Blackfin arch: include linux/mm.h since we use PAGE_ALIGN and such
[linux-2.6/mini2440.git] / fs / nfsd / vfs.c
blob44aa92aba891c96fa7061ee5dda5bc6aaf6f0fec
1 #define MSNFS /* HACK HACK */
2 /*
3 * linux/fs/nfsd/vfs.c
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
19 #include <linux/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/fs.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/splice.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
34 #include <linux/in.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
41 #ifdef CONFIG_NFSD_V3
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
51 #ifdef CONFIG_NFSD_V4
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
57 #include <linux/jhash.h>
59 #include <asm/uaccess.h>
61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
65 * This is a cache of readahead params that help us choose the proper
66 * readahead strategy. Initially, we set all readahead parameters to 0
67 * and let the VFS handle things.
68 * If you increase the number of cached files very much, you'll need to
69 * add a hash table here.
71 struct raparms {
72 struct raparms *p_next;
73 unsigned int p_count;
74 ino_t p_ino;
75 dev_t p_dev;
76 int p_set;
77 struct file_ra_state p_ra;
78 unsigned int p_hindex;
81 struct raparm_hbucket {
82 struct raparms *pb_head;
83 spinlock_t pb_lock;
84 } ____cacheline_aligned_in_smp;
86 #define RAPARM_HASH_BITS 4
87 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
88 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
89 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
91 /*
92 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
93 * a mount point.
94 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
95 * or nfs_ok having possibly changed *dpp and *expp
97 int
98 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
99 struct svc_export **expp)
101 struct svc_export *exp = *expp, *exp2 = NULL;
102 struct dentry *dentry = *dpp;
103 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
104 struct dentry *mounts = dget(dentry);
105 int err = 0;
107 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
109 exp2 = rqst_exp_get_by_name(rqstp, mnt, mounts);
110 if (IS_ERR(exp2)) {
111 if (PTR_ERR(exp2) != -ENOENT)
112 err = PTR_ERR(exp2);
113 dput(mounts);
114 mntput(mnt);
115 goto out;
117 if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
118 /* successfully crossed mount point */
119 exp_put(exp);
120 *expp = exp2;
121 dput(dentry);
122 *dpp = mounts;
123 } else {
124 exp_put(exp2);
125 dput(mounts);
127 mntput(mnt);
128 out:
129 return err;
132 __be32
133 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
134 const char *name, unsigned int len,
135 struct svc_export **exp_ret, struct dentry **dentry_ret)
137 struct svc_export *exp;
138 struct dentry *dparent;
139 struct dentry *dentry;
140 __be32 err;
141 int host_err;
143 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
145 /* Obtain dentry and export. */
146 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
147 if (err)
148 return err;
150 dparent = fhp->fh_dentry;
151 exp = fhp->fh_export;
152 exp_get(exp);
154 /* Lookup the name, but don't follow links */
155 if (isdotent(name, len)) {
156 if (len==1)
157 dentry = dget(dparent);
158 else if (dparent != exp->ex_path.dentry)
159 dentry = dget_parent(dparent);
160 else if (!EX_NOHIDE(exp))
161 dentry = dget(dparent); /* .. == . just like at / */
162 else {
163 /* checking mountpoint crossing is very different when stepping up */
164 struct svc_export *exp2 = NULL;
165 struct dentry *dp;
166 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
167 dentry = dget(dparent);
168 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
170 dp = dget_parent(dentry);
171 dput(dentry);
172 dentry = dp;
174 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
175 if (PTR_ERR(exp2) == -ENOENT) {
176 dput(dentry);
177 dentry = dget(dparent);
178 } else if (IS_ERR(exp2)) {
179 host_err = PTR_ERR(exp2);
180 dput(dentry);
181 mntput(mnt);
182 goto out_nfserr;
183 } else {
184 exp_put(exp);
185 exp = exp2;
187 mntput(mnt);
189 } else {
190 fh_lock(fhp);
191 dentry = lookup_one_len(name, dparent, len);
192 host_err = PTR_ERR(dentry);
193 if (IS_ERR(dentry))
194 goto out_nfserr;
196 * check if we have crossed a mount point ...
198 if (d_mountpoint(dentry)) {
199 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
200 dput(dentry);
201 goto out_nfserr;
205 *dentry_ret = dentry;
206 *exp_ret = exp;
207 return 0;
209 out_nfserr:
210 exp_put(exp);
211 return nfserrno(host_err);
215 * Look up one component of a pathname.
216 * N.B. After this call _both_ fhp and resfh need an fh_put
218 * If the lookup would cross a mountpoint, and the mounted filesystem
219 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
220 * accepted as it stands and the mounted directory is
221 * returned. Otherwise the covered directory is returned.
222 * NOTE: this mountpoint crossing is not supported properly by all
223 * clients and is explicitly disallowed for NFSv3
224 * NeilBrown <neilb@cse.unsw.edu.au>
226 __be32
227 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
228 unsigned int len, struct svc_fh *resfh)
230 struct svc_export *exp;
231 struct dentry *dentry;
232 __be32 err;
234 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
235 if (err)
236 return err;
237 err = check_nfsd_access(exp, rqstp);
238 if (err)
239 goto out;
241 * Note: we compose the file handle now, but as the
242 * dentry may be negative, it may need to be updated.
244 err = fh_compose(resfh, exp, dentry, fhp);
245 if (!err && !dentry->d_inode)
246 err = nfserr_noent;
247 out:
248 dput(dentry);
249 exp_put(exp);
250 return err;
255 * Set various file attributes.
256 * N.B. After this call fhp needs an fh_put
258 __be32
259 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
260 int check_guard, time_t guardtime)
262 struct dentry *dentry;
263 struct inode *inode;
264 int accmode = NFSD_MAY_SATTR;
265 int ftype = 0;
266 __be32 err;
267 int host_err;
268 int size_change = 0;
270 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
271 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
272 if (iap->ia_valid & ATTR_SIZE)
273 ftype = S_IFREG;
275 /* Get inode */
276 err = fh_verify(rqstp, fhp, ftype, accmode);
277 if (err)
278 goto out;
280 dentry = fhp->fh_dentry;
281 inode = dentry->d_inode;
283 /* Ignore any mode updates on symlinks */
284 if (S_ISLNK(inode->i_mode))
285 iap->ia_valid &= ~ATTR_MODE;
287 if (!iap->ia_valid)
288 goto out;
291 * NFSv2 does not differentiate between "set-[ac]time-to-now"
292 * which only requires access, and "set-[ac]time-to-X" which
293 * requires ownership.
294 * So if it looks like it might be "set both to the same time which
295 * is close to now", and if inode_change_ok fails, then we
296 * convert to "set to now" instead of "set to explicit time"
298 * We only call inode_change_ok as the last test as technically
299 * it is not an interface that we should be using. It is only
300 * valid if the filesystem does not define it's own i_op->setattr.
302 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
303 #define MAX_TOUCH_TIME_ERROR (30*60)
304 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
305 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
307 * Looks probable.
309 * Now just make sure time is in the right ballpark.
310 * Solaris, at least, doesn't seem to care what the time
311 * request is. We require it be within 30 minutes of now.
313 time_t delta = iap->ia_atime.tv_sec - get_seconds();
314 if (delta < 0)
315 delta = -delta;
316 if (delta < MAX_TOUCH_TIME_ERROR &&
317 inode_change_ok(inode, iap) != 0) {
319 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
320 * This will cause notify_change to set these times
321 * to "now"
323 iap->ia_valid &= ~BOTH_TIME_SET;
328 * The size case is special.
329 * It changes the file as well as the attributes.
331 if (iap->ia_valid & ATTR_SIZE) {
332 if (iap->ia_size < inode->i_size) {
333 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
334 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
335 if (err)
336 goto out;
340 * If we are changing the size of the file, then
341 * we need to break all leases.
343 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
344 if (host_err == -EWOULDBLOCK)
345 host_err = -ETIMEDOUT;
346 if (host_err) /* ENOMEM or EWOULDBLOCK */
347 goto out_nfserr;
349 host_err = get_write_access(inode);
350 if (host_err)
351 goto out_nfserr;
353 size_change = 1;
354 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
355 if (host_err) {
356 put_write_access(inode);
357 goto out_nfserr;
359 DQUOT_INIT(inode);
362 /* sanitize the mode change */
363 if (iap->ia_valid & ATTR_MODE) {
364 iap->ia_mode &= S_IALLUGO;
365 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
368 /* Revoke setuid/setgid on chown */
369 if (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
370 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)) {
371 iap->ia_valid |= ATTR_KILL_PRIV;
372 if (iap->ia_valid & ATTR_MODE) {
373 /* we're setting mode too, just clear the s*id bits */
374 iap->ia_mode &= ~S_ISUID;
375 if (iap->ia_mode & S_IXGRP)
376 iap->ia_mode &= ~S_ISGID;
377 } else {
378 /* set ATTR_KILL_* bits and let VFS handle it */
379 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
383 /* Change the attributes. */
385 iap->ia_valid |= ATTR_CTIME;
387 err = nfserr_notsync;
388 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
389 fh_lock(fhp);
390 host_err = notify_change(dentry, iap);
391 err = nfserrno(host_err);
392 fh_unlock(fhp);
394 if (size_change)
395 put_write_access(inode);
396 if (!err)
397 if (EX_ISSYNC(fhp->fh_export))
398 write_inode_now(inode, 1);
399 out:
400 return err;
402 out_nfserr:
403 err = nfserrno(host_err);
404 goto out;
407 #if defined(CONFIG_NFSD_V2_ACL) || \
408 defined(CONFIG_NFSD_V3_ACL) || \
409 defined(CONFIG_NFSD_V4)
410 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
412 ssize_t buflen;
413 ssize_t ret;
415 buflen = vfs_getxattr(dentry, key, NULL, 0);
416 if (buflen <= 0)
417 return buflen;
419 *buf = kmalloc(buflen, GFP_KERNEL);
420 if (!*buf)
421 return -ENOMEM;
423 ret = vfs_getxattr(dentry, key, *buf, buflen);
424 if (ret < 0)
425 kfree(*buf);
426 return ret;
428 #endif
430 #if defined(CONFIG_NFSD_V4)
431 static int
432 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
434 int len;
435 size_t buflen;
436 char *buf = NULL;
437 int error = 0;
439 buflen = posix_acl_xattr_size(pacl->a_count);
440 buf = kmalloc(buflen, GFP_KERNEL);
441 error = -ENOMEM;
442 if (buf == NULL)
443 goto out;
445 len = posix_acl_to_xattr(pacl, buf, buflen);
446 if (len < 0) {
447 error = len;
448 goto out;
451 error = vfs_setxattr(dentry, key, buf, len, 0);
452 out:
453 kfree(buf);
454 return error;
457 __be32
458 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
459 struct nfs4_acl *acl)
461 __be32 error;
462 int host_error;
463 struct dentry *dentry;
464 struct inode *inode;
465 struct posix_acl *pacl = NULL, *dpacl = NULL;
466 unsigned int flags = 0;
468 /* Get inode */
469 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
470 if (error)
471 return error;
473 dentry = fhp->fh_dentry;
474 inode = dentry->d_inode;
475 if (S_ISDIR(inode->i_mode))
476 flags = NFS4_ACL_DIR;
478 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
479 if (host_error == -EINVAL) {
480 return nfserr_attrnotsupp;
481 } else if (host_error < 0)
482 goto out_nfserr;
484 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
485 if (host_error < 0)
486 goto out_release;
488 if (S_ISDIR(inode->i_mode))
489 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
491 out_release:
492 posix_acl_release(pacl);
493 posix_acl_release(dpacl);
494 out_nfserr:
495 if (host_error == -EOPNOTSUPP)
496 return nfserr_attrnotsupp;
497 else
498 return nfserrno(host_error);
501 static struct posix_acl *
502 _get_posix_acl(struct dentry *dentry, char *key)
504 void *buf = NULL;
505 struct posix_acl *pacl = NULL;
506 int buflen;
508 buflen = nfsd_getxattr(dentry, key, &buf);
509 if (!buflen)
510 buflen = -ENODATA;
511 if (buflen <= 0)
512 return ERR_PTR(buflen);
514 pacl = posix_acl_from_xattr(buf, buflen);
515 kfree(buf);
516 return pacl;
520 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
522 struct inode *inode = dentry->d_inode;
523 int error = 0;
524 struct posix_acl *pacl = NULL, *dpacl = NULL;
525 unsigned int flags = 0;
527 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
528 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
529 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
530 if (IS_ERR(pacl)) {
531 error = PTR_ERR(pacl);
532 pacl = NULL;
533 goto out;
536 if (S_ISDIR(inode->i_mode)) {
537 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
538 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
539 dpacl = NULL;
540 else if (IS_ERR(dpacl)) {
541 error = PTR_ERR(dpacl);
542 dpacl = NULL;
543 goto out;
545 flags = NFS4_ACL_DIR;
548 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
549 if (IS_ERR(*acl)) {
550 error = PTR_ERR(*acl);
551 *acl = NULL;
553 out:
554 posix_acl_release(pacl);
555 posix_acl_release(dpacl);
556 return error;
559 #endif /* defined(CONFIG_NFS_V4) */
561 #ifdef CONFIG_NFSD_V3
563 * Check server access rights to a file system object
565 struct accessmap {
566 u32 access;
567 int how;
569 static struct accessmap nfs3_regaccess[] = {
570 { NFS3_ACCESS_READ, NFSD_MAY_READ },
571 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
572 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
573 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
575 { 0, 0 }
578 static struct accessmap nfs3_diraccess[] = {
579 { NFS3_ACCESS_READ, NFSD_MAY_READ },
580 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
581 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
582 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
583 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
585 { 0, 0 }
588 static struct accessmap nfs3_anyaccess[] = {
589 /* Some clients - Solaris 2.6 at least, make an access call
590 * to the server to check for access for things like /dev/null
591 * (which really, the server doesn't care about). So
592 * We provide simple access checking for them, looking
593 * mainly at mode bits, and we make sure to ignore read-only
594 * filesystem checks
596 { NFS3_ACCESS_READ, NFSD_MAY_READ },
597 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
598 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
599 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
601 { 0, 0 }
604 __be32
605 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
607 struct accessmap *map;
608 struct svc_export *export;
609 struct dentry *dentry;
610 u32 query, result = 0, sresult = 0;
611 __be32 error;
613 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
614 if (error)
615 goto out;
617 export = fhp->fh_export;
618 dentry = fhp->fh_dentry;
620 if (S_ISREG(dentry->d_inode->i_mode))
621 map = nfs3_regaccess;
622 else if (S_ISDIR(dentry->d_inode->i_mode))
623 map = nfs3_diraccess;
624 else
625 map = nfs3_anyaccess;
628 query = *access;
629 for (; map->access; map++) {
630 if (map->access & query) {
631 __be32 err2;
633 sresult |= map->access;
635 err2 = nfsd_permission(rqstp, export, dentry, map->how);
636 switch (err2) {
637 case nfs_ok:
638 result |= map->access;
639 break;
641 /* the following error codes just mean the access was not allowed,
642 * rather than an error occurred */
643 case nfserr_rofs:
644 case nfserr_acces:
645 case nfserr_perm:
646 /* simply don't "or" in the access bit. */
647 break;
648 default:
649 error = err2;
650 goto out;
654 *access = result;
655 if (supported)
656 *supported = sresult;
658 out:
659 return error;
661 #endif /* CONFIG_NFSD_V3 */
666 * Open an existing file or directory.
667 * The access argument indicates the type of open (read/write/lock)
668 * N.B. After this call fhp needs an fh_put
670 __be32
671 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
672 int access, struct file **filp)
674 const struct cred *cred = current_cred();
675 struct dentry *dentry;
676 struct inode *inode;
677 int flags = O_RDONLY|O_LARGEFILE;
678 __be32 err;
679 int host_err;
682 * If we get here, then the client has already done an "open",
683 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
684 * in case a chmod has now revoked permission.
686 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
687 if (err)
688 goto out;
690 dentry = fhp->fh_dentry;
691 inode = dentry->d_inode;
693 /* Disallow write access to files with the append-only bit set
694 * or any access when mandatory locking enabled
696 err = nfserr_perm;
697 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
698 goto out;
700 * We must ignore files (but only files) which might have mandatory
701 * locks on them because there is no way to know if the accesser has
702 * the lock.
704 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
705 goto out;
707 if (!inode->i_fop)
708 goto out;
711 * Check to see if there are any leases on this file.
712 * This may block while leases are broken.
714 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
715 if (host_err == -EWOULDBLOCK)
716 host_err = -ETIMEDOUT;
717 if (host_err) /* NOMEM or WOULDBLOCK */
718 goto out_nfserr;
720 if (access & NFSD_MAY_WRITE) {
721 if (access & NFSD_MAY_READ)
722 flags = O_RDWR|O_LARGEFILE;
723 else
724 flags = O_WRONLY|O_LARGEFILE;
726 DQUOT_INIT(inode);
728 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
729 flags, cred);
730 if (IS_ERR(*filp))
731 host_err = PTR_ERR(*filp);
732 out_nfserr:
733 err = nfserrno(host_err);
734 out:
735 return err;
739 * Close a file.
741 void
742 nfsd_close(struct file *filp)
744 fput(filp);
747 static int
748 nfsd_sync(struct file *filp)
750 return vfs_fsync(filp, filp->f_path.dentry, 0);
754 nfsd_sync_dir(struct dentry *dentry)
756 return vfs_fsync(NULL, dentry, 0);
760 * Obtain the readahead parameters for the file
761 * specified by (dev, ino).
764 static inline struct raparms *
765 nfsd_get_raparms(dev_t dev, ino_t ino)
767 struct raparms *ra, **rap, **frap = NULL;
768 int depth = 0;
769 unsigned int hash;
770 struct raparm_hbucket *rab;
772 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
773 rab = &raparm_hash[hash];
775 spin_lock(&rab->pb_lock);
776 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
777 if (ra->p_ino == ino && ra->p_dev == dev)
778 goto found;
779 depth++;
780 if (ra->p_count == 0)
781 frap = rap;
783 depth = nfsdstats.ra_size*11/10;
784 if (!frap) {
785 spin_unlock(&rab->pb_lock);
786 return NULL;
788 rap = frap;
789 ra = *frap;
790 ra->p_dev = dev;
791 ra->p_ino = ino;
792 ra->p_set = 0;
793 ra->p_hindex = hash;
794 found:
795 if (rap != &rab->pb_head) {
796 *rap = ra->p_next;
797 ra->p_next = rab->pb_head;
798 rab->pb_head = ra;
800 ra->p_count++;
801 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
802 spin_unlock(&rab->pb_lock);
803 return ra;
807 * Grab and keep cached pages associated with a file in the svc_rqst
808 * so that they can be passed to the network sendmsg/sendpage routines
809 * directly. They will be released after the sending has completed.
811 static int
812 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
813 struct splice_desc *sd)
815 struct svc_rqst *rqstp = sd->u.data;
816 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
817 struct page *page = buf->page;
818 size_t size;
819 int ret;
821 ret = buf->ops->confirm(pipe, buf);
822 if (unlikely(ret))
823 return ret;
825 size = sd->len;
827 if (rqstp->rq_res.page_len == 0) {
828 get_page(page);
829 put_page(*pp);
830 *pp = page;
831 rqstp->rq_resused++;
832 rqstp->rq_res.page_base = buf->offset;
833 rqstp->rq_res.page_len = size;
834 } else if (page != pp[-1]) {
835 get_page(page);
836 if (*pp)
837 put_page(*pp);
838 *pp = page;
839 rqstp->rq_resused++;
840 rqstp->rq_res.page_len += size;
841 } else
842 rqstp->rq_res.page_len += size;
844 return size;
847 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
848 struct splice_desc *sd)
850 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
853 static inline int svc_msnfs(struct svc_fh *ffhp)
855 #ifdef MSNFS
856 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
857 #else
858 return 0;
859 #endif
862 static __be32
863 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
864 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
866 struct inode *inode;
867 struct raparms *ra;
868 mm_segment_t oldfs;
869 __be32 err;
870 int host_err;
872 err = nfserr_perm;
873 inode = file->f_path.dentry->d_inode;
875 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
876 goto out;
878 /* Get readahead parameters */
879 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
881 if (ra && ra->p_set)
882 file->f_ra = ra->p_ra;
884 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
885 struct splice_desc sd = {
886 .len = 0,
887 .total_len = *count,
888 .pos = offset,
889 .u.data = rqstp,
892 rqstp->rq_resused = 1;
893 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
894 } else {
895 oldfs = get_fs();
896 set_fs(KERNEL_DS);
897 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
898 set_fs(oldfs);
901 /* Write back readahead params */
902 if (ra) {
903 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
904 spin_lock(&rab->pb_lock);
905 ra->p_ra = file->f_ra;
906 ra->p_set = 1;
907 ra->p_count--;
908 spin_unlock(&rab->pb_lock);
911 if (host_err >= 0) {
912 nfsdstats.io_read += host_err;
913 *count = host_err;
914 err = 0;
915 fsnotify_access(file->f_path.dentry);
916 } else
917 err = nfserrno(host_err);
918 out:
919 return err;
922 static void kill_suid(struct dentry *dentry)
924 struct iattr ia;
925 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
927 mutex_lock(&dentry->d_inode->i_mutex);
928 notify_change(dentry, &ia);
929 mutex_unlock(&dentry->d_inode->i_mutex);
932 static __be32
933 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
934 loff_t offset, struct kvec *vec, int vlen,
935 unsigned long cnt, int *stablep)
937 struct svc_export *exp;
938 struct dentry *dentry;
939 struct inode *inode;
940 mm_segment_t oldfs;
941 __be32 err = 0;
942 int host_err;
943 int stable = *stablep;
945 #ifdef MSNFS
946 err = nfserr_perm;
948 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
949 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
950 goto out;
951 #endif
953 dentry = file->f_path.dentry;
954 inode = dentry->d_inode;
955 exp = fhp->fh_export;
958 * Request sync writes if
959 * - the sync export option has been set, or
960 * - the client requested O_SYNC behavior (NFSv3 feature).
961 * - The file system doesn't support fsync().
962 * When gathered writes have been configured for this volume,
963 * flushing the data to disk is handled separately below.
966 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
967 stable = 2;
968 *stablep = 2; /* FILE_SYNC */
971 if (!EX_ISSYNC(exp))
972 stable = 0;
973 if (stable && !EX_WGATHER(exp))
974 file->f_flags |= O_SYNC;
976 /* Write the data. */
977 oldfs = get_fs(); set_fs(KERNEL_DS);
978 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
979 set_fs(oldfs);
980 if (host_err >= 0) {
981 nfsdstats.io_write += cnt;
982 fsnotify_modify(file->f_path.dentry);
985 /* clear setuid/setgid flag after write */
986 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
987 kill_suid(dentry);
989 if (host_err >= 0 && stable) {
990 static ino_t last_ino;
991 static dev_t last_dev;
994 * Gathered writes: If another process is currently
995 * writing to the file, there's a high chance
996 * this is another nfsd (triggered by a bulk write
997 * from a client's biod). Rather than syncing the
998 * file with each write request, we sleep for 10 msec.
1000 * I don't know if this roughly approximates
1001 * C. Juszak's idea of gathered writes, but it's a
1002 * nice and simple solution (IMHO), and it seems to
1003 * work:-)
1005 if (EX_WGATHER(exp)) {
1006 if (atomic_read(&inode->i_writecount) > 1
1007 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1008 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1009 msleep(10);
1010 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1013 if (inode->i_state & I_DIRTY) {
1014 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1015 host_err=nfsd_sync(file);
1017 #if 0
1018 wake_up(&inode->i_wait);
1019 #endif
1021 last_ino = inode->i_ino;
1022 last_dev = inode->i_sb->s_dev;
1025 dprintk("nfsd: write complete host_err=%d\n", host_err);
1026 if (host_err >= 0)
1027 err = 0;
1028 else
1029 err = nfserrno(host_err);
1030 out:
1031 return err;
1035 * Read data from a file. count must contain the requested read count
1036 * on entry. On return, *count contains the number of bytes actually read.
1037 * N.B. After this call fhp needs an fh_put
1039 __be32
1040 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1041 loff_t offset, struct kvec *vec, int vlen,
1042 unsigned long *count)
1044 __be32 err;
1046 if (file) {
1047 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1048 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1049 if (err)
1050 goto out;
1051 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1052 } else {
1053 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1054 if (err)
1055 goto out;
1056 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1057 nfsd_close(file);
1059 out:
1060 return err;
1064 * Write data to a file.
1065 * The stable flag requests synchronous writes.
1066 * N.B. After this call fhp needs an fh_put
1068 __be32
1069 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1070 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1071 int *stablep)
1073 __be32 err = 0;
1075 if (file) {
1076 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1077 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1078 if (err)
1079 goto out;
1080 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1081 stablep);
1082 } else {
1083 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1084 if (err)
1085 goto out;
1087 if (cnt)
1088 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1089 cnt, stablep);
1090 nfsd_close(file);
1092 out:
1093 return err;
1096 #ifdef CONFIG_NFSD_V3
1098 * Commit all pending writes to stable storage.
1099 * Strictly speaking, we could sync just the indicated file region here,
1100 * but there's currently no way we can ask the VFS to do so.
1102 * Unfortunately we cannot lock the file to make sure we return full WCC
1103 * data to the client, as locking happens lower down in the filesystem.
1105 __be32
1106 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1107 loff_t offset, unsigned long count)
1109 struct file *file;
1110 __be32 err;
1112 if ((u64)count > ~(u64)offset)
1113 return nfserr_inval;
1115 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1116 if (err)
1117 return err;
1118 if (EX_ISSYNC(fhp->fh_export)) {
1119 if (file->f_op && file->f_op->fsync) {
1120 err = nfserrno(nfsd_sync(file));
1121 } else {
1122 err = nfserr_notsupp;
1126 nfsd_close(file);
1127 return err;
1129 #endif /* CONFIG_NFSD_V3 */
1131 static __be32
1132 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1133 struct iattr *iap)
1136 * Mode has already been set earlier in create:
1138 iap->ia_valid &= ~ATTR_MODE;
1140 * Setting uid/gid works only for root. Irix appears to
1141 * send along the gid on create when it tries to implement
1142 * setgid directories via NFS:
1144 if (current_fsuid() != 0)
1145 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1146 if (iap->ia_valid)
1147 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1148 return 0;
1152 * Create a file (regular, directory, device, fifo); UNIX sockets
1153 * not yet implemented.
1154 * If the response fh has been verified, the parent directory should
1155 * already be locked. Note that the parent directory is left locked.
1157 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1159 __be32
1160 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1161 char *fname, int flen, struct iattr *iap,
1162 int type, dev_t rdev, struct svc_fh *resfhp)
1164 struct dentry *dentry, *dchild = NULL;
1165 struct inode *dirp;
1166 __be32 err;
1167 __be32 err2;
1168 int host_err;
1170 err = nfserr_perm;
1171 if (!flen)
1172 goto out;
1173 err = nfserr_exist;
1174 if (isdotent(fname, flen))
1175 goto out;
1177 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1178 if (err)
1179 goto out;
1181 dentry = fhp->fh_dentry;
1182 dirp = dentry->d_inode;
1184 err = nfserr_notdir;
1185 if (!dirp->i_op->lookup)
1186 goto out;
1188 * Check whether the response file handle has been verified yet.
1189 * If it has, the parent directory should already be locked.
1191 if (!resfhp->fh_dentry) {
1192 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1193 fh_lock_nested(fhp, I_MUTEX_PARENT);
1194 dchild = lookup_one_len(fname, dentry, flen);
1195 host_err = PTR_ERR(dchild);
1196 if (IS_ERR(dchild))
1197 goto out_nfserr;
1198 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1199 if (err)
1200 goto out;
1201 } else {
1202 /* called from nfsd_proc_create */
1203 dchild = dget(resfhp->fh_dentry);
1204 if (!fhp->fh_locked) {
1205 /* not actually possible */
1206 printk(KERN_ERR
1207 "nfsd_create: parent %s/%s not locked!\n",
1208 dentry->d_parent->d_name.name,
1209 dentry->d_name.name);
1210 err = nfserr_io;
1211 goto out;
1215 * Make sure the child dentry is still negative ...
1217 err = nfserr_exist;
1218 if (dchild->d_inode) {
1219 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1220 dentry->d_name.name, dchild->d_name.name);
1221 goto out;
1224 if (!(iap->ia_valid & ATTR_MODE))
1225 iap->ia_mode = 0;
1226 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1228 err = nfserr_inval;
1229 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1230 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1231 type);
1232 goto out;
1235 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1236 if (host_err)
1237 goto out_nfserr;
1240 * Get the dir op function pointer.
1242 err = 0;
1243 switch (type) {
1244 case S_IFREG:
1245 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1246 break;
1247 case S_IFDIR:
1248 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1249 break;
1250 case S_IFCHR:
1251 case S_IFBLK:
1252 case S_IFIFO:
1253 case S_IFSOCK:
1254 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1255 break;
1257 if (host_err < 0) {
1258 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1259 goto out_nfserr;
1262 if (EX_ISSYNC(fhp->fh_export)) {
1263 err = nfserrno(nfsd_sync_dir(dentry));
1264 write_inode_now(dchild->d_inode, 1);
1267 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1268 if (err2)
1269 err = err2;
1270 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1272 * Update the file handle to get the new inode info.
1274 if (!err)
1275 err = fh_update(resfhp);
1276 out:
1277 if (dchild && !IS_ERR(dchild))
1278 dput(dchild);
1279 return err;
1281 out_nfserr:
1282 err = nfserrno(host_err);
1283 goto out;
1286 #ifdef CONFIG_NFSD_V3
1288 * NFSv3 version of nfsd_create
1290 __be32
1291 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1292 char *fname, int flen, struct iattr *iap,
1293 struct svc_fh *resfhp, int createmode, u32 *verifier,
1294 int *truncp, int *created)
1296 struct dentry *dentry, *dchild = NULL;
1297 struct inode *dirp;
1298 __be32 err;
1299 __be32 err2;
1300 int host_err;
1301 __u32 v_mtime=0, v_atime=0;
1303 err = nfserr_perm;
1304 if (!flen)
1305 goto out;
1306 err = nfserr_exist;
1307 if (isdotent(fname, flen))
1308 goto out;
1309 if (!(iap->ia_valid & ATTR_MODE))
1310 iap->ia_mode = 0;
1311 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1312 if (err)
1313 goto out;
1315 dentry = fhp->fh_dentry;
1316 dirp = dentry->d_inode;
1318 /* Get all the sanity checks out of the way before
1319 * we lock the parent. */
1320 err = nfserr_notdir;
1321 if (!dirp->i_op->lookup)
1322 goto out;
1323 fh_lock_nested(fhp, I_MUTEX_PARENT);
1326 * Compose the response file handle.
1328 dchild = lookup_one_len(fname, dentry, flen);
1329 host_err = PTR_ERR(dchild);
1330 if (IS_ERR(dchild))
1331 goto out_nfserr;
1333 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1334 if (err)
1335 goto out;
1337 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1338 /* solaris7 gets confused (bugid 4218508) if these have
1339 * the high bit set, so just clear the high bits. If this is
1340 * ever changed to use different attrs for storing the
1341 * verifier, then do_open_lookup() will also need to be fixed
1342 * accordingly.
1344 v_mtime = verifier[0]&0x7fffffff;
1345 v_atime = verifier[1]&0x7fffffff;
1348 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1349 if (host_err)
1350 goto out_nfserr;
1351 if (dchild->d_inode) {
1352 err = 0;
1354 switch (createmode) {
1355 case NFS3_CREATE_UNCHECKED:
1356 if (! S_ISREG(dchild->d_inode->i_mode))
1357 err = nfserr_exist;
1358 else if (truncp) {
1359 /* in nfsv4, we need to treat this case a little
1360 * differently. we don't want to truncate the
1361 * file now; this would be wrong if the OPEN
1362 * fails for some other reason. furthermore,
1363 * if the size is nonzero, we should ignore it
1364 * according to spec!
1366 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1368 else {
1369 iap->ia_valid &= ATTR_SIZE;
1370 goto set_attr;
1372 break;
1373 case NFS3_CREATE_EXCLUSIVE:
1374 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1375 && dchild->d_inode->i_atime.tv_sec == v_atime
1376 && dchild->d_inode->i_size == 0 )
1377 break;
1378 /* fallthru */
1379 case NFS3_CREATE_GUARDED:
1380 err = nfserr_exist;
1382 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1383 goto out;
1386 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1387 if (host_err < 0) {
1388 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1389 goto out_nfserr;
1391 if (created)
1392 *created = 1;
1394 if (EX_ISSYNC(fhp->fh_export)) {
1395 err = nfserrno(nfsd_sync_dir(dentry));
1396 /* setattr will sync the child (or not) */
1399 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1400 /* Cram the verifier into atime/mtime */
1401 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1402 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1403 /* XXX someone who knows this better please fix it for nsec */
1404 iap->ia_mtime.tv_sec = v_mtime;
1405 iap->ia_atime.tv_sec = v_atime;
1406 iap->ia_mtime.tv_nsec = 0;
1407 iap->ia_atime.tv_nsec = 0;
1410 set_attr:
1411 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1412 if (err2)
1413 err = err2;
1415 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1417 * Update the filehandle to get the new inode info.
1419 if (!err)
1420 err = fh_update(resfhp);
1422 out:
1423 fh_unlock(fhp);
1424 if (dchild && !IS_ERR(dchild))
1425 dput(dchild);
1426 return err;
1428 out_nfserr:
1429 err = nfserrno(host_err);
1430 goto out;
1432 #endif /* CONFIG_NFSD_V3 */
1435 * Read a symlink. On entry, *lenp must contain the maximum path length that
1436 * fits into the buffer. On return, it contains the true length.
1437 * N.B. After this call fhp needs an fh_put
1439 __be32
1440 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1442 struct dentry *dentry;
1443 struct inode *inode;
1444 mm_segment_t oldfs;
1445 __be32 err;
1446 int host_err;
1448 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1449 if (err)
1450 goto out;
1452 dentry = fhp->fh_dentry;
1453 inode = dentry->d_inode;
1455 err = nfserr_inval;
1456 if (!inode->i_op->readlink)
1457 goto out;
1459 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1460 /* N.B. Why does this call need a get_fs()??
1461 * Remove the set_fs and watch the fireworks:-) --okir
1464 oldfs = get_fs(); set_fs(KERNEL_DS);
1465 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1466 set_fs(oldfs);
1468 if (host_err < 0)
1469 goto out_nfserr;
1470 *lenp = host_err;
1471 err = 0;
1472 out:
1473 return err;
1475 out_nfserr:
1476 err = nfserrno(host_err);
1477 goto out;
1481 * Create a symlink and look up its inode
1482 * N.B. After this call _both_ fhp and resfhp need an fh_put
1484 __be32
1485 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1486 char *fname, int flen,
1487 char *path, int plen,
1488 struct svc_fh *resfhp,
1489 struct iattr *iap)
1491 struct dentry *dentry, *dnew;
1492 __be32 err, cerr;
1493 int host_err;
1495 err = nfserr_noent;
1496 if (!flen || !plen)
1497 goto out;
1498 err = nfserr_exist;
1499 if (isdotent(fname, flen))
1500 goto out;
1502 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1503 if (err)
1504 goto out;
1505 fh_lock(fhp);
1506 dentry = fhp->fh_dentry;
1507 dnew = lookup_one_len(fname, dentry, flen);
1508 host_err = PTR_ERR(dnew);
1509 if (IS_ERR(dnew))
1510 goto out_nfserr;
1512 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1513 if (host_err)
1514 goto out_nfserr;
1516 if (unlikely(path[plen] != 0)) {
1517 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1518 if (path_alloced == NULL)
1519 host_err = -ENOMEM;
1520 else {
1521 strncpy(path_alloced, path, plen);
1522 path_alloced[plen] = 0;
1523 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1524 kfree(path_alloced);
1526 } else
1527 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1529 if (!host_err) {
1530 if (EX_ISSYNC(fhp->fh_export))
1531 host_err = nfsd_sync_dir(dentry);
1533 err = nfserrno(host_err);
1534 fh_unlock(fhp);
1536 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1538 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1539 dput(dnew);
1540 if (err==0) err = cerr;
1541 out:
1542 return err;
1544 out_nfserr:
1545 err = nfserrno(host_err);
1546 goto out;
1550 * Create a hardlink
1551 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1553 __be32
1554 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1555 char *name, int len, struct svc_fh *tfhp)
1557 struct dentry *ddir, *dnew, *dold;
1558 struct inode *dirp, *dest;
1559 __be32 err;
1560 int host_err;
1562 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1563 if (err)
1564 goto out;
1565 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1566 if (err)
1567 goto out;
1569 err = nfserr_perm;
1570 if (!len)
1571 goto out;
1572 err = nfserr_exist;
1573 if (isdotent(name, len))
1574 goto out;
1576 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1577 ddir = ffhp->fh_dentry;
1578 dirp = ddir->d_inode;
1580 dnew = lookup_one_len(name, ddir, len);
1581 host_err = PTR_ERR(dnew);
1582 if (IS_ERR(dnew))
1583 goto out_nfserr;
1585 dold = tfhp->fh_dentry;
1586 dest = dold->d_inode;
1588 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1589 if (host_err) {
1590 err = nfserrno(host_err);
1591 goto out_dput;
1593 host_err = vfs_link(dold, dirp, dnew);
1594 if (!host_err) {
1595 if (EX_ISSYNC(ffhp->fh_export)) {
1596 err = nfserrno(nfsd_sync_dir(ddir));
1597 write_inode_now(dest, 1);
1599 err = 0;
1600 } else {
1601 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1602 err = nfserr_acces;
1603 else
1604 err = nfserrno(host_err);
1606 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1607 out_dput:
1608 dput(dnew);
1609 out_unlock:
1610 fh_unlock(ffhp);
1611 out:
1612 return err;
1614 out_nfserr:
1615 err = nfserrno(host_err);
1616 goto out_unlock;
1620 * Rename a file
1621 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1623 __be32
1624 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1625 struct svc_fh *tfhp, char *tname, int tlen)
1627 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1628 struct inode *fdir, *tdir;
1629 __be32 err;
1630 int host_err;
1632 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1633 if (err)
1634 goto out;
1635 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1636 if (err)
1637 goto out;
1639 fdentry = ffhp->fh_dentry;
1640 fdir = fdentry->d_inode;
1642 tdentry = tfhp->fh_dentry;
1643 tdir = tdentry->d_inode;
1645 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1646 if (ffhp->fh_export != tfhp->fh_export)
1647 goto out;
1649 err = nfserr_perm;
1650 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1651 goto out;
1653 /* cannot use fh_lock as we need deadlock protective ordering
1654 * so do it by hand */
1655 trap = lock_rename(tdentry, fdentry);
1656 ffhp->fh_locked = tfhp->fh_locked = 1;
1657 fill_pre_wcc(ffhp);
1658 fill_pre_wcc(tfhp);
1660 odentry = lookup_one_len(fname, fdentry, flen);
1661 host_err = PTR_ERR(odentry);
1662 if (IS_ERR(odentry))
1663 goto out_nfserr;
1665 host_err = -ENOENT;
1666 if (!odentry->d_inode)
1667 goto out_dput_old;
1668 host_err = -EINVAL;
1669 if (odentry == trap)
1670 goto out_dput_old;
1672 ndentry = lookup_one_len(tname, tdentry, tlen);
1673 host_err = PTR_ERR(ndentry);
1674 if (IS_ERR(ndentry))
1675 goto out_dput_old;
1676 host_err = -ENOTEMPTY;
1677 if (ndentry == trap)
1678 goto out_dput_new;
1680 if (svc_msnfs(ffhp) &&
1681 ((atomic_read(&odentry->d_count) > 1)
1682 || (atomic_read(&ndentry->d_count) > 1))) {
1683 host_err = -EPERM;
1684 goto out_dput_new;
1687 host_err = -EXDEV;
1688 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1689 goto out_dput_new;
1690 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1691 if (host_err)
1692 goto out_dput_new;
1694 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1695 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1696 host_err = nfsd_sync_dir(tdentry);
1697 if (!host_err)
1698 host_err = nfsd_sync_dir(fdentry);
1701 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1703 out_dput_new:
1704 dput(ndentry);
1705 out_dput_old:
1706 dput(odentry);
1707 out_nfserr:
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;
1719 out:
1720 return err;
1724 * Unlink a file or directory
1725 * N.B. After this call fhp needs an fh_put
1727 __be32
1728 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1729 char *fname, int flen)
1731 struct dentry *dentry, *rdentry;
1732 struct inode *dirp;
1733 __be32 err;
1734 int host_err;
1736 err = nfserr_acces;
1737 if (!flen || isdotent(fname, flen))
1738 goto out;
1739 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1740 if (err)
1741 goto out;
1743 fh_lock_nested(fhp, I_MUTEX_PARENT);
1744 dentry = fhp->fh_dentry;
1745 dirp = dentry->d_inode;
1747 rdentry = lookup_one_len(fname, dentry, flen);
1748 host_err = PTR_ERR(rdentry);
1749 if (IS_ERR(rdentry))
1750 goto out_nfserr;
1752 if (!rdentry->d_inode) {
1753 dput(rdentry);
1754 err = nfserr_noent;
1755 goto out;
1758 if (!type)
1759 type = rdentry->d_inode->i_mode & S_IFMT;
1761 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1762 if (host_err)
1763 goto out_nfserr;
1765 if (type != S_IFDIR) { /* It's UNLINK */
1766 #ifdef MSNFS
1767 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1768 (atomic_read(&rdentry->d_count) > 1)) {
1769 host_err = -EPERM;
1770 } else
1771 #endif
1772 host_err = vfs_unlink(dirp, rdentry);
1773 } else { /* It's RMDIR */
1774 host_err = vfs_rmdir(dirp, rdentry);
1777 dput(rdentry);
1779 if (host_err)
1780 goto out_drop;
1781 if (EX_ISSYNC(fhp->fh_export))
1782 host_err = nfsd_sync_dir(dentry);
1784 out_drop:
1785 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1786 out_nfserr:
1787 err = nfserrno(host_err);
1788 out:
1789 return err;
1793 * We do this buffering because we must not call back into the file
1794 * system's ->lookup() method from the filldir callback. That may well
1795 * deadlock a number of file systems.
1797 * This is based heavily on the implementation of same in XFS.
1799 struct buffered_dirent {
1800 u64 ino;
1801 loff_t offset;
1802 int namlen;
1803 unsigned int d_type;
1804 char name[];
1807 struct readdir_data {
1808 char *dirent;
1809 size_t used;
1810 int full;
1813 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1814 loff_t offset, u64 ino, unsigned int d_type)
1816 struct readdir_data *buf = __buf;
1817 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1818 unsigned int reclen;
1820 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1821 if (buf->used + reclen > PAGE_SIZE) {
1822 buf->full = 1;
1823 return -EINVAL;
1826 de->namlen = namlen;
1827 de->offset = offset;
1828 de->ino = ino;
1829 de->d_type = d_type;
1830 memcpy(de->name, name, namlen);
1831 buf->used += reclen;
1833 return 0;
1836 static int nfsd_buffered_readdir(struct file *file, filldir_t func,
1837 struct readdir_cd *cdp, loff_t *offsetp)
1839 struct readdir_data buf;
1840 struct buffered_dirent *de;
1841 int host_err;
1842 int size;
1843 loff_t offset;
1845 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1846 if (!buf.dirent)
1847 return -ENOMEM;
1849 offset = *offsetp;
1851 while (1) {
1852 unsigned int reclen;
1854 cdp->err = nfserr_eof; /* will be cleared on successful read */
1855 buf.used = 0;
1856 buf.full = 0;
1858 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1859 if (buf.full)
1860 host_err = 0;
1862 if (host_err < 0)
1863 break;
1865 size = buf.used;
1867 if (!size)
1868 break;
1870 de = (struct buffered_dirent *)buf.dirent;
1871 while (size > 0) {
1872 offset = de->offset;
1874 if (func(cdp, de->name, de->namlen, de->offset,
1875 de->ino, de->d_type))
1876 goto done;
1878 if (cdp->err != nfs_ok)
1879 goto done;
1881 reclen = ALIGN(sizeof(*de) + de->namlen,
1882 sizeof(u64));
1883 size -= reclen;
1884 de = (struct buffered_dirent *)((char *)de + reclen);
1886 offset = vfs_llseek(file, 0, SEEK_CUR);
1889 done:
1890 free_page((unsigned long)(buf.dirent));
1892 if (host_err)
1893 return nfserrno(host_err);
1895 *offsetp = offset;
1896 return cdp->err;
1900 * Read entries from a directory.
1901 * The NFSv3/4 verifier we ignore for now.
1903 __be32
1904 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1905 struct readdir_cd *cdp, filldir_t func)
1907 __be32 err;
1908 struct file *file;
1909 loff_t offset = *offsetp;
1911 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1912 if (err)
1913 goto out;
1915 offset = vfs_llseek(file, offset, 0);
1916 if (offset < 0) {
1917 err = nfserrno((int)offset);
1918 goto out_close;
1921 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1923 if (err == nfserr_eof || err == nfserr_toosmall)
1924 err = nfs_ok; /* can still be found in ->err */
1925 out_close:
1926 nfsd_close(file);
1927 out:
1928 return err;
1932 * Get file system stats
1933 * N.B. After this call fhp needs an fh_put
1935 __be32
1936 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1938 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1939 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1940 err = nfserr_io;
1941 return err;
1944 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1946 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1950 * Check for a user's access permissions to this inode.
1952 __be32
1953 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1954 struct dentry *dentry, int acc)
1956 struct inode *inode = dentry->d_inode;
1957 int err;
1959 if (acc == NFSD_MAY_NOP)
1960 return 0;
1961 #if 0
1962 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1963 acc,
1964 (acc & NFSD_MAY_READ)? " read" : "",
1965 (acc & NFSD_MAY_WRITE)? " write" : "",
1966 (acc & NFSD_MAY_EXEC)? " exec" : "",
1967 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1968 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1969 (acc & NFSD_MAY_LOCK)? " lock" : "",
1970 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1971 inode->i_mode,
1972 IS_IMMUTABLE(inode)? " immut" : "",
1973 IS_APPEND(inode)? " append" : "",
1974 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
1975 dprintk(" owner %d/%d user %d/%d\n",
1976 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
1977 #endif
1979 /* Normally we reject any write/sattr etc access on a read-only file
1980 * system. But if it is IRIX doing check on write-access for a
1981 * device special file, we ignore rofs.
1983 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
1984 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
1985 if (exp_rdonly(rqstp, exp) ||
1986 __mnt_is_readonly(exp->ex_path.mnt))
1987 return nfserr_rofs;
1988 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
1989 return nfserr_perm;
1991 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
1992 return nfserr_perm;
1994 if (acc & NFSD_MAY_LOCK) {
1995 /* If we cannot rely on authentication in NLM requests,
1996 * just allow locks, otherwise require read permission, or
1997 * ownership
1999 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2000 return 0;
2001 else
2002 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2005 * The file owner always gets access permission for accesses that
2006 * would normally be checked at open time. This is to make
2007 * file access work even when the client has done a fchmod(fd, 0).
2009 * However, `cp foo bar' should fail nevertheless when bar is
2010 * readonly. A sensible way to do this might be to reject all
2011 * attempts to truncate a read-only file, because a creat() call
2012 * always implies file truncation.
2013 * ... but this isn't really fair. A process may reasonably call
2014 * ftruncate on an open file descriptor on a file with perm 000.
2015 * We must trust the client to do permission checking - using "ACCESS"
2016 * with NFSv3.
2018 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2019 inode->i_uid == current_fsuid())
2020 return 0;
2022 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2023 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2025 /* Allow read access to binaries even when mode 111 */
2026 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2027 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2028 err = inode_permission(inode, MAY_EXEC);
2030 return err? nfserrno(err) : 0;
2033 void
2034 nfsd_racache_shutdown(void)
2036 struct raparms *raparm, *last_raparm;
2037 unsigned int i;
2039 dprintk("nfsd: freeing readahead buffers.\n");
2041 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2042 raparm = raparm_hash[i].pb_head;
2043 while(raparm) {
2044 last_raparm = raparm;
2045 raparm = raparm->p_next;
2046 kfree(last_raparm);
2048 raparm_hash[i].pb_head = NULL;
2052 * Initialize readahead param cache
2055 nfsd_racache_init(int cache_size)
2057 int i;
2058 int j = 0;
2059 int nperbucket;
2060 struct raparms **raparm = NULL;
2063 if (raparm_hash[0].pb_head)
2064 return 0;
2065 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2066 if (nperbucket < 2)
2067 nperbucket = 2;
2068 cache_size = nperbucket * RAPARM_HASH_SIZE;
2070 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2072 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2073 spin_lock_init(&raparm_hash[i].pb_lock);
2075 raparm = &raparm_hash[i].pb_head;
2076 for (j = 0; j < nperbucket; j++) {
2077 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2078 if (!*raparm)
2079 goto out_nomem;
2080 raparm = &(*raparm)->p_next;
2082 *raparm = NULL;
2085 nfsdstats.ra_size = cache_size;
2086 return 0;
2088 out_nomem:
2089 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2090 nfsd_racache_shutdown();
2091 return -ENOMEM;
2094 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2095 struct posix_acl *
2096 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2098 struct inode *inode = fhp->fh_dentry->d_inode;
2099 char *name;
2100 void *value = NULL;
2101 ssize_t size;
2102 struct posix_acl *acl;
2104 if (!IS_POSIXACL(inode))
2105 return ERR_PTR(-EOPNOTSUPP);
2107 switch (type) {
2108 case ACL_TYPE_ACCESS:
2109 name = POSIX_ACL_XATTR_ACCESS;
2110 break;
2111 case ACL_TYPE_DEFAULT:
2112 name = POSIX_ACL_XATTR_DEFAULT;
2113 break;
2114 default:
2115 return ERR_PTR(-EOPNOTSUPP);
2118 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2119 if (size < 0)
2120 return ERR_PTR(size);
2122 acl = posix_acl_from_xattr(value, size);
2123 kfree(value);
2124 return acl;
2128 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2130 struct inode *inode = fhp->fh_dentry->d_inode;
2131 char *name;
2132 void *value = NULL;
2133 size_t size;
2134 int error;
2136 if (!IS_POSIXACL(inode) ||
2137 !inode->i_op->setxattr || !inode->i_op->removexattr)
2138 return -EOPNOTSUPP;
2139 switch(type) {
2140 case ACL_TYPE_ACCESS:
2141 name = POSIX_ACL_XATTR_ACCESS;
2142 break;
2143 case ACL_TYPE_DEFAULT:
2144 name = POSIX_ACL_XATTR_DEFAULT;
2145 break;
2146 default:
2147 return -EOPNOTSUPP;
2150 if (acl && acl->a_count) {
2151 size = posix_acl_xattr_size(acl->a_count);
2152 value = kmalloc(size, GFP_KERNEL);
2153 if (!value)
2154 return -ENOMEM;
2155 error = posix_acl_to_xattr(acl, value, size);
2156 if (error < 0)
2157 goto getout;
2158 size = error;
2159 } else
2160 size = 0;
2162 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2163 if (error)
2164 goto getout;
2165 if (size)
2166 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2167 else {
2168 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2169 error = 0;
2170 else {
2171 error = vfs_removexattr(fhp->fh_dentry, name);
2172 if (error == -ENODATA)
2173 error = 0;
2176 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2178 getout:
2179 kfree(value);
2180 return error;
2182 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */