Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #define MSNFS /* HACK HACK */ |
2 | /* | |
1da177e4 LT |
3 | * File operations used by nfsd. Some of these have been ripped from |
4 | * other parts of the kernel because they weren't exported, others | |
5 | * are partial duplicates with added or changed functionality. | |
6 | * | |
7 | * Note that several functions dget() the dentry upon which they want | |
8 | * to act, most notably those that create directory entries. Response | |
9 | * dentry's are dput()'d if necessary in the release callback. | |
10 | * So if you notice code paths that apparently fail to dput() the | |
11 | * dentry, don't worry--they have been taken care of. | |
12 | * | |
13 | * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de> | |
14 | * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp> | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/fs.h> |
18 | #include <linux/file.h> | |
d6b29d7c | 19 | #include <linux/splice.h> |
1da177e4 | 20 | #include <linux/fcntl.h> |
1da177e4 | 21 | #include <linux/namei.h> |
1da177e4 | 22 | #include <linux/delay.h> |
0eeca283 | 23 | #include <linux/fsnotify.h> |
1da177e4 | 24 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 25 | #include <linux/xattr.h> |
9a74af21 BH |
26 | #include <linux/jhash.h> |
27 | #include <linux/ima.h> | |
28 | #include <asm/uaccess.h> | |
29 | ||
30 | #ifdef CONFIG_NFSD_V3 | |
31 | #include "xdr3.h" | |
32 | #endif /* CONFIG_NFSD_V3 */ | |
33 | ||
5be196e5 | 34 | #ifdef CONFIG_NFSD_V4 |
1da177e4 LT |
35 | #include <linux/nfs4_acl.h> |
36 | #include <linux/nfsd_idmap.h> | |
1da177e4 LT |
37 | #endif /* CONFIG_NFSD_V4 */ |
38 | ||
9a74af21 BH |
39 | #include "nfsd.h" |
40 | #include "vfs.h" | |
1da177e4 LT |
41 | |
42 | #define NFSDDBG_FACILITY NFSDDBG_FILEOP | |
1da177e4 LT |
43 | |
44 | ||
1da177e4 LT |
45 | /* |
46 | * This is a cache of readahead params that help us choose the proper | |
47 | * readahead strategy. Initially, we set all readahead parameters to 0 | |
48 | * and let the VFS handle things. | |
49 | * If you increase the number of cached files very much, you'll need to | |
50 | * add a hash table here. | |
51 | */ | |
52 | struct raparms { | |
53 | struct raparms *p_next; | |
54 | unsigned int p_count; | |
55 | ino_t p_ino; | |
56 | dev_t p_dev; | |
57 | int p_set; | |
58 | struct file_ra_state p_ra; | |
fce1456a | 59 | unsigned int p_hindex; |
1da177e4 LT |
60 | }; |
61 | ||
fce1456a GB |
62 | struct raparm_hbucket { |
63 | struct raparms *pb_head; | |
64 | spinlock_t pb_lock; | |
65 | } ____cacheline_aligned_in_smp; | |
66 | ||
fce1456a GB |
67 | #define RAPARM_HASH_BITS 4 |
68 | #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) | |
69 | #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) | |
70 | static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; | |
1da177e4 LT |
71 | |
72 | /* | |
73 | * Called from nfsd_lookup and encode_dirent. Check if we have crossed | |
74 | * a mount point. | |
e0bb89ef | 75 | * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, |
1da177e4 LT |
76 | * or nfs_ok having possibly changed *dpp and *expp |
77 | */ | |
78 | int | |
79 | nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, | |
80 | struct svc_export **expp) | |
81 | { | |
82 | struct svc_export *exp = *expp, *exp2 = NULL; | |
83 | struct dentry *dentry = *dpp; | |
91c9fa8f AV |
84 | struct path path = {.mnt = mntget(exp->ex_path.mnt), |
85 | .dentry = dget(dentry)}; | |
6264d69d | 86 | int err = 0; |
1da177e4 | 87 | |
9393bd07 | 88 | while (d_mountpoint(path.dentry) && follow_down(&path)) |
91c9fa8f | 89 | ; |
1da177e4 | 90 | |
91c9fa8f | 91 | exp2 = rqst_exp_get_by_name(rqstp, &path); |
1da177e4 | 92 | if (IS_ERR(exp2)) { |
3b6cee7b BF |
93 | err = PTR_ERR(exp2); |
94 | /* | |
95 | * We normally allow NFS clients to continue | |
96 | * "underneath" a mountpoint that is not exported. | |
97 | * The exception is V4ROOT, where no traversal is ever | |
98 | * allowed without an explicit export of the new | |
99 | * directory. | |
100 | */ | |
101 | if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT)) | |
102 | err = 0; | |
91c9fa8f | 103 | path_put(&path); |
1da177e4 LT |
104 | goto out; |
105 | } | |
3c394dda SD |
106 | if (nfsd_v4client(rqstp) || |
107 | (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { | |
1da177e4 | 108 | /* successfully crossed mount point */ |
1644ccc8 | 109 | /* |
91c9fa8f AV |
110 | * This is subtle: path.dentry is *not* on path.mnt |
111 | * at this point. The only reason we are safe is that | |
112 | * original mnt is pinned down by exp, so we should | |
113 | * put path *before* putting exp | |
1644ccc8 | 114 | */ |
91c9fa8f AV |
115 | *dpp = path.dentry; |
116 | path.dentry = dentry; | |
1644ccc8 | 117 | *expp = exp2; |
91c9fa8f | 118 | exp2 = exp; |
1da177e4 | 119 | } |
91c9fa8f AV |
120 | path_put(&path); |
121 | exp_put(exp2); | |
1da177e4 LT |
122 | out: |
123 | return err; | |
124 | } | |
125 | ||
289ede45 BF |
126 | static void follow_to_parent(struct path *path) |
127 | { | |
128 | struct dentry *dp; | |
129 | ||
130 | while (path->dentry == path->mnt->mnt_root && follow_up(path)) | |
131 | ; | |
132 | dp = dget_parent(path->dentry); | |
133 | dput(path->dentry); | |
134 | path->dentry = dp; | |
135 | } | |
136 | ||
137 | static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp) | |
138 | { | |
139 | struct svc_export *exp2; | |
140 | struct path path = {.mnt = mntget((*exp)->ex_path.mnt), | |
141 | .dentry = dget(dparent)}; | |
142 | ||
143 | follow_to_parent(&path); | |
144 | ||
145 | exp2 = rqst_exp_parent(rqstp, &path); | |
146 | if (PTR_ERR(exp2) == -ENOENT) { | |
147 | *dentryp = dget(dparent); | |
148 | } else if (IS_ERR(exp2)) { | |
149 | path_put(&path); | |
150 | return PTR_ERR(exp2); | |
151 | } else { | |
152 | *dentryp = dget(path.dentry); | |
153 | exp_put(*exp); | |
154 | *exp = exp2; | |
155 | } | |
156 | path_put(&path); | |
157 | return 0; | |
158 | } | |
159 | ||
82ead7fe BF |
160 | /* |
161 | * For nfsd purposes, we treat V4ROOT exports as though there was an | |
162 | * export at *every* directory. | |
163 | */ | |
3227fa41 | 164 | int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp) |
82ead7fe BF |
165 | { |
166 | if (d_mountpoint(dentry)) | |
167 | return 1; | |
168 | if (!(exp->ex_flags & NFSEXP_V4ROOT)) | |
169 | return 0; | |
170 | return dentry->d_inode != NULL; | |
171 | } | |
172 | ||
6264d69d | 173 | __be32 |
6c0a654d | 174 | nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp, |
5a022fc8 | 175 | const char *name, unsigned int len, |
6c0a654d | 176 | struct svc_export **exp_ret, struct dentry **dentry_ret) |
1da177e4 LT |
177 | { |
178 | struct svc_export *exp; | |
179 | struct dentry *dparent; | |
180 | struct dentry *dentry; | |
6264d69d AV |
181 | __be32 err; |
182 | int host_err; | |
1da177e4 LT |
183 | |
184 | dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); | |
185 | ||
186 | /* Obtain dentry and export. */ | |
8837abca | 187 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); |
1da177e4 LT |
188 | if (err) |
189 | return err; | |
190 | ||
191 | dparent = fhp->fh_dentry; | |
192 | exp = fhp->fh_export; | |
193 | exp_get(exp); | |
194 | ||
1da177e4 LT |
195 | /* Lookup the name, but don't follow links */ |
196 | if (isdotent(name, len)) { | |
197 | if (len==1) | |
198 | dentry = dget(dparent); | |
54775491 | 199 | else if (dparent != exp->ex_path.dentry) |
1da177e4 | 200 | dentry = dget_parent(dparent); |
fed83811 | 201 | else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp)) |
1da177e4 LT |
202 | dentry = dget(dparent); /* .. == . just like at / */ |
203 | else { | |
204 | /* checking mountpoint crossing is very different when stepping up */ | |
289ede45 BF |
205 | host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry); |
206 | if (host_err) | |
1da177e4 | 207 | goto out_nfserr; |
1da177e4 LT |
208 | } |
209 | } else { | |
210 | fh_lock(fhp); | |
211 | dentry = lookup_one_len(name, dparent, len); | |
6264d69d | 212 | host_err = PTR_ERR(dentry); |
1da177e4 LT |
213 | if (IS_ERR(dentry)) |
214 | goto out_nfserr; | |
215 | /* | |
216 | * check if we have crossed a mount point ... | |
217 | */ | |
82ead7fe | 218 | if (nfsd_mountpoint(dentry, exp)) { |
6264d69d | 219 | if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { |
1da177e4 LT |
220 | dput(dentry); |
221 | goto out_nfserr; | |
222 | } | |
223 | } | |
224 | } | |
6c0a654d BF |
225 | *dentry_ret = dentry; |
226 | *exp_ret = exp; | |
227 | return 0; | |
228 | ||
229 | out_nfserr: | |
230 | exp_put(exp); | |
231 | return nfserrno(host_err); | |
232 | } | |
233 | ||
234 | /* | |
235 | * Look up one component of a pathname. | |
236 | * N.B. After this call _both_ fhp and resfh need an fh_put | |
237 | * | |
238 | * If the lookup would cross a mountpoint, and the mounted filesystem | |
239 | * is exported to the client with NFSEXP_NOHIDE, then the lookup is | |
240 | * accepted as it stands and the mounted directory is | |
241 | * returned. Otherwise the covered directory is returned. | |
242 | * NOTE: this mountpoint crossing is not supported properly by all | |
243 | * clients and is explicitly disallowed for NFSv3 | |
244 | * NeilBrown <neilb@cse.unsw.edu.au> | |
245 | */ | |
246 | __be32 | |
247 | nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, | |
5a022fc8 | 248 | unsigned int len, struct svc_fh *resfh) |
6c0a654d BF |
249 | { |
250 | struct svc_export *exp; | |
251 | struct dentry *dentry; | |
252 | __be32 err; | |
253 | ||
254 | err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); | |
255 | if (err) | |
256 | return err; | |
32c1eb0c AA |
257 | err = check_nfsd_access(exp, rqstp); |
258 | if (err) | |
259 | goto out; | |
1da177e4 LT |
260 | /* |
261 | * Note: we compose the file handle now, but as the | |
262 | * dentry may be negative, it may need to be updated. | |
263 | */ | |
264 | err = fh_compose(resfh, exp, dentry, fhp); | |
265 | if (!err && !dentry->d_inode) | |
266 | err = nfserr_noent; | |
32c1eb0c | 267 | out: |
1da177e4 | 268 | dput(dentry); |
1da177e4 LT |
269 | exp_put(exp); |
270 | return err; | |
1da177e4 LT |
271 | } |
272 | ||
6c0a654d | 273 | |
1da177e4 LT |
274 | /* |
275 | * Set various file attributes. | |
276 | * N.B. After this call fhp needs an fh_put | |
277 | */ | |
6264d69d | 278 | __be32 |
1da177e4 LT |
279 | nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, |
280 | int check_guard, time_t guardtime) | |
281 | { | |
282 | struct dentry *dentry; | |
283 | struct inode *inode; | |
8837abca | 284 | int accmode = NFSD_MAY_SATTR; |
1da177e4 | 285 | int ftype = 0; |
6264d69d AV |
286 | __be32 err; |
287 | int host_err; | |
1da177e4 LT |
288 | int size_change = 0; |
289 | ||
290 | if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) | |
8837abca | 291 | accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; |
1da177e4 LT |
292 | if (iap->ia_valid & ATTR_SIZE) |
293 | ftype = S_IFREG; | |
294 | ||
295 | /* Get inode */ | |
296 | err = fh_verify(rqstp, fhp, ftype, accmode); | |
15b7a1b8 | 297 | if (err) |
1da177e4 LT |
298 | goto out; |
299 | ||
300 | dentry = fhp->fh_dentry; | |
301 | inode = dentry->d_inode; | |
302 | ||
15b7a1b8 N |
303 | /* Ignore any mode updates on symlinks */ |
304 | if (S_ISLNK(inode->i_mode)) | |
305 | iap->ia_valid &= ~ATTR_MODE; | |
306 | ||
307 | if (!iap->ia_valid) | |
308 | goto out; | |
309 | ||
9c85fca5 CH |
310 | /* |
311 | * NFSv2 does not differentiate between "set-[ac]time-to-now" | |
1da177e4 LT |
312 | * which only requires access, and "set-[ac]time-to-X" which |
313 | * requires ownership. | |
314 | * So if it looks like it might be "set both to the same time which | |
315 | * is close to now", and if inode_change_ok fails, then we | |
316 | * convert to "set to now" instead of "set to explicit time" | |
317 | * | |
318 | * We only call inode_change_ok as the last test as technically | |
319 | * it is not an interface that we should be using. It is only | |
320 | * valid if the filesystem does not define it's own i_op->setattr. | |
321 | */ | |
322 | #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) | |
323 | #define MAX_TOUCH_TIME_ERROR (30*60) | |
9c85fca5 CH |
324 | if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET && |
325 | iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) { | |
326 | /* | |
327 | * Looks probable. | |
328 | * | |
329 | * Now just make sure time is in the right ballpark. | |
330 | * Solaris, at least, doesn't seem to care what the time | |
331 | * request is. We require it be within 30 minutes of now. | |
1da177e4 | 332 | */ |
9c85fca5 CH |
333 | time_t delta = iap->ia_atime.tv_sec - get_seconds(); |
334 | if (delta < 0) | |
335 | delta = -delta; | |
336 | if (delta < MAX_TOUCH_TIME_ERROR && | |
337 | inode_change_ok(inode, iap) != 0) { | |
338 | /* | |
339 | * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME. | |
340 | * This will cause notify_change to set these times | |
341 | * to "now" | |
342 | */ | |
343 | iap->ia_valid &= ~BOTH_TIME_SET; | |
344 | } | |
1da177e4 LT |
345 | } |
346 | ||
9c85fca5 CH |
347 | /* |
348 | * The size case is special. | |
349 | * It changes the file as well as the attributes. | |
350 | */ | |
1da177e4 LT |
351 | if (iap->ia_valid & ATTR_SIZE) { |
352 | if (iap->ia_size < inode->i_size) { | |
8837abca MS |
353 | err = nfsd_permission(rqstp, fhp->fh_export, dentry, |
354 | NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE); | |
1da177e4 LT |
355 | if (err) |
356 | goto out; | |
357 | } | |
358 | ||
359 | /* | |
360 | * If we are changing the size of the file, then | |
361 | * we need to break all leases. | |
362 | */ | |
6264d69d AV |
363 | host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK); |
364 | if (host_err == -EWOULDBLOCK) | |
365 | host_err = -ETIMEDOUT; | |
366 | if (host_err) /* ENOMEM or EWOULDBLOCK */ | |
1da177e4 LT |
367 | goto out_nfserr; |
368 | ||
6264d69d AV |
369 | host_err = get_write_access(inode); |
370 | if (host_err) | |
1da177e4 LT |
371 | goto out_nfserr; |
372 | ||
373 | size_change = 1; | |
6264d69d AV |
374 | host_err = locks_verify_truncate(inode, NULL, iap->ia_size); |
375 | if (host_err) { | |
1da177e4 LT |
376 | put_write_access(inode); |
377 | goto out_nfserr; | |
378 | } | |
1da177e4 LT |
379 | } |
380 | ||
ca456252 | 381 | /* sanitize the mode change */ |
1da177e4 LT |
382 | if (iap->ia_valid & ATTR_MODE) { |
383 | iap->ia_mode &= S_IALLUGO; | |
dee3209d | 384 | iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); |
ca456252 JL |
385 | } |
386 | ||
387 | /* Revoke setuid/setgid on chown */ | |
0953e620 SP |
388 | if (!S_ISDIR(inode->i_mode) && |
389 | (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) || | |
390 | ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) { | |
ca456252 JL |
391 | iap->ia_valid |= ATTR_KILL_PRIV; |
392 | if (iap->ia_valid & ATTR_MODE) { | |
393 | /* we're setting mode too, just clear the s*id bits */ | |
8a0ce7d9 | 394 | iap->ia_mode &= ~S_ISUID; |
ca456252 JL |
395 | if (iap->ia_mode & S_IXGRP) |
396 | iap->ia_mode &= ~S_ISGID; | |
397 | } else { | |
398 | /* set ATTR_KILL_* bits and let VFS handle it */ | |
399 | iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); | |
8a0ce7d9 | 400 | } |
1da177e4 LT |
401 | } |
402 | ||
1da177e4 LT |
403 | /* Change the attributes. */ |
404 | ||
405 | iap->ia_valid |= ATTR_CTIME; | |
406 | ||
407 | err = nfserr_notsync; | |
408 | if (!check_guard || guardtime == inode->i_ctime.tv_sec) { | |
409 | fh_lock(fhp); | |
6264d69d AV |
410 | host_err = notify_change(dentry, iap); |
411 | err = nfserrno(host_err); | |
1da177e4 LT |
412 | fh_unlock(fhp); |
413 | } | |
414 | if (size_change) | |
415 | put_write_access(inode); | |
416 | if (!err) | |
417 | if (EX_ISSYNC(fhp->fh_export)) | |
418 | write_inode_now(inode, 1); | |
419 | out: | |
420 | return err; | |
421 | ||
422 | out_nfserr: | |
6264d69d | 423 | err = nfserrno(host_err); |
1da177e4 LT |
424 | goto out; |
425 | } | |
426 | ||
5be196e5 CH |
427 | #if defined(CONFIG_NFSD_V2_ACL) || \ |
428 | defined(CONFIG_NFSD_V3_ACL) || \ | |
429 | defined(CONFIG_NFSD_V4) | |
430 | static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) | |
431 | { | |
432 | ssize_t buflen; | |
6c6a426f | 433 | ssize_t ret; |
5be196e5 CH |
434 | |
435 | buflen = vfs_getxattr(dentry, key, NULL, 0); | |
436 | if (buflen <= 0) | |
437 | return buflen; | |
1da177e4 | 438 | |
5be196e5 CH |
439 | *buf = kmalloc(buflen, GFP_KERNEL); |
440 | if (!*buf) | |
441 | return -ENOMEM; | |
442 | ||
6c6a426f KK |
443 | ret = vfs_getxattr(dentry, key, *buf, buflen); |
444 | if (ret < 0) | |
445 | kfree(*buf); | |
446 | return ret; | |
5be196e5 CH |
447 | } |
448 | #endif | |
449 | ||
450 | #if defined(CONFIG_NFSD_V4) | |
1da177e4 LT |
451 | static int |
452 | set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) | |
453 | { | |
454 | int len; | |
455 | size_t buflen; | |
456 | char *buf = NULL; | |
457 | int error = 0; | |
1da177e4 LT |
458 | |
459 | buflen = posix_acl_xattr_size(pacl->a_count); | |
460 | buf = kmalloc(buflen, GFP_KERNEL); | |
461 | error = -ENOMEM; | |
462 | if (buf == NULL) | |
463 | goto out; | |
464 | ||
465 | len = posix_acl_to_xattr(pacl, buf, buflen); | |
466 | if (len < 0) { | |
467 | error = len; | |
468 | goto out; | |
469 | } | |
470 | ||
5be196e5 | 471 | error = vfs_setxattr(dentry, key, buf, len, 0); |
1da177e4 LT |
472 | out: |
473 | kfree(buf); | |
474 | return error; | |
475 | } | |
476 | ||
6264d69d | 477 | __be32 |
1da177e4 LT |
478 | nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, |
479 | struct nfs4_acl *acl) | |
480 | { | |
6264d69d AV |
481 | __be32 error; |
482 | int host_error; | |
1da177e4 LT |
483 | struct dentry *dentry; |
484 | struct inode *inode; | |
485 | struct posix_acl *pacl = NULL, *dpacl = NULL; | |
486 | unsigned int flags = 0; | |
487 | ||
488 | /* Get inode */ | |
8837abca | 489 | error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); |
1da177e4 | 490 | if (error) |
4b2ca38a | 491 | return error; |
1da177e4 LT |
492 | |
493 | dentry = fhp->fh_dentry; | |
494 | inode = dentry->d_inode; | |
495 | if (S_ISDIR(inode->i_mode)) | |
496 | flags = NFS4_ACL_DIR; | |
497 | ||
6264d69d AV |
498 | host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); |
499 | if (host_error == -EINVAL) { | |
4b2ca38a | 500 | return nfserr_attrnotsupp; |
6264d69d | 501 | } else if (host_error < 0) |
1da177e4 LT |
502 | goto out_nfserr; |
503 | ||
6264d69d AV |
504 | host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); |
505 | if (host_error < 0) | |
4b2ca38a | 506 | goto out_release; |
1da177e4 | 507 | |
4b2ca38a | 508 | if (S_ISDIR(inode->i_mode)) |
6264d69d | 509 | host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); |
1da177e4 | 510 | |
4b2ca38a | 511 | out_release: |
1da177e4 LT |
512 | posix_acl_release(pacl); |
513 | posix_acl_release(dpacl); | |
1da177e4 | 514 | out_nfserr: |
f34f9242 | 515 | if (host_error == -EOPNOTSUPP) |
4b2ca38a | 516 | return nfserr_attrnotsupp; |
f34f9242 | 517 | else |
4b2ca38a | 518 | return nfserrno(host_error); |
1da177e4 LT |
519 | } |
520 | ||
521 | static struct posix_acl * | |
522 | _get_posix_acl(struct dentry *dentry, char *key) | |
523 | { | |
5be196e5 | 524 | void *buf = NULL; |
1da177e4 | 525 | struct posix_acl *pacl = NULL; |
5be196e5 | 526 | int buflen; |
1da177e4 | 527 | |
5be196e5 CH |
528 | buflen = nfsd_getxattr(dentry, key, &buf); |
529 | if (!buflen) | |
530 | buflen = -ENODATA; | |
531 | if (buflen <= 0) | |
532 | return ERR_PTR(buflen); | |
1da177e4 LT |
533 | |
534 | pacl = posix_acl_from_xattr(buf, buflen); | |
1da177e4 LT |
535 | kfree(buf); |
536 | return pacl; | |
1da177e4 LT |
537 | } |
538 | ||
539 | int | |
540 | nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) | |
541 | { | |
542 | struct inode *inode = dentry->d_inode; | |
543 | int error = 0; | |
544 | struct posix_acl *pacl = NULL, *dpacl = NULL; | |
545 | unsigned int flags = 0; | |
546 | ||
9a59f452 | 547 | pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); |
1da177e4 LT |
548 | if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) |
549 | pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); | |
550 | if (IS_ERR(pacl)) { | |
551 | error = PTR_ERR(pacl); | |
552 | pacl = NULL; | |
553 | goto out; | |
554 | } | |
555 | ||
556 | if (S_ISDIR(inode->i_mode)) { | |
9a59f452 | 557 | dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); |
1da177e4 LT |
558 | if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) |
559 | dpacl = NULL; | |
560 | else if (IS_ERR(dpacl)) { | |
561 | error = PTR_ERR(dpacl); | |
562 | dpacl = NULL; | |
563 | goto out; | |
564 | } | |
565 | flags = NFS4_ACL_DIR; | |
566 | } | |
567 | ||
568 | *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); | |
569 | if (IS_ERR(*acl)) { | |
570 | error = PTR_ERR(*acl); | |
571 | *acl = NULL; | |
572 | } | |
573 | out: | |
574 | posix_acl_release(pacl); | |
575 | posix_acl_release(dpacl); | |
576 | return error; | |
577 | } | |
578 | ||
579 | #endif /* defined(CONFIG_NFS_V4) */ | |
580 | ||
581 | #ifdef CONFIG_NFSD_V3 | |
582 | /* | |
583 | * Check server access rights to a file system object | |
584 | */ | |
585 | struct accessmap { | |
586 | u32 access; | |
587 | int how; | |
588 | }; | |
589 | static struct accessmap nfs3_regaccess[] = { | |
8837abca MS |
590 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
591 | { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, | |
592 | { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, | |
593 | { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, | |
1da177e4 LT |
594 | |
595 | { 0, 0 } | |
596 | }; | |
597 | ||
598 | static struct accessmap nfs3_diraccess[] = { | |
8837abca MS |
599 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
600 | { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, | |
601 | { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, | |
602 | { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, | |
603 | { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, | |
1da177e4 LT |
604 | |
605 | { 0, 0 } | |
606 | }; | |
607 | ||
608 | static struct accessmap nfs3_anyaccess[] = { | |
609 | /* Some clients - Solaris 2.6 at least, make an access call | |
610 | * to the server to check for access for things like /dev/null | |
611 | * (which really, the server doesn't care about). So | |
612 | * We provide simple access checking for them, looking | |
613 | * mainly at mode bits, and we make sure to ignore read-only | |
614 | * filesystem checks | |
615 | */ | |
8837abca MS |
616 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
617 | { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, | |
618 | { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, | |
619 | { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, | |
1da177e4 LT |
620 | |
621 | { 0, 0 } | |
622 | }; | |
623 | ||
6264d69d | 624 | __be32 |
1da177e4 LT |
625 | nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) |
626 | { | |
627 | struct accessmap *map; | |
628 | struct svc_export *export; | |
629 | struct dentry *dentry; | |
630 | u32 query, result = 0, sresult = 0; | |
6264d69d | 631 | __be32 error; |
1da177e4 | 632 | |
8837abca | 633 | error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); |
1da177e4 LT |
634 | if (error) |
635 | goto out; | |
636 | ||
637 | export = fhp->fh_export; | |
638 | dentry = fhp->fh_dentry; | |
639 | ||
640 | if (S_ISREG(dentry->d_inode->i_mode)) | |
641 | map = nfs3_regaccess; | |
642 | else if (S_ISDIR(dentry->d_inode->i_mode)) | |
643 | map = nfs3_diraccess; | |
644 | else | |
645 | map = nfs3_anyaccess; | |
646 | ||
647 | ||
648 | query = *access; | |
649 | for (; map->access; map++) { | |
650 | if (map->access & query) { | |
6264d69d | 651 | __be32 err2; |
1da177e4 LT |
652 | |
653 | sresult |= map->access; | |
654 | ||
0ec757df | 655 | err2 = nfsd_permission(rqstp, export, dentry, map->how); |
1da177e4 LT |
656 | switch (err2) { |
657 | case nfs_ok: | |
658 | result |= map->access; | |
659 | break; | |
660 | ||
661 | /* the following error codes just mean the access was not allowed, | |
662 | * rather than an error occurred */ | |
663 | case nfserr_rofs: | |
664 | case nfserr_acces: | |
665 | case nfserr_perm: | |
666 | /* simply don't "or" in the access bit. */ | |
667 | break; | |
668 | default: | |
669 | error = err2; | |
670 | goto out; | |
671 | } | |
672 | } | |
673 | } | |
674 | *access = result; | |
675 | if (supported) | |
676 | *supported = sresult; | |
677 | ||
678 | out: | |
679 | return error; | |
680 | } | |
681 | #endif /* CONFIG_NFSD_V3 */ | |
682 | ||
683 | ||
684 | ||
685 | /* | |
686 | * Open an existing file or directory. | |
687 | * The access argument indicates the type of open (read/write/lock) | |
688 | * N.B. After this call fhp needs an fh_put | |
689 | */ | |
6264d69d | 690 | __be32 |
1da177e4 LT |
691 | nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, |
692 | int access, struct file **filp) | |
693 | { | |
694 | struct dentry *dentry; | |
695 | struct inode *inode; | |
6264d69d AV |
696 | int flags = O_RDONLY|O_LARGEFILE; |
697 | __be32 err; | |
698 | int host_err; | |
1da177e4 | 699 | |
e0e81739 DH |
700 | validate_process_creds(); |
701 | ||
1da177e4 LT |
702 | /* |
703 | * If we get here, then the client has already done an "open", | |
704 | * and (hopefully) checked permission - so allow OWNER_OVERRIDE | |
705 | * in case a chmod has now revoked permission. | |
706 | */ | |
8837abca | 707 | err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE); |
1da177e4 LT |
708 | if (err) |
709 | goto out; | |
710 | ||
711 | dentry = fhp->fh_dentry; | |
712 | inode = dentry->d_inode; | |
713 | ||
714 | /* Disallow write access to files with the append-only bit set | |
715 | * or any access when mandatory locking enabled | |
716 | */ | |
717 | err = nfserr_perm; | |
8837abca | 718 | if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE)) |
1da177e4 | 719 | goto out; |
5e7fc436 BF |
720 | /* |
721 | * We must ignore files (but only files) which might have mandatory | |
722 | * locks on them because there is no way to know if the accesser has | |
723 | * the lock. | |
724 | */ | |
725 | if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) | |
1da177e4 LT |
726 | goto out; |
727 | ||
728 | if (!inode->i_fop) | |
729 | goto out; | |
730 | ||
731 | /* | |
732 | * Check to see if there are any leases on this file. | |
733 | * This may block while leases are broken. | |
734 | */ | |
8837abca | 735 | host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0)); |
6264d69d AV |
736 | if (host_err == -EWOULDBLOCK) |
737 | host_err = -ETIMEDOUT; | |
738 | if (host_err) /* NOMEM or WOULDBLOCK */ | |
1da177e4 LT |
739 | goto out_nfserr; |
740 | ||
8837abca MS |
741 | if (access & NFSD_MAY_WRITE) { |
742 | if (access & NFSD_MAY_READ) | |
9ecb6a08 BF |
743 | flags = O_RDWR|O_LARGEFILE; |
744 | else | |
745 | flags = O_WRONLY|O_LARGEFILE; | |
1da177e4 | 746 | } |
54775491 | 747 | *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt), |
033a666c | 748 | flags, current_cred()); |
1da177e4 | 749 | if (IS_ERR(*filp)) |
6264d69d | 750 | host_err = PTR_ERR(*filp); |
aeaa5ccd CE |
751 | else |
752 | host_err = ima_file_check(*filp, access); | |
1da177e4 | 753 | out_nfserr: |
6264d69d | 754 | err = nfserrno(host_err); |
1da177e4 | 755 | out: |
e0e81739 | 756 | validate_process_creds(); |
1da177e4 LT |
757 | return err; |
758 | } | |
759 | ||
760 | /* | |
761 | * Close a file. | |
762 | */ | |
763 | void | |
764 | nfsd_close(struct file *filp) | |
765 | { | |
766 | fput(filp); | |
767 | } | |
768 | ||
9a8d248e BF |
769 | /* |
770 | * Sync a file | |
771 | * As this calls fsync (not fdatasync) there is no need for a write_inode | |
772 | * after it. | |
773 | */ | |
774 | static inline int nfsd_dosync(struct file *filp, struct dentry *dp, | |
775 | const struct file_operations *fop) | |
776 | { | |
777 | struct inode *inode = dp->d_inode; | |
778 | int (*fsync) (struct file *, struct dentry *, int); | |
779 | int err; | |
780 | ||
7211a4e8 | 781 | err = filemap_write_and_wait(inode->i_mapping); |
9a8d248e BF |
782 | if (err == 0 && fop && (fsync = fop->fsync)) |
783 | err = fsync(filp, dp, 0); | |
9a8d248e BF |
784 | return err; |
785 | } | |
786 | ||
a334de28 | 787 | static int |
1da177e4 LT |
788 | nfsd_sync(struct file *filp) |
789 | { | |
9a8d248e BF |
790 | int err; |
791 | struct inode *inode = filp->f_path.dentry->d_inode; | |
792 | dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name); | |
793 | mutex_lock(&inode->i_mutex); | |
794 | err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op); | |
795 | mutex_unlock(&inode->i_mutex); | |
796 | ||
797 | return err; | |
1da177e4 LT |
798 | } |
799 | ||
f193fbab | 800 | int |
9a8d248e | 801 | nfsd_sync_dir(struct dentry *dp) |
1da177e4 | 802 | { |
9a8d248e | 803 | return nfsd_dosync(NULL, dp, dp->d_inode->i_fop); |
1da177e4 LT |
804 | } |
805 | ||
806 | /* | |
807 | * Obtain the readahead parameters for the file | |
808 | * specified by (dev, ino). | |
809 | */ | |
1da177e4 LT |
810 | |
811 | static inline struct raparms * | |
812 | nfsd_get_raparms(dev_t dev, ino_t ino) | |
813 | { | |
814 | struct raparms *ra, **rap, **frap = NULL; | |
815 | int depth = 0; | |
fce1456a GB |
816 | unsigned int hash; |
817 | struct raparm_hbucket *rab; | |
818 | ||
819 | hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; | |
820 | rab = &raparm_hash[hash]; | |
1da177e4 | 821 | |
fce1456a GB |
822 | spin_lock(&rab->pb_lock); |
823 | for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { | |
1da177e4 LT |
824 | if (ra->p_ino == ino && ra->p_dev == dev) |
825 | goto found; | |
826 | depth++; | |
827 | if (ra->p_count == 0) | |
828 | frap = rap; | |
829 | } | |
830 | depth = nfsdstats.ra_size*11/10; | |
831 | if (!frap) { | |
fce1456a | 832 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
833 | return NULL; |
834 | } | |
835 | rap = frap; | |
836 | ra = *frap; | |
837 | ra->p_dev = dev; | |
838 | ra->p_ino = ino; | |
839 | ra->p_set = 0; | |
fce1456a | 840 | ra->p_hindex = hash; |
1da177e4 | 841 | found: |
fce1456a | 842 | if (rap != &rab->pb_head) { |
1da177e4 | 843 | *rap = ra->p_next; |
fce1456a GB |
844 | ra->p_next = rab->pb_head; |
845 | rab->pb_head = ra; | |
1da177e4 LT |
846 | } |
847 | ra->p_count++; | |
848 | nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; | |
fce1456a | 849 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
850 | return ra; |
851 | } | |
852 | ||
853 | /* | |
cf8208d0 JA |
854 | * Grab and keep cached pages associated with a file in the svc_rqst |
855 | * so that they can be passed to the network sendmsg/sendpage routines | |
856 | * directly. They will be released after the sending has completed. | |
1da177e4 LT |
857 | */ |
858 | static int | |
cf8208d0 JA |
859 | nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
860 | struct splice_desc *sd) | |
1da177e4 | 861 | { |
cf8208d0 | 862 | struct svc_rqst *rqstp = sd->u.data; |
44524359 | 863 | struct page **pp = rqstp->rq_respages + rqstp->rq_resused; |
cf8208d0 JA |
864 | struct page *page = buf->page; |
865 | size_t size; | |
866 | int ret; | |
1da177e4 | 867 | |
cac36bb0 | 868 | ret = buf->ops->confirm(pipe, buf); |
cf8208d0 JA |
869 | if (unlikely(ret)) |
870 | return ret; | |
871 | ||
872 | size = sd->len; | |
1da177e4 LT |
873 | |
874 | if (rqstp->rq_res.page_len == 0) { | |
875 | get_page(page); | |
44524359 N |
876 | put_page(*pp); |
877 | *pp = page; | |
878 | rqstp->rq_resused++; | |
cf8208d0 | 879 | rqstp->rq_res.page_base = buf->offset; |
1da177e4 | 880 | rqstp->rq_res.page_len = size; |
44524359 | 881 | } else if (page != pp[-1]) { |
1da177e4 | 882 | get_page(page); |
250f3915 N |
883 | if (*pp) |
884 | put_page(*pp); | |
44524359 N |
885 | *pp = page; |
886 | rqstp->rq_resused++; | |
1da177e4 | 887 | rqstp->rq_res.page_len += size; |
44524359 | 888 | } else |
1da177e4 | 889 | rqstp->rq_res.page_len += size; |
1da177e4 | 890 | |
1da177e4 LT |
891 | return size; |
892 | } | |
893 | ||
cf8208d0 JA |
894 | static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, |
895 | struct splice_desc *sd) | |
896 | { | |
897 | return __splice_from_pipe(pipe, sd, nfsd_splice_actor); | |
898 | } | |
899 | ||
a8754bee DH |
900 | static inline int svc_msnfs(struct svc_fh *ffhp) |
901 | { | |
902 | #ifdef MSNFS | |
903 | return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS); | |
904 | #else | |
905 | return 0; | |
906 | #endif | |
907 | } | |
908 | ||
6264d69d | 909 | static __be32 |
1da177e4 LT |
910 | nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
911 | loff_t offset, struct kvec *vec, int vlen, unsigned long *count) | |
912 | { | |
913 | struct inode *inode; | |
914 | struct raparms *ra; | |
915 | mm_segment_t oldfs; | |
6264d69d AV |
916 | __be32 err; |
917 | int host_err; | |
1da177e4 LT |
918 | |
919 | err = nfserr_perm; | |
7eaa36e2 | 920 | inode = file->f_path.dentry->d_inode; |
a8754bee DH |
921 | |
922 | if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count)) | |
1da177e4 | 923 | goto out; |
1da177e4 LT |
924 | |
925 | /* Get readahead parameters */ | |
926 | ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); | |
927 | ||
928 | if (ra && ra->p_set) | |
929 | file->f_ra = ra->p_ra; | |
930 | ||
cf8208d0 JA |
931 | if (file->f_op->splice_read && rqstp->rq_splice_ok) { |
932 | struct splice_desc sd = { | |
933 | .len = 0, | |
934 | .total_len = *count, | |
935 | .pos = offset, | |
936 | .u.data = rqstp, | |
937 | }; | |
938 | ||
4fbef206 | 939 | rqstp->rq_resused = 1; |
cf8208d0 | 940 | host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); |
1da177e4 LT |
941 | } else { |
942 | oldfs = get_fs(); | |
943 | set_fs(KERNEL_DS); | |
6264d69d | 944 | host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); |
1da177e4 LT |
945 | set_fs(oldfs); |
946 | } | |
947 | ||
948 | /* Write back readahead params */ | |
949 | if (ra) { | |
fce1456a GB |
950 | struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; |
951 | spin_lock(&rab->pb_lock); | |
1da177e4 LT |
952 | ra->p_ra = file->f_ra; |
953 | ra->p_set = 1; | |
954 | ra->p_count--; | |
fce1456a | 955 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
956 | } |
957 | ||
6264d69d AV |
958 | if (host_err >= 0) { |
959 | nfsdstats.io_read += host_err; | |
960 | *count = host_err; | |
1da177e4 | 961 | err = 0; |
7eaa36e2 | 962 | fsnotify_access(file->f_path.dentry); |
1da177e4 | 963 | } else |
6264d69d | 964 | err = nfserrno(host_err); |
1da177e4 LT |
965 | out: |
966 | return err; | |
967 | } | |
968 | ||
9f708e40 NB |
969 | static void kill_suid(struct dentry *dentry) |
970 | { | |
971 | struct iattr ia; | |
b5376771 | 972 | ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; |
9f708e40 | 973 | |
1b1dcc1b | 974 | mutex_lock(&dentry->d_inode->i_mutex); |
9f708e40 | 975 | notify_change(dentry, &ia); |
1b1dcc1b | 976 | mutex_unlock(&dentry->d_inode->i_mutex); |
9f708e40 NB |
977 | } |
978 | ||
d911df7b BF |
979 | /* |
980 | * Gathered writes: If another process is currently writing to the file, | |
981 | * there's a high chance this is another nfsd (triggered by a bulk write | |
982 | * from a client's biod). Rather than syncing the file with each write | |
983 | * request, we sleep for 10 msec. | |
984 | * | |
985 | * I don't know if this roughly approximates C. Juszak's idea of | |
986 | * gathered writes, but it's a nice and simple solution (IMHO), and it | |
987 | * seems to work:-) | |
988 | * | |
989 | * Note: we do this only in the NFSv2 case, since v3 and higher have a | |
990 | * better tool (separate unstable writes and commits) for solving this | |
991 | * problem. | |
992 | */ | |
993 | static int wait_for_concurrent_writes(struct file *file) | |
994 | { | |
995 | struct inode *inode = file->f_path.dentry->d_inode; | |
996 | static ino_t last_ino; | |
997 | static dev_t last_dev; | |
998 | int err = 0; | |
999 | ||
1000 | if (atomic_read(&inode->i_writecount) > 1 | |
1001 | || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { | |
1002 | dprintk("nfsd: write defer %d\n", task_pid_nr(current)); | |
1003 | msleep(10); | |
1004 | dprintk("nfsd: write resume %d\n", task_pid_nr(current)); | |
1005 | } | |
1006 | ||
1007 | if (inode->i_state & I_DIRTY) { | |
1008 | dprintk("nfsd: write sync %d\n", task_pid_nr(current)); | |
1009 | err = nfsd_sync(file); | |
1010 | } | |
1011 | last_ino = inode->i_ino; | |
1012 | last_dev = inode->i_sb->s_dev; | |
1013 | return err; | |
1014 | } | |
1015 | ||
6264d69d | 1016 | static __be32 |
1da177e4 LT |
1017 | nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
1018 | loff_t offset, struct kvec *vec, int vlen, | |
31dec253 | 1019 | unsigned long *cnt, int *stablep) |
1da177e4 LT |
1020 | { |
1021 | struct svc_export *exp; | |
1022 | struct dentry *dentry; | |
1023 | struct inode *inode; | |
1024 | mm_segment_t oldfs; | |
6264d69d AV |
1025 | __be32 err = 0; |
1026 | int host_err; | |
1da177e4 | 1027 | int stable = *stablep; |
48e03bc5 | 1028 | int use_wgather; |
1da177e4 | 1029 | |
45bd3b3d | 1030 | #ifdef MSNFS |
1da177e4 LT |
1031 | err = nfserr_perm; |
1032 | ||
1da177e4 | 1033 | if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && |
31dec253 | 1034 | (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt))) |
1da177e4 LT |
1035 | goto out; |
1036 | #endif | |
1037 | ||
7eaa36e2 | 1038 | dentry = file->f_path.dentry; |
1da177e4 LT |
1039 | inode = dentry->d_inode; |
1040 | exp = fhp->fh_export; | |
1041 | ||
1042 | /* | |
1043 | * Request sync writes if | |
1044 | * - the sync export option has been set, or | |
1045 | * - the client requested O_SYNC behavior (NFSv3 feature). | |
1046 | * - The file system doesn't support fsync(). | |
48e03bc5 | 1047 | * When NFSv2 gathered writes have been configured for this volume, |
1da177e4 LT |
1048 | * flushing the data to disk is handled separately below. |
1049 | */ | |
48e03bc5 | 1050 | use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); |
1da177e4 | 1051 | |
3ba15148 | 1052 | if (!file->f_op->fsync) {/* COMMIT3 cannot work */ |
1da177e4 LT |
1053 | stable = 2; |
1054 | *stablep = 2; /* FILE_SYNC */ | |
1055 | } | |
1056 | ||
1057 | if (!EX_ISSYNC(exp)) | |
1058 | stable = 0; | |
48e03bc5 | 1059 | if (stable && !use_wgather) { |
db1dd4d3 | 1060 | spin_lock(&file->f_lock); |
1da177e4 | 1061 | file->f_flags |= O_SYNC; |
db1dd4d3 JC |
1062 | spin_unlock(&file->f_lock); |
1063 | } | |
1da177e4 LT |
1064 | |
1065 | /* Write the data. */ | |
1066 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
6264d69d | 1067 | host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); |
1da177e4 | 1068 | set_fs(oldfs); |
e4636d53 BF |
1069 | if (host_err < 0) |
1070 | goto out_nfserr; | |
1071 | *cnt = host_err; | |
1072 | nfsdstats.io_write += host_err; | |
1073 | fsnotify_modify(file->f_path.dentry); | |
1da177e4 LT |
1074 | |
1075 | /* clear setuid/setgid flag after write */ | |
e4636d53 | 1076 | if (inode->i_mode & (S_ISUID | S_ISGID)) |
9f708e40 | 1077 | kill_suid(dentry); |
1da177e4 | 1078 | |
e4636d53 | 1079 | if (stable && use_wgather) |
d911df7b | 1080 | host_err = wait_for_concurrent_writes(file); |
1da177e4 | 1081 | |
e4636d53 | 1082 | out_nfserr: |
6264d69d | 1083 | dprintk("nfsd: write complete host_err=%d\n", host_err); |
a0d24b29 | 1084 | if (host_err >= 0) |
1da177e4 | 1085 | err = 0; |
a0d24b29 | 1086 | else |
6264d69d | 1087 | err = nfserrno(host_err); |
1da177e4 LT |
1088 | out: |
1089 | return err; | |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * Read data from a file. count must contain the requested read count | |
1094 | * on entry. On return, *count contains the number of bytes actually read. | |
1095 | * N.B. After this call fhp needs an fh_put | |
1096 | */ | |
6264d69d | 1097 | __be32 |
1da177e4 LT |
1098 | nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
1099 | loff_t offset, struct kvec *vec, int vlen, | |
1100 | unsigned long *count) | |
1101 | { | |
6264d69d | 1102 | __be32 err; |
1da177e4 LT |
1103 | |
1104 | if (file) { | |
0ec757df | 1105 | err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, |
8837abca | 1106 | NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); |
1da177e4 LT |
1107 | if (err) |
1108 | goto out; | |
1109 | err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); | |
1110 | } else { | |
8837abca | 1111 | err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); |
1da177e4 LT |
1112 | if (err) |
1113 | goto out; | |
1114 | err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); | |
1115 | nfsd_close(file); | |
1116 | } | |
1117 | out: | |
1118 | return err; | |
1119 | } | |
1120 | ||
1121 | /* | |
1122 | * Write data to a file. | |
1123 | * The stable flag requests synchronous writes. | |
1124 | * N.B. After this call fhp needs an fh_put | |
1125 | */ | |
6264d69d | 1126 | __be32 |
1da177e4 | 1127 | nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
31dec253 | 1128 | loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, |
1da177e4 LT |
1129 | int *stablep) |
1130 | { | |
6264d69d | 1131 | __be32 err = 0; |
1da177e4 LT |
1132 | |
1133 | if (file) { | |
0ec757df | 1134 | err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, |
8837abca | 1135 | NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); |
1da177e4 LT |
1136 | if (err) |
1137 | goto out; | |
1138 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, | |
1139 | stablep); | |
1140 | } else { | |
8837abca | 1141 | err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); |
1da177e4 LT |
1142 | if (err) |
1143 | goto out; | |
1144 | ||
1145 | if (cnt) | |
1146 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, | |
1147 | cnt, stablep); | |
1148 | nfsd_close(file); | |
1149 | } | |
1150 | out: | |
1151 | return err; | |
1152 | } | |
1153 | ||
1154 | #ifdef CONFIG_NFSD_V3 | |
1155 | /* | |
1156 | * Commit all pending writes to stable storage. | |
1157 | * Strictly speaking, we could sync just the indicated file region here, | |
1158 | * but there's currently no way we can ask the VFS to do so. | |
1159 | * | |
1160 | * Unfortunately we cannot lock the file to make sure we return full WCC | |
1161 | * data to the client, as locking happens lower down in the filesystem. | |
1162 | */ | |
6264d69d | 1163 | __be32 |
1da177e4 LT |
1164 | nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1165 | loff_t offset, unsigned long count) | |
1166 | { | |
1167 | struct file *file; | |
6264d69d | 1168 | __be32 err; |
1da177e4 LT |
1169 | |
1170 | if ((u64)count > ~(u64)offset) | |
1171 | return nfserr_inval; | |
1172 | ||
8837abca MS |
1173 | err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); |
1174 | if (err) | |
1da177e4 LT |
1175 | return err; |
1176 | if (EX_ISSYNC(fhp->fh_export)) { | |
1177 | if (file->f_op && file->f_op->fsync) { | |
45bd3b3d | 1178 | err = nfserrno(nfsd_sync(file)); |
1da177e4 LT |
1179 | } else { |
1180 | err = nfserr_notsupp; | |
1181 | } | |
1182 | } | |
1183 | ||
1184 | nfsd_close(file); | |
1185 | return err; | |
1186 | } | |
1187 | #endif /* CONFIG_NFSD_V3 */ | |
1188 | ||
f2b0dee2 | 1189 | static __be32 |
5c002b3b BF |
1190 | nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, |
1191 | struct iattr *iap) | |
1192 | { | |
1193 | /* | |
1194 | * Mode has already been set earlier in create: | |
1195 | */ | |
1196 | iap->ia_valid &= ~ATTR_MODE; | |
1197 | /* | |
1198 | * Setting uid/gid works only for root. Irix appears to | |
1199 | * send along the gid on create when it tries to implement | |
1200 | * setgid directories via NFS: | |
1201 | */ | |
5cc0a840 | 1202 | if (current_fsuid() != 0) |
5c002b3b BF |
1203 | iap->ia_valid &= ~(ATTR_UID|ATTR_GID); |
1204 | if (iap->ia_valid) | |
1205 | return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); | |
1206 | return 0; | |
1207 | } | |
1208 | ||
4ac35c2f | 1209 | /* HPUX client sometimes creates a file in mode 000, and sets size to 0. |
1210 | * setting size to 0 may fail for some specific file systems by the permission | |
1211 | * checking which requires WRITE permission but the mode is 000. | |
1212 | * we ignore the resizing(to 0) on the just new created file, since the size is | |
1213 | * 0 after file created. | |
1214 | * | |
1215 | * call this only after vfs_create() is called. | |
1216 | * */ | |
1217 | static void | |
1218 | nfsd_check_ignore_resizing(struct iattr *iap) | |
1219 | { | |
1220 | if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) | |
1221 | iap->ia_valid &= ~ATTR_SIZE; | |
1222 | } | |
1223 | ||
1da177e4 LT |
1224 | /* |
1225 | * Create a file (regular, directory, device, fifo); UNIX sockets | |
1226 | * not yet implemented. | |
1227 | * If the response fh has been verified, the parent directory should | |
1228 | * already be locked. Note that the parent directory is left locked. | |
1229 | * | |
1230 | * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp | |
1231 | */ | |
6264d69d | 1232 | __be32 |
1da177e4 LT |
1233 | nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1234 | char *fname, int flen, struct iattr *iap, | |
1235 | int type, dev_t rdev, struct svc_fh *resfhp) | |
1236 | { | |
1237 | struct dentry *dentry, *dchild = NULL; | |
1238 | struct inode *dirp; | |
6264d69d | 1239 | __be32 err; |
5c002b3b | 1240 | __be32 err2; |
6264d69d | 1241 | int host_err; |
1da177e4 LT |
1242 | |
1243 | err = nfserr_perm; | |
1244 | if (!flen) | |
1245 | goto out; | |
1246 | err = nfserr_exist; | |
1247 | if (isdotent(fname, flen)) | |
1248 | goto out; | |
1249 | ||
8837abca | 1250 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1251 | if (err) |
1252 | goto out; | |
1253 | ||
1254 | dentry = fhp->fh_dentry; | |
1255 | dirp = dentry->d_inode; | |
1256 | ||
1257 | err = nfserr_notdir; | |
acfa4380 | 1258 | if (!dirp->i_op->lookup) |
1da177e4 LT |
1259 | goto out; |
1260 | /* | |
1261 | * Check whether the response file handle has been verified yet. | |
1262 | * If it has, the parent directory should already be locked. | |
1263 | */ | |
1264 | if (!resfhp->fh_dentry) { | |
1265 | /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ | |
12fd3520 | 1266 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 | 1267 | dchild = lookup_one_len(fname, dentry, flen); |
6264d69d | 1268 | host_err = PTR_ERR(dchild); |
1da177e4 LT |
1269 | if (IS_ERR(dchild)) |
1270 | goto out_nfserr; | |
1271 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); | |
1272 | if (err) | |
1273 | goto out; | |
1274 | } else { | |
1275 | /* called from nfsd_proc_create */ | |
1276 | dchild = dget(resfhp->fh_dentry); | |
1277 | if (!fhp->fh_locked) { | |
1278 | /* not actually possible */ | |
1279 | printk(KERN_ERR | |
1280 | "nfsd_create: parent %s/%s not locked!\n", | |
1281 | dentry->d_parent->d_name.name, | |
1282 | dentry->d_name.name); | |
d75f2b9f | 1283 | err = nfserr_io; |
1da177e4 LT |
1284 | goto out; |
1285 | } | |
1286 | } | |
1287 | /* | |
1288 | * Make sure the child dentry is still negative ... | |
1289 | */ | |
1290 | err = nfserr_exist; | |
1291 | if (dchild->d_inode) { | |
1292 | dprintk("nfsd_create: dentry %s/%s not negative!\n", | |
1293 | dentry->d_name.name, dchild->d_name.name); | |
1294 | goto out; | |
1295 | } | |
1296 | ||
1297 | if (!(iap->ia_valid & ATTR_MODE)) | |
1298 | iap->ia_mode = 0; | |
1299 | iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; | |
1300 | ||
07cad1d2 MS |
1301 | err = nfserr_inval; |
1302 | if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { | |
1303 | printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", | |
1304 | type); | |
1305 | goto out; | |
1306 | } | |
1307 | ||
1308 | host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); | |
1309 | if (host_err) | |
1310 | goto out_nfserr; | |
1311 | ||
1da177e4 LT |
1312 | /* |
1313 | * Get the dir op function pointer. | |
1314 | */ | |
088406bc | 1315 | err = 0; |
1da177e4 LT |
1316 | switch (type) { |
1317 | case S_IFREG: | |
6264d69d | 1318 | host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); |
4ac35c2f | 1319 | if (!host_err) |
1320 | nfsd_check_ignore_resizing(iap); | |
1da177e4 LT |
1321 | break; |
1322 | case S_IFDIR: | |
6264d69d | 1323 | host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); |
1da177e4 LT |
1324 | break; |
1325 | case S_IFCHR: | |
1326 | case S_IFBLK: | |
1327 | case S_IFIFO: | |
1328 | case S_IFSOCK: | |
6264d69d | 1329 | host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); |
1da177e4 | 1330 | break; |
1da177e4 | 1331 | } |
463c3197 DH |
1332 | if (host_err < 0) { |
1333 | mnt_drop_write(fhp->fh_export->ex_path.mnt); | |
1da177e4 | 1334 | goto out_nfserr; |
463c3197 | 1335 | } |
1da177e4 LT |
1336 | |
1337 | if (EX_ISSYNC(fhp->fh_export)) { | |
45bd3b3d | 1338 | err = nfserrno(nfsd_sync_dir(dentry)); |
1da177e4 LT |
1339 | write_inode_now(dchild->d_inode, 1); |
1340 | } | |
1341 | ||
5c002b3b BF |
1342 | err2 = nfsd_create_setattr(rqstp, resfhp, iap); |
1343 | if (err2) | |
1344 | err = err2; | |
463c3197 | 1345 | mnt_drop_write(fhp->fh_export->ex_path.mnt); |
1da177e4 LT |
1346 | /* |
1347 | * Update the file handle to get the new inode info. | |
1348 | */ | |
1349 | if (!err) | |
1350 | err = fh_update(resfhp); | |
1351 | out: | |
1352 | if (dchild && !IS_ERR(dchild)) | |
1353 | dput(dchild); | |
1354 | return err; | |
1355 | ||
1356 | out_nfserr: | |
6264d69d | 1357 | err = nfserrno(host_err); |
1da177e4 LT |
1358 | goto out; |
1359 | } | |
1360 | ||
1361 | #ifdef CONFIG_NFSD_V3 | |
1362 | /* | |
1363 | * NFSv3 version of nfsd_create | |
1364 | */ | |
6264d69d | 1365 | __be32 |
1da177e4 LT |
1366 | nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1367 | char *fname, int flen, struct iattr *iap, | |
1368 | struct svc_fh *resfhp, int createmode, u32 *verifier, | |
81ac95c5 | 1369 | int *truncp, int *created) |
1da177e4 LT |
1370 | { |
1371 | struct dentry *dentry, *dchild = NULL; | |
1372 | struct inode *dirp; | |
6264d69d | 1373 | __be32 err; |
5c002b3b | 1374 | __be32 err2; |
6264d69d | 1375 | int host_err; |
1da177e4 | 1376 | __u32 v_mtime=0, v_atime=0; |
1da177e4 LT |
1377 | |
1378 | err = nfserr_perm; | |
1379 | if (!flen) | |
1380 | goto out; | |
1381 | err = nfserr_exist; | |
1382 | if (isdotent(fname, flen)) | |
1383 | goto out; | |
1384 | if (!(iap->ia_valid & ATTR_MODE)) | |
1385 | iap->ia_mode = 0; | |
8837abca | 1386 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1387 | if (err) |
1388 | goto out; | |
1389 | ||
1390 | dentry = fhp->fh_dentry; | |
1391 | dirp = dentry->d_inode; | |
1392 | ||
1393 | /* Get all the sanity checks out of the way before | |
1394 | * we lock the parent. */ | |
1395 | err = nfserr_notdir; | |
acfa4380 | 1396 | if (!dirp->i_op->lookup) |
1da177e4 | 1397 | goto out; |
12fd3520 | 1398 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1399 | |
1400 | /* | |
1401 | * Compose the response file handle. | |
1402 | */ | |
1403 | dchild = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1404 | host_err = PTR_ERR(dchild); |
1da177e4 LT |
1405 | if (IS_ERR(dchild)) |
1406 | goto out_nfserr; | |
1407 | ||
1408 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); | |
1409 | if (err) | |
1410 | goto out; | |
1411 | ||
1412 | if (createmode == NFS3_CREATE_EXCLUSIVE) { | |
c397852c | 1413 | /* solaris7 gets confused (bugid 4218508) if these have |
749997e5 JL |
1414 | * the high bit set, so just clear the high bits. If this is |
1415 | * ever changed to use different attrs for storing the | |
1416 | * verifier, then do_open_lookup() will also need to be fixed | |
1417 | * accordingly. | |
1da177e4 LT |
1418 | */ |
1419 | v_mtime = verifier[0]&0x7fffffff; | |
1420 | v_atime = verifier[1]&0x7fffffff; | |
1da177e4 LT |
1421 | } |
1422 | ||
463c3197 DH |
1423 | host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); |
1424 | if (host_err) | |
1425 | goto out_nfserr; | |
1da177e4 LT |
1426 | if (dchild->d_inode) { |
1427 | err = 0; | |
1428 | ||
1429 | switch (createmode) { | |
1430 | case NFS3_CREATE_UNCHECKED: | |
1431 | if (! S_ISREG(dchild->d_inode->i_mode)) | |
1432 | err = nfserr_exist; | |
1433 | else if (truncp) { | |
1434 | /* in nfsv4, we need to treat this case a little | |
1435 | * differently. we don't want to truncate the | |
1436 | * file now; this would be wrong if the OPEN | |
1437 | * fails for some other reason. furthermore, | |
1438 | * if the size is nonzero, we should ignore it | |
1439 | * according to spec! | |
1440 | */ | |
1441 | *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; | |
1442 | } | |
1443 | else { | |
1444 | iap->ia_valid &= ATTR_SIZE; | |
1445 | goto set_attr; | |
1446 | } | |
1447 | break; | |
1448 | case NFS3_CREATE_EXCLUSIVE: | |
1449 | if ( dchild->d_inode->i_mtime.tv_sec == v_mtime | |
1450 | && dchild->d_inode->i_atime.tv_sec == v_atime | |
1da177e4 LT |
1451 | && dchild->d_inode->i_size == 0 ) |
1452 | break; | |
1453 | /* fallthru */ | |
1454 | case NFS3_CREATE_GUARDED: | |
1455 | err = nfserr_exist; | |
1456 | } | |
463c3197 | 1457 | mnt_drop_write(fhp->fh_export->ex_path.mnt); |
1da177e4 LT |
1458 | goto out; |
1459 | } | |
1460 | ||
6264d69d | 1461 | host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); |
463c3197 DH |
1462 | if (host_err < 0) { |
1463 | mnt_drop_write(fhp->fh_export->ex_path.mnt); | |
1da177e4 | 1464 | goto out_nfserr; |
463c3197 | 1465 | } |
81ac95c5 BF |
1466 | if (created) |
1467 | *created = 1; | |
1da177e4 LT |
1468 | |
1469 | if (EX_ISSYNC(fhp->fh_export)) { | |
45bd3b3d | 1470 | err = nfserrno(nfsd_sync_dir(dentry)); |
1da177e4 LT |
1471 | /* setattr will sync the child (or not) */ |
1472 | } | |
1473 | ||
4ac35c2f | 1474 | nfsd_check_ignore_resizing(iap); |
1475 | ||
1da177e4 | 1476 | if (createmode == NFS3_CREATE_EXCLUSIVE) { |
c397852c | 1477 | /* Cram the verifier into atime/mtime */ |
1da177e4 | 1478 | iap->ia_valid = ATTR_MTIME|ATTR_ATIME |
c397852c | 1479 | | ATTR_MTIME_SET|ATTR_ATIME_SET; |
1da177e4 LT |
1480 | /* XXX someone who knows this better please fix it for nsec */ |
1481 | iap->ia_mtime.tv_sec = v_mtime; | |
1482 | iap->ia_atime.tv_sec = v_atime; | |
1483 | iap->ia_mtime.tv_nsec = 0; | |
1484 | iap->ia_atime.tv_nsec = 0; | |
1da177e4 LT |
1485 | } |
1486 | ||
1da177e4 | 1487 | set_attr: |
5c002b3b BF |
1488 | err2 = nfsd_create_setattr(rqstp, resfhp, iap); |
1489 | if (err2) | |
1490 | err = err2; | |
f193fbab | 1491 | |
463c3197 | 1492 | mnt_drop_write(fhp->fh_export->ex_path.mnt); |
f193fbab YT |
1493 | /* |
1494 | * Update the filehandle to get the new inode info. | |
1495 | */ | |
1496 | if (!err) | |
1497 | err = fh_update(resfhp); | |
1da177e4 LT |
1498 | |
1499 | out: | |
1500 | fh_unlock(fhp); | |
1501 | if (dchild && !IS_ERR(dchild)) | |
1502 | dput(dchild); | |
1503 | return err; | |
1504 | ||
1505 | out_nfserr: | |
6264d69d | 1506 | err = nfserrno(host_err); |
1da177e4 LT |
1507 | goto out; |
1508 | } | |
1509 | #endif /* CONFIG_NFSD_V3 */ | |
1510 | ||
1511 | /* | |
1512 | * Read a symlink. On entry, *lenp must contain the maximum path length that | |
1513 | * fits into the buffer. On return, it contains the true length. | |
1514 | * N.B. After this call fhp needs an fh_put | |
1515 | */ | |
6264d69d | 1516 | __be32 |
1da177e4 LT |
1517 | nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) |
1518 | { | |
1519 | struct dentry *dentry; | |
1520 | struct inode *inode; | |
1521 | mm_segment_t oldfs; | |
6264d69d AV |
1522 | __be32 err; |
1523 | int host_err; | |
1da177e4 | 1524 | |
8837abca | 1525 | err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); |
1da177e4 LT |
1526 | if (err) |
1527 | goto out; | |
1528 | ||
1529 | dentry = fhp->fh_dentry; | |
1530 | inode = dentry->d_inode; | |
1531 | ||
1532 | err = nfserr_inval; | |
acfa4380 | 1533 | if (!inode->i_op->readlink) |
1da177e4 LT |
1534 | goto out; |
1535 | ||
54775491 | 1536 | touch_atime(fhp->fh_export->ex_path.mnt, dentry); |
1da177e4 LT |
1537 | /* N.B. Why does this call need a get_fs()?? |
1538 | * Remove the set_fs and watch the fireworks:-) --okir | |
1539 | */ | |
1540 | ||
1541 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
6264d69d | 1542 | host_err = inode->i_op->readlink(dentry, buf, *lenp); |
1da177e4 LT |
1543 | set_fs(oldfs); |
1544 | ||
6264d69d | 1545 | if (host_err < 0) |
1da177e4 | 1546 | goto out_nfserr; |
6264d69d | 1547 | *lenp = host_err; |
1da177e4 LT |
1548 | err = 0; |
1549 | out: | |
1550 | return err; | |
1551 | ||
1552 | out_nfserr: | |
6264d69d | 1553 | err = nfserrno(host_err); |
1da177e4 LT |
1554 | goto out; |
1555 | } | |
1556 | ||
1557 | /* | |
1558 | * Create a symlink and look up its inode | |
1559 | * N.B. After this call _both_ fhp and resfhp need an fh_put | |
1560 | */ | |
6264d69d | 1561 | __be32 |
1da177e4 LT |
1562 | nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1563 | char *fname, int flen, | |
1564 | char *path, int plen, | |
1565 | struct svc_fh *resfhp, | |
1566 | struct iattr *iap) | |
1567 | { | |
1568 | struct dentry *dentry, *dnew; | |
6264d69d AV |
1569 | __be32 err, cerr; |
1570 | int host_err; | |
1da177e4 LT |
1571 | |
1572 | err = nfserr_noent; | |
1573 | if (!flen || !plen) | |
1574 | goto out; | |
1575 | err = nfserr_exist; | |
1576 | if (isdotent(fname, flen)) | |
1577 | goto out; | |
1578 | ||
8837abca | 1579 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1580 | if (err) |
1581 | goto out; | |
1582 | fh_lock(fhp); | |
1583 | dentry = fhp->fh_dentry; | |
1584 | dnew = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1585 | host_err = PTR_ERR(dnew); |
1da177e4 LT |
1586 | if (IS_ERR(dnew)) |
1587 | goto out_nfserr; | |
1588 | ||
75c3f29d DH |
1589 | host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); |
1590 | if (host_err) | |
1591 | goto out_nfserr; | |
1592 | ||
1da177e4 LT |
1593 | if (unlikely(path[plen] != 0)) { |
1594 | char *path_alloced = kmalloc(plen+1, GFP_KERNEL); | |
1595 | if (path_alloced == NULL) | |
6264d69d | 1596 | host_err = -ENOMEM; |
1da177e4 LT |
1597 | else { |
1598 | strncpy(path_alloced, path, plen); | |
1599 | path_alloced[plen] = 0; | |
db2e747b | 1600 | host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); |
1da177e4 LT |
1601 | kfree(path_alloced); |
1602 | } | |
1603 | } else | |
db2e747b | 1604 | host_err = vfs_symlink(dentry->d_inode, dnew, path); |
1da177e4 | 1605 | |
6264d69d | 1606 | if (!host_err) { |
1da177e4 | 1607 | if (EX_ISSYNC(fhp->fh_export)) |
6264d69d AV |
1608 | host_err = nfsd_sync_dir(dentry); |
1609 | } | |
1610 | err = nfserrno(host_err); | |
1da177e4 LT |
1611 | fh_unlock(fhp); |
1612 | ||
75c3f29d DH |
1613 | mnt_drop_write(fhp->fh_export->ex_path.mnt); |
1614 | ||
1da177e4 LT |
1615 | cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); |
1616 | dput(dnew); | |
1617 | if (err==0) err = cerr; | |
1618 | out: | |
1619 | return err; | |
1620 | ||
1621 | out_nfserr: | |
6264d69d | 1622 | err = nfserrno(host_err); |
1da177e4 LT |
1623 | goto out; |
1624 | } | |
1625 | ||
1626 | /* | |
1627 | * Create a hardlink | |
1628 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1629 | */ | |
6264d69d | 1630 | __be32 |
1da177e4 LT |
1631 | nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, |
1632 | char *name, int len, struct svc_fh *tfhp) | |
1633 | { | |
1634 | struct dentry *ddir, *dnew, *dold; | |
1635 | struct inode *dirp, *dest; | |
6264d69d AV |
1636 | __be32 err; |
1637 | int host_err; | |
1da177e4 | 1638 | |
8837abca | 1639 | err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1640 | if (err) |
1641 | goto out; | |
8837abca | 1642 | err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP); |
1da177e4 LT |
1643 | if (err) |
1644 | goto out; | |
1645 | ||
1646 | err = nfserr_perm; | |
1647 | if (!len) | |
1648 | goto out; | |
1649 | err = nfserr_exist; | |
1650 | if (isdotent(name, len)) | |
1651 | goto out; | |
1652 | ||
12fd3520 | 1653 | fh_lock_nested(ffhp, I_MUTEX_PARENT); |
1da177e4 LT |
1654 | ddir = ffhp->fh_dentry; |
1655 | dirp = ddir->d_inode; | |
1656 | ||
1657 | dnew = lookup_one_len(name, ddir, len); | |
6264d69d | 1658 | host_err = PTR_ERR(dnew); |
1da177e4 LT |
1659 | if (IS_ERR(dnew)) |
1660 | goto out_nfserr; | |
1661 | ||
1662 | dold = tfhp->fh_dentry; | |
1663 | dest = dold->d_inode; | |
1664 | ||
75c3f29d DH |
1665 | host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt); |
1666 | if (host_err) { | |
1667 | err = nfserrno(host_err); | |
1668 | goto out_dput; | |
1669 | } | |
6264d69d AV |
1670 | host_err = vfs_link(dold, dirp, dnew); |
1671 | if (!host_err) { | |
1da177e4 | 1672 | if (EX_ISSYNC(ffhp->fh_export)) { |
45bd3b3d | 1673 | err = nfserrno(nfsd_sync_dir(ddir)); |
1da177e4 LT |
1674 | write_inode_now(dest, 1); |
1675 | } | |
6264d69d | 1676 | err = 0; |
1da177e4 | 1677 | } else { |
6264d69d | 1678 | if (host_err == -EXDEV && rqstp->rq_vers == 2) |
1da177e4 LT |
1679 | err = nfserr_acces; |
1680 | else | |
6264d69d | 1681 | err = nfserrno(host_err); |
1da177e4 | 1682 | } |
75c3f29d DH |
1683 | mnt_drop_write(tfhp->fh_export->ex_path.mnt); |
1684 | out_dput: | |
1da177e4 | 1685 | dput(dnew); |
270d56e5 DR |
1686 | out_unlock: |
1687 | fh_unlock(ffhp); | |
1da177e4 LT |
1688 | out: |
1689 | return err; | |
1690 | ||
1691 | out_nfserr: | |
6264d69d | 1692 | err = nfserrno(host_err); |
270d56e5 | 1693 | goto out_unlock; |
1da177e4 LT |
1694 | } |
1695 | ||
1696 | /* | |
1697 | * Rename a file | |
1698 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1699 | */ | |
6264d69d | 1700 | __be32 |
1da177e4 LT |
1701 | nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, |
1702 | struct svc_fh *tfhp, char *tname, int tlen) | |
1703 | { | |
1704 | struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; | |
1705 | struct inode *fdir, *tdir; | |
6264d69d AV |
1706 | __be32 err; |
1707 | int host_err; | |
1da177e4 | 1708 | |
8837abca | 1709 | err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); |
1da177e4 LT |
1710 | if (err) |
1711 | goto out; | |
8837abca | 1712 | err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1713 | if (err) |
1714 | goto out; | |
1715 | ||
1716 | fdentry = ffhp->fh_dentry; | |
1717 | fdir = fdentry->d_inode; | |
1718 | ||
1719 | tdentry = tfhp->fh_dentry; | |
1720 | tdir = tdentry->d_inode; | |
1721 | ||
1722 | err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; | |
a56f3937 | 1723 | if (ffhp->fh_export != tfhp->fh_export) |
1da177e4 LT |
1724 | goto out; |
1725 | ||
1726 | err = nfserr_perm; | |
1727 | if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) | |
1728 | goto out; | |
1729 | ||
1730 | /* cannot use fh_lock as we need deadlock protective ordering | |
1731 | * so do it by hand */ | |
1732 | trap = lock_rename(tdentry, fdentry); | |
1733 | ffhp->fh_locked = tfhp->fh_locked = 1; | |
1734 | fill_pre_wcc(ffhp); | |
1735 | fill_pre_wcc(tfhp); | |
1736 | ||
1737 | odentry = lookup_one_len(fname, fdentry, flen); | |
6264d69d | 1738 | host_err = PTR_ERR(odentry); |
1da177e4 LT |
1739 | if (IS_ERR(odentry)) |
1740 | goto out_nfserr; | |
1741 | ||
6264d69d | 1742 | host_err = -ENOENT; |
1da177e4 LT |
1743 | if (!odentry->d_inode) |
1744 | goto out_dput_old; | |
6264d69d | 1745 | host_err = -EINVAL; |
1da177e4 LT |
1746 | if (odentry == trap) |
1747 | goto out_dput_old; | |
1748 | ||
1749 | ndentry = lookup_one_len(tname, tdentry, tlen); | |
6264d69d | 1750 | host_err = PTR_ERR(ndentry); |
1da177e4 LT |
1751 | if (IS_ERR(ndentry)) |
1752 | goto out_dput_old; | |
6264d69d | 1753 | host_err = -ENOTEMPTY; |
1da177e4 LT |
1754 | if (ndentry == trap) |
1755 | goto out_dput_new; | |
1756 | ||
9079b1eb | 1757 | if (svc_msnfs(ffhp) && |
1da177e4 LT |
1758 | ((atomic_read(&odentry->d_count) > 1) |
1759 | || (atomic_read(&ndentry->d_count) > 1))) { | |
6264d69d | 1760 | host_err = -EPERM; |
9079b1eb DH |
1761 | goto out_dput_new; |
1762 | } | |
1763 | ||
1764 | host_err = -EXDEV; | |
1765 | if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) | |
1766 | goto out_dput_new; | |
1767 | host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt); | |
1768 | if (host_err) | |
1769 | goto out_dput_new; | |
1770 | ||
6264d69d AV |
1771 | host_err = vfs_rename(fdir, odentry, tdir, ndentry); |
1772 | if (!host_err && EX_ISSYNC(tfhp->fh_export)) { | |
1773 | host_err = nfsd_sync_dir(tdentry); | |
1774 | if (!host_err) | |
1775 | host_err = nfsd_sync_dir(fdentry); | |
1da177e4 LT |
1776 | } |
1777 | ||
9079b1eb DH |
1778 | mnt_drop_write(ffhp->fh_export->ex_path.mnt); |
1779 | ||
1da177e4 LT |
1780 | out_dput_new: |
1781 | dput(ndentry); | |
1782 | out_dput_old: | |
1783 | dput(odentry); | |
1784 | out_nfserr: | |
6264d69d | 1785 | err = nfserrno(host_err); |
1da177e4 LT |
1786 | |
1787 | /* we cannot reply on fh_unlock on the two filehandles, | |
1788 | * as that would do the wrong thing if the two directories | |
1789 | * were the same, so again we do it by hand | |
1790 | */ | |
1791 | fill_post_wcc(ffhp); | |
1792 | fill_post_wcc(tfhp); | |
1793 | unlock_rename(tdentry, fdentry); | |
1794 | ffhp->fh_locked = tfhp->fh_locked = 0; | |
1795 | ||
1796 | out: | |
1797 | return err; | |
1798 | } | |
1799 | ||
1800 | /* | |
1801 | * Unlink a file or directory | |
1802 | * N.B. After this call fhp needs an fh_put | |
1803 | */ | |
6264d69d | 1804 | __be32 |
1da177e4 LT |
1805 | nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, |
1806 | char *fname, int flen) | |
1807 | { | |
1808 | struct dentry *dentry, *rdentry; | |
1809 | struct inode *dirp; | |
6264d69d AV |
1810 | __be32 err; |
1811 | int host_err; | |
1da177e4 LT |
1812 | |
1813 | err = nfserr_acces; | |
1814 | if (!flen || isdotent(fname, flen)) | |
1815 | goto out; | |
8837abca | 1816 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); |
1da177e4 LT |
1817 | if (err) |
1818 | goto out; | |
1819 | ||
12fd3520 | 1820 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1821 | dentry = fhp->fh_dentry; |
1822 | dirp = dentry->d_inode; | |
1823 | ||
1824 | rdentry = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1825 | host_err = PTR_ERR(rdentry); |
1da177e4 LT |
1826 | if (IS_ERR(rdentry)) |
1827 | goto out_nfserr; | |
1828 | ||
1829 | if (!rdentry->d_inode) { | |
1830 | dput(rdentry); | |
1831 | err = nfserr_noent; | |
1832 | goto out; | |
1833 | } | |
1834 | ||
1835 | if (!type) | |
1836 | type = rdentry->d_inode->i_mode & S_IFMT; | |
1837 | ||
0622753b DH |
1838 | host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); |
1839 | if (host_err) | |
1840 | goto out_nfserr; | |
1841 | ||
1da177e4 LT |
1842 | if (type != S_IFDIR) { /* It's UNLINK */ |
1843 | #ifdef MSNFS | |
1844 | if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && | |
1845 | (atomic_read(&rdentry->d_count) > 1)) { | |
6264d69d | 1846 | host_err = -EPERM; |
1da177e4 LT |
1847 | } else |
1848 | #endif | |
6264d69d | 1849 | host_err = vfs_unlink(dirp, rdentry); |
1da177e4 | 1850 | } else { /* It's RMDIR */ |
6264d69d | 1851 | host_err = vfs_rmdir(dirp, rdentry); |
1da177e4 LT |
1852 | } |
1853 | ||
1854 | dput(rdentry); | |
1855 | ||
6264d69d | 1856 | if (host_err) |
0622753b | 1857 | goto out_drop; |
6264d69d AV |
1858 | if (EX_ISSYNC(fhp->fh_export)) |
1859 | host_err = nfsd_sync_dir(dentry); | |
1da177e4 | 1860 | |
0622753b DH |
1861 | out_drop: |
1862 | mnt_drop_write(fhp->fh_export->ex_path.mnt); | |
1da177e4 | 1863 | out_nfserr: |
6264d69d | 1864 | err = nfserrno(host_err); |
f193fbab YT |
1865 | out: |
1866 | return err; | |
1da177e4 LT |
1867 | } |
1868 | ||
14f7dd63 DW |
1869 | /* |
1870 | * We do this buffering because we must not call back into the file | |
1871 | * system's ->lookup() method from the filldir callback. That may well | |
1872 | * deadlock a number of file systems. | |
1873 | * | |
1874 | * This is based heavily on the implementation of same in XFS. | |
1875 | */ | |
1876 | struct buffered_dirent { | |
1877 | u64 ino; | |
1878 | loff_t offset; | |
1879 | int namlen; | |
1880 | unsigned int d_type; | |
1881 | char name[]; | |
1882 | }; | |
1883 | ||
1884 | struct readdir_data { | |
1885 | char *dirent; | |
1886 | size_t used; | |
53c9c5c0 | 1887 | int full; |
14f7dd63 DW |
1888 | }; |
1889 | ||
1890 | static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, | |
1891 | loff_t offset, u64 ino, unsigned int d_type) | |
1892 | { | |
1893 | struct readdir_data *buf = __buf; | |
1894 | struct buffered_dirent *de = (void *)(buf->dirent + buf->used); | |
1895 | unsigned int reclen; | |
1896 | ||
1897 | reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); | |
53c9c5c0 AV |
1898 | if (buf->used + reclen > PAGE_SIZE) { |
1899 | buf->full = 1; | |
14f7dd63 | 1900 | return -EINVAL; |
53c9c5c0 | 1901 | } |
14f7dd63 DW |
1902 | |
1903 | de->namlen = namlen; | |
1904 | de->offset = offset; | |
1905 | de->ino = ino; | |
1906 | de->d_type = d_type; | |
1907 | memcpy(de->name, name, namlen); | |
1908 | buf->used += reclen; | |
1909 | ||
1910 | return 0; | |
1911 | } | |
1912 | ||
2f9092e1 DW |
1913 | static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, |
1914 | struct readdir_cd *cdp, loff_t *offsetp) | |
2628b766 | 1915 | { |
14f7dd63 DW |
1916 | struct readdir_data buf; |
1917 | struct buffered_dirent *de; | |
2628b766 | 1918 | int host_err; |
14f7dd63 DW |
1919 | int size; |
1920 | loff_t offset; | |
2628b766 | 1921 | |
14f7dd63 DW |
1922 | buf.dirent = (void *)__get_free_page(GFP_KERNEL); |
1923 | if (!buf.dirent) | |
2f9092e1 | 1924 | return nfserrno(-ENOMEM); |
14f7dd63 DW |
1925 | |
1926 | offset = *offsetp; | |
2628b766 | 1927 | |
14f7dd63 | 1928 | while (1) { |
2f9092e1 | 1929 | struct inode *dir_inode = file->f_path.dentry->d_inode; |
14f7dd63 DW |
1930 | unsigned int reclen; |
1931 | ||
b726e923 | 1932 | cdp->err = nfserr_eof; /* will be cleared on successful read */ |
14f7dd63 | 1933 | buf.used = 0; |
53c9c5c0 | 1934 | buf.full = 0; |
14f7dd63 DW |
1935 | |
1936 | host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf); | |
53c9c5c0 AV |
1937 | if (buf.full) |
1938 | host_err = 0; | |
1939 | ||
1940 | if (host_err < 0) | |
14f7dd63 DW |
1941 | break; |
1942 | ||
1943 | size = buf.used; | |
1944 | ||
1945 | if (!size) | |
1946 | break; | |
1947 | ||
2f9092e1 DW |
1948 | /* |
1949 | * Various filldir functions may end up calling back into | |
1950 | * lookup_one_len() and the file system's ->lookup() method. | |
1951 | * These expect i_mutex to be held, as it would within readdir. | |
1952 | */ | |
1953 | host_err = mutex_lock_killable(&dir_inode->i_mutex); | |
1954 | if (host_err) | |
1955 | break; | |
1956 | ||
14f7dd63 DW |
1957 | de = (struct buffered_dirent *)buf.dirent; |
1958 | while (size > 0) { | |
1959 | offset = de->offset; | |
1960 | ||
1961 | if (func(cdp, de->name, de->namlen, de->offset, | |
1962 | de->ino, de->d_type)) | |
2f9092e1 | 1963 | break; |
14f7dd63 DW |
1964 | |
1965 | if (cdp->err != nfs_ok) | |
2f9092e1 | 1966 | break; |
14f7dd63 DW |
1967 | |
1968 | reclen = ALIGN(sizeof(*de) + de->namlen, | |
1969 | sizeof(u64)); | |
1970 | size -= reclen; | |
1971 | de = (struct buffered_dirent *)((char *)de + reclen); | |
1972 | } | |
2f9092e1 DW |
1973 | mutex_unlock(&dir_inode->i_mutex); |
1974 | if (size > 0) /* We bailed out early */ | |
1975 | break; | |
1976 | ||
c002a6c7 | 1977 | offset = vfs_llseek(file, 0, SEEK_CUR); |
14f7dd63 DW |
1978 | } |
1979 | ||
14f7dd63 | 1980 | free_page((unsigned long)(buf.dirent)); |
2628b766 DW |
1981 | |
1982 | if (host_err) | |
1983 | return nfserrno(host_err); | |
14f7dd63 DW |
1984 | |
1985 | *offsetp = offset; | |
1986 | return cdp->err; | |
2628b766 DW |
1987 | } |
1988 | ||
1da177e4 LT |
1989 | /* |
1990 | * Read entries from a directory. | |
1991 | * The NFSv3/4 verifier we ignore for now. | |
1992 | */ | |
6264d69d | 1993 | __be32 |
1da177e4 | 1994 | nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, |
a0ad13ef | 1995 | struct readdir_cd *cdp, filldir_t func) |
1da177e4 | 1996 | { |
6264d69d | 1997 | __be32 err; |
1da177e4 LT |
1998 | struct file *file; |
1999 | loff_t offset = *offsetp; | |
2000 | ||
8837abca | 2001 | err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file); |
1da177e4 LT |
2002 | if (err) |
2003 | goto out; | |
2004 | ||
2005 | offset = vfs_llseek(file, offset, 0); | |
2006 | if (offset < 0) { | |
2007 | err = nfserrno((int)offset); | |
2008 | goto out_close; | |
2009 | } | |
2010 | ||
14f7dd63 | 2011 | err = nfsd_buffered_readdir(file, func, cdp, offsetp); |
1da177e4 LT |
2012 | |
2013 | if (err == nfserr_eof || err == nfserr_toosmall) | |
2014 | err = nfs_ok; /* can still be found in ->err */ | |
2015 | out_close: | |
2016 | nfsd_close(file); | |
2017 | out: | |
2018 | return err; | |
2019 | } | |
2020 | ||
2021 | /* | |
2022 | * Get file system stats | |
2023 | * N.B. After this call fhp needs an fh_put | |
2024 | */ | |
6264d69d | 2025 | __be32 |
04716e66 | 2026 | nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) |
1da177e4 | 2027 | { |
04716e66 | 2028 | __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); |
726c3342 | 2029 | if (!err && vfs_statfs(fhp->fh_dentry,stat)) |
1da177e4 LT |
2030 | err = nfserr_io; |
2031 | return err; | |
2032 | } | |
2033 | ||
c7d51402 | 2034 | static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) |
e22841c6 | 2035 | { |
c7d51402 | 2036 | return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; |
e22841c6 BF |
2037 | } |
2038 | ||
1da177e4 LT |
2039 | /* |
2040 | * Check for a user's access permissions to this inode. | |
2041 | */ | |
6264d69d | 2042 | __be32 |
0ec757df BF |
2043 | nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, |
2044 | struct dentry *dentry, int acc) | |
1da177e4 LT |
2045 | { |
2046 | struct inode *inode = dentry->d_inode; | |
14dba533 | 2047 | struct path path; |
1da177e4 LT |
2048 | int err; |
2049 | ||
8837abca | 2050 | if (acc == NFSD_MAY_NOP) |
1da177e4 LT |
2051 | return 0; |
2052 | #if 0 | |
2053 | dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", | |
2054 | acc, | |
8837abca MS |
2055 | (acc & NFSD_MAY_READ)? " read" : "", |
2056 | (acc & NFSD_MAY_WRITE)? " write" : "", | |
2057 | (acc & NFSD_MAY_EXEC)? " exec" : "", | |
2058 | (acc & NFSD_MAY_SATTR)? " sattr" : "", | |
2059 | (acc & NFSD_MAY_TRUNC)? " trunc" : "", | |
2060 | (acc & NFSD_MAY_LOCK)? " lock" : "", | |
2061 | (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", | |
1da177e4 LT |
2062 | inode->i_mode, |
2063 | IS_IMMUTABLE(inode)? " immut" : "", | |
2064 | IS_APPEND(inode)? " append" : "", | |
2c463e95 | 2065 | __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); |
1da177e4 | 2066 | dprintk(" owner %d/%d user %d/%d\n", |
5cc0a840 | 2067 | inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); |
1da177e4 LT |
2068 | #endif |
2069 | ||
2070 | /* Normally we reject any write/sattr etc access on a read-only file | |
2071 | * system. But if it is IRIX doing check on write-access for a | |
2072 | * device special file, we ignore rofs. | |
2073 | */ | |
8837abca MS |
2074 | if (!(acc & NFSD_MAY_LOCAL_ACCESS)) |
2075 | if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { | |
2c463e95 DH |
2076 | if (exp_rdonly(rqstp, exp) || |
2077 | __mnt_is_readonly(exp->ex_path.mnt)) | |
1da177e4 | 2078 | return nfserr_rofs; |
8837abca | 2079 | if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) |
1da177e4 LT |
2080 | return nfserr_perm; |
2081 | } | |
8837abca | 2082 | if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) |
1da177e4 LT |
2083 | return nfserr_perm; |
2084 | ||
8837abca | 2085 | if (acc & NFSD_MAY_LOCK) { |
1da177e4 LT |
2086 | /* If we cannot rely on authentication in NLM requests, |
2087 | * just allow locks, otherwise require read permission, or | |
2088 | * ownership | |
2089 | */ | |
2090 | if (exp->ex_flags & NFSEXP_NOAUTHNLM) | |
2091 | return 0; | |
2092 | else | |
8837abca | 2093 | acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; |
1da177e4 LT |
2094 | } |
2095 | /* | |
2096 | * The file owner always gets access permission for accesses that | |
2097 | * would normally be checked at open time. This is to make | |
2098 | * file access work even when the client has done a fchmod(fd, 0). | |
2099 | * | |
2100 | * However, `cp foo bar' should fail nevertheless when bar is | |
2101 | * readonly. A sensible way to do this might be to reject all | |
2102 | * attempts to truncate a read-only file, because a creat() call | |
2103 | * always implies file truncation. | |
2104 | * ... but this isn't really fair. A process may reasonably call | |
2105 | * ftruncate on an open file descriptor on a file with perm 000. | |
2106 | * We must trust the client to do permission checking - using "ACCESS" | |
2107 | * with NFSv3. | |
2108 | */ | |
8837abca | 2109 | if ((acc & NFSD_MAY_OWNER_OVERRIDE) && |
5cc0a840 | 2110 | inode->i_uid == current_fsuid()) |
1da177e4 LT |
2111 | return 0; |
2112 | ||
8837abca | 2113 | /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ |
f419a2e3 | 2114 | err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); |
1da177e4 LT |
2115 | |
2116 | /* Allow read access to binaries even when mode 111 */ | |
2117 | if (err == -EACCES && S_ISREG(inode->i_mode) && | |
8837abca | 2118 | acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE)) |
f419a2e3 | 2119 | err = inode_permission(inode, MAY_EXEC); |
14dba533 MZ |
2120 | if (err) |
2121 | goto nfsd_out; | |
1da177e4 | 2122 | |
14dba533 MZ |
2123 | /* Do integrity (permission) checking now, but defer incrementing |
2124 | * IMA counts to the actual file open. | |
2125 | */ | |
2126 | path.mnt = exp->ex_path.mnt; | |
2127 | path.dentry = dentry; | |
14dba533 | 2128 | nfsd_out: |
1da177e4 LT |
2129 | return err? nfserrno(err) : 0; |
2130 | } | |
2131 | ||
2132 | void | |
2133 | nfsd_racache_shutdown(void) | |
2134 | { | |
54a66e54 JL |
2135 | struct raparms *raparm, *last_raparm; |
2136 | unsigned int i; | |
2137 | ||
1da177e4 | 2138 | dprintk("nfsd: freeing readahead buffers.\n"); |
54a66e54 JL |
2139 | |
2140 | for (i = 0; i < RAPARM_HASH_SIZE; i++) { | |
2141 | raparm = raparm_hash[i].pb_head; | |
2142 | while(raparm) { | |
2143 | last_raparm = raparm; | |
2144 | raparm = raparm->p_next; | |
2145 | kfree(last_raparm); | |
2146 | } | |
2147 | raparm_hash[i].pb_head = NULL; | |
2148 | } | |
1da177e4 LT |
2149 | } |
2150 | /* | |
2151 | * Initialize readahead param cache | |
2152 | */ | |
2153 | int | |
2154 | nfsd_racache_init(int cache_size) | |
2155 | { | |
2156 | int i; | |
fce1456a GB |
2157 | int j = 0; |
2158 | int nperbucket; | |
54a66e54 | 2159 | struct raparms **raparm = NULL; |
1da177e4 | 2160 | |
fce1456a | 2161 | |
54a66e54 | 2162 | if (raparm_hash[0].pb_head) |
1da177e4 | 2163 | return 0; |
54a66e54 JL |
2164 | nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); |
2165 | if (nperbucket < 2) | |
2166 | nperbucket = 2; | |
2167 | cache_size = nperbucket * RAPARM_HASH_SIZE; | |
4b3bb06b YB |
2168 | |
2169 | dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); | |
54a66e54 JL |
2170 | |
2171 | for (i = 0; i < RAPARM_HASH_SIZE; i++) { | |
4b3bb06b | 2172 | spin_lock_init(&raparm_hash[i].pb_lock); |
54a66e54 JL |
2173 | |
2174 | raparm = &raparm_hash[i].pb_head; | |
2175 | for (j = 0; j < nperbucket; j++) { | |
2176 | *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); | |
2177 | if (!*raparm) | |
2178 | goto out_nomem; | |
2179 | raparm = &(*raparm)->p_next; | |
2180 | } | |
2181 | *raparm = NULL; | |
4b3bb06b YB |
2182 | } |
2183 | ||
1da177e4 LT |
2184 | nfsdstats.ra_size = cache_size; |
2185 | return 0; | |
54a66e54 JL |
2186 | |
2187 | out_nomem: | |
2188 | dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); | |
2189 | nfsd_racache_shutdown(); | |
2190 | return -ENOMEM; | |
1da177e4 | 2191 | } |
a257cdd0 AG |
2192 | |
2193 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) | |
2194 | struct posix_acl * | |
2195 | nfsd_get_posix_acl(struct svc_fh *fhp, int type) | |
2196 | { | |
2197 | struct inode *inode = fhp->fh_dentry->d_inode; | |
2198 | char *name; | |
2199 | void *value = NULL; | |
2200 | ssize_t size; | |
2201 | struct posix_acl *acl; | |
2202 | ||
5be196e5 CH |
2203 | if (!IS_POSIXACL(inode)) |
2204 | return ERR_PTR(-EOPNOTSUPP); | |
2205 | ||
2206 | switch (type) { | |
2207 | case ACL_TYPE_ACCESS: | |
2208 | name = POSIX_ACL_XATTR_ACCESS; | |
2209 | break; | |
2210 | case ACL_TYPE_DEFAULT: | |
2211 | name = POSIX_ACL_XATTR_DEFAULT; | |
2212 | break; | |
2213 | default: | |
a257cdd0 | 2214 | return ERR_PTR(-EOPNOTSUPP); |
a257cdd0 AG |
2215 | } |
2216 | ||
5be196e5 CH |
2217 | size = nfsd_getxattr(fhp->fh_dentry, name, &value); |
2218 | if (size < 0) | |
2219 | return ERR_PTR(size); | |
a257cdd0 | 2220 | |
a257cdd0 | 2221 | acl = posix_acl_from_xattr(value, size); |
a257cdd0 AG |
2222 | kfree(value); |
2223 | return acl; | |
2224 | } | |
2225 | ||
2226 | int | |
2227 | nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) | |
2228 | { | |
2229 | struct inode *inode = fhp->fh_dentry->d_inode; | |
2230 | char *name; | |
2231 | void *value = NULL; | |
2232 | size_t size; | |
2233 | int error; | |
2234 | ||
acfa4380 | 2235 | if (!IS_POSIXACL(inode) || |
a257cdd0 AG |
2236 | !inode->i_op->setxattr || !inode->i_op->removexattr) |
2237 | return -EOPNOTSUPP; | |
2238 | switch(type) { | |
2239 | case ACL_TYPE_ACCESS: | |
334a13ec | 2240 | name = POSIX_ACL_XATTR_ACCESS; |
a257cdd0 AG |
2241 | break; |
2242 | case ACL_TYPE_DEFAULT: | |
334a13ec | 2243 | name = POSIX_ACL_XATTR_DEFAULT; |
a257cdd0 AG |
2244 | break; |
2245 | default: | |
2246 | return -EOPNOTSUPP; | |
2247 | } | |
2248 | ||
2249 | if (acl && acl->a_count) { | |
334a13ec | 2250 | size = posix_acl_xattr_size(acl->a_count); |
a257cdd0 AG |
2251 | value = kmalloc(size, GFP_KERNEL); |
2252 | if (!value) | |
2253 | return -ENOMEM; | |
9ccfc29c FM |
2254 | error = posix_acl_to_xattr(acl, value, size); |
2255 | if (error < 0) | |
a257cdd0 | 2256 | goto getout; |
9ccfc29c | 2257 | size = error; |
a257cdd0 AG |
2258 | } else |
2259 | size = 0; | |
2260 | ||
18f335af DH |
2261 | error = mnt_want_write(fhp->fh_export->ex_path.mnt); |
2262 | if (error) | |
2263 | goto getout; | |
a257cdd0 | 2264 | if (size) |
5be196e5 | 2265 | error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); |
a257cdd0 AG |
2266 | else { |
2267 | if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) | |
2268 | error = 0; | |
2269 | else { | |
5be196e5 | 2270 | error = vfs_removexattr(fhp->fh_dentry, name); |
a257cdd0 AG |
2271 | if (error == -ENODATA) |
2272 | error = 0; | |
2273 | } | |
2274 | } | |
18f335af | 2275 | mnt_drop_write(fhp->fh_export->ex_path.mnt); |
a257cdd0 AG |
2276 | |
2277 | getout: | |
2278 | kfree(value); | |
2279 | return error; | |
2280 | } | |
2281 | #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ |