projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
nfsd: warn on odd reply state in nfsd_vfs_read
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / nfsd / vfs.c
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) && iap->ia_uid != inode->i_uid) ||
405 ((iap->ia_valid & ATTR_GID) && 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_respages + rqstp->rq_resused;
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(*pp);
898 *pp = page;
899 rqstp->rq_resused++;
900 rqstp->rq_res.page_base = buf->offset;
901 rqstp->rq_res.page_len = size;
902 } else if (page != pp[-1]) {
903 get_page(page);
904 if (*pp)
905 put_page(*pp);
906 *pp = page;
907 rqstp->rq_resused++;
908 rqstp->rq_res.page_len += size;
909 } else
910 rqstp->rq_res.page_len += size;
911
912 return size;
913 }
914
915 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
916 struct splice_desc *sd)
917 {
918 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
919 }
920
921 static __be32
922 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
923 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
924 {
925 mm_segment_t oldfs;
926 __be32 err;
927 int host_err;
928
929 err = nfserr_perm;
930
931 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
932 struct splice_desc sd = {
933 .len = 0,
934 .total_len = *count,
935 .pos = offset,
936 .u.data = rqstp,
937 };
938
939 WARN_ON_ONCE(rqstp->rq_resused != 1);
940 rqstp->rq_resused = 1;
941 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
942 } else {
943 oldfs = get_fs();
944 set_fs(KERNEL_DS);
945 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
946 set_fs(oldfs);
947 }
948
949 if (host_err >= 0) {
950 nfsdstats.io_read += host_err;
951 *count = host_err;
952 err = 0;
953 fsnotify_access(file);
954 } else
955 err = nfserrno(host_err);
956 return err;
957 }
958
959 static void kill_suid(struct dentry *dentry)
960 {
961 struct iattr ia;
962 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
963
964 mutex_lock(&dentry->d_inode->i_mutex);
965 notify_change(dentry, &ia);
966 mutex_unlock(&dentry->d_inode->i_mutex);
967 }
968
969 /*
970 * Gathered writes: If another process is currently writing to the file,
971 * there's a high chance this is another nfsd (triggered by a bulk write
972 * from a client's biod). Rather than syncing the file with each write
973 * request, we sleep for 10 msec.
974 *
975 * I don't know if this roughly approximates C. Juszak's idea of
976 * gathered writes, but it's a nice and simple solution (IMHO), and it
977 * seems to work:-)
978 *
979 * Note: we do this only in the NFSv2 case, since v3 and higher have a
980 * better tool (separate unstable writes and commits) for solving this
981 * problem.
982 */
983 static int wait_for_concurrent_writes(struct file *file)
984 {
985 struct inode *inode = file->f_path.dentry->d_inode;
986 static ino_t last_ino;
987 static dev_t last_dev;
988 int err = 0;
989
990 if (atomic_read(&inode->i_writecount) > 1
991 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
992 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
993 msleep(10);
994 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
995 }
996
997 if (inode->i_state & I_DIRTY) {
998 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
999 err = vfs_fsync(file, 0);
1000 }
1001 last_ino = inode->i_ino;
1002 last_dev = inode->i_sb->s_dev;
1003 return err;
1004 }
1005
1006 static __be32
1007 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1008 loff_t offset, struct kvec *vec, int vlen,
1009 unsigned long *cnt, int *stablep)
1010 {
1011 struct svc_export *exp;
1012 struct dentry *dentry;
1013 struct inode *inode;
1014 mm_segment_t oldfs;
1015 __be32 err = 0;
1016 int host_err;
1017 int stable = *stablep;
1018 int use_wgather;
1019
1020 dentry = file->f_path.dentry;
1021 inode = dentry->d_inode;
1022 exp = fhp->fh_export;
1023
1024 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1025
1026 if (!EX_ISSYNC(exp))
1027 stable = 0;
1028
1029 /* Write the data. */
1030 oldfs = get_fs(); set_fs(KERNEL_DS);
1031 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1032 set_fs(oldfs);
1033 if (host_err < 0)
1034 goto out_nfserr;
1035 *cnt = host_err;
1036 nfsdstats.io_write += host_err;
1037 fsnotify_modify(file);
1038
1039 /* clear setuid/setgid flag after write */
1040 if (inode->i_mode & (S_ISUID | S_ISGID))
1041 kill_suid(dentry);
1042
1043 if (stable) {
1044 if (use_wgather)
1045 host_err = wait_for_concurrent_writes(file);
1046 else
1047 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
1048 }
1049
1050 out_nfserr:
1051 dprintk("nfsd: write complete host_err=%d\n", host_err);
1052 if (host_err >= 0)
1053 err = 0;
1054 else
1055 err = nfserrno(host_err);
1056 return err;
1057 }
1058
1059 /*
1060 * Read data from a file. count must contain the requested read count
1061 * on entry. On return, *count contains the number of bytes actually read.
1062 * N.B. After this call fhp needs an fh_put
1063 */
1064 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1065 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1066 {
1067 struct file *file;
1068 struct inode *inode;
1069 struct raparms *ra;
1070 __be32 err;
1071
1072 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1073 if (err)
1074 return err;
1075
1076 inode = file->f_path.dentry->d_inode;
1077
1078 /* Get readahead parameters */
1079 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1080
1081 if (ra && ra->p_set)
1082 file->f_ra = ra->p_ra;
1083
1084 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1085
1086 /* Write back readahead params */
1087 if (ra) {
1088 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1089 spin_lock(&rab->pb_lock);
1090 ra->p_ra = file->f_ra;
1091 ra->p_set = 1;
1092 ra->p_count--;
1093 spin_unlock(&rab->pb_lock);
1094 }
1095
1096 nfsd_close(file);
1097 return err;
1098 }
1099
1100 /* As above, but use the provided file descriptor. */
1101 __be32
1102 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1103 loff_t offset, struct kvec *vec, int vlen,
1104 unsigned long *count)
1105 {
1106 __be32 err;
1107
1108 if (file) {
1109 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1110 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1111 if (err)
1112 goto out;
1113 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1114 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1115 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1116 out:
1117 return err;
1118 }
1119
1120 /*
1121 * Write data to a file.
1122 * The stable flag requests synchronous writes.
1123 * N.B. After this call fhp needs an fh_put
1124 */
1125 __be32
1126 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1127 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1128 int *stablep)
1129 {
1130 __be32 err = 0;
1131
1132 if (file) {
1133 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1134 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1135 if (err)
1136 goto out;
1137 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1138 stablep);
1139 } else {
1140 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1141 if (err)
1142 goto out;
1143
1144 if (cnt)
1145 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1146 cnt, stablep);
1147 nfsd_close(file);
1148 }
1149 out:
1150 return err;
1151 }
1152
1153 #ifdef CONFIG_NFSD_V3
1154 /*
1155 * Commit all pending writes to stable storage.
1156 *
1157 * Note: we only guarantee that data that lies within the range specified
1158 * by the 'offset' and 'count' parameters will be synced.
1159 *
1160 * Unfortunately we cannot lock the file to make sure we return full WCC
1161 * data to the client, as locking happens lower down in the filesystem.
1162 */
1163 __be32
1164 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1165 loff_t offset, unsigned long count)
1166 {
1167 struct file *file;
1168 loff_t end = LLONG_MAX;
1169 __be32 err = nfserr_inval;
1170
1171 if (offset < 0)
1172 goto out;
1173 if (count != 0) {
1174 end = offset + (loff_t)count - 1;
1175 if (end < offset)
1176 goto out;
1177 }
1178
1179 err = nfsd_open(rqstp, fhp, S_IFREG,
1180 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1181 if (err)
1182 goto out;
1183 if (EX_ISSYNC(fhp->fh_export)) {
1184 int err2 = vfs_fsync_range(file, offset, end, 0);
1185
1186 if (err2 != -EINVAL)
1187 err = nfserrno(err2);
1188 else
1189 err = nfserr_notsupp;
1190 }
1191
1192 nfsd_close(file);
1193 out:
1194 return err;
1195 }
1196 #endif /* CONFIG_NFSD_V3 */
1197
1198 static __be32
1199 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1200 struct iattr *iap)
1201 {
1202 /*
1203 * Mode has already been set earlier in create:
1204 */
1205 iap->ia_valid &= ~ATTR_MODE;
1206 /*
1207 * Setting uid/gid works only for root. Irix appears to
1208 * send along the gid on create when it tries to implement
1209 * setgid directories via NFS:
1210 */
1211 if (current_fsuid() != 0)
1212 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1213 if (iap->ia_valid)
1214 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1215 return 0;
1216 }
1217
1218 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1219 * setting size to 0 may fail for some specific file systems by the permission
1220 * checking which requires WRITE permission but the mode is 000.
1221 * we ignore the resizing(to 0) on the just new created file, since the size is
1222 * 0 after file created.
1223 *
1224 * call this only after vfs_create() is called.
1225 * */
1226 static void
1227 nfsd_check_ignore_resizing(struct iattr *iap)
1228 {
1229 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1230 iap->ia_valid &= ~ATTR_SIZE;
1231 }
1232
1233 /*
1234 * Create a file (regular, directory, device, fifo); UNIX sockets
1235 * not yet implemented.
1236 * If the response fh has been verified, the parent directory should
1237 * already be locked. Note that the parent directory is left locked.
1238 *
1239 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1240 */
1241 __be32
1242 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1243 char *fname, int flen, struct iattr *iap,
1244 int type, dev_t rdev, struct svc_fh *resfhp)
1245 {
1246 struct dentry *dentry, *dchild = NULL;
1247 struct inode *dirp;
1248 __be32 err;
1249 __be32 err2;
1250 int host_err;
1251
1252 err = nfserr_perm;
1253 if (!flen)
1254 goto out;
1255 err = nfserr_exist;
1256 if (isdotent(fname, flen))
1257 goto out;
1258
1259 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1260 if (err)
1261 goto out;
1262
1263 dentry = fhp->fh_dentry;
1264 dirp = dentry->d_inode;
1265
1266 err = nfserr_notdir;
1267 if (!dirp->i_op->lookup)
1268 goto out;
1269 /*
1270 * Check whether the response file handle has been verified yet.
1271 * If it has, the parent directory should already be locked.
1272 */
1273 if (!resfhp->fh_dentry) {
1274 host_err = fh_want_write(fhp);
1275 if (host_err)
1276 goto out_nfserr;
1277
1278 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1279 fh_lock_nested(fhp, I_MUTEX_PARENT);
1280 dchild = lookup_one_len(fname, dentry, flen);
1281 host_err = PTR_ERR(dchild);
1282 if (IS_ERR(dchild))
1283 goto out_nfserr;
1284 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1285 if (err)
1286 goto out;
1287 } else {
1288 /* called from nfsd_proc_create */
1289 dchild = dget(resfhp->fh_dentry);
1290 if (!fhp->fh_locked) {
1291 /* not actually possible */
1292 printk(KERN_ERR
1293 "nfsd_create: parent %s/%s not locked!\n",
1294 dentry->d_parent->d_name.name,
1295 dentry->d_name.name);
1296 err = nfserr_io;
1297 goto out;
1298 }
1299 }
1300 /*
1301 * Make sure the child dentry is still negative ...
1302 */
1303 err = nfserr_exist;
1304 if (dchild->d_inode) {
1305 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1306 dentry->d_name.name, dchild->d_name.name);
1307 goto out;
1308 }
1309
1310 if (!(iap->ia_valid & ATTR_MODE))
1311 iap->ia_mode = 0;
1312 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1313
1314 err = nfserr_inval;
1315 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1316 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1317 type);
1318 goto out;
1319 }
1320
1321 /*
1322 * Get the dir op function pointer.
1323 */
1324 err = 0;
1325 host_err = 0;
1326 switch (type) {
1327 case S_IFREG:
1328 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1329 if (!host_err)
1330 nfsd_check_ignore_resizing(iap);
1331 break;
1332 case S_IFDIR:
1333 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1334 break;
1335 case S_IFCHR:
1336 case S_IFBLK:
1337 case S_IFIFO:
1338 case S_IFSOCK:
1339 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1340 break;
1341 }
1342 if (host_err < 0)
1343 goto out_nfserr;
1344
1345 err = nfsd_create_setattr(rqstp, resfhp, iap);
1346
1347 /*
1348 * nfsd_setattr already committed the child. Transactional filesystems
1349 * had a chance to commit changes for both parent and child
1350 * simultaneously making the following commit_metadata a noop.
1351 */
1352 err2 = nfserrno(commit_metadata(fhp));
1353 if (err2)
1354 err = err2;
1355 /*
1356 * Update the file handle to get the new inode info.
1357 */
1358 if (!err)
1359 err = fh_update(resfhp);
1360 out:
1361 if (dchild && !IS_ERR(dchild))
1362 dput(dchild);
1363 return err;
1364
1365 out_nfserr:
1366 err = nfserrno(host_err);
1367 goto out;
1368 }
1369
1370 #ifdef CONFIG_NFSD_V3
1371
1372 static inline int nfsd_create_is_exclusive(int createmode)
1373 {
1374 return createmode == NFS3_CREATE_EXCLUSIVE
1375 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1376 }
1377
1378 /*
1379 * NFSv3 and NFSv4 version of nfsd_create
1380 */
1381 __be32
1382 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1383 char *fname, int flen, struct iattr *iap,
1384 struct svc_fh *resfhp, int createmode, u32 *verifier,
1385 bool *truncp, bool *created)
1386 {
1387 struct dentry *dentry, *dchild = NULL;
1388 struct inode *dirp;
1389 __be32 err;
1390 int host_err;
1391 __u32 v_mtime=0, v_atime=0;
1392
1393 err = nfserr_perm;
1394 if (!flen)
1395 goto out;
1396 err = nfserr_exist;
1397 if (isdotent(fname, flen))
1398 goto out;
1399 if (!(iap->ia_valid & ATTR_MODE))
1400 iap->ia_mode = 0;
1401 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1402 if (err)
1403 goto out;
1404
1405 dentry = fhp->fh_dentry;
1406 dirp = dentry->d_inode;
1407
1408 /* Get all the sanity checks out of the way before
1409 * we lock the parent. */
1410 err = nfserr_notdir;
1411 if (!dirp->i_op->lookup)
1412 goto out;
1413
1414 host_err = fh_want_write(fhp);
1415 if (host_err)
1416 goto out_nfserr;
1417
1418 fh_lock_nested(fhp, I_MUTEX_PARENT);
1419
1420 /*
1421 * Compose the response file handle.
1422 */
1423 dchild = lookup_one_len(fname, dentry, flen);
1424 host_err = PTR_ERR(dchild);
1425 if (IS_ERR(dchild))
1426 goto out_nfserr;
1427
1428 /* If file doesn't exist, check for permissions to create one */
1429 if (!dchild->d_inode) {
1430 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1431 if (err)
1432 goto out;
1433 }
1434
1435 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1436 if (err)
1437 goto out;
1438
1439 if (nfsd_create_is_exclusive(createmode)) {
1440 /* solaris7 gets confused (bugid 4218508) if these have
1441 * the high bit set, so just clear the high bits. If this is
1442 * ever changed to use different attrs for storing the
1443 * verifier, then do_open_lookup() will also need to be fixed
1444 * accordingly.
1445 */
1446 v_mtime = verifier[0]&0x7fffffff;
1447 v_atime = verifier[1]&0x7fffffff;
1448 }
1449
1450 if (dchild->d_inode) {
1451 err = 0;
1452
1453 switch (createmode) {
1454 case NFS3_CREATE_UNCHECKED:
1455 if (! S_ISREG(dchild->d_inode->i_mode))
1456 goto out;
1457 else if (truncp) {
1458 /* in nfsv4, we need to treat this case a little
1459 * differently. we don't want to truncate the
1460 * file now; this would be wrong if the OPEN
1461 * fails for some other reason. furthermore,
1462 * if the size is nonzero, we should ignore it
1463 * according to spec!
1464 */
1465 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1466 }
1467 else {
1468 iap->ia_valid &= ATTR_SIZE;
1469 goto set_attr;
1470 }
1471 break;
1472 case NFS3_CREATE_EXCLUSIVE:
1473 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1474 && dchild->d_inode->i_atime.tv_sec == v_atime
1475 && dchild->d_inode->i_size == 0 ) {
1476 if (created)
1477 *created = 1;
1478 break;
1479 }
1480 case NFS4_CREATE_EXCLUSIVE4_1:
1481 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1482 && dchild->d_inode->i_atime.tv_sec == v_atime
1483 && dchild->d_inode->i_size == 0 ) {
1484 if (created)
1485 *created = 1;
1486 goto set_attr;
1487 }
1488 /* fallthru */
1489 case NFS3_CREATE_GUARDED:
1490 err = nfserr_exist;
1491 }
1492 fh_drop_write(fhp);
1493 goto out;
1494 }
1495
1496 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1497 if (host_err < 0) {
1498 fh_drop_write(fhp);
1499 goto out_nfserr;
1500 }
1501 if (created)
1502 *created = 1;
1503
1504 nfsd_check_ignore_resizing(iap);
1505
1506 if (nfsd_create_is_exclusive(createmode)) {
1507 /* Cram the verifier into atime/mtime */
1508 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1509 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1510 /* XXX someone who knows this better please fix it for nsec */
1511 iap->ia_mtime.tv_sec = v_mtime;
1512 iap->ia_atime.tv_sec = v_atime;
1513 iap->ia_mtime.tv_nsec = 0;
1514 iap->ia_atime.tv_nsec = 0;
1515 }
1516
1517 set_attr:
1518 err = nfsd_create_setattr(rqstp, resfhp, iap);
1519
1520 /*
1521 * nfsd_setattr already committed the child (and possibly also the parent).
1522 */
1523 if (!err)
1524 err = nfserrno(commit_metadata(fhp));
1525
1526 /*
1527 * Update the filehandle to get the new inode info.
1528 */
1529 if (!err)
1530 err = fh_update(resfhp);
1531
1532 out:
1533 fh_unlock(fhp);
1534 if (dchild && !IS_ERR(dchild))
1535 dput(dchild);
1536 fh_drop_write(fhp);
1537 return err;
1538
1539 out_nfserr:
1540 err = nfserrno(host_err);
1541 goto out;
1542 }
1543 #endif /* CONFIG_NFSD_V3 */
1544
1545 /*
1546 * Read a symlink. On entry, *lenp must contain the maximum path length that
1547 * fits into the buffer. On return, it contains the true length.
1548 * N.B. After this call fhp needs an fh_put
1549 */
1550 __be32
1551 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1552 {
1553 struct inode *inode;
1554 mm_segment_t oldfs;
1555 __be32 err;
1556 int host_err;
1557 struct path path;
1558
1559 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1560 if (err)
1561 goto out;
1562
1563 path.mnt = fhp->fh_export->ex_path.mnt;
1564 path.dentry = fhp->fh_dentry;
1565 inode = path.dentry->d_inode;
1566
1567 err = nfserr_inval;
1568 if (!inode->i_op->readlink)
1569 goto out;
1570
1571 touch_atime(&path);
1572 /* N.B. Why does this call need a get_fs()??
1573 * Remove the set_fs and watch the fireworks:-) --okir
1574 */
1575
1576 oldfs = get_fs(); set_fs(KERNEL_DS);
1577 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1578 set_fs(oldfs);
1579
1580 if (host_err < 0)
1581 goto out_nfserr;
1582 *lenp = host_err;
1583 err = 0;
1584 out:
1585 return err;
1586
1587 out_nfserr:
1588 err = nfserrno(host_err);
1589 goto out;
1590 }
1591
1592 /*
1593 * Create a symlink and look up its inode
1594 * N.B. After this call _both_ fhp and resfhp need an fh_put
1595 */
1596 __be32
1597 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1598 char *fname, int flen,
1599 char *path, int plen,
1600 struct svc_fh *resfhp,
1601 struct iattr *iap)
1602 {
1603 struct dentry *dentry, *dnew;
1604 __be32 err, cerr;
1605 int host_err;
1606
1607 err = nfserr_noent;
1608 if (!flen || !plen)
1609 goto out;
1610 err = nfserr_exist;
1611 if (isdotent(fname, flen))
1612 goto out;
1613
1614 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1615 if (err)
1616 goto out;
1617
1618 host_err = fh_want_write(fhp);
1619 if (host_err)
1620 goto out_nfserr;
1621
1622 fh_lock(fhp);
1623 dentry = fhp->fh_dentry;
1624 dnew = lookup_one_len(fname, dentry, flen);
1625 host_err = PTR_ERR(dnew);
1626 if (IS_ERR(dnew))
1627 goto out_nfserr;
1628
1629 if (unlikely(path[plen] != 0)) {
1630 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1631 if (path_alloced == NULL)
1632 host_err = -ENOMEM;
1633 else {
1634 strncpy(path_alloced, path, plen);
1635 path_alloced[plen] = 0;
1636 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1637 kfree(path_alloced);
1638 }
1639 } else
1640 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1641 err = nfserrno(host_err);
1642 if (!err)
1643 err = nfserrno(commit_metadata(fhp));
1644 fh_unlock(fhp);
1645
1646 fh_drop_write(fhp);
1647
1648 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1649 dput(dnew);
1650 if (err==0) err = cerr;
1651 out:
1652 return err;
1653
1654 out_nfserr:
1655 err = nfserrno(host_err);
1656 goto out;
1657 }
1658
1659 /*
1660 * Create a hardlink
1661 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1662 */
1663 __be32
1664 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1665 char *name, int len, struct svc_fh *tfhp)
1666 {
1667 struct dentry *ddir, *dnew, *dold;
1668 struct inode *dirp;
1669 __be32 err;
1670 int host_err;
1671
1672 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1673 if (err)
1674 goto out;
1675 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1676 if (err)
1677 goto out;
1678 err = nfserr_isdir;
1679 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1680 goto out;
1681 err = nfserr_perm;
1682 if (!len)
1683 goto out;
1684 err = nfserr_exist;
1685 if (isdotent(name, len))
1686 goto out;
1687
1688 host_err = fh_want_write(tfhp);
1689 if (host_err) {
1690 err = nfserrno(host_err);
1691 goto out;
1692 }
1693
1694 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1695 ddir = ffhp->fh_dentry;
1696 dirp = ddir->d_inode;
1697
1698 dnew = lookup_one_len(name, ddir, len);
1699 host_err = PTR_ERR(dnew);
1700 if (IS_ERR(dnew))
1701 goto out_nfserr;
1702
1703 dold = tfhp->fh_dentry;
1704
1705 err = nfserr_noent;
1706 if (!dold->d_inode)
1707 goto out_dput;
1708 host_err = nfsd_break_lease(dold->d_inode);
1709 if (host_err) {
1710 err = nfserrno(host_err);
1711 goto out_dput;
1712 }
1713 host_err = vfs_link(dold, dirp, dnew);
1714 if (!host_err) {
1715 err = nfserrno(commit_metadata(ffhp));
1716 if (!err)
1717 err = nfserrno(commit_metadata(tfhp));
1718 } else {
1719 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1720 err = nfserr_acces;
1721 else
1722 err = nfserrno(host_err);
1723 }
1724 out_dput:
1725 dput(dnew);
1726 out_unlock:
1727 fh_unlock(ffhp);
1728 fh_drop_write(tfhp);
1729 out:
1730 return err;
1731
1732 out_nfserr:
1733 err = nfserrno(host_err);
1734 goto out_unlock;
1735 }
1736
1737 /*
1738 * Rename a file
1739 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1740 */
1741 __be32
1742 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1743 struct svc_fh *tfhp, char *tname, int tlen)
1744 {
1745 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1746 struct inode *fdir, *tdir;
1747 __be32 err;
1748 int host_err;
1749
1750 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1751 if (err)
1752 goto out;
1753 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1754 if (err)
1755 goto out;
1756
1757 fdentry = ffhp->fh_dentry;
1758 fdir = fdentry->d_inode;
1759
1760 tdentry = tfhp->fh_dentry;
1761 tdir = tdentry->d_inode;
1762
1763 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1764 if (ffhp->fh_export != tfhp->fh_export)
1765 goto out;
1766
1767 err = nfserr_perm;
1768 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1769 goto out;
1770
1771 host_err = fh_want_write(ffhp);
1772 if (host_err) {
1773 err = nfserrno(host_err);
1774 goto out;
1775 }
1776
1777 /* cannot use fh_lock as we need deadlock protective ordering
1778 * so do it by hand */
1779 trap = lock_rename(tdentry, fdentry);
1780 ffhp->fh_locked = tfhp->fh_locked = 1;
1781 fill_pre_wcc(ffhp);
1782 fill_pre_wcc(tfhp);
1783
1784 odentry = lookup_one_len(fname, fdentry, flen);
1785 host_err = PTR_ERR(odentry);
1786 if (IS_ERR(odentry))
1787 goto out_nfserr;
1788
1789 host_err = -ENOENT;
1790 if (!odentry->d_inode)
1791 goto out_dput_old;
1792 host_err = -EINVAL;
1793 if (odentry == trap)
1794 goto out_dput_old;
1795
1796 ndentry = lookup_one_len(tname, tdentry, tlen);
1797 host_err = PTR_ERR(ndentry);
1798 if (IS_ERR(ndentry))
1799 goto out_dput_old;
1800 host_err = -ENOTEMPTY;
1801 if (ndentry == trap)
1802 goto out_dput_new;
1803
1804 host_err = -EXDEV;
1805 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1806 goto out_dput_new;
1807
1808 host_err = nfsd_break_lease(odentry->d_inode);
1809 if (host_err)
1810 goto out_dput_new;
1811 if (ndentry->d_inode) {
1812 host_err = nfsd_break_lease(ndentry->d_inode);
1813 if (host_err)
1814 goto out_dput_new;
1815 }
1816 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1817 if (!host_err) {
1818 host_err = commit_metadata(tfhp);
1819 if (!host_err)
1820 host_err = commit_metadata(ffhp);
1821 }
1822 out_dput_new:
1823 dput(ndentry);
1824 out_dput_old:
1825 dput(odentry);
1826 out_nfserr:
1827 err = nfserrno(host_err);
1828
1829 /* we cannot reply on fh_unlock on the two filehandles,
1830 * as that would do the wrong thing if the two directories
1831 * were the same, so again we do it by hand
1832 */
1833 fill_post_wcc(ffhp);
1834 fill_post_wcc(tfhp);
1835 unlock_rename(tdentry, fdentry);
1836 ffhp->fh_locked = tfhp->fh_locked = 0;
1837 fh_drop_write(ffhp);
1838
1839 out:
1840 return err;
1841 }
1842
1843 /*
1844 * Unlink a file or directory
1845 * N.B. After this call fhp needs an fh_put
1846 */
1847 __be32
1848 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1849 char *fname, int flen)
1850 {
1851 struct dentry *dentry, *rdentry;
1852 struct inode *dirp;
1853 __be32 err;
1854 int host_err;
1855
1856 err = nfserr_acces;
1857 if (!flen || isdotent(fname, flen))
1858 goto out;
1859 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1860 if (err)
1861 goto out;
1862
1863 host_err = fh_want_write(fhp);
1864 if (host_err)
1865 goto out_nfserr;
1866
1867 fh_lock_nested(fhp, I_MUTEX_PARENT);
1868 dentry = fhp->fh_dentry;
1869 dirp = dentry->d_inode;
1870
1871 rdentry = lookup_one_len(fname, dentry, flen);
1872 host_err = PTR_ERR(rdentry);
1873 if (IS_ERR(rdentry))
1874 goto out_nfserr;
1875
1876 if (!rdentry->d_inode) {
1877 dput(rdentry);
1878 err = nfserr_noent;
1879 goto out;
1880 }
1881
1882 if (!type)
1883 type = rdentry->d_inode->i_mode & S_IFMT;
1884
1885 host_err = nfsd_break_lease(rdentry->d_inode);
1886 if (host_err)
1887 goto out_put;
1888 if (type != S_IFDIR)
1889 host_err = vfs_unlink(dirp, rdentry);
1890 else
1891 host_err = vfs_rmdir(dirp, rdentry);
1892 if (!host_err)
1893 host_err = commit_metadata(fhp);
1894 out_put:
1895 dput(rdentry);
1896
1897 out_nfserr:
1898 err = nfserrno(host_err);
1899 out:
1900 return err;
1901 }
1902
1903 /*
1904 * We do this buffering because we must not call back into the file
1905 * system's ->lookup() method from the filldir callback. That may well
1906 * deadlock a number of file systems.
1907 *
1908 * This is based heavily on the implementation of same in XFS.
1909 */
1910 struct buffered_dirent {
1911 u64 ino;
1912 loff_t offset;
1913 int namlen;
1914 unsigned int d_type;
1915 char name[];
1916 };
1917
1918 struct readdir_data {
1919 char *dirent;
1920 size_t used;
1921 int full;
1922 };
1923
1924 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1925 loff_t offset, u64 ino, unsigned int d_type)
1926 {
1927 struct readdir_data *buf = __buf;
1928 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1929 unsigned int reclen;
1930
1931 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1932 if (buf->used + reclen > PAGE_SIZE) {
1933 buf->full = 1;
1934 return -EINVAL;
1935 }
1936
1937 de->namlen = namlen;
1938 de->offset = offset;
1939 de->ino = ino;
1940 de->d_type = d_type;
1941 memcpy(de->name, name, namlen);
1942 buf->used += reclen;
1943
1944 return 0;
1945 }
1946
1947 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1948 struct readdir_cd *cdp, loff_t *offsetp)
1949 {
1950 struct readdir_data buf;
1951 struct buffered_dirent *de;
1952 int host_err;
1953 int size;
1954 loff_t offset;
1955
1956 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1957 if (!buf.dirent)
1958 return nfserrno(-ENOMEM);
1959
1960 offset = *offsetp;
1961
1962 while (1) {
1963 struct inode *dir_inode = file->f_path.dentry->d_inode;
1964 unsigned int reclen;
1965
1966 cdp->err = nfserr_eof; /* will be cleared on successful read */
1967 buf.used = 0;
1968 buf.full = 0;
1969
1970 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1971 if (buf.full)
1972 host_err = 0;
1973
1974 if (host_err < 0)
1975 break;
1976
1977 size = buf.used;
1978
1979 if (!size)
1980 break;
1981
1982 /*
1983 * Various filldir functions may end up calling back into
1984 * lookup_one_len() and the file system's ->lookup() method.
1985 * These expect i_mutex to be held, as it would within readdir.
1986 */
1987 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1988 if (host_err)
1989 break;
1990
1991 de = (struct buffered_dirent *)buf.dirent;
1992 while (size > 0) {
1993 offset = de->offset;
1994
1995 if (func(cdp, de->name, de->namlen, de->offset,
1996 de->ino, de->d_type))
1997 break;
1998
1999 if (cdp->err != nfs_ok)
2000 break;
2001
2002 reclen = ALIGN(sizeof(*de) + de->namlen,
2003 sizeof(u64));
2004 size -= reclen;
2005 de = (struct buffered_dirent *)((char *)de + reclen);
2006 }
2007 mutex_unlock(&dir_inode->i_mutex);
2008 if (size > 0) /* We bailed out early */
2009 break;
2010
2011 offset = vfs_llseek(file, 0, SEEK_CUR);
2012 }
2013
2014 free_page((unsigned long)(buf.dirent));
2015
2016 if (host_err)
2017 return nfserrno(host_err);
2018
2019 *offsetp = offset;
2020 return cdp->err;
2021 }
2022
2023 /*
2024 * Read entries from a directory.
2025 * The NFSv3/4 verifier we ignore for now.
2026 */
2027 __be32
2028 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2029 struct readdir_cd *cdp, filldir_t func)
2030 {
2031 __be32 err;
2032 struct file *file;
2033 loff_t offset = *offsetp;
2034 int may_flags = NFSD_MAY_READ;
2035
2036 /* NFSv2 only supports 32 bit cookies */
2037 if (rqstp->rq_vers > 2)
2038 may_flags |= NFSD_MAY_64BIT_COOKIE;
2039
2040 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2041 if (err)
2042 goto out;
2043
2044 offset = vfs_llseek(file, offset, SEEK_SET);
2045 if (offset < 0) {
2046 err = nfserrno((int)offset);
2047 goto out_close;
2048 }
2049
2050 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2051
2052 if (err == nfserr_eof || err == nfserr_toosmall)
2053 err = nfs_ok; /* can still be found in ->err */
2054 out_close:
2055 nfsd_close(file);
2056 out:
2057 return err;
2058 }
2059
2060 /*
2061 * Get file system stats
2062 * N.B. After this call fhp needs an fh_put
2063 */
2064 __be32
2065 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2066 {
2067 __be32 err;
2068
2069 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2070 if (!err) {
2071 struct path path = {
2072 .mnt = fhp->fh_export->ex_path.mnt,
2073 .dentry = fhp->fh_dentry,
2074 };
2075 if (vfs_statfs(&path, stat))
2076 err = nfserr_io;
2077 }
2078 return err;
2079 }
2080
2081 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2082 {
2083 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2084 }
2085
2086 /*
2087 * Check for a user's access permissions to this inode.
2088 */
2089 __be32
2090 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2091 struct dentry *dentry, int acc)
2092 {
2093 struct inode *inode = dentry->d_inode;
2094 int err;
2095
2096 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2097 return 0;
2098 #if 0
2099 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2100 acc,
2101 (acc & NFSD_MAY_READ)? " read" : "",
2102 (acc & NFSD_MAY_WRITE)? " write" : "",
2103 (acc & NFSD_MAY_EXEC)? " exec" : "",
2104 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2105 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2106 (acc & NFSD_MAY_LOCK)? " lock" : "",
2107 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2108 inode->i_mode,
2109 IS_IMMUTABLE(inode)? " immut" : "",
2110 IS_APPEND(inode)? " append" : "",
2111 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2112 dprintk(" owner %d/%d user %d/%d\n",
2113 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2114 #endif
2115
2116 /* Normally we reject any write/sattr etc access on a read-only file
2117 * system. But if it is IRIX doing check on write-access for a
2118 * device special file, we ignore rofs.
2119 */
2120 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2121 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2122 if (exp_rdonly(rqstp, exp) ||
2123 __mnt_is_readonly(exp->ex_path.mnt))
2124 return nfserr_rofs;
2125 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2126 return nfserr_perm;
2127 }
2128 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2129 return nfserr_perm;
2130
2131 if (acc & NFSD_MAY_LOCK) {
2132 /* If we cannot rely on authentication in NLM requests,
2133 * just allow locks, otherwise require read permission, or
2134 * ownership
2135 */
2136 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2137 return 0;
2138 else
2139 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2140 }
2141 /*
2142 * The file owner always gets access permission for accesses that
2143 * would normally be checked at open time. This is to make
2144 * file access work even when the client has done a fchmod(fd, 0).
2145 *
2146 * However, `cp foo bar' should fail nevertheless when bar is
2147 * readonly. A sensible way to do this might be to reject all
2148 * attempts to truncate a read-only file, because a creat() call
2149 * always implies file truncation.
2150 * ... but this isn't really fair. A process may reasonably call
2151 * ftruncate on an open file descriptor on a file with perm 000.
2152 * We must trust the client to do permission checking - using "ACCESS"
2153 * with NFSv3.
2154 */
2155 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2156 inode->i_uid == current_fsuid())
2157 return 0;
2158
2159 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2160 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2161
2162 /* Allow read access to binaries even when mode 111 */
2163 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2164 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2165 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2166 err = inode_permission(inode, MAY_EXEC);
2167
2168 return err? nfserrno(err) : 0;
2169 }
2170
2171 void
2172 nfsd_racache_shutdown(void)
2173 {
2174 struct raparms *raparm, *last_raparm;
2175 unsigned int i;
2176
2177 dprintk("nfsd: freeing readahead buffers.\n");
2178
2179 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2180 raparm = raparm_hash[i].pb_head;
2181 while(raparm) {
2182 last_raparm = raparm;
2183 raparm = raparm->p_next;
2184 kfree(last_raparm);
2185 }
2186 raparm_hash[i].pb_head = NULL;
2187 }
2188 }
2189 /*
2190 * Initialize readahead param cache
2191 */
2192 int
2193 nfsd_racache_init(int cache_size)
2194 {
2195 int i;
2196 int j = 0;
2197 int nperbucket;
2198 struct raparms **raparm = NULL;
2199
2200
2201 if (raparm_hash[0].pb_head)
2202 return 0;
2203 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2204 if (nperbucket < 2)
2205 nperbucket = 2;
2206 cache_size = nperbucket * RAPARM_HASH_SIZE;
2207
2208 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2209
2210 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2211 spin_lock_init(&raparm_hash[i].pb_lock);
2212
2213 raparm = &raparm_hash[i].pb_head;
2214 for (j = 0; j < nperbucket; j++) {
2215 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2216 if (!*raparm)
2217 goto out_nomem;
2218 raparm = &(*raparm)->p_next;
2219 }
2220 *raparm = NULL;
2221 }
2222
2223 nfsdstats.ra_size = cache_size;
2224 return 0;
2225
2226 out_nomem:
2227 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2228 nfsd_racache_shutdown();
2229 return -ENOMEM;
2230 }
2231
2232 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2233 struct posix_acl *
2234 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2235 {
2236 struct inode *inode = fhp->fh_dentry->d_inode;
2237 char *name;
2238 void *value = NULL;
2239 ssize_t size;
2240 struct posix_acl *acl;
2241
2242 if (!IS_POSIXACL(inode))
2243 return ERR_PTR(-EOPNOTSUPP);
2244
2245 switch (type) {
2246 case ACL_TYPE_ACCESS:
2247 name = POSIX_ACL_XATTR_ACCESS;
2248 break;
2249 case ACL_TYPE_DEFAULT:
2250 name = POSIX_ACL_XATTR_DEFAULT;
2251 break;
2252 default:
2253 return ERR_PTR(-EOPNOTSUPP);
2254 }
2255
2256 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2257 if (size < 0)
2258 return ERR_PTR(size);
2259
2260 acl = posix_acl_from_xattr(&init_user_ns, value, size);
2261 kfree(value);
2262 return acl;
2263 }
2264
2265 int
2266 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2267 {
2268 struct inode *inode = fhp->fh_dentry->d_inode;
2269 char *name;
2270 void *value = NULL;
2271 size_t size;
2272 int error;
2273
2274 if (!IS_POSIXACL(inode) ||
2275 !inode->i_op->setxattr || !inode->i_op->removexattr)
2276 return -EOPNOTSUPP;
2277 switch(type) {
2278 case ACL_TYPE_ACCESS:
2279 name = POSIX_ACL_XATTR_ACCESS;
2280 break;
2281 case ACL_TYPE_DEFAULT:
2282 name = POSIX_ACL_XATTR_DEFAULT;
2283 break;
2284 default:
2285 return -EOPNOTSUPP;
2286 }
2287
2288 if (acl && acl->a_count) {
2289 size = posix_acl_xattr_size(acl->a_count);
2290 value = kmalloc(size, GFP_KERNEL);
2291 if (!value)
2292 return -ENOMEM;
2293 error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2294 if (error < 0)
2295 goto getout;
2296 size = error;
2297 } else
2298 size = 0;
2299
2300 error = fh_want_write(fhp);
2301 if (error)
2302 goto getout;
2303 if (size)
2304 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2305 else {
2306 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2307 error = 0;
2308 else {
2309 error = vfs_removexattr(fhp->fh_dentry, name);
2310 if (error == -ENODATA)
2311 error = 0;
2312 }
2313 }
2314 fh_drop_write(fhp);
2315
2316 getout:
2317 kfree(value);
2318 return error;
2319 }
2320 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
kernel source for telekom puls (alcatel/mediatek)