projects / deliverable / linux.git / blob
? search:


eef0576a77857577805e053d428ebe13bd6ce9a6
[deliverable/linux.git] / fs / nfsd / vfs.c
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 struct file_operations *fop)
710 {
711 struct inode *inode = dp->d_inode;
712 int (*fsync) (struct file *, struct dentry *, int);
713 int err = nfs_ok;
714
715 filemap_fdatawrite(inode->i_mapping);
716 if (fop && (fsync = fop->fsync))
717 err=fsync(filp, dp, 0);
718 filemap_fdatawait(inode->i_mapping);
719
720 return nfserrno(err);
721 }
722
723
724 static int
725 nfsd_sync(struct file *filp)
726 {
727 int err;
728 struct inode *inode = filp->f_dentry->d_inode;
729 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
730 mutex_lock(&inode->i_mutex);
731 err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
732 mutex_unlock(&inode->i_mutex);
733
734 return err;
735 }
736
737 void
738 nfsd_sync_dir(struct dentry *dp)
739 {
740 nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
741 }
742
743 /*
744 * Obtain the readahead parameters for the file
745 * specified by (dev, ino).
746 */
747 static DEFINE_SPINLOCK(ra_lock);
748
749 static inline struct raparms *
750 nfsd_get_raparms(dev_t dev, ino_t ino)
751 {
752 struct raparms *ra, **rap, **frap = NULL;
753 int depth = 0;
754
755 spin_lock(&ra_lock);
756 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
757 if (ra->p_ino == ino && ra->p_dev == dev)
758 goto found;
759 depth++;
760 if (ra->p_count == 0)
761 frap = rap;
762 }
763 depth = nfsdstats.ra_size*11/10;
764 if (!frap) {
765 spin_unlock(&ra_lock);
766 return NULL;
767 }
768 rap = frap;
769 ra = *frap;
770 ra->p_dev = dev;
771 ra->p_ino = ino;
772 ra->p_set = 0;
773 found:
774 if (rap != &raparm_cache) {
775 *rap = ra->p_next;
776 ra->p_next = raparm_cache;
777 raparm_cache = ra;
778 }
779 ra->p_count++;
780 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
781 spin_unlock(&ra_lock);
782 return ra;
783 }
784
785 /*
786 * Grab and keep cached pages assosiated with a file in the svc_rqst
787 * so that they can be passed to the netowork sendmsg/sendpage routines
788 * directrly. They will be released after the sending has completed.
789 */
790 static int
791 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
792 {
793 unsigned long count = desc->count;
794 struct svc_rqst *rqstp = desc->arg.data;
795
796 if (size > count)
797 size = count;
798
799 if (rqstp->rq_res.page_len == 0) {
800 get_page(page);
801 rqstp->rq_respages[rqstp->rq_resused++] = page;
802 rqstp->rq_res.page_base = offset;
803 rqstp->rq_res.page_len = size;
804 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
805 get_page(page);
806 rqstp->rq_respages[rqstp->rq_resused++] = page;
807 rqstp->rq_res.page_len += size;
808 } else {
809 rqstp->rq_res.page_len += size;
810 }
811
812 desc->count = count - size;
813 desc->written += size;
814 return size;
815 }
816
817 static inline int
818 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
819 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
820 {
821 struct inode *inode;
822 struct raparms *ra;
823 mm_segment_t oldfs;
824 int err;
825
826 err = nfserr_perm;
827 inode = file->f_dentry->d_inode;
828 #ifdef MSNFS
829 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
830 (!lock_may_read(inode, offset, *count)))
831 goto out;
832 #endif
833
834 /* Get readahead parameters */
835 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
836
837 if (ra && ra->p_set)
838 file->f_ra = ra->p_ra;
839
840 if (file->f_op->sendfile) {
841 svc_pushback_unused_pages(rqstp);
842 err = file->f_op->sendfile(file, &offset, *count,
843 nfsd_read_actor, rqstp);
844 } else {
845 oldfs = get_fs();
846 set_fs(KERNEL_DS);
847 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
848 set_fs(oldfs);
849 }
850
851 /* Write back readahead params */
852 if (ra) {
853 spin_lock(&ra_lock);
854 ra->p_ra = file->f_ra;
855 ra->p_set = 1;
856 ra->p_count--;
857 spin_unlock(&ra_lock);
858 }
859
860 if (err >= 0) {
861 nfsdstats.io_read += err;
862 *count = err;
863 err = 0;
864 fsnotify_access(file->f_dentry);
865 } else
866 err = nfserrno(err);
867 out:
868 return err;
869 }
870
871 static void kill_suid(struct dentry *dentry)
872 {
873 struct iattr ia;
874 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
875
876 mutex_lock(&dentry->d_inode->i_mutex);
877 notify_change(dentry, &ia);
878 mutex_unlock(&dentry->d_inode->i_mutex);
879 }
880
881 static inline int
882 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
883 loff_t offset, struct kvec *vec, int vlen,
884 unsigned long cnt, int *stablep)
885 {
886 struct svc_export *exp;
887 struct dentry *dentry;
888 struct inode *inode;
889 mm_segment_t oldfs;
890 int err = 0;
891 int stable = *stablep;
892
893 err = nfserr_perm;
894
895 #ifdef MSNFS
896 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
897 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
898 goto out;
899 #endif
900
901 dentry = file->f_dentry;
902 inode = dentry->d_inode;
903 exp = fhp->fh_export;
904
905 /*
906 * Request sync writes if
907 * - the sync export option has been set, or
908 * - the client requested O_SYNC behavior (NFSv3 feature).
909 * - The file system doesn't support fsync().
910 * When gathered writes have been configured for this volume,
911 * flushing the data to disk is handled separately below.
912 */
913
914 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
915 stable = 2;
916 *stablep = 2; /* FILE_SYNC */
917 }
918
919 if (!EX_ISSYNC(exp))
920 stable = 0;
921 if (stable && !EX_WGATHER(exp))
922 file->f_flags |= O_SYNC;
923
924 /* Write the data. */
925 oldfs = get_fs(); set_fs(KERNEL_DS);
926 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
927 set_fs(oldfs);
928 if (err >= 0) {
929 nfsdstats.io_write += cnt;
930 fsnotify_modify(file->f_dentry);
931 }
932
933 /* clear setuid/setgid flag after write */
934 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
935 kill_suid(dentry);
936
937 if (err >= 0 && stable) {
938 static ino_t last_ino;
939 static dev_t last_dev;
940
941 /*
942 * Gathered writes: If another process is currently
943 * writing to the file, there's a high chance
944 * this is another nfsd (triggered by a bulk write
945 * from a client's biod). Rather than syncing the
946 * file with each write request, we sleep for 10 msec.
947 *
948 * I don't know if this roughly approximates
949 * C. Juszak's idea of gathered writes, but it's a
950 * nice and simple solution (IMHO), and it seems to
951 * work:-)
952 */
953 if (EX_WGATHER(exp)) {
954 if (atomic_read(&inode->i_writecount) > 1
955 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
956 dprintk("nfsd: write defer %d\n", current->pid);
957 msleep(10);
958 dprintk("nfsd: write resume %d\n", current->pid);
959 }
960
961 if (inode->i_state & I_DIRTY) {
962 dprintk("nfsd: write sync %d\n", current->pid);
963 err=nfsd_sync(file);
964 }
965 #if 0
966 wake_up(&inode->i_wait);
967 #endif
968 }
969 last_ino = inode->i_ino;
970 last_dev = inode->i_sb->s_dev;
971 }
972
973 dprintk("nfsd: write complete err=%d\n", err);
974 if (err >= 0)
975 err = 0;
976 else
977 err = nfserrno(err);
978 out:
979 return err;
980 }
981
982 /*
983 * Read data from a file. count must contain the requested read count
984 * on entry. On return, *count contains the number of bytes actually read.
985 * N.B. After this call fhp needs an fh_put
986 */
987 int
988 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
989 loff_t offset, struct kvec *vec, int vlen,
990 unsigned long *count)
991 {
992 int err;
993
994 if (file) {
995 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
996 MAY_READ|MAY_OWNER_OVERRIDE);
997 if (err)
998 goto out;
999 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1000 } else {
1001 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1002 if (err)
1003 goto out;
1004 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1005 nfsd_close(file);
1006 }
1007 out:
1008 return err;
1009 }
1010
1011 /*
1012 * Write data to a file.
1013 * The stable flag requests synchronous writes.
1014 * N.B. After this call fhp needs an fh_put
1015 */
1016 int
1017 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1018 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1019 int *stablep)
1020 {
1021 int err = 0;
1022
1023 if (file) {
1024 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1025 MAY_WRITE|MAY_OWNER_OVERRIDE);
1026 if (err)
1027 goto out;
1028 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1029 stablep);
1030 } else {
1031 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1032 if (err)
1033 goto out;
1034
1035 if (cnt)
1036 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1037 cnt, stablep);
1038 nfsd_close(file);
1039 }
1040 out:
1041 return err;
1042 }
1043
1044 #ifdef CONFIG_NFSD_V3
1045 /*
1046 * Commit all pending writes to stable storage.
1047 * Strictly speaking, we could sync just the indicated file region here,
1048 * but there's currently no way we can ask the VFS to do so.
1049 *
1050 * Unfortunately we cannot lock the file to make sure we return full WCC
1051 * data to the client, as locking happens lower down in the filesystem.
1052 */
1053 int
1054 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1055 loff_t offset, unsigned long count)
1056 {
1057 struct file *file;
1058 int err;
1059
1060 if ((u64)count > ~(u64)offset)
1061 return nfserr_inval;
1062
1063 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1064 return err;
1065 if (EX_ISSYNC(fhp->fh_export)) {
1066 if (file->f_op && file->f_op->fsync) {
1067 err = nfsd_sync(file);
1068 } else {
1069 err = nfserr_notsupp;
1070 }
1071 }
1072
1073 nfsd_close(file);
1074 return err;
1075 }
1076 #endif /* CONFIG_NFSD_V3 */
1077
1078 /*
1079 * Create a file (regular, directory, device, fifo); UNIX sockets
1080 * not yet implemented.
1081 * If the response fh has been verified, the parent directory should
1082 * already be locked. Note that the parent directory is left locked.
1083 *
1084 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1085 */
1086 int
1087 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1088 char *fname, int flen, struct iattr *iap,
1089 int type, dev_t rdev, struct svc_fh *resfhp)
1090 {
1091 struct dentry *dentry, *dchild = NULL;
1092 struct inode *dirp;
1093 int err;
1094
1095 err = nfserr_perm;
1096 if (!flen)
1097 goto out;
1098 err = nfserr_exist;
1099 if (isdotent(fname, flen))
1100 goto out;
1101
1102 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1103 if (err)
1104 goto out;
1105
1106 dentry = fhp->fh_dentry;
1107 dirp = dentry->d_inode;
1108
1109 err = nfserr_notdir;
1110 if(!dirp->i_op || !dirp->i_op->lookup)
1111 goto out;
1112 /*
1113 * Check whether the response file handle has been verified yet.
1114 * If it has, the parent directory should already be locked.
1115 */
1116 if (!resfhp->fh_dentry) {
1117 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1118 fh_lock(fhp);
1119 dchild = lookup_one_len(fname, dentry, flen);
1120 err = PTR_ERR(dchild);
1121 if (IS_ERR(dchild))
1122 goto out_nfserr;
1123 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1124 if (err)
1125 goto out;
1126 } else {
1127 /* called from nfsd_proc_create */
1128 dchild = dget(resfhp->fh_dentry);
1129 if (!fhp->fh_locked) {
1130 /* not actually possible */
1131 printk(KERN_ERR
1132 "nfsd_create: parent %s/%s not locked!\n",
1133 dentry->d_parent->d_name.name,
1134 dentry->d_name.name);
1135 err = -EIO;
1136 goto out;
1137 }
1138 }
1139 /*
1140 * Make sure the child dentry is still negative ...
1141 */
1142 err = nfserr_exist;
1143 if (dchild->d_inode) {
1144 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1145 dentry->d_name.name, dchild->d_name.name);
1146 goto out;
1147 }
1148
1149 if (!(iap->ia_valid & ATTR_MODE))
1150 iap->ia_mode = 0;
1151 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1152
1153 /*
1154 * Get the dir op function pointer.
1155 */
1156 err = nfserr_perm;
1157 switch (type) {
1158 case S_IFREG:
1159 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1160 break;
1161 case S_IFDIR:
1162 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1163 break;
1164 case S_IFCHR:
1165 case S_IFBLK:
1166 case S_IFIFO:
1167 case S_IFSOCK:
1168 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1169 break;
1170 default:
1171 printk("nfsd: bad file type %o in nfsd_create\n", type);
1172 err = -EINVAL;
1173 }
1174 if (err < 0)
1175 goto out_nfserr;
1176
1177 if (EX_ISSYNC(fhp->fh_export)) {
1178 nfsd_sync_dir(dentry);
1179 write_inode_now(dchild->d_inode, 1);
1180 }
1181
1182
1183 /* Set file attributes. Mode has already been set and
1184 * setting uid/gid works only for root. Irix appears to
1185 * send along the gid when it tries to implement setgid
1186 * directories via NFS.
1187 */
1188 err = 0;
1189 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
1190 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1191 /*
1192 * Update the file handle to get the new inode info.
1193 */
1194 if (!err)
1195 err = fh_update(resfhp);
1196 out:
1197 if (dchild && !IS_ERR(dchild))
1198 dput(dchild);
1199 return err;
1200
1201 out_nfserr:
1202 err = nfserrno(err);
1203 goto out;
1204 }
1205
1206 #ifdef CONFIG_NFSD_V3
1207 /*
1208 * NFSv3 version of nfsd_create
1209 */
1210 int
1211 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1212 char *fname, int flen, struct iattr *iap,
1213 struct svc_fh *resfhp, int createmode, u32 *verifier,
1214 int *truncp)
1215 {
1216 struct dentry *dentry, *dchild = NULL;
1217 struct inode *dirp;
1218 int err;
1219 __u32 v_mtime=0, v_atime=0;
1220 int v_mode=0;
1221
1222 err = nfserr_perm;
1223 if (!flen)
1224 goto out;
1225 err = nfserr_exist;
1226 if (isdotent(fname, flen))
1227 goto out;
1228 if (!(iap->ia_valid & ATTR_MODE))
1229 iap->ia_mode = 0;
1230 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1231 if (err)
1232 goto out;
1233
1234 dentry = fhp->fh_dentry;
1235 dirp = dentry->d_inode;
1236
1237 /* Get all the sanity checks out of the way before
1238 * we lock the parent. */
1239 err = nfserr_notdir;
1240 if(!dirp->i_op || !dirp->i_op->lookup)
1241 goto out;
1242 fh_lock(fhp);
1243
1244 /*
1245 * Compose the response file handle.
1246 */
1247 dchild = lookup_one_len(fname, dentry, flen);
1248 err = PTR_ERR(dchild);
1249 if (IS_ERR(dchild))
1250 goto out_nfserr;
1251
1252 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1253 if (err)
1254 goto out;
1255
1256 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1257 /* while the verifier would fit in mtime+atime,
1258 * solaris7 gets confused (bugid 4218508) if these have
1259 * the high bit set, so we use the mode as well
1260 */
1261 v_mtime = verifier[0]&0x7fffffff;
1262 v_atime = verifier[1]&0x7fffffff;
1263 v_mode = S_IFREG
1264 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1265 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1266 ;
1267 }
1268
1269 if (dchild->d_inode) {
1270 err = 0;
1271
1272 switch (createmode) {
1273 case NFS3_CREATE_UNCHECKED:
1274 if (! S_ISREG(dchild->d_inode->i_mode))
1275 err = nfserr_exist;
1276 else if (truncp) {
1277 /* in nfsv4, we need to treat this case a little
1278 * differently. we don't want to truncate the
1279 * file now; this would be wrong if the OPEN
1280 * fails for some other reason. furthermore,
1281 * if the size is nonzero, we should ignore it
1282 * according to spec!
1283 */
1284 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1285 }
1286 else {
1287 iap->ia_valid &= ATTR_SIZE;
1288 goto set_attr;
1289 }
1290 break;
1291 case NFS3_CREATE_EXCLUSIVE:
1292 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1293 && dchild->d_inode->i_atime.tv_sec == v_atime
1294 && dchild->d_inode->i_mode == v_mode
1295 && dchild->d_inode->i_size == 0 )
1296 break;
1297 /* fallthru */
1298 case NFS3_CREATE_GUARDED:
1299 err = nfserr_exist;
1300 }
1301 goto out;
1302 }
1303
1304 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1305 if (err < 0)
1306 goto out_nfserr;
1307
1308 if (EX_ISSYNC(fhp->fh_export)) {
1309 nfsd_sync_dir(dentry);
1310 /* setattr will sync the child (or not) */
1311 }
1312
1313 /*
1314 * Update the filehandle to get the new inode info.
1315 */
1316 err = fh_update(resfhp);
1317 if (err)
1318 goto out;
1319
1320 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1321 /* Cram the verifier into atime/mtime/mode */
1322 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1323 | ATTR_MTIME_SET|ATTR_ATIME_SET
1324 | ATTR_MODE;
1325 /* XXX someone who knows this better please fix it for nsec */
1326 iap->ia_mtime.tv_sec = v_mtime;
1327 iap->ia_atime.tv_sec = v_atime;
1328 iap->ia_mtime.tv_nsec = 0;
1329 iap->ia_atime.tv_nsec = 0;
1330 iap->ia_mode = v_mode;
1331 }
1332
1333 /* Set file attributes.
1334 * Mode has already been set but we might need to reset it
1335 * for CREATE_EXCLUSIVE
1336 * Irix appears to send along the gid when it tries to
1337 * implement setgid directories via NFS. Clear out all that cruft.
1338 */
1339 set_attr:
1340 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1341 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1342
1343 out:
1344 fh_unlock(fhp);
1345 if (dchild && !IS_ERR(dchild))
1346 dput(dchild);
1347 return err;
1348
1349 out_nfserr:
1350 err = nfserrno(err);
1351 goto out;
1352 }
1353 #endif /* CONFIG_NFSD_V3 */
1354
1355 /*
1356 * Read a symlink. On entry, *lenp must contain the maximum path length that
1357 * fits into the buffer. On return, it contains the true length.
1358 * N.B. After this call fhp needs an fh_put
1359 */
1360 int
1361 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1362 {
1363 struct dentry *dentry;
1364 struct inode *inode;
1365 mm_segment_t oldfs;
1366 int err;
1367
1368 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1369 if (err)
1370 goto out;
1371
1372 dentry = fhp->fh_dentry;
1373 inode = dentry->d_inode;
1374
1375 err = nfserr_inval;
1376 if (!inode->i_op || !inode->i_op->readlink)
1377 goto out;
1378
1379 touch_atime(fhp->fh_export->ex_mnt, dentry);
1380 /* N.B. Why does this call need a get_fs()??
1381 * Remove the set_fs and watch the fireworks:-) --okir
1382 */
1383
1384 oldfs = get_fs(); set_fs(KERNEL_DS);
1385 err = inode->i_op->readlink(dentry, buf, *lenp);
1386 set_fs(oldfs);
1387
1388 if (err < 0)
1389 goto out_nfserr;
1390 *lenp = err;
1391 err = 0;
1392 out:
1393 return err;
1394
1395 out_nfserr:
1396 err = nfserrno(err);
1397 goto out;
1398 }
1399
1400 /*
1401 * Create a symlink and look up its inode
1402 * N.B. After this call _both_ fhp and resfhp need an fh_put
1403 */
1404 int
1405 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1406 char *fname, int flen,
1407 char *path, int plen,
1408 struct svc_fh *resfhp,
1409 struct iattr *iap)
1410 {
1411 struct dentry *dentry, *dnew;
1412 int err, cerr;
1413 umode_t mode;
1414
1415 err = nfserr_noent;
1416 if (!flen || !plen)
1417 goto out;
1418 err = nfserr_exist;
1419 if (isdotent(fname, flen))
1420 goto out;
1421
1422 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1423 if (err)
1424 goto out;
1425 fh_lock(fhp);
1426 dentry = fhp->fh_dentry;
1427 dnew = lookup_one_len(fname, dentry, flen);
1428 err = PTR_ERR(dnew);
1429 if (IS_ERR(dnew))
1430 goto out_nfserr;
1431
1432 mode = S_IALLUGO;
1433 /* Only the MODE ATTRibute is even vaguely meaningful */
1434 if (iap && (iap->ia_valid & ATTR_MODE))
1435 mode = iap->ia_mode & S_IALLUGO;
1436
1437 if (unlikely(path[plen] != 0)) {
1438 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1439 if (path_alloced == NULL)
1440 err = -ENOMEM;
1441 else {
1442 strncpy(path_alloced, path, plen);
1443 path_alloced[plen] = 0;
1444 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1445 kfree(path_alloced);
1446 }
1447 } else
1448 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1449
1450 if (!err) {
1451 if (EX_ISSYNC(fhp->fh_export))
1452 nfsd_sync_dir(dentry);
1453 } else
1454 err = nfserrno(err);
1455 fh_unlock(fhp);
1456
1457 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1458 dput(dnew);
1459 if (err==0) err = cerr;
1460 out:
1461 return err;
1462
1463 out_nfserr:
1464 err = nfserrno(err);
1465 goto out;
1466 }
1467
1468 /*
1469 * Create a hardlink
1470 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1471 */
1472 int
1473 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1474 char *name, int len, struct svc_fh *tfhp)
1475 {
1476 struct dentry *ddir, *dnew, *dold;
1477 struct inode *dirp, *dest;
1478 int err;
1479
1480 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1481 if (err)
1482 goto out;
1483 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1484 if (err)
1485 goto out;
1486
1487 err = nfserr_perm;
1488 if (!len)
1489 goto out;
1490 err = nfserr_exist;
1491 if (isdotent(name, len))
1492 goto out;
1493
1494 fh_lock(ffhp);
1495 ddir = ffhp->fh_dentry;
1496 dirp = ddir->d_inode;
1497
1498 dnew = lookup_one_len(name, ddir, len);
1499 err = PTR_ERR(dnew);
1500 if (IS_ERR(dnew))
1501 goto out_nfserr;
1502
1503 dold = tfhp->fh_dentry;
1504 dest = dold->d_inode;
1505
1506 err = vfs_link(dold, dirp, dnew);
1507 if (!err) {
1508 if (EX_ISSYNC(ffhp->fh_export)) {
1509 nfsd_sync_dir(ddir);
1510 write_inode_now(dest, 1);
1511 }
1512 } else {
1513 if (err == -EXDEV && rqstp->rq_vers == 2)
1514 err = nfserr_acces;
1515 else
1516 err = nfserrno(err);
1517 }
1518
1519 fh_unlock(ffhp);
1520 dput(dnew);
1521 out:
1522 return err;
1523
1524 out_nfserr:
1525 err = nfserrno(err);
1526 goto out;
1527 }
1528
1529 /*
1530 * Rename a file
1531 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1532 */
1533 int
1534 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1535 struct svc_fh *tfhp, char *tname, int tlen)
1536 {
1537 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1538 struct inode *fdir, *tdir;
1539 int err;
1540
1541 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1542 if (err)
1543 goto out;
1544 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1545 if (err)
1546 goto out;
1547
1548 fdentry = ffhp->fh_dentry;
1549 fdir = fdentry->d_inode;
1550
1551 tdentry = tfhp->fh_dentry;
1552 tdir = tdentry->d_inode;
1553
1554 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1555 if (fdir->i_sb != tdir->i_sb)
1556 goto out;
1557
1558 err = nfserr_perm;
1559 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1560 goto out;
1561
1562 /* cannot use fh_lock as we need deadlock protective ordering
1563 * so do it by hand */
1564 trap = lock_rename(tdentry, fdentry);
1565 ffhp->fh_locked = tfhp->fh_locked = 1;
1566 fill_pre_wcc(ffhp);
1567 fill_pre_wcc(tfhp);
1568
1569 odentry = lookup_one_len(fname, fdentry, flen);
1570 err = PTR_ERR(odentry);
1571 if (IS_ERR(odentry))
1572 goto out_nfserr;
1573
1574 err = -ENOENT;
1575 if (!odentry->d_inode)
1576 goto out_dput_old;
1577 err = -EINVAL;
1578 if (odentry == trap)
1579 goto out_dput_old;
1580
1581 ndentry = lookup_one_len(tname, tdentry, tlen);
1582 err = PTR_ERR(ndentry);
1583 if (IS_ERR(ndentry))
1584 goto out_dput_old;
1585 err = -ENOTEMPTY;
1586 if (ndentry == trap)
1587 goto out_dput_new;
1588
1589 #ifdef MSNFS
1590 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1591 ((atomic_read(&odentry->d_count) > 1)
1592 || (atomic_read(&ndentry->d_count) > 1))) {
1593 err = nfserr_perm;
1594 } else
1595 #endif
1596 err = vfs_rename(fdir, odentry, tdir, ndentry);
1597 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1598 nfsd_sync_dir(tdentry);
1599 nfsd_sync_dir(fdentry);
1600 }
1601
1602 out_dput_new:
1603 dput(ndentry);
1604 out_dput_old:
1605 dput(odentry);
1606 out_nfserr:
1607 if (err)
1608 err = nfserrno(err);
1609
1610 /* we cannot reply on fh_unlock on the two filehandles,
1611 * as that would do the wrong thing if the two directories
1612 * were the same, so again we do it by hand
1613 */
1614 fill_post_wcc(ffhp);
1615 fill_post_wcc(tfhp);
1616 unlock_rename(tdentry, fdentry);
1617 ffhp->fh_locked = tfhp->fh_locked = 0;
1618
1619 out:
1620 return err;
1621 }
1622
1623 /*
1624 * Unlink a file or directory
1625 * N.B. After this call fhp needs an fh_put
1626 */
1627 int
1628 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1629 char *fname, int flen)
1630 {
1631 struct dentry *dentry, *rdentry;
1632 struct inode *dirp;
1633 int err;
1634
1635 err = nfserr_acces;
1636 if (!flen || isdotent(fname, flen))
1637 goto out;
1638 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1639 if (err)
1640 goto out;
1641
1642 fh_lock(fhp);
1643 dentry = fhp->fh_dentry;
1644 dirp = dentry->d_inode;
1645
1646 rdentry = lookup_one_len(fname, dentry, flen);
1647 err = PTR_ERR(rdentry);
1648 if (IS_ERR(rdentry))
1649 goto out_nfserr;
1650
1651 if (!rdentry->d_inode) {
1652 dput(rdentry);
1653 err = nfserr_noent;
1654 goto out;
1655 }
1656
1657 if (!type)
1658 type = rdentry->d_inode->i_mode & S_IFMT;
1659
1660 if (type != S_IFDIR) { /* It's UNLINK */
1661 #ifdef MSNFS
1662 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1663 (atomic_read(&rdentry->d_count) > 1)) {
1664 err = nfserr_perm;
1665 } else
1666 #endif
1667 err = vfs_unlink(dirp, rdentry);
1668 } else { /* It's RMDIR */
1669 err = vfs_rmdir(dirp, rdentry);
1670 }
1671
1672 dput(rdentry);
1673
1674 if (err)
1675 goto out_nfserr;
1676 if (EX_ISSYNC(fhp->fh_export))
1677 nfsd_sync_dir(dentry);
1678
1679 out:
1680 return err;
1681
1682 out_nfserr:
1683 err = nfserrno(err);
1684 goto out;
1685 }
1686
1687 /*
1688 * Read entries from a directory.
1689 * The NFSv3/4 verifier we ignore for now.
1690 */
1691 int
1692 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1693 struct readdir_cd *cdp, encode_dent_fn func)
1694 {
1695 int err;
1696 struct file *file;
1697 loff_t offset = *offsetp;
1698
1699 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1700 if (err)
1701 goto out;
1702
1703 offset = vfs_llseek(file, offset, 0);
1704 if (offset < 0) {
1705 err = nfserrno((int)offset);
1706 goto out_close;
1707 }
1708
1709 /*
1710 * Read the directory entries. This silly loop is necessary because
1711 * readdir() is not guaranteed to fill up the entire buffer, but
1712 * may choose to do less.
1713 */
1714
1715 do {
1716 cdp->err = nfserr_eof; /* will be cleared on successful read */
1717 err = vfs_readdir(file, (filldir_t) func, cdp);
1718 } while (err >=0 && cdp->err == nfs_ok);
1719 if (err)
1720 err = nfserrno(err);
1721 else
1722 err = cdp->err;
1723 *offsetp = vfs_llseek(file, 0, 1);
1724
1725 if (err == nfserr_eof || err == nfserr_toosmall)
1726 err = nfs_ok; /* can still be found in ->err */
1727 out_close:
1728 nfsd_close(file);
1729 out:
1730 return err;
1731 }
1732
1733 /*
1734 * Get file system stats
1735 * N.B. After this call fhp needs an fh_put
1736 */
1737 int
1738 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1739 {
1740 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1741 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1742 err = nfserr_io;
1743 return err;
1744 }
1745
1746 /*
1747 * Check for a user's access permissions to this inode.
1748 */
1749 int
1750 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1751 {
1752 struct inode *inode = dentry->d_inode;
1753 int err;
1754
1755 if (acc == MAY_NOP)
1756 return 0;
1757 #if 0
1758 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1759 acc,
1760 (acc & MAY_READ)? " read" : "",
1761 (acc & MAY_WRITE)? " write" : "",
1762 (acc & MAY_EXEC)? " exec" : "",
1763 (acc & MAY_SATTR)? " sattr" : "",
1764 (acc & MAY_TRUNC)? " trunc" : "",
1765 (acc & MAY_LOCK)? " lock" : "",
1766 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1767 inode->i_mode,
1768 IS_IMMUTABLE(inode)? " immut" : "",
1769 IS_APPEND(inode)? " append" : "",
1770 IS_RDONLY(inode)? " ro" : "");
1771 dprintk(" owner %d/%d user %d/%d\n",
1772 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1773 #endif
1774
1775 /* Normally we reject any write/sattr etc access on a read-only file
1776 * system. But if it is IRIX doing check on write-access for a
1777 * device special file, we ignore rofs.
1778 */
1779 if (!(acc & MAY_LOCAL_ACCESS))
1780 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1781 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1782 return nfserr_rofs;
1783 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1784 return nfserr_perm;
1785 }
1786 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1787 return nfserr_perm;
1788
1789 if (acc & MAY_LOCK) {
1790 /* If we cannot rely on authentication in NLM requests,
1791 * just allow locks, otherwise require read permission, or
1792 * ownership
1793 */
1794 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1795 return 0;
1796 else
1797 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1798 }
1799 /*
1800 * The file owner always gets access permission for accesses that
1801 * would normally be checked at open time. This is to make
1802 * file access work even when the client has done a fchmod(fd, 0).
1803 *
1804 * However, `cp foo bar' should fail nevertheless when bar is
1805 * readonly. A sensible way to do this might be to reject all
1806 * attempts to truncate a read-only file, because a creat() call
1807 * always implies file truncation.
1808 * ... but this isn't really fair. A process may reasonably call
1809 * ftruncate on an open file descriptor on a file with perm 000.
1810 * We must trust the client to do permission checking - using "ACCESS"
1811 * with NFSv3.
1812 */
1813 if ((acc & MAY_OWNER_OVERRIDE) &&
1814 inode->i_uid == current->fsuid)
1815 return 0;
1816
1817 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1818
1819 /* Allow read access to binaries even when mode 111 */
1820 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1821 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1822 err = permission(inode, MAY_EXEC, NULL);
1823
1824 return err? nfserrno(err) : 0;
1825 }
1826
1827 void
1828 nfsd_racache_shutdown(void)
1829 {
1830 if (!raparm_cache)
1831 return;
1832 dprintk("nfsd: freeing readahead buffers.\n");
1833 kfree(raparml);
1834 raparm_cache = raparml = NULL;
1835 }
1836 /*
1837 * Initialize readahead param cache
1838 */
1839 int
1840 nfsd_racache_init(int cache_size)
1841 {
1842 int i;
1843
1844 if (raparm_cache)
1845 return 0;
1846 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1847
1848 if (raparml != NULL) {
1849 dprintk("nfsd: allocating %d readahead buffers.\n",
1850 cache_size);
1851 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1852 for (i = 0; i < cache_size - 1; i++) {
1853 raparml[i].p_next = raparml + i + 1;
1854 }
1855 raparm_cache = raparml;
1856 } else {
1857 printk(KERN_WARNING
1858 "nfsd: Could not allocate memory read-ahead cache.\n");
1859 return -ENOMEM;
1860 }
1861 nfsdstats.ra_size = cache_size;
1862 return 0;
1863 }
1864
1865 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1866 struct posix_acl *
1867 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1868 {
1869 struct inode *inode = fhp->fh_dentry->d_inode;
1870 char *name;
1871 void *value = NULL;
1872 ssize_t size;
1873 struct posix_acl *acl;
1874
1875 if (!IS_POSIXACL(inode))
1876 return ERR_PTR(-EOPNOTSUPP);
1877
1878 switch (type) {
1879 case ACL_TYPE_ACCESS:
1880 name = POSIX_ACL_XATTR_ACCESS;
1881 break;
1882 case ACL_TYPE_DEFAULT:
1883 name = POSIX_ACL_XATTR_DEFAULT;
1884 break;
1885 default:
1886 return ERR_PTR(-EOPNOTSUPP);
1887 }
1888
1889 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1890 if (size < 0)
1891 return ERR_PTR(size);
1892
1893 acl = posix_acl_from_xattr(value, size);
1894 kfree(value);
1895 return acl;
1896 }
1897
1898 int
1899 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1900 {
1901 struct inode *inode = fhp->fh_dentry->d_inode;
1902 char *name;
1903 void *value = NULL;
1904 size_t size;
1905 int error;
1906
1907 if (!IS_POSIXACL(inode) || !inode->i_op ||
1908 !inode->i_op->setxattr || !inode->i_op->removexattr)
1909 return -EOPNOTSUPP;
1910 switch(type) {
1911 case ACL_TYPE_ACCESS:
1912 name = POSIX_ACL_XATTR_ACCESS;
1913 break;
1914 case ACL_TYPE_DEFAULT:
1915 name = POSIX_ACL_XATTR_DEFAULT;
1916 break;
1917 default:
1918 return -EOPNOTSUPP;
1919 }
1920
1921 if (acl && acl->a_count) {
1922 size = posix_acl_xattr_size(acl->a_count);
1923 value = kmalloc(size, GFP_KERNEL);
1924 if (!value)
1925 return -ENOMEM;
1926 size = posix_acl_to_xattr(acl, value, size);
1927 if (size < 0) {
1928 error = size;
1929 goto getout;
1930 }
1931 } else
1932 size = 0;
1933
1934 if (size)
1935 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1936 else {
1937 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1938 error = 0;
1939 else {
1940 error = vfs_removexattr(fhp->fh_dentry, name);
1941 if (error == -ENODATA)
1942 error = 0;
1943 }
1944 }
1945
1946 getout:
1947 kfree(value);
1948 return error;
1949 }
1950 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
Linux kernel - Deliverables
This page took 0.070445 seconds and 4 git commands to generate.