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 static void loop_free_idle_workers(struct timer_list *timer); 808 809 #ifdef CONFIG_BLK_CGROUP 810 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 811 { 812 return !css || css == blkcg_root_css; 813 } 814 #else 815 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 816 { 817 return !css; 818 } 819 #endif 820 821 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 822 { 823 struct rb_node **node, *parent = NULL; 824 struct loop_worker *cur_worker, *worker = NULL; 825 struct work_struct *work; 826 struct list_head *cmd_list; 827 828 spin_lock_irq(&lo->lo_work_lock); 829 830 if (queue_on_root_worker(cmd->blkcg_css)) 831 goto queue_work; 832 833 node = &lo->worker_tree.rb_node; 834 835 while (*node) { 836 parent = *node; 837 cur_worker = container_of(*node, struct loop_worker, rb_node); 838 if (cur_worker->blkcg_css == cmd->blkcg_css) { 839 worker = cur_worker; 840 break; 841 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { 842 node = &(*node)->rb_left; 843 } else { 844 node = &(*node)->rb_right; 845 } 846 } 847 if (worker) 848 goto queue_work; 849 850 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 851 /* 852 * In the event we cannot allocate a worker, just queue on the 853 * rootcg worker and issue the I/O as the rootcg 854 */ 855 if (!worker) { 856 cmd->blkcg_css = NULL; 857 if (cmd->memcg_css) 858 css_put(cmd->memcg_css); 859 cmd->memcg_css = NULL; 860 goto queue_work; 861 } 862 863 worker->blkcg_css = cmd->blkcg_css; 864 css_get(worker->blkcg_css); 865 INIT_WORK(&worker->work, loop_workfn); 866 INIT_LIST_HEAD(&worker->cmd_list); 867 INIT_LIST_HEAD(&worker->idle_list); 868 worker->lo = lo; 869 rb_link_node(&worker->rb_node, parent, node); 870 rb_insert_color(&worker->rb_node, &lo->worker_tree); 871 queue_work: 872 if (worker) { 873 /* 874 * We need to remove from the idle list here while 875 * holding the lock so that the idle timer doesn't 876 * free the worker 877 */ 878 if (!list_empty(&worker->idle_list)) 879 list_del_init(&worker->idle_list); 880 work = &worker->work; 881 cmd_list = &worker->cmd_list; 882 } else { 883 work = &lo->rootcg_work; 884 cmd_list = &lo->rootcg_cmd_list; 885 } 886 list_add_tail(&cmd->list_entry, cmd_list); 887 queue_work(lo->workqueue, work); 888 spin_unlock_irq(&lo->lo_work_lock); 889 } 890 891 static void loop_update_rotational(struct loop_device *lo) 892 { 893 struct file *file = lo->lo_backing_file; 894 struct inode *file_inode = file->f_mapping->host; 895 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 896 struct request_queue *q = lo->lo_queue; 897 bool nonrot = true; 898 899 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 900 if (file_bdev) 901 nonrot = bdev_nonrot(file_bdev); 902 903 if (nonrot) 904 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 905 else 906 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 907 } 908 909 /** 910 * loop_set_status_from_info - configure device from loop_info 911 * @lo: struct loop_device to configure 912 * @info: struct loop_info64 to configure the device with 913 * 914 * Configures the loop device parameters according to the passed 915 * in loop_info64 configuration. 916 */ 917 static int 918 loop_set_status_from_info(struct loop_device *lo, 919 const struct loop_info64 *info) 920 { 921 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 922 return -EINVAL; 923 924 switch (info->lo_encrypt_type) { 925 case LO_CRYPT_NONE: 926 break; 927 case LO_CRYPT_XOR: 928 pr_warn("support for the xor transformation has been removed.\n"); 929 return -EINVAL; 930 case LO_CRYPT_CRYPTOAPI: 931 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n"); 932 return -EINVAL; 933 default: 934 return -EINVAL; 935 } 936 937 lo->lo_offset = info->lo_offset; 938 lo->lo_sizelimit = info->lo_sizelimit; 939 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 940 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 941 lo->lo_flags = info->lo_flags; 942 return 0; 943 } 944 945 static int loop_configure(struct loop_device *lo, fmode_t mode, 946 struct block_device *bdev, 947 const struct loop_config *config) 948 { 949 struct file *file = fget(config->fd); 950 struct inode *inode; 951 struct address_space *mapping; 952 int error; 953 loff_t size; 954 bool partscan; 955 unsigned short bsize; 956 bool is_loop; 957 958 if (!file) 959 return -EBADF; 960 is_loop = is_loop_device(file); 961 962 /* This is safe, since we have a reference from open(). */ 963 __module_get(THIS_MODULE); 964 965 /* 966 * If we don't hold exclusive handle for the device, upgrade to it 967 * here to avoid changing device under exclusive owner. 968 */ 969 if (!(mode & FMODE_EXCL)) { 970 error = bd_prepare_to_claim(bdev, loop_configure); 971 if (error) 972 goto out_putf; 973 } 974 975 error = loop_global_lock_killable(lo, is_loop); 976 if (error) 977 goto out_bdev; 978 979 error = -EBUSY; 980 if (lo->lo_state != Lo_unbound) 981 goto out_unlock; 982 983 error = loop_validate_file(file, bdev); 984 if (error) 985 goto out_unlock; 986 987 mapping = file->f_mapping; 988 inode = mapping->host; 989 990 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 991 error = -EINVAL; 992 goto out_unlock; 993 } 994 995 if (config->block_size) { 996 error = blk_validate_block_size(config->block_size); 997 if (error) 998 goto out_unlock; 999 } 1000 1001 error = loop_set_status_from_info(lo, &config->info); 1002 if (error) 1003 goto out_unlock; 1004 1005 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1006 !file->f_op->write_iter) 1007 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1008 1009 lo->workqueue = alloc_workqueue("loop%d", 1010 WQ_UNBOUND | WQ_FREEZABLE, 1011 0, 1012 lo->lo_number); 1013 if (!lo->workqueue) { 1014 error = -ENOMEM; 1015 goto out_unlock; 1016 } 1017 1018 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1019 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1020 1021 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 1022 INIT_LIST_HEAD(&lo->rootcg_cmd_list); 1023 INIT_LIST_HEAD(&lo->idle_worker_list); 1024 lo->worker_tree = RB_ROOT; 1025 timer_setup(&lo->timer, loop_free_idle_workers, 1026 TIMER_DEFERRABLE); 1027 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1028 lo->lo_device = bdev; 1029 lo->lo_backing_file = file; 1030 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1031 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1032 1033 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1034 blk_queue_write_cache(lo->lo_queue, true, false); 1035 1036 if (config->block_size) 1037 bsize = config->block_size; 1038 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1039 /* In case of direct I/O, match underlying block size */ 1040 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1041 else 1042 bsize = 512; 1043 1044 blk_queue_logical_block_size(lo->lo_queue, bsize); 1045 blk_queue_physical_block_size(lo->lo_queue, bsize); 1046 blk_queue_io_min(lo->lo_queue, bsize); 1047 1048 loop_config_discard(lo); 1049 loop_update_rotational(lo); 1050 loop_update_dio(lo); 1051 loop_sysfs_init(lo); 1052 1053 size = get_loop_size(lo, file); 1054 loop_set_size(lo, size); 1055 1056 /* Order wrt reading lo_state in loop_validate_file(). */ 1057 wmb(); 1058 1059 lo->lo_state = Lo_bound; 1060 if (part_shift) 1061 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1062 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1063 if (partscan) 1064 lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 1065 1066 loop_global_unlock(lo, is_loop); 1067 if (partscan) 1068 loop_reread_partitions(lo); 1069 if (!(mode & FMODE_EXCL)) 1070 bd_abort_claiming(bdev, loop_configure); 1071 return 0; 1072 1073 out_unlock: 1074 loop_global_unlock(lo, is_loop); 1075 out_bdev: 1076 if (!(mode & FMODE_EXCL)) 1077 bd_abort_claiming(bdev, loop_configure); 1078 out_putf: 1079 fput(file); 1080 /* This is safe: open() is still holding a reference. */ 1081 module_put(THIS_MODULE); 1082 return error; 1083 } 1084 1085 static void __loop_clr_fd(struct loop_device *lo, bool release) 1086 { 1087 struct file *filp; 1088 gfp_t gfp = lo->old_gfp_mask; 1089 struct loop_worker *pos, *worker; 1090 1091 /* 1092 * Flush loop_configure() and loop_change_fd(). It is acceptable for 1093 * loop_validate_file() to succeed, for actual clear operation has not 1094 * started yet. 1095 */ 1096 mutex_lock(&loop_validate_mutex); 1097 mutex_unlock(&loop_validate_mutex); 1098 /* 1099 * loop_validate_file() now fails because l->lo_state != Lo_bound 1100 * became visible. 1101 */ 1102 1103 /* 1104 * Since this function is called upon "ioctl(LOOP_CLR_FD)" xor "close() 1105 * after ioctl(LOOP_CLR_FD)", it is a sign of something going wrong if 1106 * lo->lo_state has changed while waiting for lo->lo_mutex. 1107 */ 1108 mutex_lock(&lo->lo_mutex); 1109 BUG_ON(lo->lo_state != Lo_rundown); 1110 mutex_unlock(&lo->lo_mutex); 1111 1112 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1113 blk_queue_write_cache(lo->lo_queue, false, false); 1114 1115 /* freeze request queue during the transition */ 1116 blk_mq_freeze_queue(lo->lo_queue); 1117 1118 destroy_workqueue(lo->workqueue); 1119 spin_lock_irq(&lo->lo_work_lock); 1120 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 1121 idle_list) { 1122 list_del(&worker->idle_list); 1123 rb_erase(&worker->rb_node, &lo->worker_tree); 1124 css_put(worker->blkcg_css); 1125 kfree(worker); 1126 } 1127 spin_unlock_irq(&lo->lo_work_lock); 1128 del_timer_sync(&lo->timer); 1129 1130 spin_lock_irq(&lo->lo_lock); 1131 filp = lo->lo_backing_file; 1132 lo->lo_backing_file = NULL; 1133 spin_unlock_irq(&lo->lo_lock); 1134 1135 lo->lo_device = NULL; 1136 lo->lo_offset = 0; 1137 lo->lo_sizelimit = 0; 1138 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1139 blk_queue_logical_block_size(lo->lo_queue, 512); 1140 blk_queue_physical_block_size(lo->lo_queue, 512); 1141 blk_queue_io_min(lo->lo_queue, 512); 1142 invalidate_disk(lo->lo_disk); 1143 loop_sysfs_exit(lo); 1144 /* let user-space know about this change */ 1145 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 1146 mapping_set_gfp_mask(filp->f_mapping, gfp); 1147 /* This is safe: open() is still holding a reference. */ 1148 module_put(THIS_MODULE); 1149 blk_mq_unfreeze_queue(lo->lo_queue); 1150 1151 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1152 1153 if (lo->lo_flags & LO_FLAGS_PARTSCAN) { 1154 int err; 1155 1156 /* 1157 * open_mutex has been held already in release path, so don't 1158 * acquire it if this function is called in such case. 1159 * 1160 * If the reread partition isn't from release path, lo_refcnt 1161 * must be at least one and it can only become zero when the 1162 * current holder is released. 1163 */ 1164 if (!release) 1165 mutex_lock(&lo->lo_disk->open_mutex); 1166 err = bdev_disk_changed(lo->lo_disk, false); 1167 if (!release) 1168 mutex_unlock(&lo->lo_disk->open_mutex); 1169 if (err) 1170 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1171 __func__, lo->lo_number, err); 1172 /* Device is gone, no point in returning error */ 1173 } 1174 1175 /* 1176 * lo->lo_state is set to Lo_unbound here after above partscan has 1177 * finished. There cannot be anybody else entering __loop_clr_fd() as 1178 * Lo_rundown state protects us from all the other places trying to 1179 * change the 'lo' device. 1180 */ 1181 lo->lo_flags = 0; 1182 if (!part_shift) 1183 lo->lo_disk->flags |= GENHD_FL_NO_PART; 1184 mutex_lock(&lo->lo_mutex); 1185 lo->lo_state = Lo_unbound; 1186 mutex_unlock(&lo->lo_mutex); 1187 1188 /* 1189 * Need not hold lo_mutex to fput backing file. Calling fput holding 1190 * lo_mutex triggers a circular lock dependency possibility warning as 1191 * fput can take open_mutex which is usually taken before lo_mutex. 1192 */ 1193 fput(filp); 1194 } 1195 1196 static int loop_clr_fd(struct loop_device *lo) 1197 { 1198 int err; 1199 1200 err = mutex_lock_killable(&lo->lo_mutex); 1201 if (err) 1202 return err; 1203 if (lo->lo_state != Lo_bound) { 1204 mutex_unlock(&lo->lo_mutex); 1205 return -ENXIO; 1206 } 1207 /* 1208 * If we've explicitly asked to tear down the loop device, 1209 * and it has an elevated reference count, set it for auto-teardown when 1210 * the last reference goes away. This stops $!~#$@ udev from 1211 * preventing teardown because it decided that it needs to run blkid on 1212 * the loopback device whenever they appear. xfstests is notorious for 1213 * failing tests because blkid via udev races with a losetup 1214 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1215 * command to fail with EBUSY. 1216 */ 1217 if (atomic_read(&lo->lo_refcnt) > 1) { 1218 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1219 mutex_unlock(&lo->lo_mutex); 1220 return 0; 1221 } 1222 lo->lo_state = Lo_rundown; 1223 mutex_unlock(&lo->lo_mutex); 1224 1225 __loop_clr_fd(lo, false); 1226 return 0; 1227 } 1228 1229 static int 1230 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1231 { 1232 int err; 1233 int prev_lo_flags; 1234 bool partscan = false; 1235 bool size_changed = false; 1236 1237 err = mutex_lock_killable(&lo->lo_mutex); 1238 if (err) 1239 return err; 1240 if (lo->lo_state != Lo_bound) { 1241 err = -ENXIO; 1242 goto out_unlock; 1243 } 1244 1245 if (lo->lo_offset != info->lo_offset || 1246 lo->lo_sizelimit != info->lo_sizelimit) { 1247 size_changed = true; 1248 sync_blockdev(lo->lo_device); 1249 invalidate_bdev(lo->lo_device); 1250 } 1251 1252 /* I/O need to be drained during transfer transition */ 1253 blk_mq_freeze_queue(lo->lo_queue); 1254 1255 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1256 /* If any pages were dirtied after invalidate_bdev(), try again */ 1257 err = -EAGAIN; 1258 pr_warn("%s: loop%d (%s) still has dirty pages (nrpages=%lu)\n", 1259 __func__, lo->lo_number, lo->lo_file_name, 1260 lo->lo_device->bd_inode->i_mapping->nrpages); 1261 goto out_unfreeze; 1262 } 1263 1264 prev_lo_flags = lo->lo_flags; 1265 1266 err = loop_set_status_from_info(lo, info); 1267 if (err) 1268 goto out_unfreeze; 1269 1270 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1271 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1272 /* For those flags, use the previous values instead */ 1273 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1274 /* For flags that can't be cleared, use previous values too */ 1275 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1276 1277 if (size_changed) { 1278 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1279 lo->lo_backing_file); 1280 loop_set_size(lo, new_size); 1281 } 1282 1283 loop_config_discard(lo); 1284 1285 /* update dio if lo_offset or transfer is changed */ 1286 __loop_update_dio(lo, lo->use_dio); 1287 1288 out_unfreeze: 1289 blk_mq_unfreeze_queue(lo->lo_queue); 1290 1291 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1292 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1293 lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 1294 partscan = true; 1295 } 1296 out_unlock: 1297 mutex_unlock(&lo->lo_mutex); 1298 if (partscan) 1299 loop_reread_partitions(lo); 1300 1301 return err; 1302 } 1303 1304 static int 1305 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1306 { 1307 struct path path; 1308 struct kstat stat; 1309 int ret; 1310 1311 ret = mutex_lock_killable(&lo->lo_mutex); 1312 if (ret) 1313 return ret; 1314 if (lo->lo_state != Lo_bound) { 1315 mutex_unlock(&lo->lo_mutex); 1316 return -ENXIO; 1317 } 1318 1319 memset(info, 0, sizeof(*info)); 1320 info->lo_number = lo->lo_number; 1321 info->lo_offset = lo->lo_offset; 1322 info->lo_sizelimit = lo->lo_sizelimit; 1323 info->lo_flags = lo->lo_flags; 1324 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1325 1326 /* Drop lo_mutex while we call into the filesystem. */ 1327 path = lo->lo_backing_file->f_path; 1328 path_get(&path); 1329 mutex_unlock(&lo->lo_mutex); 1330 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1331 if (!ret) { 1332 info->lo_device = huge_encode_dev(stat.dev); 1333 info->lo_inode = stat.ino; 1334 info->lo_rdevice = huge_encode_dev(stat.rdev); 1335 } 1336 path_put(&path); 1337 return ret; 1338 } 1339 1340 static void 1341 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1342 { 1343 memset(info64, 0, sizeof(*info64)); 1344 info64->lo_number = info->lo_number; 1345 info64->lo_device = info->lo_device; 1346 info64->lo_inode = info->lo_inode; 1347 info64->lo_rdevice = info->lo_rdevice; 1348 info64->lo_offset = info->lo_offset; 1349 info64->lo_sizelimit = 0; 1350 info64->lo_flags = info->lo_flags; 1351 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1352 } 1353 1354 static int 1355 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1356 { 1357 memset(info, 0, sizeof(*info)); 1358 info->lo_number = info64->lo_number; 1359 info->lo_device = info64->lo_device; 1360 info->lo_inode = info64->lo_inode; 1361 info->lo_rdevice = info64->lo_rdevice; 1362 info->lo_offset = info64->lo_offset; 1363 info->lo_flags = info64->lo_flags; 1364 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1365 1366 /* error in case values were truncated */ 1367 if (info->lo_device != info64->lo_device || 1368 info->lo_rdevice != info64->lo_rdevice || 1369 info->lo_inode != info64->lo_inode || 1370 info->lo_offset != info64->lo_offset) 1371 return -EOVERFLOW; 1372 1373 return 0; 1374 } 1375 1376 static int 1377 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1378 { 1379 struct loop_info info; 1380 struct loop_info64 info64; 1381 1382 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1383 return -EFAULT; 1384 loop_info64_from_old(&info, &info64); 1385 return loop_set_status(lo, &info64); 1386 } 1387 1388 static int 1389 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1390 { 1391 struct loop_info64 info64; 1392 1393 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1394 return -EFAULT; 1395 return loop_set_status(lo, &info64); 1396 } 1397 1398 static int 1399 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1400 struct loop_info info; 1401 struct loop_info64 info64; 1402 int err; 1403 1404 if (!arg) 1405 return -EINVAL; 1406 err = loop_get_status(lo, &info64); 1407 if (!err) 1408 err = loop_info64_to_old(&info64, &info); 1409 if (!err && copy_to_user(arg, &info, sizeof(info))) 1410 err = -EFAULT; 1411 1412 return err; 1413 } 1414 1415 static int 1416 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 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 && copy_to_user(arg, &info64, sizeof(info64))) 1424 err = -EFAULT; 1425 1426 return err; 1427 } 1428 1429 static int loop_set_capacity(struct loop_device *lo) 1430 { 1431 loff_t size; 1432 1433 if (unlikely(lo->lo_state != Lo_bound)) 1434 return -ENXIO; 1435 1436 size = get_loop_size(lo, lo->lo_backing_file); 1437 loop_set_size(lo, size); 1438 1439 return 0; 1440 } 1441 1442 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1443 { 1444 int error = -ENXIO; 1445 if (lo->lo_state != Lo_bound) 1446 goto out; 1447 1448 __loop_update_dio(lo, !!arg); 1449 if (lo->use_dio == !!arg) 1450 return 0; 1451 error = -EINVAL; 1452 out: 1453 return error; 1454 } 1455 1456 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1457 { 1458 int err = 0; 1459 1460 if (lo->lo_state != Lo_bound) 1461 return -ENXIO; 1462 1463 err = blk_validate_block_size(arg); 1464 if (err) 1465 return err; 1466 1467 if (lo->lo_queue->limits.logical_block_size == arg) 1468 return 0; 1469 1470 sync_blockdev(lo->lo_device); 1471 invalidate_bdev(lo->lo_device); 1472 1473 blk_mq_freeze_queue(lo->lo_queue); 1474 1475 /* invalidate_bdev should have truncated all the pages */ 1476 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1477 err = -EAGAIN; 1478 pr_warn("%s: loop%d (%s) still has dirty pages (nrpages=%lu)\n", 1479 __func__, lo->lo_number, lo->lo_file_name, 1480 lo->lo_device->bd_inode->i_mapping->nrpages); 1481 goto out_unfreeze; 1482 } 1483 1484 blk_queue_logical_block_size(lo->lo_queue, arg); 1485 blk_queue_physical_block_size(lo->lo_queue, arg); 1486 blk_queue_io_min(lo->lo_queue, arg); 1487 loop_update_dio(lo); 1488 out_unfreeze: 1489 blk_mq_unfreeze_queue(lo->lo_queue); 1490 1491 return err; 1492 } 1493 1494 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1495 unsigned long arg) 1496 { 1497 int err; 1498 1499 err = mutex_lock_killable(&lo->lo_mutex); 1500 if (err) 1501 return err; 1502 switch (cmd) { 1503 case LOOP_SET_CAPACITY: 1504 err = loop_set_capacity(lo); 1505 break; 1506 case LOOP_SET_DIRECT_IO: 1507 err = loop_set_dio(lo, arg); 1508 break; 1509 case LOOP_SET_BLOCK_SIZE: 1510 err = loop_set_block_size(lo, arg); 1511 break; 1512 default: 1513 err = -EINVAL; 1514 } 1515 mutex_unlock(&lo->lo_mutex); 1516 return err; 1517 } 1518 1519 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1520 unsigned int cmd, unsigned long arg) 1521 { 1522 struct loop_device *lo = bdev->bd_disk->private_data; 1523 void __user *argp = (void __user *) arg; 1524 int err; 1525 1526 switch (cmd) { 1527 case LOOP_SET_FD: { 1528 /* 1529 * Legacy case - pass in a zeroed out struct loop_config with 1530 * only the file descriptor set , which corresponds with the 1531 * default parameters we'd have used otherwise. 1532 */ 1533 struct loop_config config; 1534 1535 memset(&config, 0, sizeof(config)); 1536 config.fd = arg; 1537 1538 return loop_configure(lo, mode, bdev, &config); 1539 } 1540 case LOOP_CONFIGURE: { 1541 struct loop_config config; 1542 1543 if (copy_from_user(&config, argp, sizeof(config))) 1544 return -EFAULT; 1545 1546 return loop_configure(lo, mode, bdev, &config); 1547 } 1548 case LOOP_CHANGE_FD: 1549 return loop_change_fd(lo, bdev, arg); 1550 case LOOP_CLR_FD: 1551 return loop_clr_fd(lo); 1552 case LOOP_SET_STATUS: 1553 err = -EPERM; 1554 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1555 err = loop_set_status_old(lo, argp); 1556 } 1557 break; 1558 case LOOP_GET_STATUS: 1559 return loop_get_status_old(lo, argp); 1560 case LOOP_SET_STATUS64: 1561 err = -EPERM; 1562 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1563 err = loop_set_status64(lo, argp); 1564 } 1565 break; 1566 case LOOP_GET_STATUS64: 1567 return loop_get_status64(lo, argp); 1568 case LOOP_SET_CAPACITY: 1569 case LOOP_SET_DIRECT_IO: 1570 case LOOP_SET_BLOCK_SIZE: 1571 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1572 return -EPERM; 1573 fallthrough; 1574 default: 1575 err = lo_simple_ioctl(lo, cmd, arg); 1576 break; 1577 } 1578 1579 return err; 1580 } 1581 1582 #ifdef CONFIG_COMPAT 1583 struct compat_loop_info { 1584 compat_int_t lo_number; /* ioctl r/o */ 1585 compat_dev_t lo_device; /* ioctl r/o */ 1586 compat_ulong_t lo_inode; /* ioctl r/o */ 1587 compat_dev_t lo_rdevice; /* ioctl r/o */ 1588 compat_int_t lo_offset; 1589 compat_int_t lo_encrypt_type; /* obsolete, ignored */ 1590 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1591 compat_int_t lo_flags; /* ioctl r/o */ 1592 char lo_name[LO_NAME_SIZE]; 1593 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1594 compat_ulong_t lo_init[2]; 1595 char reserved[4]; 1596 }; 1597 1598 /* 1599 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1600 * - noinlined to reduce stack space usage in main part of driver 1601 */ 1602 static noinline int 1603 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1604 struct loop_info64 *info64) 1605 { 1606 struct compat_loop_info info; 1607 1608 if (copy_from_user(&info, arg, sizeof(info))) 1609 return -EFAULT; 1610 1611 memset(info64, 0, sizeof(*info64)); 1612 info64->lo_number = info.lo_number; 1613 info64->lo_device = info.lo_device; 1614 info64->lo_inode = info.lo_inode; 1615 info64->lo_rdevice = info.lo_rdevice; 1616 info64->lo_offset = info.lo_offset; 1617 info64->lo_sizelimit = 0; 1618 info64->lo_flags = info.lo_flags; 1619 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1620 return 0; 1621 } 1622 1623 /* 1624 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1625 * - noinlined to reduce stack space usage in main part of driver 1626 */ 1627 static noinline int 1628 loop_info64_to_compat(const struct loop_info64 *info64, 1629 struct compat_loop_info __user *arg) 1630 { 1631 struct compat_loop_info info; 1632 1633 memset(&info, 0, sizeof(info)); 1634 info.lo_number = info64->lo_number; 1635 info.lo_device = info64->lo_device; 1636 info.lo_inode = info64->lo_inode; 1637 info.lo_rdevice = info64->lo_rdevice; 1638 info.lo_offset = info64->lo_offset; 1639 info.lo_flags = info64->lo_flags; 1640 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1641 1642 /* error in case values were truncated */ 1643 if (info.lo_device != info64->lo_device || 1644 info.lo_rdevice != info64->lo_rdevice || 1645 info.lo_inode != info64->lo_inode || 1646 info.lo_offset != info64->lo_offset) 1647 return -EOVERFLOW; 1648 1649 if (copy_to_user(arg, &info, sizeof(info))) 1650 return -EFAULT; 1651 return 0; 1652 } 1653 1654 static int 1655 loop_set_status_compat(struct loop_device *lo, 1656 const struct compat_loop_info __user *arg) 1657 { 1658 struct loop_info64 info64; 1659 int ret; 1660 1661 ret = loop_info64_from_compat(arg, &info64); 1662 if (ret < 0) 1663 return ret; 1664 return loop_set_status(lo, &info64); 1665 } 1666 1667 static int 1668 loop_get_status_compat(struct loop_device *lo, 1669 struct compat_loop_info __user *arg) 1670 { 1671 struct loop_info64 info64; 1672 int err; 1673 1674 if (!arg) 1675 return -EINVAL; 1676 err = loop_get_status(lo, &info64); 1677 if (!err) 1678 err = loop_info64_to_compat(&info64, arg); 1679 return err; 1680 } 1681 1682 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1683 unsigned int cmd, unsigned long arg) 1684 { 1685 struct loop_device *lo = bdev->bd_disk->private_data; 1686 int err; 1687 1688 switch(cmd) { 1689 case LOOP_SET_STATUS: 1690 err = loop_set_status_compat(lo, 1691 (const struct compat_loop_info __user *)arg); 1692 break; 1693 case LOOP_GET_STATUS: 1694 err = loop_get_status_compat(lo, 1695 (struct compat_loop_info __user *)arg); 1696 break; 1697 case LOOP_SET_CAPACITY: 1698 case LOOP_CLR_FD: 1699 case LOOP_GET_STATUS64: 1700 case LOOP_SET_STATUS64: 1701 case LOOP_CONFIGURE: 1702 arg = (unsigned long) compat_ptr(arg); 1703 fallthrough; 1704 case LOOP_SET_FD: 1705 case LOOP_CHANGE_FD: 1706 case LOOP_SET_BLOCK_SIZE: 1707 case LOOP_SET_DIRECT_IO: 1708 err = lo_ioctl(bdev, mode, cmd, arg); 1709 break; 1710 default: 1711 err = -ENOIOCTLCMD; 1712 break; 1713 } 1714 return err; 1715 } 1716 #endif 1717 1718 static int lo_open(struct block_device *bdev, fmode_t mode) 1719 { 1720 struct loop_device *lo = bdev->bd_disk->private_data; 1721 int err; 1722 1723 err = mutex_lock_killable(&lo->lo_mutex); 1724 if (err) 1725 return err; 1726 if (lo->lo_state == Lo_deleting) 1727 err = -ENXIO; 1728 else 1729 atomic_inc(&lo->lo_refcnt); 1730 mutex_unlock(&lo->lo_mutex); 1731 return err; 1732 } 1733 1734 static void lo_release(struct gendisk *disk, fmode_t mode) 1735 { 1736 struct loop_device *lo = disk->private_data; 1737 1738 mutex_lock(&lo->lo_mutex); 1739 if (atomic_dec_return(&lo->lo_refcnt)) 1740 goto out_unlock; 1741 1742 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1743 if (lo->lo_state != Lo_bound) 1744 goto out_unlock; 1745 lo->lo_state = Lo_rundown; 1746 mutex_unlock(&lo->lo_mutex); 1747 /* 1748 * In autoclear mode, stop the loop thread 1749 * and remove configuration after last close. 1750 */ 1751 __loop_clr_fd(lo, true); 1752 return; 1753 } else if (lo->lo_state == Lo_bound) { 1754 /* 1755 * Otherwise keep thread (if running) and config, 1756 * but flush possible ongoing bios in thread. 1757 */ 1758 blk_mq_freeze_queue(lo->lo_queue); 1759 blk_mq_unfreeze_queue(lo->lo_queue); 1760 } 1761 1762 out_unlock: 1763 mutex_unlock(&lo->lo_mutex); 1764 } 1765 1766 static const struct block_device_operations lo_fops = { 1767 .owner = THIS_MODULE, 1768 .open = lo_open, 1769 .release = lo_release, 1770 .ioctl = lo_ioctl, 1771 #ifdef CONFIG_COMPAT 1772 .compat_ioctl = lo_compat_ioctl, 1773 #endif 1774 }; 1775 1776 /* 1777 * And now the modules code and kernel interface. 1778 */ 1779 static int max_loop; 1780 module_param(max_loop, int, 0444); 1781 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1782 module_param(max_part, int, 0444); 1783 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1784 1785 static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH; 1786 1787 static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p) 1788 { 1789 int ret = kstrtoint(s, 10, &hw_queue_depth); 1790 1791 return (ret || (hw_queue_depth < 1)) ? -EINVAL : 0; 1792 } 1793 1794 static const struct kernel_param_ops loop_hw_qdepth_param_ops = { 1795 .set = loop_set_hw_queue_depth, 1796 .get = param_get_int, 1797 }; 1798 1799 device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444); 1800 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 128"); 1801 1802 MODULE_LICENSE("GPL"); 1803 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1804 1805 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1806 const struct blk_mq_queue_data *bd) 1807 { 1808 struct request *rq = bd->rq; 1809 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1810 struct loop_device *lo = rq->q->queuedata; 1811 1812 blk_mq_start_request(rq); 1813 1814 if (lo->lo_state != Lo_bound) 1815 return BLK_STS_IOERR; 1816 1817 switch (req_op(rq)) { 1818 case REQ_OP_FLUSH: 1819 case REQ_OP_DISCARD: 1820 case REQ_OP_WRITE_ZEROES: 1821 cmd->use_aio = false; 1822 break; 1823 default: 1824 cmd->use_aio = lo->use_dio; 1825 break; 1826 } 1827 1828 /* always use the first bio's css */ 1829 cmd->blkcg_css = NULL; 1830 cmd->memcg_css = NULL; 1831 #ifdef CONFIG_BLK_CGROUP 1832 if (rq->bio && rq->bio->bi_blkg) { 1833 cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 1834 #ifdef CONFIG_MEMCG 1835 cmd->memcg_css = 1836 cgroup_get_e_css(cmd->blkcg_css->cgroup, 1837 &memory_cgrp_subsys); 1838 #endif 1839 } 1840 #endif 1841 loop_queue_work(lo, cmd); 1842 1843 return BLK_STS_OK; 1844 } 1845 1846 static void loop_handle_cmd(struct loop_cmd *cmd) 1847 { 1848 struct request *rq = blk_mq_rq_from_pdu(cmd); 1849 const bool write = op_is_write(req_op(rq)); 1850 struct loop_device *lo = rq->q->queuedata; 1851 int ret = 0; 1852 struct mem_cgroup *old_memcg = NULL; 1853 1854 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1855 ret = -EIO; 1856 goto failed; 1857 } 1858 1859 if (cmd->blkcg_css) 1860 kthread_associate_blkcg(cmd->blkcg_css); 1861 if (cmd->memcg_css) 1862 old_memcg = set_active_memcg( 1863 mem_cgroup_from_css(cmd->memcg_css)); 1864 1865 ret = do_req_filebacked(lo, rq); 1866 1867 if (cmd->blkcg_css) 1868 kthread_associate_blkcg(NULL); 1869 1870 if (cmd->memcg_css) { 1871 set_active_memcg(old_memcg); 1872 css_put(cmd->memcg_css); 1873 } 1874 failed: 1875 /* complete non-aio request */ 1876 if (!cmd->use_aio || ret) { 1877 if (ret == -EOPNOTSUPP) 1878 cmd->ret = ret; 1879 else 1880 cmd->ret = ret ? -EIO : 0; 1881 if (likely(!blk_should_fake_timeout(rq->q))) 1882 blk_mq_complete_request(rq); 1883 } 1884 } 1885 1886 static void loop_set_timer(struct loop_device *lo) 1887 { 1888 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 1889 } 1890 1891 static void loop_process_work(struct loop_worker *worker, 1892 struct list_head *cmd_list, struct loop_device *lo) 1893 { 1894 int orig_flags = current->flags; 1895 struct loop_cmd *cmd; 1896 1897 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 1898 spin_lock_irq(&lo->lo_work_lock); 1899 while (!list_empty(cmd_list)) { 1900 cmd = container_of( 1901 cmd_list->next, struct loop_cmd, list_entry); 1902 list_del(cmd_list->next); 1903 spin_unlock_irq(&lo->lo_work_lock); 1904 1905 loop_handle_cmd(cmd); 1906 cond_resched(); 1907 1908 spin_lock_irq(&lo->lo_work_lock); 1909 } 1910 1911 /* 1912 * We only add to the idle list if there are no pending cmds 1913 * *and* the worker will not run again which ensures that it 1914 * is safe to free any worker on the idle list 1915 */ 1916 if (worker && !work_pending(&worker->work)) { 1917 worker->last_ran_at = jiffies; 1918 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 1919 loop_set_timer(lo); 1920 } 1921 spin_unlock_irq(&lo->lo_work_lock); 1922 current->flags = orig_flags; 1923 } 1924 1925 static void loop_workfn(struct work_struct *work) 1926 { 1927 struct loop_worker *worker = 1928 container_of(work, struct loop_worker, work); 1929 loop_process_work(worker, &worker->cmd_list, worker->lo); 1930 } 1931 1932 static void loop_rootcg_workfn(struct work_struct *work) 1933 { 1934 struct loop_device *lo = 1935 container_of(work, struct loop_device, rootcg_work); 1936 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 1937 } 1938 1939 static void loop_free_idle_workers(struct timer_list *timer) 1940 { 1941 struct loop_device *lo = container_of(timer, struct loop_device, timer); 1942 struct loop_worker *pos, *worker; 1943 1944 spin_lock_irq(&lo->lo_work_lock); 1945 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 1946 idle_list) { 1947 if (time_is_after_jiffies(worker->last_ran_at + 1948 LOOP_IDLE_WORKER_TIMEOUT)) 1949 break; 1950 list_del(&worker->idle_list); 1951 rb_erase(&worker->rb_node, &lo->worker_tree); 1952 css_put(worker->blkcg_css); 1953 kfree(worker); 1954 } 1955 if (!list_empty(&lo->idle_worker_list)) 1956 loop_set_timer(lo); 1957 spin_unlock_irq(&lo->lo_work_lock); 1958 } 1959 1960 static const struct blk_mq_ops loop_mq_ops = { 1961 .queue_rq = loop_queue_rq, 1962 .complete = lo_complete_rq, 1963 }; 1964 1965 static int loop_add(int i) 1966 { 1967 struct loop_device *lo; 1968 struct gendisk *disk; 1969 int err; 1970 1971 err = -ENOMEM; 1972 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1973 if (!lo) 1974 goto out; 1975 lo->lo_state = Lo_unbound; 1976 1977 err = mutex_lock_killable(&loop_ctl_mutex); 1978 if (err) 1979 goto out_free_dev; 1980 1981 /* allocate id, if @id >= 0, we're requesting that specific id */ 1982 if (i >= 0) { 1983 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1984 if (err == -ENOSPC) 1985 err = -EEXIST; 1986 } else { 1987 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1988 } 1989 mutex_unlock(&loop_ctl_mutex); 1990 if (err < 0) 1991 goto out_free_dev; 1992 i = err; 1993 1994 lo->tag_set.ops = &loop_mq_ops; 1995 lo->tag_set.nr_hw_queues = 1; 1996 lo->tag_set.queue_depth = hw_queue_depth; 1997 lo->tag_set.numa_node = NUMA_NO_NODE; 1998 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1999 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | 2000 BLK_MQ_F_NO_SCHED_BY_DEFAULT; 2001 lo->tag_set.driver_data = lo; 2002 2003 err = blk_mq_alloc_tag_set(&lo->tag_set); 2004 if (err) 2005 goto out_free_idr; 2006 2007 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 2008 if (IS_ERR(disk)) { 2009 err = PTR_ERR(disk); 2010 goto out_cleanup_tags; 2011 } 2012 lo->lo_queue = lo->lo_disk->queue; 2013 2014 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2015 2016 /* 2017 * By default, we do buffer IO, so it doesn't make sense to enable 2018 * merge because the I/O submitted to backing file is handled page by 2019 * page. For directio mode, merge does help to dispatch bigger request 2020 * to underlayer disk. We will enable merge once directio is enabled. 2021 */ 2022 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2023 2024 /* 2025 * Disable partition scanning by default. The in-kernel partition 2026 * scanning can be requested individually per-device during its 2027 * setup. Userspace can always add and remove partitions from all 2028 * devices. The needed partition minors are allocated from the 2029 * extended minor space, the main loop device numbers will continue 2030 * to match the loop minors, regardless of the number of partitions 2031 * used. 2032 * 2033 * If max_part is given, partition scanning is globally enabled for 2034 * all loop devices. The minors for the main loop devices will be 2035 * multiples of max_part. 2036 * 2037 * Note: Global-for-all-devices, set-only-at-init, read-only module 2038 * parameteters like 'max_loop' and 'max_part' make things needlessly 2039 * complicated, are too static, inflexible and may surprise 2040 * userspace tools. Parameters like this in general should be avoided. 2041 */ 2042 if (!part_shift) 2043 disk->flags |= GENHD_FL_NO_PART; 2044 atomic_set(&lo->lo_refcnt, 0); 2045 mutex_init(&lo->lo_mutex); 2046 lo->lo_number = i; 2047 spin_lock_init(&lo->lo_lock); 2048 spin_lock_init(&lo->lo_work_lock); 2049 disk->major = LOOP_MAJOR; 2050 disk->first_minor = i << part_shift; 2051 disk->minors = 1 << part_shift; 2052 disk->fops = &lo_fops; 2053 disk->private_data = lo; 2054 disk->queue = lo->lo_queue; 2055 disk->events = DISK_EVENT_MEDIA_CHANGE; 2056 disk->event_flags = DISK_EVENT_FLAG_UEVENT; 2057 sprintf(disk->disk_name, "loop%d", i); 2058 /* Make this loop device reachable from pathname. */ 2059 err = add_disk(disk); 2060 if (err) 2061 goto out_cleanup_disk; 2062 2063 /* Show this loop device. */ 2064 mutex_lock(&loop_ctl_mutex); 2065 lo->idr_visible = true; 2066 mutex_unlock(&loop_ctl_mutex); 2067 2068 return i; 2069 2070 out_cleanup_disk: 2071 blk_cleanup_disk(disk); 2072 out_cleanup_tags: 2073 blk_mq_free_tag_set(&lo->tag_set); 2074 out_free_idr: 2075 mutex_lock(&loop_ctl_mutex); 2076 idr_remove(&loop_index_idr, i); 2077 mutex_unlock(&loop_ctl_mutex); 2078 out_free_dev: 2079 kfree(lo); 2080 out: 2081 return err; 2082 } 2083 2084 static void loop_remove(struct loop_device *lo) 2085 { 2086 /* Make this loop device unreachable from pathname. */ 2087 del_gendisk(lo->lo_disk); 2088 blk_cleanup_disk(lo->lo_disk); 2089 blk_mq_free_tag_set(&lo->tag_set); 2090 2091 mutex_lock(&loop_ctl_mutex); 2092 idr_remove(&loop_index_idr, lo->lo_number); 2093 mutex_unlock(&loop_ctl_mutex); 2094 /* There is no route which can find this loop device. */ 2095 mutex_destroy(&lo->lo_mutex); 2096 kfree(lo); 2097 } 2098 2099 static void loop_probe(dev_t dev) 2100 { 2101 int idx = MINOR(dev) >> part_shift; 2102 2103 if (max_loop && idx >= max_loop) 2104 return; 2105 loop_add(idx); 2106 } 2107 2108 static int loop_control_remove(int idx) 2109 { 2110 struct loop_device *lo; 2111 int ret; 2112 2113 if (idx < 0) { 2114 pr_warn_once("deleting an unspecified loop device is not supported.\n"); 2115 return -EINVAL; 2116 } 2117 2118 /* Hide this loop device for serialization. */ 2119 ret = mutex_lock_killable(&loop_ctl_mutex); 2120 if (ret) 2121 return ret; 2122 lo = idr_find(&loop_index_idr, idx); 2123 if (!lo || !lo->idr_visible) 2124 ret = -ENODEV; 2125 else 2126 lo->idr_visible = false; 2127 mutex_unlock(&loop_ctl_mutex); 2128 if (ret) 2129 return ret; 2130 2131 /* Check whether this loop device can be removed. */ 2132 ret = mutex_lock_killable(&lo->lo_mutex); 2133 if (ret) 2134 goto mark_visible; 2135 if (lo->lo_state != Lo_unbound || 2136 atomic_read(&lo->lo_refcnt) > 0) { 2137 mutex_unlock(&lo->lo_mutex); 2138 ret = -EBUSY; 2139 goto mark_visible; 2140 } 2141 /* Mark this loop device no longer open()-able. */ 2142 lo->lo_state = Lo_deleting; 2143 mutex_unlock(&lo->lo_mutex); 2144 2145 loop_remove(lo); 2146 return 0; 2147 2148 mark_visible: 2149 /* Show this loop device again. */ 2150 mutex_lock(&loop_ctl_mutex); 2151 lo->idr_visible = true; 2152 mutex_unlock(&loop_ctl_mutex); 2153 return ret; 2154 } 2155 2156 static int loop_control_get_free(int idx) 2157 { 2158 struct loop_device *lo; 2159 int id, ret; 2160 2161 ret = mutex_lock_killable(&loop_ctl_mutex); 2162 if (ret) 2163 return ret; 2164 idr_for_each_entry(&loop_index_idr, lo, id) { 2165 /* Hitting a race results in creating a new loop device which is harmless. */ 2166 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) 2167 goto found; 2168 } 2169 mutex_unlock(&loop_ctl_mutex); 2170 return loop_add(-1); 2171 found: 2172 mutex_unlock(&loop_ctl_mutex); 2173 return id; 2174 } 2175 2176 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2177 unsigned long parm) 2178 { 2179 switch (cmd) { 2180 case LOOP_CTL_ADD: 2181 return loop_add(parm); 2182 case LOOP_CTL_REMOVE: 2183 return loop_control_remove(parm); 2184 case LOOP_CTL_GET_FREE: 2185 return loop_control_get_free(parm); 2186 default: 2187 return -ENOSYS; 2188 } 2189 } 2190 2191 static const struct file_operations loop_ctl_fops = { 2192 .open = nonseekable_open, 2193 .unlocked_ioctl = loop_control_ioctl, 2194 .compat_ioctl = loop_control_ioctl, 2195 .owner = THIS_MODULE, 2196 .llseek = noop_llseek, 2197 }; 2198 2199 static struct miscdevice loop_misc = { 2200 .minor = LOOP_CTRL_MINOR, 2201 .name = "loop-control", 2202 .fops = &loop_ctl_fops, 2203 }; 2204 2205 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2206 MODULE_ALIAS("devname:loop-control"); 2207 2208 static int __init loop_init(void) 2209 { 2210 int i, nr; 2211 int err; 2212 2213 part_shift = 0; 2214 if (max_part > 0) { 2215 part_shift = fls(max_part); 2216 2217 /* 2218 * Adjust max_part according to part_shift as it is exported 2219 * to user space so that user can decide correct minor number 2220 * if [s]he want to create more devices. 2221 * 2222 * Note that -1 is required because partition 0 is reserved 2223 * for the whole disk. 2224 */ 2225 max_part = (1UL << part_shift) - 1; 2226 } 2227 2228 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2229 err = -EINVAL; 2230 goto err_out; 2231 } 2232 2233 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2234 err = -EINVAL; 2235 goto err_out; 2236 } 2237 2238 /* 2239 * If max_loop is specified, create that many devices upfront. 2240 * This also becomes a hard limit. If max_loop is not specified, 2241 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2242 * init time. Loop devices can be requested on-demand with the 2243 * /dev/loop-control interface, or be instantiated by accessing 2244 * a 'dead' device node. 2245 */ 2246 if (max_loop) 2247 nr = max_loop; 2248 else 2249 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2250 2251 err = misc_register(&loop_misc); 2252 if (err < 0) 2253 goto err_out; 2254 2255 2256 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2257 err = -EIO; 2258 goto misc_out; 2259 } 2260 2261 /* pre-create number of devices given by config or max_loop */ 2262 for (i = 0; i < nr; i++) 2263 loop_add(i); 2264 2265 printk(KERN_INFO "loop: module loaded\n"); 2266 return 0; 2267 2268 misc_out: 2269 misc_deregister(&loop_misc); 2270 err_out: 2271 return err; 2272 } 2273 2274 static void __exit loop_exit(void) 2275 { 2276 struct loop_device *lo; 2277 int id; 2278 2279 unregister_blkdev(LOOP_MAJOR, "loop"); 2280 misc_deregister(&loop_misc); 2281 2282 /* 2283 * There is no need to use loop_ctl_mutex here, for nobody else can 2284 * access loop_index_idr when this module is unloading (unless forced 2285 * module unloading is requested). If this is not a clean unloading, 2286 * we have no means to avoid kernel crash. 2287 */ 2288 idr_for_each_entry(&loop_index_idr, lo, id) 2289 loop_remove(lo); 2290 2291 idr_destroy(&loop_index_idr); 2292 } 2293 2294 module_init(loop_init); 2295 module_exit(loop_exit); 2296 2297 #ifndef MODULE 2298 static int __init max_loop_setup(char *str) 2299 { 2300 max_loop = simple_strtol(str, NULL, 0); 2301 return 1; 2302 } 2303 2304 __setup("max_loop=", max_loop_setup); 2305 #endif 2306