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 #include <linux/sched/mm.h> 82 83 #include "loop.h" 84 85 #include <linux/uaccess.h> 86 87 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) 88 89 static DEFINE_IDR(loop_index_idr); 90 static DEFINE_MUTEX(loop_ctl_mutex); 91 static DEFINE_MUTEX(loop_validate_mutex); 92 93 /** 94 * loop_global_lock_killable() - take locks for safe loop_validate_file() test 95 * 96 * @lo: struct loop_device 97 * @global: true if @lo is about to bind another "struct loop_device", false otherwise 98 * 99 * Returns 0 on success, -EINTR otherwise. 100 * 101 * Since loop_validate_file() traverses on other "struct loop_device" if 102 * is_loop_device() is true, we need a global lock for serializing concurrent 103 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls. 104 */ 105 static int loop_global_lock_killable(struct loop_device *lo, bool global) 106 { 107 int err; 108 109 if (global) { 110 err = mutex_lock_killable(&loop_validate_mutex); 111 if (err) 112 return err; 113 } 114 err = mutex_lock_killable(&lo->lo_mutex); 115 if (err && global) 116 mutex_unlock(&loop_validate_mutex); 117 return err; 118 } 119 120 /** 121 * loop_global_unlock() - release locks taken by loop_global_lock_killable() 122 * 123 * @lo: struct loop_device 124 * @global: true if @lo was about to bind another "struct loop_device", false otherwise 125 */ 126 static void loop_global_unlock(struct loop_device *lo, bool global) 127 { 128 mutex_unlock(&lo->lo_mutex); 129 if (global) 130 mutex_unlock(&loop_validate_mutex); 131 } 132 133 static int max_part; 134 static int part_shift; 135 136 static int transfer_xor(struct loop_device *lo, int cmd, 137 struct page *raw_page, unsigned raw_off, 138 struct page *loop_page, unsigned loop_off, 139 int size, sector_t real_block) 140 { 141 char *raw_buf = kmap_atomic(raw_page) + raw_off; 142 char *loop_buf = kmap_atomic(loop_page) + loop_off; 143 char *in, *out, *key; 144 int i, keysize; 145 146 if (cmd == READ) { 147 in = raw_buf; 148 out = loop_buf; 149 } else { 150 in = loop_buf; 151 out = raw_buf; 152 } 153 154 key = lo->lo_encrypt_key; 155 keysize = lo->lo_encrypt_key_size; 156 for (i = 0; i < size; i++) 157 *out++ = *in++ ^ key[(i & 511) % keysize]; 158 159 kunmap_atomic(loop_buf); 160 kunmap_atomic(raw_buf); 161 cond_resched(); 162 return 0; 163 } 164 165 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 166 { 167 if (unlikely(info->lo_encrypt_key_size <= 0)) 168 return -EINVAL; 169 return 0; 170 } 171 172 static struct loop_func_table none_funcs = { 173 .number = LO_CRYPT_NONE, 174 }; 175 176 static struct loop_func_table xor_funcs = { 177 .number = LO_CRYPT_XOR, 178 .transfer = transfer_xor, 179 .init = xor_init 180 }; 181 182 /* xfer_funcs[0] is special - its release function is never called */ 183 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 184 &none_funcs, 185 &xor_funcs 186 }; 187 188 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 189 { 190 loff_t loopsize; 191 192 /* Compute loopsize in bytes */ 193 loopsize = i_size_read(file->f_mapping->host); 194 if (offset > 0) 195 loopsize -= offset; 196 /* offset is beyond i_size, weird but possible */ 197 if (loopsize < 0) 198 return 0; 199 200 if (sizelimit > 0 && sizelimit < loopsize) 201 loopsize = sizelimit; 202 /* 203 * Unfortunately, if we want to do I/O on the device, 204 * the number of 512-byte sectors has to fit into a sector_t. 205 */ 206 return loopsize >> 9; 207 } 208 209 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 210 { 211 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 212 } 213 214 static void __loop_update_dio(struct loop_device *lo, bool dio) 215 { 216 struct file *file = lo->lo_backing_file; 217 struct address_space *mapping = file->f_mapping; 218 struct inode *inode = mapping->host; 219 unsigned short sb_bsize = 0; 220 unsigned dio_align = 0; 221 bool use_dio; 222 223 if (inode->i_sb->s_bdev) { 224 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 225 dio_align = sb_bsize - 1; 226 } 227 228 /* 229 * We support direct I/O only if lo_offset is aligned with the 230 * logical I/O size of backing device, and the logical block 231 * size of loop is bigger than the backing device's and the loop 232 * needn't transform transfer. 233 * 234 * TODO: the above condition may be loosed in the future, and 235 * direct I/O may be switched runtime at that time because most 236 * of requests in sane applications should be PAGE_SIZE aligned 237 */ 238 if (dio) { 239 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 240 !(lo->lo_offset & dio_align) && 241 mapping->a_ops->direct_IO && 242 !lo->transfer) 243 use_dio = true; 244 else 245 use_dio = false; 246 } else { 247 use_dio = false; 248 } 249 250 if (lo->use_dio == use_dio) 251 return; 252 253 /* flush dirty pages before changing direct IO */ 254 vfs_fsync(file, 0); 255 256 /* 257 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 258 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 259 * will get updated by ioctl(LOOP_GET_STATUS) 260 */ 261 if (lo->lo_state == Lo_bound) 262 blk_mq_freeze_queue(lo->lo_queue); 263 lo->use_dio = use_dio; 264 if (use_dio) { 265 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 266 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 267 } else { 268 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 269 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 270 } 271 if (lo->lo_state == Lo_bound) 272 blk_mq_unfreeze_queue(lo->lo_queue); 273 } 274 275 /** 276 * loop_set_size() - sets device size and notifies userspace 277 * @lo: struct loop_device to set the size for 278 * @size: new size of the loop device 279 * 280 * Callers must validate that the size passed into this function fits into 281 * a sector_t, eg using loop_validate_size() 282 */ 283 static void loop_set_size(struct loop_device *lo, loff_t size) 284 { 285 if (!set_capacity_and_notify(lo->lo_disk, size)) 286 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 287 } 288 289 static inline int 290 lo_do_transfer(struct loop_device *lo, int cmd, 291 struct page *rpage, unsigned roffs, 292 struct page *lpage, unsigned loffs, 293 int size, sector_t rblock) 294 { 295 int ret; 296 297 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 298 if (likely(!ret)) 299 return 0; 300 301 printk_ratelimited(KERN_ERR 302 "loop: Transfer error at byte offset %llu, length %i.\n", 303 (unsigned long long)rblock << 9, size); 304 return ret; 305 } 306 307 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 308 { 309 struct iov_iter i; 310 ssize_t bw; 311 312 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 313 314 file_start_write(file); 315 bw = vfs_iter_write(file, &i, ppos, 0); 316 file_end_write(file); 317 318 if (likely(bw == bvec->bv_len)) 319 return 0; 320 321 printk_ratelimited(KERN_ERR 322 "loop: Write error at byte offset %llu, length %i.\n", 323 (unsigned long long)*ppos, bvec->bv_len); 324 if (bw >= 0) 325 bw = -EIO; 326 return bw; 327 } 328 329 static int lo_write_simple(struct loop_device *lo, struct request *rq, 330 loff_t pos) 331 { 332 struct bio_vec bvec; 333 struct req_iterator iter; 334 int ret = 0; 335 336 rq_for_each_segment(bvec, rq, iter) { 337 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 338 if (ret < 0) 339 break; 340 cond_resched(); 341 } 342 343 return ret; 344 } 345 346 /* 347 * This is the slow, transforming version that needs to double buffer the 348 * data as it cannot do the transformations in place without having direct 349 * access to the destination pages of the backing file. 350 */ 351 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 352 loff_t pos) 353 { 354 struct bio_vec bvec, b; 355 struct req_iterator iter; 356 struct page *page; 357 int ret = 0; 358 359 page = alloc_page(GFP_NOIO); 360 if (unlikely(!page)) 361 return -ENOMEM; 362 363 rq_for_each_segment(bvec, rq, iter) { 364 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 365 bvec.bv_offset, bvec.bv_len, pos >> 9); 366 if (unlikely(ret)) 367 break; 368 369 b.bv_page = page; 370 b.bv_offset = 0; 371 b.bv_len = bvec.bv_len; 372 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 373 if (ret < 0) 374 break; 375 } 376 377 __free_page(page); 378 return ret; 379 } 380 381 static int lo_read_simple(struct loop_device *lo, struct request *rq, 382 loff_t pos) 383 { 384 struct bio_vec bvec; 385 struct req_iterator iter; 386 struct iov_iter i; 387 ssize_t len; 388 389 rq_for_each_segment(bvec, rq, iter) { 390 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 391 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 392 if (len < 0) 393 return len; 394 395 flush_dcache_page(bvec.bv_page); 396 397 if (len != bvec.bv_len) { 398 struct bio *bio; 399 400 __rq_for_each_bio(bio, rq) 401 zero_fill_bio(bio); 402 break; 403 } 404 cond_resched(); 405 } 406 407 return 0; 408 } 409 410 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 411 loff_t pos) 412 { 413 struct bio_vec bvec, b; 414 struct req_iterator iter; 415 struct iov_iter i; 416 struct page *page; 417 ssize_t len; 418 int ret = 0; 419 420 page = alloc_page(GFP_NOIO); 421 if (unlikely(!page)) 422 return -ENOMEM; 423 424 rq_for_each_segment(bvec, rq, iter) { 425 loff_t offset = pos; 426 427 b.bv_page = page; 428 b.bv_offset = 0; 429 b.bv_len = bvec.bv_len; 430 431 iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 432 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 433 if (len < 0) { 434 ret = len; 435 goto out_free_page; 436 } 437 438 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 439 bvec.bv_offset, len, offset >> 9); 440 if (ret) 441 goto out_free_page; 442 443 flush_dcache_page(bvec.bv_page); 444 445 if (len != bvec.bv_len) { 446 struct bio *bio; 447 448 __rq_for_each_bio(bio, rq) 449 zero_fill_bio(bio); 450 break; 451 } 452 } 453 454 ret = 0; 455 out_free_page: 456 __free_page(page); 457 return ret; 458 } 459 460 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 461 int mode) 462 { 463 /* 464 * We use fallocate to manipulate the space mappings used by the image 465 * a.k.a. discard/zerorange. However we do not support this if 466 * encryption is enabled, because it may give an attacker useful 467 * information. 468 */ 469 struct file *file = lo->lo_backing_file; 470 struct request_queue *q = lo->lo_queue; 471 int ret; 472 473 mode |= FALLOC_FL_KEEP_SIZE; 474 475 if (!blk_queue_discard(q)) { 476 ret = -EOPNOTSUPP; 477 goto out; 478 } 479 480 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 481 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 482 ret = -EIO; 483 out: 484 return ret; 485 } 486 487 static int lo_req_flush(struct loop_device *lo, struct request *rq) 488 { 489 struct file *file = lo->lo_backing_file; 490 int ret = vfs_fsync(file, 0); 491 if (unlikely(ret && ret != -EINVAL)) 492 ret = -EIO; 493 494 return ret; 495 } 496 497 static void lo_complete_rq(struct request *rq) 498 { 499 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 500 blk_status_t ret = BLK_STS_OK; 501 502 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 503 req_op(rq) != REQ_OP_READ) { 504 if (cmd->ret < 0) 505 ret = errno_to_blk_status(cmd->ret); 506 goto end_io; 507 } 508 509 /* 510 * Short READ - if we got some data, advance our request and 511 * retry it. If we got no data, end the rest with EIO. 512 */ 513 if (cmd->ret) { 514 blk_update_request(rq, BLK_STS_OK, cmd->ret); 515 cmd->ret = 0; 516 blk_mq_requeue_request(rq, true); 517 } else { 518 if (cmd->use_aio) { 519 struct bio *bio = rq->bio; 520 521 while (bio) { 522 zero_fill_bio(bio); 523 bio = bio->bi_next; 524 } 525 } 526 ret = BLK_STS_IOERR; 527 end_io: 528 blk_mq_end_request(rq, ret); 529 } 530 } 531 532 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 533 { 534 struct request *rq = blk_mq_rq_from_pdu(cmd); 535 536 if (!atomic_dec_and_test(&cmd->ref)) 537 return; 538 kfree(cmd->bvec); 539 cmd->bvec = NULL; 540 if (likely(!blk_should_fake_timeout(rq->q))) 541 blk_mq_complete_request(rq); 542 } 543 544 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 545 { 546 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 547 548 cmd->ret = ret; 549 lo_rw_aio_do_completion(cmd); 550 } 551 552 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 553 loff_t pos, bool rw) 554 { 555 struct iov_iter iter; 556 struct req_iterator rq_iter; 557 struct bio_vec *bvec; 558 struct request *rq = blk_mq_rq_from_pdu(cmd); 559 struct bio *bio = rq->bio; 560 struct file *file = lo->lo_backing_file; 561 struct bio_vec tmp; 562 unsigned int offset; 563 int nr_bvec = 0; 564 int ret; 565 566 rq_for_each_bvec(tmp, rq, rq_iter) 567 nr_bvec++; 568 569 if (rq->bio != rq->biotail) { 570 571 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 572 GFP_NOIO); 573 if (!bvec) 574 return -EIO; 575 cmd->bvec = bvec; 576 577 /* 578 * The bios of the request may be started from the middle of 579 * the 'bvec' because of bio splitting, so we can't directly 580 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 581 * API will take care of all details for us. 582 */ 583 rq_for_each_bvec(tmp, rq, rq_iter) { 584 *bvec = tmp; 585 bvec++; 586 } 587 bvec = cmd->bvec; 588 offset = 0; 589 } else { 590 /* 591 * Same here, this bio may be started from the middle of the 592 * 'bvec' because of bio splitting, so offset from the bvec 593 * must be passed to iov iterator 594 */ 595 offset = bio->bi_iter.bi_bvec_done; 596 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 597 } 598 atomic_set(&cmd->ref, 2); 599 600 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 601 iter.iov_offset = offset; 602 603 cmd->iocb.ki_pos = pos; 604 cmd->iocb.ki_filp = file; 605 cmd->iocb.ki_complete = lo_rw_aio_complete; 606 cmd->iocb.ki_flags = IOCB_DIRECT; 607 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 608 609 if (rw == WRITE) 610 ret = call_write_iter(file, &cmd->iocb, &iter); 611 else 612 ret = call_read_iter(file, &cmd->iocb, &iter); 613 614 lo_rw_aio_do_completion(cmd); 615 616 if (ret != -EIOCBQUEUED) 617 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 618 return 0; 619 } 620 621 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 622 { 623 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 624 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 625 626 /* 627 * lo_write_simple and lo_read_simple should have been covered 628 * by io submit style function like lo_rw_aio(), one blocker 629 * is that lo_read_simple() need to call flush_dcache_page after 630 * the page is written from kernel, and it isn't easy to handle 631 * this in io submit style function which submits all segments 632 * of the req at one time. And direct read IO doesn't need to 633 * run flush_dcache_page(). 634 */ 635 switch (req_op(rq)) { 636 case REQ_OP_FLUSH: 637 return lo_req_flush(lo, rq); 638 case REQ_OP_WRITE_ZEROES: 639 /* 640 * If the caller doesn't want deallocation, call zeroout to 641 * write zeroes the range. Otherwise, punch them out. 642 */ 643 return lo_fallocate(lo, rq, pos, 644 (rq->cmd_flags & REQ_NOUNMAP) ? 645 FALLOC_FL_ZERO_RANGE : 646 FALLOC_FL_PUNCH_HOLE); 647 case REQ_OP_DISCARD: 648 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 649 case REQ_OP_WRITE: 650 if (lo->transfer) 651 return lo_write_transfer(lo, rq, pos); 652 else if (cmd->use_aio) 653 return lo_rw_aio(lo, cmd, pos, WRITE); 654 else 655 return lo_write_simple(lo, rq, pos); 656 case REQ_OP_READ: 657 if (lo->transfer) 658 return lo_read_transfer(lo, rq, pos); 659 else if (cmd->use_aio) 660 return lo_rw_aio(lo, cmd, pos, READ); 661 else 662 return lo_read_simple(lo, rq, pos); 663 default: 664 WARN_ON_ONCE(1); 665 return -EIO; 666 } 667 } 668 669 static inline void loop_update_dio(struct loop_device *lo) 670 { 671 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 672 lo->use_dio); 673 } 674 675 static void loop_reread_partitions(struct loop_device *lo) 676 { 677 int rc; 678 679 mutex_lock(&lo->lo_disk->open_mutex); 680 rc = bdev_disk_changed(lo->lo_disk, false); 681 mutex_unlock(&lo->lo_disk->open_mutex); 682 if (rc) 683 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 684 __func__, lo->lo_number, lo->lo_file_name, rc); 685 } 686 687 static inline int is_loop_device(struct file *file) 688 { 689 struct inode *i = file->f_mapping->host; 690 691 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; 692 } 693 694 static int loop_validate_file(struct file *file, struct block_device *bdev) 695 { 696 struct inode *inode = file->f_mapping->host; 697 struct file *f = file; 698 699 /* Avoid recursion */ 700 while (is_loop_device(f)) { 701 struct loop_device *l; 702 703 lockdep_assert_held(&loop_validate_mutex); 704 if (f->f_mapping->host->i_rdev == bdev->bd_dev) 705 return -EBADF; 706 707 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 708 if (l->lo_state != Lo_bound) 709 return -EINVAL; 710 /* Order wrt setting lo->lo_backing_file in loop_configure(). */ 711 rmb(); 712 f = l->lo_backing_file; 713 } 714 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 715 return -EINVAL; 716 return 0; 717 } 718 719 /* 720 * loop_change_fd switched the backing store of a loopback device to 721 * a new file. This is useful for operating system installers to free up 722 * the original file and in High Availability environments to switch to 723 * an alternative location for the content in case of server meltdown. 724 * This can only work if the loop device is used read-only, and if the 725 * new backing store is the same size and type as the old backing store. 726 */ 727 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 728 unsigned int arg) 729 { 730 struct file *file = fget(arg); 731 struct file *old_file; 732 int error; 733 bool partscan; 734 bool is_loop; 735 736 if (!file) 737 return -EBADF; 738 is_loop = is_loop_device(file); 739 error = loop_global_lock_killable(lo, is_loop); 740 if (error) 741 goto out_putf; 742 error = -ENXIO; 743 if (lo->lo_state != Lo_bound) 744 goto out_err; 745 746 /* the loop device has to be read-only */ 747 error = -EINVAL; 748 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 749 goto out_err; 750 751 error = loop_validate_file(file, bdev); 752 if (error) 753 goto out_err; 754 755 old_file = lo->lo_backing_file; 756 757 error = -EINVAL; 758 759 /* size of the new backing store needs to be the same */ 760 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 761 goto out_err; 762 763 /* and ... switch */ 764 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 765 blk_mq_freeze_queue(lo->lo_queue); 766 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 767 lo->lo_backing_file = file; 768 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 769 mapping_set_gfp_mask(file->f_mapping, 770 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 771 loop_update_dio(lo); 772 blk_mq_unfreeze_queue(lo->lo_queue); 773 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 774 loop_global_unlock(lo, is_loop); 775 776 /* 777 * Flush loop_validate_file() before fput(), for l->lo_backing_file 778 * might be pointing at old_file which might be the last reference. 779 */ 780 if (!is_loop) { 781 mutex_lock(&loop_validate_mutex); 782 mutex_unlock(&loop_validate_mutex); 783 } 784 /* 785 * We must drop file reference outside of lo_mutex as dropping 786 * the file ref can take open_mutex which creates circular locking 787 * dependency. 788 */ 789 fput(old_file); 790 if (partscan) 791 loop_reread_partitions(lo); 792 return 0; 793 794 out_err: 795 loop_global_unlock(lo, is_loop); 796 out_putf: 797 fput(file); 798 return error; 799 } 800 801 /* loop sysfs attributes */ 802 803 static ssize_t loop_attr_show(struct device *dev, char *page, 804 ssize_t (*callback)(struct loop_device *, char *)) 805 { 806 struct gendisk *disk = dev_to_disk(dev); 807 struct loop_device *lo = disk->private_data; 808 809 return callback(lo, page); 810 } 811 812 #define LOOP_ATTR_RO(_name) \ 813 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 814 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 815 struct device_attribute *attr, char *b) \ 816 { \ 817 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 818 } \ 819 static struct device_attribute loop_attr_##_name = \ 820 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 821 822 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 823 { 824 ssize_t ret; 825 char *p = NULL; 826 827 spin_lock_irq(&lo->lo_lock); 828 if (lo->lo_backing_file) 829 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 830 spin_unlock_irq(&lo->lo_lock); 831 832 if (IS_ERR_OR_NULL(p)) 833 ret = PTR_ERR(p); 834 else { 835 ret = strlen(p); 836 memmove(buf, p, ret); 837 buf[ret++] = '\n'; 838 buf[ret] = 0; 839 } 840 841 return ret; 842 } 843 844 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 845 { 846 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 847 } 848 849 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 850 { 851 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 852 } 853 854 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 855 { 856 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 857 858 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 859 } 860 861 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 862 { 863 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 864 865 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 866 } 867 868 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 869 { 870 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 871 872 return sprintf(buf, "%s\n", dio ? "1" : "0"); 873 } 874 875 LOOP_ATTR_RO(backing_file); 876 LOOP_ATTR_RO(offset); 877 LOOP_ATTR_RO(sizelimit); 878 LOOP_ATTR_RO(autoclear); 879 LOOP_ATTR_RO(partscan); 880 LOOP_ATTR_RO(dio); 881 882 static struct attribute *loop_attrs[] = { 883 &loop_attr_backing_file.attr, 884 &loop_attr_offset.attr, 885 &loop_attr_sizelimit.attr, 886 &loop_attr_autoclear.attr, 887 &loop_attr_partscan.attr, 888 &loop_attr_dio.attr, 889 NULL, 890 }; 891 892 static struct attribute_group loop_attribute_group = { 893 .name = "loop", 894 .attrs= loop_attrs, 895 }; 896 897 static void loop_sysfs_init(struct loop_device *lo) 898 { 899 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 900 &loop_attribute_group); 901 } 902 903 static void loop_sysfs_exit(struct loop_device *lo) 904 { 905 if (lo->sysfs_inited) 906 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 907 &loop_attribute_group); 908 } 909 910 static void loop_config_discard(struct loop_device *lo) 911 { 912 struct file *file = lo->lo_backing_file; 913 struct inode *inode = file->f_mapping->host; 914 struct request_queue *q = lo->lo_queue; 915 u32 granularity, max_discard_sectors; 916 917 /* 918 * If the backing device is a block device, mirror its zeroing 919 * capability. Set the discard sectors to the block device's zeroing 920 * capabilities because loop discards result in blkdev_issue_zeroout(), 921 * not blkdev_issue_discard(). This maintains consistent behavior with 922 * file-backed loop devices: discarded regions read back as zero. 923 */ 924 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { 925 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 926 927 max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 928 granularity = backingq->limits.discard_granularity ?: 929 queue_physical_block_size(backingq); 930 931 /* 932 * We use punch hole to reclaim the free space used by the 933 * image a.k.a. discard. However we do not support discard if 934 * encryption is enabled, because it may give an attacker 935 * useful information. 936 */ 937 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { 938 max_discard_sectors = 0; 939 granularity = 0; 940 941 } else { 942 max_discard_sectors = UINT_MAX >> 9; 943 granularity = inode->i_sb->s_blocksize; 944 } 945 946 if (max_discard_sectors) { 947 q->limits.discard_granularity = granularity; 948 blk_queue_max_discard_sectors(q, max_discard_sectors); 949 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 950 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 951 } else { 952 q->limits.discard_granularity = 0; 953 blk_queue_max_discard_sectors(q, 0); 954 blk_queue_max_write_zeroes_sectors(q, 0); 955 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 956 } 957 q->limits.discard_alignment = 0; 958 } 959 960 struct loop_worker { 961 struct rb_node rb_node; 962 struct work_struct work; 963 struct list_head cmd_list; 964 struct list_head idle_list; 965 struct loop_device *lo; 966 struct cgroup_subsys_state *blkcg_css; 967 unsigned long last_ran_at; 968 }; 969 970 static void loop_workfn(struct work_struct *work); 971 static void loop_rootcg_workfn(struct work_struct *work); 972 static void loop_free_idle_workers(struct timer_list *timer); 973 974 #ifdef CONFIG_BLK_CGROUP 975 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 976 { 977 return !css || css == blkcg_root_css; 978 } 979 #else 980 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 981 { 982 return !css; 983 } 984 #endif 985 986 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 987 { 988 struct rb_node **node = &(lo->worker_tree.rb_node), *parent = NULL; 989 struct loop_worker *cur_worker, *worker = NULL; 990 struct work_struct *work; 991 struct list_head *cmd_list; 992 993 spin_lock_irq(&lo->lo_work_lock); 994 995 if (queue_on_root_worker(cmd->blkcg_css)) 996 goto queue_work; 997 998 node = &lo->worker_tree.rb_node; 999 1000 while (*node) { 1001 parent = *node; 1002 cur_worker = container_of(*node, struct loop_worker, rb_node); 1003 if (cur_worker->blkcg_css == cmd->blkcg_css) { 1004 worker = cur_worker; 1005 break; 1006 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { 1007 node = &(*node)->rb_left; 1008 } else { 1009 node = &(*node)->rb_right; 1010 } 1011 } 1012 if (worker) 1013 goto queue_work; 1014 1015 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 1016 /* 1017 * In the event we cannot allocate a worker, just queue on the 1018 * rootcg worker and issue the I/O as the rootcg 1019 */ 1020 if (!worker) { 1021 cmd->blkcg_css = NULL; 1022 if (cmd->memcg_css) 1023 css_put(cmd->memcg_css); 1024 cmd->memcg_css = NULL; 1025 goto queue_work; 1026 } 1027 1028 worker->blkcg_css = cmd->blkcg_css; 1029 css_get(worker->blkcg_css); 1030 INIT_WORK(&worker->work, loop_workfn); 1031 INIT_LIST_HEAD(&worker->cmd_list); 1032 INIT_LIST_HEAD(&worker->idle_list); 1033 worker->lo = lo; 1034 rb_link_node(&worker->rb_node, parent, node); 1035 rb_insert_color(&worker->rb_node, &lo->worker_tree); 1036 queue_work: 1037 if (worker) { 1038 /* 1039 * We need to remove from the idle list here while 1040 * holding the lock so that the idle timer doesn't 1041 * free the worker 1042 */ 1043 if (!list_empty(&worker->idle_list)) 1044 list_del_init(&worker->idle_list); 1045 work = &worker->work; 1046 cmd_list = &worker->cmd_list; 1047 } else { 1048 work = &lo->rootcg_work; 1049 cmd_list = &lo->rootcg_cmd_list; 1050 } 1051 list_add_tail(&cmd->list_entry, cmd_list); 1052 queue_work(lo->workqueue, work); 1053 spin_unlock_irq(&lo->lo_work_lock); 1054 } 1055 1056 static void loop_update_rotational(struct loop_device *lo) 1057 { 1058 struct file *file = lo->lo_backing_file; 1059 struct inode *file_inode = file->f_mapping->host; 1060 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 1061 struct request_queue *q = lo->lo_queue; 1062 bool nonrot = true; 1063 1064 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 1065 if (file_bdev) 1066 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 1067 1068 if (nonrot) 1069 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 1070 else 1071 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 1072 } 1073 1074 static int 1075 loop_release_xfer(struct loop_device *lo) 1076 { 1077 int err = 0; 1078 struct loop_func_table *xfer = lo->lo_encryption; 1079 1080 if (xfer) { 1081 if (xfer->release) 1082 err = xfer->release(lo); 1083 lo->transfer = NULL; 1084 lo->lo_encryption = NULL; 1085 module_put(xfer->owner); 1086 } 1087 return err; 1088 } 1089 1090 static int 1091 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1092 const struct loop_info64 *i) 1093 { 1094 int err = 0; 1095 1096 if (xfer) { 1097 struct module *owner = xfer->owner; 1098 1099 if (!try_module_get(owner)) 1100 return -EINVAL; 1101 if (xfer->init) 1102 err = xfer->init(lo, i); 1103 if (err) 1104 module_put(owner); 1105 else 1106 lo->lo_encryption = xfer; 1107 } 1108 return err; 1109 } 1110 1111 /** 1112 * loop_set_status_from_info - configure device from loop_info 1113 * @lo: struct loop_device to configure 1114 * @info: struct loop_info64 to configure the device with 1115 * 1116 * Configures the loop device parameters according to the passed 1117 * in loop_info64 configuration. 1118 */ 1119 static int 1120 loop_set_status_from_info(struct loop_device *lo, 1121 const struct loop_info64 *info) 1122 { 1123 int err; 1124 struct loop_func_table *xfer; 1125 kuid_t uid = current_uid(); 1126 1127 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1128 return -EINVAL; 1129 1130 err = loop_release_xfer(lo); 1131 if (err) 1132 return err; 1133 1134 if (info->lo_encrypt_type) { 1135 unsigned int type = info->lo_encrypt_type; 1136 1137 if (type >= MAX_LO_CRYPT) 1138 return -EINVAL; 1139 xfer = xfer_funcs[type]; 1140 if (xfer == NULL) 1141 return -EINVAL; 1142 } else 1143 xfer = NULL; 1144 1145 err = loop_init_xfer(lo, xfer, info); 1146 if (err) 1147 return err; 1148 1149 lo->lo_offset = info->lo_offset; 1150 lo->lo_sizelimit = info->lo_sizelimit; 1151 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1152 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1153 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1154 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1155 1156 if (!xfer) 1157 xfer = &none_funcs; 1158 lo->transfer = xfer->transfer; 1159 lo->ioctl = xfer->ioctl; 1160 1161 lo->lo_flags = info->lo_flags; 1162 1163 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1164 lo->lo_init[0] = info->lo_init[0]; 1165 lo->lo_init[1] = info->lo_init[1]; 1166 if (info->lo_encrypt_key_size) { 1167 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1168 info->lo_encrypt_key_size); 1169 lo->lo_key_owner = uid; 1170 } 1171 1172 return 0; 1173 } 1174 1175 static int loop_configure(struct loop_device *lo, fmode_t mode, 1176 struct block_device *bdev, 1177 const struct loop_config *config) 1178 { 1179 struct file *file = fget(config->fd); 1180 struct inode *inode; 1181 struct address_space *mapping; 1182 int error; 1183 loff_t size; 1184 bool partscan; 1185 unsigned short bsize; 1186 bool is_loop; 1187 1188 if (!file) 1189 return -EBADF; 1190 is_loop = is_loop_device(file); 1191 1192 /* This is safe, since we have a reference from open(). */ 1193 __module_get(THIS_MODULE); 1194 1195 /* 1196 * If we don't hold exclusive handle for the device, upgrade to it 1197 * here to avoid changing device under exclusive owner. 1198 */ 1199 if (!(mode & FMODE_EXCL)) { 1200 error = bd_prepare_to_claim(bdev, loop_configure); 1201 if (error) 1202 goto out_putf; 1203 } 1204 1205 error = loop_global_lock_killable(lo, is_loop); 1206 if (error) 1207 goto out_bdev; 1208 1209 error = -EBUSY; 1210 if (lo->lo_state != Lo_unbound) 1211 goto out_unlock; 1212 1213 error = loop_validate_file(file, bdev); 1214 if (error) 1215 goto out_unlock; 1216 1217 mapping = file->f_mapping; 1218 inode = mapping->host; 1219 1220 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1221 error = -EINVAL; 1222 goto out_unlock; 1223 } 1224 1225 if (config->block_size) { 1226 error = blk_validate_block_size(config->block_size); 1227 if (error) 1228 goto out_unlock; 1229 } 1230 1231 error = loop_set_status_from_info(lo, &config->info); 1232 if (error) 1233 goto out_unlock; 1234 1235 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1236 !file->f_op->write_iter) 1237 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1238 1239 lo->workqueue = alloc_workqueue("loop%d", 1240 WQ_UNBOUND | WQ_FREEZABLE, 1241 0, 1242 lo->lo_number); 1243 if (!lo->workqueue) { 1244 error = -ENOMEM; 1245 goto out_unlock; 1246 } 1247 1248 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1249 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1250 1251 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 1252 INIT_LIST_HEAD(&lo->rootcg_cmd_list); 1253 INIT_LIST_HEAD(&lo->idle_worker_list); 1254 lo->worker_tree = RB_ROOT; 1255 timer_setup(&lo->timer, loop_free_idle_workers, 1256 TIMER_DEFERRABLE); 1257 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1258 lo->lo_device = bdev; 1259 lo->lo_backing_file = file; 1260 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1261 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1262 1263 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1264 blk_queue_write_cache(lo->lo_queue, true, false); 1265 1266 if (config->block_size) 1267 bsize = config->block_size; 1268 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1269 /* In case of direct I/O, match underlying block size */ 1270 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1271 else 1272 bsize = 512; 1273 1274 blk_queue_logical_block_size(lo->lo_queue, bsize); 1275 blk_queue_physical_block_size(lo->lo_queue, bsize); 1276 blk_queue_io_min(lo->lo_queue, bsize); 1277 1278 loop_config_discard(lo); 1279 loop_update_rotational(lo); 1280 loop_update_dio(lo); 1281 loop_sysfs_init(lo); 1282 1283 size = get_loop_size(lo, file); 1284 loop_set_size(lo, size); 1285 1286 /* Order wrt reading lo_state in loop_validate_file(). */ 1287 wmb(); 1288 1289 lo->lo_state = Lo_bound; 1290 if (part_shift) 1291 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1292 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1293 if (partscan) 1294 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1295 1296 loop_global_unlock(lo, is_loop); 1297 if (partscan) 1298 loop_reread_partitions(lo); 1299 if (!(mode & FMODE_EXCL)) 1300 bd_abort_claiming(bdev, loop_configure); 1301 return 0; 1302 1303 out_unlock: 1304 loop_global_unlock(lo, is_loop); 1305 out_bdev: 1306 if (!(mode & FMODE_EXCL)) 1307 bd_abort_claiming(bdev, loop_configure); 1308 out_putf: 1309 fput(file); 1310 /* This is safe: open() is still holding a reference. */ 1311 module_put(THIS_MODULE); 1312 return error; 1313 } 1314 1315 static int __loop_clr_fd(struct loop_device *lo, bool release) 1316 { 1317 struct file *filp = NULL; 1318 gfp_t gfp = lo->old_gfp_mask; 1319 struct block_device *bdev = lo->lo_device; 1320 int err = 0; 1321 bool partscan = false; 1322 int lo_number; 1323 struct loop_worker *pos, *worker; 1324 1325 /* 1326 * Flush loop_configure() and loop_change_fd(). It is acceptable for 1327 * loop_validate_file() to succeed, for actual clear operation has not 1328 * started yet. 1329 */ 1330 mutex_lock(&loop_validate_mutex); 1331 mutex_unlock(&loop_validate_mutex); 1332 /* 1333 * loop_validate_file() now fails because l->lo_state != Lo_bound 1334 * became visible. 1335 */ 1336 1337 mutex_lock(&lo->lo_mutex); 1338 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1339 err = -ENXIO; 1340 goto out_unlock; 1341 } 1342 1343 filp = lo->lo_backing_file; 1344 if (filp == NULL) { 1345 err = -EINVAL; 1346 goto out_unlock; 1347 } 1348 1349 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1350 blk_queue_write_cache(lo->lo_queue, false, false); 1351 1352 /* freeze request queue during the transition */ 1353 blk_mq_freeze_queue(lo->lo_queue); 1354 1355 destroy_workqueue(lo->workqueue); 1356 spin_lock_irq(&lo->lo_work_lock); 1357 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 1358 idle_list) { 1359 list_del(&worker->idle_list); 1360 rb_erase(&worker->rb_node, &lo->worker_tree); 1361 css_put(worker->blkcg_css); 1362 kfree(worker); 1363 } 1364 spin_unlock_irq(&lo->lo_work_lock); 1365 del_timer_sync(&lo->timer); 1366 1367 spin_lock_irq(&lo->lo_lock); 1368 lo->lo_backing_file = NULL; 1369 spin_unlock_irq(&lo->lo_lock); 1370 1371 loop_release_xfer(lo); 1372 lo->transfer = NULL; 1373 lo->ioctl = NULL; 1374 lo->lo_device = NULL; 1375 lo->lo_encryption = NULL; 1376 lo->lo_offset = 0; 1377 lo->lo_sizelimit = 0; 1378 lo->lo_encrypt_key_size = 0; 1379 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1380 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1381 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1382 blk_queue_logical_block_size(lo->lo_queue, 512); 1383 blk_queue_physical_block_size(lo->lo_queue, 512); 1384 blk_queue_io_min(lo->lo_queue, 512); 1385 if (bdev) { 1386 invalidate_bdev(bdev); 1387 bdev->bd_inode->i_mapping->wb_err = 0; 1388 } 1389 set_capacity(lo->lo_disk, 0); 1390 loop_sysfs_exit(lo); 1391 if (bdev) { 1392 /* let user-space know about this change */ 1393 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1394 } 1395 mapping_set_gfp_mask(filp->f_mapping, gfp); 1396 /* This is safe: open() is still holding a reference. */ 1397 module_put(THIS_MODULE); 1398 blk_mq_unfreeze_queue(lo->lo_queue); 1399 1400 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1401 lo_number = lo->lo_number; 1402 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1403 out_unlock: 1404 mutex_unlock(&lo->lo_mutex); 1405 if (partscan) { 1406 /* 1407 * open_mutex has been held already in release path, so don't 1408 * acquire it if this function is called in such case. 1409 * 1410 * If the reread partition isn't from release path, lo_refcnt 1411 * must be at least one and it can only become zero when the 1412 * current holder is released. 1413 */ 1414 if (!release) 1415 mutex_lock(&lo->lo_disk->open_mutex); 1416 err = bdev_disk_changed(lo->lo_disk, false); 1417 if (!release) 1418 mutex_unlock(&lo->lo_disk->open_mutex); 1419 if (err) 1420 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1421 __func__, lo_number, err); 1422 /* Device is gone, no point in returning error */ 1423 err = 0; 1424 } 1425 1426 /* 1427 * lo->lo_state is set to Lo_unbound here after above partscan has 1428 * finished. 1429 * 1430 * There cannot be anybody else entering __loop_clr_fd() as 1431 * lo->lo_backing_file is already cleared and Lo_rundown state 1432 * protects us from all the other places trying to change the 'lo' 1433 * device. 1434 */ 1435 mutex_lock(&lo->lo_mutex); 1436 lo->lo_flags = 0; 1437 if (!part_shift) 1438 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1439 lo->lo_state = Lo_unbound; 1440 mutex_unlock(&lo->lo_mutex); 1441 1442 /* 1443 * Need not hold lo_mutex to fput backing file. Calling fput holding 1444 * lo_mutex triggers a circular lock dependency possibility warning as 1445 * fput can take open_mutex which is usually taken before lo_mutex. 1446 */ 1447 if (filp) 1448 fput(filp); 1449 return err; 1450 } 1451 1452 static int loop_clr_fd(struct loop_device *lo) 1453 { 1454 int err; 1455 1456 err = mutex_lock_killable(&lo->lo_mutex); 1457 if (err) 1458 return err; 1459 if (lo->lo_state != Lo_bound) { 1460 mutex_unlock(&lo->lo_mutex); 1461 return -ENXIO; 1462 } 1463 /* 1464 * If we've explicitly asked to tear down the loop device, 1465 * and it has an elevated reference count, set it for auto-teardown when 1466 * the last reference goes away. This stops $!~#$@ udev from 1467 * preventing teardown because it decided that it needs to run blkid on 1468 * the loopback device whenever they appear. xfstests is notorious for 1469 * failing tests because blkid via udev races with a losetup 1470 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1471 * command to fail with EBUSY. 1472 */ 1473 if (atomic_read(&lo->lo_refcnt) > 1) { 1474 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1475 mutex_unlock(&lo->lo_mutex); 1476 return 0; 1477 } 1478 lo->lo_state = Lo_rundown; 1479 mutex_unlock(&lo->lo_mutex); 1480 1481 return __loop_clr_fd(lo, false); 1482 } 1483 1484 static int 1485 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1486 { 1487 int err; 1488 kuid_t uid = current_uid(); 1489 int prev_lo_flags; 1490 bool partscan = false; 1491 bool size_changed = false; 1492 1493 err = mutex_lock_killable(&lo->lo_mutex); 1494 if (err) 1495 return err; 1496 if (lo->lo_encrypt_key_size && 1497 !uid_eq(lo->lo_key_owner, uid) && 1498 !capable(CAP_SYS_ADMIN)) { 1499 err = -EPERM; 1500 goto out_unlock; 1501 } 1502 if (lo->lo_state != Lo_bound) { 1503 err = -ENXIO; 1504 goto out_unlock; 1505 } 1506 1507 if (lo->lo_offset != info->lo_offset || 1508 lo->lo_sizelimit != info->lo_sizelimit) { 1509 size_changed = true; 1510 sync_blockdev(lo->lo_device); 1511 invalidate_bdev(lo->lo_device); 1512 } 1513 1514 /* I/O need to be drained during transfer transition */ 1515 blk_mq_freeze_queue(lo->lo_queue); 1516 1517 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1518 /* If any pages were dirtied after invalidate_bdev(), try again */ 1519 err = -EAGAIN; 1520 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1521 __func__, lo->lo_number, lo->lo_file_name, 1522 lo->lo_device->bd_inode->i_mapping->nrpages); 1523 goto out_unfreeze; 1524 } 1525 1526 prev_lo_flags = lo->lo_flags; 1527 1528 err = loop_set_status_from_info(lo, info); 1529 if (err) 1530 goto out_unfreeze; 1531 1532 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1533 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1534 /* For those flags, use the previous values instead */ 1535 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1536 /* For flags that can't be cleared, use previous values too */ 1537 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1538 1539 if (size_changed) { 1540 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1541 lo->lo_backing_file); 1542 loop_set_size(lo, new_size); 1543 } 1544 1545 loop_config_discard(lo); 1546 1547 /* update dio if lo_offset or transfer is changed */ 1548 __loop_update_dio(lo, lo->use_dio); 1549 1550 out_unfreeze: 1551 blk_mq_unfreeze_queue(lo->lo_queue); 1552 1553 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1554 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1555 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1556 partscan = true; 1557 } 1558 out_unlock: 1559 mutex_unlock(&lo->lo_mutex); 1560 if (partscan) 1561 loop_reread_partitions(lo); 1562 1563 return err; 1564 } 1565 1566 static int 1567 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1568 { 1569 struct path path; 1570 struct kstat stat; 1571 int ret; 1572 1573 ret = mutex_lock_killable(&lo->lo_mutex); 1574 if (ret) 1575 return ret; 1576 if (lo->lo_state != Lo_bound) { 1577 mutex_unlock(&lo->lo_mutex); 1578 return -ENXIO; 1579 } 1580 1581 memset(info, 0, sizeof(*info)); 1582 info->lo_number = lo->lo_number; 1583 info->lo_offset = lo->lo_offset; 1584 info->lo_sizelimit = lo->lo_sizelimit; 1585 info->lo_flags = lo->lo_flags; 1586 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1587 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1588 info->lo_encrypt_type = 1589 lo->lo_encryption ? lo->lo_encryption->number : 0; 1590 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1591 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1592 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1593 lo->lo_encrypt_key_size); 1594 } 1595 1596 /* Drop lo_mutex while we call into the filesystem. */ 1597 path = lo->lo_backing_file->f_path; 1598 path_get(&path); 1599 mutex_unlock(&lo->lo_mutex); 1600 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1601 if (!ret) { 1602 info->lo_device = huge_encode_dev(stat.dev); 1603 info->lo_inode = stat.ino; 1604 info->lo_rdevice = huge_encode_dev(stat.rdev); 1605 } 1606 path_put(&path); 1607 return ret; 1608 } 1609 1610 static void 1611 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1612 { 1613 memset(info64, 0, sizeof(*info64)); 1614 info64->lo_number = info->lo_number; 1615 info64->lo_device = info->lo_device; 1616 info64->lo_inode = info->lo_inode; 1617 info64->lo_rdevice = info->lo_rdevice; 1618 info64->lo_offset = info->lo_offset; 1619 info64->lo_sizelimit = 0; 1620 info64->lo_encrypt_type = info->lo_encrypt_type; 1621 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1622 info64->lo_flags = info->lo_flags; 1623 info64->lo_init[0] = info->lo_init[0]; 1624 info64->lo_init[1] = info->lo_init[1]; 1625 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1626 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1627 else 1628 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1629 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1630 } 1631 1632 static int 1633 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1634 { 1635 memset(info, 0, sizeof(*info)); 1636 info->lo_number = info64->lo_number; 1637 info->lo_device = info64->lo_device; 1638 info->lo_inode = info64->lo_inode; 1639 info->lo_rdevice = info64->lo_rdevice; 1640 info->lo_offset = info64->lo_offset; 1641 info->lo_encrypt_type = info64->lo_encrypt_type; 1642 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1643 info->lo_flags = info64->lo_flags; 1644 info->lo_init[0] = info64->lo_init[0]; 1645 info->lo_init[1] = info64->lo_init[1]; 1646 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1647 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1648 else 1649 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1650 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1651 1652 /* error in case values were truncated */ 1653 if (info->lo_device != info64->lo_device || 1654 info->lo_rdevice != info64->lo_rdevice || 1655 info->lo_inode != info64->lo_inode || 1656 info->lo_offset != info64->lo_offset) 1657 return -EOVERFLOW; 1658 1659 return 0; 1660 } 1661 1662 static int 1663 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1664 { 1665 struct loop_info info; 1666 struct loop_info64 info64; 1667 1668 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1669 return -EFAULT; 1670 loop_info64_from_old(&info, &info64); 1671 return loop_set_status(lo, &info64); 1672 } 1673 1674 static int 1675 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1676 { 1677 struct loop_info64 info64; 1678 1679 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1680 return -EFAULT; 1681 return loop_set_status(lo, &info64); 1682 } 1683 1684 static int 1685 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1686 struct loop_info info; 1687 struct loop_info64 info64; 1688 int err; 1689 1690 if (!arg) 1691 return -EINVAL; 1692 err = loop_get_status(lo, &info64); 1693 if (!err) 1694 err = loop_info64_to_old(&info64, &info); 1695 if (!err && copy_to_user(arg, &info, sizeof(info))) 1696 err = -EFAULT; 1697 1698 return err; 1699 } 1700 1701 static int 1702 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1703 struct loop_info64 info64; 1704 int err; 1705 1706 if (!arg) 1707 return -EINVAL; 1708 err = loop_get_status(lo, &info64); 1709 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1710 err = -EFAULT; 1711 1712 return err; 1713 } 1714 1715 static int loop_set_capacity(struct loop_device *lo) 1716 { 1717 loff_t size; 1718 1719 if (unlikely(lo->lo_state != Lo_bound)) 1720 return -ENXIO; 1721 1722 size = get_loop_size(lo, lo->lo_backing_file); 1723 loop_set_size(lo, size); 1724 1725 return 0; 1726 } 1727 1728 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1729 { 1730 int error = -ENXIO; 1731 if (lo->lo_state != Lo_bound) 1732 goto out; 1733 1734 __loop_update_dio(lo, !!arg); 1735 if (lo->use_dio == !!arg) 1736 return 0; 1737 error = -EINVAL; 1738 out: 1739 return error; 1740 } 1741 1742 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1743 { 1744 int err = 0; 1745 1746 if (lo->lo_state != Lo_bound) 1747 return -ENXIO; 1748 1749 err = blk_validate_block_size(arg); 1750 if (err) 1751 return err; 1752 1753 if (lo->lo_queue->limits.logical_block_size == arg) 1754 return 0; 1755 1756 sync_blockdev(lo->lo_device); 1757 invalidate_bdev(lo->lo_device); 1758 1759 blk_mq_freeze_queue(lo->lo_queue); 1760 1761 /* invalidate_bdev should have truncated all the pages */ 1762 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1763 err = -EAGAIN; 1764 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1765 __func__, lo->lo_number, lo->lo_file_name, 1766 lo->lo_device->bd_inode->i_mapping->nrpages); 1767 goto out_unfreeze; 1768 } 1769 1770 blk_queue_logical_block_size(lo->lo_queue, arg); 1771 blk_queue_physical_block_size(lo->lo_queue, arg); 1772 blk_queue_io_min(lo->lo_queue, arg); 1773 loop_update_dio(lo); 1774 out_unfreeze: 1775 blk_mq_unfreeze_queue(lo->lo_queue); 1776 1777 return err; 1778 } 1779 1780 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1781 unsigned long arg) 1782 { 1783 int err; 1784 1785 err = mutex_lock_killable(&lo->lo_mutex); 1786 if (err) 1787 return err; 1788 switch (cmd) { 1789 case LOOP_SET_CAPACITY: 1790 err = loop_set_capacity(lo); 1791 break; 1792 case LOOP_SET_DIRECT_IO: 1793 err = loop_set_dio(lo, arg); 1794 break; 1795 case LOOP_SET_BLOCK_SIZE: 1796 err = loop_set_block_size(lo, arg); 1797 break; 1798 default: 1799 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1800 } 1801 mutex_unlock(&lo->lo_mutex); 1802 return err; 1803 } 1804 1805 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1806 unsigned int cmd, unsigned long arg) 1807 { 1808 struct loop_device *lo = bdev->bd_disk->private_data; 1809 void __user *argp = (void __user *) arg; 1810 int err; 1811 1812 switch (cmd) { 1813 case LOOP_SET_FD: { 1814 /* 1815 * Legacy case - pass in a zeroed out struct loop_config with 1816 * only the file descriptor set , which corresponds with the 1817 * default parameters we'd have used otherwise. 1818 */ 1819 struct loop_config config; 1820 1821 memset(&config, 0, sizeof(config)); 1822 config.fd = arg; 1823 1824 return loop_configure(lo, mode, bdev, &config); 1825 } 1826 case LOOP_CONFIGURE: { 1827 struct loop_config config; 1828 1829 if (copy_from_user(&config, argp, sizeof(config))) 1830 return -EFAULT; 1831 1832 return loop_configure(lo, mode, bdev, &config); 1833 } 1834 case LOOP_CHANGE_FD: 1835 return loop_change_fd(lo, bdev, arg); 1836 case LOOP_CLR_FD: 1837 return loop_clr_fd(lo); 1838 case LOOP_SET_STATUS: 1839 err = -EPERM; 1840 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1841 err = loop_set_status_old(lo, argp); 1842 } 1843 break; 1844 case LOOP_GET_STATUS: 1845 return loop_get_status_old(lo, argp); 1846 case LOOP_SET_STATUS64: 1847 err = -EPERM; 1848 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1849 err = loop_set_status64(lo, argp); 1850 } 1851 break; 1852 case LOOP_GET_STATUS64: 1853 return loop_get_status64(lo, argp); 1854 case LOOP_SET_CAPACITY: 1855 case LOOP_SET_DIRECT_IO: 1856 case LOOP_SET_BLOCK_SIZE: 1857 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1858 return -EPERM; 1859 fallthrough; 1860 default: 1861 err = lo_simple_ioctl(lo, cmd, arg); 1862 break; 1863 } 1864 1865 return err; 1866 } 1867 1868 #ifdef CONFIG_COMPAT 1869 struct compat_loop_info { 1870 compat_int_t lo_number; /* ioctl r/o */ 1871 compat_dev_t lo_device; /* ioctl r/o */ 1872 compat_ulong_t lo_inode; /* ioctl r/o */ 1873 compat_dev_t lo_rdevice; /* ioctl r/o */ 1874 compat_int_t lo_offset; 1875 compat_int_t lo_encrypt_type; 1876 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1877 compat_int_t lo_flags; /* ioctl r/o */ 1878 char lo_name[LO_NAME_SIZE]; 1879 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1880 compat_ulong_t lo_init[2]; 1881 char reserved[4]; 1882 }; 1883 1884 /* 1885 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1886 * - noinlined to reduce stack space usage in main part of driver 1887 */ 1888 static noinline int 1889 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1890 struct loop_info64 *info64) 1891 { 1892 struct compat_loop_info info; 1893 1894 if (copy_from_user(&info, arg, sizeof(info))) 1895 return -EFAULT; 1896 1897 memset(info64, 0, sizeof(*info64)); 1898 info64->lo_number = info.lo_number; 1899 info64->lo_device = info.lo_device; 1900 info64->lo_inode = info.lo_inode; 1901 info64->lo_rdevice = info.lo_rdevice; 1902 info64->lo_offset = info.lo_offset; 1903 info64->lo_sizelimit = 0; 1904 info64->lo_encrypt_type = info.lo_encrypt_type; 1905 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1906 info64->lo_flags = info.lo_flags; 1907 info64->lo_init[0] = info.lo_init[0]; 1908 info64->lo_init[1] = info.lo_init[1]; 1909 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1910 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1911 else 1912 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1913 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1914 return 0; 1915 } 1916 1917 /* 1918 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1919 * - noinlined to reduce stack space usage in main part of driver 1920 */ 1921 static noinline int 1922 loop_info64_to_compat(const struct loop_info64 *info64, 1923 struct compat_loop_info __user *arg) 1924 { 1925 struct compat_loop_info info; 1926 1927 memset(&info, 0, sizeof(info)); 1928 info.lo_number = info64->lo_number; 1929 info.lo_device = info64->lo_device; 1930 info.lo_inode = info64->lo_inode; 1931 info.lo_rdevice = info64->lo_rdevice; 1932 info.lo_offset = info64->lo_offset; 1933 info.lo_encrypt_type = info64->lo_encrypt_type; 1934 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1935 info.lo_flags = info64->lo_flags; 1936 info.lo_init[0] = info64->lo_init[0]; 1937 info.lo_init[1] = info64->lo_init[1]; 1938 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1939 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1940 else 1941 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1942 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1943 1944 /* error in case values were truncated */ 1945 if (info.lo_device != info64->lo_device || 1946 info.lo_rdevice != info64->lo_rdevice || 1947 info.lo_inode != info64->lo_inode || 1948 info.lo_offset != info64->lo_offset || 1949 info.lo_init[0] != info64->lo_init[0] || 1950 info.lo_init[1] != info64->lo_init[1]) 1951 return -EOVERFLOW; 1952 1953 if (copy_to_user(arg, &info, sizeof(info))) 1954 return -EFAULT; 1955 return 0; 1956 } 1957 1958 static int 1959 loop_set_status_compat(struct loop_device *lo, 1960 const struct compat_loop_info __user *arg) 1961 { 1962 struct loop_info64 info64; 1963 int ret; 1964 1965 ret = loop_info64_from_compat(arg, &info64); 1966 if (ret < 0) 1967 return ret; 1968 return loop_set_status(lo, &info64); 1969 } 1970 1971 static int 1972 loop_get_status_compat(struct loop_device *lo, 1973 struct compat_loop_info __user *arg) 1974 { 1975 struct loop_info64 info64; 1976 int err; 1977 1978 if (!arg) 1979 return -EINVAL; 1980 err = loop_get_status(lo, &info64); 1981 if (!err) 1982 err = loop_info64_to_compat(&info64, arg); 1983 return err; 1984 } 1985 1986 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1987 unsigned int cmd, unsigned long arg) 1988 { 1989 struct loop_device *lo = bdev->bd_disk->private_data; 1990 int err; 1991 1992 switch(cmd) { 1993 case LOOP_SET_STATUS: 1994 err = loop_set_status_compat(lo, 1995 (const struct compat_loop_info __user *)arg); 1996 break; 1997 case LOOP_GET_STATUS: 1998 err = loop_get_status_compat(lo, 1999 (struct compat_loop_info __user *)arg); 2000 break; 2001 case LOOP_SET_CAPACITY: 2002 case LOOP_CLR_FD: 2003 case LOOP_GET_STATUS64: 2004 case LOOP_SET_STATUS64: 2005 case LOOP_CONFIGURE: 2006 arg = (unsigned long) compat_ptr(arg); 2007 fallthrough; 2008 case LOOP_SET_FD: 2009 case LOOP_CHANGE_FD: 2010 case LOOP_SET_BLOCK_SIZE: 2011 case LOOP_SET_DIRECT_IO: 2012 err = lo_ioctl(bdev, mode, cmd, arg); 2013 break; 2014 default: 2015 err = -ENOIOCTLCMD; 2016 break; 2017 } 2018 return err; 2019 } 2020 #endif 2021 2022 static int lo_open(struct block_device *bdev, fmode_t mode) 2023 { 2024 struct loop_device *lo = bdev->bd_disk->private_data; 2025 int err; 2026 2027 err = mutex_lock_killable(&lo->lo_mutex); 2028 if (err) 2029 return err; 2030 if (lo->lo_state == Lo_deleting) 2031 err = -ENXIO; 2032 else 2033 atomic_inc(&lo->lo_refcnt); 2034 mutex_unlock(&lo->lo_mutex); 2035 return err; 2036 } 2037 2038 static void lo_release(struct gendisk *disk, fmode_t mode) 2039 { 2040 struct loop_device *lo = disk->private_data; 2041 2042 mutex_lock(&lo->lo_mutex); 2043 if (atomic_dec_return(&lo->lo_refcnt)) 2044 goto out_unlock; 2045 2046 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 2047 if (lo->lo_state != Lo_bound) 2048 goto out_unlock; 2049 lo->lo_state = Lo_rundown; 2050 mutex_unlock(&lo->lo_mutex); 2051 /* 2052 * In autoclear mode, stop the loop thread 2053 * and remove configuration after last close. 2054 */ 2055 __loop_clr_fd(lo, true); 2056 return; 2057 } else if (lo->lo_state == Lo_bound) { 2058 /* 2059 * Otherwise keep thread (if running) and config, 2060 * but flush possible ongoing bios in thread. 2061 */ 2062 blk_mq_freeze_queue(lo->lo_queue); 2063 blk_mq_unfreeze_queue(lo->lo_queue); 2064 } 2065 2066 out_unlock: 2067 mutex_unlock(&lo->lo_mutex); 2068 } 2069 2070 static const struct block_device_operations lo_fops = { 2071 .owner = THIS_MODULE, 2072 .open = lo_open, 2073 .release = lo_release, 2074 .ioctl = lo_ioctl, 2075 #ifdef CONFIG_COMPAT 2076 .compat_ioctl = lo_compat_ioctl, 2077 #endif 2078 }; 2079 2080 /* 2081 * And now the modules code and kernel interface. 2082 */ 2083 static int max_loop; 2084 module_param(max_loop, int, 0444); 2085 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 2086 module_param(max_part, int, 0444); 2087 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 2088 MODULE_LICENSE("GPL"); 2089 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 2090 2091 int loop_register_transfer(struct loop_func_table *funcs) 2092 { 2093 unsigned int n = funcs->number; 2094 2095 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 2096 return -EINVAL; 2097 xfer_funcs[n] = funcs; 2098 return 0; 2099 } 2100 2101 int loop_unregister_transfer(int number) 2102 { 2103 unsigned int n = number; 2104 struct loop_func_table *xfer; 2105 2106 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 2107 return -EINVAL; 2108 /* 2109 * This function is called from only cleanup_cryptoloop(). 2110 * Given that each loop device that has a transfer enabled holds a 2111 * reference to the module implementing it we should never get here 2112 * with a transfer that is set (unless forced module unloading is 2113 * requested). Thus, check module's refcount and warn if this is 2114 * not a clean unloading. 2115 */ 2116 #ifdef CONFIG_MODULE_UNLOAD 2117 if (xfer->owner && module_refcount(xfer->owner) != -1) 2118 pr_err("Danger! Unregistering an in use transfer function.\n"); 2119 #endif 2120 2121 xfer_funcs[n] = NULL; 2122 return 0; 2123 } 2124 2125 EXPORT_SYMBOL(loop_register_transfer); 2126 EXPORT_SYMBOL(loop_unregister_transfer); 2127 2128 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 2129 const struct blk_mq_queue_data *bd) 2130 { 2131 struct request *rq = bd->rq; 2132 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2133 struct loop_device *lo = rq->q->queuedata; 2134 2135 blk_mq_start_request(rq); 2136 2137 if (lo->lo_state != Lo_bound) 2138 return BLK_STS_IOERR; 2139 2140 switch (req_op(rq)) { 2141 case REQ_OP_FLUSH: 2142 case REQ_OP_DISCARD: 2143 case REQ_OP_WRITE_ZEROES: 2144 cmd->use_aio = false; 2145 break; 2146 default: 2147 cmd->use_aio = lo->use_dio; 2148 break; 2149 } 2150 2151 /* always use the first bio's css */ 2152 cmd->blkcg_css = NULL; 2153 cmd->memcg_css = NULL; 2154 #ifdef CONFIG_BLK_CGROUP 2155 if (rq->bio && rq->bio->bi_blkg) { 2156 cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 2157 #ifdef CONFIG_MEMCG 2158 cmd->memcg_css = 2159 cgroup_get_e_css(cmd->blkcg_css->cgroup, 2160 &memory_cgrp_subsys); 2161 #endif 2162 } 2163 #endif 2164 loop_queue_work(lo, cmd); 2165 2166 return BLK_STS_OK; 2167 } 2168 2169 static void loop_handle_cmd(struct loop_cmd *cmd) 2170 { 2171 struct request *rq = blk_mq_rq_from_pdu(cmd); 2172 const bool write = op_is_write(req_op(rq)); 2173 struct loop_device *lo = rq->q->queuedata; 2174 int ret = 0; 2175 struct mem_cgroup *old_memcg = NULL; 2176 2177 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 2178 ret = -EIO; 2179 goto failed; 2180 } 2181 2182 if (cmd->blkcg_css) 2183 kthread_associate_blkcg(cmd->blkcg_css); 2184 if (cmd->memcg_css) 2185 old_memcg = set_active_memcg( 2186 mem_cgroup_from_css(cmd->memcg_css)); 2187 2188 ret = do_req_filebacked(lo, rq); 2189 2190 if (cmd->blkcg_css) 2191 kthread_associate_blkcg(NULL); 2192 2193 if (cmd->memcg_css) { 2194 set_active_memcg(old_memcg); 2195 css_put(cmd->memcg_css); 2196 } 2197 failed: 2198 /* complete non-aio request */ 2199 if (!cmd->use_aio || ret) { 2200 if (ret == -EOPNOTSUPP) 2201 cmd->ret = ret; 2202 else 2203 cmd->ret = ret ? -EIO : 0; 2204 if (likely(!blk_should_fake_timeout(rq->q))) 2205 blk_mq_complete_request(rq); 2206 } 2207 } 2208 2209 static void loop_set_timer(struct loop_device *lo) 2210 { 2211 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 2212 } 2213 2214 static void loop_process_work(struct loop_worker *worker, 2215 struct list_head *cmd_list, struct loop_device *lo) 2216 { 2217 int orig_flags = current->flags; 2218 struct loop_cmd *cmd; 2219 2220 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 2221 spin_lock_irq(&lo->lo_work_lock); 2222 while (!list_empty(cmd_list)) { 2223 cmd = container_of( 2224 cmd_list->next, struct loop_cmd, list_entry); 2225 list_del(cmd_list->next); 2226 spin_unlock_irq(&lo->lo_work_lock); 2227 2228 loop_handle_cmd(cmd); 2229 cond_resched(); 2230 2231 spin_lock_irq(&lo->lo_work_lock); 2232 } 2233 2234 /* 2235 * We only add to the idle list if there are no pending cmds 2236 * *and* the worker will not run again which ensures that it 2237 * is safe to free any worker on the idle list 2238 */ 2239 if (worker && !work_pending(&worker->work)) { 2240 worker->last_ran_at = jiffies; 2241 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 2242 loop_set_timer(lo); 2243 } 2244 spin_unlock_irq(&lo->lo_work_lock); 2245 current->flags = orig_flags; 2246 } 2247 2248 static void loop_workfn(struct work_struct *work) 2249 { 2250 struct loop_worker *worker = 2251 container_of(work, struct loop_worker, work); 2252 loop_process_work(worker, &worker->cmd_list, worker->lo); 2253 } 2254 2255 static void loop_rootcg_workfn(struct work_struct *work) 2256 { 2257 struct loop_device *lo = 2258 container_of(work, struct loop_device, rootcg_work); 2259 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 2260 } 2261 2262 static void loop_free_idle_workers(struct timer_list *timer) 2263 { 2264 struct loop_device *lo = container_of(timer, struct loop_device, timer); 2265 struct loop_worker *pos, *worker; 2266 2267 spin_lock_irq(&lo->lo_work_lock); 2268 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 2269 idle_list) { 2270 if (time_is_after_jiffies(worker->last_ran_at + 2271 LOOP_IDLE_WORKER_TIMEOUT)) 2272 break; 2273 list_del(&worker->idle_list); 2274 rb_erase(&worker->rb_node, &lo->worker_tree); 2275 css_put(worker->blkcg_css); 2276 kfree(worker); 2277 } 2278 if (!list_empty(&lo->idle_worker_list)) 2279 loop_set_timer(lo); 2280 spin_unlock_irq(&lo->lo_work_lock); 2281 } 2282 2283 static const struct blk_mq_ops loop_mq_ops = { 2284 .queue_rq = loop_queue_rq, 2285 .complete = lo_complete_rq, 2286 }; 2287 2288 static int loop_add(int i) 2289 { 2290 struct loop_device *lo; 2291 struct gendisk *disk; 2292 int err; 2293 2294 err = -ENOMEM; 2295 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 2296 if (!lo) 2297 goto out; 2298 lo->lo_state = Lo_unbound; 2299 2300 err = mutex_lock_killable(&loop_ctl_mutex); 2301 if (err) 2302 goto out_free_dev; 2303 2304 /* allocate id, if @id >= 0, we're requesting that specific id */ 2305 if (i >= 0) { 2306 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2307 if (err == -ENOSPC) 2308 err = -EEXIST; 2309 } else { 2310 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2311 } 2312 mutex_unlock(&loop_ctl_mutex); 2313 if (err < 0) 2314 goto out_free_dev; 2315 i = err; 2316 2317 err = -ENOMEM; 2318 lo->tag_set.ops = &loop_mq_ops; 2319 lo->tag_set.nr_hw_queues = 1; 2320 lo->tag_set.queue_depth = 128; 2321 lo->tag_set.numa_node = NUMA_NO_NODE; 2322 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2323 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | 2324 BLK_MQ_F_NO_SCHED_BY_DEFAULT; 2325 lo->tag_set.driver_data = lo; 2326 2327 err = blk_mq_alloc_tag_set(&lo->tag_set); 2328 if (err) 2329 goto out_free_idr; 2330 2331 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 2332 if (IS_ERR(disk)) { 2333 err = PTR_ERR(disk); 2334 goto out_cleanup_tags; 2335 } 2336 lo->lo_queue = lo->lo_disk->queue; 2337 2338 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2339 2340 /* 2341 * By default, we do buffer IO, so it doesn't make sense to enable 2342 * merge because the I/O submitted to backing file is handled page by 2343 * page. For directio mode, merge does help to dispatch bigger request 2344 * to underlayer disk. We will enable merge once directio is enabled. 2345 */ 2346 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2347 2348 /* 2349 * Disable partition scanning by default. The in-kernel partition 2350 * scanning can be requested individually per-device during its 2351 * setup. Userspace can always add and remove partitions from all 2352 * devices. The needed partition minors are allocated from the 2353 * extended minor space, the main loop device numbers will continue 2354 * to match the loop minors, regardless of the number of partitions 2355 * used. 2356 * 2357 * If max_part is given, partition scanning is globally enabled for 2358 * all loop devices. The minors for the main loop devices will be 2359 * multiples of max_part. 2360 * 2361 * Note: Global-for-all-devices, set-only-at-init, read-only module 2362 * parameteters like 'max_loop' and 'max_part' make things needlessly 2363 * complicated, are too static, inflexible and may surprise 2364 * userspace tools. Parameters like this in general should be avoided. 2365 */ 2366 if (!part_shift) 2367 disk->flags |= GENHD_FL_NO_PART_SCAN; 2368 disk->flags |= GENHD_FL_EXT_DEVT; 2369 atomic_set(&lo->lo_refcnt, 0); 2370 mutex_init(&lo->lo_mutex); 2371 lo->lo_number = i; 2372 spin_lock_init(&lo->lo_lock); 2373 spin_lock_init(&lo->lo_work_lock); 2374 disk->major = LOOP_MAJOR; 2375 disk->first_minor = i << part_shift; 2376 disk->minors = 1 << part_shift; 2377 disk->fops = &lo_fops; 2378 disk->private_data = lo; 2379 disk->queue = lo->lo_queue; 2380 disk->events = DISK_EVENT_MEDIA_CHANGE; 2381 disk->event_flags = DISK_EVENT_FLAG_UEVENT; 2382 sprintf(disk->disk_name, "loop%d", i); 2383 /* Make this loop device reachable from pathname. */ 2384 add_disk(disk); 2385 /* Show this loop device. */ 2386 mutex_lock(&loop_ctl_mutex); 2387 lo->idr_visible = true; 2388 mutex_unlock(&loop_ctl_mutex); 2389 return i; 2390 2391 out_cleanup_tags: 2392 blk_mq_free_tag_set(&lo->tag_set); 2393 out_free_idr: 2394 mutex_lock(&loop_ctl_mutex); 2395 idr_remove(&loop_index_idr, i); 2396 mutex_unlock(&loop_ctl_mutex); 2397 out_free_dev: 2398 kfree(lo); 2399 out: 2400 return err; 2401 } 2402 2403 static void loop_remove(struct loop_device *lo) 2404 { 2405 /* Make this loop device unreachable from pathname. */ 2406 del_gendisk(lo->lo_disk); 2407 blk_cleanup_disk(lo->lo_disk); 2408 blk_mq_free_tag_set(&lo->tag_set); 2409 mutex_lock(&loop_ctl_mutex); 2410 idr_remove(&loop_index_idr, lo->lo_number); 2411 mutex_unlock(&loop_ctl_mutex); 2412 /* There is no route which can find this loop device. */ 2413 mutex_destroy(&lo->lo_mutex); 2414 kfree(lo); 2415 } 2416 2417 static void loop_probe(dev_t dev) 2418 { 2419 int idx = MINOR(dev) >> part_shift; 2420 2421 if (max_loop && idx >= max_loop) 2422 return; 2423 loop_add(idx); 2424 } 2425 2426 static int loop_control_remove(int idx) 2427 { 2428 struct loop_device *lo; 2429 int ret; 2430 2431 if (idx < 0) { 2432 pr_warn_once("deleting an unspecified loop device is not supported.\n"); 2433 return -EINVAL; 2434 } 2435 2436 /* Hide this loop device for serialization. */ 2437 ret = mutex_lock_killable(&loop_ctl_mutex); 2438 if (ret) 2439 return ret; 2440 lo = idr_find(&loop_index_idr, idx); 2441 if (!lo || !lo->idr_visible) 2442 ret = -ENODEV; 2443 else 2444 lo->idr_visible = false; 2445 mutex_unlock(&loop_ctl_mutex); 2446 if (ret) 2447 return ret; 2448 2449 /* Check whether this loop device can be removed. */ 2450 ret = mutex_lock_killable(&lo->lo_mutex); 2451 if (ret) 2452 goto mark_visible; 2453 if (lo->lo_state != Lo_unbound || 2454 atomic_read(&lo->lo_refcnt) > 0) { 2455 mutex_unlock(&lo->lo_mutex); 2456 ret = -EBUSY; 2457 goto mark_visible; 2458 } 2459 /* Mark this loop device no longer open()-able. */ 2460 lo->lo_state = Lo_deleting; 2461 mutex_unlock(&lo->lo_mutex); 2462 2463 loop_remove(lo); 2464 return 0; 2465 2466 mark_visible: 2467 /* Show this loop device again. */ 2468 mutex_lock(&loop_ctl_mutex); 2469 lo->idr_visible = true; 2470 mutex_unlock(&loop_ctl_mutex); 2471 return ret; 2472 } 2473 2474 static int loop_control_get_free(int idx) 2475 { 2476 struct loop_device *lo; 2477 int id, ret; 2478 2479 ret = mutex_lock_killable(&loop_ctl_mutex); 2480 if (ret) 2481 return ret; 2482 idr_for_each_entry(&loop_index_idr, lo, id) { 2483 /* Hitting a race results in creating a new loop device which is harmless. */ 2484 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) 2485 goto found; 2486 } 2487 mutex_unlock(&loop_ctl_mutex); 2488 return loop_add(-1); 2489 found: 2490 mutex_unlock(&loop_ctl_mutex); 2491 return id; 2492 } 2493 2494 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2495 unsigned long parm) 2496 { 2497 switch (cmd) { 2498 case LOOP_CTL_ADD: 2499 return loop_add(parm); 2500 case LOOP_CTL_REMOVE: 2501 return loop_control_remove(parm); 2502 case LOOP_CTL_GET_FREE: 2503 return loop_control_get_free(parm); 2504 default: 2505 return -ENOSYS; 2506 } 2507 } 2508 2509 static const struct file_operations loop_ctl_fops = { 2510 .open = nonseekable_open, 2511 .unlocked_ioctl = loop_control_ioctl, 2512 .compat_ioctl = loop_control_ioctl, 2513 .owner = THIS_MODULE, 2514 .llseek = noop_llseek, 2515 }; 2516 2517 static struct miscdevice loop_misc = { 2518 .minor = LOOP_CTRL_MINOR, 2519 .name = "loop-control", 2520 .fops = &loop_ctl_fops, 2521 }; 2522 2523 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2524 MODULE_ALIAS("devname:loop-control"); 2525 2526 static int __init loop_init(void) 2527 { 2528 int i, nr; 2529 int err; 2530 2531 part_shift = 0; 2532 if (max_part > 0) { 2533 part_shift = fls(max_part); 2534 2535 /* 2536 * Adjust max_part according to part_shift as it is exported 2537 * to user space so that user can decide correct minor number 2538 * if [s]he want to create more devices. 2539 * 2540 * Note that -1 is required because partition 0 is reserved 2541 * for the whole disk. 2542 */ 2543 max_part = (1UL << part_shift) - 1; 2544 } 2545 2546 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2547 err = -EINVAL; 2548 goto err_out; 2549 } 2550 2551 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2552 err = -EINVAL; 2553 goto err_out; 2554 } 2555 2556 /* 2557 * If max_loop is specified, create that many devices upfront. 2558 * This also becomes a hard limit. If max_loop is not specified, 2559 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2560 * init time. Loop devices can be requested on-demand with the 2561 * /dev/loop-control interface, or be instantiated by accessing 2562 * a 'dead' device node. 2563 */ 2564 if (max_loop) 2565 nr = max_loop; 2566 else 2567 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2568 2569 err = misc_register(&loop_misc); 2570 if (err < 0) 2571 goto err_out; 2572 2573 2574 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2575 err = -EIO; 2576 goto misc_out; 2577 } 2578 2579 /* pre-create number of devices given by config or max_loop */ 2580 for (i = 0; i < nr; i++) 2581 loop_add(i); 2582 2583 printk(KERN_INFO "loop: module loaded\n"); 2584 return 0; 2585 2586 misc_out: 2587 misc_deregister(&loop_misc); 2588 err_out: 2589 return err; 2590 } 2591 2592 static void __exit loop_exit(void) 2593 { 2594 struct loop_device *lo; 2595 int id; 2596 2597 unregister_blkdev(LOOP_MAJOR, "loop"); 2598 misc_deregister(&loop_misc); 2599 2600 /* 2601 * There is no need to use loop_ctl_mutex here, for nobody else can 2602 * access loop_index_idr when this module is unloading (unless forced 2603 * module unloading is requested). If this is not a clean unloading, 2604 * we have no means to avoid kernel crash. 2605 */ 2606 idr_for_each_entry(&loop_index_idr, lo, id) 2607 loop_remove(lo); 2608 2609 idr_destroy(&loop_index_idr); 2610 } 2611 2612 module_init(loop_init); 2613 module_exit(loop_exit); 2614 2615 #ifndef MODULE 2616 static int __init max_loop_setup(char *str) 2617 { 2618 max_loop = simple_strtol(str, NULL, 0); 2619 return 1; 2620 } 2621 2622 __setup("max_loop=", max_loop_setup); 2623 #endif 2624