FreeBSD/Linux Kernel Cross Reference
sys/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 in ksyms.c, 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 */
17
18 #include <linux/config.h>
19 #include <linux/version.h>
20 #include <linux/string.h>
21 #include <linux/sched.h>
22 #include <linux/errno.h>
23 #include <linux/locks.h>
24 #include <linux/fs.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/in.h>
34 #define __NO_VERSION__
35 #include <linux/module.h>
36
37 #include <linux/sunrpc/svc.h>
38 #include <linux/nfsd/nfsd.h>
39 #ifdef CONFIG_NFSD_V3
40 #include <linux/nfs3.h>
41 #include <linux/nfsd/xdr3.h>
42 #endif /* CONFIG_NFSD_V3 */
43 #include <linux/nfsd/nfsfh.h>
44 #include <linux/quotaops.h>
45
46 #include <asm/uaccess.h>
47
48 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 #define NFSD_PARANOIA
50
51
52 /* We must ignore files (but only files) which might have mandatory
53 * locks on them because there is no way to know if the accesser has
54 * the lock.
55 */
56 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
57
58 /*
59 * This is a cache of readahead params that help us choose the proper
60 * readahead strategy. Initially, we set all readahead parameters to 0
61 * and let the VFS handle things.
62 * If you increase the number of cached files very much, you'll need to
63 * add a hash table here.
64 */
65 struct raparms {
66 struct raparms *p_next;
67 unsigned int p_count;
68 ino_t p_ino;
69 dev_t p_dev;
70 unsigned long p_reada,
71 p_ramax,
72 p_raend,
73 p_ralen,
74 p_rawin;
75 };
76
77 static struct raparms * raparml;
78 static struct raparms * raparm_cache;
79
80 /*
81 * Look up one component of a pathname.
82 * N.B. After this call _both_ fhp and resfh need an fh_put
83 *
84 * If the lookup would cross a mountpoint, and the mounted filesystem
85 * is exported to the client with NFSEXP_CROSSMNT, then the lookup is
86 * accepted as it stands and the mounted directory is
87 * returned. Otherwise the covered directory is returned.
88 * NOTE: this mountpoint crossing is not supported properly by all
89 * clients and is explicitly disallowed for NFSv3
90 * NeilBrown <neilb@cse.unsw.edu.au>
91 */
92 int
93 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
94 int len, struct svc_fh *resfh)
95 {
96 struct svc_export *exp;
97 struct dentry *dparent;
98 struct dentry *dentry;
99 int err;
100
101 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
102
103 /* Obtain dentry and export. */
104 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
105 if (err)
106 goto out;
107
108 dparent = fhp->fh_dentry;
109 exp = fhp->fh_export;
110
111 err = nfserr_acces;
112
113 /* Lookup the name, but don't follow links */
114 if (isdotent(name, len)) {
115 if (len==1)
116 dentry = dget(dparent);
117 else if (dparent != exp->ex_dentry)
118 dentry = dget(dparent->d_parent);
119 else if (!EX_CROSSMNT(exp))
120 dentry = dget(dparent); /* .. == . just like at / */
121 else {
122 /* checking mountpoint crossing is very different when stepping up */
123 struct svc_export *exp2 = NULL;
124 struct dentry *dp;
125 struct vfsmount *mnt = mntget(exp->ex_mnt);
126 dentry = dget(dparent);
127 while(follow_up(&mnt, &dentry))
128 ;
129 dp = dget(dentry->d_parent);
130 dput(dentry);
131 dentry = dp;
132 for ( ; exp2 == NULL && dp->d_parent != dp;
133 dp=dp->d_parent)
134 exp2 = exp_get(exp->ex_client, dp->d_inode->i_dev, dp->d_inode->i_ino);
135 if (exp2==NULL) {
136 dput(dentry);
137 dentry = dget(dparent);
138 } else {
139 exp = exp2;
140 }
141 mntput(mnt);
142 }
143 } else {
144 fh_lock(fhp);
145 dentry = lookup_one_len(name, dparent, len);
146 err = PTR_ERR(dentry);
147 if (IS_ERR(dentry))
148 goto out_nfserr;
149 /*
150 * check if we have crossed a mount point ...
151 */
152 if (d_mountpoint(dentry)) {
153 struct svc_export *exp2 = NULL;
154 struct vfsmount *mnt = mntget(exp->ex_mnt);
155 struct dentry *mounts = dget(dentry);
156 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts))
157 ;
158 exp2 = exp_get(rqstp->rq_client,
159 mounts->d_inode->i_dev,
160 mounts->d_inode->i_ino);
161 if (exp2 && EX_CROSSMNT(exp2)) {
162 /* successfully crossed mount point */
163 exp = exp2;
164 dput(dentry);
165 dentry = mounts;
166 } else
167 dput(mounts);
168 mntput(mnt);
169 }
170 }
171
172 if (dentry->d_inode && dentry->d_inode->i_op &&
173 dentry->d_inode->i_op->revalidate &&
174 dentry->d_inode->i_op->revalidate(dentry))
175 err = nfserr_noent;
176 else
177 err = fh_compose(resfh, exp, dentry, fhp);
178 if (!err && !dentry->d_inode)
179 err = nfserr_noent;
180 out:
181 return err;
182
183 out_nfserr:
184 err = nfserrno(err);
185 goto out;
186 }
187
188 /*
189 * Set various file attributes.
190 * N.B. After this call fhp needs an fh_put
191 */
192 int
193 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
194 int check_guard, time_t guardtime)
195 {
196 struct dentry *dentry;
197 struct inode *inode;
198 int accmode = MAY_SATTR;
199 int ftype = 0;
200 int imode;
201 int err;
202 int size_change = 0;
203
204 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
205 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
206 if (iap->ia_valid & ATTR_SIZE)
207 ftype = S_IFREG;
208
209 /* Get inode */
210 err = fh_verify(rqstp, fhp, ftype, accmode);
211 if (err || !iap->ia_valid)
212 goto out;
213
214 dentry = fhp->fh_dentry;
215 inode = dentry->d_inode;
216
217 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
218 * which only requires access, and "set-[ac]time-to-X" which
219 * requires ownership.
220 * So if it looks like it might be "set both to the same time which
221 * is close to now", and if inode_change_ok fails, then we
222 * convert to "set to now" instead of "set to explicit time"
223 *
224 * We only call inode_change_ok as the last test as technically
225 * it is not an interface that we should be using. It is only
226 * valid if the filesystem does not define it's own i_op->setattr.
227 */
228 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
229 #define MAX_TOUCH_TIME_ERROR (30*60)
230 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
231 && iap->ia_mtime == iap->ia_atime
232 ) {
233 /* Looks probable. Now just make sure time is in the right ballpark.
234 * Solaris, at least, doesn't seem to care what the time request is.
235 * We require it be within 30 minutes of now.
236 */
237 time_t delta = iap->ia_atime - CURRENT_TIME;
238 if (delta<0) delta = -delta;
239 if (delta < MAX_TOUCH_TIME_ERROR &&
240 inode_change_ok(inode, iap) != 0) {
241 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
242 * this will cause notify_change to set these times to "now"
243 */
244 iap->ia_valid &= ~BOTH_TIME_SET;
245 }
246 }
247
248 /* The size case is special. It changes the file as well as the attributes. */
249 if (iap->ia_valid & ATTR_SIZE) {
250 if (iap->ia_size < inode->i_size) {
251 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
252 if (err)
253 goto out;
254 }
255
256 /*
257 * If we are changing the size of the file, then
258 * we need to break all leases.
259 */
260 err = get_lease(inode, FMODE_WRITE);
261 if (err)
262 goto out_nfserr;
263
264 err = get_write_access(inode);
265 if (err)
266 goto out_nfserr;
267
268 err = locks_verify_truncate(inode, NULL, iap->ia_size);
269 if (err) {
270 put_write_access(inode);
271 goto out_nfserr;
272 }
273 DQUOT_INIT(inode);
274 }
275
276 imode = inode->i_mode;
277 if (iap->ia_valid & ATTR_MODE) {
278 iap->ia_mode &= S_IALLUGO;
279 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
280 }
281
282 /* Revoke setuid/setgid bit on chown/chgrp */
283 if ((iap->ia_valid & ATTR_UID) && (imode & S_ISUID)
284 && iap->ia_uid != inode->i_uid) {
285 iap->ia_valid |= ATTR_MODE;
286 iap->ia_mode = imode &= ~S_ISUID;
287 }
288 if ((iap->ia_valid & ATTR_GID) && (imode & S_ISGID)
289 && iap->ia_gid != inode->i_gid) {
290 iap->ia_valid |= ATTR_MODE;
291 iap->ia_mode = imode &= ~S_ISGID;
292 }
293
294 /* Change the attributes. */
295
296
297 iap->ia_valid |= ATTR_CTIME;
298
299 if (iap->ia_valid & ATTR_SIZE) {
300 fh_lock(fhp);
301 size_change = 1;
302 }
303 err = nfserr_notsync;
304 if (!check_guard || guardtime == inode->i_ctime) {
305 err = notify_change(dentry, iap);
306 err = nfserrno(err);
307 }
308 if (size_change) {
309 fh_unlock(fhp);
310 put_write_access(inode);
311 }
312 if (!err)
313 if (EX_ISSYNC(fhp->fh_export))
314 write_inode_now(inode, 1);
315 out:
316 return err;
317
318 out_nfserr:
319 err = nfserrno(err);
320 goto out;
321 }
322
323 #ifdef CONFIG_NFSD_V3
324 /*
325 * Check server access rights to a file system object
326 */
327 struct accessmap {
328 u32 access;
329 int how;
330 };
331 static struct accessmap nfs3_regaccess[] = {
332 { NFS3_ACCESS_READ, MAY_READ },
333 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
334 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
335 { NFS3_ACCESS_EXTEND, MAY_WRITE },
336
337 { 0, 0 }
338 };
339
340 static struct accessmap nfs3_diraccess[] = {
341 { NFS3_ACCESS_READ, MAY_READ },
342 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
343 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
344 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
345 { NFS3_ACCESS_DELETE, MAY_REMOVE },
346
347 { 0, 0 }
348 };
349
350 static struct accessmap nfs3_anyaccess[] = {
351 /* Some clients - Solaris 2.6 at least, make an access call
352 * to the server to check for access for things like /dev/null
353 * (which really, the server doesn't care about). So
354 * We provide simple access checking for them, looking
355 * mainly at mode bits
356 */
357 { NFS3_ACCESS_READ, MAY_READ },
358 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
359 { NFS3_ACCESS_MODIFY, MAY_WRITE },
360 { NFS3_ACCESS_EXTEND, MAY_WRITE },
361
362 { 0, 0 }
363 };
364
365 int
366 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access)
367 {
368 struct accessmap *map;
369 struct svc_export *export;
370 struct dentry *dentry;
371 u32 query, result = 0;
372 unsigned int error;
373
374 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
375 if (error)
376 goto out;
377
378 export = fhp->fh_export;
379 dentry = fhp->fh_dentry;
380
381 if (S_ISREG(dentry->d_inode->i_mode))
382 map = nfs3_regaccess;
383 else if (S_ISDIR(dentry->d_inode->i_mode))
384 map = nfs3_diraccess;
385 else
386 map = nfs3_anyaccess;
387
388
389 query = *access;
390 for (; map->access; map++) {
391 if (map->access & query) {
392 unsigned int err2;
393 err2 = nfsd_permission(export, dentry, map->how);
394 switch (err2) {
395 case nfs_ok:
396 result |= map->access;
397 break;
398
399 /* the following error codes just mean the access was not allowed,
400 * rather than an error occurred */
401 case nfserr_rofs:
402 case nfserr_acces:
403 case nfserr_perm:
404 /* simply don't "or" in the access bit. */
405 break;
406 default:
407 error = err2;
408 goto out;
409 }
410 }
411 }
412 *access = result;
413
414 out:
415 return error;
416 }
417 #endif /* CONFIG_NFSD_V3 */
418
419
420
421 /*
422 * Open an existing file or directory.
423 * The access argument indicates the type of open (read/write/lock)
424 * N.B. After this call fhp needs an fh_put
425 */
426 int
427 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
428 int access, struct file *filp)
429 {
430 struct dentry *dentry;
431 struct inode *inode;
432 int err;
433
434 /* If we get here, then the client has already done an "open", and (hopefully)
435 * checked permission - so allow OWNER_OVERRIDE in case a chmod has now revoked
436 * permission */
437 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
438 if (err)
439 goto out;
440
441 dentry = fhp->fh_dentry;
442 inode = dentry->d_inode;
443
444 /* Disallow access to files with the append-only bit set or
445 * with mandatory locking enabled
446 */
447 err = nfserr_perm;
448 if (IS_APPEND(inode) || IS_ISMNDLK(inode))
449 goto out;
450 if (!inode->i_fop)
451 goto out;
452
453 /*
454 * Check to see if there are any leases on this file.
455 * This may block while leases are broken.
456 */
457 err = get_lease(inode, (access & MAY_WRITE) ? FMODE_WRITE : 0);
458 if (err)
459 goto out_nfserr;
460
461 if ((access & MAY_WRITE) && (err = get_write_access(inode)) != 0)
462 goto out_nfserr;
463
464 memset(filp, 0, sizeof(*filp));
465 filp->f_op = fops_get(inode->i_fop);
466 atomic_set(&filp->f_count, 1);
467 filp->f_dentry = dentry;
468 filp->f_vfsmnt = fhp->fh_export->ex_mnt;
469 if (access & MAY_WRITE) {
470 filp->f_flags = O_WRONLY|O_LARGEFILE;
471 filp->f_mode = FMODE_WRITE;
472 DQUOT_INIT(inode);
473 } else {
474 filp->f_flags = O_RDONLY|O_LARGEFILE;
475 filp->f_mode = FMODE_READ;
476 }
477
478 err = 0;
479 if (filp->f_op && filp->f_op->open) {
480 err = filp->f_op->open(inode, filp);
481 if (err) {
482 fops_put(filp->f_op);
483 if (access & MAY_WRITE)
484 put_write_access(inode);
485
486 /* I nearly added put_filp() call here, but this filp
487 * is really on callers stack frame. -DaveM
488 */
489 atomic_dec(&filp->f_count);
490 }
491 }
492 out_nfserr:
493 if (err)
494 err = nfserrno(err);
495 out:
496 return err;
497 }
498
499 /*
500 * Close a file.
501 */
502 void
503 nfsd_close(struct file *filp)
504 {
505 struct dentry *dentry = filp->f_dentry;
506 struct inode *inode = dentry->d_inode;
507
508 if (filp->f_op && filp->f_op->release)
509 filp->f_op->release(inode, filp);
510 fops_put(filp->f_op);
511 if (filp->f_mode & FMODE_WRITE)
512 put_write_access(inode);
513 }
514
515 /*
516 * Sync a file
517 * As this calls fsync (not fdatasync) there is no need for a write_inode
518 * after it.
519 */
520 inline void nfsd_dosync(struct file *filp, struct dentry *dp,
521 struct file_operations *fop)
522 {
523 struct inode *inode = dp->d_inode;
524 int (*fsync) (struct file *, struct dentry *, int);
525
526 filemap_fdatasync(inode->i_mapping);
527 if (fop && (fsync = fop->fsync))
528 fsync(filp, dp, 0);
529 filemap_fdatawait(inode->i_mapping);
530 }
531
532
533 void
534 nfsd_sync(struct file *filp)
535 {
536 struct inode *inode = filp->f_dentry->d_inode;
537 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
538 down(&inode->i_sem);
539 nfsd_dosync(filp, filp->f_dentry, filp->f_op);
540 up(&inode->i_sem);
541 }
542
543 void
544 nfsd_sync_dir(struct dentry *dp)
545 {
546 nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
547 }
548
549 /*
550 * Obtain the readahead parameters for the file
551 * specified by (dev, ino).
552 */
553 static inline struct raparms *
554 nfsd_get_raparms(dev_t dev, ino_t ino)
555 {
556 struct raparms *ra, **rap, **frap = NULL;
557 int depth = 0;
558
559 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
560 if (ra->p_ino == ino && ra->p_dev == dev)
561 goto found;
562 depth++;
563 if (ra->p_count == 0)
564 frap = rap;
565 }
566 depth = nfsdstats.ra_size*11/10;
567 if (!frap)
568 return NULL;
569 rap = frap;
570 ra = *frap;
571 ra->p_dev = dev;
572 ra->p_ino = ino;
573 ra->p_reada = 0;
574 ra->p_ramax = 0;
575 ra->p_raend = 0;
576 ra->p_ralen = 0;
577 ra->p_rawin = 0;
578 found:
579 if (rap != &raparm_cache) {
580 *rap = ra->p_next;
581 ra->p_next = raparm_cache;
582 raparm_cache = ra;
583 }
584 ra->p_count++;
585 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
586 return ra;
587 }
588
589 /*
590 * Read data from a file. count must contain the requested read count
591 * on entry. On return, *count contains the number of bytes actually read.
592 * N.B. After this call fhp needs an fh_put
593 */
594 int
595 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
596 char *buf, unsigned long *count)
597 {
598 struct raparms *ra;
599 mm_segment_t oldfs;
600 int err;
601 struct file file;
602
603 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
604 if (err)
605 goto out;
606 err = nfserr_perm;
607 if (!file.f_op->read)
608 goto out_close;
609 #ifdef MSNFS
610 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
611 (!lock_may_read(file.f_dentry->d_inode, offset, *count)))
612 goto out_close;
613 #endif
614
615 /* Get readahead parameters */
616 ra = nfsd_get_raparms(fhp->fh_export->ex_dev, fhp->fh_dentry->d_inode->i_ino);
617 if (ra) {
618 file.f_reada = ra->p_reada;
619 file.f_ramax = ra->p_ramax;
620 file.f_raend = ra->p_raend;
621 file.f_ralen = ra->p_ralen;
622 file.f_rawin = ra->p_rawin;
623 }
624 file.f_pos = offset;
625
626 oldfs = get_fs(); set_fs(KERNEL_DS);
627 err = file.f_op->read(&file, buf, *count, &file.f_pos);
628 set_fs(oldfs);
629
630 /* Write back readahead params */
631 if (ra != NULL) {
632 dprintk("nfsd: raparms %ld %ld %ld %ld %ld\n",
633 file.f_reada, file.f_ramax, file.f_raend,
634 file.f_ralen, file.f_rawin);
635 ra->p_reada = file.f_reada;
636 ra->p_ramax = file.f_ramax;
637 ra->p_raend = file.f_raend;
638 ra->p_ralen = file.f_ralen;
639 ra->p_rawin = file.f_rawin;
640 ra->p_count -= 1;
641 }
642
643 if (err >= 0) {
644 nfsdstats.io_read += err;
645 *count = err;
646 err = 0;
647 } else
648 err = nfserrno(err);
649 out_close:
650 nfsd_close(&file);
651 out:
652 return err;
653 }
654
655 /*
656 * Write data to a file.
657 * The stable flag requests synchronous writes.
658 * N.B. After this call fhp needs an fh_put
659 */
660 int
661 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
662 char *buf, unsigned long cnt, int *stablep)
663 {
664 struct svc_export *exp;
665 struct file file;
666 struct dentry *dentry;
667 struct inode *inode;
668 mm_segment_t oldfs;
669 int err = 0;
670 int stable = *stablep;
671
672 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
673 if (err)
674 goto out;
675 if (!cnt)
676 goto out_close;
677 err = nfserr_perm;
678 if (!file.f_op->write)
679 goto out_close;
680 #ifdef MSNFS
681 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
682 (!lock_may_write(file.f_dentry->d_inode, offset, cnt)))
683 goto out_close;
684 #endif
685
686 dentry = file.f_dentry;
687 inode = dentry->d_inode;
688 exp = fhp->fh_export;
689
690 /*
691 * Request sync writes if
692 * - the sync export option has been set, or
693 * - the client requested O_SYNC behavior (NFSv3 feature).
694 * - The file system doesn't support fsync().
695 * When gathered writes have been configured for this volume,
696 * flushing the data to disk is handled separately below.
697 */
698
699 if (file.f_op->fsync == 0) {/* COMMIT3 cannot work */
700 stable = 2;
701 *stablep = 2; /* FILE_SYNC */
702 }
703
704 if (!EX_ISSYNC(exp))
705 stable = 0;
706 if (stable && !EX_WGATHER(exp))
707 file.f_flags |= O_SYNC;
708
709 file.f_pos = offset; /* set write offset */
710
711 /* Write the data. */
712 oldfs = get_fs(); set_fs(KERNEL_DS);
713 err = file.f_op->write(&file, buf, cnt, &file.f_pos);
714 if (err >= 0)
715 nfsdstats.io_write += cnt;
716 set_fs(oldfs);
717
718 /* clear setuid/setgid flag after write */
719 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) {
720 struct iattr ia;
721
722 ia.ia_valid = ATTR_MODE;
723 ia.ia_mode = inode->i_mode & ~(S_ISUID | S_ISGID);
724 notify_change(dentry, &ia);
725 }
726
727 if (err >= 0 && stable) {
728 static unsigned long last_ino;
729 static kdev_t last_dev = NODEV;
730
731 /*
732 * Gathered writes: If another process is currently
733 * writing to the file, there's a high chance
734 * this is another nfsd (triggered by a bulk write
735 * from a client's biod). Rather than syncing the
736 * file with each write request, we sleep for 10 msec.
737 *
738 * I don't know if this roughly approximates
739 * C. Juszak's idea of gathered writes, but it's a
740 * nice and simple solution (IMHO), and it seems to
741 * work:-)
742 */
743 if (EX_WGATHER(exp)) {
744 if (atomic_read(&inode->i_writecount) > 1
745 || (last_ino == inode->i_ino && last_dev == inode->i_dev)) {
746 dprintk("nfsd: write defer %d\n", current->pid);
747 set_current_state(TASK_UNINTERRUPTIBLE);
748 schedule_timeout((HZ+99)/100);
749 current->state = TASK_RUNNING;
750 dprintk("nfsd: write resume %d\n", current->pid);
751 }
752
753 if (inode->i_state & I_DIRTY) {
754 dprintk("nfsd: write sync %d\n", current->pid);
755 nfsd_sync(&file);
756 }
757 #if 0
758 wake_up(&inode->i_wait);
759 #endif
760 }
761 last_ino = inode->i_ino;
762 last_dev = inode->i_dev;
763 }
764
765 dprintk("nfsd: write complete err=%d\n", err);
766 if (err >= 0)
767 err = 0;
768 else
769 err = nfserrno(err);
770 out_close:
771 nfsd_close(&file);
772 out:
773 return err;
774 }
775
776
777 #ifdef CONFIG_NFSD_V3
778 /*
779 * Commit all pending writes to stable storage.
780 * Strictly speaking, we could sync just the indicated file region here,
781 * but there's currently no way we can ask the VFS to do so.
782 *
783 * Unfortunately we cannot lock the file to make sure we return full WCC
784 * data to the client, as locking happens lower down in the filesystem.
785 */
786 int
787 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
788 off_t offset, unsigned long count)
789 {
790 struct file file;
791 int err;
792
793 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
794 return err;
795 if (EX_ISSYNC(fhp->fh_export)) {
796 if (file.f_op && file.f_op->fsync) {
797 nfsd_sync(&file);
798 } else {
799 err = nfserr_notsupp;
800 }
801 }
802
803 nfsd_close(&file);
804 return err;
805 }
806 #endif /* CONFIG_NFSD_V3 */
807
808 /*
809 * Create a file (regular, directory, device, fifo); UNIX sockets
810 * not yet implemented.
811 * If the response fh has been verified, the parent directory should
812 * already be locked. Note that the parent directory is left locked.
813 *
814 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
815 */
816 int
817 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
818 char *fname, int flen, struct iattr *iap,
819 int type, dev_t rdev, struct svc_fh *resfhp)
820 {
821 struct dentry *dentry, *dchild;
822 struct inode *dirp;
823 int err;
824
825 err = nfserr_perm;
826 if (!flen)
827 goto out;
828 err = nfserr_exist;
829 if (isdotent(fname, flen))
830 goto out;
831
832 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
833 if (err)
834 goto out;
835
836 dentry = fhp->fh_dentry;
837 dirp = dentry->d_inode;
838
839 err = nfserr_notdir;
840 if(!dirp->i_op || !dirp->i_op->lookup)
841 goto out;
842 /*
843 * Check whether the response file handle has been verified yet.
844 * If it has, the parent directory should already be locked.
845 */
846 if (!resfhp->fh_dentry) {
847 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
848 fh_lock(fhp);
849 dchild = lookup_one_len(fname, dentry, flen);
850 err = PTR_ERR(dchild);
851 if (IS_ERR(dchild))
852 goto out_nfserr;
853 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
854 if (err)
855 goto out;
856 } else {
857 /* called from nfsd_proc_create */
858 dchild = resfhp->fh_dentry;
859 if (!fhp->fh_locked) {
860 /* not actually possible */
861 printk(KERN_ERR
862 "nfsd_create: parent %s/%s not locked!\n",
863 dentry->d_parent->d_name.name,
864 dentry->d_name.name);
865 err = -EIO;
866 goto out;
867 }
868 }
869 /*
870 * Make sure the child dentry is still negative ...
871 */
872 err = nfserr_exist;
873 if (dchild->d_inode) {
874 dprintk("nfsd_create: dentry %s/%s not negative!\n",
875 dentry->d_name.name, dchild->d_name.name);
876 goto out;
877 }
878
879 if (!(iap->ia_valid & ATTR_MODE))
880 iap->ia_mode = 0;
881 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
882
883 /*
884 * Get the dir op function pointer.
885 */
886 err = nfserr_perm;
887 switch (type) {
888 case S_IFREG:
889 err = vfs_create(dirp, dchild, iap->ia_mode);
890 break;
891 case S_IFDIR:
892 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
893 break;
894 case S_IFCHR:
895 case S_IFBLK:
896 case S_IFIFO:
897 case S_IFSOCK:
898 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
899 break;
900 default:
901 printk("nfsd: bad file type %o in nfsd_create\n", type);
902 err = -EINVAL;
903 }
904 if (err < 0)
905 goto out_nfserr;
906
907 if (EX_ISSYNC(fhp->fh_export)) {
908 nfsd_sync_dir(dentry);
909 write_inode_now(dchild->d_inode, 1);
910 }
911
912
913 /* Set file attributes. Mode has already been set and
914 * setting uid/gid works only for root. Irix appears to
915 * send along the gid when it tries to implement setgid
916 * directories via NFS.
917 */
918 err = 0;
919 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
920 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
921 /*
922 * Update the file handle to get the new inode info.
923 */
924 if (!err)
925 err = fh_update(resfhp);
926 out:
927 return err;
928
929 out_nfserr:
930 err = nfserrno(err);
931 goto out;
932 }
933
934 #ifdef CONFIG_NFSD_V3
935 /*
936 * NFSv3 version of nfsd_create
937 */
938 int
939 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
940 char *fname, int flen, struct iattr *iap,
941 struct svc_fh *resfhp, int createmode, u32 *verifier)
942 {
943 struct dentry *dentry, *dchild;
944 struct inode *dirp;
945 int err;
946 __u32 v_mtime=0, v_atime=0;
947 int v_mode=0;
948
949 err = nfserr_perm;
950 if (!flen)
951 goto out;
952 err = nfserr_exist;
953 if (isdotent(fname, flen))
954 goto out;
955 if (!(iap->ia_valid & ATTR_MODE))
956 iap->ia_mode = 0;
957 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
958 if (err)
959 goto out;
960
961 dentry = fhp->fh_dentry;
962 dirp = dentry->d_inode;
963
964 /* Get all the sanity checks out of the way before
965 * we lock the parent. */
966 err = nfserr_notdir;
967 if(!dirp->i_op || !dirp->i_op->lookup)
968 goto out;
969 fh_lock(fhp);
970
971 /*
972 * Compose the response file handle.
973 */
974 dchild = lookup_one_len(fname, dentry, flen);
975 err = PTR_ERR(dchild);
976 if (IS_ERR(dchild))
977 goto out_nfserr;
978
979 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
980 if (err)
981 goto out;
982
983 if (createmode == NFS3_CREATE_EXCLUSIVE) {
984 /* while the verifier would fit in mtime+atime,
985 * solaris7 gets confused (bugid 4218508) if these have
986 * the high bit set, so we use the mode as well
987 */
988 v_mtime = verifier[0]&0x7fffffff;
989 v_atime = verifier[1]&0x7fffffff;
990 v_mode = S_IFREG
991 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
992 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
993 ;
994 }
995
996 if (dchild->d_inode) {
997 err = 0;
998
999 switch (createmode) {
1000 case NFS3_CREATE_UNCHECKED:
1001 if (! S_ISREG(dchild->d_inode->i_mode))
1002 err = nfserr_exist;
1003 else {
1004 iap->ia_valid &= ATTR_SIZE;
1005 goto set_attr;
1006 }
1007 break;
1008 case NFS3_CREATE_EXCLUSIVE:
1009 if ( dchild->d_inode->i_mtime == v_mtime
1010 && dchild->d_inode->i_atime == v_atime
1011 && dchild->d_inode->i_mode == v_mode
1012 && dchild->d_inode->i_size == 0 )
1013 break;
1014 /* fallthru */
1015 case NFS3_CREATE_GUARDED:
1016 err = nfserr_exist;
1017 }
1018 goto out;
1019 }
1020
1021 err = vfs_create(dirp, dchild, iap->ia_mode);
1022 if (err < 0)
1023 goto out_nfserr;
1024
1025 if (EX_ISSYNC(fhp->fh_export)) {
1026 nfsd_sync_dir(dentry);
1027 /* setattr will sync the child (or not) */
1028 }
1029
1030 /*
1031 * Update the filehandle to get the new inode info.
1032 */
1033 err = fh_update(resfhp);
1034 if (err)
1035 goto out;
1036
1037 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1038 /* Cram the verifier into atime/mtime/mode */
1039 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1040 | ATTR_MTIME_SET|ATTR_ATIME_SET
1041 | ATTR_MODE;
1042 iap->ia_mtime = v_mtime;
1043 iap->ia_atime = v_atime;
1044 iap->ia_mode = v_mode;
1045 }
1046
1047 /* Set file attributes.
1048 * Mode has already been set but we might need to reset it
1049 * for CREATE_EXCLUSIVE
1050 * Irix appears to send along the gid when it tries to
1051 * implement setgid directories via NFS. Clear out all that cruft.
1052 */
1053 set_attr:
1054 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1055 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1056
1057 out:
1058 fh_unlock(fhp);
1059 return err;
1060
1061 out_nfserr:
1062 err = nfserrno(err);
1063 goto out;
1064 }
1065 #endif /* CONFIG_NFSD_V3 */
1066
1067 /*
1068 * Read a symlink. On entry, *lenp must contain the maximum path length that
1069 * fits into the buffer. On return, it contains the true length.
1070 * N.B. After this call fhp needs an fh_put
1071 */
1072 int
1073 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1074 {
1075 struct dentry *dentry;
1076 struct inode *inode;
1077 mm_segment_t oldfs;
1078 int err;
1079
1080 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1081 if (err)
1082 goto out;
1083
1084 dentry = fhp->fh_dentry;
1085 inode = dentry->d_inode;
1086
1087 err = nfserr_inval;
1088 if (!inode->i_op || !inode->i_op->readlink)
1089 goto out;
1090
1091 UPDATE_ATIME(inode);
1092 /* N.B. Why does this call need a get_fs()??
1093 * Remove the set_fs and watch the fireworks:-) --okir
1094 */
1095
1096 oldfs = get_fs(); set_fs(KERNEL_DS);
1097 err = inode->i_op->readlink(dentry, buf, *lenp);
1098 set_fs(oldfs);
1099
1100 if (err < 0)
1101 goto out_nfserr;
1102 *lenp = err;
1103 err = 0;
1104 out:
1105 return err;
1106
1107 out_nfserr:
1108 err = nfserrno(err);
1109 goto out;
1110 }
1111
1112 /*
1113 * Create a symlink and look up its inode
1114 * N.B. After this call _both_ fhp and resfhp need an fh_put
1115 */
1116 int
1117 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1118 char *fname, int flen,
1119 char *path, int plen,
1120 struct svc_fh *resfhp,
1121 struct iattr *iap)
1122 {
1123 struct dentry *dentry, *dnew;
1124 int err, cerr;
1125
1126 err = nfserr_noent;
1127 if (!flen || !plen)
1128 goto out;
1129 err = nfserr_exist;
1130 if (isdotent(fname, flen))
1131 goto out;
1132
1133 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1134 if (err)
1135 goto out;
1136 fh_lock(fhp);
1137 dentry = fhp->fh_dentry;
1138 dnew = lookup_one_len(fname, dentry, flen);
1139 err = PTR_ERR(dnew);
1140 if (IS_ERR(dnew))
1141 goto out_nfserr;
1142
1143 err = vfs_symlink(dentry->d_inode, dnew, path);
1144 if (!err) {
1145 if (EX_ISSYNC(fhp->fh_export))
1146 nfsd_sync_dir(dentry);
1147 if (iap) {
1148 iap->ia_valid &= ATTR_MODE /* ~(ATTR_MODE|ATTR_UID|ATTR_GID)*/;
1149 if (iap->ia_valid) {
1150 iap->ia_valid |= ATTR_CTIME;
1151 iap->ia_mode = (iap->ia_mode&S_IALLUGO)
1152 | S_IFLNK;
1153 err = notify_change(dnew, iap);
1154 if (err)
1155 err = nfserrno(err);
1156 else if (EX_ISSYNC(fhp->fh_export))
1157 write_inode_now(dentry->d_inode, 1);
1158 }
1159 }
1160 } else
1161 err = nfserrno(err);
1162 fh_unlock(fhp);
1163
1164 /* Compose the fh so the dentry will be freed ... */
1165 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1166 if (err==0) err = cerr;
1167 out:
1168 return err;
1169
1170 out_nfserr:
1171 err = nfserrno(err);
1172 goto out;
1173 }
1174
1175 /*
1176 * Create a hardlink
1177 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1178 */
1179 int
1180 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1181 char *name, int len, struct svc_fh *tfhp)
1182 {
1183 struct dentry *ddir, *dnew, *dold;
1184 struct inode *dirp, *dest;
1185 int err;
1186
1187 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1188 if (err)
1189 goto out;
1190 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1191 if (err)
1192 goto out;
1193
1194 err = nfserr_perm;
1195 if (!len)
1196 goto out;
1197 err = nfserr_exist;
1198 if (isdotent(name, len))
1199 goto out;
1200
1201 fh_lock(ffhp);
1202 ddir = ffhp->fh_dentry;
1203 dirp = ddir->d_inode;
1204
1205 dnew = lookup_one_len(name, ddir, len);
1206 err = PTR_ERR(dnew);
1207 if (IS_ERR(dnew))
1208 goto out_nfserr;
1209
1210 dold = tfhp->fh_dentry;
1211 dest = dold->d_inode;
1212
1213 err = vfs_link(dold, dirp, dnew);
1214 if (!err) {
1215 if (EX_ISSYNC(ffhp->fh_export)) {
1216 nfsd_sync_dir(ddir);
1217 write_inode_now(dest, 1);
1218 }
1219 } else {
1220 if (err == -EXDEV && rqstp->rq_vers == 2)
1221 err = nfserr_acces;
1222 else
1223 err = nfserrno(err);
1224 }
1225
1226 fh_unlock(ffhp);
1227 dput(dnew);
1228 out:
1229 return err;
1230
1231 out_nfserr:
1232 err = nfserrno(err);
1233 goto out;
1234 }
1235
1236 /*
1237 * Rename a file
1238 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1239 */
1240 int
1241 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1242 struct svc_fh *tfhp, char *tname, int tlen)
1243 {
1244 struct dentry *fdentry, *tdentry, *odentry, *ndentry;
1245 struct inode *fdir, *tdir;
1246 int err;
1247
1248 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1249 if (err)
1250 goto out;
1251 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1252 if (err)
1253 goto out;
1254
1255 fdentry = ffhp->fh_dentry;
1256 fdir = fdentry->d_inode;
1257
1258 tdentry = tfhp->fh_dentry;
1259 tdir = tdentry->d_inode;
1260
1261 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1262 if (fdir->i_dev != tdir->i_dev)
1263 goto out;
1264
1265 err = nfserr_perm;
1266 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1267 goto out;
1268
1269 /* cannot use fh_lock as we need deadlock protective ordering
1270 * so do it by hand */
1271 double_down(&tdir->i_sem, &fdir->i_sem);
1272 ffhp->fh_locked = tfhp->fh_locked = 1;
1273 fill_pre_wcc(ffhp);
1274 fill_pre_wcc(tfhp);
1275
1276 odentry = lookup_one_len(fname, fdentry, flen);
1277 err = PTR_ERR(odentry);
1278 if (IS_ERR(odentry))
1279 goto out_nfserr;
1280
1281 err = -ENOENT;
1282 if (!odentry->d_inode)
1283 goto out_dput_old;
1284
1285 ndentry = lookup_one_len(tname, tdentry, tlen);
1286 err = PTR_ERR(ndentry);
1287 if (IS_ERR(ndentry))
1288 goto out_dput_old;
1289
1290
1291 #ifdef MSNFS
1292 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1293 ((atomic_read(&odentry->d_count) > 1)
1294 || (atomic_read(&ndentry->d_count) > 1))) {
1295 err = nfserr_perm;
1296 } else
1297 #endif
1298 err = vfs_rename(fdir, odentry, tdir, ndentry);
1299 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1300 nfsd_sync_dir(tdentry);
1301 nfsd_sync_dir(fdentry);
1302 }
1303 dput(ndentry);
1304
1305 out_dput_old:
1306 dput(odentry);
1307 out_nfserr:
1308 if (err)
1309 err = nfserrno(err);
1310
1311 /* we cannot reply on fh_unlock on the two filehandles,
1312 * as that would do the wrong thing if the two directories
1313 * were the same, so again we do it by hand
1314 */
1315 fill_post_wcc(ffhp);
1316 fill_post_wcc(tfhp);
1317 double_up(&tdir->i_sem, &fdir->i_sem);
1318 ffhp->fh_locked = tfhp->fh_locked = 0;
1319
1320 out:
1321 return err;
1322 }
1323
1324 /*
1325 * Unlink a file or directory
1326 * N.B. After this call fhp needs an fh_put
1327 */
1328 int
1329 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1330 char *fname, int flen)
1331 {
1332 struct dentry *dentry, *rdentry;
1333 struct inode *dirp;
1334 int err;
1335
1336 err = nfserr_acces;
1337 if (!flen || isdotent(fname, flen))
1338 goto out;
1339 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1340 if (err)
1341 goto out;
1342
1343 fh_lock(fhp);
1344 dentry = fhp->fh_dentry;
1345 dirp = dentry->d_inode;
1346
1347 rdentry = lookup_one_len(fname, dentry, flen);
1348 err = PTR_ERR(rdentry);
1349 if (IS_ERR(rdentry))
1350 goto out_nfserr;
1351
1352 if (!rdentry->d_inode) {
1353 dput(rdentry);
1354 err = nfserr_noent;
1355 goto out;
1356 }
1357
1358 if (type != S_IFDIR) { /* It's UNLINK */
1359 #ifdef MSNFS
1360 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1361 (atomic_read(&rdentry->d_count) > 1)) {
1362 err = nfserr_perm;
1363 } else
1364 #endif
1365 err = vfs_unlink(dirp, rdentry);
1366 } else { /* It's RMDIR */
1367 err = vfs_rmdir(dirp, rdentry);
1368 }
1369
1370 dput(rdentry);
1371
1372 if (err)
1373 goto out_nfserr;
1374 if (EX_ISSYNC(fhp->fh_export))
1375 nfsd_sync_dir(dentry);
1376
1377 out:
1378 return err;
1379
1380 out_nfserr:
1381 err = nfserrno(err);
1382 goto out;
1383 }
1384
1385 /*
1386 * Read entries from a directory.
1387 * The verifier is an NFSv3 thing we ignore for now.
1388 */
1389 int
1390 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
1391 encode_dent_fn func, u32 *buffer, int *countp, u32 *verf)
1392 {
1393 u32 *p;
1394 int oldlen, eof, err;
1395 struct file file;
1396 struct readdir_cd cd;
1397
1398 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1399 if (err)
1400 goto out;
1401 if (offset > ~(u32) 0)
1402 goto out_close;
1403
1404 file.f_pos = offset;
1405
1406 /* Set up the readdir context */
1407 memset(&cd, 0, sizeof(cd));
1408 cd.rqstp = rqstp;
1409 cd.buffer = buffer;
1410 cd.buflen = *countp; /* count of words */
1411 cd.dirfh = fhp;
1412
1413 /*
1414 * Read the directory entries. This silly loop is necessary because
1415 * readdir() is not guaranteed to fill up the entire buffer, but
1416 * may choose to do less.
1417 */
1418
1419 do {
1420 oldlen = cd.buflen;
1421
1422 err = vfs_readdir(&file, (filldir_t) func, &cd);
1423
1424 if (err < 0)
1425 goto out_nfserr;
1426
1427 } while (oldlen != cd.buflen && !cd.eob);
1428
1429 /* If we didn't fill the buffer completely, we're at EOF */
1430 eof = !cd.eob;
1431
1432 if (cd.offset) {
1433 if (rqstp->rq_vers == 3)
1434 (void)xdr_encode_hyper(cd.offset, file.f_pos);
1435 else
1436 *cd.offset = htonl(file.f_pos);
1437 }
1438
1439 p = cd.buffer;
1440 *p++ = 0; /* no more entries */
1441 *p++ = htonl(eof); /* end of directory */
1442 *countp = (caddr_t) p - (caddr_t) buffer;
1443
1444 dprintk("nfsd: readdir result %d bytes, eof %d offset %d\n",
1445 *countp, eof,
1446 cd.offset? ntohl(*cd.offset) : -1);
1447 err = 0;
1448 out_close:
1449 nfsd_close(&file);
1450 out:
1451 return err;
1452
1453 out_nfserr:
1454 err = nfserrno(err);
1455 goto out_close;
1456 }
1457
1458 /*
1459 * Get file system stats
1460 * N.B. After this call fhp needs an fh_put
1461 */
1462 int
1463 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct statfs *stat)
1464 {
1465 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1466 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1467 err = nfserr_io;
1468 return err;
1469 }
1470
1471 /*
1472 * Check for a user's access permissions to this inode.
1473 */
1474 int
1475 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1476 {
1477 struct inode *inode = dentry->d_inode;
1478 int err;
1479
1480 if (acc == MAY_NOP)
1481 return 0;
1482 #if 0
1483 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1484 acc,
1485 (acc & MAY_READ)? " read" : "",
1486 (acc & MAY_WRITE)? " write" : "",
1487 (acc & MAY_EXEC)? " exec" : "",
1488 (acc & MAY_SATTR)? " sattr" : "",
1489 (acc & MAY_TRUNC)? " trunc" : "",
1490 (acc & MAY_LOCK)? " lock" : "",
1491 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1492 inode->i_mode,
1493 IS_IMMUTABLE(inode)? " immut" : "",
1494 IS_APPEND(inode)? " append" : "",
1495 IS_RDONLY(inode)? " ro" : "");
1496 dprintk(" owner %d/%d user %d/%d\n",
1497 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1498 #endif
1499
1500 /* The following code is here to make IRIX happy, which
1501 * does a permission check every time a user does
1502 * echo yaddayadda > special-file
1503 * by sending a CREATE request.
1504 * The original code would check read-only export status
1505 * only for regular files and directories, allowing
1506 * clients to chown/chmod device files and fifos even
1507 * on volumes exported read-only. */
1508 if (!(acc & _NFSD_IRIX_BOGOSITY)
1509 && (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC))) {
1510 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1511 return nfserr_rofs;
1512 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1513 return nfserr_perm;
1514 }
1515 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1516 return nfserr_perm;
1517
1518 if (acc & MAY_LOCK) {
1519 /* If we cannot rely on authentication in NLM requests,
1520 * just allow locks, otherwise require read permission, or
1521 * ownership
1522 */
1523 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1524 return 0;
1525 else
1526 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1527 }
1528 /*
1529 * The file owner always gets access permission for accesses that
1530 * would normally be checked at open time. This is to make
1531 * file access work even when the client has done a fchmod(fd, 0).
1532 *
1533 * However, `cp foo bar' should fail nevertheless when bar is
1534 * readonly. A sensible way to do this might be to reject all
1535 * attempts to truncate a read-only file, because a creat() call
1536 * always implies file truncation.
1537 * ... but this isn't really fair. A process may reasonably call
1538 * ftruncate on an open file descriptor on a file with perm 000.
1539 * We must trust the client to do permission checking - using "ACCESS"
1540 * with NFSv3.
1541 */
1542 if ((acc & MAY_OWNER_OVERRIDE) &&
1543 inode->i_uid == current->fsuid)
1544 return 0;
1545
1546 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
1547
1548 /* Allow read access to binaries even when mode 111 */
1549 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1550 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1551 err = permission(inode, MAY_EXEC);
1552
1553 return err? nfserrno(err) : 0;
1554 }
1555
1556 void
1557 nfsd_racache_shutdown(void)
1558 {
1559 if (!raparm_cache)
1560 return;
1561 dprintk("nfsd: freeing readahead buffers.\n");
1562 kfree(raparml);
1563 raparm_cache = raparml = NULL;
1564 }
1565 /*
1566 * Initialize readahead param cache
1567 */
1568 int
1569 nfsd_racache_init(int cache_size)
1570 {
1571 int i;
1572
1573 if (raparm_cache)
1574 return 0;
1575 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1576
1577 if (raparml != NULL) {
1578 dprintk("nfsd: allocating %d readahead buffers.\n",
1579 cache_size);
1580 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1581 for (i = 0; i < cache_size - 1; i++) {
1582 raparml[i].p_next = raparml + i + 1;
1583 }
1584 raparm_cache = raparml;
1585 } else {
1586 printk(KERN_WARNING
1587 "nfsd: Could not allocate memory read-ahead cache.\n");
1588 return -ENOMEM;
1589 }
1590 nfsdstats.ra_size = cache_size;
1591 return 0;
1592 }
Cache object: 25934574cac5c76f79d075bc8a9af860
|