1 /* 2 * linux/drivers/block/loop.c 3 * 4 * Written by Theodore Ts'o, 3/29/93 5 * 6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is 7 * permitted under the GNU General Public License. 8 * 9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993 10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996 11 * 12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994 13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996 14 * 15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997 16 * 17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998 18 * 19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998 20 * 21 * Loadable modules and other fixes by AK, 1998 22 * 23 * Make real block number available to downstream transfer functions, enables 24 * CBC (and relatives) mode encryption requiring unique IVs per data block. 25 * Reed H. Petty, rhp@draper.net 26 * 27 * Maximum number of loop devices now dynamic via max_loop module parameter. 28 * Russell Kroll <rkroll@exploits.org> 19990701 29 * 30 * Maximum number of loop devices when compiled-in now selectable by passing 31 * max_loop=<1-255> to the kernel on boot. 32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999 33 * 34 * Completely rewrite request handling to be make_request_fn style and 35 * non blocking, pushing work to a helper thread. Lots of fixes from 36 * Al Viro too. 37 * Jens Axboe <axboe@suse.de>, Nov 2000 38 * 39 * Support up to 256 loop devices 40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002 41 * 42 * Support for falling back on the write file operation when the address space 43 * operations write_begin is not available on the backing filesystem. 44 * Anton Altaparmakov, 16 Feb 2005 45 * 46 * Still To Fix: 47 * - Advisory locking is ignored here. 48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN 49 * 50 */ 51 52 #include <linux/module.h> 53 #include <linux/moduleparam.h> 54 #include <linux/sched.h> 55 #include <linux/fs.h> 56 #include <linux/pagemap.h> 57 #include <linux/file.h> 58 #include <linux/stat.h> 59 #include <linux/errno.h> 60 #include <linux/major.h> 61 #include <linux/wait.h> 62 #include <linux/blkdev.h> 63 #include <linux/blkpg.h> 64 #include <linux/init.h> 65 #include <linux/swap.h> 66 #include <linux/slab.h> 67 #include <linux/compat.h> 68 #include <linux/suspend.h> 69 #include <linux/freezer.h> 70 #include <linux/mutex.h> 71 #include <linux/writeback.h> 72 #include <linux/completion.h> 73 #include <linux/highmem.h> 74 #include <linux/splice.h> 75 #include <linux/sysfs.h> 76 #include <linux/miscdevice.h> 77 #include <linux/falloc.h> 78 #include <linux/uio.h> 79 #include <linux/ioprio.h> 80 #include <linux/blk-cgroup.h> 81 #include <linux/sched/mm.h> 82 83 #include "loop.h" 84 85 #include <linux/uaccess.h> 86 87 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) 88 89 static DEFINE_IDR(loop_index_idr); 90 static DEFINE_MUTEX(loop_ctl_mutex); 91 static DEFINE_MUTEX(loop_validate_mutex); 92 93 /** 94 * loop_global_lock_killable() - take locks for safe loop_validate_file() test 95 * 96 * @lo: struct loop_device 97 * @global: true if @lo is about to bind another "struct loop_device", false otherwise 98 * 99 * Returns 0 on success, -EINTR otherwise. 100 * 101 * Since loop_validate_file() traverses on other "struct loop_device" if 102 * is_loop_device() is true, we need a global lock for serializing concurrent 103 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls. 104 */ 105 static int loop_global_lock_killable(struct loop_device *lo, bool global) 106 { 107 int err; 108 109 if (global) { 110 err = mutex_lock_killable(&loop_validate_mutex); 111 if (err) 112 return err; 113 } 114 err = mutex_lock_killable(&lo->lo_mutex); 115 if (err && global) 116 mutex_unlock(&loop_validate_mutex); 117 return err; 118 } 119 120 /** 121 * loop_global_unlock() - release locks taken by loop_global_lock_killable() 122 * 123 * @lo: struct loop_device 124 * @global: true if @lo was about to bind another "struct loop_device", false otherwise 125 */ 126 static void loop_global_unlock(struct loop_device *lo, bool global) 127 { 128 mutex_unlock(&lo->lo_mutex); 129 if (global) 130 mutex_unlock(&loop_validate_mutex); 131 } 132 133 static int max_part; 134 static int part_shift; 135 136 static int transfer_xor(struct loop_device *lo, int cmd, 137 struct page *raw_page, unsigned raw_off, 138 struct page *loop_page, unsigned loop_off, 139 int size, sector_t real_block) 140 { 141 char *raw_buf = kmap_atomic(raw_page) + raw_off; 142 char *loop_buf = kmap_atomic(loop_page) + loop_off; 143 char *in, *out, *key; 144 int i, keysize; 145 146 if (cmd == READ) { 147 in = raw_buf; 148 out = loop_buf; 149 } else { 150 in = loop_buf; 151 out = raw_buf; 152 } 153 154 key = lo->lo_encrypt_key; 155 keysize = lo->lo_encrypt_key_size; 156 for (i = 0; i < size; i++) 157 *out++ = *in++ ^ key[(i & 511) % keysize]; 158 159 kunmap_atomic(loop_buf); 160 kunmap_atomic(raw_buf); 161 cond_resched(); 162 return 0; 163 } 164 165 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 166 { 167 if (unlikely(info->lo_encrypt_key_size <= 0)) 168 return -EINVAL; 169 return 0; 170 } 171 172 static struct loop_func_table none_funcs = { 173 .number = LO_CRYPT_NONE, 174 }; 175 176 static struct loop_func_table xor_funcs = { 177 .number = LO_CRYPT_XOR, 178 .transfer = transfer_xor, 179 .init = xor_init 180 }; 181 182 /* xfer_funcs[0] is special - its release function is never called */ 183 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 184 &none_funcs, 185 &xor_funcs 186 }; 187 188 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 189 { 190 loff_t loopsize; 191 192 /* Compute loopsize in bytes */ 193 loopsize = i_size_read(file->f_mapping->host); 194 if (offset > 0) 195 loopsize -= offset; 196 /* offset is beyond i_size, weird but possible */ 197 if (loopsize < 0) 198 return 0; 199 200 if (sizelimit > 0 && sizelimit < loopsize) 201 loopsize = sizelimit; 202 /* 203 * Unfortunately, if we want to do I/O on the device, 204 * the number of 512-byte sectors has to fit into a sector_t. 205 */ 206 return loopsize >> 9; 207 } 208 209 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 210 { 211 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 212 } 213 214 static void __loop_update_dio(struct loop_device *lo, bool dio) 215 { 216 struct file *file = lo->lo_backing_file; 217 struct address_space *mapping = file->f_mapping; 218 struct inode *inode = mapping->host; 219 unsigned short sb_bsize = 0; 220 unsigned dio_align = 0; 221 bool use_dio; 222 223 if (inode->i_sb->s_bdev) { 224 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 225 dio_align = sb_bsize - 1; 226 } 227 228 /* 229 * We support direct I/O only if lo_offset is aligned with the 230 * logical I/O size of backing device, and the logical block 231 * size of loop is bigger than the backing device's and the loop 232 * needn't transform transfer. 233 * 234 * TODO: the above condition may be loosed in the future, and 235 * direct I/O may be switched runtime at that time because most 236 * of requests in sane applications should be PAGE_SIZE aligned 237 */ 238 if (dio) { 239 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 240 !(lo->lo_offset & dio_align) && 241 mapping->a_ops->direct_IO && 242 !lo->transfer) 243 use_dio = true; 244 else 245 use_dio = false; 246 } else { 247 use_dio = false; 248 } 249 250 if (lo->use_dio == use_dio) 251 return; 252 253 /* flush dirty pages before changing direct IO */ 254 vfs_fsync(file, 0); 255 256 /* 257 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 258 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 259 * will get updated by ioctl(LOOP_GET_STATUS) 260 */ 261 if (lo->lo_state == Lo_bound) 262 blk_mq_freeze_queue(lo->lo_queue); 263 lo->use_dio = use_dio; 264 if (use_dio) { 265 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 266 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 267 } else { 268 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 269 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 270 } 271 if (lo->lo_state == Lo_bound) 272 blk_mq_unfreeze_queue(lo->lo_queue); 273 } 274 275 /** 276 * loop_validate_block_size() - validates the passed in block size 277 * @bsize: size to validate 278 */ 279 static int 280 loop_validate_block_size(unsigned short bsize) 281 { 282 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize)) 283 return -EINVAL; 284 285 return 0; 286 } 287 288 /** 289 * loop_set_size() - sets device size and notifies userspace 290 * @lo: struct loop_device to set the size for 291 * @size: new size of the loop device 292 * 293 * Callers must validate that the size passed into this function fits into 294 * a sector_t, eg using loop_validate_size() 295 */ 296 static void loop_set_size(struct loop_device *lo, loff_t size) 297 { 298 if (!set_capacity_and_notify(lo->lo_disk, size)) 299 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 300 } 301 302 static inline int 303 lo_do_transfer(struct loop_device *lo, int cmd, 304 struct page *rpage, unsigned roffs, 305 struct page *lpage, unsigned loffs, 306 int size, sector_t rblock) 307 { 308 int ret; 309 310 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 311 if (likely(!ret)) 312 return 0; 313 314 printk_ratelimited(KERN_ERR 315 "loop: Transfer error at byte offset %llu, length %i.\n", 316 (unsigned long long)rblock << 9, size); 317 return ret; 318 } 319 320 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 321 { 322 struct iov_iter i; 323 ssize_t bw; 324 325 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 326 327 file_start_write(file); 328 bw = vfs_iter_write(file, &i, ppos, 0); 329 file_end_write(file); 330 331 if (likely(bw == bvec->bv_len)) 332 return 0; 333 334 printk_ratelimited(KERN_ERR 335 "loop: Write error at byte offset %llu, length %i.\n", 336 (unsigned long long)*ppos, bvec->bv_len); 337 if (bw >= 0) 338 bw = -EIO; 339 return bw; 340 } 341 342 static int lo_write_simple(struct loop_device *lo, struct request *rq, 343 loff_t pos) 344 { 345 struct bio_vec bvec; 346 struct req_iterator iter; 347 int ret = 0; 348 349 rq_for_each_segment(bvec, rq, iter) { 350 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 351 if (ret < 0) 352 break; 353 cond_resched(); 354 } 355 356 return ret; 357 } 358 359 /* 360 * This is the slow, transforming version that needs to double buffer the 361 * data as it cannot do the transformations in place without having direct 362 * access to the destination pages of the backing file. 363 */ 364 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 365 loff_t pos) 366 { 367 struct bio_vec bvec, b; 368 struct req_iterator iter; 369 struct page *page; 370 int ret = 0; 371 372 page = alloc_page(GFP_NOIO); 373 if (unlikely(!page)) 374 return -ENOMEM; 375 376 rq_for_each_segment(bvec, rq, iter) { 377 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 378 bvec.bv_offset, bvec.bv_len, pos >> 9); 379 if (unlikely(ret)) 380 break; 381 382 b.bv_page = page; 383 b.bv_offset = 0; 384 b.bv_len = bvec.bv_len; 385 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 386 if (ret < 0) 387 break; 388 } 389 390 __free_page(page); 391 return ret; 392 } 393 394 static int lo_read_simple(struct loop_device *lo, struct request *rq, 395 loff_t pos) 396 { 397 struct bio_vec bvec; 398 struct req_iterator iter; 399 struct iov_iter i; 400 ssize_t len; 401 402 rq_for_each_segment(bvec, rq, iter) { 403 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 404 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 405 if (len < 0) 406 return len; 407 408 flush_dcache_page(bvec.bv_page); 409 410 if (len != bvec.bv_len) { 411 struct bio *bio; 412 413 __rq_for_each_bio(bio, rq) 414 zero_fill_bio(bio); 415 break; 416 } 417 cond_resched(); 418 } 419 420 return 0; 421 } 422 423 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 424 loff_t pos) 425 { 426 struct bio_vec bvec, b; 427 struct req_iterator iter; 428 struct iov_iter i; 429 struct page *page; 430 ssize_t len; 431 int ret = 0; 432 433 page = alloc_page(GFP_NOIO); 434 if (unlikely(!page)) 435 return -ENOMEM; 436 437 rq_for_each_segment(bvec, rq, iter) { 438 loff_t offset = pos; 439 440 b.bv_page = page; 441 b.bv_offset = 0; 442 b.bv_len = bvec.bv_len; 443 444 iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 445 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 446 if (len < 0) { 447 ret = len; 448 goto out_free_page; 449 } 450 451 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 452 bvec.bv_offset, len, offset >> 9); 453 if (ret) 454 goto out_free_page; 455 456 flush_dcache_page(bvec.bv_page); 457 458 if (len != bvec.bv_len) { 459 struct bio *bio; 460 461 __rq_for_each_bio(bio, rq) 462 zero_fill_bio(bio); 463 break; 464 } 465 } 466 467 ret = 0; 468 out_free_page: 469 __free_page(page); 470 return ret; 471 } 472 473 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 474 int mode) 475 { 476 /* 477 * We use fallocate to manipulate the space mappings used by the image 478 * a.k.a. discard/zerorange. However we do not support this if 479 * encryption is enabled, because it may give an attacker useful 480 * information. 481 */ 482 struct file *file = lo->lo_backing_file; 483 struct request_queue *q = lo->lo_queue; 484 int ret; 485 486 mode |= FALLOC_FL_KEEP_SIZE; 487 488 if (!blk_queue_discard(q)) { 489 ret = -EOPNOTSUPP; 490 goto out; 491 } 492 493 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 494 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 495 ret = -EIO; 496 out: 497 return ret; 498 } 499 500 static int lo_req_flush(struct loop_device *lo, struct request *rq) 501 { 502 struct file *file = lo->lo_backing_file; 503 int ret = vfs_fsync(file, 0); 504 if (unlikely(ret && ret != -EINVAL)) 505 ret = -EIO; 506 507 return ret; 508 } 509 510 static void lo_complete_rq(struct request *rq) 511 { 512 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 513 blk_status_t ret = BLK_STS_OK; 514 515 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 516 req_op(rq) != REQ_OP_READ) { 517 if (cmd->ret < 0) 518 ret = errno_to_blk_status(cmd->ret); 519 goto end_io; 520 } 521 522 /* 523 * Short READ - if we got some data, advance our request and 524 * retry it. If we got no data, end the rest with EIO. 525 */ 526 if (cmd->ret) { 527 blk_update_request(rq, BLK_STS_OK, cmd->ret); 528 cmd->ret = 0; 529 blk_mq_requeue_request(rq, true); 530 } else { 531 if (cmd->use_aio) { 532 struct bio *bio = rq->bio; 533 534 while (bio) { 535 zero_fill_bio(bio); 536 bio = bio->bi_next; 537 } 538 } 539 ret = BLK_STS_IOERR; 540 end_io: 541 blk_mq_end_request(rq, ret); 542 } 543 } 544 545 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 546 { 547 struct request *rq = blk_mq_rq_from_pdu(cmd); 548 549 if (!atomic_dec_and_test(&cmd->ref)) 550 return; 551 kfree(cmd->bvec); 552 cmd->bvec = NULL; 553 if (likely(!blk_should_fake_timeout(rq->q))) 554 blk_mq_complete_request(rq); 555 } 556 557 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 558 { 559 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 560 561 cmd->ret = ret; 562 lo_rw_aio_do_completion(cmd); 563 } 564 565 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 566 loff_t pos, bool rw) 567 { 568 struct iov_iter iter; 569 struct req_iterator rq_iter; 570 struct bio_vec *bvec; 571 struct request *rq = blk_mq_rq_from_pdu(cmd); 572 struct bio *bio = rq->bio; 573 struct file *file = lo->lo_backing_file; 574 struct bio_vec tmp; 575 unsigned int offset; 576 int nr_bvec = 0; 577 int ret; 578 579 rq_for_each_bvec(tmp, rq, rq_iter) 580 nr_bvec++; 581 582 if (rq->bio != rq->biotail) { 583 584 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 585 GFP_NOIO); 586 if (!bvec) 587 return -EIO; 588 cmd->bvec = bvec; 589 590 /* 591 * The bios of the request may be started from the middle of 592 * the 'bvec' because of bio splitting, so we can't directly 593 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 594 * API will take care of all details for us. 595 */ 596 rq_for_each_bvec(tmp, rq, rq_iter) { 597 *bvec = tmp; 598 bvec++; 599 } 600 bvec = cmd->bvec; 601 offset = 0; 602 } else { 603 /* 604 * Same here, this bio may be started from the middle of the 605 * 'bvec' because of bio splitting, so offset from the bvec 606 * must be passed to iov iterator 607 */ 608 offset = bio->bi_iter.bi_bvec_done; 609 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 610 } 611 atomic_set(&cmd->ref, 2); 612 613 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 614 iter.iov_offset = offset; 615 616 cmd->iocb.ki_pos = pos; 617 cmd->iocb.ki_filp = file; 618 cmd->iocb.ki_complete = lo_rw_aio_complete; 619 cmd->iocb.ki_flags = IOCB_DIRECT; 620 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 621 622 if (rw == WRITE) 623 ret = call_write_iter(file, &cmd->iocb, &iter); 624 else 625 ret = call_read_iter(file, &cmd->iocb, &iter); 626 627 lo_rw_aio_do_completion(cmd); 628 629 if (ret != -EIOCBQUEUED) 630 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 631 return 0; 632 } 633 634 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 635 { 636 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 637 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 638 639 /* 640 * lo_write_simple and lo_read_simple should have been covered 641 * by io submit style function like lo_rw_aio(), one blocker 642 * is that lo_read_simple() need to call flush_dcache_page after 643 * the page is written from kernel, and it isn't easy to handle 644 * this in io submit style function which submits all segments 645 * of the req at one time. And direct read IO doesn't need to 646 * run flush_dcache_page(). 647 */ 648 switch (req_op(rq)) { 649 case REQ_OP_FLUSH: 650 return lo_req_flush(lo, rq); 651 case REQ_OP_WRITE_ZEROES: 652 /* 653 * If the caller doesn't want deallocation, call zeroout to 654 * write zeroes the range. Otherwise, punch them out. 655 */ 656 return lo_fallocate(lo, rq, pos, 657 (rq->cmd_flags & REQ_NOUNMAP) ? 658 FALLOC_FL_ZERO_RANGE : 659 FALLOC_FL_PUNCH_HOLE); 660 case REQ_OP_DISCARD: 661 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 662 case REQ_OP_WRITE: 663 if (lo->transfer) 664 return lo_write_transfer(lo, rq, pos); 665 else if (cmd->use_aio) 666 return lo_rw_aio(lo, cmd, pos, WRITE); 667 else 668 return lo_write_simple(lo, rq, pos); 669 case REQ_OP_READ: 670 if (lo->transfer) 671 return lo_read_transfer(lo, rq, pos); 672 else if (cmd->use_aio) 673 return lo_rw_aio(lo, cmd, pos, READ); 674 else 675 return lo_read_simple(lo, rq, pos); 676 default: 677 WARN_ON_ONCE(1); 678 return -EIO; 679 } 680 } 681 682 static inline void loop_update_dio(struct loop_device *lo) 683 { 684 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 685 lo->use_dio); 686 } 687 688 static void loop_reread_partitions(struct loop_device *lo) 689 { 690 int rc; 691 692 mutex_lock(&lo->lo_disk->open_mutex); 693 rc = bdev_disk_changed(lo->lo_disk, false); 694 mutex_unlock(&lo->lo_disk->open_mutex); 695 if (rc) 696 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 697 __func__, lo->lo_number, lo->lo_file_name, rc); 698 } 699 700 static inline int is_loop_device(struct file *file) 701 { 702 struct inode *i = file->f_mapping->host; 703 704 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; 705 } 706 707 static int loop_validate_file(struct file *file, struct block_device *bdev) 708 { 709 struct inode *inode = file->f_mapping->host; 710 struct file *f = file; 711 712 /* Avoid recursion */ 713 while (is_loop_device(f)) { 714 struct loop_device *l; 715 716 lockdep_assert_held(&loop_validate_mutex); 717 if (f->f_mapping->host->i_rdev == bdev->bd_dev) 718 return -EBADF; 719 720 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 721 if (l->lo_state != Lo_bound) 722 return -EINVAL; 723 /* Order wrt setting lo->lo_backing_file in loop_configure(). */ 724 rmb(); 725 f = l->lo_backing_file; 726 } 727 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 728 return -EINVAL; 729 return 0; 730 } 731 732 /* 733 * loop_change_fd switched the backing store of a loopback device to 734 * a new file. This is useful for operating system installers to free up 735 * the original file and in High Availability environments to switch to 736 * an alternative location for the content in case of server meltdown. 737 * This can only work if the loop device is used read-only, and if the 738 * new backing store is the same size and type as the old backing store. 739 */ 740 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 741 unsigned int arg) 742 { 743 struct file *file = fget(arg); 744 struct file *old_file; 745 int error; 746 bool partscan; 747 bool is_loop; 748 749 if (!file) 750 return -EBADF; 751 is_loop = is_loop_device(file); 752 error = loop_global_lock_killable(lo, is_loop); 753 if (error) 754 goto out_putf; 755 error = -ENXIO; 756 if (lo->lo_state != Lo_bound) 757 goto out_err; 758 759 /* the loop device has to be read-only */ 760 error = -EINVAL; 761 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 762 goto out_err; 763 764 error = loop_validate_file(file, bdev); 765 if (error) 766 goto out_err; 767 768 old_file = lo->lo_backing_file; 769 770 error = -EINVAL; 771 772 /* size of the new backing store needs to be the same */ 773 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 774 goto out_err; 775 776 /* and ... switch */ 777 blk_mq_freeze_queue(lo->lo_queue); 778 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 779 lo->lo_backing_file = file; 780 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 781 mapping_set_gfp_mask(file->f_mapping, 782 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 783 loop_update_dio(lo); 784 blk_mq_unfreeze_queue(lo->lo_queue); 785 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 786 loop_global_unlock(lo, is_loop); 787 788 /* 789 * Flush loop_validate_file() before fput(), for l->lo_backing_file 790 * might be pointing at old_file which might be the last reference. 791 */ 792 if (!is_loop) { 793 mutex_lock(&loop_validate_mutex); 794 mutex_unlock(&loop_validate_mutex); 795 } 796 /* 797 * We must drop file reference outside of lo_mutex as dropping 798 * the file ref can take open_mutex which creates circular locking 799 * dependency. 800 */ 801 fput(old_file); 802 if (partscan) 803 loop_reread_partitions(lo); 804 return 0; 805 806 out_err: 807 loop_global_unlock(lo, is_loop); 808 out_putf: 809 fput(file); 810 return error; 811 } 812 813 /* loop sysfs attributes */ 814 815 static ssize_t loop_attr_show(struct device *dev, char *page, 816 ssize_t (*callback)(struct loop_device *, char *)) 817 { 818 struct gendisk *disk = dev_to_disk(dev); 819 struct loop_device *lo = disk->private_data; 820 821 return callback(lo, page); 822 } 823 824 #define LOOP_ATTR_RO(_name) \ 825 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 826 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 827 struct device_attribute *attr, char *b) \ 828 { \ 829 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 830 } \ 831 static struct device_attribute loop_attr_##_name = \ 832 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 833 834 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 835 { 836 ssize_t ret; 837 char *p = NULL; 838 839 spin_lock_irq(&lo->lo_lock); 840 if (lo->lo_backing_file) 841 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 842 spin_unlock_irq(&lo->lo_lock); 843 844 if (IS_ERR_OR_NULL(p)) 845 ret = PTR_ERR(p); 846 else { 847 ret = strlen(p); 848 memmove(buf, p, ret); 849 buf[ret++] = '\n'; 850 buf[ret] = 0; 851 } 852 853 return ret; 854 } 855 856 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 857 { 858 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 859 } 860 861 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 862 { 863 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 864 } 865 866 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 867 { 868 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 869 870 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 871 } 872 873 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 874 { 875 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 876 877 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 878 } 879 880 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 881 { 882 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 883 884 return sprintf(buf, "%s\n", dio ? "1" : "0"); 885 } 886 887 LOOP_ATTR_RO(backing_file); 888 LOOP_ATTR_RO(offset); 889 LOOP_ATTR_RO(sizelimit); 890 LOOP_ATTR_RO(autoclear); 891 LOOP_ATTR_RO(partscan); 892 LOOP_ATTR_RO(dio); 893 894 static struct attribute *loop_attrs[] = { 895 &loop_attr_backing_file.attr, 896 &loop_attr_offset.attr, 897 &loop_attr_sizelimit.attr, 898 &loop_attr_autoclear.attr, 899 &loop_attr_partscan.attr, 900 &loop_attr_dio.attr, 901 NULL, 902 }; 903 904 static struct attribute_group loop_attribute_group = { 905 .name = "loop", 906 .attrs= loop_attrs, 907 }; 908 909 static void loop_sysfs_init(struct loop_device *lo) 910 { 911 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 912 &loop_attribute_group); 913 } 914 915 static void loop_sysfs_exit(struct loop_device *lo) 916 { 917 if (lo->sysfs_inited) 918 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 919 &loop_attribute_group); 920 } 921 922 static void loop_config_discard(struct loop_device *lo) 923 { 924 struct file *file = lo->lo_backing_file; 925 struct inode *inode = file->f_mapping->host; 926 struct request_queue *q = lo->lo_queue; 927 u32 granularity, max_discard_sectors; 928 929 /* 930 * If the backing device is a block device, mirror its zeroing 931 * capability. Set the discard sectors to the block device's zeroing 932 * capabilities because loop discards result in blkdev_issue_zeroout(), 933 * not blkdev_issue_discard(). This maintains consistent behavior with 934 * file-backed loop devices: discarded regions read back as zero. 935 */ 936 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { 937 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 938 939 max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 940 granularity = backingq->limits.discard_granularity ?: 941 queue_physical_block_size(backingq); 942 943 /* 944 * We use punch hole to reclaim the free space used by the 945 * image a.k.a. discard. However we do not support discard if 946 * encryption is enabled, because it may give an attacker 947 * useful information. 948 */ 949 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { 950 max_discard_sectors = 0; 951 granularity = 0; 952 953 } else { 954 max_discard_sectors = UINT_MAX >> 9; 955 granularity = inode->i_sb->s_blocksize; 956 } 957 958 if (max_discard_sectors) { 959 q->limits.discard_granularity = granularity; 960 blk_queue_max_discard_sectors(q, max_discard_sectors); 961 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 962 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 963 } else { 964 q->limits.discard_granularity = 0; 965 blk_queue_max_discard_sectors(q, 0); 966 blk_queue_max_write_zeroes_sectors(q, 0); 967 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 968 } 969 q->limits.discard_alignment = 0; 970 } 971 972 struct loop_worker { 973 struct rb_node rb_node; 974 struct work_struct work; 975 struct list_head cmd_list; 976 struct list_head idle_list; 977 struct loop_device *lo; 978 struct cgroup_subsys_state *blkcg_css; 979 unsigned long last_ran_at; 980 }; 981 982 static void loop_workfn(struct work_struct *work); 983 static void loop_rootcg_workfn(struct work_struct *work); 984 static void loop_free_idle_workers(struct timer_list *timer); 985 986 #ifdef CONFIG_BLK_CGROUP 987 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 988 { 989 return !css || css == blkcg_root_css; 990 } 991 #else 992 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 993 { 994 return !css; 995 } 996 #endif 997 998 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 999 { 1000 struct rb_node **node = &(lo->worker_tree.rb_node), *parent = NULL; 1001 struct loop_worker *cur_worker, *worker = NULL; 1002 struct work_struct *work; 1003 struct list_head *cmd_list; 1004 1005 spin_lock_irq(&lo->lo_work_lock); 1006 1007 if (queue_on_root_worker(cmd->blkcg_css)) 1008 goto queue_work; 1009 1010 node = &lo->worker_tree.rb_node; 1011 1012 while (*node) { 1013 parent = *node; 1014 cur_worker = container_of(*node, struct loop_worker, rb_node); 1015 if (cur_worker->blkcg_css == cmd->blkcg_css) { 1016 worker = cur_worker; 1017 break; 1018 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { 1019 node = &(*node)->rb_left; 1020 } else { 1021 node = &(*node)->rb_right; 1022 } 1023 } 1024 if (worker) 1025 goto queue_work; 1026 1027 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 1028 /* 1029 * In the event we cannot allocate a worker, just queue on the 1030 * rootcg worker and issue the I/O as the rootcg 1031 */ 1032 if (!worker) { 1033 cmd->blkcg_css = NULL; 1034 if (cmd->memcg_css) 1035 css_put(cmd->memcg_css); 1036 cmd->memcg_css = NULL; 1037 goto queue_work; 1038 } 1039 1040 worker->blkcg_css = cmd->blkcg_css; 1041 css_get(worker->blkcg_css); 1042 INIT_WORK(&worker->work, loop_workfn); 1043 INIT_LIST_HEAD(&worker->cmd_list); 1044 INIT_LIST_HEAD(&worker->idle_list); 1045 worker->lo = lo; 1046 rb_link_node(&worker->rb_node, parent, node); 1047 rb_insert_color(&worker->rb_node, &lo->worker_tree); 1048 queue_work: 1049 if (worker) { 1050 /* 1051 * We need to remove from the idle list here while 1052 * holding the lock so that the idle timer doesn't 1053 * free the worker 1054 */ 1055 if (!list_empty(&worker->idle_list)) 1056 list_del_init(&worker->idle_list); 1057 work = &worker->work; 1058 cmd_list = &worker->cmd_list; 1059 } else { 1060 work = &lo->rootcg_work; 1061 cmd_list = &lo->rootcg_cmd_list; 1062 } 1063 list_add_tail(&cmd->list_entry, cmd_list); 1064 queue_work(lo->workqueue, work); 1065 spin_unlock_irq(&lo->lo_work_lock); 1066 } 1067 1068 static void loop_update_rotational(struct loop_device *lo) 1069 { 1070 struct file *file = lo->lo_backing_file; 1071 struct inode *file_inode = file->f_mapping->host; 1072 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 1073 struct request_queue *q = lo->lo_queue; 1074 bool nonrot = true; 1075 1076 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 1077 if (file_bdev) 1078 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 1079 1080 if (nonrot) 1081 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 1082 else 1083 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 1084 } 1085 1086 static int 1087 loop_release_xfer(struct loop_device *lo) 1088 { 1089 int err = 0; 1090 struct loop_func_table *xfer = lo->lo_encryption; 1091 1092 if (xfer) { 1093 if (xfer->release) 1094 err = xfer->release(lo); 1095 lo->transfer = NULL; 1096 lo->lo_encryption = NULL; 1097 module_put(xfer->owner); 1098 } 1099 return err; 1100 } 1101 1102 static int 1103 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1104 const struct loop_info64 *i) 1105 { 1106 int err = 0; 1107 1108 if (xfer) { 1109 struct module *owner = xfer->owner; 1110 1111 if (!try_module_get(owner)) 1112 return -EINVAL; 1113 if (xfer->init) 1114 err = xfer->init(lo, i); 1115 if (err) 1116 module_put(owner); 1117 else 1118 lo->lo_encryption = xfer; 1119 } 1120 return err; 1121 } 1122 1123 /** 1124 * loop_set_status_from_info - configure device from loop_info 1125 * @lo: struct loop_device to configure 1126 * @info: struct loop_info64 to configure the device with 1127 * 1128 * Configures the loop device parameters according to the passed 1129 * in loop_info64 configuration. 1130 */ 1131 static int 1132 loop_set_status_from_info(struct loop_device *lo, 1133 const struct loop_info64 *info) 1134 { 1135 int err; 1136 struct loop_func_table *xfer; 1137 kuid_t uid = current_uid(); 1138 1139 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1140 return -EINVAL; 1141 1142 err = loop_release_xfer(lo); 1143 if (err) 1144 return err; 1145 1146 if (info->lo_encrypt_type) { 1147 unsigned int type = info->lo_encrypt_type; 1148 1149 if (type >= MAX_LO_CRYPT) 1150 return -EINVAL; 1151 xfer = xfer_funcs[type]; 1152 if (xfer == NULL) 1153 return -EINVAL; 1154 } else 1155 xfer = NULL; 1156 1157 err = loop_init_xfer(lo, xfer, info); 1158 if (err) 1159 return err; 1160 1161 lo->lo_offset = info->lo_offset; 1162 lo->lo_sizelimit = info->lo_sizelimit; 1163 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1164 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1165 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1166 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1167 1168 if (!xfer) 1169 xfer = &none_funcs; 1170 lo->transfer = xfer->transfer; 1171 lo->ioctl = xfer->ioctl; 1172 1173 lo->lo_flags = info->lo_flags; 1174 1175 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1176 lo->lo_init[0] = info->lo_init[0]; 1177 lo->lo_init[1] = info->lo_init[1]; 1178 if (info->lo_encrypt_key_size) { 1179 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1180 info->lo_encrypt_key_size); 1181 lo->lo_key_owner = uid; 1182 } 1183 1184 return 0; 1185 } 1186 1187 static int loop_configure(struct loop_device *lo, fmode_t mode, 1188 struct block_device *bdev, 1189 const struct loop_config *config) 1190 { 1191 struct file *file = fget(config->fd); 1192 struct inode *inode; 1193 struct address_space *mapping; 1194 int error; 1195 loff_t size; 1196 bool partscan; 1197 unsigned short bsize; 1198 bool is_loop; 1199 1200 if (!file) 1201 return -EBADF; 1202 is_loop = is_loop_device(file); 1203 1204 /* This is safe, since we have a reference from open(). */ 1205 __module_get(THIS_MODULE); 1206 1207 /* 1208 * If we don't hold exclusive handle for the device, upgrade to it 1209 * here to avoid changing device under exclusive owner. 1210 */ 1211 if (!(mode & FMODE_EXCL)) { 1212 error = bd_prepare_to_claim(bdev, loop_configure); 1213 if (error) 1214 goto out_putf; 1215 } 1216 1217 error = loop_global_lock_killable(lo, is_loop); 1218 if (error) 1219 goto out_bdev; 1220 1221 error = -EBUSY; 1222 if (lo->lo_state != Lo_unbound) 1223 goto out_unlock; 1224 1225 error = loop_validate_file(file, bdev); 1226 if (error) 1227 goto out_unlock; 1228 1229 mapping = file->f_mapping; 1230 inode = mapping->host; 1231 1232 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1233 error = -EINVAL; 1234 goto out_unlock; 1235 } 1236 1237 if (config->block_size) { 1238 error = loop_validate_block_size(config->block_size); 1239 if (error) 1240 goto out_unlock; 1241 } 1242 1243 error = loop_set_status_from_info(lo, &config->info); 1244 if (error) 1245 goto out_unlock; 1246 1247 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1248 !file->f_op->write_iter) 1249 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1250 1251 lo->workqueue = alloc_workqueue("loop%d", 1252 WQ_UNBOUND | WQ_FREEZABLE, 1253 0, 1254 lo->lo_number); 1255 if (!lo->workqueue) { 1256 error = -ENOMEM; 1257 goto out_unlock; 1258 } 1259 1260 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1261 1262 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 1263 INIT_LIST_HEAD(&lo->rootcg_cmd_list); 1264 INIT_LIST_HEAD(&lo->idle_worker_list); 1265 lo->worker_tree = RB_ROOT; 1266 timer_setup(&lo->timer, loop_free_idle_workers, 1267 TIMER_DEFERRABLE); 1268 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1269 lo->lo_device = bdev; 1270 lo->lo_backing_file = file; 1271 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1272 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1273 1274 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1275 blk_queue_write_cache(lo->lo_queue, true, false); 1276 1277 if (config->block_size) 1278 bsize = config->block_size; 1279 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1280 /* In case of direct I/O, match underlying block size */ 1281 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1282 else 1283 bsize = 512; 1284 1285 blk_queue_logical_block_size(lo->lo_queue, bsize); 1286 blk_queue_physical_block_size(lo->lo_queue, bsize); 1287 blk_queue_io_min(lo->lo_queue, bsize); 1288 1289 loop_config_discard(lo); 1290 loop_update_rotational(lo); 1291 loop_update_dio(lo); 1292 loop_sysfs_init(lo); 1293 1294 size = get_loop_size(lo, file); 1295 loop_set_size(lo, size); 1296 1297 /* Order wrt reading lo_state in loop_validate_file(). */ 1298 wmb(); 1299 1300 lo->lo_state = Lo_bound; 1301 if (part_shift) 1302 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1303 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1304 if (partscan) 1305 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1306 1307 /* Grab the block_device to prevent its destruction after we 1308 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 1309 */ 1310 bdgrab(bdev); 1311 loop_global_unlock(lo, is_loop); 1312 if (partscan) 1313 loop_reread_partitions(lo); 1314 if (!(mode & FMODE_EXCL)) 1315 bd_abort_claiming(bdev, loop_configure); 1316 return 0; 1317 1318 out_unlock: 1319 loop_global_unlock(lo, is_loop); 1320 out_bdev: 1321 if (!(mode & FMODE_EXCL)) 1322 bd_abort_claiming(bdev, loop_configure); 1323 out_putf: 1324 fput(file); 1325 /* This is safe: open() is still holding a reference. */ 1326 module_put(THIS_MODULE); 1327 return error; 1328 } 1329 1330 static int __loop_clr_fd(struct loop_device *lo, bool release) 1331 { 1332 struct file *filp = NULL; 1333 gfp_t gfp = lo->old_gfp_mask; 1334 struct block_device *bdev = lo->lo_device; 1335 int err = 0; 1336 bool partscan = false; 1337 int lo_number; 1338 struct loop_worker *pos, *worker; 1339 1340 /* 1341 * Flush loop_configure() and loop_change_fd(). It is acceptable for 1342 * loop_validate_file() to succeed, for actual clear operation has not 1343 * started yet. 1344 */ 1345 mutex_lock(&loop_validate_mutex); 1346 mutex_unlock(&loop_validate_mutex); 1347 /* 1348 * loop_validate_file() now fails because l->lo_state != Lo_bound 1349 * became visible. 1350 */ 1351 1352 mutex_lock(&lo->lo_mutex); 1353 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1354 err = -ENXIO; 1355 goto out_unlock; 1356 } 1357 1358 filp = lo->lo_backing_file; 1359 if (filp == NULL) { 1360 err = -EINVAL; 1361 goto out_unlock; 1362 } 1363 1364 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1365 blk_queue_write_cache(lo->lo_queue, false, false); 1366 1367 /* freeze request queue during the transition */ 1368 blk_mq_freeze_queue(lo->lo_queue); 1369 1370 destroy_workqueue(lo->workqueue); 1371 spin_lock_irq(&lo->lo_work_lock); 1372 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 1373 idle_list) { 1374 list_del(&worker->idle_list); 1375 rb_erase(&worker->rb_node, &lo->worker_tree); 1376 css_put(worker->blkcg_css); 1377 kfree(worker); 1378 } 1379 spin_unlock_irq(&lo->lo_work_lock); 1380 del_timer_sync(&lo->timer); 1381 1382 spin_lock_irq(&lo->lo_lock); 1383 lo->lo_backing_file = NULL; 1384 spin_unlock_irq(&lo->lo_lock); 1385 1386 loop_release_xfer(lo); 1387 lo->transfer = NULL; 1388 lo->ioctl = NULL; 1389 lo->lo_device = NULL; 1390 lo->lo_encryption = NULL; 1391 lo->lo_offset = 0; 1392 lo->lo_sizelimit = 0; 1393 lo->lo_encrypt_key_size = 0; 1394 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1395 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1396 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1397 blk_queue_logical_block_size(lo->lo_queue, 512); 1398 blk_queue_physical_block_size(lo->lo_queue, 512); 1399 blk_queue_io_min(lo->lo_queue, 512); 1400 if (bdev) { 1401 bdput(bdev); 1402 invalidate_bdev(bdev); 1403 bdev->bd_inode->i_mapping->wb_err = 0; 1404 } 1405 set_capacity(lo->lo_disk, 0); 1406 loop_sysfs_exit(lo); 1407 if (bdev) { 1408 /* let user-space know about this change */ 1409 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1410 } 1411 mapping_set_gfp_mask(filp->f_mapping, gfp); 1412 /* This is safe: open() is still holding a reference. */ 1413 module_put(THIS_MODULE); 1414 blk_mq_unfreeze_queue(lo->lo_queue); 1415 1416 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1417 lo_number = lo->lo_number; 1418 out_unlock: 1419 mutex_unlock(&lo->lo_mutex); 1420 if (partscan) { 1421 /* 1422 * open_mutex has been held already in release path, so don't 1423 * acquire it if this function is called in such case. 1424 * 1425 * If the reread partition isn't from release path, lo_refcnt 1426 * must be at least one and it can only become zero when the 1427 * current holder is released. 1428 */ 1429 if (!release) 1430 mutex_lock(&lo->lo_disk->open_mutex); 1431 err = bdev_disk_changed(lo->lo_disk, false); 1432 if (!release) 1433 mutex_unlock(&lo->lo_disk->open_mutex); 1434 if (err) 1435 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1436 __func__, lo_number, err); 1437 /* Device is gone, no point in returning error */ 1438 err = 0; 1439 } 1440 1441 /* 1442 * lo->lo_state is set to Lo_unbound here after above partscan has 1443 * finished. 1444 * 1445 * There cannot be anybody else entering __loop_clr_fd() as 1446 * lo->lo_backing_file is already cleared and Lo_rundown state 1447 * protects us from all the other places trying to change the 'lo' 1448 * device. 1449 */ 1450 mutex_lock(&lo->lo_mutex); 1451 lo->lo_flags = 0; 1452 if (!part_shift) 1453 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1454 lo->lo_state = Lo_unbound; 1455 mutex_unlock(&lo->lo_mutex); 1456 1457 /* 1458 * Need not hold lo_mutex to fput backing file. Calling fput holding 1459 * lo_mutex triggers a circular lock dependency possibility warning as 1460 * fput can take open_mutex which is usually taken before lo_mutex. 1461 */ 1462 if (filp) 1463 fput(filp); 1464 return err; 1465 } 1466 1467 static int loop_clr_fd(struct loop_device *lo) 1468 { 1469 int err; 1470 1471 err = mutex_lock_killable(&lo->lo_mutex); 1472 if (err) 1473 return err; 1474 if (lo->lo_state != Lo_bound) { 1475 mutex_unlock(&lo->lo_mutex); 1476 return -ENXIO; 1477 } 1478 /* 1479 * If we've explicitly asked to tear down the loop device, 1480 * and it has an elevated reference count, set it for auto-teardown when 1481 * the last reference goes away. This stops $!~#$@ udev from 1482 * preventing teardown because it decided that it needs to run blkid on 1483 * the loopback device whenever they appear. xfstests is notorious for 1484 * failing tests because blkid via udev races with a losetup 1485 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1486 * command to fail with EBUSY. 1487 */ 1488 if (atomic_read(&lo->lo_refcnt) > 1) { 1489 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1490 mutex_unlock(&lo->lo_mutex); 1491 return 0; 1492 } 1493 lo->lo_state = Lo_rundown; 1494 mutex_unlock(&lo->lo_mutex); 1495 1496 return __loop_clr_fd(lo, false); 1497 } 1498 1499 static int 1500 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1501 { 1502 int err; 1503 kuid_t uid = current_uid(); 1504 int prev_lo_flags; 1505 bool partscan = false; 1506 bool size_changed = false; 1507 1508 err = mutex_lock_killable(&lo->lo_mutex); 1509 if (err) 1510 return err; 1511 if (lo->lo_encrypt_key_size && 1512 !uid_eq(lo->lo_key_owner, uid) && 1513 !capable(CAP_SYS_ADMIN)) { 1514 err = -EPERM; 1515 goto out_unlock; 1516 } 1517 if (lo->lo_state != Lo_bound) { 1518 err = -ENXIO; 1519 goto out_unlock; 1520 } 1521 1522 if (lo->lo_offset != info->lo_offset || 1523 lo->lo_sizelimit != info->lo_sizelimit) { 1524 size_changed = true; 1525 sync_blockdev(lo->lo_device); 1526 invalidate_bdev(lo->lo_device); 1527 } 1528 1529 /* I/O need to be drained during transfer transition */ 1530 blk_mq_freeze_queue(lo->lo_queue); 1531 1532 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1533 /* If any pages were dirtied after invalidate_bdev(), try again */ 1534 err = -EAGAIN; 1535 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1536 __func__, lo->lo_number, lo->lo_file_name, 1537 lo->lo_device->bd_inode->i_mapping->nrpages); 1538 goto out_unfreeze; 1539 } 1540 1541 prev_lo_flags = lo->lo_flags; 1542 1543 err = loop_set_status_from_info(lo, info); 1544 if (err) 1545 goto out_unfreeze; 1546 1547 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1548 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1549 /* For those flags, use the previous values instead */ 1550 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1551 /* For flags that can't be cleared, use previous values too */ 1552 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1553 1554 if (size_changed) { 1555 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1556 lo->lo_backing_file); 1557 loop_set_size(lo, new_size); 1558 } 1559 1560 loop_config_discard(lo); 1561 1562 /* update dio if lo_offset or transfer is changed */ 1563 __loop_update_dio(lo, lo->use_dio); 1564 1565 out_unfreeze: 1566 blk_mq_unfreeze_queue(lo->lo_queue); 1567 1568 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1569 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1570 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1571 partscan = true; 1572 } 1573 out_unlock: 1574 mutex_unlock(&lo->lo_mutex); 1575 if (partscan) 1576 loop_reread_partitions(lo); 1577 1578 return err; 1579 } 1580 1581 static int 1582 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1583 { 1584 struct path path; 1585 struct kstat stat; 1586 int ret; 1587 1588 ret = mutex_lock_killable(&lo->lo_mutex); 1589 if (ret) 1590 return ret; 1591 if (lo->lo_state != Lo_bound) { 1592 mutex_unlock(&lo->lo_mutex); 1593 return -ENXIO; 1594 } 1595 1596 memset(info, 0, sizeof(*info)); 1597 info->lo_number = lo->lo_number; 1598 info->lo_offset = lo->lo_offset; 1599 info->lo_sizelimit = lo->lo_sizelimit; 1600 info->lo_flags = lo->lo_flags; 1601 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1602 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1603 info->lo_encrypt_type = 1604 lo->lo_encryption ? lo->lo_encryption->number : 0; 1605 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1606 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1607 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1608 lo->lo_encrypt_key_size); 1609 } 1610 1611 /* Drop lo_mutex while we call into the filesystem. */ 1612 path = lo->lo_backing_file->f_path; 1613 path_get(&path); 1614 mutex_unlock(&lo->lo_mutex); 1615 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1616 if (!ret) { 1617 info->lo_device = huge_encode_dev(stat.dev); 1618 info->lo_inode = stat.ino; 1619 info->lo_rdevice = huge_encode_dev(stat.rdev); 1620 } 1621 path_put(&path); 1622 return ret; 1623 } 1624 1625 static void 1626 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1627 { 1628 memset(info64, 0, sizeof(*info64)); 1629 info64->lo_number = info->lo_number; 1630 info64->lo_device = info->lo_device; 1631 info64->lo_inode = info->lo_inode; 1632 info64->lo_rdevice = info->lo_rdevice; 1633 info64->lo_offset = info->lo_offset; 1634 info64->lo_sizelimit = 0; 1635 info64->lo_encrypt_type = info->lo_encrypt_type; 1636 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1637 info64->lo_flags = info->lo_flags; 1638 info64->lo_init[0] = info->lo_init[0]; 1639 info64->lo_init[1] = info->lo_init[1]; 1640 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1641 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1642 else 1643 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1644 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1645 } 1646 1647 static int 1648 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1649 { 1650 memset(info, 0, sizeof(*info)); 1651 info->lo_number = info64->lo_number; 1652 info->lo_device = info64->lo_device; 1653 info->lo_inode = info64->lo_inode; 1654 info->lo_rdevice = info64->lo_rdevice; 1655 info->lo_offset = info64->lo_offset; 1656 info->lo_encrypt_type = info64->lo_encrypt_type; 1657 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1658 info->lo_flags = info64->lo_flags; 1659 info->lo_init[0] = info64->lo_init[0]; 1660 info->lo_init[1] = info64->lo_init[1]; 1661 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1662 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1663 else 1664 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1665 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1666 1667 /* error in case values were truncated */ 1668 if (info->lo_device != info64->lo_device || 1669 info->lo_rdevice != info64->lo_rdevice || 1670 info->lo_inode != info64->lo_inode || 1671 info->lo_offset != info64->lo_offset) 1672 return -EOVERFLOW; 1673 1674 return 0; 1675 } 1676 1677 static int 1678 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1679 { 1680 struct loop_info info; 1681 struct loop_info64 info64; 1682 1683 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1684 return -EFAULT; 1685 loop_info64_from_old(&info, &info64); 1686 return loop_set_status(lo, &info64); 1687 } 1688 1689 static int 1690 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1691 { 1692 struct loop_info64 info64; 1693 1694 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1695 return -EFAULT; 1696 return loop_set_status(lo, &info64); 1697 } 1698 1699 static int 1700 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1701 struct loop_info info; 1702 struct loop_info64 info64; 1703 int err; 1704 1705 if (!arg) 1706 return -EINVAL; 1707 err = loop_get_status(lo, &info64); 1708 if (!err) 1709 err = loop_info64_to_old(&info64, &info); 1710 if (!err && copy_to_user(arg, &info, sizeof(info))) 1711 err = -EFAULT; 1712 1713 return err; 1714 } 1715 1716 static int 1717 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1718 struct loop_info64 info64; 1719 int err; 1720 1721 if (!arg) 1722 return -EINVAL; 1723 err = loop_get_status(lo, &info64); 1724 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1725 err = -EFAULT; 1726 1727 return err; 1728 } 1729 1730 static int loop_set_capacity(struct loop_device *lo) 1731 { 1732 loff_t size; 1733 1734 if (unlikely(lo->lo_state != Lo_bound)) 1735 return -ENXIO; 1736 1737 size = get_loop_size(lo, lo->lo_backing_file); 1738 loop_set_size(lo, size); 1739 1740 return 0; 1741 } 1742 1743 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1744 { 1745 int error = -ENXIO; 1746 if (lo->lo_state != Lo_bound) 1747 goto out; 1748 1749 __loop_update_dio(lo, !!arg); 1750 if (lo->use_dio == !!arg) 1751 return 0; 1752 error = -EINVAL; 1753 out: 1754 return error; 1755 } 1756 1757 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1758 { 1759 int err = 0; 1760 1761 if (lo->lo_state != Lo_bound) 1762 return -ENXIO; 1763 1764 err = loop_validate_block_size(arg); 1765 if (err) 1766 return err; 1767 1768 if (lo->lo_queue->limits.logical_block_size == arg) 1769 return 0; 1770 1771 sync_blockdev(lo->lo_device); 1772 invalidate_bdev(lo->lo_device); 1773 1774 blk_mq_freeze_queue(lo->lo_queue); 1775 1776 /* invalidate_bdev should have truncated all the pages */ 1777 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1778 err = -EAGAIN; 1779 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1780 __func__, lo->lo_number, lo->lo_file_name, 1781 lo->lo_device->bd_inode->i_mapping->nrpages); 1782 goto out_unfreeze; 1783 } 1784 1785 blk_queue_logical_block_size(lo->lo_queue, arg); 1786 blk_queue_physical_block_size(lo->lo_queue, arg); 1787 blk_queue_io_min(lo->lo_queue, arg); 1788 loop_update_dio(lo); 1789 out_unfreeze: 1790 blk_mq_unfreeze_queue(lo->lo_queue); 1791 1792 return err; 1793 } 1794 1795 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1796 unsigned long arg) 1797 { 1798 int err; 1799 1800 err = mutex_lock_killable(&lo->lo_mutex); 1801 if (err) 1802 return err; 1803 switch (cmd) { 1804 case LOOP_SET_CAPACITY: 1805 err = loop_set_capacity(lo); 1806 break; 1807 case LOOP_SET_DIRECT_IO: 1808 err = loop_set_dio(lo, arg); 1809 break; 1810 case LOOP_SET_BLOCK_SIZE: 1811 err = loop_set_block_size(lo, arg); 1812 break; 1813 default: 1814 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1815 } 1816 mutex_unlock(&lo->lo_mutex); 1817 return err; 1818 } 1819 1820 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1821 unsigned int cmd, unsigned long arg) 1822 { 1823 struct loop_device *lo = bdev->bd_disk->private_data; 1824 void __user *argp = (void __user *) arg; 1825 int err; 1826 1827 switch (cmd) { 1828 case LOOP_SET_FD: { 1829 /* 1830 * Legacy case - pass in a zeroed out struct loop_config with 1831 * only the file descriptor set , which corresponds with the 1832 * default parameters we'd have used otherwise. 1833 */ 1834 struct loop_config config; 1835 1836 memset(&config, 0, sizeof(config)); 1837 config.fd = arg; 1838 1839 return loop_configure(lo, mode, bdev, &config); 1840 } 1841 case LOOP_CONFIGURE: { 1842 struct loop_config config; 1843 1844 if (copy_from_user(&config, argp, sizeof(config))) 1845 return -EFAULT; 1846 1847 return loop_configure(lo, mode, bdev, &config); 1848 } 1849 case LOOP_CHANGE_FD: 1850 return loop_change_fd(lo, bdev, arg); 1851 case LOOP_CLR_FD: 1852 return loop_clr_fd(lo); 1853 case LOOP_SET_STATUS: 1854 err = -EPERM; 1855 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1856 err = loop_set_status_old(lo, argp); 1857 } 1858 break; 1859 case LOOP_GET_STATUS: 1860 return loop_get_status_old(lo, argp); 1861 case LOOP_SET_STATUS64: 1862 err = -EPERM; 1863 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1864 err = loop_set_status64(lo, argp); 1865 } 1866 break; 1867 case LOOP_GET_STATUS64: 1868 return loop_get_status64(lo, argp); 1869 case LOOP_SET_CAPACITY: 1870 case LOOP_SET_DIRECT_IO: 1871 case LOOP_SET_BLOCK_SIZE: 1872 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1873 return -EPERM; 1874 fallthrough; 1875 default: 1876 err = lo_simple_ioctl(lo, cmd, arg); 1877 break; 1878 } 1879 1880 return err; 1881 } 1882 1883 #ifdef CONFIG_COMPAT 1884 struct compat_loop_info { 1885 compat_int_t lo_number; /* ioctl r/o */ 1886 compat_dev_t lo_device; /* ioctl r/o */ 1887 compat_ulong_t lo_inode; /* ioctl r/o */ 1888 compat_dev_t lo_rdevice; /* ioctl r/o */ 1889 compat_int_t lo_offset; 1890 compat_int_t lo_encrypt_type; 1891 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1892 compat_int_t lo_flags; /* ioctl r/o */ 1893 char lo_name[LO_NAME_SIZE]; 1894 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1895 compat_ulong_t lo_init[2]; 1896 char reserved[4]; 1897 }; 1898 1899 /* 1900 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1901 * - noinlined to reduce stack space usage in main part of driver 1902 */ 1903 static noinline int 1904 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1905 struct loop_info64 *info64) 1906 { 1907 struct compat_loop_info info; 1908 1909 if (copy_from_user(&info, arg, sizeof(info))) 1910 return -EFAULT; 1911 1912 memset(info64, 0, sizeof(*info64)); 1913 info64->lo_number = info.lo_number; 1914 info64->lo_device = info.lo_device; 1915 info64->lo_inode = info.lo_inode; 1916 info64->lo_rdevice = info.lo_rdevice; 1917 info64->lo_offset = info.lo_offset; 1918 info64->lo_sizelimit = 0; 1919 info64->lo_encrypt_type = info.lo_encrypt_type; 1920 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1921 info64->lo_flags = info.lo_flags; 1922 info64->lo_init[0] = info.lo_init[0]; 1923 info64->lo_init[1] = info.lo_init[1]; 1924 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1925 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1926 else 1927 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1928 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1929 return 0; 1930 } 1931 1932 /* 1933 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1934 * - noinlined to reduce stack space usage in main part of driver 1935 */ 1936 static noinline int 1937 loop_info64_to_compat(const struct loop_info64 *info64, 1938 struct compat_loop_info __user *arg) 1939 { 1940 struct compat_loop_info info; 1941 1942 memset(&info, 0, sizeof(info)); 1943 info.lo_number = info64->lo_number; 1944 info.lo_device = info64->lo_device; 1945 info.lo_inode = info64->lo_inode; 1946 info.lo_rdevice = info64->lo_rdevice; 1947 info.lo_offset = info64->lo_offset; 1948 info.lo_encrypt_type = info64->lo_encrypt_type; 1949 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1950 info.lo_flags = info64->lo_flags; 1951 info.lo_init[0] = info64->lo_init[0]; 1952 info.lo_init[1] = info64->lo_init[1]; 1953 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1954 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1955 else 1956 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1957 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1958 1959 /* error in case values were truncated */ 1960 if (info.lo_device != info64->lo_device || 1961 info.lo_rdevice != info64->lo_rdevice || 1962 info.lo_inode != info64->lo_inode || 1963 info.lo_offset != info64->lo_offset || 1964 info.lo_init[0] != info64->lo_init[0] || 1965 info.lo_init[1] != info64->lo_init[1]) 1966 return -EOVERFLOW; 1967 1968 if (copy_to_user(arg, &info, sizeof(info))) 1969 return -EFAULT; 1970 return 0; 1971 } 1972 1973 static int 1974 loop_set_status_compat(struct loop_device *lo, 1975 const struct compat_loop_info __user *arg) 1976 { 1977 struct loop_info64 info64; 1978 int ret; 1979 1980 ret = loop_info64_from_compat(arg, &info64); 1981 if (ret < 0) 1982 return ret; 1983 return loop_set_status(lo, &info64); 1984 } 1985 1986 static int 1987 loop_get_status_compat(struct loop_device *lo, 1988 struct compat_loop_info __user *arg) 1989 { 1990 struct loop_info64 info64; 1991 int err; 1992 1993 if (!arg) 1994 return -EINVAL; 1995 err = loop_get_status(lo, &info64); 1996 if (!err) 1997 err = loop_info64_to_compat(&info64, arg); 1998 return err; 1999 } 2000 2001 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 2002 unsigned int cmd, unsigned long arg) 2003 { 2004 struct loop_device *lo = bdev->bd_disk->private_data; 2005 int err; 2006 2007 switch(cmd) { 2008 case LOOP_SET_STATUS: 2009 err = loop_set_status_compat(lo, 2010 (const struct compat_loop_info __user *)arg); 2011 break; 2012 case LOOP_GET_STATUS: 2013 err = loop_get_status_compat(lo, 2014 (struct compat_loop_info __user *)arg); 2015 break; 2016 case LOOP_SET_CAPACITY: 2017 case LOOP_CLR_FD: 2018 case LOOP_GET_STATUS64: 2019 case LOOP_SET_STATUS64: 2020 case LOOP_CONFIGURE: 2021 arg = (unsigned long) compat_ptr(arg); 2022 fallthrough; 2023 case LOOP_SET_FD: 2024 case LOOP_CHANGE_FD: 2025 case LOOP_SET_BLOCK_SIZE: 2026 case LOOP_SET_DIRECT_IO: 2027 err = lo_ioctl(bdev, mode, cmd, arg); 2028 break; 2029 default: 2030 err = -ENOIOCTLCMD; 2031 break; 2032 } 2033 return err; 2034 } 2035 #endif 2036 2037 static int lo_open(struct block_device *bdev, fmode_t mode) 2038 { 2039 struct loop_device *lo = bdev->bd_disk->private_data; 2040 int err; 2041 2042 err = mutex_lock_killable(&lo->lo_mutex); 2043 if (err) 2044 return err; 2045 if (lo->lo_state == Lo_deleting) 2046 err = -ENXIO; 2047 else 2048 atomic_inc(&lo->lo_refcnt); 2049 mutex_unlock(&lo->lo_mutex); 2050 return err; 2051 } 2052 2053 static void lo_release(struct gendisk *disk, fmode_t mode) 2054 { 2055 struct loop_device *lo = disk->private_data; 2056 2057 mutex_lock(&lo->lo_mutex); 2058 if (atomic_dec_return(&lo->lo_refcnt)) 2059 goto out_unlock; 2060 2061 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 2062 if (lo->lo_state != Lo_bound) 2063 goto out_unlock; 2064 lo->lo_state = Lo_rundown; 2065 mutex_unlock(&lo->lo_mutex); 2066 /* 2067 * In autoclear mode, stop the loop thread 2068 * and remove configuration after last close. 2069 */ 2070 __loop_clr_fd(lo, true); 2071 return; 2072 } else if (lo->lo_state == Lo_bound) { 2073 /* 2074 * Otherwise keep thread (if running) and config, 2075 * but flush possible ongoing bios in thread. 2076 */ 2077 blk_mq_freeze_queue(lo->lo_queue); 2078 blk_mq_unfreeze_queue(lo->lo_queue); 2079 } 2080 2081 out_unlock: 2082 mutex_unlock(&lo->lo_mutex); 2083 } 2084 2085 static const struct block_device_operations lo_fops = { 2086 .owner = THIS_MODULE, 2087 .open = lo_open, 2088 .release = lo_release, 2089 .ioctl = lo_ioctl, 2090 #ifdef CONFIG_COMPAT 2091 .compat_ioctl = lo_compat_ioctl, 2092 #endif 2093 }; 2094 2095 /* 2096 * And now the modules code and kernel interface. 2097 */ 2098 static int max_loop; 2099 module_param(max_loop, int, 0444); 2100 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 2101 module_param(max_part, int, 0444); 2102 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 2103 MODULE_LICENSE("GPL"); 2104 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 2105 2106 int loop_register_transfer(struct loop_func_table *funcs) 2107 { 2108 unsigned int n = funcs->number; 2109 2110 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 2111 return -EINVAL; 2112 xfer_funcs[n] = funcs; 2113 return 0; 2114 } 2115 2116 static int unregister_transfer_cb(int id, void *ptr, void *data) 2117 { 2118 struct loop_device *lo = ptr; 2119 struct loop_func_table *xfer = data; 2120 2121 mutex_lock(&lo->lo_mutex); 2122 if (lo->lo_encryption == xfer) 2123 loop_release_xfer(lo); 2124 mutex_unlock(&lo->lo_mutex); 2125 return 0; 2126 } 2127 2128 int loop_unregister_transfer(int number) 2129 { 2130 unsigned int n = number; 2131 struct loop_func_table *xfer; 2132 2133 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 2134 return -EINVAL; 2135 2136 xfer_funcs[n] = NULL; 2137 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 2138 return 0; 2139 } 2140 2141 EXPORT_SYMBOL(loop_register_transfer); 2142 EXPORT_SYMBOL(loop_unregister_transfer); 2143 2144 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 2145 const struct blk_mq_queue_data *bd) 2146 { 2147 struct request *rq = bd->rq; 2148 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2149 struct loop_device *lo = rq->q->queuedata; 2150 2151 blk_mq_start_request(rq); 2152 2153 if (lo->lo_state != Lo_bound) 2154 return BLK_STS_IOERR; 2155 2156 switch (req_op(rq)) { 2157 case REQ_OP_FLUSH: 2158 case REQ_OP_DISCARD: 2159 case REQ_OP_WRITE_ZEROES: 2160 cmd->use_aio = false; 2161 break; 2162 default: 2163 cmd->use_aio = lo->use_dio; 2164 break; 2165 } 2166 2167 /* always use the first bio's css */ 2168 cmd->blkcg_css = NULL; 2169 cmd->memcg_css = NULL; 2170 #ifdef CONFIG_BLK_CGROUP 2171 if (rq->bio && rq->bio->bi_blkg) { 2172 cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 2173 #ifdef CONFIG_MEMCG 2174 cmd->memcg_css = 2175 cgroup_get_e_css(cmd->blkcg_css->cgroup, 2176 &memory_cgrp_subsys); 2177 #endif 2178 } 2179 #endif 2180 loop_queue_work(lo, cmd); 2181 2182 return BLK_STS_OK; 2183 } 2184 2185 static void loop_handle_cmd(struct loop_cmd *cmd) 2186 { 2187 struct request *rq = blk_mq_rq_from_pdu(cmd); 2188 const bool write = op_is_write(req_op(rq)); 2189 struct loop_device *lo = rq->q->queuedata; 2190 int ret = 0; 2191 struct mem_cgroup *old_memcg = NULL; 2192 2193 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 2194 ret = -EIO; 2195 goto failed; 2196 } 2197 2198 if (cmd->blkcg_css) 2199 kthread_associate_blkcg(cmd->blkcg_css); 2200 if (cmd->memcg_css) 2201 old_memcg = set_active_memcg( 2202 mem_cgroup_from_css(cmd->memcg_css)); 2203 2204 ret = do_req_filebacked(lo, rq); 2205 2206 if (cmd->blkcg_css) 2207 kthread_associate_blkcg(NULL); 2208 2209 if (cmd->memcg_css) { 2210 set_active_memcg(old_memcg); 2211 css_put(cmd->memcg_css); 2212 } 2213 failed: 2214 /* complete non-aio request */ 2215 if (!cmd->use_aio || ret) { 2216 if (ret == -EOPNOTSUPP) 2217 cmd->ret = ret; 2218 else 2219 cmd->ret = ret ? -EIO : 0; 2220 if (likely(!blk_should_fake_timeout(rq->q))) 2221 blk_mq_complete_request(rq); 2222 } 2223 } 2224 2225 static void loop_set_timer(struct loop_device *lo) 2226 { 2227 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 2228 } 2229 2230 static void loop_process_work(struct loop_worker *worker, 2231 struct list_head *cmd_list, struct loop_device *lo) 2232 { 2233 int orig_flags = current->flags; 2234 struct loop_cmd *cmd; 2235 2236 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 2237 spin_lock_irq(&lo->lo_work_lock); 2238 while (!list_empty(cmd_list)) { 2239 cmd = container_of( 2240 cmd_list->next, struct loop_cmd, list_entry); 2241 list_del(cmd_list->next); 2242 spin_unlock_irq(&lo->lo_work_lock); 2243 2244 loop_handle_cmd(cmd); 2245 cond_resched(); 2246 2247 spin_lock_irq(&lo->lo_work_lock); 2248 } 2249 2250 /* 2251 * We only add to the idle list if there are no pending cmds 2252 * *and* the worker will not run again which ensures that it 2253 * is safe to free any worker on the idle list 2254 */ 2255 if (worker && !work_pending(&worker->work)) { 2256 worker->last_ran_at = jiffies; 2257 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 2258 loop_set_timer(lo); 2259 } 2260 spin_unlock_irq(&lo->lo_work_lock); 2261 current->flags = orig_flags; 2262 } 2263 2264 static void loop_workfn(struct work_struct *work) 2265 { 2266 struct loop_worker *worker = 2267 container_of(work, struct loop_worker, work); 2268 loop_process_work(worker, &worker->cmd_list, worker->lo); 2269 } 2270 2271 static void loop_rootcg_workfn(struct work_struct *work) 2272 { 2273 struct loop_device *lo = 2274 container_of(work, struct loop_device, rootcg_work); 2275 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 2276 } 2277 2278 static void loop_free_idle_workers(struct timer_list *timer) 2279 { 2280 struct loop_device *lo = container_of(timer, struct loop_device, timer); 2281 struct loop_worker *pos, *worker; 2282 2283 spin_lock_irq(&lo->lo_work_lock); 2284 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 2285 idle_list) { 2286 if (time_is_after_jiffies(worker->last_ran_at + 2287 LOOP_IDLE_WORKER_TIMEOUT)) 2288 break; 2289 list_del(&worker->idle_list); 2290 rb_erase(&worker->rb_node, &lo->worker_tree); 2291 css_put(worker->blkcg_css); 2292 kfree(worker); 2293 } 2294 if (!list_empty(&lo->idle_worker_list)) 2295 loop_set_timer(lo); 2296 spin_unlock_irq(&lo->lo_work_lock); 2297 } 2298 2299 static const struct blk_mq_ops loop_mq_ops = { 2300 .queue_rq = loop_queue_rq, 2301 .complete = lo_complete_rq, 2302 }; 2303 2304 static int loop_add(int i) 2305 { 2306 struct loop_device *lo; 2307 struct gendisk *disk; 2308 int err; 2309 2310 err = -ENOMEM; 2311 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 2312 if (!lo) 2313 goto out; 2314 lo->lo_state = Lo_unbound; 2315 2316 err = mutex_lock_killable(&loop_ctl_mutex); 2317 if (err) 2318 goto out_free_dev; 2319 2320 /* allocate id, if @id >= 0, we're requesting that specific id */ 2321 if (i >= 0) { 2322 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2323 if (err == -ENOSPC) 2324 err = -EEXIST; 2325 } else { 2326 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2327 } 2328 if (err < 0) 2329 goto out_unlock; 2330 i = err; 2331 2332 err = -ENOMEM; 2333 lo->tag_set.ops = &loop_mq_ops; 2334 lo->tag_set.nr_hw_queues = 1; 2335 lo->tag_set.queue_depth = 128; 2336 lo->tag_set.numa_node = NUMA_NO_NODE; 2337 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2338 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING; 2339 lo->tag_set.driver_data = lo; 2340 2341 err = blk_mq_alloc_tag_set(&lo->tag_set); 2342 if (err) 2343 goto out_free_idr; 2344 2345 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 2346 if (IS_ERR(disk)) { 2347 err = PTR_ERR(disk); 2348 goto out_cleanup_tags; 2349 } 2350 lo->lo_queue = lo->lo_disk->queue; 2351 2352 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2353 2354 /* 2355 * By default, we do buffer IO, so it doesn't make sense to enable 2356 * merge because the I/O submitted to backing file is handled page by 2357 * page. For directio mode, merge does help to dispatch bigger request 2358 * to underlayer disk. We will enable merge once directio is enabled. 2359 */ 2360 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2361 2362 /* 2363 * Disable partition scanning by default. The in-kernel partition 2364 * scanning can be requested individually per-device during its 2365 * setup. Userspace can always add and remove partitions from all 2366 * devices. The needed partition minors are allocated from the 2367 * extended minor space, the main loop device numbers will continue 2368 * to match the loop minors, regardless of the number of partitions 2369 * used. 2370 * 2371 * If max_part is given, partition scanning is globally enabled for 2372 * all loop devices. The minors for the main loop devices will be 2373 * multiples of max_part. 2374 * 2375 * Note: Global-for-all-devices, set-only-at-init, read-only module 2376 * parameteters like 'max_loop' and 'max_part' make things needlessly 2377 * complicated, are too static, inflexible and may surprise 2378 * userspace tools. Parameters like this in general should be avoided. 2379 */ 2380 if (!part_shift) 2381 disk->flags |= GENHD_FL_NO_PART_SCAN; 2382 disk->flags |= GENHD_FL_EXT_DEVT; 2383 atomic_set(&lo->lo_refcnt, 0); 2384 mutex_init(&lo->lo_mutex); 2385 lo->lo_number = i; 2386 spin_lock_init(&lo->lo_lock); 2387 spin_lock_init(&lo->lo_work_lock); 2388 disk->major = LOOP_MAJOR; 2389 disk->first_minor = i << part_shift; 2390 disk->minors = 1 << part_shift; 2391 disk->fops = &lo_fops; 2392 disk->private_data = lo; 2393 disk->queue = lo->lo_queue; 2394 sprintf(disk->disk_name, "loop%d", i); 2395 add_disk(disk); 2396 mutex_unlock(&loop_ctl_mutex); 2397 return i; 2398 2399 out_cleanup_tags: 2400 blk_mq_free_tag_set(&lo->tag_set); 2401 out_free_idr: 2402 idr_remove(&loop_index_idr, i); 2403 out_unlock: 2404 mutex_unlock(&loop_ctl_mutex); 2405 out_free_dev: 2406 kfree(lo); 2407 out: 2408 return err; 2409 } 2410 2411 static void loop_remove(struct loop_device *lo) 2412 { 2413 del_gendisk(lo->lo_disk); 2414 blk_cleanup_disk(lo->lo_disk); 2415 blk_mq_free_tag_set(&lo->tag_set); 2416 mutex_destroy(&lo->lo_mutex); 2417 kfree(lo); 2418 } 2419 2420 static void loop_probe(dev_t dev) 2421 { 2422 int idx = MINOR(dev) >> part_shift; 2423 2424 if (max_loop && idx >= max_loop) 2425 return; 2426 loop_add(idx); 2427 } 2428 2429 static int loop_control_remove(int idx) 2430 { 2431 struct loop_device *lo; 2432 int ret; 2433 2434 if (idx < 0) { 2435 pr_warn("deleting an unspecified loop device is not supported.\n"); 2436 return -EINVAL; 2437 } 2438 2439 ret = mutex_lock_killable(&loop_ctl_mutex); 2440 if (ret) 2441 return ret; 2442 2443 lo = idr_find(&loop_index_idr, idx); 2444 if (!lo) { 2445 ret = -ENODEV; 2446 goto out_unlock_ctrl; 2447 } 2448 2449 ret = mutex_lock_killable(&lo->lo_mutex); 2450 if (ret) 2451 goto out_unlock_ctrl; 2452 if (lo->lo_state != Lo_unbound || 2453 atomic_read(&lo->lo_refcnt) > 0) { 2454 mutex_unlock(&lo->lo_mutex); 2455 ret = -EBUSY; 2456 goto out_unlock_ctrl; 2457 } 2458 lo->lo_state = Lo_deleting; 2459 mutex_unlock(&lo->lo_mutex); 2460 2461 idr_remove(&loop_index_idr, lo->lo_number); 2462 loop_remove(lo); 2463 out_unlock_ctrl: 2464 mutex_unlock(&loop_ctl_mutex); 2465 return ret; 2466 } 2467 2468 static int loop_control_get_free(int idx) 2469 { 2470 struct loop_device *lo; 2471 int id, ret; 2472 2473 ret = mutex_lock_killable(&loop_ctl_mutex); 2474 if (ret) 2475 return ret; 2476 idr_for_each_entry(&loop_index_idr, lo, id) { 2477 if (lo->lo_state == Lo_unbound) 2478 goto found; 2479 } 2480 mutex_unlock(&loop_ctl_mutex); 2481 return loop_add(-1); 2482 found: 2483 mutex_unlock(&loop_ctl_mutex); 2484 return id; 2485 } 2486 2487 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2488 unsigned long parm) 2489 { 2490 switch (cmd) { 2491 case LOOP_CTL_ADD: 2492 return loop_add(parm); 2493 case LOOP_CTL_REMOVE: 2494 return loop_control_remove(parm); 2495 case LOOP_CTL_GET_FREE: 2496 return loop_control_get_free(parm); 2497 default: 2498 return -ENOSYS; 2499 } 2500 } 2501 2502 static const struct file_operations loop_ctl_fops = { 2503 .open = nonseekable_open, 2504 .unlocked_ioctl = loop_control_ioctl, 2505 .compat_ioctl = loop_control_ioctl, 2506 .owner = THIS_MODULE, 2507 .llseek = noop_llseek, 2508 }; 2509 2510 static struct miscdevice loop_misc = { 2511 .minor = LOOP_CTRL_MINOR, 2512 .name = "loop-control", 2513 .fops = &loop_ctl_fops, 2514 }; 2515 2516 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2517 MODULE_ALIAS("devname:loop-control"); 2518 2519 static int __init loop_init(void) 2520 { 2521 int i, nr; 2522 int err; 2523 2524 part_shift = 0; 2525 if (max_part > 0) { 2526 part_shift = fls(max_part); 2527 2528 /* 2529 * Adjust max_part according to part_shift as it is exported 2530 * to user space so that user can decide correct minor number 2531 * if [s]he want to create more devices. 2532 * 2533 * Note that -1 is required because partition 0 is reserved 2534 * for the whole disk. 2535 */ 2536 max_part = (1UL << part_shift) - 1; 2537 } 2538 2539 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2540 err = -EINVAL; 2541 goto err_out; 2542 } 2543 2544 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2545 err = -EINVAL; 2546 goto err_out; 2547 } 2548 2549 /* 2550 * If max_loop is specified, create that many devices upfront. 2551 * This also becomes a hard limit. If max_loop is not specified, 2552 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2553 * init time. Loop devices can be requested on-demand with the 2554 * /dev/loop-control interface, or be instantiated by accessing 2555 * a 'dead' device node. 2556 */ 2557 if (max_loop) 2558 nr = max_loop; 2559 else 2560 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2561 2562 err = misc_register(&loop_misc); 2563 if (err < 0) 2564 goto err_out; 2565 2566 2567 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2568 err = -EIO; 2569 goto misc_out; 2570 } 2571 2572 /* pre-create number of devices given by config or max_loop */ 2573 for (i = 0; i < nr; i++) 2574 loop_add(i); 2575 2576 printk(KERN_INFO "loop: module loaded\n"); 2577 return 0; 2578 2579 misc_out: 2580 misc_deregister(&loop_misc); 2581 err_out: 2582 return err; 2583 } 2584 2585 static void __exit loop_exit(void) 2586 { 2587 struct loop_device *lo; 2588 int id; 2589 2590 unregister_blkdev(LOOP_MAJOR, "loop"); 2591 misc_deregister(&loop_misc); 2592 2593 mutex_lock(&loop_ctl_mutex); 2594 idr_for_each_entry(&loop_index_idr, lo, id) 2595 loop_remove(lo); 2596 mutex_unlock(&loop_ctl_mutex); 2597 2598 idr_destroy(&loop_index_idr); 2599 } 2600 2601 module_init(loop_init); 2602 module_exit(loop_exit); 2603 2604 #ifndef MODULE 2605 static int __init max_loop_setup(char *str) 2606 { 2607 max_loop = simple_strtol(str, NULL, 0); 2608 return 1; 2609 } 2610 2611 __setup("max_loop=", max_loop_setup); 2612 #endif 2613