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/init.h> 12 #include <linux/module.h> 13 #include <linux/poll.h> 14 #include <linux/sched/signal.h> 15 #include <linux/uio.h> 16 #include <linux/miscdevice.h> 17 #include <linux/pagemap.h> 18 #include <linux/file.h> 19 #include <linux/slab.h> 20 #include <linux/pipe_fs_i.h> 21 #include <linux/swap.h> 22 #include <linux/splice.h> 23 #include <linux/sched.h> 24 25 MODULE_ALIAS_MISCDEV(FUSE_MINOR); 26 MODULE_ALIAS("devname:fuse"); 27 28 /* Ordinary requests have even IDs, while interrupts IDs are odd */ 29 #define FUSE_INT_REQ_BIT (1ULL << 0) 30 #define FUSE_REQ_ID_STEP (1ULL << 1) 31 32 static struct kmem_cache *fuse_req_cachep; 33 34 static struct fuse_dev *fuse_get_dev(struct file *file) 35 { 36 /* 37 * Lockless access is OK, because file->private data is set 38 * once during mount and is valid until the file is released. 39 */ 40 return READ_ONCE(file->private_data); 41 } 42 43 static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req) 44 { 45 INIT_LIST_HEAD(&req->list); 46 INIT_LIST_HEAD(&req->intr_entry); 47 init_waitqueue_head(&req->waitq); 48 refcount_set(&req->count, 1); 49 __set_bit(FR_PENDING, &req->flags); 50 req->fm = fm; 51 } 52 53 static struct fuse_req *fuse_request_alloc(struct fuse_mount *fm, gfp_t flags) 54 { 55 struct fuse_req *req = kmem_cache_zalloc(fuse_req_cachep, flags); 56 if (req) 57 fuse_request_init(fm, req); 58 59 return req; 60 } 61 62 static void fuse_request_free(struct fuse_req *req) 63 { 64 kmem_cache_free(fuse_req_cachep, req); 65 } 66 67 static void __fuse_get_request(struct fuse_req *req) 68 { 69 refcount_inc(&req->count); 70 } 71 72 /* Must be called with > 1 refcount */ 73 static void __fuse_put_request(struct fuse_req *req) 74 { 75 refcount_dec(&req->count); 76 } 77 78 void fuse_set_initialized(struct fuse_conn *fc) 79 { 80 /* Make sure stores before this are seen on another CPU */ 81 smp_wmb(); 82 fc->initialized = 1; 83 } 84 85 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background) 86 { 87 return !fc->initialized || (for_background && fc->blocked); 88 } 89 90 static void fuse_drop_waiting(struct fuse_conn *fc) 91 { 92 /* 93 * lockess check of fc->connected is okay, because atomic_dec_and_test() 94 * provides a memory barrier mached with the one in fuse_wait_aborted() 95 * to ensure no wake-up is missed. 96 */ 97 if (atomic_dec_and_test(&fc->num_waiting) && 98 !READ_ONCE(fc->connected)) { 99 /* wake up aborters */ 100 wake_up_all(&fc->blocked_waitq); 101 } 102 } 103 104 static void fuse_put_request(struct fuse_req *req); 105 106 static struct fuse_req *fuse_get_req(struct fuse_mount *fm, bool for_background) 107 { 108 struct fuse_conn *fc = fm->fc; 109 struct fuse_req *req; 110 int err; 111 atomic_inc(&fc->num_waiting); 112 113 if (fuse_block_alloc(fc, for_background)) { 114 err = -EINTR; 115 if (wait_event_killable_exclusive(fc->blocked_waitq, 116 !fuse_block_alloc(fc, for_background))) 117 goto out; 118 } 119 /* Matches smp_wmb() in fuse_set_initialized() */ 120 smp_rmb(); 121 122 err = -ENOTCONN; 123 if (!fc->connected) 124 goto out; 125 126 err = -ECONNREFUSED; 127 if (fc->conn_error) 128 goto out; 129 130 req = fuse_request_alloc(fm, GFP_KERNEL); 131 err = -ENOMEM; 132 if (!req) { 133 if (for_background) 134 wake_up(&fc->blocked_waitq); 135 goto out; 136 } 137 138 req->in.h.uid = from_kuid(fc->user_ns, current_fsuid()); 139 req->in.h.gid = from_kgid(fc->user_ns, current_fsgid()); 140 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns); 141 142 __set_bit(FR_WAITING, &req->flags); 143 if (for_background) 144 __set_bit(FR_BACKGROUND, &req->flags); 145 146 if (unlikely(req->in.h.uid == ((uid_t)-1) || 147 req->in.h.gid == ((gid_t)-1))) { 148 fuse_put_request(req); 149 return ERR_PTR(-EOVERFLOW); 150 } 151 return req; 152 153 out: 154 fuse_drop_waiting(fc); 155 return ERR_PTR(err); 156 } 157 158 static void fuse_put_request(struct fuse_req *req) 159 { 160 struct fuse_conn *fc = req->fm->fc; 161 162 if (refcount_dec_and_test(&req->count)) { 163 if (test_bit(FR_BACKGROUND, &req->flags)) { 164 /* 165 * We get here in the unlikely case that a background 166 * request was allocated but not sent 167 */ 168 spin_lock(&fc->bg_lock); 169 if (!fc->blocked) 170 wake_up(&fc->blocked_waitq); 171 spin_unlock(&fc->bg_lock); 172 } 173 174 if (test_bit(FR_WAITING, &req->flags)) { 175 __clear_bit(FR_WAITING, &req->flags); 176 fuse_drop_waiting(fc); 177 } 178 179 fuse_request_free(req); 180 } 181 } 182 183 unsigned int fuse_len_args(unsigned int numargs, struct fuse_arg *args) 184 { 185 unsigned nbytes = 0; 186 unsigned i; 187 188 for (i = 0; i < numargs; i++) 189 nbytes += args[i].size; 190 191 return nbytes; 192 } 193 EXPORT_SYMBOL_GPL(fuse_len_args); 194 195 u64 fuse_get_unique(struct fuse_iqueue *fiq) 196 { 197 fiq->reqctr += FUSE_REQ_ID_STEP; 198 return fiq->reqctr; 199 } 200 EXPORT_SYMBOL_GPL(fuse_get_unique); 201 202 static unsigned int fuse_req_hash(u64 unique) 203 { 204 return hash_long(unique & ~FUSE_INT_REQ_BIT, FUSE_PQ_HASH_BITS); 205 } 206 207 /** 208 * A new request is available, wake fiq->waitq 209 */ 210 static void fuse_dev_wake_and_unlock(struct fuse_iqueue *fiq) 211 __releases(fiq->lock) 212 { 213 wake_up(&fiq->waitq); 214 kill_fasync(&fiq->fasync, SIGIO, POLL_IN); 215 spin_unlock(&fiq->lock); 216 } 217 218 const struct fuse_iqueue_ops fuse_dev_fiq_ops = { 219 .wake_forget_and_unlock = fuse_dev_wake_and_unlock, 220 .wake_interrupt_and_unlock = fuse_dev_wake_and_unlock, 221 .wake_pending_and_unlock = fuse_dev_wake_and_unlock, 222 }; 223 EXPORT_SYMBOL_GPL(fuse_dev_fiq_ops); 224 225 static void queue_request_and_unlock(struct fuse_iqueue *fiq, 226 struct fuse_req *req) 227 __releases(fiq->lock) 228 { 229 req->in.h.len = sizeof(struct fuse_in_header) + 230 fuse_len_args(req->args->in_numargs, 231 (struct fuse_arg *) req->args->in_args); 232 list_add_tail(&req->list, &fiq->pending); 233 fiq->ops->wake_pending_and_unlock(fiq); 234 } 235 236 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget, 237 u64 nodeid, u64 nlookup) 238 { 239 struct fuse_iqueue *fiq = &fc->iq; 240 241 forget->forget_one.nodeid = nodeid; 242 forget->forget_one.nlookup = nlookup; 243 244 spin_lock(&fiq->lock); 245 if (fiq->connected) { 246 fiq->forget_list_tail->next = forget; 247 fiq->forget_list_tail = forget; 248 fiq->ops->wake_forget_and_unlock(fiq); 249 } else { 250 kfree(forget); 251 spin_unlock(&fiq->lock); 252 } 253 } 254 255 static void flush_bg_queue(struct fuse_conn *fc) 256 { 257 struct fuse_iqueue *fiq = &fc->iq; 258 259 while (fc->active_background < fc->max_background && 260 !list_empty(&fc->bg_queue)) { 261 struct fuse_req *req; 262 263 req = list_first_entry(&fc->bg_queue, struct fuse_req, list); 264 list_del(&req->list); 265 fc->active_background++; 266 spin_lock(&fiq->lock); 267 req->in.h.unique = fuse_get_unique(fiq); 268 queue_request_and_unlock(fiq, req); 269 } 270 } 271 272 /* 273 * This function is called when a request is finished. Either a reply 274 * has arrived or it was aborted (and not yet sent) or some error 275 * occurred during communication with userspace, or the device file 276 * was closed. The requester thread is woken up (if still waiting), 277 * the 'end' callback is called if given, else the reference to the 278 * request is released 279 */ 280 void fuse_request_end(struct fuse_req *req) 281 { 282 struct fuse_mount *fm = req->fm; 283 struct fuse_conn *fc = fm->fc; 284 struct fuse_iqueue *fiq = &fc->iq; 285 286 if (test_and_set_bit(FR_FINISHED, &req->flags)) 287 goto put_request; 288 289 /* 290 * test_and_set_bit() implies smp_mb() between bit 291 * changing and below intr_entry check. Pairs with 292 * smp_mb() from queue_interrupt(). 293 */ 294 if (!list_empty(&req->intr_entry)) { 295 spin_lock(&fiq->lock); 296 list_del_init(&req->intr_entry); 297 spin_unlock(&fiq->lock); 298 } 299 WARN_ON(test_bit(FR_PENDING, &req->flags)); 300 WARN_ON(test_bit(FR_SENT, &req->flags)); 301 if (test_bit(FR_BACKGROUND, &req->flags)) { 302 spin_lock(&fc->bg_lock); 303 clear_bit(FR_BACKGROUND, &req->flags); 304 if (fc->num_background == fc->max_background) { 305 fc->blocked = 0; 306 wake_up(&fc->blocked_waitq); 307 } else if (!fc->blocked) { 308 /* 309 * Wake up next waiter, if any. It's okay to use 310 * waitqueue_active(), as we've already synced up 311 * fc->blocked with waiters with the wake_up() call 312 * above. 313 */ 314 if (waitqueue_active(&fc->blocked_waitq)) 315 wake_up(&fc->blocked_waitq); 316 } 317 318 if (fc->num_background == fc->congestion_threshold && fm->sb) { 319 clear_bdi_congested(fm->sb->s_bdi, BLK_RW_SYNC); 320 clear_bdi_congested(fm->sb->s_bdi, BLK_RW_ASYNC); 321 } 322 fc->num_background--; 323 fc->active_background--; 324 flush_bg_queue(fc); 325 spin_unlock(&fc->bg_lock); 326 } else { 327 /* Wake up waiter sleeping in request_wait_answer() */ 328 wake_up(&req->waitq); 329 } 330 331 if (test_bit(FR_ASYNC, &req->flags)) 332 req->args->end(fm, req->args, req->out.h.error); 333 put_request: 334 fuse_put_request(req); 335 } 336 EXPORT_SYMBOL_GPL(fuse_request_end); 337 338 static int queue_interrupt(struct fuse_req *req) 339 { 340 struct fuse_iqueue *fiq = &req->fm->fc->iq; 341 342 spin_lock(&fiq->lock); 343 /* Check for we've sent request to interrupt this req */ 344 if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags))) { 345 spin_unlock(&fiq->lock); 346 return -EINVAL; 347 } 348 349 if (list_empty(&req->intr_entry)) { 350 list_add_tail(&req->intr_entry, &fiq->interrupts); 351 /* 352 * Pairs with smp_mb() implied by test_and_set_bit() 353 * from fuse_request_end(). 354 */ 355 smp_mb(); 356 if (test_bit(FR_FINISHED, &req->flags)) { 357 list_del_init(&req->intr_entry); 358 spin_unlock(&fiq->lock); 359 return 0; 360 } 361 fiq->ops->wake_interrupt_and_unlock(fiq); 362 } else { 363 spin_unlock(&fiq->lock); 364 } 365 return 0; 366 } 367 368 static void request_wait_answer(struct fuse_req *req) 369 { 370 struct fuse_conn *fc = req->fm->fc; 371 struct fuse_iqueue *fiq = &fc->iq; 372 int err; 373 374 if (!fc->no_interrupt) { 375 /* Any signal may interrupt this */ 376 err = wait_event_interruptible(req->waitq, 377 test_bit(FR_FINISHED, &req->flags)); 378 if (!err) 379 return; 380 381 set_bit(FR_INTERRUPTED, &req->flags); 382 /* matches barrier in fuse_dev_do_read() */ 383 smp_mb__after_atomic(); 384 if (test_bit(FR_SENT, &req->flags)) 385 queue_interrupt(req); 386 } 387 388 if (!test_bit(FR_FORCE, &req->flags)) { 389 /* Only fatal signals may interrupt this */ 390 err = wait_event_killable(req->waitq, 391 test_bit(FR_FINISHED, &req->flags)); 392 if (!err) 393 return; 394 395 spin_lock(&fiq->lock); 396 /* Request is not yet in userspace, bail out */ 397 if (test_bit(FR_PENDING, &req->flags)) { 398 list_del(&req->list); 399 spin_unlock(&fiq->lock); 400 __fuse_put_request(req); 401 req->out.h.error = -EINTR; 402 return; 403 } 404 spin_unlock(&fiq->lock); 405 } 406 407 /* 408 * Either request is already in userspace, or it was forced. 409 * Wait it out. 410 */ 411 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags)); 412 } 413 414 static void __fuse_request_send(struct fuse_req *req) 415 { 416 struct fuse_iqueue *fiq = &req->fm->fc->iq; 417 418 BUG_ON(test_bit(FR_BACKGROUND, &req->flags)); 419 spin_lock(&fiq->lock); 420 if (!fiq->connected) { 421 spin_unlock(&fiq->lock); 422 req->out.h.error = -ENOTCONN; 423 } else { 424 req->in.h.unique = fuse_get_unique(fiq); 425 /* acquire extra reference, since request is still needed 426 after fuse_request_end() */ 427 __fuse_get_request(req); 428 queue_request_and_unlock(fiq, req); 429 430 request_wait_answer(req); 431 /* Pairs with smp_wmb() in fuse_request_end() */ 432 smp_rmb(); 433 } 434 } 435 436 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args) 437 { 438 if (fc->minor < 4 && args->opcode == FUSE_STATFS) 439 args->out_args[0].size = FUSE_COMPAT_STATFS_SIZE; 440 441 if (fc->minor < 9) { 442 switch (args->opcode) { 443 case FUSE_LOOKUP: 444 case FUSE_CREATE: 445 case FUSE_MKNOD: 446 case FUSE_MKDIR: 447 case FUSE_SYMLINK: 448 case FUSE_LINK: 449 args->out_args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE; 450 break; 451 case FUSE_GETATTR: 452 case FUSE_SETATTR: 453 args->out_args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE; 454 break; 455 } 456 } 457 if (fc->minor < 12) { 458 switch (args->opcode) { 459 case FUSE_CREATE: 460 args->in_args[0].size = sizeof(struct fuse_open_in); 461 break; 462 case FUSE_MKNOD: 463 args->in_args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE; 464 break; 465 } 466 } 467 } 468 469 static void fuse_force_creds(struct fuse_req *req) 470 { 471 struct fuse_conn *fc = req->fm->fc; 472 473 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid()); 474 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid()); 475 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns); 476 } 477 478 static void fuse_args_to_req(struct fuse_req *req, struct fuse_args *args) 479 { 480 req->in.h.opcode = args->opcode; 481 req->in.h.nodeid = args->nodeid; 482 req->args = args; 483 if (args->end) 484 __set_bit(FR_ASYNC, &req->flags); 485 } 486 487 ssize_t fuse_simple_request(struct fuse_mount *fm, struct fuse_args *args) 488 { 489 struct fuse_conn *fc = fm->fc; 490 struct fuse_req *req; 491 ssize_t ret; 492 493 if (args->force) { 494 atomic_inc(&fc->num_waiting); 495 req = fuse_request_alloc(fm, GFP_KERNEL | __GFP_NOFAIL); 496 497 if (!args->nocreds) 498 fuse_force_creds(req); 499 500 __set_bit(FR_WAITING, &req->flags); 501 __set_bit(FR_FORCE, &req->flags); 502 } else { 503 WARN_ON(args->nocreds); 504 req = fuse_get_req(fm, false); 505 if (IS_ERR(req)) 506 return PTR_ERR(req); 507 } 508 509 /* Needs to be done after fuse_get_req() so that fc->minor is valid */ 510 fuse_adjust_compat(fc, args); 511 fuse_args_to_req(req, args); 512 513 if (!args->noreply) 514 __set_bit(FR_ISREPLY, &req->flags); 515 __fuse_request_send(req); 516 ret = req->out.h.error; 517 if (!ret && args->out_argvar) { 518 BUG_ON(args->out_numargs == 0); 519 ret = args->out_args[args->out_numargs - 1].size; 520 } 521 fuse_put_request(req); 522 523 return ret; 524 } 525 526 static bool fuse_request_queue_background(struct fuse_req *req) 527 { 528 struct fuse_mount *fm = req->fm; 529 struct fuse_conn *fc = fm->fc; 530 bool queued = false; 531 532 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags)); 533 if (!test_bit(FR_WAITING, &req->flags)) { 534 __set_bit(FR_WAITING, &req->flags); 535 atomic_inc(&fc->num_waiting); 536 } 537 __set_bit(FR_ISREPLY, &req->flags); 538 spin_lock(&fc->bg_lock); 539 if (likely(fc->connected)) { 540 fc->num_background++; 541 if (fc->num_background == fc->max_background) 542 fc->blocked = 1; 543 if (fc->num_background == fc->congestion_threshold && fm->sb) { 544 set_bdi_congested(fm->sb->s_bdi, BLK_RW_SYNC); 545 set_bdi_congested(fm->sb->s_bdi, BLK_RW_ASYNC); 546 } 547 list_add_tail(&req->list, &fc->bg_queue); 548 flush_bg_queue(fc); 549 queued = true; 550 } 551 spin_unlock(&fc->bg_lock); 552 553 return queued; 554 } 555 556 int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args, 557 gfp_t gfp_flags) 558 { 559 struct fuse_req *req; 560 561 if (args->force) { 562 WARN_ON(!args->nocreds); 563 req = fuse_request_alloc(fm, gfp_flags); 564 if (!req) 565 return -ENOMEM; 566 __set_bit(FR_BACKGROUND, &req->flags); 567 } else { 568 WARN_ON(args->nocreds); 569 req = fuse_get_req(fm, true); 570 if (IS_ERR(req)) 571 return PTR_ERR(req); 572 } 573 574 fuse_args_to_req(req, args); 575 576 if (!fuse_request_queue_background(req)) { 577 fuse_put_request(req); 578 return -ENOTCONN; 579 } 580 581 return 0; 582 } 583 EXPORT_SYMBOL_GPL(fuse_simple_background); 584 585 static int fuse_simple_notify_reply(struct fuse_mount *fm, 586 struct fuse_args *args, u64 unique) 587 { 588 struct fuse_req *req; 589 struct fuse_iqueue *fiq = &fm->fc->iq; 590 int err = 0; 591 592 req = fuse_get_req(fm, false); 593 if (IS_ERR(req)) 594 return PTR_ERR(req); 595 596 __clear_bit(FR_ISREPLY, &req->flags); 597 req->in.h.unique = unique; 598 599 fuse_args_to_req(req, args); 600 601 spin_lock(&fiq->lock); 602 if (fiq->connected) { 603 queue_request_and_unlock(fiq, req); 604 } else { 605 err = -ENODEV; 606 spin_unlock(&fiq->lock); 607 fuse_put_request(req); 608 } 609 610 return err; 611 } 612 613 /* 614 * Lock the request. Up to the next unlock_request() there mustn't be 615 * anything that could cause a page-fault. If the request was already 616 * aborted bail out. 617 */ 618 static int lock_request(struct fuse_req *req) 619 { 620 int err = 0; 621 if (req) { 622 spin_lock(&req->waitq.lock); 623 if (test_bit(FR_ABORTED, &req->flags)) 624 err = -ENOENT; 625 else 626 set_bit(FR_LOCKED, &req->flags); 627 spin_unlock(&req->waitq.lock); 628 } 629 return err; 630 } 631 632 /* 633 * Unlock request. If it was aborted while locked, caller is responsible 634 * for unlocking and ending the request. 635 */ 636 static int unlock_request(struct fuse_req *req) 637 { 638 int err = 0; 639 if (req) { 640 spin_lock(&req->waitq.lock); 641 if (test_bit(FR_ABORTED, &req->flags)) 642 err = -ENOENT; 643 else 644 clear_bit(FR_LOCKED, &req->flags); 645 spin_unlock(&req->waitq.lock); 646 } 647 return err; 648 } 649 650 struct fuse_copy_state { 651 int write; 652 struct fuse_req *req; 653 struct iov_iter *iter; 654 struct pipe_buffer *pipebufs; 655 struct pipe_buffer *currbuf; 656 struct pipe_inode_info *pipe; 657 unsigned long nr_segs; 658 struct page *pg; 659 unsigned len; 660 unsigned offset; 661 unsigned move_pages:1; 662 }; 663 664 static void fuse_copy_init(struct fuse_copy_state *cs, int write, 665 struct iov_iter *iter) 666 { 667 memset(cs, 0, sizeof(*cs)); 668 cs->write = write; 669 cs->iter = iter; 670 } 671 672 /* Unmap and put previous page of userspace buffer */ 673 static void fuse_copy_finish(struct fuse_copy_state *cs) 674 { 675 if (cs->currbuf) { 676 struct pipe_buffer *buf = cs->currbuf; 677 678 if (cs->write) 679 buf->len = PAGE_SIZE - cs->len; 680 cs->currbuf = NULL; 681 } else if (cs->pg) { 682 if (cs->write) { 683 flush_dcache_page(cs->pg); 684 set_page_dirty_lock(cs->pg); 685 } 686 put_page(cs->pg); 687 } 688 cs->pg = NULL; 689 } 690 691 /* 692 * Get another pagefull of userspace buffer, and map it to kernel 693 * address space, and lock request 694 */ 695 static int fuse_copy_fill(struct fuse_copy_state *cs) 696 { 697 struct page *page; 698 int err; 699 700 err = unlock_request(cs->req); 701 if (err) 702 return err; 703 704 fuse_copy_finish(cs); 705 if (cs->pipebufs) { 706 struct pipe_buffer *buf = cs->pipebufs; 707 708 if (!cs->write) { 709 err = pipe_buf_confirm(cs->pipe, buf); 710 if (err) 711 return err; 712 713 BUG_ON(!cs->nr_segs); 714 cs->currbuf = buf; 715 cs->pg = buf->page; 716 cs->offset = buf->offset; 717 cs->len = buf->len; 718 cs->pipebufs++; 719 cs->nr_segs--; 720 } else { 721 if (cs->nr_segs >= cs->pipe->max_usage) 722 return -EIO; 723 724 page = alloc_page(GFP_HIGHUSER); 725 if (!page) 726 return -ENOMEM; 727 728 buf->page = page; 729 buf->offset = 0; 730 buf->len = 0; 731 732 cs->currbuf = buf; 733 cs->pg = page; 734 cs->offset = 0; 735 cs->len = PAGE_SIZE; 736 cs->pipebufs++; 737 cs->nr_segs++; 738 } 739 } else { 740 size_t off; 741 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off); 742 if (err < 0) 743 return err; 744 BUG_ON(!err); 745 cs->len = err; 746 cs->offset = off; 747 cs->pg = page; 748 iov_iter_advance(cs->iter, err); 749 } 750 751 return lock_request(cs->req); 752 } 753 754 /* Do as much copy to/from userspace buffer as we can */ 755 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) 756 { 757 unsigned ncpy = min(*size, cs->len); 758 if (val) { 759 void *pgaddr = kmap_atomic(cs->pg); 760 void *buf = pgaddr + cs->offset; 761 762 if (cs->write) 763 memcpy(buf, *val, ncpy); 764 else 765 memcpy(*val, buf, ncpy); 766 767 kunmap_atomic(pgaddr); 768 *val += ncpy; 769 } 770 *size -= ncpy; 771 cs->len -= ncpy; 772 cs->offset += ncpy; 773 return ncpy; 774 } 775 776 static int fuse_check_page(struct page *page) 777 { 778 if (page_mapcount(page) || 779 page->mapping != NULL || 780 (page->flags & PAGE_FLAGS_CHECK_AT_PREP & 781 ~(1 << PG_locked | 782 1 << PG_referenced | 783 1 << PG_uptodate | 784 1 << PG_lru | 785 1 << PG_active | 786 1 << PG_workingset | 787 1 << PG_reclaim | 788 1 << PG_waiters))) { 789 dump_page(page, "fuse: trying to steal weird page"); 790 return 1; 791 } 792 return 0; 793 } 794 795 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep) 796 { 797 int err; 798 struct page *oldpage = *pagep; 799 struct page *newpage; 800 struct pipe_buffer *buf = cs->pipebufs; 801 802 get_page(oldpage); 803 err = unlock_request(cs->req); 804 if (err) 805 goto out_put_old; 806 807 fuse_copy_finish(cs); 808 809 err = pipe_buf_confirm(cs->pipe, buf); 810 if (err) 811 goto out_put_old; 812 813 BUG_ON(!cs->nr_segs); 814 cs->currbuf = buf; 815 cs->len = buf->len; 816 cs->pipebufs++; 817 cs->nr_segs--; 818 819 if (cs->len != PAGE_SIZE) 820 goto out_fallback; 821 822 if (!pipe_buf_try_steal(cs->pipe, buf)) 823 goto out_fallback; 824 825 newpage = buf->page; 826 827 if (!PageUptodate(newpage)) 828 SetPageUptodate(newpage); 829 830 ClearPageMappedToDisk(newpage); 831 832 if (fuse_check_page(newpage) != 0) 833 goto out_fallback_unlock; 834 835 /* 836 * This is a new and locked page, it shouldn't be mapped or 837 * have any special flags on it 838 */ 839 if (WARN_ON(page_mapped(oldpage))) 840 goto out_fallback_unlock; 841 if (WARN_ON(page_has_private(oldpage))) 842 goto out_fallback_unlock; 843 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage))) 844 goto out_fallback_unlock; 845 if (WARN_ON(PageMlocked(oldpage))) 846 goto out_fallback_unlock; 847 848 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL); 849 if (err) { 850 unlock_page(newpage); 851 goto out_put_old; 852 } 853 854 get_page(newpage); 855 856 if (!(buf->flags & PIPE_BUF_FLAG_LRU)) 857 lru_cache_add(newpage); 858 859 err = 0; 860 spin_lock(&cs->req->waitq.lock); 861 if (test_bit(FR_ABORTED, &cs->req->flags)) 862 err = -ENOENT; 863 else 864 *pagep = newpage; 865 spin_unlock(&cs->req->waitq.lock); 866 867 if (err) { 868 unlock_page(newpage); 869 put_page(newpage); 870 goto out_put_old; 871 } 872 873 unlock_page(oldpage); 874 /* Drop ref for ap->pages[] array */ 875 put_page(oldpage); 876 cs->len = 0; 877 878 err = 0; 879 out_put_old: 880 /* Drop ref obtained in this function */ 881 put_page(oldpage); 882 return err; 883 884 out_fallback_unlock: 885 unlock_page(newpage); 886 out_fallback: 887 cs->pg = buf->page; 888 cs->offset = buf->offset; 889 890 err = lock_request(cs->req); 891 if (!err) 892 err = 1; 893 894 goto out_put_old; 895 } 896 897 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page, 898 unsigned offset, unsigned count) 899 { 900 struct pipe_buffer *buf; 901 int err; 902 903 if (cs->nr_segs >= cs->pipe->max_usage) 904 return -EIO; 905 906 get_page(page); 907 err = unlock_request(cs->req); 908 if (err) { 909 put_page(page); 910 return err; 911 } 912 913 fuse_copy_finish(cs); 914 915 buf = cs->pipebufs; 916 buf->page = page; 917 buf->offset = offset; 918 buf->len = count; 919 920 cs->pipebufs++; 921 cs->nr_segs++; 922 cs->len = 0; 923 924 return 0; 925 } 926 927 /* 928 * Copy a page in the request to/from the userspace buffer. Must be 929 * done atomically 930 */ 931 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep, 932 unsigned offset, unsigned count, int zeroing) 933 { 934 int err; 935 struct page *page = *pagep; 936 937 if (page && zeroing && count < PAGE_SIZE) 938 clear_highpage(page); 939 940 while (count) { 941 if (cs->write && cs->pipebufs && page) { 942 return fuse_ref_page(cs, page, offset, count); 943 } else if (!cs->len) { 944 if (cs->move_pages && page && 945 offset == 0 && count == PAGE_SIZE) { 946 err = fuse_try_move_page(cs, pagep); 947 if (err <= 0) 948 return err; 949 } else { 950 err = fuse_copy_fill(cs); 951 if (err) 952 return err; 953 } 954 } 955 if (page) { 956 void *mapaddr = kmap_atomic(page); 957 void *buf = mapaddr + offset; 958 offset += fuse_copy_do(cs, &buf, &count); 959 kunmap_atomic(mapaddr); 960 } else 961 offset += fuse_copy_do(cs, NULL, &count); 962 } 963 if (page && !cs->write) 964 flush_dcache_page(page); 965 return 0; 966 } 967 968 /* Copy pages in the request to/from userspace buffer */ 969 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, 970 int zeroing) 971 { 972 unsigned i; 973 struct fuse_req *req = cs->req; 974 struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args); 975 976 977 for (i = 0; i < ap->num_pages && (nbytes || zeroing); i++) { 978 int err; 979 unsigned int offset = ap->descs[i].offset; 980 unsigned int count = min(nbytes, ap->descs[i].length); 981 982 err = fuse_copy_page(cs, &ap->pages[i], offset, count, zeroing); 983 if (err) 984 return err; 985 986 nbytes -= count; 987 } 988 return 0; 989 } 990 991 /* Copy a single argument in the request to/from userspace buffer */ 992 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) 993 { 994 while (size) { 995 if (!cs->len) { 996 int err = fuse_copy_fill(cs); 997 if (err) 998 return err; 999 } 1000 fuse_copy_do(cs, &val, &size); 1001 } 1002 return 0; 1003 } 1004 1005 /* Copy request arguments to/from userspace buffer */ 1006 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, 1007 unsigned argpages, struct fuse_arg *args, 1008 int zeroing) 1009 { 1010 int err = 0; 1011 unsigned i; 1012 1013 for (i = 0; !err && i < numargs; i++) { 1014 struct fuse_arg *arg = &args[i]; 1015 if (i == numargs - 1 && argpages) 1016 err = fuse_copy_pages(cs, arg->size, zeroing); 1017 else 1018 err = fuse_copy_one(cs, arg->value, arg->size); 1019 } 1020 return err; 1021 } 1022 1023 static int forget_pending(struct fuse_iqueue *fiq) 1024 { 1025 return fiq->forget_list_head.next != NULL; 1026 } 1027 1028 static int request_pending(struct fuse_iqueue *fiq) 1029 { 1030 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) || 1031 forget_pending(fiq); 1032 } 1033 1034 /* 1035 * Transfer an interrupt request to userspace 1036 * 1037 * Unlike other requests this is assembled on demand, without a need 1038 * to allocate a separate fuse_req structure. 1039 * 1040 * Called with fiq->lock held, releases it 1041 */ 1042 static int fuse_read_interrupt(struct fuse_iqueue *fiq, 1043 struct fuse_copy_state *cs, 1044 size_t nbytes, struct fuse_req *req) 1045 __releases(fiq->lock) 1046 { 1047 struct fuse_in_header ih; 1048 struct fuse_interrupt_in arg; 1049 unsigned reqsize = sizeof(ih) + sizeof(arg); 1050 int err; 1051 1052 list_del_init(&req->intr_entry); 1053 memset(&ih, 0, sizeof(ih)); 1054 memset(&arg, 0, sizeof(arg)); 1055 ih.len = reqsize; 1056 ih.opcode = FUSE_INTERRUPT; 1057 ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT); 1058 arg.unique = req->in.h.unique; 1059 1060 spin_unlock(&fiq->lock); 1061 if (nbytes < reqsize) 1062 return -EINVAL; 1063 1064 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1065 if (!err) 1066 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1067 fuse_copy_finish(cs); 1068 1069 return err ? err : reqsize; 1070 } 1071 1072 struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq, 1073 unsigned int max, 1074 unsigned int *countp) 1075 { 1076 struct fuse_forget_link *head = fiq->forget_list_head.next; 1077 struct fuse_forget_link **newhead = &head; 1078 unsigned count; 1079 1080 for (count = 0; *newhead != NULL && count < max; count++) 1081 newhead = &(*newhead)->next; 1082 1083 fiq->forget_list_head.next = *newhead; 1084 *newhead = NULL; 1085 if (fiq->forget_list_head.next == NULL) 1086 fiq->forget_list_tail = &fiq->forget_list_head; 1087 1088 if (countp != NULL) 1089 *countp = count; 1090 1091 return head; 1092 } 1093 EXPORT_SYMBOL(fuse_dequeue_forget); 1094 1095 static int fuse_read_single_forget(struct fuse_iqueue *fiq, 1096 struct fuse_copy_state *cs, 1097 size_t nbytes) 1098 __releases(fiq->lock) 1099 { 1100 int err; 1101 struct fuse_forget_link *forget = fuse_dequeue_forget(fiq, 1, NULL); 1102 struct fuse_forget_in arg = { 1103 .nlookup = forget->forget_one.nlookup, 1104 }; 1105 struct fuse_in_header ih = { 1106 .opcode = FUSE_FORGET, 1107 .nodeid = forget->forget_one.nodeid, 1108 .unique = fuse_get_unique(fiq), 1109 .len = sizeof(ih) + sizeof(arg), 1110 }; 1111 1112 spin_unlock(&fiq->lock); 1113 kfree(forget); 1114 if (nbytes < ih.len) 1115 return -EINVAL; 1116 1117 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1118 if (!err) 1119 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1120 fuse_copy_finish(cs); 1121 1122 if (err) 1123 return err; 1124 1125 return ih.len; 1126 } 1127 1128 static int fuse_read_batch_forget(struct fuse_iqueue *fiq, 1129 struct fuse_copy_state *cs, size_t nbytes) 1130 __releases(fiq->lock) 1131 { 1132 int err; 1133 unsigned max_forgets; 1134 unsigned count; 1135 struct fuse_forget_link *head; 1136 struct fuse_batch_forget_in arg = { .count = 0 }; 1137 struct fuse_in_header ih = { 1138 .opcode = FUSE_BATCH_FORGET, 1139 .unique = fuse_get_unique(fiq), 1140 .len = sizeof(ih) + sizeof(arg), 1141 }; 1142 1143 if (nbytes < ih.len) { 1144 spin_unlock(&fiq->lock); 1145 return -EINVAL; 1146 } 1147 1148 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one); 1149 head = fuse_dequeue_forget(fiq, max_forgets, &count); 1150 spin_unlock(&fiq->lock); 1151 1152 arg.count = count; 1153 ih.len += count * sizeof(struct fuse_forget_one); 1154 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1155 if (!err) 1156 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1157 1158 while (head) { 1159 struct fuse_forget_link *forget = head; 1160 1161 if (!err) { 1162 err = fuse_copy_one(cs, &forget->forget_one, 1163 sizeof(forget->forget_one)); 1164 } 1165 head = forget->next; 1166 kfree(forget); 1167 } 1168 1169 fuse_copy_finish(cs); 1170 1171 if (err) 1172 return err; 1173 1174 return ih.len; 1175 } 1176 1177 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq, 1178 struct fuse_copy_state *cs, 1179 size_t nbytes) 1180 __releases(fiq->lock) 1181 { 1182 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL) 1183 return fuse_read_single_forget(fiq, cs, nbytes); 1184 else 1185 return fuse_read_batch_forget(fiq, cs, nbytes); 1186 } 1187 1188 /* 1189 * Read a single request into the userspace filesystem's buffer. This 1190 * function waits until a request is available, then removes it from 1191 * the pending list and copies request data to userspace buffer. If 1192 * no reply is needed (FORGET) or request has been aborted or there 1193 * was an error during the copying then it's finished by calling 1194 * fuse_request_end(). Otherwise add it to the processing list, and set 1195 * the 'sent' flag. 1196 */ 1197 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file, 1198 struct fuse_copy_state *cs, size_t nbytes) 1199 { 1200 ssize_t err; 1201 struct fuse_conn *fc = fud->fc; 1202 struct fuse_iqueue *fiq = &fc->iq; 1203 struct fuse_pqueue *fpq = &fud->pq; 1204 struct fuse_req *req; 1205 struct fuse_args *args; 1206 unsigned reqsize; 1207 unsigned int hash; 1208 1209 /* 1210 * Require sane minimum read buffer - that has capacity for fixed part 1211 * of any request header + negotiated max_write room for data. 1212 * 1213 * Historically libfuse reserves 4K for fixed header room, but e.g. 1214 * GlusterFS reserves only 80 bytes 1215 * 1216 * = `sizeof(fuse_in_header) + sizeof(fuse_write_in)` 1217 * 1218 * which is the absolute minimum any sane filesystem should be using 1219 * for header room. 1220 */ 1221 if (nbytes < max_t(size_t, FUSE_MIN_READ_BUFFER, 1222 sizeof(struct fuse_in_header) + 1223 sizeof(struct fuse_write_in) + 1224 fc->max_write)) 1225 return -EINVAL; 1226 1227 restart: 1228 for (;;) { 1229 spin_lock(&fiq->lock); 1230 if (!fiq->connected || request_pending(fiq)) 1231 break; 1232 spin_unlock(&fiq->lock); 1233 1234 if (file->f_flags & O_NONBLOCK) 1235 return -EAGAIN; 1236 err = wait_event_interruptible_exclusive(fiq->waitq, 1237 !fiq->connected || request_pending(fiq)); 1238 if (err) 1239 return err; 1240 } 1241 1242 if (!fiq->connected) { 1243 err = fc->aborted ? -ECONNABORTED : -ENODEV; 1244 goto err_unlock; 1245 } 1246 1247 if (!list_empty(&fiq->interrupts)) { 1248 req = list_entry(fiq->interrupts.next, struct fuse_req, 1249 intr_entry); 1250 return fuse_read_interrupt(fiq, cs, nbytes, req); 1251 } 1252 1253 if (forget_pending(fiq)) { 1254 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0) 1255 return fuse_read_forget(fc, fiq, cs, nbytes); 1256 1257 if (fiq->forget_batch <= -8) 1258 fiq->forget_batch = 16; 1259 } 1260 1261 req = list_entry(fiq->pending.next, struct fuse_req, list); 1262 clear_bit(FR_PENDING, &req->flags); 1263 list_del_init(&req->list); 1264 spin_unlock(&fiq->lock); 1265 1266 args = req->args; 1267 reqsize = req->in.h.len; 1268 1269 /* If request is too large, reply with an error and restart the read */ 1270 if (nbytes < reqsize) { 1271 req->out.h.error = -EIO; 1272 /* SETXATTR is special, since it may contain too large data */ 1273 if (args->opcode == FUSE_SETXATTR) 1274 req->out.h.error = -E2BIG; 1275 fuse_request_end(req); 1276 goto restart; 1277 } 1278 spin_lock(&fpq->lock); 1279 /* 1280 * Must not put request on fpq->io queue after having been shut down by 1281 * fuse_abort_conn() 1282 */ 1283 if (!fpq->connected) { 1284 req->out.h.error = err = -ECONNABORTED; 1285 goto out_end; 1286 1287 } 1288 list_add(&req->list, &fpq->io); 1289 spin_unlock(&fpq->lock); 1290 cs->req = req; 1291 err = fuse_copy_one(cs, &req->in.h, sizeof(req->in.h)); 1292 if (!err) 1293 err = fuse_copy_args(cs, args->in_numargs, args->in_pages, 1294 (struct fuse_arg *) args->in_args, 0); 1295 fuse_copy_finish(cs); 1296 spin_lock(&fpq->lock); 1297 clear_bit(FR_LOCKED, &req->flags); 1298 if (!fpq->connected) { 1299 err = fc->aborted ? -ECONNABORTED : -ENODEV; 1300 goto out_end; 1301 } 1302 if (err) { 1303 req->out.h.error = -EIO; 1304 goto out_end; 1305 } 1306 if (!test_bit(FR_ISREPLY, &req->flags)) { 1307 err = reqsize; 1308 goto out_end; 1309 } 1310 hash = fuse_req_hash(req->in.h.unique); 1311 list_move_tail(&req->list, &fpq->processing[hash]); 1312 __fuse_get_request(req); 1313 set_bit(FR_SENT, &req->flags); 1314 spin_unlock(&fpq->lock); 1315 /* matches barrier in request_wait_answer() */ 1316 smp_mb__after_atomic(); 1317 if (test_bit(FR_INTERRUPTED, &req->flags)) 1318 queue_interrupt(req); 1319 fuse_put_request(req); 1320 1321 return reqsize; 1322 1323 out_end: 1324 if (!test_bit(FR_PRIVATE, &req->flags)) 1325 list_del_init(&req->list); 1326 spin_unlock(&fpq->lock); 1327 fuse_request_end(req); 1328 return err; 1329 1330 err_unlock: 1331 spin_unlock(&fiq->lock); 1332 return err; 1333 } 1334 1335 static int fuse_dev_open(struct inode *inode, struct file *file) 1336 { 1337 /* 1338 * The fuse device's file's private_data is used to hold 1339 * the fuse_conn(ection) when it is mounted, and is used to 1340 * keep track of whether the file has been mounted already. 1341 */ 1342 file->private_data = NULL; 1343 return 0; 1344 } 1345 1346 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to) 1347 { 1348 struct fuse_copy_state cs; 1349 struct file *file = iocb->ki_filp; 1350 struct fuse_dev *fud = fuse_get_dev(file); 1351 1352 if (!fud) 1353 return -EPERM; 1354 1355 if (!iter_is_iovec(to)) 1356 return -EINVAL; 1357 1358 fuse_copy_init(&cs, 1, to); 1359 1360 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to)); 1361 } 1362 1363 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos, 1364 struct pipe_inode_info *pipe, 1365 size_t len, unsigned int flags) 1366 { 1367 int total, ret; 1368 int page_nr = 0; 1369 struct pipe_buffer *bufs; 1370 struct fuse_copy_state cs; 1371 struct fuse_dev *fud = fuse_get_dev(in); 1372 1373 if (!fud) 1374 return -EPERM; 1375 1376 bufs = kvmalloc_array(pipe->max_usage, sizeof(struct pipe_buffer), 1377 GFP_KERNEL); 1378 if (!bufs) 1379 return -ENOMEM; 1380 1381 fuse_copy_init(&cs, 1, NULL); 1382 cs.pipebufs = bufs; 1383 cs.pipe = pipe; 1384 ret = fuse_dev_do_read(fud, in, &cs, len); 1385 if (ret < 0) 1386 goto out; 1387 1388 if (pipe_occupancy(pipe->head, pipe->tail) + cs.nr_segs > pipe->max_usage) { 1389 ret = -EIO; 1390 goto out; 1391 } 1392 1393 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) { 1394 /* 1395 * Need to be careful about this. Having buf->ops in module 1396 * code can Oops if the buffer persists after module unload. 1397 */ 1398 bufs[page_nr].ops = &nosteal_pipe_buf_ops; 1399 bufs[page_nr].flags = 0; 1400 ret = add_to_pipe(pipe, &bufs[page_nr++]); 1401 if (unlikely(ret < 0)) 1402 break; 1403 } 1404 if (total) 1405 ret = total; 1406 out: 1407 for (; page_nr < cs.nr_segs; page_nr++) 1408 put_page(bufs[page_nr].page); 1409 1410 kvfree(bufs); 1411 return ret; 1412 } 1413 1414 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size, 1415 struct fuse_copy_state *cs) 1416 { 1417 struct fuse_notify_poll_wakeup_out outarg; 1418 int err = -EINVAL; 1419 1420 if (size != sizeof(outarg)) 1421 goto err; 1422 1423 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1424 if (err) 1425 goto err; 1426 1427 fuse_copy_finish(cs); 1428 return fuse_notify_poll_wakeup(fc, &outarg); 1429 1430 err: 1431 fuse_copy_finish(cs); 1432 return err; 1433 } 1434 1435 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size, 1436 struct fuse_copy_state *cs) 1437 { 1438 struct fuse_notify_inval_inode_out outarg; 1439 int err = -EINVAL; 1440 1441 if (size != sizeof(outarg)) 1442 goto err; 1443 1444 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1445 if (err) 1446 goto err; 1447 fuse_copy_finish(cs); 1448 1449 down_read(&fc->killsb); 1450 err = fuse_reverse_inval_inode(fc, outarg.ino, 1451 outarg.off, outarg.len); 1452 up_read(&fc->killsb); 1453 return err; 1454 1455 err: 1456 fuse_copy_finish(cs); 1457 return err; 1458 } 1459 1460 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size, 1461 struct fuse_copy_state *cs) 1462 { 1463 struct fuse_notify_inval_entry_out outarg; 1464 int err = -ENOMEM; 1465 char *buf; 1466 struct qstr name; 1467 1468 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); 1469 if (!buf) 1470 goto err; 1471 1472 err = -EINVAL; 1473 if (size < sizeof(outarg)) 1474 goto err; 1475 1476 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1477 if (err) 1478 goto err; 1479 1480 err = -ENAMETOOLONG; 1481 if (outarg.namelen > FUSE_NAME_MAX) 1482 goto err; 1483 1484 err = -EINVAL; 1485 if (size != sizeof(outarg) + outarg.namelen + 1) 1486 goto err; 1487 1488 name.name = buf; 1489 name.len = outarg.namelen; 1490 err = fuse_copy_one(cs, buf, outarg.namelen + 1); 1491 if (err) 1492 goto err; 1493 fuse_copy_finish(cs); 1494 buf[outarg.namelen] = 0; 1495 1496 down_read(&fc->killsb); 1497 err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name); 1498 up_read(&fc->killsb); 1499 kfree(buf); 1500 return err; 1501 1502 err: 1503 kfree(buf); 1504 fuse_copy_finish(cs); 1505 return err; 1506 } 1507 1508 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size, 1509 struct fuse_copy_state *cs) 1510 { 1511 struct fuse_notify_delete_out outarg; 1512 int err = -ENOMEM; 1513 char *buf; 1514 struct qstr name; 1515 1516 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); 1517 if (!buf) 1518 goto err; 1519 1520 err = -EINVAL; 1521 if (size < sizeof(outarg)) 1522 goto err; 1523 1524 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1525 if (err) 1526 goto err; 1527 1528 err = -ENAMETOOLONG; 1529 if (outarg.namelen > FUSE_NAME_MAX) 1530 goto err; 1531 1532 err = -EINVAL; 1533 if (size != sizeof(outarg) + outarg.namelen + 1) 1534 goto err; 1535 1536 name.name = buf; 1537 name.len = outarg.namelen; 1538 err = fuse_copy_one(cs, buf, outarg.namelen + 1); 1539 if (err) 1540 goto err; 1541 fuse_copy_finish(cs); 1542 buf[outarg.namelen] = 0; 1543 1544 down_read(&fc->killsb); 1545 err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name); 1546 up_read(&fc->killsb); 1547 kfree(buf); 1548 return err; 1549 1550 err: 1551 kfree(buf); 1552 fuse_copy_finish(cs); 1553 return err; 1554 } 1555 1556 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size, 1557 struct fuse_copy_state *cs) 1558 { 1559 struct fuse_notify_store_out outarg; 1560 struct inode *inode; 1561 struct address_space *mapping; 1562 u64 nodeid; 1563 int err; 1564 pgoff_t index; 1565 unsigned int offset; 1566 unsigned int num; 1567 loff_t file_size; 1568 loff_t end; 1569 1570 err = -EINVAL; 1571 if (size < sizeof(outarg)) 1572 goto out_finish; 1573 1574 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1575 if (err) 1576 goto out_finish; 1577 1578 err = -EINVAL; 1579 if (size - sizeof(outarg) != outarg.size) 1580 goto out_finish; 1581 1582 nodeid = outarg.nodeid; 1583 1584 down_read(&fc->killsb); 1585 1586 err = -ENOENT; 1587 inode = fuse_ilookup(fc, nodeid, NULL); 1588 if (!inode) 1589 goto out_up_killsb; 1590 1591 mapping = inode->i_mapping; 1592 index = outarg.offset >> PAGE_SHIFT; 1593 offset = outarg.offset & ~PAGE_MASK; 1594 file_size = i_size_read(inode); 1595 end = outarg.offset + outarg.size; 1596 if (end > file_size) { 1597 file_size = end; 1598 fuse_write_update_size(inode, file_size); 1599 } 1600 1601 num = outarg.size; 1602 while (num) { 1603 struct page *page; 1604 unsigned int this_num; 1605 1606 err = -ENOMEM; 1607 page = find_or_create_page(mapping, index, 1608 mapping_gfp_mask(mapping)); 1609 if (!page) 1610 goto out_iput; 1611 1612 this_num = min_t(unsigned, num, PAGE_SIZE - offset); 1613 err = fuse_copy_page(cs, &page, offset, this_num, 0); 1614 if (!err && offset == 0 && 1615 (this_num == PAGE_SIZE || file_size == end)) 1616 SetPageUptodate(page); 1617 unlock_page(page); 1618 put_page(page); 1619 1620 if (err) 1621 goto out_iput; 1622 1623 num -= this_num; 1624 offset = 0; 1625 index++; 1626 } 1627 1628 err = 0; 1629 1630 out_iput: 1631 iput(inode); 1632 out_up_killsb: 1633 up_read(&fc->killsb); 1634 out_finish: 1635 fuse_copy_finish(cs); 1636 return err; 1637 } 1638 1639 struct fuse_retrieve_args { 1640 struct fuse_args_pages ap; 1641 struct fuse_notify_retrieve_in inarg; 1642 }; 1643 1644 static void fuse_retrieve_end(struct fuse_mount *fm, struct fuse_args *args, 1645 int error) 1646 { 1647 struct fuse_retrieve_args *ra = 1648 container_of(args, typeof(*ra), ap.args); 1649 1650 release_pages(ra->ap.pages, ra->ap.num_pages); 1651 kfree(ra); 1652 } 1653 1654 static int fuse_retrieve(struct fuse_mount *fm, struct inode *inode, 1655 struct fuse_notify_retrieve_out *outarg) 1656 { 1657 int err; 1658 struct address_space *mapping = inode->i_mapping; 1659 pgoff_t index; 1660 loff_t file_size; 1661 unsigned int num; 1662 unsigned int offset; 1663 size_t total_len = 0; 1664 unsigned int num_pages; 1665 struct fuse_conn *fc = fm->fc; 1666 struct fuse_retrieve_args *ra; 1667 size_t args_size = sizeof(*ra); 1668 struct fuse_args_pages *ap; 1669 struct fuse_args *args; 1670 1671 offset = outarg->offset & ~PAGE_MASK; 1672 file_size = i_size_read(inode); 1673 1674 num = min(outarg->size, fc->max_write); 1675 if (outarg->offset > file_size) 1676 num = 0; 1677 else if (outarg->offset + num > file_size) 1678 num = file_size - outarg->offset; 1679 1680 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; 1681 num_pages = min(num_pages, fc->max_pages); 1682 1683 args_size += num_pages * (sizeof(ap->pages[0]) + sizeof(ap->descs[0])); 1684 1685 ra = kzalloc(args_size, GFP_KERNEL); 1686 if (!ra) 1687 return -ENOMEM; 1688 1689 ap = &ra->ap; 1690 ap->pages = (void *) (ra + 1); 1691 ap->descs = (void *) (ap->pages + num_pages); 1692 1693 args = &ap->args; 1694 args->nodeid = outarg->nodeid; 1695 args->opcode = FUSE_NOTIFY_REPLY; 1696 args->in_numargs = 2; 1697 args->in_pages = true; 1698 args->end = fuse_retrieve_end; 1699 1700 index = outarg->offset >> PAGE_SHIFT; 1701 1702 while (num && ap->num_pages < num_pages) { 1703 struct page *page; 1704 unsigned int this_num; 1705 1706 page = find_get_page(mapping, index); 1707 if (!page) 1708 break; 1709 1710 this_num = min_t(unsigned, num, PAGE_SIZE - offset); 1711 ap->pages[ap->num_pages] = page; 1712 ap->descs[ap->num_pages].offset = offset; 1713 ap->descs[ap->num_pages].length = this_num; 1714 ap->num_pages++; 1715 1716 offset = 0; 1717 num -= this_num; 1718 total_len += this_num; 1719 index++; 1720 } 1721 ra->inarg.offset = outarg->offset; 1722 ra->inarg.size = total_len; 1723 args->in_args[0].size = sizeof(ra->inarg); 1724 args->in_args[0].value = &ra->inarg; 1725 args->in_args[1].size = total_len; 1726 1727 err = fuse_simple_notify_reply(fm, args, outarg->notify_unique); 1728 if (err) 1729 fuse_retrieve_end(fm, args, err); 1730 1731 return err; 1732 } 1733 1734 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size, 1735 struct fuse_copy_state *cs) 1736 { 1737 struct fuse_notify_retrieve_out outarg; 1738 struct fuse_mount *fm; 1739 struct inode *inode; 1740 u64 nodeid; 1741 int err; 1742 1743 err = -EINVAL; 1744 if (size != sizeof(outarg)) 1745 goto copy_finish; 1746 1747 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1748 if (err) 1749 goto copy_finish; 1750 1751 fuse_copy_finish(cs); 1752 1753 down_read(&fc->killsb); 1754 err = -ENOENT; 1755 nodeid = outarg.nodeid; 1756 1757 inode = fuse_ilookup(fc, nodeid, &fm); 1758 if (inode) { 1759 err = fuse_retrieve(fm, inode, &outarg); 1760 iput(inode); 1761 } 1762 up_read(&fc->killsb); 1763 1764 return err; 1765 1766 copy_finish: 1767 fuse_copy_finish(cs); 1768 return err; 1769 } 1770 1771 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code, 1772 unsigned int size, struct fuse_copy_state *cs) 1773 { 1774 /* Don't try to move pages (yet) */ 1775 cs->move_pages = 0; 1776 1777 switch (code) { 1778 case FUSE_NOTIFY_POLL: 1779 return fuse_notify_poll(fc, size, cs); 1780 1781 case FUSE_NOTIFY_INVAL_INODE: 1782 return fuse_notify_inval_inode(fc, size, cs); 1783 1784 case FUSE_NOTIFY_INVAL_ENTRY: 1785 return fuse_notify_inval_entry(fc, size, cs); 1786 1787 case FUSE_NOTIFY_STORE: 1788 return fuse_notify_store(fc, size, cs); 1789 1790 case FUSE_NOTIFY_RETRIEVE: 1791 return fuse_notify_retrieve(fc, size, cs); 1792 1793 case FUSE_NOTIFY_DELETE: 1794 return fuse_notify_delete(fc, size, cs); 1795 1796 default: 1797 fuse_copy_finish(cs); 1798 return -EINVAL; 1799 } 1800 } 1801 1802 /* Look up request on processing list by unique ID */ 1803 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique) 1804 { 1805 unsigned int hash = fuse_req_hash(unique); 1806 struct fuse_req *req; 1807 1808 list_for_each_entry(req, &fpq->processing[hash], list) { 1809 if (req->in.h.unique == unique) 1810 return req; 1811 } 1812 return NULL; 1813 } 1814 1815 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args, 1816 unsigned nbytes) 1817 { 1818 unsigned reqsize = sizeof(struct fuse_out_header); 1819 1820 reqsize += fuse_len_args(args->out_numargs, args->out_args); 1821 1822 if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar)) 1823 return -EINVAL; 1824 else if (reqsize > nbytes) { 1825 struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1]; 1826 unsigned diffsize = reqsize - nbytes; 1827 1828 if (diffsize > lastarg->size) 1829 return -EINVAL; 1830 lastarg->size -= diffsize; 1831 } 1832 return fuse_copy_args(cs, args->out_numargs, args->out_pages, 1833 args->out_args, args->page_zeroing); 1834 } 1835 1836 /* 1837 * Write a single reply to a request. First the header is copied from 1838 * the write buffer. The request is then searched on the processing 1839 * list by the unique ID found in the header. If found, then remove 1840 * it from the list and copy the rest of the buffer to the request. 1841 * The request is finished by calling fuse_request_end(). 1842 */ 1843 static ssize_t fuse_dev_do_write(struct fuse_dev *fud, 1844 struct fuse_copy_state *cs, size_t nbytes) 1845 { 1846 int err; 1847 struct fuse_conn *fc = fud->fc; 1848 struct fuse_pqueue *fpq = &fud->pq; 1849 struct fuse_req *req; 1850 struct fuse_out_header oh; 1851 1852 err = -EINVAL; 1853 if (nbytes < sizeof(struct fuse_out_header)) 1854 goto out; 1855 1856 err = fuse_copy_one(cs, &oh, sizeof(oh)); 1857 if (err) 1858 goto copy_finish; 1859 1860 err = -EINVAL; 1861 if (oh.len != nbytes) 1862 goto copy_finish; 1863 1864 /* 1865 * Zero oh.unique indicates unsolicited notification message 1866 * and error contains notification code. 1867 */ 1868 if (!oh.unique) { 1869 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs); 1870 goto out; 1871 } 1872 1873 err = -EINVAL; 1874 if (oh.error <= -512 || oh.error > 0) 1875 goto copy_finish; 1876 1877 spin_lock(&fpq->lock); 1878 req = NULL; 1879 if (fpq->connected) 1880 req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT); 1881 1882 err = -ENOENT; 1883 if (!req) { 1884 spin_unlock(&fpq->lock); 1885 goto copy_finish; 1886 } 1887 1888 /* Is it an interrupt reply ID? */ 1889 if (oh.unique & FUSE_INT_REQ_BIT) { 1890 __fuse_get_request(req); 1891 spin_unlock(&fpq->lock); 1892 1893 err = 0; 1894 if (nbytes != sizeof(struct fuse_out_header)) 1895 err = -EINVAL; 1896 else if (oh.error == -ENOSYS) 1897 fc->no_interrupt = 1; 1898 else if (oh.error == -EAGAIN) 1899 err = queue_interrupt(req); 1900 1901 fuse_put_request(req); 1902 1903 goto copy_finish; 1904 } 1905 1906 clear_bit(FR_SENT, &req->flags); 1907 list_move(&req->list, &fpq->io); 1908 req->out.h = oh; 1909 set_bit(FR_LOCKED, &req->flags); 1910 spin_unlock(&fpq->lock); 1911 cs->req = req; 1912 if (!req->args->page_replace) 1913 cs->move_pages = 0; 1914 1915 if (oh.error) 1916 err = nbytes != sizeof(oh) ? -EINVAL : 0; 1917 else 1918 err = copy_out_args(cs, req->args, nbytes); 1919 fuse_copy_finish(cs); 1920 1921 spin_lock(&fpq->lock); 1922 clear_bit(FR_LOCKED, &req->flags); 1923 if (!fpq->connected) 1924 err = -ENOENT; 1925 else if (err) 1926 req->out.h.error = -EIO; 1927 if (!test_bit(FR_PRIVATE, &req->flags)) 1928 list_del_init(&req->list); 1929 spin_unlock(&fpq->lock); 1930 1931 fuse_request_end(req); 1932 out: 1933 return err ? err : nbytes; 1934 1935 copy_finish: 1936 fuse_copy_finish(cs); 1937 goto out; 1938 } 1939 1940 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from) 1941 { 1942 struct fuse_copy_state cs; 1943 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp); 1944 1945 if (!fud) 1946 return -EPERM; 1947 1948 if (!iter_is_iovec(from)) 1949 return -EINVAL; 1950 1951 fuse_copy_init(&cs, 0, from); 1952 1953 return fuse_dev_do_write(fud, &cs, iov_iter_count(from)); 1954 } 1955 1956 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe, 1957 struct file *out, loff_t *ppos, 1958 size_t len, unsigned int flags) 1959 { 1960 unsigned int head, tail, mask, count; 1961 unsigned nbuf; 1962 unsigned idx; 1963 struct pipe_buffer *bufs; 1964 struct fuse_copy_state cs; 1965 struct fuse_dev *fud; 1966 size_t rem; 1967 ssize_t ret; 1968 1969 fud = fuse_get_dev(out); 1970 if (!fud) 1971 return -EPERM; 1972 1973 pipe_lock(pipe); 1974 1975 head = pipe->head; 1976 tail = pipe->tail; 1977 mask = pipe->ring_size - 1; 1978 count = head - tail; 1979 1980 bufs = kvmalloc_array(count, sizeof(struct pipe_buffer), GFP_KERNEL); 1981 if (!bufs) { 1982 pipe_unlock(pipe); 1983 return -ENOMEM; 1984 } 1985 1986 nbuf = 0; 1987 rem = 0; 1988 for (idx = tail; idx != head && rem < len; idx++) 1989 rem += pipe->bufs[idx & mask].len; 1990 1991 ret = -EINVAL; 1992 if (rem < len) 1993 goto out_free; 1994 1995 rem = len; 1996 while (rem) { 1997 struct pipe_buffer *ibuf; 1998 struct pipe_buffer *obuf; 1999 2000 if (WARN_ON(nbuf >= count || tail == head)) 2001 goto out_free; 2002 2003 ibuf = &pipe->bufs[tail & mask]; 2004 obuf = &bufs[nbuf]; 2005 2006 if (rem >= ibuf->len) { 2007 *obuf = *ibuf; 2008 ibuf->ops = NULL; 2009 tail++; 2010 pipe->tail = tail; 2011 } else { 2012 if (!pipe_buf_get(pipe, ibuf)) 2013 goto out_free; 2014 2015 *obuf = *ibuf; 2016 obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 2017 obuf->len = rem; 2018 ibuf->offset += obuf->len; 2019 ibuf->len -= obuf->len; 2020 } 2021 nbuf++; 2022 rem -= obuf->len; 2023 } 2024 pipe_unlock(pipe); 2025 2026 fuse_copy_init(&cs, 0, NULL); 2027 cs.pipebufs = bufs; 2028 cs.nr_segs = nbuf; 2029 cs.pipe = pipe; 2030 2031 if (flags & SPLICE_F_MOVE) 2032 cs.move_pages = 1; 2033 2034 ret = fuse_dev_do_write(fud, &cs, len); 2035 2036 pipe_lock(pipe); 2037 out_free: 2038 for (idx = 0; idx < nbuf; idx++) 2039 pipe_buf_release(pipe, &bufs[idx]); 2040 pipe_unlock(pipe); 2041 2042 kvfree(bufs); 2043 return ret; 2044 } 2045 2046 static __poll_t fuse_dev_poll(struct file *file, poll_table *wait) 2047 { 2048 __poll_t mask = EPOLLOUT | EPOLLWRNORM; 2049 struct fuse_iqueue *fiq; 2050 struct fuse_dev *fud = fuse_get_dev(file); 2051 2052 if (!fud) 2053 return EPOLLERR; 2054 2055 fiq = &fud->fc->iq; 2056 poll_wait(file, &fiq->waitq, wait); 2057 2058 spin_lock(&fiq->lock); 2059 if (!fiq->connected) 2060 mask = EPOLLERR; 2061 else if (request_pending(fiq)) 2062 mask |= EPOLLIN | EPOLLRDNORM; 2063 spin_unlock(&fiq->lock); 2064 2065 return mask; 2066 } 2067 2068 /* Abort all requests on the given list (pending or processing) */ 2069 static void end_requests(struct list_head *head) 2070 { 2071 while (!list_empty(head)) { 2072 struct fuse_req *req; 2073 req = list_entry(head->next, struct fuse_req, list); 2074 req->out.h.error = -ECONNABORTED; 2075 clear_bit(FR_SENT, &req->flags); 2076 list_del_init(&req->list); 2077 fuse_request_end(req); 2078 } 2079 } 2080 2081 static void end_polls(struct fuse_conn *fc) 2082 { 2083 struct rb_node *p; 2084 2085 p = rb_first(&fc->polled_files); 2086 2087 while (p) { 2088 struct fuse_file *ff; 2089 ff = rb_entry(p, struct fuse_file, polled_node); 2090 wake_up_interruptible_all(&ff->poll_wait); 2091 2092 p = rb_next(p); 2093 } 2094 } 2095 2096 /* 2097 * Abort all requests. 2098 * 2099 * Emergency exit in case of a malicious or accidental deadlock, or just a hung 2100 * filesystem. 2101 * 2102 * The same effect is usually achievable through killing the filesystem daemon 2103 * and all users of the filesystem. The exception is the combination of an 2104 * asynchronous request and the tricky deadlock (see 2105 * Documentation/filesystems/fuse.rst). 2106 * 2107 * Aborting requests under I/O goes as follows: 1: Separate out unlocked 2108 * requests, they should be finished off immediately. Locked requests will be 2109 * finished after unlock; see unlock_request(). 2: Finish off the unlocked 2110 * requests. It is possible that some request will finish before we can. This 2111 * is OK, the request will in that case be removed from the list before we touch 2112 * it. 2113 */ 2114 void fuse_abort_conn(struct fuse_conn *fc) 2115 { 2116 struct fuse_iqueue *fiq = &fc->iq; 2117 2118 spin_lock(&fc->lock); 2119 if (fc->connected) { 2120 struct fuse_dev *fud; 2121 struct fuse_req *req, *next; 2122 LIST_HEAD(to_end); 2123 unsigned int i; 2124 2125 /* Background queuing checks fc->connected under bg_lock */ 2126 spin_lock(&fc->bg_lock); 2127 fc->connected = 0; 2128 spin_unlock(&fc->bg_lock); 2129 2130 fuse_set_initialized(fc); 2131 list_for_each_entry(fud, &fc->devices, entry) { 2132 struct fuse_pqueue *fpq = &fud->pq; 2133 2134 spin_lock(&fpq->lock); 2135 fpq->connected = 0; 2136 list_for_each_entry_safe(req, next, &fpq->io, list) { 2137 req->out.h.error = -ECONNABORTED; 2138 spin_lock(&req->waitq.lock); 2139 set_bit(FR_ABORTED, &req->flags); 2140 if (!test_bit(FR_LOCKED, &req->flags)) { 2141 set_bit(FR_PRIVATE, &req->flags); 2142 __fuse_get_request(req); 2143 list_move(&req->list, &to_end); 2144 } 2145 spin_unlock(&req->waitq.lock); 2146 } 2147 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 2148 list_splice_tail_init(&fpq->processing[i], 2149 &to_end); 2150 spin_unlock(&fpq->lock); 2151 } 2152 spin_lock(&fc->bg_lock); 2153 fc->blocked = 0; 2154 fc->max_background = UINT_MAX; 2155 flush_bg_queue(fc); 2156 spin_unlock(&fc->bg_lock); 2157 2158 spin_lock(&fiq->lock); 2159 fiq->connected = 0; 2160 list_for_each_entry(req, &fiq->pending, list) 2161 clear_bit(FR_PENDING, &req->flags); 2162 list_splice_tail_init(&fiq->pending, &to_end); 2163 while (forget_pending(fiq)) 2164 kfree(fuse_dequeue_forget(fiq, 1, NULL)); 2165 wake_up_all(&fiq->waitq); 2166 spin_unlock(&fiq->lock); 2167 kill_fasync(&fiq->fasync, SIGIO, POLL_IN); 2168 end_polls(fc); 2169 wake_up_all(&fc->blocked_waitq); 2170 spin_unlock(&fc->lock); 2171 2172 end_requests(&to_end); 2173 } else { 2174 spin_unlock(&fc->lock); 2175 } 2176 } 2177 EXPORT_SYMBOL_GPL(fuse_abort_conn); 2178 2179 void fuse_wait_aborted(struct fuse_conn *fc) 2180 { 2181 /* matches implicit memory barrier in fuse_drop_waiting() */ 2182 smp_mb(); 2183 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0); 2184 } 2185 2186 int fuse_dev_release(struct inode *inode, struct file *file) 2187 { 2188 struct fuse_dev *fud = fuse_get_dev(file); 2189 2190 if (fud) { 2191 struct fuse_conn *fc = fud->fc; 2192 struct fuse_pqueue *fpq = &fud->pq; 2193 LIST_HEAD(to_end); 2194 unsigned int i; 2195 2196 spin_lock(&fpq->lock); 2197 WARN_ON(!list_empty(&fpq->io)); 2198 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 2199 list_splice_init(&fpq->processing[i], &to_end); 2200 spin_unlock(&fpq->lock); 2201 2202 end_requests(&to_end); 2203 2204 /* Are we the last open device? */ 2205 if (atomic_dec_and_test(&fc->dev_count)) { 2206 WARN_ON(fc->iq.fasync != NULL); 2207 fuse_abort_conn(fc); 2208 } 2209 fuse_dev_free(fud); 2210 } 2211 return 0; 2212 } 2213 EXPORT_SYMBOL_GPL(fuse_dev_release); 2214 2215 static int fuse_dev_fasync(int fd, struct file *file, int on) 2216 { 2217 struct fuse_dev *fud = fuse_get_dev(file); 2218 2219 if (!fud) 2220 return -EPERM; 2221 2222 /* No locking - fasync_helper does its own locking */ 2223 return fasync_helper(fd, file, on, &fud->fc->iq.fasync); 2224 } 2225 2226 static int fuse_device_clone(struct fuse_conn *fc, struct file *new) 2227 { 2228 struct fuse_dev *fud; 2229 2230 if (new->private_data) 2231 return -EINVAL; 2232 2233 fud = fuse_dev_alloc_install(fc); 2234 if (!fud) 2235 return -ENOMEM; 2236 2237 new->private_data = fud; 2238 atomic_inc(&fc->dev_count); 2239 2240 return 0; 2241 } 2242 2243 static long fuse_dev_ioctl(struct file *file, unsigned int cmd, 2244 unsigned long arg) 2245 { 2246 int err = -ENOTTY; 2247 2248 if (cmd == FUSE_DEV_IOC_CLONE) { 2249 int oldfd; 2250 2251 err = -EFAULT; 2252 if (!get_user(oldfd, (__u32 __user *) arg)) { 2253 struct file *old = fget(oldfd); 2254 2255 err = -EINVAL; 2256 if (old) { 2257 struct fuse_dev *fud = NULL; 2258 2259 /* 2260 * Check against file->f_op because CUSE 2261 * uses the same ioctl handler. 2262 */ 2263 if (old->f_op == file->f_op && 2264 old->f_cred->user_ns == file->f_cred->user_ns) 2265 fud = fuse_get_dev(old); 2266 2267 if (fud) { 2268 mutex_lock(&fuse_mutex); 2269 err = fuse_device_clone(fud->fc, file); 2270 mutex_unlock(&fuse_mutex); 2271 } 2272 fput(old); 2273 } 2274 } 2275 } 2276 return err; 2277 } 2278 2279 const struct file_operations fuse_dev_operations = { 2280 .owner = THIS_MODULE, 2281 .open = fuse_dev_open, 2282 .llseek = no_llseek, 2283 .read_iter = fuse_dev_read, 2284 .splice_read = fuse_dev_splice_read, 2285 .write_iter = fuse_dev_write, 2286 .splice_write = fuse_dev_splice_write, 2287 .poll = fuse_dev_poll, 2288 .release = fuse_dev_release, 2289 .fasync = fuse_dev_fasync, 2290 .unlocked_ioctl = fuse_dev_ioctl, 2291 .compat_ioctl = compat_ptr_ioctl, 2292 }; 2293 EXPORT_SYMBOL_GPL(fuse_dev_operations); 2294 2295 static struct miscdevice fuse_miscdevice = { 2296 .minor = FUSE_MINOR, 2297 .name = "fuse", 2298 .fops = &fuse_dev_operations, 2299 }; 2300 2301 int __init fuse_dev_init(void) 2302 { 2303 int err = -ENOMEM; 2304 fuse_req_cachep = kmem_cache_create("fuse_request", 2305 sizeof(struct fuse_req), 2306 0, 0, NULL); 2307 if (!fuse_req_cachep) 2308 goto out; 2309 2310 err = misc_register(&fuse_miscdevice); 2311 if (err) 2312 goto out_cache_clean; 2313 2314 return 0; 2315 2316 out_cache_clean: 2317 kmem_cache_destroy(fuse_req_cachep); 2318 out: 2319 return err; 2320 } 2321 2322 void fuse_dev_cleanup(void) 2323 { 2324 misc_deregister(&fuse_miscdevice); 2325 kmem_cache_destroy(fuse_req_cachep); 2326 } 2327