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