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 mutex_init(&fi->mutex); 101 fi->forget = fuse_alloc_forget(); 102 if (!fi->forget) { 103 kmem_cache_free(fuse_inode_cachep, inode); 104 return NULL; 105 } 106 107 return inode; 108 } 109 110 static void fuse_i_callback(struct rcu_head *head) 111 { 112 struct inode *inode = container_of(head, struct inode, i_rcu); 113 kmem_cache_free(fuse_inode_cachep, inode); 114 } 115 116 static void fuse_destroy_inode(struct inode *inode) 117 { 118 struct fuse_inode *fi = get_fuse_inode(inode); 119 BUG_ON(!list_empty(&fi->write_files)); 120 BUG_ON(!list_empty(&fi->queued_writes)); 121 mutex_destroy(&fi->mutex); 122 kfree(fi->forget); 123 call_rcu(&inode->i_rcu, fuse_i_callback); 124 } 125 126 static void fuse_evict_inode(struct inode *inode) 127 { 128 truncate_inode_pages_final(&inode->i_data); 129 clear_inode(inode); 130 if (inode->i_sb->s_flags & MS_ACTIVE) { 131 struct fuse_conn *fc = get_fuse_conn(inode); 132 struct fuse_inode *fi = get_fuse_inode(inode); 133 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup); 134 fi->forget = NULL; 135 } 136 } 137 138 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data) 139 { 140 sync_filesystem(sb); 141 if (*flags & MS_MANDLOCK) 142 return -EINVAL; 143 144 return 0; 145 } 146 147 /* 148 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down 149 * so that it will fit. 150 */ 151 static ino_t fuse_squash_ino(u64 ino64) 152 { 153 ino_t ino = (ino_t) ino64; 154 if (sizeof(ino_t) < sizeof(u64)) 155 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8; 156 return ino; 157 } 158 159 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr, 160 u64 attr_valid) 161 { 162 struct fuse_conn *fc = get_fuse_conn(inode); 163 struct fuse_inode *fi = get_fuse_inode(inode); 164 165 fi->attr_version = ++fc->attr_version; 166 fi->i_time = attr_valid; 167 168 inode->i_ino = fuse_squash_ino(attr->ino); 169 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777); 170 set_nlink(inode, attr->nlink); 171 inode->i_uid = make_kuid(&init_user_ns, attr->uid); 172 inode->i_gid = make_kgid(&init_user_ns, attr->gid); 173 inode->i_blocks = attr->blocks; 174 inode->i_atime.tv_sec = attr->atime; 175 inode->i_atime.tv_nsec = attr->atimensec; 176 /* mtime from server may be stale due to local buffered write */ 177 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) { 178 inode->i_mtime.tv_sec = attr->mtime; 179 inode->i_mtime.tv_nsec = attr->mtimensec; 180 inode->i_ctime.tv_sec = attr->ctime; 181 inode->i_ctime.tv_nsec = attr->ctimensec; 182 } 183 184 if (attr->blksize != 0) 185 inode->i_blkbits = ilog2(attr->blksize); 186 else 187 inode->i_blkbits = inode->i_sb->s_blocksize_bits; 188 189 /* 190 * Don't set the sticky bit in i_mode, unless we want the VFS 191 * to check permissions. This prevents failures due to the 192 * check in may_delete(). 193 */ 194 fi->orig_i_mode = inode->i_mode; 195 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS)) 196 inode->i_mode &= ~S_ISVTX; 197 198 fi->orig_ino = attr->ino; 199 } 200 201 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr, 202 u64 attr_valid, u64 attr_version) 203 { 204 struct fuse_conn *fc = get_fuse_conn(inode); 205 struct fuse_inode *fi = get_fuse_inode(inode); 206 bool is_wb = fc->writeback_cache; 207 loff_t oldsize; 208 struct timespec old_mtime; 209 210 spin_lock(&fc->lock); 211 if ((attr_version != 0 && fi->attr_version > attr_version) || 212 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { 213 spin_unlock(&fc->lock); 214 return; 215 } 216 217 old_mtime = inode->i_mtime; 218 fuse_change_attributes_common(inode, attr, attr_valid); 219 220 oldsize = inode->i_size; 221 /* 222 * In case of writeback_cache enabled, the cached writes beyond EOF 223 * extend local i_size without keeping userspace server in sync. So, 224 * attr->size coming from server can be stale. We cannot trust it. 225 */ 226 if (!is_wb || !S_ISREG(inode->i_mode)) 227 i_size_write(inode, attr->size); 228 spin_unlock(&fc->lock); 229 230 if (!is_wb && S_ISREG(inode->i_mode)) { 231 bool inval = false; 232 233 if (oldsize != attr->size) { 234 truncate_pagecache(inode, attr->size); 235 inval = true; 236 } else if (fc->auto_inval_data) { 237 struct timespec new_mtime = { 238 .tv_sec = attr->mtime, 239 .tv_nsec = attr->mtimensec, 240 }; 241 242 /* 243 * Auto inval mode also checks and invalidates if mtime 244 * has changed. 245 */ 246 if (!timespec_equal(&old_mtime, &new_mtime)) 247 inval = true; 248 } 249 250 if (inval) 251 invalidate_inode_pages2(inode->i_mapping); 252 } 253 } 254 255 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr) 256 { 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(); 277 } 278 279 int fuse_inode_eq(struct inode *inode, void *_nodeidp) 280 { 281 u64 nodeid = *(u64 *) _nodeidp; 282 if (get_node_id(inode) == nodeid) 283 return 1; 284 else 285 return 0; 286 } 287 288 static int fuse_inode_set(struct inode *inode, void *_nodeidp) 289 { 290 u64 nodeid = *(u64 *) _nodeidp; 291 get_fuse_inode(inode)->nodeid = nodeid; 292 return 0; 293 } 294 295 struct inode *fuse_iget(struct super_block *sb, u64 nodeid, 296 int generation, struct fuse_attr *attr, 297 u64 attr_valid, u64 attr_version) 298 { 299 struct inode *inode; 300 struct fuse_inode *fi; 301 struct fuse_conn *fc = get_fuse_conn_super(sb); 302 303 retry: 304 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid); 305 if (!inode) 306 return NULL; 307 308 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; 320 } 321 322 fi = get_fuse_inode(inode); 323 spin_lock(&fc->lock); 324 fi->nlookup++; 325 spin_unlock(&fc->lock); 326 fuse_change_attributes(inode, attr, attr_valid, attr_version); 327 328 return inode; 329 } 330 331 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid, 332 loff_t offset, loff_t len) 333 { 334 struct inode *inode; 335 pgoff_t pg_start; 336 pgoff_t pg_end; 337 338 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid); 339 if (!inode) 340 return -ENOENT; 341 342 fuse_invalidate_attr(inode); 343 if (offset >= 0) { 344 pg_start = offset >> PAGE_SHIFT; 345 if (len <= 0) 346 pg_end = -1; 347 else 348 pg_end = (offset + len - 1) >> PAGE_SHIFT; 349 invalidate_inode_pages2_range(inode->i_mapping, 350 pg_start, pg_end); 351 } 352 iput(inode); 353 return 0; 354 } 355 356 void fuse_lock_inode(struct inode *inode) 357 { 358 if (!get_fuse_conn(inode)->parallel_dirops) 359 mutex_lock(&get_fuse_inode(inode)->mutex); 360 } 361 362 void fuse_unlock_inode(struct inode *inode) 363 { 364 if (!get_fuse_conn(inode)->parallel_dirops) 365 mutex_unlock(&get_fuse_inode(inode)->mutex); 366 } 367 368 static void fuse_umount_begin(struct super_block *sb) 369 { 370 fuse_abort_conn(get_fuse_conn_super(sb)); 371 } 372 373 static void fuse_send_destroy(struct fuse_conn *fc) 374 { 375 struct fuse_req *req = fc->destroy_req; 376 if (req && fc->conn_init) { 377 fc->destroy_req = NULL; 378 req->in.h.opcode = FUSE_DESTROY; 379 __set_bit(FR_FORCE, &req->flags); 380 __clear_bit(FR_BACKGROUND, &req->flags); 381 fuse_request_send(fc, req); 382 fuse_put_request(fc, req); 383 } 384 } 385 386 static void fuse_bdi_destroy(struct fuse_conn *fc) 387 { 388 if (fc->bdi_initialized) 389 bdi_destroy(&fc->bdi); 390 } 391 392 static void fuse_put_super(struct super_block *sb) 393 { 394 struct fuse_conn *fc = get_fuse_conn_super(sb); 395 396 fuse_send_destroy(fc); 397 398 fuse_abort_conn(fc); 399 mutex_lock(&fuse_mutex); 400 list_del(&fc->entry); 401 fuse_ctl_remove_conn(fc); 402 mutex_unlock(&fuse_mutex); 403 fuse_bdi_destroy(fc); 404 405 fuse_conn_put(fc); 406 } 407 408 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr) 409 { 410 stbuf->f_type = FUSE_SUPER_MAGIC; 411 stbuf->f_bsize = attr->bsize; 412 stbuf->f_frsize = attr->frsize; 413 stbuf->f_blocks = attr->blocks; 414 stbuf->f_bfree = attr->bfree; 415 stbuf->f_bavail = attr->bavail; 416 stbuf->f_files = attr->files; 417 stbuf->f_ffree = attr->ffree; 418 stbuf->f_namelen = attr->namelen; 419 /* fsid is left zero */ 420 } 421 422 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf) 423 { 424 struct super_block *sb = dentry->d_sb; 425 struct fuse_conn *fc = get_fuse_conn_super(sb); 426 FUSE_ARGS(args); 427 struct fuse_statfs_out outarg; 428 int err; 429 430 if (!fuse_allow_current_process(fc)) { 431 buf->f_type = FUSE_SUPER_MAGIC; 432 return 0; 433 } 434 435 memset(&outarg, 0, sizeof(outarg)); 436 args.in.numargs = 0; 437 args.in.h.opcode = FUSE_STATFS; 438 args.in.h.nodeid = get_node_id(d_inode(dentry)); 439 args.out.numargs = 1; 440 args.out.args[0].size = sizeof(outarg); 441 args.out.args[0].value = &outarg; 442 err = fuse_simple_request(fc, &args); 443 if (!err) 444 convert_fuse_statfs(buf, &outarg.st); 445 return err; 446 } 447 448 enum { 449 OPT_FD, 450 OPT_ROOTMODE, 451 OPT_USER_ID, 452 OPT_GROUP_ID, 453 OPT_DEFAULT_PERMISSIONS, 454 OPT_ALLOW_OTHER, 455 OPT_MAX_READ, 456 OPT_BLKSIZE, 457 OPT_ERR 458 }; 459 460 static const match_table_t tokens = { 461 {OPT_FD, "fd=%u"}, 462 {OPT_ROOTMODE, "rootmode=%o"}, 463 {OPT_USER_ID, "user_id=%u"}, 464 {OPT_GROUP_ID, "group_id=%u"}, 465 {OPT_DEFAULT_PERMISSIONS, "default_permissions"}, 466 {OPT_ALLOW_OTHER, "allow_other"}, 467 {OPT_MAX_READ, "max_read=%u"}, 468 {OPT_BLKSIZE, "blksize=%u"}, 469 {OPT_ERR, NULL} 470 }; 471 472 static int fuse_match_uint(substring_t *s, unsigned int *res) 473 { 474 int err = -ENOMEM; 475 char *buf = match_strdup(s); 476 if (buf) { 477 err = kstrtouint(buf, 10, res); 478 kfree(buf); 479 } 480 return err; 481 } 482 483 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev) 484 { 485 char *p; 486 memset(d, 0, sizeof(struct fuse_mount_data)); 487 d->max_read = ~0; 488 d->blksize = FUSE_DEFAULT_BLKSIZE; 489 490 while ((p = strsep(&opt, ",")) != NULL) { 491 int token; 492 int value; 493 unsigned uv; 494 substring_t args[MAX_OPT_ARGS]; 495 if (!*p) 496 continue; 497 498 token = match_token(p, tokens, args); 499 switch (token) { 500 case OPT_FD: 501 if (match_int(&args[0], &value)) 502 return 0; 503 d->fd = value; 504 d->fd_present = 1; 505 break; 506 507 case OPT_ROOTMODE: 508 if (match_octal(&args[0], &value)) 509 return 0; 510 if (!fuse_valid_type(value)) 511 return 0; 512 d->rootmode = value; 513 d->rootmode_present = 1; 514 break; 515 516 case OPT_USER_ID: 517 if (fuse_match_uint(&args[0], &uv)) 518 return 0; 519 d->user_id = make_kuid(current_user_ns(), uv); 520 if (!uid_valid(d->user_id)) 521 return 0; 522 d->user_id_present = 1; 523 break; 524 525 case OPT_GROUP_ID: 526 if (fuse_match_uint(&args[0], &uv)) 527 return 0; 528 d->group_id = make_kgid(current_user_ns(), uv); 529 if (!gid_valid(d->group_id)) 530 return 0; 531 d->group_id_present = 1; 532 break; 533 534 case OPT_DEFAULT_PERMISSIONS: 535 d->flags |= FUSE_DEFAULT_PERMISSIONS; 536 break; 537 538 case OPT_ALLOW_OTHER: 539 d->flags |= FUSE_ALLOW_OTHER; 540 break; 541 542 case OPT_MAX_READ: 543 if (match_int(&args[0], &value)) 544 return 0; 545 d->max_read = value; 546 break; 547 548 case OPT_BLKSIZE: 549 if (!is_bdev || match_int(&args[0], &value)) 550 return 0; 551 d->blksize = value; 552 break; 553 554 default: 555 return 0; 556 } 557 } 558 559 if (!d->fd_present || !d->rootmode_present || 560 !d->user_id_present || !d->group_id_present) 561 return 0; 562 563 return 1; 564 } 565 566 static int fuse_show_options(struct seq_file *m, struct dentry *root) 567 { 568 struct super_block *sb = root->d_sb; 569 struct fuse_conn *fc = get_fuse_conn_super(sb); 570 571 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id)); 572 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id)); 573 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) 574 seq_puts(m, ",default_permissions"); 575 if (fc->flags & FUSE_ALLOW_OTHER) 576 seq_puts(m, ",allow_other"); 577 if (fc->max_read != ~0) 578 seq_printf(m, ",max_read=%u", fc->max_read); 579 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE) 580 seq_printf(m, ",blksize=%lu", sb->s_blocksize); 581 return 0; 582 } 583 584 static void fuse_iqueue_init(struct fuse_iqueue *fiq) 585 { 586 memset(fiq, 0, sizeof(struct fuse_iqueue)); 587 init_waitqueue_head(&fiq->waitq); 588 INIT_LIST_HEAD(&fiq->pending); 589 INIT_LIST_HEAD(&fiq->interrupts); 590 fiq->forget_list_tail = &fiq->forget_list_head; 591 fiq->connected = 1; 592 } 593 594 static void fuse_pqueue_init(struct fuse_pqueue *fpq) 595 { 596 memset(fpq, 0, sizeof(struct fuse_pqueue)); 597 spin_lock_init(&fpq->lock); 598 INIT_LIST_HEAD(&fpq->processing); 599 INIT_LIST_HEAD(&fpq->io); 600 fpq->connected = 1; 601 } 602 603 void fuse_conn_init(struct fuse_conn *fc) 604 { 605 memset(fc, 0, sizeof(*fc)); 606 spin_lock_init(&fc->lock); 607 init_rwsem(&fc->killsb); 608 atomic_set(&fc->count, 1); 609 atomic_set(&fc->dev_count, 1); 610 init_waitqueue_head(&fc->blocked_waitq); 611 init_waitqueue_head(&fc->reserved_req_waitq); 612 fuse_iqueue_init(&fc->iq); 613 INIT_LIST_HEAD(&fc->bg_queue); 614 INIT_LIST_HEAD(&fc->entry); 615 INIT_LIST_HEAD(&fc->devices); 616 atomic_set(&fc->num_waiting, 0); 617 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND; 618 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD; 619 fc->khctr = 0; 620 fc->polled_files = RB_ROOT; 621 fc->blocked = 0; 622 fc->initialized = 0; 623 fc->connected = 1; 624 fc->attr_version = 1; 625 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key)); 626 } 627 EXPORT_SYMBOL_GPL(fuse_conn_init); 628 629 void fuse_conn_put(struct fuse_conn *fc) 630 { 631 if (atomic_dec_and_test(&fc->count)) { 632 if (fc->destroy_req) 633 fuse_request_free(fc->destroy_req); 634 fc->release(fc); 635 } 636 } 637 EXPORT_SYMBOL_GPL(fuse_conn_put); 638 639 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc) 640 { 641 atomic_inc(&fc->count); 642 return fc; 643 } 644 EXPORT_SYMBOL_GPL(fuse_conn_get); 645 646 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode) 647 { 648 struct fuse_attr attr; 649 memset(&attr, 0, sizeof(attr)); 650 651 attr.mode = mode; 652 attr.ino = FUSE_ROOT_ID; 653 attr.nlink = 1; 654 return fuse_iget(sb, 1, 0, &attr, 0, 0); 655 } 656 657 struct fuse_inode_handle { 658 u64 nodeid; 659 u32 generation; 660 }; 661 662 static struct dentry *fuse_get_dentry(struct super_block *sb, 663 struct fuse_inode_handle *handle) 664 { 665 struct fuse_conn *fc = get_fuse_conn_super(sb); 666 struct inode *inode; 667 struct dentry *entry; 668 int err = -ESTALE; 669 670 if (handle->nodeid == 0) 671 goto out_err; 672 673 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid); 674 if (!inode) { 675 struct fuse_entry_out outarg; 676 const struct qstr name = QSTR_INIT(".", 1); 677 678 if (!fc->export_support) 679 goto out_err; 680 681 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg, 682 &inode); 683 if (err && err != -ENOENT) 684 goto out_err; 685 if (err || !inode) { 686 err = -ESTALE; 687 goto out_err; 688 } 689 err = -EIO; 690 if (get_node_id(inode) != handle->nodeid) 691 goto out_iput; 692 } 693 err = -ESTALE; 694 if (inode->i_generation != handle->generation) 695 goto out_iput; 696 697 entry = d_obtain_alias(inode); 698 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) 699 fuse_invalidate_entry_cache(entry); 700 701 return entry; 702 703 out_iput: 704 iput(inode); 705 out_err: 706 return ERR_PTR(err); 707 } 708 709 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len, 710 struct inode *parent) 711 { 712 int len = parent ? 6 : 3; 713 u64 nodeid; 714 u32 generation; 715 716 if (*max_len < len) { 717 *max_len = len; 718 return FILEID_INVALID; 719 } 720 721 nodeid = get_fuse_inode(inode)->nodeid; 722 generation = inode->i_generation; 723 724 fh[0] = (u32)(nodeid >> 32); 725 fh[1] = (u32)(nodeid & 0xffffffff); 726 fh[2] = generation; 727 728 if (parent) { 729 nodeid = get_fuse_inode(parent)->nodeid; 730 generation = parent->i_generation; 731 732 fh[3] = (u32)(nodeid >> 32); 733 fh[4] = (u32)(nodeid & 0xffffffff); 734 fh[5] = generation; 735 } 736 737 *max_len = len; 738 return parent ? 0x82 : 0x81; 739 } 740 741 static struct dentry *fuse_fh_to_dentry(struct super_block *sb, 742 struct fid *fid, int fh_len, int fh_type) 743 { 744 struct fuse_inode_handle handle; 745 746 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3) 747 return NULL; 748 749 handle.nodeid = (u64) fid->raw[0] << 32; 750 handle.nodeid |= (u64) fid->raw[1]; 751 handle.generation = fid->raw[2]; 752 return fuse_get_dentry(sb, &handle); 753 } 754 755 static struct dentry *fuse_fh_to_parent(struct super_block *sb, 756 struct fid *fid, int fh_len, int fh_type) 757 { 758 struct fuse_inode_handle parent; 759 760 if (fh_type != 0x82 || fh_len < 6) 761 return NULL; 762 763 parent.nodeid = (u64) fid->raw[3] << 32; 764 parent.nodeid |= (u64) fid->raw[4]; 765 parent.generation = fid->raw[5]; 766 return fuse_get_dentry(sb, &parent); 767 } 768 769 static struct dentry *fuse_get_parent(struct dentry *child) 770 { 771 struct inode *child_inode = d_inode(child); 772 struct fuse_conn *fc = get_fuse_conn(child_inode); 773 struct inode *inode; 774 struct dentry *parent; 775 struct fuse_entry_out outarg; 776 const struct qstr name = QSTR_INIT("..", 2); 777 int err; 778 779 if (!fc->export_support) 780 return ERR_PTR(-ESTALE); 781 782 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode), 783 &name, &outarg, &inode); 784 if (err) { 785 if (err == -ENOENT) 786 return ERR_PTR(-ESTALE); 787 return ERR_PTR(err); 788 } 789 790 parent = d_obtain_alias(inode); 791 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) 792 fuse_invalidate_entry_cache(parent); 793 794 return parent; 795 } 796 797 static const struct export_operations fuse_export_operations = { 798 .fh_to_dentry = fuse_fh_to_dentry, 799 .fh_to_parent = fuse_fh_to_parent, 800 .encode_fh = fuse_encode_fh, 801 .get_parent = fuse_get_parent, 802 }; 803 804 static const struct super_operations fuse_super_operations = { 805 .alloc_inode = fuse_alloc_inode, 806 .destroy_inode = fuse_destroy_inode, 807 .evict_inode = fuse_evict_inode, 808 .write_inode = fuse_write_inode, 809 .drop_inode = generic_delete_inode, 810 .remount_fs = fuse_remount_fs, 811 .put_super = fuse_put_super, 812 .umount_begin = fuse_umount_begin, 813 .statfs = fuse_statfs, 814 .show_options = fuse_show_options, 815 }; 816 817 static void sanitize_global_limit(unsigned *limit) 818 { 819 if (*limit == 0) 820 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) / 821 sizeof(struct fuse_req); 822 823 if (*limit >= 1 << 16) 824 *limit = (1 << 16) - 1; 825 } 826 827 static int set_global_limit(const char *val, struct kernel_param *kp) 828 { 829 int rv; 830 831 rv = param_set_uint(val, kp); 832 if (rv) 833 return rv; 834 835 sanitize_global_limit((unsigned *)kp->arg); 836 837 return 0; 838 } 839 840 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg) 841 { 842 int cap_sys_admin = capable(CAP_SYS_ADMIN); 843 844 if (arg->minor < 13) 845 return; 846 847 sanitize_global_limit(&max_user_bgreq); 848 sanitize_global_limit(&max_user_congthresh); 849 850 if (arg->max_background) { 851 fc->max_background = arg->max_background; 852 853 if (!cap_sys_admin && fc->max_background > max_user_bgreq) 854 fc->max_background = max_user_bgreq; 855 } 856 if (arg->congestion_threshold) { 857 fc->congestion_threshold = arg->congestion_threshold; 858 859 if (!cap_sys_admin && 860 fc->congestion_threshold > max_user_congthresh) 861 fc->congestion_threshold = max_user_congthresh; 862 } 863 } 864 865 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req) 866 { 867 struct fuse_init_out *arg = &req->misc.init_out; 868 869 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION) 870 fc->conn_error = 1; 871 else { 872 unsigned long ra_pages; 873 874 process_init_limits(fc, arg); 875 876 if (arg->minor >= 6) { 877 ra_pages = arg->max_readahead / PAGE_SIZE; 878 if (arg->flags & FUSE_ASYNC_READ) 879 fc->async_read = 1; 880 if (!(arg->flags & FUSE_POSIX_LOCKS)) 881 fc->no_lock = 1; 882 if (arg->minor >= 17) { 883 if (!(arg->flags & FUSE_FLOCK_LOCKS)) 884 fc->no_flock = 1; 885 } else { 886 if (!(arg->flags & FUSE_POSIX_LOCKS)) 887 fc->no_flock = 1; 888 } 889 if (arg->flags & FUSE_ATOMIC_O_TRUNC) 890 fc->atomic_o_trunc = 1; 891 if (arg->minor >= 9) { 892 /* LOOKUP has dependency on proto version */ 893 if (arg->flags & FUSE_EXPORT_SUPPORT) 894 fc->export_support = 1; 895 } 896 if (arg->flags & FUSE_BIG_WRITES) 897 fc->big_writes = 1; 898 if (arg->flags & FUSE_DONT_MASK) 899 fc->dont_mask = 1; 900 if (arg->flags & FUSE_AUTO_INVAL_DATA) 901 fc->auto_inval_data = 1; 902 if (arg->flags & FUSE_DO_READDIRPLUS) { 903 fc->do_readdirplus = 1; 904 if (arg->flags & FUSE_READDIRPLUS_AUTO) 905 fc->readdirplus_auto = 1; 906 } 907 if (arg->flags & FUSE_ASYNC_DIO) 908 fc->async_dio = 1; 909 if (arg->flags & FUSE_WRITEBACK_CACHE) 910 fc->writeback_cache = 1; 911 if (arg->flags & FUSE_PARALLEL_DIROPS) 912 fc->parallel_dirops = 1; 913 if (arg->time_gran && arg->time_gran <= 1000000000) 914 fc->sb->s_time_gran = arg->time_gran; 915 } else { 916 ra_pages = fc->max_read / PAGE_SIZE; 917 fc->no_lock = 1; 918 fc->no_flock = 1; 919 } 920 921 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages); 922 fc->minor = arg->minor; 923 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write; 924 fc->max_write = max_t(unsigned, 4096, fc->max_write); 925 fc->conn_init = 1; 926 } 927 fuse_set_initialized(fc); 928 wake_up_all(&fc->blocked_waitq); 929 } 930 931 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req) 932 { 933 struct fuse_init_in *arg = &req->misc.init_in; 934 935 arg->major = FUSE_KERNEL_VERSION; 936 arg->minor = FUSE_KERNEL_MINOR_VERSION; 937 arg->max_readahead = fc->bdi.ra_pages * PAGE_SIZE; 938 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC | 939 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK | 940 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ | 941 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA | 942 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO | 943 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT | 944 FUSE_PARALLEL_DIROPS; 945 req->in.h.opcode = FUSE_INIT; 946 req->in.numargs = 1; 947 req->in.args[0].size = sizeof(*arg); 948 req->in.args[0].value = arg; 949 req->out.numargs = 1; 950 /* Variable length argument used for backward compatibility 951 with interface version < 7.5. Rest of init_out is zeroed 952 by do_get_request(), so a short reply is not a problem */ 953 req->out.argvar = 1; 954 req->out.args[0].size = sizeof(struct fuse_init_out); 955 req->out.args[0].value = &req->misc.init_out; 956 req->end = process_init_reply; 957 fuse_request_send_background(fc, req); 958 } 959 960 static void fuse_free_conn(struct fuse_conn *fc) 961 { 962 WARN_ON(!list_empty(&fc->devices)); 963 kfree_rcu(fc, rcu); 964 } 965 966 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb) 967 { 968 int err; 969 970 fc->bdi.name = "fuse"; 971 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; 972 /* fuse does it's own writeback accounting */ 973 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT; 974 975 err = bdi_init(&fc->bdi); 976 if (err) 977 return err; 978 979 fc->bdi_initialized = 1; 980 981 if (sb->s_bdev) { 982 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk", 983 MAJOR(fc->dev), MINOR(fc->dev)); 984 } else { 985 err = bdi_register_dev(&fc->bdi, fc->dev); 986 } 987 988 if (err) 989 return err; 990 991 /* 992 * For a single fuse filesystem use max 1% of dirty + 993 * writeback threshold. 994 * 995 * This gives about 1M of write buffer for memory maps on a 996 * machine with 1G and 10% dirty_ratio, which should be more 997 * than enough. 998 * 999 * Privileged users can raise it by writing to 1000 * 1001 * /sys/class/bdi/<bdi>/max_ratio 1002 */ 1003 bdi_set_max_ratio(&fc->bdi, 1); 1004 1005 return 0; 1006 } 1007 1008 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc) 1009 { 1010 struct fuse_dev *fud; 1011 1012 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL); 1013 if (fud) { 1014 fud->fc = fuse_conn_get(fc); 1015 fuse_pqueue_init(&fud->pq); 1016 1017 spin_lock(&fc->lock); 1018 list_add_tail(&fud->entry, &fc->devices); 1019 spin_unlock(&fc->lock); 1020 } 1021 1022 return fud; 1023 } 1024 EXPORT_SYMBOL_GPL(fuse_dev_alloc); 1025 1026 void fuse_dev_free(struct fuse_dev *fud) 1027 { 1028 struct fuse_conn *fc = fud->fc; 1029 1030 if (fc) { 1031 spin_lock(&fc->lock); 1032 list_del(&fud->entry); 1033 spin_unlock(&fc->lock); 1034 1035 fuse_conn_put(fc); 1036 } 1037 kfree(fud); 1038 } 1039 EXPORT_SYMBOL_GPL(fuse_dev_free); 1040 1041 static int fuse_fill_super(struct super_block *sb, void *data, int silent) 1042 { 1043 struct fuse_dev *fud; 1044 struct fuse_conn *fc; 1045 struct inode *root; 1046 struct fuse_mount_data d; 1047 struct file *file; 1048 struct dentry *root_dentry; 1049 struct fuse_req *init_req; 1050 int err; 1051 int is_bdev = sb->s_bdev != NULL; 1052 1053 err = -EINVAL; 1054 if (sb->s_flags & MS_MANDLOCK) 1055 goto err; 1056 1057 sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION); 1058 1059 if (!parse_fuse_opt(data, &d, is_bdev)) 1060 goto err; 1061 1062 if (is_bdev) { 1063 #ifdef CONFIG_BLOCK 1064 err = -EINVAL; 1065 if (!sb_set_blocksize(sb, d.blksize)) 1066 goto err; 1067 #endif 1068 } else { 1069 sb->s_blocksize = PAGE_SIZE; 1070 sb->s_blocksize_bits = PAGE_SHIFT; 1071 } 1072 sb->s_magic = FUSE_SUPER_MAGIC; 1073 sb->s_op = &fuse_super_operations; 1074 sb->s_maxbytes = MAX_LFS_FILESIZE; 1075 sb->s_time_gran = 1; 1076 sb->s_export_op = &fuse_export_operations; 1077 1078 file = fget(d.fd); 1079 err = -EINVAL; 1080 if (!file) 1081 goto err; 1082 1083 if ((file->f_op != &fuse_dev_operations) || 1084 (file->f_cred->user_ns != &init_user_ns)) 1085 goto err_fput; 1086 1087 fc = kmalloc(sizeof(*fc), GFP_KERNEL); 1088 err = -ENOMEM; 1089 if (!fc) 1090 goto err_fput; 1091 1092 fuse_conn_init(fc); 1093 fc->release = fuse_free_conn; 1094 1095 fud = fuse_dev_alloc(fc); 1096 if (!fud) 1097 goto err_put_conn; 1098 1099 fc->dev = sb->s_dev; 1100 fc->sb = sb; 1101 err = fuse_bdi_init(fc, sb); 1102 if (err) 1103 goto err_dev_free; 1104 1105 sb->s_bdi = &fc->bdi; 1106 1107 /* Handle umasking inside the fuse code */ 1108 if (sb->s_flags & MS_POSIXACL) 1109 fc->dont_mask = 1; 1110 sb->s_flags |= MS_POSIXACL; 1111 1112 fc->flags = d.flags; 1113 fc->user_id = d.user_id; 1114 fc->group_id = d.group_id; 1115 fc->max_read = max_t(unsigned, 4096, d.max_read); 1116 1117 /* Used by get_root_inode() */ 1118 sb->s_fs_info = fc; 1119 1120 err = -ENOMEM; 1121 root = fuse_get_root_inode(sb, d.rootmode); 1122 root_dentry = d_make_root(root); 1123 if (!root_dentry) 1124 goto err_dev_free; 1125 /* only now - we want root dentry with NULL ->d_op */ 1126 sb->s_d_op = &fuse_dentry_operations; 1127 1128 init_req = fuse_request_alloc(0); 1129 if (!init_req) 1130 goto err_put_root; 1131 __set_bit(FR_BACKGROUND, &init_req->flags); 1132 1133 if (is_bdev) { 1134 fc->destroy_req = fuse_request_alloc(0); 1135 if (!fc->destroy_req) 1136 goto err_free_init_req; 1137 } 1138 1139 mutex_lock(&fuse_mutex); 1140 err = -EINVAL; 1141 if (file->private_data) 1142 goto err_unlock; 1143 1144 err = fuse_ctl_add_conn(fc); 1145 if (err) 1146 goto err_unlock; 1147 1148 list_add_tail(&fc->entry, &fuse_conn_list); 1149 sb->s_root = root_dentry; 1150 file->private_data = fud; 1151 mutex_unlock(&fuse_mutex); 1152 /* 1153 * atomic_dec_and_test() in fput() provides the necessary 1154 * memory barrier for file->private_data to be visible on all 1155 * CPUs after this 1156 */ 1157 fput(file); 1158 1159 fuse_send_init(fc, init_req); 1160 1161 return 0; 1162 1163 err_unlock: 1164 mutex_unlock(&fuse_mutex); 1165 err_free_init_req: 1166 fuse_request_free(init_req); 1167 err_put_root: 1168 dput(root_dentry); 1169 err_dev_free: 1170 fuse_dev_free(fud); 1171 err_put_conn: 1172 fuse_bdi_destroy(fc); 1173 fuse_conn_put(fc); 1174 err_fput: 1175 fput(file); 1176 err: 1177 return err; 1178 } 1179 1180 static struct dentry *fuse_mount(struct file_system_type *fs_type, 1181 int flags, const char *dev_name, 1182 void *raw_data) 1183 { 1184 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super); 1185 } 1186 1187 static void fuse_kill_sb_anon(struct super_block *sb) 1188 { 1189 struct fuse_conn *fc = get_fuse_conn_super(sb); 1190 1191 if (fc) { 1192 down_write(&fc->killsb); 1193 fc->sb = NULL; 1194 up_write(&fc->killsb); 1195 } 1196 1197 kill_anon_super(sb); 1198 } 1199 1200 static struct file_system_type fuse_fs_type = { 1201 .owner = THIS_MODULE, 1202 .name = "fuse", 1203 .fs_flags = FS_HAS_SUBTYPE, 1204 .mount = fuse_mount, 1205 .kill_sb = fuse_kill_sb_anon, 1206 }; 1207 MODULE_ALIAS_FS("fuse"); 1208 1209 #ifdef CONFIG_BLOCK 1210 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type, 1211 int flags, const char *dev_name, 1212 void *raw_data) 1213 { 1214 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super); 1215 } 1216 1217 static void fuse_kill_sb_blk(struct super_block *sb) 1218 { 1219 struct fuse_conn *fc = get_fuse_conn_super(sb); 1220 1221 if (fc) { 1222 down_write(&fc->killsb); 1223 fc->sb = NULL; 1224 up_write(&fc->killsb); 1225 } 1226 1227 kill_block_super(sb); 1228 } 1229 1230 static struct file_system_type fuseblk_fs_type = { 1231 .owner = THIS_MODULE, 1232 .name = "fuseblk", 1233 .mount = fuse_mount_blk, 1234 .kill_sb = fuse_kill_sb_blk, 1235 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE, 1236 }; 1237 MODULE_ALIAS_FS("fuseblk"); 1238 1239 static inline int register_fuseblk(void) 1240 { 1241 return register_filesystem(&fuseblk_fs_type); 1242 } 1243 1244 static inline void unregister_fuseblk(void) 1245 { 1246 unregister_filesystem(&fuseblk_fs_type); 1247 } 1248 #else 1249 static inline int register_fuseblk(void) 1250 { 1251 return 0; 1252 } 1253 1254 static inline void unregister_fuseblk(void) 1255 { 1256 } 1257 #endif 1258 1259 static void fuse_inode_init_once(void *foo) 1260 { 1261 struct inode *inode = foo; 1262 1263 inode_init_once(inode); 1264 } 1265 1266 static int __init fuse_fs_init(void) 1267 { 1268 int err; 1269 1270 fuse_inode_cachep = kmem_cache_create("fuse_inode", 1271 sizeof(struct fuse_inode), 0, 1272 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, 1273 fuse_inode_init_once); 1274 err = -ENOMEM; 1275 if (!fuse_inode_cachep) 1276 goto out; 1277 1278 err = register_fuseblk(); 1279 if (err) 1280 goto out2; 1281 1282 err = register_filesystem(&fuse_fs_type); 1283 if (err) 1284 goto out3; 1285 1286 return 0; 1287 1288 out3: 1289 unregister_fuseblk(); 1290 out2: 1291 kmem_cache_destroy(fuse_inode_cachep); 1292 out: 1293 return err; 1294 } 1295 1296 static void fuse_fs_cleanup(void) 1297 { 1298 unregister_filesystem(&fuse_fs_type); 1299 unregister_fuseblk(); 1300 1301 /* 1302 * Make sure all delayed rcu free inodes are flushed before we 1303 * destroy cache. 1304 */ 1305 rcu_barrier(); 1306 kmem_cache_destroy(fuse_inode_cachep); 1307 } 1308 1309 static struct kobject *fuse_kobj; 1310 1311 static int fuse_sysfs_init(void) 1312 { 1313 int err; 1314 1315 fuse_kobj = kobject_create_and_add("fuse", fs_kobj); 1316 if (!fuse_kobj) { 1317 err = -ENOMEM; 1318 goto out_err; 1319 } 1320 1321 err = sysfs_create_mount_point(fuse_kobj, "connections"); 1322 if (err) 1323 goto out_fuse_unregister; 1324 1325 return 0; 1326 1327 out_fuse_unregister: 1328 kobject_put(fuse_kobj); 1329 out_err: 1330 return err; 1331 } 1332 1333 static void fuse_sysfs_cleanup(void) 1334 { 1335 sysfs_remove_mount_point(fuse_kobj, "connections"); 1336 kobject_put(fuse_kobj); 1337 } 1338 1339 static int __init fuse_init(void) 1340 { 1341 int res; 1342 1343 printk(KERN_INFO "fuse init (API version %i.%i)\n", 1344 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION); 1345 1346 INIT_LIST_HEAD(&fuse_conn_list); 1347 res = fuse_fs_init(); 1348 if (res) 1349 goto err; 1350 1351 res = fuse_dev_init(); 1352 if (res) 1353 goto err_fs_cleanup; 1354 1355 res = fuse_sysfs_init(); 1356 if (res) 1357 goto err_dev_cleanup; 1358 1359 res = fuse_ctl_init(); 1360 if (res) 1361 goto err_sysfs_cleanup; 1362 1363 sanitize_global_limit(&max_user_bgreq); 1364 sanitize_global_limit(&max_user_congthresh); 1365 1366 return 0; 1367 1368 err_sysfs_cleanup: 1369 fuse_sysfs_cleanup(); 1370 err_dev_cleanup: 1371 fuse_dev_cleanup(); 1372 err_fs_cleanup: 1373 fuse_fs_cleanup(); 1374 err: 1375 return res; 1376 } 1377 1378 static void __exit fuse_exit(void) 1379 { 1380 printk(KERN_DEBUG "fuse exit\n"); 1381 1382 fuse_ctl_cleanup(); 1383 fuse_sysfs_cleanup(); 1384 fuse_fs_cleanup(); 1385 fuse_dev_cleanup(); 1386 } 1387 1388 module_init(fuse_init); 1389 module_exit(fuse_exit); 1390