2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.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 */
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) /* ENOMEM or EWOULDBLOCK */
385 host_err = get_write_access(inode);
390 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
392 put_write_access(inode);
397 /* sanitize the mode change */
398 if (iap->ia_valid & ATTR_MODE) {
399 iap->ia_mode &= S_IALLUGO;
400 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
403 /* Revoke setuid/setgid on chown */
404 if (!S_ISDIR(inode->i_mode) &&
405 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
406 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
407 iap->ia_valid |= ATTR_KILL_PRIV;
408 if (iap->ia_valid & ATTR_MODE) {
409 /* we're setting mode too, just clear the s*id bits */
410 iap->ia_mode &= ~S_ISUID;
411 if (iap->ia_mode & S_IXGRP)
412 iap->ia_mode &= ~S_ISGID;
414 /* set ATTR_KILL_* bits and let VFS handle it */
415 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
419 /* Change the attributes. */
421 iap->ia_valid |= ATTR_CTIME;
423 err = nfserr_notsync;
424 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
426 host_err = notify_change(dentry, iap);
427 err = nfserrno(host_err);
431 put_write_access(inode);
433 commit_metadata(fhp);
438 err = nfserrno(host_err);
442 #if defined(CONFIG_NFSD_V2_ACL) || \
443 defined(CONFIG_NFSD_V3_ACL) || \
444 defined(CONFIG_NFSD_V4)
445 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
450 buflen = vfs_getxattr(dentry, key, NULL, 0);
454 *buf = kmalloc(buflen, GFP_KERNEL);
458 ret = vfs_getxattr(dentry, key, *buf, buflen);
465 #if defined(CONFIG_NFSD_V4)
467 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
474 buflen = posix_acl_xattr_size(pacl->a_count);
475 buf = kmalloc(buflen, GFP_KERNEL);
480 len = posix_acl_to_xattr(pacl, buf, buflen);
486 error = vfs_setxattr(dentry, key, buf, len, 0);
493 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
494 struct nfs4_acl *acl)
498 struct dentry *dentry;
500 struct posix_acl *pacl = NULL, *dpacl = NULL;
501 unsigned int flags = 0;
504 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
508 dentry = fhp->fh_dentry;
509 inode = dentry->d_inode;
510 if (S_ISDIR(inode->i_mode))
511 flags = NFS4_ACL_DIR;
513 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
514 if (host_error == -EINVAL) {
515 return nfserr_attrnotsupp;
516 } else if (host_error < 0)
519 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
523 if (S_ISDIR(inode->i_mode))
524 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
527 posix_acl_release(pacl);
528 posix_acl_release(dpacl);
530 if (host_error == -EOPNOTSUPP)
531 return nfserr_attrnotsupp;
533 return nfserrno(host_error);
536 static struct posix_acl *
537 _get_posix_acl(struct dentry *dentry, char *key)
540 struct posix_acl *pacl = NULL;
543 buflen = nfsd_getxattr(dentry, key, &buf);
547 return ERR_PTR(buflen);
549 pacl = posix_acl_from_xattr(buf, buflen);
555 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
557 struct inode *inode = dentry->d_inode;
559 struct posix_acl *pacl = NULL, *dpacl = NULL;
560 unsigned int flags = 0;
562 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
563 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
564 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
566 error = PTR_ERR(pacl);
571 if (S_ISDIR(inode->i_mode)) {
572 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
573 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
575 else if (IS_ERR(dpacl)) {
576 error = PTR_ERR(dpacl);
580 flags = NFS4_ACL_DIR;
583 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
585 error = PTR_ERR(*acl);
589 posix_acl_release(pacl);
590 posix_acl_release(dpacl);
594 #endif /* defined(CONFIG_NFSD_V4) */
596 #ifdef CONFIG_NFSD_V3
598 * Check server access rights to a file system object
604 static struct accessmap nfs3_regaccess[] = {
605 { NFS3_ACCESS_READ, NFSD_MAY_READ },
606 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
607 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
608 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
613 static struct accessmap nfs3_diraccess[] = {
614 { NFS3_ACCESS_READ, NFSD_MAY_READ },
615 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
616 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
617 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
618 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
623 static struct accessmap nfs3_anyaccess[] = {
624 /* Some clients - Solaris 2.6 at least, make an access call
625 * to the server to check for access for things like /dev/null
626 * (which really, the server doesn't care about). So
627 * We provide simple access checking for them, looking
628 * mainly at mode bits, and we make sure to ignore read-only
631 { NFS3_ACCESS_READ, NFSD_MAY_READ },
632 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
633 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
634 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
640 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
642 struct accessmap *map;
643 struct svc_export *export;
644 struct dentry *dentry;
645 u32 query, result = 0, sresult = 0;
648 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
652 export = fhp->fh_export;
653 dentry = fhp->fh_dentry;
655 if (S_ISREG(dentry->d_inode->i_mode))
656 map = nfs3_regaccess;
657 else if (S_ISDIR(dentry->d_inode->i_mode))
658 map = nfs3_diraccess;
660 map = nfs3_anyaccess;
664 for (; map->access; map++) {
665 if (map->access & query) {
668 sresult |= map->access;
670 err2 = nfsd_permission(rqstp, export, dentry, map->how);
673 result |= map->access;
676 /* the following error codes just mean the access was not allowed,
677 * rather than an error occurred */
681 /* simply don't "or" in the access bit. */
691 *supported = sresult;
696 #endif /* CONFIG_NFSD_V3 */
701 * Open an existing file or directory.
702 * The access argument indicates the type of open (read/write/lock)
703 * N.B. After this call fhp needs an fh_put
706 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
707 int access, struct file **filp)
709 struct dentry *dentry;
711 int flags = O_RDONLY|O_LARGEFILE;
715 validate_process_creds();
718 * If we get here, then the client has already done an "open",
719 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
720 * in case a chmod has now revoked permission.
722 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
726 dentry = fhp->fh_dentry;
727 inode = dentry->d_inode;
729 /* Disallow write access to files with the append-only bit set
730 * or any access when mandatory locking enabled
733 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
736 * We must ignore files (but only files) which might have mandatory
737 * locks on them because there is no way to know if the accesser has
740 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
747 * Check to see if there are any leases on this file.
748 * This may block while leases are broken.
750 if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
751 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
752 if (host_err) /* NOMEM or WOULDBLOCK */
755 if (access & NFSD_MAY_WRITE) {
756 if (access & NFSD_MAY_READ)
757 flags = O_RDWR|O_LARGEFILE;
759 flags = O_WRONLY|O_LARGEFILE;
761 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
762 flags, current_cred());
764 host_err = PTR_ERR(*filp);
766 host_err = ima_file_check(*filp, access);
768 err = nfserrno(host_err);
770 validate_process_creds();
778 nfsd_close(struct file *filp)
784 * Obtain the readahead parameters for the file
785 * specified by (dev, ino).
788 static inline struct raparms *
789 nfsd_get_raparms(dev_t dev, ino_t ino)
791 struct raparms *ra, **rap, **frap = NULL;
794 struct raparm_hbucket *rab;
796 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
797 rab = &raparm_hash[hash];
799 spin_lock(&rab->pb_lock);
800 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
801 if (ra->p_ino == ino && ra->p_dev == dev)
804 if (ra->p_count == 0)
807 depth = nfsdstats.ra_size*11/10;
809 spin_unlock(&rab->pb_lock);
819 if (rap != &rab->pb_head) {
821 ra->p_next = rab->pb_head;
825 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
826 spin_unlock(&rab->pb_lock);
831 * Grab and keep cached pages associated with a file in the svc_rqst
832 * so that they can be passed to the network sendmsg/sendpage routines
833 * directly. They will be released after the sending has completed.
836 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
837 struct splice_desc *sd)
839 struct svc_rqst *rqstp = sd->u.data;
840 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
841 struct page *page = buf->page;
845 ret = buf->ops->confirm(pipe, buf);
851 if (rqstp->rq_res.page_len == 0) {
856 rqstp->rq_res.page_base = buf->offset;
857 rqstp->rq_res.page_len = size;
858 } else if (page != pp[-1]) {
864 rqstp->rq_res.page_len += size;
866 rqstp->rq_res.page_len += size;
871 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
872 struct splice_desc *sd)
874 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
878 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
879 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
887 inode = file->f_path.dentry->d_inode;
889 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
890 struct splice_desc sd = {
897 rqstp->rq_resused = 1;
898 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
902 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
907 nfsdstats.io_read += host_err;
910 fsnotify_access(file);
912 err = nfserrno(host_err);
916 static void kill_suid(struct dentry *dentry)
919 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
921 mutex_lock(&dentry->d_inode->i_mutex);
922 notify_change(dentry, &ia);
923 mutex_unlock(&dentry->d_inode->i_mutex);
927 * Gathered writes: If another process is currently writing to the file,
928 * there's a high chance this is another nfsd (triggered by a bulk write
929 * from a client's biod). Rather than syncing the file with each write
930 * request, we sleep for 10 msec.
932 * I don't know if this roughly approximates C. Juszak's idea of
933 * gathered writes, but it's a nice and simple solution (IMHO), and it
936 * Note: we do this only in the NFSv2 case, since v3 and higher have a
937 * better tool (separate unstable writes and commits) for solving this
940 static int wait_for_concurrent_writes(struct file *file)
942 struct inode *inode = file->f_path.dentry->d_inode;
943 static ino_t last_ino;
944 static dev_t last_dev;
947 if (atomic_read(&inode->i_writecount) > 1
948 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
949 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
951 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
954 if (inode->i_state & I_DIRTY) {
955 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
956 err = vfs_fsync(file, 0);
958 last_ino = inode->i_ino;
959 last_dev = inode->i_sb->s_dev;
964 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
965 loff_t offset, struct kvec *vec, int vlen,
966 unsigned long *cnt, int *stablep)
968 struct svc_export *exp;
969 struct dentry *dentry;
974 int stable = *stablep;
977 dentry = file->f_path.dentry;
978 inode = dentry->d_inode;
979 exp = fhp->fh_export;
982 * Request sync writes if
983 * - the sync export option has been set, or
984 * - the client requested O_SYNC behavior (NFSv3 feature).
985 * - The file system doesn't support fsync().
986 * When NFSv2 gathered writes have been configured for this volume,
987 * flushing the data to disk is handled separately below.
989 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
991 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
993 *stablep = 2; /* FILE_SYNC */
998 if (stable && !use_wgather) {
999 spin_lock(&file->f_lock);
1000 file->f_flags |= O_SYNC;
1001 spin_unlock(&file->f_lock);
1004 /* Write the data. */
1005 oldfs = get_fs(); set_fs(KERNEL_DS);
1006 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1011 nfsdstats.io_write += host_err;
1012 fsnotify_modify(file);
1014 /* clear setuid/setgid flag after write */
1015 if (inode->i_mode & (S_ISUID | S_ISGID))
1018 if (stable && use_wgather)
1019 host_err = wait_for_concurrent_writes(file);
1022 dprintk("nfsd: write complete host_err=%d\n", host_err);
1026 err = nfserrno(host_err);
1031 * Read data from a file. count must contain the requested read count
1032 * on entry. On return, *count contains the number of bytes actually read.
1033 * N.B. After this call fhp needs an fh_put
1035 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1036 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1039 struct inode *inode;
1043 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1047 inode = file->f_path.dentry->d_inode;
1049 /* Get readahead parameters */
1050 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1052 if (ra && ra->p_set)
1053 file->f_ra = ra->p_ra;
1055 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1057 /* Write back readahead params */
1059 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1060 spin_lock(&rab->pb_lock);
1061 ra->p_ra = file->f_ra;
1064 spin_unlock(&rab->pb_lock);
1071 /* As above, but use the provided file descriptor. */
1073 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1074 loff_t offset, struct kvec *vec, int vlen,
1075 unsigned long *count)
1080 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1081 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1084 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1085 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1086 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1092 * Write data to a file.
1093 * The stable flag requests synchronous writes.
1094 * N.B. After this call fhp needs an fh_put
1097 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1098 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1104 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1105 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1108 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1111 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1116 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1124 #ifdef CONFIG_NFSD_V3
1126 * Commit all pending writes to stable storage.
1128 * Note: we only guarantee that data that lies within the range specified
1129 * by the 'offset' and 'count' parameters will be synced.
1131 * Unfortunately we cannot lock the file to make sure we return full WCC
1132 * data to the client, as locking happens lower down in the filesystem.
1135 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1136 loff_t offset, unsigned long count)
1139 loff_t end = LLONG_MAX;
1140 __be32 err = nfserr_inval;
1145 end = offset + (loff_t)count - 1;
1150 err = nfsd_open(rqstp, fhp, S_IFREG,
1151 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1154 if (EX_ISSYNC(fhp->fh_export)) {
1155 int err2 = vfs_fsync_range(file, offset, end, 0);
1157 if (err2 != -EINVAL)
1158 err = nfserrno(err2);
1160 err = nfserr_notsupp;
1167 #endif /* CONFIG_NFSD_V3 */
1170 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1174 * Mode has already been set earlier in create:
1176 iap->ia_valid &= ~ATTR_MODE;
1178 * Setting uid/gid works only for root. Irix appears to
1179 * send along the gid on create when it tries to implement
1180 * setgid directories via NFS:
1182 if (current_fsuid() != 0)
1183 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1185 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1189 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1190 * setting size to 0 may fail for some specific file systems by the permission
1191 * checking which requires WRITE permission but the mode is 000.
1192 * we ignore the resizing(to 0) on the just new created file, since the size is
1193 * 0 after file created.
1195 * call this only after vfs_create() is called.
1198 nfsd_check_ignore_resizing(struct iattr *iap)
1200 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1201 iap->ia_valid &= ~ATTR_SIZE;
1205 * Create a file (regular, directory, device, fifo); UNIX sockets
1206 * not yet implemented.
1207 * If the response fh has been verified, the parent directory should
1208 * already be locked. Note that the parent directory is left locked.
1210 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1213 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1214 char *fname, int flen, struct iattr *iap,
1215 int type, dev_t rdev, struct svc_fh *resfhp)
1217 struct dentry *dentry, *dchild = NULL;
1227 if (isdotent(fname, flen))
1230 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1234 dentry = fhp->fh_dentry;
1235 dirp = dentry->d_inode;
1237 err = nfserr_notdir;
1238 if (!dirp->i_op->lookup)
1241 * Check whether the response file handle has been verified yet.
1242 * If it has, the parent directory should already be locked.
1244 if (!resfhp->fh_dentry) {
1245 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1246 fh_lock_nested(fhp, I_MUTEX_PARENT);
1247 dchild = lookup_one_len(fname, dentry, flen);
1248 host_err = PTR_ERR(dchild);
1251 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1255 /* called from nfsd_proc_create */
1256 dchild = dget(resfhp->fh_dentry);
1257 if (!fhp->fh_locked) {
1258 /* not actually possible */
1260 "nfsd_create: parent %s/%s not locked!\n",
1261 dentry->d_parent->d_name.name,
1262 dentry->d_name.name);
1268 * Make sure the child dentry is still negative ...
1271 if (dchild->d_inode) {
1272 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1273 dentry->d_name.name, dchild->d_name.name);
1277 if (!(iap->ia_valid & ATTR_MODE))
1279 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1282 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1283 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1288 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1293 * Get the dir op function pointer.
1298 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1300 nfsd_check_ignore_resizing(iap);
1303 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1309 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1313 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1317 err = nfsd_create_setattr(rqstp, resfhp, iap);
1320 * nfsd_setattr already committed the child. Transactional filesystems
1321 * had a chance to commit changes for both parent and child
1322 * simultaneously making the following commit_metadata a noop.
1324 err2 = nfserrno(commit_metadata(fhp));
1327 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1329 * Update the file handle to get the new inode info.
1332 err = fh_update(resfhp);
1334 if (dchild && !IS_ERR(dchild))
1339 err = nfserrno(host_err);
1343 #ifdef CONFIG_NFSD_V3
1345 * NFSv3 version of nfsd_create
1348 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1349 char *fname, int flen, struct iattr *iap,
1350 struct svc_fh *resfhp, int createmode, u32 *verifier,
1351 int *truncp, int *created)
1353 struct dentry *dentry, *dchild = NULL;
1357 __u32 v_mtime=0, v_atime=0;
1363 if (isdotent(fname, flen))
1365 if (!(iap->ia_valid & ATTR_MODE))
1367 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1371 dentry = fhp->fh_dentry;
1372 dirp = dentry->d_inode;
1374 /* Get all the sanity checks out of the way before
1375 * we lock the parent. */
1376 err = nfserr_notdir;
1377 if (!dirp->i_op->lookup)
1379 fh_lock_nested(fhp, I_MUTEX_PARENT);
1382 * Compose the response file handle.
1384 dchild = lookup_one_len(fname, dentry, flen);
1385 host_err = PTR_ERR(dchild);
1389 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1393 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1394 /* solaris7 gets confused (bugid 4218508) if these have
1395 * the high bit set, so just clear the high bits. If this is
1396 * ever changed to use different attrs for storing the
1397 * verifier, then do_open_lookup() will also need to be fixed
1400 v_mtime = verifier[0]&0x7fffffff;
1401 v_atime = verifier[1]&0x7fffffff;
1404 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1407 if (dchild->d_inode) {
1410 switch (createmode) {
1411 case NFS3_CREATE_UNCHECKED:
1412 if (! S_ISREG(dchild->d_inode->i_mode))
1415 /* in nfsv4, we need to treat this case a little
1416 * differently. we don't want to truncate the
1417 * file now; this would be wrong if the OPEN
1418 * fails for some other reason. furthermore,
1419 * if the size is nonzero, we should ignore it
1420 * according to spec!
1422 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1425 iap->ia_valid &= ATTR_SIZE;
1429 case NFS3_CREATE_EXCLUSIVE:
1430 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1431 && dchild->d_inode->i_atime.tv_sec == v_atime
1432 && dchild->d_inode->i_size == 0 )
1435 case NFS3_CREATE_GUARDED:
1438 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1442 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1444 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1450 nfsd_check_ignore_resizing(iap);
1452 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1453 /* Cram the verifier into atime/mtime */
1454 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1455 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1456 /* XXX someone who knows this better please fix it for nsec */
1457 iap->ia_mtime.tv_sec = v_mtime;
1458 iap->ia_atime.tv_sec = v_atime;
1459 iap->ia_mtime.tv_nsec = 0;
1460 iap->ia_atime.tv_nsec = 0;
1464 err = nfsd_create_setattr(rqstp, resfhp, iap);
1467 * nfsd_setattr already committed the child (and possibly also the parent).
1470 err = nfserrno(commit_metadata(fhp));
1472 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1474 * Update the filehandle to get the new inode info.
1477 err = fh_update(resfhp);
1481 if (dchild && !IS_ERR(dchild))
1486 err = nfserrno(host_err);
1489 #endif /* CONFIG_NFSD_V3 */
1492 * Read a symlink. On entry, *lenp must contain the maximum path length that
1493 * fits into the buffer. On return, it contains the true length.
1494 * N.B. After this call fhp needs an fh_put
1497 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1499 struct dentry *dentry;
1500 struct inode *inode;
1505 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1509 dentry = fhp->fh_dentry;
1510 inode = dentry->d_inode;
1513 if (!inode->i_op->readlink)
1516 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1517 /* N.B. Why does this call need a get_fs()??
1518 * Remove the set_fs and watch the fireworks:-) --okir
1521 oldfs = get_fs(); set_fs(KERNEL_DS);
1522 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1533 err = nfserrno(host_err);
1538 * Create a symlink and look up its inode
1539 * N.B. After this call _both_ fhp and resfhp need an fh_put
1542 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1543 char *fname, int flen,
1544 char *path, int plen,
1545 struct svc_fh *resfhp,
1548 struct dentry *dentry, *dnew;
1556 if (isdotent(fname, flen))
1559 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1563 dentry = fhp->fh_dentry;
1564 dnew = lookup_one_len(fname, dentry, flen);
1565 host_err = PTR_ERR(dnew);
1569 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1573 if (unlikely(path[plen] != 0)) {
1574 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1575 if (path_alloced == NULL)
1578 strncpy(path_alloced, path, plen);
1579 path_alloced[plen] = 0;
1580 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1581 kfree(path_alloced);
1584 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1585 err = nfserrno(host_err);
1587 err = nfserrno(commit_metadata(fhp));
1590 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1592 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1594 if (err==0) err = cerr;
1599 err = nfserrno(host_err);
1605 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1608 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1609 char *name, int len, struct svc_fh *tfhp)
1611 struct dentry *ddir, *dnew, *dold;
1616 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1619 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1627 if (isdotent(name, len))
1630 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1631 ddir = ffhp->fh_dentry;
1632 dirp = ddir->d_inode;
1634 dnew = lookup_one_len(name, ddir, len);
1635 host_err = PTR_ERR(dnew);
1639 dold = tfhp->fh_dentry;
1641 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1643 err = nfserrno(host_err);
1646 host_err = vfs_link(dold, dirp, dnew);
1648 err = nfserrno(commit_metadata(ffhp));
1650 err = nfserrno(commit_metadata(tfhp));
1652 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1655 err = nfserrno(host_err);
1657 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1666 err = nfserrno(host_err);
1672 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1675 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1676 struct svc_fh *tfhp, char *tname, int tlen)
1678 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1679 struct inode *fdir, *tdir;
1683 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1686 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1690 fdentry = ffhp->fh_dentry;
1691 fdir = fdentry->d_inode;
1693 tdentry = tfhp->fh_dentry;
1694 tdir = tdentry->d_inode;
1696 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1697 if (ffhp->fh_export != tfhp->fh_export)
1701 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1704 /* cannot use fh_lock as we need deadlock protective ordering
1705 * so do it by hand */
1706 trap = lock_rename(tdentry, fdentry);
1707 ffhp->fh_locked = tfhp->fh_locked = 1;
1711 odentry = lookup_one_len(fname, fdentry, flen);
1712 host_err = PTR_ERR(odentry);
1713 if (IS_ERR(odentry))
1717 if (!odentry->d_inode)
1720 if (odentry == trap)
1723 ndentry = lookup_one_len(tname, tdentry, tlen);
1724 host_err = PTR_ERR(ndentry);
1725 if (IS_ERR(ndentry))
1727 host_err = -ENOTEMPTY;
1728 if (ndentry == trap)
1732 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1734 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1738 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1740 host_err = commit_metadata(tfhp);
1742 host_err = commit_metadata(ffhp);
1745 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1752 err = nfserrno(host_err);
1754 /* we cannot reply on fh_unlock on the two filehandles,
1755 * as that would do the wrong thing if the two directories
1756 * were the same, so again we do it by hand
1758 fill_post_wcc(ffhp);
1759 fill_post_wcc(tfhp);
1760 unlock_rename(tdentry, fdentry);
1761 ffhp->fh_locked = tfhp->fh_locked = 0;
1768 * Unlink a file or directory
1769 * N.B. After this call fhp needs an fh_put
1772 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1773 char *fname, int flen)
1775 struct dentry *dentry, *rdentry;
1781 if (!flen || isdotent(fname, flen))
1783 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1787 fh_lock_nested(fhp, I_MUTEX_PARENT);
1788 dentry = fhp->fh_dentry;
1789 dirp = dentry->d_inode;
1791 rdentry = lookup_one_len(fname, dentry, flen);
1792 host_err = PTR_ERR(rdentry);
1793 if (IS_ERR(rdentry))
1796 if (!rdentry->d_inode) {
1803 type = rdentry->d_inode->i_mode & S_IFMT;
1805 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1809 if (type != S_IFDIR)
1810 host_err = vfs_unlink(dirp, rdentry);
1812 host_err = vfs_rmdir(dirp, rdentry);
1817 host_err = commit_metadata(fhp);
1819 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1821 err = nfserrno(host_err);
1827 * We do this buffering because we must not call back into the file
1828 * system's ->lookup() method from the filldir callback. That may well
1829 * deadlock a number of file systems.
1831 * This is based heavily on the implementation of same in XFS.
1833 struct buffered_dirent {
1837 unsigned int d_type;
1841 struct readdir_data {
1847 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1848 loff_t offset, u64 ino, unsigned int d_type)
1850 struct readdir_data *buf = __buf;
1851 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1852 unsigned int reclen;
1854 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1855 if (buf->used + reclen > PAGE_SIZE) {
1860 de->namlen = namlen;
1861 de->offset = offset;
1863 de->d_type = d_type;
1864 memcpy(de->name, name, namlen);
1865 buf->used += reclen;
1870 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1871 struct readdir_cd *cdp, loff_t *offsetp)
1873 struct readdir_data buf;
1874 struct buffered_dirent *de;
1879 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1881 return nfserrno(-ENOMEM);
1886 struct inode *dir_inode = file->f_path.dentry->d_inode;
1887 unsigned int reclen;
1889 cdp->err = nfserr_eof; /* will be cleared on successful read */
1893 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1906 * Various filldir functions may end up calling back into
1907 * lookup_one_len() and the file system's ->lookup() method.
1908 * These expect i_mutex to be held, as it would within readdir.
1910 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1914 de = (struct buffered_dirent *)buf.dirent;
1916 offset = de->offset;
1918 if (func(cdp, de->name, de->namlen, de->offset,
1919 de->ino, de->d_type))
1922 if (cdp->err != nfs_ok)
1925 reclen = ALIGN(sizeof(*de) + de->namlen,
1928 de = (struct buffered_dirent *)((char *)de + reclen);
1930 mutex_unlock(&dir_inode->i_mutex);
1931 if (size > 0) /* We bailed out early */
1934 offset = vfs_llseek(file, 0, SEEK_CUR);
1937 free_page((unsigned long)(buf.dirent));
1940 return nfserrno(host_err);
1947 * Read entries from a directory.
1948 * The NFSv3/4 verifier we ignore for now.
1951 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1952 struct readdir_cd *cdp, filldir_t func)
1956 loff_t offset = *offsetp;
1958 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1962 offset = vfs_llseek(file, offset, 0);
1964 err = nfserrno((int)offset);
1968 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1970 if (err == nfserr_eof || err == nfserr_toosmall)
1971 err = nfs_ok; /* can still be found in ->err */
1979 * Get file system stats
1980 * N.B. After this call fhp needs an fh_put
1983 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1987 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1989 struct path path = {
1990 .mnt = fhp->fh_export->ex_path.mnt,
1991 .dentry = fhp->fh_dentry,
1993 if (vfs_statfs(&path, stat))
1999 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2001 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2005 * Check for a user's access permissions to this inode.
2008 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2009 struct dentry *dentry, int acc)
2011 struct inode *inode = dentry->d_inode;
2014 if (acc == NFSD_MAY_NOP)
2017 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2019 (acc & NFSD_MAY_READ)? " read" : "",
2020 (acc & NFSD_MAY_WRITE)? " write" : "",
2021 (acc & NFSD_MAY_EXEC)? " exec" : "",
2022 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2023 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2024 (acc & NFSD_MAY_LOCK)? " lock" : "",
2025 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2027 IS_IMMUTABLE(inode)? " immut" : "",
2028 IS_APPEND(inode)? " append" : "",
2029 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2030 dprintk(" owner %d/%d user %d/%d\n",
2031 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2034 /* Normally we reject any write/sattr etc access on a read-only file
2035 * system. But if it is IRIX doing check on write-access for a
2036 * device special file, we ignore rofs.
2038 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2039 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2040 if (exp_rdonly(rqstp, exp) ||
2041 __mnt_is_readonly(exp->ex_path.mnt))
2043 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2046 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2049 if (acc & NFSD_MAY_LOCK) {
2050 /* If we cannot rely on authentication in NLM requests,
2051 * just allow locks, otherwise require read permission, or
2054 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2057 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2060 * The file owner always gets access permission for accesses that
2061 * would normally be checked at open time. This is to make
2062 * file access work even when the client has done a fchmod(fd, 0).
2064 * However, `cp foo bar' should fail nevertheless when bar is
2065 * readonly. A sensible way to do this might be to reject all
2066 * attempts to truncate a read-only file, because a creat() call
2067 * always implies file truncation.
2068 * ... but this isn't really fair. A process may reasonably call
2069 * ftruncate on an open file descriptor on a file with perm 000.
2070 * We must trust the client to do permission checking - using "ACCESS"
2073 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2074 inode->i_uid == current_fsuid())
2077 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2078 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2080 /* Allow read access to binaries even when mode 111 */
2081 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2082 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2083 err = inode_permission(inode, MAY_EXEC);
2085 return err? nfserrno(err) : 0;
2089 nfsd_racache_shutdown(void)
2091 struct raparms *raparm, *last_raparm;
2094 dprintk("nfsd: freeing readahead buffers.\n");
2096 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2097 raparm = raparm_hash[i].pb_head;
2099 last_raparm = raparm;
2100 raparm = raparm->p_next;
2103 raparm_hash[i].pb_head = NULL;
2107 * Initialize readahead param cache
2110 nfsd_racache_init(int cache_size)
2115 struct raparms **raparm = NULL;
2118 if (raparm_hash[0].pb_head)
2120 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2123 cache_size = nperbucket * RAPARM_HASH_SIZE;
2125 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2127 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2128 spin_lock_init(&raparm_hash[i].pb_lock);
2130 raparm = &raparm_hash[i].pb_head;
2131 for (j = 0; j < nperbucket; j++) {
2132 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2135 raparm = &(*raparm)->p_next;
2140 nfsdstats.ra_size = cache_size;
2144 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2145 nfsd_racache_shutdown();
2149 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2151 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2153 struct inode *inode = fhp->fh_dentry->d_inode;
2157 struct posix_acl *acl;
2159 if (!IS_POSIXACL(inode))
2160 return ERR_PTR(-EOPNOTSUPP);
2163 case ACL_TYPE_ACCESS:
2164 name = POSIX_ACL_XATTR_ACCESS;
2166 case ACL_TYPE_DEFAULT:
2167 name = POSIX_ACL_XATTR_DEFAULT;
2170 return ERR_PTR(-EOPNOTSUPP);
2173 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2175 return ERR_PTR(size);
2177 acl = posix_acl_from_xattr(value, size);
2183 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2185 struct inode *inode = fhp->fh_dentry->d_inode;
2191 if (!IS_POSIXACL(inode) ||
2192 !inode->i_op->setxattr || !inode->i_op->removexattr)
2195 case ACL_TYPE_ACCESS:
2196 name = POSIX_ACL_XATTR_ACCESS;
2198 case ACL_TYPE_DEFAULT:
2199 name = POSIX_ACL_XATTR_DEFAULT;
2205 if (acl && acl->a_count) {
2206 size = posix_acl_xattr_size(acl->a_count);
2207 value = kmalloc(size, GFP_KERNEL);
2210 error = posix_acl_to_xattr(acl, value, size);
2217 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2221 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2223 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2226 error = vfs_removexattr(fhp->fh_dentry, name);
2227 if (error == -ENODATA)
2231 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2237 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */