?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge master.kernel.org:/home/rmk/linux-2.6-arm
[linux/fpc-iii.git] / fs / nfsd / vfs.c
blob31018333dc383f58905f6348ac7a949de04849ca
1 #define MSNFS /* HACK HACK */
2 /*
3 * linux/fs/nfsd/vfs.c
4 *
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.
8 *
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.
14 *
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 */
18
19 #include <linux/config.h>
20 #include <linux/string.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/fs.h>
24 #include <linux/file.h>
25 #include <linux/mount.h>
26 #include <linux/major.h>
27 #include <linux/ext2_fs.h>
28 #include <linux/proc_fs.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/net.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/pagemap.h>
35 #include <linux/in.h>
36 #include <linux/module.h>
37 #include <linux/namei.h>
38 #include <linux/vfs.h>
39 #include <linux/delay.h>
40 #include <linux/sunrpc/svc.h>
41 #include <linux/nfsd/nfsd.h>
42 #ifdef CONFIG_NFSD_V3
43 #include <linux/nfs3.h>
44 #include <linux/nfsd/xdr3.h>
45 #endif /* CONFIG_NFSD_V3 */
46 #include <linux/nfsd/nfsfh.h>
47 #include <linux/quotaops.h>
48 #include <linux/fsnotify.h>
49 #include <linux/posix_acl.h>
50 #include <linux/posix_acl_xattr.h>
51 #include <linux/xattr.h>
52 #ifdef CONFIG_NFSD_V4
53 #include <linux/nfs4.h>
54 #include <linux/nfs4_acl.h>
55 #include <linux/nfsd_idmap.h>
56 #include <linux/security.h>
57 #endif /* CONFIG_NFSD_V4 */
58
59 #include <asm/uaccess.h>
60
61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
62 #define NFSD_PARANOIA
63
64
65 /* We must ignore files (but only files) which might have mandatory
66 * locks on them because there is no way to know if the accesser has
67 * the lock.
68 */
69 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
70
71 /*
72 * This is a cache of readahead params that help us choose the proper
73 * readahead strategy. Initially, we set all readahead parameters to 0
74 * and let the VFS handle things.
75 * If you increase the number of cached files very much, you'll need to
76 * add a hash table here.
77 */
78 struct raparms {
79 struct raparms *p_next;
80 unsigned int p_count;
81 ino_t p_ino;
82 dev_t p_dev;
83 int p_set;
84 struct file_ra_state p_ra;
85 };
86
87 static struct raparms * raparml;
88 static struct raparms * raparm_cache;
89
90 /*
91 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
92 * a mount point.
93 * Returns -EAGAIN leaving *dpp and *expp unchanged,
94 * or nfs_ok having possibly changed *dpp and *expp
95 */
96 int
97 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
98 struct svc_export **expp)
99 {
100 struct svc_export *exp = *expp, *exp2 = NULL;
101 struct dentry *dentry = *dpp;
102 struct vfsmount *mnt = mntget(exp->ex_mnt);
103 struct dentry *mounts = dget(dentry);
104 int err = nfs_ok;
105
106 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
107
108 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
109 if (IS_ERR(exp2)) {
110 err = PTR_ERR(exp2);
111 dput(mounts);
112 mntput(mnt);
113 goto out;
114 }
115 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
116 /* successfully crossed mount point */
117 exp_put(exp);
118 *expp = exp2;
119 dput(dentry);
120 *dpp = mounts;
121 } else {
122 if (exp2) exp_put(exp2);
123 dput(mounts);
124 }
125 mntput(mnt);
126 out:
127 return err;
128 }
129
130 /*
131 * Look up one component of a pathname.
132 * N.B. After this call _both_ fhp and resfh need an fh_put
133 *
134 * If the lookup would cross a mountpoint, and the mounted filesystem
135 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
136 * accepted as it stands and the mounted directory is
137 * returned. Otherwise the covered directory is returned.
138 * NOTE: this mountpoint crossing is not supported properly by all
139 * clients and is explicitly disallowed for NFSv3
140 * NeilBrown <neilb@cse.unsw.edu.au>
141 */
142 int
143 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
144 int len, struct svc_fh *resfh)
145 {
146 struct svc_export *exp;
147 struct dentry *dparent;
148 struct dentry *dentry;
149 int err;
150
151 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
152
153 /* Obtain dentry and export. */
154 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
155 if (err)
156 return err;
157
158 dparent = fhp->fh_dentry;
159 exp = fhp->fh_export;
160 exp_get(exp);
161
162 err = nfserr_acces;
163
164 /* Lookup the name, but don't follow links */
165 if (isdotent(name, len)) {
166 if (len==1)
167 dentry = dget(dparent);
168 else if (dparent != exp->ex_dentry) {
169 dentry = dget_parent(dparent);
170 } else if (!EX_NOHIDE(exp))
171 dentry = dget(dparent); /* .. == . just like at / */
172 else {
173 /* checking mountpoint crossing is very different when stepping up */
174 struct svc_export *exp2 = NULL;
175 struct dentry *dp;
176 struct vfsmount *mnt = mntget(exp->ex_mnt);
177 dentry = dget(dparent);
178 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
179 ;
180 dp = dget_parent(dentry);
181 dput(dentry);
182 dentry = dp;
183
184 exp2 = exp_parent(exp->ex_client, mnt, dentry,
185 &rqstp->rq_chandle);
186 if (IS_ERR(exp2)) {
187 err = PTR_ERR(exp2);
188 dput(dentry);
189 mntput(mnt);
190 goto out_nfserr;
191 }
192 if (!exp2) {
193 dput(dentry);
194 dentry = dget(dparent);
195 } else {
196 exp_put(exp);
197 exp = exp2;
198 }
199 mntput(mnt);
200 }
201 } else {
202 fh_lock(fhp);
203 dentry = lookup_one_len(name, dparent, len);
204 err = PTR_ERR(dentry);
205 if (IS_ERR(dentry))
206 goto out_nfserr;
207 /*
208 * check if we have crossed a mount point ...
209 */
210 if (d_mountpoint(dentry)) {
211 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
212 dput(dentry);
213 goto out_nfserr;
214 }
215 }
216 }
217 /*
218 * Note: we compose the file handle now, but as the
219 * dentry may be negative, it may need to be updated.
220 */
221 err = fh_compose(resfh, exp, dentry, fhp);
222 if (!err && !dentry->d_inode)
223 err = nfserr_noent;
224 dput(dentry);
225 out:
226 exp_put(exp);
227 return err;
228
229 out_nfserr:
230 err = nfserrno(err);
231 goto out;
232 }
233
234 /*
235 * Set various file attributes.
236 * N.B. After this call fhp needs an fh_put
237 */
238 int
239 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
240 int check_guard, time_t guardtime)
241 {
242 struct dentry *dentry;
243 struct inode *inode;
244 int accmode = MAY_SATTR;
245 int ftype = 0;
246 int imode;
247 int err;
248 int size_change = 0;
249
250 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
251 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
252 if (iap->ia_valid & ATTR_SIZE)
253 ftype = S_IFREG;
254
255 /* Get inode */
256 err = fh_verify(rqstp, fhp, ftype, accmode);
257 if (err)
258 goto out;
259
260 dentry = fhp->fh_dentry;
261 inode = dentry->d_inode;
262
263 /* Ignore any mode updates on symlinks */
264 if (S_ISLNK(inode->i_mode))
265 iap->ia_valid &= ~ATTR_MODE;
266
267 if (!iap->ia_valid)
268 goto out;
269
270 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
271 * which only requires access, and "set-[ac]time-to-X" which
272 * requires ownership.
273 * So if it looks like it might be "set both to the same time which
274 * is close to now", and if inode_change_ok fails, then we
275 * convert to "set to now" instead of "set to explicit time"
276 *
277 * We only call inode_change_ok as the last test as technically
278 * it is not an interface that we should be using. It is only
279 * valid if the filesystem does not define it's own i_op->setattr.
280 */
281 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
282 #define MAX_TOUCH_TIME_ERROR (30*60)
283 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
284 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
285 ) {
286 /* Looks probable. Now just make sure time is in the right ballpark.
287 * Solaris, at least, doesn't seem to care what the time request is.
288 * We require it be within 30 minutes of now.
289 */
290 time_t delta = iap->ia_atime.tv_sec - get_seconds();
291 if (delta<0) delta = -delta;
292 if (delta < MAX_TOUCH_TIME_ERROR &&
293 inode_change_ok(inode, iap) != 0) {
294 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
295 * this will cause notify_change to set these times to "now"
296 */
297 iap->ia_valid &= ~BOTH_TIME_SET;
298 }
299 }
300
301 /* The size case is special. It changes the file as well as the attributes. */
302 if (iap->ia_valid & ATTR_SIZE) {
303 if (iap->ia_size < inode->i_size) {
304 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
305 if (err)
306 goto out;
307 }
308
309 /*
310 * If we are changing the size of the file, then
311 * we need to break all leases.
312 */
313 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
314 if (err == -EWOULDBLOCK)
315 err = -ETIMEDOUT;
316 if (err) /* ENOMEM or EWOULDBLOCK */
317 goto out_nfserr;
318
319 err = get_write_access(inode);
320 if (err)
321 goto out_nfserr;
322
323 size_change = 1;
324 err = locks_verify_truncate(inode, NULL, iap->ia_size);
325 if (err) {
326 put_write_access(inode);
327 goto out_nfserr;
328 }
329 DQUOT_INIT(inode);
330 }
331
332 imode = inode->i_mode;
333 if (iap->ia_valid & ATTR_MODE) {
334 iap->ia_mode &= S_IALLUGO;
335 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
336 }
337
338 /* Revoke setuid/setgid bit on chown/chgrp */
339 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
340 iap->ia_valid |= ATTR_KILL_SUID;
341 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
342 iap->ia_valid |= ATTR_KILL_SGID;
343
344 /* Change the attributes. */
345
346 iap->ia_valid |= ATTR_CTIME;
347
348 err = nfserr_notsync;
349 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
350 fh_lock(fhp);
351 err = notify_change(dentry, iap);
352 err = nfserrno(err);
353 fh_unlock(fhp);
354 }
355 if (size_change)
356 put_write_access(inode);
357 if (!err)
358 if (EX_ISSYNC(fhp->fh_export))
359 write_inode_now(inode, 1);
360 out:
361 return err;
362
363 out_nfserr:
364 err = nfserrno(err);
365 goto out;
366 }
367
368 #if defined(CONFIG_NFSD_V2_ACL) || \
369 defined(CONFIG_NFSD_V3_ACL) || \
370 defined(CONFIG_NFSD_V4)
371 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
372 {
373 ssize_t buflen;
374 int error;
375
376 buflen = vfs_getxattr(dentry, key, NULL, 0);
377 if (buflen <= 0)
378 return buflen;
379
380 *buf = kmalloc(buflen, GFP_KERNEL);
381 if (!*buf)
382 return -ENOMEM;
383
384 error = vfs_getxattr(dentry, key, *buf, buflen);
385 if (error < 0)
386 return error;
387 return buflen;
388 }
389 #endif
390
391 #if defined(CONFIG_NFSD_V4)
392 static int
393 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
394 {
395 int len;
396 size_t buflen;
397 char *buf = NULL;
398 int error = 0;
399
400 buflen = posix_acl_xattr_size(pacl->a_count);
401 buf = kmalloc(buflen, GFP_KERNEL);
402 error = -ENOMEM;
403 if (buf == NULL)
404 goto out;
405
406 len = posix_acl_to_xattr(pacl, buf, buflen);
407 if (len < 0) {
408 error = len;
409 goto out;
410 }
411
412 error = vfs_setxattr(dentry, key, buf, len, 0);
413 out:
414 kfree(buf);
415 return error;
416 }
417
418 int
419 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
420 struct nfs4_acl *acl)
421 {
422 int error;
423 struct dentry *dentry;
424 struct inode *inode;
425 struct posix_acl *pacl = NULL, *dpacl = NULL;
426 unsigned int flags = 0;
427
428 /* Get inode */
429 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
430 if (error)
431 goto out;
432
433 dentry = fhp->fh_dentry;
434 inode = dentry->d_inode;
435 if (S_ISDIR(inode->i_mode))
436 flags = NFS4_ACL_DIR;
437
438 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
439 if (error == -EINVAL) {
440 error = nfserr_attrnotsupp;
441 goto out;
442 } else if (error < 0)
443 goto out_nfserr;
444
445 if (pacl) {
446 error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
447 if (error < 0)
448 goto out_nfserr;
449 }
450
451 if (dpacl) {
452 error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
453 if (error < 0)
454 goto out_nfserr;
455 }
456
457 error = nfs_ok;
458
459 out:
460 posix_acl_release(pacl);
461 posix_acl_release(dpacl);
462 return (error);
463 out_nfserr:
464 error = nfserrno(error);
465 goto out;
466 }
467
468 static struct posix_acl *
469 _get_posix_acl(struct dentry *dentry, char *key)
470 {
471 void *buf = NULL;
472 struct posix_acl *pacl = NULL;
473 int buflen;
474
475 buflen = nfsd_getxattr(dentry, key, &buf);
476 if (!buflen)
477 buflen = -ENODATA;
478 if (buflen <= 0)
479 return ERR_PTR(buflen);
480
481 pacl = posix_acl_from_xattr(buf, buflen);
482 kfree(buf);
483 return pacl;
484 }
485
486 int
487 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
488 {
489 struct inode *inode = dentry->d_inode;
490 int error = 0;
491 struct posix_acl *pacl = NULL, *dpacl = NULL;
492 unsigned int flags = 0;
493
494 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
495 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
496 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
497 if (IS_ERR(pacl)) {
498 error = PTR_ERR(pacl);
499 pacl = NULL;
500 goto out;
501 }
502
503 if (S_ISDIR(inode->i_mode)) {
504 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
505 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
506 dpacl = NULL;
507 else if (IS_ERR(dpacl)) {
508 error = PTR_ERR(dpacl);
509 dpacl = NULL;
510 goto out;
511 }
512 flags = NFS4_ACL_DIR;
513 }
514
515 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
516 if (IS_ERR(*acl)) {
517 error = PTR_ERR(*acl);
518 *acl = NULL;
519 }
520 out:
521 posix_acl_release(pacl);
522 posix_acl_release(dpacl);
523 return error;
524 }
525
526 #endif /* defined(CONFIG_NFS_V4) */
527
528 #ifdef CONFIG_NFSD_V3
529 /*
530 * Check server access rights to a file system object
531 */
532 struct accessmap {
533 u32 access;
534 int how;
535 };
536 static struct accessmap nfs3_regaccess[] = {
537 { NFS3_ACCESS_READ, MAY_READ },
538 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
539 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
540 { NFS3_ACCESS_EXTEND, MAY_WRITE },
541
542 { 0, 0 }
543 };
544
545 static struct accessmap nfs3_diraccess[] = {
546 { NFS3_ACCESS_READ, MAY_READ },
547 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
548 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
549 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
550 { NFS3_ACCESS_DELETE, MAY_REMOVE },
551
552 { 0, 0 }
553 };
554
555 static struct accessmap nfs3_anyaccess[] = {
556 /* Some clients - Solaris 2.6 at least, make an access call
557 * to the server to check for access for things like /dev/null
558 * (which really, the server doesn't care about). So
559 * We provide simple access checking for them, looking
560 * mainly at mode bits, and we make sure to ignore read-only
561 * filesystem checks
562 */
563 { NFS3_ACCESS_READ, MAY_READ },
564 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
565 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
566 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
567
568 { 0, 0 }
569 };
570
571 int
572 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
573 {
574 struct accessmap *map;
575 struct svc_export *export;
576 struct dentry *dentry;
577 u32 query, result = 0, sresult = 0;
578 unsigned int error;
579
580 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
581 if (error)
582 goto out;
583
584 export = fhp->fh_export;
585 dentry = fhp->fh_dentry;
586
587 if (S_ISREG(dentry->d_inode->i_mode))
588 map = nfs3_regaccess;
589 else if (S_ISDIR(dentry->d_inode->i_mode))
590 map = nfs3_diraccess;
591 else
592 map = nfs3_anyaccess;
593
594
595 query = *access;
596 for (; map->access; map++) {
597 if (map->access & query) {
598 unsigned int err2;
599
600 sresult |= map->access;
601
602 err2 = nfsd_permission(export, dentry, map->how);
603 switch (err2) {
604 case nfs_ok:
605 result |= map->access;
606 break;
607
608 /* the following error codes just mean the access was not allowed,
609 * rather than an error occurred */
610 case nfserr_rofs:
611 case nfserr_acces:
612 case nfserr_perm:
613 /* simply don't "or" in the access bit. */
614 break;
615 default:
616 error = err2;
617 goto out;
618 }
619 }
620 }
621 *access = result;
622 if (supported)
623 *supported = sresult;
624
625 out:
626 return error;
627 }
628 #endif /* CONFIG_NFSD_V3 */
629
630
631
632 /*
633 * Open an existing file or directory.
634 * The access argument indicates the type of open (read/write/lock)
635 * N.B. After this call fhp needs an fh_put
636 */
637 int
638 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
639 int access, struct file **filp)
640 {
641 struct dentry *dentry;
642 struct inode *inode;
643 int flags = O_RDONLY|O_LARGEFILE, err;
644
645 /*
646 * If we get here, then the client has already done an "open",
647 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
648 * in case a chmod has now revoked permission.
649 */
650 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
651 if (err)
652 goto out;
653
654 dentry = fhp->fh_dentry;
655 inode = dentry->d_inode;
656
657 /* Disallow write access to files with the append-only bit set
658 * or any access when mandatory locking enabled
659 */
660 err = nfserr_perm;
661 if (IS_APPEND(inode) && (access & MAY_WRITE))
662 goto out;
663 if (IS_ISMNDLK(inode))
664 goto out;
665
666 if (!inode->i_fop)
667 goto out;
668
669 /*
670 * Check to see if there are any leases on this file.
671 * This may block while leases are broken.
672 */
673 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
674 if (err == -EWOULDBLOCK)
675 err = -ETIMEDOUT;
676 if (err) /* NOMEM or WOULDBLOCK */
677 goto out_nfserr;
678
679 if (access & MAY_WRITE) {
680 flags = O_WRONLY|O_LARGEFILE;
681
682 DQUOT_INIT(inode);
683 }
684 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
685 if (IS_ERR(*filp))
686 err = PTR_ERR(*filp);
687 out_nfserr:
688 if (err)
689 err = nfserrno(err);
690 out:
691 return err;
692 }
693
694 /*
695 * Close a file.
696 */
697 void
698 nfsd_close(struct file *filp)
699 {
700 fput(filp);
701 }
702
703 /*
704 * Sync a file
705 * As this calls fsync (not fdatasync) there is no need for a write_inode
706 * after it.
707 */
708 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
709 const struct file_operations *fop)
710 {
711 struct inode *inode = dp->d_inode;
712 int (*fsync) (struct file *, struct dentry *, int);
713 int err;
714
715 err = filemap_fdatawrite(inode->i_mapping);
716 if (err == 0 && fop && (fsync = fop->fsync))
717 err = fsync(filp, dp, 0);
718 if (err == 0)
719 err = filemap_fdatawait(inode->i_mapping);
720
721 return err;
722 }
723
724
725 static int
726 nfsd_sync(struct file *filp)
727 {
728 int err;
729 struct inode *inode = filp->f_dentry->d_inode;
730 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
731 mutex_lock(&inode->i_mutex);
732 err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
733 mutex_unlock(&inode->i_mutex);
734
735 return err;
736 }
737
738 int
739 nfsd_sync_dir(struct dentry *dp)
740 {
741 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
742 }
743
744 /*
745 * Obtain the readahead parameters for the file
746 * specified by (dev, ino).
747 */
748 static DEFINE_SPINLOCK(ra_lock);
749
750 static inline struct raparms *
751 nfsd_get_raparms(dev_t dev, ino_t ino)
752 {
753 struct raparms *ra, **rap, **frap = NULL;
754 int depth = 0;
755
756 spin_lock(&ra_lock);
757 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
758 if (ra->p_ino == ino && ra->p_dev == dev)
759 goto found;
760 depth++;
761 if (ra->p_count == 0)
762 frap = rap;
763 }
764 depth = nfsdstats.ra_size*11/10;
765 if (!frap) {
766 spin_unlock(&ra_lock);
767 return NULL;
768 }
769 rap = frap;
770 ra = *frap;
771 ra->p_dev = dev;
772 ra->p_ino = ino;
773 ra->p_set = 0;
774 found:
775 if (rap != &raparm_cache) {
776 *rap = ra->p_next;
777 ra->p_next = raparm_cache;
778 raparm_cache = ra;
779 }
780 ra->p_count++;
781 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
782 spin_unlock(&ra_lock);
783 return ra;
784 }
785
786 /*
787 * Grab and keep cached pages assosiated with a file in the svc_rqst
788 * so that they can be passed to the netowork sendmsg/sendpage routines
789 * directrly. They will be released after the sending has completed.
790 */
791 static int
792 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
793 {
794 unsigned long count = desc->count;
795 struct svc_rqst *rqstp = desc->arg.data;
796
797 if (size > count)
798 size = count;
799
800 if (rqstp->rq_res.page_len == 0) {
801 get_page(page);
802 rqstp->rq_respages[rqstp->rq_resused++] = page;
803 rqstp->rq_res.page_base = offset;
804 rqstp->rq_res.page_len = size;
805 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
806 get_page(page);
807 rqstp->rq_respages[rqstp->rq_resused++] = page;
808 rqstp->rq_res.page_len += size;
809 } else {
810 rqstp->rq_res.page_len += size;
811 }
812
813 desc->count = count - size;
814 desc->written += size;
815 return size;
816 }
817
818 static int
819 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
820 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
821 {
822 struct inode *inode;
823 struct raparms *ra;
824 mm_segment_t oldfs;
825 int err;
826
827 err = nfserr_perm;
828 inode = file->f_dentry->d_inode;
829 #ifdef MSNFS
830 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
831 (!lock_may_read(inode, offset, *count)))
832 goto out;
833 #endif
834
835 /* Get readahead parameters */
836 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
837
838 if (ra && ra->p_set)
839 file->f_ra = ra->p_ra;
840
841 if (file->f_op->sendfile) {
842 svc_pushback_unused_pages(rqstp);
843 err = file->f_op->sendfile(file, &offset, *count,
844 nfsd_read_actor, rqstp);
845 } else {
846 oldfs = get_fs();
847 set_fs(KERNEL_DS);
848 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
849 set_fs(oldfs);
850 }
851
852 /* Write back readahead params */
853 if (ra) {
854 spin_lock(&ra_lock);
855 ra->p_ra = file->f_ra;
856 ra->p_set = 1;
857 ra->p_count--;
858 spin_unlock(&ra_lock);
859 }
860
861 if (err >= 0) {
862 nfsdstats.io_read += err;
863 *count = err;
864 err = 0;
865 fsnotify_access(file->f_dentry);
866 } else
867 err = nfserrno(err);
868 out:
869 return err;
870 }
871
872 static void kill_suid(struct dentry *dentry)
873 {
874 struct iattr ia;
875 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
876
877 mutex_lock(&dentry->d_inode->i_mutex);
878 notify_change(dentry, &ia);
879 mutex_unlock(&dentry->d_inode->i_mutex);
880 }
881
882 static int
883 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
884 loff_t offset, struct kvec *vec, int vlen,
885 unsigned long cnt, int *stablep)
886 {
887 struct svc_export *exp;
888 struct dentry *dentry;
889 struct inode *inode;
890 mm_segment_t oldfs;
891 int err = 0;
892 int stable = *stablep;
893
894 #ifdef MSNFS
895 err = nfserr_perm;
896
897 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
898 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
899 goto out;
900 #endif
901
902 dentry = file->f_dentry;
903 inode = dentry->d_inode;
904 exp = fhp->fh_export;
905
906 /*
907 * Request sync writes if
908 * - the sync export option has been set, or
909 * - the client requested O_SYNC behavior (NFSv3 feature).
910 * - The file system doesn't support fsync().
911 * When gathered writes have been configured for this volume,
912 * flushing the data to disk is handled separately below.
913 */
914
915 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
916 stable = 2;
917 *stablep = 2; /* FILE_SYNC */
918 }
919
920 if (!EX_ISSYNC(exp))
921 stable = 0;
922 if (stable && !EX_WGATHER(exp))
923 file->f_flags |= O_SYNC;
924
925 /* Write the data. */
926 oldfs = get_fs(); set_fs(KERNEL_DS);
927 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
928 set_fs(oldfs);
929 if (err >= 0) {
930 nfsdstats.io_write += cnt;
931 fsnotify_modify(file->f_dentry);
932 }
933
934 /* clear setuid/setgid flag after write */
935 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
936 kill_suid(dentry);
937
938 if (err >= 0 && stable) {
939 static ino_t last_ino;
940 static dev_t last_dev;
941
942 /*
943 * Gathered writes: If another process is currently
944 * writing to the file, there's a high chance
945 * this is another nfsd (triggered by a bulk write
946 * from a client's biod). Rather than syncing the
947 * file with each write request, we sleep for 10 msec.
948 *
949 * I don't know if this roughly approximates
950 * C. Juszak's idea of gathered writes, but it's a
951 * nice and simple solution (IMHO), and it seems to
952 * work:-)
953 */
954 if (EX_WGATHER(exp)) {
955 if (atomic_read(&inode->i_writecount) > 1
956 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
957 dprintk("nfsd: write defer %d\n", current->pid);
958 msleep(10);
959 dprintk("nfsd: write resume %d\n", current->pid);
960 }
961
962 if (inode->i_state & I_DIRTY) {
963 dprintk("nfsd: write sync %d\n", current->pid);
964 err=nfsd_sync(file);
965 }
966 #if 0
967 wake_up(&inode->i_wait);
968 #endif
969 }
970 last_ino = inode->i_ino;
971 last_dev = inode->i_sb->s_dev;
972 }
973
974 dprintk("nfsd: write complete err=%d\n", err);
975 if (err >= 0)
976 err = 0;
977 else
978 err = nfserrno(err);
979 out:
980 return err;
981 }
982
983 /*
984 * Read data from a file. count must contain the requested read count
985 * on entry. On return, *count contains the number of bytes actually read.
986 * N.B. After this call fhp needs an fh_put
987 */
988 int
989 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
990 loff_t offset, struct kvec *vec, int vlen,
991 unsigned long *count)
992 {
993 int err;
994
995 if (file) {
996 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
997 MAY_READ|MAY_OWNER_OVERRIDE);
998 if (err)
999 goto out;
1000 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1001 } else {
1002 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1003 if (err)
1004 goto out;
1005 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1006 nfsd_close(file);
1007 }
1008 out:
1009 return err;
1010 }
1011
1012 /*
1013 * Write data to a file.
1014 * The stable flag requests synchronous writes.
1015 * N.B. After this call fhp needs an fh_put
1016 */
1017 int
1018 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1019 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1020 int *stablep)
1021 {
1022 int err = 0;
1023
1024 if (file) {
1025 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1026 MAY_WRITE|MAY_OWNER_OVERRIDE);
1027 if (err)
1028 goto out;
1029 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1030 stablep);
1031 } else {
1032 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1033 if (err)
1034 goto out;
1035
1036 if (cnt)
1037 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1038 cnt, stablep);
1039 nfsd_close(file);
1040 }
1041 out:
1042 return err;
1043 }
1044
1045 #ifdef CONFIG_NFSD_V3
1046 /*
1047 * Commit all pending writes to stable storage.
1048 * Strictly speaking, we could sync just the indicated file region here,
1049 * but there's currently no way we can ask the VFS to do so.
1050 *
1051 * Unfortunately we cannot lock the file to make sure we return full WCC
1052 * data to the client, as locking happens lower down in the filesystem.
1053 */
1054 int
1055 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1056 loff_t offset, unsigned long count)
1057 {
1058 struct file *file;
1059 int err;
1060
1061 if ((u64)count > ~(u64)offset)
1062 return nfserr_inval;
1063
1064 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1065 return err;
1066 if (EX_ISSYNC(fhp->fh_export)) {
1067 if (file->f_op && file->f_op->fsync) {
1068 err = nfserrno(nfsd_sync(file));
1069 } else {
1070 err = nfserr_notsupp;
1071 }
1072 }
1073
1074 nfsd_close(file);
1075 return err;
1076 }
1077 #endif /* CONFIG_NFSD_V3 */
1078
1079 /*
1080 * Create a file (regular, directory, device, fifo); UNIX sockets
1081 * not yet implemented.
1082 * If the response fh has been verified, the parent directory should
1083 * already be locked. Note that the parent directory is left locked.
1084 *
1085 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1086 */
1087 int
1088 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1089 char *fname, int flen, struct iattr *iap,
1090 int type, dev_t rdev, struct svc_fh *resfhp)
1091 {
1092 struct dentry *dentry, *dchild = NULL;
1093 struct inode *dirp;
1094 int err;
1095
1096 err = nfserr_perm;
1097 if (!flen)
1098 goto out;
1099 err = nfserr_exist;
1100 if (isdotent(fname, flen))
1101 goto out;
1102
1103 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1104 if (err)
1105 goto out;
1106
1107 dentry = fhp->fh_dentry;
1108 dirp = dentry->d_inode;
1109
1110 err = nfserr_notdir;
1111 if(!dirp->i_op || !dirp->i_op->lookup)
1112 goto out;
1113 /*
1114 * Check whether the response file handle has been verified yet.
1115 * If it has, the parent directory should already be locked.
1116 */
1117 if (!resfhp->fh_dentry) {
1118 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1119 fh_lock(fhp);
1120 dchild = lookup_one_len(fname, dentry, flen);
1121 err = PTR_ERR(dchild);
1122 if (IS_ERR(dchild))
1123 goto out_nfserr;
1124 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1125 if (err)
1126 goto out;
1127 } else {
1128 /* called from nfsd_proc_create */
1129 dchild = dget(resfhp->fh_dentry);
1130 if (!fhp->fh_locked) {
1131 /* not actually possible */
1132 printk(KERN_ERR
1133 "nfsd_create: parent %s/%s not locked!\n",
1134 dentry->d_parent->d_name.name,
1135 dentry->d_name.name);
1136 err = nfserr_io;
1137 goto out;
1138 }
1139 }
1140 /*
1141 * Make sure the child dentry is still negative ...
1142 */
1143 err = nfserr_exist;
1144 if (dchild->d_inode) {
1145 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1146 dentry->d_name.name, dchild->d_name.name);
1147 goto out;
1148 }
1149
1150 if (!(iap->ia_valid & ATTR_MODE))
1151 iap->ia_mode = 0;
1152 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1153
1154 /*
1155 * Get the dir op function pointer.
1156 */
1157 err = nfserr_perm;
1158 switch (type) {
1159 case S_IFREG:
1160 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1161 break;
1162 case S_IFDIR:
1163 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1164 break;
1165 case S_IFCHR:
1166 case S_IFBLK:
1167 case S_IFIFO:
1168 case S_IFSOCK:
1169 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1170 break;
1171 default:
1172 printk("nfsd: bad file type %o in nfsd_create\n", type);
1173 err = -EINVAL;
1174 }
1175 if (err < 0)
1176 goto out_nfserr;
1177
1178 if (EX_ISSYNC(fhp->fh_export)) {
1179 err = nfserrno(nfsd_sync_dir(dentry));
1180 write_inode_now(dchild->d_inode, 1);
1181 }
1182
1183
1184 /* Set file attributes. Mode has already been set and
1185 * setting uid/gid works only for root. Irix appears to
1186 * send along the gid when it tries to implement setgid
1187 * directories via NFS.
1188 */
1189 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1190 int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1191 if (err2)
1192 err = err2;
1193 }
1194 /*
1195 * Update the file handle to get the new inode info.
1196 */
1197 if (!err)
1198 err = fh_update(resfhp);
1199 out:
1200 if (dchild && !IS_ERR(dchild))
1201 dput(dchild);
1202 return err;
1203
1204 out_nfserr:
1205 err = nfserrno(err);
1206 goto out;
1207 }
1208
1209 #ifdef CONFIG_NFSD_V3
1210 /*
1211 * NFSv3 version of nfsd_create
1212 */
1213 int
1214 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1215 char *fname, int flen, struct iattr *iap,
1216 struct svc_fh *resfhp, int createmode, u32 *verifier,
1217 int *truncp)
1218 {
1219 struct dentry *dentry, *dchild = NULL;
1220 struct inode *dirp;
1221 int err;
1222 __u32 v_mtime=0, v_atime=0;
1223 int v_mode=0;
1224
1225 err = nfserr_perm;
1226 if (!flen)
1227 goto out;
1228 err = nfserr_exist;
1229 if (isdotent(fname, flen))
1230 goto out;
1231 if (!(iap->ia_valid & ATTR_MODE))
1232 iap->ia_mode = 0;
1233 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1234 if (err)
1235 goto out;
1236
1237 dentry = fhp->fh_dentry;
1238 dirp = dentry->d_inode;
1239
1240 /* Get all the sanity checks out of the way before
1241 * we lock the parent. */
1242 err = nfserr_notdir;
1243 if(!dirp->i_op || !dirp->i_op->lookup)
1244 goto out;
1245 fh_lock(fhp);
1246
1247 /*
1248 * Compose the response file handle.
1249 */
1250 dchild = lookup_one_len(fname, dentry, flen);
1251 err = PTR_ERR(dchild);
1252 if (IS_ERR(dchild))
1253 goto out_nfserr;
1254
1255 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1256 if (err)
1257 goto out;
1258
1259 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1260 /* while the verifier would fit in mtime+atime,
1261 * solaris7 gets confused (bugid 4218508) if these have
1262 * the high bit set, so we use the mode as well
1263 */
1264 v_mtime = verifier[0]&0x7fffffff;
1265 v_atime = verifier[1]&0x7fffffff;
1266 v_mode = S_IFREG
1267 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1268 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1269 ;
1270 }
1271
1272 if (dchild->d_inode) {
1273 err = 0;
1274
1275 switch (createmode) {
1276 case NFS3_CREATE_UNCHECKED:
1277 if (! S_ISREG(dchild->d_inode->i_mode))
1278 err = nfserr_exist;
1279 else if (truncp) {
1280 /* in nfsv4, we need to treat this case a little
1281 * differently. we don't want to truncate the
1282 * file now; this would be wrong if the OPEN
1283 * fails for some other reason. furthermore,
1284 * if the size is nonzero, we should ignore it
1285 * according to spec!
1286 */
1287 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1288 }
1289 else {
1290 iap->ia_valid &= ATTR_SIZE;
1291 goto set_attr;
1292 }
1293 break;
1294 case NFS3_CREATE_EXCLUSIVE:
1295 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1296 && dchild->d_inode->i_atime.tv_sec == v_atime
1297 && dchild->d_inode->i_mode == v_mode
1298 && dchild->d_inode->i_size == 0 )
1299 break;
1300 /* fallthru */
1301 case NFS3_CREATE_GUARDED:
1302 err = nfserr_exist;
1303 }
1304 goto out;
1305 }
1306
1307 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1308 if (err < 0)
1309 goto out_nfserr;
1310
1311 if (EX_ISSYNC(fhp->fh_export)) {
1312 err = nfserrno(nfsd_sync_dir(dentry));
1313 /* setattr will sync the child (or not) */
1314 }
1315
1316 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1317 /* Cram the verifier into atime/mtime/mode */
1318 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1319 | ATTR_MTIME_SET|ATTR_ATIME_SET
1320 | ATTR_MODE;
1321 /* XXX someone who knows this better please fix it for nsec */
1322 iap->ia_mtime.tv_sec = v_mtime;
1323 iap->ia_atime.tv_sec = v_atime;
1324 iap->ia_mtime.tv_nsec = 0;
1325 iap->ia_atime.tv_nsec = 0;
1326 iap->ia_mode = v_mode;
1327 }
1328
1329 /* Set file attributes.
1330 * Mode has already been set but we might need to reset it
1331 * for CREATE_EXCLUSIVE
1332 * Irix appears to send along the gid when it tries to
1333 * implement setgid directories via NFS. Clear out all that cruft.
1334 */
1335 set_attr:
1336 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) {
1337 int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1338 if (err2)
1339 err = err2;
1340 }
1341
1342 /*
1343 * Update the filehandle to get the new inode info.
1344 */
1345 if (!err)
1346 err = fh_update(resfhp);
1347
1348 out:
1349 fh_unlock(fhp);
1350 if (dchild && !IS_ERR(dchild))
1351 dput(dchild);
1352 return err;
1353
1354 out_nfserr:
1355 err = nfserrno(err);
1356 goto out;
1357 }
1358 #endif /* CONFIG_NFSD_V3 */
1359
1360 /*
1361 * Read a symlink. On entry, *lenp must contain the maximum path length that
1362 * fits into the buffer. On return, it contains the true length.
1363 * N.B. After this call fhp needs an fh_put
1364 */
1365 int
1366 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1367 {
1368 struct dentry *dentry;
1369 struct inode *inode;
1370 mm_segment_t oldfs;
1371 int err;
1372
1373 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1374 if (err)
1375 goto out;
1376
1377 dentry = fhp->fh_dentry;
1378 inode = dentry->d_inode;
1379
1380 err = nfserr_inval;
1381 if (!inode->i_op || !inode->i_op->readlink)
1382 goto out;
1383
1384 touch_atime(fhp->fh_export->ex_mnt, dentry);
1385 /* N.B. Why does this call need a get_fs()??
1386 * Remove the set_fs and watch the fireworks:-) --okir
1387 */
1388
1389 oldfs = get_fs(); set_fs(KERNEL_DS);
1390 err = inode->i_op->readlink(dentry, buf, *lenp);
1391 set_fs(oldfs);
1392
1393 if (err < 0)
1394 goto out_nfserr;
1395 *lenp = err;
1396 err = 0;
1397 out:
1398 return err;
1399
1400 out_nfserr:
1401 err = nfserrno(err);
1402 goto out;
1403 }
1404
1405 /*
1406 * Create a symlink and look up its inode
1407 * N.B. After this call _both_ fhp and resfhp need an fh_put
1408 */
1409 int
1410 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1411 char *fname, int flen,
1412 char *path, int plen,
1413 struct svc_fh *resfhp,
1414 struct iattr *iap)
1415 {
1416 struct dentry *dentry, *dnew;
1417 int err, cerr;
1418 umode_t mode;
1419
1420 err = nfserr_noent;
1421 if (!flen || !plen)
1422 goto out;
1423 err = nfserr_exist;
1424 if (isdotent(fname, flen))
1425 goto out;
1426
1427 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1428 if (err)
1429 goto out;
1430 fh_lock(fhp);
1431 dentry = fhp->fh_dentry;
1432 dnew = lookup_one_len(fname, dentry, flen);
1433 err = PTR_ERR(dnew);
1434 if (IS_ERR(dnew))
1435 goto out_nfserr;
1436
1437 mode = S_IALLUGO;
1438 /* Only the MODE ATTRibute is even vaguely meaningful */
1439 if (iap && (iap->ia_valid & ATTR_MODE))
1440 mode = iap->ia_mode & S_IALLUGO;
1441
1442 if (unlikely(path[plen] != 0)) {
1443 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1444 if (path_alloced == NULL)
1445 err = -ENOMEM;
1446 else {
1447 strncpy(path_alloced, path, plen);
1448 path_alloced[plen] = 0;
1449 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1450 kfree(path_alloced);
1451 }
1452 } else
1453 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1454
1455 if (!err)
1456 if (EX_ISSYNC(fhp->fh_export))
1457 err = nfsd_sync_dir(dentry);
1458 if (err)
1459 err = nfserrno(err);
1460 fh_unlock(fhp);
1461
1462 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1463 dput(dnew);
1464 if (err==0) err = cerr;
1465 out:
1466 return err;
1467
1468 out_nfserr:
1469 err = nfserrno(err);
1470 goto out;
1471 }
1472
1473 /*
1474 * Create a hardlink
1475 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1476 */
1477 int
1478 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1479 char *name, int len, struct svc_fh *tfhp)
1480 {
1481 struct dentry *ddir, *dnew, *dold;
1482 struct inode *dirp, *dest;
1483 int err;
1484
1485 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1486 if (err)
1487 goto out;
1488 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1489 if (err)
1490 goto out;
1491
1492 err = nfserr_perm;
1493 if (!len)
1494 goto out;
1495 err = nfserr_exist;
1496 if (isdotent(name, len))
1497 goto out;
1498
1499 fh_lock(ffhp);
1500 ddir = ffhp->fh_dentry;
1501 dirp = ddir->d_inode;
1502
1503 dnew = lookup_one_len(name, ddir, len);
1504 err = PTR_ERR(dnew);
1505 if (IS_ERR(dnew))
1506 goto out_nfserr;
1507
1508 dold = tfhp->fh_dentry;
1509 dest = dold->d_inode;
1510
1511 err = vfs_link(dold, dirp, dnew);
1512 if (!err) {
1513 if (EX_ISSYNC(ffhp->fh_export)) {
1514 err = nfserrno(nfsd_sync_dir(ddir));
1515 write_inode_now(dest, 1);
1516 }
1517 } else {
1518 if (err == -EXDEV && rqstp->rq_vers == 2)
1519 err = nfserr_acces;
1520 else
1521 err = nfserrno(err);
1522 }
1523
1524 fh_unlock(ffhp);
1525 dput(dnew);
1526 out:
1527 return err;
1528
1529 out_nfserr:
1530 err = nfserrno(err);
1531 goto out;
1532 }
1533
1534 /*
1535 * Rename a file
1536 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1537 */
1538 int
1539 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1540 struct svc_fh *tfhp, char *tname, int tlen)
1541 {
1542 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1543 struct inode *fdir, *tdir;
1544 int err;
1545
1546 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1547 if (err)
1548 goto out;
1549 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1550 if (err)
1551 goto out;
1552
1553 fdentry = ffhp->fh_dentry;
1554 fdir = fdentry->d_inode;
1555
1556 tdentry = tfhp->fh_dentry;
1557 tdir = tdentry->d_inode;
1558
1559 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1560 if (fdir->i_sb != tdir->i_sb)
1561 goto out;
1562
1563 err = nfserr_perm;
1564 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1565 goto out;
1566
1567 /* cannot use fh_lock as we need deadlock protective ordering
1568 * so do it by hand */
1569 trap = lock_rename(tdentry, fdentry);
1570 ffhp->fh_locked = tfhp->fh_locked = 1;
1571 fill_pre_wcc(ffhp);
1572 fill_pre_wcc(tfhp);
1573
1574 odentry = lookup_one_len(fname, fdentry, flen);
1575 err = PTR_ERR(odentry);
1576 if (IS_ERR(odentry))
1577 goto out_nfserr;
1578
1579 err = -ENOENT;
1580 if (!odentry->d_inode)
1581 goto out_dput_old;
1582 err = -EINVAL;
1583 if (odentry == trap)
1584 goto out_dput_old;
1585
1586 ndentry = lookup_one_len(tname, tdentry, tlen);
1587 err = PTR_ERR(ndentry);
1588 if (IS_ERR(ndentry))
1589 goto out_dput_old;
1590 err = -ENOTEMPTY;
1591 if (ndentry == trap)
1592 goto out_dput_new;
1593
1594 #ifdef MSNFS
1595 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1596 ((atomic_read(&odentry->d_count) > 1)
1597 || (atomic_read(&ndentry->d_count) > 1))) {
1598 err = -EPERM;
1599 } else
1600 #endif
1601 err = vfs_rename(fdir, odentry, tdir, ndentry);
1602 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1603 err = nfsd_sync_dir(tdentry);
1604 if (!err)
1605 err = nfsd_sync_dir(fdentry);
1606 }
1607
1608 out_dput_new:
1609 dput(ndentry);
1610 out_dput_old:
1611 dput(odentry);
1612 out_nfserr:
1613 if (err)
1614 err = nfserrno(err);
1615
1616 /* we cannot reply on fh_unlock on the two filehandles,
1617 * as that would do the wrong thing if the two directories
1618 * were the same, so again we do it by hand
1619 */
1620 fill_post_wcc(ffhp);
1621 fill_post_wcc(tfhp);
1622 unlock_rename(tdentry, fdentry);
1623 ffhp->fh_locked = tfhp->fh_locked = 0;
1624
1625 out:
1626 return err;
1627 }
1628
1629 /*
1630 * Unlink a file or directory
1631 * N.B. After this call fhp needs an fh_put
1632 */
1633 int
1634 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1635 char *fname, int flen)
1636 {
1637 struct dentry *dentry, *rdentry;
1638 struct inode *dirp;
1639 int err;
1640
1641 err = nfserr_acces;
1642 if (!flen || isdotent(fname, flen))
1643 goto out;
1644 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1645 if (err)
1646 goto out;
1647
1648 fh_lock(fhp);
1649 dentry = fhp->fh_dentry;
1650 dirp = dentry->d_inode;
1651
1652 rdentry = lookup_one_len(fname, dentry, flen);
1653 err = PTR_ERR(rdentry);
1654 if (IS_ERR(rdentry))
1655 goto out_nfserr;
1656
1657 if (!rdentry->d_inode) {
1658 dput(rdentry);
1659 err = nfserr_noent;
1660 goto out;
1661 }
1662
1663 if (!type)
1664 type = rdentry->d_inode->i_mode & S_IFMT;
1665
1666 if (type != S_IFDIR) { /* It's UNLINK */
1667 #ifdef MSNFS
1668 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1669 (atomic_read(&rdentry->d_count) > 1)) {
1670 err = -EPERM;
1671 } else
1672 #endif
1673 err = vfs_unlink(dirp, rdentry);
1674 } else { /* It's RMDIR */
1675 err = vfs_rmdir(dirp, rdentry);
1676 }
1677
1678 dput(rdentry);
1679
1680 if (err == 0 &&
1681 EX_ISSYNC(fhp->fh_export))
1682 err = nfsd_sync_dir(dentry);
1683
1684 out_nfserr:
1685 err = nfserrno(err);
1686 out:
1687 return err;
1688 }
1689
1690 /*
1691 * Read entries from a directory.
1692 * The NFSv3/4 verifier we ignore for now.
1693 */
1694 int
1695 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1696 struct readdir_cd *cdp, encode_dent_fn func)
1697 {
1698 int err;
1699 struct file *file;
1700 loff_t offset = *offsetp;
1701
1702 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1703 if (err)
1704 goto out;
1705
1706 offset = vfs_llseek(file, offset, 0);
1707 if (offset < 0) {
1708 err = nfserrno((int)offset);
1709 goto out_close;
1710 }
1711
1712 /*
1713 * Read the directory entries. This silly loop is necessary because
1714 * readdir() is not guaranteed to fill up the entire buffer, but
1715 * may choose to do less.
1716 */
1717
1718 do {
1719 cdp->err = nfserr_eof; /* will be cleared on successful read */
1720 err = vfs_readdir(file, (filldir_t) func, cdp);
1721 } while (err >=0 && cdp->err == nfs_ok);
1722 if (err)
1723 err = nfserrno(err);
1724 else
1725 err = cdp->err;
1726 *offsetp = vfs_llseek(file, 0, 1);
1727
1728 if (err == nfserr_eof || err == nfserr_toosmall)
1729 err = nfs_ok; /* can still be found in ->err */
1730 out_close:
1731 nfsd_close(file);
1732 out:
1733 return err;
1734 }
1735
1736 /*
1737 * Get file system stats
1738 * N.B. After this call fhp needs an fh_put
1739 */
1740 int
1741 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1742 {
1743 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1744 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1745 err = nfserr_io;
1746 return err;
1747 }
1748
1749 /*
1750 * Check for a user's access permissions to this inode.
1751 */
1752 int
1753 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1754 {
1755 struct inode *inode = dentry->d_inode;
1756 int err;
1757
1758 if (acc == MAY_NOP)
1759 return 0;
1760 #if 0
1761 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1762 acc,
1763 (acc & MAY_READ)? " read" : "",
1764 (acc & MAY_WRITE)? " write" : "",
1765 (acc & MAY_EXEC)? " exec" : "",
1766 (acc & MAY_SATTR)? " sattr" : "",
1767 (acc & MAY_TRUNC)? " trunc" : "",
1768 (acc & MAY_LOCK)? " lock" : "",
1769 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1770 inode->i_mode,
1771 IS_IMMUTABLE(inode)? " immut" : "",
1772 IS_APPEND(inode)? " append" : "",
1773 IS_RDONLY(inode)? " ro" : "");
1774 dprintk(" owner %d/%d user %d/%d\n",
1775 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1776 #endif
1777
1778 /* Normally we reject any write/sattr etc access on a read-only file
1779 * system. But if it is IRIX doing check on write-access for a
1780 * device special file, we ignore rofs.
1781 */
1782 if (!(acc & MAY_LOCAL_ACCESS))
1783 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1784 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1785 return nfserr_rofs;
1786 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1787 return nfserr_perm;
1788 }
1789 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1790 return nfserr_perm;
1791
1792 if (acc & MAY_LOCK) {
1793 /* If we cannot rely on authentication in NLM requests,
1794 * just allow locks, otherwise require read permission, or
1795 * ownership
1796 */
1797 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1798 return 0;
1799 else
1800 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1801 }
1802 /*
1803 * The file owner always gets access permission for accesses that
1804 * would normally be checked at open time. This is to make
1805 * file access work even when the client has done a fchmod(fd, 0).
1806 *
1807 * However, `cp foo bar' should fail nevertheless when bar is
1808 * readonly. A sensible way to do this might be to reject all
1809 * attempts to truncate a read-only file, because a creat() call
1810 * always implies file truncation.
1811 * ... but this isn't really fair. A process may reasonably call
1812 * ftruncate on an open file descriptor on a file with perm 000.
1813 * We must trust the client to do permission checking - using "ACCESS"
1814 * with NFSv3.
1815 */
1816 if ((acc & MAY_OWNER_OVERRIDE) &&
1817 inode->i_uid == current->fsuid)
1818 return 0;
1819
1820 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1821
1822 /* Allow read access to binaries even when mode 111 */
1823 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1824 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1825 err = permission(inode, MAY_EXEC, NULL);
1826
1827 return err? nfserrno(err) : 0;
1828 }
1829
1830 void
1831 nfsd_racache_shutdown(void)
1832 {
1833 if (!raparm_cache)
1834 return;
1835 dprintk("nfsd: freeing readahead buffers.\n");
1836 kfree(raparml);
1837 raparm_cache = raparml = NULL;
1838 }
1839 /*
1840 * Initialize readahead param cache
1841 */
1842 int
1843 nfsd_racache_init(int cache_size)
1844 {
1845 int i;
1846
1847 if (raparm_cache)
1848 return 0;
1849 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1850
1851 if (raparml != NULL) {
1852 dprintk("nfsd: allocating %d readahead buffers.\n",
1853 cache_size);
1854 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1855 for (i = 0; i < cache_size - 1; i++) {
1856 raparml[i].p_next = raparml + i + 1;
1857 }
1858 raparm_cache = raparml;
1859 } else {
1860 printk(KERN_WARNING
1861 "nfsd: Could not allocate memory read-ahead cache.\n");
1862 return -ENOMEM;
1863 }
1864 nfsdstats.ra_size = cache_size;
1865 return 0;
1866 }
1867
1868 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1869 struct posix_acl *
1870 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1871 {
1872 struct inode *inode = fhp->fh_dentry->d_inode;
1873 char *name;
1874 void *value = NULL;
1875 ssize_t size;
1876 struct posix_acl *acl;
1877
1878 if (!IS_POSIXACL(inode))
1879 return ERR_PTR(-EOPNOTSUPP);
1880
1881 switch (type) {
1882 case ACL_TYPE_ACCESS:
1883 name = POSIX_ACL_XATTR_ACCESS;
1884 break;
1885 case ACL_TYPE_DEFAULT:
1886 name = POSIX_ACL_XATTR_DEFAULT;
1887 break;
1888 default:
1889 return ERR_PTR(-EOPNOTSUPP);
1890 }
1891
1892 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1893 if (size < 0)
1894 return ERR_PTR(size);
1895
1896 acl = posix_acl_from_xattr(value, size);
1897 kfree(value);
1898 return acl;
1899 }
1900
1901 int
1902 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1903 {
1904 struct inode *inode = fhp->fh_dentry->d_inode;
1905 char *name;
1906 void *value = NULL;
1907 size_t size;
1908 int error;
1909
1910 if (!IS_POSIXACL(inode) || !inode->i_op ||
1911 !inode->i_op->setxattr || !inode->i_op->removexattr)
1912 return -EOPNOTSUPP;
1913 switch(type) {
1914 case ACL_TYPE_ACCESS:
1915 name = POSIX_ACL_XATTR_ACCESS;
1916 break;
1917 case ACL_TYPE_DEFAULT:
1918 name = POSIX_ACL_XATTR_DEFAULT;
1919 break;
1920 default:
1921 return -EOPNOTSUPP;
1922 }
1923
1924 if (acl && acl->a_count) {
1925 size = posix_acl_xattr_size(acl->a_count);
1926 value = kmalloc(size, GFP_KERNEL);
1927 if (!value)
1928 return -ENOMEM;
1929 size = posix_acl_to_xattr(acl, value, size);
1930 if (size < 0) {
1931 error = size;
1932 goto getout;
1933 }
1934 } else
1935 size = 0;
1936
1937 if (size)
1938 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1939 else {
1940 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1941 error = 0;
1942 else {
1943 error = vfs_removexattr(fhp->fh_dentry, name);
1944 if (error == -ENODATA)
1945 error = 0;
1946 }
1947 }
1948
1949 getout:
1950 kfree(value);
1951 return error;
1952 }
1953 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
Linux with added FPC-III support