]>
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 */ | |
fce1456a | 57 | #include <linux/jhash.h> |
1da177e4 LT |
58 | |
59 | #include <asm/uaccess.h> | |
60 | ||
61 | #define NFSDDBG_FACILITY NFSDDBG_FILEOP | |
62 | #define NFSD_PARANOIA | |
63 | ||
64 | ||
65 | /* We must ignore files (but only files) which might have mandatory | |
66 | * locks on them because there is no way to know if the accesser has | |
67 | * the lock. | |
68 | */ | |
69 | #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i)) | |
70 | ||
71 | /* | |
72 | * This is a cache of readahead params that help us choose the proper | |
73 | * readahead strategy. Initially, we set all readahead parameters to 0 | |
74 | * and let the VFS handle things. | |
75 | * If you increase the number of cached files very much, you'll need to | |
76 | * add a hash table here. | |
77 | */ | |
78 | struct raparms { | |
79 | struct raparms *p_next; | |
80 | unsigned int p_count; | |
81 | ino_t p_ino; | |
82 | dev_t p_dev; | |
83 | int p_set; | |
84 | struct file_ra_state p_ra; | |
fce1456a | 85 | unsigned int p_hindex; |
1da177e4 LT |
86 | }; |
87 | ||
fce1456a GB |
88 | struct raparm_hbucket { |
89 | struct raparms *pb_head; | |
90 | spinlock_t pb_lock; | |
91 | } ____cacheline_aligned_in_smp; | |
92 | ||
1da177e4 | 93 | static struct raparms * raparml; |
fce1456a GB |
94 | #define RAPARM_HASH_BITS 4 |
95 | #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) | |
96 | #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) | |
97 | static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; | |
1da177e4 LT |
98 | |
99 | /* | |
100 | * Called from nfsd_lookup and encode_dirent. Check if we have crossed | |
101 | * a mount point. | |
102 | * Returns -EAGAIN leaving *dpp and *expp unchanged, | |
103 | * or nfs_ok having possibly changed *dpp and *expp | |
104 | */ | |
105 | int | |
106 | nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, | |
107 | struct svc_export **expp) | |
108 | { | |
109 | struct svc_export *exp = *expp, *exp2 = NULL; | |
110 | struct dentry *dentry = *dpp; | |
111 | struct vfsmount *mnt = mntget(exp->ex_mnt); | |
112 | struct dentry *mounts = dget(dentry); | |
113 | int err = nfs_ok; | |
114 | ||
115 | while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts)); | |
116 | ||
117 | exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle); | |
118 | if (IS_ERR(exp2)) { | |
119 | err = PTR_ERR(exp2); | |
120 | dput(mounts); | |
121 | mntput(mnt); | |
122 | goto out; | |
123 | } | |
124 | if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) { | |
125 | /* successfully crossed mount point */ | |
126 | exp_put(exp); | |
127 | *expp = exp2; | |
128 | dput(dentry); | |
129 | *dpp = mounts; | |
130 | } else { | |
131 | if (exp2) exp_put(exp2); | |
132 | dput(mounts); | |
133 | } | |
134 | mntput(mnt); | |
135 | out: | |
136 | return err; | |
137 | } | |
138 | ||
139 | /* | |
140 | * Look up one component of a pathname. | |
141 | * N.B. After this call _both_ fhp and resfh need an fh_put | |
142 | * | |
143 | * If the lookup would cross a mountpoint, and the mounted filesystem | |
144 | * is exported to the client with NFSEXP_NOHIDE, then the lookup is | |
145 | * accepted as it stands and the mounted directory is | |
146 | * returned. Otherwise the covered directory is returned. | |
147 | * NOTE: this mountpoint crossing is not supported properly by all | |
148 | * clients and is explicitly disallowed for NFSv3 | |
149 | * NeilBrown <neilb@cse.unsw.edu.au> | |
150 | */ | |
151 | int | |
152 | nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, | |
153 | int len, struct svc_fh *resfh) | |
154 | { | |
155 | struct svc_export *exp; | |
156 | struct dentry *dparent; | |
157 | struct dentry *dentry; | |
158 | int err; | |
159 | ||
160 | dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); | |
161 | ||
162 | /* Obtain dentry and export. */ | |
163 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC); | |
164 | if (err) | |
165 | return err; | |
166 | ||
167 | dparent = fhp->fh_dentry; | |
168 | exp = fhp->fh_export; | |
169 | exp_get(exp); | |
170 | ||
171 | err = nfserr_acces; | |
172 | ||
173 | /* Lookup the name, but don't follow links */ | |
174 | if (isdotent(name, len)) { | |
175 | if (len==1) | |
176 | dentry = dget(dparent); | |
177 | else if (dparent != exp->ex_dentry) { | |
178 | dentry = dget_parent(dparent); | |
179 | } else if (!EX_NOHIDE(exp)) | |
180 | dentry = dget(dparent); /* .. == . just like at / */ | |
181 | else { | |
182 | /* checking mountpoint crossing is very different when stepping up */ | |
183 | struct svc_export *exp2 = NULL; | |
184 | struct dentry *dp; | |
185 | struct vfsmount *mnt = mntget(exp->ex_mnt); | |
186 | dentry = dget(dparent); | |
187 | while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry)) | |
188 | ; | |
189 | dp = dget_parent(dentry); | |
190 | dput(dentry); | |
191 | dentry = dp; | |
192 | ||
193 | exp2 = exp_parent(exp->ex_client, mnt, dentry, | |
194 | &rqstp->rq_chandle); | |
195 | if (IS_ERR(exp2)) { | |
196 | err = PTR_ERR(exp2); | |
197 | dput(dentry); | |
198 | mntput(mnt); | |
199 | goto out_nfserr; | |
200 | } | |
201 | if (!exp2) { | |
202 | dput(dentry); | |
203 | dentry = dget(dparent); | |
204 | } else { | |
205 | exp_put(exp); | |
206 | exp = exp2; | |
207 | } | |
208 | mntput(mnt); | |
209 | } | |
210 | } else { | |
211 | fh_lock(fhp); | |
212 | dentry = lookup_one_len(name, dparent, len); | |
213 | err = PTR_ERR(dentry); | |
214 | if (IS_ERR(dentry)) | |
215 | goto out_nfserr; | |
216 | /* | |
217 | * check if we have crossed a mount point ... | |
218 | */ | |
219 | if (d_mountpoint(dentry)) { | |
220 | if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { | |
221 | dput(dentry); | |
222 | goto out_nfserr; | |
223 | } | |
224 | } | |
225 | } | |
226 | /* | |
227 | * Note: we compose the file handle now, but as the | |
228 | * dentry may be negative, it may need to be updated. | |
229 | */ | |
230 | err = fh_compose(resfh, exp, dentry, fhp); | |
231 | if (!err && !dentry->d_inode) | |
232 | err = nfserr_noent; | |
233 | dput(dentry); | |
234 | out: | |
235 | exp_put(exp); | |
236 | return err; | |
237 | ||
238 | out_nfserr: | |
239 | err = nfserrno(err); | |
240 | goto out; | |
241 | } | |
242 | ||
243 | /* | |
244 | * Set various file attributes. | |
245 | * N.B. After this call fhp needs an fh_put | |
246 | */ | |
247 | int | |
248 | nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, | |
249 | int check_guard, time_t guardtime) | |
250 | { | |
251 | struct dentry *dentry; | |
252 | struct inode *inode; | |
253 | int accmode = MAY_SATTR; | |
254 | int ftype = 0; | |
255 | int imode; | |
256 | int err; | |
257 | int size_change = 0; | |
258 | ||
259 | if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) | |
260 | accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE; | |
261 | if (iap->ia_valid & ATTR_SIZE) | |
262 | ftype = S_IFREG; | |
263 | ||
264 | /* Get inode */ | |
265 | err = fh_verify(rqstp, fhp, ftype, accmode); | |
15b7a1b8 | 266 | if (err) |
1da177e4 LT |
267 | goto out; |
268 | ||
269 | dentry = fhp->fh_dentry; | |
270 | inode = dentry->d_inode; | |
271 | ||
15b7a1b8 N |
272 | /* Ignore any mode updates on symlinks */ |
273 | if (S_ISLNK(inode->i_mode)) | |
274 | iap->ia_valid &= ~ATTR_MODE; | |
275 | ||
276 | if (!iap->ia_valid) | |
277 | goto out; | |
278 | ||
1da177e4 LT |
279 | /* NFSv2 does not differentiate between "set-[ac]time-to-now" |
280 | * which only requires access, and "set-[ac]time-to-X" which | |
281 | * requires ownership. | |
282 | * So if it looks like it might be "set both to the same time which | |
283 | * is close to now", and if inode_change_ok fails, then we | |
284 | * convert to "set to now" instead of "set to explicit time" | |
285 | * | |
286 | * We only call inode_change_ok as the last test as technically | |
287 | * it is not an interface that we should be using. It is only | |
288 | * valid if the filesystem does not define it's own i_op->setattr. | |
289 | */ | |
290 | #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) | |
291 | #define MAX_TOUCH_TIME_ERROR (30*60) | |
292 | if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET | |
293 | && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec | |
294 | ) { | |
295 | /* Looks probable. Now just make sure time is in the right ballpark. | |
296 | * Solaris, at least, doesn't seem to care what the time request is. | |
297 | * We require it be within 30 minutes of now. | |
298 | */ | |
299 | time_t delta = iap->ia_atime.tv_sec - get_seconds(); | |
300 | if (delta<0) delta = -delta; | |
301 | if (delta < MAX_TOUCH_TIME_ERROR && | |
302 | inode_change_ok(inode, iap) != 0) { | |
303 | /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME | |
304 | * this will cause notify_change to set these times to "now" | |
305 | */ | |
306 | iap->ia_valid &= ~BOTH_TIME_SET; | |
307 | } | |
308 | } | |
309 | ||
310 | /* The size case is special. It changes the file as well as the attributes. */ | |
311 | if (iap->ia_valid & ATTR_SIZE) { | |
312 | if (iap->ia_size < inode->i_size) { | |
313 | err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE); | |
314 | if (err) | |
315 | goto out; | |
316 | } | |
317 | ||
318 | /* | |
319 | * If we are changing the size of the file, then | |
320 | * we need to break all leases. | |
321 | */ | |
322 | err = break_lease(inode, FMODE_WRITE | O_NONBLOCK); | |
323 | if (err == -EWOULDBLOCK) | |
324 | err = -ETIMEDOUT; | |
325 | if (err) /* ENOMEM or EWOULDBLOCK */ | |
326 | goto out_nfserr; | |
327 | ||
328 | err = get_write_access(inode); | |
329 | if (err) | |
330 | goto out_nfserr; | |
331 | ||
332 | size_change = 1; | |
333 | err = locks_verify_truncate(inode, NULL, iap->ia_size); | |
334 | if (err) { | |
335 | put_write_access(inode); | |
336 | goto out_nfserr; | |
337 | } | |
338 | DQUOT_INIT(inode); | |
339 | } | |
340 | ||
341 | imode = inode->i_mode; | |
342 | if (iap->ia_valid & ATTR_MODE) { | |
343 | iap->ia_mode &= S_IALLUGO; | |
344 | imode = iap->ia_mode |= (imode & ~S_IALLUGO); | |
345 | } | |
346 | ||
347 | /* Revoke setuid/setgid bit on chown/chgrp */ | |
348 | if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) | |
349 | iap->ia_valid |= ATTR_KILL_SUID; | |
350 | if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid) | |
351 | iap->ia_valid |= ATTR_KILL_SGID; | |
352 | ||
353 | /* Change the attributes. */ | |
354 | ||
355 | iap->ia_valid |= ATTR_CTIME; | |
356 | ||
357 | err = nfserr_notsync; | |
358 | if (!check_guard || guardtime == inode->i_ctime.tv_sec) { | |
359 | fh_lock(fhp); | |
360 | err = notify_change(dentry, iap); | |
361 | err = nfserrno(err); | |
362 | fh_unlock(fhp); | |
363 | } | |
364 | if (size_change) | |
365 | put_write_access(inode); | |
366 | if (!err) | |
367 | if (EX_ISSYNC(fhp->fh_export)) | |
368 | write_inode_now(inode, 1); | |
369 | out: | |
370 | return err; | |
371 | ||
372 | out_nfserr: | |
373 | err = nfserrno(err); | |
374 | goto out; | |
375 | } | |
376 | ||
5be196e5 CH |
377 | #if defined(CONFIG_NFSD_V2_ACL) || \ |
378 | defined(CONFIG_NFSD_V3_ACL) || \ | |
379 | defined(CONFIG_NFSD_V4) | |
380 | static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) | |
381 | { | |
382 | ssize_t buflen; | |
5be196e5 CH |
383 | |
384 | buflen = vfs_getxattr(dentry, key, NULL, 0); | |
385 | if (buflen <= 0) | |
386 | return buflen; | |
1da177e4 | 387 | |
5be196e5 CH |
388 | *buf = kmalloc(buflen, GFP_KERNEL); |
389 | if (!*buf) | |
390 | return -ENOMEM; | |
391 | ||
b5872b0d | 392 | return vfs_getxattr(dentry, key, *buf, buflen); |
5be196e5 CH |
393 | } |
394 | #endif | |
395 | ||
396 | #if defined(CONFIG_NFSD_V4) | |
1da177e4 LT |
397 | static int |
398 | set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) | |
399 | { | |
400 | int len; | |
401 | size_t buflen; | |
402 | char *buf = NULL; | |
403 | int error = 0; | |
1da177e4 LT |
404 | |
405 | buflen = posix_acl_xattr_size(pacl->a_count); | |
406 | buf = kmalloc(buflen, GFP_KERNEL); | |
407 | error = -ENOMEM; | |
408 | if (buf == NULL) | |
409 | goto out; | |
410 | ||
411 | len = posix_acl_to_xattr(pacl, buf, buflen); | |
412 | if (len < 0) { | |
413 | error = len; | |
414 | goto out; | |
415 | } | |
416 | ||
5be196e5 | 417 | error = vfs_setxattr(dentry, key, buf, len, 0); |
1da177e4 LT |
418 | out: |
419 | kfree(buf); | |
420 | return error; | |
421 | } | |
422 | ||
423 | int | |
424 | nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
425 | struct nfs4_acl *acl) | |
426 | { | |
427 | int error; | |
428 | struct dentry *dentry; | |
429 | struct inode *inode; | |
430 | struct posix_acl *pacl = NULL, *dpacl = NULL; | |
431 | unsigned int flags = 0; | |
432 | ||
433 | /* Get inode */ | |
434 | error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR); | |
435 | if (error) | |
436 | goto out; | |
437 | ||
438 | dentry = fhp->fh_dentry; | |
439 | inode = dentry->d_inode; | |
440 | if (S_ISDIR(inode->i_mode)) | |
441 | flags = NFS4_ACL_DIR; | |
442 | ||
443 | error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); | |
444 | if (error == -EINVAL) { | |
445 | error = nfserr_attrnotsupp; | |
446 | goto out; | |
447 | } else if (error < 0) | |
448 | goto out_nfserr; | |
449 | ||
b66285ce BF |
450 | error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); |
451 | if (error < 0) | |
452 | goto out_nfserr; | |
1da177e4 | 453 | |
b66285ce | 454 | if (S_ISDIR(inode->i_mode)) { |
9a59f452 | 455 | error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); |
1da177e4 LT |
456 | if (error < 0) |
457 | goto out_nfserr; | |
458 | } | |
459 | ||
460 | error = nfs_ok; | |
461 | ||
462 | out: | |
463 | posix_acl_release(pacl); | |
464 | posix_acl_release(dpacl); | |
465 | return (error); | |
466 | out_nfserr: | |
467 | error = nfserrno(error); | |
468 | goto out; | |
469 | } | |
470 | ||
471 | static struct posix_acl * | |
472 | _get_posix_acl(struct dentry *dentry, char *key) | |
473 | { | |
5be196e5 | 474 | void *buf = NULL; |
1da177e4 | 475 | struct posix_acl *pacl = NULL; |
5be196e5 | 476 | int buflen; |
1da177e4 | 477 | |
5be196e5 CH |
478 | buflen = nfsd_getxattr(dentry, key, &buf); |
479 | if (!buflen) | |
480 | buflen = -ENODATA; | |
481 | if (buflen <= 0) | |
482 | return ERR_PTR(buflen); | |
1da177e4 LT |
483 | |
484 | pacl = posix_acl_from_xattr(buf, buflen); | |
1da177e4 LT |
485 | kfree(buf); |
486 | return pacl; | |
1da177e4 LT |
487 | } |
488 | ||
489 | int | |
490 | nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) | |
491 | { | |
492 | struct inode *inode = dentry->d_inode; | |
493 | int error = 0; | |
494 | struct posix_acl *pacl = NULL, *dpacl = NULL; | |
495 | unsigned int flags = 0; | |
496 | ||
9a59f452 | 497 | pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); |
1da177e4 LT |
498 | if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) |
499 | pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); | |
500 | if (IS_ERR(pacl)) { | |
501 | error = PTR_ERR(pacl); | |
502 | pacl = NULL; | |
503 | goto out; | |
504 | } | |
505 | ||
506 | if (S_ISDIR(inode->i_mode)) { | |
9a59f452 | 507 | dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); |
1da177e4 LT |
508 | if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) |
509 | dpacl = NULL; | |
510 | else if (IS_ERR(dpacl)) { | |
511 | error = PTR_ERR(dpacl); | |
512 | dpacl = NULL; | |
513 | goto out; | |
514 | } | |
515 | flags = NFS4_ACL_DIR; | |
516 | } | |
517 | ||
518 | *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); | |
519 | if (IS_ERR(*acl)) { | |
520 | error = PTR_ERR(*acl); | |
521 | *acl = NULL; | |
522 | } | |
523 | out: | |
524 | posix_acl_release(pacl); | |
525 | posix_acl_release(dpacl); | |
526 | return error; | |
527 | } | |
528 | ||
529 | #endif /* defined(CONFIG_NFS_V4) */ | |
530 | ||
531 | #ifdef CONFIG_NFSD_V3 | |
532 | /* | |
533 | * Check server access rights to a file system object | |
534 | */ | |
535 | struct accessmap { | |
536 | u32 access; | |
537 | int how; | |
538 | }; | |
539 | static struct accessmap nfs3_regaccess[] = { | |
540 | { NFS3_ACCESS_READ, MAY_READ }, | |
541 | { NFS3_ACCESS_EXECUTE, MAY_EXEC }, | |
542 | { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC }, | |
543 | { NFS3_ACCESS_EXTEND, MAY_WRITE }, | |
544 | ||
545 | { 0, 0 } | |
546 | }; | |
547 | ||
548 | static struct accessmap nfs3_diraccess[] = { | |
549 | { NFS3_ACCESS_READ, MAY_READ }, | |
550 | { NFS3_ACCESS_LOOKUP, MAY_EXEC }, | |
551 | { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC }, | |
552 | { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE }, | |
553 | { NFS3_ACCESS_DELETE, MAY_REMOVE }, | |
554 | ||
555 | { 0, 0 } | |
556 | }; | |
557 | ||
558 | static struct accessmap nfs3_anyaccess[] = { | |
559 | /* Some clients - Solaris 2.6 at least, make an access call | |
560 | * to the server to check for access for things like /dev/null | |
561 | * (which really, the server doesn't care about). So | |
562 | * We provide simple access checking for them, looking | |
563 | * mainly at mode bits, and we make sure to ignore read-only | |
564 | * filesystem checks | |
565 | */ | |
566 | { NFS3_ACCESS_READ, MAY_READ }, | |
567 | { NFS3_ACCESS_EXECUTE, MAY_EXEC }, | |
568 | { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS }, | |
569 | { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS }, | |
570 | ||
571 | { 0, 0 } | |
572 | }; | |
573 | ||
574 | int | |
575 | nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) | |
576 | { | |
577 | struct accessmap *map; | |
578 | struct svc_export *export; | |
579 | struct dentry *dentry; | |
580 | u32 query, result = 0, sresult = 0; | |
581 | unsigned int error; | |
582 | ||
583 | error = fh_verify(rqstp, fhp, 0, MAY_NOP); | |
584 | if (error) | |
585 | goto out; | |
586 | ||
587 | export = fhp->fh_export; | |
588 | dentry = fhp->fh_dentry; | |
589 | ||
590 | if (S_ISREG(dentry->d_inode->i_mode)) | |
591 | map = nfs3_regaccess; | |
592 | else if (S_ISDIR(dentry->d_inode->i_mode)) | |
593 | map = nfs3_diraccess; | |
594 | else | |
595 | map = nfs3_anyaccess; | |
596 | ||
597 | ||
598 | query = *access; | |
599 | for (; map->access; map++) { | |
600 | if (map->access & query) { | |
601 | unsigned int err2; | |
602 | ||
603 | sresult |= map->access; | |
604 | ||
605 | err2 = nfsd_permission(export, dentry, map->how); | |
606 | switch (err2) { | |
607 | case nfs_ok: | |
608 | result |= map->access; | |
609 | break; | |
610 | ||
611 | /* the following error codes just mean the access was not allowed, | |
612 | * rather than an error occurred */ | |
613 | case nfserr_rofs: | |
614 | case nfserr_acces: | |
615 | case nfserr_perm: | |
616 | /* simply don't "or" in the access bit. */ | |
617 | break; | |
618 | default: | |
619 | error = err2; | |
620 | goto out; | |
621 | } | |
622 | } | |
623 | } | |
624 | *access = result; | |
625 | if (supported) | |
626 | *supported = sresult; | |
627 | ||
628 | out: | |
629 | return error; | |
630 | } | |
631 | #endif /* CONFIG_NFSD_V3 */ | |
632 | ||
633 | ||
634 | ||
635 | /* | |
636 | * Open an existing file or directory. | |
637 | * The access argument indicates the type of open (read/write/lock) | |
638 | * N.B. After this call fhp needs an fh_put | |
639 | */ | |
640 | int | |
641 | nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, | |
642 | int access, struct file **filp) | |
643 | { | |
644 | struct dentry *dentry; | |
645 | struct inode *inode; | |
646 | int flags = O_RDONLY|O_LARGEFILE, err; | |
647 | ||
648 | /* | |
649 | * If we get here, then the client has already done an "open", | |
650 | * and (hopefully) checked permission - so allow OWNER_OVERRIDE | |
651 | * in case a chmod has now revoked permission. | |
652 | */ | |
653 | err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE); | |
654 | if (err) | |
655 | goto out; | |
656 | ||
657 | dentry = fhp->fh_dentry; | |
658 | inode = dentry->d_inode; | |
659 | ||
660 | /* Disallow write access to files with the append-only bit set | |
661 | * or any access when mandatory locking enabled | |
662 | */ | |
663 | err = nfserr_perm; | |
664 | if (IS_APPEND(inode) && (access & MAY_WRITE)) | |
665 | goto out; | |
666 | if (IS_ISMNDLK(inode)) | |
667 | goto out; | |
668 | ||
669 | if (!inode->i_fop) | |
670 | goto out; | |
671 | ||
672 | /* | |
673 | * Check to see if there are any leases on this file. | |
674 | * This may block while leases are broken. | |
675 | */ | |
676 | err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0)); | |
677 | if (err == -EWOULDBLOCK) | |
678 | err = -ETIMEDOUT; | |
679 | if (err) /* NOMEM or WOULDBLOCK */ | |
680 | goto out_nfserr; | |
681 | ||
682 | if (access & MAY_WRITE) { | |
9ecb6a08 BF |
683 | if (access & MAY_READ) |
684 | flags = O_RDWR|O_LARGEFILE; | |
685 | else | |
686 | flags = O_WRONLY|O_LARGEFILE; | |
1da177e4 LT |
687 | |
688 | DQUOT_INIT(inode); | |
689 | } | |
690 | *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags); | |
691 | if (IS_ERR(*filp)) | |
692 | err = PTR_ERR(*filp); | |
693 | out_nfserr: | |
694 | if (err) | |
695 | err = nfserrno(err); | |
696 | out: | |
697 | return err; | |
698 | } | |
699 | ||
700 | /* | |
701 | * Close a file. | |
702 | */ | |
703 | void | |
704 | nfsd_close(struct file *filp) | |
705 | { | |
706 | fput(filp); | |
707 | } | |
708 | ||
709 | /* | |
710 | * Sync a file | |
711 | * As this calls fsync (not fdatasync) there is no need for a write_inode | |
712 | * after it. | |
713 | */ | |
a334de28 | 714 | static inline int nfsd_dosync(struct file *filp, struct dentry *dp, |
99ac48f5 | 715 | const struct file_operations *fop) |
1da177e4 LT |
716 | { |
717 | struct inode *inode = dp->d_inode; | |
718 | int (*fsync) (struct file *, struct dentry *, int); | |
f193fbab | 719 | int err; |
1da177e4 | 720 | |
f193fbab YT |
721 | err = filemap_fdatawrite(inode->i_mapping); |
722 | if (err == 0 && fop && (fsync = fop->fsync)) | |
723 | err = fsync(filp, dp, 0); | |
724 | if (err == 0) | |
725 | err = filemap_fdatawait(inode->i_mapping); | |
a334de28 | 726 | |
f193fbab | 727 | return err; |
1da177e4 LT |
728 | } |
729 | ||
730 | ||
a334de28 | 731 | static int |
1da177e4 LT |
732 | nfsd_sync(struct file *filp) |
733 | { | |
a334de28 | 734 | int err; |
1da177e4 LT |
735 | struct inode *inode = filp->f_dentry->d_inode; |
736 | dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name); | |
1b1dcc1b | 737 | mutex_lock(&inode->i_mutex); |
a334de28 | 738 | err=nfsd_dosync(filp, filp->f_dentry, filp->f_op); |
1b1dcc1b | 739 | mutex_unlock(&inode->i_mutex); |
a334de28 DS |
740 | |
741 | return err; | |
1da177e4 LT |
742 | } |
743 | ||
f193fbab | 744 | int |
1da177e4 LT |
745 | nfsd_sync_dir(struct dentry *dp) |
746 | { | |
f193fbab | 747 | return nfsd_dosync(NULL, dp, dp->d_inode->i_fop); |
1da177e4 LT |
748 | } |
749 | ||
750 | /* | |
751 | * Obtain the readahead parameters for the file | |
752 | * specified by (dev, ino). | |
753 | */ | |
1da177e4 LT |
754 | |
755 | static inline struct raparms * | |
756 | nfsd_get_raparms(dev_t dev, ino_t ino) | |
757 | { | |
758 | struct raparms *ra, **rap, **frap = NULL; | |
759 | int depth = 0; | |
fce1456a GB |
760 | unsigned int hash; |
761 | struct raparm_hbucket *rab; | |
762 | ||
763 | hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; | |
764 | rab = &raparm_hash[hash]; | |
1da177e4 | 765 | |
fce1456a GB |
766 | spin_lock(&rab->pb_lock); |
767 | for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { | |
1da177e4 LT |
768 | if (ra->p_ino == ino && ra->p_dev == dev) |
769 | goto found; | |
770 | depth++; | |
771 | if (ra->p_count == 0) | |
772 | frap = rap; | |
773 | } | |
774 | depth = nfsdstats.ra_size*11/10; | |
775 | if (!frap) { | |
fce1456a | 776 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
777 | return NULL; |
778 | } | |
779 | rap = frap; | |
780 | ra = *frap; | |
781 | ra->p_dev = dev; | |
782 | ra->p_ino = ino; | |
783 | ra->p_set = 0; | |
fce1456a | 784 | ra->p_hindex = hash; |
1da177e4 | 785 | found: |
fce1456a | 786 | if (rap != &rab->pb_head) { |
1da177e4 | 787 | *rap = ra->p_next; |
fce1456a GB |
788 | ra->p_next = rab->pb_head; |
789 | rab->pb_head = ra; | |
1da177e4 LT |
790 | } |
791 | ra->p_count++; | |
792 | nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; | |
fce1456a | 793 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
794 | return ra; |
795 | } | |
796 | ||
797 | /* | |
798 | * Grab and keep cached pages assosiated with a file in the svc_rqst | |
799 | * so that they can be passed to the netowork sendmsg/sendpage routines | |
800 | * directrly. They will be released after the sending has completed. | |
801 | */ | |
802 | static int | |
803 | nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size) | |
804 | { | |
805 | unsigned long count = desc->count; | |
806 | struct svc_rqst *rqstp = desc->arg.data; | |
44524359 | 807 | struct page **pp = rqstp->rq_respages + rqstp->rq_resused; |
1da177e4 LT |
808 | |
809 | if (size > count) | |
810 | size = count; | |
811 | ||
812 | if (rqstp->rq_res.page_len == 0) { | |
813 | get_page(page); | |
44524359 N |
814 | put_page(*pp); |
815 | *pp = page; | |
816 | rqstp->rq_resused++; | |
1da177e4 LT |
817 | rqstp->rq_res.page_base = offset; |
818 | rqstp->rq_res.page_len = size; | |
44524359 | 819 | } else if (page != pp[-1]) { |
1da177e4 | 820 | get_page(page); |
44524359 N |
821 | put_page(*pp); |
822 | *pp = page; | |
823 | rqstp->rq_resused++; | |
1da177e4 | 824 | rqstp->rq_res.page_len += size; |
44524359 | 825 | } else |
1da177e4 | 826 | rqstp->rq_res.page_len += size; |
1da177e4 LT |
827 | |
828 | desc->count = count - size; | |
829 | desc->written += size; | |
830 | return size; | |
831 | } | |
832 | ||
7e8f0593 | 833 | static int |
1da177e4 LT |
834 | nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
835 | loff_t offset, struct kvec *vec, int vlen, unsigned long *count) | |
836 | { | |
837 | struct inode *inode; | |
838 | struct raparms *ra; | |
839 | mm_segment_t oldfs; | |
840 | int err; | |
841 | ||
842 | err = nfserr_perm; | |
843 | inode = file->f_dentry->d_inode; | |
844 | #ifdef MSNFS | |
845 | if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && | |
846 | (!lock_may_read(inode, offset, *count))) | |
847 | goto out; | |
848 | #endif | |
849 | ||
850 | /* Get readahead parameters */ | |
851 | ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); | |
852 | ||
853 | if (ra && ra->p_set) | |
854 | file->f_ra = ra->p_ra; | |
855 | ||
5c04c46a | 856 | if (file->f_op->sendfile && rqstp->rq_sendfile_ok) { |
44524359 | 857 | rqstp->rq_resused = 1; |
1da177e4 LT |
858 | err = file->f_op->sendfile(file, &offset, *count, |
859 | nfsd_read_actor, rqstp); | |
860 | } else { | |
861 | oldfs = get_fs(); | |
862 | set_fs(KERNEL_DS); | |
863 | err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); | |
864 | set_fs(oldfs); | |
865 | } | |
866 | ||
867 | /* Write back readahead params */ | |
868 | if (ra) { | |
fce1456a GB |
869 | struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; |
870 | spin_lock(&rab->pb_lock); | |
1da177e4 LT |
871 | ra->p_ra = file->f_ra; |
872 | ra->p_set = 1; | |
873 | ra->p_count--; | |
fce1456a | 874 | spin_unlock(&rab->pb_lock); |
1da177e4 LT |
875 | } |
876 | ||
877 | if (err >= 0) { | |
878 | nfsdstats.io_read += err; | |
879 | *count = err; | |
880 | err = 0; | |
0eeca283 | 881 | fsnotify_access(file->f_dentry); |
1da177e4 LT |
882 | } else |
883 | err = nfserrno(err); | |
884 | out: | |
885 | return err; | |
886 | } | |
887 | ||
9f708e40 NB |
888 | static void kill_suid(struct dentry *dentry) |
889 | { | |
890 | struct iattr ia; | |
891 | ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID; | |
892 | ||
1b1dcc1b | 893 | mutex_lock(&dentry->d_inode->i_mutex); |
9f708e40 | 894 | notify_change(dentry, &ia); |
1b1dcc1b | 895 | mutex_unlock(&dentry->d_inode->i_mutex); |
9f708e40 NB |
896 | } |
897 | ||
7e8f0593 | 898 | static int |
1da177e4 LT |
899 | nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, |
900 | loff_t offset, struct kvec *vec, int vlen, | |
901 | unsigned long cnt, int *stablep) | |
902 | { | |
903 | struct svc_export *exp; | |
904 | struct dentry *dentry; | |
905 | struct inode *inode; | |
906 | mm_segment_t oldfs; | |
907 | int err = 0; | |
908 | int stable = *stablep; | |
909 | ||
45bd3b3d | 910 | #ifdef MSNFS |
1da177e4 LT |
911 | err = nfserr_perm; |
912 | ||
1da177e4 LT |
913 | if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && |
914 | (!lock_may_write(file->f_dentry->d_inode, offset, cnt))) | |
915 | goto out; | |
916 | #endif | |
917 | ||
918 | dentry = file->f_dentry; | |
919 | inode = dentry->d_inode; | |
920 | exp = fhp->fh_export; | |
921 | ||
922 | /* | |
923 | * Request sync writes if | |
924 | * - the sync export option has been set, or | |
925 | * - the client requested O_SYNC behavior (NFSv3 feature). | |
926 | * - The file system doesn't support fsync(). | |
927 | * When gathered writes have been configured for this volume, | |
928 | * flushing the data to disk is handled separately below. | |
929 | */ | |
930 | ||
931 | if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */ | |
932 | stable = 2; | |
933 | *stablep = 2; /* FILE_SYNC */ | |
934 | } | |
935 | ||
936 | if (!EX_ISSYNC(exp)) | |
937 | stable = 0; | |
938 | if (stable && !EX_WGATHER(exp)) | |
939 | file->f_flags |= O_SYNC; | |
940 | ||
941 | /* Write the data. */ | |
942 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
943 | err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); | |
944 | set_fs(oldfs); | |
945 | if (err >= 0) { | |
946 | nfsdstats.io_write += cnt; | |
0eeca283 | 947 | fsnotify_modify(file->f_dentry); |
1da177e4 LT |
948 | } |
949 | ||
950 | /* clear setuid/setgid flag after write */ | |
9f708e40 NB |
951 | if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) |
952 | kill_suid(dentry); | |
1da177e4 LT |
953 | |
954 | if (err >= 0 && stable) { | |
955 | static ino_t last_ino; | |
956 | static dev_t last_dev; | |
957 | ||
958 | /* | |
959 | * Gathered writes: If another process is currently | |
960 | * writing to the file, there's a high chance | |
961 | * this is another nfsd (triggered by a bulk write | |
962 | * from a client's biod). Rather than syncing the | |
963 | * file with each write request, we sleep for 10 msec. | |
964 | * | |
965 | * I don't know if this roughly approximates | |
966 | * C. Juszak's idea of gathered writes, but it's a | |
967 | * nice and simple solution (IMHO), and it seems to | |
968 | * work:-) | |
969 | */ | |
970 | if (EX_WGATHER(exp)) { | |
971 | if (atomic_read(&inode->i_writecount) > 1 | |
972 | || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { | |
973 | dprintk("nfsd: write defer %d\n", current->pid); | |
974 | msleep(10); | |
975 | dprintk("nfsd: write resume %d\n", current->pid); | |
976 | } | |
977 | ||
978 | if (inode->i_state & I_DIRTY) { | |
979 | dprintk("nfsd: write sync %d\n", current->pid); | |
a334de28 | 980 | err=nfsd_sync(file); |
1da177e4 LT |
981 | } |
982 | #if 0 | |
983 | wake_up(&inode->i_wait); | |
984 | #endif | |
985 | } | |
986 | last_ino = inode->i_ino; | |
987 | last_dev = inode->i_sb->s_dev; | |
988 | } | |
989 | ||
990 | dprintk("nfsd: write complete err=%d\n", err); | |
991 | if (err >= 0) | |
992 | err = 0; | |
993 | else | |
994 | err = nfserrno(err); | |
995 | out: | |
996 | return err; | |
997 | } | |
998 | ||
999 | /* | |
1000 | * Read data from a file. count must contain the requested read count | |
1001 | * on entry. On return, *count contains the number of bytes actually read. | |
1002 | * N.B. After this call fhp needs an fh_put | |
1003 | */ | |
1004 | int | |
1005 | nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, | |
1006 | loff_t offset, struct kvec *vec, int vlen, | |
1007 | unsigned long *count) | |
1008 | { | |
1009 | int err; | |
1010 | ||
1011 | if (file) { | |
1012 | err = nfsd_permission(fhp->fh_export, fhp->fh_dentry, | |
1013 | MAY_READ|MAY_OWNER_OVERRIDE); | |
1014 | if (err) | |
1015 | goto out; | |
1016 | err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); | |
1017 | } else { | |
1018 | err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file); | |
1019 | if (err) | |
1020 | goto out; | |
1021 | err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); | |
1022 | nfsd_close(file); | |
1023 | } | |
1024 | out: | |
1025 | return err; | |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * Write data to a file. | |
1030 | * The stable flag requests synchronous writes. | |
1031 | * N.B. After this call fhp needs an fh_put | |
1032 | */ | |
1033 | int | |
1034 | nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, | |
1035 | loff_t offset, struct kvec *vec, int vlen, unsigned long cnt, | |
1036 | int *stablep) | |
1037 | { | |
1038 | int err = 0; | |
1039 | ||
1040 | if (file) { | |
1041 | err = nfsd_permission(fhp->fh_export, fhp->fh_dentry, | |
1042 | MAY_WRITE|MAY_OWNER_OVERRIDE); | |
1043 | if (err) | |
1044 | goto out; | |
1045 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, | |
1046 | stablep); | |
1047 | } else { | |
1048 | err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file); | |
1049 | if (err) | |
1050 | goto out; | |
1051 | ||
1052 | if (cnt) | |
1053 | err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, | |
1054 | cnt, stablep); | |
1055 | nfsd_close(file); | |
1056 | } | |
1057 | out: | |
1058 | return err; | |
1059 | } | |
1060 | ||
1061 | #ifdef CONFIG_NFSD_V3 | |
1062 | /* | |
1063 | * Commit all pending writes to stable storage. | |
1064 | * Strictly speaking, we could sync just the indicated file region here, | |
1065 | * but there's currently no way we can ask the VFS to do so. | |
1066 | * | |
1067 | * Unfortunately we cannot lock the file to make sure we return full WCC | |
1068 | * data to the client, as locking happens lower down in the filesystem. | |
1069 | */ | |
1070 | int | |
1071 | nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1072 | loff_t offset, unsigned long count) | |
1073 | { | |
1074 | struct file *file; | |
1075 | int err; | |
1076 | ||
1077 | if ((u64)count > ~(u64)offset) | |
1078 | return nfserr_inval; | |
1079 | ||
1080 | if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0) | |
1081 | return err; | |
1082 | if (EX_ISSYNC(fhp->fh_export)) { | |
1083 | if (file->f_op && file->f_op->fsync) { | |
45bd3b3d | 1084 | err = nfserrno(nfsd_sync(file)); |
1da177e4 LT |
1085 | } else { |
1086 | err = nfserr_notsupp; | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | nfsd_close(file); | |
1091 | return err; | |
1092 | } | |
1093 | #endif /* CONFIG_NFSD_V3 */ | |
1094 | ||
1095 | /* | |
1096 | * Create a file (regular, directory, device, fifo); UNIX sockets | |
1097 | * not yet implemented. | |
1098 | * If the response fh has been verified, the parent directory should | |
1099 | * already be locked. Note that the parent directory is left locked. | |
1100 | * | |
1101 | * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp | |
1102 | */ | |
1103 | int | |
1104 | nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1105 | char *fname, int flen, struct iattr *iap, | |
1106 | int type, dev_t rdev, struct svc_fh *resfhp) | |
1107 | { | |
1108 | struct dentry *dentry, *dchild = NULL; | |
1109 | struct inode *dirp; | |
1110 | int err; | |
1111 | ||
1112 | err = nfserr_perm; | |
1113 | if (!flen) | |
1114 | goto out; | |
1115 | err = nfserr_exist; | |
1116 | if (isdotent(fname, flen)) | |
1117 | goto out; | |
1118 | ||
1119 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); | |
1120 | if (err) | |
1121 | goto out; | |
1122 | ||
1123 | dentry = fhp->fh_dentry; | |
1124 | dirp = dentry->d_inode; | |
1125 | ||
1126 | err = nfserr_notdir; | |
1127 | if(!dirp->i_op || !dirp->i_op->lookup) | |
1128 | goto out; | |
1129 | /* | |
1130 | * Check whether the response file handle has been verified yet. | |
1131 | * If it has, the parent directory should already be locked. | |
1132 | */ | |
1133 | if (!resfhp->fh_dentry) { | |
1134 | /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ | |
12fd3520 | 1135 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1136 | dchild = lookup_one_len(fname, dentry, flen); |
1137 | err = PTR_ERR(dchild); | |
1138 | if (IS_ERR(dchild)) | |
1139 | goto out_nfserr; | |
1140 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); | |
1141 | if (err) | |
1142 | goto out; | |
1143 | } else { | |
1144 | /* called from nfsd_proc_create */ | |
1145 | dchild = dget(resfhp->fh_dentry); | |
1146 | if (!fhp->fh_locked) { | |
1147 | /* not actually possible */ | |
1148 | printk(KERN_ERR | |
1149 | "nfsd_create: parent %s/%s not locked!\n", | |
1150 | dentry->d_parent->d_name.name, | |
1151 | dentry->d_name.name); | |
d75f2b9f | 1152 | err = nfserr_io; |
1da177e4 LT |
1153 | goto out; |
1154 | } | |
1155 | } | |
1156 | /* | |
1157 | * Make sure the child dentry is still negative ... | |
1158 | */ | |
1159 | err = nfserr_exist; | |
1160 | if (dchild->d_inode) { | |
1161 | dprintk("nfsd_create: dentry %s/%s not negative!\n", | |
1162 | dentry->d_name.name, dchild->d_name.name); | |
1163 | goto out; | |
1164 | } | |
1165 | ||
1166 | if (!(iap->ia_valid & ATTR_MODE)) | |
1167 | iap->ia_mode = 0; | |
1168 | iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; | |
1169 | ||
1170 | /* | |
1171 | * Get the dir op function pointer. | |
1172 | */ | |
1173 | err = nfserr_perm; | |
1174 | switch (type) { | |
1175 | case S_IFREG: | |
1176 | err = vfs_create(dirp, dchild, iap->ia_mode, NULL); | |
1177 | break; | |
1178 | case S_IFDIR: | |
1179 | err = vfs_mkdir(dirp, dchild, iap->ia_mode); | |
1180 | break; | |
1181 | case S_IFCHR: | |
1182 | case S_IFBLK: | |
1183 | case S_IFIFO: | |
1184 | case S_IFSOCK: | |
1185 | err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); | |
1186 | break; | |
1187 | default: | |
1188 | printk("nfsd: bad file type %o in nfsd_create\n", type); | |
1189 | err = -EINVAL; | |
1190 | } | |
1191 | if (err < 0) | |
1192 | goto out_nfserr; | |
1193 | ||
1194 | if (EX_ISSYNC(fhp->fh_export)) { | |
45bd3b3d | 1195 | err = nfserrno(nfsd_sync_dir(dentry)); |
1da177e4 LT |
1196 | write_inode_now(dchild->d_inode, 1); |
1197 | } | |
1198 | ||
1199 | ||
1200 | /* Set file attributes. Mode has already been set and | |
1201 | * setting uid/gid works only for root. Irix appears to | |
1202 | * send along the gid when it tries to implement setgid | |
1203 | * directories via NFS. | |
1204 | */ | |
f193fbab YT |
1205 | if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) { |
1206 | int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); | |
1207 | if (err2) | |
1208 | err = err2; | |
1209 | } | |
1da177e4 LT |
1210 | /* |
1211 | * Update the file handle to get the new inode info. | |
1212 | */ | |
1213 | if (!err) | |
1214 | err = fh_update(resfhp); | |
1215 | out: | |
1216 | if (dchild && !IS_ERR(dchild)) | |
1217 | dput(dchild); | |
1218 | return err; | |
1219 | ||
1220 | out_nfserr: | |
1221 | err = nfserrno(err); | |
1222 | goto out; | |
1223 | } | |
1224 | ||
1225 | #ifdef CONFIG_NFSD_V3 | |
1226 | /* | |
1227 | * NFSv3 version of nfsd_create | |
1228 | */ | |
1229 | int | |
1230 | nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1231 | char *fname, int flen, struct iattr *iap, | |
1232 | struct svc_fh *resfhp, int createmode, u32 *verifier, | |
1233 | int *truncp) | |
1234 | { | |
1235 | struct dentry *dentry, *dchild = NULL; | |
1236 | struct inode *dirp; | |
1237 | int err; | |
1238 | __u32 v_mtime=0, v_atime=0; | |
1239 | int v_mode=0; | |
1240 | ||
1241 | err = nfserr_perm; | |
1242 | if (!flen) | |
1243 | goto out; | |
1244 | err = nfserr_exist; | |
1245 | if (isdotent(fname, flen)) | |
1246 | goto out; | |
1247 | if (!(iap->ia_valid & ATTR_MODE)) | |
1248 | iap->ia_mode = 0; | |
1249 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); | |
1250 | if (err) | |
1251 | goto out; | |
1252 | ||
1253 | dentry = fhp->fh_dentry; | |
1254 | dirp = dentry->d_inode; | |
1255 | ||
1256 | /* Get all the sanity checks out of the way before | |
1257 | * we lock the parent. */ | |
1258 | err = nfserr_notdir; | |
1259 | if(!dirp->i_op || !dirp->i_op->lookup) | |
1260 | goto out; | |
12fd3520 | 1261 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1262 | |
1263 | /* | |
1264 | * Compose the response file handle. | |
1265 | */ | |
1266 | dchild = lookup_one_len(fname, dentry, flen); | |
1267 | err = PTR_ERR(dchild); | |
1268 | if (IS_ERR(dchild)) | |
1269 | goto out_nfserr; | |
1270 | ||
1271 | err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); | |
1272 | if (err) | |
1273 | goto out; | |
1274 | ||
1275 | if (createmode == NFS3_CREATE_EXCLUSIVE) { | |
1276 | /* while the verifier would fit in mtime+atime, | |
1277 | * solaris7 gets confused (bugid 4218508) if these have | |
1278 | * the high bit set, so we use the mode as well | |
1279 | */ | |
1280 | v_mtime = verifier[0]&0x7fffffff; | |
1281 | v_atime = verifier[1]&0x7fffffff; | |
1282 | v_mode = S_IFREG | |
1283 | | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */ | |
1284 | | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */ | |
1285 | ; | |
1286 | } | |
1287 | ||
1288 | if (dchild->d_inode) { | |
1289 | err = 0; | |
1290 | ||
1291 | switch (createmode) { | |
1292 | case NFS3_CREATE_UNCHECKED: | |
1293 | if (! S_ISREG(dchild->d_inode->i_mode)) | |
1294 | err = nfserr_exist; | |
1295 | else if (truncp) { | |
1296 | /* in nfsv4, we need to treat this case a little | |
1297 | * differently. we don't want to truncate the | |
1298 | * file now; this would be wrong if the OPEN | |
1299 | * fails for some other reason. furthermore, | |
1300 | * if the size is nonzero, we should ignore it | |
1301 | * according to spec! | |
1302 | */ | |
1303 | *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; | |
1304 | } | |
1305 | else { | |
1306 | iap->ia_valid &= ATTR_SIZE; | |
1307 | goto set_attr; | |
1308 | } | |
1309 | break; | |
1310 | case NFS3_CREATE_EXCLUSIVE: | |
1311 | if ( dchild->d_inode->i_mtime.tv_sec == v_mtime | |
1312 | && dchild->d_inode->i_atime.tv_sec == v_atime | |
1313 | && dchild->d_inode->i_mode == v_mode | |
1314 | && dchild->d_inode->i_size == 0 ) | |
1315 | break; | |
1316 | /* fallthru */ | |
1317 | case NFS3_CREATE_GUARDED: | |
1318 | err = nfserr_exist; | |
1319 | } | |
1320 | goto out; | |
1321 | } | |
1322 | ||
1323 | err = vfs_create(dirp, dchild, iap->ia_mode, NULL); | |
1324 | if (err < 0) | |
1325 | goto out_nfserr; | |
1326 | ||
1327 | if (EX_ISSYNC(fhp->fh_export)) { | |
45bd3b3d | 1328 | err = nfserrno(nfsd_sync_dir(dentry)); |
1da177e4 LT |
1329 | /* setattr will sync the child (or not) */ |
1330 | } | |
1331 | ||
1da177e4 LT |
1332 | if (createmode == NFS3_CREATE_EXCLUSIVE) { |
1333 | /* Cram the verifier into atime/mtime/mode */ | |
1334 | iap->ia_valid = ATTR_MTIME|ATTR_ATIME | |
1335 | | ATTR_MTIME_SET|ATTR_ATIME_SET | |
1336 | | ATTR_MODE; | |
1337 | /* XXX someone who knows this better please fix it for nsec */ | |
1338 | iap->ia_mtime.tv_sec = v_mtime; | |
1339 | iap->ia_atime.tv_sec = v_atime; | |
1340 | iap->ia_mtime.tv_nsec = 0; | |
1341 | iap->ia_atime.tv_nsec = 0; | |
1342 | iap->ia_mode = v_mode; | |
1343 | } | |
1344 | ||
1345 | /* Set file attributes. | |
1346 | * Mode has already been set but we might need to reset it | |
1347 | * for CREATE_EXCLUSIVE | |
1348 | * Irix appears to send along the gid when it tries to | |
1349 | * implement setgid directories via NFS. Clear out all that cruft. | |
1350 | */ | |
1351 | set_attr: | |
f193fbab YT |
1352 | if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) { |
1353 | int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); | |
1354 | if (err2) | |
45bd3b3d | 1355 | err = err2; |
f193fbab YT |
1356 | } |
1357 | ||
1358 | /* | |
1359 | * Update the filehandle to get the new inode info. | |
1360 | */ | |
1361 | if (!err) | |
1362 | err = fh_update(resfhp); | |
1da177e4 LT |
1363 | |
1364 | out: | |
1365 | fh_unlock(fhp); | |
1366 | if (dchild && !IS_ERR(dchild)) | |
1367 | dput(dchild); | |
1368 | return err; | |
1369 | ||
1370 | out_nfserr: | |
1371 | err = nfserrno(err); | |
1372 | goto out; | |
1373 | } | |
1374 | #endif /* CONFIG_NFSD_V3 */ | |
1375 | ||
1376 | /* | |
1377 | * Read a symlink. On entry, *lenp must contain the maximum path length that | |
1378 | * fits into the buffer. On return, it contains the true length. | |
1379 | * N.B. After this call fhp needs an fh_put | |
1380 | */ | |
1381 | int | |
1382 | nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) | |
1383 | { | |
1384 | struct dentry *dentry; | |
1385 | struct inode *inode; | |
1386 | mm_segment_t oldfs; | |
1387 | int err; | |
1388 | ||
1389 | err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP); | |
1390 | if (err) | |
1391 | goto out; | |
1392 | ||
1393 | dentry = fhp->fh_dentry; | |
1394 | inode = dentry->d_inode; | |
1395 | ||
1396 | err = nfserr_inval; | |
1397 | if (!inode->i_op || !inode->i_op->readlink) | |
1398 | goto out; | |
1399 | ||
1400 | touch_atime(fhp->fh_export->ex_mnt, dentry); | |
1401 | /* N.B. Why does this call need a get_fs()?? | |
1402 | * Remove the set_fs and watch the fireworks:-) --okir | |
1403 | */ | |
1404 | ||
1405 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
1406 | err = inode->i_op->readlink(dentry, buf, *lenp); | |
1407 | set_fs(oldfs); | |
1408 | ||
1409 | if (err < 0) | |
1410 | goto out_nfserr; | |
1411 | *lenp = err; | |
1412 | err = 0; | |
1413 | out: | |
1414 | return err; | |
1415 | ||
1416 | out_nfserr: | |
1417 | err = nfserrno(err); | |
1418 | goto out; | |
1419 | } | |
1420 | ||
1421 | /* | |
1422 | * Create a symlink and look up its inode | |
1423 | * N.B. After this call _both_ fhp and resfhp need an fh_put | |
1424 | */ | |
1425 | int | |
1426 | nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1427 | char *fname, int flen, | |
1428 | char *path, int plen, | |
1429 | struct svc_fh *resfhp, | |
1430 | struct iattr *iap) | |
1431 | { | |
1432 | struct dentry *dentry, *dnew; | |
1433 | int err, cerr; | |
1434 | umode_t mode; | |
1435 | ||
1436 | err = nfserr_noent; | |
1437 | if (!flen || !plen) | |
1438 | goto out; | |
1439 | err = nfserr_exist; | |
1440 | if (isdotent(fname, flen)) | |
1441 | goto out; | |
1442 | ||
1443 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); | |
1444 | if (err) | |
1445 | goto out; | |
1446 | fh_lock(fhp); | |
1447 | dentry = fhp->fh_dentry; | |
1448 | dnew = lookup_one_len(fname, dentry, flen); | |
1449 | err = PTR_ERR(dnew); | |
1450 | if (IS_ERR(dnew)) | |
1451 | goto out_nfserr; | |
1452 | ||
1453 | mode = S_IALLUGO; | |
1454 | /* Only the MODE ATTRibute is even vaguely meaningful */ | |
1455 | if (iap && (iap->ia_valid & ATTR_MODE)) | |
1456 | mode = iap->ia_mode & S_IALLUGO; | |
1457 | ||
1458 | if (unlikely(path[plen] != 0)) { | |
1459 | char *path_alloced = kmalloc(plen+1, GFP_KERNEL); | |
1460 | if (path_alloced == NULL) | |
1461 | err = -ENOMEM; | |
1462 | else { | |
1463 | strncpy(path_alloced, path, plen); | |
1464 | path_alloced[plen] = 0; | |
1465 | err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode); | |
1466 | kfree(path_alloced); | |
1467 | } | |
1468 | } else | |
1469 | err = vfs_symlink(dentry->d_inode, dnew, path, mode); | |
1470 | ||
f193fbab | 1471 | if (!err) |
1da177e4 | 1472 | if (EX_ISSYNC(fhp->fh_export)) |
f193fbab YT |
1473 | err = nfsd_sync_dir(dentry); |
1474 | if (err) | |
1da177e4 LT |
1475 | err = nfserrno(err); |
1476 | fh_unlock(fhp); | |
1477 | ||
1478 | cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); | |
1479 | dput(dnew); | |
1480 | if (err==0) err = cerr; | |
1481 | out: | |
1482 | return err; | |
1483 | ||
1484 | out_nfserr: | |
1485 | err = nfserrno(err); | |
1486 | goto out; | |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * Create a hardlink | |
1491 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1492 | */ | |
1493 | int | |
1494 | nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, | |
1495 | char *name, int len, struct svc_fh *tfhp) | |
1496 | { | |
1497 | struct dentry *ddir, *dnew, *dold; | |
1498 | struct inode *dirp, *dest; | |
1499 | int err; | |
1500 | ||
1501 | err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE); | |
1502 | if (err) | |
1503 | goto out; | |
1504 | err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP); | |
1505 | if (err) | |
1506 | goto out; | |
1507 | ||
1508 | err = nfserr_perm; | |
1509 | if (!len) | |
1510 | goto out; | |
1511 | err = nfserr_exist; | |
1512 | if (isdotent(name, len)) | |
1513 | goto out; | |
1514 | ||
12fd3520 | 1515 | fh_lock_nested(ffhp, I_MUTEX_PARENT); |
1da177e4 LT |
1516 | ddir = ffhp->fh_dentry; |
1517 | dirp = ddir->d_inode; | |
1518 | ||
1519 | dnew = lookup_one_len(name, ddir, len); | |
1520 | err = PTR_ERR(dnew); | |
1521 | if (IS_ERR(dnew)) | |
1522 | goto out_nfserr; | |
1523 | ||
1524 | dold = tfhp->fh_dentry; | |
1525 | dest = dold->d_inode; | |
1526 | ||
1527 | err = vfs_link(dold, dirp, dnew); | |
1528 | if (!err) { | |
1529 | if (EX_ISSYNC(ffhp->fh_export)) { | |
45bd3b3d | 1530 | err = nfserrno(nfsd_sync_dir(ddir)); |
1da177e4 LT |
1531 | write_inode_now(dest, 1); |
1532 | } | |
1533 | } else { | |
1534 | if (err == -EXDEV && rqstp->rq_vers == 2) | |
1535 | err = nfserr_acces; | |
1536 | else | |
1537 | err = nfserrno(err); | |
1538 | } | |
1539 | ||
1da177e4 | 1540 | dput(dnew); |
270d56e5 DR |
1541 | out_unlock: |
1542 | fh_unlock(ffhp); | |
1da177e4 LT |
1543 | out: |
1544 | return err; | |
1545 | ||
1546 | out_nfserr: | |
1547 | err = nfserrno(err); | |
270d56e5 | 1548 | goto out_unlock; |
1da177e4 LT |
1549 | } |
1550 | ||
1551 | /* | |
1552 | * Rename a file | |
1553 | * N.B. After this call _both_ ffhp and tfhp need an fh_put | |
1554 | */ | |
1555 | int | |
1556 | nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, | |
1557 | struct svc_fh *tfhp, char *tname, int tlen) | |
1558 | { | |
1559 | struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; | |
1560 | struct inode *fdir, *tdir; | |
1561 | int err; | |
1562 | ||
1563 | err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE); | |
1564 | if (err) | |
1565 | goto out; | |
1566 | err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE); | |
1567 | if (err) | |
1568 | goto out; | |
1569 | ||
1570 | fdentry = ffhp->fh_dentry; | |
1571 | fdir = fdentry->d_inode; | |
1572 | ||
1573 | tdentry = tfhp->fh_dentry; | |
1574 | tdir = tdentry->d_inode; | |
1575 | ||
1576 | err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; | |
a56f3937 | 1577 | if (ffhp->fh_export != tfhp->fh_export) |
1da177e4 LT |
1578 | goto out; |
1579 | ||
1580 | err = nfserr_perm; | |
1581 | if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) | |
1582 | goto out; | |
1583 | ||
1584 | /* cannot use fh_lock as we need deadlock protective ordering | |
1585 | * so do it by hand */ | |
1586 | trap = lock_rename(tdentry, fdentry); | |
1587 | ffhp->fh_locked = tfhp->fh_locked = 1; | |
1588 | fill_pre_wcc(ffhp); | |
1589 | fill_pre_wcc(tfhp); | |
1590 | ||
1591 | odentry = lookup_one_len(fname, fdentry, flen); | |
1592 | err = PTR_ERR(odentry); | |
1593 | if (IS_ERR(odentry)) | |
1594 | goto out_nfserr; | |
1595 | ||
1596 | err = -ENOENT; | |
1597 | if (!odentry->d_inode) | |
1598 | goto out_dput_old; | |
1599 | err = -EINVAL; | |
1600 | if (odentry == trap) | |
1601 | goto out_dput_old; | |
1602 | ||
1603 | ndentry = lookup_one_len(tname, tdentry, tlen); | |
1604 | err = PTR_ERR(ndentry); | |
1605 | if (IS_ERR(ndentry)) | |
1606 | goto out_dput_old; | |
1607 | err = -ENOTEMPTY; | |
1608 | if (ndentry == trap) | |
1609 | goto out_dput_new; | |
1610 | ||
1611 | #ifdef MSNFS | |
1612 | if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) && | |
1613 | ((atomic_read(&odentry->d_count) > 1) | |
1614 | || (atomic_read(&ndentry->d_count) > 1))) { | |
d75f2b9f | 1615 | err = -EPERM; |
1da177e4 LT |
1616 | } else |
1617 | #endif | |
1618 | err = vfs_rename(fdir, odentry, tdir, ndentry); | |
1619 | if (!err && EX_ISSYNC(tfhp->fh_export)) { | |
f193fbab YT |
1620 | err = nfsd_sync_dir(tdentry); |
1621 | if (!err) | |
1622 | err = nfsd_sync_dir(fdentry); | |
1da177e4 LT |
1623 | } |
1624 | ||
1625 | out_dput_new: | |
1626 | dput(ndentry); | |
1627 | out_dput_old: | |
1628 | dput(odentry); | |
1629 | out_nfserr: | |
1630 | if (err) | |
1631 | err = nfserrno(err); | |
1632 | ||
1633 | /* we cannot reply on fh_unlock on the two filehandles, | |
1634 | * as that would do the wrong thing if the two directories | |
1635 | * were the same, so again we do it by hand | |
1636 | */ | |
1637 | fill_post_wcc(ffhp); | |
1638 | fill_post_wcc(tfhp); | |
1639 | unlock_rename(tdentry, fdentry); | |
1640 | ffhp->fh_locked = tfhp->fh_locked = 0; | |
1641 | ||
1642 | out: | |
1643 | return err; | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * Unlink a file or directory | |
1648 | * N.B. After this call fhp needs an fh_put | |
1649 | */ | |
1650 | int | |
1651 | nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, | |
1652 | char *fname, int flen) | |
1653 | { | |
1654 | struct dentry *dentry, *rdentry; | |
1655 | struct inode *dirp; | |
1656 | int err; | |
1657 | ||
1658 | err = nfserr_acces; | |
1659 | if (!flen || isdotent(fname, flen)) | |
1660 | goto out; | |
1661 | err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE); | |
1662 | if (err) | |
1663 | goto out; | |
1664 | ||
12fd3520 | 1665 | fh_lock_nested(fhp, I_MUTEX_PARENT); |
1da177e4 LT |
1666 | dentry = fhp->fh_dentry; |
1667 | dirp = dentry->d_inode; | |
1668 | ||
1669 | rdentry = lookup_one_len(fname, dentry, flen); | |
1670 | err = PTR_ERR(rdentry); | |
1671 | if (IS_ERR(rdentry)) | |
1672 | goto out_nfserr; | |
1673 | ||
1674 | if (!rdentry->d_inode) { | |
1675 | dput(rdentry); | |
1676 | err = nfserr_noent; | |
1677 | goto out; | |
1678 | } | |
1679 | ||
1680 | if (!type) | |
1681 | type = rdentry->d_inode->i_mode & S_IFMT; | |
1682 | ||
1683 | if (type != S_IFDIR) { /* It's UNLINK */ | |
1684 | #ifdef MSNFS | |
1685 | if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && | |
1686 | (atomic_read(&rdentry->d_count) > 1)) { | |
d75f2b9f | 1687 | err = -EPERM; |
1da177e4 LT |
1688 | } else |
1689 | #endif | |
1690 | err = vfs_unlink(dirp, rdentry); | |
1691 | } else { /* It's RMDIR */ | |
1692 | err = vfs_rmdir(dirp, rdentry); | |
1693 | } | |
1694 | ||
1695 | dput(rdentry); | |
1696 | ||
f193fbab YT |
1697 | if (err == 0 && |
1698 | EX_ISSYNC(fhp->fh_export)) | |
1699 | err = nfsd_sync_dir(dentry); | |
1da177e4 LT |
1700 | |
1701 | out_nfserr: | |
1702 | err = nfserrno(err); | |
f193fbab YT |
1703 | out: |
1704 | return err; | |
1da177e4 LT |
1705 | } |
1706 | ||
1707 | /* | |
1708 | * Read entries from a directory. | |
1709 | * The NFSv3/4 verifier we ignore for now. | |
1710 | */ | |
1711 | int | |
1712 | nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, | |
1713 | struct readdir_cd *cdp, encode_dent_fn func) | |
1714 | { | |
1715 | int err; | |
1716 | struct file *file; | |
1717 | loff_t offset = *offsetp; | |
1718 | ||
1719 | err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file); | |
1720 | if (err) | |
1721 | goto out; | |
1722 | ||
1723 | offset = vfs_llseek(file, offset, 0); | |
1724 | if (offset < 0) { | |
1725 | err = nfserrno((int)offset); | |
1726 | goto out_close; | |
1727 | } | |
1728 | ||
1729 | /* | |
1730 | * Read the directory entries. This silly loop is necessary because | |
1731 | * readdir() is not guaranteed to fill up the entire buffer, but | |
1732 | * may choose to do less. | |
1733 | */ | |
1734 | ||
1735 | do { | |
1736 | cdp->err = nfserr_eof; /* will be cleared on successful read */ | |
1737 | err = vfs_readdir(file, (filldir_t) func, cdp); | |
1738 | } while (err >=0 && cdp->err == nfs_ok); | |
1739 | if (err) | |
1740 | err = nfserrno(err); | |
1741 | else | |
1742 | err = cdp->err; | |
1743 | *offsetp = vfs_llseek(file, 0, 1); | |
1744 | ||
1745 | if (err == nfserr_eof || err == nfserr_toosmall) | |
1746 | err = nfs_ok; /* can still be found in ->err */ | |
1747 | out_close: | |
1748 | nfsd_close(file); | |
1749 | out: | |
1750 | return err; | |
1751 | } | |
1752 | ||
1753 | /* | |
1754 | * Get file system stats | |
1755 | * N.B. After this call fhp needs an fh_put | |
1756 | */ | |
1757 | int | |
1758 | nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat) | |
1759 | { | |
1760 | int err = fh_verify(rqstp, fhp, 0, MAY_NOP); | |
726c3342 | 1761 | if (!err && vfs_statfs(fhp->fh_dentry,stat)) |
1da177e4 LT |
1762 | err = nfserr_io; |
1763 | return err; | |
1764 | } | |
1765 | ||
1766 | /* | |
1767 | * Check for a user's access permissions to this inode. | |
1768 | */ | |
1769 | int | |
1770 | nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc) | |
1771 | { | |
1772 | struct inode *inode = dentry->d_inode; | |
1773 | int err; | |
1774 | ||
1775 | if (acc == MAY_NOP) | |
1776 | return 0; | |
1777 | #if 0 | |
1778 | dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", | |
1779 | acc, | |
1780 | (acc & MAY_READ)? " read" : "", | |
1781 | (acc & MAY_WRITE)? " write" : "", | |
1782 | (acc & MAY_EXEC)? " exec" : "", | |
1783 | (acc & MAY_SATTR)? " sattr" : "", | |
1784 | (acc & MAY_TRUNC)? " trunc" : "", | |
1785 | (acc & MAY_LOCK)? " lock" : "", | |
1786 | (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "", | |
1787 | inode->i_mode, | |
1788 | IS_IMMUTABLE(inode)? " immut" : "", | |
1789 | IS_APPEND(inode)? " append" : "", | |
1790 | IS_RDONLY(inode)? " ro" : ""); | |
1791 | dprintk(" owner %d/%d user %d/%d\n", | |
1792 | inode->i_uid, inode->i_gid, current->fsuid, current->fsgid); | |
1793 | #endif | |
1794 | ||
1795 | /* Normally we reject any write/sattr etc access on a read-only file | |
1796 | * system. But if it is IRIX doing check on write-access for a | |
1797 | * device special file, we ignore rofs. | |
1798 | */ | |
1799 | if (!(acc & MAY_LOCAL_ACCESS)) | |
1800 | if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) { | |
1801 | if (EX_RDONLY(exp) || IS_RDONLY(inode)) | |
1802 | return nfserr_rofs; | |
1803 | if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode)) | |
1804 | return nfserr_perm; | |
1805 | } | |
1806 | if ((acc & MAY_TRUNC) && IS_APPEND(inode)) | |
1807 | return nfserr_perm; | |
1808 | ||
1809 | if (acc & MAY_LOCK) { | |
1810 | /* If we cannot rely on authentication in NLM requests, | |
1811 | * just allow locks, otherwise require read permission, or | |
1812 | * ownership | |
1813 | */ | |
1814 | if (exp->ex_flags & NFSEXP_NOAUTHNLM) | |
1815 | return 0; | |
1816 | else | |
1817 | acc = MAY_READ | MAY_OWNER_OVERRIDE; | |
1818 | } | |
1819 | /* | |
1820 | * The file owner always gets access permission for accesses that | |
1821 | * would normally be checked at open time. This is to make | |
1822 | * file access work even when the client has done a fchmod(fd, 0). | |
1823 | * | |
1824 | * However, `cp foo bar' should fail nevertheless when bar is | |
1825 | * readonly. A sensible way to do this might be to reject all | |
1826 | * attempts to truncate a read-only file, because a creat() call | |
1827 | * always implies file truncation. | |
1828 | * ... but this isn't really fair. A process may reasonably call | |
1829 | * ftruncate on an open file descriptor on a file with perm 000. | |
1830 | * We must trust the client to do permission checking - using "ACCESS" | |
1831 | * with NFSv3. | |
1832 | */ | |
1833 | if ((acc & MAY_OWNER_OVERRIDE) && | |
1834 | inode->i_uid == current->fsuid) | |
1835 | return 0; | |
1836 | ||
1837 | err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL); | |
1838 | ||
1839 | /* Allow read access to binaries even when mode 111 */ | |
1840 | if (err == -EACCES && S_ISREG(inode->i_mode) && | |
1841 | acc == (MAY_READ | MAY_OWNER_OVERRIDE)) | |
1842 | err = permission(inode, MAY_EXEC, NULL); | |
1843 | ||
1844 | return err? nfserrno(err) : 0; | |
1845 | } | |
1846 | ||
1847 | void | |
1848 | nfsd_racache_shutdown(void) | |
1849 | { | |
fce1456a | 1850 | if (!raparml) |
1da177e4 LT |
1851 | return; |
1852 | dprintk("nfsd: freeing readahead buffers.\n"); | |
1853 | kfree(raparml); | |
fce1456a | 1854 | raparml = NULL; |
1da177e4 LT |
1855 | } |
1856 | /* | |
1857 | * Initialize readahead param cache | |
1858 | */ | |
1859 | int | |
1860 | nfsd_racache_init(int cache_size) | |
1861 | { | |
1862 | int i; | |
fce1456a GB |
1863 | int j = 0; |
1864 | int nperbucket; | |
1da177e4 | 1865 | |
fce1456a GB |
1866 | |
1867 | if (raparml) | |
1da177e4 | 1868 | return 0; |
fce1456a GB |
1869 | if (cache_size < 2*RAPARM_HASH_SIZE) |
1870 | cache_size = 2*RAPARM_HASH_SIZE; | |
1da177e4 LT |
1871 | raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL); |
1872 | ||
1873 | if (raparml != NULL) { | |
1874 | dprintk("nfsd: allocating %d readahead buffers.\n", | |
1875 | cache_size); | |
fce1456a GB |
1876 | for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) { |
1877 | raparm_hash[i].pb_head = NULL; | |
1878 | spin_lock_init(&raparm_hash[i].pb_lock); | |
1879 | } | |
1880 | nperbucket = cache_size >> RAPARM_HASH_BITS; | |
1da177e4 LT |
1881 | memset(raparml, 0, sizeof(struct raparms) * cache_size); |
1882 | for (i = 0; i < cache_size - 1; i++) { | |
fce1456a GB |
1883 | if (i % nperbucket == 0) |
1884 | raparm_hash[j++].pb_head = raparml + i; | |
1885 | if (i % nperbucket < nperbucket-1) | |
1886 | raparml[i].p_next = raparml + i + 1; | |
1da177e4 | 1887 | } |
1da177e4 LT |
1888 | } else { |
1889 | printk(KERN_WARNING | |
1890 | "nfsd: Could not allocate memory read-ahead cache.\n"); | |
1891 | return -ENOMEM; | |
1892 | } | |
1893 | nfsdstats.ra_size = cache_size; | |
1894 | return 0; | |
1895 | } | |
a257cdd0 AG |
1896 | |
1897 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) | |
1898 | struct posix_acl * | |
1899 | nfsd_get_posix_acl(struct svc_fh *fhp, int type) | |
1900 | { | |
1901 | struct inode *inode = fhp->fh_dentry->d_inode; | |
1902 | char *name; | |
1903 | void *value = NULL; | |
1904 | ssize_t size; | |
1905 | struct posix_acl *acl; | |
1906 | ||
5be196e5 CH |
1907 | if (!IS_POSIXACL(inode)) |
1908 | return ERR_PTR(-EOPNOTSUPP); | |
1909 | ||
1910 | switch (type) { | |
1911 | case ACL_TYPE_ACCESS: | |
1912 | name = POSIX_ACL_XATTR_ACCESS; | |
1913 | break; | |
1914 | case ACL_TYPE_DEFAULT: | |
1915 | name = POSIX_ACL_XATTR_DEFAULT; | |
1916 | break; | |
1917 | default: | |
a257cdd0 | 1918 | return ERR_PTR(-EOPNOTSUPP); |
a257cdd0 AG |
1919 | } |
1920 | ||
5be196e5 CH |
1921 | size = nfsd_getxattr(fhp->fh_dentry, name, &value); |
1922 | if (size < 0) | |
1923 | return ERR_PTR(size); | |
a257cdd0 | 1924 | |
a257cdd0 | 1925 | acl = posix_acl_from_xattr(value, size); |
a257cdd0 AG |
1926 | kfree(value); |
1927 | return acl; | |
1928 | } | |
1929 | ||
1930 | int | |
1931 | nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) | |
1932 | { | |
1933 | struct inode *inode = fhp->fh_dentry->d_inode; | |
1934 | char *name; | |
1935 | void *value = NULL; | |
1936 | size_t size; | |
1937 | int error; | |
1938 | ||
1939 | if (!IS_POSIXACL(inode) || !inode->i_op || | |
1940 | !inode->i_op->setxattr || !inode->i_op->removexattr) | |
1941 | return -EOPNOTSUPP; | |
1942 | switch(type) { | |
1943 | case ACL_TYPE_ACCESS: | |
334a13ec | 1944 | name = POSIX_ACL_XATTR_ACCESS; |
a257cdd0 AG |
1945 | break; |
1946 | case ACL_TYPE_DEFAULT: | |
334a13ec | 1947 | name = POSIX_ACL_XATTR_DEFAULT; |
a257cdd0 AG |
1948 | break; |
1949 | default: | |
1950 | return -EOPNOTSUPP; | |
1951 | } | |
1952 | ||
1953 | if (acl && acl->a_count) { | |
334a13ec | 1954 | size = posix_acl_xattr_size(acl->a_count); |
a257cdd0 AG |
1955 | value = kmalloc(size, GFP_KERNEL); |
1956 | if (!value) | |
1957 | return -ENOMEM; | |
9ccfc29c FM |
1958 | error = posix_acl_to_xattr(acl, value, size); |
1959 | if (error < 0) | |
a257cdd0 | 1960 | goto getout; |
9ccfc29c | 1961 | size = error; |
a257cdd0 AG |
1962 | } else |
1963 | size = 0; | |
1964 | ||
a257cdd0 | 1965 | if (size) |
5be196e5 | 1966 | error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); |
a257cdd0 AG |
1967 | else { |
1968 | if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) | |
1969 | error = 0; | |
1970 | else { | |
5be196e5 | 1971 | error = vfs_removexattr(fhp->fh_dentry, name); |
a257cdd0 AG |
1972 | if (error == -ENODATA) |
1973 | error = 0; | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | getout: | |
1978 | kfree(value); | |
1979 | return error; | |
1980 | } | |
1981 | #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ |