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 /* freeze request queue during the transition */ 1143 blk_mq_freeze_queue(lo->lo_queue); 1144 1145 destroy_workqueue(lo->workqueue); 1146 loop_free_idle_workers(lo, true); 1147 del_timer_sync(&lo->timer); 1148 1149 spin_lock_irq(&lo->lo_lock); 1150 filp = lo->lo_backing_file; 1151 lo->lo_backing_file = NULL; 1152 spin_unlock_irq(&lo->lo_lock); 1153 1154 lo->lo_device = NULL; 1155 lo->lo_offset = 0; 1156 lo->lo_sizelimit = 0; 1157 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1158 blk_queue_logical_block_size(lo->lo_queue, 512); 1159 blk_queue_physical_block_size(lo->lo_queue, 512); 1160 blk_queue_io_min(lo->lo_queue, 512); 1161 invalidate_disk(lo->lo_disk); 1162 loop_sysfs_exit(lo); 1163 /* let user-space know about this change */ 1164 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 1165 mapping_set_gfp_mask(filp->f_mapping, gfp); 1166 /* This is safe: open() is still holding a reference. */ 1167 module_put(THIS_MODULE); 1168 blk_mq_unfreeze_queue(lo->lo_queue); 1169 1170 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1171 1172 if (lo->lo_flags & LO_FLAGS_PARTSCAN) { 1173 int err; 1174 1175 /* 1176 * open_mutex has been held already in release path, so don't 1177 * acquire it if this function is called in such case. 1178 * 1179 * If the reread partition isn't from release path, lo_refcnt 1180 * must be at least one and it can only become zero when the 1181 * current holder is released. 1182 */ 1183 if (!release) 1184 mutex_lock(&lo->lo_disk->open_mutex); 1185 err = bdev_disk_changed(lo->lo_disk, false); 1186 if (!release) 1187 mutex_unlock(&lo->lo_disk->open_mutex); 1188 if (err) 1189 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1190 __func__, lo->lo_number, err); 1191 /* Device is gone, no point in returning error */ 1192 } 1193 1194 /* 1195 * lo->lo_state is set to Lo_unbound here after above partscan has 1196 * finished. There cannot be anybody else entering __loop_clr_fd() as 1197 * Lo_rundown state protects us from all the other places trying to 1198 * change the 'lo' device. 1199 */ 1200 lo->lo_flags = 0; 1201 if (!part_shift) 1202 lo->lo_disk->flags |= GENHD_FL_NO_PART; 1203 mutex_lock(&lo->lo_mutex); 1204 lo->lo_state = Lo_unbound; 1205 mutex_unlock(&lo->lo_mutex); 1206 1207 /* 1208 * Need not hold lo_mutex to fput backing file. Calling fput holding 1209 * lo_mutex triggers a circular lock dependency possibility warning as 1210 * fput can take open_mutex which is usually taken before lo_mutex. 1211 */ 1212 fput(filp); 1213 } 1214 1215 static int loop_clr_fd(struct loop_device *lo) 1216 { 1217 int err; 1218 1219 err = mutex_lock_killable(&lo->lo_mutex); 1220 if (err) 1221 return err; 1222 if (lo->lo_state != Lo_bound) { 1223 mutex_unlock(&lo->lo_mutex); 1224 return -ENXIO; 1225 } 1226 /* 1227 * If we've explicitly asked to tear down the loop device, 1228 * and it has an elevated reference count, set it for auto-teardown when 1229 * the last reference goes away. This stops $!~#$@ udev from 1230 * preventing teardown because it decided that it needs to run blkid on 1231 * the loopback device whenever they appear. xfstests is notorious for 1232 * failing tests because blkid via udev races with a losetup 1233 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1234 * command to fail with EBUSY. 1235 */ 1236 if (atomic_read(&lo->lo_refcnt) > 1) { 1237 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1238 mutex_unlock(&lo->lo_mutex); 1239 return 0; 1240 } 1241 lo->lo_state = Lo_rundown; 1242 mutex_unlock(&lo->lo_mutex); 1243 1244 __loop_clr_fd(lo, false); 1245 return 0; 1246 } 1247 1248 static int 1249 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1250 { 1251 int err; 1252 int prev_lo_flags; 1253 bool partscan = false; 1254 bool size_changed = false; 1255 1256 err = mutex_lock_killable(&lo->lo_mutex); 1257 if (err) 1258 return err; 1259 if (lo->lo_state != Lo_bound) { 1260 err = -ENXIO; 1261 goto out_unlock; 1262 } 1263 1264 if (lo->lo_offset != info->lo_offset || 1265 lo->lo_sizelimit != info->lo_sizelimit) { 1266 size_changed = true; 1267 sync_blockdev(lo->lo_device); 1268 invalidate_bdev(lo->lo_device); 1269 } 1270 1271 /* I/O need to be drained during transfer transition */ 1272 blk_mq_freeze_queue(lo->lo_queue); 1273 1274 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1275 /* If any pages were dirtied after invalidate_bdev(), try again */ 1276 err = -EAGAIN; 1277 pr_warn("%s: loop%d (%s) still has dirty pages (nrpages=%lu)\n", 1278 __func__, lo->lo_number, lo->lo_file_name, 1279 lo->lo_device->bd_inode->i_mapping->nrpages); 1280 goto out_unfreeze; 1281 } 1282 1283 prev_lo_flags = lo->lo_flags; 1284 1285 err = loop_set_status_from_info(lo, info); 1286 if (err) 1287 goto out_unfreeze; 1288 1289 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1290 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1291 /* For those flags, use the previous values instead */ 1292 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1293 /* For flags that can't be cleared, use previous values too */ 1294 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1295 1296 if (size_changed) { 1297 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1298 lo->lo_backing_file); 1299 loop_set_size(lo, new_size); 1300 } 1301 1302 loop_config_discard(lo); 1303 1304 /* update dio if lo_offset or transfer is changed */ 1305 __loop_update_dio(lo, lo->use_dio); 1306 1307 out_unfreeze: 1308 blk_mq_unfreeze_queue(lo->lo_queue); 1309 1310 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1311 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1312 lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 1313 partscan = true; 1314 } 1315 out_unlock: 1316 mutex_unlock(&lo->lo_mutex); 1317 if (partscan) 1318 loop_reread_partitions(lo); 1319 1320 return err; 1321 } 1322 1323 static int 1324 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1325 { 1326 struct path path; 1327 struct kstat stat; 1328 int ret; 1329 1330 ret = mutex_lock_killable(&lo->lo_mutex); 1331 if (ret) 1332 return ret; 1333 if (lo->lo_state != Lo_bound) { 1334 mutex_unlock(&lo->lo_mutex); 1335 return -ENXIO; 1336 } 1337 1338 memset(info, 0, sizeof(*info)); 1339 info->lo_number = lo->lo_number; 1340 info->lo_offset = lo->lo_offset; 1341 info->lo_sizelimit = lo->lo_sizelimit; 1342 info->lo_flags = lo->lo_flags; 1343 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1344 1345 /* Drop lo_mutex while we call into the filesystem. */ 1346 path = lo->lo_backing_file->f_path; 1347 path_get(&path); 1348 mutex_unlock(&lo->lo_mutex); 1349 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1350 if (!ret) { 1351 info->lo_device = huge_encode_dev(stat.dev); 1352 info->lo_inode = stat.ino; 1353 info->lo_rdevice = huge_encode_dev(stat.rdev); 1354 } 1355 path_put(&path); 1356 return ret; 1357 } 1358 1359 static void 1360 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1361 { 1362 memset(info64, 0, sizeof(*info64)); 1363 info64->lo_number = info->lo_number; 1364 info64->lo_device = info->lo_device; 1365 info64->lo_inode = info->lo_inode; 1366 info64->lo_rdevice = info->lo_rdevice; 1367 info64->lo_offset = info->lo_offset; 1368 info64->lo_sizelimit = 0; 1369 info64->lo_flags = info->lo_flags; 1370 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1371 } 1372 1373 static int 1374 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1375 { 1376 memset(info, 0, sizeof(*info)); 1377 info->lo_number = info64->lo_number; 1378 info->lo_device = info64->lo_device; 1379 info->lo_inode = info64->lo_inode; 1380 info->lo_rdevice = info64->lo_rdevice; 1381 info->lo_offset = info64->lo_offset; 1382 info->lo_flags = info64->lo_flags; 1383 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1384 1385 /* error in case values were truncated */ 1386 if (info->lo_device != info64->lo_device || 1387 info->lo_rdevice != info64->lo_rdevice || 1388 info->lo_inode != info64->lo_inode || 1389 info->lo_offset != info64->lo_offset) 1390 return -EOVERFLOW; 1391 1392 return 0; 1393 } 1394 1395 static int 1396 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1397 { 1398 struct loop_info info; 1399 struct loop_info64 info64; 1400 1401 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1402 return -EFAULT; 1403 loop_info64_from_old(&info, &info64); 1404 return loop_set_status(lo, &info64); 1405 } 1406 1407 static int 1408 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1409 { 1410 struct loop_info64 info64; 1411 1412 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1413 return -EFAULT; 1414 return loop_set_status(lo, &info64); 1415 } 1416 1417 static int 1418 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1419 struct loop_info info; 1420 struct loop_info64 info64; 1421 int err; 1422 1423 if (!arg) 1424 return -EINVAL; 1425 err = loop_get_status(lo, &info64); 1426 if (!err) 1427 err = loop_info64_to_old(&info64, &info); 1428 if (!err && copy_to_user(arg, &info, sizeof(info))) 1429 err = -EFAULT; 1430 1431 return err; 1432 } 1433 1434 static int 1435 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1436 struct loop_info64 info64; 1437 int err; 1438 1439 if (!arg) 1440 return -EINVAL; 1441 err = loop_get_status(lo, &info64); 1442 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1443 err = -EFAULT; 1444 1445 return err; 1446 } 1447 1448 static int loop_set_capacity(struct loop_device *lo) 1449 { 1450 loff_t size; 1451 1452 if (unlikely(lo->lo_state != Lo_bound)) 1453 return -ENXIO; 1454 1455 size = get_loop_size(lo, lo->lo_backing_file); 1456 loop_set_size(lo, size); 1457 1458 return 0; 1459 } 1460 1461 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1462 { 1463 int error = -ENXIO; 1464 if (lo->lo_state != Lo_bound) 1465 goto out; 1466 1467 __loop_update_dio(lo, !!arg); 1468 if (lo->use_dio == !!arg) 1469 return 0; 1470 error = -EINVAL; 1471 out: 1472 return error; 1473 } 1474 1475 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1476 { 1477 int err = 0; 1478 1479 if (lo->lo_state != Lo_bound) 1480 return -ENXIO; 1481 1482 err = blk_validate_block_size(arg); 1483 if (err) 1484 return err; 1485 1486 if (lo->lo_queue->limits.logical_block_size == arg) 1487 return 0; 1488 1489 sync_blockdev(lo->lo_device); 1490 invalidate_bdev(lo->lo_device); 1491 1492 blk_mq_freeze_queue(lo->lo_queue); 1493 1494 /* invalidate_bdev should have truncated all the pages */ 1495 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1496 err = -EAGAIN; 1497 pr_warn("%s: loop%d (%s) still has dirty pages (nrpages=%lu)\n", 1498 __func__, lo->lo_number, lo->lo_file_name, 1499 lo->lo_device->bd_inode->i_mapping->nrpages); 1500 goto out_unfreeze; 1501 } 1502 1503 blk_queue_logical_block_size(lo->lo_queue, arg); 1504 blk_queue_physical_block_size(lo->lo_queue, arg); 1505 blk_queue_io_min(lo->lo_queue, arg); 1506 loop_update_dio(lo); 1507 out_unfreeze: 1508 blk_mq_unfreeze_queue(lo->lo_queue); 1509 1510 return err; 1511 } 1512 1513 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1514 unsigned long arg) 1515 { 1516 int err; 1517 1518 err = mutex_lock_killable(&lo->lo_mutex); 1519 if (err) 1520 return err; 1521 switch (cmd) { 1522 case LOOP_SET_CAPACITY: 1523 err = loop_set_capacity(lo); 1524 break; 1525 case LOOP_SET_DIRECT_IO: 1526 err = loop_set_dio(lo, arg); 1527 break; 1528 case LOOP_SET_BLOCK_SIZE: 1529 err = loop_set_block_size(lo, arg); 1530 break; 1531 default: 1532 err = -EINVAL; 1533 } 1534 mutex_unlock(&lo->lo_mutex); 1535 return err; 1536 } 1537 1538 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1539 unsigned int cmd, unsigned long arg) 1540 { 1541 struct loop_device *lo = bdev->bd_disk->private_data; 1542 void __user *argp = (void __user *) arg; 1543 int err; 1544 1545 switch (cmd) { 1546 case LOOP_SET_FD: { 1547 /* 1548 * Legacy case - pass in a zeroed out struct loop_config with 1549 * only the file descriptor set , which corresponds with the 1550 * default parameters we'd have used otherwise. 1551 */ 1552 struct loop_config config; 1553 1554 memset(&config, 0, sizeof(config)); 1555 config.fd = arg; 1556 1557 return loop_configure(lo, mode, bdev, &config); 1558 } 1559 case LOOP_CONFIGURE: { 1560 struct loop_config config; 1561 1562 if (copy_from_user(&config, argp, sizeof(config))) 1563 return -EFAULT; 1564 1565 return loop_configure(lo, mode, bdev, &config); 1566 } 1567 case LOOP_CHANGE_FD: 1568 return loop_change_fd(lo, bdev, arg); 1569 case LOOP_CLR_FD: 1570 return loop_clr_fd(lo); 1571 case LOOP_SET_STATUS: 1572 err = -EPERM; 1573 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1574 err = loop_set_status_old(lo, argp); 1575 } 1576 break; 1577 case LOOP_GET_STATUS: 1578 return loop_get_status_old(lo, argp); 1579 case LOOP_SET_STATUS64: 1580 err = -EPERM; 1581 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1582 err = loop_set_status64(lo, argp); 1583 } 1584 break; 1585 case LOOP_GET_STATUS64: 1586 return loop_get_status64(lo, argp); 1587 case LOOP_SET_CAPACITY: 1588 case LOOP_SET_DIRECT_IO: 1589 case LOOP_SET_BLOCK_SIZE: 1590 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1591 return -EPERM; 1592 fallthrough; 1593 default: 1594 err = lo_simple_ioctl(lo, cmd, arg); 1595 break; 1596 } 1597 1598 return err; 1599 } 1600 1601 #ifdef CONFIG_COMPAT 1602 struct compat_loop_info { 1603 compat_int_t lo_number; /* ioctl r/o */ 1604 compat_dev_t lo_device; /* ioctl r/o */ 1605 compat_ulong_t lo_inode; /* ioctl r/o */ 1606 compat_dev_t lo_rdevice; /* ioctl r/o */ 1607 compat_int_t lo_offset; 1608 compat_int_t lo_encrypt_type; /* obsolete, ignored */ 1609 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1610 compat_int_t lo_flags; /* ioctl r/o */ 1611 char lo_name[LO_NAME_SIZE]; 1612 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1613 compat_ulong_t lo_init[2]; 1614 char reserved[4]; 1615 }; 1616 1617 /* 1618 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1619 * - noinlined to reduce stack space usage in main part of driver 1620 */ 1621 static noinline int 1622 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1623 struct loop_info64 *info64) 1624 { 1625 struct compat_loop_info info; 1626 1627 if (copy_from_user(&info, arg, sizeof(info))) 1628 return -EFAULT; 1629 1630 memset(info64, 0, sizeof(*info64)); 1631 info64->lo_number = info.lo_number; 1632 info64->lo_device = info.lo_device; 1633 info64->lo_inode = info.lo_inode; 1634 info64->lo_rdevice = info.lo_rdevice; 1635 info64->lo_offset = info.lo_offset; 1636 info64->lo_sizelimit = 0; 1637 info64->lo_flags = info.lo_flags; 1638 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1639 return 0; 1640 } 1641 1642 /* 1643 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1644 * - noinlined to reduce stack space usage in main part of driver 1645 */ 1646 static noinline int 1647 loop_info64_to_compat(const struct loop_info64 *info64, 1648 struct compat_loop_info __user *arg) 1649 { 1650 struct compat_loop_info info; 1651 1652 memset(&info, 0, sizeof(info)); 1653 info.lo_number = info64->lo_number; 1654 info.lo_device = info64->lo_device; 1655 info.lo_inode = info64->lo_inode; 1656 info.lo_rdevice = info64->lo_rdevice; 1657 info.lo_offset = info64->lo_offset; 1658 info.lo_flags = info64->lo_flags; 1659 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1660 1661 /* error in case values were truncated */ 1662 if (info.lo_device != info64->lo_device || 1663 info.lo_rdevice != info64->lo_rdevice || 1664 info.lo_inode != info64->lo_inode || 1665 info.lo_offset != info64->lo_offset) 1666 return -EOVERFLOW; 1667 1668 if (copy_to_user(arg, &info, sizeof(info))) 1669 return -EFAULT; 1670 return 0; 1671 } 1672 1673 static int 1674 loop_set_status_compat(struct loop_device *lo, 1675 const struct compat_loop_info __user *arg) 1676 { 1677 struct loop_info64 info64; 1678 int ret; 1679 1680 ret = loop_info64_from_compat(arg, &info64); 1681 if (ret < 0) 1682 return ret; 1683 return loop_set_status(lo, &info64); 1684 } 1685 1686 static int 1687 loop_get_status_compat(struct loop_device *lo, 1688 struct compat_loop_info __user *arg) 1689 { 1690 struct loop_info64 info64; 1691 int err; 1692 1693 if (!arg) 1694 return -EINVAL; 1695 err = loop_get_status(lo, &info64); 1696 if (!err) 1697 err = loop_info64_to_compat(&info64, arg); 1698 return err; 1699 } 1700 1701 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1702 unsigned int cmd, unsigned long arg) 1703 { 1704 struct loop_device *lo = bdev->bd_disk->private_data; 1705 int err; 1706 1707 switch(cmd) { 1708 case LOOP_SET_STATUS: 1709 err = loop_set_status_compat(lo, 1710 (const struct compat_loop_info __user *)arg); 1711 break; 1712 case LOOP_GET_STATUS: 1713 err = loop_get_status_compat(lo, 1714 (struct compat_loop_info __user *)arg); 1715 break; 1716 case LOOP_SET_CAPACITY: 1717 case LOOP_CLR_FD: 1718 case LOOP_GET_STATUS64: 1719 case LOOP_SET_STATUS64: 1720 case LOOP_CONFIGURE: 1721 arg = (unsigned long) compat_ptr(arg); 1722 fallthrough; 1723 case LOOP_SET_FD: 1724 case LOOP_CHANGE_FD: 1725 case LOOP_SET_BLOCK_SIZE: 1726 case LOOP_SET_DIRECT_IO: 1727 err = lo_ioctl(bdev, mode, cmd, arg); 1728 break; 1729 default: 1730 err = -ENOIOCTLCMD; 1731 break; 1732 } 1733 return err; 1734 } 1735 #endif 1736 1737 static int lo_open(struct block_device *bdev, fmode_t mode) 1738 { 1739 struct loop_device *lo = bdev->bd_disk->private_data; 1740 int err; 1741 1742 err = mutex_lock_killable(&lo->lo_mutex); 1743 if (err) 1744 return err; 1745 if (lo->lo_state == Lo_deleting) 1746 err = -ENXIO; 1747 else 1748 atomic_inc(&lo->lo_refcnt); 1749 mutex_unlock(&lo->lo_mutex); 1750 return err; 1751 } 1752 1753 static void lo_release(struct gendisk *disk, fmode_t mode) 1754 { 1755 struct loop_device *lo = disk->private_data; 1756 1757 mutex_lock(&lo->lo_mutex); 1758 if (atomic_dec_return(&lo->lo_refcnt)) 1759 goto out_unlock; 1760 1761 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1762 if (lo->lo_state != Lo_bound) 1763 goto out_unlock; 1764 lo->lo_state = Lo_rundown; 1765 mutex_unlock(&lo->lo_mutex); 1766 /* 1767 * In autoclear mode, stop the loop thread 1768 * and remove configuration after last close. 1769 */ 1770 __loop_clr_fd(lo, true); 1771 return; 1772 } else if (lo->lo_state == Lo_bound) { 1773 /* 1774 * Otherwise keep thread (if running) and config, 1775 * but flush possible ongoing bios in thread. 1776 */ 1777 blk_mq_freeze_queue(lo->lo_queue); 1778 blk_mq_unfreeze_queue(lo->lo_queue); 1779 } 1780 1781 out_unlock: 1782 mutex_unlock(&lo->lo_mutex); 1783 } 1784 1785 static const struct block_device_operations lo_fops = { 1786 .owner = THIS_MODULE, 1787 .open = lo_open, 1788 .release = lo_release, 1789 .ioctl = lo_ioctl, 1790 #ifdef CONFIG_COMPAT 1791 .compat_ioctl = lo_compat_ioctl, 1792 #endif 1793 }; 1794 1795 /* 1796 * And now the modules code and kernel interface. 1797 */ 1798 static int max_loop; 1799 module_param(max_loop, int, 0444); 1800 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1801 module_param(max_part, int, 0444); 1802 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1803 1804 static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH; 1805 1806 static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p) 1807 { 1808 int ret = kstrtoint(s, 10, &hw_queue_depth); 1809 1810 return (ret || (hw_queue_depth < 1)) ? -EINVAL : 0; 1811 } 1812 1813 static const struct kernel_param_ops loop_hw_qdepth_param_ops = { 1814 .set = loop_set_hw_queue_depth, 1815 .get = param_get_int, 1816 }; 1817 1818 device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444); 1819 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 128"); 1820 1821 MODULE_LICENSE("GPL"); 1822 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1823 1824 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1825 const struct blk_mq_queue_data *bd) 1826 { 1827 struct request *rq = bd->rq; 1828 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1829 struct loop_device *lo = rq->q->queuedata; 1830 1831 blk_mq_start_request(rq); 1832 1833 if (lo->lo_state != Lo_bound) 1834 return BLK_STS_IOERR; 1835 1836 switch (req_op(rq)) { 1837 case REQ_OP_FLUSH: 1838 case REQ_OP_DISCARD: 1839 case REQ_OP_WRITE_ZEROES: 1840 cmd->use_aio = false; 1841 break; 1842 default: 1843 cmd->use_aio = lo->use_dio; 1844 break; 1845 } 1846 1847 /* always use the first bio's css */ 1848 cmd->blkcg_css = NULL; 1849 cmd->memcg_css = NULL; 1850 #ifdef CONFIG_BLK_CGROUP 1851 if (rq->bio && rq->bio->bi_blkg) { 1852 cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 1853 #ifdef CONFIG_MEMCG 1854 cmd->memcg_css = 1855 cgroup_get_e_css(cmd->blkcg_css->cgroup, 1856 &memory_cgrp_subsys); 1857 #endif 1858 } 1859 #endif 1860 loop_queue_work(lo, cmd); 1861 1862 return BLK_STS_OK; 1863 } 1864 1865 static void loop_handle_cmd(struct loop_cmd *cmd) 1866 { 1867 struct request *rq = blk_mq_rq_from_pdu(cmd); 1868 const bool write = op_is_write(req_op(rq)); 1869 struct loop_device *lo = rq->q->queuedata; 1870 int ret = 0; 1871 struct mem_cgroup *old_memcg = NULL; 1872 1873 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1874 ret = -EIO; 1875 goto failed; 1876 } 1877 1878 if (cmd->blkcg_css) 1879 kthread_associate_blkcg(cmd->blkcg_css); 1880 if (cmd->memcg_css) 1881 old_memcg = set_active_memcg( 1882 mem_cgroup_from_css(cmd->memcg_css)); 1883 1884 ret = do_req_filebacked(lo, rq); 1885 1886 if (cmd->blkcg_css) 1887 kthread_associate_blkcg(NULL); 1888 1889 if (cmd->memcg_css) { 1890 set_active_memcg(old_memcg); 1891 css_put(cmd->memcg_css); 1892 } 1893 failed: 1894 /* complete non-aio request */ 1895 if (!cmd->use_aio || ret) { 1896 if (ret == -EOPNOTSUPP) 1897 cmd->ret = ret; 1898 else 1899 cmd->ret = ret ? -EIO : 0; 1900 if (likely(!blk_should_fake_timeout(rq->q))) 1901 blk_mq_complete_request(rq); 1902 } 1903 } 1904 1905 static void loop_process_work(struct loop_worker *worker, 1906 struct list_head *cmd_list, struct loop_device *lo) 1907 { 1908 int orig_flags = current->flags; 1909 struct loop_cmd *cmd; 1910 1911 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 1912 spin_lock_irq(&lo->lo_work_lock); 1913 while (!list_empty(cmd_list)) { 1914 cmd = container_of( 1915 cmd_list->next, struct loop_cmd, list_entry); 1916 list_del(cmd_list->next); 1917 spin_unlock_irq(&lo->lo_work_lock); 1918 1919 loop_handle_cmd(cmd); 1920 cond_resched(); 1921 1922 spin_lock_irq(&lo->lo_work_lock); 1923 } 1924 1925 /* 1926 * We only add to the idle list if there are no pending cmds 1927 * *and* the worker will not run again which ensures that it 1928 * is safe to free any worker on the idle list 1929 */ 1930 if (worker && !work_pending(&worker->work)) { 1931 worker->last_ran_at = jiffies; 1932 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 1933 loop_set_timer(lo); 1934 } 1935 spin_unlock_irq(&lo->lo_work_lock); 1936 current->flags = orig_flags; 1937 } 1938 1939 static void loop_workfn(struct work_struct *work) 1940 { 1941 struct loop_worker *worker = 1942 container_of(work, struct loop_worker, work); 1943 loop_process_work(worker, &worker->cmd_list, worker->lo); 1944 } 1945 1946 static void loop_rootcg_workfn(struct work_struct *work) 1947 { 1948 struct loop_device *lo = 1949 container_of(work, struct loop_device, rootcg_work); 1950 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 1951 } 1952 1953 static const struct blk_mq_ops loop_mq_ops = { 1954 .queue_rq = loop_queue_rq, 1955 .complete = lo_complete_rq, 1956 }; 1957 1958 static int loop_add(int i) 1959 { 1960 struct loop_device *lo; 1961 struct gendisk *disk; 1962 int err; 1963 1964 err = -ENOMEM; 1965 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1966 if (!lo) 1967 goto out; 1968 lo->worker_tree = RB_ROOT; 1969 INIT_LIST_HEAD(&lo->idle_worker_list); 1970 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE); 1971 lo->lo_state = Lo_unbound; 1972 1973 err = mutex_lock_killable(&loop_ctl_mutex); 1974 if (err) 1975 goto out_free_dev; 1976 1977 /* allocate id, if @id >= 0, we're requesting that specific id */ 1978 if (i >= 0) { 1979 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1980 if (err == -ENOSPC) 1981 err = -EEXIST; 1982 } else { 1983 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1984 } 1985 mutex_unlock(&loop_ctl_mutex); 1986 if (err < 0) 1987 goto out_free_dev; 1988 i = err; 1989 1990 lo->tag_set.ops = &loop_mq_ops; 1991 lo->tag_set.nr_hw_queues = 1; 1992 lo->tag_set.queue_depth = hw_queue_depth; 1993 lo->tag_set.numa_node = NUMA_NO_NODE; 1994 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1995 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | 1996 BLK_MQ_F_NO_SCHED_BY_DEFAULT; 1997 lo->tag_set.driver_data = lo; 1998 1999 err = blk_mq_alloc_tag_set(&lo->tag_set); 2000 if (err) 2001 goto out_free_idr; 2002 2003 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 2004 if (IS_ERR(disk)) { 2005 err = PTR_ERR(disk); 2006 goto out_cleanup_tags; 2007 } 2008 lo->lo_queue = lo->lo_disk->queue; 2009 2010 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2011 2012 /* 2013 * By default, we do buffer IO, so it doesn't make sense to enable 2014 * merge because the I/O submitted to backing file is handled page by 2015 * page. For directio mode, merge does help to dispatch bigger request 2016 * to underlayer disk. We will enable merge once directio is enabled. 2017 */ 2018 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2019 2020 /* 2021 * Disable partition scanning by default. The in-kernel partition 2022 * scanning can be requested individually per-device during its 2023 * setup. Userspace can always add and remove partitions from all 2024 * devices. The needed partition minors are allocated from the 2025 * extended minor space, the main loop device numbers will continue 2026 * to match the loop minors, regardless of the number of partitions 2027 * used. 2028 * 2029 * If max_part is given, partition scanning is globally enabled for 2030 * all loop devices. The minors for the main loop devices will be 2031 * multiples of max_part. 2032 * 2033 * Note: Global-for-all-devices, set-only-at-init, read-only module 2034 * parameteters like 'max_loop' and 'max_part' make things needlessly 2035 * complicated, are too static, inflexible and may surprise 2036 * userspace tools. Parameters like this in general should be avoided. 2037 */ 2038 if (!part_shift) 2039 disk->flags |= GENHD_FL_NO_PART; 2040 atomic_set(&lo->lo_refcnt, 0); 2041 mutex_init(&lo->lo_mutex); 2042 lo->lo_number = i; 2043 spin_lock_init(&lo->lo_lock); 2044 spin_lock_init(&lo->lo_work_lock); 2045 disk->major = LOOP_MAJOR; 2046 disk->first_minor = i << part_shift; 2047 disk->minors = 1 << part_shift; 2048 disk->fops = &lo_fops; 2049 disk->private_data = lo; 2050 disk->queue = lo->lo_queue; 2051 disk->events = DISK_EVENT_MEDIA_CHANGE; 2052 disk->event_flags = DISK_EVENT_FLAG_UEVENT; 2053 sprintf(disk->disk_name, "loop%d", i); 2054 /* Make this loop device reachable from pathname. */ 2055 err = add_disk(disk); 2056 if (err) 2057 goto out_cleanup_disk; 2058 2059 /* Show this loop device. */ 2060 mutex_lock(&loop_ctl_mutex); 2061 lo->idr_visible = true; 2062 mutex_unlock(&loop_ctl_mutex); 2063 2064 return i; 2065 2066 out_cleanup_disk: 2067 blk_cleanup_disk(disk); 2068 out_cleanup_tags: 2069 blk_mq_free_tag_set(&lo->tag_set); 2070 out_free_idr: 2071 mutex_lock(&loop_ctl_mutex); 2072 idr_remove(&loop_index_idr, i); 2073 mutex_unlock(&loop_ctl_mutex); 2074 out_free_dev: 2075 kfree(lo); 2076 out: 2077 return err; 2078 } 2079 2080 static void loop_remove(struct loop_device *lo) 2081 { 2082 /* Make this loop device unreachable from pathname. */ 2083 del_gendisk(lo->lo_disk); 2084 blk_cleanup_disk(lo->lo_disk); 2085 blk_mq_free_tag_set(&lo->tag_set); 2086 2087 mutex_lock(&loop_ctl_mutex); 2088 idr_remove(&loop_index_idr, lo->lo_number); 2089 mutex_unlock(&loop_ctl_mutex); 2090 /* There is no route which can find this loop device. */ 2091 mutex_destroy(&lo->lo_mutex); 2092 kfree(lo); 2093 } 2094 2095 static void loop_probe(dev_t dev) 2096 { 2097 int idx = MINOR(dev) >> part_shift; 2098 2099 if (max_loop && idx >= max_loop) 2100 return; 2101 loop_add(idx); 2102 } 2103 2104 static int loop_control_remove(int idx) 2105 { 2106 struct loop_device *lo; 2107 int ret; 2108 2109 if (idx < 0) { 2110 pr_warn_once("deleting an unspecified loop device is not supported.\n"); 2111 return -EINVAL; 2112 } 2113 2114 /* Hide this loop device for serialization. */ 2115 ret = mutex_lock_killable(&loop_ctl_mutex); 2116 if (ret) 2117 return ret; 2118 lo = idr_find(&loop_index_idr, idx); 2119 if (!lo || !lo->idr_visible) 2120 ret = -ENODEV; 2121 else 2122 lo->idr_visible = false; 2123 mutex_unlock(&loop_ctl_mutex); 2124 if (ret) 2125 return ret; 2126 2127 /* Check whether this loop device can be removed. */ 2128 ret = mutex_lock_killable(&lo->lo_mutex); 2129 if (ret) 2130 goto mark_visible; 2131 if (lo->lo_state != Lo_unbound || 2132 atomic_read(&lo->lo_refcnt) > 0) { 2133 mutex_unlock(&lo->lo_mutex); 2134 ret = -EBUSY; 2135 goto mark_visible; 2136 } 2137 /* Mark this loop device no longer open()-able. */ 2138 lo->lo_state = Lo_deleting; 2139 mutex_unlock(&lo->lo_mutex); 2140 2141 loop_remove(lo); 2142 return 0; 2143 2144 mark_visible: 2145 /* Show this loop device again. */ 2146 mutex_lock(&loop_ctl_mutex); 2147 lo->idr_visible = true; 2148 mutex_unlock(&loop_ctl_mutex); 2149 return ret; 2150 } 2151 2152 static int loop_control_get_free(int idx) 2153 { 2154 struct loop_device *lo; 2155 int id, ret; 2156 2157 ret = mutex_lock_killable(&loop_ctl_mutex); 2158 if (ret) 2159 return ret; 2160 idr_for_each_entry(&loop_index_idr, lo, id) { 2161 /* Hitting a race results in creating a new loop device which is harmless. */ 2162 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) 2163 goto found; 2164 } 2165 mutex_unlock(&loop_ctl_mutex); 2166 return loop_add(-1); 2167 found: 2168 mutex_unlock(&loop_ctl_mutex); 2169 return id; 2170 } 2171 2172 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2173 unsigned long parm) 2174 { 2175 switch (cmd) { 2176 case LOOP_CTL_ADD: 2177 return loop_add(parm); 2178 case LOOP_CTL_REMOVE: 2179 return loop_control_remove(parm); 2180 case LOOP_CTL_GET_FREE: 2181 return loop_control_get_free(parm); 2182 default: 2183 return -ENOSYS; 2184 } 2185 } 2186 2187 static const struct file_operations loop_ctl_fops = { 2188 .open = nonseekable_open, 2189 .unlocked_ioctl = loop_control_ioctl, 2190 .compat_ioctl = loop_control_ioctl, 2191 .owner = THIS_MODULE, 2192 .llseek = noop_llseek, 2193 }; 2194 2195 static struct miscdevice loop_misc = { 2196 .minor = LOOP_CTRL_MINOR, 2197 .name = "loop-control", 2198 .fops = &loop_ctl_fops, 2199 }; 2200 2201 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2202 MODULE_ALIAS("devname:loop-control"); 2203 2204 static int __init loop_init(void) 2205 { 2206 int i, nr; 2207 int err; 2208 2209 part_shift = 0; 2210 if (max_part > 0) { 2211 part_shift = fls(max_part); 2212 2213 /* 2214 * Adjust max_part according to part_shift as it is exported 2215 * to user space so that user can decide correct minor number 2216 * if [s]he want to create more devices. 2217 * 2218 * Note that -1 is required because partition 0 is reserved 2219 * for the whole disk. 2220 */ 2221 max_part = (1UL << part_shift) - 1; 2222 } 2223 2224 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2225 err = -EINVAL; 2226 goto err_out; 2227 } 2228 2229 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2230 err = -EINVAL; 2231 goto err_out; 2232 } 2233 2234 /* 2235 * If max_loop is specified, create that many devices upfront. 2236 * This also becomes a hard limit. If max_loop is not specified, 2237 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2238 * init time. Loop devices can be requested on-demand with the 2239 * /dev/loop-control interface, or be instantiated by accessing 2240 * a 'dead' device node. 2241 */ 2242 if (max_loop) 2243 nr = max_loop; 2244 else 2245 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2246 2247 err = misc_register(&loop_misc); 2248 if (err < 0) 2249 goto err_out; 2250 2251 2252 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2253 err = -EIO; 2254 goto misc_out; 2255 } 2256 2257 /* pre-create number of devices given by config or max_loop */ 2258 for (i = 0; i < nr; i++) 2259 loop_add(i); 2260 2261 printk(KERN_INFO "loop: module loaded\n"); 2262 return 0; 2263 2264 misc_out: 2265 misc_deregister(&loop_misc); 2266 err_out: 2267 return err; 2268 } 2269 2270 static void __exit loop_exit(void) 2271 { 2272 struct loop_device *lo; 2273 int id; 2274 2275 unregister_blkdev(LOOP_MAJOR, "loop"); 2276 misc_deregister(&loop_misc); 2277 2278 /* 2279 * There is no need to use loop_ctl_mutex here, for nobody else can 2280 * access loop_index_idr when this module is unloading (unless forced 2281 * module unloading is requested). If this is not a clean unloading, 2282 * we have no means to avoid kernel crash. 2283 */ 2284 idr_for_each_entry(&loop_index_idr, lo, id) 2285 loop_remove(lo); 2286 2287 idr_destroy(&loop_index_idr); 2288 } 2289 2290 module_init(loop_init); 2291 module_exit(loop_exit); 2292 2293 #ifndef MODULE 2294 static int __init max_loop_setup(char *str) 2295 { 2296 max_loop = simple_strtol(str, NULL, 0); 2297 return 1; 2298 } 2299 2300 __setup("max_loop=", max_loop_setup); 2301 #endif 2302