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 83 #include "loop.h" 84 85 #include <linux/uaccess.h> 86 87 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) 88 89 static DEFINE_IDR(loop_index_idr); 90 static DEFINE_MUTEX(loop_ctl_mutex); 91 static DEFINE_MUTEX(loop_validate_mutex); 92 93 /** 94 * loop_global_lock_killable() - take locks for safe loop_validate_file() test 95 * 96 * @lo: struct loop_device 97 * @global: true if @lo is about to bind another "struct loop_device", false otherwise 98 * 99 * Returns 0 on success, -EINTR otherwise. 100 * 101 * Since loop_validate_file() traverses on other "struct loop_device" if 102 * is_loop_device() is true, we need a global lock for serializing concurrent 103 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls. 104 */ 105 static int loop_global_lock_killable(struct loop_device *lo, bool global) 106 { 107 int err; 108 109 if (global) { 110 err = mutex_lock_killable(&loop_validate_mutex); 111 if (err) 112 return err; 113 } 114 err = mutex_lock_killable(&lo->lo_mutex); 115 if (err && global) 116 mutex_unlock(&loop_validate_mutex); 117 return err; 118 } 119 120 /** 121 * loop_global_unlock() - release locks taken by loop_global_lock_killable() 122 * 123 * @lo: struct loop_device 124 * @global: true if @lo was about to bind another "struct loop_device", false otherwise 125 */ 126 static void loop_global_unlock(struct loop_device *lo, bool global) 127 { 128 mutex_unlock(&lo->lo_mutex); 129 if (global) 130 mutex_unlock(&loop_validate_mutex); 131 } 132 133 static int max_part; 134 static int part_shift; 135 136 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 137 { 138 loff_t loopsize; 139 140 /* Compute loopsize in bytes */ 141 loopsize = i_size_read(file->f_mapping->host); 142 if (offset > 0) 143 loopsize -= offset; 144 /* offset is beyond i_size, weird but possible */ 145 if (loopsize < 0) 146 return 0; 147 148 if (sizelimit > 0 && sizelimit < loopsize) 149 loopsize = sizelimit; 150 /* 151 * Unfortunately, if we want to do I/O on the device, 152 * the number of 512-byte sectors has to fit into a sector_t. 153 */ 154 return loopsize >> 9; 155 } 156 157 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 158 { 159 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 160 } 161 162 static void __loop_update_dio(struct loop_device *lo, bool dio) 163 { 164 struct file *file = lo->lo_backing_file; 165 struct address_space *mapping = file->f_mapping; 166 struct inode *inode = mapping->host; 167 unsigned short sb_bsize = 0; 168 unsigned dio_align = 0; 169 bool use_dio; 170 171 if (inode->i_sb->s_bdev) { 172 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 173 dio_align = sb_bsize - 1; 174 } 175 176 /* 177 * We support direct I/O only if lo_offset is aligned with the 178 * logical I/O size of backing device, and the logical block 179 * size of loop is bigger than the backing device's. 180 * 181 * TODO: the above condition may be loosed in the future, and 182 * direct I/O may be switched runtime at that time because most 183 * of requests in sane applications should be PAGE_SIZE aligned 184 */ 185 if (dio) { 186 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 187 !(lo->lo_offset & dio_align) && 188 mapping->a_ops->direct_IO) 189 use_dio = true; 190 else 191 use_dio = false; 192 } else { 193 use_dio = false; 194 } 195 196 if (lo->use_dio == use_dio) 197 return; 198 199 /* flush dirty pages before changing direct IO */ 200 vfs_fsync(file, 0); 201 202 /* 203 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 204 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 205 * will get updated by ioctl(LOOP_GET_STATUS) 206 */ 207 if (lo->lo_state == Lo_bound) 208 blk_mq_freeze_queue(lo->lo_queue); 209 lo->use_dio = use_dio; 210 if (use_dio) { 211 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 212 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 213 } else { 214 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 215 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 216 } 217 if (lo->lo_state == Lo_bound) 218 blk_mq_unfreeze_queue(lo->lo_queue); 219 } 220 221 /** 222 * loop_validate_block_size() - validates the passed in block size 223 * @bsize: size to validate 224 */ 225 static int 226 loop_validate_block_size(unsigned short bsize) 227 { 228 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize)) 229 return -EINVAL; 230 231 return 0; 232 } 233 234 /** 235 * loop_set_size() - sets device size and notifies userspace 236 * @lo: struct loop_device to set the size for 237 * @size: new size of the loop device 238 * 239 * Callers must validate that the size passed into this function fits into 240 * a sector_t, eg using loop_validate_size() 241 */ 242 static void loop_set_size(struct loop_device *lo, loff_t size) 243 { 244 if (!set_capacity_and_notify(lo->lo_disk, size)) 245 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 246 } 247 248 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 249 { 250 struct iov_iter i; 251 ssize_t bw; 252 253 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 254 255 file_start_write(file); 256 bw = vfs_iter_write(file, &i, ppos, 0); 257 file_end_write(file); 258 259 if (likely(bw == bvec->bv_len)) 260 return 0; 261 262 printk_ratelimited(KERN_ERR 263 "loop: Write error at byte offset %llu, length %i.\n", 264 (unsigned long long)*ppos, bvec->bv_len); 265 if (bw >= 0) 266 bw = -EIO; 267 return bw; 268 } 269 270 static int lo_write_simple(struct loop_device *lo, struct request *rq, 271 loff_t pos) 272 { 273 struct bio_vec bvec; 274 struct req_iterator iter; 275 int ret = 0; 276 277 rq_for_each_segment(bvec, rq, iter) { 278 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 279 if (ret < 0) 280 break; 281 cond_resched(); 282 } 283 284 return ret; 285 } 286 287 static int lo_read_simple(struct loop_device *lo, struct request *rq, 288 loff_t pos) 289 { 290 struct bio_vec bvec; 291 struct req_iterator iter; 292 struct iov_iter i; 293 ssize_t len; 294 295 rq_for_each_segment(bvec, rq, iter) { 296 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 297 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 298 if (len < 0) 299 return len; 300 301 flush_dcache_page(bvec.bv_page); 302 303 if (len != bvec.bv_len) { 304 struct bio *bio; 305 306 __rq_for_each_bio(bio, rq) 307 zero_fill_bio(bio); 308 break; 309 } 310 cond_resched(); 311 } 312 313 return 0; 314 } 315 316 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 317 int mode) 318 { 319 /* 320 * We use fallocate to manipulate the space mappings used by the image 321 * a.k.a. discard/zerorange. 322 */ 323 struct file *file = lo->lo_backing_file; 324 struct request_queue *q = lo->lo_queue; 325 int ret; 326 327 mode |= FALLOC_FL_KEEP_SIZE; 328 329 if (!blk_queue_discard(q)) { 330 ret = -EOPNOTSUPP; 331 goto out; 332 } 333 334 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 335 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 336 ret = -EIO; 337 out: 338 return ret; 339 } 340 341 static int lo_req_flush(struct loop_device *lo, struct request *rq) 342 { 343 struct file *file = lo->lo_backing_file; 344 int ret = vfs_fsync(file, 0); 345 if (unlikely(ret && ret != -EINVAL)) 346 ret = -EIO; 347 348 return ret; 349 } 350 351 static void lo_complete_rq(struct request *rq) 352 { 353 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 354 blk_status_t ret = BLK_STS_OK; 355 356 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 357 req_op(rq) != REQ_OP_READ) { 358 if (cmd->ret < 0) 359 ret = errno_to_blk_status(cmd->ret); 360 goto end_io; 361 } 362 363 /* 364 * Short READ - if we got some data, advance our request and 365 * retry it. If we got no data, end the rest with EIO. 366 */ 367 if (cmd->ret) { 368 blk_update_request(rq, BLK_STS_OK, cmd->ret); 369 cmd->ret = 0; 370 blk_mq_requeue_request(rq, true); 371 } else { 372 if (cmd->use_aio) { 373 struct bio *bio = rq->bio; 374 375 while (bio) { 376 zero_fill_bio(bio); 377 bio = bio->bi_next; 378 } 379 } 380 ret = BLK_STS_IOERR; 381 end_io: 382 blk_mq_end_request(rq, ret); 383 } 384 } 385 386 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 387 { 388 struct request *rq = blk_mq_rq_from_pdu(cmd); 389 390 if (!atomic_dec_and_test(&cmd->ref)) 391 return; 392 kfree(cmd->bvec); 393 cmd->bvec = NULL; 394 if (likely(!blk_should_fake_timeout(rq->q))) 395 blk_mq_complete_request(rq); 396 } 397 398 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 399 { 400 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 401 402 cmd->ret = ret; 403 lo_rw_aio_do_completion(cmd); 404 } 405 406 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 407 loff_t pos, bool rw) 408 { 409 struct iov_iter iter; 410 struct req_iterator rq_iter; 411 struct bio_vec *bvec; 412 struct request *rq = blk_mq_rq_from_pdu(cmd); 413 struct bio *bio = rq->bio; 414 struct file *file = lo->lo_backing_file; 415 struct bio_vec tmp; 416 unsigned int offset; 417 int nr_bvec = 0; 418 int ret; 419 420 rq_for_each_bvec(tmp, rq, rq_iter) 421 nr_bvec++; 422 423 if (rq->bio != rq->biotail) { 424 425 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 426 GFP_NOIO); 427 if (!bvec) 428 return -EIO; 429 cmd->bvec = bvec; 430 431 /* 432 * The bios of the request may be started from the middle of 433 * the 'bvec' because of bio splitting, so we can't directly 434 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 435 * API will take care of all details for us. 436 */ 437 rq_for_each_bvec(tmp, rq, rq_iter) { 438 *bvec = tmp; 439 bvec++; 440 } 441 bvec = cmd->bvec; 442 offset = 0; 443 } else { 444 /* 445 * Same here, this bio may be started from the middle of the 446 * 'bvec' because of bio splitting, so offset from the bvec 447 * must be passed to iov iterator 448 */ 449 offset = bio->bi_iter.bi_bvec_done; 450 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 451 } 452 atomic_set(&cmd->ref, 2); 453 454 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 455 iter.iov_offset = offset; 456 457 cmd->iocb.ki_pos = pos; 458 cmd->iocb.ki_filp = file; 459 cmd->iocb.ki_complete = lo_rw_aio_complete; 460 cmd->iocb.ki_flags = IOCB_DIRECT; 461 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 462 463 if (rw == WRITE) 464 ret = call_write_iter(file, &cmd->iocb, &iter); 465 else 466 ret = call_read_iter(file, &cmd->iocb, &iter); 467 468 lo_rw_aio_do_completion(cmd); 469 470 if (ret != -EIOCBQUEUED) 471 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 472 return 0; 473 } 474 475 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 476 { 477 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 478 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 479 480 /* 481 * lo_write_simple and lo_read_simple should have been covered 482 * by io submit style function like lo_rw_aio(), one blocker 483 * is that lo_read_simple() need to call flush_dcache_page after 484 * the page is written from kernel, and it isn't easy to handle 485 * this in io submit style function which submits all segments 486 * of the req at one time. And direct read IO doesn't need to 487 * run flush_dcache_page(). 488 */ 489 switch (req_op(rq)) { 490 case REQ_OP_FLUSH: 491 return lo_req_flush(lo, rq); 492 case REQ_OP_WRITE_ZEROES: 493 /* 494 * If the caller doesn't want deallocation, call zeroout to 495 * write zeroes the range. Otherwise, punch them out. 496 */ 497 return lo_fallocate(lo, rq, pos, 498 (rq->cmd_flags & REQ_NOUNMAP) ? 499 FALLOC_FL_ZERO_RANGE : 500 FALLOC_FL_PUNCH_HOLE); 501 case REQ_OP_DISCARD: 502 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 503 case REQ_OP_WRITE: 504 if (cmd->use_aio) 505 return lo_rw_aio(lo, cmd, pos, WRITE); 506 else 507 return lo_write_simple(lo, rq, pos); 508 case REQ_OP_READ: 509 if (cmd->use_aio) 510 return lo_rw_aio(lo, cmd, pos, READ); 511 else 512 return lo_read_simple(lo, rq, pos); 513 default: 514 WARN_ON_ONCE(1); 515 return -EIO; 516 } 517 } 518 519 static inline void loop_update_dio(struct loop_device *lo) 520 { 521 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 522 lo->use_dio); 523 } 524 525 static void loop_reread_partitions(struct loop_device *lo) 526 { 527 int rc; 528 529 mutex_lock(&lo->lo_disk->open_mutex); 530 rc = bdev_disk_changed(lo->lo_disk, false); 531 mutex_unlock(&lo->lo_disk->open_mutex); 532 if (rc) 533 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 534 __func__, lo->lo_number, lo->lo_file_name, rc); 535 } 536 537 static inline int is_loop_device(struct file *file) 538 { 539 struct inode *i = file->f_mapping->host; 540 541 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; 542 } 543 544 static int loop_validate_file(struct file *file, struct block_device *bdev) 545 { 546 struct inode *inode = file->f_mapping->host; 547 struct file *f = file; 548 549 /* Avoid recursion */ 550 while (is_loop_device(f)) { 551 struct loop_device *l; 552 553 lockdep_assert_held(&loop_validate_mutex); 554 if (f->f_mapping->host->i_rdev == bdev->bd_dev) 555 return -EBADF; 556 557 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 558 if (l->lo_state != Lo_bound) 559 return -EINVAL; 560 /* Order wrt setting lo->lo_backing_file in loop_configure(). */ 561 rmb(); 562 f = l->lo_backing_file; 563 } 564 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 565 return -EINVAL; 566 return 0; 567 } 568 569 /* 570 * loop_change_fd switched the backing store of a loopback device to 571 * a new file. This is useful for operating system installers to free up 572 * the original file and in High Availability environments to switch to 573 * an alternative location for the content in case of server meltdown. 574 * This can only work if the loop device is used read-only, and if the 575 * new backing store is the same size and type as the old backing store. 576 */ 577 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 578 unsigned int arg) 579 { 580 struct file *file = fget(arg); 581 struct file *old_file; 582 int error; 583 bool partscan; 584 bool is_loop; 585 586 if (!file) 587 return -EBADF; 588 is_loop = is_loop_device(file); 589 error = loop_global_lock_killable(lo, is_loop); 590 if (error) 591 goto out_putf; 592 error = -ENXIO; 593 if (lo->lo_state != Lo_bound) 594 goto out_err; 595 596 /* the loop device has to be read-only */ 597 error = -EINVAL; 598 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 599 goto out_err; 600 601 error = loop_validate_file(file, bdev); 602 if (error) 603 goto out_err; 604 605 old_file = lo->lo_backing_file; 606 607 error = -EINVAL; 608 609 /* size of the new backing store needs to be the same */ 610 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 611 goto out_err; 612 613 /* and ... switch */ 614 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 615 blk_mq_freeze_queue(lo->lo_queue); 616 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 617 lo->lo_backing_file = file; 618 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 619 mapping_set_gfp_mask(file->f_mapping, 620 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 621 loop_update_dio(lo); 622 blk_mq_unfreeze_queue(lo->lo_queue); 623 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 624 loop_global_unlock(lo, is_loop); 625 626 /* 627 * Flush loop_validate_file() before fput(), for l->lo_backing_file 628 * might be pointing at old_file which might be the last reference. 629 */ 630 if (!is_loop) { 631 mutex_lock(&loop_validate_mutex); 632 mutex_unlock(&loop_validate_mutex); 633 } 634 /* 635 * We must drop file reference outside of lo_mutex as dropping 636 * the file ref can take open_mutex which creates circular locking 637 * dependency. 638 */ 639 fput(old_file); 640 if (partscan) 641 loop_reread_partitions(lo); 642 return 0; 643 644 out_err: 645 loop_global_unlock(lo, is_loop); 646 out_putf: 647 fput(file); 648 return error; 649 } 650 651 /* loop sysfs attributes */ 652 653 static ssize_t loop_attr_show(struct device *dev, char *page, 654 ssize_t (*callback)(struct loop_device *, char *)) 655 { 656 struct gendisk *disk = dev_to_disk(dev); 657 struct loop_device *lo = disk->private_data; 658 659 return callback(lo, page); 660 } 661 662 #define LOOP_ATTR_RO(_name) \ 663 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 664 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 665 struct device_attribute *attr, char *b) \ 666 { \ 667 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 668 } \ 669 static struct device_attribute loop_attr_##_name = \ 670 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 671 672 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 673 { 674 ssize_t ret; 675 char *p = NULL; 676 677 spin_lock_irq(&lo->lo_lock); 678 if (lo->lo_backing_file) 679 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 680 spin_unlock_irq(&lo->lo_lock); 681 682 if (IS_ERR_OR_NULL(p)) 683 ret = PTR_ERR(p); 684 else { 685 ret = strlen(p); 686 memmove(buf, p, ret); 687 buf[ret++] = '\n'; 688 buf[ret] = 0; 689 } 690 691 return ret; 692 } 693 694 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 695 { 696 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 697 } 698 699 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 700 { 701 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 702 } 703 704 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 705 { 706 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 707 708 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 709 } 710 711 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 712 { 713 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 714 715 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 716 } 717 718 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 719 { 720 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 721 722 return sprintf(buf, "%s\n", dio ? "1" : "0"); 723 } 724 725 LOOP_ATTR_RO(backing_file); 726 LOOP_ATTR_RO(offset); 727 LOOP_ATTR_RO(sizelimit); 728 LOOP_ATTR_RO(autoclear); 729 LOOP_ATTR_RO(partscan); 730 LOOP_ATTR_RO(dio); 731 732 static struct attribute *loop_attrs[] = { 733 &loop_attr_backing_file.attr, 734 &loop_attr_offset.attr, 735 &loop_attr_sizelimit.attr, 736 &loop_attr_autoclear.attr, 737 &loop_attr_partscan.attr, 738 &loop_attr_dio.attr, 739 NULL, 740 }; 741 742 static struct attribute_group loop_attribute_group = { 743 .name = "loop", 744 .attrs= loop_attrs, 745 }; 746 747 static void loop_sysfs_init(struct loop_device *lo) 748 { 749 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 750 &loop_attribute_group); 751 } 752 753 static void loop_sysfs_exit(struct loop_device *lo) 754 { 755 if (lo->sysfs_inited) 756 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 757 &loop_attribute_group); 758 } 759 760 static void loop_config_discard(struct loop_device *lo) 761 { 762 struct file *file = lo->lo_backing_file; 763 struct inode *inode = file->f_mapping->host; 764 struct request_queue *q = lo->lo_queue; 765 u32 granularity, max_discard_sectors; 766 767 /* 768 * If the backing device is a block device, mirror its zeroing 769 * capability. Set the discard sectors to the block device's zeroing 770 * capabilities because loop discards result in blkdev_issue_zeroout(), 771 * not blkdev_issue_discard(). This maintains consistent behavior with 772 * file-backed loop devices: discarded regions read back as zero. 773 */ 774 if (S_ISBLK(inode->i_mode)) { 775 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 776 777 max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 778 granularity = backingq->limits.discard_granularity ?: 779 queue_physical_block_size(backingq); 780 781 /* 782 * We use punch hole to reclaim the free space used by the 783 * image a.k.a. discard. 784 */ 785 } else if (!file->f_op->fallocate) { 786 max_discard_sectors = 0; 787 granularity = 0; 788 789 } else { 790 max_discard_sectors = UINT_MAX >> 9; 791 granularity = inode->i_sb->s_blocksize; 792 } 793 794 if (max_discard_sectors) { 795 q->limits.discard_granularity = granularity; 796 blk_queue_max_discard_sectors(q, max_discard_sectors); 797 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 798 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 799 } else { 800 q->limits.discard_granularity = 0; 801 blk_queue_max_discard_sectors(q, 0); 802 blk_queue_max_write_zeroes_sectors(q, 0); 803 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 804 } 805 q->limits.discard_alignment = 0; 806 } 807 808 struct loop_worker { 809 struct rb_node rb_node; 810 struct work_struct work; 811 struct list_head cmd_list; 812 struct list_head idle_list; 813 struct loop_device *lo; 814 struct cgroup_subsys_state *blkcg_css; 815 unsigned long last_ran_at; 816 }; 817 818 static void loop_workfn(struct work_struct *work); 819 static void loop_rootcg_workfn(struct work_struct *work); 820 static void loop_free_idle_workers(struct timer_list *timer); 821 822 #ifdef CONFIG_BLK_CGROUP 823 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 824 { 825 return !css || css == blkcg_root_css; 826 } 827 #else 828 static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 829 { 830 return !css; 831 } 832 #endif 833 834 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 835 { 836 struct rb_node **node = &(lo->worker_tree.rb_node), *parent = NULL; 837 struct loop_worker *cur_worker, *worker = NULL; 838 struct work_struct *work; 839 struct list_head *cmd_list; 840 841 spin_lock_irq(&lo->lo_work_lock); 842 843 if (queue_on_root_worker(cmd->blkcg_css)) 844 goto queue_work; 845 846 node = &lo->worker_tree.rb_node; 847 848 while (*node) { 849 parent = *node; 850 cur_worker = container_of(*node, struct loop_worker, rb_node); 851 if (cur_worker->blkcg_css == cmd->blkcg_css) { 852 worker = cur_worker; 853 break; 854 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { 855 node = &(*node)->rb_left; 856 } else { 857 node = &(*node)->rb_right; 858 } 859 } 860 if (worker) 861 goto queue_work; 862 863 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 864 /* 865 * In the event we cannot allocate a worker, just queue on the 866 * rootcg worker and issue the I/O as the rootcg 867 */ 868 if (!worker) { 869 cmd->blkcg_css = NULL; 870 if (cmd->memcg_css) 871 css_put(cmd->memcg_css); 872 cmd->memcg_css = NULL; 873 goto queue_work; 874 } 875 876 worker->blkcg_css = cmd->blkcg_css; 877 css_get(worker->blkcg_css); 878 INIT_WORK(&worker->work, loop_workfn); 879 INIT_LIST_HEAD(&worker->cmd_list); 880 INIT_LIST_HEAD(&worker->idle_list); 881 worker->lo = lo; 882 rb_link_node(&worker->rb_node, parent, node); 883 rb_insert_color(&worker->rb_node, &lo->worker_tree); 884 queue_work: 885 if (worker) { 886 /* 887 * We need to remove from the idle list here while 888 * holding the lock so that the idle timer doesn't 889 * free the worker 890 */ 891 if (!list_empty(&worker->idle_list)) 892 list_del_init(&worker->idle_list); 893 work = &worker->work; 894 cmd_list = &worker->cmd_list; 895 } else { 896 work = &lo->rootcg_work; 897 cmd_list = &lo->rootcg_cmd_list; 898 } 899 list_add_tail(&cmd->list_entry, cmd_list); 900 queue_work(lo->workqueue, work); 901 spin_unlock_irq(&lo->lo_work_lock); 902 } 903 904 static void loop_update_rotational(struct loop_device *lo) 905 { 906 struct file *file = lo->lo_backing_file; 907 struct inode *file_inode = file->f_mapping->host; 908 struct block_device *file_bdev = file_inode->i_sb->s_bdev; 909 struct request_queue *q = lo->lo_queue; 910 bool nonrot = true; 911 912 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 913 if (file_bdev) 914 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 915 916 if (nonrot) 917 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 918 else 919 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 920 } 921 922 /** 923 * loop_set_status_from_info - configure device from loop_info 924 * @lo: struct loop_device to configure 925 * @info: struct loop_info64 to configure the device with 926 * 927 * Configures the loop device parameters according to the passed 928 * in loop_info64 configuration. 929 */ 930 static int 931 loop_set_status_from_info(struct loop_device *lo, 932 const struct loop_info64 *info) 933 { 934 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 935 return -EINVAL; 936 937 switch (info->lo_encrypt_type) { 938 case LO_CRYPT_NONE: 939 break; 940 case LO_CRYPT_XOR: 941 pr_warn("support for the xor transformation has been removed.\n"); 942 return -EINVAL; 943 case LO_CRYPT_CRYPTOAPI: 944 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n"); 945 return -EINVAL; 946 default: 947 return -EINVAL; 948 } 949 950 lo->lo_offset = info->lo_offset; 951 lo->lo_sizelimit = info->lo_sizelimit; 952 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 953 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 954 lo->lo_flags = info->lo_flags; 955 return 0; 956 } 957 958 static int loop_configure(struct loop_device *lo, fmode_t mode, 959 struct block_device *bdev, 960 const struct loop_config *config) 961 { 962 struct file *file = fget(config->fd); 963 struct inode *inode; 964 struct address_space *mapping; 965 int error; 966 loff_t size; 967 bool partscan; 968 unsigned short bsize; 969 bool is_loop; 970 971 if (!file) 972 return -EBADF; 973 is_loop = is_loop_device(file); 974 975 /* This is safe, since we have a reference from open(). */ 976 __module_get(THIS_MODULE); 977 978 /* 979 * If we don't hold exclusive handle for the device, upgrade to it 980 * here to avoid changing device under exclusive owner. 981 */ 982 if (!(mode & FMODE_EXCL)) { 983 error = bd_prepare_to_claim(bdev, loop_configure); 984 if (error) 985 goto out_putf; 986 } 987 988 error = loop_global_lock_killable(lo, is_loop); 989 if (error) 990 goto out_bdev; 991 992 error = -EBUSY; 993 if (lo->lo_state != Lo_unbound) 994 goto out_unlock; 995 996 error = loop_validate_file(file, bdev); 997 if (error) 998 goto out_unlock; 999 1000 mapping = file->f_mapping; 1001 inode = mapping->host; 1002 1003 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 1004 error = -EINVAL; 1005 goto out_unlock; 1006 } 1007 1008 if (config->block_size) { 1009 error = loop_validate_block_size(config->block_size); 1010 if (error) 1011 goto out_unlock; 1012 } 1013 1014 error = loop_set_status_from_info(lo, &config->info); 1015 if (error) 1016 goto out_unlock; 1017 1018 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1019 !file->f_op->write_iter) 1020 lo->lo_flags |= LO_FLAGS_READ_ONLY; 1021 1022 lo->workqueue = alloc_workqueue("loop%d", 1023 WQ_UNBOUND | WQ_FREEZABLE, 1024 0, 1025 lo->lo_number); 1026 if (!lo->workqueue) { 1027 error = -ENOMEM; 1028 goto out_unlock; 1029 } 1030 1031 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1032 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 1033 1034 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 1035 INIT_LIST_HEAD(&lo->rootcg_cmd_list); 1036 INIT_LIST_HEAD(&lo->idle_worker_list); 1037 lo->worker_tree = RB_ROOT; 1038 timer_setup(&lo->timer, loop_free_idle_workers, 1039 TIMER_DEFERRABLE); 1040 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 1041 lo->lo_device = bdev; 1042 lo->lo_backing_file = file; 1043 lo->old_gfp_mask = mapping_gfp_mask(mapping); 1044 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 1045 1046 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 1047 blk_queue_write_cache(lo->lo_queue, true, false); 1048 1049 if (config->block_size) 1050 bsize = config->block_size; 1051 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 1052 /* In case of direct I/O, match underlying block size */ 1053 bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1054 else 1055 bsize = 512; 1056 1057 blk_queue_logical_block_size(lo->lo_queue, bsize); 1058 blk_queue_physical_block_size(lo->lo_queue, bsize); 1059 blk_queue_io_min(lo->lo_queue, bsize); 1060 1061 loop_config_discard(lo); 1062 loop_update_rotational(lo); 1063 loop_update_dio(lo); 1064 loop_sysfs_init(lo); 1065 1066 size = get_loop_size(lo, file); 1067 loop_set_size(lo, size); 1068 1069 /* Order wrt reading lo_state in loop_validate_file(). */ 1070 wmb(); 1071 1072 lo->lo_state = Lo_bound; 1073 if (part_shift) 1074 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1075 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1076 if (partscan) 1077 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1078 1079 loop_global_unlock(lo, is_loop); 1080 if (partscan) 1081 loop_reread_partitions(lo); 1082 if (!(mode & FMODE_EXCL)) 1083 bd_abort_claiming(bdev, loop_configure); 1084 return 0; 1085 1086 out_unlock: 1087 loop_global_unlock(lo, is_loop); 1088 out_bdev: 1089 if (!(mode & FMODE_EXCL)) 1090 bd_abort_claiming(bdev, loop_configure); 1091 out_putf: 1092 fput(file); 1093 /* This is safe: open() is still holding a reference. */ 1094 module_put(THIS_MODULE); 1095 return error; 1096 } 1097 1098 static int __loop_clr_fd(struct loop_device *lo, bool release) 1099 { 1100 struct file *filp = NULL; 1101 gfp_t gfp = lo->old_gfp_mask; 1102 struct block_device *bdev = lo->lo_device; 1103 int err = 0; 1104 bool partscan = false; 1105 int lo_number; 1106 struct loop_worker *pos, *worker; 1107 1108 /* 1109 * Flush loop_configure() and loop_change_fd(). It is acceptable for 1110 * loop_validate_file() to succeed, for actual clear operation has not 1111 * started yet. 1112 */ 1113 mutex_lock(&loop_validate_mutex); 1114 mutex_unlock(&loop_validate_mutex); 1115 /* 1116 * loop_validate_file() now fails because l->lo_state != Lo_bound 1117 * became visible. 1118 */ 1119 1120 mutex_lock(&lo->lo_mutex); 1121 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1122 err = -ENXIO; 1123 goto out_unlock; 1124 } 1125 1126 filp = lo->lo_backing_file; 1127 if (filp == NULL) { 1128 err = -EINVAL; 1129 goto out_unlock; 1130 } 1131 1132 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 1133 blk_queue_write_cache(lo->lo_queue, false, false); 1134 1135 /* freeze request queue during the transition */ 1136 blk_mq_freeze_queue(lo->lo_queue); 1137 1138 destroy_workqueue(lo->workqueue); 1139 spin_lock_irq(&lo->lo_work_lock); 1140 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 1141 idle_list) { 1142 list_del(&worker->idle_list); 1143 rb_erase(&worker->rb_node, &lo->worker_tree); 1144 css_put(worker->blkcg_css); 1145 kfree(worker); 1146 } 1147 spin_unlock_irq(&lo->lo_work_lock); 1148 del_timer_sync(&lo->timer); 1149 1150 spin_lock_irq(&lo->lo_lock); 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 if (bdev) { 1162 invalidate_bdev(bdev); 1163 bdev->bd_inode->i_mapping->wb_err = 0; 1164 } 1165 set_capacity(lo->lo_disk, 0); 1166 loop_sysfs_exit(lo); 1167 if (bdev) { 1168 /* let user-space know about this change */ 1169 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1170 } 1171 mapping_set_gfp_mask(filp->f_mapping, gfp); 1172 /* This is safe: open() is still holding a reference. */ 1173 module_put(THIS_MODULE); 1174 blk_mq_unfreeze_queue(lo->lo_queue); 1175 1176 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1177 lo_number = lo->lo_number; 1178 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 1179 out_unlock: 1180 mutex_unlock(&lo->lo_mutex); 1181 if (partscan) { 1182 /* 1183 * open_mutex has been held already in release path, so don't 1184 * acquire it if this function is called in such case. 1185 * 1186 * If the reread partition isn't from release path, lo_refcnt 1187 * must be at least one and it can only become zero when the 1188 * current holder is released. 1189 */ 1190 if (!release) 1191 mutex_lock(&lo->lo_disk->open_mutex); 1192 err = bdev_disk_changed(lo->lo_disk, false); 1193 if (!release) 1194 mutex_unlock(&lo->lo_disk->open_mutex); 1195 if (err) 1196 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1197 __func__, lo_number, err); 1198 /* Device is gone, no point in returning error */ 1199 err = 0; 1200 } 1201 1202 /* 1203 * lo->lo_state is set to Lo_unbound here after above partscan has 1204 * finished. 1205 * 1206 * There cannot be anybody else entering __loop_clr_fd() as 1207 * lo->lo_backing_file is already cleared and Lo_rundown state 1208 * protects us from all the other places trying to change the 'lo' 1209 * device. 1210 */ 1211 mutex_lock(&lo->lo_mutex); 1212 lo->lo_flags = 0; 1213 if (!part_shift) 1214 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1215 lo->lo_state = Lo_unbound; 1216 mutex_unlock(&lo->lo_mutex); 1217 1218 /* 1219 * Need not hold lo_mutex to fput backing file. Calling fput holding 1220 * lo_mutex triggers a circular lock dependency possibility warning as 1221 * fput can take open_mutex which is usually taken before lo_mutex. 1222 */ 1223 if (filp) 1224 fput(filp); 1225 return err; 1226 } 1227 1228 static int loop_clr_fd(struct loop_device *lo) 1229 { 1230 int err; 1231 1232 err = mutex_lock_killable(&lo->lo_mutex); 1233 if (err) 1234 return err; 1235 if (lo->lo_state != Lo_bound) { 1236 mutex_unlock(&lo->lo_mutex); 1237 return -ENXIO; 1238 } 1239 /* 1240 * If we've explicitly asked to tear down the loop device, 1241 * and it has an elevated reference count, set it for auto-teardown when 1242 * the last reference goes away. This stops $!~#$@ udev from 1243 * preventing teardown because it decided that it needs to run blkid on 1244 * the loopback device whenever they appear. xfstests is notorious for 1245 * failing tests because blkid via udev races with a losetup 1246 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1247 * command to fail with EBUSY. 1248 */ 1249 if (atomic_read(&lo->lo_refcnt) > 1) { 1250 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1251 mutex_unlock(&lo->lo_mutex); 1252 return 0; 1253 } 1254 lo->lo_state = Lo_rundown; 1255 mutex_unlock(&lo->lo_mutex); 1256 1257 return __loop_clr_fd(lo, false); 1258 } 1259 1260 static int 1261 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1262 { 1263 int err; 1264 int prev_lo_flags; 1265 bool partscan = false; 1266 bool size_changed = false; 1267 1268 err = mutex_lock_killable(&lo->lo_mutex); 1269 if (err) 1270 return err; 1271 if (lo->lo_state != Lo_bound) { 1272 err = -ENXIO; 1273 goto out_unlock; 1274 } 1275 1276 if (lo->lo_offset != info->lo_offset || 1277 lo->lo_sizelimit != info->lo_sizelimit) { 1278 size_changed = true; 1279 sync_blockdev(lo->lo_device); 1280 invalidate_bdev(lo->lo_device); 1281 } 1282 1283 /* I/O need to be drained during transfer transition */ 1284 blk_mq_freeze_queue(lo->lo_queue); 1285 1286 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1287 /* If any pages were dirtied after invalidate_bdev(), try again */ 1288 err = -EAGAIN; 1289 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1290 __func__, lo->lo_number, lo->lo_file_name, 1291 lo->lo_device->bd_inode->i_mapping->nrpages); 1292 goto out_unfreeze; 1293 } 1294 1295 prev_lo_flags = lo->lo_flags; 1296 1297 err = loop_set_status_from_info(lo, info); 1298 if (err) 1299 goto out_unfreeze; 1300 1301 /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 1302 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1303 /* For those flags, use the previous values instead */ 1304 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1305 /* For flags that can't be cleared, use previous values too */ 1306 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1307 1308 if (size_changed) { 1309 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1310 lo->lo_backing_file); 1311 loop_set_size(lo, new_size); 1312 } 1313 1314 loop_config_discard(lo); 1315 1316 /* update dio if lo_offset or transfer is changed */ 1317 __loop_update_dio(lo, lo->use_dio); 1318 1319 out_unfreeze: 1320 blk_mq_unfreeze_queue(lo->lo_queue); 1321 1322 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1323 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 1324 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1325 partscan = true; 1326 } 1327 out_unlock: 1328 mutex_unlock(&lo->lo_mutex); 1329 if (partscan) 1330 loop_reread_partitions(lo); 1331 1332 return err; 1333 } 1334 1335 static int 1336 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1337 { 1338 struct path path; 1339 struct kstat stat; 1340 int ret; 1341 1342 ret = mutex_lock_killable(&lo->lo_mutex); 1343 if (ret) 1344 return ret; 1345 if (lo->lo_state != Lo_bound) { 1346 mutex_unlock(&lo->lo_mutex); 1347 return -ENXIO; 1348 } 1349 1350 memset(info, 0, sizeof(*info)); 1351 info->lo_number = lo->lo_number; 1352 info->lo_offset = lo->lo_offset; 1353 info->lo_sizelimit = lo->lo_sizelimit; 1354 info->lo_flags = lo->lo_flags; 1355 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1356 1357 /* Drop lo_mutex while we call into the filesystem. */ 1358 path = lo->lo_backing_file->f_path; 1359 path_get(&path); 1360 mutex_unlock(&lo->lo_mutex); 1361 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1362 if (!ret) { 1363 info->lo_device = huge_encode_dev(stat.dev); 1364 info->lo_inode = stat.ino; 1365 info->lo_rdevice = huge_encode_dev(stat.rdev); 1366 } 1367 path_put(&path); 1368 return ret; 1369 } 1370 1371 static void 1372 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1373 { 1374 memset(info64, 0, sizeof(*info64)); 1375 info64->lo_number = info->lo_number; 1376 info64->lo_device = info->lo_device; 1377 info64->lo_inode = info->lo_inode; 1378 info64->lo_rdevice = info->lo_rdevice; 1379 info64->lo_offset = info->lo_offset; 1380 info64->lo_sizelimit = 0; 1381 info64->lo_flags = info->lo_flags; 1382 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1383 } 1384 1385 static int 1386 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1387 { 1388 memset(info, 0, sizeof(*info)); 1389 info->lo_number = info64->lo_number; 1390 info->lo_device = info64->lo_device; 1391 info->lo_inode = info64->lo_inode; 1392 info->lo_rdevice = info64->lo_rdevice; 1393 info->lo_offset = info64->lo_offset; 1394 info->lo_flags = info64->lo_flags; 1395 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1396 1397 /* error in case values were truncated */ 1398 if (info->lo_device != info64->lo_device || 1399 info->lo_rdevice != info64->lo_rdevice || 1400 info->lo_inode != info64->lo_inode || 1401 info->lo_offset != info64->lo_offset) 1402 return -EOVERFLOW; 1403 1404 return 0; 1405 } 1406 1407 static int 1408 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1409 { 1410 struct loop_info info; 1411 struct loop_info64 info64; 1412 1413 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1414 return -EFAULT; 1415 loop_info64_from_old(&info, &info64); 1416 return loop_set_status(lo, &info64); 1417 } 1418 1419 static int 1420 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1421 { 1422 struct loop_info64 info64; 1423 1424 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1425 return -EFAULT; 1426 return loop_set_status(lo, &info64); 1427 } 1428 1429 static int 1430 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1431 struct loop_info info; 1432 struct loop_info64 info64; 1433 int err; 1434 1435 if (!arg) 1436 return -EINVAL; 1437 err = loop_get_status(lo, &info64); 1438 if (!err) 1439 err = loop_info64_to_old(&info64, &info); 1440 if (!err && copy_to_user(arg, &info, sizeof(info))) 1441 err = -EFAULT; 1442 1443 return err; 1444 } 1445 1446 static int 1447 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1448 struct loop_info64 info64; 1449 int err; 1450 1451 if (!arg) 1452 return -EINVAL; 1453 err = loop_get_status(lo, &info64); 1454 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1455 err = -EFAULT; 1456 1457 return err; 1458 } 1459 1460 static int loop_set_capacity(struct loop_device *lo) 1461 { 1462 loff_t size; 1463 1464 if (unlikely(lo->lo_state != Lo_bound)) 1465 return -ENXIO; 1466 1467 size = get_loop_size(lo, lo->lo_backing_file); 1468 loop_set_size(lo, size); 1469 1470 return 0; 1471 } 1472 1473 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1474 { 1475 int error = -ENXIO; 1476 if (lo->lo_state != Lo_bound) 1477 goto out; 1478 1479 __loop_update_dio(lo, !!arg); 1480 if (lo->use_dio == !!arg) 1481 return 0; 1482 error = -EINVAL; 1483 out: 1484 return error; 1485 } 1486 1487 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1488 { 1489 int err = 0; 1490 1491 if (lo->lo_state != Lo_bound) 1492 return -ENXIO; 1493 1494 err = loop_validate_block_size(arg); 1495 if (err) 1496 return err; 1497 1498 if (lo->lo_queue->limits.logical_block_size == arg) 1499 return 0; 1500 1501 sync_blockdev(lo->lo_device); 1502 invalidate_bdev(lo->lo_device); 1503 1504 blk_mq_freeze_queue(lo->lo_queue); 1505 1506 /* invalidate_bdev should have truncated all the pages */ 1507 if (lo->lo_device->bd_inode->i_mapping->nrpages) { 1508 err = -EAGAIN; 1509 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1510 __func__, lo->lo_number, lo->lo_file_name, 1511 lo->lo_device->bd_inode->i_mapping->nrpages); 1512 goto out_unfreeze; 1513 } 1514 1515 blk_queue_logical_block_size(lo->lo_queue, arg); 1516 blk_queue_physical_block_size(lo->lo_queue, arg); 1517 blk_queue_io_min(lo->lo_queue, arg); 1518 loop_update_dio(lo); 1519 out_unfreeze: 1520 blk_mq_unfreeze_queue(lo->lo_queue); 1521 1522 return err; 1523 } 1524 1525 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1526 unsigned long arg) 1527 { 1528 int err; 1529 1530 err = mutex_lock_killable(&lo->lo_mutex); 1531 if (err) 1532 return err; 1533 switch (cmd) { 1534 case LOOP_SET_CAPACITY: 1535 err = loop_set_capacity(lo); 1536 break; 1537 case LOOP_SET_DIRECT_IO: 1538 err = loop_set_dio(lo, arg); 1539 break; 1540 case LOOP_SET_BLOCK_SIZE: 1541 err = loop_set_block_size(lo, arg); 1542 break; 1543 default: 1544 err = -EINVAL; 1545 } 1546 mutex_unlock(&lo->lo_mutex); 1547 return err; 1548 } 1549 1550 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1551 unsigned int cmd, unsigned long arg) 1552 { 1553 struct loop_device *lo = bdev->bd_disk->private_data; 1554 void __user *argp = (void __user *) arg; 1555 int err; 1556 1557 switch (cmd) { 1558 case LOOP_SET_FD: { 1559 /* 1560 * Legacy case - pass in a zeroed out struct loop_config with 1561 * only the file descriptor set , which corresponds with the 1562 * default parameters we'd have used otherwise. 1563 */ 1564 struct loop_config config; 1565 1566 memset(&config, 0, sizeof(config)); 1567 config.fd = arg; 1568 1569 return loop_configure(lo, mode, bdev, &config); 1570 } 1571 case LOOP_CONFIGURE: { 1572 struct loop_config config; 1573 1574 if (copy_from_user(&config, argp, sizeof(config))) 1575 return -EFAULT; 1576 1577 return loop_configure(lo, mode, bdev, &config); 1578 } 1579 case LOOP_CHANGE_FD: 1580 return loop_change_fd(lo, bdev, arg); 1581 case LOOP_CLR_FD: 1582 return loop_clr_fd(lo); 1583 case LOOP_SET_STATUS: 1584 err = -EPERM; 1585 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1586 err = loop_set_status_old(lo, argp); 1587 } 1588 break; 1589 case LOOP_GET_STATUS: 1590 return loop_get_status_old(lo, argp); 1591 case LOOP_SET_STATUS64: 1592 err = -EPERM; 1593 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1594 err = loop_set_status64(lo, argp); 1595 } 1596 break; 1597 case LOOP_GET_STATUS64: 1598 return loop_get_status64(lo, argp); 1599 case LOOP_SET_CAPACITY: 1600 case LOOP_SET_DIRECT_IO: 1601 case LOOP_SET_BLOCK_SIZE: 1602 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1603 return -EPERM; 1604 fallthrough; 1605 default: 1606 err = lo_simple_ioctl(lo, cmd, arg); 1607 break; 1608 } 1609 1610 return err; 1611 } 1612 1613 #ifdef CONFIG_COMPAT 1614 struct compat_loop_info { 1615 compat_int_t lo_number; /* ioctl r/o */ 1616 compat_dev_t lo_device; /* ioctl r/o */ 1617 compat_ulong_t lo_inode; /* ioctl r/o */ 1618 compat_dev_t lo_rdevice; /* ioctl r/o */ 1619 compat_int_t lo_offset; 1620 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1621 compat_int_t lo_flags; /* ioctl r/o */ 1622 char lo_name[LO_NAME_SIZE]; 1623 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1624 compat_ulong_t lo_init[2]; 1625 char reserved[4]; 1626 }; 1627 1628 /* 1629 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1630 * - noinlined to reduce stack space usage in main part of driver 1631 */ 1632 static noinline int 1633 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1634 struct loop_info64 *info64) 1635 { 1636 struct compat_loop_info info; 1637 1638 if (copy_from_user(&info, arg, sizeof(info))) 1639 return -EFAULT; 1640 1641 memset(info64, 0, sizeof(*info64)); 1642 info64->lo_number = info.lo_number; 1643 info64->lo_device = info.lo_device; 1644 info64->lo_inode = info.lo_inode; 1645 info64->lo_rdevice = info.lo_rdevice; 1646 info64->lo_offset = info.lo_offset; 1647 info64->lo_sizelimit = 0; 1648 info64->lo_flags = info.lo_flags; 1649 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1650 return 0; 1651 } 1652 1653 /* 1654 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1655 * - noinlined to reduce stack space usage in main part of driver 1656 */ 1657 static noinline int 1658 loop_info64_to_compat(const struct loop_info64 *info64, 1659 struct compat_loop_info __user *arg) 1660 { 1661 struct compat_loop_info info; 1662 1663 memset(&info, 0, sizeof(info)); 1664 info.lo_number = info64->lo_number; 1665 info.lo_device = info64->lo_device; 1666 info.lo_inode = info64->lo_inode; 1667 info.lo_rdevice = info64->lo_rdevice; 1668 info.lo_offset = info64->lo_offset; 1669 info.lo_flags = info64->lo_flags; 1670 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1671 1672 /* error in case values were truncated */ 1673 if (info.lo_device != info64->lo_device || 1674 info.lo_rdevice != info64->lo_rdevice || 1675 info.lo_inode != info64->lo_inode || 1676 info.lo_offset != info64->lo_offset) 1677 return -EOVERFLOW; 1678 1679 if (copy_to_user(arg, &info, sizeof(info))) 1680 return -EFAULT; 1681 return 0; 1682 } 1683 1684 static int 1685 loop_set_status_compat(struct loop_device *lo, 1686 const struct compat_loop_info __user *arg) 1687 { 1688 struct loop_info64 info64; 1689 int ret; 1690 1691 ret = loop_info64_from_compat(arg, &info64); 1692 if (ret < 0) 1693 return ret; 1694 return loop_set_status(lo, &info64); 1695 } 1696 1697 static int 1698 loop_get_status_compat(struct loop_device *lo, 1699 struct compat_loop_info __user *arg) 1700 { 1701 struct loop_info64 info64; 1702 int err; 1703 1704 if (!arg) 1705 return -EINVAL; 1706 err = loop_get_status(lo, &info64); 1707 if (!err) 1708 err = loop_info64_to_compat(&info64, arg); 1709 return err; 1710 } 1711 1712 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1713 unsigned int cmd, unsigned long arg) 1714 { 1715 struct loop_device *lo = bdev->bd_disk->private_data; 1716 int err; 1717 1718 switch(cmd) { 1719 case LOOP_SET_STATUS: 1720 err = loop_set_status_compat(lo, 1721 (const struct compat_loop_info __user *)arg); 1722 break; 1723 case LOOP_GET_STATUS: 1724 err = loop_get_status_compat(lo, 1725 (struct compat_loop_info __user *)arg); 1726 break; 1727 case LOOP_SET_CAPACITY: 1728 case LOOP_CLR_FD: 1729 case LOOP_GET_STATUS64: 1730 case LOOP_SET_STATUS64: 1731 case LOOP_CONFIGURE: 1732 arg = (unsigned long) compat_ptr(arg); 1733 fallthrough; 1734 case LOOP_SET_FD: 1735 case LOOP_CHANGE_FD: 1736 case LOOP_SET_BLOCK_SIZE: 1737 case LOOP_SET_DIRECT_IO: 1738 err = lo_ioctl(bdev, mode, cmd, arg); 1739 break; 1740 default: 1741 err = -ENOIOCTLCMD; 1742 break; 1743 } 1744 return err; 1745 } 1746 #endif 1747 1748 static int lo_open(struct block_device *bdev, fmode_t mode) 1749 { 1750 struct loop_device *lo = bdev->bd_disk->private_data; 1751 int err; 1752 1753 err = mutex_lock_killable(&lo->lo_mutex); 1754 if (err) 1755 return err; 1756 if (lo->lo_state == Lo_deleting) 1757 err = -ENXIO; 1758 else 1759 atomic_inc(&lo->lo_refcnt); 1760 mutex_unlock(&lo->lo_mutex); 1761 return err; 1762 } 1763 1764 static void lo_release(struct gendisk *disk, fmode_t mode) 1765 { 1766 struct loop_device *lo = disk->private_data; 1767 1768 mutex_lock(&lo->lo_mutex); 1769 if (atomic_dec_return(&lo->lo_refcnt)) 1770 goto out_unlock; 1771 1772 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1773 if (lo->lo_state != Lo_bound) 1774 goto out_unlock; 1775 lo->lo_state = Lo_rundown; 1776 mutex_unlock(&lo->lo_mutex); 1777 /* 1778 * In autoclear mode, stop the loop thread 1779 * and remove configuration after last close. 1780 */ 1781 __loop_clr_fd(lo, true); 1782 return; 1783 } else if (lo->lo_state == Lo_bound) { 1784 /* 1785 * Otherwise keep thread (if running) and config, 1786 * but flush possible ongoing bios in thread. 1787 */ 1788 blk_mq_freeze_queue(lo->lo_queue); 1789 blk_mq_unfreeze_queue(lo->lo_queue); 1790 } 1791 1792 out_unlock: 1793 mutex_unlock(&lo->lo_mutex); 1794 } 1795 1796 static const struct block_device_operations lo_fops = { 1797 .owner = THIS_MODULE, 1798 .open = lo_open, 1799 .release = lo_release, 1800 .ioctl = lo_ioctl, 1801 #ifdef CONFIG_COMPAT 1802 .compat_ioctl = lo_compat_ioctl, 1803 #endif 1804 }; 1805 1806 /* 1807 * And now the modules code and kernel interface. 1808 */ 1809 static int max_loop; 1810 module_param(max_loop, int, 0444); 1811 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1812 module_param(max_part, int, 0444); 1813 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1814 MODULE_LICENSE("GPL"); 1815 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1816 1817 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1818 const struct blk_mq_queue_data *bd) 1819 { 1820 struct request *rq = bd->rq; 1821 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1822 struct loop_device *lo = rq->q->queuedata; 1823 1824 blk_mq_start_request(rq); 1825 1826 if (lo->lo_state != Lo_bound) 1827 return BLK_STS_IOERR; 1828 1829 switch (req_op(rq)) { 1830 case REQ_OP_FLUSH: 1831 case REQ_OP_DISCARD: 1832 case REQ_OP_WRITE_ZEROES: 1833 cmd->use_aio = false; 1834 break; 1835 default: 1836 cmd->use_aio = lo->use_dio; 1837 break; 1838 } 1839 1840 /* always use the first bio's css */ 1841 cmd->blkcg_css = NULL; 1842 cmd->memcg_css = NULL; 1843 #ifdef CONFIG_BLK_CGROUP 1844 if (rq->bio && rq->bio->bi_blkg) { 1845 cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 1846 #ifdef CONFIG_MEMCG 1847 cmd->memcg_css = 1848 cgroup_get_e_css(cmd->blkcg_css->cgroup, 1849 &memory_cgrp_subsys); 1850 #endif 1851 } 1852 #endif 1853 loop_queue_work(lo, cmd); 1854 1855 return BLK_STS_OK; 1856 } 1857 1858 static void loop_handle_cmd(struct loop_cmd *cmd) 1859 { 1860 struct request *rq = blk_mq_rq_from_pdu(cmd); 1861 const bool write = op_is_write(req_op(rq)); 1862 struct loop_device *lo = rq->q->queuedata; 1863 int ret = 0; 1864 struct mem_cgroup *old_memcg = NULL; 1865 1866 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1867 ret = -EIO; 1868 goto failed; 1869 } 1870 1871 if (cmd->blkcg_css) 1872 kthread_associate_blkcg(cmd->blkcg_css); 1873 if (cmd->memcg_css) 1874 old_memcg = set_active_memcg( 1875 mem_cgroup_from_css(cmd->memcg_css)); 1876 1877 ret = do_req_filebacked(lo, rq); 1878 1879 if (cmd->blkcg_css) 1880 kthread_associate_blkcg(NULL); 1881 1882 if (cmd->memcg_css) { 1883 set_active_memcg(old_memcg); 1884 css_put(cmd->memcg_css); 1885 } 1886 failed: 1887 /* complete non-aio request */ 1888 if (!cmd->use_aio || ret) { 1889 if (ret == -EOPNOTSUPP) 1890 cmd->ret = ret; 1891 else 1892 cmd->ret = ret ? -EIO : 0; 1893 if (likely(!blk_should_fake_timeout(rq->q))) 1894 blk_mq_complete_request(rq); 1895 } 1896 } 1897 1898 static void loop_set_timer(struct loop_device *lo) 1899 { 1900 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 1901 } 1902 1903 static void loop_process_work(struct loop_worker *worker, 1904 struct list_head *cmd_list, struct loop_device *lo) 1905 { 1906 int orig_flags = current->flags; 1907 struct loop_cmd *cmd; 1908 1909 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 1910 spin_lock_irq(&lo->lo_work_lock); 1911 while (!list_empty(cmd_list)) { 1912 cmd = container_of( 1913 cmd_list->next, struct loop_cmd, list_entry); 1914 list_del(cmd_list->next); 1915 spin_unlock_irq(&lo->lo_work_lock); 1916 1917 loop_handle_cmd(cmd); 1918 cond_resched(); 1919 1920 spin_lock_irq(&lo->lo_work_lock); 1921 } 1922 1923 /* 1924 * We only add to the idle list if there are no pending cmds 1925 * *and* the worker will not run again which ensures that it 1926 * is safe to free any worker on the idle list 1927 */ 1928 if (worker && !work_pending(&worker->work)) { 1929 worker->last_ran_at = jiffies; 1930 list_add_tail(&worker->idle_list, &lo->idle_worker_list); 1931 loop_set_timer(lo); 1932 } 1933 spin_unlock_irq(&lo->lo_work_lock); 1934 current->flags = orig_flags; 1935 } 1936 1937 static void loop_workfn(struct work_struct *work) 1938 { 1939 struct loop_worker *worker = 1940 container_of(work, struct loop_worker, work); 1941 loop_process_work(worker, &worker->cmd_list, worker->lo); 1942 } 1943 1944 static void loop_rootcg_workfn(struct work_struct *work) 1945 { 1946 struct loop_device *lo = 1947 container_of(work, struct loop_device, rootcg_work); 1948 loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 1949 } 1950 1951 static void loop_free_idle_workers(struct timer_list *timer) 1952 { 1953 struct loop_device *lo = container_of(timer, struct loop_device, timer); 1954 struct loop_worker *pos, *worker; 1955 1956 spin_lock_irq(&lo->lo_work_lock); 1957 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 1958 idle_list) { 1959 if (time_is_after_jiffies(worker->last_ran_at + 1960 LOOP_IDLE_WORKER_TIMEOUT)) 1961 break; 1962 list_del(&worker->idle_list); 1963 rb_erase(&worker->rb_node, &lo->worker_tree); 1964 css_put(worker->blkcg_css); 1965 kfree(worker); 1966 } 1967 if (!list_empty(&lo->idle_worker_list)) 1968 loop_set_timer(lo); 1969 spin_unlock_irq(&lo->lo_work_lock); 1970 } 1971 1972 static const struct blk_mq_ops loop_mq_ops = { 1973 .queue_rq = loop_queue_rq, 1974 .complete = lo_complete_rq, 1975 }; 1976 1977 static int loop_add(int i) 1978 { 1979 struct loop_device *lo; 1980 struct gendisk *disk; 1981 int err; 1982 1983 err = -ENOMEM; 1984 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1985 if (!lo) 1986 goto out; 1987 lo->lo_state = Lo_unbound; 1988 1989 err = mutex_lock_killable(&loop_ctl_mutex); 1990 if (err) 1991 goto out_free_dev; 1992 1993 /* allocate id, if @id >= 0, we're requesting that specific id */ 1994 if (i >= 0) { 1995 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1996 if (err == -ENOSPC) 1997 err = -EEXIST; 1998 } else { 1999 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 2000 } 2001 mutex_unlock(&loop_ctl_mutex); 2002 if (err < 0) 2003 goto out_free_dev; 2004 i = err; 2005 2006 err = -ENOMEM; 2007 lo->tag_set.ops = &loop_mq_ops; 2008 lo->tag_set.nr_hw_queues = 1; 2009 lo->tag_set.queue_depth = 128; 2010 lo->tag_set.numa_node = NUMA_NO_NODE; 2011 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 2012 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | 2013 BLK_MQ_F_NO_SCHED_BY_DEFAULT; 2014 lo->tag_set.driver_data = lo; 2015 2016 err = blk_mq_alloc_tag_set(&lo->tag_set); 2017 if (err) 2018 goto out_free_idr; 2019 2020 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 2021 if (IS_ERR(disk)) { 2022 err = PTR_ERR(disk); 2023 goto out_cleanup_tags; 2024 } 2025 lo->lo_queue = lo->lo_disk->queue; 2026 2027 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 2028 2029 /* 2030 * By default, we do buffer IO, so it doesn't make sense to enable 2031 * merge because the I/O submitted to backing file is handled page by 2032 * page. For directio mode, merge does help to dispatch bigger request 2033 * to underlayer disk. We will enable merge once directio is enabled. 2034 */ 2035 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2036 2037 /* 2038 * Disable partition scanning by default. The in-kernel partition 2039 * scanning can be requested individually per-device during its 2040 * setup. Userspace can always add and remove partitions from all 2041 * devices. The needed partition minors are allocated from the 2042 * extended minor space, the main loop device numbers will continue 2043 * to match the loop minors, regardless of the number of partitions 2044 * used. 2045 * 2046 * If max_part is given, partition scanning is globally enabled for 2047 * all loop devices. The minors for the main loop devices will be 2048 * multiples of max_part. 2049 * 2050 * Note: Global-for-all-devices, set-only-at-init, read-only module 2051 * parameteters like 'max_loop' and 'max_part' make things needlessly 2052 * complicated, are too static, inflexible and may surprise 2053 * userspace tools. Parameters like this in general should be avoided. 2054 */ 2055 if (!part_shift) 2056 disk->flags |= GENHD_FL_NO_PART_SCAN; 2057 disk->flags |= GENHD_FL_EXT_DEVT; 2058 atomic_set(&lo->lo_refcnt, 0); 2059 mutex_init(&lo->lo_mutex); 2060 lo->lo_number = i; 2061 spin_lock_init(&lo->lo_lock); 2062 spin_lock_init(&lo->lo_work_lock); 2063 disk->major = LOOP_MAJOR; 2064 disk->first_minor = i << part_shift; 2065 disk->minors = 1 << part_shift; 2066 disk->fops = &lo_fops; 2067 disk->private_data = lo; 2068 disk->queue = lo->lo_queue; 2069 disk->events = DISK_EVENT_MEDIA_CHANGE; 2070 disk->event_flags = DISK_EVENT_FLAG_UEVENT; 2071 sprintf(disk->disk_name, "loop%d", i); 2072 /* Make this loop device reachable from pathname. */ 2073 err = add_disk(disk); 2074 if (err) 2075 goto out_cleanup_disk; 2076 2077 /* Show this loop device. */ 2078 mutex_lock(&loop_ctl_mutex); 2079 lo->idr_visible = true; 2080 mutex_unlock(&loop_ctl_mutex); 2081 2082 return i; 2083 2084 out_cleanup_disk: 2085 blk_cleanup_disk(disk); 2086 out_cleanup_tags: 2087 blk_mq_free_tag_set(&lo->tag_set); 2088 out_free_idr: 2089 mutex_lock(&loop_ctl_mutex); 2090 idr_remove(&loop_index_idr, i); 2091 mutex_unlock(&loop_ctl_mutex); 2092 out_free_dev: 2093 kfree(lo); 2094 out: 2095 return err; 2096 } 2097 2098 static void loop_remove(struct loop_device *lo) 2099 { 2100 /* Make this loop device unreachable from pathname. */ 2101 del_gendisk(lo->lo_disk); 2102 blk_cleanup_disk(lo->lo_disk); 2103 blk_mq_free_tag_set(&lo->tag_set); 2104 mutex_lock(&loop_ctl_mutex); 2105 idr_remove(&loop_index_idr, lo->lo_number); 2106 mutex_unlock(&loop_ctl_mutex); 2107 /* There is no route which can find this loop device. */ 2108 mutex_destroy(&lo->lo_mutex); 2109 kfree(lo); 2110 } 2111 2112 static void loop_probe(dev_t dev) 2113 { 2114 int idx = MINOR(dev) >> part_shift; 2115 2116 if (max_loop && idx >= max_loop) 2117 return; 2118 loop_add(idx); 2119 } 2120 2121 static int loop_control_remove(int idx) 2122 { 2123 struct loop_device *lo; 2124 int ret; 2125 2126 if (idx < 0) { 2127 pr_warn("deleting an unspecified loop device is not supported.\n"); 2128 return -EINVAL; 2129 } 2130 2131 /* Hide this loop device for serialization. */ 2132 ret = mutex_lock_killable(&loop_ctl_mutex); 2133 if (ret) 2134 return ret; 2135 lo = idr_find(&loop_index_idr, idx); 2136 if (!lo || !lo->idr_visible) 2137 ret = -ENODEV; 2138 else 2139 lo->idr_visible = false; 2140 mutex_unlock(&loop_ctl_mutex); 2141 if (ret) 2142 return ret; 2143 2144 /* Check whether this loop device can be removed. */ 2145 ret = mutex_lock_killable(&lo->lo_mutex); 2146 if (ret) 2147 goto mark_visible; 2148 if (lo->lo_state != Lo_unbound || 2149 atomic_read(&lo->lo_refcnt) > 0) { 2150 mutex_unlock(&lo->lo_mutex); 2151 ret = -EBUSY; 2152 goto mark_visible; 2153 } 2154 /* Mark this loop device no longer open()-able. */ 2155 lo->lo_state = Lo_deleting; 2156 mutex_unlock(&lo->lo_mutex); 2157 2158 loop_remove(lo); 2159 return 0; 2160 2161 mark_visible: 2162 /* Show this loop device again. */ 2163 mutex_lock(&loop_ctl_mutex); 2164 lo->idr_visible = true; 2165 mutex_unlock(&loop_ctl_mutex); 2166 return ret; 2167 } 2168 2169 static int loop_control_get_free(int idx) 2170 { 2171 struct loop_device *lo; 2172 int id, ret; 2173 2174 ret = mutex_lock_killable(&loop_ctl_mutex); 2175 if (ret) 2176 return ret; 2177 idr_for_each_entry(&loop_index_idr, lo, id) { 2178 /* Hitting a race results in creating a new loop device which is harmless. */ 2179 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) 2180 goto found; 2181 } 2182 mutex_unlock(&loop_ctl_mutex); 2183 return loop_add(-1); 2184 found: 2185 mutex_unlock(&loop_ctl_mutex); 2186 return id; 2187 } 2188 2189 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2190 unsigned long parm) 2191 { 2192 switch (cmd) { 2193 case LOOP_CTL_ADD: 2194 return loop_add(parm); 2195 case LOOP_CTL_REMOVE: 2196 return loop_control_remove(parm); 2197 case LOOP_CTL_GET_FREE: 2198 return loop_control_get_free(parm); 2199 default: 2200 return -ENOSYS; 2201 } 2202 } 2203 2204 static const struct file_operations loop_ctl_fops = { 2205 .open = nonseekable_open, 2206 .unlocked_ioctl = loop_control_ioctl, 2207 .compat_ioctl = loop_control_ioctl, 2208 .owner = THIS_MODULE, 2209 .llseek = noop_llseek, 2210 }; 2211 2212 static struct miscdevice loop_misc = { 2213 .minor = LOOP_CTRL_MINOR, 2214 .name = "loop-control", 2215 .fops = &loop_ctl_fops, 2216 }; 2217 2218 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2219 MODULE_ALIAS("devname:loop-control"); 2220 2221 static int __init loop_init(void) 2222 { 2223 int i, nr; 2224 int err; 2225 2226 part_shift = 0; 2227 if (max_part > 0) { 2228 part_shift = fls(max_part); 2229 2230 /* 2231 * Adjust max_part according to part_shift as it is exported 2232 * to user space so that user can decide correct minor number 2233 * if [s]he want to create more devices. 2234 * 2235 * Note that -1 is required because partition 0 is reserved 2236 * for the whole disk. 2237 */ 2238 max_part = (1UL << part_shift) - 1; 2239 } 2240 2241 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2242 err = -EINVAL; 2243 goto err_out; 2244 } 2245 2246 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2247 err = -EINVAL; 2248 goto err_out; 2249 } 2250 2251 /* 2252 * If max_loop is specified, create that many devices upfront. 2253 * This also becomes a hard limit. If max_loop is not specified, 2254 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2255 * init time. Loop devices can be requested on-demand with the 2256 * /dev/loop-control interface, or be instantiated by accessing 2257 * a 'dead' device node. 2258 */ 2259 if (max_loop) 2260 nr = max_loop; 2261 else 2262 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2263 2264 err = misc_register(&loop_misc); 2265 if (err < 0) 2266 goto err_out; 2267 2268 2269 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2270 err = -EIO; 2271 goto misc_out; 2272 } 2273 2274 /* pre-create number of devices given by config or max_loop */ 2275 for (i = 0; i < nr; i++) 2276 loop_add(i); 2277 2278 printk(KERN_INFO "loop: module loaded\n"); 2279 return 0; 2280 2281 misc_out: 2282 misc_deregister(&loop_misc); 2283 err_out: 2284 return err; 2285 } 2286 2287 static void __exit loop_exit(void) 2288 { 2289 struct loop_device *lo; 2290 int id; 2291 2292 unregister_blkdev(LOOP_MAJOR, "loop"); 2293 misc_deregister(&loop_misc); 2294 2295 /* 2296 * There is no need to use loop_ctl_mutex here, for nobody else can 2297 * access loop_index_idr when this module is unloading (unless forced 2298 * module unloading is requested). If this is not a clean unloading, 2299 * we have no means to avoid kernel crash. 2300 */ 2301 idr_for_each_entry(&loop_index_idr, lo, id) 2302 loop_remove(lo); 2303 2304 idr_destroy(&loop_index_idr); 2305 } 2306 2307 module_init(loop_init); 2308 module_exit(loop_exit); 2309 2310 #ifndef MODULE 2311 static int __init max_loop_setup(char *str) 2312 { 2313 max_loop = simple_strtol(str, NULL, 0); 2314 return 1; 2315 } 2316 2317 __setup("max_loop=", max_loop_setup); 2318 #endif 2319