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_SHIFT; 343 if (len <= 0) 344 pg_end = -1; 345 else 346 pg_end = (offset + len - 1) >> PAGE_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 atomic_set(&fc->dev_count, 1); 596 init_waitqueue_head(&fc->blocked_waitq); 597 init_waitqueue_head(&fc->reserved_req_waitq); 598 fuse_iqueue_init(&fc->iq); 599 INIT_LIST_HEAD(&fc->bg_queue); 600 INIT_LIST_HEAD(&fc->entry); 601 INIT_LIST_HEAD(&fc->devices); 602 atomic_set(&fc->num_waiting, 0); 603 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND; 604 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD; 605 fc->khctr = 0; 606 fc->polled_files = RB_ROOT; 607 fc->blocked = 0; 608 fc->initialized = 0; 609 fc->connected = 1; 610 fc->attr_version = 1; 611 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key)); 612 } 613 EXPORT_SYMBOL_GPL(fuse_conn_init); 614 615 void fuse_conn_put(struct fuse_conn *fc) 616 { 617 if (atomic_dec_and_test(&fc->count)) { 618 if (fc->destroy_req) 619 fuse_request_free(fc->destroy_req); 620 fc->release(fc); 621 } 622 } 623 EXPORT_SYMBOL_GPL(fuse_conn_put); 624 625 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc) 626 { 627 atomic_inc(&fc->count); 628 return fc; 629 } 630 EXPORT_SYMBOL_GPL(fuse_conn_get); 631 632 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode) 633 { 634 struct fuse_attr attr; 635 memset(&attr, 0, sizeof(attr)); 636 637 attr.mode = mode; 638 attr.ino = FUSE_ROOT_ID; 639 attr.nlink = 1; 640 return fuse_iget(sb, 1, 0, &attr, 0, 0); 641 } 642 643 struct fuse_inode_handle { 644 u64 nodeid; 645 u32 generation; 646 }; 647 648 static struct dentry *fuse_get_dentry(struct super_block *sb, 649 struct fuse_inode_handle *handle) 650 { 651 struct fuse_conn *fc = get_fuse_conn_super(sb); 652 struct inode *inode; 653 struct dentry *entry; 654 int err = -ESTALE; 655 656 if (handle->nodeid == 0) 657 goto out_err; 658 659 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid); 660 if (!inode) { 661 struct fuse_entry_out outarg; 662 struct qstr name; 663 664 if (!fc->export_support) 665 goto out_err; 666 667 name.len = 1; 668 name.name = "."; 669 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg, 670 &inode); 671 if (err && err != -ENOENT) 672 goto out_err; 673 if (err || !inode) { 674 err = -ESTALE; 675 goto out_err; 676 } 677 err = -EIO; 678 if (get_node_id(inode) != handle->nodeid) 679 goto out_iput; 680 } 681 err = -ESTALE; 682 if (inode->i_generation != handle->generation) 683 goto out_iput; 684 685 entry = d_obtain_alias(inode); 686 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) 687 fuse_invalidate_entry_cache(entry); 688 689 return entry; 690 691 out_iput: 692 iput(inode); 693 out_err: 694 return ERR_PTR(err); 695 } 696 697 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len, 698 struct inode *parent) 699 { 700 int len = parent ? 6 : 3; 701 u64 nodeid; 702 u32 generation; 703 704 if (*max_len < len) { 705 *max_len = len; 706 return FILEID_INVALID; 707 } 708 709 nodeid = get_fuse_inode(inode)->nodeid; 710 generation = inode->i_generation; 711 712 fh[0] = (u32)(nodeid >> 32); 713 fh[1] = (u32)(nodeid & 0xffffffff); 714 fh[2] = generation; 715 716 if (parent) { 717 nodeid = get_fuse_inode(parent)->nodeid; 718 generation = parent->i_generation; 719 720 fh[3] = (u32)(nodeid >> 32); 721 fh[4] = (u32)(nodeid & 0xffffffff); 722 fh[5] = generation; 723 } 724 725 *max_len = len; 726 return parent ? 0x82 : 0x81; 727 } 728 729 static struct dentry *fuse_fh_to_dentry(struct super_block *sb, 730 struct fid *fid, int fh_len, int fh_type) 731 { 732 struct fuse_inode_handle handle; 733 734 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3) 735 return NULL; 736 737 handle.nodeid = (u64) fid->raw[0] << 32; 738 handle.nodeid |= (u64) fid->raw[1]; 739 handle.generation = fid->raw[2]; 740 return fuse_get_dentry(sb, &handle); 741 } 742 743 static struct dentry *fuse_fh_to_parent(struct super_block *sb, 744 struct fid *fid, int fh_len, int fh_type) 745 { 746 struct fuse_inode_handle parent; 747 748 if (fh_type != 0x82 || fh_len < 6) 749 return NULL; 750 751 parent.nodeid = (u64) fid->raw[3] << 32; 752 parent.nodeid |= (u64) fid->raw[4]; 753 parent.generation = fid->raw[5]; 754 return fuse_get_dentry(sb, &parent); 755 } 756 757 static struct dentry *fuse_get_parent(struct dentry *child) 758 { 759 struct inode *child_inode = d_inode(child); 760 struct fuse_conn *fc = get_fuse_conn(child_inode); 761 struct inode *inode; 762 struct dentry *parent; 763 struct fuse_entry_out outarg; 764 struct qstr name; 765 int err; 766 767 if (!fc->export_support) 768 return ERR_PTR(-ESTALE); 769 770 name.len = 2; 771 name.name = ".."; 772 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode), 773 &name, &outarg, &inode); 774 if (err) { 775 if (err == -ENOENT) 776 return ERR_PTR(-ESTALE); 777 return ERR_PTR(err); 778 } 779 780 parent = d_obtain_alias(inode); 781 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) 782 fuse_invalidate_entry_cache(parent); 783 784 return parent; 785 } 786 787 static const struct export_operations fuse_export_operations = { 788 .fh_to_dentry = fuse_fh_to_dentry, 789 .fh_to_parent = fuse_fh_to_parent, 790 .encode_fh = fuse_encode_fh, 791 .get_parent = fuse_get_parent, 792 }; 793 794 static const struct super_operations fuse_super_operations = { 795 .alloc_inode = fuse_alloc_inode, 796 .destroy_inode = fuse_destroy_inode, 797 .evict_inode = fuse_evict_inode, 798 .write_inode = fuse_write_inode, 799 .drop_inode = generic_delete_inode, 800 .remount_fs = fuse_remount_fs, 801 .put_super = fuse_put_super, 802 .umount_begin = fuse_umount_begin, 803 .statfs = fuse_statfs, 804 .show_options = fuse_show_options, 805 }; 806 807 static void sanitize_global_limit(unsigned *limit) 808 { 809 if (*limit == 0) 810 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) / 811 sizeof(struct fuse_req); 812 813 if (*limit >= 1 << 16) 814 *limit = (1 << 16) - 1; 815 } 816 817 static int set_global_limit(const char *val, struct kernel_param *kp) 818 { 819 int rv; 820 821 rv = param_set_uint(val, kp); 822 if (rv) 823 return rv; 824 825 sanitize_global_limit((unsigned *)kp->arg); 826 827 return 0; 828 } 829 830 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg) 831 { 832 int cap_sys_admin = capable(CAP_SYS_ADMIN); 833 834 if (arg->minor < 13) 835 return; 836 837 sanitize_global_limit(&max_user_bgreq); 838 sanitize_global_limit(&max_user_congthresh); 839 840 if (arg->max_background) { 841 fc->max_background = arg->max_background; 842 843 if (!cap_sys_admin && fc->max_background > max_user_bgreq) 844 fc->max_background = max_user_bgreq; 845 } 846 if (arg->congestion_threshold) { 847 fc->congestion_threshold = arg->congestion_threshold; 848 849 if (!cap_sys_admin && 850 fc->congestion_threshold > max_user_congthresh) 851 fc->congestion_threshold = max_user_congthresh; 852 } 853 } 854 855 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req) 856 { 857 struct fuse_init_out *arg = &req->misc.init_out; 858 859 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION) 860 fc->conn_error = 1; 861 else { 862 unsigned long ra_pages; 863 864 process_init_limits(fc, arg); 865 866 if (arg->minor >= 6) { 867 ra_pages = arg->max_readahead / PAGE_SIZE; 868 if (arg->flags & FUSE_ASYNC_READ) 869 fc->async_read = 1; 870 if (!(arg->flags & FUSE_POSIX_LOCKS)) 871 fc->no_lock = 1; 872 if (arg->minor >= 17) { 873 if (!(arg->flags & FUSE_FLOCK_LOCKS)) 874 fc->no_flock = 1; 875 } else { 876 if (!(arg->flags & FUSE_POSIX_LOCKS)) 877 fc->no_flock = 1; 878 } 879 if (arg->flags & FUSE_ATOMIC_O_TRUNC) 880 fc->atomic_o_trunc = 1; 881 if (arg->minor >= 9) { 882 /* LOOKUP has dependency on proto version */ 883 if (arg->flags & FUSE_EXPORT_SUPPORT) 884 fc->export_support = 1; 885 } 886 if (arg->flags & FUSE_BIG_WRITES) 887 fc->big_writes = 1; 888 if (arg->flags & FUSE_DONT_MASK) 889 fc->dont_mask = 1; 890 if (arg->flags & FUSE_AUTO_INVAL_DATA) 891 fc->auto_inval_data = 1; 892 if (arg->flags & FUSE_DO_READDIRPLUS) { 893 fc->do_readdirplus = 1; 894 if (arg->flags & FUSE_READDIRPLUS_AUTO) 895 fc->readdirplus_auto = 1; 896 } 897 if (arg->flags & FUSE_ASYNC_DIO) 898 fc->async_dio = 1; 899 if (arg->flags & FUSE_WRITEBACK_CACHE) 900 fc->writeback_cache = 1; 901 if (arg->time_gran && arg->time_gran <= 1000000000) 902 fc->sb->s_time_gran = arg->time_gran; 903 } else { 904 ra_pages = fc->max_read / PAGE_SIZE; 905 fc->no_lock = 1; 906 fc->no_flock = 1; 907 } 908 909 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages); 910 fc->minor = arg->minor; 911 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write; 912 fc->max_write = max_t(unsigned, 4096, fc->max_write); 913 fc->conn_init = 1; 914 } 915 fuse_set_initialized(fc); 916 wake_up_all(&fc->blocked_waitq); 917 } 918 919 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req) 920 { 921 struct fuse_init_in *arg = &req->misc.init_in; 922 923 arg->major = FUSE_KERNEL_VERSION; 924 arg->minor = FUSE_KERNEL_MINOR_VERSION; 925 arg->max_readahead = fc->bdi.ra_pages * PAGE_SIZE; 926 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC | 927 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK | 928 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ | 929 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA | 930 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO | 931 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT; 932 req->in.h.opcode = FUSE_INIT; 933 req->in.numargs = 1; 934 req->in.args[0].size = sizeof(*arg); 935 req->in.args[0].value = arg; 936 req->out.numargs = 1; 937 /* Variable length argument used for backward compatibility 938 with interface version < 7.5. Rest of init_out is zeroed 939 by do_get_request(), so a short reply is not a problem */ 940 req->out.argvar = 1; 941 req->out.args[0].size = sizeof(struct fuse_init_out); 942 req->out.args[0].value = &req->misc.init_out; 943 req->end = process_init_reply; 944 fuse_request_send_background(fc, req); 945 } 946 947 static void fuse_free_conn(struct fuse_conn *fc) 948 { 949 WARN_ON(!list_empty(&fc->devices)); 950 kfree_rcu(fc, rcu); 951 } 952 953 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb) 954 { 955 int err; 956 957 fc->bdi.name = "fuse"; 958 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; 959 /* fuse does it's own writeback accounting */ 960 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT; 961 962 err = bdi_init(&fc->bdi); 963 if (err) 964 return err; 965 966 fc->bdi_initialized = 1; 967 968 if (sb->s_bdev) { 969 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk", 970 MAJOR(fc->dev), MINOR(fc->dev)); 971 } else { 972 err = bdi_register_dev(&fc->bdi, fc->dev); 973 } 974 975 if (err) 976 return err; 977 978 /* 979 * For a single fuse filesystem use max 1% of dirty + 980 * writeback threshold. 981 * 982 * This gives about 1M of write buffer for memory maps on a 983 * machine with 1G and 10% dirty_ratio, which should be more 984 * than enough. 985 * 986 * Privileged users can raise it by writing to 987 * 988 * /sys/class/bdi/<bdi>/max_ratio 989 */ 990 bdi_set_max_ratio(&fc->bdi, 1); 991 992 return 0; 993 } 994 995 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc) 996 { 997 struct fuse_dev *fud; 998 999 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL); 1000 if (fud) { 1001 fud->fc = fuse_conn_get(fc); 1002 fuse_pqueue_init(&fud->pq); 1003 1004 spin_lock(&fc->lock); 1005 list_add_tail(&fud->entry, &fc->devices); 1006 spin_unlock(&fc->lock); 1007 } 1008 1009 return fud; 1010 } 1011 EXPORT_SYMBOL_GPL(fuse_dev_alloc); 1012 1013 void fuse_dev_free(struct fuse_dev *fud) 1014 { 1015 struct fuse_conn *fc = fud->fc; 1016 1017 if (fc) { 1018 spin_lock(&fc->lock); 1019 list_del(&fud->entry); 1020 spin_unlock(&fc->lock); 1021 1022 fuse_conn_put(fc); 1023 } 1024 kfree(fud); 1025 } 1026 EXPORT_SYMBOL_GPL(fuse_dev_free); 1027 1028 static int fuse_fill_super(struct super_block *sb, void *data, int silent) 1029 { 1030 struct fuse_dev *fud; 1031 struct fuse_conn *fc; 1032 struct inode *root; 1033 struct fuse_mount_data d; 1034 struct file *file; 1035 struct dentry *root_dentry; 1036 struct fuse_req *init_req; 1037 int err; 1038 int is_bdev = sb->s_bdev != NULL; 1039 1040 err = -EINVAL; 1041 if (sb->s_flags & MS_MANDLOCK) 1042 goto err; 1043 1044 sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION); 1045 1046 if (!parse_fuse_opt(data, &d, is_bdev)) 1047 goto err; 1048 1049 if (is_bdev) { 1050 #ifdef CONFIG_BLOCK 1051 err = -EINVAL; 1052 if (!sb_set_blocksize(sb, d.blksize)) 1053 goto err; 1054 #endif 1055 } else { 1056 sb->s_blocksize = PAGE_SIZE; 1057 sb->s_blocksize_bits = PAGE_SHIFT; 1058 } 1059 sb->s_magic = FUSE_SUPER_MAGIC; 1060 sb->s_op = &fuse_super_operations; 1061 sb->s_maxbytes = MAX_LFS_FILESIZE; 1062 sb->s_time_gran = 1; 1063 sb->s_export_op = &fuse_export_operations; 1064 1065 file = fget(d.fd); 1066 err = -EINVAL; 1067 if (!file) 1068 goto err; 1069 1070 if ((file->f_op != &fuse_dev_operations) || 1071 (file->f_cred->user_ns != &init_user_ns)) 1072 goto err_fput; 1073 1074 fc = kmalloc(sizeof(*fc), GFP_KERNEL); 1075 err = -ENOMEM; 1076 if (!fc) 1077 goto err_fput; 1078 1079 fuse_conn_init(fc); 1080 fc->release = fuse_free_conn; 1081 1082 fud = fuse_dev_alloc(fc); 1083 if (!fud) 1084 goto err_put_conn; 1085 1086 fc->dev = sb->s_dev; 1087 fc->sb = sb; 1088 err = fuse_bdi_init(fc, sb); 1089 if (err) 1090 goto err_dev_free; 1091 1092 sb->s_bdi = &fc->bdi; 1093 1094 /* Handle umasking inside the fuse code */ 1095 if (sb->s_flags & MS_POSIXACL) 1096 fc->dont_mask = 1; 1097 sb->s_flags |= MS_POSIXACL; 1098 1099 fc->flags = d.flags; 1100 fc->user_id = d.user_id; 1101 fc->group_id = d.group_id; 1102 fc->max_read = max_t(unsigned, 4096, d.max_read); 1103 1104 /* Used by get_root_inode() */ 1105 sb->s_fs_info = fc; 1106 1107 err = -ENOMEM; 1108 root = fuse_get_root_inode(sb, d.rootmode); 1109 root_dentry = d_make_root(root); 1110 if (!root_dentry) 1111 goto err_dev_free; 1112 /* only now - we want root dentry with NULL ->d_op */ 1113 sb->s_d_op = &fuse_dentry_operations; 1114 1115 init_req = fuse_request_alloc(0); 1116 if (!init_req) 1117 goto err_put_root; 1118 __set_bit(FR_BACKGROUND, &init_req->flags); 1119 1120 if (is_bdev) { 1121 fc->destroy_req = fuse_request_alloc(0); 1122 if (!fc->destroy_req) 1123 goto err_free_init_req; 1124 } 1125 1126 mutex_lock(&fuse_mutex); 1127 err = -EINVAL; 1128 if (file->private_data) 1129 goto err_unlock; 1130 1131 err = fuse_ctl_add_conn(fc); 1132 if (err) 1133 goto err_unlock; 1134 1135 list_add_tail(&fc->entry, &fuse_conn_list); 1136 sb->s_root = root_dentry; 1137 file->private_data = fud; 1138 mutex_unlock(&fuse_mutex); 1139 /* 1140 * atomic_dec_and_test() in fput() provides the necessary 1141 * memory barrier for file->private_data to be visible on all 1142 * CPUs after this 1143 */ 1144 fput(file); 1145 1146 fuse_send_init(fc, init_req); 1147 1148 return 0; 1149 1150 err_unlock: 1151 mutex_unlock(&fuse_mutex); 1152 err_free_init_req: 1153 fuse_request_free(init_req); 1154 err_put_root: 1155 dput(root_dentry); 1156 err_dev_free: 1157 fuse_dev_free(fud); 1158 err_put_conn: 1159 fuse_bdi_destroy(fc); 1160 fuse_conn_put(fc); 1161 err_fput: 1162 fput(file); 1163 err: 1164 return err; 1165 } 1166 1167 static struct dentry *fuse_mount(struct file_system_type *fs_type, 1168 int flags, const char *dev_name, 1169 void *raw_data) 1170 { 1171 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super); 1172 } 1173 1174 static void fuse_kill_sb_anon(struct super_block *sb) 1175 { 1176 struct fuse_conn *fc = get_fuse_conn_super(sb); 1177 1178 if (fc) { 1179 down_write(&fc->killsb); 1180 fc->sb = NULL; 1181 up_write(&fc->killsb); 1182 } 1183 1184 kill_anon_super(sb); 1185 } 1186 1187 static struct file_system_type fuse_fs_type = { 1188 .owner = THIS_MODULE, 1189 .name = "fuse", 1190 .fs_flags = FS_HAS_SUBTYPE, 1191 .mount = fuse_mount, 1192 .kill_sb = fuse_kill_sb_anon, 1193 }; 1194 MODULE_ALIAS_FS("fuse"); 1195 1196 #ifdef CONFIG_BLOCK 1197 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type, 1198 int flags, const char *dev_name, 1199 void *raw_data) 1200 { 1201 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super); 1202 } 1203 1204 static void fuse_kill_sb_blk(struct super_block *sb) 1205 { 1206 struct fuse_conn *fc = get_fuse_conn_super(sb); 1207 1208 if (fc) { 1209 down_write(&fc->killsb); 1210 fc->sb = NULL; 1211 up_write(&fc->killsb); 1212 } 1213 1214 kill_block_super(sb); 1215 } 1216 1217 static struct file_system_type fuseblk_fs_type = { 1218 .owner = THIS_MODULE, 1219 .name = "fuseblk", 1220 .mount = fuse_mount_blk, 1221 .kill_sb = fuse_kill_sb_blk, 1222 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE, 1223 }; 1224 MODULE_ALIAS_FS("fuseblk"); 1225 1226 static inline int register_fuseblk(void) 1227 { 1228 return register_filesystem(&fuseblk_fs_type); 1229 } 1230 1231 static inline void unregister_fuseblk(void) 1232 { 1233 unregister_filesystem(&fuseblk_fs_type); 1234 } 1235 #else 1236 static inline int register_fuseblk(void) 1237 { 1238 return 0; 1239 } 1240 1241 static inline void unregister_fuseblk(void) 1242 { 1243 } 1244 #endif 1245 1246 static void fuse_inode_init_once(void *foo) 1247 { 1248 struct inode *inode = foo; 1249 1250 inode_init_once(inode); 1251 } 1252 1253 static int __init fuse_fs_init(void) 1254 { 1255 int err; 1256 1257 fuse_inode_cachep = kmem_cache_create("fuse_inode", 1258 sizeof(struct fuse_inode), 0, 1259 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, 1260 fuse_inode_init_once); 1261 err = -ENOMEM; 1262 if (!fuse_inode_cachep) 1263 goto out; 1264 1265 err = register_fuseblk(); 1266 if (err) 1267 goto out2; 1268 1269 err = register_filesystem(&fuse_fs_type); 1270 if (err) 1271 goto out3; 1272 1273 return 0; 1274 1275 out3: 1276 unregister_fuseblk(); 1277 out2: 1278 kmem_cache_destroy(fuse_inode_cachep); 1279 out: 1280 return err; 1281 } 1282 1283 static void fuse_fs_cleanup(void) 1284 { 1285 unregister_filesystem(&fuse_fs_type); 1286 unregister_fuseblk(); 1287 1288 /* 1289 * Make sure all delayed rcu free inodes are flushed before we 1290 * destroy cache. 1291 */ 1292 rcu_barrier(); 1293 kmem_cache_destroy(fuse_inode_cachep); 1294 } 1295 1296 static struct kobject *fuse_kobj; 1297 1298 static int fuse_sysfs_init(void) 1299 { 1300 int err; 1301 1302 fuse_kobj = kobject_create_and_add("fuse", fs_kobj); 1303 if (!fuse_kobj) { 1304 err = -ENOMEM; 1305 goto out_err; 1306 } 1307 1308 err = sysfs_create_mount_point(fuse_kobj, "connections"); 1309 if (err) 1310 goto out_fuse_unregister; 1311 1312 return 0; 1313 1314 out_fuse_unregister: 1315 kobject_put(fuse_kobj); 1316 out_err: 1317 return err; 1318 } 1319 1320 static void fuse_sysfs_cleanup(void) 1321 { 1322 sysfs_remove_mount_point(fuse_kobj, "connections"); 1323 kobject_put(fuse_kobj); 1324 } 1325 1326 static int __init fuse_init(void) 1327 { 1328 int res; 1329 1330 printk(KERN_INFO "fuse init (API version %i.%i)\n", 1331 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION); 1332 1333 INIT_LIST_HEAD(&fuse_conn_list); 1334 res = fuse_fs_init(); 1335 if (res) 1336 goto err; 1337 1338 res = fuse_dev_init(); 1339 if (res) 1340 goto err_fs_cleanup; 1341 1342 res = fuse_sysfs_init(); 1343 if (res) 1344 goto err_dev_cleanup; 1345 1346 res = fuse_ctl_init(); 1347 if (res) 1348 goto err_sysfs_cleanup; 1349 1350 sanitize_global_limit(&max_user_bgreq); 1351 sanitize_global_limit(&max_user_congthresh); 1352 1353 return 0; 1354 1355 err_sysfs_cleanup: 1356 fuse_sysfs_cleanup(); 1357 err_dev_cleanup: 1358 fuse_dev_cleanup(); 1359 err_fs_cleanup: 1360 fuse_fs_cleanup(); 1361 err: 1362 return res; 1363 } 1364 1365 static void __exit fuse_exit(void) 1366 { 1367 printk(KERN_DEBUG "fuse exit\n"); 1368 1369 fuse_ctl_cleanup(); 1370 fuse_sysfs_cleanup(); 1371 fuse_fs_cleanup(); 1372 fuse_dev_cleanup(); 1373 } 1374 1375 module_init(fuse_init); 1376 module_exit(fuse_exit); 1377