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 loff_t logical_blocksize) 226 { 227 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 228 sector_t x = (sector_t)size; 229 struct block_device *bdev = lo->lo_device; 230 231 if (unlikely((loff_t)x != size)) 232 return -EFBIG; 233 if (lo->lo_offset != offset) 234 lo->lo_offset = offset; 235 if (lo->lo_sizelimit != sizelimit) 236 lo->lo_sizelimit = sizelimit; 237 if (lo->lo_flags & LO_FLAGS_BLOCKSIZE) { 238 lo->lo_logical_blocksize = logical_blocksize; 239 blk_queue_physical_block_size(lo->lo_queue, lo->lo_blocksize); 240 blk_queue_logical_block_size(lo->lo_queue, 241 lo->lo_logical_blocksize); 242 } 243 set_capacity(lo->lo_disk, x); 244 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 245 /* let user-space know about the new size */ 246 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 247 return 0; 248 } 249 250 static inline int 251 lo_do_transfer(struct loop_device *lo, int cmd, 252 struct page *rpage, unsigned roffs, 253 struct page *lpage, unsigned loffs, 254 int size, sector_t rblock) 255 { 256 int ret; 257 258 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 259 if (likely(!ret)) 260 return 0; 261 262 printk_ratelimited(KERN_ERR 263 "loop: Transfer error at byte offset %llu, length %i.\n", 264 (unsigned long long)rblock << 9, size); 265 return ret; 266 } 267 268 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 269 { 270 struct iov_iter i; 271 ssize_t bw; 272 273 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len); 274 275 file_start_write(file); 276 bw = vfs_iter_write(file, &i, ppos); 277 file_end_write(file); 278 279 if (likely(bw == bvec->bv_len)) 280 return 0; 281 282 printk_ratelimited(KERN_ERR 283 "loop: Write error at byte offset %llu, length %i.\n", 284 (unsigned long long)*ppos, bvec->bv_len); 285 if (bw >= 0) 286 bw = -EIO; 287 return bw; 288 } 289 290 static int lo_write_simple(struct loop_device *lo, struct request *rq, 291 loff_t pos) 292 { 293 struct bio_vec bvec; 294 struct req_iterator iter; 295 int ret = 0; 296 297 rq_for_each_segment(bvec, rq, iter) { 298 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 299 if (ret < 0) 300 break; 301 cond_resched(); 302 } 303 304 return ret; 305 } 306 307 /* 308 * This is the slow, transforming version that needs to double buffer the 309 * data as it cannot do the transformations in place without having direct 310 * access to the destination pages of the backing file. 311 */ 312 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 313 loff_t pos) 314 { 315 struct bio_vec bvec, b; 316 struct req_iterator iter; 317 struct page *page; 318 int ret = 0; 319 320 page = alloc_page(GFP_NOIO); 321 if (unlikely(!page)) 322 return -ENOMEM; 323 324 rq_for_each_segment(bvec, rq, iter) { 325 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 326 bvec.bv_offset, bvec.bv_len, pos >> 9); 327 if (unlikely(ret)) 328 break; 329 330 b.bv_page = page; 331 b.bv_offset = 0; 332 b.bv_len = bvec.bv_len; 333 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 334 if (ret < 0) 335 break; 336 } 337 338 __free_page(page); 339 return ret; 340 } 341 342 static int lo_read_simple(struct loop_device *lo, struct request *rq, 343 loff_t pos) 344 { 345 struct bio_vec bvec; 346 struct req_iterator iter; 347 struct iov_iter i; 348 ssize_t len; 349 350 rq_for_each_segment(bvec, rq, iter) { 351 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); 352 len = vfs_iter_read(lo->lo_backing_file, &i, &pos); 353 if (len < 0) 354 return len; 355 356 flush_dcache_page(bvec.bv_page); 357 358 if (len != bvec.bv_len) { 359 struct bio *bio; 360 361 __rq_for_each_bio(bio, rq) 362 zero_fill_bio(bio); 363 break; 364 } 365 cond_resched(); 366 } 367 368 return 0; 369 } 370 371 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 372 loff_t pos) 373 { 374 struct bio_vec bvec, b; 375 struct req_iterator iter; 376 struct iov_iter i; 377 struct page *page; 378 ssize_t len; 379 int ret = 0; 380 381 page = alloc_page(GFP_NOIO); 382 if (unlikely(!page)) 383 return -ENOMEM; 384 385 rq_for_each_segment(bvec, rq, iter) { 386 loff_t offset = pos; 387 388 b.bv_page = page; 389 b.bv_offset = 0; 390 b.bv_len = bvec.bv_len; 391 392 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); 393 len = vfs_iter_read(lo->lo_backing_file, &i, &pos); 394 if (len < 0) { 395 ret = len; 396 goto out_free_page; 397 } 398 399 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 400 bvec.bv_offset, len, offset >> 9); 401 if (ret) 402 goto out_free_page; 403 404 flush_dcache_page(bvec.bv_page); 405 406 if (len != bvec.bv_len) { 407 struct bio *bio; 408 409 __rq_for_each_bio(bio, rq) 410 zero_fill_bio(bio); 411 break; 412 } 413 } 414 415 ret = 0; 416 out_free_page: 417 __free_page(page); 418 return ret; 419 } 420 421 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 422 { 423 /* 424 * We use punch hole to reclaim the free space used by the 425 * image a.k.a. discard. However we do not support discard if 426 * encryption is enabled, because it may give an attacker 427 * useful information. 428 */ 429 struct file *file = lo->lo_backing_file; 430 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 431 int ret; 432 433 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 434 ret = -EOPNOTSUPP; 435 goto out; 436 } 437 438 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 439 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 440 ret = -EIO; 441 out: 442 return ret; 443 } 444 445 static int lo_req_flush(struct loop_device *lo, struct request *rq) 446 { 447 struct file *file = lo->lo_backing_file; 448 int ret = vfs_fsync(file, 0); 449 if (unlikely(ret && ret != -EINVAL)) 450 ret = -EIO; 451 452 return ret; 453 } 454 455 static void lo_complete_rq(struct request *rq) 456 { 457 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 458 459 if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio && 460 cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) { 461 struct bio *bio = cmd->rq->bio; 462 463 bio_advance(bio, cmd->ret); 464 zero_fill_bio(bio); 465 } 466 467 blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK); 468 } 469 470 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 471 { 472 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 473 474 cmd->ret = ret; 475 blk_mq_complete_request(cmd->rq); 476 } 477 478 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 479 loff_t pos, bool rw) 480 { 481 struct iov_iter iter; 482 struct bio_vec *bvec; 483 struct bio *bio = cmd->rq->bio; 484 struct file *file = lo->lo_backing_file; 485 int ret; 486 487 /* nomerge for loop request queue */ 488 WARN_ON(cmd->rq->bio != cmd->rq->biotail); 489 490 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 491 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, 492 bio_segments(bio), blk_rq_bytes(cmd->rq)); 493 /* 494 * This bio may be started from the middle of the 'bvec' 495 * because of bio splitting, so offset from the bvec must 496 * be passed to iov iterator 497 */ 498 iter.iov_offset = bio->bi_iter.bi_bvec_done; 499 500 cmd->iocb.ki_pos = pos; 501 cmd->iocb.ki_filp = file; 502 cmd->iocb.ki_complete = lo_rw_aio_complete; 503 cmd->iocb.ki_flags = IOCB_DIRECT; 504 505 if (rw == WRITE) 506 ret = call_write_iter(file, &cmd->iocb, &iter); 507 else 508 ret = call_read_iter(file, &cmd->iocb, &iter); 509 510 if (ret != -EIOCBQUEUED) 511 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 512 return 0; 513 } 514 515 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 516 { 517 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 518 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 519 520 /* 521 * lo_write_simple and lo_read_simple should have been covered 522 * by io submit style function like lo_rw_aio(), one blocker 523 * is that lo_read_simple() need to call flush_dcache_page after 524 * the page is written from kernel, and it isn't easy to handle 525 * this in io submit style function which submits all segments 526 * of the req at one time. And direct read IO doesn't need to 527 * run flush_dcache_page(). 528 */ 529 switch (req_op(rq)) { 530 case REQ_OP_FLUSH: 531 return lo_req_flush(lo, rq); 532 case REQ_OP_DISCARD: 533 case REQ_OP_WRITE_ZEROES: 534 return lo_discard(lo, rq, pos); 535 case REQ_OP_WRITE: 536 if (lo->transfer) 537 return lo_write_transfer(lo, rq, pos); 538 else if (cmd->use_aio) 539 return lo_rw_aio(lo, cmd, pos, WRITE); 540 else 541 return lo_write_simple(lo, rq, pos); 542 case REQ_OP_READ: 543 if (lo->transfer) 544 return lo_read_transfer(lo, rq, pos); 545 else if (cmd->use_aio) 546 return lo_rw_aio(lo, cmd, pos, READ); 547 else 548 return lo_read_simple(lo, rq, pos); 549 default: 550 WARN_ON_ONCE(1); 551 return -EIO; 552 break; 553 } 554 } 555 556 struct switch_request { 557 struct file *file; 558 struct completion wait; 559 }; 560 561 static inline void loop_update_dio(struct loop_device *lo) 562 { 563 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 564 lo->use_dio); 565 } 566 567 /* 568 * Do the actual switch; called from the BIO completion routine 569 */ 570 static void do_loop_switch(struct loop_device *lo, struct switch_request *p) 571 { 572 struct file *file = p->file; 573 struct file *old_file = lo->lo_backing_file; 574 struct address_space *mapping; 575 576 /* if no new file, only flush of queued bios requested */ 577 if (!file) 578 return; 579 580 mapping = file->f_mapping; 581 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 582 lo->lo_backing_file = file; 583 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? 584 mapping->host->i_bdev->bd_block_size : PAGE_SIZE; 585 lo->old_gfp_mask = mapping_gfp_mask(mapping); 586 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 587 loop_update_dio(lo); 588 } 589 590 /* 591 * loop_switch performs the hard work of switching a backing store. 592 * First it needs to flush existing IO, it does this by sending a magic 593 * BIO down the pipe. The completion of this BIO does the actual switch. 594 */ 595 static int loop_switch(struct loop_device *lo, struct file *file) 596 { 597 struct switch_request w; 598 599 w.file = file; 600 601 /* freeze queue and wait for completion of scheduled requests */ 602 blk_mq_freeze_queue(lo->lo_queue); 603 604 /* do the switch action */ 605 do_loop_switch(lo, &w); 606 607 /* unfreeze */ 608 blk_mq_unfreeze_queue(lo->lo_queue); 609 610 return 0; 611 } 612 613 /* 614 * Helper to flush the IOs in loop, but keeping loop thread running 615 */ 616 static int loop_flush(struct loop_device *lo) 617 { 618 return loop_switch(lo, NULL); 619 } 620 621 static void loop_reread_partitions(struct loop_device *lo, 622 struct block_device *bdev) 623 { 624 int rc; 625 626 /* 627 * bd_mutex has been held already in release path, so don't 628 * acquire it if this function is called in such case. 629 * 630 * If the reread partition isn't from release path, lo_refcnt 631 * must be at least one and it can only become zero when the 632 * current holder is released. 633 */ 634 if (!atomic_read(&lo->lo_refcnt)) 635 rc = __blkdev_reread_part(bdev); 636 else 637 rc = blkdev_reread_part(bdev); 638 if (rc) 639 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 640 __func__, lo->lo_number, lo->lo_file_name, rc); 641 } 642 643 /* 644 * loop_change_fd switched the backing store of a loopback device to 645 * a new file. This is useful for operating system installers to free up 646 * the original file and in High Availability environments to switch to 647 * an alternative location for the content in case of server meltdown. 648 * This can only work if the loop device is used read-only, and if the 649 * new backing store is the same size and type as the old backing store. 650 */ 651 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 652 unsigned int arg) 653 { 654 struct file *file, *old_file; 655 struct inode *inode; 656 int error; 657 658 error = -ENXIO; 659 if (lo->lo_state != Lo_bound) 660 goto out; 661 662 /* the loop device has to be read-only */ 663 error = -EINVAL; 664 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 665 goto out; 666 667 error = -EBADF; 668 file = fget(arg); 669 if (!file) 670 goto out; 671 672 inode = file->f_mapping->host; 673 old_file = lo->lo_backing_file; 674 675 error = -EINVAL; 676 677 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 678 goto out_putf; 679 680 /* size of the new backing store needs to be the same */ 681 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 682 goto out_putf; 683 684 /* and ... switch */ 685 error = loop_switch(lo, file); 686 if (error) 687 goto out_putf; 688 689 fput(old_file); 690 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 691 loop_reread_partitions(lo, bdev); 692 return 0; 693 694 out_putf: 695 fput(file); 696 out: 697 return error; 698 } 699 700 static inline int is_loop_device(struct file *file) 701 { 702 struct inode *i = file->f_mapping->host; 703 704 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 705 } 706 707 /* loop sysfs attributes */ 708 709 static ssize_t loop_attr_show(struct device *dev, char *page, 710 ssize_t (*callback)(struct loop_device *, char *)) 711 { 712 struct gendisk *disk = dev_to_disk(dev); 713 struct loop_device *lo = disk->private_data; 714 715 return callback(lo, page); 716 } 717 718 #define LOOP_ATTR_RO(_name) \ 719 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 720 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 721 struct device_attribute *attr, char *b) \ 722 { \ 723 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 724 } \ 725 static struct device_attribute loop_attr_##_name = \ 726 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); 727 728 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 729 { 730 ssize_t ret; 731 char *p = NULL; 732 733 spin_lock_irq(&lo->lo_lock); 734 if (lo->lo_backing_file) 735 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 736 spin_unlock_irq(&lo->lo_lock); 737 738 if (IS_ERR_OR_NULL(p)) 739 ret = PTR_ERR(p); 740 else { 741 ret = strlen(p); 742 memmove(buf, p, ret); 743 buf[ret++] = '\n'; 744 buf[ret] = 0; 745 } 746 747 return ret; 748 } 749 750 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 751 { 752 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 753 } 754 755 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 756 { 757 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 758 } 759 760 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 761 { 762 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 763 764 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 765 } 766 767 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 768 { 769 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 770 771 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 772 } 773 774 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 775 { 776 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 777 778 return sprintf(buf, "%s\n", dio ? "1" : "0"); 779 } 780 781 LOOP_ATTR_RO(backing_file); 782 LOOP_ATTR_RO(offset); 783 LOOP_ATTR_RO(sizelimit); 784 LOOP_ATTR_RO(autoclear); 785 LOOP_ATTR_RO(partscan); 786 LOOP_ATTR_RO(dio); 787 788 static struct attribute *loop_attrs[] = { 789 &loop_attr_backing_file.attr, 790 &loop_attr_offset.attr, 791 &loop_attr_sizelimit.attr, 792 &loop_attr_autoclear.attr, 793 &loop_attr_partscan.attr, 794 &loop_attr_dio.attr, 795 NULL, 796 }; 797 798 static struct attribute_group loop_attribute_group = { 799 .name = "loop", 800 .attrs= loop_attrs, 801 }; 802 803 static int loop_sysfs_init(struct loop_device *lo) 804 { 805 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 806 &loop_attribute_group); 807 } 808 809 static void loop_sysfs_exit(struct loop_device *lo) 810 { 811 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 812 &loop_attribute_group); 813 } 814 815 static void loop_config_discard(struct loop_device *lo) 816 { 817 struct file *file = lo->lo_backing_file; 818 struct inode *inode = file->f_mapping->host; 819 struct request_queue *q = lo->lo_queue; 820 int lo_bits = 9; 821 822 /* 823 * We use punch hole to reclaim the free space used by the 824 * image a.k.a. discard. However we do not support discard if 825 * encryption is enabled, because it may give an attacker 826 * useful information. 827 */ 828 if ((!file->f_op->fallocate) || 829 lo->lo_encrypt_key_size) { 830 q->limits.discard_granularity = 0; 831 q->limits.discard_alignment = 0; 832 blk_queue_max_discard_sectors(q, 0); 833 blk_queue_max_write_zeroes_sectors(q, 0); 834 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 835 return; 836 } 837 838 q->limits.discard_granularity = inode->i_sb->s_blocksize; 839 q->limits.discard_alignment = 0; 840 if (lo->lo_flags & LO_FLAGS_BLOCKSIZE) 841 lo_bits = blksize_bits(lo->lo_logical_blocksize); 842 843 blk_queue_max_discard_sectors(q, UINT_MAX >> lo_bits); 844 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> lo_bits); 845 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 846 } 847 848 static void loop_unprepare_queue(struct loop_device *lo) 849 { 850 kthread_flush_worker(&lo->worker); 851 kthread_stop(lo->worker_task); 852 } 853 854 static int loop_prepare_queue(struct loop_device *lo) 855 { 856 kthread_init_worker(&lo->worker); 857 lo->worker_task = kthread_run(kthread_worker_fn, 858 &lo->worker, "loop%d", lo->lo_number); 859 if (IS_ERR(lo->worker_task)) 860 return -ENOMEM; 861 set_user_nice(lo->worker_task, MIN_NICE); 862 return 0; 863 } 864 865 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 866 struct block_device *bdev, unsigned int arg) 867 { 868 struct file *file, *f; 869 struct inode *inode; 870 struct address_space *mapping; 871 unsigned lo_blocksize; 872 int lo_flags = 0; 873 int error; 874 loff_t size; 875 876 /* This is safe, since we have a reference from open(). */ 877 __module_get(THIS_MODULE); 878 879 error = -EBADF; 880 file = fget(arg); 881 if (!file) 882 goto out; 883 884 error = -EBUSY; 885 if (lo->lo_state != Lo_unbound) 886 goto out_putf; 887 888 /* Avoid recursion */ 889 f = file; 890 while (is_loop_device(f)) { 891 struct loop_device *l; 892 893 if (f->f_mapping->host->i_bdev == bdev) 894 goto out_putf; 895 896 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 897 if (l->lo_state == Lo_unbound) { 898 error = -EINVAL; 899 goto out_putf; 900 } 901 f = l->lo_backing_file; 902 } 903 904 mapping = file->f_mapping; 905 inode = mapping->host; 906 907 error = -EINVAL; 908 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 909 goto out_putf; 910 911 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 912 !file->f_op->write_iter) 913 lo_flags |= LO_FLAGS_READ_ONLY; 914 915 lo_blocksize = S_ISBLK(inode->i_mode) ? 916 inode->i_bdev->bd_block_size : PAGE_SIZE; 917 918 error = -EFBIG; 919 size = get_loop_size(lo, file); 920 if ((loff_t)(sector_t)size != size) 921 goto out_putf; 922 error = loop_prepare_queue(lo); 923 if (error) 924 goto out_putf; 925 926 error = 0; 927 928 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 929 930 lo->use_dio = false; 931 lo->lo_blocksize = lo_blocksize; 932 lo->lo_logical_blocksize = 512; 933 lo->lo_device = bdev; 934 lo->lo_flags = lo_flags; 935 lo->lo_backing_file = file; 936 lo->transfer = NULL; 937 lo->ioctl = NULL; 938 lo->lo_sizelimit = 0; 939 lo->old_gfp_mask = mapping_gfp_mask(mapping); 940 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 941 942 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 943 blk_queue_write_cache(lo->lo_queue, true, false); 944 945 loop_update_dio(lo); 946 set_capacity(lo->lo_disk, size); 947 bd_set_size(bdev, size << 9); 948 loop_sysfs_init(lo); 949 /* let user-space know about the new size */ 950 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 951 952 set_blocksize(bdev, lo_blocksize); 953 954 lo->lo_state = Lo_bound; 955 if (part_shift) 956 lo->lo_flags |= LO_FLAGS_PARTSCAN; 957 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 958 loop_reread_partitions(lo, bdev); 959 960 /* Grab the block_device to prevent its destruction after we 961 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 962 */ 963 bdgrab(bdev); 964 return 0; 965 966 out_putf: 967 fput(file); 968 out: 969 /* This is safe: open() is still holding a reference. */ 970 module_put(THIS_MODULE); 971 return error; 972 } 973 974 static int 975 loop_release_xfer(struct loop_device *lo) 976 { 977 int err = 0; 978 struct loop_func_table *xfer = lo->lo_encryption; 979 980 if (xfer) { 981 if (xfer->release) 982 err = xfer->release(lo); 983 lo->transfer = NULL; 984 lo->lo_encryption = NULL; 985 module_put(xfer->owner); 986 } 987 return err; 988 } 989 990 static int 991 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 992 const struct loop_info64 *i) 993 { 994 int err = 0; 995 996 if (xfer) { 997 struct module *owner = xfer->owner; 998 999 if (!try_module_get(owner)) 1000 return -EINVAL; 1001 if (xfer->init) 1002 err = xfer->init(lo, i); 1003 if (err) 1004 module_put(owner); 1005 else 1006 lo->lo_encryption = xfer; 1007 } 1008 return err; 1009 } 1010 1011 static int loop_clr_fd(struct loop_device *lo) 1012 { 1013 struct file *filp = lo->lo_backing_file; 1014 gfp_t gfp = lo->old_gfp_mask; 1015 struct block_device *bdev = lo->lo_device; 1016 1017 if (lo->lo_state != Lo_bound) 1018 return -ENXIO; 1019 1020 /* 1021 * If we've explicitly asked to tear down the loop device, 1022 * and it has an elevated reference count, set it for auto-teardown when 1023 * the last reference goes away. This stops $!~#$@ udev from 1024 * preventing teardown because it decided that it needs to run blkid on 1025 * the loopback device whenever they appear. xfstests is notorious for 1026 * failing tests because blkid via udev races with a losetup 1027 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1028 * command to fail with EBUSY. 1029 */ 1030 if (atomic_read(&lo->lo_refcnt) > 1) { 1031 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1032 mutex_unlock(&lo->lo_ctl_mutex); 1033 return 0; 1034 } 1035 1036 if (filp == NULL) 1037 return -EINVAL; 1038 1039 /* freeze request queue during the transition */ 1040 blk_mq_freeze_queue(lo->lo_queue); 1041 1042 spin_lock_irq(&lo->lo_lock); 1043 lo->lo_state = Lo_rundown; 1044 lo->lo_backing_file = NULL; 1045 spin_unlock_irq(&lo->lo_lock); 1046 1047 loop_release_xfer(lo); 1048 lo->transfer = NULL; 1049 lo->ioctl = NULL; 1050 lo->lo_device = NULL; 1051 lo->lo_encryption = NULL; 1052 lo->lo_offset = 0; 1053 lo->lo_sizelimit = 0; 1054 lo->lo_encrypt_key_size = 0; 1055 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1056 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1057 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1058 if (bdev) { 1059 bdput(bdev); 1060 invalidate_bdev(bdev); 1061 } 1062 set_capacity(lo->lo_disk, 0); 1063 loop_sysfs_exit(lo); 1064 if (bdev) { 1065 bd_set_size(bdev, 0); 1066 /* let user-space know about this change */ 1067 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1068 } 1069 mapping_set_gfp_mask(filp->f_mapping, gfp); 1070 lo->lo_state = Lo_unbound; 1071 /* This is safe: open() is still holding a reference. */ 1072 module_put(THIS_MODULE); 1073 blk_mq_unfreeze_queue(lo->lo_queue); 1074 1075 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 1076 loop_reread_partitions(lo, bdev); 1077 lo->lo_flags = 0; 1078 if (!part_shift) 1079 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1080 loop_unprepare_queue(lo); 1081 mutex_unlock(&lo->lo_ctl_mutex); 1082 /* 1083 * Need not hold lo_ctl_mutex to fput backing file. 1084 * Calling fput holding lo_ctl_mutex triggers a circular 1085 * lock dependency possibility warning as fput can take 1086 * bd_mutex which is usually taken before lo_ctl_mutex. 1087 */ 1088 fput(filp); 1089 return 0; 1090 } 1091 1092 static int 1093 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1094 { 1095 int err; 1096 struct loop_func_table *xfer; 1097 kuid_t uid = current_uid(); 1098 int lo_flags = lo->lo_flags; 1099 1100 if (lo->lo_encrypt_key_size && 1101 !uid_eq(lo->lo_key_owner, uid) && 1102 !capable(CAP_SYS_ADMIN)) 1103 return -EPERM; 1104 if (lo->lo_state != Lo_bound) 1105 return -ENXIO; 1106 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1107 return -EINVAL; 1108 1109 /* I/O need to be drained during transfer transition */ 1110 blk_mq_freeze_queue(lo->lo_queue); 1111 1112 err = loop_release_xfer(lo); 1113 if (err) 1114 goto exit; 1115 1116 if (info->lo_encrypt_type) { 1117 unsigned int type = info->lo_encrypt_type; 1118 1119 if (type >= MAX_LO_CRYPT) 1120 return -EINVAL; 1121 xfer = xfer_funcs[type]; 1122 if (xfer == NULL) 1123 return -EINVAL; 1124 } else 1125 xfer = NULL; 1126 1127 err = loop_init_xfer(lo, xfer, info); 1128 if (err) 1129 goto exit; 1130 1131 if (info->lo_flags & LO_FLAGS_BLOCKSIZE) { 1132 if (!(lo->lo_flags & LO_FLAGS_BLOCKSIZE)) 1133 lo->lo_logical_blocksize = 512; 1134 lo->lo_flags |= LO_FLAGS_BLOCKSIZE; 1135 if (LO_INFO_BLOCKSIZE(info) != 512 && 1136 LO_INFO_BLOCKSIZE(info) != 1024 && 1137 LO_INFO_BLOCKSIZE(info) != 2048 && 1138 LO_INFO_BLOCKSIZE(info) != 4096) 1139 return -EINVAL; 1140 if (LO_INFO_BLOCKSIZE(info) > lo->lo_blocksize) 1141 return -EINVAL; 1142 } 1143 1144 if (lo->lo_offset != info->lo_offset || 1145 lo->lo_sizelimit != info->lo_sizelimit || 1146 lo->lo_flags != lo_flags || 1147 ((lo->lo_flags & LO_FLAGS_BLOCKSIZE) && 1148 lo->lo_logical_blocksize != LO_INFO_BLOCKSIZE(info))) { 1149 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit, 1150 LO_INFO_BLOCKSIZE(info))) { 1151 err = -EFBIG; 1152 goto exit; 1153 } 1154 } 1155 1156 loop_config_discard(lo); 1157 1158 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1159 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1160 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1161 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1162 1163 if (!xfer) 1164 xfer = &none_funcs; 1165 lo->transfer = xfer->transfer; 1166 lo->ioctl = xfer->ioctl; 1167 1168 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1169 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1170 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1171 1172 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1173 lo->lo_init[0] = info->lo_init[0]; 1174 lo->lo_init[1] = info->lo_init[1]; 1175 if (info->lo_encrypt_key_size) { 1176 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1177 info->lo_encrypt_key_size); 1178 lo->lo_key_owner = uid; 1179 } 1180 1181 /* update dio if lo_offset or transfer is changed */ 1182 __loop_update_dio(lo, lo->use_dio); 1183 1184 exit: 1185 blk_mq_unfreeze_queue(lo->lo_queue); 1186 1187 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1188 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1189 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1190 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1191 loop_reread_partitions(lo, lo->lo_device); 1192 } 1193 1194 return err; 1195 } 1196 1197 static int 1198 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1199 { 1200 struct file *file = lo->lo_backing_file; 1201 struct kstat stat; 1202 int error; 1203 1204 if (lo->lo_state != Lo_bound) 1205 return -ENXIO; 1206 error = vfs_getattr(&file->f_path, &stat, 1207 STATX_INO, AT_STATX_SYNC_AS_STAT); 1208 if (error) 1209 return error; 1210 memset(info, 0, sizeof(*info)); 1211 info->lo_number = lo->lo_number; 1212 info->lo_device = huge_encode_dev(stat.dev); 1213 info->lo_inode = stat.ino; 1214 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); 1215 info->lo_offset = lo->lo_offset; 1216 info->lo_sizelimit = lo->lo_sizelimit; 1217 info->lo_flags = lo->lo_flags; 1218 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1219 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1220 info->lo_encrypt_type = 1221 lo->lo_encryption ? lo->lo_encryption->number : 0; 1222 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1223 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1224 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1225 lo->lo_encrypt_key_size); 1226 } 1227 return 0; 1228 } 1229 1230 static void 1231 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1232 { 1233 memset(info64, 0, sizeof(*info64)); 1234 info64->lo_number = info->lo_number; 1235 info64->lo_device = info->lo_device; 1236 info64->lo_inode = info->lo_inode; 1237 info64->lo_rdevice = info->lo_rdevice; 1238 info64->lo_offset = info->lo_offset; 1239 info64->lo_sizelimit = 0; 1240 info64->lo_encrypt_type = info->lo_encrypt_type; 1241 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1242 info64->lo_flags = info->lo_flags; 1243 info64->lo_init[0] = info->lo_init[0]; 1244 info64->lo_init[1] = info->lo_init[1]; 1245 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1246 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1247 else 1248 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1249 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1250 } 1251 1252 static int 1253 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1254 { 1255 memset(info, 0, sizeof(*info)); 1256 info->lo_number = info64->lo_number; 1257 info->lo_device = info64->lo_device; 1258 info->lo_inode = info64->lo_inode; 1259 info->lo_rdevice = info64->lo_rdevice; 1260 info->lo_offset = info64->lo_offset; 1261 info->lo_encrypt_type = info64->lo_encrypt_type; 1262 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1263 info->lo_flags = info64->lo_flags; 1264 info->lo_init[0] = info64->lo_init[0]; 1265 info->lo_init[1] = info64->lo_init[1]; 1266 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1267 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1268 else 1269 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1270 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1271 1272 /* error in case values were truncated */ 1273 if (info->lo_device != info64->lo_device || 1274 info->lo_rdevice != info64->lo_rdevice || 1275 info->lo_inode != info64->lo_inode || 1276 info->lo_offset != info64->lo_offset) 1277 return -EOVERFLOW; 1278 1279 return 0; 1280 } 1281 1282 static int 1283 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1284 { 1285 struct loop_info info; 1286 struct loop_info64 info64; 1287 1288 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1289 return -EFAULT; 1290 loop_info64_from_old(&info, &info64); 1291 return loop_set_status(lo, &info64); 1292 } 1293 1294 static int 1295 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1296 { 1297 struct loop_info64 info64; 1298 1299 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1300 return -EFAULT; 1301 return loop_set_status(lo, &info64); 1302 } 1303 1304 static int 1305 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1306 struct loop_info info; 1307 struct loop_info64 info64; 1308 int err = 0; 1309 1310 if (!arg) 1311 err = -EINVAL; 1312 if (!err) 1313 err = loop_get_status(lo, &info64); 1314 if (!err) 1315 err = loop_info64_to_old(&info64, &info); 1316 if (!err && copy_to_user(arg, &info, sizeof(info))) 1317 err = -EFAULT; 1318 1319 return err; 1320 } 1321 1322 static int 1323 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1324 struct loop_info64 info64; 1325 int err = 0; 1326 1327 if (!arg) 1328 err = -EINVAL; 1329 if (!err) 1330 err = loop_get_status(lo, &info64); 1331 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1332 err = -EFAULT; 1333 1334 return err; 1335 } 1336 1337 static int loop_set_capacity(struct loop_device *lo) 1338 { 1339 if (unlikely(lo->lo_state != Lo_bound)) 1340 return -ENXIO; 1341 1342 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit, 1343 lo->lo_logical_blocksize); 1344 } 1345 1346 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1347 { 1348 int error = -ENXIO; 1349 if (lo->lo_state != Lo_bound) 1350 goto out; 1351 1352 __loop_update_dio(lo, !!arg); 1353 if (lo->use_dio == !!arg) 1354 return 0; 1355 error = -EINVAL; 1356 out: 1357 return error; 1358 } 1359 1360 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1361 unsigned int cmd, unsigned long arg) 1362 { 1363 struct loop_device *lo = bdev->bd_disk->private_data; 1364 int err; 1365 1366 mutex_lock_nested(&lo->lo_ctl_mutex, 1); 1367 switch (cmd) { 1368 case LOOP_SET_FD: 1369 err = loop_set_fd(lo, mode, bdev, arg); 1370 break; 1371 case LOOP_CHANGE_FD: 1372 err = loop_change_fd(lo, bdev, arg); 1373 break; 1374 case LOOP_CLR_FD: 1375 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1376 err = loop_clr_fd(lo); 1377 if (!err) 1378 goto out_unlocked; 1379 break; 1380 case LOOP_SET_STATUS: 1381 err = -EPERM; 1382 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1383 err = loop_set_status_old(lo, 1384 (struct loop_info __user *)arg); 1385 break; 1386 case LOOP_GET_STATUS: 1387 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1388 break; 1389 case LOOP_SET_STATUS64: 1390 err = -EPERM; 1391 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1392 err = loop_set_status64(lo, 1393 (struct loop_info64 __user *) arg); 1394 break; 1395 case LOOP_GET_STATUS64: 1396 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1397 break; 1398 case LOOP_SET_CAPACITY: 1399 err = -EPERM; 1400 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1401 err = loop_set_capacity(lo); 1402 break; 1403 case LOOP_SET_DIRECT_IO: 1404 err = -EPERM; 1405 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1406 err = loop_set_dio(lo, arg); 1407 break; 1408 default: 1409 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1410 } 1411 mutex_unlock(&lo->lo_ctl_mutex); 1412 1413 out_unlocked: 1414 return err; 1415 } 1416 1417 #ifdef CONFIG_COMPAT 1418 struct compat_loop_info { 1419 compat_int_t lo_number; /* ioctl r/o */ 1420 compat_dev_t lo_device; /* ioctl r/o */ 1421 compat_ulong_t lo_inode; /* ioctl r/o */ 1422 compat_dev_t lo_rdevice; /* ioctl r/o */ 1423 compat_int_t lo_offset; 1424 compat_int_t lo_encrypt_type; 1425 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1426 compat_int_t lo_flags; /* ioctl r/o */ 1427 char lo_name[LO_NAME_SIZE]; 1428 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1429 compat_ulong_t lo_init[2]; 1430 char reserved[4]; 1431 }; 1432 1433 /* 1434 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1435 * - noinlined to reduce stack space usage in main part of driver 1436 */ 1437 static noinline int 1438 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1439 struct loop_info64 *info64) 1440 { 1441 struct compat_loop_info info; 1442 1443 if (copy_from_user(&info, arg, sizeof(info))) 1444 return -EFAULT; 1445 1446 memset(info64, 0, sizeof(*info64)); 1447 info64->lo_number = info.lo_number; 1448 info64->lo_device = info.lo_device; 1449 info64->lo_inode = info.lo_inode; 1450 info64->lo_rdevice = info.lo_rdevice; 1451 info64->lo_offset = info.lo_offset; 1452 info64->lo_sizelimit = 0; 1453 info64->lo_encrypt_type = info.lo_encrypt_type; 1454 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1455 info64->lo_flags = info.lo_flags; 1456 info64->lo_init[0] = info.lo_init[0]; 1457 info64->lo_init[1] = info.lo_init[1]; 1458 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1459 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1460 else 1461 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1462 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1463 return 0; 1464 } 1465 1466 /* 1467 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1468 * - noinlined to reduce stack space usage in main part of driver 1469 */ 1470 static noinline int 1471 loop_info64_to_compat(const struct loop_info64 *info64, 1472 struct compat_loop_info __user *arg) 1473 { 1474 struct compat_loop_info info; 1475 1476 memset(&info, 0, sizeof(info)); 1477 info.lo_number = info64->lo_number; 1478 info.lo_device = info64->lo_device; 1479 info.lo_inode = info64->lo_inode; 1480 info.lo_rdevice = info64->lo_rdevice; 1481 info.lo_offset = info64->lo_offset; 1482 info.lo_encrypt_type = info64->lo_encrypt_type; 1483 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1484 info.lo_flags = info64->lo_flags; 1485 info.lo_init[0] = info64->lo_init[0]; 1486 info.lo_init[1] = info64->lo_init[1]; 1487 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1488 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1489 else 1490 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1491 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1492 1493 /* error in case values were truncated */ 1494 if (info.lo_device != info64->lo_device || 1495 info.lo_rdevice != info64->lo_rdevice || 1496 info.lo_inode != info64->lo_inode || 1497 info.lo_offset != info64->lo_offset || 1498 info.lo_init[0] != info64->lo_init[0] || 1499 info.lo_init[1] != info64->lo_init[1]) 1500 return -EOVERFLOW; 1501 1502 if (copy_to_user(arg, &info, sizeof(info))) 1503 return -EFAULT; 1504 return 0; 1505 } 1506 1507 static int 1508 loop_set_status_compat(struct loop_device *lo, 1509 const struct compat_loop_info __user *arg) 1510 { 1511 struct loop_info64 info64; 1512 int ret; 1513 1514 ret = loop_info64_from_compat(arg, &info64); 1515 if (ret < 0) 1516 return ret; 1517 return loop_set_status(lo, &info64); 1518 } 1519 1520 static int 1521 loop_get_status_compat(struct loop_device *lo, 1522 struct compat_loop_info __user *arg) 1523 { 1524 struct loop_info64 info64; 1525 int err = 0; 1526 1527 if (!arg) 1528 err = -EINVAL; 1529 if (!err) 1530 err = loop_get_status(lo, &info64); 1531 if (!err) 1532 err = loop_info64_to_compat(&info64, arg); 1533 return err; 1534 } 1535 1536 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1537 unsigned int cmd, unsigned long arg) 1538 { 1539 struct loop_device *lo = bdev->bd_disk->private_data; 1540 int err; 1541 1542 switch(cmd) { 1543 case LOOP_SET_STATUS: 1544 mutex_lock(&lo->lo_ctl_mutex); 1545 err = loop_set_status_compat( 1546 lo, (const struct compat_loop_info __user *) arg); 1547 mutex_unlock(&lo->lo_ctl_mutex); 1548 break; 1549 case LOOP_GET_STATUS: 1550 mutex_lock(&lo->lo_ctl_mutex); 1551 err = loop_get_status_compat( 1552 lo, (struct compat_loop_info __user *) arg); 1553 mutex_unlock(&lo->lo_ctl_mutex); 1554 break; 1555 case LOOP_SET_CAPACITY: 1556 case LOOP_CLR_FD: 1557 case LOOP_GET_STATUS64: 1558 case LOOP_SET_STATUS64: 1559 arg = (unsigned long) compat_ptr(arg); 1560 case LOOP_SET_FD: 1561 case LOOP_CHANGE_FD: 1562 err = lo_ioctl(bdev, mode, cmd, arg); 1563 break; 1564 default: 1565 err = -ENOIOCTLCMD; 1566 break; 1567 } 1568 return err; 1569 } 1570 #endif 1571 1572 static int lo_open(struct block_device *bdev, fmode_t mode) 1573 { 1574 struct loop_device *lo; 1575 int err = 0; 1576 1577 mutex_lock(&loop_index_mutex); 1578 lo = bdev->bd_disk->private_data; 1579 if (!lo) { 1580 err = -ENXIO; 1581 goto out; 1582 } 1583 1584 atomic_inc(&lo->lo_refcnt); 1585 out: 1586 mutex_unlock(&loop_index_mutex); 1587 return err; 1588 } 1589 1590 static void lo_release(struct gendisk *disk, fmode_t mode) 1591 { 1592 struct loop_device *lo = disk->private_data; 1593 int err; 1594 1595 if (atomic_dec_return(&lo->lo_refcnt)) 1596 return; 1597 1598 mutex_lock(&lo->lo_ctl_mutex); 1599 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1600 /* 1601 * In autoclear mode, stop the loop thread 1602 * and remove configuration after last close. 1603 */ 1604 err = loop_clr_fd(lo); 1605 if (!err) 1606 return; 1607 } else { 1608 /* 1609 * Otherwise keep thread (if running) and config, 1610 * but flush possible ongoing bios in thread. 1611 */ 1612 loop_flush(lo); 1613 } 1614 1615 mutex_unlock(&lo->lo_ctl_mutex); 1616 } 1617 1618 static const struct block_device_operations lo_fops = { 1619 .owner = THIS_MODULE, 1620 .open = lo_open, 1621 .release = lo_release, 1622 .ioctl = lo_ioctl, 1623 #ifdef CONFIG_COMPAT 1624 .compat_ioctl = lo_compat_ioctl, 1625 #endif 1626 }; 1627 1628 /* 1629 * And now the modules code and kernel interface. 1630 */ 1631 static int max_loop; 1632 module_param(max_loop, int, S_IRUGO); 1633 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1634 module_param(max_part, int, S_IRUGO); 1635 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1636 MODULE_LICENSE("GPL"); 1637 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1638 1639 int loop_register_transfer(struct loop_func_table *funcs) 1640 { 1641 unsigned int n = funcs->number; 1642 1643 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1644 return -EINVAL; 1645 xfer_funcs[n] = funcs; 1646 return 0; 1647 } 1648 1649 static int unregister_transfer_cb(int id, void *ptr, void *data) 1650 { 1651 struct loop_device *lo = ptr; 1652 struct loop_func_table *xfer = data; 1653 1654 mutex_lock(&lo->lo_ctl_mutex); 1655 if (lo->lo_encryption == xfer) 1656 loop_release_xfer(lo); 1657 mutex_unlock(&lo->lo_ctl_mutex); 1658 return 0; 1659 } 1660 1661 int loop_unregister_transfer(int number) 1662 { 1663 unsigned int n = number; 1664 struct loop_func_table *xfer; 1665 1666 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1667 return -EINVAL; 1668 1669 xfer_funcs[n] = NULL; 1670 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1671 return 0; 1672 } 1673 1674 EXPORT_SYMBOL(loop_register_transfer); 1675 EXPORT_SYMBOL(loop_unregister_transfer); 1676 1677 static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1678 const struct blk_mq_queue_data *bd) 1679 { 1680 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); 1681 struct loop_device *lo = cmd->rq->q->queuedata; 1682 1683 blk_mq_start_request(bd->rq); 1684 1685 if (lo->lo_state != Lo_bound) 1686 return BLK_MQ_RQ_QUEUE_ERROR; 1687 1688 switch (req_op(cmd->rq)) { 1689 case REQ_OP_FLUSH: 1690 case REQ_OP_DISCARD: 1691 case REQ_OP_WRITE_ZEROES: 1692 cmd->use_aio = false; 1693 break; 1694 default: 1695 cmd->use_aio = lo->use_dio; 1696 break; 1697 } 1698 1699 kthread_queue_work(&lo->worker, &cmd->work); 1700 1701 return BLK_MQ_RQ_QUEUE_OK; 1702 } 1703 1704 static void loop_handle_cmd(struct loop_cmd *cmd) 1705 { 1706 const bool write = op_is_write(req_op(cmd->rq)); 1707 struct loop_device *lo = cmd->rq->q->queuedata; 1708 int ret = 0; 1709 1710 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1711 ret = -EIO; 1712 goto failed; 1713 } 1714 1715 ret = do_req_filebacked(lo, cmd->rq); 1716 failed: 1717 /* complete non-aio request */ 1718 if (!cmd->use_aio || ret) { 1719 cmd->ret = ret ? -EIO : 0; 1720 blk_mq_complete_request(cmd->rq); 1721 } 1722 } 1723 1724 static void loop_queue_work(struct kthread_work *work) 1725 { 1726 struct loop_cmd *cmd = 1727 container_of(work, struct loop_cmd, work); 1728 1729 loop_handle_cmd(cmd); 1730 } 1731 1732 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1733 unsigned int hctx_idx, unsigned int numa_node) 1734 { 1735 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1736 1737 cmd->rq = rq; 1738 kthread_init_work(&cmd->work, loop_queue_work); 1739 1740 return 0; 1741 } 1742 1743 static const struct blk_mq_ops loop_mq_ops = { 1744 .queue_rq = loop_queue_rq, 1745 .init_request = loop_init_request, 1746 .complete = lo_complete_rq, 1747 }; 1748 1749 static int loop_add(struct loop_device **l, int i) 1750 { 1751 struct loop_device *lo; 1752 struct gendisk *disk; 1753 int err; 1754 1755 err = -ENOMEM; 1756 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1757 if (!lo) 1758 goto out; 1759 1760 lo->lo_state = Lo_unbound; 1761 1762 /* allocate id, if @id >= 0, we're requesting that specific id */ 1763 if (i >= 0) { 1764 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1765 if (err == -ENOSPC) 1766 err = -EEXIST; 1767 } else { 1768 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1769 } 1770 if (err < 0) 1771 goto out_free_dev; 1772 i = err; 1773 1774 err = -ENOMEM; 1775 lo->tag_set.ops = &loop_mq_ops; 1776 lo->tag_set.nr_hw_queues = 1; 1777 lo->tag_set.queue_depth = 128; 1778 lo->tag_set.numa_node = NUMA_NO_NODE; 1779 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1780 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1781 lo->tag_set.driver_data = lo; 1782 1783 err = blk_mq_alloc_tag_set(&lo->tag_set); 1784 if (err) 1785 goto out_free_idr; 1786 1787 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1788 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1789 err = PTR_ERR(lo->lo_queue); 1790 goto out_cleanup_tags; 1791 } 1792 lo->lo_queue->queuedata = lo; 1793 1794 /* 1795 * It doesn't make sense to enable merge because the I/O 1796 * submitted to backing file is handled page by page. 1797 */ 1798 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1799 1800 err = -ENOMEM; 1801 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1802 if (!disk) 1803 goto out_free_queue; 1804 1805 /* 1806 * Disable partition scanning by default. The in-kernel partition 1807 * scanning can be requested individually per-device during its 1808 * setup. Userspace can always add and remove partitions from all 1809 * devices. The needed partition minors are allocated from the 1810 * extended minor space, the main loop device numbers will continue 1811 * to match the loop minors, regardless of the number of partitions 1812 * used. 1813 * 1814 * If max_part is given, partition scanning is globally enabled for 1815 * all loop devices. The minors for the main loop devices will be 1816 * multiples of max_part. 1817 * 1818 * Note: Global-for-all-devices, set-only-at-init, read-only module 1819 * parameteters like 'max_loop' and 'max_part' make things needlessly 1820 * complicated, are too static, inflexible and may surprise 1821 * userspace tools. Parameters like this in general should be avoided. 1822 */ 1823 if (!part_shift) 1824 disk->flags |= GENHD_FL_NO_PART_SCAN; 1825 disk->flags |= GENHD_FL_EXT_DEVT; 1826 mutex_init(&lo->lo_ctl_mutex); 1827 atomic_set(&lo->lo_refcnt, 0); 1828 lo->lo_number = i; 1829 spin_lock_init(&lo->lo_lock); 1830 disk->major = LOOP_MAJOR; 1831 disk->first_minor = i << part_shift; 1832 disk->fops = &lo_fops; 1833 disk->private_data = lo; 1834 disk->queue = lo->lo_queue; 1835 sprintf(disk->disk_name, "loop%d", i); 1836 add_disk(disk); 1837 *l = lo; 1838 return lo->lo_number; 1839 1840 out_free_queue: 1841 blk_cleanup_queue(lo->lo_queue); 1842 out_cleanup_tags: 1843 blk_mq_free_tag_set(&lo->tag_set); 1844 out_free_idr: 1845 idr_remove(&loop_index_idr, i); 1846 out_free_dev: 1847 kfree(lo); 1848 out: 1849 return err; 1850 } 1851 1852 static void loop_remove(struct loop_device *lo) 1853 { 1854 blk_cleanup_queue(lo->lo_queue); 1855 del_gendisk(lo->lo_disk); 1856 blk_mq_free_tag_set(&lo->tag_set); 1857 put_disk(lo->lo_disk); 1858 kfree(lo); 1859 } 1860 1861 static int find_free_cb(int id, void *ptr, void *data) 1862 { 1863 struct loop_device *lo = ptr; 1864 struct loop_device **l = data; 1865 1866 if (lo->lo_state == Lo_unbound) { 1867 *l = lo; 1868 return 1; 1869 } 1870 return 0; 1871 } 1872 1873 static int loop_lookup(struct loop_device **l, int i) 1874 { 1875 struct loop_device *lo; 1876 int ret = -ENODEV; 1877 1878 if (i < 0) { 1879 int err; 1880 1881 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1882 if (err == 1) { 1883 *l = lo; 1884 ret = lo->lo_number; 1885 } 1886 goto out; 1887 } 1888 1889 /* lookup and return a specific i */ 1890 lo = idr_find(&loop_index_idr, i); 1891 if (lo) { 1892 *l = lo; 1893 ret = lo->lo_number; 1894 } 1895 out: 1896 return ret; 1897 } 1898 1899 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1900 { 1901 struct loop_device *lo; 1902 struct kobject *kobj; 1903 int err; 1904 1905 mutex_lock(&loop_index_mutex); 1906 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1907 if (err < 0) 1908 err = loop_add(&lo, MINOR(dev) >> part_shift); 1909 if (err < 0) 1910 kobj = NULL; 1911 else 1912 kobj = get_disk(lo->lo_disk); 1913 mutex_unlock(&loop_index_mutex); 1914 1915 *part = 0; 1916 return kobj; 1917 } 1918 1919 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1920 unsigned long parm) 1921 { 1922 struct loop_device *lo; 1923 int ret = -ENOSYS; 1924 1925 mutex_lock(&loop_index_mutex); 1926 switch (cmd) { 1927 case LOOP_CTL_ADD: 1928 ret = loop_lookup(&lo, parm); 1929 if (ret >= 0) { 1930 ret = -EEXIST; 1931 break; 1932 } 1933 ret = loop_add(&lo, parm); 1934 break; 1935 case LOOP_CTL_REMOVE: 1936 ret = loop_lookup(&lo, parm); 1937 if (ret < 0) 1938 break; 1939 mutex_lock(&lo->lo_ctl_mutex); 1940 if (lo->lo_state != Lo_unbound) { 1941 ret = -EBUSY; 1942 mutex_unlock(&lo->lo_ctl_mutex); 1943 break; 1944 } 1945 if (atomic_read(&lo->lo_refcnt) > 0) { 1946 ret = -EBUSY; 1947 mutex_unlock(&lo->lo_ctl_mutex); 1948 break; 1949 } 1950 lo->lo_disk->private_data = NULL; 1951 mutex_unlock(&lo->lo_ctl_mutex); 1952 idr_remove(&loop_index_idr, lo->lo_number); 1953 loop_remove(lo); 1954 break; 1955 case LOOP_CTL_GET_FREE: 1956 ret = loop_lookup(&lo, -1); 1957 if (ret >= 0) 1958 break; 1959 ret = loop_add(&lo, -1); 1960 } 1961 mutex_unlock(&loop_index_mutex); 1962 1963 return ret; 1964 } 1965 1966 static const struct file_operations loop_ctl_fops = { 1967 .open = nonseekable_open, 1968 .unlocked_ioctl = loop_control_ioctl, 1969 .compat_ioctl = loop_control_ioctl, 1970 .owner = THIS_MODULE, 1971 .llseek = noop_llseek, 1972 }; 1973 1974 static struct miscdevice loop_misc = { 1975 .minor = LOOP_CTRL_MINOR, 1976 .name = "loop-control", 1977 .fops = &loop_ctl_fops, 1978 }; 1979 1980 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 1981 MODULE_ALIAS("devname:loop-control"); 1982 1983 static int __init loop_init(void) 1984 { 1985 int i, nr; 1986 unsigned long range; 1987 struct loop_device *lo; 1988 int err; 1989 1990 err = misc_register(&loop_misc); 1991 if (err < 0) 1992 return err; 1993 1994 part_shift = 0; 1995 if (max_part > 0) { 1996 part_shift = fls(max_part); 1997 1998 /* 1999 * Adjust max_part according to part_shift as it is exported 2000 * to user space so that user can decide correct minor number 2001 * if [s]he want to create more devices. 2002 * 2003 * Note that -1 is required because partition 0 is reserved 2004 * for the whole disk. 2005 */ 2006 max_part = (1UL << part_shift) - 1; 2007 } 2008 2009 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2010 err = -EINVAL; 2011 goto misc_out; 2012 } 2013 2014 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2015 err = -EINVAL; 2016 goto misc_out; 2017 } 2018 2019 /* 2020 * If max_loop is specified, create that many devices upfront. 2021 * This also becomes a hard limit. If max_loop is not specified, 2022 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2023 * init time. Loop devices can be requested on-demand with the 2024 * /dev/loop-control interface, or be instantiated by accessing 2025 * a 'dead' device node. 2026 */ 2027 if (max_loop) { 2028 nr = max_loop; 2029 range = max_loop << part_shift; 2030 } else { 2031 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2032 range = 1UL << MINORBITS; 2033 } 2034 2035 if (register_blkdev(LOOP_MAJOR, "loop")) { 2036 err = -EIO; 2037 goto misc_out; 2038 } 2039 2040 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2041 THIS_MODULE, loop_probe, NULL, NULL); 2042 2043 /* pre-create number of devices given by config or max_loop */ 2044 mutex_lock(&loop_index_mutex); 2045 for (i = 0; i < nr; i++) 2046 loop_add(&lo, i); 2047 mutex_unlock(&loop_index_mutex); 2048 2049 printk(KERN_INFO "loop: module loaded\n"); 2050 return 0; 2051 2052 misc_out: 2053 misc_deregister(&loop_misc); 2054 return err; 2055 } 2056 2057 static int loop_exit_cb(int id, void *ptr, void *data) 2058 { 2059 struct loop_device *lo = ptr; 2060 2061 loop_remove(lo); 2062 return 0; 2063 } 2064 2065 static void __exit loop_exit(void) 2066 { 2067 unsigned long range; 2068 2069 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2070 2071 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2072 idr_destroy(&loop_index_idr); 2073 2074 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2075 unregister_blkdev(LOOP_MAJOR, "loop"); 2076 2077 misc_deregister(&loop_misc); 2078 } 2079 2080 module_init(loop_init); 2081 module_exit(loop_exit); 2082 2083 #ifndef MODULE 2084 static int __init max_loop_setup(char *str) 2085 { 2086 max_loop = simple_strtol(str, NULL, 0); 2087 return 1; 2088 } 2089 2090 __setup("max_loop=", max_loop_setup); 2091 #endif 2092