projects / deliverable / linux.git / blob
? search:


8373f59dc2a88947e9e3c603018454002d16c4f1
[deliverable/linux.git] / fs / fuse / inode.c
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
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
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
27
28 static struct kmem_cache *fuse_inode_cachep;
29 struct list_head fuse_conn_list;
30 DEFINE_MUTEX(fuse_mutex);
31
32 static int set_global_limit(const char *val, struct kernel_param *kp);
33
34 unsigned max_user_bgreq;
35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
36 &max_user_bgreq, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
38 MODULE_PARM_DESC(max_user_bgreq,
39 "Global limit for the maximum number of backgrounded requests an "
40 "unprivileged user can set");
41
42 unsigned max_user_congthresh;
43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
44 &max_user_congthresh, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
46 MODULE_PARM_DESC(max_user_congthresh,
47 "Global limit for the maximum congestion threshold an "
48 "unprivileged user can set");
49
50 #define FUSE_SUPER_MAGIC 0x65735546
51
52 #define FUSE_DEFAULT_BLKSIZE 512
53
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
56
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
59
60 struct fuse_mount_data {
61 int fd;
62 unsigned rootmode;
63 kuid_t user_id;
64 kgid_t group_id;
65 unsigned fd_present:1;
66 unsigned rootmode_present:1;
67 unsigned user_id_present:1;
68 unsigned group_id_present:1;
69 unsigned flags;
70 unsigned max_read;
71 unsigned blksize;
72 };
73
74 struct fuse_forget_link *fuse_alloc_forget(void)
75 {
76 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
77 }
78
79 static struct inode *fuse_alloc_inode(struct super_block *sb)
80 {
81 struct inode *inode;
82 struct fuse_inode *fi;
83
84 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
85 if (!inode)
86 return NULL;
87
88 fi = get_fuse_inode(inode);
89 fi->i_time = 0;
90 fi->nodeid = 0;
91 fi->nlookup = 0;
92 fi->attr_version = 0;
93 fi->writectr = 0;
94 fi->orig_ino = 0;
95 fi->state = 0;
96 INIT_LIST_HEAD(&fi->write_files);
97 INIT_LIST_HEAD(&fi->queued_writes);
98 INIT_LIST_HEAD(&fi->writepages);
99 init_waitqueue_head(&fi->page_waitq);
100 fi->forget = fuse_alloc_forget();
101 if (!fi->forget) {
102 kmem_cache_free(fuse_inode_cachep, inode);
103 return NULL;
104 }
105
106 return inode;
107 }
108
109 static void fuse_i_callback(struct rcu_head *head)
110 {
111 struct inode *inode = container_of(head, struct inode, i_rcu);
112 kmem_cache_free(fuse_inode_cachep, inode);
113 }
114
115 static void fuse_destroy_inode(struct inode *inode)
116 {
117 struct fuse_inode *fi = get_fuse_inode(inode);
118 BUG_ON(!list_empty(&fi->write_files));
119 BUG_ON(!list_empty(&fi->queued_writes));
120 kfree(fi->forget);
121 call_rcu(&inode->i_rcu, fuse_i_callback);
122 }
123
124 static void fuse_evict_inode(struct inode *inode)
125 {
126 truncate_inode_pages_final(&inode->i_data);
127 clear_inode(inode);
128 if (inode->i_sb->s_flags & MS_ACTIVE) {
129 struct fuse_conn *fc = get_fuse_conn(inode);
130 struct fuse_inode *fi = get_fuse_inode(inode);
131 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
132 fi->forget = NULL;
133 }
134 }
135
136 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
137 {
138 sync_filesystem(sb);
139 if (*flags & MS_MANDLOCK)
140 return -EINVAL;
141
142 return 0;
143 }
144
145 /*
146 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
147 * so that it will fit.
148 */
149 static ino_t fuse_squash_ino(u64 ino64)
150 {
151 ino_t ino = (ino_t) ino64;
152 if (sizeof(ino_t) < sizeof(u64))
153 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
154 return ino;
155 }
156
157 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
158 u64 attr_valid)
159 {
160 struct fuse_conn *fc = get_fuse_conn(inode);
161 struct fuse_inode *fi = get_fuse_inode(inode);
162
163 fi->attr_version = ++fc->attr_version;
164 fi->i_time = attr_valid;
165
166 inode->i_ino = fuse_squash_ino(attr->ino);
167 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
168 set_nlink(inode, attr->nlink);
169 inode->i_uid = make_kuid(&init_user_ns, attr->uid);
170 inode->i_gid = make_kgid(&init_user_ns, attr->gid);
171 inode->i_blocks = attr->blocks;
172 inode->i_atime.tv_sec = attr->atime;
173 inode->i_atime.tv_nsec = attr->atimensec;
174 /* mtime from server may be stale due to local buffered write */
175 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
176 inode->i_mtime.tv_sec = attr->mtime;
177 inode->i_mtime.tv_nsec = attr->mtimensec;
178 inode->i_ctime.tv_sec = attr->ctime;
179 inode->i_ctime.tv_nsec = attr->ctimensec;
180 }
181
182 if (attr->blksize != 0)
183 inode->i_blkbits = ilog2(attr->blksize);
184 else
185 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
186
187 /*
188 * Don't set the sticky bit in i_mode, unless we want the VFS
189 * to check permissions. This prevents failures due to the
190 * check in may_delete().
191 */
192 fi->orig_i_mode = inode->i_mode;
193 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
194 inode->i_mode &= ~S_ISVTX;
195
196 fi->orig_ino = attr->ino;
197 }
198
199 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
200 u64 attr_valid, u64 attr_version)
201 {
202 struct fuse_conn *fc = get_fuse_conn(inode);
203 struct fuse_inode *fi = get_fuse_inode(inode);
204 bool is_wb = fc->writeback_cache;
205 loff_t oldsize;
206 struct timespec old_mtime;
207
208 spin_lock(&fc->lock);
209 if ((attr_version != 0 && fi->attr_version > attr_version) ||
210 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
211 spin_unlock(&fc->lock);
212 return;
213 }
214
215 old_mtime = inode->i_mtime;
216 fuse_change_attributes_common(inode, attr, attr_valid);
217
218 oldsize = inode->i_size;
219 /*
220 * In case of writeback_cache enabled, the cached writes beyond EOF
221 * extend local i_size without keeping userspace server in sync. So,
222 * attr->size coming from server can be stale. We cannot trust it.
223 */
224 if (!is_wb || !S_ISREG(inode->i_mode))
225 i_size_write(inode, attr->size);
226 spin_unlock(&fc->lock);
227
228 if (!is_wb && S_ISREG(inode->i_mode)) {
229 bool inval = false;
230
231 if (oldsize != attr->size) {
232 truncate_pagecache(inode, attr->size);
233 inval = true;
234 } else if (fc->auto_inval_data) {
235 struct timespec new_mtime = {
236 .tv_sec = attr->mtime,
237 .tv_nsec = attr->mtimensec,
238 };
239
240 /*
241 * Auto inval mode also checks and invalidates if mtime
242 * has changed.
243 */
244 if (!timespec_equal(&old_mtime, &new_mtime))
245 inval = true;
246 }
247
248 if (inval)
249 invalidate_inode_pages2(inode->i_mapping);
250 }
251 }
252
253 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
254 {
255 inode->i_mode = attr->mode & S_IFMT;
256 inode->i_size = attr->size;
257 inode->i_mtime.tv_sec = attr->mtime;
258 inode->i_mtime.tv_nsec = attr->mtimensec;
259 inode->i_ctime.tv_sec = attr->ctime;
260 inode->i_ctime.tv_nsec = attr->ctimensec;
261 if (S_ISREG(inode->i_mode)) {
262 fuse_init_common(inode);
263 fuse_init_file_inode(inode);
264 } else if (S_ISDIR(inode->i_mode))
265 fuse_init_dir(inode);
266 else if (S_ISLNK(inode->i_mode))
267 fuse_init_symlink(inode);
268 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
269 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
270 fuse_init_common(inode);
271 init_special_inode(inode, inode->i_mode,
272 new_decode_dev(attr->rdev));
273 } else
274 BUG();
275 }
276
277 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
278 {
279 u64 nodeid = *(u64 *) _nodeidp;
280 if (get_node_id(inode) == nodeid)
281 return 1;
282 else
283 return 0;
284 }
285
286 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
287 {
288 u64 nodeid = *(u64 *) _nodeidp;
289 get_fuse_inode(inode)->nodeid = nodeid;
290 return 0;
291 }
292
293 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
294 int generation, struct fuse_attr *attr,
295 u64 attr_valid, u64 attr_version)
296 {
297 struct inode *inode;
298 struct fuse_inode *fi;
299 struct fuse_conn *fc = get_fuse_conn_super(sb);
300
301 retry:
302 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
303 if (!inode)
304 return NULL;
305
306 if ((inode->i_state & I_NEW)) {
307 inode->i_flags |= S_NOATIME;
308 if (!fc->writeback_cache || !S_ISREG(attr->mode))
309 inode->i_flags |= S_NOCMTIME;
310 inode->i_generation = generation;
311 fuse_init_inode(inode, attr);
312 unlock_new_inode(inode);
313 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
314 /* Inode has changed type, any I/O on the old should fail */
315 make_bad_inode(inode);
316 iput(inode);
317 goto retry;
318 }
319
320 fi = get_fuse_inode(inode);
321 spin_lock(&fc->lock);
322 fi->nlookup++;
323 spin_unlock(&fc->lock);
324 fuse_change_attributes(inode, attr, attr_valid, attr_version);
325
326 return inode;
327 }
328
329 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
330 loff_t offset, loff_t len)
331 {
332 struct inode *inode;
333 pgoff_t pg_start;
334 pgoff_t pg_end;
335
336 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
337 if (!inode)
338 return -ENOENT;
339
340 fuse_invalidate_attr(inode);
341 if (offset >= 0) {
342 pg_start = offset >> PAGE_CACHE_SHIFT;
343 if (len <= 0)
344 pg_end = -1;
345 else
346 pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
347 invalidate_inode_pages2_range(inode->i_mapping,
348 pg_start, pg_end);
349 }
350 iput(inode);
351 return 0;
352 }
353
354 static void fuse_umount_begin(struct super_block *sb)
355 {
356 fuse_abort_conn(get_fuse_conn_super(sb));
357 }
358
359 static void fuse_send_destroy(struct fuse_conn *fc)
360 {
361 struct fuse_req *req = fc->destroy_req;
362 if (req && fc->conn_init) {
363 fc->destroy_req = NULL;
364 req->in.h.opcode = FUSE_DESTROY;
365 __set_bit(FR_FORCE, &req->flags);
366 __clear_bit(FR_BACKGROUND, &req->flags);
367 fuse_request_send(fc, req);
368 fuse_put_request(fc, req);
369 }
370 }
371
372 static void fuse_bdi_destroy(struct fuse_conn *fc)
373 {
374 if (fc->bdi_initialized)
375 bdi_destroy(&fc->bdi);
376 }
377
378 static void fuse_put_super(struct super_block *sb)
379 {
380 struct fuse_conn *fc = get_fuse_conn_super(sb);
381
382 fuse_send_destroy(fc);
383
384 fuse_abort_conn(fc);
385 mutex_lock(&fuse_mutex);
386 list_del(&fc->entry);
387 fuse_ctl_remove_conn(fc);
388 mutex_unlock(&fuse_mutex);
389 fuse_bdi_destroy(fc);
390
391 fuse_conn_put(fc);
392 }
393
394 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
395 {
396 stbuf->f_type = FUSE_SUPER_MAGIC;
397 stbuf->f_bsize = attr->bsize;
398 stbuf->f_frsize = attr->frsize;
399 stbuf->f_blocks = attr->blocks;
400 stbuf->f_bfree = attr->bfree;
401 stbuf->f_bavail = attr->bavail;
402 stbuf->f_files = attr->files;
403 stbuf->f_ffree = attr->ffree;
404 stbuf->f_namelen = attr->namelen;
405 /* fsid is left zero */
406 }
407
408 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
409 {
410 struct super_block *sb = dentry->d_sb;
411 struct fuse_conn *fc = get_fuse_conn_super(sb);
412 FUSE_ARGS(args);
413 struct fuse_statfs_out outarg;
414 int err;
415
416 if (!fuse_allow_current_process(fc)) {
417 buf->f_type = FUSE_SUPER_MAGIC;
418 return 0;
419 }
420
421 memset(&outarg, 0, sizeof(outarg));
422 args.in.numargs = 0;
423 args.in.h.opcode = FUSE_STATFS;
424 args.in.h.nodeid = get_node_id(d_inode(dentry));
425 args.out.numargs = 1;
426 args.out.args[0].size = sizeof(outarg);
427 args.out.args[0].value = &outarg;
428 err = fuse_simple_request(fc, &args);
429 if (!err)
430 convert_fuse_statfs(buf, &outarg.st);
431 return err;
432 }
433
434 enum {
435 OPT_FD,
436 OPT_ROOTMODE,
437 OPT_USER_ID,
438 OPT_GROUP_ID,
439 OPT_DEFAULT_PERMISSIONS,
440 OPT_ALLOW_OTHER,
441 OPT_MAX_READ,
442 OPT_BLKSIZE,
443 OPT_ERR
444 };
445
446 static const match_table_t tokens = {
447 {OPT_FD, "fd=%u"},
448 {OPT_ROOTMODE, "rootmode=%o"},
449 {OPT_USER_ID, "user_id=%u"},
450 {OPT_GROUP_ID, "group_id=%u"},
451 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
452 {OPT_ALLOW_OTHER, "allow_other"},
453 {OPT_MAX_READ, "max_read=%u"},
454 {OPT_BLKSIZE, "blksize=%u"},
455 {OPT_ERR, NULL}
456 };
457
458 static int fuse_match_uint(substring_t *s, unsigned int *res)
459 {
460 int err = -ENOMEM;
461 char *buf = match_strdup(s);
462 if (buf) {
463 err = kstrtouint(buf, 10, res);
464 kfree(buf);
465 }
466 return err;
467 }
468
469 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
470 {
471 char *p;
472 memset(d, 0, sizeof(struct fuse_mount_data));
473 d->max_read = ~0;
474 d->blksize = FUSE_DEFAULT_BLKSIZE;
475
476 while ((p = strsep(&opt, ",")) != NULL) {
477 int token;
478 int value;
479 unsigned uv;
480 substring_t args[MAX_OPT_ARGS];
481 if (!*p)
482 continue;
483
484 token = match_token(p, tokens, args);
485 switch (token) {
486 case OPT_FD:
487 if (match_int(&args[0], &value))
488 return 0;
489 d->fd = value;
490 d->fd_present = 1;
491 break;
492
493 case OPT_ROOTMODE:
494 if (match_octal(&args[0], &value))
495 return 0;
496 if (!fuse_valid_type(value))
497 return 0;
498 d->rootmode = value;
499 d->rootmode_present = 1;
500 break;
501
502 case OPT_USER_ID:
503 if (fuse_match_uint(&args[0], &uv))
504 return 0;
505 d->user_id = make_kuid(current_user_ns(), uv);
506 if (!uid_valid(d->user_id))
507 return 0;
508 d->user_id_present = 1;
509 break;
510
511 case OPT_GROUP_ID:
512 if (fuse_match_uint(&args[0], &uv))
513 return 0;
514 d->group_id = make_kgid(current_user_ns(), uv);
515 if (!gid_valid(d->group_id))
516 return 0;
517 d->group_id_present = 1;
518 break;
519
520 case OPT_DEFAULT_PERMISSIONS:
521 d->flags |= FUSE_DEFAULT_PERMISSIONS;
522 break;
523
524 case OPT_ALLOW_OTHER:
525 d->flags |= FUSE_ALLOW_OTHER;
526 break;
527
528 case OPT_MAX_READ:
529 if (match_int(&args[0], &value))
530 return 0;
531 d->max_read = value;
532 break;
533
534 case OPT_BLKSIZE:
535 if (!is_bdev || match_int(&args[0], &value))
536 return 0;
537 d->blksize = value;
538 break;
539
540 default:
541 return 0;
542 }
543 }
544
545 if (!d->fd_present || !d->rootmode_present ||
546 !d->user_id_present || !d->group_id_present)
547 return 0;
548
549 return 1;
550 }
551
552 static int fuse_show_options(struct seq_file *m, struct dentry *root)
553 {
554 struct super_block *sb = root->d_sb;
555 struct fuse_conn *fc = get_fuse_conn_super(sb);
556
557 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
558 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
559 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
560 seq_puts(m, ",default_permissions");
561 if (fc->flags & FUSE_ALLOW_OTHER)
562 seq_puts(m, ",allow_other");
563 if (fc->max_read != ~0)
564 seq_printf(m, ",max_read=%u", fc->max_read);
565 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
566 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
567 return 0;
568 }
569
570 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
571 {
572 memset(fiq, 0, sizeof(struct fuse_iqueue));
573 init_waitqueue_head(&fiq->waitq);
574 INIT_LIST_HEAD(&fiq->pending);
575 INIT_LIST_HEAD(&fiq->interrupts);
576 fiq->forget_list_tail = &fiq->forget_list_head;
577 fiq->connected = 1;
578 }
579
580 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
581 {
582 memset(fpq, 0, sizeof(struct fuse_pqueue));
583 spin_lock_init(&fpq->lock);
584 INIT_LIST_HEAD(&fpq->processing);
585 INIT_LIST_HEAD(&fpq->io);
586 fpq->connected = 1;
587 }
588
589 void fuse_conn_init(struct fuse_conn *fc)
590 {
591 memset(fc, 0, sizeof(*fc));
592 spin_lock_init(&fc->lock);
593 init_rwsem(&fc->killsb);
594 atomic_set(&fc->count, 1);
595 init_waitqueue_head(&fc->blocked_waitq);
596 init_waitqueue_head(&fc->reserved_req_waitq);
597 fuse_iqueue_init(&fc->iq);
598 fuse_pqueue_init(&fc->pq);
599 INIT_LIST_HEAD(&fc->bg_queue);
600 INIT_LIST_HEAD(&fc->entry);
601 atomic_set(&fc->num_waiting, 0);
602 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
603 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
604 fc->khctr = 0;
605 fc->polled_files = RB_ROOT;
606 fc->blocked = 0;
607 fc->initialized = 0;
608 fc->connected = 1;
609 fc->attr_version = 1;
610 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
611 }
612 EXPORT_SYMBOL_GPL(fuse_conn_init);
613
614 void fuse_conn_put(struct fuse_conn *fc)
615 {
616 if (atomic_dec_and_test(&fc->count)) {
617 if (fc->destroy_req)
618 fuse_request_free(fc->destroy_req);
619 fc->release(fc);
620 }
621 }
622 EXPORT_SYMBOL_GPL(fuse_conn_put);
623
624 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
625 {
626 atomic_inc(&fc->count);
627 return fc;
628 }
629 EXPORT_SYMBOL_GPL(fuse_conn_get);
630
631 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
632 {
633 struct fuse_attr attr;
634 memset(&attr, 0, sizeof(attr));
635
636 attr.mode = mode;
637 attr.ino = FUSE_ROOT_ID;
638 attr.nlink = 1;
639 return fuse_iget(sb, 1, 0, &attr, 0, 0);
640 }
641
642 struct fuse_inode_handle {
643 u64 nodeid;
644 u32 generation;
645 };
646
647 static struct dentry *fuse_get_dentry(struct super_block *sb,
648 struct fuse_inode_handle *handle)
649 {
650 struct fuse_conn *fc = get_fuse_conn_super(sb);
651 struct inode *inode;
652 struct dentry *entry;
653 int err = -ESTALE;
654
655 if (handle->nodeid == 0)
656 goto out_err;
657
658 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
659 if (!inode) {
660 struct fuse_entry_out outarg;
661 struct qstr name;
662
663 if (!fc->export_support)
664 goto out_err;
665
666 name.len = 1;
667 name.name = ".";
668 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
669 &inode);
670 if (err && err != -ENOENT)
671 goto out_err;
672 if (err || !inode) {
673 err = -ESTALE;
674 goto out_err;
675 }
676 err = -EIO;
677 if (get_node_id(inode) != handle->nodeid)
678 goto out_iput;
679 }
680 err = -ESTALE;
681 if (inode->i_generation != handle->generation)
682 goto out_iput;
683
684 entry = d_obtain_alias(inode);
685 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
686 fuse_invalidate_entry_cache(entry);
687
688 return entry;
689
690 out_iput:
691 iput(inode);
692 out_err:
693 return ERR_PTR(err);
694 }
695
696 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
697 struct inode *parent)
698 {
699 int len = parent ? 6 : 3;
700 u64 nodeid;
701 u32 generation;
702
703 if (*max_len < len) {
704 *max_len = len;
705 return FILEID_INVALID;
706 }
707
708 nodeid = get_fuse_inode(inode)->nodeid;
709 generation = inode->i_generation;
710
711 fh[0] = (u32)(nodeid >> 32);
712 fh[1] = (u32)(nodeid & 0xffffffff);
713 fh[2] = generation;
714
715 if (parent) {
716 nodeid = get_fuse_inode(parent)->nodeid;
717 generation = parent->i_generation;
718
719 fh[3] = (u32)(nodeid >> 32);
720 fh[4] = (u32)(nodeid & 0xffffffff);
721 fh[5] = generation;
722 }
723
724 *max_len = len;
725 return parent ? 0x82 : 0x81;
726 }
727
728 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
729 struct fid *fid, int fh_len, int fh_type)
730 {
731 struct fuse_inode_handle handle;
732
733 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
734 return NULL;
735
736 handle.nodeid = (u64) fid->raw[0] << 32;
737 handle.nodeid |= (u64) fid->raw[1];
738 handle.generation = fid->raw[2];
739 return fuse_get_dentry(sb, &handle);
740 }
741
742 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
743 struct fid *fid, int fh_len, int fh_type)
744 {
745 struct fuse_inode_handle parent;
746
747 if (fh_type != 0x82 || fh_len < 6)
748 return NULL;
749
750 parent.nodeid = (u64) fid->raw[3] << 32;
751 parent.nodeid |= (u64) fid->raw[4];
752 parent.generation = fid->raw[5];
753 return fuse_get_dentry(sb, &parent);
754 }
755
756 static struct dentry *fuse_get_parent(struct dentry *child)
757 {
758 struct inode *child_inode = d_inode(child);
759 struct fuse_conn *fc = get_fuse_conn(child_inode);
760 struct inode *inode;
761 struct dentry *parent;
762 struct fuse_entry_out outarg;
763 struct qstr name;
764 int err;
765
766 if (!fc->export_support)
767 return ERR_PTR(-ESTALE);
768
769 name.len = 2;
770 name.name = "..";
771 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
772 &name, &outarg, &inode);
773 if (err) {
774 if (err == -ENOENT)
775 return ERR_PTR(-ESTALE);
776 return ERR_PTR(err);
777 }
778
779 parent = d_obtain_alias(inode);
780 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
781 fuse_invalidate_entry_cache(parent);
782
783 return parent;
784 }
785
786 static const struct export_operations fuse_export_operations = {
787 .fh_to_dentry = fuse_fh_to_dentry,
788 .fh_to_parent = fuse_fh_to_parent,
789 .encode_fh = fuse_encode_fh,
790 .get_parent = fuse_get_parent,
791 };
792
793 static const struct super_operations fuse_super_operations = {
794 .alloc_inode = fuse_alloc_inode,
795 .destroy_inode = fuse_destroy_inode,
796 .evict_inode = fuse_evict_inode,
797 .write_inode = fuse_write_inode,
798 .drop_inode = generic_delete_inode,
799 .remount_fs = fuse_remount_fs,
800 .put_super = fuse_put_super,
801 .umount_begin = fuse_umount_begin,
802 .statfs = fuse_statfs,
803 .show_options = fuse_show_options,
804 };
805
806 static void sanitize_global_limit(unsigned *limit)
807 {
808 if (*limit == 0)
809 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
810 sizeof(struct fuse_req);
811
812 if (*limit >= 1 << 16)
813 *limit = (1 << 16) - 1;
814 }
815
816 static int set_global_limit(const char *val, struct kernel_param *kp)
817 {
818 int rv;
819
820 rv = param_set_uint(val, kp);
821 if (rv)
822 return rv;
823
824 sanitize_global_limit((unsigned *)kp->arg);
825
826 return 0;
827 }
828
829 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
830 {
831 int cap_sys_admin = capable(CAP_SYS_ADMIN);
832
833 if (arg->minor < 13)
834 return;
835
836 sanitize_global_limit(&max_user_bgreq);
837 sanitize_global_limit(&max_user_congthresh);
838
839 if (arg->max_background) {
840 fc->max_background = arg->max_background;
841
842 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
843 fc->max_background = max_user_bgreq;
844 }
845 if (arg->congestion_threshold) {
846 fc->congestion_threshold = arg->congestion_threshold;
847
848 if (!cap_sys_admin &&
849 fc->congestion_threshold > max_user_congthresh)
850 fc->congestion_threshold = max_user_congthresh;
851 }
852 }
853
854 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
855 {
856 struct fuse_init_out *arg = &req->misc.init_out;
857
858 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
859 fc->conn_error = 1;
860 else {
861 unsigned long ra_pages;
862
863 process_init_limits(fc, arg);
864
865 if (arg->minor >= 6) {
866 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
867 if (arg->flags & FUSE_ASYNC_READ)
868 fc->async_read = 1;
869 if (!(arg->flags & FUSE_POSIX_LOCKS))
870 fc->no_lock = 1;
871 if (arg->minor >= 17) {
872 if (!(arg->flags & FUSE_FLOCK_LOCKS))
873 fc->no_flock = 1;
874 } else {
875 if (!(arg->flags & FUSE_POSIX_LOCKS))
876 fc->no_flock = 1;
877 }
878 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
879 fc->atomic_o_trunc = 1;
880 if (arg->minor >= 9) {
881 /* LOOKUP has dependency on proto version */
882 if (arg->flags & FUSE_EXPORT_SUPPORT)
883 fc->export_support = 1;
884 }
885 if (arg->flags & FUSE_BIG_WRITES)
886 fc->big_writes = 1;
887 if (arg->flags & FUSE_DONT_MASK)
888 fc->dont_mask = 1;
889 if (arg->flags & FUSE_AUTO_INVAL_DATA)
890 fc->auto_inval_data = 1;
891 if (arg->flags & FUSE_DO_READDIRPLUS) {
892 fc->do_readdirplus = 1;
893 if (arg->flags & FUSE_READDIRPLUS_AUTO)
894 fc->readdirplus_auto = 1;
895 }
896 if (arg->flags & FUSE_ASYNC_DIO)
897 fc->async_dio = 1;
898 if (arg->flags & FUSE_WRITEBACK_CACHE)
899 fc->writeback_cache = 1;
900 if (arg->time_gran && arg->time_gran <= 1000000000)
901 fc->sb->s_time_gran = arg->time_gran;
902 } else {
903 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
904 fc->no_lock = 1;
905 fc->no_flock = 1;
906 }
907
908 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
909 fc->minor = arg->minor;
910 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
911 fc->max_write = max_t(unsigned, 4096, fc->max_write);
912 fc->conn_init = 1;
913 }
914 fuse_set_initialized(fc);
915 wake_up_all(&fc->blocked_waitq);
916 }
917
918 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
919 {
920 struct fuse_init_in *arg = &req->misc.init_in;
921
922 arg->major = FUSE_KERNEL_VERSION;
923 arg->minor = FUSE_KERNEL_MINOR_VERSION;
924 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
925 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
926 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
927 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
928 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
929 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
930 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT;
931 req->in.h.opcode = FUSE_INIT;
932 req->in.numargs = 1;
933 req->in.args[0].size = sizeof(*arg);
934 req->in.args[0].value = arg;
935 req->out.numargs = 1;
936 /* Variable length argument used for backward compatibility
937 with interface version < 7.5. Rest of init_out is zeroed
938 by do_get_request(), so a short reply is not a problem */
939 req->out.argvar = 1;
940 req->out.args[0].size = sizeof(struct fuse_init_out);
941 req->out.args[0].value = &req->misc.init_out;
942 req->end = process_init_reply;
943 fuse_request_send_background(fc, req);
944 }
945
946 static void fuse_free_conn(struct fuse_conn *fc)
947 {
948 kfree_rcu(fc, rcu);
949 }
950
951 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
952 {
953 int err;
954
955 fc->bdi.name = "fuse";
956 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
957 /* fuse does it's own writeback accounting */
958 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
959
960 err = bdi_init(&fc->bdi);
961 if (err)
962 return err;
963
964 fc->bdi_initialized = 1;
965
966 if (sb->s_bdev) {
967 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
968 MAJOR(fc->dev), MINOR(fc->dev));
969 } else {
970 err = bdi_register_dev(&fc->bdi, fc->dev);
971 }
972
973 if (err)
974 return err;
975
976 /*
977 * For a single fuse filesystem use max 1% of dirty +
978 * writeback threshold.
979 *
980 * This gives about 1M of write buffer for memory maps on a
981 * machine with 1G and 10% dirty_ratio, which should be more
982 * than enough.
983 *
984 * Privileged users can raise it by writing to
985 *
986 * /sys/class/bdi/<bdi>/max_ratio
987 */
988 bdi_set_max_ratio(&fc->bdi, 1);
989
990 return 0;
991 }
992
993 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
994 {
995 struct fuse_conn *fc;
996 struct inode *root;
997 struct fuse_mount_data d;
998 struct file *file;
999 struct dentry *root_dentry;
1000 struct fuse_req *init_req;
1001 int err;
1002 int is_bdev = sb->s_bdev != NULL;
1003
1004 err = -EINVAL;
1005 if (sb->s_flags & MS_MANDLOCK)
1006 goto err;
1007
1008 sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1009
1010 if (!parse_fuse_opt(data, &d, is_bdev))
1011 goto err;
1012
1013 if (is_bdev) {
1014 #ifdef CONFIG_BLOCK
1015 err = -EINVAL;
1016 if (!sb_set_blocksize(sb, d.blksize))
1017 goto err;
1018 #endif
1019 } else {
1020 sb->s_blocksize = PAGE_CACHE_SIZE;
1021 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1022 }
1023 sb->s_magic = FUSE_SUPER_MAGIC;
1024 sb->s_op = &fuse_super_operations;
1025 sb->s_maxbytes = MAX_LFS_FILESIZE;
1026 sb->s_time_gran = 1;
1027 sb->s_export_op = &fuse_export_operations;
1028
1029 file = fget(d.fd);
1030 err = -EINVAL;
1031 if (!file)
1032 goto err;
1033
1034 if ((file->f_op != &fuse_dev_operations) ||
1035 (file->f_cred->user_ns != &init_user_ns))
1036 goto err_fput;
1037
1038 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1039 err = -ENOMEM;
1040 if (!fc)
1041 goto err_fput;
1042
1043 fuse_conn_init(fc);
1044 fc->release = fuse_free_conn;
1045
1046 fc->dev = sb->s_dev;
1047 fc->sb = sb;
1048 err = fuse_bdi_init(fc, sb);
1049 if (err)
1050 goto err_put_conn;
1051
1052 sb->s_bdi = &fc->bdi;
1053
1054 /* Handle umasking inside the fuse code */
1055 if (sb->s_flags & MS_POSIXACL)
1056 fc->dont_mask = 1;
1057 sb->s_flags |= MS_POSIXACL;
1058
1059 fc->flags = d.flags;
1060 fc->user_id = d.user_id;
1061 fc->group_id = d.group_id;
1062 fc->max_read = max_t(unsigned, 4096, d.max_read);
1063
1064 /* Used by get_root_inode() */
1065 sb->s_fs_info = fc;
1066
1067 err = -ENOMEM;
1068 root = fuse_get_root_inode(sb, d.rootmode);
1069 root_dentry = d_make_root(root);
1070 if (!root_dentry)
1071 goto err_put_conn;
1072 /* only now - we want root dentry with NULL ->d_op */
1073 sb->s_d_op = &fuse_dentry_operations;
1074
1075 init_req = fuse_request_alloc(0);
1076 if (!init_req)
1077 goto err_put_root;
1078 __set_bit(FR_BACKGROUND, &init_req->flags);
1079
1080 if (is_bdev) {
1081 fc->destroy_req = fuse_request_alloc(0);
1082 if (!fc->destroy_req)
1083 goto err_free_init_req;
1084 }
1085
1086 mutex_lock(&fuse_mutex);
1087 err = -EINVAL;
1088 if (file->private_data)
1089 goto err_unlock;
1090
1091 err = fuse_ctl_add_conn(fc);
1092 if (err)
1093 goto err_unlock;
1094
1095 list_add_tail(&fc->entry, &fuse_conn_list);
1096 sb->s_root = root_dentry;
1097 file->private_data = fuse_conn_get(fc);
1098 mutex_unlock(&fuse_mutex);
1099 /*
1100 * atomic_dec_and_test() in fput() provides the necessary
1101 * memory barrier for file->private_data to be visible on all
1102 * CPUs after this
1103 */
1104 fput(file);
1105
1106 fuse_send_init(fc, init_req);
1107
1108 return 0;
1109
1110 err_unlock:
1111 mutex_unlock(&fuse_mutex);
1112 err_free_init_req:
1113 fuse_request_free(init_req);
1114 err_put_root:
1115 dput(root_dentry);
1116 err_put_conn:
1117 fuse_bdi_destroy(fc);
1118 fuse_conn_put(fc);
1119 err_fput:
1120 fput(file);
1121 err:
1122 return err;
1123 }
1124
1125 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1126 int flags, const char *dev_name,
1127 void *raw_data)
1128 {
1129 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1130 }
1131
1132 static void fuse_kill_sb_anon(struct super_block *sb)
1133 {
1134 struct fuse_conn *fc = get_fuse_conn_super(sb);
1135
1136 if (fc) {
1137 down_write(&fc->killsb);
1138 fc->sb = NULL;
1139 up_write(&fc->killsb);
1140 }
1141
1142 kill_anon_super(sb);
1143 }
1144
1145 static struct file_system_type fuse_fs_type = {
1146 .owner = THIS_MODULE,
1147 .name = "fuse",
1148 .fs_flags = FS_HAS_SUBTYPE,
1149 .mount = fuse_mount,
1150 .kill_sb = fuse_kill_sb_anon,
1151 };
1152 MODULE_ALIAS_FS("fuse");
1153
1154 #ifdef CONFIG_BLOCK
1155 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1156 int flags, const char *dev_name,
1157 void *raw_data)
1158 {
1159 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1160 }
1161
1162 static void fuse_kill_sb_blk(struct super_block *sb)
1163 {
1164 struct fuse_conn *fc = get_fuse_conn_super(sb);
1165
1166 if (fc) {
1167 down_write(&fc->killsb);
1168 fc->sb = NULL;
1169 up_write(&fc->killsb);
1170 }
1171
1172 kill_block_super(sb);
1173 }
1174
1175 static struct file_system_type fuseblk_fs_type = {
1176 .owner = THIS_MODULE,
1177 .name = "fuseblk",
1178 .mount = fuse_mount_blk,
1179 .kill_sb = fuse_kill_sb_blk,
1180 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1181 };
1182 MODULE_ALIAS_FS("fuseblk");
1183
1184 static inline int register_fuseblk(void)
1185 {
1186 return register_filesystem(&fuseblk_fs_type);
1187 }
1188
1189 static inline void unregister_fuseblk(void)
1190 {
1191 unregister_filesystem(&fuseblk_fs_type);
1192 }
1193 #else
1194 static inline int register_fuseblk(void)
1195 {
1196 return 0;
1197 }
1198
1199 static inline void unregister_fuseblk(void)
1200 {
1201 }
1202 #endif
1203
1204 static void fuse_inode_init_once(void *foo)
1205 {
1206 struct inode *inode = foo;
1207
1208 inode_init_once(inode);
1209 }
1210
1211 static int __init fuse_fs_init(void)
1212 {
1213 int err;
1214
1215 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1216 sizeof(struct fuse_inode),
1217 0, SLAB_HWCACHE_ALIGN,
1218 fuse_inode_init_once);
1219 err = -ENOMEM;
1220 if (!fuse_inode_cachep)
1221 goto out;
1222
1223 err = register_fuseblk();
1224 if (err)
1225 goto out2;
1226
1227 err = register_filesystem(&fuse_fs_type);
1228 if (err)
1229 goto out3;
1230
1231 return 0;
1232
1233 out3:
1234 unregister_fuseblk();
1235 out2:
1236 kmem_cache_destroy(fuse_inode_cachep);
1237 out:
1238 return err;
1239 }
1240
1241 static void fuse_fs_cleanup(void)
1242 {
1243 unregister_filesystem(&fuse_fs_type);
1244 unregister_fuseblk();
1245
1246 /*
1247 * Make sure all delayed rcu free inodes are flushed before we
1248 * destroy cache.
1249 */
1250 rcu_barrier();
1251 kmem_cache_destroy(fuse_inode_cachep);
1252 }
1253
1254 static struct kobject *fuse_kobj;
1255 static struct kobject *connections_kobj;
1256
1257 static int fuse_sysfs_init(void)
1258 {
1259 int err;
1260
1261 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1262 if (!fuse_kobj) {
1263 err = -ENOMEM;
1264 goto out_err;
1265 }
1266
1267 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1268 if (!connections_kobj) {
1269 err = -ENOMEM;
1270 goto out_fuse_unregister;
1271 }
1272
1273 return 0;
1274
1275 out_fuse_unregister:
1276 kobject_put(fuse_kobj);
1277 out_err:
1278 return err;
1279 }
1280
1281 static void fuse_sysfs_cleanup(void)
1282 {
1283 kobject_put(connections_kobj);
1284 kobject_put(fuse_kobj);
1285 }
1286
1287 static int __init fuse_init(void)
1288 {
1289 int res;
1290
1291 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1292 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1293
1294 INIT_LIST_HEAD(&fuse_conn_list);
1295 res = fuse_fs_init();
1296 if (res)
1297 goto err;
1298
1299 res = fuse_dev_init();
1300 if (res)
1301 goto err_fs_cleanup;
1302
1303 res = fuse_sysfs_init();
1304 if (res)
1305 goto err_dev_cleanup;
1306
1307 res = fuse_ctl_init();
1308 if (res)
1309 goto err_sysfs_cleanup;
1310
1311 sanitize_global_limit(&max_user_bgreq);
1312 sanitize_global_limit(&max_user_congthresh);
1313
1314 return 0;
1315
1316 err_sysfs_cleanup:
1317 fuse_sysfs_cleanup();
1318 err_dev_cleanup:
1319 fuse_dev_cleanup();
1320 err_fs_cleanup:
1321 fuse_fs_cleanup();
1322 err:
1323 return res;
1324 }
1325
1326 static void __exit fuse_exit(void)
1327 {
1328 printk(KERN_DEBUG "fuse exit\n");
1329
1330 fuse_ctl_cleanup();
1331 fuse_sysfs_cleanup();
1332 fuse_fs_cleanup();
1333 fuse_dev_cleanup();
1334 }
1335
1336 module_init(fuse_init);
1337 module_exit(fuse_exit);
Linux kernel - Deliverables
This page took 0.160093 seconds and 5 git commands to generate.