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_ctl_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, 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, READ, &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, READ, &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 req_iterator rq_iter; 515 struct bio_vec *bvec; 516 struct request *rq = blk_mq_rq_from_pdu(cmd); 517 struct bio *bio = rq->bio; 518 struct file *file = lo->lo_backing_file; 519 struct bio_vec tmp; 520 unsigned int offset; 521 int nr_bvec = 0; 522 int ret; 523 524 rq_for_each_bvec(tmp, rq, rq_iter) 525 nr_bvec++; 526 527 if (rq->bio != rq->biotail) { 528 529 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 530 GFP_NOIO); 531 if (!bvec) 532 return -EIO; 533 cmd->bvec = bvec; 534 535 /* 536 * The bios of the request may be started from the middle of 537 * the 'bvec' because of bio splitting, so we can't directly 538 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 539 * API will take care of all details for us. 540 */ 541 rq_for_each_bvec(tmp, rq, rq_iter) { 542 *bvec = tmp; 543 bvec++; 544 } 545 bvec = cmd->bvec; 546 offset = 0; 547 } else { 548 /* 549 * Same here, this bio may be started from the middle of the 550 * 'bvec' because of bio splitting, so offset from the bvec 551 * must be passed to iov iterator 552 */ 553 offset = bio->bi_iter.bi_bvec_done; 554 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 555 } 556 atomic_set(&cmd->ref, 2); 557 558 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 559 iter.iov_offset = offset; 560 561 cmd->iocb.ki_pos = pos; 562 cmd->iocb.ki_filp = file; 563 cmd->iocb.ki_complete = lo_rw_aio_complete; 564 cmd->iocb.ki_flags = IOCB_DIRECT; 565 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 566 if (cmd->css) 567 kthread_associate_blkcg(cmd->css); 568 569 if (rw == WRITE) 570 ret = call_write_iter(file, &cmd->iocb, &iter); 571 else 572 ret = call_read_iter(file, &cmd->iocb, &iter); 573 574 lo_rw_aio_do_completion(cmd); 575 kthread_associate_blkcg(NULL); 576 577 if (ret != -EIOCBQUEUED) 578 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 579 return 0; 580 } 581 582 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 583 { 584 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 585 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 586 587 /* 588 * lo_write_simple and lo_read_simple should have been covered 589 * by io submit style function like lo_rw_aio(), one blocker 590 * is that lo_read_simple() need to call flush_dcache_page after 591 * the page is written from kernel, and it isn't easy to handle 592 * this in io submit style function which submits all segments 593 * of the req at one time. And direct read IO doesn't need to 594 * run flush_dcache_page(). 595 */ 596 switch (req_op(rq)) { 597 case REQ_OP_FLUSH: 598 return lo_req_flush(lo, rq); 599 case REQ_OP_DISCARD: 600 case REQ_OP_WRITE_ZEROES: 601 return lo_discard(lo, rq, pos); 602 case REQ_OP_WRITE: 603 if (lo->transfer) 604 return lo_write_transfer(lo, rq, pos); 605 else if (cmd->use_aio) 606 return lo_rw_aio(lo, cmd, pos, WRITE); 607 else 608 return lo_write_simple(lo, rq, pos); 609 case REQ_OP_READ: 610 if (lo->transfer) 611 return lo_read_transfer(lo, rq, pos); 612 else if (cmd->use_aio) 613 return lo_rw_aio(lo, cmd, pos, READ); 614 else 615 return lo_read_simple(lo, rq, pos); 616 default: 617 WARN_ON_ONCE(1); 618 return -EIO; 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 = NULL, *old_file; 681 int error; 682 bool partscan; 683 684 error = mutex_lock_killable(&loop_ctl_mutex); 685 if (error) 686 return error; 687 error = -ENXIO; 688 if (lo->lo_state != Lo_bound) 689 goto out_err; 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_err; 695 696 error = -EBADF; 697 file = fget(arg); 698 if (!file) 699 goto out_err; 700 701 error = loop_validate_file(file, bdev); 702 if (error) 703 goto out_err; 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_err; 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 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 723 mutex_unlock(&loop_ctl_mutex); 724 /* 725 * We must drop file reference outside of loop_ctl_mutex as dropping 726 * the file ref can take bd_mutex which creates circular locking 727 * dependency. 728 */ 729 fput(old_file); 730 if (partscan) 731 loop_reread_partitions(lo, bdev); 732 return 0; 733 734 out_err: 735 mutex_unlock(&loop_ctl_mutex); 736 if (file) 737 fput(file); 738 return error; 739 } 740 741 /* loop sysfs attributes */ 742 743 static ssize_t loop_attr_show(struct device *dev, char *page, 744 ssize_t (*callback)(struct loop_device *, char *)) 745 { 746 struct gendisk *disk = dev_to_disk(dev); 747 struct loop_device *lo = disk->private_data; 748 749 return callback(lo, page); 750 } 751 752 #define LOOP_ATTR_RO(_name) \ 753 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 754 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 755 struct device_attribute *attr, char *b) \ 756 { \ 757 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 758 } \ 759 static struct device_attribute loop_attr_##_name = \ 760 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 761 762 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 763 { 764 ssize_t ret; 765 char *p = NULL; 766 767 spin_lock_irq(&lo->lo_lock); 768 if (lo->lo_backing_file) 769 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 770 spin_unlock_irq(&lo->lo_lock); 771 772 if (IS_ERR_OR_NULL(p)) 773 ret = PTR_ERR(p); 774 else { 775 ret = strlen(p); 776 memmove(buf, p, ret); 777 buf[ret++] = '\n'; 778 buf[ret] = 0; 779 } 780 781 return ret; 782 } 783 784 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 785 { 786 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 787 } 788 789 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 790 { 791 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 792 } 793 794 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 795 { 796 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 797 798 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 799 } 800 801 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 802 { 803 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 804 805 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 806 } 807 808 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 809 { 810 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 811 812 return sprintf(buf, "%s\n", dio ? "1" : "0"); 813 } 814 815 LOOP_ATTR_RO(backing_file); 816 LOOP_ATTR_RO(offset); 817 LOOP_ATTR_RO(sizelimit); 818 LOOP_ATTR_RO(autoclear); 819 LOOP_ATTR_RO(partscan); 820 LOOP_ATTR_RO(dio); 821 822 static struct attribute *loop_attrs[] = { 823 &loop_attr_backing_file.attr, 824 &loop_attr_offset.attr, 825 &loop_attr_sizelimit.attr, 826 &loop_attr_autoclear.attr, 827 &loop_attr_partscan.attr, 828 &loop_attr_dio.attr, 829 NULL, 830 }; 831 832 static struct attribute_group loop_attribute_group = { 833 .name = "loop", 834 .attrs= loop_attrs, 835 }; 836 837 static void loop_sysfs_init(struct loop_device *lo) 838 { 839 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 840 &loop_attribute_group); 841 } 842 843 static void loop_sysfs_exit(struct loop_device *lo) 844 { 845 if (lo->sysfs_inited) 846 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 847 &loop_attribute_group); 848 } 849 850 static void loop_config_discard(struct loop_device *lo) 851 { 852 struct file *file = lo->lo_backing_file; 853 struct inode *inode = file->f_mapping->host; 854 struct request_queue *q = lo->lo_queue; 855 856 /* 857 * We use punch hole to reclaim the free space used by the 858 * image a.k.a. discard. However we do not support discard if 859 * encryption is enabled, because it may give an attacker 860 * useful information. 861 */ 862 if ((!file->f_op->fallocate) || 863 lo->lo_encrypt_key_size) { 864 q->limits.discard_granularity = 0; 865 q->limits.discard_alignment = 0; 866 blk_queue_max_discard_sectors(q, 0); 867 blk_queue_max_write_zeroes_sectors(q, 0); 868 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 869 return; 870 } 871 872 q->limits.discard_granularity = inode->i_sb->s_blocksize; 873 q->limits.discard_alignment = 0; 874 875 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 876 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 877 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 878 } 879 880 static void loop_unprepare_queue(struct loop_device *lo) 881 { 882 kthread_flush_worker(&lo->worker); 883 kthread_stop(lo->worker_task); 884 } 885 886 static int loop_kthread_worker_fn(void *worker_ptr) 887 { 888 current->flags |= PF_LESS_THROTTLE; 889 return kthread_worker_fn(worker_ptr); 890 } 891 892 static int loop_prepare_queue(struct loop_device *lo) 893 { 894 kthread_init_worker(&lo->worker); 895 lo->worker_task = kthread_run(loop_kthread_worker_fn, 896 &lo->worker, "loop%d", lo->lo_number); 897 if (IS_ERR(lo->worker_task)) 898 return -ENOMEM; 899 set_user_nice(lo->worker_task, MIN_NICE); 900 return 0; 901 } 902 903 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 904 struct block_device *bdev, unsigned int arg) 905 { 906 struct file *file; 907 struct inode *inode; 908 struct address_space *mapping; 909 int lo_flags = 0; 910 int error; 911 loff_t size; 912 bool partscan; 913 914 /* This is safe, since we have a reference from open(). */ 915 __module_get(THIS_MODULE); 916 917 error = -EBADF; 918 file = fget(arg); 919 if (!file) 920 goto out; 921 922 error = mutex_lock_killable(&loop_ctl_mutex); 923 if (error) 924 goto out_putf; 925 926 error = -EBUSY; 927 if (lo->lo_state != Lo_unbound) 928 goto out_unlock; 929 930 error = loop_validate_file(file, bdev); 931 if (error) 932 goto out_unlock; 933 934 mapping = file->f_mapping; 935 inode = mapping->host; 936 937 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 938 !file->f_op->write_iter) 939 lo_flags |= LO_FLAGS_READ_ONLY; 940 941 error = -EFBIG; 942 size = get_loop_size(lo, file); 943 if ((loff_t)(sector_t)size != size) 944 goto out_unlock; 945 error = loop_prepare_queue(lo); 946 if (error) 947 goto out_unlock; 948 949 error = 0; 950 951 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 952 953 lo->use_dio = false; 954 lo->lo_device = bdev; 955 lo->lo_flags = lo_flags; 956 lo->lo_backing_file = file; 957 lo->transfer = NULL; 958 lo->ioctl = NULL; 959 lo->lo_sizelimit = 0; 960 lo->old_gfp_mask = mapping_gfp_mask(mapping); 961 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 962 963 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 964 blk_queue_write_cache(lo->lo_queue, true, false); 965 966 loop_update_dio(lo); 967 set_capacity(lo->lo_disk, size); 968 bd_set_size(bdev, size << 9); 969 loop_sysfs_init(lo); 970 /* let user-space know about the new size */ 971 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 972 973 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 974 block_size(inode->i_bdev) : PAGE_SIZE); 975 976 lo->lo_state = Lo_bound; 977 if (part_shift) 978 lo->lo_flags |= LO_FLAGS_PARTSCAN; 979 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 980 981 /* Grab the block_device to prevent its destruction after we 982 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 983 */ 984 bdgrab(bdev); 985 mutex_unlock(&loop_ctl_mutex); 986 if (partscan) 987 loop_reread_partitions(lo, bdev); 988 return 0; 989 990 out_unlock: 991 mutex_unlock(&loop_ctl_mutex); 992 out_putf: 993 fput(file); 994 out: 995 /* This is safe: open() is still holding a reference. */ 996 module_put(THIS_MODULE); 997 return error; 998 } 999 1000 static int 1001 loop_release_xfer(struct loop_device *lo) 1002 { 1003 int err = 0; 1004 struct loop_func_table *xfer = lo->lo_encryption; 1005 1006 if (xfer) { 1007 if (xfer->release) 1008 err = xfer->release(lo); 1009 lo->transfer = NULL; 1010 lo->lo_encryption = NULL; 1011 module_put(xfer->owner); 1012 } 1013 return err; 1014 } 1015 1016 static int 1017 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1018 const struct loop_info64 *i) 1019 { 1020 int err = 0; 1021 1022 if (xfer) { 1023 struct module *owner = xfer->owner; 1024 1025 if (!try_module_get(owner)) 1026 return -EINVAL; 1027 if (xfer->init) 1028 err = xfer->init(lo, i); 1029 if (err) 1030 module_put(owner); 1031 else 1032 lo->lo_encryption = xfer; 1033 } 1034 return err; 1035 } 1036 1037 static int __loop_clr_fd(struct loop_device *lo, bool release) 1038 { 1039 struct file *filp = NULL; 1040 gfp_t gfp = lo->old_gfp_mask; 1041 struct block_device *bdev = lo->lo_device; 1042 int err = 0; 1043 bool partscan = false; 1044 int lo_number; 1045 1046 mutex_lock(&loop_ctl_mutex); 1047 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1048 err = -ENXIO; 1049 goto out_unlock; 1050 } 1051 1052 filp = lo->lo_backing_file; 1053 if (filp == NULL) { 1054 err = -EINVAL; 1055 goto out_unlock; 1056 } 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_backing_file = NULL; 1063 spin_unlock_irq(&lo->lo_lock); 1064 1065 loop_release_xfer(lo); 1066 lo->transfer = NULL; 1067 lo->ioctl = NULL; 1068 lo->lo_device = NULL; 1069 lo->lo_encryption = NULL; 1070 lo->lo_offset = 0; 1071 lo->lo_sizelimit = 0; 1072 lo->lo_encrypt_key_size = 0; 1073 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1074 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1075 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1076 blk_queue_logical_block_size(lo->lo_queue, 512); 1077 blk_queue_physical_block_size(lo->lo_queue, 512); 1078 blk_queue_io_min(lo->lo_queue, 512); 1079 if (bdev) { 1080 bdput(bdev); 1081 invalidate_bdev(bdev); 1082 bdev->bd_inode->i_mapping->wb_err = 0; 1083 } 1084 set_capacity(lo->lo_disk, 0); 1085 loop_sysfs_exit(lo); 1086 if (bdev) { 1087 bd_set_size(bdev, 0); 1088 /* let user-space know about this change */ 1089 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1090 } 1091 mapping_set_gfp_mask(filp->f_mapping, gfp); 1092 lo->lo_state = Lo_unbound; 1093 /* This is safe: open() is still holding a reference. */ 1094 module_put(THIS_MODULE); 1095 blk_mq_unfreeze_queue(lo->lo_queue); 1096 1097 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1098 lo_number = lo->lo_number; 1099 lo->lo_flags = 0; 1100 if (!part_shift) 1101 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1102 loop_unprepare_queue(lo); 1103 out_unlock: 1104 mutex_unlock(&loop_ctl_mutex); 1105 if (partscan) { 1106 /* 1107 * bd_mutex has been held already in release path, so don't 1108 * acquire it if this function is called in such case. 1109 * 1110 * If the reread partition isn't from release path, lo_refcnt 1111 * must be at least one and it can only become zero when the 1112 * current holder is released. 1113 */ 1114 if (release) 1115 err = __blkdev_reread_part(bdev); 1116 else 1117 err = blkdev_reread_part(bdev); 1118 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1119 __func__, lo_number, err); 1120 /* Device is gone, no point in returning error */ 1121 err = 0; 1122 } 1123 /* 1124 * Need not hold loop_ctl_mutex to fput backing file. 1125 * Calling fput holding loop_ctl_mutex triggers a circular 1126 * lock dependency possibility warning as fput can take 1127 * bd_mutex which is usually taken before loop_ctl_mutex. 1128 */ 1129 if (filp) 1130 fput(filp); 1131 return err; 1132 } 1133 1134 static int loop_clr_fd(struct loop_device *lo) 1135 { 1136 int err; 1137 1138 err = mutex_lock_killable(&loop_ctl_mutex); 1139 if (err) 1140 return err; 1141 if (lo->lo_state != Lo_bound) { 1142 mutex_unlock(&loop_ctl_mutex); 1143 return -ENXIO; 1144 } 1145 /* 1146 * If we've explicitly asked to tear down the loop device, 1147 * and it has an elevated reference count, set it for auto-teardown when 1148 * the last reference goes away. This stops $!~#$@ udev from 1149 * preventing teardown because it decided that it needs to run blkid on 1150 * the loopback device whenever they appear. xfstests is notorious for 1151 * failing tests because blkid via udev races with a losetup 1152 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1153 * command to fail with EBUSY. 1154 */ 1155 if (atomic_read(&lo->lo_refcnt) > 1) { 1156 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1157 mutex_unlock(&loop_ctl_mutex); 1158 return 0; 1159 } 1160 lo->lo_state = Lo_rundown; 1161 mutex_unlock(&loop_ctl_mutex); 1162 1163 return __loop_clr_fd(lo, false); 1164 } 1165 1166 static int 1167 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1168 { 1169 int err; 1170 struct loop_func_table *xfer; 1171 kuid_t uid = current_uid(); 1172 struct block_device *bdev; 1173 bool partscan = false; 1174 1175 err = mutex_lock_killable(&loop_ctl_mutex); 1176 if (err) 1177 return err; 1178 if (lo->lo_encrypt_key_size && 1179 !uid_eq(lo->lo_key_owner, uid) && 1180 !capable(CAP_SYS_ADMIN)) { 1181 err = -EPERM; 1182 goto out_unlock; 1183 } 1184 if (lo->lo_state != Lo_bound) { 1185 err = -ENXIO; 1186 goto out_unlock; 1187 } 1188 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) { 1189 err = -EINVAL; 1190 goto out_unlock; 1191 } 1192 1193 if (lo->lo_offset != info->lo_offset || 1194 lo->lo_sizelimit != info->lo_sizelimit) { 1195 sync_blockdev(lo->lo_device); 1196 kill_bdev(lo->lo_device); 1197 } 1198 1199 /* I/O need to be drained during transfer transition */ 1200 blk_mq_freeze_queue(lo->lo_queue); 1201 1202 err = loop_release_xfer(lo); 1203 if (err) 1204 goto out_unfreeze; 1205 1206 if (info->lo_encrypt_type) { 1207 unsigned int type = info->lo_encrypt_type; 1208 1209 if (type >= MAX_LO_CRYPT) { 1210 err = -EINVAL; 1211 goto out_unfreeze; 1212 } 1213 xfer = xfer_funcs[type]; 1214 if (xfer == NULL) { 1215 err = -EINVAL; 1216 goto out_unfreeze; 1217 } 1218 } else 1219 xfer = NULL; 1220 1221 err = loop_init_xfer(lo, xfer, info); 1222 if (err) 1223 goto out_unfreeze; 1224 1225 if (lo->lo_offset != info->lo_offset || 1226 lo->lo_sizelimit != info->lo_sizelimit) { 1227 /* kill_bdev should have truncated all the pages */ 1228 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1229 err = -EAGAIN; 1230 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1231 __func__, lo->lo_number, lo->lo_file_name, 1232 lo->lo_device->bd_inode->i_mapping->nrpages); 1233 goto out_unfreeze; 1234 } 1235 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1236 err = -EFBIG; 1237 goto out_unfreeze; 1238 } 1239 } 1240 1241 loop_config_discard(lo); 1242 1243 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1244 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1245 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1246 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1247 1248 if (!xfer) 1249 xfer = &none_funcs; 1250 lo->transfer = xfer->transfer; 1251 lo->ioctl = xfer->ioctl; 1252 1253 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1254 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1255 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1256 1257 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1258 lo->lo_init[0] = info->lo_init[0]; 1259 lo->lo_init[1] = info->lo_init[1]; 1260 if (info->lo_encrypt_key_size) { 1261 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1262 info->lo_encrypt_key_size); 1263 lo->lo_key_owner = uid; 1264 } 1265 1266 /* update dio if lo_offset or transfer is changed */ 1267 __loop_update_dio(lo, lo->use_dio); 1268 1269 out_unfreeze: 1270 blk_mq_unfreeze_queue(lo->lo_queue); 1271 1272 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1273 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1274 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1275 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1276 bdev = lo->lo_device; 1277 partscan = true; 1278 } 1279 out_unlock: 1280 mutex_unlock(&loop_ctl_mutex); 1281 if (partscan) 1282 loop_reread_partitions(lo, bdev); 1283 1284 return err; 1285 } 1286 1287 static int 1288 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1289 { 1290 struct path path; 1291 struct kstat stat; 1292 int ret; 1293 1294 ret = mutex_lock_killable(&loop_ctl_mutex); 1295 if (ret) 1296 return ret; 1297 if (lo->lo_state != Lo_bound) { 1298 mutex_unlock(&loop_ctl_mutex); 1299 return -ENXIO; 1300 } 1301 1302 memset(info, 0, sizeof(*info)); 1303 info->lo_number = lo->lo_number; 1304 info->lo_offset = lo->lo_offset; 1305 info->lo_sizelimit = lo->lo_sizelimit; 1306 info->lo_flags = lo->lo_flags; 1307 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1308 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1309 info->lo_encrypt_type = 1310 lo->lo_encryption ? lo->lo_encryption->number : 0; 1311 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1312 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1313 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1314 lo->lo_encrypt_key_size); 1315 } 1316 1317 /* Drop loop_ctl_mutex while we call into the filesystem. */ 1318 path = lo->lo_backing_file->f_path; 1319 path_get(&path); 1320 mutex_unlock(&loop_ctl_mutex); 1321 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1322 if (!ret) { 1323 info->lo_device = huge_encode_dev(stat.dev); 1324 info->lo_inode = stat.ino; 1325 info->lo_rdevice = huge_encode_dev(stat.rdev); 1326 } 1327 path_put(&path); 1328 return ret; 1329 } 1330 1331 static void 1332 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1333 { 1334 memset(info64, 0, sizeof(*info64)); 1335 info64->lo_number = info->lo_number; 1336 info64->lo_device = info->lo_device; 1337 info64->lo_inode = info->lo_inode; 1338 info64->lo_rdevice = info->lo_rdevice; 1339 info64->lo_offset = info->lo_offset; 1340 info64->lo_sizelimit = 0; 1341 info64->lo_encrypt_type = info->lo_encrypt_type; 1342 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1343 info64->lo_flags = info->lo_flags; 1344 info64->lo_init[0] = info->lo_init[0]; 1345 info64->lo_init[1] = info->lo_init[1]; 1346 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1347 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1348 else 1349 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1350 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1351 } 1352 1353 static int 1354 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1355 { 1356 memset(info, 0, sizeof(*info)); 1357 info->lo_number = info64->lo_number; 1358 info->lo_device = info64->lo_device; 1359 info->lo_inode = info64->lo_inode; 1360 info->lo_rdevice = info64->lo_rdevice; 1361 info->lo_offset = info64->lo_offset; 1362 info->lo_encrypt_type = info64->lo_encrypt_type; 1363 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1364 info->lo_flags = info64->lo_flags; 1365 info->lo_init[0] = info64->lo_init[0]; 1366 info->lo_init[1] = info64->lo_init[1]; 1367 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1368 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1369 else 1370 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1371 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1372 1373 /* error in case values were truncated */ 1374 if (info->lo_device != info64->lo_device || 1375 info->lo_rdevice != info64->lo_rdevice || 1376 info->lo_inode != info64->lo_inode || 1377 info->lo_offset != info64->lo_offset) 1378 return -EOVERFLOW; 1379 1380 return 0; 1381 } 1382 1383 static int 1384 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1385 { 1386 struct loop_info info; 1387 struct loop_info64 info64; 1388 1389 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1390 return -EFAULT; 1391 loop_info64_from_old(&info, &info64); 1392 return loop_set_status(lo, &info64); 1393 } 1394 1395 static int 1396 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1397 { 1398 struct loop_info64 info64; 1399 1400 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1401 return -EFAULT; 1402 return loop_set_status(lo, &info64); 1403 } 1404 1405 static int 1406 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1407 struct loop_info info; 1408 struct loop_info64 info64; 1409 int err; 1410 1411 if (!arg) 1412 return -EINVAL; 1413 err = loop_get_status(lo, &info64); 1414 if (!err) 1415 err = loop_info64_to_old(&info64, &info); 1416 if (!err && copy_to_user(arg, &info, sizeof(info))) 1417 err = -EFAULT; 1418 1419 return err; 1420 } 1421 1422 static int 1423 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1424 struct loop_info64 info64; 1425 int err; 1426 1427 if (!arg) 1428 return -EINVAL; 1429 err = loop_get_status(lo, &info64); 1430 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1431 err = -EFAULT; 1432 1433 return err; 1434 } 1435 1436 static int loop_set_capacity(struct loop_device *lo) 1437 { 1438 if (unlikely(lo->lo_state != Lo_bound)) 1439 return -ENXIO; 1440 1441 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1442 } 1443 1444 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1445 { 1446 int error = -ENXIO; 1447 if (lo->lo_state != Lo_bound) 1448 goto out; 1449 1450 __loop_update_dio(lo, !!arg); 1451 if (lo->use_dio == !!arg) 1452 return 0; 1453 error = -EINVAL; 1454 out: 1455 return error; 1456 } 1457 1458 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1459 { 1460 int err = 0; 1461 1462 if (lo->lo_state != Lo_bound) 1463 return -ENXIO; 1464 1465 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1466 return -EINVAL; 1467 1468 if (lo->lo_queue->limits.logical_block_size != arg) { 1469 sync_blockdev(lo->lo_device); 1470 kill_bdev(lo->lo_device); 1471 } 1472 1473 blk_mq_freeze_queue(lo->lo_queue); 1474 1475 /* kill_bdev should have truncated all the pages */ 1476 if (lo->lo_queue->limits.logical_block_size != arg && 1477 lo->lo_device->bd_inode->i_mapping->nrpages) { 1478 err = -EAGAIN; 1479 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1480 __func__, lo->lo_number, lo->lo_file_name, 1481 lo->lo_device->bd_inode->i_mapping->nrpages); 1482 goto out_unfreeze; 1483 } 1484 1485 blk_queue_logical_block_size(lo->lo_queue, arg); 1486 blk_queue_physical_block_size(lo->lo_queue, arg); 1487 blk_queue_io_min(lo->lo_queue, arg); 1488 loop_update_dio(lo); 1489 out_unfreeze: 1490 blk_mq_unfreeze_queue(lo->lo_queue); 1491 1492 return err; 1493 } 1494 1495 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1496 unsigned long arg) 1497 { 1498 int err; 1499 1500 err = mutex_lock_killable(&loop_ctl_mutex); 1501 if (err) 1502 return err; 1503 switch (cmd) { 1504 case LOOP_SET_CAPACITY: 1505 err = loop_set_capacity(lo); 1506 break; 1507 case LOOP_SET_DIRECT_IO: 1508 err = loop_set_dio(lo, arg); 1509 break; 1510 case LOOP_SET_BLOCK_SIZE: 1511 err = loop_set_block_size(lo, arg); 1512 break; 1513 default: 1514 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1515 } 1516 mutex_unlock(&loop_ctl_mutex); 1517 return err; 1518 } 1519 1520 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1521 unsigned int cmd, unsigned long arg) 1522 { 1523 struct loop_device *lo = bdev->bd_disk->private_data; 1524 int err; 1525 1526 switch (cmd) { 1527 case LOOP_SET_FD: 1528 return loop_set_fd(lo, mode, bdev, arg); 1529 case LOOP_CHANGE_FD: 1530 return loop_change_fd(lo, bdev, arg); 1531 case LOOP_CLR_FD: 1532 return loop_clr_fd(lo); 1533 case LOOP_SET_STATUS: 1534 err = -EPERM; 1535 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1536 err = loop_set_status_old(lo, 1537 (struct loop_info __user *)arg); 1538 } 1539 break; 1540 case LOOP_GET_STATUS: 1541 return loop_get_status_old(lo, (struct loop_info __user *) arg); 1542 case LOOP_SET_STATUS64: 1543 err = -EPERM; 1544 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1545 err = loop_set_status64(lo, 1546 (struct loop_info64 __user *) arg); 1547 } 1548 break; 1549 case LOOP_GET_STATUS64: 1550 return loop_get_status64(lo, (struct loop_info64 __user *) arg); 1551 case LOOP_SET_CAPACITY: 1552 case LOOP_SET_DIRECT_IO: 1553 case LOOP_SET_BLOCK_SIZE: 1554 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1555 return -EPERM; 1556 /* Fall through */ 1557 default: 1558 err = lo_simple_ioctl(lo, cmd, arg); 1559 break; 1560 } 1561 1562 return err; 1563 } 1564 1565 #ifdef CONFIG_COMPAT 1566 struct compat_loop_info { 1567 compat_int_t lo_number; /* ioctl r/o */ 1568 compat_dev_t lo_device; /* ioctl r/o */ 1569 compat_ulong_t lo_inode; /* ioctl r/o */ 1570 compat_dev_t lo_rdevice; /* ioctl r/o */ 1571 compat_int_t lo_offset; 1572 compat_int_t lo_encrypt_type; 1573 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1574 compat_int_t lo_flags; /* ioctl r/o */ 1575 char lo_name[LO_NAME_SIZE]; 1576 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1577 compat_ulong_t lo_init[2]; 1578 char reserved[4]; 1579 }; 1580 1581 /* 1582 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1583 * - noinlined to reduce stack space usage in main part of driver 1584 */ 1585 static noinline int 1586 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1587 struct loop_info64 *info64) 1588 { 1589 struct compat_loop_info info; 1590 1591 if (copy_from_user(&info, arg, sizeof(info))) 1592 return -EFAULT; 1593 1594 memset(info64, 0, sizeof(*info64)); 1595 info64->lo_number = info.lo_number; 1596 info64->lo_device = info.lo_device; 1597 info64->lo_inode = info.lo_inode; 1598 info64->lo_rdevice = info.lo_rdevice; 1599 info64->lo_offset = info.lo_offset; 1600 info64->lo_sizelimit = 0; 1601 info64->lo_encrypt_type = info.lo_encrypt_type; 1602 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1603 info64->lo_flags = info.lo_flags; 1604 info64->lo_init[0] = info.lo_init[0]; 1605 info64->lo_init[1] = info.lo_init[1]; 1606 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1607 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1608 else 1609 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1610 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1611 return 0; 1612 } 1613 1614 /* 1615 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1616 * - noinlined to reduce stack space usage in main part of driver 1617 */ 1618 static noinline int 1619 loop_info64_to_compat(const struct loop_info64 *info64, 1620 struct compat_loop_info __user *arg) 1621 { 1622 struct compat_loop_info info; 1623 1624 memset(&info, 0, sizeof(info)); 1625 info.lo_number = info64->lo_number; 1626 info.lo_device = info64->lo_device; 1627 info.lo_inode = info64->lo_inode; 1628 info.lo_rdevice = info64->lo_rdevice; 1629 info.lo_offset = info64->lo_offset; 1630 info.lo_encrypt_type = info64->lo_encrypt_type; 1631 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1632 info.lo_flags = info64->lo_flags; 1633 info.lo_init[0] = info64->lo_init[0]; 1634 info.lo_init[1] = info64->lo_init[1]; 1635 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1636 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1637 else 1638 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1639 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1640 1641 /* error in case values were truncated */ 1642 if (info.lo_device != info64->lo_device || 1643 info.lo_rdevice != info64->lo_rdevice || 1644 info.lo_inode != info64->lo_inode || 1645 info.lo_offset != info64->lo_offset || 1646 info.lo_init[0] != info64->lo_init[0] || 1647 info.lo_init[1] != info64->lo_init[1]) 1648 return -EOVERFLOW; 1649 1650 if (copy_to_user(arg, &info, sizeof(info))) 1651 return -EFAULT; 1652 return 0; 1653 } 1654 1655 static int 1656 loop_set_status_compat(struct loop_device *lo, 1657 const struct compat_loop_info __user *arg) 1658 { 1659 struct loop_info64 info64; 1660 int ret; 1661 1662 ret = loop_info64_from_compat(arg, &info64); 1663 if (ret < 0) 1664 return ret; 1665 return loop_set_status(lo, &info64); 1666 } 1667 1668 static int 1669 loop_get_status_compat(struct loop_device *lo, 1670 struct compat_loop_info __user *arg) 1671 { 1672 struct loop_info64 info64; 1673 int err; 1674 1675 if (!arg) 1676 return -EINVAL; 1677 err = loop_get_status(lo, &info64); 1678 if (!err) 1679 err = loop_info64_to_compat(&info64, arg); 1680 return err; 1681 } 1682 1683 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1684 unsigned int cmd, unsigned long arg) 1685 { 1686 struct loop_device *lo = bdev->bd_disk->private_data; 1687 int err; 1688 1689 switch(cmd) { 1690 case LOOP_SET_STATUS: 1691 err = loop_set_status_compat(lo, 1692 (const struct compat_loop_info __user *)arg); 1693 break; 1694 case LOOP_GET_STATUS: 1695 err = loop_get_status_compat(lo, 1696 (struct compat_loop_info __user *)arg); 1697 break; 1698 case LOOP_SET_CAPACITY: 1699 case LOOP_CLR_FD: 1700 case LOOP_GET_STATUS64: 1701 case LOOP_SET_STATUS64: 1702 arg = (unsigned long) compat_ptr(arg); 1703 /* fall through */ 1704 case LOOP_SET_FD: 1705 case LOOP_CHANGE_FD: 1706 case LOOP_SET_BLOCK_SIZE: 1707 err = lo_ioctl(bdev, mode, cmd, arg); 1708 break; 1709 default: 1710 err = -ENOIOCTLCMD; 1711 break; 1712 } 1713 return err; 1714 } 1715 #endif 1716 1717 static int lo_open(struct block_device *bdev, fmode_t mode) 1718 { 1719 struct loop_device *lo; 1720 int err; 1721 1722 err = mutex_lock_killable(&loop_ctl_mutex); 1723 if (err) 1724 return err; 1725 lo = bdev->bd_disk->private_data; 1726 if (!lo) { 1727 err = -ENXIO; 1728 goto out; 1729 } 1730 1731 atomic_inc(&lo->lo_refcnt); 1732 out: 1733 mutex_unlock(&loop_ctl_mutex); 1734 return err; 1735 } 1736 1737 static void lo_release(struct gendisk *disk, fmode_t mode) 1738 { 1739 struct loop_device *lo; 1740 1741 mutex_lock(&loop_ctl_mutex); 1742 lo = disk->private_data; 1743 if (atomic_dec_return(&lo->lo_refcnt)) 1744 goto out_unlock; 1745 1746 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1747 if (lo->lo_state != Lo_bound) 1748 goto out_unlock; 1749 lo->lo_state = Lo_rundown; 1750 mutex_unlock(&loop_ctl_mutex); 1751 /* 1752 * In autoclear mode, stop the loop thread 1753 * and remove configuration after last close. 1754 */ 1755 __loop_clr_fd(lo, true); 1756 return; 1757 } else if (lo->lo_state == Lo_bound) { 1758 /* 1759 * Otherwise keep thread (if running) and config, 1760 * but flush possible ongoing bios in thread. 1761 */ 1762 blk_mq_freeze_queue(lo->lo_queue); 1763 blk_mq_unfreeze_queue(lo->lo_queue); 1764 } 1765 1766 out_unlock: 1767 mutex_unlock(&loop_ctl_mutex); 1768 } 1769 1770 static const struct block_device_operations lo_fops = { 1771 .owner = THIS_MODULE, 1772 .open = lo_open, 1773 .release = lo_release, 1774 .ioctl = lo_ioctl, 1775 #ifdef CONFIG_COMPAT 1776 .compat_ioctl = lo_compat_ioctl, 1777 #endif 1778 }; 1779 1780 /* 1781 * And now the modules code and kernel interface. 1782 */ 1783 static int max_loop; 1784 module_param(max_loop, int, 0444); 1785 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1786 module_param(max_part, int, 0444); 1787 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1788 MODULE_LICENSE("GPL"); 1789 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1790 1791 int loop_register_transfer(struct loop_func_table *funcs) 1792 { 1793 unsigned int n = funcs->number; 1794 1795 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1796 return -EINVAL; 1797 xfer_funcs[n] = funcs; 1798 return 0; 1799 } 1800 1801 static int unregister_transfer_cb(int id, void *ptr, void *data) 1802 { 1803 struct loop_device *lo = ptr; 1804 struct loop_func_table *xfer = data; 1805 1806 mutex_lock(&loop_ctl_mutex); 1807 if (lo->lo_encryption == xfer) 1808 loop_release_xfer(lo); 1809 mutex_unlock(&loop_ctl_mutex); 1810 return 0; 1811 } 1812 1813 int loop_unregister_transfer(int number) 1814 { 1815 unsigned int n = number; 1816 struct loop_func_table *xfer; 1817 1818 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1819 return -EINVAL; 1820 1821 xfer_funcs[n] = NULL; 1822 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1823 return 0; 1824 } 1825 1826 EXPORT_SYMBOL(loop_register_transfer); 1827 EXPORT_SYMBOL(loop_unregister_transfer); 1828 1829 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1830 const struct blk_mq_queue_data *bd) 1831 { 1832 struct request *rq = bd->rq; 1833 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1834 struct loop_device *lo = rq->q->queuedata; 1835 1836 blk_mq_start_request(rq); 1837 1838 if (lo->lo_state != Lo_bound) 1839 return BLK_STS_IOERR; 1840 1841 switch (req_op(rq)) { 1842 case REQ_OP_FLUSH: 1843 case REQ_OP_DISCARD: 1844 case REQ_OP_WRITE_ZEROES: 1845 cmd->use_aio = false; 1846 break; 1847 default: 1848 cmd->use_aio = lo->use_dio; 1849 break; 1850 } 1851 1852 /* always use the first bio's css */ 1853 #ifdef CONFIG_BLK_CGROUP 1854 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 1855 cmd->css = &bio_blkcg(rq->bio)->css; 1856 css_get(cmd->css); 1857 } else 1858 #endif 1859 cmd->css = NULL; 1860 kthread_queue_work(&lo->worker, &cmd->work); 1861 1862 return BLK_STS_OK; 1863 } 1864 1865 static void loop_handle_cmd(struct loop_cmd *cmd) 1866 { 1867 struct request *rq = blk_mq_rq_from_pdu(cmd); 1868 const bool write = op_is_write(req_op(rq)); 1869 struct loop_device *lo = rq->q->queuedata; 1870 int ret = 0; 1871 1872 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1873 ret = -EIO; 1874 goto failed; 1875 } 1876 1877 ret = do_req_filebacked(lo, rq); 1878 failed: 1879 /* complete non-aio request */ 1880 if (!cmd->use_aio || ret) { 1881 cmd->ret = ret ? -EIO : 0; 1882 blk_mq_complete_request(rq); 1883 } 1884 } 1885 1886 static void loop_queue_work(struct kthread_work *work) 1887 { 1888 struct loop_cmd *cmd = 1889 container_of(work, struct loop_cmd, work); 1890 1891 loop_handle_cmd(cmd); 1892 } 1893 1894 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1895 unsigned int hctx_idx, unsigned int numa_node) 1896 { 1897 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1898 1899 kthread_init_work(&cmd->work, loop_queue_work); 1900 return 0; 1901 } 1902 1903 static const struct blk_mq_ops loop_mq_ops = { 1904 .queue_rq = loop_queue_rq, 1905 .init_request = loop_init_request, 1906 .complete = lo_complete_rq, 1907 }; 1908 1909 static int loop_add(struct loop_device **l, int i) 1910 { 1911 struct loop_device *lo; 1912 struct gendisk *disk; 1913 int err; 1914 1915 err = -ENOMEM; 1916 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1917 if (!lo) 1918 goto out; 1919 1920 lo->lo_state = Lo_unbound; 1921 1922 /* allocate id, if @id >= 0, we're requesting that specific id */ 1923 if (i >= 0) { 1924 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1925 if (err == -ENOSPC) 1926 err = -EEXIST; 1927 } else { 1928 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1929 } 1930 if (err < 0) 1931 goto out_free_dev; 1932 i = err; 1933 1934 err = -ENOMEM; 1935 lo->tag_set.ops = &loop_mq_ops; 1936 lo->tag_set.nr_hw_queues = 1; 1937 lo->tag_set.queue_depth = 128; 1938 lo->tag_set.numa_node = NUMA_NO_NODE; 1939 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1940 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; 1941 lo->tag_set.driver_data = lo; 1942 1943 err = blk_mq_alloc_tag_set(&lo->tag_set); 1944 if (err) 1945 goto out_free_idr; 1946 1947 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1948 if (IS_ERR(lo->lo_queue)) { 1949 err = PTR_ERR(lo->lo_queue); 1950 goto out_cleanup_tags; 1951 } 1952 lo->lo_queue->queuedata = lo; 1953 1954 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1955 1956 /* 1957 * By default, we do buffer IO, so it doesn't make sense to enable 1958 * merge because the I/O submitted to backing file is handled page by 1959 * page. For directio mode, merge does help to dispatch bigger request 1960 * to underlayer disk. We will enable merge once directio is enabled. 1961 */ 1962 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1963 1964 err = -ENOMEM; 1965 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1966 if (!disk) 1967 goto out_free_queue; 1968 1969 /* 1970 * Disable partition scanning by default. The in-kernel partition 1971 * scanning can be requested individually per-device during its 1972 * setup. Userspace can always add and remove partitions from all 1973 * devices. The needed partition minors are allocated from the 1974 * extended minor space, the main loop device numbers will continue 1975 * to match the loop minors, regardless of the number of partitions 1976 * used. 1977 * 1978 * If max_part is given, partition scanning is globally enabled for 1979 * all loop devices. The minors for the main loop devices will be 1980 * multiples of max_part. 1981 * 1982 * Note: Global-for-all-devices, set-only-at-init, read-only module 1983 * parameteters like 'max_loop' and 'max_part' make things needlessly 1984 * complicated, are too static, inflexible and may surprise 1985 * userspace tools. Parameters like this in general should be avoided. 1986 */ 1987 if (!part_shift) 1988 disk->flags |= GENHD_FL_NO_PART_SCAN; 1989 disk->flags |= GENHD_FL_EXT_DEVT; 1990 atomic_set(&lo->lo_refcnt, 0); 1991 lo->lo_number = i; 1992 spin_lock_init(&lo->lo_lock); 1993 disk->major = LOOP_MAJOR; 1994 disk->first_minor = i << part_shift; 1995 disk->fops = &lo_fops; 1996 disk->private_data = lo; 1997 disk->queue = lo->lo_queue; 1998 sprintf(disk->disk_name, "loop%d", i); 1999 add_disk(disk); 2000 *l = lo; 2001 return lo->lo_number; 2002 2003 out_free_queue: 2004 blk_cleanup_queue(lo->lo_queue); 2005 out_cleanup_tags: 2006 blk_mq_free_tag_set(&lo->tag_set); 2007 out_free_idr: 2008 idr_remove(&loop_index_idr, i); 2009 out_free_dev: 2010 kfree(lo); 2011 out: 2012 return err; 2013 } 2014 2015 static void loop_remove(struct loop_device *lo) 2016 { 2017 del_gendisk(lo->lo_disk); 2018 blk_cleanup_queue(lo->lo_queue); 2019 blk_mq_free_tag_set(&lo->tag_set); 2020 put_disk(lo->lo_disk); 2021 kfree(lo); 2022 } 2023 2024 static int find_free_cb(int id, void *ptr, void *data) 2025 { 2026 struct loop_device *lo = ptr; 2027 struct loop_device **l = data; 2028 2029 if (lo->lo_state == Lo_unbound) { 2030 *l = lo; 2031 return 1; 2032 } 2033 return 0; 2034 } 2035 2036 static int loop_lookup(struct loop_device **l, int i) 2037 { 2038 struct loop_device *lo; 2039 int ret = -ENODEV; 2040 2041 if (i < 0) { 2042 int err; 2043 2044 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 2045 if (err == 1) { 2046 *l = lo; 2047 ret = lo->lo_number; 2048 } 2049 goto out; 2050 } 2051 2052 /* lookup and return a specific i */ 2053 lo = idr_find(&loop_index_idr, i); 2054 if (lo) { 2055 *l = lo; 2056 ret = lo->lo_number; 2057 } 2058 out: 2059 return ret; 2060 } 2061 2062 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 2063 { 2064 struct loop_device *lo; 2065 struct kobject *kobj; 2066 int err; 2067 2068 mutex_lock(&loop_ctl_mutex); 2069 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 2070 if (err < 0) 2071 err = loop_add(&lo, MINOR(dev) >> part_shift); 2072 if (err < 0) 2073 kobj = NULL; 2074 else 2075 kobj = get_disk_and_module(lo->lo_disk); 2076 mutex_unlock(&loop_ctl_mutex); 2077 2078 *part = 0; 2079 return kobj; 2080 } 2081 2082 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2083 unsigned long parm) 2084 { 2085 struct loop_device *lo; 2086 int ret; 2087 2088 ret = mutex_lock_killable(&loop_ctl_mutex); 2089 if (ret) 2090 return ret; 2091 2092 ret = -ENOSYS; 2093 switch (cmd) { 2094 case LOOP_CTL_ADD: 2095 ret = loop_lookup(&lo, parm); 2096 if (ret >= 0) { 2097 ret = -EEXIST; 2098 break; 2099 } 2100 ret = loop_add(&lo, parm); 2101 break; 2102 case LOOP_CTL_REMOVE: 2103 ret = loop_lookup(&lo, parm); 2104 if (ret < 0) 2105 break; 2106 if (lo->lo_state != Lo_unbound) { 2107 ret = -EBUSY; 2108 break; 2109 } 2110 if (atomic_read(&lo->lo_refcnt) > 0) { 2111 ret = -EBUSY; 2112 break; 2113 } 2114 lo->lo_disk->private_data = NULL; 2115 idr_remove(&loop_index_idr, lo->lo_number); 2116 loop_remove(lo); 2117 break; 2118 case LOOP_CTL_GET_FREE: 2119 ret = loop_lookup(&lo, -1); 2120 if (ret >= 0) 2121 break; 2122 ret = loop_add(&lo, -1); 2123 } 2124 mutex_unlock(&loop_ctl_mutex); 2125 2126 return ret; 2127 } 2128 2129 static const struct file_operations loop_ctl_fops = { 2130 .open = nonseekable_open, 2131 .unlocked_ioctl = loop_control_ioctl, 2132 .compat_ioctl = loop_control_ioctl, 2133 .owner = THIS_MODULE, 2134 .llseek = noop_llseek, 2135 }; 2136 2137 static struct miscdevice loop_misc = { 2138 .minor = LOOP_CTRL_MINOR, 2139 .name = "loop-control", 2140 .fops = &loop_ctl_fops, 2141 }; 2142 2143 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2144 MODULE_ALIAS("devname:loop-control"); 2145 2146 static int __init loop_init(void) 2147 { 2148 int i, nr; 2149 unsigned long range; 2150 struct loop_device *lo; 2151 int err; 2152 2153 part_shift = 0; 2154 if (max_part > 0) { 2155 part_shift = fls(max_part); 2156 2157 /* 2158 * Adjust max_part according to part_shift as it is exported 2159 * to user space so that user can decide correct minor number 2160 * if [s]he want to create more devices. 2161 * 2162 * Note that -1 is required because partition 0 is reserved 2163 * for the whole disk. 2164 */ 2165 max_part = (1UL << part_shift) - 1; 2166 } 2167 2168 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2169 err = -EINVAL; 2170 goto err_out; 2171 } 2172 2173 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2174 err = -EINVAL; 2175 goto err_out; 2176 } 2177 2178 /* 2179 * If max_loop is specified, create that many devices upfront. 2180 * This also becomes a hard limit. If max_loop is not specified, 2181 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2182 * init time. Loop devices can be requested on-demand with the 2183 * /dev/loop-control interface, or be instantiated by accessing 2184 * a 'dead' device node. 2185 */ 2186 if (max_loop) { 2187 nr = max_loop; 2188 range = max_loop << part_shift; 2189 } else { 2190 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2191 range = 1UL << MINORBITS; 2192 } 2193 2194 err = misc_register(&loop_misc); 2195 if (err < 0) 2196 goto err_out; 2197 2198 2199 if (register_blkdev(LOOP_MAJOR, "loop")) { 2200 err = -EIO; 2201 goto misc_out; 2202 } 2203 2204 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2205 THIS_MODULE, loop_probe, NULL, NULL); 2206 2207 /* pre-create number of devices given by config or max_loop */ 2208 mutex_lock(&loop_ctl_mutex); 2209 for (i = 0; i < nr; i++) 2210 loop_add(&lo, i); 2211 mutex_unlock(&loop_ctl_mutex); 2212 2213 printk(KERN_INFO "loop: module loaded\n"); 2214 return 0; 2215 2216 misc_out: 2217 misc_deregister(&loop_misc); 2218 err_out: 2219 return err; 2220 } 2221 2222 static int loop_exit_cb(int id, void *ptr, void *data) 2223 { 2224 struct loop_device *lo = ptr; 2225 2226 loop_remove(lo); 2227 return 0; 2228 } 2229 2230 static void __exit loop_exit(void) 2231 { 2232 unsigned long range; 2233 2234 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2235 2236 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2237 idr_destroy(&loop_index_idr); 2238 2239 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2240 unregister_blkdev(LOOP_MAJOR, "loop"); 2241 2242 misc_deregister(&loop_misc); 2243 } 2244 2245 module_init(loop_init); 2246 module_exit(loop_exit); 2247 2248 #ifndef MODULE 2249 static int __init max_loop_setup(char *str) 2250 { 2251 max_loop = simple_strtol(str, NULL, 0); 2252 return 1; 2253 } 2254 2255 __setup("max_loop=", max_loop_setup); 2256 #endif 2257