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/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 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) 87 88 static DEFINE_IDR(loop_index_idr); 89 static DEFINE_MUTEX(loop_ctl_mutex); 90 91 static int max_part; 92 static int part_shift; 93 94 static int transfer_xor(struct loop_device *lo, int cmd, 95 struct page *raw_page, unsigned raw_off, 96 struct page *loop_page, unsigned loop_off, 97 int size, sector_t real_block) 98 { 99 char *raw_buf = kmap_atomic(raw_page) + raw_off; 100 char *loop_buf = kmap_atomic(loop_page) + loop_off; 101 char *in, *out, *key; 102 int i, keysize; 103 104 if (cmd == READ) { 105 in = raw_buf; 106 out = loop_buf; 107 } else { 108 in = loop_buf; 109 out = raw_buf; 110 } 111 112 key = lo->lo_encrypt_key; 113 keysize = lo->lo_encrypt_key_size; 114 for (i = 0; i < size; i++) 115 *out++ = *in++ ^ key[(i & 511) % keysize]; 116 117 kunmap_atomic(loop_buf); 118 kunmap_atomic(raw_buf); 119 cond_resched(); 120 return 0; 121 } 122 123 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 124 { 125 if (unlikely(info->lo_encrypt_key_size <= 0)) 126 return -EINVAL; 127 return 0; 128 } 129 130 static struct loop_func_table none_funcs = { 131 .number = LO_CRYPT_NONE, 132 }; 133 134 static struct loop_func_table xor_funcs = { 135 .number = LO_CRYPT_XOR, 136 .transfer = transfer_xor, 137 .init = xor_init 138 }; 139 140 /* xfer_funcs[0] is special - its release function is never called */ 141 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 142 &none_funcs, 143 &xor_funcs 144 }; 145 146 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 147 { 148 loff_t loopsize; 149 150 /* Compute loopsize in bytes */ 151 loopsize = i_size_read(file->f_mapping->host); 152 if (offset > 0) 153 loopsize -= offset; 154 /* offset is beyond i_size, weird but possible */ 155 if (loopsize < 0) 156 return 0; 157 158 if (sizelimit > 0 && sizelimit < loopsize) 159 loopsize = sizelimit; 160 /* 161 * Unfortunately, if we want to do I/O on the device, 162 * the number of 512-byte sectors has to fit into a sector_t. 163 */ 164 return loopsize >> 9; 165 } 166 167 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 168 { 169 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 170 } 171 172 static void __loop_update_dio(struct loop_device *lo, bool dio) 173 { 174 struct file *file = lo->lo_backing_file; 175 struct address_space *mapping = file->f_mapping; 176 struct inode *inode = mapping->host; 177 unsigned short sb_bsize = 0; 178 unsigned dio_align = 0; 179 bool use_dio; 180 181 if (inode->i_sb->s_bdev) { 182 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 183 dio_align = sb_bsize - 1; 184 } 185 186 /* 187 * We support direct I/O only if lo_offset is aligned with the 188 * logical I/O size of backing device, and the logical block 189 * size of loop is bigger than the backing device's and the loop 190 * needn't transform transfer. 191 * 192 * TODO: the above condition may be loosed in the future, and 193 * direct I/O may be switched runtime at that time because most 194 * of requests in sane applications should be PAGE_SIZE aligned 195 */ 196 if (dio) { 197 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 198 !(lo->lo_offset & dio_align) && 199 mapping->a_ops->direct_IO && 200 !lo->transfer) 201 use_dio = true; 202 else 203 use_dio = false; 204 } else { 205 use_dio = false; 206 } 207 208 if (lo->use_dio == use_dio) 209 return; 210 211 /* flush dirty pages before changing direct IO */ 212 vfs_fsync(file, 0); 213 214 /* 215 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 216 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 217 * will get updated by ioctl(LOOP_GET_STATUS) 218 */ 219 if (lo->lo_state == Lo_bound) 220 blk_mq_freeze_queue(lo->lo_queue); 221 lo->use_dio = use_dio; 222 if (use_dio) { 223 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 224 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 225 } else { 226 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 227 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 228 } 229 if (lo->lo_state == Lo_bound) 230 blk_mq_unfreeze_queue(lo->lo_queue); 231 } 232 233 /** 234 * loop_validate_block_size() - validates the passed in block size 235 * @bsize: size to validate 236 */ 237 static int 238 loop_validate_block_size(unsigned short bsize) 239 { 240 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize)) 241 return -EINVAL; 242 243 return 0; 244 } 245 246 /** 247 * loop_set_size() - sets device size and notifies userspace 248 * @lo: struct loop_device to set the size for 249 * @size: new size of the loop device 250 * 251 * Callers must validate that the size passed into this function fits into 252 * a sector_t, eg using loop_validate_size() 253 */ 254 static void loop_set_size(struct loop_device *lo, loff_t size) 255 { 256 if (!set_capacity_and_notify(lo->lo_disk, size)) 257 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 258 } 259 260 static inline int 261 lo_do_transfer(struct loop_device *lo, int cmd, 262 struct page *rpage, unsigned roffs, 263 struct page *lpage, unsigned loffs, 264 int size, sector_t rblock) 265 { 266 int ret; 267 268 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 269 if (likely(!ret)) 270 return 0; 271 272 printk_ratelimited(KERN_ERR 273 "loop: Transfer error at byte offset %llu, length %i.\n", 274 (unsigned long long)rblock << 9, size); 275 return ret; 276 } 277 278 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 279 { 280 struct iov_iter i; 281 ssize_t bw; 282 283 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 284 285 file_start_write(file); 286 bw = vfs_iter_write(file, &i, ppos, 0); 287 file_end_write(file); 288 289 if (likely(bw == bvec->bv_len)) 290 return 0; 291 292 printk_ratelimited(KERN_ERR 293 "loop: Write error at byte offset %llu, length %i.\n", 294 (unsigned long long)*ppos, bvec->bv_len); 295 if (bw >= 0) 296 bw = -EIO; 297 return bw; 298 } 299 300 static int lo_write_simple(struct loop_device *lo, struct request *rq, 301 loff_t pos) 302 { 303 struct bio_vec bvec; 304 struct req_iterator iter; 305 int ret = 0; 306 307 rq_for_each_segment(bvec, rq, iter) { 308 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 309 if (ret < 0) 310 break; 311 cond_resched(); 312 } 313 314 return ret; 315 } 316 317 /* 318 * This is the slow, transforming version that needs to double buffer the 319 * data as it cannot do the transformations in place without having direct 320 * access to the destination pages of the backing file. 321 */ 322 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 323 loff_t pos) 324 { 325 struct bio_vec bvec, b; 326 struct req_iterator iter; 327 struct page *page; 328 int ret = 0; 329 330 page = alloc_page(GFP_NOIO); 331 if (unlikely(!page)) 332 return -ENOMEM; 333 334 rq_for_each_segment(bvec, rq, iter) { 335 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 336 bvec.bv_offset, bvec.bv_len, pos >> 9); 337 if (unlikely(ret)) 338 break; 339 340 b.bv_page = page; 341 b.bv_offset = 0; 342 b.bv_len = bvec.bv_len; 343 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 344 if (ret < 0) 345 break; 346 } 347 348 __free_page(page); 349 return ret; 350 } 351 352 static int lo_read_simple(struct loop_device *lo, struct request *rq, 353 loff_t pos) 354 { 355 struct bio_vec bvec; 356 struct req_iterator iter; 357 struct iov_iter i; 358 ssize_t len; 359 360 rq_for_each_segment(bvec, rq, iter) { 361 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 362 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 363 if (len < 0) 364 return len; 365 366 flush_dcache_page(bvec.bv_page); 367 368 if (len != bvec.bv_len) { 369 struct bio *bio; 370 371 __rq_for_each_bio(bio, rq) 372 zero_fill_bio(bio); 373 break; 374 } 375 cond_resched(); 376 } 377 378 return 0; 379 } 380 381 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 382 loff_t pos) 383 { 384 struct bio_vec bvec, b; 385 struct req_iterator iter; 386 struct iov_iter i; 387 struct page *page; 388 ssize_t len; 389 int ret = 0; 390 391 page = alloc_page(GFP_NOIO); 392 if (unlikely(!page)) 393 return -ENOMEM; 394 395 rq_for_each_segment(bvec, rq, iter) { 396 loff_t offset = pos; 397 398 b.bv_page = page; 399 b.bv_offset = 0; 400 b.bv_len = bvec.bv_len; 401 402 iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 403 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 404 if (len < 0) { 405 ret = len; 406 goto out_free_page; 407 } 408 409 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 410 bvec.bv_offset, len, offset >> 9); 411 if (ret) 412 goto out_free_page; 413 414 flush_dcache_page(bvec.bv_page); 415 416 if (len != bvec.bv_len) { 417 struct bio *bio; 418 419 __rq_for_each_bio(bio, rq) 420 zero_fill_bio(bio); 421 break; 422 } 423 } 424 425 ret = 0; 426 out_free_page: 427 __free_page(page); 428 return ret; 429 } 430 431 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 432 int mode) 433 { 434 /* 435 * We use fallocate to manipulate the space mappings used by the image 436 * a.k.a. discard/zerorange. However we do not support this if 437 * encryption is enabled, because it may give an attacker useful 438 * information. 439 */ 440 struct file *file = lo->lo_backing_file; 441 struct request_queue *q = lo->lo_queue; 442 int ret; 443 444 mode |= FALLOC_FL_KEEP_SIZE; 445 446 if (!blk_queue_discard(q)) { 447 ret = -EOPNOTSUPP; 448 goto out; 449 } 450 451 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 452 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 453 ret = -EIO; 454 out: 455 return ret; 456 } 457 458 static int lo_req_flush(struct loop_device *lo, struct request *rq) 459 { 460 struct file *file = lo->lo_backing_file; 461 int ret = vfs_fsync(file, 0); 462 if (unlikely(ret && ret != -EINVAL)) 463 ret = -EIO; 464 465 return ret; 466 } 467 468 static void lo_complete_rq(struct request *rq) 469 { 470 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 471 blk_status_t ret = BLK_STS_OK; 472 473 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 474 req_op(rq) != REQ_OP_READ) { 475 if (cmd->ret < 0) 476 ret = errno_to_blk_status(cmd->ret); 477 goto end_io; 478 } 479 480 /* 481 * Short READ - if we got some data, advance our request and 482 * retry it. If we got no data, end the rest with EIO. 483 */ 484 if (cmd->ret) { 485 blk_update_request(rq, BLK_STS_OK, cmd->ret); 486 cmd->ret = 0; 487 blk_mq_requeue_request(rq, true); 488 } else { 489 if (cmd->use_aio) { 490 struct bio *bio = rq->bio; 491 492 while (bio) { 493 zero_fill_bio(bio); 494 bio = bio->bi_next; 495 } 496 } 497 ret = BLK_STS_IOERR; 498 end_io: 499 blk_mq_end_request(rq, ret); 500 } 501 } 502 503 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 504 { 505 struct request *rq = blk_mq_rq_from_pdu(cmd); 506 507 if (!atomic_dec_and_test(&cmd->ref)) 508 return; 509 kfree(cmd->bvec); 510 cmd->bvec = NULL; 511 if (likely(!blk_should_fake_timeout(rq->q))) 512 blk_mq_complete_request(rq); 513 } 514 515 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 516 { 517 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 518 519 if (cmd->css) 520 css_put(cmd->css); 521 cmd->ret = ret; 522 lo_rw_aio_do_completion(cmd); 523 } 524 525 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 526 loff_t pos, bool rw) 527 { 528 struct iov_iter iter; 529 struct req_iterator rq_iter; 530 struct bio_vec *bvec; 531 struct request *rq = blk_mq_rq_from_pdu(cmd); 532 struct bio *bio = rq->bio; 533 struct file *file = lo->lo_backing_file; 534 struct bio_vec tmp; 535 unsigned int offset; 536 int nr_bvec = 0; 537 int ret; 538 539 rq_for_each_bvec(tmp, rq, rq_iter) 540 nr_bvec++; 541 542 if (rq->bio != rq->biotail) { 543 544 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 545 GFP_NOIO); 546 if (!bvec) 547 return -EIO; 548 cmd->bvec = bvec; 549 550 /* 551 * The bios of the request may be started from the middle of 552 * the 'bvec' because of bio splitting, so we can't directly 553 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 554 * API will take care of all details for us. 555 */ 556 rq_for_each_bvec(tmp, rq, rq_iter) { 557 *bvec = tmp; 558 bvec++; 559 } 560 bvec = cmd->bvec; 561 offset = 0; 562 } else { 563 /* 564 * Same here, this bio may be started from the middle of the 565 * 'bvec' because of bio splitting, so offset from the bvec 566 * must be passed to iov iterator 567 */ 568 offset = bio->bi_iter.bi_bvec_done; 569 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 570 } 571 atomic_set(&cmd->ref, 2); 572 573 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 574 iter.iov_offset = offset; 575 576 cmd->iocb.ki_pos = pos; 577 cmd->iocb.ki_filp = file; 578 cmd->iocb.ki_complete = lo_rw_aio_complete; 579 cmd->iocb.ki_flags = IOCB_DIRECT; 580 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 581 if (cmd->css) 582 kthread_associate_blkcg(cmd->css); 583 584 if (rw == WRITE) 585 ret = call_write_iter(file, &cmd->iocb, &iter); 586 else 587 ret = call_read_iter(file, &cmd->iocb, &iter); 588 589 lo_rw_aio_do_completion(cmd); 590 kthread_associate_blkcg(NULL); 591 592 if (ret != -EIOCBQUEUED) 593 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 594 return 0; 595 } 596 597 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 598 { 599 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 600 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 601 602 /* 603 * lo_write_simple and lo_read_simple should have been covered 604 * by io submit style function like lo_rw_aio(), one blocker 605 * is that lo_read_simple() need to call flush_dcache_page after 606 * the page is written from kernel, and it isn't easy to handle 607 * this in io submit style function which submits all segments 608 * of the req at one time. And direct read IO doesn't need to 609 * run flush_dcache_page(). 610 */ 611 switch (req_op(rq)) { 612 case REQ_OP_FLUSH: 613 return lo_req_flush(lo, rq); 614 case REQ_OP_WRITE_ZEROES: 615 /* 616 * If the caller doesn't want deallocation, call zeroout to 617 * write zeroes the range. Otherwise, punch them out. 618 */ 619 return lo_fallocate(lo, rq, pos, 620 (rq->cmd_flags & REQ_NOUNMAP) ? 621 FALLOC_FL_ZERO_RANGE : 622 FALLOC_FL_PUNCH_HOLE); 623 case REQ_OP_DISCARD: 624 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 625 case REQ_OP_WRITE: 626 if (lo->transfer) 627 return lo_write_transfer(lo, rq, pos); 628 else if (cmd->use_aio) 629 return lo_rw_aio(lo, cmd, pos, WRITE); 630 else 631 return lo_write_simple(lo, rq, pos); 632 case REQ_OP_READ: 633 if (lo->transfer) 634 return lo_read_transfer(lo, rq, pos); 635 else if (cmd->use_aio) 636 return lo_rw_aio(lo, cmd, pos, READ); 637 else 638 return lo_read_simple(lo, rq, pos); 639 default: 640 WARN_ON_ONCE(1); 641 return -EIO; 642 } 643 } 644 645 static inline void loop_update_dio(struct loop_device *lo) 646 { 647 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 648 lo->use_dio); 649 } 650 651 static void loop_reread_partitions(struct loop_device *lo, 652 struct block_device *bdev) 653 { 654 int rc; 655 656 mutex_lock(&bdev->bd_mutex); 657 rc = bdev_disk_changed(bdev, false); 658 mutex_unlock(&bdev->bd_mutex); 659 if (rc) 660 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 661 __func__, lo->lo_number, lo->lo_file_name, rc); 662 } 663 664 static inline int is_loop_device(struct file *file) 665 { 666 struct inode *i = file->f_mapping->host; 667 668 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; 669 } 670 671 static int loop_validate_file(struct file *file, struct block_device *bdev) 672 { 673 struct inode *inode = file->f_mapping->host; 674 struct file *f = file; 675 676 /* Avoid recursion */ 677 while (is_loop_device(f)) { 678 struct loop_device *l; 679 680 if (f->f_mapping->host->i_rdev == bdev->bd_dev) 681 return -EBADF; 682 683 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 684 if (l->lo_state != Lo_bound) { 685 return -EINVAL; 686 } 687 f = l->lo_backing_file; 688 } 689 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 690 return -EINVAL; 691 return 0; 692 } 693 694 /* 695 * loop_change_fd switched the backing store of a loopback device to 696 * a new file. This is useful for operating system installers to free up 697 * the original file and in High Availability environments to switch to 698 * an alternative location for the content in case of server meltdown. 699 * This can only work if the loop device is used read-only, and if the 700 * new backing store is the same size and type as the old backing store. 701 */ 702 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 703 unsigned int arg) 704 { 705 struct file *file = NULL, *old_file; 706 int error; 707 bool partscan; 708 709 error = mutex_lock_killable(&lo->lo_mutex); 710 if (error) 711 return error; 712 error = -ENXIO; 713 if (lo->lo_state != Lo_bound) 714 goto out_err; 715 716 /* the loop device has to be read-only */ 717 error = -EINVAL; 718 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 719 goto out_err; 720 721 error = -EBADF; 722 file = fget(arg); 723 if (!file) 724 goto out_err; 725 726 error = loop_validate_file(file, bdev); 727 if (error) 728 goto out_err; 729 730 old_file = lo->lo_backing_file; 731 732 error = -EINVAL; 733 734 /* size of the new backing store needs to be the same */ 735 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 736 goto out_err; 737 738 /* and ... switch */ 739 blk_mq_freeze_queue(lo->lo_queue); 740 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 741 lo->lo_backing_file = file; 742 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 743 mapping_set_gfp_mask(file->f_mapping, 744 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 745 loop_update_dio(lo); 746 blk_mq_unfreeze_queue(lo->lo_queue); 747 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 748 mutex_unlock(&lo->lo_mutex); 749 /* 750 * We must drop file reference outside of lo_mutex as dropping 751 * the file ref can take bd_mutex which creates circular locking 752 * dependency. 753 */ 754 fput(old_file); 755 if (partscan) 756 loop_reread_partitions(lo, bdev); 757 return 0; 758 759 out_err: 760 mutex_unlock(&lo->lo_mutex); 761 if (file) 762 fput(file); 763 return error; 764 } 765 766 /* loop sysfs attributes */ 767 768 static ssize_t loop_attr_show(struct device *dev, char *page, 769 ssize_t (*callback)(struct loop_device *, char *)) 770 { 771 struct gendisk *disk = dev_to_disk(dev); 772 struct loop_device *lo = disk->private_data; 773 774 return callback(lo, page); 775 } 776 777 #define LOOP_ATTR_RO(_name) \ 778 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 779 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 780 struct device_attribute *attr, char *b) \ 781 { \ 782 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 783 } \ 784 static struct device_attribute loop_attr_##_name = \ 785 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 786 787 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 788 { 789 ssize_t ret; 790 char *p = NULL; 791 792 spin_lock_irq(&lo->lo_lock); 793 if (lo->lo_backing_file) 794 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 795 spin_unlock_irq(&lo->lo_lock); 796 797 if (IS_ERR_OR_NULL(p)) 798 ret = PTR_ERR(p); 799 else { 800 ret = strlen(p); 801 memmove(buf, p, ret); 802 buf[ret++] = '\n'; 803 buf[ret] = 0; 804 } 805 806 return ret; 807 } 808 809 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 810 { 811 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 812 } 813 814 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 815 { 816 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 817 } 818 819 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 820 { 821 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 822 823 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 824 } 825 826 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 827 { 828 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 829 830 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 831 } 832 833 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 834 { 835 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 836 837 return sprintf(buf, "%s\n", dio ? "1" : "0"); 838 } 839 840 LOOP_ATTR_RO(backing_file); 841 LOOP_ATTR_RO(offset); 842 LOOP_ATTR_RO(sizelimit); 843 LOOP_ATTR_RO(autoclear); 844 LOOP_ATTR_RO(partscan); 845 LOOP_ATTR_RO(dio); 846 847 static struct attribute *loop_attrs[] = { 848 &loop_attr_backing_file.attr, 849 &loop_attr_offset.attr, 850 &loop_attr_sizelimit.attr, 851 &loop_attr_autoclear.attr, 852 &loop_attr_partscan.attr, 853 &loop_attr_dio.attr, 854 NULL, 855 }; 856 857 static struct attribute_group loop_attribute_group = { 858 .name = "loop", 859 .attrs= loop_attrs, 860 }; 861 862 static void loop_sysfs_init(struct loop_device *lo) 863 { 864 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 865 &loop_attribute_group); 866 } 867 868 static void loop_sysfs_exit(struct loop_device *lo) 869 { 870 if (lo->sysfs_inited) 871 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 872 &loop_attribute_group); 873 } 874 875 static void loop_config_discard(struct loop_device *lo) 876 { 877 struct file *file = lo->lo_backing_file; 878 struct inode *inode = file->f_mapping->host; 879 struct request_queue *q = lo->lo_queue; 880 u32 granularity, max_discard_sectors; 881 882 /* 883 * If the backing device is a block device, mirror its zeroing 884 * capability. Set the discard sectors to the block device's zeroing 885 * capabilities because loop discards result in blkdev_issue_zeroout(), 886 * not blkdev_issue_discard(). This maintains consistent behavior with 887 * file-backed loop devices: discarded regions read back as zero. 888 */ 889 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { 890 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 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 struct loop_worker { 926 struct rb_node rb_node; 927 struct work_struct work; 928 struct list_head cmd_list; 929 struct list_head idle_list; 930 struct loop_device *lo; 931 struct cgroup_subsys_state *css; 932 unsigned long last_ran_at; 933 }; 934 935 static void loop_workfn(struct work_struct *work); 936 static void loop_rootcg_workfn(struct work_struct *work); 937 static void loop_free_idle_workers(struct timer_list *timer); 938 939 #ifdef CONFIG_BLK_CGROUP 940 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 941 { 942 return !css || css == blkcg_root_css; 943 } 944 #else 945 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 946 { 947 return !css; 948 } 949 #endif 950 951 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 952 { 953 struct rb_node **node = &(lo->worker_tree.rb_node), *parent = NULL; 954 struct loop_worker *cur_worker, *worker = NULL; 955 struct work_struct *work; 956 struct list_head *cmd_list; 957 958 spin_lock_irq(&lo->lo_work_lock); 959 960 if (queue_on_root_worker(cmd->css)) 961 goto queue_work; 962 963 node = &lo->worker_tree.rb_node; 964 965 while (*node) { 966 parent = *node; 967 cur_worker = container_of(*node, struct loop_worker, rb_node); 968 if (cur_worker->css == cmd->css) { 969 worker = cur_worker; 970 break; 971 } else if ((long)cur_worker->css < (long)cmd->css) { 972 node = &(*node)->rb_left; 973 } else { 974 node = &(*node)->rb_right; 975 } 976 } 977 if (worker) 978 goto queue_work; 979 980 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 981 /* 982 * In the event we cannot allocate a worker, just queue on the 983 * rootcg worker 984 */ 985 if (!worker) 986 goto queue_work; 987 988 worker->css = cmd->css; 989 css_get(worker->css); 990 INIT_WORK(&worker->work, loop_workfn); 991 INIT_LIST_HEAD(&worker->cmd_list); 992 INIT_LIST_HEAD(&worker->idle_list); 993 worker->lo = lo; 994 rb_link_node(&worker->rb_node, parent, node); 995 rb_insert_color(&worker->rb_node, &lo->worker_tree); 996 queue_work: 997 if (worker) { 998 /* 999 * We need to remove from the idle list here while 1000 * holding the lock so that the idle timer doesn't 1001 * free the worker 1002 */ 1003 if (!list_empty(&worker->idle_list)) 1004 list_del_init(&worker->idle_list); 1005 work = &worker->work; 1006 cmd_list = &worker->cmd_list; 1007 } else { 1008 work = &lo->rootcg_work; 1009 cmd_list = &lo->rootcg_cmd_list; 1010 } 1011 list_add_tail(&cmd->list_entry, cmd_list); 1012 queue_work(lo->workqueue, work); 1013 spin_unlock_irq(&lo->lo_work_lock); 1014 } 1015 1016 static void loop_update_rotational(struct loop_device *lo) 1017 { 1018 struct file *file = lo->lo_backing_file; 1019 struct inode *file_inode = file->f_mapping->host; 1020 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 1021 struct request_queue *q = lo->lo_queue; 1022 bool nonrot = true; 1023 1024 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 1025 if (file_bdev) 1026 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 1027 1028 if (nonrot) 1029 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 1030 else 1031 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 1032 } 1033 1034 static int 1035 loop_release_xfer(struct loop_device *lo) 1036 { 1037 int err = 0; 1038 struct loop_func_table *xfer = lo->lo_encryption; 1039 1040 if (xfer) { 1041 if (xfer->release) 1042 err = xfer->release(lo); 1043 lo->transfer = NULL; 1044 lo->lo_encryption = NULL; 1045 module_put(xfer->owner); 1046 } 1047 return err; 1048 } 1049 1050 static int 1051 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1052 const struct loop_info64 *i) 1053 { 1054 int err = 0; 1055 1056 if (xfer) { 1057 struct module *owner = xfer->owner; 1058 1059 if (!try_module_get(owner)) 1060 return -EINVAL; 1061 if (xfer->init) 1062 err = xfer->init(lo, i); 1063 if (err) 1064 module_put(owner); 1065 else 1066 lo->lo_encryption = xfer; 1067 } 1068 return err; 1069 } 1070 1071 /** 1072 * loop_set_status_from_info - configure device from loop_info 1073 * @lo: struct loop_device to configure 1074 * @info: struct loop_info64 to configure the device with 1075 * 1076 * Configures the loop device parameters according to the passed 1077 * in loop_info64 configuration. 1078 */ 1079 static int 1080 loop_set_status_from_info(struct loop_device *lo, 1081 const struct loop_info64 *info) 1082 { 1083 int err; 1084 struct loop_func_table *xfer; 1085 kuid_t uid = current_uid(); 1086 1087 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1088 return -EINVAL; 1089 1090 err = loop_release_xfer(lo); 1091 if (err) 1092 return err; 1093 1094 if (info->lo_encrypt_type) { 1095 unsigned int type = info->lo_encrypt_type; 1096 1097 if (type >= MAX_LO_CRYPT) 1098 return -EINVAL; 1099 xfer = xfer_funcs[type]; 1100 if (xfer == NULL) 1101 return -EINVAL; 1102 } else 1103 xfer = NULL; 1104 1105 err = loop_init_xfer(lo, xfer, info); 1106 if (err) 1107 return err; 1108 1109 lo->lo_offset = info->lo_offset; 1110 lo->lo_sizelimit = info->lo_sizelimit; 1111 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1112 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1113 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1114 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1115 1116 if (!xfer) 1117 xfer = &none_funcs; 1118 lo->transfer = xfer->transfer; 1119 lo->ioctl = xfer->ioctl; 1120 1121 lo->lo_flags = info->lo_flags; 1122 1123 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1124 lo->lo_init[0] = info->lo_init[0]; 1125 lo->lo_init[1] = info->lo_init[1]; 1126 if (info->lo_encrypt_key_size) { 1127 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1128 info->lo_encrypt_key_size); 1129 lo->lo_key_owner = uid; 1130 } 1131 1132 return 0; 1133 } 1134 1135 static int loop_configure(struct loop_device *lo, fmode_t mode, 1136 struct block_device *bdev, 1137 const struct loop_config *config) 1138 { 1139 struct file *file; 1140 struct inode *inode; 1141 struct address_space *mapping; 1142 int error; 1143 loff_t size; 1144 bool partscan; 1145 unsigned short bsize; 1146 1147 /* This is safe, since we have a reference from open(). */ 1148 __module_get(THIS_MODULE); 1149 1150 error = -EBADF; 1151 file = fget(config->fd); 1152 if (!file) 1153 goto out; 1154 1155 /* 1156 * If we don't hold exclusive handle for the device, upgrade to it 1157 * here to avoid changing device under exclusive owner. 1158 */ 1159 if (!(mode & FMODE_EXCL)) { 1160 error = bd_prepare_to_claim(bdev, loop_configure); 1161 if (error) 1162 goto out_putf; 1163 } 1164 1165 error = mutex_lock_killable(&lo->lo_mutex); 1166 if (error) 1167 goto out_bdev; 1168 1169 error = -EBUSY; 1170 if (lo->lo_state != Lo_unbound) 1171 goto out_unlock; 1172 1173 error = loop_validate_file(file, bdev); 1174 if (error) 1175 goto out_unlock; 1176 1177 mapping = file->f_mapping; 1178 inode = mapping->host; 1179 1180 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1181 error = -EINVAL; 1182 goto out_unlock; 1183 } 1184 1185 if (config->block_size) { 1186 error = loop_validate_block_size(config->block_size); 1187 if (error) 1188 goto out_unlock; 1189 } 1190 1191 error = loop_set_status_from_info(lo, &config->info); 1192 if (error) 1193 goto out_unlock; 1194 1195 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1196 !file->f_op->write_iter) 1197 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1198 1199 lo->workqueue = alloc_workqueue("loop%d", 1200 WQ_UNBOUND | WQ_FREEZABLE, 1201 0, 1202 lo->lo_number); 1203 if (!lo->workqueue) { 1204 error = -ENOMEM; 1205 goto out_unlock; 1206 } 1207 1208 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1209 1210 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 1211 INIT_LIST_HEAD(&lo->rootcg_cmd_list); 1212 INIT_LIST_HEAD(&lo->idle_worker_list); 1213 lo->worker_tree = RB_ROOT; 1214 timer_setup(&lo->timer, loop_free_idle_workers, 1215 TIMER_DEFERRABLE); 1216 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1217 lo->lo_device = bdev; 1218 lo->lo_backing_file = file; 1219 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1220 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1221 1222 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1223 blk_queue_write_cache(lo->lo_queue, true, false); 1224 1225 if (config->block_size) 1226 bsize = config->block_size; 1227 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1228 /* In case of direct I/O, match underlying block size */ 1229 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1230 else 1231 bsize = 512; 1232 1233 blk_queue_logical_block_size(lo->lo_queue, bsize); 1234 blk_queue_physical_block_size(lo->lo_queue, bsize); 1235 blk_queue_io_min(lo->lo_queue, bsize); 1236 1237 loop_update_rotational(lo); 1238 loop_update_dio(lo); 1239 loop_sysfs_init(lo); 1240 1241 size = get_loop_size(lo, file); 1242 loop_set_size(lo, size); 1243 1244 lo->lo_state = Lo_bound; 1245 if (part_shift) 1246 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1247 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1248 if (partscan) 1249 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1250 1251 /* Grab the block_device to prevent its destruction after we 1252 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 1253 */ 1254 bdgrab(bdev); 1255 mutex_unlock(&lo->lo_mutex); 1256 if (partscan) 1257 loop_reread_partitions(lo, bdev); 1258 if (!(mode & FMODE_EXCL)) 1259 bd_abort_claiming(bdev, loop_configure); 1260 return 0; 1261 1262 out_unlock: 1263 mutex_unlock(&lo->lo_mutex); 1264 out_bdev: 1265 if (!(mode & FMODE_EXCL)) 1266 bd_abort_claiming(bdev, loop_configure); 1267 out_putf: 1268 fput(file); 1269 out: 1270 /* This is safe: open() is still holding a reference. */ 1271 module_put(THIS_MODULE); 1272 return error; 1273 } 1274 1275 static int __loop_clr_fd(struct loop_device *lo, bool release) 1276 { 1277 struct file *filp = NULL; 1278 gfp_t gfp = lo->old_gfp_mask; 1279 struct block_device *bdev = lo->lo_device; 1280 int err = 0; 1281 bool partscan = false; 1282 int lo_number; 1283 struct loop_worker *pos, *worker; 1284 1285 mutex_lock(&lo->lo_mutex); 1286 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1287 err = -ENXIO; 1288 goto out_unlock; 1289 } 1290 1291 filp = lo->lo_backing_file; 1292 if (filp == NULL) { 1293 err = -EINVAL; 1294 goto out_unlock; 1295 } 1296 1297 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1298 blk_queue_write_cache(lo->lo_queue, false, false); 1299 1300 /* freeze request queue during the transition */ 1301 blk_mq_freeze_queue(lo->lo_queue); 1302 1303 destroy_workqueue(lo->workqueue); 1304 spin_lock_irq(&lo->lo_work_lock); 1305 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 1306 idle_list) { 1307 list_del(&worker->idle_list); 1308 rb_erase(&worker->rb_node, &lo->worker_tree); 1309 css_put(worker->css); 1310 kfree(worker); 1311 } 1312 spin_unlock_irq(&lo->lo_work_lock); 1313 del_timer_sync(&lo->timer); 1314 1315 spin_lock_irq(&lo->lo_lock); 1316 lo->lo_backing_file = NULL; 1317 spin_unlock_irq(&lo->lo_lock); 1318 1319 loop_release_xfer(lo); 1320 lo->transfer = NULL; 1321 lo->ioctl = NULL; 1322 lo->lo_device = NULL; 1323 lo->lo_encryption = NULL; 1324 lo->lo_offset = 0; 1325 lo->lo_sizelimit = 0; 1326 lo->lo_encrypt_key_size = 0; 1327 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1328 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1329 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1330 blk_queue_logical_block_size(lo->lo_queue, 512); 1331 blk_queue_physical_block_size(lo->lo_queue, 512); 1332 blk_queue_io_min(lo->lo_queue, 512); 1333 if (bdev) { 1334 bdput(bdev); 1335 invalidate_bdev(bdev); 1336 bdev->bd_inode->i_mapping->wb_err = 0; 1337 } 1338 set_capacity(lo->lo_disk, 0); 1339 loop_sysfs_exit(lo); 1340 if (bdev) { 1341 /* let user-space know about this change */ 1342 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1343 } 1344 mapping_set_gfp_mask(filp->f_mapping, gfp); 1345 /* This is safe: open() is still holding a reference. */ 1346 module_put(THIS_MODULE); 1347 blk_mq_unfreeze_queue(lo->lo_queue); 1348 1349 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1350 lo_number = lo->lo_number; 1351 out_unlock: 1352 mutex_unlock(&lo->lo_mutex); 1353 if (partscan) { 1354 /* 1355 * bd_mutex has been held already in release path, so don't 1356 * acquire it if this function is called in such case. 1357 * 1358 * If the reread partition isn't from release path, lo_refcnt 1359 * must be at least one and it can only become zero when the 1360 * current holder is released. 1361 */ 1362 if (!release) 1363 mutex_lock(&bdev->bd_mutex); 1364 err = bdev_disk_changed(bdev, false); 1365 if (!release) 1366 mutex_unlock(&bdev->bd_mutex); 1367 if (err) 1368 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1369 __func__, lo_number, err); 1370 /* Device is gone, no point in returning error */ 1371 err = 0; 1372 } 1373 1374 /* 1375 * lo->lo_state is set to Lo_unbound here after above partscan has 1376 * finished. 1377 * 1378 * There cannot be anybody else entering __loop_clr_fd() as 1379 * lo->lo_backing_file is already cleared and Lo_rundown state 1380 * protects us from all the other places trying to change the 'lo' 1381 * device. 1382 */ 1383 mutex_lock(&lo->lo_mutex); 1384 lo->lo_flags = 0; 1385 if (!part_shift) 1386 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1387 lo->lo_state = Lo_unbound; 1388 mutex_unlock(&lo->lo_mutex); 1389 1390 /* 1391 * Need not hold lo_mutex to fput backing file. Calling fput holding 1392 * lo_mutex triggers a circular lock dependency possibility warning as 1393 * fput can take bd_mutex which is usually taken before lo_mutex. 1394 */ 1395 if (filp) 1396 fput(filp); 1397 return err; 1398 } 1399 1400 static int loop_clr_fd(struct loop_device *lo) 1401 { 1402 int err; 1403 1404 err = mutex_lock_killable(&lo->lo_mutex); 1405 if (err) 1406 return err; 1407 if (lo->lo_state != Lo_bound) { 1408 mutex_unlock(&lo->lo_mutex); 1409 return -ENXIO; 1410 } 1411 /* 1412 * If we've explicitly asked to tear down the loop device, 1413 * and it has an elevated reference count, set it for auto-teardown when 1414 * the last reference goes away. This stops $!~#$@ udev from 1415 * preventing teardown because it decided that it needs to run blkid on 1416 * the loopback device whenever they appear. xfstests is notorious for 1417 * failing tests because blkid via udev races with a losetup 1418 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1419 * command to fail with EBUSY. 1420 */ 1421 if (atomic_read(&lo->lo_refcnt) > 1) { 1422 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1423 mutex_unlock(&lo->lo_mutex); 1424 return 0; 1425 } 1426 lo->lo_state = Lo_rundown; 1427 mutex_unlock(&lo->lo_mutex); 1428 1429 return __loop_clr_fd(lo, false); 1430 } 1431 1432 static int 1433 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1434 { 1435 int err; 1436 struct block_device *bdev; 1437 kuid_t uid = current_uid(); 1438 int prev_lo_flags; 1439 bool partscan = false; 1440 bool size_changed = false; 1441 1442 err = mutex_lock_killable(&lo->lo_mutex); 1443 if (err) 1444 return err; 1445 if (lo->lo_encrypt_key_size && 1446 !uid_eq(lo->lo_key_owner, uid) && 1447 !capable(CAP_SYS_ADMIN)) { 1448 err = -EPERM; 1449 goto out_unlock; 1450 } 1451 if (lo->lo_state != Lo_bound) { 1452 err = -ENXIO; 1453 goto out_unlock; 1454 } 1455 1456 if (lo->lo_offset != info->lo_offset || 1457 lo->lo_sizelimit != info->lo_sizelimit) { 1458 size_changed = true; 1459 sync_blockdev(lo->lo_device); 1460 invalidate_bdev(lo->lo_device); 1461 } 1462 1463 /* I/O need to be drained during transfer transition */ 1464 blk_mq_freeze_queue(lo->lo_queue); 1465 1466 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1467 /* If any pages were dirtied after invalidate_bdev(), try again */ 1468 err = -EAGAIN; 1469 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1470 __func__, lo->lo_number, lo->lo_file_name, 1471 lo->lo_device->bd_inode->i_mapping->nrpages); 1472 goto out_unfreeze; 1473 } 1474 1475 prev_lo_flags = lo->lo_flags; 1476 1477 err = loop_set_status_from_info(lo, info); 1478 if (err) 1479 goto out_unfreeze; 1480 1481 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1482 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1483 /* For those flags, use the previous values instead */ 1484 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1485 /* For flags that can't be cleared, use previous values too */ 1486 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1487 1488 if (size_changed) { 1489 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1490 lo->lo_backing_file); 1491 loop_set_size(lo, new_size); 1492 } 1493 1494 loop_config_discard(lo); 1495 1496 /* update dio if lo_offset or transfer is changed */ 1497 __loop_update_dio(lo, lo->use_dio); 1498 1499 out_unfreeze: 1500 blk_mq_unfreeze_queue(lo->lo_queue); 1501 1502 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1503 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1504 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1505 bdev = lo->lo_device; 1506 partscan = true; 1507 } 1508 out_unlock: 1509 mutex_unlock(&lo->lo_mutex); 1510 if (partscan) 1511 loop_reread_partitions(lo, bdev); 1512 1513 return err; 1514 } 1515 1516 static int 1517 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1518 { 1519 struct path path; 1520 struct kstat stat; 1521 int ret; 1522 1523 ret = mutex_lock_killable(&lo->lo_mutex); 1524 if (ret) 1525 return ret; 1526 if (lo->lo_state != Lo_bound) { 1527 mutex_unlock(&lo->lo_mutex); 1528 return -ENXIO; 1529 } 1530 1531 memset(info, 0, sizeof(*info)); 1532 info->lo_number = lo->lo_number; 1533 info->lo_offset = lo->lo_offset; 1534 info->lo_sizelimit = lo->lo_sizelimit; 1535 info->lo_flags = lo->lo_flags; 1536 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1537 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1538 info->lo_encrypt_type = 1539 lo->lo_encryption ? lo->lo_encryption->number : 0; 1540 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1541 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1542 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1543 lo->lo_encrypt_key_size); 1544 } 1545 1546 /* Drop lo_mutex while we call into the filesystem. */ 1547 path = lo->lo_backing_file->f_path; 1548 path_get(&path); 1549 mutex_unlock(&lo->lo_mutex); 1550 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1551 if (!ret) { 1552 info->lo_device = huge_encode_dev(stat.dev); 1553 info->lo_inode = stat.ino; 1554 info->lo_rdevice = huge_encode_dev(stat.rdev); 1555 } 1556 path_put(&path); 1557 return ret; 1558 } 1559 1560 static void 1561 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1562 { 1563 memset(info64, 0, sizeof(*info64)); 1564 info64->lo_number = info->lo_number; 1565 info64->lo_device = info->lo_device; 1566 info64->lo_inode = info->lo_inode; 1567 info64->lo_rdevice = info->lo_rdevice; 1568 info64->lo_offset = info->lo_offset; 1569 info64->lo_sizelimit = 0; 1570 info64->lo_encrypt_type = info->lo_encrypt_type; 1571 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1572 info64->lo_flags = info->lo_flags; 1573 info64->lo_init[0] = info->lo_init[0]; 1574 info64->lo_init[1] = info->lo_init[1]; 1575 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1576 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1577 else 1578 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1579 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1580 } 1581 1582 static int 1583 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1584 { 1585 memset(info, 0, sizeof(*info)); 1586 info->lo_number = info64->lo_number; 1587 info->lo_device = info64->lo_device; 1588 info->lo_inode = info64->lo_inode; 1589 info->lo_rdevice = info64->lo_rdevice; 1590 info->lo_offset = info64->lo_offset; 1591 info->lo_encrypt_type = info64->lo_encrypt_type; 1592 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1593 info->lo_flags = info64->lo_flags; 1594 info->lo_init[0] = info64->lo_init[0]; 1595 info->lo_init[1] = info64->lo_init[1]; 1596 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1597 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1598 else 1599 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1600 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1601 1602 /* error in case values were truncated */ 1603 if (info->lo_device != info64->lo_device || 1604 info->lo_rdevice != info64->lo_rdevice || 1605 info->lo_inode != info64->lo_inode || 1606 info->lo_offset != info64->lo_offset) 1607 return -EOVERFLOW; 1608 1609 return 0; 1610 } 1611 1612 static int 1613 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1614 { 1615 struct loop_info info; 1616 struct loop_info64 info64; 1617 1618 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1619 return -EFAULT; 1620 loop_info64_from_old(&info, &info64); 1621 return loop_set_status(lo, &info64); 1622 } 1623 1624 static int 1625 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1626 { 1627 struct loop_info64 info64; 1628 1629 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1630 return -EFAULT; 1631 return loop_set_status(lo, &info64); 1632 } 1633 1634 static int 1635 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1636 struct loop_info info; 1637 struct loop_info64 info64; 1638 int err; 1639 1640 if (!arg) 1641 return -EINVAL; 1642 err = loop_get_status(lo, &info64); 1643 if (!err) 1644 err = loop_info64_to_old(&info64, &info); 1645 if (!err && copy_to_user(arg, &info, sizeof(info))) 1646 err = -EFAULT; 1647 1648 return err; 1649 } 1650 1651 static int 1652 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1653 struct loop_info64 info64; 1654 int err; 1655 1656 if (!arg) 1657 return -EINVAL; 1658 err = loop_get_status(lo, &info64); 1659 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1660 err = -EFAULT; 1661 1662 return err; 1663 } 1664 1665 static int loop_set_capacity(struct loop_device *lo) 1666 { 1667 loff_t size; 1668 1669 if (unlikely(lo->lo_state != Lo_bound)) 1670 return -ENXIO; 1671 1672 size = get_loop_size(lo, lo->lo_backing_file); 1673 loop_set_size(lo, size); 1674 1675 return 0; 1676 } 1677 1678 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1679 { 1680 int error = -ENXIO; 1681 if (lo->lo_state != Lo_bound) 1682 goto out; 1683 1684 __loop_update_dio(lo, !!arg); 1685 if (lo->use_dio == !!arg) 1686 return 0; 1687 error = -EINVAL; 1688 out: 1689 return error; 1690 } 1691 1692 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1693 { 1694 int err = 0; 1695 1696 if (lo->lo_state != Lo_bound) 1697 return -ENXIO; 1698 1699 err = loop_validate_block_size(arg); 1700 if (err) 1701 return err; 1702 1703 if (lo->lo_queue->limits.logical_block_size == arg) 1704 return 0; 1705 1706 sync_blockdev(lo->lo_device); 1707 invalidate_bdev(lo->lo_device); 1708 1709 blk_mq_freeze_queue(lo->lo_queue); 1710 1711 /* invalidate_bdev should have truncated all the pages */ 1712 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1713 err = -EAGAIN; 1714 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1715 __func__, lo->lo_number, lo->lo_file_name, 1716 lo->lo_device->bd_inode->i_mapping->nrpages); 1717 goto out_unfreeze; 1718 } 1719 1720 blk_queue_logical_block_size(lo->lo_queue, arg); 1721 blk_queue_physical_block_size(lo->lo_queue, arg); 1722 blk_queue_io_min(lo->lo_queue, arg); 1723 loop_update_dio(lo); 1724 out_unfreeze: 1725 blk_mq_unfreeze_queue(lo->lo_queue); 1726 1727 return err; 1728 } 1729 1730 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1731 unsigned long arg) 1732 { 1733 int err; 1734 1735 err = mutex_lock_killable(&lo->lo_mutex); 1736 if (err) 1737 return err; 1738 switch (cmd) { 1739 case LOOP_SET_CAPACITY: 1740 err = loop_set_capacity(lo); 1741 break; 1742 case LOOP_SET_DIRECT_IO: 1743 err = loop_set_dio(lo, arg); 1744 break; 1745 case LOOP_SET_BLOCK_SIZE: 1746 err = loop_set_block_size(lo, arg); 1747 break; 1748 default: 1749 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1750 } 1751 mutex_unlock(&lo->lo_mutex); 1752 return err; 1753 } 1754 1755 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1756 unsigned int cmd, unsigned long arg) 1757 { 1758 struct loop_device *lo = bdev->bd_disk->private_data; 1759 void __user *argp = (void __user *) arg; 1760 int err; 1761 1762 switch (cmd) { 1763 case LOOP_SET_FD: { 1764 /* 1765 * Legacy case - pass in a zeroed out struct loop_config with 1766 * only the file descriptor set , which corresponds with the 1767 * default parameters we'd have used otherwise. 1768 */ 1769 struct loop_config config; 1770 1771 memset(&config, 0, sizeof(config)); 1772 config.fd = arg; 1773 1774 return loop_configure(lo, mode, bdev, &config); 1775 } 1776 case LOOP_CONFIGURE: { 1777 struct loop_config config; 1778 1779 if (copy_from_user(&config, argp, sizeof(config))) 1780 return -EFAULT; 1781 1782 return loop_configure(lo, mode, bdev, &config); 1783 } 1784 case LOOP_CHANGE_FD: 1785 return loop_change_fd(lo, bdev, arg); 1786 case LOOP_CLR_FD: 1787 return loop_clr_fd(lo); 1788 case LOOP_SET_STATUS: 1789 err = -EPERM; 1790 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1791 err = loop_set_status_old(lo, argp); 1792 } 1793 break; 1794 case LOOP_GET_STATUS: 1795 return loop_get_status_old(lo, argp); 1796 case LOOP_SET_STATUS64: 1797 err = -EPERM; 1798 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1799 err = loop_set_status64(lo, argp); 1800 } 1801 break; 1802 case LOOP_GET_STATUS64: 1803 return loop_get_status64(lo, argp); 1804 case LOOP_SET_CAPACITY: 1805 case LOOP_SET_DIRECT_IO: 1806 case LOOP_SET_BLOCK_SIZE: 1807 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1808 return -EPERM; 1809 fallthrough; 1810 default: 1811 err = lo_simple_ioctl(lo, cmd, arg); 1812 break; 1813 } 1814 1815 return err; 1816 } 1817 1818 #ifdef CONFIG_COMPAT 1819 struct compat_loop_info { 1820 compat_int_t lo_number; /* ioctl r/o */ 1821 compat_dev_t lo_device; /* ioctl r/o */ 1822 compat_ulong_t lo_inode; /* ioctl r/o */ 1823 compat_dev_t lo_rdevice; /* ioctl r/o */ 1824 compat_int_t lo_offset; 1825 compat_int_t lo_encrypt_type; 1826 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1827 compat_int_t lo_flags; /* ioctl r/o */ 1828 char lo_name[LO_NAME_SIZE]; 1829 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1830 compat_ulong_t lo_init[2]; 1831 char reserved[4]; 1832 }; 1833 1834 /* 1835 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1836 * - noinlined to reduce stack space usage in main part of driver 1837 */ 1838 static noinline int 1839 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1840 struct loop_info64 *info64) 1841 { 1842 struct compat_loop_info info; 1843 1844 if (copy_from_user(&info, arg, sizeof(info))) 1845 return -EFAULT; 1846 1847 memset(info64, 0, sizeof(*info64)); 1848 info64->lo_number = info.lo_number; 1849 info64->lo_device = info.lo_device; 1850 info64->lo_inode = info.lo_inode; 1851 info64->lo_rdevice = info.lo_rdevice; 1852 info64->lo_offset = info.lo_offset; 1853 info64->lo_sizelimit = 0; 1854 info64->lo_encrypt_type = info.lo_encrypt_type; 1855 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1856 info64->lo_flags = info.lo_flags; 1857 info64->lo_init[0] = info.lo_init[0]; 1858 info64->lo_init[1] = info.lo_init[1]; 1859 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1860 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1861 else 1862 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1863 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1864 return 0; 1865 } 1866 1867 /* 1868 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1869 * - noinlined to reduce stack space usage in main part of driver 1870 */ 1871 static noinline int 1872 loop_info64_to_compat(const struct loop_info64 *info64, 1873 struct compat_loop_info __user *arg) 1874 { 1875 struct compat_loop_info info; 1876 1877 memset(&info, 0, sizeof(info)); 1878 info.lo_number = info64->lo_number; 1879 info.lo_device = info64->lo_device; 1880 info.lo_inode = info64->lo_inode; 1881 info.lo_rdevice = info64->lo_rdevice; 1882 info.lo_offset = info64->lo_offset; 1883 info.lo_encrypt_type = info64->lo_encrypt_type; 1884 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1885 info.lo_flags = info64->lo_flags; 1886 info.lo_init[0] = info64->lo_init[0]; 1887 info.lo_init[1] = info64->lo_init[1]; 1888 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1889 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1890 else 1891 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1892 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1893 1894 /* error in case values were truncated */ 1895 if (info.lo_device != info64->lo_device || 1896 info.lo_rdevice != info64->lo_rdevice || 1897 info.lo_inode != info64->lo_inode || 1898 info.lo_offset != info64->lo_offset || 1899 info.lo_init[0] != info64->lo_init[0] || 1900 info.lo_init[1] != info64->lo_init[1]) 1901 return -EOVERFLOW; 1902 1903 if (copy_to_user(arg, &info, sizeof(info))) 1904 return -EFAULT; 1905 return 0; 1906 } 1907 1908 static int 1909 loop_set_status_compat(struct loop_device *lo, 1910 const struct compat_loop_info __user *arg) 1911 { 1912 struct loop_info64 info64; 1913 int ret; 1914 1915 ret = loop_info64_from_compat(arg, &info64); 1916 if (ret < 0) 1917 return ret; 1918 return loop_set_status(lo, &info64); 1919 } 1920 1921 static int 1922 loop_get_status_compat(struct loop_device *lo, 1923 struct compat_loop_info __user *arg) 1924 { 1925 struct loop_info64 info64; 1926 int err; 1927 1928 if (!arg) 1929 return -EINVAL; 1930 err = loop_get_status(lo, &info64); 1931 if (!err) 1932 err = loop_info64_to_compat(&info64, arg); 1933 return err; 1934 } 1935 1936 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1937 unsigned int cmd, unsigned long arg) 1938 { 1939 struct loop_device *lo = bdev->bd_disk->private_data; 1940 int err; 1941 1942 switch(cmd) { 1943 case LOOP_SET_STATUS: 1944 err = loop_set_status_compat(lo, 1945 (const struct compat_loop_info __user *)arg); 1946 break; 1947 case LOOP_GET_STATUS: 1948 err = loop_get_status_compat(lo, 1949 (struct compat_loop_info __user *)arg); 1950 break; 1951 case LOOP_SET_CAPACITY: 1952 case LOOP_CLR_FD: 1953 case LOOP_GET_STATUS64: 1954 case LOOP_SET_STATUS64: 1955 case LOOP_CONFIGURE: 1956 arg = (unsigned long) compat_ptr(arg); 1957 fallthrough; 1958 case LOOP_SET_FD: 1959 case LOOP_CHANGE_FD: 1960 case LOOP_SET_BLOCK_SIZE: 1961 case LOOP_SET_DIRECT_IO: 1962 err = lo_ioctl(bdev, mode, cmd, arg); 1963 break; 1964 default: 1965 err = -ENOIOCTLCMD; 1966 break; 1967 } 1968 return err; 1969 } 1970 #endif 1971 1972 static int lo_open(struct block_device *bdev, fmode_t mode) 1973 { 1974 struct loop_device *lo = bdev->bd_disk->private_data; 1975 int err; 1976 1977 err = mutex_lock_killable(&lo->lo_mutex); 1978 if (err) 1979 return err; 1980 if (lo->lo_state == Lo_deleting) 1981 err = -ENXIO; 1982 else 1983 atomic_inc(&lo->lo_refcnt); 1984 mutex_unlock(&lo->lo_mutex); 1985 return err; 1986 } 1987 1988 static void lo_release(struct gendisk *disk, fmode_t mode) 1989 { 1990 struct loop_device *lo = disk->private_data; 1991 1992 mutex_lock(&lo->lo_mutex); 1993 if (atomic_dec_return(&lo->lo_refcnt)) 1994 goto out_unlock; 1995 1996 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1997 if (lo->lo_state != Lo_bound) 1998 goto out_unlock; 1999 lo->lo_state = Lo_rundown; 2000 mutex_unlock(&lo->lo_mutex); 2001 /* 2002 * In autoclear mode, stop the loop thread 2003 * and remove configuration after last close. 2004 */ 2005 __loop_clr_fd(lo, true); 2006 return; 2007 } else if (lo->lo_state == Lo_bound) { 2008 /* 2009 * Otherwise keep thread (if running) and config, 2010 * but flush possible ongoing bios in thread. 2011 */ 2012 blk_mq_freeze_queue(lo->lo_queue); 2013 blk_mq_unfreeze_queue(lo->lo_queue); 2014 } 2015 2016 out_unlock: 2017 mutex_unlock(&lo->lo_mutex); 2018 } 2019 2020 static const struct block_device_operations lo_fops = { 2021 .owner = THIS_MODULE, 2022 .open = lo_open, 2023 .release = lo_release, 2024 .ioctl = lo_ioctl, 2025 #ifdef CONFIG_COMPAT 2026 .compat_ioctl = lo_compat_ioctl, 2027 #endif 2028 }; 2029 2030 /* 2031 * And now the modules code and kernel interface. 2032 */ 2033 static int max_loop; 2034 module_param(max_loop, int, 0444); 2035 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 2036 module_param(max_part, int, 0444); 2037 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 2038 MODULE_LICENSE("GPL"); 2039 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 2040 2041 int loop_register_transfer(struct loop_func_table *funcs) 2042 { 2043 unsigned int n = funcs->number; 2044 2045 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 2046 return -EINVAL; 2047 xfer_funcs[n] = funcs; 2048 return 0; 2049 } 2050 2051 static int unregister_transfer_cb(int id, void *ptr, void *data) 2052 { 2053 struct loop_device *lo = ptr; 2054 struct loop_func_table *xfer = data; 2055 2056 mutex_lock(&lo->lo_mutex); 2057 if (lo->lo_encryption == xfer) 2058 loop_release_xfer(lo); 2059 mutex_unlock(&lo->lo_mutex); 2060 return 0; 2061 } 2062 2063 int loop_unregister_transfer(int number) 2064 { 2065 unsigned int n = number; 2066 struct loop_func_table *xfer; 2067 2068 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 2069 return -EINVAL; 2070 2071 xfer_funcs[n] = NULL; 2072 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 2073 return 0; 2074 } 2075 2076 EXPORT_SYMBOL(loop_register_transfer); 2077 EXPORT_SYMBOL(loop_unregister_transfer); 2078 2079 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 2080 const struct blk_mq_queue_data *bd) 2081 { 2082 struct request *rq = bd->rq; 2083 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2084 struct loop_device *lo = rq->q->queuedata; 2085 2086 blk_mq_start_request(rq); 2087 2088 if (lo->lo_state != Lo_bound) 2089 return BLK_STS_IOERR; 2090 2091 switch (req_op(rq)) { 2092 case REQ_OP_FLUSH: 2093 case REQ_OP_DISCARD: 2094 case REQ_OP_WRITE_ZEROES: 2095 cmd->use_aio = false; 2096 break; 2097 default: 2098 cmd->use_aio = lo->use_dio; 2099 break; 2100 } 2101 2102 /* always use the first bio's css */ 2103 #ifdef CONFIG_BLK_CGROUP 2104 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 2105 cmd->css = &bio_blkcg(rq->bio)->css; 2106 css_get(cmd->css); 2107 } else 2108 #endif 2109 cmd->css = NULL; 2110 loop_queue_work(lo, cmd); 2111 2112 return BLK_STS_OK; 2113 } 2114 2115 static void loop_handle_cmd(struct loop_cmd *cmd) 2116 { 2117 struct request *rq = blk_mq_rq_from_pdu(cmd); 2118 const bool write = op_is_write(req_op(rq)); 2119 struct loop_device *lo = rq->q->queuedata; 2120 int ret = 0; 2121 2122 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 2123 ret = -EIO; 2124 goto failed; 2125 } 2126 2127 ret = do_req_filebacked(lo, rq); 2128 failed: 2129 /* complete non-aio request */ 2130 if (!cmd->use_aio || ret) { 2131 if (ret == -EOPNOTSUPP) 2132 cmd->ret = ret; 2133 else 2134 cmd->ret = ret ? -EIO : 0; 2135 if (likely(!blk_should_fake_timeout(rq->q))) 2136 blk_mq_complete_request(rq); 2137 } 2138 } 2139 2140 static void loop_set_timer(struct loop_device *lo) 2141 { 2142 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 2143 } 2144 2145 static void loop_process_work(struct loop_worker *worker, 2146 struct list_head *cmd_list, struct loop_device *lo) 2147 { 2148 int orig_flags = current->flags; 2149 struct loop_cmd *cmd; 2150 2151 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 2152 spin_lock_irq(&lo->lo_work_lock); 2153 while (!list_empty(cmd_list)) { 2154 cmd = container_of( 2155 cmd_list->next, struct loop_cmd, list_entry); 2156 list_del(cmd_list->next); 2157 spin_unlock_irq(&lo->lo_work_lock); 2158 2159 loop_handle_cmd(cmd); 2160 cond_resched(); 2161 2162 spin_lock_irq(&lo->lo_work_lock); 2163 } 2164 2165 /* 2166 * We only add to the idle list if there are no pending cmds 2167 * *and* the worker will not run again which ensures that it 2168 * is safe to free any worker on the idle list 2169 */ 2170 if (worker && !work_pending(&worker->work)) { 2171 worker->last_ran_at = jiffies; 2172 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 2173 loop_set_timer(lo); 2174 } 2175 spin_unlock_irq(&lo->lo_work_lock); 2176 current->flags = orig_flags; 2177 } 2178 2179 static void loop_workfn(struct work_struct *work) 2180 { 2181 struct loop_worker *worker = 2182 container_of(work, struct loop_worker, work); 2183 loop_process_work(worker, &worker->cmd_list, worker->lo); 2184 } 2185 2186 static void loop_rootcg_workfn(struct work_struct *work) 2187 { 2188 struct loop_device *lo = 2189 container_of(work, struct loop_device, rootcg_work); 2190 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 2191 } 2192 2193 static void loop_free_idle_workers(struct timer_list *timer) 2194 { 2195 struct loop_device *lo = container_of(timer, struct loop_device, timer); 2196 struct loop_worker *pos, *worker; 2197 2198 spin_lock_irq(&lo->lo_work_lock); 2199 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 2200 idle_list) { 2201 if (time_is_after_jiffies(worker->last_ran_at + 2202 LOOP_IDLE_WORKER_TIMEOUT)) 2203 break; 2204 list_del(&worker->idle_list); 2205 rb_erase(&worker->rb_node, &lo->worker_tree); 2206 css_put(worker->css); 2207 kfree(worker); 2208 } 2209 if (!list_empty(&lo->idle_worker_list)) 2210 loop_set_timer(lo); 2211 spin_unlock_irq(&lo->lo_work_lock); 2212 } 2213 2214 static const struct blk_mq_ops loop_mq_ops = { 2215 .queue_rq = loop_queue_rq, 2216 .complete = lo_complete_rq, 2217 }; 2218 2219 static int loop_add(struct loop_device **l, int i) 2220 { 2221 struct loop_device *lo; 2222 struct gendisk *disk; 2223 int err; 2224 2225 err = -ENOMEM; 2226 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 2227 if (!lo) 2228 goto out; 2229 2230 lo->lo_state = Lo_unbound; 2231 2232 /* allocate id, if @id >= 0, we're requesting that specific id */ 2233 if (i >= 0) { 2234 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2235 if (err == -ENOSPC) 2236 err = -EEXIST; 2237 } else { 2238 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2239 } 2240 if (err < 0) 2241 goto out_free_dev; 2242 i = err; 2243 2244 err = -ENOMEM; 2245 lo->tag_set.ops = &loop_mq_ops; 2246 lo->tag_set.nr_hw_queues = 1; 2247 lo->tag_set.queue_depth = 128; 2248 lo->tag_set.numa_node = NUMA_NO_NODE; 2249 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2250 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING; 2251 lo->tag_set.driver_data = lo; 2252 2253 err = blk_mq_alloc_tag_set(&lo->tag_set); 2254 if (err) 2255 goto out_free_idr; 2256 2257 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 2258 if (IS_ERR(lo->lo_queue)) { 2259 err = PTR_ERR(lo->lo_queue); 2260 goto out_cleanup_tags; 2261 } 2262 lo->lo_queue->queuedata = lo; 2263 2264 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2265 2266 /* 2267 * By default, we do buffer IO, so it doesn't make sense to enable 2268 * merge because the I/O submitted to backing file is handled page by 2269 * page. For directio mode, merge does help to dispatch bigger request 2270 * to underlayer disk. We will enable merge once directio is enabled. 2271 */ 2272 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2273 2274 err = -ENOMEM; 2275 disk = lo->lo_disk = alloc_disk(1 << part_shift); 2276 if (!disk) 2277 goto out_free_queue; 2278 2279 /* 2280 * Disable partition scanning by default. The in-kernel partition 2281 * scanning can be requested individually per-device during its 2282 * setup. Userspace can always add and remove partitions from all 2283 * devices. The needed partition minors are allocated from the 2284 * extended minor space, the main loop device numbers will continue 2285 * to match the loop minors, regardless of the number of partitions 2286 * used. 2287 * 2288 * If max_part is given, partition scanning is globally enabled for 2289 * all loop devices. The minors for the main loop devices will be 2290 * multiples of max_part. 2291 * 2292 * Note: Global-for-all-devices, set-only-at-init, read-only module 2293 * parameteters like 'max_loop' and 'max_part' make things needlessly 2294 * complicated, are too static, inflexible and may surprise 2295 * userspace tools. Parameters like this in general should be avoided. 2296 */ 2297 if (!part_shift) 2298 disk->flags |= GENHD_FL_NO_PART_SCAN; 2299 disk->flags |= GENHD_FL_EXT_DEVT; 2300 atomic_set(&lo->lo_refcnt, 0); 2301 mutex_init(&lo->lo_mutex); 2302 lo->lo_number = i; 2303 spin_lock_init(&lo->lo_lock); 2304 spin_lock_init(&lo->lo_work_lock); 2305 disk->major = LOOP_MAJOR; 2306 disk->first_minor = i << part_shift; 2307 disk->fops = &lo_fops; 2308 disk->private_data = lo; 2309 disk->queue = lo->lo_queue; 2310 sprintf(disk->disk_name, "loop%d", i); 2311 add_disk(disk); 2312 *l = lo; 2313 return lo->lo_number; 2314 2315 out_free_queue: 2316 blk_cleanup_queue(lo->lo_queue); 2317 out_cleanup_tags: 2318 blk_mq_free_tag_set(&lo->tag_set); 2319 out_free_idr: 2320 idr_remove(&loop_index_idr, i); 2321 out_free_dev: 2322 kfree(lo); 2323 out: 2324 return err; 2325 } 2326 2327 static void loop_remove(struct loop_device *lo) 2328 { 2329 del_gendisk(lo->lo_disk); 2330 blk_cleanup_queue(lo->lo_queue); 2331 blk_mq_free_tag_set(&lo->tag_set); 2332 put_disk(lo->lo_disk); 2333 mutex_destroy(&lo->lo_mutex); 2334 kfree(lo); 2335 } 2336 2337 static int find_free_cb(int id, void *ptr, void *data) 2338 { 2339 struct loop_device *lo = ptr; 2340 struct loop_device **l = data; 2341 2342 if (lo->lo_state == Lo_unbound) { 2343 *l = lo; 2344 return 1; 2345 } 2346 return 0; 2347 } 2348 2349 static int loop_lookup(struct loop_device **l, int i) 2350 { 2351 struct loop_device *lo; 2352 int ret = -ENODEV; 2353 2354 if (i < 0) { 2355 int err; 2356 2357 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 2358 if (err == 1) { 2359 *l = lo; 2360 ret = lo->lo_number; 2361 } 2362 goto out; 2363 } 2364 2365 /* lookup and return a specific i */ 2366 lo = idr_find(&loop_index_idr, i); 2367 if (lo) { 2368 *l = lo; 2369 ret = lo->lo_number; 2370 } 2371 out: 2372 return ret; 2373 } 2374 2375 static void loop_probe(dev_t dev) 2376 { 2377 int idx = MINOR(dev) >> part_shift; 2378 struct loop_device *lo; 2379 2380 if (max_loop && idx >= max_loop) 2381 return; 2382 2383 mutex_lock(&loop_ctl_mutex); 2384 if (loop_lookup(&lo, idx) < 0) 2385 loop_add(&lo, idx); 2386 mutex_unlock(&loop_ctl_mutex); 2387 } 2388 2389 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2390 unsigned long parm) 2391 { 2392 struct loop_device *lo; 2393 int ret; 2394 2395 ret = mutex_lock_killable(&loop_ctl_mutex); 2396 if (ret) 2397 return ret; 2398 2399 ret = -ENOSYS; 2400 switch (cmd) { 2401 case LOOP_CTL_ADD: 2402 ret = loop_lookup(&lo, parm); 2403 if (ret >= 0) { 2404 ret = -EEXIST; 2405 break; 2406 } 2407 ret = loop_add(&lo, parm); 2408 break; 2409 case LOOP_CTL_REMOVE: 2410 ret = loop_lookup(&lo, parm); 2411 if (ret < 0) 2412 break; 2413 ret = mutex_lock_killable(&lo->lo_mutex); 2414 if (ret) 2415 break; 2416 if (lo->lo_state != Lo_unbound) { 2417 ret = -EBUSY; 2418 mutex_unlock(&lo->lo_mutex); 2419 break; 2420 } 2421 if (atomic_read(&lo->lo_refcnt) > 0) { 2422 ret = -EBUSY; 2423 mutex_unlock(&lo->lo_mutex); 2424 break; 2425 } 2426 lo->lo_state = Lo_deleting; 2427 mutex_unlock(&lo->lo_mutex); 2428 idr_remove(&loop_index_idr, lo->lo_number); 2429 loop_remove(lo); 2430 break; 2431 case LOOP_CTL_GET_FREE: 2432 ret = loop_lookup(&lo, -1); 2433 if (ret >= 0) 2434 break; 2435 ret = loop_add(&lo, -1); 2436 } 2437 mutex_unlock(&loop_ctl_mutex); 2438 2439 return ret; 2440 } 2441 2442 static const struct file_operations loop_ctl_fops = { 2443 .open = nonseekable_open, 2444 .unlocked_ioctl = loop_control_ioctl, 2445 .compat_ioctl = loop_control_ioctl, 2446 .owner = THIS_MODULE, 2447 .llseek = noop_llseek, 2448 }; 2449 2450 static struct miscdevice loop_misc = { 2451 .minor = LOOP_CTRL_MINOR, 2452 .name = "loop-control", 2453 .fops = &loop_ctl_fops, 2454 }; 2455 2456 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2457 MODULE_ALIAS("devname:loop-control"); 2458 2459 static int __init loop_init(void) 2460 { 2461 int i, nr; 2462 struct loop_device *lo; 2463 int err; 2464 2465 part_shift = 0; 2466 if (max_part > 0) { 2467 part_shift = fls(max_part); 2468 2469 /* 2470 * Adjust max_part according to part_shift as it is exported 2471 * to user space so that user can decide correct minor number 2472 * if [s]he want to create more devices. 2473 * 2474 * Note that -1 is required because partition 0 is reserved 2475 * for the whole disk. 2476 */ 2477 max_part = (1UL << part_shift) - 1; 2478 } 2479 2480 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2481 err = -EINVAL; 2482 goto err_out; 2483 } 2484 2485 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2486 err = -EINVAL; 2487 goto err_out; 2488 } 2489 2490 /* 2491 * If max_loop is specified, create that many devices upfront. 2492 * This also becomes a hard limit. If max_loop is not specified, 2493 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2494 * init time. Loop devices can be requested on-demand with the 2495 * /dev/loop-control interface, or be instantiated by accessing 2496 * a 'dead' device node. 2497 */ 2498 if (max_loop) 2499 nr = max_loop; 2500 else 2501 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2502 2503 err = misc_register(&loop_misc); 2504 if (err < 0) 2505 goto err_out; 2506 2507 2508 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2509 err = -EIO; 2510 goto misc_out; 2511 } 2512 2513 /* pre-create number of devices given by config or max_loop */ 2514 mutex_lock(&loop_ctl_mutex); 2515 for (i = 0; i < nr; i++) 2516 loop_add(&lo, i); 2517 mutex_unlock(&loop_ctl_mutex); 2518 2519 printk(KERN_INFO "loop: module loaded\n"); 2520 return 0; 2521 2522 misc_out: 2523 misc_deregister(&loop_misc); 2524 err_out: 2525 return err; 2526 } 2527 2528 static int loop_exit_cb(int id, void *ptr, void *data) 2529 { 2530 struct loop_device *lo = ptr; 2531 2532 loop_remove(lo); 2533 return 0; 2534 } 2535 2536 static void __exit loop_exit(void) 2537 { 2538 mutex_lock(&loop_ctl_mutex); 2539 2540 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2541 idr_destroy(&loop_index_idr); 2542 2543 unregister_blkdev(LOOP_MAJOR, "loop"); 2544 2545 misc_deregister(&loop_misc); 2546 2547 mutex_unlock(&loop_ctl_mutex); 2548 } 2549 2550 module_init(loop_init); 2551 module_exit(loop_exit); 2552 2553 #ifndef MODULE 2554 static int __init max_loop_setup(char *str) 2555 { 2556 max_loop = simple_strtol(str, NULL, 0); 2557 return 1; 2558 } 2559 2560 __setup("max_loop=", max_loop_setup); 2561 #endif 2562