blob: 46f59d5365a0d49986512c54a25ecdc6daeb54f1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#define MSNFS /* HACK HACK */
2/*
3 * linux/fs/nfsd/vfs.c
4 *
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
8 *
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
14 *
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 */
18
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/string.h>
20#include <linux/time.h>
21#include <linux/errno.h>
22#include <linux/fs.h>
23#include <linux/file.h>
24#include <linux/mount.h>
25#include <linux/major.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020026#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/proc_fs.h>
28#include <linux/stat.h>
29#include <linux/fcntl.h>
30#include <linux/net.h>
31#include <linux/unistd.h>
32#include <linux/slab.h>
33#include <linux/pagemap.h>
34#include <linux/in.h>
35#include <linux/module.h>
36#include <linux/namei.h>
37#include <linux/vfs.h>
38#include <linux/delay.h>
39#include <linux/sunrpc/svc.h>
40#include <linux/nfsd/nfsd.h>
41#ifdef CONFIG_NFSD_V3
42#include <linux/nfs3.h>
43#include <linux/nfsd/xdr3.h>
44#endif /* CONFIG_NFSD_V3 */
45#include <linux/nfsd/nfsfh.h>
46#include <linux/quotaops.h>
Robert Love0eeca282005-07-12 17:06:03 -040047#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/posix_acl.h>
49#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/xattr.h>
Christoph Hellwig5be196e2006-01-09 20:51:55 -080051#ifdef CONFIG_NFSD_V4
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/nfs4.h>
53#include <linux/nfs4_acl.h>
54#include <linux/nfsd_idmap.h>
55#include <linux/security.h>
56#endif /* CONFIG_NFSD_V4 */
Greg Banksfce14562006-10-04 02:15:49 -070057#include <linux/jhash.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59#include <asm/uaccess.h>
60
61#define NFSDDBG_FACILITY NFSDDBG_FILEOP
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/*
65 * This is a cache of readahead params that help us choose the proper
66 * readahead strategy. Initially, we set all readahead parameters to 0
67 * and let the VFS handle things.
68 * If you increase the number of cached files very much, you'll need to
69 * add a hash table here.
70 */
71struct raparms {
72 struct raparms *p_next;
73 unsigned int p_count;
74 ino_t p_ino;
75 dev_t p_dev;
76 int p_set;
77 struct file_ra_state p_ra;
Greg Banksfce14562006-10-04 02:15:49 -070078 unsigned int p_hindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079};
80
Greg Banksfce14562006-10-04 02:15:49 -070081struct raparm_hbucket {
82 struct raparms *pb_head;
83 spinlock_t pb_lock;
84} ____cacheline_aligned_in_smp;
85
Linus Torvalds1da177e2005-04-16 15:20:36 -070086static struct raparms * raparml;
Greg Banksfce14562006-10-04 02:15:49 -070087#define RAPARM_HASH_BITS 4
88#define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
89#define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
90static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
92/*
93 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
94 * a mount point.
J.Bruce Fieldse0bb89e2006-12-13 00:35:25 -080095 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * or nfs_ok having possibly changed *dpp and *expp
97 */
98int
99nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
100 struct svc_export **expp)
101{
102 struct svc_export *exp = *expp, *exp2 = NULL;
103 struct dentry *dentry = *dpp;
Jan Blunck54775492008-02-14 19:38:39 -0800104 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 struct dentry *mounts = dget(dentry);
Al Viro6264d692006-10-19 23:28:58 -0700106 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
109
J. Bruce Fields0989a782007-07-17 04:04:44 -0700110 exp2 = rqst_exp_get_by_name(rqstp, mnt, mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 if (IS_ERR(exp2)) {
Neil Browna1033be2007-09-05 17:22:12 -0400112 if (PTR_ERR(exp2) != -ENOENT)
113 err = PTR_ERR(exp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 dput(mounts);
115 mntput(mnt);
116 goto out;
117 }
J. Bruce Fields5d3dbbe2007-07-19 01:49:19 -0700118 if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 /* successfully crossed mount point */
120 exp_put(exp);
121 *expp = exp2;
122 dput(dentry);
123 *dpp = mounts;
124 } else {
J. Bruce Fields5d3dbbe2007-07-19 01:49:19 -0700125 exp_put(exp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 dput(mounts);
127 }
128 mntput(mnt);
129out:
130 return err;
131}
132
Al Viro6264d692006-10-19 23:28:58 -0700133__be32
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700134nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
Chuck Lever5a022fc2007-11-01 16:57:09 -0400135 const char *name, unsigned int len,
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700136 struct svc_export **exp_ret, struct dentry **dentry_ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137{
138 struct svc_export *exp;
139 struct dentry *dparent;
140 struct dentry *dentry;
Al Viro6264d692006-10-19 23:28:58 -0700141 __be32 err;
142 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
144 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
145
146 /* Obtain dentry and export. */
147 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
148 if (err)
149 return err;
150
151 dparent = fhp->fh_dentry;
152 exp = fhp->fh_export;
153 exp_get(exp);
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 /* Lookup the name, but don't follow links */
156 if (isdotent(name, len)) {
157 if (len==1)
158 dentry = dget(dparent);
Jan Blunck54775492008-02-14 19:38:39 -0800159 else if (dparent != exp->ex_path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 dentry = dget_parent(dparent);
Jan Blunck54775492008-02-14 19:38:39 -0800161 else if (!EX_NOHIDE(exp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 dentry = dget(dparent); /* .. == . just like at / */
163 else {
164 /* checking mountpoint crossing is very different when stepping up */
165 struct svc_export *exp2 = NULL;
166 struct dentry *dp;
Jan Blunck54775492008-02-14 19:38:39 -0800167 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 dentry = dget(dparent);
169 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
170 ;
171 dp = dget_parent(dentry);
172 dput(dentry);
173 dentry = dp;
174
J. Bruce Fields0989a782007-07-17 04:04:44 -0700175 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
J. Bruce Fields2d3bb252007-07-17 04:04:40 -0700176 if (PTR_ERR(exp2) == -ENOENT) {
177 dput(dentry);
178 dentry = dget(dparent);
179 } else if (IS_ERR(exp2)) {
Al Viro6264d692006-10-19 23:28:58 -0700180 host_err = PTR_ERR(exp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 dput(dentry);
182 mntput(mnt);
183 goto out_nfserr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 } else {
185 exp_put(exp);
186 exp = exp2;
187 }
188 mntput(mnt);
189 }
190 } else {
191 fh_lock(fhp);
192 dentry = lookup_one_len(name, dparent, len);
Al Viro6264d692006-10-19 23:28:58 -0700193 host_err = PTR_ERR(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 if (IS_ERR(dentry))
195 goto out_nfserr;
196 /*
197 * check if we have crossed a mount point ...
198 */
199 if (d_mountpoint(dentry)) {
Al Viro6264d692006-10-19 23:28:58 -0700200 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 dput(dentry);
202 goto out_nfserr;
203 }
204 }
205 }
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700206 *dentry_ret = dentry;
207 *exp_ret = exp;
208 return 0;
209
210out_nfserr:
211 exp_put(exp);
212 return nfserrno(host_err);
213}
214
215/*
216 * Look up one component of a pathname.
217 * N.B. After this call _both_ fhp and resfh need an fh_put
218 *
219 * If the lookup would cross a mountpoint, and the mounted filesystem
220 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
221 * accepted as it stands and the mounted directory is
222 * returned. Otherwise the covered directory is returned.
223 * NOTE: this mountpoint crossing is not supported properly by all
224 * clients and is explicitly disallowed for NFSv3
225 * NeilBrown <neilb@cse.unsw.edu.au>
226 */
227__be32
228nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
Chuck Lever5a022fc2007-11-01 16:57:09 -0400229 unsigned int len, struct svc_fh *resfh)
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700230{
231 struct svc_export *exp;
232 struct dentry *dentry;
233 __be32 err;
234
235 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
236 if (err)
237 return err;
Andy Adamson32c1eb02007-07-17 04:04:48 -0700238 err = check_nfsd_access(exp, rqstp);
239 if (err)
240 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 /*
242 * Note: we compose the file handle now, but as the
243 * dentry may be negative, it may need to be updated.
244 */
245 err = fh_compose(resfh, exp, dentry, fhp);
246 if (!err && !dentry->d_inode)
247 err = nfserr_noent;
Andy Adamson32c1eb02007-07-17 04:04:48 -0700248out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 dput(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 exp_put(exp);
251 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252}
253
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700254
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255/*
256 * Set various file attributes.
257 * N.B. After this call fhp needs an fh_put
258 */
Al Viro6264d692006-10-19 23:28:58 -0700259__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
261 int check_guard, time_t guardtime)
262{
263 struct dentry *dentry;
264 struct inode *inode;
265 int accmode = MAY_SATTR;
266 int ftype = 0;
267 int imode;
Al Viro6264d692006-10-19 23:28:58 -0700268 __be32 err;
269 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 int size_change = 0;
271
272 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
273 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
274 if (iap->ia_valid & ATTR_SIZE)
275 ftype = S_IFREG;
276
277 /* Get inode */
278 err = fh_verify(rqstp, fhp, ftype, accmode);
NeilBrown15b7a1b2005-11-07 01:00:23 -0800279 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 goto out;
281
282 dentry = fhp->fh_dentry;
283 inode = dentry->d_inode;
284
NeilBrown15b7a1b2005-11-07 01:00:23 -0800285 /* Ignore any mode updates on symlinks */
286 if (S_ISLNK(inode->i_mode))
287 iap->ia_valid &= ~ATTR_MODE;
288
289 if (!iap->ia_valid)
290 goto out;
291
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000292 /*
293 * NFSv2 does not differentiate between "set-[ac]time-to-now"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 * which only requires access, and "set-[ac]time-to-X" which
295 * requires ownership.
296 * So if it looks like it might be "set both to the same time which
297 * is close to now", and if inode_change_ok fails, then we
298 * convert to "set to now" instead of "set to explicit time"
299 *
300 * We only call inode_change_ok as the last test as technically
301 * it is not an interface that we should be using. It is only
302 * valid if the filesystem does not define it's own i_op->setattr.
303 */
304#define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
305#define MAX_TOUCH_TIME_ERROR (30*60)
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000306 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
307 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
308 /*
309 * Looks probable.
310 *
311 * Now just make sure time is in the right ballpark.
312 * Solaris, at least, doesn't seem to care what the time
313 * request is. We require it be within 30 minutes of now.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 */
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000315 time_t delta = iap->ia_atime.tv_sec - get_seconds();
316 if (delta < 0)
317 delta = -delta;
318 if (delta < MAX_TOUCH_TIME_ERROR &&
319 inode_change_ok(inode, iap) != 0) {
320 /*
321 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
322 * This will cause notify_change to set these times
323 * to "now"
324 */
325 iap->ia_valid &= ~BOTH_TIME_SET;
326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 }
328
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000329 /*
330 * The size case is special.
331 * It changes the file as well as the attributes.
332 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (iap->ia_valid & ATTR_SIZE) {
334 if (iap->ia_size < inode->i_size) {
J. Bruce Fields0ec757d2007-07-17 04:04:48 -0700335 err = nfsd_permission(rqstp, fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 if (err)
337 goto out;
338 }
339
340 /*
341 * If we are changing the size of the file, then
342 * we need to break all leases.
343 */
Al Viro6264d692006-10-19 23:28:58 -0700344 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
345 if (host_err == -EWOULDBLOCK)
346 host_err = -ETIMEDOUT;
347 if (host_err) /* ENOMEM or EWOULDBLOCK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 goto out_nfserr;
349
Al Viro6264d692006-10-19 23:28:58 -0700350 host_err = get_write_access(inode);
351 if (host_err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 goto out_nfserr;
353
354 size_change = 1;
Al Viro6264d692006-10-19 23:28:58 -0700355 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
356 if (host_err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 put_write_access(inode);
358 goto out_nfserr;
359 }
360 DQUOT_INIT(inode);
361 }
362
363 imode = inode->i_mode;
364 if (iap->ia_valid & ATTR_MODE) {
365 iap->ia_mode &= S_IALLUGO;
366 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
Jeff Layton8a0ce7d2007-10-18 03:05:19 -0700367 /* if changing uid/gid revoke setuid/setgid in mode */
368 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) {
369 iap->ia_valid |= ATTR_KILL_PRIV;
370 iap->ia_mode &= ~S_ISUID;
371 }
372 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
373 iap->ia_mode &= ~S_ISGID;
374 } else {
375 /*
376 * Revoke setuid/setgid bit on chown/chgrp
377 */
378 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
379 iap->ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV;
380 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
381 iap->ia_valid |= ATTR_KILL_SGID;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
383
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 /* Change the attributes. */
385
386 iap->ia_valid |= ATTR_CTIME;
387
388 err = nfserr_notsync;
389 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
390 fh_lock(fhp);
Al Viro6264d692006-10-19 23:28:58 -0700391 host_err = notify_change(dentry, iap);
392 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 fh_unlock(fhp);
394 }
395 if (size_change)
396 put_write_access(inode);
397 if (!err)
398 if (EX_ISSYNC(fhp->fh_export))
399 write_inode_now(inode, 1);
400out:
401 return err;
402
403out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -0700404 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 goto out;
406}
407
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800408#if defined(CONFIG_NFSD_V2_ACL) || \
409 defined(CONFIG_NFSD_V3_ACL) || \
410 defined(CONFIG_NFSD_V4)
411static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
412{
413 ssize_t buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800415 buflen = vfs_getxattr(dentry, key, NULL, 0);
416 if (buflen <= 0)
417 return buflen;
418
419 *buf = kmalloc(buflen, GFP_KERNEL);
420 if (!*buf)
421 return -ENOMEM;
422
NeilBrownb5872b02006-04-10 22:55:26 -0700423 return vfs_getxattr(dentry, key, *buf, buflen);
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800424}
425#endif
426
427#if defined(CONFIG_NFSD_V4)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428static int
429set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
430{
431 int len;
432 size_t buflen;
433 char *buf = NULL;
434 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436 buflen = posix_acl_xattr_size(pacl->a_count);
437 buf = kmalloc(buflen, GFP_KERNEL);
438 error = -ENOMEM;
439 if (buf == NULL)
440 goto out;
441
442 len = posix_acl_to_xattr(pacl, buf, buflen);
443 if (len < 0) {
444 error = len;
445 goto out;
446 }
447
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800448 error = vfs_setxattr(dentry, key, buf, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449out:
450 kfree(buf);
451 return error;
452}
453
Al Viro6264d692006-10-19 23:28:58 -0700454__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
456 struct nfs4_acl *acl)
457{
Al Viro6264d692006-10-19 23:28:58 -0700458 __be32 error;
459 int host_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 struct dentry *dentry;
461 struct inode *inode;
462 struct posix_acl *pacl = NULL, *dpacl = NULL;
463 unsigned int flags = 0;
464
465 /* Get inode */
466 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
467 if (error)
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700468 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 dentry = fhp->fh_dentry;
471 inode = dentry->d_inode;
472 if (S_ISDIR(inode->i_mode))
473 flags = NFS4_ACL_DIR;
474
Al Viro6264d692006-10-19 23:28:58 -0700475 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
476 if (host_error == -EINVAL) {
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700477 return nfserr_attrnotsupp;
Al Viro6264d692006-10-19 23:28:58 -0700478 } else if (host_error < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 goto out_nfserr;
480
Al Viro6264d692006-10-19 23:28:58 -0700481 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
482 if (host_error < 0)
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700483 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700485 if (S_ISDIR(inode->i_mode))
Al Viro6264d692006-10-19 23:28:58 -0700486 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700488out_release:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 posix_acl_release(pacl);
490 posix_acl_release(dpacl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491out_nfserr:
J. Bruce Fieldsf34f9242007-02-16 01:28:34 -0800492 if (host_error == -EOPNOTSUPP)
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700493 return nfserr_attrnotsupp;
J. Bruce Fieldsf34f9242007-02-16 01:28:34 -0800494 else
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700495 return nfserrno(host_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496}
497
498static struct posix_acl *
499_get_posix_acl(struct dentry *dentry, char *key)
500{
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800501 void *buf = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 struct posix_acl *pacl = NULL;
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800503 int buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800505 buflen = nfsd_getxattr(dentry, key, &buf);
506 if (!buflen)
507 buflen = -ENODATA;
508 if (buflen <= 0)
509 return ERR_PTR(buflen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
511 pacl = posix_acl_from_xattr(buf, buflen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 kfree(buf);
513 return pacl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514}
515
516int
517nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
518{
519 struct inode *inode = dentry->d_inode;
520 int error = 0;
521 struct posix_acl *pacl = NULL, *dpacl = NULL;
522 unsigned int flags = 0;
523
Christoph Hellwig9a59f452005-06-23 00:10:19 -0700524 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
526 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
527 if (IS_ERR(pacl)) {
528 error = PTR_ERR(pacl);
529 pacl = NULL;
530 goto out;
531 }
532
533 if (S_ISDIR(inode->i_mode)) {
Christoph Hellwig9a59f452005-06-23 00:10:19 -0700534 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
536 dpacl = NULL;
537 else if (IS_ERR(dpacl)) {
538 error = PTR_ERR(dpacl);
539 dpacl = NULL;
540 goto out;
541 }
542 flags = NFS4_ACL_DIR;
543 }
544
545 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
546 if (IS_ERR(*acl)) {
547 error = PTR_ERR(*acl);
548 *acl = NULL;
549 }
550 out:
551 posix_acl_release(pacl);
552 posix_acl_release(dpacl);
553 return error;
554}
555
556#endif /* defined(CONFIG_NFS_V4) */
557
558#ifdef CONFIG_NFSD_V3
559/*
560 * Check server access rights to a file system object
561 */
562struct accessmap {
563 u32 access;
564 int how;
565};
566static struct accessmap nfs3_regaccess[] = {
567 { NFS3_ACCESS_READ, MAY_READ },
568 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
569 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
570 { NFS3_ACCESS_EXTEND, MAY_WRITE },
571
572 { 0, 0 }
573};
574
575static struct accessmap nfs3_diraccess[] = {
576 { NFS3_ACCESS_READ, MAY_READ },
577 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
578 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
579 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
580 { NFS3_ACCESS_DELETE, MAY_REMOVE },
581
582 { 0, 0 }
583};
584
585static struct accessmap nfs3_anyaccess[] = {
586 /* Some clients - Solaris 2.6 at least, make an access call
587 * to the server to check for access for things like /dev/null
588 * (which really, the server doesn't care about). So
589 * We provide simple access checking for them, looking
590 * mainly at mode bits, and we make sure to ignore read-only
591 * filesystem checks
592 */
593 { NFS3_ACCESS_READ, MAY_READ },
594 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
595 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
596 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
597
598 { 0, 0 }
599};
600
Al Viro6264d692006-10-19 23:28:58 -0700601__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
603{
604 struct accessmap *map;
605 struct svc_export *export;
606 struct dentry *dentry;
607 u32 query, result = 0, sresult = 0;
Al Viro6264d692006-10-19 23:28:58 -0700608 __be32 error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
611 if (error)
612 goto out;
613
614 export = fhp->fh_export;
615 dentry = fhp->fh_dentry;
616
617 if (S_ISREG(dentry->d_inode->i_mode))
618 map = nfs3_regaccess;
619 else if (S_ISDIR(dentry->d_inode->i_mode))
620 map = nfs3_diraccess;
621 else
622 map = nfs3_anyaccess;
623
624
625 query = *access;
626 for (; map->access; map++) {
627 if (map->access & query) {
Al Viro6264d692006-10-19 23:28:58 -0700628 __be32 err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630 sresult |= map->access;
631
J. Bruce Fields0ec757d2007-07-17 04:04:48 -0700632 err2 = nfsd_permission(rqstp, export, dentry, map->how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 switch (err2) {
634 case nfs_ok:
635 result |= map->access;
636 break;
637
638 /* the following error codes just mean the access was not allowed,
639 * rather than an error occurred */
640 case nfserr_rofs:
641 case nfserr_acces:
642 case nfserr_perm:
643 /* simply don't "or" in the access bit. */
644 break;
645 default:
646 error = err2;
647 goto out;
648 }
649 }
650 }
651 *access = result;
652 if (supported)
653 *supported = sresult;
654
655 out:
656 return error;
657}
658#endif /* CONFIG_NFSD_V3 */
659
660
661
662/*
663 * Open an existing file or directory.
664 * The access argument indicates the type of open (read/write/lock)
665 * N.B. After this call fhp needs an fh_put
666 */
Al Viro6264d692006-10-19 23:28:58 -0700667__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
669 int access, struct file **filp)
670{
671 struct dentry *dentry;
672 struct inode *inode;
Al Viro6264d692006-10-19 23:28:58 -0700673 int flags = O_RDONLY|O_LARGEFILE;
674 __be32 err;
675 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 /*
678 * If we get here, then the client has already done an "open",
679 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
680 * in case a chmod has now revoked permission.
681 */
682 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
683 if (err)
684 goto out;
685
686 dentry = fhp->fh_dentry;
687 inode = dentry->d_inode;
688
689 /* Disallow write access to files with the append-only bit set
690 * or any access when mandatory locking enabled
691 */
692 err = nfserr_perm;
693 if (IS_APPEND(inode) && (access & MAY_WRITE))
694 goto out;
J. Bruce Fields5e7fc432007-10-02 14:18:12 -0400695 /*
696 * We must ignore files (but only files) which might have mandatory
697 * locks on them because there is no way to know if the accesser has
698 * the lock.
699 */
700 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 goto out;
702
703 if (!inode->i_fop)
704 goto out;
705
706 /*
707 * Check to see if there are any leases on this file.
708 * This may block while leases are broken.
709 */
Al Viro6264d692006-10-19 23:28:58 -0700710 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
711 if (host_err == -EWOULDBLOCK)
712 host_err = -ETIMEDOUT;
713 if (host_err) /* NOMEM or WOULDBLOCK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 goto out_nfserr;
715
716 if (access & MAY_WRITE) {
J. Bruce Fields9ecb6a02006-06-30 01:56:17 -0700717 if (access & MAY_READ)
718 flags = O_RDWR|O_LARGEFILE;
719 else
720 flags = O_WRONLY|O_LARGEFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
722 DQUOT_INIT(inode);
723 }
Jan Blunck54775492008-02-14 19:38:39 -0800724 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
725 flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 if (IS_ERR(*filp))
Al Viro6264d692006-10-19 23:28:58 -0700727 host_err = PTR_ERR(*filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -0700729 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730out:
731 return err;
732}
733
734/*
735 * Close a file.
736 */
737void
738nfsd_close(struct file *filp)
739{
740 fput(filp);
741}
742
743/*
744 * Sync a file
745 * As this calls fsync (not fdatasync) there is no need for a write_inode
746 * after it.
747 */
David Shawa334de22006-01-06 00:19:58 -0800748static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800749 const struct file_operations *fop)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
751 struct inode *inode = dp->d_inode;
752 int (*fsync) (struct file *, struct dentry *, int);
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -0800753 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -0800755 err = filemap_fdatawrite(inode->i_mapping);
756 if (err == 0 && fop && (fsync = fop->fsync))
757 err = fsync(filp, dp, 0);
758 if (err == 0)
759 err = filemap_fdatawait(inode->i_mapping);
David Shawa334de22006-01-06 00:19:58 -0800760
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -0800761 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
764
David Shawa334de22006-01-06 00:19:58 -0800765static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766nfsd_sync(struct file *filp)
767{
David Shawa334de22006-01-06 00:19:58 -0800768 int err;
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800769 struct inode *inode = filp->f_path.dentry->d_inode;
770 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800771 mutex_lock(&inode->i_mutex);
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800772 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800773 mutex_unlock(&inode->i_mutex);
David Shawa334de22006-01-06 00:19:58 -0800774
775 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -0800778int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779nfsd_sync_dir(struct dentry *dp)
780{
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -0800781 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782}
783
784/*
785 * Obtain the readahead parameters for the file
786 * specified by (dev, ino).
787 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
789static inline struct raparms *
790nfsd_get_raparms(dev_t dev, ino_t ino)
791{
792 struct raparms *ra, **rap, **frap = NULL;
793 int depth = 0;
Greg Banksfce14562006-10-04 02:15:49 -0700794 unsigned int hash;
795 struct raparm_hbucket *rab;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Greg Banksfce14562006-10-04 02:15:49 -0700797 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
798 rab = &raparm_hash[hash];
799
800 spin_lock(&rab->pb_lock);
801 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 if (ra->p_ino == ino && ra->p_dev == dev)
803 goto found;
804 depth++;
805 if (ra->p_count == 0)
806 frap = rap;
807 }
808 depth = nfsdstats.ra_size*11/10;
809 if (!frap) {
Greg Banksfce14562006-10-04 02:15:49 -0700810 spin_unlock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 return NULL;
812 }
813 rap = frap;
814 ra = *frap;
815 ra->p_dev = dev;
816 ra->p_ino = ino;
817 ra->p_set = 0;
Greg Banksfce14562006-10-04 02:15:49 -0700818 ra->p_hindex = hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819found:
Greg Banksfce14562006-10-04 02:15:49 -0700820 if (rap != &rab->pb_head) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 *rap = ra->p_next;
Greg Banksfce14562006-10-04 02:15:49 -0700822 ra->p_next = rab->pb_head;
823 rab->pb_head = ra;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 }
825 ra->p_count++;
826 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
Greg Banksfce14562006-10-04 02:15:49 -0700827 spin_unlock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 return ra;
829}
830
831/*
Jens Axboecf8208d2007-06-12 21:22:14 +0200832 * Grab and keep cached pages associated with a file in the svc_rqst
833 * so that they can be passed to the network sendmsg/sendpage routines
834 * directly. They will be released after the sending has completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 */
836static int
Jens Axboecf8208d2007-06-12 21:22:14 +0200837nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
838 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
Jens Axboecf8208d2007-06-12 21:22:14 +0200840 struct svc_rqst *rqstp = sd->u.data;
NeilBrown44524352006-10-04 02:15:46 -0700841 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
Jens Axboecf8208d2007-06-12 21:22:14 +0200842 struct page *page = buf->page;
843 size_t size;
844 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845
Jens Axboecac36bb02007-06-14 13:10:48 +0200846 ret = buf->ops->confirm(pipe, buf);
Jens Axboecf8208d2007-06-12 21:22:14 +0200847 if (unlikely(ret))
848 return ret;
849
850 size = sd->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
852 if (rqstp->rq_res.page_len == 0) {
853 get_page(page);
NeilBrown44524352006-10-04 02:15:46 -0700854 put_page(*pp);
855 *pp = page;
856 rqstp->rq_resused++;
Jens Axboecf8208d2007-06-12 21:22:14 +0200857 rqstp->rq_res.page_base = buf->offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 rqstp->rq_res.page_len = size;
NeilBrown44524352006-10-04 02:15:46 -0700859 } else if (page != pp[-1]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 get_page(page);
NeilBrown250f3912007-01-26 00:56:59 -0800861 if (*pp)
862 put_page(*pp);
NeilBrown44524352006-10-04 02:15:46 -0700863 *pp = page;
864 rqstp->rq_resused++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 rqstp->rq_res.page_len += size;
NeilBrown44524352006-10-04 02:15:46 -0700866 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 rqstp->rq_res.page_len += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 return size;
870}
871
Jens Axboecf8208d2007-06-12 21:22:14 +0200872static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
873 struct splice_desc *sd)
874{
875 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
876}
877
Dave Hansena8754be2007-10-16 23:31:15 -0700878static inline int svc_msnfs(struct svc_fh *ffhp)
879{
880#ifdef MSNFS
881 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
882#else
883 return 0;
884#endif
885}
886
Al Viro6264d692006-10-19 23:28:58 -0700887static __be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
889 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
890{
891 struct inode *inode;
892 struct raparms *ra;
893 mm_segment_t oldfs;
Al Viro6264d692006-10-19 23:28:58 -0700894 __be32 err;
895 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 err = nfserr_perm;
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800898 inode = file->f_path.dentry->d_inode;
Dave Hansena8754be2007-10-16 23:31:15 -0700899
900 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 /* Get readahead parameters */
904 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
905
906 if (ra && ra->p_set)
907 file->f_ra = ra->p_ra;
908
Jens Axboecf8208d2007-06-12 21:22:14 +0200909 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
910 struct splice_desc sd = {
911 .len = 0,
912 .total_len = *count,
913 .pos = offset,
914 .u.data = rqstp,
915 };
916
Jens Axboe4fbef202007-07-13 22:42:20 +0200917 rqstp->rq_resused = 1;
Jens Axboecf8208d2007-06-12 21:22:14 +0200918 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 } else {
920 oldfs = get_fs();
921 set_fs(KERNEL_DS);
Al Viro6264d692006-10-19 23:28:58 -0700922 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 set_fs(oldfs);
924 }
925
926 /* Write back readahead params */
927 if (ra) {
Greg Banksfce14562006-10-04 02:15:49 -0700928 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
929 spin_lock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 ra->p_ra = file->f_ra;
931 ra->p_set = 1;
932 ra->p_count--;
Greg Banksfce14562006-10-04 02:15:49 -0700933 spin_unlock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 }
935
Al Viro6264d692006-10-19 23:28:58 -0700936 if (host_err >= 0) {
937 nfsdstats.io_read += host_err;
938 *count = host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 err = 0;
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800940 fsnotify_access(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 } else
Al Viro6264d692006-10-19 23:28:58 -0700942 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943out:
944 return err;
945}
946
Neil Brown9f708e42006-01-06 00:19:59 -0800947static void kill_suid(struct dentry *dentry)
948{
949 struct iattr ia;
Serge E. Hallynb5376772007-10-16 23:31:36 -0700950 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
Neil Brown9f708e42006-01-06 00:19:59 -0800951
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800952 mutex_lock(&dentry->d_inode->i_mutex);
Neil Brown9f708e42006-01-06 00:19:59 -0800953 notify_change(dentry, &ia);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800954 mutex_unlock(&dentry->d_inode->i_mutex);
Neil Brown9f708e42006-01-06 00:19:59 -0800955}
956
Al Viro6264d692006-10-19 23:28:58 -0700957static __be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
959 loff_t offset, struct kvec *vec, int vlen,
960 unsigned long cnt, int *stablep)
961{
962 struct svc_export *exp;
963 struct dentry *dentry;
964 struct inode *inode;
965 mm_segment_t oldfs;
Al Viro6264d692006-10-19 23:28:58 -0700966 __be32 err = 0;
967 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 int stable = *stablep;
969
NeilBrown45bd3b32006-01-18 17:43:50 -0800970#ifdef MSNFS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 err = nfserr_perm;
972
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800974 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 goto out;
976#endif
977
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800978 dentry = file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 inode = dentry->d_inode;
980 exp = fhp->fh_export;
981
982 /*
983 * Request sync writes if
984 * - the sync export option has been set, or
985 * - the client requested O_SYNC behavior (NFSv3 feature).
986 * - The file system doesn't support fsync().
987 * When gathered writes have been configured for this volume,
988 * flushing the data to disk is handled separately below.
989 */
990
991 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
992 stable = 2;
993 *stablep = 2; /* FILE_SYNC */
994 }
995
996 if (!EX_ISSYNC(exp))
997 stable = 0;
998 if (stable && !EX_WGATHER(exp))
999 file->f_flags |= O_SYNC;
1000
1001 /* Write the data. */
1002 oldfs = get_fs(); set_fs(KERNEL_DS);
Al Viro6264d692006-10-19 23:28:58 -07001003 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 set_fs(oldfs);
Al Viro6264d692006-10-19 23:28:58 -07001005 if (host_err >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 nfsdstats.io_write += cnt;
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -08001007 fsnotify_modify(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
1009
1010 /* clear setuid/setgid flag after write */
Al Viro6264d692006-10-19 23:28:58 -07001011 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
Neil Brown9f708e42006-01-06 00:19:59 -08001012 kill_suid(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Al Viro6264d692006-10-19 23:28:58 -07001014 if (host_err >= 0 && stable) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 static ino_t last_ino;
1016 static dev_t last_dev;
1017
1018 /*
1019 * Gathered writes: If another process is currently
1020 * writing to the file, there's a high chance
1021 * this is another nfsd (triggered by a bulk write
1022 * from a client's biod). Rather than syncing the
1023 * file with each write request, we sleep for 10 msec.
1024 *
1025 * I don't know if this roughly approximates
1026 * C. Juszak's idea of gathered writes, but it's a
1027 * nice and simple solution (IMHO), and it seems to
1028 * work:-)
1029 */
1030 if (EX_WGATHER(exp)) {
1031 if (atomic_read(&inode->i_writecount) > 1
1032 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001033 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 msleep(10);
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001035 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
1037
1038 if (inode->i_state & I_DIRTY) {
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001039 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
Al Viro6264d692006-10-19 23:28:58 -07001040 host_err=nfsd_sync(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 }
1042#if 0
1043 wake_up(&inode->i_wait);
1044#endif
1045 }
1046 last_ino = inode->i_ino;
1047 last_dev = inode->i_sb->s_dev;
1048 }
1049
Al Viro6264d692006-10-19 23:28:58 -07001050 dprintk("nfsd: write complete host_err=%d\n", host_err);
1051 if (host_err >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 err = 0;
1053 else
Al Viro6264d692006-10-19 23:28:58 -07001054 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055out:
1056 return err;
1057}
1058
1059/*
1060 * Read data from a file. count must contain the requested read count
1061 * on entry. On return, *count contains the number of bytes actually read.
1062 * N.B. After this call fhp needs an fh_put
1063 */
Al Viro6264d692006-10-19 23:28:58 -07001064__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1066 loff_t offset, struct kvec *vec, int vlen,
1067 unsigned long *count)
1068{
Al Viro6264d692006-10-19 23:28:58 -07001069 __be32 err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
1071 if (file) {
J. Bruce Fields0ec757d2007-07-17 04:04:48 -07001072 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 MAY_READ|MAY_OWNER_OVERRIDE);
1074 if (err)
1075 goto out;
1076 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1077 } else {
1078 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1079 if (err)
1080 goto out;
1081 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1082 nfsd_close(file);
1083 }
1084out:
1085 return err;
1086}
1087
1088/*
1089 * Write data to a file.
1090 * The stable flag requests synchronous writes.
1091 * N.B. After this call fhp needs an fh_put
1092 */
Al Viro6264d692006-10-19 23:28:58 -07001093__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1095 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1096 int *stablep)
1097{
Al Viro6264d692006-10-19 23:28:58 -07001098 __be32 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
1100 if (file) {
J. Bruce Fields0ec757d2007-07-17 04:04:48 -07001101 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 MAY_WRITE|MAY_OWNER_OVERRIDE);
1103 if (err)
1104 goto out;
1105 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1106 stablep);
1107 } else {
1108 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1109 if (err)
1110 goto out;
1111
1112 if (cnt)
1113 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1114 cnt, stablep);
1115 nfsd_close(file);
1116 }
1117out:
1118 return err;
1119}
1120
1121#ifdef CONFIG_NFSD_V3
1122/*
1123 * Commit all pending writes to stable storage.
1124 * Strictly speaking, we could sync just the indicated file region here,
1125 * but there's currently no way we can ask the VFS to do so.
1126 *
1127 * Unfortunately we cannot lock the file to make sure we return full WCC
1128 * data to the client, as locking happens lower down in the filesystem.
1129 */
Al Viro6264d692006-10-19 23:28:58 -07001130__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1132 loff_t offset, unsigned long count)
1133{
1134 struct file *file;
Al Viro6264d692006-10-19 23:28:58 -07001135 __be32 err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137 if ((u64)count > ~(u64)offset)
1138 return nfserr_inval;
1139
1140 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1141 return err;
1142 if (EX_ISSYNC(fhp->fh_export)) {
1143 if (file->f_op && file->f_op->fsync) {
NeilBrown45bd3b32006-01-18 17:43:50 -08001144 err = nfserrno(nfsd_sync(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 } else {
1146 err = nfserr_notsupp;
1147 }
1148 }
1149
1150 nfsd_close(file);
1151 return err;
1152}
1153#endif /* CONFIG_NFSD_V3 */
1154
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001155__be32
1156nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1157 struct iattr *iap)
1158{
1159 /*
1160 * Mode has already been set earlier in create:
1161 */
1162 iap->ia_valid &= ~ATTR_MODE;
1163 /*
1164 * Setting uid/gid works only for root. Irix appears to
1165 * send along the gid on create when it tries to implement
1166 * setgid directories via NFS:
1167 */
1168 if (current->fsuid != 0)
1169 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1170 if (iap->ia_valid)
1171 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1172 return 0;
1173}
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175/*
1176 * Create a file (regular, directory, device, fifo); UNIX sockets
1177 * not yet implemented.
1178 * If the response fh has been verified, the parent directory should
1179 * already be locked. Note that the parent directory is left locked.
1180 *
1181 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1182 */
Al Viro6264d692006-10-19 23:28:58 -07001183__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1185 char *fname, int flen, struct iattr *iap,
1186 int type, dev_t rdev, struct svc_fh *resfhp)
1187{
1188 struct dentry *dentry, *dchild = NULL;
1189 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001190 __be32 err;
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001191 __be32 err2;
Al Viro6264d692006-10-19 23:28:58 -07001192 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194 err = nfserr_perm;
1195 if (!flen)
1196 goto out;
1197 err = nfserr_exist;
1198 if (isdotent(fname, flen))
1199 goto out;
1200
1201 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1202 if (err)
1203 goto out;
1204
1205 dentry = fhp->fh_dentry;
1206 dirp = dentry->d_inode;
1207
1208 err = nfserr_notdir;
1209 if(!dirp->i_op || !dirp->i_op->lookup)
1210 goto out;
1211 /*
1212 * Check whether the response file handle has been verified yet.
1213 * If it has, the parent directory should already be locked.
1214 */
1215 if (!resfhp->fh_dentry) {
1216 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001217 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 dchild = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001219 host_err = PTR_ERR(dchild);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 if (IS_ERR(dchild))
1221 goto out_nfserr;
1222 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1223 if (err)
1224 goto out;
1225 } else {
1226 /* called from nfsd_proc_create */
1227 dchild = dget(resfhp->fh_dentry);
1228 if (!fhp->fh_locked) {
1229 /* not actually possible */
1230 printk(KERN_ERR
1231 "nfsd_create: parent %s/%s not locked!\n",
1232 dentry->d_parent->d_name.name,
1233 dentry->d_name.name);
Al Virod75f2b92006-01-18 17:43:44 -08001234 err = nfserr_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 goto out;
1236 }
1237 }
1238 /*
1239 * Make sure the child dentry is still negative ...
1240 */
1241 err = nfserr_exist;
1242 if (dchild->d_inode) {
1243 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1244 dentry->d_name.name, dchild->d_name.name);
1245 goto out;
1246 }
1247
1248 if (!(iap->ia_valid & ATTR_MODE))
1249 iap->ia_mode = 0;
1250 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1251
1252 /*
1253 * Get the dir op function pointer.
1254 */
J. Bruce Fields088406b2006-11-08 17:44:59 -08001255 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 switch (type) {
1257 case S_IFREG:
Al Viro6264d692006-10-19 23:28:58 -07001258 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 break;
1260 case S_IFDIR:
Al Viro6264d692006-10-19 23:28:58 -07001261 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 break;
1263 case S_IFCHR:
1264 case S_IFBLK:
1265 case S_IFIFO:
1266 case S_IFSOCK:
Al Viro6264d692006-10-19 23:28:58 -07001267 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 break;
1269 default:
1270 printk("nfsd: bad file type %o in nfsd_create\n", type);
Al Viro6264d692006-10-19 23:28:58 -07001271 host_err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 }
Al Viro6264d692006-10-19 23:28:58 -07001273 if (host_err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 goto out_nfserr;
1275
1276 if (EX_ISSYNC(fhp->fh_export)) {
NeilBrown45bd3b32006-01-18 17:43:50 -08001277 err = nfserrno(nfsd_sync_dir(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 write_inode_now(dchild->d_inode, 1);
1279 }
1280
1281
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001282 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1283 if (err2)
1284 err = err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 /*
1286 * Update the file handle to get the new inode info.
1287 */
1288 if (!err)
1289 err = fh_update(resfhp);
1290out:
1291 if (dchild && !IS_ERR(dchild))
1292 dput(dchild);
1293 return err;
1294
1295out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001296 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 goto out;
1298}
1299
1300#ifdef CONFIG_NFSD_V3
1301/*
1302 * NFSv3 version of nfsd_create
1303 */
Al Viro6264d692006-10-19 23:28:58 -07001304__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1306 char *fname, int flen, struct iattr *iap,
1307 struct svc_fh *resfhp, int createmode, u32 *verifier,
J. Bruce Fields81ac95c2006-11-08 17:44:40 -08001308 int *truncp, int *created)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309{
1310 struct dentry *dentry, *dchild = NULL;
1311 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001312 __be32 err;
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001313 __be32 err2;
Al Viro6264d692006-10-19 23:28:58 -07001314 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 __u32 v_mtime=0, v_atime=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317 err = nfserr_perm;
1318 if (!flen)
1319 goto out;
1320 err = nfserr_exist;
1321 if (isdotent(fname, flen))
1322 goto out;
1323 if (!(iap->ia_valid & ATTR_MODE))
1324 iap->ia_mode = 0;
1325 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1326 if (err)
1327 goto out;
1328
1329 dentry = fhp->fh_dentry;
1330 dirp = dentry->d_inode;
1331
1332 /* Get all the sanity checks out of the way before
1333 * we lock the parent. */
1334 err = nfserr_notdir;
1335 if(!dirp->i_op || !dirp->i_op->lookup)
1336 goto out;
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001337 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
1339 /*
1340 * Compose the response file handle.
1341 */
1342 dchild = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001343 host_err = PTR_ERR(dchild);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 if (IS_ERR(dchild))
1345 goto out_nfserr;
1346
1347 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1348 if (err)
1349 goto out;
1350
1351 if (createmode == NFS3_CREATE_EXCLUSIVE) {
Peter Staubachc3978522007-01-26 00:57:00 -08001352 /* solaris7 gets confused (bugid 4218508) if these have
Jeff Layton749997e2007-07-31 00:37:51 -07001353 * the high bit set, so just clear the high bits. If this is
1354 * ever changed to use different attrs for storing the
1355 * verifier, then do_open_lookup() will also need to be fixed
1356 * accordingly.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 */
1358 v_mtime = verifier[0]&0x7fffffff;
1359 v_atime = verifier[1]&0x7fffffff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 }
1361
1362 if (dchild->d_inode) {
1363 err = 0;
1364
1365 switch (createmode) {
1366 case NFS3_CREATE_UNCHECKED:
1367 if (! S_ISREG(dchild->d_inode->i_mode))
1368 err = nfserr_exist;
1369 else if (truncp) {
1370 /* in nfsv4, we need to treat this case a little
1371 * differently. we don't want to truncate the
1372 * file now; this would be wrong if the OPEN
1373 * fails for some other reason. furthermore,
1374 * if the size is nonzero, we should ignore it
1375 * according to spec!
1376 */
1377 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1378 }
1379 else {
1380 iap->ia_valid &= ATTR_SIZE;
1381 goto set_attr;
1382 }
1383 break;
1384 case NFS3_CREATE_EXCLUSIVE:
1385 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1386 && dchild->d_inode->i_atime.tv_sec == v_atime
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 && dchild->d_inode->i_size == 0 )
1388 break;
1389 /* fallthru */
1390 case NFS3_CREATE_GUARDED:
1391 err = nfserr_exist;
1392 }
1393 goto out;
1394 }
1395
Al Viro6264d692006-10-19 23:28:58 -07001396 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1397 if (host_err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 goto out_nfserr;
J. Bruce Fields81ac95c2006-11-08 17:44:40 -08001399 if (created)
1400 *created = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
1402 if (EX_ISSYNC(fhp->fh_export)) {
NeilBrown45bd3b32006-01-18 17:43:50 -08001403 err = nfserrno(nfsd_sync_dir(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 /* setattr will sync the child (or not) */
1405 }
1406
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 if (createmode == NFS3_CREATE_EXCLUSIVE) {
Peter Staubachc3978522007-01-26 00:57:00 -08001408 /* Cram the verifier into atime/mtime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
Peter Staubachc3978522007-01-26 00:57:00 -08001410 | ATTR_MTIME_SET|ATTR_ATIME_SET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 /* XXX someone who knows this better please fix it for nsec */
1412 iap->ia_mtime.tv_sec = v_mtime;
1413 iap->ia_atime.tv_sec = v_atime;
1414 iap->ia_mtime.tv_nsec = 0;
1415 iap->ia_atime.tv_nsec = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 }
1417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 set_attr:
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001419 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1420 if (err2)
1421 err = err2;
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -08001422
1423 /*
1424 * Update the filehandle to get the new inode info.
1425 */
1426 if (!err)
1427 err = fh_update(resfhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 out:
1430 fh_unlock(fhp);
1431 if (dchild && !IS_ERR(dchild))
1432 dput(dchild);
1433 return err;
1434
1435 out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001436 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 goto out;
1438}
1439#endif /* CONFIG_NFSD_V3 */
1440
1441/*
1442 * Read a symlink. On entry, *lenp must contain the maximum path length that
1443 * fits into the buffer. On return, it contains the true length.
1444 * N.B. After this call fhp needs an fh_put
1445 */
Al Viro6264d692006-10-19 23:28:58 -07001446__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1448{
1449 struct dentry *dentry;
1450 struct inode *inode;
1451 mm_segment_t oldfs;
Al Viro6264d692006-10-19 23:28:58 -07001452 __be32 err;
1453 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
1455 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1456 if (err)
1457 goto out;
1458
1459 dentry = fhp->fh_dentry;
1460 inode = dentry->d_inode;
1461
1462 err = nfserr_inval;
1463 if (!inode->i_op || !inode->i_op->readlink)
1464 goto out;
1465
Jan Blunck54775492008-02-14 19:38:39 -08001466 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 /* N.B. Why does this call need a get_fs()??
1468 * Remove the set_fs and watch the fireworks:-) --okir
1469 */
1470
1471 oldfs = get_fs(); set_fs(KERNEL_DS);
Al Viro6264d692006-10-19 23:28:58 -07001472 host_err = inode->i_op->readlink(dentry, buf, *lenp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 set_fs(oldfs);
1474
Al Viro6264d692006-10-19 23:28:58 -07001475 if (host_err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 goto out_nfserr;
Al Viro6264d692006-10-19 23:28:58 -07001477 *lenp = host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 err = 0;
1479out:
1480 return err;
1481
1482out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001483 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 goto out;
1485}
1486
1487/*
1488 * Create a symlink and look up its inode
1489 * N.B. After this call _both_ fhp and resfhp need an fh_put
1490 */
Al Viro6264d692006-10-19 23:28:58 -07001491__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1493 char *fname, int flen,
1494 char *path, int plen,
1495 struct svc_fh *resfhp,
1496 struct iattr *iap)
1497{
1498 struct dentry *dentry, *dnew;
Al Viro6264d692006-10-19 23:28:58 -07001499 __be32 err, cerr;
1500 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 umode_t mode;
1502
1503 err = nfserr_noent;
1504 if (!flen || !plen)
1505 goto out;
1506 err = nfserr_exist;
1507 if (isdotent(fname, flen))
1508 goto out;
1509
1510 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1511 if (err)
1512 goto out;
1513 fh_lock(fhp);
1514 dentry = fhp->fh_dentry;
1515 dnew = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001516 host_err = PTR_ERR(dnew);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 if (IS_ERR(dnew))
1518 goto out_nfserr;
1519
1520 mode = S_IALLUGO;
1521 /* Only the MODE ATTRibute is even vaguely meaningful */
1522 if (iap && (iap->ia_valid & ATTR_MODE))
1523 mode = iap->ia_mode & S_IALLUGO;
1524
1525 if (unlikely(path[plen] != 0)) {
1526 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1527 if (path_alloced == NULL)
Al Viro6264d692006-10-19 23:28:58 -07001528 host_err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 else {
1530 strncpy(path_alloced, path, plen);
1531 path_alloced[plen] = 0;
Al Viro6264d692006-10-19 23:28:58 -07001532 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 kfree(path_alloced);
1534 }
1535 } else
Al Viro6264d692006-10-19 23:28:58 -07001536 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Al Viro6264d692006-10-19 23:28:58 -07001538 if (!host_err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if (EX_ISSYNC(fhp->fh_export))
Al Viro6264d692006-10-19 23:28:58 -07001540 host_err = nfsd_sync_dir(dentry);
1541 }
1542 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 fh_unlock(fhp);
1544
1545 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1546 dput(dnew);
1547 if (err==0) err = cerr;
1548out:
1549 return err;
1550
1551out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001552 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 goto out;
1554}
1555
1556/*
1557 * Create a hardlink
1558 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1559 */
Al Viro6264d692006-10-19 23:28:58 -07001560__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1562 char *name, int len, struct svc_fh *tfhp)
1563{
1564 struct dentry *ddir, *dnew, *dold;
1565 struct inode *dirp, *dest;
Al Viro6264d692006-10-19 23:28:58 -07001566 __be32 err;
1567 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
1569 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1570 if (err)
1571 goto out;
1572 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1573 if (err)
1574 goto out;
1575
1576 err = nfserr_perm;
1577 if (!len)
1578 goto out;
1579 err = nfserr_exist;
1580 if (isdotent(name, len))
1581 goto out;
1582
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001583 fh_lock_nested(ffhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 ddir = ffhp->fh_dentry;
1585 dirp = ddir->d_inode;
1586
1587 dnew = lookup_one_len(name, ddir, len);
Al Viro6264d692006-10-19 23:28:58 -07001588 host_err = PTR_ERR(dnew);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 if (IS_ERR(dnew))
1590 goto out_nfserr;
1591
1592 dold = tfhp->fh_dentry;
1593 dest = dold->d_inode;
1594
Al Viro6264d692006-10-19 23:28:58 -07001595 host_err = vfs_link(dold, dirp, dnew);
1596 if (!host_err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 if (EX_ISSYNC(ffhp->fh_export)) {
NeilBrown45bd3b32006-01-18 17:43:50 -08001598 err = nfserrno(nfsd_sync_dir(ddir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 write_inode_now(dest, 1);
1600 }
Al Viro6264d692006-10-19 23:28:58 -07001601 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 } else {
Al Viro6264d692006-10-19 23:28:58 -07001603 if (host_err == -EXDEV && rqstp->rq_vers == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 err = nfserr_acces;
1605 else
Al Viro6264d692006-10-19 23:28:58 -07001606 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 }
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 dput(dnew);
David M. Richter270d56e2006-06-30 01:56:15 -07001610out_unlock:
1611 fh_unlock(ffhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612out:
1613 return err;
1614
1615out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001616 err = nfserrno(host_err);
David M. Richter270d56e2006-06-30 01:56:15 -07001617 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618}
1619
1620/*
1621 * Rename a file
1622 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1623 */
Al Viro6264d692006-10-19 23:28:58 -07001624__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1626 struct svc_fh *tfhp, char *tname, int tlen)
1627{
1628 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1629 struct inode *fdir, *tdir;
Al Viro6264d692006-10-19 23:28:58 -07001630 __be32 err;
1631 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
1633 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1634 if (err)
1635 goto out;
1636 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1637 if (err)
1638 goto out;
1639
1640 fdentry = ffhp->fh_dentry;
1641 fdir = fdentry->d_inode;
1642
1643 tdentry = tfhp->fh_dentry;
1644 tdir = tdentry->d_inode;
1645
1646 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
NeilBrowna56f3932006-06-30 01:56:10 -07001647 if (ffhp->fh_export != tfhp->fh_export)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 goto out;
1649
1650 err = nfserr_perm;
1651 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1652 goto out;
1653
1654 /* cannot use fh_lock as we need deadlock protective ordering
1655 * so do it by hand */
1656 trap = lock_rename(tdentry, fdentry);
1657 ffhp->fh_locked = tfhp->fh_locked = 1;
1658 fill_pre_wcc(ffhp);
1659 fill_pre_wcc(tfhp);
1660
1661 odentry = lookup_one_len(fname, fdentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001662 host_err = PTR_ERR(odentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (IS_ERR(odentry))
1664 goto out_nfserr;
1665
Al Viro6264d692006-10-19 23:28:58 -07001666 host_err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 if (!odentry->d_inode)
1668 goto out_dput_old;
Al Viro6264d692006-10-19 23:28:58 -07001669 host_err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 if (odentry == trap)
1671 goto out_dput_old;
1672
1673 ndentry = lookup_one_len(tname, tdentry, tlen);
Al Viro6264d692006-10-19 23:28:58 -07001674 host_err = PTR_ERR(ndentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 if (IS_ERR(ndentry))
1676 goto out_dput_old;
Al Viro6264d692006-10-19 23:28:58 -07001677 host_err = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 if (ndentry == trap)
1679 goto out_dput_new;
1680
1681#ifdef MSNFS
1682 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1683 ((atomic_read(&odentry->d_count) > 1)
1684 || (atomic_read(&ndentry->d_count) > 1))) {
Al Viro6264d692006-10-19 23:28:58 -07001685 host_err = -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 } else
1687#endif
Al Viro6264d692006-10-19 23:28:58 -07001688 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1689 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1690 host_err = nfsd_sync_dir(tdentry);
1691 if (!host_err)
1692 host_err = nfsd_sync_dir(fdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 }
1694
1695 out_dput_new:
1696 dput(ndentry);
1697 out_dput_old:
1698 dput(odentry);
1699 out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001700 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
1702 /* we cannot reply on fh_unlock on the two filehandles,
1703 * as that would do the wrong thing if the two directories
1704 * were the same, so again we do it by hand
1705 */
1706 fill_post_wcc(ffhp);
1707 fill_post_wcc(tfhp);
1708 unlock_rename(tdentry, fdentry);
1709 ffhp->fh_locked = tfhp->fh_locked = 0;
1710
1711out:
1712 return err;
1713}
1714
1715/*
1716 * Unlink a file or directory
1717 * N.B. After this call fhp needs an fh_put
1718 */
Al Viro6264d692006-10-19 23:28:58 -07001719__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1721 char *fname, int flen)
1722{
1723 struct dentry *dentry, *rdentry;
1724 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001725 __be32 err;
1726 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
1728 err = nfserr_acces;
1729 if (!flen || isdotent(fname, flen))
1730 goto out;
1731 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1732 if (err)
1733 goto out;
1734
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001735 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 dentry = fhp->fh_dentry;
1737 dirp = dentry->d_inode;
1738
1739 rdentry = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001740 host_err = PTR_ERR(rdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 if (IS_ERR(rdentry))
1742 goto out_nfserr;
1743
1744 if (!rdentry->d_inode) {
1745 dput(rdentry);
1746 err = nfserr_noent;
1747 goto out;
1748 }
1749
1750 if (!type)
1751 type = rdentry->d_inode->i_mode & S_IFMT;
1752
1753 if (type != S_IFDIR) { /* It's UNLINK */
1754#ifdef MSNFS
1755 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1756 (atomic_read(&rdentry->d_count) > 1)) {
Al Viro6264d692006-10-19 23:28:58 -07001757 host_err = -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 } else
1759#endif
Al Viro6264d692006-10-19 23:28:58 -07001760 host_err = vfs_unlink(dirp, rdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 } else { /* It's RMDIR */
Al Viro6264d692006-10-19 23:28:58 -07001762 host_err = vfs_rmdir(dirp, rdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
1764
1765 dput(rdentry);
1766
Al Viro6264d692006-10-19 23:28:58 -07001767 if (host_err)
1768 goto out_nfserr;
1769 if (EX_ISSYNC(fhp->fh_export))
1770 host_err = nfsd_sync_dir(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
1772out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001773 err = nfserrno(host_err);
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -08001774out:
1775 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776}
1777
1778/*
1779 * Read entries from a directory.
1780 * The NFSv3/4 verifier we ignore for now.
1781 */
Al Viro6264d692006-10-19 23:28:58 -07001782__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
NeilBrowna0ad13e2007-01-26 00:57:10 -08001784 struct readdir_cd *cdp, filldir_t func)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
Al Viro6264d692006-10-19 23:28:58 -07001786 __be32 err;
1787 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 struct file *file;
1789 loff_t offset = *offsetp;
1790
1791 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1792 if (err)
1793 goto out;
1794
1795 offset = vfs_llseek(file, offset, 0);
1796 if (offset < 0) {
1797 err = nfserrno((int)offset);
1798 goto out_close;
1799 }
1800
1801 /*
1802 * Read the directory entries. This silly loop is necessary because
1803 * readdir() is not guaranteed to fill up the entire buffer, but
1804 * may choose to do less.
1805 */
1806
1807 do {
1808 cdp->err = nfserr_eof; /* will be cleared on successful read */
NeilBrowna0ad13e2007-01-26 00:57:10 -08001809 host_err = vfs_readdir(file, func, cdp);
Al Viro6264d692006-10-19 23:28:58 -07001810 } while (host_err >=0 && cdp->err == nfs_ok);
1811 if (host_err)
1812 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 else
1814 err = cdp->err;
1815 *offsetp = vfs_llseek(file, 0, 1);
1816
1817 if (err == nfserr_eof || err == nfserr_toosmall)
1818 err = nfs_ok; /* can still be found in ->err */
1819out_close:
1820 nfsd_close(file);
1821out:
1822 return err;
1823}
1824
1825/*
1826 * Get file system stats
1827 * N.B. After this call fhp needs an fh_put
1828 */
Al Viro6264d692006-10-19 23:28:58 -07001829__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1831{
Al Viro6264d692006-10-19 23:28:58 -07001832 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
David Howells726c3342006-06-23 02:02:58 -07001833 if (!err && vfs_statfs(fhp->fh_dentry,stat))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 err = nfserr_io;
1835 return err;
1836}
1837
J. Bruce Fieldsc7d51402007-07-19 01:49:20 -07001838static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
J. Bruce Fieldse22841c2007-07-19 01:49:20 -07001839{
J. Bruce Fieldsc7d51402007-07-19 01:49:20 -07001840 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
J. Bruce Fieldse22841c2007-07-19 01:49:20 -07001841}
1842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843/*
1844 * Check for a user's access permissions to this inode.
1845 */
Al Viro6264d692006-10-19 23:28:58 -07001846__be32
J. Bruce Fields0ec757d2007-07-17 04:04:48 -07001847nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1848 struct dentry *dentry, int acc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849{
1850 struct inode *inode = dentry->d_inode;
1851 int err;
1852
1853 if (acc == MAY_NOP)
1854 return 0;
1855#if 0
1856 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1857 acc,
1858 (acc & MAY_READ)? " read" : "",
1859 (acc & MAY_WRITE)? " write" : "",
1860 (acc & MAY_EXEC)? " exec" : "",
1861 (acc & MAY_SATTR)? " sattr" : "",
1862 (acc & MAY_TRUNC)? " trunc" : "",
1863 (acc & MAY_LOCK)? " lock" : "",
1864 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1865 inode->i_mode,
1866 IS_IMMUTABLE(inode)? " immut" : "",
1867 IS_APPEND(inode)? " append" : "",
1868 IS_RDONLY(inode)? " ro" : "");
1869 dprintk(" owner %d/%d user %d/%d\n",
1870 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1871#endif
1872
1873 /* Normally we reject any write/sattr etc access on a read-only file
1874 * system. But if it is IRIX doing check on write-access for a
1875 * device special file, we ignore rofs.
1876 */
1877 if (!(acc & MAY_LOCAL_ACCESS))
1878 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
J. Bruce Fieldsc7d51402007-07-19 01:49:20 -07001879 if (exp_rdonly(rqstp, exp) || IS_RDONLY(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 return nfserr_rofs;
1881 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1882 return nfserr_perm;
1883 }
1884 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1885 return nfserr_perm;
1886
1887 if (acc & MAY_LOCK) {
1888 /* If we cannot rely on authentication in NLM requests,
1889 * just allow locks, otherwise require read permission, or
1890 * ownership
1891 */
1892 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1893 return 0;
1894 else
1895 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1896 }
1897 /*
1898 * The file owner always gets access permission for accesses that
1899 * would normally be checked at open time. This is to make
1900 * file access work even when the client has done a fchmod(fd, 0).
1901 *
1902 * However, `cp foo bar' should fail nevertheless when bar is
1903 * readonly. A sensible way to do this might be to reject all
1904 * attempts to truncate a read-only file, because a creat() call
1905 * always implies file truncation.
1906 * ... but this isn't really fair. A process may reasonably call
1907 * ftruncate on an open file descriptor on a file with perm 000.
1908 * We must trust the client to do permission checking - using "ACCESS"
1909 * with NFSv3.
1910 */
1911 if ((acc & MAY_OWNER_OVERRIDE) &&
1912 inode->i_uid == current->fsuid)
1913 return 0;
1914
1915 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1916
1917 /* Allow read access to binaries even when mode 111 */
1918 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1919 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1920 err = permission(inode, MAY_EXEC, NULL);
1921
1922 return err? nfserrno(err) : 0;
1923}
1924
1925void
1926nfsd_racache_shutdown(void)
1927{
Greg Banksfce14562006-10-04 02:15:49 -07001928 if (!raparml)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 return;
1930 dprintk("nfsd: freeing readahead buffers.\n");
1931 kfree(raparml);
Greg Banksfce14562006-10-04 02:15:49 -07001932 raparml = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933}
1934/*
1935 * Initialize readahead param cache
1936 */
1937int
1938nfsd_racache_init(int cache_size)
1939{
1940 int i;
Greg Banksfce14562006-10-04 02:15:49 -07001941 int j = 0;
1942 int nperbucket;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
Greg Banksfce14562006-10-04 02:15:49 -07001944
1945 if (raparml)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 return 0;
Greg Banksfce14562006-10-04 02:15:49 -07001947 if (cache_size < 2*RAPARM_HASH_SIZE)
1948 cache_size = 2*RAPARM_HASH_SIZE;
Yan Burman4b3bb062006-12-08 02:39:41 -08001949 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
Yan Burman4b3bb062006-12-08 02:39:41 -08001951 if (!raparml) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 printk(KERN_WARNING
Yan Burman4b3bb062006-12-08 02:39:41 -08001953 "nfsd: Could not allocate memory read-ahead cache.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return -ENOMEM;
1955 }
Yan Burman4b3bb062006-12-08 02:39:41 -08001956
1957 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1958 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1959 raparm_hash[i].pb_head = NULL;
1960 spin_lock_init(&raparm_hash[i].pb_lock);
1961 }
J. Bruce Fieldsa280df32007-07-19 01:49:18 -07001962 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
Yan Burman4b3bb062006-12-08 02:39:41 -08001963 for (i = 0; i < cache_size - 1; i++) {
1964 if (i % nperbucket == 0)
1965 raparm_hash[j++].pb_head = raparml + i;
1966 if (i % nperbucket < nperbucket-1)
1967 raparml[i].p_next = raparml + i + 1;
1968 }
1969
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 nfsdstats.ra_size = cache_size;
1971 return 0;
1972}
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001973
1974#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1975struct posix_acl *
1976nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1977{
1978 struct inode *inode = fhp->fh_dentry->d_inode;
1979 char *name;
1980 void *value = NULL;
1981 ssize_t size;
1982 struct posix_acl *acl;
1983
Christoph Hellwig5be196e2006-01-09 20:51:55 -08001984 if (!IS_POSIXACL(inode))
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001985 return ERR_PTR(-EOPNOTSUPP);
Christoph Hellwig5be196e2006-01-09 20:51:55 -08001986
1987 switch (type) {
1988 case ACL_TYPE_ACCESS:
1989 name = POSIX_ACL_XATTR_ACCESS;
1990 break;
1991 case ACL_TYPE_DEFAULT:
1992 name = POSIX_ACL_XATTR_DEFAULT;
1993 break;
1994 default:
1995 return ERR_PTR(-EOPNOTSUPP);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00001996 }
1997
Christoph Hellwig5be196e2006-01-09 20:51:55 -08001998 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1999 if (size < 0)
2000 return ERR_PTR(size);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002001
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002002 acl = posix_acl_from_xattr(value, size);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002003 kfree(value);
2004 return acl;
2005}
2006
2007int
2008nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2009{
2010 struct inode *inode = fhp->fh_dentry->d_inode;
2011 char *name;
2012 void *value = NULL;
2013 size_t size;
2014 int error;
2015
2016 if (!IS_POSIXACL(inode) || !inode->i_op ||
2017 !inode->i_op->setxattr || !inode->i_op->removexattr)
2018 return -EOPNOTSUPP;
2019 switch(type) {
2020 case ACL_TYPE_ACCESS:
Christoph Hellwig334a13e2005-06-28 20:44:58 -07002021 name = POSIX_ACL_XATTR_ACCESS;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002022 break;
2023 case ACL_TYPE_DEFAULT:
Christoph Hellwig334a13e2005-06-28 20:44:58 -07002024 name = POSIX_ACL_XATTR_DEFAULT;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002025 break;
2026 default:
2027 return -EOPNOTSUPP;
2028 }
2029
2030 if (acl && acl->a_count) {
Christoph Hellwig334a13e2005-06-28 20:44:58 -07002031 size = posix_acl_xattr_size(acl->a_count);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002032 value = kmalloc(size, GFP_KERNEL);
2033 if (!value)
2034 return -ENOMEM;
Florin Malita9ccfc292006-05-20 14:59:58 -07002035 error = posix_acl_to_xattr(acl, value, size);
2036 if (error < 0)
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002037 goto getout;
Florin Malita9ccfc292006-05-20 14:59:58 -07002038 size = error;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002039 } else
2040 size = 0;
2041
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002042 if (size)
Christoph Hellwig5be196e2006-01-09 20:51:55 -08002043 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002044 else {
2045 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2046 error = 0;
2047 else {
Christoph Hellwig5be196e2006-01-09 20:51:55 -08002048 error = vfs_removexattr(fhp->fh_dentry, name);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002049 if (error == -ENODATA)
2050 error = 0;
2051 }
2052 }
2053
2054getout:
2055 kfree(value);
2056 return error;
2057}
2058#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */