spi: davinci: fix up dma_mapping_error() incorrect patch
[linux/fpc-iii.git] / fs / fuse / dir.c
blob4bbad745415a0228c81d9553924ebe2de928e133
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>
16 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
19 static bool fuse_use_readdirplus(struct inode *dir, struct dir_context *ctx)
21 struct fuse_conn *fc = get_fuse_conn(dir);
22 struct fuse_inode *fi = get_fuse_inode(dir);
24 if (!fc->do_readdirplus)
25 return false;
26 if (!fc->readdirplus_auto)
27 return true;
28 if (test_and_clear_bit(FUSE_I_ADVISE_RDPLUS, &fi->state))
29 return true;
30 if (ctx->pos == 0)
31 return true;
32 return false;
35 static void fuse_advise_use_readdirplus(struct inode *dir)
37 struct fuse_inode *fi = get_fuse_inode(dir);
39 set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
42 union fuse_dentry {
43 u64 time;
44 struct rcu_head rcu;
47 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
49 ((union fuse_dentry *) entry->d_fsdata)->time = time;
52 static inline u64 fuse_dentry_time(struct dentry *entry)
54 return ((union fuse_dentry *) entry->d_fsdata)->time;
58 * FUSE caches dentries and attributes with separate timeout. The
59 * time in jiffies until the dentry/attributes are valid is stored in
60 * dentry->d_fsdata and fuse_inode->i_time respectively.
64 * Calculate the time in jiffies until a dentry/attributes are valid
66 static u64 time_to_jiffies(u64 sec, u32 nsec)
68 if (sec || nsec) {
69 struct timespec64 ts = {
70 sec,
71 min_t(u32, nsec, NSEC_PER_SEC - 1)
74 return get_jiffies_64() + timespec64_to_jiffies(&ts);
75 } else
76 return 0;
80 * Set dentry and possibly attribute timeouts from the lookup/mk*
81 * replies
83 static void fuse_change_entry_timeout(struct dentry *entry,
84 struct fuse_entry_out *o)
86 fuse_dentry_settime(entry,
87 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
90 static u64 attr_timeout(struct fuse_attr_out *o)
92 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
95 static u64 entry_attr_timeout(struct fuse_entry_out *o)
97 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
101 * Mark the attributes as stale, so that at the next call to
102 * ->getattr() they will be fetched from userspace
104 void fuse_invalidate_attr(struct inode *inode)
106 get_fuse_inode(inode)->i_time = 0;
110 * Mark the attributes as stale due to an atime change. Avoid the invalidate if
111 * atime is not used.
113 void fuse_invalidate_atime(struct inode *inode)
115 if (!IS_RDONLY(inode))
116 fuse_invalidate_attr(inode);
120 * Just mark the entry as stale, so that a next attempt to look it up
121 * will result in a new lookup call to userspace
123 * This is called when a dentry is about to become negative and the
124 * timeout is unknown (unlink, rmdir, rename and in some cases
125 * lookup)
127 void fuse_invalidate_entry_cache(struct dentry *entry)
129 fuse_dentry_settime(entry, 0);
133 * Same as fuse_invalidate_entry_cache(), but also try to remove the
134 * dentry from the hash
136 static void fuse_invalidate_entry(struct dentry *entry)
138 d_invalidate(entry);
139 fuse_invalidate_entry_cache(entry);
142 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
143 u64 nodeid, const struct qstr *name,
144 struct fuse_entry_out *outarg)
146 memset(outarg, 0, sizeof(struct fuse_entry_out));
147 args->in.h.opcode = FUSE_LOOKUP;
148 args->in.h.nodeid = nodeid;
149 args->in.numargs = 1;
150 args->in.args[0].size = name->len + 1;
151 args->in.args[0].value = name->name;
152 args->out.numargs = 1;
153 args->out.args[0].size = sizeof(struct fuse_entry_out);
154 args->out.args[0].value = outarg;
157 u64 fuse_get_attr_version(struct fuse_conn *fc)
159 u64 curr_version;
162 * The spin lock isn't actually needed on 64bit archs, but we
163 * don't yet care too much about such optimizations.
165 spin_lock(&fc->lock);
166 curr_version = fc->attr_version;
167 spin_unlock(&fc->lock);
169 return curr_version;
173 * Check whether the dentry is still valid
175 * If the entry validity timeout has expired and the dentry is
176 * positive, try to redo the lookup. If the lookup results in a
177 * different inode, then let the VFS invalidate the dentry and redo
178 * the lookup once more. If the lookup results in the same inode,
179 * then refresh the attributes, timeouts and mark the dentry valid.
181 static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
183 struct inode *inode;
184 struct dentry *parent;
185 struct fuse_conn *fc;
186 struct fuse_inode *fi;
187 int ret;
189 inode = d_inode_rcu(entry);
190 if (inode && is_bad_inode(inode))
191 goto invalid;
192 else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
193 (flags & LOOKUP_REVAL)) {
194 struct fuse_entry_out outarg;
195 FUSE_ARGS(args);
196 struct fuse_forget_link *forget;
197 u64 attr_version;
199 /* For negative dentries, always do a fresh lookup */
200 if (!inode)
201 goto invalid;
203 ret = -ECHILD;
204 if (flags & LOOKUP_RCU)
205 goto out;
207 fc = get_fuse_conn(inode);
209 forget = fuse_alloc_forget();
210 ret = -ENOMEM;
211 if (!forget)
212 goto out;
214 attr_version = fuse_get_attr_version(fc);
216 parent = dget_parent(entry);
217 fuse_lookup_init(fc, &args, get_node_id(d_inode(parent)),
218 &entry->d_name, &outarg);
219 ret = fuse_simple_request(fc, &args);
220 dput(parent);
221 /* Zero nodeid is same as -ENOENT */
222 if (!ret && !outarg.nodeid)
223 ret = -ENOENT;
224 if (!ret) {
225 fi = get_fuse_inode(inode);
226 if (outarg.nodeid != get_node_id(inode)) {
227 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
228 goto invalid;
230 spin_lock(&fc->lock);
231 fi->nlookup++;
232 spin_unlock(&fc->lock);
234 kfree(forget);
235 if (ret == -ENOMEM)
236 goto out;
237 if (ret || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
238 goto invalid;
240 forget_all_cached_acls(inode);
241 fuse_change_attributes(inode, &outarg.attr,
242 entry_attr_timeout(&outarg),
243 attr_version);
244 fuse_change_entry_timeout(entry, &outarg);
245 } else if (inode) {
246 fi = get_fuse_inode(inode);
247 if (flags & LOOKUP_RCU) {
248 if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
249 return -ECHILD;
250 } else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
251 parent = dget_parent(entry);
252 fuse_advise_use_readdirplus(d_inode(parent));
253 dput(parent);
256 ret = 1;
257 out:
258 return ret;
260 invalid:
261 ret = 0;
262 goto out;
265 static int invalid_nodeid(u64 nodeid)
267 return !nodeid || nodeid == FUSE_ROOT_ID;
270 static int fuse_dentry_init(struct dentry *dentry)
272 dentry->d_fsdata = kzalloc(sizeof(union fuse_dentry), GFP_KERNEL);
274 return dentry->d_fsdata ? 0 : -ENOMEM;
276 static void fuse_dentry_release(struct dentry *dentry)
278 union fuse_dentry *fd = dentry->d_fsdata;
280 kfree_rcu(fd, rcu);
283 const struct dentry_operations fuse_dentry_operations = {
284 .d_revalidate = fuse_dentry_revalidate,
285 .d_init = fuse_dentry_init,
286 .d_release = fuse_dentry_release,
289 const struct dentry_operations fuse_root_dentry_operations = {
290 .d_init = fuse_dentry_init,
291 .d_release = fuse_dentry_release,
294 int fuse_valid_type(int m)
296 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
297 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
300 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
301 struct fuse_entry_out *outarg, struct inode **inode)
303 struct fuse_conn *fc = get_fuse_conn_super(sb);
304 FUSE_ARGS(args);
305 struct fuse_forget_link *forget;
306 u64 attr_version;
307 int err;
309 *inode = NULL;
310 err = -ENAMETOOLONG;
311 if (name->len > FUSE_NAME_MAX)
312 goto out;
315 forget = fuse_alloc_forget();
316 err = -ENOMEM;
317 if (!forget)
318 goto out;
320 attr_version = fuse_get_attr_version(fc);
322 fuse_lookup_init(fc, &args, nodeid, name, outarg);
323 err = fuse_simple_request(fc, &args);
324 /* Zero nodeid is same as -ENOENT, but with valid timeout */
325 if (err || !outarg->nodeid)
326 goto out_put_forget;
328 err = -EIO;
329 if (!outarg->nodeid)
330 goto out_put_forget;
331 if (!fuse_valid_type(outarg->attr.mode))
332 goto out_put_forget;
334 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
335 &outarg->attr, entry_attr_timeout(outarg),
336 attr_version);
337 err = -ENOMEM;
338 if (!*inode) {
339 fuse_queue_forget(fc, forget, outarg->nodeid, 1);
340 goto out;
342 err = 0;
344 out_put_forget:
345 kfree(forget);
346 out:
347 return err;
350 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
351 unsigned int flags)
353 int err;
354 struct fuse_entry_out outarg;
355 struct inode *inode;
356 struct dentry *newent;
357 bool outarg_valid = true;
359 fuse_lock_inode(dir);
360 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
361 &outarg, &inode);
362 fuse_unlock_inode(dir);
363 if (err == -ENOENT) {
364 outarg_valid = false;
365 err = 0;
367 if (err)
368 goto out_err;
370 err = -EIO;
371 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
372 goto out_iput;
374 newent = d_splice_alias(inode, entry);
375 err = PTR_ERR(newent);
376 if (IS_ERR(newent))
377 goto out_err;
379 entry = newent ? newent : entry;
380 if (outarg_valid)
381 fuse_change_entry_timeout(entry, &outarg);
382 else
383 fuse_invalidate_entry_cache(entry);
385 fuse_advise_use_readdirplus(dir);
386 return newent;
388 out_iput:
389 iput(inode);
390 out_err:
391 return ERR_PTR(err);
395 * Atomic create+open operation
397 * If the filesystem doesn't support this, then fall back to separate
398 * 'mknod' + 'open' requests.
400 static int fuse_create_open(struct inode *dir, struct dentry *entry,
401 struct file *file, unsigned flags,
402 umode_t mode, int *opened)
404 int err;
405 struct inode *inode;
406 struct fuse_conn *fc = get_fuse_conn(dir);
407 FUSE_ARGS(args);
408 struct fuse_forget_link *forget;
409 struct fuse_create_in inarg;
410 struct fuse_open_out outopen;
411 struct fuse_entry_out outentry;
412 struct fuse_file *ff;
414 /* Userspace expects S_IFREG in create mode */
415 BUG_ON((mode & S_IFMT) != S_IFREG);
417 forget = fuse_alloc_forget();
418 err = -ENOMEM;
419 if (!forget)
420 goto out_err;
422 err = -ENOMEM;
423 ff = fuse_file_alloc(fc);
424 if (!ff)
425 goto out_put_forget_req;
427 if (!fc->dont_mask)
428 mode &= ~current_umask();
430 flags &= ~O_NOCTTY;
431 memset(&inarg, 0, sizeof(inarg));
432 memset(&outentry, 0, sizeof(outentry));
433 inarg.flags = flags;
434 inarg.mode = mode;
435 inarg.umask = current_umask();
436 args.in.h.opcode = FUSE_CREATE;
437 args.in.h.nodeid = get_node_id(dir);
438 args.in.numargs = 2;
439 args.in.args[0].size = sizeof(inarg);
440 args.in.args[0].value = &inarg;
441 args.in.args[1].size = entry->d_name.len + 1;
442 args.in.args[1].value = entry->d_name.name;
443 args.out.numargs = 2;
444 args.out.args[0].size = sizeof(outentry);
445 args.out.args[0].value = &outentry;
446 args.out.args[1].size = sizeof(outopen);
447 args.out.args[1].value = &outopen;
448 err = fuse_simple_request(fc, &args);
449 if (err)
450 goto out_free_ff;
452 err = -EIO;
453 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
454 goto out_free_ff;
456 ff->fh = outopen.fh;
457 ff->nodeid = outentry.nodeid;
458 ff->open_flags = outopen.open_flags;
459 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
460 &outentry.attr, entry_attr_timeout(&outentry), 0);
461 if (!inode) {
462 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
463 fuse_sync_release(ff, flags);
464 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
465 err = -ENOMEM;
466 goto out_err;
468 kfree(forget);
469 d_instantiate(entry, inode);
470 fuse_change_entry_timeout(entry, &outentry);
471 fuse_invalidate_attr(dir);
472 err = finish_open(file, entry, generic_file_open, opened);
473 if (err) {
474 fuse_sync_release(ff, flags);
475 } else {
476 file->private_data = fuse_file_get(ff);
477 fuse_finish_open(inode, file);
479 return err;
481 out_free_ff:
482 fuse_file_free(ff);
483 out_put_forget_req:
484 kfree(forget);
485 out_err:
486 return err;
489 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
490 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
491 struct file *file, unsigned flags,
492 umode_t mode, int *opened)
494 int err;
495 struct fuse_conn *fc = get_fuse_conn(dir);
496 struct dentry *res = NULL;
498 if (d_in_lookup(entry)) {
499 res = fuse_lookup(dir, entry, 0);
500 if (IS_ERR(res))
501 return PTR_ERR(res);
503 if (res)
504 entry = res;
507 if (!(flags & O_CREAT) || d_really_is_positive(entry))
508 goto no_open;
510 /* Only creates */
511 *opened |= FILE_CREATED;
513 if (fc->no_create)
514 goto mknod;
516 err = fuse_create_open(dir, entry, file, flags, mode, opened);
517 if (err == -ENOSYS) {
518 fc->no_create = 1;
519 goto mknod;
521 out_dput:
522 dput(res);
523 return err;
525 mknod:
526 err = fuse_mknod(dir, entry, mode, 0);
527 if (err)
528 goto out_dput;
529 no_open:
530 return finish_no_open(file, res);
534 * Code shared between mknod, mkdir, symlink and link
536 static int create_new_entry(struct fuse_conn *fc, struct fuse_args *args,
537 struct inode *dir, struct dentry *entry,
538 umode_t mode)
540 struct fuse_entry_out outarg;
541 struct inode *inode;
542 int err;
543 struct fuse_forget_link *forget;
545 forget = fuse_alloc_forget();
546 if (!forget)
547 return -ENOMEM;
549 memset(&outarg, 0, sizeof(outarg));
550 args->in.h.nodeid = get_node_id(dir);
551 args->out.numargs = 1;
552 args->out.args[0].size = sizeof(outarg);
553 args->out.args[0].value = &outarg;
554 err = fuse_simple_request(fc, args);
555 if (err)
556 goto out_put_forget_req;
558 err = -EIO;
559 if (invalid_nodeid(outarg.nodeid))
560 goto out_put_forget_req;
562 if ((outarg.attr.mode ^ mode) & S_IFMT)
563 goto out_put_forget_req;
565 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
566 &outarg.attr, entry_attr_timeout(&outarg), 0);
567 if (!inode) {
568 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
569 return -ENOMEM;
571 kfree(forget);
573 err = d_instantiate_no_diralias(entry, inode);
574 if (err)
575 return err;
577 fuse_change_entry_timeout(entry, &outarg);
578 fuse_invalidate_attr(dir);
579 return 0;
581 out_put_forget_req:
582 kfree(forget);
583 return err;
586 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
587 dev_t rdev)
589 struct fuse_mknod_in inarg;
590 struct fuse_conn *fc = get_fuse_conn(dir);
591 FUSE_ARGS(args);
593 if (!fc->dont_mask)
594 mode &= ~current_umask();
596 memset(&inarg, 0, sizeof(inarg));
597 inarg.mode = mode;
598 inarg.rdev = new_encode_dev(rdev);
599 inarg.umask = current_umask();
600 args.in.h.opcode = FUSE_MKNOD;
601 args.in.numargs = 2;
602 args.in.args[0].size = sizeof(inarg);
603 args.in.args[0].value = &inarg;
604 args.in.args[1].size = entry->d_name.len + 1;
605 args.in.args[1].value = entry->d_name.name;
606 return create_new_entry(fc, &args, dir, entry, mode);
609 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
610 bool excl)
612 return fuse_mknod(dir, entry, mode, 0);
615 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
617 struct fuse_mkdir_in inarg;
618 struct fuse_conn *fc = get_fuse_conn(dir);
619 FUSE_ARGS(args);
621 if (!fc->dont_mask)
622 mode &= ~current_umask();
624 memset(&inarg, 0, sizeof(inarg));
625 inarg.mode = mode;
626 inarg.umask = current_umask();
627 args.in.h.opcode = FUSE_MKDIR;
628 args.in.numargs = 2;
629 args.in.args[0].size = sizeof(inarg);
630 args.in.args[0].value = &inarg;
631 args.in.args[1].size = entry->d_name.len + 1;
632 args.in.args[1].value = entry->d_name.name;
633 return create_new_entry(fc, &args, dir, entry, S_IFDIR);
636 static int fuse_symlink(struct inode *dir, struct dentry *entry,
637 const char *link)
639 struct fuse_conn *fc = get_fuse_conn(dir);
640 unsigned len = strlen(link) + 1;
641 FUSE_ARGS(args);
643 args.in.h.opcode = FUSE_SYMLINK;
644 args.in.numargs = 2;
645 args.in.args[0].size = entry->d_name.len + 1;
646 args.in.args[0].value = entry->d_name.name;
647 args.in.args[1].size = len;
648 args.in.args[1].value = link;
649 return create_new_entry(fc, &args, dir, entry, S_IFLNK);
652 void fuse_update_ctime(struct inode *inode)
654 if (!IS_NOCMTIME(inode)) {
655 inode->i_ctime = current_time(inode);
656 mark_inode_dirty_sync(inode);
660 static int fuse_unlink(struct inode *dir, struct dentry *entry)
662 int err;
663 struct fuse_conn *fc = get_fuse_conn(dir);
664 FUSE_ARGS(args);
666 args.in.h.opcode = FUSE_UNLINK;
667 args.in.h.nodeid = get_node_id(dir);
668 args.in.numargs = 1;
669 args.in.args[0].size = entry->d_name.len + 1;
670 args.in.args[0].value = entry->d_name.name;
671 err = fuse_simple_request(fc, &args);
672 if (!err) {
673 struct inode *inode = d_inode(entry);
674 struct fuse_inode *fi = get_fuse_inode(inode);
676 spin_lock(&fc->lock);
677 fi->attr_version = ++fc->attr_version;
679 * If i_nlink == 0 then unlink doesn't make sense, yet this can
680 * happen if userspace filesystem is careless. It would be
681 * difficult to enforce correct nlink usage so just ignore this
682 * condition here
684 if (inode->i_nlink > 0)
685 drop_nlink(inode);
686 spin_unlock(&fc->lock);
687 fuse_invalidate_attr(inode);
688 fuse_invalidate_attr(dir);
689 fuse_invalidate_entry_cache(entry);
690 fuse_update_ctime(inode);
691 } else if (err == -EINTR)
692 fuse_invalidate_entry(entry);
693 return err;
696 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
698 int err;
699 struct fuse_conn *fc = get_fuse_conn(dir);
700 FUSE_ARGS(args);
702 args.in.h.opcode = FUSE_RMDIR;
703 args.in.h.nodeid = get_node_id(dir);
704 args.in.numargs = 1;
705 args.in.args[0].size = entry->d_name.len + 1;
706 args.in.args[0].value = entry->d_name.name;
707 err = fuse_simple_request(fc, &args);
708 if (!err) {
709 clear_nlink(d_inode(entry));
710 fuse_invalidate_attr(dir);
711 fuse_invalidate_entry_cache(entry);
712 } else if (err == -EINTR)
713 fuse_invalidate_entry(entry);
714 return err;
717 static int fuse_rename_common(struct inode *olddir, struct dentry *oldent,
718 struct inode *newdir, struct dentry *newent,
719 unsigned int flags, int opcode, size_t argsize)
721 int err;
722 struct fuse_rename2_in inarg;
723 struct fuse_conn *fc = get_fuse_conn(olddir);
724 FUSE_ARGS(args);
726 memset(&inarg, 0, argsize);
727 inarg.newdir = get_node_id(newdir);
728 inarg.flags = flags;
729 args.in.h.opcode = opcode;
730 args.in.h.nodeid = get_node_id(olddir);
731 args.in.numargs = 3;
732 args.in.args[0].size = argsize;
733 args.in.args[0].value = &inarg;
734 args.in.args[1].size = oldent->d_name.len + 1;
735 args.in.args[1].value = oldent->d_name.name;
736 args.in.args[2].size = newent->d_name.len + 1;
737 args.in.args[2].value = newent->d_name.name;
738 err = fuse_simple_request(fc, &args);
739 if (!err) {
740 /* ctime changes */
741 fuse_invalidate_attr(d_inode(oldent));
742 fuse_update_ctime(d_inode(oldent));
744 if (flags & RENAME_EXCHANGE) {
745 fuse_invalidate_attr(d_inode(newent));
746 fuse_update_ctime(d_inode(newent));
749 fuse_invalidate_attr(olddir);
750 if (olddir != newdir)
751 fuse_invalidate_attr(newdir);
753 /* newent will end up negative */
754 if (!(flags & RENAME_EXCHANGE) && d_really_is_positive(newent)) {
755 fuse_invalidate_attr(d_inode(newent));
756 fuse_invalidate_entry_cache(newent);
757 fuse_update_ctime(d_inode(newent));
759 } else if (err == -EINTR) {
760 /* If request was interrupted, DEITY only knows if the
761 rename actually took place. If the invalidation
762 fails (e.g. some process has CWD under the renamed
763 directory), then there can be inconsistency between
764 the dcache and the real filesystem. Tough luck. */
765 fuse_invalidate_entry(oldent);
766 if (d_really_is_positive(newent))
767 fuse_invalidate_entry(newent);
770 return err;
773 static int fuse_rename2(struct inode *olddir, struct dentry *oldent,
774 struct inode *newdir, struct dentry *newent,
775 unsigned int flags)
777 struct fuse_conn *fc = get_fuse_conn(olddir);
778 int err;
780 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
781 return -EINVAL;
783 if (flags) {
784 if (fc->no_rename2 || fc->minor < 23)
785 return -EINVAL;
787 err = fuse_rename_common(olddir, oldent, newdir, newent, flags,
788 FUSE_RENAME2,
789 sizeof(struct fuse_rename2_in));
790 if (err == -ENOSYS) {
791 fc->no_rename2 = 1;
792 err = -EINVAL;
794 } else {
795 err = fuse_rename_common(olddir, oldent, newdir, newent, 0,
796 FUSE_RENAME,
797 sizeof(struct fuse_rename_in));
800 return err;
803 static int fuse_link(struct dentry *entry, struct inode *newdir,
804 struct dentry *newent)
806 int err;
807 struct fuse_link_in inarg;
808 struct inode *inode = d_inode(entry);
809 struct fuse_conn *fc = get_fuse_conn(inode);
810 FUSE_ARGS(args);
812 memset(&inarg, 0, sizeof(inarg));
813 inarg.oldnodeid = get_node_id(inode);
814 args.in.h.opcode = FUSE_LINK;
815 args.in.numargs = 2;
816 args.in.args[0].size = sizeof(inarg);
817 args.in.args[0].value = &inarg;
818 args.in.args[1].size = newent->d_name.len + 1;
819 args.in.args[1].value = newent->d_name.name;
820 err = create_new_entry(fc, &args, newdir, newent, inode->i_mode);
821 /* Contrary to "normal" filesystems it can happen that link
822 makes two "logical" inodes point to the same "physical"
823 inode. We invalidate the attributes of the old one, so it
824 will reflect changes in the backing inode (link count,
825 etc.)
827 if (!err) {
828 struct fuse_inode *fi = get_fuse_inode(inode);
830 spin_lock(&fc->lock);
831 fi->attr_version = ++fc->attr_version;
832 inc_nlink(inode);
833 spin_unlock(&fc->lock);
834 fuse_invalidate_attr(inode);
835 fuse_update_ctime(inode);
836 } else if (err == -EINTR) {
837 fuse_invalidate_attr(inode);
839 return err;
842 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
843 struct kstat *stat)
845 unsigned int blkbits;
846 struct fuse_conn *fc = get_fuse_conn(inode);
848 /* see the comment in fuse_change_attributes() */
849 if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
850 attr->size = i_size_read(inode);
851 attr->mtime = inode->i_mtime.tv_sec;
852 attr->mtimensec = inode->i_mtime.tv_nsec;
853 attr->ctime = inode->i_ctime.tv_sec;
854 attr->ctimensec = inode->i_ctime.tv_nsec;
857 stat->dev = inode->i_sb->s_dev;
858 stat->ino = attr->ino;
859 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
860 stat->nlink = attr->nlink;
861 stat->uid = make_kuid(&init_user_ns, attr->uid);
862 stat->gid = make_kgid(&init_user_ns, attr->gid);
863 stat->rdev = inode->i_rdev;
864 stat->atime.tv_sec = attr->atime;
865 stat->atime.tv_nsec = attr->atimensec;
866 stat->mtime.tv_sec = attr->mtime;
867 stat->mtime.tv_nsec = attr->mtimensec;
868 stat->ctime.tv_sec = attr->ctime;
869 stat->ctime.tv_nsec = attr->ctimensec;
870 stat->size = attr->size;
871 stat->blocks = attr->blocks;
873 if (attr->blksize != 0)
874 blkbits = ilog2(attr->blksize);
875 else
876 blkbits = inode->i_sb->s_blocksize_bits;
878 stat->blksize = 1 << blkbits;
881 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
882 struct file *file)
884 int err;
885 struct fuse_getattr_in inarg;
886 struct fuse_attr_out outarg;
887 struct fuse_conn *fc = get_fuse_conn(inode);
888 FUSE_ARGS(args);
889 u64 attr_version;
891 attr_version = fuse_get_attr_version(fc);
893 memset(&inarg, 0, sizeof(inarg));
894 memset(&outarg, 0, sizeof(outarg));
895 /* Directories have separate file-handle space */
896 if (file && S_ISREG(inode->i_mode)) {
897 struct fuse_file *ff = file->private_data;
899 inarg.getattr_flags |= FUSE_GETATTR_FH;
900 inarg.fh = ff->fh;
902 args.in.h.opcode = FUSE_GETATTR;
903 args.in.h.nodeid = get_node_id(inode);
904 args.in.numargs = 1;
905 args.in.args[0].size = sizeof(inarg);
906 args.in.args[0].value = &inarg;
907 args.out.numargs = 1;
908 args.out.args[0].size = sizeof(outarg);
909 args.out.args[0].value = &outarg;
910 err = fuse_simple_request(fc, &args);
911 if (!err) {
912 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
913 make_bad_inode(inode);
914 err = -EIO;
915 } else {
916 fuse_change_attributes(inode, &outarg.attr,
917 attr_timeout(&outarg),
918 attr_version);
919 if (stat)
920 fuse_fillattr(inode, &outarg.attr, stat);
923 return err;
926 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
927 struct file *file, bool *refreshed)
929 struct fuse_inode *fi = get_fuse_inode(inode);
930 int err;
931 bool r;
933 if (time_before64(fi->i_time, get_jiffies_64())) {
934 r = true;
935 forget_all_cached_acls(inode);
936 err = fuse_do_getattr(inode, stat, file);
937 } else {
938 r = false;
939 err = 0;
940 if (stat) {
941 generic_fillattr(inode, stat);
942 stat->mode = fi->orig_i_mode;
943 stat->ino = fi->orig_ino;
947 if (refreshed != NULL)
948 *refreshed = r;
950 return err;
953 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
954 u64 child_nodeid, struct qstr *name)
956 int err = -ENOTDIR;
957 struct inode *parent;
958 struct dentry *dir;
959 struct dentry *entry;
961 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
962 if (!parent)
963 return -ENOENT;
965 inode_lock(parent);
966 if (!S_ISDIR(parent->i_mode))
967 goto unlock;
969 err = -ENOENT;
970 dir = d_find_alias(parent);
971 if (!dir)
972 goto unlock;
974 name->hash = full_name_hash(dir, name->name, name->len);
975 entry = d_lookup(dir, name);
976 dput(dir);
977 if (!entry)
978 goto unlock;
980 fuse_invalidate_attr(parent);
981 fuse_invalidate_entry(entry);
983 if (child_nodeid != 0 && d_really_is_positive(entry)) {
984 inode_lock(d_inode(entry));
985 if (get_node_id(d_inode(entry)) != child_nodeid) {
986 err = -ENOENT;
987 goto badentry;
989 if (d_mountpoint(entry)) {
990 err = -EBUSY;
991 goto badentry;
993 if (d_is_dir(entry)) {
994 shrink_dcache_parent(entry);
995 if (!simple_empty(entry)) {
996 err = -ENOTEMPTY;
997 goto badentry;
999 d_inode(entry)->i_flags |= S_DEAD;
1001 dont_mount(entry);
1002 clear_nlink(d_inode(entry));
1003 err = 0;
1004 badentry:
1005 inode_unlock(d_inode(entry));
1006 if (!err)
1007 d_delete(entry);
1008 } else {
1009 err = 0;
1011 dput(entry);
1013 unlock:
1014 inode_unlock(parent);
1015 iput(parent);
1016 return err;
1020 * Calling into a user-controlled filesystem gives the filesystem
1021 * daemon ptrace-like capabilities over the current process. This
1022 * means, that the filesystem daemon is able to record the exact
1023 * filesystem operations performed, and can also control the behavior
1024 * of the requester process in otherwise impossible ways. For example
1025 * it can delay the operation for arbitrary length of time allowing
1026 * DoS against the requester.
1028 * For this reason only those processes can call into the filesystem,
1029 * for which the owner of the mount has ptrace privilege. This
1030 * excludes processes started by other users, suid or sgid processes.
1032 int fuse_allow_current_process(struct fuse_conn *fc)
1034 const struct cred *cred;
1036 if (fc->allow_other)
1037 return 1;
1039 cred = current_cred();
1040 if (uid_eq(cred->euid, fc->user_id) &&
1041 uid_eq(cred->suid, fc->user_id) &&
1042 uid_eq(cred->uid, fc->user_id) &&
1043 gid_eq(cred->egid, fc->group_id) &&
1044 gid_eq(cred->sgid, fc->group_id) &&
1045 gid_eq(cred->gid, fc->group_id))
1046 return 1;
1048 return 0;
1051 static int fuse_access(struct inode *inode, int mask)
1053 struct fuse_conn *fc = get_fuse_conn(inode);
1054 FUSE_ARGS(args);
1055 struct fuse_access_in inarg;
1056 int err;
1058 BUG_ON(mask & MAY_NOT_BLOCK);
1060 if (fc->no_access)
1061 return 0;
1063 memset(&inarg, 0, sizeof(inarg));
1064 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1065 args.in.h.opcode = FUSE_ACCESS;
1066 args.in.h.nodeid = get_node_id(inode);
1067 args.in.numargs = 1;
1068 args.in.args[0].size = sizeof(inarg);
1069 args.in.args[0].value = &inarg;
1070 err = fuse_simple_request(fc, &args);
1071 if (err == -ENOSYS) {
1072 fc->no_access = 1;
1073 err = 0;
1075 return err;
1078 static int fuse_perm_getattr(struct inode *inode, int mask)
1080 if (mask & MAY_NOT_BLOCK)
1081 return -ECHILD;
1083 forget_all_cached_acls(inode);
1084 return fuse_do_getattr(inode, NULL, NULL);
1088 * Check permission. The two basic access models of FUSE are:
1090 * 1) Local access checking ('default_permissions' mount option) based
1091 * on file mode. This is the plain old disk filesystem permission
1092 * modell.
1094 * 2) "Remote" access checking, where server is responsible for
1095 * checking permission in each inode operation. An exception to this
1096 * is if ->permission() was invoked from sys_access() in which case an
1097 * access request is sent. Execute permission is still checked
1098 * locally based on file mode.
1100 static int fuse_permission(struct inode *inode, int mask)
1102 struct fuse_conn *fc = get_fuse_conn(inode);
1103 bool refreshed = false;
1104 int err = 0;
1106 if (!fuse_allow_current_process(fc))
1107 return -EACCES;
1110 * If attributes are needed, refresh them before proceeding
1112 if (fc->default_permissions ||
1113 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1114 struct fuse_inode *fi = get_fuse_inode(inode);
1116 if (time_before64(fi->i_time, get_jiffies_64())) {
1117 refreshed = true;
1119 err = fuse_perm_getattr(inode, mask);
1120 if (err)
1121 return err;
1125 if (fc->default_permissions) {
1126 err = generic_permission(inode, mask);
1128 /* If permission is denied, try to refresh file
1129 attributes. This is also needed, because the root
1130 node will at first have no permissions */
1131 if (err == -EACCES && !refreshed) {
1132 err = fuse_perm_getattr(inode, mask);
1133 if (!err)
1134 err = generic_permission(inode, mask);
1137 /* Note: the opposite of the above test does not
1138 exist. So if permissions are revoked this won't be
1139 noticed immediately, only after the attribute
1140 timeout has expired */
1141 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1142 err = fuse_access(inode, mask);
1143 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1144 if (!(inode->i_mode & S_IXUGO)) {
1145 if (refreshed)
1146 return -EACCES;
1148 err = fuse_perm_getattr(inode, mask);
1149 if (!err && !(inode->i_mode & S_IXUGO))
1150 return -EACCES;
1153 return err;
1156 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1157 struct dir_context *ctx)
1159 while (nbytes >= FUSE_NAME_OFFSET) {
1160 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1161 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1162 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1163 return -EIO;
1164 if (reclen > nbytes)
1165 break;
1166 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1167 return -EIO;
1169 if (!dir_emit(ctx, dirent->name, dirent->namelen,
1170 dirent->ino, dirent->type))
1171 break;
1173 buf += reclen;
1174 nbytes -= reclen;
1175 ctx->pos = dirent->off;
1178 return 0;
1181 static int fuse_direntplus_link(struct file *file,
1182 struct fuse_direntplus *direntplus,
1183 u64 attr_version)
1185 struct fuse_entry_out *o = &direntplus->entry_out;
1186 struct fuse_dirent *dirent = &direntplus->dirent;
1187 struct dentry *parent = file->f_path.dentry;
1188 struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
1189 struct dentry *dentry;
1190 struct dentry *alias;
1191 struct inode *dir = d_inode(parent);
1192 struct fuse_conn *fc;
1193 struct inode *inode;
1194 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1196 if (!o->nodeid) {
1198 * Unlike in the case of fuse_lookup, zero nodeid does not mean
1199 * ENOENT. Instead, it only means the userspace filesystem did
1200 * not want to return attributes/handle for this entry.
1202 * So do nothing.
1204 return 0;
1207 if (name.name[0] == '.') {
1209 * We could potentially refresh the attributes of the directory
1210 * and its parent?
1212 if (name.len == 1)
1213 return 0;
1214 if (name.name[1] == '.' && name.len == 2)
1215 return 0;
1218 if (invalid_nodeid(o->nodeid))
1219 return -EIO;
1220 if (!fuse_valid_type(o->attr.mode))
1221 return -EIO;
1223 fc = get_fuse_conn(dir);
1225 name.hash = full_name_hash(parent, name.name, name.len);
1226 dentry = d_lookup(parent, &name);
1227 if (!dentry) {
1228 retry:
1229 dentry = d_alloc_parallel(parent, &name, &wq);
1230 if (IS_ERR(dentry))
1231 return PTR_ERR(dentry);
1233 if (!d_in_lookup(dentry)) {
1234 struct fuse_inode *fi;
1235 inode = d_inode(dentry);
1236 if (!inode ||
1237 get_node_id(inode) != o->nodeid ||
1238 ((o->attr.mode ^ inode->i_mode) & S_IFMT)) {
1239 d_invalidate(dentry);
1240 dput(dentry);
1241 goto retry;
1243 if (is_bad_inode(inode)) {
1244 dput(dentry);
1245 return -EIO;
1248 fi = get_fuse_inode(inode);
1249 spin_lock(&fc->lock);
1250 fi->nlookup++;
1251 spin_unlock(&fc->lock);
1253 forget_all_cached_acls(inode);
1254 fuse_change_attributes(inode, &o->attr,
1255 entry_attr_timeout(o),
1256 attr_version);
1258 * The other branch comes via fuse_iget()
1259 * which bumps nlookup inside
1261 } else {
1262 inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
1263 &o->attr, entry_attr_timeout(o),
1264 attr_version);
1265 if (!inode)
1266 inode = ERR_PTR(-ENOMEM);
1268 alias = d_splice_alias(inode, dentry);
1269 d_lookup_done(dentry);
1270 if (alias) {
1271 dput(dentry);
1272 dentry = alias;
1274 if (IS_ERR(dentry))
1275 return PTR_ERR(dentry);
1277 if (fc->readdirplus_auto)
1278 set_bit(FUSE_I_INIT_RDPLUS, &get_fuse_inode(inode)->state);
1279 fuse_change_entry_timeout(dentry, o);
1281 dput(dentry);
1282 return 0;
1285 static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
1286 struct dir_context *ctx, u64 attr_version)
1288 struct fuse_direntplus *direntplus;
1289 struct fuse_dirent *dirent;
1290 size_t reclen;
1291 int over = 0;
1292 int ret;
1294 while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
1295 direntplus = (struct fuse_direntplus *) buf;
1296 dirent = &direntplus->dirent;
1297 reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
1299 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1300 return -EIO;
1301 if (reclen > nbytes)
1302 break;
1303 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1304 return -EIO;
1306 if (!over) {
1307 /* We fill entries into dstbuf only as much as
1308 it can hold. But we still continue iterating
1309 over remaining entries to link them. If not,
1310 we need to send a FORGET for each of those
1311 which we did not link.
1313 over = !dir_emit(ctx, dirent->name, dirent->namelen,
1314 dirent->ino, dirent->type);
1315 if (!over)
1316 ctx->pos = dirent->off;
1319 buf += reclen;
1320 nbytes -= reclen;
1322 ret = fuse_direntplus_link(file, direntplus, attr_version);
1323 if (ret)
1324 fuse_force_forget(file, direntplus->entry_out.nodeid);
1327 return 0;
1330 static int fuse_readdir(struct file *file, struct dir_context *ctx)
1332 int plus, err;
1333 size_t nbytes;
1334 struct page *page;
1335 struct inode *inode = file_inode(file);
1336 struct fuse_conn *fc = get_fuse_conn(inode);
1337 struct fuse_req *req;
1338 u64 attr_version = 0;
1340 if (is_bad_inode(inode))
1341 return -EIO;
1343 req = fuse_get_req(fc, 1);
1344 if (IS_ERR(req))
1345 return PTR_ERR(req);
1347 page = alloc_page(GFP_KERNEL);
1348 if (!page) {
1349 fuse_put_request(fc, req);
1350 return -ENOMEM;
1353 plus = fuse_use_readdirplus(inode, ctx);
1354 req->out.argpages = 1;
1355 req->num_pages = 1;
1356 req->pages[0] = page;
1357 req->page_descs[0].length = PAGE_SIZE;
1358 if (plus) {
1359 attr_version = fuse_get_attr_version(fc);
1360 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1361 FUSE_READDIRPLUS);
1362 } else {
1363 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1364 FUSE_READDIR);
1366 fuse_lock_inode(inode);
1367 fuse_request_send(fc, req);
1368 fuse_unlock_inode(inode);
1369 nbytes = req->out.args[0].size;
1370 err = req->out.h.error;
1371 fuse_put_request(fc, req);
1372 if (!err) {
1373 if (plus) {
1374 err = parse_dirplusfile(page_address(page), nbytes,
1375 file, ctx,
1376 attr_version);
1377 } else {
1378 err = parse_dirfile(page_address(page), nbytes, file,
1379 ctx);
1383 __free_page(page);
1384 fuse_invalidate_atime(inode);
1385 return err;
1388 static const char *fuse_get_link(struct dentry *dentry,
1389 struct inode *inode,
1390 struct delayed_call *done)
1392 struct fuse_conn *fc = get_fuse_conn(inode);
1393 FUSE_ARGS(args);
1394 char *link;
1395 ssize_t ret;
1397 if (!dentry)
1398 return ERR_PTR(-ECHILD);
1400 link = kmalloc(PAGE_SIZE, GFP_KERNEL);
1401 if (!link)
1402 return ERR_PTR(-ENOMEM);
1404 args.in.h.opcode = FUSE_READLINK;
1405 args.in.h.nodeid = get_node_id(inode);
1406 args.out.argvar = 1;
1407 args.out.numargs = 1;
1408 args.out.args[0].size = PAGE_SIZE - 1;
1409 args.out.args[0].value = link;
1410 ret = fuse_simple_request(fc, &args);
1411 if (ret < 0) {
1412 kfree(link);
1413 link = ERR_PTR(ret);
1414 } else {
1415 link[ret] = '\0';
1416 set_delayed_call(done, kfree_link, link);
1418 fuse_invalidate_atime(inode);
1419 return link;
1422 static int fuse_dir_open(struct inode *inode, struct file *file)
1424 return fuse_open_common(inode, file, true);
1427 static int fuse_dir_release(struct inode *inode, struct file *file)
1429 fuse_release_common(file, FUSE_RELEASEDIR);
1431 return 0;
1434 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1435 int datasync)
1437 return fuse_fsync_common(file, start, end, datasync, 1);
1440 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1441 unsigned long arg)
1443 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1445 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1446 if (fc->minor < 18)
1447 return -ENOTTY;
1449 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1452 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1453 unsigned long arg)
1455 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1457 if (fc->minor < 18)
1458 return -ENOTTY;
1460 return fuse_ioctl_common(file, cmd, arg,
1461 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1464 static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
1466 /* Always update if mtime is explicitly set */
1467 if (ivalid & ATTR_MTIME_SET)
1468 return true;
1470 /* Or if kernel i_mtime is the official one */
1471 if (trust_local_mtime)
1472 return true;
1474 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1475 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1476 return false;
1478 /* In all other cases update */
1479 return true;
1482 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg,
1483 bool trust_local_cmtime)
1485 unsigned ivalid = iattr->ia_valid;
1487 if (ivalid & ATTR_MODE)
1488 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1489 if (ivalid & ATTR_UID)
1490 arg->valid |= FATTR_UID, arg->uid = from_kuid(&init_user_ns, iattr->ia_uid);
1491 if (ivalid & ATTR_GID)
1492 arg->valid |= FATTR_GID, arg->gid = from_kgid(&init_user_ns, iattr->ia_gid);
1493 if (ivalid & ATTR_SIZE)
1494 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1495 if (ivalid & ATTR_ATIME) {
1496 arg->valid |= FATTR_ATIME;
1497 arg->atime = iattr->ia_atime.tv_sec;
1498 arg->atimensec = iattr->ia_atime.tv_nsec;
1499 if (!(ivalid & ATTR_ATIME_SET))
1500 arg->valid |= FATTR_ATIME_NOW;
1502 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_cmtime)) {
1503 arg->valid |= FATTR_MTIME;
1504 arg->mtime = iattr->ia_mtime.tv_sec;
1505 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1506 if (!(ivalid & ATTR_MTIME_SET) && !trust_local_cmtime)
1507 arg->valid |= FATTR_MTIME_NOW;
1509 if ((ivalid & ATTR_CTIME) && trust_local_cmtime) {
1510 arg->valid |= FATTR_CTIME;
1511 arg->ctime = iattr->ia_ctime.tv_sec;
1512 arg->ctimensec = iattr->ia_ctime.tv_nsec;
1517 * Prevent concurrent writepages on inode
1519 * This is done by adding a negative bias to the inode write counter
1520 * and waiting for all pending writes to finish.
1522 void fuse_set_nowrite(struct inode *inode)
1524 struct fuse_conn *fc = get_fuse_conn(inode);
1525 struct fuse_inode *fi = get_fuse_inode(inode);
1527 BUG_ON(!inode_is_locked(inode));
1529 spin_lock(&fc->lock);
1530 BUG_ON(fi->writectr < 0);
1531 fi->writectr += FUSE_NOWRITE;
1532 spin_unlock(&fc->lock);
1533 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1537 * Allow writepages on inode
1539 * Remove the bias from the writecounter and send any queued
1540 * writepages.
1542 static void __fuse_release_nowrite(struct inode *inode)
1544 struct fuse_inode *fi = get_fuse_inode(inode);
1546 BUG_ON(fi->writectr != FUSE_NOWRITE);
1547 fi->writectr = 0;
1548 fuse_flush_writepages(inode);
1551 void fuse_release_nowrite(struct inode *inode)
1553 struct fuse_conn *fc = get_fuse_conn(inode);
1555 spin_lock(&fc->lock);
1556 __fuse_release_nowrite(inode);
1557 spin_unlock(&fc->lock);
1560 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_args *args,
1561 struct inode *inode,
1562 struct fuse_setattr_in *inarg_p,
1563 struct fuse_attr_out *outarg_p)
1565 args->in.h.opcode = FUSE_SETATTR;
1566 args->in.h.nodeid = get_node_id(inode);
1567 args->in.numargs = 1;
1568 args->in.args[0].size = sizeof(*inarg_p);
1569 args->in.args[0].value = inarg_p;
1570 args->out.numargs = 1;
1571 args->out.args[0].size = sizeof(*outarg_p);
1572 args->out.args[0].value = outarg_p;
1576 * Flush inode->i_mtime to the server
1578 int fuse_flush_times(struct inode *inode, struct fuse_file *ff)
1580 struct fuse_conn *fc = get_fuse_conn(inode);
1581 FUSE_ARGS(args);
1582 struct fuse_setattr_in inarg;
1583 struct fuse_attr_out outarg;
1585 memset(&inarg, 0, sizeof(inarg));
1586 memset(&outarg, 0, sizeof(outarg));
1588 inarg.valid = FATTR_MTIME;
1589 inarg.mtime = inode->i_mtime.tv_sec;
1590 inarg.mtimensec = inode->i_mtime.tv_nsec;
1591 if (fc->minor >= 23) {
1592 inarg.valid |= FATTR_CTIME;
1593 inarg.ctime = inode->i_ctime.tv_sec;
1594 inarg.ctimensec = inode->i_ctime.tv_nsec;
1596 if (ff) {
1597 inarg.valid |= FATTR_FH;
1598 inarg.fh = ff->fh;
1600 fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1602 return fuse_simple_request(fc, &args);
1606 * Set attributes, and at the same time refresh them.
1608 * Truncation is slightly complicated, because the 'truncate' request
1609 * may fail, in which case we don't want to touch the mapping.
1610 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1611 * and the actual truncation by hand.
1613 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1614 struct file *file)
1616 struct inode *inode = d_inode(dentry);
1617 struct fuse_conn *fc = get_fuse_conn(inode);
1618 struct fuse_inode *fi = get_fuse_inode(inode);
1619 FUSE_ARGS(args);
1620 struct fuse_setattr_in inarg;
1621 struct fuse_attr_out outarg;
1622 bool is_truncate = false;
1623 bool is_wb = fc->writeback_cache;
1624 loff_t oldsize;
1625 int err;
1626 bool trust_local_cmtime = is_wb && S_ISREG(inode->i_mode);
1628 if (!fc->default_permissions)
1629 attr->ia_valid |= ATTR_FORCE;
1631 err = setattr_prepare(dentry, attr);
1632 if (err)
1633 return err;
1635 if (attr->ia_valid & ATTR_OPEN) {
1636 if (fc->atomic_o_trunc)
1637 return 0;
1638 file = NULL;
1641 if (attr->ia_valid & ATTR_SIZE)
1642 is_truncate = true;
1644 if (is_truncate) {
1645 fuse_set_nowrite(inode);
1646 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1647 if (trust_local_cmtime && attr->ia_size != inode->i_size)
1648 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1651 memset(&inarg, 0, sizeof(inarg));
1652 memset(&outarg, 0, sizeof(outarg));
1653 iattr_to_fattr(attr, &inarg, trust_local_cmtime);
1654 if (file) {
1655 struct fuse_file *ff = file->private_data;
1656 inarg.valid |= FATTR_FH;
1657 inarg.fh = ff->fh;
1659 if (attr->ia_valid & ATTR_SIZE) {
1660 /* For mandatory locking in truncate */
1661 inarg.valid |= FATTR_LOCKOWNER;
1662 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1664 fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1665 err = fuse_simple_request(fc, &args);
1666 if (err) {
1667 if (err == -EINTR)
1668 fuse_invalidate_attr(inode);
1669 goto error;
1672 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1673 make_bad_inode(inode);
1674 err = -EIO;
1675 goto error;
1678 spin_lock(&fc->lock);
1679 /* the kernel maintains i_mtime locally */
1680 if (trust_local_cmtime) {
1681 if (attr->ia_valid & ATTR_MTIME)
1682 inode->i_mtime = attr->ia_mtime;
1683 if (attr->ia_valid & ATTR_CTIME)
1684 inode->i_ctime = attr->ia_ctime;
1685 /* FIXME: clear I_DIRTY_SYNC? */
1688 fuse_change_attributes_common(inode, &outarg.attr,
1689 attr_timeout(&outarg));
1690 oldsize = inode->i_size;
1691 /* see the comment in fuse_change_attributes() */
1692 if (!is_wb || is_truncate || !S_ISREG(inode->i_mode))
1693 i_size_write(inode, outarg.attr.size);
1695 if (is_truncate) {
1696 /* NOTE: this may release/reacquire fc->lock */
1697 __fuse_release_nowrite(inode);
1699 spin_unlock(&fc->lock);
1702 * Only call invalidate_inode_pages2() after removing
1703 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1705 if ((is_truncate || !is_wb) &&
1706 S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1707 truncate_pagecache(inode, outarg.attr.size);
1708 invalidate_inode_pages2(inode->i_mapping);
1711 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1712 return 0;
1714 error:
1715 if (is_truncate)
1716 fuse_release_nowrite(inode);
1718 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1719 return err;
1722 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1724 struct inode *inode = d_inode(entry);
1725 struct fuse_conn *fc = get_fuse_conn(inode);
1726 struct file *file = (attr->ia_valid & ATTR_FILE) ? attr->ia_file : NULL;
1727 int ret;
1729 if (!fuse_allow_current_process(get_fuse_conn(inode)))
1730 return -EACCES;
1732 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) {
1733 attr->ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID |
1734 ATTR_MODE);
1737 * The only sane way to reliably kill suid/sgid is to do it in
1738 * the userspace filesystem
1740 * This should be done on write(), truncate() and chown().
1742 if (!fc->handle_killpriv) {
1744 * ia_mode calculation may have used stale i_mode.
1745 * Refresh and recalculate.
1747 ret = fuse_do_getattr(inode, NULL, file);
1748 if (ret)
1749 return ret;
1751 attr->ia_mode = inode->i_mode;
1752 if (inode->i_mode & S_ISUID) {
1753 attr->ia_valid |= ATTR_MODE;
1754 attr->ia_mode &= ~S_ISUID;
1756 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1757 attr->ia_valid |= ATTR_MODE;
1758 attr->ia_mode &= ~S_ISGID;
1762 if (!attr->ia_valid)
1763 return 0;
1765 ret = fuse_do_setattr(entry, attr, file);
1766 if (!ret) {
1768 * If filesystem supports acls it may have updated acl xattrs in
1769 * the filesystem, so forget cached acls for the inode.
1771 if (fc->posix_acl)
1772 forget_all_cached_acls(inode);
1774 /* Directory mode changed, may need to revalidate access */
1775 if (d_is_dir(entry) && (attr->ia_valid & ATTR_MODE))
1776 fuse_invalidate_entry_cache(entry);
1778 return ret;
1781 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1782 struct kstat *stat)
1784 struct inode *inode = d_inode(entry);
1785 struct fuse_conn *fc = get_fuse_conn(inode);
1787 if (!fuse_allow_current_process(fc))
1788 return -EACCES;
1790 return fuse_update_attributes(inode, stat, NULL, NULL);
1793 static const struct inode_operations fuse_dir_inode_operations = {
1794 .lookup = fuse_lookup,
1795 .mkdir = fuse_mkdir,
1796 .symlink = fuse_symlink,
1797 .unlink = fuse_unlink,
1798 .rmdir = fuse_rmdir,
1799 .rename = fuse_rename2,
1800 .link = fuse_link,
1801 .setattr = fuse_setattr,
1802 .create = fuse_create,
1803 .atomic_open = fuse_atomic_open,
1804 .mknod = fuse_mknod,
1805 .permission = fuse_permission,
1806 .getattr = fuse_getattr,
1807 .listxattr = fuse_listxattr,
1808 .get_acl = fuse_get_acl,
1809 .set_acl = fuse_set_acl,
1812 static const struct file_operations fuse_dir_operations = {
1813 .llseek = generic_file_llseek,
1814 .read = generic_read_dir,
1815 .iterate_shared = fuse_readdir,
1816 .open = fuse_dir_open,
1817 .release = fuse_dir_release,
1818 .fsync = fuse_dir_fsync,
1819 .unlocked_ioctl = fuse_dir_ioctl,
1820 .compat_ioctl = fuse_dir_compat_ioctl,
1823 static const struct inode_operations fuse_common_inode_operations = {
1824 .setattr = fuse_setattr,
1825 .permission = fuse_permission,
1826 .getattr = fuse_getattr,
1827 .listxattr = fuse_listxattr,
1828 .get_acl = fuse_get_acl,
1829 .set_acl = fuse_set_acl,
1832 static const struct inode_operations fuse_symlink_inode_operations = {
1833 .setattr = fuse_setattr,
1834 .get_link = fuse_get_link,
1835 .readlink = generic_readlink,
1836 .getattr = fuse_getattr,
1837 .listxattr = fuse_listxattr,
1840 void fuse_init_common(struct inode *inode)
1842 inode->i_op = &fuse_common_inode_operations;
1845 void fuse_init_dir(struct inode *inode)
1847 inode->i_op = &fuse_dir_inode_operations;
1848 inode->i_fop = &fuse_dir_operations;
1851 void fuse_init_symlink(struct inode *inode)
1853 inode->i_op = &fuse_symlink_inode_operations;