]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/nfsd/vfs.c
Merge branch 'bdw-fixes' of git://people.freedesktop.org/~danvet/drm-intel into drm...
[mirror_ubuntu-artful-kernel.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 #include <linux/security.h>
32
33 #ifdef CONFIG_NFSD_V3
34 #include "xdr3.h"
35 #endif /* CONFIG_NFSD_V3 */
36
37 #ifdef CONFIG_NFSD_V4
38 #include "acl.h"
39 #include "idmap.h"
40 #endif /* CONFIG_NFSD_V4 */
41
42 #include "nfsd.h"
43 #include "vfs.h"
44
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
46
47
48 /*
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
54 */
55 struct raparms {
56 struct raparms *p_next;
57 unsigned int p_count;
58 ino_t p_ino;
59 dev_t p_dev;
60 int p_set;
61 struct file_ra_state p_ra;
62 unsigned int p_hindex;
63 };
64
65 struct raparm_hbucket {
66 struct raparms *pb_head;
67 spinlock_t pb_lock;
68 } ____cacheline_aligned_in_smp;
69
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
74
75 /*
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * a mount point.
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
80 */
81 int
82 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
83 struct svc_export **expp)
84 {
85 struct svc_export *exp = *expp, *exp2 = NULL;
86 struct dentry *dentry = *dpp;
87 struct path path = {.mnt = mntget(exp->ex_path.mnt),
88 .dentry = dget(dentry)};
89 int err = 0;
90
91 err = follow_down(&path);
92 if (err < 0)
93 goto out;
94
95 exp2 = rqst_exp_get_by_name(rqstp, &path);
96 if (IS_ERR(exp2)) {
97 err = PTR_ERR(exp2);
98 /*
99 * We normally allow NFS clients to continue
100 * "underneath" a mountpoint that is not exported.
101 * The exception is V4ROOT, where no traversal is ever
102 * allowed without an explicit export of the new
103 * directory.
104 */
105 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
106 err = 0;
107 path_put(&path);
108 goto out;
109 }
110 if (nfsd_v4client(rqstp) ||
111 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
112 /* successfully crossed mount point */
113 /*
114 * This is subtle: path.dentry is *not* on path.mnt
115 * at this point. The only reason we are safe is that
116 * original mnt is pinned down by exp, so we should
117 * put path *before* putting exp
118 */
119 *dpp = path.dentry;
120 path.dentry = dentry;
121 *expp = exp2;
122 exp2 = exp;
123 }
124 path_put(&path);
125 exp_put(exp2);
126 out:
127 return err;
128 }
129
130 static void follow_to_parent(struct path *path)
131 {
132 struct dentry *dp;
133
134 while (path->dentry == path->mnt->mnt_root && follow_up(path))
135 ;
136 dp = dget_parent(path->dentry);
137 dput(path->dentry);
138 path->dentry = dp;
139 }
140
141 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
142 {
143 struct svc_export *exp2;
144 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
145 .dentry = dget(dparent)};
146
147 follow_to_parent(&path);
148
149 exp2 = rqst_exp_parent(rqstp, &path);
150 if (PTR_ERR(exp2) == -ENOENT) {
151 *dentryp = dget(dparent);
152 } else if (IS_ERR(exp2)) {
153 path_put(&path);
154 return PTR_ERR(exp2);
155 } else {
156 *dentryp = dget(path.dentry);
157 exp_put(*exp);
158 *exp = exp2;
159 }
160 path_put(&path);
161 return 0;
162 }
163
164 /*
165 * For nfsd purposes, we treat V4ROOT exports as though there was an
166 * export at *every* directory.
167 */
168 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
169 {
170 if (d_mountpoint(dentry))
171 return 1;
172 if (nfsd4_is_junction(dentry))
173 return 1;
174 if (!(exp->ex_flags & NFSEXP_V4ROOT))
175 return 0;
176 return dentry->d_inode != NULL;
177 }
178
179 __be32
180 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
181 const char *name, unsigned int len,
182 struct svc_export **exp_ret, struct dentry **dentry_ret)
183 {
184 struct svc_export *exp;
185 struct dentry *dparent;
186 struct dentry *dentry;
187 int host_err;
188
189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
190
191 dparent = fhp->fh_dentry;
192 exp = fhp->fh_export;
193 exp_get(exp);
194
195 /* Lookup the name, but don't follow links */
196 if (isdotent(name, len)) {
197 if (len==1)
198 dentry = dget(dparent);
199 else if (dparent != exp->ex_path.dentry)
200 dentry = dget_parent(dparent);
201 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
202 dentry = dget(dparent); /* .. == . just like at / */
203 else {
204 /* checking mountpoint crossing is very different when stepping up */
205 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
206 if (host_err)
207 goto out_nfserr;
208 }
209 } else {
210 fh_lock(fhp);
211 dentry = lookup_one_len(name, dparent, len);
212 host_err = PTR_ERR(dentry);
213 if (IS_ERR(dentry))
214 goto out_nfserr;
215 /*
216 * check if we have crossed a mount point ...
217 */
218 if (nfsd_mountpoint(dentry, exp)) {
219 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
220 dput(dentry);
221 goto out_nfserr;
222 }
223 }
224 }
225 *dentry_ret = dentry;
226 *exp_ret = exp;
227 return 0;
228
229 out_nfserr:
230 exp_put(exp);
231 return nfserrno(host_err);
232 }
233
234 /*
235 * Look up one component of a pathname.
236 * N.B. After this call _both_ fhp and resfh need an fh_put
237 *
238 * If the lookup would cross a mountpoint, and the mounted filesystem
239 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
240 * accepted as it stands and the mounted directory is
241 * returned. Otherwise the covered directory is returned.
242 * NOTE: this mountpoint crossing is not supported properly by all
243 * clients and is explicitly disallowed for NFSv3
244 * NeilBrown <neilb@cse.unsw.edu.au>
245 */
246 __be32
247 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
248 unsigned int len, struct svc_fh *resfh)
249 {
250 struct svc_export *exp;
251 struct dentry *dentry;
252 __be32 err;
253
254 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
255 if (err)
256 return err;
257 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
258 if (err)
259 return err;
260 err = check_nfsd_access(exp, rqstp);
261 if (err)
262 goto out;
263 /*
264 * Note: we compose the file handle now, but as the
265 * dentry may be negative, it may need to be updated.
266 */
267 err = fh_compose(resfh, exp, dentry, fhp);
268 if (!err && !dentry->d_inode)
269 err = nfserr_noent;
270 out:
271 dput(dentry);
272 exp_put(exp);
273 return err;
274 }
275
276 static int nfsd_break_lease(struct inode *inode)
277 {
278 if (!S_ISREG(inode->i_mode))
279 return 0;
280 return break_lease(inode, O_WRONLY | O_NONBLOCK);
281 }
282
283 /*
284 * Commit metadata changes to stable storage.
285 */
286 static int
287 commit_metadata(struct svc_fh *fhp)
288 {
289 struct inode *inode = fhp->fh_dentry->d_inode;
290 const struct export_operations *export_ops = inode->i_sb->s_export_op;
291
292 if (!EX_ISSYNC(fhp->fh_export))
293 return 0;
294
295 if (export_ops->commit_metadata)
296 return export_ops->commit_metadata(inode);
297 return sync_inode_metadata(inode, 1);
298 }
299
300 /*
301 * Go over the attributes and take care of the small differences between
302 * NFS semantics and what Linux expects.
303 */
304 static void
305 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
306 {
307 /*
308 * NFSv2 does not differentiate between "set-[ac]time-to-now"
309 * which only requires access, and "set-[ac]time-to-X" which
310 * requires ownership.
311 * So if it looks like it might be "set both to the same time which
312 * is close to now", and if inode_change_ok fails, then we
313 * convert to "set to now" instead of "set to explicit time"
314 *
315 * We only call inode_change_ok as the last test as technically
316 * it is not an interface that we should be using.
317 */
318 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
319 #define MAX_TOUCH_TIME_ERROR (30*60)
320 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
321 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
322 /*
323 * Looks probable.
324 *
325 * Now just make sure time is in the right ballpark.
326 * Solaris, at least, doesn't seem to care what the time
327 * request is. We require it be within 30 minutes of now.
328 */
329 time_t delta = iap->ia_atime.tv_sec - get_seconds();
330 if (delta < 0)
331 delta = -delta;
332 if (delta < MAX_TOUCH_TIME_ERROR &&
333 inode_change_ok(inode, iap) != 0) {
334 /*
335 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
336 * This will cause notify_change to set these times
337 * to "now"
338 */
339 iap->ia_valid &= ~BOTH_TIME_SET;
340 }
341 }
342
343 /* sanitize the mode change */
344 if (iap->ia_valid & ATTR_MODE) {
345 iap->ia_mode &= S_IALLUGO;
346 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
347 }
348
349 /* Revoke setuid/setgid on chown */
350 if (!S_ISDIR(inode->i_mode) &&
351 (((iap->ia_valid & ATTR_UID) && !uid_eq(iap->ia_uid, inode->i_uid)) ||
352 ((iap->ia_valid & ATTR_GID) && !gid_eq(iap->ia_gid, inode->i_gid)))) {
353 iap->ia_valid |= ATTR_KILL_PRIV;
354 if (iap->ia_valid & ATTR_MODE) {
355 /* we're setting mode too, just clear the s*id bits */
356 iap->ia_mode &= ~S_ISUID;
357 if (iap->ia_mode & S_IXGRP)
358 iap->ia_mode &= ~S_ISGID;
359 } else {
360 /* set ATTR_KILL_* bits and let VFS handle it */
361 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
362 }
363 }
364 }
365
366 static __be32
367 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
368 struct iattr *iap)
369 {
370 struct inode *inode = fhp->fh_dentry->d_inode;
371 int host_err;
372
373 if (iap->ia_size < inode->i_size) {
374 __be32 err;
375
376 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
377 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
378 if (err)
379 return err;
380 }
381
382 host_err = get_write_access(inode);
383 if (host_err)
384 goto out_nfserrno;
385
386 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
387 if (host_err)
388 goto out_put_write_access;
389 return 0;
390
391 out_put_write_access:
392 put_write_access(inode);
393 out_nfserrno:
394 return nfserrno(host_err);
395 }
396
397 /*
398 * Set various file attributes. After this call fhp needs an fh_put.
399 */
400 __be32
401 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
402 int check_guard, time_t guardtime)
403 {
404 struct dentry *dentry;
405 struct inode *inode;
406 int accmode = NFSD_MAY_SATTR;
407 umode_t ftype = 0;
408 __be32 err;
409 int host_err;
410 int size_change = 0;
411
412 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
413 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
414 if (iap->ia_valid & ATTR_SIZE)
415 ftype = S_IFREG;
416
417 /* Get inode */
418 err = fh_verify(rqstp, fhp, ftype, accmode);
419 if (err)
420 goto out;
421
422 dentry = fhp->fh_dentry;
423 inode = dentry->d_inode;
424
425 /* Ignore any mode updates on symlinks */
426 if (S_ISLNK(inode->i_mode))
427 iap->ia_valid &= ~ATTR_MODE;
428
429 if (!iap->ia_valid)
430 goto out;
431
432 nfsd_sanitize_attrs(inode, iap);
433
434 /*
435 * The size case is special, it changes the file in addition to the
436 * attributes.
437 */
438 if (iap->ia_valid & ATTR_SIZE) {
439 err = nfsd_get_write_access(rqstp, fhp, iap);
440 if (err)
441 goto out;
442 size_change = 1;
443 }
444
445 iap->ia_valid |= ATTR_CTIME;
446
447 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
448 err = nfserr_notsync;
449 goto out_put_write_access;
450 }
451
452 host_err = nfsd_break_lease(inode);
453 if (host_err)
454 goto out_put_write_access_nfserror;
455
456 fh_lock(fhp);
457 host_err = notify_change(dentry, iap, NULL);
458 fh_unlock(fhp);
459
460 out_put_write_access_nfserror:
461 err = nfserrno(host_err);
462 out_put_write_access:
463 if (size_change)
464 put_write_access(inode);
465 if (!err)
466 commit_metadata(fhp);
467 out:
468 return err;
469 }
470
471 #if defined(CONFIG_NFSD_V2_ACL) || \
472 defined(CONFIG_NFSD_V3_ACL) || \
473 defined(CONFIG_NFSD_V4)
474 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
475 {
476 ssize_t buflen;
477 ssize_t ret;
478
479 buflen = vfs_getxattr(dentry, key, NULL, 0);
480 if (buflen <= 0)
481 return buflen;
482
483 *buf = kmalloc(buflen, GFP_KERNEL);
484 if (!*buf)
485 return -ENOMEM;
486
487 ret = vfs_getxattr(dentry, key, *buf, buflen);
488 if (ret < 0)
489 kfree(*buf);
490 return ret;
491 }
492 #endif
493
494 #if defined(CONFIG_NFSD_V4)
495 static int
496 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
497 {
498 int len;
499 size_t buflen;
500 char *buf = NULL;
501 int error = 0;
502
503 buflen = posix_acl_xattr_size(pacl->a_count);
504 buf = kmalloc(buflen, GFP_KERNEL);
505 error = -ENOMEM;
506 if (buf == NULL)
507 goto out;
508
509 len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen);
510 if (len < 0) {
511 error = len;
512 goto out;
513 }
514
515 error = vfs_setxattr(dentry, key, buf, len, 0);
516 out:
517 kfree(buf);
518 return error;
519 }
520
521 __be32
522 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
523 struct nfs4_acl *acl)
524 {
525 __be32 error;
526 int host_error;
527 struct dentry *dentry;
528 struct inode *inode;
529 struct posix_acl *pacl = NULL, *dpacl = NULL;
530 unsigned int flags = 0;
531
532 /* Get inode */
533 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
534 if (error)
535 return error;
536
537 dentry = fhp->fh_dentry;
538 inode = dentry->d_inode;
539 if (S_ISDIR(inode->i_mode))
540 flags = NFS4_ACL_DIR;
541
542 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
543 if (host_error == -EINVAL) {
544 return nfserr_attrnotsupp;
545 } else if (host_error < 0)
546 goto out_nfserr;
547
548 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
549 if (host_error < 0)
550 goto out_release;
551
552 if (S_ISDIR(inode->i_mode))
553 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
554
555 out_release:
556 posix_acl_release(pacl);
557 posix_acl_release(dpacl);
558 out_nfserr:
559 if (host_error == -EOPNOTSUPP)
560 return nfserr_attrnotsupp;
561 else
562 return nfserrno(host_error);
563 }
564
565 static struct posix_acl *
566 _get_posix_acl(struct dentry *dentry, char *key)
567 {
568 void *buf = NULL;
569 struct posix_acl *pacl = NULL;
570 int buflen;
571
572 buflen = nfsd_getxattr(dentry, key, &buf);
573 if (!buflen)
574 buflen = -ENODATA;
575 if (buflen <= 0)
576 return ERR_PTR(buflen);
577
578 pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen);
579 kfree(buf);
580 return pacl;
581 }
582
583 int
584 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
585 {
586 struct inode *inode = dentry->d_inode;
587 int error = 0;
588 struct posix_acl *pacl = NULL, *dpacl = NULL;
589 unsigned int flags = 0;
590
591 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
592 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
593 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
594 if (IS_ERR(pacl)) {
595 error = PTR_ERR(pacl);
596 pacl = NULL;
597 goto out;
598 }
599
600 if (S_ISDIR(inode->i_mode)) {
601 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
602 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
603 dpacl = NULL;
604 else if (IS_ERR(dpacl)) {
605 error = PTR_ERR(dpacl);
606 dpacl = NULL;
607 goto out;
608 }
609 flags = NFS4_ACL_DIR;
610 }
611
612 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
613 if (IS_ERR(*acl)) {
614 error = PTR_ERR(*acl);
615 *acl = NULL;
616 }
617 out:
618 posix_acl_release(pacl);
619 posix_acl_release(dpacl);
620 return error;
621 }
622
623 /*
624 * NFS junction information is stored in an extended attribute.
625 */
626 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
627
628 /**
629 * nfsd4_is_junction - Test if an object could be an NFS junction
630 *
631 * @dentry: object to test
632 *
633 * Returns 1 if "dentry" appears to contain NFS junction information.
634 * Otherwise 0 is returned.
635 */
636 int nfsd4_is_junction(struct dentry *dentry)
637 {
638 struct inode *inode = dentry->d_inode;
639
640 if (inode == NULL)
641 return 0;
642 if (inode->i_mode & S_IXUGO)
643 return 0;
644 if (!(inode->i_mode & S_ISVTX))
645 return 0;
646 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
647 return 0;
648 return 1;
649 }
650 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
651 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
652 struct xdr_netobj *label)
653 {
654 __be32 error;
655 int host_error;
656 struct dentry *dentry;
657
658 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
659 if (error)
660 return error;
661
662 dentry = fhp->fh_dentry;
663
664 mutex_lock(&dentry->d_inode->i_mutex);
665 host_error = security_inode_setsecctx(dentry, label->data, label->len);
666 mutex_unlock(&dentry->d_inode->i_mutex);
667 return nfserrno(host_error);
668 }
669 #else
670 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
671 struct xdr_netobj *label)
672 {
673 return nfserr_notsupp;
674 }
675 #endif
676
677 #endif /* defined(CONFIG_NFSD_V4) */
678
679 #ifdef CONFIG_NFSD_V3
680 /*
681 * Check server access rights to a file system object
682 */
683 struct accessmap {
684 u32 access;
685 int how;
686 };
687 static struct accessmap nfs3_regaccess[] = {
688 { NFS3_ACCESS_READ, NFSD_MAY_READ },
689 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
690 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
691 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
692
693 { 0, 0 }
694 };
695
696 static struct accessmap nfs3_diraccess[] = {
697 { NFS3_ACCESS_READ, NFSD_MAY_READ },
698 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
699 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
700 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
701 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
702
703 { 0, 0 }
704 };
705
706 static struct accessmap nfs3_anyaccess[] = {
707 /* Some clients - Solaris 2.6 at least, make an access call
708 * to the server to check for access for things like /dev/null
709 * (which really, the server doesn't care about). So
710 * We provide simple access checking for them, looking
711 * mainly at mode bits, and we make sure to ignore read-only
712 * filesystem checks
713 */
714 { NFS3_ACCESS_READ, NFSD_MAY_READ },
715 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
716 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
717 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
718
719 { 0, 0 }
720 };
721
722 __be32
723 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
724 {
725 struct accessmap *map;
726 struct svc_export *export;
727 struct dentry *dentry;
728 u32 query, result = 0, sresult = 0;
729 __be32 error;
730
731 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
732 if (error)
733 goto out;
734
735 export = fhp->fh_export;
736 dentry = fhp->fh_dentry;
737
738 if (S_ISREG(dentry->d_inode->i_mode))
739 map = nfs3_regaccess;
740 else if (S_ISDIR(dentry->d_inode->i_mode))
741 map = nfs3_diraccess;
742 else
743 map = nfs3_anyaccess;
744
745
746 query = *access;
747 for (; map->access; map++) {
748 if (map->access & query) {
749 __be32 err2;
750
751 sresult |= map->access;
752
753 err2 = nfsd_permission(rqstp, export, dentry, map->how);
754 switch (err2) {
755 case nfs_ok:
756 result |= map->access;
757 break;
758
759 /* the following error codes just mean the access was not allowed,
760 * rather than an error occurred */
761 case nfserr_rofs:
762 case nfserr_acces:
763 case nfserr_perm:
764 /* simply don't "or" in the access bit. */
765 break;
766 default:
767 error = err2;
768 goto out;
769 }
770 }
771 }
772 *access = result;
773 if (supported)
774 *supported = sresult;
775
776 out:
777 return error;
778 }
779 #endif /* CONFIG_NFSD_V3 */
780
781 static int nfsd_open_break_lease(struct inode *inode, int access)
782 {
783 unsigned int mode;
784
785 if (access & NFSD_MAY_NOT_BREAK_LEASE)
786 return 0;
787 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
788 return break_lease(inode, mode | O_NONBLOCK);
789 }
790
791 /*
792 * Open an existing file or directory.
793 * The may_flags argument indicates the type of open (read/write/lock)
794 * and additional flags.
795 * N.B. After this call fhp needs an fh_put
796 */
797 __be32
798 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
799 int may_flags, struct file **filp)
800 {
801 struct path path;
802 struct inode *inode;
803 int flags = O_RDONLY|O_LARGEFILE;
804 __be32 err;
805 int host_err = 0;
806
807 validate_process_creds();
808
809 /*
810 * If we get here, then the client has already done an "open",
811 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
812 * in case a chmod has now revoked permission.
813 *
814 * Arguably we should also allow the owner override for
815 * directories, but we never have and it doesn't seem to have
816 * caused anyone a problem. If we were to change this, note
817 * also that our filldir callbacks would need a variant of
818 * lookup_one_len that doesn't check permissions.
819 */
820 if (type == S_IFREG)
821 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
822 err = fh_verify(rqstp, fhp, type, may_flags);
823 if (err)
824 goto out;
825
826 path.mnt = fhp->fh_export->ex_path.mnt;
827 path.dentry = fhp->fh_dentry;
828 inode = path.dentry->d_inode;
829
830 /* Disallow write access to files with the append-only bit set
831 * or any access when mandatory locking enabled
832 */
833 err = nfserr_perm;
834 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
835 goto out;
836 /*
837 * We must ignore files (but only files) which might have mandatory
838 * locks on them because there is no way to know if the accesser has
839 * the lock.
840 */
841 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
842 goto out;
843
844 if (!inode->i_fop)
845 goto out;
846
847 host_err = nfsd_open_break_lease(inode, may_flags);
848 if (host_err) /* NOMEM or WOULDBLOCK */
849 goto out_nfserr;
850
851 if (may_flags & NFSD_MAY_WRITE) {
852 if (may_flags & NFSD_MAY_READ)
853 flags = O_RDWR|O_LARGEFILE;
854 else
855 flags = O_WRONLY|O_LARGEFILE;
856 }
857 *filp = dentry_open(&path, flags, current_cred());
858 if (IS_ERR(*filp)) {
859 host_err = PTR_ERR(*filp);
860 *filp = NULL;
861 } else {
862 host_err = ima_file_check(*filp, may_flags);
863
864 if (may_flags & NFSD_MAY_64BIT_COOKIE)
865 (*filp)->f_mode |= FMODE_64BITHASH;
866 else
867 (*filp)->f_mode |= FMODE_32BITHASH;
868 }
869
870 out_nfserr:
871 err = nfserrno(host_err);
872 out:
873 validate_process_creds();
874 return err;
875 }
876
877 /*
878 * Close a file.
879 */
880 void
881 nfsd_close(struct file *filp)
882 {
883 fput(filp);
884 }
885
886 /*
887 * Obtain the readahead parameters for the file
888 * specified by (dev, ino).
889 */
890
891 static inline struct raparms *
892 nfsd_get_raparms(dev_t dev, ino_t ino)
893 {
894 struct raparms *ra, **rap, **frap = NULL;
895 int depth = 0;
896 unsigned int hash;
897 struct raparm_hbucket *rab;
898
899 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
900 rab = &raparm_hash[hash];
901
902 spin_lock(&rab->pb_lock);
903 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
904 if (ra->p_ino == ino && ra->p_dev == dev)
905 goto found;
906 depth++;
907 if (ra->p_count == 0)
908 frap = rap;
909 }
910 depth = nfsdstats.ra_size;
911 if (!frap) {
912 spin_unlock(&rab->pb_lock);
913 return NULL;
914 }
915 rap = frap;
916 ra = *frap;
917 ra->p_dev = dev;
918 ra->p_ino = ino;
919 ra->p_set = 0;
920 ra->p_hindex = hash;
921 found:
922 if (rap != &rab->pb_head) {
923 *rap = ra->p_next;
924 ra->p_next = rab->pb_head;
925 rab->pb_head = ra;
926 }
927 ra->p_count++;
928 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
929 spin_unlock(&rab->pb_lock);
930 return ra;
931 }
932
933 /*
934 * Grab and keep cached pages associated with a file in the svc_rqst
935 * so that they can be passed to the network sendmsg/sendpage routines
936 * directly. They will be released after the sending has completed.
937 */
938 static int
939 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
940 struct splice_desc *sd)
941 {
942 struct svc_rqst *rqstp = sd->u.data;
943 struct page **pp = rqstp->rq_next_page;
944 struct page *page = buf->page;
945 size_t size;
946
947 size = sd->len;
948
949 if (rqstp->rq_res.page_len == 0) {
950 get_page(page);
951 put_page(*rqstp->rq_next_page);
952 *(rqstp->rq_next_page++) = page;
953 rqstp->rq_res.page_base = buf->offset;
954 rqstp->rq_res.page_len = size;
955 } else if (page != pp[-1]) {
956 get_page(page);
957 if (*rqstp->rq_next_page)
958 put_page(*rqstp->rq_next_page);
959 *(rqstp->rq_next_page++) = page;
960 rqstp->rq_res.page_len += size;
961 } else
962 rqstp->rq_res.page_len += size;
963
964 return size;
965 }
966
967 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
968 struct splice_desc *sd)
969 {
970 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
971 }
972
973 static __be32
974 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
975 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
976 {
977 mm_segment_t oldfs;
978 __be32 err;
979 int host_err;
980
981 err = nfserr_perm;
982
983 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
984 struct splice_desc sd = {
985 .len = 0,
986 .total_len = *count,
987 .pos = offset,
988 .u.data = rqstp,
989 };
990
991 rqstp->rq_next_page = rqstp->rq_respages + 1;
992 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
993 } else {
994 oldfs = get_fs();
995 set_fs(KERNEL_DS);
996 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
997 set_fs(oldfs);
998 }
999
1000 if (host_err >= 0) {
1001 nfsdstats.io_read += host_err;
1002 *count = host_err;
1003 err = 0;
1004 fsnotify_access(file);
1005 } else
1006 err = nfserrno(host_err);
1007 return err;
1008 }
1009
1010 static void kill_suid(struct dentry *dentry)
1011 {
1012 struct iattr ia;
1013 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
1014
1015 mutex_lock(&dentry->d_inode->i_mutex);
1016 /*
1017 * Note we call this on write, so notify_change will not
1018 * encounter any conflicting delegations:
1019 */
1020 notify_change(dentry, &ia, NULL);
1021 mutex_unlock(&dentry->d_inode->i_mutex);
1022 }
1023
1024 /*
1025 * Gathered writes: If another process is currently writing to the file,
1026 * there's a high chance this is another nfsd (triggered by a bulk write
1027 * from a client's biod). Rather than syncing the file with each write
1028 * request, we sleep for 10 msec.
1029 *
1030 * I don't know if this roughly approximates C. Juszak's idea of
1031 * gathered writes, but it's a nice and simple solution (IMHO), and it
1032 * seems to work:-)
1033 *
1034 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1035 * better tool (separate unstable writes and commits) for solving this
1036 * problem.
1037 */
1038 static int wait_for_concurrent_writes(struct file *file)
1039 {
1040 struct inode *inode = file_inode(file);
1041 static ino_t last_ino;
1042 static dev_t last_dev;
1043 int err = 0;
1044
1045 if (atomic_read(&inode->i_writecount) > 1
1046 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1047 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1048 msleep(10);
1049 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1050 }
1051
1052 if (inode->i_state & I_DIRTY) {
1053 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1054 err = vfs_fsync(file, 0);
1055 }
1056 last_ino = inode->i_ino;
1057 last_dev = inode->i_sb->s_dev;
1058 return err;
1059 }
1060
1061 static __be32
1062 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1063 loff_t offset, struct kvec *vec, int vlen,
1064 unsigned long *cnt, int *stablep)
1065 {
1066 struct svc_export *exp;
1067 struct dentry *dentry;
1068 struct inode *inode;
1069 mm_segment_t oldfs;
1070 __be32 err = 0;
1071 int host_err;
1072 int stable = *stablep;
1073 int use_wgather;
1074 loff_t pos = offset;
1075
1076 dentry = file->f_path.dentry;
1077 inode = dentry->d_inode;
1078 exp = fhp->fh_export;
1079
1080 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1081
1082 if (!EX_ISSYNC(exp))
1083 stable = 0;
1084
1085 /* Write the data. */
1086 oldfs = get_fs(); set_fs(KERNEL_DS);
1087 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
1088 set_fs(oldfs);
1089 if (host_err < 0)
1090 goto out_nfserr;
1091 *cnt = host_err;
1092 nfsdstats.io_write += host_err;
1093 fsnotify_modify(file);
1094
1095 /* clear setuid/setgid flag after write */
1096 if (inode->i_mode & (S_ISUID | S_ISGID))
1097 kill_suid(dentry);
1098
1099 if (stable) {
1100 if (use_wgather)
1101 host_err = wait_for_concurrent_writes(file);
1102 else
1103 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
1104 }
1105
1106 out_nfserr:
1107 dprintk("nfsd: write complete host_err=%d\n", host_err);
1108 if (host_err >= 0)
1109 err = 0;
1110 else
1111 err = nfserrno(host_err);
1112 return err;
1113 }
1114
1115 /*
1116 * Read data from a file. count must contain the requested read count
1117 * on entry. On return, *count contains the number of bytes actually read.
1118 * N.B. After this call fhp needs an fh_put
1119 */
1120 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1121 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1122 {
1123 struct file *file;
1124 struct inode *inode;
1125 struct raparms *ra;
1126 __be32 err;
1127
1128 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1129 if (err)
1130 return err;
1131
1132 inode = file_inode(file);
1133
1134 /* Get readahead parameters */
1135 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1136
1137 if (ra && ra->p_set)
1138 file->f_ra = ra->p_ra;
1139
1140 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1141
1142 /* Write back readahead params */
1143 if (ra) {
1144 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1145 spin_lock(&rab->pb_lock);
1146 ra->p_ra = file->f_ra;
1147 ra->p_set = 1;
1148 ra->p_count--;
1149 spin_unlock(&rab->pb_lock);
1150 }
1151
1152 nfsd_close(file);
1153 return err;
1154 }
1155
1156 /* As above, but use the provided file descriptor. */
1157 __be32
1158 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1159 loff_t offset, struct kvec *vec, int vlen,
1160 unsigned long *count)
1161 {
1162 __be32 err;
1163
1164 if (file) {
1165 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1166 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1167 if (err)
1168 goto out;
1169 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1170 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1171 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1172 out:
1173 return err;
1174 }
1175
1176 /*
1177 * Write data to a file.
1178 * The stable flag requests synchronous writes.
1179 * N.B. After this call fhp needs an fh_put
1180 */
1181 __be32
1182 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1183 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1184 int *stablep)
1185 {
1186 __be32 err = 0;
1187
1188 if (file) {
1189 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1190 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1191 if (err)
1192 goto out;
1193 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1194 stablep);
1195 } else {
1196 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1197 if (err)
1198 goto out;
1199
1200 if (cnt)
1201 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1202 cnt, stablep);
1203 nfsd_close(file);
1204 }
1205 out:
1206 return err;
1207 }
1208
1209 #ifdef CONFIG_NFSD_V3
1210 /*
1211 * Commit all pending writes to stable storage.
1212 *
1213 * Note: we only guarantee that data that lies within the range specified
1214 * by the 'offset' and 'count' parameters will be synced.
1215 *
1216 * Unfortunately we cannot lock the file to make sure we return full WCC
1217 * data to the client, as locking happens lower down in the filesystem.
1218 */
1219 __be32
1220 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1221 loff_t offset, unsigned long count)
1222 {
1223 struct file *file;
1224 loff_t end = LLONG_MAX;
1225 __be32 err = nfserr_inval;
1226
1227 if (offset < 0)
1228 goto out;
1229 if (count != 0) {
1230 end = offset + (loff_t)count - 1;
1231 if (end < offset)
1232 goto out;
1233 }
1234
1235 err = nfsd_open(rqstp, fhp, S_IFREG,
1236 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1237 if (err)
1238 goto out;
1239 if (EX_ISSYNC(fhp->fh_export)) {
1240 int err2 = vfs_fsync_range(file, offset, end, 0);
1241
1242 if (err2 != -EINVAL)
1243 err = nfserrno(err2);
1244 else
1245 err = nfserr_notsupp;
1246 }
1247
1248 nfsd_close(file);
1249 out:
1250 return err;
1251 }
1252 #endif /* CONFIG_NFSD_V3 */
1253
1254 static __be32
1255 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1256 struct iattr *iap)
1257 {
1258 /*
1259 * Mode has already been set earlier in create:
1260 */
1261 iap->ia_valid &= ~ATTR_MODE;
1262 /*
1263 * Setting uid/gid works only for root. Irix appears to
1264 * send along the gid on create when it tries to implement
1265 * setgid directories via NFS:
1266 */
1267 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1268 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1269 if (iap->ia_valid)
1270 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1271 return 0;
1272 }
1273
1274 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1275 * setting size to 0 may fail for some specific file systems by the permission
1276 * checking which requires WRITE permission but the mode is 000.
1277 * we ignore the resizing(to 0) on the just new created file, since the size is
1278 * 0 after file created.
1279 *
1280 * call this only after vfs_create() is called.
1281 * */
1282 static void
1283 nfsd_check_ignore_resizing(struct iattr *iap)
1284 {
1285 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1286 iap->ia_valid &= ~ATTR_SIZE;
1287 }
1288
1289 /*
1290 * Create a file (regular, directory, device, fifo); UNIX sockets
1291 * not yet implemented.
1292 * If the response fh has been verified, the parent directory should
1293 * already be locked. Note that the parent directory is left locked.
1294 *
1295 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1296 */
1297 __be32
1298 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1299 char *fname, int flen, struct iattr *iap,
1300 int type, dev_t rdev, struct svc_fh *resfhp)
1301 {
1302 struct dentry *dentry, *dchild = NULL;
1303 struct inode *dirp;
1304 __be32 err;
1305 __be32 err2;
1306 int host_err;
1307
1308 err = nfserr_perm;
1309 if (!flen)
1310 goto out;
1311 err = nfserr_exist;
1312 if (isdotent(fname, flen))
1313 goto out;
1314
1315 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1316 if (err)
1317 goto out;
1318
1319 dentry = fhp->fh_dentry;
1320 dirp = dentry->d_inode;
1321
1322 err = nfserr_notdir;
1323 if (!dirp->i_op->lookup)
1324 goto out;
1325 /*
1326 * Check whether the response file handle has been verified yet.
1327 * If it has, the parent directory should already be locked.
1328 */
1329 if (!resfhp->fh_dentry) {
1330 host_err = fh_want_write(fhp);
1331 if (host_err)
1332 goto out_nfserr;
1333
1334 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1335 fh_lock_nested(fhp, I_MUTEX_PARENT);
1336 dchild = lookup_one_len(fname, dentry, flen);
1337 host_err = PTR_ERR(dchild);
1338 if (IS_ERR(dchild))
1339 goto out_nfserr;
1340 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1341 if (err)
1342 goto out;
1343 } else {
1344 /* called from nfsd_proc_create */
1345 dchild = dget(resfhp->fh_dentry);
1346 if (!fhp->fh_locked) {
1347 /* not actually possible */
1348 printk(KERN_ERR
1349 "nfsd_create: parent %pd2 not locked!\n",
1350 dentry);
1351 err = nfserr_io;
1352 goto out;
1353 }
1354 }
1355 /*
1356 * Make sure the child dentry is still negative ...
1357 */
1358 err = nfserr_exist;
1359 if (dchild->d_inode) {
1360 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1361 dentry, dchild);
1362 goto out;
1363 }
1364
1365 if (!(iap->ia_valid & ATTR_MODE))
1366 iap->ia_mode = 0;
1367 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1368
1369 err = nfserr_inval;
1370 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1371 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1372 type);
1373 goto out;
1374 }
1375
1376 /*
1377 * Get the dir op function pointer.
1378 */
1379 err = 0;
1380 host_err = 0;
1381 switch (type) {
1382 case S_IFREG:
1383 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1384 if (!host_err)
1385 nfsd_check_ignore_resizing(iap);
1386 break;
1387 case S_IFDIR:
1388 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1389 break;
1390 case S_IFCHR:
1391 case S_IFBLK:
1392 case S_IFIFO:
1393 case S_IFSOCK:
1394 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1395 break;
1396 }
1397 if (host_err < 0)
1398 goto out_nfserr;
1399
1400 err = nfsd_create_setattr(rqstp, resfhp, iap);
1401
1402 /*
1403 * nfsd_setattr already committed the child. Transactional filesystems
1404 * had a chance to commit changes for both parent and child
1405 * simultaneously making the following commit_metadata a noop.
1406 */
1407 err2 = nfserrno(commit_metadata(fhp));
1408 if (err2)
1409 err = err2;
1410 /*
1411 * Update the file handle to get the new inode info.
1412 */
1413 if (!err)
1414 err = fh_update(resfhp);
1415 out:
1416 if (dchild && !IS_ERR(dchild))
1417 dput(dchild);
1418 return err;
1419
1420 out_nfserr:
1421 err = nfserrno(host_err);
1422 goto out;
1423 }
1424
1425 #ifdef CONFIG_NFSD_V3
1426
1427 static inline int nfsd_create_is_exclusive(int createmode)
1428 {
1429 return createmode == NFS3_CREATE_EXCLUSIVE
1430 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1431 }
1432
1433 /*
1434 * NFSv3 and NFSv4 version of nfsd_create
1435 */
1436 __be32
1437 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1438 char *fname, int flen, struct iattr *iap,
1439 struct svc_fh *resfhp, int createmode, u32 *verifier,
1440 bool *truncp, bool *created)
1441 {
1442 struct dentry *dentry, *dchild = NULL;
1443 struct inode *dirp;
1444 __be32 err;
1445 int host_err;
1446 __u32 v_mtime=0, v_atime=0;
1447
1448 err = nfserr_perm;
1449 if (!flen)
1450 goto out;
1451 err = nfserr_exist;
1452 if (isdotent(fname, flen))
1453 goto out;
1454 if (!(iap->ia_valid & ATTR_MODE))
1455 iap->ia_mode = 0;
1456 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1457 if (err)
1458 goto out;
1459
1460 dentry = fhp->fh_dentry;
1461 dirp = dentry->d_inode;
1462
1463 /* Get all the sanity checks out of the way before
1464 * we lock the parent. */
1465 err = nfserr_notdir;
1466 if (!dirp->i_op->lookup)
1467 goto out;
1468
1469 host_err = fh_want_write(fhp);
1470 if (host_err)
1471 goto out_nfserr;
1472
1473 fh_lock_nested(fhp, I_MUTEX_PARENT);
1474
1475 /*
1476 * Compose the response file handle.
1477 */
1478 dchild = lookup_one_len(fname, dentry, flen);
1479 host_err = PTR_ERR(dchild);
1480 if (IS_ERR(dchild))
1481 goto out_nfserr;
1482
1483 /* If file doesn't exist, check for permissions to create one */
1484 if (!dchild->d_inode) {
1485 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1486 if (err)
1487 goto out;
1488 }
1489
1490 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1491 if (err)
1492 goto out;
1493
1494 if (nfsd_create_is_exclusive(createmode)) {
1495 /* solaris7 gets confused (bugid 4218508) if these have
1496 * the high bit set, so just clear the high bits. If this is
1497 * ever changed to use different attrs for storing the
1498 * verifier, then do_open_lookup() will also need to be fixed
1499 * accordingly.
1500 */
1501 v_mtime = verifier[0]&0x7fffffff;
1502 v_atime = verifier[1]&0x7fffffff;
1503 }
1504
1505 if (dchild->d_inode) {
1506 err = 0;
1507
1508 switch (createmode) {
1509 case NFS3_CREATE_UNCHECKED:
1510 if (! S_ISREG(dchild->d_inode->i_mode))
1511 goto out;
1512 else if (truncp) {
1513 /* in nfsv4, we need to treat this case a little
1514 * differently. we don't want to truncate the
1515 * file now; this would be wrong if the OPEN
1516 * fails for some other reason. furthermore,
1517 * if the size is nonzero, we should ignore it
1518 * according to spec!
1519 */
1520 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1521 }
1522 else {
1523 iap->ia_valid &= ATTR_SIZE;
1524 goto set_attr;
1525 }
1526 break;
1527 case NFS3_CREATE_EXCLUSIVE:
1528 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1529 && dchild->d_inode->i_atime.tv_sec == v_atime
1530 && dchild->d_inode->i_size == 0 ) {
1531 if (created)
1532 *created = 1;
1533 break;
1534 }
1535 case NFS4_CREATE_EXCLUSIVE4_1:
1536 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1537 && dchild->d_inode->i_atime.tv_sec == v_atime
1538 && dchild->d_inode->i_size == 0 ) {
1539 if (created)
1540 *created = 1;
1541 goto set_attr;
1542 }
1543 /* fallthru */
1544 case NFS3_CREATE_GUARDED:
1545 err = nfserr_exist;
1546 }
1547 fh_drop_write(fhp);
1548 goto out;
1549 }
1550
1551 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1552 if (host_err < 0) {
1553 fh_drop_write(fhp);
1554 goto out_nfserr;
1555 }
1556 if (created)
1557 *created = 1;
1558
1559 nfsd_check_ignore_resizing(iap);
1560
1561 if (nfsd_create_is_exclusive(createmode)) {
1562 /* Cram the verifier into atime/mtime */
1563 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1564 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1565 /* XXX someone who knows this better please fix it for nsec */
1566 iap->ia_mtime.tv_sec = v_mtime;
1567 iap->ia_atime.tv_sec = v_atime;
1568 iap->ia_mtime.tv_nsec = 0;
1569 iap->ia_atime.tv_nsec = 0;
1570 }
1571
1572 set_attr:
1573 err = nfsd_create_setattr(rqstp, resfhp, iap);
1574
1575 /*
1576 * nfsd_setattr already committed the child (and possibly also the parent).
1577 */
1578 if (!err)
1579 err = nfserrno(commit_metadata(fhp));
1580
1581 /*
1582 * Update the filehandle to get the new inode info.
1583 */
1584 if (!err)
1585 err = fh_update(resfhp);
1586
1587 out:
1588 fh_unlock(fhp);
1589 if (dchild && !IS_ERR(dchild))
1590 dput(dchild);
1591 fh_drop_write(fhp);
1592 return err;
1593
1594 out_nfserr:
1595 err = nfserrno(host_err);
1596 goto out;
1597 }
1598 #endif /* CONFIG_NFSD_V3 */
1599
1600 /*
1601 * Read a symlink. On entry, *lenp must contain the maximum path length that
1602 * fits into the buffer. On return, it contains the true length.
1603 * N.B. After this call fhp needs an fh_put
1604 */
1605 __be32
1606 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1607 {
1608 struct inode *inode;
1609 mm_segment_t oldfs;
1610 __be32 err;
1611 int host_err;
1612 struct path path;
1613
1614 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1615 if (err)
1616 goto out;
1617
1618 path.mnt = fhp->fh_export->ex_path.mnt;
1619 path.dentry = fhp->fh_dentry;
1620 inode = path.dentry->d_inode;
1621
1622 err = nfserr_inval;
1623 if (!inode->i_op->readlink)
1624 goto out;
1625
1626 touch_atime(&path);
1627 /* N.B. Why does this call need a get_fs()??
1628 * Remove the set_fs and watch the fireworks:-) --okir
1629 */
1630
1631 oldfs = get_fs(); set_fs(KERNEL_DS);
1632 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1633 set_fs(oldfs);
1634
1635 if (host_err < 0)
1636 goto out_nfserr;
1637 *lenp = host_err;
1638 err = 0;
1639 out:
1640 return err;
1641
1642 out_nfserr:
1643 err = nfserrno(host_err);
1644 goto out;
1645 }
1646
1647 /*
1648 * Create a symlink and look up its inode
1649 * N.B. After this call _both_ fhp and resfhp need an fh_put
1650 */
1651 __be32
1652 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1653 char *fname, int flen,
1654 char *path, int plen,
1655 struct svc_fh *resfhp,
1656 struct iattr *iap)
1657 {
1658 struct dentry *dentry, *dnew;
1659 __be32 err, cerr;
1660 int host_err;
1661
1662 err = nfserr_noent;
1663 if (!flen || !plen)
1664 goto out;
1665 err = nfserr_exist;
1666 if (isdotent(fname, flen))
1667 goto out;
1668
1669 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1670 if (err)
1671 goto out;
1672
1673 host_err = fh_want_write(fhp);
1674 if (host_err)
1675 goto out_nfserr;
1676
1677 fh_lock(fhp);
1678 dentry = fhp->fh_dentry;
1679 dnew = lookup_one_len(fname, dentry, flen);
1680 host_err = PTR_ERR(dnew);
1681 if (IS_ERR(dnew))
1682 goto out_nfserr;
1683
1684 if (unlikely(path[plen] != 0)) {
1685 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1686 if (path_alloced == NULL)
1687 host_err = -ENOMEM;
1688 else {
1689 strncpy(path_alloced, path, plen);
1690 path_alloced[plen] = 0;
1691 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1692 kfree(path_alloced);
1693 }
1694 } else
1695 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1696 err = nfserrno(host_err);
1697 if (!err)
1698 err = nfserrno(commit_metadata(fhp));
1699 fh_unlock(fhp);
1700
1701 fh_drop_write(fhp);
1702
1703 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1704 dput(dnew);
1705 if (err==0) err = cerr;
1706 out:
1707 return err;
1708
1709 out_nfserr:
1710 err = nfserrno(host_err);
1711 goto out;
1712 }
1713
1714 /*
1715 * Create a hardlink
1716 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1717 */
1718 __be32
1719 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1720 char *name, int len, struct svc_fh *tfhp)
1721 {
1722 struct dentry *ddir, *dnew, *dold;
1723 struct inode *dirp;
1724 __be32 err;
1725 int host_err;
1726
1727 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1728 if (err)
1729 goto out;
1730 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1731 if (err)
1732 goto out;
1733 err = nfserr_isdir;
1734 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1735 goto out;
1736 err = nfserr_perm;
1737 if (!len)
1738 goto out;
1739 err = nfserr_exist;
1740 if (isdotent(name, len))
1741 goto out;
1742
1743 host_err = fh_want_write(tfhp);
1744 if (host_err) {
1745 err = nfserrno(host_err);
1746 goto out;
1747 }
1748
1749 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1750 ddir = ffhp->fh_dentry;
1751 dirp = ddir->d_inode;
1752
1753 dnew = lookup_one_len(name, ddir, len);
1754 host_err = PTR_ERR(dnew);
1755 if (IS_ERR(dnew))
1756 goto out_nfserr;
1757
1758 dold = tfhp->fh_dentry;
1759
1760 err = nfserr_noent;
1761 if (!dold->d_inode)
1762 goto out_dput;
1763 host_err = nfsd_break_lease(dold->d_inode);
1764 if (host_err) {
1765 err = nfserrno(host_err);
1766 goto out_dput;
1767 }
1768 host_err = vfs_link(dold, dirp, dnew, NULL);
1769 if (!host_err) {
1770 err = nfserrno(commit_metadata(ffhp));
1771 if (!err)
1772 err = nfserrno(commit_metadata(tfhp));
1773 } else {
1774 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1775 err = nfserr_acces;
1776 else
1777 err = nfserrno(host_err);
1778 }
1779 out_dput:
1780 dput(dnew);
1781 out_unlock:
1782 fh_unlock(ffhp);
1783 fh_drop_write(tfhp);
1784 out:
1785 return err;
1786
1787 out_nfserr:
1788 err = nfserrno(host_err);
1789 goto out_unlock;
1790 }
1791
1792 /*
1793 * Rename a file
1794 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1795 */
1796 __be32
1797 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1798 struct svc_fh *tfhp, char *tname, int tlen)
1799 {
1800 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1801 struct inode *fdir, *tdir;
1802 __be32 err;
1803 int host_err;
1804
1805 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1806 if (err)
1807 goto out;
1808 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1809 if (err)
1810 goto out;
1811
1812 fdentry = ffhp->fh_dentry;
1813 fdir = fdentry->d_inode;
1814
1815 tdentry = tfhp->fh_dentry;
1816 tdir = tdentry->d_inode;
1817
1818 err = nfserr_perm;
1819 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1820 goto out;
1821
1822 host_err = fh_want_write(ffhp);
1823 if (host_err) {
1824 err = nfserrno(host_err);
1825 goto out;
1826 }
1827
1828 /* cannot use fh_lock as we need deadlock protective ordering
1829 * so do it by hand */
1830 trap = lock_rename(tdentry, fdentry);
1831 ffhp->fh_locked = tfhp->fh_locked = 1;
1832 fill_pre_wcc(ffhp);
1833 fill_pre_wcc(tfhp);
1834
1835 odentry = lookup_one_len(fname, fdentry, flen);
1836 host_err = PTR_ERR(odentry);
1837 if (IS_ERR(odentry))
1838 goto out_nfserr;
1839
1840 host_err = -ENOENT;
1841 if (!odentry->d_inode)
1842 goto out_dput_old;
1843 host_err = -EINVAL;
1844 if (odentry == trap)
1845 goto out_dput_old;
1846
1847 ndentry = lookup_one_len(tname, tdentry, tlen);
1848 host_err = PTR_ERR(ndentry);
1849 if (IS_ERR(ndentry))
1850 goto out_dput_old;
1851 host_err = -ENOTEMPTY;
1852 if (ndentry == trap)
1853 goto out_dput_new;
1854
1855 host_err = -EXDEV;
1856 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1857 goto out_dput_new;
1858 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1859 goto out_dput_new;
1860
1861 host_err = nfsd_break_lease(odentry->d_inode);
1862 if (host_err)
1863 goto out_dput_new;
1864 if (ndentry->d_inode) {
1865 host_err = nfsd_break_lease(ndentry->d_inode);
1866 if (host_err)
1867 goto out_dput_new;
1868 }
1869 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL);
1870 if (!host_err) {
1871 host_err = commit_metadata(tfhp);
1872 if (!host_err)
1873 host_err = commit_metadata(ffhp);
1874 }
1875 out_dput_new:
1876 dput(ndentry);
1877 out_dput_old:
1878 dput(odentry);
1879 out_nfserr:
1880 err = nfserrno(host_err);
1881
1882 /* we cannot reply on fh_unlock on the two filehandles,
1883 * as that would do the wrong thing if the two directories
1884 * were the same, so again we do it by hand
1885 */
1886 fill_post_wcc(ffhp);
1887 fill_post_wcc(tfhp);
1888 unlock_rename(tdentry, fdentry);
1889 ffhp->fh_locked = tfhp->fh_locked = 0;
1890 fh_drop_write(ffhp);
1891
1892 out:
1893 return err;
1894 }
1895
1896 /*
1897 * Unlink a file or directory
1898 * N.B. After this call fhp needs an fh_put
1899 */
1900 __be32
1901 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1902 char *fname, int flen)
1903 {
1904 struct dentry *dentry, *rdentry;
1905 struct inode *dirp;
1906 __be32 err;
1907 int host_err;
1908
1909 err = nfserr_acces;
1910 if (!flen || isdotent(fname, flen))
1911 goto out;
1912 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1913 if (err)
1914 goto out;
1915
1916 host_err = fh_want_write(fhp);
1917 if (host_err)
1918 goto out_nfserr;
1919
1920 fh_lock_nested(fhp, I_MUTEX_PARENT);
1921 dentry = fhp->fh_dentry;
1922 dirp = dentry->d_inode;
1923
1924 rdentry = lookup_one_len(fname, dentry, flen);
1925 host_err = PTR_ERR(rdentry);
1926 if (IS_ERR(rdentry))
1927 goto out_nfserr;
1928
1929 if (!rdentry->d_inode) {
1930 dput(rdentry);
1931 err = nfserr_noent;
1932 goto out;
1933 }
1934
1935 if (!type)
1936 type = rdentry->d_inode->i_mode & S_IFMT;
1937
1938 host_err = nfsd_break_lease(rdentry->d_inode);
1939 if (host_err)
1940 goto out_put;
1941 if (type != S_IFDIR)
1942 host_err = vfs_unlink(dirp, rdentry, NULL);
1943 else
1944 host_err = vfs_rmdir(dirp, rdentry);
1945 if (!host_err)
1946 host_err = commit_metadata(fhp);
1947 out_put:
1948 dput(rdentry);
1949
1950 out_nfserr:
1951 err = nfserrno(host_err);
1952 out:
1953 return err;
1954 }
1955
1956 /*
1957 * We do this buffering because we must not call back into the file
1958 * system's ->lookup() method from the filldir callback. That may well
1959 * deadlock a number of file systems.
1960 *
1961 * This is based heavily on the implementation of same in XFS.
1962 */
1963 struct buffered_dirent {
1964 u64 ino;
1965 loff_t offset;
1966 int namlen;
1967 unsigned int d_type;
1968 char name[];
1969 };
1970
1971 struct readdir_data {
1972 struct dir_context ctx;
1973 char *dirent;
1974 size_t used;
1975 int full;
1976 };
1977
1978 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1979 loff_t offset, u64 ino, unsigned int d_type)
1980 {
1981 struct readdir_data *buf = __buf;
1982 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1983 unsigned int reclen;
1984
1985 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1986 if (buf->used + reclen > PAGE_SIZE) {
1987 buf->full = 1;
1988 return -EINVAL;
1989 }
1990
1991 de->namlen = namlen;
1992 de->offset = offset;
1993 de->ino = ino;
1994 de->d_type = d_type;
1995 memcpy(de->name, name, namlen);
1996 buf->used += reclen;
1997
1998 return 0;
1999 }
2000
2001 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
2002 struct readdir_cd *cdp, loff_t *offsetp)
2003 {
2004 struct buffered_dirent *de;
2005 int host_err;
2006 int size;
2007 loff_t offset;
2008 struct readdir_data buf = {
2009 .ctx.actor = nfsd_buffered_filldir,
2010 .dirent = (void *)__get_free_page(GFP_KERNEL)
2011 };
2012
2013 if (!buf.dirent)
2014 return nfserrno(-ENOMEM);
2015
2016 offset = *offsetp;
2017
2018 while (1) {
2019 struct inode *dir_inode = file_inode(file);
2020 unsigned int reclen;
2021
2022 cdp->err = nfserr_eof; /* will be cleared on successful read */
2023 buf.used = 0;
2024 buf.full = 0;
2025
2026 host_err = iterate_dir(file, &buf.ctx);
2027 if (buf.full)
2028 host_err = 0;
2029
2030 if (host_err < 0)
2031 break;
2032
2033 size = buf.used;
2034
2035 if (!size)
2036 break;
2037
2038 /*
2039 * Various filldir functions may end up calling back into
2040 * lookup_one_len() and the file system's ->lookup() method.
2041 * These expect i_mutex to be held, as it would within readdir.
2042 */
2043 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2044 if (host_err)
2045 break;
2046
2047 de = (struct buffered_dirent *)buf.dirent;
2048 while (size > 0) {
2049 offset = de->offset;
2050
2051 if (func(cdp, de->name, de->namlen, de->offset,
2052 de->ino, de->d_type))
2053 break;
2054
2055 if (cdp->err != nfs_ok)
2056 break;
2057
2058 reclen = ALIGN(sizeof(*de) + de->namlen,
2059 sizeof(u64));
2060 size -= reclen;
2061 de = (struct buffered_dirent *)((char *)de + reclen);
2062 }
2063 mutex_unlock(&dir_inode->i_mutex);
2064 if (size > 0) /* We bailed out early */
2065 break;
2066
2067 offset = vfs_llseek(file, 0, SEEK_CUR);
2068 }
2069
2070 free_page((unsigned long)(buf.dirent));
2071
2072 if (host_err)
2073 return nfserrno(host_err);
2074
2075 *offsetp = offset;
2076 return cdp->err;
2077 }
2078
2079 /*
2080 * Read entries from a directory.
2081 * The NFSv3/4 verifier we ignore for now.
2082 */
2083 __be32
2084 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2085 struct readdir_cd *cdp, filldir_t func)
2086 {
2087 __be32 err;
2088 struct file *file;
2089 loff_t offset = *offsetp;
2090 int may_flags = NFSD_MAY_READ;
2091
2092 /* NFSv2 only supports 32 bit cookies */
2093 if (rqstp->rq_vers > 2)
2094 may_flags |= NFSD_MAY_64BIT_COOKIE;
2095
2096 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2097 if (err)
2098 goto out;
2099
2100 offset = vfs_llseek(file, offset, SEEK_SET);
2101 if (offset < 0) {
2102 err = nfserrno((int)offset);
2103 goto out_close;
2104 }
2105
2106 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2107
2108 if (err == nfserr_eof || err == nfserr_toosmall)
2109 err = nfs_ok; /* can still be found in ->err */
2110 out_close:
2111 nfsd_close(file);
2112 out:
2113 return err;
2114 }
2115
2116 /*
2117 * Get file system stats
2118 * N.B. After this call fhp needs an fh_put
2119 */
2120 __be32
2121 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2122 {
2123 __be32 err;
2124
2125 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2126 if (!err) {
2127 struct path path = {
2128 .mnt = fhp->fh_export->ex_path.mnt,
2129 .dentry = fhp->fh_dentry,
2130 };
2131 if (vfs_statfs(&path, stat))
2132 err = nfserr_io;
2133 }
2134 return err;
2135 }
2136
2137 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2138 {
2139 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2140 }
2141
2142 /*
2143 * Check for a user's access permissions to this inode.
2144 */
2145 __be32
2146 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2147 struct dentry *dentry, int acc)
2148 {
2149 struct inode *inode = dentry->d_inode;
2150 int err;
2151
2152 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2153 return 0;
2154 #if 0
2155 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2156 acc,
2157 (acc & NFSD_MAY_READ)? " read" : "",
2158 (acc & NFSD_MAY_WRITE)? " write" : "",
2159 (acc & NFSD_MAY_EXEC)? " exec" : "",
2160 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2161 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2162 (acc & NFSD_MAY_LOCK)? " lock" : "",
2163 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2164 inode->i_mode,
2165 IS_IMMUTABLE(inode)? " immut" : "",
2166 IS_APPEND(inode)? " append" : "",
2167 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2168 dprintk(" owner %d/%d user %d/%d\n",
2169 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2170 #endif
2171
2172 /* Normally we reject any write/sattr etc access on a read-only file
2173 * system. But if it is IRIX doing check on write-access for a
2174 * device special file, we ignore rofs.
2175 */
2176 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2177 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2178 if (exp_rdonly(rqstp, exp) ||
2179 __mnt_is_readonly(exp->ex_path.mnt))
2180 return nfserr_rofs;
2181 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2182 return nfserr_perm;
2183 }
2184 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2185 return nfserr_perm;
2186
2187 if (acc & NFSD_MAY_LOCK) {
2188 /* If we cannot rely on authentication in NLM requests,
2189 * just allow locks, otherwise require read permission, or
2190 * ownership
2191 */
2192 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2193 return 0;
2194 else
2195 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2196 }
2197 /*
2198 * The file owner always gets access permission for accesses that
2199 * would normally be checked at open time. This is to make
2200 * file access work even when the client has done a fchmod(fd, 0).
2201 *
2202 * However, `cp foo bar' should fail nevertheless when bar is
2203 * readonly. A sensible way to do this might be to reject all
2204 * attempts to truncate a read-only file, because a creat() call
2205 * always implies file truncation.
2206 * ... but this isn't really fair. A process may reasonably call
2207 * ftruncate on an open file descriptor on a file with perm 000.
2208 * We must trust the client to do permission checking - using "ACCESS"
2209 * with NFSv3.
2210 */
2211 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2212 uid_eq(inode->i_uid, current_fsuid()))
2213 return 0;
2214
2215 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2216 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2217
2218 /* Allow read access to binaries even when mode 111 */
2219 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2220 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2221 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2222 err = inode_permission(inode, MAY_EXEC);
2223
2224 return err? nfserrno(err) : 0;
2225 }
2226
2227 void
2228 nfsd_racache_shutdown(void)
2229 {
2230 struct raparms *raparm, *last_raparm;
2231 unsigned int i;
2232
2233 dprintk("nfsd: freeing readahead buffers.\n");
2234
2235 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2236 raparm = raparm_hash[i].pb_head;
2237 while(raparm) {
2238 last_raparm = raparm;
2239 raparm = raparm->p_next;
2240 kfree(last_raparm);
2241 }
2242 raparm_hash[i].pb_head = NULL;
2243 }
2244 }
2245 /*
2246 * Initialize readahead param cache
2247 */
2248 int
2249 nfsd_racache_init(int cache_size)
2250 {
2251 int i;
2252 int j = 0;
2253 int nperbucket;
2254 struct raparms **raparm = NULL;
2255
2256
2257 if (raparm_hash[0].pb_head)
2258 return 0;
2259 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2260 if (nperbucket < 2)
2261 nperbucket = 2;
2262 cache_size = nperbucket * RAPARM_HASH_SIZE;
2263
2264 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2265
2266 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2267 spin_lock_init(&raparm_hash[i].pb_lock);
2268
2269 raparm = &raparm_hash[i].pb_head;
2270 for (j = 0; j < nperbucket; j++) {
2271 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2272 if (!*raparm)
2273 goto out_nomem;
2274 raparm = &(*raparm)->p_next;
2275 }
2276 *raparm = NULL;
2277 }
2278
2279 nfsdstats.ra_size = cache_size;
2280 return 0;
2281
2282 out_nomem:
2283 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2284 nfsd_racache_shutdown();
2285 return -ENOMEM;
2286 }
2287
2288 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2289 struct posix_acl *
2290 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2291 {
2292 struct inode *inode = fhp->fh_dentry->d_inode;
2293 char *name;
2294 void *value = NULL;
2295 ssize_t size;
2296 struct posix_acl *acl;
2297
2298 if (!IS_POSIXACL(inode))
2299 return ERR_PTR(-EOPNOTSUPP);
2300
2301 switch (type) {
2302 case ACL_TYPE_ACCESS:
2303 name = POSIX_ACL_XATTR_ACCESS;
2304 break;
2305 case ACL_TYPE_DEFAULT:
2306 name = POSIX_ACL_XATTR_DEFAULT;
2307 break;
2308 default:
2309 return ERR_PTR(-EOPNOTSUPP);
2310 }
2311
2312 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2313 if (size < 0)
2314 return ERR_PTR(size);
2315
2316 acl = posix_acl_from_xattr(&init_user_ns, value, size);
2317 kfree(value);
2318 return acl;
2319 }
2320
2321 int
2322 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2323 {
2324 struct inode *inode = fhp->fh_dentry->d_inode;
2325 char *name;
2326 void *value = NULL;
2327 size_t size;
2328 int error;
2329
2330 if (!IS_POSIXACL(inode) ||
2331 !inode->i_op->setxattr || !inode->i_op->removexattr)
2332 return -EOPNOTSUPP;
2333 switch(type) {
2334 case ACL_TYPE_ACCESS:
2335 name = POSIX_ACL_XATTR_ACCESS;
2336 break;
2337 case ACL_TYPE_DEFAULT:
2338 name = POSIX_ACL_XATTR_DEFAULT;
2339 break;
2340 default:
2341 return -EOPNOTSUPP;
2342 }
2343
2344 if (acl && acl->a_count) {
2345 size = posix_acl_xattr_size(acl->a_count);
2346 value = kmalloc(size, GFP_KERNEL);
2347 if (!value)
2348 return -ENOMEM;
2349 error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2350 if (error < 0)
2351 goto getout;
2352 size = error;
2353 } else
2354 size = 0;
2355
2356 error = fh_want_write(fhp);
2357 if (error)
2358 goto getout;
2359 if (size)
2360 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2361 else {
2362 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2363 error = 0;
2364 else {
2365 error = vfs_removexattr(fhp->fh_dentry, name);
2366 if (error == -ENODATA)
2367 error = 0;
2368 }
2369 }
2370 fh_drop_write(fhp);
2371
2372 getout:
2373 kfree(value);
2374 return error;
2375 }
2376 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */