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 <linux/ioprio.h> 80 #include <linux/blk-cgroup.h> 81 82 #include "loop.h" 83 84 #include <linux/uaccess.h> 85 86 static DEFINE_IDR(loop_index_idr); 87 static DEFINE_MUTEX(loop_index_mutex); 88 89 static int max_part; 90 static int part_shift; 91 92 static int transfer_xor(struct loop_device *lo, int cmd, 93 struct page *raw_page, unsigned raw_off, 94 struct page *loop_page, unsigned loop_off, 95 int size, sector_t real_block) 96 { 97 char *raw_buf = kmap_atomic(raw_page) + raw_off; 98 char *loop_buf = kmap_atomic(loop_page) + loop_off; 99 char *in, *out, *key; 100 int i, keysize; 101 102 if (cmd == READ) { 103 in = raw_buf; 104 out = loop_buf; 105 } else { 106 in = loop_buf; 107 out = raw_buf; 108 } 109 110 key = lo->lo_encrypt_key; 111 keysize = lo->lo_encrypt_key_size; 112 for (i = 0; i < size; i++) 113 *out++ = *in++ ^ key[(i & 511) % keysize]; 114 115 kunmap_atomic(loop_buf); 116 kunmap_atomic(raw_buf); 117 cond_resched(); 118 return 0; 119 } 120 121 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 122 { 123 if (unlikely(info->lo_encrypt_key_size <= 0)) 124 return -EINVAL; 125 return 0; 126 } 127 128 static struct loop_func_table none_funcs = { 129 .number = LO_CRYPT_NONE, 130 }; 131 132 static struct loop_func_table xor_funcs = { 133 .number = LO_CRYPT_XOR, 134 .transfer = transfer_xor, 135 .init = xor_init 136 }; 137 138 /* xfer_funcs[0] is special - its release function is never called */ 139 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 140 &none_funcs, 141 &xor_funcs 142 }; 143 144 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 145 { 146 loff_t loopsize; 147 148 /* Compute loopsize in bytes */ 149 loopsize = i_size_read(file->f_mapping->host); 150 if (offset > 0) 151 loopsize -= offset; 152 /* offset is beyond i_size, weird but possible */ 153 if (loopsize < 0) 154 return 0; 155 156 if (sizelimit > 0 && sizelimit < loopsize) 157 loopsize = sizelimit; 158 /* 159 * Unfortunately, if we want to do I/O on the device, 160 * the number of 512-byte sectors has to fit into a sector_t. 161 */ 162 return loopsize >> 9; 163 } 164 165 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 166 { 167 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 168 } 169 170 static void __loop_update_dio(struct loop_device *lo, bool dio) 171 { 172 struct file *file = lo->lo_backing_file; 173 struct address_space *mapping = file->f_mapping; 174 struct inode *inode = mapping->host; 175 unsigned short sb_bsize = 0; 176 unsigned dio_align = 0; 177 bool use_dio; 178 179 if (inode->i_sb->s_bdev) { 180 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 181 dio_align = sb_bsize - 1; 182 } 183 184 /* 185 * We support direct I/O only if lo_offset is aligned with the 186 * logical I/O size of backing device, and the logical block 187 * size of loop is bigger than the backing device's and the loop 188 * needn't transform transfer. 189 * 190 * TODO: the above condition may be loosed in the future, and 191 * direct I/O may be switched runtime at that time because most 192 * of requests in sane applications should be PAGE_SIZE aligned 193 */ 194 if (dio) { 195 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 196 !(lo->lo_offset & dio_align) && 197 mapping->a_ops->direct_IO && 198 !lo->transfer) 199 use_dio = true; 200 else 201 use_dio = false; 202 } else { 203 use_dio = false; 204 } 205 206 if (lo->use_dio == use_dio) 207 return; 208 209 /* flush dirty pages before changing direct IO */ 210 vfs_fsync(file, 0); 211 212 /* 213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 215 * will get updated by ioctl(LOOP_GET_STATUS) 216 */ 217 blk_mq_freeze_queue(lo->lo_queue); 218 lo->use_dio = use_dio; 219 if (use_dio) { 220 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 221 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 222 } else { 223 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 224 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 225 } 226 blk_mq_unfreeze_queue(lo->lo_queue); 227 } 228 229 static int 230 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 231 { 232 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 233 sector_t x = (sector_t)size; 234 struct block_device *bdev = lo->lo_device; 235 236 if (unlikely((loff_t)x != size)) 237 return -EFBIG; 238 if (lo->lo_offset != offset) 239 lo->lo_offset = offset; 240 if (lo->lo_sizelimit != sizelimit) 241 lo->lo_sizelimit = sizelimit; 242 set_capacity(lo->lo_disk, x); 243 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 244 /* let user-space know about the new size */ 245 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 246 return 0; 247 } 248 249 static inline int 250 lo_do_transfer(struct loop_device *lo, int cmd, 251 struct page *rpage, unsigned roffs, 252 struct page *lpage, unsigned loffs, 253 int size, sector_t rblock) 254 { 255 int ret; 256 257 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 258 if (likely(!ret)) 259 return 0; 260 261 printk_ratelimited(KERN_ERR 262 "loop: Transfer error at byte offset %llu, length %i.\n", 263 (unsigned long long)rblock << 9, size); 264 return ret; 265 } 266 267 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 268 { 269 struct iov_iter i; 270 ssize_t bw; 271 272 iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len); 273 274 file_start_write(file); 275 bw = vfs_iter_write(file, &i, ppos, 0); 276 file_end_write(file); 277 278 if (likely(bw == bvec->bv_len)) 279 return 0; 280 281 printk_ratelimited(KERN_ERR 282 "loop: Write error at byte offset %llu, length %i.\n", 283 (unsigned long long)*ppos, bvec->bv_len); 284 if (bw >= 0) 285 bw = -EIO; 286 return bw; 287 } 288 289 static int lo_write_simple(struct loop_device *lo, struct request *rq, 290 loff_t pos) 291 { 292 struct bio_vec bvec; 293 struct req_iterator iter; 294 int ret = 0; 295 296 rq_for_each_segment(bvec, rq, iter) { 297 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 298 if (ret < 0) 299 break; 300 cond_resched(); 301 } 302 303 return ret; 304 } 305 306 /* 307 * This is the slow, transforming version that needs to double buffer the 308 * data as it cannot do the transformations in place without having direct 309 * access to the destination pages of the backing file. 310 */ 311 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 312 loff_t pos) 313 { 314 struct bio_vec bvec, b; 315 struct req_iterator iter; 316 struct page *page; 317 int ret = 0; 318 319 page = alloc_page(GFP_NOIO); 320 if (unlikely(!page)) 321 return -ENOMEM; 322 323 rq_for_each_segment(bvec, rq, iter) { 324 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 325 bvec.bv_offset, bvec.bv_len, pos >> 9); 326 if (unlikely(ret)) 327 break; 328 329 b.bv_page = page; 330 b.bv_offset = 0; 331 b.bv_len = bvec.bv_len; 332 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 333 if (ret < 0) 334 break; 335 } 336 337 __free_page(page); 338 return ret; 339 } 340 341 static int lo_read_simple(struct loop_device *lo, struct request *rq, 342 loff_t pos) 343 { 344 struct bio_vec bvec; 345 struct req_iterator iter; 346 struct iov_iter i; 347 ssize_t len; 348 349 rq_for_each_segment(bvec, rq, iter) { 350 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); 351 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 352 if (len < 0) 353 return len; 354 355 flush_dcache_page(bvec.bv_page); 356 357 if (len != bvec.bv_len) { 358 struct bio *bio; 359 360 __rq_for_each_bio(bio, rq) 361 zero_fill_bio(bio); 362 break; 363 } 364 cond_resched(); 365 } 366 367 return 0; 368 } 369 370 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 371 loff_t pos) 372 { 373 struct bio_vec bvec, b; 374 struct req_iterator iter; 375 struct iov_iter i; 376 struct page *page; 377 ssize_t len; 378 int ret = 0; 379 380 page = alloc_page(GFP_NOIO); 381 if (unlikely(!page)) 382 return -ENOMEM; 383 384 rq_for_each_segment(bvec, rq, iter) { 385 loff_t offset = pos; 386 387 b.bv_page = page; 388 b.bv_offset = 0; 389 b.bv_len = bvec.bv_len; 390 391 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); 392 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 393 if (len < 0) { 394 ret = len; 395 goto out_free_page; 396 } 397 398 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 399 bvec.bv_offset, len, offset >> 9); 400 if (ret) 401 goto out_free_page; 402 403 flush_dcache_page(bvec.bv_page); 404 405 if (len != bvec.bv_len) { 406 struct bio *bio; 407 408 __rq_for_each_bio(bio, rq) 409 zero_fill_bio(bio); 410 break; 411 } 412 } 413 414 ret = 0; 415 out_free_page: 416 __free_page(page); 417 return ret; 418 } 419 420 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 421 { 422 /* 423 * We use punch hole to reclaim the free space used by the 424 * image a.k.a. discard. However we do not support discard if 425 * encryption is enabled, because it may give an attacker 426 * useful information. 427 */ 428 struct file *file = lo->lo_backing_file; 429 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 430 int ret; 431 432 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 433 ret = -EOPNOTSUPP; 434 goto out; 435 } 436 437 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 438 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 439 ret = -EIO; 440 out: 441 return ret; 442 } 443 444 static int lo_req_flush(struct loop_device *lo, struct request *rq) 445 { 446 struct file *file = lo->lo_backing_file; 447 int ret = vfs_fsync(file, 0); 448 if (unlikely(ret && ret != -EINVAL)) 449 ret = -EIO; 450 451 return ret; 452 } 453 454 static void lo_complete_rq(struct request *rq) 455 { 456 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 457 blk_status_t ret = BLK_STS_OK; 458 459 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 460 req_op(rq) != REQ_OP_READ) { 461 if (cmd->ret < 0) 462 ret = BLK_STS_IOERR; 463 goto end_io; 464 } 465 466 /* 467 * Short READ - if we got some data, advance our request and 468 * retry it. If we got no data, end the rest with EIO. 469 */ 470 if (cmd->ret) { 471 blk_update_request(rq, BLK_STS_OK, cmd->ret); 472 cmd->ret = 0; 473 blk_mq_requeue_request(rq, true); 474 } else { 475 if (cmd->use_aio) { 476 struct bio *bio = rq->bio; 477 478 while (bio) { 479 zero_fill_bio(bio); 480 bio = bio->bi_next; 481 } 482 } 483 ret = BLK_STS_IOERR; 484 end_io: 485 blk_mq_end_request(rq, ret); 486 } 487 } 488 489 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 490 { 491 struct request *rq = blk_mq_rq_from_pdu(cmd); 492 493 if (!atomic_dec_and_test(&cmd->ref)) 494 return; 495 kfree(cmd->bvec); 496 cmd->bvec = NULL; 497 blk_mq_complete_request(rq); 498 } 499 500 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 501 { 502 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 503 504 if (cmd->css) 505 css_put(cmd->css); 506 cmd->ret = ret; 507 lo_rw_aio_do_completion(cmd); 508 } 509 510 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 511 loff_t pos, bool rw) 512 { 513 struct iov_iter iter; 514 struct bio_vec *bvec; 515 struct request *rq = blk_mq_rq_from_pdu(cmd); 516 struct bio *bio = rq->bio; 517 struct file *file = lo->lo_backing_file; 518 unsigned int offset; 519 int segments = 0; 520 int ret; 521 522 if (rq->bio != rq->biotail) { 523 struct req_iterator iter; 524 struct bio_vec tmp; 525 526 __rq_for_each_bio(bio, rq) 527 segments += bio_segments(bio); 528 bvec = kmalloc_array(segments, sizeof(struct bio_vec), 529 GFP_NOIO); 530 if (!bvec) 531 return -EIO; 532 cmd->bvec = bvec; 533 534 /* 535 * The bios of the request may be started from the middle of 536 * the 'bvec' because of bio splitting, so we can't directly 537 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment 538 * API will take care of all details for us. 539 */ 540 rq_for_each_segment(tmp, rq, iter) { 541 *bvec = tmp; 542 bvec++; 543 } 544 bvec = cmd->bvec; 545 offset = 0; 546 } else { 547 /* 548 * Same here, this bio may be started from the middle of the 549 * 'bvec' because of bio splitting, so offset from the bvec 550 * must be passed to iov iterator 551 */ 552 offset = bio->bi_iter.bi_bvec_done; 553 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 554 segments = bio_segments(bio); 555 } 556 atomic_set(&cmd->ref, 2); 557 558 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, 559 segments, blk_rq_bytes(rq)); 560 iter.iov_offset = offset; 561 562 cmd->iocb.ki_pos = pos; 563 cmd->iocb.ki_filp = file; 564 cmd->iocb.ki_complete = lo_rw_aio_complete; 565 cmd->iocb.ki_flags = IOCB_DIRECT; 566 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 567 if (cmd->css) 568 kthread_associate_blkcg(cmd->css); 569 570 if (rw == WRITE) 571 ret = call_write_iter(file, &cmd->iocb, &iter); 572 else 573 ret = call_read_iter(file, &cmd->iocb, &iter); 574 575 lo_rw_aio_do_completion(cmd); 576 kthread_associate_blkcg(NULL); 577 578 if (ret != -EIOCBQUEUED) 579 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 580 return 0; 581 } 582 583 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 584 { 585 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 586 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 587 588 /* 589 * lo_write_simple and lo_read_simple should have been covered 590 * by io submit style function like lo_rw_aio(), one blocker 591 * is that lo_read_simple() need to call flush_dcache_page after 592 * the page is written from kernel, and it isn't easy to handle 593 * this in io submit style function which submits all segments 594 * of the req at one time. And direct read IO doesn't need to 595 * run flush_dcache_page(). 596 */ 597 switch (req_op(rq)) { 598 case REQ_OP_FLUSH: 599 return lo_req_flush(lo, rq); 600 case REQ_OP_DISCARD: 601 case REQ_OP_WRITE_ZEROES: 602 return lo_discard(lo, rq, pos); 603 case REQ_OP_WRITE: 604 if (lo->transfer) 605 return lo_write_transfer(lo, rq, pos); 606 else if (cmd->use_aio) 607 return lo_rw_aio(lo, cmd, pos, WRITE); 608 else 609 return lo_write_simple(lo, rq, pos); 610 case REQ_OP_READ: 611 if (lo->transfer) 612 return lo_read_transfer(lo, rq, pos); 613 else if (cmd->use_aio) 614 return lo_rw_aio(lo, cmd, pos, READ); 615 else 616 return lo_read_simple(lo, rq, pos); 617 default: 618 WARN_ON_ONCE(1); 619 return -EIO; 620 break; 621 } 622 } 623 624 static inline void loop_update_dio(struct loop_device *lo) 625 { 626 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 627 lo->use_dio); 628 } 629 630 static void loop_reread_partitions(struct loop_device *lo, 631 struct block_device *bdev) 632 { 633 int rc; 634 635 /* 636 * bd_mutex has been held already in release path, so don't 637 * acquire it if this function is called in such case. 638 * 639 * If the reread partition isn't from release path, lo_refcnt 640 * must be at least one and it can only become zero when the 641 * current holder is released. 642 */ 643 if (!atomic_read(&lo->lo_refcnt)) 644 rc = __blkdev_reread_part(bdev); 645 else 646 rc = blkdev_reread_part(bdev); 647 if (rc) 648 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 649 __func__, lo->lo_number, lo->lo_file_name, rc); 650 } 651 652 static inline int is_loop_device(struct file *file) 653 { 654 struct inode *i = file->f_mapping->host; 655 656 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 657 } 658 659 static int loop_validate_file(struct file *file, struct block_device *bdev) 660 { 661 struct inode *inode = file->f_mapping->host; 662 struct file *f = file; 663 664 /* Avoid recursion */ 665 while (is_loop_device(f)) { 666 struct loop_device *l; 667 668 if (f->f_mapping->host->i_bdev == bdev) 669 return -EBADF; 670 671 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 672 if (l->lo_state == Lo_unbound) { 673 return -EINVAL; 674 } 675 f = l->lo_backing_file; 676 } 677 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 678 return -EINVAL; 679 return 0; 680 } 681 682 /* 683 * loop_change_fd switched the backing store of a loopback device to 684 * a new file. This is useful for operating system installers to free up 685 * the original file and in High Availability environments to switch to 686 * an alternative location for the content in case of server meltdown. 687 * This can only work if the loop device is used read-only, and if the 688 * new backing store is the same size and type as the old backing store. 689 */ 690 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 691 unsigned int arg) 692 { 693 struct file *file, *old_file; 694 int error; 695 696 error = -ENXIO; 697 if (lo->lo_state != Lo_bound) 698 goto out; 699 700 /* the loop device has to be read-only */ 701 error = -EINVAL; 702 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 703 goto out; 704 705 error = -EBADF; 706 file = fget(arg); 707 if (!file) 708 goto out; 709 710 error = loop_validate_file(file, bdev); 711 if (error) 712 goto out_putf; 713 714 old_file = lo->lo_backing_file; 715 716 error = -EINVAL; 717 718 /* size of the new backing store needs to be the same */ 719 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 720 goto out_putf; 721 722 /* and ... switch */ 723 blk_mq_freeze_queue(lo->lo_queue); 724 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 725 lo->lo_backing_file = file; 726 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 727 mapping_set_gfp_mask(file->f_mapping, 728 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 729 loop_update_dio(lo); 730 blk_mq_unfreeze_queue(lo->lo_queue); 731 732 fput(old_file); 733 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 734 loop_reread_partitions(lo, bdev); 735 return 0; 736 737 out_putf: 738 fput(file); 739 out: 740 return error; 741 } 742 743 /* loop sysfs attributes */ 744 745 static ssize_t loop_attr_show(struct device *dev, char *page, 746 ssize_t (*callback)(struct loop_device *, char *)) 747 { 748 struct gendisk *disk = dev_to_disk(dev); 749 struct loop_device *lo = disk->private_data; 750 751 return callback(lo, page); 752 } 753 754 #define LOOP_ATTR_RO(_name) \ 755 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 756 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 757 struct device_attribute *attr, char *b) \ 758 { \ 759 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 760 } \ 761 static struct device_attribute loop_attr_##_name = \ 762 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 763 764 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 765 { 766 ssize_t ret; 767 char *p = NULL; 768 769 spin_lock_irq(&lo->lo_lock); 770 if (lo->lo_backing_file) 771 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 772 spin_unlock_irq(&lo->lo_lock); 773 774 if (IS_ERR_OR_NULL(p)) 775 ret = PTR_ERR(p); 776 else { 777 ret = strlen(p); 778 memmove(buf, p, ret); 779 buf[ret++] = '\n'; 780 buf[ret] = 0; 781 } 782 783 return ret; 784 } 785 786 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 787 { 788 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 789 } 790 791 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 792 { 793 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 794 } 795 796 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 797 { 798 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 799 800 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 801 } 802 803 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 804 { 805 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 806 807 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 808 } 809 810 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 811 { 812 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 813 814 return sprintf(buf, "%s\n", dio ? "1" : "0"); 815 } 816 817 LOOP_ATTR_RO(backing_file); 818 LOOP_ATTR_RO(offset); 819 LOOP_ATTR_RO(sizelimit); 820 LOOP_ATTR_RO(autoclear); 821 LOOP_ATTR_RO(partscan); 822 LOOP_ATTR_RO(dio); 823 824 static struct attribute *loop_attrs[] = { 825 &loop_attr_backing_file.attr, 826 &loop_attr_offset.attr, 827 &loop_attr_sizelimit.attr, 828 &loop_attr_autoclear.attr, 829 &loop_attr_partscan.attr, 830 &loop_attr_dio.attr, 831 NULL, 832 }; 833 834 static struct attribute_group loop_attribute_group = { 835 .name = "loop", 836 .attrs= loop_attrs, 837 }; 838 839 static void loop_sysfs_init(struct loop_device *lo) 840 { 841 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 842 &loop_attribute_group); 843 } 844 845 static void loop_sysfs_exit(struct loop_device *lo) 846 { 847 if (lo->sysfs_inited) 848 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 849 &loop_attribute_group); 850 } 851 852 static void loop_config_discard(struct loop_device *lo) 853 { 854 struct file *file = lo->lo_backing_file; 855 struct inode *inode = file->f_mapping->host; 856 struct request_queue *q = lo->lo_queue; 857 858 /* 859 * We use punch hole to reclaim the free space used by the 860 * image a.k.a. discard. However we do not support discard if 861 * encryption is enabled, because it may give an attacker 862 * useful information. 863 */ 864 if ((!file->f_op->fallocate) || 865 lo->lo_encrypt_key_size) { 866 q->limits.discard_granularity = 0; 867 q->limits.discard_alignment = 0; 868 blk_queue_max_discard_sectors(q, 0); 869 blk_queue_max_write_zeroes_sectors(q, 0); 870 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 871 return; 872 } 873 874 q->limits.discard_granularity = inode->i_sb->s_blocksize; 875 q->limits.discard_alignment = 0; 876 877 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 878 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 879 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 880 } 881 882 static void loop_unprepare_queue(struct loop_device *lo) 883 { 884 kthread_flush_worker(&lo->worker); 885 kthread_stop(lo->worker_task); 886 } 887 888 static int loop_kthread_worker_fn(void *worker_ptr) 889 { 890 current->flags |= PF_LESS_THROTTLE; 891 return kthread_worker_fn(worker_ptr); 892 } 893 894 static int loop_prepare_queue(struct loop_device *lo) 895 { 896 kthread_init_worker(&lo->worker); 897 lo->worker_task = kthread_run(loop_kthread_worker_fn, 898 &lo->worker, "loop%d", lo->lo_number); 899 if (IS_ERR(lo->worker_task)) 900 return -ENOMEM; 901 set_user_nice(lo->worker_task, MIN_NICE); 902 return 0; 903 } 904 905 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 906 struct block_device *bdev, unsigned int arg) 907 { 908 struct file *file; 909 struct inode *inode; 910 struct address_space *mapping; 911 int lo_flags = 0; 912 int error; 913 loff_t size; 914 915 /* This is safe, since we have a reference from open(). */ 916 __module_get(THIS_MODULE); 917 918 error = -EBADF; 919 file = fget(arg); 920 if (!file) 921 goto out; 922 923 error = -EBUSY; 924 if (lo->lo_state != Lo_unbound) 925 goto out_putf; 926 927 error = loop_validate_file(file, bdev); 928 if (error) 929 goto out_putf; 930 931 mapping = file->f_mapping; 932 inode = mapping->host; 933 934 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 935 !file->f_op->write_iter) 936 lo_flags |= LO_FLAGS_READ_ONLY; 937 938 error = -EFBIG; 939 size = get_loop_size(lo, file); 940 if ((loff_t)(sector_t)size != size) 941 goto out_putf; 942 error = loop_prepare_queue(lo); 943 if (error) 944 goto out_putf; 945 946 error = 0; 947 948 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 949 950 lo->use_dio = false; 951 lo->lo_device = bdev; 952 lo->lo_flags = lo_flags; 953 lo->lo_backing_file = file; 954 lo->transfer = NULL; 955 lo->ioctl = NULL; 956 lo->lo_sizelimit = 0; 957 lo->old_gfp_mask = mapping_gfp_mask(mapping); 958 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 959 960 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 961 blk_queue_write_cache(lo->lo_queue, true, false); 962 963 loop_update_dio(lo); 964 set_capacity(lo->lo_disk, size); 965 bd_set_size(bdev, size << 9); 966 loop_sysfs_init(lo); 967 /* let user-space know about the new size */ 968 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 969 970 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 971 block_size(inode->i_bdev) : PAGE_SIZE); 972 973 lo->lo_state = Lo_bound; 974 if (part_shift) 975 lo->lo_flags |= LO_FLAGS_PARTSCAN; 976 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 977 loop_reread_partitions(lo, bdev); 978 979 /* Grab the block_device to prevent its destruction after we 980 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 981 */ 982 bdgrab(bdev); 983 return 0; 984 985 out_putf: 986 fput(file); 987 out: 988 /* This is safe: open() is still holding a reference. */ 989 module_put(THIS_MODULE); 990 return error; 991 } 992 993 static int 994 loop_release_xfer(struct loop_device *lo) 995 { 996 int err = 0; 997 struct loop_func_table *xfer = lo->lo_encryption; 998 999 if (xfer) { 1000 if (xfer->release) 1001 err = xfer->release(lo); 1002 lo->transfer = NULL; 1003 lo->lo_encryption = NULL; 1004 module_put(xfer->owner); 1005 } 1006 return err; 1007 } 1008 1009 static int 1010 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1011 const struct loop_info64 *i) 1012 { 1013 int err = 0; 1014 1015 if (xfer) { 1016 struct module *owner = xfer->owner; 1017 1018 if (!try_module_get(owner)) 1019 return -EINVAL; 1020 if (xfer->init) 1021 err = xfer->init(lo, i); 1022 if (err) 1023 module_put(owner); 1024 else 1025 lo->lo_encryption = xfer; 1026 } 1027 return err; 1028 } 1029 1030 static int loop_clr_fd(struct loop_device *lo) 1031 { 1032 struct file *filp = lo->lo_backing_file; 1033 gfp_t gfp = lo->old_gfp_mask; 1034 struct block_device *bdev = lo->lo_device; 1035 1036 if (lo->lo_state != Lo_bound) 1037 return -ENXIO; 1038 1039 /* 1040 * If we've explicitly asked to tear down the loop device, 1041 * and it has an elevated reference count, set it for auto-teardown when 1042 * the last reference goes away. This stops $!~#$@ udev from 1043 * preventing teardown because it decided that it needs to run blkid on 1044 * the loopback device whenever they appear. xfstests is notorious for 1045 * failing tests because blkid via udev races with a losetup 1046 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1047 * command to fail with EBUSY. 1048 */ 1049 if (atomic_read(&lo->lo_refcnt) > 1) { 1050 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1051 mutex_unlock(&lo->lo_ctl_mutex); 1052 return 0; 1053 } 1054 1055 if (filp == NULL) 1056 return -EINVAL; 1057 1058 /* freeze request queue during the transition */ 1059 blk_mq_freeze_queue(lo->lo_queue); 1060 1061 spin_lock_irq(&lo->lo_lock); 1062 lo->lo_state = Lo_rundown; 1063 lo->lo_backing_file = NULL; 1064 spin_unlock_irq(&lo->lo_lock); 1065 1066 loop_release_xfer(lo); 1067 lo->transfer = NULL; 1068 lo->ioctl = NULL; 1069 lo->lo_device = NULL; 1070 lo->lo_encryption = NULL; 1071 lo->lo_offset = 0; 1072 lo->lo_sizelimit = 0; 1073 lo->lo_encrypt_key_size = 0; 1074 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1075 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1076 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1077 blk_queue_logical_block_size(lo->lo_queue, 512); 1078 blk_queue_physical_block_size(lo->lo_queue, 512); 1079 blk_queue_io_min(lo->lo_queue, 512); 1080 if (bdev) { 1081 bdput(bdev); 1082 invalidate_bdev(bdev); 1083 bdev->bd_inode->i_mapping->wb_err = 0; 1084 } 1085 set_capacity(lo->lo_disk, 0); 1086 loop_sysfs_exit(lo); 1087 if (bdev) { 1088 bd_set_size(bdev, 0); 1089 /* let user-space know about this change */ 1090 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1091 } 1092 mapping_set_gfp_mask(filp->f_mapping, gfp); 1093 lo->lo_state = Lo_unbound; 1094 /* This is safe: open() is still holding a reference. */ 1095 module_put(THIS_MODULE); 1096 blk_mq_unfreeze_queue(lo->lo_queue); 1097 1098 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 1099 loop_reread_partitions(lo, bdev); 1100 lo->lo_flags = 0; 1101 if (!part_shift) 1102 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1103 loop_unprepare_queue(lo); 1104 mutex_unlock(&lo->lo_ctl_mutex); 1105 /* 1106 * Need not hold lo_ctl_mutex to fput backing file. 1107 * Calling fput holding lo_ctl_mutex triggers a circular 1108 * lock dependency possibility warning as fput can take 1109 * bd_mutex which is usually taken before lo_ctl_mutex. 1110 */ 1111 fput(filp); 1112 return 0; 1113 } 1114 1115 static int 1116 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1117 { 1118 int err; 1119 struct loop_func_table *xfer; 1120 kuid_t uid = current_uid(); 1121 1122 if (lo->lo_encrypt_key_size && 1123 !uid_eq(lo->lo_key_owner, uid) && 1124 !capable(CAP_SYS_ADMIN)) 1125 return -EPERM; 1126 if (lo->lo_state != Lo_bound) 1127 return -ENXIO; 1128 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1129 return -EINVAL; 1130 1131 /* I/O need to be drained during transfer transition */ 1132 blk_mq_freeze_queue(lo->lo_queue); 1133 1134 err = loop_release_xfer(lo); 1135 if (err) 1136 goto exit; 1137 1138 if (info->lo_encrypt_type) { 1139 unsigned int type = info->lo_encrypt_type; 1140 1141 if (type >= MAX_LO_CRYPT) { 1142 err = -EINVAL; 1143 goto exit; 1144 } 1145 xfer = xfer_funcs[type]; 1146 if (xfer == NULL) { 1147 err = -EINVAL; 1148 goto exit; 1149 } 1150 } else 1151 xfer = NULL; 1152 1153 err = loop_init_xfer(lo, xfer, info); 1154 if (err) 1155 goto exit; 1156 1157 if (lo->lo_offset != info->lo_offset || 1158 lo->lo_sizelimit != info->lo_sizelimit) { 1159 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1160 err = -EFBIG; 1161 goto exit; 1162 } 1163 } 1164 1165 loop_config_discard(lo); 1166 1167 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1168 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1169 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1170 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1171 1172 if (!xfer) 1173 xfer = &none_funcs; 1174 lo->transfer = xfer->transfer; 1175 lo->ioctl = xfer->ioctl; 1176 1177 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1178 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1179 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1180 1181 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1182 lo->lo_init[0] = info->lo_init[0]; 1183 lo->lo_init[1] = info->lo_init[1]; 1184 if (info->lo_encrypt_key_size) { 1185 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1186 info->lo_encrypt_key_size); 1187 lo->lo_key_owner = uid; 1188 } 1189 1190 /* update dio if lo_offset or transfer is changed */ 1191 __loop_update_dio(lo, lo->use_dio); 1192 1193 exit: 1194 blk_mq_unfreeze_queue(lo->lo_queue); 1195 1196 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1197 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1198 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1199 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1200 loop_reread_partitions(lo, lo->lo_device); 1201 } 1202 1203 return err; 1204 } 1205 1206 static int 1207 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1208 { 1209 struct file *file; 1210 struct kstat stat; 1211 int ret; 1212 1213 if (lo->lo_state != Lo_bound) { 1214 mutex_unlock(&lo->lo_ctl_mutex); 1215 return -ENXIO; 1216 } 1217 1218 memset(info, 0, sizeof(*info)); 1219 info->lo_number = lo->lo_number; 1220 info->lo_offset = lo->lo_offset; 1221 info->lo_sizelimit = lo->lo_sizelimit; 1222 info->lo_flags = lo->lo_flags; 1223 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1224 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1225 info->lo_encrypt_type = 1226 lo->lo_encryption ? lo->lo_encryption->number : 0; 1227 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1228 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1229 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1230 lo->lo_encrypt_key_size); 1231 } 1232 1233 /* Drop lo_ctl_mutex while we call into the filesystem. */ 1234 file = get_file(lo->lo_backing_file); 1235 mutex_unlock(&lo->lo_ctl_mutex); 1236 ret = vfs_getattr(&file->f_path, &stat, STATX_INO, 1237 AT_STATX_SYNC_AS_STAT); 1238 if (!ret) { 1239 info->lo_device = huge_encode_dev(stat.dev); 1240 info->lo_inode = stat.ino; 1241 info->lo_rdevice = huge_encode_dev(stat.rdev); 1242 } 1243 fput(file); 1244 return ret; 1245 } 1246 1247 static void 1248 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1249 { 1250 memset(info64, 0, sizeof(*info64)); 1251 info64->lo_number = info->lo_number; 1252 info64->lo_device = info->lo_device; 1253 info64->lo_inode = info->lo_inode; 1254 info64->lo_rdevice = info->lo_rdevice; 1255 info64->lo_offset = info->lo_offset; 1256 info64->lo_sizelimit = 0; 1257 info64->lo_encrypt_type = info->lo_encrypt_type; 1258 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1259 info64->lo_flags = info->lo_flags; 1260 info64->lo_init[0] = info->lo_init[0]; 1261 info64->lo_init[1] = info->lo_init[1]; 1262 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1263 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1264 else 1265 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1266 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1267 } 1268 1269 static int 1270 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1271 { 1272 memset(info, 0, sizeof(*info)); 1273 info->lo_number = info64->lo_number; 1274 info->lo_device = info64->lo_device; 1275 info->lo_inode = info64->lo_inode; 1276 info->lo_rdevice = info64->lo_rdevice; 1277 info->lo_offset = info64->lo_offset; 1278 info->lo_encrypt_type = info64->lo_encrypt_type; 1279 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1280 info->lo_flags = info64->lo_flags; 1281 info->lo_init[0] = info64->lo_init[0]; 1282 info->lo_init[1] = info64->lo_init[1]; 1283 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1284 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1285 else 1286 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1287 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1288 1289 /* error in case values were truncated */ 1290 if (info->lo_device != info64->lo_device || 1291 info->lo_rdevice != info64->lo_rdevice || 1292 info->lo_inode != info64->lo_inode || 1293 info->lo_offset != info64->lo_offset) 1294 return -EOVERFLOW; 1295 1296 return 0; 1297 } 1298 1299 static int 1300 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1301 { 1302 struct loop_info info; 1303 struct loop_info64 info64; 1304 1305 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1306 return -EFAULT; 1307 loop_info64_from_old(&info, &info64); 1308 return loop_set_status(lo, &info64); 1309 } 1310 1311 static int 1312 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1313 { 1314 struct loop_info64 info64; 1315 1316 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1317 return -EFAULT; 1318 return loop_set_status(lo, &info64); 1319 } 1320 1321 static int 1322 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1323 struct loop_info info; 1324 struct loop_info64 info64; 1325 int err; 1326 1327 if (!arg) { 1328 mutex_unlock(&lo->lo_ctl_mutex); 1329 return -EINVAL; 1330 } 1331 err = loop_get_status(lo, &info64); 1332 if (!err) 1333 err = loop_info64_to_old(&info64, &info); 1334 if (!err && copy_to_user(arg, &info, sizeof(info))) 1335 err = -EFAULT; 1336 1337 return err; 1338 } 1339 1340 static int 1341 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1342 struct loop_info64 info64; 1343 int err; 1344 1345 if (!arg) { 1346 mutex_unlock(&lo->lo_ctl_mutex); 1347 return -EINVAL; 1348 } 1349 err = loop_get_status(lo, &info64); 1350 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1351 err = -EFAULT; 1352 1353 return err; 1354 } 1355 1356 static int loop_set_capacity(struct loop_device *lo) 1357 { 1358 if (unlikely(lo->lo_state != Lo_bound)) 1359 return -ENXIO; 1360 1361 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1362 } 1363 1364 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1365 { 1366 int error = -ENXIO; 1367 if (lo->lo_state != Lo_bound) 1368 goto out; 1369 1370 __loop_update_dio(lo, !!arg); 1371 if (lo->use_dio == !!arg) 1372 return 0; 1373 error = -EINVAL; 1374 out: 1375 return error; 1376 } 1377 1378 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1379 { 1380 if (lo->lo_state != Lo_bound) 1381 return -ENXIO; 1382 1383 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1384 return -EINVAL; 1385 1386 blk_mq_freeze_queue(lo->lo_queue); 1387 1388 blk_queue_logical_block_size(lo->lo_queue, arg); 1389 blk_queue_physical_block_size(lo->lo_queue, arg); 1390 blk_queue_io_min(lo->lo_queue, arg); 1391 loop_update_dio(lo); 1392 1393 blk_mq_unfreeze_queue(lo->lo_queue); 1394 1395 return 0; 1396 } 1397 1398 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1399 unsigned int cmd, unsigned long arg) 1400 { 1401 struct loop_device *lo = bdev->bd_disk->private_data; 1402 int err; 1403 1404 err = mutex_lock_killable_nested(&lo->lo_ctl_mutex, 1); 1405 if (err) 1406 goto out_unlocked; 1407 1408 switch (cmd) { 1409 case LOOP_SET_FD: 1410 err = loop_set_fd(lo, mode, bdev, arg); 1411 break; 1412 case LOOP_CHANGE_FD: 1413 err = loop_change_fd(lo, bdev, arg); 1414 break; 1415 case LOOP_CLR_FD: 1416 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1417 err = loop_clr_fd(lo); 1418 if (!err) 1419 goto out_unlocked; 1420 break; 1421 case LOOP_SET_STATUS: 1422 err = -EPERM; 1423 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1424 err = loop_set_status_old(lo, 1425 (struct loop_info __user *)arg); 1426 break; 1427 case LOOP_GET_STATUS: 1428 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1429 /* loop_get_status() unlocks lo_ctl_mutex */ 1430 goto out_unlocked; 1431 case LOOP_SET_STATUS64: 1432 err = -EPERM; 1433 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1434 err = loop_set_status64(lo, 1435 (struct loop_info64 __user *) arg); 1436 break; 1437 case LOOP_GET_STATUS64: 1438 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1439 /* loop_get_status() unlocks lo_ctl_mutex */ 1440 goto out_unlocked; 1441 case LOOP_SET_CAPACITY: 1442 err = -EPERM; 1443 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1444 err = loop_set_capacity(lo); 1445 break; 1446 case LOOP_SET_DIRECT_IO: 1447 err = -EPERM; 1448 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1449 err = loop_set_dio(lo, arg); 1450 break; 1451 case LOOP_SET_BLOCK_SIZE: 1452 err = -EPERM; 1453 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1454 err = loop_set_block_size(lo, arg); 1455 break; 1456 default: 1457 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1458 } 1459 mutex_unlock(&lo->lo_ctl_mutex); 1460 1461 out_unlocked: 1462 return err; 1463 } 1464 1465 #ifdef CONFIG_COMPAT 1466 struct compat_loop_info { 1467 compat_int_t lo_number; /* ioctl r/o */ 1468 compat_dev_t lo_device; /* ioctl r/o */ 1469 compat_ulong_t lo_inode; /* ioctl r/o */ 1470 compat_dev_t lo_rdevice; /* ioctl r/o */ 1471 compat_int_t lo_offset; 1472 compat_int_t lo_encrypt_type; 1473 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1474 compat_int_t lo_flags; /* ioctl r/o */ 1475 char lo_name[LO_NAME_SIZE]; 1476 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1477 compat_ulong_t lo_init[2]; 1478 char reserved[4]; 1479 }; 1480 1481 /* 1482 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1483 * - noinlined to reduce stack space usage in main part of driver 1484 */ 1485 static noinline int 1486 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1487 struct loop_info64 *info64) 1488 { 1489 struct compat_loop_info info; 1490 1491 if (copy_from_user(&info, arg, sizeof(info))) 1492 return -EFAULT; 1493 1494 memset(info64, 0, sizeof(*info64)); 1495 info64->lo_number = info.lo_number; 1496 info64->lo_device = info.lo_device; 1497 info64->lo_inode = info.lo_inode; 1498 info64->lo_rdevice = info.lo_rdevice; 1499 info64->lo_offset = info.lo_offset; 1500 info64->lo_sizelimit = 0; 1501 info64->lo_encrypt_type = info.lo_encrypt_type; 1502 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1503 info64->lo_flags = info.lo_flags; 1504 info64->lo_init[0] = info.lo_init[0]; 1505 info64->lo_init[1] = info.lo_init[1]; 1506 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1507 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1508 else 1509 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1510 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1511 return 0; 1512 } 1513 1514 /* 1515 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1516 * - noinlined to reduce stack space usage in main part of driver 1517 */ 1518 static noinline int 1519 loop_info64_to_compat(const struct loop_info64 *info64, 1520 struct compat_loop_info __user *arg) 1521 { 1522 struct compat_loop_info info; 1523 1524 memset(&info, 0, sizeof(info)); 1525 info.lo_number = info64->lo_number; 1526 info.lo_device = info64->lo_device; 1527 info.lo_inode = info64->lo_inode; 1528 info.lo_rdevice = info64->lo_rdevice; 1529 info.lo_offset = info64->lo_offset; 1530 info.lo_encrypt_type = info64->lo_encrypt_type; 1531 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1532 info.lo_flags = info64->lo_flags; 1533 info.lo_init[0] = info64->lo_init[0]; 1534 info.lo_init[1] = info64->lo_init[1]; 1535 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1536 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1537 else 1538 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1539 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1540 1541 /* error in case values were truncated */ 1542 if (info.lo_device != info64->lo_device || 1543 info.lo_rdevice != info64->lo_rdevice || 1544 info.lo_inode != info64->lo_inode || 1545 info.lo_offset != info64->lo_offset || 1546 info.lo_init[0] != info64->lo_init[0] || 1547 info.lo_init[1] != info64->lo_init[1]) 1548 return -EOVERFLOW; 1549 1550 if (copy_to_user(arg, &info, sizeof(info))) 1551 return -EFAULT; 1552 return 0; 1553 } 1554 1555 static int 1556 loop_set_status_compat(struct loop_device *lo, 1557 const struct compat_loop_info __user *arg) 1558 { 1559 struct loop_info64 info64; 1560 int ret; 1561 1562 ret = loop_info64_from_compat(arg, &info64); 1563 if (ret < 0) 1564 return ret; 1565 return loop_set_status(lo, &info64); 1566 } 1567 1568 static int 1569 loop_get_status_compat(struct loop_device *lo, 1570 struct compat_loop_info __user *arg) 1571 { 1572 struct loop_info64 info64; 1573 int err; 1574 1575 if (!arg) { 1576 mutex_unlock(&lo->lo_ctl_mutex); 1577 return -EINVAL; 1578 } 1579 err = loop_get_status(lo, &info64); 1580 if (!err) 1581 err = loop_info64_to_compat(&info64, arg); 1582 return err; 1583 } 1584 1585 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1586 unsigned int cmd, unsigned long arg) 1587 { 1588 struct loop_device *lo = bdev->bd_disk->private_data; 1589 int err; 1590 1591 switch(cmd) { 1592 case LOOP_SET_STATUS: 1593 err = mutex_lock_killable(&lo->lo_ctl_mutex); 1594 if (!err) { 1595 err = loop_set_status_compat(lo, 1596 (const struct compat_loop_info __user *)arg); 1597 mutex_unlock(&lo->lo_ctl_mutex); 1598 } 1599 break; 1600 case LOOP_GET_STATUS: 1601 err = mutex_lock_killable(&lo->lo_ctl_mutex); 1602 if (!err) { 1603 err = loop_get_status_compat(lo, 1604 (struct compat_loop_info __user *)arg); 1605 /* loop_get_status() unlocks lo_ctl_mutex */ 1606 } 1607 break; 1608 case LOOP_SET_CAPACITY: 1609 case LOOP_CLR_FD: 1610 case LOOP_GET_STATUS64: 1611 case LOOP_SET_STATUS64: 1612 arg = (unsigned long) compat_ptr(arg); 1613 /* fall through */ 1614 case LOOP_SET_FD: 1615 case LOOP_CHANGE_FD: 1616 case LOOP_SET_BLOCK_SIZE: 1617 err = lo_ioctl(bdev, mode, cmd, arg); 1618 break; 1619 default: 1620 err = -ENOIOCTLCMD; 1621 break; 1622 } 1623 return err; 1624 } 1625 #endif 1626 1627 static int lo_open(struct block_device *bdev, fmode_t mode) 1628 { 1629 struct loop_device *lo; 1630 int err = 0; 1631 1632 mutex_lock(&loop_index_mutex); 1633 lo = bdev->bd_disk->private_data; 1634 if (!lo) { 1635 err = -ENXIO; 1636 goto out; 1637 } 1638 1639 atomic_inc(&lo->lo_refcnt); 1640 out: 1641 mutex_unlock(&loop_index_mutex); 1642 return err; 1643 } 1644 1645 static void __lo_release(struct loop_device *lo) 1646 { 1647 int err; 1648 1649 if (atomic_dec_return(&lo->lo_refcnt)) 1650 return; 1651 1652 mutex_lock(&lo->lo_ctl_mutex); 1653 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1654 /* 1655 * In autoclear mode, stop the loop thread 1656 * and remove configuration after last close. 1657 */ 1658 err = loop_clr_fd(lo); 1659 if (!err) 1660 return; 1661 } else if (lo->lo_state == Lo_bound) { 1662 /* 1663 * Otherwise keep thread (if running) and config, 1664 * but flush possible ongoing bios in thread. 1665 */ 1666 blk_mq_freeze_queue(lo->lo_queue); 1667 blk_mq_unfreeze_queue(lo->lo_queue); 1668 } 1669 1670 mutex_unlock(&lo->lo_ctl_mutex); 1671 } 1672 1673 static void lo_release(struct gendisk *disk, fmode_t mode) 1674 { 1675 mutex_lock(&loop_index_mutex); 1676 __lo_release(disk->private_data); 1677 mutex_unlock(&loop_index_mutex); 1678 } 1679 1680 static const struct block_device_operations lo_fops = { 1681 .owner = THIS_MODULE, 1682 .open = lo_open, 1683 .release = lo_release, 1684 .ioctl = lo_ioctl, 1685 #ifdef CONFIG_COMPAT 1686 .compat_ioctl = lo_compat_ioctl, 1687 #endif 1688 }; 1689 1690 /* 1691 * And now the modules code and kernel interface. 1692 */ 1693 static int max_loop; 1694 module_param(max_loop, int, 0444); 1695 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1696 module_param(max_part, int, 0444); 1697 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1698 MODULE_LICENSE("GPL"); 1699 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1700 1701 int loop_register_transfer(struct loop_func_table *funcs) 1702 { 1703 unsigned int n = funcs->number; 1704 1705 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1706 return -EINVAL; 1707 xfer_funcs[n] = funcs; 1708 return 0; 1709 } 1710 1711 static int unregister_transfer_cb(int id, void *ptr, void *data) 1712 { 1713 struct loop_device *lo = ptr; 1714 struct loop_func_table *xfer = data; 1715 1716 mutex_lock(&lo->lo_ctl_mutex); 1717 if (lo->lo_encryption == xfer) 1718 loop_release_xfer(lo); 1719 mutex_unlock(&lo->lo_ctl_mutex); 1720 return 0; 1721 } 1722 1723 int loop_unregister_transfer(int number) 1724 { 1725 unsigned int n = number; 1726 struct loop_func_table *xfer; 1727 1728 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1729 return -EINVAL; 1730 1731 xfer_funcs[n] = NULL; 1732 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1733 return 0; 1734 } 1735 1736 EXPORT_SYMBOL(loop_register_transfer); 1737 EXPORT_SYMBOL(loop_unregister_transfer); 1738 1739 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1740 const struct blk_mq_queue_data *bd) 1741 { 1742 struct request *rq = bd->rq; 1743 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1744 struct loop_device *lo = rq->q->queuedata; 1745 1746 blk_mq_start_request(rq); 1747 1748 if (lo->lo_state != Lo_bound) 1749 return BLK_STS_IOERR; 1750 1751 switch (req_op(rq)) { 1752 case REQ_OP_FLUSH: 1753 case REQ_OP_DISCARD: 1754 case REQ_OP_WRITE_ZEROES: 1755 cmd->use_aio = false; 1756 break; 1757 default: 1758 cmd->use_aio = lo->use_dio; 1759 break; 1760 } 1761 1762 /* always use the first bio's css */ 1763 #ifdef CONFIG_BLK_CGROUP 1764 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 1765 cmd->css = &bio_blkcg(rq->bio)->css; 1766 css_get(cmd->css); 1767 } else 1768 #endif 1769 cmd->css = NULL; 1770 kthread_queue_work(&lo->worker, &cmd->work); 1771 1772 return BLK_STS_OK; 1773 } 1774 1775 static void loop_handle_cmd(struct loop_cmd *cmd) 1776 { 1777 struct request *rq = blk_mq_rq_from_pdu(cmd); 1778 const bool write = op_is_write(req_op(rq)); 1779 struct loop_device *lo = rq->q->queuedata; 1780 int ret = 0; 1781 1782 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1783 ret = -EIO; 1784 goto failed; 1785 } 1786 1787 ret = do_req_filebacked(lo, rq); 1788 failed: 1789 /* complete non-aio request */ 1790 if (!cmd->use_aio || ret) { 1791 cmd->ret = ret ? -EIO : 0; 1792 blk_mq_complete_request(rq); 1793 } 1794 } 1795 1796 static void loop_queue_work(struct kthread_work *work) 1797 { 1798 struct loop_cmd *cmd = 1799 container_of(work, struct loop_cmd, work); 1800 1801 loop_handle_cmd(cmd); 1802 } 1803 1804 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1805 unsigned int hctx_idx, unsigned int numa_node) 1806 { 1807 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1808 1809 kthread_init_work(&cmd->work, loop_queue_work); 1810 return 0; 1811 } 1812 1813 static const struct blk_mq_ops loop_mq_ops = { 1814 .queue_rq = loop_queue_rq, 1815 .init_request = loop_init_request, 1816 .complete = lo_complete_rq, 1817 }; 1818 1819 static int loop_add(struct loop_device **l, int i) 1820 { 1821 struct loop_device *lo; 1822 struct gendisk *disk; 1823 int err; 1824 1825 err = -ENOMEM; 1826 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1827 if (!lo) 1828 goto out; 1829 1830 lo->lo_state = Lo_unbound; 1831 1832 /* allocate id, if @id >= 0, we're requesting that specific id */ 1833 if (i >= 0) { 1834 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1835 if (err == -ENOSPC) 1836 err = -EEXIST; 1837 } else { 1838 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1839 } 1840 if (err < 0) 1841 goto out_free_dev; 1842 i = err; 1843 1844 err = -ENOMEM; 1845 lo->tag_set.ops = &loop_mq_ops; 1846 lo->tag_set.nr_hw_queues = 1; 1847 lo->tag_set.queue_depth = 128; 1848 lo->tag_set.numa_node = NUMA_NO_NODE; 1849 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1850 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1851 lo->tag_set.driver_data = lo; 1852 1853 err = blk_mq_alloc_tag_set(&lo->tag_set); 1854 if (err) 1855 goto out_free_idr; 1856 1857 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1858 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1859 err = PTR_ERR(lo->lo_queue); 1860 goto out_cleanup_tags; 1861 } 1862 lo->lo_queue->queuedata = lo; 1863 1864 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1865 1866 /* 1867 * By default, we do buffer IO, so it doesn't make sense to enable 1868 * merge because the I/O submitted to backing file is handled page by 1869 * page. For directio mode, merge does help to dispatch bigger request 1870 * to underlayer disk. We will enable merge once directio is enabled. 1871 */ 1872 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1873 1874 err = -ENOMEM; 1875 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1876 if (!disk) 1877 goto out_free_queue; 1878 1879 /* 1880 * Disable partition scanning by default. The in-kernel partition 1881 * scanning can be requested individually per-device during its 1882 * setup. Userspace can always add and remove partitions from all 1883 * devices. The needed partition minors are allocated from the 1884 * extended minor space, the main loop device numbers will continue 1885 * to match the loop minors, regardless of the number of partitions 1886 * used. 1887 * 1888 * If max_part is given, partition scanning is globally enabled for 1889 * all loop devices. The minors for the main loop devices will be 1890 * multiples of max_part. 1891 * 1892 * Note: Global-for-all-devices, set-only-at-init, read-only module 1893 * parameteters like 'max_loop' and 'max_part' make things needlessly 1894 * complicated, are too static, inflexible and may surprise 1895 * userspace tools. Parameters like this in general should be avoided. 1896 */ 1897 if (!part_shift) 1898 disk->flags |= GENHD_FL_NO_PART_SCAN; 1899 disk->flags |= GENHD_FL_EXT_DEVT; 1900 mutex_init(&lo->lo_ctl_mutex); 1901 atomic_set(&lo->lo_refcnt, 0); 1902 lo->lo_number = i; 1903 spin_lock_init(&lo->lo_lock); 1904 disk->major = LOOP_MAJOR; 1905 disk->first_minor = i << part_shift; 1906 disk->fops = &lo_fops; 1907 disk->private_data = lo; 1908 disk->queue = lo->lo_queue; 1909 sprintf(disk->disk_name, "loop%d", i); 1910 add_disk(disk); 1911 *l = lo; 1912 return lo->lo_number; 1913 1914 out_free_queue: 1915 blk_cleanup_queue(lo->lo_queue); 1916 out_cleanup_tags: 1917 blk_mq_free_tag_set(&lo->tag_set); 1918 out_free_idr: 1919 idr_remove(&loop_index_idr, i); 1920 out_free_dev: 1921 kfree(lo); 1922 out: 1923 return err; 1924 } 1925 1926 static void loop_remove(struct loop_device *lo) 1927 { 1928 del_gendisk(lo->lo_disk); 1929 blk_cleanup_queue(lo->lo_queue); 1930 blk_mq_free_tag_set(&lo->tag_set); 1931 put_disk(lo->lo_disk); 1932 kfree(lo); 1933 } 1934 1935 static int find_free_cb(int id, void *ptr, void *data) 1936 { 1937 struct loop_device *lo = ptr; 1938 struct loop_device **l = data; 1939 1940 if (lo->lo_state == Lo_unbound) { 1941 *l = lo; 1942 return 1; 1943 } 1944 return 0; 1945 } 1946 1947 static int loop_lookup(struct loop_device **l, int i) 1948 { 1949 struct loop_device *lo; 1950 int ret = -ENODEV; 1951 1952 if (i < 0) { 1953 int err; 1954 1955 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1956 if (err == 1) { 1957 *l = lo; 1958 ret = lo->lo_number; 1959 } 1960 goto out; 1961 } 1962 1963 /* lookup and return a specific i */ 1964 lo = idr_find(&loop_index_idr, i); 1965 if (lo) { 1966 *l = lo; 1967 ret = lo->lo_number; 1968 } 1969 out: 1970 return ret; 1971 } 1972 1973 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1974 { 1975 struct loop_device *lo; 1976 struct kobject *kobj; 1977 int err; 1978 1979 mutex_lock(&loop_index_mutex); 1980 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1981 if (err < 0) 1982 err = loop_add(&lo, MINOR(dev) >> part_shift); 1983 if (err < 0) 1984 kobj = NULL; 1985 else 1986 kobj = get_disk_and_module(lo->lo_disk); 1987 mutex_unlock(&loop_index_mutex); 1988 1989 *part = 0; 1990 return kobj; 1991 } 1992 1993 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1994 unsigned long parm) 1995 { 1996 struct loop_device *lo; 1997 int ret = -ENOSYS; 1998 1999 mutex_lock(&loop_index_mutex); 2000 switch (cmd) { 2001 case LOOP_CTL_ADD: 2002 ret = loop_lookup(&lo, parm); 2003 if (ret >= 0) { 2004 ret = -EEXIST; 2005 break; 2006 } 2007 ret = loop_add(&lo, parm); 2008 break; 2009 case LOOP_CTL_REMOVE: 2010 ret = loop_lookup(&lo, parm); 2011 if (ret < 0) 2012 break; 2013 ret = mutex_lock_killable(&lo->lo_ctl_mutex); 2014 if (ret) 2015 break; 2016 if (lo->lo_state != Lo_unbound) { 2017 ret = -EBUSY; 2018 mutex_unlock(&lo->lo_ctl_mutex); 2019 break; 2020 } 2021 if (atomic_read(&lo->lo_refcnt) > 0) { 2022 ret = -EBUSY; 2023 mutex_unlock(&lo->lo_ctl_mutex); 2024 break; 2025 } 2026 lo->lo_disk->private_data = NULL; 2027 mutex_unlock(&lo->lo_ctl_mutex); 2028 idr_remove(&loop_index_idr, lo->lo_number); 2029 loop_remove(lo); 2030 break; 2031 case LOOP_CTL_GET_FREE: 2032 ret = loop_lookup(&lo, -1); 2033 if (ret >= 0) 2034 break; 2035 ret = loop_add(&lo, -1); 2036 } 2037 mutex_unlock(&loop_index_mutex); 2038 2039 return ret; 2040 } 2041 2042 static const struct file_operations loop_ctl_fops = { 2043 .open = nonseekable_open, 2044 .unlocked_ioctl = loop_control_ioctl, 2045 .compat_ioctl = loop_control_ioctl, 2046 .owner = THIS_MODULE, 2047 .llseek = noop_llseek, 2048 }; 2049 2050 static struct miscdevice loop_misc = { 2051 .minor = LOOP_CTRL_MINOR, 2052 .name = "loop-control", 2053 .fops = &loop_ctl_fops, 2054 }; 2055 2056 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2057 MODULE_ALIAS("devname:loop-control"); 2058 2059 static int __init loop_init(void) 2060 { 2061 int i, nr; 2062 unsigned long range; 2063 struct loop_device *lo; 2064 int err; 2065 2066 part_shift = 0; 2067 if (max_part > 0) { 2068 part_shift = fls(max_part); 2069 2070 /* 2071 * Adjust max_part according to part_shift as it is exported 2072 * to user space so that user can decide correct minor number 2073 * if [s]he want to create more devices. 2074 * 2075 * Note that -1 is required because partition 0 is reserved 2076 * for the whole disk. 2077 */ 2078 max_part = (1UL << part_shift) - 1; 2079 } 2080 2081 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2082 err = -EINVAL; 2083 goto err_out; 2084 } 2085 2086 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2087 err = -EINVAL; 2088 goto err_out; 2089 } 2090 2091 /* 2092 * If max_loop is specified, create that many devices upfront. 2093 * This also becomes a hard limit. If max_loop is not specified, 2094 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2095 * init time. Loop devices can be requested on-demand with the 2096 * /dev/loop-control interface, or be instantiated by accessing 2097 * a 'dead' device node. 2098 */ 2099 if (max_loop) { 2100 nr = max_loop; 2101 range = max_loop << part_shift; 2102 } else { 2103 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2104 range = 1UL << MINORBITS; 2105 } 2106 2107 err = misc_register(&loop_misc); 2108 if (err < 0) 2109 goto err_out; 2110 2111 2112 if (register_blkdev(LOOP_MAJOR, "loop")) { 2113 err = -EIO; 2114 goto misc_out; 2115 } 2116 2117 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2118 THIS_MODULE, loop_probe, NULL, NULL); 2119 2120 /* pre-create number of devices given by config or max_loop */ 2121 mutex_lock(&loop_index_mutex); 2122 for (i = 0; i < nr; i++) 2123 loop_add(&lo, i); 2124 mutex_unlock(&loop_index_mutex); 2125 2126 printk(KERN_INFO "loop: module loaded\n"); 2127 return 0; 2128 2129 misc_out: 2130 misc_deregister(&loop_misc); 2131 err_out: 2132 return err; 2133 } 2134 2135 static int loop_exit_cb(int id, void *ptr, void *data) 2136 { 2137 struct loop_device *lo = ptr; 2138 2139 loop_remove(lo); 2140 return 0; 2141 } 2142 2143 static void __exit loop_exit(void) 2144 { 2145 unsigned long range; 2146 2147 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2148 2149 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2150 idr_destroy(&loop_index_idr); 2151 2152 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2153 unregister_blkdev(LOOP_MAJOR, "loop"); 2154 2155 misc_deregister(&loop_misc); 2156 } 2157 2158 module_init(loop_init); 2159 module_exit(loop_exit); 2160 2161 #ifndef MODULE 2162 static int __init max_loop_setup(char *str) 2163 { 2164 max_loop = simple_strtol(str, NULL, 0); 2165 return 1; 2166 } 2167 2168 __setup("max_loop=", max_loop_setup); 2169 #endif 2170