]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/nfsd/vfs.c
Merge remote-tracking branches 'asoc/topic/sgtl5000', 'asoc/topic/simple', 'asoc...
[mirror_ubuntu-bionic-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/falloc.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <linux/slab.h>
29 #include <asm/uaccess.h>
30 #include <linux/exportfs.h>
31 #include <linux/writeback.h>
32 #include <linux/security.h>
33
34 #ifdef CONFIG_NFSD_V3
35 #include "xdr3.h"
36 #endif /* CONFIG_NFSD_V3 */
37
38 #ifdef CONFIG_NFSD_V4
39 #include "../internal.h"
40 #include "acl.h"
41 #include "idmap.h"
42 #endif /* CONFIG_NFSD_V4 */
43
44 #include "nfsd.h"
45 #include "vfs.h"
46 #include "trace.h"
47
48 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49
50
51 /*
52 * This is a cache of readahead params that help us choose the proper
53 * readahead strategy. Initially, we set all readahead parameters to 0
54 * and let the VFS handle things.
55 * If you increase the number of cached files very much, you'll need to
56 * add a hash table here.
57 */
58 struct raparms {
59 struct raparms *p_next;
60 unsigned int p_count;
61 ino_t p_ino;
62 dev_t p_dev;
63 int p_set;
64 struct file_ra_state p_ra;
65 unsigned int p_hindex;
66 };
67
68 struct raparm_hbucket {
69 struct raparms *pb_head;
70 spinlock_t pb_lock;
71 } ____cacheline_aligned_in_smp;
72
73 #define RAPARM_HASH_BITS 4
74 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
75 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
76 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
77
78 /*
79 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
80 * a mount point.
81 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
82 * or nfs_ok having possibly changed *dpp and *expp
83 */
84 int
85 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
86 struct svc_export **expp)
87 {
88 struct svc_export *exp = *expp, *exp2 = NULL;
89 struct dentry *dentry = *dpp;
90 struct path path = {.mnt = mntget(exp->ex_path.mnt),
91 .dentry = dget(dentry)};
92 int err = 0;
93
94 err = follow_down(&path);
95 if (err < 0)
96 goto out;
97
98 exp2 = rqst_exp_get_by_name(rqstp, &path);
99 if (IS_ERR(exp2)) {
100 err = PTR_ERR(exp2);
101 /*
102 * We normally allow NFS clients to continue
103 * "underneath" a mountpoint that is not exported.
104 * The exception is V4ROOT, where no traversal is ever
105 * allowed without an explicit export of the new
106 * directory.
107 */
108 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
109 err = 0;
110 path_put(&path);
111 goto out;
112 }
113 if (nfsd_v4client(rqstp) ||
114 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
115 /* successfully crossed mount point */
116 /*
117 * This is subtle: path.dentry is *not* on path.mnt
118 * at this point. The only reason we are safe is that
119 * original mnt is pinned down by exp, so we should
120 * put path *before* putting exp
121 */
122 *dpp = path.dentry;
123 path.dentry = dentry;
124 *expp = exp2;
125 exp2 = exp;
126 }
127 path_put(&path);
128 exp_put(exp2);
129 out:
130 return err;
131 }
132
133 static void follow_to_parent(struct path *path)
134 {
135 struct dentry *dp;
136
137 while (path->dentry == path->mnt->mnt_root && follow_up(path))
138 ;
139 dp = dget_parent(path->dentry);
140 dput(path->dentry);
141 path->dentry = dp;
142 }
143
144 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
145 {
146 struct svc_export *exp2;
147 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
148 .dentry = dget(dparent)};
149
150 follow_to_parent(&path);
151
152 exp2 = rqst_exp_parent(rqstp, &path);
153 if (PTR_ERR(exp2) == -ENOENT) {
154 *dentryp = dget(dparent);
155 } else if (IS_ERR(exp2)) {
156 path_put(&path);
157 return PTR_ERR(exp2);
158 } else {
159 *dentryp = dget(path.dentry);
160 exp_put(*exp);
161 *exp = exp2;
162 }
163 path_put(&path);
164 return 0;
165 }
166
167 /*
168 * For nfsd purposes, we treat V4ROOT exports as though there was an
169 * export at *every* directory.
170 */
171 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
172 {
173 if (d_mountpoint(dentry))
174 return 1;
175 if (nfsd4_is_junction(dentry))
176 return 1;
177 if (!(exp->ex_flags & NFSEXP_V4ROOT))
178 return 0;
179 return d_inode(dentry) != NULL;
180 }
181
182 __be32
183 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
184 const char *name, unsigned int len,
185 struct svc_export **exp_ret, struct dentry **dentry_ret)
186 {
187 struct svc_export *exp;
188 struct dentry *dparent;
189 struct dentry *dentry;
190 int host_err;
191
192 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
193
194 dparent = fhp->fh_dentry;
195 exp = exp_get(fhp->fh_export);
196
197 /* Lookup the name, but don't follow links */
198 if (isdotent(name, len)) {
199 if (len==1)
200 dentry = dget(dparent);
201 else if (dparent != exp->ex_path.dentry)
202 dentry = dget_parent(dparent);
203 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
204 dentry = dget(dparent); /* .. == . just like at / */
205 else {
206 /* checking mountpoint crossing is very different when stepping up */
207 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
208 if (host_err)
209 goto out_nfserr;
210 }
211 } else {
212 /*
213 * In the nfsd4_open() case, this may be held across
214 * subsequent open and delegation acquisition which may
215 * need to take the child's i_mutex:
216 */
217 fh_lock_nested(fhp, I_MUTEX_PARENT);
218 dentry = lookup_one_len(name, dparent, len);
219 host_err = PTR_ERR(dentry);
220 if (IS_ERR(dentry))
221 goto out_nfserr;
222 if (nfsd_mountpoint(dentry, exp)) {
223 /*
224 * We don't need the i_mutex after all. It's
225 * still possible we could open this (regular
226 * files can be mountpoints too), but the
227 * i_mutex is just there to prevent renames of
228 * something that we might be about to delegate,
229 * and a mountpoint won't be renamed:
230 */
231 fh_unlock(fhp);
232 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
233 dput(dentry);
234 goto out_nfserr;
235 }
236 }
237 }
238 *dentry_ret = dentry;
239 *exp_ret = exp;
240 return 0;
241
242 out_nfserr:
243 exp_put(exp);
244 return nfserrno(host_err);
245 }
246
247 /*
248 * Look up one component of a pathname.
249 * N.B. After this call _both_ fhp and resfh need an fh_put
250 *
251 * If the lookup would cross a mountpoint, and the mounted filesystem
252 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
253 * accepted as it stands and the mounted directory is
254 * returned. Otherwise the covered directory is returned.
255 * NOTE: this mountpoint crossing is not supported properly by all
256 * clients and is explicitly disallowed for NFSv3
257 * NeilBrown <neilb@cse.unsw.edu.au>
258 */
259 __be32
260 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
261 unsigned int len, struct svc_fh *resfh)
262 {
263 struct svc_export *exp;
264 struct dentry *dentry;
265 __be32 err;
266
267 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
268 if (err)
269 return err;
270 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
271 if (err)
272 return err;
273 err = check_nfsd_access(exp, rqstp);
274 if (err)
275 goto out;
276 /*
277 * Note: we compose the file handle now, but as the
278 * dentry may be negative, it may need to be updated.
279 */
280 err = fh_compose(resfh, exp, dentry, fhp);
281 if (!err && d_really_is_negative(dentry))
282 err = nfserr_noent;
283 out:
284 dput(dentry);
285 exp_put(exp);
286 return err;
287 }
288
289 /*
290 * Commit metadata changes to stable storage.
291 */
292 static int
293 commit_metadata(struct svc_fh *fhp)
294 {
295 struct inode *inode = d_inode(fhp->fh_dentry);
296 const struct export_operations *export_ops = inode->i_sb->s_export_op;
297
298 if (!EX_ISSYNC(fhp->fh_export))
299 return 0;
300
301 if (export_ops->commit_metadata)
302 return export_ops->commit_metadata(inode);
303 return sync_inode_metadata(inode, 1);
304 }
305
306 /*
307 * Go over the attributes and take care of the small differences between
308 * NFS semantics and what Linux expects.
309 */
310 static void
311 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
312 {
313 /* sanitize the mode change */
314 if (iap->ia_valid & ATTR_MODE) {
315 iap->ia_mode &= S_IALLUGO;
316 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
317 }
318
319 /* Revoke setuid/setgid on chown */
320 if (!S_ISDIR(inode->i_mode) &&
321 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) {
322 iap->ia_valid |= ATTR_KILL_PRIV;
323 if (iap->ia_valid & ATTR_MODE) {
324 /* we're setting mode too, just clear the s*id bits */
325 iap->ia_mode &= ~S_ISUID;
326 if (iap->ia_mode & S_IXGRP)
327 iap->ia_mode &= ~S_ISGID;
328 } else {
329 /* set ATTR_KILL_* bits and let VFS handle it */
330 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
331 }
332 }
333 }
334
335 static __be32
336 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
337 struct iattr *iap)
338 {
339 struct inode *inode = d_inode(fhp->fh_dentry);
340 int host_err;
341
342 if (iap->ia_size < inode->i_size) {
343 __be32 err;
344
345 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
346 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
347 if (err)
348 return err;
349 }
350
351 host_err = get_write_access(inode);
352 if (host_err)
353 goto out_nfserrno;
354
355 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
356 if (host_err)
357 goto out_put_write_access;
358 return 0;
359
360 out_put_write_access:
361 put_write_access(inode);
362 out_nfserrno:
363 return nfserrno(host_err);
364 }
365
366 /*
367 * Set various file attributes. After this call fhp needs an fh_put.
368 */
369 __be32
370 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
371 int check_guard, time_t guardtime)
372 {
373 struct dentry *dentry;
374 struct inode *inode;
375 int accmode = NFSD_MAY_SATTR;
376 umode_t ftype = 0;
377 __be32 err;
378 int host_err;
379 bool get_write_count;
380 int size_change = 0;
381
382 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
383 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
384 if (iap->ia_valid & ATTR_SIZE)
385 ftype = S_IFREG;
386
387 /* Callers that do fh_verify should do the fh_want_write: */
388 get_write_count = !fhp->fh_dentry;
389
390 /* Get inode */
391 err = fh_verify(rqstp, fhp, ftype, accmode);
392 if (err)
393 goto out;
394 if (get_write_count) {
395 host_err = fh_want_write(fhp);
396 if (host_err)
397 return nfserrno(host_err);
398 }
399
400 dentry = fhp->fh_dentry;
401 inode = d_inode(dentry);
402
403 /* Ignore any mode updates on symlinks */
404 if (S_ISLNK(inode->i_mode))
405 iap->ia_valid &= ~ATTR_MODE;
406
407 if (!iap->ia_valid)
408 goto out;
409
410 nfsd_sanitize_attrs(inode, iap);
411
412 /*
413 * The size case is special, it changes the file in addition to the
414 * attributes.
415 */
416 if (iap->ia_valid & ATTR_SIZE) {
417 err = nfsd_get_write_access(rqstp, fhp, iap);
418 if (err)
419 goto out;
420 size_change = 1;
421
422 /*
423 * RFC5661, Section 18.30.4:
424 * Changing the size of a file with SETATTR indirectly
425 * changes the time_modify and change attributes.
426 *
427 * (and similar for the older RFCs)
428 */
429 if (iap->ia_size != i_size_read(inode))
430 iap->ia_valid |= ATTR_MTIME;
431 }
432
433 iap->ia_valid |= ATTR_CTIME;
434
435 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
436 err = nfserr_notsync;
437 goto out_put_write_access;
438 }
439
440 fh_lock(fhp);
441 host_err = notify_change(dentry, iap, NULL);
442 fh_unlock(fhp);
443 err = nfserrno(host_err);
444
445 out_put_write_access:
446 if (size_change)
447 put_write_access(inode);
448 if (!err)
449 err = nfserrno(commit_metadata(fhp));
450 out:
451 return err;
452 }
453
454 #if defined(CONFIG_NFSD_V4)
455 /*
456 * NFS junction information is stored in an extended attribute.
457 */
458 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
459
460 /**
461 * nfsd4_is_junction - Test if an object could be an NFS junction
462 *
463 * @dentry: object to test
464 *
465 * Returns 1 if "dentry" appears to contain NFS junction information.
466 * Otherwise 0 is returned.
467 */
468 int nfsd4_is_junction(struct dentry *dentry)
469 {
470 struct inode *inode = d_inode(dentry);
471
472 if (inode == NULL)
473 return 0;
474 if (inode->i_mode & S_IXUGO)
475 return 0;
476 if (!(inode->i_mode & S_ISVTX))
477 return 0;
478 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
479 return 0;
480 return 1;
481 }
482 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
483 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
484 struct xdr_netobj *label)
485 {
486 __be32 error;
487 int host_error;
488 struct dentry *dentry;
489
490 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
491 if (error)
492 return error;
493
494 dentry = fhp->fh_dentry;
495
496 inode_lock(d_inode(dentry));
497 host_error = security_inode_setsecctx(dentry, label->data, label->len);
498 inode_unlock(d_inode(dentry));
499 return nfserrno(host_error);
500 }
501 #else
502 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
503 struct xdr_netobj *label)
504 {
505 return nfserr_notsupp;
506 }
507 #endif
508
509 __be32 nfsd4_clone_file_range(struct file *src, u64 src_pos, struct file *dst,
510 u64 dst_pos, u64 count)
511 {
512 return nfserrno(vfs_clone_file_range(src, src_pos, dst, dst_pos,
513 count));
514 }
515
516 __be32 nfsd4_vfs_fallocate(struct svc_rqst *rqstp, struct svc_fh *fhp,
517 struct file *file, loff_t offset, loff_t len,
518 int flags)
519 {
520 int error;
521
522 if (!S_ISREG(file_inode(file)->i_mode))
523 return nfserr_inval;
524
525 error = vfs_fallocate(file, flags, offset, len);
526 if (!error)
527 error = commit_metadata(fhp);
528
529 return nfserrno(error);
530 }
531 #endif /* defined(CONFIG_NFSD_V4) */
532
533 #ifdef CONFIG_NFSD_V3
534 /*
535 * Check server access rights to a file system object
536 */
537 struct accessmap {
538 u32 access;
539 int how;
540 };
541 static struct accessmap nfs3_regaccess[] = {
542 { NFS3_ACCESS_READ, NFSD_MAY_READ },
543 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
544 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
545 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
546
547 { 0, 0 }
548 };
549
550 static struct accessmap nfs3_diraccess[] = {
551 { NFS3_ACCESS_READ, NFSD_MAY_READ },
552 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
553 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
554 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
555 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
556
557 { 0, 0 }
558 };
559
560 static struct accessmap nfs3_anyaccess[] = {
561 /* Some clients - Solaris 2.6 at least, make an access call
562 * to the server to check for access for things like /dev/null
563 * (which really, the server doesn't care about). So
564 * We provide simple access checking for them, looking
565 * mainly at mode bits, and we make sure to ignore read-only
566 * filesystem checks
567 */
568 { NFS3_ACCESS_READ, NFSD_MAY_READ },
569 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
570 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
571 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
572
573 { 0, 0 }
574 };
575
576 __be32
577 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
578 {
579 struct accessmap *map;
580 struct svc_export *export;
581 struct dentry *dentry;
582 u32 query, result = 0, sresult = 0;
583 __be32 error;
584
585 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
586 if (error)
587 goto out;
588
589 export = fhp->fh_export;
590 dentry = fhp->fh_dentry;
591
592 if (d_is_reg(dentry))
593 map = nfs3_regaccess;
594 else if (d_is_dir(dentry))
595 map = nfs3_diraccess;
596 else
597 map = nfs3_anyaccess;
598
599
600 query = *access;
601 for (; map->access; map++) {
602 if (map->access & query) {
603 __be32 err2;
604
605 sresult |= map->access;
606
607 err2 = nfsd_permission(rqstp, export, dentry, map->how);
608 switch (err2) {
609 case nfs_ok:
610 result |= map->access;
611 break;
612
613 /* the following error codes just mean the access was not allowed,
614 * rather than an error occurred */
615 case nfserr_rofs:
616 case nfserr_acces:
617 case nfserr_perm:
618 /* simply don't "or" in the access bit. */
619 break;
620 default:
621 error = err2;
622 goto out;
623 }
624 }
625 }
626 *access = result;
627 if (supported)
628 *supported = sresult;
629
630 out:
631 return error;
632 }
633 #endif /* CONFIG_NFSD_V3 */
634
635 static int nfsd_open_break_lease(struct inode *inode, int access)
636 {
637 unsigned int mode;
638
639 if (access & NFSD_MAY_NOT_BREAK_LEASE)
640 return 0;
641 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
642 return break_lease(inode, mode | O_NONBLOCK);
643 }
644
645 /*
646 * Open an existing file or directory.
647 * The may_flags argument indicates the type of open (read/write/lock)
648 * and additional flags.
649 * N.B. After this call fhp needs an fh_put
650 */
651 __be32
652 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
653 int may_flags, struct file **filp)
654 {
655 struct path path;
656 struct inode *inode;
657 struct file *file;
658 int flags = O_RDONLY|O_LARGEFILE;
659 __be32 err;
660 int host_err = 0;
661
662 validate_process_creds();
663
664 /*
665 * If we get here, then the client has already done an "open",
666 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
667 * in case a chmod has now revoked permission.
668 *
669 * Arguably we should also allow the owner override for
670 * directories, but we never have and it doesn't seem to have
671 * caused anyone a problem. If we were to change this, note
672 * also that our filldir callbacks would need a variant of
673 * lookup_one_len that doesn't check permissions.
674 */
675 if (type == S_IFREG)
676 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
677 err = fh_verify(rqstp, fhp, type, may_flags);
678 if (err)
679 goto out;
680
681 path.mnt = fhp->fh_export->ex_path.mnt;
682 path.dentry = fhp->fh_dentry;
683 inode = d_inode(path.dentry);
684
685 /* Disallow write access to files with the append-only bit set
686 * or any access when mandatory locking enabled
687 */
688 err = nfserr_perm;
689 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
690 goto out;
691 /*
692 * We must ignore files (but only files) which might have mandatory
693 * locks on them because there is no way to know if the accesser has
694 * the lock.
695 */
696 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
697 goto out;
698
699 if (!inode->i_fop)
700 goto out;
701
702 host_err = nfsd_open_break_lease(inode, may_flags);
703 if (host_err) /* NOMEM or WOULDBLOCK */
704 goto out_nfserr;
705
706 if (may_flags & NFSD_MAY_WRITE) {
707 if (may_flags & NFSD_MAY_READ)
708 flags = O_RDWR|O_LARGEFILE;
709 else
710 flags = O_WRONLY|O_LARGEFILE;
711 }
712
713 file = dentry_open(&path, flags, current_cred());
714 if (IS_ERR(file)) {
715 host_err = PTR_ERR(file);
716 goto out_nfserr;
717 }
718
719 host_err = ima_file_check(file, may_flags, 0);
720 if (host_err) {
721 fput(file);
722 goto out_nfserr;
723 }
724
725 if (may_flags & NFSD_MAY_64BIT_COOKIE)
726 file->f_mode |= FMODE_64BITHASH;
727 else
728 file->f_mode |= FMODE_32BITHASH;
729
730 *filp = file;
731 out_nfserr:
732 err = nfserrno(host_err);
733 out:
734 validate_process_creds();
735 return err;
736 }
737
738 struct raparms *
739 nfsd_init_raparms(struct file *file)
740 {
741 struct inode *inode = file_inode(file);
742 dev_t dev = inode->i_sb->s_dev;
743 ino_t ino = inode->i_ino;
744 struct raparms *ra, **rap, **frap = NULL;
745 int depth = 0;
746 unsigned int hash;
747 struct raparm_hbucket *rab;
748
749 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
750 rab = &raparm_hash[hash];
751
752 spin_lock(&rab->pb_lock);
753 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
754 if (ra->p_ino == ino && ra->p_dev == dev)
755 goto found;
756 depth++;
757 if (ra->p_count == 0)
758 frap = rap;
759 }
760 depth = nfsdstats.ra_size;
761 if (!frap) {
762 spin_unlock(&rab->pb_lock);
763 return NULL;
764 }
765 rap = frap;
766 ra = *frap;
767 ra->p_dev = dev;
768 ra->p_ino = ino;
769 ra->p_set = 0;
770 ra->p_hindex = hash;
771 found:
772 if (rap != &rab->pb_head) {
773 *rap = ra->p_next;
774 ra->p_next = rab->pb_head;
775 rab->pb_head = ra;
776 }
777 ra->p_count++;
778 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
779 spin_unlock(&rab->pb_lock);
780
781 if (ra->p_set)
782 file->f_ra = ra->p_ra;
783 return ra;
784 }
785
786 void nfsd_put_raparams(struct file *file, struct raparms *ra)
787 {
788 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
789
790 spin_lock(&rab->pb_lock);
791 ra->p_ra = file->f_ra;
792 ra->p_set = 1;
793 ra->p_count--;
794 spin_unlock(&rab->pb_lock);
795 }
796
797 /*
798 * Grab and keep cached pages associated with a file in the svc_rqst
799 * so that they can be passed to the network sendmsg/sendpage routines
800 * directly. They will be released after the sending has completed.
801 */
802 static int
803 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
804 struct splice_desc *sd)
805 {
806 struct svc_rqst *rqstp = sd->u.data;
807 struct page **pp = rqstp->rq_next_page;
808 struct page *page = buf->page;
809 size_t size;
810
811 size = sd->len;
812
813 if (rqstp->rq_res.page_len == 0) {
814 get_page(page);
815 put_page(*rqstp->rq_next_page);
816 *(rqstp->rq_next_page++) = page;
817 rqstp->rq_res.page_base = buf->offset;
818 rqstp->rq_res.page_len = size;
819 } else if (page != pp[-1]) {
820 get_page(page);
821 if (*rqstp->rq_next_page)
822 put_page(*rqstp->rq_next_page);
823 *(rqstp->rq_next_page++) = page;
824 rqstp->rq_res.page_len += size;
825 } else
826 rqstp->rq_res.page_len += size;
827
828 return size;
829 }
830
831 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
832 struct splice_desc *sd)
833 {
834 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
835 }
836
837 static __be32
838 nfsd_finish_read(struct file *file, unsigned long *count, int host_err)
839 {
840 if (host_err >= 0) {
841 nfsdstats.io_read += host_err;
842 *count = host_err;
843 fsnotify_access(file);
844 return 0;
845 } else
846 return nfserrno(host_err);
847 }
848
849 __be32 nfsd_splice_read(struct svc_rqst *rqstp,
850 struct file *file, loff_t offset, unsigned long *count)
851 {
852 struct splice_desc sd = {
853 .len = 0,
854 .total_len = *count,
855 .pos = offset,
856 .u.data = rqstp,
857 };
858 int host_err;
859
860 rqstp->rq_next_page = rqstp->rq_respages + 1;
861 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
862 return nfsd_finish_read(file, count, host_err);
863 }
864
865 __be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen,
866 unsigned long *count)
867 {
868 mm_segment_t oldfs;
869 int host_err;
870
871 oldfs = get_fs();
872 set_fs(KERNEL_DS);
873 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset, 0);
874 set_fs(oldfs);
875 return nfsd_finish_read(file, count, host_err);
876 }
877
878 static __be32
879 nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file,
880 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
881 {
882 if (file->f_op->splice_read && test_bit(RQ_SPLICE_OK, &rqstp->rq_flags))
883 return nfsd_splice_read(rqstp, file, offset, count);
884 else
885 return nfsd_readv(file, offset, vec, vlen, count);
886 }
887
888 /*
889 * Gathered writes: If another process is currently writing to the file,
890 * there's a high chance this is another nfsd (triggered by a bulk write
891 * from a client's biod). Rather than syncing the file with each write
892 * request, we sleep for 10 msec.
893 *
894 * I don't know if this roughly approximates C. Juszak's idea of
895 * gathered writes, but it's a nice and simple solution (IMHO), and it
896 * seems to work:-)
897 *
898 * Note: we do this only in the NFSv2 case, since v3 and higher have a
899 * better tool (separate unstable writes and commits) for solving this
900 * problem.
901 */
902 static int wait_for_concurrent_writes(struct file *file)
903 {
904 struct inode *inode = file_inode(file);
905 static ino_t last_ino;
906 static dev_t last_dev;
907 int err = 0;
908
909 if (atomic_read(&inode->i_writecount) > 1
910 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
911 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
912 msleep(10);
913 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
914 }
915
916 if (inode->i_state & I_DIRTY) {
917 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
918 err = vfs_fsync(file, 0);
919 }
920 last_ino = inode->i_ino;
921 last_dev = inode->i_sb->s_dev;
922 return err;
923 }
924
925 __be32
926 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
927 loff_t offset, struct kvec *vec, int vlen,
928 unsigned long *cnt, int *stablep)
929 {
930 struct svc_export *exp;
931 struct inode *inode;
932 mm_segment_t oldfs;
933 __be32 err = 0;
934 int host_err;
935 int stable = *stablep;
936 int use_wgather;
937 loff_t pos = offset;
938 unsigned int pflags = current->flags;
939 int flags = 0;
940
941 if (test_bit(RQ_LOCAL, &rqstp->rq_flags))
942 /*
943 * We want less throttling in balance_dirty_pages()
944 * and shrink_inactive_list() so that nfs to
945 * localhost doesn't cause nfsd to lock up due to all
946 * the client's dirty pages or its congested queue.
947 */
948 current->flags |= PF_LESS_THROTTLE;
949
950 inode = file_inode(file);
951 exp = fhp->fh_export;
952
953 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
954
955 if (!EX_ISSYNC(exp))
956 stable = 0;
957
958 if (stable && !use_wgather)
959 flags |= RWF_SYNC;
960
961 /* Write the data. */
962 oldfs = get_fs(); set_fs(KERNEL_DS);
963 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos, flags);
964 set_fs(oldfs);
965 if (host_err < 0)
966 goto out_nfserr;
967 *cnt = host_err;
968 nfsdstats.io_write += host_err;
969 fsnotify_modify(file);
970
971 if (stable && use_wgather)
972 host_err = wait_for_concurrent_writes(file);
973
974 out_nfserr:
975 dprintk("nfsd: write complete host_err=%d\n", host_err);
976 if (host_err >= 0)
977 err = 0;
978 else
979 err = nfserrno(host_err);
980 if (test_bit(RQ_LOCAL, &rqstp->rq_flags))
981 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE);
982 return err;
983 }
984
985 /*
986 * Read data from a file. count must contain the requested read count
987 * on entry. On return, *count contains the number of bytes actually read.
988 * N.B. After this call fhp needs an fh_put
989 */
990 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
991 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
992 {
993 struct file *file;
994 struct raparms *ra;
995 __be32 err;
996
997 trace_read_start(rqstp, fhp, offset, vlen);
998 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
999 if (err)
1000 return err;
1001
1002 ra = nfsd_init_raparms(file);
1003
1004 trace_read_opened(rqstp, fhp, offset, vlen);
1005 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count);
1006 trace_read_io_done(rqstp, fhp, offset, vlen);
1007
1008 if (ra)
1009 nfsd_put_raparams(file, ra);
1010 fput(file);
1011
1012 trace_read_done(rqstp, fhp, offset, vlen);
1013
1014 return err;
1015 }
1016
1017 /*
1018 * Write data to a file.
1019 * The stable flag requests synchronous writes.
1020 * N.B. After this call fhp needs an fh_put
1021 */
1022 __be32
1023 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1024 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1025 int *stablep)
1026 {
1027 __be32 err = 0;
1028
1029 trace_write_start(rqstp, fhp, offset, vlen);
1030
1031 if (file) {
1032 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1033 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1034 if (err)
1035 goto out;
1036 trace_write_opened(rqstp, fhp, offset, vlen);
1037 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1038 stablep);
1039 trace_write_io_done(rqstp, fhp, offset, vlen);
1040 } else {
1041 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1042 if (err)
1043 goto out;
1044
1045 trace_write_opened(rqstp, fhp, offset, vlen);
1046 if (cnt)
1047 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1048 cnt, stablep);
1049 trace_write_io_done(rqstp, fhp, offset, vlen);
1050 fput(file);
1051 }
1052 out:
1053 trace_write_done(rqstp, fhp, offset, vlen);
1054 return err;
1055 }
1056
1057 #ifdef CONFIG_NFSD_V3
1058 /*
1059 * Commit all pending writes to stable storage.
1060 *
1061 * Note: we only guarantee that data that lies within the range specified
1062 * by the 'offset' and 'count' parameters will be synced.
1063 *
1064 * Unfortunately we cannot lock the file to make sure we return full WCC
1065 * data to the client, as locking happens lower down in the filesystem.
1066 */
1067 __be32
1068 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1069 loff_t offset, unsigned long count)
1070 {
1071 struct file *file;
1072 loff_t end = LLONG_MAX;
1073 __be32 err = nfserr_inval;
1074
1075 if (offset < 0)
1076 goto out;
1077 if (count != 0) {
1078 end = offset + (loff_t)count - 1;
1079 if (end < offset)
1080 goto out;
1081 }
1082
1083 err = nfsd_open(rqstp, fhp, S_IFREG,
1084 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1085 if (err)
1086 goto out;
1087 if (EX_ISSYNC(fhp->fh_export)) {
1088 int err2 = vfs_fsync_range(file, offset, end, 0);
1089
1090 if (err2 != -EINVAL)
1091 err = nfserrno(err2);
1092 else
1093 err = nfserr_notsupp;
1094 }
1095
1096 fput(file);
1097 out:
1098 return err;
1099 }
1100 #endif /* CONFIG_NFSD_V3 */
1101
1102 static __be32
1103 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1104 struct iattr *iap)
1105 {
1106 /*
1107 * Mode has already been set earlier in create:
1108 */
1109 iap->ia_valid &= ~ATTR_MODE;
1110 /*
1111 * Setting uid/gid works only for root. Irix appears to
1112 * send along the gid on create when it tries to implement
1113 * setgid directories via NFS:
1114 */
1115 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1116 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1117 if (iap->ia_valid)
1118 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1119 /* Callers expect file metadata to be committed here */
1120 return nfserrno(commit_metadata(resfhp));
1121 }
1122
1123 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1124 * setting size to 0 may fail for some specific file systems by the permission
1125 * checking which requires WRITE permission but the mode is 000.
1126 * we ignore the resizing(to 0) on the just new created file, since the size is
1127 * 0 after file created.
1128 *
1129 * call this only after vfs_create() is called.
1130 * */
1131 static void
1132 nfsd_check_ignore_resizing(struct iattr *iap)
1133 {
1134 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1135 iap->ia_valid &= ~ATTR_SIZE;
1136 }
1137
1138 /* The parent directory should already be locked: */
1139 __be32
1140 nfsd_create_locked(struct svc_rqst *rqstp, struct svc_fh *fhp,
1141 char *fname, int flen, struct iattr *iap,
1142 int type, dev_t rdev, struct svc_fh *resfhp)
1143 {
1144 struct dentry *dentry, *dchild;
1145 struct inode *dirp;
1146 __be32 err;
1147 __be32 err2;
1148 int host_err;
1149
1150 dentry = fhp->fh_dentry;
1151 dirp = d_inode(dentry);
1152
1153 dchild = dget(resfhp->fh_dentry);
1154 if (!fhp->fh_locked) {
1155 WARN_ONCE(1, "nfsd_create: parent %pd2 not locked!\n",
1156 dentry);
1157 err = nfserr_io;
1158 goto out;
1159 }
1160
1161 err = nfsd_permission(rqstp, fhp->fh_export, dentry, NFSD_MAY_CREATE);
1162 if (err)
1163 goto out;
1164
1165 if (!(iap->ia_valid & ATTR_MODE))
1166 iap->ia_mode = 0;
1167 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1168
1169 err = 0;
1170 host_err = 0;
1171 switch (type) {
1172 case S_IFREG:
1173 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1174 if (!host_err)
1175 nfsd_check_ignore_resizing(iap);
1176 break;
1177 case S_IFDIR:
1178 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1179 break;
1180 case S_IFCHR:
1181 case S_IFBLK:
1182 case S_IFIFO:
1183 case S_IFSOCK:
1184 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1185 break;
1186 default:
1187 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1188 type);
1189 host_err = -EINVAL;
1190 }
1191 if (host_err < 0)
1192 goto out_nfserr;
1193
1194 err = nfsd_create_setattr(rqstp, resfhp, iap);
1195
1196 /*
1197 * nfsd_create_setattr already committed the child. Transactional
1198 * filesystems had a chance to commit changes for both parent and
1199 * child simultaneously making the following commit_metadata a
1200 * noop.
1201 */
1202 err2 = nfserrno(commit_metadata(fhp));
1203 if (err2)
1204 err = err2;
1205 /*
1206 * Update the file handle to get the new inode info.
1207 */
1208 if (!err)
1209 err = fh_update(resfhp);
1210 out:
1211 dput(dchild);
1212 return err;
1213
1214 out_nfserr:
1215 err = nfserrno(host_err);
1216 goto out;
1217 }
1218
1219 /*
1220 * Create a filesystem object (regular, directory, special).
1221 * Note that the parent directory is left locked.
1222 *
1223 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1224 */
1225 __be32
1226 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1227 char *fname, int flen, struct iattr *iap,
1228 int type, dev_t rdev, struct svc_fh *resfhp)
1229 {
1230 struct dentry *dentry, *dchild = NULL;
1231 struct inode *dirp;
1232 __be32 err;
1233 int host_err;
1234
1235 if (isdotent(fname, flen))
1236 return nfserr_exist;
1237
1238 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_NOP);
1239 if (err)
1240 return err;
1241
1242 dentry = fhp->fh_dentry;
1243 dirp = d_inode(dentry);
1244
1245 host_err = fh_want_write(fhp);
1246 if (host_err)
1247 return nfserrno(host_err);
1248
1249 fh_lock_nested(fhp, I_MUTEX_PARENT);
1250 dchild = lookup_one_len(fname, dentry, flen);
1251 host_err = PTR_ERR(dchild);
1252 if (IS_ERR(dchild))
1253 return nfserrno(host_err);
1254 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1255 /*
1256 * We unconditionally drop our ref to dchild as fh_compose will have
1257 * already grabbed its own ref for it.
1258 */
1259 dput(dchild);
1260 if (err)
1261 return err;
1262 return nfsd_create_locked(rqstp, fhp, fname, flen, iap, type,
1263 rdev, resfhp);
1264 }
1265
1266 #ifdef CONFIG_NFSD_V3
1267
1268 /*
1269 * NFSv3 and NFSv4 version of nfsd_create
1270 */
1271 __be32
1272 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1273 char *fname, int flen, struct iattr *iap,
1274 struct svc_fh *resfhp, int createmode, u32 *verifier,
1275 bool *truncp, bool *created)
1276 {
1277 struct dentry *dentry, *dchild = NULL;
1278 struct inode *dirp;
1279 __be32 err;
1280 int host_err;
1281 __u32 v_mtime=0, v_atime=0;
1282
1283 err = nfserr_perm;
1284 if (!flen)
1285 goto out;
1286 err = nfserr_exist;
1287 if (isdotent(fname, flen))
1288 goto out;
1289 if (!(iap->ia_valid & ATTR_MODE))
1290 iap->ia_mode = 0;
1291 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1292 if (err)
1293 goto out;
1294
1295 dentry = fhp->fh_dentry;
1296 dirp = d_inode(dentry);
1297
1298 host_err = fh_want_write(fhp);
1299 if (host_err)
1300 goto out_nfserr;
1301
1302 fh_lock_nested(fhp, I_MUTEX_PARENT);
1303
1304 /*
1305 * Compose the response file handle.
1306 */
1307 dchild = lookup_one_len(fname, dentry, flen);
1308 host_err = PTR_ERR(dchild);
1309 if (IS_ERR(dchild))
1310 goto out_nfserr;
1311
1312 /* If file doesn't exist, check for permissions to create one */
1313 if (d_really_is_negative(dchild)) {
1314 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1315 if (err)
1316 goto out;
1317 }
1318
1319 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1320 if (err)
1321 goto out;
1322
1323 if (nfsd_create_is_exclusive(createmode)) {
1324 /* solaris7 gets confused (bugid 4218508) if these have
1325 * the high bit set, so just clear the high bits. If this is
1326 * ever changed to use different attrs for storing the
1327 * verifier, then do_open_lookup() will also need to be fixed
1328 * accordingly.
1329 */
1330 v_mtime = verifier[0]&0x7fffffff;
1331 v_atime = verifier[1]&0x7fffffff;
1332 }
1333
1334 if (d_really_is_positive(dchild)) {
1335 err = 0;
1336
1337 switch (createmode) {
1338 case NFS3_CREATE_UNCHECKED:
1339 if (! d_is_reg(dchild))
1340 goto out;
1341 else if (truncp) {
1342 /* in nfsv4, we need to treat this case a little
1343 * differently. we don't want to truncate the
1344 * file now; this would be wrong if the OPEN
1345 * fails for some other reason. furthermore,
1346 * if the size is nonzero, we should ignore it
1347 * according to spec!
1348 */
1349 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1350 }
1351 else {
1352 iap->ia_valid &= ATTR_SIZE;
1353 goto set_attr;
1354 }
1355 break;
1356 case NFS3_CREATE_EXCLUSIVE:
1357 if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime
1358 && d_inode(dchild)->i_atime.tv_sec == v_atime
1359 && d_inode(dchild)->i_size == 0 ) {
1360 if (created)
1361 *created = 1;
1362 break;
1363 }
1364 case NFS4_CREATE_EXCLUSIVE4_1:
1365 if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime
1366 && d_inode(dchild)->i_atime.tv_sec == v_atime
1367 && d_inode(dchild)->i_size == 0 ) {
1368 if (created)
1369 *created = 1;
1370 goto set_attr;
1371 }
1372 /* fallthru */
1373 case NFS3_CREATE_GUARDED:
1374 err = nfserr_exist;
1375 }
1376 fh_drop_write(fhp);
1377 goto out;
1378 }
1379
1380 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1381 if (host_err < 0) {
1382 fh_drop_write(fhp);
1383 goto out_nfserr;
1384 }
1385 if (created)
1386 *created = 1;
1387
1388 nfsd_check_ignore_resizing(iap);
1389
1390 if (nfsd_create_is_exclusive(createmode)) {
1391 /* Cram the verifier into atime/mtime */
1392 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1393 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1394 /* XXX someone who knows this better please fix it for nsec */
1395 iap->ia_mtime.tv_sec = v_mtime;
1396 iap->ia_atime.tv_sec = v_atime;
1397 iap->ia_mtime.tv_nsec = 0;
1398 iap->ia_atime.tv_nsec = 0;
1399 }
1400
1401 set_attr:
1402 err = nfsd_create_setattr(rqstp, resfhp, iap);
1403
1404 /*
1405 * nfsd_create_setattr already committed the child
1406 * (and possibly also the parent).
1407 */
1408 if (!err)
1409 err = nfserrno(commit_metadata(fhp));
1410
1411 /*
1412 * Update the filehandle to get the new inode info.
1413 */
1414 if (!err)
1415 err = fh_update(resfhp);
1416
1417 out:
1418 fh_unlock(fhp);
1419 if (dchild && !IS_ERR(dchild))
1420 dput(dchild);
1421 fh_drop_write(fhp);
1422 return err;
1423
1424 out_nfserr:
1425 err = nfserrno(host_err);
1426 goto out;
1427 }
1428 #endif /* CONFIG_NFSD_V3 */
1429
1430 /*
1431 * Read a symlink. On entry, *lenp must contain the maximum path length that
1432 * fits into the buffer. On return, it contains the true length.
1433 * N.B. After this call fhp needs an fh_put
1434 */
1435 __be32
1436 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1437 {
1438 struct inode *inode;
1439 mm_segment_t oldfs;
1440 __be32 err;
1441 int host_err;
1442 struct path path;
1443
1444 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1445 if (err)
1446 goto out;
1447
1448 path.mnt = fhp->fh_export->ex_path.mnt;
1449 path.dentry = fhp->fh_dentry;
1450 inode = d_inode(path.dentry);
1451
1452 err = nfserr_inval;
1453 if (!inode->i_op->readlink)
1454 goto out;
1455
1456 touch_atime(&path);
1457 /* N.B. Why does this call need a get_fs()??
1458 * Remove the set_fs and watch the fireworks:-) --okir
1459 */
1460
1461 oldfs = get_fs(); set_fs(KERNEL_DS);
1462 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1463 set_fs(oldfs);
1464
1465 if (host_err < 0)
1466 goto out_nfserr;
1467 *lenp = host_err;
1468 err = 0;
1469 out:
1470 return err;
1471
1472 out_nfserr:
1473 err = nfserrno(host_err);
1474 goto out;
1475 }
1476
1477 /*
1478 * Create a symlink and look up its inode
1479 * N.B. After this call _both_ fhp and resfhp need an fh_put
1480 */
1481 __be32
1482 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1483 char *fname, int flen,
1484 char *path,
1485 struct svc_fh *resfhp)
1486 {
1487 struct dentry *dentry, *dnew;
1488 __be32 err, cerr;
1489 int host_err;
1490
1491 err = nfserr_noent;
1492 if (!flen || path[0] == '\0')
1493 goto out;
1494 err = nfserr_exist;
1495 if (isdotent(fname, flen))
1496 goto out;
1497
1498 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1499 if (err)
1500 goto out;
1501
1502 host_err = fh_want_write(fhp);
1503 if (host_err)
1504 goto out_nfserr;
1505
1506 fh_lock(fhp);
1507 dentry = fhp->fh_dentry;
1508 dnew = lookup_one_len(fname, dentry, flen);
1509 host_err = PTR_ERR(dnew);
1510 if (IS_ERR(dnew))
1511 goto out_nfserr;
1512
1513 host_err = vfs_symlink(d_inode(dentry), dnew, path);
1514 err = nfserrno(host_err);
1515 if (!err)
1516 err = nfserrno(commit_metadata(fhp));
1517 fh_unlock(fhp);
1518
1519 fh_drop_write(fhp);
1520
1521 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1522 dput(dnew);
1523 if (err==0) err = cerr;
1524 out:
1525 return err;
1526
1527 out_nfserr:
1528 err = nfserrno(host_err);
1529 goto out;
1530 }
1531
1532 /*
1533 * Create a hardlink
1534 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1535 */
1536 __be32
1537 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1538 char *name, int len, struct svc_fh *tfhp)
1539 {
1540 struct dentry *ddir, *dnew, *dold;
1541 struct inode *dirp;
1542 __be32 err;
1543 int host_err;
1544
1545 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1546 if (err)
1547 goto out;
1548 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1549 if (err)
1550 goto out;
1551 err = nfserr_isdir;
1552 if (d_is_dir(tfhp->fh_dentry))
1553 goto out;
1554 err = nfserr_perm;
1555 if (!len)
1556 goto out;
1557 err = nfserr_exist;
1558 if (isdotent(name, len))
1559 goto out;
1560
1561 host_err = fh_want_write(tfhp);
1562 if (host_err) {
1563 err = nfserrno(host_err);
1564 goto out;
1565 }
1566
1567 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1568 ddir = ffhp->fh_dentry;
1569 dirp = d_inode(ddir);
1570
1571 dnew = lookup_one_len(name, ddir, len);
1572 host_err = PTR_ERR(dnew);
1573 if (IS_ERR(dnew))
1574 goto out_nfserr;
1575
1576 dold = tfhp->fh_dentry;
1577
1578 err = nfserr_noent;
1579 if (d_really_is_negative(dold))
1580 goto out_dput;
1581 host_err = vfs_link(dold, dirp, dnew, NULL);
1582 if (!host_err) {
1583 err = nfserrno(commit_metadata(ffhp));
1584 if (!err)
1585 err = nfserrno(commit_metadata(tfhp));
1586 } else {
1587 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1588 err = nfserr_acces;
1589 else
1590 err = nfserrno(host_err);
1591 }
1592 out_dput:
1593 dput(dnew);
1594 out_unlock:
1595 fh_unlock(ffhp);
1596 fh_drop_write(tfhp);
1597 out:
1598 return err;
1599
1600 out_nfserr:
1601 err = nfserrno(host_err);
1602 goto out_unlock;
1603 }
1604
1605 /*
1606 * Rename a file
1607 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1608 */
1609 __be32
1610 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1611 struct svc_fh *tfhp, char *tname, int tlen)
1612 {
1613 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1614 struct inode *fdir, *tdir;
1615 __be32 err;
1616 int host_err;
1617
1618 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1619 if (err)
1620 goto out;
1621 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1622 if (err)
1623 goto out;
1624
1625 fdentry = ffhp->fh_dentry;
1626 fdir = d_inode(fdentry);
1627
1628 tdentry = tfhp->fh_dentry;
1629 tdir = d_inode(tdentry);
1630
1631 err = nfserr_perm;
1632 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1633 goto out;
1634
1635 host_err = fh_want_write(ffhp);
1636 if (host_err) {
1637 err = nfserrno(host_err);
1638 goto out;
1639 }
1640
1641 /* cannot use fh_lock as we need deadlock protective ordering
1642 * so do it by hand */
1643 trap = lock_rename(tdentry, fdentry);
1644 ffhp->fh_locked = tfhp->fh_locked = true;
1645 fill_pre_wcc(ffhp);
1646 fill_pre_wcc(tfhp);
1647
1648 odentry = lookup_one_len(fname, fdentry, flen);
1649 host_err = PTR_ERR(odentry);
1650 if (IS_ERR(odentry))
1651 goto out_nfserr;
1652
1653 host_err = -ENOENT;
1654 if (d_really_is_negative(odentry))
1655 goto out_dput_old;
1656 host_err = -EINVAL;
1657 if (odentry == trap)
1658 goto out_dput_old;
1659
1660 ndentry = lookup_one_len(tname, tdentry, tlen);
1661 host_err = PTR_ERR(ndentry);
1662 if (IS_ERR(ndentry))
1663 goto out_dput_old;
1664 host_err = -ENOTEMPTY;
1665 if (ndentry == trap)
1666 goto out_dput_new;
1667
1668 host_err = -EXDEV;
1669 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1670 goto out_dput_new;
1671 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1672 goto out_dput_new;
1673
1674 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0);
1675 if (!host_err) {
1676 host_err = commit_metadata(tfhp);
1677 if (!host_err)
1678 host_err = commit_metadata(ffhp);
1679 }
1680 out_dput_new:
1681 dput(ndentry);
1682 out_dput_old:
1683 dput(odentry);
1684 out_nfserr:
1685 err = nfserrno(host_err);
1686 /*
1687 * We cannot rely on fh_unlock on the two filehandles,
1688 * as that would do the wrong thing if the two directories
1689 * were the same, so again we do it by hand.
1690 */
1691 fill_post_wcc(ffhp);
1692 fill_post_wcc(tfhp);
1693 unlock_rename(tdentry, fdentry);
1694 ffhp->fh_locked = tfhp->fh_locked = false;
1695 fh_drop_write(ffhp);
1696
1697 out:
1698 return err;
1699 }
1700
1701 /*
1702 * Unlink a file or directory
1703 * N.B. After this call fhp needs an fh_put
1704 */
1705 __be32
1706 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1707 char *fname, int flen)
1708 {
1709 struct dentry *dentry, *rdentry;
1710 struct inode *dirp;
1711 __be32 err;
1712 int host_err;
1713
1714 err = nfserr_acces;
1715 if (!flen || isdotent(fname, flen))
1716 goto out;
1717 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1718 if (err)
1719 goto out;
1720
1721 host_err = fh_want_write(fhp);
1722 if (host_err)
1723 goto out_nfserr;
1724
1725 fh_lock_nested(fhp, I_MUTEX_PARENT);
1726 dentry = fhp->fh_dentry;
1727 dirp = d_inode(dentry);
1728
1729 rdentry = lookup_one_len(fname, dentry, flen);
1730 host_err = PTR_ERR(rdentry);
1731 if (IS_ERR(rdentry))
1732 goto out_nfserr;
1733
1734 if (d_really_is_negative(rdentry)) {
1735 dput(rdentry);
1736 err = nfserr_noent;
1737 goto out;
1738 }
1739
1740 if (!type)
1741 type = d_inode(rdentry)->i_mode & S_IFMT;
1742
1743 if (type != S_IFDIR)
1744 host_err = vfs_unlink(dirp, rdentry, NULL);
1745 else
1746 host_err = vfs_rmdir(dirp, rdentry);
1747 if (!host_err)
1748 host_err = commit_metadata(fhp);
1749 dput(rdentry);
1750
1751 out_nfserr:
1752 err = nfserrno(host_err);
1753 out:
1754 return err;
1755 }
1756
1757 /*
1758 * We do this buffering because we must not call back into the file
1759 * system's ->lookup() method from the filldir callback. That may well
1760 * deadlock a number of file systems.
1761 *
1762 * This is based heavily on the implementation of same in XFS.
1763 */
1764 struct buffered_dirent {
1765 u64 ino;
1766 loff_t offset;
1767 int namlen;
1768 unsigned int d_type;
1769 char name[];
1770 };
1771
1772 struct readdir_data {
1773 struct dir_context ctx;
1774 char *dirent;
1775 size_t used;
1776 int full;
1777 };
1778
1779 static int nfsd_buffered_filldir(struct dir_context *ctx, const char *name,
1780 int namlen, loff_t offset, u64 ino,
1781 unsigned int d_type)
1782 {
1783 struct readdir_data *buf =
1784 container_of(ctx, struct readdir_data, ctx);
1785 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1786 unsigned int reclen;
1787
1788 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1789 if (buf->used + reclen > PAGE_SIZE) {
1790 buf->full = 1;
1791 return -EINVAL;
1792 }
1793
1794 de->namlen = namlen;
1795 de->offset = offset;
1796 de->ino = ino;
1797 de->d_type = d_type;
1798 memcpy(de->name, name, namlen);
1799 buf->used += reclen;
1800
1801 return 0;
1802 }
1803
1804 static __be32 nfsd_buffered_readdir(struct file *file, nfsd_filldir_t func,
1805 struct readdir_cd *cdp, loff_t *offsetp)
1806 {
1807 struct buffered_dirent *de;
1808 int host_err;
1809 int size;
1810 loff_t offset;
1811 struct readdir_data buf = {
1812 .ctx.actor = nfsd_buffered_filldir,
1813 .dirent = (void *)__get_free_page(GFP_KERNEL)
1814 };
1815
1816 if (!buf.dirent)
1817 return nfserrno(-ENOMEM);
1818
1819 offset = *offsetp;
1820
1821 while (1) {
1822 unsigned int reclen;
1823
1824 cdp->err = nfserr_eof; /* will be cleared on successful read */
1825 buf.used = 0;
1826 buf.full = 0;
1827
1828 host_err = iterate_dir(file, &buf.ctx);
1829 if (buf.full)
1830 host_err = 0;
1831
1832 if (host_err < 0)
1833 break;
1834
1835 size = buf.used;
1836
1837 if (!size)
1838 break;
1839
1840 de = (struct buffered_dirent *)buf.dirent;
1841 while (size > 0) {
1842 offset = de->offset;
1843
1844 if (func(cdp, de->name, de->namlen, de->offset,
1845 de->ino, de->d_type))
1846 break;
1847
1848 if (cdp->err != nfs_ok)
1849 break;
1850
1851 reclen = ALIGN(sizeof(*de) + de->namlen,
1852 sizeof(u64));
1853 size -= reclen;
1854 de = (struct buffered_dirent *)((char *)de + reclen);
1855 }
1856 if (size > 0) /* We bailed out early */
1857 break;
1858
1859 offset = vfs_llseek(file, 0, SEEK_CUR);
1860 }
1861
1862 free_page((unsigned long)(buf.dirent));
1863
1864 if (host_err)
1865 return nfserrno(host_err);
1866
1867 *offsetp = offset;
1868 return cdp->err;
1869 }
1870
1871 /*
1872 * Read entries from a directory.
1873 * The NFSv3/4 verifier we ignore for now.
1874 */
1875 __be32
1876 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1877 struct readdir_cd *cdp, nfsd_filldir_t func)
1878 {
1879 __be32 err;
1880 struct file *file;
1881 loff_t offset = *offsetp;
1882 int may_flags = NFSD_MAY_READ;
1883
1884 /* NFSv2 only supports 32 bit cookies */
1885 if (rqstp->rq_vers > 2)
1886 may_flags |= NFSD_MAY_64BIT_COOKIE;
1887
1888 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
1889 if (err)
1890 goto out;
1891
1892 offset = vfs_llseek(file, offset, SEEK_SET);
1893 if (offset < 0) {
1894 err = nfserrno((int)offset);
1895 goto out_close;
1896 }
1897
1898 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1899
1900 if (err == nfserr_eof || err == nfserr_toosmall)
1901 err = nfs_ok; /* can still be found in ->err */
1902 out_close:
1903 fput(file);
1904 out:
1905 return err;
1906 }
1907
1908 /*
1909 * Get file system stats
1910 * N.B. After this call fhp needs an fh_put
1911 */
1912 __be32
1913 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1914 {
1915 __be32 err;
1916
1917 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1918 if (!err) {
1919 struct path path = {
1920 .mnt = fhp->fh_export->ex_path.mnt,
1921 .dentry = fhp->fh_dentry,
1922 };
1923 if (vfs_statfs(&path, stat))
1924 err = nfserr_io;
1925 }
1926 return err;
1927 }
1928
1929 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1930 {
1931 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1932 }
1933
1934 /*
1935 * Check for a user's access permissions to this inode.
1936 */
1937 __be32
1938 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1939 struct dentry *dentry, int acc)
1940 {
1941 struct inode *inode = d_inode(dentry);
1942 int err;
1943
1944 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
1945 return 0;
1946 #if 0
1947 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1948 acc,
1949 (acc & NFSD_MAY_READ)? " read" : "",
1950 (acc & NFSD_MAY_WRITE)? " write" : "",
1951 (acc & NFSD_MAY_EXEC)? " exec" : "",
1952 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1953 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1954 (acc & NFSD_MAY_LOCK)? " lock" : "",
1955 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1956 inode->i_mode,
1957 IS_IMMUTABLE(inode)? " immut" : "",
1958 IS_APPEND(inode)? " append" : "",
1959 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
1960 dprintk(" owner %d/%d user %d/%d\n",
1961 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
1962 #endif
1963
1964 /* Normally we reject any write/sattr etc access on a read-only file
1965 * system. But if it is IRIX doing check on write-access for a
1966 * device special file, we ignore rofs.
1967 */
1968 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
1969 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
1970 if (exp_rdonly(rqstp, exp) ||
1971 __mnt_is_readonly(exp->ex_path.mnt))
1972 return nfserr_rofs;
1973 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
1974 return nfserr_perm;
1975 }
1976 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
1977 return nfserr_perm;
1978
1979 if (acc & NFSD_MAY_LOCK) {
1980 /* If we cannot rely on authentication in NLM requests,
1981 * just allow locks, otherwise require read permission, or
1982 * ownership
1983 */
1984 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1985 return 0;
1986 else
1987 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
1988 }
1989 /*
1990 * The file owner always gets access permission for accesses that
1991 * would normally be checked at open time. This is to make
1992 * file access work even when the client has done a fchmod(fd, 0).
1993 *
1994 * However, `cp foo bar' should fail nevertheless when bar is
1995 * readonly. A sensible way to do this might be to reject all
1996 * attempts to truncate a read-only file, because a creat() call
1997 * always implies file truncation.
1998 * ... but this isn't really fair. A process may reasonably call
1999 * ftruncate on an open file descriptor on a file with perm 000.
2000 * We must trust the client to do permission checking - using "ACCESS"
2001 * with NFSv3.
2002 */
2003 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2004 uid_eq(inode->i_uid, current_fsuid()))
2005 return 0;
2006
2007 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2008 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2009
2010 /* Allow read access to binaries even when mode 111 */
2011 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2012 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2013 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2014 err = inode_permission(inode, MAY_EXEC);
2015
2016 return err? nfserrno(err) : 0;
2017 }
2018
2019 void
2020 nfsd_racache_shutdown(void)
2021 {
2022 struct raparms *raparm, *last_raparm;
2023 unsigned int i;
2024
2025 dprintk("nfsd: freeing readahead buffers.\n");
2026
2027 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2028 raparm = raparm_hash[i].pb_head;
2029 while(raparm) {
2030 last_raparm = raparm;
2031 raparm = raparm->p_next;
2032 kfree(last_raparm);
2033 }
2034 raparm_hash[i].pb_head = NULL;
2035 }
2036 }
2037 /*
2038 * Initialize readahead param cache
2039 */
2040 int
2041 nfsd_racache_init(int cache_size)
2042 {
2043 int i;
2044 int j = 0;
2045 int nperbucket;
2046 struct raparms **raparm = NULL;
2047
2048
2049 if (raparm_hash[0].pb_head)
2050 return 0;
2051 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2052 nperbucket = max(2, nperbucket);
2053 cache_size = nperbucket * RAPARM_HASH_SIZE;
2054
2055 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2056
2057 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2058 spin_lock_init(&raparm_hash[i].pb_lock);
2059
2060 raparm = &raparm_hash[i].pb_head;
2061 for (j = 0; j < nperbucket; j++) {
2062 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2063 if (!*raparm)
2064 goto out_nomem;
2065 raparm = &(*raparm)->p_next;
2066 }
2067 *raparm = NULL;
2068 }
2069
2070 nfsdstats.ra_size = cache_size;
2071 return 0;
2072
2073 out_nomem:
2074 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2075 nfsd_racache_shutdown();
2076 return -ENOMEM;
2077 }