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 #include <linux/statfs.h> 83 84 #include "loop.h" 85 86 #include <linux/uaccess.h> 87 88 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) 89 90 static DEFINE_IDR(loop_index_idr); 91 static DEFINE_MUTEX(loop_ctl_mutex); 92 static DEFINE_MUTEX(loop_validate_mutex); 93 94 /** 95 * loop_global_lock_killable() - take locks for safe loop_validate_file() test 96 * 97 * @lo: struct loop_device 98 * @global: true if @lo is about to bind another "struct loop_device", false otherwise 99 * 100 * Returns 0 on success, -EINTR otherwise. 101 * 102 * Since loop_validate_file() traverses on other "struct loop_device" if 103 * is_loop_device() is true, we need a global lock for serializing concurrent 104 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls. 105 */ 106 static int loop_global_lock_killable(struct loop_device *lo, bool global) 107 { 108 int err; 109 110 if (global) { 111 err = mutex_lock_killable(&loop_validate_mutex); 112 if (err) 113 return err; 114 } 115 err = mutex_lock_killable(&lo->lo_mutex); 116 if (err && global) 117 mutex_unlock(&loop_validate_mutex); 118 return err; 119 } 120 121 /** 122 * loop_global_unlock() - release locks taken by loop_global_lock_killable() 123 * 124 * @lo: struct loop_device 125 * @global: true if @lo was about to bind another "struct loop_device", false otherwise 126 */ 127 static void loop_global_unlock(struct loop_device *lo, bool global) 128 { 129 mutex_unlock(&lo->lo_mutex); 130 if (global) 131 mutex_unlock(&loop_validate_mutex); 132 } 133 134 static int max_part; 135 static int part_shift; 136 137 static int transfer_xor(struct loop_device *lo, int cmd, 138 struct page *raw_page, unsigned raw_off, 139 struct page *loop_page, unsigned loop_off, 140 int size, sector_t real_block) 141 { 142 char *raw_buf = kmap_atomic(raw_page) + raw_off; 143 char *loop_buf = kmap_atomic(loop_page) + loop_off; 144 char *in, *out, *key; 145 int i, keysize; 146 147 if (cmd == READ) { 148 in = raw_buf; 149 out = loop_buf; 150 } else { 151 in = loop_buf; 152 out = raw_buf; 153 } 154 155 key = lo->lo_encrypt_key; 156 keysize = lo->lo_encrypt_key_size; 157 for (i = 0; i < size; i++) 158 *out++ = *in++ ^ key[(i & 511) % keysize]; 159 160 kunmap_atomic(loop_buf); 161 kunmap_atomic(raw_buf); 162 cond_resched(); 163 return 0; 164 } 165 166 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 167 { 168 if (unlikely(info->lo_encrypt_key_size <= 0)) 169 return -EINVAL; 170 return 0; 171 } 172 173 static struct loop_func_table none_funcs = { 174 .number = LO_CRYPT_NONE, 175 }; 176 177 static struct loop_func_table xor_funcs = { 178 .number = LO_CRYPT_XOR, 179 .transfer = transfer_xor, 180 .init = xor_init 181 }; 182 183 /* xfer_funcs[0] is special - its release function is never called */ 184 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 185 &none_funcs, 186 &xor_funcs 187 }; 188 189 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 190 { 191 loff_t loopsize; 192 193 /* Compute loopsize in bytes */ 194 loopsize = i_size_read(file->f_mapping->host); 195 if (offset > 0) 196 loopsize -= offset; 197 /* offset is beyond i_size, weird but possible */ 198 if (loopsize < 0) 199 return 0; 200 201 if (sizelimit > 0 && sizelimit < loopsize) 202 loopsize = sizelimit; 203 /* 204 * Unfortunately, if we want to do I/O on the device, 205 * the number of 512-byte sectors has to fit into a sector_t. 206 */ 207 return loopsize >> 9; 208 } 209 210 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 211 { 212 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 213 } 214 215 static void __loop_update_dio(struct loop_device *lo, bool dio) 216 { 217 struct file *file = lo->lo_backing_file; 218 struct address_space *mapping = file->f_mapping; 219 struct inode *inode = mapping->host; 220 unsigned short sb_bsize = 0; 221 unsigned dio_align = 0; 222 bool use_dio; 223 224 if (inode->i_sb->s_bdev) { 225 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 226 dio_align = sb_bsize - 1; 227 } 228 229 /* 230 * We support direct I/O only if lo_offset is aligned with the 231 * logical I/O size of backing device, and the logical block 232 * size of loop is bigger than the backing device's and the loop 233 * needn't transform transfer. 234 * 235 * TODO: the above condition may be loosed in the future, and 236 * direct I/O may be switched runtime at that time because most 237 * of requests in sane applications should be PAGE_SIZE aligned 238 */ 239 if (dio) { 240 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 241 !(lo->lo_offset & dio_align) && 242 mapping->a_ops->direct_IO && 243 !lo->transfer) 244 use_dio = true; 245 else 246 use_dio = false; 247 } else { 248 use_dio = false; 249 } 250 251 if (lo->use_dio == use_dio) 252 return; 253 254 /* flush dirty pages before changing direct IO */ 255 vfs_fsync(file, 0); 256 257 /* 258 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 259 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 260 * will get updated by ioctl(LOOP_GET_STATUS) 261 */ 262 if (lo->lo_state == Lo_bound) 263 blk_mq_freeze_queue(lo->lo_queue); 264 lo->use_dio = use_dio; 265 if (use_dio) { 266 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 267 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 268 } else { 269 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 270 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 271 } 272 if (lo->lo_state == Lo_bound) 273 blk_mq_unfreeze_queue(lo->lo_queue); 274 } 275 276 /** 277 * loop_set_size() - sets device size and notifies userspace 278 * @lo: struct loop_device to set the size for 279 * @size: new size of the loop device 280 * 281 * Callers must validate that the size passed into this function fits into 282 * a sector_t, eg using loop_validate_size() 283 */ 284 static void loop_set_size(struct loop_device *lo, loff_t size) 285 { 286 if (!set_capacity_and_notify(lo->lo_disk, size)) 287 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 288 } 289 290 static inline int 291 lo_do_transfer(struct loop_device *lo, int cmd, 292 struct page *rpage, unsigned roffs, 293 struct page *lpage, unsigned loffs, 294 int size, sector_t rblock) 295 { 296 int ret; 297 298 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 299 if (likely(!ret)) 300 return 0; 301 302 printk_ratelimited(KERN_ERR 303 "loop: Transfer error at byte offset %llu, length %i.\n", 304 (unsigned long long)rblock << 9, size); 305 return ret; 306 } 307 308 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 309 { 310 struct iov_iter i; 311 ssize_t bw; 312 313 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 314 315 file_start_write(file); 316 bw = vfs_iter_write(file, &i, ppos, 0); 317 file_end_write(file); 318 319 if (likely(bw == bvec->bv_len)) 320 return 0; 321 322 printk_ratelimited(KERN_ERR 323 "loop: Write error at byte offset %llu, length %i.\n", 324 (unsigned long long)*ppos, bvec->bv_len); 325 if (bw >= 0) 326 bw = -EIO; 327 return bw; 328 } 329 330 static int lo_write_simple(struct loop_device *lo, struct request *rq, 331 loff_t pos) 332 { 333 struct bio_vec bvec; 334 struct req_iterator iter; 335 int ret = 0; 336 337 rq_for_each_segment(bvec, rq, iter) { 338 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 339 if (ret < 0) 340 break; 341 cond_resched(); 342 } 343 344 return ret; 345 } 346 347 /* 348 * This is the slow, transforming version that needs to double buffer the 349 * data as it cannot do the transformations in place without having direct 350 * access to the destination pages of the backing file. 351 */ 352 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 353 loff_t pos) 354 { 355 struct bio_vec bvec, b; 356 struct req_iterator iter; 357 struct page *page; 358 int ret = 0; 359 360 page = alloc_page(GFP_NOIO); 361 if (unlikely(!page)) 362 return -ENOMEM; 363 364 rq_for_each_segment(bvec, rq, iter) { 365 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 366 bvec.bv_offset, bvec.bv_len, pos >> 9); 367 if (unlikely(ret)) 368 break; 369 370 b.bv_page = page; 371 b.bv_offset = 0; 372 b.bv_len = bvec.bv_len; 373 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 374 if (ret < 0) 375 break; 376 } 377 378 __free_page(page); 379 return ret; 380 } 381 382 static int lo_read_simple(struct loop_device *lo, struct request *rq, 383 loff_t pos) 384 { 385 struct bio_vec bvec; 386 struct req_iterator iter; 387 struct iov_iter i; 388 ssize_t len; 389 390 rq_for_each_segment(bvec, rq, iter) { 391 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 392 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 393 if (len < 0) 394 return len; 395 396 flush_dcache_page(bvec.bv_page); 397 398 if (len != bvec.bv_len) { 399 struct bio *bio; 400 401 __rq_for_each_bio(bio, rq) 402 zero_fill_bio(bio); 403 break; 404 } 405 cond_resched(); 406 } 407 408 return 0; 409 } 410 411 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 412 loff_t pos) 413 { 414 struct bio_vec bvec, b; 415 struct req_iterator iter; 416 struct iov_iter i; 417 struct page *page; 418 ssize_t len; 419 int ret = 0; 420 421 page = alloc_page(GFP_NOIO); 422 if (unlikely(!page)) 423 return -ENOMEM; 424 425 rq_for_each_segment(bvec, rq, iter) { 426 loff_t offset = pos; 427 428 b.bv_page = page; 429 b.bv_offset = 0; 430 b.bv_len = bvec.bv_len; 431 432 iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 433 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 434 if (len < 0) { 435 ret = len; 436 goto out_free_page; 437 } 438 439 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 440 bvec.bv_offset, len, offset >> 9); 441 if (ret) 442 goto out_free_page; 443 444 flush_dcache_page(bvec.bv_page); 445 446 if (len != bvec.bv_len) { 447 struct bio *bio; 448 449 __rq_for_each_bio(bio, rq) 450 zero_fill_bio(bio); 451 break; 452 } 453 } 454 455 ret = 0; 456 out_free_page: 457 __free_page(page); 458 return ret; 459 } 460 461 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 462 int mode) 463 { 464 /* 465 * We use fallocate to manipulate the space mappings used by the image 466 * a.k.a. discard/zerorange. However we do not support this if 467 * encryption is enabled, because it may give an attacker useful 468 * information. 469 */ 470 struct file *file = lo->lo_backing_file; 471 struct request_queue *q = lo->lo_queue; 472 int ret; 473 474 mode |= FALLOC_FL_KEEP_SIZE; 475 476 if (!blk_queue_discard(q)) { 477 ret = -EOPNOTSUPP; 478 goto out; 479 } 480 481 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 482 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 483 ret = -EIO; 484 out: 485 return ret; 486 } 487 488 static int lo_req_flush(struct loop_device *lo, struct request *rq) 489 { 490 struct file *file = lo->lo_backing_file; 491 int ret = vfs_fsync(file, 0); 492 if (unlikely(ret && ret != -EINVAL)) 493 ret = -EIO; 494 495 return ret; 496 } 497 498 static void lo_complete_rq(struct request *rq) 499 { 500 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 501 blk_status_t ret = BLK_STS_OK; 502 503 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 504 req_op(rq) != REQ_OP_READ) { 505 if (cmd->ret < 0) 506 ret = errno_to_blk_status(cmd->ret); 507 goto end_io; 508 } 509 510 /* 511 * Short READ - if we got some data, advance our request and 512 * retry it. If we got no data, end the rest with EIO. 513 */ 514 if (cmd->ret) { 515 blk_update_request(rq, BLK_STS_OK, cmd->ret); 516 cmd->ret = 0; 517 blk_mq_requeue_request(rq, true); 518 } else { 519 if (cmd->use_aio) { 520 struct bio *bio = rq->bio; 521 522 while (bio) { 523 zero_fill_bio(bio); 524 bio = bio->bi_next; 525 } 526 } 527 ret = BLK_STS_IOERR; 528 end_io: 529 blk_mq_end_request(rq, ret); 530 } 531 } 532 533 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 534 { 535 struct request *rq = blk_mq_rq_from_pdu(cmd); 536 537 if (!atomic_dec_and_test(&cmd->ref)) 538 return; 539 kfree(cmd->bvec); 540 cmd->bvec = NULL; 541 if (likely(!blk_should_fake_timeout(rq->q))) 542 blk_mq_complete_request(rq); 543 } 544 545 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 546 { 547 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 548 549 cmd->ret = ret; 550 lo_rw_aio_do_completion(cmd); 551 } 552 553 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 554 loff_t pos, bool rw) 555 { 556 struct iov_iter iter; 557 struct req_iterator rq_iter; 558 struct bio_vec *bvec; 559 struct request *rq = blk_mq_rq_from_pdu(cmd); 560 struct bio *bio = rq->bio; 561 struct file *file = lo->lo_backing_file; 562 struct bio_vec tmp; 563 unsigned int offset; 564 int nr_bvec = 0; 565 int ret; 566 567 rq_for_each_bvec(tmp, rq, rq_iter) 568 nr_bvec++; 569 570 if (rq->bio != rq->biotail) { 571 572 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 573 GFP_NOIO); 574 if (!bvec) 575 return -EIO; 576 cmd->bvec = bvec; 577 578 /* 579 * The bios of the request may be started from the middle of 580 * the 'bvec' because of bio splitting, so we can't directly 581 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 582 * API will take care of all details for us. 583 */ 584 rq_for_each_bvec(tmp, rq, rq_iter) { 585 *bvec = tmp; 586 bvec++; 587 } 588 bvec = cmd->bvec; 589 offset = 0; 590 } else { 591 /* 592 * Same here, this bio may be started from the middle of the 593 * 'bvec' because of bio splitting, so offset from the bvec 594 * must be passed to iov iterator 595 */ 596 offset = bio->bi_iter.bi_bvec_done; 597 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 598 } 599 atomic_set(&cmd->ref, 2); 600 601 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 602 iter.iov_offset = offset; 603 604 cmd->iocb.ki_pos = pos; 605 cmd->iocb.ki_filp = file; 606 cmd->iocb.ki_complete = lo_rw_aio_complete; 607 cmd->iocb.ki_flags = IOCB_DIRECT; 608 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 609 610 if (rw == WRITE) 611 ret = call_write_iter(file, &cmd->iocb, &iter); 612 else 613 ret = call_read_iter(file, &cmd->iocb, &iter); 614 615 lo_rw_aio_do_completion(cmd); 616 617 if (ret != -EIOCBQUEUED) 618 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 619 return 0; 620 } 621 622 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 623 { 624 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 625 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 626 627 /* 628 * lo_write_simple and lo_read_simple should have been covered 629 * by io submit style function like lo_rw_aio(), one blocker 630 * is that lo_read_simple() need to call flush_dcache_page after 631 * the page is written from kernel, and it isn't easy to handle 632 * this in io submit style function which submits all segments 633 * of the req at one time. And direct read IO doesn't need to 634 * run flush_dcache_page(). 635 */ 636 switch (req_op(rq)) { 637 case REQ_OP_FLUSH: 638 return lo_req_flush(lo, rq); 639 case REQ_OP_WRITE_ZEROES: 640 /* 641 * If the caller doesn't want deallocation, call zeroout to 642 * write zeroes the range. Otherwise, punch them out. 643 */ 644 return lo_fallocate(lo, rq, pos, 645 (rq->cmd_flags & REQ_NOUNMAP) ? 646 FALLOC_FL_ZERO_RANGE : 647 FALLOC_FL_PUNCH_HOLE); 648 case REQ_OP_DISCARD: 649 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 650 case REQ_OP_WRITE: 651 if (lo->transfer) 652 return lo_write_transfer(lo, rq, pos); 653 else if (cmd->use_aio) 654 return lo_rw_aio(lo, cmd, pos, WRITE); 655 else 656 return lo_write_simple(lo, rq, pos); 657 case REQ_OP_READ: 658 if (lo->transfer) 659 return lo_read_transfer(lo, rq, pos); 660 else if (cmd->use_aio) 661 return lo_rw_aio(lo, cmd, pos, READ); 662 else 663 return lo_read_simple(lo, rq, pos); 664 default: 665 WARN_ON_ONCE(1); 666 return -EIO; 667 } 668 } 669 670 static inline void loop_update_dio(struct loop_device *lo) 671 { 672 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 673 lo->use_dio); 674 } 675 676 static void loop_reread_partitions(struct loop_device *lo) 677 { 678 int rc; 679 680 mutex_lock(&lo->lo_disk->open_mutex); 681 rc = bdev_disk_changed(lo->lo_disk, false); 682 mutex_unlock(&lo->lo_disk->open_mutex); 683 if (rc) 684 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 685 __func__, lo->lo_number, lo->lo_file_name, rc); 686 } 687 688 static inline int is_loop_device(struct file *file) 689 { 690 struct inode *i = file->f_mapping->host; 691 692 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; 693 } 694 695 static int loop_validate_file(struct file *file, struct block_device *bdev) 696 { 697 struct inode *inode = file->f_mapping->host; 698 struct file *f = file; 699 700 /* Avoid recursion */ 701 while (is_loop_device(f)) { 702 struct loop_device *l; 703 704 lockdep_assert_held(&loop_validate_mutex); 705 if (f->f_mapping->host->i_rdev == bdev->bd_dev) 706 return -EBADF; 707 708 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 709 if (l->lo_state != Lo_bound) 710 return -EINVAL; 711 /* Order wrt setting lo->lo_backing_file in loop_configure(). */ 712 rmb(); 713 f = l->lo_backing_file; 714 } 715 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 716 return -EINVAL; 717 return 0; 718 } 719 720 /* 721 * loop_change_fd switched the backing store of a loopback device to 722 * a new file. This is useful for operating system installers to free up 723 * the original file and in High Availability environments to switch to 724 * an alternative location for the content in case of server meltdown. 725 * This can only work if the loop device is used read-only, and if the 726 * new backing store is the same size and type as the old backing store. 727 */ 728 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 729 unsigned int arg) 730 { 731 struct file *file = fget(arg); 732 struct file *old_file; 733 int error; 734 bool partscan; 735 bool is_loop; 736 737 if (!file) 738 return -EBADF; 739 is_loop = is_loop_device(file); 740 error = loop_global_lock_killable(lo, is_loop); 741 if (error) 742 goto out_putf; 743 error = -ENXIO; 744 if (lo->lo_state != Lo_bound) 745 goto out_err; 746 747 /* the loop device has to be read-only */ 748 error = -EINVAL; 749 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 750 goto out_err; 751 752 error = loop_validate_file(file, bdev); 753 if (error) 754 goto out_err; 755 756 old_file = lo->lo_backing_file; 757 758 error = -EINVAL; 759 760 /* size of the new backing store needs to be the same */ 761 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 762 goto out_err; 763 764 /* and ... switch */ 765 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 766 blk_mq_freeze_queue(lo->lo_queue); 767 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 768 lo->lo_backing_file = file; 769 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 770 mapping_set_gfp_mask(file->f_mapping, 771 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 772 loop_update_dio(lo); 773 blk_mq_unfreeze_queue(lo->lo_queue); 774 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 775 loop_global_unlock(lo, is_loop); 776 777 /* 778 * Flush loop_validate_file() before fput(), for l->lo_backing_file 779 * might be pointing at old_file which might be the last reference. 780 */ 781 if (!is_loop) { 782 mutex_lock(&loop_validate_mutex); 783 mutex_unlock(&loop_validate_mutex); 784 } 785 /* 786 * We must drop file reference outside of lo_mutex as dropping 787 * the file ref can take open_mutex which creates circular locking 788 * dependency. 789 */ 790 fput(old_file); 791 if (partscan) 792 loop_reread_partitions(lo); 793 return 0; 794 795 out_err: 796 loop_global_unlock(lo, is_loop); 797 out_putf: 798 fput(file); 799 return error; 800 } 801 802 /* loop sysfs attributes */ 803 804 static ssize_t loop_attr_show(struct device *dev, char *page, 805 ssize_t (*callback)(struct loop_device *, char *)) 806 { 807 struct gendisk *disk = dev_to_disk(dev); 808 struct loop_device *lo = disk->private_data; 809 810 return callback(lo, page); 811 } 812 813 #define LOOP_ATTR_RO(_name) \ 814 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 815 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 816 struct device_attribute *attr, char *b) \ 817 { \ 818 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 819 } \ 820 static struct device_attribute loop_attr_##_name = \ 821 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 822 823 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 824 { 825 ssize_t ret; 826 char *p = NULL; 827 828 spin_lock_irq(&lo->lo_lock); 829 if (lo->lo_backing_file) 830 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 831 spin_unlock_irq(&lo->lo_lock); 832 833 if (IS_ERR_OR_NULL(p)) 834 ret = PTR_ERR(p); 835 else { 836 ret = strlen(p); 837 memmove(buf, p, ret); 838 buf[ret++] = '\n'; 839 buf[ret] = 0; 840 } 841 842 return ret; 843 } 844 845 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 846 { 847 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 848 } 849 850 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 851 { 852 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 853 } 854 855 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 856 { 857 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 858 859 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 860 } 861 862 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 863 { 864 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 865 866 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 867 } 868 869 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 870 { 871 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 872 873 return sprintf(buf, "%s\n", dio ? "1" : "0"); 874 } 875 876 LOOP_ATTR_RO(backing_file); 877 LOOP_ATTR_RO(offset); 878 LOOP_ATTR_RO(sizelimit); 879 LOOP_ATTR_RO(autoclear); 880 LOOP_ATTR_RO(partscan); 881 LOOP_ATTR_RO(dio); 882 883 static struct attribute *loop_attrs[] = { 884 &loop_attr_backing_file.attr, 885 &loop_attr_offset.attr, 886 &loop_attr_sizelimit.attr, 887 &loop_attr_autoclear.attr, 888 &loop_attr_partscan.attr, 889 &loop_attr_dio.attr, 890 NULL, 891 }; 892 893 static struct attribute_group loop_attribute_group = { 894 .name = "loop", 895 .attrs= loop_attrs, 896 }; 897 898 static void loop_sysfs_init(struct loop_device *lo) 899 { 900 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 901 &loop_attribute_group); 902 } 903 904 static void loop_sysfs_exit(struct loop_device *lo) 905 { 906 if (lo->sysfs_inited) 907 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 908 &loop_attribute_group); 909 } 910 911 static void loop_config_discard(struct loop_device *lo) 912 { 913 struct file *file = lo->lo_backing_file; 914 struct inode *inode = file->f_mapping->host; 915 struct request_queue *q = lo->lo_queue; 916 u32 granularity, max_discard_sectors; 917 918 /* 919 * If the backing device is a block device, mirror its zeroing 920 * capability. Set the discard sectors to the block device's zeroing 921 * capabilities because loop discards result in blkdev_issue_zeroout(), 922 * not blkdev_issue_discard(). This maintains consistent behavior with 923 * file-backed loop devices: discarded regions read back as zero. 924 */ 925 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { 926 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 927 928 max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 929 granularity = backingq->limits.discard_granularity ?: 930 queue_physical_block_size(backingq); 931 932 /* 933 * We use punch hole to reclaim the free space used by the 934 * image a.k.a. discard. However we do not support discard if 935 * encryption is enabled, because it may give an attacker 936 * useful information. 937 */ 938 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { 939 max_discard_sectors = 0; 940 granularity = 0; 941 942 } else { 943 struct kstatfs sbuf; 944 945 max_discard_sectors = UINT_MAX >> 9; 946 if (!vfs_statfs(&file->f_path, &sbuf)) 947 granularity = sbuf.f_bsize; 948 else 949 max_discard_sectors = 0; 950 } 951 952 if (max_discard_sectors) { 953 q->limits.discard_granularity = granularity; 954 blk_queue_max_discard_sectors(q, max_discard_sectors); 955 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 956 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 957 } else { 958 q->limits.discard_granularity = 0; 959 blk_queue_max_discard_sectors(q, 0); 960 blk_queue_max_write_zeroes_sectors(q, 0); 961 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 962 } 963 q->limits.discard_alignment = 0; 964 } 965 966 struct loop_worker { 967 struct rb_node rb_node; 968 struct work_struct work; 969 struct list_head cmd_list; 970 struct list_head idle_list; 971 struct loop_device *lo; 972 struct cgroup_subsys_state *blkcg_css; 973 unsigned long last_ran_at; 974 }; 975 976 static void loop_workfn(struct work_struct *work); 977 static void loop_rootcg_workfn(struct work_struct *work); 978 static void loop_free_idle_workers(struct timer_list *timer); 979 980 #ifdef CONFIG_BLK_CGROUP 981 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 982 { 983 return !css || css == blkcg_root_css; 984 } 985 #else 986 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 987 { 988 return !css; 989 } 990 #endif 991 992 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 993 { 994 struct rb_node **node = &(lo->worker_tree.rb_node), *parent = NULL; 995 struct loop_worker *cur_worker, *worker = NULL; 996 struct work_struct *work; 997 struct list_head *cmd_list; 998 999 spin_lock_irq(&lo->lo_work_lock); 1000 1001 if (queue_on_root_worker(cmd->blkcg_css)) 1002 goto queue_work; 1003 1004 node = &lo->worker_tree.rb_node; 1005 1006 while (*node) { 1007 parent = *node; 1008 cur_worker = container_of(*node, struct loop_worker, rb_node); 1009 if (cur_worker->blkcg_css == cmd->blkcg_css) { 1010 worker = cur_worker; 1011 break; 1012 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { 1013 node = &(*node)->rb_left; 1014 } else { 1015 node = &(*node)->rb_right; 1016 } 1017 } 1018 if (worker) 1019 goto queue_work; 1020 1021 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 1022 /* 1023 * In the event we cannot allocate a worker, just queue on the 1024 * rootcg worker and issue the I/O as the rootcg 1025 */ 1026 if (!worker) { 1027 cmd->blkcg_css = NULL; 1028 if (cmd->memcg_css) 1029 css_put(cmd->memcg_css); 1030 cmd->memcg_css = NULL; 1031 goto queue_work; 1032 } 1033 1034 worker->blkcg_css = cmd->blkcg_css; 1035 css_get(worker->blkcg_css); 1036 INIT_WORK(&worker->work, loop_workfn); 1037 INIT_LIST_HEAD(&worker->cmd_list); 1038 INIT_LIST_HEAD(&worker->idle_list); 1039 worker->lo = lo; 1040 rb_link_node(&worker->rb_node, parent, node); 1041 rb_insert_color(&worker->rb_node, &lo->worker_tree); 1042 queue_work: 1043 if (worker) { 1044 /* 1045 * We need to remove from the idle list here while 1046 * holding the lock so that the idle timer doesn't 1047 * free the worker 1048 */ 1049 if (!list_empty(&worker->idle_list)) 1050 list_del_init(&worker->idle_list); 1051 work = &worker->work; 1052 cmd_list = &worker->cmd_list; 1053 } else { 1054 work = &lo->rootcg_work; 1055 cmd_list = &lo->rootcg_cmd_list; 1056 } 1057 list_add_tail(&cmd->list_entry, cmd_list); 1058 queue_work(lo->workqueue, work); 1059 spin_unlock_irq(&lo->lo_work_lock); 1060 } 1061 1062 static void loop_update_rotational(struct loop_device *lo) 1063 { 1064 struct file *file = lo->lo_backing_file; 1065 struct inode *file_inode = file->f_mapping->host; 1066 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 1067 struct request_queue *q = lo->lo_queue; 1068 bool nonrot = true; 1069 1070 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 1071 if (file_bdev) 1072 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 1073 1074 if (nonrot) 1075 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 1076 else 1077 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 1078 } 1079 1080 static int 1081 loop_release_xfer(struct loop_device *lo) 1082 { 1083 int err = 0; 1084 struct loop_func_table *xfer = lo->lo_encryption; 1085 1086 if (xfer) { 1087 if (xfer->release) 1088 err = xfer->release(lo); 1089 lo->transfer = NULL; 1090 lo->lo_encryption = NULL; 1091 module_put(xfer->owner); 1092 } 1093 return err; 1094 } 1095 1096 static int 1097 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1098 const struct loop_info64 *i) 1099 { 1100 int err = 0; 1101 1102 if (xfer) { 1103 struct module *owner = xfer->owner; 1104 1105 if (!try_module_get(owner)) 1106 return -EINVAL; 1107 if (xfer->init) 1108 err = xfer->init(lo, i); 1109 if (err) 1110 module_put(owner); 1111 else 1112 lo->lo_encryption = xfer; 1113 } 1114 return err; 1115 } 1116 1117 /** 1118 * loop_set_status_from_info - configure device from loop_info 1119 * @lo: struct loop_device to configure 1120 * @info: struct loop_info64 to configure the device with 1121 * 1122 * Configures the loop device parameters according to the passed 1123 * in loop_info64 configuration. 1124 */ 1125 static int 1126 loop_set_status_from_info(struct loop_device *lo, 1127 const struct loop_info64 *info) 1128 { 1129 int err; 1130 struct loop_func_table *xfer; 1131 kuid_t uid = current_uid(); 1132 1133 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1134 return -EINVAL; 1135 1136 err = loop_release_xfer(lo); 1137 if (err) 1138 return err; 1139 1140 if (info->lo_encrypt_type) { 1141 unsigned int type = info->lo_encrypt_type; 1142 1143 if (type >= MAX_LO_CRYPT) 1144 return -EINVAL; 1145 xfer = xfer_funcs[type]; 1146 if (xfer == NULL) 1147 return -EINVAL; 1148 } else 1149 xfer = NULL; 1150 1151 err = loop_init_xfer(lo, xfer, info); 1152 if (err) 1153 return err; 1154 1155 lo->lo_offset = info->lo_offset; 1156 lo->lo_sizelimit = info->lo_sizelimit; 1157 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1158 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1159 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1160 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1161 1162 if (!xfer) 1163 xfer = &none_funcs; 1164 lo->transfer = xfer->transfer; 1165 lo->ioctl = xfer->ioctl; 1166 1167 lo->lo_flags = info->lo_flags; 1168 1169 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1170 lo->lo_init[0] = info->lo_init[0]; 1171 lo->lo_init[1] = info->lo_init[1]; 1172 if (info->lo_encrypt_key_size) { 1173 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1174 info->lo_encrypt_key_size); 1175 lo->lo_key_owner = uid; 1176 } 1177 1178 return 0; 1179 } 1180 1181 static int loop_configure(struct loop_device *lo, fmode_t mode, 1182 struct block_device *bdev, 1183 const struct loop_config *config) 1184 { 1185 struct file *file = fget(config->fd); 1186 struct inode *inode; 1187 struct address_space *mapping; 1188 int error; 1189 loff_t size; 1190 bool partscan; 1191 unsigned short bsize; 1192 bool is_loop; 1193 1194 if (!file) 1195 return -EBADF; 1196 is_loop = is_loop_device(file); 1197 1198 /* This is safe, since we have a reference from open(). */ 1199 __module_get(THIS_MODULE); 1200 1201 /* 1202 * If we don't hold exclusive handle for the device, upgrade to it 1203 * here to avoid changing device under exclusive owner. 1204 */ 1205 if (!(mode & FMODE_EXCL)) { 1206 error = bd_prepare_to_claim(bdev, loop_configure); 1207 if (error) 1208 goto out_putf; 1209 } 1210 1211 error = loop_global_lock_killable(lo, is_loop); 1212 if (error) 1213 goto out_bdev; 1214 1215 error = -EBUSY; 1216 if (lo->lo_state != Lo_unbound) 1217 goto out_unlock; 1218 1219 error = loop_validate_file(file, bdev); 1220 if (error) 1221 goto out_unlock; 1222 1223 mapping = file->f_mapping; 1224 inode = mapping->host; 1225 1226 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1227 error = -EINVAL; 1228 goto out_unlock; 1229 } 1230 1231 if (config->block_size) { 1232 error = blk_validate_block_size(config->block_size); 1233 if (error) 1234 goto out_unlock; 1235 } 1236 1237 error = loop_set_status_from_info(lo, &config->info); 1238 if (error) 1239 goto out_unlock; 1240 1241 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1242 !file->f_op->write_iter) 1243 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1244 1245 lo->workqueue = alloc_workqueue("loop%d", 1246 WQ_UNBOUND | WQ_FREEZABLE, 1247 0, 1248 lo->lo_number); 1249 if (!lo->workqueue) { 1250 error = -ENOMEM; 1251 goto out_unlock; 1252 } 1253 1254 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1255 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1256 1257 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 1258 INIT_LIST_HEAD(&lo->rootcg_cmd_list); 1259 INIT_LIST_HEAD(&lo->idle_worker_list); 1260 lo->worker_tree = RB_ROOT; 1261 timer_setup(&lo->timer, loop_free_idle_workers, 1262 TIMER_DEFERRABLE); 1263 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1264 lo->lo_device = bdev; 1265 lo->lo_backing_file = file; 1266 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1267 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1268 1269 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1270 blk_queue_write_cache(lo->lo_queue, true, false); 1271 1272 if (config->block_size) 1273 bsize = config->block_size; 1274 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1275 /* In case of direct I/O, match underlying block size */ 1276 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1277 else 1278 bsize = 512; 1279 1280 blk_queue_logical_block_size(lo->lo_queue, bsize); 1281 blk_queue_physical_block_size(lo->lo_queue, bsize); 1282 blk_queue_io_min(lo->lo_queue, bsize); 1283 1284 loop_config_discard(lo); 1285 loop_update_rotational(lo); 1286 loop_update_dio(lo); 1287 loop_sysfs_init(lo); 1288 1289 size = get_loop_size(lo, file); 1290 loop_set_size(lo, size); 1291 1292 /* Order wrt reading lo_state in loop_validate_file(). */ 1293 wmb(); 1294 1295 lo->lo_state = Lo_bound; 1296 if (part_shift) 1297 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1298 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1299 if (partscan) 1300 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1301 1302 loop_global_unlock(lo, is_loop); 1303 if (partscan) 1304 loop_reread_partitions(lo); 1305 if (!(mode & FMODE_EXCL)) 1306 bd_abort_claiming(bdev, loop_configure); 1307 return 0; 1308 1309 out_unlock: 1310 loop_global_unlock(lo, is_loop); 1311 out_bdev: 1312 if (!(mode & FMODE_EXCL)) 1313 bd_abort_claiming(bdev, loop_configure); 1314 out_putf: 1315 fput(file); 1316 /* This is safe: open() is still holding a reference. */ 1317 module_put(THIS_MODULE); 1318 return error; 1319 } 1320 1321 static int __loop_clr_fd(struct loop_device *lo, bool release) 1322 { 1323 struct file *filp = NULL; 1324 gfp_t gfp = lo->old_gfp_mask; 1325 struct block_device *bdev = lo->lo_device; 1326 int err = 0; 1327 bool partscan = false; 1328 int lo_number; 1329 struct loop_worker *pos, *worker; 1330 1331 /* 1332 * Flush loop_configure() and loop_change_fd(). It is acceptable for 1333 * loop_validate_file() to succeed, for actual clear operation has not 1334 * started yet. 1335 */ 1336 mutex_lock(&loop_validate_mutex); 1337 mutex_unlock(&loop_validate_mutex); 1338 /* 1339 * loop_validate_file() now fails because l->lo_state != Lo_bound 1340 * became visible. 1341 */ 1342 1343 mutex_lock(&lo->lo_mutex); 1344 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1345 err = -ENXIO; 1346 goto out_unlock; 1347 } 1348 1349 filp = lo->lo_backing_file; 1350 if (filp == NULL) { 1351 err = -EINVAL; 1352 goto out_unlock; 1353 } 1354 1355 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1356 blk_queue_write_cache(lo->lo_queue, false, false); 1357 1358 /* freeze request queue during the transition */ 1359 blk_mq_freeze_queue(lo->lo_queue); 1360 1361 destroy_workqueue(lo->workqueue); 1362 spin_lock_irq(&lo->lo_work_lock); 1363 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 1364 idle_list) { 1365 list_del(&worker->idle_list); 1366 rb_erase(&worker->rb_node, &lo->worker_tree); 1367 css_put(worker->blkcg_css); 1368 kfree(worker); 1369 } 1370 spin_unlock_irq(&lo->lo_work_lock); 1371 del_timer_sync(&lo->timer); 1372 1373 spin_lock_irq(&lo->lo_lock); 1374 lo->lo_backing_file = NULL; 1375 spin_unlock_irq(&lo->lo_lock); 1376 1377 loop_release_xfer(lo); 1378 lo->transfer = NULL; 1379 lo->ioctl = NULL; 1380 lo->lo_device = NULL; 1381 lo->lo_encryption = NULL; 1382 lo->lo_offset = 0; 1383 lo->lo_sizelimit = 0; 1384 lo->lo_encrypt_key_size = 0; 1385 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1386 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1387 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1388 blk_queue_logical_block_size(lo->lo_queue, 512); 1389 blk_queue_physical_block_size(lo->lo_queue, 512); 1390 blk_queue_io_min(lo->lo_queue, 512); 1391 if (bdev) { 1392 invalidate_bdev(bdev); 1393 bdev->bd_inode->i_mapping->wb_err = 0; 1394 } 1395 set_capacity(lo->lo_disk, 0); 1396 loop_sysfs_exit(lo); 1397 if (bdev) { 1398 /* let user-space know about this change */ 1399 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1400 } 1401 mapping_set_gfp_mask(filp->f_mapping, gfp); 1402 /* This is safe: open() is still holding a reference. */ 1403 module_put(THIS_MODULE); 1404 blk_mq_unfreeze_queue(lo->lo_queue); 1405 1406 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1407 lo_number = lo->lo_number; 1408 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1409 out_unlock: 1410 mutex_unlock(&lo->lo_mutex); 1411 if (partscan) { 1412 /* 1413 * open_mutex has been held already in release path, so don't 1414 * acquire it if this function is called in such case. 1415 * 1416 * If the reread partition isn't from release path, lo_refcnt 1417 * must be at least one and it can only become zero when the 1418 * current holder is released. 1419 */ 1420 if (!release) 1421 mutex_lock(&lo->lo_disk->open_mutex); 1422 err = bdev_disk_changed(lo->lo_disk, false); 1423 if (!release) 1424 mutex_unlock(&lo->lo_disk->open_mutex); 1425 if (err) 1426 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1427 __func__, lo_number, err); 1428 /* Device is gone, no point in returning error */ 1429 err = 0; 1430 } 1431 1432 /* 1433 * lo->lo_state is set to Lo_unbound here after above partscan has 1434 * finished. 1435 * 1436 * There cannot be anybody else entering __loop_clr_fd() as 1437 * lo->lo_backing_file is already cleared and Lo_rundown state 1438 * protects us from all the other places trying to change the 'lo' 1439 * device. 1440 */ 1441 mutex_lock(&lo->lo_mutex); 1442 lo->lo_flags = 0; 1443 if (!part_shift) 1444 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1445 lo->lo_state = Lo_unbound; 1446 mutex_unlock(&lo->lo_mutex); 1447 1448 /* 1449 * Need not hold lo_mutex to fput backing file. Calling fput holding 1450 * lo_mutex triggers a circular lock dependency possibility warning as 1451 * fput can take open_mutex which is usually taken before lo_mutex. 1452 */ 1453 if (filp) 1454 fput(filp); 1455 return err; 1456 } 1457 1458 static int loop_clr_fd(struct loop_device *lo) 1459 { 1460 int err; 1461 1462 err = mutex_lock_killable(&lo->lo_mutex); 1463 if (err) 1464 return err; 1465 if (lo->lo_state != Lo_bound) { 1466 mutex_unlock(&lo->lo_mutex); 1467 return -ENXIO; 1468 } 1469 /* 1470 * If we've explicitly asked to tear down the loop device, 1471 * and it has an elevated reference count, set it for auto-teardown when 1472 * the last reference goes away. This stops $!~#$@ udev from 1473 * preventing teardown because it decided that it needs to run blkid on 1474 * the loopback device whenever they appear. xfstests is notorious for 1475 * failing tests because blkid via udev races with a losetup 1476 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1477 * command to fail with EBUSY. 1478 */ 1479 if (atomic_read(&lo->lo_refcnt) > 1) { 1480 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1481 mutex_unlock(&lo->lo_mutex); 1482 return 0; 1483 } 1484 lo->lo_state = Lo_rundown; 1485 mutex_unlock(&lo->lo_mutex); 1486 1487 return __loop_clr_fd(lo, false); 1488 } 1489 1490 static int 1491 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1492 { 1493 int err; 1494 kuid_t uid = current_uid(); 1495 int prev_lo_flags; 1496 bool partscan = false; 1497 bool size_changed = false; 1498 1499 err = mutex_lock_killable(&lo->lo_mutex); 1500 if (err) 1501 return err; 1502 if (lo->lo_encrypt_key_size && 1503 !uid_eq(lo->lo_key_owner, uid) && 1504 !capable(CAP_SYS_ADMIN)) { 1505 err = -EPERM; 1506 goto out_unlock; 1507 } 1508 if (lo->lo_state != Lo_bound) { 1509 err = -ENXIO; 1510 goto out_unlock; 1511 } 1512 1513 if (lo->lo_offset != info->lo_offset || 1514 lo->lo_sizelimit != info->lo_sizelimit) { 1515 size_changed = true; 1516 sync_blockdev(lo->lo_device); 1517 invalidate_bdev(lo->lo_device); 1518 } 1519 1520 /* I/O need to be drained during transfer transition */ 1521 blk_mq_freeze_queue(lo->lo_queue); 1522 1523 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1524 /* If any pages were dirtied after invalidate_bdev(), try again */ 1525 err = -EAGAIN; 1526 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1527 __func__, lo->lo_number, lo->lo_file_name, 1528 lo->lo_device->bd_inode->i_mapping->nrpages); 1529 goto out_unfreeze; 1530 } 1531 1532 prev_lo_flags = lo->lo_flags; 1533 1534 err = loop_set_status_from_info(lo, info); 1535 if (err) 1536 goto out_unfreeze; 1537 1538 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1539 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1540 /* For those flags, use the previous values instead */ 1541 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1542 /* For flags that can't be cleared, use previous values too */ 1543 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1544 1545 if (size_changed) { 1546 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1547 lo->lo_backing_file); 1548 loop_set_size(lo, new_size); 1549 } 1550 1551 loop_config_discard(lo); 1552 1553 /* update dio if lo_offset or transfer is changed */ 1554 __loop_update_dio(lo, lo->use_dio); 1555 1556 out_unfreeze: 1557 blk_mq_unfreeze_queue(lo->lo_queue); 1558 1559 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1560 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1561 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1562 partscan = true; 1563 } 1564 out_unlock: 1565 mutex_unlock(&lo->lo_mutex); 1566 if (partscan) 1567 loop_reread_partitions(lo); 1568 1569 return err; 1570 } 1571 1572 static int 1573 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1574 { 1575 struct path path; 1576 struct kstat stat; 1577 int ret; 1578 1579 ret = mutex_lock_killable(&lo->lo_mutex); 1580 if (ret) 1581 return ret; 1582 if (lo->lo_state != Lo_bound) { 1583 mutex_unlock(&lo->lo_mutex); 1584 return -ENXIO; 1585 } 1586 1587 memset(info, 0, sizeof(*info)); 1588 info->lo_number = lo->lo_number; 1589 info->lo_offset = lo->lo_offset; 1590 info->lo_sizelimit = lo->lo_sizelimit; 1591 info->lo_flags = lo->lo_flags; 1592 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1593 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1594 info->lo_encrypt_type = 1595 lo->lo_encryption ? lo->lo_encryption->number : 0; 1596 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1597 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1598 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1599 lo->lo_encrypt_key_size); 1600 } 1601 1602 /* Drop lo_mutex while we call into the filesystem. */ 1603 path = lo->lo_backing_file->f_path; 1604 path_get(&path); 1605 mutex_unlock(&lo->lo_mutex); 1606 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1607 if (!ret) { 1608 info->lo_device = huge_encode_dev(stat.dev); 1609 info->lo_inode = stat.ino; 1610 info->lo_rdevice = huge_encode_dev(stat.rdev); 1611 } 1612 path_put(&path); 1613 return ret; 1614 } 1615 1616 static void 1617 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1618 { 1619 memset(info64, 0, sizeof(*info64)); 1620 info64->lo_number = info->lo_number; 1621 info64->lo_device = info->lo_device; 1622 info64->lo_inode = info->lo_inode; 1623 info64->lo_rdevice = info->lo_rdevice; 1624 info64->lo_offset = info->lo_offset; 1625 info64->lo_sizelimit = 0; 1626 info64->lo_encrypt_type = info->lo_encrypt_type; 1627 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1628 info64->lo_flags = info->lo_flags; 1629 info64->lo_init[0] = info->lo_init[0]; 1630 info64->lo_init[1] = info->lo_init[1]; 1631 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1632 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1633 else 1634 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1635 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1636 } 1637 1638 static int 1639 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1640 { 1641 memset(info, 0, sizeof(*info)); 1642 info->lo_number = info64->lo_number; 1643 info->lo_device = info64->lo_device; 1644 info->lo_inode = info64->lo_inode; 1645 info->lo_rdevice = info64->lo_rdevice; 1646 info->lo_offset = info64->lo_offset; 1647 info->lo_encrypt_type = info64->lo_encrypt_type; 1648 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1649 info->lo_flags = info64->lo_flags; 1650 info->lo_init[0] = info64->lo_init[0]; 1651 info->lo_init[1] = info64->lo_init[1]; 1652 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1653 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1654 else 1655 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1656 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1657 1658 /* error in case values were truncated */ 1659 if (info->lo_device != info64->lo_device || 1660 info->lo_rdevice != info64->lo_rdevice || 1661 info->lo_inode != info64->lo_inode || 1662 info->lo_offset != info64->lo_offset) 1663 return -EOVERFLOW; 1664 1665 return 0; 1666 } 1667 1668 static int 1669 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1670 { 1671 struct loop_info info; 1672 struct loop_info64 info64; 1673 1674 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1675 return -EFAULT; 1676 loop_info64_from_old(&info, &info64); 1677 return loop_set_status(lo, &info64); 1678 } 1679 1680 static int 1681 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1682 { 1683 struct loop_info64 info64; 1684 1685 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1686 return -EFAULT; 1687 return loop_set_status(lo, &info64); 1688 } 1689 1690 static int 1691 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1692 struct loop_info info; 1693 struct loop_info64 info64; 1694 int err; 1695 1696 if (!arg) 1697 return -EINVAL; 1698 err = loop_get_status(lo, &info64); 1699 if (!err) 1700 err = loop_info64_to_old(&info64, &info); 1701 if (!err && copy_to_user(arg, &info, sizeof(info))) 1702 err = -EFAULT; 1703 1704 return err; 1705 } 1706 1707 static int 1708 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1709 struct loop_info64 info64; 1710 int err; 1711 1712 if (!arg) 1713 return -EINVAL; 1714 err = loop_get_status(lo, &info64); 1715 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1716 err = -EFAULT; 1717 1718 return err; 1719 } 1720 1721 static int loop_set_capacity(struct loop_device *lo) 1722 { 1723 loff_t size; 1724 1725 if (unlikely(lo->lo_state != Lo_bound)) 1726 return -ENXIO; 1727 1728 size = get_loop_size(lo, lo->lo_backing_file); 1729 loop_set_size(lo, size); 1730 1731 return 0; 1732 } 1733 1734 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1735 { 1736 int error = -ENXIO; 1737 if (lo->lo_state != Lo_bound) 1738 goto out; 1739 1740 __loop_update_dio(lo, !!arg); 1741 if (lo->use_dio == !!arg) 1742 return 0; 1743 error = -EINVAL; 1744 out: 1745 return error; 1746 } 1747 1748 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1749 { 1750 int err = 0; 1751 1752 if (lo->lo_state != Lo_bound) 1753 return -ENXIO; 1754 1755 err = blk_validate_block_size(arg); 1756 if (err) 1757 return err; 1758 1759 if (lo->lo_queue->limits.logical_block_size == arg) 1760 return 0; 1761 1762 sync_blockdev(lo->lo_device); 1763 invalidate_bdev(lo->lo_device); 1764 1765 blk_mq_freeze_queue(lo->lo_queue); 1766 1767 /* invalidate_bdev should have truncated all the pages */ 1768 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1769 err = -EAGAIN; 1770 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1771 __func__, lo->lo_number, lo->lo_file_name, 1772 lo->lo_device->bd_inode->i_mapping->nrpages); 1773 goto out_unfreeze; 1774 } 1775 1776 blk_queue_logical_block_size(lo->lo_queue, arg); 1777 blk_queue_physical_block_size(lo->lo_queue, arg); 1778 blk_queue_io_min(lo->lo_queue, arg); 1779 loop_update_dio(lo); 1780 out_unfreeze: 1781 blk_mq_unfreeze_queue(lo->lo_queue); 1782 1783 return err; 1784 } 1785 1786 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1787 unsigned long arg) 1788 { 1789 int err; 1790 1791 err = mutex_lock_killable(&lo->lo_mutex); 1792 if (err) 1793 return err; 1794 switch (cmd) { 1795 case LOOP_SET_CAPACITY: 1796 err = loop_set_capacity(lo); 1797 break; 1798 case LOOP_SET_DIRECT_IO: 1799 err = loop_set_dio(lo, arg); 1800 break; 1801 case LOOP_SET_BLOCK_SIZE: 1802 err = loop_set_block_size(lo, arg); 1803 break; 1804 default: 1805 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1806 } 1807 mutex_unlock(&lo->lo_mutex); 1808 return err; 1809 } 1810 1811 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1812 unsigned int cmd, unsigned long arg) 1813 { 1814 struct loop_device *lo = bdev->bd_disk->private_data; 1815 void __user *argp = (void __user *) arg; 1816 int err; 1817 1818 switch (cmd) { 1819 case LOOP_SET_FD: { 1820 /* 1821 * Legacy case - pass in a zeroed out struct loop_config with 1822 * only the file descriptor set , which corresponds with the 1823 * default parameters we'd have used otherwise. 1824 */ 1825 struct loop_config config; 1826 1827 memset(&config, 0, sizeof(config)); 1828 config.fd = arg; 1829 1830 return loop_configure(lo, mode, bdev, &config); 1831 } 1832 case LOOP_CONFIGURE: { 1833 struct loop_config config; 1834 1835 if (copy_from_user(&config, argp, sizeof(config))) 1836 return -EFAULT; 1837 1838 return loop_configure(lo, mode, bdev, &config); 1839 } 1840 case LOOP_CHANGE_FD: 1841 return loop_change_fd(lo, bdev, arg); 1842 case LOOP_CLR_FD: 1843 return loop_clr_fd(lo); 1844 case LOOP_SET_STATUS: 1845 err = -EPERM; 1846 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1847 err = loop_set_status_old(lo, argp); 1848 } 1849 break; 1850 case LOOP_GET_STATUS: 1851 return loop_get_status_old(lo, argp); 1852 case LOOP_SET_STATUS64: 1853 err = -EPERM; 1854 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1855 err = loop_set_status64(lo, argp); 1856 } 1857 break; 1858 case LOOP_GET_STATUS64: 1859 return loop_get_status64(lo, argp); 1860 case LOOP_SET_CAPACITY: 1861 case LOOP_SET_DIRECT_IO: 1862 case LOOP_SET_BLOCK_SIZE: 1863 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1864 return -EPERM; 1865 fallthrough; 1866 default: 1867 err = lo_simple_ioctl(lo, cmd, arg); 1868 break; 1869 } 1870 1871 return err; 1872 } 1873 1874 #ifdef CONFIG_COMPAT 1875 struct compat_loop_info { 1876 compat_int_t lo_number; /* ioctl r/o */ 1877 compat_dev_t lo_device; /* ioctl r/o */ 1878 compat_ulong_t lo_inode; /* ioctl r/o */ 1879 compat_dev_t lo_rdevice; /* ioctl r/o */ 1880 compat_int_t lo_offset; 1881 compat_int_t lo_encrypt_type; 1882 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1883 compat_int_t lo_flags; /* ioctl r/o */ 1884 char lo_name[LO_NAME_SIZE]; 1885 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1886 compat_ulong_t lo_init[2]; 1887 char reserved[4]; 1888 }; 1889 1890 /* 1891 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1892 * - noinlined to reduce stack space usage in main part of driver 1893 */ 1894 static noinline int 1895 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1896 struct loop_info64 *info64) 1897 { 1898 struct compat_loop_info info; 1899 1900 if (copy_from_user(&info, arg, sizeof(info))) 1901 return -EFAULT; 1902 1903 memset(info64, 0, sizeof(*info64)); 1904 info64->lo_number = info.lo_number; 1905 info64->lo_device = info.lo_device; 1906 info64->lo_inode = info.lo_inode; 1907 info64->lo_rdevice = info.lo_rdevice; 1908 info64->lo_offset = info.lo_offset; 1909 info64->lo_sizelimit = 0; 1910 info64->lo_encrypt_type = info.lo_encrypt_type; 1911 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1912 info64->lo_flags = info.lo_flags; 1913 info64->lo_init[0] = info.lo_init[0]; 1914 info64->lo_init[1] = info.lo_init[1]; 1915 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1916 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1917 else 1918 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1919 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1920 return 0; 1921 } 1922 1923 /* 1924 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1925 * - noinlined to reduce stack space usage in main part of driver 1926 */ 1927 static noinline int 1928 loop_info64_to_compat(const struct loop_info64 *info64, 1929 struct compat_loop_info __user *arg) 1930 { 1931 struct compat_loop_info info; 1932 1933 memset(&info, 0, sizeof(info)); 1934 info.lo_number = info64->lo_number; 1935 info.lo_device = info64->lo_device; 1936 info.lo_inode = info64->lo_inode; 1937 info.lo_rdevice = info64->lo_rdevice; 1938 info.lo_offset = info64->lo_offset; 1939 info.lo_encrypt_type = info64->lo_encrypt_type; 1940 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1941 info.lo_flags = info64->lo_flags; 1942 info.lo_init[0] = info64->lo_init[0]; 1943 info.lo_init[1] = info64->lo_init[1]; 1944 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1945 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1946 else 1947 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1948 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1949 1950 /* error in case values were truncated */ 1951 if (info.lo_device != info64->lo_device || 1952 info.lo_rdevice != info64->lo_rdevice || 1953 info.lo_inode != info64->lo_inode || 1954 info.lo_offset != info64->lo_offset || 1955 info.lo_init[0] != info64->lo_init[0] || 1956 info.lo_init[1] != info64->lo_init[1]) 1957 return -EOVERFLOW; 1958 1959 if (copy_to_user(arg, &info, sizeof(info))) 1960 return -EFAULT; 1961 return 0; 1962 } 1963 1964 static int 1965 loop_set_status_compat(struct loop_device *lo, 1966 const struct compat_loop_info __user *arg) 1967 { 1968 struct loop_info64 info64; 1969 int ret; 1970 1971 ret = loop_info64_from_compat(arg, &info64); 1972 if (ret < 0) 1973 return ret; 1974 return loop_set_status(lo, &info64); 1975 } 1976 1977 static int 1978 loop_get_status_compat(struct loop_device *lo, 1979 struct compat_loop_info __user *arg) 1980 { 1981 struct loop_info64 info64; 1982 int err; 1983 1984 if (!arg) 1985 return -EINVAL; 1986 err = loop_get_status(lo, &info64); 1987 if (!err) 1988 err = loop_info64_to_compat(&info64, arg); 1989 return err; 1990 } 1991 1992 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1993 unsigned int cmd, unsigned long arg) 1994 { 1995 struct loop_device *lo = bdev->bd_disk->private_data; 1996 int err; 1997 1998 switch(cmd) { 1999 case LOOP_SET_STATUS: 2000 err = loop_set_status_compat(lo, 2001 (const struct compat_loop_info __user *)arg); 2002 break; 2003 case LOOP_GET_STATUS: 2004 err = loop_get_status_compat(lo, 2005 (struct compat_loop_info __user *)arg); 2006 break; 2007 case LOOP_SET_CAPACITY: 2008 case LOOP_CLR_FD: 2009 case LOOP_GET_STATUS64: 2010 case LOOP_SET_STATUS64: 2011 case LOOP_CONFIGURE: 2012 arg = (unsigned long) compat_ptr(arg); 2013 fallthrough; 2014 case LOOP_SET_FD: 2015 case LOOP_CHANGE_FD: 2016 case LOOP_SET_BLOCK_SIZE: 2017 case LOOP_SET_DIRECT_IO: 2018 err = lo_ioctl(bdev, mode, cmd, arg); 2019 break; 2020 default: 2021 err = -ENOIOCTLCMD; 2022 break; 2023 } 2024 return err; 2025 } 2026 #endif 2027 2028 static int lo_open(struct block_device *bdev, fmode_t mode) 2029 { 2030 struct loop_device *lo = bdev->bd_disk->private_data; 2031 int err; 2032 2033 err = mutex_lock_killable(&lo->lo_mutex); 2034 if (err) 2035 return err; 2036 if (lo->lo_state == Lo_deleting) 2037 err = -ENXIO; 2038 else 2039 atomic_inc(&lo->lo_refcnt); 2040 mutex_unlock(&lo->lo_mutex); 2041 return err; 2042 } 2043 2044 static void lo_release(struct gendisk *disk, fmode_t mode) 2045 { 2046 struct loop_device *lo = disk->private_data; 2047 2048 mutex_lock(&lo->lo_mutex); 2049 if (atomic_dec_return(&lo->lo_refcnt)) 2050 goto out_unlock; 2051 2052 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 2053 if (lo->lo_state != Lo_bound) 2054 goto out_unlock; 2055 lo->lo_state = Lo_rundown; 2056 mutex_unlock(&lo->lo_mutex); 2057 /* 2058 * In autoclear mode, stop the loop thread 2059 * and remove configuration after last close. 2060 */ 2061 __loop_clr_fd(lo, true); 2062 return; 2063 } else if (lo->lo_state == Lo_bound) { 2064 /* 2065 * Otherwise keep thread (if running) and config, 2066 * but flush possible ongoing bios in thread. 2067 */ 2068 blk_mq_freeze_queue(lo->lo_queue); 2069 blk_mq_unfreeze_queue(lo->lo_queue); 2070 } 2071 2072 out_unlock: 2073 mutex_unlock(&lo->lo_mutex); 2074 } 2075 2076 static const struct block_device_operations lo_fops = { 2077 .owner = THIS_MODULE, 2078 .open = lo_open, 2079 .release = lo_release, 2080 .ioctl = lo_ioctl, 2081 #ifdef CONFIG_COMPAT 2082 .compat_ioctl = lo_compat_ioctl, 2083 #endif 2084 }; 2085 2086 /* 2087 * And now the modules code and kernel interface. 2088 */ 2089 static int max_loop; 2090 module_param(max_loop, int, 0444); 2091 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 2092 module_param(max_part, int, 0444); 2093 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 2094 MODULE_LICENSE("GPL"); 2095 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 2096 2097 int loop_register_transfer(struct loop_func_table *funcs) 2098 { 2099 unsigned int n = funcs->number; 2100 2101 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 2102 return -EINVAL; 2103 xfer_funcs[n] = funcs; 2104 return 0; 2105 } 2106 2107 int loop_unregister_transfer(int number) 2108 { 2109 unsigned int n = number; 2110 struct loop_func_table *xfer; 2111 2112 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 2113 return -EINVAL; 2114 /* 2115 * This function is called from only cleanup_cryptoloop(). 2116 * Given that each loop device that has a transfer enabled holds a 2117 * reference to the module implementing it we should never get here 2118 * with a transfer that is set (unless forced module unloading is 2119 * requested). Thus, check module's refcount and warn if this is 2120 * not a clean unloading. 2121 */ 2122 #ifdef CONFIG_MODULE_UNLOAD 2123 if (xfer->owner && module_refcount(xfer->owner) != -1) 2124 pr_err("Danger! Unregistering an in use transfer function.\n"); 2125 #endif 2126 2127 xfer_funcs[n] = NULL; 2128 return 0; 2129 } 2130 2131 EXPORT_SYMBOL(loop_register_transfer); 2132 EXPORT_SYMBOL(loop_unregister_transfer); 2133 2134 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 2135 const struct blk_mq_queue_data *bd) 2136 { 2137 struct request *rq = bd->rq; 2138 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2139 struct loop_device *lo = rq->q->queuedata; 2140 2141 blk_mq_start_request(rq); 2142 2143 if (lo->lo_state != Lo_bound) 2144 return BLK_STS_IOERR; 2145 2146 switch (req_op(rq)) { 2147 case REQ_OP_FLUSH: 2148 case REQ_OP_DISCARD: 2149 case REQ_OP_WRITE_ZEROES: 2150 cmd->use_aio = false; 2151 break; 2152 default: 2153 cmd->use_aio = lo->use_dio; 2154 break; 2155 } 2156 2157 /* always use the first bio's css */ 2158 cmd->blkcg_css = NULL; 2159 cmd->memcg_css = NULL; 2160 #ifdef CONFIG_BLK_CGROUP 2161 if (rq->bio && rq->bio->bi_blkg) { 2162 cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 2163 #ifdef CONFIG_MEMCG 2164 cmd->memcg_css = 2165 cgroup_get_e_css(cmd->blkcg_css->cgroup, 2166 &memory_cgrp_subsys); 2167 #endif 2168 } 2169 #endif 2170 loop_queue_work(lo, cmd); 2171 2172 return BLK_STS_OK; 2173 } 2174 2175 static void loop_handle_cmd(struct loop_cmd *cmd) 2176 { 2177 struct request *rq = blk_mq_rq_from_pdu(cmd); 2178 const bool write = op_is_write(req_op(rq)); 2179 struct loop_device *lo = rq->q->queuedata; 2180 int ret = 0; 2181 struct mem_cgroup *old_memcg = NULL; 2182 2183 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 2184 ret = -EIO; 2185 goto failed; 2186 } 2187 2188 if (cmd->blkcg_css) 2189 kthread_associate_blkcg(cmd->blkcg_css); 2190 if (cmd->memcg_css) 2191 old_memcg = set_active_memcg( 2192 mem_cgroup_from_css(cmd->memcg_css)); 2193 2194 ret = do_req_filebacked(lo, rq); 2195 2196 if (cmd->blkcg_css) 2197 kthread_associate_blkcg(NULL); 2198 2199 if (cmd->memcg_css) { 2200 set_active_memcg(old_memcg); 2201 css_put(cmd->memcg_css); 2202 } 2203 failed: 2204 /* complete non-aio request */ 2205 if (!cmd->use_aio || ret) { 2206 if (ret == -EOPNOTSUPP) 2207 cmd->ret = ret; 2208 else 2209 cmd->ret = ret ? -EIO : 0; 2210 if (likely(!blk_should_fake_timeout(rq->q))) 2211 blk_mq_complete_request(rq); 2212 } 2213 } 2214 2215 static void loop_set_timer(struct loop_device *lo) 2216 { 2217 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 2218 } 2219 2220 static void loop_process_work(struct loop_worker *worker, 2221 struct list_head *cmd_list, struct loop_device *lo) 2222 { 2223 int orig_flags = current->flags; 2224 struct loop_cmd *cmd; 2225 2226 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 2227 spin_lock_irq(&lo->lo_work_lock); 2228 while (!list_empty(cmd_list)) { 2229 cmd = container_of( 2230 cmd_list->next, struct loop_cmd, list_entry); 2231 list_del(cmd_list->next); 2232 spin_unlock_irq(&lo->lo_work_lock); 2233 2234 loop_handle_cmd(cmd); 2235 cond_resched(); 2236 2237 spin_lock_irq(&lo->lo_work_lock); 2238 } 2239 2240 /* 2241 * We only add to the idle list if there are no pending cmds 2242 * *and* the worker will not run again which ensures that it 2243 * is safe to free any worker on the idle list 2244 */ 2245 if (worker && !work_pending(&worker->work)) { 2246 worker->last_ran_at = jiffies; 2247 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 2248 loop_set_timer(lo); 2249 } 2250 spin_unlock_irq(&lo->lo_work_lock); 2251 current->flags = orig_flags; 2252 } 2253 2254 static void loop_workfn(struct work_struct *work) 2255 { 2256 struct loop_worker *worker = 2257 container_of(work, struct loop_worker, work); 2258 loop_process_work(worker, &worker->cmd_list, worker->lo); 2259 } 2260 2261 static void loop_rootcg_workfn(struct work_struct *work) 2262 { 2263 struct loop_device *lo = 2264 container_of(work, struct loop_device, rootcg_work); 2265 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 2266 } 2267 2268 static void loop_free_idle_workers(struct timer_list *timer) 2269 { 2270 struct loop_device *lo = container_of(timer, struct loop_device, timer); 2271 struct loop_worker *pos, *worker; 2272 2273 spin_lock_irq(&lo->lo_work_lock); 2274 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 2275 idle_list) { 2276 if (time_is_after_jiffies(worker->last_ran_at + 2277 LOOP_IDLE_WORKER_TIMEOUT)) 2278 break; 2279 list_del(&worker->idle_list); 2280 rb_erase(&worker->rb_node, &lo->worker_tree); 2281 css_put(worker->blkcg_css); 2282 kfree(worker); 2283 } 2284 if (!list_empty(&lo->idle_worker_list)) 2285 loop_set_timer(lo); 2286 spin_unlock_irq(&lo->lo_work_lock); 2287 } 2288 2289 static const struct blk_mq_ops loop_mq_ops = { 2290 .queue_rq = loop_queue_rq, 2291 .complete = lo_complete_rq, 2292 }; 2293 2294 static int loop_add(int i) 2295 { 2296 struct loop_device *lo; 2297 struct gendisk *disk; 2298 int err; 2299 2300 err = -ENOMEM; 2301 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 2302 if (!lo) 2303 goto out; 2304 lo->lo_state = Lo_unbound; 2305 2306 err = mutex_lock_killable(&loop_ctl_mutex); 2307 if (err) 2308 goto out_free_dev; 2309 2310 /* allocate id, if @id >= 0, we're requesting that specific id */ 2311 if (i >= 0) { 2312 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2313 if (err == -ENOSPC) 2314 err = -EEXIST; 2315 } else { 2316 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2317 } 2318 mutex_unlock(&loop_ctl_mutex); 2319 if (err < 0) 2320 goto out_free_dev; 2321 i = err; 2322 2323 err = -ENOMEM; 2324 lo->tag_set.ops = &loop_mq_ops; 2325 lo->tag_set.nr_hw_queues = 1; 2326 lo->tag_set.queue_depth = 128; 2327 lo->tag_set.numa_node = NUMA_NO_NODE; 2328 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2329 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | 2330 BLK_MQ_F_NO_SCHED_BY_DEFAULT; 2331 lo->tag_set.driver_data = lo; 2332 2333 err = blk_mq_alloc_tag_set(&lo->tag_set); 2334 if (err) 2335 goto out_free_idr; 2336 2337 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 2338 if (IS_ERR(disk)) { 2339 err = PTR_ERR(disk); 2340 goto out_cleanup_tags; 2341 } 2342 lo->lo_queue = lo->lo_disk->queue; 2343 2344 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2345 2346 /* 2347 * By default, we do buffer IO, so it doesn't make sense to enable 2348 * merge because the I/O submitted to backing file is handled page by 2349 * page. For directio mode, merge does help to dispatch bigger request 2350 * to underlayer disk. We will enable merge once directio is enabled. 2351 */ 2352 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2353 2354 /* 2355 * Disable partition scanning by default. The in-kernel partition 2356 * scanning can be requested individually per-device during its 2357 * setup. Userspace can always add and remove partitions from all 2358 * devices. The needed partition minors are allocated from the 2359 * extended minor space, the main loop device numbers will continue 2360 * to match the loop minors, regardless of the number of partitions 2361 * used. 2362 * 2363 * If max_part is given, partition scanning is globally enabled for 2364 * all loop devices. The minors for the main loop devices will be 2365 * multiples of max_part. 2366 * 2367 * Note: Global-for-all-devices, set-only-at-init, read-only module 2368 * parameteters like 'max_loop' and 'max_part' make things needlessly 2369 * complicated, are too static, inflexible and may surprise 2370 * userspace tools. Parameters like this in general should be avoided. 2371 */ 2372 if (!part_shift) 2373 disk->flags |= GENHD_FL_NO_PART_SCAN; 2374 disk->flags |= GENHD_FL_EXT_DEVT; 2375 atomic_set(&lo->lo_refcnt, 0); 2376 mutex_init(&lo->lo_mutex); 2377 lo->lo_number = i; 2378 spin_lock_init(&lo->lo_lock); 2379 spin_lock_init(&lo->lo_work_lock); 2380 disk->major = LOOP_MAJOR; 2381 disk->first_minor = i << part_shift; 2382 disk->minors = 1 << part_shift; 2383 disk->fops = &lo_fops; 2384 disk->private_data = lo; 2385 disk->queue = lo->lo_queue; 2386 disk->events = DISK_EVENT_MEDIA_CHANGE; 2387 disk->event_flags = DISK_EVENT_FLAG_UEVENT; 2388 sprintf(disk->disk_name, "loop%d", i); 2389 /* Make this loop device reachable from pathname. */ 2390 add_disk(disk); 2391 /* Show this loop device. */ 2392 mutex_lock(&loop_ctl_mutex); 2393 lo->idr_visible = true; 2394 mutex_unlock(&loop_ctl_mutex); 2395 return i; 2396 2397 out_cleanup_tags: 2398 blk_mq_free_tag_set(&lo->tag_set); 2399 out_free_idr: 2400 mutex_lock(&loop_ctl_mutex); 2401 idr_remove(&loop_index_idr, i); 2402 mutex_unlock(&loop_ctl_mutex); 2403 out_free_dev: 2404 kfree(lo); 2405 out: 2406 return err; 2407 } 2408 2409 static void loop_remove(struct loop_device *lo) 2410 { 2411 /* Make this loop device unreachable from pathname. */ 2412 del_gendisk(lo->lo_disk); 2413 blk_cleanup_disk(lo->lo_disk); 2414 blk_mq_free_tag_set(&lo->tag_set); 2415 mutex_lock(&loop_ctl_mutex); 2416 idr_remove(&loop_index_idr, lo->lo_number); 2417 mutex_unlock(&loop_ctl_mutex); 2418 /* There is no route which can find this loop device. */ 2419 mutex_destroy(&lo->lo_mutex); 2420 kfree(lo); 2421 } 2422 2423 static void loop_probe(dev_t dev) 2424 { 2425 int idx = MINOR(dev) >> part_shift; 2426 2427 if (max_loop && idx >= max_loop) 2428 return; 2429 loop_add(idx); 2430 } 2431 2432 static int loop_control_remove(int idx) 2433 { 2434 struct loop_device *lo; 2435 int ret; 2436 2437 if (idx < 0) { 2438 pr_warn_once("deleting an unspecified loop device is not supported.\n"); 2439 return -EINVAL; 2440 } 2441 2442 /* Hide this loop device for serialization. */ 2443 ret = mutex_lock_killable(&loop_ctl_mutex); 2444 if (ret) 2445 return ret; 2446 lo = idr_find(&loop_index_idr, idx); 2447 if (!lo || !lo->idr_visible) 2448 ret = -ENODEV; 2449 else 2450 lo->idr_visible = false; 2451 mutex_unlock(&loop_ctl_mutex); 2452 if (ret) 2453 return ret; 2454 2455 /* Check whether this loop device can be removed. */ 2456 ret = mutex_lock_killable(&lo->lo_mutex); 2457 if (ret) 2458 goto mark_visible; 2459 if (lo->lo_state != Lo_unbound || 2460 atomic_read(&lo->lo_refcnt) > 0) { 2461 mutex_unlock(&lo->lo_mutex); 2462 ret = -EBUSY; 2463 goto mark_visible; 2464 } 2465 /* Mark this loop device no longer open()-able. */ 2466 lo->lo_state = Lo_deleting; 2467 mutex_unlock(&lo->lo_mutex); 2468 2469 loop_remove(lo); 2470 return 0; 2471 2472 mark_visible: 2473 /* Show this loop device again. */ 2474 mutex_lock(&loop_ctl_mutex); 2475 lo->idr_visible = true; 2476 mutex_unlock(&loop_ctl_mutex); 2477 return ret; 2478 } 2479 2480 static int loop_control_get_free(int idx) 2481 { 2482 struct loop_device *lo; 2483 int id, ret; 2484 2485 ret = mutex_lock_killable(&loop_ctl_mutex); 2486 if (ret) 2487 return ret; 2488 idr_for_each_entry(&loop_index_idr, lo, id) { 2489 /* Hitting a race results in creating a new loop device which is harmless. */ 2490 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) 2491 goto found; 2492 } 2493 mutex_unlock(&loop_ctl_mutex); 2494 return loop_add(-1); 2495 found: 2496 mutex_unlock(&loop_ctl_mutex); 2497 return id; 2498 } 2499 2500 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2501 unsigned long parm) 2502 { 2503 switch (cmd) { 2504 case LOOP_CTL_ADD: 2505 return loop_add(parm); 2506 case LOOP_CTL_REMOVE: 2507 return loop_control_remove(parm); 2508 case LOOP_CTL_GET_FREE: 2509 return loop_control_get_free(parm); 2510 default: 2511 return -ENOSYS; 2512 } 2513 } 2514 2515 static const struct file_operations loop_ctl_fops = { 2516 .open = nonseekable_open, 2517 .unlocked_ioctl = loop_control_ioctl, 2518 .compat_ioctl = loop_control_ioctl, 2519 .owner = THIS_MODULE, 2520 .llseek = noop_llseek, 2521 }; 2522 2523 static struct miscdevice loop_misc = { 2524 .minor = LOOP_CTRL_MINOR, 2525 .name = "loop-control", 2526 .fops = &loop_ctl_fops, 2527 }; 2528 2529 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2530 MODULE_ALIAS("devname:loop-control"); 2531 2532 static int __init loop_init(void) 2533 { 2534 int i, nr; 2535 int err; 2536 2537 part_shift = 0; 2538 if (max_part > 0) { 2539 part_shift = fls(max_part); 2540 2541 /* 2542 * Adjust max_part according to part_shift as it is exported 2543 * to user space so that user can decide correct minor number 2544 * if [s]he want to create more devices. 2545 * 2546 * Note that -1 is required because partition 0 is reserved 2547 * for the whole disk. 2548 */ 2549 max_part = (1UL << part_shift) - 1; 2550 } 2551 2552 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2553 err = -EINVAL; 2554 goto err_out; 2555 } 2556 2557 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2558 err = -EINVAL; 2559 goto err_out; 2560 } 2561 2562 /* 2563 * If max_loop is specified, create that many devices upfront. 2564 * This also becomes a hard limit. If max_loop is not specified, 2565 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2566 * init time. Loop devices can be requested on-demand with the 2567 * /dev/loop-control interface, or be instantiated by accessing 2568 * a 'dead' device node. 2569 */ 2570 if (max_loop) 2571 nr = max_loop; 2572 else 2573 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2574 2575 err = misc_register(&loop_misc); 2576 if (err < 0) 2577 goto err_out; 2578 2579 2580 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2581 err = -EIO; 2582 goto misc_out; 2583 } 2584 2585 /* pre-create number of devices given by config or max_loop */ 2586 for (i = 0; i < nr; i++) 2587 loop_add(i); 2588 2589 printk(KERN_INFO "loop: module loaded\n"); 2590 return 0; 2591 2592 misc_out: 2593 misc_deregister(&loop_misc); 2594 err_out: 2595 return err; 2596 } 2597 2598 static void __exit loop_exit(void) 2599 { 2600 struct loop_device *lo; 2601 int id; 2602 2603 unregister_blkdev(LOOP_MAJOR, "loop"); 2604 misc_deregister(&loop_misc); 2605 2606 /* 2607 * There is no need to use loop_ctl_mutex here, for nobody else can 2608 * access loop_index_idr when this module is unloading (unless forced 2609 * module unloading is requested). If this is not a clean unloading, 2610 * we have no means to avoid kernel crash. 2611 */ 2612 idr_for_each_entry(&loop_index_idr, lo, id) 2613 loop_remove(lo); 2614 2615 idr_destroy(&loop_index_idr); 2616 } 2617 2618 module_init(loop_init); 2619 module_exit(loop_exit); 2620 2621 #ifndef MODULE 2622 static int __init max_loop_setup(char *str) 2623 { 2624 max_loop = simple_strtol(str, NULL, 0); 2625 return 1; 2626 } 2627 2628 __setup("max_loop=", max_loop_setup); 2629 #endif 2630