Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #define MSNFS /* HACK HACK */ |
2 | /* | |
3 | * linux/fs/nfsd/vfs.c | |
4 | * | |
5 | * File operations used by nfsd. Some of these have been ripped from | |
6 | * other parts of the kernel because they weren't exported, others | |
7 | * are partial duplicates with added or changed functionality. | |
8 | * | |
9 | * Note that several functions dget() the dentry upon which they want | |
10 | * to act, most notably those that create directory entries. Response | |
11 | * dentry's are dput()'d if necessary in the release callback. | |
12 | * So if you notice code paths that apparently fail to dput() the | |
13 | * dentry, don't worry--they have been taken care of. | |
14 | * | |
15 | * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de> | |
16 | * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp> | |
17 | */ | |
18 | ||
1da177e4 LT |
19 | #include <linux/string.h> |
20 | #include <linux/time.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/fs.h> | |
23 | #include <linux/file.h> | |
24 | #include <linux/mount.h> | |
25 | #include <linux/major.h> | |
26 | #include <linux/ext2_fs.h> | |
27 | #include <linux/proc_fs.h> | |
28 | #include <linux/stat.h> | |
29 | #include <linux/fcntl.h> | |
30 | #include <linux/net.h> | |
31 | #include <linux/unistd.h> | |
32 | #include <linux/slab.h> | |
33 | #include <linux/pagemap.h> | |
34 | #include <linux/in.h> | |
35 | #include <linux/module.h> | |
36 | #include <linux/namei.h> | |
37 | #include <linux/vfs.h> | |
38 | #include <linux/delay.h> | |
39 | #include <linux/sunrpc/svc.h> | |
40 | #include <linux/nfsd/nfsd.h> | |
41 | #ifdef CONFIG_NFSD_V3 | |
42 | #include <linux/nfs3.h> | |
43 | #include <linux/nfsd/xdr3.h> | |
44 | #endif /* CONFIG_NFSD_V3 */ | |
45 | #include <linux/nfsd/nfsfh.h> | |
46 | #include <linux/quotaops.h> | |
0eeca283 | 47 | #include <linux/fsnotify.h> |
1da177e4 LT |
48 | #include <linux/posix_acl.h> |
49 | #include <linux/posix_acl_xattr.h> | |
1da177e4 | 50 | #include <linux/xattr.h> |
5be196e5 | 51 | #ifdef CONFIG_NFSD_V4 |
1da177e4 LT |
52 | #include <linux/nfs4.h> |
53 | #include <linux/nfs4_acl.h> | |
54 | #include <linux/nfsd_idmap.h> | |
55 | #include <linux/security.h> | |
56 | #endif /* CONFIG_NFSD_V4 */ | |
57 | ||
58 | #include <asm/uaccess.h> | |
59 | ||
60 | #define NFSDDBG_FACILITY NFSDDBG_FILEOP | |
61 | #define NFSD_PARANOIA | |
62 | ||
63 | ||
64 | /* We must ignore files (but only files) which might have mandatory | |
65 | * locks on them because there is no way to know if the accesser has | |
66 | * the lock. | |
67 | */ | |
68 | #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i)) | |
69 | ||
70 | /* | |
71 | * This is a cache of readahead params that help us choose the proper | |
72 | * readahead strategy. Initially, we set all readahead parameters to 0 | |
73 | * and let the VFS handle things. | |
74 | * If you increase the number of cached files very much, you'll need to | |
75 | * add a hash table here. | |
76 | */ | |
77 | struct raparms { | |
78 | struct raparms *p_next; | |
79 | unsigned int p_count; | |
80 | ino_t p_ino; | |
81 | dev_t p_dev; | |
82 | int p_set; | |
83 | struct file_ra_state p_ra; | |
84 | }; | |
85 | ||
86 | static struct raparms * raparml; | |
87 | static struct raparms * raparm_cache; | |
88 | ||
89 | /* | |
90 | * Called from nfsd_lookup and encode_dirent. Check if we have crossed | |
91 | * a mount point. | |
92 | * Returns -EAGAIN leaving *dpp and *expp unchanged, | |
93 | * or nfs_ok having possibly changed *dpp and *expp | |
94 | */ | |
95 | int | |
96 | nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, | |
97 | struct svc_export **expp) | |
98 | { | |
99 | struct svc_export *exp = *expp, *exp2 = NULL; | |
100 | struct dentry *dentry = *dpp; | |
101 | struct vfsmount *mnt = mntget(exp->ex_mnt); | |
102 | struct dentry *mounts = dget(dentry); | |
103 | int err = nfs_ok; | |
104 | ||
105 | while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts)); | |
106 | ||
107 | exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle); | |
108 | if (IS_ERR(exp2)) { | |
109 | err = PTR_ERR(exp2); | |
110 | dput(mounts); | |
111 | mntput(mnt); | |
112 | goto out; | |
113 | } | |
114 | if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) { | |
115 | /* successfully crossed mount point */ | |
116 | exp_put(exp); | |
117 | *expp = exp2; | |
118 | dput(dentry); | |
119 | *dpp = mounts; | |
120 | } else { | |
121 | if (exp2) exp_put(exp2); | |
122 | dput(mounts); | |
123 | } | |
124 | mntput(mnt); | |
125 | out: | |
126 | return err; | |
127 | } | |
128 | ||
129 | /* | |
130 | * Look up one component of a pathname. | |
131 | * N.B. After this call _both_ fhp and resfh need an fh_put | |
132 | * | |
133 | * If the lookup would cross a mountpoint, and the mounted filesystem | |
134 | * is exported to the client with NFSEXP_NOHIDE, then the lookup is | |
135 | * accepted as it stands and the mounted directory is | |
136 | * returned. Otherwise the covered directory is returned. | |
137 | * NOTE: this mountpoint crossing is not supported properly by all | |
138 | * clients and is explicitly disallowed for NFSv3 | |
139 | * NeilBrown <neilb@cse.unsw.edu.au> | |
140 | */ | |
141 | int | |
142 | nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, | |
143 | int len, struct svc_fh *resfh) | |
144 | { | |
145 | struct svc_export *exp; | |
146 | struct dentry *dparent; | |
147 | struct dentry *dentry; | |
148 | int err; | |
149 | ||
150 | dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); | |
151 | ||
152 | /* Obtain dentry and export. */ | |
153 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC); | |
154 | if (err) | |
155 | return err; | |
156 | ||
157 | dparent = fhp->fh_dentry; | |
158 | exp = fhp->fh_export; | |
159 | exp_get(exp); | |
160 | ||
161 | err = nfserr_acces; | |
162 | ||
163 | /* Lookup the name, but don't follow links */ | |
164 | if (isdotent(name, len)) { | |
165 | if (len==1) | |
166 | dentry = dget(dparent); | |
167 | else if (dparent != exp->ex_dentry) { | |
168 | dentry = dget_parent(dparent); | |
169 | } else if (!EX_NOHIDE(exp)) | |
170 | dentry = dget(dparent); /* .. == . just like at / */ | |
171 | else { | |
172 | /* checking mountpoint crossing is very different when stepping up */ | |
173 | struct svc_export *exp2 = NULL; | |
174 | struct dentry *dp; | |
175 | struct vfsmount *mnt = mntget(exp->ex_mnt); | |
176 | dentry = dget(dparent); | |
177 | while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry)) | |
178 | ; | |
179 | dp = dget_parent(dentry); | |
180 | dput(dentry); | |
181 | dentry = dp; | |
182 | ||
183 | exp2 = exp_parent(exp->ex_client, mnt, dentry, | |
184 | &rqstp->rq_chandle); | |
185 | if (IS_ERR(exp2)) { | |
186 | err = PTR_ERR(exp2); | |
187 | dput(dentry); | |
188 | mntput(mnt); | |
189 | goto out_nfserr; | |
190 | } | |
191 | if (!exp2) { | |
192 | dput(dentry); | |
193 | dentry = dget(dparent); | |
194 | } else { | |
195 | exp_put(exp); | |
196 | exp = exp2; | |
197 | } | |
198 | mntput(mnt); | |
199 | } | |
200 | } else { | |
201 | fh_lock(fhp); | |
202 | dentry = lookup_one_len(name, dparent, len); | |
203 | err = PTR_ERR(dentry); | |
204 | if (IS_ERR(dentry)) | |
205 | goto out_nfserr; | |
206 | /* | |
207 | * check if we have crossed a mount point ... | |
208 | */ | |
209 | if (d_mountpoint(dentry)) { | |
210 | if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { | |
211 | dput(dentry); | |
212 | goto out_nfserr; | |
213 | } | |
214 | } | |
215 | } | |
216 | /* | |
217 | * Note: we compose the file handle now, but as the | |
218 | * dentry may be negative, it may need to be updated. | |
219 | */ | |
220 | err = fh_compose(resfh, exp, dentry, fhp); | |
221 | if (!err && !dentry->d_inode) | |
222 | err = nfserr_noent; | |
223 | dput(dentry); | |
224 | out: | |
225 | exp_put(exp); | |
226 | return err; | |
227 | ||
228 | out_nfserr: | |
229 | err = nfserrno(err); | |
230 | goto out; | |
231 | } | |
232 | ||
233 | /* | |
234 | * Set various file attributes. | |
235 | * N.B. After this call fhp needs an fh_put | |
236 | */ | |
237 | int | |
238 | nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, | |
239 | int check_guard, time_t guardtime) | |
240 | { | |
241 | struct dentry *dentry; | |
242 | struct inode *inode; | |
243 | int accmode = MAY_SATTR; | |
244 | int ftype = 0; | |
245 | int imode; | |
246 | int err; | |
247 | int size_change = 0; | |
248 | ||
249 | if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) | |
250 | accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE; | |
251 | if (iap->ia_valid & ATTR_SIZE) | |
252 | ftype = S_IFREG; | |
253 | ||
254 | /* Get inode */ | |
255 | err = fh_verify(rqstp, fhp, ftype, accmode); | |
15b7a1b8 | 256 | if (err) |
1da177e4 LT |
257 | goto out; |
258 | ||
259 | dentry = fhp->fh_dentry; | |
260 | inode = dentry->d_inode; | |
261 | ||
15b7a1b8 N |
262 | /* Ignore any mode updates on symlinks */ |
263 | if (S_ISLNK(inode->i_mode)) | |
264 | iap->ia_valid &= ~ATTR_MODE; | |
265 | ||
266 | if (!iap->ia_valid) | |
267 | goto out; | |
268 | ||
1da177e4 LT |
269 | /* NFSv2 does not differentiate between "set-[ac]time-to-now" |
270 | * which only requires access, and "set-[ac]time-to-X" which | |
271 | * requires ownership. | |
272 | * So if it looks like it might be "set both to the same time which | |
273 | * is close to now", and if inode_change_ok fails, then we | |
274 | * convert to "set to now" instead of "set to explicit time" | |
275 | * | |
276 | * We only call inode_change_ok as the last test as technically | |
277 | * it is not an interface that we should be using. It is only | |
278 | * valid if the filesystem does not define it's own i_op->setattr. | |
279 | */ | |
280 | #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) | |
281 | #define MAX_TOUCH_TIME_ERROR (30*60) | |
282 | if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET | |
283 | && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec | |
284 | ) { | |
285 | /* Looks probable. Now just make sure time is in the right ballpark. | |
286 | * Solaris, at least, doesn't seem to care what the time request is. | |
287 | * We require it be within 30 minutes of now. | |
288 | */ | |
289 | time_t delta = iap->ia_atime.tv_sec - get_seconds(); | |
290 | if (delta<0) delta = -delta; | |
291 | if (delta < MAX_TOUCH_TIME_ERROR && | |
292 | inode_change_ok(inode, iap) != 0) { | |
293 | /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME | |
294 | * this will cause notify_change to set these times to "now" | |
295 | */ | |
296 | iap->ia_valid &= ~BOTH_TIME_SET; | |
297 | } | |
298 | } | |
299 | ||
300 | /* The size case is special. It changes the file as well as the attributes. */ | |
301 | if (iap->ia_valid & ATTR_SIZE) { | |
302 | if (iap->ia_size < inode->i_size) { | |
303 | err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE); | |
304 | if (err) | |
305 | goto out; | |
306 | } | |
307 | ||
308 | /* | |
309 | * If we are changing the size of the file, then | |
310 | * we need to break all leases. | |
311 | */ | |
312 | err = break_lease(inode, FMODE_WRITE | O_NONBLOCK); | |
313 | if (err == -EWOULDBLOCK) | |
314 | err = -ETIMEDOUT; | |
315 | if (err) /* ENOMEM or EWOULDBLOCK */ | |
316 | goto out_nfserr; | |
317 | ||
318 | err = get_write_access(inode); | |
319 | if (err) | |
320 | goto out_nfserr; | |
321 | ||
322 | size_change = 1; | |
323 | err = locks_verify_truncate(inode, NULL, iap->ia_size); | |
324 | if (err) { | |
325 | put_write_access(inode); | |
326 | goto out_nfserr; | |
327 | } | |
328 | DQUOT_INIT(inode); | |
329 | } | |
330 | ||
331 | imode = inode->i_mode; | |
332 | if (iap->ia_valid & ATTR_MODE) { | |
333 | iap->ia_mode &= S_IALLUGO; | |
334 | imode = iap->ia_mode |= (imode & ~S_IALLUGO); | |
335 | } | |
336 | ||
337 | /* Revoke setuid/setgid bit on chown/chgrp */ | |
338 | if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) | |
339 | iap->ia_valid |= ATTR_KILL_SUID; | |
340 | if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid) | |
341 | iap->ia_valid |= ATTR_KILL_SGID; | |
342 | ||
343 | /* Change the attributes. */ | |
344 | ||
345 | iap->ia_valid |= ATTR_CTIME; | |
346 | ||
347 | err = nfserr_notsync; | |
348 | if (!check_guard || guardtime == inode->i_ctime.tv_sec) { | |
349 | fh_lock(fhp); | |
350 | err = notify_change(dentry, iap); | |
351 | err = nfserrno(err); | |
352 | fh_unlock(fhp); | |
353 | } | |
354 | if (size_change) | |
355 | put_write_access(inode); | |
356 | if (!err) | |
357 | if (EX_ISSYNC(fhp->fh_export)) | |
358 | write_inode_now(inode, 1); | |
359 | out: | |
360 | return err; | |
361 | ||
362 | out_nfserr: | |
363 | err = nfserrno(err); | |
364 | goto out; | |
365 | } | |
366 | ||
5be196e5 CH |
367 | #if defined(CONFIG_NFSD_V2_ACL) || \ |
368 | defined(CONFIG_NFSD_V3_ACL) || \ | |
369 | defined(CONFIG_NFSD_V4) | |
370 | static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) | |
371 | { | |
372 | ssize_t buflen; | |
5be196e5 CH |
373 | |
374 | buflen = vfs_getxattr(dentry, key, NULL, 0); | |
375 | if (buflen <= 0) | |
376 | return buflen; | |
1da177e4 | 377 | |
5be196e5 CH |
378 | *buf = kmalloc(buflen, GFP_KERNEL); |
379 | if (!*buf) | |
380 | return -ENOMEM; | |
381 | ||
b5872b0d | 382 | return vfs_getxattr(dentry, key, *buf, buflen); |
5be196e5 CH |
383 | } |
384 | #endif | |
385 | ||
386 | #if defined(CONFIG_NFSD_V4) | |
1da177e4 LT |
387 | static int |
388 | set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) | |
389 | { | |
390 | int len; | |
391 | size_t buflen; | |
392 | char *buf = NULL; | |
393 | int error = 0; | |
1da177e4 LT |
394 | |
395 | buflen = posix_acl_xattr_size(pacl->a_count); | |
396 | buf = kmalloc(buflen, GFP_KERNEL); | |
397 | error = -ENOMEM; | |
398 | if (buf == NULL) | |
399 | goto out; | |
400 | ||
401 | len = posix_acl_to_xattr(pacl, buf, buflen); | |
402 | if (len < 0) { | |
403 | error = len; | |
404 | goto out; | |
405 | } | |
406 | ||
5be196e5 | 407 | error = vfs_setxattr(dentry, key, buf, len, 0); |
1da177e4 LT |
408 | out: |
409 | kfree(buf); | |
410 | return error; | |
411 | } | |
412 | ||
413 | int | |
414 | nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
415 | struct nfs4_acl *acl) | |
416 | { | |
417 | int error; | |
418 | struct dentry *dentry; | |
419 | struct inode *inode; | |
420 | struct posix_acl *pacl = NULL, *dpacl = NULL; | |
421 | unsigned int flags = 0; | |
422 | ||
423 | /* Get inode */ | |
424 | error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR); | |
425 | if (error) | |
426 | goto out; | |
427 | ||
428 | dentry = fhp->fh_dentry; | |
429 | inode = dentry->d_inode; | |
430 | if (S_ISDIR(inode->i_mode)) | |
431 | flags = NFS4_ACL_DIR; | |
432 | ||
433 | error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); | |
434 | if (error == -EINVAL) { | |
435 | error = nfserr_attrnotsupp; | |
436 | goto out; | |
437 | } else if (error < 0) | |
438 | goto out_nfserr; | |
439 | ||
440 | if (pacl) { | |
9a59f452 | 441 | error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); |
1da177e4 LT |
442 | if (error < 0) |
443 | goto out_nfserr; | |
444 | } | |
445 | ||
446 | if (dpacl) { | |
9a59f452 | 447 | error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); |
1da177e4 LT |
448 | if (error < 0) |
449 | goto out_nfserr; | |
450 | } | |
451 | ||
452 | error = nfs_ok; | |
453 | ||
454 | out: | |
455 | posix_acl_release(pacl); | |
456 | posix_acl_release(dpacl); | |
457 | return (error); | |
458 | out_nfserr: | |
459 | error = nfserrno(error); | |
460 | goto out; | |
461 | } | |
462 | ||
463 | static struct posix_acl * | |
464 | _get_posix_acl(struct dentry *dentry, char *key) | |
465 | { | |
5be196e5 | 466 | void *buf = NULL; |
1da177e4 | 467 | struct posix_acl *pacl = NULL; |
5be196e5 | 468 | int buflen; |
1da177e4 | 469 | |
5be196e5 CH |
470 | buflen = nfsd_getxattr(dentry, key, &buf); |
471 | if (!buflen) | |
472 | buflen = -ENODATA; | |
473 | if (buflen <= 0) | |
474 | return ERR_PTR(buflen); | |
1da177e4 LT |
475 | |
476 | pacl = posix_acl_from_xattr(buf, buflen); | |
1da177e4 LT |
477 | kfree(buf); |
478 | return pacl; | |
1da177e4 LT |
479 | } |
480 | ||
481 | int | |
482 | nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) | |
483 | { | |
484 | struct inode *inode = dentry->d_inode; | |
485 | int error = 0; | |
486 | struct posix_acl *pacl = NULL, *dpacl = NULL; | |
487 | unsigned int flags = 0; | |
488 | ||
9a59f452 | 489 | pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); |
1da177e4 LT |
490 | if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) |
491 | pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); | |
492 | if (IS_ERR(pacl)) { | |
493 | error = PTR_ERR(pacl); | |
494 | pacl = NULL; | |
495 | goto out; | |
496 | } | |
497 | ||
498 | if (S_ISDIR(inode->i_mode)) { | |
9a59f452 | 499 | dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); |
1da177e4 LT |
500 | if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) |
501 | dpacl = NULL; | |
502 | else if (IS_ERR(dpacl)) { | |
503 | error = PTR_ERR(dpacl); | |
504 | dpacl = NULL; | |
505 | goto out; | |
506 | } | |
507 | flags = NFS4_ACL_DIR; | |
508 | } | |
509 | ||
510 | *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); | |
511 | if (IS_ERR(*acl)) { | |
512 | error = PTR_ERR(*acl); | |
513 | *acl = NULL; | |
514 | } | |
515 | out: | |
516 | posix_acl_release(pacl); | |
517 | posix_acl_release(dpacl); | |
518 | return error; | |
519 | } | |
520 | ||
521 | #endif /* defined(CONFIG_NFS_V4) */ | |
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[] = { | |
532 | { NFS3_ACCESS_READ, MAY_READ }, | |
533 | { NFS3_ACCESS_EXECUTE, MAY_EXEC }, | |
534 | { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC }, | |
535 | { NFS3_ACCESS_EXTEND, MAY_WRITE }, | |
536 | ||
537 | { 0, 0 } | |
538 | }; | |
539 | ||
540 | static struct accessmap nfs3_diraccess[] = { | |
541 | { NFS3_ACCESS_READ, MAY_READ }, | |
542 | { NFS3_ACCESS_LOOKUP, MAY_EXEC }, | |
543 | { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC }, | |
544 | { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE }, | |
545 | { NFS3_ACCESS_DELETE, MAY_REMOVE }, | |
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 | */ | |
558 | { NFS3_ACCESS_READ, MAY_READ }, | |
559 | { NFS3_ACCESS_EXECUTE, MAY_EXEC }, | |
560 | { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS }, | |
561 | { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS }, | |
562 | ||
563 | { 0, 0 } | |
564 | }; | |
565 | ||
566 | int | |
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; | |
573 | unsigned int error; | |
574 | ||
575 | error = fh_verify(rqstp, fhp, 0, MAY_NOP); | |
576 | if (error) | |
577 | goto out; | |
578 | ||
579 | export = fhp->fh_export; | |
580 | dentry = fhp->fh_dentry; | |
581 | ||
582 | if (S_ISREG(dentry->d_inode->i_mode)) | |
583 | map = nfs3_regaccess; | |
584 | else if (S_ISDIR(dentry->d_inode->i_mode)) | |
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) { | |
593 | unsigned int err2; | |
594 | ||
595 | sresult |= map->access; | |
596 | ||
597 | err2 = nfsd_permission(export, dentry, map->how); | |
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 | ||
625 | ||
626 | ||
627 | /* | |
628 | * Open an existing file or directory. | |
629 | * The access argument indicates the type of open (read/write/lock) | |
630 | * N.B. After this call fhp needs an fh_put | |
631 | */ | |
632 | int | |
633 | nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, | |
634 | int access, struct file **filp) | |
635 | { | |
636 | struct dentry *dentry; | |
637 | struct inode *inode; | |
638 | int flags = O_RDONLY|O_LARGEFILE, err; | |
639 | ||
640 | /* | |
641 | * If we get here, then the client has already done an "open", | |
642 | * and (hopefully) checked permission - so allow OWNER_OVERRIDE | |
643 | * in case a chmod has now revoked permission. | |
644 | */ | |
645 | err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE); | |
646 | if (err) | |
647 | goto out; | |
648 | ||
649 | dentry = fhp->fh_dentry; | |
650 | inode = dentry->d_inode; | |
651 | ||
652 | /* Disallow write access to files with the append-only bit set | |
653 | * or any access when mandatory locking enabled | |
654 | */ | |
655 | err = nfserr_perm; | |
656 | if (IS_APPEND(inode) && (access & MAY_WRITE)) | |
657 | goto out; | |
658 | if (IS_ISMNDLK(inode)) | |
659 | goto out; | |
660 | ||
661 | if (!inode->i_fop) | |
662 | goto out; | |
663 | ||
664 | /* | |
665 | * Check to see if there are any leases on this file. | |
666 | * This may block while leases are broken. | |
667 | */ | |
668 | err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0)); | |
669 | if (err == -EWOULDBLOCK) | |
670 | err = -ETIMEDOUT; | |
671 | if (err) /* NOMEM or WOULDBLOCK */ | |
672 | goto out_nfserr; | |
673 | ||
674 | if (access & MAY_WRITE) { | |
9ecb6a08 BF |
675 | if (access & MAY_READ) |
676 | flags = O_RDWR|O_LARGEFILE; | |
677 | else | |
678 | flags = O_WRONLY|O_LARGEFILE; | |
1da177e4 LT |
679 | |
680 | DQUOT_INIT(inode); | |
681 | } | |
682 | *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags); | |
683 | if (IS_ERR(*filp)) | |
684 | err = PTR_ERR(*filp); | |
685 | out_nfserr: | |
686 | if (err) | |
687 | err = nfserrno(err); | |
688 | out: | |
689 | return err; | |
690 | } | |
691 | ||
692 | /* | |
693 | * Close a file. | |
694 | */ | |
695 | void | |
696 | nfsd_close(struct file *filp) | |
697 | { | |
698 | fput(filp); | |
699 | } | |
700 | ||
701 | /* | |
702 | * Sync a file | |
703 | * As this calls fsync (not fdatasync) there is no need for a write_inode | |
704 | * after it. | |
705 | */ | |
a334de28 | 706 | static inline int nfsd_dosync(struct file *filp, struct dentry *dp, |
99ac48f5 | 707 | const struct file_operations *fop) |
1da177e4 LT |
708 | { |
709 | struct inode *inode = dp->d_inode; | |
710 | int (*fsync) (struct file *, struct dentry *, int); | |
f193fbab | 711 | int err; |
1da177e4 | 712 | |
f193fbab YT |
713 | err = filemap_fdatawrite(inode->i_mapping); |
714 | if (err == 0 && fop && (fsync = fop->fsync)) | |
715 | err = fsync(filp, dp, 0); | |
716 | if (err == 0) | |
717 | err = filemap_fdatawait(inode->i_mapping); | |
a334de28 | 718 | |
f193fbab | 719 | return err; |
1da177e4 LT |
720 | } |
721 | ||
722 | ||
a334de28 | 723 | static int |
1da177e4 LT |
724 | nfsd_sync(struct file *filp) |
725 | { | |
a334de28 | 726 | int err; |
1da177e4 LT |
727 | struct inode *inode = filp->f_dentry->d_inode; |
728 | dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name); | |
1b1dcc1b | 729 | mutex_lock(&inode->i_mutex); |
a334de28 | 730 | err=nfsd_dosync(filp, filp->f_dentry, filp->f_op); |
1b1dcc1b | 731 | mutex_unlock(&inode->i_mutex); |
a334de28 DS |
732 | |
733 | return err; | |
1da177e4 LT |
734 | } |
735 | ||
f193fbab | 736 | int |
1da177e4 LT |
737 | nfsd_sync_dir(struct dentry *dp) |
738 | { | |
f193fbab | 739 | return nfsd_dosync(NULL, dp, dp->d_inode->i_fop); |
1da177e4 LT |
740 | } |
741 | ||
742 | /* | |
743 | * Obtain the readahead parameters for the file | |
744 | * specified by (dev, ino). | |
745 | */ | |
746 | static DEFINE_SPINLOCK(ra_lock); | |
747 | ||
748 | static inline struct raparms * | |
749 | nfsd_get_raparms(dev_t dev, ino_t ino) | |
750 | { | |
751 | struct raparms *ra, **rap, **frap = NULL; | |
752 | int depth = 0; | |
753 | ||
754 | spin_lock(&ra_lock); | |
755 | for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) { | |
756 | if (ra->p_ino == ino && ra->p_dev == dev) | |
757 | goto found; | |
758 | depth++; | |
759 | if (ra->p_count == 0) | |
760 | frap = rap; | |
761 | } | |
762 | depth = nfsdstats.ra_size*11/10; | |
763 | if (!frap) { | |
764 | spin_unlock(&ra_lock); | |
765 | return NULL; | |
766 | } | |
767 | rap = frap; | |
768 | ra = *frap; | |
769 | ra->p_dev = dev; | |
770 | ra->p_ino = ino; | |
771 | ra->p_set = 0; | |
772 | found: | |
773 | if (rap != &raparm_cache) { | |
774 | *rap = ra->p_next; | |
775 | ra->p_next = raparm_cache; | |
776 | raparm_cache = ra; | |
777 | } | |
778 | ra->p_count++; | |
779 | nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; | |
780 | spin_unlock(&ra_lock); | |
781 | return ra; | |
782 | } | |
783 | ||
784 | /* | |
785 | * Grab and keep cached pages assosiated with a file in the svc_rqst | |
786 | * so that they can be passed to the netowork sendmsg/sendpage routines | |
787 | * directrly. They will be released after the sending has completed. | |
788 | */ | |
789 | static int | |
790 | nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size) | |
791 | { | |
792 | unsigned long count = desc->count; | |
793 | struct svc_rqst *rqstp = desc->arg.data; | |
44524359 | 794 | struct page **pp = rqstp->rq_respages + rqstp->rq_resused; |
1da177e4 LT |
795 | |
796 | if (size > count) | |
797 | size = count; | |
798 | ||
799 | if (rqstp->rq_res.page_len == 0) { | |
800 | get_page(page); | |
44524359 N |
801 | put_page(*pp); |
802 | *pp = page; | |
803 | rqstp->rq_resused++; | |
1da177e4 LT |
804 | rqstp->rq_res.page_base = offset; |
805 | rqstp->rq_res.page_len = size; | |
44524359 | 806 | } else if (page != pp[-1]) { |
1da177e4 | 807 | get_page(page); |
44524359 N |
808 | put_page(*pp); |
809 | *pp = page; | |
810 | rqstp->rq_resused++; | |
1da177e4 | 811 | rqstp->rq_res.page_len += size; |
44524359 | 812 | } else |
1da177e4 | 813 | rqstp->rq_res.page_len += size; |
1da177e4 LT |
814 | |
815 | desc->count = count - size; | |
816 | desc->written += size; | |
817 | return size; | |
818 | } | |
819 | ||
7e8f0593 | 820 | static int |
1da177e4 LT |
821 | nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
822 | loff_t offset, struct kvec *vec, int vlen, unsigned long *count) | |
823 | { | |
824 | struct inode *inode; | |
825 | struct raparms *ra; | |
826 | mm_segment_t oldfs; | |
827 | int err; | |
828 | ||
829 | err = nfserr_perm; | |
830 | inode = file->f_dentry->d_inode; | |
831 | #ifdef MSNFS | |
832 | if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && | |
833 | (!lock_may_read(inode, offset, *count))) | |
834 | goto out; | |
835 | #endif | |
836 | ||
837 | /* Get readahead parameters */ | |
838 | ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); | |
839 | ||
840 | if (ra && ra->p_set) | |
841 | file->f_ra = ra->p_ra; | |
842 | ||
5c04c46a | 843 | if (file->f_op->sendfile && rqstp->rq_sendfile_ok) { |
44524359 | 844 | rqstp->rq_resused = 1; |
1da177e4 LT |
845 | err = file->f_op->sendfile(file, &offset, *count, |
846 | nfsd_read_actor, rqstp); | |
847 | } else { | |
848 | oldfs = get_fs(); | |
849 | set_fs(KERNEL_DS); | |
850 | err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); | |
851 | set_fs(oldfs); | |
852 | } | |
853 | ||
854 | /* Write back readahead params */ | |
855 | if (ra) { | |
856 | spin_lock(&ra_lock); | |
857 | ra->p_ra = file->f_ra; | |
858 | ra->p_set = 1; | |
859 | ra->p_count--; | |
860 | spin_unlock(&ra_lock); | |
861 | } | |
862 | ||
863 | if (err >= 0) { | |
864 | nfsdstats.io_read += err; | |
865 | *count = err; | |
866 | err = 0; | |
0eeca283 | 867 | fsnotify_access(file->f_dentry); |
1da177e4 LT |
868 | } else |
869 | err = nfserrno(err); | |
870 | out: | |
871 | return err; | |
872 | } | |
873 | ||
9f708e40 NB |
874 | static void kill_suid(struct dentry *dentry) |
875 | { | |
876 | struct iattr ia; | |
877 | ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID; | |
878 | ||
1b1dcc1b | 879 | mutex_lock(&dentry->d_inode->i_mutex); |
9f708e40 | 880 | notify_change(dentry, &ia); |
1b1dcc1b | 881 | mutex_unlock(&dentry->d_inode->i_mutex); |
9f708e40 NB |
882 | } |
883 | ||
7e8f0593 | 884 | static int |
1da177e4 LT |
885 | nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
886 | loff_t offset, struct kvec *vec, int vlen, | |
887 | unsigned long cnt, int *stablep) | |
888 | { | |
889 | struct svc_export *exp; | |
890 | struct dentry *dentry; | |
891 | struct inode *inode; | |
892 | mm_segment_t oldfs; | |
893 | int err = 0; | |
894 | int stable = *stablep; | |
895 | ||
45bd3b3d | 896 | #ifdef MSNFS |
1da177e4 LT |
897 | err = nfserr_perm; |
898 | ||
1da177e4 LT |
899 | if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && |
900 | (!lock_may_write(file->f_dentry->d_inode, offset, cnt))) | |
901 | goto out; | |
902 | #endif | |
903 | ||
904 | dentry = file->f_dentry; | |
905 | inode = dentry->d_inode; | |
906 | exp = fhp->fh_export; | |
907 | ||
908 | /* | |
909 | * Request sync writes if | |
910 | * - the sync export option has been set, or | |
911 | * - the client requested O_SYNC behavior (NFSv3 feature). | |
912 | * - The file system doesn't support fsync(). | |
913 | * When gathered writes have been configured for this volume, | |
914 | * flushing the data to disk is handled separately below. | |
915 | */ | |
916 | ||
917 | if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */ | |
918 | stable = 2; | |
919 | *stablep = 2; /* FILE_SYNC */ | |
920 | } | |
921 | ||
922 | if (!EX_ISSYNC(exp)) | |
923 | stable = 0; | |
924 | if (stable && !EX_WGATHER(exp)) | |
925 | file->f_flags |= O_SYNC; | |
926 | ||
927 | /* Write the data. */ | |
928 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
929 | err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); | |
930 | set_fs(oldfs); | |
931 | if (err >= 0) { | |
932 | nfsdstats.io_write += cnt; | |
0eeca283 | 933 | fsnotify_modify(file->f_dentry); |
1da177e4 LT |
934 | } |
935 | ||
936 | /* clear setuid/setgid flag after write */ | |
9f708e40 NB |
937 | if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) |
938 | kill_suid(dentry); | |
1da177e4 LT |
939 | |
940 | if (err >= 0 && stable) { | |
941 | static ino_t last_ino; | |
942 | static dev_t last_dev; | |
943 | ||
944 | /* | |
945 | * Gathered writes: If another process is currently | |
946 | * writing to the file, there's a high chance | |
947 | * this is another nfsd (triggered by a bulk write | |
948 | * from a client's biod). Rather than syncing the | |
949 | * file with each write request, we sleep for 10 msec. | |
950 | * | |
951 | * I don't know if this roughly approximates | |
952 | * C. Juszak's idea of gathered writes, but it's a | |
953 | * nice and simple solution (IMHO), and it seems to | |
954 | * work:-) | |
955 | */ | |
956 | if (EX_WGATHER(exp)) { | |
957 | if (atomic_read(&inode->i_writecount) > 1 | |
958 | || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { | |
959 | dprintk("nfsd: write defer %d\n", current->pid); | |
960 | msleep(10); | |
961 | dprintk("nfsd: write resume %d\n", current->pid); | |
962 | } | |
963 | ||
964 | if (inode->i_state & I_DIRTY) { | |
965 | dprintk("nfsd: write sync %d\n", current->pid); | |
a334de28 | 966 | err=nfsd_sync(file); |
1da177e4 LT |
967 | } |
968 | #if 0 | |
969 | wake_up(&inode->i_wait); | |
970 | #endif | |
971 | } | |
972 | last_ino = inode->i_ino; | |
973 | last_dev = inode->i_sb->s_dev; | |
974 | } | |
975 | ||
976 | dprintk("nfsd: write complete err=%d\n", err); | |
977 | if (err >= 0) | |
978 | err = 0; | |
979 | else | |
980 | err = nfserrno(err); | |
981 | out: | |
982 | return err; | |
983 | } | |
984 | ||
985 | /* | |
986 | * Read data from a file. count must contain the requested read count | |
987 | * on entry. On return, *count contains the number of bytes actually read. | |
988 | * N.B. After this call fhp needs an fh_put | |
989 | */ | |
990 | int | |
991 | nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, | |
992 | loff_t offset, struct kvec *vec, int vlen, | |
993 | unsigned long *count) | |
994 | { | |
995 | int err; | |
996 | ||
997 | if (file) { | |
998 | err = nfsd_permission(fhp->fh_export, fhp->fh_dentry, | |
999 | MAY_READ|MAY_OWNER_OVERRIDE); | |
1000 | if (err) | |
1001 | goto out; | |
1002 | err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); | |
1003 | } else { | |
1004 | err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file); | |
1005 | if (err) | |
1006 | goto out; | |
1007 | err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); | |
1008 | nfsd_close(file); | |
1009 | } | |
1010 | out: | |
1011 | return err; | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * Write data to a file. | |
1016 | * The stable flag requests synchronous writes. | |
1017 | * N.B. After this call fhp needs an fh_put | |
1018 | */ | |
1019 | int | |
1020 | nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, | |
1021 | loff_t offset, struct kvec *vec, int vlen, unsigned long cnt, | |
1022 | int *stablep) | |
1023 | { | |
1024 | int err = 0; | |
1025 | ||
1026 | if (file) { | |
1027 | err = nfsd_permission(fhp->fh_export, fhp->fh_dentry, | |
1028 | MAY_WRITE|MAY_OWNER_OVERRIDE); | |
1029 | if (err) | |
1030 | goto out; | |
1031 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, | |
1032 | stablep); | |
1033 | } else { | |
1034 | err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file); | |
1035 | if (err) | |
1036 | goto out; | |
1037 | ||
1038 | if (cnt) | |
1039 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, | |
1040 | cnt, stablep); | |
1041 | nfsd_close(file); | |
1042 | } | |
1043 | out: | |
1044 | return err; | |
1045 | } | |
1046 | ||
1047 | #ifdef CONFIG_NFSD_V3 | |
1048 | /* | |
1049 | * Commit all pending writes to stable storage. | |
1050 | * Strictly speaking, we could sync just the indicated file region here, | |
1051 | * but there's currently no way we can ask the VFS to do so. | |
1052 | * | |
1053 | * Unfortunately we cannot lock the file to make sure we return full WCC | |
1054 | * data to the client, as locking happens lower down in the filesystem. | |
1055 | */ | |
1056 | int | |
1057 | nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1058 | loff_t offset, unsigned long count) | |
1059 | { | |
1060 | struct file *file; | |
1061 | int err; | |
1062 | ||
1063 | if ((u64)count > ~(u64)offset) | |
1064 | return nfserr_inval; | |
1065 | ||
1066 | if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0) | |
1067 | return err; | |
1068 | if (EX_ISSYNC(fhp->fh_export)) { | |
1069 | if (file->f_op && file->f_op->fsync) { | |
45bd3b3d | 1070 | err = nfserrno(nfsd_sync(file)); |
1da177e4 LT |
1071 | } else { |
1072 | err = nfserr_notsupp; | |
1073 | } | |
1074 | } | |
1075 | ||
1076 | nfsd_close(file); | |
1077 | return err; | |
1078 | } | |
1079 | #endif /* CONFIG_NFSD_V3 */ | |
1080 | ||
1081 | /* | |
1082 | * Create a file (regular, directory, device, fifo); UNIX sockets | |
1083 | * not yet implemented. | |
1084 | * If the response fh has been verified, the parent directory should | |
1085 | * already be locked. Note that the parent directory is left locked. | |
1086 | * | |
1087 | * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp | |
1088 | */ | |
1089 | int | |
1090 | nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1091 | char *fname, int flen, struct iattr *iap, | |
1092 | int type, dev_t rdev, struct svc_fh *resfhp) | |
1093 | { | |
1094 | struct dentry *dentry, *dchild = NULL; | |
1095 | struct inode *dirp; | |
1096 | int err; | |
1097 | ||
1098 | err = nfserr_perm; | |
1099 | if (!flen) | |
1100 | goto out; | |
1101 | err = nfserr_exist; | |
1102 | if (isdotent(fname, flen)) | |
1103 | goto out; | |
1104 | ||
1105 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); | |
1106 | if (err) | |
1107 | goto out; | |
1108 | ||
1109 | dentry = fhp->fh_dentry; | |
1110 | dirp = dentry->d_inode; | |
1111 | ||
1112 | err = nfserr_notdir; | |
1113 | if(!dirp->i_op || !dirp->i_op->lookup) | |
1114 | goto out; | |
1115 | /* | |
1116 | * Check whether the response file handle has been verified yet. | |
1117 | * If it has, the parent directory should already be locked. | |
1118 | */ | |
1119 | if (!resfhp->fh_dentry) { | |
1120 | /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ | |
12fd3520 | 1121 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1122 | dchild = lookup_one_len(fname, dentry, flen); |
1123 | err = PTR_ERR(dchild); | |
1124 | if (IS_ERR(dchild)) | |
1125 | goto out_nfserr; | |
1126 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); | |
1127 | if (err) | |
1128 | goto out; | |
1129 | } else { | |
1130 | /* called from nfsd_proc_create */ | |
1131 | dchild = dget(resfhp->fh_dentry); | |
1132 | if (!fhp->fh_locked) { | |
1133 | /* not actually possible */ | |
1134 | printk(KERN_ERR | |
1135 | "nfsd_create: parent %s/%s not locked!\n", | |
1136 | dentry->d_parent->d_name.name, | |
1137 | dentry->d_name.name); | |
d75f2b9f | 1138 | err = nfserr_io; |
1da177e4 LT |
1139 | goto out; |
1140 | } | |
1141 | } | |
1142 | /* | |
1143 | * Make sure the child dentry is still negative ... | |
1144 | */ | |
1145 | err = nfserr_exist; | |
1146 | if (dchild->d_inode) { | |
1147 | dprintk("nfsd_create: dentry %s/%s not negative!\n", | |
1148 | dentry->d_name.name, dchild->d_name.name); | |
1149 | goto out; | |
1150 | } | |
1151 | ||
1152 | if (!(iap->ia_valid & ATTR_MODE)) | |
1153 | iap->ia_mode = 0; | |
1154 | iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; | |
1155 | ||
1156 | /* | |
1157 | * Get the dir op function pointer. | |
1158 | */ | |
1159 | err = nfserr_perm; | |
1160 | switch (type) { | |
1161 | case S_IFREG: | |
1162 | err = vfs_create(dirp, dchild, iap->ia_mode, NULL); | |
1163 | break; | |
1164 | case S_IFDIR: | |
1165 | err = vfs_mkdir(dirp, dchild, iap->ia_mode); | |
1166 | break; | |
1167 | case S_IFCHR: | |
1168 | case S_IFBLK: | |
1169 | case S_IFIFO: | |
1170 | case S_IFSOCK: | |
1171 | err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); | |
1172 | break; | |
1173 | default: | |
1174 | printk("nfsd: bad file type %o in nfsd_create\n", type); | |
1175 | err = -EINVAL; | |
1176 | } | |
1177 | if (err < 0) | |
1178 | goto out_nfserr; | |
1179 | ||
1180 | if (EX_ISSYNC(fhp->fh_export)) { | |
45bd3b3d | 1181 | err = nfserrno(nfsd_sync_dir(dentry)); |
1da177e4 LT |
1182 | write_inode_now(dchild->d_inode, 1); |
1183 | } | |
1184 | ||
1185 | ||
1186 | /* Set file attributes. Mode has already been set and | |
1187 | * setting uid/gid works only for root. Irix appears to | |
1188 | * send along the gid when it tries to implement setgid | |
1189 | * directories via NFS. | |
1190 | */ | |
f193fbab YT |
1191 | if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) { |
1192 | int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); | |
1193 | if (err2) | |
1194 | err = err2; | |
1195 | } | |
1da177e4 LT |
1196 | /* |
1197 | * Update the file handle to get the new inode info. | |
1198 | */ | |
1199 | if (!err) | |
1200 | err = fh_update(resfhp); | |
1201 | out: | |
1202 | if (dchild && !IS_ERR(dchild)) | |
1203 | dput(dchild); | |
1204 | return err; | |
1205 | ||
1206 | out_nfserr: | |
1207 | err = nfserrno(err); | |
1208 | goto out; | |
1209 | } | |
1210 | ||
1211 | #ifdef CONFIG_NFSD_V3 | |
1212 | /* | |
1213 | * NFSv3 version of nfsd_create | |
1214 | */ | |
1215 | int | |
1216 | nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1217 | char *fname, int flen, struct iattr *iap, | |
1218 | struct svc_fh *resfhp, int createmode, u32 *verifier, | |
1219 | int *truncp) | |
1220 | { | |
1221 | struct dentry *dentry, *dchild = NULL; | |
1222 | struct inode *dirp; | |
1223 | int err; | |
1224 | __u32 v_mtime=0, v_atime=0; | |
1225 | int v_mode=0; | |
1226 | ||
1227 | err = nfserr_perm; | |
1228 | if (!flen) | |
1229 | goto out; | |
1230 | err = nfserr_exist; | |
1231 | if (isdotent(fname, flen)) | |
1232 | goto out; | |
1233 | if (!(iap->ia_valid & ATTR_MODE)) | |
1234 | iap->ia_mode = 0; | |
1235 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); | |
1236 | if (err) | |
1237 | goto out; | |
1238 | ||
1239 | dentry = fhp->fh_dentry; | |
1240 | dirp = dentry->d_inode; | |
1241 | ||
1242 | /* Get all the sanity checks out of the way before | |
1243 | * we lock the parent. */ | |
1244 | err = nfserr_notdir; | |
1245 | if(!dirp->i_op || !dirp->i_op->lookup) | |
1246 | goto out; | |
12fd3520 | 1247 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1248 | |
1249 | /* | |
1250 | * Compose the response file handle. | |
1251 | */ | |
1252 | dchild = lookup_one_len(fname, dentry, flen); | |
1253 | err = PTR_ERR(dchild); | |
1254 | if (IS_ERR(dchild)) | |
1255 | goto out_nfserr; | |
1256 | ||
1257 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); | |
1258 | if (err) | |
1259 | goto out; | |
1260 | ||
1261 | if (createmode == NFS3_CREATE_EXCLUSIVE) { | |
1262 | /* while the verifier would fit in mtime+atime, | |
1263 | * solaris7 gets confused (bugid 4218508) if these have | |
1264 | * the high bit set, so we use the mode as well | |
1265 | */ | |
1266 | v_mtime = verifier[0]&0x7fffffff; | |
1267 | v_atime = verifier[1]&0x7fffffff; | |
1268 | v_mode = S_IFREG | |
1269 | | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */ | |
1270 | | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */ | |
1271 | ; | |
1272 | } | |
1273 | ||
1274 | if (dchild->d_inode) { | |
1275 | err = 0; | |
1276 | ||
1277 | switch (createmode) { | |
1278 | case NFS3_CREATE_UNCHECKED: | |
1279 | if (! S_ISREG(dchild->d_inode->i_mode)) | |
1280 | err = nfserr_exist; | |
1281 | else if (truncp) { | |
1282 | /* in nfsv4, we need to treat this case a little | |
1283 | * differently. we don't want to truncate the | |
1284 | * file now; this would be wrong if the OPEN | |
1285 | * fails for some other reason. furthermore, | |
1286 | * if the size is nonzero, we should ignore it | |
1287 | * according to spec! | |
1288 | */ | |
1289 | *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; | |
1290 | } | |
1291 | else { | |
1292 | iap->ia_valid &= ATTR_SIZE; | |
1293 | goto set_attr; | |
1294 | } | |
1295 | break; | |
1296 | case NFS3_CREATE_EXCLUSIVE: | |
1297 | if ( dchild->d_inode->i_mtime.tv_sec == v_mtime | |
1298 | && dchild->d_inode->i_atime.tv_sec == v_atime | |
1299 | && dchild->d_inode->i_mode == v_mode | |
1300 | && dchild->d_inode->i_size == 0 ) | |
1301 | break; | |
1302 | /* fallthru */ | |
1303 | case NFS3_CREATE_GUARDED: | |
1304 | err = nfserr_exist; | |
1305 | } | |
1306 | goto out; | |
1307 | } | |
1308 | ||
1309 | err = vfs_create(dirp, dchild, iap->ia_mode, NULL); | |
1310 | if (err < 0) | |
1311 | goto out_nfserr; | |
1312 | ||
1313 | if (EX_ISSYNC(fhp->fh_export)) { | |
45bd3b3d | 1314 | err = nfserrno(nfsd_sync_dir(dentry)); |
1da177e4 LT |
1315 | /* setattr will sync the child (or not) */ |
1316 | } | |
1317 | ||
1da177e4 LT |
1318 | if (createmode == NFS3_CREATE_EXCLUSIVE) { |
1319 | /* Cram the verifier into atime/mtime/mode */ | |
1320 | iap->ia_valid = ATTR_MTIME|ATTR_ATIME | |
1321 | | ATTR_MTIME_SET|ATTR_ATIME_SET | |
1322 | | ATTR_MODE; | |
1323 | /* XXX someone who knows this better please fix it for nsec */ | |
1324 | iap->ia_mtime.tv_sec = v_mtime; | |
1325 | iap->ia_atime.tv_sec = v_atime; | |
1326 | iap->ia_mtime.tv_nsec = 0; | |
1327 | iap->ia_atime.tv_nsec = 0; | |
1328 | iap->ia_mode = v_mode; | |
1329 | } | |
1330 | ||
1331 | /* Set file attributes. | |
1332 | * Mode has already been set but we might need to reset it | |
1333 | * for CREATE_EXCLUSIVE | |
1334 | * Irix appears to send along the gid when it tries to | |
1335 | * implement setgid directories via NFS. Clear out all that cruft. | |
1336 | */ | |
1337 | set_attr: | |
f193fbab YT |
1338 | if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) { |
1339 | int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); | |
1340 | if (err2) | |
45bd3b3d | 1341 | err = err2; |
f193fbab YT |
1342 | } |
1343 | ||
1344 | /* | |
1345 | * Update the filehandle to get the new inode info. | |
1346 | */ | |
1347 | if (!err) | |
1348 | err = fh_update(resfhp); | |
1da177e4 LT |
1349 | |
1350 | out: | |
1351 | fh_unlock(fhp); | |
1352 | if (dchild && !IS_ERR(dchild)) | |
1353 | dput(dchild); | |
1354 | return err; | |
1355 | ||
1356 | out_nfserr: | |
1357 | err = nfserrno(err); | |
1358 | goto out; | |
1359 | } | |
1360 | #endif /* CONFIG_NFSD_V3 */ | |
1361 | ||
1362 | /* | |
1363 | * Read a symlink. On entry, *lenp must contain the maximum path length that | |
1364 | * fits into the buffer. On return, it contains the true length. | |
1365 | * N.B. After this call fhp needs an fh_put | |
1366 | */ | |
1367 | int | |
1368 | nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) | |
1369 | { | |
1370 | struct dentry *dentry; | |
1371 | struct inode *inode; | |
1372 | mm_segment_t oldfs; | |
1373 | int err; | |
1374 | ||
1375 | err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP); | |
1376 | if (err) | |
1377 | goto out; | |
1378 | ||
1379 | dentry = fhp->fh_dentry; | |
1380 | inode = dentry->d_inode; | |
1381 | ||
1382 | err = nfserr_inval; | |
1383 | if (!inode->i_op || !inode->i_op->readlink) | |
1384 | goto out; | |
1385 | ||
1386 | touch_atime(fhp->fh_export->ex_mnt, dentry); | |
1387 | /* N.B. Why does this call need a get_fs()?? | |
1388 | * Remove the set_fs and watch the fireworks:-) --okir | |
1389 | */ | |
1390 | ||
1391 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
1392 | err = inode->i_op->readlink(dentry, buf, *lenp); | |
1393 | set_fs(oldfs); | |
1394 | ||
1395 | if (err < 0) | |
1396 | goto out_nfserr; | |
1397 | *lenp = err; | |
1398 | err = 0; | |
1399 | out: | |
1400 | return err; | |
1401 | ||
1402 | out_nfserr: | |
1403 | err = nfserrno(err); | |
1404 | goto out; | |
1405 | } | |
1406 | ||
1407 | /* | |
1408 | * Create a symlink and look up its inode | |
1409 | * N.B. After this call _both_ fhp and resfhp need an fh_put | |
1410 | */ | |
1411 | int | |
1412 | nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1413 | char *fname, int flen, | |
1414 | char *path, int plen, | |
1415 | struct svc_fh *resfhp, | |
1416 | struct iattr *iap) | |
1417 | { | |
1418 | struct dentry *dentry, *dnew; | |
1419 | int err, cerr; | |
1420 | umode_t mode; | |
1421 | ||
1422 | err = nfserr_noent; | |
1423 | if (!flen || !plen) | |
1424 | goto out; | |
1425 | err = nfserr_exist; | |
1426 | if (isdotent(fname, flen)) | |
1427 | goto out; | |
1428 | ||
1429 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); | |
1430 | if (err) | |
1431 | goto out; | |
1432 | fh_lock(fhp); | |
1433 | dentry = fhp->fh_dentry; | |
1434 | dnew = lookup_one_len(fname, dentry, flen); | |
1435 | err = PTR_ERR(dnew); | |
1436 | if (IS_ERR(dnew)) | |
1437 | goto out_nfserr; | |
1438 | ||
1439 | mode = S_IALLUGO; | |
1440 | /* Only the MODE ATTRibute is even vaguely meaningful */ | |
1441 | if (iap && (iap->ia_valid & ATTR_MODE)) | |
1442 | mode = iap->ia_mode & S_IALLUGO; | |
1443 | ||
1444 | if (unlikely(path[plen] != 0)) { | |
1445 | char *path_alloced = kmalloc(plen+1, GFP_KERNEL); | |
1446 | if (path_alloced == NULL) | |
1447 | err = -ENOMEM; | |
1448 | else { | |
1449 | strncpy(path_alloced, path, plen); | |
1450 | path_alloced[plen] = 0; | |
1451 | err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode); | |
1452 | kfree(path_alloced); | |
1453 | } | |
1454 | } else | |
1455 | err = vfs_symlink(dentry->d_inode, dnew, path, mode); | |
1456 | ||
f193fbab | 1457 | if (!err) |
1da177e4 | 1458 | if (EX_ISSYNC(fhp->fh_export)) |
f193fbab YT |
1459 | err = nfsd_sync_dir(dentry); |
1460 | if (err) | |
1da177e4 LT |
1461 | err = nfserrno(err); |
1462 | fh_unlock(fhp); | |
1463 | ||
1464 | cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); | |
1465 | dput(dnew); | |
1466 | if (err==0) err = cerr; | |
1467 | out: | |
1468 | return err; | |
1469 | ||
1470 | out_nfserr: | |
1471 | err = nfserrno(err); | |
1472 | goto out; | |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * Create a hardlink | |
1477 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1478 | */ | |
1479 | int | |
1480 | nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, | |
1481 | char *name, int len, struct svc_fh *tfhp) | |
1482 | { | |
1483 | struct dentry *ddir, *dnew, *dold; | |
1484 | struct inode *dirp, *dest; | |
1485 | int err; | |
1486 | ||
1487 | err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE); | |
1488 | if (err) | |
1489 | goto out; | |
1490 | err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP); | |
1491 | if (err) | |
1492 | goto out; | |
1493 | ||
1494 | err = nfserr_perm; | |
1495 | if (!len) | |
1496 | goto out; | |
1497 | err = nfserr_exist; | |
1498 | if (isdotent(name, len)) | |
1499 | goto out; | |
1500 | ||
12fd3520 | 1501 | fh_lock_nested(ffhp, I_MUTEX_PARENT); |
1da177e4 LT |
1502 | ddir = ffhp->fh_dentry; |
1503 | dirp = ddir->d_inode; | |
1504 | ||
1505 | dnew = lookup_one_len(name, ddir, len); | |
1506 | err = PTR_ERR(dnew); | |
1507 | if (IS_ERR(dnew)) | |
1508 | goto out_nfserr; | |
1509 | ||
1510 | dold = tfhp->fh_dentry; | |
1511 | dest = dold->d_inode; | |
1512 | ||
1513 | err = vfs_link(dold, dirp, dnew); | |
1514 | if (!err) { | |
1515 | if (EX_ISSYNC(ffhp->fh_export)) { | |
45bd3b3d | 1516 | err = nfserrno(nfsd_sync_dir(ddir)); |
1da177e4 LT |
1517 | write_inode_now(dest, 1); |
1518 | } | |
1519 | } else { | |
1520 | if (err == -EXDEV && rqstp->rq_vers == 2) | |
1521 | err = nfserr_acces; | |
1522 | else | |
1523 | err = nfserrno(err); | |
1524 | } | |
1525 | ||
1da177e4 | 1526 | dput(dnew); |
270d56e5 DR |
1527 | out_unlock: |
1528 | fh_unlock(ffhp); | |
1da177e4 LT |
1529 | out: |
1530 | return err; | |
1531 | ||
1532 | out_nfserr: | |
1533 | err = nfserrno(err); | |
270d56e5 | 1534 | goto out_unlock; |
1da177e4 LT |
1535 | } |
1536 | ||
1537 | /* | |
1538 | * Rename a file | |
1539 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1540 | */ | |
1541 | int | |
1542 | nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, | |
1543 | struct svc_fh *tfhp, char *tname, int tlen) | |
1544 | { | |
1545 | struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; | |
1546 | struct inode *fdir, *tdir; | |
1547 | int err; | |
1548 | ||
1549 | err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE); | |
1550 | if (err) | |
1551 | goto out; | |
1552 | err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE); | |
1553 | if (err) | |
1554 | goto out; | |
1555 | ||
1556 | fdentry = ffhp->fh_dentry; | |
1557 | fdir = fdentry->d_inode; | |
1558 | ||
1559 | tdentry = tfhp->fh_dentry; | |
1560 | tdir = tdentry->d_inode; | |
1561 | ||
1562 | err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; | |
a56f3937 | 1563 | if (ffhp->fh_export != tfhp->fh_export) |
1da177e4 LT |
1564 | goto out; |
1565 | ||
1566 | err = nfserr_perm; | |
1567 | if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) | |
1568 | goto out; | |
1569 | ||
1570 | /* cannot use fh_lock as we need deadlock protective ordering | |
1571 | * so do it by hand */ | |
1572 | trap = lock_rename(tdentry, fdentry); | |
1573 | ffhp->fh_locked = tfhp->fh_locked = 1; | |
1574 | fill_pre_wcc(ffhp); | |
1575 | fill_pre_wcc(tfhp); | |
1576 | ||
1577 | odentry = lookup_one_len(fname, fdentry, flen); | |
1578 | err = PTR_ERR(odentry); | |
1579 | if (IS_ERR(odentry)) | |
1580 | goto out_nfserr; | |
1581 | ||
1582 | err = -ENOENT; | |
1583 | if (!odentry->d_inode) | |
1584 | goto out_dput_old; | |
1585 | err = -EINVAL; | |
1586 | if (odentry == trap) | |
1587 | goto out_dput_old; | |
1588 | ||
1589 | ndentry = lookup_one_len(tname, tdentry, tlen); | |
1590 | err = PTR_ERR(ndentry); | |
1591 | if (IS_ERR(ndentry)) | |
1592 | goto out_dput_old; | |
1593 | err = -ENOTEMPTY; | |
1594 | if (ndentry == trap) | |
1595 | goto out_dput_new; | |
1596 | ||
1597 | #ifdef MSNFS | |
1598 | if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) && | |
1599 | ((atomic_read(&odentry->d_count) > 1) | |
1600 | || (atomic_read(&ndentry->d_count) > 1))) { | |
d75f2b9f | 1601 | err = -EPERM; |
1da177e4 LT |
1602 | } else |
1603 | #endif | |
1604 | err = vfs_rename(fdir, odentry, tdir, ndentry); | |
1605 | if (!err && EX_ISSYNC(tfhp->fh_export)) { | |
f193fbab YT |
1606 | err = nfsd_sync_dir(tdentry); |
1607 | if (!err) | |
1608 | err = nfsd_sync_dir(fdentry); | |
1da177e4 LT |
1609 | } |
1610 | ||
1611 | out_dput_new: | |
1612 | dput(ndentry); | |
1613 | out_dput_old: | |
1614 | dput(odentry); | |
1615 | out_nfserr: | |
1616 | if (err) | |
1617 | err = nfserrno(err); | |
1618 | ||
1619 | /* we cannot reply on fh_unlock on the two filehandles, | |
1620 | * as that would do the wrong thing if the two directories | |
1621 | * were the same, so again we do it by hand | |
1622 | */ | |
1623 | fill_post_wcc(ffhp); | |
1624 | fill_post_wcc(tfhp); | |
1625 | unlock_rename(tdentry, fdentry); | |
1626 | ffhp->fh_locked = tfhp->fh_locked = 0; | |
1627 | ||
1628 | out: | |
1629 | return err; | |
1630 | } | |
1631 | ||
1632 | /* | |
1633 | * Unlink a file or directory | |
1634 | * N.B. After this call fhp needs an fh_put | |
1635 | */ | |
1636 | int | |
1637 | nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, | |
1638 | char *fname, int flen) | |
1639 | { | |
1640 | struct dentry *dentry, *rdentry; | |
1641 | struct inode *dirp; | |
1642 | int err; | |
1643 | ||
1644 | err = nfserr_acces; | |
1645 | if (!flen || isdotent(fname, flen)) | |
1646 | goto out; | |
1647 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE); | |
1648 | if (err) | |
1649 | goto out; | |
1650 | ||
12fd3520 | 1651 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1652 | dentry = fhp->fh_dentry; |
1653 | dirp = dentry->d_inode; | |
1654 | ||
1655 | rdentry = lookup_one_len(fname, dentry, flen); | |
1656 | err = PTR_ERR(rdentry); | |
1657 | if (IS_ERR(rdentry)) | |
1658 | goto out_nfserr; | |
1659 | ||
1660 | if (!rdentry->d_inode) { | |
1661 | dput(rdentry); | |
1662 | err = nfserr_noent; | |
1663 | goto out; | |
1664 | } | |
1665 | ||
1666 | if (!type) | |
1667 | type = rdentry->d_inode->i_mode & S_IFMT; | |
1668 | ||
1669 | if (type != S_IFDIR) { /* It's UNLINK */ | |
1670 | #ifdef MSNFS | |
1671 | if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && | |
1672 | (atomic_read(&rdentry->d_count) > 1)) { | |
d75f2b9f | 1673 | err = -EPERM; |
1da177e4 LT |
1674 | } else |
1675 | #endif | |
1676 | err = vfs_unlink(dirp, rdentry); | |
1677 | } else { /* It's RMDIR */ | |
1678 | err = vfs_rmdir(dirp, rdentry); | |
1679 | } | |
1680 | ||
1681 | dput(rdentry); | |
1682 | ||
f193fbab YT |
1683 | if (err == 0 && |
1684 | EX_ISSYNC(fhp->fh_export)) | |
1685 | err = nfsd_sync_dir(dentry); | |
1da177e4 LT |
1686 | |
1687 | out_nfserr: | |
1688 | err = nfserrno(err); | |
f193fbab YT |
1689 | out: |
1690 | return err; | |
1da177e4 LT |
1691 | } |
1692 | ||
1693 | /* | |
1694 | * Read entries from a directory. | |
1695 | * The NFSv3/4 verifier we ignore for now. | |
1696 | */ | |
1697 | int | |
1698 | nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, | |
1699 | struct readdir_cd *cdp, encode_dent_fn func) | |
1700 | { | |
1701 | int err; | |
1702 | struct file *file; | |
1703 | loff_t offset = *offsetp; | |
1704 | ||
1705 | err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file); | |
1706 | if (err) | |
1707 | goto out; | |
1708 | ||
1709 | offset = vfs_llseek(file, offset, 0); | |
1710 | if (offset < 0) { | |
1711 | err = nfserrno((int)offset); | |
1712 | goto out_close; | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * Read the directory entries. This silly loop is necessary because | |
1717 | * readdir() is not guaranteed to fill up the entire buffer, but | |
1718 | * may choose to do less. | |
1719 | */ | |
1720 | ||
1721 | do { | |
1722 | cdp->err = nfserr_eof; /* will be cleared on successful read */ | |
1723 | err = vfs_readdir(file, (filldir_t) func, cdp); | |
1724 | } while (err >=0 && cdp->err == nfs_ok); | |
1725 | if (err) | |
1726 | err = nfserrno(err); | |
1727 | else | |
1728 | err = cdp->err; | |
1729 | *offsetp = vfs_llseek(file, 0, 1); | |
1730 | ||
1731 | if (err == nfserr_eof || err == nfserr_toosmall) | |
1732 | err = nfs_ok; /* can still be found in ->err */ | |
1733 | out_close: | |
1734 | nfsd_close(file); | |
1735 | out: | |
1736 | return err; | |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * Get file system stats | |
1741 | * N.B. After this call fhp needs an fh_put | |
1742 | */ | |
1743 | int | |
1744 | nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat) | |
1745 | { | |
1746 | int err = fh_verify(rqstp, fhp, 0, MAY_NOP); | |
726c3342 | 1747 | if (!err && vfs_statfs(fhp->fh_dentry,stat)) |
1da177e4 LT |
1748 | err = nfserr_io; |
1749 | return err; | |
1750 | } | |
1751 | ||
1752 | /* | |
1753 | * Check for a user's access permissions to this inode. | |
1754 | */ | |
1755 | int | |
1756 | nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc) | |
1757 | { | |
1758 | struct inode *inode = dentry->d_inode; | |
1759 | int err; | |
1760 | ||
1761 | if (acc == MAY_NOP) | |
1762 | return 0; | |
1763 | #if 0 | |
1764 | dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", | |
1765 | acc, | |
1766 | (acc & MAY_READ)? " read" : "", | |
1767 | (acc & MAY_WRITE)? " write" : "", | |
1768 | (acc & MAY_EXEC)? " exec" : "", | |
1769 | (acc & MAY_SATTR)? " sattr" : "", | |
1770 | (acc & MAY_TRUNC)? " trunc" : "", | |
1771 | (acc & MAY_LOCK)? " lock" : "", | |
1772 | (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "", | |
1773 | inode->i_mode, | |
1774 | IS_IMMUTABLE(inode)? " immut" : "", | |
1775 | IS_APPEND(inode)? " append" : "", | |
1776 | IS_RDONLY(inode)? " ro" : ""); | |
1777 | dprintk(" owner %d/%d user %d/%d\n", | |
1778 | inode->i_uid, inode->i_gid, current->fsuid, current->fsgid); | |
1779 | #endif | |
1780 | ||
1781 | /* Normally we reject any write/sattr etc access on a read-only file | |
1782 | * system. But if it is IRIX doing check on write-access for a | |
1783 | * device special file, we ignore rofs. | |
1784 | */ | |
1785 | if (!(acc & MAY_LOCAL_ACCESS)) | |
1786 | if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) { | |
1787 | if (EX_RDONLY(exp) || IS_RDONLY(inode)) | |
1788 | return nfserr_rofs; | |
1789 | if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode)) | |
1790 | return nfserr_perm; | |
1791 | } | |
1792 | if ((acc & MAY_TRUNC) && IS_APPEND(inode)) | |
1793 | return nfserr_perm; | |
1794 | ||
1795 | if (acc & MAY_LOCK) { | |
1796 | /* If we cannot rely on authentication in NLM requests, | |
1797 | * just allow locks, otherwise require read permission, or | |
1798 | * ownership | |
1799 | */ | |
1800 | if (exp->ex_flags & NFSEXP_NOAUTHNLM) | |
1801 | return 0; | |
1802 | else | |
1803 | acc = MAY_READ | MAY_OWNER_OVERRIDE; | |
1804 | } | |
1805 | /* | |
1806 | * The file owner always gets access permission for accesses that | |
1807 | * would normally be checked at open time. This is to make | |
1808 | * file access work even when the client has done a fchmod(fd, 0). | |
1809 | * | |
1810 | * However, `cp foo bar' should fail nevertheless when bar is | |
1811 | * readonly. A sensible way to do this might be to reject all | |
1812 | * attempts to truncate a read-only file, because a creat() call | |
1813 | * always implies file truncation. | |
1814 | * ... but this isn't really fair. A process may reasonably call | |
1815 | * ftruncate on an open file descriptor on a file with perm 000. | |
1816 | * We must trust the client to do permission checking - using "ACCESS" | |
1817 | * with NFSv3. | |
1818 | */ | |
1819 | if ((acc & MAY_OWNER_OVERRIDE) && | |
1820 | inode->i_uid == current->fsuid) | |
1821 | return 0; | |
1822 | ||
1823 | err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL); | |
1824 | ||
1825 | /* Allow read access to binaries even when mode 111 */ | |
1826 | if (err == -EACCES && S_ISREG(inode->i_mode) && | |
1827 | acc == (MAY_READ | MAY_OWNER_OVERRIDE)) | |
1828 | err = permission(inode, MAY_EXEC, NULL); | |
1829 | ||
1830 | return err? nfserrno(err) : 0; | |
1831 | } | |
1832 | ||
1833 | void | |
1834 | nfsd_racache_shutdown(void) | |
1835 | { | |
1836 | if (!raparm_cache) | |
1837 | return; | |
1838 | dprintk("nfsd: freeing readahead buffers.\n"); | |
1839 | kfree(raparml); | |
1840 | raparm_cache = raparml = NULL; | |
1841 | } | |
1842 | /* | |
1843 | * Initialize readahead param cache | |
1844 | */ | |
1845 | int | |
1846 | nfsd_racache_init(int cache_size) | |
1847 | { | |
1848 | int i; | |
1849 | ||
1850 | if (raparm_cache) | |
1851 | return 0; | |
1852 | raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL); | |
1853 | ||
1854 | if (raparml != NULL) { | |
1855 | dprintk("nfsd: allocating %d readahead buffers.\n", | |
1856 | cache_size); | |
1857 | memset(raparml, 0, sizeof(struct raparms) * cache_size); | |
1858 | for (i = 0; i < cache_size - 1; i++) { | |
1859 | raparml[i].p_next = raparml + i + 1; | |
1860 | } | |
1861 | raparm_cache = raparml; | |
1862 | } else { | |
1863 | printk(KERN_WARNING | |
1864 | "nfsd: Could not allocate memory read-ahead cache.\n"); | |
1865 | return -ENOMEM; | |
1866 | } | |
1867 | nfsdstats.ra_size = cache_size; | |
1868 | return 0; | |
1869 | } | |
a257cdd0 AG |
1870 | |
1871 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) | |
1872 | struct posix_acl * | |
1873 | nfsd_get_posix_acl(struct svc_fh *fhp, int type) | |
1874 | { | |
1875 | struct inode *inode = fhp->fh_dentry->d_inode; | |
1876 | char *name; | |
1877 | void *value = NULL; | |
1878 | ssize_t size; | |
1879 | struct posix_acl *acl; | |
1880 | ||
5be196e5 CH |
1881 | if (!IS_POSIXACL(inode)) |
1882 | return ERR_PTR(-EOPNOTSUPP); | |
1883 | ||
1884 | switch (type) { | |
1885 | case ACL_TYPE_ACCESS: | |
1886 | name = POSIX_ACL_XATTR_ACCESS; | |
1887 | break; | |
1888 | case ACL_TYPE_DEFAULT: | |
1889 | name = POSIX_ACL_XATTR_DEFAULT; | |
1890 | break; | |
1891 | default: | |
a257cdd0 | 1892 | return ERR_PTR(-EOPNOTSUPP); |
a257cdd0 AG |
1893 | } |
1894 | ||
5be196e5 CH |
1895 | size = nfsd_getxattr(fhp->fh_dentry, name, &value); |
1896 | if (size < 0) | |
1897 | return ERR_PTR(size); | |
a257cdd0 | 1898 | |
a257cdd0 | 1899 | acl = posix_acl_from_xattr(value, size); |
a257cdd0 AG |
1900 | kfree(value); |
1901 | return acl; | |
1902 | } | |
1903 | ||
1904 | int | |
1905 | nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) | |
1906 | { | |
1907 | struct inode *inode = fhp->fh_dentry->d_inode; | |
1908 | char *name; | |
1909 | void *value = NULL; | |
1910 | size_t size; | |
1911 | int error; | |
1912 | ||
1913 | if (!IS_POSIXACL(inode) || !inode->i_op || | |
1914 | !inode->i_op->setxattr || !inode->i_op->removexattr) | |
1915 | return -EOPNOTSUPP; | |
1916 | switch(type) { | |
1917 | case ACL_TYPE_ACCESS: | |
334a13ec | 1918 | name = POSIX_ACL_XATTR_ACCESS; |
a257cdd0 AG |
1919 | break; |
1920 | case ACL_TYPE_DEFAULT: | |
334a13ec | 1921 | name = POSIX_ACL_XATTR_DEFAULT; |
a257cdd0 AG |
1922 | break; |
1923 | default: | |
1924 | return -EOPNOTSUPP; | |
1925 | } | |
1926 | ||
1927 | if (acl && acl->a_count) { | |
334a13ec | 1928 | size = posix_acl_xattr_size(acl->a_count); |
a257cdd0 AG |
1929 | value = kmalloc(size, GFP_KERNEL); |
1930 | if (!value) | |
1931 | return -ENOMEM; | |
9ccfc29c FM |
1932 | error = posix_acl_to_xattr(acl, value, size); |
1933 | if (error < 0) | |
a257cdd0 | 1934 | goto getout; |
9ccfc29c | 1935 | size = error; |
a257cdd0 AG |
1936 | } else |
1937 | size = 0; | |
1938 | ||
a257cdd0 | 1939 | if (size) |
5be196e5 | 1940 | error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); |
a257cdd0 AG |
1941 | else { |
1942 | if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) | |
1943 | error = 0; | |
1944 | else { | |
5be196e5 | 1945 | error = vfs_removexattr(fhp->fh_dentry, name); |
a257cdd0 AG |
1946 | if (error == -ENODATA) |
1947 | error = 0; | |
1948 | } | |
1949 | } | |
1950 | ||
1951 | getout: | |
1952 | kfree(value); | |
1953 | return error; | |
1954 | } | |
1955 | #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ |