btrfs: use file:line format for assertion report
[linux/fpc-iii.git] / fs / fuse / inode.c
blob4bb885b0f03224e9f793612ea05f74c1d9c0038e
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/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
23 #include <linux/posix_acl.h>
24 #include <linux/pid_namespace.h>
26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
27 MODULE_DESCRIPTION("Filesystem in Userspace");
28 MODULE_LICENSE("GPL");
30 static struct kmem_cache *fuse_inode_cachep;
31 struct list_head fuse_conn_list;
32 DEFINE_MUTEX(fuse_mutex);
34 static int set_global_limit(const char *val, const struct kernel_param *kp);
36 unsigned max_user_bgreq;
37 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
38 &max_user_bgreq, 0644);
39 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
40 MODULE_PARM_DESC(max_user_bgreq,
41 "Global limit for the maximum number of backgrounded requests an "
42 "unprivileged user can set");
44 unsigned max_user_congthresh;
45 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
46 &max_user_congthresh, 0644);
47 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
48 MODULE_PARM_DESC(max_user_congthresh,
49 "Global limit for the maximum congestion threshold an "
50 "unprivileged user can set");
52 #define FUSE_SUPER_MAGIC 0x65735546
54 #define FUSE_DEFAULT_BLKSIZE 512
56 /** Maximum number of outstanding background requests */
57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
59 /** Congestion starts at 75% of maximum */
60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62 struct fuse_mount_data {
63 int fd;
64 unsigned rootmode;
65 kuid_t user_id;
66 kgid_t group_id;
67 unsigned fd_present:1;
68 unsigned rootmode_present:1;
69 unsigned user_id_present:1;
70 unsigned group_id_present:1;
71 unsigned default_permissions:1;
72 unsigned allow_other:1;
73 unsigned max_read;
74 unsigned blksize;
77 struct fuse_forget_link *fuse_alloc_forget(void)
79 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
82 static struct inode *fuse_alloc_inode(struct super_block *sb)
84 struct fuse_inode *fi;
86 fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
87 if (!fi)
88 return NULL;
90 fi->i_time = 0;
91 fi->inval_mask = 0;
92 fi->nodeid = 0;
93 fi->nlookup = 0;
94 fi->attr_version = 0;
95 fi->orig_ino = 0;
96 fi->state = 0;
97 mutex_init(&fi->mutex);
98 spin_lock_init(&fi->lock);
99 fi->forget = fuse_alloc_forget();
100 if (!fi->forget) {
101 kmem_cache_free(fuse_inode_cachep, fi);
102 return NULL;
105 return &fi->inode;
108 static void fuse_free_inode(struct inode *inode)
110 struct fuse_inode *fi = get_fuse_inode(inode);
112 mutex_destroy(&fi->mutex);
113 kfree(fi->forget);
114 kmem_cache_free(fuse_inode_cachep, fi);
117 static void fuse_evict_inode(struct inode *inode)
119 struct fuse_inode *fi = get_fuse_inode(inode);
121 truncate_inode_pages_final(&inode->i_data);
122 clear_inode(inode);
123 if (inode->i_sb->s_flags & SB_ACTIVE) {
124 struct fuse_conn *fc = get_fuse_conn(inode);
125 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
126 fi->forget = NULL;
128 if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
129 WARN_ON(!list_empty(&fi->write_files));
130 WARN_ON(!list_empty(&fi->queued_writes));
134 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
136 sync_filesystem(sb);
137 if (*flags & SB_MANDLOCK)
138 return -EINVAL;
140 return 0;
144 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
145 * so that it will fit.
147 static ino_t fuse_squash_ino(u64 ino64)
149 ino_t ino = (ino_t) ino64;
150 if (sizeof(ino_t) < sizeof(u64))
151 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
152 return ino;
155 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
156 u64 attr_valid)
158 struct fuse_conn *fc = get_fuse_conn(inode);
159 struct fuse_inode *fi = get_fuse_inode(inode);
161 lockdep_assert_held(&fi->lock);
163 fi->attr_version = atomic64_inc_return(&fc->attr_version);
164 fi->i_time = attr_valid;
165 WRITE_ONCE(fi->inval_mask, 0);
167 inode->i_ino = fuse_squash_ino(attr->ino);
168 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
169 set_nlink(inode, attr->nlink);
170 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
171 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
172 inode->i_blocks = attr->blocks;
173 inode->i_atime.tv_sec = attr->atime;
174 inode->i_atime.tv_nsec = attr->atimensec;
175 /* mtime from server may be stale due to local buffered write */
176 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
177 inode->i_mtime.tv_sec = attr->mtime;
178 inode->i_mtime.tv_nsec = attr->mtimensec;
179 inode->i_ctime.tv_sec = attr->ctime;
180 inode->i_ctime.tv_nsec = attr->ctimensec;
183 if (attr->blksize != 0)
184 inode->i_blkbits = ilog2(attr->blksize);
185 else
186 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
189 * Don't set the sticky bit in i_mode, unless we want the VFS
190 * to check permissions. This prevents failures due to the
191 * check in may_delete().
193 fi->orig_i_mode = inode->i_mode;
194 if (!fc->default_permissions)
195 inode->i_mode &= ~S_ISVTX;
197 fi->orig_ino = attr->ino;
200 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
201 u64 attr_valid, u64 attr_version)
203 struct fuse_conn *fc = get_fuse_conn(inode);
204 struct fuse_inode *fi = get_fuse_inode(inode);
205 bool is_wb = fc->writeback_cache;
206 loff_t oldsize;
207 struct timespec64 old_mtime;
209 spin_lock(&fi->lock);
210 if ((attr_version != 0 && fi->attr_version > attr_version) ||
211 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
212 spin_unlock(&fi->lock);
213 return;
216 old_mtime = inode->i_mtime;
217 fuse_change_attributes_common(inode, attr, attr_valid);
219 oldsize = inode->i_size;
221 * In case of writeback_cache enabled, the cached writes beyond EOF
222 * extend local i_size without keeping userspace server in sync. So,
223 * attr->size coming from server can be stale. We cannot trust it.
225 if (!is_wb || !S_ISREG(inode->i_mode))
226 i_size_write(inode, attr->size);
227 spin_unlock(&fi->lock);
229 if (!is_wb && S_ISREG(inode->i_mode)) {
230 bool inval = false;
232 if (oldsize != attr->size) {
233 truncate_pagecache(inode, attr->size);
234 if (!fc->explicit_inval_data)
235 inval = true;
236 } else if (fc->auto_inval_data) {
237 struct timespec64 new_mtime = {
238 .tv_sec = attr->mtime,
239 .tv_nsec = attr->mtimensec,
243 * Auto inval mode also checks and invalidates if mtime
244 * has changed.
246 if (!timespec64_equal(&old_mtime, &new_mtime))
247 inval = true;
250 if (inval)
251 invalidate_inode_pages2(inode->i_mapping);
255 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
257 inode->i_mode = attr->mode & S_IFMT;
258 inode->i_size = attr->size;
259 inode->i_mtime.tv_sec = attr->mtime;
260 inode->i_mtime.tv_nsec = attr->mtimensec;
261 inode->i_ctime.tv_sec = attr->ctime;
262 inode->i_ctime.tv_nsec = attr->ctimensec;
263 if (S_ISREG(inode->i_mode)) {
264 fuse_init_common(inode);
265 fuse_init_file_inode(inode);
266 } else if (S_ISDIR(inode->i_mode))
267 fuse_init_dir(inode);
268 else if (S_ISLNK(inode->i_mode))
269 fuse_init_symlink(inode);
270 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
271 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
272 fuse_init_common(inode);
273 init_special_inode(inode, inode->i_mode,
274 new_decode_dev(attr->rdev));
275 } else
276 BUG();
279 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
281 u64 nodeid = *(u64 *) _nodeidp;
282 if (get_node_id(inode) == nodeid)
283 return 1;
284 else
285 return 0;
288 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
290 u64 nodeid = *(u64 *) _nodeidp;
291 get_fuse_inode(inode)->nodeid = nodeid;
292 return 0;
295 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
296 int generation, struct fuse_attr *attr,
297 u64 attr_valid, u64 attr_version)
299 struct inode *inode;
300 struct fuse_inode *fi;
301 struct fuse_conn *fc = get_fuse_conn_super(sb);
303 retry:
304 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
305 if (!inode)
306 return NULL;
308 if ((inode->i_state & I_NEW)) {
309 inode->i_flags |= S_NOATIME;
310 if (!fc->writeback_cache || !S_ISREG(attr->mode))
311 inode->i_flags |= S_NOCMTIME;
312 inode->i_generation = generation;
313 fuse_init_inode(inode, attr);
314 unlock_new_inode(inode);
315 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
316 /* Inode has changed type, any I/O on the old should fail */
317 make_bad_inode(inode);
318 iput(inode);
319 goto retry;
322 fi = get_fuse_inode(inode);
323 spin_lock(&fi->lock);
324 fi->nlookup++;
325 spin_unlock(&fi->lock);
326 fuse_change_attributes(inode, attr, attr_valid, attr_version);
328 return inode;
331 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
332 loff_t offset, loff_t len)
334 struct inode *inode;
335 pgoff_t pg_start;
336 pgoff_t pg_end;
338 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
339 if (!inode)
340 return -ENOENT;
342 fuse_invalidate_attr(inode);
343 forget_all_cached_acls(inode);
344 if (offset >= 0) {
345 pg_start = offset >> PAGE_SHIFT;
346 if (len <= 0)
347 pg_end = -1;
348 else
349 pg_end = (offset + len - 1) >> PAGE_SHIFT;
350 invalidate_inode_pages2_range(inode->i_mapping,
351 pg_start, pg_end);
353 iput(inode);
354 return 0;
357 bool fuse_lock_inode(struct inode *inode)
359 bool locked = false;
361 if (!get_fuse_conn(inode)->parallel_dirops) {
362 mutex_lock(&get_fuse_inode(inode)->mutex);
363 locked = true;
366 return locked;
369 void fuse_unlock_inode(struct inode *inode, bool locked)
371 if (locked)
372 mutex_unlock(&get_fuse_inode(inode)->mutex);
375 static void fuse_umount_begin(struct super_block *sb)
377 fuse_abort_conn(get_fuse_conn_super(sb));
380 static void fuse_send_destroy(struct fuse_conn *fc)
382 struct fuse_req *req = fc->destroy_req;
383 if (req && fc->conn_init) {
384 fc->destroy_req = NULL;
385 req->in.h.opcode = FUSE_DESTROY;
386 __set_bit(FR_FORCE, &req->flags);
387 __clear_bit(FR_BACKGROUND, &req->flags);
388 fuse_request_send(fc, req);
389 fuse_put_request(fc, req);
393 static void fuse_put_super(struct super_block *sb)
395 struct fuse_conn *fc = get_fuse_conn_super(sb);
397 mutex_lock(&fuse_mutex);
398 list_del(&fc->entry);
399 fuse_ctl_remove_conn(fc);
400 mutex_unlock(&fuse_mutex);
402 fuse_conn_put(fc);
405 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
407 stbuf->f_type = FUSE_SUPER_MAGIC;
408 stbuf->f_bsize = attr->bsize;
409 stbuf->f_frsize = attr->frsize;
410 stbuf->f_blocks = attr->blocks;
411 stbuf->f_bfree = attr->bfree;
412 stbuf->f_bavail = attr->bavail;
413 stbuf->f_files = attr->files;
414 stbuf->f_ffree = attr->ffree;
415 stbuf->f_namelen = attr->namelen;
416 /* fsid is left zero */
419 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
421 struct super_block *sb = dentry->d_sb;
422 struct fuse_conn *fc = get_fuse_conn_super(sb);
423 FUSE_ARGS(args);
424 struct fuse_statfs_out outarg;
425 int err;
427 if (!fuse_allow_current_process(fc)) {
428 buf->f_type = FUSE_SUPER_MAGIC;
429 return 0;
432 memset(&outarg, 0, sizeof(outarg));
433 args.in.numargs = 0;
434 args.in.h.opcode = FUSE_STATFS;
435 args.in.h.nodeid = get_node_id(d_inode(dentry));
436 args.out.numargs = 1;
437 args.out.args[0].size = sizeof(outarg);
438 args.out.args[0].value = &outarg;
439 err = fuse_simple_request(fc, &args);
440 if (!err)
441 convert_fuse_statfs(buf, &outarg.st);
442 return err;
445 enum {
446 OPT_FD,
447 OPT_ROOTMODE,
448 OPT_USER_ID,
449 OPT_GROUP_ID,
450 OPT_DEFAULT_PERMISSIONS,
451 OPT_ALLOW_OTHER,
452 OPT_MAX_READ,
453 OPT_BLKSIZE,
454 OPT_ERR
457 static const match_table_t tokens = {
458 {OPT_FD, "fd=%u"},
459 {OPT_ROOTMODE, "rootmode=%o"},
460 {OPT_USER_ID, "user_id=%u"},
461 {OPT_GROUP_ID, "group_id=%u"},
462 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
463 {OPT_ALLOW_OTHER, "allow_other"},
464 {OPT_MAX_READ, "max_read=%u"},
465 {OPT_BLKSIZE, "blksize=%u"},
466 {OPT_ERR, NULL}
469 static int fuse_match_uint(substring_t *s, unsigned int *res)
471 int err = -ENOMEM;
472 char *buf = match_strdup(s);
473 if (buf) {
474 err = kstrtouint(buf, 10, res);
475 kfree(buf);
477 return err;
480 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
481 struct user_namespace *user_ns)
483 char *p;
484 memset(d, 0, sizeof(struct fuse_mount_data));
485 d->max_read = ~0;
486 d->blksize = FUSE_DEFAULT_BLKSIZE;
488 while ((p = strsep(&opt, ",")) != NULL) {
489 int token;
490 int value;
491 unsigned uv;
492 substring_t args[MAX_OPT_ARGS];
493 if (!*p)
494 continue;
496 token = match_token(p, tokens, args);
497 switch (token) {
498 case OPT_FD:
499 if (match_int(&args[0], &value))
500 return 0;
501 d->fd = value;
502 d->fd_present = 1;
503 break;
505 case OPT_ROOTMODE:
506 if (match_octal(&args[0], &value))
507 return 0;
508 if (!fuse_valid_type(value))
509 return 0;
510 d->rootmode = value;
511 d->rootmode_present = 1;
512 break;
514 case OPT_USER_ID:
515 if (fuse_match_uint(&args[0], &uv))
516 return 0;
517 d->user_id = make_kuid(user_ns, uv);
518 if (!uid_valid(d->user_id))
519 return 0;
520 d->user_id_present = 1;
521 break;
523 case OPT_GROUP_ID:
524 if (fuse_match_uint(&args[0], &uv))
525 return 0;
526 d->group_id = make_kgid(user_ns, uv);
527 if (!gid_valid(d->group_id))
528 return 0;
529 d->group_id_present = 1;
530 break;
532 case OPT_DEFAULT_PERMISSIONS:
533 d->default_permissions = 1;
534 break;
536 case OPT_ALLOW_OTHER:
537 d->allow_other = 1;
538 break;
540 case OPT_MAX_READ:
541 if (match_int(&args[0], &value))
542 return 0;
543 d->max_read = value;
544 break;
546 case OPT_BLKSIZE:
547 if (!is_bdev || match_int(&args[0], &value))
548 return 0;
549 d->blksize = value;
550 break;
552 default:
553 return 0;
557 if (!d->fd_present || !d->rootmode_present ||
558 !d->user_id_present || !d->group_id_present)
559 return 0;
561 return 1;
564 static int fuse_show_options(struct seq_file *m, struct dentry *root)
566 struct super_block *sb = root->d_sb;
567 struct fuse_conn *fc = get_fuse_conn_super(sb);
569 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
570 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
571 if (fc->default_permissions)
572 seq_puts(m, ",default_permissions");
573 if (fc->allow_other)
574 seq_puts(m, ",allow_other");
575 if (fc->max_read != ~0)
576 seq_printf(m, ",max_read=%u", fc->max_read);
577 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
578 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
579 return 0;
582 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
584 memset(fiq, 0, sizeof(struct fuse_iqueue));
585 init_waitqueue_head(&fiq->waitq);
586 INIT_LIST_HEAD(&fiq->pending);
587 INIT_LIST_HEAD(&fiq->interrupts);
588 fiq->forget_list_tail = &fiq->forget_list_head;
589 fiq->connected = 1;
592 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
594 unsigned int i;
596 spin_lock_init(&fpq->lock);
597 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
598 INIT_LIST_HEAD(&fpq->processing[i]);
599 INIT_LIST_HEAD(&fpq->io);
600 fpq->connected = 1;
603 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
605 memset(fc, 0, sizeof(*fc));
606 spin_lock_init(&fc->lock);
607 spin_lock_init(&fc->bg_lock);
608 init_rwsem(&fc->killsb);
609 refcount_set(&fc->count, 1);
610 atomic_set(&fc->dev_count, 1);
611 init_waitqueue_head(&fc->blocked_waitq);
612 init_waitqueue_head(&fc->reserved_req_waitq);
613 fuse_iqueue_init(&fc->iq);
614 INIT_LIST_HEAD(&fc->bg_queue);
615 INIT_LIST_HEAD(&fc->entry);
616 INIT_LIST_HEAD(&fc->devices);
617 atomic_set(&fc->num_waiting, 0);
618 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
619 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
620 atomic64_set(&fc->khctr, 0);
621 fc->polled_files = RB_ROOT;
622 fc->blocked = 0;
623 fc->initialized = 0;
624 fc->connected = 1;
625 atomic64_set(&fc->attr_version, 1);
626 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
627 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
628 fc->user_ns = get_user_ns(user_ns);
629 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
631 EXPORT_SYMBOL_GPL(fuse_conn_init);
633 void fuse_conn_put(struct fuse_conn *fc)
635 if (refcount_dec_and_test(&fc->count)) {
636 if (fc->destroy_req)
637 fuse_request_free(fc->destroy_req);
638 put_pid_ns(fc->pid_ns);
639 put_user_ns(fc->user_ns);
640 fc->release(fc);
643 EXPORT_SYMBOL_GPL(fuse_conn_put);
645 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
647 refcount_inc(&fc->count);
648 return fc;
650 EXPORT_SYMBOL_GPL(fuse_conn_get);
652 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
654 struct fuse_attr attr;
655 memset(&attr, 0, sizeof(attr));
657 attr.mode = mode;
658 attr.ino = FUSE_ROOT_ID;
659 attr.nlink = 1;
660 return fuse_iget(sb, 1, 0, &attr, 0, 0);
663 struct fuse_inode_handle {
664 u64 nodeid;
665 u32 generation;
668 static struct dentry *fuse_get_dentry(struct super_block *sb,
669 struct fuse_inode_handle *handle)
671 struct fuse_conn *fc = get_fuse_conn_super(sb);
672 struct inode *inode;
673 struct dentry *entry;
674 int err = -ESTALE;
676 if (handle->nodeid == 0)
677 goto out_err;
679 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
680 if (!inode) {
681 struct fuse_entry_out outarg;
682 const struct qstr name = QSTR_INIT(".", 1);
684 if (!fc->export_support)
685 goto out_err;
687 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
688 &inode);
689 if (err && err != -ENOENT)
690 goto out_err;
691 if (err || !inode) {
692 err = -ESTALE;
693 goto out_err;
695 err = -EIO;
696 if (get_node_id(inode) != handle->nodeid)
697 goto out_iput;
699 err = -ESTALE;
700 if (inode->i_generation != handle->generation)
701 goto out_iput;
703 entry = d_obtain_alias(inode);
704 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
705 fuse_invalidate_entry_cache(entry);
707 return entry;
709 out_iput:
710 iput(inode);
711 out_err:
712 return ERR_PTR(err);
715 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
716 struct inode *parent)
718 int len = parent ? 6 : 3;
719 u64 nodeid;
720 u32 generation;
722 if (*max_len < len) {
723 *max_len = len;
724 return FILEID_INVALID;
727 nodeid = get_fuse_inode(inode)->nodeid;
728 generation = inode->i_generation;
730 fh[0] = (u32)(nodeid >> 32);
731 fh[1] = (u32)(nodeid & 0xffffffff);
732 fh[2] = generation;
734 if (parent) {
735 nodeid = get_fuse_inode(parent)->nodeid;
736 generation = parent->i_generation;
738 fh[3] = (u32)(nodeid >> 32);
739 fh[4] = (u32)(nodeid & 0xffffffff);
740 fh[5] = generation;
743 *max_len = len;
744 return parent ? 0x82 : 0x81;
747 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
748 struct fid *fid, int fh_len, int fh_type)
750 struct fuse_inode_handle handle;
752 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
753 return NULL;
755 handle.nodeid = (u64) fid->raw[0] << 32;
756 handle.nodeid |= (u64) fid->raw[1];
757 handle.generation = fid->raw[2];
758 return fuse_get_dentry(sb, &handle);
761 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
762 struct fid *fid, int fh_len, int fh_type)
764 struct fuse_inode_handle parent;
766 if (fh_type != 0x82 || fh_len < 6)
767 return NULL;
769 parent.nodeid = (u64) fid->raw[3] << 32;
770 parent.nodeid |= (u64) fid->raw[4];
771 parent.generation = fid->raw[5];
772 return fuse_get_dentry(sb, &parent);
775 static struct dentry *fuse_get_parent(struct dentry *child)
777 struct inode *child_inode = d_inode(child);
778 struct fuse_conn *fc = get_fuse_conn(child_inode);
779 struct inode *inode;
780 struct dentry *parent;
781 struct fuse_entry_out outarg;
782 const struct qstr name = QSTR_INIT("..", 2);
783 int err;
785 if (!fc->export_support)
786 return ERR_PTR(-ESTALE);
788 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
789 &name, &outarg, &inode);
790 if (err) {
791 if (err == -ENOENT)
792 return ERR_PTR(-ESTALE);
793 return ERR_PTR(err);
796 parent = d_obtain_alias(inode);
797 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
798 fuse_invalidate_entry_cache(parent);
800 return parent;
803 static const struct export_operations fuse_export_operations = {
804 .fh_to_dentry = fuse_fh_to_dentry,
805 .fh_to_parent = fuse_fh_to_parent,
806 .encode_fh = fuse_encode_fh,
807 .get_parent = fuse_get_parent,
810 static const struct super_operations fuse_super_operations = {
811 .alloc_inode = fuse_alloc_inode,
812 .free_inode = fuse_free_inode,
813 .evict_inode = fuse_evict_inode,
814 .write_inode = fuse_write_inode,
815 .drop_inode = generic_delete_inode,
816 .remount_fs = fuse_remount_fs,
817 .put_super = fuse_put_super,
818 .umount_begin = fuse_umount_begin,
819 .statfs = fuse_statfs,
820 .show_options = fuse_show_options,
823 static void sanitize_global_limit(unsigned *limit)
825 if (*limit == 0)
826 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) /
827 sizeof(struct fuse_req);
829 if (*limit >= 1 << 16)
830 *limit = (1 << 16) - 1;
833 static int set_global_limit(const char *val, const struct kernel_param *kp)
835 int rv;
837 rv = param_set_uint(val, kp);
838 if (rv)
839 return rv;
841 sanitize_global_limit((unsigned *)kp->arg);
843 return 0;
846 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
848 int cap_sys_admin = capable(CAP_SYS_ADMIN);
850 if (arg->minor < 13)
851 return;
853 sanitize_global_limit(&max_user_bgreq);
854 sanitize_global_limit(&max_user_congthresh);
856 spin_lock(&fc->bg_lock);
857 if (arg->max_background) {
858 fc->max_background = arg->max_background;
860 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
861 fc->max_background = max_user_bgreq;
863 if (arg->congestion_threshold) {
864 fc->congestion_threshold = arg->congestion_threshold;
866 if (!cap_sys_admin &&
867 fc->congestion_threshold > max_user_congthresh)
868 fc->congestion_threshold = max_user_congthresh;
870 spin_unlock(&fc->bg_lock);
873 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
875 struct fuse_init_out *arg = &req->misc.init_out;
877 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
878 fc->conn_error = 1;
879 else {
880 unsigned long ra_pages;
882 process_init_limits(fc, arg);
884 if (arg->minor >= 6) {
885 ra_pages = arg->max_readahead / PAGE_SIZE;
886 if (arg->flags & FUSE_ASYNC_READ)
887 fc->async_read = 1;
888 if (!(arg->flags & FUSE_POSIX_LOCKS))
889 fc->no_lock = 1;
890 if (arg->minor >= 17) {
891 if (!(arg->flags & FUSE_FLOCK_LOCKS))
892 fc->no_flock = 1;
893 } else {
894 if (!(arg->flags & FUSE_POSIX_LOCKS))
895 fc->no_flock = 1;
897 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
898 fc->atomic_o_trunc = 1;
899 if (arg->minor >= 9) {
900 /* LOOKUP has dependency on proto version */
901 if (arg->flags & FUSE_EXPORT_SUPPORT)
902 fc->export_support = 1;
904 if (arg->flags & FUSE_BIG_WRITES)
905 fc->big_writes = 1;
906 if (arg->flags & FUSE_DONT_MASK)
907 fc->dont_mask = 1;
908 if (arg->flags & FUSE_AUTO_INVAL_DATA)
909 fc->auto_inval_data = 1;
910 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
911 fc->explicit_inval_data = 1;
912 if (arg->flags & FUSE_DO_READDIRPLUS) {
913 fc->do_readdirplus = 1;
914 if (arg->flags & FUSE_READDIRPLUS_AUTO)
915 fc->readdirplus_auto = 1;
917 if (arg->flags & FUSE_ASYNC_DIO)
918 fc->async_dio = 1;
919 if (arg->flags & FUSE_WRITEBACK_CACHE)
920 fc->writeback_cache = 1;
921 if (arg->flags & FUSE_PARALLEL_DIROPS)
922 fc->parallel_dirops = 1;
923 if (arg->flags & FUSE_HANDLE_KILLPRIV)
924 fc->handle_killpriv = 1;
925 if (arg->time_gran && arg->time_gran <= 1000000000)
926 fc->sb->s_time_gran = arg->time_gran;
927 if ((arg->flags & FUSE_POSIX_ACL)) {
928 fc->default_permissions = 1;
929 fc->posix_acl = 1;
930 fc->sb->s_xattr = fuse_acl_xattr_handlers;
932 if (arg->flags & FUSE_CACHE_SYMLINKS)
933 fc->cache_symlinks = 1;
934 if (arg->flags & FUSE_ABORT_ERROR)
935 fc->abort_err = 1;
936 if (arg->flags & FUSE_MAX_PAGES) {
937 fc->max_pages =
938 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
939 max_t(unsigned int, arg->max_pages, 1));
941 } else {
942 ra_pages = fc->max_read / PAGE_SIZE;
943 fc->no_lock = 1;
944 fc->no_flock = 1;
947 fc->sb->s_bdi->ra_pages =
948 min(fc->sb->s_bdi->ra_pages, ra_pages);
949 fc->minor = arg->minor;
950 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
951 fc->max_write = max_t(unsigned, 4096, fc->max_write);
952 fc->conn_init = 1;
954 fuse_set_initialized(fc);
955 wake_up_all(&fc->blocked_waitq);
958 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
960 struct fuse_init_in *arg = &req->misc.init_in;
962 arg->major = FUSE_KERNEL_VERSION;
963 arg->minor = FUSE_KERNEL_MINOR_VERSION;
964 arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
965 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
966 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
967 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
968 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
969 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
970 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
971 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
972 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
973 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
974 req->in.h.opcode = FUSE_INIT;
975 req->in.numargs = 1;
976 req->in.args[0].size = sizeof(*arg);
977 req->in.args[0].value = arg;
978 req->out.numargs = 1;
979 /* Variable length argument used for backward compatibility
980 with interface version < 7.5. Rest of init_out is zeroed
981 by do_get_request(), so a short reply is not a problem */
982 req->out.argvar = 1;
983 req->out.args[0].size = sizeof(struct fuse_init_out);
984 req->out.args[0].value = &req->misc.init_out;
985 req->end = process_init_reply;
986 fuse_request_send_background(fc, req);
989 static void fuse_free_conn(struct fuse_conn *fc)
991 WARN_ON(!list_empty(&fc->devices));
992 kfree_rcu(fc, rcu);
995 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
997 int err;
998 char *suffix = "";
1000 if (sb->s_bdev) {
1001 suffix = "-fuseblk";
1003 * sb->s_bdi points to blkdev's bdi however we want to redirect
1004 * it to our private bdi...
1006 bdi_put(sb->s_bdi);
1007 sb->s_bdi = &noop_backing_dev_info;
1009 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1010 MINOR(fc->dev), suffix);
1011 if (err)
1012 return err;
1014 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1015 /* fuse does it's own writeback accounting */
1016 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1019 * For a single fuse filesystem use max 1% of dirty +
1020 * writeback threshold.
1022 * This gives about 1M of write buffer for memory maps on a
1023 * machine with 1G and 10% dirty_ratio, which should be more
1024 * than enough.
1026 * Privileged users can raise it by writing to
1028 * /sys/class/bdi/<bdi>/max_ratio
1030 bdi_set_max_ratio(sb->s_bdi, 1);
1032 return 0;
1035 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1037 struct fuse_dev *fud;
1038 struct list_head *pq;
1040 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1041 if (!fud)
1042 return NULL;
1044 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1045 if (!pq) {
1046 kfree(fud);
1047 return NULL;
1050 fud->pq.processing = pq;
1051 fud->fc = fuse_conn_get(fc);
1052 fuse_pqueue_init(&fud->pq);
1054 spin_lock(&fc->lock);
1055 list_add_tail(&fud->entry, &fc->devices);
1056 spin_unlock(&fc->lock);
1058 return fud;
1060 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1062 void fuse_dev_free(struct fuse_dev *fud)
1064 struct fuse_conn *fc = fud->fc;
1066 if (fc) {
1067 spin_lock(&fc->lock);
1068 list_del(&fud->entry);
1069 spin_unlock(&fc->lock);
1071 fuse_conn_put(fc);
1073 kfree(fud->pq.processing);
1074 kfree(fud);
1076 EXPORT_SYMBOL_GPL(fuse_dev_free);
1078 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1080 struct fuse_dev *fud;
1081 struct fuse_conn *fc;
1082 struct inode *root;
1083 struct fuse_mount_data d;
1084 struct file *file;
1085 struct dentry *root_dentry;
1086 struct fuse_req *init_req;
1087 int err;
1088 int is_bdev = sb->s_bdev != NULL;
1090 err = -EINVAL;
1091 if (sb->s_flags & SB_MANDLOCK)
1092 goto err;
1094 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1096 if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1097 goto err;
1099 if (is_bdev) {
1100 #ifdef CONFIG_BLOCK
1101 err = -EINVAL;
1102 if (!sb_set_blocksize(sb, d.blksize))
1103 goto err;
1104 #endif
1105 } else {
1106 sb->s_blocksize = PAGE_SIZE;
1107 sb->s_blocksize_bits = PAGE_SHIFT;
1109 sb->s_magic = FUSE_SUPER_MAGIC;
1110 sb->s_op = &fuse_super_operations;
1111 sb->s_xattr = fuse_xattr_handlers;
1112 sb->s_maxbytes = MAX_LFS_FILESIZE;
1113 sb->s_time_gran = 1;
1114 sb->s_export_op = &fuse_export_operations;
1115 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1116 if (sb->s_user_ns != &init_user_ns)
1117 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1119 file = fget(d.fd);
1120 err = -EINVAL;
1121 if (!file)
1122 goto err;
1125 * Require mount to happen from the same user namespace which
1126 * opened /dev/fuse to prevent potential attacks.
1128 if (file->f_op != &fuse_dev_operations ||
1129 file->f_cred->user_ns != sb->s_user_ns)
1130 goto err_fput;
1133 * If we are not in the initial user namespace posix
1134 * acls must be translated.
1136 if (sb->s_user_ns != &init_user_ns)
1137 sb->s_xattr = fuse_no_acl_xattr_handlers;
1139 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1140 err = -ENOMEM;
1141 if (!fc)
1142 goto err_fput;
1144 fuse_conn_init(fc, sb->s_user_ns);
1145 fc->release = fuse_free_conn;
1147 fud = fuse_dev_alloc(fc);
1148 if (!fud)
1149 goto err_put_conn;
1151 fc->dev = sb->s_dev;
1152 fc->sb = sb;
1153 err = fuse_bdi_init(fc, sb);
1154 if (err)
1155 goto err_dev_free;
1157 /* Handle umasking inside the fuse code */
1158 if (sb->s_flags & SB_POSIXACL)
1159 fc->dont_mask = 1;
1160 sb->s_flags |= SB_POSIXACL;
1162 fc->default_permissions = d.default_permissions;
1163 fc->allow_other = d.allow_other;
1164 fc->user_id = d.user_id;
1165 fc->group_id = d.group_id;
1166 fc->max_read = max_t(unsigned, 4096, d.max_read);
1168 /* Used by get_root_inode() */
1169 sb->s_fs_info = fc;
1171 err = -ENOMEM;
1172 root = fuse_get_root_inode(sb, d.rootmode);
1173 sb->s_d_op = &fuse_root_dentry_operations;
1174 root_dentry = d_make_root(root);
1175 if (!root_dentry)
1176 goto err_dev_free;
1177 /* Root dentry doesn't have .d_revalidate */
1178 sb->s_d_op = &fuse_dentry_operations;
1180 init_req = fuse_request_alloc(0);
1181 if (!init_req)
1182 goto err_put_root;
1183 __set_bit(FR_BACKGROUND, &init_req->flags);
1185 if (is_bdev) {
1186 fc->destroy_req = fuse_request_alloc(0);
1187 if (!fc->destroy_req)
1188 goto err_free_init_req;
1191 mutex_lock(&fuse_mutex);
1192 err = -EINVAL;
1193 if (file->private_data)
1194 goto err_unlock;
1196 err = fuse_ctl_add_conn(fc);
1197 if (err)
1198 goto err_unlock;
1200 list_add_tail(&fc->entry, &fuse_conn_list);
1201 sb->s_root = root_dentry;
1202 file->private_data = fud;
1203 mutex_unlock(&fuse_mutex);
1205 * atomic_dec_and_test() in fput() provides the necessary
1206 * memory barrier for file->private_data to be visible on all
1207 * CPUs after this
1209 fput(file);
1211 fuse_send_init(fc, init_req);
1213 return 0;
1215 err_unlock:
1216 mutex_unlock(&fuse_mutex);
1217 err_free_init_req:
1218 fuse_request_free(init_req);
1219 err_put_root:
1220 dput(root_dentry);
1221 err_dev_free:
1222 fuse_dev_free(fud);
1223 err_put_conn:
1224 fuse_conn_put(fc);
1225 sb->s_fs_info = NULL;
1226 err_fput:
1227 fput(file);
1228 err:
1229 return err;
1232 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1233 int flags, const char *dev_name,
1234 void *raw_data)
1236 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1239 static void fuse_sb_destroy(struct super_block *sb)
1241 struct fuse_conn *fc = get_fuse_conn_super(sb);
1243 if (fc) {
1244 fuse_send_destroy(fc);
1246 fuse_abort_conn(fc);
1247 fuse_wait_aborted(fc);
1249 down_write(&fc->killsb);
1250 fc->sb = NULL;
1251 up_write(&fc->killsb);
1255 static void fuse_kill_sb_anon(struct super_block *sb)
1257 fuse_sb_destroy(sb);
1258 kill_anon_super(sb);
1261 static struct file_system_type fuse_fs_type = {
1262 .owner = THIS_MODULE,
1263 .name = "fuse",
1264 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1265 .mount = fuse_mount,
1266 .kill_sb = fuse_kill_sb_anon,
1268 MODULE_ALIAS_FS("fuse");
1270 #ifdef CONFIG_BLOCK
1271 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1272 int flags, const char *dev_name,
1273 void *raw_data)
1275 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1278 static void fuse_kill_sb_blk(struct super_block *sb)
1280 fuse_sb_destroy(sb);
1281 kill_block_super(sb);
1284 static struct file_system_type fuseblk_fs_type = {
1285 .owner = THIS_MODULE,
1286 .name = "fuseblk",
1287 .mount = fuse_mount_blk,
1288 .kill_sb = fuse_kill_sb_blk,
1289 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1291 MODULE_ALIAS_FS("fuseblk");
1293 static inline int register_fuseblk(void)
1295 return register_filesystem(&fuseblk_fs_type);
1298 static inline void unregister_fuseblk(void)
1300 unregister_filesystem(&fuseblk_fs_type);
1302 #else
1303 static inline int register_fuseblk(void)
1305 return 0;
1308 static inline void unregister_fuseblk(void)
1311 #endif
1313 static void fuse_inode_init_once(void *foo)
1315 struct inode *inode = foo;
1317 inode_init_once(inode);
1320 static int __init fuse_fs_init(void)
1322 int err;
1324 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1325 sizeof(struct fuse_inode), 0,
1326 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1327 fuse_inode_init_once);
1328 err = -ENOMEM;
1329 if (!fuse_inode_cachep)
1330 goto out;
1332 err = register_fuseblk();
1333 if (err)
1334 goto out2;
1336 err = register_filesystem(&fuse_fs_type);
1337 if (err)
1338 goto out3;
1340 return 0;
1342 out3:
1343 unregister_fuseblk();
1344 out2:
1345 kmem_cache_destroy(fuse_inode_cachep);
1346 out:
1347 return err;
1350 static void fuse_fs_cleanup(void)
1352 unregister_filesystem(&fuse_fs_type);
1353 unregister_fuseblk();
1356 * Make sure all delayed rcu free inodes are flushed before we
1357 * destroy cache.
1359 rcu_barrier();
1360 kmem_cache_destroy(fuse_inode_cachep);
1363 static struct kobject *fuse_kobj;
1365 static int fuse_sysfs_init(void)
1367 int err;
1369 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1370 if (!fuse_kobj) {
1371 err = -ENOMEM;
1372 goto out_err;
1375 err = sysfs_create_mount_point(fuse_kobj, "connections");
1376 if (err)
1377 goto out_fuse_unregister;
1379 return 0;
1381 out_fuse_unregister:
1382 kobject_put(fuse_kobj);
1383 out_err:
1384 return err;
1387 static void fuse_sysfs_cleanup(void)
1389 sysfs_remove_mount_point(fuse_kobj, "connections");
1390 kobject_put(fuse_kobj);
1393 static int __init fuse_init(void)
1395 int res;
1397 pr_info("init (API version %i.%i)\n",
1398 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1400 INIT_LIST_HEAD(&fuse_conn_list);
1401 res = fuse_fs_init();
1402 if (res)
1403 goto err;
1405 res = fuse_dev_init();
1406 if (res)
1407 goto err_fs_cleanup;
1409 res = fuse_sysfs_init();
1410 if (res)
1411 goto err_dev_cleanup;
1413 res = fuse_ctl_init();
1414 if (res)
1415 goto err_sysfs_cleanup;
1417 sanitize_global_limit(&max_user_bgreq);
1418 sanitize_global_limit(&max_user_congthresh);
1420 return 0;
1422 err_sysfs_cleanup:
1423 fuse_sysfs_cleanup();
1424 err_dev_cleanup:
1425 fuse_dev_cleanup();
1426 err_fs_cleanup:
1427 fuse_fs_cleanup();
1428 err:
1429 return res;
1432 static void __exit fuse_exit(void)
1434 pr_debug("exit\n");
1436 fuse_ctl_cleanup();
1437 fuse_sysfs_cleanup();
1438 fuse_fs_cleanup();
1439 fuse_dev_cleanup();
1442 module_init(fuse_init);
1443 module_exit(fuse_exit);