ipv4: RCU conversion of ip_route_input_slow/ip_route_input_mc
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / nfsd / vfs.c
blobebbf3b6b24577128c3182072d34933770d56e644
1 #define MSNFS /* HACK HACK */
2 /*
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/fs.h>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/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>
33 #ifdef CONFIG_NFSD_V3
34 #include "xdr3.h"
35 #endif /* CONFIG_NFSD_V3 */
37 #ifdef CONFIG_NFSD_V4
38 #include <linux/nfs4_acl.h>
39 #include <linux/nfsd_idmap.h>
40 #endif /* CONFIG_NFSD_V4 */
42 #include "nfsd.h"
43 #include "vfs.h"
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.
55 struct raparms {
56 struct raparms *p_next;
57 unsigned int p_count;
58 ino_t p_ino;
59 dev_t p_dev;
60 int p_set;
61 struct file_ra_state p_ra;
62 unsigned int p_hindex;
65 struct raparm_hbucket {
66 struct raparms *pb_head;
67 spinlock_t pb_lock;
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];
75 /*
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * a mount point.
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
81 int
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)};
89 int err = 0;
91 while (d_mountpoint(path.dentry) && follow_down(&path))
94 exp2 = rqst_exp_get_by_name(rqstp, &path);
95 if (IS_ERR(exp2)) {
96 err = PTR_ERR(exp2);
98 * We normally allow NFS clients to continue
99 * "underneath" a mountpoint that is not exported.
100 * The exception is V4ROOT, where no traversal is ever
101 * allowed without an explicit export of the new
102 * directory.
104 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
105 err = 0;
106 path_put(&path);
107 goto out;
109 if (nfsd_v4client(rqstp) ||
110 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
111 /* successfully crossed mount point */
113 * This is subtle: path.dentry is *not* on path.mnt
114 * at this point. The only reason we are safe is that
115 * original mnt is pinned down by exp, so we should
116 * put path *before* putting exp
118 *dpp = path.dentry;
119 path.dentry = dentry;
120 *expp = exp2;
121 exp2 = exp;
123 path_put(&path);
124 exp_put(exp2);
125 out:
126 return err;
129 static void follow_to_parent(struct path *path)
131 struct dentry *dp;
133 while (path->dentry == path->mnt->mnt_root && follow_up(path))
135 dp = dget_parent(path->dentry);
136 dput(path->dentry);
137 path->dentry = dp;
140 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
142 struct svc_export *exp2;
143 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
144 .dentry = dget(dparent)};
146 follow_to_parent(&path);
148 exp2 = rqst_exp_parent(rqstp, &path);
149 if (PTR_ERR(exp2) == -ENOENT) {
150 *dentryp = dget(dparent);
151 } else if (IS_ERR(exp2)) {
152 path_put(&path);
153 return PTR_ERR(exp2);
154 } else {
155 *dentryp = dget(path.dentry);
156 exp_put(*exp);
157 *exp = exp2;
159 path_put(&path);
160 return 0;
164 * For nfsd purposes, we treat V4ROOT exports as though there was an
165 * export at *every* directory.
167 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
169 if (d_mountpoint(dentry))
170 return 1;
171 if (!(exp->ex_flags & NFSEXP_V4ROOT))
172 return 0;
173 return dentry->d_inode != NULL;
176 __be32
177 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
178 const char *name, unsigned int len,
179 struct svc_export **exp_ret, struct dentry **dentry_ret)
181 struct svc_export *exp;
182 struct dentry *dparent;
183 struct dentry *dentry;
184 __be32 err;
185 int host_err;
187 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
189 /* Obtain dentry and export. */
190 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
191 if (err)
192 return err;
194 dparent = fhp->fh_dentry;
195 exp = fhp->fh_export;
196 exp_get(exp);
198 /* Lookup the name, but don't follow links */
199 if (isdotent(name, len)) {
200 if (len==1)
201 dentry = dget(dparent);
202 else if (dparent != exp->ex_path.dentry)
203 dentry = dget_parent(dparent);
204 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
205 dentry = dget(dparent); /* .. == . just like at / */
206 else {
207 /* checking mountpoint crossing is very different when stepping up */
208 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
209 if (host_err)
210 goto out_nfserr;
212 } else {
213 fh_lock(fhp);
214 dentry = lookup_one_len(name, dparent, len);
215 host_err = PTR_ERR(dentry);
216 if (IS_ERR(dentry))
217 goto out_nfserr;
219 * check if we have crossed a mount point ...
221 if (nfsd_mountpoint(dentry, exp)) {
222 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
223 dput(dentry);
224 goto out_nfserr;
228 *dentry_ret = dentry;
229 *exp_ret = exp;
230 return 0;
232 out_nfserr:
233 exp_put(exp);
234 return nfserrno(host_err);
238 * Look up one component of a pathname.
239 * N.B. After this call _both_ fhp and resfh need an fh_put
241 * If the lookup would cross a mountpoint, and the mounted filesystem
242 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
243 * accepted as it stands and the mounted directory is
244 * returned. Otherwise the covered directory is returned.
245 * NOTE: this mountpoint crossing is not supported properly by all
246 * clients and is explicitly disallowed for NFSv3
247 * NeilBrown <neilb@cse.unsw.edu.au>
249 __be32
250 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
251 unsigned int len, struct svc_fh *resfh)
253 struct svc_export *exp;
254 struct dentry *dentry;
255 __be32 err;
257 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
258 if (err)
259 return err;
260 err = check_nfsd_access(exp, rqstp);
261 if (err)
262 goto out;
264 * Note: we compose the file handle now, but as the
265 * dentry may be negative, it may need to be updated.
267 err = fh_compose(resfh, exp, dentry, fhp);
268 if (!err && !dentry->d_inode)
269 err = nfserr_noent;
270 out:
271 dput(dentry);
272 exp_put(exp);
273 return err;
277 * Commit metadata changes to stable storage.
279 static int
280 commit_metadata(struct svc_fh *fhp)
282 struct inode *inode = fhp->fh_dentry->d_inode;
283 const struct export_operations *export_ops = inode->i_sb->s_export_op;
284 int error = 0;
286 if (!EX_ISSYNC(fhp->fh_export))
287 return 0;
289 if (export_ops->commit_metadata) {
290 error = export_ops->commit_metadata(inode);
291 } else {
292 struct writeback_control wbc = {
293 .sync_mode = WB_SYNC_ALL,
294 .nr_to_write = 0, /* metadata only */
297 error = sync_inode(inode, &wbc);
300 return error;
304 * Set various file attributes.
305 * N.B. After this call fhp needs an fh_put
307 __be32
308 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
309 int check_guard, time_t guardtime)
311 struct dentry *dentry;
312 struct inode *inode;
313 int accmode = NFSD_MAY_SATTR;
314 int ftype = 0;
315 __be32 err;
316 int host_err;
317 int size_change = 0;
319 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
320 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
321 if (iap->ia_valid & ATTR_SIZE)
322 ftype = S_IFREG;
324 /* Get inode */
325 err = fh_verify(rqstp, fhp, ftype, accmode);
326 if (err)
327 goto out;
329 dentry = fhp->fh_dentry;
330 inode = dentry->d_inode;
332 /* Ignore any mode updates on symlinks */
333 if (S_ISLNK(inode->i_mode))
334 iap->ia_valid &= ~ATTR_MODE;
336 if (!iap->ia_valid)
337 goto out;
340 * NFSv2 does not differentiate between "set-[ac]time-to-now"
341 * which only requires access, and "set-[ac]time-to-X" which
342 * requires ownership.
343 * So if it looks like it might be "set both to the same time which
344 * is close to now", and if inode_change_ok fails, then we
345 * convert to "set to now" instead of "set to explicit time"
347 * We only call inode_change_ok as the last test as technically
348 * it is not an interface that we should be using. It is only
349 * valid if the filesystem does not define it's own i_op->setattr.
351 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
352 #define MAX_TOUCH_TIME_ERROR (30*60)
353 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
354 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
356 * Looks probable.
358 * Now just make sure time is in the right ballpark.
359 * Solaris, at least, doesn't seem to care what the time
360 * request is. We require it be within 30 minutes of now.
362 time_t delta = iap->ia_atime.tv_sec - get_seconds();
363 if (delta < 0)
364 delta = -delta;
365 if (delta < MAX_TOUCH_TIME_ERROR &&
366 inode_change_ok(inode, iap) != 0) {
368 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
369 * This will cause notify_change to set these times
370 * to "now"
372 iap->ia_valid &= ~BOTH_TIME_SET;
377 * The size case is special.
378 * It changes the file as well as the attributes.
380 if (iap->ia_valid & ATTR_SIZE) {
381 if (iap->ia_size < inode->i_size) {
382 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
383 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
384 if (err)
385 goto out;
389 * If we are changing the size of the file, then
390 * we need to break all leases.
392 host_err = break_lease(inode, O_WRONLY | O_NONBLOCK);
393 if (host_err == -EWOULDBLOCK)
394 host_err = -ETIMEDOUT;
395 if (host_err) /* ENOMEM or EWOULDBLOCK */
396 goto out_nfserr;
398 host_err = get_write_access(inode);
399 if (host_err)
400 goto out_nfserr;
402 size_change = 1;
403 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
404 if (host_err) {
405 put_write_access(inode);
406 goto out_nfserr;
410 /* sanitize the mode change */
411 if (iap->ia_valid & ATTR_MODE) {
412 iap->ia_mode &= S_IALLUGO;
413 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
416 /* Revoke setuid/setgid on chown */
417 if (!S_ISDIR(inode->i_mode) &&
418 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
419 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
420 iap->ia_valid |= ATTR_KILL_PRIV;
421 if (iap->ia_valid & ATTR_MODE) {
422 /* we're setting mode too, just clear the s*id bits */
423 iap->ia_mode &= ~S_ISUID;
424 if (iap->ia_mode & S_IXGRP)
425 iap->ia_mode &= ~S_ISGID;
426 } else {
427 /* set ATTR_KILL_* bits and let VFS handle it */
428 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
432 /* Change the attributes. */
434 iap->ia_valid |= ATTR_CTIME;
436 err = nfserr_notsync;
437 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
438 fh_lock(fhp);
439 host_err = notify_change(dentry, iap);
440 err = nfserrno(host_err);
441 fh_unlock(fhp);
443 if (size_change)
444 put_write_access(inode);
445 if (!err)
446 if (EX_ISSYNC(fhp->fh_export))
447 write_inode_now(inode, 1);
448 out:
449 return err;
451 out_nfserr:
452 err = nfserrno(host_err);
453 goto out;
456 #if defined(CONFIG_NFSD_V2_ACL) || \
457 defined(CONFIG_NFSD_V3_ACL) || \
458 defined(CONFIG_NFSD_V4)
459 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
461 ssize_t buflen;
462 ssize_t ret;
464 buflen = vfs_getxattr(dentry, key, NULL, 0);
465 if (buflen <= 0)
466 return buflen;
468 *buf = kmalloc(buflen, GFP_KERNEL);
469 if (!*buf)
470 return -ENOMEM;
472 ret = vfs_getxattr(dentry, key, *buf, buflen);
473 if (ret < 0)
474 kfree(*buf);
475 return ret;
477 #endif
479 #if defined(CONFIG_NFSD_V4)
480 static int
481 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
483 int len;
484 size_t buflen;
485 char *buf = NULL;
486 int error = 0;
488 buflen = posix_acl_xattr_size(pacl->a_count);
489 buf = kmalloc(buflen, GFP_KERNEL);
490 error = -ENOMEM;
491 if (buf == NULL)
492 goto out;
494 len = posix_acl_to_xattr(pacl, buf, buflen);
495 if (len < 0) {
496 error = len;
497 goto out;
500 error = vfs_setxattr(dentry, key, buf, len, 0);
501 out:
502 kfree(buf);
503 return error;
506 __be32
507 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
508 struct nfs4_acl *acl)
510 __be32 error;
511 int host_error;
512 struct dentry *dentry;
513 struct inode *inode;
514 struct posix_acl *pacl = NULL, *dpacl = NULL;
515 unsigned int flags = 0;
517 /* Get inode */
518 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
519 if (error)
520 return error;
522 dentry = fhp->fh_dentry;
523 inode = dentry->d_inode;
524 if (S_ISDIR(inode->i_mode))
525 flags = NFS4_ACL_DIR;
527 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
528 if (host_error == -EINVAL) {
529 return nfserr_attrnotsupp;
530 } else if (host_error < 0)
531 goto out_nfserr;
533 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
534 if (host_error < 0)
535 goto out_release;
537 if (S_ISDIR(inode->i_mode))
538 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
540 out_release:
541 posix_acl_release(pacl);
542 posix_acl_release(dpacl);
543 out_nfserr:
544 if (host_error == -EOPNOTSUPP)
545 return nfserr_attrnotsupp;
546 else
547 return nfserrno(host_error);
550 static struct posix_acl *
551 _get_posix_acl(struct dentry *dentry, char *key)
553 void *buf = NULL;
554 struct posix_acl *pacl = NULL;
555 int buflen;
557 buflen = nfsd_getxattr(dentry, key, &buf);
558 if (!buflen)
559 buflen = -ENODATA;
560 if (buflen <= 0)
561 return ERR_PTR(buflen);
563 pacl = posix_acl_from_xattr(buf, buflen);
564 kfree(buf);
565 return pacl;
569 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
571 struct inode *inode = dentry->d_inode;
572 int error = 0;
573 struct posix_acl *pacl = NULL, *dpacl = NULL;
574 unsigned int flags = 0;
576 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
577 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
578 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
579 if (IS_ERR(pacl)) {
580 error = PTR_ERR(pacl);
581 pacl = NULL;
582 goto out;
585 if (S_ISDIR(inode->i_mode)) {
586 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
587 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
588 dpacl = NULL;
589 else if (IS_ERR(dpacl)) {
590 error = PTR_ERR(dpacl);
591 dpacl = NULL;
592 goto out;
594 flags = NFS4_ACL_DIR;
597 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
598 if (IS_ERR(*acl)) {
599 error = PTR_ERR(*acl);
600 *acl = NULL;
602 out:
603 posix_acl_release(pacl);
604 posix_acl_release(dpacl);
605 return error;
608 #endif /* defined(CONFIG_NFS_V4) */
610 #ifdef CONFIG_NFSD_V3
612 * Check server access rights to a file system object
614 struct accessmap {
615 u32 access;
616 int how;
618 static struct accessmap nfs3_regaccess[] = {
619 { NFS3_ACCESS_READ, NFSD_MAY_READ },
620 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
621 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
622 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
624 { 0, 0 }
627 static struct accessmap nfs3_diraccess[] = {
628 { NFS3_ACCESS_READ, NFSD_MAY_READ },
629 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
630 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
631 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
632 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
634 { 0, 0 }
637 static struct accessmap nfs3_anyaccess[] = {
638 /* Some clients - Solaris 2.6 at least, make an access call
639 * to the server to check for access for things like /dev/null
640 * (which really, the server doesn't care about). So
641 * We provide simple access checking for them, looking
642 * mainly at mode bits, and we make sure to ignore read-only
643 * filesystem checks
645 { NFS3_ACCESS_READ, NFSD_MAY_READ },
646 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
647 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
648 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
650 { 0, 0 }
653 __be32
654 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
656 struct accessmap *map;
657 struct svc_export *export;
658 struct dentry *dentry;
659 u32 query, result = 0, sresult = 0;
660 __be32 error;
662 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
663 if (error)
664 goto out;
666 export = fhp->fh_export;
667 dentry = fhp->fh_dentry;
669 if (S_ISREG(dentry->d_inode->i_mode))
670 map = nfs3_regaccess;
671 else if (S_ISDIR(dentry->d_inode->i_mode))
672 map = nfs3_diraccess;
673 else
674 map = nfs3_anyaccess;
677 query = *access;
678 for (; map->access; map++) {
679 if (map->access & query) {
680 __be32 err2;
682 sresult |= map->access;
684 err2 = nfsd_permission(rqstp, export, dentry, map->how);
685 switch (err2) {
686 case nfs_ok:
687 result |= map->access;
688 break;
690 /* the following error codes just mean the access was not allowed,
691 * rather than an error occurred */
692 case nfserr_rofs:
693 case nfserr_acces:
694 case nfserr_perm:
695 /* simply don't "or" in the access bit. */
696 break;
697 default:
698 error = err2;
699 goto out;
703 *access = result;
704 if (supported)
705 *supported = sresult;
707 out:
708 return error;
710 #endif /* CONFIG_NFSD_V3 */
715 * Open an existing file or directory.
716 * The access argument indicates the type of open (read/write/lock)
717 * N.B. After this call fhp needs an fh_put
719 __be32
720 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
721 int access, struct file **filp)
723 struct dentry *dentry;
724 struct inode *inode;
725 int flags = O_RDONLY|O_LARGEFILE;
726 __be32 err;
727 int host_err = 0;
729 validate_process_creds();
732 * If we get here, then the client has already done an "open",
733 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
734 * in case a chmod has now revoked permission.
736 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
737 if (err)
738 goto out;
740 dentry = fhp->fh_dentry;
741 inode = dentry->d_inode;
743 /* Disallow write access to files with the append-only bit set
744 * or any access when mandatory locking enabled
746 err = nfserr_perm;
747 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
748 goto out;
750 * We must ignore files (but only files) which might have mandatory
751 * locks on them because there is no way to know if the accesser has
752 * the lock.
754 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
755 goto out;
757 if (!inode->i_fop)
758 goto out;
761 * Check to see if there are any leases on this file.
762 * This may block while leases are broken.
764 if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
765 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
766 if (host_err == -EWOULDBLOCK)
767 host_err = -ETIMEDOUT;
768 if (host_err) /* NOMEM or WOULDBLOCK */
769 goto out_nfserr;
771 if (access & NFSD_MAY_WRITE) {
772 if (access & NFSD_MAY_READ)
773 flags = O_RDWR|O_LARGEFILE;
774 else
775 flags = O_WRONLY|O_LARGEFILE;
777 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
778 flags, current_cred());
779 if (IS_ERR(*filp))
780 host_err = PTR_ERR(*filp);
781 else
782 host_err = ima_file_check(*filp, access);
783 out_nfserr:
784 err = nfserrno(host_err);
785 out:
786 validate_process_creds();
787 return err;
791 * Close a file.
793 void
794 nfsd_close(struct file *filp)
796 fput(filp);
800 * Obtain the readahead parameters for the file
801 * specified by (dev, ino).
804 static inline struct raparms *
805 nfsd_get_raparms(dev_t dev, ino_t ino)
807 struct raparms *ra, **rap, **frap = NULL;
808 int depth = 0;
809 unsigned int hash;
810 struct raparm_hbucket *rab;
812 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
813 rab = &raparm_hash[hash];
815 spin_lock(&rab->pb_lock);
816 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
817 if (ra->p_ino == ino && ra->p_dev == dev)
818 goto found;
819 depth++;
820 if (ra->p_count == 0)
821 frap = rap;
823 depth = nfsdstats.ra_size*11/10;
824 if (!frap) {
825 spin_unlock(&rab->pb_lock);
826 return NULL;
828 rap = frap;
829 ra = *frap;
830 ra->p_dev = dev;
831 ra->p_ino = ino;
832 ra->p_set = 0;
833 ra->p_hindex = hash;
834 found:
835 if (rap != &rab->pb_head) {
836 *rap = ra->p_next;
837 ra->p_next = rab->pb_head;
838 rab->pb_head = ra;
840 ra->p_count++;
841 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
842 spin_unlock(&rab->pb_lock);
843 return ra;
847 * Grab and keep cached pages associated with a file in the svc_rqst
848 * so that they can be passed to the network sendmsg/sendpage routines
849 * directly. They will be released after the sending has completed.
851 static int
852 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
853 struct splice_desc *sd)
855 struct svc_rqst *rqstp = sd->u.data;
856 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
857 struct page *page = buf->page;
858 size_t size;
859 int ret;
861 ret = buf->ops->confirm(pipe, buf);
862 if (unlikely(ret))
863 return ret;
865 size = sd->len;
867 if (rqstp->rq_res.page_len == 0) {
868 get_page(page);
869 put_page(*pp);
870 *pp = page;
871 rqstp->rq_resused++;
872 rqstp->rq_res.page_base = buf->offset;
873 rqstp->rq_res.page_len = size;
874 } else if (page != pp[-1]) {
875 get_page(page);
876 if (*pp)
877 put_page(*pp);
878 *pp = page;
879 rqstp->rq_resused++;
880 rqstp->rq_res.page_len += size;
881 } else
882 rqstp->rq_res.page_len += size;
884 return size;
887 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
888 struct splice_desc *sd)
890 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
893 static inline int svc_msnfs(struct svc_fh *ffhp)
895 #ifdef MSNFS
896 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
897 #else
898 return 0;
899 #endif
902 static __be32
903 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
904 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
906 struct inode *inode;
907 struct raparms *ra;
908 mm_segment_t oldfs;
909 __be32 err;
910 int host_err;
912 err = nfserr_perm;
913 inode = file->f_path.dentry->d_inode;
915 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
916 goto out;
918 /* Get readahead parameters */
919 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
921 if (ra && ra->p_set)
922 file->f_ra = ra->p_ra;
924 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
925 struct splice_desc sd = {
926 .len = 0,
927 .total_len = *count,
928 .pos = offset,
929 .u.data = rqstp,
932 rqstp->rq_resused = 1;
933 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
934 } else {
935 oldfs = get_fs();
936 set_fs(KERNEL_DS);
937 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
938 set_fs(oldfs);
941 /* Write back readahead params */
942 if (ra) {
943 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
944 spin_lock(&rab->pb_lock);
945 ra->p_ra = file->f_ra;
946 ra->p_set = 1;
947 ra->p_count--;
948 spin_unlock(&rab->pb_lock);
951 if (host_err >= 0) {
952 nfsdstats.io_read += host_err;
953 *count = host_err;
954 err = 0;
955 fsnotify_access(file->f_path.dentry);
956 } else
957 err = nfserrno(host_err);
958 out:
959 return err;
962 static void kill_suid(struct dentry *dentry)
964 struct iattr ia;
965 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
967 mutex_lock(&dentry->d_inode->i_mutex);
968 notify_change(dentry, &ia);
969 mutex_unlock(&dentry->d_inode->i_mutex);
973 * Gathered writes: If another process is currently writing to the file,
974 * there's a high chance this is another nfsd (triggered by a bulk write
975 * from a client's biod). Rather than syncing the file with each write
976 * request, we sleep for 10 msec.
978 * I don't know if this roughly approximates C. Juszak's idea of
979 * gathered writes, but it's a nice and simple solution (IMHO), and it
980 * seems to work:-)
982 * Note: we do this only in the NFSv2 case, since v3 and higher have a
983 * better tool (separate unstable writes and commits) for solving this
984 * problem.
986 static int wait_for_concurrent_writes(struct file *file)
988 struct inode *inode = file->f_path.dentry->d_inode;
989 static ino_t last_ino;
990 static dev_t last_dev;
991 int err = 0;
993 if (atomic_read(&inode->i_writecount) > 1
994 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
995 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
996 msleep(10);
997 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1000 if (inode->i_state & I_DIRTY) {
1001 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1002 err = vfs_fsync(file, 0);
1004 last_ino = inode->i_ino;
1005 last_dev = inode->i_sb->s_dev;
1006 return err;
1009 static __be32
1010 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1011 loff_t offset, struct kvec *vec, int vlen,
1012 unsigned long *cnt, int *stablep)
1014 struct svc_export *exp;
1015 struct dentry *dentry;
1016 struct inode *inode;
1017 mm_segment_t oldfs;
1018 __be32 err = 0;
1019 int host_err;
1020 int stable = *stablep;
1021 int use_wgather;
1023 #ifdef MSNFS
1024 err = nfserr_perm;
1026 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1027 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1028 goto out;
1029 #endif
1031 dentry = file->f_path.dentry;
1032 inode = dentry->d_inode;
1033 exp = fhp->fh_export;
1036 * Request sync writes if
1037 * - the sync export option has been set, or
1038 * - the client requested O_SYNC behavior (NFSv3 feature).
1039 * - The file system doesn't support fsync().
1040 * When NFSv2 gathered writes have been configured for this volume,
1041 * flushing the data to disk is handled separately below.
1043 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1045 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1046 stable = 2;
1047 *stablep = 2; /* FILE_SYNC */
1050 if (!EX_ISSYNC(exp))
1051 stable = 0;
1052 if (stable && !use_wgather) {
1053 spin_lock(&file->f_lock);
1054 file->f_flags |= O_SYNC;
1055 spin_unlock(&file->f_lock);
1058 /* Write the data. */
1059 oldfs = get_fs(); set_fs(KERNEL_DS);
1060 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1061 set_fs(oldfs);
1062 if (host_err < 0)
1063 goto out_nfserr;
1064 *cnt = host_err;
1065 nfsdstats.io_write += host_err;
1066 fsnotify_modify(file->f_path.dentry);
1068 /* clear setuid/setgid flag after write */
1069 if (inode->i_mode & (S_ISUID | S_ISGID))
1070 kill_suid(dentry);
1072 if (stable && use_wgather)
1073 host_err = wait_for_concurrent_writes(file);
1075 out_nfserr:
1076 dprintk("nfsd: write complete host_err=%d\n", host_err);
1077 if (host_err >= 0)
1078 err = 0;
1079 else
1080 err = nfserrno(host_err);
1081 out:
1082 return err;
1086 * Read data from a file. count must contain the requested read count
1087 * on entry. On return, *count contains the number of bytes actually read.
1088 * N.B. After this call fhp needs an fh_put
1090 __be32
1091 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1092 loff_t offset, struct kvec *vec, int vlen,
1093 unsigned long *count)
1095 __be32 err;
1097 if (file) {
1098 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1099 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1100 if (err)
1101 goto out;
1102 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1103 } else {
1104 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1105 if (err)
1106 goto out;
1107 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1108 nfsd_close(file);
1110 out:
1111 return err;
1115 * Write data to a file.
1116 * The stable flag requests synchronous writes.
1117 * N.B. After this call fhp needs an fh_put
1119 __be32
1120 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1121 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1122 int *stablep)
1124 __be32 err = 0;
1126 if (file) {
1127 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1128 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1129 if (err)
1130 goto out;
1131 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1132 stablep);
1133 } else {
1134 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1135 if (err)
1136 goto out;
1138 if (cnt)
1139 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1140 cnt, stablep);
1141 nfsd_close(file);
1143 out:
1144 return err;
1147 #ifdef CONFIG_NFSD_V3
1149 * Commit all pending writes to stable storage.
1151 * Note: we only guarantee that data that lies within the range specified
1152 * by the 'offset' and 'count' parameters will be synced.
1154 * Unfortunately we cannot lock the file to make sure we return full WCC
1155 * data to the client, as locking happens lower down in the filesystem.
1157 __be32
1158 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1159 loff_t offset, unsigned long count)
1161 struct file *file;
1162 loff_t end = LLONG_MAX;
1163 __be32 err = nfserr_inval;
1165 if (offset < 0)
1166 goto out;
1167 if (count != 0) {
1168 end = offset + (loff_t)count - 1;
1169 if (end < offset)
1170 goto out;
1173 err = nfsd_open(rqstp, fhp, S_IFREG,
1174 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1175 if (err)
1176 goto out;
1177 if (EX_ISSYNC(fhp->fh_export)) {
1178 int err2 = vfs_fsync_range(file, offset, end, 0);
1180 if (err2 != -EINVAL)
1181 err = nfserrno(err2);
1182 else
1183 err = nfserr_notsupp;
1186 nfsd_close(file);
1187 out:
1188 return err;
1190 #endif /* CONFIG_NFSD_V3 */
1192 static __be32
1193 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1194 struct iattr *iap)
1197 * Mode has already been set earlier in create:
1199 iap->ia_valid &= ~ATTR_MODE;
1201 * Setting uid/gid works only for root. Irix appears to
1202 * send along the gid on create when it tries to implement
1203 * setgid directories via NFS:
1205 if (current_fsuid() != 0)
1206 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1207 if (iap->ia_valid)
1208 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1209 return 0;
1212 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1213 * setting size to 0 may fail for some specific file systems by the permission
1214 * checking which requires WRITE permission but the mode is 000.
1215 * we ignore the resizing(to 0) on the just new created file, since the size is
1216 * 0 after file created.
1218 * call this only after vfs_create() is called.
1219 * */
1220 static void
1221 nfsd_check_ignore_resizing(struct iattr *iap)
1223 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1224 iap->ia_valid &= ~ATTR_SIZE;
1228 * Create a file (regular, directory, device, fifo); UNIX sockets
1229 * not yet implemented.
1230 * If the response fh has been verified, the parent directory should
1231 * already be locked. Note that the parent directory is left locked.
1233 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1235 __be32
1236 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1237 char *fname, int flen, struct iattr *iap,
1238 int type, dev_t rdev, struct svc_fh *resfhp)
1240 struct dentry *dentry, *dchild = NULL;
1241 struct inode *dirp;
1242 __be32 err;
1243 __be32 err2;
1244 int host_err;
1246 err = nfserr_perm;
1247 if (!flen)
1248 goto out;
1249 err = nfserr_exist;
1250 if (isdotent(fname, flen))
1251 goto out;
1253 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1254 if (err)
1255 goto out;
1257 dentry = fhp->fh_dentry;
1258 dirp = dentry->d_inode;
1260 err = nfserr_notdir;
1261 if (!dirp->i_op->lookup)
1262 goto out;
1264 * Check whether the response file handle has been verified yet.
1265 * If it has, the parent directory should already be locked.
1267 if (!resfhp->fh_dentry) {
1268 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1269 fh_lock_nested(fhp, I_MUTEX_PARENT);
1270 dchild = lookup_one_len(fname, dentry, flen);
1271 host_err = PTR_ERR(dchild);
1272 if (IS_ERR(dchild))
1273 goto out_nfserr;
1274 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1275 if (err)
1276 goto out;
1277 } else {
1278 /* called from nfsd_proc_create */
1279 dchild = dget(resfhp->fh_dentry);
1280 if (!fhp->fh_locked) {
1281 /* not actually possible */
1282 printk(KERN_ERR
1283 "nfsd_create: parent %s/%s not locked!\n",
1284 dentry->d_parent->d_name.name,
1285 dentry->d_name.name);
1286 err = nfserr_io;
1287 goto out;
1291 * Make sure the child dentry is still negative ...
1293 err = nfserr_exist;
1294 if (dchild->d_inode) {
1295 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1296 dentry->d_name.name, dchild->d_name.name);
1297 goto out;
1300 if (!(iap->ia_valid & ATTR_MODE))
1301 iap->ia_mode = 0;
1302 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1304 err = nfserr_inval;
1305 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1306 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1307 type);
1308 goto out;
1311 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1312 if (host_err)
1313 goto out_nfserr;
1316 * Get the dir op function pointer.
1318 err = 0;
1319 switch (type) {
1320 case S_IFREG:
1321 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1322 if (!host_err)
1323 nfsd_check_ignore_resizing(iap);
1324 break;
1325 case S_IFDIR:
1326 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1327 break;
1328 case S_IFCHR:
1329 case S_IFBLK:
1330 case S_IFIFO:
1331 case S_IFSOCK:
1332 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1333 break;
1335 if (host_err < 0) {
1336 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1337 goto out_nfserr;
1340 err = nfsd_create_setattr(rqstp, resfhp, iap);
1343 * nfsd_setattr already committed the child. Transactional filesystems
1344 * had a chance to commit changes for both parent and child
1345 * simultaneously making the following commit_metadata a noop.
1347 err2 = nfserrno(commit_metadata(fhp));
1348 if (err2)
1349 err = err2;
1350 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1352 * Update the file handle to get the new inode info.
1354 if (!err)
1355 err = fh_update(resfhp);
1356 out:
1357 if (dchild && !IS_ERR(dchild))
1358 dput(dchild);
1359 return err;
1361 out_nfserr:
1362 err = nfserrno(host_err);
1363 goto out;
1366 #ifdef CONFIG_NFSD_V3
1368 * NFSv3 version of nfsd_create
1370 __be32
1371 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1372 char *fname, int flen, struct iattr *iap,
1373 struct svc_fh *resfhp, int createmode, u32 *verifier,
1374 int *truncp, int *created)
1376 struct dentry *dentry, *dchild = NULL;
1377 struct inode *dirp;
1378 __be32 err;
1379 int host_err;
1380 __u32 v_mtime=0, v_atime=0;
1382 err = nfserr_perm;
1383 if (!flen)
1384 goto out;
1385 err = nfserr_exist;
1386 if (isdotent(fname, flen))
1387 goto out;
1388 if (!(iap->ia_valid & ATTR_MODE))
1389 iap->ia_mode = 0;
1390 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1391 if (err)
1392 goto out;
1394 dentry = fhp->fh_dentry;
1395 dirp = dentry->d_inode;
1397 /* Get all the sanity checks out of the way before
1398 * we lock the parent. */
1399 err = nfserr_notdir;
1400 if (!dirp->i_op->lookup)
1401 goto out;
1402 fh_lock_nested(fhp, I_MUTEX_PARENT);
1405 * Compose the response file handle.
1407 dchild = lookup_one_len(fname, dentry, flen);
1408 host_err = PTR_ERR(dchild);
1409 if (IS_ERR(dchild))
1410 goto out_nfserr;
1412 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1413 if (err)
1414 goto out;
1416 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1417 /* solaris7 gets confused (bugid 4218508) if these have
1418 * the high bit set, so just clear the high bits. If this is
1419 * ever changed to use different attrs for storing the
1420 * verifier, then do_open_lookup() will also need to be fixed
1421 * accordingly.
1423 v_mtime = verifier[0]&0x7fffffff;
1424 v_atime = verifier[1]&0x7fffffff;
1427 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1428 if (host_err)
1429 goto out_nfserr;
1430 if (dchild->d_inode) {
1431 err = 0;
1433 switch (createmode) {
1434 case NFS3_CREATE_UNCHECKED:
1435 if (! S_ISREG(dchild->d_inode->i_mode))
1436 err = nfserr_exist;
1437 else if (truncp) {
1438 /* in nfsv4, we need to treat this case a little
1439 * differently. we don't want to truncate the
1440 * file now; this would be wrong if the OPEN
1441 * fails for some other reason. furthermore,
1442 * if the size is nonzero, we should ignore it
1443 * according to spec!
1445 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1447 else {
1448 iap->ia_valid &= ATTR_SIZE;
1449 goto set_attr;
1451 break;
1452 case NFS3_CREATE_EXCLUSIVE:
1453 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1454 && dchild->d_inode->i_atime.tv_sec == v_atime
1455 && dchild->d_inode->i_size == 0 )
1456 break;
1457 /* fallthru */
1458 case NFS3_CREATE_GUARDED:
1459 err = nfserr_exist;
1461 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1462 goto out;
1465 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1466 if (host_err < 0) {
1467 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1468 goto out_nfserr;
1470 if (created)
1471 *created = 1;
1473 nfsd_check_ignore_resizing(iap);
1475 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1476 /* Cram the verifier into atime/mtime */
1477 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1478 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1479 /* XXX someone who knows this better please fix it for nsec */
1480 iap->ia_mtime.tv_sec = v_mtime;
1481 iap->ia_atime.tv_sec = v_atime;
1482 iap->ia_mtime.tv_nsec = 0;
1483 iap->ia_atime.tv_nsec = 0;
1486 set_attr:
1487 err = nfsd_create_setattr(rqstp, resfhp, iap);
1490 * nfsd_setattr already committed the child (and possibly also the parent).
1492 if (!err)
1493 err = nfserrno(commit_metadata(fhp));
1495 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1497 * Update the filehandle to get the new inode info.
1499 if (!err)
1500 err = fh_update(resfhp);
1502 out:
1503 fh_unlock(fhp);
1504 if (dchild && !IS_ERR(dchild))
1505 dput(dchild);
1506 return err;
1508 out_nfserr:
1509 err = nfserrno(host_err);
1510 goto out;
1512 #endif /* CONFIG_NFSD_V3 */
1515 * Read a symlink. On entry, *lenp must contain the maximum path length that
1516 * fits into the buffer. On return, it contains the true length.
1517 * N.B. After this call fhp needs an fh_put
1519 __be32
1520 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1522 struct dentry *dentry;
1523 struct inode *inode;
1524 mm_segment_t oldfs;
1525 __be32 err;
1526 int host_err;
1528 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1529 if (err)
1530 goto out;
1532 dentry = fhp->fh_dentry;
1533 inode = dentry->d_inode;
1535 err = nfserr_inval;
1536 if (!inode->i_op->readlink)
1537 goto out;
1539 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1540 /* N.B. Why does this call need a get_fs()??
1541 * Remove the set_fs and watch the fireworks:-) --okir
1544 oldfs = get_fs(); set_fs(KERNEL_DS);
1545 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1546 set_fs(oldfs);
1548 if (host_err < 0)
1549 goto out_nfserr;
1550 *lenp = host_err;
1551 err = 0;
1552 out:
1553 return err;
1555 out_nfserr:
1556 err = nfserrno(host_err);
1557 goto out;
1561 * Create a symlink and look up its inode
1562 * N.B. After this call _both_ fhp and resfhp need an fh_put
1564 __be32
1565 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1566 char *fname, int flen,
1567 char *path, int plen,
1568 struct svc_fh *resfhp,
1569 struct iattr *iap)
1571 struct dentry *dentry, *dnew;
1572 __be32 err, cerr;
1573 int host_err;
1575 err = nfserr_noent;
1576 if (!flen || !plen)
1577 goto out;
1578 err = nfserr_exist;
1579 if (isdotent(fname, flen))
1580 goto out;
1582 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1583 if (err)
1584 goto out;
1585 fh_lock(fhp);
1586 dentry = fhp->fh_dentry;
1587 dnew = lookup_one_len(fname, dentry, flen);
1588 host_err = PTR_ERR(dnew);
1589 if (IS_ERR(dnew))
1590 goto out_nfserr;
1592 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1593 if (host_err)
1594 goto out_nfserr;
1596 if (unlikely(path[plen] != 0)) {
1597 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1598 if (path_alloced == NULL)
1599 host_err = -ENOMEM;
1600 else {
1601 strncpy(path_alloced, path, plen);
1602 path_alloced[plen] = 0;
1603 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1604 kfree(path_alloced);
1606 } else
1607 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1608 err = nfserrno(host_err);
1609 if (!err)
1610 err = nfserrno(commit_metadata(fhp));
1611 fh_unlock(fhp);
1613 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1615 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1616 dput(dnew);
1617 if (err==0) err = cerr;
1618 out:
1619 return err;
1621 out_nfserr:
1622 err = nfserrno(host_err);
1623 goto out;
1627 * Create a hardlink
1628 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1630 __be32
1631 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1632 char *name, int len, struct svc_fh *tfhp)
1634 struct dentry *ddir, *dnew, *dold;
1635 struct inode *dirp, *dest;
1636 __be32 err;
1637 int host_err;
1639 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1640 if (err)
1641 goto out;
1642 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1643 if (err)
1644 goto out;
1646 err = nfserr_perm;
1647 if (!len)
1648 goto out;
1649 err = nfserr_exist;
1650 if (isdotent(name, len))
1651 goto out;
1653 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1654 ddir = ffhp->fh_dentry;
1655 dirp = ddir->d_inode;
1657 dnew = lookup_one_len(name, ddir, len);
1658 host_err = PTR_ERR(dnew);
1659 if (IS_ERR(dnew))
1660 goto out_nfserr;
1662 dold = tfhp->fh_dentry;
1663 dest = dold->d_inode;
1665 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1666 if (host_err) {
1667 err = nfserrno(host_err);
1668 goto out_dput;
1670 host_err = vfs_link(dold, dirp, dnew);
1671 if (!host_err) {
1672 err = nfserrno(commit_metadata(ffhp));
1673 if (!err)
1674 err = nfserrno(commit_metadata(tfhp));
1675 } else {
1676 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1677 err = nfserr_acces;
1678 else
1679 err = nfserrno(host_err);
1681 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1682 out_dput:
1683 dput(dnew);
1684 out_unlock:
1685 fh_unlock(ffhp);
1686 out:
1687 return err;
1689 out_nfserr:
1690 err = nfserrno(host_err);
1691 goto out_unlock;
1695 * Rename a file
1696 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1698 __be32
1699 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1700 struct svc_fh *tfhp, char *tname, int tlen)
1702 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1703 struct inode *fdir, *tdir;
1704 __be32 err;
1705 int host_err;
1707 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1708 if (err)
1709 goto out;
1710 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1711 if (err)
1712 goto out;
1714 fdentry = ffhp->fh_dentry;
1715 fdir = fdentry->d_inode;
1717 tdentry = tfhp->fh_dentry;
1718 tdir = tdentry->d_inode;
1720 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1721 if (ffhp->fh_export != tfhp->fh_export)
1722 goto out;
1724 err = nfserr_perm;
1725 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1726 goto out;
1728 /* cannot use fh_lock as we need deadlock protective ordering
1729 * so do it by hand */
1730 trap = lock_rename(tdentry, fdentry);
1731 ffhp->fh_locked = tfhp->fh_locked = 1;
1732 fill_pre_wcc(ffhp);
1733 fill_pre_wcc(tfhp);
1735 odentry = lookup_one_len(fname, fdentry, flen);
1736 host_err = PTR_ERR(odentry);
1737 if (IS_ERR(odentry))
1738 goto out_nfserr;
1740 host_err = -ENOENT;
1741 if (!odentry->d_inode)
1742 goto out_dput_old;
1743 host_err = -EINVAL;
1744 if (odentry == trap)
1745 goto out_dput_old;
1747 ndentry = lookup_one_len(tname, tdentry, tlen);
1748 host_err = PTR_ERR(ndentry);
1749 if (IS_ERR(ndentry))
1750 goto out_dput_old;
1751 host_err = -ENOTEMPTY;
1752 if (ndentry == trap)
1753 goto out_dput_new;
1755 if (svc_msnfs(ffhp) &&
1756 ((atomic_read(&odentry->d_count) > 1)
1757 || (atomic_read(&ndentry->d_count) > 1))) {
1758 host_err = -EPERM;
1759 goto out_dput_new;
1762 host_err = -EXDEV;
1763 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1764 goto out_dput_new;
1765 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1766 if (host_err)
1767 goto out_dput_new;
1769 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1770 if (!host_err) {
1771 host_err = commit_metadata(tfhp);
1772 if (!host_err)
1773 host_err = commit_metadata(ffhp);
1776 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1778 out_dput_new:
1779 dput(ndentry);
1780 out_dput_old:
1781 dput(odentry);
1782 out_nfserr:
1783 err = nfserrno(host_err);
1785 /* we cannot reply on fh_unlock on the two filehandles,
1786 * as that would do the wrong thing if the two directories
1787 * were the same, so again we do it by hand
1789 fill_post_wcc(ffhp);
1790 fill_post_wcc(tfhp);
1791 unlock_rename(tdentry, fdentry);
1792 ffhp->fh_locked = tfhp->fh_locked = 0;
1794 out:
1795 return err;
1799 * Unlink a file or directory
1800 * N.B. After this call fhp needs an fh_put
1802 __be32
1803 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1804 char *fname, int flen)
1806 struct dentry *dentry, *rdentry;
1807 struct inode *dirp;
1808 __be32 err;
1809 int host_err;
1811 err = nfserr_acces;
1812 if (!flen || isdotent(fname, flen))
1813 goto out;
1814 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1815 if (err)
1816 goto out;
1818 fh_lock_nested(fhp, I_MUTEX_PARENT);
1819 dentry = fhp->fh_dentry;
1820 dirp = dentry->d_inode;
1822 rdentry = lookup_one_len(fname, dentry, flen);
1823 host_err = PTR_ERR(rdentry);
1824 if (IS_ERR(rdentry))
1825 goto out_nfserr;
1827 if (!rdentry->d_inode) {
1828 dput(rdentry);
1829 err = nfserr_noent;
1830 goto out;
1833 if (!type)
1834 type = rdentry->d_inode->i_mode & S_IFMT;
1836 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1837 if (host_err)
1838 goto out_nfserr;
1840 if (type != S_IFDIR) { /* It's UNLINK */
1841 #ifdef MSNFS
1842 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1843 (atomic_read(&rdentry->d_count) > 1)) {
1844 host_err = -EPERM;
1845 } else
1846 #endif
1847 host_err = vfs_unlink(dirp, rdentry);
1848 } else { /* It's RMDIR */
1849 host_err = vfs_rmdir(dirp, rdentry);
1852 dput(rdentry);
1854 if (!host_err)
1855 host_err = commit_metadata(fhp);
1857 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1858 out_nfserr:
1859 err = nfserrno(host_err);
1860 out:
1861 return err;
1865 * We do this buffering because we must not call back into the file
1866 * system's ->lookup() method from the filldir callback. That may well
1867 * deadlock a number of file systems.
1869 * This is based heavily on the implementation of same in XFS.
1871 struct buffered_dirent {
1872 u64 ino;
1873 loff_t offset;
1874 int namlen;
1875 unsigned int d_type;
1876 char name[];
1879 struct readdir_data {
1880 char *dirent;
1881 size_t used;
1882 int full;
1885 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1886 loff_t offset, u64 ino, unsigned int d_type)
1888 struct readdir_data *buf = __buf;
1889 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1890 unsigned int reclen;
1892 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1893 if (buf->used + reclen > PAGE_SIZE) {
1894 buf->full = 1;
1895 return -EINVAL;
1898 de->namlen = namlen;
1899 de->offset = offset;
1900 de->ino = ino;
1901 de->d_type = d_type;
1902 memcpy(de->name, name, namlen);
1903 buf->used += reclen;
1905 return 0;
1908 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1909 struct readdir_cd *cdp, loff_t *offsetp)
1911 struct readdir_data buf;
1912 struct buffered_dirent *de;
1913 int host_err;
1914 int size;
1915 loff_t offset;
1917 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1918 if (!buf.dirent)
1919 return nfserrno(-ENOMEM);
1921 offset = *offsetp;
1923 while (1) {
1924 struct inode *dir_inode = file->f_path.dentry->d_inode;
1925 unsigned int reclen;
1927 cdp->err = nfserr_eof; /* will be cleared on successful read */
1928 buf.used = 0;
1929 buf.full = 0;
1931 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1932 if (buf.full)
1933 host_err = 0;
1935 if (host_err < 0)
1936 break;
1938 size = buf.used;
1940 if (!size)
1941 break;
1944 * Various filldir functions may end up calling back into
1945 * lookup_one_len() and the file system's ->lookup() method.
1946 * These expect i_mutex to be held, as it would within readdir.
1948 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1949 if (host_err)
1950 break;
1952 de = (struct buffered_dirent *)buf.dirent;
1953 while (size > 0) {
1954 offset = de->offset;
1956 if (func(cdp, de->name, de->namlen, de->offset,
1957 de->ino, de->d_type))
1958 break;
1960 if (cdp->err != nfs_ok)
1961 break;
1963 reclen = ALIGN(sizeof(*de) + de->namlen,
1964 sizeof(u64));
1965 size -= reclen;
1966 de = (struct buffered_dirent *)((char *)de + reclen);
1968 mutex_unlock(&dir_inode->i_mutex);
1969 if (size > 0) /* We bailed out early */
1970 break;
1972 offset = vfs_llseek(file, 0, SEEK_CUR);
1975 free_page((unsigned long)(buf.dirent));
1977 if (host_err)
1978 return nfserrno(host_err);
1980 *offsetp = offset;
1981 return cdp->err;
1985 * Read entries from a directory.
1986 * The NFSv3/4 verifier we ignore for now.
1988 __be32
1989 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1990 struct readdir_cd *cdp, filldir_t func)
1992 __be32 err;
1993 struct file *file;
1994 loff_t offset = *offsetp;
1996 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1997 if (err)
1998 goto out;
2000 offset = vfs_llseek(file, offset, 0);
2001 if (offset < 0) {
2002 err = nfserrno((int)offset);
2003 goto out_close;
2006 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2008 if (err == nfserr_eof || err == nfserr_toosmall)
2009 err = nfs_ok; /* can still be found in ->err */
2010 out_close:
2011 nfsd_close(file);
2012 out:
2013 return err;
2017 * Get file system stats
2018 * N.B. After this call fhp needs an fh_put
2020 __be32
2021 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2023 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2024 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2025 err = nfserr_io;
2026 return err;
2029 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2031 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2035 * Check for a user's access permissions to this inode.
2037 __be32
2038 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2039 struct dentry *dentry, int acc)
2041 struct inode *inode = dentry->d_inode;
2042 struct path path;
2043 int err;
2045 if (acc == NFSD_MAY_NOP)
2046 return 0;
2047 #if 0
2048 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2049 acc,
2050 (acc & NFSD_MAY_READ)? " read" : "",
2051 (acc & NFSD_MAY_WRITE)? " write" : "",
2052 (acc & NFSD_MAY_EXEC)? " exec" : "",
2053 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2054 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2055 (acc & NFSD_MAY_LOCK)? " lock" : "",
2056 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2057 inode->i_mode,
2058 IS_IMMUTABLE(inode)? " immut" : "",
2059 IS_APPEND(inode)? " append" : "",
2060 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2061 dprintk(" owner %d/%d user %d/%d\n",
2062 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2063 #endif
2065 /* Normally we reject any write/sattr etc access on a read-only file
2066 * system. But if it is IRIX doing check on write-access for a
2067 * device special file, we ignore rofs.
2069 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2070 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2071 if (exp_rdonly(rqstp, exp) ||
2072 __mnt_is_readonly(exp->ex_path.mnt))
2073 return nfserr_rofs;
2074 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2075 return nfserr_perm;
2077 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2078 return nfserr_perm;
2080 if (acc & NFSD_MAY_LOCK) {
2081 /* If we cannot rely on authentication in NLM requests,
2082 * just allow locks, otherwise require read permission, or
2083 * ownership
2085 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2086 return 0;
2087 else
2088 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2091 * The file owner always gets access permission for accesses that
2092 * would normally be checked at open time. This is to make
2093 * file access work even when the client has done a fchmod(fd, 0).
2095 * However, `cp foo bar' should fail nevertheless when bar is
2096 * readonly. A sensible way to do this might be to reject all
2097 * attempts to truncate a read-only file, because a creat() call
2098 * always implies file truncation.
2099 * ... but this isn't really fair. A process may reasonably call
2100 * ftruncate on an open file descriptor on a file with perm 000.
2101 * We must trust the client to do permission checking - using "ACCESS"
2102 * with NFSv3.
2104 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2105 inode->i_uid == current_fsuid())
2106 return 0;
2108 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2109 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2111 /* Allow read access to binaries even when mode 111 */
2112 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2113 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2114 err = inode_permission(inode, MAY_EXEC);
2115 if (err)
2116 goto nfsd_out;
2118 /* Do integrity (permission) checking now, but defer incrementing
2119 * IMA counts to the actual file open.
2121 path.mnt = exp->ex_path.mnt;
2122 path.dentry = dentry;
2123 nfsd_out:
2124 return err? nfserrno(err) : 0;
2127 void
2128 nfsd_racache_shutdown(void)
2130 struct raparms *raparm, *last_raparm;
2131 unsigned int i;
2133 dprintk("nfsd: freeing readahead buffers.\n");
2135 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2136 raparm = raparm_hash[i].pb_head;
2137 while(raparm) {
2138 last_raparm = raparm;
2139 raparm = raparm->p_next;
2140 kfree(last_raparm);
2142 raparm_hash[i].pb_head = NULL;
2146 * Initialize readahead param cache
2149 nfsd_racache_init(int cache_size)
2151 int i;
2152 int j = 0;
2153 int nperbucket;
2154 struct raparms **raparm = NULL;
2157 if (raparm_hash[0].pb_head)
2158 return 0;
2159 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2160 if (nperbucket < 2)
2161 nperbucket = 2;
2162 cache_size = nperbucket * RAPARM_HASH_SIZE;
2164 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2166 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2167 spin_lock_init(&raparm_hash[i].pb_lock);
2169 raparm = &raparm_hash[i].pb_head;
2170 for (j = 0; j < nperbucket; j++) {
2171 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2172 if (!*raparm)
2173 goto out_nomem;
2174 raparm = &(*raparm)->p_next;
2176 *raparm = NULL;
2179 nfsdstats.ra_size = cache_size;
2180 return 0;
2182 out_nomem:
2183 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2184 nfsd_racache_shutdown();
2185 return -ENOMEM;
2188 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2189 struct posix_acl *
2190 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2192 struct inode *inode = fhp->fh_dentry->d_inode;
2193 char *name;
2194 void *value = NULL;
2195 ssize_t size;
2196 struct posix_acl *acl;
2198 if (!IS_POSIXACL(inode))
2199 return ERR_PTR(-EOPNOTSUPP);
2201 switch (type) {
2202 case ACL_TYPE_ACCESS:
2203 name = POSIX_ACL_XATTR_ACCESS;
2204 break;
2205 case ACL_TYPE_DEFAULT:
2206 name = POSIX_ACL_XATTR_DEFAULT;
2207 break;
2208 default:
2209 return ERR_PTR(-EOPNOTSUPP);
2212 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2213 if (size < 0)
2214 return ERR_PTR(size);
2216 acl = posix_acl_from_xattr(value, size);
2217 kfree(value);
2218 return acl;
2222 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2224 struct inode *inode = fhp->fh_dentry->d_inode;
2225 char *name;
2226 void *value = NULL;
2227 size_t size;
2228 int error;
2230 if (!IS_POSIXACL(inode) ||
2231 !inode->i_op->setxattr || !inode->i_op->removexattr)
2232 return -EOPNOTSUPP;
2233 switch(type) {
2234 case ACL_TYPE_ACCESS:
2235 name = POSIX_ACL_XATTR_ACCESS;
2236 break;
2237 case ACL_TYPE_DEFAULT:
2238 name = POSIX_ACL_XATTR_DEFAULT;
2239 break;
2240 default:
2241 return -EOPNOTSUPP;
2244 if (acl && acl->a_count) {
2245 size = posix_acl_xattr_size(acl->a_count);
2246 value = kmalloc(size, GFP_KERNEL);
2247 if (!value)
2248 return -ENOMEM;
2249 error = posix_acl_to_xattr(acl, value, size);
2250 if (error < 0)
2251 goto getout;
2252 size = error;
2253 } else
2254 size = 0;
2256 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2257 if (error)
2258 goto getout;
2259 if (size)
2260 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2261 else {
2262 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2263 error = 0;
2264 else {
2265 error = vfs_removexattr(fhp->fh_dentry, name);
2266 if (error == -ENODATA)
2267 error = 0;
2270 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2272 getout:
2273 kfree(value);
2274 return error;
2276 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */