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/pagemap.h> 57 #include <linux/file.h> 58 #include <linux/stat.h> 59 #include <linux/errno.h> 60 #include <linux/major.h> 61 #include <linux/wait.h> 62 #include <linux/blkdev.h> 63 #include <linux/blkpg.h> 64 #include <linux/init.h> 65 #include <linux/swap.h> 66 #include <linux/slab.h> 67 #include <linux/compat.h> 68 #include <linux/suspend.h> 69 #include <linux/freezer.h> 70 #include <linux/mutex.h> 71 #include <linux/writeback.h> 72 #include <linux/completion.h> 73 #include <linux/highmem.h> 74 #include <linux/kthread.h> 75 #include <linux/splice.h> 76 #include <linux/sysfs.h> 77 #include <linux/miscdevice.h> 78 #include <linux/falloc.h> 79 #include <linux/uio.h> 80 #include <linux/ioprio.h> 81 #include <linux/blk-cgroup.h> 82 83 #include "loop.h" 84 85 #include <linux/uaccess.h> 86 87 static DEFINE_IDR(loop_index_idr); 88 static DEFINE_MUTEX(loop_ctl_mutex); 89 90 static int max_part; 91 static int part_shift; 92 93 static int transfer_xor(struct loop_device *lo, int cmd, 94 struct page *raw_page, unsigned raw_off, 95 struct page *loop_page, unsigned loop_off, 96 int size, sector_t real_block) 97 { 98 char *raw_buf = kmap_atomic(raw_page) + raw_off; 99 char *loop_buf = kmap_atomic(loop_page) + loop_off; 100 char *in, *out, *key; 101 int i, keysize; 102 103 if (cmd == READ) { 104 in = raw_buf; 105 out = loop_buf; 106 } else { 107 in = loop_buf; 108 out = raw_buf; 109 } 110 111 key = lo->lo_encrypt_key; 112 keysize = lo->lo_encrypt_key_size; 113 for (i = 0; i < size; i++) 114 *out++ = *in++ ^ key[(i & 511) % keysize]; 115 116 kunmap_atomic(loop_buf); 117 kunmap_atomic(raw_buf); 118 cond_resched(); 119 return 0; 120 } 121 122 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 123 { 124 if (unlikely(info->lo_encrypt_key_size <= 0)) 125 return -EINVAL; 126 return 0; 127 } 128 129 static struct loop_func_table none_funcs = { 130 .number = LO_CRYPT_NONE, 131 }; 132 133 static struct loop_func_table xor_funcs = { 134 .number = LO_CRYPT_XOR, 135 .transfer = transfer_xor, 136 .init = xor_init 137 }; 138 139 /* xfer_funcs[0] is special - its release function is never called */ 140 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 141 &none_funcs, 142 &xor_funcs 143 }; 144 145 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 146 { 147 loff_t loopsize; 148 149 /* Compute loopsize in bytes */ 150 loopsize = i_size_read(file->f_mapping->host); 151 if (offset > 0) 152 loopsize -= offset; 153 /* offset is beyond i_size, weird but possible */ 154 if (loopsize < 0) 155 return 0; 156 157 if (sizelimit > 0 && sizelimit < loopsize) 158 loopsize = sizelimit; 159 /* 160 * Unfortunately, if we want to do I/O on the device, 161 * the number of 512-byte sectors has to fit into a sector_t. 162 */ 163 return loopsize >> 9; 164 } 165 166 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 167 { 168 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 169 } 170 171 static void __loop_update_dio(struct loop_device *lo, bool dio) 172 { 173 struct file *file = lo->lo_backing_file; 174 struct address_space *mapping = file->f_mapping; 175 struct inode *inode = mapping->host; 176 unsigned short sb_bsize = 0; 177 unsigned dio_align = 0; 178 bool use_dio; 179 180 if (inode->i_sb->s_bdev) { 181 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 182 dio_align = sb_bsize - 1; 183 } 184 185 /* 186 * We support direct I/O only if lo_offset is aligned with the 187 * logical I/O size of backing device, and the logical block 188 * size of loop is bigger than the backing device's and the loop 189 * needn't transform transfer. 190 * 191 * TODO: the above condition may be loosed in the future, and 192 * direct I/O may be switched runtime at that time because most 193 * of requests in sane applications should be PAGE_SIZE aligned 194 */ 195 if (dio) { 196 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 197 !(lo->lo_offset & dio_align) && 198 mapping->a_ops->direct_IO && 199 !lo->transfer) 200 use_dio = true; 201 else 202 use_dio = false; 203 } else { 204 use_dio = false; 205 } 206 207 if (lo->use_dio == use_dio) 208 return; 209 210 /* flush dirty pages before changing direct IO */ 211 vfs_fsync(file, 0); 212 213 /* 214 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 215 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 216 * will get updated by ioctl(LOOP_GET_STATUS) 217 */ 218 if (lo->lo_state == Lo_bound) 219 blk_mq_freeze_queue(lo->lo_queue); 220 lo->use_dio = use_dio; 221 if (use_dio) { 222 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 223 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 224 } else { 225 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 226 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 227 } 228 if (lo->lo_state == Lo_bound) 229 blk_mq_unfreeze_queue(lo->lo_queue); 230 } 231 232 /** 233 * loop_validate_block_size() - validates the passed in block size 234 * @bsize: size to validate 235 */ 236 static int 237 loop_validate_block_size(unsigned short bsize) 238 { 239 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize)) 240 return -EINVAL; 241 242 return 0; 243 } 244 245 /** 246 * loop_set_size() - sets device size and notifies userspace 247 * @lo: struct loop_device to set the size for 248 * @size: new size of the loop device 249 * 250 * Callers must validate that the size passed into this function fits into 251 * a sector_t, eg using loop_validate_size() 252 */ 253 static void loop_set_size(struct loop_device *lo, loff_t size) 254 { 255 if (!set_capacity_and_notify(lo->lo_disk, size)) 256 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 257 } 258 259 static inline int 260 lo_do_transfer(struct loop_device *lo, int cmd, 261 struct page *rpage, unsigned roffs, 262 struct page *lpage, unsigned loffs, 263 int size, sector_t rblock) 264 { 265 int ret; 266 267 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 268 if (likely(!ret)) 269 return 0; 270 271 printk_ratelimited(KERN_ERR 272 "loop: Transfer error at byte offset %llu, length %i.\n", 273 (unsigned long long)rblock << 9, size); 274 return ret; 275 } 276 277 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 278 { 279 struct iov_iter i; 280 ssize_t bw; 281 282 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 283 284 file_start_write(file); 285 bw = vfs_iter_write(file, &i, ppos, 0); 286 file_end_write(file); 287 288 if (likely(bw == bvec->bv_len)) 289 return 0; 290 291 printk_ratelimited(KERN_ERR 292 "loop: Write error at byte offset %llu, length %i.\n", 293 (unsigned long long)*ppos, bvec->bv_len); 294 if (bw >= 0) 295 bw = -EIO; 296 return bw; 297 } 298 299 static int lo_write_simple(struct loop_device *lo, struct request *rq, 300 loff_t pos) 301 { 302 struct bio_vec bvec; 303 struct req_iterator iter; 304 int ret = 0; 305 306 rq_for_each_segment(bvec, rq, iter) { 307 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 308 if (ret < 0) 309 break; 310 cond_resched(); 311 } 312 313 return ret; 314 } 315 316 /* 317 * This is the slow, transforming version that needs to double buffer the 318 * data as it cannot do the transformations in place without having direct 319 * access to the destination pages of the backing file. 320 */ 321 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 322 loff_t pos) 323 { 324 struct bio_vec bvec, b; 325 struct req_iterator iter; 326 struct page *page; 327 int ret = 0; 328 329 page = alloc_page(GFP_NOIO); 330 if (unlikely(!page)) 331 return -ENOMEM; 332 333 rq_for_each_segment(bvec, rq, iter) { 334 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 335 bvec.bv_offset, bvec.bv_len, pos >> 9); 336 if (unlikely(ret)) 337 break; 338 339 b.bv_page = page; 340 b.bv_offset = 0; 341 b.bv_len = bvec.bv_len; 342 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 343 if (ret < 0) 344 break; 345 } 346 347 __free_page(page); 348 return ret; 349 } 350 351 static int lo_read_simple(struct loop_device *lo, struct request *rq, 352 loff_t pos) 353 { 354 struct bio_vec bvec; 355 struct req_iterator iter; 356 struct iov_iter i; 357 ssize_t len; 358 359 rq_for_each_segment(bvec, rq, iter) { 360 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 361 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 362 if (len < 0) 363 return len; 364 365 flush_dcache_page(bvec.bv_page); 366 367 if (len != bvec.bv_len) { 368 struct bio *bio; 369 370 __rq_for_each_bio(bio, rq) 371 zero_fill_bio(bio); 372 break; 373 } 374 cond_resched(); 375 } 376 377 return 0; 378 } 379 380 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 381 loff_t pos) 382 { 383 struct bio_vec bvec, b; 384 struct req_iterator iter; 385 struct iov_iter i; 386 struct page *page; 387 ssize_t len; 388 int ret = 0; 389 390 page = alloc_page(GFP_NOIO); 391 if (unlikely(!page)) 392 return -ENOMEM; 393 394 rq_for_each_segment(bvec, rq, iter) { 395 loff_t offset = pos; 396 397 b.bv_page = page; 398 b.bv_offset = 0; 399 b.bv_len = bvec.bv_len; 400 401 iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 402 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 403 if (len < 0) { 404 ret = len; 405 goto out_free_page; 406 } 407 408 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 409 bvec.bv_offset, len, offset >> 9); 410 if (ret) 411 goto out_free_page; 412 413 flush_dcache_page(bvec.bv_page); 414 415 if (len != bvec.bv_len) { 416 struct bio *bio; 417 418 __rq_for_each_bio(bio, rq) 419 zero_fill_bio(bio); 420 break; 421 } 422 } 423 424 ret = 0; 425 out_free_page: 426 __free_page(page); 427 return ret; 428 } 429 430 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 431 int mode) 432 { 433 /* 434 * We use fallocate to manipulate the space mappings used by the image 435 * a.k.a. discard/zerorange. However we do not support this if 436 * encryption is enabled, because it may give an attacker useful 437 * information. 438 */ 439 struct file *file = lo->lo_backing_file; 440 struct request_queue *q = lo->lo_queue; 441 int ret; 442 443 mode |= FALLOC_FL_KEEP_SIZE; 444 445 if (!blk_queue_discard(q)) { 446 ret = -EOPNOTSUPP; 447 goto out; 448 } 449 450 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 451 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 452 ret = -EIO; 453 out: 454 return ret; 455 } 456 457 static int lo_req_flush(struct loop_device *lo, struct request *rq) 458 { 459 struct file *file = lo->lo_backing_file; 460 int ret = vfs_fsync(file, 0); 461 if (unlikely(ret && ret != -EINVAL)) 462 ret = -EIO; 463 464 return ret; 465 } 466 467 static void lo_complete_rq(struct request *rq) 468 { 469 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 470 blk_status_t ret = BLK_STS_OK; 471 472 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 473 req_op(rq) != REQ_OP_READ) { 474 if (cmd->ret < 0) 475 ret = errno_to_blk_status(cmd->ret); 476 goto end_io; 477 } 478 479 /* 480 * Short READ - if we got some data, advance our request and 481 * retry it. If we got no data, end the rest with EIO. 482 */ 483 if (cmd->ret) { 484 blk_update_request(rq, BLK_STS_OK, cmd->ret); 485 cmd->ret = 0; 486 blk_mq_requeue_request(rq, true); 487 } else { 488 if (cmd->use_aio) { 489 struct bio *bio = rq->bio; 490 491 while (bio) { 492 zero_fill_bio(bio); 493 bio = bio->bi_next; 494 } 495 } 496 ret = BLK_STS_IOERR; 497 end_io: 498 blk_mq_end_request(rq, ret); 499 } 500 } 501 502 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 503 { 504 struct request *rq = blk_mq_rq_from_pdu(cmd); 505 506 if (!atomic_dec_and_test(&cmd->ref)) 507 return; 508 kfree(cmd->bvec); 509 cmd->bvec = NULL; 510 if (likely(!blk_should_fake_timeout(rq->q))) 511 blk_mq_complete_request(rq); 512 } 513 514 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 515 { 516 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 517 518 if (cmd->css) 519 css_put(cmd->css); 520 cmd->ret = ret; 521 lo_rw_aio_do_completion(cmd); 522 } 523 524 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 525 loff_t pos, bool rw) 526 { 527 struct iov_iter iter; 528 struct req_iterator rq_iter; 529 struct bio_vec *bvec; 530 struct request *rq = blk_mq_rq_from_pdu(cmd); 531 struct bio *bio = rq->bio; 532 struct file *file = lo->lo_backing_file; 533 struct bio_vec tmp; 534 unsigned int offset; 535 int nr_bvec = 0; 536 int ret; 537 538 rq_for_each_bvec(tmp, rq, rq_iter) 539 nr_bvec++; 540 541 if (rq->bio != rq->biotail) { 542 543 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 544 GFP_NOIO); 545 if (!bvec) 546 return -EIO; 547 cmd->bvec = bvec; 548 549 /* 550 * The bios of the request may be started from the middle of 551 * the 'bvec' because of bio splitting, so we can't directly 552 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 553 * API will take care of all details for us. 554 */ 555 rq_for_each_bvec(tmp, rq, rq_iter) { 556 *bvec = tmp; 557 bvec++; 558 } 559 bvec = cmd->bvec; 560 offset = 0; 561 } else { 562 /* 563 * Same here, this bio may be started from the middle of the 564 * 'bvec' because of bio splitting, so offset from the bvec 565 * must be passed to iov iterator 566 */ 567 offset = bio->bi_iter.bi_bvec_done; 568 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 569 } 570 atomic_set(&cmd->ref, 2); 571 572 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 573 iter.iov_offset = offset; 574 575 cmd->iocb.ki_pos = pos; 576 cmd->iocb.ki_filp = file; 577 cmd->iocb.ki_complete = lo_rw_aio_complete; 578 cmd->iocb.ki_flags = IOCB_DIRECT; 579 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 580 if (cmd->css) 581 kthread_associate_blkcg(cmd->css); 582 583 if (rw == WRITE) 584 ret = call_write_iter(file, &cmd->iocb, &iter); 585 else 586 ret = call_read_iter(file, &cmd->iocb, &iter); 587 588 lo_rw_aio_do_completion(cmd); 589 kthread_associate_blkcg(NULL); 590 591 if (ret != -EIOCBQUEUED) 592 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 593 return 0; 594 } 595 596 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 597 { 598 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 599 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 600 601 /* 602 * lo_write_simple and lo_read_simple should have been covered 603 * by io submit style function like lo_rw_aio(), one blocker 604 * is that lo_read_simple() need to call flush_dcache_page after 605 * the page is written from kernel, and it isn't easy to handle 606 * this in io submit style function which submits all segments 607 * of the req at one time. And direct read IO doesn't need to 608 * run flush_dcache_page(). 609 */ 610 switch (req_op(rq)) { 611 case REQ_OP_FLUSH: 612 return lo_req_flush(lo, rq); 613 case REQ_OP_WRITE_ZEROES: 614 /* 615 * If the caller doesn't want deallocation, call zeroout to 616 * write zeroes the range. Otherwise, punch them out. 617 */ 618 return lo_fallocate(lo, rq, pos, 619 (rq->cmd_flags & REQ_NOUNMAP) ? 620 FALLOC_FL_ZERO_RANGE : 621 FALLOC_FL_PUNCH_HOLE); 622 case REQ_OP_DISCARD: 623 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 624 case REQ_OP_WRITE: 625 if (lo->transfer) 626 return lo_write_transfer(lo, rq, pos); 627 else if (cmd->use_aio) 628 return lo_rw_aio(lo, cmd, pos, WRITE); 629 else 630 return lo_write_simple(lo, rq, pos); 631 case REQ_OP_READ: 632 if (lo->transfer) 633 return lo_read_transfer(lo, rq, pos); 634 else if (cmd->use_aio) 635 return lo_rw_aio(lo, cmd, pos, READ); 636 else 637 return lo_read_simple(lo, rq, pos); 638 default: 639 WARN_ON_ONCE(1); 640 return -EIO; 641 } 642 } 643 644 static inline void loop_update_dio(struct loop_device *lo) 645 { 646 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 647 lo->use_dio); 648 } 649 650 static void loop_reread_partitions(struct loop_device *lo) 651 { 652 int rc; 653 654 mutex_lock(&lo->lo_disk->open_mutex); 655 rc = bdev_disk_changed(lo->lo_disk, false); 656 mutex_unlock(&lo->lo_disk->open_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) && imajor(i) == 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(&lo->lo_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(&lo->lo_mutex); 747 /* 748 * We must drop file reference outside of lo_mutex as dropping 749 * the file ref can take open_mutex which creates circular locking 750 * dependency. 751 */ 752 fput(old_file); 753 if (partscan) 754 loop_reread_partitions(lo); 755 return 0; 756 757 out_err: 758 mutex_unlock(&lo->lo_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 int error; 1073 loff_t size; 1074 bool partscan; 1075 unsigned short bsize; 1076 1077 /* This is safe, since we have a reference from open(). */ 1078 __module_get(THIS_MODULE); 1079 1080 error = -EBADF; 1081 file = fget(config->fd); 1082 if (!file) 1083 goto out; 1084 1085 /* 1086 * If we don't hold exclusive handle for the device, upgrade to it 1087 * here to avoid changing device under exclusive owner. 1088 */ 1089 if (!(mode & FMODE_EXCL)) { 1090 error = bd_prepare_to_claim(bdev, loop_configure); 1091 if (error) 1092 goto out_putf; 1093 } 1094 1095 error = mutex_lock_killable(&lo->lo_mutex); 1096 if (error) 1097 goto out_bdev; 1098 1099 error = -EBUSY; 1100 if (lo->lo_state != Lo_unbound) 1101 goto out_unlock; 1102 1103 error = loop_validate_file(file, bdev); 1104 if (error) 1105 goto out_unlock; 1106 1107 mapping = file->f_mapping; 1108 inode = mapping->host; 1109 1110 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1111 error = -EINVAL; 1112 goto out_unlock; 1113 } 1114 1115 if (config->block_size) { 1116 error = loop_validate_block_size(config->block_size); 1117 if (error) 1118 goto out_unlock; 1119 } 1120 1121 error = loop_set_status_from_info(lo, &config->info); 1122 if (error) 1123 goto out_unlock; 1124 1125 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1126 !file->f_op->write_iter) 1127 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1128 1129 error = loop_prepare_queue(lo); 1130 if (error) 1131 goto out_unlock; 1132 1133 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1134 1135 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1136 lo->lo_device = bdev; 1137 lo->lo_backing_file = file; 1138 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1139 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1140 1141 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1142 blk_queue_write_cache(lo->lo_queue, true, false); 1143 1144 if (config->block_size) 1145 bsize = config->block_size; 1146 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1147 /* In case of direct I/O, match underlying block size */ 1148 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1149 else 1150 bsize = 512; 1151 1152 blk_queue_logical_block_size(lo->lo_queue, bsize); 1153 blk_queue_physical_block_size(lo->lo_queue, bsize); 1154 blk_queue_io_min(lo->lo_queue, bsize); 1155 1156 loop_update_rotational(lo); 1157 loop_update_dio(lo); 1158 loop_sysfs_init(lo); 1159 1160 size = get_loop_size(lo, file); 1161 loop_set_size(lo, size); 1162 1163 lo->lo_state = Lo_bound; 1164 if (part_shift) 1165 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1166 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1167 if (partscan) 1168 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1169 1170 /* Grab the block_device to prevent its destruction after we 1171 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 1172 */ 1173 bdgrab(bdev); 1174 mutex_unlock(&lo->lo_mutex); 1175 if (partscan) 1176 loop_reread_partitions(lo); 1177 if (!(mode & FMODE_EXCL)) 1178 bd_abort_claiming(bdev, loop_configure); 1179 return 0; 1180 1181 out_unlock: 1182 mutex_unlock(&lo->lo_mutex); 1183 out_bdev: 1184 if (!(mode & FMODE_EXCL)) 1185 bd_abort_claiming(bdev, loop_configure); 1186 out_putf: 1187 fput(file); 1188 out: 1189 /* This is safe: open() is still holding a reference. */ 1190 module_put(THIS_MODULE); 1191 return error; 1192 } 1193 1194 static int __loop_clr_fd(struct loop_device *lo, bool release) 1195 { 1196 struct file *filp = NULL; 1197 gfp_t gfp = lo->old_gfp_mask; 1198 struct block_device *bdev = lo->lo_device; 1199 int err = 0; 1200 bool partscan = false; 1201 int lo_number; 1202 1203 mutex_lock(&lo->lo_mutex); 1204 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1205 err = -ENXIO; 1206 goto out_unlock; 1207 } 1208 1209 filp = lo->lo_backing_file; 1210 if (filp == NULL) { 1211 err = -EINVAL; 1212 goto out_unlock; 1213 } 1214 1215 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1216 blk_queue_write_cache(lo->lo_queue, false, false); 1217 1218 /* freeze request queue during the transition */ 1219 blk_mq_freeze_queue(lo->lo_queue); 1220 1221 spin_lock_irq(&lo->lo_lock); 1222 lo->lo_backing_file = NULL; 1223 spin_unlock_irq(&lo->lo_lock); 1224 1225 loop_release_xfer(lo); 1226 lo->transfer = NULL; 1227 lo->ioctl = NULL; 1228 lo->lo_device = NULL; 1229 lo->lo_encryption = NULL; 1230 lo->lo_offset = 0; 1231 lo->lo_sizelimit = 0; 1232 lo->lo_encrypt_key_size = 0; 1233 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1234 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1235 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1236 blk_queue_logical_block_size(lo->lo_queue, 512); 1237 blk_queue_physical_block_size(lo->lo_queue, 512); 1238 blk_queue_io_min(lo->lo_queue, 512); 1239 if (bdev) { 1240 bdput(bdev); 1241 invalidate_bdev(bdev); 1242 bdev->bd_inode->i_mapping->wb_err = 0; 1243 } 1244 set_capacity(lo->lo_disk, 0); 1245 loop_sysfs_exit(lo); 1246 if (bdev) { 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(&lo->lo_mutex); 1260 if (partscan) { 1261 /* 1262 * open_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(&lo->lo_disk->open_mutex); 1271 err = bdev_disk_changed(lo->lo_disk, false); 1272 if (!release) 1273 mutex_unlock(&lo->lo_disk->open_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(&lo->lo_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(&lo->lo_mutex); 1296 1297 /* 1298 * Need not hold lo_mutex to fput backing file. Calling fput holding 1299 * lo_mutex triggers a circular lock dependency possibility warning as 1300 * fput can take open_mutex which is usually taken before lo_mutex. 1301 */ 1302 if (filp) 1303 fput(filp); 1304 return err; 1305 } 1306 1307 static int loop_clr_fd(struct loop_device *lo) 1308 { 1309 int err; 1310 1311 err = mutex_lock_killable(&lo->lo_mutex); 1312 if (err) 1313 return err; 1314 if (lo->lo_state != Lo_bound) { 1315 mutex_unlock(&lo->lo_mutex); 1316 return -ENXIO; 1317 } 1318 /* 1319 * If we've explicitly asked to tear down the loop device, 1320 * and it has an elevated reference count, set it for auto-teardown when 1321 * the last reference goes away. This stops $!~#$@ udev from 1322 * preventing teardown because it decided that it needs to run blkid on 1323 * the loopback device whenever they appear. xfstests is notorious for 1324 * failing tests because blkid via udev races with a losetup 1325 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1326 * command to fail with EBUSY. 1327 */ 1328 if (atomic_read(&lo->lo_refcnt) > 1) { 1329 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1330 mutex_unlock(&lo->lo_mutex); 1331 return 0; 1332 } 1333 lo->lo_state = Lo_rundown; 1334 mutex_unlock(&lo->lo_mutex); 1335 1336 return __loop_clr_fd(lo, false); 1337 } 1338 1339 static int 1340 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1341 { 1342 int err; 1343 struct block_device *bdev; 1344 kuid_t uid = current_uid(); 1345 int prev_lo_flags; 1346 bool partscan = false; 1347 bool size_changed = false; 1348 1349 err = mutex_lock_killable(&lo->lo_mutex); 1350 if (err) 1351 return err; 1352 if (lo->lo_encrypt_key_size && 1353 !uid_eq(lo->lo_key_owner, uid) && 1354 !capable(CAP_SYS_ADMIN)) { 1355 err = -EPERM; 1356 goto out_unlock; 1357 } 1358 if (lo->lo_state != Lo_bound) { 1359 err = -ENXIO; 1360 goto out_unlock; 1361 } 1362 1363 if (lo->lo_offset != info->lo_offset || 1364 lo->lo_sizelimit != info->lo_sizelimit) { 1365 size_changed = true; 1366 sync_blockdev(lo->lo_device); 1367 invalidate_bdev(lo->lo_device); 1368 } 1369 1370 /* I/O need to be drained during transfer transition */ 1371 blk_mq_freeze_queue(lo->lo_queue); 1372 1373 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1374 /* If any pages were dirtied after invalidate_bdev(), try again */ 1375 err = -EAGAIN; 1376 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1377 __func__, lo->lo_number, lo->lo_file_name, 1378 lo->lo_device->bd_inode->i_mapping->nrpages); 1379 goto out_unfreeze; 1380 } 1381 1382 prev_lo_flags = lo->lo_flags; 1383 1384 err = loop_set_status_from_info(lo, info); 1385 if (err) 1386 goto out_unfreeze; 1387 1388 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1389 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1390 /* For those flags, use the previous values instead */ 1391 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1392 /* For flags that can't be cleared, use previous values too */ 1393 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1394 1395 if (size_changed) { 1396 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1397 lo->lo_backing_file); 1398 loop_set_size(lo, new_size); 1399 } 1400 1401 loop_config_discard(lo); 1402 1403 /* update dio if lo_offset or transfer is changed */ 1404 __loop_update_dio(lo, lo->use_dio); 1405 1406 out_unfreeze: 1407 blk_mq_unfreeze_queue(lo->lo_queue); 1408 1409 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1410 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1411 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1412 bdev = lo->lo_device; 1413 partscan = true; 1414 } 1415 out_unlock: 1416 mutex_unlock(&lo->lo_mutex); 1417 if (partscan) 1418 loop_reread_partitions(lo); 1419 1420 return err; 1421 } 1422 1423 static int 1424 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1425 { 1426 struct path path; 1427 struct kstat stat; 1428 int ret; 1429 1430 ret = mutex_lock_killable(&lo->lo_mutex); 1431 if (ret) 1432 return ret; 1433 if (lo->lo_state != Lo_bound) { 1434 mutex_unlock(&lo->lo_mutex); 1435 return -ENXIO; 1436 } 1437 1438 memset(info, 0, sizeof(*info)); 1439 info->lo_number = lo->lo_number; 1440 info->lo_offset = lo->lo_offset; 1441 info->lo_sizelimit = lo->lo_sizelimit; 1442 info->lo_flags = lo->lo_flags; 1443 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1444 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1445 info->lo_encrypt_type = 1446 lo->lo_encryption ? lo->lo_encryption->number : 0; 1447 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1448 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1449 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1450 lo->lo_encrypt_key_size); 1451 } 1452 1453 /* Drop lo_mutex while we call into the filesystem. */ 1454 path = lo->lo_backing_file->f_path; 1455 path_get(&path); 1456 mutex_unlock(&lo->lo_mutex); 1457 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1458 if (!ret) { 1459 info->lo_device = huge_encode_dev(stat.dev); 1460 info->lo_inode = stat.ino; 1461 info->lo_rdevice = huge_encode_dev(stat.rdev); 1462 } 1463 path_put(&path); 1464 return ret; 1465 } 1466 1467 static void 1468 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1469 { 1470 memset(info64, 0, sizeof(*info64)); 1471 info64->lo_number = info->lo_number; 1472 info64->lo_device = info->lo_device; 1473 info64->lo_inode = info->lo_inode; 1474 info64->lo_rdevice = info->lo_rdevice; 1475 info64->lo_offset = info->lo_offset; 1476 info64->lo_sizelimit = 0; 1477 info64->lo_encrypt_type = info->lo_encrypt_type; 1478 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1479 info64->lo_flags = info->lo_flags; 1480 info64->lo_init[0] = info->lo_init[0]; 1481 info64->lo_init[1] = info->lo_init[1]; 1482 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1483 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1484 else 1485 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1486 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1487 } 1488 1489 static int 1490 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1491 { 1492 memset(info, 0, sizeof(*info)); 1493 info->lo_number = info64->lo_number; 1494 info->lo_device = info64->lo_device; 1495 info->lo_inode = info64->lo_inode; 1496 info->lo_rdevice = info64->lo_rdevice; 1497 info->lo_offset = info64->lo_offset; 1498 info->lo_encrypt_type = info64->lo_encrypt_type; 1499 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1500 info->lo_flags = info64->lo_flags; 1501 info->lo_init[0] = info64->lo_init[0]; 1502 info->lo_init[1] = info64->lo_init[1]; 1503 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1504 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1505 else 1506 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1507 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1508 1509 /* error in case values were truncated */ 1510 if (info->lo_device != info64->lo_device || 1511 info->lo_rdevice != info64->lo_rdevice || 1512 info->lo_inode != info64->lo_inode || 1513 info->lo_offset != info64->lo_offset) 1514 return -EOVERFLOW; 1515 1516 return 0; 1517 } 1518 1519 static int 1520 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1521 { 1522 struct loop_info info; 1523 struct loop_info64 info64; 1524 1525 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1526 return -EFAULT; 1527 loop_info64_from_old(&info, &info64); 1528 return loop_set_status(lo, &info64); 1529 } 1530 1531 static int 1532 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1533 { 1534 struct loop_info64 info64; 1535 1536 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1537 return -EFAULT; 1538 return loop_set_status(lo, &info64); 1539 } 1540 1541 static int 1542 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1543 struct loop_info info; 1544 struct loop_info64 info64; 1545 int err; 1546 1547 if (!arg) 1548 return -EINVAL; 1549 err = loop_get_status(lo, &info64); 1550 if (!err) 1551 err = loop_info64_to_old(&info64, &info); 1552 if (!err && copy_to_user(arg, &info, sizeof(info))) 1553 err = -EFAULT; 1554 1555 return err; 1556 } 1557 1558 static int 1559 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1560 struct loop_info64 info64; 1561 int err; 1562 1563 if (!arg) 1564 return -EINVAL; 1565 err = loop_get_status(lo, &info64); 1566 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1567 err = -EFAULT; 1568 1569 return err; 1570 } 1571 1572 static int loop_set_capacity(struct loop_device *lo) 1573 { 1574 loff_t size; 1575 1576 if (unlikely(lo->lo_state != Lo_bound)) 1577 return -ENXIO; 1578 1579 size = get_loop_size(lo, lo->lo_backing_file); 1580 loop_set_size(lo, size); 1581 1582 return 0; 1583 } 1584 1585 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1586 { 1587 int error = -ENXIO; 1588 if (lo->lo_state != Lo_bound) 1589 goto out; 1590 1591 __loop_update_dio(lo, !!arg); 1592 if (lo->use_dio == !!arg) 1593 return 0; 1594 error = -EINVAL; 1595 out: 1596 return error; 1597 } 1598 1599 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1600 { 1601 int err = 0; 1602 1603 if (lo->lo_state != Lo_bound) 1604 return -ENXIO; 1605 1606 err = loop_validate_block_size(arg); 1607 if (err) 1608 return err; 1609 1610 if (lo->lo_queue->limits.logical_block_size == arg) 1611 return 0; 1612 1613 sync_blockdev(lo->lo_device); 1614 invalidate_bdev(lo->lo_device); 1615 1616 blk_mq_freeze_queue(lo->lo_queue); 1617 1618 /* invalidate_bdev should have truncated all the pages */ 1619 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1620 err = -EAGAIN; 1621 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1622 __func__, lo->lo_number, lo->lo_file_name, 1623 lo->lo_device->bd_inode->i_mapping->nrpages); 1624 goto out_unfreeze; 1625 } 1626 1627 blk_queue_logical_block_size(lo->lo_queue, arg); 1628 blk_queue_physical_block_size(lo->lo_queue, arg); 1629 blk_queue_io_min(lo->lo_queue, arg); 1630 loop_update_dio(lo); 1631 out_unfreeze: 1632 blk_mq_unfreeze_queue(lo->lo_queue); 1633 1634 return err; 1635 } 1636 1637 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1638 unsigned long arg) 1639 { 1640 int err; 1641 1642 err = mutex_lock_killable(&lo->lo_mutex); 1643 if (err) 1644 return err; 1645 switch (cmd) { 1646 case LOOP_SET_CAPACITY: 1647 err = loop_set_capacity(lo); 1648 break; 1649 case LOOP_SET_DIRECT_IO: 1650 err = loop_set_dio(lo, arg); 1651 break; 1652 case LOOP_SET_BLOCK_SIZE: 1653 err = loop_set_block_size(lo, arg); 1654 break; 1655 default: 1656 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1657 } 1658 mutex_unlock(&lo->lo_mutex); 1659 return err; 1660 } 1661 1662 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1663 unsigned int cmd, unsigned long arg) 1664 { 1665 struct loop_device *lo = bdev->bd_disk->private_data; 1666 void __user *argp = (void __user *) arg; 1667 int err; 1668 1669 switch (cmd) { 1670 case LOOP_SET_FD: { 1671 /* 1672 * Legacy case - pass in a zeroed out struct loop_config with 1673 * only the file descriptor set , which corresponds with the 1674 * default parameters we'd have used otherwise. 1675 */ 1676 struct loop_config config; 1677 1678 memset(&config, 0, sizeof(config)); 1679 config.fd = arg; 1680 1681 return loop_configure(lo, mode, bdev, &config); 1682 } 1683 case LOOP_CONFIGURE: { 1684 struct loop_config config; 1685 1686 if (copy_from_user(&config, argp, sizeof(config))) 1687 return -EFAULT; 1688 1689 return loop_configure(lo, mode, bdev, &config); 1690 } 1691 case LOOP_CHANGE_FD: 1692 return loop_change_fd(lo, bdev, arg); 1693 case LOOP_CLR_FD: 1694 return loop_clr_fd(lo); 1695 case LOOP_SET_STATUS: 1696 err = -EPERM; 1697 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1698 err = loop_set_status_old(lo, argp); 1699 } 1700 break; 1701 case LOOP_GET_STATUS: 1702 return loop_get_status_old(lo, argp); 1703 case LOOP_SET_STATUS64: 1704 err = -EPERM; 1705 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1706 err = loop_set_status64(lo, argp); 1707 } 1708 break; 1709 case LOOP_GET_STATUS64: 1710 return loop_get_status64(lo, argp); 1711 case LOOP_SET_CAPACITY: 1712 case LOOP_SET_DIRECT_IO: 1713 case LOOP_SET_BLOCK_SIZE: 1714 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1715 return -EPERM; 1716 fallthrough; 1717 default: 1718 err = lo_simple_ioctl(lo, cmd, arg); 1719 break; 1720 } 1721 1722 return err; 1723 } 1724 1725 #ifdef CONFIG_COMPAT 1726 struct compat_loop_info { 1727 compat_int_t lo_number; /* ioctl r/o */ 1728 compat_dev_t lo_device; /* ioctl r/o */ 1729 compat_ulong_t lo_inode; /* ioctl r/o */ 1730 compat_dev_t lo_rdevice; /* ioctl r/o */ 1731 compat_int_t lo_offset; 1732 compat_int_t lo_encrypt_type; 1733 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1734 compat_int_t lo_flags; /* ioctl r/o */ 1735 char lo_name[LO_NAME_SIZE]; 1736 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1737 compat_ulong_t lo_init[2]; 1738 char reserved[4]; 1739 }; 1740 1741 /* 1742 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1743 * - noinlined to reduce stack space usage in main part of driver 1744 */ 1745 static noinline int 1746 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1747 struct loop_info64 *info64) 1748 { 1749 struct compat_loop_info info; 1750 1751 if (copy_from_user(&info, arg, sizeof(info))) 1752 return -EFAULT; 1753 1754 memset(info64, 0, sizeof(*info64)); 1755 info64->lo_number = info.lo_number; 1756 info64->lo_device = info.lo_device; 1757 info64->lo_inode = info.lo_inode; 1758 info64->lo_rdevice = info.lo_rdevice; 1759 info64->lo_offset = info.lo_offset; 1760 info64->lo_sizelimit = 0; 1761 info64->lo_encrypt_type = info.lo_encrypt_type; 1762 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1763 info64->lo_flags = info.lo_flags; 1764 info64->lo_init[0] = info.lo_init[0]; 1765 info64->lo_init[1] = info.lo_init[1]; 1766 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1767 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1768 else 1769 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1770 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1771 return 0; 1772 } 1773 1774 /* 1775 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1776 * - noinlined to reduce stack space usage in main part of driver 1777 */ 1778 static noinline int 1779 loop_info64_to_compat(const struct loop_info64 *info64, 1780 struct compat_loop_info __user *arg) 1781 { 1782 struct compat_loop_info info; 1783 1784 memset(&info, 0, sizeof(info)); 1785 info.lo_number = info64->lo_number; 1786 info.lo_device = info64->lo_device; 1787 info.lo_inode = info64->lo_inode; 1788 info.lo_rdevice = info64->lo_rdevice; 1789 info.lo_offset = info64->lo_offset; 1790 info.lo_encrypt_type = info64->lo_encrypt_type; 1791 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1792 info.lo_flags = info64->lo_flags; 1793 info.lo_init[0] = info64->lo_init[0]; 1794 info.lo_init[1] = info64->lo_init[1]; 1795 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1796 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1797 else 1798 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1799 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1800 1801 /* error in case values were truncated */ 1802 if (info.lo_device != info64->lo_device || 1803 info.lo_rdevice != info64->lo_rdevice || 1804 info.lo_inode != info64->lo_inode || 1805 info.lo_offset != info64->lo_offset || 1806 info.lo_init[0] != info64->lo_init[0] || 1807 info.lo_init[1] != info64->lo_init[1]) 1808 return -EOVERFLOW; 1809 1810 if (copy_to_user(arg, &info, sizeof(info))) 1811 return -EFAULT; 1812 return 0; 1813 } 1814 1815 static int 1816 loop_set_status_compat(struct loop_device *lo, 1817 const struct compat_loop_info __user *arg) 1818 { 1819 struct loop_info64 info64; 1820 int ret; 1821 1822 ret = loop_info64_from_compat(arg, &info64); 1823 if (ret < 0) 1824 return ret; 1825 return loop_set_status(lo, &info64); 1826 } 1827 1828 static int 1829 loop_get_status_compat(struct loop_device *lo, 1830 struct compat_loop_info __user *arg) 1831 { 1832 struct loop_info64 info64; 1833 int err; 1834 1835 if (!arg) 1836 return -EINVAL; 1837 err = loop_get_status(lo, &info64); 1838 if (!err) 1839 err = loop_info64_to_compat(&info64, arg); 1840 return err; 1841 } 1842 1843 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1844 unsigned int cmd, unsigned long arg) 1845 { 1846 struct loop_device *lo = bdev->bd_disk->private_data; 1847 int err; 1848 1849 switch(cmd) { 1850 case LOOP_SET_STATUS: 1851 err = loop_set_status_compat(lo, 1852 (const struct compat_loop_info __user *)arg); 1853 break; 1854 case LOOP_GET_STATUS: 1855 err = loop_get_status_compat(lo, 1856 (struct compat_loop_info __user *)arg); 1857 break; 1858 case LOOP_SET_CAPACITY: 1859 case LOOP_CLR_FD: 1860 case LOOP_GET_STATUS64: 1861 case LOOP_SET_STATUS64: 1862 case LOOP_CONFIGURE: 1863 arg = (unsigned long) compat_ptr(arg); 1864 fallthrough; 1865 case LOOP_SET_FD: 1866 case LOOP_CHANGE_FD: 1867 case LOOP_SET_BLOCK_SIZE: 1868 case LOOP_SET_DIRECT_IO: 1869 err = lo_ioctl(bdev, mode, cmd, arg); 1870 break; 1871 default: 1872 err = -ENOIOCTLCMD; 1873 break; 1874 } 1875 return err; 1876 } 1877 #endif 1878 1879 static int lo_open(struct block_device *bdev, fmode_t mode) 1880 { 1881 struct loop_device *lo; 1882 int err; 1883 1884 /* 1885 * take loop_ctl_mutex to protect lo pointer from race with 1886 * loop_control_ioctl(LOOP_CTL_REMOVE), however, to reduce contention 1887 * release it prior to updating lo->lo_refcnt. 1888 */ 1889 err = mutex_lock_killable(&loop_ctl_mutex); 1890 if (err) 1891 return err; 1892 lo = bdev->bd_disk->private_data; 1893 if (!lo) { 1894 mutex_unlock(&loop_ctl_mutex); 1895 return -ENXIO; 1896 } 1897 err = mutex_lock_killable(&lo->lo_mutex); 1898 mutex_unlock(&loop_ctl_mutex); 1899 if (err) 1900 return err; 1901 atomic_inc(&lo->lo_refcnt); 1902 mutex_unlock(&lo->lo_mutex); 1903 return 0; 1904 } 1905 1906 static void lo_release(struct gendisk *disk, fmode_t mode) 1907 { 1908 struct loop_device *lo = disk->private_data; 1909 1910 mutex_lock(&lo->lo_mutex); 1911 if (atomic_dec_return(&lo->lo_refcnt)) 1912 goto out_unlock; 1913 1914 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1915 if (lo->lo_state != Lo_bound) 1916 goto out_unlock; 1917 lo->lo_state = Lo_rundown; 1918 mutex_unlock(&lo->lo_mutex); 1919 /* 1920 * In autoclear mode, stop the loop thread 1921 * and remove configuration after last close. 1922 */ 1923 __loop_clr_fd(lo, true); 1924 return; 1925 } else if (lo->lo_state == Lo_bound) { 1926 /* 1927 * Otherwise keep thread (if running) and config, 1928 * but flush possible ongoing bios in thread. 1929 */ 1930 blk_mq_freeze_queue(lo->lo_queue); 1931 blk_mq_unfreeze_queue(lo->lo_queue); 1932 } 1933 1934 out_unlock: 1935 mutex_unlock(&lo->lo_mutex); 1936 } 1937 1938 static const struct block_device_operations lo_fops = { 1939 .owner = THIS_MODULE, 1940 .open = lo_open, 1941 .release = lo_release, 1942 .ioctl = lo_ioctl, 1943 #ifdef CONFIG_COMPAT 1944 .compat_ioctl = lo_compat_ioctl, 1945 #endif 1946 }; 1947 1948 /* 1949 * And now the modules code and kernel interface. 1950 */ 1951 static int max_loop; 1952 module_param(max_loop, int, 0444); 1953 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1954 module_param(max_part, int, 0444); 1955 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1956 MODULE_LICENSE("GPL"); 1957 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1958 1959 int loop_register_transfer(struct loop_func_table *funcs) 1960 { 1961 unsigned int n = funcs->number; 1962 1963 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1964 return -EINVAL; 1965 xfer_funcs[n] = funcs; 1966 return 0; 1967 } 1968 1969 static int unregister_transfer_cb(int id, void *ptr, void *data) 1970 { 1971 struct loop_device *lo = ptr; 1972 struct loop_func_table *xfer = data; 1973 1974 mutex_lock(&lo->lo_mutex); 1975 if (lo->lo_encryption == xfer) 1976 loop_release_xfer(lo); 1977 mutex_unlock(&lo->lo_mutex); 1978 return 0; 1979 } 1980 1981 int loop_unregister_transfer(int number) 1982 { 1983 unsigned int n = number; 1984 struct loop_func_table *xfer; 1985 1986 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1987 return -EINVAL; 1988 1989 xfer_funcs[n] = NULL; 1990 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1991 return 0; 1992 } 1993 1994 EXPORT_SYMBOL(loop_register_transfer); 1995 EXPORT_SYMBOL(loop_unregister_transfer); 1996 1997 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1998 const struct blk_mq_queue_data *bd) 1999 { 2000 struct request *rq = bd->rq; 2001 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2002 struct loop_device *lo = rq->q->queuedata; 2003 2004 blk_mq_start_request(rq); 2005 2006 if (lo->lo_state != Lo_bound) 2007 return BLK_STS_IOERR; 2008 2009 switch (req_op(rq)) { 2010 case REQ_OP_FLUSH: 2011 case REQ_OP_DISCARD: 2012 case REQ_OP_WRITE_ZEROES: 2013 cmd->use_aio = false; 2014 break; 2015 default: 2016 cmd->use_aio = lo->use_dio; 2017 break; 2018 } 2019 2020 /* always use the first bio's css */ 2021 #ifdef CONFIG_BLK_CGROUP 2022 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 2023 cmd->css = &bio_blkcg(rq->bio)->css; 2024 css_get(cmd->css); 2025 } else 2026 #endif 2027 cmd->css = NULL; 2028 kthread_queue_work(&lo->worker, &cmd->work); 2029 2030 return BLK_STS_OK; 2031 } 2032 2033 static void loop_handle_cmd(struct loop_cmd *cmd) 2034 { 2035 struct request *rq = blk_mq_rq_from_pdu(cmd); 2036 const bool write = op_is_write(req_op(rq)); 2037 struct loop_device *lo = rq->q->queuedata; 2038 int ret = 0; 2039 2040 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 2041 ret = -EIO; 2042 goto failed; 2043 } 2044 2045 ret = do_req_filebacked(lo, rq); 2046 failed: 2047 /* complete non-aio request */ 2048 if (!cmd->use_aio || ret) { 2049 if (ret == -EOPNOTSUPP) 2050 cmd->ret = ret; 2051 else 2052 cmd->ret = ret ? -EIO : 0; 2053 if (likely(!blk_should_fake_timeout(rq->q))) 2054 blk_mq_complete_request(rq); 2055 } 2056 } 2057 2058 static void loop_queue_work(struct kthread_work *work) 2059 { 2060 struct loop_cmd *cmd = 2061 container_of(work, struct loop_cmd, work); 2062 2063 loop_handle_cmd(cmd); 2064 } 2065 2066 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 2067 unsigned int hctx_idx, unsigned int numa_node) 2068 { 2069 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2070 2071 kthread_init_work(&cmd->work, loop_queue_work); 2072 return 0; 2073 } 2074 2075 static const struct blk_mq_ops loop_mq_ops = { 2076 .queue_rq = loop_queue_rq, 2077 .init_request = loop_init_request, 2078 .complete = lo_complete_rq, 2079 }; 2080 2081 static int loop_add(struct loop_device **l, int i) 2082 { 2083 struct loop_device *lo; 2084 struct gendisk *disk; 2085 int err; 2086 2087 err = -ENOMEM; 2088 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 2089 if (!lo) 2090 goto out; 2091 2092 lo->lo_state = Lo_unbound; 2093 2094 /* allocate id, if @id >= 0, we're requesting that specific id */ 2095 if (i >= 0) { 2096 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2097 if (err == -ENOSPC) 2098 err = -EEXIST; 2099 } else { 2100 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2101 } 2102 if (err < 0) 2103 goto out_free_dev; 2104 i = err; 2105 2106 err = -ENOMEM; 2107 lo->tag_set.ops = &loop_mq_ops; 2108 lo->tag_set.nr_hw_queues = 1; 2109 lo->tag_set.queue_depth = 128; 2110 lo->tag_set.numa_node = NUMA_NO_NODE; 2111 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2112 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING; 2113 lo->tag_set.driver_data = lo; 2114 2115 err = blk_mq_alloc_tag_set(&lo->tag_set); 2116 if (err) 2117 goto out_free_idr; 2118 2119 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 2120 if (IS_ERR(disk)) { 2121 err = PTR_ERR(disk); 2122 goto out_cleanup_tags; 2123 } 2124 lo->lo_queue = lo->lo_disk->queue; 2125 2126 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2127 2128 /* 2129 * By default, we do buffer IO, so it doesn't make sense to enable 2130 * merge because the I/O submitted to backing file is handled page by 2131 * page. For directio mode, merge does help to dispatch bigger request 2132 * to underlayer disk. We will enable merge once directio is enabled. 2133 */ 2134 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_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 mutex_init(&lo->lo_mutex); 2159 lo->lo_number = i; 2160 spin_lock_init(&lo->lo_lock); 2161 disk->major = LOOP_MAJOR; 2162 disk->first_minor = i << part_shift; 2163 disk->minors = 1 << 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_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_disk(lo->lo_disk); 2186 blk_mq_free_tag_set(&lo->tag_set); 2187 mutex_destroy(&lo->lo_mutex); 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 ret = mutex_lock_killable(&lo->lo_mutex); 2268 if (ret) 2269 break; 2270 if (lo->lo_state != Lo_unbound) { 2271 ret = -EBUSY; 2272 mutex_unlock(&lo->lo_mutex); 2273 break; 2274 } 2275 if (atomic_read(&lo->lo_refcnt) > 0) { 2276 ret = -EBUSY; 2277 mutex_unlock(&lo->lo_mutex); 2278 break; 2279 } 2280 lo->lo_disk->private_data = NULL; 2281 mutex_unlock(&lo->lo_mutex); 2282 idr_remove(&loop_index_idr, lo->lo_number); 2283 loop_remove(lo); 2284 break; 2285 case LOOP_CTL_GET_FREE: 2286 ret = loop_lookup(&lo, -1); 2287 if (ret >= 0) 2288 break; 2289 ret = loop_add(&lo, -1); 2290 } 2291 mutex_unlock(&loop_ctl_mutex); 2292 2293 return ret; 2294 } 2295 2296 static const struct file_operations loop_ctl_fops = { 2297 .open = nonseekable_open, 2298 .unlocked_ioctl = loop_control_ioctl, 2299 .compat_ioctl = loop_control_ioctl, 2300 .owner = THIS_MODULE, 2301 .llseek = noop_llseek, 2302 }; 2303 2304 static struct miscdevice loop_misc = { 2305 .minor = LOOP_CTRL_MINOR, 2306 .name = "loop-control", 2307 .fops = &loop_ctl_fops, 2308 }; 2309 2310 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2311 MODULE_ALIAS("devname:loop-control"); 2312 2313 static int __init loop_init(void) 2314 { 2315 int i, nr; 2316 struct loop_device *lo; 2317 int err; 2318 2319 part_shift = 0; 2320 if (max_part > 0) { 2321 part_shift = fls(max_part); 2322 2323 /* 2324 * Adjust max_part according to part_shift as it is exported 2325 * to user space so that user can decide correct minor number 2326 * if [s]he want to create more devices. 2327 * 2328 * Note that -1 is required because partition 0 is reserved 2329 * for the whole disk. 2330 */ 2331 max_part = (1UL << part_shift) - 1; 2332 } 2333 2334 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2335 err = -EINVAL; 2336 goto err_out; 2337 } 2338 2339 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2340 err = -EINVAL; 2341 goto err_out; 2342 } 2343 2344 /* 2345 * If max_loop is specified, create that many devices upfront. 2346 * This also becomes a hard limit. If max_loop is not specified, 2347 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2348 * init time. Loop devices can be requested on-demand with the 2349 * /dev/loop-control interface, or be instantiated by accessing 2350 * a 'dead' device node. 2351 */ 2352 if (max_loop) 2353 nr = max_loop; 2354 else 2355 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 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