nfsd: factor out hash functions for export caches.
[deliverable/linux.git] / fs / nfsd / vfs.c
CommitLineData
1da177e4
LT
1#define MSNFS /* HACK HACK */
2/*
1da177e4
LT
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
6 *
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
12 *
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
15 */
16
1da177e4
LT
17#include <linux/fs.h>
18#include <linux/file.h>
d6b29d7c 19#include <linux/splice.h>
1da177e4 20#include <linux/fcntl.h>
1da177e4 21#include <linux/namei.h>
1da177e4 22#include <linux/delay.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4 24#include <linux/posix_acl_xattr.h>
1da177e4 25#include <linux/xattr.h>
9a74af21
BH
26#include <linux/jhash.h>
27#include <linux/ima.h>
28#include <asm/uaccess.h>
f501912a
BM
29#include <linux/exportfs.h>
30#include <linux/writeback.h>
9a74af21
BH
31
32#ifdef CONFIG_NFSD_V3
33#include "xdr3.h"
34#endif /* CONFIG_NFSD_V3 */
35
5be196e5 36#ifdef CONFIG_NFSD_V4
1da177e4
LT
37#include <linux/nfs4_acl.h>
38#include <linux/nfsd_idmap.h>
1da177e4
LT
39#endif /* CONFIG_NFSD_V4 */
40
9a74af21
BH
41#include "nfsd.h"
42#include "vfs.h"
1da177e4
LT
43
44#define NFSDDBG_FACILITY NFSDDBG_FILEOP
1da177e4
LT
45
46
1da177e4
LT
47/*
48 * This is a cache of readahead params that help us choose the proper
49 * readahead strategy. Initially, we set all readahead parameters to 0
50 * and let the VFS handle things.
51 * If you increase the number of cached files very much, you'll need to
52 * add a hash table here.
53 */
54struct raparms {
55 struct raparms *p_next;
56 unsigned int p_count;
57 ino_t p_ino;
58 dev_t p_dev;
59 int p_set;
60 struct file_ra_state p_ra;
fce1456a 61 unsigned int p_hindex;
1da177e4
LT
62};
63
fce1456a
GB
64struct raparm_hbucket {
65 struct raparms *pb_head;
66 spinlock_t pb_lock;
67} ____cacheline_aligned_in_smp;
68
fce1456a
GB
69#define RAPARM_HASH_BITS 4
70#define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
71#define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
72static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
1da177e4
LT
73
74/*
75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
76 * a mount point.
e0bb89ef 77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
1da177e4
LT
78 * or nfs_ok having possibly changed *dpp and *expp
79 */
80int
81nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
82 struct svc_export **expp)
83{
84 struct svc_export *exp = *expp, *exp2 = NULL;
85 struct dentry *dentry = *dpp;
91c9fa8f
AV
86 struct path path = {.mnt = mntget(exp->ex_path.mnt),
87 .dentry = dget(dentry)};
6264d69d 88 int err = 0;
1da177e4 89
9393bd07 90 while (d_mountpoint(path.dentry) && follow_down(&path))
91c9fa8f 91 ;
1da177e4 92
91c9fa8f 93 exp2 = rqst_exp_get_by_name(rqstp, &path);
1da177e4 94 if (IS_ERR(exp2)) {
3b6cee7b
BF
95 err = PTR_ERR(exp2);
96 /*
97 * We normally allow NFS clients to continue
98 * "underneath" a mountpoint that is not exported.
99 * The exception is V4ROOT, where no traversal is ever
100 * allowed without an explicit export of the new
101 * directory.
102 */
103 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
104 err = 0;
91c9fa8f 105 path_put(&path);
1da177e4
LT
106 goto out;
107 }
3c394dda
SD
108 if (nfsd_v4client(rqstp) ||
109 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
1da177e4 110 /* successfully crossed mount point */
1644ccc8 111 /*
91c9fa8f
AV
112 * This is subtle: path.dentry is *not* on path.mnt
113 * at this point. The only reason we are safe is that
114 * original mnt is pinned down by exp, so we should
115 * put path *before* putting exp
1644ccc8 116 */
91c9fa8f
AV
117 *dpp = path.dentry;
118 path.dentry = dentry;
1644ccc8 119 *expp = exp2;
91c9fa8f 120 exp2 = exp;
1da177e4 121 }
91c9fa8f
AV
122 path_put(&path);
123 exp_put(exp2);
1da177e4
LT
124out:
125 return err;
126}
127
289ede45
BF
128static void follow_to_parent(struct path *path)
129{
130 struct dentry *dp;
131
132 while (path->dentry == path->mnt->mnt_root && follow_up(path))
133 ;
134 dp = dget_parent(path->dentry);
135 dput(path->dentry);
136 path->dentry = dp;
137}
138
139static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
140{
141 struct svc_export *exp2;
142 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
143 .dentry = dget(dparent)};
144
145 follow_to_parent(&path);
146
147 exp2 = rqst_exp_parent(rqstp, &path);
148 if (PTR_ERR(exp2) == -ENOENT) {
149 *dentryp = dget(dparent);
150 } else if (IS_ERR(exp2)) {
151 path_put(&path);
152 return PTR_ERR(exp2);
153 } else {
154 *dentryp = dget(path.dentry);
155 exp_put(*exp);
156 *exp = exp2;
157 }
158 path_put(&path);
159 return 0;
160}
161
82ead7fe
BF
162/*
163 * For nfsd purposes, we treat V4ROOT exports as though there was an
164 * export at *every* directory.
165 */
3227fa41 166int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
82ead7fe
BF
167{
168 if (d_mountpoint(dentry))
169 return 1;
170 if (!(exp->ex_flags & NFSEXP_V4ROOT))
171 return 0;
172 return dentry->d_inode != NULL;
173}
174
6264d69d 175__be32
6c0a654d 176nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
5a022fc8 177 const char *name, unsigned int len,
6c0a654d 178 struct svc_export **exp_ret, struct dentry **dentry_ret)
1da177e4
LT
179{
180 struct svc_export *exp;
181 struct dentry *dparent;
182 struct dentry *dentry;
6264d69d
AV
183 __be32 err;
184 int host_err;
1da177e4
LT
185
186 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
187
188 /* Obtain dentry and export. */
8837abca 189 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1da177e4
LT
190 if (err)
191 return err;
192
193 dparent = fhp->fh_dentry;
194 exp = fhp->fh_export;
195 exp_get(exp);
196
1da177e4
LT
197 /* Lookup the name, but don't follow links */
198 if (isdotent(name, len)) {
199 if (len==1)
200 dentry = dget(dparent);
54775491 201 else if (dparent != exp->ex_path.dentry)
1da177e4 202 dentry = dget_parent(dparent);
fed83811 203 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
1da177e4
LT
204 dentry = dget(dparent); /* .. == . just like at / */
205 else {
206 /* checking mountpoint crossing is very different when stepping up */
289ede45
BF
207 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
208 if (host_err)
1da177e4 209 goto out_nfserr;
1da177e4
LT
210 }
211 } else {
212 fh_lock(fhp);
213 dentry = lookup_one_len(name, dparent, len);
6264d69d 214 host_err = PTR_ERR(dentry);
1da177e4
LT
215 if (IS_ERR(dentry))
216 goto out_nfserr;
217 /*
218 * check if we have crossed a mount point ...
219 */
82ead7fe 220 if (nfsd_mountpoint(dentry, exp)) {
6264d69d 221 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
1da177e4
LT
222 dput(dentry);
223 goto out_nfserr;
224 }
225 }
226 }
6c0a654d
BF
227 *dentry_ret = dentry;
228 *exp_ret = exp;
229 return 0;
230
231out_nfserr:
232 exp_put(exp);
233 return nfserrno(host_err);
234}
235
236/*
237 * Look up one component of a pathname.
238 * N.B. After this call _both_ fhp and resfh need an fh_put
239 *
240 * If the lookup would cross a mountpoint, and the mounted filesystem
241 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
242 * accepted as it stands and the mounted directory is
243 * returned. Otherwise the covered directory is returned.
244 * NOTE: this mountpoint crossing is not supported properly by all
245 * clients and is explicitly disallowed for NFSv3
246 * NeilBrown <neilb@cse.unsw.edu.au>
247 */
248__be32
249nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
5a022fc8 250 unsigned int len, struct svc_fh *resfh)
6c0a654d
BF
251{
252 struct svc_export *exp;
253 struct dentry *dentry;
254 __be32 err;
255
256 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
257 if (err)
258 return err;
32c1eb0c
AA
259 err = check_nfsd_access(exp, rqstp);
260 if (err)
261 goto out;
1da177e4
LT
262 /*
263 * Note: we compose the file handle now, but as the
264 * dentry may be negative, it may need to be updated.
265 */
266 err = fh_compose(resfh, exp, dentry, fhp);
267 if (!err && !dentry->d_inode)
268 err = nfserr_noent;
32c1eb0c 269out:
1da177e4 270 dput(dentry);
1da177e4
LT
271 exp_put(exp);
272 return err;
1da177e4
LT
273}
274
f501912a
BM
275/*
276 * Commit metadata changes to stable storage.
277 */
278static int
279commit_metadata(struct svc_fh *fhp)
280{
281 struct inode *inode = fhp->fh_dentry->d_inode;
282 const struct export_operations *export_ops = inode->i_sb->s_export_op;
283 int error = 0;
284
285 if (!EX_ISSYNC(fhp->fh_export))
286 return 0;
287
288 if (export_ops->commit_metadata) {
289 error = export_ops->commit_metadata(inode);
290 } else {
291 struct writeback_control wbc = {
292 .sync_mode = WB_SYNC_ALL,
293 .nr_to_write = 0, /* metadata only */
294 };
295
296 error = sync_inode(inode, &wbc);
297 }
298
299 return error;
300}
6c0a654d 301
1da177e4
LT
302/*
303 * Set various file attributes.
304 * N.B. After this call fhp needs an fh_put
305 */
6264d69d 306__be32
1da177e4
LT
307nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
308 int check_guard, time_t guardtime)
309{
310 struct dentry *dentry;
311 struct inode *inode;
8837abca 312 int accmode = NFSD_MAY_SATTR;
1da177e4 313 int ftype = 0;
6264d69d
AV
314 __be32 err;
315 int host_err;
1da177e4
LT
316 int size_change = 0;
317
318 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
8837abca 319 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
1da177e4
LT
320 if (iap->ia_valid & ATTR_SIZE)
321 ftype = S_IFREG;
322
323 /* Get inode */
324 err = fh_verify(rqstp, fhp, ftype, accmode);
15b7a1b8 325 if (err)
1da177e4
LT
326 goto out;
327
328 dentry = fhp->fh_dentry;
329 inode = dentry->d_inode;
330
15b7a1b8
N
331 /* Ignore any mode updates on symlinks */
332 if (S_ISLNK(inode->i_mode))
333 iap->ia_valid &= ~ATTR_MODE;
334
335 if (!iap->ia_valid)
336 goto out;
337
9c85fca5
CH
338 /*
339 * NFSv2 does not differentiate between "set-[ac]time-to-now"
1da177e4
LT
340 * which only requires access, and "set-[ac]time-to-X" which
341 * requires ownership.
342 * So if it looks like it might be "set both to the same time which
343 * is close to now", and if inode_change_ok fails, then we
344 * convert to "set to now" instead of "set to explicit time"
345 *
346 * We only call inode_change_ok as the last test as technically
347 * it is not an interface that we should be using. It is only
348 * valid if the filesystem does not define it's own i_op->setattr.
349 */
350#define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
351#define MAX_TOUCH_TIME_ERROR (30*60)
9c85fca5
CH
352 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
353 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
354 /*
355 * Looks probable.
356 *
357 * Now just make sure time is in the right ballpark.
358 * Solaris, at least, doesn't seem to care what the time
359 * request is. We require it be within 30 minutes of now.
1da177e4 360 */
9c85fca5
CH
361 time_t delta = iap->ia_atime.tv_sec - get_seconds();
362 if (delta < 0)
363 delta = -delta;
364 if (delta < MAX_TOUCH_TIME_ERROR &&
365 inode_change_ok(inode, iap) != 0) {
366 /*
367 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
368 * This will cause notify_change to set these times
369 * to "now"
370 */
371 iap->ia_valid &= ~BOTH_TIME_SET;
372 }
1da177e4
LT
373 }
374
9c85fca5
CH
375 /*
376 * The size case is special.
377 * It changes the file as well as the attributes.
378 */
1da177e4
LT
379 if (iap->ia_valid & ATTR_SIZE) {
380 if (iap->ia_size < inode->i_size) {
8837abca
MS
381 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
382 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
1da177e4
LT
383 if (err)
384 goto out;
385 }
386
387 /*
388 * If we are changing the size of the file, then
389 * we need to break all leases.
390 */
8737c930 391 host_err = break_lease(inode, O_WRONLY | O_NONBLOCK);
6264d69d
AV
392 if (host_err == -EWOULDBLOCK)
393 host_err = -ETIMEDOUT;
394 if (host_err) /* ENOMEM or EWOULDBLOCK */
1da177e4
LT
395 goto out_nfserr;
396
6264d69d
AV
397 host_err = get_write_access(inode);
398 if (host_err)
1da177e4
LT
399 goto out_nfserr;
400
401 size_change = 1;
6264d69d
AV
402 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
403 if (host_err) {
1da177e4
LT
404 put_write_access(inode);
405 goto out_nfserr;
406 }
1da177e4
LT
407 }
408
ca456252 409 /* sanitize the mode change */
1da177e4
LT
410 if (iap->ia_valid & ATTR_MODE) {
411 iap->ia_mode &= S_IALLUGO;
dee3209d 412 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
ca456252
JL
413 }
414
415 /* Revoke setuid/setgid on chown */
0953e620
SP
416 if (!S_ISDIR(inode->i_mode) &&
417 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
418 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
ca456252
JL
419 iap->ia_valid |= ATTR_KILL_PRIV;
420 if (iap->ia_valid & ATTR_MODE) {
421 /* we're setting mode too, just clear the s*id bits */
8a0ce7d9 422 iap->ia_mode &= ~S_ISUID;
ca456252
JL
423 if (iap->ia_mode & S_IXGRP)
424 iap->ia_mode &= ~S_ISGID;
425 } else {
426 /* set ATTR_KILL_* bits and let VFS handle it */
427 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
8a0ce7d9 428 }
1da177e4
LT
429 }
430
1da177e4
LT
431 /* Change the attributes. */
432
433 iap->ia_valid |= ATTR_CTIME;
434
435 err = nfserr_notsync;
436 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
437 fh_lock(fhp);
6264d69d
AV
438 host_err = notify_change(dentry, iap);
439 err = nfserrno(host_err);
1da177e4
LT
440 fh_unlock(fhp);
441 }
442 if (size_change)
443 put_write_access(inode);
444 if (!err)
445 if (EX_ISSYNC(fhp->fh_export))
446 write_inode_now(inode, 1);
447out:
448 return err;
449
450out_nfserr:
6264d69d 451 err = nfserrno(host_err);
1da177e4
LT
452 goto out;
453}
454
5be196e5
CH
455#if defined(CONFIG_NFSD_V2_ACL) || \
456 defined(CONFIG_NFSD_V3_ACL) || \
457 defined(CONFIG_NFSD_V4)
458static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
459{
460 ssize_t buflen;
6c6a426f 461 ssize_t ret;
5be196e5
CH
462
463 buflen = vfs_getxattr(dentry, key, NULL, 0);
464 if (buflen <= 0)
465 return buflen;
1da177e4 466
5be196e5
CH
467 *buf = kmalloc(buflen, GFP_KERNEL);
468 if (!*buf)
469 return -ENOMEM;
470
6c6a426f
KK
471 ret = vfs_getxattr(dentry, key, *buf, buflen);
472 if (ret < 0)
473 kfree(*buf);
474 return ret;
5be196e5
CH
475}
476#endif
477
478#if defined(CONFIG_NFSD_V4)
1da177e4
LT
479static int
480set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
481{
482 int len;
483 size_t buflen;
484 char *buf = NULL;
485 int error = 0;
1da177e4
LT
486
487 buflen = posix_acl_xattr_size(pacl->a_count);
488 buf = kmalloc(buflen, GFP_KERNEL);
489 error = -ENOMEM;
490 if (buf == NULL)
491 goto out;
492
493 len = posix_acl_to_xattr(pacl, buf, buflen);
494 if (len < 0) {
495 error = len;
496 goto out;
497 }
498
5be196e5 499 error = vfs_setxattr(dentry, key, buf, len, 0);
1da177e4
LT
500out:
501 kfree(buf);
502 return error;
503}
504
6264d69d 505__be32
1da177e4
LT
506nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
507 struct nfs4_acl *acl)
508{
6264d69d
AV
509 __be32 error;
510 int host_error;
1da177e4
LT
511 struct dentry *dentry;
512 struct inode *inode;
513 struct posix_acl *pacl = NULL, *dpacl = NULL;
514 unsigned int flags = 0;
515
516 /* Get inode */
8837abca 517 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
1da177e4 518 if (error)
4b2ca38a 519 return error;
1da177e4
LT
520
521 dentry = fhp->fh_dentry;
522 inode = dentry->d_inode;
523 if (S_ISDIR(inode->i_mode))
524 flags = NFS4_ACL_DIR;
525
6264d69d
AV
526 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
527 if (host_error == -EINVAL) {
4b2ca38a 528 return nfserr_attrnotsupp;
6264d69d 529 } else if (host_error < 0)
1da177e4
LT
530 goto out_nfserr;
531
6264d69d
AV
532 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
533 if (host_error < 0)
4b2ca38a 534 goto out_release;
1da177e4 535
4b2ca38a 536 if (S_ISDIR(inode->i_mode))
6264d69d 537 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
1da177e4 538
4b2ca38a 539out_release:
1da177e4
LT
540 posix_acl_release(pacl);
541 posix_acl_release(dpacl);
1da177e4 542out_nfserr:
f34f9242 543 if (host_error == -EOPNOTSUPP)
4b2ca38a 544 return nfserr_attrnotsupp;
f34f9242 545 else
4b2ca38a 546 return nfserrno(host_error);
1da177e4
LT
547}
548
549static struct posix_acl *
550_get_posix_acl(struct dentry *dentry, char *key)
551{
5be196e5 552 void *buf = NULL;
1da177e4 553 struct posix_acl *pacl = NULL;
5be196e5 554 int buflen;
1da177e4 555
5be196e5
CH
556 buflen = nfsd_getxattr(dentry, key, &buf);
557 if (!buflen)
558 buflen = -ENODATA;
559 if (buflen <= 0)
560 return ERR_PTR(buflen);
1da177e4
LT
561
562 pacl = posix_acl_from_xattr(buf, buflen);
1da177e4
LT
563 kfree(buf);
564 return pacl;
1da177e4
LT
565}
566
567int
568nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
569{
570 struct inode *inode = dentry->d_inode;
571 int error = 0;
572 struct posix_acl *pacl = NULL, *dpacl = NULL;
573 unsigned int flags = 0;
574
9a59f452 575 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
1da177e4
LT
576 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
577 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
578 if (IS_ERR(pacl)) {
579 error = PTR_ERR(pacl);
580 pacl = NULL;
581 goto out;
582 }
583
584 if (S_ISDIR(inode->i_mode)) {
9a59f452 585 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
1da177e4
LT
586 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
587 dpacl = NULL;
588 else if (IS_ERR(dpacl)) {
589 error = PTR_ERR(dpacl);
590 dpacl = NULL;
591 goto out;
592 }
593 flags = NFS4_ACL_DIR;
594 }
595
596 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
597 if (IS_ERR(*acl)) {
598 error = PTR_ERR(*acl);
599 *acl = NULL;
600 }
601 out:
602 posix_acl_release(pacl);
603 posix_acl_release(dpacl);
604 return error;
605}
606
607#endif /* defined(CONFIG_NFS_V4) */
608
609#ifdef CONFIG_NFSD_V3
610/*
611 * Check server access rights to a file system object
612 */
613struct accessmap {
614 u32 access;
615 int how;
616};
617static struct accessmap nfs3_regaccess[] = {
8837abca
MS
618 { NFS3_ACCESS_READ, NFSD_MAY_READ },
619 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
620 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
621 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
1da177e4
LT
622
623 { 0, 0 }
624};
625
626static struct accessmap nfs3_diraccess[] = {
8837abca
MS
627 { NFS3_ACCESS_READ, NFSD_MAY_READ },
628 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
629 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
630 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
631 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
1da177e4
LT
632
633 { 0, 0 }
634};
635
636static struct accessmap nfs3_anyaccess[] = {
637 /* Some clients - Solaris 2.6 at least, make an access call
638 * to the server to check for access for things like /dev/null
639 * (which really, the server doesn't care about). So
640 * We provide simple access checking for them, looking
641 * mainly at mode bits, and we make sure to ignore read-only
642 * filesystem checks
643 */
8837abca
MS
644 { NFS3_ACCESS_READ, NFSD_MAY_READ },
645 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
646 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
647 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
1da177e4
LT
648
649 { 0, 0 }
650};
651
6264d69d 652__be32
1da177e4
LT
653nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
654{
655 struct accessmap *map;
656 struct svc_export *export;
657 struct dentry *dentry;
658 u32 query, result = 0, sresult = 0;
6264d69d 659 __be32 error;
1da177e4 660
8837abca 661 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
1da177e4
LT
662 if (error)
663 goto out;
664
665 export = fhp->fh_export;
666 dentry = fhp->fh_dentry;
667
668 if (S_ISREG(dentry->d_inode->i_mode))
669 map = nfs3_regaccess;
670 else if (S_ISDIR(dentry->d_inode->i_mode))
671 map = nfs3_diraccess;
672 else
673 map = nfs3_anyaccess;
674
675
676 query = *access;
677 for (; map->access; map++) {
678 if (map->access & query) {
6264d69d 679 __be32 err2;
1da177e4
LT
680
681 sresult |= map->access;
682
0ec757df 683 err2 = nfsd_permission(rqstp, export, dentry, map->how);
1da177e4
LT
684 switch (err2) {
685 case nfs_ok:
686 result |= map->access;
687 break;
688
689 /* the following error codes just mean the access was not allowed,
690 * rather than an error occurred */
691 case nfserr_rofs:
692 case nfserr_acces:
693 case nfserr_perm:
694 /* simply don't "or" in the access bit. */
695 break;
696 default:
697 error = err2;
698 goto out;
699 }
700 }
701 }
702 *access = result;
703 if (supported)
704 *supported = sresult;
705
706 out:
707 return error;
708}
709#endif /* CONFIG_NFSD_V3 */
710
711
712
713/*
714 * Open an existing file or directory.
715 * The access argument indicates the type of open (read/write/lock)
716 * N.B. After this call fhp needs an fh_put
717 */
6264d69d 718__be32
1da177e4
LT
719nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
720 int access, struct file **filp)
721{
722 struct dentry *dentry;
723 struct inode *inode;
6264d69d
AV
724 int flags = O_RDONLY|O_LARGEFILE;
725 __be32 err;
726 int host_err;
1da177e4 727
e0e81739
DH
728 validate_process_creds();
729
1da177e4
LT
730 /*
731 * If we get here, then the client has already done an "open",
732 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
733 * in case a chmod has now revoked permission.
734 */
8837abca 735 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
1da177e4
LT
736 if (err)
737 goto out;
738
739 dentry = fhp->fh_dentry;
740 inode = dentry->d_inode;
741
742 /* Disallow write access to files with the append-only bit set
743 * or any access when mandatory locking enabled
744 */
745 err = nfserr_perm;
8837abca 746 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
1da177e4 747 goto out;
5e7fc436
BF
748 /*
749 * We must ignore files (but only files) which might have mandatory
750 * locks on them because there is no way to know if the accesser has
751 * the lock.
752 */
753 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
1da177e4
LT
754 goto out;
755
756 if (!inode->i_fop)
757 goto out;
758
759 /*
760 * Check to see if there are any leases on this file.
761 * This may block while leases are broken.
762 */
8737c930 763 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
6264d69d
AV
764 if (host_err == -EWOULDBLOCK)
765 host_err = -ETIMEDOUT;
766 if (host_err) /* NOMEM or WOULDBLOCK */
1da177e4
LT
767 goto out_nfserr;
768
8837abca
MS
769 if (access & NFSD_MAY_WRITE) {
770 if (access & NFSD_MAY_READ)
9ecb6a08
BF
771 flags = O_RDWR|O_LARGEFILE;
772 else
773 flags = O_WRONLY|O_LARGEFILE;
1da177e4 774 }
54775491 775 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
033a666c 776 flags, current_cred());
1da177e4 777 if (IS_ERR(*filp))
6264d69d 778 host_err = PTR_ERR(*filp);
aeaa5ccd
CE
779 else
780 host_err = ima_file_check(*filp, access);
1da177e4 781out_nfserr:
6264d69d 782 err = nfserrno(host_err);
1da177e4 783out:
e0e81739 784 validate_process_creds();
1da177e4
LT
785 return err;
786}
787
788/*
789 * Close a file.
790 */
791void
792nfsd_close(struct file *filp)
793{
794 fput(filp);
795}
796
1da177e4
LT
797/*
798 * Obtain the readahead parameters for the file
799 * specified by (dev, ino).
800 */
1da177e4
LT
801
802static inline struct raparms *
803nfsd_get_raparms(dev_t dev, ino_t ino)
804{
805 struct raparms *ra, **rap, **frap = NULL;
806 int depth = 0;
fce1456a
GB
807 unsigned int hash;
808 struct raparm_hbucket *rab;
809
810 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
811 rab = &raparm_hash[hash];
1da177e4 812
fce1456a
GB
813 spin_lock(&rab->pb_lock);
814 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
1da177e4
LT
815 if (ra->p_ino == ino && ra->p_dev == dev)
816 goto found;
817 depth++;
818 if (ra->p_count == 0)
819 frap = rap;
820 }
821 depth = nfsdstats.ra_size*11/10;
822 if (!frap) {
fce1456a 823 spin_unlock(&rab->pb_lock);
1da177e4
LT
824 return NULL;
825 }
826 rap = frap;
827 ra = *frap;
828 ra->p_dev = dev;
829 ra->p_ino = ino;
830 ra->p_set = 0;
fce1456a 831 ra->p_hindex = hash;
1da177e4 832found:
fce1456a 833 if (rap != &rab->pb_head) {
1da177e4 834 *rap = ra->p_next;
fce1456a
GB
835 ra->p_next = rab->pb_head;
836 rab->pb_head = ra;
1da177e4
LT
837 }
838 ra->p_count++;
839 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
fce1456a 840 spin_unlock(&rab->pb_lock);
1da177e4
LT
841 return ra;
842}
843
844/*
cf8208d0
JA
845 * Grab and keep cached pages associated with a file in the svc_rqst
846 * so that they can be passed to the network sendmsg/sendpage routines
847 * directly. They will be released after the sending has completed.
1da177e4
LT
848 */
849static int
cf8208d0
JA
850nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
851 struct splice_desc *sd)
1da177e4 852{
cf8208d0 853 struct svc_rqst *rqstp = sd->u.data;
44524359 854 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
cf8208d0
JA
855 struct page *page = buf->page;
856 size_t size;
857 int ret;
1da177e4 858
cac36bb0 859 ret = buf->ops->confirm(pipe, buf);
cf8208d0
JA
860 if (unlikely(ret))
861 return ret;
862
863 size = sd->len;
1da177e4
LT
864
865 if (rqstp->rq_res.page_len == 0) {
866 get_page(page);
44524359
N
867 put_page(*pp);
868 *pp = page;
869 rqstp->rq_resused++;
cf8208d0 870 rqstp->rq_res.page_base = buf->offset;
1da177e4 871 rqstp->rq_res.page_len = size;
44524359 872 } else if (page != pp[-1]) {
1da177e4 873 get_page(page);
250f3915
N
874 if (*pp)
875 put_page(*pp);
44524359
N
876 *pp = page;
877 rqstp->rq_resused++;
1da177e4 878 rqstp->rq_res.page_len += size;
44524359 879 } else
1da177e4 880 rqstp->rq_res.page_len += size;
1da177e4 881
1da177e4
LT
882 return size;
883}
884
cf8208d0
JA
885static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
886 struct splice_desc *sd)
887{
888 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
889}
890
a8754bee
DH
891static inline int svc_msnfs(struct svc_fh *ffhp)
892{
893#ifdef MSNFS
894 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
895#else
896 return 0;
897#endif
898}
899
6264d69d 900static __be32
1da177e4
LT
901nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
902 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
903{
904 struct inode *inode;
905 struct raparms *ra;
906 mm_segment_t oldfs;
6264d69d
AV
907 __be32 err;
908 int host_err;
1da177e4
LT
909
910 err = nfserr_perm;
7eaa36e2 911 inode = file->f_path.dentry->d_inode;
a8754bee
DH
912
913 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
1da177e4 914 goto out;
1da177e4
LT
915
916 /* Get readahead parameters */
917 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
918
919 if (ra && ra->p_set)
920 file->f_ra = ra->p_ra;
921
cf8208d0
JA
922 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
923 struct splice_desc sd = {
924 .len = 0,
925 .total_len = *count,
926 .pos = offset,
927 .u.data = rqstp,
928 };
929
4fbef206 930 rqstp->rq_resused = 1;
cf8208d0 931 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
1da177e4
LT
932 } else {
933 oldfs = get_fs();
934 set_fs(KERNEL_DS);
6264d69d 935 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
1da177e4
LT
936 set_fs(oldfs);
937 }
938
939 /* Write back readahead params */
940 if (ra) {
fce1456a
GB
941 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
942 spin_lock(&rab->pb_lock);
1da177e4
LT
943 ra->p_ra = file->f_ra;
944 ra->p_set = 1;
945 ra->p_count--;
fce1456a 946 spin_unlock(&rab->pb_lock);
1da177e4
LT
947 }
948
6264d69d
AV
949 if (host_err >= 0) {
950 nfsdstats.io_read += host_err;
951 *count = host_err;
1da177e4 952 err = 0;
7eaa36e2 953 fsnotify_access(file->f_path.dentry);
1da177e4 954 } else
6264d69d 955 err = nfserrno(host_err);
1da177e4
LT
956out:
957 return err;
958}
959
9f708e40
NB
960static void kill_suid(struct dentry *dentry)
961{
962 struct iattr ia;
b5376771 963 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
9f708e40 964
1b1dcc1b 965 mutex_lock(&dentry->d_inode->i_mutex);
9f708e40 966 notify_change(dentry, &ia);
1b1dcc1b 967 mutex_unlock(&dentry->d_inode->i_mutex);
9f708e40
NB
968}
969
d911df7b
BF
970/*
971 * Gathered writes: If another process is currently writing to the file,
972 * there's a high chance this is another nfsd (triggered by a bulk write
973 * from a client's biod). Rather than syncing the file with each write
974 * request, we sleep for 10 msec.
975 *
976 * I don't know if this roughly approximates C. Juszak's idea of
977 * gathered writes, but it's a nice and simple solution (IMHO), and it
978 * seems to work:-)
979 *
980 * Note: we do this only in the NFSv2 case, since v3 and higher have a
981 * better tool (separate unstable writes and commits) for solving this
982 * problem.
983 */
984static int wait_for_concurrent_writes(struct file *file)
985{
986 struct inode *inode = file->f_path.dentry->d_inode;
987 static ino_t last_ino;
988 static dev_t last_dev;
989 int err = 0;
990
991 if (atomic_read(&inode->i_writecount) > 1
992 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
993 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
994 msleep(10);
995 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
996 }
997
998 if (inode->i_state & I_DIRTY) {
999 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
6a68f89e 1000 err = vfs_fsync(file, file->f_path.dentry, 0);
d911df7b
BF
1001 }
1002 last_ino = inode->i_ino;
1003 last_dev = inode->i_sb->s_dev;
1004 return err;
1005}
1006
6264d69d 1007static __be32
1da177e4
LT
1008nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1009 loff_t offset, struct kvec *vec, int vlen,
31dec253 1010 unsigned long *cnt, int *stablep)
1da177e4
LT
1011{
1012 struct svc_export *exp;
1013 struct dentry *dentry;
1014 struct inode *inode;
1015 mm_segment_t oldfs;
6264d69d
AV
1016 __be32 err = 0;
1017 int host_err;
1da177e4 1018 int stable = *stablep;
48e03bc5 1019 int use_wgather;
1da177e4 1020
45bd3b3d 1021#ifdef MSNFS
1da177e4
LT
1022 err = nfserr_perm;
1023
1da177e4 1024 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
31dec253 1025 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1da177e4
LT
1026 goto out;
1027#endif
1028
7eaa36e2 1029 dentry = file->f_path.dentry;
1da177e4
LT
1030 inode = dentry->d_inode;
1031 exp = fhp->fh_export;
1032
1033 /*
1034 * Request sync writes if
1035 * - the sync export option has been set, or
1036 * - the client requested O_SYNC behavior (NFSv3 feature).
1037 * - The file system doesn't support fsync().
48e03bc5 1038 * When NFSv2 gathered writes have been configured for this volume,
1da177e4
LT
1039 * flushing the data to disk is handled separately below.
1040 */
48e03bc5 1041 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1da177e4 1042
3ba15148 1043 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1da177e4
LT
1044 stable = 2;
1045 *stablep = 2; /* FILE_SYNC */
1046 }
1047
1048 if (!EX_ISSYNC(exp))
1049 stable = 0;
48e03bc5 1050 if (stable && !use_wgather) {
db1dd4d3 1051 spin_lock(&file->f_lock);
1da177e4 1052 file->f_flags |= O_SYNC;
db1dd4d3
JC
1053 spin_unlock(&file->f_lock);
1054 }
1da177e4
LT
1055
1056 /* Write the data. */
1057 oldfs = get_fs(); set_fs(KERNEL_DS);
6264d69d 1058 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1da177e4 1059 set_fs(oldfs);
e4636d53
BF
1060 if (host_err < 0)
1061 goto out_nfserr;
1062 *cnt = host_err;
1063 nfsdstats.io_write += host_err;
1064 fsnotify_modify(file->f_path.dentry);
1da177e4
LT
1065
1066 /* clear setuid/setgid flag after write */
e4636d53 1067 if (inode->i_mode & (S_ISUID | S_ISGID))
9f708e40 1068 kill_suid(dentry);
1da177e4 1069
e4636d53 1070 if (stable && use_wgather)
d911df7b 1071 host_err = wait_for_concurrent_writes(file);
1da177e4 1072
e4636d53 1073out_nfserr:
6264d69d 1074 dprintk("nfsd: write complete host_err=%d\n", host_err);
a0d24b29 1075 if (host_err >= 0)
1da177e4 1076 err = 0;
a0d24b29 1077 else
6264d69d 1078 err = nfserrno(host_err);
1da177e4
LT
1079out:
1080 return err;
1081}
1082
1083/*
1084 * Read data from a file. count must contain the requested read count
1085 * on entry. On return, *count contains the number of bytes actually read.
1086 * N.B. After this call fhp needs an fh_put
1087 */
6264d69d 1088__be32
1da177e4
LT
1089nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1090 loff_t offset, struct kvec *vec, int vlen,
1091 unsigned long *count)
1092{
6264d69d 1093 __be32 err;
1da177e4
LT
1094
1095 if (file) {
0ec757df 1096 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
8837abca 1097 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1da177e4
LT
1098 if (err)
1099 goto out;
1100 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1101 } else {
8837abca 1102 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1da177e4
LT
1103 if (err)
1104 goto out;
1105 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1106 nfsd_close(file);
1107 }
1108out:
1109 return err;
1110}
1111
1112/*
1113 * Write data to a file.
1114 * The stable flag requests synchronous writes.
1115 * N.B. After this call fhp needs an fh_put
1116 */
6264d69d 1117__be32
1da177e4 1118nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
31dec253 1119 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1da177e4
LT
1120 int *stablep)
1121{
6264d69d 1122 __be32 err = 0;
1da177e4
LT
1123
1124 if (file) {
0ec757df 1125 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
8837abca 1126 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1da177e4
LT
1127 if (err)
1128 goto out;
1129 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1130 stablep);
1131 } else {
8837abca 1132 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1da177e4
LT
1133 if (err)
1134 goto out;
1135
1136 if (cnt)
1137 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1138 cnt, stablep);
1139 nfsd_close(file);
1140 }
1141out:
1142 return err;
1143}
1144
1145#ifdef CONFIG_NFSD_V3
1146/*
1147 * Commit all pending writes to stable storage.
aa696a6f
TM
1148 *
1149 * Note: we only guarantee that data that lies within the range specified
1150 * by the 'offset' and 'count' parameters will be synced.
1da177e4
LT
1151 *
1152 * Unfortunately we cannot lock the file to make sure we return full WCC
1153 * data to the client, as locking happens lower down in the filesystem.
1154 */
6264d69d 1155__be32
1da177e4
LT
1156nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1157 loff_t offset, unsigned long count)
1158{
1159 struct file *file;
aa696a6f
TM
1160 loff_t end = LLONG_MAX;
1161 __be32 err = nfserr_inval;
1da177e4 1162
aa696a6f
TM
1163 if (offset < 0)
1164 goto out;
1165 if (count != 0) {
1166 end = offset + (loff_t)count - 1;
1167 if (end < offset)
1168 goto out;
1169 }
1da177e4 1170
8837abca
MS
1171 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1172 if (err)
aa696a6f 1173 goto out;
1da177e4 1174 if (EX_ISSYNC(fhp->fh_export)) {
aa696a6f
TM
1175 int err2 = vfs_fsync_range(file, file->f_path.dentry,
1176 offset, end, 0);
1177
1178 if (err2 != -EINVAL)
1179 err = nfserrno(err2);
1180 else
1da177e4 1181 err = nfserr_notsupp;
1da177e4
LT
1182 }
1183
1184 nfsd_close(file);
aa696a6f 1185out:
1da177e4
LT
1186 return err;
1187}
1188#endif /* CONFIG_NFSD_V3 */
1189
f2b0dee2 1190static __be32
5c002b3b
BF
1191nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1192 struct iattr *iap)
1193{
1194 /*
1195 * Mode has already been set earlier in create:
1196 */
1197 iap->ia_valid &= ~ATTR_MODE;
1198 /*
1199 * Setting uid/gid works only for root. Irix appears to
1200 * send along the gid on create when it tries to implement
1201 * setgid directories via NFS:
1202 */
5cc0a840 1203 if (current_fsuid() != 0)
5c002b3b
BF
1204 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1205 if (iap->ia_valid)
1206 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1207 return 0;
1208}
1209
4ac35c2f 1210/* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1211 * setting size to 0 may fail for some specific file systems by the permission
1212 * checking which requires WRITE permission but the mode is 000.
1213 * we ignore the resizing(to 0) on the just new created file, since the size is
1214 * 0 after file created.
1215 *
1216 * call this only after vfs_create() is called.
1217 * */
1218static void
1219nfsd_check_ignore_resizing(struct iattr *iap)
1220{
1221 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1222 iap->ia_valid &= ~ATTR_SIZE;
1223}
1224
1da177e4
LT
1225/*
1226 * Create a file (regular, directory, device, fifo); UNIX sockets
1227 * not yet implemented.
1228 * If the response fh has been verified, the parent directory should
1229 * already be locked. Note that the parent directory is left locked.
1230 *
1231 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1232 */
6264d69d 1233__be32
1da177e4
LT
1234nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1235 char *fname, int flen, struct iattr *iap,
1236 int type, dev_t rdev, struct svc_fh *resfhp)
1237{
1238 struct dentry *dentry, *dchild = NULL;
1239 struct inode *dirp;
6264d69d 1240 __be32 err;
5c002b3b 1241 __be32 err2;
6264d69d 1242 int host_err;
1da177e4
LT
1243
1244 err = nfserr_perm;
1245 if (!flen)
1246 goto out;
1247 err = nfserr_exist;
1248 if (isdotent(fname, flen))
1249 goto out;
1250
8837abca 1251 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1da177e4
LT
1252 if (err)
1253 goto out;
1254
1255 dentry = fhp->fh_dentry;
1256 dirp = dentry->d_inode;
1257
1258 err = nfserr_notdir;
acfa4380 1259 if (!dirp->i_op->lookup)
1da177e4
LT
1260 goto out;
1261 /*
1262 * Check whether the response file handle has been verified yet.
1263 * If it has, the parent directory should already be locked.
1264 */
1265 if (!resfhp->fh_dentry) {
1266 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
12fd3520 1267 fh_lock_nested(fhp, I_MUTEX_PARENT);
1da177e4 1268 dchild = lookup_one_len(fname, dentry, flen);
6264d69d 1269 host_err = PTR_ERR(dchild);
1da177e4
LT
1270 if (IS_ERR(dchild))
1271 goto out_nfserr;
1272 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1273 if (err)
1274 goto out;
1275 } else {
1276 /* called from nfsd_proc_create */
1277 dchild = dget(resfhp->fh_dentry);
1278 if (!fhp->fh_locked) {
1279 /* not actually possible */
1280 printk(KERN_ERR
1281 "nfsd_create: parent %s/%s not locked!\n",
1282 dentry->d_parent->d_name.name,
1283 dentry->d_name.name);
d75f2b9f 1284 err = nfserr_io;
1da177e4
LT
1285 goto out;
1286 }
1287 }
1288 /*
1289 * Make sure the child dentry is still negative ...
1290 */
1291 err = nfserr_exist;
1292 if (dchild->d_inode) {
1293 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1294 dentry->d_name.name, dchild->d_name.name);
1295 goto out;
1296 }
1297
1298 if (!(iap->ia_valid & ATTR_MODE))
1299 iap->ia_mode = 0;
1300 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1301
07cad1d2
MS
1302 err = nfserr_inval;
1303 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1304 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1305 type);
1306 goto out;
1307 }
1308
1309 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1310 if (host_err)
1311 goto out_nfserr;
1312
1da177e4
LT
1313 /*
1314 * Get the dir op function pointer.
1315 */
088406bc 1316 err = 0;
1da177e4
LT
1317 switch (type) {
1318 case S_IFREG:
6264d69d 1319 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
4ac35c2f 1320 if (!host_err)
1321 nfsd_check_ignore_resizing(iap);
1da177e4
LT
1322 break;
1323 case S_IFDIR:
6264d69d 1324 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1da177e4
LT
1325 break;
1326 case S_IFCHR:
1327 case S_IFBLK:
1328 case S_IFIFO:
1329 case S_IFSOCK:
6264d69d 1330 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1da177e4 1331 break;
1da177e4 1332 }
463c3197
DH
1333 if (host_err < 0) {
1334 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1da177e4 1335 goto out_nfserr;
463c3197 1336 }
1da177e4 1337
f501912a 1338 err = nfsd_create_setattr(rqstp, resfhp, iap);
1da177e4 1339
f501912a
BM
1340 /*
1341 * nfsd_setattr already committed the child. Transactional filesystems
1342 * had a chance to commit changes for both parent and child
1343 * simultaneously making the following commit_metadata a noop.
1344 */
1345 err2 = nfserrno(commit_metadata(fhp));
5c002b3b
BF
1346 if (err2)
1347 err = err2;
463c3197 1348 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1da177e4
LT
1349 /*
1350 * Update the file handle to get the new inode info.
1351 */
1352 if (!err)
1353 err = fh_update(resfhp);
1354out:
1355 if (dchild && !IS_ERR(dchild))
1356 dput(dchild);
1357 return err;
1358
1359out_nfserr:
6264d69d 1360 err = nfserrno(host_err);
1da177e4
LT
1361 goto out;
1362}
1363
1364#ifdef CONFIG_NFSD_V3
1365/*
1366 * NFSv3 version of nfsd_create
1367 */
6264d69d 1368__be32
1da177e4
LT
1369nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1370 char *fname, int flen, struct iattr *iap,
1371 struct svc_fh *resfhp, int createmode, u32 *verifier,
81ac95c5 1372 int *truncp, int *created)
1da177e4
LT
1373{
1374 struct dentry *dentry, *dchild = NULL;
1375 struct inode *dirp;
6264d69d
AV
1376 __be32 err;
1377 int host_err;
1da177e4 1378 __u32 v_mtime=0, v_atime=0;
1da177e4
LT
1379
1380 err = nfserr_perm;
1381 if (!flen)
1382 goto out;
1383 err = nfserr_exist;
1384 if (isdotent(fname, flen))
1385 goto out;
1386 if (!(iap->ia_valid & ATTR_MODE))
1387 iap->ia_mode = 0;
8837abca 1388 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1da177e4
LT
1389 if (err)
1390 goto out;
1391
1392 dentry = fhp->fh_dentry;
1393 dirp = dentry->d_inode;
1394
1395 /* Get all the sanity checks out of the way before
1396 * we lock the parent. */
1397 err = nfserr_notdir;
acfa4380 1398 if (!dirp->i_op->lookup)
1da177e4 1399 goto out;
12fd3520 1400 fh_lock_nested(fhp, I_MUTEX_PARENT);
1da177e4
LT
1401
1402 /*
1403 * Compose the response file handle.
1404 */
1405 dchild = lookup_one_len(fname, dentry, flen);
6264d69d 1406 host_err = PTR_ERR(dchild);
1da177e4
LT
1407 if (IS_ERR(dchild))
1408 goto out_nfserr;
1409
1410 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1411 if (err)
1412 goto out;
1413
1414 if (createmode == NFS3_CREATE_EXCLUSIVE) {
c397852c 1415 /* solaris7 gets confused (bugid 4218508) if these have
749997e5
JL
1416 * the high bit set, so just clear the high bits. If this is
1417 * ever changed to use different attrs for storing the
1418 * verifier, then do_open_lookup() will also need to be fixed
1419 * accordingly.
1da177e4
LT
1420 */
1421 v_mtime = verifier[0]&0x7fffffff;
1422 v_atime = verifier[1]&0x7fffffff;
1da177e4
LT
1423 }
1424
463c3197
DH
1425 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1426 if (host_err)
1427 goto out_nfserr;
1da177e4
LT
1428 if (dchild->d_inode) {
1429 err = 0;
1430
1431 switch (createmode) {
1432 case NFS3_CREATE_UNCHECKED:
1433 if (! S_ISREG(dchild->d_inode->i_mode))
1434 err = nfserr_exist;
1435 else if (truncp) {
1436 /* in nfsv4, we need to treat this case a little
1437 * differently. we don't want to truncate the
1438 * file now; this would be wrong if the OPEN
1439 * fails for some other reason. furthermore,
1440 * if the size is nonzero, we should ignore it
1441 * according to spec!
1442 */
1443 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1444 }
1445 else {
1446 iap->ia_valid &= ATTR_SIZE;
1447 goto set_attr;
1448 }
1449 break;
1450 case NFS3_CREATE_EXCLUSIVE:
1451 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1452 && dchild->d_inode->i_atime.tv_sec == v_atime
1da177e4
LT
1453 && dchild->d_inode->i_size == 0 )
1454 break;
1455 /* fallthru */
1456 case NFS3_CREATE_GUARDED:
1457 err = nfserr_exist;
1458 }
463c3197 1459 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1da177e4
LT
1460 goto out;
1461 }
1462
6264d69d 1463 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
463c3197
DH
1464 if (host_err < 0) {
1465 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1da177e4 1466 goto out_nfserr;
463c3197 1467 }
81ac95c5
BF
1468 if (created)
1469 *created = 1;
1da177e4 1470
4ac35c2f 1471 nfsd_check_ignore_resizing(iap);
1472
1da177e4 1473 if (createmode == NFS3_CREATE_EXCLUSIVE) {
c397852c 1474 /* Cram the verifier into atime/mtime */
1da177e4 1475 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
c397852c 1476 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1da177e4
LT
1477 /* XXX someone who knows this better please fix it for nsec */
1478 iap->ia_mtime.tv_sec = v_mtime;
1479 iap->ia_atime.tv_sec = v_atime;
1480 iap->ia_mtime.tv_nsec = 0;
1481 iap->ia_atime.tv_nsec = 0;
1da177e4
LT
1482 }
1483
1da177e4 1484 set_attr:
f501912a
BM
1485 err = nfsd_create_setattr(rqstp, resfhp, iap);
1486
1487 /*
1488 * nfsd_setattr already committed the child (and possibly also the parent).
1489 */
1490 if (!err)
1491 err = nfserrno(commit_metadata(fhp));
f193fbab 1492
463c3197 1493 mnt_drop_write(fhp->fh_export->ex_path.mnt);
f193fbab
YT
1494 /*
1495 * Update the filehandle to get the new inode info.
1496 */
1497 if (!err)
1498 err = fh_update(resfhp);
1da177e4
LT
1499
1500 out:
1501 fh_unlock(fhp);
1502 if (dchild && !IS_ERR(dchild))
1503 dput(dchild);
1504 return err;
1505
1506 out_nfserr:
6264d69d 1507 err = nfserrno(host_err);
1da177e4
LT
1508 goto out;
1509}
1510#endif /* CONFIG_NFSD_V3 */
1511
1512/*
1513 * Read a symlink. On entry, *lenp must contain the maximum path length that
1514 * fits into the buffer. On return, it contains the true length.
1515 * N.B. After this call fhp needs an fh_put
1516 */
6264d69d 1517__be32
1da177e4
LT
1518nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1519{
1520 struct dentry *dentry;
1521 struct inode *inode;
1522 mm_segment_t oldfs;
6264d69d
AV
1523 __be32 err;
1524 int host_err;
1da177e4 1525
8837abca 1526 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1da177e4
LT
1527 if (err)
1528 goto out;
1529
1530 dentry = fhp->fh_dentry;
1531 inode = dentry->d_inode;
1532
1533 err = nfserr_inval;
acfa4380 1534 if (!inode->i_op->readlink)
1da177e4
LT
1535 goto out;
1536
54775491 1537 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1da177e4
LT
1538 /* N.B. Why does this call need a get_fs()??
1539 * Remove the set_fs and watch the fireworks:-) --okir
1540 */
1541
1542 oldfs = get_fs(); set_fs(KERNEL_DS);
6264d69d 1543 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1da177e4
LT
1544 set_fs(oldfs);
1545
6264d69d 1546 if (host_err < 0)
1da177e4 1547 goto out_nfserr;
6264d69d 1548 *lenp = host_err;
1da177e4
LT
1549 err = 0;
1550out:
1551 return err;
1552
1553out_nfserr:
6264d69d 1554 err = nfserrno(host_err);
1da177e4
LT
1555 goto out;
1556}
1557
1558/*
1559 * Create a symlink and look up its inode
1560 * N.B. After this call _both_ fhp and resfhp need an fh_put
1561 */
6264d69d 1562__be32
1da177e4
LT
1563nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1564 char *fname, int flen,
1565 char *path, int plen,
1566 struct svc_fh *resfhp,
1567 struct iattr *iap)
1568{
1569 struct dentry *dentry, *dnew;
6264d69d
AV
1570 __be32 err, cerr;
1571 int host_err;
1da177e4
LT
1572
1573 err = nfserr_noent;
1574 if (!flen || !plen)
1575 goto out;
1576 err = nfserr_exist;
1577 if (isdotent(fname, flen))
1578 goto out;
1579
8837abca 1580 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1da177e4
LT
1581 if (err)
1582 goto out;
1583 fh_lock(fhp);
1584 dentry = fhp->fh_dentry;
1585 dnew = lookup_one_len(fname, dentry, flen);
6264d69d 1586 host_err = PTR_ERR(dnew);
1da177e4
LT
1587 if (IS_ERR(dnew))
1588 goto out_nfserr;
1589
75c3f29d
DH
1590 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1591 if (host_err)
1592 goto out_nfserr;
1593
1da177e4
LT
1594 if (unlikely(path[plen] != 0)) {
1595 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1596 if (path_alloced == NULL)
6264d69d 1597 host_err = -ENOMEM;
1da177e4
LT
1598 else {
1599 strncpy(path_alloced, path, plen);
1600 path_alloced[plen] = 0;
db2e747b 1601 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1da177e4
LT
1602 kfree(path_alloced);
1603 }
1604 } else
db2e747b 1605 host_err = vfs_symlink(dentry->d_inode, dnew, path);
6264d69d 1606 err = nfserrno(host_err);
f501912a
BM
1607 if (!err)
1608 err = nfserrno(commit_metadata(fhp));
1da177e4
LT
1609 fh_unlock(fhp);
1610
75c3f29d
DH
1611 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1612
1da177e4
LT
1613 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1614 dput(dnew);
1615 if (err==0) err = cerr;
1616out:
1617 return err;
1618
1619out_nfserr:
6264d69d 1620 err = nfserrno(host_err);
1da177e4
LT
1621 goto out;
1622}
1623
1624/*
1625 * Create a hardlink
1626 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1627 */
6264d69d 1628__be32
1da177e4
LT
1629nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1630 char *name, int len, struct svc_fh *tfhp)
1631{
1632 struct dentry *ddir, *dnew, *dold;
1633 struct inode *dirp, *dest;
6264d69d
AV
1634 __be32 err;
1635 int host_err;
1da177e4 1636
8837abca 1637 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1da177e4
LT
1638 if (err)
1639 goto out;
8837abca 1640 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1da177e4
LT
1641 if (err)
1642 goto out;
1643
1644 err = nfserr_perm;
1645 if (!len)
1646 goto out;
1647 err = nfserr_exist;
1648 if (isdotent(name, len))
1649 goto out;
1650
12fd3520 1651 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1da177e4
LT
1652 ddir = ffhp->fh_dentry;
1653 dirp = ddir->d_inode;
1654
1655 dnew = lookup_one_len(name, ddir, len);
6264d69d 1656 host_err = PTR_ERR(dnew);
1da177e4
LT
1657 if (IS_ERR(dnew))
1658 goto out_nfserr;
1659
1660 dold = tfhp->fh_dentry;
1661 dest = dold->d_inode;
1662
75c3f29d
DH
1663 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1664 if (host_err) {
1665 err = nfserrno(host_err);
1666 goto out_dput;
1667 }
6264d69d
AV
1668 host_err = vfs_link(dold, dirp, dnew);
1669 if (!host_err) {
f501912a
BM
1670 err = nfserrno(commit_metadata(ffhp));
1671 if (!err)
1672 err = nfserrno(commit_metadata(tfhp));
1da177e4 1673 } else {
6264d69d 1674 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1da177e4
LT
1675 err = nfserr_acces;
1676 else
6264d69d 1677 err = nfserrno(host_err);
1da177e4 1678 }
75c3f29d
DH
1679 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1680out_dput:
1da177e4 1681 dput(dnew);
270d56e5
DR
1682out_unlock:
1683 fh_unlock(ffhp);
1da177e4
LT
1684out:
1685 return err;
1686
1687out_nfserr:
6264d69d 1688 err = nfserrno(host_err);
270d56e5 1689 goto out_unlock;
1da177e4
LT
1690}
1691
1692/*
1693 * Rename a file
1694 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1695 */
6264d69d 1696__be32
1da177e4
LT
1697nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1698 struct svc_fh *tfhp, char *tname, int tlen)
1699{
1700 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1701 struct inode *fdir, *tdir;
6264d69d
AV
1702 __be32 err;
1703 int host_err;
1da177e4 1704
8837abca 1705 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1da177e4
LT
1706 if (err)
1707 goto out;
8837abca 1708 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1da177e4
LT
1709 if (err)
1710 goto out;
1711
1712 fdentry = ffhp->fh_dentry;
1713 fdir = fdentry->d_inode;
1714
1715 tdentry = tfhp->fh_dentry;
1716 tdir = tdentry->d_inode;
1717
1718 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
a56f3937 1719 if (ffhp->fh_export != tfhp->fh_export)
1da177e4
LT
1720 goto out;
1721
1722 err = nfserr_perm;
1723 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1724 goto out;
1725
1726 /* cannot use fh_lock as we need deadlock protective ordering
1727 * so do it by hand */
1728 trap = lock_rename(tdentry, fdentry);
1729 ffhp->fh_locked = tfhp->fh_locked = 1;
1730 fill_pre_wcc(ffhp);
1731 fill_pre_wcc(tfhp);
1732
1733 odentry = lookup_one_len(fname, fdentry, flen);
6264d69d 1734 host_err = PTR_ERR(odentry);
1da177e4
LT
1735 if (IS_ERR(odentry))
1736 goto out_nfserr;
1737
6264d69d 1738 host_err = -ENOENT;
1da177e4
LT
1739 if (!odentry->d_inode)
1740 goto out_dput_old;
6264d69d 1741 host_err = -EINVAL;
1da177e4
LT
1742 if (odentry == trap)
1743 goto out_dput_old;
1744
1745 ndentry = lookup_one_len(tname, tdentry, tlen);
6264d69d 1746 host_err = PTR_ERR(ndentry);
1da177e4
LT
1747 if (IS_ERR(ndentry))
1748 goto out_dput_old;
6264d69d 1749 host_err = -ENOTEMPTY;
1da177e4
LT
1750 if (ndentry == trap)
1751 goto out_dput_new;
1752
9079b1eb 1753 if (svc_msnfs(ffhp) &&
1da177e4
LT
1754 ((atomic_read(&odentry->d_count) > 1)
1755 || (atomic_read(&ndentry->d_count) > 1))) {
6264d69d 1756 host_err = -EPERM;
9079b1eb
DH
1757 goto out_dput_new;
1758 }
1759
1760 host_err = -EXDEV;
1761 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1762 goto out_dput_new;
1763 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1764 if (host_err)
1765 goto out_dput_new;
1766
6264d69d 1767 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
f501912a
BM
1768 if (!host_err) {
1769 host_err = commit_metadata(tfhp);
6264d69d 1770 if (!host_err)
f501912a 1771 host_err = commit_metadata(ffhp);
1da177e4
LT
1772 }
1773
9079b1eb
DH
1774 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1775
1da177e4
LT
1776 out_dput_new:
1777 dput(ndentry);
1778 out_dput_old:
1779 dput(odentry);
1780 out_nfserr:
6264d69d 1781 err = nfserrno(host_err);
1da177e4
LT
1782
1783 /* we cannot reply on fh_unlock on the two filehandles,
1784 * as that would do the wrong thing if the two directories
1785 * were the same, so again we do it by hand
1786 */
1787 fill_post_wcc(ffhp);
1788 fill_post_wcc(tfhp);
1789 unlock_rename(tdentry, fdentry);
1790 ffhp->fh_locked = tfhp->fh_locked = 0;
1791
1792out:
1793 return err;
1794}
1795
1796/*
1797 * Unlink a file or directory
1798 * N.B. After this call fhp needs an fh_put
1799 */
6264d69d 1800__be32
1da177e4
LT
1801nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1802 char *fname, int flen)
1803{
1804 struct dentry *dentry, *rdentry;
1805 struct inode *dirp;
6264d69d
AV
1806 __be32 err;
1807 int host_err;
1da177e4
LT
1808
1809 err = nfserr_acces;
1810 if (!flen || isdotent(fname, flen))
1811 goto out;
8837abca 1812 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1da177e4
LT
1813 if (err)
1814 goto out;
1815
12fd3520 1816 fh_lock_nested(fhp, I_MUTEX_PARENT);
1da177e4
LT
1817 dentry = fhp->fh_dentry;
1818 dirp = dentry->d_inode;
1819
1820 rdentry = lookup_one_len(fname, dentry, flen);
6264d69d 1821 host_err = PTR_ERR(rdentry);
1da177e4
LT
1822 if (IS_ERR(rdentry))
1823 goto out_nfserr;
1824
1825 if (!rdentry->d_inode) {
1826 dput(rdentry);
1827 err = nfserr_noent;
1828 goto out;
1829 }
1830
1831 if (!type)
1832 type = rdentry->d_inode->i_mode & S_IFMT;
1833
0622753b
DH
1834 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1835 if (host_err)
1836 goto out_nfserr;
1837
1da177e4
LT
1838 if (type != S_IFDIR) { /* It's UNLINK */
1839#ifdef MSNFS
1840 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1841 (atomic_read(&rdentry->d_count) > 1)) {
6264d69d 1842 host_err = -EPERM;
1da177e4
LT
1843 } else
1844#endif
6264d69d 1845 host_err = vfs_unlink(dirp, rdentry);
1da177e4 1846 } else { /* It's RMDIR */
6264d69d 1847 host_err = vfs_rmdir(dirp, rdentry);
1da177e4
LT
1848 }
1849
1850 dput(rdentry);
1851
f501912a
BM
1852 if (!host_err)
1853 host_err = commit_metadata(fhp);
1da177e4 1854
0622753b 1855 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1da177e4 1856out_nfserr:
6264d69d 1857 err = nfserrno(host_err);
f193fbab
YT
1858out:
1859 return err;
1da177e4
LT
1860}
1861
14f7dd63
DW
1862/*
1863 * We do this buffering because we must not call back into the file
1864 * system's ->lookup() method from the filldir callback. That may well
1865 * deadlock a number of file systems.
1866 *
1867 * This is based heavily on the implementation of same in XFS.
1868 */
1869struct buffered_dirent {
1870 u64 ino;
1871 loff_t offset;
1872 int namlen;
1873 unsigned int d_type;
1874 char name[];
1875};
1876
1877struct readdir_data {
1878 char *dirent;
1879 size_t used;
53c9c5c0 1880 int full;
14f7dd63
DW
1881};
1882
1883static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1884 loff_t offset, u64 ino, unsigned int d_type)
1885{
1886 struct readdir_data *buf = __buf;
1887 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1888 unsigned int reclen;
1889
1890 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
53c9c5c0
AV
1891 if (buf->used + reclen > PAGE_SIZE) {
1892 buf->full = 1;
14f7dd63 1893 return -EINVAL;
53c9c5c0 1894 }
14f7dd63
DW
1895
1896 de->namlen = namlen;
1897 de->offset = offset;
1898 de->ino = ino;
1899 de->d_type = d_type;
1900 memcpy(de->name, name, namlen);
1901 buf->used += reclen;
1902
1903 return 0;
1904}
1905
2f9092e1
DW
1906static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1907 struct readdir_cd *cdp, loff_t *offsetp)
2628b766 1908{
14f7dd63
DW
1909 struct readdir_data buf;
1910 struct buffered_dirent *de;
2628b766 1911 int host_err;
14f7dd63
DW
1912 int size;
1913 loff_t offset;
2628b766 1914
14f7dd63
DW
1915 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1916 if (!buf.dirent)
2f9092e1 1917 return nfserrno(-ENOMEM);
14f7dd63
DW
1918
1919 offset = *offsetp;
2628b766 1920
14f7dd63 1921 while (1) {
2f9092e1 1922 struct inode *dir_inode = file->f_path.dentry->d_inode;
14f7dd63
DW
1923 unsigned int reclen;
1924
b726e923 1925 cdp->err = nfserr_eof; /* will be cleared on successful read */
14f7dd63 1926 buf.used = 0;
53c9c5c0 1927 buf.full = 0;
14f7dd63
DW
1928
1929 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
53c9c5c0
AV
1930 if (buf.full)
1931 host_err = 0;
1932
1933 if (host_err < 0)
14f7dd63
DW
1934 break;
1935
1936 size = buf.used;
1937
1938 if (!size)
1939 break;
1940
2f9092e1
DW
1941 /*
1942 * Various filldir functions may end up calling back into
1943 * lookup_one_len() and the file system's ->lookup() method.
1944 * These expect i_mutex to be held, as it would within readdir.
1945 */
1946 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1947 if (host_err)
1948 break;
1949
14f7dd63
DW
1950 de = (struct buffered_dirent *)buf.dirent;
1951 while (size > 0) {
1952 offset = de->offset;
1953
1954 if (func(cdp, de->name, de->namlen, de->offset,
1955 de->ino, de->d_type))
2f9092e1 1956 break;
14f7dd63
DW
1957
1958 if (cdp->err != nfs_ok)
2f9092e1 1959 break;
14f7dd63
DW
1960
1961 reclen = ALIGN(sizeof(*de) + de->namlen,
1962 sizeof(u64));
1963 size -= reclen;
1964 de = (struct buffered_dirent *)((char *)de + reclen);
1965 }
2f9092e1
DW
1966 mutex_unlock(&dir_inode->i_mutex);
1967 if (size > 0) /* We bailed out early */
1968 break;
1969
c002a6c7 1970 offset = vfs_llseek(file, 0, SEEK_CUR);
14f7dd63
DW
1971 }
1972
14f7dd63 1973 free_page((unsigned long)(buf.dirent));
2628b766
DW
1974
1975 if (host_err)
1976 return nfserrno(host_err);
14f7dd63
DW
1977
1978 *offsetp = offset;
1979 return cdp->err;
2628b766
DW
1980}
1981
1da177e4
LT
1982/*
1983 * Read entries from a directory.
1984 * The NFSv3/4 verifier we ignore for now.
1985 */
6264d69d 1986__be32
1da177e4 1987nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
a0ad13ef 1988 struct readdir_cd *cdp, filldir_t func)
1da177e4 1989{
6264d69d 1990 __be32 err;
1da177e4
LT
1991 struct file *file;
1992 loff_t offset = *offsetp;
1993
8837abca 1994 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1da177e4
LT
1995 if (err)
1996 goto out;
1997
1998 offset = vfs_llseek(file, offset, 0);
1999 if (offset < 0) {
2000 err = nfserrno((int)offset);
2001 goto out_close;
2002 }
2003
14f7dd63 2004 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1da177e4
LT
2005
2006 if (err == nfserr_eof || err == nfserr_toosmall)
2007 err = nfs_ok; /* can still be found in ->err */
2008out_close:
2009 nfsd_close(file);
2010out:
2011 return err;
2012}
2013
2014/*
2015 * Get file system stats
2016 * N.B. After this call fhp needs an fh_put
2017 */
6264d69d 2018__be32
04716e66 2019nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1da177e4 2020{
04716e66 2021 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
726c3342 2022 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1da177e4
LT
2023 err = nfserr_io;
2024 return err;
2025}
2026
c7d51402 2027static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
e22841c6 2028{
c7d51402 2029 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
e22841c6
BF
2030}
2031
1da177e4
LT
2032/*
2033 * Check for a user's access permissions to this inode.
2034 */
6264d69d 2035__be32
0ec757df
BF
2036nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2037 struct dentry *dentry, int acc)
1da177e4
LT
2038{
2039 struct inode *inode = dentry->d_inode;
14dba533 2040 struct path path;
1da177e4
LT
2041 int err;
2042
8837abca 2043 if (acc == NFSD_MAY_NOP)
1da177e4
LT
2044 return 0;
2045#if 0
2046 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2047 acc,
8837abca
MS
2048 (acc & NFSD_MAY_READ)? " read" : "",
2049 (acc & NFSD_MAY_WRITE)? " write" : "",
2050 (acc & NFSD_MAY_EXEC)? " exec" : "",
2051 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2052 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2053 (acc & NFSD_MAY_LOCK)? " lock" : "",
2054 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1da177e4
LT
2055 inode->i_mode,
2056 IS_IMMUTABLE(inode)? " immut" : "",
2057 IS_APPEND(inode)? " append" : "",
2c463e95 2058 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
1da177e4 2059 dprintk(" owner %d/%d user %d/%d\n",
5cc0a840 2060 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
1da177e4
LT
2061#endif
2062
2063 /* Normally we reject any write/sattr etc access on a read-only file
2064 * system. But if it is IRIX doing check on write-access for a
2065 * device special file, we ignore rofs.
2066 */
8837abca
MS
2067 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2068 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2c463e95
DH
2069 if (exp_rdonly(rqstp, exp) ||
2070 __mnt_is_readonly(exp->ex_path.mnt))
1da177e4 2071 return nfserr_rofs;
8837abca 2072 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
1da177e4
LT
2073 return nfserr_perm;
2074 }
8837abca 2075 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
1da177e4
LT
2076 return nfserr_perm;
2077
8837abca 2078 if (acc & NFSD_MAY_LOCK) {
1da177e4
LT
2079 /* If we cannot rely on authentication in NLM requests,
2080 * just allow locks, otherwise require read permission, or
2081 * ownership
2082 */
2083 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2084 return 0;
2085 else
8837abca 2086 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
1da177e4
LT
2087 }
2088 /*
2089 * The file owner always gets access permission for accesses that
2090 * would normally be checked at open time. This is to make
2091 * file access work even when the client has done a fchmod(fd, 0).
2092 *
2093 * However, `cp foo bar' should fail nevertheless when bar is
2094 * readonly. A sensible way to do this might be to reject all
2095 * attempts to truncate a read-only file, because a creat() call
2096 * always implies file truncation.
2097 * ... but this isn't really fair. A process may reasonably call
2098 * ftruncate on an open file descriptor on a file with perm 000.
2099 * We must trust the client to do permission checking - using "ACCESS"
2100 * with NFSv3.
2101 */
8837abca 2102 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
5cc0a840 2103 inode->i_uid == current_fsuid())
1da177e4
LT
2104 return 0;
2105
8837abca 2106 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
f419a2e3 2107 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
1da177e4
LT
2108
2109 /* Allow read access to binaries even when mode 111 */
2110 if (err == -EACCES && S_ISREG(inode->i_mode) &&
8837abca 2111 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
f419a2e3 2112 err = inode_permission(inode, MAY_EXEC);
14dba533
MZ
2113 if (err)
2114 goto nfsd_out;
1da177e4 2115
14dba533
MZ
2116 /* Do integrity (permission) checking now, but defer incrementing
2117 * IMA counts to the actual file open.
2118 */
2119 path.mnt = exp->ex_path.mnt;
2120 path.dentry = dentry;
14dba533 2121nfsd_out:
1da177e4
LT
2122 return err? nfserrno(err) : 0;
2123}
2124
2125void
2126nfsd_racache_shutdown(void)
2127{
54a66e54
JL
2128 struct raparms *raparm, *last_raparm;
2129 unsigned int i;
2130
1da177e4 2131 dprintk("nfsd: freeing readahead buffers.\n");
54a66e54
JL
2132
2133 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2134 raparm = raparm_hash[i].pb_head;
2135 while(raparm) {
2136 last_raparm = raparm;
2137 raparm = raparm->p_next;
2138 kfree(last_raparm);
2139 }
2140 raparm_hash[i].pb_head = NULL;
2141 }
1da177e4
LT
2142}
2143/*
2144 * Initialize readahead param cache
2145 */
2146int
2147nfsd_racache_init(int cache_size)
2148{
2149 int i;
fce1456a
GB
2150 int j = 0;
2151 int nperbucket;
54a66e54 2152 struct raparms **raparm = NULL;
1da177e4 2153
fce1456a 2154
54a66e54 2155 if (raparm_hash[0].pb_head)
1da177e4 2156 return 0;
54a66e54
JL
2157 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2158 if (nperbucket < 2)
2159 nperbucket = 2;
2160 cache_size = nperbucket * RAPARM_HASH_SIZE;
4b3bb06b
YB
2161
2162 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
54a66e54
JL
2163
2164 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
4b3bb06b 2165 spin_lock_init(&raparm_hash[i].pb_lock);
54a66e54
JL
2166
2167 raparm = &raparm_hash[i].pb_head;
2168 for (j = 0; j < nperbucket; j++) {
2169 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2170 if (!*raparm)
2171 goto out_nomem;
2172 raparm = &(*raparm)->p_next;
2173 }
2174 *raparm = NULL;
4b3bb06b
YB
2175 }
2176
1da177e4
LT
2177 nfsdstats.ra_size = cache_size;
2178 return 0;
54a66e54
JL
2179
2180out_nomem:
2181 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2182 nfsd_racache_shutdown();
2183 return -ENOMEM;
1da177e4 2184}
a257cdd0
AG
2185
2186#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2187struct posix_acl *
2188nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2189{
2190 struct inode *inode = fhp->fh_dentry->d_inode;
2191 char *name;
2192 void *value = NULL;
2193 ssize_t size;
2194 struct posix_acl *acl;
2195
5be196e5
CH
2196 if (!IS_POSIXACL(inode))
2197 return ERR_PTR(-EOPNOTSUPP);
2198
2199 switch (type) {
2200 case ACL_TYPE_ACCESS:
2201 name = POSIX_ACL_XATTR_ACCESS;
2202 break;
2203 case ACL_TYPE_DEFAULT:
2204 name = POSIX_ACL_XATTR_DEFAULT;
2205 break;
2206 default:
a257cdd0 2207 return ERR_PTR(-EOPNOTSUPP);
a257cdd0
AG
2208 }
2209
5be196e5
CH
2210 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2211 if (size < 0)
2212 return ERR_PTR(size);
a257cdd0 2213
a257cdd0 2214 acl = posix_acl_from_xattr(value, size);
a257cdd0
AG
2215 kfree(value);
2216 return acl;
2217}
2218
2219int
2220nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2221{
2222 struct inode *inode = fhp->fh_dentry->d_inode;
2223 char *name;
2224 void *value = NULL;
2225 size_t size;
2226 int error;
2227
acfa4380 2228 if (!IS_POSIXACL(inode) ||
a257cdd0
AG
2229 !inode->i_op->setxattr || !inode->i_op->removexattr)
2230 return -EOPNOTSUPP;
2231 switch(type) {
2232 case ACL_TYPE_ACCESS:
334a13ec 2233 name = POSIX_ACL_XATTR_ACCESS;
a257cdd0
AG
2234 break;
2235 case ACL_TYPE_DEFAULT:
334a13ec 2236 name = POSIX_ACL_XATTR_DEFAULT;
a257cdd0
AG
2237 break;
2238 default:
2239 return -EOPNOTSUPP;
2240 }
2241
2242 if (acl && acl->a_count) {
334a13ec 2243 size = posix_acl_xattr_size(acl->a_count);
a257cdd0
AG
2244 value = kmalloc(size, GFP_KERNEL);
2245 if (!value)
2246 return -ENOMEM;
9ccfc29c
FM
2247 error = posix_acl_to_xattr(acl, value, size);
2248 if (error < 0)
a257cdd0 2249 goto getout;
9ccfc29c 2250 size = error;
a257cdd0
AG
2251 } else
2252 size = 0;
2253
18f335af
DH
2254 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2255 if (error)
2256 goto getout;
a257cdd0 2257 if (size)
5be196e5 2258 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
a257cdd0
AG
2259 else {
2260 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2261 error = 0;
2262 else {
5be196e5 2263 error = vfs_removexattr(fhp->fh_dentry, name);
a257cdd0
AG
2264 if (error == -ENODATA)
2265 error = 0;
2266 }
2267 }
18f335af 2268 mnt_drop_write(fhp->fh_export->ex_path.mnt);
a257cdd0
AG
2269
2270getout:
2271 kfree(value);
2272 return error;
2273}
2274#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
This page took 0.771454 seconds and 5 git commands to generate.