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_bdev == bdev) 679 return -EBADF; 680 681 l = f->f_mapping->host->i_bdev->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; 889 890 backingq = bdev_get_queue(inode->i_bdev); 891 892 max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 893 granularity = backingq->limits.discard_granularity ?: 894 queue_physical_block_size(backingq); 895 896 /* 897 * We use punch hole to reclaim the free space used by the 898 * image a.k.a. discard. However we do not support discard if 899 * encryption is enabled, because it may give an attacker 900 * useful information. 901 */ 902 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { 903 max_discard_sectors = 0; 904 granularity = 0; 905 906 } else { 907 max_discard_sectors = UINT_MAX >> 9; 908 granularity = inode->i_sb->s_blocksize; 909 } 910 911 if (max_discard_sectors) { 912 q->limits.discard_granularity = granularity; 913 blk_queue_max_discard_sectors(q, max_discard_sectors); 914 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 915 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 916 } else { 917 q->limits.discard_granularity = 0; 918 blk_queue_max_discard_sectors(q, 0); 919 blk_queue_max_write_zeroes_sectors(q, 0); 920 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 921 } 922 q->limits.discard_alignment = 0; 923 } 924 925 static void loop_unprepare_queue(struct loop_device *lo) 926 { 927 kthread_flush_worker(&lo->worker); 928 kthread_stop(lo->worker_task); 929 } 930 931 static int loop_kthread_worker_fn(void *worker_ptr) 932 { 933 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 934 return kthread_worker_fn(worker_ptr); 935 } 936 937 static int loop_prepare_queue(struct loop_device *lo) 938 { 939 kthread_init_worker(&lo->worker); 940 lo->worker_task = kthread_run(loop_kthread_worker_fn, 941 &lo->worker, "loop%d", lo->lo_number); 942 if (IS_ERR(lo->worker_task)) 943 return -ENOMEM; 944 set_user_nice(lo->worker_task, MIN_NICE); 945 return 0; 946 } 947 948 static void loop_update_rotational(struct loop_device *lo) 949 { 950 struct file *file = lo->lo_backing_file; 951 struct inode *file_inode = file->f_mapping->host; 952 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 953 struct request_queue *q = lo->lo_queue; 954 bool nonrot = true; 955 956 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 957 if (file_bdev) 958 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 959 960 if (nonrot) 961 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 962 else 963 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 964 } 965 966 static int 967 loop_release_xfer(struct loop_device *lo) 968 { 969 int err = 0; 970 struct loop_func_table *xfer = lo->lo_encryption; 971 972 if (xfer) { 973 if (xfer->release) 974 err = xfer->release(lo); 975 lo->transfer = NULL; 976 lo->lo_encryption = NULL; 977 module_put(xfer->owner); 978 } 979 return err; 980 } 981 982 static int 983 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 984 const struct loop_info64 *i) 985 { 986 int err = 0; 987 988 if (xfer) { 989 struct module *owner = xfer->owner; 990 991 if (!try_module_get(owner)) 992 return -EINVAL; 993 if (xfer->init) 994 err = xfer->init(lo, i); 995 if (err) 996 module_put(owner); 997 else 998 lo->lo_encryption = xfer; 999 } 1000 return err; 1001 } 1002 1003 /** 1004 * loop_set_status_from_info - configure device from loop_info 1005 * @lo: struct loop_device to configure 1006 * @info: struct loop_info64 to configure the device with 1007 * 1008 * Configures the loop device parameters according to the passed 1009 * in loop_info64 configuration. 1010 */ 1011 static int 1012 loop_set_status_from_info(struct loop_device *lo, 1013 const struct loop_info64 *info) 1014 { 1015 int err; 1016 struct loop_func_table *xfer; 1017 kuid_t uid = current_uid(); 1018 1019 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1020 return -EINVAL; 1021 1022 err = loop_release_xfer(lo); 1023 if (err) 1024 return err; 1025 1026 if (info->lo_encrypt_type) { 1027 unsigned int type = info->lo_encrypt_type; 1028 1029 if (type >= MAX_LO_CRYPT) 1030 return -EINVAL; 1031 xfer = xfer_funcs[type]; 1032 if (xfer == NULL) 1033 return -EINVAL; 1034 } else 1035 xfer = NULL; 1036 1037 err = loop_init_xfer(lo, xfer, info); 1038 if (err) 1039 return err; 1040 1041 lo->lo_offset = info->lo_offset; 1042 lo->lo_sizelimit = info->lo_sizelimit; 1043 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1044 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1045 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1046 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1047 1048 if (!xfer) 1049 xfer = &none_funcs; 1050 lo->transfer = xfer->transfer; 1051 lo->ioctl = xfer->ioctl; 1052 1053 lo->lo_flags = info->lo_flags; 1054 1055 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1056 lo->lo_init[0] = info->lo_init[0]; 1057 lo->lo_init[1] = info->lo_init[1]; 1058 if (info->lo_encrypt_key_size) { 1059 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1060 info->lo_encrypt_key_size); 1061 lo->lo_key_owner = uid; 1062 } 1063 1064 return 0; 1065 } 1066 1067 static int loop_configure(struct loop_device *lo, fmode_t mode, 1068 struct block_device *bdev, 1069 const struct loop_config *config) 1070 { 1071 struct file *file; 1072 struct inode *inode; 1073 struct address_space *mapping; 1074 struct block_device *claimed_bdev = NULL; 1075 int error; 1076 loff_t size; 1077 bool partscan; 1078 unsigned short bsize; 1079 1080 /* This is safe, since we have a reference from open(). */ 1081 __module_get(THIS_MODULE); 1082 1083 error = -EBADF; 1084 file = fget(config->fd); 1085 if (!file) 1086 goto out; 1087 1088 /* 1089 * If we don't hold exclusive handle for the device, upgrade to it 1090 * here to avoid changing device under exclusive owner. 1091 */ 1092 if (!(mode & FMODE_EXCL)) { 1093 claimed_bdev = bdev->bd_contains; 1094 error = bd_prepare_to_claim(bdev, claimed_bdev, loop_configure); 1095 if (error) 1096 goto out_putf; 1097 } 1098 1099 error = mutex_lock_killable(&loop_ctl_mutex); 1100 if (error) 1101 goto out_bdev; 1102 1103 error = -EBUSY; 1104 if (lo->lo_state != Lo_unbound) 1105 goto out_unlock; 1106 1107 error = loop_validate_file(file, bdev); 1108 if (error) 1109 goto out_unlock; 1110 1111 mapping = file->f_mapping; 1112 inode = mapping->host; 1113 1114 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1115 error = -EINVAL; 1116 goto out_unlock; 1117 } 1118 1119 if (config->block_size) { 1120 error = loop_validate_block_size(config->block_size); 1121 if (error) 1122 goto out_unlock; 1123 } 1124 1125 error = loop_set_status_from_info(lo, &config->info); 1126 if (error) 1127 goto out_unlock; 1128 1129 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1130 !file->f_op->write_iter) 1131 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1132 1133 error = loop_prepare_queue(lo); 1134 if (error) 1135 goto out_unlock; 1136 1137 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1138 1139 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1140 lo->lo_device = bdev; 1141 lo->lo_backing_file = file; 1142 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1143 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1144 1145 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1146 blk_queue_write_cache(lo->lo_queue, true, false); 1147 1148 if (config->block_size) 1149 bsize = config->block_size; 1150 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1151 /* In case of direct I/O, match underlying block size */ 1152 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1153 else 1154 bsize = 512; 1155 1156 blk_queue_logical_block_size(lo->lo_queue, bsize); 1157 blk_queue_physical_block_size(lo->lo_queue, bsize); 1158 blk_queue_io_min(lo->lo_queue, bsize); 1159 1160 loop_update_rotational(lo); 1161 loop_update_dio(lo); 1162 loop_sysfs_init(lo); 1163 1164 size = get_loop_size(lo, file); 1165 loop_set_size(lo, size); 1166 1167 lo->lo_state = Lo_bound; 1168 if (part_shift) 1169 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1170 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1171 if (partscan) 1172 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1173 1174 /* Grab the block_device to prevent its destruction after we 1175 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 1176 */ 1177 bdgrab(bdev); 1178 mutex_unlock(&loop_ctl_mutex); 1179 if (partscan) 1180 loop_reread_partitions(lo, bdev); 1181 if (claimed_bdev) 1182 bd_abort_claiming(bdev, claimed_bdev, loop_configure); 1183 return 0; 1184 1185 out_unlock: 1186 mutex_unlock(&loop_ctl_mutex); 1187 out_bdev: 1188 if (claimed_bdev) 1189 bd_abort_claiming(bdev, claimed_bdev, loop_configure); 1190 out_putf: 1191 fput(file); 1192 out: 1193 /* This is safe: open() is still holding a reference. */ 1194 module_put(THIS_MODULE); 1195 return error; 1196 } 1197 1198 static int __loop_clr_fd(struct loop_device *lo, bool release) 1199 { 1200 struct file *filp = NULL; 1201 gfp_t gfp = lo->old_gfp_mask; 1202 struct block_device *bdev = lo->lo_device; 1203 int err = 0; 1204 bool partscan = false; 1205 int lo_number; 1206 1207 mutex_lock(&loop_ctl_mutex); 1208 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1209 err = -ENXIO; 1210 goto out_unlock; 1211 } 1212 1213 filp = lo->lo_backing_file; 1214 if (filp == NULL) { 1215 err = -EINVAL; 1216 goto out_unlock; 1217 } 1218 1219 /* freeze request queue during the transition */ 1220 blk_mq_freeze_queue(lo->lo_queue); 1221 1222 spin_lock_irq(&lo->lo_lock); 1223 lo->lo_backing_file = NULL; 1224 spin_unlock_irq(&lo->lo_lock); 1225 1226 loop_release_xfer(lo); 1227 lo->transfer = NULL; 1228 lo->ioctl = NULL; 1229 lo->lo_device = NULL; 1230 lo->lo_encryption = NULL; 1231 lo->lo_offset = 0; 1232 lo->lo_sizelimit = 0; 1233 lo->lo_encrypt_key_size = 0; 1234 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1235 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1236 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1237 blk_queue_logical_block_size(lo->lo_queue, 512); 1238 blk_queue_physical_block_size(lo->lo_queue, 512); 1239 blk_queue_io_min(lo->lo_queue, 512); 1240 if (bdev) { 1241 bdput(bdev); 1242 invalidate_bdev(bdev); 1243 bdev->bd_inode->i_mapping->wb_err = 0; 1244 } 1245 set_capacity(lo->lo_disk, 0); 1246 loop_sysfs_exit(lo); 1247 if (bdev) { 1248 bd_set_nr_sectors(bdev, 0); 1249 /* let user-space know about this change */ 1250 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1251 } 1252 mapping_set_gfp_mask(filp->f_mapping, gfp); 1253 /* This is safe: open() is still holding a reference. */ 1254 module_put(THIS_MODULE); 1255 blk_mq_unfreeze_queue(lo->lo_queue); 1256 1257 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1258 lo_number = lo->lo_number; 1259 loop_unprepare_queue(lo); 1260 out_unlock: 1261 mutex_unlock(&loop_ctl_mutex); 1262 if (partscan) { 1263 /* 1264 * bd_mutex has been held already in release path, so don't 1265 * acquire it if this function is called in such case. 1266 * 1267 * If the reread partition isn't from release path, lo_refcnt 1268 * must be at least one and it can only become zero when the 1269 * current holder is released. 1270 */ 1271 if (!release) 1272 mutex_lock(&bdev->bd_mutex); 1273 err = bdev_disk_changed(bdev, false); 1274 if (!release) 1275 mutex_unlock(&bdev->bd_mutex); 1276 if (err) 1277 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1278 __func__, lo_number, err); 1279 /* Device is gone, no point in returning error */ 1280 err = 0; 1281 } 1282 1283 /* 1284 * lo->lo_state is set to Lo_unbound here after above partscan has 1285 * finished. 1286 * 1287 * There cannot be anybody else entering __loop_clr_fd() as 1288 * lo->lo_backing_file is already cleared and Lo_rundown state 1289 * protects us from all the other places trying to change the 'lo' 1290 * device. 1291 */ 1292 mutex_lock(&loop_ctl_mutex); 1293 lo->lo_flags = 0; 1294 if (!part_shift) 1295 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1296 lo->lo_state = Lo_unbound; 1297 mutex_unlock(&loop_ctl_mutex); 1298 1299 /* 1300 * Need not hold loop_ctl_mutex to fput backing file. 1301 * Calling fput holding loop_ctl_mutex triggers a circular 1302 * lock dependency possibility warning as fput can take 1303 * bd_mutex which is usually taken before loop_ctl_mutex. 1304 */ 1305 if (filp) 1306 fput(filp); 1307 return err; 1308 } 1309 1310 static int loop_clr_fd(struct loop_device *lo) 1311 { 1312 int err; 1313 1314 err = mutex_lock_killable(&loop_ctl_mutex); 1315 if (err) 1316 return err; 1317 if (lo->lo_state != Lo_bound) { 1318 mutex_unlock(&loop_ctl_mutex); 1319 return -ENXIO; 1320 } 1321 /* 1322 * If we've explicitly asked to tear down the loop device, 1323 * and it has an elevated reference count, set it for auto-teardown when 1324 * the last reference goes away. This stops $!~#$@ udev from 1325 * preventing teardown because it decided that it needs to run blkid on 1326 * the loopback device whenever they appear. xfstests is notorious for 1327 * failing tests because blkid via udev races with a losetup 1328 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1329 * command to fail with EBUSY. 1330 */ 1331 if (atomic_read(&lo->lo_refcnt) > 1) { 1332 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1333 mutex_unlock(&loop_ctl_mutex); 1334 return 0; 1335 } 1336 lo->lo_state = Lo_rundown; 1337 mutex_unlock(&loop_ctl_mutex); 1338 1339 return __loop_clr_fd(lo, false); 1340 } 1341 1342 static int 1343 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1344 { 1345 int err; 1346 struct block_device *bdev; 1347 kuid_t uid = current_uid(); 1348 int prev_lo_flags; 1349 bool partscan = false; 1350 bool size_changed = false; 1351 1352 err = mutex_lock_killable(&loop_ctl_mutex); 1353 if (err) 1354 return err; 1355 if (lo->lo_encrypt_key_size && 1356 !uid_eq(lo->lo_key_owner, uid) && 1357 !capable(CAP_SYS_ADMIN)) { 1358 err = -EPERM; 1359 goto out_unlock; 1360 } 1361 if (lo->lo_state != Lo_bound) { 1362 err = -ENXIO; 1363 goto out_unlock; 1364 } 1365 1366 if (lo->lo_offset != info->lo_offset || 1367 lo->lo_sizelimit != info->lo_sizelimit) { 1368 size_changed = true; 1369 sync_blockdev(lo->lo_device); 1370 invalidate_bdev(lo->lo_device); 1371 } 1372 1373 /* I/O need to be drained during transfer transition */ 1374 blk_mq_freeze_queue(lo->lo_queue); 1375 1376 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1377 /* If any pages were dirtied after invalidate_bdev(), try again */ 1378 err = -EAGAIN; 1379 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1380 __func__, lo->lo_number, lo->lo_file_name, 1381 lo->lo_device->bd_inode->i_mapping->nrpages); 1382 goto out_unfreeze; 1383 } 1384 1385 prev_lo_flags = lo->lo_flags; 1386 1387 err = loop_set_status_from_info(lo, info); 1388 if (err) 1389 goto out_unfreeze; 1390 1391 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1392 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1393 /* For those flags, use the previous values instead */ 1394 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1395 /* For flags that can't be cleared, use previous values too */ 1396 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1397 1398 if (size_changed) { 1399 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1400 lo->lo_backing_file); 1401 loop_set_size(lo, new_size); 1402 } 1403 1404 loop_config_discard(lo); 1405 1406 /* update dio if lo_offset or transfer is changed */ 1407 __loop_update_dio(lo, lo->use_dio); 1408 1409 out_unfreeze: 1410 blk_mq_unfreeze_queue(lo->lo_queue); 1411 1412 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1413 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1414 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1415 bdev = lo->lo_device; 1416 partscan = true; 1417 } 1418 out_unlock: 1419 mutex_unlock(&loop_ctl_mutex); 1420 if (partscan) 1421 loop_reread_partitions(lo, bdev); 1422 1423 return err; 1424 } 1425 1426 static int 1427 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1428 { 1429 struct path path; 1430 struct kstat stat; 1431 int ret; 1432 1433 ret = mutex_lock_killable(&loop_ctl_mutex); 1434 if (ret) 1435 return ret; 1436 if (lo->lo_state != Lo_bound) { 1437 mutex_unlock(&loop_ctl_mutex); 1438 return -ENXIO; 1439 } 1440 1441 memset(info, 0, sizeof(*info)); 1442 info->lo_number = lo->lo_number; 1443 info->lo_offset = lo->lo_offset; 1444 info->lo_sizelimit = lo->lo_sizelimit; 1445 info->lo_flags = lo->lo_flags; 1446 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1447 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1448 info->lo_encrypt_type = 1449 lo->lo_encryption ? lo->lo_encryption->number : 0; 1450 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1451 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1452 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1453 lo->lo_encrypt_key_size); 1454 } 1455 1456 /* Drop loop_ctl_mutex while we call into the filesystem. */ 1457 path = lo->lo_backing_file->f_path; 1458 path_get(&path); 1459 mutex_unlock(&loop_ctl_mutex); 1460 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1461 if (!ret) { 1462 info->lo_device = huge_encode_dev(stat.dev); 1463 info->lo_inode = stat.ino; 1464 info->lo_rdevice = huge_encode_dev(stat.rdev); 1465 } 1466 path_put(&path); 1467 return ret; 1468 } 1469 1470 static void 1471 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1472 { 1473 memset(info64, 0, sizeof(*info64)); 1474 info64->lo_number = info->lo_number; 1475 info64->lo_device = info->lo_device; 1476 info64->lo_inode = info->lo_inode; 1477 info64->lo_rdevice = info->lo_rdevice; 1478 info64->lo_offset = info->lo_offset; 1479 info64->lo_sizelimit = 0; 1480 info64->lo_encrypt_type = info->lo_encrypt_type; 1481 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1482 info64->lo_flags = info->lo_flags; 1483 info64->lo_init[0] = info->lo_init[0]; 1484 info64->lo_init[1] = info->lo_init[1]; 1485 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1486 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1487 else 1488 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1489 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1490 } 1491 1492 static int 1493 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1494 { 1495 memset(info, 0, sizeof(*info)); 1496 info->lo_number = info64->lo_number; 1497 info->lo_device = info64->lo_device; 1498 info->lo_inode = info64->lo_inode; 1499 info->lo_rdevice = info64->lo_rdevice; 1500 info->lo_offset = info64->lo_offset; 1501 info->lo_encrypt_type = info64->lo_encrypt_type; 1502 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1503 info->lo_flags = info64->lo_flags; 1504 info->lo_init[0] = info64->lo_init[0]; 1505 info->lo_init[1] = info64->lo_init[1]; 1506 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1507 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1508 else 1509 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1510 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1511 1512 /* error in case values were truncated */ 1513 if (info->lo_device != info64->lo_device || 1514 info->lo_rdevice != info64->lo_rdevice || 1515 info->lo_inode != info64->lo_inode || 1516 info->lo_offset != info64->lo_offset) 1517 return -EOVERFLOW; 1518 1519 return 0; 1520 } 1521 1522 static int 1523 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1524 { 1525 struct loop_info info; 1526 struct loop_info64 info64; 1527 1528 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1529 return -EFAULT; 1530 loop_info64_from_old(&info, &info64); 1531 return loop_set_status(lo, &info64); 1532 } 1533 1534 static int 1535 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1536 { 1537 struct loop_info64 info64; 1538 1539 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1540 return -EFAULT; 1541 return loop_set_status(lo, &info64); 1542 } 1543 1544 static int 1545 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1546 struct loop_info info; 1547 struct loop_info64 info64; 1548 int err; 1549 1550 if (!arg) 1551 return -EINVAL; 1552 err = loop_get_status(lo, &info64); 1553 if (!err) 1554 err = loop_info64_to_old(&info64, &info); 1555 if (!err && copy_to_user(arg, &info, sizeof(info))) 1556 err = -EFAULT; 1557 1558 return err; 1559 } 1560 1561 static int 1562 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1563 struct loop_info64 info64; 1564 int err; 1565 1566 if (!arg) 1567 return -EINVAL; 1568 err = loop_get_status(lo, &info64); 1569 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1570 err = -EFAULT; 1571 1572 return err; 1573 } 1574 1575 static int loop_set_capacity(struct loop_device *lo) 1576 { 1577 loff_t size; 1578 1579 if (unlikely(lo->lo_state != Lo_bound)) 1580 return -ENXIO; 1581 1582 size = get_loop_size(lo, lo->lo_backing_file); 1583 loop_set_size(lo, size); 1584 1585 return 0; 1586 } 1587 1588 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1589 { 1590 int error = -ENXIO; 1591 if (lo->lo_state != Lo_bound) 1592 goto out; 1593 1594 __loop_update_dio(lo, !!arg); 1595 if (lo->use_dio == !!arg) 1596 return 0; 1597 error = -EINVAL; 1598 out: 1599 return error; 1600 } 1601 1602 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1603 { 1604 int err = 0; 1605 1606 if (lo->lo_state != Lo_bound) 1607 return -ENXIO; 1608 1609 err = loop_validate_block_size(arg); 1610 if (err) 1611 return err; 1612 1613 if (lo->lo_queue->limits.logical_block_size == arg) 1614 return 0; 1615 1616 sync_blockdev(lo->lo_device); 1617 invalidate_bdev(lo->lo_device); 1618 1619 blk_mq_freeze_queue(lo->lo_queue); 1620 1621 /* invalidate_bdev should have truncated all the pages */ 1622 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1623 err = -EAGAIN; 1624 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1625 __func__, lo->lo_number, lo->lo_file_name, 1626 lo->lo_device->bd_inode->i_mapping->nrpages); 1627 goto out_unfreeze; 1628 } 1629 1630 blk_queue_logical_block_size(lo->lo_queue, arg); 1631 blk_queue_physical_block_size(lo->lo_queue, arg); 1632 blk_queue_io_min(lo->lo_queue, arg); 1633 loop_update_dio(lo); 1634 out_unfreeze: 1635 blk_mq_unfreeze_queue(lo->lo_queue); 1636 1637 return err; 1638 } 1639 1640 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1641 unsigned long arg) 1642 { 1643 int err; 1644 1645 err = mutex_lock_killable(&loop_ctl_mutex); 1646 if (err) 1647 return err; 1648 switch (cmd) { 1649 case LOOP_SET_CAPACITY: 1650 err = loop_set_capacity(lo); 1651 break; 1652 case LOOP_SET_DIRECT_IO: 1653 err = loop_set_dio(lo, arg); 1654 break; 1655 case LOOP_SET_BLOCK_SIZE: 1656 err = loop_set_block_size(lo, arg); 1657 break; 1658 default: 1659 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1660 } 1661 mutex_unlock(&loop_ctl_mutex); 1662 return err; 1663 } 1664 1665 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1666 unsigned int cmd, unsigned long arg) 1667 { 1668 struct loop_device *lo = bdev->bd_disk->private_data; 1669 void __user *argp = (void __user *) arg; 1670 int err; 1671 1672 switch (cmd) { 1673 case LOOP_SET_FD: { 1674 /* 1675 * Legacy case - pass in a zeroed out struct loop_config with 1676 * only the file descriptor set , which corresponds with the 1677 * default parameters we'd have used otherwise. 1678 */ 1679 struct loop_config config; 1680 1681 memset(&config, 0, sizeof(config)); 1682 config.fd = arg; 1683 1684 return loop_configure(lo, mode, bdev, &config); 1685 } 1686 case LOOP_CONFIGURE: { 1687 struct loop_config config; 1688 1689 if (copy_from_user(&config, argp, sizeof(config))) 1690 return -EFAULT; 1691 1692 return loop_configure(lo, mode, bdev, &config); 1693 } 1694 case LOOP_CHANGE_FD: 1695 return loop_change_fd(lo, bdev, arg); 1696 case LOOP_CLR_FD: 1697 return loop_clr_fd(lo); 1698 case LOOP_SET_STATUS: 1699 err = -EPERM; 1700 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1701 err = loop_set_status_old(lo, argp); 1702 } 1703 break; 1704 case LOOP_GET_STATUS: 1705 return loop_get_status_old(lo, argp); 1706 case LOOP_SET_STATUS64: 1707 err = -EPERM; 1708 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1709 err = loop_set_status64(lo, argp); 1710 } 1711 break; 1712 case LOOP_GET_STATUS64: 1713 return loop_get_status64(lo, argp); 1714 case LOOP_SET_CAPACITY: 1715 case LOOP_SET_DIRECT_IO: 1716 case LOOP_SET_BLOCK_SIZE: 1717 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1718 return -EPERM; 1719 fallthrough; 1720 default: 1721 err = lo_simple_ioctl(lo, cmd, arg); 1722 break; 1723 } 1724 1725 return err; 1726 } 1727 1728 #ifdef CONFIG_COMPAT 1729 struct compat_loop_info { 1730 compat_int_t lo_number; /* ioctl r/o */ 1731 compat_dev_t lo_device; /* ioctl r/o */ 1732 compat_ulong_t lo_inode; /* ioctl r/o */ 1733 compat_dev_t lo_rdevice; /* ioctl r/o */ 1734 compat_int_t lo_offset; 1735 compat_int_t lo_encrypt_type; 1736 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1737 compat_int_t lo_flags; /* ioctl r/o */ 1738 char lo_name[LO_NAME_SIZE]; 1739 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1740 compat_ulong_t lo_init[2]; 1741 char reserved[4]; 1742 }; 1743 1744 /* 1745 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1746 * - noinlined to reduce stack space usage in main part of driver 1747 */ 1748 static noinline int 1749 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1750 struct loop_info64 *info64) 1751 { 1752 struct compat_loop_info info; 1753 1754 if (copy_from_user(&info, arg, sizeof(info))) 1755 return -EFAULT; 1756 1757 memset(info64, 0, sizeof(*info64)); 1758 info64->lo_number = info.lo_number; 1759 info64->lo_device = info.lo_device; 1760 info64->lo_inode = info.lo_inode; 1761 info64->lo_rdevice = info.lo_rdevice; 1762 info64->lo_offset = info.lo_offset; 1763 info64->lo_sizelimit = 0; 1764 info64->lo_encrypt_type = info.lo_encrypt_type; 1765 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1766 info64->lo_flags = info.lo_flags; 1767 info64->lo_init[0] = info.lo_init[0]; 1768 info64->lo_init[1] = info.lo_init[1]; 1769 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1770 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1771 else 1772 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1773 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1774 return 0; 1775 } 1776 1777 /* 1778 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1779 * - noinlined to reduce stack space usage in main part of driver 1780 */ 1781 static noinline int 1782 loop_info64_to_compat(const struct loop_info64 *info64, 1783 struct compat_loop_info __user *arg) 1784 { 1785 struct compat_loop_info info; 1786 1787 memset(&info, 0, sizeof(info)); 1788 info.lo_number = info64->lo_number; 1789 info.lo_device = info64->lo_device; 1790 info.lo_inode = info64->lo_inode; 1791 info.lo_rdevice = info64->lo_rdevice; 1792 info.lo_offset = info64->lo_offset; 1793 info.lo_encrypt_type = info64->lo_encrypt_type; 1794 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1795 info.lo_flags = info64->lo_flags; 1796 info.lo_init[0] = info64->lo_init[0]; 1797 info.lo_init[1] = info64->lo_init[1]; 1798 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1799 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1800 else 1801 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1802 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1803 1804 /* error in case values were truncated */ 1805 if (info.lo_device != info64->lo_device || 1806 info.lo_rdevice != info64->lo_rdevice || 1807 info.lo_inode != info64->lo_inode || 1808 info.lo_offset != info64->lo_offset || 1809 info.lo_init[0] != info64->lo_init[0] || 1810 info.lo_init[1] != info64->lo_init[1]) 1811 return -EOVERFLOW; 1812 1813 if (copy_to_user(arg, &info, sizeof(info))) 1814 return -EFAULT; 1815 return 0; 1816 } 1817 1818 static int 1819 loop_set_status_compat(struct loop_device *lo, 1820 const struct compat_loop_info __user *arg) 1821 { 1822 struct loop_info64 info64; 1823 int ret; 1824 1825 ret = loop_info64_from_compat(arg, &info64); 1826 if (ret < 0) 1827 return ret; 1828 return loop_set_status(lo, &info64); 1829 } 1830 1831 static int 1832 loop_get_status_compat(struct loop_device *lo, 1833 struct compat_loop_info __user *arg) 1834 { 1835 struct loop_info64 info64; 1836 int err; 1837 1838 if (!arg) 1839 return -EINVAL; 1840 err = loop_get_status(lo, &info64); 1841 if (!err) 1842 err = loop_info64_to_compat(&info64, arg); 1843 return err; 1844 } 1845 1846 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1847 unsigned int cmd, unsigned long arg) 1848 { 1849 struct loop_device *lo = bdev->bd_disk->private_data; 1850 int err; 1851 1852 switch(cmd) { 1853 case LOOP_SET_STATUS: 1854 err = loop_set_status_compat(lo, 1855 (const struct compat_loop_info __user *)arg); 1856 break; 1857 case LOOP_GET_STATUS: 1858 err = loop_get_status_compat(lo, 1859 (struct compat_loop_info __user *)arg); 1860 break; 1861 case LOOP_SET_CAPACITY: 1862 case LOOP_CLR_FD: 1863 case LOOP_GET_STATUS64: 1864 case LOOP_SET_STATUS64: 1865 case LOOP_CONFIGURE: 1866 arg = (unsigned long) compat_ptr(arg); 1867 fallthrough; 1868 case LOOP_SET_FD: 1869 case LOOP_CHANGE_FD: 1870 case LOOP_SET_BLOCK_SIZE: 1871 case LOOP_SET_DIRECT_IO: 1872 err = lo_ioctl(bdev, mode, cmd, arg); 1873 break; 1874 default: 1875 err = -ENOIOCTLCMD; 1876 break; 1877 } 1878 return err; 1879 } 1880 #endif 1881 1882 static int lo_open(struct block_device *bdev, fmode_t mode) 1883 { 1884 struct loop_device *lo; 1885 int err; 1886 1887 err = mutex_lock_killable(&loop_ctl_mutex); 1888 if (err) 1889 return err; 1890 lo = bdev->bd_disk->private_data; 1891 if (!lo) { 1892 err = -ENXIO; 1893 goto out; 1894 } 1895 1896 atomic_inc(&lo->lo_refcnt); 1897 out: 1898 mutex_unlock(&loop_ctl_mutex); 1899 return err; 1900 } 1901 1902 static void lo_release(struct gendisk *disk, fmode_t mode) 1903 { 1904 struct loop_device *lo; 1905 1906 mutex_lock(&loop_ctl_mutex); 1907 lo = disk->private_data; 1908 if (atomic_dec_return(&lo->lo_refcnt)) 1909 goto out_unlock; 1910 1911 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1912 if (lo->lo_state != Lo_bound) 1913 goto out_unlock; 1914 lo->lo_state = Lo_rundown; 1915 mutex_unlock(&loop_ctl_mutex); 1916 /* 1917 * In autoclear mode, stop the loop thread 1918 * and remove configuration after last close. 1919 */ 1920 __loop_clr_fd(lo, true); 1921 return; 1922 } else if (lo->lo_state == Lo_bound) { 1923 /* 1924 * Otherwise keep thread (if running) and config, 1925 * but flush possible ongoing bios in thread. 1926 */ 1927 blk_mq_freeze_queue(lo->lo_queue); 1928 blk_mq_unfreeze_queue(lo->lo_queue); 1929 } 1930 1931 out_unlock: 1932 mutex_unlock(&loop_ctl_mutex); 1933 } 1934 1935 static const struct block_device_operations lo_fops = { 1936 .owner = THIS_MODULE, 1937 .open = lo_open, 1938 .release = lo_release, 1939 .ioctl = lo_ioctl, 1940 #ifdef CONFIG_COMPAT 1941 .compat_ioctl = lo_compat_ioctl, 1942 #endif 1943 }; 1944 1945 /* 1946 * And now the modules code and kernel interface. 1947 */ 1948 static int max_loop; 1949 module_param(max_loop, int, 0444); 1950 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1951 module_param(max_part, int, 0444); 1952 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1953 MODULE_LICENSE("GPL"); 1954 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1955 1956 int loop_register_transfer(struct loop_func_table *funcs) 1957 { 1958 unsigned int n = funcs->number; 1959 1960 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1961 return -EINVAL; 1962 xfer_funcs[n] = funcs; 1963 return 0; 1964 } 1965 1966 static int unregister_transfer_cb(int id, void *ptr, void *data) 1967 { 1968 struct loop_device *lo = ptr; 1969 struct loop_func_table *xfer = data; 1970 1971 mutex_lock(&loop_ctl_mutex); 1972 if (lo->lo_encryption == xfer) 1973 loop_release_xfer(lo); 1974 mutex_unlock(&loop_ctl_mutex); 1975 return 0; 1976 } 1977 1978 int loop_unregister_transfer(int number) 1979 { 1980 unsigned int n = number; 1981 struct loop_func_table *xfer; 1982 1983 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1984 return -EINVAL; 1985 1986 xfer_funcs[n] = NULL; 1987 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1988 return 0; 1989 } 1990 1991 EXPORT_SYMBOL(loop_register_transfer); 1992 EXPORT_SYMBOL(loop_unregister_transfer); 1993 1994 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1995 const struct blk_mq_queue_data *bd) 1996 { 1997 struct request *rq = bd->rq; 1998 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1999 struct loop_device *lo = rq->q->queuedata; 2000 2001 blk_mq_start_request(rq); 2002 2003 if (lo->lo_state != Lo_bound) 2004 return BLK_STS_IOERR; 2005 2006 switch (req_op(rq)) { 2007 case REQ_OP_FLUSH: 2008 case REQ_OP_DISCARD: 2009 case REQ_OP_WRITE_ZEROES: 2010 cmd->use_aio = false; 2011 break; 2012 default: 2013 cmd->use_aio = lo->use_dio; 2014 break; 2015 } 2016 2017 /* always use the first bio's css */ 2018 #ifdef CONFIG_BLK_CGROUP 2019 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 2020 cmd->css = &bio_blkcg(rq->bio)->css; 2021 css_get(cmd->css); 2022 } else 2023 #endif 2024 cmd->css = NULL; 2025 kthread_queue_work(&lo->worker, &cmd->work); 2026 2027 return BLK_STS_OK; 2028 } 2029 2030 static void loop_handle_cmd(struct loop_cmd *cmd) 2031 { 2032 struct request *rq = blk_mq_rq_from_pdu(cmd); 2033 const bool write = op_is_write(req_op(rq)); 2034 struct loop_device *lo = rq->q->queuedata; 2035 int ret = 0; 2036 2037 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 2038 ret = -EIO; 2039 goto failed; 2040 } 2041 2042 ret = do_req_filebacked(lo, rq); 2043 failed: 2044 /* complete non-aio request */ 2045 if (!cmd->use_aio || ret) { 2046 if (ret == -EOPNOTSUPP) 2047 cmd->ret = ret; 2048 else 2049 cmd->ret = ret ? -EIO : 0; 2050 if (likely(!blk_should_fake_timeout(rq->q))) 2051 blk_mq_complete_request(rq); 2052 } 2053 } 2054 2055 static void loop_queue_work(struct kthread_work *work) 2056 { 2057 struct loop_cmd *cmd = 2058 container_of(work, struct loop_cmd, work); 2059 2060 loop_handle_cmd(cmd); 2061 } 2062 2063 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 2064 unsigned int hctx_idx, unsigned int numa_node) 2065 { 2066 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2067 2068 kthread_init_work(&cmd->work, loop_queue_work); 2069 return 0; 2070 } 2071 2072 static const struct blk_mq_ops loop_mq_ops = { 2073 .queue_rq = loop_queue_rq, 2074 .init_request = loop_init_request, 2075 .complete = lo_complete_rq, 2076 }; 2077 2078 static int loop_add(struct loop_device **l, int i) 2079 { 2080 struct loop_device *lo; 2081 struct gendisk *disk; 2082 int err; 2083 2084 err = -ENOMEM; 2085 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 2086 if (!lo) 2087 goto out; 2088 2089 lo->lo_state = Lo_unbound; 2090 2091 /* allocate id, if @id >= 0, we're requesting that specific id */ 2092 if (i >= 0) { 2093 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2094 if (err == -ENOSPC) 2095 err = -EEXIST; 2096 } else { 2097 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2098 } 2099 if (err < 0) 2100 goto out_free_dev; 2101 i = err; 2102 2103 err = -ENOMEM; 2104 lo->tag_set.ops = &loop_mq_ops; 2105 lo->tag_set.nr_hw_queues = 1; 2106 lo->tag_set.queue_depth = 128; 2107 lo->tag_set.numa_node = NUMA_NO_NODE; 2108 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2109 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING; 2110 lo->tag_set.driver_data = lo; 2111 2112 err = blk_mq_alloc_tag_set(&lo->tag_set); 2113 if (err) 2114 goto out_free_idr; 2115 2116 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 2117 if (IS_ERR(lo->lo_queue)) { 2118 err = PTR_ERR(lo->lo_queue); 2119 goto out_cleanup_tags; 2120 } 2121 lo->lo_queue->queuedata = lo; 2122 2123 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2124 2125 /* 2126 * By default, we do buffer IO, so it doesn't make sense to enable 2127 * merge because the I/O submitted to backing file is handled page by 2128 * page. For directio mode, merge does help to dispatch bigger request 2129 * to underlayer disk. We will enable merge once directio is enabled. 2130 */ 2131 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2132 2133 err = -ENOMEM; 2134 disk = lo->lo_disk = alloc_disk(1 << part_shift); 2135 if (!disk) 2136 goto out_free_queue; 2137 2138 /* 2139 * Disable partition scanning by default. The in-kernel partition 2140 * scanning can be requested individually per-device during its 2141 * setup. Userspace can always add and remove partitions from all 2142 * devices. The needed partition minors are allocated from the 2143 * extended minor space, the main loop device numbers will continue 2144 * to match the loop minors, regardless of the number of partitions 2145 * used. 2146 * 2147 * If max_part is given, partition scanning is globally enabled for 2148 * all loop devices. The minors for the main loop devices will be 2149 * multiples of max_part. 2150 * 2151 * Note: Global-for-all-devices, set-only-at-init, read-only module 2152 * parameteters like 'max_loop' and 'max_part' make things needlessly 2153 * complicated, are too static, inflexible and may surprise 2154 * userspace tools. Parameters like this in general should be avoided. 2155 */ 2156 if (!part_shift) 2157 disk->flags |= GENHD_FL_NO_PART_SCAN; 2158 disk->flags |= GENHD_FL_EXT_DEVT; 2159 atomic_set(&lo->lo_refcnt, 0); 2160 lo->lo_number = i; 2161 spin_lock_init(&lo->lo_lock); 2162 disk->major = LOOP_MAJOR; 2163 disk->first_minor = i << part_shift; 2164 disk->fops = &lo_fops; 2165 disk->private_data = lo; 2166 disk->queue = lo->lo_queue; 2167 sprintf(disk->disk_name, "loop%d", i); 2168 add_disk(disk); 2169 *l = lo; 2170 return lo->lo_number; 2171 2172 out_free_queue: 2173 blk_cleanup_queue(lo->lo_queue); 2174 out_cleanup_tags: 2175 blk_mq_free_tag_set(&lo->tag_set); 2176 out_free_idr: 2177 idr_remove(&loop_index_idr, i); 2178 out_free_dev: 2179 kfree(lo); 2180 out: 2181 return err; 2182 } 2183 2184 static void loop_remove(struct loop_device *lo) 2185 { 2186 del_gendisk(lo->lo_disk); 2187 blk_cleanup_queue(lo->lo_queue); 2188 blk_mq_free_tag_set(&lo->tag_set); 2189 put_disk(lo->lo_disk); 2190 kfree(lo); 2191 } 2192 2193 static int find_free_cb(int id, void *ptr, void *data) 2194 { 2195 struct loop_device *lo = ptr; 2196 struct loop_device **l = data; 2197 2198 if (lo->lo_state == Lo_unbound) { 2199 *l = lo; 2200 return 1; 2201 } 2202 return 0; 2203 } 2204 2205 static int loop_lookup(struct loop_device **l, int i) 2206 { 2207 struct loop_device *lo; 2208 int ret = -ENODEV; 2209 2210 if (i < 0) { 2211 int err; 2212 2213 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 2214 if (err == 1) { 2215 *l = lo; 2216 ret = lo->lo_number; 2217 } 2218 goto out; 2219 } 2220 2221 /* lookup and return a specific i */ 2222 lo = idr_find(&loop_index_idr, i); 2223 if (lo) { 2224 *l = lo; 2225 ret = lo->lo_number; 2226 } 2227 out: 2228 return ret; 2229 } 2230 2231 static void loop_probe(dev_t dev) 2232 { 2233 int idx = MINOR(dev) >> part_shift; 2234 struct loop_device *lo; 2235 2236 if (max_loop && idx >= max_loop) 2237 return; 2238 2239 mutex_lock(&loop_ctl_mutex); 2240 if (loop_lookup(&lo, idx) < 0) 2241 loop_add(&lo, idx); 2242 mutex_unlock(&loop_ctl_mutex); 2243 } 2244 2245 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2246 unsigned long parm) 2247 { 2248 struct loop_device *lo; 2249 int ret; 2250 2251 ret = mutex_lock_killable(&loop_ctl_mutex); 2252 if (ret) 2253 return ret; 2254 2255 ret = -ENOSYS; 2256 switch (cmd) { 2257 case LOOP_CTL_ADD: 2258 ret = loop_lookup(&lo, parm); 2259 if (ret >= 0) { 2260 ret = -EEXIST; 2261 break; 2262 } 2263 ret = loop_add(&lo, parm); 2264 break; 2265 case LOOP_CTL_REMOVE: 2266 ret = loop_lookup(&lo, parm); 2267 if (ret < 0) 2268 break; 2269 if (lo->lo_state != Lo_unbound) { 2270 ret = -EBUSY; 2271 break; 2272 } 2273 if (atomic_read(&lo->lo_refcnt) > 0) { 2274 ret = -EBUSY; 2275 break; 2276 } 2277 lo->lo_disk->private_data = NULL; 2278 idr_remove(&loop_index_idr, lo->lo_number); 2279 loop_remove(lo); 2280 break; 2281 case LOOP_CTL_GET_FREE: 2282 ret = loop_lookup(&lo, -1); 2283 if (ret >= 0) 2284 break; 2285 ret = loop_add(&lo, -1); 2286 } 2287 mutex_unlock(&loop_ctl_mutex); 2288 2289 return ret; 2290 } 2291 2292 static const struct file_operations loop_ctl_fops = { 2293 .open = nonseekable_open, 2294 .unlocked_ioctl = loop_control_ioctl, 2295 .compat_ioctl = loop_control_ioctl, 2296 .owner = THIS_MODULE, 2297 .llseek = noop_llseek, 2298 }; 2299 2300 static struct miscdevice loop_misc = { 2301 .minor = LOOP_CTRL_MINOR, 2302 .name = "loop-control", 2303 .fops = &loop_ctl_fops, 2304 }; 2305 2306 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2307 MODULE_ALIAS("devname:loop-control"); 2308 2309 static int __init loop_init(void) 2310 { 2311 int i, nr; 2312 unsigned long range; 2313 struct loop_device *lo; 2314 int err; 2315 2316 part_shift = 0; 2317 if (max_part > 0) { 2318 part_shift = fls(max_part); 2319 2320 /* 2321 * Adjust max_part according to part_shift as it is exported 2322 * to user space so that user can decide correct minor number 2323 * if [s]he want to create more devices. 2324 * 2325 * Note that -1 is required because partition 0 is reserved 2326 * for the whole disk. 2327 */ 2328 max_part = (1UL << part_shift) - 1; 2329 } 2330 2331 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2332 err = -EINVAL; 2333 goto err_out; 2334 } 2335 2336 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2337 err = -EINVAL; 2338 goto err_out; 2339 } 2340 2341 /* 2342 * If max_loop is specified, create that many devices upfront. 2343 * This also becomes a hard limit. If max_loop is not specified, 2344 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2345 * init time. Loop devices can be requested on-demand with the 2346 * /dev/loop-control interface, or be instantiated by accessing 2347 * a 'dead' device node. 2348 */ 2349 if (max_loop) { 2350 nr = max_loop; 2351 range = max_loop << part_shift; 2352 } else { 2353 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2354 range = 1UL << MINORBITS; 2355 } 2356 2357 err = misc_register(&loop_misc); 2358 if (err < 0) 2359 goto err_out; 2360 2361 2362 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2363 err = -EIO; 2364 goto misc_out; 2365 } 2366 2367 /* pre-create number of devices given by config or max_loop */ 2368 mutex_lock(&loop_ctl_mutex); 2369 for (i = 0; i < nr; i++) 2370 loop_add(&lo, i); 2371 mutex_unlock(&loop_ctl_mutex); 2372 2373 printk(KERN_INFO "loop: module loaded\n"); 2374 return 0; 2375 2376 misc_out: 2377 misc_deregister(&loop_misc); 2378 err_out: 2379 return err; 2380 } 2381 2382 static int loop_exit_cb(int id, void *ptr, void *data) 2383 { 2384 struct loop_device *lo = ptr; 2385 2386 loop_remove(lo); 2387 return 0; 2388 } 2389 2390 static void __exit loop_exit(void) 2391 { 2392 mutex_lock(&loop_ctl_mutex); 2393 2394 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2395 idr_destroy(&loop_index_idr); 2396 2397 unregister_blkdev(LOOP_MAJOR, "loop"); 2398 2399 misc_deregister(&loop_misc); 2400 2401 mutex_unlock(&loop_ctl_mutex); 2402 } 2403 2404 module_init(loop_init); 2405 module_exit(loop_exit); 2406 2407 #ifndef MODULE 2408 static int __init max_loop_setup(char *str) 2409 { 2410 max_loop = simple_strtol(str, NULL, 0); 2411 return 1; 2412 } 2413 2414 __setup("max_loop=", max_loop_setup); 2415 #endif 2416