1 #define MSNFS /* HACK HACK */
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
34 #endif /* CONFIG_NFSD_V3 */
37 #include <linux/nfs4_acl.h>
38 #include <linux/nfsd_idmap.h>
39 #endif /* CONFIG_NFSD_V4 */
44 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
48 * This is a cache of readahead params that help us choose the proper
49 * readahead strategy. Initially, we set all readahead parameters to 0
50 * and let the VFS handle things.
51 * If you increase the number of cached files very much, you'll need to
52 * add a hash table here.
55 struct raparms *p_next;
60 struct file_ra_state p_ra;
61 unsigned int p_hindex;
64 struct raparm_hbucket {
65 struct raparms *pb_head;
67 } ____cacheline_aligned_in_smp;
69 #define RAPARM_HASH_BITS 4
70 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
71 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
72 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
78 * or nfs_ok having possibly changed *dpp and *expp
81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
82 struct svc_export **expp)
84 struct svc_export *exp = *expp, *exp2 = NULL;
85 struct dentry *dentry = *dpp;
86 struct path path = {.mnt = mntget(exp->ex_path.mnt),
87 .dentry = dget(dentry)};
90 while (d_mountpoint(path.dentry) && follow_down(&path))
93 exp2 = rqst_exp_get_by_name(rqstp, &path);
97 * We normally allow NFS clients to continue
98 * "underneath" a mountpoint that is not exported.
99 * The exception is V4ROOT, where no traversal is ever
100 * allowed without an explicit export of the new
103 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
108 if (nfsd_v4client(rqstp) ||
109 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
110 /* successfully crossed mount point */
112 * This is subtle: path.dentry is *not* on path.mnt
113 * at this point. The only reason we are safe is that
114 * original mnt is pinned down by exp, so we should
115 * put path *before* putting exp
118 path.dentry = dentry;
128 static void follow_to_parent(struct path *path)
132 while (path->dentry == path->mnt->mnt_root && follow_up(path))
134 dp = dget_parent(path->dentry);
139 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
141 struct svc_export *exp2;
142 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
143 .dentry = dget(dparent)};
145 follow_to_parent(&path);
147 exp2 = rqst_exp_parent(rqstp, &path);
148 if (PTR_ERR(exp2) == -ENOENT) {
149 *dentryp = dget(dparent);
150 } else if (IS_ERR(exp2)) {
152 return PTR_ERR(exp2);
154 *dentryp = dget(path.dentry);
163 * For nfsd purposes, we treat V4ROOT exports as though there was an
164 * export at *every* directory.
166 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
168 if (d_mountpoint(dentry))
170 if (!(exp->ex_flags & NFSEXP_V4ROOT))
172 return dentry->d_inode != NULL;
176 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
177 const char *name, unsigned int len,
178 struct svc_export **exp_ret, struct dentry **dentry_ret)
180 struct svc_export *exp;
181 struct dentry *dparent;
182 struct dentry *dentry;
186 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
188 /* Obtain dentry and export. */
189 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
193 dparent = fhp->fh_dentry;
194 exp = fhp->fh_export;
197 /* Lookup the name, but don't follow links */
198 if (isdotent(name, len)) {
200 dentry = dget(dparent);
201 else if (dparent != exp->ex_path.dentry)
202 dentry = dget_parent(dparent);
203 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
204 dentry = dget(dparent); /* .. == . just like at / */
206 /* checking mountpoint crossing is very different when stepping up */
207 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
213 dentry = lookup_one_len(name, dparent, len);
214 host_err = PTR_ERR(dentry);
218 * check if we have crossed a mount point ...
220 if (nfsd_mountpoint(dentry, exp)) {
221 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
227 *dentry_ret = dentry;
233 return nfserrno(host_err);
237 * Look up one component of a pathname.
238 * N.B. After this call _both_ fhp and resfh need an fh_put
240 * If the lookup would cross a mountpoint, and the mounted filesystem
241 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
242 * accepted as it stands and the mounted directory is
243 * returned. Otherwise the covered directory is returned.
244 * NOTE: this mountpoint crossing is not supported properly by all
245 * clients and is explicitly disallowed for NFSv3
246 * NeilBrown <neilb@cse.unsw.edu.au>
249 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
250 unsigned int len, struct svc_fh *resfh)
252 struct svc_export *exp;
253 struct dentry *dentry;
256 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
259 err = check_nfsd_access(exp, rqstp);
263 * Note: we compose the file handle now, but as the
264 * dentry may be negative, it may need to be updated.
266 err = fh_compose(resfh, exp, dentry, fhp);
267 if (!err && !dentry->d_inode)
276 * Commit metadata changes to stable storage.
279 commit_metadata(struct svc_fh *fhp)
281 struct inode *inode = fhp->fh_dentry->d_inode;
282 const struct export_operations *export_ops = inode->i_sb->s_export_op;
284 if (!EX_ISSYNC(fhp->fh_export))
287 if (export_ops->commit_metadata)
288 return export_ops->commit_metadata(inode);
289 return sync_inode_metadata(inode, 1);
293 * Set various file attributes.
294 * N.B. After this call fhp needs an fh_put
297 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
298 int check_guard, time_t guardtime)
300 struct dentry *dentry;
302 int accmode = NFSD_MAY_SATTR;
308 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
309 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
310 if (iap->ia_valid & ATTR_SIZE)
314 err = fh_verify(rqstp, fhp, ftype, accmode);
318 dentry = fhp->fh_dentry;
319 inode = dentry->d_inode;
321 /* Ignore any mode updates on symlinks */
322 if (S_ISLNK(inode->i_mode))
323 iap->ia_valid &= ~ATTR_MODE;
329 * NFSv2 does not differentiate between "set-[ac]time-to-now"
330 * which only requires access, and "set-[ac]time-to-X" which
331 * requires ownership.
332 * So if it looks like it might be "set both to the same time which
333 * is close to now", and if inode_change_ok fails, then we
334 * convert to "set to now" instead of "set to explicit time"
336 * We only call inode_change_ok as the last test as technically
337 * it is not an interface that we should be using. It is only
338 * valid if the filesystem does not define it's own i_op->setattr.
340 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
341 #define MAX_TOUCH_TIME_ERROR (30*60)
342 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
343 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
347 * Now just make sure time is in the right ballpark.
348 * Solaris, at least, doesn't seem to care what the time
349 * request is. We require it be within 30 minutes of now.
351 time_t delta = iap->ia_atime.tv_sec - get_seconds();
354 if (delta < MAX_TOUCH_TIME_ERROR &&
355 inode_change_ok(inode, iap) != 0) {
357 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
358 * This will cause notify_change to set these times
361 iap->ia_valid &= ~BOTH_TIME_SET;
366 * The size case is special.
367 * It changes the file as well as the attributes.
369 if (iap->ia_valid & ATTR_SIZE) {
370 if (iap->ia_size < inode->i_size) {
371 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
372 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
378 * If we are changing the size of the file, then
379 * we need to break all leases.
381 host_err = break_lease(inode, O_WRONLY | O_NONBLOCK);
382 if (host_err == -EWOULDBLOCK)
383 host_err = -ETIMEDOUT;
384 if (host_err) /* ENOMEM or EWOULDBLOCK */
387 host_err = get_write_access(inode);
392 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
394 put_write_access(inode);
399 /* sanitize the mode change */
400 if (iap->ia_valid & ATTR_MODE) {
401 iap->ia_mode &= S_IALLUGO;
402 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
405 /* Revoke setuid/setgid on chown */
406 if (!S_ISDIR(inode->i_mode) &&
407 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
408 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
409 iap->ia_valid |= ATTR_KILL_PRIV;
410 if (iap->ia_valid & ATTR_MODE) {
411 /* we're setting mode too, just clear the s*id bits */
412 iap->ia_mode &= ~S_ISUID;
413 if (iap->ia_mode & S_IXGRP)
414 iap->ia_mode &= ~S_ISGID;
416 /* set ATTR_KILL_* bits and let VFS handle it */
417 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
421 /* Change the attributes. */
423 iap->ia_valid |= ATTR_CTIME;
425 err = nfserr_notsync;
426 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
428 host_err = notify_change(dentry, iap);
429 err = nfserrno(host_err);
433 put_write_access(inode);
435 commit_metadata(fhp);
440 err = nfserrno(host_err);
444 #if defined(CONFIG_NFSD_V2_ACL) || \
445 defined(CONFIG_NFSD_V3_ACL) || \
446 defined(CONFIG_NFSD_V4)
447 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
452 buflen = vfs_getxattr(dentry, key, NULL, 0);
456 *buf = kmalloc(buflen, GFP_KERNEL);
460 ret = vfs_getxattr(dentry, key, *buf, buflen);
467 #if defined(CONFIG_NFSD_V4)
469 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
476 buflen = posix_acl_xattr_size(pacl->a_count);
477 buf = kmalloc(buflen, GFP_KERNEL);
482 len = posix_acl_to_xattr(pacl, buf, buflen);
488 error = vfs_setxattr(dentry, key, buf, len, 0);
495 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
496 struct nfs4_acl *acl)
500 struct dentry *dentry;
502 struct posix_acl *pacl = NULL, *dpacl = NULL;
503 unsigned int flags = 0;
506 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
510 dentry = fhp->fh_dentry;
511 inode = dentry->d_inode;
512 if (S_ISDIR(inode->i_mode))
513 flags = NFS4_ACL_DIR;
515 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
516 if (host_error == -EINVAL) {
517 return nfserr_attrnotsupp;
518 } else if (host_error < 0)
521 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
525 if (S_ISDIR(inode->i_mode))
526 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
529 posix_acl_release(pacl);
530 posix_acl_release(dpacl);
532 if (host_error == -EOPNOTSUPP)
533 return nfserr_attrnotsupp;
535 return nfserrno(host_error);
538 static struct posix_acl *
539 _get_posix_acl(struct dentry *dentry, char *key)
542 struct posix_acl *pacl = NULL;
545 buflen = nfsd_getxattr(dentry, key, &buf);
549 return ERR_PTR(buflen);
551 pacl = posix_acl_from_xattr(buf, buflen);
557 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
559 struct inode *inode = dentry->d_inode;
561 struct posix_acl *pacl = NULL, *dpacl = NULL;
562 unsigned int flags = 0;
564 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
565 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
566 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
568 error = PTR_ERR(pacl);
573 if (S_ISDIR(inode->i_mode)) {
574 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
575 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
577 else if (IS_ERR(dpacl)) {
578 error = PTR_ERR(dpacl);
582 flags = NFS4_ACL_DIR;
585 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
587 error = PTR_ERR(*acl);
591 posix_acl_release(pacl);
592 posix_acl_release(dpacl);
596 #endif /* defined(CONFIG_NFSD_V4) */
598 #ifdef CONFIG_NFSD_V3
600 * Check server access rights to a file system object
606 static struct accessmap nfs3_regaccess[] = {
607 { NFS3_ACCESS_READ, NFSD_MAY_READ },
608 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
609 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
610 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
615 static struct accessmap nfs3_diraccess[] = {
616 { NFS3_ACCESS_READ, NFSD_MAY_READ },
617 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
618 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
619 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
620 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
625 static struct accessmap nfs3_anyaccess[] = {
626 /* Some clients - Solaris 2.6 at least, make an access call
627 * to the server to check for access for things like /dev/null
628 * (which really, the server doesn't care about). So
629 * We provide simple access checking for them, looking
630 * mainly at mode bits, and we make sure to ignore read-only
633 { NFS3_ACCESS_READ, NFSD_MAY_READ },
634 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
635 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
636 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
642 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
644 struct accessmap *map;
645 struct svc_export *export;
646 struct dentry *dentry;
647 u32 query, result = 0, sresult = 0;
650 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
654 export = fhp->fh_export;
655 dentry = fhp->fh_dentry;
657 if (S_ISREG(dentry->d_inode->i_mode))
658 map = nfs3_regaccess;
659 else if (S_ISDIR(dentry->d_inode->i_mode))
660 map = nfs3_diraccess;
662 map = nfs3_anyaccess;
666 for (; map->access; map++) {
667 if (map->access & query) {
670 sresult |= map->access;
672 err2 = nfsd_permission(rqstp, export, dentry, map->how);
675 result |= map->access;
678 /* the following error codes just mean the access was not allowed,
679 * rather than an error occurred */
683 /* simply don't "or" in the access bit. */
693 *supported = sresult;
698 #endif /* CONFIG_NFSD_V3 */
703 * Open an existing file or directory.
704 * The access argument indicates the type of open (read/write/lock)
705 * N.B. After this call fhp needs an fh_put
708 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
709 int access, struct file **filp)
711 struct dentry *dentry;
713 int flags = O_RDONLY|O_LARGEFILE;
717 validate_process_creds();
720 * If we get here, then the client has already done an "open",
721 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
722 * in case a chmod has now revoked permission.
724 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
728 dentry = fhp->fh_dentry;
729 inode = dentry->d_inode;
731 /* Disallow write access to files with the append-only bit set
732 * or any access when mandatory locking enabled
735 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
738 * We must ignore files (but only files) which might have mandatory
739 * locks on them because there is no way to know if the accesser has
742 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
749 * Check to see if there are any leases on this file.
750 * This may block while leases are broken.
752 if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
753 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
754 if (host_err == -EWOULDBLOCK)
755 host_err = -ETIMEDOUT;
756 if (host_err) /* NOMEM or WOULDBLOCK */
759 if (access & NFSD_MAY_WRITE) {
760 if (access & NFSD_MAY_READ)
761 flags = O_RDWR|O_LARGEFILE;
763 flags = O_WRONLY|O_LARGEFILE;
765 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
766 flags, current_cred());
768 host_err = PTR_ERR(*filp);
770 host_err = ima_file_check(*filp, access);
772 err = nfserrno(host_err);
774 validate_process_creds();
782 nfsd_close(struct file *filp)
788 * Obtain the readahead parameters for the file
789 * specified by (dev, ino).
792 static inline struct raparms *
793 nfsd_get_raparms(dev_t dev, ino_t ino)
795 struct raparms *ra, **rap, **frap = NULL;
798 struct raparm_hbucket *rab;
800 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
801 rab = &raparm_hash[hash];
803 spin_lock(&rab->pb_lock);
804 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
805 if (ra->p_ino == ino && ra->p_dev == dev)
808 if (ra->p_count == 0)
811 depth = nfsdstats.ra_size*11/10;
813 spin_unlock(&rab->pb_lock);
823 if (rap != &rab->pb_head) {
825 ra->p_next = rab->pb_head;
829 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
830 spin_unlock(&rab->pb_lock);
835 * Grab and keep cached pages associated with a file in the svc_rqst
836 * so that they can be passed to the network sendmsg/sendpage routines
837 * directly. They will be released after the sending has completed.
840 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
841 struct splice_desc *sd)
843 struct svc_rqst *rqstp = sd->u.data;
844 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
845 struct page *page = buf->page;
849 ret = buf->ops->confirm(pipe, buf);
855 if (rqstp->rq_res.page_len == 0) {
860 rqstp->rq_res.page_base = buf->offset;
861 rqstp->rq_res.page_len = size;
862 } else if (page != pp[-1]) {
868 rqstp->rq_res.page_len += size;
870 rqstp->rq_res.page_len += size;
875 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
876 struct splice_desc *sd)
878 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
881 static inline int svc_msnfs(struct svc_fh *ffhp)
884 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
891 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
892 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
900 inode = file->f_path.dentry->d_inode;
902 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
905 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
906 struct splice_desc sd = {
913 rqstp->rq_resused = 1;
914 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
918 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
923 nfsdstats.io_read += host_err;
927 err = nfserrno(host_err);
932 static void kill_suid(struct dentry *dentry)
935 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
937 mutex_lock(&dentry->d_inode->i_mutex);
938 notify_change(dentry, &ia);
939 mutex_unlock(&dentry->d_inode->i_mutex);
943 * Gathered writes: If another process is currently writing to the file,
944 * there's a high chance this is another nfsd (triggered by a bulk write
945 * from a client's biod). Rather than syncing the file with each write
946 * request, we sleep for 10 msec.
948 * I don't know if this roughly approximates C. Juszak's idea of
949 * gathered writes, but it's a nice and simple solution (IMHO), and it
952 * Note: we do this only in the NFSv2 case, since v3 and higher have a
953 * better tool (separate unstable writes and commits) for solving this
956 static int wait_for_concurrent_writes(struct file *file)
958 struct inode *inode = file->f_path.dentry->d_inode;
959 static ino_t last_ino;
960 static dev_t last_dev;
963 if (atomic_read(&inode->i_writecount) > 1
964 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
965 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
967 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
970 if (inode->i_state & I_DIRTY) {
971 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
972 err = vfs_fsync(file, 0);
974 last_ino = inode->i_ino;
975 last_dev = inode->i_sb->s_dev;
980 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
981 loff_t offset, struct kvec *vec, int vlen,
982 unsigned long *cnt, int *stablep)
984 struct svc_export *exp;
985 struct dentry *dentry;
990 int stable = *stablep;
996 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
997 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1001 dentry = file->f_path.dentry;
1002 inode = dentry->d_inode;
1003 exp = fhp->fh_export;
1006 * Request sync writes if
1007 * - the sync export option has been set, or
1008 * - the client requested O_SYNC behavior (NFSv3 feature).
1009 * - The file system doesn't support fsync().
1010 * When NFSv2 gathered writes have been configured for this volume,
1011 * flushing the data to disk is handled separately below.
1013 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1015 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1017 *stablep = 2; /* FILE_SYNC */
1020 if (!EX_ISSYNC(exp))
1022 if (stable && !use_wgather) {
1023 spin_lock(&file->f_lock);
1024 file->f_flags |= O_SYNC;
1025 spin_unlock(&file->f_lock);
1028 /* Write the data. */
1029 oldfs = get_fs(); set_fs(KERNEL_DS);
1030 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1035 nfsdstats.io_write += host_err;
1037 /* clear setuid/setgid flag after write */
1038 if (inode->i_mode & (S_ISUID | S_ISGID))
1041 if (stable && use_wgather)
1042 host_err = wait_for_concurrent_writes(file);
1045 dprintk("nfsd: write complete host_err=%d\n", host_err);
1049 err = nfserrno(host_err);
1055 * Read data from a file. count must contain the requested read count
1056 * on entry. On return, *count contains the number of bytes actually read.
1057 * N.B. After this call fhp needs an fh_put
1059 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1060 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1063 struct inode *inode;
1067 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1071 inode = file->f_path.dentry->d_inode;
1073 /* Get readahead parameters */
1074 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1076 if (ra && ra->p_set)
1077 file->f_ra = ra->p_ra;
1079 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1081 /* Write back readahead params */
1083 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1084 spin_lock(&rab->pb_lock);
1085 ra->p_ra = file->f_ra;
1088 spin_unlock(&rab->pb_lock);
1095 /* As above, but use the provided file descriptor. */
1097 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1098 loff_t offset, struct kvec *vec, int vlen,
1099 unsigned long *count)
1104 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1105 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1108 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1109 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1110 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1116 * Write data to a file.
1117 * The stable flag requests synchronous writes.
1118 * N.B. After this call fhp needs an fh_put
1121 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1122 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1128 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1129 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1132 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1135 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1140 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1148 #ifdef CONFIG_NFSD_V3
1150 * Commit all pending writes to stable storage.
1152 * Note: we only guarantee that data that lies within the range specified
1153 * by the 'offset' and 'count' parameters will be synced.
1155 * Unfortunately we cannot lock the file to make sure we return full WCC
1156 * data to the client, as locking happens lower down in the filesystem.
1159 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1160 loff_t offset, unsigned long count)
1163 loff_t end = LLONG_MAX;
1164 __be32 err = nfserr_inval;
1169 end = offset + (loff_t)count - 1;
1174 err = nfsd_open(rqstp, fhp, S_IFREG,
1175 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1178 if (EX_ISSYNC(fhp->fh_export)) {
1179 int err2 = vfs_fsync_range(file, offset, end, 0);
1181 if (err2 != -EINVAL)
1182 err = nfserrno(err2);
1184 err = nfserr_notsupp;
1191 #endif /* CONFIG_NFSD_V3 */
1194 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1198 * Mode has already been set earlier in create:
1200 iap->ia_valid &= ~ATTR_MODE;
1202 * Setting uid/gid works only for root. Irix appears to
1203 * send along the gid on create when it tries to implement
1204 * setgid directories via NFS:
1206 if (current_fsuid() != 0)
1207 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1209 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1213 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1214 * setting size to 0 may fail for some specific file systems by the permission
1215 * checking which requires WRITE permission but the mode is 000.
1216 * we ignore the resizing(to 0) on the just new created file, since the size is
1217 * 0 after file created.
1219 * call this only after vfs_create() is called.
1222 nfsd_check_ignore_resizing(struct iattr *iap)
1224 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1225 iap->ia_valid &= ~ATTR_SIZE;
1229 * Create a file (regular, directory, device, fifo); UNIX sockets
1230 * not yet implemented.
1231 * If the response fh has been verified, the parent directory should
1232 * already be locked. Note that the parent directory is left locked.
1234 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1237 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1238 char *fname, int flen, struct iattr *iap,
1239 int type, dev_t rdev, struct svc_fh *resfhp)
1241 struct dentry *dentry, *dchild = NULL;
1251 if (isdotent(fname, flen))
1254 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1258 dentry = fhp->fh_dentry;
1259 dirp = dentry->d_inode;
1261 err = nfserr_notdir;
1262 if (!dirp->i_op->lookup)
1265 * Check whether the response file handle has been verified yet.
1266 * If it has, the parent directory should already be locked.
1268 if (!resfhp->fh_dentry) {
1269 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1270 fh_lock_nested(fhp, I_MUTEX_PARENT);
1271 dchild = lookup_one_len(fname, dentry, flen);
1272 host_err = PTR_ERR(dchild);
1275 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1279 /* called from nfsd_proc_create */
1280 dchild = dget(resfhp->fh_dentry);
1281 if (!fhp->fh_locked) {
1282 /* not actually possible */
1284 "nfsd_create: parent %s/%s not locked!\n",
1285 dentry->d_parent->d_name.name,
1286 dentry->d_name.name);
1292 * Make sure the child dentry is still negative ...
1295 if (dchild->d_inode) {
1296 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1297 dentry->d_name.name, dchild->d_name.name);
1301 if (!(iap->ia_valid & ATTR_MODE))
1303 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1306 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1307 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1312 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1317 * Get the dir op function pointer.
1322 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1324 nfsd_check_ignore_resizing(iap);
1327 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1333 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1337 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1341 err = nfsd_create_setattr(rqstp, resfhp, iap);
1344 * nfsd_setattr already committed the child. Transactional filesystems
1345 * had a chance to commit changes for both parent and child
1346 * simultaneously making the following commit_metadata a noop.
1348 err2 = nfserrno(commit_metadata(fhp));
1351 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1353 * Update the file handle to get the new inode info.
1356 err = fh_update(resfhp);
1358 if (dchild && !IS_ERR(dchild))
1363 err = nfserrno(host_err);
1367 #ifdef CONFIG_NFSD_V3
1369 * NFSv3 version of nfsd_create
1372 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1373 char *fname, int flen, struct iattr *iap,
1374 struct svc_fh *resfhp, int createmode, u32 *verifier,
1375 int *truncp, int *created)
1377 struct dentry *dentry, *dchild = NULL;
1381 __u32 v_mtime=0, v_atime=0;
1387 if (isdotent(fname, flen))
1389 if (!(iap->ia_valid & ATTR_MODE))
1391 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1395 dentry = fhp->fh_dentry;
1396 dirp = dentry->d_inode;
1398 /* Get all the sanity checks out of the way before
1399 * we lock the parent. */
1400 err = nfserr_notdir;
1401 if (!dirp->i_op->lookup)
1403 fh_lock_nested(fhp, I_MUTEX_PARENT);
1406 * Compose the response file handle.
1408 dchild = lookup_one_len(fname, dentry, flen);
1409 host_err = PTR_ERR(dchild);
1413 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1417 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1418 /* solaris7 gets confused (bugid 4218508) if these have
1419 * the high bit set, so just clear the high bits. If this is
1420 * ever changed to use different attrs for storing the
1421 * verifier, then do_open_lookup() will also need to be fixed
1424 v_mtime = verifier[0]&0x7fffffff;
1425 v_atime = verifier[1]&0x7fffffff;
1428 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1431 if (dchild->d_inode) {
1434 switch (createmode) {
1435 case NFS3_CREATE_UNCHECKED:
1436 if (! S_ISREG(dchild->d_inode->i_mode))
1439 /* in nfsv4, we need to treat this case a little
1440 * differently. we don't want to truncate the
1441 * file now; this would be wrong if the OPEN
1442 * fails for some other reason. furthermore,
1443 * if the size is nonzero, we should ignore it
1444 * according to spec!
1446 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1449 iap->ia_valid &= ATTR_SIZE;
1453 case NFS3_CREATE_EXCLUSIVE:
1454 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1455 && dchild->d_inode->i_atime.tv_sec == v_atime
1456 && dchild->d_inode->i_size == 0 )
1459 case NFS3_CREATE_GUARDED:
1462 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1466 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1468 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1474 nfsd_check_ignore_resizing(iap);
1476 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1477 /* Cram the verifier into atime/mtime */
1478 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1479 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1480 /* XXX someone who knows this better please fix it for nsec */
1481 iap->ia_mtime.tv_sec = v_mtime;
1482 iap->ia_atime.tv_sec = v_atime;
1483 iap->ia_mtime.tv_nsec = 0;
1484 iap->ia_atime.tv_nsec = 0;
1488 err = nfsd_create_setattr(rqstp, resfhp, iap);
1491 * nfsd_setattr already committed the child (and possibly also the parent).
1494 err = nfserrno(commit_metadata(fhp));
1496 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1498 * Update the filehandle to get the new inode info.
1501 err = fh_update(resfhp);
1505 if (dchild && !IS_ERR(dchild))
1510 err = nfserrno(host_err);
1513 #endif /* CONFIG_NFSD_V3 */
1516 * Read a symlink. On entry, *lenp must contain the maximum path length that
1517 * fits into the buffer. On return, it contains the true length.
1518 * N.B. After this call fhp needs an fh_put
1521 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1523 struct dentry *dentry;
1524 struct inode *inode;
1529 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1533 dentry = fhp->fh_dentry;
1534 inode = dentry->d_inode;
1537 if (!inode->i_op->readlink)
1540 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1541 /* N.B. Why does this call need a get_fs()??
1542 * Remove the set_fs and watch the fireworks:-) --okir
1545 oldfs = get_fs(); set_fs(KERNEL_DS);
1546 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1557 err = nfserrno(host_err);
1562 * Create a symlink and look up its inode
1563 * N.B. After this call _both_ fhp and resfhp need an fh_put
1566 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1567 char *fname, int flen,
1568 char *path, int plen,
1569 struct svc_fh *resfhp,
1572 struct dentry *dentry, *dnew;
1580 if (isdotent(fname, flen))
1583 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1587 dentry = fhp->fh_dentry;
1588 dnew = lookup_one_len(fname, dentry, flen);
1589 host_err = PTR_ERR(dnew);
1593 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1597 if (unlikely(path[plen] != 0)) {
1598 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1599 if (path_alloced == NULL)
1602 strncpy(path_alloced, path, plen);
1603 path_alloced[plen] = 0;
1604 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1605 kfree(path_alloced);
1608 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1609 err = nfserrno(host_err);
1611 err = nfserrno(commit_metadata(fhp));
1614 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1616 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1618 if (err==0) err = cerr;
1623 err = nfserrno(host_err);
1629 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1632 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1633 char *name, int len, struct svc_fh *tfhp)
1635 struct dentry *ddir, *dnew, *dold;
1640 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1643 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1651 if (isdotent(name, len))
1654 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1655 ddir = ffhp->fh_dentry;
1656 dirp = ddir->d_inode;
1658 dnew = lookup_one_len(name, ddir, len);
1659 host_err = PTR_ERR(dnew);
1663 dold = tfhp->fh_dentry;
1665 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1667 err = nfserrno(host_err);
1670 host_err = vfs_link(dold, dirp, dnew);
1672 err = nfserrno(commit_metadata(ffhp));
1674 err = nfserrno(commit_metadata(tfhp));
1676 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1679 err = nfserrno(host_err);
1681 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1690 err = nfserrno(host_err);
1696 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1699 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1700 struct svc_fh *tfhp, char *tname, int tlen)
1702 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1703 struct inode *fdir, *tdir;
1707 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1710 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1714 fdentry = ffhp->fh_dentry;
1715 fdir = fdentry->d_inode;
1717 tdentry = tfhp->fh_dentry;
1718 tdir = tdentry->d_inode;
1720 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1721 if (ffhp->fh_export != tfhp->fh_export)
1725 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1728 /* cannot use fh_lock as we need deadlock protective ordering
1729 * so do it by hand */
1730 trap = lock_rename(tdentry, fdentry);
1731 ffhp->fh_locked = tfhp->fh_locked = 1;
1735 odentry = lookup_one_len(fname, fdentry, flen);
1736 host_err = PTR_ERR(odentry);
1737 if (IS_ERR(odentry))
1741 if (!odentry->d_inode)
1744 if (odentry == trap)
1747 ndentry = lookup_one_len(tname, tdentry, tlen);
1748 host_err = PTR_ERR(ndentry);
1749 if (IS_ERR(ndentry))
1751 host_err = -ENOTEMPTY;
1752 if (ndentry == trap)
1755 if (svc_msnfs(ffhp) &&
1756 ((atomic_read(&odentry->d_count) > 1)
1757 || (atomic_read(&ndentry->d_count) > 1))) {
1763 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1765 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1769 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1771 host_err = commit_metadata(tfhp);
1773 host_err = commit_metadata(ffhp);
1776 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1783 err = nfserrno(host_err);
1785 /* we cannot reply on fh_unlock on the two filehandles,
1786 * as that would do the wrong thing if the two directories
1787 * were the same, so again we do it by hand
1789 fill_post_wcc(ffhp);
1790 fill_post_wcc(tfhp);
1791 unlock_rename(tdentry, fdentry);
1792 ffhp->fh_locked = tfhp->fh_locked = 0;
1799 * Unlink a file or directory
1800 * N.B. After this call fhp needs an fh_put
1803 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1804 char *fname, int flen)
1806 struct dentry *dentry, *rdentry;
1812 if (!flen || isdotent(fname, flen))
1814 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1818 fh_lock_nested(fhp, I_MUTEX_PARENT);
1819 dentry = fhp->fh_dentry;
1820 dirp = dentry->d_inode;
1822 rdentry = lookup_one_len(fname, dentry, flen);
1823 host_err = PTR_ERR(rdentry);
1824 if (IS_ERR(rdentry))
1827 if (!rdentry->d_inode) {
1834 type = rdentry->d_inode->i_mode & S_IFMT;
1836 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1840 if (type != S_IFDIR) { /* It's UNLINK */
1842 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1843 (atomic_read(&rdentry->d_count) > 1)) {
1847 host_err = vfs_unlink(dirp, rdentry);
1848 } else { /* It's RMDIR */
1849 host_err = vfs_rmdir(dirp, rdentry);
1855 host_err = commit_metadata(fhp);
1857 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1859 err = nfserrno(host_err);
1865 * We do this buffering because we must not call back into the file
1866 * system's ->lookup() method from the filldir callback. That may well
1867 * deadlock a number of file systems.
1869 * This is based heavily on the implementation of same in XFS.
1871 struct buffered_dirent {
1875 unsigned int d_type;
1879 struct readdir_data {
1885 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1886 loff_t offset, u64 ino, unsigned int d_type)
1888 struct readdir_data *buf = __buf;
1889 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1890 unsigned int reclen;
1892 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1893 if (buf->used + reclen > PAGE_SIZE) {
1898 de->namlen = namlen;
1899 de->offset = offset;
1901 de->d_type = d_type;
1902 memcpy(de->name, name, namlen);
1903 buf->used += reclen;
1908 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1909 struct readdir_cd *cdp, loff_t *offsetp)
1911 struct readdir_data buf;
1912 struct buffered_dirent *de;
1917 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1919 return nfserrno(-ENOMEM);
1924 struct inode *dir_inode = file->f_path.dentry->d_inode;
1925 unsigned int reclen;
1927 cdp->err = nfserr_eof; /* will be cleared on successful read */
1931 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1944 * Various filldir functions may end up calling back into
1945 * lookup_one_len() and the file system's ->lookup() method.
1946 * These expect i_mutex to be held, as it would within readdir.
1948 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1952 de = (struct buffered_dirent *)buf.dirent;
1954 offset = de->offset;
1956 if (func(cdp, de->name, de->namlen, de->offset,
1957 de->ino, de->d_type))
1960 if (cdp->err != nfs_ok)
1963 reclen = ALIGN(sizeof(*de) + de->namlen,
1966 de = (struct buffered_dirent *)((char *)de + reclen);
1968 mutex_unlock(&dir_inode->i_mutex);
1969 if (size > 0) /* We bailed out early */
1972 offset = vfs_llseek(file, 0, SEEK_CUR);
1975 free_page((unsigned long)(buf.dirent));
1978 return nfserrno(host_err);
1985 * Read entries from a directory.
1986 * The NFSv3/4 verifier we ignore for now.
1989 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1990 struct readdir_cd *cdp, filldir_t func)
1994 loff_t offset = *offsetp;
1996 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2000 offset = vfs_llseek(file, offset, 0);
2002 err = nfserrno((int)offset);
2006 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2008 if (err == nfserr_eof || err == nfserr_toosmall)
2009 err = nfs_ok; /* can still be found in ->err */
2017 * Get file system stats
2018 * N.B. After this call fhp needs an fh_put
2021 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2025 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2027 struct path path = {
2028 .mnt = fhp->fh_export->ex_path.mnt,
2029 .dentry = fhp->fh_dentry,
2031 if (vfs_statfs(&path, stat))
2037 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2039 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2043 * Check for a user's access permissions to this inode.
2046 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2047 struct dentry *dentry, int acc)
2049 struct inode *inode = dentry->d_inode;
2052 if (acc == NFSD_MAY_NOP)
2055 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2057 (acc & NFSD_MAY_READ)? " read" : "",
2058 (acc & NFSD_MAY_WRITE)? " write" : "",
2059 (acc & NFSD_MAY_EXEC)? " exec" : "",
2060 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2061 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2062 (acc & NFSD_MAY_LOCK)? " lock" : "",
2063 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2065 IS_IMMUTABLE(inode)? " immut" : "",
2066 IS_APPEND(inode)? " append" : "",
2067 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2068 dprintk(" owner %d/%d user %d/%d\n",
2069 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2072 /* Normally we reject any write/sattr etc access on a read-only file
2073 * system. But if it is IRIX doing check on write-access for a
2074 * device special file, we ignore rofs.
2076 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2077 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2078 if (exp_rdonly(rqstp, exp) ||
2079 __mnt_is_readonly(exp->ex_path.mnt))
2081 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2084 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2087 if (acc & NFSD_MAY_LOCK) {
2088 /* If we cannot rely on authentication in NLM requests,
2089 * just allow locks, otherwise require read permission, or
2092 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2095 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2098 * The file owner always gets access permission for accesses that
2099 * would normally be checked at open time. This is to make
2100 * file access work even when the client has done a fchmod(fd, 0).
2102 * However, `cp foo bar' should fail nevertheless when bar is
2103 * readonly. A sensible way to do this might be to reject all
2104 * attempts to truncate a read-only file, because a creat() call
2105 * always implies file truncation.
2106 * ... but this isn't really fair. A process may reasonably call
2107 * ftruncate on an open file descriptor on a file with perm 000.
2108 * We must trust the client to do permission checking - using "ACCESS"
2111 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2112 inode->i_uid == current_fsuid())
2115 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2116 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2118 /* Allow read access to binaries even when mode 111 */
2119 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2120 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2121 err = inode_permission(inode, MAY_EXEC);
2123 return err? nfserrno(err) : 0;
2127 nfsd_racache_shutdown(void)
2129 struct raparms *raparm, *last_raparm;
2132 dprintk("nfsd: freeing readahead buffers.\n");
2134 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2135 raparm = raparm_hash[i].pb_head;
2137 last_raparm = raparm;
2138 raparm = raparm->p_next;
2141 raparm_hash[i].pb_head = NULL;
2145 * Initialize readahead param cache
2148 nfsd_racache_init(int cache_size)
2153 struct raparms **raparm = NULL;
2156 if (raparm_hash[0].pb_head)
2158 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2161 cache_size = nperbucket * RAPARM_HASH_SIZE;
2163 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2165 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2166 spin_lock_init(&raparm_hash[i].pb_lock);
2168 raparm = &raparm_hash[i].pb_head;
2169 for (j = 0; j < nperbucket; j++) {
2170 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2173 raparm = &(*raparm)->p_next;
2178 nfsdstats.ra_size = cache_size;
2182 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2183 nfsd_racache_shutdown();
2187 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2189 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2191 struct inode *inode = fhp->fh_dentry->d_inode;
2195 struct posix_acl *acl;
2197 if (!IS_POSIXACL(inode))
2198 return ERR_PTR(-EOPNOTSUPP);
2201 case ACL_TYPE_ACCESS:
2202 name = POSIX_ACL_XATTR_ACCESS;
2204 case ACL_TYPE_DEFAULT:
2205 name = POSIX_ACL_XATTR_DEFAULT;
2208 return ERR_PTR(-EOPNOTSUPP);
2211 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2213 return ERR_PTR(size);
2215 acl = posix_acl_from_xattr(value, size);
2221 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2223 struct inode *inode = fhp->fh_dentry->d_inode;
2229 if (!IS_POSIXACL(inode) ||
2230 !inode->i_op->setxattr || !inode->i_op->removexattr)
2233 case ACL_TYPE_ACCESS:
2234 name = POSIX_ACL_XATTR_ACCESS;
2236 case ACL_TYPE_DEFAULT:
2237 name = POSIX_ACL_XATTR_DEFAULT;
2243 if (acl && acl->a_count) {
2244 size = posix_acl_xattr_size(acl->a_count);
2245 value = kmalloc(size, GFP_KERNEL);
2248 error = posix_acl_to_xattr(acl, value, size);
2255 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2259 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2261 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2264 error = vfs_removexattr(fhp->fh_dentry, name);
2265 if (error == -ENODATA)
2269 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2275 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */