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/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>
31 #include <linux/security.h>
35 #endif /* CONFIG_NFSD_V3 */
40 #endif /* CONFIG_NFSD_V4 */
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
56 struct raparms *p_next;
61 struct file_ra_state p_ra;
62 unsigned int p_hindex;
65 struct raparm_hbucket {
66 struct raparms *pb_head;
68 } ____cacheline_aligned_in_smp;
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
82 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
83 struct svc_export **expp)
85 struct svc_export *exp = *expp, *exp2 = NULL;
86 struct dentry *dentry = *dpp;
87 struct path path = {.mnt = mntget(exp->ex_path.mnt),
88 .dentry = dget(dentry)};
91 err = follow_down(&path);
95 exp2 = rqst_exp_get_by_name(rqstp, &path);
99 * We normally allow NFS clients to continue
100 * "underneath" a mountpoint that is not exported.
101 * The exception is V4ROOT, where no traversal is ever
102 * allowed without an explicit export of the new
105 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
110 if (nfsd_v4client(rqstp) ||
111 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
112 /* successfully crossed mount point */
114 * This is subtle: path.dentry is *not* on path.mnt
115 * at this point. The only reason we are safe is that
116 * original mnt is pinned down by exp, so we should
117 * put path *before* putting exp
120 path.dentry = dentry;
130 static void follow_to_parent(struct path *path)
134 while (path->dentry == path->mnt->mnt_root && follow_up(path))
136 dp = dget_parent(path->dentry);
141 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
143 struct svc_export *exp2;
144 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
145 .dentry = dget(dparent)};
147 follow_to_parent(&path);
149 exp2 = rqst_exp_parent(rqstp, &path);
150 if (PTR_ERR(exp2) == -ENOENT) {
151 *dentryp = dget(dparent);
152 } else if (IS_ERR(exp2)) {
154 return PTR_ERR(exp2);
156 *dentryp = dget(path.dentry);
165 * For nfsd purposes, we treat V4ROOT exports as though there was an
166 * export at *every* directory.
168 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
170 if (d_mountpoint(dentry))
172 if (nfsd4_is_junction(dentry))
174 if (!(exp->ex_flags & NFSEXP_V4ROOT))
176 return dentry->d_inode != NULL;
180 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
181 const char *name, unsigned int len,
182 struct svc_export **exp_ret, struct dentry **dentry_ret)
184 struct svc_export *exp;
185 struct dentry *dparent;
186 struct dentry *dentry;
189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
191 dparent = fhp->fh_dentry;
192 exp = 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;
409 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
410 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
411 if (iap->ia_valid & ATTR_SIZE)
415 err = fh_verify(rqstp, fhp, ftype, accmode);
419 dentry = fhp->fh_dentry;
420 inode = dentry->d_inode;
422 /* Ignore any mode updates on symlinks */
423 if (S_ISLNK(inode->i_mode))
424 iap->ia_valid &= ~ATTR_MODE;
429 nfsd_sanitize_attrs(inode, iap);
432 * The size case is special, it changes the file in addition to the
435 if (iap->ia_valid & ATTR_SIZE) {
436 err = nfsd_get_write_access(rqstp, fhp, iap);
442 iap->ia_valid |= ATTR_CTIME;
444 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
445 err = nfserr_notsync;
446 goto out_put_write_access;
450 host_err = notify_change(dentry, iap, NULL);
452 err = nfserrno(host_err);
454 out_put_write_access:
456 put_write_access(inode);
458 commit_metadata(fhp);
463 #if defined(CONFIG_NFSD_V4)
465 * NFS junction information is stored in an extended attribute.
467 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
470 * nfsd4_is_junction - Test if an object could be an NFS junction
472 * @dentry: object to test
474 * Returns 1 if "dentry" appears to contain NFS junction information.
475 * Otherwise 0 is returned.
477 int nfsd4_is_junction(struct dentry *dentry)
479 struct inode *inode = dentry->d_inode;
483 if (inode->i_mode & S_IXUGO)
485 if (!(inode->i_mode & S_ISVTX))
487 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
491 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
492 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
493 struct xdr_netobj *label)
497 struct dentry *dentry;
499 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
503 dentry = fhp->fh_dentry;
505 mutex_lock(&dentry->d_inode->i_mutex);
506 host_error = security_inode_setsecctx(dentry, label->data, label->len);
507 mutex_unlock(&dentry->d_inode->i_mutex);
508 return nfserrno(host_error);
511 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
512 struct xdr_netobj *label)
514 return nfserr_notsupp;
518 #endif /* defined(CONFIG_NFSD_V4) */
520 #ifdef CONFIG_NFSD_V3
522 * Check server access rights to a file system object
528 static struct accessmap nfs3_regaccess[] = {
529 { NFS3_ACCESS_READ, NFSD_MAY_READ },
530 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
531 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
532 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
537 static struct accessmap nfs3_diraccess[] = {
538 { NFS3_ACCESS_READ, NFSD_MAY_READ },
539 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
540 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
541 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
542 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
547 static struct accessmap nfs3_anyaccess[] = {
548 /* Some clients - Solaris 2.6 at least, make an access call
549 * to the server to check for access for things like /dev/null
550 * (which really, the server doesn't care about). So
551 * We provide simple access checking for them, looking
552 * mainly at mode bits, and we make sure to ignore read-only
555 { NFS3_ACCESS_READ, NFSD_MAY_READ },
556 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
557 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
558 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
564 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
566 struct accessmap *map;
567 struct svc_export *export;
568 struct dentry *dentry;
569 u32 query, result = 0, sresult = 0;
572 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
576 export = fhp->fh_export;
577 dentry = fhp->fh_dentry;
579 if (S_ISREG(dentry->d_inode->i_mode))
580 map = nfs3_regaccess;
581 else if (S_ISDIR(dentry->d_inode->i_mode))
582 map = nfs3_diraccess;
584 map = nfs3_anyaccess;
588 for (; map->access; map++) {
589 if (map->access & query) {
592 sresult |= map->access;
594 err2 = nfsd_permission(rqstp, export, dentry, map->how);
597 result |= map->access;
600 /* the following error codes just mean the access was not allowed,
601 * rather than an error occurred */
605 /* simply don't "or" in the access bit. */
615 *supported = sresult;
620 #endif /* CONFIG_NFSD_V3 */
622 static int nfsd_open_break_lease(struct inode *inode, int access)
626 if (access & NFSD_MAY_NOT_BREAK_LEASE)
628 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
629 return break_lease(inode, mode | O_NONBLOCK);
633 * Open an existing file or directory.
634 * The may_flags argument indicates the type of open (read/write/lock)
635 * and additional flags.
636 * N.B. After this call fhp needs an fh_put
639 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
640 int may_flags, struct file **filp)
644 int flags = O_RDONLY|O_LARGEFILE;
648 validate_process_creds();
651 * If we get here, then the client has already done an "open",
652 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
653 * in case a chmod has now revoked permission.
655 * Arguably we should also allow the owner override for
656 * directories, but we never have and it doesn't seem to have
657 * caused anyone a problem. If we were to change this, note
658 * also that our filldir callbacks would need a variant of
659 * lookup_one_len that doesn't check permissions.
662 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
663 err = fh_verify(rqstp, fhp, type, may_flags);
667 path.mnt = fhp->fh_export->ex_path.mnt;
668 path.dentry = fhp->fh_dentry;
669 inode = path.dentry->d_inode;
671 /* Disallow write access to files with the append-only bit set
672 * or any access when mandatory locking enabled
675 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
678 * We must ignore files (but only files) which might have mandatory
679 * locks on them because there is no way to know if the accesser has
682 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
688 host_err = nfsd_open_break_lease(inode, may_flags);
689 if (host_err) /* NOMEM or WOULDBLOCK */
692 if (may_flags & NFSD_MAY_WRITE) {
693 if (may_flags & NFSD_MAY_READ)
694 flags = O_RDWR|O_LARGEFILE;
696 flags = O_WRONLY|O_LARGEFILE;
698 *filp = dentry_open(&path, flags, current_cred());
700 host_err = PTR_ERR(*filp);
703 host_err = ima_file_check(*filp, may_flags);
705 if (may_flags & NFSD_MAY_64BIT_COOKIE)
706 (*filp)->f_mode |= FMODE_64BITHASH;
708 (*filp)->f_mode |= FMODE_32BITHASH;
712 err = nfserrno(host_err);
714 validate_process_creds();
722 nfsd_close(struct file *filp)
728 * Obtain the readahead parameters for the file
729 * specified by (dev, ino).
732 static inline struct raparms *
733 nfsd_get_raparms(dev_t dev, ino_t ino)
735 struct raparms *ra, **rap, **frap = NULL;
738 struct raparm_hbucket *rab;
740 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
741 rab = &raparm_hash[hash];
743 spin_lock(&rab->pb_lock);
744 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
745 if (ra->p_ino == ino && ra->p_dev == dev)
748 if (ra->p_count == 0)
751 depth = nfsdstats.ra_size;
753 spin_unlock(&rab->pb_lock);
763 if (rap != &rab->pb_head) {
765 ra->p_next = rab->pb_head;
769 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
770 spin_unlock(&rab->pb_lock);
775 * Grab and keep cached pages associated with a file in the svc_rqst
776 * so that they can be passed to the network sendmsg/sendpage routines
777 * directly. They will be released after the sending has completed.
780 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
781 struct splice_desc *sd)
783 struct svc_rqst *rqstp = sd->u.data;
784 struct page **pp = rqstp->rq_next_page;
785 struct page *page = buf->page;
790 if (rqstp->rq_res.page_len == 0) {
792 put_page(*rqstp->rq_next_page);
793 *(rqstp->rq_next_page++) = page;
794 rqstp->rq_res.page_base = buf->offset;
795 rqstp->rq_res.page_len = size;
796 } else if (page != pp[-1]) {
798 if (*rqstp->rq_next_page)
799 put_page(*rqstp->rq_next_page);
800 *(rqstp->rq_next_page++) = page;
801 rqstp->rq_res.page_len += size;
803 rqstp->rq_res.page_len += size;
808 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
809 struct splice_desc *sd)
811 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
815 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
816 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
824 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
825 struct splice_desc sd = {
832 rqstp->rq_next_page = rqstp->rq_respages + 1;
833 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
837 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
842 nfsdstats.io_read += host_err;
845 fsnotify_access(file);
847 err = nfserrno(host_err);
851 static void kill_suid(struct dentry *dentry)
854 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
856 mutex_lock(&dentry->d_inode->i_mutex);
858 * Note we call this on write, so notify_change will not
859 * encounter any conflicting delegations:
861 notify_change(dentry, &ia, NULL);
862 mutex_unlock(&dentry->d_inode->i_mutex);
866 * Gathered writes: If another process is currently writing to the file,
867 * there's a high chance this is another nfsd (triggered by a bulk write
868 * from a client's biod). Rather than syncing the file with each write
869 * request, we sleep for 10 msec.
871 * I don't know if this roughly approximates C. Juszak's idea of
872 * gathered writes, but it's a nice and simple solution (IMHO), and it
875 * Note: we do this only in the NFSv2 case, since v3 and higher have a
876 * better tool (separate unstable writes and commits) for solving this
879 static int wait_for_concurrent_writes(struct file *file)
881 struct inode *inode = file_inode(file);
882 static ino_t last_ino;
883 static dev_t last_dev;
886 if (atomic_read(&inode->i_writecount) > 1
887 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
888 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
890 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
893 if (inode->i_state & I_DIRTY) {
894 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
895 err = vfs_fsync(file, 0);
897 last_ino = inode->i_ino;
898 last_dev = inode->i_sb->s_dev;
903 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
904 loff_t offset, struct kvec *vec, int vlen,
905 unsigned long *cnt, int *stablep)
907 struct svc_export *exp;
908 struct dentry *dentry;
913 int stable = *stablep;
917 dentry = file->f_path.dentry;
918 inode = dentry->d_inode;
919 exp = fhp->fh_export;
921 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
926 /* Write the data. */
927 oldfs = get_fs(); set_fs(KERNEL_DS);
928 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
933 nfsdstats.io_write += host_err;
934 fsnotify_modify(file);
936 /* clear setuid/setgid flag after write */
937 if (inode->i_mode & (S_ISUID | S_ISGID))
942 host_err = wait_for_concurrent_writes(file);
944 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
948 dprintk("nfsd: write complete host_err=%d\n", host_err);
952 err = nfserrno(host_err);
957 * Read data from a file. count must contain the requested read count
958 * on entry. On return, *count contains the number of bytes actually read.
959 * N.B. After this call fhp needs an fh_put
961 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
962 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
969 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
973 inode = file_inode(file);
975 /* Get readahead parameters */
976 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
979 file->f_ra = ra->p_ra;
981 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
983 /* Write back readahead params */
985 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
986 spin_lock(&rab->pb_lock);
987 ra->p_ra = file->f_ra;
990 spin_unlock(&rab->pb_lock);
997 /* As above, but use the provided file descriptor. */
999 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1000 loff_t offset, struct kvec *vec, int vlen,
1001 unsigned long *count)
1006 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1007 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1010 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1011 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1012 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1018 * Write data to a file.
1019 * The stable flag requests synchronous writes.
1020 * N.B. After this call fhp needs an fh_put
1023 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1024 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1030 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1031 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1034 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1037 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1042 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1050 #ifdef CONFIG_NFSD_V3
1052 * Commit all pending writes to stable storage.
1054 * Note: we only guarantee that data that lies within the range specified
1055 * by the 'offset' and 'count' parameters will be synced.
1057 * Unfortunately we cannot lock the file to make sure we return full WCC
1058 * data to the client, as locking happens lower down in the filesystem.
1061 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1062 loff_t offset, unsigned long count)
1065 loff_t end = LLONG_MAX;
1066 __be32 err = nfserr_inval;
1071 end = offset + (loff_t)count - 1;
1076 err = nfsd_open(rqstp, fhp, S_IFREG,
1077 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1080 if (EX_ISSYNC(fhp->fh_export)) {
1081 int err2 = vfs_fsync_range(file, offset, end, 0);
1083 if (err2 != -EINVAL)
1084 err = nfserrno(err2);
1086 err = nfserr_notsupp;
1093 #endif /* CONFIG_NFSD_V3 */
1096 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1100 * Mode has already been set earlier in create:
1102 iap->ia_valid &= ~ATTR_MODE;
1104 * Setting uid/gid works only for root. Irix appears to
1105 * send along the gid on create when it tries to implement
1106 * setgid directories via NFS:
1108 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1109 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1111 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1115 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1116 * setting size to 0 may fail for some specific file systems by the permission
1117 * checking which requires WRITE permission but the mode is 000.
1118 * we ignore the resizing(to 0) on the just new created file, since the size is
1119 * 0 after file created.
1121 * call this only after vfs_create() is called.
1124 nfsd_check_ignore_resizing(struct iattr *iap)
1126 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1127 iap->ia_valid &= ~ATTR_SIZE;
1131 * Create a file (regular, directory, device, fifo); UNIX sockets
1132 * not yet implemented.
1133 * If the response fh has been verified, the parent directory should
1134 * already be locked. Note that the parent directory is left locked.
1136 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1139 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1140 char *fname, int flen, struct iattr *iap,
1141 int type, dev_t rdev, struct svc_fh *resfhp)
1143 struct dentry *dentry, *dchild = NULL;
1153 if (isdotent(fname, flen))
1156 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1160 dentry = fhp->fh_dentry;
1161 dirp = dentry->d_inode;
1163 err = nfserr_notdir;
1164 if (!dirp->i_op->lookup)
1167 * Check whether the response file handle has been verified yet.
1168 * If it has, the parent directory should already be locked.
1170 if (!resfhp->fh_dentry) {
1171 host_err = fh_want_write(fhp);
1175 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1176 fh_lock_nested(fhp, I_MUTEX_PARENT);
1177 dchild = lookup_one_len(fname, dentry, flen);
1178 host_err = PTR_ERR(dchild);
1181 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1185 /* called from nfsd_proc_create */
1186 dchild = dget(resfhp->fh_dentry);
1187 if (!fhp->fh_locked) {
1188 /* not actually possible */
1190 "nfsd_create: parent %pd2 not locked!\n",
1197 * Make sure the child dentry is still negative ...
1200 if (dchild->d_inode) {
1201 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1206 if (!(iap->ia_valid & ATTR_MODE))
1208 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1211 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1212 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1218 * Get the dir op function pointer.
1224 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1226 nfsd_check_ignore_resizing(iap);
1229 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1235 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1241 err = nfsd_create_setattr(rqstp, resfhp, iap);
1244 * nfsd_setattr already committed the child. Transactional filesystems
1245 * had a chance to commit changes for both parent and child
1246 * simultaneously making the following commit_metadata a noop.
1248 err2 = nfserrno(commit_metadata(fhp));
1252 * Update the file handle to get the new inode info.
1255 err = fh_update(resfhp);
1257 if (dchild && !IS_ERR(dchild))
1262 err = nfserrno(host_err);
1266 #ifdef CONFIG_NFSD_V3
1268 static inline int nfsd_create_is_exclusive(int createmode)
1270 return createmode == NFS3_CREATE_EXCLUSIVE
1271 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1275 * NFSv3 and NFSv4 version of nfsd_create
1278 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1279 char *fname, int flen, struct iattr *iap,
1280 struct svc_fh *resfhp, int createmode, u32 *verifier,
1281 bool *truncp, bool *created)
1283 struct dentry *dentry, *dchild = NULL;
1287 __u32 v_mtime=0, v_atime=0;
1293 if (isdotent(fname, flen))
1295 if (!(iap->ia_valid & ATTR_MODE))
1297 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1301 dentry = fhp->fh_dentry;
1302 dirp = dentry->d_inode;
1304 /* Get all the sanity checks out of the way before
1305 * we lock the parent. */
1306 err = nfserr_notdir;
1307 if (!dirp->i_op->lookup)
1310 host_err = fh_want_write(fhp);
1314 fh_lock_nested(fhp, I_MUTEX_PARENT);
1317 * Compose the response file handle.
1319 dchild = lookup_one_len(fname, dentry, flen);
1320 host_err = PTR_ERR(dchild);
1324 /* If file doesn't exist, check for permissions to create one */
1325 if (!dchild->d_inode) {
1326 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1331 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1335 if (nfsd_create_is_exclusive(createmode)) {
1336 /* solaris7 gets confused (bugid 4218508) if these have
1337 * the high bit set, so just clear the high bits. If this is
1338 * ever changed to use different attrs for storing the
1339 * verifier, then do_open_lookup() will also need to be fixed
1342 v_mtime = verifier[0]&0x7fffffff;
1343 v_atime = verifier[1]&0x7fffffff;
1346 if (dchild->d_inode) {
1349 switch (createmode) {
1350 case NFS3_CREATE_UNCHECKED:
1351 if (! S_ISREG(dchild->d_inode->i_mode))
1354 /* in nfsv4, we need to treat this case a little
1355 * differently. we don't want to truncate the
1356 * file now; this would be wrong if the OPEN
1357 * fails for some other reason. furthermore,
1358 * if the size is nonzero, we should ignore it
1359 * according to spec!
1361 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1364 iap->ia_valid &= ATTR_SIZE;
1368 case NFS3_CREATE_EXCLUSIVE:
1369 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1370 && dchild->d_inode->i_atime.tv_sec == v_atime
1371 && dchild->d_inode->i_size == 0 ) {
1376 case NFS4_CREATE_EXCLUSIVE4_1:
1377 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1378 && dchild->d_inode->i_atime.tv_sec == v_atime
1379 && dchild->d_inode->i_size == 0 ) {
1385 case NFS3_CREATE_GUARDED:
1392 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1400 nfsd_check_ignore_resizing(iap);
1402 if (nfsd_create_is_exclusive(createmode)) {
1403 /* Cram the verifier into atime/mtime */
1404 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1405 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1406 /* XXX someone who knows this better please fix it for nsec */
1407 iap->ia_mtime.tv_sec = v_mtime;
1408 iap->ia_atime.tv_sec = v_atime;
1409 iap->ia_mtime.tv_nsec = 0;
1410 iap->ia_atime.tv_nsec = 0;
1414 err = nfsd_create_setattr(rqstp, resfhp, iap);
1417 * nfsd_setattr already committed the child (and possibly also the parent).
1420 err = nfserrno(commit_metadata(fhp));
1423 * Update the filehandle to get the new inode info.
1426 err = fh_update(resfhp);
1430 if (dchild && !IS_ERR(dchild))
1436 err = nfserrno(host_err);
1439 #endif /* CONFIG_NFSD_V3 */
1442 * Read a symlink. On entry, *lenp must contain the maximum path length that
1443 * fits into the buffer. On return, it contains the true length.
1444 * N.B. After this call fhp needs an fh_put
1447 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1449 struct inode *inode;
1455 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1459 path.mnt = fhp->fh_export->ex_path.mnt;
1460 path.dentry = fhp->fh_dentry;
1461 inode = path.dentry->d_inode;
1464 if (!inode->i_op->readlink)
1468 /* N.B. Why does this call need a get_fs()??
1469 * Remove the set_fs and watch the fireworks:-) --okir
1472 oldfs = get_fs(); set_fs(KERNEL_DS);
1473 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1484 err = nfserrno(host_err);
1489 * Create a symlink and look up its inode
1490 * N.B. After this call _both_ fhp and resfhp need an fh_put
1493 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1494 char *fname, int flen,
1495 char *path, int plen,
1496 struct svc_fh *resfhp,
1499 struct dentry *dentry, *dnew;
1507 if (isdotent(fname, flen))
1510 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1514 host_err = fh_want_write(fhp);
1519 dentry = fhp->fh_dentry;
1520 dnew = lookup_one_len(fname, dentry, flen);
1521 host_err = PTR_ERR(dnew);
1525 if (unlikely(path[plen] != 0)) {
1526 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1527 if (path_alloced == NULL)
1530 strncpy(path_alloced, path, plen);
1531 path_alloced[plen] = 0;
1532 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1533 kfree(path_alloced);
1536 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1537 err = nfserrno(host_err);
1539 err = nfserrno(commit_metadata(fhp));
1544 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1546 if (err==0) err = cerr;
1551 err = nfserrno(host_err);
1557 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1560 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1561 char *name, int len, struct svc_fh *tfhp)
1563 struct dentry *ddir, *dnew, *dold;
1568 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1571 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1575 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1581 if (isdotent(name, len))
1584 host_err = fh_want_write(tfhp);
1586 err = nfserrno(host_err);
1590 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1591 ddir = ffhp->fh_dentry;
1592 dirp = ddir->d_inode;
1594 dnew = lookup_one_len(name, ddir, len);
1595 host_err = PTR_ERR(dnew);
1599 dold = tfhp->fh_dentry;
1604 host_err = vfs_link(dold, dirp, dnew, NULL);
1606 err = nfserrno(commit_metadata(ffhp));
1608 err = nfserrno(commit_metadata(tfhp));
1610 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1613 err = nfserrno(host_err);
1619 fh_drop_write(tfhp);
1624 err = nfserrno(host_err);
1630 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1633 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1634 struct svc_fh *tfhp, char *tname, int tlen)
1636 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1637 struct inode *fdir, *tdir;
1641 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1644 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1648 fdentry = ffhp->fh_dentry;
1649 fdir = fdentry->d_inode;
1651 tdentry = tfhp->fh_dentry;
1652 tdir = tdentry->d_inode;
1655 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1658 host_err = fh_want_write(ffhp);
1660 err = nfserrno(host_err);
1664 /* cannot use fh_lock as we need deadlock protective ordering
1665 * so do it by hand */
1666 trap = lock_rename(tdentry, fdentry);
1667 ffhp->fh_locked = tfhp->fh_locked = 1;
1671 odentry = lookup_one_len(fname, fdentry, flen);
1672 host_err = PTR_ERR(odentry);
1673 if (IS_ERR(odentry))
1677 if (!odentry->d_inode)
1680 if (odentry == trap)
1683 ndentry = lookup_one_len(tname, tdentry, tlen);
1684 host_err = PTR_ERR(ndentry);
1685 if (IS_ERR(ndentry))
1687 host_err = -ENOTEMPTY;
1688 if (ndentry == trap)
1692 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1694 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1697 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL);
1699 host_err = commit_metadata(tfhp);
1701 host_err = commit_metadata(ffhp);
1708 err = nfserrno(host_err);
1710 /* we cannot reply on fh_unlock on the two filehandles,
1711 * as that would do the wrong thing if the two directories
1712 * were the same, so again we do it by hand
1714 fill_post_wcc(ffhp);
1715 fill_post_wcc(tfhp);
1716 unlock_rename(tdentry, fdentry);
1717 ffhp->fh_locked = tfhp->fh_locked = 0;
1718 fh_drop_write(ffhp);
1725 * Unlink a file or directory
1726 * N.B. After this call fhp needs an fh_put
1729 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1730 char *fname, int flen)
1732 struct dentry *dentry, *rdentry;
1738 if (!flen || isdotent(fname, flen))
1740 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1744 host_err = fh_want_write(fhp);
1748 fh_lock_nested(fhp, I_MUTEX_PARENT);
1749 dentry = fhp->fh_dentry;
1750 dirp = dentry->d_inode;
1752 rdentry = lookup_one_len(fname, dentry, flen);
1753 host_err = PTR_ERR(rdentry);
1754 if (IS_ERR(rdentry))
1757 if (!rdentry->d_inode) {
1764 type = rdentry->d_inode->i_mode & S_IFMT;
1766 if (type != S_IFDIR)
1767 host_err = vfs_unlink(dirp, rdentry, NULL);
1769 host_err = vfs_rmdir(dirp, rdentry);
1771 host_err = commit_metadata(fhp);
1775 err = nfserrno(host_err);
1781 * We do this buffering because we must not call back into the file
1782 * system's ->lookup() method from the filldir callback. That may well
1783 * deadlock a number of file systems.
1785 * This is based heavily on the implementation of same in XFS.
1787 struct buffered_dirent {
1791 unsigned int d_type;
1795 struct readdir_data {
1796 struct dir_context ctx;
1802 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1803 loff_t offset, u64 ino, unsigned int d_type)
1805 struct readdir_data *buf = __buf;
1806 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1807 unsigned int reclen;
1809 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1810 if (buf->used + reclen > PAGE_SIZE) {
1815 de->namlen = namlen;
1816 de->offset = offset;
1818 de->d_type = d_type;
1819 memcpy(de->name, name, namlen);
1820 buf->used += reclen;
1825 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1826 struct readdir_cd *cdp, loff_t *offsetp)
1828 struct buffered_dirent *de;
1832 struct readdir_data buf = {
1833 .ctx.actor = nfsd_buffered_filldir,
1834 .dirent = (void *)__get_free_page(GFP_KERNEL)
1838 return nfserrno(-ENOMEM);
1843 struct inode *dir_inode = file_inode(file);
1844 unsigned int reclen;
1846 cdp->err = nfserr_eof; /* will be cleared on successful read */
1850 host_err = iterate_dir(file, &buf.ctx);
1863 * Various filldir functions may end up calling back into
1864 * lookup_one_len() and the file system's ->lookup() method.
1865 * These expect i_mutex to be held, as it would within readdir.
1867 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1871 de = (struct buffered_dirent *)buf.dirent;
1873 offset = de->offset;
1875 if (func(cdp, de->name, de->namlen, de->offset,
1876 de->ino, de->d_type))
1879 if (cdp->err != nfs_ok)
1882 reclen = ALIGN(sizeof(*de) + de->namlen,
1885 de = (struct buffered_dirent *)((char *)de + reclen);
1887 mutex_unlock(&dir_inode->i_mutex);
1888 if (size > 0) /* We bailed out early */
1891 offset = vfs_llseek(file, 0, SEEK_CUR);
1894 free_page((unsigned long)(buf.dirent));
1897 return nfserrno(host_err);
1904 * Read entries from a directory.
1905 * The NFSv3/4 verifier we ignore for now.
1908 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1909 struct readdir_cd *cdp, filldir_t func)
1913 loff_t offset = *offsetp;
1914 int may_flags = NFSD_MAY_READ;
1916 /* NFSv2 only supports 32 bit cookies */
1917 if (rqstp->rq_vers > 2)
1918 may_flags |= NFSD_MAY_64BIT_COOKIE;
1920 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
1924 offset = vfs_llseek(file, offset, SEEK_SET);
1926 err = nfserrno((int)offset);
1930 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1932 if (err == nfserr_eof || err == nfserr_toosmall)
1933 err = nfs_ok; /* can still be found in ->err */
1941 * Get file system stats
1942 * N.B. After this call fhp needs an fh_put
1945 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1949 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1951 struct path path = {
1952 .mnt = fhp->fh_export->ex_path.mnt,
1953 .dentry = fhp->fh_dentry,
1955 if (vfs_statfs(&path, stat))
1961 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1963 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1967 * Check for a user's access permissions to this inode.
1970 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1971 struct dentry *dentry, int acc)
1973 struct inode *inode = dentry->d_inode;
1976 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
1979 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1981 (acc & NFSD_MAY_READ)? " read" : "",
1982 (acc & NFSD_MAY_WRITE)? " write" : "",
1983 (acc & NFSD_MAY_EXEC)? " exec" : "",
1984 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1985 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1986 (acc & NFSD_MAY_LOCK)? " lock" : "",
1987 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1989 IS_IMMUTABLE(inode)? " immut" : "",
1990 IS_APPEND(inode)? " append" : "",
1991 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
1992 dprintk(" owner %d/%d user %d/%d\n",
1993 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
1996 /* Normally we reject any write/sattr etc access on a read-only file
1997 * system. But if it is IRIX doing check on write-access for a
1998 * device special file, we ignore rofs.
2000 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2001 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2002 if (exp_rdonly(rqstp, exp) ||
2003 __mnt_is_readonly(exp->ex_path.mnt))
2005 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2008 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2011 if (acc & NFSD_MAY_LOCK) {
2012 /* If we cannot rely on authentication in NLM requests,
2013 * just allow locks, otherwise require read permission, or
2016 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2019 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2022 * The file owner always gets access permission for accesses that
2023 * would normally be checked at open time. This is to make
2024 * file access work even when the client has done a fchmod(fd, 0).
2026 * However, `cp foo bar' should fail nevertheless when bar is
2027 * readonly. A sensible way to do this might be to reject all
2028 * attempts to truncate a read-only file, because a creat() call
2029 * always implies file truncation.
2030 * ... but this isn't really fair. A process may reasonably call
2031 * ftruncate on an open file descriptor on a file with perm 000.
2032 * We must trust the client to do permission checking - using "ACCESS"
2035 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2036 uid_eq(inode->i_uid, current_fsuid()))
2039 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2040 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2042 /* Allow read access to binaries even when mode 111 */
2043 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2044 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2045 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2046 err = inode_permission(inode, MAY_EXEC);
2048 return err? nfserrno(err) : 0;
2052 nfsd_racache_shutdown(void)
2054 struct raparms *raparm, *last_raparm;
2057 dprintk("nfsd: freeing readahead buffers.\n");
2059 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2060 raparm = raparm_hash[i].pb_head;
2062 last_raparm = raparm;
2063 raparm = raparm->p_next;
2066 raparm_hash[i].pb_head = NULL;
2070 * Initialize readahead param cache
2073 nfsd_racache_init(int cache_size)
2078 struct raparms **raparm = NULL;
2081 if (raparm_hash[0].pb_head)
2083 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2086 cache_size = nperbucket * RAPARM_HASH_SIZE;
2088 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2090 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2091 spin_lock_init(&raparm_hash[i].pb_lock);
2093 raparm = &raparm_hash[i].pb_head;
2094 for (j = 0; j < nperbucket; j++) {
2095 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2098 raparm = &(*raparm)->p_next;
2103 nfsdstats.ra_size = cache_size;
2107 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2108 nfsd_racache_shutdown();