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 if (lo->lo_state == Lo_bound) 218 blk_mq_freeze_queue(lo->lo_queue); 219 lo->use_dio = use_dio; 220 if (use_dio) { 221 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 222 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 223 } else { 224 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 225 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 226 } 227 if (lo->lo_state == Lo_bound) 228 blk_mq_unfreeze_queue(lo->lo_queue); 229 } 230 231 /** 232 * loop_set_size() - sets device size and notifies userspace 233 * @lo: struct loop_device to set the size for 234 * @size: new size of the loop device 235 * 236 * Callers must validate that the size passed into this function fits into 237 * a sector_t, eg using loop_validate_size() 238 */ 239 static void loop_set_size(struct loop_device *lo, loff_t size) 240 { 241 struct block_device *bdev = lo->lo_device; 242 243 bd_set_size(bdev, size << SECTOR_SHIFT); 244 245 set_capacity_revalidate_and_notify(lo->lo_disk, size, false); 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_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 420 int mode) 421 { 422 /* 423 * We use fallocate to manipulate the space mappings used by the image 424 * a.k.a. discard/zerorange. However we do not support this if 425 * encryption is enabled, because it may give an attacker useful 426 * information. 427 */ 428 struct file *file = lo->lo_backing_file; 429 struct request_queue *q = lo->lo_queue; 430 int ret; 431 432 mode |= FALLOC_FL_KEEP_SIZE; 433 434 if (!blk_queue_discard(q)) { 435 ret = -EOPNOTSUPP; 436 goto out; 437 } 438 439 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 440 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 441 ret = -EIO; 442 out: 443 return ret; 444 } 445 446 static int lo_req_flush(struct loop_device *lo, struct request *rq) 447 { 448 struct file *file = lo->lo_backing_file; 449 int ret = vfs_fsync(file, 0); 450 if (unlikely(ret && ret != -EINVAL)) 451 ret = -EIO; 452 453 return ret; 454 } 455 456 static void lo_complete_rq(struct request *rq) 457 { 458 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 459 blk_status_t ret = BLK_STS_OK; 460 461 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 462 req_op(rq) != REQ_OP_READ) { 463 if (cmd->ret < 0) 464 ret = errno_to_blk_status(cmd->ret); 465 goto end_io; 466 } 467 468 /* 469 * Short READ - if we got some data, advance our request and 470 * retry it. If we got no data, end the rest with EIO. 471 */ 472 if (cmd->ret) { 473 blk_update_request(rq, BLK_STS_OK, cmd->ret); 474 cmd->ret = 0; 475 blk_mq_requeue_request(rq, true); 476 } else { 477 if (cmd->use_aio) { 478 struct bio *bio = rq->bio; 479 480 while (bio) { 481 zero_fill_bio(bio); 482 bio = bio->bi_next; 483 } 484 } 485 ret = BLK_STS_IOERR; 486 end_io: 487 blk_mq_end_request(rq, ret); 488 } 489 } 490 491 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 492 { 493 struct request *rq = blk_mq_rq_from_pdu(cmd); 494 495 if (!atomic_dec_and_test(&cmd->ref)) 496 return; 497 kfree(cmd->bvec); 498 cmd->bvec = NULL; 499 blk_mq_complete_request(rq); 500 } 501 502 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 503 { 504 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 505 506 if (cmd->css) 507 css_put(cmd->css); 508 cmd->ret = ret; 509 lo_rw_aio_do_completion(cmd); 510 } 511 512 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 513 loff_t pos, bool rw) 514 { 515 struct iov_iter iter; 516 struct req_iterator rq_iter; 517 struct bio_vec *bvec; 518 struct request *rq = blk_mq_rq_from_pdu(cmd); 519 struct bio *bio = rq->bio; 520 struct file *file = lo->lo_backing_file; 521 struct bio_vec tmp; 522 unsigned int offset; 523 int nr_bvec = 0; 524 int ret; 525 526 rq_for_each_bvec(tmp, rq, rq_iter) 527 nr_bvec++; 528 529 if (rq->bio != rq->biotail) { 530 531 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 532 GFP_NOIO); 533 if (!bvec) 534 return -EIO; 535 cmd->bvec = bvec; 536 537 /* 538 * The bios of the request may be started from the middle of 539 * the 'bvec' because of bio splitting, so we can't directly 540 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 541 * API will take care of all details for us. 542 */ 543 rq_for_each_bvec(tmp, rq, rq_iter) { 544 *bvec = tmp; 545 bvec++; 546 } 547 bvec = cmd->bvec; 548 offset = 0; 549 } else { 550 /* 551 * Same here, this bio may be started from the middle of the 552 * 'bvec' because of bio splitting, so offset from the bvec 553 * must be passed to iov iterator 554 */ 555 offset = bio->bi_iter.bi_bvec_done; 556 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 557 } 558 atomic_set(&cmd->ref, 2); 559 560 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 561 iter.iov_offset = offset; 562 563 cmd->iocb.ki_pos = pos; 564 cmd->iocb.ki_filp = file; 565 cmd->iocb.ki_complete = lo_rw_aio_complete; 566 cmd->iocb.ki_flags = IOCB_DIRECT; 567 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 568 if (cmd->css) 569 kthread_associate_blkcg(cmd->css); 570 571 if (rw == WRITE) 572 ret = call_write_iter(file, &cmd->iocb, &iter); 573 else 574 ret = call_read_iter(file, &cmd->iocb, &iter); 575 576 lo_rw_aio_do_completion(cmd); 577 kthread_associate_blkcg(NULL); 578 579 if (ret != -EIOCBQUEUED) 580 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 581 return 0; 582 } 583 584 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 585 { 586 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 587 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 588 589 /* 590 * lo_write_simple and lo_read_simple should have been covered 591 * by io submit style function like lo_rw_aio(), one blocker 592 * is that lo_read_simple() need to call flush_dcache_page after 593 * the page is written from kernel, and it isn't easy to handle 594 * this in io submit style function which submits all segments 595 * of the req at one time. And direct read IO doesn't need to 596 * run flush_dcache_page(). 597 */ 598 switch (req_op(rq)) { 599 case REQ_OP_FLUSH: 600 return lo_req_flush(lo, rq); 601 case REQ_OP_WRITE_ZEROES: 602 /* 603 * If the caller doesn't want deallocation, call zeroout to 604 * write zeroes the range. Otherwise, punch them out. 605 */ 606 return lo_fallocate(lo, rq, pos, 607 (rq->cmd_flags & REQ_NOUNMAP) ? 608 FALLOC_FL_ZERO_RANGE : 609 FALLOC_FL_PUNCH_HOLE); 610 case REQ_OP_DISCARD: 611 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 612 case REQ_OP_WRITE: 613 if (lo->transfer) 614 return lo_write_transfer(lo, rq, pos); 615 else if (cmd->use_aio) 616 return lo_rw_aio(lo, cmd, pos, WRITE); 617 else 618 return lo_write_simple(lo, rq, pos); 619 case REQ_OP_READ: 620 if (lo->transfer) 621 return lo_read_transfer(lo, rq, pos); 622 else if (cmd->use_aio) 623 return lo_rw_aio(lo, cmd, pos, READ); 624 else 625 return lo_read_simple(lo, rq, pos); 626 default: 627 WARN_ON_ONCE(1); 628 return -EIO; 629 } 630 } 631 632 static inline void loop_update_dio(struct loop_device *lo) 633 { 634 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 635 lo->use_dio); 636 } 637 638 static void loop_reread_partitions(struct loop_device *lo, 639 struct block_device *bdev) 640 { 641 int rc; 642 643 mutex_lock(&bdev->bd_mutex); 644 rc = bdev_disk_changed(bdev, false); 645 mutex_unlock(&bdev->bd_mutex); 646 if (rc) 647 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 648 __func__, lo->lo_number, lo->lo_file_name, rc); 649 } 650 651 static inline int is_loop_device(struct file *file) 652 { 653 struct inode *i = file->f_mapping->host; 654 655 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 656 } 657 658 static int loop_validate_file(struct file *file, struct block_device *bdev) 659 { 660 struct inode *inode = file->f_mapping->host; 661 struct file *f = file; 662 663 /* Avoid recursion */ 664 while (is_loop_device(f)) { 665 struct loop_device *l; 666 667 if (f->f_mapping->host->i_bdev == bdev) 668 return -EBADF; 669 670 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 671 if (l->lo_state != Lo_bound) { 672 return -EINVAL; 673 } 674 f = l->lo_backing_file; 675 } 676 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 677 return -EINVAL; 678 return 0; 679 } 680 681 /* 682 * loop_change_fd switched the backing store of a loopback device to 683 * a new file. This is useful for operating system installers to free up 684 * the original file and in High Availability environments to switch to 685 * an alternative location for the content in case of server meltdown. 686 * This can only work if the loop device is used read-only, and if the 687 * new backing store is the same size and type as the old backing store. 688 */ 689 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 690 unsigned int arg) 691 { 692 struct file *file = NULL, *old_file; 693 int error; 694 bool partscan; 695 696 error = mutex_lock_killable(&loop_ctl_mutex); 697 if (error) 698 return error; 699 error = -ENXIO; 700 if (lo->lo_state != Lo_bound) 701 goto out_err; 702 703 /* the loop device has to be read-only */ 704 error = -EINVAL; 705 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 706 goto out_err; 707 708 error = -EBADF; 709 file = fget(arg); 710 if (!file) 711 goto out_err; 712 713 error = loop_validate_file(file, bdev); 714 if (error) 715 goto out_err; 716 717 old_file = lo->lo_backing_file; 718 719 error = -EINVAL; 720 721 /* size of the new backing store needs to be the same */ 722 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 723 goto out_err; 724 725 /* and ... switch */ 726 blk_mq_freeze_queue(lo->lo_queue); 727 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 728 lo->lo_backing_file = file; 729 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 730 mapping_set_gfp_mask(file->f_mapping, 731 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 732 loop_update_dio(lo); 733 blk_mq_unfreeze_queue(lo->lo_queue); 734 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 735 mutex_unlock(&loop_ctl_mutex); 736 /* 737 * We must drop file reference outside of loop_ctl_mutex as dropping 738 * the file ref can take bd_mutex which creates circular locking 739 * dependency. 740 */ 741 fput(old_file); 742 if (partscan) 743 loop_reread_partitions(lo, bdev); 744 return 0; 745 746 out_err: 747 mutex_unlock(&loop_ctl_mutex); 748 if (file) 749 fput(file); 750 return error; 751 } 752 753 /* loop sysfs attributes */ 754 755 static ssize_t loop_attr_show(struct device *dev, char *page, 756 ssize_t (*callback)(struct loop_device *, char *)) 757 { 758 struct gendisk *disk = dev_to_disk(dev); 759 struct loop_device *lo = disk->private_data; 760 761 return callback(lo, page); 762 } 763 764 #define LOOP_ATTR_RO(_name) \ 765 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 766 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 767 struct device_attribute *attr, char *b) \ 768 { \ 769 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 770 } \ 771 static struct device_attribute loop_attr_##_name = \ 772 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 773 774 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 775 { 776 ssize_t ret; 777 char *p = NULL; 778 779 spin_lock_irq(&lo->lo_lock); 780 if (lo->lo_backing_file) 781 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 782 spin_unlock_irq(&lo->lo_lock); 783 784 if (IS_ERR_OR_NULL(p)) 785 ret = PTR_ERR(p); 786 else { 787 ret = strlen(p); 788 memmove(buf, p, ret); 789 buf[ret++] = '\n'; 790 buf[ret] = 0; 791 } 792 793 return ret; 794 } 795 796 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 797 { 798 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 799 } 800 801 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 802 { 803 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 804 } 805 806 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 807 { 808 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 809 810 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 811 } 812 813 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 814 { 815 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 816 817 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 818 } 819 820 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 821 { 822 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 823 824 return sprintf(buf, "%s\n", dio ? "1" : "0"); 825 } 826 827 LOOP_ATTR_RO(backing_file); 828 LOOP_ATTR_RO(offset); 829 LOOP_ATTR_RO(sizelimit); 830 LOOP_ATTR_RO(autoclear); 831 LOOP_ATTR_RO(partscan); 832 LOOP_ATTR_RO(dio); 833 834 static struct attribute *loop_attrs[] = { 835 &loop_attr_backing_file.attr, 836 &loop_attr_offset.attr, 837 &loop_attr_sizelimit.attr, 838 &loop_attr_autoclear.attr, 839 &loop_attr_partscan.attr, 840 &loop_attr_dio.attr, 841 NULL, 842 }; 843 844 static struct attribute_group loop_attribute_group = { 845 .name = "loop", 846 .attrs= loop_attrs, 847 }; 848 849 static void loop_sysfs_init(struct loop_device *lo) 850 { 851 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 852 &loop_attribute_group); 853 } 854 855 static void loop_sysfs_exit(struct loop_device *lo) 856 { 857 if (lo->sysfs_inited) 858 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 859 &loop_attribute_group); 860 } 861 862 static void loop_config_discard(struct loop_device *lo) 863 { 864 struct file *file = lo->lo_backing_file; 865 struct inode *inode = file->f_mapping->host; 866 struct request_queue *q = lo->lo_queue; 867 868 /* 869 * If the backing device is a block device, mirror its zeroing 870 * capability. Set the discard sectors to the block device's zeroing 871 * capabilities because loop discards result in blkdev_issue_zeroout(), 872 * not blkdev_issue_discard(). This maintains consistent behavior with 873 * file-backed loop devices: discarded regions read back as zero. 874 */ 875 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { 876 struct request_queue *backingq; 877 878 backingq = bdev_get_queue(inode->i_bdev); 879 blk_queue_max_discard_sectors(q, 880 backingq->limits.max_write_zeroes_sectors); 881 882 blk_queue_max_write_zeroes_sectors(q, 883 backingq->limits.max_write_zeroes_sectors); 884 885 /* 886 * We use punch hole to reclaim the free space used by the 887 * image a.k.a. discard. However we do not support discard if 888 * encryption is enabled, because it may give an attacker 889 * useful information. 890 */ 891 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { 892 q->limits.discard_granularity = 0; 893 q->limits.discard_alignment = 0; 894 blk_queue_max_discard_sectors(q, 0); 895 blk_queue_max_write_zeroes_sectors(q, 0); 896 897 } else { 898 q->limits.discard_granularity = inode->i_sb->s_blocksize; 899 q->limits.discard_alignment = 0; 900 901 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 902 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 903 } 904 905 if (q->limits.max_write_zeroes_sectors) 906 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 907 else 908 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 909 } 910 911 static void loop_unprepare_queue(struct loop_device *lo) 912 { 913 kthread_flush_worker(&lo->worker); 914 kthread_stop(lo->worker_task); 915 } 916 917 static int loop_kthread_worker_fn(void *worker_ptr) 918 { 919 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO; 920 return kthread_worker_fn(worker_ptr); 921 } 922 923 static int loop_prepare_queue(struct loop_device *lo) 924 { 925 kthread_init_worker(&lo->worker); 926 lo->worker_task = kthread_run(loop_kthread_worker_fn, 927 &lo->worker, "loop%d", lo->lo_number); 928 if (IS_ERR(lo->worker_task)) 929 return -ENOMEM; 930 set_user_nice(lo->worker_task, MIN_NICE); 931 return 0; 932 } 933 934 static void loop_update_rotational(struct loop_device *lo) 935 { 936 struct file *file = lo->lo_backing_file; 937 struct inode *file_inode = file->f_mapping->host; 938 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 939 struct request_queue *q = lo->lo_queue; 940 bool nonrot = true; 941 942 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 943 if (file_bdev) 944 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 945 946 if (nonrot) 947 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 948 else 949 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 950 } 951 952 static int 953 loop_release_xfer(struct loop_device *lo) 954 { 955 int err = 0; 956 struct loop_func_table *xfer = lo->lo_encryption; 957 958 if (xfer) { 959 if (xfer->release) 960 err = xfer->release(lo); 961 lo->transfer = NULL; 962 lo->lo_encryption = NULL; 963 module_put(xfer->owner); 964 } 965 return err; 966 } 967 968 static int 969 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 970 const struct loop_info64 *i) 971 { 972 int err = 0; 973 974 if (xfer) { 975 struct module *owner = xfer->owner; 976 977 if (!try_module_get(owner)) 978 return -EINVAL; 979 if (xfer->init) 980 err = xfer->init(lo, i); 981 if (err) 982 module_put(owner); 983 else 984 lo->lo_encryption = xfer; 985 } 986 return err; 987 } 988 989 /** 990 * loop_set_status_from_info - configure device from loop_info 991 * @lo: struct loop_device to configure 992 * @info: struct loop_info64 to configure the device with 993 * 994 * Configures the loop device parameters according to the passed 995 * in loop_info64 configuration. 996 */ 997 static int 998 loop_set_status_from_info(struct loop_device *lo, 999 const struct loop_info64 *info) 1000 { 1001 int err; 1002 struct loop_func_table *xfer; 1003 kuid_t uid = current_uid(); 1004 1005 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1006 return -EINVAL; 1007 1008 err = loop_release_xfer(lo); 1009 if (err) 1010 return err; 1011 1012 if (info->lo_encrypt_type) { 1013 unsigned int type = info->lo_encrypt_type; 1014 1015 if (type >= MAX_LO_CRYPT) 1016 return -EINVAL; 1017 xfer = xfer_funcs[type]; 1018 if (xfer == NULL) 1019 return -EINVAL; 1020 } else 1021 xfer = NULL; 1022 1023 err = loop_init_xfer(lo, xfer, info); 1024 if (err) 1025 return err; 1026 1027 lo->lo_offset = info->lo_offset; 1028 lo->lo_sizelimit = info->lo_sizelimit; 1029 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1030 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1031 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1032 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1033 1034 if (!xfer) 1035 xfer = &none_funcs; 1036 lo->transfer = xfer->transfer; 1037 lo->ioctl = xfer->ioctl; 1038 1039 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1040 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1041 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1042 1043 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1044 lo->lo_init[0] = info->lo_init[0]; 1045 lo->lo_init[1] = info->lo_init[1]; 1046 if (info->lo_encrypt_key_size) { 1047 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1048 info->lo_encrypt_key_size); 1049 lo->lo_key_owner = uid; 1050 } 1051 1052 return 0; 1053 } 1054 1055 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 1056 struct block_device *bdev, unsigned int arg) 1057 { 1058 struct file *file; 1059 struct inode *inode; 1060 struct address_space *mapping; 1061 struct block_device *claimed_bdev = NULL; 1062 int lo_flags = 0; 1063 int error; 1064 loff_t size; 1065 bool partscan; 1066 1067 /* This is safe, since we have a reference from open(). */ 1068 __module_get(THIS_MODULE); 1069 1070 error = -EBADF; 1071 file = fget(arg); 1072 if (!file) 1073 goto out; 1074 1075 /* 1076 * If we don't hold exclusive handle for the device, upgrade to it 1077 * here to avoid changing device under exclusive owner. 1078 */ 1079 if (!(mode & FMODE_EXCL)) { 1080 claimed_bdev = bd_start_claiming(bdev, loop_set_fd); 1081 if (IS_ERR(claimed_bdev)) { 1082 error = PTR_ERR(claimed_bdev); 1083 goto out_putf; 1084 } 1085 } 1086 1087 error = mutex_lock_killable(&loop_ctl_mutex); 1088 if (error) 1089 goto out_bdev; 1090 1091 error = -EBUSY; 1092 if (lo->lo_state != Lo_unbound) 1093 goto out_unlock; 1094 1095 error = loop_validate_file(file, bdev); 1096 if (error) 1097 goto out_unlock; 1098 1099 mapping = file->f_mapping; 1100 inode = mapping->host; 1101 1102 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1103 !file->f_op->write_iter) 1104 lo_flags |= LO_FLAGS_READ_ONLY; 1105 1106 size = get_loop_size(lo, file); 1107 1108 error = loop_prepare_queue(lo); 1109 if (error) 1110 goto out_unlock; 1111 1112 error = 0; 1113 1114 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 1115 1116 lo->use_dio = false; 1117 lo->lo_device = bdev; 1118 lo->lo_flags = lo_flags; 1119 lo->lo_backing_file = file; 1120 lo->transfer = NULL; 1121 lo->ioctl = NULL; 1122 lo->lo_sizelimit = 0; 1123 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1124 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1125 1126 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1127 blk_queue_write_cache(lo->lo_queue, true, false); 1128 1129 if (io_is_direct(lo->lo_backing_file) && inode->i_sb->s_bdev) { 1130 /* In case of direct I/O, match underlying block size */ 1131 unsigned short bsize = bdev_logical_block_size( 1132 inode->i_sb->s_bdev); 1133 1134 blk_queue_logical_block_size(lo->lo_queue, bsize); 1135 blk_queue_physical_block_size(lo->lo_queue, bsize); 1136 blk_queue_io_min(lo->lo_queue, bsize); 1137 } 1138 1139 loop_update_rotational(lo); 1140 loop_update_dio(lo); 1141 loop_sysfs_init(lo); 1142 loop_set_size(lo, size); 1143 1144 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 1145 block_size(inode->i_bdev) : PAGE_SIZE); 1146 1147 lo->lo_state = Lo_bound; 1148 if (part_shift) 1149 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1150 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1151 1152 /* Grab the block_device to prevent its destruction after we 1153 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 1154 */ 1155 bdgrab(bdev); 1156 mutex_unlock(&loop_ctl_mutex); 1157 if (partscan) 1158 loop_reread_partitions(lo, bdev); 1159 if (claimed_bdev) 1160 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd); 1161 return 0; 1162 1163 out_unlock: 1164 mutex_unlock(&loop_ctl_mutex); 1165 out_bdev: 1166 if (claimed_bdev) 1167 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd); 1168 out_putf: 1169 fput(file); 1170 out: 1171 /* This is safe: open() is still holding a reference. */ 1172 module_put(THIS_MODULE); 1173 return error; 1174 } 1175 1176 static int __loop_clr_fd(struct loop_device *lo, bool release) 1177 { 1178 struct file *filp = NULL; 1179 gfp_t gfp = lo->old_gfp_mask; 1180 struct block_device *bdev = lo->lo_device; 1181 int err = 0; 1182 bool partscan = false; 1183 int lo_number; 1184 1185 mutex_lock(&loop_ctl_mutex); 1186 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1187 err = -ENXIO; 1188 goto out_unlock; 1189 } 1190 1191 filp = lo->lo_backing_file; 1192 if (filp == NULL) { 1193 err = -EINVAL; 1194 goto out_unlock; 1195 } 1196 1197 /* freeze request queue during the transition */ 1198 blk_mq_freeze_queue(lo->lo_queue); 1199 1200 spin_lock_irq(&lo->lo_lock); 1201 lo->lo_backing_file = NULL; 1202 spin_unlock_irq(&lo->lo_lock); 1203 1204 loop_release_xfer(lo); 1205 lo->transfer = NULL; 1206 lo->ioctl = NULL; 1207 lo->lo_device = NULL; 1208 lo->lo_encryption = NULL; 1209 lo->lo_offset = 0; 1210 lo->lo_sizelimit = 0; 1211 lo->lo_encrypt_key_size = 0; 1212 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1213 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1214 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1215 blk_queue_logical_block_size(lo->lo_queue, 512); 1216 blk_queue_physical_block_size(lo->lo_queue, 512); 1217 blk_queue_io_min(lo->lo_queue, 512); 1218 if (bdev) { 1219 bdput(bdev); 1220 invalidate_bdev(bdev); 1221 bdev->bd_inode->i_mapping->wb_err = 0; 1222 } 1223 set_capacity(lo->lo_disk, 0); 1224 loop_sysfs_exit(lo); 1225 if (bdev) { 1226 bd_set_size(bdev, 0); 1227 /* let user-space know about this change */ 1228 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1229 } 1230 mapping_set_gfp_mask(filp->f_mapping, gfp); 1231 /* This is safe: open() is still holding a reference. */ 1232 module_put(THIS_MODULE); 1233 blk_mq_unfreeze_queue(lo->lo_queue); 1234 1235 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1236 lo_number = lo->lo_number; 1237 loop_unprepare_queue(lo); 1238 out_unlock: 1239 mutex_unlock(&loop_ctl_mutex); 1240 if (partscan) { 1241 /* 1242 * bd_mutex has been held already in release path, so don't 1243 * acquire it if this function is called in such case. 1244 * 1245 * If the reread partition isn't from release path, lo_refcnt 1246 * must be at least one and it can only become zero when the 1247 * current holder is released. 1248 */ 1249 if (!release) 1250 mutex_lock(&bdev->bd_mutex); 1251 err = bdev_disk_changed(bdev, false); 1252 if (!release) 1253 mutex_unlock(&bdev->bd_mutex); 1254 if (err) 1255 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1256 __func__, lo_number, err); 1257 /* Device is gone, no point in returning error */ 1258 err = 0; 1259 } 1260 1261 /* 1262 * lo->lo_state is set to Lo_unbound here after above partscan has 1263 * finished. 1264 * 1265 * There cannot be anybody else entering __loop_clr_fd() as 1266 * lo->lo_backing_file is already cleared and Lo_rundown state 1267 * protects us from all the other places trying to change the 'lo' 1268 * device. 1269 */ 1270 mutex_lock(&loop_ctl_mutex); 1271 lo->lo_flags = 0; 1272 if (!part_shift) 1273 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1274 lo->lo_state = Lo_unbound; 1275 mutex_unlock(&loop_ctl_mutex); 1276 1277 /* 1278 * Need not hold loop_ctl_mutex to fput backing file. 1279 * Calling fput holding loop_ctl_mutex triggers a circular 1280 * lock dependency possibility warning as fput can take 1281 * bd_mutex which is usually taken before loop_ctl_mutex. 1282 */ 1283 if (filp) 1284 fput(filp); 1285 return err; 1286 } 1287 1288 static int loop_clr_fd(struct loop_device *lo) 1289 { 1290 int err; 1291 1292 err = mutex_lock_killable(&loop_ctl_mutex); 1293 if (err) 1294 return err; 1295 if (lo->lo_state != Lo_bound) { 1296 mutex_unlock(&loop_ctl_mutex); 1297 return -ENXIO; 1298 } 1299 /* 1300 * If we've explicitly asked to tear down the loop device, 1301 * and it has an elevated reference count, set it for auto-teardown when 1302 * the last reference goes away. This stops $!~#$@ udev from 1303 * preventing teardown because it decided that it needs to run blkid on 1304 * the loopback device whenever they appear. xfstests is notorious for 1305 * failing tests because blkid via udev races with a losetup 1306 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1307 * command to fail with EBUSY. 1308 */ 1309 if (atomic_read(&lo->lo_refcnt) > 1) { 1310 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1311 mutex_unlock(&loop_ctl_mutex); 1312 return 0; 1313 } 1314 lo->lo_state = Lo_rundown; 1315 mutex_unlock(&loop_ctl_mutex); 1316 1317 return __loop_clr_fd(lo, false); 1318 } 1319 1320 static int 1321 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1322 { 1323 int err; 1324 struct block_device *bdev; 1325 kuid_t uid = current_uid(); 1326 bool partscan = false; 1327 bool size_changed = false; 1328 1329 err = mutex_lock_killable(&loop_ctl_mutex); 1330 if (err) 1331 return err; 1332 if (lo->lo_encrypt_key_size && 1333 !uid_eq(lo->lo_key_owner, uid) && 1334 !capable(CAP_SYS_ADMIN)) { 1335 err = -EPERM; 1336 goto out_unlock; 1337 } 1338 if (lo->lo_state != Lo_bound) { 1339 err = -ENXIO; 1340 goto out_unlock; 1341 } 1342 1343 if (lo->lo_offset != info->lo_offset || 1344 lo->lo_sizelimit != info->lo_sizelimit) { 1345 size_changed = true; 1346 sync_blockdev(lo->lo_device); 1347 kill_bdev(lo->lo_device); 1348 } 1349 1350 /* I/O need to be drained during transfer transition */ 1351 blk_mq_freeze_queue(lo->lo_queue); 1352 1353 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1354 /* If any pages were dirtied after kill_bdev(), try again */ 1355 err = -EAGAIN; 1356 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1357 __func__, lo->lo_number, lo->lo_file_name, 1358 lo->lo_device->bd_inode->i_mapping->nrpages); 1359 goto out_unfreeze; 1360 } 1361 1362 err = loop_set_status_from_info(lo, info); 1363 if (err) 1364 goto out_unfreeze; 1365 1366 if (size_changed) { 1367 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1368 lo->lo_backing_file); 1369 loop_set_size(lo, new_size); 1370 } 1371 1372 loop_config_discard(lo); 1373 1374 /* update dio if lo_offset or transfer is changed */ 1375 __loop_update_dio(lo, lo->use_dio); 1376 1377 out_unfreeze: 1378 blk_mq_unfreeze_queue(lo->lo_queue); 1379 1380 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1381 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1382 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1383 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1384 bdev = lo->lo_device; 1385 partscan = true; 1386 } 1387 out_unlock: 1388 mutex_unlock(&loop_ctl_mutex); 1389 if (partscan) 1390 loop_reread_partitions(lo, bdev); 1391 1392 return err; 1393 } 1394 1395 static int 1396 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1397 { 1398 struct path path; 1399 struct kstat stat; 1400 int ret; 1401 1402 ret = mutex_lock_killable(&loop_ctl_mutex); 1403 if (ret) 1404 return ret; 1405 if (lo->lo_state != Lo_bound) { 1406 mutex_unlock(&loop_ctl_mutex); 1407 return -ENXIO; 1408 } 1409 1410 memset(info, 0, sizeof(*info)); 1411 info->lo_number = lo->lo_number; 1412 info->lo_offset = lo->lo_offset; 1413 info->lo_sizelimit = lo->lo_sizelimit; 1414 info->lo_flags = lo->lo_flags; 1415 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1416 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1417 info->lo_encrypt_type = 1418 lo->lo_encryption ? lo->lo_encryption->number : 0; 1419 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1420 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1421 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1422 lo->lo_encrypt_key_size); 1423 } 1424 1425 /* Drop loop_ctl_mutex while we call into the filesystem. */ 1426 path = lo->lo_backing_file->f_path; 1427 path_get(&path); 1428 mutex_unlock(&loop_ctl_mutex); 1429 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1430 if (!ret) { 1431 info->lo_device = huge_encode_dev(stat.dev); 1432 info->lo_inode = stat.ino; 1433 info->lo_rdevice = huge_encode_dev(stat.rdev); 1434 } 1435 path_put(&path); 1436 return ret; 1437 } 1438 1439 static void 1440 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1441 { 1442 memset(info64, 0, sizeof(*info64)); 1443 info64->lo_number = info->lo_number; 1444 info64->lo_device = info->lo_device; 1445 info64->lo_inode = info->lo_inode; 1446 info64->lo_rdevice = info->lo_rdevice; 1447 info64->lo_offset = info->lo_offset; 1448 info64->lo_sizelimit = 0; 1449 info64->lo_encrypt_type = info->lo_encrypt_type; 1450 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1451 info64->lo_flags = info->lo_flags; 1452 info64->lo_init[0] = info->lo_init[0]; 1453 info64->lo_init[1] = info->lo_init[1]; 1454 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1455 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1456 else 1457 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1458 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1459 } 1460 1461 static int 1462 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1463 { 1464 memset(info, 0, sizeof(*info)); 1465 info->lo_number = info64->lo_number; 1466 info->lo_device = info64->lo_device; 1467 info->lo_inode = info64->lo_inode; 1468 info->lo_rdevice = info64->lo_rdevice; 1469 info->lo_offset = info64->lo_offset; 1470 info->lo_encrypt_type = info64->lo_encrypt_type; 1471 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1472 info->lo_flags = info64->lo_flags; 1473 info->lo_init[0] = info64->lo_init[0]; 1474 info->lo_init[1] = info64->lo_init[1]; 1475 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1476 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1477 else 1478 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1479 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1480 1481 /* error in case values were truncated */ 1482 if (info->lo_device != info64->lo_device || 1483 info->lo_rdevice != info64->lo_rdevice || 1484 info->lo_inode != info64->lo_inode || 1485 info->lo_offset != info64->lo_offset) 1486 return -EOVERFLOW; 1487 1488 return 0; 1489 } 1490 1491 static int 1492 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1493 { 1494 struct loop_info info; 1495 struct loop_info64 info64; 1496 1497 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1498 return -EFAULT; 1499 loop_info64_from_old(&info, &info64); 1500 return loop_set_status(lo, &info64); 1501 } 1502 1503 static int 1504 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1505 { 1506 struct loop_info64 info64; 1507 1508 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1509 return -EFAULT; 1510 return loop_set_status(lo, &info64); 1511 } 1512 1513 static int 1514 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1515 struct loop_info info; 1516 struct loop_info64 info64; 1517 int err; 1518 1519 if (!arg) 1520 return -EINVAL; 1521 err = loop_get_status(lo, &info64); 1522 if (!err) 1523 err = loop_info64_to_old(&info64, &info); 1524 if (!err && copy_to_user(arg, &info, sizeof(info))) 1525 err = -EFAULT; 1526 1527 return err; 1528 } 1529 1530 static int 1531 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1532 struct loop_info64 info64; 1533 int err; 1534 1535 if (!arg) 1536 return -EINVAL; 1537 err = loop_get_status(lo, &info64); 1538 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1539 err = -EFAULT; 1540 1541 return err; 1542 } 1543 1544 static int loop_set_capacity(struct loop_device *lo) 1545 { 1546 loff_t size; 1547 1548 if (unlikely(lo->lo_state != Lo_bound)) 1549 return -ENXIO; 1550 1551 size = get_loop_size(lo, lo->lo_backing_file); 1552 loop_set_size(lo, size); 1553 1554 return 0; 1555 } 1556 1557 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1558 { 1559 int error = -ENXIO; 1560 if (lo->lo_state != Lo_bound) 1561 goto out; 1562 1563 __loop_update_dio(lo, !!arg); 1564 if (lo->use_dio == !!arg) 1565 return 0; 1566 error = -EINVAL; 1567 out: 1568 return error; 1569 } 1570 1571 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1572 { 1573 int err = 0; 1574 1575 if (lo->lo_state != Lo_bound) 1576 return -ENXIO; 1577 1578 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1579 return -EINVAL; 1580 1581 if (lo->lo_queue->limits.logical_block_size == arg) 1582 return 0; 1583 1584 sync_blockdev(lo->lo_device); 1585 kill_bdev(lo->lo_device); 1586 1587 blk_mq_freeze_queue(lo->lo_queue); 1588 1589 /* kill_bdev should have truncated all the pages */ 1590 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1591 err = -EAGAIN; 1592 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1593 __func__, lo->lo_number, lo->lo_file_name, 1594 lo->lo_device->bd_inode->i_mapping->nrpages); 1595 goto out_unfreeze; 1596 } 1597 1598 blk_queue_logical_block_size(lo->lo_queue, arg); 1599 blk_queue_physical_block_size(lo->lo_queue, arg); 1600 blk_queue_io_min(lo->lo_queue, arg); 1601 loop_update_dio(lo); 1602 out_unfreeze: 1603 blk_mq_unfreeze_queue(lo->lo_queue); 1604 1605 return err; 1606 } 1607 1608 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1609 unsigned long arg) 1610 { 1611 int err; 1612 1613 err = mutex_lock_killable(&loop_ctl_mutex); 1614 if (err) 1615 return err; 1616 switch (cmd) { 1617 case LOOP_SET_CAPACITY: 1618 err = loop_set_capacity(lo); 1619 break; 1620 case LOOP_SET_DIRECT_IO: 1621 err = loop_set_dio(lo, arg); 1622 break; 1623 case LOOP_SET_BLOCK_SIZE: 1624 err = loop_set_block_size(lo, arg); 1625 break; 1626 default: 1627 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1628 } 1629 mutex_unlock(&loop_ctl_mutex); 1630 return err; 1631 } 1632 1633 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1634 unsigned int cmd, unsigned long arg) 1635 { 1636 struct loop_device *lo = bdev->bd_disk->private_data; 1637 void __user *argp = (void __user *) arg; 1638 int err; 1639 1640 switch (cmd) { 1641 case LOOP_SET_FD: 1642 return loop_set_fd(lo, mode, bdev, arg); 1643 case LOOP_CHANGE_FD: 1644 return loop_change_fd(lo, bdev, arg); 1645 case LOOP_CLR_FD: 1646 return loop_clr_fd(lo); 1647 case LOOP_SET_STATUS: 1648 err = -EPERM; 1649 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1650 err = loop_set_status_old(lo, argp); 1651 } 1652 break; 1653 case LOOP_GET_STATUS: 1654 return loop_get_status_old(lo, argp); 1655 case LOOP_SET_STATUS64: 1656 err = -EPERM; 1657 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1658 err = loop_set_status64(lo, argp); 1659 } 1660 break; 1661 case LOOP_GET_STATUS64: 1662 return loop_get_status64(lo, argp); 1663 case LOOP_SET_CAPACITY: 1664 case LOOP_SET_DIRECT_IO: 1665 case LOOP_SET_BLOCK_SIZE: 1666 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1667 return -EPERM; 1668 /* Fall through */ 1669 default: 1670 err = lo_simple_ioctl(lo, cmd, arg); 1671 break; 1672 } 1673 1674 return err; 1675 } 1676 1677 #ifdef CONFIG_COMPAT 1678 struct compat_loop_info { 1679 compat_int_t lo_number; /* ioctl r/o */ 1680 compat_dev_t lo_device; /* ioctl r/o */ 1681 compat_ulong_t lo_inode; /* ioctl r/o */ 1682 compat_dev_t lo_rdevice; /* ioctl r/o */ 1683 compat_int_t lo_offset; 1684 compat_int_t lo_encrypt_type; 1685 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1686 compat_int_t lo_flags; /* ioctl r/o */ 1687 char lo_name[LO_NAME_SIZE]; 1688 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1689 compat_ulong_t lo_init[2]; 1690 char reserved[4]; 1691 }; 1692 1693 /* 1694 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1695 * - noinlined to reduce stack space usage in main part of driver 1696 */ 1697 static noinline int 1698 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1699 struct loop_info64 *info64) 1700 { 1701 struct compat_loop_info info; 1702 1703 if (copy_from_user(&info, arg, sizeof(info))) 1704 return -EFAULT; 1705 1706 memset(info64, 0, sizeof(*info64)); 1707 info64->lo_number = info.lo_number; 1708 info64->lo_device = info.lo_device; 1709 info64->lo_inode = info.lo_inode; 1710 info64->lo_rdevice = info.lo_rdevice; 1711 info64->lo_offset = info.lo_offset; 1712 info64->lo_sizelimit = 0; 1713 info64->lo_encrypt_type = info.lo_encrypt_type; 1714 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1715 info64->lo_flags = info.lo_flags; 1716 info64->lo_init[0] = info.lo_init[0]; 1717 info64->lo_init[1] = info.lo_init[1]; 1718 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1719 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1720 else 1721 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1722 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1723 return 0; 1724 } 1725 1726 /* 1727 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1728 * - noinlined to reduce stack space usage in main part of driver 1729 */ 1730 static noinline int 1731 loop_info64_to_compat(const struct loop_info64 *info64, 1732 struct compat_loop_info __user *arg) 1733 { 1734 struct compat_loop_info info; 1735 1736 memset(&info, 0, sizeof(info)); 1737 info.lo_number = info64->lo_number; 1738 info.lo_device = info64->lo_device; 1739 info.lo_inode = info64->lo_inode; 1740 info.lo_rdevice = info64->lo_rdevice; 1741 info.lo_offset = info64->lo_offset; 1742 info.lo_encrypt_type = info64->lo_encrypt_type; 1743 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1744 info.lo_flags = info64->lo_flags; 1745 info.lo_init[0] = info64->lo_init[0]; 1746 info.lo_init[1] = info64->lo_init[1]; 1747 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1748 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1749 else 1750 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1751 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1752 1753 /* error in case values were truncated */ 1754 if (info.lo_device != info64->lo_device || 1755 info.lo_rdevice != info64->lo_rdevice || 1756 info.lo_inode != info64->lo_inode || 1757 info.lo_offset != info64->lo_offset || 1758 info.lo_init[0] != info64->lo_init[0] || 1759 info.lo_init[1] != info64->lo_init[1]) 1760 return -EOVERFLOW; 1761 1762 if (copy_to_user(arg, &info, sizeof(info))) 1763 return -EFAULT; 1764 return 0; 1765 } 1766 1767 static int 1768 loop_set_status_compat(struct loop_device *lo, 1769 const struct compat_loop_info __user *arg) 1770 { 1771 struct loop_info64 info64; 1772 int ret; 1773 1774 ret = loop_info64_from_compat(arg, &info64); 1775 if (ret < 0) 1776 return ret; 1777 return loop_set_status(lo, &info64); 1778 } 1779 1780 static int 1781 loop_get_status_compat(struct loop_device *lo, 1782 struct compat_loop_info __user *arg) 1783 { 1784 struct loop_info64 info64; 1785 int err; 1786 1787 if (!arg) 1788 return -EINVAL; 1789 err = loop_get_status(lo, &info64); 1790 if (!err) 1791 err = loop_info64_to_compat(&info64, arg); 1792 return err; 1793 } 1794 1795 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1796 unsigned int cmd, unsigned long arg) 1797 { 1798 struct loop_device *lo = bdev->bd_disk->private_data; 1799 int err; 1800 1801 switch(cmd) { 1802 case LOOP_SET_STATUS: 1803 err = loop_set_status_compat(lo, 1804 (const struct compat_loop_info __user *)arg); 1805 break; 1806 case LOOP_GET_STATUS: 1807 err = loop_get_status_compat(lo, 1808 (struct compat_loop_info __user *)arg); 1809 break; 1810 case LOOP_SET_CAPACITY: 1811 case LOOP_CLR_FD: 1812 case LOOP_GET_STATUS64: 1813 case LOOP_SET_STATUS64: 1814 arg = (unsigned long) compat_ptr(arg); 1815 /* fall through */ 1816 case LOOP_SET_FD: 1817 case LOOP_CHANGE_FD: 1818 case LOOP_SET_BLOCK_SIZE: 1819 case LOOP_SET_DIRECT_IO: 1820 err = lo_ioctl(bdev, mode, cmd, arg); 1821 break; 1822 default: 1823 err = -ENOIOCTLCMD; 1824 break; 1825 } 1826 return err; 1827 } 1828 #endif 1829 1830 static int lo_open(struct block_device *bdev, fmode_t mode) 1831 { 1832 struct loop_device *lo; 1833 int err; 1834 1835 err = mutex_lock_killable(&loop_ctl_mutex); 1836 if (err) 1837 return err; 1838 lo = bdev->bd_disk->private_data; 1839 if (!lo) { 1840 err = -ENXIO; 1841 goto out; 1842 } 1843 1844 atomic_inc(&lo->lo_refcnt); 1845 out: 1846 mutex_unlock(&loop_ctl_mutex); 1847 return err; 1848 } 1849 1850 static void lo_release(struct gendisk *disk, fmode_t mode) 1851 { 1852 struct loop_device *lo; 1853 1854 mutex_lock(&loop_ctl_mutex); 1855 lo = disk->private_data; 1856 if (atomic_dec_return(&lo->lo_refcnt)) 1857 goto out_unlock; 1858 1859 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1860 if (lo->lo_state != Lo_bound) 1861 goto out_unlock; 1862 lo->lo_state = Lo_rundown; 1863 mutex_unlock(&loop_ctl_mutex); 1864 /* 1865 * In autoclear mode, stop the loop thread 1866 * and remove configuration after last close. 1867 */ 1868 __loop_clr_fd(lo, true); 1869 return; 1870 } else if (lo->lo_state == Lo_bound) { 1871 /* 1872 * Otherwise keep thread (if running) and config, 1873 * but flush possible ongoing bios in thread. 1874 */ 1875 blk_mq_freeze_queue(lo->lo_queue); 1876 blk_mq_unfreeze_queue(lo->lo_queue); 1877 } 1878 1879 out_unlock: 1880 mutex_unlock(&loop_ctl_mutex); 1881 } 1882 1883 static const struct block_device_operations lo_fops = { 1884 .owner = THIS_MODULE, 1885 .open = lo_open, 1886 .release = lo_release, 1887 .ioctl = lo_ioctl, 1888 #ifdef CONFIG_COMPAT 1889 .compat_ioctl = lo_compat_ioctl, 1890 #endif 1891 }; 1892 1893 /* 1894 * And now the modules code and kernel interface. 1895 */ 1896 static int max_loop; 1897 module_param(max_loop, int, 0444); 1898 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1899 module_param(max_part, int, 0444); 1900 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1901 MODULE_LICENSE("GPL"); 1902 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1903 1904 int loop_register_transfer(struct loop_func_table *funcs) 1905 { 1906 unsigned int n = funcs->number; 1907 1908 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1909 return -EINVAL; 1910 xfer_funcs[n] = funcs; 1911 return 0; 1912 } 1913 1914 static int unregister_transfer_cb(int id, void *ptr, void *data) 1915 { 1916 struct loop_device *lo = ptr; 1917 struct loop_func_table *xfer = data; 1918 1919 mutex_lock(&loop_ctl_mutex); 1920 if (lo->lo_encryption == xfer) 1921 loop_release_xfer(lo); 1922 mutex_unlock(&loop_ctl_mutex); 1923 return 0; 1924 } 1925 1926 int loop_unregister_transfer(int number) 1927 { 1928 unsigned int n = number; 1929 struct loop_func_table *xfer; 1930 1931 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1932 return -EINVAL; 1933 1934 xfer_funcs[n] = NULL; 1935 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1936 return 0; 1937 } 1938 1939 EXPORT_SYMBOL(loop_register_transfer); 1940 EXPORT_SYMBOL(loop_unregister_transfer); 1941 1942 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1943 const struct blk_mq_queue_data *bd) 1944 { 1945 struct request *rq = bd->rq; 1946 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1947 struct loop_device *lo = rq->q->queuedata; 1948 1949 blk_mq_start_request(rq); 1950 1951 if (lo->lo_state != Lo_bound) 1952 return BLK_STS_IOERR; 1953 1954 switch (req_op(rq)) { 1955 case REQ_OP_FLUSH: 1956 case REQ_OP_DISCARD: 1957 case REQ_OP_WRITE_ZEROES: 1958 cmd->use_aio = false; 1959 break; 1960 default: 1961 cmd->use_aio = lo->use_dio; 1962 break; 1963 } 1964 1965 /* always use the first bio's css */ 1966 #ifdef CONFIG_BLK_CGROUP 1967 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 1968 cmd->css = &bio_blkcg(rq->bio)->css; 1969 css_get(cmd->css); 1970 } else 1971 #endif 1972 cmd->css = NULL; 1973 kthread_queue_work(&lo->worker, &cmd->work); 1974 1975 return BLK_STS_OK; 1976 } 1977 1978 static void loop_handle_cmd(struct loop_cmd *cmd) 1979 { 1980 struct request *rq = blk_mq_rq_from_pdu(cmd); 1981 const bool write = op_is_write(req_op(rq)); 1982 struct loop_device *lo = rq->q->queuedata; 1983 int ret = 0; 1984 1985 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1986 ret = -EIO; 1987 goto failed; 1988 } 1989 1990 ret = do_req_filebacked(lo, rq); 1991 failed: 1992 /* complete non-aio request */ 1993 if (!cmd->use_aio || ret) { 1994 if (ret == -EOPNOTSUPP) 1995 cmd->ret = ret; 1996 else 1997 cmd->ret = ret ? -EIO : 0; 1998 blk_mq_complete_request(rq); 1999 } 2000 } 2001 2002 static void loop_queue_work(struct kthread_work *work) 2003 { 2004 struct loop_cmd *cmd = 2005 container_of(work, struct loop_cmd, work); 2006 2007 loop_handle_cmd(cmd); 2008 } 2009 2010 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 2011 unsigned int hctx_idx, unsigned int numa_node) 2012 { 2013 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2014 2015 kthread_init_work(&cmd->work, loop_queue_work); 2016 return 0; 2017 } 2018 2019 static const struct blk_mq_ops loop_mq_ops = { 2020 .queue_rq = loop_queue_rq, 2021 .init_request = loop_init_request, 2022 .complete = lo_complete_rq, 2023 }; 2024 2025 static int loop_add(struct loop_device **l, int i) 2026 { 2027 struct loop_device *lo; 2028 struct gendisk *disk; 2029 int err; 2030 2031 err = -ENOMEM; 2032 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 2033 if (!lo) 2034 goto out; 2035 2036 lo->lo_state = Lo_unbound; 2037 2038 /* allocate id, if @id >= 0, we're requesting that specific id */ 2039 if (i >= 0) { 2040 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2041 if (err == -ENOSPC) 2042 err = -EEXIST; 2043 } else { 2044 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2045 } 2046 if (err < 0) 2047 goto out_free_dev; 2048 i = err; 2049 2050 err = -ENOMEM; 2051 lo->tag_set.ops = &loop_mq_ops; 2052 lo->tag_set.nr_hw_queues = 1; 2053 lo->tag_set.queue_depth = 128; 2054 lo->tag_set.numa_node = NUMA_NO_NODE; 2055 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2056 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; 2057 lo->tag_set.driver_data = lo; 2058 2059 err = blk_mq_alloc_tag_set(&lo->tag_set); 2060 if (err) 2061 goto out_free_idr; 2062 2063 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 2064 if (IS_ERR(lo->lo_queue)) { 2065 err = PTR_ERR(lo->lo_queue); 2066 goto out_cleanup_tags; 2067 } 2068 lo->lo_queue->queuedata = lo; 2069 2070 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2071 2072 /* 2073 * By default, we do buffer IO, so it doesn't make sense to enable 2074 * merge because the I/O submitted to backing file is handled page by 2075 * page. For directio mode, merge does help to dispatch bigger request 2076 * to underlayer disk. We will enable merge once directio is enabled. 2077 */ 2078 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2079 2080 err = -ENOMEM; 2081 disk = lo->lo_disk = alloc_disk(1 << part_shift); 2082 if (!disk) 2083 goto out_free_queue; 2084 2085 /* 2086 * Disable partition scanning by default. The in-kernel partition 2087 * scanning can be requested individually per-device during its 2088 * setup. Userspace can always add and remove partitions from all 2089 * devices. The needed partition minors are allocated from the 2090 * extended minor space, the main loop device numbers will continue 2091 * to match the loop minors, regardless of the number of partitions 2092 * used. 2093 * 2094 * If max_part is given, partition scanning is globally enabled for 2095 * all loop devices. The minors for the main loop devices will be 2096 * multiples of max_part. 2097 * 2098 * Note: Global-for-all-devices, set-only-at-init, read-only module 2099 * parameteters like 'max_loop' and 'max_part' make things needlessly 2100 * complicated, are too static, inflexible and may surprise 2101 * userspace tools. Parameters like this in general should be avoided. 2102 */ 2103 if (!part_shift) 2104 disk->flags |= GENHD_FL_NO_PART_SCAN; 2105 disk->flags |= GENHD_FL_EXT_DEVT; 2106 atomic_set(&lo->lo_refcnt, 0); 2107 lo->lo_number = i; 2108 spin_lock_init(&lo->lo_lock); 2109 disk->major = LOOP_MAJOR; 2110 disk->first_minor = i << part_shift; 2111 disk->fops = &lo_fops; 2112 disk->private_data = lo; 2113 disk->queue = lo->lo_queue; 2114 sprintf(disk->disk_name, "loop%d", i); 2115 add_disk(disk); 2116 *l = lo; 2117 return lo->lo_number; 2118 2119 out_free_queue: 2120 blk_cleanup_queue(lo->lo_queue); 2121 out_cleanup_tags: 2122 blk_mq_free_tag_set(&lo->tag_set); 2123 out_free_idr: 2124 idr_remove(&loop_index_idr, i); 2125 out_free_dev: 2126 kfree(lo); 2127 out: 2128 return err; 2129 } 2130 2131 static void loop_remove(struct loop_device *lo) 2132 { 2133 del_gendisk(lo->lo_disk); 2134 blk_cleanup_queue(lo->lo_queue); 2135 blk_mq_free_tag_set(&lo->tag_set); 2136 put_disk(lo->lo_disk); 2137 kfree(lo); 2138 } 2139 2140 static int find_free_cb(int id, void *ptr, void *data) 2141 { 2142 struct loop_device *lo = ptr; 2143 struct loop_device **l = data; 2144 2145 if (lo->lo_state == Lo_unbound) { 2146 *l = lo; 2147 return 1; 2148 } 2149 return 0; 2150 } 2151 2152 static int loop_lookup(struct loop_device **l, int i) 2153 { 2154 struct loop_device *lo; 2155 int ret = -ENODEV; 2156 2157 if (i < 0) { 2158 int err; 2159 2160 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 2161 if (err == 1) { 2162 *l = lo; 2163 ret = lo->lo_number; 2164 } 2165 goto out; 2166 } 2167 2168 /* lookup and return a specific i */ 2169 lo = idr_find(&loop_index_idr, i); 2170 if (lo) { 2171 *l = lo; 2172 ret = lo->lo_number; 2173 } 2174 out: 2175 return ret; 2176 } 2177 2178 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 2179 { 2180 struct loop_device *lo; 2181 struct kobject *kobj; 2182 int err; 2183 2184 mutex_lock(&loop_ctl_mutex); 2185 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 2186 if (err < 0) 2187 err = loop_add(&lo, MINOR(dev) >> part_shift); 2188 if (err < 0) 2189 kobj = NULL; 2190 else 2191 kobj = get_disk_and_module(lo->lo_disk); 2192 mutex_unlock(&loop_ctl_mutex); 2193 2194 *part = 0; 2195 return kobj; 2196 } 2197 2198 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2199 unsigned long parm) 2200 { 2201 struct loop_device *lo; 2202 int ret; 2203 2204 ret = mutex_lock_killable(&loop_ctl_mutex); 2205 if (ret) 2206 return ret; 2207 2208 ret = -ENOSYS; 2209 switch (cmd) { 2210 case LOOP_CTL_ADD: 2211 ret = loop_lookup(&lo, parm); 2212 if (ret >= 0) { 2213 ret = -EEXIST; 2214 break; 2215 } 2216 ret = loop_add(&lo, parm); 2217 break; 2218 case LOOP_CTL_REMOVE: 2219 ret = loop_lookup(&lo, parm); 2220 if (ret < 0) 2221 break; 2222 if (lo->lo_state != Lo_unbound) { 2223 ret = -EBUSY; 2224 break; 2225 } 2226 if (atomic_read(&lo->lo_refcnt) > 0) { 2227 ret = -EBUSY; 2228 break; 2229 } 2230 lo->lo_disk->private_data = NULL; 2231 idr_remove(&loop_index_idr, lo->lo_number); 2232 loop_remove(lo); 2233 break; 2234 case LOOP_CTL_GET_FREE: 2235 ret = loop_lookup(&lo, -1); 2236 if (ret >= 0) 2237 break; 2238 ret = loop_add(&lo, -1); 2239 } 2240 mutex_unlock(&loop_ctl_mutex); 2241 2242 return ret; 2243 } 2244 2245 static const struct file_operations loop_ctl_fops = { 2246 .open = nonseekable_open, 2247 .unlocked_ioctl = loop_control_ioctl, 2248 .compat_ioctl = loop_control_ioctl, 2249 .owner = THIS_MODULE, 2250 .llseek = noop_llseek, 2251 }; 2252 2253 static struct miscdevice loop_misc = { 2254 .minor = LOOP_CTRL_MINOR, 2255 .name = "loop-control", 2256 .fops = &loop_ctl_fops, 2257 }; 2258 2259 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2260 MODULE_ALIAS("devname:loop-control"); 2261 2262 static int __init loop_init(void) 2263 { 2264 int i, nr; 2265 unsigned long range; 2266 struct loop_device *lo; 2267 int err; 2268 2269 part_shift = 0; 2270 if (max_part > 0) { 2271 part_shift = fls(max_part); 2272 2273 /* 2274 * Adjust max_part according to part_shift as it is exported 2275 * to user space so that user can decide correct minor number 2276 * if [s]he want to create more devices. 2277 * 2278 * Note that -1 is required because partition 0 is reserved 2279 * for the whole disk. 2280 */ 2281 max_part = (1UL << part_shift) - 1; 2282 } 2283 2284 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2285 err = -EINVAL; 2286 goto err_out; 2287 } 2288 2289 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2290 err = -EINVAL; 2291 goto err_out; 2292 } 2293 2294 /* 2295 * If max_loop is specified, create that many devices upfront. 2296 * This also becomes a hard limit. If max_loop is not specified, 2297 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2298 * init time. Loop devices can be requested on-demand with the 2299 * /dev/loop-control interface, or be instantiated by accessing 2300 * a 'dead' device node. 2301 */ 2302 if (max_loop) { 2303 nr = max_loop; 2304 range = max_loop << part_shift; 2305 } else { 2306 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2307 range = 1UL << MINORBITS; 2308 } 2309 2310 err = misc_register(&loop_misc); 2311 if (err < 0) 2312 goto err_out; 2313 2314 2315 if (register_blkdev(LOOP_MAJOR, "loop")) { 2316 err = -EIO; 2317 goto misc_out; 2318 } 2319 2320 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2321 THIS_MODULE, loop_probe, NULL, NULL); 2322 2323 /* pre-create number of devices given by config or max_loop */ 2324 mutex_lock(&loop_ctl_mutex); 2325 for (i = 0; i < nr; i++) 2326 loop_add(&lo, i); 2327 mutex_unlock(&loop_ctl_mutex); 2328 2329 printk(KERN_INFO "loop: module loaded\n"); 2330 return 0; 2331 2332 misc_out: 2333 misc_deregister(&loop_misc); 2334 err_out: 2335 return err; 2336 } 2337 2338 static int loop_exit_cb(int id, void *ptr, void *data) 2339 { 2340 struct loop_device *lo = ptr; 2341 2342 loop_remove(lo); 2343 return 0; 2344 } 2345 2346 static void __exit loop_exit(void) 2347 { 2348 unsigned long range; 2349 2350 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2351 2352 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2353 idr_destroy(&loop_index_idr); 2354 2355 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2356 unregister_blkdev(LOOP_MAJOR, "loop"); 2357 2358 misc_deregister(&loop_misc); 2359 } 2360 2361 module_init(loop_init); 2362 module_exit(loop_exit); 2363 2364 #ifndef MODULE 2365 static int __init max_loop_setup(char *str) 2366 { 2367 max_loop = simple_strtol(str, NULL, 0); 2368 return 1; 2369 } 2370 2371 __setup("max_loop=", max_loop_setup); 2372 #endif 2373