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_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 } 51 52 static struct fuse_req *fuse_request_alloc(gfp_t flags) 53 { 54 struct fuse_req *req = kmem_cache_zalloc(fuse_req_cachep, flags); 55 if (req) 56 fuse_request_init(req); 57 58 return req; 59 } 60 61 static void fuse_request_free(struct fuse_req *req) 62 { 63 kmem_cache_free(fuse_req_cachep, req); 64 } 65 66 static void __fuse_get_request(struct fuse_req *req) 67 { 68 refcount_inc(&req->count); 69 } 70 71 /* Must be called with > 1 refcount */ 72 static void __fuse_put_request(struct fuse_req *req) 73 { 74 refcount_dec(&req->count); 75 } 76 77 void fuse_set_initialized(struct fuse_conn *fc) 78 { 79 /* Make sure stores before this are seen on another CPU */ 80 smp_wmb(); 81 fc->initialized = 1; 82 } 83 84 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background) 85 { 86 return !fc->initialized || (for_background && fc->blocked); 87 } 88 89 static void fuse_drop_waiting(struct fuse_conn *fc) 90 { 91 /* 92 * lockess check of fc->connected is okay, because atomic_dec_and_test() 93 * provides a memory barrier mached with the one in fuse_wait_aborted() 94 * to ensure no wake-up is missed. 95 */ 96 if (atomic_dec_and_test(&fc->num_waiting) && 97 !READ_ONCE(fc->connected)) { 98 /* wake up aborters */ 99 wake_up_all(&fc->blocked_waitq); 100 } 101 } 102 103 static void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req); 104 105 static struct fuse_req *fuse_get_req(struct fuse_conn *fc, bool for_background) 106 { 107 struct fuse_req *req; 108 int err; 109 atomic_inc(&fc->num_waiting); 110 111 if (fuse_block_alloc(fc, for_background)) { 112 err = -EINTR; 113 if (wait_event_killable_exclusive(fc->blocked_waitq, 114 !fuse_block_alloc(fc, for_background))) 115 goto out; 116 } 117 /* Matches smp_wmb() in fuse_set_initialized() */ 118 smp_rmb(); 119 120 err = -ENOTCONN; 121 if (!fc->connected) 122 goto out; 123 124 err = -ECONNREFUSED; 125 if (fc->conn_error) 126 goto out; 127 128 req = fuse_request_alloc(GFP_KERNEL); 129 err = -ENOMEM; 130 if (!req) { 131 if (for_background) 132 wake_up(&fc->blocked_waitq); 133 goto out; 134 } 135 136 req->in.h.uid = from_kuid(fc->user_ns, current_fsuid()); 137 req->in.h.gid = from_kgid(fc->user_ns, current_fsgid()); 138 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns); 139 140 __set_bit(FR_WAITING, &req->flags); 141 if (for_background) 142 __set_bit(FR_BACKGROUND, &req->flags); 143 144 if (unlikely(req->in.h.uid == ((uid_t)-1) || 145 req->in.h.gid == ((gid_t)-1))) { 146 fuse_put_request(fc, req); 147 return ERR_PTR(-EOVERFLOW); 148 } 149 return req; 150 151 out: 152 fuse_drop_waiting(fc); 153 return ERR_PTR(err); 154 } 155 156 static void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) 157 { 158 if (refcount_dec_and_test(&req->count)) { 159 if (test_bit(FR_BACKGROUND, &req->flags)) { 160 /* 161 * We get here in the unlikely case that a background 162 * request was allocated but not sent 163 */ 164 spin_lock(&fc->bg_lock); 165 if (!fc->blocked) 166 wake_up(&fc->blocked_waitq); 167 spin_unlock(&fc->bg_lock); 168 } 169 170 if (test_bit(FR_WAITING, &req->flags)) { 171 __clear_bit(FR_WAITING, &req->flags); 172 fuse_drop_waiting(fc); 173 } 174 175 fuse_request_free(req); 176 } 177 } 178 EXPORT_SYMBOL_GPL(fuse_put_request); 179 180 static unsigned len_args(unsigned numargs, struct fuse_arg *args) 181 { 182 unsigned nbytes = 0; 183 unsigned i; 184 185 for (i = 0; i < numargs; i++) 186 nbytes += args[i].size; 187 188 return nbytes; 189 } 190 191 static u64 fuse_get_unique(struct fuse_iqueue *fiq) 192 { 193 fiq->reqctr += FUSE_REQ_ID_STEP; 194 return fiq->reqctr; 195 } 196 197 static unsigned int fuse_req_hash(u64 unique) 198 { 199 return hash_long(unique & ~FUSE_INT_REQ_BIT, FUSE_PQ_HASH_BITS); 200 } 201 202 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req) 203 { 204 req->in.h.len = sizeof(struct fuse_in_header) + 205 len_args(req->args->in_numargs, 206 (struct fuse_arg *) req->args->in_args); 207 list_add_tail(&req->list, &fiq->pending); 208 wake_up(&fiq->waitq); 209 kill_fasync(&fiq->fasync, SIGIO, POLL_IN); 210 } 211 212 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget, 213 u64 nodeid, u64 nlookup) 214 { 215 struct fuse_iqueue *fiq = &fc->iq; 216 217 forget->forget_one.nodeid = nodeid; 218 forget->forget_one.nlookup = nlookup; 219 220 spin_lock(&fiq->lock); 221 if (fiq->connected) { 222 fiq->forget_list_tail->next = forget; 223 fiq->forget_list_tail = forget; 224 wake_up(&fiq->waitq); 225 kill_fasync(&fiq->fasync, SIGIO, POLL_IN); 226 } else { 227 kfree(forget); 228 } 229 spin_unlock(&fiq->lock); 230 } 231 232 static void flush_bg_queue(struct fuse_conn *fc) 233 { 234 struct fuse_iqueue *fiq = &fc->iq; 235 236 while (fc->active_background < fc->max_background && 237 !list_empty(&fc->bg_queue)) { 238 struct fuse_req *req; 239 240 req = list_first_entry(&fc->bg_queue, struct fuse_req, list); 241 list_del(&req->list); 242 fc->active_background++; 243 spin_lock(&fiq->lock); 244 req->in.h.unique = fuse_get_unique(fiq); 245 queue_request(fiq, req); 246 spin_unlock(&fiq->lock); 247 } 248 } 249 250 /* 251 * This function is called when a request is finished. Either a reply 252 * has arrived or it was aborted (and not yet sent) or some error 253 * occurred during communication with userspace, or the device file 254 * was closed. The requester thread is woken up (if still waiting), 255 * the 'end' callback is called if given, else the reference to the 256 * request is released 257 */ 258 void fuse_request_end(struct fuse_conn *fc, struct fuse_req *req) 259 { 260 struct fuse_iqueue *fiq = &fc->iq; 261 bool async = req->args->end; 262 263 if (test_and_set_bit(FR_FINISHED, &req->flags)) 264 goto put_request; 265 /* 266 * test_and_set_bit() implies smp_mb() between bit 267 * changing and below intr_entry check. Pairs with 268 * smp_mb() from queue_interrupt(). 269 */ 270 if (!list_empty(&req->intr_entry)) { 271 spin_lock(&fiq->lock); 272 list_del_init(&req->intr_entry); 273 spin_unlock(&fiq->lock); 274 } 275 WARN_ON(test_bit(FR_PENDING, &req->flags)); 276 WARN_ON(test_bit(FR_SENT, &req->flags)); 277 if (test_bit(FR_BACKGROUND, &req->flags)) { 278 spin_lock(&fc->bg_lock); 279 clear_bit(FR_BACKGROUND, &req->flags); 280 if (fc->num_background == fc->max_background) { 281 fc->blocked = 0; 282 wake_up(&fc->blocked_waitq); 283 } else if (!fc->blocked) { 284 /* 285 * Wake up next waiter, if any. It's okay to use 286 * waitqueue_active(), as we've already synced up 287 * fc->blocked with waiters with the wake_up() call 288 * above. 289 */ 290 if (waitqueue_active(&fc->blocked_waitq)) 291 wake_up(&fc->blocked_waitq); 292 } 293 294 if (fc->num_background == fc->congestion_threshold && fc->sb) { 295 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC); 296 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC); 297 } 298 fc->num_background--; 299 fc->active_background--; 300 flush_bg_queue(fc); 301 spin_unlock(&fc->bg_lock); 302 } else { 303 /* Wake up waiter sleeping in request_wait_answer() */ 304 wake_up(&req->waitq); 305 } 306 307 if (async) 308 req->args->end(fc, req->args, req->out.h.error); 309 put_request: 310 fuse_put_request(fc, req); 311 } 312 EXPORT_SYMBOL_GPL(fuse_request_end); 313 314 static int queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req) 315 { 316 spin_lock(&fiq->lock); 317 /* Check for we've sent request to interrupt this req */ 318 if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags))) { 319 spin_unlock(&fiq->lock); 320 return -EINVAL; 321 } 322 323 if (list_empty(&req->intr_entry)) { 324 list_add_tail(&req->intr_entry, &fiq->interrupts); 325 /* 326 * Pairs with smp_mb() implied by test_and_set_bit() 327 * from request_end(). 328 */ 329 smp_mb(); 330 if (test_bit(FR_FINISHED, &req->flags)) { 331 list_del_init(&req->intr_entry); 332 spin_unlock(&fiq->lock); 333 return 0; 334 } 335 wake_up(&fiq->waitq); 336 kill_fasync(&fiq->fasync, SIGIO, POLL_IN); 337 } 338 spin_unlock(&fiq->lock); 339 return 0; 340 } 341 342 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req) 343 { 344 struct fuse_iqueue *fiq = &fc->iq; 345 int err; 346 347 if (!fc->no_interrupt) { 348 /* Any signal may interrupt this */ 349 err = wait_event_interruptible(req->waitq, 350 test_bit(FR_FINISHED, &req->flags)); 351 if (!err) 352 return; 353 354 set_bit(FR_INTERRUPTED, &req->flags); 355 /* matches barrier in fuse_dev_do_read() */ 356 smp_mb__after_atomic(); 357 if (test_bit(FR_SENT, &req->flags)) 358 queue_interrupt(fiq, req); 359 } 360 361 if (!test_bit(FR_FORCE, &req->flags)) { 362 /* Only fatal signals may interrupt this */ 363 err = wait_event_killable(req->waitq, 364 test_bit(FR_FINISHED, &req->flags)); 365 if (!err) 366 return; 367 368 spin_lock(&fiq->lock); 369 /* Request is not yet in userspace, bail out */ 370 if (test_bit(FR_PENDING, &req->flags)) { 371 list_del(&req->list); 372 spin_unlock(&fiq->lock); 373 __fuse_put_request(req); 374 req->out.h.error = -EINTR; 375 return; 376 } 377 spin_unlock(&fiq->lock); 378 } 379 380 /* 381 * Either request is already in userspace, or it was forced. 382 * Wait it out. 383 */ 384 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags)); 385 } 386 387 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req) 388 { 389 struct fuse_iqueue *fiq = &fc->iq; 390 391 BUG_ON(test_bit(FR_BACKGROUND, &req->flags)); 392 spin_lock(&fiq->lock); 393 if (!fiq->connected) { 394 spin_unlock(&fiq->lock); 395 req->out.h.error = -ENOTCONN; 396 } else { 397 req->in.h.unique = fuse_get_unique(fiq); 398 queue_request(fiq, req); 399 /* acquire extra reference, since request is still needed 400 after fuse_request_end() */ 401 __fuse_get_request(req); 402 spin_unlock(&fiq->lock); 403 404 request_wait_answer(fc, req); 405 /* Pairs with smp_wmb() in fuse_request_end() */ 406 smp_rmb(); 407 } 408 } 409 410 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args) 411 { 412 if (fc->minor < 4 && args->opcode == FUSE_STATFS) 413 args->out_args[0].size = FUSE_COMPAT_STATFS_SIZE; 414 415 if (fc->minor < 9) { 416 switch (args->opcode) { 417 case FUSE_LOOKUP: 418 case FUSE_CREATE: 419 case FUSE_MKNOD: 420 case FUSE_MKDIR: 421 case FUSE_SYMLINK: 422 case FUSE_LINK: 423 args->out_args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE; 424 break; 425 case FUSE_GETATTR: 426 case FUSE_SETATTR: 427 args->out_args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE; 428 break; 429 } 430 } 431 if (fc->minor < 12) { 432 switch (args->opcode) { 433 case FUSE_CREATE: 434 args->in_args[0].size = sizeof(struct fuse_open_in); 435 break; 436 case FUSE_MKNOD: 437 args->in_args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE; 438 break; 439 } 440 } 441 } 442 443 static void fuse_force_creds(struct fuse_conn *fc, struct fuse_req *req) 444 { 445 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid()); 446 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid()); 447 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns); 448 } 449 450 void fuse_args_to_req(struct fuse_req *req, struct fuse_args *args) 451 { 452 req->in.h.opcode = args->opcode; 453 req->in.h.nodeid = args->nodeid; 454 req->args = args; 455 } 456 457 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args) 458 { 459 struct fuse_req *req; 460 ssize_t ret; 461 462 if (args->force) { 463 atomic_inc(&fc->num_waiting); 464 req = fuse_request_alloc(GFP_KERNEL | __GFP_NOFAIL); 465 466 if (!args->nocreds) 467 fuse_force_creds(fc, req); 468 469 __set_bit(FR_WAITING, &req->flags); 470 __set_bit(FR_FORCE, &req->flags); 471 } else { 472 WARN_ON(args->nocreds); 473 req = fuse_get_req(fc, false); 474 if (IS_ERR(req)) 475 return PTR_ERR(req); 476 } 477 478 /* Needs to be done after fuse_get_req() so that fc->minor is valid */ 479 fuse_adjust_compat(fc, args); 480 fuse_args_to_req(req, args); 481 482 if (!args->noreply) 483 __set_bit(FR_ISREPLY, &req->flags); 484 __fuse_request_send(fc, req); 485 ret = req->out.h.error; 486 if (!ret && args->out_argvar) { 487 BUG_ON(args->out_numargs == 0); 488 ret = args->out_args[args->out_numargs - 1].size; 489 } 490 fuse_put_request(fc, req); 491 492 return ret; 493 } 494 495 static bool fuse_request_queue_background(struct fuse_conn *fc, 496 struct fuse_req *req) 497 { 498 bool queued = false; 499 500 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags)); 501 if (!test_bit(FR_WAITING, &req->flags)) { 502 __set_bit(FR_WAITING, &req->flags); 503 atomic_inc(&fc->num_waiting); 504 } 505 __set_bit(FR_ISREPLY, &req->flags); 506 spin_lock(&fc->bg_lock); 507 if (likely(fc->connected)) { 508 fc->num_background++; 509 if (fc->num_background == fc->max_background) 510 fc->blocked = 1; 511 if (fc->num_background == fc->congestion_threshold && fc->sb) { 512 set_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC); 513 set_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC); 514 } 515 list_add_tail(&req->list, &fc->bg_queue); 516 flush_bg_queue(fc); 517 queued = true; 518 } 519 spin_unlock(&fc->bg_lock); 520 521 return queued; 522 } 523 524 int fuse_simple_background(struct fuse_conn *fc, struct fuse_args *args, 525 gfp_t gfp_flags) 526 { 527 struct fuse_req *req; 528 529 if (args->force) { 530 WARN_ON(!args->nocreds); 531 req = fuse_request_alloc(gfp_flags); 532 if (!req) 533 return -ENOMEM; 534 __set_bit(FR_BACKGROUND, &req->flags); 535 } else { 536 WARN_ON(args->nocreds); 537 req = fuse_get_req(fc, true); 538 if (IS_ERR(req)) 539 return PTR_ERR(req); 540 } 541 542 fuse_args_to_req(req, args); 543 544 if (!fuse_request_queue_background(fc, req)) { 545 fuse_put_request(fc, req); 546 return -ENOTCONN; 547 } 548 549 return 0; 550 } 551 EXPORT_SYMBOL_GPL(fuse_simple_background); 552 553 static int fuse_simple_notify_reply(struct fuse_conn *fc, 554 struct fuse_args *args, u64 unique) 555 { 556 struct fuse_req *req; 557 struct fuse_iqueue *fiq = &fc->iq; 558 int err = 0; 559 560 req = fuse_get_req(fc, false); 561 if (IS_ERR(req)) 562 return PTR_ERR(req); 563 564 __clear_bit(FR_ISREPLY, &req->flags); 565 req->in.h.unique = unique; 566 567 fuse_args_to_req(req, args); 568 569 spin_lock(&fiq->lock); 570 if (fiq->connected) { 571 queue_request(fiq, req); 572 spin_unlock(&fiq->lock); 573 } else { 574 err = -ENODEV; 575 spin_unlock(&fiq->lock); 576 fuse_put_request(fc, req); 577 } 578 spin_unlock(&fiq->lock); 579 580 return err; 581 } 582 583 /* 584 * Lock the request. Up to the next unlock_request() there mustn't be 585 * anything that could cause a page-fault. If the request was already 586 * aborted bail out. 587 */ 588 static int lock_request(struct fuse_req *req) 589 { 590 int err = 0; 591 if (req) { 592 spin_lock(&req->waitq.lock); 593 if (test_bit(FR_ABORTED, &req->flags)) 594 err = -ENOENT; 595 else 596 set_bit(FR_LOCKED, &req->flags); 597 spin_unlock(&req->waitq.lock); 598 } 599 return err; 600 } 601 602 /* 603 * Unlock request. If it was aborted while locked, caller is responsible 604 * for unlocking and ending the request. 605 */ 606 static int unlock_request(struct fuse_req *req) 607 { 608 int err = 0; 609 if (req) { 610 spin_lock(&req->waitq.lock); 611 if (test_bit(FR_ABORTED, &req->flags)) 612 err = -ENOENT; 613 else 614 clear_bit(FR_LOCKED, &req->flags); 615 spin_unlock(&req->waitq.lock); 616 } 617 return err; 618 } 619 620 struct fuse_copy_state { 621 int write; 622 struct fuse_req *req; 623 struct iov_iter *iter; 624 struct pipe_buffer *pipebufs; 625 struct pipe_buffer *currbuf; 626 struct pipe_inode_info *pipe; 627 unsigned long nr_segs; 628 struct page *pg; 629 unsigned len; 630 unsigned offset; 631 unsigned move_pages:1; 632 }; 633 634 static void fuse_copy_init(struct fuse_copy_state *cs, int write, 635 struct iov_iter *iter) 636 { 637 memset(cs, 0, sizeof(*cs)); 638 cs->write = write; 639 cs->iter = iter; 640 } 641 642 /* Unmap and put previous page of userspace buffer */ 643 static void fuse_copy_finish(struct fuse_copy_state *cs) 644 { 645 if (cs->currbuf) { 646 struct pipe_buffer *buf = cs->currbuf; 647 648 if (cs->write) 649 buf->len = PAGE_SIZE - cs->len; 650 cs->currbuf = NULL; 651 } else if (cs->pg) { 652 if (cs->write) { 653 flush_dcache_page(cs->pg); 654 set_page_dirty_lock(cs->pg); 655 } 656 put_page(cs->pg); 657 } 658 cs->pg = NULL; 659 } 660 661 /* 662 * Get another pagefull of userspace buffer, and map it to kernel 663 * address space, and lock request 664 */ 665 static int fuse_copy_fill(struct fuse_copy_state *cs) 666 { 667 struct page *page; 668 int err; 669 670 err = unlock_request(cs->req); 671 if (err) 672 return err; 673 674 fuse_copy_finish(cs); 675 if (cs->pipebufs) { 676 struct pipe_buffer *buf = cs->pipebufs; 677 678 if (!cs->write) { 679 err = pipe_buf_confirm(cs->pipe, buf); 680 if (err) 681 return err; 682 683 BUG_ON(!cs->nr_segs); 684 cs->currbuf = buf; 685 cs->pg = buf->page; 686 cs->offset = buf->offset; 687 cs->len = buf->len; 688 cs->pipebufs++; 689 cs->nr_segs--; 690 } else { 691 if (cs->nr_segs == cs->pipe->buffers) 692 return -EIO; 693 694 page = alloc_page(GFP_HIGHUSER); 695 if (!page) 696 return -ENOMEM; 697 698 buf->page = page; 699 buf->offset = 0; 700 buf->len = 0; 701 702 cs->currbuf = buf; 703 cs->pg = page; 704 cs->offset = 0; 705 cs->len = PAGE_SIZE; 706 cs->pipebufs++; 707 cs->nr_segs++; 708 } 709 } else { 710 size_t off; 711 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off); 712 if (err < 0) 713 return err; 714 BUG_ON(!err); 715 cs->len = err; 716 cs->offset = off; 717 cs->pg = page; 718 iov_iter_advance(cs->iter, err); 719 } 720 721 return lock_request(cs->req); 722 } 723 724 /* Do as much copy to/from userspace buffer as we can */ 725 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) 726 { 727 unsigned ncpy = min(*size, cs->len); 728 if (val) { 729 void *pgaddr = kmap_atomic(cs->pg); 730 void *buf = pgaddr + cs->offset; 731 732 if (cs->write) 733 memcpy(buf, *val, ncpy); 734 else 735 memcpy(*val, buf, ncpy); 736 737 kunmap_atomic(pgaddr); 738 *val += ncpy; 739 } 740 *size -= ncpy; 741 cs->len -= ncpy; 742 cs->offset += ncpy; 743 return ncpy; 744 } 745 746 static int fuse_check_page(struct page *page) 747 { 748 if (page_mapcount(page) || 749 page->mapping != NULL || 750 page_count(page) != 1 || 751 (page->flags & PAGE_FLAGS_CHECK_AT_PREP & 752 ~(1 << PG_locked | 753 1 << PG_referenced | 754 1 << PG_uptodate | 755 1 << PG_lru | 756 1 << PG_active | 757 1 << PG_reclaim))) { 758 pr_warn("trying to steal weird page\n"); 759 pr_warn(" page=%p index=%li flags=%08lx, count=%i, mapcount=%i, mapping=%p\n", page, page->index, page->flags, page_count(page), page_mapcount(page), page->mapping); 760 return 1; 761 } 762 return 0; 763 } 764 765 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep) 766 { 767 int err; 768 struct page *oldpage = *pagep; 769 struct page *newpage; 770 struct pipe_buffer *buf = cs->pipebufs; 771 772 err = unlock_request(cs->req); 773 if (err) 774 return err; 775 776 fuse_copy_finish(cs); 777 778 err = pipe_buf_confirm(cs->pipe, buf); 779 if (err) 780 return err; 781 782 BUG_ON(!cs->nr_segs); 783 cs->currbuf = buf; 784 cs->len = buf->len; 785 cs->pipebufs++; 786 cs->nr_segs--; 787 788 if (cs->len != PAGE_SIZE) 789 goto out_fallback; 790 791 if (pipe_buf_steal(cs->pipe, buf) != 0) 792 goto out_fallback; 793 794 newpage = buf->page; 795 796 if (!PageUptodate(newpage)) 797 SetPageUptodate(newpage); 798 799 ClearPageMappedToDisk(newpage); 800 801 if (fuse_check_page(newpage) != 0) 802 goto out_fallback_unlock; 803 804 /* 805 * This is a new and locked page, it shouldn't be mapped or 806 * have any special flags on it 807 */ 808 if (WARN_ON(page_mapped(oldpage))) 809 goto out_fallback_unlock; 810 if (WARN_ON(page_has_private(oldpage))) 811 goto out_fallback_unlock; 812 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage))) 813 goto out_fallback_unlock; 814 if (WARN_ON(PageMlocked(oldpage))) 815 goto out_fallback_unlock; 816 817 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL); 818 if (err) { 819 unlock_page(newpage); 820 return err; 821 } 822 823 get_page(newpage); 824 825 if (!(buf->flags & PIPE_BUF_FLAG_LRU)) 826 lru_cache_add_file(newpage); 827 828 err = 0; 829 spin_lock(&cs->req->waitq.lock); 830 if (test_bit(FR_ABORTED, &cs->req->flags)) 831 err = -ENOENT; 832 else 833 *pagep = newpage; 834 spin_unlock(&cs->req->waitq.lock); 835 836 if (err) { 837 unlock_page(newpage); 838 put_page(newpage); 839 return err; 840 } 841 842 unlock_page(oldpage); 843 put_page(oldpage); 844 cs->len = 0; 845 846 return 0; 847 848 out_fallback_unlock: 849 unlock_page(newpage); 850 out_fallback: 851 cs->pg = buf->page; 852 cs->offset = buf->offset; 853 854 err = lock_request(cs->req); 855 if (err) 856 return err; 857 858 return 1; 859 } 860 861 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page, 862 unsigned offset, unsigned count) 863 { 864 struct pipe_buffer *buf; 865 int err; 866 867 if (cs->nr_segs == cs->pipe->buffers) 868 return -EIO; 869 870 err = unlock_request(cs->req); 871 if (err) 872 return err; 873 874 fuse_copy_finish(cs); 875 876 buf = cs->pipebufs; 877 get_page(page); 878 buf->page = page; 879 buf->offset = offset; 880 buf->len = count; 881 882 cs->pipebufs++; 883 cs->nr_segs++; 884 cs->len = 0; 885 886 return 0; 887 } 888 889 /* 890 * Copy a page in the request to/from the userspace buffer. Must be 891 * done atomically 892 */ 893 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep, 894 unsigned offset, unsigned count, int zeroing) 895 { 896 int err; 897 struct page *page = *pagep; 898 899 if (page && zeroing && count < PAGE_SIZE) 900 clear_highpage(page); 901 902 while (count) { 903 if (cs->write && cs->pipebufs && page) { 904 return fuse_ref_page(cs, page, offset, count); 905 } else if (!cs->len) { 906 if (cs->move_pages && page && 907 offset == 0 && count == PAGE_SIZE) { 908 err = fuse_try_move_page(cs, pagep); 909 if (err <= 0) 910 return err; 911 } else { 912 err = fuse_copy_fill(cs); 913 if (err) 914 return err; 915 } 916 } 917 if (page) { 918 void *mapaddr = kmap_atomic(page); 919 void *buf = mapaddr + offset; 920 offset += fuse_copy_do(cs, &buf, &count); 921 kunmap_atomic(mapaddr); 922 } else 923 offset += fuse_copy_do(cs, NULL, &count); 924 } 925 if (page && !cs->write) 926 flush_dcache_page(page); 927 return 0; 928 } 929 930 /* Copy pages in the request to/from userspace buffer */ 931 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, 932 int zeroing) 933 { 934 unsigned i; 935 struct fuse_req *req = cs->req; 936 struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args); 937 938 939 for (i = 0; i < ap->num_pages && (nbytes || zeroing); i++) { 940 int err; 941 unsigned int offset = ap->descs[i].offset; 942 unsigned int count = min(nbytes, ap->descs[i].length); 943 944 err = fuse_copy_page(cs, &ap->pages[i], offset, count, zeroing); 945 if (err) 946 return err; 947 948 nbytes -= count; 949 } 950 return 0; 951 } 952 953 /* Copy a single argument in the request to/from userspace buffer */ 954 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) 955 { 956 while (size) { 957 if (!cs->len) { 958 int err = fuse_copy_fill(cs); 959 if (err) 960 return err; 961 } 962 fuse_copy_do(cs, &val, &size); 963 } 964 return 0; 965 } 966 967 /* Copy request arguments to/from userspace buffer */ 968 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, 969 unsigned argpages, struct fuse_arg *args, 970 int zeroing) 971 { 972 int err = 0; 973 unsigned i; 974 975 for (i = 0; !err && i < numargs; i++) { 976 struct fuse_arg *arg = &args[i]; 977 if (i == numargs - 1 && argpages) 978 err = fuse_copy_pages(cs, arg->size, zeroing); 979 else 980 err = fuse_copy_one(cs, arg->value, arg->size); 981 } 982 return err; 983 } 984 985 static int forget_pending(struct fuse_iqueue *fiq) 986 { 987 return fiq->forget_list_head.next != NULL; 988 } 989 990 static int request_pending(struct fuse_iqueue *fiq) 991 { 992 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) || 993 forget_pending(fiq); 994 } 995 996 /* 997 * Transfer an interrupt request to userspace 998 * 999 * Unlike other requests this is assembled on demand, without a need 1000 * to allocate a separate fuse_req structure. 1001 * 1002 * Called with fiq->lock held, releases it 1003 */ 1004 static int fuse_read_interrupt(struct fuse_iqueue *fiq, 1005 struct fuse_copy_state *cs, 1006 size_t nbytes, struct fuse_req *req) 1007 __releases(fiq->lock) 1008 { 1009 struct fuse_in_header ih; 1010 struct fuse_interrupt_in arg; 1011 unsigned reqsize = sizeof(ih) + sizeof(arg); 1012 int err; 1013 1014 list_del_init(&req->intr_entry); 1015 memset(&ih, 0, sizeof(ih)); 1016 memset(&arg, 0, sizeof(arg)); 1017 ih.len = reqsize; 1018 ih.opcode = FUSE_INTERRUPT; 1019 ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT); 1020 arg.unique = req->in.h.unique; 1021 1022 spin_unlock(&fiq->lock); 1023 if (nbytes < reqsize) 1024 return -EINVAL; 1025 1026 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1027 if (!err) 1028 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1029 fuse_copy_finish(cs); 1030 1031 return err ? err : reqsize; 1032 } 1033 1034 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq, 1035 unsigned max, 1036 unsigned *countp) 1037 { 1038 struct fuse_forget_link *head = fiq->forget_list_head.next; 1039 struct fuse_forget_link **newhead = &head; 1040 unsigned count; 1041 1042 for (count = 0; *newhead != NULL && count < max; count++) 1043 newhead = &(*newhead)->next; 1044 1045 fiq->forget_list_head.next = *newhead; 1046 *newhead = NULL; 1047 if (fiq->forget_list_head.next == NULL) 1048 fiq->forget_list_tail = &fiq->forget_list_head; 1049 1050 if (countp != NULL) 1051 *countp = count; 1052 1053 return head; 1054 } 1055 1056 static int fuse_read_single_forget(struct fuse_iqueue *fiq, 1057 struct fuse_copy_state *cs, 1058 size_t nbytes) 1059 __releases(fiq->lock) 1060 { 1061 int err; 1062 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL); 1063 struct fuse_forget_in arg = { 1064 .nlookup = forget->forget_one.nlookup, 1065 }; 1066 struct fuse_in_header ih = { 1067 .opcode = FUSE_FORGET, 1068 .nodeid = forget->forget_one.nodeid, 1069 .unique = fuse_get_unique(fiq), 1070 .len = sizeof(ih) + sizeof(arg), 1071 }; 1072 1073 spin_unlock(&fiq->lock); 1074 kfree(forget); 1075 if (nbytes < ih.len) 1076 return -EINVAL; 1077 1078 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1079 if (!err) 1080 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1081 fuse_copy_finish(cs); 1082 1083 if (err) 1084 return err; 1085 1086 return ih.len; 1087 } 1088 1089 static int fuse_read_batch_forget(struct fuse_iqueue *fiq, 1090 struct fuse_copy_state *cs, size_t nbytes) 1091 __releases(fiq->lock) 1092 { 1093 int err; 1094 unsigned max_forgets; 1095 unsigned count; 1096 struct fuse_forget_link *head; 1097 struct fuse_batch_forget_in arg = { .count = 0 }; 1098 struct fuse_in_header ih = { 1099 .opcode = FUSE_BATCH_FORGET, 1100 .unique = fuse_get_unique(fiq), 1101 .len = sizeof(ih) + sizeof(arg), 1102 }; 1103 1104 if (nbytes < ih.len) { 1105 spin_unlock(&fiq->lock); 1106 return -EINVAL; 1107 } 1108 1109 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one); 1110 head = dequeue_forget(fiq, max_forgets, &count); 1111 spin_unlock(&fiq->lock); 1112 1113 arg.count = count; 1114 ih.len += count * sizeof(struct fuse_forget_one); 1115 err = fuse_copy_one(cs, &ih, sizeof(ih)); 1116 if (!err) 1117 err = fuse_copy_one(cs, &arg, sizeof(arg)); 1118 1119 while (head) { 1120 struct fuse_forget_link *forget = head; 1121 1122 if (!err) { 1123 err = fuse_copy_one(cs, &forget->forget_one, 1124 sizeof(forget->forget_one)); 1125 } 1126 head = forget->next; 1127 kfree(forget); 1128 } 1129 1130 fuse_copy_finish(cs); 1131 1132 if (err) 1133 return err; 1134 1135 return ih.len; 1136 } 1137 1138 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq, 1139 struct fuse_copy_state *cs, 1140 size_t nbytes) 1141 __releases(fiq->lock) 1142 { 1143 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL) 1144 return fuse_read_single_forget(fiq, cs, nbytes); 1145 else 1146 return fuse_read_batch_forget(fiq, cs, nbytes); 1147 } 1148 1149 /* 1150 * Read a single request into the userspace filesystem's buffer. This 1151 * function waits until a request is available, then removes it from 1152 * the pending list and copies request data to userspace buffer. If 1153 * no reply is needed (FORGET) or request has been aborted or there 1154 * was an error during the copying then it's finished by calling 1155 * fuse_request_end(). Otherwise add it to the processing list, and set 1156 * the 'sent' flag. 1157 */ 1158 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file, 1159 struct fuse_copy_state *cs, size_t nbytes) 1160 { 1161 ssize_t err; 1162 struct fuse_conn *fc = fud->fc; 1163 struct fuse_iqueue *fiq = &fc->iq; 1164 struct fuse_pqueue *fpq = &fud->pq; 1165 struct fuse_req *req; 1166 struct fuse_args *args; 1167 unsigned reqsize; 1168 unsigned int hash; 1169 1170 /* 1171 * Require sane minimum read buffer - that has capacity for fixed part 1172 * of any request header + negotiated max_write room for data. 1173 * 1174 * Historically libfuse reserves 4K for fixed header room, but e.g. 1175 * GlusterFS reserves only 80 bytes 1176 * 1177 * = `sizeof(fuse_in_header) + sizeof(fuse_write_in)` 1178 * 1179 * which is the absolute minimum any sane filesystem should be using 1180 * for header room. 1181 */ 1182 if (nbytes < max_t(size_t, FUSE_MIN_READ_BUFFER, 1183 sizeof(struct fuse_in_header) + 1184 sizeof(struct fuse_write_in) + 1185 fc->max_write)) 1186 return -EINVAL; 1187 1188 restart: 1189 for (;;) { 1190 spin_lock(&fiq->lock); 1191 if (!fiq->connected || request_pending(fiq)) 1192 break; 1193 spin_unlock(&fiq->lock); 1194 1195 if (file->f_flags & O_NONBLOCK) 1196 return -EAGAIN; 1197 err = wait_event_interruptible_exclusive(fiq->waitq, 1198 !fiq->connected || request_pending(fiq)); 1199 if (err) 1200 return err; 1201 } 1202 1203 if (!fiq->connected) { 1204 err = fc->aborted ? -ECONNABORTED : -ENODEV; 1205 goto err_unlock; 1206 } 1207 1208 if (!list_empty(&fiq->interrupts)) { 1209 req = list_entry(fiq->interrupts.next, struct fuse_req, 1210 intr_entry); 1211 return fuse_read_interrupt(fiq, cs, nbytes, req); 1212 } 1213 1214 if (forget_pending(fiq)) { 1215 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0) 1216 return fuse_read_forget(fc, fiq, cs, nbytes); 1217 1218 if (fiq->forget_batch <= -8) 1219 fiq->forget_batch = 16; 1220 } 1221 1222 req = list_entry(fiq->pending.next, struct fuse_req, list); 1223 clear_bit(FR_PENDING, &req->flags); 1224 list_del_init(&req->list); 1225 spin_unlock(&fiq->lock); 1226 1227 args = req->args; 1228 reqsize = req->in.h.len; 1229 1230 /* If request is too large, reply with an error and restart the read */ 1231 if (nbytes < reqsize) { 1232 req->out.h.error = -EIO; 1233 /* SETXATTR is special, since it may contain too large data */ 1234 if (args->opcode == FUSE_SETXATTR) 1235 req->out.h.error = -E2BIG; 1236 fuse_request_end(fc, req); 1237 goto restart; 1238 } 1239 spin_lock(&fpq->lock); 1240 list_add(&req->list, &fpq->io); 1241 spin_unlock(&fpq->lock); 1242 cs->req = req; 1243 err = fuse_copy_one(cs, &req->in.h, sizeof(req->in.h)); 1244 if (!err) 1245 err = fuse_copy_args(cs, args->in_numargs, args->in_pages, 1246 (struct fuse_arg *) args->in_args, 0); 1247 fuse_copy_finish(cs); 1248 spin_lock(&fpq->lock); 1249 clear_bit(FR_LOCKED, &req->flags); 1250 if (!fpq->connected) { 1251 err = fc->aborted ? -ECONNABORTED : -ENODEV; 1252 goto out_end; 1253 } 1254 if (err) { 1255 req->out.h.error = -EIO; 1256 goto out_end; 1257 } 1258 if (!test_bit(FR_ISREPLY, &req->flags)) { 1259 err = reqsize; 1260 goto out_end; 1261 } 1262 hash = fuse_req_hash(req->in.h.unique); 1263 list_move_tail(&req->list, &fpq->processing[hash]); 1264 __fuse_get_request(req); 1265 set_bit(FR_SENT, &req->flags); 1266 spin_unlock(&fpq->lock); 1267 /* matches barrier in request_wait_answer() */ 1268 smp_mb__after_atomic(); 1269 if (test_bit(FR_INTERRUPTED, &req->flags)) 1270 queue_interrupt(fiq, req); 1271 fuse_put_request(fc, req); 1272 1273 return reqsize; 1274 1275 out_end: 1276 if (!test_bit(FR_PRIVATE, &req->flags)) 1277 list_del_init(&req->list); 1278 spin_unlock(&fpq->lock); 1279 fuse_request_end(fc, req); 1280 return err; 1281 1282 err_unlock: 1283 spin_unlock(&fiq->lock); 1284 return err; 1285 } 1286 1287 static int fuse_dev_open(struct inode *inode, struct file *file) 1288 { 1289 /* 1290 * The fuse device's file's private_data is used to hold 1291 * the fuse_conn(ection) when it is mounted, and is used to 1292 * keep track of whether the file has been mounted already. 1293 */ 1294 file->private_data = NULL; 1295 return 0; 1296 } 1297 1298 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to) 1299 { 1300 struct fuse_copy_state cs; 1301 struct file *file = iocb->ki_filp; 1302 struct fuse_dev *fud = fuse_get_dev(file); 1303 1304 if (!fud) 1305 return -EPERM; 1306 1307 if (!iter_is_iovec(to)) 1308 return -EINVAL; 1309 1310 fuse_copy_init(&cs, 1, to); 1311 1312 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to)); 1313 } 1314 1315 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos, 1316 struct pipe_inode_info *pipe, 1317 size_t len, unsigned int flags) 1318 { 1319 int total, ret; 1320 int page_nr = 0; 1321 struct pipe_buffer *bufs; 1322 struct fuse_copy_state cs; 1323 struct fuse_dev *fud = fuse_get_dev(in); 1324 1325 if (!fud) 1326 return -EPERM; 1327 1328 bufs = kvmalloc_array(pipe->buffers, sizeof(struct pipe_buffer), 1329 GFP_KERNEL); 1330 if (!bufs) 1331 return -ENOMEM; 1332 1333 fuse_copy_init(&cs, 1, NULL); 1334 cs.pipebufs = bufs; 1335 cs.pipe = pipe; 1336 ret = fuse_dev_do_read(fud, in, &cs, len); 1337 if (ret < 0) 1338 goto out; 1339 1340 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) { 1341 ret = -EIO; 1342 goto out; 1343 } 1344 1345 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) { 1346 /* 1347 * Need to be careful about this. Having buf->ops in module 1348 * code can Oops if the buffer persists after module unload. 1349 */ 1350 bufs[page_nr].ops = &nosteal_pipe_buf_ops; 1351 bufs[page_nr].flags = 0; 1352 ret = add_to_pipe(pipe, &bufs[page_nr++]); 1353 if (unlikely(ret < 0)) 1354 break; 1355 } 1356 if (total) 1357 ret = total; 1358 out: 1359 for (; page_nr < cs.nr_segs; page_nr++) 1360 put_page(bufs[page_nr].page); 1361 1362 kvfree(bufs); 1363 return ret; 1364 } 1365 1366 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size, 1367 struct fuse_copy_state *cs) 1368 { 1369 struct fuse_notify_poll_wakeup_out outarg; 1370 int err = -EINVAL; 1371 1372 if (size != sizeof(outarg)) 1373 goto err; 1374 1375 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1376 if (err) 1377 goto err; 1378 1379 fuse_copy_finish(cs); 1380 return fuse_notify_poll_wakeup(fc, &outarg); 1381 1382 err: 1383 fuse_copy_finish(cs); 1384 return err; 1385 } 1386 1387 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size, 1388 struct fuse_copy_state *cs) 1389 { 1390 struct fuse_notify_inval_inode_out outarg; 1391 int err = -EINVAL; 1392 1393 if (size != sizeof(outarg)) 1394 goto err; 1395 1396 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1397 if (err) 1398 goto err; 1399 fuse_copy_finish(cs); 1400 1401 down_read(&fc->killsb); 1402 err = -ENOENT; 1403 if (fc->sb) { 1404 err = fuse_reverse_inval_inode(fc->sb, outarg.ino, 1405 outarg.off, outarg.len); 1406 } 1407 up_read(&fc->killsb); 1408 return err; 1409 1410 err: 1411 fuse_copy_finish(cs); 1412 return err; 1413 } 1414 1415 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size, 1416 struct fuse_copy_state *cs) 1417 { 1418 struct fuse_notify_inval_entry_out outarg; 1419 int err = -ENOMEM; 1420 char *buf; 1421 struct qstr name; 1422 1423 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); 1424 if (!buf) 1425 goto err; 1426 1427 err = -EINVAL; 1428 if (size < sizeof(outarg)) 1429 goto err; 1430 1431 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1432 if (err) 1433 goto err; 1434 1435 err = -ENAMETOOLONG; 1436 if (outarg.namelen > FUSE_NAME_MAX) 1437 goto err; 1438 1439 err = -EINVAL; 1440 if (size != sizeof(outarg) + outarg.namelen + 1) 1441 goto err; 1442 1443 name.name = buf; 1444 name.len = outarg.namelen; 1445 err = fuse_copy_one(cs, buf, outarg.namelen + 1); 1446 if (err) 1447 goto err; 1448 fuse_copy_finish(cs); 1449 buf[outarg.namelen] = 0; 1450 1451 down_read(&fc->killsb); 1452 err = -ENOENT; 1453 if (fc->sb) 1454 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name); 1455 up_read(&fc->killsb); 1456 kfree(buf); 1457 return err; 1458 1459 err: 1460 kfree(buf); 1461 fuse_copy_finish(cs); 1462 return err; 1463 } 1464 1465 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size, 1466 struct fuse_copy_state *cs) 1467 { 1468 struct fuse_notify_delete_out outarg; 1469 int err = -ENOMEM; 1470 char *buf; 1471 struct qstr name; 1472 1473 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); 1474 if (!buf) 1475 goto err; 1476 1477 err = -EINVAL; 1478 if (size < sizeof(outarg)) 1479 goto err; 1480 1481 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1482 if (err) 1483 goto err; 1484 1485 err = -ENAMETOOLONG; 1486 if (outarg.namelen > FUSE_NAME_MAX) 1487 goto err; 1488 1489 err = -EINVAL; 1490 if (size != sizeof(outarg) + outarg.namelen + 1) 1491 goto err; 1492 1493 name.name = buf; 1494 name.len = outarg.namelen; 1495 err = fuse_copy_one(cs, buf, outarg.namelen + 1); 1496 if (err) 1497 goto err; 1498 fuse_copy_finish(cs); 1499 buf[outarg.namelen] = 0; 1500 1501 down_read(&fc->killsb); 1502 err = -ENOENT; 1503 if (fc->sb) 1504 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 1505 outarg.child, &name); 1506 up_read(&fc->killsb); 1507 kfree(buf); 1508 return err; 1509 1510 err: 1511 kfree(buf); 1512 fuse_copy_finish(cs); 1513 return err; 1514 } 1515 1516 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size, 1517 struct fuse_copy_state *cs) 1518 { 1519 struct fuse_notify_store_out outarg; 1520 struct inode *inode; 1521 struct address_space *mapping; 1522 u64 nodeid; 1523 int err; 1524 pgoff_t index; 1525 unsigned int offset; 1526 unsigned int num; 1527 loff_t file_size; 1528 loff_t end; 1529 1530 err = -EINVAL; 1531 if (size < sizeof(outarg)) 1532 goto out_finish; 1533 1534 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1535 if (err) 1536 goto out_finish; 1537 1538 err = -EINVAL; 1539 if (size - sizeof(outarg) != outarg.size) 1540 goto out_finish; 1541 1542 nodeid = outarg.nodeid; 1543 1544 down_read(&fc->killsb); 1545 1546 err = -ENOENT; 1547 if (!fc->sb) 1548 goto out_up_killsb; 1549 1550 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid); 1551 if (!inode) 1552 goto out_up_killsb; 1553 1554 mapping = inode->i_mapping; 1555 index = outarg.offset >> PAGE_SHIFT; 1556 offset = outarg.offset & ~PAGE_MASK; 1557 file_size = i_size_read(inode); 1558 end = outarg.offset + outarg.size; 1559 if (end > file_size) { 1560 file_size = end; 1561 fuse_write_update_size(inode, file_size); 1562 } 1563 1564 num = outarg.size; 1565 while (num) { 1566 struct page *page; 1567 unsigned int this_num; 1568 1569 err = -ENOMEM; 1570 page = find_or_create_page(mapping, index, 1571 mapping_gfp_mask(mapping)); 1572 if (!page) 1573 goto out_iput; 1574 1575 this_num = min_t(unsigned, num, PAGE_SIZE - offset); 1576 err = fuse_copy_page(cs, &page, offset, this_num, 0); 1577 if (!err && offset == 0 && 1578 (this_num == PAGE_SIZE || file_size == end)) 1579 SetPageUptodate(page); 1580 unlock_page(page); 1581 put_page(page); 1582 1583 if (err) 1584 goto out_iput; 1585 1586 num -= this_num; 1587 offset = 0; 1588 index++; 1589 } 1590 1591 err = 0; 1592 1593 out_iput: 1594 iput(inode); 1595 out_up_killsb: 1596 up_read(&fc->killsb); 1597 out_finish: 1598 fuse_copy_finish(cs); 1599 return err; 1600 } 1601 1602 struct fuse_retrieve_args { 1603 struct fuse_args_pages ap; 1604 struct fuse_notify_retrieve_in inarg; 1605 }; 1606 1607 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_args *args, 1608 int error) 1609 { 1610 struct fuse_retrieve_args *ra = 1611 container_of(args, typeof(*ra), ap.args); 1612 1613 release_pages(ra->ap.pages, ra->ap.num_pages); 1614 kfree(ra); 1615 } 1616 1617 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode, 1618 struct fuse_notify_retrieve_out *outarg) 1619 { 1620 int err; 1621 struct address_space *mapping = inode->i_mapping; 1622 pgoff_t index; 1623 loff_t file_size; 1624 unsigned int num; 1625 unsigned int offset; 1626 size_t total_len = 0; 1627 unsigned int num_pages; 1628 struct fuse_retrieve_args *ra; 1629 size_t args_size = sizeof(*ra); 1630 struct fuse_args_pages *ap; 1631 struct fuse_args *args; 1632 1633 offset = outarg->offset & ~PAGE_MASK; 1634 file_size = i_size_read(inode); 1635 1636 num = min(outarg->size, fc->max_write); 1637 if (outarg->offset > file_size) 1638 num = 0; 1639 else if (outarg->offset + num > file_size) 1640 num = file_size - outarg->offset; 1641 1642 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; 1643 num_pages = min(num_pages, fc->max_pages); 1644 1645 args_size += num_pages * (sizeof(ap->pages[0]) + sizeof(ap->descs[0])); 1646 1647 ra = kzalloc(args_size, GFP_KERNEL); 1648 if (!ra) 1649 return -ENOMEM; 1650 1651 ap = &ra->ap; 1652 ap->pages = (void *) (ra + 1); 1653 ap->descs = (void *) (ap->pages + num_pages); 1654 1655 args = &ap->args; 1656 args->nodeid = outarg->nodeid; 1657 args->opcode = FUSE_NOTIFY_REPLY; 1658 args->in_numargs = 2; 1659 args->in_pages = true; 1660 args->end = fuse_retrieve_end; 1661 1662 index = outarg->offset >> PAGE_SHIFT; 1663 1664 while (num && ap->num_pages < num_pages) { 1665 struct page *page; 1666 unsigned int this_num; 1667 1668 page = find_get_page(mapping, index); 1669 if (!page) 1670 break; 1671 1672 this_num = min_t(unsigned, num, PAGE_SIZE - offset); 1673 ap->pages[ap->num_pages] = page; 1674 ap->descs[ap->num_pages].offset = offset; 1675 ap->descs[ap->num_pages].length = this_num; 1676 ap->num_pages++; 1677 1678 offset = 0; 1679 num -= this_num; 1680 total_len += this_num; 1681 index++; 1682 } 1683 ra->inarg.offset = outarg->offset; 1684 ra->inarg.size = total_len; 1685 args->in_args[0].size = sizeof(ra->inarg); 1686 args->in_args[0].value = &ra->inarg; 1687 args->in_args[1].size = total_len; 1688 1689 err = fuse_simple_notify_reply(fc, args, outarg->notify_unique); 1690 if (err) 1691 fuse_retrieve_end(fc, args, err); 1692 1693 return err; 1694 } 1695 1696 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size, 1697 struct fuse_copy_state *cs) 1698 { 1699 struct fuse_notify_retrieve_out outarg; 1700 struct inode *inode; 1701 int err; 1702 1703 err = -EINVAL; 1704 if (size != sizeof(outarg)) 1705 goto copy_finish; 1706 1707 err = fuse_copy_one(cs, &outarg, sizeof(outarg)); 1708 if (err) 1709 goto copy_finish; 1710 1711 fuse_copy_finish(cs); 1712 1713 down_read(&fc->killsb); 1714 err = -ENOENT; 1715 if (fc->sb) { 1716 u64 nodeid = outarg.nodeid; 1717 1718 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid); 1719 if (inode) { 1720 err = fuse_retrieve(fc, inode, &outarg); 1721 iput(inode); 1722 } 1723 } 1724 up_read(&fc->killsb); 1725 1726 return err; 1727 1728 copy_finish: 1729 fuse_copy_finish(cs); 1730 return err; 1731 } 1732 1733 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code, 1734 unsigned int size, struct fuse_copy_state *cs) 1735 { 1736 /* Don't try to move pages (yet) */ 1737 cs->move_pages = 0; 1738 1739 switch (code) { 1740 case FUSE_NOTIFY_POLL: 1741 return fuse_notify_poll(fc, size, cs); 1742 1743 case FUSE_NOTIFY_INVAL_INODE: 1744 return fuse_notify_inval_inode(fc, size, cs); 1745 1746 case FUSE_NOTIFY_INVAL_ENTRY: 1747 return fuse_notify_inval_entry(fc, size, cs); 1748 1749 case FUSE_NOTIFY_STORE: 1750 return fuse_notify_store(fc, size, cs); 1751 1752 case FUSE_NOTIFY_RETRIEVE: 1753 return fuse_notify_retrieve(fc, size, cs); 1754 1755 case FUSE_NOTIFY_DELETE: 1756 return fuse_notify_delete(fc, size, cs); 1757 1758 default: 1759 fuse_copy_finish(cs); 1760 return -EINVAL; 1761 } 1762 } 1763 1764 /* Look up request on processing list by unique ID */ 1765 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique) 1766 { 1767 unsigned int hash = fuse_req_hash(unique); 1768 struct fuse_req *req; 1769 1770 list_for_each_entry(req, &fpq->processing[hash], list) { 1771 if (req->in.h.unique == unique) 1772 return req; 1773 } 1774 return NULL; 1775 } 1776 1777 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args, 1778 unsigned nbytes) 1779 { 1780 unsigned reqsize = sizeof(struct fuse_out_header); 1781 1782 reqsize += len_args(args->out_numargs, args->out_args); 1783 1784 if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar)) 1785 return -EINVAL; 1786 else if (reqsize > nbytes) { 1787 struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1]; 1788 unsigned diffsize = reqsize - nbytes; 1789 1790 if (diffsize > lastarg->size) 1791 return -EINVAL; 1792 lastarg->size -= diffsize; 1793 } 1794 return fuse_copy_args(cs, args->out_numargs, args->out_pages, 1795 args->out_args, args->page_zeroing); 1796 } 1797 1798 /* 1799 * Write a single reply to a request. First the header is copied from 1800 * the write buffer. The request is then searched on the processing 1801 * list by the unique ID found in the header. If found, then remove 1802 * it from the list and copy the rest of the buffer to the request. 1803 * The request is finished by calling fuse_request_end(). 1804 */ 1805 static ssize_t fuse_dev_do_write(struct fuse_dev *fud, 1806 struct fuse_copy_state *cs, size_t nbytes) 1807 { 1808 int err; 1809 struct fuse_conn *fc = fud->fc; 1810 struct fuse_pqueue *fpq = &fud->pq; 1811 struct fuse_req *req; 1812 struct fuse_out_header oh; 1813 1814 err = -EINVAL; 1815 if (nbytes < sizeof(struct fuse_out_header)) 1816 goto out; 1817 1818 err = fuse_copy_one(cs, &oh, sizeof(oh)); 1819 if (err) 1820 goto copy_finish; 1821 1822 err = -EINVAL; 1823 if (oh.len != nbytes) 1824 goto copy_finish; 1825 1826 /* 1827 * Zero oh.unique indicates unsolicited notification message 1828 * and error contains notification code. 1829 */ 1830 if (!oh.unique) { 1831 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs); 1832 goto out; 1833 } 1834 1835 err = -EINVAL; 1836 if (oh.error <= -1000 || oh.error > 0) 1837 goto copy_finish; 1838 1839 spin_lock(&fpq->lock); 1840 req = NULL; 1841 if (fpq->connected) 1842 req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT); 1843 1844 err = -ENOENT; 1845 if (!req) { 1846 spin_unlock(&fpq->lock); 1847 goto copy_finish; 1848 } 1849 1850 /* Is it an interrupt reply ID? */ 1851 if (oh.unique & FUSE_INT_REQ_BIT) { 1852 __fuse_get_request(req); 1853 spin_unlock(&fpq->lock); 1854 1855 err = 0; 1856 if (nbytes != sizeof(struct fuse_out_header)) 1857 err = -EINVAL; 1858 else if (oh.error == -ENOSYS) 1859 fc->no_interrupt = 1; 1860 else if (oh.error == -EAGAIN) 1861 err = queue_interrupt(&fc->iq, req); 1862 1863 fuse_put_request(fc, req); 1864 1865 goto copy_finish; 1866 } 1867 1868 clear_bit(FR_SENT, &req->flags); 1869 list_move(&req->list, &fpq->io); 1870 req->out.h = oh; 1871 set_bit(FR_LOCKED, &req->flags); 1872 spin_unlock(&fpq->lock); 1873 cs->req = req; 1874 if (!req->args->page_replace) 1875 cs->move_pages = 0; 1876 1877 if (oh.error) 1878 err = nbytes != sizeof(oh) ? -EINVAL : 0; 1879 else 1880 err = copy_out_args(cs, req->args, nbytes); 1881 fuse_copy_finish(cs); 1882 1883 spin_lock(&fpq->lock); 1884 clear_bit(FR_LOCKED, &req->flags); 1885 if (!fpq->connected) 1886 err = -ENOENT; 1887 else if (err) 1888 req->out.h.error = -EIO; 1889 if (!test_bit(FR_PRIVATE, &req->flags)) 1890 list_del_init(&req->list); 1891 spin_unlock(&fpq->lock); 1892 1893 fuse_request_end(fc, req); 1894 out: 1895 return err ? err : nbytes; 1896 1897 copy_finish: 1898 fuse_copy_finish(cs); 1899 goto out; 1900 } 1901 1902 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from) 1903 { 1904 struct fuse_copy_state cs; 1905 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp); 1906 1907 if (!fud) 1908 return -EPERM; 1909 1910 if (!iter_is_iovec(from)) 1911 return -EINVAL; 1912 1913 fuse_copy_init(&cs, 0, from); 1914 1915 return fuse_dev_do_write(fud, &cs, iov_iter_count(from)); 1916 } 1917 1918 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe, 1919 struct file *out, loff_t *ppos, 1920 size_t len, unsigned int flags) 1921 { 1922 unsigned nbuf; 1923 unsigned idx; 1924 struct pipe_buffer *bufs; 1925 struct fuse_copy_state cs; 1926 struct fuse_dev *fud; 1927 size_t rem; 1928 ssize_t ret; 1929 1930 fud = fuse_get_dev(out); 1931 if (!fud) 1932 return -EPERM; 1933 1934 pipe_lock(pipe); 1935 1936 bufs = kvmalloc_array(pipe->nrbufs, sizeof(struct pipe_buffer), 1937 GFP_KERNEL); 1938 if (!bufs) { 1939 pipe_unlock(pipe); 1940 return -ENOMEM; 1941 } 1942 1943 nbuf = 0; 1944 rem = 0; 1945 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++) 1946 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len; 1947 1948 ret = -EINVAL; 1949 if (rem < len) 1950 goto out_free; 1951 1952 rem = len; 1953 while (rem) { 1954 struct pipe_buffer *ibuf; 1955 struct pipe_buffer *obuf; 1956 1957 BUG_ON(nbuf >= pipe->buffers); 1958 BUG_ON(!pipe->nrbufs); 1959 ibuf = &pipe->bufs[pipe->curbuf]; 1960 obuf = &bufs[nbuf]; 1961 1962 if (rem >= ibuf->len) { 1963 *obuf = *ibuf; 1964 ibuf->ops = NULL; 1965 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); 1966 pipe->nrbufs--; 1967 } else { 1968 if (!pipe_buf_get(pipe, ibuf)) 1969 goto out_free; 1970 1971 *obuf = *ibuf; 1972 obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 1973 obuf->len = rem; 1974 ibuf->offset += obuf->len; 1975 ibuf->len -= obuf->len; 1976 } 1977 nbuf++; 1978 rem -= obuf->len; 1979 } 1980 pipe_unlock(pipe); 1981 1982 fuse_copy_init(&cs, 0, NULL); 1983 cs.pipebufs = bufs; 1984 cs.nr_segs = nbuf; 1985 cs.pipe = pipe; 1986 1987 if (flags & SPLICE_F_MOVE) 1988 cs.move_pages = 1; 1989 1990 ret = fuse_dev_do_write(fud, &cs, len); 1991 1992 pipe_lock(pipe); 1993 out_free: 1994 for (idx = 0; idx < nbuf; idx++) 1995 pipe_buf_release(pipe, &bufs[idx]); 1996 pipe_unlock(pipe); 1997 1998 kvfree(bufs); 1999 return ret; 2000 } 2001 2002 static __poll_t fuse_dev_poll(struct file *file, poll_table *wait) 2003 { 2004 __poll_t mask = EPOLLOUT | EPOLLWRNORM; 2005 struct fuse_iqueue *fiq; 2006 struct fuse_dev *fud = fuse_get_dev(file); 2007 2008 if (!fud) 2009 return EPOLLERR; 2010 2011 fiq = &fud->fc->iq; 2012 poll_wait(file, &fiq->waitq, wait); 2013 2014 spin_lock(&fiq->lock); 2015 if (!fiq->connected) 2016 mask = EPOLLERR; 2017 else if (request_pending(fiq)) 2018 mask |= EPOLLIN | EPOLLRDNORM; 2019 spin_unlock(&fiq->lock); 2020 2021 return mask; 2022 } 2023 2024 /* Abort all requests on the given list (pending or processing) */ 2025 static void end_requests(struct fuse_conn *fc, struct list_head *head) 2026 { 2027 while (!list_empty(head)) { 2028 struct fuse_req *req; 2029 req = list_entry(head->next, struct fuse_req, list); 2030 req->out.h.error = -ECONNABORTED; 2031 clear_bit(FR_SENT, &req->flags); 2032 list_del_init(&req->list); 2033 fuse_request_end(fc, req); 2034 } 2035 } 2036 2037 static void end_polls(struct fuse_conn *fc) 2038 { 2039 struct rb_node *p; 2040 2041 p = rb_first(&fc->polled_files); 2042 2043 while (p) { 2044 struct fuse_file *ff; 2045 ff = rb_entry(p, struct fuse_file, polled_node); 2046 wake_up_interruptible_all(&ff->poll_wait); 2047 2048 p = rb_next(p); 2049 } 2050 } 2051 2052 /* 2053 * Abort all requests. 2054 * 2055 * Emergency exit in case of a malicious or accidental deadlock, or just a hung 2056 * filesystem. 2057 * 2058 * The same effect is usually achievable through killing the filesystem daemon 2059 * and all users of the filesystem. The exception is the combination of an 2060 * asynchronous request and the tricky deadlock (see 2061 * Documentation/filesystems/fuse.txt). 2062 * 2063 * Aborting requests under I/O goes as follows: 1: Separate out unlocked 2064 * requests, they should be finished off immediately. Locked requests will be 2065 * finished after unlock; see unlock_request(). 2: Finish off the unlocked 2066 * requests. It is possible that some request will finish before we can. This 2067 * is OK, the request will in that case be removed from the list before we touch 2068 * it. 2069 */ 2070 void fuse_abort_conn(struct fuse_conn *fc) 2071 { 2072 struct fuse_iqueue *fiq = &fc->iq; 2073 2074 spin_lock(&fc->lock); 2075 if (fc->connected) { 2076 struct fuse_dev *fud; 2077 struct fuse_req *req, *next; 2078 LIST_HEAD(to_end); 2079 unsigned int i; 2080 2081 /* Background queuing checks fc->connected under bg_lock */ 2082 spin_lock(&fc->bg_lock); 2083 fc->connected = 0; 2084 spin_unlock(&fc->bg_lock); 2085 2086 fuse_set_initialized(fc); 2087 list_for_each_entry(fud, &fc->devices, entry) { 2088 struct fuse_pqueue *fpq = &fud->pq; 2089 2090 spin_lock(&fpq->lock); 2091 fpq->connected = 0; 2092 list_for_each_entry_safe(req, next, &fpq->io, list) { 2093 req->out.h.error = -ECONNABORTED; 2094 spin_lock(&req->waitq.lock); 2095 set_bit(FR_ABORTED, &req->flags); 2096 if (!test_bit(FR_LOCKED, &req->flags)) { 2097 set_bit(FR_PRIVATE, &req->flags); 2098 __fuse_get_request(req); 2099 list_move(&req->list, &to_end); 2100 } 2101 spin_unlock(&req->waitq.lock); 2102 } 2103 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 2104 list_splice_tail_init(&fpq->processing[i], 2105 &to_end); 2106 spin_unlock(&fpq->lock); 2107 } 2108 spin_lock(&fc->bg_lock); 2109 fc->blocked = 0; 2110 fc->max_background = UINT_MAX; 2111 flush_bg_queue(fc); 2112 spin_unlock(&fc->bg_lock); 2113 2114 spin_lock(&fiq->lock); 2115 fiq->connected = 0; 2116 list_for_each_entry(req, &fiq->pending, list) 2117 clear_bit(FR_PENDING, &req->flags); 2118 list_splice_tail_init(&fiq->pending, &to_end); 2119 while (forget_pending(fiq)) 2120 kfree(dequeue_forget(fiq, 1, NULL)); 2121 wake_up_all(&fiq->waitq); 2122 spin_unlock(&fiq->lock); 2123 kill_fasync(&fiq->fasync, SIGIO, POLL_IN); 2124 end_polls(fc); 2125 wake_up_all(&fc->blocked_waitq); 2126 spin_unlock(&fc->lock); 2127 2128 end_requests(fc, &to_end); 2129 } else { 2130 spin_unlock(&fc->lock); 2131 } 2132 } 2133 EXPORT_SYMBOL_GPL(fuse_abort_conn); 2134 2135 void fuse_wait_aborted(struct fuse_conn *fc) 2136 { 2137 /* matches implicit memory barrier in fuse_drop_waiting() */ 2138 smp_mb(); 2139 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0); 2140 } 2141 2142 int fuse_dev_release(struct inode *inode, struct file *file) 2143 { 2144 struct fuse_dev *fud = fuse_get_dev(file); 2145 2146 if (fud) { 2147 struct fuse_conn *fc = fud->fc; 2148 struct fuse_pqueue *fpq = &fud->pq; 2149 LIST_HEAD(to_end); 2150 unsigned int i; 2151 2152 spin_lock(&fpq->lock); 2153 WARN_ON(!list_empty(&fpq->io)); 2154 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 2155 list_splice_init(&fpq->processing[i], &to_end); 2156 spin_unlock(&fpq->lock); 2157 2158 end_requests(fc, &to_end); 2159 2160 /* Are we the last open device? */ 2161 if (atomic_dec_and_test(&fc->dev_count)) { 2162 WARN_ON(fc->iq.fasync != NULL); 2163 fuse_abort_conn(fc); 2164 } 2165 fuse_dev_free(fud); 2166 } 2167 return 0; 2168 } 2169 EXPORT_SYMBOL_GPL(fuse_dev_release); 2170 2171 static int fuse_dev_fasync(int fd, struct file *file, int on) 2172 { 2173 struct fuse_dev *fud = fuse_get_dev(file); 2174 2175 if (!fud) 2176 return -EPERM; 2177 2178 /* No locking - fasync_helper does its own locking */ 2179 return fasync_helper(fd, file, on, &fud->fc->iq.fasync); 2180 } 2181 2182 static int fuse_device_clone(struct fuse_conn *fc, struct file *new) 2183 { 2184 struct fuse_dev *fud; 2185 2186 if (new->private_data) 2187 return -EINVAL; 2188 2189 fud = fuse_dev_alloc(fc); 2190 if (!fud) 2191 return -ENOMEM; 2192 2193 new->private_data = fud; 2194 atomic_inc(&fc->dev_count); 2195 2196 return 0; 2197 } 2198 2199 static long fuse_dev_ioctl(struct file *file, unsigned int cmd, 2200 unsigned long arg) 2201 { 2202 int err = -ENOTTY; 2203 2204 if (cmd == FUSE_DEV_IOC_CLONE) { 2205 int oldfd; 2206 2207 err = -EFAULT; 2208 if (!get_user(oldfd, (__u32 __user *) arg)) { 2209 struct file *old = fget(oldfd); 2210 2211 err = -EINVAL; 2212 if (old) { 2213 struct fuse_dev *fud = NULL; 2214 2215 /* 2216 * Check against file->f_op because CUSE 2217 * uses the same ioctl handler. 2218 */ 2219 if (old->f_op == file->f_op && 2220 old->f_cred->user_ns == file->f_cred->user_ns) 2221 fud = fuse_get_dev(old); 2222 2223 if (fud) { 2224 mutex_lock(&fuse_mutex); 2225 err = fuse_device_clone(fud->fc, file); 2226 mutex_unlock(&fuse_mutex); 2227 } 2228 fput(old); 2229 } 2230 } 2231 } 2232 return err; 2233 } 2234 2235 const struct file_operations fuse_dev_operations = { 2236 .owner = THIS_MODULE, 2237 .open = fuse_dev_open, 2238 .llseek = no_llseek, 2239 .read_iter = fuse_dev_read, 2240 .splice_read = fuse_dev_splice_read, 2241 .write_iter = fuse_dev_write, 2242 .splice_write = fuse_dev_splice_write, 2243 .poll = fuse_dev_poll, 2244 .release = fuse_dev_release, 2245 .fasync = fuse_dev_fasync, 2246 .unlocked_ioctl = fuse_dev_ioctl, 2247 .compat_ioctl = fuse_dev_ioctl, 2248 }; 2249 EXPORT_SYMBOL_GPL(fuse_dev_operations); 2250 2251 static struct miscdevice fuse_miscdevice = { 2252 .minor = FUSE_MINOR, 2253 .name = "fuse", 2254 .fops = &fuse_dev_operations, 2255 }; 2256 2257 int __init fuse_dev_init(void) 2258 { 2259 int err = -ENOMEM; 2260 fuse_req_cachep = kmem_cache_create("fuse_request", 2261 sizeof(struct fuse_req), 2262 0, 0, NULL); 2263 if (!fuse_req_cachep) 2264 goto out; 2265 2266 err = misc_register(&fuse_miscdevice); 2267 if (err) 2268 goto out_cache_clean; 2269 2270 return 0; 2271 2272 out_cache_clean: 2273 kmem_cache_destroy(fuse_req_cachep); 2274 out: 2275 return err; 2276 } 2277 2278 void fuse_dev_cleanup(void) 2279 { 2280 misc_deregister(&fuse_miscdevice); 2281 kmem_cache_destroy(fuse_req_cachep); 2282 } 2283