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 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 218 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 219 } else { 220 blk_queue_flag_set(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 | WRITE, 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_do_completion(struct loop_cmd *cmd) 467 { 468 if (!atomic_dec_and_test(&cmd->ref)) 469 return; 470 kfree(cmd->bvec); 471 cmd->bvec = NULL; 472 blk_mq_complete_request(cmd->rq); 473 } 474 475 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 476 { 477 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 478 479 if (cmd->css) 480 css_put(cmd->css); 481 cmd->ret = ret; 482 lo_rw_aio_do_completion(cmd); 483 } 484 485 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 486 loff_t pos, bool rw) 487 { 488 struct iov_iter iter; 489 struct bio_vec *bvec; 490 struct request *rq = cmd->rq; 491 struct bio *bio = rq->bio; 492 struct file *file = lo->lo_backing_file; 493 unsigned int offset; 494 int segments = 0; 495 int ret; 496 497 if (rq->bio != rq->biotail) { 498 struct req_iterator iter; 499 struct bio_vec tmp; 500 501 __rq_for_each_bio(bio, rq) 502 segments += bio_segments(bio); 503 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO); 504 if (!bvec) 505 return -EIO; 506 cmd->bvec = bvec; 507 508 /* 509 * The bios of the request may be started from the middle of 510 * the 'bvec' because of bio splitting, so we can't directly 511 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment 512 * API will take care of all details for us. 513 */ 514 rq_for_each_segment(tmp, rq, iter) { 515 *bvec = tmp; 516 bvec++; 517 } 518 bvec = cmd->bvec; 519 offset = 0; 520 } else { 521 /* 522 * Same here, this bio may be started from the middle of the 523 * 'bvec' because of bio splitting, so offset from the bvec 524 * must be passed to iov iterator 525 */ 526 offset = bio->bi_iter.bi_bvec_done; 527 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 528 segments = bio_segments(bio); 529 } 530 atomic_set(&cmd->ref, 2); 531 532 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, 533 segments, blk_rq_bytes(rq)); 534 iter.iov_offset = offset; 535 536 cmd->iocb.ki_pos = pos; 537 cmd->iocb.ki_filp = file; 538 cmd->iocb.ki_complete = lo_rw_aio_complete; 539 cmd->iocb.ki_flags = IOCB_DIRECT; 540 if (cmd->css) 541 kthread_associate_blkcg(cmd->css); 542 543 if (rw == WRITE) 544 ret = call_write_iter(file, &cmd->iocb, &iter); 545 else 546 ret = call_read_iter(file, &cmd->iocb, &iter); 547 548 lo_rw_aio_do_completion(cmd); 549 kthread_associate_blkcg(NULL); 550 551 if (ret != -EIOCBQUEUED) 552 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 553 return 0; 554 } 555 556 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 557 { 558 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 559 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 560 561 /* 562 * lo_write_simple and lo_read_simple should have been covered 563 * by io submit style function like lo_rw_aio(), one blocker 564 * is that lo_read_simple() need to call flush_dcache_page after 565 * the page is written from kernel, and it isn't easy to handle 566 * this in io submit style function which submits all segments 567 * of the req at one time. And direct read IO doesn't need to 568 * run flush_dcache_page(). 569 */ 570 switch (req_op(rq)) { 571 case REQ_OP_FLUSH: 572 return lo_req_flush(lo, rq); 573 case REQ_OP_DISCARD: 574 case REQ_OP_WRITE_ZEROES: 575 return lo_discard(lo, rq, pos); 576 case REQ_OP_WRITE: 577 if (lo->transfer) 578 return lo_write_transfer(lo, rq, pos); 579 else if (cmd->use_aio) 580 return lo_rw_aio(lo, cmd, pos, WRITE); 581 else 582 return lo_write_simple(lo, rq, pos); 583 case REQ_OP_READ: 584 if (lo->transfer) 585 return lo_read_transfer(lo, rq, pos); 586 else if (cmd->use_aio) 587 return lo_rw_aio(lo, cmd, pos, READ); 588 else 589 return lo_read_simple(lo, rq, pos); 590 default: 591 WARN_ON_ONCE(1); 592 return -EIO; 593 break; 594 } 595 } 596 597 static inline void loop_update_dio(struct loop_device *lo) 598 { 599 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 600 lo->use_dio); 601 } 602 603 static void loop_reread_partitions(struct loop_device *lo, 604 struct block_device *bdev) 605 { 606 int rc; 607 608 /* 609 * bd_mutex has been held already in release path, so don't 610 * acquire it if this function is called in such case. 611 * 612 * If the reread partition isn't from release path, lo_refcnt 613 * must be at least one and it can only become zero when the 614 * current holder is released. 615 */ 616 if (!atomic_read(&lo->lo_refcnt)) 617 rc = __blkdev_reread_part(bdev); 618 else 619 rc = blkdev_reread_part(bdev); 620 if (rc) 621 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 622 __func__, lo->lo_number, lo->lo_file_name, rc); 623 } 624 625 /* 626 * loop_change_fd switched the backing store of a loopback device to 627 * a new file. This is useful for operating system installers to free up 628 * the original file and in High Availability environments to switch to 629 * an alternative location for the content in case of server meltdown. 630 * This can only work if the loop device is used read-only, and if the 631 * new backing store is the same size and type as the old backing store. 632 */ 633 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 634 unsigned int arg) 635 { 636 struct file *file, *old_file; 637 struct inode *inode; 638 int error; 639 640 error = -ENXIO; 641 if (lo->lo_state != Lo_bound) 642 goto out; 643 644 /* the loop device has to be read-only */ 645 error = -EINVAL; 646 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 647 goto out; 648 649 error = -EBADF; 650 file = fget(arg); 651 if (!file) 652 goto out; 653 654 inode = file->f_mapping->host; 655 old_file = lo->lo_backing_file; 656 657 error = -EINVAL; 658 659 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 660 goto out_putf; 661 662 /* size of the new backing store needs to be the same */ 663 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 664 goto out_putf; 665 666 /* and ... switch */ 667 blk_mq_freeze_queue(lo->lo_queue); 668 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 669 lo->lo_backing_file = file; 670 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 671 mapping_set_gfp_mask(file->f_mapping, 672 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 673 loop_update_dio(lo); 674 blk_mq_unfreeze_queue(lo->lo_queue); 675 676 fput(old_file); 677 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 678 loop_reread_partitions(lo, bdev); 679 return 0; 680 681 out_putf: 682 fput(file); 683 out: 684 return error; 685 } 686 687 static inline int is_loop_device(struct file *file) 688 { 689 struct inode *i = file->f_mapping->host; 690 691 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 692 } 693 694 /* loop sysfs attributes */ 695 696 static ssize_t loop_attr_show(struct device *dev, char *page, 697 ssize_t (*callback)(struct loop_device *, char *)) 698 { 699 struct gendisk *disk = dev_to_disk(dev); 700 struct loop_device *lo = disk->private_data; 701 702 return callback(lo, page); 703 } 704 705 #define LOOP_ATTR_RO(_name) \ 706 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 707 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 708 struct device_attribute *attr, char *b) \ 709 { \ 710 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 711 } \ 712 static struct device_attribute loop_attr_##_name = \ 713 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); 714 715 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 716 { 717 ssize_t ret; 718 char *p = NULL; 719 720 spin_lock_irq(&lo->lo_lock); 721 if (lo->lo_backing_file) 722 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 723 spin_unlock_irq(&lo->lo_lock); 724 725 if (IS_ERR_OR_NULL(p)) 726 ret = PTR_ERR(p); 727 else { 728 ret = strlen(p); 729 memmove(buf, p, ret); 730 buf[ret++] = '\n'; 731 buf[ret] = 0; 732 } 733 734 return ret; 735 } 736 737 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 738 { 739 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 740 } 741 742 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 743 { 744 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 745 } 746 747 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 748 { 749 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 750 751 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 752 } 753 754 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 755 { 756 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 757 758 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 759 } 760 761 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 762 { 763 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 764 765 return sprintf(buf, "%s\n", dio ? "1" : "0"); 766 } 767 768 LOOP_ATTR_RO(backing_file); 769 LOOP_ATTR_RO(offset); 770 LOOP_ATTR_RO(sizelimit); 771 LOOP_ATTR_RO(autoclear); 772 LOOP_ATTR_RO(partscan); 773 LOOP_ATTR_RO(dio); 774 775 static struct attribute *loop_attrs[] = { 776 &loop_attr_backing_file.attr, 777 &loop_attr_offset.attr, 778 &loop_attr_sizelimit.attr, 779 &loop_attr_autoclear.attr, 780 &loop_attr_partscan.attr, 781 &loop_attr_dio.attr, 782 NULL, 783 }; 784 785 static struct attribute_group loop_attribute_group = { 786 .name = "loop", 787 .attrs= loop_attrs, 788 }; 789 790 static int loop_sysfs_init(struct loop_device *lo) 791 { 792 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 793 &loop_attribute_group); 794 } 795 796 static void loop_sysfs_exit(struct loop_device *lo) 797 { 798 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 799 &loop_attribute_group); 800 } 801 802 static void loop_config_discard(struct loop_device *lo) 803 { 804 struct file *file = lo->lo_backing_file; 805 struct inode *inode = file->f_mapping->host; 806 struct request_queue *q = lo->lo_queue; 807 808 /* 809 * We use punch hole to reclaim the free space used by the 810 * image a.k.a. discard. However we do not support discard if 811 * encryption is enabled, because it may give an attacker 812 * useful information. 813 */ 814 if ((!file->f_op->fallocate) || 815 lo->lo_encrypt_key_size) { 816 q->limits.discard_granularity = 0; 817 q->limits.discard_alignment = 0; 818 blk_queue_max_discard_sectors(q, 0); 819 blk_queue_max_write_zeroes_sectors(q, 0); 820 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 821 return; 822 } 823 824 q->limits.discard_granularity = inode->i_sb->s_blocksize; 825 q->limits.discard_alignment = 0; 826 827 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 828 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 829 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 830 } 831 832 static void loop_unprepare_queue(struct loop_device *lo) 833 { 834 kthread_flush_worker(&lo->worker); 835 kthread_stop(lo->worker_task); 836 } 837 838 static int loop_kthread_worker_fn(void *worker_ptr) 839 { 840 current->flags |= PF_LESS_THROTTLE; 841 return kthread_worker_fn(worker_ptr); 842 } 843 844 static int loop_prepare_queue(struct loop_device *lo) 845 { 846 kthread_init_worker(&lo->worker); 847 lo->worker_task = kthread_run(loop_kthread_worker_fn, 848 &lo->worker, "loop%d", lo->lo_number); 849 if (IS_ERR(lo->worker_task)) 850 return -ENOMEM; 851 set_user_nice(lo->worker_task, MIN_NICE); 852 return 0; 853 } 854 855 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 856 struct block_device *bdev, unsigned int arg) 857 { 858 struct file *file, *f; 859 struct inode *inode; 860 struct address_space *mapping; 861 int lo_flags = 0; 862 int error; 863 loff_t size; 864 865 /* This is safe, since we have a reference from open(). */ 866 __module_get(THIS_MODULE); 867 868 error = -EBADF; 869 file = fget(arg); 870 if (!file) 871 goto out; 872 873 error = -EBUSY; 874 if (lo->lo_state != Lo_unbound) 875 goto out_putf; 876 877 /* Avoid recursion */ 878 f = file; 879 while (is_loop_device(f)) { 880 struct loop_device *l; 881 882 if (f->f_mapping->host->i_bdev == bdev) 883 goto out_putf; 884 885 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 886 if (l->lo_state == Lo_unbound) { 887 error = -EINVAL; 888 goto out_putf; 889 } 890 f = l->lo_backing_file; 891 } 892 893 mapping = file->f_mapping; 894 inode = mapping->host; 895 896 error = -EINVAL; 897 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 898 goto out_putf; 899 900 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 901 !file->f_op->write_iter) 902 lo_flags |= LO_FLAGS_READ_ONLY; 903 904 error = -EFBIG; 905 size = get_loop_size(lo, file); 906 if ((loff_t)(sector_t)size != size) 907 goto out_putf; 908 error = loop_prepare_queue(lo); 909 if (error) 910 goto out_putf; 911 912 error = 0; 913 914 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 915 916 lo->use_dio = false; 917 lo->lo_device = bdev; 918 lo->lo_flags = lo_flags; 919 lo->lo_backing_file = file; 920 lo->transfer = NULL; 921 lo->ioctl = NULL; 922 lo->lo_sizelimit = 0; 923 lo->old_gfp_mask = mapping_gfp_mask(mapping); 924 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 925 926 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 927 blk_queue_write_cache(lo->lo_queue, true, false); 928 929 loop_update_dio(lo); 930 set_capacity(lo->lo_disk, size); 931 bd_set_size(bdev, size << 9); 932 loop_sysfs_init(lo); 933 /* let user-space know about the new size */ 934 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 935 936 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 937 block_size(inode->i_bdev) : PAGE_SIZE); 938 939 lo->lo_state = Lo_bound; 940 if (part_shift) 941 lo->lo_flags |= LO_FLAGS_PARTSCAN; 942 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 943 loop_reread_partitions(lo, bdev); 944 945 /* Grab the block_device to prevent its destruction after we 946 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 947 */ 948 bdgrab(bdev); 949 return 0; 950 951 out_putf: 952 fput(file); 953 out: 954 /* This is safe: open() is still holding a reference. */ 955 module_put(THIS_MODULE); 956 return error; 957 } 958 959 static int 960 loop_release_xfer(struct loop_device *lo) 961 { 962 int err = 0; 963 struct loop_func_table *xfer = lo->lo_encryption; 964 965 if (xfer) { 966 if (xfer->release) 967 err = xfer->release(lo); 968 lo->transfer = NULL; 969 lo->lo_encryption = NULL; 970 module_put(xfer->owner); 971 } 972 return err; 973 } 974 975 static int 976 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 977 const struct loop_info64 *i) 978 { 979 int err = 0; 980 981 if (xfer) { 982 struct module *owner = xfer->owner; 983 984 if (!try_module_get(owner)) 985 return -EINVAL; 986 if (xfer->init) 987 err = xfer->init(lo, i); 988 if (err) 989 module_put(owner); 990 else 991 lo->lo_encryption = xfer; 992 } 993 return err; 994 } 995 996 static int loop_clr_fd(struct loop_device *lo) 997 { 998 struct file *filp = lo->lo_backing_file; 999 gfp_t gfp = lo->old_gfp_mask; 1000 struct block_device *bdev = lo->lo_device; 1001 1002 if (lo->lo_state != Lo_bound) 1003 return -ENXIO; 1004 1005 /* 1006 * If we've explicitly asked to tear down the loop device, 1007 * and it has an elevated reference count, set it for auto-teardown when 1008 * the last reference goes away. This stops $!~#$@ udev from 1009 * preventing teardown because it decided that it needs to run blkid on 1010 * the loopback device whenever they appear. xfstests is notorious for 1011 * failing tests because blkid via udev races with a losetup 1012 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1013 * command to fail with EBUSY. 1014 */ 1015 if (atomic_read(&lo->lo_refcnt) > 1) { 1016 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1017 mutex_unlock(&lo->lo_ctl_mutex); 1018 return 0; 1019 } 1020 1021 if (filp == NULL) 1022 return -EINVAL; 1023 1024 /* freeze request queue during the transition */ 1025 blk_mq_freeze_queue(lo->lo_queue); 1026 1027 spin_lock_irq(&lo->lo_lock); 1028 lo->lo_state = Lo_rundown; 1029 lo->lo_backing_file = NULL; 1030 spin_unlock_irq(&lo->lo_lock); 1031 1032 loop_release_xfer(lo); 1033 lo->transfer = NULL; 1034 lo->ioctl = NULL; 1035 lo->lo_device = NULL; 1036 lo->lo_encryption = NULL; 1037 lo->lo_offset = 0; 1038 lo->lo_sizelimit = 0; 1039 lo->lo_encrypt_key_size = 0; 1040 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1041 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1042 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1043 blk_queue_logical_block_size(lo->lo_queue, 512); 1044 blk_queue_physical_block_size(lo->lo_queue, 512); 1045 blk_queue_io_min(lo->lo_queue, 512); 1046 if (bdev) { 1047 bdput(bdev); 1048 invalidate_bdev(bdev); 1049 } 1050 set_capacity(lo->lo_disk, 0); 1051 loop_sysfs_exit(lo); 1052 if (bdev) { 1053 bd_set_size(bdev, 0); 1054 /* let user-space know about this change */ 1055 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1056 } 1057 mapping_set_gfp_mask(filp->f_mapping, gfp); 1058 lo->lo_state = Lo_unbound; 1059 /* This is safe: open() is still holding a reference. */ 1060 module_put(THIS_MODULE); 1061 blk_mq_unfreeze_queue(lo->lo_queue); 1062 1063 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 1064 loop_reread_partitions(lo, bdev); 1065 lo->lo_flags = 0; 1066 if (!part_shift) 1067 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1068 loop_unprepare_queue(lo); 1069 mutex_unlock(&lo->lo_ctl_mutex); 1070 /* 1071 * Need not hold lo_ctl_mutex to fput backing file. 1072 * Calling fput holding lo_ctl_mutex triggers a circular 1073 * lock dependency possibility warning as fput can take 1074 * bd_mutex which is usually taken before lo_ctl_mutex. 1075 */ 1076 fput(filp); 1077 return 0; 1078 } 1079 1080 static int 1081 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1082 { 1083 int err; 1084 struct loop_func_table *xfer; 1085 kuid_t uid = current_uid(); 1086 1087 if (lo->lo_encrypt_key_size && 1088 !uid_eq(lo->lo_key_owner, uid) && 1089 !capable(CAP_SYS_ADMIN)) 1090 return -EPERM; 1091 if (lo->lo_state != Lo_bound) 1092 return -ENXIO; 1093 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1094 return -EINVAL; 1095 1096 /* I/O need to be drained during transfer transition */ 1097 blk_mq_freeze_queue(lo->lo_queue); 1098 1099 err = loop_release_xfer(lo); 1100 if (err) 1101 goto exit; 1102 1103 if (info->lo_encrypt_type) { 1104 unsigned int type = info->lo_encrypt_type; 1105 1106 if (type >= MAX_LO_CRYPT) { 1107 err = -EINVAL; 1108 goto exit; 1109 } 1110 xfer = xfer_funcs[type]; 1111 if (xfer == NULL) { 1112 err = -EINVAL; 1113 goto exit; 1114 } 1115 } else 1116 xfer = NULL; 1117 1118 err = loop_init_xfer(lo, xfer, info); 1119 if (err) 1120 goto exit; 1121 1122 if (lo->lo_offset != info->lo_offset || 1123 lo->lo_sizelimit != info->lo_sizelimit) { 1124 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1125 err = -EFBIG; 1126 goto exit; 1127 } 1128 } 1129 1130 loop_config_discard(lo); 1131 1132 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1133 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1134 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1135 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1136 1137 if (!xfer) 1138 xfer = &none_funcs; 1139 lo->transfer = xfer->transfer; 1140 lo->ioctl = xfer->ioctl; 1141 1142 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1143 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1144 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1145 1146 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1147 lo->lo_init[0] = info->lo_init[0]; 1148 lo->lo_init[1] = info->lo_init[1]; 1149 if (info->lo_encrypt_key_size) { 1150 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1151 info->lo_encrypt_key_size); 1152 lo->lo_key_owner = uid; 1153 } 1154 1155 /* update dio if lo_offset or transfer is changed */ 1156 __loop_update_dio(lo, lo->use_dio); 1157 1158 exit: 1159 blk_mq_unfreeze_queue(lo->lo_queue); 1160 1161 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1162 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1163 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1164 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1165 loop_reread_partitions(lo, lo->lo_device); 1166 } 1167 1168 return err; 1169 } 1170 1171 static int 1172 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1173 { 1174 struct file *file; 1175 struct kstat stat; 1176 int ret; 1177 1178 if (lo->lo_state != Lo_bound) { 1179 mutex_unlock(&lo->lo_ctl_mutex); 1180 return -ENXIO; 1181 } 1182 1183 memset(info, 0, sizeof(*info)); 1184 info->lo_number = lo->lo_number; 1185 info->lo_offset = lo->lo_offset; 1186 info->lo_sizelimit = lo->lo_sizelimit; 1187 info->lo_flags = lo->lo_flags; 1188 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1189 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1190 info->lo_encrypt_type = 1191 lo->lo_encryption ? lo->lo_encryption->number : 0; 1192 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1193 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1194 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1195 lo->lo_encrypt_key_size); 1196 } 1197 1198 /* Drop lo_ctl_mutex while we call into the filesystem. */ 1199 file = get_file(lo->lo_backing_file); 1200 mutex_unlock(&lo->lo_ctl_mutex); 1201 ret = vfs_getattr(&file->f_path, &stat, STATX_INO, 1202 AT_STATX_SYNC_AS_STAT); 1203 if (!ret) { 1204 info->lo_device = huge_encode_dev(stat.dev); 1205 info->lo_inode = stat.ino; 1206 info->lo_rdevice = huge_encode_dev(stat.rdev); 1207 } 1208 fput(file); 1209 return ret; 1210 } 1211 1212 static void 1213 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1214 { 1215 memset(info64, 0, sizeof(*info64)); 1216 info64->lo_number = info->lo_number; 1217 info64->lo_device = info->lo_device; 1218 info64->lo_inode = info->lo_inode; 1219 info64->lo_rdevice = info->lo_rdevice; 1220 info64->lo_offset = info->lo_offset; 1221 info64->lo_sizelimit = 0; 1222 info64->lo_encrypt_type = info->lo_encrypt_type; 1223 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1224 info64->lo_flags = info->lo_flags; 1225 info64->lo_init[0] = info->lo_init[0]; 1226 info64->lo_init[1] = info->lo_init[1]; 1227 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1228 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1229 else 1230 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1231 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1232 } 1233 1234 static int 1235 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1236 { 1237 memset(info, 0, sizeof(*info)); 1238 info->lo_number = info64->lo_number; 1239 info->lo_device = info64->lo_device; 1240 info->lo_inode = info64->lo_inode; 1241 info->lo_rdevice = info64->lo_rdevice; 1242 info->lo_offset = info64->lo_offset; 1243 info->lo_encrypt_type = info64->lo_encrypt_type; 1244 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1245 info->lo_flags = info64->lo_flags; 1246 info->lo_init[0] = info64->lo_init[0]; 1247 info->lo_init[1] = info64->lo_init[1]; 1248 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1249 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1250 else 1251 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1252 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1253 1254 /* error in case values were truncated */ 1255 if (info->lo_device != info64->lo_device || 1256 info->lo_rdevice != info64->lo_rdevice || 1257 info->lo_inode != info64->lo_inode || 1258 info->lo_offset != info64->lo_offset) 1259 return -EOVERFLOW; 1260 1261 return 0; 1262 } 1263 1264 static int 1265 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1266 { 1267 struct loop_info info; 1268 struct loop_info64 info64; 1269 1270 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1271 return -EFAULT; 1272 loop_info64_from_old(&info, &info64); 1273 return loop_set_status(lo, &info64); 1274 } 1275 1276 static int 1277 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1278 { 1279 struct loop_info64 info64; 1280 1281 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1282 return -EFAULT; 1283 return loop_set_status(lo, &info64); 1284 } 1285 1286 static int 1287 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1288 struct loop_info info; 1289 struct loop_info64 info64; 1290 int err = 0; 1291 1292 if (!arg) 1293 err = -EINVAL; 1294 if (!err) 1295 err = loop_get_status(lo, &info64); 1296 if (!err) 1297 err = loop_info64_to_old(&info64, &info); 1298 if (!err && copy_to_user(arg, &info, sizeof(info))) 1299 err = -EFAULT; 1300 1301 return err; 1302 } 1303 1304 static int 1305 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1306 struct loop_info64 info64; 1307 int err = 0; 1308 1309 if (!arg) 1310 err = -EINVAL; 1311 if (!err) 1312 err = loop_get_status(lo, &info64); 1313 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1314 err = -EFAULT; 1315 1316 return err; 1317 } 1318 1319 static int loop_set_capacity(struct loop_device *lo) 1320 { 1321 if (unlikely(lo->lo_state != Lo_bound)) 1322 return -ENXIO; 1323 1324 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1325 } 1326 1327 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1328 { 1329 int error = -ENXIO; 1330 if (lo->lo_state != Lo_bound) 1331 goto out; 1332 1333 __loop_update_dio(lo, !!arg); 1334 if (lo->use_dio == !!arg) 1335 return 0; 1336 error = -EINVAL; 1337 out: 1338 return error; 1339 } 1340 1341 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1342 { 1343 if (lo->lo_state != Lo_bound) 1344 return -ENXIO; 1345 1346 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1347 return -EINVAL; 1348 1349 blk_mq_freeze_queue(lo->lo_queue); 1350 1351 blk_queue_logical_block_size(lo->lo_queue, arg); 1352 blk_queue_physical_block_size(lo->lo_queue, arg); 1353 blk_queue_io_min(lo->lo_queue, arg); 1354 loop_update_dio(lo); 1355 1356 blk_mq_unfreeze_queue(lo->lo_queue); 1357 1358 return 0; 1359 } 1360 1361 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1362 unsigned int cmd, unsigned long arg) 1363 { 1364 struct loop_device *lo = bdev->bd_disk->private_data; 1365 int err; 1366 1367 err = mutex_lock_killable_nested(&lo->lo_ctl_mutex, 1); 1368 if (err) 1369 goto out_unlocked; 1370 1371 switch (cmd) { 1372 case LOOP_SET_FD: 1373 err = loop_set_fd(lo, mode, bdev, arg); 1374 break; 1375 case LOOP_CHANGE_FD: 1376 err = loop_change_fd(lo, bdev, arg); 1377 break; 1378 case LOOP_CLR_FD: 1379 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1380 err = loop_clr_fd(lo); 1381 if (!err) 1382 goto out_unlocked; 1383 break; 1384 case LOOP_SET_STATUS: 1385 err = -EPERM; 1386 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1387 err = loop_set_status_old(lo, 1388 (struct loop_info __user *)arg); 1389 break; 1390 case LOOP_GET_STATUS: 1391 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1392 /* loop_get_status() unlocks lo_ctl_mutex */ 1393 goto out_unlocked; 1394 case LOOP_SET_STATUS64: 1395 err = -EPERM; 1396 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1397 err = loop_set_status64(lo, 1398 (struct loop_info64 __user *) arg); 1399 break; 1400 case LOOP_GET_STATUS64: 1401 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1402 /* loop_get_status() unlocks lo_ctl_mutex */ 1403 goto out_unlocked; 1404 case LOOP_SET_CAPACITY: 1405 err = -EPERM; 1406 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1407 err = loop_set_capacity(lo); 1408 break; 1409 case LOOP_SET_DIRECT_IO: 1410 err = -EPERM; 1411 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1412 err = loop_set_dio(lo, arg); 1413 break; 1414 case LOOP_SET_BLOCK_SIZE: 1415 err = -EPERM; 1416 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1417 err = loop_set_block_size(lo, arg); 1418 break; 1419 default: 1420 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1421 } 1422 mutex_unlock(&lo->lo_ctl_mutex); 1423 1424 out_unlocked: 1425 return err; 1426 } 1427 1428 #ifdef CONFIG_COMPAT 1429 struct compat_loop_info { 1430 compat_int_t lo_number; /* ioctl r/o */ 1431 compat_dev_t lo_device; /* ioctl r/o */ 1432 compat_ulong_t lo_inode; /* ioctl r/o */ 1433 compat_dev_t lo_rdevice; /* ioctl r/o */ 1434 compat_int_t lo_offset; 1435 compat_int_t lo_encrypt_type; 1436 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1437 compat_int_t lo_flags; /* ioctl r/o */ 1438 char lo_name[LO_NAME_SIZE]; 1439 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1440 compat_ulong_t lo_init[2]; 1441 char reserved[4]; 1442 }; 1443 1444 /* 1445 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1446 * - noinlined to reduce stack space usage in main part of driver 1447 */ 1448 static noinline int 1449 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1450 struct loop_info64 *info64) 1451 { 1452 struct compat_loop_info info; 1453 1454 if (copy_from_user(&info, arg, sizeof(info))) 1455 return -EFAULT; 1456 1457 memset(info64, 0, sizeof(*info64)); 1458 info64->lo_number = info.lo_number; 1459 info64->lo_device = info.lo_device; 1460 info64->lo_inode = info.lo_inode; 1461 info64->lo_rdevice = info.lo_rdevice; 1462 info64->lo_offset = info.lo_offset; 1463 info64->lo_sizelimit = 0; 1464 info64->lo_encrypt_type = info.lo_encrypt_type; 1465 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1466 info64->lo_flags = info.lo_flags; 1467 info64->lo_init[0] = info.lo_init[0]; 1468 info64->lo_init[1] = info.lo_init[1]; 1469 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1470 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1471 else 1472 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1473 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1474 return 0; 1475 } 1476 1477 /* 1478 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1479 * - noinlined to reduce stack space usage in main part of driver 1480 */ 1481 static noinline int 1482 loop_info64_to_compat(const struct loop_info64 *info64, 1483 struct compat_loop_info __user *arg) 1484 { 1485 struct compat_loop_info info; 1486 1487 memset(&info, 0, sizeof(info)); 1488 info.lo_number = info64->lo_number; 1489 info.lo_device = info64->lo_device; 1490 info.lo_inode = info64->lo_inode; 1491 info.lo_rdevice = info64->lo_rdevice; 1492 info.lo_offset = info64->lo_offset; 1493 info.lo_encrypt_type = info64->lo_encrypt_type; 1494 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1495 info.lo_flags = info64->lo_flags; 1496 info.lo_init[0] = info64->lo_init[0]; 1497 info.lo_init[1] = info64->lo_init[1]; 1498 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1499 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1500 else 1501 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1502 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1503 1504 /* error in case values were truncated */ 1505 if (info.lo_device != info64->lo_device || 1506 info.lo_rdevice != info64->lo_rdevice || 1507 info.lo_inode != info64->lo_inode || 1508 info.lo_offset != info64->lo_offset || 1509 info.lo_init[0] != info64->lo_init[0] || 1510 info.lo_init[1] != info64->lo_init[1]) 1511 return -EOVERFLOW; 1512 1513 if (copy_to_user(arg, &info, sizeof(info))) 1514 return -EFAULT; 1515 return 0; 1516 } 1517 1518 static int 1519 loop_set_status_compat(struct loop_device *lo, 1520 const struct compat_loop_info __user *arg) 1521 { 1522 struct loop_info64 info64; 1523 int ret; 1524 1525 ret = loop_info64_from_compat(arg, &info64); 1526 if (ret < 0) 1527 return ret; 1528 return loop_set_status(lo, &info64); 1529 } 1530 1531 static int 1532 loop_get_status_compat(struct loop_device *lo, 1533 struct compat_loop_info __user *arg) 1534 { 1535 struct loop_info64 info64; 1536 int err = 0; 1537 1538 if (!arg) 1539 err = -EINVAL; 1540 if (!err) 1541 err = loop_get_status(lo, &info64); 1542 if (!err) 1543 err = loop_info64_to_compat(&info64, arg); 1544 return err; 1545 } 1546 1547 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1548 unsigned int cmd, unsigned long arg) 1549 { 1550 struct loop_device *lo = bdev->bd_disk->private_data; 1551 int err; 1552 1553 switch(cmd) { 1554 case LOOP_SET_STATUS: 1555 err = mutex_lock_killable(&lo->lo_ctl_mutex); 1556 if (!err) { 1557 err = loop_set_status_compat(lo, 1558 (const struct compat_loop_info __user *)arg); 1559 mutex_unlock(&lo->lo_ctl_mutex); 1560 } 1561 break; 1562 case LOOP_GET_STATUS: 1563 err = mutex_lock_killable(&lo->lo_ctl_mutex); 1564 if (!err) { 1565 err = loop_get_status_compat(lo, 1566 (struct compat_loop_info __user *)arg); 1567 /* loop_get_status() unlocks lo_ctl_mutex */ 1568 } 1569 break; 1570 case LOOP_SET_CAPACITY: 1571 case LOOP_CLR_FD: 1572 case LOOP_GET_STATUS64: 1573 case LOOP_SET_STATUS64: 1574 arg = (unsigned long) compat_ptr(arg); 1575 case LOOP_SET_FD: 1576 case LOOP_CHANGE_FD: 1577 err = lo_ioctl(bdev, mode, cmd, arg); 1578 break; 1579 default: 1580 err = -ENOIOCTLCMD; 1581 break; 1582 } 1583 return err; 1584 } 1585 #endif 1586 1587 static int lo_open(struct block_device *bdev, fmode_t mode) 1588 { 1589 struct loop_device *lo; 1590 int err = 0; 1591 1592 mutex_lock(&loop_index_mutex); 1593 lo = bdev->bd_disk->private_data; 1594 if (!lo) { 1595 err = -ENXIO; 1596 goto out; 1597 } 1598 1599 atomic_inc(&lo->lo_refcnt); 1600 out: 1601 mutex_unlock(&loop_index_mutex); 1602 return err; 1603 } 1604 1605 static void __lo_release(struct loop_device *lo) 1606 { 1607 int err; 1608 1609 if (atomic_dec_return(&lo->lo_refcnt)) 1610 return; 1611 1612 mutex_lock(&lo->lo_ctl_mutex); 1613 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1614 /* 1615 * In autoclear mode, stop the loop thread 1616 * and remove configuration after last close. 1617 */ 1618 err = loop_clr_fd(lo); 1619 if (!err) 1620 return; 1621 } else if (lo->lo_state == Lo_bound) { 1622 /* 1623 * Otherwise keep thread (if running) and config, 1624 * but flush possible ongoing bios in thread. 1625 */ 1626 blk_mq_freeze_queue(lo->lo_queue); 1627 blk_mq_unfreeze_queue(lo->lo_queue); 1628 } 1629 1630 mutex_unlock(&lo->lo_ctl_mutex); 1631 } 1632 1633 static void lo_release(struct gendisk *disk, fmode_t mode) 1634 { 1635 mutex_lock(&loop_index_mutex); 1636 __lo_release(disk->private_data); 1637 mutex_unlock(&loop_index_mutex); 1638 } 1639 1640 static const struct block_device_operations lo_fops = { 1641 .owner = THIS_MODULE, 1642 .open = lo_open, 1643 .release = lo_release, 1644 .ioctl = lo_ioctl, 1645 #ifdef CONFIG_COMPAT 1646 .compat_ioctl = lo_compat_ioctl, 1647 #endif 1648 }; 1649 1650 /* 1651 * And now the modules code and kernel interface. 1652 */ 1653 static int max_loop; 1654 module_param(max_loop, int, S_IRUGO); 1655 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1656 module_param(max_part, int, S_IRUGO); 1657 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1658 MODULE_LICENSE("GPL"); 1659 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1660 1661 int loop_register_transfer(struct loop_func_table *funcs) 1662 { 1663 unsigned int n = funcs->number; 1664 1665 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1666 return -EINVAL; 1667 xfer_funcs[n] = funcs; 1668 return 0; 1669 } 1670 1671 static int unregister_transfer_cb(int id, void *ptr, void *data) 1672 { 1673 struct loop_device *lo = ptr; 1674 struct loop_func_table *xfer = data; 1675 1676 mutex_lock(&lo->lo_ctl_mutex); 1677 if (lo->lo_encryption == xfer) 1678 loop_release_xfer(lo); 1679 mutex_unlock(&lo->lo_ctl_mutex); 1680 return 0; 1681 } 1682 1683 int loop_unregister_transfer(int number) 1684 { 1685 unsigned int n = number; 1686 struct loop_func_table *xfer; 1687 1688 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1689 return -EINVAL; 1690 1691 xfer_funcs[n] = NULL; 1692 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1693 return 0; 1694 } 1695 1696 EXPORT_SYMBOL(loop_register_transfer); 1697 EXPORT_SYMBOL(loop_unregister_transfer); 1698 1699 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1700 const struct blk_mq_queue_data *bd) 1701 { 1702 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); 1703 struct loop_device *lo = cmd->rq->q->queuedata; 1704 1705 blk_mq_start_request(bd->rq); 1706 1707 if (lo->lo_state != Lo_bound) 1708 return BLK_STS_IOERR; 1709 1710 switch (req_op(cmd->rq)) { 1711 case REQ_OP_FLUSH: 1712 case REQ_OP_DISCARD: 1713 case REQ_OP_WRITE_ZEROES: 1714 cmd->use_aio = false; 1715 break; 1716 default: 1717 cmd->use_aio = lo->use_dio; 1718 break; 1719 } 1720 1721 /* always use the first bio's css */ 1722 #ifdef CONFIG_BLK_CGROUP 1723 if (cmd->use_aio && cmd->rq->bio && cmd->rq->bio->bi_css) { 1724 cmd->css = cmd->rq->bio->bi_css; 1725 css_get(cmd->css); 1726 } else 1727 #endif 1728 cmd->css = NULL; 1729 kthread_queue_work(&lo->worker, &cmd->work); 1730 1731 return BLK_STS_OK; 1732 } 1733 1734 static void loop_handle_cmd(struct loop_cmd *cmd) 1735 { 1736 const bool write = op_is_write(req_op(cmd->rq)); 1737 struct loop_device *lo = cmd->rq->q->queuedata; 1738 int ret = 0; 1739 1740 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1741 ret = -EIO; 1742 goto failed; 1743 } 1744 1745 ret = do_req_filebacked(lo, cmd->rq); 1746 failed: 1747 /* complete non-aio request */ 1748 if (!cmd->use_aio || ret) { 1749 cmd->ret = ret ? -EIO : 0; 1750 blk_mq_complete_request(cmd->rq); 1751 } 1752 } 1753 1754 static void loop_queue_work(struct kthread_work *work) 1755 { 1756 struct loop_cmd *cmd = 1757 container_of(work, struct loop_cmd, work); 1758 1759 loop_handle_cmd(cmd); 1760 } 1761 1762 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1763 unsigned int hctx_idx, unsigned int numa_node) 1764 { 1765 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1766 1767 cmd->rq = rq; 1768 kthread_init_work(&cmd->work, loop_queue_work); 1769 1770 return 0; 1771 } 1772 1773 static const struct blk_mq_ops loop_mq_ops = { 1774 .queue_rq = loop_queue_rq, 1775 .init_request = loop_init_request, 1776 .complete = lo_complete_rq, 1777 }; 1778 1779 static int loop_add(struct loop_device **l, int i) 1780 { 1781 struct loop_device *lo; 1782 struct gendisk *disk; 1783 int err; 1784 1785 err = -ENOMEM; 1786 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1787 if (!lo) 1788 goto out; 1789 1790 lo->lo_state = Lo_unbound; 1791 1792 /* allocate id, if @id >= 0, we're requesting that specific id */ 1793 if (i >= 0) { 1794 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1795 if (err == -ENOSPC) 1796 err = -EEXIST; 1797 } else { 1798 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1799 } 1800 if (err < 0) 1801 goto out_free_dev; 1802 i = err; 1803 1804 err = -ENOMEM; 1805 lo->tag_set.ops = &loop_mq_ops; 1806 lo->tag_set.nr_hw_queues = 1; 1807 lo->tag_set.queue_depth = 128; 1808 lo->tag_set.numa_node = NUMA_NO_NODE; 1809 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1810 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1811 lo->tag_set.driver_data = lo; 1812 1813 err = blk_mq_alloc_tag_set(&lo->tag_set); 1814 if (err) 1815 goto out_free_idr; 1816 1817 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1818 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1819 err = PTR_ERR(lo->lo_queue); 1820 goto out_cleanup_tags; 1821 } 1822 lo->lo_queue->queuedata = lo; 1823 1824 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1825 1826 /* 1827 * By default, we do buffer IO, so it doesn't make sense to enable 1828 * merge because the I/O submitted to backing file is handled page by 1829 * page. For directio mode, merge does help to dispatch bigger request 1830 * to underlayer disk. We will enable merge once directio is enabled. 1831 */ 1832 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1833 1834 err = -ENOMEM; 1835 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1836 if (!disk) 1837 goto out_free_queue; 1838 1839 /* 1840 * Disable partition scanning by default. The in-kernel partition 1841 * scanning can be requested individually per-device during its 1842 * setup. Userspace can always add and remove partitions from all 1843 * devices. The needed partition minors are allocated from the 1844 * extended minor space, the main loop device numbers will continue 1845 * to match the loop minors, regardless of the number of partitions 1846 * used. 1847 * 1848 * If max_part is given, partition scanning is globally enabled for 1849 * all loop devices. The minors for the main loop devices will be 1850 * multiples of max_part. 1851 * 1852 * Note: Global-for-all-devices, set-only-at-init, read-only module 1853 * parameteters like 'max_loop' and 'max_part' make things needlessly 1854 * complicated, are too static, inflexible and may surprise 1855 * userspace tools. Parameters like this in general should be avoided. 1856 */ 1857 if (!part_shift) 1858 disk->flags |= GENHD_FL_NO_PART_SCAN; 1859 disk->flags |= GENHD_FL_EXT_DEVT; 1860 mutex_init(&lo->lo_ctl_mutex); 1861 atomic_set(&lo->lo_refcnt, 0); 1862 lo->lo_number = i; 1863 spin_lock_init(&lo->lo_lock); 1864 disk->major = LOOP_MAJOR; 1865 disk->first_minor = i << part_shift; 1866 disk->fops = &lo_fops; 1867 disk->private_data = lo; 1868 disk->queue = lo->lo_queue; 1869 sprintf(disk->disk_name, "loop%d", i); 1870 add_disk(disk); 1871 *l = lo; 1872 return lo->lo_number; 1873 1874 out_free_queue: 1875 blk_cleanup_queue(lo->lo_queue); 1876 out_cleanup_tags: 1877 blk_mq_free_tag_set(&lo->tag_set); 1878 out_free_idr: 1879 idr_remove(&loop_index_idr, i); 1880 out_free_dev: 1881 kfree(lo); 1882 out: 1883 return err; 1884 } 1885 1886 static void loop_remove(struct loop_device *lo) 1887 { 1888 del_gendisk(lo->lo_disk); 1889 blk_cleanup_queue(lo->lo_queue); 1890 blk_mq_free_tag_set(&lo->tag_set); 1891 put_disk(lo->lo_disk); 1892 kfree(lo); 1893 } 1894 1895 static int find_free_cb(int id, void *ptr, void *data) 1896 { 1897 struct loop_device *lo = ptr; 1898 struct loop_device **l = data; 1899 1900 if (lo->lo_state == Lo_unbound) { 1901 *l = lo; 1902 return 1; 1903 } 1904 return 0; 1905 } 1906 1907 static int loop_lookup(struct loop_device **l, int i) 1908 { 1909 struct loop_device *lo; 1910 int ret = -ENODEV; 1911 1912 if (i < 0) { 1913 int err; 1914 1915 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1916 if (err == 1) { 1917 *l = lo; 1918 ret = lo->lo_number; 1919 } 1920 goto out; 1921 } 1922 1923 /* lookup and return a specific i */ 1924 lo = idr_find(&loop_index_idr, i); 1925 if (lo) { 1926 *l = lo; 1927 ret = lo->lo_number; 1928 } 1929 out: 1930 return ret; 1931 } 1932 1933 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1934 { 1935 struct loop_device *lo; 1936 struct kobject *kobj; 1937 int err; 1938 1939 mutex_lock(&loop_index_mutex); 1940 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1941 if (err < 0) 1942 err = loop_add(&lo, MINOR(dev) >> part_shift); 1943 if (err < 0) 1944 kobj = NULL; 1945 else 1946 kobj = get_disk_and_module(lo->lo_disk); 1947 mutex_unlock(&loop_index_mutex); 1948 1949 *part = 0; 1950 return kobj; 1951 } 1952 1953 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1954 unsigned long parm) 1955 { 1956 struct loop_device *lo; 1957 int ret = -ENOSYS; 1958 1959 mutex_lock(&loop_index_mutex); 1960 switch (cmd) { 1961 case LOOP_CTL_ADD: 1962 ret = loop_lookup(&lo, parm); 1963 if (ret >= 0) { 1964 ret = -EEXIST; 1965 break; 1966 } 1967 ret = loop_add(&lo, parm); 1968 break; 1969 case LOOP_CTL_REMOVE: 1970 ret = loop_lookup(&lo, parm); 1971 if (ret < 0) 1972 break; 1973 ret = mutex_lock_killable(&lo->lo_ctl_mutex); 1974 if (ret) 1975 break; 1976 if (lo->lo_state != Lo_unbound) { 1977 ret = -EBUSY; 1978 mutex_unlock(&lo->lo_ctl_mutex); 1979 break; 1980 } 1981 if (atomic_read(&lo->lo_refcnt) > 0) { 1982 ret = -EBUSY; 1983 mutex_unlock(&lo->lo_ctl_mutex); 1984 break; 1985 } 1986 lo->lo_disk->private_data = NULL; 1987 mutex_unlock(&lo->lo_ctl_mutex); 1988 idr_remove(&loop_index_idr, lo->lo_number); 1989 loop_remove(lo); 1990 break; 1991 case LOOP_CTL_GET_FREE: 1992 ret = loop_lookup(&lo, -1); 1993 if (ret >= 0) 1994 break; 1995 ret = loop_add(&lo, -1); 1996 } 1997 mutex_unlock(&loop_index_mutex); 1998 1999 return ret; 2000 } 2001 2002 static const struct file_operations loop_ctl_fops = { 2003 .open = nonseekable_open, 2004 .unlocked_ioctl = loop_control_ioctl, 2005 .compat_ioctl = loop_control_ioctl, 2006 .owner = THIS_MODULE, 2007 .llseek = noop_llseek, 2008 }; 2009 2010 static struct miscdevice loop_misc = { 2011 .minor = LOOP_CTRL_MINOR, 2012 .name = "loop-control", 2013 .fops = &loop_ctl_fops, 2014 }; 2015 2016 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2017 MODULE_ALIAS("devname:loop-control"); 2018 2019 static int __init loop_init(void) 2020 { 2021 int i, nr; 2022 unsigned long range; 2023 struct loop_device *lo; 2024 int err; 2025 2026 part_shift = 0; 2027 if (max_part > 0) { 2028 part_shift = fls(max_part); 2029 2030 /* 2031 * Adjust max_part according to part_shift as it is exported 2032 * to user space so that user can decide correct minor number 2033 * if [s]he want to create more devices. 2034 * 2035 * Note that -1 is required because partition 0 is reserved 2036 * for the whole disk. 2037 */ 2038 max_part = (1UL << part_shift) - 1; 2039 } 2040 2041 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2042 err = -EINVAL; 2043 goto err_out; 2044 } 2045 2046 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2047 err = -EINVAL; 2048 goto err_out; 2049 } 2050 2051 /* 2052 * If max_loop is specified, create that many devices upfront. 2053 * This also becomes a hard limit. If max_loop is not specified, 2054 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2055 * init time. Loop devices can be requested on-demand with the 2056 * /dev/loop-control interface, or be instantiated by accessing 2057 * a 'dead' device node. 2058 */ 2059 if (max_loop) { 2060 nr = max_loop; 2061 range = max_loop << part_shift; 2062 } else { 2063 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2064 range = 1UL << MINORBITS; 2065 } 2066 2067 err = misc_register(&loop_misc); 2068 if (err < 0) 2069 goto err_out; 2070 2071 2072 if (register_blkdev(LOOP_MAJOR, "loop")) { 2073 err = -EIO; 2074 goto misc_out; 2075 } 2076 2077 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2078 THIS_MODULE, loop_probe, NULL, NULL); 2079 2080 /* pre-create number of devices given by config or max_loop */ 2081 mutex_lock(&loop_index_mutex); 2082 for (i = 0; i < nr; i++) 2083 loop_add(&lo, i); 2084 mutex_unlock(&loop_index_mutex); 2085 2086 printk(KERN_INFO "loop: module loaded\n"); 2087 return 0; 2088 2089 misc_out: 2090 misc_deregister(&loop_misc); 2091 err_out: 2092 return err; 2093 } 2094 2095 static int loop_exit_cb(int id, void *ptr, void *data) 2096 { 2097 struct loop_device *lo = ptr; 2098 2099 loop_remove(lo); 2100 return 0; 2101 } 2102 2103 static void __exit loop_exit(void) 2104 { 2105 unsigned long range; 2106 2107 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2108 2109 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2110 idr_destroy(&loop_index_idr); 2111 2112 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2113 unregister_blkdev(LOOP_MAJOR, "loop"); 2114 2115 misc_deregister(&loop_misc); 2116 } 2117 2118 module_init(loop_init); 2119 module_exit(loop_exit); 2120 2121 #ifndef MODULE 2122 static int __init max_loop_setup(char *str) 2123 { 2124 max_loop = simple_strtol(str, NULL, 0); 2125 return 1; 2126 } 2127 2128 __setup("max_loop=", max_loop_setup); 2129 #endif 2130