Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * File operations used by nfsd. Some of these have been ripped from |
3 | * other parts of the kernel because they weren't exported, others | |
4 | * are partial duplicates with added or changed functionality. | |
5 | * | |
6 | * Note that several functions dget() the dentry upon which they want | |
7 | * to act, most notably those that create directory entries. Response | |
8 | * dentry's are dput()'d if necessary in the release callback. | |
9 | * So if you notice code paths that apparently fail to dput() the | |
10 | * dentry, don't worry--they have been taken care of. | |
11 | * | |
12 | * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de> | |
13 | * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp> | |
14 | */ | |
15 | ||
1da177e4 LT |
16 | #include <linux/fs.h> |
17 | #include <linux/file.h> | |
d6b29d7c | 18 | #include <linux/splice.h> |
1da177e4 | 19 | #include <linux/fcntl.h> |
1da177e4 | 20 | #include <linux/namei.h> |
1da177e4 | 21 | #include <linux/delay.h> |
0eeca283 | 22 | #include <linux/fsnotify.h> |
1da177e4 | 23 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 24 | #include <linux/xattr.h> |
9a74af21 BH |
25 | #include <linux/jhash.h> |
26 | #include <linux/ima.h> | |
5a0e3ad6 | 27 | #include <linux/slab.h> |
9a74af21 | 28 | #include <asm/uaccess.h> |
f501912a BM |
29 | #include <linux/exportfs.h> |
30 | #include <linux/writeback.h> | |
18032ca0 | 31 | #include <linux/security.h> |
9a74af21 BH |
32 | |
33 | #ifdef CONFIG_NFSD_V3 | |
34 | #include "xdr3.h" | |
35 | #endif /* CONFIG_NFSD_V3 */ | |
36 | ||
5be196e5 | 37 | #ifdef CONFIG_NFSD_V4 |
2ca72e17 BF |
38 | #include "acl.h" |
39 | #include "idmap.h" | |
1da177e4 LT |
40 | #endif /* CONFIG_NFSD_V4 */ |
41 | ||
9a74af21 BH |
42 | #include "nfsd.h" |
43 | #include "vfs.h" | |
1da177e4 LT |
44 | |
45 | #define NFSDDBG_FACILITY NFSDDBG_FILEOP | |
1da177e4 LT |
46 | |
47 | ||
1da177e4 LT |
48 | /* |
49 | * This is a cache of readahead params that help us choose the proper | |
50 | * readahead strategy. Initially, we set all readahead parameters to 0 | |
51 | * and let the VFS handle things. | |
52 | * If you increase the number of cached files very much, you'll need to | |
53 | * add a hash table here. | |
54 | */ | |
55 | struct raparms { | |
56 | struct raparms *p_next; | |
57 | unsigned int p_count; | |
58 | ino_t p_ino; | |
59 | dev_t p_dev; | |
60 | int p_set; | |
61 | struct file_ra_state p_ra; | |
fce1456a | 62 | unsigned int p_hindex; |
1da177e4 LT |
63 | }; |
64 | ||
fce1456a GB |
65 | struct raparm_hbucket { |
66 | struct raparms *pb_head; | |
67 | spinlock_t pb_lock; | |
68 | } ____cacheline_aligned_in_smp; | |
69 | ||
fce1456a GB |
70 | #define RAPARM_HASH_BITS 4 |
71 | #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) | |
72 | #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) | |
73 | static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; | |
1da177e4 LT |
74 | |
75 | /* | |
76 | * Called from nfsd_lookup and encode_dirent. Check if we have crossed | |
77 | * a mount point. | |
e0bb89ef | 78 | * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, |
1da177e4 LT |
79 | * or nfs_ok having possibly changed *dpp and *expp |
80 | */ | |
81 | int | |
82 | nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, | |
83 | struct svc_export **expp) | |
84 | { | |
85 | struct svc_export *exp = *expp, *exp2 = NULL; | |
86 | struct dentry *dentry = *dpp; | |
91c9fa8f AV |
87 | struct path path = {.mnt = mntget(exp->ex_path.mnt), |
88 | .dentry = dget(dentry)}; | |
6264d69d | 89 | int err = 0; |
1da177e4 | 90 | |
7cc90cc3 | 91 | err = follow_down(&path); |
cc53ce53 DH |
92 | if (err < 0) |
93 | goto out; | |
1da177e4 | 94 | |
91c9fa8f | 95 | exp2 = rqst_exp_get_by_name(rqstp, &path); |
1da177e4 | 96 | if (IS_ERR(exp2)) { |
3b6cee7b BF |
97 | err = PTR_ERR(exp2); |
98 | /* | |
99 | * We normally allow NFS clients to continue | |
100 | * "underneath" a mountpoint that is not exported. | |
101 | * The exception is V4ROOT, where no traversal is ever | |
102 | * allowed without an explicit export of the new | |
103 | * directory. | |
104 | */ | |
105 | if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT)) | |
106 | err = 0; | |
91c9fa8f | 107 | path_put(&path); |
1da177e4 LT |
108 | goto out; |
109 | } | |
3c394dda SD |
110 | if (nfsd_v4client(rqstp) || |
111 | (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { | |
1da177e4 | 112 | /* successfully crossed mount point */ |
1644ccc8 | 113 | /* |
91c9fa8f AV |
114 | * This is subtle: path.dentry is *not* on path.mnt |
115 | * at this point. The only reason we are safe is that | |
116 | * original mnt is pinned down by exp, so we should | |
117 | * put path *before* putting exp | |
1644ccc8 | 118 | */ |
91c9fa8f AV |
119 | *dpp = path.dentry; |
120 | path.dentry = dentry; | |
1644ccc8 | 121 | *expp = exp2; |
91c9fa8f | 122 | exp2 = exp; |
1da177e4 | 123 | } |
91c9fa8f AV |
124 | path_put(&path); |
125 | exp_put(exp2); | |
1da177e4 LT |
126 | out: |
127 | return err; | |
128 | } | |
129 | ||
289ede45 BF |
130 | static void follow_to_parent(struct path *path) |
131 | { | |
132 | struct dentry *dp; | |
133 | ||
134 | while (path->dentry == path->mnt->mnt_root && follow_up(path)) | |
135 | ; | |
136 | dp = dget_parent(path->dentry); | |
137 | dput(path->dentry); | |
138 | path->dentry = dp; | |
139 | } | |
140 | ||
141 | static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp) | |
142 | { | |
143 | struct svc_export *exp2; | |
144 | struct path path = {.mnt = mntget((*exp)->ex_path.mnt), | |
145 | .dentry = dget(dparent)}; | |
146 | ||
147 | follow_to_parent(&path); | |
148 | ||
149 | exp2 = rqst_exp_parent(rqstp, &path); | |
150 | if (PTR_ERR(exp2) == -ENOENT) { | |
151 | *dentryp = dget(dparent); | |
152 | } else if (IS_ERR(exp2)) { | |
153 | path_put(&path); | |
154 | return PTR_ERR(exp2); | |
155 | } else { | |
156 | *dentryp = dget(path.dentry); | |
157 | exp_put(*exp); | |
158 | *exp = exp2; | |
159 | } | |
160 | path_put(&path); | |
161 | return 0; | |
162 | } | |
163 | ||
82ead7fe BF |
164 | /* |
165 | * For nfsd purposes, we treat V4ROOT exports as though there was an | |
166 | * export at *every* directory. | |
167 | */ | |
3227fa41 | 168 | int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp) |
82ead7fe BF |
169 | { |
170 | if (d_mountpoint(dentry)) | |
171 | return 1; | |
11fcee02 TM |
172 | if (nfsd4_is_junction(dentry)) |
173 | return 1; | |
82ead7fe BF |
174 | if (!(exp->ex_flags & NFSEXP_V4ROOT)) |
175 | return 0; | |
176 | return dentry->d_inode != NULL; | |
177 | } | |
178 | ||
6264d69d | 179 | __be32 |
6c0a654d | 180 | nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp, |
5a022fc8 | 181 | const char *name, unsigned int len, |
6c0a654d | 182 | struct svc_export **exp_ret, struct dentry **dentry_ret) |
1da177e4 LT |
183 | { |
184 | struct svc_export *exp; | |
185 | struct dentry *dparent; | |
186 | struct dentry *dentry; | |
6264d69d | 187 | int host_err; |
1da177e4 LT |
188 | |
189 | dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); | |
190 | ||
1da177e4 LT |
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 | ||
29a78a3e BF |
254 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); |
255 | if (err) | |
256 | return err; | |
6c0a654d BF |
257 | err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); |
258 | if (err) | |
259 | return err; | |
32c1eb0c AA |
260 | err = check_nfsd_access(exp, rqstp); |
261 | if (err) | |
262 | goto out; | |
1da177e4 LT |
263 | /* |
264 | * Note: we compose the file handle now, but as the | |
265 | * dentry may be negative, it may need to be updated. | |
266 | */ | |
267 | err = fh_compose(resfh, exp, dentry, fhp); | |
268 | if (!err && !dentry->d_inode) | |
269 | err = nfserr_noent; | |
32c1eb0c | 270 | out: |
1da177e4 | 271 | dput(dentry); |
1da177e4 LT |
272 | exp_put(exp); |
273 | return err; | |
1da177e4 LT |
274 | } |
275 | ||
4795bb37 BF |
276 | static int nfsd_break_lease(struct inode *inode) |
277 | { | |
278 | if (!S_ISREG(inode->i_mode)) | |
279 | return 0; | |
280 | return break_lease(inode, O_WRONLY | O_NONBLOCK); | |
281 | } | |
282 | ||
f501912a BM |
283 | /* |
284 | * Commit metadata changes to stable storage. | |
285 | */ | |
286 | static int | |
287 | commit_metadata(struct svc_fh *fhp) | |
288 | { | |
289 | struct inode *inode = fhp->fh_dentry->d_inode; | |
290 | const struct export_operations *export_ops = inode->i_sb->s_export_op; | |
f501912a BM |
291 | |
292 | if (!EX_ISSYNC(fhp->fh_export)) | |
293 | return 0; | |
294 | ||
c3765016 CH |
295 | if (export_ops->commit_metadata) |
296 | return export_ops->commit_metadata(inode); | |
297 | return sync_inode_metadata(inode, 1); | |
f501912a | 298 | } |
6c0a654d | 299 | |
1da177e4 LT |
300 | /* |
301 | * Set various file attributes. | |
302 | * N.B. After this call fhp needs an fh_put | |
303 | */ | |
6264d69d | 304 | __be32 |
1da177e4 LT |
305 | nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, |
306 | int check_guard, time_t guardtime) | |
307 | { | |
308 | struct dentry *dentry; | |
309 | struct inode *inode; | |
8837abca | 310 | int accmode = NFSD_MAY_SATTR; |
175a4eb7 | 311 | umode_t ftype = 0; |
6264d69d AV |
312 | __be32 err; |
313 | int host_err; | |
1da177e4 LT |
314 | int size_change = 0; |
315 | ||
316 | if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) | |
8837abca | 317 | accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; |
1da177e4 LT |
318 | if (iap->ia_valid & ATTR_SIZE) |
319 | ftype = S_IFREG; | |
320 | ||
321 | /* Get inode */ | |
322 | err = fh_verify(rqstp, fhp, ftype, accmode); | |
15b7a1b8 | 323 | if (err) |
1da177e4 LT |
324 | goto out; |
325 | ||
326 | dentry = fhp->fh_dentry; | |
327 | inode = dentry->d_inode; | |
328 | ||
15b7a1b8 N |
329 | /* Ignore any mode updates on symlinks */ |
330 | if (S_ISLNK(inode->i_mode)) | |
331 | iap->ia_valid &= ~ATTR_MODE; | |
332 | ||
333 | if (!iap->ia_valid) | |
334 | goto out; | |
335 | ||
9c85fca5 CH |
336 | /* |
337 | * NFSv2 does not differentiate between "set-[ac]time-to-now" | |
1da177e4 LT |
338 | * which only requires access, and "set-[ac]time-to-X" which |
339 | * requires ownership. | |
340 | * So if it looks like it might be "set both to the same time which | |
341 | * is close to now", and if inode_change_ok fails, then we | |
342 | * convert to "set to now" instead of "set to explicit time" | |
343 | * | |
344 | * We only call inode_change_ok as the last test as technically | |
345 | * it is not an interface that we should be using. It is only | |
346 | * valid if the filesystem does not define it's own i_op->setattr. | |
347 | */ | |
348 | #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) | |
349 | #define MAX_TOUCH_TIME_ERROR (30*60) | |
9c85fca5 CH |
350 | if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET && |
351 | iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) { | |
352 | /* | |
353 | * Looks probable. | |
354 | * | |
355 | * Now just make sure time is in the right ballpark. | |
356 | * Solaris, at least, doesn't seem to care what the time | |
357 | * request is. We require it be within 30 minutes of now. | |
1da177e4 | 358 | */ |
9c85fca5 CH |
359 | time_t delta = iap->ia_atime.tv_sec - get_seconds(); |
360 | if (delta < 0) | |
361 | delta = -delta; | |
362 | if (delta < MAX_TOUCH_TIME_ERROR && | |
363 | inode_change_ok(inode, iap) != 0) { | |
364 | /* | |
365 | * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME. | |
366 | * This will cause notify_change to set these times | |
367 | * to "now" | |
368 | */ | |
369 | iap->ia_valid &= ~BOTH_TIME_SET; | |
370 | } | |
1da177e4 LT |
371 | } |
372 | ||
9c85fca5 CH |
373 | /* |
374 | * The size case is special. | |
375 | * It changes the file as well as the attributes. | |
376 | */ | |
1da177e4 LT |
377 | if (iap->ia_valid & ATTR_SIZE) { |
378 | if (iap->ia_size < inode->i_size) { | |
8837abca MS |
379 | err = nfsd_permission(rqstp, fhp->fh_export, dentry, |
380 | NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE); | |
1da177e4 LT |
381 | if (err) |
382 | goto out; | |
383 | } | |
384 | ||
6264d69d AV |
385 | host_err = get_write_access(inode); |
386 | if (host_err) | |
1da177e4 LT |
387 | goto out_nfserr; |
388 | ||
389 | size_change = 1; | |
6264d69d AV |
390 | host_err = locks_verify_truncate(inode, NULL, iap->ia_size); |
391 | if (host_err) { | |
1da177e4 LT |
392 | put_write_access(inode); |
393 | goto out_nfserr; | |
394 | } | |
1da177e4 LT |
395 | } |
396 | ||
ca456252 | 397 | /* sanitize the mode change */ |
1da177e4 LT |
398 | if (iap->ia_valid & ATTR_MODE) { |
399 | iap->ia_mode &= S_IALLUGO; | |
dee3209d | 400 | iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); |
ca456252 JL |
401 | } |
402 | ||
403 | /* Revoke setuid/setgid on chown */ | |
0953e620 | 404 | if (!S_ISDIR(inode->i_mode) && |
6fab8779 EB |
405 | (((iap->ia_valid & ATTR_UID) && !uid_eq(iap->ia_uid, inode->i_uid)) || |
406 | ((iap->ia_valid & ATTR_GID) && !gid_eq(iap->ia_gid, inode->i_gid)))) { | |
ca456252 JL |
407 | iap->ia_valid |= ATTR_KILL_PRIV; |
408 | if (iap->ia_valid & ATTR_MODE) { | |
409 | /* we're setting mode too, just clear the s*id bits */ | |
8a0ce7d9 | 410 | iap->ia_mode &= ~S_ISUID; |
ca456252 JL |
411 | if (iap->ia_mode & S_IXGRP) |
412 | iap->ia_mode &= ~S_ISGID; | |
413 | } else { | |
414 | /* set ATTR_KILL_* bits and let VFS handle it */ | |
415 | iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); | |
8a0ce7d9 | 416 | } |
1da177e4 LT |
417 | } |
418 | ||
1da177e4 LT |
419 | /* Change the attributes. */ |
420 | ||
421 | iap->ia_valid |= ATTR_CTIME; | |
422 | ||
423 | err = nfserr_notsync; | |
424 | if (!check_guard || guardtime == inode->i_ctime.tv_sec) { | |
4795bb37 | 425 | host_err = nfsd_break_lease(inode); |
6a76bebe BF |
426 | if (host_err) |
427 | goto out_nfserr; | |
1da177e4 | 428 | fh_lock(fhp); |
6a76bebe | 429 | |
27ac0ffe | 430 | host_err = notify_change(dentry, iap, NULL); |
6264d69d | 431 | err = nfserrno(host_err); |
1da177e4 LT |
432 | fh_unlock(fhp); |
433 | } | |
434 | if (size_change) | |
435 | put_write_access(inode); | |
436 | if (!err) | |
b160fdab | 437 | commit_metadata(fhp); |
1da177e4 LT |
438 | out: |
439 | return err; | |
440 | ||
441 | out_nfserr: | |
6264d69d | 442 | err = nfserrno(host_err); |
1da177e4 LT |
443 | goto out; |
444 | } | |
445 | ||
5be196e5 CH |
446 | #if defined(CONFIG_NFSD_V2_ACL) || \ |
447 | defined(CONFIG_NFSD_V3_ACL) || \ | |
448 | defined(CONFIG_NFSD_V4) | |
449 | static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) | |
450 | { | |
451 | ssize_t buflen; | |
6c6a426f | 452 | ssize_t ret; |
5be196e5 CH |
453 | |
454 | buflen = vfs_getxattr(dentry, key, NULL, 0); | |
455 | if (buflen <= 0) | |
456 | return buflen; | |
1da177e4 | 457 | |
5be196e5 CH |
458 | *buf = kmalloc(buflen, GFP_KERNEL); |
459 | if (!*buf) | |
460 | return -ENOMEM; | |
461 | ||
6c6a426f KK |
462 | ret = vfs_getxattr(dentry, key, *buf, buflen); |
463 | if (ret < 0) | |
464 | kfree(*buf); | |
465 | return ret; | |
5be196e5 CH |
466 | } |
467 | #endif | |
468 | ||
469 | #if defined(CONFIG_NFSD_V4) | |
1da177e4 LT |
470 | static int |
471 | set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) | |
472 | { | |
473 | int len; | |
474 | size_t buflen; | |
475 | char *buf = NULL; | |
476 | int error = 0; | |
1da177e4 LT |
477 | |
478 | buflen = posix_acl_xattr_size(pacl->a_count); | |
479 | buf = kmalloc(buflen, GFP_KERNEL); | |
480 | error = -ENOMEM; | |
481 | if (buf == NULL) | |
482 | goto out; | |
483 | ||
5f3a4a28 | 484 | len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen); |
1da177e4 LT |
485 | if (len < 0) { |
486 | error = len; | |
487 | goto out; | |
488 | } | |
489 | ||
5be196e5 | 490 | error = vfs_setxattr(dentry, key, buf, len, 0); |
1da177e4 LT |
491 | out: |
492 | kfree(buf); | |
493 | return error; | |
494 | } | |
495 | ||
6264d69d | 496 | __be32 |
1da177e4 LT |
497 | nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, |
498 | struct nfs4_acl *acl) | |
499 | { | |
6264d69d AV |
500 | __be32 error; |
501 | int host_error; | |
1da177e4 LT |
502 | struct dentry *dentry; |
503 | struct inode *inode; | |
504 | struct posix_acl *pacl = NULL, *dpacl = NULL; | |
505 | unsigned int flags = 0; | |
506 | ||
507 | /* Get inode */ | |
e281d810 | 508 | error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR); |
1da177e4 | 509 | if (error) |
4b2ca38a | 510 | return error; |
1da177e4 LT |
511 | |
512 | dentry = fhp->fh_dentry; | |
513 | inode = dentry->d_inode; | |
514 | if (S_ISDIR(inode->i_mode)) | |
515 | flags = NFS4_ACL_DIR; | |
516 | ||
6264d69d AV |
517 | host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); |
518 | if (host_error == -EINVAL) { | |
4b2ca38a | 519 | return nfserr_attrnotsupp; |
6264d69d | 520 | } else if (host_error < 0) |
1da177e4 LT |
521 | goto out_nfserr; |
522 | ||
6264d69d AV |
523 | host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); |
524 | if (host_error < 0) | |
4b2ca38a | 525 | goto out_release; |
1da177e4 | 526 | |
4b2ca38a | 527 | if (S_ISDIR(inode->i_mode)) |
6264d69d | 528 | host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); |
1da177e4 | 529 | |
4b2ca38a | 530 | out_release: |
1da177e4 LT |
531 | posix_acl_release(pacl); |
532 | posix_acl_release(dpacl); | |
1da177e4 | 533 | out_nfserr: |
f34f9242 | 534 | if (host_error == -EOPNOTSUPP) |
4b2ca38a | 535 | return nfserr_attrnotsupp; |
f34f9242 | 536 | else |
4b2ca38a | 537 | return nfserrno(host_error); |
1da177e4 LT |
538 | } |
539 | ||
540 | static struct posix_acl * | |
541 | _get_posix_acl(struct dentry *dentry, char *key) | |
542 | { | |
5be196e5 | 543 | void *buf = NULL; |
1da177e4 | 544 | struct posix_acl *pacl = NULL; |
5be196e5 | 545 | int buflen; |
1da177e4 | 546 | |
5be196e5 CH |
547 | buflen = nfsd_getxattr(dentry, key, &buf); |
548 | if (!buflen) | |
549 | buflen = -ENODATA; | |
550 | if (buflen <= 0) | |
551 | return ERR_PTR(buflen); | |
1da177e4 | 552 | |
5f3a4a28 | 553 | pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen); |
1da177e4 LT |
554 | kfree(buf); |
555 | return pacl; | |
1da177e4 LT |
556 | } |
557 | ||
558 | int | |
559 | nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) | |
560 | { | |
561 | struct inode *inode = dentry->d_inode; | |
562 | int error = 0; | |
563 | struct posix_acl *pacl = NULL, *dpacl = NULL; | |
564 | unsigned int flags = 0; | |
565 | ||
9a59f452 | 566 | pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); |
1da177e4 LT |
567 | if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) |
568 | pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); | |
569 | if (IS_ERR(pacl)) { | |
570 | error = PTR_ERR(pacl); | |
571 | pacl = NULL; | |
572 | goto out; | |
573 | } | |
574 | ||
575 | if (S_ISDIR(inode->i_mode)) { | |
9a59f452 | 576 | dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); |
1da177e4 LT |
577 | if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) |
578 | dpacl = NULL; | |
579 | else if (IS_ERR(dpacl)) { | |
580 | error = PTR_ERR(dpacl); | |
581 | dpacl = NULL; | |
582 | goto out; | |
583 | } | |
584 | flags = NFS4_ACL_DIR; | |
585 | } | |
586 | ||
587 | *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); | |
588 | if (IS_ERR(*acl)) { | |
589 | error = PTR_ERR(*acl); | |
590 | *acl = NULL; | |
591 | } | |
592 | out: | |
593 | posix_acl_release(pacl); | |
594 | posix_acl_release(dpacl); | |
595 | return error; | |
596 | } | |
597 | ||
9b4146e8 CL |
598 | /* |
599 | * NFS junction information is stored in an extended attribute. | |
600 | */ | |
601 | #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs" | |
602 | ||
603 | /** | |
604 | * nfsd4_is_junction - Test if an object could be an NFS junction | |
605 | * | |
606 | * @dentry: object to test | |
607 | * | |
608 | * Returns 1 if "dentry" appears to contain NFS junction information. | |
609 | * Otherwise 0 is returned. | |
610 | */ | |
11fcee02 TM |
611 | int nfsd4_is_junction(struct dentry *dentry) |
612 | { | |
613 | struct inode *inode = dentry->d_inode; | |
614 | ||
615 | if (inode == NULL) | |
616 | return 0; | |
617 | if (inode->i_mode & S_IXUGO) | |
618 | return 0; | |
619 | if (!(inode->i_mode & S_ISVTX)) | |
620 | return 0; | |
9b4146e8 | 621 | if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0) |
11fcee02 TM |
622 | return 0; |
623 | return 1; | |
624 | } | |
18032ca0 DQ |
625 | #ifdef CONFIG_NFSD_V4_SECURITY_LABEL |
626 | __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
627 | struct xdr_netobj *label) | |
628 | { | |
629 | __be32 error; | |
630 | int host_error; | |
631 | struct dentry *dentry; | |
632 | ||
633 | error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); | |
634 | if (error) | |
635 | return error; | |
636 | ||
637 | dentry = fhp->fh_dentry; | |
638 | ||
639 | mutex_lock(&dentry->d_inode->i_mutex); | |
640 | host_error = security_inode_setsecctx(dentry, label->data, label->len); | |
641 | mutex_unlock(&dentry->d_inode->i_mutex); | |
642 | return nfserrno(host_error); | |
643 | } | |
644 | #else | |
645 | __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
646 | struct xdr_netobj *label) | |
647 | { | |
648 | return nfserr_notsupp; | |
649 | } | |
650 | #endif | |
651 | ||
6a85d6c7 | 652 | #endif /* defined(CONFIG_NFSD_V4) */ |
1da177e4 LT |
653 | |
654 | #ifdef CONFIG_NFSD_V3 | |
655 | /* | |
656 | * Check server access rights to a file system object | |
657 | */ | |
658 | struct accessmap { | |
659 | u32 access; | |
660 | int how; | |
661 | }; | |
662 | static struct accessmap nfs3_regaccess[] = { | |
8837abca MS |
663 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
664 | { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, | |
665 | { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, | |
666 | { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, | |
1da177e4 LT |
667 | |
668 | { 0, 0 } | |
669 | }; | |
670 | ||
671 | static struct accessmap nfs3_diraccess[] = { | |
8837abca MS |
672 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
673 | { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, | |
674 | { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, | |
675 | { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, | |
676 | { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, | |
1da177e4 LT |
677 | |
678 | { 0, 0 } | |
679 | }; | |
680 | ||
681 | static struct accessmap nfs3_anyaccess[] = { | |
682 | /* Some clients - Solaris 2.6 at least, make an access call | |
683 | * to the server to check for access for things like /dev/null | |
684 | * (which really, the server doesn't care about). So | |
685 | * We provide simple access checking for them, looking | |
686 | * mainly at mode bits, and we make sure to ignore read-only | |
687 | * filesystem checks | |
688 | */ | |
8837abca MS |
689 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
690 | { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, | |
691 | { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, | |
692 | { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, | |
1da177e4 LT |
693 | |
694 | { 0, 0 } | |
695 | }; | |
696 | ||
6264d69d | 697 | __be32 |
1da177e4 LT |
698 | nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) |
699 | { | |
700 | struct accessmap *map; | |
701 | struct svc_export *export; | |
702 | struct dentry *dentry; | |
703 | u32 query, result = 0, sresult = 0; | |
6264d69d | 704 | __be32 error; |
1da177e4 | 705 | |
8837abca | 706 | error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); |
1da177e4 LT |
707 | if (error) |
708 | goto out; | |
709 | ||
710 | export = fhp->fh_export; | |
711 | dentry = fhp->fh_dentry; | |
712 | ||
713 | if (S_ISREG(dentry->d_inode->i_mode)) | |
714 | map = nfs3_regaccess; | |
715 | else if (S_ISDIR(dentry->d_inode->i_mode)) | |
716 | map = nfs3_diraccess; | |
717 | else | |
718 | map = nfs3_anyaccess; | |
719 | ||
720 | ||
721 | query = *access; | |
722 | for (; map->access; map++) { | |
723 | if (map->access & query) { | |
6264d69d | 724 | __be32 err2; |
1da177e4 LT |
725 | |
726 | sresult |= map->access; | |
727 | ||
0ec757df | 728 | err2 = nfsd_permission(rqstp, export, dentry, map->how); |
1da177e4 LT |
729 | switch (err2) { |
730 | case nfs_ok: | |
731 | result |= map->access; | |
732 | break; | |
733 | ||
734 | /* the following error codes just mean the access was not allowed, | |
735 | * rather than an error occurred */ | |
736 | case nfserr_rofs: | |
737 | case nfserr_acces: | |
738 | case nfserr_perm: | |
739 | /* simply don't "or" in the access bit. */ | |
740 | break; | |
741 | default: | |
742 | error = err2; | |
743 | goto out; | |
744 | } | |
745 | } | |
746 | } | |
747 | *access = result; | |
748 | if (supported) | |
749 | *supported = sresult; | |
750 | ||
751 | out: | |
752 | return error; | |
753 | } | |
754 | #endif /* CONFIG_NFSD_V3 */ | |
755 | ||
105f4622 BF |
756 | static int nfsd_open_break_lease(struct inode *inode, int access) |
757 | { | |
758 | unsigned int mode; | |
1da177e4 | 759 | |
105f4622 BF |
760 | if (access & NFSD_MAY_NOT_BREAK_LEASE) |
761 | return 0; | |
762 | mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY; | |
763 | return break_lease(inode, mode | O_NONBLOCK); | |
764 | } | |
1da177e4 LT |
765 | |
766 | /* | |
767 | * Open an existing file or directory. | |
999448a8 BS |
768 | * The may_flags argument indicates the type of open (read/write/lock) |
769 | * and additional flags. | |
1da177e4 LT |
770 | * N.B. After this call fhp needs an fh_put |
771 | */ | |
6264d69d | 772 | __be32 |
175a4eb7 | 773 | nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, |
999448a8 | 774 | int may_flags, struct file **filp) |
1da177e4 | 775 | { |
765927b2 | 776 | struct path path; |
1da177e4 | 777 | struct inode *inode; |
6264d69d AV |
778 | int flags = O_RDONLY|O_LARGEFILE; |
779 | __be32 err; | |
91885258 | 780 | int host_err = 0; |
1da177e4 | 781 | |
e0e81739 DH |
782 | validate_process_creds(); |
783 | ||
1da177e4 LT |
784 | /* |
785 | * If we get here, then the client has already done an "open", | |
786 | * and (hopefully) checked permission - so allow OWNER_OVERRIDE | |
787 | * in case a chmod has now revoked permission. | |
d91d0b56 BF |
788 | * |
789 | * Arguably we should also allow the owner override for | |
790 | * directories, but we never have and it doesn't seem to have | |
791 | * caused anyone a problem. If we were to change this, note | |
792 | * also that our filldir callbacks would need a variant of | |
793 | * lookup_one_len that doesn't check permissions. | |
1da177e4 | 794 | */ |
d91d0b56 BF |
795 | if (type == S_IFREG) |
796 | may_flags |= NFSD_MAY_OWNER_OVERRIDE; | |
797 | err = fh_verify(rqstp, fhp, type, may_flags); | |
1da177e4 LT |
798 | if (err) |
799 | goto out; | |
800 | ||
765927b2 AV |
801 | path.mnt = fhp->fh_export->ex_path.mnt; |
802 | path.dentry = fhp->fh_dentry; | |
803 | inode = path.dentry->d_inode; | |
1da177e4 LT |
804 | |
805 | /* Disallow write access to files with the append-only bit set | |
806 | * or any access when mandatory locking enabled | |
807 | */ | |
808 | err = nfserr_perm; | |
999448a8 | 809 | if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE)) |
1da177e4 | 810 | goto out; |
5e7fc436 BF |
811 | /* |
812 | * We must ignore files (but only files) which might have mandatory | |
813 | * locks on them because there is no way to know if the accesser has | |
814 | * the lock. | |
815 | */ | |
816 | if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) | |
1da177e4 LT |
817 | goto out; |
818 | ||
819 | if (!inode->i_fop) | |
820 | goto out; | |
821 | ||
999448a8 | 822 | host_err = nfsd_open_break_lease(inode, may_flags); |
6264d69d | 823 | if (host_err) /* NOMEM or WOULDBLOCK */ |
1da177e4 LT |
824 | goto out_nfserr; |
825 | ||
999448a8 BS |
826 | if (may_flags & NFSD_MAY_WRITE) { |
827 | if (may_flags & NFSD_MAY_READ) | |
9ecb6a08 BF |
828 | flags = O_RDWR|O_LARGEFILE; |
829 | else | |
830 | flags = O_WRONLY|O_LARGEFILE; | |
1da177e4 | 831 | } |
765927b2 | 832 | *filp = dentry_open(&path, flags, current_cred()); |
e4daf1ff | 833 | if (IS_ERR(*filp)) { |
6264d69d | 834 | host_err = PTR_ERR(*filp); |
e4daf1ff HJ |
835 | *filp = NULL; |
836 | } else { | |
999448a8 BS |
837 | host_err = ima_file_check(*filp, may_flags); |
838 | ||
06effdbb BS |
839 | if (may_flags & NFSD_MAY_64BIT_COOKIE) |
840 | (*filp)->f_mode |= FMODE_64BITHASH; | |
841 | else | |
842 | (*filp)->f_mode |= FMODE_32BITHASH; | |
843 | } | |
844 | ||
1da177e4 | 845 | out_nfserr: |
6264d69d | 846 | err = nfserrno(host_err); |
1da177e4 | 847 | out: |
e0e81739 | 848 | validate_process_creds(); |
1da177e4 LT |
849 | return err; |
850 | } | |
851 | ||
852 | /* | |
853 | * Close a file. | |
854 | */ | |
855 | void | |
856 | nfsd_close(struct file *filp) | |
857 | { | |
858 | fput(filp); | |
859 | } | |
860 | ||
1da177e4 LT |
861 | /* |
862 | * Obtain the readahead parameters for the file | |
863 | * specified by (dev, ino). | |
864 | */ | |
1da177e4 LT |
865 | |
866 | static inline struct raparms * | |
867 | nfsd_get_raparms(dev_t dev, ino_t ino) | |
868 | { | |
869 | struct raparms *ra, **rap, **frap = NULL; | |
870 | int depth = 0; | |
fce1456a GB |
871 | unsigned int hash; |
872 | struct raparm_hbucket *rab; | |
873 | ||
874 | hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; | |
875 | rab = &raparm_hash[hash]; | |
1da177e4 | 876 | |
fce1456a GB |
877 | spin_lock(&rab->pb_lock); |
878 | for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { | |
1da177e4 LT |
879 | if (ra->p_ino == ino && ra->p_dev == dev) |
880 | goto found; | |
881 | depth++; | |
882 | if (ra->p_count == 0) | |
883 | frap = rap; | |
884 | } | |
3aa6e0aa | 885 | depth = nfsdstats.ra_size; |
1da177e4 | 886 | if (!frap) { |
fce1456a | 887 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
888 | return NULL; |
889 | } | |
890 | rap = frap; | |
891 | ra = *frap; | |
892 | ra->p_dev = dev; | |
893 | ra->p_ino = ino; | |
894 | ra->p_set = 0; | |
fce1456a | 895 | ra->p_hindex = hash; |
1da177e4 | 896 | found: |
fce1456a | 897 | if (rap != &rab->pb_head) { |
1da177e4 | 898 | *rap = ra->p_next; |
fce1456a GB |
899 | ra->p_next = rab->pb_head; |
900 | rab->pb_head = ra; | |
1da177e4 LT |
901 | } |
902 | ra->p_count++; | |
903 | nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; | |
fce1456a | 904 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
905 | return ra; |
906 | } | |
907 | ||
908 | /* | |
cf8208d0 JA |
909 | * Grab and keep cached pages associated with a file in the svc_rqst |
910 | * so that they can be passed to the network sendmsg/sendpage routines | |
911 | * directly. They will be released after the sending has completed. | |
1da177e4 LT |
912 | */ |
913 | static int | |
cf8208d0 JA |
914 | nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
915 | struct splice_desc *sd) | |
1da177e4 | 916 | { |
cf8208d0 | 917 | struct svc_rqst *rqstp = sd->u.data; |
afc59400 | 918 | struct page **pp = rqstp->rq_next_page; |
cf8208d0 JA |
919 | struct page *page = buf->page; |
920 | size_t size; | |
cf8208d0 JA |
921 | |
922 | size = sd->len; | |
1da177e4 LT |
923 | |
924 | if (rqstp->rq_res.page_len == 0) { | |
925 | get_page(page); | |
afc59400 BF |
926 | put_page(*rqstp->rq_next_page); |
927 | *(rqstp->rq_next_page++) = page; | |
cf8208d0 | 928 | rqstp->rq_res.page_base = buf->offset; |
1da177e4 | 929 | rqstp->rq_res.page_len = size; |
44524359 | 930 | } else if (page != pp[-1]) { |
1da177e4 | 931 | get_page(page); |
afc59400 BF |
932 | if (*rqstp->rq_next_page) |
933 | put_page(*rqstp->rq_next_page); | |
934 | *(rqstp->rq_next_page++) = page; | |
1da177e4 | 935 | rqstp->rq_res.page_len += size; |
44524359 | 936 | } else |
1da177e4 | 937 | rqstp->rq_res.page_len += size; |
1da177e4 | 938 | |
1da177e4 LT |
939 | return size; |
940 | } | |
941 | ||
cf8208d0 JA |
942 | static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, |
943 | struct splice_desc *sd) | |
944 | { | |
945 | return __splice_from_pipe(pipe, sd, nfsd_splice_actor); | |
946 | } | |
947 | ||
6264d69d | 948 | static __be32 |
1da177e4 LT |
949 | nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
950 | loff_t offset, struct kvec *vec, int vlen, unsigned long *count) | |
951 | { | |
1da177e4 | 952 | mm_segment_t oldfs; |
6264d69d AV |
953 | __be32 err; |
954 | int host_err; | |
1da177e4 LT |
955 | |
956 | err = nfserr_perm; | |
a8754bee | 957 | |
cf8208d0 JA |
958 | if (file->f_op->splice_read && rqstp->rq_splice_ok) { |
959 | struct splice_desc sd = { | |
960 | .len = 0, | |
961 | .total_len = *count, | |
962 | .pos = offset, | |
963 | .u.data = rqstp, | |
964 | }; | |
965 | ||
afc59400 | 966 | rqstp->rq_next_page = rqstp->rq_respages + 1; |
cf8208d0 | 967 | host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); |
1da177e4 LT |
968 | } else { |
969 | oldfs = get_fs(); | |
970 | set_fs(KERNEL_DS); | |
6264d69d | 971 | host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); |
1da177e4 LT |
972 | set_fs(oldfs); |
973 | } | |
974 | ||
6264d69d AV |
975 | if (host_err >= 0) { |
976 | nfsdstats.io_read += host_err; | |
977 | *count = host_err; | |
1da177e4 | 978 | err = 0; |
2a12a9d7 | 979 | fsnotify_access(file); |
1da177e4 | 980 | } else |
6264d69d | 981 | err = nfserrno(host_err); |
1da177e4 LT |
982 | return err; |
983 | } | |
984 | ||
9f708e40 NB |
985 | static void kill_suid(struct dentry *dentry) |
986 | { | |
987 | struct iattr ia; | |
b5376771 | 988 | ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; |
9f708e40 | 989 | |
1b1dcc1b | 990 | mutex_lock(&dentry->d_inode->i_mutex); |
27ac0ffe BF |
991 | /* |
992 | * Note we call this on write, so notify_change will not | |
993 | * encounter any conflicting delegations: | |
994 | */ | |
995 | notify_change(dentry, &ia, NULL); | |
1b1dcc1b | 996 | mutex_unlock(&dentry->d_inode->i_mutex); |
9f708e40 NB |
997 | } |
998 | ||
d911df7b BF |
999 | /* |
1000 | * Gathered writes: If another process is currently writing to the file, | |
1001 | * there's a high chance this is another nfsd (triggered by a bulk write | |
1002 | * from a client's biod). Rather than syncing the file with each write | |
1003 | * request, we sleep for 10 msec. | |
1004 | * | |
1005 | * I don't know if this roughly approximates C. Juszak's idea of | |
1006 | * gathered writes, but it's a nice and simple solution (IMHO), and it | |
1007 | * seems to work:-) | |
1008 | * | |
1009 | * Note: we do this only in the NFSv2 case, since v3 and higher have a | |
1010 | * better tool (separate unstable writes and commits) for solving this | |
1011 | * problem. | |
1012 | */ | |
1013 | static int wait_for_concurrent_writes(struct file *file) | |
1014 | { | |
496ad9aa | 1015 | struct inode *inode = file_inode(file); |
d911df7b BF |
1016 | static ino_t last_ino; |
1017 | static dev_t last_dev; | |
1018 | int err = 0; | |
1019 | ||
1020 | if (atomic_read(&inode->i_writecount) > 1 | |
1021 | || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { | |
1022 | dprintk("nfsd: write defer %d\n", task_pid_nr(current)); | |
1023 | msleep(10); | |
1024 | dprintk("nfsd: write resume %d\n", task_pid_nr(current)); | |
1025 | } | |
1026 | ||
1027 | if (inode->i_state & I_DIRTY) { | |
1028 | dprintk("nfsd: write sync %d\n", task_pid_nr(current)); | |
8018ab05 | 1029 | err = vfs_fsync(file, 0); |
d911df7b BF |
1030 | } |
1031 | last_ino = inode->i_ino; | |
1032 | last_dev = inode->i_sb->s_dev; | |
1033 | return err; | |
1034 | } | |
1035 | ||
6264d69d | 1036 | static __be32 |
1da177e4 LT |
1037 | nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
1038 | loff_t offset, struct kvec *vec, int vlen, | |
31dec253 | 1039 | unsigned long *cnt, int *stablep) |
1da177e4 LT |
1040 | { |
1041 | struct svc_export *exp; | |
1042 | struct dentry *dentry; | |
1043 | struct inode *inode; | |
1044 | mm_segment_t oldfs; | |
6264d69d AV |
1045 | __be32 err = 0; |
1046 | int host_err; | |
1da177e4 | 1047 | int stable = *stablep; |
48e03bc5 | 1048 | int use_wgather; |
e49dbbf3 | 1049 | loff_t pos = offset; |
1da177e4 | 1050 | |
7eaa36e2 | 1051 | dentry = file->f_path.dentry; |
1da177e4 LT |
1052 | inode = dentry->d_inode; |
1053 | exp = fhp->fh_export; | |
1054 | ||
48e03bc5 | 1055 | use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); |
1da177e4 | 1056 | |
1da177e4 LT |
1057 | if (!EX_ISSYNC(exp)) |
1058 | stable = 0; | |
1da177e4 LT |
1059 | |
1060 | /* Write the data. */ | |
1061 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
e49dbbf3 | 1062 | host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos); |
1da177e4 | 1063 | set_fs(oldfs); |
e4636d53 BF |
1064 | if (host_err < 0) |
1065 | goto out_nfserr; | |
1066 | *cnt = host_err; | |
1067 | nfsdstats.io_write += host_err; | |
2a12a9d7 | 1068 | fsnotify_modify(file); |
1da177e4 LT |
1069 | |
1070 | /* clear setuid/setgid flag after write */ | |
e4636d53 | 1071 | if (inode->i_mode & (S_ISUID | S_ISGID)) |
9f708e40 | 1072 | kill_suid(dentry); |
1da177e4 | 1073 | |
face1502 BF |
1074 | if (stable) { |
1075 | if (use_wgather) | |
1076 | host_err = wait_for_concurrent_writes(file); | |
1077 | else | |
1078 | host_err = vfs_fsync_range(file, offset, offset+*cnt, 0); | |
1079 | } | |
1da177e4 | 1080 | |
e4636d53 | 1081 | out_nfserr: |
6264d69d | 1082 | dprintk("nfsd: write complete host_err=%d\n", host_err); |
a0d24b29 | 1083 | if (host_err >= 0) |
1da177e4 | 1084 | err = 0; |
a0d24b29 | 1085 | else |
6264d69d | 1086 | err = nfserrno(host_err); |
1da177e4 LT |
1087 | return err; |
1088 | } | |
1089 | ||
1090 | /* | |
1091 | * Read data from a file. count must contain the requested read count | |
1092 | * on entry. On return, *count contains the number of bytes actually read. | |
1093 | * N.B. After this call fhp needs an fh_put | |
1094 | */ | |
039a87ca | 1095 | __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, |
fa0a2126 BF |
1096 | loff_t offset, struct kvec *vec, int vlen, unsigned long *count) |
1097 | { | |
1098 | struct file *file; | |
1099 | struct inode *inode; | |
1100 | struct raparms *ra; | |
1101 | __be32 err; | |
1102 | ||
1103 | err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); | |
1104 | if (err) | |
1105 | return err; | |
1106 | ||
496ad9aa | 1107 | inode = file_inode(file); |
fa0a2126 BF |
1108 | |
1109 | /* Get readahead parameters */ | |
1110 | ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); | |
1111 | ||
1112 | if (ra && ra->p_set) | |
1113 | file->f_ra = ra->p_ra; | |
1114 | ||
1115 | err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); | |
1116 | ||
1117 | /* Write back readahead params */ | |
1118 | if (ra) { | |
1119 | struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; | |
1120 | spin_lock(&rab->pb_lock); | |
1121 | ra->p_ra = file->f_ra; | |
1122 | ra->p_set = 1; | |
1123 | ra->p_count--; | |
1124 | spin_unlock(&rab->pb_lock); | |
1125 | } | |
1126 | ||
1127 | nfsd_close(file); | |
1128 | return err; | |
1129 | } | |
1130 | ||
039a87ca | 1131 | /* As above, but use the provided file descriptor. */ |
6264d69d | 1132 | __be32 |
039a87ca | 1133 | nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
1da177e4 LT |
1134 | loff_t offset, struct kvec *vec, int vlen, |
1135 | unsigned long *count) | |
1136 | { | |
6264d69d | 1137 | __be32 err; |
1da177e4 LT |
1138 | |
1139 | if (file) { | |
0ec757df | 1140 | err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, |
8837abca | 1141 | NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); |
1da177e4 LT |
1142 | if (err) |
1143 | goto out; | |
1144 | err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); | |
039a87ca BF |
1145 | } else /* Note file may still be NULL in NFSv4 special stateid case: */ |
1146 | err = nfsd_read(rqstp, fhp, offset, vec, vlen, count); | |
1da177e4 LT |
1147 | out: |
1148 | return err; | |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * Write data to a file. | |
1153 | * The stable flag requests synchronous writes. | |
1154 | * N.B. After this call fhp needs an fh_put | |
1155 | */ | |
6264d69d | 1156 | __be32 |
1da177e4 | 1157 | nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
31dec253 | 1158 | loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, |
1da177e4 LT |
1159 | int *stablep) |
1160 | { | |
6264d69d | 1161 | __be32 err = 0; |
1da177e4 LT |
1162 | |
1163 | if (file) { | |
0ec757df | 1164 | err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, |
8837abca | 1165 | NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); |
1da177e4 LT |
1166 | if (err) |
1167 | goto out; | |
1168 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, | |
1169 | stablep); | |
1170 | } else { | |
8837abca | 1171 | err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); |
1da177e4 LT |
1172 | if (err) |
1173 | goto out; | |
1174 | ||
1175 | if (cnt) | |
1176 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, | |
1177 | cnt, stablep); | |
1178 | nfsd_close(file); | |
1179 | } | |
1180 | out: | |
1181 | return err; | |
1182 | } | |
1183 | ||
1184 | #ifdef CONFIG_NFSD_V3 | |
1185 | /* | |
1186 | * Commit all pending writes to stable storage. | |
aa696a6f TM |
1187 | * |
1188 | * Note: we only guarantee that data that lies within the range specified | |
1189 | * by the 'offset' and 'count' parameters will be synced. | |
1da177e4 LT |
1190 | * |
1191 | * Unfortunately we cannot lock the file to make sure we return full WCC | |
1192 | * data to the client, as locking happens lower down in the filesystem. | |
1193 | */ | |
6264d69d | 1194 | __be32 |
1da177e4 LT |
1195 | nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1196 | loff_t offset, unsigned long count) | |
1197 | { | |
1198 | struct file *file; | |
aa696a6f TM |
1199 | loff_t end = LLONG_MAX; |
1200 | __be32 err = nfserr_inval; | |
1da177e4 | 1201 | |
aa696a6f TM |
1202 | if (offset < 0) |
1203 | goto out; | |
1204 | if (count != 0) { | |
1205 | end = offset + (loff_t)count - 1; | |
1206 | if (end < offset) | |
1207 | goto out; | |
1208 | } | |
1da177e4 | 1209 | |
91885258 JL |
1210 | err = nfsd_open(rqstp, fhp, S_IFREG, |
1211 | NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); | |
8837abca | 1212 | if (err) |
aa696a6f | 1213 | goto out; |
1da177e4 | 1214 | if (EX_ISSYNC(fhp->fh_export)) { |
8018ab05 | 1215 | int err2 = vfs_fsync_range(file, offset, end, 0); |
aa696a6f TM |
1216 | |
1217 | if (err2 != -EINVAL) | |
1218 | err = nfserrno(err2); | |
1219 | else | |
1da177e4 | 1220 | err = nfserr_notsupp; |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | nfsd_close(file); | |
aa696a6f | 1224 | out: |
1da177e4 LT |
1225 | return err; |
1226 | } | |
1227 | #endif /* CONFIG_NFSD_V3 */ | |
1228 | ||
f2b0dee2 | 1229 | static __be32 |
5c002b3b BF |
1230 | nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, |
1231 | struct iattr *iap) | |
1232 | { | |
1233 | /* | |
1234 | * Mode has already been set earlier in create: | |
1235 | */ | |
1236 | iap->ia_valid &= ~ATTR_MODE; | |
1237 | /* | |
1238 | * Setting uid/gid works only for root. Irix appears to | |
1239 | * send along the gid on create when it tries to implement | |
1240 | * setgid directories via NFS: | |
1241 | */ | |
6fab8779 | 1242 | if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) |
5c002b3b BF |
1243 | iap->ia_valid &= ~(ATTR_UID|ATTR_GID); |
1244 | if (iap->ia_valid) | |
1245 | return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); | |
1246 | return 0; | |
1247 | } | |
1248 | ||
4ac35c2f | 1249 | /* HPUX client sometimes creates a file in mode 000, and sets size to 0. |
1250 | * setting size to 0 may fail for some specific file systems by the permission | |
1251 | * checking which requires WRITE permission but the mode is 000. | |
1252 | * we ignore the resizing(to 0) on the just new created file, since the size is | |
1253 | * 0 after file created. | |
1254 | * | |
1255 | * call this only after vfs_create() is called. | |
1256 | * */ | |
1257 | static void | |
1258 | nfsd_check_ignore_resizing(struct iattr *iap) | |
1259 | { | |
1260 | if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) | |
1261 | iap->ia_valid &= ~ATTR_SIZE; | |
1262 | } | |
1263 | ||
1da177e4 LT |
1264 | /* |
1265 | * Create a file (regular, directory, device, fifo); UNIX sockets | |
1266 | * not yet implemented. | |
1267 | * If the response fh has been verified, the parent directory should | |
1268 | * already be locked. Note that the parent directory is left locked. | |
1269 | * | |
1270 | * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp | |
1271 | */ | |
6264d69d | 1272 | __be32 |
1da177e4 LT |
1273 | nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1274 | char *fname, int flen, struct iattr *iap, | |
1275 | int type, dev_t rdev, struct svc_fh *resfhp) | |
1276 | { | |
1277 | struct dentry *dentry, *dchild = NULL; | |
1278 | struct inode *dirp; | |
6264d69d | 1279 | __be32 err; |
5c002b3b | 1280 | __be32 err2; |
6264d69d | 1281 | int host_err; |
1da177e4 LT |
1282 | |
1283 | err = nfserr_perm; | |
1284 | if (!flen) | |
1285 | goto out; | |
1286 | err = nfserr_exist; | |
1287 | if (isdotent(fname, flen)) | |
1288 | goto out; | |
1289 | ||
8837abca | 1290 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1291 | if (err) |
1292 | goto out; | |
1293 | ||
1294 | dentry = fhp->fh_dentry; | |
1295 | dirp = dentry->d_inode; | |
1296 | ||
1297 | err = nfserr_notdir; | |
acfa4380 | 1298 | if (!dirp->i_op->lookup) |
1da177e4 LT |
1299 | goto out; |
1300 | /* | |
1301 | * Check whether the response file handle has been verified yet. | |
1302 | * If it has, the parent directory should already be locked. | |
1303 | */ | |
1304 | if (!resfhp->fh_dentry) { | |
4a55c101 JK |
1305 | host_err = fh_want_write(fhp); |
1306 | if (host_err) | |
1307 | goto out_nfserr; | |
1308 | ||
1da177e4 | 1309 | /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ |
12fd3520 | 1310 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 | 1311 | dchild = lookup_one_len(fname, dentry, flen); |
6264d69d | 1312 | host_err = PTR_ERR(dchild); |
1da177e4 LT |
1313 | if (IS_ERR(dchild)) |
1314 | goto out_nfserr; | |
1315 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); | |
1316 | if (err) | |
1317 | goto out; | |
1318 | } else { | |
1319 | /* called from nfsd_proc_create */ | |
1320 | dchild = dget(resfhp->fh_dentry); | |
1321 | if (!fhp->fh_locked) { | |
1322 | /* not actually possible */ | |
1323 | printk(KERN_ERR | |
a6a9f18f AV |
1324 | "nfsd_create: parent %pd2 not locked!\n", |
1325 | dentry); | |
d75f2b9f | 1326 | err = nfserr_io; |
1da177e4 LT |
1327 | goto out; |
1328 | } | |
1329 | } | |
1330 | /* | |
1331 | * Make sure the child dentry is still negative ... | |
1332 | */ | |
1333 | err = nfserr_exist; | |
1334 | if (dchild->d_inode) { | |
a6a9f18f AV |
1335 | dprintk("nfsd_create: dentry %pd/%pd not negative!\n", |
1336 | dentry, dchild); | |
1da177e4 LT |
1337 | goto out; |
1338 | } | |
1339 | ||
1340 | if (!(iap->ia_valid & ATTR_MODE)) | |
1341 | iap->ia_mode = 0; | |
1342 | iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; | |
1343 | ||
07cad1d2 MS |
1344 | err = nfserr_inval; |
1345 | if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { | |
1346 | printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", | |
1347 | type); | |
1348 | goto out; | |
1349 | } | |
1350 | ||
1da177e4 LT |
1351 | /* |
1352 | * Get the dir op function pointer. | |
1353 | */ | |
088406bc | 1354 | err = 0; |
4a55c101 | 1355 | host_err = 0; |
1da177e4 LT |
1356 | switch (type) { |
1357 | case S_IFREG: | |
312b63fb | 1358 | host_err = vfs_create(dirp, dchild, iap->ia_mode, true); |
4ac35c2f | 1359 | if (!host_err) |
1360 | nfsd_check_ignore_resizing(iap); | |
1da177e4 LT |
1361 | break; |
1362 | case S_IFDIR: | |
6264d69d | 1363 | host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); |
1da177e4 LT |
1364 | break; |
1365 | case S_IFCHR: | |
1366 | case S_IFBLK: | |
1367 | case S_IFIFO: | |
1368 | case S_IFSOCK: | |
6264d69d | 1369 | host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); |
1da177e4 | 1370 | break; |
1da177e4 | 1371 | } |
4a55c101 | 1372 | if (host_err < 0) |
1da177e4 LT |
1373 | goto out_nfserr; |
1374 | ||
f501912a | 1375 | err = nfsd_create_setattr(rqstp, resfhp, iap); |
1da177e4 | 1376 | |
f501912a BM |
1377 | /* |
1378 | * nfsd_setattr already committed the child. Transactional filesystems | |
1379 | * had a chance to commit changes for both parent and child | |
1380 | * simultaneously making the following commit_metadata a noop. | |
1381 | */ | |
1382 | err2 = nfserrno(commit_metadata(fhp)); | |
5c002b3b BF |
1383 | if (err2) |
1384 | err = err2; | |
1da177e4 LT |
1385 | /* |
1386 | * Update the file handle to get the new inode info. | |
1387 | */ | |
1388 | if (!err) | |
1389 | err = fh_update(resfhp); | |
1390 | out: | |
1391 | if (dchild && !IS_ERR(dchild)) | |
1392 | dput(dchild); | |
1393 | return err; | |
1394 | ||
1395 | out_nfserr: | |
6264d69d | 1396 | err = nfserrno(host_err); |
1da177e4 LT |
1397 | goto out; |
1398 | } | |
1399 | ||
1400 | #ifdef CONFIG_NFSD_V3 | |
ac6721a1 MJ |
1401 | |
1402 | static inline int nfsd_create_is_exclusive(int createmode) | |
1403 | { | |
1404 | return createmode == NFS3_CREATE_EXCLUSIVE | |
1405 | || createmode == NFS4_CREATE_EXCLUSIVE4_1; | |
1406 | } | |
1407 | ||
1da177e4 | 1408 | /* |
ac6721a1 | 1409 | * NFSv3 and NFSv4 version of nfsd_create |
1da177e4 | 1410 | */ |
6264d69d | 1411 | __be32 |
ac6721a1 | 1412 | do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1da177e4 LT |
1413 | char *fname, int flen, struct iattr *iap, |
1414 | struct svc_fh *resfhp, int createmode, u32 *verifier, | |
856121b2 | 1415 | bool *truncp, bool *created) |
1da177e4 LT |
1416 | { |
1417 | struct dentry *dentry, *dchild = NULL; | |
1418 | struct inode *dirp; | |
6264d69d AV |
1419 | __be32 err; |
1420 | int host_err; | |
1da177e4 | 1421 | __u32 v_mtime=0, v_atime=0; |
1da177e4 LT |
1422 | |
1423 | err = nfserr_perm; | |
1424 | if (!flen) | |
1425 | goto out; | |
1426 | err = nfserr_exist; | |
1427 | if (isdotent(fname, flen)) | |
1428 | goto out; | |
1429 | if (!(iap->ia_valid & ATTR_MODE)) | |
1430 | iap->ia_mode = 0; | |
1574dff8 | 1431 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); |
1da177e4 LT |
1432 | if (err) |
1433 | goto out; | |
1434 | ||
1435 | dentry = fhp->fh_dentry; | |
1436 | dirp = dentry->d_inode; | |
1437 | ||
1438 | /* Get all the sanity checks out of the way before | |
1439 | * we lock the parent. */ | |
1440 | err = nfserr_notdir; | |
acfa4380 | 1441 | if (!dirp->i_op->lookup) |
1da177e4 | 1442 | goto out; |
4a55c101 JK |
1443 | |
1444 | host_err = fh_want_write(fhp); | |
1445 | if (host_err) | |
1446 | goto out_nfserr; | |
1447 | ||
12fd3520 | 1448 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1449 | |
1450 | /* | |
1451 | * Compose the response file handle. | |
1452 | */ | |
1453 | dchild = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1454 | host_err = PTR_ERR(dchild); |
1da177e4 LT |
1455 | if (IS_ERR(dchild)) |
1456 | goto out_nfserr; | |
1457 | ||
1574dff8 SP |
1458 | /* If file doesn't exist, check for permissions to create one */ |
1459 | if (!dchild->d_inode) { | |
1460 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); | |
1461 | if (err) | |
1462 | goto out; | |
1463 | } | |
1464 | ||
1da177e4 LT |
1465 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); |
1466 | if (err) | |
1467 | goto out; | |
1468 | ||
ac6721a1 | 1469 | if (nfsd_create_is_exclusive(createmode)) { |
c397852c | 1470 | /* solaris7 gets confused (bugid 4218508) if these have |
749997e5 JL |
1471 | * the high bit set, so just clear the high bits. If this is |
1472 | * ever changed to use different attrs for storing the | |
1473 | * verifier, then do_open_lookup() will also need to be fixed | |
1474 | * accordingly. | |
1da177e4 LT |
1475 | */ |
1476 | v_mtime = verifier[0]&0x7fffffff; | |
1477 | v_atime = verifier[1]&0x7fffffff; | |
1da177e4 LT |
1478 | } |
1479 | ||
1480 | if (dchild->d_inode) { | |
1481 | err = 0; | |
1482 | ||
1483 | switch (createmode) { | |
1484 | case NFS3_CREATE_UNCHECKED: | |
1485 | if (! S_ISREG(dchild->d_inode->i_mode)) | |
9dc4e6c4 | 1486 | goto out; |
1da177e4 LT |
1487 | else if (truncp) { |
1488 | /* in nfsv4, we need to treat this case a little | |
1489 | * differently. we don't want to truncate the | |
1490 | * file now; this would be wrong if the OPEN | |
1491 | * fails for some other reason. furthermore, | |
1492 | * if the size is nonzero, we should ignore it | |
1493 | * according to spec! | |
1494 | */ | |
1495 | *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; | |
1496 | } | |
1497 | else { | |
1498 | iap->ia_valid &= ATTR_SIZE; | |
1499 | goto set_attr; | |
1500 | } | |
1501 | break; | |
1502 | case NFS3_CREATE_EXCLUSIVE: | |
1503 | if ( dchild->d_inode->i_mtime.tv_sec == v_mtime | |
1504 | && dchild->d_inode->i_atime.tv_sec == v_atime | |
7007c90f NB |
1505 | && dchild->d_inode->i_size == 0 ) { |
1506 | if (created) | |
1507 | *created = 1; | |
1da177e4 | 1508 | break; |
7007c90f | 1509 | } |
ac6721a1 MJ |
1510 | case NFS4_CREATE_EXCLUSIVE4_1: |
1511 | if ( dchild->d_inode->i_mtime.tv_sec == v_mtime | |
1512 | && dchild->d_inode->i_atime.tv_sec == v_atime | |
7007c90f NB |
1513 | && dchild->d_inode->i_size == 0 ) { |
1514 | if (created) | |
1515 | *created = 1; | |
ac6721a1 | 1516 | goto set_attr; |
7007c90f | 1517 | } |
1da177e4 LT |
1518 | /* fallthru */ |
1519 | case NFS3_CREATE_GUARDED: | |
1520 | err = nfserr_exist; | |
1521 | } | |
bad0dcff | 1522 | fh_drop_write(fhp); |
1da177e4 LT |
1523 | goto out; |
1524 | } | |
1525 | ||
312b63fb | 1526 | host_err = vfs_create(dirp, dchild, iap->ia_mode, true); |
463c3197 | 1527 | if (host_err < 0) { |
bad0dcff | 1528 | fh_drop_write(fhp); |
1da177e4 | 1529 | goto out_nfserr; |
463c3197 | 1530 | } |
81ac95c5 BF |
1531 | if (created) |
1532 | *created = 1; | |
1da177e4 | 1533 | |
4ac35c2f | 1534 | nfsd_check_ignore_resizing(iap); |
1535 | ||
ac6721a1 | 1536 | if (nfsd_create_is_exclusive(createmode)) { |
c397852c | 1537 | /* Cram the verifier into atime/mtime */ |
1da177e4 | 1538 | iap->ia_valid = ATTR_MTIME|ATTR_ATIME |
c397852c | 1539 | | ATTR_MTIME_SET|ATTR_ATIME_SET; |
1da177e4 LT |
1540 | /* XXX someone who knows this better please fix it for nsec */ |
1541 | iap->ia_mtime.tv_sec = v_mtime; | |
1542 | iap->ia_atime.tv_sec = v_atime; | |
1543 | iap->ia_mtime.tv_nsec = 0; | |
1544 | iap->ia_atime.tv_nsec = 0; | |
1da177e4 LT |
1545 | } |
1546 | ||
1da177e4 | 1547 | set_attr: |
f501912a BM |
1548 | err = nfsd_create_setattr(rqstp, resfhp, iap); |
1549 | ||
1550 | /* | |
1551 | * nfsd_setattr already committed the child (and possibly also the parent). | |
1552 | */ | |
1553 | if (!err) | |
1554 | err = nfserrno(commit_metadata(fhp)); | |
f193fbab YT |
1555 | |
1556 | /* | |
1557 | * Update the filehandle to get the new inode info. | |
1558 | */ | |
1559 | if (!err) | |
1560 | err = fh_update(resfhp); | |
1da177e4 LT |
1561 | |
1562 | out: | |
1563 | fh_unlock(fhp); | |
1564 | if (dchild && !IS_ERR(dchild)) | |
1565 | dput(dchild); | |
4a55c101 | 1566 | fh_drop_write(fhp); |
1da177e4 LT |
1567 | return err; |
1568 | ||
1569 | out_nfserr: | |
6264d69d | 1570 | err = nfserrno(host_err); |
1da177e4 LT |
1571 | goto out; |
1572 | } | |
1573 | #endif /* CONFIG_NFSD_V3 */ | |
1574 | ||
1575 | /* | |
1576 | * Read a symlink. On entry, *lenp must contain the maximum path length that | |
1577 | * fits into the buffer. On return, it contains the true length. | |
1578 | * N.B. After this call fhp needs an fh_put | |
1579 | */ | |
6264d69d | 1580 | __be32 |
1da177e4 LT |
1581 | nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) |
1582 | { | |
1da177e4 LT |
1583 | struct inode *inode; |
1584 | mm_segment_t oldfs; | |
6264d69d AV |
1585 | __be32 err; |
1586 | int host_err; | |
68ac1234 | 1587 | struct path path; |
1da177e4 | 1588 | |
8837abca | 1589 | err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); |
1da177e4 LT |
1590 | if (err) |
1591 | goto out; | |
1592 | ||
68ac1234 AV |
1593 | path.mnt = fhp->fh_export->ex_path.mnt; |
1594 | path.dentry = fhp->fh_dentry; | |
1595 | inode = path.dentry->d_inode; | |
1da177e4 LT |
1596 | |
1597 | err = nfserr_inval; | |
acfa4380 | 1598 | if (!inode->i_op->readlink) |
1da177e4 LT |
1599 | goto out; |
1600 | ||
68ac1234 | 1601 | touch_atime(&path); |
1da177e4 LT |
1602 | /* N.B. Why does this call need a get_fs()?? |
1603 | * Remove the set_fs and watch the fireworks:-) --okir | |
1604 | */ | |
1605 | ||
1606 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
fac7a17b | 1607 | host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp); |
1da177e4 LT |
1608 | set_fs(oldfs); |
1609 | ||
6264d69d | 1610 | if (host_err < 0) |
1da177e4 | 1611 | goto out_nfserr; |
6264d69d | 1612 | *lenp = host_err; |
1da177e4 LT |
1613 | err = 0; |
1614 | out: | |
1615 | return err; | |
1616 | ||
1617 | out_nfserr: | |
6264d69d | 1618 | err = nfserrno(host_err); |
1da177e4 LT |
1619 | goto out; |
1620 | } | |
1621 | ||
1622 | /* | |
1623 | * Create a symlink and look up its inode | |
1624 | * N.B. After this call _both_ fhp and resfhp need an fh_put | |
1625 | */ | |
6264d69d | 1626 | __be32 |
1da177e4 LT |
1627 | nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1628 | char *fname, int flen, | |
1629 | char *path, int plen, | |
1630 | struct svc_fh *resfhp, | |
1631 | struct iattr *iap) | |
1632 | { | |
1633 | struct dentry *dentry, *dnew; | |
6264d69d AV |
1634 | __be32 err, cerr; |
1635 | int host_err; | |
1da177e4 LT |
1636 | |
1637 | err = nfserr_noent; | |
1638 | if (!flen || !plen) | |
1639 | goto out; | |
1640 | err = nfserr_exist; | |
1641 | if (isdotent(fname, flen)) | |
1642 | goto out; | |
1643 | ||
8837abca | 1644 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1645 | if (err) |
1646 | goto out; | |
4a55c101 JK |
1647 | |
1648 | host_err = fh_want_write(fhp); | |
1649 | if (host_err) | |
1650 | goto out_nfserr; | |
1651 | ||
1da177e4 LT |
1652 | fh_lock(fhp); |
1653 | dentry = fhp->fh_dentry; | |
1654 | dnew = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1655 | host_err = PTR_ERR(dnew); |
1da177e4 LT |
1656 | if (IS_ERR(dnew)) |
1657 | goto out_nfserr; | |
1658 | ||
1da177e4 LT |
1659 | if (unlikely(path[plen] != 0)) { |
1660 | char *path_alloced = kmalloc(plen+1, GFP_KERNEL); | |
1661 | if (path_alloced == NULL) | |
6264d69d | 1662 | host_err = -ENOMEM; |
1da177e4 LT |
1663 | else { |
1664 | strncpy(path_alloced, path, plen); | |
1665 | path_alloced[plen] = 0; | |
db2e747b | 1666 | host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); |
1da177e4 LT |
1667 | kfree(path_alloced); |
1668 | } | |
1669 | } else | |
db2e747b | 1670 | host_err = vfs_symlink(dentry->d_inode, dnew, path); |
6264d69d | 1671 | err = nfserrno(host_err); |
f501912a BM |
1672 | if (!err) |
1673 | err = nfserrno(commit_metadata(fhp)); | |
1da177e4 LT |
1674 | fh_unlock(fhp); |
1675 | ||
bad0dcff | 1676 | fh_drop_write(fhp); |
75c3f29d | 1677 | |
1da177e4 LT |
1678 | cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); |
1679 | dput(dnew); | |
1680 | if (err==0) err = cerr; | |
1681 | out: | |
1682 | return err; | |
1683 | ||
1684 | out_nfserr: | |
6264d69d | 1685 | err = nfserrno(host_err); |
1da177e4 LT |
1686 | goto out; |
1687 | } | |
1688 | ||
1689 | /* | |
1690 | * Create a hardlink | |
1691 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1692 | */ | |
6264d69d | 1693 | __be32 |
1da177e4 LT |
1694 | nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, |
1695 | char *name, int len, struct svc_fh *tfhp) | |
1696 | { | |
1697 | struct dentry *ddir, *dnew, *dold; | |
55b13354 | 1698 | struct inode *dirp; |
6264d69d AV |
1699 | __be32 err; |
1700 | int host_err; | |
1da177e4 | 1701 | |
8837abca | 1702 | err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1703 | if (err) |
1704 | goto out; | |
7d818a7b | 1705 | err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); |
1da177e4 LT |
1706 | if (err) |
1707 | goto out; | |
7d818a7b BF |
1708 | err = nfserr_isdir; |
1709 | if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode)) | |
1710 | goto out; | |
1da177e4 LT |
1711 | err = nfserr_perm; |
1712 | if (!len) | |
1713 | goto out; | |
1714 | err = nfserr_exist; | |
1715 | if (isdotent(name, len)) | |
1716 | goto out; | |
1717 | ||
4a55c101 JK |
1718 | host_err = fh_want_write(tfhp); |
1719 | if (host_err) { | |
1720 | err = nfserrno(host_err); | |
1721 | goto out; | |
1722 | } | |
1723 | ||
12fd3520 | 1724 | fh_lock_nested(ffhp, I_MUTEX_PARENT); |
1da177e4 LT |
1725 | ddir = ffhp->fh_dentry; |
1726 | dirp = ddir->d_inode; | |
1727 | ||
1728 | dnew = lookup_one_len(name, ddir, len); | |
6264d69d | 1729 | host_err = PTR_ERR(dnew); |
1da177e4 LT |
1730 | if (IS_ERR(dnew)) |
1731 | goto out_nfserr; | |
1732 | ||
1733 | dold = tfhp->fh_dentry; | |
1da177e4 | 1734 | |
4795bb37 BF |
1735 | err = nfserr_noent; |
1736 | if (!dold->d_inode) | |
4a55c101 | 1737 | goto out_dput; |
4795bb37 | 1738 | host_err = nfsd_break_lease(dold->d_inode); |
7d751f6f CB |
1739 | if (host_err) { |
1740 | err = nfserrno(host_err); | |
4a55c101 | 1741 | goto out_dput; |
7d751f6f | 1742 | } |
146a8595 | 1743 | host_err = vfs_link(dold, dirp, dnew, NULL); |
6264d69d | 1744 | if (!host_err) { |
f501912a BM |
1745 | err = nfserrno(commit_metadata(ffhp)); |
1746 | if (!err) | |
1747 | err = nfserrno(commit_metadata(tfhp)); | |
1da177e4 | 1748 | } else { |
6264d69d | 1749 | if (host_err == -EXDEV && rqstp->rq_vers == 2) |
1da177e4 LT |
1750 | err = nfserr_acces; |
1751 | else | |
6264d69d | 1752 | err = nfserrno(host_err); |
1da177e4 | 1753 | } |
75c3f29d | 1754 | out_dput: |
1da177e4 | 1755 | dput(dnew); |
270d56e5 DR |
1756 | out_unlock: |
1757 | fh_unlock(ffhp); | |
4a55c101 | 1758 | fh_drop_write(tfhp); |
1da177e4 LT |
1759 | out: |
1760 | return err; | |
1761 | ||
1762 | out_nfserr: | |
6264d69d | 1763 | err = nfserrno(host_err); |
270d56e5 | 1764 | goto out_unlock; |
1da177e4 LT |
1765 | } |
1766 | ||
1767 | /* | |
1768 | * Rename a file | |
1769 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1770 | */ | |
6264d69d | 1771 | __be32 |
1da177e4 LT |
1772 | nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, |
1773 | struct svc_fh *tfhp, char *tname, int tlen) | |
1774 | { | |
1775 | struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; | |
1776 | struct inode *fdir, *tdir; | |
6264d69d AV |
1777 | __be32 err; |
1778 | int host_err; | |
1da177e4 | 1779 | |
8837abca | 1780 | err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); |
1da177e4 LT |
1781 | if (err) |
1782 | goto out; | |
8837abca | 1783 | err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1784 | if (err) |
1785 | goto out; | |
1786 | ||
1787 | fdentry = ffhp->fh_dentry; | |
1788 | fdir = fdentry->d_inode; | |
1789 | ||
1790 | tdentry = tfhp->fh_dentry; | |
1791 | tdir = tdentry->d_inode; | |
1792 | ||
1da177e4 LT |
1793 | err = nfserr_perm; |
1794 | if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) | |
1795 | goto out; | |
1796 | ||
4a55c101 JK |
1797 | host_err = fh_want_write(ffhp); |
1798 | if (host_err) { | |
1799 | err = nfserrno(host_err); | |
1800 | goto out; | |
1801 | } | |
1802 | ||
1da177e4 LT |
1803 | /* cannot use fh_lock as we need deadlock protective ordering |
1804 | * so do it by hand */ | |
1805 | trap = lock_rename(tdentry, fdentry); | |
1806 | ffhp->fh_locked = tfhp->fh_locked = 1; | |
1807 | fill_pre_wcc(ffhp); | |
1808 | fill_pre_wcc(tfhp); | |
1809 | ||
1810 | odentry = lookup_one_len(fname, fdentry, flen); | |
6264d69d | 1811 | host_err = PTR_ERR(odentry); |
1da177e4 LT |
1812 | if (IS_ERR(odentry)) |
1813 | goto out_nfserr; | |
1814 | ||
6264d69d | 1815 | host_err = -ENOENT; |
1da177e4 LT |
1816 | if (!odentry->d_inode) |
1817 | goto out_dput_old; | |
6264d69d | 1818 | host_err = -EINVAL; |
1da177e4 LT |
1819 | if (odentry == trap) |
1820 | goto out_dput_old; | |
1821 | ||
1822 | ndentry = lookup_one_len(tname, tdentry, tlen); | |
6264d69d | 1823 | host_err = PTR_ERR(ndentry); |
1da177e4 LT |
1824 | if (IS_ERR(ndentry)) |
1825 | goto out_dput_old; | |
6264d69d | 1826 | host_err = -ENOTEMPTY; |
1da177e4 LT |
1827 | if (ndentry == trap) |
1828 | goto out_dput_new; | |
1829 | ||
9079b1eb DH |
1830 | host_err = -EXDEV; |
1831 | if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) | |
1832 | goto out_dput_new; | |
aa387d6c BF |
1833 | if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry) |
1834 | goto out_dput_new; | |
9079b1eb | 1835 | |
4795bb37 | 1836 | host_err = nfsd_break_lease(odentry->d_inode); |
83f6b0c1 | 1837 | if (host_err) |
4a55c101 | 1838 | goto out_dput_new; |
83f6b0c1 BF |
1839 | if (ndentry->d_inode) { |
1840 | host_err = nfsd_break_lease(ndentry->d_inode); | |
1841 | if (host_err) | |
4a55c101 | 1842 | goto out_dput_new; |
83f6b0c1 | 1843 | } |
8e6d782c | 1844 | host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL); |
f501912a BM |
1845 | if (!host_err) { |
1846 | host_err = commit_metadata(tfhp); | |
6264d69d | 1847 | if (!host_err) |
f501912a | 1848 | host_err = commit_metadata(ffhp); |
1da177e4 | 1849 | } |
1da177e4 LT |
1850 | out_dput_new: |
1851 | dput(ndentry); | |
1852 | out_dput_old: | |
1853 | dput(odentry); | |
1854 | out_nfserr: | |
6264d69d | 1855 | err = nfserrno(host_err); |
1da177e4 LT |
1856 | |
1857 | /* we cannot reply on fh_unlock on the two filehandles, | |
1858 | * as that would do the wrong thing if the two directories | |
1859 | * were the same, so again we do it by hand | |
1860 | */ | |
1861 | fill_post_wcc(ffhp); | |
1862 | fill_post_wcc(tfhp); | |
1863 | unlock_rename(tdentry, fdentry); | |
1864 | ffhp->fh_locked = tfhp->fh_locked = 0; | |
4a55c101 | 1865 | fh_drop_write(ffhp); |
1da177e4 LT |
1866 | |
1867 | out: | |
1868 | return err; | |
1869 | } | |
1870 | ||
1871 | /* | |
1872 | * Unlink a file or directory | |
1873 | * N.B. After this call fhp needs an fh_put | |
1874 | */ | |
6264d69d | 1875 | __be32 |
1da177e4 LT |
1876 | nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, |
1877 | char *fname, int flen) | |
1878 | { | |
1879 | struct dentry *dentry, *rdentry; | |
1880 | struct inode *dirp; | |
6264d69d AV |
1881 | __be32 err; |
1882 | int host_err; | |
1da177e4 LT |
1883 | |
1884 | err = nfserr_acces; | |
1885 | if (!flen || isdotent(fname, flen)) | |
1886 | goto out; | |
8837abca | 1887 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); |
1da177e4 LT |
1888 | if (err) |
1889 | goto out; | |
1890 | ||
4a55c101 JK |
1891 | host_err = fh_want_write(fhp); |
1892 | if (host_err) | |
1893 | goto out_nfserr; | |
1894 | ||
12fd3520 | 1895 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1896 | dentry = fhp->fh_dentry; |
1897 | dirp = dentry->d_inode; | |
1898 | ||
1899 | rdentry = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1900 | host_err = PTR_ERR(rdentry); |
1da177e4 LT |
1901 | if (IS_ERR(rdentry)) |
1902 | goto out_nfserr; | |
1903 | ||
1904 | if (!rdentry->d_inode) { | |
1905 | dput(rdentry); | |
1906 | err = nfserr_noent; | |
1907 | goto out; | |
1908 | } | |
1909 | ||
1910 | if (!type) | |
1911 | type = rdentry->d_inode->i_mode & S_IFMT; | |
1912 | ||
4795bb37 BF |
1913 | host_err = nfsd_break_lease(rdentry->d_inode); |
1914 | if (host_err) | |
4a55c101 | 1915 | goto out_put; |
9ce137ee | 1916 | if (type != S_IFDIR) |
b21996e3 | 1917 | host_err = vfs_unlink(dirp, rdentry, NULL); |
9ce137ee | 1918 | else |
6264d69d | 1919 | host_err = vfs_rmdir(dirp, rdentry); |
f501912a BM |
1920 | if (!host_err) |
1921 | host_err = commit_metadata(fhp); | |
541ce98c BF |
1922 | out_put: |
1923 | dput(rdentry); | |
1924 | ||
1da177e4 | 1925 | out_nfserr: |
6264d69d | 1926 | err = nfserrno(host_err); |
f193fbab YT |
1927 | out: |
1928 | return err; | |
1da177e4 LT |
1929 | } |
1930 | ||
14f7dd63 DW |
1931 | /* |
1932 | * We do this buffering because we must not call back into the file | |
1933 | * system's ->lookup() method from the filldir callback. That may well | |
1934 | * deadlock a number of file systems. | |
1935 | * | |
1936 | * This is based heavily on the implementation of same in XFS. | |
1937 | */ | |
1938 | struct buffered_dirent { | |
1939 | u64 ino; | |
1940 | loff_t offset; | |
1941 | int namlen; | |
1942 | unsigned int d_type; | |
1943 | char name[]; | |
1944 | }; | |
1945 | ||
1946 | struct readdir_data { | |
5c0ba4e0 | 1947 | struct dir_context ctx; |
14f7dd63 DW |
1948 | char *dirent; |
1949 | size_t used; | |
53c9c5c0 | 1950 | int full; |
14f7dd63 DW |
1951 | }; |
1952 | ||
1953 | static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, | |
1954 | loff_t offset, u64 ino, unsigned int d_type) | |
1955 | { | |
1956 | struct readdir_data *buf = __buf; | |
1957 | struct buffered_dirent *de = (void *)(buf->dirent + buf->used); | |
1958 | unsigned int reclen; | |
1959 | ||
1960 | reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); | |
53c9c5c0 AV |
1961 | if (buf->used + reclen > PAGE_SIZE) { |
1962 | buf->full = 1; | |
14f7dd63 | 1963 | return -EINVAL; |
53c9c5c0 | 1964 | } |
14f7dd63 DW |
1965 | |
1966 | de->namlen = namlen; | |
1967 | de->offset = offset; | |
1968 | de->ino = ino; | |
1969 | de->d_type = d_type; | |
1970 | memcpy(de->name, name, namlen); | |
1971 | buf->used += reclen; | |
1972 | ||
1973 | return 0; | |
1974 | } | |
1975 | ||
2f9092e1 DW |
1976 | static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, |
1977 | struct readdir_cd *cdp, loff_t *offsetp) | |
2628b766 | 1978 | { |
14f7dd63 | 1979 | struct buffered_dirent *de; |
2628b766 | 1980 | int host_err; |
14f7dd63 DW |
1981 | int size; |
1982 | loff_t offset; | |
ac6614b7 AV |
1983 | struct readdir_data buf = { |
1984 | .ctx.actor = nfsd_buffered_filldir, | |
1985 | .dirent = (void *)__get_free_page(GFP_KERNEL) | |
1986 | }; | |
2628b766 | 1987 | |
14f7dd63 | 1988 | if (!buf.dirent) |
2f9092e1 | 1989 | return nfserrno(-ENOMEM); |
14f7dd63 DW |
1990 | |
1991 | offset = *offsetp; | |
2628b766 | 1992 | |
14f7dd63 | 1993 | while (1) { |
496ad9aa | 1994 | struct inode *dir_inode = file_inode(file); |
14f7dd63 DW |
1995 | unsigned int reclen; |
1996 | ||
b726e923 | 1997 | cdp->err = nfserr_eof; /* will be cleared on successful read */ |
14f7dd63 | 1998 | buf.used = 0; |
53c9c5c0 | 1999 | buf.full = 0; |
14f7dd63 | 2000 | |
5c0ba4e0 | 2001 | host_err = iterate_dir(file, &buf.ctx); |
53c9c5c0 AV |
2002 | if (buf.full) |
2003 | host_err = 0; | |
2004 | ||
2005 | if (host_err < 0) | |
14f7dd63 DW |
2006 | break; |
2007 | ||
2008 | size = buf.used; | |
2009 | ||
2010 | if (!size) | |
2011 | break; | |
2012 | ||
2f9092e1 DW |
2013 | /* |
2014 | * Various filldir functions may end up calling back into | |
2015 | * lookup_one_len() and the file system's ->lookup() method. | |
2016 | * These expect i_mutex to be held, as it would within readdir. | |
2017 | */ | |
2018 | host_err = mutex_lock_killable(&dir_inode->i_mutex); | |
2019 | if (host_err) | |
2020 | break; | |
2021 | ||
14f7dd63 DW |
2022 | de = (struct buffered_dirent *)buf.dirent; |
2023 | while (size > 0) { | |
2024 | offset = de->offset; | |
2025 | ||
2026 | if (func(cdp, de->name, de->namlen, de->offset, | |
2027 | de->ino, de->d_type)) | |
2f9092e1 | 2028 | break; |
14f7dd63 DW |
2029 | |
2030 | if (cdp->err != nfs_ok) | |
2f9092e1 | 2031 | break; |
14f7dd63 DW |
2032 | |
2033 | reclen = ALIGN(sizeof(*de) + de->namlen, | |
2034 | sizeof(u64)); | |
2035 | size -= reclen; | |
2036 | de = (struct buffered_dirent *)((char *)de + reclen); | |
2037 | } | |
2f9092e1 DW |
2038 | mutex_unlock(&dir_inode->i_mutex); |
2039 | if (size > 0) /* We bailed out early */ | |
2040 | break; | |
2041 | ||
c002a6c7 | 2042 | offset = vfs_llseek(file, 0, SEEK_CUR); |
14f7dd63 DW |
2043 | } |
2044 | ||
14f7dd63 | 2045 | free_page((unsigned long)(buf.dirent)); |
2628b766 DW |
2046 | |
2047 | if (host_err) | |
2048 | return nfserrno(host_err); | |
14f7dd63 DW |
2049 | |
2050 | *offsetp = offset; | |
2051 | return cdp->err; | |
2628b766 DW |
2052 | } |
2053 | ||
1da177e4 LT |
2054 | /* |
2055 | * Read entries from a directory. | |
2056 | * The NFSv3/4 verifier we ignore for now. | |
2057 | */ | |
6264d69d | 2058 | __be32 |
1da177e4 | 2059 | nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, |
a0ad13ef | 2060 | struct readdir_cd *cdp, filldir_t func) |
1da177e4 | 2061 | { |
6264d69d | 2062 | __be32 err; |
1da177e4 LT |
2063 | struct file *file; |
2064 | loff_t offset = *offsetp; | |
06effdbb BS |
2065 | int may_flags = NFSD_MAY_READ; |
2066 | ||
2067 | /* NFSv2 only supports 32 bit cookies */ | |
2068 | if (rqstp->rq_vers > 2) | |
2069 | may_flags |= NFSD_MAY_64BIT_COOKIE; | |
1da177e4 | 2070 | |
06effdbb | 2071 | err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file); |
1da177e4 LT |
2072 | if (err) |
2073 | goto out; | |
2074 | ||
b108fe6b | 2075 | offset = vfs_llseek(file, offset, SEEK_SET); |
1da177e4 LT |
2076 | if (offset < 0) { |
2077 | err = nfserrno((int)offset); | |
2078 | goto out_close; | |
2079 | } | |
2080 | ||
14f7dd63 | 2081 | err = nfsd_buffered_readdir(file, func, cdp, offsetp); |
1da177e4 LT |
2082 | |
2083 | if (err == nfserr_eof || err == nfserr_toosmall) | |
2084 | err = nfs_ok; /* can still be found in ->err */ | |
2085 | out_close: | |
2086 | nfsd_close(file); | |
2087 | out: | |
2088 | return err; | |
2089 | } | |
2090 | ||
2091 | /* | |
2092 | * Get file system stats | |
2093 | * N.B. After this call fhp needs an fh_put | |
2094 | */ | |
6264d69d | 2095 | __be32 |
04716e66 | 2096 | nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) |
1da177e4 | 2097 | { |
ebabe9a9 CH |
2098 | __be32 err; |
2099 | ||
2100 | err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); | |
f6360efb TI |
2101 | if (!err) { |
2102 | struct path path = { | |
2103 | .mnt = fhp->fh_export->ex_path.mnt, | |
2104 | .dentry = fhp->fh_dentry, | |
2105 | }; | |
2106 | if (vfs_statfs(&path, stat)) | |
2107 | err = nfserr_io; | |
2108 | } | |
1da177e4 LT |
2109 | return err; |
2110 | } | |
2111 | ||
c7d51402 | 2112 | static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) |
e22841c6 | 2113 | { |
c7d51402 | 2114 | return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; |
e22841c6 BF |
2115 | } |
2116 | ||
1da177e4 LT |
2117 | /* |
2118 | * Check for a user's access permissions to this inode. | |
2119 | */ | |
6264d69d | 2120 | __be32 |
0ec757df BF |
2121 | nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, |
2122 | struct dentry *dentry, int acc) | |
1da177e4 LT |
2123 | { |
2124 | struct inode *inode = dentry->d_inode; | |
2125 | int err; | |
2126 | ||
aea93397 | 2127 | if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP) |
1da177e4 LT |
2128 | return 0; |
2129 | #if 0 | |
2130 | dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", | |
2131 | acc, | |
8837abca MS |
2132 | (acc & NFSD_MAY_READ)? " read" : "", |
2133 | (acc & NFSD_MAY_WRITE)? " write" : "", | |
2134 | (acc & NFSD_MAY_EXEC)? " exec" : "", | |
2135 | (acc & NFSD_MAY_SATTR)? " sattr" : "", | |
2136 | (acc & NFSD_MAY_TRUNC)? " trunc" : "", | |
2137 | (acc & NFSD_MAY_LOCK)? " lock" : "", | |
2138 | (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", | |
1da177e4 LT |
2139 | inode->i_mode, |
2140 | IS_IMMUTABLE(inode)? " immut" : "", | |
2141 | IS_APPEND(inode)? " append" : "", | |
2c463e95 | 2142 | __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); |
1da177e4 | 2143 | dprintk(" owner %d/%d user %d/%d\n", |
5cc0a840 | 2144 | inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); |
1da177e4 LT |
2145 | #endif |
2146 | ||
2147 | /* Normally we reject any write/sattr etc access on a read-only file | |
2148 | * system. But if it is IRIX doing check on write-access for a | |
2149 | * device special file, we ignore rofs. | |
2150 | */ | |
8837abca MS |
2151 | if (!(acc & NFSD_MAY_LOCAL_ACCESS)) |
2152 | if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { | |
2c463e95 DH |
2153 | if (exp_rdonly(rqstp, exp) || |
2154 | __mnt_is_readonly(exp->ex_path.mnt)) | |
1da177e4 | 2155 | return nfserr_rofs; |
8837abca | 2156 | if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) |
1da177e4 LT |
2157 | return nfserr_perm; |
2158 | } | |
8837abca | 2159 | if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) |
1da177e4 LT |
2160 | return nfserr_perm; |
2161 | ||
8837abca | 2162 | if (acc & NFSD_MAY_LOCK) { |
1da177e4 LT |
2163 | /* If we cannot rely on authentication in NLM requests, |
2164 | * just allow locks, otherwise require read permission, or | |
2165 | * ownership | |
2166 | */ | |
2167 | if (exp->ex_flags & NFSEXP_NOAUTHNLM) | |
2168 | return 0; | |
2169 | else | |
8837abca | 2170 | acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; |
1da177e4 LT |
2171 | } |
2172 | /* | |
2173 | * The file owner always gets access permission for accesses that | |
2174 | * would normally be checked at open time. This is to make | |
2175 | * file access work even when the client has done a fchmod(fd, 0). | |
2176 | * | |
2177 | * However, `cp foo bar' should fail nevertheless when bar is | |
2178 | * readonly. A sensible way to do this might be to reject all | |
2179 | * attempts to truncate a read-only file, because a creat() call | |
2180 | * always implies file truncation. | |
2181 | * ... but this isn't really fair. A process may reasonably call | |
2182 | * ftruncate on an open file descriptor on a file with perm 000. | |
2183 | * We must trust the client to do permission checking - using "ACCESS" | |
2184 | * with NFSv3. | |
2185 | */ | |
8837abca | 2186 | if ((acc & NFSD_MAY_OWNER_OVERRIDE) && |
6fab8779 | 2187 | uid_eq(inode->i_uid, current_fsuid())) |
1da177e4 LT |
2188 | return 0; |
2189 | ||
8837abca | 2190 | /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ |
f419a2e3 | 2191 | err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); |
1da177e4 LT |
2192 | |
2193 | /* Allow read access to binaries even when mode 111 */ | |
2194 | if (err == -EACCES && S_ISREG(inode->i_mode) && | |
a043226b BF |
2195 | (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) || |
2196 | acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC))) | |
f419a2e3 | 2197 | err = inode_permission(inode, MAY_EXEC); |
1da177e4 LT |
2198 | |
2199 | return err? nfserrno(err) : 0; | |
2200 | } | |
2201 | ||
2202 | void | |
2203 | nfsd_racache_shutdown(void) | |
2204 | { | |
54a66e54 JL |
2205 | struct raparms *raparm, *last_raparm; |
2206 | unsigned int i; | |
2207 | ||
1da177e4 | 2208 | dprintk("nfsd: freeing readahead buffers.\n"); |
54a66e54 JL |
2209 | |
2210 | for (i = 0; i < RAPARM_HASH_SIZE; i++) { | |
2211 | raparm = raparm_hash[i].pb_head; | |
2212 | while(raparm) { | |
2213 | last_raparm = raparm; | |
2214 | raparm = raparm->p_next; | |
2215 | kfree(last_raparm); | |
2216 | } | |
2217 | raparm_hash[i].pb_head = NULL; | |
2218 | } | |
1da177e4 LT |
2219 | } |
2220 | /* | |
2221 | * Initialize readahead param cache | |
2222 | */ | |
2223 | int | |
2224 | nfsd_racache_init(int cache_size) | |
2225 | { | |
2226 | int i; | |
fce1456a GB |
2227 | int j = 0; |
2228 | int nperbucket; | |
54a66e54 | 2229 | struct raparms **raparm = NULL; |
1da177e4 | 2230 | |
fce1456a | 2231 | |
54a66e54 | 2232 | if (raparm_hash[0].pb_head) |
1da177e4 | 2233 | return 0; |
54a66e54 JL |
2234 | nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); |
2235 | if (nperbucket < 2) | |
2236 | nperbucket = 2; | |
2237 | cache_size = nperbucket * RAPARM_HASH_SIZE; | |
4b3bb06b YB |
2238 | |
2239 | dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); | |
54a66e54 JL |
2240 | |
2241 | for (i = 0; i < RAPARM_HASH_SIZE; i++) { | |
4b3bb06b | 2242 | spin_lock_init(&raparm_hash[i].pb_lock); |
54a66e54 JL |
2243 | |
2244 | raparm = &raparm_hash[i].pb_head; | |
2245 | for (j = 0; j < nperbucket; j++) { | |
2246 | *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); | |
2247 | if (!*raparm) | |
2248 | goto out_nomem; | |
2249 | raparm = &(*raparm)->p_next; | |
2250 | } | |
2251 | *raparm = NULL; | |
4b3bb06b YB |
2252 | } |
2253 | ||
1da177e4 LT |
2254 | nfsdstats.ra_size = cache_size; |
2255 | return 0; | |
54a66e54 JL |
2256 | |
2257 | out_nomem: | |
2258 | dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); | |
2259 | nfsd_racache_shutdown(); | |
2260 | return -ENOMEM; | |
1da177e4 | 2261 | } |
a257cdd0 AG |
2262 | |
2263 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) | |
2264 | struct posix_acl * | |
2265 | nfsd_get_posix_acl(struct svc_fh *fhp, int type) | |
2266 | { | |
2267 | struct inode *inode = fhp->fh_dentry->d_inode; | |
2268 | char *name; | |
2269 | void *value = NULL; | |
2270 | ssize_t size; | |
2271 | struct posix_acl *acl; | |
2272 | ||
5be196e5 CH |
2273 | if (!IS_POSIXACL(inode)) |
2274 | return ERR_PTR(-EOPNOTSUPP); | |
2275 | ||
2276 | switch (type) { | |
2277 | case ACL_TYPE_ACCESS: | |
2278 | name = POSIX_ACL_XATTR_ACCESS; | |
2279 | break; | |
2280 | case ACL_TYPE_DEFAULT: | |
2281 | name = POSIX_ACL_XATTR_DEFAULT; | |
2282 | break; | |
2283 | default: | |
a257cdd0 | 2284 | return ERR_PTR(-EOPNOTSUPP); |
a257cdd0 AG |
2285 | } |
2286 | ||
5be196e5 CH |
2287 | size = nfsd_getxattr(fhp->fh_dentry, name, &value); |
2288 | if (size < 0) | |
2289 | return ERR_PTR(size); | |
a257cdd0 | 2290 | |
5f3a4a28 | 2291 | acl = posix_acl_from_xattr(&init_user_ns, value, size); |
a257cdd0 AG |
2292 | kfree(value); |
2293 | return acl; | |
2294 | } | |
2295 | ||
2296 | int | |
2297 | nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) | |
2298 | { | |
2299 | struct inode *inode = fhp->fh_dentry->d_inode; | |
2300 | char *name; | |
2301 | void *value = NULL; | |
2302 | size_t size; | |
2303 | int error; | |
2304 | ||
acfa4380 | 2305 | if (!IS_POSIXACL(inode) || |
a257cdd0 AG |
2306 | !inode->i_op->setxattr || !inode->i_op->removexattr) |
2307 | return -EOPNOTSUPP; | |
2308 | switch(type) { | |
2309 | case ACL_TYPE_ACCESS: | |
334a13ec | 2310 | name = POSIX_ACL_XATTR_ACCESS; |
a257cdd0 AG |
2311 | break; |
2312 | case ACL_TYPE_DEFAULT: | |
334a13ec | 2313 | name = POSIX_ACL_XATTR_DEFAULT; |
a257cdd0 AG |
2314 | break; |
2315 | default: | |
2316 | return -EOPNOTSUPP; | |
2317 | } | |
2318 | ||
2319 | if (acl && acl->a_count) { | |
334a13ec | 2320 | size = posix_acl_xattr_size(acl->a_count); |
a257cdd0 AG |
2321 | value = kmalloc(size, GFP_KERNEL); |
2322 | if (!value) | |
2323 | return -ENOMEM; | |
5f3a4a28 | 2324 | error = posix_acl_to_xattr(&init_user_ns, acl, value, size); |
9ccfc29c | 2325 | if (error < 0) |
a257cdd0 | 2326 | goto getout; |
9ccfc29c | 2327 | size = error; |
a257cdd0 AG |
2328 | } else |
2329 | size = 0; | |
2330 | ||
bad0dcff | 2331 | error = fh_want_write(fhp); |
18f335af DH |
2332 | if (error) |
2333 | goto getout; | |
a257cdd0 | 2334 | if (size) |
5be196e5 | 2335 | error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); |
a257cdd0 AG |
2336 | else { |
2337 | if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) | |
2338 | error = 0; | |
2339 | else { | |
5be196e5 | 2340 | error = vfs_removexattr(fhp->fh_dentry, name); |
a257cdd0 AG |
2341 | if (error == -ENODATA) |
2342 | error = 0; | |
2343 | } | |
2344 | } | |
bad0dcff | 2345 | fh_drop_write(fhp); |
a257cdd0 AG |
2346 | |
2347 | getout: | |
2348 | kfree(value); | |
2349 | return error; | |
2350 | } | |
2351 | #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ |