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 queue_flag_clear_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); 218 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 219 } else { 220 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); 221 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 222 } 223 blk_mq_unfreeze_queue(lo->lo_queue); 224 } 225 226 static int 227 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 228 { 229 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 230 sector_t x = (sector_t)size; 231 struct block_device *bdev = lo->lo_device; 232 233 if (unlikely((loff_t)x != size)) 234 return -EFBIG; 235 if (lo->lo_offset != offset) 236 lo->lo_offset = offset; 237 if (lo->lo_sizelimit != sizelimit) 238 lo->lo_sizelimit = sizelimit; 239 set_capacity(lo->lo_disk, x); 240 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 241 /* let user-space know about the new size */ 242 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 243 return 0; 244 } 245 246 static inline int 247 lo_do_transfer(struct loop_device *lo, int cmd, 248 struct page *rpage, unsigned roffs, 249 struct page *lpage, unsigned loffs, 250 int size, sector_t rblock) 251 { 252 int ret; 253 254 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 255 if (likely(!ret)) 256 return 0; 257 258 printk_ratelimited(KERN_ERR 259 "loop: Transfer error at byte offset %llu, length %i.\n", 260 (unsigned long long)rblock << 9, size); 261 return ret; 262 } 263 264 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 265 { 266 struct iov_iter i; 267 ssize_t bw; 268 269 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len); 270 271 file_start_write(file); 272 bw = vfs_iter_write(file, &i, ppos, 0); 273 file_end_write(file); 274 275 if (likely(bw == bvec->bv_len)) 276 return 0; 277 278 printk_ratelimited(KERN_ERR 279 "loop: Write error at byte offset %llu, length %i.\n", 280 (unsigned long long)*ppos, bvec->bv_len); 281 if (bw >= 0) 282 bw = -EIO; 283 return bw; 284 } 285 286 static int lo_write_simple(struct loop_device *lo, struct request *rq, 287 loff_t pos) 288 { 289 struct bio_vec bvec; 290 struct req_iterator iter; 291 int ret = 0; 292 293 rq_for_each_segment(bvec, rq, iter) { 294 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 295 if (ret < 0) 296 break; 297 cond_resched(); 298 } 299 300 return ret; 301 } 302 303 /* 304 * This is the slow, transforming version that needs to double buffer the 305 * data as it cannot do the transformations in place without having direct 306 * access to the destination pages of the backing file. 307 */ 308 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 309 loff_t pos) 310 { 311 struct bio_vec bvec, b; 312 struct req_iterator iter; 313 struct page *page; 314 int ret = 0; 315 316 page = alloc_page(GFP_NOIO); 317 if (unlikely(!page)) 318 return -ENOMEM; 319 320 rq_for_each_segment(bvec, rq, iter) { 321 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 322 bvec.bv_offset, bvec.bv_len, pos >> 9); 323 if (unlikely(ret)) 324 break; 325 326 b.bv_page = page; 327 b.bv_offset = 0; 328 b.bv_len = bvec.bv_len; 329 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 330 if (ret < 0) 331 break; 332 } 333 334 __free_page(page); 335 return ret; 336 } 337 338 static int lo_read_simple(struct loop_device *lo, struct request *rq, 339 loff_t pos) 340 { 341 struct bio_vec bvec; 342 struct req_iterator iter; 343 struct iov_iter i; 344 ssize_t len; 345 346 rq_for_each_segment(bvec, rq, iter) { 347 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); 348 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 349 if (len < 0) 350 return len; 351 352 flush_dcache_page(bvec.bv_page); 353 354 if (len != bvec.bv_len) { 355 struct bio *bio; 356 357 __rq_for_each_bio(bio, rq) 358 zero_fill_bio(bio); 359 break; 360 } 361 cond_resched(); 362 } 363 364 return 0; 365 } 366 367 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 368 loff_t pos) 369 { 370 struct bio_vec bvec, b; 371 struct req_iterator iter; 372 struct iov_iter i; 373 struct page *page; 374 ssize_t len; 375 int ret = 0; 376 377 page = alloc_page(GFP_NOIO); 378 if (unlikely(!page)) 379 return -ENOMEM; 380 381 rq_for_each_segment(bvec, rq, iter) { 382 loff_t offset = pos; 383 384 b.bv_page = page; 385 b.bv_offset = 0; 386 b.bv_len = bvec.bv_len; 387 388 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); 389 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 390 if (len < 0) { 391 ret = len; 392 goto out_free_page; 393 } 394 395 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 396 bvec.bv_offset, len, offset >> 9); 397 if (ret) 398 goto out_free_page; 399 400 flush_dcache_page(bvec.bv_page); 401 402 if (len != bvec.bv_len) { 403 struct bio *bio; 404 405 __rq_for_each_bio(bio, rq) 406 zero_fill_bio(bio); 407 break; 408 } 409 } 410 411 ret = 0; 412 out_free_page: 413 __free_page(page); 414 return ret; 415 } 416 417 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 418 { 419 /* 420 * We use punch hole to reclaim the free space used by the 421 * image a.k.a. discard. However we do not support discard if 422 * encryption is enabled, because it may give an attacker 423 * useful information. 424 */ 425 struct file *file = lo->lo_backing_file; 426 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 427 int ret; 428 429 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 430 ret = -EOPNOTSUPP; 431 goto out; 432 } 433 434 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 435 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 436 ret = -EIO; 437 out: 438 return ret; 439 } 440 441 static int lo_req_flush(struct loop_device *lo, struct request *rq) 442 { 443 struct file *file = lo->lo_backing_file; 444 int ret = vfs_fsync(file, 0); 445 if (unlikely(ret && ret != -EINVAL)) 446 ret = -EIO; 447 448 return ret; 449 } 450 451 static void lo_complete_rq(struct request *rq) 452 { 453 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 454 455 if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio && 456 cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) { 457 struct bio *bio = cmd->rq->bio; 458 459 bio_advance(bio, cmd->ret); 460 zero_fill_bio(bio); 461 } 462 463 blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK); 464 } 465 466 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 467 { 468 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 469 470 kfree(cmd->bvec); 471 cmd->bvec = NULL; 472 cmd->ret = ret; 473 blk_mq_complete_request(cmd->rq); 474 } 475 476 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 477 loff_t pos, bool rw) 478 { 479 struct iov_iter iter; 480 struct bio_vec *bvec; 481 struct request *rq = cmd->rq; 482 struct bio *bio = rq->bio; 483 struct file *file = lo->lo_backing_file; 484 unsigned int offset; 485 int segments = 0; 486 int ret; 487 488 if (rq->bio != rq->biotail) { 489 struct req_iterator iter; 490 struct bio_vec tmp; 491 492 __rq_for_each_bio(bio, rq) 493 segments += bio_segments(bio); 494 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO); 495 if (!bvec) 496 return -EIO; 497 cmd->bvec = bvec; 498 499 /* 500 * The bios of the request may be started from the middle of 501 * the 'bvec' because of bio splitting, so we can't directly 502 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment 503 * API will take care of all details for us. 504 */ 505 rq_for_each_segment(tmp, rq, iter) { 506 *bvec = tmp; 507 bvec++; 508 } 509 bvec = cmd->bvec; 510 offset = 0; 511 } else { 512 /* 513 * Same here, this bio may be started from the middle of the 514 * 'bvec' because of bio splitting, so offset from the bvec 515 * must be passed to iov iterator 516 */ 517 offset = bio->bi_iter.bi_bvec_done; 518 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 519 segments = bio_segments(bio); 520 } 521 522 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, 523 segments, blk_rq_bytes(rq)); 524 iter.iov_offset = offset; 525 526 cmd->iocb.ki_pos = pos; 527 cmd->iocb.ki_filp = file; 528 cmd->iocb.ki_complete = lo_rw_aio_complete; 529 cmd->iocb.ki_flags = IOCB_DIRECT; 530 531 if (rw == WRITE) 532 ret = call_write_iter(file, &cmd->iocb, &iter); 533 else 534 ret = call_read_iter(file, &cmd->iocb, &iter); 535 536 if (ret != -EIOCBQUEUED) 537 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 538 return 0; 539 } 540 541 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 542 { 543 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 544 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 545 546 /* 547 * lo_write_simple and lo_read_simple should have been covered 548 * by io submit style function like lo_rw_aio(), one blocker 549 * is that lo_read_simple() need to call flush_dcache_page after 550 * the page is written from kernel, and it isn't easy to handle 551 * this in io submit style function which submits all segments 552 * of the req at one time. And direct read IO doesn't need to 553 * run flush_dcache_page(). 554 */ 555 switch (req_op(rq)) { 556 case REQ_OP_FLUSH: 557 return lo_req_flush(lo, rq); 558 case REQ_OP_DISCARD: 559 case REQ_OP_WRITE_ZEROES: 560 return lo_discard(lo, rq, pos); 561 case REQ_OP_WRITE: 562 if (lo->transfer) 563 return lo_write_transfer(lo, rq, pos); 564 else if (cmd->use_aio) 565 return lo_rw_aio(lo, cmd, pos, WRITE); 566 else 567 return lo_write_simple(lo, rq, pos); 568 case REQ_OP_READ: 569 if (lo->transfer) 570 return lo_read_transfer(lo, rq, pos); 571 else if (cmd->use_aio) 572 return lo_rw_aio(lo, cmd, pos, READ); 573 else 574 return lo_read_simple(lo, rq, pos); 575 default: 576 WARN_ON_ONCE(1); 577 return -EIO; 578 break; 579 } 580 } 581 582 static inline void loop_update_dio(struct loop_device *lo) 583 { 584 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 585 lo->use_dio); 586 } 587 588 static void loop_reread_partitions(struct loop_device *lo, 589 struct block_device *bdev) 590 { 591 int rc; 592 593 /* 594 * bd_mutex has been held already in release path, so don't 595 * acquire it if this function is called in such case. 596 * 597 * If the reread partition isn't from release path, lo_refcnt 598 * must be at least one and it can only become zero when the 599 * current holder is released. 600 */ 601 if (!atomic_read(&lo->lo_refcnt)) 602 rc = __blkdev_reread_part(bdev); 603 else 604 rc = blkdev_reread_part(bdev); 605 if (rc) 606 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 607 __func__, lo->lo_number, lo->lo_file_name, rc); 608 } 609 610 /* 611 * loop_change_fd switched the backing store of a loopback device to 612 * a new file. This is useful for operating system installers to free up 613 * the original file and in High Availability environments to switch to 614 * an alternative location for the content in case of server meltdown. 615 * This can only work if the loop device is used read-only, and if the 616 * new backing store is the same size and type as the old backing store. 617 */ 618 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 619 unsigned int arg) 620 { 621 struct file *file, *old_file; 622 struct inode *inode; 623 int error; 624 625 error = -ENXIO; 626 if (lo->lo_state != Lo_bound) 627 goto out; 628 629 /* the loop device has to be read-only */ 630 error = -EINVAL; 631 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 632 goto out; 633 634 error = -EBADF; 635 file = fget(arg); 636 if (!file) 637 goto out; 638 639 inode = file->f_mapping->host; 640 old_file = lo->lo_backing_file; 641 642 error = -EINVAL; 643 644 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 645 goto out_putf; 646 647 /* size of the new backing store needs to be the same */ 648 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 649 goto out_putf; 650 651 /* and ... switch */ 652 blk_mq_freeze_queue(lo->lo_queue); 653 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 654 lo->lo_backing_file = file; 655 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 656 mapping_set_gfp_mask(file->f_mapping, 657 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 658 loop_update_dio(lo); 659 blk_mq_unfreeze_queue(lo->lo_queue); 660 661 fput(old_file); 662 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 663 loop_reread_partitions(lo, bdev); 664 return 0; 665 666 out_putf: 667 fput(file); 668 out: 669 return error; 670 } 671 672 static inline int is_loop_device(struct file *file) 673 { 674 struct inode *i = file->f_mapping->host; 675 676 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 677 } 678 679 /* loop sysfs attributes */ 680 681 static ssize_t loop_attr_show(struct device *dev, char *page, 682 ssize_t (*callback)(struct loop_device *, char *)) 683 { 684 struct gendisk *disk = dev_to_disk(dev); 685 struct loop_device *lo = disk->private_data; 686 687 return callback(lo, page); 688 } 689 690 #define LOOP_ATTR_RO(_name) \ 691 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 692 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 693 struct device_attribute *attr, char *b) \ 694 { \ 695 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 696 } \ 697 static struct device_attribute loop_attr_##_name = \ 698 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); 699 700 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 701 { 702 ssize_t ret; 703 char *p = NULL; 704 705 spin_lock_irq(&lo->lo_lock); 706 if (lo->lo_backing_file) 707 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 708 spin_unlock_irq(&lo->lo_lock); 709 710 if (IS_ERR_OR_NULL(p)) 711 ret = PTR_ERR(p); 712 else { 713 ret = strlen(p); 714 memmove(buf, p, ret); 715 buf[ret++] = '\n'; 716 buf[ret] = 0; 717 } 718 719 return ret; 720 } 721 722 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 723 { 724 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 725 } 726 727 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 728 { 729 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 730 } 731 732 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 733 { 734 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 735 736 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 737 } 738 739 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 740 { 741 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 742 743 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 744 } 745 746 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 747 { 748 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 749 750 return sprintf(buf, "%s\n", dio ? "1" : "0"); 751 } 752 753 LOOP_ATTR_RO(backing_file); 754 LOOP_ATTR_RO(offset); 755 LOOP_ATTR_RO(sizelimit); 756 LOOP_ATTR_RO(autoclear); 757 LOOP_ATTR_RO(partscan); 758 LOOP_ATTR_RO(dio); 759 760 static struct attribute *loop_attrs[] = { 761 &loop_attr_backing_file.attr, 762 &loop_attr_offset.attr, 763 &loop_attr_sizelimit.attr, 764 &loop_attr_autoclear.attr, 765 &loop_attr_partscan.attr, 766 &loop_attr_dio.attr, 767 NULL, 768 }; 769 770 static struct attribute_group loop_attribute_group = { 771 .name = "loop", 772 .attrs= loop_attrs, 773 }; 774 775 static int loop_sysfs_init(struct loop_device *lo) 776 { 777 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 778 &loop_attribute_group); 779 } 780 781 static void loop_sysfs_exit(struct loop_device *lo) 782 { 783 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 784 &loop_attribute_group); 785 } 786 787 static void loop_config_discard(struct loop_device *lo) 788 { 789 struct file *file = lo->lo_backing_file; 790 struct inode *inode = file->f_mapping->host; 791 struct request_queue *q = lo->lo_queue; 792 793 /* 794 * We use punch hole to reclaim the free space used by the 795 * image a.k.a. discard. However we do not support discard if 796 * encryption is enabled, because it may give an attacker 797 * useful information. 798 */ 799 if ((!file->f_op->fallocate) || 800 lo->lo_encrypt_key_size) { 801 q->limits.discard_granularity = 0; 802 q->limits.discard_alignment = 0; 803 blk_queue_max_discard_sectors(q, 0); 804 blk_queue_max_write_zeroes_sectors(q, 0); 805 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 806 return; 807 } 808 809 q->limits.discard_granularity = inode->i_sb->s_blocksize; 810 q->limits.discard_alignment = 0; 811 812 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 813 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 814 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 815 } 816 817 static void loop_unprepare_queue(struct loop_device *lo) 818 { 819 kthread_flush_worker(&lo->worker); 820 kthread_stop(lo->worker_task); 821 } 822 823 static int loop_kthread_worker_fn(void *worker_ptr) 824 { 825 current->flags |= PF_LESS_THROTTLE; 826 return kthread_worker_fn(worker_ptr); 827 } 828 829 static int loop_prepare_queue(struct loop_device *lo) 830 { 831 kthread_init_worker(&lo->worker); 832 lo->worker_task = kthread_run(loop_kthread_worker_fn, 833 &lo->worker, "loop%d", lo->lo_number); 834 if (IS_ERR(lo->worker_task)) 835 return -ENOMEM; 836 set_user_nice(lo->worker_task, MIN_NICE); 837 return 0; 838 } 839 840 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 841 struct block_device *bdev, unsigned int arg) 842 { 843 struct file *file, *f; 844 struct inode *inode; 845 struct address_space *mapping; 846 int lo_flags = 0; 847 int error; 848 loff_t size; 849 850 /* This is safe, since we have a reference from open(). */ 851 __module_get(THIS_MODULE); 852 853 error = -EBADF; 854 file = fget(arg); 855 if (!file) 856 goto out; 857 858 error = -EBUSY; 859 if (lo->lo_state != Lo_unbound) 860 goto out_putf; 861 862 /* Avoid recursion */ 863 f = file; 864 while (is_loop_device(f)) { 865 struct loop_device *l; 866 867 if (f->f_mapping->host->i_bdev == bdev) 868 goto out_putf; 869 870 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 871 if (l->lo_state == Lo_unbound) { 872 error = -EINVAL; 873 goto out_putf; 874 } 875 f = l->lo_backing_file; 876 } 877 878 mapping = file->f_mapping; 879 inode = mapping->host; 880 881 error = -EINVAL; 882 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 883 goto out_putf; 884 885 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 886 !file->f_op->write_iter) 887 lo_flags |= LO_FLAGS_READ_ONLY; 888 889 error = -EFBIG; 890 size = get_loop_size(lo, file); 891 if ((loff_t)(sector_t)size != size) 892 goto out_putf; 893 error = loop_prepare_queue(lo); 894 if (error) 895 goto out_putf; 896 897 error = 0; 898 899 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 900 901 lo->use_dio = false; 902 lo->lo_device = bdev; 903 lo->lo_flags = lo_flags; 904 lo->lo_backing_file = file; 905 lo->transfer = NULL; 906 lo->ioctl = NULL; 907 lo->lo_sizelimit = 0; 908 lo->old_gfp_mask = mapping_gfp_mask(mapping); 909 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 910 911 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 912 blk_queue_write_cache(lo->lo_queue, true, false); 913 914 loop_update_dio(lo); 915 set_capacity(lo->lo_disk, size); 916 bd_set_size(bdev, size << 9); 917 loop_sysfs_init(lo); 918 /* let user-space know about the new size */ 919 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 920 921 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 922 block_size(inode->i_bdev) : PAGE_SIZE); 923 924 lo->lo_state = Lo_bound; 925 if (part_shift) 926 lo->lo_flags |= LO_FLAGS_PARTSCAN; 927 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 928 loop_reread_partitions(lo, bdev); 929 930 /* Grab the block_device to prevent its destruction after we 931 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 932 */ 933 bdgrab(bdev); 934 return 0; 935 936 out_putf: 937 fput(file); 938 out: 939 /* This is safe: open() is still holding a reference. */ 940 module_put(THIS_MODULE); 941 return error; 942 } 943 944 static int 945 loop_release_xfer(struct loop_device *lo) 946 { 947 int err = 0; 948 struct loop_func_table *xfer = lo->lo_encryption; 949 950 if (xfer) { 951 if (xfer->release) 952 err = xfer->release(lo); 953 lo->transfer = NULL; 954 lo->lo_encryption = NULL; 955 module_put(xfer->owner); 956 } 957 return err; 958 } 959 960 static int 961 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 962 const struct loop_info64 *i) 963 { 964 int err = 0; 965 966 if (xfer) { 967 struct module *owner = xfer->owner; 968 969 if (!try_module_get(owner)) 970 return -EINVAL; 971 if (xfer->init) 972 err = xfer->init(lo, i); 973 if (err) 974 module_put(owner); 975 else 976 lo->lo_encryption = xfer; 977 } 978 return err; 979 } 980 981 static int loop_clr_fd(struct loop_device *lo) 982 { 983 struct file *filp = lo->lo_backing_file; 984 gfp_t gfp = lo->old_gfp_mask; 985 struct block_device *bdev = lo->lo_device; 986 987 if (lo->lo_state != Lo_bound) 988 return -ENXIO; 989 990 /* 991 * If we've explicitly asked to tear down the loop device, 992 * and it has an elevated reference count, set it for auto-teardown when 993 * the last reference goes away. This stops $!~#$@ udev from 994 * preventing teardown because it decided that it needs to run blkid on 995 * the loopback device whenever they appear. xfstests is notorious for 996 * failing tests because blkid via udev races with a losetup 997 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 998 * command to fail with EBUSY. 999 */ 1000 if (atomic_read(&lo->lo_refcnt) > 1) { 1001 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1002 mutex_unlock(&lo->lo_ctl_mutex); 1003 return 0; 1004 } 1005 1006 if (filp == NULL) 1007 return -EINVAL; 1008 1009 /* freeze request queue during the transition */ 1010 blk_mq_freeze_queue(lo->lo_queue); 1011 1012 spin_lock_irq(&lo->lo_lock); 1013 lo->lo_state = Lo_rundown; 1014 lo->lo_backing_file = NULL; 1015 spin_unlock_irq(&lo->lo_lock); 1016 1017 loop_release_xfer(lo); 1018 lo->transfer = NULL; 1019 lo->ioctl = NULL; 1020 lo->lo_device = NULL; 1021 lo->lo_encryption = NULL; 1022 lo->lo_offset = 0; 1023 lo->lo_sizelimit = 0; 1024 lo->lo_encrypt_key_size = 0; 1025 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1026 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1027 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1028 blk_queue_logical_block_size(lo->lo_queue, 512); 1029 if (bdev) { 1030 bdput(bdev); 1031 invalidate_bdev(bdev); 1032 } 1033 set_capacity(lo->lo_disk, 0); 1034 loop_sysfs_exit(lo); 1035 if (bdev) { 1036 bd_set_size(bdev, 0); 1037 /* let user-space know about this change */ 1038 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1039 } 1040 mapping_set_gfp_mask(filp->f_mapping, gfp); 1041 lo->lo_state = Lo_unbound; 1042 /* This is safe: open() is still holding a reference. */ 1043 module_put(THIS_MODULE); 1044 blk_mq_unfreeze_queue(lo->lo_queue); 1045 1046 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 1047 loop_reread_partitions(lo, bdev); 1048 lo->lo_flags = 0; 1049 if (!part_shift) 1050 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1051 loop_unprepare_queue(lo); 1052 mutex_unlock(&lo->lo_ctl_mutex); 1053 /* 1054 * Need not hold lo_ctl_mutex to fput backing file. 1055 * Calling fput holding lo_ctl_mutex triggers a circular 1056 * lock dependency possibility warning as fput can take 1057 * bd_mutex which is usually taken before lo_ctl_mutex. 1058 */ 1059 fput(filp); 1060 return 0; 1061 } 1062 1063 static int 1064 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1065 { 1066 int err; 1067 struct loop_func_table *xfer; 1068 kuid_t uid = current_uid(); 1069 1070 if (lo->lo_encrypt_key_size && 1071 !uid_eq(lo->lo_key_owner, uid) && 1072 !capable(CAP_SYS_ADMIN)) 1073 return -EPERM; 1074 if (lo->lo_state != Lo_bound) 1075 return -ENXIO; 1076 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1077 return -EINVAL; 1078 1079 /* I/O need to be drained during transfer transition */ 1080 blk_mq_freeze_queue(lo->lo_queue); 1081 1082 err = loop_release_xfer(lo); 1083 if (err) 1084 goto exit; 1085 1086 if (info->lo_encrypt_type) { 1087 unsigned int type = info->lo_encrypt_type; 1088 1089 if (type >= MAX_LO_CRYPT) 1090 return -EINVAL; 1091 xfer = xfer_funcs[type]; 1092 if (xfer == NULL) 1093 return -EINVAL; 1094 } else 1095 xfer = NULL; 1096 1097 err = loop_init_xfer(lo, xfer, info); 1098 if (err) 1099 goto exit; 1100 1101 if (lo->lo_offset != info->lo_offset || 1102 lo->lo_sizelimit != info->lo_sizelimit) { 1103 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1104 err = -EFBIG; 1105 goto exit; 1106 } 1107 } 1108 1109 loop_config_discard(lo); 1110 1111 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1112 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1113 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1114 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1115 1116 if (!xfer) 1117 xfer = &none_funcs; 1118 lo->transfer = xfer->transfer; 1119 lo->ioctl = xfer->ioctl; 1120 1121 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1122 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1123 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1124 1125 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1126 lo->lo_init[0] = info->lo_init[0]; 1127 lo->lo_init[1] = info->lo_init[1]; 1128 if (info->lo_encrypt_key_size) { 1129 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1130 info->lo_encrypt_key_size); 1131 lo->lo_key_owner = uid; 1132 } 1133 1134 /* update dio if lo_offset or transfer is changed */ 1135 __loop_update_dio(lo, lo->use_dio); 1136 1137 exit: 1138 blk_mq_unfreeze_queue(lo->lo_queue); 1139 1140 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1141 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1142 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1143 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1144 loop_reread_partitions(lo, lo->lo_device); 1145 } 1146 1147 return err; 1148 } 1149 1150 static int 1151 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1152 { 1153 struct file *file = lo->lo_backing_file; 1154 struct kstat stat; 1155 int error; 1156 1157 if (lo->lo_state != Lo_bound) 1158 return -ENXIO; 1159 error = vfs_getattr(&file->f_path, &stat, 1160 STATX_INO, AT_STATX_SYNC_AS_STAT); 1161 if (error) 1162 return error; 1163 memset(info, 0, sizeof(*info)); 1164 info->lo_number = lo->lo_number; 1165 info->lo_device = huge_encode_dev(stat.dev); 1166 info->lo_inode = stat.ino; 1167 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); 1168 info->lo_offset = lo->lo_offset; 1169 info->lo_sizelimit = lo->lo_sizelimit; 1170 info->lo_flags = lo->lo_flags; 1171 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1172 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1173 info->lo_encrypt_type = 1174 lo->lo_encryption ? lo->lo_encryption->number : 0; 1175 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1176 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1177 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1178 lo->lo_encrypt_key_size); 1179 } 1180 return 0; 1181 } 1182 1183 static void 1184 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1185 { 1186 memset(info64, 0, sizeof(*info64)); 1187 info64->lo_number = info->lo_number; 1188 info64->lo_device = info->lo_device; 1189 info64->lo_inode = info->lo_inode; 1190 info64->lo_rdevice = info->lo_rdevice; 1191 info64->lo_offset = info->lo_offset; 1192 info64->lo_sizelimit = 0; 1193 info64->lo_encrypt_type = info->lo_encrypt_type; 1194 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1195 info64->lo_flags = info->lo_flags; 1196 info64->lo_init[0] = info->lo_init[0]; 1197 info64->lo_init[1] = info->lo_init[1]; 1198 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1199 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1200 else 1201 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1202 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1203 } 1204 1205 static int 1206 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1207 { 1208 memset(info, 0, sizeof(*info)); 1209 info->lo_number = info64->lo_number; 1210 info->lo_device = info64->lo_device; 1211 info->lo_inode = info64->lo_inode; 1212 info->lo_rdevice = info64->lo_rdevice; 1213 info->lo_offset = info64->lo_offset; 1214 info->lo_encrypt_type = info64->lo_encrypt_type; 1215 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1216 info->lo_flags = info64->lo_flags; 1217 info->lo_init[0] = info64->lo_init[0]; 1218 info->lo_init[1] = info64->lo_init[1]; 1219 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1220 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1221 else 1222 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1223 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1224 1225 /* error in case values were truncated */ 1226 if (info->lo_device != info64->lo_device || 1227 info->lo_rdevice != info64->lo_rdevice || 1228 info->lo_inode != info64->lo_inode || 1229 info->lo_offset != info64->lo_offset) 1230 return -EOVERFLOW; 1231 1232 return 0; 1233 } 1234 1235 static int 1236 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1237 { 1238 struct loop_info info; 1239 struct loop_info64 info64; 1240 1241 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1242 return -EFAULT; 1243 loop_info64_from_old(&info, &info64); 1244 return loop_set_status(lo, &info64); 1245 } 1246 1247 static int 1248 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1249 { 1250 struct loop_info64 info64; 1251 1252 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1253 return -EFAULT; 1254 return loop_set_status(lo, &info64); 1255 } 1256 1257 static int 1258 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1259 struct loop_info info; 1260 struct loop_info64 info64; 1261 int err = 0; 1262 1263 if (!arg) 1264 err = -EINVAL; 1265 if (!err) 1266 err = loop_get_status(lo, &info64); 1267 if (!err) 1268 err = loop_info64_to_old(&info64, &info); 1269 if (!err && copy_to_user(arg, &info, sizeof(info))) 1270 err = -EFAULT; 1271 1272 return err; 1273 } 1274 1275 static int 1276 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1277 struct loop_info64 info64; 1278 int err = 0; 1279 1280 if (!arg) 1281 err = -EINVAL; 1282 if (!err) 1283 err = loop_get_status(lo, &info64); 1284 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1285 err = -EFAULT; 1286 1287 return err; 1288 } 1289 1290 static int loop_set_capacity(struct loop_device *lo) 1291 { 1292 if (unlikely(lo->lo_state != Lo_bound)) 1293 return -ENXIO; 1294 1295 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1296 } 1297 1298 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1299 { 1300 int error = -ENXIO; 1301 if (lo->lo_state != Lo_bound) 1302 goto out; 1303 1304 __loop_update_dio(lo, !!arg); 1305 if (lo->use_dio == !!arg) 1306 return 0; 1307 error = -EINVAL; 1308 out: 1309 return error; 1310 } 1311 1312 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1313 { 1314 if (lo->lo_state != Lo_bound) 1315 return -ENXIO; 1316 1317 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1318 return -EINVAL; 1319 1320 blk_mq_freeze_queue(lo->lo_queue); 1321 1322 blk_queue_logical_block_size(lo->lo_queue, arg); 1323 loop_update_dio(lo); 1324 1325 blk_mq_unfreeze_queue(lo->lo_queue); 1326 1327 return 0; 1328 } 1329 1330 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1331 unsigned int cmd, unsigned long arg) 1332 { 1333 struct loop_device *lo = bdev->bd_disk->private_data; 1334 int err; 1335 1336 mutex_lock_nested(&lo->lo_ctl_mutex, 1); 1337 switch (cmd) { 1338 case LOOP_SET_FD: 1339 err = loop_set_fd(lo, mode, bdev, arg); 1340 break; 1341 case LOOP_CHANGE_FD: 1342 err = loop_change_fd(lo, bdev, arg); 1343 break; 1344 case LOOP_CLR_FD: 1345 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1346 err = loop_clr_fd(lo); 1347 if (!err) 1348 goto out_unlocked; 1349 break; 1350 case LOOP_SET_STATUS: 1351 err = -EPERM; 1352 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1353 err = loop_set_status_old(lo, 1354 (struct loop_info __user *)arg); 1355 break; 1356 case LOOP_GET_STATUS: 1357 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1358 break; 1359 case LOOP_SET_STATUS64: 1360 err = -EPERM; 1361 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1362 err = loop_set_status64(lo, 1363 (struct loop_info64 __user *) arg); 1364 break; 1365 case LOOP_GET_STATUS64: 1366 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1367 break; 1368 case LOOP_SET_CAPACITY: 1369 err = -EPERM; 1370 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1371 err = loop_set_capacity(lo); 1372 break; 1373 case LOOP_SET_DIRECT_IO: 1374 err = -EPERM; 1375 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1376 err = loop_set_dio(lo, arg); 1377 break; 1378 case LOOP_SET_BLOCK_SIZE: 1379 err = -EPERM; 1380 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1381 err = loop_set_block_size(lo, arg); 1382 break; 1383 default: 1384 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1385 } 1386 mutex_unlock(&lo->lo_ctl_mutex); 1387 1388 out_unlocked: 1389 return err; 1390 } 1391 1392 #ifdef CONFIG_COMPAT 1393 struct compat_loop_info { 1394 compat_int_t lo_number; /* ioctl r/o */ 1395 compat_dev_t lo_device; /* ioctl r/o */ 1396 compat_ulong_t lo_inode; /* ioctl r/o */ 1397 compat_dev_t lo_rdevice; /* ioctl r/o */ 1398 compat_int_t lo_offset; 1399 compat_int_t lo_encrypt_type; 1400 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1401 compat_int_t lo_flags; /* ioctl r/o */ 1402 char lo_name[LO_NAME_SIZE]; 1403 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1404 compat_ulong_t lo_init[2]; 1405 char reserved[4]; 1406 }; 1407 1408 /* 1409 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1410 * - noinlined to reduce stack space usage in main part of driver 1411 */ 1412 static noinline int 1413 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1414 struct loop_info64 *info64) 1415 { 1416 struct compat_loop_info info; 1417 1418 if (copy_from_user(&info, arg, sizeof(info))) 1419 return -EFAULT; 1420 1421 memset(info64, 0, sizeof(*info64)); 1422 info64->lo_number = info.lo_number; 1423 info64->lo_device = info.lo_device; 1424 info64->lo_inode = info.lo_inode; 1425 info64->lo_rdevice = info.lo_rdevice; 1426 info64->lo_offset = info.lo_offset; 1427 info64->lo_sizelimit = 0; 1428 info64->lo_encrypt_type = info.lo_encrypt_type; 1429 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1430 info64->lo_flags = info.lo_flags; 1431 info64->lo_init[0] = info.lo_init[0]; 1432 info64->lo_init[1] = info.lo_init[1]; 1433 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1434 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1435 else 1436 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1437 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1438 return 0; 1439 } 1440 1441 /* 1442 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1443 * - noinlined to reduce stack space usage in main part of driver 1444 */ 1445 static noinline int 1446 loop_info64_to_compat(const struct loop_info64 *info64, 1447 struct compat_loop_info __user *arg) 1448 { 1449 struct compat_loop_info info; 1450 1451 memset(&info, 0, sizeof(info)); 1452 info.lo_number = info64->lo_number; 1453 info.lo_device = info64->lo_device; 1454 info.lo_inode = info64->lo_inode; 1455 info.lo_rdevice = info64->lo_rdevice; 1456 info.lo_offset = info64->lo_offset; 1457 info.lo_encrypt_type = info64->lo_encrypt_type; 1458 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1459 info.lo_flags = info64->lo_flags; 1460 info.lo_init[0] = info64->lo_init[0]; 1461 info.lo_init[1] = info64->lo_init[1]; 1462 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1463 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1464 else 1465 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1466 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1467 1468 /* error in case values were truncated */ 1469 if (info.lo_device != info64->lo_device || 1470 info.lo_rdevice != info64->lo_rdevice || 1471 info.lo_inode != info64->lo_inode || 1472 info.lo_offset != info64->lo_offset || 1473 info.lo_init[0] != info64->lo_init[0] || 1474 info.lo_init[1] != info64->lo_init[1]) 1475 return -EOVERFLOW; 1476 1477 if (copy_to_user(arg, &info, sizeof(info))) 1478 return -EFAULT; 1479 return 0; 1480 } 1481 1482 static int 1483 loop_set_status_compat(struct loop_device *lo, 1484 const struct compat_loop_info __user *arg) 1485 { 1486 struct loop_info64 info64; 1487 int ret; 1488 1489 ret = loop_info64_from_compat(arg, &info64); 1490 if (ret < 0) 1491 return ret; 1492 return loop_set_status(lo, &info64); 1493 } 1494 1495 static int 1496 loop_get_status_compat(struct loop_device *lo, 1497 struct compat_loop_info __user *arg) 1498 { 1499 struct loop_info64 info64; 1500 int err = 0; 1501 1502 if (!arg) 1503 err = -EINVAL; 1504 if (!err) 1505 err = loop_get_status(lo, &info64); 1506 if (!err) 1507 err = loop_info64_to_compat(&info64, arg); 1508 return err; 1509 } 1510 1511 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1512 unsigned int cmd, unsigned long arg) 1513 { 1514 struct loop_device *lo = bdev->bd_disk->private_data; 1515 int err; 1516 1517 switch(cmd) { 1518 case LOOP_SET_STATUS: 1519 mutex_lock(&lo->lo_ctl_mutex); 1520 err = loop_set_status_compat( 1521 lo, (const struct compat_loop_info __user *) arg); 1522 mutex_unlock(&lo->lo_ctl_mutex); 1523 break; 1524 case LOOP_GET_STATUS: 1525 mutex_lock(&lo->lo_ctl_mutex); 1526 err = loop_get_status_compat( 1527 lo, (struct compat_loop_info __user *) arg); 1528 mutex_unlock(&lo->lo_ctl_mutex); 1529 break; 1530 case LOOP_SET_CAPACITY: 1531 case LOOP_CLR_FD: 1532 case LOOP_GET_STATUS64: 1533 case LOOP_SET_STATUS64: 1534 arg = (unsigned long) compat_ptr(arg); 1535 case LOOP_SET_FD: 1536 case LOOP_CHANGE_FD: 1537 err = lo_ioctl(bdev, mode, cmd, arg); 1538 break; 1539 default: 1540 err = -ENOIOCTLCMD; 1541 break; 1542 } 1543 return err; 1544 } 1545 #endif 1546 1547 static int lo_open(struct block_device *bdev, fmode_t mode) 1548 { 1549 struct loop_device *lo; 1550 int err = 0; 1551 1552 mutex_lock(&loop_index_mutex); 1553 lo = bdev->bd_disk->private_data; 1554 if (!lo) { 1555 err = -ENXIO; 1556 goto out; 1557 } 1558 1559 atomic_inc(&lo->lo_refcnt); 1560 out: 1561 mutex_unlock(&loop_index_mutex); 1562 return err; 1563 } 1564 1565 static void lo_release(struct gendisk *disk, fmode_t mode) 1566 { 1567 struct loop_device *lo = disk->private_data; 1568 int err; 1569 1570 if (atomic_dec_return(&lo->lo_refcnt)) 1571 return; 1572 1573 mutex_lock(&lo->lo_ctl_mutex); 1574 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1575 /* 1576 * In autoclear mode, stop the loop thread 1577 * and remove configuration after last close. 1578 */ 1579 err = loop_clr_fd(lo); 1580 if (!err) 1581 return; 1582 } else if (lo->lo_state == Lo_bound) { 1583 /* 1584 * Otherwise keep thread (if running) and config, 1585 * but flush possible ongoing bios in thread. 1586 */ 1587 blk_mq_freeze_queue(lo->lo_queue); 1588 blk_mq_unfreeze_queue(lo->lo_queue); 1589 } 1590 1591 mutex_unlock(&lo->lo_ctl_mutex); 1592 } 1593 1594 static const struct block_device_operations lo_fops = { 1595 .owner = THIS_MODULE, 1596 .open = lo_open, 1597 .release = lo_release, 1598 .ioctl = lo_ioctl, 1599 #ifdef CONFIG_COMPAT 1600 .compat_ioctl = lo_compat_ioctl, 1601 #endif 1602 }; 1603 1604 /* 1605 * And now the modules code and kernel interface. 1606 */ 1607 static int max_loop; 1608 module_param(max_loop, int, S_IRUGO); 1609 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1610 module_param(max_part, int, S_IRUGO); 1611 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1612 MODULE_LICENSE("GPL"); 1613 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1614 1615 int loop_register_transfer(struct loop_func_table *funcs) 1616 { 1617 unsigned int n = funcs->number; 1618 1619 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1620 return -EINVAL; 1621 xfer_funcs[n] = funcs; 1622 return 0; 1623 } 1624 1625 static int unregister_transfer_cb(int id, void *ptr, void *data) 1626 { 1627 struct loop_device *lo = ptr; 1628 struct loop_func_table *xfer = data; 1629 1630 mutex_lock(&lo->lo_ctl_mutex); 1631 if (lo->lo_encryption == xfer) 1632 loop_release_xfer(lo); 1633 mutex_unlock(&lo->lo_ctl_mutex); 1634 return 0; 1635 } 1636 1637 int loop_unregister_transfer(int number) 1638 { 1639 unsigned int n = number; 1640 struct loop_func_table *xfer; 1641 1642 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1643 return -EINVAL; 1644 1645 xfer_funcs[n] = NULL; 1646 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1647 return 0; 1648 } 1649 1650 EXPORT_SYMBOL(loop_register_transfer); 1651 EXPORT_SYMBOL(loop_unregister_transfer); 1652 1653 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1654 const struct blk_mq_queue_data *bd) 1655 { 1656 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); 1657 struct loop_device *lo = cmd->rq->q->queuedata; 1658 1659 blk_mq_start_request(bd->rq); 1660 1661 if (lo->lo_state != Lo_bound) 1662 return BLK_STS_IOERR; 1663 1664 switch (req_op(cmd->rq)) { 1665 case REQ_OP_FLUSH: 1666 case REQ_OP_DISCARD: 1667 case REQ_OP_WRITE_ZEROES: 1668 cmd->use_aio = false; 1669 break; 1670 default: 1671 cmd->use_aio = lo->use_dio; 1672 break; 1673 } 1674 1675 kthread_queue_work(&lo->worker, &cmd->work); 1676 1677 return BLK_STS_OK; 1678 } 1679 1680 static void loop_handle_cmd(struct loop_cmd *cmd) 1681 { 1682 const bool write = op_is_write(req_op(cmd->rq)); 1683 struct loop_device *lo = cmd->rq->q->queuedata; 1684 int ret = 0; 1685 1686 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1687 ret = -EIO; 1688 goto failed; 1689 } 1690 1691 ret = do_req_filebacked(lo, cmd->rq); 1692 failed: 1693 /* complete non-aio request */ 1694 if (!cmd->use_aio || ret) { 1695 cmd->ret = ret ? -EIO : 0; 1696 blk_mq_complete_request(cmd->rq); 1697 } 1698 } 1699 1700 static void loop_queue_work(struct kthread_work *work) 1701 { 1702 struct loop_cmd *cmd = 1703 container_of(work, struct loop_cmd, work); 1704 1705 loop_handle_cmd(cmd); 1706 } 1707 1708 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1709 unsigned int hctx_idx, unsigned int numa_node) 1710 { 1711 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1712 1713 cmd->rq = rq; 1714 kthread_init_work(&cmd->work, loop_queue_work); 1715 1716 return 0; 1717 } 1718 1719 static const struct blk_mq_ops loop_mq_ops = { 1720 .queue_rq = loop_queue_rq, 1721 .init_request = loop_init_request, 1722 .complete = lo_complete_rq, 1723 }; 1724 1725 static int loop_add(struct loop_device **l, int i) 1726 { 1727 struct loop_device *lo; 1728 struct gendisk *disk; 1729 int err; 1730 1731 err = -ENOMEM; 1732 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1733 if (!lo) 1734 goto out; 1735 1736 lo->lo_state = Lo_unbound; 1737 1738 /* allocate id, if @id >= 0, we're requesting that specific id */ 1739 if (i >= 0) { 1740 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1741 if (err == -ENOSPC) 1742 err = -EEXIST; 1743 } else { 1744 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1745 } 1746 if (err < 0) 1747 goto out_free_dev; 1748 i = err; 1749 1750 err = -ENOMEM; 1751 lo->tag_set.ops = &loop_mq_ops; 1752 lo->tag_set.nr_hw_queues = 1; 1753 lo->tag_set.queue_depth = 128; 1754 lo->tag_set.numa_node = NUMA_NO_NODE; 1755 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1756 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1757 lo->tag_set.driver_data = lo; 1758 1759 err = blk_mq_alloc_tag_set(&lo->tag_set); 1760 if (err) 1761 goto out_free_idr; 1762 1763 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1764 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1765 err = PTR_ERR(lo->lo_queue); 1766 goto out_cleanup_tags; 1767 } 1768 lo->lo_queue->queuedata = lo; 1769 1770 blk_queue_physical_block_size(lo->lo_queue, PAGE_SIZE); 1771 1772 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1773 1774 /* 1775 * By default, we do buffer IO, so it doesn't make sense to enable 1776 * merge because the I/O submitted to backing file is handled page by 1777 * page. For directio mode, merge does help to dispatch bigger request 1778 * to underlayer disk. We will enable merge once directio is enabled. 1779 */ 1780 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1781 1782 err = -ENOMEM; 1783 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1784 if (!disk) 1785 goto out_free_queue; 1786 1787 /* 1788 * Disable partition scanning by default. The in-kernel partition 1789 * scanning can be requested individually per-device during its 1790 * setup. Userspace can always add and remove partitions from all 1791 * devices. The needed partition minors are allocated from the 1792 * extended minor space, the main loop device numbers will continue 1793 * to match the loop minors, regardless of the number of partitions 1794 * used. 1795 * 1796 * If max_part is given, partition scanning is globally enabled for 1797 * all loop devices. The minors for the main loop devices will be 1798 * multiples of max_part. 1799 * 1800 * Note: Global-for-all-devices, set-only-at-init, read-only module 1801 * parameteters like 'max_loop' and 'max_part' make things needlessly 1802 * complicated, are too static, inflexible and may surprise 1803 * userspace tools. Parameters like this in general should be avoided. 1804 */ 1805 if (!part_shift) 1806 disk->flags |= GENHD_FL_NO_PART_SCAN; 1807 disk->flags |= GENHD_FL_EXT_DEVT; 1808 mutex_init(&lo->lo_ctl_mutex); 1809 atomic_set(&lo->lo_refcnt, 0); 1810 lo->lo_number = i; 1811 spin_lock_init(&lo->lo_lock); 1812 disk->major = LOOP_MAJOR; 1813 disk->first_minor = i << part_shift; 1814 disk->fops = &lo_fops; 1815 disk->private_data = lo; 1816 disk->queue = lo->lo_queue; 1817 sprintf(disk->disk_name, "loop%d", i); 1818 add_disk(disk); 1819 *l = lo; 1820 return lo->lo_number; 1821 1822 out_free_queue: 1823 blk_cleanup_queue(lo->lo_queue); 1824 out_cleanup_tags: 1825 blk_mq_free_tag_set(&lo->tag_set); 1826 out_free_idr: 1827 idr_remove(&loop_index_idr, i); 1828 out_free_dev: 1829 kfree(lo); 1830 out: 1831 return err; 1832 } 1833 1834 static void loop_remove(struct loop_device *lo) 1835 { 1836 blk_cleanup_queue(lo->lo_queue); 1837 del_gendisk(lo->lo_disk); 1838 blk_mq_free_tag_set(&lo->tag_set); 1839 put_disk(lo->lo_disk); 1840 kfree(lo); 1841 } 1842 1843 static int find_free_cb(int id, void *ptr, void *data) 1844 { 1845 struct loop_device *lo = ptr; 1846 struct loop_device **l = data; 1847 1848 if (lo->lo_state == Lo_unbound) { 1849 *l = lo; 1850 return 1; 1851 } 1852 return 0; 1853 } 1854 1855 static int loop_lookup(struct loop_device **l, int i) 1856 { 1857 struct loop_device *lo; 1858 int ret = -ENODEV; 1859 1860 if (i < 0) { 1861 int err; 1862 1863 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1864 if (err == 1) { 1865 *l = lo; 1866 ret = lo->lo_number; 1867 } 1868 goto out; 1869 } 1870 1871 /* lookup and return a specific i */ 1872 lo = idr_find(&loop_index_idr, i); 1873 if (lo) { 1874 *l = lo; 1875 ret = lo->lo_number; 1876 } 1877 out: 1878 return ret; 1879 } 1880 1881 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1882 { 1883 struct loop_device *lo; 1884 struct kobject *kobj; 1885 int err; 1886 1887 mutex_lock(&loop_index_mutex); 1888 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1889 if (err < 0) 1890 err = loop_add(&lo, MINOR(dev) >> part_shift); 1891 if (err < 0) 1892 kobj = NULL; 1893 else 1894 kobj = get_disk(lo->lo_disk); 1895 mutex_unlock(&loop_index_mutex); 1896 1897 *part = 0; 1898 return kobj; 1899 } 1900 1901 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1902 unsigned long parm) 1903 { 1904 struct loop_device *lo; 1905 int ret = -ENOSYS; 1906 1907 mutex_lock(&loop_index_mutex); 1908 switch (cmd) { 1909 case LOOP_CTL_ADD: 1910 ret = loop_lookup(&lo, parm); 1911 if (ret >= 0) { 1912 ret = -EEXIST; 1913 break; 1914 } 1915 ret = loop_add(&lo, parm); 1916 break; 1917 case LOOP_CTL_REMOVE: 1918 ret = loop_lookup(&lo, parm); 1919 if (ret < 0) 1920 break; 1921 mutex_lock(&lo->lo_ctl_mutex); 1922 if (lo->lo_state != Lo_unbound) { 1923 ret = -EBUSY; 1924 mutex_unlock(&lo->lo_ctl_mutex); 1925 break; 1926 } 1927 if (atomic_read(&lo->lo_refcnt) > 0) { 1928 ret = -EBUSY; 1929 mutex_unlock(&lo->lo_ctl_mutex); 1930 break; 1931 } 1932 lo->lo_disk->private_data = NULL; 1933 mutex_unlock(&lo->lo_ctl_mutex); 1934 idr_remove(&loop_index_idr, lo->lo_number); 1935 loop_remove(lo); 1936 break; 1937 case LOOP_CTL_GET_FREE: 1938 ret = loop_lookup(&lo, -1); 1939 if (ret >= 0) 1940 break; 1941 ret = loop_add(&lo, -1); 1942 } 1943 mutex_unlock(&loop_index_mutex); 1944 1945 return ret; 1946 } 1947 1948 static const struct file_operations loop_ctl_fops = { 1949 .open = nonseekable_open, 1950 .unlocked_ioctl = loop_control_ioctl, 1951 .compat_ioctl = loop_control_ioctl, 1952 .owner = THIS_MODULE, 1953 .llseek = noop_llseek, 1954 }; 1955 1956 static struct miscdevice loop_misc = { 1957 .minor = LOOP_CTRL_MINOR, 1958 .name = "loop-control", 1959 .fops = &loop_ctl_fops, 1960 }; 1961 1962 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 1963 MODULE_ALIAS("devname:loop-control"); 1964 1965 static int __init loop_init(void) 1966 { 1967 int i, nr; 1968 unsigned long range; 1969 struct loop_device *lo; 1970 int err; 1971 1972 part_shift = 0; 1973 if (max_part > 0) { 1974 part_shift = fls(max_part); 1975 1976 /* 1977 * Adjust max_part according to part_shift as it is exported 1978 * to user space so that user can decide correct minor number 1979 * if [s]he want to create more devices. 1980 * 1981 * Note that -1 is required because partition 0 is reserved 1982 * for the whole disk. 1983 */ 1984 max_part = (1UL << part_shift) - 1; 1985 } 1986 1987 if ((1UL << part_shift) > DISK_MAX_PARTS) { 1988 err = -EINVAL; 1989 goto err_out; 1990 } 1991 1992 if (max_loop > 1UL << (MINORBITS - part_shift)) { 1993 err = -EINVAL; 1994 goto err_out; 1995 } 1996 1997 /* 1998 * If max_loop is specified, create that many devices upfront. 1999 * This also becomes a hard limit. If max_loop is not specified, 2000 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2001 * init time. Loop devices can be requested on-demand with the 2002 * /dev/loop-control interface, or be instantiated by accessing 2003 * a 'dead' device node. 2004 */ 2005 if (max_loop) { 2006 nr = max_loop; 2007 range = max_loop << part_shift; 2008 } else { 2009 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2010 range = 1UL << MINORBITS; 2011 } 2012 2013 err = misc_register(&loop_misc); 2014 if (err < 0) 2015 goto err_out; 2016 2017 2018 if (register_blkdev(LOOP_MAJOR, "loop")) { 2019 err = -EIO; 2020 goto misc_out; 2021 } 2022 2023 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2024 THIS_MODULE, loop_probe, NULL, NULL); 2025 2026 /* pre-create number of devices given by config or max_loop */ 2027 mutex_lock(&loop_index_mutex); 2028 for (i = 0; i < nr; i++) 2029 loop_add(&lo, i); 2030 mutex_unlock(&loop_index_mutex); 2031 2032 printk(KERN_INFO "loop: module loaded\n"); 2033 return 0; 2034 2035 misc_out: 2036 misc_deregister(&loop_misc); 2037 err_out: 2038 return err; 2039 } 2040 2041 static int loop_exit_cb(int id, void *ptr, void *data) 2042 { 2043 struct loop_device *lo = ptr; 2044 2045 loop_remove(lo); 2046 return 0; 2047 } 2048 2049 static void __exit loop_exit(void) 2050 { 2051 unsigned long range; 2052 2053 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2054 2055 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2056 idr_destroy(&loop_index_idr); 2057 2058 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2059 unregister_blkdev(LOOP_MAJOR, "loop"); 2060 2061 misc_deregister(&loop_misc); 2062 } 2063 2064 module_init(loop_init); 2065 module_exit(loop_exit); 2066 2067 #ifndef MODULE 2068 static int __init max_loop_setup(char *str) 2069 { 2070 max_loop = simple_strtol(str, NULL, 0); 2071 return 1; 2072 } 2073 2074 __setup("max_loop=", max_loop_setup); 2075 #endif 2076