blob: 4f2cd3d2756665a1f600f8bcac96b4f2756b1ded [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#define MSNFS /* HACK HACK */
2/*
3 * linux/fs/nfsd/vfs.c
4 *
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
8 *
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
14 *
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 */
18
19#include <linux/config.h>
20#include <linux/string.h>
21#include <linux/time.h>
22#include <linux/errno.h>
23#include <linux/fs.h>
24#include <linux/file.h>
25#include <linux/mount.h>
26#include <linux/major.h>
27#include <linux/ext2_fs.h>
28#include <linux/proc_fs.h>
29#include <linux/stat.h>
30#include <linux/fcntl.h>
31#include <linux/net.h>
32#include <linux/unistd.h>
33#include <linux/slab.h>
34#include <linux/pagemap.h>
35#include <linux/in.h>
36#include <linux/module.h>
37#include <linux/namei.h>
38#include <linux/vfs.h>
39#include <linux/delay.h>
40#include <linux/sunrpc/svc.h>
41#include <linux/nfsd/nfsd.h>
42#ifdef CONFIG_NFSD_V3
43#include <linux/nfs3.h>
44#include <linux/nfsd/xdr3.h>
45#endif /* CONFIG_NFSD_V3 */
46#include <linux/nfsd/nfsfh.h>
47#include <linux/quotaops.h>
Robert Love0eeca282005-07-12 17:06:03 -040048#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/posix_acl.h>
50#include <linux/posix_acl_xattr.h>
Christoph Hellwig334a13e2005-06-28 20:44:58 -070051#ifdef CONFIG_NFSD_V4
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/xattr.h>
53#include <linux/nfs4.h>
54#include <linux/nfs4_acl.h>
55#include <linux/nfsd_idmap.h>
56#include <linux/security.h>
57#endif /* CONFIG_NFSD_V4 */
58
59#include <asm/uaccess.h>
60
61#define NFSDDBG_FACILITY NFSDDBG_FILEOP
62#define NFSD_PARANOIA
63
64
65/* We must ignore files (but only files) which might have mandatory
66 * locks on them because there is no way to know if the accesser has
67 * the lock.
68 */
69#define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
70
71/*
72 * This is a cache of readahead params that help us choose the proper
73 * readahead strategy. Initially, we set all readahead parameters to 0
74 * and let the VFS handle things.
75 * If you increase the number of cached files very much, you'll need to
76 * add a hash table here.
77 */
78struct raparms {
79 struct raparms *p_next;
80 unsigned int p_count;
81 ino_t p_ino;
82 dev_t p_dev;
83 int p_set;
84 struct file_ra_state p_ra;
85};
86
87static struct raparms * raparml;
88static struct raparms * raparm_cache;
89
90/*
91 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
92 * a mount point.
93 * Returns -EAGAIN leaving *dpp and *expp unchanged,
94 * or nfs_ok having possibly changed *dpp and *expp
95 */
96int
97nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
98 struct svc_export **expp)
99{
100 struct svc_export *exp = *expp, *exp2 = NULL;
101 struct dentry *dentry = *dpp;
102 struct vfsmount *mnt = mntget(exp->ex_mnt);
103 struct dentry *mounts = dget(dentry);
104 int err = nfs_ok;
105
106 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
107
108 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
109 if (IS_ERR(exp2)) {
110 err = PTR_ERR(exp2);
111 dput(mounts);
112 mntput(mnt);
113 goto out;
114 }
115 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
116 /* successfully crossed mount point */
117 exp_put(exp);
118 *expp = exp2;
119 dput(dentry);
120 *dpp = mounts;
121 } else {
122 if (exp2) exp_put(exp2);
123 dput(mounts);
124 }
125 mntput(mnt);
126out:
127 return err;
128}
129
130/*
131 * Look up one component of a pathname.
132 * N.B. After this call _both_ fhp and resfh need an fh_put
133 *
134 * If the lookup would cross a mountpoint, and the mounted filesystem
135 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
136 * accepted as it stands and the mounted directory is
137 * returned. Otherwise the covered directory is returned.
138 * NOTE: this mountpoint crossing is not supported properly by all
139 * clients and is explicitly disallowed for NFSv3
140 * NeilBrown <neilb@cse.unsw.edu.au>
141 */
142int
143nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
144 int len, struct svc_fh *resfh)
145{
146 struct svc_export *exp;
147 struct dentry *dparent;
148 struct dentry *dentry;
149 int err;
150
151 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
152
153 /* Obtain dentry and export. */
154 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
155 if (err)
156 return err;
157
158 dparent = fhp->fh_dentry;
159 exp = fhp->fh_export;
160 exp_get(exp);
161
162 err = nfserr_acces;
163
164 /* Lookup the name, but don't follow links */
165 if (isdotent(name, len)) {
166 if (len==1)
167 dentry = dget(dparent);
168 else if (dparent != exp->ex_dentry) {
169 dentry = dget_parent(dparent);
170 } else if (!EX_NOHIDE(exp))
171 dentry = dget(dparent); /* .. == . just like at / */
172 else {
173 /* checking mountpoint crossing is very different when stepping up */
174 struct svc_export *exp2 = NULL;
175 struct dentry *dp;
176 struct vfsmount *mnt = mntget(exp->ex_mnt);
177 dentry = dget(dparent);
178 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
179 ;
180 dp = dget_parent(dentry);
181 dput(dentry);
182 dentry = dp;
183
184 exp2 = exp_parent(exp->ex_client, mnt, dentry,
185 &rqstp->rq_chandle);
186 if (IS_ERR(exp2)) {
187 err = PTR_ERR(exp2);
188 dput(dentry);
189 mntput(mnt);
190 goto out_nfserr;
191 }
192 if (!exp2) {
193 dput(dentry);
194 dentry = dget(dparent);
195 } else {
196 exp_put(exp);
197 exp = exp2;
198 }
199 mntput(mnt);
200 }
201 } else {
202 fh_lock(fhp);
203 dentry = lookup_one_len(name, dparent, len);
204 err = PTR_ERR(dentry);
205 if (IS_ERR(dentry))
206 goto out_nfserr;
207 /*
208 * check if we have crossed a mount point ...
209 */
210 if (d_mountpoint(dentry)) {
211 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
212 dput(dentry);
213 goto out_nfserr;
214 }
215 }
216 }
217 /*
218 * Note: we compose the file handle now, but as the
219 * dentry may be negative, it may need to be updated.
220 */
221 err = fh_compose(resfh, exp, dentry, fhp);
222 if (!err && !dentry->d_inode)
223 err = nfserr_noent;
224 dput(dentry);
225out:
226 exp_put(exp);
227 return err;
228
229out_nfserr:
230 err = nfserrno(err);
231 goto out;
232}
233
234/*
235 * Set various file attributes.
236 * N.B. After this call fhp needs an fh_put
237 */
238int
239nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
240 int check_guard, time_t guardtime)
241{
242 struct dentry *dentry;
243 struct inode *inode;
244 int accmode = MAY_SATTR;
245 int ftype = 0;
246 int imode;
247 int err;
248 int size_change = 0;
249
250 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
251 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
252 if (iap->ia_valid & ATTR_SIZE)
253 ftype = S_IFREG;
254
255 /* Get inode */
256 err = fh_verify(rqstp, fhp, ftype, accmode);
257 if (err || !iap->ia_valid)
258 goto out;
259
260 dentry = fhp->fh_dentry;
261 inode = dentry->d_inode;
262
263 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
264 * which only requires access, and "set-[ac]time-to-X" which
265 * requires ownership.
266 * So if it looks like it might be "set both to the same time which
267 * is close to now", and if inode_change_ok fails, then we
268 * convert to "set to now" instead of "set to explicit time"
269 *
270 * We only call inode_change_ok as the last test as technically
271 * it is not an interface that we should be using. It is only
272 * valid if the filesystem does not define it's own i_op->setattr.
273 */
274#define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
275#define MAX_TOUCH_TIME_ERROR (30*60)
276 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
277 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
278 ) {
279 /* Looks probable. Now just make sure time is in the right ballpark.
280 * Solaris, at least, doesn't seem to care what the time request is.
281 * We require it be within 30 minutes of now.
282 */
283 time_t delta = iap->ia_atime.tv_sec - get_seconds();
284 if (delta<0) delta = -delta;
285 if (delta < MAX_TOUCH_TIME_ERROR &&
286 inode_change_ok(inode, iap) != 0) {
287 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
288 * this will cause notify_change to set these times to "now"
289 */
290 iap->ia_valid &= ~BOTH_TIME_SET;
291 }
292 }
293
294 /* The size case is special. It changes the file as well as the attributes. */
295 if (iap->ia_valid & ATTR_SIZE) {
296 if (iap->ia_size < inode->i_size) {
297 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
298 if (err)
299 goto out;
300 }
301
302 /*
303 * If we are changing the size of the file, then
304 * we need to break all leases.
305 */
306 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
307 if (err == -EWOULDBLOCK)
308 err = -ETIMEDOUT;
309 if (err) /* ENOMEM or EWOULDBLOCK */
310 goto out_nfserr;
311
312 err = get_write_access(inode);
313 if (err)
314 goto out_nfserr;
315
316 size_change = 1;
317 err = locks_verify_truncate(inode, NULL, iap->ia_size);
318 if (err) {
319 put_write_access(inode);
320 goto out_nfserr;
321 }
322 DQUOT_INIT(inode);
323 }
324
325 imode = inode->i_mode;
326 if (iap->ia_valid & ATTR_MODE) {
327 iap->ia_mode &= S_IALLUGO;
328 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
329 }
330
331 /* Revoke setuid/setgid bit on chown/chgrp */
332 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
333 iap->ia_valid |= ATTR_KILL_SUID;
334 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
335 iap->ia_valid |= ATTR_KILL_SGID;
336
337 /* Change the attributes. */
338
339 iap->ia_valid |= ATTR_CTIME;
340
341 err = nfserr_notsync;
342 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
343 fh_lock(fhp);
344 err = notify_change(dentry, iap);
345 err = nfserrno(err);
346 fh_unlock(fhp);
347 }
348 if (size_change)
349 put_write_access(inode);
350 if (!err)
351 if (EX_ISSYNC(fhp->fh_export))
352 write_inode_now(inode, 1);
353out:
354 return err;
355
356out_nfserr:
357 err = nfserrno(err);
358 goto out;
359}
360
361#if defined(CONFIG_NFSD_V4)
362
363static int
364set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
365{
366 int len;
367 size_t buflen;
368 char *buf = NULL;
369 int error = 0;
370 struct inode *inode = dentry->d_inode;
371
372 buflen = posix_acl_xattr_size(pacl->a_count);
373 buf = kmalloc(buflen, GFP_KERNEL);
374 error = -ENOMEM;
375 if (buf == NULL)
376 goto out;
377
378 len = posix_acl_to_xattr(pacl, buf, buflen);
379 if (len < 0) {
380 error = len;
381 goto out;
382 }
383
384 error = -EOPNOTSUPP;
385 if (inode->i_op && inode->i_op->setxattr) {
386 down(&inode->i_sem);
387 security_inode_setxattr(dentry, key, buf, len, 0);
388 error = inode->i_op->setxattr(dentry, key, buf, len, 0);
389 if (!error)
390 security_inode_post_setxattr(dentry, key, buf, len, 0);
391 up(&inode->i_sem);
392 }
393out:
394 kfree(buf);
395 return error;
396}
397
398int
399nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
400 struct nfs4_acl *acl)
401{
402 int error;
403 struct dentry *dentry;
404 struct inode *inode;
405 struct posix_acl *pacl = NULL, *dpacl = NULL;
406 unsigned int flags = 0;
407
408 /* Get inode */
409 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
410 if (error)
411 goto out;
412
413 dentry = fhp->fh_dentry;
414 inode = dentry->d_inode;
415 if (S_ISDIR(inode->i_mode))
416 flags = NFS4_ACL_DIR;
417
418 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
419 if (error == -EINVAL) {
420 error = nfserr_attrnotsupp;
421 goto out;
422 } else if (error < 0)
423 goto out_nfserr;
424
425 if (pacl) {
Christoph Hellwig9a59f452005-06-23 00:10:19 -0700426 error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (error < 0)
428 goto out_nfserr;
429 }
430
431 if (dpacl) {
Christoph Hellwig9a59f452005-06-23 00:10:19 -0700432 error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (error < 0)
434 goto out_nfserr;
435 }
436
437 error = nfs_ok;
438
439out:
440 posix_acl_release(pacl);
441 posix_acl_release(dpacl);
442 return (error);
443out_nfserr:
444 error = nfserrno(error);
445 goto out;
446}
447
448static struct posix_acl *
449_get_posix_acl(struct dentry *dentry, char *key)
450{
451 struct inode *inode = dentry->d_inode;
452 char *buf = NULL;
453 int buflen, error = 0;
454 struct posix_acl *pacl = NULL;
455
456 error = -EOPNOTSUPP;
457 if (inode->i_op == NULL)
458 goto out_err;
459 if (inode->i_op->getxattr == NULL)
460 goto out_err;
461
462 error = security_inode_getxattr(dentry, key);
463 if (error)
464 goto out_err;
465
466 buflen = inode->i_op->getxattr(dentry, key, NULL, 0);
467 if (buflen <= 0) {
468 error = buflen < 0 ? buflen : -ENODATA;
469 goto out_err;
470 }
471
472 buf = kmalloc(buflen, GFP_KERNEL);
473 if (buf == NULL) {
474 error = -ENOMEM;
475 goto out_err;
476 }
477
478 error = inode->i_op->getxattr(dentry, key, buf, buflen);
479 if (error < 0)
480 goto out_err;
481
482 pacl = posix_acl_from_xattr(buf, buflen);
483 out:
484 kfree(buf);
485 return pacl;
486 out_err:
487 pacl = ERR_PTR(error);
488 goto out;
489}
490
491int
492nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
493{
494 struct inode *inode = dentry->d_inode;
495 int error = 0;
496 struct posix_acl *pacl = NULL, *dpacl = NULL;
497 unsigned int flags = 0;
498
Christoph Hellwig9a59f452005-06-23 00:10:19 -0700499 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
501 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
502 if (IS_ERR(pacl)) {
503 error = PTR_ERR(pacl);
504 pacl = NULL;
505 goto out;
506 }
507
508 if (S_ISDIR(inode->i_mode)) {
Christoph Hellwig9a59f452005-06-23 00:10:19 -0700509 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
511 dpacl = NULL;
512 else if (IS_ERR(dpacl)) {
513 error = PTR_ERR(dpacl);
514 dpacl = NULL;
515 goto out;
516 }
517 flags = NFS4_ACL_DIR;
518 }
519
520 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
521 if (IS_ERR(*acl)) {
522 error = PTR_ERR(*acl);
523 *acl = NULL;
524 }
525 out:
526 posix_acl_release(pacl);
527 posix_acl_release(dpacl);
528 return error;
529}
530
531#endif /* defined(CONFIG_NFS_V4) */
532
533#ifdef CONFIG_NFSD_V3
534/*
535 * Check server access rights to a file system object
536 */
537struct accessmap {
538 u32 access;
539 int how;
540};
541static struct accessmap nfs3_regaccess[] = {
542 { NFS3_ACCESS_READ, MAY_READ },
543 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
544 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
545 { NFS3_ACCESS_EXTEND, MAY_WRITE },
546
547 { 0, 0 }
548};
549
550static struct accessmap nfs3_diraccess[] = {
551 { NFS3_ACCESS_READ, MAY_READ },
552 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
553 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
554 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
555 { NFS3_ACCESS_DELETE, MAY_REMOVE },
556
557 { 0, 0 }
558};
559
560static struct accessmap nfs3_anyaccess[] = {
561 /* Some clients - Solaris 2.6 at least, make an access call
562 * to the server to check for access for things like /dev/null
563 * (which really, the server doesn't care about). So
564 * We provide simple access checking for them, looking
565 * mainly at mode bits, and we make sure to ignore read-only
566 * filesystem checks
567 */
568 { NFS3_ACCESS_READ, MAY_READ },
569 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
570 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
571 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
572
573 { 0, 0 }
574};
575
576int
577nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
578{
579 struct accessmap *map;
580 struct svc_export *export;
581 struct dentry *dentry;
582 u32 query, result = 0, sresult = 0;
583 unsigned int error;
584
585 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
586 if (error)
587 goto out;
588
589 export = fhp->fh_export;
590 dentry = fhp->fh_dentry;
591
592 if (S_ISREG(dentry->d_inode->i_mode))
593 map = nfs3_regaccess;
594 else if (S_ISDIR(dentry->d_inode->i_mode))
595 map = nfs3_diraccess;
596 else
597 map = nfs3_anyaccess;
598
599
600 query = *access;
601 for (; map->access; map++) {
602 if (map->access & query) {
603 unsigned int err2;
604
605 sresult |= map->access;
606
607 err2 = nfsd_permission(export, dentry, map->how);
608 switch (err2) {
609 case nfs_ok:
610 result |= map->access;
611 break;
612
613 /* the following error codes just mean the access was not allowed,
614 * rather than an error occurred */
615 case nfserr_rofs:
616 case nfserr_acces:
617 case nfserr_perm:
618 /* simply don't "or" in the access bit. */
619 break;
620 default:
621 error = err2;
622 goto out;
623 }
624 }
625 }
626 *access = result;
627 if (supported)
628 *supported = sresult;
629
630 out:
631 return error;
632}
633#endif /* CONFIG_NFSD_V3 */
634
635
636
637/*
638 * Open an existing file or directory.
639 * The access argument indicates the type of open (read/write/lock)
640 * N.B. After this call fhp needs an fh_put
641 */
642int
643nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
644 int access, struct file **filp)
645{
646 struct dentry *dentry;
647 struct inode *inode;
648 int flags = O_RDONLY|O_LARGEFILE, err;
649
650 /*
651 * If we get here, then the client has already done an "open",
652 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
653 * in case a chmod has now revoked permission.
654 */
655 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
656 if (err)
657 goto out;
658
659 dentry = fhp->fh_dentry;
660 inode = dentry->d_inode;
661
662 /* Disallow write access to files with the append-only bit set
663 * or any access when mandatory locking enabled
664 */
665 err = nfserr_perm;
666 if (IS_APPEND(inode) && (access & MAY_WRITE))
667 goto out;
668 if (IS_ISMNDLK(inode))
669 goto out;
670
671 if (!inode->i_fop)
672 goto out;
673
674 /*
675 * Check to see if there are any leases on this file.
676 * This may block while leases are broken.
677 */
678 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
679 if (err == -EWOULDBLOCK)
680 err = -ETIMEDOUT;
681 if (err) /* NOMEM or WOULDBLOCK */
682 goto out_nfserr;
683
684 if (access & MAY_WRITE) {
685 flags = O_WRONLY|O_LARGEFILE;
686
687 DQUOT_INIT(inode);
688 }
689 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
690 if (IS_ERR(*filp))
691 err = PTR_ERR(*filp);
692out_nfserr:
693 if (err)
694 err = nfserrno(err);
695out:
696 return err;
697}
698
699/*
700 * Close a file.
701 */
702void
703nfsd_close(struct file *filp)
704{
705 fput(filp);
706}
707
708/*
709 * Sync a file
710 * As this calls fsync (not fdatasync) there is no need for a write_inode
711 * after it.
712 */
713static inline void nfsd_dosync(struct file *filp, struct dentry *dp,
714 struct file_operations *fop)
715{
716 struct inode *inode = dp->d_inode;
717 int (*fsync) (struct file *, struct dentry *, int);
718
719 filemap_fdatawrite(inode->i_mapping);
720 if (fop && (fsync = fop->fsync))
721 fsync(filp, dp, 0);
722 filemap_fdatawait(inode->i_mapping);
723}
724
725
726static void
727nfsd_sync(struct file *filp)
728{
729 struct inode *inode = filp->f_dentry->d_inode;
730 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
731 down(&inode->i_sem);
732 nfsd_dosync(filp, filp->f_dentry, filp->f_op);
733 up(&inode->i_sem);
734}
735
NeilBrowna6ccbbb2005-07-07 17:59:11 -0700736void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737nfsd_sync_dir(struct dentry *dp)
738{
739 nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
740}
741
742/*
743 * Obtain the readahead parameters for the file
744 * specified by (dev, ino).
745 */
746static DEFINE_SPINLOCK(ra_lock);
747
748static inline struct raparms *
749nfsd_get_raparms(dev_t dev, ino_t ino)
750{
751 struct raparms *ra, **rap, **frap = NULL;
752 int depth = 0;
753
754 spin_lock(&ra_lock);
755 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
756 if (ra->p_ino == ino && ra->p_dev == dev)
757 goto found;
758 depth++;
759 if (ra->p_count == 0)
760 frap = rap;
761 }
762 depth = nfsdstats.ra_size*11/10;
763 if (!frap) {
764 spin_unlock(&ra_lock);
765 return NULL;
766 }
767 rap = frap;
768 ra = *frap;
769 ra->p_dev = dev;
770 ra->p_ino = ino;
771 ra->p_set = 0;
772found:
773 if (rap != &raparm_cache) {
774 *rap = ra->p_next;
775 ra->p_next = raparm_cache;
776 raparm_cache = ra;
777 }
778 ra->p_count++;
779 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
780 spin_unlock(&ra_lock);
781 return ra;
782}
783
784/*
785 * Grab and keep cached pages assosiated with a file in the svc_rqst
786 * so that they can be passed to the netowork sendmsg/sendpage routines
787 * directrly. They will be released after the sending has completed.
788 */
789static int
790nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
791{
792 unsigned long count = desc->count;
793 struct svc_rqst *rqstp = desc->arg.data;
794
795 if (size > count)
796 size = count;
797
798 if (rqstp->rq_res.page_len == 0) {
799 get_page(page);
800 rqstp->rq_respages[rqstp->rq_resused++] = page;
801 rqstp->rq_res.page_base = offset;
802 rqstp->rq_res.page_len = size;
803 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
804 get_page(page);
805 rqstp->rq_respages[rqstp->rq_resused++] = page;
806 rqstp->rq_res.page_len += size;
807 } else {
808 rqstp->rq_res.page_len += size;
809 }
810
811 desc->count = count - size;
812 desc->written += size;
813 return size;
814}
815
816static inline int
817nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
818 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
819{
820 struct inode *inode;
821 struct raparms *ra;
822 mm_segment_t oldfs;
823 int err;
824
825 err = nfserr_perm;
826 inode = file->f_dentry->d_inode;
827#ifdef MSNFS
828 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
829 (!lock_may_read(inode, offset, *count)))
830 goto out;
831#endif
832
833 /* Get readahead parameters */
834 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
835
836 if (ra && ra->p_set)
837 file->f_ra = ra->p_ra;
838
839 if (file->f_op->sendfile) {
840 svc_pushback_unused_pages(rqstp);
841 err = file->f_op->sendfile(file, &offset, *count,
842 nfsd_read_actor, rqstp);
843 } else {
844 oldfs = get_fs();
845 set_fs(KERNEL_DS);
846 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
847 set_fs(oldfs);
848 }
849
850 /* Write back readahead params */
851 if (ra) {
852 spin_lock(&ra_lock);
853 ra->p_ra = file->f_ra;
854 ra->p_set = 1;
855 ra->p_count--;
856 spin_unlock(&ra_lock);
857 }
858
859 if (err >= 0) {
860 nfsdstats.io_read += err;
861 *count = err;
862 err = 0;
Robert Love0eeca282005-07-12 17:06:03 -0400863 fsnotify_access(file->f_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 } else
865 err = nfserrno(err);
866out:
867 return err;
868}
869
870static inline int
871nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
872 loff_t offset, struct kvec *vec, int vlen,
873 unsigned long cnt, int *stablep)
874{
875 struct svc_export *exp;
876 struct dentry *dentry;
877 struct inode *inode;
878 mm_segment_t oldfs;
879 int err = 0;
880 int stable = *stablep;
881
882 err = nfserr_perm;
883
884#ifdef MSNFS
885 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
886 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
887 goto out;
888#endif
889
890 dentry = file->f_dentry;
891 inode = dentry->d_inode;
892 exp = fhp->fh_export;
893
894 /*
895 * Request sync writes if
896 * - the sync export option has been set, or
897 * - the client requested O_SYNC behavior (NFSv3 feature).
898 * - The file system doesn't support fsync().
899 * When gathered writes have been configured for this volume,
900 * flushing the data to disk is handled separately below.
901 */
902
903 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
904 stable = 2;
905 *stablep = 2; /* FILE_SYNC */
906 }
907
908 if (!EX_ISSYNC(exp))
909 stable = 0;
910 if (stable && !EX_WGATHER(exp))
911 file->f_flags |= O_SYNC;
912
913 /* Write the data. */
914 oldfs = get_fs(); set_fs(KERNEL_DS);
915 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
916 set_fs(oldfs);
917 if (err >= 0) {
918 nfsdstats.io_write += cnt;
Robert Love0eeca282005-07-12 17:06:03 -0400919 fsnotify_modify(file->f_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 }
921
922 /* clear setuid/setgid flag after write */
923 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) {
924 struct iattr ia;
925 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
926
927 down(&inode->i_sem);
928 notify_change(dentry, &ia);
929 up(&inode->i_sem);
930 }
931
932 if (err >= 0 && stable) {
933 static ino_t last_ino;
934 static dev_t last_dev;
935
936 /*
937 * Gathered writes: If another process is currently
938 * writing to the file, there's a high chance
939 * this is another nfsd (triggered by a bulk write
940 * from a client's biod). Rather than syncing the
941 * file with each write request, we sleep for 10 msec.
942 *
943 * I don't know if this roughly approximates
944 * C. Juszak's idea of gathered writes, but it's a
945 * nice and simple solution (IMHO), and it seems to
946 * work:-)
947 */
948 if (EX_WGATHER(exp)) {
949 if (atomic_read(&inode->i_writecount) > 1
950 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
951 dprintk("nfsd: write defer %d\n", current->pid);
952 msleep(10);
953 dprintk("nfsd: write resume %d\n", current->pid);
954 }
955
956 if (inode->i_state & I_DIRTY) {
957 dprintk("nfsd: write sync %d\n", current->pid);
958 nfsd_sync(file);
959 }
960#if 0
961 wake_up(&inode->i_wait);
962#endif
963 }
964 last_ino = inode->i_ino;
965 last_dev = inode->i_sb->s_dev;
966 }
967
968 dprintk("nfsd: write complete err=%d\n", err);
969 if (err >= 0)
970 err = 0;
971 else
972 err = nfserrno(err);
973out:
974 return err;
975}
976
977/*
978 * Read data from a file. count must contain the requested read count
979 * on entry. On return, *count contains the number of bytes actually read.
980 * N.B. After this call fhp needs an fh_put
981 */
982int
983nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
984 loff_t offset, struct kvec *vec, int vlen,
985 unsigned long *count)
986{
987 int err;
988
989 if (file) {
990 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
991 MAY_READ|MAY_OWNER_OVERRIDE);
992 if (err)
993 goto out;
994 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
995 } else {
996 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
997 if (err)
998 goto out;
999 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1000 nfsd_close(file);
1001 }
1002out:
1003 return err;
1004}
1005
1006/*
1007 * Write data to a file.
1008 * The stable flag requests synchronous writes.
1009 * N.B. After this call fhp needs an fh_put
1010 */
1011int
1012nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1013 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1014 int *stablep)
1015{
1016 int err = 0;
1017
1018 if (file) {
1019 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1020 MAY_WRITE|MAY_OWNER_OVERRIDE);
1021 if (err)
1022 goto out;
1023 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1024 stablep);
1025 } else {
1026 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1027 if (err)
1028 goto out;
1029
1030 if (cnt)
1031 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1032 cnt, stablep);
1033 nfsd_close(file);
1034 }
1035out:
1036 return err;
1037}
1038
1039#ifdef CONFIG_NFSD_V3
1040/*
1041 * Commit all pending writes to stable storage.
1042 * Strictly speaking, we could sync just the indicated file region here,
1043 * but there's currently no way we can ask the VFS to do so.
1044 *
1045 * Unfortunately we cannot lock the file to make sure we return full WCC
1046 * data to the client, as locking happens lower down in the filesystem.
1047 */
1048int
1049nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1050 loff_t offset, unsigned long count)
1051{
1052 struct file *file;
1053 int err;
1054
1055 if ((u64)count > ~(u64)offset)
1056 return nfserr_inval;
1057
1058 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1059 return err;
1060 if (EX_ISSYNC(fhp->fh_export)) {
1061 if (file->f_op && file->f_op->fsync) {
1062 nfsd_sync(file);
1063 } else {
1064 err = nfserr_notsupp;
1065 }
1066 }
1067
1068 nfsd_close(file);
1069 return err;
1070}
1071#endif /* CONFIG_NFSD_V3 */
1072
1073/*
1074 * Create a file (regular, directory, device, fifo); UNIX sockets
1075 * not yet implemented.
1076 * If the response fh has been verified, the parent directory should
1077 * already be locked. Note that the parent directory is left locked.
1078 *
1079 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1080 */
1081int
1082nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1083 char *fname, int flen, struct iattr *iap,
1084 int type, dev_t rdev, struct svc_fh *resfhp)
1085{
1086 struct dentry *dentry, *dchild = NULL;
1087 struct inode *dirp;
1088 int err;
1089
1090 err = nfserr_perm;
1091 if (!flen)
1092 goto out;
1093 err = nfserr_exist;
1094 if (isdotent(fname, flen))
1095 goto out;
1096
1097 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1098 if (err)
1099 goto out;
1100
1101 dentry = fhp->fh_dentry;
1102 dirp = dentry->d_inode;
1103
1104 err = nfserr_notdir;
1105 if(!dirp->i_op || !dirp->i_op->lookup)
1106 goto out;
1107 /*
1108 * Check whether the response file handle has been verified yet.
1109 * If it has, the parent directory should already be locked.
1110 */
1111 if (!resfhp->fh_dentry) {
1112 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1113 fh_lock(fhp);
1114 dchild = lookup_one_len(fname, dentry, flen);
1115 err = PTR_ERR(dchild);
1116 if (IS_ERR(dchild))
1117 goto out_nfserr;
1118 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1119 if (err)
1120 goto out;
1121 } else {
1122 /* called from nfsd_proc_create */
1123 dchild = dget(resfhp->fh_dentry);
1124 if (!fhp->fh_locked) {
1125 /* not actually possible */
1126 printk(KERN_ERR
1127 "nfsd_create: parent %s/%s not locked!\n",
1128 dentry->d_parent->d_name.name,
1129 dentry->d_name.name);
1130 err = -EIO;
1131 goto out;
1132 }
1133 }
1134 /*
1135 * Make sure the child dentry is still negative ...
1136 */
1137 err = nfserr_exist;
1138 if (dchild->d_inode) {
1139 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1140 dentry->d_name.name, dchild->d_name.name);
1141 goto out;
1142 }
1143
1144 if (!(iap->ia_valid & ATTR_MODE))
1145 iap->ia_mode = 0;
1146 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1147
1148 /*
1149 * Get the dir op function pointer.
1150 */
1151 err = nfserr_perm;
1152 switch (type) {
1153 case S_IFREG:
1154 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1155 break;
1156 case S_IFDIR:
1157 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1158 break;
1159 case S_IFCHR:
1160 case S_IFBLK:
1161 case S_IFIFO:
1162 case S_IFSOCK:
1163 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1164 break;
1165 default:
1166 printk("nfsd: bad file type %o in nfsd_create\n", type);
1167 err = -EINVAL;
1168 }
1169 if (err < 0)
1170 goto out_nfserr;
1171
1172 if (EX_ISSYNC(fhp->fh_export)) {
1173 nfsd_sync_dir(dentry);
1174 write_inode_now(dchild->d_inode, 1);
1175 }
1176
1177
1178 /* Set file attributes. Mode has already been set and
1179 * setting uid/gid works only for root. Irix appears to
1180 * send along the gid when it tries to implement setgid
1181 * directories via NFS.
1182 */
1183 err = 0;
1184 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
1185 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1186 /*
1187 * Update the file handle to get the new inode info.
1188 */
1189 if (!err)
1190 err = fh_update(resfhp);
1191out:
1192 if (dchild && !IS_ERR(dchild))
1193 dput(dchild);
1194 return err;
1195
1196out_nfserr:
1197 err = nfserrno(err);
1198 goto out;
1199}
1200
1201#ifdef CONFIG_NFSD_V3
1202/*
1203 * NFSv3 version of nfsd_create
1204 */
1205int
1206nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1207 char *fname, int flen, struct iattr *iap,
1208 struct svc_fh *resfhp, int createmode, u32 *verifier,
1209 int *truncp)
1210{
1211 struct dentry *dentry, *dchild = NULL;
1212 struct inode *dirp;
1213 int err;
1214 __u32 v_mtime=0, v_atime=0;
1215 int v_mode=0;
1216
1217 err = nfserr_perm;
1218 if (!flen)
1219 goto out;
1220 err = nfserr_exist;
1221 if (isdotent(fname, flen))
1222 goto out;
1223 if (!(iap->ia_valid & ATTR_MODE))
1224 iap->ia_mode = 0;
1225 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1226 if (err)
1227 goto out;
1228
1229 dentry = fhp->fh_dentry;
1230 dirp = dentry->d_inode;
1231
1232 /* Get all the sanity checks out of the way before
1233 * we lock the parent. */
1234 err = nfserr_notdir;
1235 if(!dirp->i_op || !dirp->i_op->lookup)
1236 goto out;
1237 fh_lock(fhp);
1238
1239 /*
1240 * Compose the response file handle.
1241 */
1242 dchild = lookup_one_len(fname, dentry, flen);
1243 err = PTR_ERR(dchild);
1244 if (IS_ERR(dchild))
1245 goto out_nfserr;
1246
1247 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1248 if (err)
1249 goto out;
1250
1251 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1252 /* while the verifier would fit in mtime+atime,
1253 * solaris7 gets confused (bugid 4218508) if these have
1254 * the high bit set, so we use the mode as well
1255 */
1256 v_mtime = verifier[0]&0x7fffffff;
1257 v_atime = verifier[1]&0x7fffffff;
1258 v_mode = S_IFREG
1259 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1260 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1261 ;
1262 }
1263
1264 if (dchild->d_inode) {
1265 err = 0;
1266
1267 switch (createmode) {
1268 case NFS3_CREATE_UNCHECKED:
1269 if (! S_ISREG(dchild->d_inode->i_mode))
1270 err = nfserr_exist;
1271 else if (truncp) {
1272 /* in nfsv4, we need to treat this case a little
1273 * differently. we don't want to truncate the
1274 * file now; this would be wrong if the OPEN
1275 * fails for some other reason. furthermore,
1276 * if the size is nonzero, we should ignore it
1277 * according to spec!
1278 */
1279 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1280 }
1281 else {
1282 iap->ia_valid &= ATTR_SIZE;
1283 goto set_attr;
1284 }
1285 break;
1286 case NFS3_CREATE_EXCLUSIVE:
1287 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1288 && dchild->d_inode->i_atime.tv_sec == v_atime
1289 && dchild->d_inode->i_mode == v_mode
1290 && dchild->d_inode->i_size == 0 )
1291 break;
1292 /* fallthru */
1293 case NFS3_CREATE_GUARDED:
1294 err = nfserr_exist;
1295 }
1296 goto out;
1297 }
1298
1299 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1300 if (err < 0)
1301 goto out_nfserr;
1302
1303 if (EX_ISSYNC(fhp->fh_export)) {
1304 nfsd_sync_dir(dentry);
1305 /* setattr will sync the child (or not) */
1306 }
1307
1308 /*
1309 * Update the filehandle to get the new inode info.
1310 */
1311 err = fh_update(resfhp);
1312 if (err)
1313 goto out;
1314
1315 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1316 /* Cram the verifier into atime/mtime/mode */
1317 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1318 | ATTR_MTIME_SET|ATTR_ATIME_SET
1319 | ATTR_MODE;
1320 /* XXX someone who knows this better please fix it for nsec */
1321 iap->ia_mtime.tv_sec = v_mtime;
1322 iap->ia_atime.tv_sec = v_atime;
1323 iap->ia_mtime.tv_nsec = 0;
1324 iap->ia_atime.tv_nsec = 0;
1325 iap->ia_mode = v_mode;
1326 }
1327
1328 /* Set file attributes.
1329 * Mode has already been set but we might need to reset it
1330 * for CREATE_EXCLUSIVE
1331 * Irix appears to send along the gid when it tries to
1332 * implement setgid directories via NFS. Clear out all that cruft.
1333 */
1334 set_attr:
1335 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1336 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1337
1338 out:
1339 fh_unlock(fhp);
1340 if (dchild && !IS_ERR(dchild))
1341 dput(dchild);
1342 return err;
1343
1344 out_nfserr:
1345 err = nfserrno(err);
1346 goto out;
1347}
1348#endif /* CONFIG_NFSD_V3 */
1349
1350/*
1351 * Read a symlink. On entry, *lenp must contain the maximum path length that
1352 * fits into the buffer. On return, it contains the true length.
1353 * N.B. After this call fhp needs an fh_put
1354 */
1355int
1356nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1357{
1358 struct dentry *dentry;
1359 struct inode *inode;
1360 mm_segment_t oldfs;
1361 int err;
1362
1363 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1364 if (err)
1365 goto out;
1366
1367 dentry = fhp->fh_dentry;
1368 inode = dentry->d_inode;
1369
1370 err = nfserr_inval;
1371 if (!inode->i_op || !inode->i_op->readlink)
1372 goto out;
1373
1374 touch_atime(fhp->fh_export->ex_mnt, dentry);
1375 /* N.B. Why does this call need a get_fs()??
1376 * Remove the set_fs and watch the fireworks:-) --okir
1377 */
1378
1379 oldfs = get_fs(); set_fs(KERNEL_DS);
1380 err = inode->i_op->readlink(dentry, buf, *lenp);
1381 set_fs(oldfs);
1382
1383 if (err < 0)
1384 goto out_nfserr;
1385 *lenp = err;
1386 err = 0;
1387out:
1388 return err;
1389
1390out_nfserr:
1391 err = nfserrno(err);
1392 goto out;
1393}
1394
1395/*
1396 * Create a symlink and look up its inode
1397 * N.B. After this call _both_ fhp and resfhp need an fh_put
1398 */
1399int
1400nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1401 char *fname, int flen,
1402 char *path, int plen,
1403 struct svc_fh *resfhp,
1404 struct iattr *iap)
1405{
1406 struct dentry *dentry, *dnew;
1407 int err, cerr;
1408 umode_t mode;
1409
1410 err = nfserr_noent;
1411 if (!flen || !plen)
1412 goto out;
1413 err = nfserr_exist;
1414 if (isdotent(fname, flen))
1415 goto out;
1416
1417 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1418 if (err)
1419 goto out;
1420 fh_lock(fhp);
1421 dentry = fhp->fh_dentry;
1422 dnew = lookup_one_len(fname, dentry, flen);
1423 err = PTR_ERR(dnew);
1424 if (IS_ERR(dnew))
1425 goto out_nfserr;
1426
1427 mode = S_IALLUGO;
1428 /* Only the MODE ATTRibute is even vaguely meaningful */
1429 if (iap && (iap->ia_valid & ATTR_MODE))
1430 mode = iap->ia_mode & S_IALLUGO;
1431
1432 if (unlikely(path[plen] != 0)) {
1433 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1434 if (path_alloced == NULL)
1435 err = -ENOMEM;
1436 else {
1437 strncpy(path_alloced, path, plen);
1438 path_alloced[plen] = 0;
1439 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1440 kfree(path_alloced);
1441 }
1442 } else
1443 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1444
1445 if (!err) {
1446 if (EX_ISSYNC(fhp->fh_export))
1447 nfsd_sync_dir(dentry);
1448 } else
1449 err = nfserrno(err);
1450 fh_unlock(fhp);
1451
1452 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1453 dput(dnew);
1454 if (err==0) err = cerr;
1455out:
1456 return err;
1457
1458out_nfserr:
1459 err = nfserrno(err);
1460 goto out;
1461}
1462
1463/*
1464 * Create a hardlink
1465 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1466 */
1467int
1468nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1469 char *name, int len, struct svc_fh *tfhp)
1470{
1471 struct dentry *ddir, *dnew, *dold;
1472 struct inode *dirp, *dest;
1473 int err;
1474
1475 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1476 if (err)
1477 goto out;
1478 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1479 if (err)
1480 goto out;
1481
1482 err = nfserr_perm;
1483 if (!len)
1484 goto out;
1485 err = nfserr_exist;
1486 if (isdotent(name, len))
1487 goto out;
1488
1489 fh_lock(ffhp);
1490 ddir = ffhp->fh_dentry;
1491 dirp = ddir->d_inode;
1492
1493 dnew = lookup_one_len(name, ddir, len);
1494 err = PTR_ERR(dnew);
1495 if (IS_ERR(dnew))
1496 goto out_nfserr;
1497
1498 dold = tfhp->fh_dentry;
1499 dest = dold->d_inode;
1500
1501 err = vfs_link(dold, dirp, dnew);
1502 if (!err) {
1503 if (EX_ISSYNC(ffhp->fh_export)) {
1504 nfsd_sync_dir(ddir);
1505 write_inode_now(dest, 1);
1506 }
1507 } else {
1508 if (err == -EXDEV && rqstp->rq_vers == 2)
1509 err = nfserr_acces;
1510 else
1511 err = nfserrno(err);
1512 }
1513
1514 fh_unlock(ffhp);
1515 dput(dnew);
1516out:
1517 return err;
1518
1519out_nfserr:
1520 err = nfserrno(err);
1521 goto out;
1522}
1523
1524/*
1525 * Rename a file
1526 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1527 */
1528int
1529nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1530 struct svc_fh *tfhp, char *tname, int tlen)
1531{
1532 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1533 struct inode *fdir, *tdir;
1534 int err;
1535
1536 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1537 if (err)
1538 goto out;
1539 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1540 if (err)
1541 goto out;
1542
1543 fdentry = ffhp->fh_dentry;
1544 fdir = fdentry->d_inode;
1545
1546 tdentry = tfhp->fh_dentry;
1547 tdir = tdentry->d_inode;
1548
1549 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1550 if (fdir->i_sb != tdir->i_sb)
1551 goto out;
1552
1553 err = nfserr_perm;
1554 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1555 goto out;
1556
1557 /* cannot use fh_lock as we need deadlock protective ordering
1558 * so do it by hand */
1559 trap = lock_rename(tdentry, fdentry);
1560 ffhp->fh_locked = tfhp->fh_locked = 1;
1561 fill_pre_wcc(ffhp);
1562 fill_pre_wcc(tfhp);
1563
1564 odentry = lookup_one_len(fname, fdentry, flen);
1565 err = PTR_ERR(odentry);
1566 if (IS_ERR(odentry))
1567 goto out_nfserr;
1568
1569 err = -ENOENT;
1570 if (!odentry->d_inode)
1571 goto out_dput_old;
1572 err = -EINVAL;
1573 if (odentry == trap)
1574 goto out_dput_old;
1575
1576 ndentry = lookup_one_len(tname, tdentry, tlen);
1577 err = PTR_ERR(ndentry);
1578 if (IS_ERR(ndentry))
1579 goto out_dput_old;
1580 err = -ENOTEMPTY;
1581 if (ndentry == trap)
1582 goto out_dput_new;
1583
1584#ifdef MSNFS
1585 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1586 ((atomic_read(&odentry->d_count) > 1)
1587 || (atomic_read(&ndentry->d_count) > 1))) {
1588 err = nfserr_perm;
1589 } else
1590#endif
1591 err = vfs_rename(fdir, odentry, tdir, ndentry);
1592 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1593 nfsd_sync_dir(tdentry);
1594 nfsd_sync_dir(fdentry);
1595 }
1596
1597 out_dput_new:
1598 dput(ndentry);
1599 out_dput_old:
1600 dput(odentry);
1601 out_nfserr:
1602 if (err)
1603 err = nfserrno(err);
1604
1605 /* we cannot reply on fh_unlock on the two filehandles,
1606 * as that would do the wrong thing if the two directories
1607 * were the same, so again we do it by hand
1608 */
1609 fill_post_wcc(ffhp);
1610 fill_post_wcc(tfhp);
1611 unlock_rename(tdentry, fdentry);
1612 ffhp->fh_locked = tfhp->fh_locked = 0;
1613
1614out:
1615 return err;
1616}
1617
1618/*
1619 * Unlink a file or directory
1620 * N.B. After this call fhp needs an fh_put
1621 */
1622int
1623nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1624 char *fname, int flen)
1625{
1626 struct dentry *dentry, *rdentry;
1627 struct inode *dirp;
1628 int err;
1629
1630 err = nfserr_acces;
1631 if (!flen || isdotent(fname, flen))
1632 goto out;
1633 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1634 if (err)
1635 goto out;
1636
1637 fh_lock(fhp);
1638 dentry = fhp->fh_dentry;
1639 dirp = dentry->d_inode;
1640
1641 rdentry = lookup_one_len(fname, dentry, flen);
1642 err = PTR_ERR(rdentry);
1643 if (IS_ERR(rdentry))
1644 goto out_nfserr;
1645
1646 if (!rdentry->d_inode) {
1647 dput(rdentry);
1648 err = nfserr_noent;
1649 goto out;
1650 }
1651
1652 if (!type)
1653 type = rdentry->d_inode->i_mode & S_IFMT;
1654
1655 if (type != S_IFDIR) { /* It's UNLINK */
1656#ifdef MSNFS
1657 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1658 (atomic_read(&rdentry->d_count) > 1)) {
1659 err = nfserr_perm;
1660 } else
1661#endif
1662 err = vfs_unlink(dirp, rdentry);
1663 } else { /* It's RMDIR */
1664 err = vfs_rmdir(dirp, rdentry);
1665 }
1666
1667 dput(rdentry);
1668
1669 if (err)
1670 goto out_nfserr;
1671 if (EX_ISSYNC(fhp->fh_export))
1672 nfsd_sync_dir(dentry);
1673
1674out:
1675 return err;
1676
1677out_nfserr:
1678 err = nfserrno(err);
1679 goto out;
1680}
1681
1682/*
1683 * Read entries from a directory.
1684 * The NFSv3/4 verifier we ignore for now.
1685 */
1686int
1687nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1688 struct readdir_cd *cdp, encode_dent_fn func)
1689{
1690 int err;
1691 struct file *file;
1692 loff_t offset = *offsetp;
1693
1694 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1695 if (err)
1696 goto out;
1697
1698 offset = vfs_llseek(file, offset, 0);
1699 if (offset < 0) {
1700 err = nfserrno((int)offset);
1701 goto out_close;
1702 }
1703
1704 /*
1705 * Read the directory entries. This silly loop is necessary because
1706 * readdir() is not guaranteed to fill up the entire buffer, but
1707 * may choose to do less.
1708 */
1709
1710 do {
1711 cdp->err = nfserr_eof; /* will be cleared on successful read */
1712 err = vfs_readdir(file, (filldir_t) func, cdp);
1713 } while (err >=0 && cdp->err == nfs_ok);
1714 if (err)
1715 err = nfserrno(err);
1716 else
1717 err = cdp->err;
1718 *offsetp = vfs_llseek(file, 0, 1);
1719
1720 if (err == nfserr_eof || err == nfserr_toosmall)
1721 err = nfs_ok; /* can still be found in ->err */
1722out_close:
1723 nfsd_close(file);
1724out:
1725 return err;
1726}
1727
1728/*
1729 * Get file system stats
1730 * N.B. After this call fhp needs an fh_put
1731 */
1732int
1733nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1734{
1735 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1736 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1737 err = nfserr_io;
1738 return err;
1739}
1740
1741/*
1742 * Check for a user's access permissions to this inode.
1743 */
1744int
1745nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1746{
1747 struct inode *inode = dentry->d_inode;
1748 int err;
1749
1750 if (acc == MAY_NOP)
1751 return 0;
1752#if 0
1753 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1754 acc,
1755 (acc & MAY_READ)? " read" : "",
1756 (acc & MAY_WRITE)? " write" : "",
1757 (acc & MAY_EXEC)? " exec" : "",
1758 (acc & MAY_SATTR)? " sattr" : "",
1759 (acc & MAY_TRUNC)? " trunc" : "",
1760 (acc & MAY_LOCK)? " lock" : "",
1761 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1762 inode->i_mode,
1763 IS_IMMUTABLE(inode)? " immut" : "",
1764 IS_APPEND(inode)? " append" : "",
1765 IS_RDONLY(inode)? " ro" : "");
1766 dprintk(" owner %d/%d user %d/%d\n",
1767 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1768#endif
1769
1770 /* Normally we reject any write/sattr etc access on a read-only file
1771 * system. But if it is IRIX doing check on write-access for a
1772 * device special file, we ignore rofs.
1773 */
1774 if (!(acc & MAY_LOCAL_ACCESS))
1775 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1776 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1777 return nfserr_rofs;
1778 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1779 return nfserr_perm;
1780 }
1781 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1782 return nfserr_perm;
1783
1784 if (acc & MAY_LOCK) {
1785 /* If we cannot rely on authentication in NLM requests,
1786 * just allow locks, otherwise require read permission, or
1787 * ownership
1788 */
1789 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1790 return 0;
1791 else
1792 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1793 }
1794 /*
1795 * The file owner always gets access permission for accesses that
1796 * would normally be checked at open time. This is to make
1797 * file access work even when the client has done a fchmod(fd, 0).
1798 *
1799 * However, `cp foo bar' should fail nevertheless when bar is
1800 * readonly. A sensible way to do this might be to reject all
1801 * attempts to truncate a read-only file, because a creat() call
1802 * always implies file truncation.
1803 * ... but this isn't really fair. A process may reasonably call
1804 * ftruncate on an open file descriptor on a file with perm 000.
1805 * We must trust the client to do permission checking - using "ACCESS"
1806 * with NFSv3.
1807 */
1808 if ((acc & MAY_OWNER_OVERRIDE) &&
1809 inode->i_uid == current->fsuid)
1810 return 0;
1811
1812 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1813
1814 /* Allow read access to binaries even when mode 111 */
1815 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1816 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1817 err = permission(inode, MAY_EXEC, NULL);
1818
1819 return err? nfserrno(err) : 0;
1820}
1821
1822void
1823nfsd_racache_shutdown(void)
1824{
1825 if (!raparm_cache)
1826 return;
1827 dprintk("nfsd: freeing readahead buffers.\n");
1828 kfree(raparml);
1829 raparm_cache = raparml = NULL;
1830}
1831/*
1832 * Initialize readahead param cache
1833 */
1834int
1835nfsd_racache_init(int cache_size)
1836{
1837 int i;
1838
1839 if (raparm_cache)
1840 return 0;
1841 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1842
1843 if (raparml != NULL) {
1844 dprintk("nfsd: allocating %d readahead buffers.\n",
1845 cache_size);
1846 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1847 for (i = 0; i < cache_size - 1; i++) {
1848 raparml[i].p_next = raparml + i + 1;
1849 }
1850 raparm_cache = raparml;
1851 } else {
1852 printk(KERN_WARNING
1853 "nfsd: Could not allocate memory read-ahead cache.\n");
1854 return -ENOMEM;
1855 }
1856 nfsdstats.ra_size = cache_size;
1857 return 0;
1858}
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001859
1860#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1861struct posix_acl *
1862nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1863{
1864 struct inode *inode = fhp->fh_dentry->d_inode;
1865 char *name;
1866 void *value = NULL;
1867 ssize_t size;
1868 struct posix_acl *acl;
1869
1870 if (!IS_POSIXACL(inode) || !inode->i_op || !inode->i_op->getxattr)
1871 return ERR_PTR(-EOPNOTSUPP);
1872 switch(type) {
1873 case ACL_TYPE_ACCESS:
Christoph Hellwig334a13e2005-06-28 20:44:58 -07001874 name = POSIX_ACL_XATTR_ACCESS;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001875 break;
1876 case ACL_TYPE_DEFAULT:
Christoph Hellwig334a13e2005-06-28 20:44:58 -07001877 name = POSIX_ACL_XATTR_DEFAULT;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001878 break;
1879 default:
1880 return ERR_PTR(-EOPNOTSUPP);
1881 }
1882
1883 size = inode->i_op->getxattr(fhp->fh_dentry, name, NULL, 0);
1884
1885 if (size < 0) {
1886 acl = ERR_PTR(size);
1887 goto getout;
1888 } else if (size > 0) {
1889 value = kmalloc(size, GFP_KERNEL);
1890 if (!value) {
1891 acl = ERR_PTR(-ENOMEM);
1892 goto getout;
1893 }
1894 size = inode->i_op->getxattr(fhp->fh_dentry, name, value, size);
1895 if (size < 0) {
1896 acl = ERR_PTR(size);
1897 goto getout;
1898 }
1899 }
1900 acl = posix_acl_from_xattr(value, size);
1901
1902getout:
1903 kfree(value);
1904 return acl;
1905}
1906
1907int
1908nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1909{
1910 struct inode *inode = fhp->fh_dentry->d_inode;
1911 char *name;
1912 void *value = NULL;
1913 size_t size;
1914 int error;
1915
1916 if (!IS_POSIXACL(inode) || !inode->i_op ||
1917 !inode->i_op->setxattr || !inode->i_op->removexattr)
1918 return -EOPNOTSUPP;
1919 switch(type) {
1920 case ACL_TYPE_ACCESS:
Christoph Hellwig334a13e2005-06-28 20:44:58 -07001921 name = POSIX_ACL_XATTR_ACCESS;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001922 break;
1923 case ACL_TYPE_DEFAULT:
Christoph Hellwig334a13e2005-06-28 20:44:58 -07001924 name = POSIX_ACL_XATTR_DEFAULT;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001925 break;
1926 default:
1927 return -EOPNOTSUPP;
1928 }
1929
1930 if (acl && acl->a_count) {
Christoph Hellwig334a13e2005-06-28 20:44:58 -07001931 size = posix_acl_xattr_size(acl->a_count);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001932 value = kmalloc(size, GFP_KERNEL);
1933 if (!value)
1934 return -ENOMEM;
1935 size = posix_acl_to_xattr(acl, value, size);
1936 if (size < 0) {
1937 error = size;
1938 goto getout;
1939 }
1940 } else
1941 size = 0;
1942
1943 if (!fhp->fh_locked)
1944 fh_lock(fhp); /* unlocking is done automatically */
1945 if (size)
1946 error = inode->i_op->setxattr(fhp->fh_dentry, name,
1947 value, size, 0);
1948 else {
1949 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1950 error = 0;
1951 else {
1952 error = inode->i_op->removexattr(fhp->fh_dentry, name);
1953 if (error == -ENODATA)
1954 error = 0;
1955 }
1956 }
1957
1958getout:
1959 kfree(value);
1960 return error;
1961}
1962#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */