ocfs2: fix locking for res->tracking and dlm->tracking_list
[linux/fpc-iii.git] / fs / fuse / dir.c
blob49b7b40f75987119b772ed50be529bc27caef2ba
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 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/sched.h>
14 #include <linux/namei.h>
15 #include <linux/slab.h>
17 static bool fuse_use_readdirplus(struct inode *dir, struct dir_context *ctx)
19 struct fuse_conn *fc = get_fuse_conn(dir);
20 struct fuse_inode *fi = get_fuse_inode(dir);
22 if (!fc->do_readdirplus)
23 return false;
24 if (!fc->readdirplus_auto)
25 return true;
26 if (test_and_clear_bit(FUSE_I_ADVISE_RDPLUS, &fi->state))
27 return true;
28 if (ctx->pos == 0)
29 return true;
30 return false;
33 static void fuse_advise_use_readdirplus(struct inode *dir)
35 struct fuse_inode *fi = get_fuse_inode(dir);
37 set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
40 #if BITS_PER_LONG >= 64
41 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
43 entry->d_time = time;
46 static inline u64 fuse_dentry_time(struct dentry *entry)
48 return entry->d_time;
50 #else
52 * On 32 bit archs store the high 32 bits of time in d_fsdata
54 static void fuse_dentry_settime(struct dentry *entry, u64 time)
56 entry->d_time = time;
57 entry->d_fsdata = (void *) (unsigned long) (time >> 32);
60 static u64 fuse_dentry_time(struct dentry *entry)
62 return (u64) entry->d_time +
63 ((u64) (unsigned long) entry->d_fsdata << 32);
65 #endif
68 * FUSE caches dentries and attributes with separate timeout. The
69 * time in jiffies until the dentry/attributes are valid is stored in
70 * dentry->d_time and fuse_inode->i_time respectively.
74 * Calculate the time in jiffies until a dentry/attributes are valid
76 static u64 time_to_jiffies(unsigned long sec, unsigned long nsec)
78 if (sec || nsec) {
79 struct timespec ts = {sec, nsec};
80 return get_jiffies_64() + timespec_to_jiffies(&ts);
81 } else
82 return 0;
86 * Set dentry and possibly attribute timeouts from the lookup/mk*
87 * replies
89 static void fuse_change_entry_timeout(struct dentry *entry,
90 struct fuse_entry_out *o)
92 fuse_dentry_settime(entry,
93 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
96 static u64 attr_timeout(struct fuse_attr_out *o)
98 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
101 static u64 entry_attr_timeout(struct fuse_entry_out *o)
103 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
107 * Mark the attributes as stale, so that at the next call to
108 * ->getattr() they will be fetched from userspace
110 void fuse_invalidate_attr(struct inode *inode)
112 get_fuse_inode(inode)->i_time = 0;
116 * Mark the attributes as stale due to an atime change. Avoid the invalidate if
117 * atime is not used.
119 void fuse_invalidate_atime(struct inode *inode)
121 if (!IS_RDONLY(inode))
122 fuse_invalidate_attr(inode);
126 * Just mark the entry as stale, so that a next attempt to look it up
127 * will result in a new lookup call to userspace
129 * This is called when a dentry is about to become negative and the
130 * timeout is unknown (unlink, rmdir, rename and in some cases
131 * lookup)
133 void fuse_invalidate_entry_cache(struct dentry *entry)
135 fuse_dentry_settime(entry, 0);
139 * Same as fuse_invalidate_entry_cache(), but also try to remove the
140 * dentry from the hash
142 static void fuse_invalidate_entry(struct dentry *entry)
144 d_invalidate(entry);
145 fuse_invalidate_entry_cache(entry);
148 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
149 u64 nodeid, struct qstr *name,
150 struct fuse_entry_out *outarg)
152 memset(outarg, 0, sizeof(struct fuse_entry_out));
153 args->in.h.opcode = FUSE_LOOKUP;
154 args->in.h.nodeid = nodeid;
155 args->in.numargs = 1;
156 args->in.args[0].size = name->len + 1;
157 args->in.args[0].value = name->name;
158 args->out.numargs = 1;
159 args->out.args[0].size = sizeof(struct fuse_entry_out);
160 args->out.args[0].value = outarg;
163 u64 fuse_get_attr_version(struct fuse_conn *fc)
165 u64 curr_version;
168 * The spin lock isn't actually needed on 64bit archs, but we
169 * don't yet care too much about such optimizations.
171 spin_lock(&fc->lock);
172 curr_version = fc->attr_version;
173 spin_unlock(&fc->lock);
175 return curr_version;
179 * Check whether the dentry is still valid
181 * If the entry validity timeout has expired and the dentry is
182 * positive, try to redo the lookup. If the lookup results in a
183 * different inode, then let the VFS invalidate the dentry and redo
184 * the lookup once more. If the lookup results in the same inode,
185 * then refresh the attributes, timeouts and mark the dentry valid.
187 static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
189 struct inode *inode;
190 struct dentry *parent;
191 struct fuse_conn *fc;
192 struct fuse_inode *fi;
193 int ret;
195 inode = d_inode_rcu(entry);
196 if (inode && is_bad_inode(inode))
197 goto invalid;
198 else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
199 (flags & LOOKUP_REVAL)) {
200 struct fuse_entry_out outarg;
201 FUSE_ARGS(args);
202 struct fuse_forget_link *forget;
203 u64 attr_version;
205 /* For negative dentries, always do a fresh lookup */
206 if (!inode)
207 goto invalid;
209 ret = -ECHILD;
210 if (flags & LOOKUP_RCU)
211 goto out;
213 fc = get_fuse_conn(inode);
215 forget = fuse_alloc_forget();
216 ret = -ENOMEM;
217 if (!forget)
218 goto out;
220 attr_version = fuse_get_attr_version(fc);
222 parent = dget_parent(entry);
223 fuse_lookup_init(fc, &args, get_node_id(d_inode(parent)),
224 &entry->d_name, &outarg);
225 ret = fuse_simple_request(fc, &args);
226 dput(parent);
227 /* Zero nodeid is same as -ENOENT */
228 if (!ret && !outarg.nodeid)
229 ret = -ENOENT;
230 if (!ret) {
231 fi = get_fuse_inode(inode);
232 if (outarg.nodeid != get_node_id(inode)) {
233 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
234 goto invalid;
236 spin_lock(&fc->lock);
237 fi->nlookup++;
238 spin_unlock(&fc->lock);
240 kfree(forget);
241 if (ret == -ENOMEM)
242 goto out;
243 if (ret || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
244 goto invalid;
246 fuse_change_attributes(inode, &outarg.attr,
247 entry_attr_timeout(&outarg),
248 attr_version);
249 fuse_change_entry_timeout(entry, &outarg);
250 } else if (inode) {
251 fi = get_fuse_inode(inode);
252 if (flags & LOOKUP_RCU) {
253 if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
254 return -ECHILD;
255 } else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
256 parent = dget_parent(entry);
257 fuse_advise_use_readdirplus(d_inode(parent));
258 dput(parent);
261 ret = 1;
262 out:
263 return ret;
265 invalid:
266 ret = 0;
267 goto out;
270 static int invalid_nodeid(u64 nodeid)
272 return !nodeid || nodeid == FUSE_ROOT_ID;
275 const struct dentry_operations fuse_dentry_operations = {
276 .d_revalidate = fuse_dentry_revalidate,
279 int fuse_valid_type(int m)
281 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
282 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
285 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
286 struct fuse_entry_out *outarg, struct inode **inode)
288 struct fuse_conn *fc = get_fuse_conn_super(sb);
289 FUSE_ARGS(args);
290 struct fuse_forget_link *forget;
291 u64 attr_version;
292 int err;
294 *inode = NULL;
295 err = -ENAMETOOLONG;
296 if (name->len > FUSE_NAME_MAX)
297 goto out;
300 forget = fuse_alloc_forget();
301 err = -ENOMEM;
302 if (!forget)
303 goto out;
305 attr_version = fuse_get_attr_version(fc);
307 fuse_lookup_init(fc, &args, nodeid, name, outarg);
308 err = fuse_simple_request(fc, &args);
309 /* Zero nodeid is same as -ENOENT, but with valid timeout */
310 if (err || !outarg->nodeid)
311 goto out_put_forget;
313 err = -EIO;
314 if (!outarg->nodeid)
315 goto out_put_forget;
316 if (!fuse_valid_type(outarg->attr.mode))
317 goto out_put_forget;
319 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
320 &outarg->attr, entry_attr_timeout(outarg),
321 attr_version);
322 err = -ENOMEM;
323 if (!*inode) {
324 fuse_queue_forget(fc, forget, outarg->nodeid, 1);
325 goto out;
327 err = 0;
329 out_put_forget:
330 kfree(forget);
331 out:
332 return err;
335 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
336 unsigned int flags)
338 int err;
339 struct fuse_entry_out outarg;
340 struct inode *inode;
341 struct dentry *newent;
342 bool outarg_valid = true;
344 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
345 &outarg, &inode);
346 if (err == -ENOENT) {
347 outarg_valid = false;
348 err = 0;
350 if (err)
351 goto out_err;
353 err = -EIO;
354 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
355 goto out_iput;
357 newent = d_splice_alias(inode, entry);
358 err = PTR_ERR(newent);
359 if (IS_ERR(newent))
360 goto out_err;
362 entry = newent ? newent : entry;
363 if (outarg_valid)
364 fuse_change_entry_timeout(entry, &outarg);
365 else
366 fuse_invalidate_entry_cache(entry);
368 fuse_advise_use_readdirplus(dir);
369 return newent;
371 out_iput:
372 iput(inode);
373 out_err:
374 return ERR_PTR(err);
378 * Atomic create+open operation
380 * If the filesystem doesn't support this, then fall back to separate
381 * 'mknod' + 'open' requests.
383 static int fuse_create_open(struct inode *dir, struct dentry *entry,
384 struct file *file, unsigned flags,
385 umode_t mode, int *opened)
387 int err;
388 struct inode *inode;
389 struct fuse_conn *fc = get_fuse_conn(dir);
390 FUSE_ARGS(args);
391 struct fuse_forget_link *forget;
392 struct fuse_create_in inarg;
393 struct fuse_open_out outopen;
394 struct fuse_entry_out outentry;
395 struct fuse_file *ff;
397 /* Userspace expects S_IFREG in create mode */
398 BUG_ON((mode & S_IFMT) != S_IFREG);
400 forget = fuse_alloc_forget();
401 err = -ENOMEM;
402 if (!forget)
403 goto out_err;
405 err = -ENOMEM;
406 ff = fuse_file_alloc(fc);
407 if (!ff)
408 goto out_put_forget_req;
410 if (!fc->dont_mask)
411 mode &= ~current_umask();
413 flags &= ~O_NOCTTY;
414 memset(&inarg, 0, sizeof(inarg));
415 memset(&outentry, 0, sizeof(outentry));
416 inarg.flags = flags;
417 inarg.mode = mode;
418 inarg.umask = current_umask();
419 args.in.h.opcode = FUSE_CREATE;
420 args.in.h.nodeid = get_node_id(dir);
421 args.in.numargs = 2;
422 args.in.args[0].size = sizeof(inarg);
423 args.in.args[0].value = &inarg;
424 args.in.args[1].size = entry->d_name.len + 1;
425 args.in.args[1].value = entry->d_name.name;
426 args.out.numargs = 2;
427 args.out.args[0].size = sizeof(outentry);
428 args.out.args[0].value = &outentry;
429 args.out.args[1].size = sizeof(outopen);
430 args.out.args[1].value = &outopen;
431 err = fuse_simple_request(fc, &args);
432 if (err)
433 goto out_free_ff;
435 err = -EIO;
436 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
437 goto out_free_ff;
439 ff->fh = outopen.fh;
440 ff->nodeid = outentry.nodeid;
441 ff->open_flags = outopen.open_flags;
442 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
443 &outentry.attr, entry_attr_timeout(&outentry), 0);
444 if (!inode) {
445 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
446 fuse_sync_release(ff, flags);
447 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
448 err = -ENOMEM;
449 goto out_err;
451 kfree(forget);
452 d_instantiate(entry, inode);
453 fuse_change_entry_timeout(entry, &outentry);
454 fuse_invalidate_attr(dir);
455 err = finish_open(file, entry, generic_file_open, opened);
456 if (err) {
457 fuse_sync_release(ff, flags);
458 } else {
459 file->private_data = fuse_file_get(ff);
460 fuse_finish_open(inode, file);
462 return err;
464 out_free_ff:
465 fuse_file_free(ff);
466 out_put_forget_req:
467 kfree(forget);
468 out_err:
469 return err;
472 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
473 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
474 struct file *file, unsigned flags,
475 umode_t mode, int *opened)
477 int err;
478 struct fuse_conn *fc = get_fuse_conn(dir);
479 struct dentry *res = NULL;
481 if (d_unhashed(entry)) {
482 res = fuse_lookup(dir, entry, 0);
483 if (IS_ERR(res))
484 return PTR_ERR(res);
486 if (res)
487 entry = res;
490 if (!(flags & O_CREAT) || d_really_is_positive(entry))
491 goto no_open;
493 /* Only creates */
494 *opened |= FILE_CREATED;
496 if (fc->no_create)
497 goto mknod;
499 err = fuse_create_open(dir, entry, file, flags, mode, opened);
500 if (err == -ENOSYS) {
501 fc->no_create = 1;
502 goto mknod;
504 out_dput:
505 dput(res);
506 return err;
508 mknod:
509 err = fuse_mknod(dir, entry, mode, 0);
510 if (err)
511 goto out_dput;
512 no_open:
513 return finish_no_open(file, res);
517 * Code shared between mknod, mkdir, symlink and link
519 static int create_new_entry(struct fuse_conn *fc, struct fuse_args *args,
520 struct inode *dir, struct dentry *entry,
521 umode_t mode)
523 struct fuse_entry_out outarg;
524 struct inode *inode;
525 int err;
526 struct fuse_forget_link *forget;
528 forget = fuse_alloc_forget();
529 if (!forget)
530 return -ENOMEM;
532 memset(&outarg, 0, sizeof(outarg));
533 args->in.h.nodeid = get_node_id(dir);
534 args->out.numargs = 1;
535 args->out.args[0].size = sizeof(outarg);
536 args->out.args[0].value = &outarg;
537 err = fuse_simple_request(fc, args);
538 if (err)
539 goto out_put_forget_req;
541 err = -EIO;
542 if (invalid_nodeid(outarg.nodeid))
543 goto out_put_forget_req;
545 if ((outarg.attr.mode ^ mode) & S_IFMT)
546 goto out_put_forget_req;
548 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
549 &outarg.attr, entry_attr_timeout(&outarg), 0);
550 if (!inode) {
551 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
552 return -ENOMEM;
554 kfree(forget);
556 err = d_instantiate_no_diralias(entry, inode);
557 if (err)
558 return err;
560 fuse_change_entry_timeout(entry, &outarg);
561 fuse_invalidate_attr(dir);
562 return 0;
564 out_put_forget_req:
565 kfree(forget);
566 return err;
569 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
570 dev_t rdev)
572 struct fuse_mknod_in inarg;
573 struct fuse_conn *fc = get_fuse_conn(dir);
574 FUSE_ARGS(args);
576 if (!fc->dont_mask)
577 mode &= ~current_umask();
579 memset(&inarg, 0, sizeof(inarg));
580 inarg.mode = mode;
581 inarg.rdev = new_encode_dev(rdev);
582 inarg.umask = current_umask();
583 args.in.h.opcode = FUSE_MKNOD;
584 args.in.numargs = 2;
585 args.in.args[0].size = sizeof(inarg);
586 args.in.args[0].value = &inarg;
587 args.in.args[1].size = entry->d_name.len + 1;
588 args.in.args[1].value = entry->d_name.name;
589 return create_new_entry(fc, &args, dir, entry, mode);
592 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
593 bool excl)
595 return fuse_mknod(dir, entry, mode, 0);
598 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
600 struct fuse_mkdir_in inarg;
601 struct fuse_conn *fc = get_fuse_conn(dir);
602 FUSE_ARGS(args);
604 if (!fc->dont_mask)
605 mode &= ~current_umask();
607 memset(&inarg, 0, sizeof(inarg));
608 inarg.mode = mode;
609 inarg.umask = current_umask();
610 args.in.h.opcode = FUSE_MKDIR;
611 args.in.numargs = 2;
612 args.in.args[0].size = sizeof(inarg);
613 args.in.args[0].value = &inarg;
614 args.in.args[1].size = entry->d_name.len + 1;
615 args.in.args[1].value = entry->d_name.name;
616 return create_new_entry(fc, &args, dir, entry, S_IFDIR);
619 static int fuse_symlink(struct inode *dir, struct dentry *entry,
620 const char *link)
622 struct fuse_conn *fc = get_fuse_conn(dir);
623 unsigned len = strlen(link) + 1;
624 FUSE_ARGS(args);
626 args.in.h.opcode = FUSE_SYMLINK;
627 args.in.numargs = 2;
628 args.in.args[0].size = entry->d_name.len + 1;
629 args.in.args[0].value = entry->d_name.name;
630 args.in.args[1].size = len;
631 args.in.args[1].value = link;
632 return create_new_entry(fc, &args, dir, entry, S_IFLNK);
635 static inline void fuse_update_ctime(struct inode *inode)
637 if (!IS_NOCMTIME(inode)) {
638 inode->i_ctime = current_fs_time(inode->i_sb);
639 mark_inode_dirty_sync(inode);
643 static int fuse_unlink(struct inode *dir, struct dentry *entry)
645 int err;
646 struct fuse_conn *fc = get_fuse_conn(dir);
647 FUSE_ARGS(args);
649 args.in.h.opcode = FUSE_UNLINK;
650 args.in.h.nodeid = get_node_id(dir);
651 args.in.numargs = 1;
652 args.in.args[0].size = entry->d_name.len + 1;
653 args.in.args[0].value = entry->d_name.name;
654 err = fuse_simple_request(fc, &args);
655 if (!err) {
656 struct inode *inode = d_inode(entry);
657 struct fuse_inode *fi = get_fuse_inode(inode);
659 spin_lock(&fc->lock);
660 fi->attr_version = ++fc->attr_version;
662 * If i_nlink == 0 then unlink doesn't make sense, yet this can
663 * happen if userspace filesystem is careless. It would be
664 * difficult to enforce correct nlink usage so just ignore this
665 * condition here
667 if (inode->i_nlink > 0)
668 drop_nlink(inode);
669 spin_unlock(&fc->lock);
670 fuse_invalidate_attr(inode);
671 fuse_invalidate_attr(dir);
672 fuse_invalidate_entry_cache(entry);
673 fuse_update_ctime(inode);
674 } else if (err == -EINTR)
675 fuse_invalidate_entry(entry);
676 return err;
679 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
681 int err;
682 struct fuse_conn *fc = get_fuse_conn(dir);
683 FUSE_ARGS(args);
685 args.in.h.opcode = FUSE_RMDIR;
686 args.in.h.nodeid = get_node_id(dir);
687 args.in.numargs = 1;
688 args.in.args[0].size = entry->d_name.len + 1;
689 args.in.args[0].value = entry->d_name.name;
690 err = fuse_simple_request(fc, &args);
691 if (!err) {
692 clear_nlink(d_inode(entry));
693 fuse_invalidate_attr(dir);
694 fuse_invalidate_entry_cache(entry);
695 } else if (err == -EINTR)
696 fuse_invalidate_entry(entry);
697 return err;
700 static int fuse_rename_common(struct inode *olddir, struct dentry *oldent,
701 struct inode *newdir, struct dentry *newent,
702 unsigned int flags, int opcode, size_t argsize)
704 int err;
705 struct fuse_rename2_in inarg;
706 struct fuse_conn *fc = get_fuse_conn(olddir);
707 FUSE_ARGS(args);
709 memset(&inarg, 0, argsize);
710 inarg.newdir = get_node_id(newdir);
711 inarg.flags = flags;
712 args.in.h.opcode = opcode;
713 args.in.h.nodeid = get_node_id(olddir);
714 args.in.numargs = 3;
715 args.in.args[0].size = argsize;
716 args.in.args[0].value = &inarg;
717 args.in.args[1].size = oldent->d_name.len + 1;
718 args.in.args[1].value = oldent->d_name.name;
719 args.in.args[2].size = newent->d_name.len + 1;
720 args.in.args[2].value = newent->d_name.name;
721 err = fuse_simple_request(fc, &args);
722 if (!err) {
723 /* ctime changes */
724 fuse_invalidate_attr(d_inode(oldent));
725 fuse_update_ctime(d_inode(oldent));
727 if (flags & RENAME_EXCHANGE) {
728 fuse_invalidate_attr(d_inode(newent));
729 fuse_update_ctime(d_inode(newent));
732 fuse_invalidate_attr(olddir);
733 if (olddir != newdir)
734 fuse_invalidate_attr(newdir);
736 /* newent will end up negative */
737 if (!(flags & RENAME_EXCHANGE) && d_really_is_positive(newent)) {
738 fuse_invalidate_attr(d_inode(newent));
739 fuse_invalidate_entry_cache(newent);
740 fuse_update_ctime(d_inode(newent));
742 } else if (err == -EINTR) {
743 /* If request was interrupted, DEITY only knows if the
744 rename actually took place. If the invalidation
745 fails (e.g. some process has CWD under the renamed
746 directory), then there can be inconsistency between
747 the dcache and the real filesystem. Tough luck. */
748 fuse_invalidate_entry(oldent);
749 if (d_really_is_positive(newent))
750 fuse_invalidate_entry(newent);
753 return err;
756 static int fuse_rename2(struct inode *olddir, struct dentry *oldent,
757 struct inode *newdir, struct dentry *newent,
758 unsigned int flags)
760 struct fuse_conn *fc = get_fuse_conn(olddir);
761 int err;
763 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
764 return -EINVAL;
766 if (flags) {
767 if (fc->no_rename2 || fc->minor < 23)
768 return -EINVAL;
770 err = fuse_rename_common(olddir, oldent, newdir, newent, flags,
771 FUSE_RENAME2,
772 sizeof(struct fuse_rename2_in));
773 if (err == -ENOSYS) {
774 fc->no_rename2 = 1;
775 err = -EINVAL;
777 } else {
778 err = fuse_rename_common(olddir, oldent, newdir, newent, 0,
779 FUSE_RENAME,
780 sizeof(struct fuse_rename_in));
783 return err;
786 static int fuse_link(struct dentry *entry, struct inode *newdir,
787 struct dentry *newent)
789 int err;
790 struct fuse_link_in inarg;
791 struct inode *inode = d_inode(entry);
792 struct fuse_conn *fc = get_fuse_conn(inode);
793 FUSE_ARGS(args);
795 memset(&inarg, 0, sizeof(inarg));
796 inarg.oldnodeid = get_node_id(inode);
797 args.in.h.opcode = FUSE_LINK;
798 args.in.numargs = 2;
799 args.in.args[0].size = sizeof(inarg);
800 args.in.args[0].value = &inarg;
801 args.in.args[1].size = newent->d_name.len + 1;
802 args.in.args[1].value = newent->d_name.name;
803 err = create_new_entry(fc, &args, newdir, newent, inode->i_mode);
804 /* Contrary to "normal" filesystems it can happen that link
805 makes two "logical" inodes point to the same "physical"
806 inode. We invalidate the attributes of the old one, so it
807 will reflect changes in the backing inode (link count,
808 etc.)
810 if (!err) {
811 struct fuse_inode *fi = get_fuse_inode(inode);
813 spin_lock(&fc->lock);
814 fi->attr_version = ++fc->attr_version;
815 inc_nlink(inode);
816 spin_unlock(&fc->lock);
817 fuse_invalidate_attr(inode);
818 fuse_update_ctime(inode);
819 } else if (err == -EINTR) {
820 fuse_invalidate_attr(inode);
822 return err;
825 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
826 struct kstat *stat)
828 unsigned int blkbits;
829 struct fuse_conn *fc = get_fuse_conn(inode);
831 /* see the comment in fuse_change_attributes() */
832 if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
833 attr->size = i_size_read(inode);
834 attr->mtime = inode->i_mtime.tv_sec;
835 attr->mtimensec = inode->i_mtime.tv_nsec;
836 attr->ctime = inode->i_ctime.tv_sec;
837 attr->ctimensec = inode->i_ctime.tv_nsec;
840 stat->dev = inode->i_sb->s_dev;
841 stat->ino = attr->ino;
842 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
843 stat->nlink = attr->nlink;
844 stat->uid = make_kuid(&init_user_ns, attr->uid);
845 stat->gid = make_kgid(&init_user_ns, attr->gid);
846 stat->rdev = inode->i_rdev;
847 stat->atime.tv_sec = attr->atime;
848 stat->atime.tv_nsec = attr->atimensec;
849 stat->mtime.tv_sec = attr->mtime;
850 stat->mtime.tv_nsec = attr->mtimensec;
851 stat->ctime.tv_sec = attr->ctime;
852 stat->ctime.tv_nsec = attr->ctimensec;
853 stat->size = attr->size;
854 stat->blocks = attr->blocks;
856 if (attr->blksize != 0)
857 blkbits = ilog2(attr->blksize);
858 else
859 blkbits = inode->i_sb->s_blocksize_bits;
861 stat->blksize = 1 << blkbits;
864 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
865 struct file *file)
867 int err;
868 struct fuse_getattr_in inarg;
869 struct fuse_attr_out outarg;
870 struct fuse_conn *fc = get_fuse_conn(inode);
871 FUSE_ARGS(args);
872 u64 attr_version;
874 attr_version = fuse_get_attr_version(fc);
876 memset(&inarg, 0, sizeof(inarg));
877 memset(&outarg, 0, sizeof(outarg));
878 /* Directories have separate file-handle space */
879 if (file && S_ISREG(inode->i_mode)) {
880 struct fuse_file *ff = file->private_data;
882 inarg.getattr_flags |= FUSE_GETATTR_FH;
883 inarg.fh = ff->fh;
885 args.in.h.opcode = FUSE_GETATTR;
886 args.in.h.nodeid = get_node_id(inode);
887 args.in.numargs = 1;
888 args.in.args[0].size = sizeof(inarg);
889 args.in.args[0].value = &inarg;
890 args.out.numargs = 1;
891 args.out.args[0].size = sizeof(outarg);
892 args.out.args[0].value = &outarg;
893 err = fuse_simple_request(fc, &args);
894 if (!err) {
895 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
896 make_bad_inode(inode);
897 err = -EIO;
898 } else {
899 fuse_change_attributes(inode, &outarg.attr,
900 attr_timeout(&outarg),
901 attr_version);
902 if (stat)
903 fuse_fillattr(inode, &outarg.attr, stat);
906 return err;
909 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
910 struct file *file, bool *refreshed)
912 struct fuse_inode *fi = get_fuse_inode(inode);
913 int err;
914 bool r;
916 if (time_before64(fi->i_time, get_jiffies_64())) {
917 r = true;
918 err = fuse_do_getattr(inode, stat, file);
919 } else {
920 r = false;
921 err = 0;
922 if (stat) {
923 generic_fillattr(inode, stat);
924 stat->mode = fi->orig_i_mode;
925 stat->ino = fi->orig_ino;
929 if (refreshed != NULL)
930 *refreshed = r;
932 return err;
935 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
936 u64 child_nodeid, struct qstr *name)
938 int err = -ENOTDIR;
939 struct inode *parent;
940 struct dentry *dir;
941 struct dentry *entry;
943 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
944 if (!parent)
945 return -ENOENT;
947 mutex_lock(&parent->i_mutex);
948 if (!S_ISDIR(parent->i_mode))
949 goto unlock;
951 err = -ENOENT;
952 dir = d_find_alias(parent);
953 if (!dir)
954 goto unlock;
956 entry = d_lookup(dir, name);
957 dput(dir);
958 if (!entry)
959 goto unlock;
961 fuse_invalidate_attr(parent);
962 fuse_invalidate_entry(entry);
964 if (child_nodeid != 0 && d_really_is_positive(entry)) {
965 mutex_lock(&d_inode(entry)->i_mutex);
966 if (get_node_id(d_inode(entry)) != child_nodeid) {
967 err = -ENOENT;
968 goto badentry;
970 if (d_mountpoint(entry)) {
971 err = -EBUSY;
972 goto badentry;
974 if (d_is_dir(entry)) {
975 shrink_dcache_parent(entry);
976 if (!simple_empty(entry)) {
977 err = -ENOTEMPTY;
978 goto badentry;
980 d_inode(entry)->i_flags |= S_DEAD;
982 dont_mount(entry);
983 clear_nlink(d_inode(entry));
984 err = 0;
985 badentry:
986 mutex_unlock(&d_inode(entry)->i_mutex);
987 if (!err)
988 d_delete(entry);
989 } else {
990 err = 0;
992 dput(entry);
994 unlock:
995 mutex_unlock(&parent->i_mutex);
996 iput(parent);
997 return err;
1001 * Calling into a user-controlled filesystem gives the filesystem
1002 * daemon ptrace-like capabilities over the current process. This
1003 * means, that the filesystem daemon is able to record the exact
1004 * filesystem operations performed, and can also control the behavior
1005 * of the requester process in otherwise impossible ways. For example
1006 * it can delay the operation for arbitrary length of time allowing
1007 * DoS against the requester.
1009 * For this reason only those processes can call into the filesystem,
1010 * for which the owner of the mount has ptrace privilege. This
1011 * excludes processes started by other users, suid or sgid processes.
1013 int fuse_allow_current_process(struct fuse_conn *fc)
1015 const struct cred *cred;
1017 if (fc->flags & FUSE_ALLOW_OTHER)
1018 return 1;
1020 cred = current_cred();
1021 if (uid_eq(cred->euid, fc->user_id) &&
1022 uid_eq(cred->suid, fc->user_id) &&
1023 uid_eq(cred->uid, fc->user_id) &&
1024 gid_eq(cred->egid, fc->group_id) &&
1025 gid_eq(cred->sgid, fc->group_id) &&
1026 gid_eq(cred->gid, fc->group_id))
1027 return 1;
1029 return 0;
1032 static int fuse_access(struct inode *inode, int mask)
1034 struct fuse_conn *fc = get_fuse_conn(inode);
1035 FUSE_ARGS(args);
1036 struct fuse_access_in inarg;
1037 int err;
1039 BUG_ON(mask & MAY_NOT_BLOCK);
1041 if (fc->no_access)
1042 return 0;
1044 memset(&inarg, 0, sizeof(inarg));
1045 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1046 args.in.h.opcode = FUSE_ACCESS;
1047 args.in.h.nodeid = get_node_id(inode);
1048 args.in.numargs = 1;
1049 args.in.args[0].size = sizeof(inarg);
1050 args.in.args[0].value = &inarg;
1051 err = fuse_simple_request(fc, &args);
1052 if (err == -ENOSYS) {
1053 fc->no_access = 1;
1054 err = 0;
1056 return err;
1059 static int fuse_perm_getattr(struct inode *inode, int mask)
1061 if (mask & MAY_NOT_BLOCK)
1062 return -ECHILD;
1064 return fuse_do_getattr(inode, NULL, NULL);
1068 * Check permission. The two basic access models of FUSE are:
1070 * 1) Local access checking ('default_permissions' mount option) based
1071 * on file mode. This is the plain old disk filesystem permission
1072 * modell.
1074 * 2) "Remote" access checking, where server is responsible for
1075 * checking permission in each inode operation. An exception to this
1076 * is if ->permission() was invoked from sys_access() in which case an
1077 * access request is sent. Execute permission is still checked
1078 * locally based on file mode.
1080 static int fuse_permission(struct inode *inode, int mask)
1082 struct fuse_conn *fc = get_fuse_conn(inode);
1083 bool refreshed = false;
1084 int err = 0;
1086 if (!fuse_allow_current_process(fc))
1087 return -EACCES;
1090 * If attributes are needed, refresh them before proceeding
1092 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1093 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1094 struct fuse_inode *fi = get_fuse_inode(inode);
1096 if (time_before64(fi->i_time, get_jiffies_64())) {
1097 refreshed = true;
1099 err = fuse_perm_getattr(inode, mask);
1100 if (err)
1101 return err;
1105 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1106 err = generic_permission(inode, mask);
1108 /* If permission is denied, try to refresh file
1109 attributes. This is also needed, because the root
1110 node will at first have no permissions */
1111 if (err == -EACCES && !refreshed) {
1112 err = fuse_perm_getattr(inode, mask);
1113 if (!err)
1114 err = generic_permission(inode, mask);
1117 /* Note: the opposite of the above test does not
1118 exist. So if permissions are revoked this won't be
1119 noticed immediately, only after the attribute
1120 timeout has expired */
1121 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1122 err = fuse_access(inode, mask);
1123 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1124 if (!(inode->i_mode & S_IXUGO)) {
1125 if (refreshed)
1126 return -EACCES;
1128 err = fuse_perm_getattr(inode, mask);
1129 if (!err && !(inode->i_mode & S_IXUGO))
1130 return -EACCES;
1133 return err;
1136 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1137 struct dir_context *ctx)
1139 while (nbytes >= FUSE_NAME_OFFSET) {
1140 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1141 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1142 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1143 return -EIO;
1144 if (reclen > nbytes)
1145 break;
1146 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1147 return -EIO;
1149 if (!dir_emit(ctx, dirent->name, dirent->namelen,
1150 dirent->ino, dirent->type))
1151 break;
1153 buf += reclen;
1154 nbytes -= reclen;
1155 ctx->pos = dirent->off;
1158 return 0;
1161 static int fuse_direntplus_link(struct file *file,
1162 struct fuse_direntplus *direntplus,
1163 u64 attr_version)
1165 int err;
1166 struct fuse_entry_out *o = &direntplus->entry_out;
1167 struct fuse_dirent *dirent = &direntplus->dirent;
1168 struct dentry *parent = file->f_path.dentry;
1169 struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
1170 struct dentry *dentry;
1171 struct dentry *alias;
1172 struct inode *dir = d_inode(parent);
1173 struct fuse_conn *fc;
1174 struct inode *inode;
1176 if (!o->nodeid) {
1178 * Unlike in the case of fuse_lookup, zero nodeid does not mean
1179 * ENOENT. Instead, it only means the userspace filesystem did
1180 * not want to return attributes/handle for this entry.
1182 * So do nothing.
1184 return 0;
1187 if (name.name[0] == '.') {
1189 * We could potentially refresh the attributes of the directory
1190 * and its parent?
1192 if (name.len == 1)
1193 return 0;
1194 if (name.name[1] == '.' && name.len == 2)
1195 return 0;
1198 if (invalid_nodeid(o->nodeid))
1199 return -EIO;
1200 if (!fuse_valid_type(o->attr.mode))
1201 return -EIO;
1203 fc = get_fuse_conn(dir);
1205 name.hash = full_name_hash(name.name, name.len);
1206 dentry = d_lookup(parent, &name);
1207 if (dentry) {
1208 inode = d_inode(dentry);
1209 if (!inode) {
1210 d_drop(dentry);
1211 } else if (get_node_id(inode) != o->nodeid ||
1212 ((o->attr.mode ^ inode->i_mode) & S_IFMT)) {
1213 d_invalidate(dentry);
1214 } else if (is_bad_inode(inode)) {
1215 err = -EIO;
1216 goto out;
1217 } else {
1218 struct fuse_inode *fi;
1219 fi = get_fuse_inode(inode);
1220 spin_lock(&fc->lock);
1221 fi->nlookup++;
1222 spin_unlock(&fc->lock);
1224 fuse_change_attributes(inode, &o->attr,
1225 entry_attr_timeout(o),
1226 attr_version);
1229 * The other branch to 'found' comes via fuse_iget()
1230 * which bumps nlookup inside
1232 goto found;
1234 dput(dentry);
1237 dentry = d_alloc(parent, &name);
1238 err = -ENOMEM;
1239 if (!dentry)
1240 goto out;
1242 inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
1243 &o->attr, entry_attr_timeout(o), attr_version);
1244 if (!inode)
1245 goto out;
1247 alias = d_splice_alias(inode, dentry);
1248 err = PTR_ERR(alias);
1249 if (IS_ERR(alias))
1250 goto out;
1252 if (alias) {
1253 dput(dentry);
1254 dentry = alias;
1257 found:
1258 if (fc->readdirplus_auto)
1259 set_bit(FUSE_I_INIT_RDPLUS, &get_fuse_inode(inode)->state);
1260 fuse_change_entry_timeout(dentry, o);
1262 err = 0;
1263 out:
1264 dput(dentry);
1265 return err;
1268 static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
1269 struct dir_context *ctx, u64 attr_version)
1271 struct fuse_direntplus *direntplus;
1272 struct fuse_dirent *dirent;
1273 size_t reclen;
1274 int over = 0;
1275 int ret;
1277 while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
1278 direntplus = (struct fuse_direntplus *) buf;
1279 dirent = &direntplus->dirent;
1280 reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
1282 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1283 return -EIO;
1284 if (reclen > nbytes)
1285 break;
1286 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1287 return -EIO;
1289 if (!over) {
1290 /* We fill entries into dstbuf only as much as
1291 it can hold. But we still continue iterating
1292 over remaining entries to link them. If not,
1293 we need to send a FORGET for each of those
1294 which we did not link.
1296 over = !dir_emit(ctx, dirent->name, dirent->namelen,
1297 dirent->ino, dirent->type);
1298 if (!over)
1299 ctx->pos = dirent->off;
1302 buf += reclen;
1303 nbytes -= reclen;
1305 ret = fuse_direntplus_link(file, direntplus, attr_version);
1306 if (ret)
1307 fuse_force_forget(file, direntplus->entry_out.nodeid);
1310 return 0;
1313 static int fuse_readdir(struct file *file, struct dir_context *ctx)
1315 int plus, err;
1316 size_t nbytes;
1317 struct page *page;
1318 struct inode *inode = file_inode(file);
1319 struct fuse_conn *fc = get_fuse_conn(inode);
1320 struct fuse_req *req;
1321 u64 attr_version = 0;
1323 if (is_bad_inode(inode))
1324 return -EIO;
1326 req = fuse_get_req(fc, 1);
1327 if (IS_ERR(req))
1328 return PTR_ERR(req);
1330 page = alloc_page(GFP_KERNEL);
1331 if (!page) {
1332 fuse_put_request(fc, req);
1333 return -ENOMEM;
1336 plus = fuse_use_readdirplus(inode, ctx);
1337 req->out.argpages = 1;
1338 req->num_pages = 1;
1339 req->pages[0] = page;
1340 req->page_descs[0].length = PAGE_SIZE;
1341 if (plus) {
1342 attr_version = fuse_get_attr_version(fc);
1343 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1344 FUSE_READDIRPLUS);
1345 } else {
1346 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1347 FUSE_READDIR);
1349 fuse_request_send(fc, req);
1350 nbytes = req->out.args[0].size;
1351 err = req->out.h.error;
1352 fuse_put_request(fc, req);
1353 if (!err) {
1354 if (plus) {
1355 err = parse_dirplusfile(page_address(page), nbytes,
1356 file, ctx,
1357 attr_version);
1358 } else {
1359 err = parse_dirfile(page_address(page), nbytes, file,
1360 ctx);
1364 __free_page(page);
1365 fuse_invalidate_atime(inode);
1366 return err;
1369 static const char *fuse_follow_link(struct dentry *dentry, void **cookie)
1371 struct inode *inode = d_inode(dentry);
1372 struct fuse_conn *fc = get_fuse_conn(inode);
1373 FUSE_ARGS(args);
1374 char *link;
1375 ssize_t ret;
1377 link = (char *) __get_free_page(GFP_KERNEL);
1378 if (!link)
1379 return ERR_PTR(-ENOMEM);
1381 args.in.h.opcode = FUSE_READLINK;
1382 args.in.h.nodeid = get_node_id(inode);
1383 args.out.argvar = 1;
1384 args.out.numargs = 1;
1385 args.out.args[0].size = PAGE_SIZE - 1;
1386 args.out.args[0].value = link;
1387 ret = fuse_simple_request(fc, &args);
1388 if (ret < 0) {
1389 free_page((unsigned long) link);
1390 link = ERR_PTR(ret);
1391 } else {
1392 link[ret] = '\0';
1393 *cookie = link;
1395 fuse_invalidate_atime(inode);
1396 return link;
1399 static int fuse_dir_open(struct inode *inode, struct file *file)
1401 return fuse_open_common(inode, file, true);
1404 static int fuse_dir_release(struct inode *inode, struct file *file)
1406 fuse_release_common(file, FUSE_RELEASEDIR);
1408 return 0;
1411 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1412 int datasync)
1414 return fuse_fsync_common(file, start, end, datasync, 1);
1417 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1418 unsigned long arg)
1420 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1422 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1423 if (fc->minor < 18)
1424 return -ENOTTY;
1426 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1429 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1430 unsigned long arg)
1432 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1434 if (fc->minor < 18)
1435 return -ENOTTY;
1437 return fuse_ioctl_common(file, cmd, arg,
1438 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1441 static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
1443 /* Always update if mtime is explicitly set */
1444 if (ivalid & ATTR_MTIME_SET)
1445 return true;
1447 /* Or if kernel i_mtime is the official one */
1448 if (trust_local_mtime)
1449 return true;
1451 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1452 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1453 return false;
1455 /* In all other cases update */
1456 return true;
1459 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg,
1460 bool trust_local_cmtime)
1462 unsigned ivalid = iattr->ia_valid;
1464 if (ivalid & ATTR_MODE)
1465 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1466 if (ivalid & ATTR_UID)
1467 arg->valid |= FATTR_UID, arg->uid = from_kuid(&init_user_ns, iattr->ia_uid);
1468 if (ivalid & ATTR_GID)
1469 arg->valid |= FATTR_GID, arg->gid = from_kgid(&init_user_ns, iattr->ia_gid);
1470 if (ivalid & ATTR_SIZE)
1471 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1472 if (ivalid & ATTR_ATIME) {
1473 arg->valid |= FATTR_ATIME;
1474 arg->atime = iattr->ia_atime.tv_sec;
1475 arg->atimensec = iattr->ia_atime.tv_nsec;
1476 if (!(ivalid & ATTR_ATIME_SET))
1477 arg->valid |= FATTR_ATIME_NOW;
1479 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_cmtime)) {
1480 arg->valid |= FATTR_MTIME;
1481 arg->mtime = iattr->ia_mtime.tv_sec;
1482 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1483 if (!(ivalid & ATTR_MTIME_SET) && !trust_local_cmtime)
1484 arg->valid |= FATTR_MTIME_NOW;
1486 if ((ivalid & ATTR_CTIME) && trust_local_cmtime) {
1487 arg->valid |= FATTR_CTIME;
1488 arg->ctime = iattr->ia_ctime.tv_sec;
1489 arg->ctimensec = iattr->ia_ctime.tv_nsec;
1494 * Prevent concurrent writepages on inode
1496 * This is done by adding a negative bias to the inode write counter
1497 * and waiting for all pending writes to finish.
1499 void fuse_set_nowrite(struct inode *inode)
1501 struct fuse_conn *fc = get_fuse_conn(inode);
1502 struct fuse_inode *fi = get_fuse_inode(inode);
1504 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1506 spin_lock(&fc->lock);
1507 BUG_ON(fi->writectr < 0);
1508 fi->writectr += FUSE_NOWRITE;
1509 spin_unlock(&fc->lock);
1510 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1514 * Allow writepages on inode
1516 * Remove the bias from the writecounter and send any queued
1517 * writepages.
1519 static void __fuse_release_nowrite(struct inode *inode)
1521 struct fuse_inode *fi = get_fuse_inode(inode);
1523 BUG_ON(fi->writectr != FUSE_NOWRITE);
1524 fi->writectr = 0;
1525 fuse_flush_writepages(inode);
1528 void fuse_release_nowrite(struct inode *inode)
1530 struct fuse_conn *fc = get_fuse_conn(inode);
1532 spin_lock(&fc->lock);
1533 __fuse_release_nowrite(inode);
1534 spin_unlock(&fc->lock);
1537 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_args *args,
1538 struct inode *inode,
1539 struct fuse_setattr_in *inarg_p,
1540 struct fuse_attr_out *outarg_p)
1542 args->in.h.opcode = FUSE_SETATTR;
1543 args->in.h.nodeid = get_node_id(inode);
1544 args->in.numargs = 1;
1545 args->in.args[0].size = sizeof(*inarg_p);
1546 args->in.args[0].value = inarg_p;
1547 args->out.numargs = 1;
1548 args->out.args[0].size = sizeof(*outarg_p);
1549 args->out.args[0].value = outarg_p;
1553 * Flush inode->i_mtime to the server
1555 int fuse_flush_times(struct inode *inode, struct fuse_file *ff)
1557 struct fuse_conn *fc = get_fuse_conn(inode);
1558 FUSE_ARGS(args);
1559 struct fuse_setattr_in inarg;
1560 struct fuse_attr_out outarg;
1562 memset(&inarg, 0, sizeof(inarg));
1563 memset(&outarg, 0, sizeof(outarg));
1565 inarg.valid = FATTR_MTIME;
1566 inarg.mtime = inode->i_mtime.tv_sec;
1567 inarg.mtimensec = inode->i_mtime.tv_nsec;
1568 if (fc->minor >= 23) {
1569 inarg.valid |= FATTR_CTIME;
1570 inarg.ctime = inode->i_ctime.tv_sec;
1571 inarg.ctimensec = inode->i_ctime.tv_nsec;
1573 if (ff) {
1574 inarg.valid |= FATTR_FH;
1575 inarg.fh = ff->fh;
1577 fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1579 return fuse_simple_request(fc, &args);
1583 * Set attributes, and at the same time refresh them.
1585 * Truncation is slightly complicated, because the 'truncate' request
1586 * may fail, in which case we don't want to touch the mapping.
1587 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1588 * and the actual truncation by hand.
1590 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
1591 struct file *file)
1593 struct fuse_conn *fc = get_fuse_conn(inode);
1594 struct fuse_inode *fi = get_fuse_inode(inode);
1595 FUSE_ARGS(args);
1596 struct fuse_setattr_in inarg;
1597 struct fuse_attr_out outarg;
1598 bool is_truncate = false;
1599 bool is_wb = fc->writeback_cache;
1600 loff_t oldsize;
1601 int err;
1602 bool trust_local_cmtime = is_wb && S_ISREG(inode->i_mode);
1604 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1605 attr->ia_valid |= ATTR_FORCE;
1607 err = inode_change_ok(inode, attr);
1608 if (err)
1609 return err;
1611 if (attr->ia_valid & ATTR_OPEN) {
1612 /* This is coming from open(..., ... | O_TRUNC); */
1613 WARN_ON(!(attr->ia_valid & ATTR_SIZE));
1614 WARN_ON(attr->ia_size != 0);
1615 if (fc->atomic_o_trunc) {
1617 * No need to send request to userspace, since actual
1618 * truncation has already been done by OPEN. But still
1619 * need to truncate page cache.
1621 i_size_write(inode, 0);
1622 truncate_pagecache(inode, 0);
1623 return 0;
1625 file = NULL;
1628 if (attr->ia_valid & ATTR_SIZE)
1629 is_truncate = true;
1631 if (is_truncate) {
1632 fuse_set_nowrite(inode);
1633 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1634 if (trust_local_cmtime && attr->ia_size != inode->i_size)
1635 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1638 memset(&inarg, 0, sizeof(inarg));
1639 memset(&outarg, 0, sizeof(outarg));
1640 iattr_to_fattr(attr, &inarg, trust_local_cmtime);
1641 if (file) {
1642 struct fuse_file *ff = file->private_data;
1643 inarg.valid |= FATTR_FH;
1644 inarg.fh = ff->fh;
1646 if (attr->ia_valid & ATTR_SIZE) {
1647 /* For mandatory locking in truncate */
1648 inarg.valid |= FATTR_LOCKOWNER;
1649 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1651 fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1652 err = fuse_simple_request(fc, &args);
1653 if (err) {
1654 if (err == -EINTR)
1655 fuse_invalidate_attr(inode);
1656 goto error;
1659 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1660 make_bad_inode(inode);
1661 err = -EIO;
1662 goto error;
1665 spin_lock(&fc->lock);
1666 /* the kernel maintains i_mtime locally */
1667 if (trust_local_cmtime) {
1668 if (attr->ia_valid & ATTR_MTIME)
1669 inode->i_mtime = attr->ia_mtime;
1670 if (attr->ia_valid & ATTR_CTIME)
1671 inode->i_ctime = attr->ia_ctime;
1672 /* FIXME: clear I_DIRTY_SYNC? */
1675 fuse_change_attributes_common(inode, &outarg.attr,
1676 attr_timeout(&outarg));
1677 oldsize = inode->i_size;
1678 /* see the comment in fuse_change_attributes() */
1679 if (!is_wb || is_truncate || !S_ISREG(inode->i_mode))
1680 i_size_write(inode, outarg.attr.size);
1682 if (is_truncate) {
1683 /* NOTE: this may release/reacquire fc->lock */
1684 __fuse_release_nowrite(inode);
1686 spin_unlock(&fc->lock);
1689 * Only call invalidate_inode_pages2() after removing
1690 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1692 if ((is_truncate || !is_wb) &&
1693 S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1694 truncate_pagecache(inode, outarg.attr.size);
1695 invalidate_inode_pages2(inode->i_mapping);
1698 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1699 return 0;
1701 error:
1702 if (is_truncate)
1703 fuse_release_nowrite(inode);
1705 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1706 return err;
1709 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1711 struct inode *inode = d_inode(entry);
1712 struct file *file = (attr->ia_valid & ATTR_FILE) ? attr->ia_file : NULL;
1713 int ret;
1715 if (!fuse_allow_current_process(get_fuse_conn(inode)))
1716 return -EACCES;
1718 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) {
1719 int kill;
1721 attr->ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID |
1722 ATTR_MODE);
1724 * ia_mode calculation may have used stale i_mode. Refresh and
1725 * recalculate.
1727 ret = fuse_do_getattr(inode, NULL, file);
1728 if (ret)
1729 return ret;
1731 attr->ia_mode = inode->i_mode;
1732 kill = should_remove_suid(entry);
1733 if (kill & ATTR_KILL_SUID) {
1734 attr->ia_valid |= ATTR_MODE;
1735 attr->ia_mode &= ~S_ISUID;
1737 if (kill & ATTR_KILL_SGID) {
1738 attr->ia_valid |= ATTR_MODE;
1739 attr->ia_mode &= ~S_ISGID;
1742 if (!attr->ia_valid)
1743 return 0;
1745 ret = fuse_do_setattr(inode, attr, file);
1746 if (!ret) {
1747 /* Directory mode changed, may need to revalidate access */
1748 if (d_is_dir(entry) && (attr->ia_valid & ATTR_MODE))
1749 fuse_invalidate_entry_cache(entry);
1751 return ret;
1754 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1755 struct kstat *stat)
1757 struct inode *inode = d_inode(entry);
1758 struct fuse_conn *fc = get_fuse_conn(inode);
1760 if (!fuse_allow_current_process(fc))
1761 return -EACCES;
1763 return fuse_update_attributes(inode, stat, NULL, NULL);
1766 static int fuse_setxattr(struct dentry *entry, const char *name,
1767 const void *value, size_t size, int flags)
1769 struct inode *inode = d_inode(entry);
1770 struct fuse_conn *fc = get_fuse_conn(inode);
1771 FUSE_ARGS(args);
1772 struct fuse_setxattr_in inarg;
1773 int err;
1775 if (fc->no_setxattr)
1776 return -EOPNOTSUPP;
1778 memset(&inarg, 0, sizeof(inarg));
1779 inarg.size = size;
1780 inarg.flags = flags;
1781 args.in.h.opcode = FUSE_SETXATTR;
1782 args.in.h.nodeid = get_node_id(inode);
1783 args.in.numargs = 3;
1784 args.in.args[0].size = sizeof(inarg);
1785 args.in.args[0].value = &inarg;
1786 args.in.args[1].size = strlen(name) + 1;
1787 args.in.args[1].value = name;
1788 args.in.args[2].size = size;
1789 args.in.args[2].value = value;
1790 err = fuse_simple_request(fc, &args);
1791 if (err == -ENOSYS) {
1792 fc->no_setxattr = 1;
1793 err = -EOPNOTSUPP;
1795 if (!err) {
1796 fuse_invalidate_attr(inode);
1797 fuse_update_ctime(inode);
1799 return err;
1802 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1803 void *value, size_t size)
1805 struct inode *inode = d_inode(entry);
1806 struct fuse_conn *fc = get_fuse_conn(inode);
1807 FUSE_ARGS(args);
1808 struct fuse_getxattr_in inarg;
1809 struct fuse_getxattr_out outarg;
1810 ssize_t ret;
1812 if (fc->no_getxattr)
1813 return -EOPNOTSUPP;
1815 memset(&inarg, 0, sizeof(inarg));
1816 inarg.size = size;
1817 args.in.h.opcode = FUSE_GETXATTR;
1818 args.in.h.nodeid = get_node_id(inode);
1819 args.in.numargs = 2;
1820 args.in.args[0].size = sizeof(inarg);
1821 args.in.args[0].value = &inarg;
1822 args.in.args[1].size = strlen(name) + 1;
1823 args.in.args[1].value = name;
1824 /* This is really two different operations rolled into one */
1825 args.out.numargs = 1;
1826 if (size) {
1827 args.out.argvar = 1;
1828 args.out.args[0].size = size;
1829 args.out.args[0].value = value;
1830 } else {
1831 args.out.args[0].size = sizeof(outarg);
1832 args.out.args[0].value = &outarg;
1834 ret = fuse_simple_request(fc, &args);
1835 if (!ret && !size)
1836 ret = outarg.size;
1837 if (ret == -ENOSYS) {
1838 fc->no_getxattr = 1;
1839 ret = -EOPNOTSUPP;
1841 return ret;
1844 static int fuse_verify_xattr_list(char *list, size_t size)
1846 size_t origsize = size;
1848 while (size) {
1849 size_t thislen = strnlen(list, size);
1851 if (!thislen || thislen == size)
1852 return -EIO;
1854 size -= thislen + 1;
1855 list += thislen + 1;
1858 return origsize;
1861 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1863 struct inode *inode = d_inode(entry);
1864 struct fuse_conn *fc = get_fuse_conn(inode);
1865 FUSE_ARGS(args);
1866 struct fuse_getxattr_in inarg;
1867 struct fuse_getxattr_out outarg;
1868 ssize_t ret;
1870 if (!fuse_allow_current_process(fc))
1871 return -EACCES;
1873 if (fc->no_listxattr)
1874 return -EOPNOTSUPP;
1876 memset(&inarg, 0, sizeof(inarg));
1877 inarg.size = size;
1878 args.in.h.opcode = FUSE_LISTXATTR;
1879 args.in.h.nodeid = get_node_id(inode);
1880 args.in.numargs = 1;
1881 args.in.args[0].size = sizeof(inarg);
1882 args.in.args[0].value = &inarg;
1883 /* This is really two different operations rolled into one */
1884 args.out.numargs = 1;
1885 if (size) {
1886 args.out.argvar = 1;
1887 args.out.args[0].size = size;
1888 args.out.args[0].value = list;
1889 } else {
1890 args.out.args[0].size = sizeof(outarg);
1891 args.out.args[0].value = &outarg;
1893 ret = fuse_simple_request(fc, &args);
1894 if (!ret && !size)
1895 ret = outarg.size;
1896 if (ret > 0 && size)
1897 ret = fuse_verify_xattr_list(list, ret);
1898 if (ret == -ENOSYS) {
1899 fc->no_listxattr = 1;
1900 ret = -EOPNOTSUPP;
1902 return ret;
1905 static int fuse_removexattr(struct dentry *entry, const char *name)
1907 struct inode *inode = d_inode(entry);
1908 struct fuse_conn *fc = get_fuse_conn(inode);
1909 FUSE_ARGS(args);
1910 int err;
1912 if (fc->no_removexattr)
1913 return -EOPNOTSUPP;
1915 args.in.h.opcode = FUSE_REMOVEXATTR;
1916 args.in.h.nodeid = get_node_id(inode);
1917 args.in.numargs = 1;
1918 args.in.args[0].size = strlen(name) + 1;
1919 args.in.args[0].value = name;
1920 err = fuse_simple_request(fc, &args);
1921 if (err == -ENOSYS) {
1922 fc->no_removexattr = 1;
1923 err = -EOPNOTSUPP;
1925 if (!err) {
1926 fuse_invalidate_attr(inode);
1927 fuse_update_ctime(inode);
1929 return err;
1932 static const struct inode_operations fuse_dir_inode_operations = {
1933 .lookup = fuse_lookup,
1934 .mkdir = fuse_mkdir,
1935 .symlink = fuse_symlink,
1936 .unlink = fuse_unlink,
1937 .rmdir = fuse_rmdir,
1938 .rename2 = fuse_rename2,
1939 .link = fuse_link,
1940 .setattr = fuse_setattr,
1941 .create = fuse_create,
1942 .atomic_open = fuse_atomic_open,
1943 .mknod = fuse_mknod,
1944 .permission = fuse_permission,
1945 .getattr = fuse_getattr,
1946 .setxattr = fuse_setxattr,
1947 .getxattr = fuse_getxattr,
1948 .listxattr = fuse_listxattr,
1949 .removexattr = fuse_removexattr,
1952 static const struct file_operations fuse_dir_operations = {
1953 .llseek = generic_file_llseek,
1954 .read = generic_read_dir,
1955 .iterate = fuse_readdir,
1956 .open = fuse_dir_open,
1957 .release = fuse_dir_release,
1958 .fsync = fuse_dir_fsync,
1959 .unlocked_ioctl = fuse_dir_ioctl,
1960 .compat_ioctl = fuse_dir_compat_ioctl,
1963 static const struct inode_operations fuse_common_inode_operations = {
1964 .setattr = fuse_setattr,
1965 .permission = fuse_permission,
1966 .getattr = fuse_getattr,
1967 .setxattr = fuse_setxattr,
1968 .getxattr = fuse_getxattr,
1969 .listxattr = fuse_listxattr,
1970 .removexattr = fuse_removexattr,
1973 static const struct inode_operations fuse_symlink_inode_operations = {
1974 .setattr = fuse_setattr,
1975 .follow_link = fuse_follow_link,
1976 .put_link = free_page_put_link,
1977 .readlink = generic_readlink,
1978 .getattr = fuse_getattr,
1979 .setxattr = fuse_setxattr,
1980 .getxattr = fuse_getxattr,
1981 .listxattr = fuse_listxattr,
1982 .removexattr = fuse_removexattr,
1985 void fuse_init_common(struct inode *inode)
1987 inode->i_op = &fuse_common_inode_operations;
1990 void fuse_init_dir(struct inode *inode)
1992 inode->i_op = &fuse_dir_inode_operations;
1993 inode->i_fop = &fuse_dir_operations;
1996 void fuse_init_symlink(struct inode *inode)
1998 inode->i_op = &fuse_symlink_inode_operations;