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 #define LOOP_DEFAULT_HW_Q_DEPTH (128) 90 91 static DEFINE_IDR(loop_index_idr); 92 static DEFINE_MUTEX(loop_ctl_mutex); 93 static DEFINE_MUTEX(loop_validate_mutex); 94 95 /** 96 * loop_global_lock_killable() - take locks for safe loop_validate_file() test 97 * 98 * @lo: struct loop_device 99 * @global: true if @lo is about to bind another "struct loop_device", false otherwise 100 * 101 * Returns 0 on success, -EINTR otherwise. 102 * 103 * Since loop_validate_file() traverses on other "struct loop_device" if 104 * is_loop_device() is true, we need a global lock for serializing concurrent 105 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls. 106 */ 107 static int loop_global_lock_killable(struct loop_device *lo, bool global) 108 { 109 int err; 110 111 if (global) { 112 err = mutex_lock_killable(&loop_validate_mutex); 113 if (err) 114 return err; 115 } 116 err = mutex_lock_killable(&lo->lo_mutex); 117 if (err && global) 118 mutex_unlock(&loop_validate_mutex); 119 return err; 120 } 121 122 /** 123 * loop_global_unlock() - release locks taken by loop_global_lock_killable() 124 * 125 * @lo: struct loop_device 126 * @global: true if @lo was about to bind another "struct loop_device", false otherwise 127 */ 128 static void loop_global_unlock(struct loop_device *lo, bool global) 129 { 130 mutex_unlock(&lo->lo_mutex); 131 if (global) 132 mutex_unlock(&loop_validate_mutex); 133 } 134 135 static int max_part; 136 static int part_shift; 137 138 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 139 { 140 loff_t loopsize; 141 142 /* Compute loopsize in bytes */ 143 loopsize = i_size_read(file->f_mapping->host); 144 if (offset > 0) 145 loopsize -= offset; 146 /* offset is beyond i_size, weird but possible */ 147 if (loopsize < 0) 148 return 0; 149 150 if (sizelimit > 0 && sizelimit < loopsize) 151 loopsize = sizelimit; 152 /* 153 * Unfortunately, if we want to do I/O on the device, 154 * the number of 512-byte sectors has to fit into a sector_t. 155 */ 156 return loopsize >> 9; 157 } 158 159 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 160 { 161 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 162 } 163 164 static void __loop_update_dio(struct loop_device *lo, bool dio) 165 { 166 struct file *file = lo->lo_backing_file; 167 struct address_space *mapping = file->f_mapping; 168 struct inode *inode = mapping->host; 169 unsigned short sb_bsize = 0; 170 unsigned dio_align = 0; 171 bool use_dio; 172 173 if (inode->i_sb->s_bdev) { 174 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 175 dio_align = sb_bsize - 1; 176 } 177 178 /* 179 * We support direct I/O only if lo_offset is aligned with the 180 * logical I/O size of backing device, and the logical block 181 * size of loop is bigger than the backing device's. 182 * 183 * TODO: the above condition may be loosed in the future, and 184 * direct I/O may be switched runtime at that time because most 185 * of requests in sane applications should be PAGE_SIZE aligned 186 */ 187 if (dio) { 188 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 189 !(lo->lo_offset & dio_align) && 190 mapping->a_ops->direct_IO) 191 use_dio = true; 192 else 193 use_dio = false; 194 } else { 195 use_dio = false; 196 } 197 198 if (lo->use_dio == use_dio) 199 return; 200 201 /* flush dirty pages before changing direct IO */ 202 vfs_fsync(file, 0); 203 204 /* 205 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 206 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 207 * will get updated by ioctl(LOOP_GET_STATUS) 208 */ 209 if (lo->lo_state == Lo_bound) 210 blk_mq_freeze_queue(lo->lo_queue); 211 lo->use_dio = use_dio; 212 if (use_dio) { 213 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 214 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 215 } else { 216 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 217 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 218 } 219 if (lo->lo_state == Lo_bound) 220 blk_mq_unfreeze_queue(lo->lo_queue); 221 } 222 223 /** 224 * loop_set_size() - sets device size and notifies userspace 225 * @lo: struct loop_device to set the size for 226 * @size: new size of the loop device 227 * 228 * Callers must validate that the size passed into this function fits into 229 * a sector_t, eg using loop_validate_size() 230 */ 231 static void loop_set_size(struct loop_device *lo, loff_t size) 232 { 233 if (!set_capacity_and_notify(lo->lo_disk, size)) 234 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 235 } 236 237 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 238 { 239 struct iov_iter i; 240 ssize_t bw; 241 242 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 243 244 file_start_write(file); 245 bw = vfs_iter_write(file, &i, ppos, 0); 246 file_end_write(file); 247 248 if (likely(bw == bvec->bv_len)) 249 return 0; 250 251 printk_ratelimited(KERN_ERR 252 "loop: Write error at byte offset %llu, length %i.\n", 253 (unsigned long long)*ppos, bvec->bv_len); 254 if (bw >= 0) 255 bw = -EIO; 256 return bw; 257 } 258 259 static int lo_write_simple(struct loop_device *lo, struct request *rq, 260 loff_t pos) 261 { 262 struct bio_vec bvec; 263 struct req_iterator iter; 264 int ret = 0; 265 266 rq_for_each_segment(bvec, rq, iter) { 267 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 268 if (ret < 0) 269 break; 270 cond_resched(); 271 } 272 273 return ret; 274 } 275 276 static int lo_read_simple(struct loop_device *lo, struct request *rq, 277 loff_t pos) 278 { 279 struct bio_vec bvec; 280 struct req_iterator iter; 281 struct iov_iter i; 282 ssize_t len; 283 284 rq_for_each_segment(bvec, rq, iter) { 285 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 286 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 287 if (len < 0) 288 return len; 289 290 flush_dcache_page(bvec.bv_page); 291 292 if (len != bvec.bv_len) { 293 struct bio *bio; 294 295 __rq_for_each_bio(bio, rq) 296 zero_fill_bio(bio); 297 break; 298 } 299 cond_resched(); 300 } 301 302 return 0; 303 } 304 305 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 306 int mode) 307 { 308 /* 309 * We use fallocate to manipulate the space mappings used by the image 310 * a.k.a. discard/zerorange. 311 */ 312 struct file *file = lo->lo_backing_file; 313 int ret; 314 315 mode |= FALLOC_FL_KEEP_SIZE; 316 317 if (!bdev_max_discard_sectors(lo->lo_device)) 318 return -EOPNOTSUPP; 319 320 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 321 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 322 return -EIO; 323 return ret; 324 } 325 326 static int lo_req_flush(struct loop_device *lo, struct request *rq) 327 { 328 int ret = vfs_fsync(lo->lo_backing_file, 0); 329 if (unlikely(ret && ret != -EINVAL)) 330 ret = -EIO; 331 332 return ret; 333 } 334 335 static void lo_complete_rq(struct request *rq) 336 { 337 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 338 blk_status_t ret = BLK_STS_OK; 339 340 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 341 req_op(rq) != REQ_OP_READ) { 342 if (cmd->ret < 0) 343 ret = errno_to_blk_status(cmd->ret); 344 goto end_io; 345 } 346 347 /* 348 * Short READ - if we got some data, advance our request and 349 * retry it. If we got no data, end the rest with EIO. 350 */ 351 if (cmd->ret) { 352 blk_update_request(rq, BLK_STS_OK, cmd->ret); 353 cmd->ret = 0; 354 blk_mq_requeue_request(rq, true); 355 } else { 356 if (cmd->use_aio) { 357 struct bio *bio = rq->bio; 358 359 while (bio) { 360 zero_fill_bio(bio); 361 bio = bio->bi_next; 362 } 363 } 364 ret = BLK_STS_IOERR; 365 end_io: 366 blk_mq_end_request(rq, ret); 367 } 368 } 369 370 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 371 { 372 struct request *rq = blk_mq_rq_from_pdu(cmd); 373 374 if (!atomic_dec_and_test(&cmd->ref)) 375 return; 376 kfree(cmd->bvec); 377 cmd->bvec = NULL; 378 if (likely(!blk_should_fake_timeout(rq->q))) 379 blk_mq_complete_request(rq); 380 } 381 382 static void lo_rw_aio_complete(struct kiocb *iocb, long ret) 383 { 384 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 385 386 cmd->ret = ret; 387 lo_rw_aio_do_completion(cmd); 388 } 389 390 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 391 loff_t pos, bool rw) 392 { 393 struct iov_iter iter; 394 struct req_iterator rq_iter; 395 struct bio_vec *bvec; 396 struct request *rq = blk_mq_rq_from_pdu(cmd); 397 struct bio *bio = rq->bio; 398 struct file *file = lo->lo_backing_file; 399 struct bio_vec tmp; 400 unsigned int offset; 401 int nr_bvec = 0; 402 int ret; 403 404 rq_for_each_bvec(tmp, rq, rq_iter) 405 nr_bvec++; 406 407 if (rq->bio != rq->biotail) { 408 409 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 410 GFP_NOIO); 411 if (!bvec) 412 return -EIO; 413 cmd->bvec = bvec; 414 415 /* 416 * The bios of the request may be started from the middle of 417 * the 'bvec' because of bio splitting, so we can't directly 418 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 419 * API will take care of all details for us. 420 */ 421 rq_for_each_bvec(tmp, rq, rq_iter) { 422 *bvec = tmp; 423 bvec++; 424 } 425 bvec = cmd->bvec; 426 offset = 0; 427 } else { 428 /* 429 * Same here, this bio may be started from the middle of the 430 * 'bvec' because of bio splitting, so offset from the bvec 431 * must be passed to iov iterator 432 */ 433 offset = bio->bi_iter.bi_bvec_done; 434 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 435 } 436 atomic_set(&cmd->ref, 2); 437 438 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 439 iter.iov_offset = offset; 440 441 cmd->iocb.ki_pos = pos; 442 cmd->iocb.ki_filp = file; 443 cmd->iocb.ki_complete = lo_rw_aio_complete; 444 cmd->iocb.ki_flags = IOCB_DIRECT; 445 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 446 447 if (rw == WRITE) 448 ret = call_write_iter(file, &cmd->iocb, &iter); 449 else 450 ret = call_read_iter(file, &cmd->iocb, &iter); 451 452 lo_rw_aio_do_completion(cmd); 453 454 if (ret != -EIOCBQUEUED) 455 lo_rw_aio_complete(&cmd->iocb, ret); 456 return 0; 457 } 458 459 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 460 { 461 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 462 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 463 464 /* 465 * lo_write_simple and lo_read_simple should have been covered 466 * by io submit style function like lo_rw_aio(), one blocker 467 * is that lo_read_simple() need to call flush_dcache_page after 468 * the page is written from kernel, and it isn't easy to handle 469 * this in io submit style function which submits all segments 470 * of the req at one time. And direct read IO doesn't need to 471 * run flush_dcache_page(). 472 */ 473 switch (req_op(rq)) { 474 case REQ_OP_FLUSH: 475 return lo_req_flush(lo, rq); 476 case REQ_OP_WRITE_ZEROES: 477 /* 478 * If the caller doesn't want deallocation, call zeroout to 479 * write zeroes the range. Otherwise, punch them out. 480 */ 481 return lo_fallocate(lo, rq, pos, 482 (rq->cmd_flags & REQ_NOUNMAP) ? 483 FALLOC_FL_ZERO_RANGE : 484 FALLOC_FL_PUNCH_HOLE); 485 case REQ_OP_DISCARD: 486 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 487 case REQ_OP_WRITE: 488 if (cmd->use_aio) 489 return lo_rw_aio(lo, cmd, pos, WRITE); 490 else 491 return lo_write_simple(lo, rq, pos); 492 case REQ_OP_READ: 493 if (cmd->use_aio) 494 return lo_rw_aio(lo, cmd, pos, READ); 495 else 496 return lo_read_simple(lo, rq, pos); 497 default: 498 WARN_ON_ONCE(1); 499 return -EIO; 500 } 501 } 502 503 static inline void loop_update_dio(struct loop_device *lo) 504 { 505 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 506 lo->use_dio); 507 } 508 509 static void loop_reread_partitions(struct loop_device *lo) 510 { 511 int rc; 512 513 mutex_lock(&lo->lo_disk->open_mutex); 514 rc = bdev_disk_changed(lo->lo_disk, false); 515 mutex_unlock(&lo->lo_disk->open_mutex); 516 if (rc) 517 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 518 __func__, lo->lo_number, lo->lo_file_name, rc); 519 } 520 521 static inline int is_loop_device(struct file *file) 522 { 523 struct inode *i = file->f_mapping->host; 524 525 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; 526 } 527 528 static int loop_validate_file(struct file *file, struct block_device *bdev) 529 { 530 struct inode *inode = file->f_mapping->host; 531 struct file *f = file; 532 533 /* Avoid recursion */ 534 while (is_loop_device(f)) { 535 struct loop_device *l; 536 537 lockdep_assert_held(&loop_validate_mutex); 538 if (f->f_mapping->host->i_rdev == bdev->bd_dev) 539 return -EBADF; 540 541 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 542 if (l->lo_state != Lo_bound) 543 return -EINVAL; 544 /* Order wrt setting lo->lo_backing_file in loop_configure(). */ 545 rmb(); 546 f = l->lo_backing_file; 547 } 548 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 549 return -EINVAL; 550 return 0; 551 } 552 553 /* 554 * loop_change_fd switched the backing store of a loopback device to 555 * a new file. This is useful for operating system installers to free up 556 * the original file and in High Availability environments to switch to 557 * an alternative location for the content in case of server meltdown. 558 * This can only work if the loop device is used read-only, and if the 559 * new backing store is the same size and type as the old backing store. 560 */ 561 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 562 unsigned int arg) 563 { 564 struct file *file = fget(arg); 565 struct file *old_file; 566 int error; 567 bool partscan; 568 bool is_loop; 569 570 if (!file) 571 return -EBADF; 572 is_loop = is_loop_device(file); 573 error = loop_global_lock_killable(lo, is_loop); 574 if (error) 575 goto out_putf; 576 error = -ENXIO; 577 if (lo->lo_state != Lo_bound) 578 goto out_err; 579 580 /* the loop device has to be read-only */ 581 error = -EINVAL; 582 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 583 goto out_err; 584 585 error = loop_validate_file(file, bdev); 586 if (error) 587 goto out_err; 588 589 old_file = lo->lo_backing_file; 590 591 error = -EINVAL; 592 593 /* size of the new backing store needs to be the same */ 594 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 595 goto out_err; 596 597 /* and ... switch */ 598 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 599 blk_mq_freeze_queue(lo->lo_queue); 600 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 601 lo->lo_backing_file = file; 602 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 603 mapping_set_gfp_mask(file->f_mapping, 604 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 605 loop_update_dio(lo); 606 blk_mq_unfreeze_queue(lo->lo_queue); 607 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 608 loop_global_unlock(lo, is_loop); 609 610 /* 611 * Flush loop_validate_file() before fput(), for l->lo_backing_file 612 * might be pointing at old_file which might be the last reference. 613 */ 614 if (!is_loop) { 615 mutex_lock(&loop_validate_mutex); 616 mutex_unlock(&loop_validate_mutex); 617 } 618 /* 619 * We must drop file reference outside of lo_mutex as dropping 620 * the file ref can take open_mutex which creates circular locking 621 * dependency. 622 */ 623 fput(old_file); 624 if (partscan) 625 loop_reread_partitions(lo); 626 return 0; 627 628 out_err: 629 loop_global_unlock(lo, is_loop); 630 out_putf: 631 fput(file); 632 return error; 633 } 634 635 /* loop sysfs attributes */ 636 637 static ssize_t loop_attr_show(struct device *dev, char *page, 638 ssize_t (*callback)(struct loop_device *, char *)) 639 { 640 struct gendisk *disk = dev_to_disk(dev); 641 struct loop_device *lo = disk->private_data; 642 643 return callback(lo, page); 644 } 645 646 #define LOOP_ATTR_RO(_name) \ 647 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 648 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 649 struct device_attribute *attr, char *b) \ 650 { \ 651 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 652 } \ 653 static struct device_attribute loop_attr_##_name = \ 654 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 655 656 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 657 { 658 ssize_t ret; 659 char *p = NULL; 660 661 spin_lock_irq(&lo->lo_lock); 662 if (lo->lo_backing_file) 663 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 664 spin_unlock_irq(&lo->lo_lock); 665 666 if (IS_ERR_OR_NULL(p)) 667 ret = PTR_ERR(p); 668 else { 669 ret = strlen(p); 670 memmove(buf, p, ret); 671 buf[ret++] = '\n'; 672 buf[ret] = 0; 673 } 674 675 return ret; 676 } 677 678 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 679 { 680 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset); 681 } 682 683 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 684 { 685 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 686 } 687 688 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 689 { 690 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 691 692 return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0"); 693 } 694 695 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 696 { 697 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 698 699 return sysfs_emit(buf, "%s\n", partscan ? "1" : "0"); 700 } 701 702 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 703 { 704 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 705 706 return sysfs_emit(buf, "%s\n", dio ? "1" : "0"); 707 } 708 709 LOOP_ATTR_RO(backing_file); 710 LOOP_ATTR_RO(offset); 711 LOOP_ATTR_RO(sizelimit); 712 LOOP_ATTR_RO(autoclear); 713 LOOP_ATTR_RO(partscan); 714 LOOP_ATTR_RO(dio); 715 716 static struct attribute *loop_attrs[] = { 717 &loop_attr_backing_file.attr, 718 &loop_attr_offset.attr, 719 &loop_attr_sizelimit.attr, 720 &loop_attr_autoclear.attr, 721 &loop_attr_partscan.attr, 722 &loop_attr_dio.attr, 723 NULL, 724 }; 725 726 static struct attribute_group loop_attribute_group = { 727 .name = "loop", 728 .attrs= loop_attrs, 729 }; 730 731 static void loop_sysfs_init(struct loop_device *lo) 732 { 733 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 734 &loop_attribute_group); 735 } 736 737 static void loop_sysfs_exit(struct loop_device *lo) 738 { 739 if (lo->sysfs_inited) 740 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 741 &loop_attribute_group); 742 } 743 744 static void loop_config_discard(struct loop_device *lo) 745 { 746 struct file *file = lo->lo_backing_file; 747 struct inode *inode = file->f_mapping->host; 748 struct request_queue *q = lo->lo_queue; 749 u32 granularity, max_discard_sectors; 750 751 /* 752 * If the backing device is a block device, mirror its zeroing 753 * capability. Set the discard sectors to the block device's zeroing 754 * capabilities because loop discards result in blkdev_issue_zeroout(), 755 * not blkdev_issue_discard(). This maintains consistent behavior with 756 * file-backed loop devices: discarded regions read back as zero. 757 */ 758 if (S_ISBLK(inode->i_mode)) { 759 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 760 761 max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 762 granularity = bdev_discard_granularity(I_BDEV(inode)) ?: 763 queue_physical_block_size(backingq); 764 765 /* 766 * We use punch hole to reclaim the free space used by the 767 * image a.k.a. discard. 768 */ 769 } else if (!file->f_op->fallocate) { 770 max_discard_sectors = 0; 771 granularity = 0; 772 773 } else { 774 struct kstatfs sbuf; 775 776 max_discard_sectors = UINT_MAX >> 9; 777 if (!vfs_statfs(&file->f_path, &sbuf)) 778 granularity = sbuf.f_bsize; 779 else 780 max_discard_sectors = 0; 781 } 782 783 if (max_discard_sectors) { 784 q->limits.discard_granularity = granularity; 785 blk_queue_max_discard_sectors(q, max_discard_sectors); 786 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 787 } else { 788 q->limits.discard_granularity = 0; 789 blk_queue_max_discard_sectors(q, 0); 790 blk_queue_max_write_zeroes_sectors(q, 0); 791 } 792 q->limits.discard_alignment = 0; 793 } 794 795 struct loop_worker { 796 struct rb_node rb_node; 797 struct work_struct work; 798 struct list_head cmd_list; 799 struct list_head idle_list; 800 struct loop_device *lo; 801 struct cgroup_subsys_state *blkcg_css; 802 unsigned long last_ran_at; 803 }; 804 805 static void loop_workfn(struct work_struct *work); 806 static void loop_rootcg_workfn(struct work_struct *work); 807 808 #ifdef CONFIG_BLK_CGROUP 809 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 810 { 811 return !css || css == blkcg_root_css; 812 } 813 #else 814 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 815 { 816 return !css; 817 } 818 #endif 819 820 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 821 { 822 struct rb_node **node, *parent = NULL; 823 struct loop_worker *cur_worker, *worker = NULL; 824 struct work_struct *work; 825 struct list_head *cmd_list; 826 827 spin_lock_irq(&lo->lo_work_lock); 828 829 if (queue_on_root_worker(cmd->blkcg_css)) 830 goto queue_work; 831 832 node = &lo->worker_tree.rb_node; 833 834 while (*node) { 835 parent = *node; 836 cur_worker = container_of(*node, struct loop_worker, rb_node); 837 if (cur_worker->blkcg_css == cmd->blkcg_css) { 838 worker = cur_worker; 839 break; 840 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { 841 node = &(*node)->rb_left; 842 } else { 843 node = &(*node)->rb_right; 844 } 845 } 846 if (worker) 847 goto queue_work; 848 849 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 850 /* 851 * In the event we cannot allocate a worker, just queue on the 852 * rootcg worker and issue the I/O as the rootcg 853 */ 854 if (!worker) { 855 cmd->blkcg_css = NULL; 856 if (cmd->memcg_css) 857 css_put(cmd->memcg_css); 858 cmd->memcg_css = NULL; 859 goto queue_work; 860 } 861 862 worker->blkcg_css = cmd->blkcg_css; 863 css_get(worker->blkcg_css); 864 INIT_WORK(&worker->work, loop_workfn); 865 INIT_LIST_HEAD(&worker->cmd_list); 866 INIT_LIST_HEAD(&worker->idle_list); 867 worker->lo = lo; 868 rb_link_node(&worker->rb_node, parent, node); 869 rb_insert_color(&worker->rb_node, &lo->worker_tree); 870 queue_work: 871 if (worker) { 872 /* 873 * We need to remove from the idle list here while 874 * holding the lock so that the idle timer doesn't 875 * free the worker 876 */ 877 if (!list_empty(&worker->idle_list)) 878 list_del_init(&worker->idle_list); 879 work = &worker->work; 880 cmd_list = &worker->cmd_list; 881 } else { 882 work = &lo->rootcg_work; 883 cmd_list = &lo->rootcg_cmd_list; 884 } 885 list_add_tail(&cmd->list_entry, cmd_list); 886 queue_work(lo->workqueue, work); 887 spin_unlock_irq(&lo->lo_work_lock); 888 } 889 890 static void loop_set_timer(struct loop_device *lo) 891 { 892 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 893 } 894 895 static void loop_free_idle_workers(struct loop_device *lo, bool delete_all) 896 { 897 struct loop_worker *pos, *worker; 898 899 spin_lock_irq(&lo->lo_work_lock); 900 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 901 idle_list) { 902 if (!delete_all && 903 time_is_after_jiffies(worker->last_ran_at + 904 LOOP_IDLE_WORKER_TIMEOUT)) 905 break; 906 list_del(&worker->idle_list); 907 rb_erase(&worker->rb_node, &lo->worker_tree); 908 css_put(worker->blkcg_css); 909 kfree(worker); 910 } 911 if (!list_empty(&lo->idle_worker_list)) 912 loop_set_timer(lo); 913 spin_unlock_irq(&lo->lo_work_lock); 914 } 915 916 static void loop_free_idle_workers_timer(struct timer_list *timer) 917 { 918 struct loop_device *lo = container_of(timer, struct loop_device, timer); 919 920 return loop_free_idle_workers(lo, false); 921 } 922 923 static void loop_update_rotational(struct loop_device *lo) 924 { 925 struct file *file = lo->lo_backing_file; 926 struct inode *file_inode = file->f_mapping->host; 927 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 928 struct request_queue *q = lo->lo_queue; 929 bool nonrot = true; 930 931 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 932 if (file_bdev) 933 nonrot = bdev_nonrot(file_bdev); 934 935 if (nonrot) 936 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 937 else 938 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 939 } 940 941 /** 942 * loop_set_status_from_info - configure device from loop_info 943 * @lo: struct loop_device to configure 944 * @info: struct loop_info64 to configure the device with 945 * 946 * Configures the loop device parameters according to the passed 947 * in loop_info64 configuration. 948 */ 949 static int 950 loop_set_status_from_info(struct loop_device *lo, 951 const struct loop_info64 *info) 952 { 953 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 954 return -EINVAL; 955 956 switch (info->lo_encrypt_type) { 957 case LO_CRYPT_NONE: 958 break; 959 case LO_CRYPT_XOR: 960 pr_warn("support for the xor transformation has been removed.\n"); 961 return -EINVAL; 962 case LO_CRYPT_CRYPTOAPI: 963 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n"); 964 return -EINVAL; 965 default: 966 return -EINVAL; 967 } 968 969 lo->lo_offset = info->lo_offset; 970 lo->lo_sizelimit = info->lo_sizelimit; 971 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 972 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 973 lo->lo_flags = info->lo_flags; 974 return 0; 975 } 976 977 static int loop_configure(struct loop_device *lo, fmode_t mode, 978 struct block_device *bdev, 979 const struct loop_config *config) 980 { 981 struct file *file = fget(config->fd); 982 struct inode *inode; 983 struct address_space *mapping; 984 int error; 985 loff_t size; 986 bool partscan; 987 unsigned short bsize; 988 bool is_loop; 989 990 if (!file) 991 return -EBADF; 992 is_loop = is_loop_device(file); 993 994 /* This is safe, since we have a reference from open(). */ 995 __module_get(THIS_MODULE); 996 997 /* 998 * If we don't hold exclusive handle for the device, upgrade to it 999 * here to avoid changing device under exclusive owner. 1000 */ 1001 if (!(mode & FMODE_EXCL)) { 1002 error = bd_prepare_to_claim(bdev, loop_configure); 1003 if (error) 1004 goto out_putf; 1005 } 1006 1007 error = loop_global_lock_killable(lo, is_loop); 1008 if (error) 1009 goto out_bdev; 1010 1011 error = -EBUSY; 1012 if (lo->lo_state != Lo_unbound) 1013 goto out_unlock; 1014 1015 error = loop_validate_file(file, bdev); 1016 if (error) 1017 goto out_unlock; 1018 1019 mapping = file->f_mapping; 1020 inode = mapping->host; 1021 1022 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1023 error = -EINVAL; 1024 goto out_unlock; 1025 } 1026 1027 if (config->block_size) { 1028 error = blk_validate_block_size(config->block_size); 1029 if (error) 1030 goto out_unlock; 1031 } 1032 1033 error = loop_set_status_from_info(lo, &config->info); 1034 if (error) 1035 goto out_unlock; 1036 1037 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1038 !file->f_op->write_iter) 1039 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1040 1041 lo->workqueue = alloc_workqueue("loop%d", 1042 WQ_UNBOUND | WQ_FREEZABLE, 1043 0, 1044 lo->lo_number); 1045 if (!lo->workqueue) { 1046 error = -ENOMEM; 1047 goto out_unlock; 1048 } 1049 1050 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1051 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1052 1053 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 1054 INIT_LIST_HEAD(&lo->rootcg_cmd_list); 1055 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1056 lo->lo_device = bdev; 1057 lo->lo_backing_file = file; 1058 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1059 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1060 1061 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1062 blk_queue_write_cache(lo->lo_queue, true, false); 1063 1064 if (config->block_size) 1065 bsize = config->block_size; 1066 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1067 /* In case of direct I/O, match underlying block size */ 1068 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1069 else 1070 bsize = 512; 1071 1072 blk_queue_logical_block_size(lo->lo_queue, bsize); 1073 blk_queue_physical_block_size(lo->lo_queue, bsize); 1074 blk_queue_io_min(lo->lo_queue, bsize); 1075 1076 loop_config_discard(lo); 1077 loop_update_rotational(lo); 1078 loop_update_dio(lo); 1079 loop_sysfs_init(lo); 1080 1081 size = get_loop_size(lo, file); 1082 loop_set_size(lo, size); 1083 1084 /* Order wrt reading lo_state in loop_validate_file(). */ 1085 wmb(); 1086 1087 lo->lo_state = Lo_bound; 1088 if (part_shift) 1089 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1090 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1091 if (partscan) 1092 lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 1093 1094 loop_global_unlock(lo, is_loop); 1095 if (partscan) 1096 loop_reread_partitions(lo); 1097 if (!(mode & FMODE_EXCL)) 1098 bd_abort_claiming(bdev, loop_configure); 1099 return 0; 1100 1101 out_unlock: 1102 loop_global_unlock(lo, is_loop); 1103 out_bdev: 1104 if (!(mode & FMODE_EXCL)) 1105 bd_abort_claiming(bdev, loop_configure); 1106 out_putf: 1107 fput(file); 1108 /* This is safe: open() is still holding a reference. */ 1109 module_put(THIS_MODULE); 1110 return error; 1111 } 1112 1113 static void __loop_clr_fd(struct loop_device *lo, bool release) 1114 { 1115 struct file *filp; 1116 gfp_t gfp = lo->old_gfp_mask; 1117 1118 /* 1119 * Flush loop_configure() and loop_change_fd(). It is acceptable for 1120 * loop_validate_file() to succeed, for actual clear operation has not 1121 * started yet. 1122 */ 1123 mutex_lock(&loop_validate_mutex); 1124 mutex_unlock(&loop_validate_mutex); 1125 /* 1126 * loop_validate_file() now fails because l->lo_state != Lo_bound 1127 * became visible. 1128 */ 1129 1130 /* 1131 * Since this function is called upon "ioctl(LOOP_CLR_FD)" xor "close() 1132 * after ioctl(LOOP_CLR_FD)", it is a sign of something going wrong if 1133 * lo->lo_state has changed while waiting for lo->lo_mutex. 1134 */ 1135 mutex_lock(&lo->lo_mutex); 1136 BUG_ON(lo->lo_state != Lo_rundown); 1137 mutex_unlock(&lo->lo_mutex); 1138 1139 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1140 blk_queue_write_cache(lo->lo_queue, false, false); 1141 1142 /* 1143 * Freeze the request queue when unbinding on a live file descriptor and 1144 * thus an open device. When called from ->release we are guaranteed 1145 * that there is no I/O in progress already. 1146 */ 1147 if (!release) 1148 blk_mq_freeze_queue(lo->lo_queue); 1149 1150 destroy_workqueue(lo->workqueue); 1151 loop_free_idle_workers(lo, true); 1152 del_timer_sync(&lo->timer); 1153 1154 spin_lock_irq(&lo->lo_lock); 1155 filp = lo->lo_backing_file; 1156 lo->lo_backing_file = NULL; 1157 spin_unlock_irq(&lo->lo_lock); 1158 1159 lo->lo_device = NULL; 1160 lo->lo_offset = 0; 1161 lo->lo_sizelimit = 0; 1162 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1163 blk_queue_logical_block_size(lo->lo_queue, 512); 1164 blk_queue_physical_block_size(lo->lo_queue, 512); 1165 blk_queue_io_min(lo->lo_queue, 512); 1166 invalidate_disk(lo->lo_disk); 1167 loop_sysfs_exit(lo); 1168 /* let user-space know about this change */ 1169 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 1170 mapping_set_gfp_mask(filp->f_mapping, gfp); 1171 /* This is safe: open() is still holding a reference. */ 1172 module_put(THIS_MODULE); 1173 if (!release) 1174 blk_mq_unfreeze_queue(lo->lo_queue); 1175 1176 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1177 1178 if (lo->lo_flags & LO_FLAGS_PARTSCAN) { 1179 int err; 1180 1181 /* 1182 * open_mutex has been held already in release path, so don't 1183 * acquire it if this function is called in such case. 1184 * 1185 * If the reread partition isn't from release path, lo_refcnt 1186 * must be at least one and it can only become zero when the 1187 * current holder is released. 1188 */ 1189 if (!release) 1190 mutex_lock(&lo->lo_disk->open_mutex); 1191 err = bdev_disk_changed(lo->lo_disk, false); 1192 if (!release) 1193 mutex_unlock(&lo->lo_disk->open_mutex); 1194 if (err) 1195 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1196 __func__, lo->lo_number, err); 1197 /* Device is gone, no point in returning error */ 1198 } 1199 1200 /* 1201 * lo->lo_state is set to Lo_unbound here after above partscan has 1202 * finished. There cannot be anybody else entering __loop_clr_fd() as 1203 * Lo_rundown state protects us from all the other places trying to 1204 * change the 'lo' device. 1205 */ 1206 lo->lo_flags = 0; 1207 if (!part_shift) 1208 lo->lo_disk->flags |= GENHD_FL_NO_PART; 1209 mutex_lock(&lo->lo_mutex); 1210 lo->lo_state = Lo_unbound; 1211 mutex_unlock(&lo->lo_mutex); 1212 1213 /* 1214 * Need not hold lo_mutex to fput backing file. Calling fput holding 1215 * lo_mutex triggers a circular lock dependency possibility warning as 1216 * fput can take open_mutex which is usually taken before lo_mutex. 1217 */ 1218 fput(filp); 1219 } 1220 1221 static int loop_clr_fd(struct loop_device *lo) 1222 { 1223 int err; 1224 1225 err = mutex_lock_killable(&lo->lo_mutex); 1226 if (err) 1227 return err; 1228 if (lo->lo_state != Lo_bound) { 1229 mutex_unlock(&lo->lo_mutex); 1230 return -ENXIO; 1231 } 1232 /* 1233 * If we've explicitly asked to tear down the loop device, 1234 * and it has an elevated reference count, set it for auto-teardown when 1235 * the last reference goes away. This stops $!~#$@ udev from 1236 * preventing teardown because it decided that it needs to run blkid on 1237 * the loopback device whenever they appear. xfstests is notorious for 1238 * failing tests because blkid via udev races with a losetup 1239 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1240 * command to fail with EBUSY. 1241 */ 1242 if (atomic_read(&lo->lo_refcnt) > 1) { 1243 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1244 mutex_unlock(&lo->lo_mutex); 1245 return 0; 1246 } 1247 lo->lo_state = Lo_rundown; 1248 mutex_unlock(&lo->lo_mutex); 1249 1250 __loop_clr_fd(lo, false); 1251 return 0; 1252 } 1253 1254 static int 1255 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1256 { 1257 int err; 1258 int prev_lo_flags; 1259 bool partscan = false; 1260 bool size_changed = false; 1261 1262 err = mutex_lock_killable(&lo->lo_mutex); 1263 if (err) 1264 return err; 1265 if (lo->lo_state != Lo_bound) { 1266 err = -ENXIO; 1267 goto out_unlock; 1268 } 1269 1270 if (lo->lo_offset != info->lo_offset || 1271 lo->lo_sizelimit != info->lo_sizelimit) { 1272 size_changed = true; 1273 sync_blockdev(lo->lo_device); 1274 invalidate_bdev(lo->lo_device); 1275 } 1276 1277 /* I/O need to be drained during transfer transition */ 1278 blk_mq_freeze_queue(lo->lo_queue); 1279 1280 prev_lo_flags = lo->lo_flags; 1281 1282 err = loop_set_status_from_info(lo, info); 1283 if (err) 1284 goto out_unfreeze; 1285 1286 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1287 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1288 /* For those flags, use the previous values instead */ 1289 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1290 /* For flags that can't be cleared, use previous values too */ 1291 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1292 1293 if (size_changed) { 1294 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1295 lo->lo_backing_file); 1296 loop_set_size(lo, new_size); 1297 } 1298 1299 loop_config_discard(lo); 1300 1301 /* update dio if lo_offset or transfer is changed */ 1302 __loop_update_dio(lo, lo->use_dio); 1303 1304 out_unfreeze: 1305 blk_mq_unfreeze_queue(lo->lo_queue); 1306 1307 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1308 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1309 lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 1310 partscan = true; 1311 } 1312 out_unlock: 1313 mutex_unlock(&lo->lo_mutex); 1314 if (partscan) 1315 loop_reread_partitions(lo); 1316 1317 return err; 1318 } 1319 1320 static int 1321 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1322 { 1323 struct path path; 1324 struct kstat stat; 1325 int ret; 1326 1327 ret = mutex_lock_killable(&lo->lo_mutex); 1328 if (ret) 1329 return ret; 1330 if (lo->lo_state != Lo_bound) { 1331 mutex_unlock(&lo->lo_mutex); 1332 return -ENXIO; 1333 } 1334 1335 memset(info, 0, sizeof(*info)); 1336 info->lo_number = lo->lo_number; 1337 info->lo_offset = lo->lo_offset; 1338 info->lo_sizelimit = lo->lo_sizelimit; 1339 info->lo_flags = lo->lo_flags; 1340 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1341 1342 /* Drop lo_mutex while we call into the filesystem. */ 1343 path = lo->lo_backing_file->f_path; 1344 path_get(&path); 1345 mutex_unlock(&lo->lo_mutex); 1346 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1347 if (!ret) { 1348 info->lo_device = huge_encode_dev(stat.dev); 1349 info->lo_inode = stat.ino; 1350 info->lo_rdevice = huge_encode_dev(stat.rdev); 1351 } 1352 path_put(&path); 1353 return ret; 1354 } 1355 1356 static void 1357 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1358 { 1359 memset(info64, 0, sizeof(*info64)); 1360 info64->lo_number = info->lo_number; 1361 info64->lo_device = info->lo_device; 1362 info64->lo_inode = info->lo_inode; 1363 info64->lo_rdevice = info->lo_rdevice; 1364 info64->lo_offset = info->lo_offset; 1365 info64->lo_sizelimit = 0; 1366 info64->lo_flags = info->lo_flags; 1367 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1368 } 1369 1370 static int 1371 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1372 { 1373 memset(info, 0, sizeof(*info)); 1374 info->lo_number = info64->lo_number; 1375 info->lo_device = info64->lo_device; 1376 info->lo_inode = info64->lo_inode; 1377 info->lo_rdevice = info64->lo_rdevice; 1378 info->lo_offset = info64->lo_offset; 1379 info->lo_flags = info64->lo_flags; 1380 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1381 1382 /* error in case values were truncated */ 1383 if (info->lo_device != info64->lo_device || 1384 info->lo_rdevice != info64->lo_rdevice || 1385 info->lo_inode != info64->lo_inode || 1386 info->lo_offset != info64->lo_offset) 1387 return -EOVERFLOW; 1388 1389 return 0; 1390 } 1391 1392 static int 1393 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1394 { 1395 struct loop_info info; 1396 struct loop_info64 info64; 1397 1398 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1399 return -EFAULT; 1400 loop_info64_from_old(&info, &info64); 1401 return loop_set_status(lo, &info64); 1402 } 1403 1404 static int 1405 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1406 { 1407 struct loop_info64 info64; 1408 1409 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1410 return -EFAULT; 1411 return loop_set_status(lo, &info64); 1412 } 1413 1414 static int 1415 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1416 struct loop_info info; 1417 struct loop_info64 info64; 1418 int err; 1419 1420 if (!arg) 1421 return -EINVAL; 1422 err = loop_get_status(lo, &info64); 1423 if (!err) 1424 err = loop_info64_to_old(&info64, &info); 1425 if (!err && copy_to_user(arg, &info, sizeof(info))) 1426 err = -EFAULT; 1427 1428 return err; 1429 } 1430 1431 static int 1432 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1433 struct loop_info64 info64; 1434 int err; 1435 1436 if (!arg) 1437 return -EINVAL; 1438 err = loop_get_status(lo, &info64); 1439 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1440 err = -EFAULT; 1441 1442 return err; 1443 } 1444 1445 static int loop_set_capacity(struct loop_device *lo) 1446 { 1447 loff_t size; 1448 1449 if (unlikely(lo->lo_state != Lo_bound)) 1450 return -ENXIO; 1451 1452 size = get_loop_size(lo, lo->lo_backing_file); 1453 loop_set_size(lo, size); 1454 1455 return 0; 1456 } 1457 1458 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1459 { 1460 int error = -ENXIO; 1461 if (lo->lo_state != Lo_bound) 1462 goto out; 1463 1464 __loop_update_dio(lo, !!arg); 1465 if (lo->use_dio == !!arg) 1466 return 0; 1467 error = -EINVAL; 1468 out: 1469 return error; 1470 } 1471 1472 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1473 { 1474 int err = 0; 1475 1476 if (lo->lo_state != Lo_bound) 1477 return -ENXIO; 1478 1479 err = blk_validate_block_size(arg); 1480 if (err) 1481 return err; 1482 1483 if (lo->lo_queue->limits.logical_block_size == arg) 1484 return 0; 1485 1486 sync_blockdev(lo->lo_device); 1487 invalidate_bdev(lo->lo_device); 1488 1489 blk_mq_freeze_queue(lo->lo_queue); 1490 blk_queue_logical_block_size(lo->lo_queue, arg); 1491 blk_queue_physical_block_size(lo->lo_queue, arg); 1492 blk_queue_io_min(lo->lo_queue, arg); 1493 loop_update_dio(lo); 1494 blk_mq_unfreeze_queue(lo->lo_queue); 1495 1496 return err; 1497 } 1498 1499 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1500 unsigned long arg) 1501 { 1502 int err; 1503 1504 err = mutex_lock_killable(&lo->lo_mutex); 1505 if (err) 1506 return err; 1507 switch (cmd) { 1508 case LOOP_SET_CAPACITY: 1509 err = loop_set_capacity(lo); 1510 break; 1511 case LOOP_SET_DIRECT_IO: 1512 err = loop_set_dio(lo, arg); 1513 break; 1514 case LOOP_SET_BLOCK_SIZE: 1515 err = loop_set_block_size(lo, arg); 1516 break; 1517 default: 1518 err = -EINVAL; 1519 } 1520 mutex_unlock(&lo->lo_mutex); 1521 return err; 1522 } 1523 1524 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1525 unsigned int cmd, unsigned long arg) 1526 { 1527 struct loop_device *lo = bdev->bd_disk->private_data; 1528 void __user *argp = (void __user *) arg; 1529 int err; 1530 1531 switch (cmd) { 1532 case LOOP_SET_FD: { 1533 /* 1534 * Legacy case - pass in a zeroed out struct loop_config with 1535 * only the file descriptor set , which corresponds with the 1536 * default parameters we'd have used otherwise. 1537 */ 1538 struct loop_config config; 1539 1540 memset(&config, 0, sizeof(config)); 1541 config.fd = arg; 1542 1543 return loop_configure(lo, mode, bdev, &config); 1544 } 1545 case LOOP_CONFIGURE: { 1546 struct loop_config config; 1547 1548 if (copy_from_user(&config, argp, sizeof(config))) 1549 return -EFAULT; 1550 1551 return loop_configure(lo, mode, bdev, &config); 1552 } 1553 case LOOP_CHANGE_FD: 1554 return loop_change_fd(lo, bdev, arg); 1555 case LOOP_CLR_FD: 1556 return loop_clr_fd(lo); 1557 case LOOP_SET_STATUS: 1558 err = -EPERM; 1559 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1560 err = loop_set_status_old(lo, argp); 1561 } 1562 break; 1563 case LOOP_GET_STATUS: 1564 return loop_get_status_old(lo, argp); 1565 case LOOP_SET_STATUS64: 1566 err = -EPERM; 1567 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1568 err = loop_set_status64(lo, argp); 1569 } 1570 break; 1571 case LOOP_GET_STATUS64: 1572 return loop_get_status64(lo, argp); 1573 case LOOP_SET_CAPACITY: 1574 case LOOP_SET_DIRECT_IO: 1575 case LOOP_SET_BLOCK_SIZE: 1576 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1577 return -EPERM; 1578 fallthrough; 1579 default: 1580 err = lo_simple_ioctl(lo, cmd, arg); 1581 break; 1582 } 1583 1584 return err; 1585 } 1586 1587 #ifdef CONFIG_COMPAT 1588 struct compat_loop_info { 1589 compat_int_t lo_number; /* ioctl r/o */ 1590 compat_dev_t lo_device; /* ioctl r/o */ 1591 compat_ulong_t lo_inode; /* ioctl r/o */ 1592 compat_dev_t lo_rdevice; /* ioctl r/o */ 1593 compat_int_t lo_offset; 1594 compat_int_t lo_encrypt_type; /* obsolete, ignored */ 1595 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1596 compat_int_t lo_flags; /* ioctl r/o */ 1597 char lo_name[LO_NAME_SIZE]; 1598 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1599 compat_ulong_t lo_init[2]; 1600 char reserved[4]; 1601 }; 1602 1603 /* 1604 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1605 * - noinlined to reduce stack space usage in main part of driver 1606 */ 1607 static noinline int 1608 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1609 struct loop_info64 *info64) 1610 { 1611 struct compat_loop_info info; 1612 1613 if (copy_from_user(&info, arg, sizeof(info))) 1614 return -EFAULT; 1615 1616 memset(info64, 0, sizeof(*info64)); 1617 info64->lo_number = info.lo_number; 1618 info64->lo_device = info.lo_device; 1619 info64->lo_inode = info.lo_inode; 1620 info64->lo_rdevice = info.lo_rdevice; 1621 info64->lo_offset = info.lo_offset; 1622 info64->lo_sizelimit = 0; 1623 info64->lo_flags = info.lo_flags; 1624 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1625 return 0; 1626 } 1627 1628 /* 1629 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1630 * - noinlined to reduce stack space usage in main part of driver 1631 */ 1632 static noinline int 1633 loop_info64_to_compat(const struct loop_info64 *info64, 1634 struct compat_loop_info __user *arg) 1635 { 1636 struct compat_loop_info info; 1637 1638 memset(&info, 0, sizeof(info)); 1639 info.lo_number = info64->lo_number; 1640 info.lo_device = info64->lo_device; 1641 info.lo_inode = info64->lo_inode; 1642 info.lo_rdevice = info64->lo_rdevice; 1643 info.lo_offset = info64->lo_offset; 1644 info.lo_flags = info64->lo_flags; 1645 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1646 1647 /* error in case values were truncated */ 1648 if (info.lo_device != info64->lo_device || 1649 info.lo_rdevice != info64->lo_rdevice || 1650 info.lo_inode != info64->lo_inode || 1651 info.lo_offset != info64->lo_offset) 1652 return -EOVERFLOW; 1653 1654 if (copy_to_user(arg, &info, sizeof(info))) 1655 return -EFAULT; 1656 return 0; 1657 } 1658 1659 static int 1660 loop_set_status_compat(struct loop_device *lo, 1661 const struct compat_loop_info __user *arg) 1662 { 1663 struct loop_info64 info64; 1664 int ret; 1665 1666 ret = loop_info64_from_compat(arg, &info64); 1667 if (ret < 0) 1668 return ret; 1669 return loop_set_status(lo, &info64); 1670 } 1671 1672 static int 1673 loop_get_status_compat(struct loop_device *lo, 1674 struct compat_loop_info __user *arg) 1675 { 1676 struct loop_info64 info64; 1677 int err; 1678 1679 if (!arg) 1680 return -EINVAL; 1681 err = loop_get_status(lo, &info64); 1682 if (!err) 1683 err = loop_info64_to_compat(&info64, arg); 1684 return err; 1685 } 1686 1687 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1688 unsigned int cmd, unsigned long arg) 1689 { 1690 struct loop_device *lo = bdev->bd_disk->private_data; 1691 int err; 1692 1693 switch(cmd) { 1694 case LOOP_SET_STATUS: 1695 err = loop_set_status_compat(lo, 1696 (const struct compat_loop_info __user *)arg); 1697 break; 1698 case LOOP_GET_STATUS: 1699 err = loop_get_status_compat(lo, 1700 (struct compat_loop_info __user *)arg); 1701 break; 1702 case LOOP_SET_CAPACITY: 1703 case LOOP_CLR_FD: 1704 case LOOP_GET_STATUS64: 1705 case LOOP_SET_STATUS64: 1706 case LOOP_CONFIGURE: 1707 arg = (unsigned long) compat_ptr(arg); 1708 fallthrough; 1709 case LOOP_SET_FD: 1710 case LOOP_CHANGE_FD: 1711 case LOOP_SET_BLOCK_SIZE: 1712 case LOOP_SET_DIRECT_IO: 1713 err = lo_ioctl(bdev, mode, cmd, arg); 1714 break; 1715 default: 1716 err = -ENOIOCTLCMD; 1717 break; 1718 } 1719 return err; 1720 } 1721 #endif 1722 1723 static int lo_open(struct block_device *bdev, fmode_t mode) 1724 { 1725 struct loop_device *lo = bdev->bd_disk->private_data; 1726 int err; 1727 1728 err = mutex_lock_killable(&lo->lo_mutex); 1729 if (err) 1730 return err; 1731 if (lo->lo_state == Lo_deleting) 1732 err = -ENXIO; 1733 else 1734 atomic_inc(&lo->lo_refcnt); 1735 mutex_unlock(&lo->lo_mutex); 1736 return err; 1737 } 1738 1739 static void lo_release(struct gendisk *disk, fmode_t mode) 1740 { 1741 struct loop_device *lo = disk->private_data; 1742 1743 mutex_lock(&lo->lo_mutex); 1744 if (atomic_dec_return(&lo->lo_refcnt)) 1745 goto out_unlock; 1746 1747 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1748 if (lo->lo_state != Lo_bound) 1749 goto out_unlock; 1750 lo->lo_state = Lo_rundown; 1751 mutex_unlock(&lo->lo_mutex); 1752 /* 1753 * In autoclear mode, stop the loop thread 1754 * and remove configuration after last close. 1755 */ 1756 __loop_clr_fd(lo, true); 1757 return; 1758 } 1759 1760 out_unlock: 1761 mutex_unlock(&lo->lo_mutex); 1762 } 1763 1764 static void lo_free_disk(struct gendisk *disk) 1765 { 1766 struct loop_device *lo = disk->private_data; 1767 1768 mutex_destroy(&lo->lo_mutex); 1769 kfree(lo); 1770 } 1771 1772 static const struct block_device_operations lo_fops = { 1773 .owner = THIS_MODULE, 1774 .open = lo_open, 1775 .release = lo_release, 1776 .ioctl = lo_ioctl, 1777 #ifdef CONFIG_COMPAT 1778 .compat_ioctl = lo_compat_ioctl, 1779 #endif 1780 .free_disk = lo_free_disk, 1781 }; 1782 1783 /* 1784 * And now the modules code and kernel interface. 1785 */ 1786 static int max_loop; 1787 module_param(max_loop, int, 0444); 1788 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1789 module_param(max_part, int, 0444); 1790 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1791 1792 static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH; 1793 1794 static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p) 1795 { 1796 int ret = kstrtoint(s, 10, &hw_queue_depth); 1797 1798 return (ret || (hw_queue_depth < 1)) ? -EINVAL : 0; 1799 } 1800 1801 static const struct kernel_param_ops loop_hw_qdepth_param_ops = { 1802 .set = loop_set_hw_queue_depth, 1803 .get = param_get_int, 1804 }; 1805 1806 device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444); 1807 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 128"); 1808 1809 MODULE_LICENSE("GPL"); 1810 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1811 1812 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1813 const struct blk_mq_queue_data *bd) 1814 { 1815 struct request *rq = bd->rq; 1816 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1817 struct loop_device *lo = rq->q->queuedata; 1818 1819 blk_mq_start_request(rq); 1820 1821 if (lo->lo_state != Lo_bound) 1822 return BLK_STS_IOERR; 1823 1824 switch (req_op(rq)) { 1825 case REQ_OP_FLUSH: 1826 case REQ_OP_DISCARD: 1827 case REQ_OP_WRITE_ZEROES: 1828 cmd->use_aio = false; 1829 break; 1830 default: 1831 cmd->use_aio = lo->use_dio; 1832 break; 1833 } 1834 1835 /* always use the first bio's css */ 1836 cmd->blkcg_css = NULL; 1837 cmd->memcg_css = NULL; 1838 #ifdef CONFIG_BLK_CGROUP 1839 if (rq->bio && rq->bio->bi_blkg) { 1840 cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 1841 #ifdef CONFIG_MEMCG 1842 cmd->memcg_css = 1843 cgroup_get_e_css(cmd->blkcg_css->cgroup, 1844 &memory_cgrp_subsys); 1845 #endif 1846 } 1847 #endif 1848 loop_queue_work(lo, cmd); 1849 1850 return BLK_STS_OK; 1851 } 1852 1853 static void loop_handle_cmd(struct loop_cmd *cmd) 1854 { 1855 struct request *rq = blk_mq_rq_from_pdu(cmd); 1856 const bool write = op_is_write(req_op(rq)); 1857 struct loop_device *lo = rq->q->queuedata; 1858 int ret = 0; 1859 struct mem_cgroup *old_memcg = NULL; 1860 1861 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1862 ret = -EIO; 1863 goto failed; 1864 } 1865 1866 if (cmd->blkcg_css) 1867 kthread_associate_blkcg(cmd->blkcg_css); 1868 if (cmd->memcg_css) 1869 old_memcg = set_active_memcg( 1870 mem_cgroup_from_css(cmd->memcg_css)); 1871 1872 ret = do_req_filebacked(lo, rq); 1873 1874 if (cmd->blkcg_css) 1875 kthread_associate_blkcg(NULL); 1876 1877 if (cmd->memcg_css) { 1878 set_active_memcg(old_memcg); 1879 css_put(cmd->memcg_css); 1880 } 1881 failed: 1882 /* complete non-aio request */ 1883 if (!cmd->use_aio || ret) { 1884 if (ret == -EOPNOTSUPP) 1885 cmd->ret = ret; 1886 else 1887 cmd->ret = ret ? -EIO : 0; 1888 if (likely(!blk_should_fake_timeout(rq->q))) 1889 blk_mq_complete_request(rq); 1890 } 1891 } 1892 1893 static void loop_process_work(struct loop_worker *worker, 1894 struct list_head *cmd_list, struct loop_device *lo) 1895 { 1896 int orig_flags = current->flags; 1897 struct loop_cmd *cmd; 1898 1899 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 1900 spin_lock_irq(&lo->lo_work_lock); 1901 while (!list_empty(cmd_list)) { 1902 cmd = container_of( 1903 cmd_list->next, struct loop_cmd, list_entry); 1904 list_del(cmd_list->next); 1905 spin_unlock_irq(&lo->lo_work_lock); 1906 1907 loop_handle_cmd(cmd); 1908 cond_resched(); 1909 1910 spin_lock_irq(&lo->lo_work_lock); 1911 } 1912 1913 /* 1914 * We only add to the idle list if there are no pending cmds 1915 * *and* the worker will not run again which ensures that it 1916 * is safe to free any worker on the idle list 1917 */ 1918 if (worker && !work_pending(&worker->work)) { 1919 worker->last_ran_at = jiffies; 1920 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 1921 loop_set_timer(lo); 1922 } 1923 spin_unlock_irq(&lo->lo_work_lock); 1924 current->flags = orig_flags; 1925 } 1926 1927 static void loop_workfn(struct work_struct *work) 1928 { 1929 struct loop_worker *worker = 1930 container_of(work, struct loop_worker, work); 1931 loop_process_work(worker, &worker->cmd_list, worker->lo); 1932 } 1933 1934 static void loop_rootcg_workfn(struct work_struct *work) 1935 { 1936 struct loop_device *lo = 1937 container_of(work, struct loop_device, rootcg_work); 1938 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 1939 } 1940 1941 static const struct blk_mq_ops loop_mq_ops = { 1942 .queue_rq = loop_queue_rq, 1943 .complete = lo_complete_rq, 1944 }; 1945 1946 static int loop_add(int i) 1947 { 1948 struct loop_device *lo; 1949 struct gendisk *disk; 1950 int err; 1951 1952 err = -ENOMEM; 1953 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1954 if (!lo) 1955 goto out; 1956 lo->worker_tree = RB_ROOT; 1957 INIT_LIST_HEAD(&lo->idle_worker_list); 1958 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE); 1959 lo->lo_state = Lo_unbound; 1960 1961 err = mutex_lock_killable(&loop_ctl_mutex); 1962 if (err) 1963 goto out_free_dev; 1964 1965 /* allocate id, if @id >= 0, we're requesting that specific id */ 1966 if (i >= 0) { 1967 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1968 if (err == -ENOSPC) 1969 err = -EEXIST; 1970 } else { 1971 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1972 } 1973 mutex_unlock(&loop_ctl_mutex); 1974 if (err < 0) 1975 goto out_free_dev; 1976 i = err; 1977 1978 lo->tag_set.ops = &loop_mq_ops; 1979 lo->tag_set.nr_hw_queues = 1; 1980 lo->tag_set.queue_depth = hw_queue_depth; 1981 lo->tag_set.numa_node = NUMA_NO_NODE; 1982 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1983 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | 1984 BLK_MQ_F_NO_SCHED_BY_DEFAULT; 1985 lo->tag_set.driver_data = lo; 1986 1987 err = blk_mq_alloc_tag_set(&lo->tag_set); 1988 if (err) 1989 goto out_free_idr; 1990 1991 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 1992 if (IS_ERR(disk)) { 1993 err = PTR_ERR(disk); 1994 goto out_cleanup_tags; 1995 } 1996 lo->lo_queue = lo->lo_disk->queue; 1997 1998 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1999 2000 /* 2001 * By default, we do buffer IO, so it doesn't make sense to enable 2002 * merge because the I/O submitted to backing file is handled page by 2003 * page. For directio mode, merge does help to dispatch bigger request 2004 * to underlayer disk. We will enable merge once directio is enabled. 2005 */ 2006 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2007 2008 /* 2009 * Disable partition scanning by default. The in-kernel partition 2010 * scanning can be requested individually per-device during its 2011 * setup. Userspace can always add and remove partitions from all 2012 * devices. The needed partition minors are allocated from the 2013 * extended minor space, the main loop device numbers will continue 2014 * to match the loop minors, regardless of the number of partitions 2015 * used. 2016 * 2017 * If max_part is given, partition scanning is globally enabled for 2018 * all loop devices. The minors for the main loop devices will be 2019 * multiples of max_part. 2020 * 2021 * Note: Global-for-all-devices, set-only-at-init, read-only module 2022 * parameteters like 'max_loop' and 'max_part' make things needlessly 2023 * complicated, are too static, inflexible and may surprise 2024 * userspace tools. Parameters like this in general should be avoided. 2025 */ 2026 if (!part_shift) 2027 disk->flags |= GENHD_FL_NO_PART; 2028 atomic_set(&lo->lo_refcnt, 0); 2029 mutex_init(&lo->lo_mutex); 2030 lo->lo_number = i; 2031 spin_lock_init(&lo->lo_lock); 2032 spin_lock_init(&lo->lo_work_lock); 2033 disk->major = LOOP_MAJOR; 2034 disk->first_minor = i << part_shift; 2035 disk->minors = 1 << part_shift; 2036 disk->fops = &lo_fops; 2037 disk->private_data = lo; 2038 disk->queue = lo->lo_queue; 2039 disk->events = DISK_EVENT_MEDIA_CHANGE; 2040 disk->event_flags = DISK_EVENT_FLAG_UEVENT; 2041 sprintf(disk->disk_name, "loop%d", i); 2042 /* Make this loop device reachable from pathname. */ 2043 err = add_disk(disk); 2044 if (err) 2045 goto out_cleanup_disk; 2046 2047 /* Show this loop device. */ 2048 mutex_lock(&loop_ctl_mutex); 2049 lo->idr_visible = true; 2050 mutex_unlock(&loop_ctl_mutex); 2051 2052 return i; 2053 2054 out_cleanup_disk: 2055 blk_cleanup_disk(disk); 2056 out_cleanup_tags: 2057 blk_mq_free_tag_set(&lo->tag_set); 2058 out_free_idr: 2059 mutex_lock(&loop_ctl_mutex); 2060 idr_remove(&loop_index_idr, i); 2061 mutex_unlock(&loop_ctl_mutex); 2062 out_free_dev: 2063 kfree(lo); 2064 out: 2065 return err; 2066 } 2067 2068 static void loop_remove(struct loop_device *lo) 2069 { 2070 /* Make this loop device unreachable from pathname. */ 2071 del_gendisk(lo->lo_disk); 2072 blk_cleanup_queue(lo->lo_disk->queue); 2073 blk_mq_free_tag_set(&lo->tag_set); 2074 2075 mutex_lock(&loop_ctl_mutex); 2076 idr_remove(&loop_index_idr, lo->lo_number); 2077 mutex_unlock(&loop_ctl_mutex); 2078 2079 put_disk(lo->lo_disk); 2080 } 2081 2082 static void loop_probe(dev_t dev) 2083 { 2084 int idx = MINOR(dev) >> part_shift; 2085 2086 if (max_loop && idx >= max_loop) 2087 return; 2088 loop_add(idx); 2089 } 2090 2091 static int loop_control_remove(int idx) 2092 { 2093 struct loop_device *lo; 2094 int ret; 2095 2096 if (idx < 0) { 2097 pr_warn_once("deleting an unspecified loop device is not supported.\n"); 2098 return -EINVAL; 2099 } 2100 2101 /* Hide this loop device for serialization. */ 2102 ret = mutex_lock_killable(&loop_ctl_mutex); 2103 if (ret) 2104 return ret; 2105 lo = idr_find(&loop_index_idr, idx); 2106 if (!lo || !lo->idr_visible) 2107 ret = -ENODEV; 2108 else 2109 lo->idr_visible = false; 2110 mutex_unlock(&loop_ctl_mutex); 2111 if (ret) 2112 return ret; 2113 2114 /* Check whether this loop device can be removed. */ 2115 ret = mutex_lock_killable(&lo->lo_mutex); 2116 if (ret) 2117 goto mark_visible; 2118 if (lo->lo_state != Lo_unbound || 2119 atomic_read(&lo->lo_refcnt) > 0) { 2120 mutex_unlock(&lo->lo_mutex); 2121 ret = -EBUSY; 2122 goto mark_visible; 2123 } 2124 /* Mark this loop device no longer open()-able. */ 2125 lo->lo_state = Lo_deleting; 2126 mutex_unlock(&lo->lo_mutex); 2127 2128 loop_remove(lo); 2129 return 0; 2130 2131 mark_visible: 2132 /* Show this loop device again. */ 2133 mutex_lock(&loop_ctl_mutex); 2134 lo->idr_visible = true; 2135 mutex_unlock(&loop_ctl_mutex); 2136 return ret; 2137 } 2138 2139 static int loop_control_get_free(int idx) 2140 { 2141 struct loop_device *lo; 2142 int id, ret; 2143 2144 ret = mutex_lock_killable(&loop_ctl_mutex); 2145 if (ret) 2146 return ret; 2147 idr_for_each_entry(&loop_index_idr, lo, id) { 2148 /* Hitting a race results in creating a new loop device which is harmless. */ 2149 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) 2150 goto found; 2151 } 2152 mutex_unlock(&loop_ctl_mutex); 2153 return loop_add(-1); 2154 found: 2155 mutex_unlock(&loop_ctl_mutex); 2156 return id; 2157 } 2158 2159 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2160 unsigned long parm) 2161 { 2162 switch (cmd) { 2163 case LOOP_CTL_ADD: 2164 return loop_add(parm); 2165 case LOOP_CTL_REMOVE: 2166 return loop_control_remove(parm); 2167 case LOOP_CTL_GET_FREE: 2168 return loop_control_get_free(parm); 2169 default: 2170 return -ENOSYS; 2171 } 2172 } 2173 2174 static const struct file_operations loop_ctl_fops = { 2175 .open = nonseekable_open, 2176 .unlocked_ioctl = loop_control_ioctl, 2177 .compat_ioctl = loop_control_ioctl, 2178 .owner = THIS_MODULE, 2179 .llseek = noop_llseek, 2180 }; 2181 2182 static struct miscdevice loop_misc = { 2183 .minor = LOOP_CTRL_MINOR, 2184 .name = "loop-control", 2185 .fops = &loop_ctl_fops, 2186 }; 2187 2188 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2189 MODULE_ALIAS("devname:loop-control"); 2190 2191 static int __init loop_init(void) 2192 { 2193 int i, nr; 2194 int err; 2195 2196 part_shift = 0; 2197 if (max_part > 0) { 2198 part_shift = fls(max_part); 2199 2200 /* 2201 * Adjust max_part according to part_shift as it is exported 2202 * to user space so that user can decide correct minor number 2203 * if [s]he want to create more devices. 2204 * 2205 * Note that -1 is required because partition 0 is reserved 2206 * for the whole disk. 2207 */ 2208 max_part = (1UL << part_shift) - 1; 2209 } 2210 2211 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2212 err = -EINVAL; 2213 goto err_out; 2214 } 2215 2216 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2217 err = -EINVAL; 2218 goto err_out; 2219 } 2220 2221 /* 2222 * If max_loop is specified, create that many devices upfront. 2223 * This also becomes a hard limit. If max_loop is not specified, 2224 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2225 * init time. Loop devices can be requested on-demand with the 2226 * /dev/loop-control interface, or be instantiated by accessing 2227 * a 'dead' device node. 2228 */ 2229 if (max_loop) 2230 nr = max_loop; 2231 else 2232 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2233 2234 err = misc_register(&loop_misc); 2235 if (err < 0) 2236 goto err_out; 2237 2238 2239 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2240 err = -EIO; 2241 goto misc_out; 2242 } 2243 2244 /* pre-create number of devices given by config or max_loop */ 2245 for (i = 0; i < nr; i++) 2246 loop_add(i); 2247 2248 printk(KERN_INFO "loop: module loaded\n"); 2249 return 0; 2250 2251 misc_out: 2252 misc_deregister(&loop_misc); 2253 err_out: 2254 return err; 2255 } 2256 2257 static void __exit loop_exit(void) 2258 { 2259 struct loop_device *lo; 2260 int id; 2261 2262 unregister_blkdev(LOOP_MAJOR, "loop"); 2263 misc_deregister(&loop_misc); 2264 2265 /* 2266 * There is no need to use loop_ctl_mutex here, for nobody else can 2267 * access loop_index_idr when this module is unloading (unless forced 2268 * module unloading is requested). If this is not a clean unloading, 2269 * we have no means to avoid kernel crash. 2270 */ 2271 idr_for_each_entry(&loop_index_idr, lo, id) 2272 loop_remove(lo); 2273 2274 idr_destroy(&loop_index_idr); 2275 } 2276 2277 module_init(loop_init); 2278 module_exit(loop_exit); 2279 2280 #ifndef MODULE 2281 static int __init max_loop_setup(char *str) 2282 { 2283 max_loop = simple_strtol(str, NULL, 0); 2284 return 1; 2285 } 2286 2287 __setup("max_loop=", max_loop_setup); 2288 #endif 2289