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 matched 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 FR_INTERRUPTED check. Pairs with 292 * smp_mb() from queue_interrupt(). 293 */ 294 if (test_bit(FR_INTERRUPTED, &req->flags)) { 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 fc->num_background--; 319 fc->active_background--; 320 flush_bg_queue(fc); 321 spin_unlock(&fc->bg_lock); 322 } else { 323 /* Wake up waiter sleeping in request_wait_answer() */ 324 wake_up(&req->waitq); 325 } 326 327 if (test_bit(FR_ASYNC, &req->flags)) 328 req->args->end(fm, req->args, req->out.h.error); 329 put_request: 330 fuse_put_request(req); 331 } 332 EXPORT_SYMBOL_GPL(fuse_request_end); 333 334 static int queue_interrupt(struct fuse_req *req) 335 { 336 struct fuse_iqueue *fiq = &req->fm->fc->iq; 337 338 spin_lock(&fiq->lock); 339 /* Check for we've sent request to interrupt this req */ 340 if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags))) { 341 spin_unlock(&fiq->lock); 342 return -EINVAL; 343 } 344 345 if (list_empty(&req->intr_entry)) { 346 list_add_tail(&req->intr_entry, &fiq->interrupts); 347 /* 348 * Pairs with smp_mb() implied by test_and_set_bit() 349 * from fuse_request_end(). 350 */ 351 smp_mb(); 352 if (test_bit(FR_FINISHED, &req->flags)) { 353 list_del_init(&req->intr_entry); 354 spin_unlock(&fiq->lock); 355 return 0; 356 } 357 fiq->ops->wake_interrupt_and_unlock(fiq); 358 } else { 359 spin_unlock(&fiq->lock); 360 } 361 return 0; 362 } 363 364 static void request_wait_answer(struct fuse_req *req) 365 { 366 struct fuse_conn *fc = req->fm->fc; 367 struct fuse_iqueue *fiq = &fc->iq; 368 int err; 369 370 if (!fc->no_interrupt) { 371 /* Any signal may interrupt this */ 372 err = wait_event_interruptible(req->waitq, 373 test_bit(FR_FINISHED, &req->flags)); 374 if (!err) 375 return; 376 377 set_bit(FR_INTERRUPTED, &req->flags); 378 /* matches barrier in fuse_dev_do_read() */ 379 smp_mb__after_atomic(); 380 if (test_bit(FR_SENT, &req->flags)) 381 queue_interrupt(req); 382 } 383 384 if (!test_bit(FR_FORCE, &req->flags)) { 385 /* Only fatal signals may interrupt this */ 386 err = wait_event_killable(req->waitq, 387 test_bit(FR_FINISHED, &req->flags)); 388 if (!err) 389 return; 390 391 spin_lock(&fiq->lock); 392 /* Request is not yet in userspace, bail out */ 393 if (test_bit(FR_PENDING, &req->flags)) { 394 list_del(&req->list); 395 spin_unlock(&fiq->lock); 396 __fuse_put_request(req); 397 req->out.h.error = -EINTR; 398 return; 399 } 400 spin_unlock(&fiq->lock); 401 } 402 403 /* 404 * Either request is already in userspace, or it was forced. 405 * Wait it out. 406 */ 407 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags)); 408 } 409 410 static void __fuse_request_send(struct fuse_req *req) 411 { 412 struct fuse_iqueue *fiq = &req->fm->fc->iq; 413 414 BUG_ON(test_bit(FR_BACKGROUND, &req->flags)); 415 spin_lock(&fiq->lock); 416 if (!fiq->connected) { 417 spin_unlock(&fiq->lock); 418 req->out.h.error = -ENOTCONN; 419 } else { 420 req->in.h.unique = fuse_get_unique(fiq); 421 /* acquire extra reference, since request is still needed 422 after fuse_request_end() */ 423 __fuse_get_request(req); 424 queue_request_and_unlock(fiq, req); 425 426 request_wait_answer(req); 427 /* Pairs with smp_wmb() in fuse_request_end() */ 428 smp_rmb(); 429 } 430 } 431 432 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args) 433 { 434 if (fc->minor < 4 && args->opcode == FUSE_STATFS) 435 args->out_args[0].size = FUSE_COMPAT_STATFS_SIZE; 436 437 if (fc->minor < 9) { 438 switch (args->opcode) { 439 case FUSE_LOOKUP: 440 case FUSE_CREATE: 441 case FUSE_MKNOD: 442 case FUSE_MKDIR: 443 case FUSE_SYMLINK: 444 case FUSE_LINK: 445 args->out_args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE; 446 break; 447 case FUSE_GETATTR: 448 case FUSE_SETATTR: 449 args->out_args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE; 450 break; 451 } 452 } 453 if (fc->minor < 12) { 454 switch (args->opcode) { 455 case FUSE_CREATE: 456 args->in_args[0].size = sizeof(struct fuse_open_in); 457 break; 458 case FUSE_MKNOD: 459 args->in_args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE; 460 break; 461 } 462 } 463 } 464 465 static void fuse_force_creds(struct fuse_req *req) 466 { 467 struct fuse_conn *fc = req->fm->fc; 468 469 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid()); 470 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid()); 471 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns); 472 } 473 474 static void fuse_args_to_req(struct fuse_req *req, struct fuse_args *args) 475 { 476 req->in.h.opcode = args->opcode; 477 req->in.h.nodeid = args->nodeid; 478 req->args = args; 479 if (args->end) 480 __set_bit(FR_ASYNC, &req->flags); 481 } 482 483 ssize_t fuse_simple_request(struct fuse_mount *fm, struct fuse_args *args) 484 { 485 struct fuse_conn *fc = fm->fc; 486 struct fuse_req *req; 487 ssize_t ret; 488 489 if (args->force) { 490 atomic_inc(&fc->num_waiting); 491 req = fuse_request_alloc(fm, GFP_KERNEL | __GFP_NOFAIL); 492 493 if (!args->nocreds) 494 fuse_force_creds(req); 495 496 __set_bit(FR_WAITING, &req->flags); 497 __set_bit(FR_FORCE, &req->flags); 498 } else { 499 WARN_ON(args->nocreds); 500 req = fuse_get_req(fm, false); 501 if (IS_ERR(req)) 502 return PTR_ERR(req); 503 } 504 505 /* Needs to be done after fuse_get_req() so that fc->minor is valid */ 506 fuse_adjust_compat(fc, args); 507 fuse_args_to_req(req, args); 508 509 if (!args->noreply) 510 __set_bit(FR_ISREPLY, &req->flags); 511 __fuse_request_send(req); 512 ret = req->out.h.error; 513 if (!ret && args->out_argvar) { 514 BUG_ON(args->out_numargs == 0); 515 ret = args->out_args[args->out_numargs - 1].size; 516 } 517 fuse_put_request(req); 518 519 return ret; 520 } 521 522 static bool fuse_request_queue_background(struct fuse_req *req) 523 { 524 struct fuse_mount *fm = req->fm; 525 struct fuse_conn *fc = fm->fc; 526 bool queued = false; 527 528 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags)); 529 if (!test_bit(FR_WAITING, &req->flags)) { 530 __set_bit(FR_WAITING, &req->flags); 531 atomic_inc(&fc->num_waiting); 532 } 533 __set_bit(FR_ISREPLY, &req->flags); 534 spin_lock(&fc->bg_lock); 535 if (likely(fc->connected)) { 536 fc->num_background++; 537 if (fc->num_background == fc->max_background) 538 fc->blocked = 1; 539 list_add_tail(&req->list, &fc->bg_queue); 540 flush_bg_queue(fc); 541 queued = true; 542 } 543 spin_unlock(&fc->bg_lock); 544 545 return queued; 546 } 547 548 int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args, 549 gfp_t gfp_flags) 550 { 551 struct fuse_req *req; 552 553 if (args->force) { 554 WARN_ON(!args->nocreds); 555 req = fuse_request_alloc(fm, gfp_flags); 556 if (!req) 557 return -ENOMEM; 558 __set_bit(FR_BACKGROUND, &req->flags); 559 } else { 560 WARN_ON(args->nocreds); 561 req = fuse_get_req(fm, true); 562 if (IS_ERR(req)) 563 return PTR_ERR(req); 564 } 565 566 fuse_args_to_req(req, args); 567 568 if (!fuse_request_queue_background(req)) { 569 fuse_put_request(req); 570 return -ENOTCONN; 571 } 572 573 return 0; 574 } 575 EXPORT_SYMBOL_GPL(fuse_simple_background); 576 577 static int fuse_simple_notify_reply(struct fuse_mount *fm, 578 struct fuse_args *args, u64 unique) 579 { 580 struct fuse_req *req; 581 struct fuse_iqueue *fiq = &fm->fc->iq; 582 int err = 0; 583 584 req = fuse_get_req(fm, false); 585 if (IS_ERR(req)) 586 return PTR_ERR(req); 587 588 __clear_bit(FR_ISREPLY, &req->flags); 589 req->in.h.unique = unique; 590 591 fuse_args_to_req(req, args); 592 593 spin_lock(&fiq->lock); 594 if (fiq->connected) { 595 queue_request_and_unlock(fiq, req); 596 } else { 597 err = -ENODEV; 598 spin_unlock(&fiq->lock); 599 fuse_put_request(req); 600 } 601 602 return err; 603 } 604 605 /* 606 * Lock the request. Up to the next unlock_request() there mustn't be 607 * anything that could cause a page-fault. If the request was already 608 * aborted bail out. 609 */ 610 static int lock_request(struct fuse_req *req) 611 { 612 int err = 0; 613 if (req) { 614 spin_lock(&req->waitq.lock); 615 if (test_bit(FR_ABORTED, &req->flags)) 616 err = -ENOENT; 617 else 618 set_bit(FR_LOCKED, &req->flags); 619 spin_unlock(&req->waitq.lock); 620 } 621 return err; 622 } 623 624 /* 625 * Unlock request. If it was aborted while locked, caller is responsible 626 * for unlocking and ending the request. 627 */ 628 static int unlock_request(struct fuse_req *req) 629 { 630 int err = 0; 631 if (req) { 632 spin_lock(&req->waitq.lock); 633 if (test_bit(FR_ABORTED, &req->flags)) 634 err = -ENOENT; 635 else 636 clear_bit(FR_LOCKED, &req->flags); 637 spin_unlock(&req->waitq.lock); 638 } 639 return err; 640 } 641 642 struct fuse_copy_state { 643 int write; 644 struct fuse_req *req; 645 struct iov_iter *iter; 646 struct pipe_buffer *pipebufs; 647 struct pipe_buffer *currbuf; 648 struct pipe_inode_info *pipe; 649 unsigned long nr_segs; 650 struct page *pg; 651 unsigned len; 652 unsigned offset; 653 unsigned move_pages:1; 654 }; 655 656 static void fuse_copy_init(struct fuse_copy_state *cs, int write, 657 struct iov_iter *iter) 658 { 659 memset(cs, 0, sizeof(*cs)); 660 cs->write = write; 661 cs->iter = iter; 662 } 663 664 /* Unmap and put previous page of userspace buffer */ 665 static void fuse_copy_finish(struct fuse_copy_state *cs) 666 { 667 if (cs->currbuf) { 668 struct pipe_buffer *buf = cs->currbuf; 669 670 if (cs->write) 671 buf->len = PAGE_SIZE - cs->len; 672 cs->currbuf = NULL; 673 } else if (cs->pg) { 674 if (cs->write) { 675 flush_dcache_page(cs->pg); 676 set_page_dirty_lock(cs->pg); 677 } 678 put_page(cs->pg); 679 } 680 cs->pg = NULL; 681 } 682 683 /* 684 * Get another pagefull of userspace buffer, and map it to kernel 685 * address space, and lock request 686 */ 687 static int fuse_copy_fill(struct fuse_copy_state *cs) 688 { 689 struct page *page; 690 int err; 691 692 err = unlock_request(cs->req); 693 if (err) 694 return err; 695 696 fuse_copy_finish(cs); 697 if (cs->pipebufs) { 698 struct pipe_buffer *buf = cs->pipebufs; 699 700 if (!cs->write) { 701 err = pipe_buf_confirm(cs->pipe, buf); 702 if (err) 703 return err; 704 705 BUG_ON(!cs->nr_segs); 706 cs->currbuf = buf; 707 cs->pg = buf->page; 708 cs->offset = buf->offset; 709 cs->len = buf->len; 710 cs->pipebufs++; 711 cs->nr_segs--; 712 } else { 713 if (cs->nr_segs >= cs->pipe->max_usage) 714 return -EIO; 715 716 page = alloc_page(GFP_HIGHUSER); 717 if (!page) 718 return -ENOMEM; 719 720 buf->page = page; 721 buf->offset = 0; 722 buf->len = 0; 723 724 cs->currbuf = buf; 725 cs->pg = page; 726 cs->offset = 0; 727 cs->len = PAGE_SIZE; 728 cs->pipebufs++; 729 cs->nr_segs++; 730 } 731 } else { 732 size_t off; 733 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off); 734 if (err < 0) 735 return err; 736 BUG_ON(!err); 737 cs->len = err; 738 cs->offset = off; 739 cs->pg = page; 740 iov_iter_advance(cs->iter, err); 741 } 742 743 return lock_request(cs->req); 744 } 745 746 /* Do as much copy to/from userspace buffer as we can */ 747 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) 748 { 749 unsigned ncpy = min(*size, cs->len); 750 if (val) { 751 void *pgaddr = kmap_local_page(cs->pg); 752 void *buf = pgaddr + cs->offset; 753 754 if (cs->write) 755 memcpy(buf, *val, ncpy); 756 else 757 memcpy(*val, buf, ncpy); 758 759 kunmap_local(pgaddr); 760 *val += ncpy; 761 } 762 *size -= ncpy; 763 cs->len -= ncpy; 764 cs->offset += ncpy; 765 return ncpy; 766 } 767 768 static int fuse_check_page(struct page *page) 769 { 770 if (page_mapcount(page) || 771 page->mapping != NULL || 772 (page->flags & PAGE_FLAGS_CHECK_AT_PREP & 773 ~(1 << PG_locked | 774 1 << PG_referenced | 775 1 << PG_uptodate | 776 1 << PG_lru | 777 1 << PG_active | 778 1 << PG_workingset | 779 1 << PG_reclaim | 780 1 << PG_waiters))) { 781 dump_page(page, "fuse: trying to steal weird page"); 782 return 1; 783 } 784 return 0; 785 } 786 787 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep) 788 { 789 int err; 790 struct page *oldpage = *pagep; 791 struct page *newpage; 792 struct pipe_buffer *buf = cs->pipebufs; 793 794 get_page(oldpage); 795 err = unlock_request(cs->req); 796 if (err) 797 goto out_put_old; 798 799 fuse_copy_finish(cs); 800 801 err = pipe_buf_confirm(cs->pipe, buf); 802 if (err) 803 goto out_put_old; 804 805 BUG_ON(!cs->nr_segs); 806 cs->currbuf = buf; 807 cs->len = buf->len; 808 cs->pipebufs++; 809 cs->nr_segs--; 810 811 if (cs->len != PAGE_SIZE) 812 goto out_fallback; 813 814 if (!pipe_buf_try_steal(cs->pipe, buf)) 815 goto out_fallback; 816 817 newpage = buf->page; 818 819 if (!PageUptodate(newpage)) 820 SetPageUptodate(newpage); 821 822 ClearPageMappedToDisk(newpage); 823 824 if (fuse_check_page(newpage) != 0) 825 goto out_fallback_unlock; 826 827 /* 828 * This is a new and locked page, it shouldn't be mapped or 829 * have any special flags on it 830 */ 831 if (WARN_ON(page_mapped(oldpage))) 832 goto out_fallback_unlock; 833 if (WARN_ON(page_has_private(oldpage))) 834 goto out_fallback_unlock; 835 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage))) 836 goto out_fallback_unlock; 837 if (WARN_ON(PageMlocked(oldpage))) 838 goto out_fallback_unlock; 839 840 replace_page_cache_page(oldpage, newpage); 841 842 get_page(newpage); 843 844 if (!(buf->flags & PIPE_BUF_FLAG_LRU)) 845 lru_cache_add(newpage); 846 847 /* 848 * Release while we have extra ref on stolen page. Otherwise 849 * anon_pipe_buf_release() might think the page can be reused. 850 */ 851 pipe_buf_release(cs->pipe, buf); 852 853 err = 0; 854 spin_lock(&cs->req->waitq.lock); 855 if (test_bit(FR_ABORTED, &cs->req->flags)) 856 err = -ENOENT; 857 else 858 *pagep = newpage; 859 spin_unlock(&cs->req->waitq.lock); 860 861 if (err) { 862 unlock_page(newpage); 863 put_page(newpage); 864 goto out_put_old; 865 } 866 867 unlock_page(oldpage); 868 /* Drop ref for ap->pages[] array */ 869 put_page(oldpage); 870 cs->len = 0; 871 872 err = 0; 873 out_put_old: 874 /* Drop ref obtained in this function */ 875 put_page(oldpage); 876 return err; 877 878 out_fallback_unlock: 879 unlock_page(newpage); 880 out_fallback: 881 cs->pg = buf->page; 882 cs->offset = buf->offset; 883 884 err = lock_request(cs->req); 885 if (!err) 886 err = 1; 887 888 goto out_put_old; 889 } 890 891 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page, 892 unsigned offset, unsigned count) 893 { 894 struct pipe_buffer *buf; 895 int err; 896 897 if (cs->nr_segs >= cs->pipe->max_usage) 898 return -EIO; 899 900 get_page(page); 901 err = unlock_request(cs->req); 902 if (err) { 903 put_page(page); 904 return err; 905 } 906 907 fuse_copy_finish(cs); 908 909 buf = cs->pipebufs; 910 buf->page = page; 911 buf->offset = offset; 912 buf->len = count; 913 914 cs->pipebufs++; 915 cs->nr_segs++; 916 cs->len = 0; 917 918 return 0; 919 } 920 921 /* 922 * Copy a page in the request to/from the userspace buffer. Must be 923 * done atomically 924 */ 925 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep, 926 unsigned offset, unsigned count, int zeroing) 927 { 928 int err; 929 struct page *page = *pagep; 930 931 if (page && zeroing && count < PAGE_SIZE) 932 clear_highpage(page); 933 934 while (count) { 935 if (cs->write && cs->pipebufs && page) { 936 /* 937 * Can't control lifetime of pipe buffers, so always 938 * copy user pages. 939 */ 940 if (cs->req->args->user_pages) { 941 err = fuse_copy_fill(cs); 942 if (err) 943 return err; 944 } else { 945 return fuse_ref_page(cs, page, offset, count); 946 } 947 } else if (!cs->len) { 948 if (cs->move_pages && page && 949 offset == 0 && count == PAGE_SIZE) { 950 err = fuse_try_move_page(cs, pagep); 951 if (err <= 0) 952 return err; 953 } else { 954 err = fuse_copy_fill(cs); 955 if (err) 956 return err; 957 } 958 } 959 if (page) { 960 void *mapaddr = kmap_local_page(page); 961 void *buf = mapaddr + offset; 962 offset += fuse_copy_do(cs, &buf, &count); 963 kunmap_local(mapaddr); 964 } else 965 offset += fuse_copy_do(cs, NULL, &count); 966 } 967 if (page && !cs->write) 968 flush_dcache_page(page); 969 return 0; 970 } 971 972 /* Copy pages in the request to/from userspace buffer */ 973 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, 974 int zeroing) 975 { 976 unsigned i; 977 struct fuse_req *req = cs->req; 978 struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args); 979 980 981 for (i = 0; i < ap->num_pages && (nbytes || zeroing); i++) { 982 int err; 983 unsigned int offset = ap->descs[i].offset; 984 unsigned int count = min(nbytes, ap->descs[i].length); 985 986 err = fuse_copy_page(cs, &ap->pages[i], offset, count, zeroing); 987 if (err) 988 return err; 989 990 nbytes -= count; 991 } 992 return 0; 993 } 994 995 /* Copy a single argument in the request to/from userspace buffer */ 996 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) 997 { 998 while (size) { 999 if (!cs->len) { 1000 int err = fuse_copy_fill(cs); 1001 if (err) 1002 return err; 1003 } 1004 fuse_copy_do(cs, &val, &size); 1005 } 1006 return 0; 1007 } 1008 1009 /* Copy request arguments to/from userspace buffer */ 1010 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, 1011 unsigned argpages, struct fuse_arg *args, 1012 int zeroing) 1013 { 1014 int err = 0; 1015 unsigned i; 1016 1017 for (i = 0; !err && i < numargs; i++) { 1018 struct fuse_arg *arg = &args[i]; 1019 if (i == numargs - 1 && argpages) 1020 err = fuse_copy_pages(cs, arg->size, zeroing); 1021 else 1022 err = fuse_copy_one(cs, arg->value, arg->size); 1023 } 1024 return err; 1025 } 1026 1027 static int forget_pending(struct fuse_iqueue *fiq) 1028 { 1029 return fiq->forget_list_head.next != NULL; 1030 } 1031 1032 static int request_pending(struct fuse_iqueue *fiq) 1033 { 1034 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) || 1035 forget_pending(fiq); 1036 } 1037 1038 /* 1039 * Transfer an interrupt request to userspace 1040 * 1041 * Unlike other requests this is assembled on demand, without a need 1042 * to allocate a separate fuse_req structure. 1043 * 1044 * Called with fiq->lock held, releases it 1045 */ 1046 static int fuse_read_interrupt(struct fuse_iqueue *fiq, 1047 struct fuse_copy_state *cs, 1048 size_t nbytes, struct fuse_req *req) 1049 __releases(fiq->lock) 1050 { 1051 struct fuse_in_header ih; 1052 struct fuse_interrupt_in arg; 1053 unsigned reqsize = sizeof(ih) + sizeof(arg); 1054 int err; 1055 1056 list_del_init(&req->intr_entry); 1057 memset(&ih, 0, sizeof(ih)); 1058 memset(&arg, 0, sizeof(arg)); 1059 ih.len = reqsize; 1060 ih.opcode = FUSE_INTERRUPT; 1061 ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT); 1062 arg.unique = req->in.h.unique; 1063 1064 spin_unlock(&fiq->lock); 1065 if (nbytes < reqsize) 1066 return -EINVAL; 1067 1068 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1069 if (!err) 1070 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1071 fuse_copy_finish(cs); 1072 1073 return err ? err : reqsize; 1074 } 1075 1076 struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq, 1077 unsigned int max, 1078 unsigned int *countp) 1079 { 1080 struct fuse_forget_link *head = fiq->forget_list_head.next; 1081 struct fuse_forget_link **newhead = &head; 1082 unsigned count; 1083 1084 for (count = 0; *newhead != NULL && count < max; count++) 1085 newhead = &(*newhead)->next; 1086 1087 fiq->forget_list_head.next = *newhead; 1088 *newhead = NULL; 1089 if (fiq->forget_list_head.next == NULL) 1090 fiq->forget_list_tail = &fiq->forget_list_head; 1091 1092 if (countp != NULL) 1093 *countp = count; 1094 1095 return head; 1096 } 1097 EXPORT_SYMBOL(fuse_dequeue_forget); 1098 1099 static int fuse_read_single_forget(struct fuse_iqueue *fiq, 1100 struct fuse_copy_state *cs, 1101 size_t nbytes) 1102 __releases(fiq->lock) 1103 { 1104 int err; 1105 struct fuse_forget_link *forget = fuse_dequeue_forget(fiq, 1, NULL); 1106 struct fuse_forget_in arg = { 1107 .nlookup = forget->forget_one.nlookup, 1108 }; 1109 struct fuse_in_header ih = { 1110 .opcode = FUSE_FORGET, 1111 .nodeid = forget->forget_one.nodeid, 1112 .unique = fuse_get_unique(fiq), 1113 .len = sizeof(ih) + sizeof(arg), 1114 }; 1115 1116 spin_unlock(&fiq->lock); 1117 kfree(forget); 1118 if (nbytes < ih.len) 1119 return -EINVAL; 1120 1121 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1122 if (!err) 1123 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1124 fuse_copy_finish(cs); 1125 1126 if (err) 1127 return err; 1128 1129 return ih.len; 1130 } 1131 1132 static int fuse_read_batch_forget(struct fuse_iqueue *fiq, 1133 struct fuse_copy_state *cs, size_t nbytes) 1134 __releases(fiq->lock) 1135 { 1136 int err; 1137 unsigned max_forgets; 1138 unsigned count; 1139 struct fuse_forget_link *head; 1140 struct fuse_batch_forget_in arg = { .count = 0 }; 1141 struct fuse_in_header ih = { 1142 .opcode = FUSE_BATCH_FORGET, 1143 .unique = fuse_get_unique(fiq), 1144 .len = sizeof(ih) + sizeof(arg), 1145 }; 1146 1147 if (nbytes < ih.len) { 1148 spin_unlock(&fiq->lock); 1149 return -EINVAL; 1150 } 1151 1152 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one); 1153 head = fuse_dequeue_forget(fiq, max_forgets, &count); 1154 spin_unlock(&fiq->lock); 1155 1156 arg.count = count; 1157 ih.len += count * sizeof(struct fuse_forget_one); 1158 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1159 if (!err) 1160 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1161 1162 while (head) { 1163 struct fuse_forget_link *forget = head; 1164 1165 if (!err) { 1166 err = fuse_copy_one(cs, &forget->forget_one, 1167 sizeof(forget->forget_one)); 1168 } 1169 head = forget->next; 1170 kfree(forget); 1171 } 1172 1173 fuse_copy_finish(cs); 1174 1175 if (err) 1176 return err; 1177 1178 return ih.len; 1179 } 1180 1181 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq, 1182 struct fuse_copy_state *cs, 1183 size_t nbytes) 1184 __releases(fiq->lock) 1185 { 1186 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL) 1187 return fuse_read_single_forget(fiq, cs, nbytes); 1188 else 1189 return fuse_read_batch_forget(fiq, cs, nbytes); 1190 } 1191 1192 /* 1193 * Read a single request into the userspace filesystem's buffer. This 1194 * function waits until a request is available, then removes it from 1195 * the pending list and copies request data to userspace buffer. If 1196 * no reply is needed (FORGET) or request has been aborted or there 1197 * was an error during the copying then it's finished by calling 1198 * fuse_request_end(). Otherwise add it to the processing list, and set 1199 * the 'sent' flag. 1200 */ 1201 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file, 1202 struct fuse_copy_state *cs, size_t nbytes) 1203 { 1204 ssize_t err; 1205 struct fuse_conn *fc = fud->fc; 1206 struct fuse_iqueue *fiq = &fc->iq; 1207 struct fuse_pqueue *fpq = &fud->pq; 1208 struct fuse_req *req; 1209 struct fuse_args *args; 1210 unsigned reqsize; 1211 unsigned int hash; 1212 1213 /* 1214 * Require sane minimum read buffer - that has capacity for fixed part 1215 * of any request header + negotiated max_write room for data. 1216 * 1217 * Historically libfuse reserves 4K for fixed header room, but e.g. 1218 * GlusterFS reserves only 80 bytes 1219 * 1220 * = `sizeof(fuse_in_header) + sizeof(fuse_write_in)` 1221 * 1222 * which is the absolute minimum any sane filesystem should be using 1223 * for header room. 1224 */ 1225 if (nbytes < max_t(size_t, FUSE_MIN_READ_BUFFER, 1226 sizeof(struct fuse_in_header) + 1227 sizeof(struct fuse_write_in) + 1228 fc->max_write)) 1229 return -EINVAL; 1230 1231 restart: 1232 for (;;) { 1233 spin_lock(&fiq->lock); 1234 if (!fiq->connected || request_pending(fiq)) 1235 break; 1236 spin_unlock(&fiq->lock); 1237 1238 if (file->f_flags & O_NONBLOCK) 1239 return -EAGAIN; 1240 err = wait_event_interruptible_exclusive(fiq->waitq, 1241 !fiq->connected || request_pending(fiq)); 1242 if (err) 1243 return err; 1244 } 1245 1246 if (!fiq->connected) { 1247 err = fc->aborted ? -ECONNABORTED : -ENODEV; 1248 goto err_unlock; 1249 } 1250 1251 if (!list_empty(&fiq->interrupts)) { 1252 req = list_entry(fiq->interrupts.next, struct fuse_req, 1253 intr_entry); 1254 return fuse_read_interrupt(fiq, cs, nbytes, req); 1255 } 1256 1257 if (forget_pending(fiq)) { 1258 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0) 1259 return fuse_read_forget(fc, fiq, cs, nbytes); 1260 1261 if (fiq->forget_batch <= -8) 1262 fiq->forget_batch = 16; 1263 } 1264 1265 req = list_entry(fiq->pending.next, struct fuse_req, list); 1266 clear_bit(FR_PENDING, &req->flags); 1267 list_del_init(&req->list); 1268 spin_unlock(&fiq->lock); 1269 1270 args = req->args; 1271 reqsize = req->in.h.len; 1272 1273 /* If request is too large, reply with an error and restart the read */ 1274 if (nbytes < reqsize) { 1275 req->out.h.error = -EIO; 1276 /* SETXATTR is special, since it may contain too large data */ 1277 if (args->opcode == FUSE_SETXATTR) 1278 req->out.h.error = -E2BIG; 1279 fuse_request_end(req); 1280 goto restart; 1281 } 1282 spin_lock(&fpq->lock); 1283 /* 1284 * Must not put request on fpq->io queue after having been shut down by 1285 * fuse_abort_conn() 1286 */ 1287 if (!fpq->connected) { 1288 req->out.h.error = err = -ECONNABORTED; 1289 goto out_end; 1290 1291 } 1292 list_add(&req->list, &fpq->io); 1293 spin_unlock(&fpq->lock); 1294 cs->req = req; 1295 err = fuse_copy_one(cs, &req->in.h, sizeof(req->in.h)); 1296 if (!err) 1297 err = fuse_copy_args(cs, args->in_numargs, args->in_pages, 1298 (struct fuse_arg *) args->in_args, 0); 1299 fuse_copy_finish(cs); 1300 spin_lock(&fpq->lock); 1301 clear_bit(FR_LOCKED, &req->flags); 1302 if (!fpq->connected) { 1303 err = fc->aborted ? -ECONNABORTED : -ENODEV; 1304 goto out_end; 1305 } 1306 if (err) { 1307 req->out.h.error = -EIO; 1308 goto out_end; 1309 } 1310 if (!test_bit(FR_ISREPLY, &req->flags)) { 1311 err = reqsize; 1312 goto out_end; 1313 } 1314 hash = fuse_req_hash(req->in.h.unique); 1315 list_move_tail(&req->list, &fpq->processing[hash]); 1316 __fuse_get_request(req); 1317 set_bit(FR_SENT, &req->flags); 1318 spin_unlock(&fpq->lock); 1319 /* matches barrier in request_wait_answer() */ 1320 smp_mb__after_atomic(); 1321 if (test_bit(FR_INTERRUPTED, &req->flags)) 1322 queue_interrupt(req); 1323 fuse_put_request(req); 1324 1325 return reqsize; 1326 1327 out_end: 1328 if (!test_bit(FR_PRIVATE, &req->flags)) 1329 list_del_init(&req->list); 1330 spin_unlock(&fpq->lock); 1331 fuse_request_end(req); 1332 return err; 1333 1334 err_unlock: 1335 spin_unlock(&fiq->lock); 1336 return err; 1337 } 1338 1339 static int fuse_dev_open(struct inode *inode, struct file *file) 1340 { 1341 /* 1342 * The fuse device's file's private_data is used to hold 1343 * the fuse_conn(ection) when it is mounted, and is used to 1344 * keep track of whether the file has been mounted already. 1345 */ 1346 file->private_data = NULL; 1347 return 0; 1348 } 1349 1350 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to) 1351 { 1352 struct fuse_copy_state cs; 1353 struct file *file = iocb->ki_filp; 1354 struct fuse_dev *fud = fuse_get_dev(file); 1355 1356 if (!fud) 1357 return -EPERM; 1358 1359 if (!iter_is_iovec(to)) 1360 return -EINVAL; 1361 1362 fuse_copy_init(&cs, 1, to); 1363 1364 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to)); 1365 } 1366 1367 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos, 1368 struct pipe_inode_info *pipe, 1369 size_t len, unsigned int flags) 1370 { 1371 int total, ret; 1372 int page_nr = 0; 1373 struct pipe_buffer *bufs; 1374 struct fuse_copy_state cs; 1375 struct fuse_dev *fud = fuse_get_dev(in); 1376 1377 if (!fud) 1378 return -EPERM; 1379 1380 bufs = kvmalloc_array(pipe->max_usage, sizeof(struct pipe_buffer), 1381 GFP_KERNEL); 1382 if (!bufs) 1383 return -ENOMEM; 1384 1385 fuse_copy_init(&cs, 1, NULL); 1386 cs.pipebufs = bufs; 1387 cs.pipe = pipe; 1388 ret = fuse_dev_do_read(fud, in, &cs, len); 1389 if (ret < 0) 1390 goto out; 1391 1392 if (pipe_occupancy(pipe->head, pipe->tail) + cs.nr_segs > pipe->max_usage) { 1393 ret = -EIO; 1394 goto out; 1395 } 1396 1397 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) { 1398 /* 1399 * Need to be careful about this. Having buf->ops in module 1400 * code can Oops if the buffer persists after module unload. 1401 */ 1402 bufs[page_nr].ops = &nosteal_pipe_buf_ops; 1403 bufs[page_nr].flags = 0; 1404 ret = add_to_pipe(pipe, &bufs[page_nr++]); 1405 if (unlikely(ret < 0)) 1406 break; 1407 } 1408 if (total) 1409 ret = total; 1410 out: 1411 for (; page_nr < cs.nr_segs; page_nr++) 1412 put_page(bufs[page_nr].page); 1413 1414 kvfree(bufs); 1415 return ret; 1416 } 1417 1418 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size, 1419 struct fuse_copy_state *cs) 1420 { 1421 struct fuse_notify_poll_wakeup_out outarg; 1422 int err = -EINVAL; 1423 1424 if (size != sizeof(outarg)) 1425 goto err; 1426 1427 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1428 if (err) 1429 goto err; 1430 1431 fuse_copy_finish(cs); 1432 return fuse_notify_poll_wakeup(fc, &outarg); 1433 1434 err: 1435 fuse_copy_finish(cs); 1436 return err; 1437 } 1438 1439 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size, 1440 struct fuse_copy_state *cs) 1441 { 1442 struct fuse_notify_inval_inode_out outarg; 1443 int err = -EINVAL; 1444 1445 if (size != sizeof(outarg)) 1446 goto err; 1447 1448 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1449 if (err) 1450 goto err; 1451 fuse_copy_finish(cs); 1452 1453 down_read(&fc->killsb); 1454 err = fuse_reverse_inval_inode(fc, outarg.ino, 1455 outarg.off, outarg.len); 1456 up_read(&fc->killsb); 1457 return err; 1458 1459 err: 1460 fuse_copy_finish(cs); 1461 return err; 1462 } 1463 1464 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size, 1465 struct fuse_copy_state *cs) 1466 { 1467 struct fuse_notify_inval_entry_out outarg; 1468 int err = -ENOMEM; 1469 char *buf; 1470 struct qstr name; 1471 1472 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); 1473 if (!buf) 1474 goto err; 1475 1476 err = -EINVAL; 1477 if (size < sizeof(outarg)) 1478 goto err; 1479 1480 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1481 if (err) 1482 goto err; 1483 1484 err = -ENAMETOOLONG; 1485 if (outarg.namelen > FUSE_NAME_MAX) 1486 goto err; 1487 1488 err = -EINVAL; 1489 if (size != sizeof(outarg) + outarg.namelen + 1) 1490 goto err; 1491 1492 name.name = buf; 1493 name.len = outarg.namelen; 1494 err = fuse_copy_one(cs, buf, outarg.namelen + 1); 1495 if (err) 1496 goto err; 1497 fuse_copy_finish(cs); 1498 buf[outarg.namelen] = 0; 1499 1500 down_read(&fc->killsb); 1501 err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name); 1502 up_read(&fc->killsb); 1503 kfree(buf); 1504 return err; 1505 1506 err: 1507 kfree(buf); 1508 fuse_copy_finish(cs); 1509 return err; 1510 } 1511 1512 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size, 1513 struct fuse_copy_state *cs) 1514 { 1515 struct fuse_notify_delete_out outarg; 1516 int err = -ENOMEM; 1517 char *buf; 1518 struct qstr name; 1519 1520 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); 1521 if (!buf) 1522 goto err; 1523 1524 err = -EINVAL; 1525 if (size < sizeof(outarg)) 1526 goto err; 1527 1528 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1529 if (err) 1530 goto err; 1531 1532 err = -ENAMETOOLONG; 1533 if (outarg.namelen > FUSE_NAME_MAX) 1534 goto err; 1535 1536 err = -EINVAL; 1537 if (size != sizeof(outarg) + outarg.namelen + 1) 1538 goto err; 1539 1540 name.name = buf; 1541 name.len = outarg.namelen; 1542 err = fuse_copy_one(cs, buf, outarg.namelen + 1); 1543 if (err) 1544 goto err; 1545 fuse_copy_finish(cs); 1546 buf[outarg.namelen] = 0; 1547 1548 down_read(&fc->killsb); 1549 err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name); 1550 up_read(&fc->killsb); 1551 kfree(buf); 1552 return err; 1553 1554 err: 1555 kfree(buf); 1556 fuse_copy_finish(cs); 1557 return err; 1558 } 1559 1560 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size, 1561 struct fuse_copy_state *cs) 1562 { 1563 struct fuse_notify_store_out outarg; 1564 struct inode *inode; 1565 struct address_space *mapping; 1566 u64 nodeid; 1567 int err; 1568 pgoff_t index; 1569 unsigned int offset; 1570 unsigned int num; 1571 loff_t file_size; 1572 loff_t end; 1573 1574 err = -EINVAL; 1575 if (size < sizeof(outarg)) 1576 goto out_finish; 1577 1578 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1579 if (err) 1580 goto out_finish; 1581 1582 err = -EINVAL; 1583 if (size - sizeof(outarg) != outarg.size) 1584 goto out_finish; 1585 1586 nodeid = outarg.nodeid; 1587 1588 down_read(&fc->killsb); 1589 1590 err = -ENOENT; 1591 inode = fuse_ilookup(fc, nodeid, NULL); 1592 if (!inode) 1593 goto out_up_killsb; 1594 1595 mapping = inode->i_mapping; 1596 index = outarg.offset >> PAGE_SHIFT; 1597 offset = outarg.offset & ~PAGE_MASK; 1598 file_size = i_size_read(inode); 1599 end = outarg.offset + outarg.size; 1600 if (end > file_size) { 1601 file_size = end; 1602 fuse_write_update_attr(inode, file_size, outarg.size); 1603 } 1604 1605 num = outarg.size; 1606 while (num) { 1607 struct page *page; 1608 unsigned int this_num; 1609 1610 err = -ENOMEM; 1611 page = find_or_create_page(mapping, index, 1612 mapping_gfp_mask(mapping)); 1613 if (!page) 1614 goto out_iput; 1615 1616 this_num = min_t(unsigned, num, PAGE_SIZE - offset); 1617 err = fuse_copy_page(cs, &page, offset, this_num, 0); 1618 if (!err && offset == 0 && 1619 (this_num == PAGE_SIZE || file_size == end)) 1620 SetPageUptodate(page); 1621 unlock_page(page); 1622 put_page(page); 1623 1624 if (err) 1625 goto out_iput; 1626 1627 num -= this_num; 1628 offset = 0; 1629 index++; 1630 } 1631 1632 err = 0; 1633 1634 out_iput: 1635 iput(inode); 1636 out_up_killsb: 1637 up_read(&fc->killsb); 1638 out_finish: 1639 fuse_copy_finish(cs); 1640 return err; 1641 } 1642 1643 struct fuse_retrieve_args { 1644 struct fuse_args_pages ap; 1645 struct fuse_notify_retrieve_in inarg; 1646 }; 1647 1648 static void fuse_retrieve_end(struct fuse_mount *fm, struct fuse_args *args, 1649 int error) 1650 { 1651 struct fuse_retrieve_args *ra = 1652 container_of(args, typeof(*ra), ap.args); 1653 1654 release_pages(ra->ap.pages, ra->ap.num_pages); 1655 kfree(ra); 1656 } 1657 1658 static int fuse_retrieve(struct fuse_mount *fm, struct inode *inode, 1659 struct fuse_notify_retrieve_out *outarg) 1660 { 1661 int err; 1662 struct address_space *mapping = inode->i_mapping; 1663 pgoff_t index; 1664 loff_t file_size; 1665 unsigned int num; 1666 unsigned int offset; 1667 size_t total_len = 0; 1668 unsigned int num_pages; 1669 struct fuse_conn *fc = fm->fc; 1670 struct fuse_retrieve_args *ra; 1671 size_t args_size = sizeof(*ra); 1672 struct fuse_args_pages *ap; 1673 struct fuse_args *args; 1674 1675 offset = outarg->offset & ~PAGE_MASK; 1676 file_size = i_size_read(inode); 1677 1678 num = min(outarg->size, fc->max_write); 1679 if (outarg->offset > file_size) 1680 num = 0; 1681 else if (outarg->offset + num > file_size) 1682 num = file_size - outarg->offset; 1683 1684 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; 1685 num_pages = min(num_pages, fc->max_pages); 1686 1687 args_size += num_pages * (sizeof(ap->pages[0]) + sizeof(ap->descs[0])); 1688 1689 ra = kzalloc(args_size, GFP_KERNEL); 1690 if (!ra) 1691 return -ENOMEM; 1692 1693 ap = &ra->ap; 1694 ap->pages = (void *) (ra + 1); 1695 ap->descs = (void *) (ap->pages + num_pages); 1696 1697 args = &ap->args; 1698 args->nodeid = outarg->nodeid; 1699 args->opcode = FUSE_NOTIFY_REPLY; 1700 args->in_numargs = 2; 1701 args->in_pages = true; 1702 args->end = fuse_retrieve_end; 1703 1704 index = outarg->offset >> PAGE_SHIFT; 1705 1706 while (num && ap->num_pages < num_pages) { 1707 struct page *page; 1708 unsigned int this_num; 1709 1710 page = find_get_page(mapping, index); 1711 if (!page) 1712 break; 1713 1714 this_num = min_t(unsigned, num, PAGE_SIZE - offset); 1715 ap->pages[ap->num_pages] = page; 1716 ap->descs[ap->num_pages].offset = offset; 1717 ap->descs[ap->num_pages].length = this_num; 1718 ap->num_pages++; 1719 1720 offset = 0; 1721 num -= this_num; 1722 total_len += this_num; 1723 index++; 1724 } 1725 ra->inarg.offset = outarg->offset; 1726 ra->inarg.size = total_len; 1727 args->in_args[0].size = sizeof(ra->inarg); 1728 args->in_args[0].value = &ra->inarg; 1729 args->in_args[1].size = total_len; 1730 1731 err = fuse_simple_notify_reply(fm, args, outarg->notify_unique); 1732 if (err) 1733 fuse_retrieve_end(fm, args, err); 1734 1735 return err; 1736 } 1737 1738 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size, 1739 struct fuse_copy_state *cs) 1740 { 1741 struct fuse_notify_retrieve_out outarg; 1742 struct fuse_mount *fm; 1743 struct inode *inode; 1744 u64 nodeid; 1745 int err; 1746 1747 err = -EINVAL; 1748 if (size != sizeof(outarg)) 1749 goto copy_finish; 1750 1751 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1752 if (err) 1753 goto copy_finish; 1754 1755 fuse_copy_finish(cs); 1756 1757 down_read(&fc->killsb); 1758 err = -ENOENT; 1759 nodeid = outarg.nodeid; 1760 1761 inode = fuse_ilookup(fc, nodeid, &fm); 1762 if (inode) { 1763 err = fuse_retrieve(fm, inode, &outarg); 1764 iput(inode); 1765 } 1766 up_read(&fc->killsb); 1767 1768 return err; 1769 1770 copy_finish: 1771 fuse_copy_finish(cs); 1772 return err; 1773 } 1774 1775 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code, 1776 unsigned int size, struct fuse_copy_state *cs) 1777 { 1778 /* Don't try to move pages (yet) */ 1779 cs->move_pages = 0; 1780 1781 switch (code) { 1782 case FUSE_NOTIFY_POLL: 1783 return fuse_notify_poll(fc, size, cs); 1784 1785 case FUSE_NOTIFY_INVAL_INODE: 1786 return fuse_notify_inval_inode(fc, size, cs); 1787 1788 case FUSE_NOTIFY_INVAL_ENTRY: 1789 return fuse_notify_inval_entry(fc, size, cs); 1790 1791 case FUSE_NOTIFY_STORE: 1792 return fuse_notify_store(fc, size, cs); 1793 1794 case FUSE_NOTIFY_RETRIEVE: 1795 return fuse_notify_retrieve(fc, size, cs); 1796 1797 case FUSE_NOTIFY_DELETE: 1798 return fuse_notify_delete(fc, size, cs); 1799 1800 default: 1801 fuse_copy_finish(cs); 1802 return -EINVAL; 1803 } 1804 } 1805 1806 /* Look up request on processing list by unique ID */ 1807 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique) 1808 { 1809 unsigned int hash = fuse_req_hash(unique); 1810 struct fuse_req *req; 1811 1812 list_for_each_entry(req, &fpq->processing[hash], list) { 1813 if (req->in.h.unique == unique) 1814 return req; 1815 } 1816 return NULL; 1817 } 1818 1819 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args, 1820 unsigned nbytes) 1821 { 1822 unsigned reqsize = sizeof(struct fuse_out_header); 1823 1824 reqsize += fuse_len_args(args->out_numargs, args->out_args); 1825 1826 if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar)) 1827 return -EINVAL; 1828 else if (reqsize > nbytes) { 1829 struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1]; 1830 unsigned diffsize = reqsize - nbytes; 1831 1832 if (diffsize > lastarg->size) 1833 return -EINVAL; 1834 lastarg->size -= diffsize; 1835 } 1836 return fuse_copy_args(cs, args->out_numargs, args->out_pages, 1837 args->out_args, args->page_zeroing); 1838 } 1839 1840 /* 1841 * Write a single reply to a request. First the header is copied from 1842 * the write buffer. The request is then searched on the processing 1843 * list by the unique ID found in the header. If found, then remove 1844 * it from the list and copy the rest of the buffer to the request. 1845 * The request is finished by calling fuse_request_end(). 1846 */ 1847 static ssize_t fuse_dev_do_write(struct fuse_dev *fud, 1848 struct fuse_copy_state *cs, size_t nbytes) 1849 { 1850 int err; 1851 struct fuse_conn *fc = fud->fc; 1852 struct fuse_pqueue *fpq = &fud->pq; 1853 struct fuse_req *req; 1854 struct fuse_out_header oh; 1855 1856 err = -EINVAL; 1857 if (nbytes < sizeof(struct fuse_out_header)) 1858 goto out; 1859 1860 err = fuse_copy_one(cs, &oh, sizeof(oh)); 1861 if (err) 1862 goto copy_finish; 1863 1864 err = -EINVAL; 1865 if (oh.len != nbytes) 1866 goto copy_finish; 1867 1868 /* 1869 * Zero oh.unique indicates unsolicited notification message 1870 * and error contains notification code. 1871 */ 1872 if (!oh.unique) { 1873 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs); 1874 goto out; 1875 } 1876 1877 err = -EINVAL; 1878 if (oh.error <= -512 || oh.error > 0) 1879 goto copy_finish; 1880 1881 spin_lock(&fpq->lock); 1882 req = NULL; 1883 if (fpq->connected) 1884 req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT); 1885 1886 err = -ENOENT; 1887 if (!req) { 1888 spin_unlock(&fpq->lock); 1889 goto copy_finish; 1890 } 1891 1892 /* Is it an interrupt reply ID? */ 1893 if (oh.unique & FUSE_INT_REQ_BIT) { 1894 __fuse_get_request(req); 1895 spin_unlock(&fpq->lock); 1896 1897 err = 0; 1898 if (nbytes != sizeof(struct fuse_out_header)) 1899 err = -EINVAL; 1900 else if (oh.error == -ENOSYS) 1901 fc->no_interrupt = 1; 1902 else if (oh.error == -EAGAIN) 1903 err = queue_interrupt(req); 1904 1905 fuse_put_request(req); 1906 1907 goto copy_finish; 1908 } 1909 1910 clear_bit(FR_SENT, &req->flags); 1911 list_move(&req->list, &fpq->io); 1912 req->out.h = oh; 1913 set_bit(FR_LOCKED, &req->flags); 1914 spin_unlock(&fpq->lock); 1915 cs->req = req; 1916 if (!req->args->page_replace) 1917 cs->move_pages = 0; 1918 1919 if (oh.error) 1920 err = nbytes != sizeof(oh) ? -EINVAL : 0; 1921 else 1922 err = copy_out_args(cs, req->args, nbytes); 1923 fuse_copy_finish(cs); 1924 1925 spin_lock(&fpq->lock); 1926 clear_bit(FR_LOCKED, &req->flags); 1927 if (!fpq->connected) 1928 err = -ENOENT; 1929 else if (err) 1930 req->out.h.error = -EIO; 1931 if (!test_bit(FR_PRIVATE, &req->flags)) 1932 list_del_init(&req->list); 1933 spin_unlock(&fpq->lock); 1934 1935 fuse_request_end(req); 1936 out: 1937 return err ? err : nbytes; 1938 1939 copy_finish: 1940 fuse_copy_finish(cs); 1941 goto out; 1942 } 1943 1944 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from) 1945 { 1946 struct fuse_copy_state cs; 1947 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp); 1948 1949 if (!fud) 1950 return -EPERM; 1951 1952 if (!iter_is_iovec(from)) 1953 return -EINVAL; 1954 1955 fuse_copy_init(&cs, 0, from); 1956 1957 return fuse_dev_do_write(fud, &cs, iov_iter_count(from)); 1958 } 1959 1960 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe, 1961 struct file *out, loff_t *ppos, 1962 size_t len, unsigned int flags) 1963 { 1964 unsigned int head, tail, mask, count; 1965 unsigned nbuf; 1966 unsigned idx; 1967 struct pipe_buffer *bufs; 1968 struct fuse_copy_state cs; 1969 struct fuse_dev *fud; 1970 size_t rem; 1971 ssize_t ret; 1972 1973 fud = fuse_get_dev(out); 1974 if (!fud) 1975 return -EPERM; 1976 1977 pipe_lock(pipe); 1978 1979 head = pipe->head; 1980 tail = pipe->tail; 1981 mask = pipe->ring_size - 1; 1982 count = head - tail; 1983 1984 bufs = kvmalloc_array(count, sizeof(struct pipe_buffer), GFP_KERNEL); 1985 if (!bufs) { 1986 pipe_unlock(pipe); 1987 return -ENOMEM; 1988 } 1989 1990 nbuf = 0; 1991 rem = 0; 1992 for (idx = tail; idx != head && rem < len; idx++) 1993 rem += pipe->bufs[idx & mask].len; 1994 1995 ret = -EINVAL; 1996 if (rem < len) 1997 goto out_free; 1998 1999 rem = len; 2000 while (rem) { 2001 struct pipe_buffer *ibuf; 2002 struct pipe_buffer *obuf; 2003 2004 if (WARN_ON(nbuf >= count || tail == head)) 2005 goto out_free; 2006 2007 ibuf = &pipe->bufs[tail & mask]; 2008 obuf = &bufs[nbuf]; 2009 2010 if (rem >= ibuf->len) { 2011 *obuf = *ibuf; 2012 ibuf->ops = NULL; 2013 tail++; 2014 pipe->tail = tail; 2015 } else { 2016 if (!pipe_buf_get(pipe, ibuf)) 2017 goto out_free; 2018 2019 *obuf = *ibuf; 2020 obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 2021 obuf->len = rem; 2022 ibuf->offset += obuf->len; 2023 ibuf->len -= obuf->len; 2024 } 2025 nbuf++; 2026 rem -= obuf->len; 2027 } 2028 pipe_unlock(pipe); 2029 2030 fuse_copy_init(&cs, 0, NULL); 2031 cs.pipebufs = bufs; 2032 cs.nr_segs = nbuf; 2033 cs.pipe = pipe; 2034 2035 if (flags & SPLICE_F_MOVE) 2036 cs.move_pages = 1; 2037 2038 ret = fuse_dev_do_write(fud, &cs, len); 2039 2040 pipe_lock(pipe); 2041 out_free: 2042 for (idx = 0; idx < nbuf; idx++) { 2043 struct pipe_buffer *buf = &bufs[idx]; 2044 2045 if (buf->ops) 2046 pipe_buf_release(pipe, buf); 2047 } 2048 pipe_unlock(pipe); 2049 2050 kvfree(bufs); 2051 return ret; 2052 } 2053 2054 static __poll_t fuse_dev_poll(struct file *file, poll_table *wait) 2055 { 2056 __poll_t mask = EPOLLOUT | EPOLLWRNORM; 2057 struct fuse_iqueue *fiq; 2058 struct fuse_dev *fud = fuse_get_dev(file); 2059 2060 if (!fud) 2061 return EPOLLERR; 2062 2063 fiq = &fud->fc->iq; 2064 poll_wait(file, &fiq->waitq, wait); 2065 2066 spin_lock(&fiq->lock); 2067 if (!fiq->connected) 2068 mask = EPOLLERR; 2069 else if (request_pending(fiq)) 2070 mask |= EPOLLIN | EPOLLRDNORM; 2071 spin_unlock(&fiq->lock); 2072 2073 return mask; 2074 } 2075 2076 /* Abort all requests on the given list (pending or processing) */ 2077 static void end_requests(struct list_head *head) 2078 { 2079 while (!list_empty(head)) { 2080 struct fuse_req *req; 2081 req = list_entry(head->next, struct fuse_req, list); 2082 req->out.h.error = -ECONNABORTED; 2083 clear_bit(FR_SENT, &req->flags); 2084 list_del_init(&req->list); 2085 fuse_request_end(req); 2086 } 2087 } 2088 2089 static void end_polls(struct fuse_conn *fc) 2090 { 2091 struct rb_node *p; 2092 2093 p = rb_first(&fc->polled_files); 2094 2095 while (p) { 2096 struct fuse_file *ff; 2097 ff = rb_entry(p, struct fuse_file, polled_node); 2098 wake_up_interruptible_all(&ff->poll_wait); 2099 2100 p = rb_next(p); 2101 } 2102 } 2103 2104 /* 2105 * Abort all requests. 2106 * 2107 * Emergency exit in case of a malicious or accidental deadlock, or just a hung 2108 * filesystem. 2109 * 2110 * The same effect is usually achievable through killing the filesystem daemon 2111 * and all users of the filesystem. The exception is the combination of an 2112 * asynchronous request and the tricky deadlock (see 2113 * Documentation/filesystems/fuse.rst). 2114 * 2115 * Aborting requests under I/O goes as follows: 1: Separate out unlocked 2116 * requests, they should be finished off immediately. Locked requests will be 2117 * finished after unlock; see unlock_request(). 2: Finish off the unlocked 2118 * requests. It is possible that some request will finish before we can. This 2119 * is OK, the request will in that case be removed from the list before we touch 2120 * it. 2121 */ 2122 void fuse_abort_conn(struct fuse_conn *fc) 2123 { 2124 struct fuse_iqueue *fiq = &fc->iq; 2125 2126 spin_lock(&fc->lock); 2127 if (fc->connected) { 2128 struct fuse_dev *fud; 2129 struct fuse_req *req, *next; 2130 LIST_HEAD(to_end); 2131 unsigned int i; 2132 2133 /* Background queuing checks fc->connected under bg_lock */ 2134 spin_lock(&fc->bg_lock); 2135 fc->connected = 0; 2136 spin_unlock(&fc->bg_lock); 2137 2138 fuse_set_initialized(fc); 2139 list_for_each_entry(fud, &fc->devices, entry) { 2140 struct fuse_pqueue *fpq = &fud->pq; 2141 2142 spin_lock(&fpq->lock); 2143 fpq->connected = 0; 2144 list_for_each_entry_safe(req, next, &fpq->io, list) { 2145 req->out.h.error = -ECONNABORTED; 2146 spin_lock(&req->waitq.lock); 2147 set_bit(FR_ABORTED, &req->flags); 2148 if (!test_bit(FR_LOCKED, &req->flags)) { 2149 set_bit(FR_PRIVATE, &req->flags); 2150 __fuse_get_request(req); 2151 list_move(&req->list, &to_end); 2152 } 2153 spin_unlock(&req->waitq.lock); 2154 } 2155 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 2156 list_splice_tail_init(&fpq->processing[i], 2157 &to_end); 2158 spin_unlock(&fpq->lock); 2159 } 2160 spin_lock(&fc->bg_lock); 2161 fc->blocked = 0; 2162 fc->max_background = UINT_MAX; 2163 flush_bg_queue(fc); 2164 spin_unlock(&fc->bg_lock); 2165 2166 spin_lock(&fiq->lock); 2167 fiq->connected = 0; 2168 list_for_each_entry(req, &fiq->pending, list) 2169 clear_bit(FR_PENDING, &req->flags); 2170 list_splice_tail_init(&fiq->pending, &to_end); 2171 while (forget_pending(fiq)) 2172 kfree(fuse_dequeue_forget(fiq, 1, NULL)); 2173 wake_up_all(&fiq->waitq); 2174 spin_unlock(&fiq->lock); 2175 kill_fasync(&fiq->fasync, SIGIO, POLL_IN); 2176 end_polls(fc); 2177 wake_up_all(&fc->blocked_waitq); 2178 spin_unlock(&fc->lock); 2179 2180 end_requests(&to_end); 2181 } else { 2182 spin_unlock(&fc->lock); 2183 } 2184 } 2185 EXPORT_SYMBOL_GPL(fuse_abort_conn); 2186 2187 void fuse_wait_aborted(struct fuse_conn *fc) 2188 { 2189 /* matches implicit memory barrier in fuse_drop_waiting() */ 2190 smp_mb(); 2191 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0); 2192 } 2193 2194 int fuse_dev_release(struct inode *inode, struct file *file) 2195 { 2196 struct fuse_dev *fud = fuse_get_dev(file); 2197 2198 if (fud) { 2199 struct fuse_conn *fc = fud->fc; 2200 struct fuse_pqueue *fpq = &fud->pq; 2201 LIST_HEAD(to_end); 2202 unsigned int i; 2203 2204 spin_lock(&fpq->lock); 2205 WARN_ON(!list_empty(&fpq->io)); 2206 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 2207 list_splice_init(&fpq->processing[i], &to_end); 2208 spin_unlock(&fpq->lock); 2209 2210 end_requests(&to_end); 2211 2212 /* Are we the last open device? */ 2213 if (atomic_dec_and_test(&fc->dev_count)) { 2214 WARN_ON(fc->iq.fasync != NULL); 2215 fuse_abort_conn(fc); 2216 } 2217 fuse_dev_free(fud); 2218 } 2219 return 0; 2220 } 2221 EXPORT_SYMBOL_GPL(fuse_dev_release); 2222 2223 static int fuse_dev_fasync(int fd, struct file *file, int on) 2224 { 2225 struct fuse_dev *fud = fuse_get_dev(file); 2226 2227 if (!fud) 2228 return -EPERM; 2229 2230 /* No locking - fasync_helper does its own locking */ 2231 return fasync_helper(fd, file, on, &fud->fc->iq.fasync); 2232 } 2233 2234 static int fuse_device_clone(struct fuse_conn *fc, struct file *new) 2235 { 2236 struct fuse_dev *fud; 2237 2238 if (new->private_data) 2239 return -EINVAL; 2240 2241 fud = fuse_dev_alloc_install(fc); 2242 if (!fud) 2243 return -ENOMEM; 2244 2245 new->private_data = fud; 2246 atomic_inc(&fc->dev_count); 2247 2248 return 0; 2249 } 2250 2251 static long fuse_dev_ioctl(struct file *file, unsigned int cmd, 2252 unsigned long arg) 2253 { 2254 int res; 2255 int oldfd; 2256 struct fuse_dev *fud = NULL; 2257 2258 switch (cmd) { 2259 case FUSE_DEV_IOC_CLONE: 2260 res = -EFAULT; 2261 if (!get_user(oldfd, (__u32 __user *)arg)) { 2262 struct file *old = fget(oldfd); 2263 2264 res = -EINVAL; 2265 if (old) { 2266 /* 2267 * Check against file->f_op because CUSE 2268 * uses the same ioctl handler. 2269 */ 2270 if (old->f_op == file->f_op && 2271 old->f_cred->user_ns == file->f_cred->user_ns) 2272 fud = fuse_get_dev(old); 2273 2274 if (fud) { 2275 mutex_lock(&fuse_mutex); 2276 res = fuse_device_clone(fud->fc, file); 2277 mutex_unlock(&fuse_mutex); 2278 } 2279 fput(old); 2280 } 2281 } 2282 break; 2283 default: 2284 res = -ENOTTY; 2285 break; 2286 } 2287 return res; 2288 } 2289 2290 const struct file_operations fuse_dev_operations = { 2291 .owner = THIS_MODULE, 2292 .open = fuse_dev_open, 2293 .llseek = no_llseek, 2294 .read_iter = fuse_dev_read, 2295 .splice_read = fuse_dev_splice_read, 2296 .write_iter = fuse_dev_write, 2297 .splice_write = fuse_dev_splice_write, 2298 .poll = fuse_dev_poll, 2299 .release = fuse_dev_release, 2300 .fasync = fuse_dev_fasync, 2301 .unlocked_ioctl = fuse_dev_ioctl, 2302 .compat_ioctl = compat_ptr_ioctl, 2303 }; 2304 EXPORT_SYMBOL_GPL(fuse_dev_operations); 2305 2306 static struct miscdevice fuse_miscdevice = { 2307 .minor = FUSE_MINOR, 2308 .name = "fuse", 2309 .fops = &fuse_dev_operations, 2310 }; 2311 2312 int __init fuse_dev_init(void) 2313 { 2314 int err = -ENOMEM; 2315 fuse_req_cachep = kmem_cache_create("fuse_request", 2316 sizeof(struct fuse_req), 2317 0, 0, NULL); 2318 if (!fuse_req_cachep) 2319 goto out; 2320 2321 err = misc_register(&fuse_miscdevice); 2322 if (err) 2323 goto out_cache_clean; 2324 2325 return 0; 2326 2327 out_cache_clean: 2328 kmem_cache_destroy(fuse_req_cachep); 2329 out: 2330 return err; 2331 } 2332 2333 void fuse_dev_cleanup(void) 2334 { 2335 misc_deregister(&fuse_miscdevice); 2336 kmem_cache_destroy(fuse_req_cachep); 2337 } 2338