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.
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/falloc.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <linux/slab.h>
29 #include <asm/uaccess.h>
30 #include <linux/exportfs.h>
31 #include <linux/writeback.h>
32 #include <linux/security.h>
36 #endif /* CONFIG_NFSD_V3 */
41 #endif /* CONFIG_NFSD_V4 */
46 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
50 * This is a cache of readahead params that help us choose the proper
51 * readahead strategy. Initially, we set all readahead parameters to 0
52 * and let the VFS handle things.
53 * If you increase the number of cached files very much, you'll need to
54 * add a hash table here.
57 struct raparms *p_next;
62 struct file_ra_state p_ra;
63 unsigned int p_hindex;
66 struct raparm_hbucket {
67 struct raparms *pb_head;
69 } ____cacheline_aligned_in_smp;
71 #define RAPARM_HASH_BITS 4
72 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
73 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
74 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
77 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
79 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
80 * or nfs_ok having possibly changed *dpp and *expp
83 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
84 struct svc_export **expp)
86 struct svc_export *exp = *expp, *exp2 = NULL;
87 struct dentry *dentry = *dpp;
88 struct path path = {.mnt = mntget(exp->ex_path.mnt),
89 .dentry = dget(dentry)};
92 err = follow_down(&path);
96 exp2 = rqst_exp_get_by_name(rqstp, &path);
100 * We normally allow NFS clients to continue
101 * "underneath" a mountpoint that is not exported.
102 * The exception is V4ROOT, where no traversal is ever
103 * allowed without an explicit export of the new
106 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
111 if (nfsd_v4client(rqstp) ||
112 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
113 /* successfully crossed mount point */
115 * This is subtle: path.dentry is *not* on path.mnt
116 * at this point. The only reason we are safe is that
117 * original mnt is pinned down by exp, so we should
118 * put path *before* putting exp
121 path.dentry = dentry;
131 static void follow_to_parent(struct path *path)
135 while (path->dentry == path->mnt->mnt_root && follow_up(path))
137 dp = dget_parent(path->dentry);
142 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
144 struct svc_export *exp2;
145 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
146 .dentry = dget(dparent)};
148 follow_to_parent(&path);
150 exp2 = rqst_exp_parent(rqstp, &path);
151 if (PTR_ERR(exp2) == -ENOENT) {
152 *dentryp = dget(dparent);
153 } else if (IS_ERR(exp2)) {
155 return PTR_ERR(exp2);
157 *dentryp = dget(path.dentry);
166 * For nfsd purposes, we treat V4ROOT exports as though there was an
167 * export at *every* directory.
169 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
171 if (d_mountpoint(dentry))
173 if (nfsd4_is_junction(dentry))
175 if (!(exp->ex_flags & NFSEXP_V4ROOT))
177 return dentry->d_inode != NULL;
181 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
182 const char *name, unsigned int len,
183 struct svc_export **exp_ret, struct dentry **dentry_ret)
185 struct svc_export *exp;
186 struct dentry *dparent;
187 struct dentry *dentry;
190 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
192 dparent = fhp->fh_dentry;
193 exp = exp_get(fhp->fh_export);
195 /* Lookup the name, but don't follow links */
196 if (isdotent(name, len)) {
198 dentry = dget(dparent);
199 else if (dparent != exp->ex_path.dentry)
200 dentry = dget_parent(dparent);
201 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
202 dentry = dget(dparent); /* .. == . just like at / */
204 /* checking mountpoint crossing is very different when stepping up */
205 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
211 * In the nfsd4_open() case, this may be held across
212 * subsequent open and delegation acquisition which may
213 * need to take the child's i_mutex:
215 fh_lock_nested(fhp, I_MUTEX_PARENT);
216 dentry = lookup_one_len(name, dparent, len);
217 host_err = PTR_ERR(dentry);
221 * check if we have crossed a mount point ...
223 if (nfsd_mountpoint(dentry, exp)) {
224 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
230 *dentry_ret = dentry;
236 return nfserrno(host_err);
240 * Look up one component of a pathname.
241 * N.B. After this call _both_ fhp and resfh need an fh_put
243 * If the lookup would cross a mountpoint, and the mounted filesystem
244 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
245 * accepted as it stands and the mounted directory is
246 * returned. Otherwise the covered directory is returned.
247 * NOTE: this mountpoint crossing is not supported properly by all
248 * clients and is explicitly disallowed for NFSv3
252 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
253 unsigned int len, struct svc_fh *resfh)
255 struct svc_export *exp;
256 struct dentry *dentry;
259 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
262 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
265 err = check_nfsd_access(exp, rqstp);
269 * Note: we compose the file handle now, but as the
270 * dentry may be negative, it may need to be updated.
272 err = fh_compose(resfh, exp, dentry, fhp);
273 if (!err && !dentry->d_inode)
282 * Commit metadata changes to stable storage.
285 commit_metadata(struct svc_fh *fhp)
287 struct inode *inode = fhp->fh_dentry->d_inode;
288 const struct export_operations *export_ops = inode->i_sb->s_export_op;
290 if (!EX_ISSYNC(fhp->fh_export))
293 if (export_ops->commit_metadata)
294 return export_ops->commit_metadata(inode);
295 return sync_inode_metadata(inode, 1);
299 * Go over the attributes and take care of the small differences between
300 * NFS semantics and what Linux expects.
303 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
306 * NFSv2 does not differentiate between "set-[ac]time-to-now"
307 * which only requires access, and "set-[ac]time-to-X" which
308 * requires ownership.
309 * So if it looks like it might be "set both to the same time which
310 * is close to now", and if inode_change_ok fails, then we
311 * convert to "set to now" instead of "set to explicit time"
313 * We only call inode_change_ok as the last test as technically
314 * it is not an interface that we should be using.
316 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
317 #define MAX_TOUCH_TIME_ERROR (30*60)
318 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
319 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
323 * Now just make sure time is in the right ballpark.
324 * Solaris, at least, doesn't seem to care what the time
325 * request is. We require it be within 30 minutes of now.
327 time_t delta = iap->ia_atime.tv_sec - get_seconds();
330 if (delta < MAX_TOUCH_TIME_ERROR &&
331 inode_change_ok(inode, iap) != 0) {
333 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
334 * This will cause notify_change to set these times
337 iap->ia_valid &= ~BOTH_TIME_SET;
341 /* sanitize the mode change */
342 if (iap->ia_valid & ATTR_MODE) {
343 iap->ia_mode &= S_IALLUGO;
344 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
347 /* Revoke setuid/setgid on chown */
348 if (!S_ISDIR(inode->i_mode) &&
349 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) {
350 iap->ia_valid |= ATTR_KILL_PRIV;
351 if (iap->ia_valid & ATTR_MODE) {
352 /* we're setting mode too, just clear the s*id bits */
353 iap->ia_mode &= ~S_ISUID;
354 if (iap->ia_mode & S_IXGRP)
355 iap->ia_mode &= ~S_ISGID;
357 /* set ATTR_KILL_* bits and let VFS handle it */
358 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
364 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
367 struct inode *inode = fhp->fh_dentry->d_inode;
370 if (iap->ia_size < inode->i_size) {
373 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
374 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
379 host_err = get_write_access(inode);
383 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
385 goto out_put_write_access;
388 out_put_write_access:
389 put_write_access(inode);
391 return nfserrno(host_err);
395 * Set various file attributes. After this call fhp needs an fh_put.
398 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
399 int check_guard, time_t guardtime)
401 struct dentry *dentry;
403 int accmode = NFSD_MAY_SATTR;
407 bool get_write_count;
410 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
411 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
412 if (iap->ia_valid & ATTR_SIZE)
415 /* Callers that do fh_verify should do the fh_want_write: */
416 get_write_count = !fhp->fh_dentry;
419 err = fh_verify(rqstp, fhp, ftype, accmode);
422 if (get_write_count) {
423 host_err = fh_want_write(fhp);
425 return nfserrno(host_err);
428 dentry = fhp->fh_dentry;
429 inode = dentry->d_inode;
431 /* Ignore any mode updates on symlinks */
432 if (S_ISLNK(inode->i_mode))
433 iap->ia_valid &= ~ATTR_MODE;
438 nfsd_sanitize_attrs(inode, iap);
441 * The size case is special, it changes the file in addition to the
444 if (iap->ia_valid & ATTR_SIZE) {
445 err = nfsd_get_write_access(rqstp, fhp, iap);
451 * RFC5661, Section 18.30.4:
452 * Changing the size of a file with SETATTR indirectly
453 * changes the time_modify and change attributes.
455 * (and similar for the older RFCs)
457 if (iap->ia_size != i_size_read(inode))
458 iap->ia_valid |= ATTR_MTIME;
461 iap->ia_valid |= ATTR_CTIME;
463 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
464 err = nfserr_notsync;
465 goto out_put_write_access;
469 host_err = notify_change(dentry, iap, NULL);
471 err = nfserrno(host_err);
473 out_put_write_access:
475 put_write_access(inode);
477 err = nfserrno(commit_metadata(fhp));
482 #if defined(CONFIG_NFSD_V4)
484 * NFS junction information is stored in an extended attribute.
486 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
489 * nfsd4_is_junction - Test if an object could be an NFS junction
491 * @dentry: object to test
493 * Returns 1 if "dentry" appears to contain NFS junction information.
494 * Otherwise 0 is returned.
496 int nfsd4_is_junction(struct dentry *dentry)
498 struct inode *inode = dentry->d_inode;
502 if (inode->i_mode & S_IXUGO)
504 if (!(inode->i_mode & S_ISVTX))
506 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
510 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
511 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
512 struct xdr_netobj *label)
516 struct dentry *dentry;
518 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
522 dentry = fhp->fh_dentry;
524 mutex_lock(&dentry->d_inode->i_mutex);
525 host_error = security_inode_setsecctx(dentry, label->data, label->len);
526 mutex_unlock(&dentry->d_inode->i_mutex);
527 return nfserrno(host_error);
530 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
531 struct xdr_netobj *label)
533 return nfserr_notsupp;
537 __be32 nfsd4_vfs_fallocate(struct svc_rqst *rqstp, struct svc_fh *fhp,
538 struct file *file, loff_t offset, loff_t len,
544 if (!S_ISREG(file_inode(file)->i_mode))
547 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, NFSD_MAY_WRITE);
551 error = vfs_fallocate(file, flags, offset, len);
553 error = commit_metadata(fhp);
555 return nfserrno(error);
557 #endif /* defined(CONFIG_NFSD_V4) */
559 #ifdef CONFIG_NFSD_V3
561 * Check server access rights to a file system object
567 static struct accessmap nfs3_regaccess[] = {
568 { NFS3_ACCESS_READ, NFSD_MAY_READ },
569 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
570 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
571 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
576 static struct accessmap nfs3_diraccess[] = {
577 { NFS3_ACCESS_READ, NFSD_MAY_READ },
578 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
579 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
580 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
581 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
586 static struct accessmap nfs3_anyaccess[] = {
587 /* Some clients - Solaris 2.6 at least, make an access call
588 * to the server to check for access for things like /dev/null
589 * (which really, the server doesn't care about). So
590 * We provide simple access checking for them, looking
591 * mainly at mode bits, and we make sure to ignore read-only
594 { NFS3_ACCESS_READ, NFSD_MAY_READ },
595 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
596 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
597 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
603 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
605 struct accessmap *map;
606 struct svc_export *export;
607 struct dentry *dentry;
608 u32 query, result = 0, sresult = 0;
611 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
615 export = fhp->fh_export;
616 dentry = fhp->fh_dentry;
618 if (d_is_reg(dentry))
619 map = nfs3_regaccess;
620 else if (d_is_dir(dentry))
621 map = nfs3_diraccess;
623 map = nfs3_anyaccess;
627 for (; map->access; map++) {
628 if (map->access & query) {
631 sresult |= map->access;
633 err2 = nfsd_permission(rqstp, export, dentry, map->how);
636 result |= map->access;
639 /* the following error codes just mean the access was not allowed,
640 * rather than an error occurred */
644 /* simply don't "or" in the access bit. */
654 *supported = sresult;
659 #endif /* CONFIG_NFSD_V3 */
661 static int nfsd_open_break_lease(struct inode *inode, int access)
665 if (access & NFSD_MAY_NOT_BREAK_LEASE)
667 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
668 return break_lease(inode, mode | O_NONBLOCK);
672 * Open an existing file or directory.
673 * The may_flags argument indicates the type of open (read/write/lock)
674 * and additional flags.
675 * N.B. After this call fhp needs an fh_put
678 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
679 int may_flags, struct file **filp)
684 int flags = O_RDONLY|O_LARGEFILE;
688 validate_process_creds();
691 * If we get here, then the client has already done an "open",
692 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
693 * in case a chmod has now revoked permission.
695 * Arguably we should also allow the owner override for
696 * directories, but we never have and it doesn't seem to have
697 * caused anyone a problem. If we were to change this, note
698 * also that our filldir callbacks would need a variant of
699 * lookup_one_len that doesn't check permissions.
702 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
703 err = fh_verify(rqstp, fhp, type, may_flags);
707 path.mnt = fhp->fh_export->ex_path.mnt;
708 path.dentry = fhp->fh_dentry;
709 inode = path.dentry->d_inode;
711 /* Disallow write access to files with the append-only bit set
712 * or any access when mandatory locking enabled
715 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
718 * We must ignore files (but only files) which might have mandatory
719 * locks on them because there is no way to know if the accesser has
722 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
728 host_err = nfsd_open_break_lease(inode, may_flags);
729 if (host_err) /* NOMEM or WOULDBLOCK */
732 if (may_flags & NFSD_MAY_WRITE) {
733 if (may_flags & NFSD_MAY_READ)
734 flags = O_RDWR|O_LARGEFILE;
736 flags = O_WRONLY|O_LARGEFILE;
739 file = dentry_open(&path, flags, current_cred());
741 host_err = PTR_ERR(file);
745 host_err = ima_file_check(file, may_flags, 0);
751 if (may_flags & NFSD_MAY_64BIT_COOKIE)
752 file->f_mode |= FMODE_64BITHASH;
754 file->f_mode |= FMODE_32BITHASH;
758 err = nfserrno(host_err);
760 validate_process_creds();
768 nfsd_close(struct file *filp)
774 * Obtain the readahead parameters for the file
775 * specified by (dev, ino).
778 static inline struct raparms *
779 nfsd_get_raparms(dev_t dev, ino_t ino)
781 struct raparms *ra, **rap, **frap = NULL;
784 struct raparm_hbucket *rab;
786 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
787 rab = &raparm_hash[hash];
789 spin_lock(&rab->pb_lock);
790 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
791 if (ra->p_ino == ino && ra->p_dev == dev)
794 if (ra->p_count == 0)
797 depth = nfsdstats.ra_size;
799 spin_unlock(&rab->pb_lock);
809 if (rap != &rab->pb_head) {
811 ra->p_next = rab->pb_head;
815 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
816 spin_unlock(&rab->pb_lock);
821 * Grab and keep cached pages associated with a file in the svc_rqst
822 * so that they can be passed to the network sendmsg/sendpage routines
823 * directly. They will be released after the sending has completed.
826 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
827 struct splice_desc *sd)
829 struct svc_rqst *rqstp = sd->u.data;
830 struct page **pp = rqstp->rq_next_page;
831 struct page *page = buf->page;
836 if (rqstp->rq_res.page_len == 0) {
838 put_page(*rqstp->rq_next_page);
839 *(rqstp->rq_next_page++) = page;
840 rqstp->rq_res.page_base = buf->offset;
841 rqstp->rq_res.page_len = size;
842 } else if (page != pp[-1]) {
844 if (*rqstp->rq_next_page)
845 put_page(*rqstp->rq_next_page);
846 *(rqstp->rq_next_page++) = page;
847 rqstp->rq_res.page_len += size;
849 rqstp->rq_res.page_len += size;
854 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
855 struct splice_desc *sd)
857 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
861 nfsd_finish_read(struct file *file, unsigned long *count, int host_err)
864 nfsdstats.io_read += host_err;
866 fsnotify_access(file);
869 return nfserrno(host_err);
872 __be32 nfsd_splice_read(struct svc_rqst *rqstp,
873 struct file *file, loff_t offset, unsigned long *count)
875 struct splice_desc sd = {
883 rqstp->rq_next_page = rqstp->rq_respages + 1;
884 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
885 return nfsd_finish_read(file, count, host_err);
888 __be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen,
889 unsigned long *count)
896 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
898 return nfsd_finish_read(file, count, host_err);
902 nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file,
903 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
905 if (file->f_op->splice_read && test_bit(RQ_SPLICE_OK, &rqstp->rq_flags))
906 return nfsd_splice_read(rqstp, file, offset, count);
908 return nfsd_readv(file, offset, vec, vlen, count);
912 * Gathered writes: If another process is currently writing to the file,
913 * there's a high chance this is another nfsd (triggered by a bulk write
914 * from a client's biod). Rather than syncing the file with each write
915 * request, we sleep for 10 msec.
917 * I don't know if this roughly approximates C. Juszak's idea of
918 * gathered writes, but it's a nice and simple solution (IMHO), and it
921 * Note: we do this only in the NFSv2 case, since v3 and higher have a
922 * better tool (separate unstable writes and commits) for solving this
925 static int wait_for_concurrent_writes(struct file *file)
927 struct inode *inode = file_inode(file);
928 static ino_t last_ino;
929 static dev_t last_dev;
932 if (atomic_read(&inode->i_writecount) > 1
933 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
934 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
936 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
939 if (inode->i_state & I_DIRTY) {
940 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
941 err = vfs_fsync(file, 0);
943 last_ino = inode->i_ino;
944 last_dev = inode->i_sb->s_dev;
949 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
950 loff_t offset, struct kvec *vec, int vlen,
951 unsigned long *cnt, int *stablep)
953 struct svc_export *exp;
958 int stable = *stablep;
961 loff_t end = LLONG_MAX;
962 unsigned int pflags = current->flags;
964 if (test_bit(RQ_LOCAL, &rqstp->rq_flags))
966 * We want less throttling in balance_dirty_pages()
967 * and shrink_inactive_list() so that nfs to
968 * localhost doesn't cause nfsd to lock up due to all
969 * the client's dirty pages or its congested queue.
971 current->flags |= PF_LESS_THROTTLE;
973 inode = file_inode(file);
974 exp = fhp->fh_export;
976 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
981 /* Write the data. */
982 oldfs = get_fs(); set_fs(KERNEL_DS);
983 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
988 nfsdstats.io_write += host_err;
989 fsnotify_modify(file);
993 host_err = wait_for_concurrent_writes(file);
996 end = offset + *cnt - 1;
997 host_err = vfs_fsync_range(file, offset, end, 0);
1002 dprintk("nfsd: write complete host_err=%d\n", host_err);
1006 err = nfserrno(host_err);
1007 if (test_bit(RQ_LOCAL, &rqstp->rq_flags))
1008 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE);
1012 __be32 nfsd_get_tmp_read_open(struct svc_rqst *rqstp, struct svc_fh *fhp,
1013 struct file **file, struct raparms **ra)
1015 struct inode *inode;
1018 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, file);
1022 inode = file_inode(*file);
1024 /* Get readahead parameters */
1025 *ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1027 if (*ra && (*ra)->p_set)
1028 (*file)->f_ra = (*ra)->p_ra;
1032 void nfsd_put_tmp_read_open(struct file *file, struct raparms *ra)
1034 /* Write back readahead params */
1036 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1037 spin_lock(&rab->pb_lock);
1038 ra->p_ra = file->f_ra;
1041 spin_unlock(&rab->pb_lock);
1047 * Read data from a file. count must contain the requested read count
1048 * on entry. On return, *count contains the number of bytes actually read.
1049 * N.B. After this call fhp needs an fh_put
1051 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1052 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1058 err = nfsd_get_tmp_read_open(rqstp, fhp, &file, &ra);
1062 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count);
1064 nfsd_put_tmp_read_open(file, ra);
1070 * Write data to a file.
1071 * The stable flag requests synchronous writes.
1072 * N.B. After this call fhp needs an fh_put
1075 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1076 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1082 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1083 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1086 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1089 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1094 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1102 #ifdef CONFIG_NFSD_V3
1104 * Commit all pending writes to stable storage.
1106 * Note: we only guarantee that data that lies within the range specified
1107 * by the 'offset' and 'count' parameters will be synced.
1109 * Unfortunately we cannot lock the file to make sure we return full WCC
1110 * data to the client, as locking happens lower down in the filesystem.
1113 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1114 loff_t offset, unsigned long count)
1117 loff_t end = LLONG_MAX;
1118 __be32 err = nfserr_inval;
1123 end = offset + (loff_t)count - 1;
1128 err = nfsd_open(rqstp, fhp, S_IFREG,
1129 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1132 if (EX_ISSYNC(fhp->fh_export)) {
1133 int err2 = vfs_fsync_range(file, offset, end, 0);
1135 if (err2 != -EINVAL)
1136 err = nfserrno(err2);
1138 err = nfserr_notsupp;
1145 #endif /* CONFIG_NFSD_V3 */
1148 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1152 * Mode has already been set earlier in create:
1154 iap->ia_valid &= ~ATTR_MODE;
1156 * Setting uid/gid works only for root. Irix appears to
1157 * send along the gid on create when it tries to implement
1158 * setgid directories via NFS:
1160 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1161 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1163 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1164 /* Callers expect file metadata to be committed here */
1165 return nfserrno(commit_metadata(resfhp));
1168 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1169 * setting size to 0 may fail for some specific file systems by the permission
1170 * checking which requires WRITE permission but the mode is 000.
1171 * we ignore the resizing(to 0) on the just new created file, since the size is
1172 * 0 after file created.
1174 * call this only after vfs_create() is called.
1177 nfsd_check_ignore_resizing(struct iattr *iap)
1179 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1180 iap->ia_valid &= ~ATTR_SIZE;
1184 * Create a file (regular, directory, device, fifo); UNIX sockets
1185 * not yet implemented.
1186 * If the response fh has been verified, the parent directory should
1187 * already be locked. Note that the parent directory is left locked.
1189 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1192 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1193 char *fname, int flen, struct iattr *iap,
1194 int type, dev_t rdev, struct svc_fh *resfhp)
1196 struct dentry *dentry, *dchild = NULL;
1206 if (isdotent(fname, flen))
1209 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1213 dentry = fhp->fh_dentry;
1214 dirp = dentry->d_inode;
1216 err = nfserr_notdir;
1217 if (!dirp->i_op->lookup)
1220 * Check whether the response file handle has been verified yet.
1221 * If it has, the parent directory should already be locked.
1223 if (!resfhp->fh_dentry) {
1224 host_err = fh_want_write(fhp);
1228 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1229 fh_lock_nested(fhp, I_MUTEX_PARENT);
1230 dchild = lookup_one_len(fname, dentry, flen);
1231 host_err = PTR_ERR(dchild);
1234 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1238 /* called from nfsd_proc_create */
1239 dchild = dget(resfhp->fh_dentry);
1240 if (!fhp->fh_locked) {
1241 /* not actually possible */
1243 "nfsd_create: parent %pd2 not locked!\n",
1250 * Make sure the child dentry is still negative ...
1253 if (dchild->d_inode) {
1254 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1259 if (!(iap->ia_valid & ATTR_MODE))
1261 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1264 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1265 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1271 * Get the dir op function pointer.
1277 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1279 nfsd_check_ignore_resizing(iap);
1282 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1288 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1294 err = nfsd_create_setattr(rqstp, resfhp, iap);
1297 * nfsd_create_setattr already committed the child. Transactional
1298 * filesystems had a chance to commit changes for both parent and
1299 * child * simultaneously making the following commit_metadata a
1302 err2 = nfserrno(commit_metadata(fhp));
1306 * Update the file handle to get the new inode info.
1309 err = fh_update(resfhp);
1311 if (dchild && !IS_ERR(dchild))
1316 err = nfserrno(host_err);
1320 #ifdef CONFIG_NFSD_V3
1322 static inline int nfsd_create_is_exclusive(int createmode)
1324 return createmode == NFS3_CREATE_EXCLUSIVE
1325 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1329 * NFSv3 and NFSv4 version of nfsd_create
1332 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1333 char *fname, int flen, struct iattr *iap,
1334 struct svc_fh *resfhp, int createmode, u32 *verifier,
1335 bool *truncp, bool *created)
1337 struct dentry *dentry, *dchild = NULL;
1341 __u32 v_mtime=0, v_atime=0;
1347 if (isdotent(fname, flen))
1349 if (!(iap->ia_valid & ATTR_MODE))
1351 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1355 dentry = fhp->fh_dentry;
1356 dirp = dentry->d_inode;
1358 /* Get all the sanity checks out of the way before
1359 * we lock the parent. */
1360 err = nfserr_notdir;
1361 if (!dirp->i_op->lookup)
1364 host_err = fh_want_write(fhp);
1368 fh_lock_nested(fhp, I_MUTEX_PARENT);
1371 * Compose the response file handle.
1373 dchild = lookup_one_len(fname, dentry, flen);
1374 host_err = PTR_ERR(dchild);
1378 /* If file doesn't exist, check for permissions to create one */
1379 if (!dchild->d_inode) {
1380 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1385 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1389 if (nfsd_create_is_exclusive(createmode)) {
1390 /* solaris7 gets confused (bugid 4218508) if these have
1391 * the high bit set, so just clear the high bits. If this is
1392 * ever changed to use different attrs for storing the
1393 * verifier, then do_open_lookup() will also need to be fixed
1396 v_mtime = verifier[0]&0x7fffffff;
1397 v_atime = verifier[1]&0x7fffffff;
1400 if (dchild->d_inode) {
1403 switch (createmode) {
1404 case NFS3_CREATE_UNCHECKED:
1405 if (! d_is_reg(dchild))
1408 /* in nfsv4, we need to treat this case a little
1409 * differently. we don't want to truncate the
1410 * file now; this would be wrong if the OPEN
1411 * fails for some other reason. furthermore,
1412 * if the size is nonzero, we should ignore it
1413 * according to spec!
1415 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1418 iap->ia_valid &= ATTR_SIZE;
1422 case NFS3_CREATE_EXCLUSIVE:
1423 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1424 && dchild->d_inode->i_atime.tv_sec == v_atime
1425 && dchild->d_inode->i_size == 0 ) {
1430 case NFS4_CREATE_EXCLUSIVE4_1:
1431 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1432 && dchild->d_inode->i_atime.tv_sec == v_atime
1433 && dchild->d_inode->i_size == 0 ) {
1439 case NFS3_CREATE_GUARDED:
1446 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1454 nfsd_check_ignore_resizing(iap);
1456 if (nfsd_create_is_exclusive(createmode)) {
1457 /* Cram the verifier into atime/mtime */
1458 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1459 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1460 /* XXX someone who knows this better please fix it for nsec */
1461 iap->ia_mtime.tv_sec = v_mtime;
1462 iap->ia_atime.tv_sec = v_atime;
1463 iap->ia_mtime.tv_nsec = 0;
1464 iap->ia_atime.tv_nsec = 0;
1468 err = nfsd_create_setattr(rqstp, resfhp, iap);
1471 * nfsd_create_setattr already committed the child
1472 * (and possibly also the parent).
1475 err = nfserrno(commit_metadata(fhp));
1478 * Update the filehandle to get the new inode info.
1481 err = fh_update(resfhp);
1485 if (dchild && !IS_ERR(dchild))
1491 err = nfserrno(host_err);
1494 #endif /* CONFIG_NFSD_V3 */
1497 * Read a symlink. On entry, *lenp must contain the maximum path length that
1498 * fits into the buffer. On return, it contains the true length.
1499 * N.B. After this call fhp needs an fh_put
1502 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1504 struct inode *inode;
1510 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1514 path.mnt = fhp->fh_export->ex_path.mnt;
1515 path.dentry = fhp->fh_dentry;
1516 inode = path.dentry->d_inode;
1519 if (!inode->i_op->readlink)
1523 /* N.B. Why does this call need a get_fs()??
1524 * Remove the set_fs and watch the fireworks:-) --okir
1527 oldfs = get_fs(); set_fs(KERNEL_DS);
1528 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1539 err = nfserrno(host_err);
1544 * Create a symlink and look up its inode
1545 * N.B. After this call _both_ fhp and resfhp need an fh_put
1548 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1549 char *fname, int flen,
1551 struct svc_fh *resfhp)
1553 struct dentry *dentry, *dnew;
1558 if (!flen || path[0] == '\0')
1561 if (isdotent(fname, flen))
1564 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1568 host_err = fh_want_write(fhp);
1573 dentry = fhp->fh_dentry;
1574 dnew = lookup_one_len(fname, dentry, flen);
1575 host_err = PTR_ERR(dnew);
1579 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1580 err = nfserrno(host_err);
1582 err = nfserrno(commit_metadata(fhp));
1587 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1589 if (err==0) err = cerr;
1594 err = nfserrno(host_err);
1600 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1603 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1604 char *name, int len, struct svc_fh *tfhp)
1606 struct dentry *ddir, *dnew, *dold;
1611 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1614 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1618 if (d_is_dir(tfhp->fh_dentry))
1624 if (isdotent(name, len))
1627 host_err = fh_want_write(tfhp);
1629 err = nfserrno(host_err);
1633 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1634 ddir = ffhp->fh_dentry;
1635 dirp = ddir->d_inode;
1637 dnew = lookup_one_len(name, ddir, len);
1638 host_err = PTR_ERR(dnew);
1642 dold = tfhp->fh_dentry;
1647 host_err = vfs_link(dold, dirp, dnew, NULL);
1649 err = nfserrno(commit_metadata(ffhp));
1651 err = nfserrno(commit_metadata(tfhp));
1653 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1656 err = nfserrno(host_err);
1662 fh_drop_write(tfhp);
1667 err = nfserrno(host_err);
1673 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1676 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1677 struct svc_fh *tfhp, char *tname, int tlen)
1679 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1680 struct inode *fdir, *tdir;
1684 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1687 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1691 fdentry = ffhp->fh_dentry;
1692 fdir = fdentry->d_inode;
1694 tdentry = tfhp->fh_dentry;
1695 tdir = tdentry->d_inode;
1698 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1701 host_err = fh_want_write(ffhp);
1703 err = nfserrno(host_err);
1707 /* cannot use fh_lock as we need deadlock protective ordering
1708 * so do it by hand */
1709 trap = lock_rename(tdentry, fdentry);
1710 ffhp->fh_locked = tfhp->fh_locked = 1;
1714 odentry = lookup_one_len(fname, fdentry, flen);
1715 host_err = PTR_ERR(odentry);
1716 if (IS_ERR(odentry))
1720 if (!odentry->d_inode)
1723 if (odentry == trap)
1726 ndentry = lookup_one_len(tname, tdentry, tlen);
1727 host_err = PTR_ERR(ndentry);
1728 if (IS_ERR(ndentry))
1730 host_err = -ENOTEMPTY;
1731 if (ndentry == trap)
1735 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1737 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1740 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0);
1742 host_err = commit_metadata(tfhp);
1744 host_err = commit_metadata(ffhp);
1751 err = nfserrno(host_err);
1753 * We cannot rely on fh_unlock on the two filehandles,
1754 * as that would do the wrong thing if the two directories
1755 * were the same, so again we do it by hand.
1757 fill_post_wcc(ffhp);
1758 fill_post_wcc(tfhp);
1759 unlock_rename(tdentry, fdentry);
1760 ffhp->fh_locked = tfhp->fh_locked = 0;
1761 fh_drop_write(ffhp);
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 host_err = fh_want_write(fhp);
1791 fh_lock_nested(fhp, I_MUTEX_PARENT);
1792 dentry = fhp->fh_dentry;
1793 dirp = dentry->d_inode;
1795 rdentry = lookup_one_len(fname, dentry, flen);
1796 host_err = PTR_ERR(rdentry);
1797 if (IS_ERR(rdentry))
1800 if (!rdentry->d_inode) {
1807 type = rdentry->d_inode->i_mode & S_IFMT;
1809 if (type != S_IFDIR)
1810 host_err = vfs_unlink(dirp, rdentry, NULL);
1812 host_err = vfs_rmdir(dirp, rdentry);
1814 host_err = commit_metadata(fhp);
1818 err = nfserrno(host_err);
1824 * We do this buffering because we must not call back into the file
1825 * system's ->lookup() method from the filldir callback. That may well
1826 * deadlock a number of file systems.
1828 * This is based heavily on the implementation of same in XFS.
1830 struct buffered_dirent {
1834 unsigned int d_type;
1838 struct readdir_data {
1839 struct dir_context ctx;
1845 static int nfsd_buffered_filldir(struct dir_context *ctx, const char *name,
1846 int namlen, loff_t offset, u64 ino,
1847 unsigned int d_type)
1849 struct readdir_data *buf =
1850 container_of(ctx, struct readdir_data, ctx);
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, nfsd_filldir_t func,
1871 struct readdir_cd *cdp, loff_t *offsetp)
1873 struct buffered_dirent *de;
1877 struct readdir_data buf = {
1878 .ctx.actor = nfsd_buffered_filldir,
1879 .dirent = (void *)__get_free_page(GFP_KERNEL)
1883 return nfserrno(-ENOMEM);
1888 struct inode *dir_inode = file_inode(file);
1889 unsigned int reclen;
1891 cdp->err = nfserr_eof; /* will be cleared on successful read */
1895 host_err = iterate_dir(file, &buf.ctx);
1908 * Various filldir functions may end up calling back into
1909 * lookup_one_len() and the file system's ->lookup() method.
1910 * These expect i_mutex to be held, as it would within readdir.
1912 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1916 de = (struct buffered_dirent *)buf.dirent;
1918 offset = de->offset;
1920 if (func(cdp, de->name, de->namlen, de->offset,
1921 de->ino, de->d_type))
1924 if (cdp->err != nfs_ok)
1927 reclen = ALIGN(sizeof(*de) + de->namlen,
1930 de = (struct buffered_dirent *)((char *)de + reclen);
1932 mutex_unlock(&dir_inode->i_mutex);
1933 if (size > 0) /* We bailed out early */
1936 offset = vfs_llseek(file, 0, SEEK_CUR);
1939 free_page((unsigned long)(buf.dirent));
1942 return nfserrno(host_err);
1949 * Read entries from a directory.
1950 * The NFSv3/4 verifier we ignore for now.
1953 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1954 struct readdir_cd *cdp, nfsd_filldir_t func)
1958 loff_t offset = *offsetp;
1959 int may_flags = NFSD_MAY_READ;
1961 /* NFSv2 only supports 32 bit cookies */
1962 if (rqstp->rq_vers > 2)
1963 may_flags |= NFSD_MAY_64BIT_COOKIE;
1965 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
1969 offset = vfs_llseek(file, offset, SEEK_SET);
1971 err = nfserrno((int)offset);
1975 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1977 if (err == nfserr_eof || err == nfserr_toosmall)
1978 err = nfs_ok; /* can still be found in ->err */
1986 * Get file system stats
1987 * N.B. After this call fhp needs an fh_put
1990 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1994 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1996 struct path path = {
1997 .mnt = fhp->fh_export->ex_path.mnt,
1998 .dentry = fhp->fh_dentry,
2000 if (vfs_statfs(&path, stat))
2006 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2008 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2012 * Check for a user's access permissions to this inode.
2015 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2016 struct dentry *dentry, int acc)
2018 struct inode *inode = dentry->d_inode;
2021 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2024 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2026 (acc & NFSD_MAY_READ)? " read" : "",
2027 (acc & NFSD_MAY_WRITE)? " write" : "",
2028 (acc & NFSD_MAY_EXEC)? " exec" : "",
2029 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2030 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2031 (acc & NFSD_MAY_LOCK)? " lock" : "",
2032 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2034 IS_IMMUTABLE(inode)? " immut" : "",
2035 IS_APPEND(inode)? " append" : "",
2036 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2037 dprintk(" owner %d/%d user %d/%d\n",
2038 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2041 /* Normally we reject any write/sattr etc access on a read-only file
2042 * system. But if it is IRIX doing check on write-access for a
2043 * device special file, we ignore rofs.
2045 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2046 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2047 if (exp_rdonly(rqstp, exp) ||
2048 __mnt_is_readonly(exp->ex_path.mnt))
2050 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2053 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2056 if (acc & NFSD_MAY_LOCK) {
2057 /* If we cannot rely on authentication in NLM requests,
2058 * just allow locks, otherwise require read permission, or
2061 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2064 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2067 * The file owner always gets access permission for accesses that
2068 * would normally be checked at open time. This is to make
2069 * file access work even when the client has done a fchmod(fd, 0).
2071 * However, `cp foo bar' should fail nevertheless when bar is
2072 * readonly. A sensible way to do this might be to reject all
2073 * attempts to truncate a read-only file, because a creat() call
2074 * always implies file truncation.
2075 * ... but this isn't really fair. A process may reasonably call
2076 * ftruncate on an open file descriptor on a file with perm 000.
2077 * We must trust the client to do permission checking - using "ACCESS"
2080 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2081 uid_eq(inode->i_uid, current_fsuid()))
2084 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2085 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2087 /* Allow read access to binaries even when mode 111 */
2088 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2089 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2090 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2091 err = inode_permission(inode, MAY_EXEC);
2093 return err? nfserrno(err) : 0;
2097 nfsd_racache_shutdown(void)
2099 struct raparms *raparm, *last_raparm;
2102 dprintk("nfsd: freeing readahead buffers.\n");
2104 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2105 raparm = raparm_hash[i].pb_head;
2107 last_raparm = raparm;
2108 raparm = raparm->p_next;
2111 raparm_hash[i].pb_head = NULL;
2115 * Initialize readahead param cache
2118 nfsd_racache_init(int cache_size)
2123 struct raparms **raparm = NULL;
2126 if (raparm_hash[0].pb_head)
2128 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2129 nperbucket = max(2, nperbucket);
2130 cache_size = nperbucket * RAPARM_HASH_SIZE;
2132 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2134 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2135 spin_lock_init(&raparm_hash[i].pb_lock);
2137 raparm = &raparm_hash[i].pb_head;
2138 for (j = 0; j < nperbucket; j++) {
2139 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2142 raparm = &(*raparm)->p_next;
2147 nfsdstats.ra_size = cache_size;
2151 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2152 nfsd_racache_shutdown();