projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[PATCH] knfsd: Fix type mismatch with filldir_t used by nfsd
[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/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/fs.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/ext2_fs.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
34 #include <linux/in.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
41 #ifdef CONFIG_NFSD_V3
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
51 #ifdef CONFIG_NFSD_V4
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
57 #include <linux/jhash.h>
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 unsigned int p_hindex;
86 };
87
88 struct raparm_hbucket {
89 struct raparms *pb_head;
90 spinlock_t pb_lock;
91 } ____cacheline_aligned_in_smp;
92
93 static struct raparms * raparml;
94 #define RAPARM_HASH_BITS 4
95 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
96 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
97 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
98
99 /*
100 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
101 * a mount point.
102 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
103 * or nfs_ok having possibly changed *dpp and *expp
104 */
105 int
106 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
107 struct svc_export **expp)
108 {
109 struct svc_export *exp = *expp, *exp2 = NULL;
110 struct dentry *dentry = *dpp;
111 struct vfsmount *mnt = mntget(exp->ex_mnt);
112 struct dentry *mounts = dget(dentry);
113 int err = 0;
114
115 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
116
117 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
118 if (IS_ERR(exp2)) {
119 err = PTR_ERR(exp2);
120 dput(mounts);
121 mntput(mnt);
122 goto out;
123 }
124 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
125 /* successfully crossed mount point */
126 exp_put(exp);
127 *expp = exp2;
128 dput(dentry);
129 *dpp = mounts;
130 } else {
131 if (exp2) exp_put(exp2);
132 dput(mounts);
133 }
134 mntput(mnt);
135 out:
136 return err;
137 }
138
139 /*
140 * Look up one component of a pathname.
141 * N.B. After this call _both_ fhp and resfh need an fh_put
142 *
143 * If the lookup would cross a mountpoint, and the mounted filesystem
144 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
145 * accepted as it stands and the mounted directory is
146 * returned. Otherwise the covered directory is returned.
147 * NOTE: this mountpoint crossing is not supported properly by all
148 * clients and is explicitly disallowed for NFSv3
149 * NeilBrown <neilb@cse.unsw.edu.au>
150 */
151 __be32
152 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
153 int len, struct svc_fh *resfh)
154 {
155 struct svc_export *exp;
156 struct dentry *dparent;
157 struct dentry *dentry;
158 __be32 err;
159 int host_err;
160
161 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
162
163 /* Obtain dentry and export. */
164 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
165 if (err)
166 return err;
167
168 dparent = fhp->fh_dentry;
169 exp = fhp->fh_export;
170 exp_get(exp);
171
172 err = nfserr_acces;
173
174 /* Lookup the name, but don't follow links */
175 if (isdotent(name, len)) {
176 if (len==1)
177 dentry = dget(dparent);
178 else if (dparent != exp->ex_dentry) {
179 dentry = dget_parent(dparent);
180 } else if (!EX_NOHIDE(exp))
181 dentry = dget(dparent); /* .. == . just like at / */
182 else {
183 /* checking mountpoint crossing is very different when stepping up */
184 struct svc_export *exp2 = NULL;
185 struct dentry *dp;
186 struct vfsmount *mnt = mntget(exp->ex_mnt);
187 dentry = dget(dparent);
188 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
189 ;
190 dp = dget_parent(dentry);
191 dput(dentry);
192 dentry = dp;
193
194 exp2 = exp_parent(exp->ex_client, mnt, dentry,
195 &rqstp->rq_chandle);
196 if (IS_ERR(exp2)) {
197 host_err = PTR_ERR(exp2);
198 dput(dentry);
199 mntput(mnt);
200 goto out_nfserr;
201 }
202 if (!exp2) {
203 dput(dentry);
204 dentry = dget(dparent);
205 } else {
206 exp_put(exp);
207 exp = exp2;
208 }
209 mntput(mnt);
210 }
211 } else {
212 fh_lock(fhp);
213 dentry = lookup_one_len(name, dparent, len);
214 host_err = PTR_ERR(dentry);
215 if (IS_ERR(dentry))
216 goto out_nfserr;
217 /*
218 * check if we have crossed a mount point ...
219 */
220 if (d_mountpoint(dentry)) {
221 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
222 dput(dentry);
223 goto out_nfserr;
224 }
225 }
226 }
227 /*
228 * Note: we compose the file handle now, but as the
229 * dentry may be negative, it may need to be updated.
230 */
231 err = fh_compose(resfh, exp, dentry, fhp);
232 if (!err && !dentry->d_inode)
233 err = nfserr_noent;
234 dput(dentry);
235 out:
236 exp_put(exp);
237 return err;
238
239 out_nfserr:
240 err = nfserrno(host_err);
241 goto out;
242 }
243
244 /*
245 * Set various file attributes.
246 * N.B. After this call fhp needs an fh_put
247 */
248 __be32
249 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
250 int check_guard, time_t guardtime)
251 {
252 struct dentry *dentry;
253 struct inode *inode;
254 int accmode = MAY_SATTR;
255 int ftype = 0;
256 int imode;
257 __be32 err;
258 int host_err;
259 int size_change = 0;
260
261 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
262 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
263 if (iap->ia_valid & ATTR_SIZE)
264 ftype = S_IFREG;
265
266 /* Get inode */
267 err = fh_verify(rqstp, fhp, ftype, accmode);
268 if (err)
269 goto out;
270
271 dentry = fhp->fh_dentry;
272 inode = dentry->d_inode;
273
274 /* Ignore any mode updates on symlinks */
275 if (S_ISLNK(inode->i_mode))
276 iap->ia_valid &= ~ATTR_MODE;
277
278 if (!iap->ia_valid)
279 goto out;
280
281 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
282 * which only requires access, and "set-[ac]time-to-X" which
283 * requires ownership.
284 * So if it looks like it might be "set both to the same time which
285 * is close to now", and if inode_change_ok fails, then we
286 * convert to "set to now" instead of "set to explicit time"
287 *
288 * We only call inode_change_ok as the last test as technically
289 * it is not an interface that we should be using. It is only
290 * valid if the filesystem does not define it's own i_op->setattr.
291 */
292 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
293 #define MAX_TOUCH_TIME_ERROR (30*60)
294 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
295 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
296 ) {
297 /* Looks probable. Now just make sure time is in the right ballpark.
298 * Solaris, at least, doesn't seem to care what the time request is.
299 * We require it be within 30 minutes of now.
300 */
301 time_t delta = iap->ia_atime.tv_sec - get_seconds();
302 if (delta<0) delta = -delta;
303 if (delta < MAX_TOUCH_TIME_ERROR &&
304 inode_change_ok(inode, iap) != 0) {
305 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
306 * this will cause notify_change to set these times to "now"
307 */
308 iap->ia_valid &= ~BOTH_TIME_SET;
309 }
310 }
311
312 /* The size case is special. It changes the file as well as the attributes. */
313 if (iap->ia_valid & ATTR_SIZE) {
314 if (iap->ia_size < inode->i_size) {
315 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
316 if (err)
317 goto out;
318 }
319
320 /*
321 * If we are changing the size of the file, then
322 * we need to break all leases.
323 */
324 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
325 if (host_err == -EWOULDBLOCK)
326 host_err = -ETIMEDOUT;
327 if (host_err) /* ENOMEM or EWOULDBLOCK */
328 goto out_nfserr;
329
330 host_err = get_write_access(inode);
331 if (host_err)
332 goto out_nfserr;
333
334 size_change = 1;
335 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
336 if (host_err) {
337 put_write_access(inode);
338 goto out_nfserr;
339 }
340 DQUOT_INIT(inode);
341 }
342
343 imode = inode->i_mode;
344 if (iap->ia_valid & ATTR_MODE) {
345 iap->ia_mode &= S_IALLUGO;
346 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
347 }
348
349 /* Revoke setuid/setgid bit on chown/chgrp */
350 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
351 iap->ia_valid |= ATTR_KILL_SUID;
352 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
353 iap->ia_valid |= ATTR_KILL_SGID;
354
355 /* Change the attributes. */
356
357 iap->ia_valid |= ATTR_CTIME;
358
359 err = nfserr_notsync;
360 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
361 fh_lock(fhp);
362 host_err = notify_change(dentry, iap);
363 err = nfserrno(host_err);
364 fh_unlock(fhp);
365 }
366 if (size_change)
367 put_write_access(inode);
368 if (!err)
369 if (EX_ISSYNC(fhp->fh_export))
370 write_inode_now(inode, 1);
371 out:
372 return err;
373
374 out_nfserr:
375 err = nfserrno(host_err);
376 goto out;
377 }
378
379 #if defined(CONFIG_NFSD_V2_ACL) || \
380 defined(CONFIG_NFSD_V3_ACL) || \
381 defined(CONFIG_NFSD_V4)
382 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
383 {
384 ssize_t buflen;
385
386 buflen = vfs_getxattr(dentry, key, NULL, 0);
387 if (buflen <= 0)
388 return buflen;
389
390 *buf = kmalloc(buflen, GFP_KERNEL);
391 if (!*buf)
392 return -ENOMEM;
393
394 return vfs_getxattr(dentry, key, *buf, buflen);
395 }
396 #endif
397
398 #if defined(CONFIG_NFSD_V4)
399 static int
400 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
401 {
402 int len;
403 size_t buflen;
404 char *buf = NULL;
405 int error = 0;
406
407 buflen = posix_acl_xattr_size(pacl->a_count);
408 buf = kmalloc(buflen, GFP_KERNEL);
409 error = -ENOMEM;
410 if (buf == NULL)
411 goto out;
412
413 len = posix_acl_to_xattr(pacl, buf, buflen);
414 if (len < 0) {
415 error = len;
416 goto out;
417 }
418
419 error = vfs_setxattr(dentry, key, buf, len, 0);
420 out:
421 kfree(buf);
422 return error;
423 }
424
425 __be32
426 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
427 struct nfs4_acl *acl)
428 {
429 __be32 error;
430 int host_error;
431 struct dentry *dentry;
432 struct inode *inode;
433 struct posix_acl *pacl = NULL, *dpacl = NULL;
434 unsigned int flags = 0;
435
436 /* Get inode */
437 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
438 if (error)
439 goto out;
440
441 dentry = fhp->fh_dentry;
442 inode = dentry->d_inode;
443 if (S_ISDIR(inode->i_mode))
444 flags = NFS4_ACL_DIR;
445
446 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
447 if (host_error == -EINVAL) {
448 error = nfserr_attrnotsupp;
449 goto out;
450 } else if (host_error < 0)
451 goto out_nfserr;
452
453 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
454 if (host_error < 0)
455 goto out_nfserr;
456
457 if (S_ISDIR(inode->i_mode)) {
458 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
459 if (host_error < 0)
460 goto out_nfserr;
461 }
462
463 error = nfs_ok;
464
465 out:
466 posix_acl_release(pacl);
467 posix_acl_release(dpacl);
468 return (error);
469 out_nfserr:
470 error = nfserrno(host_error);
471 goto out;
472 }
473
474 static struct posix_acl *
475 _get_posix_acl(struct dentry *dentry, char *key)
476 {
477 void *buf = NULL;
478 struct posix_acl *pacl = NULL;
479 int buflen;
480
481 buflen = nfsd_getxattr(dentry, key, &buf);
482 if (!buflen)
483 buflen = -ENODATA;
484 if (buflen <= 0)
485 return ERR_PTR(buflen);
486
487 pacl = posix_acl_from_xattr(buf, buflen);
488 kfree(buf);
489 return pacl;
490 }
491
492 int
493 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
494 {
495 struct inode *inode = dentry->d_inode;
496 int error = 0;
497 struct posix_acl *pacl = NULL, *dpacl = NULL;
498 unsigned int flags = 0;
499
500 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
501 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
502 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
503 if (IS_ERR(pacl)) {
504 error = PTR_ERR(pacl);
505 pacl = NULL;
506 goto out;
507 }
508
509 if (S_ISDIR(inode->i_mode)) {
510 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
511 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
512 dpacl = NULL;
513 else if (IS_ERR(dpacl)) {
514 error = PTR_ERR(dpacl);
515 dpacl = NULL;
516 goto out;
517 }
518 flags = NFS4_ACL_DIR;
519 }
520
521 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
522 if (IS_ERR(*acl)) {
523 error = PTR_ERR(*acl);
524 *acl = NULL;
525 }
526 out:
527 posix_acl_release(pacl);
528 posix_acl_release(dpacl);
529 return error;
530 }
531
532 #endif /* defined(CONFIG_NFS_V4) */
533
534 #ifdef CONFIG_NFSD_V3
535 /*
536 * Check server access rights to a file system object
537 */
538 struct accessmap {
539 u32 access;
540 int how;
541 };
542 static struct accessmap nfs3_regaccess[] = {
543 { NFS3_ACCESS_READ, MAY_READ },
544 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
545 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
546 { NFS3_ACCESS_EXTEND, MAY_WRITE },
547
548 { 0, 0 }
549 };
550
551 static struct accessmap nfs3_diraccess[] = {
552 { NFS3_ACCESS_READ, MAY_READ },
553 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
554 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
555 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
556 { NFS3_ACCESS_DELETE, MAY_REMOVE },
557
558 { 0, 0 }
559 };
560
561 static struct accessmap nfs3_anyaccess[] = {
562 /* Some clients - Solaris 2.6 at least, make an access call
563 * to the server to check for access for things like /dev/null
564 * (which really, the server doesn't care about). So
565 * We provide simple access checking for them, looking
566 * mainly at mode bits, and we make sure to ignore read-only
567 * filesystem checks
568 */
569 { NFS3_ACCESS_READ, MAY_READ },
570 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
571 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
572 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
573
574 { 0, 0 }
575 };
576
577 __be32
578 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
579 {
580 struct accessmap *map;
581 struct svc_export *export;
582 struct dentry *dentry;
583 u32 query, result = 0, sresult = 0;
584 __be32 error;
585
586 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
587 if (error)
588 goto out;
589
590 export = fhp->fh_export;
591 dentry = fhp->fh_dentry;
592
593 if (S_ISREG(dentry->d_inode->i_mode))
594 map = nfs3_regaccess;
595 else if (S_ISDIR(dentry->d_inode->i_mode))
596 map = nfs3_diraccess;
597 else
598 map = nfs3_anyaccess;
599
600
601 query = *access;
602 for (; map->access; map++) {
603 if (map->access & query) {
604 __be32 err2;
605
606 sresult |= map->access;
607
608 err2 = nfsd_permission(export, dentry, map->how);
609 switch (err2) {
610 case nfs_ok:
611 result |= map->access;
612 break;
613
614 /* the following error codes just mean the access was not allowed,
615 * rather than an error occurred */
616 case nfserr_rofs:
617 case nfserr_acces:
618 case nfserr_perm:
619 /* simply don't "or" in the access bit. */
620 break;
621 default:
622 error = err2;
623 goto out;
624 }
625 }
626 }
627 *access = result;
628 if (supported)
629 *supported = sresult;
630
631 out:
632 return error;
633 }
634 #endif /* CONFIG_NFSD_V3 */
635
636
637
638 /*
639 * Open an existing file or directory.
640 * The access argument indicates the type of open (read/write/lock)
641 * N.B. After this call fhp needs an fh_put
642 */
643 __be32
644 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
645 int access, struct file **filp)
646 {
647 struct dentry *dentry;
648 struct inode *inode;
649 int flags = O_RDONLY|O_LARGEFILE;
650 __be32 err;
651 int host_err;
652
653 /*
654 * If we get here, then the client has already done an "open",
655 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
656 * in case a chmod has now revoked permission.
657 */
658 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
659 if (err)
660 goto out;
661
662 dentry = fhp->fh_dentry;
663 inode = dentry->d_inode;
664
665 /* Disallow write access to files with the append-only bit set
666 * or any access when mandatory locking enabled
667 */
668 err = nfserr_perm;
669 if (IS_APPEND(inode) && (access & MAY_WRITE))
670 goto out;
671 if (IS_ISMNDLK(inode))
672 goto out;
673
674 if (!inode->i_fop)
675 goto out;
676
677 /*
678 * Check to see if there are any leases on this file.
679 * This may block while leases are broken.
680 */
681 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
682 if (host_err == -EWOULDBLOCK)
683 host_err = -ETIMEDOUT;
684 if (host_err) /* NOMEM or WOULDBLOCK */
685 goto out_nfserr;
686
687 if (access & MAY_WRITE) {
688 if (access & MAY_READ)
689 flags = O_RDWR|O_LARGEFILE;
690 else
691 flags = O_WRONLY|O_LARGEFILE;
692
693 DQUOT_INIT(inode);
694 }
695 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
696 if (IS_ERR(*filp))
697 host_err = PTR_ERR(*filp);
698 out_nfserr:
699 err = nfserrno(host_err);
700 out:
701 return err;
702 }
703
704 /*
705 * Close a file.
706 */
707 void
708 nfsd_close(struct file *filp)
709 {
710 fput(filp);
711 }
712
713 /*
714 * Sync a file
715 * As this calls fsync (not fdatasync) there is no need for a write_inode
716 * after it.
717 */
718 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
719 const struct file_operations *fop)
720 {
721 struct inode *inode = dp->d_inode;
722 int (*fsync) (struct file *, struct dentry *, int);
723 int err;
724
725 err = filemap_fdatawrite(inode->i_mapping);
726 if (err == 0 && fop && (fsync = fop->fsync))
727 err = fsync(filp, dp, 0);
728 if (err == 0)
729 err = filemap_fdatawait(inode->i_mapping);
730
731 return err;
732 }
733
734
735 static int
736 nfsd_sync(struct file *filp)
737 {
738 int err;
739 struct inode *inode = filp->f_path.dentry->d_inode;
740 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
741 mutex_lock(&inode->i_mutex);
742 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
743 mutex_unlock(&inode->i_mutex);
744
745 return err;
746 }
747
748 int
749 nfsd_sync_dir(struct dentry *dp)
750 {
751 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
752 }
753
754 /*
755 * Obtain the readahead parameters for the file
756 * specified by (dev, ino).
757 */
758
759 static inline struct raparms *
760 nfsd_get_raparms(dev_t dev, ino_t ino)
761 {
762 struct raparms *ra, **rap, **frap = NULL;
763 int depth = 0;
764 unsigned int hash;
765 struct raparm_hbucket *rab;
766
767 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
768 rab = &raparm_hash[hash];
769
770 spin_lock(&rab->pb_lock);
771 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
772 if (ra->p_ino == ino && ra->p_dev == dev)
773 goto found;
774 depth++;
775 if (ra->p_count == 0)
776 frap = rap;
777 }
778 depth = nfsdstats.ra_size*11/10;
779 if (!frap) {
780 spin_unlock(&rab->pb_lock);
781 return NULL;
782 }
783 rap = frap;
784 ra = *frap;
785 ra->p_dev = dev;
786 ra->p_ino = ino;
787 ra->p_set = 0;
788 ra->p_hindex = hash;
789 found:
790 if (rap != &rab->pb_head) {
791 *rap = ra->p_next;
792 ra->p_next = rab->pb_head;
793 rab->pb_head = ra;
794 }
795 ra->p_count++;
796 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
797 spin_unlock(&rab->pb_lock);
798 return ra;
799 }
800
801 /*
802 * Grab and keep cached pages assosiated with a file in the svc_rqst
803 * so that they can be passed to the netowork sendmsg/sendpage routines
804 * directrly. They will be released after the sending has completed.
805 */
806 static int
807 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
808 {
809 unsigned long count = desc->count;
810 struct svc_rqst *rqstp = desc->arg.data;
811 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
812
813 if (size > count)
814 size = count;
815
816 if (rqstp->rq_res.page_len == 0) {
817 get_page(page);
818 put_page(*pp);
819 *pp = page;
820 rqstp->rq_resused++;
821 rqstp->rq_res.page_base = offset;
822 rqstp->rq_res.page_len = size;
823 } else if (page != pp[-1]) {
824 get_page(page);
825 if (*pp)
826 put_page(*pp);
827 *pp = page;
828 rqstp->rq_resused++;
829 rqstp->rq_res.page_len += size;
830 } else
831 rqstp->rq_res.page_len += size;
832
833 desc->count = count - size;
834 desc->written += size;
835 return size;
836 }
837
838 static __be32
839 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
840 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
841 {
842 struct inode *inode;
843 struct raparms *ra;
844 mm_segment_t oldfs;
845 __be32 err;
846 int host_err;
847
848 err = nfserr_perm;
849 inode = file->f_path.dentry->d_inode;
850 #ifdef MSNFS
851 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
852 (!lock_may_read(inode, offset, *count)))
853 goto out;
854 #endif
855
856 /* Get readahead parameters */
857 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
858
859 if (ra && ra->p_set)
860 file->f_ra = ra->p_ra;
861
862 if (file->f_op->sendfile && rqstp->rq_sendfile_ok) {
863 rqstp->rq_resused = 1;
864 host_err = file->f_op->sendfile(file, &offset, *count,
865 nfsd_read_actor, rqstp);
866 } else {
867 oldfs = get_fs();
868 set_fs(KERNEL_DS);
869 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
870 set_fs(oldfs);
871 }
872
873 /* Write back readahead params */
874 if (ra) {
875 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
876 spin_lock(&rab->pb_lock);
877 ra->p_ra = file->f_ra;
878 ra->p_set = 1;
879 ra->p_count--;
880 spin_unlock(&rab->pb_lock);
881 }
882
883 if (host_err >= 0) {
884 nfsdstats.io_read += host_err;
885 *count = host_err;
886 err = 0;
887 fsnotify_access(file->f_path.dentry);
888 } else
889 err = nfserrno(host_err);
890 out:
891 return err;
892 }
893
894 static void kill_suid(struct dentry *dentry)
895 {
896 struct iattr ia;
897 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
898
899 mutex_lock(&dentry->d_inode->i_mutex);
900 notify_change(dentry, &ia);
901 mutex_unlock(&dentry->d_inode->i_mutex);
902 }
903
904 static __be32
905 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
906 loff_t offset, struct kvec *vec, int vlen,
907 unsigned long cnt, int *stablep)
908 {
909 struct svc_export *exp;
910 struct dentry *dentry;
911 struct inode *inode;
912 mm_segment_t oldfs;
913 __be32 err = 0;
914 int host_err;
915 int stable = *stablep;
916
917 #ifdef MSNFS
918 err = nfserr_perm;
919
920 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
921 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
922 goto out;
923 #endif
924
925 dentry = file->f_path.dentry;
926 inode = dentry->d_inode;
927 exp = fhp->fh_export;
928
929 /*
930 * Request sync writes if
931 * - the sync export option has been set, or
932 * - the client requested O_SYNC behavior (NFSv3 feature).
933 * - The file system doesn't support fsync().
934 * When gathered writes have been configured for this volume,
935 * flushing the data to disk is handled separately below.
936 */
937
938 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
939 stable = 2;
940 *stablep = 2; /* FILE_SYNC */
941 }
942
943 if (!EX_ISSYNC(exp))
944 stable = 0;
945 if (stable && !EX_WGATHER(exp))
946 file->f_flags |= O_SYNC;
947
948 /* Write the data. */
949 oldfs = get_fs(); set_fs(KERNEL_DS);
950 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
951 set_fs(oldfs);
952 if (host_err >= 0) {
953 nfsdstats.io_write += cnt;
954 fsnotify_modify(file->f_path.dentry);
955 }
956
957 /* clear setuid/setgid flag after write */
958 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
959 kill_suid(dentry);
960
961 if (host_err >= 0 && stable) {
962 static ino_t last_ino;
963 static dev_t last_dev;
964
965 /*
966 * Gathered writes: If another process is currently
967 * writing to the file, there's a high chance
968 * this is another nfsd (triggered by a bulk write
969 * from a client's biod). Rather than syncing the
970 * file with each write request, we sleep for 10 msec.
971 *
972 * I don't know if this roughly approximates
973 * C. Juszak's idea of gathered writes, but it's a
974 * nice and simple solution (IMHO), and it seems to
975 * work:-)
976 */
977 if (EX_WGATHER(exp)) {
978 if (atomic_read(&inode->i_writecount) > 1
979 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
980 dprintk("nfsd: write defer %d\n", current->pid);
981 msleep(10);
982 dprintk("nfsd: write resume %d\n", current->pid);
983 }
984
985 if (inode->i_state & I_DIRTY) {
986 dprintk("nfsd: write sync %d\n", current->pid);
987 host_err=nfsd_sync(file);
988 }
989 #if 0
990 wake_up(&inode->i_wait);
991 #endif
992 }
993 last_ino = inode->i_ino;
994 last_dev = inode->i_sb->s_dev;
995 }
996
997 dprintk("nfsd: write complete host_err=%d\n", host_err);
998 if (host_err >= 0)
999 err = 0;
1000 else
1001 err = nfserrno(host_err);
1002 out:
1003 return err;
1004 }
1005
1006 /*
1007 * Read data from a file. count must contain the requested read count
1008 * on entry. On return, *count contains the number of bytes actually read.
1009 * N.B. After this call fhp needs an fh_put
1010 */
1011 __be32
1012 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1013 loff_t offset, struct kvec *vec, int vlen,
1014 unsigned long *count)
1015 {
1016 __be32 err;
1017
1018 if (file) {
1019 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1020 MAY_READ|MAY_OWNER_OVERRIDE);
1021 if (err)
1022 goto out;
1023 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1024 } else {
1025 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1026 if (err)
1027 goto out;
1028 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1029 nfsd_close(file);
1030 }
1031 out:
1032 return err;
1033 }
1034
1035 /*
1036 * Write data to a file.
1037 * The stable flag requests synchronous writes.
1038 * N.B. After this call fhp needs an fh_put
1039 */
1040 __be32
1041 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1042 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1043 int *stablep)
1044 {
1045 __be32 err = 0;
1046
1047 if (file) {
1048 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1049 MAY_WRITE|MAY_OWNER_OVERRIDE);
1050 if (err)
1051 goto out;
1052 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1053 stablep);
1054 } else {
1055 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1056 if (err)
1057 goto out;
1058
1059 if (cnt)
1060 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1061 cnt, stablep);
1062 nfsd_close(file);
1063 }
1064 out:
1065 return err;
1066 }
1067
1068 #ifdef CONFIG_NFSD_V3
1069 /*
1070 * Commit all pending writes to stable storage.
1071 * Strictly speaking, we could sync just the indicated file region here,
1072 * but there's currently no way we can ask the VFS to do so.
1073 *
1074 * Unfortunately we cannot lock the file to make sure we return full WCC
1075 * data to the client, as locking happens lower down in the filesystem.
1076 */
1077 __be32
1078 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1079 loff_t offset, unsigned long count)
1080 {
1081 struct file *file;
1082 __be32 err;
1083
1084 if ((u64)count > ~(u64)offset)
1085 return nfserr_inval;
1086
1087 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1088 return err;
1089 if (EX_ISSYNC(fhp->fh_export)) {
1090 if (file->f_op && file->f_op->fsync) {
1091 err = nfserrno(nfsd_sync(file));
1092 } else {
1093 err = nfserr_notsupp;
1094 }
1095 }
1096
1097 nfsd_close(file);
1098 return err;
1099 }
1100 #endif /* CONFIG_NFSD_V3 */
1101
1102 /*
1103 * Create a file (regular, directory, device, fifo); UNIX sockets
1104 * not yet implemented.
1105 * If the response fh has been verified, the parent directory should
1106 * already be locked. Note that the parent directory is left locked.
1107 *
1108 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1109 */
1110 __be32
1111 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1112 char *fname, int flen, struct iattr *iap,
1113 int type, dev_t rdev, struct svc_fh *resfhp)
1114 {
1115 struct dentry *dentry, *dchild = NULL;
1116 struct inode *dirp;
1117 __be32 err;
1118 int host_err;
1119
1120 err = nfserr_perm;
1121 if (!flen)
1122 goto out;
1123 err = nfserr_exist;
1124 if (isdotent(fname, flen))
1125 goto out;
1126
1127 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1128 if (err)
1129 goto out;
1130
1131 dentry = fhp->fh_dentry;
1132 dirp = dentry->d_inode;
1133
1134 err = nfserr_notdir;
1135 if(!dirp->i_op || !dirp->i_op->lookup)
1136 goto out;
1137 /*
1138 * Check whether the response file handle has been verified yet.
1139 * If it has, the parent directory should already be locked.
1140 */
1141 if (!resfhp->fh_dentry) {
1142 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1143 fh_lock_nested(fhp, I_MUTEX_PARENT);
1144 dchild = lookup_one_len(fname, dentry, flen);
1145 host_err = PTR_ERR(dchild);
1146 if (IS_ERR(dchild))
1147 goto out_nfserr;
1148 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1149 if (err)
1150 goto out;
1151 } else {
1152 /* called from nfsd_proc_create */
1153 dchild = dget(resfhp->fh_dentry);
1154 if (!fhp->fh_locked) {
1155 /* not actually possible */
1156 printk(KERN_ERR
1157 "nfsd_create: parent %s/%s not locked!\n",
1158 dentry->d_parent->d_name.name,
1159 dentry->d_name.name);
1160 err = nfserr_io;
1161 goto out;
1162 }
1163 }
1164 /*
1165 * Make sure the child dentry is still negative ...
1166 */
1167 err = nfserr_exist;
1168 if (dchild->d_inode) {
1169 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1170 dentry->d_name.name, dchild->d_name.name);
1171 goto out;
1172 }
1173
1174 if (!(iap->ia_valid & ATTR_MODE))
1175 iap->ia_mode = 0;
1176 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1177
1178 /*
1179 * Get the dir op function pointer.
1180 */
1181 err = 0;
1182 switch (type) {
1183 case S_IFREG:
1184 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1185 break;
1186 case S_IFDIR:
1187 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1188 break;
1189 case S_IFCHR:
1190 case S_IFBLK:
1191 case S_IFIFO:
1192 case S_IFSOCK:
1193 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1194 break;
1195 default:
1196 printk("nfsd: bad file type %o in nfsd_create\n", type);
1197 host_err = -EINVAL;
1198 }
1199 if (host_err < 0)
1200 goto out_nfserr;
1201
1202 if (EX_ISSYNC(fhp->fh_export)) {
1203 err = nfserrno(nfsd_sync_dir(dentry));
1204 write_inode_now(dchild->d_inode, 1);
1205 }
1206
1207
1208 /* Set file attributes. Mode has already been set and
1209 * setting uid/gid works only for root. Irix appears to
1210 * send along the gid when it tries to implement setgid
1211 * directories via NFS.
1212 */
1213 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1214 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1215 if (err2)
1216 err = err2;
1217 }
1218 /*
1219 * Update the file handle to get the new inode info.
1220 */
1221 if (!err)
1222 err = fh_update(resfhp);
1223 out:
1224 if (dchild && !IS_ERR(dchild))
1225 dput(dchild);
1226 return err;
1227
1228 out_nfserr:
1229 err = nfserrno(host_err);
1230 goto out;
1231 }
1232
1233 #ifdef CONFIG_NFSD_V3
1234 /*
1235 * NFSv3 version of nfsd_create
1236 */
1237 __be32
1238 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1239 char *fname, int flen, struct iattr *iap,
1240 struct svc_fh *resfhp, int createmode, u32 *verifier,
1241 int *truncp, int *created)
1242 {
1243 struct dentry *dentry, *dchild = NULL;
1244 struct inode *dirp;
1245 __be32 err;
1246 int host_err;
1247 __u32 v_mtime=0, v_atime=0;
1248
1249 err = nfserr_perm;
1250 if (!flen)
1251 goto out;
1252 err = nfserr_exist;
1253 if (isdotent(fname, flen))
1254 goto out;
1255 if (!(iap->ia_valid & ATTR_MODE))
1256 iap->ia_mode = 0;
1257 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1258 if (err)
1259 goto out;
1260
1261 dentry = fhp->fh_dentry;
1262 dirp = dentry->d_inode;
1263
1264 /* Get all the sanity checks out of the way before
1265 * we lock the parent. */
1266 err = nfserr_notdir;
1267 if(!dirp->i_op || !dirp->i_op->lookup)
1268 goto out;
1269 fh_lock_nested(fhp, I_MUTEX_PARENT);
1270
1271 /*
1272 * Compose the response file handle.
1273 */
1274 dchild = lookup_one_len(fname, dentry, flen);
1275 host_err = PTR_ERR(dchild);
1276 if (IS_ERR(dchild))
1277 goto out_nfserr;
1278
1279 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1280 if (err)
1281 goto out;
1282
1283 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1284 /* solaris7 gets confused (bugid 4218508) if these have
1285 * the high bit set, so just clear the high bits.
1286 */
1287 v_mtime = verifier[0]&0x7fffffff;
1288 v_atime = verifier[1]&0x7fffffff;
1289 }
1290
1291 if (dchild->d_inode) {
1292 err = 0;
1293
1294 switch (createmode) {
1295 case NFS3_CREATE_UNCHECKED:
1296 if (! S_ISREG(dchild->d_inode->i_mode))
1297 err = nfserr_exist;
1298 else if (truncp) {
1299 /* in nfsv4, we need to treat this case a little
1300 * differently. we don't want to truncate the
1301 * file now; this would be wrong if the OPEN
1302 * fails for some other reason. furthermore,
1303 * if the size is nonzero, we should ignore it
1304 * according to spec!
1305 */
1306 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1307 }
1308 else {
1309 iap->ia_valid &= ATTR_SIZE;
1310 goto set_attr;
1311 }
1312 break;
1313 case NFS3_CREATE_EXCLUSIVE:
1314 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1315 && dchild->d_inode->i_atime.tv_sec == v_atime
1316 && dchild->d_inode->i_size == 0 )
1317 break;
1318 /* fallthru */
1319 case NFS3_CREATE_GUARDED:
1320 err = nfserr_exist;
1321 }
1322 goto out;
1323 }
1324
1325 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1326 if (host_err < 0)
1327 goto out_nfserr;
1328 if (created)
1329 *created = 1;
1330
1331 if (EX_ISSYNC(fhp->fh_export)) {
1332 err = nfserrno(nfsd_sync_dir(dentry));
1333 /* setattr will sync the child (or not) */
1334 }
1335
1336 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1337 /* Cram the verifier into atime/mtime */
1338 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1339 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1340 /* XXX someone who knows this better please fix it for nsec */
1341 iap->ia_mtime.tv_sec = v_mtime;
1342 iap->ia_atime.tv_sec = v_atime;
1343 iap->ia_mtime.tv_nsec = 0;
1344 iap->ia_atime.tv_nsec = 0;
1345 }
1346
1347 /* Set file attributes.
1348 * Irix appears to send along the gid when it tries to
1349 * implement setgid directories via NFS. Clear out all that cruft.
1350 */
1351 set_attr:
1352 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1353 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1354 if (err2)
1355 err = err2;
1356 }
1357
1358 /*
1359 * Update the filehandle to get the new inode info.
1360 */
1361 if (!err)
1362 err = fh_update(resfhp);
1363
1364 out:
1365 fh_unlock(fhp);
1366 if (dchild && !IS_ERR(dchild))
1367 dput(dchild);
1368 return err;
1369
1370 out_nfserr:
1371 err = nfserrno(host_err);
1372 goto out;
1373 }
1374 #endif /* CONFIG_NFSD_V3 */
1375
1376 /*
1377 * Read a symlink. On entry, *lenp must contain the maximum path length that
1378 * fits into the buffer. On return, it contains the true length.
1379 * N.B. After this call fhp needs an fh_put
1380 */
1381 __be32
1382 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1383 {
1384 struct dentry *dentry;
1385 struct inode *inode;
1386 mm_segment_t oldfs;
1387 __be32 err;
1388 int host_err;
1389
1390 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1391 if (err)
1392 goto out;
1393
1394 dentry = fhp->fh_dentry;
1395 inode = dentry->d_inode;
1396
1397 err = nfserr_inval;
1398 if (!inode->i_op || !inode->i_op->readlink)
1399 goto out;
1400
1401 touch_atime(fhp->fh_export->ex_mnt, dentry);
1402 /* N.B. Why does this call need a get_fs()??
1403 * Remove the set_fs and watch the fireworks:-) --okir
1404 */
1405
1406 oldfs = get_fs(); set_fs(KERNEL_DS);
1407 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1408 set_fs(oldfs);
1409
1410 if (host_err < 0)
1411 goto out_nfserr;
1412 *lenp = host_err;
1413 err = 0;
1414 out:
1415 return err;
1416
1417 out_nfserr:
1418 err = nfserrno(host_err);
1419 goto out;
1420 }
1421
1422 /*
1423 * Create a symlink and look up its inode
1424 * N.B. After this call _both_ fhp and resfhp need an fh_put
1425 */
1426 __be32
1427 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1428 char *fname, int flen,
1429 char *path, int plen,
1430 struct svc_fh *resfhp,
1431 struct iattr *iap)
1432 {
1433 struct dentry *dentry, *dnew;
1434 __be32 err, cerr;
1435 int host_err;
1436 umode_t mode;
1437
1438 err = nfserr_noent;
1439 if (!flen || !plen)
1440 goto out;
1441 err = nfserr_exist;
1442 if (isdotent(fname, flen))
1443 goto out;
1444
1445 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1446 if (err)
1447 goto out;
1448 fh_lock(fhp);
1449 dentry = fhp->fh_dentry;
1450 dnew = lookup_one_len(fname, dentry, flen);
1451 host_err = PTR_ERR(dnew);
1452 if (IS_ERR(dnew))
1453 goto out_nfserr;
1454
1455 mode = S_IALLUGO;
1456 /* Only the MODE ATTRibute is even vaguely meaningful */
1457 if (iap && (iap->ia_valid & ATTR_MODE))
1458 mode = iap->ia_mode & S_IALLUGO;
1459
1460 if (unlikely(path[plen] != 0)) {
1461 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1462 if (path_alloced == NULL)
1463 host_err = -ENOMEM;
1464 else {
1465 strncpy(path_alloced, path, plen);
1466 path_alloced[plen] = 0;
1467 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1468 kfree(path_alloced);
1469 }
1470 } else
1471 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1472
1473 if (!host_err) {
1474 if (EX_ISSYNC(fhp->fh_export))
1475 host_err = nfsd_sync_dir(dentry);
1476 }
1477 err = nfserrno(host_err);
1478 fh_unlock(fhp);
1479
1480 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1481 dput(dnew);
1482 if (err==0) err = cerr;
1483 out:
1484 return err;
1485
1486 out_nfserr:
1487 err = nfserrno(host_err);
1488 goto out;
1489 }
1490
1491 /*
1492 * Create a hardlink
1493 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1494 */
1495 __be32
1496 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1497 char *name, int len, struct svc_fh *tfhp)
1498 {
1499 struct dentry *ddir, *dnew, *dold;
1500 struct inode *dirp, *dest;
1501 __be32 err;
1502 int host_err;
1503
1504 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1505 if (err)
1506 goto out;
1507 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1508 if (err)
1509 goto out;
1510
1511 err = nfserr_perm;
1512 if (!len)
1513 goto out;
1514 err = nfserr_exist;
1515 if (isdotent(name, len))
1516 goto out;
1517
1518 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1519 ddir = ffhp->fh_dentry;
1520 dirp = ddir->d_inode;
1521
1522 dnew = lookup_one_len(name, ddir, len);
1523 host_err = PTR_ERR(dnew);
1524 if (IS_ERR(dnew))
1525 goto out_nfserr;
1526
1527 dold = tfhp->fh_dentry;
1528 dest = dold->d_inode;
1529
1530 host_err = vfs_link(dold, dirp, dnew);
1531 if (!host_err) {
1532 if (EX_ISSYNC(ffhp->fh_export)) {
1533 err = nfserrno(nfsd_sync_dir(ddir));
1534 write_inode_now(dest, 1);
1535 }
1536 err = 0;
1537 } else {
1538 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1539 err = nfserr_acces;
1540 else
1541 err = nfserrno(host_err);
1542 }
1543
1544 dput(dnew);
1545 out_unlock:
1546 fh_unlock(ffhp);
1547 out:
1548 return err;
1549
1550 out_nfserr:
1551 err = nfserrno(host_err);
1552 goto out_unlock;
1553 }
1554
1555 /*
1556 * Rename a file
1557 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1558 */
1559 __be32
1560 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1561 struct svc_fh *tfhp, char *tname, int tlen)
1562 {
1563 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1564 struct inode *fdir, *tdir;
1565 __be32 err;
1566 int host_err;
1567
1568 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1569 if (err)
1570 goto out;
1571 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1572 if (err)
1573 goto out;
1574
1575 fdentry = ffhp->fh_dentry;
1576 fdir = fdentry->d_inode;
1577
1578 tdentry = tfhp->fh_dentry;
1579 tdir = tdentry->d_inode;
1580
1581 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1582 if (ffhp->fh_export != tfhp->fh_export)
1583 goto out;
1584
1585 err = nfserr_perm;
1586 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1587 goto out;
1588
1589 /* cannot use fh_lock as we need deadlock protective ordering
1590 * so do it by hand */
1591 trap = lock_rename(tdentry, fdentry);
1592 ffhp->fh_locked = tfhp->fh_locked = 1;
1593 fill_pre_wcc(ffhp);
1594 fill_pre_wcc(tfhp);
1595
1596 odentry = lookup_one_len(fname, fdentry, flen);
1597 host_err = PTR_ERR(odentry);
1598 if (IS_ERR(odentry))
1599 goto out_nfserr;
1600
1601 host_err = -ENOENT;
1602 if (!odentry->d_inode)
1603 goto out_dput_old;
1604 host_err = -EINVAL;
1605 if (odentry == trap)
1606 goto out_dput_old;
1607
1608 ndentry = lookup_one_len(tname, tdentry, tlen);
1609 host_err = PTR_ERR(ndentry);
1610 if (IS_ERR(ndentry))
1611 goto out_dput_old;
1612 host_err = -ENOTEMPTY;
1613 if (ndentry == trap)
1614 goto out_dput_new;
1615
1616 #ifdef MSNFS
1617 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1618 ((atomic_read(&odentry->d_count) > 1)
1619 || (atomic_read(&ndentry->d_count) > 1))) {
1620 host_err = -EPERM;
1621 } else
1622 #endif
1623 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1624 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1625 host_err = nfsd_sync_dir(tdentry);
1626 if (!host_err)
1627 host_err = nfsd_sync_dir(fdentry);
1628 }
1629
1630 out_dput_new:
1631 dput(ndentry);
1632 out_dput_old:
1633 dput(odentry);
1634 out_nfserr:
1635 err = nfserrno(host_err);
1636
1637 /* we cannot reply on fh_unlock on the two filehandles,
1638 * as that would do the wrong thing if the two directories
1639 * were the same, so again we do it by hand
1640 */
1641 fill_post_wcc(ffhp);
1642 fill_post_wcc(tfhp);
1643 unlock_rename(tdentry, fdentry);
1644 ffhp->fh_locked = tfhp->fh_locked = 0;
1645
1646 out:
1647 return err;
1648 }
1649
1650 /*
1651 * Unlink a file or directory
1652 * N.B. After this call fhp needs an fh_put
1653 */
1654 __be32
1655 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1656 char *fname, int flen)
1657 {
1658 struct dentry *dentry, *rdentry;
1659 struct inode *dirp;
1660 __be32 err;
1661 int host_err;
1662
1663 err = nfserr_acces;
1664 if (!flen || isdotent(fname, flen))
1665 goto out;
1666 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1667 if (err)
1668 goto out;
1669
1670 fh_lock_nested(fhp, I_MUTEX_PARENT);
1671 dentry = fhp->fh_dentry;
1672 dirp = dentry->d_inode;
1673
1674 rdentry = lookup_one_len(fname, dentry, flen);
1675 host_err = PTR_ERR(rdentry);
1676 if (IS_ERR(rdentry))
1677 goto out_nfserr;
1678
1679 if (!rdentry->d_inode) {
1680 dput(rdentry);
1681 err = nfserr_noent;
1682 goto out;
1683 }
1684
1685 if (!type)
1686 type = rdentry->d_inode->i_mode & S_IFMT;
1687
1688 if (type != S_IFDIR) { /* It's UNLINK */
1689 #ifdef MSNFS
1690 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1691 (atomic_read(&rdentry->d_count) > 1)) {
1692 host_err = -EPERM;
1693 } else
1694 #endif
1695 host_err = vfs_unlink(dirp, rdentry);
1696 } else { /* It's RMDIR */
1697 host_err = vfs_rmdir(dirp, rdentry);
1698 }
1699
1700 dput(rdentry);
1701
1702 if (host_err)
1703 goto out_nfserr;
1704 if (EX_ISSYNC(fhp->fh_export))
1705 host_err = nfsd_sync_dir(dentry);
1706
1707 out_nfserr:
1708 err = nfserrno(host_err);
1709 out:
1710 return err;
1711 }
1712
1713 /*
1714 * Read entries from a directory.
1715 * The NFSv3/4 verifier we ignore for now.
1716 */
1717 __be32
1718 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1719 struct readdir_cd *cdp, filldir_t func)
1720 {
1721 __be32 err;
1722 int host_err;
1723 struct file *file;
1724 loff_t offset = *offsetp;
1725
1726 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1727 if (err)
1728 goto out;
1729
1730 offset = vfs_llseek(file, offset, 0);
1731 if (offset < 0) {
1732 err = nfserrno((int)offset);
1733 goto out_close;
1734 }
1735
1736 /*
1737 * Read the directory entries. This silly loop is necessary because
1738 * readdir() is not guaranteed to fill up the entire buffer, but
1739 * may choose to do less.
1740 */
1741
1742 do {
1743 cdp->err = nfserr_eof; /* will be cleared on successful read */
1744 host_err = vfs_readdir(file, func, cdp);
1745 } while (host_err >=0 && cdp->err == nfs_ok);
1746 if (host_err)
1747 err = nfserrno(host_err);
1748 else
1749 err = cdp->err;
1750 *offsetp = vfs_llseek(file, 0, 1);
1751
1752 if (err == nfserr_eof || err == nfserr_toosmall)
1753 err = nfs_ok; /* can still be found in ->err */
1754 out_close:
1755 nfsd_close(file);
1756 out:
1757 return err;
1758 }
1759
1760 /*
1761 * Get file system stats
1762 * N.B. After this call fhp needs an fh_put
1763 */
1764 __be32
1765 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1766 {
1767 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1768 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1769 err = nfserr_io;
1770 return err;
1771 }
1772
1773 /*
1774 * Check for a user's access permissions to this inode.
1775 */
1776 __be32
1777 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1778 {
1779 struct inode *inode = dentry->d_inode;
1780 int err;
1781
1782 if (acc == MAY_NOP)
1783 return 0;
1784 #if 0
1785 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1786 acc,
1787 (acc & MAY_READ)? " read" : "",
1788 (acc & MAY_WRITE)? " write" : "",
1789 (acc & MAY_EXEC)? " exec" : "",
1790 (acc & MAY_SATTR)? " sattr" : "",
1791 (acc & MAY_TRUNC)? " trunc" : "",
1792 (acc & MAY_LOCK)? " lock" : "",
1793 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1794 inode->i_mode,
1795 IS_IMMUTABLE(inode)? " immut" : "",
1796 IS_APPEND(inode)? " append" : "",
1797 IS_RDONLY(inode)? " ro" : "");
1798 dprintk(" owner %d/%d user %d/%d\n",
1799 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1800 #endif
1801
1802 /* Normally we reject any write/sattr etc access on a read-only file
1803 * system. But if it is IRIX doing check on write-access for a
1804 * device special file, we ignore rofs.
1805 */
1806 if (!(acc & MAY_LOCAL_ACCESS))
1807 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1808 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1809 return nfserr_rofs;
1810 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1811 return nfserr_perm;
1812 }
1813 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1814 return nfserr_perm;
1815
1816 if (acc & MAY_LOCK) {
1817 /* If we cannot rely on authentication in NLM requests,
1818 * just allow locks, otherwise require read permission, or
1819 * ownership
1820 */
1821 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1822 return 0;
1823 else
1824 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1825 }
1826 /*
1827 * The file owner always gets access permission for accesses that
1828 * would normally be checked at open time. This is to make
1829 * file access work even when the client has done a fchmod(fd, 0).
1830 *
1831 * However, `cp foo bar' should fail nevertheless when bar is
1832 * readonly. A sensible way to do this might be to reject all
1833 * attempts to truncate a read-only file, because a creat() call
1834 * always implies file truncation.
1835 * ... but this isn't really fair. A process may reasonably call
1836 * ftruncate on an open file descriptor on a file with perm 000.
1837 * We must trust the client to do permission checking - using "ACCESS"
1838 * with NFSv3.
1839 */
1840 if ((acc & MAY_OWNER_OVERRIDE) &&
1841 inode->i_uid == current->fsuid)
1842 return 0;
1843
1844 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1845
1846 /* Allow read access to binaries even when mode 111 */
1847 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1848 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1849 err = permission(inode, MAY_EXEC, NULL);
1850
1851 return err? nfserrno(err) : 0;
1852 }
1853
1854 void
1855 nfsd_racache_shutdown(void)
1856 {
1857 if (!raparml)
1858 return;
1859 dprintk("nfsd: freeing readahead buffers.\n");
1860 kfree(raparml);
1861 raparml = NULL;
1862 }
1863 /*
1864 * Initialize readahead param cache
1865 */
1866 int
1867 nfsd_racache_init(int cache_size)
1868 {
1869 int i;
1870 int j = 0;
1871 int nperbucket;
1872
1873
1874 if (raparml)
1875 return 0;
1876 if (cache_size < 2*RAPARM_HASH_SIZE)
1877 cache_size = 2*RAPARM_HASH_SIZE;
1878 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1879
1880 if (!raparml) {
1881 printk(KERN_WARNING
1882 "nfsd: Could not allocate memory read-ahead cache.\n");
1883 return -ENOMEM;
1884 }
1885
1886 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1887 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1888 raparm_hash[i].pb_head = NULL;
1889 spin_lock_init(&raparm_hash[i].pb_lock);
1890 }
1891 nperbucket = cache_size >> RAPARM_HASH_BITS;
1892 for (i = 0; i < cache_size - 1; i++) {
1893 if (i % nperbucket == 0)
1894 raparm_hash[j++].pb_head = raparml + i;
1895 if (i % nperbucket < nperbucket-1)
1896 raparml[i].p_next = raparml + i + 1;
1897 }
1898
1899 nfsdstats.ra_size = cache_size;
1900 return 0;
1901 }
1902
1903 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1904 struct posix_acl *
1905 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1906 {
1907 struct inode *inode = fhp->fh_dentry->d_inode;
1908 char *name;
1909 void *value = NULL;
1910 ssize_t size;
1911 struct posix_acl *acl;
1912
1913 if (!IS_POSIXACL(inode))
1914 return ERR_PTR(-EOPNOTSUPP);
1915
1916 switch (type) {
1917 case ACL_TYPE_ACCESS:
1918 name = POSIX_ACL_XATTR_ACCESS;
1919 break;
1920 case ACL_TYPE_DEFAULT:
1921 name = POSIX_ACL_XATTR_DEFAULT;
1922 break;
1923 default:
1924 return ERR_PTR(-EOPNOTSUPP);
1925 }
1926
1927 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1928 if (size < 0)
1929 return ERR_PTR(size);
1930
1931 acl = posix_acl_from_xattr(value, size);
1932 kfree(value);
1933 return acl;
1934 }
1935
1936 int
1937 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1938 {
1939 struct inode *inode = fhp->fh_dentry->d_inode;
1940 char *name;
1941 void *value = NULL;
1942 size_t size;
1943 int error;
1944
1945 if (!IS_POSIXACL(inode) || !inode->i_op ||
1946 !inode->i_op->setxattr || !inode->i_op->removexattr)
1947 return -EOPNOTSUPP;
1948 switch(type) {
1949 case ACL_TYPE_ACCESS:
1950 name = POSIX_ACL_XATTR_ACCESS;
1951 break;
1952 case ACL_TYPE_DEFAULT:
1953 name = POSIX_ACL_XATTR_DEFAULT;
1954 break;
1955 default:
1956 return -EOPNOTSUPP;
1957 }
1958
1959 if (acl && acl->a_count) {
1960 size = posix_acl_xattr_size(acl->a_count);
1961 value = kmalloc(size, GFP_KERNEL);
1962 if (!value)
1963 return -ENOMEM;
1964 error = posix_acl_to_xattr(acl, value, size);
1965 if (error < 0)
1966 goto getout;
1967 size = error;
1968 } else
1969 size = 0;
1970
1971 if (size)
1972 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1973 else {
1974 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1975 error = 0;
1976 else {
1977 error = vfs_removexattr(fhp->fh_dentry, name);
1978 if (error == -ENODATA)
1979 error = 0;
1980 }
1981 }
1982
1983 getout:
1984 kfree(value);
1985 return error;
1986 }
1987 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
Linux kernel - Deliverables
This page took 0.07363 seconds and 5 git commands to generate.