fs/orangefs/super.c

   1 /*
   2  * (C) 2001 Clemson University and The University of Chicago
   3  *
   4  * See COPYING in top-level directory.
   5  */
   6
   7 #include "protocol.h"
   8 #include "pvfs2-kernel.h"
   9 #include "pvfs2-bufmap.h"
  10
  11 #include <linux/parser.h>
  12
  13 /* a cache for pvfs2-inode objects (i.e. pvfs2 inode private data) */
  14 static struct kmem_cache *pvfs2_inode_cache;
  15
  16 /* list for storing pvfs2 specific superblocks in use */
  17 LIST_HEAD(pvfs2_superblocks);
  18
  19 DEFINE_SPINLOCK(pvfs2_superblocks_lock);
  20
  21 enum {
  22         Opt_intr,
  23         Opt_acl,
  24         Opt_local_lock,
  25
  26         Opt_err
  27 };
  28
  29 static const match_table_t tokens = {
  30         { Opt_acl,              "acl" },
  31         { Opt_intr,             "intr" },
  32         { Opt_local_lock,       "local_lock" },
  33         { Opt_err,      NULL }
  34 };
  35
  36
  37 static int parse_mount_options(struct super_block *sb, char *options,
  38                 int silent)
  39 {
  40         struct pvfs2_sb_info_s *pvfs2_sb = PVFS2_SB(sb);
  41         substring_t args[MAX_OPT_ARGS];
  42         char *p;
  43
  44         /*
  45          * Force any potential flags that might be set from the mount
  46          * to zero, ie, initialize to unset.
  47          */
  48         sb->s_flags &= ~MS_POSIXACL;
  49         pvfs2_sb->flags &= ~PVFS2_OPT_INTR;
  50         pvfs2_sb->flags &= ~PVFS2_OPT_LOCAL_LOCK;
  51
  52         while ((p = strsep(&options, ",")) != NULL) {
  53                 int token;
  54
  55                 if (!*p)
  56                         continue;
  57
  58                 token = match_token(p, tokens, args);
  59                 switch (token) {
  60                 case Opt_acl:
  61                         sb->s_flags |= MS_POSIXACL;
  62                         break;
  63                 case Opt_intr:
  64                         pvfs2_sb->flags |= PVFS2_OPT_INTR;
  65                         break;
  66                 case Opt_local_lock:
  67                         pvfs2_sb->flags |= PVFS2_OPT_LOCAL_LOCK;
  68                         break;
  69                 default:
  70                         goto fail;
  71                 }
  72         }
  73
  74         return 0;
  75 fail:
  76         if (!silent)
  77                 gossip_err("Error: mount option [%s] is not supported.\n", p);
  78         return -EINVAL;
  79 }
  80
  81 static void pvfs2_inode_cache_ctor(void *req)
  82 {
  83         struct pvfs2_inode_s *pvfs2_inode = req;
  84
  85         inode_init_once(&pvfs2_inode->vfs_inode);
  86         init_rwsem(&pvfs2_inode->xattr_sem);
  87
  88         pvfs2_inode->vfs_inode.i_version = 1;
  89 }
  90
  91 static struct inode *pvfs2_alloc_inode(struct super_block *sb)
  92 {
  93         struct pvfs2_inode_s *pvfs2_inode;
  94
  95         pvfs2_inode = kmem_cache_alloc(pvfs2_inode_cache,
  96                                        PVFS2_CACHE_ALLOC_FLAGS);
  97         if (pvfs2_inode == NULL) {
  98                 gossip_err("Failed to allocate pvfs2_inode\n");
  99                 return NULL;
 100         }
 101
 102         /*
 103          * We want to clear everything except for rw_semaphore and the
 104          * vfs_inode.
 105          */
 106         memset(&pvfs2_inode->refn.khandle, 0, 16);
 107         pvfs2_inode->refn.fs_id = PVFS_FS_ID_NULL;
 108         pvfs2_inode->last_failed_block_index_read = 0;
 109         memset(pvfs2_inode->link_target, 0, sizeof(pvfs2_inode->link_target));
 110         pvfs2_inode->pinode_flags = 0;
 111
 112         gossip_debug(GOSSIP_SUPER_DEBUG,
 113                      "pvfs2_alloc_inode: allocated %p\n",
 114                      &pvfs2_inode->vfs_inode);
 115         return &pvfs2_inode->vfs_inode;
 116 }
 117
 118 static void pvfs2_destroy_inode(struct inode *inode)
 119 {
 120         struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
 121
 122         gossip_debug(GOSSIP_SUPER_DEBUG,
 123                         "%s: deallocated %p destroying inode %pU\n",
 124                         __func__, pvfs2_inode, get_khandle_from_ino(inode));
 125
 126         kmem_cache_free(pvfs2_inode_cache, pvfs2_inode);
 127 }
 128
 129 /*
 130  * NOTE: information filled in here is typically reflected in the
 131  * output of the system command 'df'
 132 */
 133 static int pvfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 134 {
 135         int ret = -ENOMEM;
 136         struct pvfs2_kernel_op_s *new_op = NULL;
 137         int flags = 0;
 138         struct super_block *sb = NULL;
 139
 140         sb = dentry->d_sb;
 141
 142         gossip_debug(GOSSIP_SUPER_DEBUG,
 143                      "pvfs2_statfs: called on sb %p (fs_id is %d)\n",
 144                      sb,
 145                      (int)(PVFS2_SB(sb)->fs_id));
 146
 147         new_op = op_alloc(PVFS2_VFS_OP_STATFS);
 148         if (!new_op)
 149                 return ret;
 150         new_op->upcall.req.statfs.fs_id = PVFS2_SB(sb)->fs_id;
 151
 152         if (PVFS2_SB(sb)->flags & PVFS2_OPT_INTR)
 153                 flags = PVFS2_OP_INTERRUPTIBLE;
 154
 155         ret = service_operation(new_op, "pvfs2_statfs", flags);
 156
 157         if (new_op->downcall.status < 0)
 158                 goto out_op_release;
 159
 160         gossip_debug(GOSSIP_SUPER_DEBUG,
 161                      "pvfs2_statfs: got %ld blocks available | "
 162                      "%ld blocks total | %ld block size\n",
 163                      (long)new_op->downcall.resp.statfs.blocks_avail,
 164                      (long)new_op->downcall.resp.statfs.blocks_total,
 165                      (long)new_op->downcall.resp.statfs.block_size);
 166
 167         buf->f_type = sb->s_magic;
 168         memcpy(&buf->f_fsid, &PVFS2_SB(sb)->fs_id, sizeof(buf->f_fsid));
 169         buf->f_bsize = new_op->downcall.resp.statfs.block_size;
 170         buf->f_namelen = PVFS2_NAME_LEN;
 171
 172         buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total;
 173         buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
 174         buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
 175         buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total;
 176         buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail;
 177         buf->f_frsize = sb->s_blocksize;
 178
 179 out_op_release:
 180         op_release(new_op);
 181         gossip_debug(GOSSIP_SUPER_DEBUG, "pvfs2_statfs: returning %d\n", ret);
 182         return ret;
 183 }
 184
 185 /*
 186  * Remount as initiated by VFS layer.  We just need to reparse the mount
 187  * options, no need to signal pvfs2-client-core about it.
 188  */
 189 static int pvfs2_remount_fs(struct super_block *sb, int *flags, char *data)
 190 {
 191         gossip_debug(GOSSIP_SUPER_DEBUG, "pvfs2_remount_fs: called\n");
 192         return parse_mount_options(sb, data, 1);
 193 }
 194
 195 /*
 196  * Remount as initiated by pvfs2-client-core on restart.  This is used to
 197  * repopulate mount information left from previous pvfs2-client-core.
 198  *
 199  * the idea here is that given a valid superblock, we're
 200  * re-initializing the user space client with the initial mount
 201  * information specified when the super block was first initialized.
 202  * this is very different than the first initialization/creation of a
 203  * superblock.  we use the special service_priority_operation to make
 204  * sure that the mount gets ahead of any other pending operation that
 205  * is waiting for servicing.  this means that the pvfs2-client won't
 206  * fail to start several times for all other pending operations before
 207  * the client regains all of the mount information from us.
 208  * NOTE: this function assumes that the request_mutex is already acquired!
 209  */
 210 int pvfs2_remount(struct super_block *sb)
 211 {
 212         struct pvfs2_kernel_op_s *new_op;
 213         int ret = -EINVAL;
 214
 215         gossip_debug(GOSSIP_SUPER_DEBUG, "pvfs2_remount: called\n");
 216
 217         new_op = op_alloc(PVFS2_VFS_OP_FS_MOUNT);
 218         if (!new_op)
 219                 return -ENOMEM;
 220         strncpy(new_op->upcall.req.fs_mount.pvfs2_config_server,
 221                 PVFS2_SB(sb)->devname,
 222                 PVFS_MAX_SERVER_ADDR_LEN);
 223
 224         gossip_debug(GOSSIP_SUPER_DEBUG,
 225                      "Attempting PVFS2 Remount via host %s\n",
 226                      new_op->upcall.req.fs_mount.pvfs2_config_server);
 227
 228         /*
 229          * we assume that the calling function has already acquire the
 230          * request_mutex to prevent other operations from bypassing
 231          * this one
 232          */
 233         ret = service_operation(new_op, "pvfs2_remount",
 234                 PVFS2_OP_PRIORITY | PVFS2_OP_NO_SEMAPHORE);
 235         gossip_debug(GOSSIP_SUPER_DEBUG,
 236                      "pvfs2_remount: mount got return value of %d\n",
 237                      ret);
 238         if (ret == 0) {
 239                 /*
 240                  * store the id assigned to this sb -- it's just a
 241                  * short-lived mapping that the system interface uses
 242                  * to map this superblock to a particular mount entry
 243                  */
 244                 PVFS2_SB(sb)->id = new_op->downcall.resp.fs_mount.id;
 245                 PVFS2_SB(sb)->mount_pending = 0;
 246         }
 247
 248         op_release(new_op);
 249         return ret;
 250 }
 251
 252 int fsid_key_table_initialize(void)
 253 {
 254         return 0;
 255 }
 256
 257 void fsid_key_table_finalize(void)
 258 {
 259 }
 260
 261 /* Called whenever the VFS dirties the inode in response to atime updates */
 262 static void pvfs2_dirty_inode(struct inode *inode, int flags)
 263 {
 264         struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
 265
 266         gossip_debug(GOSSIP_SUPER_DEBUG,
 267                      "pvfs2_dirty_inode: %pU\n",
 268                      get_khandle_from_ino(inode));
 269         SetAtimeFlag(pvfs2_inode);
 270 }
 271
 272 struct super_operations pvfs2_s_ops = {
 273         .alloc_inode = pvfs2_alloc_inode,
 274         .destroy_inode = pvfs2_destroy_inode,
 275         .dirty_inode = pvfs2_dirty_inode,
 276         .drop_inode = generic_delete_inode,
 277         .statfs = pvfs2_statfs,
 278         .remount_fs = pvfs2_remount_fs,
 279         .show_options = generic_show_options,
 280 };
 281
 282 struct dentry *pvfs2_fh_to_dentry(struct super_block *sb,
 283                                   struct fid *fid,
 284                                   int fh_len,
 285                                   int fh_type)
 286 {
 287         struct pvfs2_object_kref refn;
 288
 289         if (fh_len < 5 || fh_type > 2)
 290                 return NULL;
 291
 292         PVFS_khandle_from(&(refn.khandle), fid->raw, 16);
 293         refn.fs_id = (u32) fid->raw[4];
 294         gossip_debug(GOSSIP_SUPER_DEBUG,
 295                      "fh_to_dentry: handle %pU, fs_id %d\n",
 296                      &refn.khandle,
 297                      refn.fs_id);
 298
 299         return d_obtain_alias(pvfs2_iget(sb, &refn));
 300 }
 301
 302 int pvfs2_encode_fh(struct inode *inode,
 303                     __u32 *fh,
 304                     int *max_len,
 305                     struct inode *parent)
 306 {
 307         int len = parent ? 10 : 5;
 308         int type = 1;
 309         struct pvfs2_object_kref refn;
 310
 311         if (*max_len < len) {
 312                 gossip_lerr("fh buffer is too small for encoding\n");
 313                 *max_len = len;
 314                 type = 255;
 315                 goto out;
 316         }
 317
 318         refn = PVFS2_I(inode)->refn;
 319         PVFS_khandle_to(&refn.khandle, fh, 16);
 320         fh[4] = refn.fs_id;
 321
 322         gossip_debug(GOSSIP_SUPER_DEBUG,
 323                      "Encoding fh: handle %pU, fsid %u\n",
 324                      &refn.khandle,
 325                      refn.fs_id);
 326
 327
 328         if (parent) {
 329                 refn = PVFS2_I(parent)->refn;
 330                 PVFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
 331                 fh[9] = refn.fs_id;
 332
 333                 type = 2;
 334                 gossip_debug(GOSSIP_SUPER_DEBUG,
 335                              "Encoding parent: handle %pU, fsid %u\n",
 336                              &refn.khandle,
 337                              refn.fs_id);
 338         }
 339         *max_len = len;
 340
 341 out:
 342         return type;
 343 }
 344
 345 static struct export_operations pvfs2_export_ops = {
 346         .encode_fh = pvfs2_encode_fh,
 347         .fh_to_dentry = pvfs2_fh_to_dentry,
 348 };
 349
 350 int pvfs2_fill_sb(struct super_block *sb, void *data, int silent)
 351 {
 352         int ret = -EINVAL;
 353         struct inode *root = NULL;
 354         struct dentry *root_dentry = NULL;
 355         struct pvfs2_mount_sb_info_s *mount_sb_info =
 356                 (struct pvfs2_mount_sb_info_s *) data;
 357         struct pvfs2_object_kref root_object;
 358
 359         /* alloc and init our private pvfs2 sb info */
 360         sb->s_fs_info =
 361                 kmalloc(sizeof(struct pvfs2_sb_info_s), PVFS2_GFP_FLAGS);
 362         if (!PVFS2_SB(sb))
 363                 return -ENOMEM;
 364         memset(sb->s_fs_info, 0, sizeof(struct pvfs2_sb_info_s));
 365         PVFS2_SB(sb)->sb = sb;
 366
 367         PVFS2_SB(sb)->root_khandle = mount_sb_info->root_khandle;
 368         PVFS2_SB(sb)->fs_id = mount_sb_info->fs_id;
 369         PVFS2_SB(sb)->id = mount_sb_info->id;
 370
 371         if (mount_sb_info->data) {
 372                 ret = parse_mount_options(sb, mount_sb_info->data,
 373                                           silent);
 374                 if (ret)
 375                         return ret;
 376         }
 377
 378         /* Hang the xattr handlers off the superblock */
 379         sb->s_xattr = pvfs2_xattr_handlers;
 380         sb->s_magic = PVFS2_SUPER_MAGIC;
 381         sb->s_op = &pvfs2_s_ops;
 382         sb->s_d_op = &pvfs2_dentry_operations;
 383
 384         sb->s_blocksize = pvfs_bufmap_size_query();
 385         sb->s_blocksize_bits = pvfs_bufmap_shift_query();
 386         sb->s_maxbytes = MAX_LFS_FILESIZE;
 387
 388         root_object.khandle = PVFS2_SB(sb)->root_khandle;
 389         root_object.fs_id = PVFS2_SB(sb)->fs_id;
 390         gossip_debug(GOSSIP_SUPER_DEBUG,
 391                      "get inode %pU, fsid %d\n",
 392                      &root_object.khandle,
 393                      root_object.fs_id);
 394
 395         root = pvfs2_iget(sb, &root_object);
 396         if (IS_ERR(root))
 397                 return PTR_ERR(root);
 398
 399         gossip_debug(GOSSIP_SUPER_DEBUG,
 400                      "Allocated root inode [%p] with mode %x\n",
 401                      root,
 402                      root->i_mode);
 403
 404         /* allocates and places root dentry in dcache */
 405         root_dentry = d_make_root(root);
 406         if (!root_dentry) {
 407                 iput(root);
 408                 return -ENOMEM;
 409         }
 410
 411         sb->s_export_op = &pvfs2_export_ops;
 412         sb->s_root = root_dentry;
 413         return 0;
 414 }
 415
 416 struct dentry *pvfs2_mount(struct file_system_type *fst,
 417                            int flags,
 418                            const char *devname,
 419                            void *data)
 420 {
 421         int ret = -EINVAL;
 422         struct super_block *sb = ERR_PTR(-EINVAL);
 423         struct pvfs2_kernel_op_s *new_op;
 424         struct pvfs2_mount_sb_info_s mount_sb_info;
 425         struct dentry *mnt_sb_d = ERR_PTR(-EINVAL);
 426
 427         gossip_debug(GOSSIP_SUPER_DEBUG,
 428                      "pvfs2_mount: called with devname %s\n",
 429                      devname);
 430
 431         if (!devname) {
 432                 gossip_err("ERROR: device name not specified.\n");
 433                 return ERR_PTR(-EINVAL);
 434         }
 435
 436         new_op = op_alloc(PVFS2_VFS_OP_FS_MOUNT);
 437         if (!new_op)
 438                 return ERR_PTR(-ENOMEM);
 439
 440         strncpy(new_op->upcall.req.fs_mount.pvfs2_config_server,
 441                 devname,
 442                 PVFS_MAX_SERVER_ADDR_LEN);
 443
 444         gossip_debug(GOSSIP_SUPER_DEBUG,
 445                      "Attempting PVFS2 Mount via host %s\n",
 446                      new_op->upcall.req.fs_mount.pvfs2_config_server);
 447
 448         ret = service_operation(new_op, "pvfs2_mount", 0);
 449         gossip_debug(GOSSIP_SUPER_DEBUG,
 450                      "pvfs2_mount: mount got return value of %d\n", ret);
 451         if (ret)
 452                 goto free_op;
 453
 454         if (new_op->downcall.resp.fs_mount.fs_id == PVFS_FS_ID_NULL) {
 455                 gossip_err("ERROR: Retrieved null fs_id\n");
 456                 ret = -EINVAL;
 457                 goto free_op;
 458         }
 459
 460         /* fill in temporary structure passed to fill_sb method */
 461         mount_sb_info.data = data;
 462         mount_sb_info.root_khandle =
 463                 new_op->downcall.resp.fs_mount.root_khandle;
 464         mount_sb_info.fs_id = new_op->downcall.resp.fs_mount.fs_id;
 465         mount_sb_info.id = new_op->downcall.resp.fs_mount.id;
 466
 467         /*
 468          * the mount_sb_info structure looks odd, but it's used because
 469          * the private sb info isn't allocated until we call
 470          * pvfs2_fill_sb, yet we have the info we need to fill it with
 471          * here.  so we store it temporarily and pass all of the info
 472          * to fill_sb where it's properly copied out
 473          */
 474         mnt_sb_d = mount_nodev(fst,
 475                                flags,
 476                                (void *)&mount_sb_info,
 477                                pvfs2_fill_sb);
 478         if (IS_ERR(mnt_sb_d)) {
 479                 sb = ERR_CAST(mnt_sb_d);
 480                 goto free_op;
 481         }
 482
 483         sb = mnt_sb_d->d_sb;
 484
 485         /*
 486          * on successful mount, store the devname and data
 487          * used
 488          */
 489         strncpy(PVFS2_SB(sb)->devname,
 490                 devname,
 491                 PVFS_MAX_SERVER_ADDR_LEN);
 492
 493         /* mount_pending must be cleared */
 494         PVFS2_SB(sb)->mount_pending = 0;
 495
 496         /*
 497          * finally, add this sb to our list of known pvfs2
 498          * sb's
 499          */
 500         add_pvfs2_sb(sb);
 501         op_release(new_op);
 502         return mnt_sb_d;
 503
 504 free_op:
 505         gossip_err("pvfs2_mount: mount request failed with %d\n", ret);
 506         if (ret == -EINVAL) {
 507                 gossip_err("Ensure that all pvfs2-servers have the same FS configuration files\n");
 508                 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
 509         }
 510
 511         op_release(new_op);
 512
 513         gossip_debug(GOSSIP_SUPER_DEBUG,
 514                      "pvfs2_mount: returning dentry %p\n",
 515                      mnt_sb_d);
 516         return mnt_sb_d;
 517 }
 518
 519 void pvfs2_kill_sb(struct super_block *sb)
 520 {
 521         gossip_debug(GOSSIP_SUPER_DEBUG, "pvfs2_kill_sb: called\n");
 522
 523         /*
 524          * issue the unmount to userspace to tell it to remove the
 525          * dynamic mount info it has for this superblock
 526          */
 527         pvfs2_unmount_sb(sb);
 528
 529         /* remove the sb from our list of pvfs2 specific sb's */
 530         remove_pvfs2_sb(sb);
 531
 532         /* provided sb cleanup */
 533         kill_anon_super(sb);
 534
 535         /* free the pvfs2 superblock private data */
 536         kfree(PVFS2_SB(sb));
 537 }
 538
 539 int pvfs2_inode_cache_initialize(void)
 540 {
 541         pvfs2_inode_cache = kmem_cache_create("pvfs2_inode_cache",
 542                                               sizeof(struct pvfs2_inode_s),
 543                                               0,
 544                                               PVFS2_CACHE_CREATE_FLAGS,
 545                                               pvfs2_inode_cache_ctor);
 546
 547         if (!pvfs2_inode_cache) {
 548                 gossip_err("Cannot create pvfs2_inode_cache\n");
 549                 return -ENOMEM;
 550         }
 551         return 0;
 552 }
 553
 554 int pvfs2_inode_cache_finalize(void)
 555 {
 556         kmem_cache_destroy(pvfs2_inode_cache);
 557         return 0;
 558 }