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 "loop.h" 80 81 #include <asm/uaccess.h> 82 83 static DEFINE_IDR(loop_index_idr); 84 static DEFINE_MUTEX(loop_index_mutex); 85 86 static int max_part; 87 static int part_shift; 88 89 static int transfer_xor(struct loop_device *lo, int cmd, 90 struct page *raw_page, unsigned raw_off, 91 struct page *loop_page, unsigned loop_off, 92 int size, sector_t real_block) 93 { 94 char *raw_buf = kmap_atomic(raw_page) + raw_off; 95 char *loop_buf = kmap_atomic(loop_page) + loop_off; 96 char *in, *out, *key; 97 int i, keysize; 98 99 if (cmd == READ) { 100 in = raw_buf; 101 out = loop_buf; 102 } else { 103 in = loop_buf; 104 out = raw_buf; 105 } 106 107 key = lo->lo_encrypt_key; 108 keysize = lo->lo_encrypt_key_size; 109 for (i = 0; i < size; i++) 110 *out++ = *in++ ^ key[(i & 511) % keysize]; 111 112 kunmap_atomic(loop_buf); 113 kunmap_atomic(raw_buf); 114 cond_resched(); 115 return 0; 116 } 117 118 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 119 { 120 if (unlikely(info->lo_encrypt_key_size <= 0)) 121 return -EINVAL; 122 return 0; 123 } 124 125 static struct loop_func_table none_funcs = { 126 .number = LO_CRYPT_NONE, 127 }; 128 129 static struct loop_func_table xor_funcs = { 130 .number = LO_CRYPT_XOR, 131 .transfer = transfer_xor, 132 .init = xor_init 133 }; 134 135 /* xfer_funcs[0] is special - its release function is never called */ 136 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 137 &none_funcs, 138 &xor_funcs 139 }; 140 141 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 142 { 143 loff_t loopsize; 144 145 /* Compute loopsize in bytes */ 146 loopsize = i_size_read(file->f_mapping->host); 147 if (offset > 0) 148 loopsize -= offset; 149 /* offset is beyond i_size, weird but possible */ 150 if (loopsize < 0) 151 return 0; 152 153 if (sizelimit > 0 && sizelimit < loopsize) 154 loopsize = sizelimit; 155 /* 156 * Unfortunately, if we want to do I/O on the device, 157 * the number of 512-byte sectors has to fit into a sector_t. 158 */ 159 return loopsize >> 9; 160 } 161 162 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 163 { 164 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 165 } 166 167 static int 168 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 169 { 170 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 171 sector_t x = (sector_t)size; 172 struct block_device *bdev = lo->lo_device; 173 174 if (unlikely((loff_t)x != size)) 175 return -EFBIG; 176 if (lo->lo_offset != offset) 177 lo->lo_offset = offset; 178 if (lo->lo_sizelimit != sizelimit) 179 lo->lo_sizelimit = sizelimit; 180 set_capacity(lo->lo_disk, x); 181 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 182 /* let user-space know about the new size */ 183 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 184 return 0; 185 } 186 187 static inline int 188 lo_do_transfer(struct loop_device *lo, int cmd, 189 struct page *rpage, unsigned roffs, 190 struct page *lpage, unsigned loffs, 191 int size, sector_t rblock) 192 { 193 int ret; 194 195 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 196 if (likely(!ret)) 197 return 0; 198 199 printk_ratelimited(KERN_ERR 200 "loop: Transfer error at byte offset %llu, length %i.\n", 201 (unsigned long long)rblock << 9, size); 202 return ret; 203 } 204 205 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 206 { 207 struct iov_iter i; 208 ssize_t bw; 209 210 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len); 211 212 file_start_write(file); 213 bw = vfs_iter_write(file, &i, ppos); 214 file_end_write(file); 215 216 if (likely(bw == bvec->bv_len)) 217 return 0; 218 219 printk_ratelimited(KERN_ERR 220 "loop: Write error at byte offset %llu, length %i.\n", 221 (unsigned long long)*ppos, bvec->bv_len); 222 if (bw >= 0) 223 bw = -EIO; 224 return bw; 225 } 226 227 static int lo_write_simple(struct loop_device *lo, struct request *rq, 228 loff_t pos) 229 { 230 struct bio_vec bvec; 231 struct req_iterator iter; 232 int ret = 0; 233 234 rq_for_each_segment(bvec, rq, iter) { 235 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 236 if (ret < 0) 237 break; 238 cond_resched(); 239 } 240 241 return ret; 242 } 243 244 /* 245 * This is the slow, transforming version that needs to double buffer the 246 * data as it cannot do the transformations in place without having direct 247 * access to the destination pages of the backing file. 248 */ 249 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 250 loff_t pos) 251 { 252 struct bio_vec bvec, b; 253 struct req_iterator iter; 254 struct page *page; 255 int ret = 0; 256 257 page = alloc_page(GFP_NOIO); 258 if (unlikely(!page)) 259 return -ENOMEM; 260 261 rq_for_each_segment(bvec, rq, iter) { 262 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 263 bvec.bv_offset, bvec.bv_len, pos >> 9); 264 if (unlikely(ret)) 265 break; 266 267 b.bv_page = page; 268 b.bv_offset = 0; 269 b.bv_len = bvec.bv_len; 270 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 271 if (ret < 0) 272 break; 273 } 274 275 __free_page(page); 276 return ret; 277 } 278 279 static int lo_read_simple(struct loop_device *lo, struct request *rq, 280 loff_t pos) 281 { 282 struct bio_vec bvec; 283 struct req_iterator iter; 284 struct iov_iter i; 285 ssize_t len; 286 287 rq_for_each_segment(bvec, rq, iter) { 288 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); 289 len = vfs_iter_read(lo->lo_backing_file, &i, &pos); 290 if (len < 0) 291 return len; 292 293 flush_dcache_page(bvec.bv_page); 294 295 if (len != bvec.bv_len) { 296 struct bio *bio; 297 298 __rq_for_each_bio(bio, rq) 299 zero_fill_bio(bio); 300 break; 301 } 302 cond_resched(); 303 } 304 305 return 0; 306 } 307 308 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 309 loff_t pos) 310 { 311 struct bio_vec bvec, b; 312 struct req_iterator iter; 313 struct iov_iter i; 314 struct page *page; 315 ssize_t len; 316 int ret = 0; 317 318 page = alloc_page(GFP_NOIO); 319 if (unlikely(!page)) 320 return -ENOMEM; 321 322 rq_for_each_segment(bvec, rq, iter) { 323 loff_t offset = pos; 324 325 b.bv_page = page; 326 b.bv_offset = 0; 327 b.bv_len = bvec.bv_len; 328 329 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); 330 len = vfs_iter_read(lo->lo_backing_file, &i, &pos); 331 if (len < 0) { 332 ret = len; 333 goto out_free_page; 334 } 335 336 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 337 bvec.bv_offset, len, offset >> 9); 338 if (ret) 339 goto out_free_page; 340 341 flush_dcache_page(bvec.bv_page); 342 343 if (len != bvec.bv_len) { 344 struct bio *bio; 345 346 __rq_for_each_bio(bio, rq) 347 zero_fill_bio(bio); 348 break; 349 } 350 } 351 352 ret = 0; 353 out_free_page: 354 __free_page(page); 355 return ret; 356 } 357 358 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 359 { 360 /* 361 * We use punch hole to reclaim the free space used by the 362 * image a.k.a. discard. However we do not support discard if 363 * encryption is enabled, because it may give an attacker 364 * useful information. 365 */ 366 struct file *file = lo->lo_backing_file; 367 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 368 int ret; 369 370 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 371 ret = -EOPNOTSUPP; 372 goto out; 373 } 374 375 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 376 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 377 ret = -EIO; 378 out: 379 return ret; 380 } 381 382 static int lo_req_flush(struct loop_device *lo, struct request *rq) 383 { 384 struct file *file = lo->lo_backing_file; 385 int ret = vfs_fsync(file, 0); 386 if (unlikely(ret && ret != -EINVAL)) 387 ret = -EIO; 388 389 return ret; 390 } 391 392 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 393 { 394 loff_t pos; 395 int ret; 396 397 pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 398 399 if (rq->cmd_flags & REQ_WRITE) { 400 if (rq->cmd_flags & REQ_FLUSH) 401 ret = lo_req_flush(lo, rq); 402 else if (rq->cmd_flags & REQ_DISCARD) 403 ret = lo_discard(lo, rq, pos); 404 else if (lo->transfer) 405 ret = lo_write_transfer(lo, rq, pos); 406 else 407 ret = lo_write_simple(lo, rq, pos); 408 409 } else { 410 if (lo->transfer) 411 ret = lo_read_transfer(lo, rq, pos); 412 else 413 ret = lo_read_simple(lo, rq, pos); 414 } 415 416 return ret; 417 } 418 419 struct switch_request { 420 struct file *file; 421 struct completion wait; 422 }; 423 424 /* 425 * Do the actual switch; called from the BIO completion routine 426 */ 427 static void do_loop_switch(struct loop_device *lo, struct switch_request *p) 428 { 429 struct file *file = p->file; 430 struct file *old_file = lo->lo_backing_file; 431 struct address_space *mapping; 432 433 /* if no new file, only flush of queued bios requested */ 434 if (!file) 435 return; 436 437 mapping = file->f_mapping; 438 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 439 lo->lo_backing_file = file; 440 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? 441 mapping->host->i_bdev->bd_block_size : PAGE_SIZE; 442 lo->old_gfp_mask = mapping_gfp_mask(mapping); 443 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 444 } 445 446 /* 447 * loop_switch performs the hard work of switching a backing store. 448 * First it needs to flush existing IO, it does this by sending a magic 449 * BIO down the pipe. The completion of this BIO does the actual switch. 450 */ 451 static int loop_switch(struct loop_device *lo, struct file *file) 452 { 453 struct switch_request w; 454 455 w.file = file; 456 457 /* freeze queue and wait for completion of scheduled requests */ 458 blk_mq_freeze_queue(lo->lo_queue); 459 460 /* do the switch action */ 461 do_loop_switch(lo, &w); 462 463 /* unfreeze */ 464 blk_mq_unfreeze_queue(lo->lo_queue); 465 466 return 0; 467 } 468 469 /* 470 * Helper to flush the IOs in loop, but keeping loop thread running 471 */ 472 static int loop_flush(struct loop_device *lo) 473 { 474 return loop_switch(lo, NULL); 475 } 476 477 static void loop_reread_partitions(struct loop_device *lo, 478 struct block_device *bdev) 479 { 480 int rc; 481 482 /* 483 * bd_mutex has been held already in release path, so don't 484 * acquire it if this function is called in such case. 485 * 486 * If the reread partition isn't from release path, lo_refcnt 487 * must be at least one and it can only become zero when the 488 * current holder is released. 489 */ 490 if (!atomic_read(&lo->lo_refcnt)) 491 rc = __blkdev_reread_part(bdev); 492 else 493 rc = blkdev_reread_part(bdev); 494 if (rc) 495 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 496 __func__, lo->lo_number, lo->lo_file_name, rc); 497 } 498 499 /* 500 * loop_change_fd switched the backing store of a loopback device to 501 * a new file. This is useful for operating system installers to free up 502 * the original file and in High Availability environments to switch to 503 * an alternative location for the content in case of server meltdown. 504 * This can only work if the loop device is used read-only, and if the 505 * new backing store is the same size and type as the old backing store. 506 */ 507 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 508 unsigned int arg) 509 { 510 struct file *file, *old_file; 511 struct inode *inode; 512 int error; 513 514 error = -ENXIO; 515 if (lo->lo_state != Lo_bound) 516 goto out; 517 518 /* the loop device has to be read-only */ 519 error = -EINVAL; 520 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 521 goto out; 522 523 error = -EBADF; 524 file = fget(arg); 525 if (!file) 526 goto out; 527 528 inode = file->f_mapping->host; 529 old_file = lo->lo_backing_file; 530 531 error = -EINVAL; 532 533 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 534 goto out_putf; 535 536 /* size of the new backing store needs to be the same */ 537 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 538 goto out_putf; 539 540 /* and ... switch */ 541 error = loop_switch(lo, file); 542 if (error) 543 goto out_putf; 544 545 fput(old_file); 546 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 547 loop_reread_partitions(lo, bdev); 548 return 0; 549 550 out_putf: 551 fput(file); 552 out: 553 return error; 554 } 555 556 static inline int is_loop_device(struct file *file) 557 { 558 struct inode *i = file->f_mapping->host; 559 560 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 561 } 562 563 /* loop sysfs attributes */ 564 565 static ssize_t loop_attr_show(struct device *dev, char *page, 566 ssize_t (*callback)(struct loop_device *, char *)) 567 { 568 struct gendisk *disk = dev_to_disk(dev); 569 struct loop_device *lo = disk->private_data; 570 571 return callback(lo, page); 572 } 573 574 #define LOOP_ATTR_RO(_name) \ 575 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 576 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 577 struct device_attribute *attr, char *b) \ 578 { \ 579 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 580 } \ 581 static struct device_attribute loop_attr_##_name = \ 582 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); 583 584 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 585 { 586 ssize_t ret; 587 char *p = NULL; 588 589 spin_lock_irq(&lo->lo_lock); 590 if (lo->lo_backing_file) 591 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 592 spin_unlock_irq(&lo->lo_lock); 593 594 if (IS_ERR_OR_NULL(p)) 595 ret = PTR_ERR(p); 596 else { 597 ret = strlen(p); 598 memmove(buf, p, ret); 599 buf[ret++] = '\n'; 600 buf[ret] = 0; 601 } 602 603 return ret; 604 } 605 606 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 607 { 608 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 609 } 610 611 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 612 { 613 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 614 } 615 616 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 617 { 618 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 619 620 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 621 } 622 623 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 624 { 625 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 626 627 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 628 } 629 630 LOOP_ATTR_RO(backing_file); 631 LOOP_ATTR_RO(offset); 632 LOOP_ATTR_RO(sizelimit); 633 LOOP_ATTR_RO(autoclear); 634 LOOP_ATTR_RO(partscan); 635 636 static struct attribute *loop_attrs[] = { 637 &loop_attr_backing_file.attr, 638 &loop_attr_offset.attr, 639 &loop_attr_sizelimit.attr, 640 &loop_attr_autoclear.attr, 641 &loop_attr_partscan.attr, 642 NULL, 643 }; 644 645 static struct attribute_group loop_attribute_group = { 646 .name = "loop", 647 .attrs= loop_attrs, 648 }; 649 650 static int loop_sysfs_init(struct loop_device *lo) 651 { 652 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 653 &loop_attribute_group); 654 } 655 656 static void loop_sysfs_exit(struct loop_device *lo) 657 { 658 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 659 &loop_attribute_group); 660 } 661 662 static void loop_config_discard(struct loop_device *lo) 663 { 664 struct file *file = lo->lo_backing_file; 665 struct inode *inode = file->f_mapping->host; 666 struct request_queue *q = lo->lo_queue; 667 668 /* 669 * We use punch hole to reclaim the free space used by the 670 * image a.k.a. discard. However we do not support discard if 671 * encryption is enabled, because it may give an attacker 672 * useful information. 673 */ 674 if ((!file->f_op->fallocate) || 675 lo->lo_encrypt_key_size) { 676 q->limits.discard_granularity = 0; 677 q->limits.discard_alignment = 0; 678 blk_queue_max_discard_sectors(q, 0); 679 q->limits.discard_zeroes_data = 0; 680 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 681 return; 682 } 683 684 q->limits.discard_granularity = inode->i_sb->s_blocksize; 685 q->limits.discard_alignment = 0; 686 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 687 q->limits.discard_zeroes_data = 1; 688 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 689 } 690 691 static void loop_unprepare_queue(struct loop_device *lo) 692 { 693 flush_kthread_worker(&lo->worker); 694 kthread_stop(lo->worker_task); 695 } 696 697 static int loop_prepare_queue(struct loop_device *lo) 698 { 699 init_kthread_worker(&lo->worker); 700 lo->worker_task = kthread_run(kthread_worker_fn, 701 &lo->worker, "loop%d", lo->lo_number); 702 if (IS_ERR(lo->worker_task)) 703 return -ENOMEM; 704 set_user_nice(lo->worker_task, MIN_NICE); 705 return 0; 706 } 707 708 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 709 struct block_device *bdev, unsigned int arg) 710 { 711 struct file *file, *f; 712 struct inode *inode; 713 struct address_space *mapping; 714 unsigned lo_blocksize; 715 int lo_flags = 0; 716 int error; 717 loff_t size; 718 719 /* This is safe, since we have a reference from open(). */ 720 __module_get(THIS_MODULE); 721 722 error = -EBADF; 723 file = fget(arg); 724 if (!file) 725 goto out; 726 727 error = -EBUSY; 728 if (lo->lo_state != Lo_unbound) 729 goto out_putf; 730 731 /* Avoid recursion */ 732 f = file; 733 while (is_loop_device(f)) { 734 struct loop_device *l; 735 736 if (f->f_mapping->host->i_bdev == bdev) 737 goto out_putf; 738 739 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 740 if (l->lo_state == Lo_unbound) { 741 error = -EINVAL; 742 goto out_putf; 743 } 744 f = l->lo_backing_file; 745 } 746 747 mapping = file->f_mapping; 748 inode = mapping->host; 749 750 error = -EINVAL; 751 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 752 goto out_putf; 753 754 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 755 !file->f_op->write_iter) 756 lo_flags |= LO_FLAGS_READ_ONLY; 757 758 lo_blocksize = S_ISBLK(inode->i_mode) ? 759 inode->i_bdev->bd_block_size : PAGE_SIZE; 760 761 error = -EFBIG; 762 size = get_loop_size(lo, file); 763 if ((loff_t)(sector_t)size != size) 764 goto out_putf; 765 error = loop_prepare_queue(lo); 766 if (error) 767 goto out_putf; 768 769 error = 0; 770 771 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 772 773 lo->lo_blocksize = lo_blocksize; 774 lo->lo_device = bdev; 775 lo->lo_flags = lo_flags; 776 lo->lo_backing_file = file; 777 lo->transfer = NULL; 778 lo->ioctl = NULL; 779 lo->lo_sizelimit = 0; 780 lo->old_gfp_mask = mapping_gfp_mask(mapping); 781 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 782 783 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 784 blk_queue_flush(lo->lo_queue, REQ_FLUSH); 785 786 set_capacity(lo->lo_disk, size); 787 bd_set_size(bdev, size << 9); 788 loop_sysfs_init(lo); 789 /* let user-space know about the new size */ 790 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 791 792 set_blocksize(bdev, lo_blocksize); 793 794 lo->lo_state = Lo_bound; 795 if (part_shift) 796 lo->lo_flags |= LO_FLAGS_PARTSCAN; 797 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 798 loop_reread_partitions(lo, bdev); 799 800 /* Grab the block_device to prevent its destruction after we 801 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 802 */ 803 bdgrab(bdev); 804 return 0; 805 806 out_putf: 807 fput(file); 808 out: 809 /* This is safe: open() is still holding a reference. */ 810 module_put(THIS_MODULE); 811 return error; 812 } 813 814 static int 815 loop_release_xfer(struct loop_device *lo) 816 { 817 int err = 0; 818 struct loop_func_table *xfer = lo->lo_encryption; 819 820 if (xfer) { 821 if (xfer->release) 822 err = xfer->release(lo); 823 lo->transfer = NULL; 824 lo->lo_encryption = NULL; 825 module_put(xfer->owner); 826 } 827 return err; 828 } 829 830 static int 831 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 832 const struct loop_info64 *i) 833 { 834 int err = 0; 835 836 if (xfer) { 837 struct module *owner = xfer->owner; 838 839 if (!try_module_get(owner)) 840 return -EINVAL; 841 if (xfer->init) 842 err = xfer->init(lo, i); 843 if (err) 844 module_put(owner); 845 else 846 lo->lo_encryption = xfer; 847 } 848 return err; 849 } 850 851 static int loop_clr_fd(struct loop_device *lo) 852 { 853 struct file *filp = lo->lo_backing_file; 854 gfp_t gfp = lo->old_gfp_mask; 855 struct block_device *bdev = lo->lo_device; 856 857 if (lo->lo_state != Lo_bound) 858 return -ENXIO; 859 860 /* 861 * If we've explicitly asked to tear down the loop device, 862 * and it has an elevated reference count, set it for auto-teardown when 863 * the last reference goes away. This stops $!~#$@ udev from 864 * preventing teardown because it decided that it needs to run blkid on 865 * the loopback device whenever they appear. xfstests is notorious for 866 * failing tests because blkid via udev races with a losetup 867 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 868 * command to fail with EBUSY. 869 */ 870 if (atomic_read(&lo->lo_refcnt) > 1) { 871 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 872 mutex_unlock(&lo->lo_ctl_mutex); 873 return 0; 874 } 875 876 if (filp == NULL) 877 return -EINVAL; 878 879 /* freeze request queue during the transition */ 880 blk_mq_freeze_queue(lo->lo_queue); 881 882 spin_lock_irq(&lo->lo_lock); 883 lo->lo_state = Lo_rundown; 884 lo->lo_backing_file = NULL; 885 spin_unlock_irq(&lo->lo_lock); 886 887 loop_release_xfer(lo); 888 lo->transfer = NULL; 889 lo->ioctl = NULL; 890 lo->lo_device = NULL; 891 lo->lo_encryption = NULL; 892 lo->lo_offset = 0; 893 lo->lo_sizelimit = 0; 894 lo->lo_encrypt_key_size = 0; 895 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 896 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 897 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 898 if (bdev) { 899 bdput(bdev); 900 invalidate_bdev(bdev); 901 } 902 set_capacity(lo->lo_disk, 0); 903 loop_sysfs_exit(lo); 904 if (bdev) { 905 bd_set_size(bdev, 0); 906 /* let user-space know about this change */ 907 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 908 } 909 mapping_set_gfp_mask(filp->f_mapping, gfp); 910 lo->lo_state = Lo_unbound; 911 /* This is safe: open() is still holding a reference. */ 912 module_put(THIS_MODULE); 913 blk_mq_unfreeze_queue(lo->lo_queue); 914 915 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 916 loop_reread_partitions(lo, bdev); 917 lo->lo_flags = 0; 918 if (!part_shift) 919 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 920 loop_unprepare_queue(lo); 921 mutex_unlock(&lo->lo_ctl_mutex); 922 /* 923 * Need not hold lo_ctl_mutex to fput backing file. 924 * Calling fput holding lo_ctl_mutex triggers a circular 925 * lock dependency possibility warning as fput can take 926 * bd_mutex which is usually taken before lo_ctl_mutex. 927 */ 928 fput(filp); 929 return 0; 930 } 931 932 static int 933 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 934 { 935 int err; 936 struct loop_func_table *xfer; 937 kuid_t uid = current_uid(); 938 939 if (lo->lo_encrypt_key_size && 940 !uid_eq(lo->lo_key_owner, uid) && 941 !capable(CAP_SYS_ADMIN)) 942 return -EPERM; 943 if (lo->lo_state != Lo_bound) 944 return -ENXIO; 945 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 946 return -EINVAL; 947 948 err = loop_release_xfer(lo); 949 if (err) 950 return err; 951 952 if (info->lo_encrypt_type) { 953 unsigned int type = info->lo_encrypt_type; 954 955 if (type >= MAX_LO_CRYPT) 956 return -EINVAL; 957 xfer = xfer_funcs[type]; 958 if (xfer == NULL) 959 return -EINVAL; 960 } else 961 xfer = NULL; 962 963 err = loop_init_xfer(lo, xfer, info); 964 if (err) 965 return err; 966 967 if (lo->lo_offset != info->lo_offset || 968 lo->lo_sizelimit != info->lo_sizelimit) 969 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) 970 return -EFBIG; 971 972 loop_config_discard(lo); 973 974 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 975 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 976 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 977 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 978 979 if (!xfer) 980 xfer = &none_funcs; 981 lo->transfer = xfer->transfer; 982 lo->ioctl = xfer->ioctl; 983 984 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 985 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 986 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 987 988 if ((info->lo_flags & LO_FLAGS_PARTSCAN) && 989 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 990 lo->lo_flags |= LO_FLAGS_PARTSCAN; 991 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 992 loop_reread_partitions(lo, lo->lo_device); 993 } 994 995 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 996 lo->lo_init[0] = info->lo_init[0]; 997 lo->lo_init[1] = info->lo_init[1]; 998 if (info->lo_encrypt_key_size) { 999 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1000 info->lo_encrypt_key_size); 1001 lo->lo_key_owner = uid; 1002 } 1003 1004 return 0; 1005 } 1006 1007 static int 1008 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1009 { 1010 struct file *file = lo->lo_backing_file; 1011 struct kstat stat; 1012 int error; 1013 1014 if (lo->lo_state != Lo_bound) 1015 return -ENXIO; 1016 error = vfs_getattr(&file->f_path, &stat); 1017 if (error) 1018 return error; 1019 memset(info, 0, sizeof(*info)); 1020 info->lo_number = lo->lo_number; 1021 info->lo_device = huge_encode_dev(stat.dev); 1022 info->lo_inode = stat.ino; 1023 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); 1024 info->lo_offset = lo->lo_offset; 1025 info->lo_sizelimit = lo->lo_sizelimit; 1026 info->lo_flags = lo->lo_flags; 1027 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1028 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1029 info->lo_encrypt_type = 1030 lo->lo_encryption ? lo->lo_encryption->number : 0; 1031 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1032 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1033 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1034 lo->lo_encrypt_key_size); 1035 } 1036 return 0; 1037 } 1038 1039 static void 1040 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1041 { 1042 memset(info64, 0, sizeof(*info64)); 1043 info64->lo_number = info->lo_number; 1044 info64->lo_device = info->lo_device; 1045 info64->lo_inode = info->lo_inode; 1046 info64->lo_rdevice = info->lo_rdevice; 1047 info64->lo_offset = info->lo_offset; 1048 info64->lo_sizelimit = 0; 1049 info64->lo_encrypt_type = info->lo_encrypt_type; 1050 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1051 info64->lo_flags = info->lo_flags; 1052 info64->lo_init[0] = info->lo_init[0]; 1053 info64->lo_init[1] = info->lo_init[1]; 1054 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1055 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1056 else 1057 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1058 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1059 } 1060 1061 static int 1062 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1063 { 1064 memset(info, 0, sizeof(*info)); 1065 info->lo_number = info64->lo_number; 1066 info->lo_device = info64->lo_device; 1067 info->lo_inode = info64->lo_inode; 1068 info->lo_rdevice = info64->lo_rdevice; 1069 info->lo_offset = info64->lo_offset; 1070 info->lo_encrypt_type = info64->lo_encrypt_type; 1071 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1072 info->lo_flags = info64->lo_flags; 1073 info->lo_init[0] = info64->lo_init[0]; 1074 info->lo_init[1] = info64->lo_init[1]; 1075 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1076 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1077 else 1078 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1079 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1080 1081 /* error in case values were truncated */ 1082 if (info->lo_device != info64->lo_device || 1083 info->lo_rdevice != info64->lo_rdevice || 1084 info->lo_inode != info64->lo_inode || 1085 info->lo_offset != info64->lo_offset) 1086 return -EOVERFLOW; 1087 1088 return 0; 1089 } 1090 1091 static int 1092 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1093 { 1094 struct loop_info info; 1095 struct loop_info64 info64; 1096 1097 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1098 return -EFAULT; 1099 loop_info64_from_old(&info, &info64); 1100 return loop_set_status(lo, &info64); 1101 } 1102 1103 static int 1104 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1105 { 1106 struct loop_info64 info64; 1107 1108 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1109 return -EFAULT; 1110 return loop_set_status(lo, &info64); 1111 } 1112 1113 static int 1114 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1115 struct loop_info info; 1116 struct loop_info64 info64; 1117 int err = 0; 1118 1119 if (!arg) 1120 err = -EINVAL; 1121 if (!err) 1122 err = loop_get_status(lo, &info64); 1123 if (!err) 1124 err = loop_info64_to_old(&info64, &info); 1125 if (!err && copy_to_user(arg, &info, sizeof(info))) 1126 err = -EFAULT; 1127 1128 return err; 1129 } 1130 1131 static int 1132 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1133 struct loop_info64 info64; 1134 int err = 0; 1135 1136 if (!arg) 1137 err = -EINVAL; 1138 if (!err) 1139 err = loop_get_status(lo, &info64); 1140 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1141 err = -EFAULT; 1142 1143 return err; 1144 } 1145 1146 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) 1147 { 1148 if (unlikely(lo->lo_state != Lo_bound)) 1149 return -ENXIO; 1150 1151 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1152 } 1153 1154 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1155 unsigned int cmd, unsigned long arg) 1156 { 1157 struct loop_device *lo = bdev->bd_disk->private_data; 1158 int err; 1159 1160 mutex_lock_nested(&lo->lo_ctl_mutex, 1); 1161 switch (cmd) { 1162 case LOOP_SET_FD: 1163 err = loop_set_fd(lo, mode, bdev, arg); 1164 break; 1165 case LOOP_CHANGE_FD: 1166 err = loop_change_fd(lo, bdev, arg); 1167 break; 1168 case LOOP_CLR_FD: 1169 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1170 err = loop_clr_fd(lo); 1171 if (!err) 1172 goto out_unlocked; 1173 break; 1174 case LOOP_SET_STATUS: 1175 err = -EPERM; 1176 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1177 err = loop_set_status_old(lo, 1178 (struct loop_info __user *)arg); 1179 break; 1180 case LOOP_GET_STATUS: 1181 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1182 break; 1183 case LOOP_SET_STATUS64: 1184 err = -EPERM; 1185 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1186 err = loop_set_status64(lo, 1187 (struct loop_info64 __user *) arg); 1188 break; 1189 case LOOP_GET_STATUS64: 1190 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1191 break; 1192 case LOOP_SET_CAPACITY: 1193 err = -EPERM; 1194 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1195 err = loop_set_capacity(lo, bdev); 1196 break; 1197 default: 1198 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1199 } 1200 mutex_unlock(&lo->lo_ctl_mutex); 1201 1202 out_unlocked: 1203 return err; 1204 } 1205 1206 #ifdef CONFIG_COMPAT 1207 struct compat_loop_info { 1208 compat_int_t lo_number; /* ioctl r/o */ 1209 compat_dev_t lo_device; /* ioctl r/o */ 1210 compat_ulong_t lo_inode; /* ioctl r/o */ 1211 compat_dev_t lo_rdevice; /* ioctl r/o */ 1212 compat_int_t lo_offset; 1213 compat_int_t lo_encrypt_type; 1214 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1215 compat_int_t lo_flags; /* ioctl r/o */ 1216 char lo_name[LO_NAME_SIZE]; 1217 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1218 compat_ulong_t lo_init[2]; 1219 char reserved[4]; 1220 }; 1221 1222 /* 1223 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1224 * - noinlined to reduce stack space usage in main part of driver 1225 */ 1226 static noinline int 1227 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1228 struct loop_info64 *info64) 1229 { 1230 struct compat_loop_info info; 1231 1232 if (copy_from_user(&info, arg, sizeof(info))) 1233 return -EFAULT; 1234 1235 memset(info64, 0, sizeof(*info64)); 1236 info64->lo_number = info.lo_number; 1237 info64->lo_device = info.lo_device; 1238 info64->lo_inode = info.lo_inode; 1239 info64->lo_rdevice = info.lo_rdevice; 1240 info64->lo_offset = info.lo_offset; 1241 info64->lo_sizelimit = 0; 1242 info64->lo_encrypt_type = info.lo_encrypt_type; 1243 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1244 info64->lo_flags = info.lo_flags; 1245 info64->lo_init[0] = info.lo_init[0]; 1246 info64->lo_init[1] = info.lo_init[1]; 1247 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1248 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1249 else 1250 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1251 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1252 return 0; 1253 } 1254 1255 /* 1256 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1257 * - noinlined to reduce stack space usage in main part of driver 1258 */ 1259 static noinline int 1260 loop_info64_to_compat(const struct loop_info64 *info64, 1261 struct compat_loop_info __user *arg) 1262 { 1263 struct compat_loop_info info; 1264 1265 memset(&info, 0, sizeof(info)); 1266 info.lo_number = info64->lo_number; 1267 info.lo_device = info64->lo_device; 1268 info.lo_inode = info64->lo_inode; 1269 info.lo_rdevice = info64->lo_rdevice; 1270 info.lo_offset = info64->lo_offset; 1271 info.lo_encrypt_type = info64->lo_encrypt_type; 1272 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1273 info.lo_flags = info64->lo_flags; 1274 info.lo_init[0] = info64->lo_init[0]; 1275 info.lo_init[1] = info64->lo_init[1]; 1276 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1277 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1278 else 1279 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1280 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1281 1282 /* error in case values were truncated */ 1283 if (info.lo_device != info64->lo_device || 1284 info.lo_rdevice != info64->lo_rdevice || 1285 info.lo_inode != info64->lo_inode || 1286 info.lo_offset != info64->lo_offset || 1287 info.lo_init[0] != info64->lo_init[0] || 1288 info.lo_init[1] != info64->lo_init[1]) 1289 return -EOVERFLOW; 1290 1291 if (copy_to_user(arg, &info, sizeof(info))) 1292 return -EFAULT; 1293 return 0; 1294 } 1295 1296 static int 1297 loop_set_status_compat(struct loop_device *lo, 1298 const struct compat_loop_info __user *arg) 1299 { 1300 struct loop_info64 info64; 1301 int ret; 1302 1303 ret = loop_info64_from_compat(arg, &info64); 1304 if (ret < 0) 1305 return ret; 1306 return loop_set_status(lo, &info64); 1307 } 1308 1309 static int 1310 loop_get_status_compat(struct loop_device *lo, 1311 struct compat_loop_info __user *arg) 1312 { 1313 struct loop_info64 info64; 1314 int err = 0; 1315 1316 if (!arg) 1317 err = -EINVAL; 1318 if (!err) 1319 err = loop_get_status(lo, &info64); 1320 if (!err) 1321 err = loop_info64_to_compat(&info64, arg); 1322 return err; 1323 } 1324 1325 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1326 unsigned int cmd, unsigned long arg) 1327 { 1328 struct loop_device *lo = bdev->bd_disk->private_data; 1329 int err; 1330 1331 switch(cmd) { 1332 case LOOP_SET_STATUS: 1333 mutex_lock(&lo->lo_ctl_mutex); 1334 err = loop_set_status_compat( 1335 lo, (const struct compat_loop_info __user *) arg); 1336 mutex_unlock(&lo->lo_ctl_mutex); 1337 break; 1338 case LOOP_GET_STATUS: 1339 mutex_lock(&lo->lo_ctl_mutex); 1340 err = loop_get_status_compat( 1341 lo, (struct compat_loop_info __user *) arg); 1342 mutex_unlock(&lo->lo_ctl_mutex); 1343 break; 1344 case LOOP_SET_CAPACITY: 1345 case LOOP_CLR_FD: 1346 case LOOP_GET_STATUS64: 1347 case LOOP_SET_STATUS64: 1348 arg = (unsigned long) compat_ptr(arg); 1349 case LOOP_SET_FD: 1350 case LOOP_CHANGE_FD: 1351 err = lo_ioctl(bdev, mode, cmd, arg); 1352 break; 1353 default: 1354 err = -ENOIOCTLCMD; 1355 break; 1356 } 1357 return err; 1358 } 1359 #endif 1360 1361 static int lo_open(struct block_device *bdev, fmode_t mode) 1362 { 1363 struct loop_device *lo; 1364 int err = 0; 1365 1366 mutex_lock(&loop_index_mutex); 1367 lo = bdev->bd_disk->private_data; 1368 if (!lo) { 1369 err = -ENXIO; 1370 goto out; 1371 } 1372 1373 atomic_inc(&lo->lo_refcnt); 1374 out: 1375 mutex_unlock(&loop_index_mutex); 1376 return err; 1377 } 1378 1379 static void lo_release(struct gendisk *disk, fmode_t mode) 1380 { 1381 struct loop_device *lo = disk->private_data; 1382 int err; 1383 1384 if (atomic_dec_return(&lo->lo_refcnt)) 1385 return; 1386 1387 mutex_lock(&lo->lo_ctl_mutex); 1388 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1389 /* 1390 * In autoclear mode, stop the loop thread 1391 * and remove configuration after last close. 1392 */ 1393 err = loop_clr_fd(lo); 1394 if (!err) 1395 return; 1396 } else { 1397 /* 1398 * Otherwise keep thread (if running) and config, 1399 * but flush possible ongoing bios in thread. 1400 */ 1401 loop_flush(lo); 1402 } 1403 1404 mutex_unlock(&lo->lo_ctl_mutex); 1405 } 1406 1407 static const struct block_device_operations lo_fops = { 1408 .owner = THIS_MODULE, 1409 .open = lo_open, 1410 .release = lo_release, 1411 .ioctl = lo_ioctl, 1412 #ifdef CONFIG_COMPAT 1413 .compat_ioctl = lo_compat_ioctl, 1414 #endif 1415 }; 1416 1417 /* 1418 * And now the modules code and kernel interface. 1419 */ 1420 static int max_loop; 1421 module_param(max_loop, int, S_IRUGO); 1422 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1423 module_param(max_part, int, S_IRUGO); 1424 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1425 MODULE_LICENSE("GPL"); 1426 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1427 1428 int loop_register_transfer(struct loop_func_table *funcs) 1429 { 1430 unsigned int n = funcs->number; 1431 1432 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1433 return -EINVAL; 1434 xfer_funcs[n] = funcs; 1435 return 0; 1436 } 1437 1438 static int unregister_transfer_cb(int id, void *ptr, void *data) 1439 { 1440 struct loop_device *lo = ptr; 1441 struct loop_func_table *xfer = data; 1442 1443 mutex_lock(&lo->lo_ctl_mutex); 1444 if (lo->lo_encryption == xfer) 1445 loop_release_xfer(lo); 1446 mutex_unlock(&lo->lo_ctl_mutex); 1447 return 0; 1448 } 1449 1450 int loop_unregister_transfer(int number) 1451 { 1452 unsigned int n = number; 1453 struct loop_func_table *xfer; 1454 1455 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1456 return -EINVAL; 1457 1458 xfer_funcs[n] = NULL; 1459 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1460 return 0; 1461 } 1462 1463 EXPORT_SYMBOL(loop_register_transfer); 1464 EXPORT_SYMBOL(loop_unregister_transfer); 1465 1466 static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1467 const struct blk_mq_queue_data *bd) 1468 { 1469 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); 1470 struct loop_device *lo = cmd->rq->q->queuedata; 1471 1472 blk_mq_start_request(bd->rq); 1473 1474 if (lo->lo_state != Lo_bound) 1475 return -EIO; 1476 1477 queue_kthread_work(&lo->worker, &cmd->work); 1478 1479 return BLK_MQ_RQ_QUEUE_OK; 1480 } 1481 1482 static void loop_handle_cmd(struct loop_cmd *cmd) 1483 { 1484 const bool write = cmd->rq->cmd_flags & REQ_WRITE; 1485 struct loop_device *lo = cmd->rq->q->queuedata; 1486 int ret = -EIO; 1487 1488 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) 1489 goto failed; 1490 1491 ret = do_req_filebacked(lo, cmd->rq); 1492 1493 failed: 1494 if (ret) 1495 cmd->rq->errors = -EIO; 1496 blk_mq_complete_request(cmd->rq); 1497 } 1498 1499 static void loop_queue_work(struct kthread_work *work) 1500 { 1501 struct loop_cmd *cmd = 1502 container_of(work, struct loop_cmd, work); 1503 1504 loop_handle_cmd(cmd); 1505 } 1506 1507 static int loop_init_request(void *data, struct request *rq, 1508 unsigned int hctx_idx, unsigned int request_idx, 1509 unsigned int numa_node) 1510 { 1511 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1512 1513 cmd->rq = rq; 1514 init_kthread_work(&cmd->work, loop_queue_work); 1515 1516 return 0; 1517 } 1518 1519 static struct blk_mq_ops loop_mq_ops = { 1520 .queue_rq = loop_queue_rq, 1521 .map_queue = blk_mq_map_queue, 1522 .init_request = loop_init_request, 1523 }; 1524 1525 static int loop_add(struct loop_device **l, int i) 1526 { 1527 struct loop_device *lo; 1528 struct gendisk *disk; 1529 int err; 1530 1531 err = -ENOMEM; 1532 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1533 if (!lo) 1534 goto out; 1535 1536 lo->lo_state = Lo_unbound; 1537 1538 /* allocate id, if @id >= 0, we're requesting that specific id */ 1539 if (i >= 0) { 1540 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1541 if (err == -ENOSPC) 1542 err = -EEXIST; 1543 } else { 1544 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1545 } 1546 if (err < 0) 1547 goto out_free_dev; 1548 i = err; 1549 1550 err = -ENOMEM; 1551 lo->tag_set.ops = &loop_mq_ops; 1552 lo->tag_set.nr_hw_queues = 1; 1553 lo->tag_set.queue_depth = 128; 1554 lo->tag_set.numa_node = NUMA_NO_NODE; 1555 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1556 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1557 lo->tag_set.driver_data = lo; 1558 1559 err = blk_mq_alloc_tag_set(&lo->tag_set); 1560 if (err) 1561 goto out_free_idr; 1562 1563 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1564 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1565 err = PTR_ERR(lo->lo_queue); 1566 goto out_cleanup_tags; 1567 } 1568 lo->lo_queue->queuedata = lo; 1569 1570 /* 1571 * It doesn't make sense to enable merge because the I/O 1572 * submitted to backing file is handled page by page. 1573 */ 1574 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1575 1576 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1577 if (!disk) 1578 goto out_free_queue; 1579 1580 /* 1581 * Disable partition scanning by default. The in-kernel partition 1582 * scanning can be requested individually per-device during its 1583 * setup. Userspace can always add and remove partitions from all 1584 * devices. The needed partition minors are allocated from the 1585 * extended minor space, the main loop device numbers will continue 1586 * to match the loop minors, regardless of the number of partitions 1587 * used. 1588 * 1589 * If max_part is given, partition scanning is globally enabled for 1590 * all loop devices. The minors for the main loop devices will be 1591 * multiples of max_part. 1592 * 1593 * Note: Global-for-all-devices, set-only-at-init, read-only module 1594 * parameteters like 'max_loop' and 'max_part' make things needlessly 1595 * complicated, are too static, inflexible and may surprise 1596 * userspace tools. Parameters like this in general should be avoided. 1597 */ 1598 if (!part_shift) 1599 disk->flags |= GENHD_FL_NO_PART_SCAN; 1600 disk->flags |= GENHD_FL_EXT_DEVT; 1601 mutex_init(&lo->lo_ctl_mutex); 1602 atomic_set(&lo->lo_refcnt, 0); 1603 lo->lo_number = i; 1604 spin_lock_init(&lo->lo_lock); 1605 disk->major = LOOP_MAJOR; 1606 disk->first_minor = i << part_shift; 1607 disk->fops = &lo_fops; 1608 disk->private_data = lo; 1609 disk->queue = lo->lo_queue; 1610 sprintf(disk->disk_name, "loop%d", i); 1611 add_disk(disk); 1612 *l = lo; 1613 return lo->lo_number; 1614 1615 out_free_queue: 1616 blk_cleanup_queue(lo->lo_queue); 1617 out_cleanup_tags: 1618 blk_mq_free_tag_set(&lo->tag_set); 1619 out_free_idr: 1620 idr_remove(&loop_index_idr, i); 1621 out_free_dev: 1622 kfree(lo); 1623 out: 1624 return err; 1625 } 1626 1627 static void loop_remove(struct loop_device *lo) 1628 { 1629 blk_cleanup_queue(lo->lo_queue); 1630 del_gendisk(lo->lo_disk); 1631 blk_mq_free_tag_set(&lo->tag_set); 1632 put_disk(lo->lo_disk); 1633 kfree(lo); 1634 } 1635 1636 static int find_free_cb(int id, void *ptr, void *data) 1637 { 1638 struct loop_device *lo = ptr; 1639 struct loop_device **l = data; 1640 1641 if (lo->lo_state == Lo_unbound) { 1642 *l = lo; 1643 return 1; 1644 } 1645 return 0; 1646 } 1647 1648 static int loop_lookup(struct loop_device **l, int i) 1649 { 1650 struct loop_device *lo; 1651 int ret = -ENODEV; 1652 1653 if (i < 0) { 1654 int err; 1655 1656 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1657 if (err == 1) { 1658 *l = lo; 1659 ret = lo->lo_number; 1660 } 1661 goto out; 1662 } 1663 1664 /* lookup and return a specific i */ 1665 lo = idr_find(&loop_index_idr, i); 1666 if (lo) { 1667 *l = lo; 1668 ret = lo->lo_number; 1669 } 1670 out: 1671 return ret; 1672 } 1673 1674 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1675 { 1676 struct loop_device *lo; 1677 struct kobject *kobj; 1678 int err; 1679 1680 mutex_lock(&loop_index_mutex); 1681 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1682 if (err < 0) 1683 err = loop_add(&lo, MINOR(dev) >> part_shift); 1684 if (err < 0) 1685 kobj = NULL; 1686 else 1687 kobj = get_disk(lo->lo_disk); 1688 mutex_unlock(&loop_index_mutex); 1689 1690 *part = 0; 1691 return kobj; 1692 } 1693 1694 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1695 unsigned long parm) 1696 { 1697 struct loop_device *lo; 1698 int ret = -ENOSYS; 1699 1700 mutex_lock(&loop_index_mutex); 1701 switch (cmd) { 1702 case LOOP_CTL_ADD: 1703 ret = loop_lookup(&lo, parm); 1704 if (ret >= 0) { 1705 ret = -EEXIST; 1706 break; 1707 } 1708 ret = loop_add(&lo, parm); 1709 break; 1710 case LOOP_CTL_REMOVE: 1711 ret = loop_lookup(&lo, parm); 1712 if (ret < 0) 1713 break; 1714 mutex_lock(&lo->lo_ctl_mutex); 1715 if (lo->lo_state != Lo_unbound) { 1716 ret = -EBUSY; 1717 mutex_unlock(&lo->lo_ctl_mutex); 1718 break; 1719 } 1720 if (atomic_read(&lo->lo_refcnt) > 0) { 1721 ret = -EBUSY; 1722 mutex_unlock(&lo->lo_ctl_mutex); 1723 break; 1724 } 1725 lo->lo_disk->private_data = NULL; 1726 mutex_unlock(&lo->lo_ctl_mutex); 1727 idr_remove(&loop_index_idr, lo->lo_number); 1728 loop_remove(lo); 1729 break; 1730 case LOOP_CTL_GET_FREE: 1731 ret = loop_lookup(&lo, -1); 1732 if (ret >= 0) 1733 break; 1734 ret = loop_add(&lo, -1); 1735 } 1736 mutex_unlock(&loop_index_mutex); 1737 1738 return ret; 1739 } 1740 1741 static const struct file_operations loop_ctl_fops = { 1742 .open = nonseekable_open, 1743 .unlocked_ioctl = loop_control_ioctl, 1744 .compat_ioctl = loop_control_ioctl, 1745 .owner = THIS_MODULE, 1746 .llseek = noop_llseek, 1747 }; 1748 1749 static struct miscdevice loop_misc = { 1750 .minor = LOOP_CTRL_MINOR, 1751 .name = "loop-control", 1752 .fops = &loop_ctl_fops, 1753 }; 1754 1755 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 1756 MODULE_ALIAS("devname:loop-control"); 1757 1758 static int __init loop_init(void) 1759 { 1760 int i, nr; 1761 unsigned long range; 1762 struct loop_device *lo; 1763 int err; 1764 1765 err = misc_register(&loop_misc); 1766 if (err < 0) 1767 return err; 1768 1769 part_shift = 0; 1770 if (max_part > 0) { 1771 part_shift = fls(max_part); 1772 1773 /* 1774 * Adjust max_part according to part_shift as it is exported 1775 * to user space so that user can decide correct minor number 1776 * if [s]he want to create more devices. 1777 * 1778 * Note that -1 is required because partition 0 is reserved 1779 * for the whole disk. 1780 */ 1781 max_part = (1UL << part_shift) - 1; 1782 } 1783 1784 if ((1UL << part_shift) > DISK_MAX_PARTS) { 1785 err = -EINVAL; 1786 goto misc_out; 1787 } 1788 1789 if (max_loop > 1UL << (MINORBITS - part_shift)) { 1790 err = -EINVAL; 1791 goto misc_out; 1792 } 1793 1794 /* 1795 * If max_loop is specified, create that many devices upfront. 1796 * This also becomes a hard limit. If max_loop is not specified, 1797 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 1798 * init time. Loop devices can be requested on-demand with the 1799 * /dev/loop-control interface, or be instantiated by accessing 1800 * a 'dead' device node. 1801 */ 1802 if (max_loop) { 1803 nr = max_loop; 1804 range = max_loop << part_shift; 1805 } else { 1806 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 1807 range = 1UL << MINORBITS; 1808 } 1809 1810 if (register_blkdev(LOOP_MAJOR, "loop")) { 1811 err = -EIO; 1812 goto misc_out; 1813 } 1814 1815 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 1816 THIS_MODULE, loop_probe, NULL, NULL); 1817 1818 /* pre-create number of devices given by config or max_loop */ 1819 mutex_lock(&loop_index_mutex); 1820 for (i = 0; i < nr; i++) 1821 loop_add(&lo, i); 1822 mutex_unlock(&loop_index_mutex); 1823 1824 printk(KERN_INFO "loop: module loaded\n"); 1825 return 0; 1826 1827 misc_out: 1828 misc_deregister(&loop_misc); 1829 return err; 1830 } 1831 1832 static int loop_exit_cb(int id, void *ptr, void *data) 1833 { 1834 struct loop_device *lo = ptr; 1835 1836 loop_remove(lo); 1837 return 0; 1838 } 1839 1840 static void __exit loop_exit(void) 1841 { 1842 unsigned long range; 1843 1844 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 1845 1846 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 1847 idr_destroy(&loop_index_idr); 1848 1849 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 1850 unregister_blkdev(LOOP_MAJOR, "loop"); 1851 1852 misc_deregister(&loop_misc); 1853 } 1854 1855 module_init(loop_init); 1856 module_exit(loop_exit); 1857 1858 #ifndef MODULE 1859 static int __init max_loop_setup(char *str) 1860 { 1861 max_loop = simple_strtol(str, NULL, 0); 1862 return 1; 1863 } 1864 1865 __setup("max_loop=", max_loop_setup); 1866 #endif 1867