSSIP_MSGPAIR_DEBUG ((__u64)1 << 30) +#define GOSSIP_CLIENTCORE_DEBUG ((__u64)1 << 31) +#define GOSSIP_CLIENTCORE_TIMING_DEBUG ((__u64)1 << 32) +#define GOSSIP_SETATTR_DEBUG ((__u64)1 << 33) +#define GOSSIP_MKDIR_DEBUG ((__u64)1 << 34) +#define GOSSIP_VARSTRIP_DEBUG ((__u64)1 << 35) +#define GOSSIP_GETEATTR_DEBUG ((__u64)1 << 36) +#define GOSSIP_SETEATTR_DEBUG ((__u64)1 << 37) +#define GOSSIP_ENDECODE_DEBUG ((__u64)1 << 38) +#define GOSSIP_DELEATTR_DEBUG ((__u64)1 << 39) +#define GOSSIP_ACCESS_DEBUG ((__u64)1 << 40) +#define GOSSIP_ACCESS_DETAIL_DEBUG ((__u64)1 << 41) +#define GOSSIP_LISTEATTR_DEBUG ((__u64)1 << 42) +#define GOSSIP_PERFCOUNTER_DEBUG ((__u64)1 << 43) +#define GOSSIP_STATE_MACHINE_DEBUG ((__u64)1 << 44) +#define GOSSIP_DBPF_KEYVAL_DEBUG ((__u64)1 << 45) +#define GOSSIP_LISTATTR_DEBUG ((__u64)1 << 46) +#define GOSSIP_DBPF_COALESCE_DEBUG ((__u64)1 << 47) +#define GOSSIP_ACCESS_HOSTNAMES ((__u64)1 << 48) +#define GOSSIP_FSCK_DEBUG ((__u64)1 << 49) +#define GOSSIP_BMI_DEBUG_MX ((__u64)1 << 50) +#define GOSSIP_BSTREAM_DEBUG ((__u64)1 << 51) +#define GOSSIP_BMI_DEBUG_PORTALS ((__u64)1 << 52) +#define GOSSIP_USER_DEV_DEBUG ((__u64)1 << 53) +#define GOSSIP_DIRECTIO_DEBUG ((__u64)1 << 54) +#define GOSSIP_MGMT_DEBUG ((__u64)1 << 55) +#define GOSSIP_MIRROR_DEBUG ((__u64)1 << 56) +#define GOSSIP_WIN_CLIENT_DEBUG ((__u64)1 << 57) +#define GOSSIP_SECURITY_DEBUG ((__u64)1 << 58) +#define GOSSIP_USRINT_DEBUG ((__u64)1 << 59) +#define GOSSIP_RCACHE_DEBUG ((__u64)1 << 60) +#define GOSSIP_SECCACHE_DEBUG ((__u64)1 << 61) + +#define GOSSIP_BMI_DEBUG_ALL ((__u64) (GOSSIP_BMI_DEBUG_TCP + \ + GOSSIP_BMI_DEBUG_CONTROL + \ + GOSSIP_BMI_DEBUG_GM + \ + GOSSIP_BMI_DEBUG_OFFSETS + \ + GOSSIP_BMI_DEBUG_IB + \ + GOSSIP_BMI_DEBUG_MX + \ + GOSSIP_BMI_DEBUG_PORTALS)) + +const char *PVFS_debug_get_next_debug_keyword(int position); + +#define GOSSIP_SUPER_DEBUG ((__u64)1 << 0) +#define GOSSIP_INODE_DEBUG ((__u64)1 << 1) +#define GOSSIP_FILE_DEBUG ((__u64)1 << 2) +#define GOSSIP_DIR_DEBUG ((__u64)1 << 3) +#define GOSSIP_UTILS_DEBUG ((__u64)1 << 4) +#define GOSSIP_WAIT_DEBUG ((__u64)1 << 5) +#define GOSSIP_ACL_DEBUG ((__u64)1 << 6) +#define GOSSIP_DCACHE_DEBUG ((__u64)1 << 7) +#define GOSSIP_DEV_DEBUG ((__u64)1 << 8) +#define GOSSIP_NAME_DEBUG ((__u64)1 << 9) +#define GOSSIP_BUFMAP_DEBUG ((__u64)1 << 10) +#define GOSSIP_CACHE_DEBUG ((__u64)1 << 11) +#define GOSSIP_DEBUGFS_DEBUG ((__u64)1 << 12) +#define GOSSIP_XATTR_DEBUG ((__u64)1 << 13) +#define GOSSIP_INIT_DEBUG ((__u64)1 << 14) +#define GOSSIP_SYSFS_DEBUG ((__u64)1 << 15) + +#define GOSSIP_MAX_NR 16 +#define GOSSIP_MAX_DEBUG (((__u64)1 << GOSSIP_MAX_NR) - 1) + +/*function prototypes*/ +__u64 PVFS_kmod_eventlog_to_mask(const char *event_logging); +__u64 PVFS_debug_eventlog_to_mask(const char *event_logging); +char *PVFS_debug_mask_to_eventlog(__u64 mask); +char *PVFS_kmod_mask_to_eventlog(__u64 mask); + +/* a private internal type */ +struct __keyword_mask_s { + const char *keyword; + __u64 mask_val; +}; + +#define __DEBUG_ALL ((__u64) -1) + +/* map all config keywords to pvfs2 debug masks here */ +static struct __keyword_mask_s s_keyword_mask_map[] = { + /* Log trove debugging info. Same as 'trove'. */ + {"storage", GOSSIP_TROVE_DEBUG}, + /* Log trove debugging info. Same as 'storage'. */ + {"trove", GOSSIP_TROVE_DEBUG}, + /* Log trove operations. */ + {"trove_op", GOSSIP_TROVE_OP_DEBUG}, + /* Log network debug info. */ + {"network", GOSSIP_BMI_DEBUG_ALL}, + /* Log server info, including new operations. */ + {"server", GOSSIP_SERVER_DEBUG}, + /* Log client sysint info. This is only useful for the client. */ + {"client", GOSSIP_CLIENT_DEBUG}, + /* Debug the varstrip distribution */ + {"varstrip", GOSSIP_VARSTRIP_DEBUG}, + /* Log job info */ + {"job", GOSSIP_JOB_DEBUG}, + /* Debug PINT_process_request calls. EXTREMELY verbose! */ + {"request", GOSSIP_REQUEST_DEBUG}, + /* Log request scheduler events */ + {"reqsched", GOSSIP_REQ_SCHED_DEBUG}, + /* Log the flow protocol events, including flowproto_multiqueue */ + {"flowproto", GOSSIP_FLOW_PROTO_DEBUG}, + /* Log flow calls */ + {"flow", GOSSIP_FLOW_DEBUG}, + /* Debug the client name cache. Only useful on the client. */ + {"ncache", GOSSIP_NCACHE_DEBUG}, + /* Debug read-ahead cache events. Only useful on the client. */ + {"mmaprcache", GOSSIP_MMAP_RCACHE_DEBUG}, + /* Debug the attribute cache. Only useful on the client. */ + {"acache", GOSSIP_ACACHE_DEBUG}, + /* Log/Debug distribution calls */ + {"distribution", GOSSIP_DIST_DEBUG}, + /* Debug the server-side dbpf attribute cache */ + {"dbpfattrcache", GOSSIP_DBPF_ATTRCACHE_DEBUG}, + /* Debug the client lookup state machine. */ + {"lookup", GOSSIP_LOOKUP_DEBUG}, + /* Debug the client remove state macine. */ + {"remove", GOSSIP_REMOVE_DEBUG}, + /* Debug the server getattr state machine. */ + {"getattr", GOSSIP_GETATTR_DEBUG}, + /* Debug the server setattr state machine. */ + {"setattr", GOSSIP_SETATTR_DEBUG}, + /* vectored getattr server state machine */ + {"listattr", GOSSIP_LISTATTR_DEBUG}, + /* Debug the client and server get ext attributes SM. */ + {"geteattr", GOSSIP_GETEATTR_DEBUG}, + /* Debug the client and server set ext attributes SM. */ + {"seteattr", GOSSIP_SETEATTR_DEBUG}, + /* Debug the readdir operation (client and server) */ + {"readdir", GOSSIP_READDIR_DEBUG}, + /* Debug the mkdir operation (server only) */ + {"mkdir", GOSSIP_MKDIR_DEBUG}, + /* + * Debug the io operation (reads and writes) + * for both the client and server. + */ + {"io", GOSSIP_IO_DEBUG}, + /* Debug the server's open file descriptor cache */ + {"open_cache", GOSSIP_DBPF_OPEN_CACHE_DEBUG}, + /* Debug permissions checking on the server */ + {"permissions", GOSSIP_PERMISSIONS_DEBUG}, + /* Debug the cancel operation */ + {"cancel", GOSSIP_CANCEL_DEBUG}, + /* Debug the msgpair state machine */ + {"msgpair", GOSSIP_MSGPAIR_DEBUG}, + /* Debug the client core app */ + {"clientcore", GOSSIP_CLIENTCORE_DEBUG}, + /* Debug the client timing state machines (job timeout, etc.) */ + {"clientcore_timing", GOSSIP_CLIENTCORE_TIMING_DEBUG}, + /* network encoding */ + {"endecode", GOSSIP_ENDECODE_DEBUG}, + /* Show server file (metadata) accesses (both modify and read-only). */ + {"access", GOSSIP_ACCESS_DEBUG}, + /* Show more detailed server file accesses */ + {"access_detail", GOSSIP_ACCESS_DETAIL_DEBUG}, + /* Debug the listeattr operation */ + {"listeattr", GOSSIP_LISTEATTR_DEBUG}, + /* Debug the state machine management code */ + {"sm", GOSSIP_STATE_MACHINE_DEBUG}, + /* Debug the metadata dbpf keyval functions */ + {"keyval", GOSSIP_DBPF_KEYVAL_DEBUG}, + /* Debug the metadata sync coalescing code */ + {"coalesce", GOSSIP_DBPF_COALESCE_DEBUG}, + /* Display the hostnames instead of IP addrs in debug output */ + {"access_hostnames", GOSSIP_ACCESS_HOSTNAMES}, + /* Show the client device events */ + {"user_dev", GOSSIP_USER_DEV_DEBUG}, + /* Debug the fsck tool */ + {"fsck", GOSSIP_FSCK_DEBUG}, + /* Debug the bstream code */ + {"bstream", GOSSIP_BSTREAM_DEBUG}, + /* Debug trove in direct io mode */ + {"directio", GOSSIP_DIRECTIO_DEBUG}, + /* Debug direct io thread management */ + {"mgmt", GOSSIP_MGMT_DEBUG}, + /* Debug mirroring process */ + {"mirror", GOSSIP_MIRROR_DEBUG}, + /* Windows client */ + {"win_client", GOSSIP_WIN_CLIENT_DEBUG}, + /* Debug robust security code */ + {"security", GOSSIP_SECURITY_DEBUG}, + /* Capability Cache */ + {"seccache", GOSSIP_SECCACHE_DEBUG}, + /* Client User Interface */ + {"usrint", GOSSIP_USRINT_DEBUG}, + /* rcache */ + {"rcache", GOSSIP_RCACHE_DEBUG}, + /* Everything except the periodic events. Useful for debugging */ + {"verbose", + (__DEBUG_ALL & + ~(GOSSIP_PERFCOUNTER_DEBUG | GOSSIP_STATE_MACHINE_DEBUG | + GOSSIP_ENDECODE_DEBUG | GOSSIP_USER_DEV_DEBUG)) + }, + /* No debug output */ + {"none", GOSSIP_NO_DEBUG}, + /* Everything */ + {"all", __DEBUG_ALL} +}; + +#undef __DEBUG_ALL + +/* + * Map all kmod keywords to kmod debug masks here. Keep this + * structure "packed": + * + * "all" is always last... + * + * keyword mask_val index + * foo 1 0 + * bar 2 1 + * baz 4 2 + * qux 8 3 + * . . . + */ +static struct __keyword_mask_s s_kmod_keyword_mask_map[] = { + {"super", GOSSIP_SUPER_DEBUG}, + {"inode", GOSSIP_INODE_DEBUG}, + {"file", GOSSIP_FILE_DEBUG}, + {"dir", GOSSIP_DIR_DEBUG}, + {"utils", GOSSIP_UTILS_DEBUG}, + {"wait", GOSSIP_WAIT_DEBUG}, + {"acl", GOSSIP_ACL_DEBUG}, + {"dcache", GOSSIP_DCACHE_DEBUG}, + {"dev", GOSSIP_DEV_DEBUG}, + {"name", GOSSIP_NAME_DEBUG}, + {"bufmap", GOSSIP_BUFMAP_DEBUG}, + {"cache", GOSSIP_CACHE_DEBUG}, + {"debugfs", GOSSIP_DEBUGFS_DEBUG}, + {"xattr", GOSSIP_XATTR_DEBUG}, + {"init", GOSSIP_INIT_DEBUG}, + {"sysfs", GOSSIP_SYSFS_DEBUG}, + {"none", GOSSIP_NO_DEBUG}, + {"all", GOSSIP_MAX_DEBUG} +}; + +static const int num_kmod_keyword_mask_map = (int) + (sizeof(s_kmod_keyword_mask_map) / sizeof(struct __keyword_mask_s)); + +static const int num_keyword_mask_map = (int) + (sizeof(s_keyword_mask_map) / sizeof(struct __keyword_mask_s)); + +#endif /* __PVFS2_DEBUG_H */ diff --git a/fs/orangefs/pvfs2-debugfs.c b/fs/orangefs/pvfs2-debugfs.c new file mode 100644 index 000000000000..ba5bfef7a3f3 --- /dev/null +++ b/fs/orangefs/pvfs2-debugfs.c @@ -0,0 +1,458 @@ +/* + * What: /sys/kernel/debug/orangefs/debug-help + * Date: June 2015 + * Contact: Mike Marshall + * Description: + * List of client and kernel debug keywords. + * + * + * What: /sys/kernel/debug/orangefs/client-debug + * Date: June 2015 + * Contact: Mike Marshall + * Description: + * Debug setting for "the client", the userspace + * helper for the kernel module. + * + * + * What: /sys/kernel/debug/orangefs/kernel-debug + * Date: June 2015 + * Contact: Mike Marshall + * Description: + * Debug setting for the orangefs kernel module. + * + * Any of the keywords, or comma-separated lists + * of keywords, from debug-help can be catted to + * client-debug or kernel-debug. + * + * "none", "all" and "verbose" are special keywords + * for client-debug. Setting client-debug to "all" + * is kind of like trying to drink water from a + * fire hose, "verbose" triggers most of the same + * output except for the constant flow of output + * from the main wait loop. + * + * "none" and "all" are similar settings for kernel-debug + * no need for a "verbose". + */ +#include +#include + +#include + +#include "pvfs2-debugfs.h" +#include "protocol.h" +#include "pvfs2-kernel.h" + +static int orangefs_debug_disabled = 1; + +static int orangefs_debug_help_open(struct inode *, struct file *); + +const struct file_operations debug_help_fops = { + .open = orangefs_debug_help_open, + .read = seq_read, + .release = seq_release, + .llseek = seq_lseek, +}; + +static void *help_start(struct seq_file *, loff_t *); +static void *help_next(struct seq_file *, void *, loff_t *); +static void help_stop(struct seq_file *, void *); +static int help_show(struct seq_file *, void *); + +static const struct seq_operations help_debug_ops = { + .start = help_start, + .next = help_next, + .stop = help_stop, + .show = help_show, +}; + +/* + * Used to protect data in ORANGEFS_KMOD_DEBUG_FILE and + * ORANGEFS_KMOD_DEBUG_FILE. + */ +static DEFINE_MUTEX(orangefs_debug_lock); + +int orangefs_debug_open(struct inode *, struct file *); + +static ssize_t orangefs_debug_read(struct file *, + char __user *, + size_t, + loff_t *); + +static ssize_t orangefs_debug_write(struct file *, + const char __user *, + size_t, + loff_t *); + +static const struct file_operations kernel_debug_fops = { + .open = orangefs_debug_open, + .read = orangefs_debug_read, + .write = orangefs_debug_write, + .llseek = generic_file_llseek, +}; + +/* + * initialize kmod debug operations, create orangefs debugfs dir and + * ORANGEFS_KMOD_DEBUG_HELP_FILE. + */ +int pvfs2_debugfs_init(void) +{ + + int rc = -ENOMEM; + + debug_dir = debugfs_create_dir("orangefs", NULL); + if (!debug_dir) + goto out; + + help_file_dentry = debugfs_create_file(ORANGEFS_KMOD_DEBUG_HELP_FILE, + 0444, + debug_dir, + debug_help_string, + &debug_help_fops); + if (!help_file_dentry) + goto out; + + orangefs_debug_disabled = 0; + rc = 0; + +out: + if (rc) + pvfs2_debugfs_cleanup(); + + return rc; +} + +void pvfs2_debugfs_cleanup(void) +{ + debugfs_remove_recursive(debug_dir); +} + +/* open ORANGEFS_KMOD_DEBUG_HELP_FILE */ +static int orangefs_debug_help_open(struct inode *inode, struct file *file) +{ + int rc = -ENODEV; + int ret; + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "orangefs_debug_help_open: start\n"); + + if (orangefs_debug_disabled) + goto out; + + ret = seq_open(file, &help_debug_ops); + if (ret) + goto out; + + ((struct seq_file *)(file->private_data))->private = inode->i_private; + + rc = 0; + +out: + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "orangefs_debug_help_open: rc:%d:\n", + rc); + return rc; +} + +/* + * I think start always gets called again after stop. Start + * needs to return NULL when it is done. The whole "payload" + * in this case is a single (long) string, so by the second + * time we get to start (pos = 1), we're done. + */ +static void *help_start(struct seq_file *m, loff_t *pos) +{ + void *payload = NULL; + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "help_start: start\n"); + + if (*pos == 0) + payload = m->private; + + return payload; +} + +static void *help_next(struct seq_file *m, void *v, loff_t *pos) +{ + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "help_next: start\n"); + + return NULL; +} + +static void help_stop(struct seq_file *m, void *p) +{ + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "help_stop: start\n"); +} + +static int help_show(struct seq_file *m, void *v) +{ + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "help_show: start\n"); + + seq_puts(m, v); + + return 0; +} + +/* + * initialize the kernel-debug file. + */ +int pvfs2_kernel_debug_init(void) +{ + + int rc = -ENOMEM; + struct dentry *ret; + char *k_buffer = NULL; + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "%s: start\n", __func__); + + k_buffer = kzalloc(PVFS2_MAX_DEBUG_STRING_LEN, GFP_KERNEL); + if (!k_buffer) + goto out; + + if (strlen(kernel_debug_string) + 1 < PVFS2_MAX_DEBUG_STRING_LEN) { + strcpy(k_buffer, kernel_debug_string); + strcat(k_buffer, "\n"); + } else { + strcpy(k_buffer, "none\n"); + pr_info("%s: overflow 1!\n", __func__); + } + + ret = debugfs_create_file(ORANGEFS_KMOD_DEBUG_FILE, + 0444, + debug_dir, + k_buffer, + &kernel_debug_fops); + if (!ret) { + pr_info("%s: failed to create %s.\n", + __func__, + ORANGEFS_KMOD_DEBUG_FILE); + goto out; + } + + rc = 0; + +out: + if (rc) + pvfs2_debugfs_cleanup(); + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "%s: rc:%d:\n", __func__, rc); + return rc; +} + +/* + * initialize the client-debug file. + */ +int pvfs2_client_debug_init(void) +{ + + int rc = -ENOMEM; + char *c_buffer = NULL; + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "%s: start\n", __func__); + + c_buffer = kzalloc(PVFS2_MAX_DEBUG_STRING_LEN, GFP_KERNEL); + if (!c_buffer) + goto out; + + if (strlen(client_debug_string) + 1 < PVFS2_MAX_DEBUG_STRING_LEN) { + strcpy(c_buffer, client_debug_string); + strcat(c_buffer, "\n"); + } else { + strcpy(c_buffer, "none\n"); + pr_info("%s: overflow! 2\n", __func__); + } + + client_debug_dentry = debugfs_create_file(ORANGEFS_CLIENT_DEBUG_FILE, + 0444, + debug_dir, + c_buffer, + &kernel_debug_fops); + if (!client_debug_dentry) { + pr_info("%s: failed to create %s.\n", + __func__, + ORANGEFS_CLIENT_DEBUG_FILE); + goto out; + } + + rc = 0; + +out: + if (rc) + pvfs2_debugfs_cleanup(); + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "%s: rc:%d:\n", __func__, rc); + return rc; +} + +/* open ORANGEFS_KMOD_DEBUG_FILE or ORANGEFS_CLIENT_DEBUG_FILE.*/ +int orangefs_debug_open(struct inode *inode, struct file *file) +{ + int rc = -ENODEV; + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "%s: orangefs_debug_disabled: %d\n", + __func__, + orangefs_debug_disabled); + + if (orangefs_debug_disabled) + goto out; + + rc = 0; + mutex_lock(&orangefs_debug_lock); + file->private_data = inode->i_private; + mutex_unlock(&orangefs_debug_lock); + +out: + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "orangefs_debug_open: rc: %d\n", + rc); + return rc; +} + +static ssize_t orangefs_debug_read(struct file *file, + char __user *ubuf, + size_t count, + loff_t *ppos) +{ + char *buf; + int sprintf_ret; + ssize_t read_ret = -ENOMEM; + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, "orangefs_debug_read: start\n"); + + buf = kmalloc(PVFS2_MAX_DEBUG_STRING_LEN, GFP_KERNEL); + if (!buf) + goto out; + + mutex_lock(&orangefs_debug_lock); + sprintf_ret = sprintf(buf, "%s", (char *)file->private_data); + mutex_unlock(&orangefs_debug_lock); + + read_ret = simple_read_from_buffer(ubuf, count, ppos, buf, sprintf_ret); + + kfree(buf); + +out: + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "orangefs_debug_read: ret: %zu\n", + read_ret); + + return read_ret; +} + +static ssize_t orangefs_debug_write(struct file *file, + const char __user *ubuf, + size_t count, + loff_t *ppos) +{ + char *buf; + int rc = -EFAULT; + size_t silly = 0; + char *debug_string; + struct pvfs2_kernel_op_s *new_op = NULL; + struct client_debug_mask c_mask = { NULL, 0, 0 }; + + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "orangefs_debug_write: %s\n", + file->f_path.dentry->d_name.name); + + /* + * Thwart users who try to jamb a ridiculous number + * of bytes into the debug file... + */ + if (count > PVFS2_MAX_DEBUG_STRING_LEN + 1) { + silly = count; + count = PVFS2_MAX_DEBUG_STRING_LEN + 1; + } + + buf = kmalloc(PVFS2_MAX_DEBUG_STRING_LEN, GFP_KERNEL); + if (!buf) + goto out; + memset(buf, 0, PVFS2_MAX_DEBUG_STRING_LEN); + + if (copy_from_user(buf, ubuf, count - 1)) { + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "%s: copy_from_user failed!\n", + __func__); + goto out; + } + + /* + * Map the keyword string from userspace into a valid debug mask. + * The mapping process involves mapping the human-inputted string + * into a valid mask, and then rebuilding the string from the + * verified valid mask. + * + * A service operation is required to set a new client-side + * debug mask. + */ + if (!strcmp(file->f_path.dentry->d_name.name, + ORANGEFS_KMOD_DEBUG_FILE)) { + debug_string_to_mask(buf, &gossip_debug_mask, 0); + debug_mask_to_string(&gossip_debug_mask, 0); + debug_string = kernel_debug_string; + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "New kernel debug string is %s\n", + kernel_debug_string); + } else { + /* Can't reset client debug mask if client is not running. */ + if (is_daemon_in_service()) { + pr_info("%s: Client not running :%d:\n", + __func__, + is_daemon_in_service()); + goto out; + } + + debug_string_to_mask(buf, &c_mask, 1); + debug_mask_to_string(&c_mask, 1); + debug_string = client_debug_string; + + new_op = op_alloc(PVFS2_VFS_OP_PARAM); + if (!new_op) { + pr_info("%s: op_alloc failed!\n", __func__); + goto out; + } + + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_TWO_MASK_VALUES; + new_op->upcall.req.param.type = PVFS2_PARAM_REQUEST_SET; + memset(new_op->upcall.req.param.s_value, + 0, + PVFS2_MAX_DEBUG_STRING_LEN); + sprintf(new_op->upcall.req.param.s_value, + "%llx %llx\n", + c_mask.mask1, + c_mask.mask2); + + /* service_operation returns 0 on success... */ + rc = service_operation(new_op, + "pvfs2_param", + PVFS2_OP_INTERRUPTIBLE); + + if (rc) + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "%s: service_operation failed! rc:%d:\n", + __func__, + rc); + + op_release(new_op); + } + + mutex_lock(&orangefs_debug_lock); + memset(file->f_inode->i_private, 0, PVFS2_MAX_DEBUG_STRING_LEN); + sprintf((char *)file->f_inode->i_private, "%s\n", debug_string); + mutex_unlock(&orangefs_debug_lock); + + *ppos += count; + if (silly) + rc = silly; + else + rc = count; + +out: + gossip_debug(GOSSIP_DEBUGFS_DEBUG, + "orangefs_debug_write: rc: %d\n", + rc); + kfree(buf); + return rc; +} diff --git a/fs/orangefs/pvfs2-debugfs.h b/fs/orangefs/pvfs2-debugfs.h new file mode 100644 index 000000000000..a66b7d08c14d --- /dev/null +++ b/fs/orangefs/pvfs2-debugfs.h @@ -0,0 +1,3 @@ +int pvfs2_debugfs_init(void); +int pvfs2_kernel_debug_init(void); +void pvfs2_debugfs_cleanup(void); diff --git a/fs/orangefs/pvfs2-dev-proto.h b/fs/orangefs/pvfs2-dev-proto.h new file mode 100644 index 000000000000..9c82e6e651f3 --- /dev/null +++ b/fs/orangefs/pvfs2-dev-proto.h @@ -0,0 +1,102 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +#ifndef _PVFS2_DEV_PROTO_H +#define _PVFS2_DEV_PROTO_H + +/* + * types and constants shared between user space and kernel space for + * device interaction using a common protocol + */ + +/* + * valid pvfs2 kernel operation types + */ +#define PVFS2_VFS_OP_INVALID 0xFF000000 +#define PVFS2_VFS_OP_FILE_IO 0xFF000001 +#define PVFS2_VFS_OP_LOOKUP 0xFF000002 +#define PVFS2_VFS_OP_CREATE 0xFF000003 +#define PVFS2_VFS_OP_GETATTR 0xFF000004 +#define PVFS2_VFS_OP_REMOVE 0xFF000005 +#define PVFS2_VFS_OP_MKDIR 0xFF000006 +#define PVFS2_VFS_OP_READDIR 0xFF000007 +#define PVFS2_VFS_OP_SETATTR 0xFF000008 +#define PVFS2_VFS_OP_SYMLINK 0xFF000009 +#define PVFS2_VFS_OP_RENAME 0xFF00000A +#define PVFS2_VFS_OP_STATFS 0xFF00000B +#define PVFS2_VFS_OP_TRUNCATE 0xFF00000C +#define PVFS2_VFS_OP_MMAP_RA_FLUSH 0xFF00000D +#define PVFS2_VFS_OP_FS_MOUNT 0xFF00000E +#define PVFS2_VFS_OP_FS_UMOUNT 0xFF00000F +#define PVFS2_VFS_OP_GETXATTR 0xFF000010 +#define PVFS2_VFS_OP_SETXATTR 0xFF000011 +#define PVFS2_VFS_OP_LISTXATTR 0xFF000012 +#define PVFS2_VFS_OP_REMOVEXATTR 0xFF000013 +#define PVFS2_VFS_OP_PARAM 0xFF000014 +#define PVFS2_VFS_OP_PERF_COUNT 0xFF000015 +#define PVFS2_VFS_OP_CANCEL 0xFF00EE00 +#define PVFS2_VFS_OP_FSYNC 0xFF00EE01 +#define PVFS2_VFS_OP_FSKEY 0xFF00EE02 +#define PVFS2_VFS_OP_READDIRPLUS 0xFF00EE03 +#define PVFS2_VFS_OP_FILE_IOX 0xFF00EE04 + +/* + * Misc constants. Please retain them as multiples of 8! + * Otherwise 32-64 bit interactions will be messed up :) + */ +#define PVFS2_NAME_LEN 0x00000100 +#define PVFS2_MAX_DEBUG_STRING_LEN 0x00000400 +#define PVFS2_MAX_DEBUG_ARRAY_LEN 0x00000800 + +/* + * MAX_DIRENT_COUNT cannot be larger than PVFS_REQ_LIMIT_LISTATTR. + * The value of PVFS_REQ_LIMIT_LISTATTR has been changed from 113 to 60 + * to accomodate an attribute object with mirrored handles. + * MAX_DIRENT_COUNT is replaced by MAX_DIRENT_COUNT_READDIR and + * MAX_DIRENT_COUNT_READDIRPLUS, since readdir doesn't trigger a listattr + * but readdirplus might. +*/ +#define MAX_DIRENT_COUNT_READDIR 0x00000060 +#define MAX_DIRENT_COUNT_READDIRPLUS 0x0000003C + +#include "upcall.h" +#include "downcall.h" + +/* + * These macros differ from proto macros in that they don't do any + * byte-swappings and are used to ensure that kernel-clientcore interactions + * don't cause any unaligned accesses etc on 64 bit machines + */ +#ifndef roundup4 +#define roundup4(x) (((x)+3) & ~3) +#endif + +#ifndef roundup8 +#define roundup8(x) (((x)+7) & ~7) +#endif + +/* strings; decoding just points into existing character data */ +#define enc_string(pptr, pbuf) do { \ + __u32 len = strlen(*pbuf); \ + *(__u32 *) *(pptr) = (len); \ + memcpy(*(pptr)+4, *pbuf, len+1); \ + *(pptr) += roundup8(4 + len + 1); \ +} while (0) + +#define dec_string(pptr, pbuf, plen) do { \ + __u32 len = (*(__u32 *) *(pptr)); \ + *pbuf = *(pptr) + 4; \ + *(pptr) += roundup8(4 + len + 1); \ + if (plen) \ + *plen = len;\ +} while (0) + +struct read_write_x { + __s64 off; + __s64 len; +}; + +#endif diff --git a/fs/orangefs/pvfs2-kernel.h b/fs/orangefs/pvfs2-kernel.h new file mode 100644 index 000000000000..29b4a48b3a25 --- /dev/null +++ b/fs/orangefs/pvfs2-kernel.h @@ -0,0 +1,861 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +/* + * The PVFS2 Linux kernel support allows PVFS2 volumes to be mounted and + * accessed through the Linux VFS (i.e. using standard I/O system calls). + * This support is only needed on clients that wish to mount the file system. + * + */ + +/* + * Declarations and macros for the PVFS2 Linux kernel support. + */ + +#ifndef __PVFS2KERNEL_H +#define __PVFS2KERNEL_H + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +#include "pvfs2-dev-proto.h" + +#ifdef PVFS2_KERNEL_DEBUG +#define PVFS2_DEFAULT_OP_TIMEOUT_SECS 10 +#else +#define PVFS2_DEFAULT_OP_TIMEOUT_SECS 20 +#endif + +#define PVFS2_BUFMAP_WAIT_TIMEOUT_SECS 30 + +#define PVFS2_DEFAULT_SLOT_TIMEOUT_SECS 900 /* 15 minutes */ + +#define PVFS2_REQDEVICE_NAME "pvfs2-req" + +#define PVFS2_DEVREQ_MAGIC 0x20030529 +#define PVFS2_LINK_MAX 0x000000FF +#define PVFS2_PURGE_RETRY_COUNT 0x00000005 +#define PVFS2_SEEK_END 0x00000002 +#define PVFS2_MAX_NUM_OPTIONS 0x00000004 +#define PVFS2_MAX_MOUNT_OPT_LEN 0x00000080 +#define PVFS2_MAX_FSKEY_LEN 64 + +#define MAX_DEV_REQ_UPSIZE (2*sizeof(__s32) + \ +sizeof(__u64) + sizeof(struct pvfs2_upcall_s)) +#define MAX_DEV_REQ_DOWNSIZE (2*sizeof(__s32) + \ +sizeof(__u64) + sizeof(struct pvfs2_downcall_s)) + +#define BITS_PER_LONG_DIV_8 (BITS_PER_LONG >> 3) + +/* borrowed from irda.h */ +#ifndef MSECS_TO_JIFFIES +#define MSECS_TO_JIFFIES(ms) (((ms)*HZ+999)/1000) +#endif + +#define MAX_ALIGNED_DEV_REQ_UPSIZE \ + (MAX_DEV_REQ_UPSIZE + \ + ((((MAX_DEV_REQ_UPSIZE / \ + (BITS_PER_LONG_DIV_8)) * \ + (BITS_PER_LONG_DIV_8)) + \ + (BITS_PER_LONG_DIV_8)) - \ + MAX_DEV_REQ_UPSIZE)) + +#define MAX_ALIGNED_DEV_REQ_DOWNSIZE \ + (MAX_DEV_REQ_DOWNSIZE + \ + ((((MAX_DEV_REQ_DOWNSIZE / \ + (BITS_PER_LONG_DIV_8)) * \ + (BITS_PER_LONG_DIV_8)) + \ + (BITS_PER_LONG_DIV_8)) - \ + MAX_DEV_REQ_DOWNSIZE)) + +/* + * valid pvfs2 kernel operation states + * + * unknown - op was just initialized + * waiting - op is on request_list (upward bound) + * inprogr - op is in progress (waiting for downcall) + * serviced - op has matching downcall; ok + * purged - op has to start a timer since client-core + * exited uncleanly before servicing op + */ +enum pvfs2_vfs_op_states { + OP_VFS_STATE_UNKNOWN = 0, + OP_VFS_STATE_WAITING = 1, + OP_VFS_STATE_INPROGR = 2, + OP_VFS_STATE_SERVICED = 4, + OP_VFS_STATE_PURGED = 8, +}; + +#define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING) +#define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR) +#define set_op_state_serviced(op) ((op)->op_state = OP_VFS_STATE_SERVICED) +#define set_op_state_purged(op) ((op)->op_state |= OP_VFS_STATE_PURGED) + +#define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING) +#define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR) +#define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED) +#define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED) + +#define get_op(op) \ + do { \ + atomic_inc(&(op)->aio_ref_count); \ + gossip_debug(GOSSIP_DEV_DEBUG, \ + "(get) Alloced OP (%p:%llu)\n", \ + op, \ + llu((op)->tag)); \ + } while (0) + +#define put_op(op) \ + do { \ + if (atomic_sub_and_test(1, &(op)->aio_ref_count) == 1) { \ + gossip_debug(GOSSIP_DEV_DEBUG, \ + "(put) Releasing OP (%p:%llu)\n", \ + op, \ + llu((op)->tag)); \ + op_release(op); \ + } \ + } while (0) + +#define op_wait(op) (atomic_read(&(op)->aio_ref_count) <= 2 ? 0 : 1) + +/* + * Defines for controlling whether I/O upcalls are for async or sync operations + */ +enum PVFS_async_io_type { + PVFS_VFS_SYNC_IO = 0, + PVFS_VFS_ASYNC_IO = 1, +}; + +/* + * An array of client_debug_mask will be built to hold debug keyword/mask + * values fetched from userspace. + */ +struct client_debug_mask { + char *keyword; + __u64 mask1; + __u64 mask2; +}; + +/* + * pvfs2 kernel memory related flags + */ + +#if ((defined PVFS2_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) +#define PVFS2_CACHE_CREATE_FLAGS SLAB_RED_ZONE +#else +#define PVFS2_CACHE_CREATE_FLAGS 0 +#endif /* ((defined PVFS2_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) */ + +#define PVFS2_CACHE_ALLOC_FLAGS (GFP_KERNEL) +#define PVFS2_GFP_FLAGS (GFP_KERNEL) +#define PVFS2_BUFMAP_GFP_FLAGS (GFP_KERNEL) + +#define pvfs2_kmap(page) kmap(page) +#define pvfs2_kunmap(page) kunmap(page) + +/* pvfs2 xattr and acl related defines */ +#define PVFS2_XATTR_INDEX_POSIX_ACL_ACCESS 1 +#define PVFS2_XATTR_INDEX_POSIX_ACL_DEFAULT 2 +#define PVFS2_XATTR_INDEX_TRUSTED 3 +#define PVFS2_XATTR_INDEX_DEFAULT 4 + +#if 0 +#ifndef POSIX_ACL_XATTR_ACCESS +#define POSIX_ACL_XATTR_ACCESS "system.posix_acl_access" +#endif +#ifndef POSIX_ACL_XATTR_DEFAULT +#define POSIX_ACL_XATTR_DEFAULT "system.posix_acl_default" +#endif +#endif + +#define PVFS2_XATTR_NAME_ACL_ACCESS POSIX_ACL_XATTR_ACCESS +#define PVFS2_XATTR_NAME_ACL_DEFAULT POSIX_ACL_XATTR_DEFAULT +#define PVFS2_XATTR_NAME_TRUSTED_PREFIX "trusted." +#define PVFS2_XATTR_NAME_DEFAULT_PREFIX "" + +/* these functions are defined in pvfs2-utils.c */ +int orangefs_prepare_cdm_array(char *debug_array_string); +int orangefs_prepare_debugfs_help_string(int); + +/* defined in pvfs2-debugfs.c */ +int pvfs2_client_debug_init(void); + +void debug_string_to_mask(char *, void *, int); +void do_c_mask(int, char *, struct client_debug_mask **); +void do_k_mask(int, char *, __u64 **); + +void debug_mask_to_string(void *, int); +void do_k_string(void *, int); +void do_c_string(void *, int); +int check_amalgam_keyword(void *, int); +int keyword_is_amalgam(char *); + +/*these variables are defined in pvfs2-mod.c */ +extern char kernel_debug_string[PVFS2_MAX_DEBUG_STRING_LEN]; +extern char client_debug_string[PVFS2_MAX_DEBUG_STRING_LEN]; +extern char client_debug_array_string[PVFS2_MAX_DEBUG_STRING_LEN]; +extern unsigned int kernel_mask_set_mod_init; + +extern int pvfs2_init_acl(struct inode *inode, struct inode *dir); +extern const struct xattr_handler *pvfs2_xattr_handlers[]; + +extern struct posix_acl *pvfs2_get_acl(struct inode *inode, int type); +extern int pvfs2_set_acl(struct inode *inode, struct posix_acl *acl, int type); + +int pvfs2_xattr_set_default(struct dentry *dentry, + const char *name, + const void *buffer, + size_t size, + int flags, + int handler_flags); + +int pvfs2_xattr_get_default(struct dentry *dentry, + const char *name, + void *buffer, + size_t size, + int handler_flags); + +/* + * Redefine xtvec structure so that we could move helper functions out of + * the define + */ +struct xtvec { + __kernel_off_t xtv_off; /* must be off_t */ + __kernel_size_t xtv_len; /* must be size_t */ +}; + +/* + * pvfs2 data structures + */ +struct pvfs2_kernel_op_s { + enum pvfs2_vfs_op_states op_state; + __u64 tag; + + /* + * Set uses_shared_memory to 1 if this operation uses shared memory. + * If true, then a retry on the op must also get a new shared memory + * buffer and re-populate it. + */ + int uses_shared_memory; + + struct pvfs2_upcall_s upcall; + struct pvfs2_downcall_s downcall; + + wait_queue_head_t waitq; + spinlock_t lock; + + int io_completed; + wait_queue_head_t io_completion_waitq; + + /* + * upcalls requiring variable length trailers require that this struct + * be in the request list even after client-core does a read() on the + * device to dequeue the upcall. + * if op_linger field goes to 0, we dequeue this op off the list. + * else we let it stay. What gets passed to the read() is + * a) if op_linger field is = 1, pvfs2_kernel_op_s itself + * b) else if = 0, we pass ->upcall.trailer_buf + * We expect to have only a single upcall trailer buffer, + * so we expect callers with trailers + * to set this field to 2 and others to set it to 1. + */ + __s32 op_linger, op_linger_tmp; + /* VFS aio fields */ + + /* used by the async I/O code to stash the pvfs2_kiocb_s structure */ + void *priv; + + /* used again for the async I/O code for deallocation */ + atomic_t aio_ref_count; + + int attempts; + + struct list_head list; +}; + +/* per inode private pvfs2 info */ +struct pvfs2_inode_s { + struct pvfs2_object_kref refn; + char link_target[PVFS_NAME_MAX]; + __s64 blksize; + /* + * Reading/Writing Extended attributes need to acquire the appropriate + * reader/writer semaphore on the pvfs2_inode_s structure. + */ + struct rw_semaphore xattr_sem; + + struct inode vfs_inode; + sector_t last_failed_block_index_read; + + /* + * State of in-memory attributes not yet flushed to disk associated + * with this object + */ + unsigned long pinode_flags; + + /* All allocated pvfs2_inode_s objects are chained to a list */ + struct list_head list; +}; + +#define P_ATIME_FLAG 0 +#define P_MTIME_FLAG 1 +#define P_CTIME_FLAG 2 +#define P_MODE_FLAG 3 + +#define ClearAtimeFlag(pinode) clear_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) +#define SetAtimeFlag(pinode) set_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) +#define AtimeFlag(pinode) test_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) + +#define ClearMtimeFlag(pinode) clear_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) +#define SetMtimeFlag(pinode) set_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) +#define MtimeFlag(pinode) test_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) + +#define ClearCtimeFlag(pinode) clear_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) +#define SetCtimeFlag(pinode) set_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) +#define CtimeFlag(pinode) test_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) + +#define ClearModeFlag(pinode) clear_bit(P_MODE_FLAG, &(pinode)->pinode_flags) +#define SetModeFlag(pinode) set_bit(P_MODE_FLAG, &(pinode)->pinode_flags) +#define ModeFlag(pinode) test_bit(P_MODE_FLAG, &(pinode)->pinode_flags) + +/* per superblock private pvfs2 info */ +struct pvfs2_sb_info_s { + struct pvfs2_khandle root_khandle; + __s32 fs_id; + int id; + int flags; +#define PVFS2_OPT_INTR 0x01 +#define PVFS2_OPT_LOCAL_LOCK 0x02 + char devname[PVFS_MAX_SERVER_ADDR_LEN]; + struct super_block *sb; + int mount_pending; + struct list_head list; +}; + +/* + * a temporary structure used only for sb mount time that groups the + * mount time data provided along with a private superblock structure + * that is allocated before a 'kernel' superblock is allocated. +*/ +struct pvfs2_mount_sb_info_s { + void *data; + struct pvfs2_khandle root_khandle; + __s32 fs_id; + int id; +}; + +/* + * structure that holds the state of any async I/O operation issued + * through the VFS. Needed especially to handle cancellation requests + * or even completion notification so that the VFS client-side daemon + * can free up its vfs_request slots. + */ +struct pvfs2_kiocb_s { + /* the pointer to the task that initiated the AIO */ + struct task_struct *tsk; + + /* pointer to the kiocb that kicked this operation */ + struct kiocb *kiocb; + + /* buffer index that was used for the I/O */ + struct pvfs2_bufmap *bufmap; + int buffer_index; + + /* pvfs2 kernel operation type */ + struct pvfs2_kernel_op_s *op; + + /* The user space buffers from/to which I/O is being staged */ + struct iovec *iov; + + /* number of elements in the iovector */ + unsigned long nr_segs; + + /* set to indicate the type of the operation */ + int rw; + + /* file offset */ + loff_t offset; + + /* and the count in bytes */ + size_t bytes_to_be_copied; + + ssize_t bytes_copied; + int needs_cleanup; +}; + +struct pvfs2_stats { + unsigned long cache_hits; + unsigned long cache_misses; + unsigned long reads; + unsigned long writes; +}; + +extern struct pvfs2_stats g_pvfs2_stats; + +/* + * NOTE: See Documentation/filesystems/porting for information + * on implementing FOO_I and properly accessing fs private data + */ +static inline struct pvfs2_inode_s *PVFS2_I(struct inode *inode) +{ + return container_of(inode, struct pvfs2_inode_s, vfs_inode); +} + +static inline struct pvfs2_sb_info_s *PVFS2_SB(struct super_block *sb) +{ + return (struct pvfs2_sb_info_s *) sb->s_fs_info; +} + +/* ino_t descends from "unsigned long", 8 bytes, 64 bits. */ +static inline ino_t pvfs2_khandle_to_ino(struct pvfs2_khandle *khandle) +{ + union { + unsigned char u[8]; + __u64 ino; + } ihandle; + + ihandle.u[0] = khandle->u[0] ^ khandle->u[4]; + ihandle.u[1] = khandle->u[1] ^ khandle->u[5]; + ihandle.u[2] = khandle->u[2] ^ khandle->u[6]; + ihandle.u[3] = khandle->u[3] ^ khandle->u[7]; + ihandle.u[4] = khandle->u[12] ^ khandle->u[8]; + ihandle.u[5] = khandle->u[13] ^ khandle->u[9]; + ihandle.u[6] = khandle->u[14] ^ khandle->u[10]; + ihandle.u[7] = khandle->u[15] ^ khandle->u[11]; + + return ihandle.ino; +} + +static inline struct pvfs2_khandle *get_khandle_from_ino(struct inode *inode) +{ + return &(PVFS2_I(inode)->refn.khandle); +} + +static inline __s32 get_fsid_from_ino(struct inode *inode) +{ + return PVFS2_I(inode)->refn.fs_id; +} + +static inline ino_t get_ino_from_khandle(struct inode *inode) +{ + struct pvfs2_khandle *khandle; + ino_t ino; + + khandle = get_khandle_from_ino(inode); + ino = pvfs2_khandle_to_ino(khandle); + return ino; +} + +static inline ino_t get_parent_ino_from_dentry(struct dentry *dentry) +{ + return get_ino_from_khandle(dentry->d_parent->d_inode); +} + +static inline int is_root_handle(struct inode *inode) +{ + gossip_debug(GOSSIP_DCACHE_DEBUG, + "%s: root handle: %pU, this handle: %pU:\n", + __func__, + &PVFS2_SB(inode->i_sb)->root_khandle, + get_khandle_from_ino(inode)); + + if (PVFS_khandle_cmp(&(PVFS2_SB(inode->i_sb)->root_khandle), + get_khandle_from_ino(inode))) + return 0; + else + return 1; +} + +static inline int match_handle(struct pvfs2_khandle resp_handle, + struct inode *inode) +{ + gossip_debug(GOSSIP_DCACHE_DEBUG, + "%s: one handle: %pU, another handle:%pU:\n", + __func__, + &resp_handle, + get_khandle_from_ino(inode)); + + if (PVFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode))) + return 0; + else + return 1; +} + +/* + * defined in pvfs2-cache.c + */ +int op_cache_initialize(void); +int op_cache_finalize(void); +struct pvfs2_kernel_op_s *op_alloc(__s32 type); +struct pvfs2_kernel_op_s *op_alloc_trailer(__s32 type); +char *get_opname_string(struct pvfs2_kernel_op_s *new_op); +void op_release(struct pvfs2_kernel_op_s *op); + +int dev_req_cache_initialize(void); +int dev_req_cache_finalize(void); +void *dev_req_alloc(void); +void dev_req_release(void *); + +int pvfs2_inode_cache_initialize(void); +int pvfs2_inode_cache_finalize(void); + +int kiocb_cache_initialize(void); +int kiocb_cache_finalize(void); +struct pvfs2_kiocb_s *kiocb_alloc(void); +void kiocb_release(struct pvfs2_kiocb_s *ptr); + +/* + * defined in pvfs2-mod.c + */ +void purge_inprogress_ops(void); + +/* + * defined in waitqueue.c + */ +int wait_for_matching_downcall(struct pvfs2_kernel_op_s *op); +int wait_for_cancellation_downcall(struct pvfs2_kernel_op_s *op); +void pvfs2_clean_up_interrupted_operation(struct pvfs2_kernel_op_s *op); +void purge_waiting_ops(void); + +/* + * defined in super.c + */ +struct dentry *pvfs2_mount(struct file_system_type *fst, + int flags, + const char *devname, + void *data); + +void pvfs2_kill_sb(struct super_block *sb); +int pvfs2_remount(struct super_block *sb); + +int fsid_key_table_initialize(void); +void fsid_key_table_finalize(void); + +/* + * defined in inode.c + */ +__u32 convert_to_pvfs2_mask(unsigned long lite_mask); +struct inode *pvfs2_new_inode(struct super_block *sb, + struct inode *dir, + int mode, + dev_t dev, + struct pvfs2_object_kref *ref); + +int pvfs2_setattr(struct dentry *dentry, struct iattr *iattr); + +int pvfs2_getattr(struct vfsmount *mnt, + struct dentry *dentry, + struct kstat *kstat); + +/* + * defined in xattr.c + */ +int pvfs2_setxattr(struct dentry *dentry, + const char *name, + const void *value, + size_t size, + int flags); + +ssize_t pvfs2_getxattr(struct dentry *dentry, + const char *name, + void *buffer, + size_t size); + +ssize_t pvfs2_listxattr(struct dentry *dentry, char *buffer, size_t size); + +/* + * defined in namei.c + */ +struct inode *pvfs2_iget(struct super_block *sb, + struct pvfs2_object_kref *ref); + +ssize_t pvfs2_inode_read(struct inode *inode, + char __user *buf, + size_t count, + loff_t *offset, + loff_t readahead_size); + +/* + * defined in devpvfs2-req.c + */ +int pvfs2_dev_init(void); +void pvfs2_dev_cleanup(void); +int is_daemon_in_service(void); +int fs_mount_pending(__s32 fsid); + +/* + * defined in pvfs2-utils.c + */ +__s32 fsid_of_op(struct pvfs2_kernel_op_s *op); + +int pvfs2_flush_inode(struct inode *inode); + +ssize_t pvfs2_inode_getxattr(struct inode *inode, + const char *prefix, + const char *name, + void *buffer, + size_t size); + +int pvfs2_inode_setxattr(struct inode *inode, + const char *prefix, + const char *name, + const void *value, + size_t size, + int flags); + +int pvfs2_inode_getattr(struct inode *inode, __u32 mask); + +int pvfs2_inode_setattr(struct inode *inode, struct iattr *iattr); + +void pvfs2_op_initialize(struct pvfs2_kernel_op_s *op); + +void pvfs2_make_bad_inode(struct inode *inode); + +void block_signals(sigset_t *); + +void set_signals(sigset_t *); + +int pvfs2_unmount_sb(struct super_block *sb); + +int pvfs2_cancel_op_in_progress(__u64 tag); + +__u64 pvfs2_convert_time_field(void *time_ptr); + +int pvfs2_normalize_to_errno(__s32 error_code); + +extern struct mutex devreq_mutex; +extern struct mutex request_mutex; +extern int debug; +extern int op_timeout_secs; +extern int slot_timeout_secs; +extern struct list_head pvfs2_superblocks; +extern spinlock_t pvfs2_superblocks_lock; +extern struct list_head pvfs2_request_list; +extern spinlock_t pvfs2_request_list_lock; +extern wait_queue_head_t pvfs2_request_list_waitq; +extern struct list_head *htable_ops_in_progress; +extern spinlock_t htable_ops_in_progress_lock; +extern int hash_table_size; + +extern const struct address_space_operations pvfs2_address_operations; +extern struct backing_dev_info pvfs2_backing_dev_info; +extern struct inode_operations pvfs2_file_inode_operations; +extern const struct file_operations pvfs2_file_operations; +extern struct inode_operations pvfs2_symlink_inode_operations; +extern struct inode_operations pvfs2_dir_inode_operations; +extern const struct file_operations pvfs2_dir_operations; +extern const struct dentry_operations pvfs2_dentry_operations; +extern const struct file_operations pvfs2_devreq_file_operations; + +extern wait_queue_head_t pvfs2_bufmap_init_waitq; + +/* + * misc convenience macros + */ +#define add_op_to_request_list(op) \ +do { \ + spin_lock(&pvfs2_request_list_lock); \ + spin_lock(&op->lock); \ + set_op_state_waiting(op); \ + list_add_tail(&op->list, &pvfs2_request_list); \ + spin_unlock(&pvfs2_request_list_lock); \ + spin_unlock(&op->lock); \ + wake_up_interruptible(&pvfs2_request_list_waitq); \ +} while (0) + +#define add_priority_op_to_request_list(op) \ + do { \ + spin_lock(&pvfs2_request_list_lock); \ + spin_lock(&op->lock); \ + set_op_state_waiting(op); \ + \ + list_add(&op->list, &pvfs2_request_list); \ + spin_unlock(&pvfs2_request_list_lock); \ + spin_unlock(&op->lock); \ + wake_up_interruptible(&pvfs2_request_list_waitq); \ +} while (0) + +#define remove_op_from_request_list(op) \ + do { \ + struct list_head *tmp = NULL; \ + struct list_head *tmp_safe = NULL; \ + struct pvfs2_kernel_op_s *tmp_op = NULL; \ + \ + spin_lock(&pvfs2_request_list_lock); \ + list_for_each_safe(tmp, tmp_safe, &pvfs2_request_list) { \ + tmp_op = list_entry(tmp, \ + struct pvfs2_kernel_op_s, \ + list); \ + if (tmp_op && (tmp_op == op)) { \ + list_del(&tmp_op->list); \ + break; \ + } \ + } \ + spin_unlock(&pvfs2_request_list_lock); \ + } while (0) + +#define PVFS2_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */ +#define PVFS2_OP_PRIORITY 2 /* service_operation() is high priority */ +#define PVFS2_OP_CANCELLATION 4 /* this is a cancellation */ +#define PVFS2_OP_NO_SEMAPHORE 8 /* don't acquire semaphore */ +#define PVFS2_OP_ASYNC 16 /* Queue it, but don't wait */ + +int service_operation(struct pvfs2_kernel_op_s *op, + const char *op_name, + int flags); + +/* + * handles two possible error cases, depending on context. + * + * by design, our vfs i/o errors need to be handled in one of two ways, + * depending on where the error occured. + * + * if the error happens in the waitqueue code because we either timed + * out or a signal was raised while waiting, we need to cancel the + * userspace i/o operation and free the op manually. this is done to + * avoid having the device start writing application data to our shared + * bufmap pages without us expecting it. + * + * FIXME: POSSIBLE OPTIMIZATION: + * However, if we timed out or if we got a signal AND our upcall was never + * picked off the queue (i.e. we were in OP_VFS_STATE_WAITING), then we don't + * need to send a cancellation upcall. The way we can handle this is + * set error_exit to 2 in such cases and 1 whenever cancellation has to be + * sent and have handle_error + * take care of this situation as well.. + * + * if a pvfs2 sysint level error occured and i/o has been completed, + * there is no need to cancel the operation, as the user has finished + * using the bufmap page and so there is no danger in this case. in + * this case, we wake up the device normally so that it may free the + * op, as normal. + * + * note the only reason this is a macro is because both read and write + * cases need the exact same handling code. + */ +#define handle_io_error() \ +do { \ + if (!op_state_serviced(new_op)) { \ + pvfs2_cancel_op_in_progress(new_op->tag); \ + op_release(new_op); \ + } else { \ + wake_up_daemon_for_return(new_op); \ + } \ + new_op = NULL; \ + pvfs_bufmap_put(bufmap, buffer_index); \ + buffer_index = -1; \ +} while (0) + +#define get_interruptible_flag(inode) \ + ((PVFS2_SB(inode->i_sb)->flags & PVFS2_OPT_INTR) ? \ + PVFS2_OP_INTERRUPTIBLE : 0) + +#define add_pvfs2_sb(sb) \ +do { \ + gossip_debug(GOSSIP_SUPER_DEBUG, \ + "Adding SB %p to pvfs2 superblocks\n", \ + PVFS2_SB(sb)); \ + spin_lock(&pvfs2_superblocks_lock); \ + list_add_tail(&PVFS2_SB(sb)->list, &pvfs2_superblocks); \ + spin_unlock(&pvfs2_superblocks_lock); \ +} while (0) + +#define remove_pvfs2_sb(sb) \ +do { \ + struct list_head *tmp = NULL; \ + struct list_head *tmp_safe = NULL; \ + struct pvfs2_sb_info_s *pvfs2_sb = NULL; \ + \ + spin_lock(&pvfs2_superblocks_lock); \ + list_for_each_safe(tmp, tmp_safe, &pvfs2_superblocks) { \ + pvfs2_sb = list_entry(tmp, \ + struct pvfs2_sb_info_s, \ + list); \ + if (pvfs2_sb && (pvfs2_sb->sb == sb)) { \ + gossip_debug(GOSSIP_SUPER_DEBUG, \ + "Removing SB %p from pvfs2 superblocks\n", \ + pvfs2_sb); \ + list_del(&pvfs2_sb->list); \ + break; \ + } \ + } \ + spin_unlock(&pvfs2_superblocks_lock); \ +} while (0) + +#define pvfs2_lock_inode(inode) spin_lock(&inode->i_lock) +#define pvfs2_unlock_inode(inode) spin_unlock(&inode->i_lock) +#define pvfs2_current_signal_lock current->sighand->siglock +#define pvfs2_current_sigaction current->sighand->action + +#define fill_default_sys_attrs(sys_attr, type, mode) \ +do { \ + sys_attr.owner = from_kuid(current_user_ns(), current_fsuid()); \ + sys_attr.group = from_kgid(current_user_ns(), current_fsgid()); \ + sys_attr.size = 0; \ + sys_attr.perms = PVFS_util_translate_mode(mode); \ + sys_attr.objtype = type; \ + sys_attr.mask = PVFS_ATTR_SYS_ALL_SETABLE; \ +} while (0) + +#define pvfs2_inode_lock(__i) mutex_lock(&(__i)->i_mutex) + +#define pvfs2_inode_unlock(__i) mutex_unlock(&(__i)->i_mutex) + +static inline void pvfs2_i_size_write(struct inode *inode, loff_t i_size) +{ +#if BITS_PER_LONG == 32 && defined(CONFIG_SMP) + pvfs2_inode_lock(inode); +#endif + i_size_write(inode, i_size); +#if BITS_PER_LONG == 32 && defined(CONFIG_SMP) + pvfs2_inode_unlock(inode); +#endif +} + +static inline unsigned int diff(struct timeval *end, struct timeval *begin) +{ + if (end->tv_usec < begin->tv_usec) { + end->tv_usec += 1000000; + end->tv_sec--; + } + end->tv_sec -= begin->tv_sec; + end->tv_usec -= begin->tv_usec; + return (end->tv_sec * 1000000) + end->tv_usec; +} + +#endif /* __PVFS2KERNEL_H */ diff --git a/fs/orangefs/pvfs2-mod.c b/fs/orangefs/pvfs2-mod.c new file mode 100644 index 000000000000..d848c90413d1 --- /dev/null +++ b/fs/orangefs/pvfs2-mod.c @@ -0,0 +1,315 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * Changes by Acxiom Corporation to add proc file handler for pvfs2 client + * parameters, Copyright Acxiom Corporation, 2005. + * + * See COPYING in top-level directory. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-debugfs.h" +#include "pvfs2-sysfs.h" + +/* PVFS2_VERSION is a ./configure define */ +#ifndef PVFS2_VERSION +#define PVFS2_VERSION "Unknown" +#endif + +/* + * global variables declared here + */ + +/* array of client debug keyword/mask values */ +struct client_debug_mask *cdm_array; +int cdm_element_count; + +char kernel_debug_string[PVFS2_MAX_DEBUG_STRING_LEN] = "none"; +char client_debug_string[PVFS2_MAX_DEBUG_STRING_LEN]; +char client_debug_array_string[PVFS2_MAX_DEBUG_STRING_LEN]; + +char *debug_help_string; +int help_string_initialized; +struct dentry *help_file_dentry; +struct dentry *client_debug_dentry; +struct dentry *debug_dir; +int client_verbose_index; +int client_all_index; +struct pvfs2_stats g_pvfs2_stats; + +/* the size of the hash tables for ops in progress */ +int hash_table_size = 509; + +static ulong module_parm_debug_mask; +__u64 gossip_debug_mask; +struct client_debug_mask client_debug_mask = { NULL, 0, 0 }; +unsigned int kernel_mask_set_mod_init; /* implicitly false */ +int op_timeout_secs = PVFS2_DEFAULT_OP_TIMEOUT_SECS; +int slot_timeout_secs = PVFS2_DEFAULT_SLOT_TIMEOUT_SECS; + +MODULE_LICENSE("GPL"); +MODULE_AUTHOR("PVFS2 Development Team"); +MODULE_DESCRIPTION("The Linux Kernel VFS interface to PVFS2"); +MODULE_PARM_DESC(module_parm_debug_mask, "debugging level (see pvfs2-debug.h for values)"); +MODULE_PARM_DESC(op_timeout_secs, "Operation timeout in seconds"); +MODULE_PARM_DESC(slot_timeout_secs, "Slot timeout in seconds"); +MODULE_PARM_DESC(hash_table_size, + "size of hash table for operations in progress"); + +static struct file_system_type pvfs2_fs_type = { + .name = "pvfs2", + .mount = pvfs2_mount, + .kill_sb = pvfs2_kill_sb, + .owner = THIS_MODULE, +}; + +module_param(hash_table_size, int, 0); +module_param(module_parm_debug_mask, ulong, 0644); +module_param(op_timeout_secs, int, 0); +module_param(slot_timeout_secs, int, 0); + +/* synchronizes the request device file */ +struct mutex devreq_mutex; + +/* + * Blocks non-priority requests from being queued for servicing. This + * could be used for protecting the request list data structure, but + * for now it's only being used to stall the op addition to the request + * list + */ +struct mutex request_mutex; + +/* hash table for storing operations waiting for matching downcall */ +struct list_head *htable_ops_in_progress; +DEFINE_SPINLOCK(htable_ops_in_progress_lock); + +/* list for queueing upcall operations */ +LIST_HEAD(pvfs2_request_list); + +/* used to protect the above pvfs2_request_list */ +DEFINE_SPINLOCK(pvfs2_request_list_lock); + +/* used for incoming request notification */ +DECLARE_WAIT_QUEUE_HEAD(pvfs2_request_list_waitq); + +static int __init pvfs2_init(void) +{ + int ret = -1; + __u32 i = 0; + + /* convert input debug mask to a 64-bit unsigned integer */ + gossip_debug_mask = (unsigned long long) module_parm_debug_mask; + + /* + * set the kernel's gossip debug string; invalid mask values will + * be ignored. + */ + debug_mask_to_string(&gossip_debug_mask, 0); + + /* remove any invalid values from the mask */ + debug_string_to_mask(kernel_debug_string, &gossip_debug_mask, 0); + + /* + * if the mask has a non-zero value, then indicate that the mask + * was set when the kernel module was loaded. The pvfs2 dev ioctl + * command will look at this boolean to determine if the kernel's + * debug mask should be overwritten when the client-core is started. + */ + if (gossip_debug_mask != 0) + kernel_mask_set_mod_init = true; + + /* print information message to the system log */ + pr_info("pvfs2: pvfs2_init called with debug mask: :%s: :%llx:\n", + kernel_debug_string, + (unsigned long long)gossip_debug_mask); + + ret = bdi_init(&pvfs2_backing_dev_info); + + if (ret) + return ret; + + if (op_timeout_secs < 0) + op_timeout_secs = 0; + + if (slot_timeout_secs < 0) + slot_timeout_secs = 0; + + /* initialize global book keeping data structures */ + ret = op_cache_initialize(); + if (ret < 0) + goto err; + + ret = dev_req_cache_initialize(); + if (ret < 0) + goto cleanup_op; + + ret = pvfs2_inode_cache_initialize(); + if (ret < 0) + goto cleanup_req; + + ret = kiocb_cache_initialize(); + if (ret < 0) + goto cleanup_inode; + + /* Initialize the pvfsdev subsystem. */ + ret = pvfs2_dev_init(); + if (ret < 0) { + gossip_err("pvfs2: could not initialize device subsystem %d!\n", + ret); + goto cleanup_kiocb; + } + + mutex_init(&devreq_mutex); + mutex_init(&request_mutex); + + htable_ops_in_progress = + kcalloc(hash_table_size, sizeof(struct list_head), GFP_KERNEL); + if (!htable_ops_in_progress) { + gossip_err("Failed to initialize op hashtable"); + ret = -ENOMEM; + goto cleanup_device; + } + + /* initialize a doubly linked at each hash table index */ + for (i = 0; i < hash_table_size; i++) + INIT_LIST_HEAD(&htable_ops_in_progress[i]); + + ret = fsid_key_table_initialize(); + if (ret < 0) + goto cleanup_progress_table; + + /* + * Build the contents of /sys/kernel/debug/orangefs/debug-help + * from the keywords in the kernel keyword/mask array. + * + * The keywords in the client keyword/mask array are + * unknown at boot time. + * + * orangefs_prepare_debugfs_help_string will be used again + * later to rebuild the debug-help file after the client starts + * and passes along the needed info. The argument signifies + * which time orangefs_prepare_debugfs_help_string is being + * called. + * + */ + ret = orangefs_prepare_debugfs_help_string(1); + if (ret) + goto out; + + pvfs2_debugfs_init(); + pvfs2_kernel_debug_init(); + orangefs_sysfs_init(); + + ret = register_filesystem(&pvfs2_fs_type); + if (ret == 0) { + pr_info("pvfs2: module version %s loaded\n", PVFS2_VERSION); + return 0; + } + + pvfs2_debugfs_cleanup(); + orangefs_sysfs_exit(); + fsid_key_table_finalize(); + +cleanup_progress_table: + kfree(htable_ops_in_progress); + +cleanup_device: + pvfs2_dev_cleanup(); + +cleanup_kiocb: + kiocb_cache_finalize(); + +cleanup_inode: + pvfs2_inode_cache_finalize(); + +cleanup_req: + dev_req_cache_finalize(); + +cleanup_op: + op_cache_finalize(); + +err: + bdi_destroy(&pvfs2_backing_dev_info); + +out: + return ret; +} + +static void __exit pvfs2_exit(void) +{ + int i = 0; + struct pvfs2_kernel_op_s *cur_op = NULL; + + gossip_debug(GOSSIP_INIT_DEBUG, "pvfs2: pvfs2_exit called\n"); + + unregister_filesystem(&pvfs2_fs_type); + pvfs2_debugfs_cleanup(); + orangefs_sysfs_exit(); + fsid_key_table_finalize(); + pvfs2_dev_cleanup(); + /* clear out all pending upcall op requests */ + spin_lock(&pvfs2_request_list_lock); + while (!list_empty(&pvfs2_request_list)) { + cur_op = list_entry(pvfs2_request_list.next, + struct pvfs2_kernel_op_s, + list); + list_del(&cur_op->list); + gossip_debug(GOSSIP_INIT_DEBUG, + "Freeing unhandled upcall request type %d\n", + cur_op->upcall.type); + op_release(cur_op); + } + spin_unlock(&pvfs2_request_list_lock); + + for (i = 0; i < hash_table_size; i++) + while (!list_empty(&htable_ops_in_progress[i])) { + cur_op = list_entry(htable_ops_in_progress[i].next, + struct pvfs2_kernel_op_s, + list); + op_release(cur_op); + } + + kiocb_cache_finalize(); + pvfs2_inode_cache_finalize(); + dev_req_cache_finalize(); + op_cache_finalize(); + + kfree(htable_ops_in_progress); + + bdi_destroy(&pvfs2_backing_dev_info); + + pr_info("pvfs2: module version %s unloaded\n", PVFS2_VERSION); +} + +/* + * What we do in this function is to walk the list of operations + * that are in progress in the hash table and mark them as purged as well. + */ +void purge_inprogress_ops(void) +{ + int i; + + for (i = 0; i < hash_table_size; i++) { + struct pvfs2_kernel_op_s *op; + struct pvfs2_kernel_op_s *next; + + list_for_each_entry_safe(op, + next, + &htable_ops_in_progress[i], + list) { + spin_lock(&op->lock); + gossip_debug(GOSSIP_INIT_DEBUG, + "pvfs2-client-core: purging in-progress op tag " + "%llu %s\n", + llu(op->tag), + get_opname_string(op)); + set_op_state_purged(op); + spin_unlock(&op->lock); + wake_up_interruptible(&op->waitq); + } + } +} + +module_init(pvfs2_init); +module_exit(pvfs2_exit); diff --git a/fs/orangefs/pvfs2-sysfs.c b/fs/orangefs/pvfs2-sysfs.c new file mode 100644 index 000000000000..ea635b5e431b --- /dev/null +++ b/fs/orangefs/pvfs2-sysfs.c @@ -0,0 +1,1787 @@ +/* + * Documentation/ABI/stable/orangefs-sysfs: + * + * What: /sys/fs/orangefs/perf_counter_reset + * Date: June 2015 + * Contact: Mike Marshall + * Description: + * echo a 0 or a 1 into perf_counter_reset to + * reset all the counters in + * /sys/fs/orangefs/perf_counters + * except ones with PINT_PERF_PRESERVE set. + * + * + * What: /sys/fs/orangefs/perf_counters/... + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * Counters and settings for various caches. + * Read only. + * + * + * What: /sys/fs/orangefs/perf_time_interval_secs + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * Length of perf counter intervals in + * seconds. + * + * + * What: /sys/fs/orangefs/perf_history_size + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * The perf_counters cache statistics have N, or + * perf_history_size, samples. The default is + * one. + * + * Every perf_time_interval_secs the (first) + * samples are reset. + * + * If N is greater than one, the "current" set + * of samples is reset, and the samples from the + * other N-1 intervals remain available. + * + * + * What: /sys/fs/orangefs/op_timeout_secs + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * Service operation timeout in seconds. + * + * + * What: /sys/fs/orangefs/slot_timeout_secs + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * "Slot" timeout in seconds. A "slot" + * is an indexed buffer in the shared + * memory segment used for communication + * between the kernel module and userspace. + * Slots are requested and waited for, + * the wait times out after slot_timeout_secs. + * + * + * What: /sys/fs/orangefs/acache/... + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * Attribute cache configurable settings. + * + * + * What: /sys/fs/orangefs/ncache/... + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * Name cache configurable settings. + * + * + * What: /sys/fs/orangefs/capcache/... + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * Capability cache configurable settings. + * + * + * What: /sys/fs/orangefs/ccache/... + * Date: Jun 2015 + * Contact: Mike Marshall + * Description: + * Credential cache configurable settings. + * + */ + +#include +#include +#include +#include +#include +#include + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-sysfs.h" + +#define ORANGEFS_KOBJ_ID "orangefs" +#define ACACHE_KOBJ_ID "acache" +#define CAPCACHE_KOBJ_ID "capcache" +#define CCACHE_KOBJ_ID "ccache" +#define NCACHE_KOBJ_ID "ncache" +#define PC_KOBJ_ID "pc" +#define STATS_KOBJ_ID "stats" + +struct orangefs_obj { + struct kobject kobj; + int op_timeout_secs; + int perf_counter_reset; + int perf_history_size; + int perf_time_interval_secs; + int slot_timeout_secs; +}; + +struct acache_orangefs_obj { + struct kobject kobj; + int hard_limit; + int reclaim_percentage; + int soft_limit; + int timeout_msecs; +}; + +struct capcache_orangefs_obj { + struct kobject kobj; + int hard_limit; + int reclaim_percentage; + int soft_limit; + int timeout_secs; +}; + +struct ccache_orangefs_obj { + struct kobject kobj; + int hard_limit; + int reclaim_percentage; + int soft_limit; + int timeout_secs; +}; + +struct ncache_orangefs_obj { + struct kobject kobj; + int hard_limit; + int reclaim_percentage; + int soft_limit; + int timeout_msecs; +}; + +struct pc_orangefs_obj { + struct kobject kobj; + char *acache; + char *capcache; + char *ncache; +}; + +struct stats_orangefs_obj { + struct kobject kobj; + int reads; + int writes; +}; + +struct orangefs_attribute { + struct attribute attr; + ssize_t (*show)(struct orangefs_obj *orangefs_obj, + struct orangefs_attribute *attr, + char *buf); + ssize_t (*store)(struct orangefs_obj *orangefs_obj, + struct orangefs_attribute *attr, + const char *buf, + size_t count); +}; + +struct acache_orangefs_attribute { + struct attribute attr; + ssize_t (*show)(struct acache_orangefs_obj *acache_orangefs_obj, + struct acache_orangefs_attribute *attr, + char *buf); + ssize_t (*store)(struct acache_orangefs_obj *acache_orangefs_obj, + struct acache_orangefs_attribute *attr, + const char *buf, + size_t count); +}; + +struct capcache_orangefs_attribute { + struct attribute attr; + ssize_t (*show)(struct capcache_orangefs_obj *capcache_orangefs_obj, + struct capcache_orangefs_attribute *attr, + char *buf); + ssize_t (*store)(struct capcache_orangefs_obj *capcache_orangefs_obj, + struct capcache_orangefs_attribute *attr, + const char *buf, + size_t count); +}; + +struct ccache_orangefs_attribute { + struct attribute attr; + ssize_t (*show)(struct ccache_orangefs_obj *ccache_orangefs_obj, + struct ccache_orangefs_attribute *attr, + char *buf); + ssize_t (*store)(struct ccache_orangefs_obj *ccache_orangefs_obj, + struct ccache_orangefs_attribute *attr, + const char *buf, + size_t count); +}; + +struct ncache_orangefs_attribute { + struct attribute attr; + ssize_t (*show)(struct ncache_orangefs_obj *ncache_orangefs_obj, + struct ncache_orangefs_attribute *attr, + char *buf); + ssize_t (*store)(struct ncache_orangefs_obj *ncache_orangefs_obj, + struct ncache_orangefs_attribute *attr, + const char *buf, + size_t count); +}; + +struct pc_orangefs_attribute { + struct attribute attr; + ssize_t (*show)(struct pc_orangefs_obj *pc_orangefs_obj, + struct pc_orangefs_attribute *attr, + char *buf); + ssize_t (*store)(struct pc_orangefs_obj *pc_orangefs_obj, + struct pc_orangefs_attribute *attr, + const char *buf, + size_t count); +}; + +struct stats_orangefs_attribute { + struct attribute attr; + ssize_t (*show)(struct stats_orangefs_obj *stats_orangefs_obj, + struct stats_orangefs_attribute *attr, + char *buf); + ssize_t (*store)(struct stats_orangefs_obj *stats_orangefs_obj, + struct stats_orangefs_attribute *attr, + const char *buf, + size_t count); +}; + +static ssize_t orangefs_attr_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + struct orangefs_attribute *attribute; + struct orangefs_obj *orangefs_obj; + int rc; + + attribute = container_of(attr, struct orangefs_attribute, attr); + orangefs_obj = container_of(kobj, struct orangefs_obj, kobj); + + if (!attribute->show) { + rc = -EIO; + goto out; + } + + rc = attribute->show(orangefs_obj, attribute, buf); + +out: + return rc; +} + +static ssize_t orangefs_attr_store(struct kobject *kobj, + struct attribute *attr, + const char *buf, + size_t len) +{ + struct orangefs_attribute *attribute; + struct orangefs_obj *orangefs_obj; + int rc; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "orangefs_attr_store: start\n"); + + attribute = container_of(attr, struct orangefs_attribute, attr); + orangefs_obj = container_of(kobj, struct orangefs_obj, kobj); + + if (!attribute->store) { + rc = -EIO; + goto out; + } + + rc = attribute->store(orangefs_obj, attribute, buf, len); + +out: + return rc; +} + +static const struct sysfs_ops orangefs_sysfs_ops = { + .show = orangefs_attr_show, + .store = orangefs_attr_store, +}; + +static ssize_t acache_orangefs_attr_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + struct acache_orangefs_attribute *attribute; + struct acache_orangefs_obj *acache_orangefs_obj; + int rc; + + attribute = container_of(attr, struct acache_orangefs_attribute, attr); + acache_orangefs_obj = + container_of(kobj, struct acache_orangefs_obj, kobj); + + if (!attribute->show) { + rc = -EIO; + goto out; + } + + rc = attribute->show(acache_orangefs_obj, attribute, buf); + +out: + return rc; +} + +static ssize_t acache_orangefs_attr_store(struct kobject *kobj, + struct attribute *attr, + const char *buf, + size_t len) +{ + struct acache_orangefs_attribute *attribute; + struct acache_orangefs_obj *acache_orangefs_obj; + int rc; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "acache_orangefs_attr_store: start\n"); + + attribute = container_of(attr, struct acache_orangefs_attribute, attr); + acache_orangefs_obj = + container_of(kobj, struct acache_orangefs_obj, kobj); + + if (!attribute->store) { + rc = -EIO; + goto out; + } + + rc = attribute->store(acache_orangefs_obj, attribute, buf, len); + +out: + return rc; +} + +static const struct sysfs_ops acache_orangefs_sysfs_ops = { + .show = acache_orangefs_attr_show, + .store = acache_orangefs_attr_store, +}; + +static ssize_t capcache_orangefs_attr_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + struct capcache_orangefs_attribute *attribute; + struct capcache_orangefs_obj *capcache_orangefs_obj; + int rc; + + attribute = + container_of(attr, struct capcache_orangefs_attribute, attr); + capcache_orangefs_obj = + container_of(kobj, struct capcache_orangefs_obj, kobj); + + if (!attribute->show) { + rc = -EIO; + goto out; + } + + rc = attribute->show(capcache_orangefs_obj, attribute, buf); + +out: + return rc; +} + +static ssize_t capcache_orangefs_attr_store(struct kobject *kobj, + struct attribute *attr, + const char *buf, + size_t len) +{ + struct capcache_orangefs_attribute *attribute; + struct capcache_orangefs_obj *capcache_orangefs_obj; + int rc; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "capcache_orangefs_attr_store: start\n"); + + attribute = + container_of(attr, struct capcache_orangefs_attribute, attr); + capcache_orangefs_obj = + container_of(kobj, struct capcache_orangefs_obj, kobj); + + if (!attribute->store) { + rc = -EIO; + goto out; + } + + rc = attribute->store(capcache_orangefs_obj, attribute, buf, len); + +out: + return rc; +} + +static const struct sysfs_ops capcache_orangefs_sysfs_ops = { + .show = capcache_orangefs_attr_show, + .store = capcache_orangefs_attr_store, +}; + +static ssize_t ccache_orangefs_attr_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + struct ccache_orangefs_attribute *attribute; + struct ccache_orangefs_obj *ccache_orangefs_obj; + int rc; + + attribute = + container_of(attr, struct ccache_orangefs_attribute, attr); + ccache_orangefs_obj = + container_of(kobj, struct ccache_orangefs_obj, kobj); + + if (!attribute->show) { + rc = -EIO; + goto out; + } + + rc = attribute->show(ccache_orangefs_obj, attribute, buf); + +out: + return rc; +} + +static ssize_t ccache_orangefs_attr_store(struct kobject *kobj, + struct attribute *attr, + const char *buf, + size_t len) +{ + struct ccache_orangefs_attribute *attribute; + struct ccache_orangefs_obj *ccache_orangefs_obj; + int rc; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "ccache_orangefs_attr_store: start\n"); + + attribute = + container_of(attr, struct ccache_orangefs_attribute, attr); + ccache_orangefs_obj = + container_of(kobj, struct ccache_orangefs_obj, kobj); + + if (!attribute->store) { + rc = -EIO; + goto out; + } + + rc = attribute->store(ccache_orangefs_obj, attribute, buf, len); + +out: + return rc; +} + +static const struct sysfs_ops ccache_orangefs_sysfs_ops = { + .show = ccache_orangefs_attr_show, + .store = ccache_orangefs_attr_store, +}; + +static ssize_t ncache_orangefs_attr_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + struct ncache_orangefs_attribute *attribute; + struct ncache_orangefs_obj *ncache_orangefs_obj; + int rc; + + attribute = container_of(attr, struct ncache_orangefs_attribute, attr); + ncache_orangefs_obj = + container_of(kobj, struct ncache_orangefs_obj, kobj); + + if (!attribute->show) { + rc = -EIO; + goto out; + } + + rc = attribute->show(ncache_orangefs_obj, attribute, buf); + +out: + return rc; +} + +static ssize_t ncache_orangefs_attr_store(struct kobject *kobj, + struct attribute *attr, + const char *buf, + size_t len) +{ + struct ncache_orangefs_attribute *attribute; + struct ncache_orangefs_obj *ncache_orangefs_obj; + int rc; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "ncache_orangefs_attr_store: start\n"); + + attribute = container_of(attr, struct ncache_orangefs_attribute, attr); + ncache_orangefs_obj = + container_of(kobj, struct ncache_orangefs_obj, kobj); + + if (!attribute->store) { + rc = -EIO; + goto out; + } + + rc = attribute->store(ncache_orangefs_obj, attribute, buf, len); + +out: + return rc; +} + +static const struct sysfs_ops ncache_orangefs_sysfs_ops = { + .show = ncache_orangefs_attr_show, + .store = ncache_orangefs_attr_store, +}; + +static ssize_t pc_orangefs_attr_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + struct pc_orangefs_attribute *attribute; + struct pc_orangefs_obj *pc_orangefs_obj; + int rc; + + attribute = container_of(attr, struct pc_orangefs_attribute, attr); + pc_orangefs_obj = + container_of(kobj, struct pc_orangefs_obj, kobj); + + if (!attribute->show) { + rc = -EIO; + goto out; + } + + rc = attribute->show(pc_orangefs_obj, attribute, buf); + +out: + return rc; +} + +static const struct sysfs_ops pc_orangefs_sysfs_ops = { + .show = pc_orangefs_attr_show, +}; + +static ssize_t stats_orangefs_attr_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + struct stats_orangefs_attribute *attribute; + struct stats_orangefs_obj *stats_orangefs_obj; + int rc; + + attribute = container_of(attr, struct stats_orangefs_attribute, attr); + stats_orangefs_obj = + container_of(kobj, struct stats_orangefs_obj, kobj); + + if (!attribute->show) { + rc = -EIO; + goto out; + } + + rc = attribute->show(stats_orangefs_obj, attribute, buf); + +out: + return rc; +} + +static const struct sysfs_ops stats_orangefs_sysfs_ops = { + .show = stats_orangefs_attr_show, +}; + +static void orangefs_release(struct kobject *kobj) +{ + struct orangefs_obj *orangefs_obj; + + orangefs_obj = container_of(kobj, struct orangefs_obj, kobj); + kfree(orangefs_obj); +} + +static void acache_orangefs_release(struct kobject *kobj) +{ + struct acache_orangefs_obj *acache_orangefs_obj; + + acache_orangefs_obj = + container_of(kobj, struct acache_orangefs_obj, kobj); + kfree(acache_orangefs_obj); +} + +static void capcache_orangefs_release(struct kobject *kobj) +{ + struct capcache_orangefs_obj *capcache_orangefs_obj; + + capcache_orangefs_obj = + container_of(kobj, struct capcache_orangefs_obj, kobj); + kfree(capcache_orangefs_obj); +} + +static void ccache_orangefs_release(struct kobject *kobj) +{ + struct ccache_orangefs_obj *ccache_orangefs_obj; + + ccache_orangefs_obj = + container_of(kobj, struct ccache_orangefs_obj, kobj); + kfree(ccache_orangefs_obj); +} + +static void ncache_orangefs_release(struct kobject *kobj) +{ + struct ncache_orangefs_obj *ncache_orangefs_obj; + + ncache_orangefs_obj = + container_of(kobj, struct ncache_orangefs_obj, kobj); + kfree(ncache_orangefs_obj); +} + +static void pc_orangefs_release(struct kobject *kobj) +{ + struct pc_orangefs_obj *pc_orangefs_obj; + + pc_orangefs_obj = + container_of(kobj, struct pc_orangefs_obj, kobj); + kfree(pc_orangefs_obj); +} + +static void stats_orangefs_release(struct kobject *kobj) +{ + struct stats_orangefs_obj *stats_orangefs_obj; + + stats_orangefs_obj = + container_of(kobj, struct stats_orangefs_obj, kobj); + kfree(stats_orangefs_obj); +} + +static ssize_t sysfs_int_show(char *kobj_id, char *buf, void *attr) +{ + int rc = -EIO; + struct orangefs_attribute *orangefs_attr; + struct stats_orangefs_attribute *stats_orangefs_attr; + + gossip_debug(GOSSIP_SYSFS_DEBUG, "sysfs_int_show: id:%s:\n", kobj_id); + + if (!strcmp(kobj_id, ORANGEFS_KOBJ_ID)) { + orangefs_attr = (struct orangefs_attribute *)attr; + + if (!strcmp(orangefs_attr->attr.name, "op_timeout_secs")) { + rc = scnprintf(buf, + PAGE_SIZE, + "%d\n", + op_timeout_secs); + goto out; + } else if (!strcmp(orangefs_attr->attr.name, + "slot_timeout_secs")) { + rc = scnprintf(buf, + PAGE_SIZE, + "%d\n", + slot_timeout_secs); + goto out; + } else { + goto out; + } + + } else if (!strcmp(kobj_id, STATS_KOBJ_ID)) { + stats_orangefs_attr = (struct stats_orangefs_attribute *)attr; + + if (!strcmp(stats_orangefs_attr->attr.name, "reads")) { + rc = scnprintf(buf, + PAGE_SIZE, + "%lu\n", + g_pvfs2_stats.reads); + goto out; + } else if (!strcmp(stats_orangefs_attr->attr.name, "writes")) { + rc = scnprintf(buf, + PAGE_SIZE, + "%lu\n", + g_pvfs2_stats.writes); + goto out; + } else { + goto out; + } + } + +out: + + return rc; +} + +static ssize_t int_orangefs_show(struct orangefs_obj *orangefs_obj, + struct orangefs_attribute *attr, + char *buf) +{ + int rc; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "int_orangefs_show:start attr->attr.name:%s:\n", + attr->attr.name); + + rc = sysfs_int_show(ORANGEFS_KOBJ_ID, buf, (void *) attr); + + return rc; +} + +static ssize_t int_stats_show(struct stats_orangefs_obj *stats_orangefs_obj, + struct stats_orangefs_attribute *attr, + char *buf) +{ + int rc; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "int_stats_show:start attr->attr.name:%s:\n", + attr->attr.name); + + rc = sysfs_int_show(STATS_KOBJ_ID, buf, (void *) attr); + + return rc; +} + +static ssize_t int_store(struct orangefs_obj *orangefs_obj, + struct orangefs_attribute *attr, + const char *buf, + size_t count) +{ + int rc = 0; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "int_store: start attr->attr.name:%s: buf:%s:\n", + attr->attr.name, buf); + + if (!strcmp(attr->attr.name, "op_timeout_secs")) { + rc = kstrtoint(buf, 0, &op_timeout_secs); + goto out; + } else if (!strcmp(attr->attr.name, "slot_timeout_secs")) { + rc = kstrtoint(buf, 0, &slot_timeout_secs); + goto out; + } else { + goto out; + } + +out: + if (rc) + rc = -EINVAL; + else + rc = count; + + return rc; +} + +/* + * obtain attribute values from userspace with a service operation. + */ +static int sysfs_service_op_show(char *kobj_id, char *buf, void *attr) +{ + struct pvfs2_kernel_op_s *new_op = NULL; + int rc = 0; + char *ser_op_type = NULL; + struct orangefs_attribute *orangefs_attr; + struct acache_orangefs_attribute *acache_attr; + struct capcache_orangefs_attribute *capcache_attr; + struct ccache_orangefs_attribute *ccache_attr; + struct ncache_orangefs_attribute *ncache_attr; + struct pc_orangefs_attribute *pc_attr; + __u32 op_alloc_type; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "sysfs_service_op_show: id:%s:\n", + kobj_id); + + if (strcmp(kobj_id, PC_KOBJ_ID)) + op_alloc_type = PVFS2_VFS_OP_PARAM; + else + op_alloc_type = PVFS2_VFS_OP_PERF_COUNT; + + new_op = op_alloc(op_alloc_type); + if (!new_op) { + rc = -ENOMEM; + goto out; + } + + /* Can't do a service_operation if the client is not running... */ + rc = is_daemon_in_service(); + if (rc) { + pr_info("%s: Client not running :%d:\n", + __func__, + is_daemon_in_service()); + goto out; + } + + if (strcmp(kobj_id, PC_KOBJ_ID)) + new_op->upcall.req.param.type = PVFS2_PARAM_REQUEST_GET; + + if (!strcmp(kobj_id, ORANGEFS_KOBJ_ID)) { + orangefs_attr = (struct orangefs_attribute *)attr; + + if (!strcmp(orangefs_attr->attr.name, "perf_history_size")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_PERF_HISTORY_SIZE; + else if (!strcmp(orangefs_attr->attr.name, + "perf_time_interval_secs")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_PERF_TIME_INTERVAL_SECS; + else if (!strcmp(orangefs_attr->attr.name, + "perf_counter_reset")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_PERF_RESET; + + } else if (!strcmp(kobj_id, ACACHE_KOBJ_ID)) { + acache_attr = (struct acache_orangefs_attribute *)attr; + + if (!strcmp(acache_attr->attr.name, "timeout_msecs")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_ACACHE_TIMEOUT_MSECS; + + if (!strcmp(acache_attr->attr.name, "hard_limit")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_ACACHE_HARD_LIMIT; + + if (!strcmp(acache_attr->attr.name, "soft_limit")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_ACACHE_SOFT_LIMIT; + + if (!strcmp(acache_attr->attr.name, "reclaim_percentage")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_ACACHE_RECLAIM_PERCENTAGE; + + } else if (!strcmp(kobj_id, CAPCACHE_KOBJ_ID)) { + capcache_attr = (struct capcache_orangefs_attribute *)attr; + + if (!strcmp(capcache_attr->attr.name, "timeout_secs")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CAPCACHE_TIMEOUT_SECS; + + if (!strcmp(capcache_attr->attr.name, "hard_limit")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CAPCACHE_HARD_LIMIT; + + if (!strcmp(capcache_attr->attr.name, "soft_limit")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CAPCACHE_SOFT_LIMIT; + + if (!strcmp(capcache_attr->attr.name, "reclaim_percentage")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CAPCACHE_RECLAIM_PERCENTAGE; + + } else if (!strcmp(kobj_id, CCACHE_KOBJ_ID)) { + ccache_attr = (struct ccache_orangefs_attribute *)attr; + + if (!strcmp(ccache_attr->attr.name, "timeout_secs")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CCACHE_TIMEOUT_SECS; + + if (!strcmp(ccache_attr->attr.name, "hard_limit")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CCACHE_HARD_LIMIT; + + if (!strcmp(ccache_attr->attr.name, "soft_limit")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CCACHE_SOFT_LIMIT; + + if (!strcmp(ccache_attr->attr.name, "reclaim_percentage")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CCACHE_RECLAIM_PERCENTAGE; + + } else if (!strcmp(kobj_id, NCACHE_KOBJ_ID)) { + ncache_attr = (struct ncache_orangefs_attribute *)attr; + + if (!strcmp(ncache_attr->attr.name, "timeout_msecs")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_NCACHE_TIMEOUT_MSECS; + + if (!strcmp(ncache_attr->attr.name, "hard_limit")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_NCACHE_HARD_LIMIT; + + if (!strcmp(ncache_attr->attr.name, "soft_limit")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_NCACHE_SOFT_LIMIT; + + if (!strcmp(ncache_attr->attr.name, "reclaim_percentage")) + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_NCACHE_RECLAIM_PERCENTAGE; + + } else if (!strcmp(kobj_id, PC_KOBJ_ID)) { + pc_attr = (struct pc_orangefs_attribute *)attr; + + if (!strcmp(pc_attr->attr.name, ACACHE_KOBJ_ID)) + new_op->upcall.req.perf_count.type = + PVFS2_PERF_COUNT_REQUEST_ACACHE; + + if (!strcmp(pc_attr->attr.name, CAPCACHE_KOBJ_ID)) + new_op->upcall.req.perf_count.type = + PVFS2_PERF_COUNT_REQUEST_CAPCACHE; + + if (!strcmp(pc_attr->attr.name, NCACHE_KOBJ_ID)) + new_op->upcall.req.perf_count.type = + PVFS2_PERF_COUNT_REQUEST_NCACHE; + + } else { + gossip_err("sysfs_service_op_show: unknown kobj_id:%s:\n", + kobj_id); + rc = -EINVAL; + goto out; + } + + + if (strcmp(kobj_id, PC_KOBJ_ID)) + ser_op_type = "pvfs2_param"; + else + ser_op_type = "pvfs2_perf_count"; + + /* + * The service_operation will return an errno return code on + * error, and zero on success. + */ + rc = service_operation(new_op, ser_op_type, PVFS2_OP_INTERRUPTIBLE); + +out: + if (!rc) { + if (strcmp(kobj_id, PC_KOBJ_ID)) { + rc = scnprintf(buf, + PAGE_SIZE, + "%d\n", + (int)new_op->downcall.resp.param.value); + } else { + rc = scnprintf( + buf, + PAGE_SIZE, + "%s", + new_op->downcall.resp.perf_count.buffer); + } + } + + /* + * if we got ENOMEM, then op_alloc probably failed... + */ + if (rc != -ENOMEM) + op_release(new_op); + + return rc; + +} + +static ssize_t service_orangefs_show(struct orangefs_obj *orangefs_obj, + struct orangefs_attribute *attr, + char *buf) +{ + int rc = 0; + + rc = sysfs_service_op_show(ORANGEFS_KOBJ_ID, buf, (void *)attr); + + return rc; +} + +static ssize_t + service_acache_show(struct acache_orangefs_obj *acache_orangefs_obj, + struct acache_orangefs_attribute *attr, + char *buf) +{ + int rc = 0; + + rc = sysfs_service_op_show(ACACHE_KOBJ_ID, buf, (void *)attr); + + return rc; +} + +static ssize_t service_capcache_show(struct capcache_orangefs_obj + *capcache_orangefs_obj, + struct capcache_orangefs_attribute *attr, + char *buf) +{ + int rc = 0; + + rc = sysfs_service_op_show(CAPCACHE_KOBJ_ID, buf, (void *)attr); + + return rc; +} + +static ssize_t service_ccache_show(struct ccache_orangefs_obj + *ccache_orangefs_obj, + struct ccache_orangefs_attribute *attr, + char *buf) +{ + int rc = 0; + + rc = sysfs_service_op_show(CCACHE_KOBJ_ID, buf, (void *)attr); + + return rc; +} + +static ssize_t + service_ncache_show(struct ncache_orangefs_obj *ncache_orangefs_obj, + struct ncache_orangefs_attribute *attr, + char *buf) +{ + int rc = 0; + + rc = sysfs_service_op_show(NCACHE_KOBJ_ID, buf, (void *)attr); + + return rc; +} + +static ssize_t + service_pc_show(struct pc_orangefs_obj *pc_orangefs_obj, + struct pc_orangefs_attribute *attr, + char *buf) +{ + int rc = 0; + + rc = sysfs_service_op_show(PC_KOBJ_ID, buf, (void *)attr); + + return rc; +} + +/* + * pass attribute values back to userspace with a service operation. + * + * We have to do a memory allocation, an sscanf and a service operation. + * And we have to evaluate what the user entered, to make sure the + * value is within the range supported by the attribute. So, there's + * a lot of return code checking and mapping going on here. + * + * We want to return 1 if we think everything went OK, and + * EINVAL if not. + */ +static int sysfs_service_op_store(char *kobj_id, const char *buf, void *attr) +{ + struct pvfs2_kernel_op_s *new_op = NULL; + int val = 0; + int rc = 0; + struct orangefs_attribute *orangefs_attr; + struct acache_orangefs_attribute *acache_attr; + struct capcache_orangefs_attribute *capcache_attr; + struct ccache_orangefs_attribute *ccache_attr; + struct ncache_orangefs_attribute *ncache_attr; + + gossip_debug(GOSSIP_SYSFS_DEBUG, + "sysfs_service_op_store: id:%s:\n", + kobj_id); + + new_op = op_alloc(PVFS2_VFS_OP_PARAM); + if (!new_op) { + rc = -ENOMEM; + goto out; + } + + /* Can't do a service_operation if the client is not running... */ + rc = is_daemon_in_service(); + if (rc) { + pr_info("%s: Client not running :%d:\n", + __func__, + is_daemon_in_service()); + goto out; + } + + /* + * The value we want to send back to userspace is in buf. + */ + rc = kstrtoint(buf, 0, &val); + if (rc) + goto out; + + if (!strcmp(kobj_id, ORANGEFS_KOBJ_ID)) { + orangefs_attr = (struct orangefs_attribute *)attr; + + if (!strcmp(orangefs_attr->attr.name, "perf_history_size")) { + if (val > 0) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_PERF_HISTORY_SIZE; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(orangefs_attr->attr.name, + "perf_time_interval_secs")) { + if (val > 0) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_PERF_TIME_INTERVAL_SECS; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(orangefs_attr->attr.name, + "perf_counter_reset")) { + if ((val == 0) || (val == 1)) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_PERF_RESET; + } else { + rc = 0; + goto out; + } + } + + } else if (!strcmp(kobj_id, ACACHE_KOBJ_ID)) { + acache_attr = (struct acache_orangefs_attribute *)attr; + + if (!strcmp(acache_attr->attr.name, "hard_limit")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_ACACHE_HARD_LIMIT; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(acache_attr->attr.name, "soft_limit")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_ACACHE_SOFT_LIMIT; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(acache_attr->attr.name, + "reclaim_percentage")) { + if ((val > -1) && (val < 101)) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_ACACHE_RECLAIM_PERCENTAGE; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(acache_attr->attr.name, "timeout_msecs")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_ACACHE_TIMEOUT_MSECS; + } else { + rc = 0; + goto out; + } + } + + } else if (!strcmp(kobj_id, CAPCACHE_KOBJ_ID)) { + capcache_attr = (struct capcache_orangefs_attribute *)attr; + + if (!strcmp(capcache_attr->attr.name, "hard_limit")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CAPCACHE_HARD_LIMIT; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(capcache_attr->attr.name, "soft_limit")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CAPCACHE_SOFT_LIMIT; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(capcache_attr->attr.name, + "reclaim_percentage")) { + if ((val > -1) && (val < 101)) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CAPCACHE_RECLAIM_PERCENTAGE; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(capcache_attr->attr.name, "timeout_secs")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CAPCACHE_TIMEOUT_SECS; + } else { + rc = 0; + goto out; + } + } + + } else if (!strcmp(kobj_id, CCACHE_KOBJ_ID)) { + ccache_attr = (struct ccache_orangefs_attribute *)attr; + + if (!strcmp(ccache_attr->attr.name, "hard_limit")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CCACHE_HARD_LIMIT; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(ccache_attr->attr.name, "soft_limit")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CCACHE_SOFT_LIMIT; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(ccache_attr->attr.name, + "reclaim_percentage")) { + if ((val > -1) && (val < 101)) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CCACHE_RECLAIM_PERCENTAGE; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(ccache_attr->attr.name, "timeout_secs")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_CCACHE_TIMEOUT_SECS; + } else { + rc = 0; + goto out; + } + } + + } else if (!strcmp(kobj_id, NCACHE_KOBJ_ID)) { + ncache_attr = (struct ncache_orangefs_attribute *)attr; + + if (!strcmp(ncache_attr->attr.name, "hard_limit")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_NCACHE_HARD_LIMIT; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(ncache_attr->attr.name, "soft_limit")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_NCACHE_SOFT_LIMIT; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(ncache_attr->attr.name, + "reclaim_percentage")) { + if ((val > -1) && (val < 101)) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_NCACHE_RECLAIM_PERCENTAGE; + } else { + rc = 0; + goto out; + } + } else if (!strcmp(ncache_attr->attr.name, "timeout_msecs")) { + if (val > -1) { + new_op->upcall.req.param.op = + PVFS2_PARAM_REQUEST_OP_NCACHE_TIMEOUT_MSECS; + } else { + rc = 0; + goto out; + } + } + + } else { + gossip_err("sysfs_service_op_store: unknown kobj_id:%s:\n", + kobj_id); + rc = -EINVAL; + goto out; + } + + new_op->upcall.req.param.type = PVFS2_PARAM_REQUEST_SET; + + new_op->upcall.req.param.value = val; + + /* + * The service_operation will return a errno return code on + * error, and zero on success. + */ + rc = service_operation(new_op, "pvfs2_param", PVFS2_OP_INTERRUPTIBLE); + + if (rc < 0) { + gossip_err("sysfs_service_op_store: service op returned:%d:\n", + rc); + rc = 0; + } else { + rc = 1; + } + +out: + /* + * if we got ENOMEM, then op_alloc probably failed... + */ + if (rc == -ENOMEM) + rc = 0; + else + op_release(new_op); + + if (rc == 0) + rc = -EINVAL; + + return rc; +} + +static ssize_t + service_orangefs_store(struct orangefs_obj *orangefs_obj, + struct orangefs_attribute *attr, + const char *buf, + size_t count) +{ + int rc = 0; + + rc = sysfs_service_op_store(ORANGEFS_KOBJ_ID, buf, (void *) attr); + + /* rc should have an errno value if the service_op went bad. */ + if (rc == 1) + rc = count; + + return rc; +} + +static ssize_t + service_acache_store(struct acache_orangefs_obj *acache_orangefs_obj, + struct acache_orangefs_attribute *attr, + const char *buf, + size_t count) +{ + int rc = 0; + + rc = sysfs_service_op_store(ACACHE_KOBJ_ID, buf, (void *) attr); + + /* rc should have an errno value if the service_op went bad. */ + if (rc == 1) + rc = count; + + return rc; +} + +static ssize_t + service_capcache_store(struct capcache_orangefs_obj + *capcache_orangefs_obj, + struct capcache_orangefs_attribute *attr, + const char *buf, + size_t count) +{ + int rc = 0; + + rc = sysfs_service_op_store(CAPCACHE_KOBJ_ID, buf, (void *) attr); + + /* rc should have an errno value if the service_op went bad. */ + if (rc == 1) + rc = count; + + return rc; +} + +static ssize_t service_ccache_store(struct ccache_orangefs_obj + *ccache_orangefs_obj, + struct ccache_orangefs_attribute *attr, + const char *buf, + size_t count) +{ + int rc = 0; + + rc = sysfs_service_op_store(CCACHE_KOBJ_ID, buf, (void *) attr); + + /* rc should have an errno value if the service_op went bad. */ + if (rc == 1) + rc = count; + + return rc; +} + +static ssize_t + service_ncache_store(struct ncache_orangefs_obj *ncache_orangefs_obj, + struct ncache_orangefs_attribute *attr, + const char *buf, + size_t count) +{ + int rc = 0; + + rc = sysfs_service_op_store(NCACHE_KOBJ_ID, buf, (void *) attr); + + /* rc should have an errno value if the service_op went bad. */ + if (rc == 1) + rc = count; + + return rc; +} + +static struct orangefs_attribute op_timeout_secs_attribute = + __ATTR(op_timeout_secs, 0664, int_orangefs_show, int_store); + +static struct orangefs_attribute slot_timeout_secs_attribute = + __ATTR(slot_timeout_secs, 0664, int_orangefs_show, int_store); + +static struct orangefs_attribute perf_counter_reset_attribute = + __ATTR(perf_counter_reset, + 0664, + service_orangefs_show, + service_orangefs_store); + +static struct orangefs_attribute perf_history_size_attribute = + __ATTR(perf_history_size, + 0664, + service_orangefs_show, + service_orangefs_store); + +static struct orangefs_attribute perf_time_interval_secs_attribute = + __ATTR(perf_time_interval_secs, + 0664, + service_orangefs_show, + service_orangefs_store); + +static struct attribute *orangefs_default_attrs[] = { + &op_timeout_secs_attribute.attr, + &slot_timeout_secs_attribute.attr, + &perf_counter_reset_attribute.attr, + &perf_history_size_attribute.attr, + &perf_time_interval_secs_attribute.attr, + NULL, +}; + +static struct kobj_type orangefs_ktype = { + .sysfs_ops = &orangefs_sysfs_ops, + .release = orangefs_release, + .default_attrs = orangefs_default_attrs, +}; + +static struct acache_orangefs_attribute acache_hard_limit_attribute = + __ATTR(hard_limit, + 0664, + service_acache_show, + service_acache_store); + +static struct acache_orangefs_attribute acache_reclaim_percent_attribute = + __ATTR(reclaim_percentage, + 0664, + service_acache_show, + service_acache_store); + +static struct acache_orangefs_attribute acache_soft_limit_attribute = + __ATTR(soft_limit, + 0664, + service_acache_show, + service_acache_store); + +static struct acache_orangefs_attribute acache_timeout_msecs_attribute = + __ATTR(timeout_msecs, + 0664, + service_acache_show, + service_acache_store); + +static struct attribute *acache_orangefs_default_attrs[] = { + &acache_hard_limit_attribute.attr, + &acache_reclaim_percent_attribute.attr, + &acache_soft_limit_attribute.attr, + &acache_timeout_msecs_attribute.attr, + NULL, +}; + +static struct kobj_type acache_orangefs_ktype = { + .sysfs_ops = &acache_orangefs_sysfs_ops, + .release = acache_orangefs_release, + .default_attrs = acache_orangefs_default_attrs, +}; + +static struct capcache_orangefs_attribute capcache_hard_limit_attribute = + __ATTR(hard_limit, + 0664, + service_capcache_show, + service_capcache_store); + +static struct capcache_orangefs_attribute capcache_reclaim_percent_attribute = + __ATTR(reclaim_percentage, + 0664, + service_capcache_show, + service_capcache_store); + +static struct capcache_orangefs_attribute capcache_soft_limit_attribute = + __ATTR(soft_limit, + 0664, + service_capcache_show, + service_capcache_store); + +static struct capcache_orangefs_attribute capcache_timeout_secs_attribute = + __ATTR(timeout_secs, + 0664, + service_capcache_show, + service_capcache_store); + +static struct attribute *capcache_orangefs_default_attrs[] = { + &capcache_hard_limit_attribute.attr, + &capcache_reclaim_percent_attribute.attr, + &capcache_soft_limit_attribute.attr, + &capcache_timeout_secs_attribute.attr, + NULL, +}; + +static struct kobj_type capcache_orangefs_ktype = { + .sysfs_ops = &capcache_orangefs_sysfs_ops, + .release = capcache_orangefs_release, + .default_attrs = capcache_orangefs_default_attrs, +}; + +static struct ccache_orangefs_attribute ccache_hard_limit_attribute = + __ATTR(hard_limit, + 0664, + service_ccache_show, + service_ccache_store); + +static struct ccache_orangefs_attribute ccache_reclaim_percent_attribute = + __ATTR(reclaim_percentage, + 0664, + service_ccache_show, + service_ccache_store); + +static struct ccache_orangefs_attribute ccache_soft_limit_attribute = + __ATTR(soft_limit, + 0664, + service_ccache_show, + service_ccache_store); + +static struct ccache_orangefs_attribute ccache_timeout_secs_attribute = + __ATTR(timeout_secs, + 0664, + service_ccache_show, + service_ccache_store); + +static struct attribute *ccache_orangefs_default_attrs[] = { + &ccache_hard_limit_attribute.attr, + &ccache_reclaim_percent_attribute.attr, + &ccache_soft_limit_attribute.attr, + &ccache_timeout_secs_attribute.attr, + NULL, +}; + +static struct kobj_type ccache_orangefs_ktype = { + .sysfs_ops = &ccache_orangefs_sysfs_ops, + .release = ccache_orangefs_release, + .default_attrs = ccache_orangefs_default_attrs, +}; + +static struct ncache_orangefs_attribute ncache_hard_limit_attribute = + __ATTR(hard_limit, + 0664, + service_ncache_show, + service_ncache_store); + +static struct ncache_orangefs_attribute ncache_reclaim_percent_attribute = + __ATTR(reclaim_percentage, + 0664, + service_ncache_show, + service_ncache_store); + +static struct ncache_orangefs_attribute ncache_soft_limit_attribute = + __ATTR(soft_limit, + 0664, + service_ncache_show, + service_ncache_store); + +static struct ncache_orangefs_attribute ncache_timeout_msecs_attribute = + __ATTR(timeout_msecs, + 0664, + service_ncache_show, + service_ncache_store); + +static struct attribute *ncache_orangefs_default_attrs[] = { + &ncache_hard_limit_attribute.attr, + &ncache_reclaim_percent_attribute.attr, + &ncache_soft_limit_attribute.attr, + &ncache_timeout_msecs_attribute.attr, + NULL, +}; + +static struct kobj_type ncache_orangefs_ktype = { + .sysfs_ops = &ncache_orangefs_sysfs_ops, + .release = ncache_orangefs_release, + .default_attrs = ncache_orangefs_default_attrs, +}; + +static struct pc_orangefs_attribute pc_acache_attribute = + __ATTR(acache, + 0664, + service_pc_show, + NULL); + +static struct pc_orangefs_attribute pc_capcache_attribute = + __ATTR(capcache, + 0664, + service_pc_show, + NULL); + +static struct pc_orangefs_attribute pc_ncache_attribute = + __ATTR(ncache, + 0664, + service_pc_show, + NULL); + +static struct attribute *pc_orangefs_default_attrs[] = { + &pc_acache_attribute.attr, + &pc_capcache_attribute.attr, + &pc_ncache_attribute.attr, + NULL, +}; + +static struct kobj_type pc_orangefs_ktype = { + .sysfs_ops = &pc_orangefs_sysfs_ops, + .release = pc_orangefs_release, + .default_attrs = pc_orangefs_default_attrs, +}; + +static struct stats_orangefs_attribute stats_reads_attribute = + __ATTR(reads, + 0664, + int_stats_show, + NULL); + +static struct stats_orangefs_attribute stats_writes_attribute = + __ATTR(writes, + 0664, + int_stats_show, + NULL); + +static struct attribute *stats_orangefs_default_attrs[] = { + &stats_reads_attribute.attr, + &stats_writes_attribute.attr, + NULL, +}; + +static struct kobj_type stats_orangefs_ktype = { + .sysfs_ops = &stats_orangefs_sysfs_ops, + .release = stats_orangefs_release, + .default_attrs = stats_orangefs_default_attrs, +}; + +static struct orangefs_obj *orangefs_obj; +static struct acache_orangefs_obj *acache_orangefs_obj; +static struct capcache_orangefs_obj *capcache_orangefs_obj; +static struct ccache_orangefs_obj *ccache_orangefs_obj; +static struct ncache_orangefs_obj *ncache_orangefs_obj; +static struct pc_orangefs_obj *pc_orangefs_obj; +static struct stats_orangefs_obj *stats_orangefs_obj; + +int orangefs_sysfs_init(void) +{ + int rc; + + gossip_debug(GOSSIP_SYSFS_DEBUG, "orangefs_sysfs_init: start\n"); + + /* create /sys/fs/orangefs. */ + orangefs_obj = kzalloc(sizeof(*orangefs_obj), GFP_KERNEL); + if (!orangefs_obj) { + rc = -EINVAL; + goto out; + } + + rc = kobject_init_and_add(&orangefs_obj->kobj, + &orangefs_ktype, + fs_kobj, + ORANGEFS_KOBJ_ID); + + if (rc) { + kobject_put(&orangefs_obj->kobj); + rc = -EINVAL; + goto out; + } + + kobject_uevent(&orangefs_obj->kobj, KOBJ_ADD); + + /* create /sys/fs/orangefs/acache. */ + acache_orangefs_obj = kzalloc(sizeof(*acache_orangefs_obj), GFP_KERNEL); + if (!acache_orangefs_obj) { + rc = -EINVAL; + goto out; + } + + rc = kobject_init_and_add(&acache_orangefs_obj->kobj, + &acache_orangefs_ktype, + &orangefs_obj->kobj, + ACACHE_KOBJ_ID); + + if (rc) { + kobject_put(&acache_orangefs_obj->kobj); + rc = -EINVAL; + goto out; + } + + kobject_uevent(&acache_orangefs_obj->kobj, KOBJ_ADD); + + /* create /sys/fs/orangefs/capcache. */ + capcache_orangefs_obj = + kzalloc(sizeof(*capcache_orangefs_obj), GFP_KERNEL); + if (!capcache_orangefs_obj) { + rc = -EINVAL; + goto out; + } + + rc = kobject_init_and_add(&capcache_orangefs_obj->kobj, + &capcache_orangefs_ktype, + &orangefs_obj->kobj, + CAPCACHE_KOBJ_ID); + if (rc) { + kobject_put(&capcache_orangefs_obj->kobj); + rc = -EINVAL; + goto out; + } + + kobject_uevent(&capcache_orangefs_obj->kobj, KOBJ_ADD); + + /* create /sys/fs/orangefs/ccache. */ + ccache_orangefs_obj = + kzalloc(sizeof(*ccache_orangefs_obj), GFP_KERNEL); + if (!ccache_orangefs_obj) { + rc = -EINVAL; + goto out; + } + + rc = kobject_init_and_add(&ccache_orangefs_obj->kobj, + &ccache_orangefs_ktype, + &orangefs_obj->kobj, + CCACHE_KOBJ_ID); + if (rc) { + kobject_put(&ccache_orangefs_obj->kobj); + rc = -EINVAL; + goto out; + } + + kobject_uevent(&ccache_orangefs_obj->kobj, KOBJ_ADD); + + /* create /sys/fs/orangefs/ncache. */ + ncache_orangefs_obj = kzalloc(sizeof(*ncache_orangefs_obj), GFP_KERNEL); + if (!ncache_orangefs_obj) { + rc = -EINVAL; + goto out; + } + + rc = kobject_init_and_add(&ncache_orangefs_obj->kobj, + &ncache_orangefs_ktype, + &orangefs_obj->kobj, + NCACHE_KOBJ_ID); + + if (rc) { + kobject_put(&ncache_orangefs_obj->kobj); + rc = -EINVAL; + goto out; + } + + kobject_uevent(&ncache_orangefs_obj->kobj, KOBJ_ADD); + + /* create /sys/fs/orangefs/perf_counters. */ + pc_orangefs_obj = kzalloc(sizeof(*pc_orangefs_obj), GFP_KERNEL); + if (!pc_orangefs_obj) { + rc = -EINVAL; + goto out; + } + + rc = kobject_init_and_add(&pc_orangefs_obj->kobj, + &pc_orangefs_ktype, + &orangefs_obj->kobj, + "perf_counters"); + + if (rc) { + kobject_put(&pc_orangefs_obj->kobj); + rc = -EINVAL; + goto out; + } + + kobject_uevent(&pc_orangefs_obj->kobj, KOBJ_ADD); + + /* create /sys/fs/orangefs/stats. */ + stats_orangefs_obj = kzalloc(sizeof(*stats_orangefs_obj), GFP_KERNEL); + if (!stats_orangefs_obj) { + rc = -EINVAL; + goto out; + } + + rc = kobject_init_and_add(&stats_orangefs_obj->kobj, + &stats_orangefs_ktype, + &orangefs_obj->kobj, + STATS_KOBJ_ID); + + if (rc) { + kobject_put(&stats_orangefs_obj->kobj); + rc = -EINVAL; + goto out; + } + + kobject_uevent(&stats_orangefs_obj->kobj, KOBJ_ADD); +out: + return rc; +} + +void orangefs_sysfs_exit(void) +{ + gossip_debug(GOSSIP_SYSFS_DEBUG, "orangefs_sysfs_exit: start\n"); + + kobject_put(&acache_orangefs_obj->kobj); + kobject_put(&capcache_orangefs_obj->kobj); + kobject_put(&ccache_orangefs_obj->kobj); + kobject_put(&ncache_orangefs_obj->kobj); + kobject_put(&pc_orangefs_obj->kobj); + kobject_put(&stats_orangefs_obj->kobj); + + kobject_put(&orangefs_obj->kobj); +} diff --git a/fs/orangefs/pvfs2-sysfs.h b/fs/orangefs/pvfs2-sysfs.h new file mode 100644 index 000000000000..f0b76382db02 --- /dev/null +++ b/fs/orangefs/pvfs2-sysfs.h @@ -0,0 +1,2 @@ +extern int orangefs_sysfs_init(void); +extern void orangefs_sysfs_exit(void); diff --git a/fs/orangefs/pvfs2-utils.c b/fs/orangefs/pvfs2-utils.c new file mode 100644 index 000000000000..c33e7193599c --- /dev/null +++ b/fs/orangefs/pvfs2-utils.c @@ -0,0 +1,1165 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-dev-proto.h" +#include "pvfs2-bufmap.h" + +__s32 fsid_of_op(struct pvfs2_kernel_op_s *op) +{ + __s32 fsid = PVFS_FS_ID_NULL; + + if (op) { + switch (op->upcall.type) { + case PVFS2_VFS_OP_FILE_IO: + fsid = op->upcall.req.io.refn.fs_id; + break; + case PVFS2_VFS_OP_LOOKUP: + fsid = op->upcall.req.lookup.parent_refn.fs_id; + break; + case PVFS2_VFS_OP_CREATE: + fsid = op->upcall.req.create.parent_refn.fs_id; + break; + case PVFS2_VFS_OP_GETATTR: + fsid = op->upcall.req.getattr.refn.fs_id; + break; + case PVFS2_VFS_OP_REMOVE: + fsid = op->upcall.req.remove.parent_refn.fs_id; + break; + case PVFS2_VFS_OP_MKDIR: + fsid = op->upcall.req.mkdir.parent_refn.fs_id; + break; + case PVFS2_VFS_OP_READDIR: + fsid = op->upcall.req.readdir.refn.fs_id; + break; + case PVFS2_VFS_OP_SETATTR: + fsid = op->upcall.req.setattr.refn.fs_id; + break; + case PVFS2_VFS_OP_SYMLINK: + fsid = op->upcall.req.sym.parent_refn.fs_id; + break; + case PVFS2_VFS_OP_RENAME: + fsid = op->upcall.req.rename.old_parent_refn.fs_id; + break; + case PVFS2_VFS_OP_STATFS: + fsid = op->upcall.req.statfs.fs_id; + break; + case PVFS2_VFS_OP_TRUNCATE: + fsid = op->upcall.req.truncate.refn.fs_id; + break; + case PVFS2_VFS_OP_MMAP_RA_FLUSH: + fsid = op->upcall.req.ra_cache_flush.refn.fs_id; + break; + case PVFS2_VFS_OP_FS_UMOUNT: + fsid = op->upcall.req.fs_umount.fs_id; + break; + case PVFS2_VFS_OP_GETXATTR: + fsid = op->upcall.req.getxattr.refn.fs_id; + break; + case PVFS2_VFS_OP_SETXATTR: + fsid = op->upcall.req.setxattr.refn.fs_id; + break; + case PVFS2_VFS_OP_LISTXATTR: + fsid = op->upcall.req.listxattr.refn.fs_id; + break; + case PVFS2_VFS_OP_REMOVEXATTR: + fsid = op->upcall.req.removexattr.refn.fs_id; + break; + case PVFS2_VFS_OP_FSYNC: + fsid = op->upcall.req.fsync.refn.fs_id; + break; + default: + break; + } + } + return fsid; +} + +static void pvfs2_set_inode_flags(struct inode *inode, + struct PVFS_sys_attr_s *attrs) +{ + if (attrs->flags & PVFS_IMMUTABLE_FL) + inode->i_flags |= S_IMMUTABLE; + else + inode->i_flags &= ~S_IMMUTABLE; + + if (attrs->flags & PVFS_APPEND_FL) + inode->i_flags |= S_APPEND; + else + inode->i_flags &= ~S_APPEND; + + if (attrs->flags & PVFS_NOATIME_FL) + inode->i_flags |= S_NOATIME; + else + inode->i_flags &= ~S_NOATIME; + +} + +/* NOTE: symname is ignored unless the inode is a sym link */ +static int copy_attributes_to_inode(struct inode *inode, + struct PVFS_sys_attr_s *attrs, + char *symname) +{ + int ret = -1; + int perm_mode = 0; + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + loff_t inode_size = 0; + loff_t rounded_up_size = 0; + + + /* + * arbitrarily set the inode block size; FIXME: we need to + * resolve the difference between the reported inode blocksize + * and the PAGE_CACHE_SIZE, since our block count will always + * be wrong. + * + * For now, we're setting the block count to be the proper + * number assuming the block size is 512 bytes, and the size is + * rounded up to the nearest 4K. This is apparently required + * to get proper size reports from the 'du' shell utility. + * + * changing the inode->i_blkbits to something other than + * PAGE_CACHE_SHIFT breaks mmap/execution as we depend on that. + */ + gossip_debug(GOSSIP_UTILS_DEBUG, + "attrs->mask = %x (objtype = %s)\n", + attrs->mask, + attrs->objtype == PVFS_TYPE_METAFILE ? "file" : + attrs->objtype == PVFS_TYPE_DIRECTORY ? "directory" : + attrs->objtype == PVFS_TYPE_SYMLINK ? "symlink" : + "invalid/unknown"); + + switch (attrs->objtype) { + case PVFS_TYPE_METAFILE: + pvfs2_set_inode_flags(inode, attrs); + if (attrs->mask & PVFS_ATTR_SYS_SIZE) { + inode_size = (loff_t) attrs->size; + rounded_up_size = + (inode_size + (4096 - (inode_size % 4096))); + + pvfs2_lock_inode(inode); + inode->i_bytes = inode_size; + inode->i_blocks = + (unsigned long)(rounded_up_size / 512); + pvfs2_unlock_inode(inode); + + /* + * NOTE: make sure all the places we're called + * from have the inode->i_sem lock. We're fine + * in 99% of the cases since we're mostly + * called from a lookup. + */ + inode->i_size = inode_size; + } + break; + case PVFS_TYPE_SYMLINK: + if (symname != NULL) { + inode->i_size = (loff_t) strlen(symname); + break; + } + /*FALLTHRU*/ + default: + inode->i_size = PAGE_CACHE_SIZE; + + pvfs2_lock_inode(inode); + inode_set_bytes(inode, inode->i_size); + pvfs2_unlock_inode(inode); + break; + } + + inode->i_uid = make_kuid(&init_user_ns, attrs->owner); + inode->i_gid = make_kgid(&init_user_ns, attrs->group); + inode->i_atime.tv_sec = (time_t) attrs->atime; + inode->i_mtime.tv_sec = (time_t) attrs->mtime; + inode->i_ctime.tv_sec = (time_t) attrs->ctime; + inode->i_atime.tv_nsec = 0; + inode->i_mtime.tv_nsec = 0; + inode->i_ctime.tv_nsec = 0; + + if (attrs->perms & PVFS_O_EXECUTE) + perm_mode |= S_IXOTH; + if (attrs->perms & PVFS_O_WRITE) + perm_mode |= S_IWOTH; + if (attrs->perms & PVFS_O_READ) + perm_mode |= S_IROTH; + + if (attrs->perms & PVFS_G_EXECUTE) + perm_mode |= S_IXGRP; + if (attrs->perms & PVFS_G_WRITE) + perm_mode |= S_IWGRP; + if (attrs->perms & PVFS_G_READ) + perm_mode |= S_IRGRP; + + if (attrs->perms & PVFS_U_EXECUTE) + perm_mode |= S_IXUSR; + if (attrs->perms & PVFS_U_WRITE) + perm_mode |= S_IWUSR; + if (attrs->perms & PVFS_U_READ) + perm_mode |= S_IRUSR; + + if (attrs->perms & PVFS_G_SGID) + perm_mode |= S_ISGID; + if (attrs->perms & PVFS_U_SUID) + perm_mode |= S_ISUID; + + inode->i_mode = perm_mode; + + if (is_root_handle(inode)) { + /* special case: mark the root inode as sticky */ + inode->i_mode |= S_ISVTX; + gossip_debug(GOSSIP_UTILS_DEBUG, + "Marking inode %pU as sticky\n", + get_khandle_from_ino(inode)); + } + + switch (attrs->objtype) { + case PVFS_TYPE_METAFILE: + inode->i_mode |= S_IFREG; + ret = 0; + break; + case PVFS_TYPE_DIRECTORY: + inode->i_mode |= S_IFDIR; + /* NOTE: we have no good way to keep nlink consistent + * for directories across clients; keep constant at 1. + * Why 1? If we go with 2, then find(1) gets confused + * and won't work properly withouth the -noleaf option + */ + set_nlink(inode, 1); + ret = 0; + break; + case PVFS_TYPE_SYMLINK: + inode->i_mode |= S_IFLNK; + + /* copy link target to inode private data */ + if (pvfs2_inode && symname) { + strncpy(pvfs2_inode->link_target, + symname, + PVFS_NAME_MAX); + gossip_debug(GOSSIP_UTILS_DEBUG, + "Copied attr link target %s\n", + pvfs2_inode->link_target); + } + gossip_debug(GOSSIP_UTILS_DEBUG, + "symlink mode %o\n", + inode->i_mode); + ret = 0; + break; + default: + gossip_err("pvfs2: copy_attributes_to_inode: got invalid attribute type %x\n", + attrs->objtype); + } + + gossip_debug(GOSSIP_UTILS_DEBUG, + "pvfs2: copy_attributes_to_inode: setting i_mode to %o, i_size to %lu\n", + inode->i_mode, + (unsigned long)i_size_read(inode)); + + return ret; +} + +/* + * NOTE: in kernel land, we never use the sys_attr->link_target for + * anything, so don't bother copying it into the sys_attr object here. + */ +static inline int copy_attributes_from_inode(struct inode *inode, + struct PVFS_sys_attr_s *attrs, + struct iattr *iattr) +{ + umode_t tmp_mode; + + if (!iattr || !inode || !attrs) { + gossip_err("NULL iattr (%p), inode (%p), attrs (%p) " + "in copy_attributes_from_inode!\n", + iattr, + inode, + attrs); + return -EINVAL; + } + /* + * We need to be careful to only copy the attributes out of the + * iattr object that we know are valid. + */ + attrs->mask = 0; + if (iattr->ia_valid & ATTR_UID) { + attrs->owner = from_kuid(current_user_ns(), iattr->ia_uid); + attrs->mask |= PVFS_ATTR_SYS_UID; + gossip_debug(GOSSIP_UTILS_DEBUG, "(UID) %d\n", attrs->owner); + } + if (iattr->ia_valid & ATTR_GID) { + attrs->group = from_kgid(current_user_ns(), iattr->ia_gid); + attrs->mask |= PVFS_ATTR_SYS_GID; + gossip_debug(GOSSIP_UTILS_DEBUG, "(GID) %d\n", attrs->group); + } + + if (iattr->ia_valid & ATTR_ATIME) { + attrs->mask |= PVFS_ATTR_SYS_ATIME; + if (iattr->ia_valid & ATTR_ATIME_SET) { + attrs->atime = + pvfs2_convert_time_field((void *)&iattr->ia_atime); + attrs->mask |= PVFS_ATTR_SYS_ATIME_SET; + } + } + if (iattr->ia_valid & ATTR_MTIME) { + attrs->mask |= PVFS_ATTR_SYS_MTIME; + if (iattr->ia_valid & ATTR_MTIME_SET) { + attrs->mtime = + pvfs2_convert_time_field((void *)&iattr->ia_mtime); + attrs->mask |= PVFS_ATTR_SYS_MTIME_SET; + } + } + if (iattr->ia_valid & ATTR_CTIME) + attrs->mask |= PVFS_ATTR_SYS_CTIME; + + /* + * PVFS2 cannot set size with a setattr operation. Probably not likely + * to be requested through the VFS, but just in case, don't worry about + * ATTR_SIZE + */ + + if (iattr->ia_valid & ATTR_MODE) { + tmp_mode = iattr->ia_mode; + if (tmp_mode & (S_ISVTX)) { + if (is_root_handle(inode)) { + /* + * allow sticky bit to be set on root (since + * it shows up that way by default anyhow), + * but don't show it to the server + */ + tmp_mode -= S_ISVTX; + } else { + gossip_debug(GOSSIP_UTILS_DEBUG, + "User attempted to set sticky bit on non-root directory; returning EINVAL.\n"); + return -EINVAL; + } + } + + if (tmp_mode & (S_ISUID)) { + gossip_debug(GOSSIP_UTILS_DEBUG, + "Attempting to set setuid bit (not supported); returning EINVAL.\n"); + return -EINVAL; + } + + attrs->perms = PVFS_util_translate_mode(tmp_mode); + attrs->mask |= PVFS_ATTR_SYS_PERM; + } + + return 0; +} + +/* + * issues a pvfs2 getattr request and fills in the appropriate inode + * attributes if successful. returns 0 on success; -errno otherwise + */ +int pvfs2_inode_getattr(struct inode *inode, __u32 getattr_mask) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_kernel_op_s *new_op; + int ret = -EINVAL; + + gossip_debug(GOSSIP_UTILS_DEBUG, + "%s: called on inode %pU\n", + __func__, + get_khandle_from_ino(inode)); + + new_op = op_alloc(PVFS2_VFS_OP_GETATTR); + if (!new_op) + return -ENOMEM; + new_op->upcall.req.getattr.refn = pvfs2_inode->refn; + new_op->upcall.req.getattr.mask = getattr_mask; + + ret = service_operation(new_op, __func__, + get_interruptible_flag(inode)); + if (ret != 0) + goto out; + + if (copy_attributes_to_inode(inode, + &new_op->downcall.resp.getattr.attributes, + new_op->downcall.resp.getattr.link_target)) { + gossip_err("%s: failed to copy attributes\n", __func__); + ret = -ENOENT; + goto out; + } + + /* + * Store blksize in pvfs2 specific part of inode structure; we are + * only going to use this to report to stat to make sure it doesn't + * perturb any inode related code paths. + */ + if (new_op->downcall.resp.getattr.attributes.objtype == + PVFS_TYPE_METAFILE) { + pvfs2_inode->blksize = + new_op->downcall.resp.getattr.attributes.blksize; + } else { + /* mimic behavior of generic_fillattr() for other types. */ + pvfs2_inode->blksize = (1 << inode->i_blkbits); + + } + +out: + gossip_debug(GOSSIP_UTILS_DEBUG, + "Getattr on handle %pU, " + "fsid %d\n (inode ct = %d) returned %d\n", + &pvfs2_inode->refn.khandle, + pvfs2_inode->refn.fs_id, + (int)atomic_read(&inode->i_count), + ret); + + op_release(new_op); + return ret; +} + +/* + * issues a pvfs2 setattr request to make sure the new attribute values + * take effect if successful. returns 0 on success; -errno otherwise + */ +int pvfs2_inode_setattr(struct inode *inode, struct iattr *iattr) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_kernel_op_s *new_op; + int ret; + + new_op = op_alloc(PVFS2_VFS_OP_SETATTR); + if (!new_op) + return -ENOMEM; + + new_op->upcall.req.setattr.refn = pvfs2_inode->refn; + ret = copy_attributes_from_inode(inode, + &new_op->upcall.req.setattr.attributes, + iattr); + if (ret < 0) { + op_release(new_op); + return ret; + } + + ret = service_operation(new_op, __func__, + get_interruptible_flag(inode)); + + gossip_debug(GOSSIP_UTILS_DEBUG, + "pvfs2_inode_setattr: returning %d\n", + ret); + + /* when request is serviced properly, free req op struct */ + op_release(new_op); + + /* + * successful setattr should clear the atime, mtime and + * ctime flags. + */ + if (ret == 0) { + ClearAtimeFlag(pvfs2_inode); + ClearMtimeFlag(pvfs2_inode); + ClearCtimeFlag(pvfs2_inode); + ClearModeFlag(pvfs2_inode); + } + + return ret; +} + +int pvfs2_flush_inode(struct inode *inode) +{ + /* + * If it is a dirty inode, this function gets called. + * Gather all the information that needs to be setattr'ed + * Right now, this will only be used for mode, atime, mtime + * and/or ctime. + */ + struct iattr wbattr; + int ret; + int mtime_flag; + int ctime_flag; + int atime_flag; + int mode_flag; + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + + memset(&wbattr, 0, sizeof(wbattr)); + + /* + * check inode flags up front, and clear them if they are set. This + * will prevent multiple processes from all trying to flush the same + * inode if they call close() simultaneously + */ + mtime_flag = MtimeFlag(pvfs2_inode); + ClearMtimeFlag(pvfs2_inode); + ctime_flag = CtimeFlag(pvfs2_inode); + ClearCtimeFlag(pvfs2_inode); + atime_flag = AtimeFlag(pvfs2_inode); + ClearAtimeFlag(pvfs2_inode); + mode_flag = ModeFlag(pvfs2_inode); + ClearModeFlag(pvfs2_inode); + + /* -- Lazy atime,mtime and ctime update -- + * Note: all times are dictated by server in the new scheme + * and not by the clients + * + * Also mode updates are being handled now.. + */ + + if (mtime_flag) + wbattr.ia_valid |= ATTR_MTIME; + if (ctime_flag) + wbattr.ia_valid |= ATTR_CTIME; + if (atime_flag) + wbattr.ia_valid |= ATTR_ATIME; + + if (mode_flag) { + wbattr.ia_mode = inode->i_mode; + wbattr.ia_valid |= ATTR_MODE; + } + + gossip_debug(GOSSIP_UTILS_DEBUG, + "*********** pvfs2_flush_inode: %pU " + "(ia_valid %d)\n", + get_khandle_from_ino(inode), + wbattr.ia_valid); + if (wbattr.ia_valid == 0) { + gossip_debug(GOSSIP_UTILS_DEBUG, + "pvfs2_flush_inode skipping setattr()\n"); + return 0; + } + + gossip_debug(GOSSIP_UTILS_DEBUG, + "pvfs2_flush_inode (%pU) writing mode %o\n", + get_khandle_from_ino(inode), + inode->i_mode); + + ret = pvfs2_inode_setattr(inode, &wbattr); + + return ret; +} + +int pvfs2_unmount_sb(struct super_block *sb) +{ + int ret = -EINVAL; + struct pvfs2_kernel_op_s *new_op = NULL; + + gossip_debug(GOSSIP_UTILS_DEBUG, + "pvfs2_unmount_sb called on sb %p\n", + sb); + + new_op = op_alloc(PVFS2_VFS_OP_FS_UMOUNT); + if (!new_op) + return -ENOMEM; + new_op->upcall.req.fs_umount.id = PVFS2_SB(sb)->id; + new_op->upcall.req.fs_umount.fs_id = PVFS2_SB(sb)->fs_id; + strncpy(new_op->upcall.req.fs_umount.pvfs2_config_server, + PVFS2_SB(sb)->devname, + PVFS_MAX_SERVER_ADDR_LEN); + + gossip_debug(GOSSIP_UTILS_DEBUG, + "Attempting PVFS2 Unmount via host %s\n", + new_op->upcall.req.fs_umount.pvfs2_config_server); + + ret = service_operation(new_op, "pvfs2_fs_umount", 0); + + gossip_debug(GOSSIP_UTILS_DEBUG, + "pvfs2_unmount: got return value of %d\n", ret); + if (ret) + sb = ERR_PTR(ret); + else + PVFS2_SB(sb)->mount_pending = 1; + + op_release(new_op); + return ret; +} + +/* + * NOTE: on successful cancellation, be sure to return -EINTR, as + * that's the return value the caller expects + */ +int pvfs2_cancel_op_in_progress(__u64 tag) +{ + int ret = -EINVAL; + struct pvfs2_kernel_op_s *new_op = NULL; + + gossip_debug(GOSSIP_UTILS_DEBUG, + "pvfs2_cancel_op_in_progress called on tag %llu\n", + llu(tag)); + + new_op = op_alloc(PVFS2_VFS_OP_CANCEL); + if (!new_op) + return -ENOMEM; + new_op->upcall.req.cancel.op_tag = tag; + + gossip_debug(GOSSIP_UTILS_DEBUG, + "Attempting PVFS2 operation cancellation of tag %llu\n", + llu(new_op->upcall.req.cancel.op_tag)); + + ret = service_operation(new_op, "pvfs2_cancel", PVFS2_OP_CANCELLATION); + + gossip_debug(GOSSIP_UTILS_DEBUG, + "pvfs2_cancel_op_in_progress: got return value of %d\n", + ret); + + op_release(new_op); + return ret; +} + +void pvfs2_op_initialize(struct pvfs2_kernel_op_s *op) +{ + if (op) { + spin_lock(&op->lock); + op->io_completed = 0; + + op->upcall.type = PVFS2_VFS_OP_INVALID; + op->downcall.type = PVFS2_VFS_OP_INVALID; + op->downcall.status = -1; + + op->op_state = OP_VFS_STATE_UNKNOWN; + op->tag = 0; + spin_unlock(&op->lock); + } +} + +void pvfs2_make_bad_inode(struct inode *inode) +{ + if (is_root_handle(inode)) { + /* + * if this occurs, the pvfs2-client-core was killed but we + * can't afford to lose the inode operations and such + * associated with the root handle in any case. + */ + gossip_debug(GOSSIP_UTILS_DEBUG, + "*** NOT making bad root inode %pU\n", + get_khandle_from_ino(inode)); + } else { + gossip_debug(GOSSIP_UTILS_DEBUG, + "*** making bad inode %pU\n", + get_khandle_from_ino(inode)); + make_bad_inode(inode); + } +} + +/* Block all blockable signals... */ +void block_signals(sigset_t *orig_sigset) +{ + sigset_t mask; + + /* + * Initialize all entries in the signal set to the + * inverse of the given mask. + */ + siginitsetinv(&mask, sigmask(SIGKILL)); + + /* Block 'em Danno... */ + sigprocmask(SIG_BLOCK, &mask, orig_sigset); +} + +/* set the signal mask to the given template... */ +void set_signals(sigset_t *sigset) +{ + sigprocmask(SIG_SETMASK, sigset, NULL); +} + +__u64 pvfs2_convert_time_field(void *time_ptr) +{ + __u64 pvfs2_time; + struct timespec *tspec = (struct timespec *)time_ptr; + + pvfs2_time = (__u64) ((time_t) tspec->tv_sec); + return pvfs2_time; +} + +/* + * The following is a very dirty hack that is now a permanent part of the + * PVFS2 protocol. See protocol.h for more error definitions. + */ + +/* The order matches include/pvfs2-types.h in the OrangeFS source. */ +static int PINT_errno_mapping[] = { + 0, EPERM, ENOENT, EINTR, EIO, ENXIO, EBADF, EAGAIN, ENOMEM, + EFAULT, EBUSY, EEXIST, ENODEV, ENOTDIR, EISDIR, EINVAL, EMFILE, + EFBIG, ENOSPC, EROFS, EMLINK, EPIPE, EDEADLK, ENAMETOOLONG, + ENOLCK, ENOSYS, ENOTEMPTY, ELOOP, EWOULDBLOCK, ENOMSG, EUNATCH, + EBADR, EDEADLOCK, ENODATA, ETIME, ENONET, EREMOTE, ECOMM, + EPROTO, EBADMSG, EOVERFLOW, ERESTART, EMSGSIZE, EPROTOTYPE, + ENOPROTOOPT, EPROTONOSUPPORT, EOPNOTSUPP, EADDRINUSE, + EADDRNOTAVAIL, ENETDOWN, ENETUNREACH, ENETRESET, ENOBUFS, + ETIMEDOUT, ECONNREFUSED, EHOSTDOWN, EHOSTUNREACH, EALREADY, + EACCES, ECONNRESET, ERANGE +}; + +int pvfs2_normalize_to_errno(__s32 error_code) +{ + __u32 i; + + /* Success */ + if (error_code == 0) { + return 0; + /* + * This shouldn't ever happen. If it does it should be fixed on the + * server. + */ + } else if (error_code > 0) { + gossip_err("pvfs2: error status receieved.\n"); + gossip_err("pvfs2: assuming error code is inverted.\n"); + error_code = -error_code; + } + + /* + * XXX: This is very bad since error codes from PVFS2 may not be + * suitable for return into userspace. + */ + + /* + * Convert PVFS2 error values into errno values suitable for return + * from the kernel. + */ + if ((-error_code) & PVFS_NON_ERRNO_ERROR_BIT) { + if (((-error_code) & + (PVFS_ERROR_NUMBER_BITS|PVFS_NON_ERRNO_ERROR_BIT| + PVFS_ERROR_BIT)) == PVFS_ECANCEL) { + /* + * cancellation error codes generally correspond to + * a timeout from the client's perspective + */ + error_code = -ETIMEDOUT; + } else { + /* assume a default error code */ + gossip_err("pvfs2: warning: got error code without errno equivalent: %d.\n", error_code); + error_code = -EINVAL; + } + + /* Convert PVFS2 encoded errno values into regular errno values. */ + } else if ((-error_code) & PVFS_ERROR_BIT) { + i = (-error_code) & ~(PVFS_ERROR_BIT|PVFS_ERROR_CLASS_BITS); + if (i < sizeof(PINT_errno_mapping)/sizeof(*PINT_errno_mapping)) + error_code = -PINT_errno_mapping[i]; + else + error_code = -EINVAL; + + /* + * Only PVFS2 protocol error codes should ever come here. Otherwise + * there is a bug somewhere. + */ + } else { + gossip_err("pvfs2: pvfs2_normalize_to_errno: got error code which is not from PVFS2.\n"); + } + return error_code; +} + +#define NUM_MODES 11 +__s32 PVFS_util_translate_mode(int mode) +{ + int ret = 0; + int i = 0; + static int modes[NUM_MODES] = { + S_IXOTH, S_IWOTH, S_IROTH, + S_IXGRP, S_IWGRP, S_IRGRP, + S_IXUSR, S_IWUSR, S_IRUSR, + S_ISGID, S_ISUID + }; + static int pvfs2_modes[NUM_MODES] = { + PVFS_O_EXECUTE, PVFS_O_WRITE, PVFS_O_READ, + PVFS_G_EXECUTE, PVFS_G_WRITE, PVFS_G_READ, + PVFS_U_EXECUTE, PVFS_U_WRITE, PVFS_U_READ, + PVFS_G_SGID, PVFS_U_SUID + }; + + for (i = 0; i < NUM_MODES; i++) + if (mode & modes[i]) + ret |= pvfs2_modes[i]; + + return ret; +} +#undef NUM_MODES + +/* + * After obtaining a string representation of the client's debug + * keywords and their associated masks, this function is called to build an + * array of these values. + */ +int orangefs_prepare_cdm_array(char *debug_array_string) +{ + int i; + int rc = -EINVAL; + char *cds_head = NULL; + char *cds_delimiter = NULL; + int keyword_len = 0; + + gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__); + + /* + * figure out how many elements the cdm_array needs. + */ + for (i = 0; i < strlen(debug_array_string); i++) + if (debug_array_string[i] == '\n') + cdm_element_count++; + + if (!cdm_element_count) { + pr_info("No elements in client debug array string!\n"); + goto out; + } + + cdm_array = + kzalloc(cdm_element_count * sizeof(struct client_debug_mask), + GFP_KERNEL); + if (!cdm_array) { + pr_info("malloc failed for cdm_array!\n"); + rc = -ENOMEM; + goto out; + } + + cds_head = debug_array_string; + + for (i = 0; i < cdm_element_count; i++) { + cds_delimiter = strchr(cds_head, '\n'); + *cds_delimiter = '\0'; + + keyword_len = strcspn(cds_head, " "); + + cdm_array[i].keyword = kzalloc(keyword_len + 1, GFP_KERNEL); + if (!cdm_array[i].keyword) { + rc = -ENOMEM; + goto out; + } + + sscanf(cds_head, + "%s %llx %llx", + cdm_array[i].keyword, + (unsigned long long *)&(cdm_array[i].mask1), + (unsigned long long *)&(cdm_array[i].mask2)); + + if (!strcmp(cdm_array[i].keyword, PVFS2_VERBOSE)) + client_verbose_index = i; + + if (!strcmp(cdm_array[i].keyword, PVFS2_ALL)) + client_all_index = i; + + cds_head = cds_delimiter + 1; + } + + rc = cdm_element_count; + + gossip_debug(GOSSIP_UTILS_DEBUG, "%s: rc:%d:\n", __func__, rc); + +out: + + return rc; + +} + +/* + * /sys/kernel/debug/orangefs/debug-help can be catted to + * see all the available kernel and client debug keywords. + * + * When the kernel boots, we have no idea what keywords the + * client supports, nor their associated masks. + * + * We pass through this function once at boot and stamp a + * boilerplate "we don't know" message for the client in the + * debug-help file. We pass through here again when the client + * starts and then we can fill out the debug-help file fully. + * + * The client might be restarted any number of times between + * reboots, we only build the debug-help file the first time. + */ +int orangefs_prepare_debugfs_help_string(int at_boot) +{ + int rc = -EINVAL; + int i; + int byte_count = 0; + char *client_title = "Client Debug Keywords:\n"; + char *kernel_title = "Kernel Debug Keywords:\n"; + + gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__); + + if (at_boot) { + byte_count += strlen(HELP_STRING_UNINITIALIZED); + client_title = HELP_STRING_UNINITIALIZED; + } else { + /* + * fill the client keyword/mask array and remember + * how many elements there were. + */ + cdm_element_count = + orangefs_prepare_cdm_array(client_debug_array_string); + if (cdm_element_count <= 0) + goto out; + + /* Count the bytes destined for debug_help_string. */ + byte_count += strlen(client_title); + + for (i = 0; i < cdm_element_count; i++) { + byte_count += strlen(cdm_array[i].keyword + 2); + if (byte_count >= DEBUG_HELP_STRING_SIZE) { + pr_info("%s: overflow 1!\n", __func__); + goto out; + } + } + + gossip_debug(GOSSIP_UTILS_DEBUG, + "%s: cdm_element_count:%d:\n", + __func__, + cdm_element_count); + } + + byte_count += strlen(kernel_title); + for (i = 0; i < num_kmod_keyword_mask_map; i++) { + byte_count += + strlen(s_kmod_keyword_mask_map[i].keyword + 2); + if (byte_count >= DEBUG_HELP_STRING_SIZE) { + pr_info("%s: overflow 2!\n", __func__); + goto out; + } + } + + /* build debug_help_string. */ + debug_help_string = kzalloc(DEBUG_HELP_STRING_SIZE, GFP_KERNEL); + if (!debug_help_string) { + rc = -ENOMEM; + goto out; + } + + strcat(debug_help_string, client_title); + + if (!at_boot) { + for (i = 0; i < cdm_element_count; i++) { + strcat(debug_help_string, "\t"); + strcat(debug_help_string, cdm_array[i].keyword); + strcat(debug_help_string, "\n"); + } + } + + strcat(debug_help_string, "\n"); + strcat(debug_help_string, kernel_title); + + for (i = 0; i < num_kmod_keyword_mask_map; i++) { + strcat(debug_help_string, "\t"); + strcat(debug_help_string, s_kmod_keyword_mask_map[i].keyword); + strcat(debug_help_string, "\n"); + } + + rc = 0; + +out: + + return rc; + +} + +/* + * kernel = type 0 + * client = type 1 + */ +void debug_mask_to_string(void *mask, int type) +{ + int i; + int len = 0; + char *debug_string; + int element_count = 0; + + gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__); + + if (type) { + debug_string = client_debug_string; + element_count = cdm_element_count; + } else { + debug_string = kernel_debug_string; + element_count = num_kmod_keyword_mask_map; + } + + memset(debug_string, 0, PVFS2_MAX_DEBUG_STRING_LEN); + + /* + * Some keywords, like "all" or "verbose", are amalgams of + * numerous other keywords. Make a special check for those + * before grinding through the whole mask only to find out + * later... + */ + if (check_amalgam_keyword(mask, type)) + goto out; + + /* Build the debug string. */ + for (i = 0; i < element_count; i++) + if (type) + do_c_string(mask, i); + else + do_k_string(mask, i); + + len = strlen(debug_string); + + if ((len) && (type)) + client_debug_string[len - 1] = '\0'; + else if (len) + kernel_debug_string[len - 1] = '\0'; + else if (type) + strcpy(client_debug_string, "none"); + else + strcpy(kernel_debug_string, "none"); + +out: +gossip_debug(GOSSIP_UTILS_DEBUG, "%s: string:%s:\n", __func__, debug_string); + + return; + +} + +void do_k_string(void *k_mask, int index) +{ + __u64 *mask = (__u64 *) k_mask; + + if (keyword_is_amalgam((char *) s_kmod_keyword_mask_map[index].keyword)) + goto out; + + if (*mask & s_kmod_keyword_mask_map[index].mask_val) { + if ((strlen(kernel_debug_string) + + strlen(s_kmod_keyword_mask_map[index].keyword)) + < PVFS2_MAX_DEBUG_STRING_LEN - 1) { + strcat(kernel_debug_string, + s_kmod_keyword_mask_map[index].keyword); + strcat(kernel_debug_string, ","); + } else { + gossip_err("%s: overflow!\n", __func__); + strcpy(kernel_debug_string, PVFS2_ALL); + goto out; + } + } + +out: + + return; +} + +void do_c_string(void *c_mask, int index) +{ + struct client_debug_mask *mask = (struct client_debug_mask *) c_mask; + + if (keyword_is_amalgam(cdm_array[index].keyword)) + goto out; + + if ((mask->mask1 & cdm_array[index].mask1) || + (mask->mask2 & cdm_array[index].mask2)) { + if ((strlen(client_debug_string) + + strlen(cdm_array[index].keyword) + 1) + < PVFS2_MAX_DEBUG_STRING_LEN - 2) { + strcat(client_debug_string, + cdm_array[index].keyword); + strcat(client_debug_string, ","); + } else { + gossip_err("%s: overflow!\n", __func__); + strcpy(client_debug_string, PVFS2_ALL); + goto out; + } + } +out: + return; +} + +int keyword_is_amalgam(char *keyword) +{ + int rc = 0; + + if ((!strcmp(keyword, PVFS2_ALL)) || (!strcmp(keyword, PVFS2_VERBOSE))) + rc = 1; + + return rc; +} + +/* + * kernel = type 0 + * client = type 1 + * + * return 1 if we found an amalgam. + */ +int check_amalgam_keyword(void *mask, int type) +{ + __u64 *k_mask; + struct client_debug_mask *c_mask; + int k_all_index = num_kmod_keyword_mask_map - 1; + int rc = 0; + + if (type) { + c_mask = (struct client_debug_mask *) mask; + + if ((c_mask->mask1 == cdm_array[client_all_index].mask1) && + (c_mask->mask2 == cdm_array[client_all_index].mask2)) { + strcpy(client_debug_string, PVFS2_ALL); + rc = 1; + goto out; + } + + if ((c_mask->mask1 == cdm_array[client_verbose_index].mask1) && + (c_mask->mask2 == cdm_array[client_verbose_index].mask2)) { + strcpy(client_debug_string, PVFS2_VERBOSE); + rc = 1; + goto out; + } + + } else { + k_mask = (__u64 *) mask; + + if (*k_mask >= s_kmod_keyword_mask_map[k_all_index].mask_val) { + strcpy(kernel_debug_string, PVFS2_ALL); + rc = 1; + goto out; + } + } + +out: + + return rc; +} + +/* + * kernel = type 0 + * client = type 1 + */ +void debug_string_to_mask(char *debug_string, void *mask, int type) +{ + char *unchecked_keyword; + int i; + char *strsep_fodder = kstrdup(debug_string, GFP_KERNEL); + char *original_pointer; + int element_count = 0; + struct client_debug_mask *c_mask; + __u64 *k_mask; + + gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__); + + if (type) { + c_mask = (struct client_debug_mask *)mask; + element_count = cdm_element_count; + } else { + k_mask = (__u64 *)mask; + *k_mask = 0; + element_count = num_kmod_keyword_mask_map; + } + + original_pointer = strsep_fodder; + while ((unchecked_keyword = strsep(&strsep_fodder, ","))) + if (strlen(unchecked_keyword)) { + for (i = 0; i < element_count; i++) + if (type) + do_c_mask(i, + unchecked_keyword, + &c_mask); + else + do_k_mask(i, + unchecked_keyword, + &k_mask); + } + + kfree(original_pointer); +} + +void do_c_mask(int i, + char *unchecked_keyword, + struct client_debug_mask **sane_mask) +{ + + if (!strcmp(cdm_array[i].keyword, unchecked_keyword)) { + (**sane_mask).mask1 = (**sane_mask).mask1 | cdm_array[i].mask1; + (**sane_mask).mask2 = (**sane_mask).mask2 | cdm_array[i].mask2; + } +} + +void do_k_mask(int i, char *unchecked_keyword, __u64 **sane_mask) +{ + + if (!strcmp(s_kmod_keyword_mask_map[i].keyword, unchecked_keyword)) + **sane_mask = (**sane_mask) | + s_kmod_keyword_mask_map[i].mask_val; +} diff --git a/fs/orangefs/super.c b/fs/orangefs/super.c new file mode 100644 index 000000000000..9dee95293599 --- /dev/null +++ b/fs/orangefs/super.c @@ -0,0 +1,553 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" + +#include + +/* a cache for pvfs2-inode objects (i.e. pvfs2 inode private data) */ +static struct kmem_cache *pvfs2_inode_cache; + +/* list for storing pvfs2 specific superblocks in use */ +LIST_HEAD(pvfs2_superblocks); + +DEFINE_SPINLOCK(pvfs2_superblocks_lock); + +enum { + Opt_intr, + Opt_acl, + Opt_local_lock, + + Opt_err +}; + +static const match_table_t tokens = { + { Opt_acl, "acl" }, + { Opt_intr, "intr" }, + { Opt_local_lock, "local_lock" }, + { Opt_err, NULL } +}; + + +static int parse_mount_options(struct super_block *sb, char *options, + int silent) +{ + struct pvfs2_sb_info_s *pvfs2_sb = PVFS2_SB(sb); + substring_t args[MAX_OPT_ARGS]; + char *p; + + /* + * Force any potential flags that might be set from the mount + * to zero, ie, initialize to unset. + */ + sb->s_flags &= ~MS_POSIXACL; + pvfs2_sb->flags &= ~PVFS2_OPT_INTR; + pvfs2_sb->flags &= ~PVFS2_OPT_LOCAL_LOCK; + + while ((p = strsep(&options, ",")) != NULL) { + int token; + + if (!*p) + continue; + + token = match_token(p, tokens, args); + switch (token) { + case Opt_acl: + sb->s_flags |= MS_POSIXACL; + break; + case Opt_intr: + pvfs2_sb->flags |= PVFS2_OPT_INTR; + break; + case Opt_local_lock: + pvfs2_sb->flags |= PVFS2_OPT_LOCAL_LOCK; + break; + default: + goto fail; + } + } + + return 0; +fail: + if (!silent) + gossip_err("Error: mount option [%s] is not supported.\n", p); + return -EINVAL; +} + +static void pvfs2_inode_cache_ctor(void *req) +{ + struct pvfs2_inode_s *pvfs2_inode = req; + + inode_init_once(&pvfs2_inode->vfs_inode); + init_rwsem(&pvfs2_inode->xattr_sem); + + pvfs2_inode->vfs_inode.i_version = 1; +} + +static struct inode *pvfs2_alloc_inode(struct super_block *sb) +{ + struct pvfs2_inode_s *pvfs2_inode; + + pvfs2_inode = kmem_cache_alloc(pvfs2_inode_cache, + PVFS2_CACHE_ALLOC_FLAGS); + if (pvfs2_inode == NULL) { + gossip_err("Failed to allocate pvfs2_inode\n"); + return NULL; + } + + /* + * We want to clear everything except for rw_semaphore and the + * vfs_inode. + */ + memset(&pvfs2_inode->refn.khandle, 0, 16); + pvfs2_inode->refn.fs_id = PVFS_FS_ID_NULL; + pvfs2_inode->last_failed_block_index_read = 0; + memset(pvfs2_inode->link_target, 0, sizeof(pvfs2_inode->link_target)); + pvfs2_inode->pinode_flags = 0; + + gossip_debug(GOSSIP_SUPER_DEBUG, + "pvfs2_alloc_inode: allocated %p\n", + &pvfs2_inode->vfs_inode); + return &pvfs2_inode->vfs_inode; +} + +static void pvfs2_destroy_inode(struct inode *inode) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + + gossip_debug(GOSSIP_SUPER_DEBUG, + "%s: deallocated %p destroying inode %pU\n", + __func__, pvfs2_inode, get_khandle_from_ino(inode)); + + kmem_cache_free(pvfs2_inode_cache, pvfs2_inode); +} + +/* + * NOTE: information filled in here is typically reflected in the + * output of the system command 'df' +*/ +static int pvfs2_statfs(struct dentry *dentry, struct kstatfs *buf) +{ + int ret = -ENOMEM; + struct pvfs2_kernel_op_s *new_op = NULL; + int flags = 0; + struct super_block *sb = NULL; + + sb = dentry->d_sb; + + gossip_debug(GOSSIP_SUPER_DEBUG, + "pvfs2_statfs: called on sb %p (fs_id is %d)\n", + sb, + (int)(PVFS2_SB(sb)->fs_id)); + + new_op = op_alloc(PVFS2_VFS_OP_STATFS); + if (!new_op) + return ret; + new_op->upcall.req.statfs.fs_id = PVFS2_SB(sb)->fs_id; + + if (PVFS2_SB(sb)->flags & PVFS2_OPT_INTR) + flags = PVFS2_OP_INTERRUPTIBLE; + + ret = service_operation(new_op, "pvfs2_statfs", flags); + + if (new_op->downcall.status < 0) + goto out_op_release; + + gossip_debug(GOSSIP_SUPER_DEBUG, + "pvfs2_statfs: got %ld blocks available | " + "%ld blocks total | %ld block size\n", + (long)new_op->downcall.resp.statfs.blocks_avail, + (long)new_op->downcall.resp.statfs.blocks_total, + (long)new_op->downcall.resp.statfs.block_size); + + buf->f_type = sb->s_magic; + memcpy(&buf->f_fsid, &PVFS2_SB(sb)->fs_id, sizeof(buf->f_fsid)); + buf->f_bsize = new_op->downcall.resp.statfs.block_size; + buf->f_namelen = PVFS2_NAME_LEN; + + buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total; + buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail; + buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail; + buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total; + buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail; + buf->f_frsize = sb->s_blocksize; + +out_op_release: + op_release(new_op); + gossip_debug(GOSSIP_SUPER_DEBUG, "pvfs2_statfs: returning %d\n", ret); + return ret; +} + +/* + * Remount as initiated by VFS layer. We just need to reparse the mount + * options, no need to signal pvfs2-client-core about it. + */ +static int pvfs2_remount_fs(struct super_block *sb, int *flags, char *data) +{ + gossip_debug(GOSSIP_SUPER_DEBUG, "pvfs2_remount_fs: called\n"); + return parse_mount_options(sb, data, 1); +} + +/* + * Remount as initiated by pvfs2-client-core on restart. This is used to + * repopulate mount information left from previous pvfs2-client-core. + * + * the idea here is that given a valid superblock, we're + * re-initializing the user space client with the initial mount + * information specified when the super block was first initialized. + * this is very different than the first initialization/creation of a + * superblock. we use the special service_priority_operation to make + * sure that the mount gets ahead of any other pending operation that + * is waiting for servicing. this means that the pvfs2-client won't + * fail to start several times for all other pending operations before + * the client regains all of the mount information from us. + * NOTE: this function assumes that the request_mutex is already acquired! + */ +int pvfs2_remount(struct super_block *sb) +{ + struct pvfs2_kernel_op_s *new_op; + int ret = -EINVAL; + + gossip_debug(GOSSIP_SUPER_DEBUG, "pvfs2_remount: called\n"); + + new_op = op_alloc(PVFS2_VFS_OP_FS_MOUNT); + if (!new_op) + return -ENOMEM; + strncpy(new_op->upcall.req.fs_mount.pvfs2_config_server, + PVFS2_SB(sb)->devname, + PVFS_MAX_SERVER_ADDR_LEN); + + gossip_debug(GOSSIP_SUPER_DEBUG, + "Attempting PVFS2 Remount via host %s\n", + new_op->upcall.req.fs_mount.pvfs2_config_server); + + /* + * we assume that the calling function has already acquire the + * request_mutex to prevent other operations from bypassing + * this one + */ + ret = service_operation(new_op, "pvfs2_remount", + PVFS2_OP_PRIORITY | PVFS2_OP_NO_SEMAPHORE); + gossip_debug(GOSSIP_SUPER_DEBUG, + "pvfs2_remount: mount got return value of %d\n", + ret); + if (ret == 0) { + /* + * store the id assigned to this sb -- it's just a + * short-lived mapping that the system interface uses + * to map this superblock to a particular mount entry + */ + PVFS2_SB(sb)->id = new_op->downcall.resp.fs_mount.id; + PVFS2_SB(sb)->mount_pending = 0; + } + + op_release(new_op); + return ret; +} + +int fsid_key_table_initialize(void) +{ + return 0; +} + +void fsid_key_table_finalize(void) +{ +} + +/* Called whenever the VFS dirties the inode in response to atime updates */ +static void pvfs2_dirty_inode(struct inode *inode, int flags) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + + gossip_debug(GOSSIP_SUPER_DEBUG, + "pvfs2_dirty_inode: %pU\n", + get_khandle_from_ino(inode)); + SetAtimeFlag(pvfs2_inode); +} + +static const struct super_operations pvfs2_s_ops = { + .alloc_inode = pvfs2_alloc_inode, + .destroy_inode = pvfs2_destroy_inode, + .dirty_inode = pvfs2_dirty_inode, + .drop_inode = generic_delete_inode, + .statfs = pvfs2_statfs, + .remount_fs = pvfs2_remount_fs, + .show_options = generic_show_options, +}; + +static struct dentry *pvfs2_fh_to_dentry(struct super_block *sb, + struct fid *fid, + int fh_len, + int fh_type) +{ + struct pvfs2_object_kref refn; + + if (fh_len < 5 || fh_type > 2) + return NULL; + + PVFS_khandle_from(&(refn.khandle), fid->raw, 16); + refn.fs_id = (u32) fid->raw[4]; + gossip_debug(GOSSIP_SUPER_DEBUG, + "fh_to_dentry: handle %pU, fs_id %d\n", + &refn.khandle, + refn.fs_id); + + return d_obtain_alias(pvfs2_iget(sb, &refn)); +} + +static int pvfs2_encode_fh(struct inode *inode, + __u32 *fh, + int *max_len, + struct inode *parent) +{ + int len = parent ? 10 : 5; + int type = 1; + struct pvfs2_object_kref refn; + + if (*max_len < len) { + gossip_lerr("fh buffer is too small for encoding\n"); + *max_len = len; + type = 255; + goto out; + } + + refn = PVFS2_I(inode)->refn; + PVFS_khandle_to(&refn.khandle, fh, 16); + fh[4] = refn.fs_id; + + gossip_debug(GOSSIP_SUPER_DEBUG, + "Encoding fh: handle %pU, fsid %u\n", + &refn.khandle, + refn.fs_id); + + + if (parent) { + refn = PVFS2_I(parent)->refn; + PVFS_khandle_to(&refn.khandle, (char *) fh + 20, 16); + fh[9] = refn.fs_id; + + type = 2; + gossip_debug(GOSSIP_SUPER_DEBUG, + "Encoding parent: handle %pU, fsid %u\n", + &refn.khandle, + refn.fs_id); + } + *max_len = len; + +out: + return type; +} + +static struct export_operations pvfs2_export_ops = { + .encode_fh = pvfs2_encode_fh, + .fh_to_dentry = pvfs2_fh_to_dentry, +}; + +static int pvfs2_fill_sb(struct super_block *sb, void *data, int silent) +{ + int ret = -EINVAL; + struct inode *root = NULL; + struct dentry *root_dentry = NULL; + struct pvfs2_mount_sb_info_s *mount_sb_info = + (struct pvfs2_mount_sb_info_s *) data; + struct pvfs2_object_kref root_object; + + /* alloc and init our private pvfs2 sb info */ + sb->s_fs_info = + kmalloc(sizeof(struct pvfs2_sb_info_s), PVFS2_GFP_FLAGS); + if (!PVFS2_SB(sb)) + return -ENOMEM; + memset(sb->s_fs_info, 0, sizeof(struct pvfs2_sb_info_s)); + PVFS2_SB(sb)->sb = sb; + + PVFS2_SB(sb)->root_khandle = mount_sb_info->root_khandle; + PVFS2_SB(sb)->fs_id = mount_sb_info->fs_id; + PVFS2_SB(sb)->id = mount_sb_info->id; + + if (mount_sb_info->data) { + ret = parse_mount_options(sb, mount_sb_info->data, + silent); + if (ret) + return ret; + } + + /* Hang the xattr handlers off the superblock */ + sb->s_xattr = pvfs2_xattr_handlers; + sb->s_magic = PVFS2_SUPER_MAGIC; + sb->s_op = &pvfs2_s_ops; + sb->s_d_op = &pvfs2_dentry_operations; + + sb->s_blocksize = pvfs_bufmap_size_query(); + sb->s_blocksize_bits = pvfs_bufmap_shift_query(); + sb->s_maxbytes = MAX_LFS_FILESIZE; + + root_object.khandle = PVFS2_SB(sb)->root_khandle; + root_object.fs_id = PVFS2_SB(sb)->fs_id; + gossip_debug(GOSSIP_SUPER_DEBUG, + "get inode %pU, fsid %d\n", + &root_object.khandle, + root_object.fs_id); + + root = pvfs2_iget(sb, &root_object); + if (IS_ERR(root)) + return PTR_ERR(root); + + gossip_debug(GOSSIP_SUPER_DEBUG, + "Allocated root inode [%p] with mode %x\n", + root, + root->i_mode); + + /* allocates and places root dentry in dcache */ + root_dentry = d_make_root(root); + if (!root_dentry) { + iput(root); + return -ENOMEM; + } + + sb->s_export_op = &pvfs2_export_ops; + sb->s_root = root_dentry; + return 0; +} + +struct dentry *pvfs2_mount(struct file_system_type *fst, + int flags, + const char *devname, + void *data) +{ + int ret = -EINVAL; + struct super_block *sb = ERR_PTR(-EINVAL); + struct pvfs2_kernel_op_s *new_op; + struct pvfs2_mount_sb_info_s mount_sb_info; + struct dentry *d = ERR_PTR(-EINVAL); + + gossip_debug(GOSSIP_SUPER_DEBUG, + "pvfs2_mount: called with devname %s\n", + devname); + + if (!devname) { + gossip_err("ERROR: device name not specified.\n"); + return ERR_PTR(-EINVAL); + } + + new_op = op_alloc(PVFS2_VFS_OP_FS_MOUNT); + if (!new_op) + return ERR_PTR(-ENOMEM); + + strncpy(new_op->upcall.req.fs_mount.pvfs2_config_server, + devname, + PVFS_MAX_SERVER_ADDR_LEN); + + gossip_debug(GOSSIP_SUPER_DEBUG, + "Attempting PVFS2 Mount via host %s\n", + new_op->upcall.req.fs_mount.pvfs2_config_server); + + ret = service_operation(new_op, "pvfs2_mount", 0); + gossip_debug(GOSSIP_SUPER_DEBUG, + "pvfs2_mount: mount got return value of %d\n", ret); + if (ret) + goto free_op; + + if (new_op->downcall.resp.fs_mount.fs_id == PVFS_FS_ID_NULL) { + gossip_err("ERROR: Retrieved null fs_id\n"); + ret = -EINVAL; + goto free_op; + } + + /* fill in temporary structure passed to fill_sb method */ + mount_sb_info.data = data; + mount_sb_info.root_khandle = + new_op->downcall.resp.fs_mount.root_khandle; + mount_sb_info.fs_id = new_op->downcall.resp.fs_mount.fs_id; + mount_sb_info.id = new_op->downcall.resp.fs_mount.id; + + sb = sget(fst, NULL, set_anon_super, flags, NULL); + + if (IS_ERR(sb)) { + d = ERR_CAST(sb); + goto free_op; + } + + ret = pvfs2_fill_sb(sb, + (void *)&mount_sb_info, + flags & MS_SILENT ? 1 : 0); + + if (ret) { + d = ERR_PTR(ret); + goto free_op; + } + + /* + * on successful mount, store the devname and data + * used + */ + strncpy(PVFS2_SB(sb)->devname, + devname, + PVFS_MAX_SERVER_ADDR_LEN); + + /* mount_pending must be cleared */ + PVFS2_SB(sb)->mount_pending = 0; + + /* + * finally, add this sb to our list of known pvfs2 + * sb's + */ + add_pvfs2_sb(sb); + op_release(new_op); + return dget(sb->s_root); + +free_op: + gossip_err("pvfs2_mount: mount request failed with %d\n", ret); + if (ret == -EINVAL) { + gossip_err("Ensure that all pvfs2-servers have the same FS configuration files\n"); + gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n"); + } + + op_release(new_op); + + return d; +} + +void pvfs2_kill_sb(struct super_block *sb) +{ + gossip_debug(GOSSIP_SUPER_DEBUG, "pvfs2_kill_sb: called\n"); + + /* + * issue the unmount to userspace to tell it to remove the + * dynamic mount info it has for this superblock + */ + pvfs2_unmount_sb(sb); + + /* remove the sb from our list of pvfs2 specific sb's */ + remove_pvfs2_sb(sb); + + /* provided sb cleanup */ + kill_anon_super(sb); + + /* free the pvfs2 superblock private data */ + kfree(PVFS2_SB(sb)); +} + +int pvfs2_inode_cache_initialize(void) +{ + pvfs2_inode_cache = kmem_cache_create("pvfs2_inode_cache", + sizeof(struct pvfs2_inode_s), + 0, + PVFS2_CACHE_CREATE_FLAGS, + pvfs2_inode_cache_ctor); + + if (!pvfs2_inode_cache) { + gossip_err("Cannot create pvfs2_inode_cache\n"); + return -ENOMEM; + } + return 0; +} + +int pvfs2_inode_cache_finalize(void) +{ + kmem_cache_destroy(pvfs2_inode_cache); + return 0; +} diff --git a/fs/orangefs/symlink.c b/fs/orangefs/symlink.c new file mode 100644 index 000000000000..2adfceff7730 --- /dev/null +++ b/fs/orangefs/symlink.c @@ -0,0 +1,31 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" + +static const char *pvfs2_follow_link(struct dentry *dentry, void **cookie) +{ + char *target = PVFS2_I(dentry->d_inode)->link_target; + + gossip_debug(GOSSIP_INODE_DEBUG, + "%s: called on %s (target is %p)\n", + __func__, (char *)dentry->d_name.name, target); + + *cookie = target; + + return target; +} + +struct inode_operations pvfs2_symlink_inode_operations = { + .readlink = generic_readlink, + .follow_link = pvfs2_follow_link, + .setattr = pvfs2_setattr, + .getattr = pvfs2_getattr, + .listxattr = pvfs2_listxattr, + .setxattr = generic_setxattr, +}; diff --git a/fs/orangefs/upcall.h b/fs/orangefs/upcall.h new file mode 100644 index 000000000000..1e07f626aac6 --- /dev/null +++ b/fs/orangefs/upcall.h @@ -0,0 +1,255 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +#ifndef __UPCALL_H +#define __UPCALL_H + +/* + * Sanitized this header file to fix + * 32-64 bit interaction issues between + * client-core and device + */ +struct pvfs2_io_request_s { + __s32 async_vfs_io; + __s32 buf_index; + __s32 count; + __s32 __pad1; + __s64 offset; + struct pvfs2_object_kref refn; + enum PVFS_io_type io_type; + __s32 readahead_size; +}; + +struct pvfs2_iox_request_s { + __s32 buf_index; + __s32 count; + struct pvfs2_object_kref refn; + enum PVFS_io_type io_type; + __s32 __pad1; +}; + +struct pvfs2_lookup_request_s { + __s32 sym_follow; + __s32 __pad1; + struct pvfs2_object_kref parent_refn; + char d_name[PVFS2_NAME_LEN]; +}; + +struct pvfs2_create_request_s { + struct pvfs2_object_kref parent_refn; + struct PVFS_sys_attr_s attributes; + char d_name[PVFS2_NAME_LEN]; +}; + +struct pvfs2_symlink_request_s { + struct pvfs2_object_kref parent_refn; + struct PVFS_sys_attr_s attributes; + char entry_name[PVFS2_NAME_LEN]; + char target[PVFS2_NAME_LEN]; +}; + +struct pvfs2_getattr_request_s { + struct pvfs2_object_kref refn; + __u32 mask; + __u32 __pad1; +}; + +struct pvfs2_setattr_request_s { + struct pvfs2_object_kref refn; + struct PVFS_sys_attr_s attributes; +}; + +struct pvfs2_remove_request_s { + struct pvfs2_object_kref parent_refn; + char d_name[PVFS2_NAME_LEN]; +}; + +struct pvfs2_mkdir_request_s { + struct pvfs2_object_kref parent_refn; + struct PVFS_sys_attr_s attributes; + char d_name[PVFS2_NAME_LEN]; +}; + +struct pvfs2_readdir_request_s { + struct pvfs2_object_kref refn; + __u64 token; + __s32 max_dirent_count; + __s32 buf_index; +}; + +struct pvfs2_readdirplus_request_s { + struct pvfs2_object_kref refn; + __u64 token; + __s32 max_dirent_count; + __u32 mask; + __s32 buf_index; + __s32 __pad1; +}; + +struct pvfs2_rename_request_s { + struct pvfs2_object_kref old_parent_refn; + struct pvfs2_object_kref new_parent_refn; + char d_old_name[PVFS2_NAME_LEN]; + char d_new_name[PVFS2_NAME_LEN]; +}; + +struct pvfs2_statfs_request_s { + __s32 fs_id; + __s32 __pad1; +}; + +struct pvfs2_truncate_request_s { + struct pvfs2_object_kref refn; + __s64 size; +}; + +struct pvfs2_mmap_ra_cache_flush_request_s { + struct pvfs2_object_kref refn; +}; + +struct pvfs2_fs_mount_request_s { + char pvfs2_config_server[PVFS_MAX_SERVER_ADDR_LEN]; +}; + +struct pvfs2_fs_umount_request_s { + __s32 id; + __s32 fs_id; + char pvfs2_config_server[PVFS_MAX_SERVER_ADDR_LEN]; +}; + +struct pvfs2_getxattr_request_s { + struct pvfs2_object_kref refn; + __s32 key_sz; + __s32 __pad1; + char key[PVFS_MAX_XATTR_NAMELEN]; +}; + +struct pvfs2_setxattr_request_s { + struct pvfs2_object_kref refn; + struct PVFS_keyval_pair keyval; + __s32 flags; + __s32 __pad1; +}; + +struct pvfs2_listxattr_request_s { + struct pvfs2_object_kref refn; + __s32 requested_count; + __s32 __pad1; + __u64 token; +}; + +struct pvfs2_removexattr_request_s { + struct pvfs2_object_kref refn; + __s32 key_sz; + __s32 __pad1; + char key[PVFS_MAX_XATTR_NAMELEN]; +}; + +struct pvfs2_op_cancel_s { + __u64 op_tag; +}; + +struct pvfs2_fsync_request_s { + struct pvfs2_object_kref refn; +}; + +enum pvfs2_param_request_type { + PVFS2_PARAM_REQUEST_SET = 1, + PVFS2_PARAM_REQUEST_GET = 2 +}; + +enum pvfs2_param_request_op { + PVFS2_PARAM_REQUEST_OP_ACACHE_TIMEOUT_MSECS = 1, + PVFS2_PARAM_REQUEST_OP_ACACHE_HARD_LIMIT = 2, + PVFS2_PARAM_REQUEST_OP_ACACHE_SOFT_LIMIT = 3, + PVFS2_PARAM_REQUEST_OP_ACACHE_RECLAIM_PERCENTAGE = 4, + PVFS2_PARAM_REQUEST_OP_PERF_TIME_INTERVAL_SECS = 5, + PVFS2_PARAM_REQUEST_OP_PERF_HISTORY_SIZE = 6, + PVFS2_PARAM_REQUEST_OP_PERF_RESET = 7, + PVFS2_PARAM_REQUEST_OP_NCACHE_TIMEOUT_MSECS = 8, + PVFS2_PARAM_REQUEST_OP_NCACHE_HARD_LIMIT = 9, + PVFS2_PARAM_REQUEST_OP_NCACHE_SOFT_LIMIT = 10, + PVFS2_PARAM_REQUEST_OP_NCACHE_RECLAIM_PERCENTAGE = 11, + PVFS2_PARAM_REQUEST_OP_STATIC_ACACHE_TIMEOUT_MSECS = 12, + PVFS2_PARAM_REQUEST_OP_STATIC_ACACHE_HARD_LIMIT = 13, + PVFS2_PARAM_REQUEST_OP_STATIC_ACACHE_SOFT_LIMIT = 14, + PVFS2_PARAM_REQUEST_OP_STATIC_ACACHE_RECLAIM_PERCENTAGE = 15, + PVFS2_PARAM_REQUEST_OP_CLIENT_DEBUG = 16, + PVFS2_PARAM_REQUEST_OP_CCACHE_TIMEOUT_SECS = 17, + PVFS2_PARAM_REQUEST_OP_CCACHE_HARD_LIMIT = 18, + PVFS2_PARAM_REQUEST_OP_CCACHE_SOFT_LIMIT = 19, + PVFS2_PARAM_REQUEST_OP_CCACHE_RECLAIM_PERCENTAGE = 20, + PVFS2_PARAM_REQUEST_OP_CAPCACHE_TIMEOUT_SECS = 21, + PVFS2_PARAM_REQUEST_OP_CAPCACHE_HARD_LIMIT = 22, + PVFS2_PARAM_REQUEST_OP_CAPCACHE_SOFT_LIMIT = 23, + PVFS2_PARAM_REQUEST_OP_CAPCACHE_RECLAIM_PERCENTAGE = 24, + PVFS2_PARAM_REQUEST_OP_TWO_MASK_VALUES = 25, +}; + +struct pvfs2_param_request_s { + enum pvfs2_param_request_type type; + enum pvfs2_param_request_op op; + __s64 value; + char s_value[PVFS2_MAX_DEBUG_STRING_LEN]; +}; + +enum pvfs2_perf_count_request_type { + PVFS2_PERF_COUNT_REQUEST_ACACHE = 1, + PVFS2_PERF_COUNT_REQUEST_NCACHE = 2, + PVFS2_PERF_COUNT_REQUEST_CAPCACHE = 3, +}; + +struct pvfs2_perf_count_request_s { + enum pvfs2_perf_count_request_type type; + __s32 __pad1; +}; + +struct pvfs2_fs_key_request_s { + __s32 fsid; + __s32 __pad1; +}; + +struct pvfs2_upcall_s { + __s32 type; + __u32 uid; + __u32 gid; + int pid; + int tgid; + /* currently trailer is used only by readx/writex (iox) */ + __s64 trailer_size; + char *trailer_buf; + + union { + struct pvfs2_io_request_s io; + struct pvfs2_iox_request_s iox; + struct pvfs2_lookup_request_s lookup; + struct pvfs2_create_request_s create; + struct pvfs2_symlink_request_s sym; + struct pvfs2_getattr_request_s getattr; + struct pvfs2_setattr_request_s setattr; + struct pvfs2_remove_request_s remove; + struct pvfs2_mkdir_request_s mkdir; + struct pvfs2_readdir_request_s readdir; + struct pvfs2_readdirplus_request_s readdirplus; + struct pvfs2_rename_request_s rename; + struct pvfs2_statfs_request_s statfs; + struct pvfs2_truncate_request_s truncate; + struct pvfs2_mmap_ra_cache_flush_request_s ra_cache_flush; + struct pvfs2_fs_mount_request_s fs_mount; + struct pvfs2_fs_umount_request_s fs_umount; + struct pvfs2_getxattr_request_s getxattr; + struct pvfs2_setxattr_request_s setxattr; + struct pvfs2_listxattr_request_s listxattr; + struct pvfs2_removexattr_request_s removexattr; + struct pvfs2_op_cancel_s cancel; + struct pvfs2_fsync_request_s fsync; + struct pvfs2_param_request_s param; + struct pvfs2_perf_count_request_s perf_count; + struct pvfs2_fs_key_request_s fs_key; + } req; +}; + +#endif /* __UPCALL_H */ diff --git a/fs/orangefs/waitqueue.c b/fs/orangefs/waitqueue.c new file mode 100644 index 000000000000..d7b0eba043ab --- /dev/null +++ b/fs/orangefs/waitqueue.c @@ -0,0 +1,529 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * (C) 2011 Omnibond Systems + * + * Changes by Acxiom Corporation to implement generic service_operation() + * function, Copyright Acxiom Corporation, 2005. + * + * See COPYING in top-level directory. + */ + +/* + * In-kernel waitqueue operations. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" + +/* + * What we do in this function is to walk the list of operations that are + * present in the request queue and mark them as purged. + * NOTE: This is called from the device close after client-core has + * guaranteed that no new operations could appear on the list since the + * client-core is anyway going to exit. + */ +void purge_waiting_ops(void) +{ + struct pvfs2_kernel_op_s *op; + + spin_lock(&pvfs2_request_list_lock); + list_for_each_entry(op, &pvfs2_request_list, list) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "pvfs2-client-core: purging op tag %llu %s\n", + llu(op->tag), + get_opname_string(op)); + spin_lock(&op->lock); + set_op_state_purged(op); + spin_unlock(&op->lock); + wake_up_interruptible(&op->waitq); + } + spin_unlock(&pvfs2_request_list_lock); +} + +/* + * submits a PVFS2 operation and waits for it to complete + * + * Note op->downcall.status will contain the status of the operation (in + * errno format), whether provided by pvfs2-client or a result of failure to + * service the operation. If the caller wishes to distinguish, then + * op->state can be checked to see if it was serviced or not. + * + * Returns contents of op->downcall.status for convenience + */ +int service_operation(struct pvfs2_kernel_op_s *op, + const char *op_name, + int flags) +{ + /* flags to modify behavior */ + sigset_t orig_sigset; + int ret = 0; + + /* irqflags and wait_entry are only used IF the client-core aborts */ + unsigned long irqflags; + + DECLARE_WAITQUEUE(wait_entry, current); + + op->upcall.tgid = current->tgid; + op->upcall.pid = current->pid; + +retry_servicing: + op->downcall.status = 0; + gossip_debug(GOSSIP_WAIT_DEBUG, + "pvfs2: service_operation: %s %p\n", + op_name, + op); + gossip_debug(GOSSIP_WAIT_DEBUG, + "pvfs2: operation posted by process: %s, pid: %i\n", + current->comm, + current->pid); + + /* mask out signals if this operation is not to be interrupted */ + if (!(flags & PVFS2_OP_INTERRUPTIBLE)) + block_signals(&orig_sigset); + + if (!(flags & PVFS2_OP_NO_SEMAPHORE)) { + ret = mutex_lock_interruptible(&request_mutex); + /* + * check to see if we were interrupted while waiting for + * semaphore + */ + if (ret < 0) { + if (!(flags & PVFS2_OP_INTERRUPTIBLE)) + set_signals(&orig_sigset); + op->downcall.status = ret; + gossip_debug(GOSSIP_WAIT_DEBUG, + "pvfs2: service_operation interrupted.\n"); + return ret; + } + } + + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:About to call is_daemon_in_service().\n", + __func__); + + if (is_daemon_in_service() < 0) { + /* + * By incrementing the per-operation attempt counter, we + * directly go into the timeout logic while waiting for + * the matching downcall to be read + */ + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:client core is NOT in service(%d).\n", + __func__, + is_daemon_in_service()); + op->attempts++; + } + + /* queue up the operation */ + if (flags & PVFS2_OP_PRIORITY) { + add_priority_op_to_request_list(op); + } else { + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:About to call add_op_to_request_list().\n", + __func__); + add_op_to_request_list(op); + } + + if (!(flags & PVFS2_OP_NO_SEMAPHORE)) + mutex_unlock(&request_mutex); + + /* + * If we are asked to service an asynchronous operation from + * VFS perspective, we are done. + */ + if (flags & PVFS2_OP_ASYNC) + return 0; + + if (flags & PVFS2_OP_CANCELLATION) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:" + "About to call wait_for_cancellation_downcall.\n", + __func__); + ret = wait_for_cancellation_downcall(op); + } else { + ret = wait_for_matching_downcall(op); + } + + if (ret < 0) { + /* failed to get matching downcall */ + if (ret == -ETIMEDOUT) { + gossip_err("pvfs2: %s -- wait timed out; aborting attempt.\n", + op_name); + } + op->downcall.status = ret; + } else { + /* got matching downcall; make sure status is in errno format */ + op->downcall.status = + pvfs2_normalize_to_errno(op->downcall.status); + ret = op->downcall.status; + } + + if (!(flags & PVFS2_OP_INTERRUPTIBLE)) + set_signals(&orig_sigset); + + BUG_ON(ret != op->downcall.status); + /* retry if operation has not been serviced and if requested */ + if (!op_state_serviced(op) && op->downcall.status == -EAGAIN) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "pvfs2: tag %llu (%s)" + " -- operation to be retried (%d attempt)\n", + llu(op->tag), + op_name, + op->attempts + 1); + + if (!op->uses_shared_memory) + /* + * this operation doesn't use the shared memory + * system + */ + goto retry_servicing; + + /* op uses shared memory */ + if (get_bufmap_init() == 0) { + /* + * This operation uses the shared memory system AND + * the system is not yet ready. This situation occurs + * when the client-core is restarted AND there were + * operations waiting to be processed or were already + * in process. + */ + gossip_debug(GOSSIP_WAIT_DEBUG, + "uses_shared_memory is true.\n"); + gossip_debug(GOSSIP_WAIT_DEBUG, + "Client core in-service status(%d).\n", + is_daemon_in_service()); + gossip_debug(GOSSIP_WAIT_DEBUG, "bufmap_init:%d.\n", + get_bufmap_init()); + gossip_debug(GOSSIP_WAIT_DEBUG, + "operation's status is 0x%0x.\n", + op->op_state); + + /* + * let process sleep for a few seconds so shared + * memory system can be initialized. + */ + spin_lock_irqsave(&op->lock, irqflags); + add_wait_queue(&pvfs2_bufmap_init_waitq, &wait_entry); + spin_unlock_irqrestore(&op->lock, irqflags); + + set_current_state(TASK_INTERRUPTIBLE); + + /* + * Wait for pvfs_bufmap_initialize() to wake me up + * within the allotted time. + */ + ret = schedule_timeout(MSECS_TO_JIFFIES + (1000 * PVFS2_BUFMAP_WAIT_TIMEOUT_SECS)); + + gossip_debug(GOSSIP_WAIT_DEBUG, + "Value returned from schedule_timeout:" + "%d.\n", + ret); + gossip_debug(GOSSIP_WAIT_DEBUG, + "Is shared memory available? (%d).\n", + get_bufmap_init()); + + spin_lock_irqsave(&op->lock, irqflags); + remove_wait_queue(&pvfs2_bufmap_init_waitq, + &wait_entry); + spin_unlock_irqrestore(&op->lock, irqflags); + + if (get_bufmap_init() == 0) { + gossip_err("%s:The shared memory system has not started in %d seconds after the client core restarted. Aborting user's request(%s).\n", + __func__, + PVFS2_BUFMAP_WAIT_TIMEOUT_SECS, + get_opname_string(op)); + return -EIO; + } + + /* + * Return to the calling function and re-populate a + * shared memory buffer. + */ + return -EAGAIN; + } + } + + gossip_debug(GOSSIP_WAIT_DEBUG, + "pvfs2: service_operation %s returning: %d for %p.\n", + op_name, + ret, + op); + return ret; +} + +void pvfs2_clean_up_interrupted_operation(struct pvfs2_kernel_op_s *op) +{ + /* + * handle interrupted cases depending on what state we were in when + * the interruption is detected. there is a coarse grained lock + * across the operation. + * + * NOTE: be sure not to reverse lock ordering by locking an op lock + * while holding the request_list lock. Here, we first lock the op + * and then lock the appropriate list. + */ + if (!op) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s: op is null, ignoring\n", + __func__); + return; + } + + /* + * one more sanity check, make sure it's in one of the possible states + * or don't try to cancel it + */ + if (!(op_state_waiting(op) || + op_state_in_progress(op) || + op_state_serviced(op) || + op_state_purged(op))) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s: op %p not in a valid state (%0x), " + "ignoring\n", + __func__, + op, + op->op_state); + return; + } + + spin_lock(&op->lock); + + if (op_state_waiting(op)) { + /* + * upcall hasn't been read; remove op from upcall request + * list. + */ + spin_unlock(&op->lock); + remove_op_from_request_list(op); + gossip_debug(GOSSIP_WAIT_DEBUG, + "Interrupted: Removed op %p from request_list\n", + op); + } else if (op_state_in_progress(op)) { + /* op must be removed from the in progress htable */ + spin_unlock(&op->lock); + spin_lock(&htable_ops_in_progress_lock); + list_del(&op->list); + spin_unlock(&htable_ops_in_progress_lock); + gossip_debug(GOSSIP_WAIT_DEBUG, + "Interrupted: Removed op %p" + " from htable_ops_in_progress\n", + op); + } else if (!op_state_serviced(op)) { + spin_unlock(&op->lock); + gossip_err("interrupted operation is in a weird state 0x%x\n", + op->op_state); + } else { + /* + * It is not intended for execution to flow here, + * but having this unlock here makes sparse happy. + */ + gossip_err("%s: can't get here.\n", __func__); + spin_unlock(&op->lock); + } +} + +/* + * sleeps on waitqueue waiting for matching downcall. + * if client-core finishes servicing, then we are good to go. + * else if client-core exits, we get woken up here, and retry with a timeout + * + * Post when this call returns to the caller, the specified op will no + * longer be on any list or htable. + * + * Returns 0 on success and -errno on failure + * Errors are: + * EAGAIN in case we want the caller to requeue and try again.. + * EINTR/EIO/ETIMEDOUT indicating we are done trying to service this + * operation since client-core seems to be exiting too often + * or if we were interrupted. + */ +int wait_for_matching_downcall(struct pvfs2_kernel_op_s *op) +{ + int ret = -EINVAL; + DECLARE_WAITQUEUE(wait_entry, current); + + spin_lock(&op->lock); + add_wait_queue(&op->waitq, &wait_entry); + spin_unlock(&op->lock); + + while (1) { + set_current_state(TASK_INTERRUPTIBLE); + + spin_lock(&op->lock); + if (op_state_serviced(op)) { + spin_unlock(&op->lock); + ret = 0; + break; + } + spin_unlock(&op->lock); + + if (!signal_pending(current)) { + /* + * if this was our first attempt and client-core + * has not purged our operation, we are happy to + * simply wait + */ + spin_lock(&op->lock); + if (op->attempts == 0 && !op_state_purged(op)) { + spin_unlock(&op->lock); + schedule(); + } else { + spin_unlock(&op->lock); + /* + * subsequent attempts, we retry exactly once + * with timeouts + */ + if (!schedule_timeout(MSECS_TO_JIFFIES + (1000 * op_timeout_secs))) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "*** %s:" + " operation timed out (tag" + " %llu, %p, att %d)\n", + __func__, + llu(op->tag), + op, + op->attempts); + ret = -ETIMEDOUT; + pvfs2_clean_up_interrupted_operation + (op); + break; + } + } + spin_lock(&op->lock); + op->attempts++; + /* + * if the operation was purged in the meantime, it + * is better to requeue it afresh but ensure that + * we have not been purged repeatedly. This could + * happen if client-core crashes when an op + * is being serviced, so we requeue the op, client + * core crashes again so we requeue the op, client + * core starts, and so on... + */ + if (op_state_purged(op)) { + ret = (op->attempts < PVFS2_PURGE_RETRY_COUNT) ? + -EAGAIN : + -EIO; + spin_unlock(&op->lock); + gossip_debug(GOSSIP_WAIT_DEBUG, + "*** %s:" + " operation purged (tag " + "%llu, %p, att %d)\n", + __func__, + llu(op->tag), + op, + op->attempts); + pvfs2_clean_up_interrupted_operation(op); + break; + } + spin_unlock(&op->lock); + continue; + } + + gossip_debug(GOSSIP_WAIT_DEBUG, + "*** %s:" + " operation interrupted by a signal (tag " + "%llu, op %p)\n", + __func__, + llu(op->tag), + op); + pvfs2_clean_up_interrupted_operation(op); + ret = -EINTR; + break; + } + + set_current_state(TASK_RUNNING); + + spin_lock(&op->lock); + remove_wait_queue(&op->waitq, &wait_entry); + spin_unlock(&op->lock); + + return ret; +} + +/* + * similar to wait_for_matching_downcall(), but used in the special case + * of I/O cancellations. + * + * Note we need a special wait function because if this is called we already + * know that a signal is pending in current and need to service the + * cancellation upcall anyway. the only way to exit this is to either + * timeout or have the cancellation be serviced properly. + */ +int wait_for_cancellation_downcall(struct pvfs2_kernel_op_s *op) +{ + int ret = -EINVAL; + DECLARE_WAITQUEUE(wait_entry, current); + + spin_lock(&op->lock); + add_wait_queue(&op->waitq, &wait_entry); + spin_unlock(&op->lock); + + while (1) { + set_current_state(TASK_INTERRUPTIBLE); + + spin_lock(&op->lock); + if (op_state_serviced(op)) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:op-state is SERVICED.\n", + __func__); + spin_unlock(&op->lock); + ret = 0; + break; + } + spin_unlock(&op->lock); + + if (signal_pending(current)) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:operation interrupted by a signal (tag" + " %llu, op %p)\n", + __func__, + llu(op->tag), + op); + pvfs2_clean_up_interrupted_operation(op); + ret = -EINTR; + break; + } + + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:About to call schedule_timeout.\n", + __func__); + ret = + schedule_timeout(MSECS_TO_JIFFIES(1000 * op_timeout_secs)); + + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:Value returned from schedule_timeout(%d).\n", + __func__, + ret); + if (!ret) { + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:*** operation timed out: %p\n", + __func__, + op); + pvfs2_clean_up_interrupted_operation(op); + ret = -ETIMEDOUT; + break; + } + + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:Breaking out of loop, regardless of value returned by schedule_timeout.\n", + __func__); + ret = -ETIMEDOUT; + break; + } + + set_current_state(TASK_RUNNING); + + spin_lock(&op->lock); + remove_wait_queue(&op->waitq, &wait_entry); + spin_unlock(&op->lock); + + gossip_debug(GOSSIP_WAIT_DEBUG, + "%s:returning ret(%d)\n", + __func__, + ret); + + return ret; +} diff --git a/fs/orangefs/xattr.c b/fs/orangefs/xattr.c new file mode 100644 index 000000000000..227eaa47b1e1 --- /dev/null +++ b/fs/orangefs/xattr.c @@ -0,0 +1,530 @@ +/* + * (C) 2001 Clemson University and The University of Chicago + * + * See COPYING in top-level directory. + */ + +/* + * Linux VFS extended attribute operations. + */ + +#include "protocol.h" +#include "pvfs2-kernel.h" +#include "pvfs2-bufmap.h" +#include +#include + + +#define SYSTEM_PVFS2_KEY "system.pvfs2." +#define SYSTEM_PVFS2_KEY_LEN 13 + +/* + * this function returns + * 0 if the key corresponding to name is not meant to be printed as part + * of a listxattr. + * 1 if the key corresponding to name is meant to be returned as part of + * a listxattr. + * The ones that start SYSTEM_PVFS2_KEY are the ones to avoid printing. + */ +static int is_reserved_key(const char *key, size_t size) +{ + + if (size < SYSTEM_PVFS2_KEY_LEN) + return 1; + + return strncmp(key, SYSTEM_PVFS2_KEY, SYSTEM_PVFS2_KEY_LEN) ? 1 : 0; +} + +static inline int convert_to_internal_xattr_flags(int setxattr_flags) +{ + int internal_flag = 0; + + if (setxattr_flags & XATTR_REPLACE) { + /* Attribute must exist! */ + internal_flag = PVFS_XATTR_REPLACE; + } else if (setxattr_flags & XATTR_CREATE) { + /* Attribute must not exist */ + internal_flag = PVFS_XATTR_CREATE; + } + return internal_flag; +} + + +/* + * Tries to get a specified key's attributes of a given + * file into a user-specified buffer. Note that the getxattr + * interface allows for the users to probe the size of an + * extended attribute by passing in a value of 0 to size. + * Thus our return value is always the size of the attribute + * unless the key does not exist for the file and/or if + * there were errors in fetching the attribute value. + */ +ssize_t pvfs2_inode_getxattr(struct inode *inode, const char *prefix, + const char *name, void *buffer, size_t size) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_kernel_op_s *new_op = NULL; + ssize_t ret = -ENOMEM; + ssize_t length = 0; + int fsuid; + int fsgid; + + gossip_debug(GOSSIP_XATTR_DEBUG, + "%s: prefix %s name %s, buffer_size %zd\n", + __func__, prefix, name, size); + + if (name == NULL || (size > 0 && buffer == NULL)) { + gossip_err("pvfs2_inode_getxattr: bogus NULL pointers\n"); + return -EINVAL; + } + if ((strlen(name) + strlen(prefix)) >= PVFS_MAX_XATTR_NAMELEN) { + gossip_err("Invalid key length (%d)\n", + (int)(strlen(name) + strlen(prefix))); + return -EINVAL; + } + + fsuid = from_kuid(current_user_ns(), current_fsuid()); + fsgid = from_kgid(current_user_ns(), current_fsgid()); + + gossip_debug(GOSSIP_XATTR_DEBUG, + "getxattr on inode %pU, name %s " + "(uid %o, gid %o)\n", + get_khandle_from_ino(inode), + name, + fsuid, + fsgid); + + down_read(&pvfs2_inode->xattr_sem); + + new_op = op_alloc(PVFS2_VFS_OP_GETXATTR); + if (!new_op) + goto out_unlock; + + new_op->upcall.req.getxattr.refn = pvfs2_inode->refn; + ret = snprintf((char *)new_op->upcall.req.getxattr.key, + PVFS_MAX_XATTR_NAMELEN, "%s%s", prefix, name); + + /* + * NOTE: Although keys are meant to be NULL terminated textual + * strings, I am going to explicitly pass the length just in case + * we change this later on... + */ + new_op->upcall.req.getxattr.key_sz = ret + 1; + + ret = service_operation(new_op, "pvfs2_inode_getxattr", + get_interruptible_flag(inode)); + if (ret != 0) { + if (ret == -ENOENT) { + ret = -ENODATA; + gossip_debug(GOSSIP_XATTR_DEBUG, + "pvfs2_inode_getxattr: inode %pU key %s" + " does not exist!\n", + get_khandle_from_ino(inode), + (char *)new_op->upcall.req.getxattr.key); + } + goto out_release_op; + } + + /* + * Length returned includes null terminator. + */ + length = new_op->downcall.resp.getxattr.val_sz; + + /* + * Just return the length of the queried attribute. + */ + if (size == 0) { + ret = length; + goto out_release_op; + } + + /* + * Check to see if key length is > provided buffer size. + */ + if (length > size) { + ret = -ERANGE; + goto out_release_op; + } + + memset(buffer, 0, size); + memcpy(buffer, new_op->downcall.resp.getxattr.val, length); + gossip_debug(GOSSIP_XATTR_DEBUG, + "pvfs2_inode_getxattr: inode %pU " + "key %s key_sz %d, val_len %d\n", + get_khandle_from_ino(inode), + (char *)new_op-> + upcall.req.getxattr.key, + (int)new_op-> + upcall.req.getxattr.key_sz, + (int)ret); + + ret = length; + +out_release_op: + op_release(new_op); +out_unlock: + up_read(&pvfs2_inode->xattr_sem); + return ret; +} + +static int pvfs2_inode_removexattr(struct inode *inode, + const char *prefix, + const char *name, + int flags) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_kernel_op_s *new_op = NULL; + int ret = -ENOMEM; + + down_write(&pvfs2_inode->xattr_sem); + new_op = op_alloc(PVFS2_VFS_OP_REMOVEXATTR); + if (!new_op) + goto out_unlock; + + new_op->upcall.req.removexattr.refn = pvfs2_inode->refn; + /* + * NOTE: Although keys are meant to be NULL terminated + * textual strings, I am going to explicitly pass the + * length just in case we change this later on... + */ + ret = snprintf((char *)new_op->upcall.req.removexattr.key, + PVFS_MAX_XATTR_NAMELEN, + "%s%s", + (prefix ? prefix : ""), + name); + new_op->upcall.req.removexattr.key_sz = ret + 1; + + gossip_debug(GOSSIP_XATTR_DEBUG, + "pvfs2_inode_removexattr: key %s, key_sz %d\n", + (char *)new_op->upcall.req.removexattr.key, + (int)new_op->upcall.req.removexattr.key_sz); + + ret = service_operation(new_op, + "pvfs2_inode_removexattr", + get_interruptible_flag(inode)); + if (ret == -ENOENT) { + /* + * Request to replace a non-existent attribute is an error. + */ + if (flags & XATTR_REPLACE) + ret = -ENODATA; + else + ret = 0; + } + + gossip_debug(GOSSIP_XATTR_DEBUG, + "pvfs2_inode_removexattr: returning %d\n", ret); + + op_release(new_op); +out_unlock: + up_write(&pvfs2_inode->xattr_sem); + return ret; +} + +/* + * Tries to set an attribute for a given key on a file. + * + * Returns a -ve number on error and 0 on success. Key is text, but value + * can be binary! + */ +int pvfs2_inode_setxattr(struct inode *inode, const char *prefix, + const char *name, const void *value, size_t size, int flags) +{ + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_kernel_op_s *new_op; + int internal_flag = 0; + int ret = -ENOMEM; + + gossip_debug(GOSSIP_XATTR_DEBUG, + "%s: prefix %s, name %s, buffer_size %zd\n", + __func__, prefix, name, size); + + if (size < 0 || + size >= PVFS_MAX_XATTR_VALUELEN || + flags < 0) { + gossip_err("pvfs2_inode_setxattr: bogus values of size(%d), flags(%d)\n", + (int)size, + flags); + return -EINVAL; + } + + if (name == NULL || + (size > 0 && value == NULL)) { + gossip_err("pvfs2_inode_setxattr: bogus NULL pointers!\n"); + return -EINVAL; + } + + internal_flag = convert_to_internal_xattr_flags(flags); + + if (prefix) { + if (strlen(name) + strlen(prefix) >= PVFS_MAX_XATTR_NAMELEN) { + gossip_err + ("pvfs2_inode_setxattr: bogus key size (%d)\n", + (int)(strlen(name) + strlen(prefix))); + return -EINVAL; + } + } else { + if (strlen(name) >= PVFS_MAX_XATTR_NAMELEN) { + gossip_err + ("pvfs2_inode_setxattr: bogus key size (%d)\n", + (int)(strlen(name))); + return -EINVAL; + } + } + + /* This is equivalent to a removexattr */ + if (size == 0 && value == NULL) { + gossip_debug(GOSSIP_XATTR_DEBUG, + "removing xattr (%s%s)\n", + prefix, + name); + return pvfs2_inode_removexattr(inode, prefix, name, flags); + } + + gossip_debug(GOSSIP_XATTR_DEBUG, + "setxattr on inode %pU, name %s\n", + get_khandle_from_ino(inode), + name); + + down_write(&pvfs2_inode->xattr_sem); + new_op = op_alloc(PVFS2_VFS_OP_SETXATTR); + if (!new_op) + goto out_unlock; + + + new_op->upcall.req.setxattr.refn = pvfs2_inode->refn; + new_op->upcall.req.setxattr.flags = internal_flag; + /* + * NOTE: Although keys are meant to be NULL terminated textual + * strings, I am going to explicitly pass the length just in + * case we change this later on... + */ + ret = snprintf((char *)new_op->upcall.req.setxattr.keyval.key, + PVFS_MAX_XATTR_NAMELEN, + "%s%s", + prefix, name); + new_op->upcall.req.setxattr.keyval.key_sz = ret + 1; + memcpy(new_op->upcall.req.setxattr.keyval.val, value, size); + new_op->upcall.req.setxattr.keyval.val_sz = size; + + gossip_debug(GOSSIP_XATTR_DEBUG, + "pvfs2_inode_setxattr: key %s, key_sz %d " + " value size %zd\n", + (char *)new_op->upcall.req.setxattr.keyval.key, + (int)new_op->upcall.req.setxattr.keyval.key_sz, + size); + + ret = service_operation(new_op, + "pvfs2_inode_setxattr", + get_interruptible_flag(inode)); + + gossip_debug(GOSSIP_XATTR_DEBUG, + "pvfs2_inode_setxattr: returning %d\n", + ret); + + /* when request is serviced properly, free req op struct */ + op_release(new_op); +out_unlock: + up_write(&pvfs2_inode->xattr_sem); + return ret; +} + +/* + * Tries to get a specified object's keys into a user-specified buffer of a + * given size. Note that like the previous instances of xattr routines, this + * also allows you to pass in a NULL pointer and 0 size to probe the size for + * subsequent memory allocations. Thus our return value is always the size of + * all the keys unless there were errors in fetching the keys! + */ +ssize_t pvfs2_listxattr(struct dentry *dentry, char *buffer, size_t size) +{ + struct inode *inode = dentry->d_inode; + struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); + struct pvfs2_kernel_op_s *new_op; + __u64 token = PVFS_ITERATE_START; + ssize_t ret = -ENOMEM; + ssize_t total = 0; + ssize_t length = 0; + int count_keys = 0; + int key_size; + int i = 0; + + if (size > 0 && buffer == NULL) { + gossip_err("%s: bogus NULL pointers\n", __func__); + return -EINVAL; + } + if (size < 0) { + gossip_err("Invalid size (%d)\n", (int)size); + return -EINVAL; + } + + down_read(&pvfs2_inode->xattr_sem); + new_op = op_alloc(PVFS2_VFS_OP_LISTXATTR); + if (!new_op) + goto out_unlock; + + if (buffer && size > 0) + memset(buffer, 0, size); + +try_again: + key_size = 0; + new_op->upcall.req.listxattr.refn = pvfs2_inode->refn; + new_op->upcall.req.listxattr.token = token; + new_op->upcall.req.listxattr.requested_count = + (size == 0) ? 0 : PVFS_MAX_XATTR_LISTLEN; + ret = service_operation(new_op, __func__, + get_interruptible_flag(inode)); + if (ret != 0) + goto done; + + if (size == 0) { + /* + * This is a bit of a big upper limit, but I did not want to + * spend too much time getting this correct, since users end + * up allocating memory rather than us... + */ + total = new_op->downcall.resp.listxattr.returned_count * + PVFS_MAX_XATTR_NAMELEN; + goto done; + } + + length = new_op->downcall.resp.listxattr.keylen; + if (length == 0) + goto done; + + /* + * Check to see how much can be fit in the buffer. Fit only whole keys. + */ + for (i = 0; i < new_op->downcall.resp.listxattr.returned_count; i++) { + if (total + new_op->downcall.resp.listxattr.lengths[i] > size) + goto done; + + /* + * Since many dumb programs try to setxattr() on our reserved + * xattrs this is a feeble attempt at defeating those by not + * listing them in the output of listxattr.. sigh + */ + if (is_reserved_key(new_op->downcall.resp.listxattr.key + + key_size, + new_op->downcall.resp. + listxattr.lengths[i])) { + gossip_debug(GOSSIP_XATTR_DEBUG, "Copying key %d -> %s\n", + i, new_op->downcall.resp.listxattr.key + + key_size); + memcpy(buffer + total, + new_op->downcall.resp.listxattr.key + key_size, + new_op->downcall.resp.listxattr.lengths[i]); + total += new_op->downcall.resp.listxattr.lengths[i]; + count_keys++; + } else { + gossip_debug(GOSSIP_XATTR_DEBUG, "[RESERVED] key %d -> %s\n", + i, new_op->downcall.resp.listxattr.key + + key_size); + } + key_size += new_op->downcall.resp.listxattr.lengths[i]; + } + + /* + * Since the buffer was large enough, we might have to continue + * fetching more keys! + */ + token = new_op->downcall.resp.listxattr.token; + if (token != PVFS_ITERATE_END) + goto try_again; + +done: + gossip_debug(GOSSIP_XATTR_DEBUG, "%s: returning %d" + " [size of buffer %ld] (filled in %d keys)\n", + __func__, + ret ? (int)ret : (int)total, + (long)size, + count_keys); + op_release(new_op); + if (ret == 0) + ret = total; +out_unlock: + up_read(&pvfs2_inode->xattr_sem); + return ret; +} + +int pvfs2_xattr_set_default(struct dentry *dentry, + const char *name, + const void *buffer, + size_t size, + int flags, + int handler_flags) +{ + return pvfs2_inode_setxattr(dentry->d_inode, + PVFS2_XATTR_NAME_DEFAULT_PREFIX, + name, + buffer, + size, + flags); +} + +int pvfs2_xattr_get_default(struct dentry *dentry, + const char *name, + void *buffer, + size_t size, + int handler_flags) +{ + return pvfs2_inode_getxattr(dentry->d_inode, + PVFS2_XATTR_NAME_DEFAULT_PREFIX, + name, + buffer, + size); + +} + +static int pvfs2_xattr_set_trusted(struct dentry *dentry, + const char *name, + const void *buffer, + size_t size, + int flags, + int handler_flags) +{ + return pvfs2_inode_setxattr(dentry->d_inode, + PVFS2_XATTR_NAME_TRUSTED_PREFIX, + name, + buffer, + size, + flags); +} + +static int pvfs2_xattr_get_trusted(struct dentry *dentry, + const char *name, + void *buffer, + size_t size, + int handler_flags) +{ + return pvfs2_inode_getxattr(dentry->d_inode, + PVFS2_XATTR_NAME_TRUSTED_PREFIX, + name, + buffer, + size); +} + +static struct xattr_handler pvfs2_xattr_trusted_handler = { + .prefix = PVFS2_XATTR_NAME_TRUSTED_PREFIX, + .get = pvfs2_xattr_get_trusted, + .set = pvfs2_xattr_set_trusted, +}; + +static struct xattr_handler pvfs2_xattr_default_handler = { + /* + * NOTE: this is set to be the empty string. + * so that all un-prefixed xattrs keys get caught + * here! + */ + .prefix = PVFS2_XATTR_NAME_DEFAULT_PREFIX, + .get = pvfs2_xattr_get_default, + .set = pvfs2_xattr_set_default, +}; + +const struct xattr_handler *pvfs2_xattr_handlers[] = { + &posix_acl_access_xattr_handler, + &posix_acl_default_xattr_handler, + &pvfs2_xattr_trusted_handler, + &pvfs2_xattr_default_handler, + NULL +};