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_validate_block_size() - validates the passed in block size 233 * @bsize: size to validate 234 */ 235 static int 236 loop_validate_block_size(unsigned short bsize) 237 { 238 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize)) 239 return -EINVAL; 240 241 return 0; 242 } 243 244 /** 245 * loop_set_size() - sets device size and notifies userspace 246 * @lo: struct loop_device to set the size for 247 * @size: new size of the loop device 248 * 249 * Callers must validate that the size passed into this function fits into 250 * a sector_t, eg using loop_validate_size() 251 */ 252 static void loop_set_size(struct loop_device *lo, loff_t size) 253 { 254 if (!set_capacity_and_notify(lo->lo_disk, size)) 255 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 256 } 257 258 static inline int 259 lo_do_transfer(struct loop_device *lo, int cmd, 260 struct page *rpage, unsigned roffs, 261 struct page *lpage, unsigned loffs, 262 int size, sector_t rblock) 263 { 264 int ret; 265 266 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 267 if (likely(!ret)) 268 return 0; 269 270 printk_ratelimited(KERN_ERR 271 "loop: Transfer error at byte offset %llu, length %i.\n", 272 (unsigned long long)rblock << 9, size); 273 return ret; 274 } 275 276 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 277 { 278 struct iov_iter i; 279 ssize_t bw; 280 281 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 282 283 file_start_write(file); 284 bw = vfs_iter_write(file, &i, ppos, 0); 285 file_end_write(file); 286 287 if (likely(bw == bvec->bv_len)) 288 return 0; 289 290 printk_ratelimited(KERN_ERR 291 "loop: Write error at byte offset %llu, length %i.\n", 292 (unsigned long long)*ppos, bvec->bv_len); 293 if (bw >= 0) 294 bw = -EIO; 295 return bw; 296 } 297 298 static int lo_write_simple(struct loop_device *lo, struct request *rq, 299 loff_t pos) 300 { 301 struct bio_vec bvec; 302 struct req_iterator iter; 303 int ret = 0; 304 305 rq_for_each_segment(bvec, rq, iter) { 306 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 307 if (ret < 0) 308 break; 309 cond_resched(); 310 } 311 312 return ret; 313 } 314 315 /* 316 * This is the slow, transforming version that needs to double buffer the 317 * data as it cannot do the transformations in place without having direct 318 * access to the destination pages of the backing file. 319 */ 320 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 321 loff_t pos) 322 { 323 struct bio_vec bvec, b; 324 struct req_iterator iter; 325 struct page *page; 326 int ret = 0; 327 328 page = alloc_page(GFP_NOIO); 329 if (unlikely(!page)) 330 return -ENOMEM; 331 332 rq_for_each_segment(bvec, rq, iter) { 333 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 334 bvec.bv_offset, bvec.bv_len, pos >> 9); 335 if (unlikely(ret)) 336 break; 337 338 b.bv_page = page; 339 b.bv_offset = 0; 340 b.bv_len = bvec.bv_len; 341 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 342 if (ret < 0) 343 break; 344 } 345 346 __free_page(page); 347 return ret; 348 } 349 350 static int lo_read_simple(struct loop_device *lo, struct request *rq, 351 loff_t pos) 352 { 353 struct bio_vec bvec; 354 struct req_iterator iter; 355 struct iov_iter i; 356 ssize_t len; 357 358 rq_for_each_segment(bvec, rq, iter) { 359 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 360 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 361 if (len < 0) 362 return len; 363 364 flush_dcache_page(bvec.bv_page); 365 366 if (len != bvec.bv_len) { 367 struct bio *bio; 368 369 __rq_for_each_bio(bio, rq) 370 zero_fill_bio(bio); 371 break; 372 } 373 cond_resched(); 374 } 375 376 return 0; 377 } 378 379 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 380 loff_t pos) 381 { 382 struct bio_vec bvec, b; 383 struct req_iterator iter; 384 struct iov_iter i; 385 struct page *page; 386 ssize_t len; 387 int ret = 0; 388 389 page = alloc_page(GFP_NOIO); 390 if (unlikely(!page)) 391 return -ENOMEM; 392 393 rq_for_each_segment(bvec, rq, iter) { 394 loff_t offset = pos; 395 396 b.bv_page = page; 397 b.bv_offset = 0; 398 b.bv_len = bvec.bv_len; 399 400 iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 401 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 402 if (len < 0) { 403 ret = len; 404 goto out_free_page; 405 } 406 407 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 408 bvec.bv_offset, len, offset >> 9); 409 if (ret) 410 goto out_free_page; 411 412 flush_dcache_page(bvec.bv_page); 413 414 if (len != bvec.bv_len) { 415 struct bio *bio; 416 417 __rq_for_each_bio(bio, rq) 418 zero_fill_bio(bio); 419 break; 420 } 421 } 422 423 ret = 0; 424 out_free_page: 425 __free_page(page); 426 return ret; 427 } 428 429 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 430 int mode) 431 { 432 /* 433 * We use fallocate to manipulate the space mappings used by the image 434 * a.k.a. discard/zerorange. However we do not support this if 435 * encryption is enabled, because it may give an attacker useful 436 * information. 437 */ 438 struct file *file = lo->lo_backing_file; 439 struct request_queue *q = lo->lo_queue; 440 int ret; 441 442 mode |= FALLOC_FL_KEEP_SIZE; 443 444 if (!blk_queue_discard(q)) { 445 ret = -EOPNOTSUPP; 446 goto out; 447 } 448 449 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 450 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 451 ret = -EIO; 452 out: 453 return ret; 454 } 455 456 static int lo_req_flush(struct loop_device *lo, struct request *rq) 457 { 458 struct file *file = lo->lo_backing_file; 459 int ret = vfs_fsync(file, 0); 460 if (unlikely(ret && ret != -EINVAL)) 461 ret = -EIO; 462 463 return ret; 464 } 465 466 static void lo_complete_rq(struct request *rq) 467 { 468 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 469 blk_status_t ret = BLK_STS_OK; 470 471 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 472 req_op(rq) != REQ_OP_READ) { 473 if (cmd->ret < 0) 474 ret = errno_to_blk_status(cmd->ret); 475 goto end_io; 476 } 477 478 /* 479 * Short READ - if we got some data, advance our request and 480 * retry it. If we got no data, end the rest with EIO. 481 */ 482 if (cmd->ret) { 483 blk_update_request(rq, BLK_STS_OK, cmd->ret); 484 cmd->ret = 0; 485 blk_mq_requeue_request(rq, true); 486 } else { 487 if (cmd->use_aio) { 488 struct bio *bio = rq->bio; 489 490 while (bio) { 491 zero_fill_bio(bio); 492 bio = bio->bi_next; 493 } 494 } 495 ret = BLK_STS_IOERR; 496 end_io: 497 blk_mq_end_request(rq, ret); 498 } 499 } 500 501 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 502 { 503 struct request *rq = blk_mq_rq_from_pdu(cmd); 504 505 if (!atomic_dec_and_test(&cmd->ref)) 506 return; 507 kfree(cmd->bvec); 508 cmd->bvec = NULL; 509 if (likely(!blk_should_fake_timeout(rq->q))) 510 blk_mq_complete_request(rq); 511 } 512 513 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 514 { 515 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 516 517 if (cmd->css) 518 css_put(cmd->css); 519 cmd->ret = ret; 520 lo_rw_aio_do_completion(cmd); 521 } 522 523 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 524 loff_t pos, bool rw) 525 { 526 struct iov_iter iter; 527 struct req_iterator rq_iter; 528 struct bio_vec *bvec; 529 struct request *rq = blk_mq_rq_from_pdu(cmd); 530 struct bio *bio = rq->bio; 531 struct file *file = lo->lo_backing_file; 532 struct bio_vec tmp; 533 unsigned int offset; 534 int nr_bvec = 0; 535 int ret; 536 537 rq_for_each_bvec(tmp, rq, rq_iter) 538 nr_bvec++; 539 540 if (rq->bio != rq->biotail) { 541 542 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 543 GFP_NOIO); 544 if (!bvec) 545 return -EIO; 546 cmd->bvec = bvec; 547 548 /* 549 * The bios of the request may be started from the middle of 550 * the 'bvec' because of bio splitting, so we can't directly 551 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 552 * API will take care of all details for us. 553 */ 554 rq_for_each_bvec(tmp, rq, rq_iter) { 555 *bvec = tmp; 556 bvec++; 557 } 558 bvec = cmd->bvec; 559 offset = 0; 560 } else { 561 /* 562 * Same here, this bio may be started from the middle of the 563 * 'bvec' because of bio splitting, so offset from the bvec 564 * must be passed to iov iterator 565 */ 566 offset = bio->bi_iter.bi_bvec_done; 567 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 568 } 569 atomic_set(&cmd->ref, 2); 570 571 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 572 iter.iov_offset = offset; 573 574 cmd->iocb.ki_pos = pos; 575 cmd->iocb.ki_filp = file; 576 cmd->iocb.ki_complete = lo_rw_aio_complete; 577 cmd->iocb.ki_flags = IOCB_DIRECT; 578 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 579 if (cmd->css) 580 kthread_associate_blkcg(cmd->css); 581 582 if (rw == WRITE) 583 ret = call_write_iter(file, &cmd->iocb, &iter); 584 else 585 ret = call_read_iter(file, &cmd->iocb, &iter); 586 587 lo_rw_aio_do_completion(cmd); 588 kthread_associate_blkcg(NULL); 589 590 if (ret != -EIOCBQUEUED) 591 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 592 return 0; 593 } 594 595 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 596 { 597 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 598 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 599 600 /* 601 * lo_write_simple and lo_read_simple should have been covered 602 * by io submit style function like lo_rw_aio(), one blocker 603 * is that lo_read_simple() need to call flush_dcache_page after 604 * the page is written from kernel, and it isn't easy to handle 605 * this in io submit style function which submits all segments 606 * of the req at one time. And direct read IO doesn't need to 607 * run flush_dcache_page(). 608 */ 609 switch (req_op(rq)) { 610 case REQ_OP_FLUSH: 611 return lo_req_flush(lo, rq); 612 case REQ_OP_WRITE_ZEROES: 613 /* 614 * If the caller doesn't want deallocation, call zeroout to 615 * write zeroes the range. Otherwise, punch them out. 616 */ 617 return lo_fallocate(lo, rq, pos, 618 (rq->cmd_flags & REQ_NOUNMAP) ? 619 FALLOC_FL_ZERO_RANGE : 620 FALLOC_FL_PUNCH_HOLE); 621 case REQ_OP_DISCARD: 622 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 623 case REQ_OP_WRITE: 624 if (lo->transfer) 625 return lo_write_transfer(lo, rq, pos); 626 else if (cmd->use_aio) 627 return lo_rw_aio(lo, cmd, pos, WRITE); 628 else 629 return lo_write_simple(lo, rq, pos); 630 case REQ_OP_READ: 631 if (lo->transfer) 632 return lo_read_transfer(lo, rq, pos); 633 else if (cmd->use_aio) 634 return lo_rw_aio(lo, cmd, pos, READ); 635 else 636 return lo_read_simple(lo, rq, pos); 637 default: 638 WARN_ON_ONCE(1); 639 return -EIO; 640 } 641 } 642 643 static inline void loop_update_dio(struct loop_device *lo) 644 { 645 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 646 lo->use_dio); 647 } 648 649 static void loop_reread_partitions(struct loop_device *lo, 650 struct block_device *bdev) 651 { 652 int rc; 653 654 mutex_lock(&bdev->bd_mutex); 655 rc = bdev_disk_changed(bdev, false); 656 mutex_unlock(&bdev->bd_mutex); 657 if (rc) 658 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 659 __func__, lo->lo_number, lo->lo_file_name, rc); 660 } 661 662 static inline int is_loop_device(struct file *file) 663 { 664 struct inode *i = file->f_mapping->host; 665 666 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 667 } 668 669 static int loop_validate_file(struct file *file, struct block_device *bdev) 670 { 671 struct inode *inode = file->f_mapping->host; 672 struct file *f = file; 673 674 /* Avoid recursion */ 675 while (is_loop_device(f)) { 676 struct loop_device *l; 677 678 if (f->f_mapping->host->i_rdev == bdev->bd_dev) 679 return -EBADF; 680 681 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 682 if (l->lo_state != Lo_bound) { 683 return -EINVAL; 684 } 685 f = l->lo_backing_file; 686 } 687 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 688 return -EINVAL; 689 return 0; 690 } 691 692 /* 693 * loop_change_fd switched the backing store of a loopback device to 694 * a new file. This is useful for operating system installers to free up 695 * the original file and in High Availability environments to switch to 696 * an alternative location for the content in case of server meltdown. 697 * This can only work if the loop device is used read-only, and if the 698 * new backing store is the same size and type as the old backing store. 699 */ 700 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 701 unsigned int arg) 702 { 703 struct file *file = NULL, *old_file; 704 int error; 705 bool partscan; 706 707 error = mutex_lock_killable(&loop_ctl_mutex); 708 if (error) 709 return error; 710 error = -ENXIO; 711 if (lo->lo_state != Lo_bound) 712 goto out_err; 713 714 /* the loop device has to be read-only */ 715 error = -EINVAL; 716 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 717 goto out_err; 718 719 error = -EBADF; 720 file = fget(arg); 721 if (!file) 722 goto out_err; 723 724 error = loop_validate_file(file, bdev); 725 if (error) 726 goto out_err; 727 728 old_file = lo->lo_backing_file; 729 730 error = -EINVAL; 731 732 /* size of the new backing store needs to be the same */ 733 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 734 goto out_err; 735 736 /* and ... switch */ 737 blk_mq_freeze_queue(lo->lo_queue); 738 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 739 lo->lo_backing_file = file; 740 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 741 mapping_set_gfp_mask(file->f_mapping, 742 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 743 loop_update_dio(lo); 744 blk_mq_unfreeze_queue(lo->lo_queue); 745 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 746 mutex_unlock(&loop_ctl_mutex); 747 /* 748 * We must drop file reference outside of loop_ctl_mutex as dropping 749 * the file ref can take bd_mutex which creates circular locking 750 * dependency. 751 */ 752 fput(old_file); 753 if (partscan) 754 loop_reread_partitions(lo, bdev); 755 return 0; 756 757 out_err: 758 mutex_unlock(&loop_ctl_mutex); 759 if (file) 760 fput(file); 761 return error; 762 } 763 764 /* loop sysfs attributes */ 765 766 static ssize_t loop_attr_show(struct device *dev, char *page, 767 ssize_t (*callback)(struct loop_device *, char *)) 768 { 769 struct gendisk *disk = dev_to_disk(dev); 770 struct loop_device *lo = disk->private_data; 771 772 return callback(lo, page); 773 } 774 775 #define LOOP_ATTR_RO(_name) \ 776 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 777 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 778 struct device_attribute *attr, char *b) \ 779 { \ 780 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 781 } \ 782 static struct device_attribute loop_attr_##_name = \ 783 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 784 785 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 786 { 787 ssize_t ret; 788 char *p = NULL; 789 790 spin_lock_irq(&lo->lo_lock); 791 if (lo->lo_backing_file) 792 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 793 spin_unlock_irq(&lo->lo_lock); 794 795 if (IS_ERR_OR_NULL(p)) 796 ret = PTR_ERR(p); 797 else { 798 ret = strlen(p); 799 memmove(buf, p, ret); 800 buf[ret++] = '\n'; 801 buf[ret] = 0; 802 } 803 804 return ret; 805 } 806 807 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 808 { 809 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 810 } 811 812 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 813 { 814 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 815 } 816 817 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 818 { 819 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 820 821 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 822 } 823 824 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 825 { 826 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 827 828 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 829 } 830 831 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 832 { 833 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 834 835 return sprintf(buf, "%s\n", dio ? "1" : "0"); 836 } 837 838 LOOP_ATTR_RO(backing_file); 839 LOOP_ATTR_RO(offset); 840 LOOP_ATTR_RO(sizelimit); 841 LOOP_ATTR_RO(autoclear); 842 LOOP_ATTR_RO(partscan); 843 LOOP_ATTR_RO(dio); 844 845 static struct attribute *loop_attrs[] = { 846 &loop_attr_backing_file.attr, 847 &loop_attr_offset.attr, 848 &loop_attr_sizelimit.attr, 849 &loop_attr_autoclear.attr, 850 &loop_attr_partscan.attr, 851 &loop_attr_dio.attr, 852 NULL, 853 }; 854 855 static struct attribute_group loop_attribute_group = { 856 .name = "loop", 857 .attrs= loop_attrs, 858 }; 859 860 static void loop_sysfs_init(struct loop_device *lo) 861 { 862 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 863 &loop_attribute_group); 864 } 865 866 static void loop_sysfs_exit(struct loop_device *lo) 867 { 868 if (lo->sysfs_inited) 869 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 870 &loop_attribute_group); 871 } 872 873 static void loop_config_discard(struct loop_device *lo) 874 { 875 struct file *file = lo->lo_backing_file; 876 struct inode *inode = file->f_mapping->host; 877 struct request_queue *q = lo->lo_queue; 878 u32 granularity, max_discard_sectors; 879 880 /* 881 * If the backing device is a block device, mirror its zeroing 882 * capability. Set the discard sectors to the block device's zeroing 883 * capabilities because loop discards result in blkdev_issue_zeroout(), 884 * not blkdev_issue_discard(). This maintains consistent behavior with 885 * file-backed loop devices: discarded regions read back as zero. 886 */ 887 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { 888 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 889 890 max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 891 granularity = backingq->limits.discard_granularity ?: 892 queue_physical_block_size(backingq); 893 894 /* 895 * We use punch hole to reclaim the free space used by the 896 * image a.k.a. discard. However we do not support discard if 897 * encryption is enabled, because it may give an attacker 898 * useful information. 899 */ 900 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { 901 max_discard_sectors = 0; 902 granularity = 0; 903 904 } else { 905 max_discard_sectors = UINT_MAX >> 9; 906 granularity = inode->i_sb->s_blocksize; 907 } 908 909 if (max_discard_sectors) { 910 q->limits.discard_granularity = granularity; 911 blk_queue_max_discard_sectors(q, max_discard_sectors); 912 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 913 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 914 } else { 915 q->limits.discard_granularity = 0; 916 blk_queue_max_discard_sectors(q, 0); 917 blk_queue_max_write_zeroes_sectors(q, 0); 918 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 919 } 920 q->limits.discard_alignment = 0; 921 } 922 923 static void loop_unprepare_queue(struct loop_device *lo) 924 { 925 kthread_flush_worker(&lo->worker); 926 kthread_stop(lo->worker_task); 927 } 928 929 static int loop_kthread_worker_fn(void *worker_ptr) 930 { 931 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 932 return kthread_worker_fn(worker_ptr); 933 } 934 935 static int loop_prepare_queue(struct loop_device *lo) 936 { 937 kthread_init_worker(&lo->worker); 938 lo->worker_task = kthread_run(loop_kthread_worker_fn, 939 &lo->worker, "loop%d", lo->lo_number); 940 if (IS_ERR(lo->worker_task)) 941 return -ENOMEM; 942 set_user_nice(lo->worker_task, MIN_NICE); 943 return 0; 944 } 945 946 static void loop_update_rotational(struct loop_device *lo) 947 { 948 struct file *file = lo->lo_backing_file; 949 struct inode *file_inode = file->f_mapping->host; 950 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 951 struct request_queue *q = lo->lo_queue; 952 bool nonrot = true; 953 954 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 955 if (file_bdev) 956 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 957 958 if (nonrot) 959 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 960 else 961 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 962 } 963 964 static int 965 loop_release_xfer(struct loop_device *lo) 966 { 967 int err = 0; 968 struct loop_func_table *xfer = lo->lo_encryption; 969 970 if (xfer) { 971 if (xfer->release) 972 err = xfer->release(lo); 973 lo->transfer = NULL; 974 lo->lo_encryption = NULL; 975 module_put(xfer->owner); 976 } 977 return err; 978 } 979 980 static int 981 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 982 const struct loop_info64 *i) 983 { 984 int err = 0; 985 986 if (xfer) { 987 struct module *owner = xfer->owner; 988 989 if (!try_module_get(owner)) 990 return -EINVAL; 991 if (xfer->init) 992 err = xfer->init(lo, i); 993 if (err) 994 module_put(owner); 995 else 996 lo->lo_encryption = xfer; 997 } 998 return err; 999 } 1000 1001 /** 1002 * loop_set_status_from_info - configure device from loop_info 1003 * @lo: struct loop_device to configure 1004 * @info: struct loop_info64 to configure the device with 1005 * 1006 * Configures the loop device parameters according to the passed 1007 * in loop_info64 configuration. 1008 */ 1009 static int 1010 loop_set_status_from_info(struct loop_device *lo, 1011 const struct loop_info64 *info) 1012 { 1013 int err; 1014 struct loop_func_table *xfer; 1015 kuid_t uid = current_uid(); 1016 1017 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1018 return -EINVAL; 1019 1020 err = loop_release_xfer(lo); 1021 if (err) 1022 return err; 1023 1024 if (info->lo_encrypt_type) { 1025 unsigned int type = info->lo_encrypt_type; 1026 1027 if (type >= MAX_LO_CRYPT) 1028 return -EINVAL; 1029 xfer = xfer_funcs[type]; 1030 if (xfer == NULL) 1031 return -EINVAL; 1032 } else 1033 xfer = NULL; 1034 1035 err = loop_init_xfer(lo, xfer, info); 1036 if (err) 1037 return err; 1038 1039 lo->lo_offset = info->lo_offset; 1040 lo->lo_sizelimit = info->lo_sizelimit; 1041 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1042 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1043 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1044 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1045 1046 if (!xfer) 1047 xfer = &none_funcs; 1048 lo->transfer = xfer->transfer; 1049 lo->ioctl = xfer->ioctl; 1050 1051 lo->lo_flags = info->lo_flags; 1052 1053 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1054 lo->lo_init[0] = info->lo_init[0]; 1055 lo->lo_init[1] = info->lo_init[1]; 1056 if (info->lo_encrypt_key_size) { 1057 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1058 info->lo_encrypt_key_size); 1059 lo->lo_key_owner = uid; 1060 } 1061 1062 return 0; 1063 } 1064 1065 static int loop_configure(struct loop_device *lo, fmode_t mode, 1066 struct block_device *bdev, 1067 const struct loop_config *config) 1068 { 1069 struct file *file; 1070 struct inode *inode; 1071 struct address_space *mapping; 1072 struct block_device *claimed_bdev = NULL; 1073 int error; 1074 loff_t size; 1075 bool partscan; 1076 unsigned short bsize; 1077 1078 /* This is safe, since we have a reference from open(). */ 1079 __module_get(THIS_MODULE); 1080 1081 error = -EBADF; 1082 file = fget(config->fd); 1083 if (!file) 1084 goto out; 1085 1086 /* 1087 * If we don't hold exclusive handle for the device, upgrade to it 1088 * here to avoid changing device under exclusive owner. 1089 */ 1090 if (!(mode & FMODE_EXCL)) { 1091 claimed_bdev = bdev->bd_contains; 1092 error = bd_prepare_to_claim(bdev, claimed_bdev, loop_configure); 1093 if (error) 1094 goto out_putf; 1095 } 1096 1097 error = mutex_lock_killable(&loop_ctl_mutex); 1098 if (error) 1099 goto out_bdev; 1100 1101 error = -EBUSY; 1102 if (lo->lo_state != Lo_unbound) 1103 goto out_unlock; 1104 1105 error = loop_validate_file(file, bdev); 1106 if (error) 1107 goto out_unlock; 1108 1109 mapping = file->f_mapping; 1110 inode = mapping->host; 1111 1112 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1113 error = -EINVAL; 1114 goto out_unlock; 1115 } 1116 1117 if (config->block_size) { 1118 error = loop_validate_block_size(config->block_size); 1119 if (error) 1120 goto out_unlock; 1121 } 1122 1123 error = loop_set_status_from_info(lo, &config->info); 1124 if (error) 1125 goto out_unlock; 1126 1127 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1128 !file->f_op->write_iter) 1129 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1130 1131 error = loop_prepare_queue(lo); 1132 if (error) 1133 goto out_unlock; 1134 1135 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1136 1137 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1138 lo->lo_device = bdev; 1139 lo->lo_backing_file = file; 1140 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1141 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1142 1143 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1144 blk_queue_write_cache(lo->lo_queue, true, false); 1145 1146 if (config->block_size) 1147 bsize = config->block_size; 1148 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1149 /* In case of direct I/O, match underlying block size */ 1150 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1151 else 1152 bsize = 512; 1153 1154 blk_queue_logical_block_size(lo->lo_queue, bsize); 1155 blk_queue_physical_block_size(lo->lo_queue, bsize); 1156 blk_queue_io_min(lo->lo_queue, bsize); 1157 1158 loop_update_rotational(lo); 1159 loop_update_dio(lo); 1160 loop_sysfs_init(lo); 1161 1162 size = get_loop_size(lo, file); 1163 loop_set_size(lo, size); 1164 1165 lo->lo_state = Lo_bound; 1166 if (part_shift) 1167 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1168 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1169 if (partscan) 1170 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1171 1172 /* Grab the block_device to prevent its destruction after we 1173 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 1174 */ 1175 bdgrab(bdev); 1176 mutex_unlock(&loop_ctl_mutex); 1177 if (partscan) 1178 loop_reread_partitions(lo, bdev); 1179 if (claimed_bdev) 1180 bd_abort_claiming(bdev, claimed_bdev, loop_configure); 1181 return 0; 1182 1183 out_unlock: 1184 mutex_unlock(&loop_ctl_mutex); 1185 out_bdev: 1186 if (claimed_bdev) 1187 bd_abort_claiming(bdev, claimed_bdev, loop_configure); 1188 out_putf: 1189 fput(file); 1190 out: 1191 /* This is safe: open() is still holding a reference. */ 1192 module_put(THIS_MODULE); 1193 return error; 1194 } 1195 1196 static int __loop_clr_fd(struct loop_device *lo, bool release) 1197 { 1198 struct file *filp = NULL; 1199 gfp_t gfp = lo->old_gfp_mask; 1200 struct block_device *bdev = lo->lo_device; 1201 int err = 0; 1202 bool partscan = false; 1203 int lo_number; 1204 1205 mutex_lock(&loop_ctl_mutex); 1206 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1207 err = -ENXIO; 1208 goto out_unlock; 1209 } 1210 1211 filp = lo->lo_backing_file; 1212 if (filp == NULL) { 1213 err = -EINVAL; 1214 goto out_unlock; 1215 } 1216 1217 /* freeze request queue during the transition */ 1218 blk_mq_freeze_queue(lo->lo_queue); 1219 1220 spin_lock_irq(&lo->lo_lock); 1221 lo->lo_backing_file = NULL; 1222 spin_unlock_irq(&lo->lo_lock); 1223 1224 loop_release_xfer(lo); 1225 lo->transfer = NULL; 1226 lo->ioctl = NULL; 1227 lo->lo_device = NULL; 1228 lo->lo_encryption = NULL; 1229 lo->lo_offset = 0; 1230 lo->lo_sizelimit = 0; 1231 lo->lo_encrypt_key_size = 0; 1232 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1233 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1234 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1235 blk_queue_logical_block_size(lo->lo_queue, 512); 1236 blk_queue_physical_block_size(lo->lo_queue, 512); 1237 blk_queue_io_min(lo->lo_queue, 512); 1238 if (bdev) { 1239 bdput(bdev); 1240 invalidate_bdev(bdev); 1241 bdev->bd_inode->i_mapping->wb_err = 0; 1242 } 1243 set_capacity(lo->lo_disk, 0); 1244 loop_sysfs_exit(lo); 1245 if (bdev) { 1246 bd_set_nr_sectors(bdev, 0); 1247 /* let user-space know about this change */ 1248 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1249 } 1250 mapping_set_gfp_mask(filp->f_mapping, gfp); 1251 /* This is safe: open() is still holding a reference. */ 1252 module_put(THIS_MODULE); 1253 blk_mq_unfreeze_queue(lo->lo_queue); 1254 1255 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1256 lo_number = lo->lo_number; 1257 loop_unprepare_queue(lo); 1258 out_unlock: 1259 mutex_unlock(&loop_ctl_mutex); 1260 if (partscan) { 1261 /* 1262 * bd_mutex has been held already in release path, so don't 1263 * acquire it if this function is called in such case. 1264 * 1265 * If the reread partition isn't from release path, lo_refcnt 1266 * must be at least one and it can only become zero when the 1267 * current holder is released. 1268 */ 1269 if (!release) 1270 mutex_lock(&bdev->bd_mutex); 1271 err = bdev_disk_changed(bdev, false); 1272 if (!release) 1273 mutex_unlock(&bdev->bd_mutex); 1274 if (err) 1275 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1276 __func__, lo_number, err); 1277 /* Device is gone, no point in returning error */ 1278 err = 0; 1279 } 1280 1281 /* 1282 * lo->lo_state is set to Lo_unbound here after above partscan has 1283 * finished. 1284 * 1285 * There cannot be anybody else entering __loop_clr_fd() as 1286 * lo->lo_backing_file is already cleared and Lo_rundown state 1287 * protects us from all the other places trying to change the 'lo' 1288 * device. 1289 */ 1290 mutex_lock(&loop_ctl_mutex); 1291 lo->lo_flags = 0; 1292 if (!part_shift) 1293 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1294 lo->lo_state = Lo_unbound; 1295 mutex_unlock(&loop_ctl_mutex); 1296 1297 /* 1298 * Need not hold loop_ctl_mutex to fput backing file. 1299 * Calling fput holding loop_ctl_mutex triggers a circular 1300 * lock dependency possibility warning as fput can take 1301 * bd_mutex which is usually taken before loop_ctl_mutex. 1302 */ 1303 if (filp) 1304 fput(filp); 1305 return err; 1306 } 1307 1308 static int loop_clr_fd(struct loop_device *lo) 1309 { 1310 int err; 1311 1312 err = mutex_lock_killable(&loop_ctl_mutex); 1313 if (err) 1314 return err; 1315 if (lo->lo_state != Lo_bound) { 1316 mutex_unlock(&loop_ctl_mutex); 1317 return -ENXIO; 1318 } 1319 /* 1320 * If we've explicitly asked to tear down the loop device, 1321 * and it has an elevated reference count, set it for auto-teardown when 1322 * the last reference goes away. This stops $!~#$@ udev from 1323 * preventing teardown because it decided that it needs to run blkid on 1324 * the loopback device whenever they appear. xfstests is notorious for 1325 * failing tests because blkid via udev races with a losetup 1326 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1327 * command to fail with EBUSY. 1328 */ 1329 if (atomic_read(&lo->lo_refcnt) > 1) { 1330 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1331 mutex_unlock(&loop_ctl_mutex); 1332 return 0; 1333 } 1334 lo->lo_state = Lo_rundown; 1335 mutex_unlock(&loop_ctl_mutex); 1336 1337 return __loop_clr_fd(lo, false); 1338 } 1339 1340 static int 1341 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1342 { 1343 int err; 1344 struct block_device *bdev; 1345 kuid_t uid = current_uid(); 1346 int prev_lo_flags; 1347 bool partscan = false; 1348 bool size_changed = false; 1349 1350 err = mutex_lock_killable(&loop_ctl_mutex); 1351 if (err) 1352 return err; 1353 if (lo->lo_encrypt_key_size && 1354 !uid_eq(lo->lo_key_owner, uid) && 1355 !capable(CAP_SYS_ADMIN)) { 1356 err = -EPERM; 1357 goto out_unlock; 1358 } 1359 if (lo->lo_state != Lo_bound) { 1360 err = -ENXIO; 1361 goto out_unlock; 1362 } 1363 1364 if (lo->lo_offset != info->lo_offset || 1365 lo->lo_sizelimit != info->lo_sizelimit) { 1366 size_changed = true; 1367 sync_blockdev(lo->lo_device); 1368 invalidate_bdev(lo->lo_device); 1369 } 1370 1371 /* I/O need to be drained during transfer transition */ 1372 blk_mq_freeze_queue(lo->lo_queue); 1373 1374 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1375 /* If any pages were dirtied after invalidate_bdev(), try again */ 1376 err = -EAGAIN; 1377 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1378 __func__, lo->lo_number, lo->lo_file_name, 1379 lo->lo_device->bd_inode->i_mapping->nrpages); 1380 goto out_unfreeze; 1381 } 1382 1383 prev_lo_flags = lo->lo_flags; 1384 1385 err = loop_set_status_from_info(lo, info); 1386 if (err) 1387 goto out_unfreeze; 1388 1389 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1390 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1391 /* For those flags, use the previous values instead */ 1392 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1393 /* For flags that can't be cleared, use previous values too */ 1394 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1395 1396 if (size_changed) { 1397 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1398 lo->lo_backing_file); 1399 loop_set_size(lo, new_size); 1400 } 1401 1402 loop_config_discard(lo); 1403 1404 /* update dio if lo_offset or transfer is changed */ 1405 __loop_update_dio(lo, lo->use_dio); 1406 1407 out_unfreeze: 1408 blk_mq_unfreeze_queue(lo->lo_queue); 1409 1410 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1411 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1412 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1413 bdev = lo->lo_device; 1414 partscan = true; 1415 } 1416 out_unlock: 1417 mutex_unlock(&loop_ctl_mutex); 1418 if (partscan) 1419 loop_reread_partitions(lo, bdev); 1420 1421 return err; 1422 } 1423 1424 static int 1425 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1426 { 1427 struct path path; 1428 struct kstat stat; 1429 int ret; 1430 1431 ret = mutex_lock_killable(&loop_ctl_mutex); 1432 if (ret) 1433 return ret; 1434 if (lo->lo_state != Lo_bound) { 1435 mutex_unlock(&loop_ctl_mutex); 1436 return -ENXIO; 1437 } 1438 1439 memset(info, 0, sizeof(*info)); 1440 info->lo_number = lo->lo_number; 1441 info->lo_offset = lo->lo_offset; 1442 info->lo_sizelimit = lo->lo_sizelimit; 1443 info->lo_flags = lo->lo_flags; 1444 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1445 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1446 info->lo_encrypt_type = 1447 lo->lo_encryption ? lo->lo_encryption->number : 0; 1448 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1449 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1450 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1451 lo->lo_encrypt_key_size); 1452 } 1453 1454 /* Drop loop_ctl_mutex while we call into the filesystem. */ 1455 path = lo->lo_backing_file->f_path; 1456 path_get(&path); 1457 mutex_unlock(&loop_ctl_mutex); 1458 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1459 if (!ret) { 1460 info->lo_device = huge_encode_dev(stat.dev); 1461 info->lo_inode = stat.ino; 1462 info->lo_rdevice = huge_encode_dev(stat.rdev); 1463 } 1464 path_put(&path); 1465 return ret; 1466 } 1467 1468 static void 1469 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1470 { 1471 memset(info64, 0, sizeof(*info64)); 1472 info64->lo_number = info->lo_number; 1473 info64->lo_device = info->lo_device; 1474 info64->lo_inode = info->lo_inode; 1475 info64->lo_rdevice = info->lo_rdevice; 1476 info64->lo_offset = info->lo_offset; 1477 info64->lo_sizelimit = 0; 1478 info64->lo_encrypt_type = info->lo_encrypt_type; 1479 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1480 info64->lo_flags = info->lo_flags; 1481 info64->lo_init[0] = info->lo_init[0]; 1482 info64->lo_init[1] = info->lo_init[1]; 1483 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1484 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1485 else 1486 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1487 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1488 } 1489 1490 static int 1491 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1492 { 1493 memset(info, 0, sizeof(*info)); 1494 info->lo_number = info64->lo_number; 1495 info->lo_device = info64->lo_device; 1496 info->lo_inode = info64->lo_inode; 1497 info->lo_rdevice = info64->lo_rdevice; 1498 info->lo_offset = info64->lo_offset; 1499 info->lo_encrypt_type = info64->lo_encrypt_type; 1500 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1501 info->lo_flags = info64->lo_flags; 1502 info->lo_init[0] = info64->lo_init[0]; 1503 info->lo_init[1] = info64->lo_init[1]; 1504 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1505 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1506 else 1507 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1508 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1509 1510 /* error in case values were truncated */ 1511 if (info->lo_device != info64->lo_device || 1512 info->lo_rdevice != info64->lo_rdevice || 1513 info->lo_inode != info64->lo_inode || 1514 info->lo_offset != info64->lo_offset) 1515 return -EOVERFLOW; 1516 1517 return 0; 1518 } 1519 1520 static int 1521 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1522 { 1523 struct loop_info info; 1524 struct loop_info64 info64; 1525 1526 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1527 return -EFAULT; 1528 loop_info64_from_old(&info, &info64); 1529 return loop_set_status(lo, &info64); 1530 } 1531 1532 static int 1533 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1534 { 1535 struct loop_info64 info64; 1536 1537 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1538 return -EFAULT; 1539 return loop_set_status(lo, &info64); 1540 } 1541 1542 static int 1543 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1544 struct loop_info info; 1545 struct loop_info64 info64; 1546 int err; 1547 1548 if (!arg) 1549 return -EINVAL; 1550 err = loop_get_status(lo, &info64); 1551 if (!err) 1552 err = loop_info64_to_old(&info64, &info); 1553 if (!err && copy_to_user(arg, &info, sizeof(info))) 1554 err = -EFAULT; 1555 1556 return err; 1557 } 1558 1559 static int 1560 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1561 struct loop_info64 info64; 1562 int err; 1563 1564 if (!arg) 1565 return -EINVAL; 1566 err = loop_get_status(lo, &info64); 1567 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1568 err = -EFAULT; 1569 1570 return err; 1571 } 1572 1573 static int loop_set_capacity(struct loop_device *lo) 1574 { 1575 loff_t size; 1576 1577 if (unlikely(lo->lo_state != Lo_bound)) 1578 return -ENXIO; 1579 1580 size = get_loop_size(lo, lo->lo_backing_file); 1581 loop_set_size(lo, size); 1582 1583 return 0; 1584 } 1585 1586 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1587 { 1588 int error = -ENXIO; 1589 if (lo->lo_state != Lo_bound) 1590 goto out; 1591 1592 __loop_update_dio(lo, !!arg); 1593 if (lo->use_dio == !!arg) 1594 return 0; 1595 error = -EINVAL; 1596 out: 1597 return error; 1598 } 1599 1600 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1601 { 1602 int err = 0; 1603 1604 if (lo->lo_state != Lo_bound) 1605 return -ENXIO; 1606 1607 err = loop_validate_block_size(arg); 1608 if (err) 1609 return err; 1610 1611 if (lo->lo_queue->limits.logical_block_size == arg) 1612 return 0; 1613 1614 sync_blockdev(lo->lo_device); 1615 invalidate_bdev(lo->lo_device); 1616 1617 blk_mq_freeze_queue(lo->lo_queue); 1618 1619 /* invalidate_bdev should have truncated all the pages */ 1620 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1621 err = -EAGAIN; 1622 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1623 __func__, lo->lo_number, lo->lo_file_name, 1624 lo->lo_device->bd_inode->i_mapping->nrpages); 1625 goto out_unfreeze; 1626 } 1627 1628 blk_queue_logical_block_size(lo->lo_queue, arg); 1629 blk_queue_physical_block_size(lo->lo_queue, arg); 1630 blk_queue_io_min(lo->lo_queue, arg); 1631 loop_update_dio(lo); 1632 out_unfreeze: 1633 blk_mq_unfreeze_queue(lo->lo_queue); 1634 1635 return err; 1636 } 1637 1638 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1639 unsigned long arg) 1640 { 1641 int err; 1642 1643 err = mutex_lock_killable(&loop_ctl_mutex); 1644 if (err) 1645 return err; 1646 switch (cmd) { 1647 case LOOP_SET_CAPACITY: 1648 err = loop_set_capacity(lo); 1649 break; 1650 case LOOP_SET_DIRECT_IO: 1651 err = loop_set_dio(lo, arg); 1652 break; 1653 case LOOP_SET_BLOCK_SIZE: 1654 err = loop_set_block_size(lo, arg); 1655 break; 1656 default: 1657 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1658 } 1659 mutex_unlock(&loop_ctl_mutex); 1660 return err; 1661 } 1662 1663 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1664 unsigned int cmd, unsigned long arg) 1665 { 1666 struct loop_device *lo = bdev->bd_disk->private_data; 1667 void __user *argp = (void __user *) arg; 1668 int err; 1669 1670 switch (cmd) { 1671 case LOOP_SET_FD: { 1672 /* 1673 * Legacy case - pass in a zeroed out struct loop_config with 1674 * only the file descriptor set , which corresponds with the 1675 * default parameters we'd have used otherwise. 1676 */ 1677 struct loop_config config; 1678 1679 memset(&config, 0, sizeof(config)); 1680 config.fd = arg; 1681 1682 return loop_configure(lo, mode, bdev, &config); 1683 } 1684 case LOOP_CONFIGURE: { 1685 struct loop_config config; 1686 1687 if (copy_from_user(&config, argp, sizeof(config))) 1688 return -EFAULT; 1689 1690 return loop_configure(lo, mode, bdev, &config); 1691 } 1692 case LOOP_CHANGE_FD: 1693 return loop_change_fd(lo, bdev, arg); 1694 case LOOP_CLR_FD: 1695 return loop_clr_fd(lo); 1696 case LOOP_SET_STATUS: 1697 err = -EPERM; 1698 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1699 err = loop_set_status_old(lo, argp); 1700 } 1701 break; 1702 case LOOP_GET_STATUS: 1703 return loop_get_status_old(lo, argp); 1704 case LOOP_SET_STATUS64: 1705 err = -EPERM; 1706 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1707 err = loop_set_status64(lo, argp); 1708 } 1709 break; 1710 case LOOP_GET_STATUS64: 1711 return loop_get_status64(lo, argp); 1712 case LOOP_SET_CAPACITY: 1713 case LOOP_SET_DIRECT_IO: 1714 case LOOP_SET_BLOCK_SIZE: 1715 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1716 return -EPERM; 1717 fallthrough; 1718 default: 1719 err = lo_simple_ioctl(lo, cmd, arg); 1720 break; 1721 } 1722 1723 return err; 1724 } 1725 1726 #ifdef CONFIG_COMPAT 1727 struct compat_loop_info { 1728 compat_int_t lo_number; /* ioctl r/o */ 1729 compat_dev_t lo_device; /* ioctl r/o */ 1730 compat_ulong_t lo_inode; /* ioctl r/o */ 1731 compat_dev_t lo_rdevice; /* ioctl r/o */ 1732 compat_int_t lo_offset; 1733 compat_int_t lo_encrypt_type; 1734 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1735 compat_int_t lo_flags; /* ioctl r/o */ 1736 char lo_name[LO_NAME_SIZE]; 1737 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1738 compat_ulong_t lo_init[2]; 1739 char reserved[4]; 1740 }; 1741 1742 /* 1743 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1744 * - noinlined to reduce stack space usage in main part of driver 1745 */ 1746 static noinline int 1747 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1748 struct loop_info64 *info64) 1749 { 1750 struct compat_loop_info info; 1751 1752 if (copy_from_user(&info, arg, sizeof(info))) 1753 return -EFAULT; 1754 1755 memset(info64, 0, sizeof(*info64)); 1756 info64->lo_number = info.lo_number; 1757 info64->lo_device = info.lo_device; 1758 info64->lo_inode = info.lo_inode; 1759 info64->lo_rdevice = info.lo_rdevice; 1760 info64->lo_offset = info.lo_offset; 1761 info64->lo_sizelimit = 0; 1762 info64->lo_encrypt_type = info.lo_encrypt_type; 1763 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1764 info64->lo_flags = info.lo_flags; 1765 info64->lo_init[0] = info.lo_init[0]; 1766 info64->lo_init[1] = info.lo_init[1]; 1767 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1768 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1769 else 1770 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1771 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1772 return 0; 1773 } 1774 1775 /* 1776 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1777 * - noinlined to reduce stack space usage in main part of driver 1778 */ 1779 static noinline int 1780 loop_info64_to_compat(const struct loop_info64 *info64, 1781 struct compat_loop_info __user *arg) 1782 { 1783 struct compat_loop_info info; 1784 1785 memset(&info, 0, sizeof(info)); 1786 info.lo_number = info64->lo_number; 1787 info.lo_device = info64->lo_device; 1788 info.lo_inode = info64->lo_inode; 1789 info.lo_rdevice = info64->lo_rdevice; 1790 info.lo_offset = info64->lo_offset; 1791 info.lo_encrypt_type = info64->lo_encrypt_type; 1792 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1793 info.lo_flags = info64->lo_flags; 1794 info.lo_init[0] = info64->lo_init[0]; 1795 info.lo_init[1] = info64->lo_init[1]; 1796 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1797 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1798 else 1799 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1800 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1801 1802 /* error in case values were truncated */ 1803 if (info.lo_device != info64->lo_device || 1804 info.lo_rdevice != info64->lo_rdevice || 1805 info.lo_inode != info64->lo_inode || 1806 info.lo_offset != info64->lo_offset || 1807 info.lo_init[0] != info64->lo_init[0] || 1808 info.lo_init[1] != info64->lo_init[1]) 1809 return -EOVERFLOW; 1810 1811 if (copy_to_user(arg, &info, sizeof(info))) 1812 return -EFAULT; 1813 return 0; 1814 } 1815 1816 static int 1817 loop_set_status_compat(struct loop_device *lo, 1818 const struct compat_loop_info __user *arg) 1819 { 1820 struct loop_info64 info64; 1821 int ret; 1822 1823 ret = loop_info64_from_compat(arg, &info64); 1824 if (ret < 0) 1825 return ret; 1826 return loop_set_status(lo, &info64); 1827 } 1828 1829 static int 1830 loop_get_status_compat(struct loop_device *lo, 1831 struct compat_loop_info __user *arg) 1832 { 1833 struct loop_info64 info64; 1834 int err; 1835 1836 if (!arg) 1837 return -EINVAL; 1838 err = loop_get_status(lo, &info64); 1839 if (!err) 1840 err = loop_info64_to_compat(&info64, arg); 1841 return err; 1842 } 1843 1844 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1845 unsigned int cmd, unsigned long arg) 1846 { 1847 struct loop_device *lo = bdev->bd_disk->private_data; 1848 int err; 1849 1850 switch(cmd) { 1851 case LOOP_SET_STATUS: 1852 err = loop_set_status_compat(lo, 1853 (const struct compat_loop_info __user *)arg); 1854 break; 1855 case LOOP_GET_STATUS: 1856 err = loop_get_status_compat(lo, 1857 (struct compat_loop_info __user *)arg); 1858 break; 1859 case LOOP_SET_CAPACITY: 1860 case LOOP_CLR_FD: 1861 case LOOP_GET_STATUS64: 1862 case LOOP_SET_STATUS64: 1863 case LOOP_CONFIGURE: 1864 arg = (unsigned long) compat_ptr(arg); 1865 fallthrough; 1866 case LOOP_SET_FD: 1867 case LOOP_CHANGE_FD: 1868 case LOOP_SET_BLOCK_SIZE: 1869 case LOOP_SET_DIRECT_IO: 1870 err = lo_ioctl(bdev, mode, cmd, arg); 1871 break; 1872 default: 1873 err = -ENOIOCTLCMD; 1874 break; 1875 } 1876 return err; 1877 } 1878 #endif 1879 1880 static int lo_open(struct block_device *bdev, fmode_t mode) 1881 { 1882 struct loop_device *lo; 1883 int err; 1884 1885 err = mutex_lock_killable(&loop_ctl_mutex); 1886 if (err) 1887 return err; 1888 lo = bdev->bd_disk->private_data; 1889 if (!lo) { 1890 err = -ENXIO; 1891 goto out; 1892 } 1893 1894 atomic_inc(&lo->lo_refcnt); 1895 out: 1896 mutex_unlock(&loop_ctl_mutex); 1897 return err; 1898 } 1899 1900 static void lo_release(struct gendisk *disk, fmode_t mode) 1901 { 1902 struct loop_device *lo; 1903 1904 mutex_lock(&loop_ctl_mutex); 1905 lo = disk->private_data; 1906 if (atomic_dec_return(&lo->lo_refcnt)) 1907 goto out_unlock; 1908 1909 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1910 if (lo->lo_state != Lo_bound) 1911 goto out_unlock; 1912 lo->lo_state = Lo_rundown; 1913 mutex_unlock(&loop_ctl_mutex); 1914 /* 1915 * In autoclear mode, stop the loop thread 1916 * and remove configuration after last close. 1917 */ 1918 __loop_clr_fd(lo, true); 1919 return; 1920 } else if (lo->lo_state == Lo_bound) { 1921 /* 1922 * Otherwise keep thread (if running) and config, 1923 * but flush possible ongoing bios in thread. 1924 */ 1925 blk_mq_freeze_queue(lo->lo_queue); 1926 blk_mq_unfreeze_queue(lo->lo_queue); 1927 } 1928 1929 out_unlock: 1930 mutex_unlock(&loop_ctl_mutex); 1931 } 1932 1933 static const struct block_device_operations lo_fops = { 1934 .owner = THIS_MODULE, 1935 .open = lo_open, 1936 .release = lo_release, 1937 .ioctl = lo_ioctl, 1938 #ifdef CONFIG_COMPAT 1939 .compat_ioctl = lo_compat_ioctl, 1940 #endif 1941 }; 1942 1943 /* 1944 * And now the modules code and kernel interface. 1945 */ 1946 static int max_loop; 1947 module_param(max_loop, int, 0444); 1948 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1949 module_param(max_part, int, 0444); 1950 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1951 MODULE_LICENSE("GPL"); 1952 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1953 1954 int loop_register_transfer(struct loop_func_table *funcs) 1955 { 1956 unsigned int n = funcs->number; 1957 1958 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1959 return -EINVAL; 1960 xfer_funcs[n] = funcs; 1961 return 0; 1962 } 1963 1964 static int unregister_transfer_cb(int id, void *ptr, void *data) 1965 { 1966 struct loop_device *lo = ptr; 1967 struct loop_func_table *xfer = data; 1968 1969 mutex_lock(&loop_ctl_mutex); 1970 if (lo->lo_encryption == xfer) 1971 loop_release_xfer(lo); 1972 mutex_unlock(&loop_ctl_mutex); 1973 return 0; 1974 } 1975 1976 int loop_unregister_transfer(int number) 1977 { 1978 unsigned int n = number; 1979 struct loop_func_table *xfer; 1980 1981 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1982 return -EINVAL; 1983 1984 xfer_funcs[n] = NULL; 1985 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1986 return 0; 1987 } 1988 1989 EXPORT_SYMBOL(loop_register_transfer); 1990 EXPORT_SYMBOL(loop_unregister_transfer); 1991 1992 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1993 const struct blk_mq_queue_data *bd) 1994 { 1995 struct request *rq = bd->rq; 1996 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1997 struct loop_device *lo = rq->q->queuedata; 1998 1999 blk_mq_start_request(rq); 2000 2001 if (lo->lo_state != Lo_bound) 2002 return BLK_STS_IOERR; 2003 2004 switch (req_op(rq)) { 2005 case REQ_OP_FLUSH: 2006 case REQ_OP_DISCARD: 2007 case REQ_OP_WRITE_ZEROES: 2008 cmd->use_aio = false; 2009 break; 2010 default: 2011 cmd->use_aio = lo->use_dio; 2012 break; 2013 } 2014 2015 /* always use the first bio's css */ 2016 #ifdef CONFIG_BLK_CGROUP 2017 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 2018 cmd->css = &bio_blkcg(rq->bio)->css; 2019 css_get(cmd->css); 2020 } else 2021 #endif 2022 cmd->css = NULL; 2023 kthread_queue_work(&lo->worker, &cmd->work); 2024 2025 return BLK_STS_OK; 2026 } 2027 2028 static void loop_handle_cmd(struct loop_cmd *cmd) 2029 { 2030 struct request *rq = blk_mq_rq_from_pdu(cmd); 2031 const bool write = op_is_write(req_op(rq)); 2032 struct loop_device *lo = rq->q->queuedata; 2033 int ret = 0; 2034 2035 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 2036 ret = -EIO; 2037 goto failed; 2038 } 2039 2040 ret = do_req_filebacked(lo, rq); 2041 failed: 2042 /* complete non-aio request */ 2043 if (!cmd->use_aio || ret) { 2044 if (ret == -EOPNOTSUPP) 2045 cmd->ret = ret; 2046 else 2047 cmd->ret = ret ? -EIO : 0; 2048 if (likely(!blk_should_fake_timeout(rq->q))) 2049 blk_mq_complete_request(rq); 2050 } 2051 } 2052 2053 static void loop_queue_work(struct kthread_work *work) 2054 { 2055 struct loop_cmd *cmd = 2056 container_of(work, struct loop_cmd, work); 2057 2058 loop_handle_cmd(cmd); 2059 } 2060 2061 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 2062 unsigned int hctx_idx, unsigned int numa_node) 2063 { 2064 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2065 2066 kthread_init_work(&cmd->work, loop_queue_work); 2067 return 0; 2068 } 2069 2070 static const struct blk_mq_ops loop_mq_ops = { 2071 .queue_rq = loop_queue_rq, 2072 .init_request = loop_init_request, 2073 .complete = lo_complete_rq, 2074 }; 2075 2076 static int loop_add(struct loop_device **l, int i) 2077 { 2078 struct loop_device *lo; 2079 struct gendisk *disk; 2080 int err; 2081 2082 err = -ENOMEM; 2083 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 2084 if (!lo) 2085 goto out; 2086 2087 lo->lo_state = Lo_unbound; 2088 2089 /* allocate id, if @id >= 0, we're requesting that specific id */ 2090 if (i >= 0) { 2091 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2092 if (err == -ENOSPC) 2093 err = -EEXIST; 2094 } else { 2095 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2096 } 2097 if (err < 0) 2098 goto out_free_dev; 2099 i = err; 2100 2101 err = -ENOMEM; 2102 lo->tag_set.ops = &loop_mq_ops; 2103 lo->tag_set.nr_hw_queues = 1; 2104 lo->tag_set.queue_depth = 128; 2105 lo->tag_set.numa_node = NUMA_NO_NODE; 2106 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2107 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING; 2108 lo->tag_set.driver_data = lo; 2109 2110 err = blk_mq_alloc_tag_set(&lo->tag_set); 2111 if (err) 2112 goto out_free_idr; 2113 2114 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 2115 if (IS_ERR(lo->lo_queue)) { 2116 err = PTR_ERR(lo->lo_queue); 2117 goto out_cleanup_tags; 2118 } 2119 lo->lo_queue->queuedata = lo; 2120 2121 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2122 2123 /* 2124 * By default, we do buffer IO, so it doesn't make sense to enable 2125 * merge because the I/O submitted to backing file is handled page by 2126 * page. For directio mode, merge does help to dispatch bigger request 2127 * to underlayer disk. We will enable merge once directio is enabled. 2128 */ 2129 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2130 2131 err = -ENOMEM; 2132 disk = lo->lo_disk = alloc_disk(1 << part_shift); 2133 if (!disk) 2134 goto out_free_queue; 2135 2136 /* 2137 * Disable partition scanning by default. The in-kernel partition 2138 * scanning can be requested individually per-device during its 2139 * setup. Userspace can always add and remove partitions from all 2140 * devices. The needed partition minors are allocated from the 2141 * extended minor space, the main loop device numbers will continue 2142 * to match the loop minors, regardless of the number of partitions 2143 * used. 2144 * 2145 * If max_part is given, partition scanning is globally enabled for 2146 * all loop devices. The minors for the main loop devices will be 2147 * multiples of max_part. 2148 * 2149 * Note: Global-for-all-devices, set-only-at-init, read-only module 2150 * parameteters like 'max_loop' and 'max_part' make things needlessly 2151 * complicated, are too static, inflexible and may surprise 2152 * userspace tools. Parameters like this in general should be avoided. 2153 */ 2154 if (!part_shift) 2155 disk->flags |= GENHD_FL_NO_PART_SCAN; 2156 disk->flags |= GENHD_FL_EXT_DEVT; 2157 atomic_set(&lo->lo_refcnt, 0); 2158 lo->lo_number = i; 2159 spin_lock_init(&lo->lo_lock); 2160 disk->major = LOOP_MAJOR; 2161 disk->first_minor = i << part_shift; 2162 disk->fops = &lo_fops; 2163 disk->private_data = lo; 2164 disk->queue = lo->lo_queue; 2165 sprintf(disk->disk_name, "loop%d", i); 2166 add_disk(disk); 2167 *l = lo; 2168 return lo->lo_number; 2169 2170 out_free_queue: 2171 blk_cleanup_queue(lo->lo_queue); 2172 out_cleanup_tags: 2173 blk_mq_free_tag_set(&lo->tag_set); 2174 out_free_idr: 2175 idr_remove(&loop_index_idr, i); 2176 out_free_dev: 2177 kfree(lo); 2178 out: 2179 return err; 2180 } 2181 2182 static void loop_remove(struct loop_device *lo) 2183 { 2184 del_gendisk(lo->lo_disk); 2185 blk_cleanup_queue(lo->lo_queue); 2186 blk_mq_free_tag_set(&lo->tag_set); 2187 put_disk(lo->lo_disk); 2188 kfree(lo); 2189 } 2190 2191 static int find_free_cb(int id, void *ptr, void *data) 2192 { 2193 struct loop_device *lo = ptr; 2194 struct loop_device **l = data; 2195 2196 if (lo->lo_state == Lo_unbound) { 2197 *l = lo; 2198 return 1; 2199 } 2200 return 0; 2201 } 2202 2203 static int loop_lookup(struct loop_device **l, int i) 2204 { 2205 struct loop_device *lo; 2206 int ret = -ENODEV; 2207 2208 if (i < 0) { 2209 int err; 2210 2211 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 2212 if (err == 1) { 2213 *l = lo; 2214 ret = lo->lo_number; 2215 } 2216 goto out; 2217 } 2218 2219 /* lookup and return a specific i */ 2220 lo = idr_find(&loop_index_idr, i); 2221 if (lo) { 2222 *l = lo; 2223 ret = lo->lo_number; 2224 } 2225 out: 2226 return ret; 2227 } 2228 2229 static void loop_probe(dev_t dev) 2230 { 2231 int idx = MINOR(dev) >> part_shift; 2232 struct loop_device *lo; 2233 2234 if (max_loop && idx >= max_loop) 2235 return; 2236 2237 mutex_lock(&loop_ctl_mutex); 2238 if (loop_lookup(&lo, idx) < 0) 2239 loop_add(&lo, idx); 2240 mutex_unlock(&loop_ctl_mutex); 2241 } 2242 2243 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2244 unsigned long parm) 2245 { 2246 struct loop_device *lo; 2247 int ret; 2248 2249 ret = mutex_lock_killable(&loop_ctl_mutex); 2250 if (ret) 2251 return ret; 2252 2253 ret = -ENOSYS; 2254 switch (cmd) { 2255 case LOOP_CTL_ADD: 2256 ret = loop_lookup(&lo, parm); 2257 if (ret >= 0) { 2258 ret = -EEXIST; 2259 break; 2260 } 2261 ret = loop_add(&lo, parm); 2262 break; 2263 case LOOP_CTL_REMOVE: 2264 ret = loop_lookup(&lo, parm); 2265 if (ret < 0) 2266 break; 2267 if (lo->lo_state != Lo_unbound) { 2268 ret = -EBUSY; 2269 break; 2270 } 2271 if (atomic_read(&lo->lo_refcnt) > 0) { 2272 ret = -EBUSY; 2273 break; 2274 } 2275 lo->lo_disk->private_data = NULL; 2276 idr_remove(&loop_index_idr, lo->lo_number); 2277 loop_remove(lo); 2278 break; 2279 case LOOP_CTL_GET_FREE: 2280 ret = loop_lookup(&lo, -1); 2281 if (ret >= 0) 2282 break; 2283 ret = loop_add(&lo, -1); 2284 } 2285 mutex_unlock(&loop_ctl_mutex); 2286 2287 return ret; 2288 } 2289 2290 static const struct file_operations loop_ctl_fops = { 2291 .open = nonseekable_open, 2292 .unlocked_ioctl = loop_control_ioctl, 2293 .compat_ioctl = loop_control_ioctl, 2294 .owner = THIS_MODULE, 2295 .llseek = noop_llseek, 2296 }; 2297 2298 static struct miscdevice loop_misc = { 2299 .minor = LOOP_CTRL_MINOR, 2300 .name = "loop-control", 2301 .fops = &loop_ctl_fops, 2302 }; 2303 2304 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2305 MODULE_ALIAS("devname:loop-control"); 2306 2307 static int __init loop_init(void) 2308 { 2309 int i, nr; 2310 unsigned long range; 2311 struct loop_device *lo; 2312 int err; 2313 2314 part_shift = 0; 2315 if (max_part > 0) { 2316 part_shift = fls(max_part); 2317 2318 /* 2319 * Adjust max_part according to part_shift as it is exported 2320 * to user space so that user can decide correct minor number 2321 * if [s]he want to create more devices. 2322 * 2323 * Note that -1 is required because partition 0 is reserved 2324 * for the whole disk. 2325 */ 2326 max_part = (1UL << part_shift) - 1; 2327 } 2328 2329 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2330 err = -EINVAL; 2331 goto err_out; 2332 } 2333 2334 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2335 err = -EINVAL; 2336 goto err_out; 2337 } 2338 2339 /* 2340 * If max_loop is specified, create that many devices upfront. 2341 * This also becomes a hard limit. If max_loop is not specified, 2342 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2343 * init time. Loop devices can be requested on-demand with the 2344 * /dev/loop-control interface, or be instantiated by accessing 2345 * a 'dead' device node. 2346 */ 2347 if (max_loop) { 2348 nr = max_loop; 2349 range = max_loop << part_shift; 2350 } else { 2351 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2352 range = 1UL << MINORBITS; 2353 } 2354 2355 err = misc_register(&loop_misc); 2356 if (err < 0) 2357 goto err_out; 2358 2359 2360 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2361 err = -EIO; 2362 goto misc_out; 2363 } 2364 2365 /* pre-create number of devices given by config or max_loop */ 2366 mutex_lock(&loop_ctl_mutex); 2367 for (i = 0; i < nr; i++) 2368 loop_add(&lo, i); 2369 mutex_unlock(&loop_ctl_mutex); 2370 2371 printk(KERN_INFO "loop: module loaded\n"); 2372 return 0; 2373 2374 misc_out: 2375 misc_deregister(&loop_misc); 2376 err_out: 2377 return err; 2378 } 2379 2380 static int loop_exit_cb(int id, void *ptr, void *data) 2381 { 2382 struct loop_device *lo = ptr; 2383 2384 loop_remove(lo); 2385 return 0; 2386 } 2387 2388 static void __exit loop_exit(void) 2389 { 2390 mutex_lock(&loop_ctl_mutex); 2391 2392 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2393 idr_destroy(&loop_index_idr); 2394 2395 unregister_blkdev(LOOP_MAJOR, "loop"); 2396 2397 misc_deregister(&loop_misc); 2398 2399 mutex_unlock(&loop_ctl_mutex); 2400 } 2401 2402 module_init(loop_init); 2403 module_exit(loop_exit); 2404 2405 #ifndef MODULE 2406 static int __init max_loop_setup(char *str) 2407 { 2408 max_loop = simple_strtol(str, NULL, 0); 2409 return 1; 2410 } 2411 2412 __setup("max_loop=", max_loop_setup); 2413 #endif 2414