x86: change 64-bit __change_page_attr() to struct page
[linux-2.6/kvm.git] / fs / fuse / dir.c
blob80d2f5292cf91f490532e903c0e37795e705de15
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2006 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
9 #include "fuse_i.h"
11 #include <linux/pagemap.h>
12 #include <linux/file.h>
13 #include <linux/gfp.h>
14 #include <linux/sched.h>
15 #include <linux/namei.h>
17 #if BITS_PER_LONG >= 64
18 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
20 entry->d_time = time;
23 static inline u64 fuse_dentry_time(struct dentry *entry)
25 return entry->d_time;
27 #else
29 * On 32 bit archs store the high 32 bits of time in d_fsdata
31 static void fuse_dentry_settime(struct dentry *entry, u64 time)
33 entry->d_time = time;
34 entry->d_fsdata = (void *) (unsigned long) (time >> 32);
37 static u64 fuse_dentry_time(struct dentry *entry)
39 return (u64) entry->d_time +
40 ((u64) (unsigned long) entry->d_fsdata << 32);
42 #endif
45 * FUSE caches dentries and attributes with separate timeout. The
46 * time in jiffies until the dentry/attributes are valid is stored in
47 * dentry->d_time and fuse_inode->i_time respectively.
51 * Calculate the time in jiffies until a dentry/attributes are valid
53 static u64 time_to_jiffies(unsigned long sec, unsigned long nsec)
55 if (sec || nsec) {
56 struct timespec ts = {sec, nsec};
57 return get_jiffies_64() + timespec_to_jiffies(&ts);
58 } else
59 return 0;
63 * Set dentry and possibly attribute timeouts from the lookup/mk*
64 * replies
66 static void fuse_change_entry_timeout(struct dentry *entry,
67 struct fuse_entry_out *o)
69 fuse_dentry_settime(entry,
70 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
73 static u64 attr_timeout(struct fuse_attr_out *o)
75 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
78 static u64 entry_attr_timeout(struct fuse_entry_out *o)
80 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
84 * Mark the attributes as stale, so that at the next call to
85 * ->getattr() they will be fetched from userspace
87 void fuse_invalidate_attr(struct inode *inode)
89 get_fuse_inode(inode)->i_time = 0;
93 * Just mark the entry as stale, so that a next attempt to look it up
94 * will result in a new lookup call to userspace
96 * This is called when a dentry is about to become negative and the
97 * timeout is unknown (unlink, rmdir, rename and in some cases
98 * lookup)
100 static void fuse_invalidate_entry_cache(struct dentry *entry)
102 fuse_dentry_settime(entry, 0);
106 * Same as fuse_invalidate_entry_cache(), but also try to remove the
107 * dentry from the hash
109 static void fuse_invalidate_entry(struct dentry *entry)
111 d_invalidate(entry);
112 fuse_invalidate_entry_cache(entry);
115 static void fuse_lookup_init(struct fuse_req *req, struct inode *dir,
116 struct dentry *entry,
117 struct fuse_entry_out *outarg)
119 struct fuse_conn *fc = get_fuse_conn(dir);
121 memset(outarg, 0, sizeof(struct fuse_entry_out));
122 req->in.h.opcode = FUSE_LOOKUP;
123 req->in.h.nodeid = get_node_id(dir);
124 req->in.numargs = 1;
125 req->in.args[0].size = entry->d_name.len + 1;
126 req->in.args[0].value = entry->d_name.name;
127 req->out.numargs = 1;
128 if (fc->minor < 9)
129 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
130 else
131 req->out.args[0].size = sizeof(struct fuse_entry_out);
132 req->out.args[0].value = outarg;
135 static u64 fuse_get_attr_version(struct fuse_conn *fc)
137 u64 curr_version;
140 * The spin lock isn't actually needed on 64bit archs, but we
141 * don't yet care too much about such optimizations.
143 spin_lock(&fc->lock);
144 curr_version = fc->attr_version;
145 spin_unlock(&fc->lock);
147 return curr_version;
151 * Check whether the dentry is still valid
153 * If the entry validity timeout has expired and the dentry is
154 * positive, try to redo the lookup. If the lookup results in a
155 * different inode, then let the VFS invalidate the dentry and redo
156 * the lookup once more. If the lookup results in the same inode,
157 * then refresh the attributes, timeouts and mark the dentry valid.
159 static int fuse_dentry_revalidate(struct dentry *entry, struct nameidata *nd)
161 struct inode *inode = entry->d_inode;
163 if (inode && is_bad_inode(inode))
164 return 0;
165 else if (fuse_dentry_time(entry) < get_jiffies_64()) {
166 int err;
167 struct fuse_entry_out outarg;
168 struct fuse_conn *fc;
169 struct fuse_req *req;
170 struct fuse_req *forget_req;
171 struct dentry *parent;
172 u64 attr_version;
174 /* For negative dentries, always do a fresh lookup */
175 if (!inode)
176 return 0;
178 fc = get_fuse_conn(inode);
179 req = fuse_get_req(fc);
180 if (IS_ERR(req))
181 return 0;
183 forget_req = fuse_get_req(fc);
184 if (IS_ERR(forget_req)) {
185 fuse_put_request(fc, req);
186 return 0;
189 attr_version = fuse_get_attr_version(fc);
191 parent = dget_parent(entry);
192 fuse_lookup_init(req, parent->d_inode, entry, &outarg);
193 request_send(fc, req);
194 dput(parent);
195 err = req->out.h.error;
196 fuse_put_request(fc, req);
197 /* Zero nodeid is same as -ENOENT */
198 if (!err && !outarg.nodeid)
199 err = -ENOENT;
200 if (!err) {
201 struct fuse_inode *fi = get_fuse_inode(inode);
202 if (outarg.nodeid != get_node_id(inode)) {
203 fuse_send_forget(fc, forget_req,
204 outarg.nodeid, 1);
205 return 0;
207 spin_lock(&fc->lock);
208 fi->nlookup ++;
209 spin_unlock(&fc->lock);
211 fuse_put_request(fc, forget_req);
212 if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
213 return 0;
215 fuse_change_attributes(inode, &outarg.attr,
216 entry_attr_timeout(&outarg),
217 attr_version);
218 fuse_change_entry_timeout(entry, &outarg);
220 return 1;
223 static int invalid_nodeid(u64 nodeid)
225 return !nodeid || nodeid == FUSE_ROOT_ID;
228 static struct dentry_operations fuse_dentry_operations = {
229 .d_revalidate = fuse_dentry_revalidate,
232 int fuse_valid_type(int m)
234 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
235 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
239 * Add a directory inode to a dentry, ensuring that no other dentry
240 * refers to this inode. Called with fc->inst_mutex.
242 static int fuse_d_add_directory(struct dentry *entry, struct inode *inode)
244 struct dentry *alias = d_find_alias(inode);
245 if (alias) {
246 /* This tries to shrink the subtree below alias */
247 fuse_invalidate_entry(alias);
248 dput(alias);
249 if (!list_empty(&inode->i_dentry))
250 return -EBUSY;
252 d_add(entry, inode);
253 return 0;
256 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
257 struct nameidata *nd)
259 int err;
260 struct fuse_entry_out outarg;
261 struct inode *inode = NULL;
262 struct fuse_conn *fc = get_fuse_conn(dir);
263 struct fuse_req *req;
264 struct fuse_req *forget_req;
265 u64 attr_version;
267 if (entry->d_name.len > FUSE_NAME_MAX)
268 return ERR_PTR(-ENAMETOOLONG);
270 req = fuse_get_req(fc);
271 if (IS_ERR(req))
272 return ERR_PTR(PTR_ERR(req));
274 forget_req = fuse_get_req(fc);
275 if (IS_ERR(forget_req)) {
276 fuse_put_request(fc, req);
277 return ERR_PTR(PTR_ERR(forget_req));
280 attr_version = fuse_get_attr_version(fc);
282 fuse_lookup_init(req, dir, entry, &outarg);
283 request_send(fc, req);
284 err = req->out.h.error;
285 fuse_put_request(fc, req);
286 /* Zero nodeid is same as -ENOENT, but with valid timeout */
287 if (!err && outarg.nodeid &&
288 (invalid_nodeid(outarg.nodeid) ||
289 !fuse_valid_type(outarg.attr.mode)))
290 err = -EIO;
291 if (!err && outarg.nodeid) {
292 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
293 &outarg.attr, entry_attr_timeout(&outarg),
294 attr_version);
295 if (!inode) {
296 fuse_send_forget(fc, forget_req, outarg.nodeid, 1);
297 return ERR_PTR(-ENOMEM);
300 fuse_put_request(fc, forget_req);
301 if (err && err != -ENOENT)
302 return ERR_PTR(err);
304 if (inode && S_ISDIR(inode->i_mode)) {
305 mutex_lock(&fc->inst_mutex);
306 err = fuse_d_add_directory(entry, inode);
307 mutex_unlock(&fc->inst_mutex);
308 if (err) {
309 iput(inode);
310 return ERR_PTR(err);
312 } else
313 d_add(entry, inode);
315 entry->d_op = &fuse_dentry_operations;
316 if (!err)
317 fuse_change_entry_timeout(entry, &outarg);
318 else
319 fuse_invalidate_entry_cache(entry);
320 return NULL;
324 * Synchronous release for the case when something goes wrong in CREATE_OPEN
326 static void fuse_sync_release(struct fuse_conn *fc, struct fuse_file *ff,
327 u64 nodeid, int flags)
329 fuse_release_fill(ff, nodeid, flags, FUSE_RELEASE);
330 ff->reserved_req->force = 1;
331 request_send(fc, ff->reserved_req);
332 fuse_put_request(fc, ff->reserved_req);
333 kfree(ff);
337 * Atomic create+open operation
339 * If the filesystem doesn't support this, then fall back to separate
340 * 'mknod' + 'open' requests.
342 static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
343 struct nameidata *nd)
345 int err;
346 struct inode *inode;
347 struct fuse_conn *fc = get_fuse_conn(dir);
348 struct fuse_req *req;
349 struct fuse_req *forget_req;
350 struct fuse_open_in inarg;
351 struct fuse_open_out outopen;
352 struct fuse_entry_out outentry;
353 struct fuse_file *ff;
354 struct file *file;
355 int flags = nd->intent.open.flags - 1;
357 if (fc->no_create)
358 return -ENOSYS;
360 forget_req = fuse_get_req(fc);
361 if (IS_ERR(forget_req))
362 return PTR_ERR(forget_req);
364 req = fuse_get_req(fc);
365 err = PTR_ERR(req);
366 if (IS_ERR(req))
367 goto out_put_forget_req;
369 err = -ENOMEM;
370 ff = fuse_file_alloc();
371 if (!ff)
372 goto out_put_request;
374 flags &= ~O_NOCTTY;
375 memset(&inarg, 0, sizeof(inarg));
376 memset(&outentry, 0, sizeof(outentry));
377 inarg.flags = flags;
378 inarg.mode = mode;
379 req->in.h.opcode = FUSE_CREATE;
380 req->in.h.nodeid = get_node_id(dir);
381 req->in.numargs = 2;
382 req->in.args[0].size = sizeof(inarg);
383 req->in.args[0].value = &inarg;
384 req->in.args[1].size = entry->d_name.len + 1;
385 req->in.args[1].value = entry->d_name.name;
386 req->out.numargs = 2;
387 if (fc->minor < 9)
388 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
389 else
390 req->out.args[0].size = sizeof(outentry);
391 req->out.args[0].value = &outentry;
392 req->out.args[1].size = sizeof(outopen);
393 req->out.args[1].value = &outopen;
394 request_send(fc, req);
395 err = req->out.h.error;
396 if (err) {
397 if (err == -ENOSYS)
398 fc->no_create = 1;
399 goto out_free_ff;
402 err = -EIO;
403 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
404 goto out_free_ff;
406 fuse_put_request(fc, req);
407 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
408 &outentry.attr, entry_attr_timeout(&outentry), 0);
409 if (!inode) {
410 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
411 ff->fh = outopen.fh;
412 fuse_sync_release(fc, ff, outentry.nodeid, flags);
413 fuse_send_forget(fc, forget_req, outentry.nodeid, 1);
414 return -ENOMEM;
416 fuse_put_request(fc, forget_req);
417 d_instantiate(entry, inode);
418 fuse_change_entry_timeout(entry, &outentry);
419 file = lookup_instantiate_filp(nd, entry, generic_file_open);
420 if (IS_ERR(file)) {
421 ff->fh = outopen.fh;
422 fuse_sync_release(fc, ff, outentry.nodeid, flags);
423 return PTR_ERR(file);
425 fuse_finish_open(inode, file, ff, &outopen);
426 return 0;
428 out_free_ff:
429 fuse_file_free(ff);
430 out_put_request:
431 fuse_put_request(fc, req);
432 out_put_forget_req:
433 fuse_put_request(fc, forget_req);
434 return err;
438 * Code shared between mknod, mkdir, symlink and link
440 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
441 struct inode *dir, struct dentry *entry,
442 int mode)
444 struct fuse_entry_out outarg;
445 struct inode *inode;
446 int err;
447 struct fuse_req *forget_req;
449 forget_req = fuse_get_req(fc);
450 if (IS_ERR(forget_req)) {
451 fuse_put_request(fc, req);
452 return PTR_ERR(forget_req);
455 memset(&outarg, 0, sizeof(outarg));
456 req->in.h.nodeid = get_node_id(dir);
457 req->out.numargs = 1;
458 if (fc->minor < 9)
459 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
460 else
461 req->out.args[0].size = sizeof(outarg);
462 req->out.args[0].value = &outarg;
463 request_send(fc, req);
464 err = req->out.h.error;
465 fuse_put_request(fc, req);
466 if (err)
467 goto out_put_forget_req;
469 err = -EIO;
470 if (invalid_nodeid(outarg.nodeid))
471 goto out_put_forget_req;
473 if ((outarg.attr.mode ^ mode) & S_IFMT)
474 goto out_put_forget_req;
476 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
477 &outarg.attr, entry_attr_timeout(&outarg), 0);
478 if (!inode) {
479 fuse_send_forget(fc, forget_req, outarg.nodeid, 1);
480 return -ENOMEM;
482 fuse_put_request(fc, forget_req);
484 if (S_ISDIR(inode->i_mode)) {
485 struct dentry *alias;
486 mutex_lock(&fc->inst_mutex);
487 alias = d_find_alias(inode);
488 if (alias) {
489 /* New directory must have moved since mkdir */
490 mutex_unlock(&fc->inst_mutex);
491 dput(alias);
492 iput(inode);
493 return -EBUSY;
495 d_instantiate(entry, inode);
496 mutex_unlock(&fc->inst_mutex);
497 } else
498 d_instantiate(entry, inode);
500 fuse_change_entry_timeout(entry, &outarg);
501 fuse_invalidate_attr(dir);
502 return 0;
504 out_put_forget_req:
505 fuse_put_request(fc, forget_req);
506 return err;
509 static int fuse_mknod(struct inode *dir, struct dentry *entry, int mode,
510 dev_t rdev)
512 struct fuse_mknod_in inarg;
513 struct fuse_conn *fc = get_fuse_conn(dir);
514 struct fuse_req *req = fuse_get_req(fc);
515 if (IS_ERR(req))
516 return PTR_ERR(req);
518 memset(&inarg, 0, sizeof(inarg));
519 inarg.mode = mode;
520 inarg.rdev = new_encode_dev(rdev);
521 req->in.h.opcode = FUSE_MKNOD;
522 req->in.numargs = 2;
523 req->in.args[0].size = sizeof(inarg);
524 req->in.args[0].value = &inarg;
525 req->in.args[1].size = entry->d_name.len + 1;
526 req->in.args[1].value = entry->d_name.name;
527 return create_new_entry(fc, req, dir, entry, mode);
530 static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
531 struct nameidata *nd)
533 if (nd && (nd->flags & LOOKUP_OPEN)) {
534 int err = fuse_create_open(dir, entry, mode, nd);
535 if (err != -ENOSYS)
536 return err;
537 /* Fall back on mknod */
539 return fuse_mknod(dir, entry, mode, 0);
542 static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
544 struct fuse_mkdir_in inarg;
545 struct fuse_conn *fc = get_fuse_conn(dir);
546 struct fuse_req *req = fuse_get_req(fc);
547 if (IS_ERR(req))
548 return PTR_ERR(req);
550 memset(&inarg, 0, sizeof(inarg));
551 inarg.mode = mode;
552 req->in.h.opcode = FUSE_MKDIR;
553 req->in.numargs = 2;
554 req->in.args[0].size = sizeof(inarg);
555 req->in.args[0].value = &inarg;
556 req->in.args[1].size = entry->d_name.len + 1;
557 req->in.args[1].value = entry->d_name.name;
558 return create_new_entry(fc, req, dir, entry, S_IFDIR);
561 static int fuse_symlink(struct inode *dir, struct dentry *entry,
562 const char *link)
564 struct fuse_conn *fc = get_fuse_conn(dir);
565 unsigned len = strlen(link) + 1;
566 struct fuse_req *req = fuse_get_req(fc);
567 if (IS_ERR(req))
568 return PTR_ERR(req);
570 req->in.h.opcode = FUSE_SYMLINK;
571 req->in.numargs = 2;
572 req->in.args[0].size = entry->d_name.len + 1;
573 req->in.args[0].value = entry->d_name.name;
574 req->in.args[1].size = len;
575 req->in.args[1].value = link;
576 return create_new_entry(fc, req, dir, entry, S_IFLNK);
579 static int fuse_unlink(struct inode *dir, struct dentry *entry)
581 int err;
582 struct fuse_conn *fc = get_fuse_conn(dir);
583 struct fuse_req *req = fuse_get_req(fc);
584 if (IS_ERR(req))
585 return PTR_ERR(req);
587 req->in.h.opcode = FUSE_UNLINK;
588 req->in.h.nodeid = get_node_id(dir);
589 req->in.numargs = 1;
590 req->in.args[0].size = entry->d_name.len + 1;
591 req->in.args[0].value = entry->d_name.name;
592 request_send(fc, req);
593 err = req->out.h.error;
594 fuse_put_request(fc, req);
595 if (!err) {
596 struct inode *inode = entry->d_inode;
598 /* Set nlink to zero so the inode can be cleared, if
599 the inode does have more links this will be
600 discovered at the next lookup/getattr */
601 clear_nlink(inode);
602 fuse_invalidate_attr(inode);
603 fuse_invalidate_attr(dir);
604 fuse_invalidate_entry_cache(entry);
605 } else if (err == -EINTR)
606 fuse_invalidate_entry(entry);
607 return err;
610 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
612 int err;
613 struct fuse_conn *fc = get_fuse_conn(dir);
614 struct fuse_req *req = fuse_get_req(fc);
615 if (IS_ERR(req))
616 return PTR_ERR(req);
618 req->in.h.opcode = FUSE_RMDIR;
619 req->in.h.nodeid = get_node_id(dir);
620 req->in.numargs = 1;
621 req->in.args[0].size = entry->d_name.len + 1;
622 req->in.args[0].value = entry->d_name.name;
623 request_send(fc, req);
624 err = req->out.h.error;
625 fuse_put_request(fc, req);
626 if (!err) {
627 clear_nlink(entry->d_inode);
628 fuse_invalidate_attr(dir);
629 fuse_invalidate_entry_cache(entry);
630 } else if (err == -EINTR)
631 fuse_invalidate_entry(entry);
632 return err;
635 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
636 struct inode *newdir, struct dentry *newent)
638 int err;
639 struct fuse_rename_in inarg;
640 struct fuse_conn *fc = get_fuse_conn(olddir);
641 struct fuse_req *req = fuse_get_req(fc);
642 if (IS_ERR(req))
643 return PTR_ERR(req);
645 memset(&inarg, 0, sizeof(inarg));
646 inarg.newdir = get_node_id(newdir);
647 req->in.h.opcode = FUSE_RENAME;
648 req->in.h.nodeid = get_node_id(olddir);
649 req->in.numargs = 3;
650 req->in.args[0].size = sizeof(inarg);
651 req->in.args[0].value = &inarg;
652 req->in.args[1].size = oldent->d_name.len + 1;
653 req->in.args[1].value = oldent->d_name.name;
654 req->in.args[2].size = newent->d_name.len + 1;
655 req->in.args[2].value = newent->d_name.name;
656 request_send(fc, req);
657 err = req->out.h.error;
658 fuse_put_request(fc, req);
659 if (!err) {
660 /* ctime changes */
661 fuse_invalidate_attr(oldent->d_inode);
663 fuse_invalidate_attr(olddir);
664 if (olddir != newdir)
665 fuse_invalidate_attr(newdir);
667 /* newent will end up negative */
668 if (newent->d_inode)
669 fuse_invalidate_entry_cache(newent);
670 } else if (err == -EINTR) {
671 /* If request was interrupted, DEITY only knows if the
672 rename actually took place. If the invalidation
673 fails (e.g. some process has CWD under the renamed
674 directory), then there can be inconsistency between
675 the dcache and the real filesystem. Tough luck. */
676 fuse_invalidate_entry(oldent);
677 if (newent->d_inode)
678 fuse_invalidate_entry(newent);
681 return err;
684 static int fuse_link(struct dentry *entry, struct inode *newdir,
685 struct dentry *newent)
687 int err;
688 struct fuse_link_in inarg;
689 struct inode *inode = entry->d_inode;
690 struct fuse_conn *fc = get_fuse_conn(inode);
691 struct fuse_req *req = fuse_get_req(fc);
692 if (IS_ERR(req))
693 return PTR_ERR(req);
695 memset(&inarg, 0, sizeof(inarg));
696 inarg.oldnodeid = get_node_id(inode);
697 req->in.h.opcode = FUSE_LINK;
698 req->in.numargs = 2;
699 req->in.args[0].size = sizeof(inarg);
700 req->in.args[0].value = &inarg;
701 req->in.args[1].size = newent->d_name.len + 1;
702 req->in.args[1].value = newent->d_name.name;
703 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
704 /* Contrary to "normal" filesystems it can happen that link
705 makes two "logical" inodes point to the same "physical"
706 inode. We invalidate the attributes of the old one, so it
707 will reflect changes in the backing inode (link count,
708 etc.)
710 if (!err || err == -EINTR)
711 fuse_invalidate_attr(inode);
712 return err;
715 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
716 struct kstat *stat)
718 stat->dev = inode->i_sb->s_dev;
719 stat->ino = attr->ino;
720 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
721 stat->nlink = attr->nlink;
722 stat->uid = attr->uid;
723 stat->gid = attr->gid;
724 stat->rdev = inode->i_rdev;
725 stat->atime.tv_sec = attr->atime;
726 stat->atime.tv_nsec = attr->atimensec;
727 stat->mtime.tv_sec = attr->mtime;
728 stat->mtime.tv_nsec = attr->mtimensec;
729 stat->ctime.tv_sec = attr->ctime;
730 stat->ctime.tv_nsec = attr->ctimensec;
731 stat->size = attr->size;
732 stat->blocks = attr->blocks;
733 stat->blksize = (1 << inode->i_blkbits);
736 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
737 struct file *file)
739 int err;
740 struct fuse_getattr_in inarg;
741 struct fuse_attr_out outarg;
742 struct fuse_conn *fc = get_fuse_conn(inode);
743 struct fuse_req *req;
744 u64 attr_version;
746 req = fuse_get_req(fc);
747 if (IS_ERR(req))
748 return PTR_ERR(req);
750 attr_version = fuse_get_attr_version(fc);
752 memset(&inarg, 0, sizeof(inarg));
753 memset(&outarg, 0, sizeof(outarg));
754 /* Directories have separate file-handle space */
755 if (file && S_ISREG(inode->i_mode)) {
756 struct fuse_file *ff = file->private_data;
758 inarg.getattr_flags |= FUSE_GETATTR_FH;
759 inarg.fh = ff->fh;
761 req->in.h.opcode = FUSE_GETATTR;
762 req->in.h.nodeid = get_node_id(inode);
763 req->in.numargs = 1;
764 req->in.args[0].size = sizeof(inarg);
765 req->in.args[0].value = &inarg;
766 req->out.numargs = 1;
767 if (fc->minor < 9)
768 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
769 else
770 req->out.args[0].size = sizeof(outarg);
771 req->out.args[0].value = &outarg;
772 request_send(fc, req);
773 err = req->out.h.error;
774 fuse_put_request(fc, req);
775 if (!err) {
776 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
777 make_bad_inode(inode);
778 err = -EIO;
779 } else {
780 fuse_change_attributes(inode, &outarg.attr,
781 attr_timeout(&outarg),
782 attr_version);
783 if (stat)
784 fuse_fillattr(inode, &outarg.attr, stat);
787 return err;
790 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
791 struct file *file, bool *refreshed)
793 struct fuse_inode *fi = get_fuse_inode(inode);
794 int err;
795 bool r;
797 if (fi->i_time < get_jiffies_64()) {
798 r = true;
799 err = fuse_do_getattr(inode, stat, file);
800 } else {
801 r = false;
802 err = 0;
803 if (stat) {
804 generic_fillattr(inode, stat);
805 stat->mode = fi->orig_i_mode;
809 if (refreshed != NULL)
810 *refreshed = r;
812 return err;
816 * Calling into a user-controlled filesystem gives the filesystem
817 * daemon ptrace-like capabilities over the requester process. This
818 * means, that the filesystem daemon is able to record the exact
819 * filesystem operations performed, and can also control the behavior
820 * of the requester process in otherwise impossible ways. For example
821 * it can delay the operation for arbitrary length of time allowing
822 * DoS against the requester.
824 * For this reason only those processes can call into the filesystem,
825 * for which the owner of the mount has ptrace privilege. This
826 * excludes processes started by other users, suid or sgid processes.
828 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
830 if (fc->flags & FUSE_ALLOW_OTHER)
831 return 1;
833 if (task->euid == fc->user_id &&
834 task->suid == fc->user_id &&
835 task->uid == fc->user_id &&
836 task->egid == fc->group_id &&
837 task->sgid == fc->group_id &&
838 task->gid == fc->group_id)
839 return 1;
841 return 0;
844 static int fuse_access(struct inode *inode, int mask)
846 struct fuse_conn *fc = get_fuse_conn(inode);
847 struct fuse_req *req;
848 struct fuse_access_in inarg;
849 int err;
851 if (fc->no_access)
852 return 0;
854 req = fuse_get_req(fc);
855 if (IS_ERR(req))
856 return PTR_ERR(req);
858 memset(&inarg, 0, sizeof(inarg));
859 inarg.mask = mask;
860 req->in.h.opcode = FUSE_ACCESS;
861 req->in.h.nodeid = get_node_id(inode);
862 req->in.numargs = 1;
863 req->in.args[0].size = sizeof(inarg);
864 req->in.args[0].value = &inarg;
865 request_send(fc, req);
866 err = req->out.h.error;
867 fuse_put_request(fc, req);
868 if (err == -ENOSYS) {
869 fc->no_access = 1;
870 err = 0;
872 return err;
876 * Check permission. The two basic access models of FUSE are:
878 * 1) Local access checking ('default_permissions' mount option) based
879 * on file mode. This is the plain old disk filesystem permission
880 * modell.
882 * 2) "Remote" access checking, where server is responsible for
883 * checking permission in each inode operation. An exception to this
884 * is if ->permission() was invoked from sys_access() in which case an
885 * access request is sent. Execute permission is still checked
886 * locally based on file mode.
888 static int fuse_permission(struct inode *inode, int mask, struct nameidata *nd)
890 struct fuse_conn *fc = get_fuse_conn(inode);
891 bool refreshed = false;
892 int err = 0;
894 if (!fuse_allow_task(fc, current))
895 return -EACCES;
898 * If attributes are needed, refresh them before proceeding
900 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
901 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
902 err = fuse_update_attributes(inode, NULL, NULL, &refreshed);
903 if (err)
904 return err;
907 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
908 int err = generic_permission(inode, mask, NULL);
910 /* If permission is denied, try to refresh file
911 attributes. This is also needed, because the root
912 node will at first have no permissions */
913 if (err == -EACCES && !refreshed) {
914 err = fuse_do_getattr(inode, NULL, NULL);
915 if (!err)
916 err = generic_permission(inode, mask, NULL);
919 /* Note: the opposite of the above test does not
920 exist. So if permissions are revoked this won't be
921 noticed immediately, only after the attribute
922 timeout has expired */
923 } else if (nd && (nd->flags & (LOOKUP_ACCESS | LOOKUP_CHDIR))) {
924 err = fuse_access(inode, mask);
925 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
926 if (!(inode->i_mode & S_IXUGO)) {
927 if (refreshed)
928 return -EACCES;
930 err = fuse_do_getattr(inode, NULL, NULL);
931 if (!err && !(inode->i_mode & S_IXUGO))
932 return -EACCES;
935 return err;
938 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
939 void *dstbuf, filldir_t filldir)
941 while (nbytes >= FUSE_NAME_OFFSET) {
942 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
943 size_t reclen = FUSE_DIRENT_SIZE(dirent);
944 int over;
945 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
946 return -EIO;
947 if (reclen > nbytes)
948 break;
950 over = filldir(dstbuf, dirent->name, dirent->namelen,
951 file->f_pos, dirent->ino, dirent->type);
952 if (over)
953 break;
955 buf += reclen;
956 nbytes -= reclen;
957 file->f_pos = dirent->off;
960 return 0;
963 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
965 int err;
966 size_t nbytes;
967 struct page *page;
968 struct inode *inode = file->f_path.dentry->d_inode;
969 struct fuse_conn *fc = get_fuse_conn(inode);
970 struct fuse_req *req;
972 if (is_bad_inode(inode))
973 return -EIO;
975 req = fuse_get_req(fc);
976 if (IS_ERR(req))
977 return PTR_ERR(req);
979 page = alloc_page(GFP_KERNEL);
980 if (!page) {
981 fuse_put_request(fc, req);
982 return -ENOMEM;
984 req->num_pages = 1;
985 req->pages[0] = page;
986 fuse_read_fill(req, file, inode, file->f_pos, PAGE_SIZE, FUSE_READDIR);
987 request_send(fc, req);
988 nbytes = req->out.args[0].size;
989 err = req->out.h.error;
990 fuse_put_request(fc, req);
991 if (!err)
992 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
993 filldir);
995 __free_page(page);
996 fuse_invalidate_attr(inode); /* atime changed */
997 return err;
1000 static char *read_link(struct dentry *dentry)
1002 struct inode *inode = dentry->d_inode;
1003 struct fuse_conn *fc = get_fuse_conn(inode);
1004 struct fuse_req *req = fuse_get_req(fc);
1005 char *link;
1007 if (IS_ERR(req))
1008 return ERR_PTR(PTR_ERR(req));
1010 link = (char *) __get_free_page(GFP_KERNEL);
1011 if (!link) {
1012 link = ERR_PTR(-ENOMEM);
1013 goto out;
1015 req->in.h.opcode = FUSE_READLINK;
1016 req->in.h.nodeid = get_node_id(inode);
1017 req->out.argvar = 1;
1018 req->out.numargs = 1;
1019 req->out.args[0].size = PAGE_SIZE - 1;
1020 req->out.args[0].value = link;
1021 request_send(fc, req);
1022 if (req->out.h.error) {
1023 free_page((unsigned long) link);
1024 link = ERR_PTR(req->out.h.error);
1025 } else
1026 link[req->out.args[0].size] = '\0';
1027 out:
1028 fuse_put_request(fc, req);
1029 fuse_invalidate_attr(inode); /* atime changed */
1030 return link;
1033 static void free_link(char *link)
1035 if (!IS_ERR(link))
1036 free_page((unsigned long) link);
1039 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1041 nd_set_link(nd, read_link(dentry));
1042 return NULL;
1045 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1047 free_link(nd_get_link(nd));
1050 static int fuse_dir_open(struct inode *inode, struct file *file)
1052 return fuse_open_common(inode, file, 1);
1055 static int fuse_dir_release(struct inode *inode, struct file *file)
1057 return fuse_release_common(inode, file, 1);
1060 static int fuse_dir_fsync(struct file *file, struct dentry *de, int datasync)
1062 /* nfsd can call this with no file */
1063 return file ? fuse_fsync_common(file, de, datasync, 1) : 0;
1066 static bool update_mtime(unsigned ivalid)
1068 /* Always update if mtime is explicitly set */
1069 if (ivalid & ATTR_MTIME_SET)
1070 return true;
1072 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1073 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1074 return false;
1076 /* In all other cases update */
1077 return true;
1080 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1082 unsigned ivalid = iattr->ia_valid;
1084 if (ivalid & ATTR_MODE)
1085 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1086 if (ivalid & ATTR_UID)
1087 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1088 if (ivalid & ATTR_GID)
1089 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1090 if (ivalid & ATTR_SIZE)
1091 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1092 if (ivalid & ATTR_ATIME) {
1093 arg->valid |= FATTR_ATIME;
1094 arg->atime = iattr->ia_atime.tv_sec;
1095 arg->atimensec = iattr->ia_atime.tv_nsec;
1096 if (!(ivalid & ATTR_ATIME_SET))
1097 arg->valid |= FATTR_ATIME_NOW;
1099 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1100 arg->valid |= FATTR_MTIME;
1101 arg->mtime = iattr->ia_mtime.tv_sec;
1102 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1103 if (!(ivalid & ATTR_MTIME_SET))
1104 arg->valid |= FATTR_MTIME_NOW;
1109 * Set attributes, and at the same time refresh them.
1111 * Truncation is slightly complicated, because the 'truncate' request
1112 * may fail, in which case we don't want to touch the mapping.
1113 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1114 * and the actual truncation by hand.
1116 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1117 struct file *file)
1119 struct inode *inode = entry->d_inode;
1120 struct fuse_conn *fc = get_fuse_conn(inode);
1121 struct fuse_req *req;
1122 struct fuse_setattr_in inarg;
1123 struct fuse_attr_out outarg;
1124 int err;
1126 if (!fuse_allow_task(fc, current))
1127 return -EACCES;
1129 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1130 err = inode_change_ok(inode, attr);
1131 if (err)
1132 return err;
1135 if ((attr->ia_valid & ATTR_OPEN) && fc->atomic_o_trunc)
1136 return 0;
1138 if (attr->ia_valid & ATTR_SIZE) {
1139 unsigned long limit;
1140 if (IS_SWAPFILE(inode))
1141 return -ETXTBSY;
1142 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1143 if (limit != RLIM_INFINITY && attr->ia_size > (loff_t) limit) {
1144 send_sig(SIGXFSZ, current, 0);
1145 return -EFBIG;
1149 req = fuse_get_req(fc);
1150 if (IS_ERR(req))
1151 return PTR_ERR(req);
1153 memset(&inarg, 0, sizeof(inarg));
1154 memset(&outarg, 0, sizeof(outarg));
1155 iattr_to_fattr(attr, &inarg);
1156 if (file) {
1157 struct fuse_file *ff = file->private_data;
1158 inarg.valid |= FATTR_FH;
1159 inarg.fh = ff->fh;
1161 if (attr->ia_valid & ATTR_SIZE) {
1162 /* For mandatory locking in truncate */
1163 inarg.valid |= FATTR_LOCKOWNER;
1164 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1166 req->in.h.opcode = FUSE_SETATTR;
1167 req->in.h.nodeid = get_node_id(inode);
1168 req->in.numargs = 1;
1169 req->in.args[0].size = sizeof(inarg);
1170 req->in.args[0].value = &inarg;
1171 req->out.numargs = 1;
1172 if (fc->minor < 9)
1173 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1174 else
1175 req->out.args[0].size = sizeof(outarg);
1176 req->out.args[0].value = &outarg;
1177 request_send(fc, req);
1178 err = req->out.h.error;
1179 fuse_put_request(fc, req);
1180 if (err) {
1181 if (err == -EINTR)
1182 fuse_invalidate_attr(inode);
1183 return err;
1186 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1187 make_bad_inode(inode);
1188 return -EIO;
1191 fuse_change_attributes(inode, &outarg.attr, attr_timeout(&outarg), 0);
1192 return 0;
1195 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1197 if (attr->ia_valid & ATTR_FILE)
1198 return fuse_do_setattr(entry, attr, attr->ia_file);
1199 else
1200 return fuse_do_setattr(entry, attr, NULL);
1203 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1204 struct kstat *stat)
1206 struct inode *inode = entry->d_inode;
1207 struct fuse_conn *fc = get_fuse_conn(inode);
1209 if (!fuse_allow_task(fc, current))
1210 return -EACCES;
1212 return fuse_update_attributes(inode, stat, NULL, NULL);
1215 static int fuse_setxattr(struct dentry *entry, const char *name,
1216 const void *value, size_t size, int flags)
1218 struct inode *inode = entry->d_inode;
1219 struct fuse_conn *fc = get_fuse_conn(inode);
1220 struct fuse_req *req;
1221 struct fuse_setxattr_in inarg;
1222 int err;
1224 if (fc->no_setxattr)
1225 return -EOPNOTSUPP;
1227 req = fuse_get_req(fc);
1228 if (IS_ERR(req))
1229 return PTR_ERR(req);
1231 memset(&inarg, 0, sizeof(inarg));
1232 inarg.size = size;
1233 inarg.flags = flags;
1234 req->in.h.opcode = FUSE_SETXATTR;
1235 req->in.h.nodeid = get_node_id(inode);
1236 req->in.numargs = 3;
1237 req->in.args[0].size = sizeof(inarg);
1238 req->in.args[0].value = &inarg;
1239 req->in.args[1].size = strlen(name) + 1;
1240 req->in.args[1].value = name;
1241 req->in.args[2].size = size;
1242 req->in.args[2].value = value;
1243 request_send(fc, req);
1244 err = req->out.h.error;
1245 fuse_put_request(fc, req);
1246 if (err == -ENOSYS) {
1247 fc->no_setxattr = 1;
1248 err = -EOPNOTSUPP;
1250 return err;
1253 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1254 void *value, size_t size)
1256 struct inode *inode = entry->d_inode;
1257 struct fuse_conn *fc = get_fuse_conn(inode);
1258 struct fuse_req *req;
1259 struct fuse_getxattr_in inarg;
1260 struct fuse_getxattr_out outarg;
1261 ssize_t ret;
1263 if (fc->no_getxattr)
1264 return -EOPNOTSUPP;
1266 req = fuse_get_req(fc);
1267 if (IS_ERR(req))
1268 return PTR_ERR(req);
1270 memset(&inarg, 0, sizeof(inarg));
1271 inarg.size = size;
1272 req->in.h.opcode = FUSE_GETXATTR;
1273 req->in.h.nodeid = get_node_id(inode);
1274 req->in.numargs = 2;
1275 req->in.args[0].size = sizeof(inarg);
1276 req->in.args[0].value = &inarg;
1277 req->in.args[1].size = strlen(name) + 1;
1278 req->in.args[1].value = name;
1279 /* This is really two different operations rolled into one */
1280 req->out.numargs = 1;
1281 if (size) {
1282 req->out.argvar = 1;
1283 req->out.args[0].size = size;
1284 req->out.args[0].value = value;
1285 } else {
1286 req->out.args[0].size = sizeof(outarg);
1287 req->out.args[0].value = &outarg;
1289 request_send(fc, req);
1290 ret = req->out.h.error;
1291 if (!ret)
1292 ret = size ? req->out.args[0].size : outarg.size;
1293 else {
1294 if (ret == -ENOSYS) {
1295 fc->no_getxattr = 1;
1296 ret = -EOPNOTSUPP;
1299 fuse_put_request(fc, req);
1300 return ret;
1303 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1305 struct inode *inode = entry->d_inode;
1306 struct fuse_conn *fc = get_fuse_conn(inode);
1307 struct fuse_req *req;
1308 struct fuse_getxattr_in inarg;
1309 struct fuse_getxattr_out outarg;
1310 ssize_t ret;
1312 if (!fuse_allow_task(fc, current))
1313 return -EACCES;
1315 if (fc->no_listxattr)
1316 return -EOPNOTSUPP;
1318 req = fuse_get_req(fc);
1319 if (IS_ERR(req))
1320 return PTR_ERR(req);
1322 memset(&inarg, 0, sizeof(inarg));
1323 inarg.size = size;
1324 req->in.h.opcode = FUSE_LISTXATTR;
1325 req->in.h.nodeid = get_node_id(inode);
1326 req->in.numargs = 1;
1327 req->in.args[0].size = sizeof(inarg);
1328 req->in.args[0].value = &inarg;
1329 /* This is really two different operations rolled into one */
1330 req->out.numargs = 1;
1331 if (size) {
1332 req->out.argvar = 1;
1333 req->out.args[0].size = size;
1334 req->out.args[0].value = list;
1335 } else {
1336 req->out.args[0].size = sizeof(outarg);
1337 req->out.args[0].value = &outarg;
1339 request_send(fc, req);
1340 ret = req->out.h.error;
1341 if (!ret)
1342 ret = size ? req->out.args[0].size : outarg.size;
1343 else {
1344 if (ret == -ENOSYS) {
1345 fc->no_listxattr = 1;
1346 ret = -EOPNOTSUPP;
1349 fuse_put_request(fc, req);
1350 return ret;
1353 static int fuse_removexattr(struct dentry *entry, const char *name)
1355 struct inode *inode = entry->d_inode;
1356 struct fuse_conn *fc = get_fuse_conn(inode);
1357 struct fuse_req *req;
1358 int err;
1360 if (fc->no_removexattr)
1361 return -EOPNOTSUPP;
1363 req = fuse_get_req(fc);
1364 if (IS_ERR(req))
1365 return PTR_ERR(req);
1367 req->in.h.opcode = FUSE_REMOVEXATTR;
1368 req->in.h.nodeid = get_node_id(inode);
1369 req->in.numargs = 1;
1370 req->in.args[0].size = strlen(name) + 1;
1371 req->in.args[0].value = name;
1372 request_send(fc, req);
1373 err = req->out.h.error;
1374 fuse_put_request(fc, req);
1375 if (err == -ENOSYS) {
1376 fc->no_removexattr = 1;
1377 err = -EOPNOTSUPP;
1379 return err;
1382 static const struct inode_operations fuse_dir_inode_operations = {
1383 .lookup = fuse_lookup,
1384 .mkdir = fuse_mkdir,
1385 .symlink = fuse_symlink,
1386 .unlink = fuse_unlink,
1387 .rmdir = fuse_rmdir,
1388 .rename = fuse_rename,
1389 .link = fuse_link,
1390 .setattr = fuse_setattr,
1391 .create = fuse_create,
1392 .mknod = fuse_mknod,
1393 .permission = fuse_permission,
1394 .getattr = fuse_getattr,
1395 .setxattr = fuse_setxattr,
1396 .getxattr = fuse_getxattr,
1397 .listxattr = fuse_listxattr,
1398 .removexattr = fuse_removexattr,
1401 static const struct file_operations fuse_dir_operations = {
1402 .llseek = generic_file_llseek,
1403 .read = generic_read_dir,
1404 .readdir = fuse_readdir,
1405 .open = fuse_dir_open,
1406 .release = fuse_dir_release,
1407 .fsync = fuse_dir_fsync,
1410 static const struct inode_operations fuse_common_inode_operations = {
1411 .setattr = fuse_setattr,
1412 .permission = fuse_permission,
1413 .getattr = fuse_getattr,
1414 .setxattr = fuse_setxattr,
1415 .getxattr = fuse_getxattr,
1416 .listxattr = fuse_listxattr,
1417 .removexattr = fuse_removexattr,
1420 static const struct inode_operations fuse_symlink_inode_operations = {
1421 .setattr = fuse_setattr,
1422 .follow_link = fuse_follow_link,
1423 .put_link = fuse_put_link,
1424 .readlink = generic_readlink,
1425 .getattr = fuse_getattr,
1426 .setxattr = fuse_setxattr,
1427 .getxattr = fuse_getxattr,
1428 .listxattr = fuse_listxattr,
1429 .removexattr = fuse_removexattr,
1432 void fuse_init_common(struct inode *inode)
1434 inode->i_op = &fuse_common_inode_operations;
1437 void fuse_init_dir(struct inode *inode)
1439 inode->i_op = &fuse_dir_inode_operations;
1440 inode->i_fop = &fuse_dir_operations;
1443 void fuse_init_symlink(struct inode *inode)
1445 inode->i_op = &fuse_symlink_inode_operations;