1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/drivers/block/loop.c 4 * 5 * Written by Theodore Ts'o, 3/29/93 6 * 7 * Copyright 1993 by Theodore Ts'o. 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/blkpg.h> 63 #include <linux/init.h> 64 #include <linux/swap.h> 65 #include <linux/slab.h> 66 #include <linux/compat.h> 67 #include <linux/suspend.h> 68 #include <linux/freezer.h> 69 #include <linux/mutex.h> 70 #include <linux/writeback.h> 71 #include <linux/completion.h> 72 #include <linux/highmem.h> 73 #include <linux/splice.h> 74 #include <linux/sysfs.h> 75 #include <linux/miscdevice.h> 76 #include <linux/falloc.h> 77 #include <linux/uio.h> 78 #include <linux/ioprio.h> 79 #include <linux/blk-cgroup.h> 80 #include <linux/sched/mm.h> 81 #include <linux/statfs.h> 82 #include <linux/uaccess.h> 83 #include <linux/blk-mq.h> 84 #include <linux/spinlock.h> 85 #include <uapi/linux/loop.h> 86 87 /* Possible states of device */ 88 enum { 89 Lo_unbound, 90 Lo_bound, 91 Lo_rundown, 92 Lo_deleting, 93 }; 94 95 struct loop_func_table; 96 97 struct loop_device { 98 int lo_number; 99 loff_t lo_offset; 100 loff_t lo_sizelimit; 101 int lo_flags; 102 char lo_file_name[LO_NAME_SIZE]; 103 104 struct file * lo_backing_file; 105 struct block_device *lo_device; 106 107 gfp_t old_gfp_mask; 108 109 spinlock_t lo_lock; 110 int lo_state; 111 spinlock_t lo_work_lock; 112 struct workqueue_struct *workqueue; 113 struct work_struct rootcg_work; 114 struct list_head rootcg_cmd_list; 115 struct list_head idle_worker_list; 116 struct rb_root worker_tree; 117 struct timer_list timer; 118 bool use_dio; 119 bool sysfs_inited; 120 121 struct request_queue *lo_queue; 122 struct blk_mq_tag_set tag_set; 123 struct gendisk *lo_disk; 124 struct mutex lo_mutex; 125 bool idr_visible; 126 }; 127 128 struct loop_cmd { 129 struct list_head list_entry; 130 bool use_aio; /* use AIO interface to handle I/O */ 131 atomic_t ref; /* only for aio */ 132 long ret; 133 struct kiocb iocb; 134 struct bio_vec *bvec; 135 struct cgroup_subsys_state *blkcg_css; 136 struct cgroup_subsys_state *memcg_css; 137 }; 138 139 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) 140 #define LOOP_DEFAULT_HW_Q_DEPTH (128) 141 142 static DEFINE_IDR(loop_index_idr); 143 static DEFINE_MUTEX(loop_ctl_mutex); 144 static DEFINE_MUTEX(loop_validate_mutex); 145 146 /** 147 * loop_global_lock_killable() - take locks for safe loop_validate_file() test 148 * 149 * @lo: struct loop_device 150 * @global: true if @lo is about to bind another "struct loop_device", false otherwise 151 * 152 * Returns 0 on success, -EINTR otherwise. 153 * 154 * Since loop_validate_file() traverses on other "struct loop_device" if 155 * is_loop_device() is true, we need a global lock for serializing concurrent 156 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls. 157 */ 158 static int loop_global_lock_killable(struct loop_device *lo, bool global) 159 { 160 int err; 161 162 if (global) { 163 err = mutex_lock_killable(&loop_validate_mutex); 164 if (err) 165 return err; 166 } 167 err = mutex_lock_killable(&lo->lo_mutex); 168 if (err && global) 169 mutex_unlock(&loop_validate_mutex); 170 return err; 171 } 172 173 /** 174 * loop_global_unlock() - release locks taken by loop_global_lock_killable() 175 * 176 * @lo: struct loop_device 177 * @global: true if @lo was about to bind another "struct loop_device", false otherwise 178 */ 179 static void loop_global_unlock(struct loop_device *lo, bool global) 180 { 181 mutex_unlock(&lo->lo_mutex); 182 if (global) 183 mutex_unlock(&loop_validate_mutex); 184 } 185 186 static int max_part; 187 static int part_shift; 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. 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 use_dio = true; 243 else 244 use_dio = false; 245 } else { 246 use_dio = false; 247 } 248 249 if (lo->use_dio == use_dio) 250 return; 251 252 /* flush dirty pages before changing direct IO */ 253 vfs_fsync(file, 0); 254 255 /* 256 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 257 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 258 * will get updated by ioctl(LOOP_GET_STATUS) 259 */ 260 if (lo->lo_state == Lo_bound) 261 blk_mq_freeze_queue(lo->lo_queue); 262 lo->use_dio = use_dio; 263 if (use_dio) { 264 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 265 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 266 } else { 267 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 268 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 269 } 270 if (lo->lo_state == Lo_bound) 271 blk_mq_unfreeze_queue(lo->lo_queue); 272 } 273 274 /** 275 * loop_set_size() - sets device size and notifies userspace 276 * @lo: struct loop_device to set the size for 277 * @size: new size of the loop device 278 * 279 * Callers must validate that the size passed into this function fits into 280 * a sector_t, eg using loop_validate_size() 281 */ 282 static void loop_set_size(struct loop_device *lo, loff_t size) 283 { 284 if (!set_capacity_and_notify(lo->lo_disk, size)) 285 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 286 } 287 288 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 289 { 290 struct iov_iter i; 291 ssize_t bw; 292 293 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 294 295 file_start_write(file); 296 bw = vfs_iter_write(file, &i, ppos, 0); 297 file_end_write(file); 298 299 if (likely(bw == bvec->bv_len)) 300 return 0; 301 302 printk_ratelimited(KERN_ERR 303 "loop: Write error at byte offset %llu, length %i.\n", 304 (unsigned long long)*ppos, bvec->bv_len); 305 if (bw >= 0) 306 bw = -EIO; 307 return bw; 308 } 309 310 static int lo_write_simple(struct loop_device *lo, struct request *rq, 311 loff_t pos) 312 { 313 struct bio_vec bvec; 314 struct req_iterator iter; 315 int ret = 0; 316 317 rq_for_each_segment(bvec, rq, iter) { 318 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 319 if (ret < 0) 320 break; 321 cond_resched(); 322 } 323 324 return ret; 325 } 326 327 static int lo_read_simple(struct loop_device *lo, struct request *rq, 328 loff_t pos) 329 { 330 struct bio_vec bvec; 331 struct req_iterator iter; 332 struct iov_iter i; 333 ssize_t len; 334 335 rq_for_each_segment(bvec, rq, iter) { 336 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 337 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 338 if (len < 0) 339 return len; 340 341 flush_dcache_page(bvec.bv_page); 342 343 if (len != bvec.bv_len) { 344 struct bio *bio; 345 346 __rq_for_each_bio(bio, rq) 347 zero_fill_bio(bio); 348 break; 349 } 350 cond_resched(); 351 } 352 353 return 0; 354 } 355 356 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 357 int mode) 358 { 359 /* 360 * We use fallocate to manipulate the space mappings used by the image 361 * a.k.a. discard/zerorange. 362 */ 363 struct file *file = lo->lo_backing_file; 364 int ret; 365 366 mode |= FALLOC_FL_KEEP_SIZE; 367 368 if (!bdev_max_discard_sectors(lo->lo_device)) 369 return -EOPNOTSUPP; 370 371 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 372 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 373 return -EIO; 374 return ret; 375 } 376 377 static int lo_req_flush(struct loop_device *lo, struct request *rq) 378 { 379 int ret = vfs_fsync(lo->lo_backing_file, 0); 380 if (unlikely(ret && ret != -EINVAL)) 381 ret = -EIO; 382 383 return ret; 384 } 385 386 static void lo_complete_rq(struct request *rq) 387 { 388 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 389 blk_status_t ret = BLK_STS_OK; 390 391 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 392 req_op(rq) != REQ_OP_READ) { 393 if (cmd->ret < 0) 394 ret = errno_to_blk_status(cmd->ret); 395 goto end_io; 396 } 397 398 /* 399 * Short READ - if we got some data, advance our request and 400 * retry it. If we got no data, end the rest with EIO. 401 */ 402 if (cmd->ret) { 403 blk_update_request(rq, BLK_STS_OK, cmd->ret); 404 cmd->ret = 0; 405 blk_mq_requeue_request(rq, true); 406 } else { 407 if (cmd->use_aio) { 408 struct bio *bio = rq->bio; 409 410 while (bio) { 411 zero_fill_bio(bio); 412 bio = bio->bi_next; 413 } 414 } 415 ret = BLK_STS_IOERR; 416 end_io: 417 blk_mq_end_request(rq, ret); 418 } 419 } 420 421 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 422 { 423 struct request *rq = blk_mq_rq_from_pdu(cmd); 424 425 if (!atomic_dec_and_test(&cmd->ref)) 426 return; 427 kfree(cmd->bvec); 428 cmd->bvec = NULL; 429 if (likely(!blk_should_fake_timeout(rq->q))) 430 blk_mq_complete_request(rq); 431 } 432 433 static void lo_rw_aio_complete(struct kiocb *iocb, long ret) 434 { 435 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 436 437 cmd->ret = ret; 438 lo_rw_aio_do_completion(cmd); 439 } 440 441 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 442 loff_t pos, bool rw) 443 { 444 struct iov_iter iter; 445 struct req_iterator rq_iter; 446 struct bio_vec *bvec; 447 struct request *rq = blk_mq_rq_from_pdu(cmd); 448 struct bio *bio = rq->bio; 449 struct file *file = lo->lo_backing_file; 450 struct bio_vec tmp; 451 unsigned int offset; 452 int nr_bvec = 0; 453 int ret; 454 455 rq_for_each_bvec(tmp, rq, rq_iter) 456 nr_bvec++; 457 458 if (rq->bio != rq->biotail) { 459 460 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 461 GFP_NOIO); 462 if (!bvec) 463 return -EIO; 464 cmd->bvec = bvec; 465 466 /* 467 * The bios of the request may be started from the middle of 468 * the 'bvec' because of bio splitting, so we can't directly 469 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 470 * API will take care of all details for us. 471 */ 472 rq_for_each_bvec(tmp, rq, rq_iter) { 473 *bvec = tmp; 474 bvec++; 475 } 476 bvec = cmd->bvec; 477 offset = 0; 478 } else { 479 /* 480 * Same here, this bio may be started from the middle of the 481 * 'bvec' because of bio splitting, so offset from the bvec 482 * must be passed to iov iterator 483 */ 484 offset = bio->bi_iter.bi_bvec_done; 485 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 486 } 487 atomic_set(&cmd->ref, 2); 488 489 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 490 iter.iov_offset = offset; 491 492 cmd->iocb.ki_pos = pos; 493 cmd->iocb.ki_filp = file; 494 cmd->iocb.ki_complete = lo_rw_aio_complete; 495 cmd->iocb.ki_flags = IOCB_DIRECT; 496 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 497 498 if (rw == WRITE) 499 ret = call_write_iter(file, &cmd->iocb, &iter); 500 else 501 ret = call_read_iter(file, &cmd->iocb, &iter); 502 503 lo_rw_aio_do_completion(cmd); 504 505 if (ret != -EIOCBQUEUED) 506 lo_rw_aio_complete(&cmd->iocb, ret); 507 return 0; 508 } 509 510 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 511 { 512 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 513 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 514 515 /* 516 * lo_write_simple and lo_read_simple should have been covered 517 * by io submit style function like lo_rw_aio(), one blocker 518 * is that lo_read_simple() need to call flush_dcache_page after 519 * the page is written from kernel, and it isn't easy to handle 520 * this in io submit style function which submits all segments 521 * of the req at one time. And direct read IO doesn't need to 522 * run flush_dcache_page(). 523 */ 524 switch (req_op(rq)) { 525 case REQ_OP_FLUSH: 526 return lo_req_flush(lo, rq); 527 case REQ_OP_WRITE_ZEROES: 528 /* 529 * If the caller doesn't want deallocation, call zeroout to 530 * write zeroes the range. Otherwise, punch them out. 531 */ 532 return lo_fallocate(lo, rq, pos, 533 (rq->cmd_flags & REQ_NOUNMAP) ? 534 FALLOC_FL_ZERO_RANGE : 535 FALLOC_FL_PUNCH_HOLE); 536 case REQ_OP_DISCARD: 537 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 538 case REQ_OP_WRITE: 539 if (cmd->use_aio) 540 return lo_rw_aio(lo, cmd, pos, WRITE); 541 else 542 return lo_write_simple(lo, rq, pos); 543 case REQ_OP_READ: 544 if (cmd->use_aio) 545 return lo_rw_aio(lo, cmd, pos, READ); 546 else 547 return lo_read_simple(lo, rq, pos); 548 default: 549 WARN_ON_ONCE(1); 550 return -EIO; 551 } 552 } 553 554 static inline void loop_update_dio(struct loop_device *lo) 555 { 556 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 557 lo->use_dio); 558 } 559 560 static void loop_reread_partitions(struct loop_device *lo) 561 { 562 int rc; 563 564 mutex_lock(&lo->lo_disk->open_mutex); 565 rc = bdev_disk_changed(lo->lo_disk, false); 566 mutex_unlock(&lo->lo_disk->open_mutex); 567 if (rc) 568 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 569 __func__, lo->lo_number, lo->lo_file_name, rc); 570 } 571 572 static inline int is_loop_device(struct file *file) 573 { 574 struct inode *i = file->f_mapping->host; 575 576 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; 577 } 578 579 static int loop_validate_file(struct file *file, struct block_device *bdev) 580 { 581 struct inode *inode = file->f_mapping->host; 582 struct file *f = file; 583 584 /* Avoid recursion */ 585 while (is_loop_device(f)) { 586 struct loop_device *l; 587 588 lockdep_assert_held(&loop_validate_mutex); 589 if (f->f_mapping->host->i_rdev == bdev->bd_dev) 590 return -EBADF; 591 592 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 593 if (l->lo_state != Lo_bound) 594 return -EINVAL; 595 /* Order wrt setting lo->lo_backing_file in loop_configure(). */ 596 rmb(); 597 f = l->lo_backing_file; 598 } 599 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 600 return -EINVAL; 601 return 0; 602 } 603 604 /* 605 * loop_change_fd switched the backing store of a loopback device to 606 * a new file. This is useful for operating system installers to free up 607 * the original file and in High Availability environments to switch to 608 * an alternative location for the content in case of server meltdown. 609 * This can only work if the loop device is used read-only, and if the 610 * new backing store is the same size and type as the old backing store. 611 */ 612 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 613 unsigned int arg) 614 { 615 struct file *file = fget(arg); 616 struct file *old_file; 617 int error; 618 bool partscan; 619 bool is_loop; 620 621 if (!file) 622 return -EBADF; 623 624 /* suppress uevents while reconfiguring the device */ 625 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1); 626 627 is_loop = is_loop_device(file); 628 error = loop_global_lock_killable(lo, is_loop); 629 if (error) 630 goto out_putf; 631 error = -ENXIO; 632 if (lo->lo_state != Lo_bound) 633 goto out_err; 634 635 /* the loop device has to be read-only */ 636 error = -EINVAL; 637 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 638 goto out_err; 639 640 error = loop_validate_file(file, bdev); 641 if (error) 642 goto out_err; 643 644 old_file = lo->lo_backing_file; 645 646 error = -EINVAL; 647 648 /* size of the new backing store needs to be the same */ 649 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 650 goto out_err; 651 652 /* and ... switch */ 653 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 654 blk_mq_freeze_queue(lo->lo_queue); 655 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 656 lo->lo_backing_file = file; 657 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 658 mapping_set_gfp_mask(file->f_mapping, 659 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 660 loop_update_dio(lo); 661 blk_mq_unfreeze_queue(lo->lo_queue); 662 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 663 loop_global_unlock(lo, is_loop); 664 665 /* 666 * Flush loop_validate_file() before fput(), for l->lo_backing_file 667 * might be pointing at old_file which might be the last reference. 668 */ 669 if (!is_loop) { 670 mutex_lock(&loop_validate_mutex); 671 mutex_unlock(&loop_validate_mutex); 672 } 673 /* 674 * We must drop file reference outside of lo_mutex as dropping 675 * the file ref can take open_mutex which creates circular locking 676 * dependency. 677 */ 678 fput(old_file); 679 if (partscan) 680 loop_reread_partitions(lo); 681 682 error = 0; 683 done: 684 /* enable and uncork uevent now that we are done */ 685 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0); 686 return error; 687 688 out_err: 689 loop_global_unlock(lo, is_loop); 690 out_putf: 691 fput(file); 692 goto done; 693 } 694 695 /* loop sysfs attributes */ 696 697 static ssize_t loop_attr_show(struct device *dev, char *page, 698 ssize_t (*callback)(struct loop_device *, char *)) 699 { 700 struct gendisk *disk = dev_to_disk(dev); 701 struct loop_device *lo = disk->private_data; 702 703 return callback(lo, page); 704 } 705 706 #define LOOP_ATTR_RO(_name) \ 707 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 708 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 709 struct device_attribute *attr, char *b) \ 710 { \ 711 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 712 } \ 713 static struct device_attribute loop_attr_##_name = \ 714 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 715 716 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 717 { 718 ssize_t ret; 719 char *p = NULL; 720 721 spin_lock_irq(&lo->lo_lock); 722 if (lo->lo_backing_file) 723 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 724 spin_unlock_irq(&lo->lo_lock); 725 726 if (IS_ERR_OR_NULL(p)) 727 ret = PTR_ERR(p); 728 else { 729 ret = strlen(p); 730 memmove(buf, p, ret); 731 buf[ret++] = '\n'; 732 buf[ret] = 0; 733 } 734 735 return ret; 736 } 737 738 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 739 { 740 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset); 741 } 742 743 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 744 { 745 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 746 } 747 748 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 749 { 750 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 751 752 return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0"); 753 } 754 755 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 756 { 757 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 758 759 return sysfs_emit(buf, "%s\n", partscan ? "1" : "0"); 760 } 761 762 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 763 { 764 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 765 766 return sysfs_emit(buf, "%s\n", dio ? "1" : "0"); 767 } 768 769 LOOP_ATTR_RO(backing_file); 770 LOOP_ATTR_RO(offset); 771 LOOP_ATTR_RO(sizelimit); 772 LOOP_ATTR_RO(autoclear); 773 LOOP_ATTR_RO(partscan); 774 LOOP_ATTR_RO(dio); 775 776 static struct attribute *loop_attrs[] = { 777 &loop_attr_backing_file.attr, 778 &loop_attr_offset.attr, 779 &loop_attr_sizelimit.attr, 780 &loop_attr_autoclear.attr, 781 &loop_attr_partscan.attr, 782 &loop_attr_dio.attr, 783 NULL, 784 }; 785 786 static struct attribute_group loop_attribute_group = { 787 .name = "loop", 788 .attrs= loop_attrs, 789 }; 790 791 static void loop_sysfs_init(struct loop_device *lo) 792 { 793 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 794 &loop_attribute_group); 795 } 796 797 static void loop_sysfs_exit(struct loop_device *lo) 798 { 799 if (lo->sysfs_inited) 800 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 801 &loop_attribute_group); 802 } 803 804 static void loop_config_discard(struct loop_device *lo) 805 { 806 struct file *file = lo->lo_backing_file; 807 struct inode *inode = file->f_mapping->host; 808 struct request_queue *q = lo->lo_queue; 809 u32 granularity, max_discard_sectors; 810 811 /* 812 * If the backing device is a block device, mirror its zeroing 813 * capability. Set the discard sectors to the block device's zeroing 814 * capabilities because loop discards result in blkdev_issue_zeroout(), 815 * not blkdev_issue_discard(). This maintains consistent behavior with 816 * file-backed loop devices: discarded regions read back as zero. 817 */ 818 if (S_ISBLK(inode->i_mode)) { 819 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 820 821 max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 822 granularity = bdev_discard_granularity(I_BDEV(inode)) ?: 823 queue_physical_block_size(backingq); 824 825 /* 826 * We use punch hole to reclaim the free space used by the 827 * image a.k.a. discard. 828 */ 829 } else if (!file->f_op->fallocate) { 830 max_discard_sectors = 0; 831 granularity = 0; 832 833 } else { 834 struct kstatfs sbuf; 835 836 max_discard_sectors = UINT_MAX >> 9; 837 if (!vfs_statfs(&file->f_path, &sbuf)) 838 granularity = sbuf.f_bsize; 839 else 840 max_discard_sectors = 0; 841 } 842 843 if (max_discard_sectors) { 844 q->limits.discard_granularity = granularity; 845 blk_queue_max_discard_sectors(q, max_discard_sectors); 846 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 847 } else { 848 q->limits.discard_granularity = 0; 849 blk_queue_max_discard_sectors(q, 0); 850 blk_queue_max_write_zeroes_sectors(q, 0); 851 } 852 } 853 854 struct loop_worker { 855 struct rb_node rb_node; 856 struct work_struct work; 857 struct list_head cmd_list; 858 struct list_head idle_list; 859 struct loop_device *lo; 860 struct cgroup_subsys_state *blkcg_css; 861 unsigned long last_ran_at; 862 }; 863 864 static void loop_workfn(struct work_struct *work); 865 866 #ifdef CONFIG_BLK_CGROUP 867 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 868 { 869 return !css || css == blkcg_root_css; 870 } 871 #else 872 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 873 { 874 return !css; 875 } 876 #endif 877 878 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 879 { 880 struct rb_node **node, *parent = NULL; 881 struct loop_worker *cur_worker, *worker = NULL; 882 struct work_struct *work; 883 struct list_head *cmd_list; 884 885 spin_lock_irq(&lo->lo_work_lock); 886 887 if (queue_on_root_worker(cmd->blkcg_css)) 888 goto queue_work; 889 890 node = &lo->worker_tree.rb_node; 891 892 while (*node) { 893 parent = *node; 894 cur_worker = container_of(*node, struct loop_worker, rb_node); 895 if (cur_worker->blkcg_css == cmd->blkcg_css) { 896 worker = cur_worker; 897 break; 898 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { 899 node = &(*node)->rb_left; 900 } else { 901 node = &(*node)->rb_right; 902 } 903 } 904 if (worker) 905 goto queue_work; 906 907 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 908 /* 909 * In the event we cannot allocate a worker, just queue on the 910 * rootcg worker and issue the I/O as the rootcg 911 */ 912 if (!worker) { 913 cmd->blkcg_css = NULL; 914 if (cmd->memcg_css) 915 css_put(cmd->memcg_css); 916 cmd->memcg_css = NULL; 917 goto queue_work; 918 } 919 920 worker->blkcg_css = cmd->blkcg_css; 921 css_get(worker->blkcg_css); 922 INIT_WORK(&worker->work, loop_workfn); 923 INIT_LIST_HEAD(&worker->cmd_list); 924 INIT_LIST_HEAD(&worker->idle_list); 925 worker->lo = lo; 926 rb_link_node(&worker->rb_node, parent, node); 927 rb_insert_color(&worker->rb_node, &lo->worker_tree); 928 queue_work: 929 if (worker) { 930 /* 931 * We need to remove from the idle list here while 932 * holding the lock so that the idle timer doesn't 933 * free the worker 934 */ 935 if (!list_empty(&worker->idle_list)) 936 list_del_init(&worker->idle_list); 937 work = &worker->work; 938 cmd_list = &worker->cmd_list; 939 } else { 940 work = &lo->rootcg_work; 941 cmd_list = &lo->rootcg_cmd_list; 942 } 943 list_add_tail(&cmd->list_entry, cmd_list); 944 queue_work(lo->workqueue, work); 945 spin_unlock_irq(&lo->lo_work_lock); 946 } 947 948 static void loop_set_timer(struct loop_device *lo) 949 { 950 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 951 } 952 953 static void loop_free_idle_workers(struct loop_device *lo, bool delete_all) 954 { 955 struct loop_worker *pos, *worker; 956 957 spin_lock_irq(&lo->lo_work_lock); 958 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 959 idle_list) { 960 if (!delete_all && 961 time_is_after_jiffies(worker->last_ran_at + 962 LOOP_IDLE_WORKER_TIMEOUT)) 963 break; 964 list_del(&worker->idle_list); 965 rb_erase(&worker->rb_node, &lo->worker_tree); 966 css_put(worker->blkcg_css); 967 kfree(worker); 968 } 969 if (!list_empty(&lo->idle_worker_list)) 970 loop_set_timer(lo); 971 spin_unlock_irq(&lo->lo_work_lock); 972 } 973 974 static void loop_free_idle_workers_timer(struct timer_list *timer) 975 { 976 struct loop_device *lo = container_of(timer, struct loop_device, timer); 977 978 return loop_free_idle_workers(lo, false); 979 } 980 981 static void loop_update_rotational(struct loop_device *lo) 982 { 983 struct file *file = lo->lo_backing_file; 984 struct inode *file_inode = file->f_mapping->host; 985 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 986 struct request_queue *q = lo->lo_queue; 987 bool nonrot = true; 988 989 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 990 if (file_bdev) 991 nonrot = bdev_nonrot(file_bdev); 992 993 if (nonrot) 994 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 995 else 996 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 997 } 998 999 /** 1000 * loop_set_status_from_info - configure device from loop_info 1001 * @lo: struct loop_device to configure 1002 * @info: struct loop_info64 to configure the device with 1003 * 1004 * Configures the loop device parameters according to the passed 1005 * in loop_info64 configuration. 1006 */ 1007 static int 1008 loop_set_status_from_info(struct loop_device *lo, 1009 const struct loop_info64 *info) 1010 { 1011 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1012 return -EINVAL; 1013 1014 switch (info->lo_encrypt_type) { 1015 case LO_CRYPT_NONE: 1016 break; 1017 case LO_CRYPT_XOR: 1018 pr_warn("support for the xor transformation has been removed.\n"); 1019 return -EINVAL; 1020 case LO_CRYPT_CRYPTOAPI: 1021 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n"); 1022 return -EINVAL; 1023 default: 1024 return -EINVAL; 1025 } 1026 1027 lo->lo_offset = info->lo_offset; 1028 lo->lo_sizelimit = info->lo_sizelimit; 1029 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1030 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1031 lo->lo_flags = info->lo_flags; 1032 return 0; 1033 } 1034 1035 static int loop_configure(struct loop_device *lo, fmode_t mode, 1036 struct block_device *bdev, 1037 const struct loop_config *config) 1038 { 1039 struct file *file = fget(config->fd); 1040 struct inode *inode; 1041 struct address_space *mapping; 1042 int error; 1043 loff_t size; 1044 bool partscan; 1045 unsigned short bsize; 1046 bool is_loop; 1047 1048 if (!file) 1049 return -EBADF; 1050 is_loop = is_loop_device(file); 1051 1052 /* This is safe, since we have a reference from open(). */ 1053 __module_get(THIS_MODULE); 1054 1055 /* suppress uevents while reconfiguring the device */ 1056 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1); 1057 1058 /* 1059 * If we don't hold exclusive handle for the device, upgrade to it 1060 * here to avoid changing device under exclusive owner. 1061 */ 1062 if (!(mode & FMODE_EXCL)) { 1063 error = bd_prepare_to_claim(bdev, loop_configure); 1064 if (error) 1065 goto out_putf; 1066 } 1067 1068 error = loop_global_lock_killable(lo, is_loop); 1069 if (error) 1070 goto out_bdev; 1071 1072 error = -EBUSY; 1073 if (lo->lo_state != Lo_unbound) 1074 goto out_unlock; 1075 1076 error = loop_validate_file(file, bdev); 1077 if (error) 1078 goto out_unlock; 1079 1080 mapping = file->f_mapping; 1081 inode = mapping->host; 1082 1083 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1084 error = -EINVAL; 1085 goto out_unlock; 1086 } 1087 1088 if (config->block_size) { 1089 error = blk_validate_block_size(config->block_size); 1090 if (error) 1091 goto out_unlock; 1092 } 1093 1094 error = loop_set_status_from_info(lo, &config->info); 1095 if (error) 1096 goto out_unlock; 1097 1098 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1099 !file->f_op->write_iter) 1100 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1101 1102 if (!lo->workqueue) { 1103 lo->workqueue = alloc_workqueue("loop%d", 1104 WQ_UNBOUND | WQ_FREEZABLE, 1105 0, lo->lo_number); 1106 if (!lo->workqueue) { 1107 error = -ENOMEM; 1108 goto out_unlock; 1109 } 1110 } 1111 1112 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1113 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1114 1115 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1116 lo->lo_device = bdev; 1117 lo->lo_backing_file = file; 1118 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1119 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1120 1121 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1122 blk_queue_write_cache(lo->lo_queue, true, false); 1123 1124 if (config->block_size) 1125 bsize = config->block_size; 1126 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1127 /* In case of direct I/O, match underlying block size */ 1128 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1129 else 1130 bsize = 512; 1131 1132 blk_queue_logical_block_size(lo->lo_queue, bsize); 1133 blk_queue_physical_block_size(lo->lo_queue, bsize); 1134 blk_queue_io_min(lo->lo_queue, bsize); 1135 1136 loop_config_discard(lo); 1137 loop_update_rotational(lo); 1138 loop_update_dio(lo); 1139 loop_sysfs_init(lo); 1140 1141 size = get_loop_size(lo, file); 1142 loop_set_size(lo, size); 1143 1144 /* Order wrt reading lo_state in loop_validate_file(). */ 1145 wmb(); 1146 1147 lo->lo_state = Lo_bound; 1148 if (part_shift) 1149 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1150 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1151 if (partscan) 1152 lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 1153 1154 loop_global_unlock(lo, is_loop); 1155 if (partscan) 1156 loop_reread_partitions(lo); 1157 if (!(mode & FMODE_EXCL)) 1158 bd_abort_claiming(bdev, loop_configure); 1159 1160 error = 0; 1161 done: 1162 /* enable and uncork uevent now that we are done */ 1163 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0); 1164 return error; 1165 1166 out_unlock: 1167 loop_global_unlock(lo, is_loop); 1168 out_bdev: 1169 if (!(mode & FMODE_EXCL)) 1170 bd_abort_claiming(bdev, loop_configure); 1171 out_putf: 1172 fput(file); 1173 /* This is safe: open() is still holding a reference. */ 1174 module_put(THIS_MODULE); 1175 goto done; 1176 } 1177 1178 static void __loop_clr_fd(struct loop_device *lo, bool release) 1179 { 1180 struct file *filp; 1181 gfp_t gfp = lo->old_gfp_mask; 1182 1183 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1184 blk_queue_write_cache(lo->lo_queue, false, false); 1185 1186 /* 1187 * Freeze the request queue when unbinding on a live file descriptor and 1188 * thus an open device. When called from ->release we are guaranteed 1189 * that there is no I/O in progress already. 1190 */ 1191 if (!release) 1192 blk_mq_freeze_queue(lo->lo_queue); 1193 1194 spin_lock_irq(&lo->lo_lock); 1195 filp = lo->lo_backing_file; 1196 lo->lo_backing_file = NULL; 1197 spin_unlock_irq(&lo->lo_lock); 1198 1199 lo->lo_device = NULL; 1200 lo->lo_offset = 0; 1201 lo->lo_sizelimit = 0; 1202 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1203 blk_queue_logical_block_size(lo->lo_queue, 512); 1204 blk_queue_physical_block_size(lo->lo_queue, 512); 1205 blk_queue_io_min(lo->lo_queue, 512); 1206 invalidate_disk(lo->lo_disk); 1207 loop_sysfs_exit(lo); 1208 /* let user-space know about this change */ 1209 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 1210 mapping_set_gfp_mask(filp->f_mapping, gfp); 1211 /* This is safe: open() is still holding a reference. */ 1212 module_put(THIS_MODULE); 1213 if (!release) 1214 blk_mq_unfreeze_queue(lo->lo_queue); 1215 1216 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1217 1218 if (lo->lo_flags & LO_FLAGS_PARTSCAN) { 1219 int err; 1220 1221 /* 1222 * open_mutex has been held already in release path, so don't 1223 * acquire it if this function is called in such case. 1224 * 1225 * If the reread partition isn't from release path, lo_refcnt 1226 * must be at least one and it can only become zero when the 1227 * current holder is released. 1228 */ 1229 if (!release) 1230 mutex_lock(&lo->lo_disk->open_mutex); 1231 err = bdev_disk_changed(lo->lo_disk, false); 1232 if (!release) 1233 mutex_unlock(&lo->lo_disk->open_mutex); 1234 if (err) 1235 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1236 __func__, lo->lo_number, err); 1237 /* Device is gone, no point in returning error */ 1238 } 1239 1240 /* 1241 * lo->lo_state is set to Lo_unbound here after above partscan has 1242 * finished. There cannot be anybody else entering __loop_clr_fd() as 1243 * Lo_rundown state protects us from all the other places trying to 1244 * change the 'lo' device. 1245 */ 1246 lo->lo_flags = 0; 1247 if (!part_shift) 1248 lo->lo_disk->flags |= GENHD_FL_NO_PART; 1249 mutex_lock(&lo->lo_mutex); 1250 lo->lo_state = Lo_unbound; 1251 mutex_unlock(&lo->lo_mutex); 1252 1253 /* 1254 * Need not hold lo_mutex to fput backing file. Calling fput holding 1255 * lo_mutex triggers a circular lock dependency possibility warning as 1256 * fput can take open_mutex which is usually taken before lo_mutex. 1257 */ 1258 fput(filp); 1259 } 1260 1261 static int loop_clr_fd(struct loop_device *lo) 1262 { 1263 int err; 1264 1265 /* 1266 * Since lo_ioctl() is called without locks held, it is possible that 1267 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel. 1268 * 1269 * Therefore, use global lock when setting Lo_rundown state in order to 1270 * make sure that loop_validate_file() will fail if the "struct file" 1271 * which loop_configure()/loop_change_fd() found via fget() was this 1272 * loop device. 1273 */ 1274 err = loop_global_lock_killable(lo, true); 1275 if (err) 1276 return err; 1277 if (lo->lo_state != Lo_bound) { 1278 loop_global_unlock(lo, true); 1279 return -ENXIO; 1280 } 1281 /* 1282 * If we've explicitly asked to tear down the loop device, 1283 * and it has an elevated reference count, set it for auto-teardown when 1284 * the last reference goes away. This stops $!~#$@ udev from 1285 * preventing teardown because it decided that it needs to run blkid on 1286 * the loopback device whenever they appear. xfstests is notorious for 1287 * failing tests because blkid via udev races with a losetup 1288 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1289 * command to fail with EBUSY. 1290 */ 1291 if (disk_openers(lo->lo_disk) > 1) { 1292 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1293 loop_global_unlock(lo, true); 1294 return 0; 1295 } 1296 lo->lo_state = Lo_rundown; 1297 loop_global_unlock(lo, true); 1298 1299 __loop_clr_fd(lo, false); 1300 return 0; 1301 } 1302 1303 static int 1304 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1305 { 1306 int err; 1307 int prev_lo_flags; 1308 bool partscan = false; 1309 bool size_changed = false; 1310 1311 err = mutex_lock_killable(&lo->lo_mutex); 1312 if (err) 1313 return err; 1314 if (lo->lo_state != Lo_bound) { 1315 err = -ENXIO; 1316 goto out_unlock; 1317 } 1318 1319 if (lo->lo_offset != info->lo_offset || 1320 lo->lo_sizelimit != info->lo_sizelimit) { 1321 size_changed = true; 1322 sync_blockdev(lo->lo_device); 1323 invalidate_bdev(lo->lo_device); 1324 } 1325 1326 /* I/O need to be drained during transfer transition */ 1327 blk_mq_freeze_queue(lo->lo_queue); 1328 1329 prev_lo_flags = lo->lo_flags; 1330 1331 err = loop_set_status_from_info(lo, info); 1332 if (err) 1333 goto out_unfreeze; 1334 1335 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1336 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1337 /* For those flags, use the previous values instead */ 1338 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1339 /* For flags that can't be cleared, use previous values too */ 1340 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1341 1342 if (size_changed) { 1343 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1344 lo->lo_backing_file); 1345 loop_set_size(lo, new_size); 1346 } 1347 1348 loop_config_discard(lo); 1349 1350 /* update dio if lo_offset or transfer is changed */ 1351 __loop_update_dio(lo, lo->use_dio); 1352 1353 out_unfreeze: 1354 blk_mq_unfreeze_queue(lo->lo_queue); 1355 1356 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1357 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1358 lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 1359 partscan = true; 1360 } 1361 out_unlock: 1362 mutex_unlock(&lo->lo_mutex); 1363 if (partscan) 1364 loop_reread_partitions(lo); 1365 1366 return err; 1367 } 1368 1369 static int 1370 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1371 { 1372 struct path path; 1373 struct kstat stat; 1374 int ret; 1375 1376 ret = mutex_lock_killable(&lo->lo_mutex); 1377 if (ret) 1378 return ret; 1379 if (lo->lo_state != Lo_bound) { 1380 mutex_unlock(&lo->lo_mutex); 1381 return -ENXIO; 1382 } 1383 1384 memset(info, 0, sizeof(*info)); 1385 info->lo_number = lo->lo_number; 1386 info->lo_offset = lo->lo_offset; 1387 info->lo_sizelimit = lo->lo_sizelimit; 1388 info->lo_flags = lo->lo_flags; 1389 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1390 1391 /* Drop lo_mutex while we call into the filesystem. */ 1392 path = lo->lo_backing_file->f_path; 1393 path_get(&path); 1394 mutex_unlock(&lo->lo_mutex); 1395 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1396 if (!ret) { 1397 info->lo_device = huge_encode_dev(stat.dev); 1398 info->lo_inode = stat.ino; 1399 info->lo_rdevice = huge_encode_dev(stat.rdev); 1400 } 1401 path_put(&path); 1402 return ret; 1403 } 1404 1405 static void 1406 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1407 { 1408 memset(info64, 0, sizeof(*info64)); 1409 info64->lo_number = info->lo_number; 1410 info64->lo_device = info->lo_device; 1411 info64->lo_inode = info->lo_inode; 1412 info64->lo_rdevice = info->lo_rdevice; 1413 info64->lo_offset = info->lo_offset; 1414 info64->lo_sizelimit = 0; 1415 info64->lo_flags = info->lo_flags; 1416 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1417 } 1418 1419 static int 1420 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1421 { 1422 memset(info, 0, sizeof(*info)); 1423 info->lo_number = info64->lo_number; 1424 info->lo_device = info64->lo_device; 1425 info->lo_inode = info64->lo_inode; 1426 info->lo_rdevice = info64->lo_rdevice; 1427 info->lo_offset = info64->lo_offset; 1428 info->lo_flags = info64->lo_flags; 1429 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1430 1431 /* error in case values were truncated */ 1432 if (info->lo_device != info64->lo_device || 1433 info->lo_rdevice != info64->lo_rdevice || 1434 info->lo_inode != info64->lo_inode || 1435 info->lo_offset != info64->lo_offset) 1436 return -EOVERFLOW; 1437 1438 return 0; 1439 } 1440 1441 static int 1442 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1443 { 1444 struct loop_info info; 1445 struct loop_info64 info64; 1446 1447 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1448 return -EFAULT; 1449 loop_info64_from_old(&info, &info64); 1450 return loop_set_status(lo, &info64); 1451 } 1452 1453 static int 1454 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1455 { 1456 struct loop_info64 info64; 1457 1458 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1459 return -EFAULT; 1460 return loop_set_status(lo, &info64); 1461 } 1462 1463 static int 1464 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1465 struct loop_info info; 1466 struct loop_info64 info64; 1467 int err; 1468 1469 if (!arg) 1470 return -EINVAL; 1471 err = loop_get_status(lo, &info64); 1472 if (!err) 1473 err = loop_info64_to_old(&info64, &info); 1474 if (!err && copy_to_user(arg, &info, sizeof(info))) 1475 err = -EFAULT; 1476 1477 return err; 1478 } 1479 1480 static int 1481 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1482 struct loop_info64 info64; 1483 int err; 1484 1485 if (!arg) 1486 return -EINVAL; 1487 err = loop_get_status(lo, &info64); 1488 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1489 err = -EFAULT; 1490 1491 return err; 1492 } 1493 1494 static int loop_set_capacity(struct loop_device *lo) 1495 { 1496 loff_t size; 1497 1498 if (unlikely(lo->lo_state != Lo_bound)) 1499 return -ENXIO; 1500 1501 size = get_loop_size(lo, lo->lo_backing_file); 1502 loop_set_size(lo, size); 1503 1504 return 0; 1505 } 1506 1507 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1508 { 1509 int error = -ENXIO; 1510 if (lo->lo_state != Lo_bound) 1511 goto out; 1512 1513 __loop_update_dio(lo, !!arg); 1514 if (lo->use_dio == !!arg) 1515 return 0; 1516 error = -EINVAL; 1517 out: 1518 return error; 1519 } 1520 1521 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1522 { 1523 int err = 0; 1524 1525 if (lo->lo_state != Lo_bound) 1526 return -ENXIO; 1527 1528 err = blk_validate_block_size(arg); 1529 if (err) 1530 return err; 1531 1532 if (lo->lo_queue->limits.logical_block_size == arg) 1533 return 0; 1534 1535 sync_blockdev(lo->lo_device); 1536 invalidate_bdev(lo->lo_device); 1537 1538 blk_mq_freeze_queue(lo->lo_queue); 1539 blk_queue_logical_block_size(lo->lo_queue, arg); 1540 blk_queue_physical_block_size(lo->lo_queue, arg); 1541 blk_queue_io_min(lo->lo_queue, arg); 1542 loop_update_dio(lo); 1543 blk_mq_unfreeze_queue(lo->lo_queue); 1544 1545 return err; 1546 } 1547 1548 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1549 unsigned long arg) 1550 { 1551 int err; 1552 1553 err = mutex_lock_killable(&lo->lo_mutex); 1554 if (err) 1555 return err; 1556 switch (cmd) { 1557 case LOOP_SET_CAPACITY: 1558 err = loop_set_capacity(lo); 1559 break; 1560 case LOOP_SET_DIRECT_IO: 1561 err = loop_set_dio(lo, arg); 1562 break; 1563 case LOOP_SET_BLOCK_SIZE: 1564 err = loop_set_block_size(lo, arg); 1565 break; 1566 default: 1567 err = -EINVAL; 1568 } 1569 mutex_unlock(&lo->lo_mutex); 1570 return err; 1571 } 1572 1573 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1574 unsigned int cmd, unsigned long arg) 1575 { 1576 struct loop_device *lo = bdev->bd_disk->private_data; 1577 void __user *argp = (void __user *) arg; 1578 int err; 1579 1580 switch (cmd) { 1581 case LOOP_SET_FD: { 1582 /* 1583 * Legacy case - pass in a zeroed out struct loop_config with 1584 * only the file descriptor set , which corresponds with the 1585 * default parameters we'd have used otherwise. 1586 */ 1587 struct loop_config config; 1588 1589 memset(&config, 0, sizeof(config)); 1590 config.fd = arg; 1591 1592 return loop_configure(lo, mode, bdev, &config); 1593 } 1594 case LOOP_CONFIGURE: { 1595 struct loop_config config; 1596 1597 if (copy_from_user(&config, argp, sizeof(config))) 1598 return -EFAULT; 1599 1600 return loop_configure(lo, mode, bdev, &config); 1601 } 1602 case LOOP_CHANGE_FD: 1603 return loop_change_fd(lo, bdev, arg); 1604 case LOOP_CLR_FD: 1605 return loop_clr_fd(lo); 1606 case LOOP_SET_STATUS: 1607 err = -EPERM; 1608 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1609 err = loop_set_status_old(lo, argp); 1610 } 1611 break; 1612 case LOOP_GET_STATUS: 1613 return loop_get_status_old(lo, argp); 1614 case LOOP_SET_STATUS64: 1615 err = -EPERM; 1616 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1617 err = loop_set_status64(lo, argp); 1618 } 1619 break; 1620 case LOOP_GET_STATUS64: 1621 return loop_get_status64(lo, argp); 1622 case LOOP_SET_CAPACITY: 1623 case LOOP_SET_DIRECT_IO: 1624 case LOOP_SET_BLOCK_SIZE: 1625 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1626 return -EPERM; 1627 fallthrough; 1628 default: 1629 err = lo_simple_ioctl(lo, cmd, arg); 1630 break; 1631 } 1632 1633 return err; 1634 } 1635 1636 #ifdef CONFIG_COMPAT 1637 struct compat_loop_info { 1638 compat_int_t lo_number; /* ioctl r/o */ 1639 compat_dev_t lo_device; /* ioctl r/o */ 1640 compat_ulong_t lo_inode; /* ioctl r/o */ 1641 compat_dev_t lo_rdevice; /* ioctl r/o */ 1642 compat_int_t lo_offset; 1643 compat_int_t lo_encrypt_type; /* obsolete, ignored */ 1644 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1645 compat_int_t lo_flags; /* ioctl r/o */ 1646 char lo_name[LO_NAME_SIZE]; 1647 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1648 compat_ulong_t lo_init[2]; 1649 char reserved[4]; 1650 }; 1651 1652 /* 1653 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1654 * - noinlined to reduce stack space usage in main part of driver 1655 */ 1656 static noinline int 1657 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1658 struct loop_info64 *info64) 1659 { 1660 struct compat_loop_info info; 1661 1662 if (copy_from_user(&info, arg, sizeof(info))) 1663 return -EFAULT; 1664 1665 memset(info64, 0, sizeof(*info64)); 1666 info64->lo_number = info.lo_number; 1667 info64->lo_device = info.lo_device; 1668 info64->lo_inode = info.lo_inode; 1669 info64->lo_rdevice = info.lo_rdevice; 1670 info64->lo_offset = info.lo_offset; 1671 info64->lo_sizelimit = 0; 1672 info64->lo_flags = info.lo_flags; 1673 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1674 return 0; 1675 } 1676 1677 /* 1678 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1679 * - noinlined to reduce stack space usage in main part of driver 1680 */ 1681 static noinline int 1682 loop_info64_to_compat(const struct loop_info64 *info64, 1683 struct compat_loop_info __user *arg) 1684 { 1685 struct compat_loop_info info; 1686 1687 memset(&info, 0, sizeof(info)); 1688 info.lo_number = info64->lo_number; 1689 info.lo_device = info64->lo_device; 1690 info.lo_inode = info64->lo_inode; 1691 info.lo_rdevice = info64->lo_rdevice; 1692 info.lo_offset = info64->lo_offset; 1693 info.lo_flags = info64->lo_flags; 1694 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1695 1696 /* error in case values were truncated */ 1697 if (info.lo_device != info64->lo_device || 1698 info.lo_rdevice != info64->lo_rdevice || 1699 info.lo_inode != info64->lo_inode || 1700 info.lo_offset != info64->lo_offset) 1701 return -EOVERFLOW; 1702 1703 if (copy_to_user(arg, &info, sizeof(info))) 1704 return -EFAULT; 1705 return 0; 1706 } 1707 1708 static int 1709 loop_set_status_compat(struct loop_device *lo, 1710 const struct compat_loop_info __user *arg) 1711 { 1712 struct loop_info64 info64; 1713 int ret; 1714 1715 ret = loop_info64_from_compat(arg, &info64); 1716 if (ret < 0) 1717 return ret; 1718 return loop_set_status(lo, &info64); 1719 } 1720 1721 static int 1722 loop_get_status_compat(struct loop_device *lo, 1723 struct compat_loop_info __user *arg) 1724 { 1725 struct loop_info64 info64; 1726 int err; 1727 1728 if (!arg) 1729 return -EINVAL; 1730 err = loop_get_status(lo, &info64); 1731 if (!err) 1732 err = loop_info64_to_compat(&info64, arg); 1733 return err; 1734 } 1735 1736 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1737 unsigned int cmd, unsigned long arg) 1738 { 1739 struct loop_device *lo = bdev->bd_disk->private_data; 1740 int err; 1741 1742 switch(cmd) { 1743 case LOOP_SET_STATUS: 1744 err = loop_set_status_compat(lo, 1745 (const struct compat_loop_info __user *)arg); 1746 break; 1747 case LOOP_GET_STATUS: 1748 err = loop_get_status_compat(lo, 1749 (struct compat_loop_info __user *)arg); 1750 break; 1751 case LOOP_SET_CAPACITY: 1752 case LOOP_CLR_FD: 1753 case LOOP_GET_STATUS64: 1754 case LOOP_SET_STATUS64: 1755 case LOOP_CONFIGURE: 1756 arg = (unsigned long) compat_ptr(arg); 1757 fallthrough; 1758 case LOOP_SET_FD: 1759 case LOOP_CHANGE_FD: 1760 case LOOP_SET_BLOCK_SIZE: 1761 case LOOP_SET_DIRECT_IO: 1762 err = lo_ioctl(bdev, mode, cmd, arg); 1763 break; 1764 default: 1765 err = -ENOIOCTLCMD; 1766 break; 1767 } 1768 return err; 1769 } 1770 #endif 1771 1772 static void lo_release(struct gendisk *disk, fmode_t mode) 1773 { 1774 struct loop_device *lo = disk->private_data; 1775 1776 if (disk_openers(disk) > 0) 1777 return; 1778 1779 mutex_lock(&lo->lo_mutex); 1780 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) { 1781 lo->lo_state = Lo_rundown; 1782 mutex_unlock(&lo->lo_mutex); 1783 /* 1784 * In autoclear mode, stop the loop thread 1785 * and remove configuration after last close. 1786 */ 1787 __loop_clr_fd(lo, true); 1788 return; 1789 } 1790 mutex_unlock(&lo->lo_mutex); 1791 } 1792 1793 static void lo_free_disk(struct gendisk *disk) 1794 { 1795 struct loop_device *lo = disk->private_data; 1796 1797 if (lo->workqueue) 1798 destroy_workqueue(lo->workqueue); 1799 loop_free_idle_workers(lo, true); 1800 del_timer_sync(&lo->timer); 1801 mutex_destroy(&lo->lo_mutex); 1802 kfree(lo); 1803 } 1804 1805 static const struct block_device_operations lo_fops = { 1806 .owner = THIS_MODULE, 1807 .release = lo_release, 1808 .ioctl = lo_ioctl, 1809 #ifdef CONFIG_COMPAT 1810 .compat_ioctl = lo_compat_ioctl, 1811 #endif 1812 .free_disk = lo_free_disk, 1813 }; 1814 1815 /* 1816 * And now the modules code and kernel interface. 1817 */ 1818 static int max_loop; 1819 module_param(max_loop, int, 0444); 1820 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1821 module_param(max_part, int, 0444); 1822 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1823 1824 static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH; 1825 1826 static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p) 1827 { 1828 int ret = kstrtoint(s, 10, &hw_queue_depth); 1829 1830 return (ret || (hw_queue_depth < 1)) ? -EINVAL : 0; 1831 } 1832 1833 static const struct kernel_param_ops loop_hw_qdepth_param_ops = { 1834 .set = loop_set_hw_queue_depth, 1835 .get = param_get_int, 1836 }; 1837 1838 device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444); 1839 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 128"); 1840 1841 MODULE_LICENSE("GPL"); 1842 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1843 1844 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1845 const struct blk_mq_queue_data *bd) 1846 { 1847 struct request *rq = bd->rq; 1848 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1849 struct loop_device *lo = rq->q->queuedata; 1850 1851 blk_mq_start_request(rq); 1852 1853 if (lo->lo_state != Lo_bound) 1854 return BLK_STS_IOERR; 1855 1856 switch (req_op(rq)) { 1857 case REQ_OP_FLUSH: 1858 case REQ_OP_DISCARD: 1859 case REQ_OP_WRITE_ZEROES: 1860 cmd->use_aio = false; 1861 break; 1862 default: 1863 cmd->use_aio = lo->use_dio; 1864 break; 1865 } 1866 1867 /* always use the first bio's css */ 1868 cmd->blkcg_css = NULL; 1869 cmd->memcg_css = NULL; 1870 #ifdef CONFIG_BLK_CGROUP 1871 if (rq->bio && rq->bio->bi_blkg) { 1872 cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 1873 #ifdef CONFIG_MEMCG 1874 cmd->memcg_css = 1875 cgroup_get_e_css(cmd->blkcg_css->cgroup, 1876 &memory_cgrp_subsys); 1877 #endif 1878 } 1879 #endif 1880 loop_queue_work(lo, cmd); 1881 1882 return BLK_STS_OK; 1883 } 1884 1885 static void loop_handle_cmd(struct loop_cmd *cmd) 1886 { 1887 struct request *rq = blk_mq_rq_from_pdu(cmd); 1888 const bool write = op_is_write(req_op(rq)); 1889 struct loop_device *lo = rq->q->queuedata; 1890 int ret = 0; 1891 struct mem_cgroup *old_memcg = NULL; 1892 1893 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1894 ret = -EIO; 1895 goto failed; 1896 } 1897 1898 if (cmd->blkcg_css) 1899 kthread_associate_blkcg(cmd->blkcg_css); 1900 if (cmd->memcg_css) 1901 old_memcg = set_active_memcg( 1902 mem_cgroup_from_css(cmd->memcg_css)); 1903 1904 ret = do_req_filebacked(lo, rq); 1905 1906 if (cmd->blkcg_css) 1907 kthread_associate_blkcg(NULL); 1908 1909 if (cmd->memcg_css) { 1910 set_active_memcg(old_memcg); 1911 css_put(cmd->memcg_css); 1912 } 1913 failed: 1914 /* complete non-aio request */ 1915 if (!cmd->use_aio || ret) { 1916 if (ret == -EOPNOTSUPP) 1917 cmd->ret = ret; 1918 else 1919 cmd->ret = ret ? -EIO : 0; 1920 if (likely(!blk_should_fake_timeout(rq->q))) 1921 blk_mq_complete_request(rq); 1922 } 1923 } 1924 1925 static void loop_process_work(struct loop_worker *worker, 1926 struct list_head *cmd_list, struct loop_device *lo) 1927 { 1928 int orig_flags = current->flags; 1929 struct loop_cmd *cmd; 1930 1931 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 1932 spin_lock_irq(&lo->lo_work_lock); 1933 while (!list_empty(cmd_list)) { 1934 cmd = container_of( 1935 cmd_list->next, struct loop_cmd, list_entry); 1936 list_del(cmd_list->next); 1937 spin_unlock_irq(&lo->lo_work_lock); 1938 1939 loop_handle_cmd(cmd); 1940 cond_resched(); 1941 1942 spin_lock_irq(&lo->lo_work_lock); 1943 } 1944 1945 /* 1946 * We only add to the idle list if there are no pending cmds 1947 * *and* the worker will not run again which ensures that it 1948 * is safe to free any worker on the idle list 1949 */ 1950 if (worker && !work_pending(&worker->work)) { 1951 worker->last_ran_at = jiffies; 1952 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 1953 loop_set_timer(lo); 1954 } 1955 spin_unlock_irq(&lo->lo_work_lock); 1956 current->flags = orig_flags; 1957 } 1958 1959 static void loop_workfn(struct work_struct *work) 1960 { 1961 struct loop_worker *worker = 1962 container_of(work, struct loop_worker, work); 1963 loop_process_work(worker, &worker->cmd_list, worker->lo); 1964 } 1965 1966 static void loop_rootcg_workfn(struct work_struct *work) 1967 { 1968 struct loop_device *lo = 1969 container_of(work, struct loop_device, rootcg_work); 1970 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 1971 } 1972 1973 static const struct blk_mq_ops loop_mq_ops = { 1974 .queue_rq = loop_queue_rq, 1975 .complete = lo_complete_rq, 1976 }; 1977 1978 static int loop_add(int i) 1979 { 1980 struct loop_device *lo; 1981 struct gendisk *disk; 1982 int err; 1983 1984 err = -ENOMEM; 1985 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1986 if (!lo) 1987 goto out; 1988 lo->worker_tree = RB_ROOT; 1989 INIT_LIST_HEAD(&lo->idle_worker_list); 1990 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE); 1991 lo->lo_state = Lo_unbound; 1992 1993 err = mutex_lock_killable(&loop_ctl_mutex); 1994 if (err) 1995 goto out_free_dev; 1996 1997 /* allocate id, if @id >= 0, we're requesting that specific id */ 1998 if (i >= 0) { 1999 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2000 if (err == -ENOSPC) 2001 err = -EEXIST; 2002 } else { 2003 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2004 } 2005 mutex_unlock(&loop_ctl_mutex); 2006 if (err < 0) 2007 goto out_free_dev; 2008 i = err; 2009 2010 lo->tag_set.ops = &loop_mq_ops; 2011 lo->tag_set.nr_hw_queues = 1; 2012 lo->tag_set.queue_depth = hw_queue_depth; 2013 lo->tag_set.numa_node = NUMA_NO_NODE; 2014 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2015 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | 2016 BLK_MQ_F_NO_SCHED_BY_DEFAULT; 2017 lo->tag_set.driver_data = lo; 2018 2019 err = blk_mq_alloc_tag_set(&lo->tag_set); 2020 if (err) 2021 goto out_free_idr; 2022 2023 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 2024 if (IS_ERR(disk)) { 2025 err = PTR_ERR(disk); 2026 goto out_cleanup_tags; 2027 } 2028 lo->lo_queue = lo->lo_disk->queue; 2029 2030 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2031 2032 /* 2033 * By default, we do buffer IO, so it doesn't make sense to enable 2034 * merge because the I/O submitted to backing file is handled page by 2035 * page. For directio mode, merge does help to dispatch bigger request 2036 * to underlayer disk. We will enable merge once directio is enabled. 2037 */ 2038 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2039 2040 /* 2041 * Disable partition scanning by default. The in-kernel partition 2042 * scanning can be requested individually per-device during its 2043 * setup. Userspace can always add and remove partitions from all 2044 * devices. The needed partition minors are allocated from the 2045 * extended minor space, the main loop device numbers will continue 2046 * to match the loop minors, regardless of the number of partitions 2047 * used. 2048 * 2049 * If max_part is given, partition scanning is globally enabled for 2050 * all loop devices. The minors for the main loop devices will be 2051 * multiples of max_part. 2052 * 2053 * Note: Global-for-all-devices, set-only-at-init, read-only module 2054 * parameteters like 'max_loop' and 'max_part' make things needlessly 2055 * complicated, are too static, inflexible and may surprise 2056 * userspace tools. Parameters like this in general should be avoided. 2057 */ 2058 if (!part_shift) 2059 disk->flags |= GENHD_FL_NO_PART; 2060 mutex_init(&lo->lo_mutex); 2061 lo->lo_number = i; 2062 spin_lock_init(&lo->lo_lock); 2063 spin_lock_init(&lo->lo_work_lock); 2064 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 2065 INIT_LIST_HEAD(&lo->rootcg_cmd_list); 2066 disk->major = LOOP_MAJOR; 2067 disk->first_minor = i << part_shift; 2068 disk->minors = 1 << part_shift; 2069 disk->fops = &lo_fops; 2070 disk->private_data = lo; 2071 disk->queue = lo->lo_queue; 2072 disk->events = DISK_EVENT_MEDIA_CHANGE; 2073 disk->event_flags = DISK_EVENT_FLAG_UEVENT; 2074 sprintf(disk->disk_name, "loop%d", i); 2075 /* Make this loop device reachable from pathname. */ 2076 err = add_disk(disk); 2077 if (err) 2078 goto out_cleanup_disk; 2079 2080 /* Show this loop device. */ 2081 mutex_lock(&loop_ctl_mutex); 2082 lo->idr_visible = true; 2083 mutex_unlock(&loop_ctl_mutex); 2084 2085 return i; 2086 2087 out_cleanup_disk: 2088 blk_cleanup_disk(disk); 2089 out_cleanup_tags: 2090 blk_mq_free_tag_set(&lo->tag_set); 2091 out_free_idr: 2092 mutex_lock(&loop_ctl_mutex); 2093 idr_remove(&loop_index_idr, i); 2094 mutex_unlock(&loop_ctl_mutex); 2095 out_free_dev: 2096 kfree(lo); 2097 out: 2098 return err; 2099 } 2100 2101 static void loop_remove(struct loop_device *lo) 2102 { 2103 /* Make this loop device unreachable from pathname. */ 2104 del_gendisk(lo->lo_disk); 2105 blk_cleanup_queue(lo->lo_disk->queue); 2106 blk_mq_free_tag_set(&lo->tag_set); 2107 2108 mutex_lock(&loop_ctl_mutex); 2109 idr_remove(&loop_index_idr, lo->lo_number); 2110 mutex_unlock(&loop_ctl_mutex); 2111 2112 put_disk(lo->lo_disk); 2113 } 2114 2115 static void loop_probe(dev_t dev) 2116 { 2117 int idx = MINOR(dev) >> part_shift; 2118 2119 if (max_loop && idx >= max_loop) 2120 return; 2121 loop_add(idx); 2122 } 2123 2124 static int loop_control_remove(int idx) 2125 { 2126 struct loop_device *lo; 2127 int ret; 2128 2129 if (idx < 0) { 2130 pr_warn_once("deleting an unspecified loop device is not supported.\n"); 2131 return -EINVAL; 2132 } 2133 2134 /* Hide this loop device for serialization. */ 2135 ret = mutex_lock_killable(&loop_ctl_mutex); 2136 if (ret) 2137 return ret; 2138 lo = idr_find(&loop_index_idr, idx); 2139 if (!lo || !lo->idr_visible) 2140 ret = -ENODEV; 2141 else 2142 lo->idr_visible = false; 2143 mutex_unlock(&loop_ctl_mutex); 2144 if (ret) 2145 return ret; 2146 2147 /* Check whether this loop device can be removed. */ 2148 ret = mutex_lock_killable(&lo->lo_mutex); 2149 if (ret) 2150 goto mark_visible; 2151 if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) { 2152 mutex_unlock(&lo->lo_mutex); 2153 ret = -EBUSY; 2154 goto mark_visible; 2155 } 2156 /* Mark this loop device as no more bound, but not quite unbound yet */ 2157 lo->lo_state = Lo_deleting; 2158 mutex_unlock(&lo->lo_mutex); 2159 2160 loop_remove(lo); 2161 return 0; 2162 2163 mark_visible: 2164 /* Show this loop device again. */ 2165 mutex_lock(&loop_ctl_mutex); 2166 lo->idr_visible = true; 2167 mutex_unlock(&loop_ctl_mutex); 2168 return ret; 2169 } 2170 2171 static int loop_control_get_free(int idx) 2172 { 2173 struct loop_device *lo; 2174 int id, ret; 2175 2176 ret = mutex_lock_killable(&loop_ctl_mutex); 2177 if (ret) 2178 return ret; 2179 idr_for_each_entry(&loop_index_idr, lo, id) { 2180 /* Hitting a race results in creating a new loop device which is harmless. */ 2181 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) 2182 goto found; 2183 } 2184 mutex_unlock(&loop_ctl_mutex); 2185 return loop_add(-1); 2186 found: 2187 mutex_unlock(&loop_ctl_mutex); 2188 return id; 2189 } 2190 2191 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2192 unsigned long parm) 2193 { 2194 switch (cmd) { 2195 case LOOP_CTL_ADD: 2196 return loop_add(parm); 2197 case LOOP_CTL_REMOVE: 2198 return loop_control_remove(parm); 2199 case LOOP_CTL_GET_FREE: 2200 return loop_control_get_free(parm); 2201 default: 2202 return -ENOSYS; 2203 } 2204 } 2205 2206 static const struct file_operations loop_ctl_fops = { 2207 .open = nonseekable_open, 2208 .unlocked_ioctl = loop_control_ioctl, 2209 .compat_ioctl = loop_control_ioctl, 2210 .owner = THIS_MODULE, 2211 .llseek = noop_llseek, 2212 }; 2213 2214 static struct miscdevice loop_misc = { 2215 .minor = LOOP_CTRL_MINOR, 2216 .name = "loop-control", 2217 .fops = &loop_ctl_fops, 2218 }; 2219 2220 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2221 MODULE_ALIAS("devname:loop-control"); 2222 2223 static int __init loop_init(void) 2224 { 2225 int i, nr; 2226 int err; 2227 2228 part_shift = 0; 2229 if (max_part > 0) { 2230 part_shift = fls(max_part); 2231 2232 /* 2233 * Adjust max_part according to part_shift as it is exported 2234 * to user space so that user can decide correct minor number 2235 * if [s]he want to create more devices. 2236 * 2237 * Note that -1 is required because partition 0 is reserved 2238 * for the whole disk. 2239 */ 2240 max_part = (1UL << part_shift) - 1; 2241 } 2242 2243 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2244 err = -EINVAL; 2245 goto err_out; 2246 } 2247 2248 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2249 err = -EINVAL; 2250 goto err_out; 2251 } 2252 2253 /* 2254 * If max_loop is specified, create that many devices upfront. 2255 * This also becomes a hard limit. If max_loop is not specified, 2256 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2257 * init time. Loop devices can be requested on-demand with the 2258 * /dev/loop-control interface, or be instantiated by accessing 2259 * a 'dead' device node. 2260 */ 2261 if (max_loop) 2262 nr = max_loop; 2263 else 2264 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2265 2266 err = misc_register(&loop_misc); 2267 if (err < 0) 2268 goto err_out; 2269 2270 2271 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2272 err = -EIO; 2273 goto misc_out; 2274 } 2275 2276 /* pre-create number of devices given by config or max_loop */ 2277 for (i = 0; i < nr; i++) 2278 loop_add(i); 2279 2280 printk(KERN_INFO "loop: module loaded\n"); 2281 return 0; 2282 2283 misc_out: 2284 misc_deregister(&loop_misc); 2285 err_out: 2286 return err; 2287 } 2288 2289 static void __exit loop_exit(void) 2290 { 2291 struct loop_device *lo; 2292 int id; 2293 2294 unregister_blkdev(LOOP_MAJOR, "loop"); 2295 misc_deregister(&loop_misc); 2296 2297 /* 2298 * There is no need to use loop_ctl_mutex here, for nobody else can 2299 * access loop_index_idr when this module is unloading (unless forced 2300 * module unloading is requested). If this is not a clean unloading, 2301 * we have no means to avoid kernel crash. 2302 */ 2303 idr_for_each_entry(&loop_index_idr, lo, id) 2304 loop_remove(lo); 2305 2306 idr_destroy(&loop_index_idr); 2307 } 2308 2309 module_init(loop_init); 2310 module_exit(loop_exit); 2311 2312 #ifndef MODULE 2313 static int __init max_loop_setup(char *str) 2314 { 2315 max_loop = simple_strtol(str, NULL, 0); 2316 return 1; 2317 } 2318 2319 __setup("max_loop=", max_loop_setup); 2320 #endif 2321