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> |
95d871f0 | 19 | #include <linux/falloc.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> | |
5a0e3ad6 | 28 | #include <linux/slab.h> |
9a74af21 | 29 | #include <asm/uaccess.h> |
f501912a BM |
30 | #include <linux/exportfs.h> |
31 | #include <linux/writeback.h> | |
18032ca0 | 32 | #include <linux/security.h> |
9a74af21 BH |
33 | |
34 | #ifdef CONFIG_NFSD_V3 | |
35 | #include "xdr3.h" | |
36 | #endif /* CONFIG_NFSD_V3 */ | |
37 | ||
5be196e5 | 38 | #ifdef CONFIG_NFSD_V4 |
2ca72e17 BF |
39 | #include "acl.h" |
40 | #include "idmap.h" | |
1da177e4 LT |
41 | #endif /* CONFIG_NFSD_V4 */ |
42 | ||
9a74af21 BH |
43 | #include "nfsd.h" |
44 | #include "vfs.h" | |
1da177e4 LT |
45 | |
46 | #define NFSDDBG_FACILITY NFSDDBG_FILEOP | |
1da177e4 LT |
47 | |
48 | ||
1da177e4 LT |
49 | /* |
50 | * This is a cache of readahead params that help us choose the proper | |
51 | * readahead strategy. Initially, we set all readahead parameters to 0 | |
52 | * and let the VFS handle things. | |
53 | * If you increase the number of cached files very much, you'll need to | |
54 | * add a hash table here. | |
55 | */ | |
56 | struct raparms { | |
57 | struct raparms *p_next; | |
58 | unsigned int p_count; | |
59 | ino_t p_ino; | |
60 | dev_t p_dev; | |
61 | int p_set; | |
62 | struct file_ra_state p_ra; | |
fce1456a | 63 | unsigned int p_hindex; |
1da177e4 LT |
64 | }; |
65 | ||
fce1456a GB |
66 | struct raparm_hbucket { |
67 | struct raparms *pb_head; | |
68 | spinlock_t pb_lock; | |
69 | } ____cacheline_aligned_in_smp; | |
70 | ||
fce1456a GB |
71 | #define RAPARM_HASH_BITS 4 |
72 | #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) | |
73 | #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) | |
74 | static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; | |
1da177e4 LT |
75 | |
76 | /* | |
77 | * Called from nfsd_lookup and encode_dirent. Check if we have crossed | |
78 | * a mount point. | |
e0bb89ef | 79 | * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, |
1da177e4 LT |
80 | * or nfs_ok having possibly changed *dpp and *expp |
81 | */ | |
82 | int | |
83 | nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, | |
84 | struct svc_export **expp) | |
85 | { | |
86 | struct svc_export *exp = *expp, *exp2 = NULL; | |
87 | struct dentry *dentry = *dpp; | |
91c9fa8f AV |
88 | struct path path = {.mnt = mntget(exp->ex_path.mnt), |
89 | .dentry = dget(dentry)}; | |
6264d69d | 90 | int err = 0; |
1da177e4 | 91 | |
7cc90cc3 | 92 | err = follow_down(&path); |
cc53ce53 DH |
93 | if (err < 0) |
94 | goto out; | |
1da177e4 | 95 | |
91c9fa8f | 96 | exp2 = rqst_exp_get_by_name(rqstp, &path); |
1da177e4 | 97 | if (IS_ERR(exp2)) { |
3b6cee7b BF |
98 | err = PTR_ERR(exp2); |
99 | /* | |
100 | * We normally allow NFS clients to continue | |
101 | * "underneath" a mountpoint that is not exported. | |
102 | * The exception is V4ROOT, where no traversal is ever | |
103 | * allowed without an explicit export of the new | |
104 | * directory. | |
105 | */ | |
106 | if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT)) | |
107 | err = 0; | |
91c9fa8f | 108 | path_put(&path); |
1da177e4 LT |
109 | goto out; |
110 | } | |
3c394dda SD |
111 | if (nfsd_v4client(rqstp) || |
112 | (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { | |
1da177e4 | 113 | /* successfully crossed mount point */ |
1644ccc8 | 114 | /* |
91c9fa8f AV |
115 | * This is subtle: path.dentry is *not* on path.mnt |
116 | * at this point. The only reason we are safe is that | |
117 | * original mnt is pinned down by exp, so we should | |
118 | * put path *before* putting exp | |
1644ccc8 | 119 | */ |
91c9fa8f AV |
120 | *dpp = path.dentry; |
121 | path.dentry = dentry; | |
1644ccc8 | 122 | *expp = exp2; |
91c9fa8f | 123 | exp2 = exp; |
1da177e4 | 124 | } |
91c9fa8f AV |
125 | path_put(&path); |
126 | exp_put(exp2); | |
1da177e4 LT |
127 | out: |
128 | return err; | |
129 | } | |
130 | ||
289ede45 BF |
131 | static void follow_to_parent(struct path *path) |
132 | { | |
133 | struct dentry *dp; | |
134 | ||
135 | while (path->dentry == path->mnt->mnt_root && follow_up(path)) | |
136 | ; | |
137 | dp = dget_parent(path->dentry); | |
138 | dput(path->dentry); | |
139 | path->dentry = dp; | |
140 | } | |
141 | ||
142 | static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp) | |
143 | { | |
144 | struct svc_export *exp2; | |
145 | struct path path = {.mnt = mntget((*exp)->ex_path.mnt), | |
146 | .dentry = dget(dparent)}; | |
147 | ||
148 | follow_to_parent(&path); | |
149 | ||
150 | exp2 = rqst_exp_parent(rqstp, &path); | |
151 | if (PTR_ERR(exp2) == -ENOENT) { | |
152 | *dentryp = dget(dparent); | |
153 | } else if (IS_ERR(exp2)) { | |
154 | path_put(&path); | |
155 | return PTR_ERR(exp2); | |
156 | } else { | |
157 | *dentryp = dget(path.dentry); | |
158 | exp_put(*exp); | |
159 | *exp = exp2; | |
160 | } | |
161 | path_put(&path); | |
162 | return 0; | |
163 | } | |
164 | ||
82ead7fe BF |
165 | /* |
166 | * For nfsd purposes, we treat V4ROOT exports as though there was an | |
167 | * export at *every* directory. | |
168 | */ | |
3227fa41 | 169 | int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp) |
82ead7fe BF |
170 | { |
171 | if (d_mountpoint(dentry)) | |
172 | return 1; | |
11fcee02 TM |
173 | if (nfsd4_is_junction(dentry)) |
174 | return 1; | |
82ead7fe BF |
175 | if (!(exp->ex_flags & NFSEXP_V4ROOT)) |
176 | return 0; | |
2b0143b5 | 177 | return d_inode(dentry) != NULL; |
82ead7fe BF |
178 | } |
179 | ||
6264d69d | 180 | __be32 |
6c0a654d | 181 | nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp, |
5a022fc8 | 182 | const char *name, unsigned int len, |
6c0a654d | 183 | struct svc_export **exp_ret, struct dentry **dentry_ret) |
1da177e4 LT |
184 | { |
185 | struct svc_export *exp; | |
186 | struct dentry *dparent; | |
187 | struct dentry *dentry; | |
6264d69d | 188 | int host_err; |
1da177e4 LT |
189 | |
190 | dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); | |
191 | ||
1da177e4 | 192 | dparent = fhp->fh_dentry; |
bf18f163 | 193 | exp = exp_get(fhp->fh_export); |
1da177e4 | 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 { | |
4335723e BF |
210 | /* |
211 | * In the nfsd4_open() case, this may be held across | |
212 | * subsequent open and delegation acquisition which may | |
213 | * need to take the child's i_mutex: | |
214 | */ | |
215 | fh_lock_nested(fhp, I_MUTEX_PARENT); | |
1da177e4 | 216 | dentry = lookup_one_len(name, dparent, len); |
6264d69d | 217 | host_err = PTR_ERR(dentry); |
1da177e4 LT |
218 | if (IS_ERR(dentry)) |
219 | goto out_nfserr; | |
220 | /* | |
221 | * check if we have crossed a mount point ... | |
222 | */ | |
82ead7fe | 223 | if (nfsd_mountpoint(dentry, exp)) { |
6264d69d | 224 | if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { |
1da177e4 LT |
225 | dput(dentry); |
226 | goto out_nfserr; | |
227 | } | |
228 | } | |
229 | } | |
6c0a654d BF |
230 | *dentry_ret = dentry; |
231 | *exp_ret = exp; | |
232 | return 0; | |
233 | ||
234 | out_nfserr: | |
235 | exp_put(exp); | |
236 | return nfserrno(host_err); | |
237 | } | |
238 | ||
239 | /* | |
240 | * Look up one component of a pathname. | |
241 | * N.B. After this call _both_ fhp and resfh need an fh_put | |
242 | * | |
243 | * If the lookup would cross a mountpoint, and the mounted filesystem | |
244 | * is exported to the client with NFSEXP_NOHIDE, then the lookup is | |
245 | * accepted as it stands and the mounted directory is | |
246 | * returned. Otherwise the covered directory is returned. | |
247 | * NOTE: this mountpoint crossing is not supported properly by all | |
248 | * clients and is explicitly disallowed for NFSv3 | |
249 | * NeilBrown <neilb@cse.unsw.edu.au> | |
250 | */ | |
251 | __be32 | |
252 | nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, | |
5a022fc8 | 253 | unsigned int len, struct svc_fh *resfh) |
6c0a654d BF |
254 | { |
255 | struct svc_export *exp; | |
256 | struct dentry *dentry; | |
257 | __be32 err; | |
258 | ||
29a78a3e BF |
259 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); |
260 | if (err) | |
261 | return err; | |
6c0a654d BF |
262 | err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); |
263 | if (err) | |
264 | return err; | |
32c1eb0c AA |
265 | err = check_nfsd_access(exp, rqstp); |
266 | if (err) | |
267 | goto out; | |
1da177e4 LT |
268 | /* |
269 | * Note: we compose the file handle now, but as the | |
270 | * dentry may be negative, it may need to be updated. | |
271 | */ | |
272 | err = fh_compose(resfh, exp, dentry, fhp); | |
2b0143b5 | 273 | if (!err && d_really_is_negative(dentry)) |
1da177e4 | 274 | err = nfserr_noent; |
32c1eb0c | 275 | out: |
1da177e4 | 276 | dput(dentry); |
1da177e4 LT |
277 | exp_put(exp); |
278 | return err; | |
1da177e4 LT |
279 | } |
280 | ||
f501912a BM |
281 | /* |
282 | * Commit metadata changes to stable storage. | |
283 | */ | |
284 | static int | |
285 | commit_metadata(struct svc_fh *fhp) | |
286 | { | |
2b0143b5 | 287 | struct inode *inode = d_inode(fhp->fh_dentry); |
f501912a | 288 | const struct export_operations *export_ops = inode->i_sb->s_export_op; |
f501912a BM |
289 | |
290 | if (!EX_ISSYNC(fhp->fh_export)) | |
291 | return 0; | |
292 | ||
c3765016 CH |
293 | if (export_ops->commit_metadata) |
294 | return export_ops->commit_metadata(inode); | |
295 | return sync_inode_metadata(inode, 1); | |
f501912a | 296 | } |
6c0a654d | 297 | |
1da177e4 | 298 | /* |
818e5a22 CH |
299 | * Go over the attributes and take care of the small differences between |
300 | * NFS semantics and what Linux expects. | |
1da177e4 | 301 | */ |
818e5a22 CH |
302 | static void |
303 | nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap) | |
1da177e4 | 304 | { |
ca456252 | 305 | /* sanitize the mode change */ |
1da177e4 LT |
306 | if (iap->ia_valid & ATTR_MODE) { |
307 | iap->ia_mode &= S_IALLUGO; | |
dee3209d | 308 | iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); |
ca456252 JL |
309 | } |
310 | ||
311 | /* Revoke setuid/setgid on chown */ | |
0953e620 | 312 | if (!S_ISDIR(inode->i_mode) && |
c4fa6d7c | 313 | ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) { |
ca456252 JL |
314 | iap->ia_valid |= ATTR_KILL_PRIV; |
315 | if (iap->ia_valid & ATTR_MODE) { | |
316 | /* we're setting mode too, just clear the s*id bits */ | |
8a0ce7d9 | 317 | iap->ia_mode &= ~S_ISUID; |
ca456252 JL |
318 | if (iap->ia_mode & S_IXGRP) |
319 | iap->ia_mode &= ~S_ISGID; | |
320 | } else { | |
321 | /* set ATTR_KILL_* bits and let VFS handle it */ | |
322 | iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); | |
8a0ce7d9 | 323 | } |
1da177e4 | 324 | } |
818e5a22 CH |
325 | } |
326 | ||
327 | static __be32 | |
328 | nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
329 | struct iattr *iap) | |
330 | { | |
2b0143b5 | 331 | struct inode *inode = d_inode(fhp->fh_dentry); |
818e5a22 CH |
332 | int host_err; |
333 | ||
334 | if (iap->ia_size < inode->i_size) { | |
335 | __be32 err; | |
336 | ||
337 | err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, | |
338 | NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE); | |
339 | if (err) | |
340 | return err; | |
341 | } | |
342 | ||
343 | host_err = get_write_access(inode); | |
344 | if (host_err) | |
345 | goto out_nfserrno; | |
346 | ||
347 | host_err = locks_verify_truncate(inode, NULL, iap->ia_size); | |
348 | if (host_err) | |
349 | goto out_put_write_access; | |
350 | return 0; | |
351 | ||
352 | out_put_write_access: | |
353 | put_write_access(inode); | |
354 | out_nfserrno: | |
355 | return nfserrno(host_err); | |
356 | } | |
357 | ||
358 | /* | |
359 | * Set various file attributes. After this call fhp needs an fh_put. | |
360 | */ | |
361 | __be32 | |
362 | nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, | |
363 | int check_guard, time_t guardtime) | |
364 | { | |
365 | struct dentry *dentry; | |
366 | struct inode *inode; | |
367 | int accmode = NFSD_MAY_SATTR; | |
368 | umode_t ftype = 0; | |
369 | __be32 err; | |
370 | int host_err; | |
9f67f189 | 371 | bool get_write_count; |
818e5a22 CH |
372 | int size_change = 0; |
373 | ||
374 | if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) | |
375 | accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; | |
376 | if (iap->ia_valid & ATTR_SIZE) | |
377 | ftype = S_IFREG; | |
378 | ||
9f67f189 BF |
379 | /* Callers that do fh_verify should do the fh_want_write: */ |
380 | get_write_count = !fhp->fh_dentry; | |
381 | ||
818e5a22 CH |
382 | /* Get inode */ |
383 | err = fh_verify(rqstp, fhp, ftype, accmode); | |
384 | if (err) | |
385 | goto out; | |
9f67f189 BF |
386 | if (get_write_count) { |
387 | host_err = fh_want_write(fhp); | |
388 | if (host_err) | |
389 | return nfserrno(host_err); | |
390 | } | |
818e5a22 CH |
391 | |
392 | dentry = fhp->fh_dentry; | |
2b0143b5 | 393 | inode = d_inode(dentry); |
818e5a22 CH |
394 | |
395 | /* Ignore any mode updates on symlinks */ | |
396 | if (S_ISLNK(inode->i_mode)) | |
397 | iap->ia_valid &= ~ATTR_MODE; | |
398 | ||
399 | if (!iap->ia_valid) | |
400 | goto out; | |
1da177e4 | 401 | |
818e5a22 CH |
402 | nfsd_sanitize_attrs(inode, iap); |
403 | ||
404 | /* | |
405 | * The size case is special, it changes the file in addition to the | |
406 | * attributes. | |
407 | */ | |
408 | if (iap->ia_valid & ATTR_SIZE) { | |
409 | err = nfsd_get_write_access(rqstp, fhp, iap); | |
410 | if (err) | |
411 | goto out; | |
412 | size_change = 1; | |
f0c63124 CH |
413 | |
414 | /* | |
415 | * RFC5661, Section 18.30.4: | |
416 | * Changing the size of a file with SETATTR indirectly | |
417 | * changes the time_modify and change attributes. | |
418 | * | |
419 | * (and similar for the older RFCs) | |
420 | */ | |
421 | if (iap->ia_size != i_size_read(inode)) | |
422 | iap->ia_valid |= ATTR_MTIME; | |
818e5a22 | 423 | } |
1da177e4 LT |
424 | |
425 | iap->ia_valid |= ATTR_CTIME; | |
426 | ||
987da479 CH |
427 | if (check_guard && guardtime != inode->i_ctime.tv_sec) { |
428 | err = nfserr_notsync; | |
429 | goto out_put_write_access; | |
1da177e4 | 430 | } |
987da479 | 431 | |
987da479 CH |
432 | fh_lock(fhp); |
433 | host_err = notify_change(dentry, iap, NULL); | |
434 | fh_unlock(fhp); | |
1406b916 | 435 | err = nfserrno(host_err); |
987da479 | 436 | |
987da479 | 437 | out_put_write_access: |
1da177e4 LT |
438 | if (size_change) |
439 | put_write_access(inode); | |
440 | if (!err) | |
722b620d | 441 | err = nfserrno(commit_metadata(fhp)); |
1da177e4 LT |
442 | out: |
443 | return err; | |
1da177e4 LT |
444 | } |
445 | ||
5be196e5 | 446 | #if defined(CONFIG_NFSD_V4) |
9b4146e8 CL |
447 | /* |
448 | * NFS junction information is stored in an extended attribute. | |
449 | */ | |
450 | #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs" | |
451 | ||
452 | /** | |
453 | * nfsd4_is_junction - Test if an object could be an NFS junction | |
454 | * | |
455 | * @dentry: object to test | |
456 | * | |
457 | * Returns 1 if "dentry" appears to contain NFS junction information. | |
458 | * Otherwise 0 is returned. | |
459 | */ | |
11fcee02 TM |
460 | int nfsd4_is_junction(struct dentry *dentry) |
461 | { | |
2b0143b5 | 462 | struct inode *inode = d_inode(dentry); |
11fcee02 TM |
463 | |
464 | if (inode == NULL) | |
465 | return 0; | |
466 | if (inode->i_mode & S_IXUGO) | |
467 | return 0; | |
468 | if (!(inode->i_mode & S_ISVTX)) | |
469 | return 0; | |
9b4146e8 | 470 | if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0) |
11fcee02 TM |
471 | return 0; |
472 | return 1; | |
473 | } | |
18032ca0 DQ |
474 | #ifdef CONFIG_NFSD_V4_SECURITY_LABEL |
475 | __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
476 | struct xdr_netobj *label) | |
477 | { | |
478 | __be32 error; | |
479 | int host_error; | |
480 | struct dentry *dentry; | |
481 | ||
482 | error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); | |
483 | if (error) | |
484 | return error; | |
485 | ||
486 | dentry = fhp->fh_dentry; | |
487 | ||
2b0143b5 | 488 | mutex_lock(&d_inode(dentry)->i_mutex); |
18032ca0 | 489 | host_error = security_inode_setsecctx(dentry, label->data, label->len); |
2b0143b5 | 490 | mutex_unlock(&d_inode(dentry)->i_mutex); |
18032ca0 DQ |
491 | return nfserrno(host_error); |
492 | } | |
493 | #else | |
494 | __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
495 | struct xdr_netobj *label) | |
496 | { | |
497 | return nfserr_notsupp; | |
498 | } | |
499 | #endif | |
500 | ||
95d871f0 AS |
501 | __be32 nfsd4_vfs_fallocate(struct svc_rqst *rqstp, struct svc_fh *fhp, |
502 | struct file *file, loff_t offset, loff_t len, | |
503 | int flags) | |
504 | { | |
505 | __be32 err; | |
506 | int error; | |
507 | ||
508 | if (!S_ISREG(file_inode(file)->i_mode)) | |
509 | return nfserr_inval; | |
510 | ||
511 | err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, NFSD_MAY_WRITE); | |
512 | if (err) | |
513 | return err; | |
514 | ||
515 | error = vfs_fallocate(file, flags, offset, len); | |
516 | if (!error) | |
517 | error = commit_metadata(fhp); | |
518 | ||
519 | return nfserrno(error); | |
520 | } | |
6a85d6c7 | 521 | #endif /* defined(CONFIG_NFSD_V4) */ |
1da177e4 LT |
522 | |
523 | #ifdef CONFIG_NFSD_V3 | |
524 | /* | |
525 | * Check server access rights to a file system object | |
526 | */ | |
527 | struct accessmap { | |
528 | u32 access; | |
529 | int how; | |
530 | }; | |
531 | static struct accessmap nfs3_regaccess[] = { | |
8837abca MS |
532 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
533 | { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, | |
534 | { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, | |
535 | { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, | |
1da177e4 LT |
536 | |
537 | { 0, 0 } | |
538 | }; | |
539 | ||
540 | static struct accessmap nfs3_diraccess[] = { | |
8837abca MS |
541 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
542 | { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, | |
543 | { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, | |
544 | { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, | |
545 | { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, | |
1da177e4 LT |
546 | |
547 | { 0, 0 } | |
548 | }; | |
549 | ||
550 | static struct accessmap nfs3_anyaccess[] = { | |
551 | /* Some clients - Solaris 2.6 at least, make an access call | |
552 | * to the server to check for access for things like /dev/null | |
553 | * (which really, the server doesn't care about). So | |
554 | * We provide simple access checking for them, looking | |
555 | * mainly at mode bits, and we make sure to ignore read-only | |
556 | * filesystem checks | |
557 | */ | |
8837abca MS |
558 | { NFS3_ACCESS_READ, NFSD_MAY_READ }, |
559 | { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, | |
560 | { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, | |
561 | { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, | |
1da177e4 LT |
562 | |
563 | { 0, 0 } | |
564 | }; | |
565 | ||
6264d69d | 566 | __be32 |
1da177e4 LT |
567 | nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) |
568 | { | |
569 | struct accessmap *map; | |
570 | struct svc_export *export; | |
571 | struct dentry *dentry; | |
572 | u32 query, result = 0, sresult = 0; | |
6264d69d | 573 | __be32 error; |
1da177e4 | 574 | |
8837abca | 575 | error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); |
1da177e4 LT |
576 | if (error) |
577 | goto out; | |
578 | ||
579 | export = fhp->fh_export; | |
580 | dentry = fhp->fh_dentry; | |
581 | ||
e36cb0b8 | 582 | if (d_is_reg(dentry)) |
1da177e4 | 583 | map = nfs3_regaccess; |
e36cb0b8 | 584 | else if (d_is_dir(dentry)) |
1da177e4 LT |
585 | map = nfs3_diraccess; |
586 | else | |
587 | map = nfs3_anyaccess; | |
588 | ||
589 | ||
590 | query = *access; | |
591 | for (; map->access; map++) { | |
592 | if (map->access & query) { | |
6264d69d | 593 | __be32 err2; |
1da177e4 LT |
594 | |
595 | sresult |= map->access; | |
596 | ||
0ec757df | 597 | err2 = nfsd_permission(rqstp, export, dentry, map->how); |
1da177e4 LT |
598 | switch (err2) { |
599 | case nfs_ok: | |
600 | result |= map->access; | |
601 | break; | |
602 | ||
603 | /* the following error codes just mean the access was not allowed, | |
604 | * rather than an error occurred */ | |
605 | case nfserr_rofs: | |
606 | case nfserr_acces: | |
607 | case nfserr_perm: | |
608 | /* simply don't "or" in the access bit. */ | |
609 | break; | |
610 | default: | |
611 | error = err2; | |
612 | goto out; | |
613 | } | |
614 | } | |
615 | } | |
616 | *access = result; | |
617 | if (supported) | |
618 | *supported = sresult; | |
619 | ||
620 | out: | |
621 | return error; | |
622 | } | |
623 | #endif /* CONFIG_NFSD_V3 */ | |
624 | ||
105f4622 BF |
625 | static int nfsd_open_break_lease(struct inode *inode, int access) |
626 | { | |
627 | unsigned int mode; | |
1da177e4 | 628 | |
105f4622 BF |
629 | if (access & NFSD_MAY_NOT_BREAK_LEASE) |
630 | return 0; | |
631 | mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY; | |
632 | return break_lease(inode, mode | O_NONBLOCK); | |
633 | } | |
1da177e4 LT |
634 | |
635 | /* | |
636 | * Open an existing file or directory. | |
999448a8 BS |
637 | * The may_flags argument indicates the type of open (read/write/lock) |
638 | * and additional flags. | |
1da177e4 LT |
639 | * N.B. After this call fhp needs an fh_put |
640 | */ | |
6264d69d | 641 | __be32 |
175a4eb7 | 642 | nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, |
999448a8 | 643 | int may_flags, struct file **filp) |
1da177e4 | 644 | { |
765927b2 | 645 | struct path path; |
1da177e4 | 646 | struct inode *inode; |
8519f994 | 647 | struct file *file; |
6264d69d AV |
648 | int flags = O_RDONLY|O_LARGEFILE; |
649 | __be32 err; | |
91885258 | 650 | int host_err = 0; |
1da177e4 | 651 | |
e0e81739 DH |
652 | validate_process_creds(); |
653 | ||
1da177e4 LT |
654 | /* |
655 | * If we get here, then the client has already done an "open", | |
656 | * and (hopefully) checked permission - so allow OWNER_OVERRIDE | |
657 | * in case a chmod has now revoked permission. | |
d91d0b56 BF |
658 | * |
659 | * Arguably we should also allow the owner override for | |
660 | * directories, but we never have and it doesn't seem to have | |
661 | * caused anyone a problem. If we were to change this, note | |
662 | * also that our filldir callbacks would need a variant of | |
663 | * lookup_one_len that doesn't check permissions. | |
1da177e4 | 664 | */ |
d91d0b56 BF |
665 | if (type == S_IFREG) |
666 | may_flags |= NFSD_MAY_OWNER_OVERRIDE; | |
667 | err = fh_verify(rqstp, fhp, type, may_flags); | |
1da177e4 LT |
668 | if (err) |
669 | goto out; | |
670 | ||
765927b2 AV |
671 | path.mnt = fhp->fh_export->ex_path.mnt; |
672 | path.dentry = fhp->fh_dentry; | |
2b0143b5 | 673 | inode = d_inode(path.dentry); |
1da177e4 LT |
674 | |
675 | /* Disallow write access to files with the append-only bit set | |
676 | * or any access when mandatory locking enabled | |
677 | */ | |
678 | err = nfserr_perm; | |
999448a8 | 679 | if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE)) |
1da177e4 | 680 | goto out; |
5e7fc436 BF |
681 | /* |
682 | * We must ignore files (but only files) which might have mandatory | |
683 | * locks on them because there is no way to know if the accesser has | |
684 | * the lock. | |
685 | */ | |
686 | if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) | |
1da177e4 LT |
687 | goto out; |
688 | ||
689 | if (!inode->i_fop) | |
690 | goto out; | |
691 | ||
999448a8 | 692 | host_err = nfsd_open_break_lease(inode, may_flags); |
6264d69d | 693 | if (host_err) /* NOMEM or WOULDBLOCK */ |
1da177e4 LT |
694 | goto out_nfserr; |
695 | ||
999448a8 BS |
696 | if (may_flags & NFSD_MAY_WRITE) { |
697 | if (may_flags & NFSD_MAY_READ) | |
9ecb6a08 BF |
698 | flags = O_RDWR|O_LARGEFILE; |
699 | else | |
700 | flags = O_WRONLY|O_LARGEFILE; | |
1da177e4 | 701 | } |
999448a8 | 702 | |
8519f994 KM |
703 | file = dentry_open(&path, flags, current_cred()); |
704 | if (IS_ERR(file)) { | |
705 | host_err = PTR_ERR(file); | |
706 | goto out_nfserr; | |
707 | } | |
708 | ||
5e40d331 | 709 | host_err = ima_file_check(file, may_flags, 0); |
8519f994 | 710 | if (host_err) { |
fd891454 | 711 | fput(file); |
8519f994 | 712 | goto out_nfserr; |
06effdbb BS |
713 | } |
714 | ||
8519f994 KM |
715 | if (may_flags & NFSD_MAY_64BIT_COOKIE) |
716 | file->f_mode |= FMODE_64BITHASH; | |
717 | else | |
718 | file->f_mode |= FMODE_32BITHASH; | |
719 | ||
720 | *filp = file; | |
1da177e4 | 721 | out_nfserr: |
6264d69d | 722 | err = nfserrno(host_err); |
1da177e4 | 723 | out: |
e0e81739 | 724 | validate_process_creds(); |
1da177e4 LT |
725 | return err; |
726 | } | |
727 | ||
1da177e4 LT |
728 | /* |
729 | * Obtain the readahead parameters for the file | |
730 | * specified by (dev, ino). | |
731 | */ | |
1da177e4 LT |
732 | |
733 | static inline struct raparms * | |
734 | nfsd_get_raparms(dev_t dev, ino_t ino) | |
735 | { | |
736 | struct raparms *ra, **rap, **frap = NULL; | |
737 | int depth = 0; | |
fce1456a GB |
738 | unsigned int hash; |
739 | struct raparm_hbucket *rab; | |
740 | ||
741 | hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; | |
742 | rab = &raparm_hash[hash]; | |
1da177e4 | 743 | |
fce1456a GB |
744 | spin_lock(&rab->pb_lock); |
745 | for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { | |
1da177e4 LT |
746 | if (ra->p_ino == ino && ra->p_dev == dev) |
747 | goto found; | |
748 | depth++; | |
749 | if (ra->p_count == 0) | |
750 | frap = rap; | |
751 | } | |
3aa6e0aa | 752 | depth = nfsdstats.ra_size; |
1da177e4 | 753 | if (!frap) { |
fce1456a | 754 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
755 | return NULL; |
756 | } | |
757 | rap = frap; | |
758 | ra = *frap; | |
759 | ra->p_dev = dev; | |
760 | ra->p_ino = ino; | |
761 | ra->p_set = 0; | |
fce1456a | 762 | ra->p_hindex = hash; |
1da177e4 | 763 | found: |
fce1456a | 764 | if (rap != &rab->pb_head) { |
1da177e4 | 765 | *rap = ra->p_next; |
fce1456a GB |
766 | ra->p_next = rab->pb_head; |
767 | rab->pb_head = ra; | |
1da177e4 LT |
768 | } |
769 | ra->p_count++; | |
770 | nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; | |
fce1456a | 771 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
772 | return ra; |
773 | } | |
774 | ||
775 | /* | |
cf8208d0 JA |
776 | * Grab and keep cached pages associated with a file in the svc_rqst |
777 | * so that they can be passed to the network sendmsg/sendpage routines | |
778 | * directly. They will be released after the sending has completed. | |
1da177e4 LT |
779 | */ |
780 | static int | |
cf8208d0 JA |
781 | nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
782 | struct splice_desc *sd) | |
1da177e4 | 783 | { |
cf8208d0 | 784 | struct svc_rqst *rqstp = sd->u.data; |
afc59400 | 785 | struct page **pp = rqstp->rq_next_page; |
cf8208d0 JA |
786 | struct page *page = buf->page; |
787 | size_t size; | |
cf8208d0 JA |
788 | |
789 | size = sd->len; | |
1da177e4 LT |
790 | |
791 | if (rqstp->rq_res.page_len == 0) { | |
792 | get_page(page); | |
afc59400 BF |
793 | put_page(*rqstp->rq_next_page); |
794 | *(rqstp->rq_next_page++) = page; | |
cf8208d0 | 795 | rqstp->rq_res.page_base = buf->offset; |
1da177e4 | 796 | rqstp->rq_res.page_len = size; |
44524359 | 797 | } else if (page != pp[-1]) { |
1da177e4 | 798 | get_page(page); |
afc59400 BF |
799 | if (*rqstp->rq_next_page) |
800 | put_page(*rqstp->rq_next_page); | |
801 | *(rqstp->rq_next_page++) = page; | |
1da177e4 | 802 | rqstp->rq_res.page_len += size; |
44524359 | 803 | } else |
1da177e4 | 804 | rqstp->rq_res.page_len += size; |
1da177e4 | 805 | |
1da177e4 LT |
806 | return size; |
807 | } | |
808 | ||
cf8208d0 JA |
809 | static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, |
810 | struct splice_desc *sd) | |
811 | { | |
812 | return __splice_from_pipe(pipe, sd, nfsd_splice_actor); | |
813 | } | |
814 | ||
e2afc819 JL |
815 | static __be32 |
816 | nfsd_finish_read(struct file *file, unsigned long *count, int host_err) | |
1da177e4 | 817 | { |
6264d69d AV |
818 | if (host_err >= 0) { |
819 | nfsdstats.io_read += host_err; | |
820 | *count = host_err; | |
2a12a9d7 | 821 | fsnotify_access(file); |
dc97618d | 822 | return 0; |
1da177e4 | 823 | } else |
dc97618d BF |
824 | return nfserrno(host_err); |
825 | } | |
826 | ||
e2afc819 | 827 | __be32 nfsd_splice_read(struct svc_rqst *rqstp, |
dc97618d BF |
828 | struct file *file, loff_t offset, unsigned long *count) |
829 | { | |
830 | struct splice_desc sd = { | |
831 | .len = 0, | |
832 | .total_len = *count, | |
833 | .pos = offset, | |
834 | .u.data = rqstp, | |
835 | }; | |
836 | int host_err; | |
837 | ||
838 | rqstp->rq_next_page = rqstp->rq_respages + 1; | |
839 | host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); | |
840 | return nfsd_finish_read(file, count, host_err); | |
841 | } | |
842 | ||
e2afc819 | 843 | __be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen, |
dc97618d BF |
844 | unsigned long *count) |
845 | { | |
846 | mm_segment_t oldfs; | |
847 | int host_err; | |
848 | ||
849 | oldfs = get_fs(); | |
850 | set_fs(KERNEL_DS); | |
851 | host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); | |
852 | set_fs(oldfs); | |
853 | return nfsd_finish_read(file, count, host_err); | |
854 | } | |
855 | ||
856 | static __be32 | |
857 | nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file, | |
858 | loff_t offset, struct kvec *vec, int vlen, unsigned long *count) | |
859 | { | |
779fb0f3 | 860 | if (file->f_op->splice_read && test_bit(RQ_SPLICE_OK, &rqstp->rq_flags)) |
dc97618d BF |
861 | return nfsd_splice_read(rqstp, file, offset, count); |
862 | else | |
863 | return nfsd_readv(file, offset, vec, vlen, count); | |
1da177e4 LT |
864 | } |
865 | ||
d911df7b BF |
866 | /* |
867 | * Gathered writes: If another process is currently writing to the file, | |
868 | * there's a high chance this is another nfsd (triggered by a bulk write | |
869 | * from a client's biod). Rather than syncing the file with each write | |
870 | * request, we sleep for 10 msec. | |
871 | * | |
872 | * I don't know if this roughly approximates C. Juszak's idea of | |
873 | * gathered writes, but it's a nice and simple solution (IMHO), and it | |
874 | * seems to work:-) | |
875 | * | |
876 | * Note: we do this only in the NFSv2 case, since v3 and higher have a | |
877 | * better tool (separate unstable writes and commits) for solving this | |
878 | * problem. | |
879 | */ | |
880 | static int wait_for_concurrent_writes(struct file *file) | |
881 | { | |
496ad9aa | 882 | struct inode *inode = file_inode(file); |
d911df7b BF |
883 | static ino_t last_ino; |
884 | static dev_t last_dev; | |
885 | int err = 0; | |
886 | ||
887 | if (atomic_read(&inode->i_writecount) > 1 | |
888 | || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { | |
889 | dprintk("nfsd: write defer %d\n", task_pid_nr(current)); | |
890 | msleep(10); | |
891 | dprintk("nfsd: write resume %d\n", task_pid_nr(current)); | |
892 | } | |
893 | ||
894 | if (inode->i_state & I_DIRTY) { | |
895 | dprintk("nfsd: write sync %d\n", task_pid_nr(current)); | |
8018ab05 | 896 | err = vfs_fsync(file, 0); |
d911df7b BF |
897 | } |
898 | last_ino = inode->i_ino; | |
899 | last_dev = inode->i_sb->s_dev; | |
900 | return err; | |
901 | } | |
902 | ||
6264d69d | 903 | static __be32 |
1da177e4 LT |
904 | nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
905 | loff_t offset, struct kvec *vec, int vlen, | |
31dec253 | 906 | unsigned long *cnt, int *stablep) |
1da177e4 LT |
907 | { |
908 | struct svc_export *exp; | |
1da177e4 LT |
909 | struct inode *inode; |
910 | mm_segment_t oldfs; | |
6264d69d AV |
911 | __be32 err = 0; |
912 | int host_err; | |
1da177e4 | 913 | int stable = *stablep; |
48e03bc5 | 914 | int use_wgather; |
e49dbbf3 | 915 | loff_t pos = offset; |
e77a7b4f | 916 | loff_t end = LLONG_MAX; |
8658452e N |
917 | unsigned int pflags = current->flags; |
918 | ||
7501cc2b | 919 | if (test_bit(RQ_LOCAL, &rqstp->rq_flags)) |
8658452e N |
920 | /* |
921 | * We want less throttling in balance_dirty_pages() | |
922 | * and shrink_inactive_list() so that nfs to | |
923 | * localhost doesn't cause nfsd to lock up due to all | |
924 | * the client's dirty pages or its congested queue. | |
925 | */ | |
926 | current->flags |= PF_LESS_THROTTLE; | |
1da177e4 | 927 | |
6f4e0d5a | 928 | inode = file_inode(file); |
1da177e4 LT |
929 | exp = fhp->fh_export; |
930 | ||
48e03bc5 | 931 | use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); |
1da177e4 | 932 | |
1da177e4 LT |
933 | if (!EX_ISSYNC(exp)) |
934 | stable = 0; | |
1da177e4 LT |
935 | |
936 | /* Write the data. */ | |
937 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
e49dbbf3 | 938 | host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos); |
1da177e4 | 939 | set_fs(oldfs); |
e4636d53 BF |
940 | if (host_err < 0) |
941 | goto out_nfserr; | |
942 | *cnt = host_err; | |
943 | nfsdstats.io_write += host_err; | |
2a12a9d7 | 944 | fsnotify_modify(file); |
1da177e4 | 945 | |
face1502 | 946 | if (stable) { |
e77a7b4f | 947 | if (use_wgather) { |
face1502 | 948 | host_err = wait_for_concurrent_writes(file); |
e77a7b4f ZB |
949 | } else { |
950 | if (*cnt) | |
951 | end = offset + *cnt - 1; | |
952 | host_err = vfs_fsync_range(file, offset, end, 0); | |
953 | } | |
face1502 | 954 | } |
1da177e4 | 955 | |
e4636d53 | 956 | out_nfserr: |
6264d69d | 957 | dprintk("nfsd: write complete host_err=%d\n", host_err); |
a0d24b29 | 958 | if (host_err >= 0) |
1da177e4 | 959 | err = 0; |
a0d24b29 | 960 | else |
6264d69d | 961 | err = nfserrno(host_err); |
7501cc2b | 962 | if (test_bit(RQ_LOCAL, &rqstp->rq_flags)) |
8658452e | 963 | tsk_restore_flags(current, pflags, PF_LESS_THROTTLE); |
1da177e4 LT |
964 | return err; |
965 | } | |
966 | ||
dc97618d BF |
967 | __be32 nfsd_get_tmp_read_open(struct svc_rqst *rqstp, struct svc_fh *fhp, |
968 | struct file **file, struct raparms **ra) | |
fa0a2126 | 969 | { |
fa0a2126 | 970 | struct inode *inode; |
fa0a2126 BF |
971 | __be32 err; |
972 | ||
dc97618d | 973 | err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, file); |
fa0a2126 BF |
974 | if (err) |
975 | return err; | |
976 | ||
dc97618d | 977 | inode = file_inode(*file); |
fa0a2126 BF |
978 | |
979 | /* Get readahead parameters */ | |
dc97618d | 980 | *ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); |
fa0a2126 | 981 | |
dc97618d BF |
982 | if (*ra && (*ra)->p_set) |
983 | (*file)->f_ra = (*ra)->p_ra; | |
984 | return nfs_ok; | |
985 | } | |
fa0a2126 | 986 | |
dc97618d BF |
987 | void nfsd_put_tmp_read_open(struct file *file, struct raparms *ra) |
988 | { | |
fa0a2126 BF |
989 | /* Write back readahead params */ |
990 | if (ra) { | |
991 | struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; | |
992 | spin_lock(&rab->pb_lock); | |
993 | ra->p_ra = file->f_ra; | |
994 | ra->p_set = 1; | |
995 | ra->p_count--; | |
996 | spin_unlock(&rab->pb_lock); | |
997 | } | |
fd891454 | 998 | fput(file); |
dc97618d BF |
999 | } |
1000 | ||
1001 | /* | |
1002 | * Read data from a file. count must contain the requested read count | |
1003 | * on entry. On return, *count contains the number of bytes actually read. | |
1004 | * N.B. After this call fhp needs an fh_put | |
1005 | */ | |
1006 | __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1007 | loff_t offset, struct kvec *vec, int vlen, unsigned long *count) | |
1008 | { | |
1009 | struct file *file; | |
1010 | struct raparms *ra; | |
1011 | __be32 err; | |
1012 | ||
1013 | err = nfsd_get_tmp_read_open(rqstp, fhp, &file, &ra); | |
1014 | if (err) | |
1015 | return err; | |
1016 | ||
1017 | err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count); | |
1018 | ||
1019 | nfsd_put_tmp_read_open(file, ra); | |
1020 | ||
fa0a2126 BF |
1021 | return err; |
1022 | } | |
1023 | ||
1da177e4 LT |
1024 | /* |
1025 | * Write data to a file. | |
1026 | * The stable flag requests synchronous writes. | |
1027 | * N.B. After this call fhp needs an fh_put | |
1028 | */ | |
6264d69d | 1029 | __be32 |
1da177e4 | 1030 | nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
31dec253 | 1031 | loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, |
1da177e4 LT |
1032 | int *stablep) |
1033 | { | |
6264d69d | 1034 | __be32 err = 0; |
1da177e4 LT |
1035 | |
1036 | if (file) { | |
0ec757df | 1037 | err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, |
8837abca | 1038 | NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); |
1da177e4 LT |
1039 | if (err) |
1040 | goto out; | |
1041 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, | |
1042 | stablep); | |
1043 | } else { | |
8837abca | 1044 | err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); |
1da177e4 LT |
1045 | if (err) |
1046 | goto out; | |
1047 | ||
1048 | if (cnt) | |
1049 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, | |
1050 | cnt, stablep); | |
fd891454 | 1051 | fput(file); |
1da177e4 LT |
1052 | } |
1053 | out: | |
1054 | return err; | |
1055 | } | |
1056 | ||
1057 | #ifdef CONFIG_NFSD_V3 | |
1058 | /* | |
1059 | * Commit all pending writes to stable storage. | |
aa696a6f TM |
1060 | * |
1061 | * Note: we only guarantee that data that lies within the range specified | |
1062 | * by the 'offset' and 'count' parameters will be synced. | |
1da177e4 LT |
1063 | * |
1064 | * Unfortunately we cannot lock the file to make sure we return full WCC | |
1065 | * data to the client, as locking happens lower down in the filesystem. | |
1066 | */ | |
6264d69d | 1067 | __be32 |
1da177e4 LT |
1068 | nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1069 | loff_t offset, unsigned long count) | |
1070 | { | |
1071 | struct file *file; | |
aa696a6f TM |
1072 | loff_t end = LLONG_MAX; |
1073 | __be32 err = nfserr_inval; | |
1da177e4 | 1074 | |
aa696a6f TM |
1075 | if (offset < 0) |
1076 | goto out; | |
1077 | if (count != 0) { | |
1078 | end = offset + (loff_t)count - 1; | |
1079 | if (end < offset) | |
1080 | goto out; | |
1081 | } | |
1da177e4 | 1082 | |
91885258 JL |
1083 | err = nfsd_open(rqstp, fhp, S_IFREG, |
1084 | NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); | |
8837abca | 1085 | if (err) |
aa696a6f | 1086 | goto out; |
1da177e4 | 1087 | if (EX_ISSYNC(fhp->fh_export)) { |
8018ab05 | 1088 | int err2 = vfs_fsync_range(file, offset, end, 0); |
aa696a6f TM |
1089 | |
1090 | if (err2 != -EINVAL) | |
1091 | err = nfserrno(err2); | |
1092 | else | |
1da177e4 | 1093 | err = nfserr_notsupp; |
1da177e4 LT |
1094 | } |
1095 | ||
fd891454 | 1096 | fput(file); |
aa696a6f | 1097 | out: |
1da177e4 LT |
1098 | return err; |
1099 | } | |
1100 | #endif /* CONFIG_NFSD_V3 */ | |
1101 | ||
f2b0dee2 | 1102 | static __be32 |
5c002b3b BF |
1103 | nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, |
1104 | struct iattr *iap) | |
1105 | { | |
1106 | /* | |
1107 | * Mode has already been set earlier in create: | |
1108 | */ | |
1109 | iap->ia_valid &= ~ATTR_MODE; | |
1110 | /* | |
1111 | * Setting uid/gid works only for root. Irix appears to | |
1112 | * send along the gid on create when it tries to implement | |
1113 | * setgid directories via NFS: | |
1114 | */ | |
6fab8779 | 1115 | if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) |
5c002b3b BF |
1116 | iap->ia_valid &= ~(ATTR_UID|ATTR_GID); |
1117 | if (iap->ia_valid) | |
1118 | return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); | |
0f3a24b4 | 1119 | /* Callers expect file metadata to be committed here */ |
722b620d | 1120 | return nfserrno(commit_metadata(resfhp)); |
5c002b3b BF |
1121 | } |
1122 | ||
4ac35c2f | 1123 | /* HPUX client sometimes creates a file in mode 000, and sets size to 0. |
1124 | * setting size to 0 may fail for some specific file systems by the permission | |
1125 | * checking which requires WRITE permission but the mode is 000. | |
1126 | * we ignore the resizing(to 0) on the just new created file, since the size is | |
1127 | * 0 after file created. | |
1128 | * | |
1129 | * call this only after vfs_create() is called. | |
1130 | * */ | |
1131 | static void | |
1132 | nfsd_check_ignore_resizing(struct iattr *iap) | |
1133 | { | |
1134 | if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) | |
1135 | iap->ia_valid &= ~ATTR_SIZE; | |
1136 | } | |
1137 | ||
1da177e4 LT |
1138 | /* |
1139 | * Create a file (regular, directory, device, fifo); UNIX sockets | |
1140 | * not yet implemented. | |
1141 | * If the response fh has been verified, the parent directory should | |
1142 | * already be locked. Note that the parent directory is left locked. | |
1143 | * | |
1144 | * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp | |
1145 | */ | |
6264d69d | 1146 | __be32 |
1da177e4 LT |
1147 | nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1148 | char *fname, int flen, struct iattr *iap, | |
1149 | int type, dev_t rdev, struct svc_fh *resfhp) | |
1150 | { | |
1151 | struct dentry *dentry, *dchild = NULL; | |
1152 | struct inode *dirp; | |
6264d69d | 1153 | __be32 err; |
5c002b3b | 1154 | __be32 err2; |
6264d69d | 1155 | int host_err; |
1da177e4 LT |
1156 | |
1157 | err = nfserr_perm; | |
1158 | if (!flen) | |
1159 | goto out; | |
1160 | err = nfserr_exist; | |
1161 | if (isdotent(fname, flen)) | |
1162 | goto out; | |
1163 | ||
8837abca | 1164 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1165 | if (err) |
1166 | goto out; | |
1167 | ||
1168 | dentry = fhp->fh_dentry; | |
2b0143b5 | 1169 | dirp = d_inode(dentry); |
1da177e4 LT |
1170 | |
1171 | err = nfserr_notdir; | |
acfa4380 | 1172 | if (!dirp->i_op->lookup) |
1da177e4 LT |
1173 | goto out; |
1174 | /* | |
1175 | * Check whether the response file handle has been verified yet. | |
1176 | * If it has, the parent directory should already be locked. | |
1177 | */ | |
1178 | if (!resfhp->fh_dentry) { | |
4a55c101 JK |
1179 | host_err = fh_want_write(fhp); |
1180 | if (host_err) | |
1181 | goto out_nfserr; | |
1182 | ||
1da177e4 | 1183 | /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ |
12fd3520 | 1184 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 | 1185 | dchild = lookup_one_len(fname, dentry, flen); |
6264d69d | 1186 | host_err = PTR_ERR(dchild); |
1da177e4 LT |
1187 | if (IS_ERR(dchild)) |
1188 | goto out_nfserr; | |
1189 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); | |
1190 | if (err) | |
1191 | goto out; | |
1192 | } else { | |
1193 | /* called from nfsd_proc_create */ | |
1194 | dchild = dget(resfhp->fh_dentry); | |
1195 | if (!fhp->fh_locked) { | |
1196 | /* not actually possible */ | |
1197 | printk(KERN_ERR | |
a6a9f18f AV |
1198 | "nfsd_create: parent %pd2 not locked!\n", |
1199 | dentry); | |
d75f2b9f | 1200 | err = nfserr_io; |
1da177e4 LT |
1201 | goto out; |
1202 | } | |
1203 | } | |
1204 | /* | |
1205 | * Make sure the child dentry is still negative ... | |
1206 | */ | |
1207 | err = nfserr_exist; | |
2b0143b5 | 1208 | if (d_really_is_positive(dchild)) { |
a6a9f18f AV |
1209 | dprintk("nfsd_create: dentry %pd/%pd not negative!\n", |
1210 | dentry, dchild); | |
1da177e4 LT |
1211 | goto out; |
1212 | } | |
1213 | ||
1214 | if (!(iap->ia_valid & ATTR_MODE)) | |
1215 | iap->ia_mode = 0; | |
1216 | iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; | |
1217 | ||
07cad1d2 MS |
1218 | err = nfserr_inval; |
1219 | if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { | |
1220 | printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", | |
1221 | type); | |
1222 | goto out; | |
1223 | } | |
1224 | ||
1da177e4 LT |
1225 | /* |
1226 | * Get the dir op function pointer. | |
1227 | */ | |
088406bc | 1228 | err = 0; |
4a55c101 | 1229 | host_err = 0; |
1da177e4 LT |
1230 | switch (type) { |
1231 | case S_IFREG: | |
312b63fb | 1232 | host_err = vfs_create(dirp, dchild, iap->ia_mode, true); |
4ac35c2f | 1233 | if (!host_err) |
1234 | nfsd_check_ignore_resizing(iap); | |
1da177e4 LT |
1235 | break; |
1236 | case S_IFDIR: | |
6264d69d | 1237 | host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); |
1da177e4 LT |
1238 | break; |
1239 | case S_IFCHR: | |
1240 | case S_IFBLK: | |
1241 | case S_IFIFO: | |
1242 | case S_IFSOCK: | |
6264d69d | 1243 | host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); |
1da177e4 | 1244 | break; |
1da177e4 | 1245 | } |
4a55c101 | 1246 | if (host_err < 0) |
1da177e4 LT |
1247 | goto out_nfserr; |
1248 | ||
f501912a | 1249 | err = nfsd_create_setattr(rqstp, resfhp, iap); |
1da177e4 | 1250 | |
f501912a | 1251 | /* |
0f3a24b4 TM |
1252 | * nfsd_create_setattr already committed the child. Transactional |
1253 | * filesystems had a chance to commit changes for both parent and | |
1254 | * child * simultaneously making the following commit_metadata a | |
1255 | * noop. | |
f501912a BM |
1256 | */ |
1257 | err2 = nfserrno(commit_metadata(fhp)); | |
5c002b3b BF |
1258 | if (err2) |
1259 | err = err2; | |
1da177e4 LT |
1260 | /* |
1261 | * Update the file handle to get the new inode info. | |
1262 | */ | |
1263 | if (!err) | |
1264 | err = fh_update(resfhp); | |
1265 | out: | |
1266 | if (dchild && !IS_ERR(dchild)) | |
1267 | dput(dchild); | |
1268 | return err; | |
1269 | ||
1270 | out_nfserr: | |
6264d69d | 1271 | err = nfserrno(host_err); |
1da177e4 LT |
1272 | goto out; |
1273 | } | |
1274 | ||
1275 | #ifdef CONFIG_NFSD_V3 | |
ac6721a1 MJ |
1276 | |
1277 | static inline int nfsd_create_is_exclusive(int createmode) | |
1278 | { | |
1279 | return createmode == NFS3_CREATE_EXCLUSIVE | |
1280 | || createmode == NFS4_CREATE_EXCLUSIVE4_1; | |
1281 | } | |
1282 | ||
1da177e4 | 1283 | /* |
ac6721a1 | 1284 | * NFSv3 and NFSv4 version of nfsd_create |
1da177e4 | 1285 | */ |
6264d69d | 1286 | __be32 |
ac6721a1 | 1287 | do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1da177e4 LT |
1288 | char *fname, int flen, struct iattr *iap, |
1289 | struct svc_fh *resfhp, int createmode, u32 *verifier, | |
856121b2 | 1290 | bool *truncp, bool *created) |
1da177e4 LT |
1291 | { |
1292 | struct dentry *dentry, *dchild = NULL; | |
1293 | struct inode *dirp; | |
6264d69d AV |
1294 | __be32 err; |
1295 | int host_err; | |
1da177e4 | 1296 | __u32 v_mtime=0, v_atime=0; |
1da177e4 LT |
1297 | |
1298 | err = nfserr_perm; | |
1299 | if (!flen) | |
1300 | goto out; | |
1301 | err = nfserr_exist; | |
1302 | if (isdotent(fname, flen)) | |
1303 | goto out; | |
1304 | if (!(iap->ia_valid & ATTR_MODE)) | |
1305 | iap->ia_mode = 0; | |
1574dff8 | 1306 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); |
1da177e4 LT |
1307 | if (err) |
1308 | goto out; | |
1309 | ||
1310 | dentry = fhp->fh_dentry; | |
2b0143b5 | 1311 | dirp = d_inode(dentry); |
1da177e4 LT |
1312 | |
1313 | /* Get all the sanity checks out of the way before | |
1314 | * we lock the parent. */ | |
1315 | err = nfserr_notdir; | |
acfa4380 | 1316 | if (!dirp->i_op->lookup) |
1da177e4 | 1317 | goto out; |
4a55c101 JK |
1318 | |
1319 | host_err = fh_want_write(fhp); | |
1320 | if (host_err) | |
1321 | goto out_nfserr; | |
1322 | ||
12fd3520 | 1323 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1324 | |
1325 | /* | |
1326 | * Compose the response file handle. | |
1327 | */ | |
1328 | dchild = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1329 | host_err = PTR_ERR(dchild); |
1da177e4 LT |
1330 | if (IS_ERR(dchild)) |
1331 | goto out_nfserr; | |
1332 | ||
1574dff8 | 1333 | /* If file doesn't exist, check for permissions to create one */ |
2b0143b5 | 1334 | if (d_really_is_negative(dchild)) { |
1574dff8 SP |
1335 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); |
1336 | if (err) | |
1337 | goto out; | |
1338 | } | |
1339 | ||
1da177e4 LT |
1340 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); |
1341 | if (err) | |
1342 | goto out; | |
1343 | ||
ac6721a1 | 1344 | if (nfsd_create_is_exclusive(createmode)) { |
c397852c | 1345 | /* solaris7 gets confused (bugid 4218508) if these have |
749997e5 JL |
1346 | * the high bit set, so just clear the high bits. If this is |
1347 | * ever changed to use different attrs for storing the | |
1348 | * verifier, then do_open_lookup() will also need to be fixed | |
1349 | * accordingly. | |
1da177e4 LT |
1350 | */ |
1351 | v_mtime = verifier[0]&0x7fffffff; | |
1352 | v_atime = verifier[1]&0x7fffffff; | |
1da177e4 LT |
1353 | } |
1354 | ||
2b0143b5 | 1355 | if (d_really_is_positive(dchild)) { |
1da177e4 LT |
1356 | err = 0; |
1357 | ||
1358 | switch (createmode) { | |
1359 | case NFS3_CREATE_UNCHECKED: | |
e36cb0b8 | 1360 | if (! d_is_reg(dchild)) |
9dc4e6c4 | 1361 | goto out; |
1da177e4 LT |
1362 | else if (truncp) { |
1363 | /* in nfsv4, we need to treat this case a little | |
1364 | * differently. we don't want to truncate the | |
1365 | * file now; this would be wrong if the OPEN | |
1366 | * fails for some other reason. furthermore, | |
1367 | * if the size is nonzero, we should ignore it | |
1368 | * according to spec! | |
1369 | */ | |
1370 | *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; | |
1371 | } | |
1372 | else { | |
1373 | iap->ia_valid &= ATTR_SIZE; | |
1374 | goto set_attr; | |
1375 | } | |
1376 | break; | |
1377 | case NFS3_CREATE_EXCLUSIVE: | |
2b0143b5 DH |
1378 | if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime |
1379 | && d_inode(dchild)->i_atime.tv_sec == v_atime | |
1380 | && d_inode(dchild)->i_size == 0 ) { | |
7007c90f NB |
1381 | if (created) |
1382 | *created = 1; | |
1da177e4 | 1383 | break; |
7007c90f | 1384 | } |
ac6721a1 | 1385 | case NFS4_CREATE_EXCLUSIVE4_1: |
2b0143b5 DH |
1386 | if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime |
1387 | && d_inode(dchild)->i_atime.tv_sec == v_atime | |
1388 | && d_inode(dchild)->i_size == 0 ) { | |
7007c90f NB |
1389 | if (created) |
1390 | *created = 1; | |
ac6721a1 | 1391 | goto set_attr; |
7007c90f | 1392 | } |
1da177e4 LT |
1393 | /* fallthru */ |
1394 | case NFS3_CREATE_GUARDED: | |
1395 | err = nfserr_exist; | |
1396 | } | |
bad0dcff | 1397 | fh_drop_write(fhp); |
1da177e4 LT |
1398 | goto out; |
1399 | } | |
1400 | ||
312b63fb | 1401 | host_err = vfs_create(dirp, dchild, iap->ia_mode, true); |
463c3197 | 1402 | if (host_err < 0) { |
bad0dcff | 1403 | fh_drop_write(fhp); |
1da177e4 | 1404 | goto out_nfserr; |
463c3197 | 1405 | } |
81ac95c5 BF |
1406 | if (created) |
1407 | *created = 1; | |
1da177e4 | 1408 | |
4ac35c2f | 1409 | nfsd_check_ignore_resizing(iap); |
1410 | ||
ac6721a1 | 1411 | if (nfsd_create_is_exclusive(createmode)) { |
c397852c | 1412 | /* Cram the verifier into atime/mtime */ |
1da177e4 | 1413 | iap->ia_valid = ATTR_MTIME|ATTR_ATIME |
c397852c | 1414 | | ATTR_MTIME_SET|ATTR_ATIME_SET; |
1da177e4 LT |
1415 | /* XXX someone who knows this better please fix it for nsec */ |
1416 | iap->ia_mtime.tv_sec = v_mtime; | |
1417 | iap->ia_atime.tv_sec = v_atime; | |
1418 | iap->ia_mtime.tv_nsec = 0; | |
1419 | iap->ia_atime.tv_nsec = 0; | |
1da177e4 LT |
1420 | } |
1421 | ||
1da177e4 | 1422 | set_attr: |
f501912a BM |
1423 | err = nfsd_create_setattr(rqstp, resfhp, iap); |
1424 | ||
1425 | /* | |
0f3a24b4 TM |
1426 | * nfsd_create_setattr already committed the child |
1427 | * (and possibly also the parent). | |
f501912a BM |
1428 | */ |
1429 | if (!err) | |
1430 | err = nfserrno(commit_metadata(fhp)); | |
f193fbab YT |
1431 | |
1432 | /* | |
1433 | * Update the filehandle to get the new inode info. | |
1434 | */ | |
1435 | if (!err) | |
1436 | err = fh_update(resfhp); | |
1da177e4 LT |
1437 | |
1438 | out: | |
1439 | fh_unlock(fhp); | |
1440 | if (dchild && !IS_ERR(dchild)) | |
1441 | dput(dchild); | |
4a55c101 | 1442 | fh_drop_write(fhp); |
1da177e4 LT |
1443 | return err; |
1444 | ||
1445 | out_nfserr: | |
6264d69d | 1446 | err = nfserrno(host_err); |
1da177e4 LT |
1447 | goto out; |
1448 | } | |
1449 | #endif /* CONFIG_NFSD_V3 */ | |
1450 | ||
1451 | /* | |
1452 | * Read a symlink. On entry, *lenp must contain the maximum path length that | |
1453 | * fits into the buffer. On return, it contains the true length. | |
1454 | * N.B. After this call fhp needs an fh_put | |
1455 | */ | |
6264d69d | 1456 | __be32 |
1da177e4 LT |
1457 | nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) |
1458 | { | |
1da177e4 LT |
1459 | struct inode *inode; |
1460 | mm_segment_t oldfs; | |
6264d69d AV |
1461 | __be32 err; |
1462 | int host_err; | |
68ac1234 | 1463 | struct path path; |
1da177e4 | 1464 | |
8837abca | 1465 | err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); |
1da177e4 LT |
1466 | if (err) |
1467 | goto out; | |
1468 | ||
68ac1234 AV |
1469 | path.mnt = fhp->fh_export->ex_path.mnt; |
1470 | path.dentry = fhp->fh_dentry; | |
2b0143b5 | 1471 | inode = d_inode(path.dentry); |
1da177e4 LT |
1472 | |
1473 | err = nfserr_inval; | |
acfa4380 | 1474 | if (!inode->i_op->readlink) |
1da177e4 LT |
1475 | goto out; |
1476 | ||
68ac1234 | 1477 | touch_atime(&path); |
1da177e4 LT |
1478 | /* N.B. Why does this call need a get_fs()?? |
1479 | * Remove the set_fs and watch the fireworks:-) --okir | |
1480 | */ | |
1481 | ||
1482 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
fac7a17b | 1483 | host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp); |
1da177e4 LT |
1484 | set_fs(oldfs); |
1485 | ||
6264d69d | 1486 | if (host_err < 0) |
1da177e4 | 1487 | goto out_nfserr; |
6264d69d | 1488 | *lenp = host_err; |
1da177e4 LT |
1489 | err = 0; |
1490 | out: | |
1491 | return err; | |
1492 | ||
1493 | out_nfserr: | |
6264d69d | 1494 | err = nfserrno(host_err); |
1da177e4 LT |
1495 | goto out; |
1496 | } | |
1497 | ||
1498 | /* | |
1499 | * Create a symlink and look up its inode | |
1500 | * N.B. After this call _both_ fhp and resfhp need an fh_put | |
1501 | */ | |
6264d69d | 1502 | __be32 |
1da177e4 LT |
1503 | nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1504 | char *fname, int flen, | |
52ee0433 | 1505 | char *path, |
1e444f5b | 1506 | struct svc_fh *resfhp) |
1da177e4 LT |
1507 | { |
1508 | struct dentry *dentry, *dnew; | |
6264d69d AV |
1509 | __be32 err, cerr; |
1510 | int host_err; | |
1da177e4 LT |
1511 | |
1512 | err = nfserr_noent; | |
52ee0433 | 1513 | if (!flen || path[0] == '\0') |
1da177e4 LT |
1514 | goto out; |
1515 | err = nfserr_exist; | |
1516 | if (isdotent(fname, flen)) | |
1517 | goto out; | |
1518 | ||
8837abca | 1519 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1520 | if (err) |
1521 | goto out; | |
4a55c101 JK |
1522 | |
1523 | host_err = fh_want_write(fhp); | |
1524 | if (host_err) | |
1525 | goto out_nfserr; | |
1526 | ||
1da177e4 LT |
1527 | fh_lock(fhp); |
1528 | dentry = fhp->fh_dentry; | |
1529 | dnew = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1530 | host_err = PTR_ERR(dnew); |
1da177e4 LT |
1531 | if (IS_ERR(dnew)) |
1532 | goto out_nfserr; | |
1533 | ||
2b0143b5 | 1534 | host_err = vfs_symlink(d_inode(dentry), dnew, path); |
6264d69d | 1535 | err = nfserrno(host_err); |
f501912a BM |
1536 | if (!err) |
1537 | err = nfserrno(commit_metadata(fhp)); | |
1da177e4 LT |
1538 | fh_unlock(fhp); |
1539 | ||
bad0dcff | 1540 | fh_drop_write(fhp); |
75c3f29d | 1541 | |
1da177e4 LT |
1542 | cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); |
1543 | dput(dnew); | |
1544 | if (err==0) err = cerr; | |
1545 | out: | |
1546 | return err; | |
1547 | ||
1548 | out_nfserr: | |
6264d69d | 1549 | err = nfserrno(host_err); |
1da177e4 LT |
1550 | goto out; |
1551 | } | |
1552 | ||
1553 | /* | |
1554 | * Create a hardlink | |
1555 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1556 | */ | |
6264d69d | 1557 | __be32 |
1da177e4 LT |
1558 | nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, |
1559 | char *name, int len, struct svc_fh *tfhp) | |
1560 | { | |
1561 | struct dentry *ddir, *dnew, *dold; | |
55b13354 | 1562 | struct inode *dirp; |
6264d69d AV |
1563 | __be32 err; |
1564 | int host_err; | |
1da177e4 | 1565 | |
8837abca | 1566 | err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1567 | if (err) |
1568 | goto out; | |
7d818a7b | 1569 | err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); |
1da177e4 LT |
1570 | if (err) |
1571 | goto out; | |
7d818a7b | 1572 | err = nfserr_isdir; |
e36cb0b8 | 1573 | if (d_is_dir(tfhp->fh_dentry)) |
7d818a7b | 1574 | goto out; |
1da177e4 LT |
1575 | err = nfserr_perm; |
1576 | if (!len) | |
1577 | goto out; | |
1578 | err = nfserr_exist; | |
1579 | if (isdotent(name, len)) | |
1580 | goto out; | |
1581 | ||
4a55c101 JK |
1582 | host_err = fh_want_write(tfhp); |
1583 | if (host_err) { | |
1584 | err = nfserrno(host_err); | |
1585 | goto out; | |
1586 | } | |
1587 | ||
12fd3520 | 1588 | fh_lock_nested(ffhp, I_MUTEX_PARENT); |
1da177e4 | 1589 | ddir = ffhp->fh_dentry; |
2b0143b5 | 1590 | dirp = d_inode(ddir); |
1da177e4 LT |
1591 | |
1592 | dnew = lookup_one_len(name, ddir, len); | |
6264d69d | 1593 | host_err = PTR_ERR(dnew); |
1da177e4 LT |
1594 | if (IS_ERR(dnew)) |
1595 | goto out_nfserr; | |
1596 | ||
1597 | dold = tfhp->fh_dentry; | |
1da177e4 | 1598 | |
4795bb37 | 1599 | err = nfserr_noent; |
2b0143b5 | 1600 | if (d_really_is_negative(dold)) |
4a55c101 | 1601 | goto out_dput; |
146a8595 | 1602 | host_err = vfs_link(dold, dirp, dnew, NULL); |
6264d69d | 1603 | if (!host_err) { |
f501912a BM |
1604 | err = nfserrno(commit_metadata(ffhp)); |
1605 | if (!err) | |
1606 | err = nfserrno(commit_metadata(tfhp)); | |
1da177e4 | 1607 | } else { |
6264d69d | 1608 | if (host_err == -EXDEV && rqstp->rq_vers == 2) |
1da177e4 LT |
1609 | err = nfserr_acces; |
1610 | else | |
6264d69d | 1611 | err = nfserrno(host_err); |
1da177e4 | 1612 | } |
75c3f29d | 1613 | out_dput: |
1da177e4 | 1614 | dput(dnew); |
270d56e5 DR |
1615 | out_unlock: |
1616 | fh_unlock(ffhp); | |
4a55c101 | 1617 | fh_drop_write(tfhp); |
1da177e4 LT |
1618 | out: |
1619 | return err; | |
1620 | ||
1621 | out_nfserr: | |
6264d69d | 1622 | err = nfserrno(host_err); |
270d56e5 | 1623 | goto out_unlock; |
1da177e4 LT |
1624 | } |
1625 | ||
1626 | /* | |
1627 | * Rename a file | |
1628 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1629 | */ | |
6264d69d | 1630 | __be32 |
1da177e4 LT |
1631 | nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, |
1632 | struct svc_fh *tfhp, char *tname, int tlen) | |
1633 | { | |
1634 | struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; | |
1635 | struct inode *fdir, *tdir; | |
6264d69d AV |
1636 | __be32 err; |
1637 | int host_err; | |
1da177e4 | 1638 | |
8837abca | 1639 | err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); |
1da177e4 LT |
1640 | if (err) |
1641 | goto out; | |
8837abca | 1642 | err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); |
1da177e4 LT |
1643 | if (err) |
1644 | goto out; | |
1645 | ||
1646 | fdentry = ffhp->fh_dentry; | |
2b0143b5 | 1647 | fdir = d_inode(fdentry); |
1da177e4 LT |
1648 | |
1649 | tdentry = tfhp->fh_dentry; | |
2b0143b5 | 1650 | tdir = d_inode(tdentry); |
1da177e4 | 1651 | |
1da177e4 LT |
1652 | err = nfserr_perm; |
1653 | if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) | |
1654 | goto out; | |
1655 | ||
4a55c101 JK |
1656 | host_err = fh_want_write(ffhp); |
1657 | if (host_err) { | |
1658 | err = nfserrno(host_err); | |
1659 | goto out; | |
1660 | } | |
1661 | ||
1da177e4 LT |
1662 | /* cannot use fh_lock as we need deadlock protective ordering |
1663 | * so do it by hand */ | |
1664 | trap = lock_rename(tdentry, fdentry); | |
1665 | ffhp->fh_locked = tfhp->fh_locked = 1; | |
1666 | fill_pre_wcc(ffhp); | |
1667 | fill_pre_wcc(tfhp); | |
1668 | ||
1669 | odentry = lookup_one_len(fname, fdentry, flen); | |
6264d69d | 1670 | host_err = PTR_ERR(odentry); |
1da177e4 LT |
1671 | if (IS_ERR(odentry)) |
1672 | goto out_nfserr; | |
1673 | ||
6264d69d | 1674 | host_err = -ENOENT; |
2b0143b5 | 1675 | if (d_really_is_negative(odentry)) |
1da177e4 | 1676 | goto out_dput_old; |
6264d69d | 1677 | host_err = -EINVAL; |
1da177e4 LT |
1678 | if (odentry == trap) |
1679 | goto out_dput_old; | |
1680 | ||
1681 | ndentry = lookup_one_len(tname, tdentry, tlen); | |
6264d69d | 1682 | host_err = PTR_ERR(ndentry); |
1da177e4 LT |
1683 | if (IS_ERR(ndentry)) |
1684 | goto out_dput_old; | |
6264d69d | 1685 | host_err = -ENOTEMPTY; |
1da177e4 LT |
1686 | if (ndentry == trap) |
1687 | goto out_dput_new; | |
1688 | ||
9079b1eb DH |
1689 | host_err = -EXDEV; |
1690 | if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) | |
1691 | goto out_dput_new; | |
aa387d6c BF |
1692 | if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry) |
1693 | goto out_dput_new; | |
9079b1eb | 1694 | |
520c8b16 | 1695 | host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0); |
f501912a BM |
1696 | if (!host_err) { |
1697 | host_err = commit_metadata(tfhp); | |
6264d69d | 1698 | if (!host_err) |
f501912a | 1699 | host_err = commit_metadata(ffhp); |
1da177e4 | 1700 | } |
1da177e4 LT |
1701 | out_dput_new: |
1702 | dput(ndentry); | |
1703 | out_dput_old: | |
1704 | dput(odentry); | |
1705 | out_nfserr: | |
6264d69d | 1706 | err = nfserrno(host_err); |
fbb74a34 BF |
1707 | /* |
1708 | * We cannot rely on fh_unlock on the two filehandles, | |
1da177e4 | 1709 | * as that would do the wrong thing if the two directories |
fbb74a34 | 1710 | * were the same, so again we do it by hand. |
1da177e4 LT |
1711 | */ |
1712 | fill_post_wcc(ffhp); | |
1713 | fill_post_wcc(tfhp); | |
1714 | unlock_rename(tdentry, fdentry); | |
1715 | ffhp->fh_locked = tfhp->fh_locked = 0; | |
4a55c101 | 1716 | fh_drop_write(ffhp); |
1da177e4 LT |
1717 | |
1718 | out: | |
1719 | return err; | |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * Unlink a file or directory | |
1724 | * N.B. After this call fhp needs an fh_put | |
1725 | */ | |
6264d69d | 1726 | __be32 |
1da177e4 LT |
1727 | nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, |
1728 | char *fname, int flen) | |
1729 | { | |
1730 | struct dentry *dentry, *rdentry; | |
1731 | struct inode *dirp; | |
6264d69d AV |
1732 | __be32 err; |
1733 | int host_err; | |
1da177e4 LT |
1734 | |
1735 | err = nfserr_acces; | |
1736 | if (!flen || isdotent(fname, flen)) | |
1737 | goto out; | |
8837abca | 1738 | err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); |
1da177e4 LT |
1739 | if (err) |
1740 | goto out; | |
1741 | ||
4a55c101 JK |
1742 | host_err = fh_want_write(fhp); |
1743 | if (host_err) | |
1744 | goto out_nfserr; | |
1745 | ||
12fd3520 | 1746 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 | 1747 | dentry = fhp->fh_dentry; |
2b0143b5 | 1748 | dirp = d_inode(dentry); |
1da177e4 LT |
1749 | |
1750 | rdentry = lookup_one_len(fname, dentry, flen); | |
6264d69d | 1751 | host_err = PTR_ERR(rdentry); |
1da177e4 LT |
1752 | if (IS_ERR(rdentry)) |
1753 | goto out_nfserr; | |
1754 | ||
2b0143b5 | 1755 | if (d_really_is_negative(rdentry)) { |
1da177e4 LT |
1756 | dput(rdentry); |
1757 | err = nfserr_noent; | |
1758 | goto out; | |
1759 | } | |
1760 | ||
1761 | if (!type) | |
2b0143b5 | 1762 | type = d_inode(rdentry)->i_mode & S_IFMT; |
1da177e4 | 1763 | |
9ce137ee | 1764 | if (type != S_IFDIR) |
b21996e3 | 1765 | host_err = vfs_unlink(dirp, rdentry, NULL); |
9ce137ee | 1766 | else |
6264d69d | 1767 | host_err = vfs_rmdir(dirp, rdentry); |
f501912a BM |
1768 | if (!host_err) |
1769 | host_err = commit_metadata(fhp); | |
541ce98c BF |
1770 | dput(rdentry); |
1771 | ||
1da177e4 | 1772 | out_nfserr: |
6264d69d | 1773 | err = nfserrno(host_err); |
f193fbab YT |
1774 | out: |
1775 | return err; | |
1da177e4 LT |
1776 | } |
1777 | ||
14f7dd63 DW |
1778 | /* |
1779 | * We do this buffering because we must not call back into the file | |
1780 | * system's ->lookup() method from the filldir callback. That may well | |
1781 | * deadlock a number of file systems. | |
1782 | * | |
1783 | * This is based heavily on the implementation of same in XFS. | |
1784 | */ | |
1785 | struct buffered_dirent { | |
1786 | u64 ino; | |
1787 | loff_t offset; | |
1788 | int namlen; | |
1789 | unsigned int d_type; | |
1790 | char name[]; | |
1791 | }; | |
1792 | ||
1793 | struct readdir_data { | |
5c0ba4e0 | 1794 | struct dir_context ctx; |
14f7dd63 DW |
1795 | char *dirent; |
1796 | size_t used; | |
53c9c5c0 | 1797 | int full; |
14f7dd63 DW |
1798 | }; |
1799 | ||
ac7576f4 MS |
1800 | static int nfsd_buffered_filldir(struct dir_context *ctx, const char *name, |
1801 | int namlen, loff_t offset, u64 ino, | |
1802 | unsigned int d_type) | |
14f7dd63 | 1803 | { |
ac7576f4 MS |
1804 | struct readdir_data *buf = |
1805 | container_of(ctx, struct readdir_data, ctx); | |
14f7dd63 DW |
1806 | struct buffered_dirent *de = (void *)(buf->dirent + buf->used); |
1807 | unsigned int reclen; | |
1808 | ||
1809 | reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); | |
53c9c5c0 AV |
1810 | if (buf->used + reclen > PAGE_SIZE) { |
1811 | buf->full = 1; | |
14f7dd63 | 1812 | return -EINVAL; |
53c9c5c0 | 1813 | } |
14f7dd63 DW |
1814 | |
1815 | de->namlen = namlen; | |
1816 | de->offset = offset; | |
1817 | de->ino = ino; | |
1818 | de->d_type = d_type; | |
1819 | memcpy(de->name, name, namlen); | |
1820 | buf->used += reclen; | |
1821 | ||
1822 | return 0; | |
1823 | } | |
1824 | ||
ac7576f4 | 1825 | static __be32 nfsd_buffered_readdir(struct file *file, nfsd_filldir_t func, |
2f9092e1 | 1826 | struct readdir_cd *cdp, loff_t *offsetp) |
2628b766 | 1827 | { |
14f7dd63 | 1828 | struct buffered_dirent *de; |
2628b766 | 1829 | int host_err; |
14f7dd63 DW |
1830 | int size; |
1831 | loff_t offset; | |
ac6614b7 AV |
1832 | struct readdir_data buf = { |
1833 | .ctx.actor = nfsd_buffered_filldir, | |
1834 | .dirent = (void *)__get_free_page(GFP_KERNEL) | |
1835 | }; | |
2628b766 | 1836 | |
14f7dd63 | 1837 | if (!buf.dirent) |
2f9092e1 | 1838 | return nfserrno(-ENOMEM); |
14f7dd63 DW |
1839 | |
1840 | offset = *offsetp; | |
2628b766 | 1841 | |
14f7dd63 | 1842 | while (1) { |
496ad9aa | 1843 | struct inode *dir_inode = file_inode(file); |
14f7dd63 DW |
1844 | unsigned int reclen; |
1845 | ||
b726e923 | 1846 | cdp->err = nfserr_eof; /* will be cleared on successful read */ |
14f7dd63 | 1847 | buf.used = 0; |
53c9c5c0 | 1848 | buf.full = 0; |
14f7dd63 | 1849 | |
5c0ba4e0 | 1850 | host_err = iterate_dir(file, &buf.ctx); |
53c9c5c0 AV |
1851 | if (buf.full) |
1852 | host_err = 0; | |
1853 | ||
1854 | if (host_err < 0) | |
14f7dd63 DW |
1855 | break; |
1856 | ||
1857 | size = buf.used; | |
1858 | ||
1859 | if (!size) | |
1860 | break; | |
1861 | ||
2f9092e1 DW |
1862 | /* |
1863 | * Various filldir functions may end up calling back into | |
1864 | * lookup_one_len() and the file system's ->lookup() method. | |
1865 | * These expect i_mutex to be held, as it would within readdir. | |
1866 | */ | |
1867 | host_err = mutex_lock_killable(&dir_inode->i_mutex); | |
1868 | if (host_err) | |
1869 | break; | |
1870 | ||
14f7dd63 DW |
1871 | de = (struct buffered_dirent *)buf.dirent; |
1872 | while (size > 0) { | |
1873 | offset = de->offset; | |
1874 | ||
1875 | if (func(cdp, de->name, de->namlen, de->offset, | |
1876 | de->ino, de->d_type)) | |
2f9092e1 | 1877 | break; |
14f7dd63 DW |
1878 | |
1879 | if (cdp->err != nfs_ok) | |
2f9092e1 | 1880 | break; |
14f7dd63 DW |
1881 | |
1882 | reclen = ALIGN(sizeof(*de) + de->namlen, | |
1883 | sizeof(u64)); | |
1884 | size -= reclen; | |
1885 | de = (struct buffered_dirent *)((char *)de + reclen); | |
1886 | } | |
2f9092e1 DW |
1887 | mutex_unlock(&dir_inode->i_mutex); |
1888 | if (size > 0) /* We bailed out early */ | |
1889 | break; | |
1890 | ||
c002a6c7 | 1891 | offset = vfs_llseek(file, 0, SEEK_CUR); |
14f7dd63 DW |
1892 | } |
1893 | ||
14f7dd63 | 1894 | free_page((unsigned long)(buf.dirent)); |
2628b766 DW |
1895 | |
1896 | if (host_err) | |
1897 | return nfserrno(host_err); | |
14f7dd63 DW |
1898 | |
1899 | *offsetp = offset; | |
1900 | return cdp->err; | |
2628b766 DW |
1901 | } |
1902 | ||
1da177e4 LT |
1903 | /* |
1904 | * Read entries from a directory. | |
1905 | * The NFSv3/4 verifier we ignore for now. | |
1906 | */ | |
6264d69d | 1907 | __be32 |
1da177e4 | 1908 | nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, |
ac7576f4 | 1909 | struct readdir_cd *cdp, nfsd_filldir_t func) |
1da177e4 | 1910 | { |
6264d69d | 1911 | __be32 err; |
1da177e4 LT |
1912 | struct file *file; |
1913 | loff_t offset = *offsetp; | |
06effdbb BS |
1914 | int may_flags = NFSD_MAY_READ; |
1915 | ||
1916 | /* NFSv2 only supports 32 bit cookies */ | |
1917 | if (rqstp->rq_vers > 2) | |
1918 | may_flags |= NFSD_MAY_64BIT_COOKIE; | |
1da177e4 | 1919 | |
06effdbb | 1920 | err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file); |
1da177e4 LT |
1921 | if (err) |
1922 | goto out; | |
1923 | ||
b108fe6b | 1924 | offset = vfs_llseek(file, offset, SEEK_SET); |
1da177e4 LT |
1925 | if (offset < 0) { |
1926 | err = nfserrno((int)offset); | |
1927 | goto out_close; | |
1928 | } | |
1929 | ||
14f7dd63 | 1930 | err = nfsd_buffered_readdir(file, func, cdp, offsetp); |
1da177e4 LT |
1931 | |
1932 | if (err == nfserr_eof || err == nfserr_toosmall) | |
1933 | err = nfs_ok; /* can still be found in ->err */ | |
1934 | out_close: | |
fd891454 | 1935 | fput(file); |
1da177e4 LT |
1936 | out: |
1937 | return err; | |
1938 | } | |
1939 | ||
1940 | /* | |
1941 | * Get file system stats | |
1942 | * N.B. After this call fhp needs an fh_put | |
1943 | */ | |
6264d69d | 1944 | __be32 |
04716e66 | 1945 | nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) |
1da177e4 | 1946 | { |
ebabe9a9 CH |
1947 | __be32 err; |
1948 | ||
1949 | err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); | |
f6360efb TI |
1950 | if (!err) { |
1951 | struct path path = { | |
1952 | .mnt = fhp->fh_export->ex_path.mnt, | |
1953 | .dentry = fhp->fh_dentry, | |
1954 | }; | |
1955 | if (vfs_statfs(&path, stat)) | |
1956 | err = nfserr_io; | |
1957 | } | |
1da177e4 LT |
1958 | return err; |
1959 | } | |
1960 | ||
c7d51402 | 1961 | static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) |
e22841c6 | 1962 | { |
c7d51402 | 1963 | return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; |
e22841c6 BF |
1964 | } |
1965 | ||
1da177e4 LT |
1966 | /* |
1967 | * Check for a user's access permissions to this inode. | |
1968 | */ | |
6264d69d | 1969 | __be32 |
0ec757df BF |
1970 | nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, |
1971 | struct dentry *dentry, int acc) | |
1da177e4 | 1972 | { |
2b0143b5 | 1973 | struct inode *inode = d_inode(dentry); |
1da177e4 LT |
1974 | int err; |
1975 | ||
aea93397 | 1976 | if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP) |
1da177e4 LT |
1977 | return 0; |
1978 | #if 0 | |
1979 | dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", | |
1980 | acc, | |
8837abca MS |
1981 | (acc & NFSD_MAY_READ)? " read" : "", |
1982 | (acc & NFSD_MAY_WRITE)? " write" : "", | |
1983 | (acc & NFSD_MAY_EXEC)? " exec" : "", | |
1984 | (acc & NFSD_MAY_SATTR)? " sattr" : "", | |
1985 | (acc & NFSD_MAY_TRUNC)? " trunc" : "", | |
1986 | (acc & NFSD_MAY_LOCK)? " lock" : "", | |
1987 | (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", | |
1da177e4 LT |
1988 | inode->i_mode, |
1989 | IS_IMMUTABLE(inode)? " immut" : "", | |
1990 | IS_APPEND(inode)? " append" : "", | |
2c463e95 | 1991 | __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); |
1da177e4 | 1992 | dprintk(" owner %d/%d user %d/%d\n", |
5cc0a840 | 1993 | inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); |
1da177e4 LT |
1994 | #endif |
1995 | ||
1996 | /* Normally we reject any write/sattr etc access on a read-only file | |
1997 | * system. But if it is IRIX doing check on write-access for a | |
1998 | * device special file, we ignore rofs. | |
1999 | */ | |
8837abca MS |
2000 | if (!(acc & NFSD_MAY_LOCAL_ACCESS)) |
2001 | if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { | |
2c463e95 DH |
2002 | if (exp_rdonly(rqstp, exp) || |
2003 | __mnt_is_readonly(exp->ex_path.mnt)) | |
1da177e4 | 2004 | return nfserr_rofs; |
8837abca | 2005 | if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) |
1da177e4 LT |
2006 | return nfserr_perm; |
2007 | } | |
8837abca | 2008 | if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) |
1da177e4 LT |
2009 | return nfserr_perm; |
2010 | ||
8837abca | 2011 | if (acc & NFSD_MAY_LOCK) { |
1da177e4 LT |
2012 | /* If we cannot rely on authentication in NLM requests, |
2013 | * just allow locks, otherwise require read permission, or | |
2014 | * ownership | |
2015 | */ | |
2016 | if (exp->ex_flags & NFSEXP_NOAUTHNLM) | |
2017 | return 0; | |
2018 | else | |
8837abca | 2019 | acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; |
1da177e4 LT |
2020 | } |
2021 | /* | |
2022 | * The file owner always gets access permission for accesses that | |
2023 | * would normally be checked at open time. This is to make | |
2024 | * file access work even when the client has done a fchmod(fd, 0). | |
2025 | * | |
2026 | * However, `cp foo bar' should fail nevertheless when bar is | |
2027 | * readonly. A sensible way to do this might be to reject all | |
2028 | * attempts to truncate a read-only file, because a creat() call | |
2029 | * always implies file truncation. | |
2030 | * ... but this isn't really fair. A process may reasonably call | |
2031 | * ftruncate on an open file descriptor on a file with perm 000. | |
2032 | * We must trust the client to do permission checking - using "ACCESS" | |
2033 | * with NFSv3. | |
2034 | */ | |
8837abca | 2035 | if ((acc & NFSD_MAY_OWNER_OVERRIDE) && |
6fab8779 | 2036 | uid_eq(inode->i_uid, current_fsuid())) |
1da177e4 LT |
2037 | return 0; |
2038 | ||
8837abca | 2039 | /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ |
f419a2e3 | 2040 | err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); |
1da177e4 LT |
2041 | |
2042 | /* Allow read access to binaries even when mode 111 */ | |
2043 | if (err == -EACCES && S_ISREG(inode->i_mode) && | |
a043226b BF |
2044 | (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) || |
2045 | acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC))) | |
f419a2e3 | 2046 | err = inode_permission(inode, MAY_EXEC); |
1da177e4 LT |
2047 | |
2048 | return err? nfserrno(err) : 0; | |
2049 | } | |
2050 | ||
2051 | void | |
2052 | nfsd_racache_shutdown(void) | |
2053 | { | |
54a66e54 JL |
2054 | struct raparms *raparm, *last_raparm; |
2055 | unsigned int i; | |
2056 | ||
1da177e4 | 2057 | dprintk("nfsd: freeing readahead buffers.\n"); |
54a66e54 JL |
2058 | |
2059 | for (i = 0; i < RAPARM_HASH_SIZE; i++) { | |
2060 | raparm = raparm_hash[i].pb_head; | |
2061 | while(raparm) { | |
2062 | last_raparm = raparm; | |
2063 | raparm = raparm->p_next; | |
2064 | kfree(last_raparm); | |
2065 | } | |
2066 | raparm_hash[i].pb_head = NULL; | |
2067 | } | |
1da177e4 LT |
2068 | } |
2069 | /* | |
2070 | * Initialize readahead param cache | |
2071 | */ | |
2072 | int | |
2073 | nfsd_racache_init(int cache_size) | |
2074 | { | |
2075 | int i; | |
fce1456a GB |
2076 | int j = 0; |
2077 | int nperbucket; | |
54a66e54 | 2078 | struct raparms **raparm = NULL; |
1da177e4 | 2079 | |
fce1456a | 2080 | |
54a66e54 | 2081 | if (raparm_hash[0].pb_head) |
1da177e4 | 2082 | return 0; |
54a66e54 | 2083 | nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); |
3c7aa15d | 2084 | nperbucket = max(2, nperbucket); |
54a66e54 | 2085 | cache_size = nperbucket * RAPARM_HASH_SIZE; |
4b3bb06b YB |
2086 | |
2087 | dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); | |
54a66e54 JL |
2088 | |
2089 | for (i = 0; i < RAPARM_HASH_SIZE; i++) { | |
4b3bb06b | 2090 | spin_lock_init(&raparm_hash[i].pb_lock); |
54a66e54 JL |
2091 | |
2092 | raparm = &raparm_hash[i].pb_head; | |
2093 | for (j = 0; j < nperbucket; j++) { | |
2094 | *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); | |
2095 | if (!*raparm) | |
2096 | goto out_nomem; | |
2097 | raparm = &(*raparm)->p_next; | |
2098 | } | |
2099 | *raparm = NULL; | |
4b3bb06b YB |
2100 | } |
2101 | ||
1da177e4 LT |
2102 | nfsdstats.ra_size = cache_size; |
2103 | return 0; | |
54a66e54 JL |
2104 | |
2105 | out_nomem: | |
2106 | dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); | |
2107 | nfsd_racache_shutdown(); | |
2108 | return -ENOMEM; | |
1da177e4 | 2109 | } |