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