?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[readdir] constify ->actor
[linux-2.6.git] / fs / nfsd / vfs.c
bloba6bc8a7423dba4c4544bd504bf5b34194a4c8d11
1 /*
2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
5 *
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
11 *
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
14 */
15
16 #include <linux/fs.h>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
31
32 #ifdef CONFIG_NFSD_V3
33 #include "xdr3.h"
34 #endif /* CONFIG_NFSD_V3 */
35
36 #ifdef CONFIG_NFSD_V4
37 #include "acl.h"
38 #include "idmap.h"
39 #endif /* CONFIG_NFSD_V4 */
40
41 #include "nfsd.h"
42 #include "vfs.h"
43
44 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
45
46
47 /*
48 * This is a cache of readahead params that help us choose the proper
49 * readahead strategy. Initially, we set all readahead parameters to 0
50 * and let the VFS handle things.
51 * If you increase the number of cached files very much, you'll need to
52 * add a hash table here.
53 */
54 struct raparms {
55 struct raparms *p_next;
56 unsigned int p_count;
57 ino_t p_ino;
58 dev_t p_dev;
59 int p_set;
60 struct file_ra_state p_ra;
61 unsigned int p_hindex;
62 };
63
64 struct raparm_hbucket {
65 struct raparms *pb_head;
66 spinlock_t pb_lock;
67 } ____cacheline_aligned_in_smp;
68
69 #define RAPARM_HASH_BITS 4
70 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
71 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
72 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
73
74 /*
75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
76 * a mount point.
77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
78 * or nfs_ok having possibly changed *dpp and *expp
79 */
80 int
81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
82 struct svc_export **expp)
83 {
84 struct svc_export *exp = *expp, *exp2 = NULL;
85 struct dentry *dentry = *dpp;
86 struct path path = {.mnt = mntget(exp->ex_path.mnt),
87 .dentry = dget(dentry)};
88 int err = 0;
89
90 err = follow_down(&path);
91 if (err < 0)
92 goto out;
93
94 exp2 = rqst_exp_get_by_name(rqstp, &path);
95 if (IS_ERR(exp2)) {
96 err = PTR_ERR(exp2);
97 /*
98 * We normally allow NFS clients to continue
99 * "underneath" a mountpoint that is not exported.
100 * The exception is V4ROOT, where no traversal is ever
101 * allowed without an explicit export of the new
102 * directory.
103 */
104 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
105 err = 0;
106 path_put(&path);
107 goto out;
108 }
109 if (nfsd_v4client(rqstp) ||
110 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
111 /* successfully crossed mount point */
112 /*
113 * This is subtle: path.dentry is *not* on path.mnt
114 * at this point. The only reason we are safe is that
115 * original mnt is pinned down by exp, so we should
116 * put path *before* putting exp
117 */
118 *dpp = path.dentry;
119 path.dentry = dentry;
120 *expp = exp2;
121 exp2 = exp;
122 }
123 path_put(&path);
124 exp_put(exp2);
125 out:
126 return err;
127 }
128
129 static void follow_to_parent(struct path *path)
130 {
131 struct dentry *dp;
132
133 while (path->dentry == path->mnt->mnt_root && follow_up(path))
134 ;
135 dp = dget_parent(path->dentry);
136 dput(path->dentry);
137 path->dentry = dp;
138 }
139
140 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
141 {
142 struct svc_export *exp2;
143 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
144 .dentry = dget(dparent)};
145
146 follow_to_parent(&path);
147
148 exp2 = rqst_exp_parent(rqstp, &path);
149 if (PTR_ERR(exp2) == -ENOENT) {
150 *dentryp = dget(dparent);
151 } else if (IS_ERR(exp2)) {
152 path_put(&path);
153 return PTR_ERR(exp2);
154 } else {
155 *dentryp = dget(path.dentry);
156 exp_put(*exp);
157 *exp = exp2;
158 }
159 path_put(&path);
160 return 0;
161 }
162
163 /*
164 * For nfsd purposes, we treat V4ROOT exports as though there was an
165 * export at *every* directory.
166 */
167 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
168 {
169 if (d_mountpoint(dentry))
170 return 1;
171 if (nfsd4_is_junction(dentry))
172 return 1;
173 if (!(exp->ex_flags & NFSEXP_V4ROOT))
174 return 0;
175 return dentry->d_inode != NULL;
176 }
177
178 __be32
179 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
180 const char *name, unsigned int len,
181 struct svc_export **exp_ret, struct dentry **dentry_ret)
182 {
183 struct svc_export *exp;
184 struct dentry *dparent;
185 struct dentry *dentry;
186 int host_err;
187
188 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
189
190 dparent = fhp->fh_dentry;
191 exp = fhp->fh_export;
192 exp_get(exp);
193
194 /* Lookup the name, but don't follow links */
195 if (isdotent(name, len)) {
196 if (len==1)
197 dentry = dget(dparent);
198 else if (dparent != exp->ex_path.dentry)
199 dentry = dget_parent(dparent);
200 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
201 dentry = dget(dparent); /* .. == . just like at / */
202 else {
203 /* checking mountpoint crossing is very different when stepping up */
204 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
205 if (host_err)
206 goto out_nfserr;
207 }
208 } else {
209 fh_lock(fhp);
210 dentry = lookup_one_len(name, dparent, len);
211 host_err = PTR_ERR(dentry);
212 if (IS_ERR(dentry))
213 goto out_nfserr;
214 /*
215 * check if we have crossed a mount point ...
216 */
217 if (nfsd_mountpoint(dentry, exp)) {
218 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
219 dput(dentry);
220 goto out_nfserr;
221 }
222 }
223 }
224 *dentry_ret = dentry;
225 *exp_ret = exp;
226 return 0;
227
228 out_nfserr:
229 exp_put(exp);
230 return nfserrno(host_err);
231 }
232
233 /*
234 * Look up one component of a pathname.
235 * N.B. After this call _both_ fhp and resfh need an fh_put
236 *
237 * If the lookup would cross a mountpoint, and the mounted filesystem
238 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
239 * accepted as it stands and the mounted directory is
240 * returned. Otherwise the covered directory is returned.
241 * NOTE: this mountpoint crossing is not supported properly by all
242 * clients and is explicitly disallowed for NFSv3
243 * NeilBrown <neilb@cse.unsw.edu.au>
244 */
245 __be32
246 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
247 unsigned int len, struct svc_fh *resfh)
248 {
249 struct svc_export *exp;
250 struct dentry *dentry;
251 __be32 err;
252
253 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
254 if (err)
255 return err;
256 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
257 if (err)
258 return err;
259 err = check_nfsd_access(exp, rqstp);
260 if (err)
261 goto out;
262 /*
263 * Note: we compose the file handle now, but as the
264 * dentry may be negative, it may need to be updated.
265 */
266 err = fh_compose(resfh, exp, dentry, fhp);
267 if (!err && !dentry->d_inode)
268 err = nfserr_noent;
269 out:
270 dput(dentry);
271 exp_put(exp);
272 return err;
273 }
274
275 static int nfsd_break_lease(struct inode *inode)
276 {
277 if (!S_ISREG(inode->i_mode))
278 return 0;
279 return break_lease(inode, O_WRONLY | O_NONBLOCK);
280 }
281
282 /*
283 * Commit metadata changes to stable storage.
284 */
285 static int
286 commit_metadata(struct svc_fh *fhp)
287 {
288 struct inode *inode = fhp->fh_dentry->d_inode;
289 const struct export_operations *export_ops = inode->i_sb->s_export_op;
290
291 if (!EX_ISSYNC(fhp->fh_export))
292 return 0;
293
294 if (export_ops->commit_metadata)
295 return export_ops->commit_metadata(inode);
296 return sync_inode_metadata(inode, 1);
297 }
298
299 /*
300 * Set various file attributes.
301 * N.B. After this call fhp needs an fh_put
302 */
303 __be32
304 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
305 int check_guard, time_t guardtime)
306 {
307 struct dentry *dentry;
308 struct inode *inode;
309 int accmode = NFSD_MAY_SATTR;
310 umode_t ftype = 0;
311 __be32 err;
312 int host_err;
313 int size_change = 0;
314
315 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
316 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
317 if (iap->ia_valid & ATTR_SIZE)
318 ftype = S_IFREG;
319
320 /* Get inode */
321 err = fh_verify(rqstp, fhp, ftype, accmode);
322 if (err)
323 goto out;
324
325 dentry = fhp->fh_dentry;
326 inode = dentry->d_inode;
327
328 /* Ignore any mode updates on symlinks */
329 if (S_ISLNK(inode->i_mode))
330 iap->ia_valid &= ~ATTR_MODE;
331
332 if (!iap->ia_valid)
333 goto out;
334
335 /*
336 * NFSv2 does not differentiate between "set-[ac]time-to-now"
337 * which only requires access, and "set-[ac]time-to-X" which
338 * requires ownership.
339 * So if it looks like it might be "set both to the same time which
340 * is close to now", and if inode_change_ok fails, then we
341 * convert to "set to now" instead of "set to explicit time"
342 *
343 * We only call inode_change_ok as the last test as technically
344 * it is not an interface that we should be using. It is only
345 * valid if the filesystem does not define it's own i_op->setattr.
346 */
347 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
348 #define MAX_TOUCH_TIME_ERROR (30*60)
349 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
350 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
351 /*
352 * Looks probable.
353 *
354 * Now just make sure time is in the right ballpark.
355 * Solaris, at least, doesn't seem to care what the time
356 * request is. We require it be within 30 minutes of now.
357 */
358 time_t delta = iap->ia_atime.tv_sec - get_seconds();
359 if (delta < 0)
360 delta = -delta;
361 if (delta < MAX_TOUCH_TIME_ERROR &&
362 inode_change_ok(inode, iap) != 0) {
363 /*
364 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
365 * This will cause notify_change to set these times
366 * to "now"
367 */
368 iap->ia_valid &= ~BOTH_TIME_SET;
369 }
370 }
371
372 /*
373 * The size case is special.
374 * It changes the file as well as the attributes.
375 */
376 if (iap->ia_valid & ATTR_SIZE) {
377 if (iap->ia_size < inode->i_size) {
378 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
379 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
380 if (err)
381 goto out;
382 }
383
384 host_err = get_write_access(inode);
385 if (host_err)
386 goto out_nfserr;
387
388 size_change = 1;
389 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
390 if (host_err) {
391 put_write_access(inode);
392 goto out_nfserr;
393 }
394 }
395
396 /* sanitize the mode change */
397 if (iap->ia_valid & ATTR_MODE) {
398 iap->ia_mode &= S_IALLUGO;
399 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
400 }
401
402 /* Revoke setuid/setgid on chown */
403 if (!S_ISDIR(inode->i_mode) &&
404 (((iap->ia_valid & ATTR_UID) && !uid_eq(iap->ia_uid, inode->i_uid)) ||
405 ((iap->ia_valid & ATTR_GID) && !gid_eq(iap->ia_gid, inode->i_gid)))) {
406 iap->ia_valid |= ATTR_KILL_PRIV;
407 if (iap->ia_valid & ATTR_MODE) {
408 /* we're setting mode too, just clear the s*id bits */
409 iap->ia_mode &= ~S_ISUID;
410 if (iap->ia_mode & S_IXGRP)
411 iap->ia_mode &= ~S_ISGID;
412 } else {
413 /* set ATTR_KILL_* bits and let VFS handle it */
414 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
415 }
416 }
417
418 /* Change the attributes. */
419
420 iap->ia_valid |= ATTR_CTIME;
421
422 err = nfserr_notsync;
423 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
424 host_err = nfsd_break_lease(inode);
425 if (host_err)
426 goto out_nfserr;
427 fh_lock(fhp);
428
429 host_err = notify_change(dentry, iap);
430 err = nfserrno(host_err);
431 fh_unlock(fhp);
432 }
433 if (size_change)
434 put_write_access(inode);
435 if (!err)
436 commit_metadata(fhp);
437 out:
438 return err;
439
440 out_nfserr:
441 err = nfserrno(host_err);
442 goto out;
443 }
444
445 #if defined(CONFIG_NFSD_V2_ACL) || \
446 defined(CONFIG_NFSD_V3_ACL) || \
447 defined(CONFIG_NFSD_V4)
448 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
449 {
450 ssize_t buflen;
451 ssize_t ret;
452
453 buflen = vfs_getxattr(dentry, key, NULL, 0);
454 if (buflen <= 0)
455 return buflen;
456
457 *buf = kmalloc(buflen, GFP_KERNEL);
458 if (!*buf)
459 return -ENOMEM;
460
461 ret = vfs_getxattr(dentry, key, *buf, buflen);
462 if (ret < 0)
463 kfree(*buf);
464 return ret;
465 }
466 #endif
467
468 #if defined(CONFIG_NFSD_V4)
469 static int
470 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
471 {
472 int len;
473 size_t buflen;
474 char *buf = NULL;
475 int error = 0;
476
477 buflen = posix_acl_xattr_size(pacl->a_count);
478 buf = kmalloc(buflen, GFP_KERNEL);
479 error = -ENOMEM;
480 if (buf == NULL)
481 goto out;
482
483 len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen);
484 if (len < 0) {
485 error = len;
486 goto out;
487 }
488
489 error = vfs_setxattr(dentry, key, buf, len, 0);
490 out:
491 kfree(buf);
492 return error;
493 }
494
495 __be32
496 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
497 struct nfs4_acl *acl)
498 {
499 __be32 error;
500 int host_error;
501 struct dentry *dentry;
502 struct inode *inode;
503 struct posix_acl *pacl = NULL, *dpacl = NULL;
504 unsigned int flags = 0;
505
506 /* Get inode */
507 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
508 if (error)
509 return error;
510
511 dentry = fhp->fh_dentry;
512 inode = dentry->d_inode;
513 if (S_ISDIR(inode->i_mode))
514 flags = NFS4_ACL_DIR;
515
516 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
517 if (host_error == -EINVAL) {
518 return nfserr_attrnotsupp;
519 } else if (host_error < 0)
520 goto out_nfserr;
521
522 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
523 if (host_error < 0)
524 goto out_release;
525
526 if (S_ISDIR(inode->i_mode))
527 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
528
529 out_release:
530 posix_acl_release(pacl);
531 posix_acl_release(dpacl);
532 out_nfserr:
533 if (host_error == -EOPNOTSUPP)
534 return nfserr_attrnotsupp;
535 else
536 return nfserrno(host_error);
537 }
538
539 static struct posix_acl *
540 _get_posix_acl(struct dentry *dentry, char *key)
541 {
542 void *buf = NULL;
543 struct posix_acl *pacl = NULL;
544 int buflen;
545
546 buflen = nfsd_getxattr(dentry, key, &buf);
547 if (!buflen)
548 buflen = -ENODATA;
549 if (buflen <= 0)
550 return ERR_PTR(buflen);
551
552 pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen);
553 kfree(buf);
554 return pacl;
555 }
556
557 int
558 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
559 {
560 struct inode *inode = dentry->d_inode;
561 int error = 0;
562 struct posix_acl *pacl = NULL, *dpacl = NULL;
563 unsigned int flags = 0;
564
565 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
566 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
567 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
568 if (IS_ERR(pacl)) {
569 error = PTR_ERR(pacl);
570 pacl = NULL;
571 goto out;
572 }
573
574 if (S_ISDIR(inode->i_mode)) {
575 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
576 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
577 dpacl = NULL;
578 else if (IS_ERR(dpacl)) {
579 error = PTR_ERR(dpacl);
580 dpacl = NULL;
581 goto out;
582 }
583 flags = NFS4_ACL_DIR;
584 }
585
586 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
587 if (IS_ERR(*acl)) {
588 error = PTR_ERR(*acl);
589 *acl = NULL;
590 }
591 out:
592 posix_acl_release(pacl);
593 posix_acl_release(dpacl);
594 return error;
595 }
596
597 /*
598 * NFS junction information is stored in an extended attribute.
599 */
600 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
601
602 /**
603 * nfsd4_is_junction - Test if an object could be an NFS junction
604 *
605 * @dentry: object to test
606 *
607 * Returns 1 if "dentry" appears to contain NFS junction information.
608 * Otherwise 0 is returned.
609 */
610 int nfsd4_is_junction(struct dentry *dentry)
611 {
612 struct inode *inode = dentry->d_inode;
613
614 if (inode == NULL)
615 return 0;
616 if (inode->i_mode & S_IXUGO)
617 return 0;
618 if (!(inode->i_mode & S_ISVTX))
619 return 0;
620 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
621 return 0;
622 return 1;
623 }
624 #endif /* defined(CONFIG_NFSD_V4) */
625
626 #ifdef CONFIG_NFSD_V3
627 /*
628 * Check server access rights to a file system object
629 */
630 struct accessmap {
631 u32 access;
632 int how;
633 };
634 static struct accessmap nfs3_regaccess[] = {
635 { NFS3_ACCESS_READ, NFSD_MAY_READ },
636 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
637 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
638 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
639
640 { 0, 0 }
641 };
642
643 static struct accessmap nfs3_diraccess[] = {
644 { NFS3_ACCESS_READ, NFSD_MAY_READ },
645 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
646 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
647 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
648 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
649
650 { 0, 0 }
651 };
652
653 static struct accessmap nfs3_anyaccess[] = {
654 /* Some clients - Solaris 2.6 at least, make an access call
655 * to the server to check for access for things like /dev/null
656 * (which really, the server doesn't care about). So
657 * We provide simple access checking for them, looking
658 * mainly at mode bits, and we make sure to ignore read-only
659 * filesystem checks
660 */
661 { NFS3_ACCESS_READ, NFSD_MAY_READ },
662 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
663 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
664 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
665
666 { 0, 0 }
667 };
668
669 __be32
670 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
671 {
672 struct accessmap *map;
673 struct svc_export *export;
674 struct dentry *dentry;
675 u32 query, result = 0, sresult = 0;
676 __be32 error;
677
678 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
679 if (error)
680 goto out;
681
682 export = fhp->fh_export;
683 dentry = fhp->fh_dentry;
684
685 if (S_ISREG(dentry->d_inode->i_mode))
686 map = nfs3_regaccess;
687 else if (S_ISDIR(dentry->d_inode->i_mode))
688 map = nfs3_diraccess;
689 else
690 map = nfs3_anyaccess;
691
692
693 query = *access;
694 for (; map->access; map++) {
695 if (map->access & query) {
696 __be32 err2;
697
698 sresult |= map->access;
699
700 err2 = nfsd_permission(rqstp, export, dentry, map->how);
701 switch (err2) {
702 case nfs_ok:
703 result |= map->access;
704 break;
705
706 /* the following error codes just mean the access was not allowed,
707 * rather than an error occurred */
708 case nfserr_rofs:
709 case nfserr_acces:
710 case nfserr_perm:
711 /* simply don't "or" in the access bit. */
712 break;
713 default:
714 error = err2;
715 goto out;
716 }
717 }
718 }
719 *access = result;
720 if (supported)
721 *supported = sresult;
722
723 out:
724 return error;
725 }
726 #endif /* CONFIG_NFSD_V3 */
727
728 static int nfsd_open_break_lease(struct inode *inode, int access)
729 {
730 unsigned int mode;
731
732 if (access & NFSD_MAY_NOT_BREAK_LEASE)
733 return 0;
734 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
735 return break_lease(inode, mode | O_NONBLOCK);
736 }
737
738 /*
739 * Open an existing file or directory.
740 * The may_flags argument indicates the type of open (read/write/lock)
741 * and additional flags.
742 * N.B. After this call fhp needs an fh_put
743 */
744 __be32
745 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
746 int may_flags, struct file **filp)
747 {
748 struct path path;
749 struct inode *inode;
750 int flags = O_RDONLY|O_LARGEFILE;
751 __be32 err;
752 int host_err = 0;
753
754 validate_process_creds();
755
756 /*
757 * If we get here, then the client has already done an "open",
758 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
759 * in case a chmod has now revoked permission.
760 *
761 * Arguably we should also allow the owner override for
762 * directories, but we never have and it doesn't seem to have
763 * caused anyone a problem. If we were to change this, note
764 * also that our filldir callbacks would need a variant of
765 * lookup_one_len that doesn't check permissions.
766 */
767 if (type == S_IFREG)
768 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
769 err = fh_verify(rqstp, fhp, type, may_flags);
770 if (err)
771 goto out;
772
773 path.mnt = fhp->fh_export->ex_path.mnt;
774 path.dentry = fhp->fh_dentry;
775 inode = path.dentry->d_inode;
776
777 /* Disallow write access to files with the append-only bit set
778 * or any access when mandatory locking enabled
779 */
780 err = nfserr_perm;
781 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
782 goto out;
783 /*
784 * We must ignore files (but only files) which might have mandatory
785 * locks on them because there is no way to know if the accesser has
786 * the lock.
787 */
788 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
789 goto out;
790
791 if (!inode->i_fop)
792 goto out;
793
794 host_err = nfsd_open_break_lease(inode, may_flags);
795 if (host_err) /* NOMEM or WOULDBLOCK */
796 goto out_nfserr;
797
798 if (may_flags & NFSD_MAY_WRITE) {
799 if (may_flags & NFSD_MAY_READ)
800 flags = O_RDWR|O_LARGEFILE;
801 else
802 flags = O_WRONLY|O_LARGEFILE;
803 }
804 *filp = dentry_open(&path, flags, current_cred());
805 if (IS_ERR(*filp))
806 host_err = PTR_ERR(*filp);
807 else {
808 host_err = ima_file_check(*filp, may_flags);
809
810 if (may_flags & NFSD_MAY_64BIT_COOKIE)
811 (*filp)->f_mode |= FMODE_64BITHASH;
812 else
813 (*filp)->f_mode |= FMODE_32BITHASH;
814 }
815
816 out_nfserr:
817 err = nfserrno(host_err);
818 out:
819 validate_process_creds();
820 return err;
821 }
822
823 /*
824 * Close a file.
825 */
826 void
827 nfsd_close(struct file *filp)
828 {
829 fput(filp);
830 }
831
832 /*
833 * Obtain the readahead parameters for the file
834 * specified by (dev, ino).
835 */
836
837 static inline struct raparms *
838 nfsd_get_raparms(dev_t dev, ino_t ino)
839 {
840 struct raparms *ra, **rap, **frap = NULL;
841 int depth = 0;
842 unsigned int hash;
843 struct raparm_hbucket *rab;
844
845 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
846 rab = &raparm_hash[hash];
847
848 spin_lock(&rab->pb_lock);
849 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
850 if (ra->p_ino == ino && ra->p_dev == dev)
851 goto found;
852 depth++;
853 if (ra->p_count == 0)
854 frap = rap;
855 }
856 depth = nfsdstats.ra_size;
857 if (!frap) {
858 spin_unlock(&rab->pb_lock);
859 return NULL;
860 }
861 rap = frap;
862 ra = *frap;
863 ra->p_dev = dev;
864 ra->p_ino = ino;
865 ra->p_set = 0;
866 ra->p_hindex = hash;
867 found:
868 if (rap != &rab->pb_head) {
869 *rap = ra->p_next;
870 ra->p_next = rab->pb_head;
871 rab->pb_head = ra;
872 }
873 ra->p_count++;
874 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
875 spin_unlock(&rab->pb_lock);
876 return ra;
877 }
878
879 /*
880 * Grab and keep cached pages associated with a file in the svc_rqst
881 * so that they can be passed to the network sendmsg/sendpage routines
882 * directly. They will be released after the sending has completed.
883 */
884 static int
885 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
886 struct splice_desc *sd)
887 {
888 struct svc_rqst *rqstp = sd->u.data;
889 struct page **pp = rqstp->rq_next_page;
890 struct page *page = buf->page;
891 size_t size;
892
893 size = sd->len;
894
895 if (rqstp->rq_res.page_len == 0) {
896 get_page(page);
897 put_page(*rqstp->rq_next_page);
898 *(rqstp->rq_next_page++) = page;
899 rqstp->rq_res.page_base = buf->offset;
900 rqstp->rq_res.page_len = size;
901 } else if (page != pp[-1]) {
902 get_page(page);
903 if (*rqstp->rq_next_page)
904 put_page(*rqstp->rq_next_page);
905 *(rqstp->rq_next_page++) = page;
906 rqstp->rq_res.page_len += size;
907 } else
908 rqstp->rq_res.page_len += size;
909
910 return size;
911 }
912
913 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
914 struct splice_desc *sd)
915 {
916 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
917 }
918
919 static __be32
920 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
921 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
922 {
923 mm_segment_t oldfs;
924 __be32 err;
925 int host_err;
926
927 err = nfserr_perm;
928
929 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
930 struct splice_desc sd = {
931 .len = 0,
932 .total_len = *count,
933 .pos = offset,
934 .u.data = rqstp,
935 };
936
937 rqstp->rq_next_page = rqstp->rq_respages + 1;
938 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
939 } else {
940 oldfs = get_fs();
941 set_fs(KERNEL_DS);
942 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
943 set_fs(oldfs);
944 }
945
946 if (host_err >= 0) {
947 nfsdstats.io_read += host_err;
948 *count = host_err;
949 err = 0;
950 fsnotify_access(file);
951 } else
952 err = nfserrno(host_err);
953 return err;
954 }
955
956 static void kill_suid(struct dentry *dentry)
957 {
958 struct iattr ia;
959 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
960
961 mutex_lock(&dentry->d_inode->i_mutex);
962 notify_change(dentry, &ia);
963 mutex_unlock(&dentry->d_inode->i_mutex);
964 }
965
966 /*
967 * Gathered writes: If another process is currently writing to the file,
968 * there's a high chance this is another nfsd (triggered by a bulk write
969 * from a client's biod). Rather than syncing the file with each write
970 * request, we sleep for 10 msec.
971 *
972 * I don't know if this roughly approximates C. Juszak's idea of
973 * gathered writes, but it's a nice and simple solution (IMHO), and it
974 * seems to work:-)
975 *
976 * Note: we do this only in the NFSv2 case, since v3 and higher have a
977 * better tool (separate unstable writes and commits) for solving this
978 * problem.
979 */
980 static int wait_for_concurrent_writes(struct file *file)
981 {
982 struct inode *inode = file_inode(file);
983 static ino_t last_ino;
984 static dev_t last_dev;
985 int err = 0;
986
987 if (atomic_read(&inode->i_writecount) > 1
988 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
989 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
990 msleep(10);
991 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
992 }
993
994 if (inode->i_state & I_DIRTY) {
995 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
996 err = vfs_fsync(file, 0);
997 }
998 last_ino = inode->i_ino;
999 last_dev = inode->i_sb->s_dev;
1000 return err;
1001 }
1002
1003 static __be32
1004 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1005 loff_t offset, struct kvec *vec, int vlen,
1006 unsigned long *cnt, int *stablep)
1007 {
1008 struct svc_export *exp;
1009 struct dentry *dentry;
1010 struct inode *inode;
1011 mm_segment_t oldfs;
1012 __be32 err = 0;
1013 int host_err;
1014 int stable = *stablep;
1015 int use_wgather;
1016 loff_t pos = offset;
1017
1018 dentry = file->f_path.dentry;
1019 inode = dentry->d_inode;
1020 exp = fhp->fh_export;
1021
1022 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1023
1024 if (!EX_ISSYNC(exp))
1025 stable = 0;
1026
1027 /* Write the data. */
1028 oldfs = get_fs(); set_fs(KERNEL_DS);
1029 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
1030 set_fs(oldfs);
1031 if (host_err < 0)
1032 goto out_nfserr;
1033 *cnt = host_err;
1034 nfsdstats.io_write += host_err;
1035 fsnotify_modify(file);
1036
1037 /* clear setuid/setgid flag after write */
1038 if (inode->i_mode & (S_ISUID | S_ISGID))
1039 kill_suid(dentry);
1040
1041 if (stable) {
1042 if (use_wgather)
1043 host_err = wait_for_concurrent_writes(file);
1044 else
1045 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
1046 }
1047
1048 out_nfserr:
1049 dprintk("nfsd: write complete host_err=%d\n", host_err);
1050 if (host_err >= 0)
1051 err = 0;
1052 else
1053 err = nfserrno(host_err);
1054 return err;
1055 }
1056
1057 /*
1058 * Read data from a file. count must contain the requested read count
1059 * on entry. On return, *count contains the number of bytes actually read.
1060 * N.B. After this call fhp needs an fh_put
1061 */
1062 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1063 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1064 {
1065 struct file *file;
1066 struct inode *inode;
1067 struct raparms *ra;
1068 __be32 err;
1069
1070 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1071 if (err)
1072 return err;
1073
1074 inode = file_inode(file);
1075
1076 /* Get readahead parameters */
1077 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1078
1079 if (ra && ra->p_set)
1080 file->f_ra = ra->p_ra;
1081
1082 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1083
1084 /* Write back readahead params */
1085 if (ra) {
1086 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1087 spin_lock(&rab->pb_lock);
1088 ra->p_ra = file->f_ra;
1089 ra->p_set = 1;
1090 ra->p_count--;
1091 spin_unlock(&rab->pb_lock);
1092 }
1093
1094 nfsd_close(file);
1095 return err;
1096 }
1097
1098 /* As above, but use the provided file descriptor. */
1099 __be32
1100 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1101 loff_t offset, struct kvec *vec, int vlen,
1102 unsigned long *count)
1103 {
1104 __be32 err;
1105
1106 if (file) {
1107 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1108 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1109 if (err)
1110 goto out;
1111 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1112 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1113 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1114 out:
1115 return err;
1116 }
1117
1118 /*
1119 * Write data to a file.
1120 * The stable flag requests synchronous writes.
1121 * N.B. After this call fhp needs an fh_put
1122 */
1123 __be32
1124 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1125 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1126 int *stablep)
1127 {
1128 __be32 err = 0;
1129
1130 if (file) {
1131 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1132 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1133 if (err)
1134 goto out;
1135 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1136 stablep);
1137 } else {
1138 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1139 if (err)
1140 goto out;
1141
1142 if (cnt)
1143 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1144 cnt, stablep);
1145 nfsd_close(file);
1146 }
1147 out:
1148 return err;
1149 }
1150
1151 #ifdef CONFIG_NFSD_V3
1152 /*
1153 * Commit all pending writes to stable storage.
1154 *
1155 * Note: we only guarantee that data that lies within the range specified
1156 * by the 'offset' and 'count' parameters will be synced.
1157 *
1158 * Unfortunately we cannot lock the file to make sure we return full WCC
1159 * data to the client, as locking happens lower down in the filesystem.
1160 */
1161 __be32
1162 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1163 loff_t offset, unsigned long count)
1164 {
1165 struct file *file;
1166 loff_t end = LLONG_MAX;
1167 __be32 err = nfserr_inval;
1168
1169 if (offset < 0)
1170 goto out;
1171 if (count != 0) {
1172 end = offset + (loff_t)count - 1;
1173 if (end < offset)
1174 goto out;
1175 }
1176
1177 err = nfsd_open(rqstp, fhp, S_IFREG,
1178 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1179 if (err)
1180 goto out;
1181 if (EX_ISSYNC(fhp->fh_export)) {
1182 int err2 = vfs_fsync_range(file, offset, end, 0);
1183
1184 if (err2 != -EINVAL)
1185 err = nfserrno(err2);
1186 else
1187 err = nfserr_notsupp;
1188 }
1189
1190 nfsd_close(file);
1191 out:
1192 return err;
1193 }
1194 #endif /* CONFIG_NFSD_V3 */
1195
1196 static __be32
1197 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1198 struct iattr *iap)
1199 {
1200 /*
1201 * Mode has already been set earlier in create:
1202 */
1203 iap->ia_valid &= ~ATTR_MODE;
1204 /*
1205 * Setting uid/gid works only for root. Irix appears to
1206 * send along the gid on create when it tries to implement
1207 * setgid directories via NFS:
1208 */
1209 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1210 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1211 if (iap->ia_valid)
1212 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1213 return 0;
1214 }
1215
1216 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1217 * setting size to 0 may fail for some specific file systems by the permission
1218 * checking which requires WRITE permission but the mode is 000.
1219 * we ignore the resizing(to 0) on the just new created file, since the size is
1220 * 0 after file created.
1221 *
1222 * call this only after vfs_create() is called.
1223 * */
1224 static void
1225 nfsd_check_ignore_resizing(struct iattr *iap)
1226 {
1227 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1228 iap->ia_valid &= ~ATTR_SIZE;
1229 }
1230
1231 /*
1232 * Create a file (regular, directory, device, fifo); UNIX sockets
1233 * not yet implemented.
1234 * If the response fh has been verified, the parent directory should
1235 * already be locked. Note that the parent directory is left locked.
1236 *
1237 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1238 */
1239 __be32
1240 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1241 char *fname, int flen, struct iattr *iap,
1242 int type, dev_t rdev, struct svc_fh *resfhp)
1243 {
1244 struct dentry *dentry, *dchild = NULL;
1245 struct inode *dirp;
1246 __be32 err;
1247 __be32 err2;
1248 int host_err;
1249
1250 err = nfserr_perm;
1251 if (!flen)
1252 goto out;
1253 err = nfserr_exist;
1254 if (isdotent(fname, flen))
1255 goto out;
1256
1257 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1258 if (err)
1259 goto out;
1260
1261 dentry = fhp->fh_dentry;
1262 dirp = dentry->d_inode;
1263
1264 err = nfserr_notdir;
1265 if (!dirp->i_op->lookup)
1266 goto out;
1267 /*
1268 * Check whether the response file handle has been verified yet.
1269 * If it has, the parent directory should already be locked.
1270 */
1271 if (!resfhp->fh_dentry) {
1272 host_err = fh_want_write(fhp);
1273 if (host_err)
1274 goto out_nfserr;
1275
1276 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1277 fh_lock_nested(fhp, I_MUTEX_PARENT);
1278 dchild = lookup_one_len(fname, dentry, flen);
1279 host_err = PTR_ERR(dchild);
1280 if (IS_ERR(dchild))
1281 goto out_nfserr;
1282 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1283 if (err)
1284 goto out;
1285 } else {
1286 /* called from nfsd_proc_create */
1287 dchild = dget(resfhp->fh_dentry);
1288 if (!fhp->fh_locked) {
1289 /* not actually possible */
1290 printk(KERN_ERR
1291 "nfsd_create: parent %s/%s not locked!\n",
1292 dentry->d_parent->d_name.name,
1293 dentry->d_name.name);
1294 err = nfserr_io;
1295 goto out;
1296 }
1297 }
1298 /*
1299 * Make sure the child dentry is still negative ...
1300 */
1301 err = nfserr_exist;
1302 if (dchild->d_inode) {
1303 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1304 dentry->d_name.name, dchild->d_name.name);
1305 goto out;
1306 }
1307
1308 if (!(iap->ia_valid & ATTR_MODE))
1309 iap->ia_mode = 0;
1310 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1311
1312 err = nfserr_inval;
1313 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1314 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1315 type);
1316 goto out;
1317 }
1318
1319 /*
1320 * Get the dir op function pointer.
1321 */
1322 err = 0;
1323 host_err = 0;
1324 switch (type) {
1325 case S_IFREG:
1326 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1327 if (!host_err)
1328 nfsd_check_ignore_resizing(iap);
1329 break;
1330 case S_IFDIR:
1331 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1332 break;
1333 case S_IFCHR:
1334 case S_IFBLK:
1335 case S_IFIFO:
1336 case S_IFSOCK:
1337 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1338 break;
1339 }
1340 if (host_err < 0)
1341 goto out_nfserr;
1342
1343 err = nfsd_create_setattr(rqstp, resfhp, iap);
1344
1345 /*
1346 * nfsd_setattr already committed the child. Transactional filesystems
1347 * had a chance to commit changes for both parent and child
1348 * simultaneously making the following commit_metadata a noop.
1349 */
1350 err2 = nfserrno(commit_metadata(fhp));
1351 if (err2)
1352 err = err2;
1353 /*
1354 * Update the file handle to get the new inode info.
1355 */
1356 if (!err)
1357 err = fh_update(resfhp);
1358 out:
1359 if (dchild && !IS_ERR(dchild))
1360 dput(dchild);
1361 return err;
1362
1363 out_nfserr:
1364 err = nfserrno(host_err);
1365 goto out;
1366 }
1367
1368 #ifdef CONFIG_NFSD_V3
1369
1370 static inline int nfsd_create_is_exclusive(int createmode)
1371 {
1372 return createmode == NFS3_CREATE_EXCLUSIVE
1373 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1374 }
1375
1376 /*
1377 * NFSv3 and NFSv4 version of nfsd_create
1378 */
1379 __be32
1380 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1381 char *fname, int flen, struct iattr *iap,
1382 struct svc_fh *resfhp, int createmode, u32 *verifier,
1383 bool *truncp, bool *created)
1384 {
1385 struct dentry *dentry, *dchild = NULL;
1386 struct inode *dirp;
1387 __be32 err;
1388 int host_err;
1389 __u32 v_mtime=0, v_atime=0;
1390
1391 err = nfserr_perm;
1392 if (!flen)
1393 goto out;
1394 err = nfserr_exist;
1395 if (isdotent(fname, flen))
1396 goto out;
1397 if (!(iap->ia_valid & ATTR_MODE))
1398 iap->ia_mode = 0;
1399 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1400 if (err)
1401 goto out;
1402
1403 dentry = fhp->fh_dentry;
1404 dirp = dentry->d_inode;
1405
1406 /* Get all the sanity checks out of the way before
1407 * we lock the parent. */
1408 err = nfserr_notdir;
1409 if (!dirp->i_op->lookup)
1410 goto out;
1411
1412 host_err = fh_want_write(fhp);
1413 if (host_err)
1414 goto out_nfserr;
1415
1416 fh_lock_nested(fhp, I_MUTEX_PARENT);
1417
1418 /*
1419 * Compose the response file handle.
1420 */
1421 dchild = lookup_one_len(fname, dentry, flen);
1422 host_err = PTR_ERR(dchild);
1423 if (IS_ERR(dchild))
1424 goto out_nfserr;
1425
1426 /* If file doesn't exist, check for permissions to create one */
1427 if (!dchild->d_inode) {
1428 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1429 if (err)
1430 goto out;
1431 }
1432
1433 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1434 if (err)
1435 goto out;
1436
1437 if (nfsd_create_is_exclusive(createmode)) {
1438 /* solaris7 gets confused (bugid 4218508) if these have
1439 * the high bit set, so just clear the high bits. If this is
1440 * ever changed to use different attrs for storing the
1441 * verifier, then do_open_lookup() will also need to be fixed
1442 * accordingly.
1443 */
1444 v_mtime = verifier[0]&0x7fffffff;
1445 v_atime = verifier[1]&0x7fffffff;
1446 }
1447
1448 if (dchild->d_inode) {
1449 err = 0;
1450
1451 switch (createmode) {
1452 case NFS3_CREATE_UNCHECKED:
1453 if (! S_ISREG(dchild->d_inode->i_mode))
1454 goto out;
1455 else if (truncp) {
1456 /* in nfsv4, we need to treat this case a little
1457 * differently. we don't want to truncate the
1458 * file now; this would be wrong if the OPEN
1459 * fails for some other reason. furthermore,
1460 * if the size is nonzero, we should ignore it
1461 * according to spec!
1462 */
1463 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1464 }
1465 else {
1466 iap->ia_valid &= ATTR_SIZE;
1467 goto set_attr;
1468 }
1469 break;
1470 case NFS3_CREATE_EXCLUSIVE:
1471 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1472 && dchild->d_inode->i_atime.tv_sec == v_atime
1473 && dchild->d_inode->i_size == 0 ) {
1474 if (created)
1475 *created = 1;
1476 break;
1477 }
1478 case NFS4_CREATE_EXCLUSIVE4_1:
1479 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1480 && dchild->d_inode->i_atime.tv_sec == v_atime
1481 && dchild->d_inode->i_size == 0 ) {
1482 if (created)
1483 *created = 1;
1484 goto set_attr;
1485 }
1486 /* fallthru */
1487 case NFS3_CREATE_GUARDED:
1488 err = nfserr_exist;
1489 }
1490 fh_drop_write(fhp);
1491 goto out;
1492 }
1493
1494 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1495 if (host_err < 0) {
1496 fh_drop_write(fhp);
1497 goto out_nfserr;
1498 }
1499 if (created)
1500 *created = 1;
1501
1502 nfsd_check_ignore_resizing(iap);
1503
1504 if (nfsd_create_is_exclusive(createmode)) {
1505 /* Cram the verifier into atime/mtime */
1506 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1507 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1508 /* XXX someone who knows this better please fix it for nsec */
1509 iap->ia_mtime.tv_sec = v_mtime;
1510 iap->ia_atime.tv_sec = v_atime;
1511 iap->ia_mtime.tv_nsec = 0;
1512 iap->ia_atime.tv_nsec = 0;
1513 }
1514
1515 set_attr:
1516 err = nfsd_create_setattr(rqstp, resfhp, iap);
1517
1518 /*
1519 * nfsd_setattr already committed the child (and possibly also the parent).
1520 */
1521 if (!err)
1522 err = nfserrno(commit_metadata(fhp));
1523
1524 /*
1525 * Update the filehandle to get the new inode info.
1526 */
1527 if (!err)
1528 err = fh_update(resfhp);
1529
1530 out:
1531 fh_unlock(fhp);
1532 if (dchild && !IS_ERR(dchild))
1533 dput(dchild);
1534 fh_drop_write(fhp);
1535 return err;
1536
1537 out_nfserr:
1538 err = nfserrno(host_err);
1539 goto out;
1540 }
1541 #endif /* CONFIG_NFSD_V3 */
1542
1543 /*
1544 * Read a symlink. On entry, *lenp must contain the maximum path length that
1545 * fits into the buffer. On return, it contains the true length.
1546 * N.B. After this call fhp needs an fh_put
1547 */
1548 __be32
1549 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1550 {
1551 struct inode *inode;
1552 mm_segment_t oldfs;
1553 __be32 err;
1554 int host_err;
1555 struct path path;
1556
1557 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1558 if (err)
1559 goto out;
1560
1561 path.mnt = fhp->fh_export->ex_path.mnt;
1562 path.dentry = fhp->fh_dentry;
1563 inode = path.dentry->d_inode;
1564
1565 err = nfserr_inval;
1566 if (!inode->i_op->readlink)
1567 goto out;
1568
1569 touch_atime(&path);
1570 /* N.B. Why does this call need a get_fs()??
1571 * Remove the set_fs and watch the fireworks:-) --okir
1572 */
1573
1574 oldfs = get_fs(); set_fs(KERNEL_DS);
1575 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1576 set_fs(oldfs);
1577
1578 if (host_err < 0)
1579 goto out_nfserr;
1580 *lenp = host_err;
1581 err = 0;
1582 out:
1583 return err;
1584
1585 out_nfserr:
1586 err = nfserrno(host_err);
1587 goto out;
1588 }
1589
1590 /*
1591 * Create a symlink and look up its inode
1592 * N.B. After this call _both_ fhp and resfhp need an fh_put
1593 */
1594 __be32
1595 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1596 char *fname, int flen,
1597 char *path, int plen,
1598 struct svc_fh *resfhp,
1599 struct iattr *iap)
1600 {
1601 struct dentry *dentry, *dnew;
1602 __be32 err, cerr;
1603 int host_err;
1604
1605 err = nfserr_noent;
1606 if (!flen || !plen)
1607 goto out;
1608 err = nfserr_exist;
1609 if (isdotent(fname, flen))
1610 goto out;
1611
1612 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1613 if (err)
1614 goto out;
1615
1616 host_err = fh_want_write(fhp);
1617 if (host_err)
1618 goto out_nfserr;
1619
1620 fh_lock(fhp);
1621 dentry = fhp->fh_dentry;
1622 dnew = lookup_one_len(fname, dentry, flen);
1623 host_err = PTR_ERR(dnew);
1624 if (IS_ERR(dnew))
1625 goto out_nfserr;
1626
1627 if (unlikely(path[plen] != 0)) {
1628 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1629 if (path_alloced == NULL)
1630 host_err = -ENOMEM;
1631 else {
1632 strncpy(path_alloced, path, plen);
1633 path_alloced[plen] = 0;
1634 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1635 kfree(path_alloced);
1636 }
1637 } else
1638 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1639 err = nfserrno(host_err);
1640 if (!err)
1641 err = nfserrno(commit_metadata(fhp));
1642 fh_unlock(fhp);
1643
1644 fh_drop_write(fhp);
1645
1646 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1647 dput(dnew);
1648 if (err==0) err = cerr;
1649 out:
1650 return err;
1651
1652 out_nfserr:
1653 err = nfserrno(host_err);
1654 goto out;
1655 }
1656
1657 /*
1658 * Create a hardlink
1659 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1660 */
1661 __be32
1662 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1663 char *name, int len, struct svc_fh *tfhp)
1664 {
1665 struct dentry *ddir, *dnew, *dold;
1666 struct inode *dirp;
1667 __be32 err;
1668 int host_err;
1669
1670 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1671 if (err)
1672 goto out;
1673 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1674 if (err)
1675 goto out;
1676 err = nfserr_isdir;
1677 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1678 goto out;
1679 err = nfserr_perm;
1680 if (!len)
1681 goto out;
1682 err = nfserr_exist;
1683 if (isdotent(name, len))
1684 goto out;
1685
1686 host_err = fh_want_write(tfhp);
1687 if (host_err) {
1688 err = nfserrno(host_err);
1689 goto out;
1690 }
1691
1692 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1693 ddir = ffhp->fh_dentry;
1694 dirp = ddir->d_inode;
1695
1696 dnew = lookup_one_len(name, ddir, len);
1697 host_err = PTR_ERR(dnew);
1698 if (IS_ERR(dnew))
1699 goto out_nfserr;
1700
1701 dold = tfhp->fh_dentry;
1702
1703 err = nfserr_noent;
1704 if (!dold->d_inode)
1705 goto out_dput;
1706 host_err = nfsd_break_lease(dold->d_inode);
1707 if (host_err) {
1708 err = nfserrno(host_err);
1709 goto out_dput;
1710 }
1711 host_err = vfs_link(dold, dirp, dnew);
1712 if (!host_err) {
1713 err = nfserrno(commit_metadata(ffhp));
1714 if (!err)
1715 err = nfserrno(commit_metadata(tfhp));
1716 } else {
1717 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1718 err = nfserr_acces;
1719 else
1720 err = nfserrno(host_err);
1721 }
1722 out_dput:
1723 dput(dnew);
1724 out_unlock:
1725 fh_unlock(ffhp);
1726 fh_drop_write(tfhp);
1727 out:
1728 return err;
1729
1730 out_nfserr:
1731 err = nfserrno(host_err);
1732 goto out_unlock;
1733 }
1734
1735 /*
1736 * Rename a file
1737 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1738 */
1739 __be32
1740 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1741 struct svc_fh *tfhp, char *tname, int tlen)
1742 {
1743 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1744 struct inode *fdir, *tdir;
1745 __be32 err;
1746 int host_err;
1747
1748 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1749 if (err)
1750 goto out;
1751 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1752 if (err)
1753 goto out;
1754
1755 fdentry = ffhp->fh_dentry;
1756 fdir = fdentry->d_inode;
1757
1758 tdentry = tfhp->fh_dentry;
1759 tdir = tdentry->d_inode;
1760
1761 err = nfserr_perm;
1762 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1763 goto out;
1764
1765 host_err = fh_want_write(ffhp);
1766 if (host_err) {
1767 err = nfserrno(host_err);
1768 goto out;
1769 }
1770
1771 /* cannot use fh_lock as we need deadlock protective ordering
1772 * so do it by hand */
1773 trap = lock_rename(tdentry, fdentry);
1774 ffhp->fh_locked = tfhp->fh_locked = 1;
1775 fill_pre_wcc(ffhp);
1776 fill_pre_wcc(tfhp);
1777
1778 odentry = lookup_one_len(fname, fdentry, flen);
1779 host_err = PTR_ERR(odentry);
1780 if (IS_ERR(odentry))
1781 goto out_nfserr;
1782
1783 host_err = -ENOENT;
1784 if (!odentry->d_inode)
1785 goto out_dput_old;
1786 host_err = -EINVAL;
1787 if (odentry == trap)
1788 goto out_dput_old;
1789
1790 ndentry = lookup_one_len(tname, tdentry, tlen);
1791 host_err = PTR_ERR(ndentry);
1792 if (IS_ERR(ndentry))
1793 goto out_dput_old;
1794 host_err = -ENOTEMPTY;
1795 if (ndentry == trap)
1796 goto out_dput_new;
1797
1798 host_err = -EXDEV;
1799 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1800 goto out_dput_new;
1801 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1802 goto out_dput_new;
1803
1804 host_err = nfsd_break_lease(odentry->d_inode);
1805 if (host_err)
1806 goto out_dput_new;
1807 if (ndentry->d_inode) {
1808 host_err = nfsd_break_lease(ndentry->d_inode);
1809 if (host_err)
1810 goto out_dput_new;
1811 }
1812 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1813 if (!host_err) {
1814 host_err = commit_metadata(tfhp);
1815 if (!host_err)
1816 host_err = commit_metadata(ffhp);
1817 }
1818 out_dput_new:
1819 dput(ndentry);
1820 out_dput_old:
1821 dput(odentry);
1822 out_nfserr:
1823 err = nfserrno(host_err);
1824
1825 /* we cannot reply on fh_unlock on the two filehandles,
1826 * as that would do the wrong thing if the two directories
1827 * were the same, so again we do it by hand
1828 */
1829 fill_post_wcc(ffhp);
1830 fill_post_wcc(tfhp);
1831 unlock_rename(tdentry, fdentry);
1832 ffhp->fh_locked = tfhp->fh_locked = 0;
1833 fh_drop_write(ffhp);
1834
1835 out:
1836 return err;
1837 }
1838
1839 /*
1840 * Unlink a file or directory
1841 * N.B. After this call fhp needs an fh_put
1842 */
1843 __be32
1844 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1845 char *fname, int flen)
1846 {
1847 struct dentry *dentry, *rdentry;
1848 struct inode *dirp;
1849 __be32 err;
1850 int host_err;
1851
1852 err = nfserr_acces;
1853 if (!flen || isdotent(fname, flen))
1854 goto out;
1855 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1856 if (err)
1857 goto out;
1858
1859 host_err = fh_want_write(fhp);
1860 if (host_err)
1861 goto out_nfserr;
1862
1863 fh_lock_nested(fhp, I_MUTEX_PARENT);
1864 dentry = fhp->fh_dentry;
1865 dirp = dentry->d_inode;
1866
1867 rdentry = lookup_one_len(fname, dentry, flen);
1868 host_err = PTR_ERR(rdentry);
1869 if (IS_ERR(rdentry))
1870 goto out_nfserr;
1871
1872 if (!rdentry->d_inode) {
1873 dput(rdentry);
1874 err = nfserr_noent;
1875 goto out;
1876 }
1877
1878 if (!type)
1879 type = rdentry->d_inode->i_mode & S_IFMT;
1880
1881 host_err = nfsd_break_lease(rdentry->d_inode);
1882 if (host_err)
1883 goto out_put;
1884 if (type != S_IFDIR)
1885 host_err = vfs_unlink(dirp, rdentry);
1886 else
1887 host_err = vfs_rmdir(dirp, rdentry);
1888 if (!host_err)
1889 host_err = commit_metadata(fhp);
1890 out_put:
1891 dput(rdentry);
1892
1893 out_nfserr:
1894 err = nfserrno(host_err);
1895 out:
1896 return err;
1897 }
1898
1899 /*
1900 * We do this buffering because we must not call back into the file
1901 * system's ->lookup() method from the filldir callback. That may well
1902 * deadlock a number of file systems.
1903 *
1904 * This is based heavily on the implementation of same in XFS.
1905 */
1906 struct buffered_dirent {
1907 u64 ino;
1908 loff_t offset;
1909 int namlen;
1910 unsigned int d_type;
1911 char name[];
1912 };
1913
1914 struct readdir_data {
1915 struct dir_context ctx;
1916 char *dirent;
1917 size_t used;
1918 int full;
1919 };
1920
1921 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1922 loff_t offset, u64 ino, unsigned int d_type)
1923 {
1924 struct readdir_data *buf = __buf;
1925 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1926 unsigned int reclen;
1927
1928 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1929 if (buf->used + reclen > PAGE_SIZE) {
1930 buf->full = 1;
1931 return -EINVAL;
1932 }
1933
1934 de->namlen = namlen;
1935 de->offset = offset;
1936 de->ino = ino;
1937 de->d_type = d_type;
1938 memcpy(de->name, name, namlen);
1939 buf->used += reclen;
1940
1941 return 0;
1942 }
1943
1944 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1945 struct readdir_cd *cdp, loff_t *offsetp)
1946 {
1947 struct buffered_dirent *de;
1948 int host_err;
1949 int size;
1950 loff_t offset;
1951 struct readdir_data buf = {
1952 .ctx.actor = nfsd_buffered_filldir,
1953 .dirent = (void *)__get_free_page(GFP_KERNEL)
1954 };
1955
1956 if (!buf.dirent)
1957 return nfserrno(-ENOMEM);
1958
1959 offset = *offsetp;
1960
1961 while (1) {
1962 struct inode *dir_inode = file_inode(file);
1963 unsigned int reclen;
1964
1965 cdp->err = nfserr_eof; /* will be cleared on successful read */
1966 buf.used = 0;
1967 buf.full = 0;
1968
1969 host_err = iterate_dir(file, &buf.ctx);
1970 if (buf.full)
1971 host_err = 0;
1972
1973 if (host_err < 0)
1974 break;
1975
1976 size = buf.used;
1977
1978 if (!size)
1979 break;
1980
1981 /*
1982 * Various filldir functions may end up calling back into
1983 * lookup_one_len() and the file system's ->lookup() method.
1984 * These expect i_mutex to be held, as it would within readdir.
1985 */
1986 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1987 if (host_err)
1988 break;
1989
1990 de = (struct buffered_dirent *)buf.dirent;
1991 while (size > 0) {
1992 offset = de->offset;
1993
1994 if (func(cdp, de->name, de->namlen, de->offset,
1995 de->ino, de->d_type))
1996 break;
1997
1998 if (cdp->err != nfs_ok)
1999 break;
2000
2001 reclen = ALIGN(sizeof(*de) + de->namlen,
2002 sizeof(u64));
2003 size -= reclen;
2004 de = (struct buffered_dirent *)((char *)de + reclen);
2005 }
2006 mutex_unlock(&dir_inode->i_mutex);
2007 if (size > 0) /* We bailed out early */
2008 break;
2009
2010 offset = vfs_llseek(file, 0, SEEK_CUR);
2011 }
2012
2013 free_page((unsigned long)(buf.dirent));
2014
2015 if (host_err)
2016 return nfserrno(host_err);
2017
2018 *offsetp = offset;
2019 return cdp->err;
2020 }
2021
2022 /*
2023 * Read entries from a directory.
2024 * The NFSv3/4 verifier we ignore for now.
2025 */
2026 __be32
2027 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2028 struct readdir_cd *cdp, filldir_t func)
2029 {
2030 __be32 err;
2031 struct file *file;
2032 loff_t offset = *offsetp;
2033 int may_flags = NFSD_MAY_READ;
2034
2035 /* NFSv2 only supports 32 bit cookies */
2036 if (rqstp->rq_vers > 2)
2037 may_flags |= NFSD_MAY_64BIT_COOKIE;
2038
2039 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2040 if (err)
2041 goto out;
2042
2043 offset = vfs_llseek(file, offset, SEEK_SET);
2044 if (offset < 0) {
2045 err = nfserrno((int)offset);
2046 goto out_close;
2047 }
2048
2049 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2050
2051 if (err == nfserr_eof || err == nfserr_toosmall)
2052 err = nfs_ok; /* can still be found in ->err */
2053 out_close:
2054 nfsd_close(file);
2055 out:
2056 return err;
2057 }
2058
2059 /*
2060 * Get file system stats
2061 * N.B. After this call fhp needs an fh_put
2062 */
2063 __be32
2064 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2065 {
2066 __be32 err;
2067
2068 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2069 if (!err) {
2070 struct path path = {
2071 .mnt = fhp->fh_export->ex_path.mnt,
2072 .dentry = fhp->fh_dentry,
2073 };
2074 if (vfs_statfs(&path, stat))
2075 err = nfserr_io;
2076 }
2077 return err;
2078 }
2079
2080 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2081 {
2082 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2083 }
2084
2085 /*
2086 * Check for a user's access permissions to this inode.
2087 */
2088 __be32
2089 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2090 struct dentry *dentry, int acc)
2091 {
2092 struct inode *inode = dentry->d_inode;
2093 int err;
2094
2095 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2096 return 0;
2097 #if 0
2098 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2099 acc,
2100 (acc & NFSD_MAY_READ)? " read" : "",
2101 (acc & NFSD_MAY_WRITE)? " write" : "",
2102 (acc & NFSD_MAY_EXEC)? " exec" : "",
2103 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2104 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2105 (acc & NFSD_MAY_LOCK)? " lock" : "",
2106 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2107 inode->i_mode,
2108 IS_IMMUTABLE(inode)? " immut" : "",
2109 IS_APPEND(inode)? " append" : "",
2110 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2111 dprintk(" owner %d/%d user %d/%d\n",
2112 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2113 #endif
2114
2115 /* Normally we reject any write/sattr etc access on a read-only file
2116 * system. But if it is IRIX doing check on write-access for a
2117 * device special file, we ignore rofs.
2118 */
2119 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2120 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2121 if (exp_rdonly(rqstp, exp) ||
2122 __mnt_is_readonly(exp->ex_path.mnt))
2123 return nfserr_rofs;
2124 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2125 return nfserr_perm;
2126 }
2127 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2128 return nfserr_perm;
2129
2130 if (acc & NFSD_MAY_LOCK) {
2131 /* If we cannot rely on authentication in NLM requests,
2132 * just allow locks, otherwise require read permission, or
2133 * ownership
2134 */
2135 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2136 return 0;
2137 else
2138 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2139 }
2140 /*
2141 * The file owner always gets access permission for accesses that
2142 * would normally be checked at open time. This is to make
2143 * file access work even when the client has done a fchmod(fd, 0).
2144 *
2145 * However, `cp foo bar' should fail nevertheless when bar is
2146 * readonly. A sensible way to do this might be to reject all
2147 * attempts to truncate a read-only file, because a creat() call
2148 * always implies file truncation.
2149 * ... but this isn't really fair. A process may reasonably call
2150 * ftruncate on an open file descriptor on a file with perm 000.
2151 * We must trust the client to do permission checking - using "ACCESS"
2152 * with NFSv3.
2153 */
2154 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2155 uid_eq(inode->i_uid, current_fsuid()))
2156 return 0;
2157
2158 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2159 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2160
2161 /* Allow read access to binaries even when mode 111 */
2162 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2163 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2164 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2165 err = inode_permission(inode, MAY_EXEC);
2166
2167 return err? nfserrno(err) : 0;
2168 }
2169
2170 void
2171 nfsd_racache_shutdown(void)
2172 {
2173 struct raparms *raparm, *last_raparm;
2174 unsigned int i;
2175
2176 dprintk("nfsd: freeing readahead buffers.\n");
2177
2178 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2179 raparm = raparm_hash[i].pb_head;
2180 while(raparm) {
2181 last_raparm = raparm;
2182 raparm = raparm->p_next;
2183 kfree(last_raparm);
2184 }
2185 raparm_hash[i].pb_head = NULL;
2186 }
2187 }
2188 /*
2189 * Initialize readahead param cache
2190 */
2191 int
2192 nfsd_racache_init(int cache_size)
2193 {
2194 int i;
2195 int j = 0;
2196 int nperbucket;
2197 struct raparms **raparm = NULL;
2198
2199
2200 if (raparm_hash[0].pb_head)
2201 return 0;
2202 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2203 if (nperbucket < 2)
2204 nperbucket = 2;
2205 cache_size = nperbucket * RAPARM_HASH_SIZE;
2206
2207 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2208
2209 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2210 spin_lock_init(&raparm_hash[i].pb_lock);
2211
2212 raparm = &raparm_hash[i].pb_head;
2213 for (j = 0; j < nperbucket; j++) {
2214 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2215 if (!*raparm)
2216 goto out_nomem;
2217 raparm = &(*raparm)->p_next;
2218 }
2219 *raparm = NULL;
2220 }
2221
2222 nfsdstats.ra_size = cache_size;
2223 return 0;
2224
2225 out_nomem:
2226 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2227 nfsd_racache_shutdown();
2228 return -ENOMEM;
2229 }
2230
2231 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2232 struct posix_acl *
2233 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2234 {
2235 struct inode *inode = fhp->fh_dentry->d_inode;
2236 char *name;
2237 void *value = NULL;
2238 ssize_t size;
2239 struct posix_acl *acl;
2240
2241 if (!IS_POSIXACL(inode))
2242 return ERR_PTR(-EOPNOTSUPP);
2243
2244 switch (type) {
2245 case ACL_TYPE_ACCESS:
2246 name = POSIX_ACL_XATTR_ACCESS;
2247 break;
2248 case ACL_TYPE_DEFAULT:
2249 name = POSIX_ACL_XATTR_DEFAULT;
2250 break;
2251 default:
2252 return ERR_PTR(-EOPNOTSUPP);
2253 }
2254
2255 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2256 if (size < 0)
2257 return ERR_PTR(size);
2258
2259 acl = posix_acl_from_xattr(&init_user_ns, value, size);
2260 kfree(value);
2261 return acl;
2262 }
2263
2264 int
2265 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2266 {
2267 struct inode *inode = fhp->fh_dentry->d_inode;
2268 char *name;
2269 void *value = NULL;
2270 size_t size;
2271 int error;
2272
2273 if (!IS_POSIXACL(inode) ||
2274 !inode->i_op->setxattr || !inode->i_op->removexattr)
2275 return -EOPNOTSUPP;
2276 switch(type) {
2277 case ACL_TYPE_ACCESS:
2278 name = POSIX_ACL_XATTR_ACCESS;
2279 break;
2280 case ACL_TYPE_DEFAULT:
2281 name = POSIX_ACL_XATTR_DEFAULT;
2282 break;
2283 default:
2284 return -EOPNOTSUPP;
2285 }
2286
2287 if (acl && acl->a_count) {
2288 size = posix_acl_xattr_size(acl->a_count);
2289 value = kmalloc(size, GFP_KERNEL);
2290 if (!value)
2291 return -ENOMEM;
2292 error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2293 if (error < 0)
2294 goto getout;
2295 size = error;
2296 } else
2297 size = 0;
2298
2299 error = fh_want_write(fhp);
2300 if (error)
2301 goto getout;
2302 if (size)
2303 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2304 else {
2305 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2306 error = 0;
2307 else {
2308 error = vfs_removexattr(fhp->fh_dentry, name);
2309 if (error == -ENODATA)
2310 error = 0;
2311 }
2312 }
2313 fh_drop_write(fhp);
2314
2315 getout:
2316 kfree(value);
2317 return error;
2318 }
2319 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
Linus' kernel tree