blob: 6e50aaa56ca25222ab71e0b92cdd66609b5c5b61 [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
Greg Banksfce14562006-10-04 02:15:49 -070086#define RAPARM_HASH_BITS 4
87#define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
88#define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
89static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
91/*
92 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
93 * a mount point.
J.Bruce Fieldse0bb89e2006-12-13 00:35:25 -080094 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * or nfs_ok having possibly changed *dpp and *expp
96 */
97int
98nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
99 struct svc_export **expp)
100{
101 struct svc_export *exp = *expp, *exp2 = NULL;
102 struct dentry *dentry = *dpp;
Jan Blunck54775492008-02-14 19:38:39 -0800103 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 struct dentry *mounts = dget(dentry);
Al Viro6264d692006-10-19 23:28:58 -0700105 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
108
J. Bruce Fields0989a782007-07-17 04:04:44 -0700109 exp2 = rqst_exp_get_by_name(rqstp, mnt, mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 if (IS_ERR(exp2)) {
Neil Browna1033be2007-09-05 17:22:12 -0400111 if (PTR_ERR(exp2) != -ENOENT)
112 err = PTR_ERR(exp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 dput(mounts);
114 mntput(mnt);
115 goto out;
116 }
J. Bruce Fields5d3dbbe2007-07-19 01:49:19 -0700117 if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 /* successfully crossed mount point */
119 exp_put(exp);
120 *expp = exp2;
121 dput(dentry);
122 *dpp = mounts;
123 } else {
J. Bruce Fields5d3dbbe2007-07-19 01:49:19 -0700124 exp_put(exp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 dput(mounts);
126 }
127 mntput(mnt);
128out:
129 return err;
130}
131
Al Viro6264d692006-10-19 23:28:58 -0700132__be32
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700133nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
Chuck Lever5a022fc2007-11-01 16:57:09 -0400134 const char *name, unsigned int len,
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700135 struct svc_export **exp_ret, struct dentry **dentry_ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
137 struct svc_export *exp;
138 struct dentry *dparent;
139 struct dentry *dentry;
Al Viro6264d692006-10-19 23:28:58 -0700140 __be32 err;
141 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
143 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
144
145 /* Obtain dentry and export. */
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200146 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 if (err)
148 return err;
149
150 dparent = fhp->fh_dentry;
151 exp = fhp->fh_export;
152 exp_get(exp);
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 /* Lookup the name, but don't follow links */
155 if (isdotent(name, len)) {
156 if (len==1)
157 dentry = dget(dparent);
Jan Blunck54775492008-02-14 19:38:39 -0800158 else if (dparent != exp->ex_path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 dentry = dget_parent(dparent);
Jan Blunck54775492008-02-14 19:38:39 -0800160 else if (!EX_NOHIDE(exp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 dentry = dget(dparent); /* .. == . just like at / */
162 else {
163 /* checking mountpoint crossing is very different when stepping up */
164 struct svc_export *exp2 = NULL;
165 struct dentry *dp;
Jan Blunck54775492008-02-14 19:38:39 -0800166 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 dentry = dget(dparent);
168 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
169 ;
170 dp = dget_parent(dentry);
171 dput(dentry);
172 dentry = dp;
173
J. Bruce Fields0989a782007-07-17 04:04:44 -0700174 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
J. Bruce Fields2d3bb252007-07-17 04:04:40 -0700175 if (PTR_ERR(exp2) == -ENOENT) {
176 dput(dentry);
177 dentry = dget(dparent);
178 } else if (IS_ERR(exp2)) {
Al Viro6264d692006-10-19 23:28:58 -0700179 host_err = PTR_ERR(exp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 dput(dentry);
181 mntput(mnt);
182 goto out_nfserr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 } else {
184 exp_put(exp);
185 exp = exp2;
186 }
187 mntput(mnt);
188 }
189 } else {
190 fh_lock(fhp);
191 dentry = lookup_one_len(name, dparent, len);
Al Viro6264d692006-10-19 23:28:58 -0700192 host_err = PTR_ERR(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 if (IS_ERR(dentry))
194 goto out_nfserr;
195 /*
196 * check if we have crossed a mount point ...
197 */
198 if (d_mountpoint(dentry)) {
Al Viro6264d692006-10-19 23:28:58 -0700199 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 dput(dentry);
201 goto out_nfserr;
202 }
203 }
204 }
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700205 *dentry_ret = dentry;
206 *exp_ret = exp;
207 return 0;
208
209out_nfserr:
210 exp_put(exp);
211 return nfserrno(host_err);
212}
213
214/*
215 * Look up one component of a pathname.
216 * N.B. After this call _both_ fhp and resfh need an fh_put
217 *
218 * If the lookup would cross a mountpoint, and the mounted filesystem
219 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
220 * accepted as it stands and the mounted directory is
221 * returned. Otherwise the covered directory is returned.
222 * NOTE: this mountpoint crossing is not supported properly by all
223 * clients and is explicitly disallowed for NFSv3
224 * NeilBrown <neilb@cse.unsw.edu.au>
225 */
226__be32
227nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
Chuck Lever5a022fc2007-11-01 16:57:09 -0400228 unsigned int len, struct svc_fh *resfh)
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700229{
230 struct svc_export *exp;
231 struct dentry *dentry;
232 __be32 err;
233
234 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
235 if (err)
236 return err;
Andy Adamson32c1eb02007-07-17 04:04:48 -0700237 err = check_nfsd_access(exp, rqstp);
238 if (err)
239 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 /*
241 * Note: we compose the file handle now, but as the
242 * dentry may be negative, it may need to be updated.
243 */
244 err = fh_compose(resfh, exp, dentry, fhp);
245 if (!err && !dentry->d_inode)
246 err = nfserr_noent;
Andy Adamson32c1eb02007-07-17 04:04:48 -0700247out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 dput(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 exp_put(exp);
250 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700253
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254/*
255 * Set various file attributes.
256 * N.B. After this call fhp needs an fh_put
257 */
Al Viro6264d692006-10-19 23:28:58 -0700258__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
260 int check_guard, time_t guardtime)
261{
262 struct dentry *dentry;
263 struct inode *inode;
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200264 int accmode = NFSD_MAY_SATTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 int ftype = 0;
Al Viro6264d692006-10-19 23:28:58 -0700266 __be32 err;
267 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 int size_change = 0;
269
270 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200271 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 if (iap->ia_valid & ATTR_SIZE)
273 ftype = S_IFREG;
274
275 /* Get inode */
276 err = fh_verify(rqstp, fhp, ftype, accmode);
NeilBrown15b7a1b2005-11-07 01:00:23 -0800277 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 goto out;
279
280 dentry = fhp->fh_dentry;
281 inode = dentry->d_inode;
282
NeilBrown15b7a1b2005-11-07 01:00:23 -0800283 /* Ignore any mode updates on symlinks */
284 if (S_ISLNK(inode->i_mode))
285 iap->ia_valid &= ~ATTR_MODE;
286
287 if (!iap->ia_valid)
288 goto out;
289
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000290 /*
291 * NFSv2 does not differentiate between "set-[ac]time-to-now"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 * which only requires access, and "set-[ac]time-to-X" which
293 * requires ownership.
294 * So if it looks like it might be "set both to the same time which
295 * is close to now", and if inode_change_ok fails, then we
296 * convert to "set to now" instead of "set to explicit time"
297 *
298 * We only call inode_change_ok as the last test as technically
299 * it is not an interface that we should be using. It is only
300 * valid if the filesystem does not define it's own i_op->setattr.
301 */
302#define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
303#define MAX_TOUCH_TIME_ERROR (30*60)
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000304 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
305 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
306 /*
307 * Looks probable.
308 *
309 * Now just make sure time is in the right ballpark.
310 * Solaris, at least, doesn't seem to care what the time
311 * request is. We require it be within 30 minutes of now.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 */
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000313 time_t delta = iap->ia_atime.tv_sec - get_seconds();
314 if (delta < 0)
315 delta = -delta;
316 if (delta < MAX_TOUCH_TIME_ERROR &&
317 inode_change_ok(inode, iap) != 0) {
318 /*
319 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
320 * This will cause notify_change to set these times
321 * to "now"
322 */
323 iap->ia_valid &= ~BOTH_TIME_SET;
324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 }
326
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000327 /*
328 * The size case is special.
329 * It changes the file as well as the attributes.
330 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 if (iap->ia_valid & ATTR_SIZE) {
332 if (iap->ia_size < inode->i_size) {
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200333 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
334 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 if (err)
336 goto out;
337 }
338
339 /*
340 * If we are changing the size of the file, then
341 * we need to break all leases.
342 */
Al Viro6264d692006-10-19 23:28:58 -0700343 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
344 if (host_err == -EWOULDBLOCK)
345 host_err = -ETIMEDOUT;
346 if (host_err) /* ENOMEM or EWOULDBLOCK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 goto out_nfserr;
348
Al Viro6264d692006-10-19 23:28:58 -0700349 host_err = get_write_access(inode);
350 if (host_err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 goto out_nfserr;
352
353 size_change = 1;
Al Viro6264d692006-10-19 23:28:58 -0700354 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
355 if (host_err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 put_write_access(inode);
357 goto out_nfserr;
358 }
359 DQUOT_INIT(inode);
360 }
361
Jeff Laytonca456252008-04-16 16:28:47 -0400362 /* sanitize the mode change */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 if (iap->ia_valid & ATTR_MODE) {
364 iap->ia_mode &= S_IALLUGO;
Jeff Laytondee32092008-04-16 16:28:46 -0400365 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
Jeff Laytonca456252008-04-16 16:28:47 -0400366 }
367
368 /* Revoke setuid/setgid on chown */
369 if (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
370 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)) {
371 iap->ia_valid |= ATTR_KILL_PRIV;
372 if (iap->ia_valid & ATTR_MODE) {
373 /* we're setting mode too, just clear the s*id bits */
Jeff Layton8a0ce7d2007-10-18 03:05:19 -0700374 iap->ia_mode &= ~S_ISUID;
Jeff Laytonca456252008-04-16 16:28:47 -0400375 if (iap->ia_mode & S_IXGRP)
376 iap->ia_mode &= ~S_ISGID;
377 } else {
378 /* set ATTR_KILL_* bits and let VFS handle it */
379 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
Jeff Layton8a0ce7d2007-10-18 03:05:19 -0700380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 }
382
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 /* Change the attributes. */
384
385 iap->ia_valid |= ATTR_CTIME;
386
387 err = nfserr_notsync;
388 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
389 fh_lock(fhp);
Al Viro6264d692006-10-19 23:28:58 -0700390 host_err = notify_change(dentry, iap);
391 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 fh_unlock(fhp);
393 }
394 if (size_change)
395 put_write_access(inode);
396 if (!err)
397 if (EX_ISSYNC(fhp->fh_export))
398 write_inode_now(inode, 1);
399out:
400 return err;
401
402out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -0700403 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 goto out;
405}
406
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800407#if defined(CONFIG_NFSD_V2_ACL) || \
408 defined(CONFIG_NFSD_V3_ACL) || \
409 defined(CONFIG_NFSD_V4)
410static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
411{
412 ssize_t buflen;
Krishna Kumar6c6a4262008-10-22 14:48:36 +0530413 ssize_t ret;
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
Krishna Kumar6c6a4262008-10-22 14:48:36 +0530423 ret = vfs_getxattr(dentry, key, *buf, buflen);
424 if (ret < 0)
425 kfree(*buf);
426 return ret;
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800427}
428#endif
429
430#if defined(CONFIG_NFSD_V4)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431static int
432set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
433{
434 int len;
435 size_t buflen;
436 char *buf = NULL;
437 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439 buflen = posix_acl_xattr_size(pacl->a_count);
440 buf = kmalloc(buflen, GFP_KERNEL);
441 error = -ENOMEM;
442 if (buf == NULL)
443 goto out;
444
445 len = posix_acl_to_xattr(pacl, buf, buflen);
446 if (len < 0) {
447 error = len;
448 goto out;
449 }
450
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800451 error = vfs_setxattr(dentry, key, buf, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452out:
453 kfree(buf);
454 return error;
455}
456
Al Viro6264d692006-10-19 23:28:58 -0700457__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
459 struct nfs4_acl *acl)
460{
Al Viro6264d692006-10-19 23:28:58 -0700461 __be32 error;
462 int host_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 struct dentry *dentry;
464 struct inode *inode;
465 struct posix_acl *pacl = NULL, *dpacl = NULL;
466 unsigned int flags = 0;
467
468 /* Get inode */
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200469 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 if (error)
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700471 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 dentry = fhp->fh_dentry;
474 inode = dentry->d_inode;
475 if (S_ISDIR(inode->i_mode))
476 flags = NFS4_ACL_DIR;
477
Al Viro6264d692006-10-19 23:28:58 -0700478 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
479 if (host_error == -EINVAL) {
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700480 return nfserr_attrnotsupp;
Al Viro6264d692006-10-19 23:28:58 -0700481 } else if (host_error < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 goto out_nfserr;
483
Al Viro6264d692006-10-19 23:28:58 -0700484 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
485 if (host_error < 0)
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700486 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700488 if (S_ISDIR(inode->i_mode))
Al Viro6264d692006-10-19 23:28:58 -0700489 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700491out_release:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 posix_acl_release(pacl);
493 posix_acl_release(dpacl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494out_nfserr:
J. Bruce Fieldsf34f9242007-02-16 01:28:34 -0800495 if (host_error == -EOPNOTSUPP)
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700496 return nfserr_attrnotsupp;
J. Bruce Fieldsf34f9242007-02-16 01:28:34 -0800497 else
J. Bruce Fields4b2ca382007-07-17 04:04:37 -0700498 return nfserrno(host_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499}
500
501static struct posix_acl *
502_get_posix_acl(struct dentry *dentry, char *key)
503{
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800504 void *buf = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 struct posix_acl *pacl = NULL;
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800506 int buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800508 buflen = nfsd_getxattr(dentry, key, &buf);
509 if (!buflen)
510 buflen = -ENODATA;
511 if (buflen <= 0)
512 return ERR_PTR(buflen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514 pacl = posix_acl_from_xattr(buf, buflen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 kfree(buf);
516 return pacl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517}
518
519int
520nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
521{
522 struct inode *inode = dentry->d_inode;
523 int error = 0;
524 struct posix_acl *pacl = NULL, *dpacl = NULL;
525 unsigned int flags = 0;
526
Christoph Hellwig9a59f452005-06-23 00:10:19 -0700527 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
529 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
530 if (IS_ERR(pacl)) {
531 error = PTR_ERR(pacl);
532 pacl = NULL;
533 goto out;
534 }
535
536 if (S_ISDIR(inode->i_mode)) {
Christoph Hellwig9a59f452005-06-23 00:10:19 -0700537 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
539 dpacl = NULL;
540 else if (IS_ERR(dpacl)) {
541 error = PTR_ERR(dpacl);
542 dpacl = NULL;
543 goto out;
544 }
545 flags = NFS4_ACL_DIR;
546 }
547
548 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
549 if (IS_ERR(*acl)) {
550 error = PTR_ERR(*acl);
551 *acl = NULL;
552 }
553 out:
554 posix_acl_release(pacl);
555 posix_acl_release(dpacl);
556 return error;
557}
558
559#endif /* defined(CONFIG_NFS_V4) */
560
561#ifdef CONFIG_NFSD_V3
562/*
563 * Check server access rights to a file system object
564 */
565struct accessmap {
566 u32 access;
567 int how;
568};
569static struct accessmap nfs3_regaccess[] = {
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200570 { NFS3_ACCESS_READ, NFSD_MAY_READ },
571 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
572 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
573 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 { 0, 0 }
576};
577
578static struct accessmap nfs3_diraccess[] = {
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200579 { NFS3_ACCESS_READ, NFSD_MAY_READ },
580 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
581 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
582 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
583 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 { 0, 0 }
586};
587
588static struct accessmap nfs3_anyaccess[] = {
589 /* Some clients - Solaris 2.6 at least, make an access call
590 * to the server to check for access for things like /dev/null
591 * (which really, the server doesn't care about). So
592 * We provide simple access checking for them, looking
593 * mainly at mode bits, and we make sure to ignore read-only
594 * filesystem checks
595 */
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200596 { NFS3_ACCESS_READ, NFSD_MAY_READ },
597 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
598 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
599 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601 { 0, 0 }
602};
603
Al Viro6264d692006-10-19 23:28:58 -0700604__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
606{
607 struct accessmap *map;
608 struct svc_export *export;
609 struct dentry *dentry;
610 u32 query, result = 0, sresult = 0;
Al Viro6264d692006-10-19 23:28:58 -0700611 __be32 error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200613 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (error)
615 goto out;
616
617 export = fhp->fh_export;
618 dentry = fhp->fh_dentry;
619
620 if (S_ISREG(dentry->d_inode->i_mode))
621 map = nfs3_regaccess;
622 else if (S_ISDIR(dentry->d_inode->i_mode))
623 map = nfs3_diraccess;
624 else
625 map = nfs3_anyaccess;
626
627
628 query = *access;
629 for (; map->access; map++) {
630 if (map->access & query) {
Al Viro6264d692006-10-19 23:28:58 -0700631 __be32 err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 sresult |= map->access;
634
J. Bruce Fields0ec757d2007-07-17 04:04:48 -0700635 err2 = nfsd_permission(rqstp, export, dentry, map->how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 switch (err2) {
637 case nfs_ok:
638 result |= map->access;
639 break;
640
641 /* the following error codes just mean the access was not allowed,
642 * rather than an error occurred */
643 case nfserr_rofs:
644 case nfserr_acces:
645 case nfserr_perm:
646 /* simply don't "or" in the access bit. */
647 break;
648 default:
649 error = err2;
650 goto out;
651 }
652 }
653 }
654 *access = result;
655 if (supported)
656 *supported = sresult;
657
658 out:
659 return error;
660}
661#endif /* CONFIG_NFSD_V3 */
662
663
664
665/*
666 * Open an existing file or directory.
667 * The access argument indicates the type of open (read/write/lock)
668 * N.B. After this call fhp needs an fh_put
669 */
Al Viro6264d692006-10-19 23:28:58 -0700670__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
672 int access, struct file **filp)
673{
David Howells745ca242008-11-14 10:39:22 +1100674 const struct cred *cred = current_cred();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 struct dentry *dentry;
676 struct inode *inode;
Al Viro6264d692006-10-19 23:28:58 -0700677 int flags = O_RDONLY|O_LARGEFILE;
678 __be32 err;
679 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
681 /*
682 * If we get here, then the client has already done an "open",
683 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
684 * in case a chmod has now revoked permission.
685 */
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200686 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 if (err)
688 goto out;
689
690 dentry = fhp->fh_dentry;
691 inode = dentry->d_inode;
692
693 /* Disallow write access to files with the append-only bit set
694 * or any access when mandatory locking enabled
695 */
696 err = nfserr_perm;
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200697 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 goto out;
J. Bruce Fields5e7fc432007-10-02 14:18:12 -0400699 /*
700 * We must ignore files (but only files) which might have mandatory
701 * locks on them because there is no way to know if the accesser has
702 * the lock.
703 */
704 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 goto out;
706
707 if (!inode->i_fop)
708 goto out;
709
710 /*
711 * Check to see if there are any leases on this file.
712 * This may block while leases are broken.
713 */
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200714 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
Al Viro6264d692006-10-19 23:28:58 -0700715 if (host_err == -EWOULDBLOCK)
716 host_err = -ETIMEDOUT;
717 if (host_err) /* NOMEM or WOULDBLOCK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 goto out_nfserr;
719
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200720 if (access & NFSD_MAY_WRITE) {
721 if (access & NFSD_MAY_READ)
J. Bruce Fields9ecb6a02006-06-30 01:56:17 -0700722 flags = O_RDWR|O_LARGEFILE;
723 else
724 flags = O_WRONLY|O_LARGEFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726 DQUOT_INIT(inode);
727 }
Jan Blunck54775492008-02-14 19:38:39 -0800728 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
David Howells745ca242008-11-14 10:39:22 +1100729 flags, cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 if (IS_ERR(*filp))
Al Viro6264d692006-10-19 23:28:58 -0700731 host_err = PTR_ERR(*filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -0700733 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734out:
735 return err;
736}
737
738/*
739 * Close a file.
740 */
741void
742nfsd_close(struct file *filp)
743{
744 fput(filp);
745}
746
J. Bruce Fields9a8d2482009-01-06 13:37:03 -0500747/*
748 * Sync a file
749 * As this calls fsync (not fdatasync) there is no need for a write_inode
750 * after it.
751 */
752static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
753 const struct file_operations *fop)
754{
755 struct inode *inode = dp->d_inode;
756 int (*fsync) (struct file *, struct dentry *, int);
757 int err;
758
759 err = filemap_fdatawrite(inode->i_mapping);
760 if (err == 0 && fop && (fsync = fop->fsync))
761 err = fsync(filp, dp, 0);
762 if (err == 0)
763 err = filemap_fdatawait(inode->i_mapping);
764
765 return err;
766}
767
David Shawa334de22006-01-06 00:19:58 -0800768static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769nfsd_sync(struct file *filp)
770{
J. Bruce Fields9a8d2482009-01-06 13:37:03 -0500771 int err;
772 struct inode *inode = filp->f_path.dentry->d_inode;
773 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
774 mutex_lock(&inode->i_mutex);
775 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
776 mutex_unlock(&inode->i_mutex);
777
778 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779}
780
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -0800781int
J. Bruce Fields9a8d2482009-01-06 13:37:03 -0500782nfsd_sync_dir(struct dentry *dp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
J. Bruce Fields9a8d2482009-01-06 13:37:03 -0500784 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
786
787/*
788 * Obtain the readahead parameters for the file
789 * specified by (dev, ino).
790 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792static inline struct raparms *
793nfsd_get_raparms(dev_t dev, ino_t ino)
794{
795 struct raparms *ra, **rap, **frap = NULL;
796 int depth = 0;
Greg Banksfce14562006-10-04 02:15:49 -0700797 unsigned int hash;
798 struct raparm_hbucket *rab;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Greg Banksfce14562006-10-04 02:15:49 -0700800 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
801 rab = &raparm_hash[hash];
802
803 spin_lock(&rab->pb_lock);
804 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 if (ra->p_ino == ino && ra->p_dev == dev)
806 goto found;
807 depth++;
808 if (ra->p_count == 0)
809 frap = rap;
810 }
811 depth = nfsdstats.ra_size*11/10;
812 if (!frap) {
Greg Banksfce14562006-10-04 02:15:49 -0700813 spin_unlock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 return NULL;
815 }
816 rap = frap;
817 ra = *frap;
818 ra->p_dev = dev;
819 ra->p_ino = ino;
820 ra->p_set = 0;
Greg Banksfce14562006-10-04 02:15:49 -0700821 ra->p_hindex = hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822found:
Greg Banksfce14562006-10-04 02:15:49 -0700823 if (rap != &rab->pb_head) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 *rap = ra->p_next;
Greg Banksfce14562006-10-04 02:15:49 -0700825 ra->p_next = rab->pb_head;
826 rab->pb_head = ra;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
828 ra->p_count++;
829 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
Greg Banksfce14562006-10-04 02:15:49 -0700830 spin_unlock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 return ra;
832}
833
834/*
Jens Axboecf8208d2007-06-12 21:22:14 +0200835 * Grab and keep cached pages associated with a file in the svc_rqst
836 * so that they can be passed to the network sendmsg/sendpage routines
837 * directly. They will be released after the sending has completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 */
839static int
Jens Axboecf8208d2007-06-12 21:22:14 +0200840nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
841 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
Jens Axboecf8208d2007-06-12 21:22:14 +0200843 struct svc_rqst *rqstp = sd->u.data;
NeilBrown44524352006-10-04 02:15:46 -0700844 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
Jens Axboecf8208d2007-06-12 21:22:14 +0200845 struct page *page = buf->page;
846 size_t size;
847 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
Jens Axboecac36bb02007-06-14 13:10:48 +0200849 ret = buf->ops->confirm(pipe, buf);
Jens Axboecf8208d2007-06-12 21:22:14 +0200850 if (unlikely(ret))
851 return ret;
852
853 size = sd->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 if (rqstp->rq_res.page_len == 0) {
856 get_page(page);
NeilBrown44524352006-10-04 02:15:46 -0700857 put_page(*pp);
858 *pp = page;
859 rqstp->rq_resused++;
Jens Axboecf8208d2007-06-12 21:22:14 +0200860 rqstp->rq_res.page_base = buf->offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 rqstp->rq_res.page_len = size;
NeilBrown44524352006-10-04 02:15:46 -0700862 } else if (page != pp[-1]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 get_page(page);
NeilBrown250f3912007-01-26 00:56:59 -0800864 if (*pp)
865 put_page(*pp);
NeilBrown44524352006-10-04 02:15:46 -0700866 *pp = page;
867 rqstp->rq_resused++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 rqstp->rq_res.page_len += size;
NeilBrown44524352006-10-04 02:15:46 -0700869 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 rqstp->rq_res.page_len += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return size;
873}
874
Jens Axboecf8208d2007-06-12 21:22:14 +0200875static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
876 struct splice_desc *sd)
877{
878 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
879}
880
Dave Hansena8754be2007-10-16 23:31:15 -0700881static inline int svc_msnfs(struct svc_fh *ffhp)
882{
883#ifdef MSNFS
884 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
885#else
886 return 0;
887#endif
888}
889
Al Viro6264d692006-10-19 23:28:58 -0700890static __be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
892 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
893{
894 struct inode *inode;
895 struct raparms *ra;
896 mm_segment_t oldfs;
Al Viro6264d692006-10-19 23:28:58 -0700897 __be32 err;
898 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
900 err = nfserr_perm;
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800901 inode = file->f_path.dentry->d_inode;
Dave Hansena8754be2007-10-16 23:31:15 -0700902
903 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 /* Get readahead parameters */
907 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
908
909 if (ra && ra->p_set)
910 file->f_ra = ra->p_ra;
911
Jens Axboecf8208d2007-06-12 21:22:14 +0200912 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
913 struct splice_desc sd = {
914 .len = 0,
915 .total_len = *count,
916 .pos = offset,
917 .u.data = rqstp,
918 };
919
Jens Axboe4fbef202007-07-13 22:42:20 +0200920 rqstp->rq_resused = 1;
Jens Axboecf8208d2007-06-12 21:22:14 +0200921 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 } else {
923 oldfs = get_fs();
924 set_fs(KERNEL_DS);
Al Viro6264d692006-10-19 23:28:58 -0700925 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 set_fs(oldfs);
927 }
928
929 /* Write back readahead params */
930 if (ra) {
Greg Banksfce14562006-10-04 02:15:49 -0700931 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
932 spin_lock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 ra->p_ra = file->f_ra;
934 ra->p_set = 1;
935 ra->p_count--;
Greg Banksfce14562006-10-04 02:15:49 -0700936 spin_unlock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 }
938
Al Viro6264d692006-10-19 23:28:58 -0700939 if (host_err >= 0) {
940 nfsdstats.io_read += host_err;
941 *count = host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 err = 0;
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800943 fsnotify_access(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 } else
Al Viro6264d692006-10-19 23:28:58 -0700945 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946out:
947 return err;
948}
949
Neil Brown9f708e42006-01-06 00:19:59 -0800950static void kill_suid(struct dentry *dentry)
951{
952 struct iattr ia;
Serge E. Hallynb5376772007-10-16 23:31:36 -0700953 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
Neil Brown9f708e42006-01-06 00:19:59 -0800954
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800955 mutex_lock(&dentry->d_inode->i_mutex);
Neil Brown9f708e42006-01-06 00:19:59 -0800956 notify_change(dentry, &ia);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800957 mutex_unlock(&dentry->d_inode->i_mutex);
Neil Brown9f708e42006-01-06 00:19:59 -0800958}
959
Al Viro6264d692006-10-19 23:28:58 -0700960static __be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
962 loff_t offset, struct kvec *vec, int vlen,
963 unsigned long cnt, int *stablep)
964{
965 struct svc_export *exp;
966 struct dentry *dentry;
967 struct inode *inode;
968 mm_segment_t oldfs;
Al Viro6264d692006-10-19 23:28:58 -0700969 __be32 err = 0;
970 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 int stable = *stablep;
972
NeilBrown45bd3b32006-01-18 17:43:50 -0800973#ifdef MSNFS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 err = nfserr_perm;
975
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800977 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 goto out;
979#endif
980
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800981 dentry = file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 inode = dentry->d_inode;
983 exp = fhp->fh_export;
984
985 /*
986 * Request sync writes if
987 * - the sync export option has been set, or
988 * - the client requested O_SYNC behavior (NFSv3 feature).
989 * - The file system doesn't support fsync().
990 * When gathered writes have been configured for this volume,
991 * flushing the data to disk is handled separately below.
992 */
993
Harvey Harrison3ba15142008-02-20 12:49:02 -0800994 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 stable = 2;
996 *stablep = 2; /* FILE_SYNC */
997 }
998
999 if (!EX_ISSYNC(exp))
1000 stable = 0;
1001 if (stable && !EX_WGATHER(exp))
1002 file->f_flags |= O_SYNC;
1003
1004 /* Write the data. */
1005 oldfs = get_fs(); set_fs(KERNEL_DS);
Al Viro6264d692006-10-19 23:28:58 -07001006 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 set_fs(oldfs);
Al Viro6264d692006-10-19 23:28:58 -07001008 if (host_err >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 nfsdstats.io_write += cnt;
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -08001010 fsnotify_modify(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 }
1012
1013 /* clear setuid/setgid flag after write */
Al Viro6264d692006-10-19 23:28:58 -07001014 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
Neil Brown9f708e42006-01-06 00:19:59 -08001015 kill_suid(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Al Viro6264d692006-10-19 23:28:58 -07001017 if (host_err >= 0 && stable) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 static ino_t last_ino;
1019 static dev_t last_dev;
1020
1021 /*
1022 * Gathered writes: If another process is currently
1023 * writing to the file, there's a high chance
1024 * this is another nfsd (triggered by a bulk write
1025 * from a client's biod). Rather than syncing the
1026 * file with each write request, we sleep for 10 msec.
1027 *
1028 * I don't know if this roughly approximates
1029 * C. Juszak's idea of gathered writes, but it's a
1030 * nice and simple solution (IMHO), and it seems to
1031 * work:-)
1032 */
1033 if (EX_WGATHER(exp)) {
1034 if (atomic_read(&inode->i_writecount) > 1
1035 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001036 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 msleep(10);
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001038 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
1040
1041 if (inode->i_state & I_DIRTY) {
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001042 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
Al Viro6264d692006-10-19 23:28:58 -07001043 host_err=nfsd_sync(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 }
1045#if 0
1046 wake_up(&inode->i_wait);
1047#endif
1048 }
1049 last_ino = inode->i_ino;
1050 last_dev = inode->i_sb->s_dev;
1051 }
1052
Al Viro6264d692006-10-19 23:28:58 -07001053 dprintk("nfsd: write complete host_err=%d\n", host_err);
1054 if (host_err >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 err = 0;
1056 else
Al Viro6264d692006-10-19 23:28:58 -07001057 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058out:
1059 return err;
1060}
1061
1062/*
1063 * Read data from a file. count must contain the requested read count
1064 * on entry. On return, *count contains the number of bytes actually read.
1065 * N.B. After this call fhp needs an fh_put
1066 */
Al Viro6264d692006-10-19 23:28:58 -07001067__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1069 loff_t offset, struct kvec *vec, int vlen,
1070 unsigned long *count)
1071{
Al Viro6264d692006-10-19 23:28:58 -07001072 __be32 err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
1074 if (file) {
J. Bruce Fields0ec757d2007-07-17 04:04:48 -07001075 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001076 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 if (err)
1078 goto out;
1079 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1080 } else {
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001081 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 if (err)
1083 goto out;
1084 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1085 nfsd_close(file);
1086 }
1087out:
1088 return err;
1089}
1090
1091/*
1092 * Write data to a file.
1093 * The stable flag requests synchronous writes.
1094 * N.B. After this call fhp needs an fh_put
1095 */
Al Viro6264d692006-10-19 23:28:58 -07001096__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1098 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1099 int *stablep)
1100{
Al Viro6264d692006-10-19 23:28:58 -07001101 __be32 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 if (file) {
J. Bruce Fields0ec757d2007-07-17 04:04:48 -07001104 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001105 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 if (err)
1107 goto out;
1108 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1109 stablep);
1110 } else {
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001111 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 if (err)
1113 goto out;
1114
1115 if (cnt)
1116 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1117 cnt, stablep);
1118 nfsd_close(file);
1119 }
1120out:
1121 return err;
1122}
1123
1124#ifdef CONFIG_NFSD_V3
1125/*
1126 * Commit all pending writes to stable storage.
1127 * Strictly speaking, we could sync just the indicated file region here,
1128 * but there's currently no way we can ask the VFS to do so.
1129 *
1130 * Unfortunately we cannot lock the file to make sure we return full WCC
1131 * data to the client, as locking happens lower down in the filesystem.
1132 */
Al Viro6264d692006-10-19 23:28:58 -07001133__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1135 loff_t offset, unsigned long count)
1136{
1137 struct file *file;
Al Viro6264d692006-10-19 23:28:58 -07001138 __be32 err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140 if ((u64)count > ~(u64)offset)
1141 return nfserr_inval;
1142
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001143 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1144 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 return err;
1146 if (EX_ISSYNC(fhp->fh_export)) {
1147 if (file->f_op && file->f_op->fsync) {
NeilBrown45bd3b32006-01-18 17:43:50 -08001148 err = nfserrno(nfsd_sync(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 } else {
1150 err = nfserr_notsupp;
1151 }
1152 }
1153
1154 nfsd_close(file);
1155 return err;
1156}
1157#endif /* CONFIG_NFSD_V3 */
1158
Adrian Bunkf2b0dee2008-02-13 23:30:26 +02001159static __be32
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001160nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1161 struct iattr *iap)
1162{
1163 /*
1164 * Mode has already been set earlier in create:
1165 */
1166 iap->ia_valid &= ~ATTR_MODE;
1167 /*
1168 * Setting uid/gid works only for root. Irix appears to
1169 * send along the gid on create when it tries to implement
1170 * setgid directories via NFS:
1171 */
David Howells5cc0a842008-11-14 10:38:58 +11001172 if (current_fsuid() != 0)
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001173 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1174 if (iap->ia_valid)
1175 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1176 return 0;
1177}
1178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179/*
1180 * Create a file (regular, directory, device, fifo); UNIX sockets
1181 * not yet implemented.
1182 * If the response fh has been verified, the parent directory should
1183 * already be locked. Note that the parent directory is left locked.
1184 *
1185 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1186 */
Al Viro6264d692006-10-19 23:28:58 -07001187__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1189 char *fname, int flen, struct iattr *iap,
1190 int type, dev_t rdev, struct svc_fh *resfhp)
1191{
1192 struct dentry *dentry, *dchild = NULL;
1193 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001194 __be32 err;
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001195 __be32 err2;
Al Viro6264d692006-10-19 23:28:58 -07001196 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
1198 err = nfserr_perm;
1199 if (!flen)
1200 goto out;
1201 err = nfserr_exist;
1202 if (isdotent(fname, flen))
1203 goto out;
1204
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001205 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 if (err)
1207 goto out;
1208
1209 dentry = fhp->fh_dentry;
1210 dirp = dentry->d_inode;
1211
1212 err = nfserr_notdir;
Al Viroacfa4382008-12-04 10:06:33 -05001213 if (!dirp->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 goto out;
1215 /*
1216 * Check whether the response file handle has been verified yet.
1217 * If it has, the parent directory should already be locked.
1218 */
1219 if (!resfhp->fh_dentry) {
1220 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001221 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 dchild = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001223 host_err = PTR_ERR(dchild);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 if (IS_ERR(dchild))
1225 goto out_nfserr;
1226 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1227 if (err)
1228 goto out;
1229 } else {
1230 /* called from nfsd_proc_create */
1231 dchild = dget(resfhp->fh_dentry);
1232 if (!fhp->fh_locked) {
1233 /* not actually possible */
1234 printk(KERN_ERR
1235 "nfsd_create: parent %s/%s not locked!\n",
1236 dentry->d_parent->d_name.name,
1237 dentry->d_name.name);
Al Virod75f2b92006-01-18 17:43:44 -08001238 err = nfserr_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 goto out;
1240 }
1241 }
1242 /*
1243 * Make sure the child dentry is still negative ...
1244 */
1245 err = nfserr_exist;
1246 if (dchild->d_inode) {
1247 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1248 dentry->d_name.name, dchild->d_name.name);
1249 goto out;
1250 }
1251
1252 if (!(iap->ia_valid & ATTR_MODE))
1253 iap->ia_mode = 0;
1254 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1255
Miklos Szeredi07cad1d2008-07-01 15:38:35 +02001256 err = nfserr_inval;
1257 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1258 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1259 type);
1260 goto out;
1261 }
1262
1263 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1264 if (host_err)
1265 goto out_nfserr;
1266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 /*
1268 * Get the dir op function pointer.
1269 */
J. Bruce Fields088406b2006-11-08 17:44:59 -08001270 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 switch (type) {
1272 case S_IFREG:
Al Viro6264d692006-10-19 23:28:58 -07001273 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 break;
1275 case S_IFDIR:
Al Viro6264d692006-10-19 23:28:58 -07001276 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 break;
1278 case S_IFCHR:
1279 case S_IFBLK:
1280 case S_IFIFO:
1281 case S_IFSOCK:
Al Viro6264d692006-10-19 23:28:58 -07001282 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 break;
Dave Hansen463c3192008-02-15 14:37:57 -08001284 }
1285 if (host_err < 0) {
1286 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1287 goto out_nfserr;
1288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
1290 if (EX_ISSYNC(fhp->fh_export)) {
NeilBrown45bd3b32006-01-18 17:43:50 -08001291 err = nfserrno(nfsd_sync_dir(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 write_inode_now(dchild->d_inode, 1);
1293 }
1294
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001295 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1296 if (err2)
1297 err = err2;
Dave Hansen463c3192008-02-15 14:37:57 -08001298 mnt_drop_write(fhp->fh_export->ex_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 /*
1300 * Update the file handle to get the new inode info.
1301 */
1302 if (!err)
1303 err = fh_update(resfhp);
1304out:
1305 if (dchild && !IS_ERR(dchild))
1306 dput(dchild);
1307 return err;
1308
1309out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001310 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 goto out;
1312}
1313
1314#ifdef CONFIG_NFSD_V3
1315/*
1316 * NFSv3 version of nfsd_create
1317 */
Al Viro6264d692006-10-19 23:28:58 -07001318__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1320 char *fname, int flen, struct iattr *iap,
1321 struct svc_fh *resfhp, int createmode, u32 *verifier,
J. Bruce Fields81ac95c2006-11-08 17:44:40 -08001322 int *truncp, int *created)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323{
1324 struct dentry *dentry, *dchild = NULL;
1325 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001326 __be32 err;
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001327 __be32 err2;
Al Viro6264d692006-10-19 23:28:58 -07001328 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 __u32 v_mtime=0, v_atime=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
1331 err = nfserr_perm;
1332 if (!flen)
1333 goto out;
1334 err = nfserr_exist;
1335 if (isdotent(fname, flen))
1336 goto out;
1337 if (!(iap->ia_valid & ATTR_MODE))
1338 iap->ia_mode = 0;
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001339 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (err)
1341 goto out;
1342
1343 dentry = fhp->fh_dentry;
1344 dirp = dentry->d_inode;
1345
1346 /* Get all the sanity checks out of the way before
1347 * we lock the parent. */
1348 err = nfserr_notdir;
Al Viroacfa4382008-12-04 10:06:33 -05001349 if (!dirp->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 goto out;
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001351 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353 /*
1354 * Compose the response file handle.
1355 */
1356 dchild = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001357 host_err = PTR_ERR(dchild);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 if (IS_ERR(dchild))
1359 goto out_nfserr;
1360
1361 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1362 if (err)
1363 goto out;
1364
1365 if (createmode == NFS3_CREATE_EXCLUSIVE) {
Peter Staubachc3978522007-01-26 00:57:00 -08001366 /* solaris7 gets confused (bugid 4218508) if these have
Jeff Layton749997e2007-07-31 00:37:51 -07001367 * the high bit set, so just clear the high bits. If this is
1368 * ever changed to use different attrs for storing the
1369 * verifier, then do_open_lookup() will also need to be fixed
1370 * accordingly.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 */
1372 v_mtime = verifier[0]&0x7fffffff;
1373 v_atime = verifier[1]&0x7fffffff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 }
1375
Dave Hansen463c3192008-02-15 14:37:57 -08001376 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1377 if (host_err)
1378 goto out_nfserr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 if (dchild->d_inode) {
1380 err = 0;
1381
1382 switch (createmode) {
1383 case NFS3_CREATE_UNCHECKED:
1384 if (! S_ISREG(dchild->d_inode->i_mode))
1385 err = nfserr_exist;
1386 else if (truncp) {
1387 /* in nfsv4, we need to treat this case a little
1388 * differently. we don't want to truncate the
1389 * file now; this would be wrong if the OPEN
1390 * fails for some other reason. furthermore,
1391 * if the size is nonzero, we should ignore it
1392 * according to spec!
1393 */
1394 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1395 }
1396 else {
1397 iap->ia_valid &= ATTR_SIZE;
1398 goto set_attr;
1399 }
1400 break;
1401 case NFS3_CREATE_EXCLUSIVE:
1402 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1403 && dchild->d_inode->i_atime.tv_sec == v_atime
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 && dchild->d_inode->i_size == 0 )
1405 break;
1406 /* fallthru */
1407 case NFS3_CREATE_GUARDED:
1408 err = nfserr_exist;
1409 }
Dave Hansen463c3192008-02-15 14:37:57 -08001410 mnt_drop_write(fhp->fh_export->ex_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 goto out;
1412 }
1413
Al Viro6264d692006-10-19 23:28:58 -07001414 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
Dave Hansen463c3192008-02-15 14:37:57 -08001415 if (host_err < 0) {
1416 mnt_drop_write(fhp->fh_export->ex_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 goto out_nfserr;
Dave Hansen463c3192008-02-15 14:37:57 -08001418 }
J. Bruce Fields81ac95c2006-11-08 17:44:40 -08001419 if (created)
1420 *created = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
1422 if (EX_ISSYNC(fhp->fh_export)) {
NeilBrown45bd3b32006-01-18 17:43:50 -08001423 err = nfserrno(nfsd_sync_dir(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 /* setattr will sync the child (or not) */
1425 }
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (createmode == NFS3_CREATE_EXCLUSIVE) {
Peter Staubachc3978522007-01-26 00:57:00 -08001428 /* Cram the verifier into atime/mtime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
Peter Staubachc3978522007-01-26 00:57:00 -08001430 | ATTR_MTIME_SET|ATTR_ATIME_SET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 /* XXX someone who knows this better please fix it for nsec */
1432 iap->ia_mtime.tv_sec = v_mtime;
1433 iap->ia_atime.tv_sec = v_atime;
1434 iap->ia_mtime.tv_nsec = 0;
1435 iap->ia_atime.tv_nsec = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 }
1437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 set_attr:
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001439 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1440 if (err2)
1441 err = err2;
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -08001442
Dave Hansen463c3192008-02-15 14:37:57 -08001443 mnt_drop_write(fhp->fh_export->ex_path.mnt);
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -08001444 /*
1445 * Update the filehandle to get the new inode info.
1446 */
1447 if (!err)
1448 err = fh_update(resfhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450 out:
1451 fh_unlock(fhp);
1452 if (dchild && !IS_ERR(dchild))
1453 dput(dchild);
1454 return err;
1455
1456 out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001457 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 goto out;
1459}
1460#endif /* CONFIG_NFSD_V3 */
1461
1462/*
1463 * Read a symlink. On entry, *lenp must contain the maximum path length that
1464 * fits into the buffer. On return, it contains the true length.
1465 * N.B. After this call fhp needs an fh_put
1466 */
Al Viro6264d692006-10-19 23:28:58 -07001467__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1469{
1470 struct dentry *dentry;
1471 struct inode *inode;
1472 mm_segment_t oldfs;
Al Viro6264d692006-10-19 23:28:58 -07001473 __be32 err;
1474 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001476 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (err)
1478 goto out;
1479
1480 dentry = fhp->fh_dentry;
1481 inode = dentry->d_inode;
1482
1483 err = nfserr_inval;
Al Viroacfa4382008-12-04 10:06:33 -05001484 if (!inode->i_op->readlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 goto out;
1486
Jan Blunck54775492008-02-14 19:38:39 -08001487 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 /* N.B. Why does this call need a get_fs()??
1489 * Remove the set_fs and watch the fireworks:-) --okir
1490 */
1491
1492 oldfs = get_fs(); set_fs(KERNEL_DS);
Al Viro6264d692006-10-19 23:28:58 -07001493 host_err = inode->i_op->readlink(dentry, buf, *lenp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 set_fs(oldfs);
1495
Al Viro6264d692006-10-19 23:28:58 -07001496 if (host_err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 goto out_nfserr;
Al Viro6264d692006-10-19 23:28:58 -07001498 *lenp = host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 err = 0;
1500out:
1501 return err;
1502
1503out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001504 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 goto out;
1506}
1507
1508/*
1509 * Create a symlink and look up its inode
1510 * N.B. After this call _both_ fhp and resfhp need an fh_put
1511 */
Al Viro6264d692006-10-19 23:28:58 -07001512__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1514 char *fname, int flen,
1515 char *path, int plen,
1516 struct svc_fh *resfhp,
1517 struct iattr *iap)
1518{
1519 struct dentry *dentry, *dnew;
Al Viro6264d692006-10-19 23:28:58 -07001520 __be32 err, cerr;
1521 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
1523 err = nfserr_noent;
1524 if (!flen || !plen)
1525 goto out;
1526 err = nfserr_exist;
1527 if (isdotent(fname, flen))
1528 goto out;
1529
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001530 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 if (err)
1532 goto out;
1533 fh_lock(fhp);
1534 dentry = fhp->fh_dentry;
1535 dnew = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001536 host_err = PTR_ERR(dnew);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 if (IS_ERR(dnew))
1538 goto out_nfserr;
1539
Dave Hansen75c3f292008-02-15 14:37:45 -08001540 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1541 if (host_err)
1542 goto out_nfserr;
1543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 if (unlikely(path[plen] != 0)) {
1545 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1546 if (path_alloced == NULL)
Al Viro6264d692006-10-19 23:28:58 -07001547 host_err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 else {
1549 strncpy(path_alloced, path, plen);
1550 path_alloced[plen] = 0;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02001551 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 kfree(path_alloced);
1553 }
1554 } else
Miklos Szeredidb2e7472008-06-24 16:50:16 +02001555 host_err = vfs_symlink(dentry->d_inode, dnew, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Al Viro6264d692006-10-19 23:28:58 -07001557 if (!host_err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (EX_ISSYNC(fhp->fh_export))
Al Viro6264d692006-10-19 23:28:58 -07001559 host_err = nfsd_sync_dir(dentry);
1560 }
1561 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 fh_unlock(fhp);
1563
Dave Hansen75c3f292008-02-15 14:37:45 -08001564 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1567 dput(dnew);
1568 if (err==0) err = cerr;
1569out:
1570 return err;
1571
1572out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001573 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 goto out;
1575}
1576
1577/*
1578 * Create a hardlink
1579 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1580 */
Al Viro6264d692006-10-19 23:28:58 -07001581__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1583 char *name, int len, struct svc_fh *tfhp)
1584{
1585 struct dentry *ddir, *dnew, *dold;
1586 struct inode *dirp, *dest;
Al Viro6264d692006-10-19 23:28:58 -07001587 __be32 err;
1588 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001590 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 if (err)
1592 goto out;
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001593 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (err)
1595 goto out;
1596
1597 err = nfserr_perm;
1598 if (!len)
1599 goto out;
1600 err = nfserr_exist;
1601 if (isdotent(name, len))
1602 goto out;
1603
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001604 fh_lock_nested(ffhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 ddir = ffhp->fh_dentry;
1606 dirp = ddir->d_inode;
1607
1608 dnew = lookup_one_len(name, ddir, len);
Al Viro6264d692006-10-19 23:28:58 -07001609 host_err = PTR_ERR(dnew);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 if (IS_ERR(dnew))
1611 goto out_nfserr;
1612
1613 dold = tfhp->fh_dentry;
1614 dest = dold->d_inode;
1615
Dave Hansen75c3f292008-02-15 14:37:45 -08001616 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1617 if (host_err) {
1618 err = nfserrno(host_err);
1619 goto out_dput;
1620 }
Al Viro6264d692006-10-19 23:28:58 -07001621 host_err = vfs_link(dold, dirp, dnew);
1622 if (!host_err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 if (EX_ISSYNC(ffhp->fh_export)) {
NeilBrown45bd3b32006-01-18 17:43:50 -08001624 err = nfserrno(nfsd_sync_dir(ddir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 write_inode_now(dest, 1);
1626 }
Al Viro6264d692006-10-19 23:28:58 -07001627 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 } else {
Al Viro6264d692006-10-19 23:28:58 -07001629 if (host_err == -EXDEV && rqstp->rq_vers == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 err = nfserr_acces;
1631 else
Al Viro6264d692006-10-19 23:28:58 -07001632 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 }
Dave Hansen75c3f292008-02-15 14:37:45 -08001634 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1635out_dput:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 dput(dnew);
David M. Richter270d56e2006-06-30 01:56:15 -07001637out_unlock:
1638 fh_unlock(ffhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639out:
1640 return err;
1641
1642out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001643 err = nfserrno(host_err);
David M. Richter270d56e2006-06-30 01:56:15 -07001644 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645}
1646
1647/*
1648 * Rename a file
1649 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1650 */
Al Viro6264d692006-10-19 23:28:58 -07001651__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1653 struct svc_fh *tfhp, char *tname, int tlen)
1654{
1655 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1656 struct inode *fdir, *tdir;
Al Viro6264d692006-10-19 23:28:58 -07001657 __be32 err;
1658 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001660 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (err)
1662 goto out;
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001663 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 if (err)
1665 goto out;
1666
1667 fdentry = ffhp->fh_dentry;
1668 fdir = fdentry->d_inode;
1669
1670 tdentry = tfhp->fh_dentry;
1671 tdir = tdentry->d_inode;
1672
1673 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
NeilBrowna56f3932006-06-30 01:56:10 -07001674 if (ffhp->fh_export != tfhp->fh_export)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 goto out;
1676
1677 err = nfserr_perm;
1678 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1679 goto out;
1680
1681 /* cannot use fh_lock as we need deadlock protective ordering
1682 * so do it by hand */
1683 trap = lock_rename(tdentry, fdentry);
1684 ffhp->fh_locked = tfhp->fh_locked = 1;
1685 fill_pre_wcc(ffhp);
1686 fill_pre_wcc(tfhp);
1687
1688 odentry = lookup_one_len(fname, fdentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001689 host_err = PTR_ERR(odentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 if (IS_ERR(odentry))
1691 goto out_nfserr;
1692
Al Viro6264d692006-10-19 23:28:58 -07001693 host_err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 if (!odentry->d_inode)
1695 goto out_dput_old;
Al Viro6264d692006-10-19 23:28:58 -07001696 host_err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 if (odentry == trap)
1698 goto out_dput_old;
1699
1700 ndentry = lookup_one_len(tname, tdentry, tlen);
Al Viro6264d692006-10-19 23:28:58 -07001701 host_err = PTR_ERR(ndentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 if (IS_ERR(ndentry))
1703 goto out_dput_old;
Al Viro6264d692006-10-19 23:28:58 -07001704 host_err = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 if (ndentry == trap)
1706 goto out_dput_new;
1707
Dave Hansen9079b1e2008-02-15 14:37:49 -08001708 if (svc_msnfs(ffhp) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 ((atomic_read(&odentry->d_count) > 1)
1710 || (atomic_read(&ndentry->d_count) > 1))) {
Al Viro6264d692006-10-19 23:28:58 -07001711 host_err = -EPERM;
Dave Hansen9079b1e2008-02-15 14:37:49 -08001712 goto out_dput_new;
1713 }
1714
1715 host_err = -EXDEV;
1716 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1717 goto out_dput_new;
1718 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1719 if (host_err)
1720 goto out_dput_new;
1721
Al Viro6264d692006-10-19 23:28:58 -07001722 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1723 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1724 host_err = nfsd_sync_dir(tdentry);
1725 if (!host_err)
1726 host_err = nfsd_sync_dir(fdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 }
1728
Dave Hansen9079b1e2008-02-15 14:37:49 -08001729 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1730
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 out_dput_new:
1732 dput(ndentry);
1733 out_dput_old:
1734 dput(odentry);
1735 out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001736 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
1738 /* we cannot reply on fh_unlock on the two filehandles,
1739 * as that would do the wrong thing if the two directories
1740 * were the same, so again we do it by hand
1741 */
1742 fill_post_wcc(ffhp);
1743 fill_post_wcc(tfhp);
1744 unlock_rename(tdentry, fdentry);
1745 ffhp->fh_locked = tfhp->fh_locked = 0;
1746
1747out:
1748 return err;
1749}
1750
1751/*
1752 * Unlink a file or directory
1753 * N.B. After this call fhp needs an fh_put
1754 */
Al Viro6264d692006-10-19 23:28:58 -07001755__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1757 char *fname, int flen)
1758{
1759 struct dentry *dentry, *rdentry;
1760 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001761 __be32 err;
1762 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
1764 err = nfserr_acces;
1765 if (!flen || isdotent(fname, flen))
1766 goto out;
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001767 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 if (err)
1769 goto out;
1770
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001771 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 dentry = fhp->fh_dentry;
1773 dirp = dentry->d_inode;
1774
1775 rdentry = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001776 host_err = PTR_ERR(rdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 if (IS_ERR(rdentry))
1778 goto out_nfserr;
1779
1780 if (!rdentry->d_inode) {
1781 dput(rdentry);
1782 err = nfserr_noent;
1783 goto out;
1784 }
1785
1786 if (!type)
1787 type = rdentry->d_inode->i_mode & S_IFMT;
1788
Dave Hansen06227532008-02-15 14:37:34 -08001789 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1790 if (host_err)
1791 goto out_nfserr;
1792
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 if (type != S_IFDIR) { /* It's UNLINK */
1794#ifdef MSNFS
1795 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1796 (atomic_read(&rdentry->d_count) > 1)) {
Al Viro6264d692006-10-19 23:28:58 -07001797 host_err = -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 } else
1799#endif
Al Viro6264d692006-10-19 23:28:58 -07001800 host_err = vfs_unlink(dirp, rdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 } else { /* It's RMDIR */
Al Viro6264d692006-10-19 23:28:58 -07001802 host_err = vfs_rmdir(dirp, rdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
1804
1805 dput(rdentry);
1806
Al Viro6264d692006-10-19 23:28:58 -07001807 if (host_err)
Dave Hansen06227532008-02-15 14:37:34 -08001808 goto out_drop;
Al Viro6264d692006-10-19 23:28:58 -07001809 if (EX_ISSYNC(fhp->fh_export))
1810 host_err = nfsd_sync_dir(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Dave Hansen06227532008-02-15 14:37:34 -08001812out_drop:
1813 mnt_drop_write(fhp->fh_export->ex_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001815 err = nfserrno(host_err);
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -08001816out:
1817 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818}
1819
1820/*
David Woodhouse14f7dd62008-07-31 20:29:12 +01001821 * We do this buffering because we must not call back into the file
1822 * system's ->lookup() method from the filldir callback. That may well
1823 * deadlock a number of file systems.
1824 *
1825 * This is based heavily on the implementation of same in XFS.
1826 */
1827struct buffered_dirent {
1828 u64 ino;
1829 loff_t offset;
1830 int namlen;
1831 unsigned int d_type;
1832 char name[];
1833};
1834
1835struct readdir_data {
1836 char *dirent;
1837 size_t used;
Al Viro53c9c5c2008-08-24 07:29:52 -04001838 int full;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001839};
1840
1841static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1842 loff_t offset, u64 ino, unsigned int d_type)
David Woodhouse2628b762008-07-31 17:16:51 +01001843{
David Woodhouse14f7dd62008-07-31 20:29:12 +01001844 struct readdir_data *buf = __buf;
1845 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1846 unsigned int reclen;
1847
1848 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
Al Viro53c9c5c2008-08-24 07:29:52 -04001849 if (buf->used + reclen > PAGE_SIZE) {
1850 buf->full = 1;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001851 return -EINVAL;
Al Viro53c9c5c2008-08-24 07:29:52 -04001852 }
David Woodhouse14f7dd62008-07-31 20:29:12 +01001853
1854 de->namlen = namlen;
1855 de->offset = offset;
1856 de->ino = ino;
1857 de->d_type = d_type;
1858 memcpy(de->name, name, namlen);
1859 buf->used += reclen;
1860
1861 return 0;
1862}
1863
1864static int nfsd_buffered_readdir(struct file *file, filldir_t func,
1865 struct readdir_cd *cdp, loff_t *offsetp)
1866{
1867 struct readdir_data buf;
1868 struct buffered_dirent *de;
David Woodhouse2628b762008-07-31 17:16:51 +01001869 int host_err;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001870 int size;
1871 loff_t offset;
David Woodhouse2628b762008-07-31 17:16:51 +01001872
David Woodhouse14f7dd62008-07-31 20:29:12 +01001873 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1874 if (!buf.dirent)
1875 return -ENOMEM;
David Woodhouse2628b762008-07-31 17:16:51 +01001876
David Woodhouse14f7dd62008-07-31 20:29:12 +01001877 offset = *offsetp;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001878
1879 while (1) {
1880 unsigned int reclen;
1881
Doug Nazarb726e922008-11-05 06:16:28 -05001882 cdp->err = nfserr_eof; /* will be cleared on successful read */
David Woodhouse14f7dd62008-07-31 20:29:12 +01001883 buf.used = 0;
Al Viro53c9c5c2008-08-24 07:29:52 -04001884 buf.full = 0;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001885
1886 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
Al Viro53c9c5c2008-08-24 07:29:52 -04001887 if (buf.full)
1888 host_err = 0;
1889
1890 if (host_err < 0)
David Woodhouse14f7dd62008-07-31 20:29:12 +01001891 break;
1892
1893 size = buf.used;
1894
1895 if (!size)
1896 break;
1897
David Woodhouse14f7dd62008-07-31 20:29:12 +01001898 de = (struct buffered_dirent *)buf.dirent;
1899 while (size > 0) {
1900 offset = de->offset;
1901
1902 if (func(cdp, de->name, de->namlen, de->offset,
1903 de->ino, de->d_type))
1904 goto done;
1905
1906 if (cdp->err != nfs_ok)
1907 goto done;
1908
1909 reclen = ALIGN(sizeof(*de) + de->namlen,
1910 sizeof(u64));
1911 size -= reclen;
1912 de = (struct buffered_dirent *)((char *)de + reclen);
1913 }
David Woodhousec002a6c2008-08-17 17:21:18 +01001914 offset = vfs_llseek(file, 0, SEEK_CUR);
David Woodhouse14f7dd62008-07-31 20:29:12 +01001915 }
1916
1917 done:
1918 free_page((unsigned long)(buf.dirent));
David Woodhouse2628b762008-07-31 17:16:51 +01001919
1920 if (host_err)
1921 return nfserrno(host_err);
David Woodhouse14f7dd62008-07-31 20:29:12 +01001922
1923 *offsetp = offset;
1924 return cdp->err;
David Woodhouse2628b762008-07-31 17:16:51 +01001925}
1926
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927/*
1928 * Read entries from a directory.
1929 * The NFSv3/4 verifier we ignore for now.
1930 */
Al Viro6264d692006-10-19 23:28:58 -07001931__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
NeilBrowna0ad13e2007-01-26 00:57:10 -08001933 struct readdir_cd *cdp, filldir_t func)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934{
Al Viro6264d692006-10-19 23:28:58 -07001935 __be32 err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 struct file *file;
1937 loff_t offset = *offsetp;
1938
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001939 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 if (err)
1941 goto out;
1942
1943 offset = vfs_llseek(file, offset, 0);
1944 if (offset < 0) {
1945 err = nfserrno((int)offset);
1946 goto out_close;
1947 }
1948
David Woodhouse14f7dd62008-07-31 20:29:12 +01001949 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
1951 if (err == nfserr_eof || err == nfserr_toosmall)
1952 err = nfs_ok; /* can still be found in ->err */
1953out_close:
1954 nfsd_close(file);
1955out:
1956 return err;
1957}
1958
1959/*
1960 * Get file system stats
1961 * N.B. After this call fhp needs an fh_put
1962 */
Al Viro6264d692006-10-19 23:28:58 -07001963__be32
J. Bruce Fields04716e62008-08-07 13:00:20 -04001964nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965{
J. Bruce Fields04716e62008-08-07 13:00:20 -04001966 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
David Howells726c3342006-06-23 02:02:58 -07001967 if (!err && vfs_statfs(fhp->fh_dentry,stat))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 err = nfserr_io;
1969 return err;
1970}
1971
J. Bruce Fieldsc7d51402007-07-19 01:49:20 -07001972static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
J. Bruce Fieldse22841c2007-07-19 01:49:20 -07001973{
J. Bruce Fieldsc7d51402007-07-19 01:49:20 -07001974 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
J. Bruce Fieldse22841c2007-07-19 01:49:20 -07001975}
1976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977/*
1978 * Check for a user's access permissions to this inode.
1979 */
Al Viro6264d692006-10-19 23:28:58 -07001980__be32
J. Bruce Fields0ec757d2007-07-17 04:04:48 -07001981nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1982 struct dentry *dentry, int acc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983{
1984 struct inode *inode = dentry->d_inode;
1985 int err;
1986
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001987 if (acc == NFSD_MAY_NOP)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 return 0;
1989#if 0
1990 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1991 acc,
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001992 (acc & NFSD_MAY_READ)? " read" : "",
1993 (acc & NFSD_MAY_WRITE)? " write" : "",
1994 (acc & NFSD_MAY_EXEC)? " exec" : "",
1995 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1996 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1997 (acc & NFSD_MAY_LOCK)? " lock" : "",
1998 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 inode->i_mode,
2000 IS_IMMUTABLE(inode)? " immut" : "",
2001 IS_APPEND(inode)? " append" : "",
Dave Hansen2c463e92008-02-15 14:37:56 -08002002 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 dprintk(" owner %d/%d user %d/%d\n",
David Howells5cc0a842008-11-14 10:38:58 +11002004 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005#endif
2006
2007 /* Normally we reject any write/sattr etc access on a read-only file
2008 * system. But if it is IRIX doing check on write-access for a
2009 * device special file, we ignore rofs.
2010 */
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002011 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2012 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
Dave Hansen2c463e92008-02-15 14:37:56 -08002013 if (exp_rdonly(rqstp, exp) ||
2014 __mnt_is_readonly(exp->ex_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 return nfserr_rofs;
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002016 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 return nfserr_perm;
2018 }
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002019 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 return nfserr_perm;
2021
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002022 if (acc & NFSD_MAY_LOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 /* If we cannot rely on authentication in NLM requests,
2024 * just allow locks, otherwise require read permission, or
2025 * ownership
2026 */
2027 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2028 return 0;
2029 else
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002030 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 }
2032 /*
2033 * The file owner always gets access permission for accesses that
2034 * would normally be checked at open time. This is to make
2035 * file access work even when the client has done a fchmod(fd, 0).
2036 *
2037 * However, `cp foo bar' should fail nevertheless when bar is
2038 * readonly. A sensible way to do this might be to reject all
2039 * attempts to truncate a read-only file, because a creat() call
2040 * always implies file truncation.
2041 * ... but this isn't really fair. A process may reasonably call
2042 * ftruncate on an open file descriptor on a file with perm 000.
2043 * We must trust the client to do permission checking - using "ACCESS"
2044 * with NFSv3.
2045 */
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002046 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
David Howells5cc0a842008-11-14 10:38:58 +11002047 inode->i_uid == current_fsuid())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return 0;
2049
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002050 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
Al Virof419a2e2008-07-22 00:07:17 -04002051 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
2053 /* Allow read access to binaries even when mode 111 */
2054 if (err == -EACCES && S_ISREG(inode->i_mode) &&
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002055 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
Al Virof419a2e2008-07-22 00:07:17 -04002056 err = inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
2058 return err? nfserrno(err) : 0;
2059}
2060
2061void
2062nfsd_racache_shutdown(void)
2063{
Jeff Layton54a66e52008-08-13 22:03:27 -04002064 struct raparms *raparm, *last_raparm;
2065 unsigned int i;
2066
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 dprintk("nfsd: freeing readahead buffers.\n");
Jeff Layton54a66e52008-08-13 22:03:27 -04002068
2069 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2070 raparm = raparm_hash[i].pb_head;
2071 while(raparm) {
2072 last_raparm = raparm;
2073 raparm = raparm->p_next;
2074 kfree(last_raparm);
2075 }
2076 raparm_hash[i].pb_head = NULL;
2077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078}
2079/*
2080 * Initialize readahead param cache
2081 */
2082int
2083nfsd_racache_init(int cache_size)
2084{
2085 int i;
Greg Banksfce14562006-10-04 02:15:49 -07002086 int j = 0;
2087 int nperbucket;
Jeff Layton54a66e52008-08-13 22:03:27 -04002088 struct raparms **raparm = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
Greg Banksfce14562006-10-04 02:15:49 -07002090
Jeff Layton54a66e52008-08-13 22:03:27 -04002091 if (raparm_hash[0].pb_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 return 0;
Jeff Layton54a66e52008-08-13 22:03:27 -04002093 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2094 if (nperbucket < 2)
2095 nperbucket = 2;
2096 cache_size = nperbucket * RAPARM_HASH_SIZE;
Yan Burman4b3bb062006-12-08 02:39:41 -08002097
2098 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
Jeff Layton54a66e52008-08-13 22:03:27 -04002099
2100 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
Yan Burman4b3bb062006-12-08 02:39:41 -08002101 spin_lock_init(&raparm_hash[i].pb_lock);
Jeff Layton54a66e52008-08-13 22:03:27 -04002102
2103 raparm = &raparm_hash[i].pb_head;
2104 for (j = 0; j < nperbucket; j++) {
2105 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2106 if (!*raparm)
2107 goto out_nomem;
2108 raparm = &(*raparm)->p_next;
2109 }
2110 *raparm = NULL;
Yan Burman4b3bb062006-12-08 02:39:41 -08002111 }
2112
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 nfsdstats.ra_size = cache_size;
2114 return 0;
Jeff Layton54a66e52008-08-13 22:03:27 -04002115
2116out_nomem:
2117 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2118 nfsd_racache_shutdown();
2119 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120}
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002121
2122#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2123struct posix_acl *
2124nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2125{
2126 struct inode *inode = fhp->fh_dentry->d_inode;
2127 char *name;
2128 void *value = NULL;
2129 ssize_t size;
2130 struct posix_acl *acl;
2131
Christoph Hellwig5be196e2006-01-09 20:51:55 -08002132 if (!IS_POSIXACL(inode))
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002133 return ERR_PTR(-EOPNOTSUPP);
Christoph Hellwig5be196e2006-01-09 20:51:55 -08002134
2135 switch (type) {
2136 case ACL_TYPE_ACCESS:
2137 name = POSIX_ACL_XATTR_ACCESS;
2138 break;
2139 case ACL_TYPE_DEFAULT:
2140 name = POSIX_ACL_XATTR_DEFAULT;
2141 break;
2142 default:
2143 return ERR_PTR(-EOPNOTSUPP);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002144 }
2145
Christoph Hellwig5be196e2006-01-09 20:51:55 -08002146 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2147 if (size < 0)
2148 return ERR_PTR(size);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002149
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002150 acl = posix_acl_from_xattr(value, size);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002151 kfree(value);
2152 return acl;
2153}
2154
2155int
2156nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2157{
2158 struct inode *inode = fhp->fh_dentry->d_inode;
2159 char *name;
2160 void *value = NULL;
2161 size_t size;
2162 int error;
2163
Al Viroacfa4382008-12-04 10:06:33 -05002164 if (!IS_POSIXACL(inode) ||
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002165 !inode->i_op->setxattr || !inode->i_op->removexattr)
2166 return -EOPNOTSUPP;
2167 switch(type) {
2168 case ACL_TYPE_ACCESS:
Christoph Hellwig334a13e2005-06-28 20:44:58 -07002169 name = POSIX_ACL_XATTR_ACCESS;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002170 break;
2171 case ACL_TYPE_DEFAULT:
Christoph Hellwig334a13e2005-06-28 20:44:58 -07002172 name = POSIX_ACL_XATTR_DEFAULT;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002173 break;
2174 default:
2175 return -EOPNOTSUPP;
2176 }
2177
2178 if (acl && acl->a_count) {
Christoph Hellwig334a13e2005-06-28 20:44:58 -07002179 size = posix_acl_xattr_size(acl->a_count);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002180 value = kmalloc(size, GFP_KERNEL);
2181 if (!value)
2182 return -ENOMEM;
Florin Malita9ccfc292006-05-20 14:59:58 -07002183 error = posix_acl_to_xattr(acl, value, size);
2184 if (error < 0)
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002185 goto getout;
Florin Malita9ccfc292006-05-20 14:59:58 -07002186 size = error;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002187 } else
2188 size = 0;
2189
Dave Hansen18f335a2008-02-15 14:37:38 -08002190 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2191 if (error)
2192 goto getout;
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002193 if (size)
Christoph Hellwig5be196e2006-01-09 20:51:55 -08002194 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002195 else {
2196 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2197 error = 0;
2198 else {
Christoph Hellwig5be196e2006-01-09 20:51:55 -08002199 error = vfs_removexattr(fhp->fh_dentry, name);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002200 if (error == -ENODATA)
2201 error = 0;
2202 }
2203 }
Dave Hansen18f335a2008-02-15 14:37:38 -08002204 mnt_drop_write(fhp->fh_export->ex_path.mnt);
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +00002205
2206getout:
2207 kfree(value);
2208 return error;
2209}
2210#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */