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