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 <linux/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 void __loop_update_dio(struct loop_device *lo, bool dio) 168 { 169 struct file *file = lo->lo_backing_file; 170 struct address_space *mapping = file->f_mapping; 171 struct inode *inode = mapping->host; 172 unsigned short sb_bsize = 0; 173 unsigned dio_align = 0; 174 bool use_dio; 175 176 if (inode->i_sb->s_bdev) { 177 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 178 dio_align = sb_bsize - 1; 179 } 180 181 /* 182 * We support direct I/O only if lo_offset is aligned with the 183 * logical I/O size of backing device, and the logical block 184 * size of loop is bigger than the backing device's and the loop 185 * needn't transform transfer. 186 * 187 * TODO: the above condition may be loosed in the future, and 188 * direct I/O may be switched runtime at that time because most 189 * of requests in sane applications should be PAGE_SIZE aligned 190 */ 191 if (dio) { 192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 193 !(lo->lo_offset & dio_align) && 194 mapping->a_ops->direct_IO && 195 !lo->transfer) 196 use_dio = true; 197 else 198 use_dio = false; 199 } else { 200 use_dio = false; 201 } 202 203 if (lo->use_dio == use_dio) 204 return; 205 206 /* flush dirty pages before changing direct IO */ 207 vfs_fsync(file, 0); 208 209 /* 210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 212 * will get updated by ioctl(LOOP_GET_STATUS) 213 */ 214 blk_mq_freeze_queue(lo->lo_queue); 215 lo->use_dio = use_dio; 216 if (use_dio) 217 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 218 else 219 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 220 blk_mq_unfreeze_queue(lo->lo_queue); 221 } 222 223 static int 224 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 225 { 226 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 227 sector_t x = (sector_t)size; 228 struct block_device *bdev = lo->lo_device; 229 230 if (unlikely((loff_t)x != size)) 231 return -EFBIG; 232 if (lo->lo_offset != offset) 233 lo->lo_offset = offset; 234 if (lo->lo_sizelimit != sizelimit) 235 lo->lo_sizelimit = sizelimit; 236 set_capacity(lo->lo_disk, x); 237 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 238 /* let user-space know about the new size */ 239 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 240 return 0; 241 } 242 243 static inline int 244 lo_do_transfer(struct loop_device *lo, int cmd, 245 struct page *rpage, unsigned roffs, 246 struct page *lpage, unsigned loffs, 247 int size, sector_t rblock) 248 { 249 int ret; 250 251 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 252 if (likely(!ret)) 253 return 0; 254 255 printk_ratelimited(KERN_ERR 256 "loop: Transfer error at byte offset %llu, length %i.\n", 257 (unsigned long long)rblock << 9, size); 258 return ret; 259 } 260 261 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 262 { 263 struct iov_iter i; 264 ssize_t bw; 265 266 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len); 267 268 file_start_write(file); 269 bw = vfs_iter_write(file, &i, ppos); 270 file_end_write(file); 271 272 if (likely(bw == bvec->bv_len)) 273 return 0; 274 275 printk_ratelimited(KERN_ERR 276 "loop: Write error at byte offset %llu, length %i.\n", 277 (unsigned long long)*ppos, bvec->bv_len); 278 if (bw >= 0) 279 bw = -EIO; 280 return bw; 281 } 282 283 static int lo_write_simple(struct loop_device *lo, struct request *rq, 284 loff_t pos) 285 { 286 struct bio_vec bvec; 287 struct req_iterator iter; 288 int ret = 0; 289 290 rq_for_each_segment(bvec, rq, iter) { 291 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 292 if (ret < 0) 293 break; 294 cond_resched(); 295 } 296 297 return ret; 298 } 299 300 /* 301 * This is the slow, transforming version that needs to double buffer the 302 * data as it cannot do the transformations in place without having direct 303 * access to the destination pages of the backing file. 304 */ 305 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 306 loff_t pos) 307 { 308 struct bio_vec bvec, b; 309 struct req_iterator iter; 310 struct page *page; 311 int ret = 0; 312 313 page = alloc_page(GFP_NOIO); 314 if (unlikely(!page)) 315 return -ENOMEM; 316 317 rq_for_each_segment(bvec, rq, iter) { 318 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 319 bvec.bv_offset, bvec.bv_len, pos >> 9); 320 if (unlikely(ret)) 321 break; 322 323 b.bv_page = page; 324 b.bv_offset = 0; 325 b.bv_len = bvec.bv_len; 326 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 327 if (ret < 0) 328 break; 329 } 330 331 __free_page(page); 332 return ret; 333 } 334 335 static int lo_read_simple(struct loop_device *lo, struct request *rq, 336 loff_t pos) 337 { 338 struct bio_vec bvec; 339 struct req_iterator iter; 340 struct iov_iter i; 341 ssize_t len; 342 343 rq_for_each_segment(bvec, rq, iter) { 344 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); 345 len = vfs_iter_read(lo->lo_backing_file, &i, &pos); 346 if (len < 0) 347 return len; 348 349 flush_dcache_page(bvec.bv_page); 350 351 if (len != bvec.bv_len) { 352 struct bio *bio; 353 354 __rq_for_each_bio(bio, rq) 355 zero_fill_bio(bio); 356 break; 357 } 358 cond_resched(); 359 } 360 361 return 0; 362 } 363 364 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 365 loff_t pos) 366 { 367 struct bio_vec bvec, b; 368 struct req_iterator iter; 369 struct iov_iter i; 370 struct page *page; 371 ssize_t len; 372 int ret = 0; 373 374 page = alloc_page(GFP_NOIO); 375 if (unlikely(!page)) 376 return -ENOMEM; 377 378 rq_for_each_segment(bvec, rq, iter) { 379 loff_t offset = pos; 380 381 b.bv_page = page; 382 b.bv_offset = 0; 383 b.bv_len = bvec.bv_len; 384 385 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); 386 len = vfs_iter_read(lo->lo_backing_file, &i, &pos); 387 if (len < 0) { 388 ret = len; 389 goto out_free_page; 390 } 391 392 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 393 bvec.bv_offset, len, offset >> 9); 394 if (ret) 395 goto out_free_page; 396 397 flush_dcache_page(bvec.bv_page); 398 399 if (len != bvec.bv_len) { 400 struct bio *bio; 401 402 __rq_for_each_bio(bio, rq) 403 zero_fill_bio(bio); 404 break; 405 } 406 } 407 408 ret = 0; 409 out_free_page: 410 __free_page(page); 411 return ret; 412 } 413 414 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 415 { 416 /* 417 * We use punch hole to reclaim the free space used by the 418 * image a.k.a. discard. However we do not support discard if 419 * encryption is enabled, because it may give an attacker 420 * useful information. 421 */ 422 struct file *file = lo->lo_backing_file; 423 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 424 int ret; 425 426 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 427 ret = -EOPNOTSUPP; 428 goto out; 429 } 430 431 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 432 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 433 ret = -EIO; 434 out: 435 return ret; 436 } 437 438 static int lo_req_flush(struct loop_device *lo, struct request *rq) 439 { 440 struct file *file = lo->lo_backing_file; 441 int ret = vfs_fsync(file, 0); 442 if (unlikely(ret && ret != -EINVAL)) 443 ret = -EIO; 444 445 return ret; 446 } 447 448 static inline void handle_partial_read(struct loop_cmd *cmd, long bytes) 449 { 450 if (bytes < 0 || op_is_write(req_op(cmd->rq))) 451 return; 452 453 if (unlikely(bytes < blk_rq_bytes(cmd->rq))) { 454 struct bio *bio = cmd->rq->bio; 455 456 bio_advance(bio, bytes); 457 zero_fill_bio(bio); 458 } 459 } 460 461 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 462 { 463 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 464 struct request *rq = cmd->rq; 465 466 handle_partial_read(cmd, ret); 467 468 if (ret > 0) 469 ret = 0; 470 else if (ret < 0) 471 ret = -EIO; 472 473 blk_mq_complete_request(rq, ret); 474 } 475 476 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 477 loff_t pos, bool rw) 478 { 479 struct iov_iter iter; 480 struct bio_vec *bvec; 481 struct bio *bio = cmd->rq->bio; 482 struct file *file = lo->lo_backing_file; 483 int ret; 484 485 /* nomerge for loop request queue */ 486 WARN_ON(cmd->rq->bio != cmd->rq->biotail); 487 488 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 489 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, 490 bio_segments(bio), blk_rq_bytes(cmd->rq)); 491 /* 492 * This bio may be started from the middle of the 'bvec' 493 * because of bio splitting, so offset from the bvec must 494 * be passed to iov iterator 495 */ 496 iter.iov_offset = bio->bi_iter.bi_bvec_done; 497 498 cmd->iocb.ki_pos = pos; 499 cmd->iocb.ki_filp = file; 500 cmd->iocb.ki_complete = lo_rw_aio_complete; 501 cmd->iocb.ki_flags = IOCB_DIRECT; 502 503 if (rw == WRITE) 504 ret = call_write_iter(file, &cmd->iocb, &iter); 505 else 506 ret = call_read_iter(file, &cmd->iocb, &iter); 507 508 if (ret != -EIOCBQUEUED) 509 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 510 return 0; 511 } 512 513 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 514 { 515 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 516 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 517 518 /* 519 * lo_write_simple and lo_read_simple should have been covered 520 * by io submit style function like lo_rw_aio(), one blocker 521 * is that lo_read_simple() need to call flush_dcache_page after 522 * the page is written from kernel, and it isn't easy to handle 523 * this in io submit style function which submits all segments 524 * of the req at one time. And direct read IO doesn't need to 525 * run flush_dcache_page(). 526 */ 527 switch (req_op(rq)) { 528 case REQ_OP_FLUSH: 529 return lo_req_flush(lo, rq); 530 case REQ_OP_DISCARD: 531 case REQ_OP_WRITE_ZEROES: 532 return lo_discard(lo, rq, pos); 533 case REQ_OP_WRITE: 534 if (lo->transfer) 535 return lo_write_transfer(lo, rq, pos); 536 else if (cmd->use_aio) 537 return lo_rw_aio(lo, cmd, pos, WRITE); 538 else 539 return lo_write_simple(lo, rq, pos); 540 case REQ_OP_READ: 541 if (lo->transfer) 542 return lo_read_transfer(lo, rq, pos); 543 else if (cmd->use_aio) 544 return lo_rw_aio(lo, cmd, pos, READ); 545 else 546 return lo_read_simple(lo, rq, pos); 547 default: 548 WARN_ON_ONCE(1); 549 return -EIO; 550 break; 551 } 552 } 553 554 struct switch_request { 555 struct file *file; 556 struct completion wait; 557 }; 558 559 static inline void loop_update_dio(struct loop_device *lo) 560 { 561 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 562 lo->use_dio); 563 } 564 565 /* 566 * Do the actual switch; called from the BIO completion routine 567 */ 568 static void do_loop_switch(struct loop_device *lo, struct switch_request *p) 569 { 570 struct file *file = p->file; 571 struct file *old_file = lo->lo_backing_file; 572 struct address_space *mapping; 573 574 /* if no new file, only flush of queued bios requested */ 575 if (!file) 576 return; 577 578 mapping = file->f_mapping; 579 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 580 lo->lo_backing_file = file; 581 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? 582 mapping->host->i_bdev->bd_block_size : PAGE_SIZE; 583 lo->old_gfp_mask = mapping_gfp_mask(mapping); 584 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 585 loop_update_dio(lo); 586 } 587 588 /* 589 * loop_switch performs the hard work of switching a backing store. 590 * First it needs to flush existing IO, it does this by sending a magic 591 * BIO down the pipe. The completion of this BIO does the actual switch. 592 */ 593 static int loop_switch(struct loop_device *lo, struct file *file) 594 { 595 struct switch_request w; 596 597 w.file = file; 598 599 /* freeze queue and wait for completion of scheduled requests */ 600 blk_mq_freeze_queue(lo->lo_queue); 601 602 /* do the switch action */ 603 do_loop_switch(lo, &w); 604 605 /* unfreeze */ 606 blk_mq_unfreeze_queue(lo->lo_queue); 607 608 return 0; 609 } 610 611 /* 612 * Helper to flush the IOs in loop, but keeping loop thread running 613 */ 614 static int loop_flush(struct loop_device *lo) 615 { 616 return loop_switch(lo, NULL); 617 } 618 619 static void loop_reread_partitions(struct loop_device *lo, 620 struct block_device *bdev) 621 { 622 int rc; 623 624 /* 625 * bd_mutex has been held already in release path, so don't 626 * acquire it if this function is called in such case. 627 * 628 * If the reread partition isn't from release path, lo_refcnt 629 * must be at least one and it can only become zero when the 630 * current holder is released. 631 */ 632 if (!atomic_read(&lo->lo_refcnt)) 633 rc = __blkdev_reread_part(bdev); 634 else 635 rc = blkdev_reread_part(bdev); 636 if (rc) 637 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 638 __func__, lo->lo_number, lo->lo_file_name, rc); 639 } 640 641 /* 642 * loop_change_fd switched the backing store of a loopback device to 643 * a new file. This is useful for operating system installers to free up 644 * the original file and in High Availability environments to switch to 645 * an alternative location for the content in case of server meltdown. 646 * This can only work if the loop device is used read-only, and if the 647 * new backing store is the same size and type as the old backing store. 648 */ 649 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 650 unsigned int arg) 651 { 652 struct file *file, *old_file; 653 struct inode *inode; 654 int error; 655 656 error = -ENXIO; 657 if (lo->lo_state != Lo_bound) 658 goto out; 659 660 /* the loop device has to be read-only */ 661 error = -EINVAL; 662 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 663 goto out; 664 665 error = -EBADF; 666 file = fget(arg); 667 if (!file) 668 goto out; 669 670 inode = file->f_mapping->host; 671 old_file = lo->lo_backing_file; 672 673 error = -EINVAL; 674 675 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 676 goto out_putf; 677 678 /* size of the new backing store needs to be the same */ 679 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 680 goto out_putf; 681 682 /* and ... switch */ 683 error = loop_switch(lo, file); 684 if (error) 685 goto out_putf; 686 687 fput(old_file); 688 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 689 loop_reread_partitions(lo, bdev); 690 return 0; 691 692 out_putf: 693 fput(file); 694 out: 695 return error; 696 } 697 698 static inline int is_loop_device(struct file *file) 699 { 700 struct inode *i = file->f_mapping->host; 701 702 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 703 } 704 705 /* loop sysfs attributes */ 706 707 static ssize_t loop_attr_show(struct device *dev, char *page, 708 ssize_t (*callback)(struct loop_device *, char *)) 709 { 710 struct gendisk *disk = dev_to_disk(dev); 711 struct loop_device *lo = disk->private_data; 712 713 return callback(lo, page); 714 } 715 716 #define LOOP_ATTR_RO(_name) \ 717 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 718 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 719 struct device_attribute *attr, char *b) \ 720 { \ 721 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 722 } \ 723 static struct device_attribute loop_attr_##_name = \ 724 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); 725 726 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 727 { 728 ssize_t ret; 729 char *p = NULL; 730 731 spin_lock_irq(&lo->lo_lock); 732 if (lo->lo_backing_file) 733 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 734 spin_unlock_irq(&lo->lo_lock); 735 736 if (IS_ERR_OR_NULL(p)) 737 ret = PTR_ERR(p); 738 else { 739 ret = strlen(p); 740 memmove(buf, p, ret); 741 buf[ret++] = '\n'; 742 buf[ret] = 0; 743 } 744 745 return ret; 746 } 747 748 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 749 { 750 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 751 } 752 753 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 754 { 755 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 756 } 757 758 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 759 { 760 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 761 762 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 763 } 764 765 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 766 { 767 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 768 769 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 770 } 771 772 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 773 { 774 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 775 776 return sprintf(buf, "%s\n", dio ? "1" : "0"); 777 } 778 779 LOOP_ATTR_RO(backing_file); 780 LOOP_ATTR_RO(offset); 781 LOOP_ATTR_RO(sizelimit); 782 LOOP_ATTR_RO(autoclear); 783 LOOP_ATTR_RO(partscan); 784 LOOP_ATTR_RO(dio); 785 786 static struct attribute *loop_attrs[] = { 787 &loop_attr_backing_file.attr, 788 &loop_attr_offset.attr, 789 &loop_attr_sizelimit.attr, 790 &loop_attr_autoclear.attr, 791 &loop_attr_partscan.attr, 792 &loop_attr_dio.attr, 793 NULL, 794 }; 795 796 static struct attribute_group loop_attribute_group = { 797 .name = "loop", 798 .attrs= loop_attrs, 799 }; 800 801 static int loop_sysfs_init(struct loop_device *lo) 802 { 803 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 804 &loop_attribute_group); 805 } 806 807 static void loop_sysfs_exit(struct loop_device *lo) 808 { 809 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 810 &loop_attribute_group); 811 } 812 813 static void loop_config_discard(struct loop_device *lo) 814 { 815 struct file *file = lo->lo_backing_file; 816 struct inode *inode = file->f_mapping->host; 817 struct request_queue *q = lo->lo_queue; 818 819 /* 820 * We use punch hole to reclaim the free space used by the 821 * image a.k.a. discard. However we do not support discard if 822 * encryption is enabled, because it may give an attacker 823 * useful information. 824 */ 825 if ((!file->f_op->fallocate) || 826 lo->lo_encrypt_key_size) { 827 q->limits.discard_granularity = 0; 828 q->limits.discard_alignment = 0; 829 blk_queue_max_discard_sectors(q, 0); 830 blk_queue_max_write_zeroes_sectors(q, 0); 831 q->limits.discard_zeroes_data = 0; 832 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 833 return; 834 } 835 836 q->limits.discard_granularity = inode->i_sb->s_blocksize; 837 q->limits.discard_alignment = 0; 838 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 839 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 840 q->limits.discard_zeroes_data = 1; 841 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 842 } 843 844 static void loop_unprepare_queue(struct loop_device *lo) 845 { 846 kthread_flush_worker(&lo->worker); 847 kthread_stop(lo->worker_task); 848 } 849 850 static int loop_prepare_queue(struct loop_device *lo) 851 { 852 kthread_init_worker(&lo->worker); 853 lo->worker_task = kthread_run(kthread_worker_fn, 854 &lo->worker, "loop%d", lo->lo_number); 855 if (IS_ERR(lo->worker_task)) 856 return -ENOMEM; 857 set_user_nice(lo->worker_task, MIN_NICE); 858 return 0; 859 } 860 861 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 862 struct block_device *bdev, unsigned int arg) 863 { 864 struct file *file, *f; 865 struct inode *inode; 866 struct address_space *mapping; 867 unsigned lo_blocksize; 868 int lo_flags = 0; 869 int error; 870 loff_t size; 871 872 /* This is safe, since we have a reference from open(). */ 873 __module_get(THIS_MODULE); 874 875 error = -EBADF; 876 file = fget(arg); 877 if (!file) 878 goto out; 879 880 error = -EBUSY; 881 if (lo->lo_state != Lo_unbound) 882 goto out_putf; 883 884 /* Avoid recursion */ 885 f = file; 886 while (is_loop_device(f)) { 887 struct loop_device *l; 888 889 if (f->f_mapping->host->i_bdev == bdev) 890 goto out_putf; 891 892 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 893 if (l->lo_state == Lo_unbound) { 894 error = -EINVAL; 895 goto out_putf; 896 } 897 f = l->lo_backing_file; 898 } 899 900 mapping = file->f_mapping; 901 inode = mapping->host; 902 903 error = -EINVAL; 904 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 905 goto out_putf; 906 907 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 908 !file->f_op->write_iter) 909 lo_flags |= LO_FLAGS_READ_ONLY; 910 911 lo_blocksize = S_ISBLK(inode->i_mode) ? 912 inode->i_bdev->bd_block_size : PAGE_SIZE; 913 914 error = -EFBIG; 915 size = get_loop_size(lo, file); 916 if ((loff_t)(sector_t)size != size) 917 goto out_putf; 918 error = loop_prepare_queue(lo); 919 if (error) 920 goto out_putf; 921 922 error = 0; 923 924 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 925 926 lo->use_dio = false; 927 lo->lo_blocksize = lo_blocksize; 928 lo->lo_device = bdev; 929 lo->lo_flags = lo_flags; 930 lo->lo_backing_file = file; 931 lo->transfer = NULL; 932 lo->ioctl = NULL; 933 lo->lo_sizelimit = 0; 934 lo->old_gfp_mask = mapping_gfp_mask(mapping); 935 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 936 937 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 938 blk_queue_write_cache(lo->lo_queue, true, false); 939 940 loop_update_dio(lo); 941 set_capacity(lo->lo_disk, size); 942 bd_set_size(bdev, size << 9); 943 loop_sysfs_init(lo); 944 /* let user-space know about the new size */ 945 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 946 947 set_blocksize(bdev, lo_blocksize); 948 949 lo->lo_state = Lo_bound; 950 if (part_shift) 951 lo->lo_flags |= LO_FLAGS_PARTSCAN; 952 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 953 loop_reread_partitions(lo, bdev); 954 955 /* Grab the block_device to prevent its destruction after we 956 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 957 */ 958 bdgrab(bdev); 959 return 0; 960 961 out_putf: 962 fput(file); 963 out: 964 /* This is safe: open() is still holding a reference. */ 965 module_put(THIS_MODULE); 966 return error; 967 } 968 969 static int 970 loop_release_xfer(struct loop_device *lo) 971 { 972 int err = 0; 973 struct loop_func_table *xfer = lo->lo_encryption; 974 975 if (xfer) { 976 if (xfer->release) 977 err = xfer->release(lo); 978 lo->transfer = NULL; 979 lo->lo_encryption = NULL; 980 module_put(xfer->owner); 981 } 982 return err; 983 } 984 985 static int 986 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 987 const struct loop_info64 *i) 988 { 989 int err = 0; 990 991 if (xfer) { 992 struct module *owner = xfer->owner; 993 994 if (!try_module_get(owner)) 995 return -EINVAL; 996 if (xfer->init) 997 err = xfer->init(lo, i); 998 if (err) 999 module_put(owner); 1000 else 1001 lo->lo_encryption = xfer; 1002 } 1003 return err; 1004 } 1005 1006 static int loop_clr_fd(struct loop_device *lo) 1007 { 1008 struct file *filp = lo->lo_backing_file; 1009 gfp_t gfp = lo->old_gfp_mask; 1010 struct block_device *bdev = lo->lo_device; 1011 1012 if (lo->lo_state != Lo_bound) 1013 return -ENXIO; 1014 1015 /* 1016 * If we've explicitly asked to tear down the loop device, 1017 * and it has an elevated reference count, set it for auto-teardown when 1018 * the last reference goes away. This stops $!~#$@ udev from 1019 * preventing teardown because it decided that it needs to run blkid on 1020 * the loopback device whenever they appear. xfstests is notorious for 1021 * failing tests because blkid via udev races with a losetup 1022 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1023 * command to fail with EBUSY. 1024 */ 1025 if (atomic_read(&lo->lo_refcnt) > 1) { 1026 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1027 mutex_unlock(&lo->lo_ctl_mutex); 1028 return 0; 1029 } 1030 1031 if (filp == NULL) 1032 return -EINVAL; 1033 1034 /* freeze request queue during the transition */ 1035 blk_mq_freeze_queue(lo->lo_queue); 1036 1037 spin_lock_irq(&lo->lo_lock); 1038 lo->lo_state = Lo_rundown; 1039 lo->lo_backing_file = NULL; 1040 spin_unlock_irq(&lo->lo_lock); 1041 1042 loop_release_xfer(lo); 1043 lo->transfer = NULL; 1044 lo->ioctl = NULL; 1045 lo->lo_device = NULL; 1046 lo->lo_encryption = NULL; 1047 lo->lo_offset = 0; 1048 lo->lo_sizelimit = 0; 1049 lo->lo_encrypt_key_size = 0; 1050 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1051 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1052 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1053 if (bdev) { 1054 bdput(bdev); 1055 invalidate_bdev(bdev); 1056 } 1057 set_capacity(lo->lo_disk, 0); 1058 loop_sysfs_exit(lo); 1059 if (bdev) { 1060 bd_set_size(bdev, 0); 1061 /* let user-space know about this change */ 1062 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1063 } 1064 mapping_set_gfp_mask(filp->f_mapping, gfp); 1065 lo->lo_state = Lo_unbound; 1066 /* This is safe: open() is still holding a reference. */ 1067 module_put(THIS_MODULE); 1068 blk_mq_unfreeze_queue(lo->lo_queue); 1069 1070 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 1071 loop_reread_partitions(lo, bdev); 1072 lo->lo_flags = 0; 1073 if (!part_shift) 1074 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1075 loop_unprepare_queue(lo); 1076 mutex_unlock(&lo->lo_ctl_mutex); 1077 /* 1078 * Need not hold lo_ctl_mutex to fput backing file. 1079 * Calling fput holding lo_ctl_mutex triggers a circular 1080 * lock dependency possibility warning as fput can take 1081 * bd_mutex which is usually taken before lo_ctl_mutex. 1082 */ 1083 fput(filp); 1084 return 0; 1085 } 1086 1087 static int 1088 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1089 { 1090 int err; 1091 struct loop_func_table *xfer; 1092 kuid_t uid = current_uid(); 1093 1094 if (lo->lo_encrypt_key_size && 1095 !uid_eq(lo->lo_key_owner, uid) && 1096 !capable(CAP_SYS_ADMIN)) 1097 return -EPERM; 1098 if (lo->lo_state != Lo_bound) 1099 return -ENXIO; 1100 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1101 return -EINVAL; 1102 1103 /* I/O need to be drained during transfer transition */ 1104 blk_mq_freeze_queue(lo->lo_queue); 1105 1106 err = loop_release_xfer(lo); 1107 if (err) 1108 goto exit; 1109 1110 if (info->lo_encrypt_type) { 1111 unsigned int type = info->lo_encrypt_type; 1112 1113 if (type >= MAX_LO_CRYPT) 1114 return -EINVAL; 1115 xfer = xfer_funcs[type]; 1116 if (xfer == NULL) 1117 return -EINVAL; 1118 } else 1119 xfer = NULL; 1120 1121 err = loop_init_xfer(lo, xfer, info); 1122 if (err) 1123 goto exit; 1124 1125 if (lo->lo_offset != info->lo_offset || 1126 lo->lo_sizelimit != info->lo_sizelimit) 1127 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1128 err = -EFBIG; 1129 goto exit; 1130 } 1131 1132 loop_config_discard(lo); 1133 1134 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1135 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1136 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1137 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1138 1139 if (!xfer) 1140 xfer = &none_funcs; 1141 lo->transfer = xfer->transfer; 1142 lo->ioctl = xfer->ioctl; 1143 1144 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1145 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1146 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1147 1148 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1149 lo->lo_init[0] = info->lo_init[0]; 1150 lo->lo_init[1] = info->lo_init[1]; 1151 if (info->lo_encrypt_key_size) { 1152 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1153 info->lo_encrypt_key_size); 1154 lo->lo_key_owner = uid; 1155 } 1156 1157 /* update dio if lo_offset or transfer is changed */ 1158 __loop_update_dio(lo, lo->use_dio); 1159 1160 exit: 1161 blk_mq_unfreeze_queue(lo->lo_queue); 1162 1163 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1164 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1165 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1166 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1167 loop_reread_partitions(lo, lo->lo_device); 1168 } 1169 1170 return err; 1171 } 1172 1173 static int 1174 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1175 { 1176 struct file *file = lo->lo_backing_file; 1177 struct kstat stat; 1178 int error; 1179 1180 if (lo->lo_state != Lo_bound) 1181 return -ENXIO; 1182 error = vfs_getattr(&file->f_path, &stat, 1183 STATX_INO, AT_STATX_SYNC_AS_STAT); 1184 if (error) 1185 return error; 1186 memset(info, 0, sizeof(*info)); 1187 info->lo_number = lo->lo_number; 1188 info->lo_device = huge_encode_dev(stat.dev); 1189 info->lo_inode = stat.ino; 1190 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); 1191 info->lo_offset = lo->lo_offset; 1192 info->lo_sizelimit = lo->lo_sizelimit; 1193 info->lo_flags = lo->lo_flags; 1194 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1195 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1196 info->lo_encrypt_type = 1197 lo->lo_encryption ? lo->lo_encryption->number : 0; 1198 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1199 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1200 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1201 lo->lo_encrypt_key_size); 1202 } 1203 return 0; 1204 } 1205 1206 static void 1207 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1208 { 1209 memset(info64, 0, sizeof(*info64)); 1210 info64->lo_number = info->lo_number; 1211 info64->lo_device = info->lo_device; 1212 info64->lo_inode = info->lo_inode; 1213 info64->lo_rdevice = info->lo_rdevice; 1214 info64->lo_offset = info->lo_offset; 1215 info64->lo_sizelimit = 0; 1216 info64->lo_encrypt_type = info->lo_encrypt_type; 1217 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1218 info64->lo_flags = info->lo_flags; 1219 info64->lo_init[0] = info->lo_init[0]; 1220 info64->lo_init[1] = info->lo_init[1]; 1221 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1222 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1223 else 1224 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1225 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1226 } 1227 1228 static int 1229 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1230 { 1231 memset(info, 0, sizeof(*info)); 1232 info->lo_number = info64->lo_number; 1233 info->lo_device = info64->lo_device; 1234 info->lo_inode = info64->lo_inode; 1235 info->lo_rdevice = info64->lo_rdevice; 1236 info->lo_offset = info64->lo_offset; 1237 info->lo_encrypt_type = info64->lo_encrypt_type; 1238 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1239 info->lo_flags = info64->lo_flags; 1240 info->lo_init[0] = info64->lo_init[0]; 1241 info->lo_init[1] = info64->lo_init[1]; 1242 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1243 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1244 else 1245 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1246 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1247 1248 /* error in case values were truncated */ 1249 if (info->lo_device != info64->lo_device || 1250 info->lo_rdevice != info64->lo_rdevice || 1251 info->lo_inode != info64->lo_inode || 1252 info->lo_offset != info64->lo_offset) 1253 return -EOVERFLOW; 1254 1255 return 0; 1256 } 1257 1258 static int 1259 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1260 { 1261 struct loop_info info; 1262 struct loop_info64 info64; 1263 1264 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1265 return -EFAULT; 1266 loop_info64_from_old(&info, &info64); 1267 return loop_set_status(lo, &info64); 1268 } 1269 1270 static int 1271 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1272 { 1273 struct loop_info64 info64; 1274 1275 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1276 return -EFAULT; 1277 return loop_set_status(lo, &info64); 1278 } 1279 1280 static int 1281 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1282 struct loop_info info; 1283 struct loop_info64 info64; 1284 int err = 0; 1285 1286 if (!arg) 1287 err = -EINVAL; 1288 if (!err) 1289 err = loop_get_status(lo, &info64); 1290 if (!err) 1291 err = loop_info64_to_old(&info64, &info); 1292 if (!err && copy_to_user(arg, &info, sizeof(info))) 1293 err = -EFAULT; 1294 1295 return err; 1296 } 1297 1298 static int 1299 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1300 struct loop_info64 info64; 1301 int err = 0; 1302 1303 if (!arg) 1304 err = -EINVAL; 1305 if (!err) 1306 err = loop_get_status(lo, &info64); 1307 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1308 err = -EFAULT; 1309 1310 return err; 1311 } 1312 1313 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) 1314 { 1315 if (unlikely(lo->lo_state != Lo_bound)) 1316 return -ENXIO; 1317 1318 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1319 } 1320 1321 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1322 { 1323 int error = -ENXIO; 1324 if (lo->lo_state != Lo_bound) 1325 goto out; 1326 1327 __loop_update_dio(lo, !!arg); 1328 if (lo->use_dio == !!arg) 1329 return 0; 1330 error = -EINVAL; 1331 out: 1332 return error; 1333 } 1334 1335 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1336 unsigned int cmd, unsigned long arg) 1337 { 1338 struct loop_device *lo = bdev->bd_disk->private_data; 1339 int err; 1340 1341 mutex_lock_nested(&lo->lo_ctl_mutex, 1); 1342 switch (cmd) { 1343 case LOOP_SET_FD: 1344 err = loop_set_fd(lo, mode, bdev, arg); 1345 break; 1346 case LOOP_CHANGE_FD: 1347 err = loop_change_fd(lo, bdev, arg); 1348 break; 1349 case LOOP_CLR_FD: 1350 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1351 err = loop_clr_fd(lo); 1352 if (!err) 1353 goto out_unlocked; 1354 break; 1355 case LOOP_SET_STATUS: 1356 err = -EPERM; 1357 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1358 err = loop_set_status_old(lo, 1359 (struct loop_info __user *)arg); 1360 break; 1361 case LOOP_GET_STATUS: 1362 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1363 break; 1364 case LOOP_SET_STATUS64: 1365 err = -EPERM; 1366 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1367 err = loop_set_status64(lo, 1368 (struct loop_info64 __user *) arg); 1369 break; 1370 case LOOP_GET_STATUS64: 1371 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1372 break; 1373 case LOOP_SET_CAPACITY: 1374 err = -EPERM; 1375 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1376 err = loop_set_capacity(lo, bdev); 1377 break; 1378 case LOOP_SET_DIRECT_IO: 1379 err = -EPERM; 1380 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1381 err = loop_set_dio(lo, arg); 1382 break; 1383 default: 1384 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1385 } 1386 mutex_unlock(&lo->lo_ctl_mutex); 1387 1388 out_unlocked: 1389 return err; 1390 } 1391 1392 #ifdef CONFIG_COMPAT 1393 struct compat_loop_info { 1394 compat_int_t lo_number; /* ioctl r/o */ 1395 compat_dev_t lo_device; /* ioctl r/o */ 1396 compat_ulong_t lo_inode; /* ioctl r/o */ 1397 compat_dev_t lo_rdevice; /* ioctl r/o */ 1398 compat_int_t lo_offset; 1399 compat_int_t lo_encrypt_type; 1400 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1401 compat_int_t lo_flags; /* ioctl r/o */ 1402 char lo_name[LO_NAME_SIZE]; 1403 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1404 compat_ulong_t lo_init[2]; 1405 char reserved[4]; 1406 }; 1407 1408 /* 1409 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1410 * - noinlined to reduce stack space usage in main part of driver 1411 */ 1412 static noinline int 1413 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1414 struct loop_info64 *info64) 1415 { 1416 struct compat_loop_info info; 1417 1418 if (copy_from_user(&info, arg, sizeof(info))) 1419 return -EFAULT; 1420 1421 memset(info64, 0, sizeof(*info64)); 1422 info64->lo_number = info.lo_number; 1423 info64->lo_device = info.lo_device; 1424 info64->lo_inode = info.lo_inode; 1425 info64->lo_rdevice = info.lo_rdevice; 1426 info64->lo_offset = info.lo_offset; 1427 info64->lo_sizelimit = 0; 1428 info64->lo_encrypt_type = info.lo_encrypt_type; 1429 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1430 info64->lo_flags = info.lo_flags; 1431 info64->lo_init[0] = info.lo_init[0]; 1432 info64->lo_init[1] = info.lo_init[1]; 1433 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1434 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1435 else 1436 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1437 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1438 return 0; 1439 } 1440 1441 /* 1442 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1443 * - noinlined to reduce stack space usage in main part of driver 1444 */ 1445 static noinline int 1446 loop_info64_to_compat(const struct loop_info64 *info64, 1447 struct compat_loop_info __user *arg) 1448 { 1449 struct compat_loop_info info; 1450 1451 memset(&info, 0, sizeof(info)); 1452 info.lo_number = info64->lo_number; 1453 info.lo_device = info64->lo_device; 1454 info.lo_inode = info64->lo_inode; 1455 info.lo_rdevice = info64->lo_rdevice; 1456 info.lo_offset = info64->lo_offset; 1457 info.lo_encrypt_type = info64->lo_encrypt_type; 1458 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1459 info.lo_flags = info64->lo_flags; 1460 info.lo_init[0] = info64->lo_init[0]; 1461 info.lo_init[1] = info64->lo_init[1]; 1462 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1463 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1464 else 1465 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1466 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1467 1468 /* error in case values were truncated */ 1469 if (info.lo_device != info64->lo_device || 1470 info.lo_rdevice != info64->lo_rdevice || 1471 info.lo_inode != info64->lo_inode || 1472 info.lo_offset != info64->lo_offset || 1473 info.lo_init[0] != info64->lo_init[0] || 1474 info.lo_init[1] != info64->lo_init[1]) 1475 return -EOVERFLOW; 1476 1477 if (copy_to_user(arg, &info, sizeof(info))) 1478 return -EFAULT; 1479 return 0; 1480 } 1481 1482 static int 1483 loop_set_status_compat(struct loop_device *lo, 1484 const struct compat_loop_info __user *arg) 1485 { 1486 struct loop_info64 info64; 1487 int ret; 1488 1489 ret = loop_info64_from_compat(arg, &info64); 1490 if (ret < 0) 1491 return ret; 1492 return loop_set_status(lo, &info64); 1493 } 1494 1495 static int 1496 loop_get_status_compat(struct loop_device *lo, 1497 struct compat_loop_info __user *arg) 1498 { 1499 struct loop_info64 info64; 1500 int err = 0; 1501 1502 if (!arg) 1503 err = -EINVAL; 1504 if (!err) 1505 err = loop_get_status(lo, &info64); 1506 if (!err) 1507 err = loop_info64_to_compat(&info64, arg); 1508 return err; 1509 } 1510 1511 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1512 unsigned int cmd, unsigned long arg) 1513 { 1514 struct loop_device *lo = bdev->bd_disk->private_data; 1515 int err; 1516 1517 switch(cmd) { 1518 case LOOP_SET_STATUS: 1519 mutex_lock(&lo->lo_ctl_mutex); 1520 err = loop_set_status_compat( 1521 lo, (const struct compat_loop_info __user *) arg); 1522 mutex_unlock(&lo->lo_ctl_mutex); 1523 break; 1524 case LOOP_GET_STATUS: 1525 mutex_lock(&lo->lo_ctl_mutex); 1526 err = loop_get_status_compat( 1527 lo, (struct compat_loop_info __user *) arg); 1528 mutex_unlock(&lo->lo_ctl_mutex); 1529 break; 1530 case LOOP_SET_CAPACITY: 1531 case LOOP_CLR_FD: 1532 case LOOP_GET_STATUS64: 1533 case LOOP_SET_STATUS64: 1534 arg = (unsigned long) compat_ptr(arg); 1535 case LOOP_SET_FD: 1536 case LOOP_CHANGE_FD: 1537 err = lo_ioctl(bdev, mode, cmd, arg); 1538 break; 1539 default: 1540 err = -ENOIOCTLCMD; 1541 break; 1542 } 1543 return err; 1544 } 1545 #endif 1546 1547 static int lo_open(struct block_device *bdev, fmode_t mode) 1548 { 1549 struct loop_device *lo; 1550 int err = 0; 1551 1552 mutex_lock(&loop_index_mutex); 1553 lo = bdev->bd_disk->private_data; 1554 if (!lo) { 1555 err = -ENXIO; 1556 goto out; 1557 } 1558 1559 atomic_inc(&lo->lo_refcnt); 1560 out: 1561 mutex_unlock(&loop_index_mutex); 1562 return err; 1563 } 1564 1565 static void lo_release(struct gendisk *disk, fmode_t mode) 1566 { 1567 struct loop_device *lo = disk->private_data; 1568 int err; 1569 1570 if (atomic_dec_return(&lo->lo_refcnt)) 1571 return; 1572 1573 mutex_lock(&lo->lo_ctl_mutex); 1574 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1575 /* 1576 * In autoclear mode, stop the loop thread 1577 * and remove configuration after last close. 1578 */ 1579 err = loop_clr_fd(lo); 1580 if (!err) 1581 return; 1582 } else { 1583 /* 1584 * Otherwise keep thread (if running) and config, 1585 * but flush possible ongoing bios in thread. 1586 */ 1587 loop_flush(lo); 1588 } 1589 1590 mutex_unlock(&lo->lo_ctl_mutex); 1591 } 1592 1593 static const struct block_device_operations lo_fops = { 1594 .owner = THIS_MODULE, 1595 .open = lo_open, 1596 .release = lo_release, 1597 .ioctl = lo_ioctl, 1598 #ifdef CONFIG_COMPAT 1599 .compat_ioctl = lo_compat_ioctl, 1600 #endif 1601 }; 1602 1603 /* 1604 * And now the modules code and kernel interface. 1605 */ 1606 static int max_loop; 1607 module_param(max_loop, int, S_IRUGO); 1608 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1609 module_param(max_part, int, S_IRUGO); 1610 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1611 MODULE_LICENSE("GPL"); 1612 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1613 1614 int loop_register_transfer(struct loop_func_table *funcs) 1615 { 1616 unsigned int n = funcs->number; 1617 1618 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1619 return -EINVAL; 1620 xfer_funcs[n] = funcs; 1621 return 0; 1622 } 1623 1624 static int unregister_transfer_cb(int id, void *ptr, void *data) 1625 { 1626 struct loop_device *lo = ptr; 1627 struct loop_func_table *xfer = data; 1628 1629 mutex_lock(&lo->lo_ctl_mutex); 1630 if (lo->lo_encryption == xfer) 1631 loop_release_xfer(lo); 1632 mutex_unlock(&lo->lo_ctl_mutex); 1633 return 0; 1634 } 1635 1636 int loop_unregister_transfer(int number) 1637 { 1638 unsigned int n = number; 1639 struct loop_func_table *xfer; 1640 1641 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1642 return -EINVAL; 1643 1644 xfer_funcs[n] = NULL; 1645 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1646 return 0; 1647 } 1648 1649 EXPORT_SYMBOL(loop_register_transfer); 1650 EXPORT_SYMBOL(loop_unregister_transfer); 1651 1652 static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1653 const struct blk_mq_queue_data *bd) 1654 { 1655 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); 1656 struct loop_device *lo = cmd->rq->q->queuedata; 1657 1658 blk_mq_start_request(bd->rq); 1659 1660 if (lo->lo_state != Lo_bound) 1661 return BLK_MQ_RQ_QUEUE_ERROR; 1662 1663 switch (req_op(cmd->rq)) { 1664 case REQ_OP_FLUSH: 1665 case REQ_OP_DISCARD: 1666 case REQ_OP_WRITE_ZEROES: 1667 cmd->use_aio = false; 1668 break; 1669 default: 1670 cmd->use_aio = lo->use_dio; 1671 break; 1672 } 1673 1674 kthread_queue_work(&lo->worker, &cmd->work); 1675 1676 return BLK_MQ_RQ_QUEUE_OK; 1677 } 1678 1679 static void loop_handle_cmd(struct loop_cmd *cmd) 1680 { 1681 const bool write = op_is_write(req_op(cmd->rq)); 1682 struct loop_device *lo = cmd->rq->q->queuedata; 1683 int ret = 0; 1684 1685 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1686 ret = -EIO; 1687 goto failed; 1688 } 1689 1690 ret = do_req_filebacked(lo, cmd->rq); 1691 failed: 1692 /* complete non-aio request */ 1693 if (!cmd->use_aio || ret) 1694 blk_mq_complete_request(cmd->rq, ret ? -EIO : 0); 1695 } 1696 1697 static void loop_queue_work(struct kthread_work *work) 1698 { 1699 struct loop_cmd *cmd = 1700 container_of(work, struct loop_cmd, work); 1701 1702 loop_handle_cmd(cmd); 1703 } 1704 1705 static int loop_init_request(void *data, struct request *rq, 1706 unsigned int hctx_idx, unsigned int request_idx, 1707 unsigned int numa_node) 1708 { 1709 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1710 1711 cmd->rq = rq; 1712 kthread_init_work(&cmd->work, loop_queue_work); 1713 1714 return 0; 1715 } 1716 1717 static const struct blk_mq_ops loop_mq_ops = { 1718 .queue_rq = loop_queue_rq, 1719 .init_request = loop_init_request, 1720 }; 1721 1722 static int loop_add(struct loop_device **l, int i) 1723 { 1724 struct loop_device *lo; 1725 struct gendisk *disk; 1726 int err; 1727 1728 err = -ENOMEM; 1729 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1730 if (!lo) 1731 goto out; 1732 1733 lo->lo_state = Lo_unbound; 1734 1735 /* allocate id, if @id >= 0, we're requesting that specific id */ 1736 if (i >= 0) { 1737 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1738 if (err == -ENOSPC) 1739 err = -EEXIST; 1740 } else { 1741 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1742 } 1743 if (err < 0) 1744 goto out_free_dev; 1745 i = err; 1746 1747 err = -ENOMEM; 1748 lo->tag_set.ops = &loop_mq_ops; 1749 lo->tag_set.nr_hw_queues = 1; 1750 lo->tag_set.queue_depth = 128; 1751 lo->tag_set.numa_node = NUMA_NO_NODE; 1752 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1753 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1754 lo->tag_set.driver_data = lo; 1755 1756 err = blk_mq_alloc_tag_set(&lo->tag_set); 1757 if (err) 1758 goto out_free_idr; 1759 1760 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1761 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1762 err = PTR_ERR(lo->lo_queue); 1763 goto out_cleanup_tags; 1764 } 1765 lo->lo_queue->queuedata = lo; 1766 1767 /* 1768 * It doesn't make sense to enable merge because the I/O 1769 * submitted to backing file is handled page by page. 1770 */ 1771 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1772 1773 err = -ENOMEM; 1774 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1775 if (!disk) 1776 goto out_free_queue; 1777 1778 /* 1779 * Disable partition scanning by default. The in-kernel partition 1780 * scanning can be requested individually per-device during its 1781 * setup. Userspace can always add and remove partitions from all 1782 * devices. The needed partition minors are allocated from the 1783 * extended minor space, the main loop device numbers will continue 1784 * to match the loop minors, regardless of the number of partitions 1785 * used. 1786 * 1787 * If max_part is given, partition scanning is globally enabled for 1788 * all loop devices. The minors for the main loop devices will be 1789 * multiples of max_part. 1790 * 1791 * Note: Global-for-all-devices, set-only-at-init, read-only module 1792 * parameteters like 'max_loop' and 'max_part' make things needlessly 1793 * complicated, are too static, inflexible and may surprise 1794 * userspace tools. Parameters like this in general should be avoided. 1795 */ 1796 if (!part_shift) 1797 disk->flags |= GENHD_FL_NO_PART_SCAN; 1798 disk->flags |= GENHD_FL_EXT_DEVT; 1799 mutex_init(&lo->lo_ctl_mutex); 1800 atomic_set(&lo->lo_refcnt, 0); 1801 lo->lo_number = i; 1802 spin_lock_init(&lo->lo_lock); 1803 disk->major = LOOP_MAJOR; 1804 disk->first_minor = i << part_shift; 1805 disk->fops = &lo_fops; 1806 disk->private_data = lo; 1807 disk->queue = lo->lo_queue; 1808 sprintf(disk->disk_name, "loop%d", i); 1809 add_disk(disk); 1810 *l = lo; 1811 return lo->lo_number; 1812 1813 out_free_queue: 1814 blk_cleanup_queue(lo->lo_queue); 1815 out_cleanup_tags: 1816 blk_mq_free_tag_set(&lo->tag_set); 1817 out_free_idr: 1818 idr_remove(&loop_index_idr, i); 1819 out_free_dev: 1820 kfree(lo); 1821 out: 1822 return err; 1823 } 1824 1825 static void loop_remove(struct loop_device *lo) 1826 { 1827 blk_cleanup_queue(lo->lo_queue); 1828 del_gendisk(lo->lo_disk); 1829 blk_mq_free_tag_set(&lo->tag_set); 1830 put_disk(lo->lo_disk); 1831 kfree(lo); 1832 } 1833 1834 static int find_free_cb(int id, void *ptr, void *data) 1835 { 1836 struct loop_device *lo = ptr; 1837 struct loop_device **l = data; 1838 1839 if (lo->lo_state == Lo_unbound) { 1840 *l = lo; 1841 return 1; 1842 } 1843 return 0; 1844 } 1845 1846 static int loop_lookup(struct loop_device **l, int i) 1847 { 1848 struct loop_device *lo; 1849 int ret = -ENODEV; 1850 1851 if (i < 0) { 1852 int err; 1853 1854 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1855 if (err == 1) { 1856 *l = lo; 1857 ret = lo->lo_number; 1858 } 1859 goto out; 1860 } 1861 1862 /* lookup and return a specific i */ 1863 lo = idr_find(&loop_index_idr, i); 1864 if (lo) { 1865 *l = lo; 1866 ret = lo->lo_number; 1867 } 1868 out: 1869 return ret; 1870 } 1871 1872 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1873 { 1874 struct loop_device *lo; 1875 struct kobject *kobj; 1876 int err; 1877 1878 mutex_lock(&loop_index_mutex); 1879 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1880 if (err < 0) 1881 err = loop_add(&lo, MINOR(dev) >> part_shift); 1882 if (err < 0) 1883 kobj = NULL; 1884 else 1885 kobj = get_disk(lo->lo_disk); 1886 mutex_unlock(&loop_index_mutex); 1887 1888 *part = 0; 1889 return kobj; 1890 } 1891 1892 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1893 unsigned long parm) 1894 { 1895 struct loop_device *lo; 1896 int ret = -ENOSYS; 1897 1898 mutex_lock(&loop_index_mutex); 1899 switch (cmd) { 1900 case LOOP_CTL_ADD: 1901 ret = loop_lookup(&lo, parm); 1902 if (ret >= 0) { 1903 ret = -EEXIST; 1904 break; 1905 } 1906 ret = loop_add(&lo, parm); 1907 break; 1908 case LOOP_CTL_REMOVE: 1909 ret = loop_lookup(&lo, parm); 1910 if (ret < 0) 1911 break; 1912 mutex_lock(&lo->lo_ctl_mutex); 1913 if (lo->lo_state != Lo_unbound) { 1914 ret = -EBUSY; 1915 mutex_unlock(&lo->lo_ctl_mutex); 1916 break; 1917 } 1918 if (atomic_read(&lo->lo_refcnt) > 0) { 1919 ret = -EBUSY; 1920 mutex_unlock(&lo->lo_ctl_mutex); 1921 break; 1922 } 1923 lo->lo_disk->private_data = NULL; 1924 mutex_unlock(&lo->lo_ctl_mutex); 1925 idr_remove(&loop_index_idr, lo->lo_number); 1926 loop_remove(lo); 1927 break; 1928 case LOOP_CTL_GET_FREE: 1929 ret = loop_lookup(&lo, -1); 1930 if (ret >= 0) 1931 break; 1932 ret = loop_add(&lo, -1); 1933 } 1934 mutex_unlock(&loop_index_mutex); 1935 1936 return ret; 1937 } 1938 1939 static const struct file_operations loop_ctl_fops = { 1940 .open = nonseekable_open, 1941 .unlocked_ioctl = loop_control_ioctl, 1942 .compat_ioctl = loop_control_ioctl, 1943 .owner = THIS_MODULE, 1944 .llseek = noop_llseek, 1945 }; 1946 1947 static struct miscdevice loop_misc = { 1948 .minor = LOOP_CTRL_MINOR, 1949 .name = "loop-control", 1950 .fops = &loop_ctl_fops, 1951 }; 1952 1953 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 1954 MODULE_ALIAS("devname:loop-control"); 1955 1956 static int __init loop_init(void) 1957 { 1958 int i, nr; 1959 unsigned long range; 1960 struct loop_device *lo; 1961 int err; 1962 1963 err = misc_register(&loop_misc); 1964 if (err < 0) 1965 return err; 1966 1967 part_shift = 0; 1968 if (max_part > 0) { 1969 part_shift = fls(max_part); 1970 1971 /* 1972 * Adjust max_part according to part_shift as it is exported 1973 * to user space so that user can decide correct minor number 1974 * if [s]he want to create more devices. 1975 * 1976 * Note that -1 is required because partition 0 is reserved 1977 * for the whole disk. 1978 */ 1979 max_part = (1UL << part_shift) - 1; 1980 } 1981 1982 if ((1UL << part_shift) > DISK_MAX_PARTS) { 1983 err = -EINVAL; 1984 goto misc_out; 1985 } 1986 1987 if (max_loop > 1UL << (MINORBITS - part_shift)) { 1988 err = -EINVAL; 1989 goto misc_out; 1990 } 1991 1992 /* 1993 * If max_loop is specified, create that many devices upfront. 1994 * This also becomes a hard limit. If max_loop is not specified, 1995 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 1996 * init time. Loop devices can be requested on-demand with the 1997 * /dev/loop-control interface, or be instantiated by accessing 1998 * a 'dead' device node. 1999 */ 2000 if (max_loop) { 2001 nr = max_loop; 2002 range = max_loop << part_shift; 2003 } else { 2004 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2005 range = 1UL << MINORBITS; 2006 } 2007 2008 if (register_blkdev(LOOP_MAJOR, "loop")) { 2009 err = -EIO; 2010 goto misc_out; 2011 } 2012 2013 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2014 THIS_MODULE, loop_probe, NULL, NULL); 2015 2016 /* pre-create number of devices given by config or max_loop */ 2017 mutex_lock(&loop_index_mutex); 2018 for (i = 0; i < nr; i++) 2019 loop_add(&lo, i); 2020 mutex_unlock(&loop_index_mutex); 2021 2022 printk(KERN_INFO "loop: module loaded\n"); 2023 return 0; 2024 2025 misc_out: 2026 misc_deregister(&loop_misc); 2027 return err; 2028 } 2029 2030 static int loop_exit_cb(int id, void *ptr, void *data) 2031 { 2032 struct loop_device *lo = ptr; 2033 2034 loop_remove(lo); 2035 return 0; 2036 } 2037 2038 static void __exit loop_exit(void) 2039 { 2040 unsigned long range; 2041 2042 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2043 2044 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2045 idr_destroy(&loop_index_idr); 2046 2047 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2048 unregister_blkdev(LOOP_MAJOR, "loop"); 2049 2050 misc_deregister(&loop_misc); 2051 } 2052 2053 module_init(loop_init); 2054 module_exit(loop_exit); 2055 2056 #ifndef MODULE 2057 static int __init max_loop_setup(char *str) 2058 { 2059 max_loop = simple_strtol(str, NULL, 0); 2060 return 1; 2061 } 2062 2063 __setup("max_loop=", max_loop_setup); 2064 #endif 2065