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 81 #include "loop.h" 82 83 #include <linux/uaccess.h> 84 85 static DEFINE_IDR(loop_index_idr); 86 static DEFINE_MUTEX(loop_ctl_mutex); 87 88 static int max_part; 89 static int part_shift; 90 91 static int transfer_xor(struct loop_device *lo, int cmd, 92 struct page *raw_page, unsigned raw_off, 93 struct page *loop_page, unsigned loop_off, 94 int size, sector_t real_block) 95 { 96 char *raw_buf = kmap_atomic(raw_page) + raw_off; 97 char *loop_buf = kmap_atomic(loop_page) + loop_off; 98 char *in, *out, *key; 99 int i, keysize; 100 101 if (cmd == READ) { 102 in = raw_buf; 103 out = loop_buf; 104 } else { 105 in = loop_buf; 106 out = raw_buf; 107 } 108 109 key = lo->lo_encrypt_key; 110 keysize = lo->lo_encrypt_key_size; 111 for (i = 0; i < size; i++) 112 *out++ = *in++ ^ key[(i & 511) % keysize]; 113 114 kunmap_atomic(loop_buf); 115 kunmap_atomic(raw_buf); 116 cond_resched(); 117 return 0; 118 } 119 120 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 121 { 122 if (unlikely(info->lo_encrypt_key_size <= 0)) 123 return -EINVAL; 124 return 0; 125 } 126 127 static struct loop_func_table none_funcs = { 128 .number = LO_CRYPT_NONE, 129 }; 130 131 static struct loop_func_table xor_funcs = { 132 .number = LO_CRYPT_XOR, 133 .transfer = transfer_xor, 134 .init = xor_init 135 }; 136 137 /* xfer_funcs[0] is special - its release function is never called */ 138 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 139 &none_funcs, 140 &xor_funcs 141 }; 142 143 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 144 { 145 loff_t loopsize; 146 147 /* Compute loopsize in bytes */ 148 loopsize = i_size_read(file->f_mapping->host); 149 if (offset > 0) 150 loopsize -= offset; 151 /* offset is beyond i_size, weird but possible */ 152 if (loopsize < 0) 153 return 0; 154 155 if (sizelimit > 0 && sizelimit < loopsize) 156 loopsize = sizelimit; 157 /* 158 * Unfortunately, if we want to do I/O on the device, 159 * the number of 512-byte sectors has to fit into a sector_t. 160 */ 161 return loopsize >> 9; 162 } 163 164 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 165 { 166 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 167 } 168 169 static void __loop_update_dio(struct loop_device *lo, bool dio) 170 { 171 struct file *file = lo->lo_backing_file; 172 struct address_space *mapping = file->f_mapping; 173 struct inode *inode = mapping->host; 174 unsigned short sb_bsize = 0; 175 unsigned dio_align = 0; 176 bool use_dio; 177 178 if (inode->i_sb->s_bdev) { 179 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 180 dio_align = sb_bsize - 1; 181 } 182 183 /* 184 * We support direct I/O only if lo_offset is aligned with the 185 * logical I/O size of backing device, and the logical block 186 * size of loop is bigger than the backing device's and the loop 187 * needn't transform transfer. 188 * 189 * TODO: the above condition may be loosed in the future, and 190 * direct I/O may be switched runtime at that time because most 191 * of requests in sane applications should be PAGE_SIZE aligned 192 */ 193 if (dio) { 194 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 195 !(lo->lo_offset & dio_align) && 196 mapping->a_ops->direct_IO && 197 !lo->transfer) 198 use_dio = true; 199 else 200 use_dio = false; 201 } else { 202 use_dio = false; 203 } 204 205 if (lo->use_dio == use_dio) 206 return; 207 208 /* flush dirty pages before changing direct IO */ 209 vfs_fsync(file, 0); 210 211 /* 212 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 213 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 214 * will get updated by ioctl(LOOP_GET_STATUS) 215 */ 216 blk_mq_freeze_queue(lo->lo_queue); 217 lo->use_dio = use_dio; 218 if (use_dio) { 219 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 220 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 221 } else { 222 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 223 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 224 } 225 blk_mq_unfreeze_queue(lo->lo_queue); 226 } 227 228 static int 229 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 230 { 231 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 232 sector_t x = (sector_t)size; 233 struct block_device *bdev = lo->lo_device; 234 235 if (unlikely((loff_t)x != size)) 236 return -EFBIG; 237 if (lo->lo_offset != offset) 238 lo->lo_offset = offset; 239 if (lo->lo_sizelimit != sizelimit) 240 lo->lo_sizelimit = sizelimit; 241 set_capacity(lo->lo_disk, x); 242 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 243 /* let user-space know about the new size */ 244 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 245 return 0; 246 } 247 248 static inline int 249 lo_do_transfer(struct loop_device *lo, int cmd, 250 struct page *rpage, unsigned roffs, 251 struct page *lpage, unsigned loffs, 252 int size, sector_t rblock) 253 { 254 int ret; 255 256 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 257 if (likely(!ret)) 258 return 0; 259 260 printk_ratelimited(KERN_ERR 261 "loop: Transfer error at byte offset %llu, length %i.\n", 262 (unsigned long long)rblock << 9, size); 263 return ret; 264 } 265 266 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 267 { 268 struct iov_iter i; 269 ssize_t bw; 270 271 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 272 273 file_start_write(file); 274 bw = vfs_iter_write(file, &i, ppos, 0); 275 file_end_write(file); 276 277 if (likely(bw == bvec->bv_len)) 278 return 0; 279 280 printk_ratelimited(KERN_ERR 281 "loop: Write error at byte offset %llu, length %i.\n", 282 (unsigned long long)*ppos, bvec->bv_len); 283 if (bw >= 0) 284 bw = -EIO; 285 return bw; 286 } 287 288 static int lo_write_simple(struct loop_device *lo, struct request *rq, 289 loff_t pos) 290 { 291 struct bio_vec bvec; 292 struct req_iterator iter; 293 int ret = 0; 294 295 rq_for_each_segment(bvec, rq, iter) { 296 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 297 if (ret < 0) 298 break; 299 cond_resched(); 300 } 301 302 return ret; 303 } 304 305 /* 306 * This is the slow, transforming version that needs to double buffer the 307 * data as it cannot do the transformations in place without having direct 308 * access to the destination pages of the backing file. 309 */ 310 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 311 loff_t pos) 312 { 313 struct bio_vec bvec, b; 314 struct req_iterator iter; 315 struct page *page; 316 int ret = 0; 317 318 page = alloc_page(GFP_NOIO); 319 if (unlikely(!page)) 320 return -ENOMEM; 321 322 rq_for_each_segment(bvec, rq, iter) { 323 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 324 bvec.bv_offset, bvec.bv_len, pos >> 9); 325 if (unlikely(ret)) 326 break; 327 328 b.bv_page = page; 329 b.bv_offset = 0; 330 b.bv_len = bvec.bv_len; 331 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 332 if (ret < 0) 333 break; 334 } 335 336 __free_page(page); 337 return ret; 338 } 339 340 static int lo_read_simple(struct loop_device *lo, struct request *rq, 341 loff_t pos) 342 { 343 struct bio_vec bvec; 344 struct req_iterator iter; 345 struct iov_iter i; 346 ssize_t len; 347 348 rq_for_each_segment(bvec, rq, iter) { 349 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 350 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 351 if (len < 0) 352 return len; 353 354 flush_dcache_page(bvec.bv_page); 355 356 if (len != bvec.bv_len) { 357 struct bio *bio; 358 359 __rq_for_each_bio(bio, rq) 360 zero_fill_bio(bio); 361 break; 362 } 363 cond_resched(); 364 } 365 366 return 0; 367 } 368 369 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 370 loff_t pos) 371 { 372 struct bio_vec bvec, b; 373 struct req_iterator iter; 374 struct iov_iter i; 375 struct page *page; 376 ssize_t len; 377 int ret = 0; 378 379 page = alloc_page(GFP_NOIO); 380 if (unlikely(!page)) 381 return -ENOMEM; 382 383 rq_for_each_segment(bvec, rq, iter) { 384 loff_t offset = pos; 385 386 b.bv_page = page; 387 b.bv_offset = 0; 388 b.bv_len = bvec.bv_len; 389 390 iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 391 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 392 if (len < 0) { 393 ret = len; 394 goto out_free_page; 395 } 396 397 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 398 bvec.bv_offset, len, offset >> 9); 399 if (ret) 400 goto out_free_page; 401 402 flush_dcache_page(bvec.bv_page); 403 404 if (len != bvec.bv_len) { 405 struct bio *bio; 406 407 __rq_for_each_bio(bio, rq) 408 zero_fill_bio(bio); 409 break; 410 } 411 } 412 413 ret = 0; 414 out_free_page: 415 __free_page(page); 416 return ret; 417 } 418 419 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 420 { 421 /* 422 * We use punch hole to reclaim the free space used by the 423 * image a.k.a. discard. However we do not support discard if 424 * encryption is enabled, because it may give an attacker 425 * useful information. 426 */ 427 struct file *file = lo->lo_backing_file; 428 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 429 int ret; 430 431 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 432 ret = -EOPNOTSUPP; 433 goto out; 434 } 435 436 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 437 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 438 ret = -EIO; 439 out: 440 return ret; 441 } 442 443 static int lo_req_flush(struct loop_device *lo, struct request *rq) 444 { 445 struct file *file = lo->lo_backing_file; 446 int ret = vfs_fsync(file, 0); 447 if (unlikely(ret && ret != -EINVAL)) 448 ret = -EIO; 449 450 return ret; 451 } 452 453 static void lo_complete_rq(struct request *rq) 454 { 455 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 456 blk_status_t ret = BLK_STS_OK; 457 458 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 459 req_op(rq) != REQ_OP_READ) { 460 if (cmd->ret < 0) 461 ret = BLK_STS_IOERR; 462 goto end_io; 463 } 464 465 /* 466 * Short READ - if we got some data, advance our request and 467 * retry it. If we got no data, end the rest with EIO. 468 */ 469 if (cmd->ret) { 470 blk_update_request(rq, BLK_STS_OK, cmd->ret); 471 cmd->ret = 0; 472 blk_mq_requeue_request(rq, true); 473 } else { 474 if (cmd->use_aio) { 475 struct bio *bio = rq->bio; 476 477 while (bio) { 478 zero_fill_bio(bio); 479 bio = bio->bi_next; 480 } 481 } 482 ret = BLK_STS_IOERR; 483 end_io: 484 blk_mq_end_request(rq, ret); 485 } 486 } 487 488 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 489 { 490 struct request *rq = blk_mq_rq_from_pdu(cmd); 491 492 if (!atomic_dec_and_test(&cmd->ref)) 493 return; 494 kfree(cmd->bvec); 495 cmd->bvec = NULL; 496 blk_mq_complete_request(rq); 497 } 498 499 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 500 { 501 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 502 503 if (cmd->css) 504 css_put(cmd->css); 505 cmd->ret = ret; 506 lo_rw_aio_do_completion(cmd); 507 } 508 509 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 510 loff_t pos, bool rw) 511 { 512 struct iov_iter iter; 513 struct bio_vec *bvec; 514 struct request *rq = blk_mq_rq_from_pdu(cmd); 515 struct bio *bio = rq->bio; 516 struct file *file = lo->lo_backing_file; 517 unsigned int offset; 518 int segments = 0; 519 int ret; 520 521 if (rq->bio != rq->biotail) { 522 struct req_iterator iter; 523 struct bio_vec tmp; 524 525 __rq_for_each_bio(bio, rq) 526 segments += bio_segments(bio); 527 bvec = kmalloc_array(segments, sizeof(struct bio_vec), 528 GFP_NOIO); 529 if (!bvec) 530 return -EIO; 531 cmd->bvec = bvec; 532 533 /* 534 * The bios of the request may be started from the middle of 535 * the 'bvec' because of bio splitting, so we can't directly 536 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment 537 * API will take care of all details for us. 538 */ 539 rq_for_each_segment(tmp, rq, iter) { 540 *bvec = tmp; 541 bvec++; 542 } 543 bvec = cmd->bvec; 544 offset = 0; 545 } else { 546 /* 547 * Same here, this bio may be started from the middle of the 548 * 'bvec' because of bio splitting, so offset from the bvec 549 * must be passed to iov iterator 550 */ 551 offset = bio->bi_iter.bi_bvec_done; 552 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 553 segments = bio_segments(bio); 554 } 555 atomic_set(&cmd->ref, 2); 556 557 iov_iter_bvec(&iter, rw, bvec, segments, blk_rq_bytes(rq)); 558 iter.iov_offset = offset; 559 560 cmd->iocb.ki_pos = pos; 561 cmd->iocb.ki_filp = file; 562 cmd->iocb.ki_complete = lo_rw_aio_complete; 563 cmd->iocb.ki_flags = IOCB_DIRECT; 564 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 565 if (cmd->css) 566 kthread_associate_blkcg(cmd->css); 567 568 if (rw == WRITE) 569 ret = call_write_iter(file, &cmd->iocb, &iter); 570 else 571 ret = call_read_iter(file, &cmd->iocb, &iter); 572 573 lo_rw_aio_do_completion(cmd); 574 kthread_associate_blkcg(NULL); 575 576 if (ret != -EIOCBQUEUED) 577 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 578 return 0; 579 } 580 581 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 582 { 583 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 584 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 585 586 /* 587 * lo_write_simple and lo_read_simple should have been covered 588 * by io submit style function like lo_rw_aio(), one blocker 589 * is that lo_read_simple() need to call flush_dcache_page after 590 * the page is written from kernel, and it isn't easy to handle 591 * this in io submit style function which submits all segments 592 * of the req at one time. And direct read IO doesn't need to 593 * run flush_dcache_page(). 594 */ 595 switch (req_op(rq)) { 596 case REQ_OP_FLUSH: 597 return lo_req_flush(lo, rq); 598 case REQ_OP_DISCARD: 599 case REQ_OP_WRITE_ZEROES: 600 return lo_discard(lo, rq, pos); 601 case REQ_OP_WRITE: 602 if (lo->transfer) 603 return lo_write_transfer(lo, rq, pos); 604 else if (cmd->use_aio) 605 return lo_rw_aio(lo, cmd, pos, WRITE); 606 else 607 return lo_write_simple(lo, rq, pos); 608 case REQ_OP_READ: 609 if (lo->transfer) 610 return lo_read_transfer(lo, rq, pos); 611 else if (cmd->use_aio) 612 return lo_rw_aio(lo, cmd, pos, READ); 613 else 614 return lo_read_simple(lo, rq, pos); 615 default: 616 WARN_ON_ONCE(1); 617 return -EIO; 618 break; 619 } 620 } 621 622 static inline void loop_update_dio(struct loop_device *lo) 623 { 624 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 625 lo->use_dio); 626 } 627 628 static void loop_reread_partitions(struct loop_device *lo, 629 struct block_device *bdev) 630 { 631 int rc; 632 633 rc = blkdev_reread_part(bdev); 634 if (rc) 635 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 636 __func__, lo->lo_number, lo->lo_file_name, rc); 637 } 638 639 static inline int is_loop_device(struct file *file) 640 { 641 struct inode *i = file->f_mapping->host; 642 643 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 644 } 645 646 static int loop_validate_file(struct file *file, struct block_device *bdev) 647 { 648 struct inode *inode = file->f_mapping->host; 649 struct file *f = file; 650 651 /* Avoid recursion */ 652 while (is_loop_device(f)) { 653 struct loop_device *l; 654 655 if (f->f_mapping->host->i_bdev == bdev) 656 return -EBADF; 657 658 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 659 if (l->lo_state == Lo_unbound) { 660 return -EINVAL; 661 } 662 f = l->lo_backing_file; 663 } 664 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 665 return -EINVAL; 666 return 0; 667 } 668 669 /* 670 * loop_change_fd switched the backing store of a loopback device to 671 * a new file. This is useful for operating system installers to free up 672 * the original file and in High Availability environments to switch to 673 * an alternative location for the content in case of server meltdown. 674 * This can only work if the loop device is used read-only, and if the 675 * new backing store is the same size and type as the old backing store. 676 */ 677 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 678 unsigned int arg) 679 { 680 struct file *file, *old_file; 681 int error; 682 bool partscan; 683 684 error = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 685 if (error) 686 return error; 687 error = -ENXIO; 688 if (lo->lo_state != Lo_bound) 689 goto out_unlock; 690 691 /* the loop device has to be read-only */ 692 error = -EINVAL; 693 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 694 goto out_unlock; 695 696 error = -EBADF; 697 file = fget(arg); 698 if (!file) 699 goto out_unlock; 700 701 error = loop_validate_file(file, bdev); 702 if (error) 703 goto out_putf; 704 705 old_file = lo->lo_backing_file; 706 707 error = -EINVAL; 708 709 /* size of the new backing store needs to be the same */ 710 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 711 goto out_putf; 712 713 /* and ... switch */ 714 blk_mq_freeze_queue(lo->lo_queue); 715 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 716 lo->lo_backing_file = file; 717 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 718 mapping_set_gfp_mask(file->f_mapping, 719 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 720 loop_update_dio(lo); 721 blk_mq_unfreeze_queue(lo->lo_queue); 722 723 fput(old_file); 724 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 725 mutex_unlock(&loop_ctl_mutex); 726 if (partscan) 727 loop_reread_partitions(lo, bdev); 728 return 0; 729 730 out_putf: 731 fput(file); 732 out_unlock: 733 mutex_unlock(&loop_ctl_mutex); 734 return error; 735 } 736 737 /* loop sysfs attributes */ 738 739 static ssize_t loop_attr_show(struct device *dev, char *page, 740 ssize_t (*callback)(struct loop_device *, char *)) 741 { 742 struct gendisk *disk = dev_to_disk(dev); 743 struct loop_device *lo = disk->private_data; 744 745 return callback(lo, page); 746 } 747 748 #define LOOP_ATTR_RO(_name) \ 749 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 750 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 751 struct device_attribute *attr, char *b) \ 752 { \ 753 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 754 } \ 755 static struct device_attribute loop_attr_##_name = \ 756 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 757 758 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 759 { 760 ssize_t ret; 761 char *p = NULL; 762 763 spin_lock_irq(&lo->lo_lock); 764 if (lo->lo_backing_file) 765 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 766 spin_unlock_irq(&lo->lo_lock); 767 768 if (IS_ERR_OR_NULL(p)) 769 ret = PTR_ERR(p); 770 else { 771 ret = strlen(p); 772 memmove(buf, p, ret); 773 buf[ret++] = '\n'; 774 buf[ret] = 0; 775 } 776 777 return ret; 778 } 779 780 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 781 { 782 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 783 } 784 785 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 786 { 787 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 788 } 789 790 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 791 { 792 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 793 794 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 795 } 796 797 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 798 { 799 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 800 801 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 802 } 803 804 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 805 { 806 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 807 808 return sprintf(buf, "%s\n", dio ? "1" : "0"); 809 } 810 811 LOOP_ATTR_RO(backing_file); 812 LOOP_ATTR_RO(offset); 813 LOOP_ATTR_RO(sizelimit); 814 LOOP_ATTR_RO(autoclear); 815 LOOP_ATTR_RO(partscan); 816 LOOP_ATTR_RO(dio); 817 818 static struct attribute *loop_attrs[] = { 819 &loop_attr_backing_file.attr, 820 &loop_attr_offset.attr, 821 &loop_attr_sizelimit.attr, 822 &loop_attr_autoclear.attr, 823 &loop_attr_partscan.attr, 824 &loop_attr_dio.attr, 825 NULL, 826 }; 827 828 static struct attribute_group loop_attribute_group = { 829 .name = "loop", 830 .attrs= loop_attrs, 831 }; 832 833 static void loop_sysfs_init(struct loop_device *lo) 834 { 835 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 836 &loop_attribute_group); 837 } 838 839 static void loop_sysfs_exit(struct loop_device *lo) 840 { 841 if (lo->sysfs_inited) 842 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 843 &loop_attribute_group); 844 } 845 846 static void loop_config_discard(struct loop_device *lo) 847 { 848 struct file *file = lo->lo_backing_file; 849 struct inode *inode = file->f_mapping->host; 850 struct request_queue *q = lo->lo_queue; 851 852 /* 853 * We use punch hole to reclaim the free space used by the 854 * image a.k.a. discard. However we do not support discard if 855 * encryption is enabled, because it may give an attacker 856 * useful information. 857 */ 858 if ((!file->f_op->fallocate) || 859 lo->lo_encrypt_key_size) { 860 q->limits.discard_granularity = 0; 861 q->limits.discard_alignment = 0; 862 blk_queue_max_discard_sectors(q, 0); 863 blk_queue_max_write_zeroes_sectors(q, 0); 864 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 865 return; 866 } 867 868 q->limits.discard_granularity = inode->i_sb->s_blocksize; 869 q->limits.discard_alignment = 0; 870 871 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 872 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 873 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 874 } 875 876 static void loop_unprepare_queue(struct loop_device *lo) 877 { 878 kthread_flush_worker(&lo->worker); 879 kthread_stop(lo->worker_task); 880 } 881 882 static int loop_kthread_worker_fn(void *worker_ptr) 883 { 884 current->flags |= PF_LESS_THROTTLE; 885 return kthread_worker_fn(worker_ptr); 886 } 887 888 static int loop_prepare_queue(struct loop_device *lo) 889 { 890 kthread_init_worker(&lo->worker); 891 lo->worker_task = kthread_run(loop_kthread_worker_fn, 892 &lo->worker, "loop%d", lo->lo_number); 893 if (IS_ERR(lo->worker_task)) 894 return -ENOMEM; 895 set_user_nice(lo->worker_task, MIN_NICE); 896 return 0; 897 } 898 899 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 900 struct block_device *bdev, unsigned int arg) 901 { 902 struct file *file; 903 struct inode *inode; 904 struct address_space *mapping; 905 int lo_flags = 0; 906 int error; 907 loff_t size; 908 bool partscan; 909 910 /* This is safe, since we have a reference from open(). */ 911 __module_get(THIS_MODULE); 912 913 error = -EBADF; 914 file = fget(arg); 915 if (!file) 916 goto out; 917 918 error = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 919 if (error) 920 goto out_putf; 921 922 error = -EBUSY; 923 if (lo->lo_state != Lo_unbound) 924 goto out_unlock; 925 926 error = loop_validate_file(file, bdev); 927 if (error) 928 goto out_unlock; 929 930 mapping = file->f_mapping; 931 inode = mapping->host; 932 933 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 934 !file->f_op->write_iter) 935 lo_flags |= LO_FLAGS_READ_ONLY; 936 937 error = -EFBIG; 938 size = get_loop_size(lo, file); 939 if ((loff_t)(sector_t)size != size) 940 goto out_unlock; 941 error = loop_prepare_queue(lo); 942 if (error) 943 goto out_unlock; 944 945 error = 0; 946 947 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 948 949 lo->use_dio = false; 950 lo->lo_device = bdev; 951 lo->lo_flags = lo_flags; 952 lo->lo_backing_file = file; 953 lo->transfer = NULL; 954 lo->ioctl = NULL; 955 lo->lo_sizelimit = 0; 956 lo->old_gfp_mask = mapping_gfp_mask(mapping); 957 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 958 959 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 960 blk_queue_write_cache(lo->lo_queue, true, false); 961 962 loop_update_dio(lo); 963 set_capacity(lo->lo_disk, size); 964 bd_set_size(bdev, size << 9); 965 loop_sysfs_init(lo); 966 /* let user-space know about the new size */ 967 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 968 969 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 970 block_size(inode->i_bdev) : PAGE_SIZE); 971 972 lo->lo_state = Lo_bound; 973 if (part_shift) 974 lo->lo_flags |= LO_FLAGS_PARTSCAN; 975 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 976 977 /* Grab the block_device to prevent its destruction after we 978 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 979 */ 980 bdgrab(bdev); 981 mutex_unlock(&loop_ctl_mutex); 982 if (partscan) 983 loop_reread_partitions(lo, bdev); 984 return 0; 985 986 out_unlock: 987 mutex_unlock(&loop_ctl_mutex); 988 out_putf: 989 fput(file); 990 out: 991 /* This is safe: open() is still holding a reference. */ 992 module_put(THIS_MODULE); 993 return error; 994 } 995 996 static int 997 loop_release_xfer(struct loop_device *lo) 998 { 999 int err = 0; 1000 struct loop_func_table *xfer = lo->lo_encryption; 1001 1002 if (xfer) { 1003 if (xfer->release) 1004 err = xfer->release(lo); 1005 lo->transfer = NULL; 1006 lo->lo_encryption = NULL; 1007 module_put(xfer->owner); 1008 } 1009 return err; 1010 } 1011 1012 static int 1013 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1014 const struct loop_info64 *i) 1015 { 1016 int err = 0; 1017 1018 if (xfer) { 1019 struct module *owner = xfer->owner; 1020 1021 if (!try_module_get(owner)) 1022 return -EINVAL; 1023 if (xfer->init) 1024 err = xfer->init(lo, i); 1025 if (err) 1026 module_put(owner); 1027 else 1028 lo->lo_encryption = xfer; 1029 } 1030 return err; 1031 } 1032 1033 static int __loop_clr_fd(struct loop_device *lo) 1034 { 1035 struct file *filp = NULL; 1036 gfp_t gfp = lo->old_gfp_mask; 1037 struct block_device *bdev = lo->lo_device; 1038 int err = 0; 1039 1040 mutex_lock(&loop_ctl_mutex); 1041 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1042 err = -ENXIO; 1043 goto out_unlock; 1044 } 1045 1046 filp = lo->lo_backing_file; 1047 if (filp == NULL) { 1048 err = -EINVAL; 1049 goto out_unlock; 1050 } 1051 1052 /* freeze request queue during the transition */ 1053 blk_mq_freeze_queue(lo->lo_queue); 1054 1055 spin_lock_irq(&lo->lo_lock); 1056 lo->lo_backing_file = NULL; 1057 spin_unlock_irq(&lo->lo_lock); 1058 1059 loop_release_xfer(lo); 1060 lo->transfer = NULL; 1061 lo->ioctl = NULL; 1062 lo->lo_device = NULL; 1063 lo->lo_encryption = NULL; 1064 lo->lo_offset = 0; 1065 lo->lo_sizelimit = 0; 1066 lo->lo_encrypt_key_size = 0; 1067 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1068 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1069 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1070 blk_queue_logical_block_size(lo->lo_queue, 512); 1071 blk_queue_physical_block_size(lo->lo_queue, 512); 1072 blk_queue_io_min(lo->lo_queue, 512); 1073 if (bdev) { 1074 bdput(bdev); 1075 invalidate_bdev(bdev); 1076 bdev->bd_inode->i_mapping->wb_err = 0; 1077 } 1078 set_capacity(lo->lo_disk, 0); 1079 loop_sysfs_exit(lo); 1080 if (bdev) { 1081 bd_set_size(bdev, 0); 1082 /* let user-space know about this change */ 1083 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1084 } 1085 mapping_set_gfp_mask(filp->f_mapping, gfp); 1086 lo->lo_state = Lo_unbound; 1087 /* This is safe: open() is still holding a reference. */ 1088 module_put(THIS_MODULE); 1089 blk_mq_unfreeze_queue(lo->lo_queue); 1090 1091 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) { 1092 /* 1093 * bd_mutex has been held already in release path, so don't 1094 * acquire it if this function is called in such case. 1095 * 1096 * If the reread partition isn't from release path, lo_refcnt 1097 * must be at least one and it can only become zero when the 1098 * current holder is released. 1099 */ 1100 if (!atomic_read(&lo->lo_refcnt)) 1101 err = __blkdev_reread_part(bdev); 1102 else 1103 err = blkdev_reread_part(bdev); 1104 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1105 __func__, lo->lo_number, err); 1106 /* Device is gone, no point in returning error */ 1107 err = 0; 1108 } 1109 lo->lo_flags = 0; 1110 if (!part_shift) 1111 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1112 loop_unprepare_queue(lo); 1113 out_unlock: 1114 mutex_unlock(&loop_ctl_mutex); 1115 /* 1116 * Need not hold loop_ctl_mutex to fput backing file. 1117 * Calling fput holding loop_ctl_mutex triggers a circular 1118 * lock dependency possibility warning as fput can take 1119 * bd_mutex which is usually taken before loop_ctl_mutex. 1120 */ 1121 if (filp) 1122 fput(filp); 1123 return err; 1124 } 1125 1126 static int loop_clr_fd(struct loop_device *lo) 1127 { 1128 int err; 1129 1130 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1131 if (err) 1132 return err; 1133 if (lo->lo_state != Lo_bound) { 1134 mutex_unlock(&loop_ctl_mutex); 1135 return -ENXIO; 1136 } 1137 /* 1138 * If we've explicitly asked to tear down the loop device, 1139 * and it has an elevated reference count, set it for auto-teardown when 1140 * the last reference goes away. This stops $!~#$@ udev from 1141 * preventing teardown because it decided that it needs to run blkid on 1142 * the loopback device whenever they appear. xfstests is notorious for 1143 * failing tests because blkid via udev races with a losetup 1144 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1145 * command to fail with EBUSY. 1146 */ 1147 if (atomic_read(&lo->lo_refcnt) > 1) { 1148 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1149 mutex_unlock(&loop_ctl_mutex); 1150 return 0; 1151 } 1152 lo->lo_state = Lo_rundown; 1153 mutex_unlock(&loop_ctl_mutex); 1154 1155 return __loop_clr_fd(lo); 1156 } 1157 1158 static int 1159 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1160 { 1161 int err; 1162 struct loop_func_table *xfer; 1163 kuid_t uid = current_uid(); 1164 struct block_device *bdev; 1165 bool partscan = false; 1166 1167 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1168 if (err) 1169 return err; 1170 if (lo->lo_encrypt_key_size && 1171 !uid_eq(lo->lo_key_owner, uid) && 1172 !capable(CAP_SYS_ADMIN)) { 1173 err = -EPERM; 1174 goto out_unlock; 1175 } 1176 if (lo->lo_state != Lo_bound) { 1177 err = -ENXIO; 1178 goto out_unlock; 1179 } 1180 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) { 1181 err = -EINVAL; 1182 goto out_unlock; 1183 } 1184 1185 /* I/O need to be drained during transfer transition */ 1186 blk_mq_freeze_queue(lo->lo_queue); 1187 1188 err = loop_release_xfer(lo); 1189 if (err) 1190 goto out_unfreeze; 1191 1192 if (info->lo_encrypt_type) { 1193 unsigned int type = info->lo_encrypt_type; 1194 1195 if (type >= MAX_LO_CRYPT) { 1196 err = -EINVAL; 1197 goto out_unfreeze; 1198 } 1199 xfer = xfer_funcs[type]; 1200 if (xfer == NULL) { 1201 err = -EINVAL; 1202 goto out_unfreeze; 1203 } 1204 } else 1205 xfer = NULL; 1206 1207 err = loop_init_xfer(lo, xfer, info); 1208 if (err) 1209 goto out_unfreeze; 1210 1211 if (lo->lo_offset != info->lo_offset || 1212 lo->lo_sizelimit != info->lo_sizelimit) { 1213 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1214 err = -EFBIG; 1215 goto out_unfreeze; 1216 } 1217 } 1218 1219 loop_config_discard(lo); 1220 1221 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1222 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1223 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1224 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1225 1226 if (!xfer) 1227 xfer = &none_funcs; 1228 lo->transfer = xfer->transfer; 1229 lo->ioctl = xfer->ioctl; 1230 1231 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1232 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1233 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1234 1235 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1236 lo->lo_init[0] = info->lo_init[0]; 1237 lo->lo_init[1] = info->lo_init[1]; 1238 if (info->lo_encrypt_key_size) { 1239 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1240 info->lo_encrypt_key_size); 1241 lo->lo_key_owner = uid; 1242 } 1243 1244 /* update dio if lo_offset or transfer is changed */ 1245 __loop_update_dio(lo, lo->use_dio); 1246 1247 out_unfreeze: 1248 blk_mq_unfreeze_queue(lo->lo_queue); 1249 1250 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1251 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1252 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1253 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1254 bdev = lo->lo_device; 1255 partscan = true; 1256 } 1257 out_unlock: 1258 mutex_unlock(&loop_ctl_mutex); 1259 if (partscan) 1260 loop_reread_partitions(lo, bdev); 1261 1262 return err; 1263 } 1264 1265 static int 1266 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1267 { 1268 struct path path; 1269 struct kstat stat; 1270 int ret; 1271 1272 ret = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1273 if (ret) 1274 return ret; 1275 if (lo->lo_state != Lo_bound) { 1276 mutex_unlock(&loop_ctl_mutex); 1277 return -ENXIO; 1278 } 1279 1280 memset(info, 0, sizeof(*info)); 1281 info->lo_number = lo->lo_number; 1282 info->lo_offset = lo->lo_offset; 1283 info->lo_sizelimit = lo->lo_sizelimit; 1284 info->lo_flags = lo->lo_flags; 1285 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1286 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1287 info->lo_encrypt_type = 1288 lo->lo_encryption ? lo->lo_encryption->number : 0; 1289 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1290 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1291 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1292 lo->lo_encrypt_key_size); 1293 } 1294 1295 /* Drop loop_ctl_mutex while we call into the filesystem. */ 1296 path = lo->lo_backing_file->f_path; 1297 path_get(&path); 1298 mutex_unlock(&loop_ctl_mutex); 1299 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1300 if (!ret) { 1301 info->lo_device = huge_encode_dev(stat.dev); 1302 info->lo_inode = stat.ino; 1303 info->lo_rdevice = huge_encode_dev(stat.rdev); 1304 } 1305 path_put(&path); 1306 return ret; 1307 } 1308 1309 static void 1310 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1311 { 1312 memset(info64, 0, sizeof(*info64)); 1313 info64->lo_number = info->lo_number; 1314 info64->lo_device = info->lo_device; 1315 info64->lo_inode = info->lo_inode; 1316 info64->lo_rdevice = info->lo_rdevice; 1317 info64->lo_offset = info->lo_offset; 1318 info64->lo_sizelimit = 0; 1319 info64->lo_encrypt_type = info->lo_encrypt_type; 1320 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1321 info64->lo_flags = info->lo_flags; 1322 info64->lo_init[0] = info->lo_init[0]; 1323 info64->lo_init[1] = info->lo_init[1]; 1324 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1325 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1326 else 1327 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1328 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1329 } 1330 1331 static int 1332 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1333 { 1334 memset(info, 0, sizeof(*info)); 1335 info->lo_number = info64->lo_number; 1336 info->lo_device = info64->lo_device; 1337 info->lo_inode = info64->lo_inode; 1338 info->lo_rdevice = info64->lo_rdevice; 1339 info->lo_offset = info64->lo_offset; 1340 info->lo_encrypt_type = info64->lo_encrypt_type; 1341 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1342 info->lo_flags = info64->lo_flags; 1343 info->lo_init[0] = info64->lo_init[0]; 1344 info->lo_init[1] = info64->lo_init[1]; 1345 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1346 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1347 else 1348 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1349 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1350 1351 /* error in case values were truncated */ 1352 if (info->lo_device != info64->lo_device || 1353 info->lo_rdevice != info64->lo_rdevice || 1354 info->lo_inode != info64->lo_inode || 1355 info->lo_offset != info64->lo_offset) 1356 return -EOVERFLOW; 1357 1358 return 0; 1359 } 1360 1361 static int 1362 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1363 { 1364 struct loop_info info; 1365 struct loop_info64 info64; 1366 1367 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1368 return -EFAULT; 1369 loop_info64_from_old(&info, &info64); 1370 return loop_set_status(lo, &info64); 1371 } 1372 1373 static int 1374 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1375 { 1376 struct loop_info64 info64; 1377 1378 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1379 return -EFAULT; 1380 return loop_set_status(lo, &info64); 1381 } 1382 1383 static int 1384 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1385 struct loop_info info; 1386 struct loop_info64 info64; 1387 int err; 1388 1389 if (!arg) 1390 return -EINVAL; 1391 err = loop_get_status(lo, &info64); 1392 if (!err) 1393 err = loop_info64_to_old(&info64, &info); 1394 if (!err && copy_to_user(arg, &info, sizeof(info))) 1395 err = -EFAULT; 1396 1397 return err; 1398 } 1399 1400 static int 1401 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1402 struct loop_info64 info64; 1403 int err; 1404 1405 if (!arg) 1406 return -EINVAL; 1407 err = loop_get_status(lo, &info64); 1408 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1409 err = -EFAULT; 1410 1411 return err; 1412 } 1413 1414 static int loop_set_capacity(struct loop_device *lo) 1415 { 1416 if (unlikely(lo->lo_state != Lo_bound)) 1417 return -ENXIO; 1418 1419 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1420 } 1421 1422 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1423 { 1424 int error = -ENXIO; 1425 if (lo->lo_state != Lo_bound) 1426 goto out; 1427 1428 __loop_update_dio(lo, !!arg); 1429 if (lo->use_dio == !!arg) 1430 return 0; 1431 error = -EINVAL; 1432 out: 1433 return error; 1434 } 1435 1436 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1437 { 1438 if (lo->lo_state != Lo_bound) 1439 return -ENXIO; 1440 1441 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1442 return -EINVAL; 1443 1444 blk_mq_freeze_queue(lo->lo_queue); 1445 1446 blk_queue_logical_block_size(lo->lo_queue, arg); 1447 blk_queue_physical_block_size(lo->lo_queue, arg); 1448 blk_queue_io_min(lo->lo_queue, arg); 1449 loop_update_dio(lo); 1450 1451 blk_mq_unfreeze_queue(lo->lo_queue); 1452 1453 return 0; 1454 } 1455 1456 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1457 unsigned long arg) 1458 { 1459 int err; 1460 1461 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1462 if (err) 1463 return err; 1464 switch (cmd) { 1465 case LOOP_SET_CAPACITY: 1466 err = loop_set_capacity(lo); 1467 break; 1468 case LOOP_SET_DIRECT_IO: 1469 err = loop_set_dio(lo, arg); 1470 break; 1471 case LOOP_SET_BLOCK_SIZE: 1472 err = loop_set_block_size(lo, arg); 1473 break; 1474 default: 1475 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1476 } 1477 mutex_unlock(&loop_ctl_mutex); 1478 return err; 1479 } 1480 1481 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1482 unsigned int cmd, unsigned long arg) 1483 { 1484 struct loop_device *lo = bdev->bd_disk->private_data; 1485 int err; 1486 1487 switch (cmd) { 1488 case LOOP_SET_FD: 1489 return loop_set_fd(lo, mode, bdev, arg); 1490 case LOOP_CHANGE_FD: 1491 return loop_change_fd(lo, bdev, arg); 1492 case LOOP_CLR_FD: 1493 return loop_clr_fd(lo); 1494 case LOOP_SET_STATUS: 1495 err = -EPERM; 1496 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1497 err = loop_set_status_old(lo, 1498 (struct loop_info __user *)arg); 1499 } 1500 break; 1501 case LOOP_GET_STATUS: 1502 return loop_get_status_old(lo, (struct loop_info __user *) arg); 1503 case LOOP_SET_STATUS64: 1504 err = -EPERM; 1505 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1506 err = loop_set_status64(lo, 1507 (struct loop_info64 __user *) arg); 1508 } 1509 break; 1510 case LOOP_GET_STATUS64: 1511 return loop_get_status64(lo, (struct loop_info64 __user *) arg); 1512 case LOOP_SET_CAPACITY: 1513 case LOOP_SET_DIRECT_IO: 1514 case LOOP_SET_BLOCK_SIZE: 1515 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1516 return -EPERM; 1517 /* Fall through */ 1518 default: 1519 err = lo_simple_ioctl(lo, cmd, arg); 1520 break; 1521 } 1522 1523 return err; 1524 } 1525 1526 #ifdef CONFIG_COMPAT 1527 struct compat_loop_info { 1528 compat_int_t lo_number; /* ioctl r/o */ 1529 compat_dev_t lo_device; /* ioctl r/o */ 1530 compat_ulong_t lo_inode; /* ioctl r/o */ 1531 compat_dev_t lo_rdevice; /* ioctl r/o */ 1532 compat_int_t lo_offset; 1533 compat_int_t lo_encrypt_type; 1534 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1535 compat_int_t lo_flags; /* ioctl r/o */ 1536 char lo_name[LO_NAME_SIZE]; 1537 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1538 compat_ulong_t lo_init[2]; 1539 char reserved[4]; 1540 }; 1541 1542 /* 1543 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1544 * - noinlined to reduce stack space usage in main part of driver 1545 */ 1546 static noinline int 1547 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1548 struct loop_info64 *info64) 1549 { 1550 struct compat_loop_info info; 1551 1552 if (copy_from_user(&info, arg, sizeof(info))) 1553 return -EFAULT; 1554 1555 memset(info64, 0, sizeof(*info64)); 1556 info64->lo_number = info.lo_number; 1557 info64->lo_device = info.lo_device; 1558 info64->lo_inode = info.lo_inode; 1559 info64->lo_rdevice = info.lo_rdevice; 1560 info64->lo_offset = info.lo_offset; 1561 info64->lo_sizelimit = 0; 1562 info64->lo_encrypt_type = info.lo_encrypt_type; 1563 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1564 info64->lo_flags = info.lo_flags; 1565 info64->lo_init[0] = info.lo_init[0]; 1566 info64->lo_init[1] = info.lo_init[1]; 1567 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1568 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1569 else 1570 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1571 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1572 return 0; 1573 } 1574 1575 /* 1576 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1577 * - noinlined to reduce stack space usage in main part of driver 1578 */ 1579 static noinline int 1580 loop_info64_to_compat(const struct loop_info64 *info64, 1581 struct compat_loop_info __user *arg) 1582 { 1583 struct compat_loop_info info; 1584 1585 memset(&info, 0, sizeof(info)); 1586 info.lo_number = info64->lo_number; 1587 info.lo_device = info64->lo_device; 1588 info.lo_inode = info64->lo_inode; 1589 info.lo_rdevice = info64->lo_rdevice; 1590 info.lo_offset = info64->lo_offset; 1591 info.lo_encrypt_type = info64->lo_encrypt_type; 1592 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1593 info.lo_flags = info64->lo_flags; 1594 info.lo_init[0] = info64->lo_init[0]; 1595 info.lo_init[1] = info64->lo_init[1]; 1596 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1597 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1598 else 1599 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1600 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1601 1602 /* error in case values were truncated */ 1603 if (info.lo_device != info64->lo_device || 1604 info.lo_rdevice != info64->lo_rdevice || 1605 info.lo_inode != info64->lo_inode || 1606 info.lo_offset != info64->lo_offset || 1607 info.lo_init[0] != info64->lo_init[0] || 1608 info.lo_init[1] != info64->lo_init[1]) 1609 return -EOVERFLOW; 1610 1611 if (copy_to_user(arg, &info, sizeof(info))) 1612 return -EFAULT; 1613 return 0; 1614 } 1615 1616 static int 1617 loop_set_status_compat(struct loop_device *lo, 1618 const struct compat_loop_info __user *arg) 1619 { 1620 struct loop_info64 info64; 1621 int ret; 1622 1623 ret = loop_info64_from_compat(arg, &info64); 1624 if (ret < 0) 1625 return ret; 1626 return loop_set_status(lo, &info64); 1627 } 1628 1629 static int 1630 loop_get_status_compat(struct loop_device *lo, 1631 struct compat_loop_info __user *arg) 1632 { 1633 struct loop_info64 info64; 1634 int err; 1635 1636 if (!arg) 1637 return -EINVAL; 1638 err = loop_get_status(lo, &info64); 1639 if (!err) 1640 err = loop_info64_to_compat(&info64, arg); 1641 return err; 1642 } 1643 1644 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1645 unsigned int cmd, unsigned long arg) 1646 { 1647 struct loop_device *lo = bdev->bd_disk->private_data; 1648 int err; 1649 1650 switch(cmd) { 1651 case LOOP_SET_STATUS: 1652 err = loop_set_status_compat(lo, 1653 (const struct compat_loop_info __user *)arg); 1654 break; 1655 case LOOP_GET_STATUS: 1656 err = loop_get_status_compat(lo, 1657 (struct compat_loop_info __user *)arg); 1658 break; 1659 case LOOP_SET_CAPACITY: 1660 case LOOP_CLR_FD: 1661 case LOOP_GET_STATUS64: 1662 case LOOP_SET_STATUS64: 1663 arg = (unsigned long) compat_ptr(arg); 1664 /* fall through */ 1665 case LOOP_SET_FD: 1666 case LOOP_CHANGE_FD: 1667 case LOOP_SET_BLOCK_SIZE: 1668 err = lo_ioctl(bdev, mode, cmd, arg); 1669 break; 1670 default: 1671 err = -ENOIOCTLCMD; 1672 break; 1673 } 1674 return err; 1675 } 1676 #endif 1677 1678 static int lo_open(struct block_device *bdev, fmode_t mode) 1679 { 1680 struct loop_device *lo; 1681 int err; 1682 1683 err = mutex_lock_killable(&loop_ctl_mutex); 1684 if (err) 1685 return err; 1686 lo = bdev->bd_disk->private_data; 1687 if (!lo) { 1688 err = -ENXIO; 1689 goto out; 1690 } 1691 1692 atomic_inc(&lo->lo_refcnt); 1693 out: 1694 mutex_unlock(&loop_ctl_mutex); 1695 return err; 1696 } 1697 1698 static void lo_release(struct gendisk *disk, fmode_t mode) 1699 { 1700 struct loop_device *lo; 1701 1702 mutex_lock(&loop_ctl_mutex); 1703 lo = disk->private_data; 1704 if (atomic_dec_return(&lo->lo_refcnt)) 1705 goto out_unlock; 1706 1707 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1708 if (lo->lo_state != Lo_bound) 1709 goto out_unlock; 1710 lo->lo_state = Lo_rundown; 1711 mutex_unlock(&loop_ctl_mutex); 1712 /* 1713 * In autoclear mode, stop the loop thread 1714 * and remove configuration after last close. 1715 */ 1716 __loop_clr_fd(lo); 1717 return; 1718 } else if (lo->lo_state == Lo_bound) { 1719 /* 1720 * Otherwise keep thread (if running) and config, 1721 * but flush possible ongoing bios in thread. 1722 */ 1723 blk_mq_freeze_queue(lo->lo_queue); 1724 blk_mq_unfreeze_queue(lo->lo_queue); 1725 } 1726 1727 out_unlock: 1728 mutex_unlock(&loop_ctl_mutex); 1729 } 1730 1731 static const struct block_device_operations lo_fops = { 1732 .owner = THIS_MODULE, 1733 .open = lo_open, 1734 .release = lo_release, 1735 .ioctl = lo_ioctl, 1736 #ifdef CONFIG_COMPAT 1737 .compat_ioctl = lo_compat_ioctl, 1738 #endif 1739 }; 1740 1741 /* 1742 * And now the modules code and kernel interface. 1743 */ 1744 static int max_loop; 1745 module_param(max_loop, int, 0444); 1746 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1747 module_param(max_part, int, 0444); 1748 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1749 MODULE_LICENSE("GPL"); 1750 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1751 1752 int loop_register_transfer(struct loop_func_table *funcs) 1753 { 1754 unsigned int n = funcs->number; 1755 1756 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1757 return -EINVAL; 1758 xfer_funcs[n] = funcs; 1759 return 0; 1760 } 1761 1762 static int unregister_transfer_cb(int id, void *ptr, void *data) 1763 { 1764 struct loop_device *lo = ptr; 1765 struct loop_func_table *xfer = data; 1766 1767 mutex_lock(&loop_ctl_mutex); 1768 if (lo->lo_encryption == xfer) 1769 loop_release_xfer(lo); 1770 mutex_unlock(&loop_ctl_mutex); 1771 return 0; 1772 } 1773 1774 int loop_unregister_transfer(int number) 1775 { 1776 unsigned int n = number; 1777 struct loop_func_table *xfer; 1778 1779 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1780 return -EINVAL; 1781 1782 xfer_funcs[n] = NULL; 1783 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1784 return 0; 1785 } 1786 1787 EXPORT_SYMBOL(loop_register_transfer); 1788 EXPORT_SYMBOL(loop_unregister_transfer); 1789 1790 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1791 const struct blk_mq_queue_data *bd) 1792 { 1793 struct request *rq = bd->rq; 1794 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1795 struct loop_device *lo = rq->q->queuedata; 1796 1797 blk_mq_start_request(rq); 1798 1799 if (lo->lo_state != Lo_bound) 1800 return BLK_STS_IOERR; 1801 1802 switch (req_op(rq)) { 1803 case REQ_OP_FLUSH: 1804 case REQ_OP_DISCARD: 1805 case REQ_OP_WRITE_ZEROES: 1806 cmd->use_aio = false; 1807 break; 1808 default: 1809 cmd->use_aio = lo->use_dio; 1810 break; 1811 } 1812 1813 /* always use the first bio's css */ 1814 #ifdef CONFIG_BLK_CGROUP 1815 if (cmd->use_aio && rq->bio && rq->bio->bi_css) { 1816 cmd->css = rq->bio->bi_css; 1817 css_get(cmd->css); 1818 } else 1819 #endif 1820 cmd->css = NULL; 1821 kthread_queue_work(&lo->worker, &cmd->work); 1822 1823 return BLK_STS_OK; 1824 } 1825 1826 static void loop_handle_cmd(struct loop_cmd *cmd) 1827 { 1828 struct request *rq = blk_mq_rq_from_pdu(cmd); 1829 const bool write = op_is_write(req_op(rq)); 1830 struct loop_device *lo = rq->q->queuedata; 1831 int ret = 0; 1832 1833 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1834 ret = -EIO; 1835 goto failed; 1836 } 1837 1838 ret = do_req_filebacked(lo, rq); 1839 failed: 1840 /* complete non-aio request */ 1841 if (!cmd->use_aio || ret) { 1842 cmd->ret = ret ? -EIO : 0; 1843 blk_mq_complete_request(rq); 1844 } 1845 } 1846 1847 static void loop_queue_work(struct kthread_work *work) 1848 { 1849 struct loop_cmd *cmd = 1850 container_of(work, struct loop_cmd, work); 1851 1852 loop_handle_cmd(cmd); 1853 } 1854 1855 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1856 unsigned int hctx_idx, unsigned int numa_node) 1857 { 1858 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1859 1860 kthread_init_work(&cmd->work, loop_queue_work); 1861 return 0; 1862 } 1863 1864 static const struct blk_mq_ops loop_mq_ops = { 1865 .queue_rq = loop_queue_rq, 1866 .init_request = loop_init_request, 1867 .complete = lo_complete_rq, 1868 }; 1869 1870 static int loop_add(struct loop_device **l, int i) 1871 { 1872 struct loop_device *lo; 1873 struct gendisk *disk; 1874 int err; 1875 1876 err = -ENOMEM; 1877 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1878 if (!lo) 1879 goto out; 1880 1881 lo->lo_state = Lo_unbound; 1882 1883 /* allocate id, if @id >= 0, we're requesting that specific id */ 1884 if (i >= 0) { 1885 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1886 if (err == -ENOSPC) 1887 err = -EEXIST; 1888 } else { 1889 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1890 } 1891 if (err < 0) 1892 goto out_free_dev; 1893 i = err; 1894 1895 err = -ENOMEM; 1896 lo->tag_set.ops = &loop_mq_ops; 1897 lo->tag_set.nr_hw_queues = 1; 1898 lo->tag_set.queue_depth = 128; 1899 lo->tag_set.numa_node = NUMA_NO_NODE; 1900 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1901 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1902 lo->tag_set.driver_data = lo; 1903 1904 err = blk_mq_alloc_tag_set(&lo->tag_set); 1905 if (err) 1906 goto out_free_idr; 1907 1908 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1909 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1910 err = PTR_ERR(lo->lo_queue); 1911 goto out_cleanup_tags; 1912 } 1913 lo->lo_queue->queuedata = lo; 1914 1915 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1916 1917 /* 1918 * By default, we do buffer IO, so it doesn't make sense to enable 1919 * merge because the I/O submitted to backing file is handled page by 1920 * page. For directio mode, merge does help to dispatch bigger request 1921 * to underlayer disk. We will enable merge once directio is enabled. 1922 */ 1923 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1924 1925 err = -ENOMEM; 1926 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1927 if (!disk) 1928 goto out_free_queue; 1929 1930 /* 1931 * Disable partition scanning by default. The in-kernel partition 1932 * scanning can be requested individually per-device during its 1933 * setup. Userspace can always add and remove partitions from all 1934 * devices. The needed partition minors are allocated from the 1935 * extended minor space, the main loop device numbers will continue 1936 * to match the loop minors, regardless of the number of partitions 1937 * used. 1938 * 1939 * If max_part is given, partition scanning is globally enabled for 1940 * all loop devices. The minors for the main loop devices will be 1941 * multiples of max_part. 1942 * 1943 * Note: Global-for-all-devices, set-only-at-init, read-only module 1944 * parameteters like 'max_loop' and 'max_part' make things needlessly 1945 * complicated, are too static, inflexible and may surprise 1946 * userspace tools. Parameters like this in general should be avoided. 1947 */ 1948 if (!part_shift) 1949 disk->flags |= GENHD_FL_NO_PART_SCAN; 1950 disk->flags |= GENHD_FL_EXT_DEVT; 1951 atomic_set(&lo->lo_refcnt, 0); 1952 lo->lo_number = i; 1953 spin_lock_init(&lo->lo_lock); 1954 disk->major = LOOP_MAJOR; 1955 disk->first_minor = i << part_shift; 1956 disk->fops = &lo_fops; 1957 disk->private_data = lo; 1958 disk->queue = lo->lo_queue; 1959 sprintf(disk->disk_name, "loop%d", i); 1960 add_disk(disk); 1961 *l = lo; 1962 return lo->lo_number; 1963 1964 out_free_queue: 1965 blk_cleanup_queue(lo->lo_queue); 1966 out_cleanup_tags: 1967 blk_mq_free_tag_set(&lo->tag_set); 1968 out_free_idr: 1969 idr_remove(&loop_index_idr, i); 1970 out_free_dev: 1971 kfree(lo); 1972 out: 1973 return err; 1974 } 1975 1976 static void loop_remove(struct loop_device *lo) 1977 { 1978 del_gendisk(lo->lo_disk); 1979 blk_cleanup_queue(lo->lo_queue); 1980 blk_mq_free_tag_set(&lo->tag_set); 1981 put_disk(lo->lo_disk); 1982 kfree(lo); 1983 } 1984 1985 static int find_free_cb(int id, void *ptr, void *data) 1986 { 1987 struct loop_device *lo = ptr; 1988 struct loop_device **l = data; 1989 1990 if (lo->lo_state == Lo_unbound) { 1991 *l = lo; 1992 return 1; 1993 } 1994 return 0; 1995 } 1996 1997 static int loop_lookup(struct loop_device **l, int i) 1998 { 1999 struct loop_device *lo; 2000 int ret = -ENODEV; 2001 2002 if (i < 0) { 2003 int err; 2004 2005 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 2006 if (err == 1) { 2007 *l = lo; 2008 ret = lo->lo_number; 2009 } 2010 goto out; 2011 } 2012 2013 /* lookup and return a specific i */ 2014 lo = idr_find(&loop_index_idr, i); 2015 if (lo) { 2016 *l = lo; 2017 ret = lo->lo_number; 2018 } 2019 out: 2020 return ret; 2021 } 2022 2023 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 2024 { 2025 struct loop_device *lo; 2026 struct kobject *kobj; 2027 int err; 2028 2029 mutex_lock(&loop_ctl_mutex); 2030 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 2031 if (err < 0) 2032 err = loop_add(&lo, MINOR(dev) >> part_shift); 2033 if (err < 0) 2034 kobj = NULL; 2035 else 2036 kobj = get_disk_and_module(lo->lo_disk); 2037 mutex_unlock(&loop_ctl_mutex); 2038 2039 *part = 0; 2040 return kobj; 2041 } 2042 2043 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2044 unsigned long parm) 2045 { 2046 struct loop_device *lo; 2047 int ret; 2048 2049 ret = mutex_lock_killable(&loop_ctl_mutex); 2050 if (ret) 2051 return ret; 2052 2053 ret = -ENOSYS; 2054 switch (cmd) { 2055 case LOOP_CTL_ADD: 2056 ret = loop_lookup(&lo, parm); 2057 if (ret >= 0) { 2058 ret = -EEXIST; 2059 break; 2060 } 2061 ret = loop_add(&lo, parm); 2062 break; 2063 case LOOP_CTL_REMOVE: 2064 ret = loop_lookup(&lo, parm); 2065 if (ret < 0) 2066 break; 2067 if (lo->lo_state != Lo_unbound) { 2068 ret = -EBUSY; 2069 mutex_unlock(&loop_ctl_mutex); 2070 break; 2071 } 2072 if (atomic_read(&lo->lo_refcnt) > 0) { 2073 ret = -EBUSY; 2074 mutex_unlock(&loop_ctl_mutex); 2075 break; 2076 } 2077 lo->lo_disk->private_data = NULL; 2078 idr_remove(&loop_index_idr, lo->lo_number); 2079 loop_remove(lo); 2080 break; 2081 case LOOP_CTL_GET_FREE: 2082 ret = loop_lookup(&lo, -1); 2083 if (ret >= 0) 2084 break; 2085 ret = loop_add(&lo, -1); 2086 } 2087 mutex_unlock(&loop_ctl_mutex); 2088 2089 return ret; 2090 } 2091 2092 static const struct file_operations loop_ctl_fops = { 2093 .open = nonseekable_open, 2094 .unlocked_ioctl = loop_control_ioctl, 2095 .compat_ioctl = loop_control_ioctl, 2096 .owner = THIS_MODULE, 2097 .llseek = noop_llseek, 2098 }; 2099 2100 static struct miscdevice loop_misc = { 2101 .minor = LOOP_CTRL_MINOR, 2102 .name = "loop-control", 2103 .fops = &loop_ctl_fops, 2104 }; 2105 2106 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2107 MODULE_ALIAS("devname:loop-control"); 2108 2109 static int __init loop_init(void) 2110 { 2111 int i, nr; 2112 unsigned long range; 2113 struct loop_device *lo; 2114 int err; 2115 2116 part_shift = 0; 2117 if (max_part > 0) { 2118 part_shift = fls(max_part); 2119 2120 /* 2121 * Adjust max_part according to part_shift as it is exported 2122 * to user space so that user can decide correct minor number 2123 * if [s]he want to create more devices. 2124 * 2125 * Note that -1 is required because partition 0 is reserved 2126 * for the whole disk. 2127 */ 2128 max_part = (1UL << part_shift) - 1; 2129 } 2130 2131 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2132 err = -EINVAL; 2133 goto err_out; 2134 } 2135 2136 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2137 err = -EINVAL; 2138 goto err_out; 2139 } 2140 2141 /* 2142 * If max_loop is specified, create that many devices upfront. 2143 * This also becomes a hard limit. If max_loop is not specified, 2144 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2145 * init time. Loop devices can be requested on-demand with the 2146 * /dev/loop-control interface, or be instantiated by accessing 2147 * a 'dead' device node. 2148 */ 2149 if (max_loop) { 2150 nr = max_loop; 2151 range = max_loop << part_shift; 2152 } else { 2153 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2154 range = 1UL << MINORBITS; 2155 } 2156 2157 err = misc_register(&loop_misc); 2158 if (err < 0) 2159 goto err_out; 2160 2161 2162 if (register_blkdev(LOOP_MAJOR, "loop")) { 2163 err = -EIO; 2164 goto misc_out; 2165 } 2166 2167 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2168 THIS_MODULE, loop_probe, NULL, NULL); 2169 2170 /* pre-create number of devices given by config or max_loop */ 2171 mutex_lock(&loop_ctl_mutex); 2172 for (i = 0; i < nr; i++) 2173 loop_add(&lo, i); 2174 mutex_unlock(&loop_ctl_mutex); 2175 2176 printk(KERN_INFO "loop: module loaded\n"); 2177 return 0; 2178 2179 misc_out: 2180 misc_deregister(&loop_misc); 2181 err_out: 2182 return err; 2183 } 2184 2185 static int loop_exit_cb(int id, void *ptr, void *data) 2186 { 2187 struct loop_device *lo = ptr; 2188 2189 loop_remove(lo); 2190 return 0; 2191 } 2192 2193 static void __exit loop_exit(void) 2194 { 2195 unsigned long range; 2196 2197 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2198 2199 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2200 idr_destroy(&loop_index_idr); 2201 2202 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2203 unregister_blkdev(LOOP_MAJOR, "loop"); 2204 2205 misc_deregister(&loop_misc); 2206 } 2207 2208 module_init(loop_init); 2209 module_exit(loop_exit); 2210 2211 #ifndef MODULE 2212 static int __init max_loop_setup(char *str) 2213 { 2214 max_loop = simple_strtol(str, NULL, 0); 2215 return 1; 2216 } 2217 2218 __setup("max_loop=", max_loop_setup); 2219 #endif 2220