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/file.h> 57 #include <linux/stat.h> 58 #include <linux/errno.h> 59 #include <linux/major.h> 60 #include <linux/wait.h> 61 #include <linux/blkdev.h> 62 #include <linux/blkpg.h> 63 #include <linux/init.h> 64 #include <linux/swap.h> 65 #include <linux/slab.h> 66 #include <linux/compat.h> 67 #include <linux/suspend.h> 68 #include <linux/freezer.h> 69 #include <linux/mutex.h> 70 #include <linux/writeback.h> 71 #include <linux/completion.h> 72 #include <linux/highmem.h> 73 #include <linux/kthread.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 "loop.h" 80 81 #include <asm/uaccess.h> 82 83 static DEFINE_IDR(loop_index_idr); 84 static DEFINE_MUTEX(loop_index_mutex); 85 86 static int max_part; 87 static int part_shift; 88 89 static int transfer_xor(struct loop_device *lo, int cmd, 90 struct page *raw_page, unsigned raw_off, 91 struct page *loop_page, unsigned loop_off, 92 int size, sector_t real_block) 93 { 94 char *raw_buf = kmap_atomic(raw_page) + raw_off; 95 char *loop_buf = kmap_atomic(loop_page) + loop_off; 96 char *in, *out, *key; 97 int i, keysize; 98 99 if (cmd == READ) { 100 in = raw_buf; 101 out = loop_buf; 102 } else { 103 in = loop_buf; 104 out = raw_buf; 105 } 106 107 key = lo->lo_encrypt_key; 108 keysize = lo->lo_encrypt_key_size; 109 for (i = 0; i < size; i++) 110 *out++ = *in++ ^ key[(i & 511) % keysize]; 111 112 kunmap_atomic(loop_buf); 113 kunmap_atomic(raw_buf); 114 cond_resched(); 115 return 0; 116 } 117 118 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 119 { 120 if (unlikely(info->lo_encrypt_key_size <= 0)) 121 return -EINVAL; 122 return 0; 123 } 124 125 static struct loop_func_table none_funcs = { 126 .number = LO_CRYPT_NONE, 127 }; 128 129 static struct loop_func_table xor_funcs = { 130 .number = LO_CRYPT_XOR, 131 .transfer = transfer_xor, 132 .init = xor_init 133 }; 134 135 /* xfer_funcs[0] is special - its release function is never called */ 136 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 137 &none_funcs, 138 &xor_funcs 139 }; 140 141 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 142 { 143 loff_t loopsize; 144 145 /* Compute loopsize in bytes */ 146 loopsize = i_size_read(file->f_mapping->host); 147 if (offset > 0) 148 loopsize -= offset; 149 /* offset is beyond i_size, weird but possible */ 150 if (loopsize < 0) 151 return 0; 152 153 if (sizelimit > 0 && sizelimit < loopsize) 154 loopsize = sizelimit; 155 /* 156 * Unfortunately, if we want to do I/O on the device, 157 * the number of 512-byte sectors has to fit into a sector_t. 158 */ 159 return loopsize >> 9; 160 } 161 162 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 163 { 164 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 165 } 166 167 static int 168 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 169 { 170 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 171 sector_t x = (sector_t)size; 172 struct block_device *bdev = lo->lo_device; 173 174 if (unlikely((loff_t)x != size)) 175 return -EFBIG; 176 if (lo->lo_offset != offset) 177 lo->lo_offset = offset; 178 if (lo->lo_sizelimit != sizelimit) 179 lo->lo_sizelimit = sizelimit; 180 set_capacity(lo->lo_disk, x); 181 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 182 /* let user-space know about the new size */ 183 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 184 return 0; 185 } 186 187 static inline int 188 lo_do_transfer(struct loop_device *lo, int cmd, 189 struct page *rpage, unsigned roffs, 190 struct page *lpage, unsigned loffs, 191 int size, sector_t rblock) 192 { 193 int ret; 194 195 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 196 if (likely(!ret)) 197 return 0; 198 199 printk_ratelimited(KERN_ERR 200 "loop: Transfer error at byte offset %llu, length %i.\n", 201 (unsigned long long)rblock << 9, size); 202 return ret; 203 } 204 205 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 206 { 207 struct iov_iter i; 208 ssize_t bw; 209 210 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len); 211 212 file_start_write(file); 213 bw = vfs_iter_write(file, &i, ppos); 214 file_end_write(file); 215 216 if (likely(bw == bvec->bv_len)) 217 return 0; 218 219 printk_ratelimited(KERN_ERR 220 "loop: Write error at byte offset %llu, length %i.\n", 221 (unsigned long long)*ppos, bvec->bv_len); 222 if (bw >= 0) 223 bw = -EIO; 224 return bw; 225 } 226 227 static int lo_write_simple(struct loop_device *lo, struct request *rq, 228 loff_t pos) 229 { 230 struct bio_vec bvec; 231 struct req_iterator iter; 232 int ret = 0; 233 234 rq_for_each_segment(bvec, rq, iter) { 235 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 236 if (ret < 0) 237 break; 238 cond_resched(); 239 } 240 241 return ret; 242 } 243 244 /* 245 * This is the slow, transforming version that needs to double buffer the 246 * data as it cannot do the transformations in place without having direct 247 * access to the destination pages of the backing file. 248 */ 249 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 250 loff_t pos) 251 { 252 struct bio_vec bvec, b; 253 struct req_iterator iter; 254 struct page *page; 255 int ret = 0; 256 257 page = alloc_page(GFP_NOIO); 258 if (unlikely(!page)) 259 return -ENOMEM; 260 261 rq_for_each_segment(bvec, rq, iter) { 262 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 263 bvec.bv_offset, bvec.bv_len, pos >> 9); 264 if (unlikely(ret)) 265 break; 266 267 b.bv_page = page; 268 b.bv_offset = 0; 269 b.bv_len = bvec.bv_len; 270 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 271 if (ret < 0) 272 break; 273 } 274 275 __free_page(page); 276 return ret; 277 } 278 279 static int lo_read_simple(struct loop_device *lo, struct request *rq, 280 loff_t pos) 281 { 282 struct bio_vec bvec; 283 struct req_iterator iter; 284 struct iov_iter i; 285 ssize_t len; 286 287 rq_for_each_segment(bvec, rq, iter) { 288 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); 289 len = vfs_iter_read(lo->lo_backing_file, &i, &pos); 290 if (len < 0) 291 return len; 292 293 flush_dcache_page(bvec.bv_page); 294 295 if (len != bvec.bv_len) { 296 struct bio *bio; 297 298 __rq_for_each_bio(bio, rq) 299 zero_fill_bio(bio); 300 break; 301 } 302 cond_resched(); 303 } 304 305 return 0; 306 } 307 308 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 309 loff_t pos) 310 { 311 struct bio_vec bvec, b; 312 struct req_iterator iter; 313 struct iov_iter i; 314 struct page *page; 315 ssize_t len; 316 int ret = 0; 317 318 page = alloc_page(GFP_NOIO); 319 if (unlikely(!page)) 320 return -ENOMEM; 321 322 rq_for_each_segment(bvec, rq, iter) { 323 loff_t offset = pos; 324 325 b.bv_page = page; 326 b.bv_offset = 0; 327 b.bv_len = bvec.bv_len; 328 329 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); 330 len = vfs_iter_read(lo->lo_backing_file, &i, &pos); 331 if (len < 0) { 332 ret = len; 333 goto out_free_page; 334 } 335 336 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 337 bvec.bv_offset, len, offset >> 9); 338 if (ret) 339 goto out_free_page; 340 341 flush_dcache_page(bvec.bv_page); 342 343 if (len != bvec.bv_len) { 344 struct bio *bio; 345 346 __rq_for_each_bio(bio, rq) 347 zero_fill_bio(bio); 348 break; 349 } 350 } 351 352 ret = 0; 353 out_free_page: 354 __free_page(page); 355 return ret; 356 } 357 358 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 359 { 360 /* 361 * We use punch hole to reclaim the free space used by the 362 * image a.k.a. discard. However we do not support discard if 363 * encryption is enabled, because it may give an attacker 364 * useful information. 365 */ 366 struct file *file = lo->lo_backing_file; 367 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 368 int ret; 369 370 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 371 ret = -EOPNOTSUPP; 372 goto out; 373 } 374 375 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 376 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 377 ret = -EIO; 378 out: 379 return ret; 380 } 381 382 static int lo_req_flush(struct loop_device *lo, struct request *rq) 383 { 384 struct file *file = lo->lo_backing_file; 385 int ret = vfs_fsync(file, 0); 386 if (unlikely(ret && ret != -EINVAL)) 387 ret = -EIO; 388 389 return ret; 390 } 391 392 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 393 { 394 loff_t pos; 395 int ret; 396 397 pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 398 399 if (rq->cmd_flags & REQ_WRITE) { 400 if (rq->cmd_flags & REQ_FLUSH) 401 ret = lo_req_flush(lo, rq); 402 else if (rq->cmd_flags & REQ_DISCARD) 403 ret = lo_discard(lo, rq, pos); 404 else if (lo->transfer) 405 ret = lo_write_transfer(lo, rq, pos); 406 else 407 ret = lo_write_simple(lo, rq, pos); 408 409 } else { 410 if (lo->transfer) 411 ret = lo_read_transfer(lo, rq, pos); 412 else 413 ret = lo_read_simple(lo, rq, pos); 414 } 415 416 return ret; 417 } 418 419 struct switch_request { 420 struct file *file; 421 struct completion wait; 422 }; 423 424 /* 425 * Do the actual switch; called from the BIO completion routine 426 */ 427 static void do_loop_switch(struct loop_device *lo, struct switch_request *p) 428 { 429 struct file *file = p->file; 430 struct file *old_file = lo->lo_backing_file; 431 struct address_space *mapping; 432 433 /* if no new file, only flush of queued bios requested */ 434 if (!file) 435 return; 436 437 mapping = file->f_mapping; 438 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 439 lo->lo_backing_file = file; 440 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? 441 mapping->host->i_bdev->bd_block_size : PAGE_SIZE; 442 lo->old_gfp_mask = mapping_gfp_mask(mapping); 443 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 444 } 445 446 /* 447 * loop_switch performs the hard work of switching a backing store. 448 * First it needs to flush existing IO, it does this by sending a magic 449 * BIO down the pipe. The completion of this BIO does the actual switch. 450 */ 451 static int loop_switch(struct loop_device *lo, struct file *file) 452 { 453 struct switch_request w; 454 455 w.file = file; 456 457 /* freeze queue and wait for completion of scheduled requests */ 458 blk_mq_freeze_queue(lo->lo_queue); 459 460 /* do the switch action */ 461 do_loop_switch(lo, &w); 462 463 /* unfreeze */ 464 blk_mq_unfreeze_queue(lo->lo_queue); 465 466 return 0; 467 } 468 469 /* 470 * Helper to flush the IOs in loop, but keeping loop thread running 471 */ 472 static int loop_flush(struct loop_device *lo) 473 { 474 return loop_switch(lo, NULL); 475 } 476 477 /* 478 * loop_change_fd switched the backing store of a loopback device to 479 * a new file. This is useful for operating system installers to free up 480 * the original file and in High Availability environments to switch to 481 * an alternative location for the content in case of server meltdown. 482 * This can only work if the loop device is used read-only, and if the 483 * new backing store is the same size and type as the old backing store. 484 */ 485 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 486 unsigned int arg) 487 { 488 struct file *file, *old_file; 489 struct inode *inode; 490 int error; 491 492 error = -ENXIO; 493 if (lo->lo_state != Lo_bound) 494 goto out; 495 496 /* the loop device has to be read-only */ 497 error = -EINVAL; 498 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 499 goto out; 500 501 error = -EBADF; 502 file = fget(arg); 503 if (!file) 504 goto out; 505 506 inode = file->f_mapping->host; 507 old_file = lo->lo_backing_file; 508 509 error = -EINVAL; 510 511 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 512 goto out_putf; 513 514 /* size of the new backing store needs to be the same */ 515 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 516 goto out_putf; 517 518 /* and ... switch */ 519 error = loop_switch(lo, file); 520 if (error) 521 goto out_putf; 522 523 fput(old_file); 524 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 525 ioctl_by_bdev(bdev, BLKRRPART, 0); 526 return 0; 527 528 out_putf: 529 fput(file); 530 out: 531 return error; 532 } 533 534 static inline int is_loop_device(struct file *file) 535 { 536 struct inode *i = file->f_mapping->host; 537 538 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 539 } 540 541 /* loop sysfs attributes */ 542 543 static ssize_t loop_attr_show(struct device *dev, char *page, 544 ssize_t (*callback)(struct loop_device *, char *)) 545 { 546 struct gendisk *disk = dev_to_disk(dev); 547 struct loop_device *lo = disk->private_data; 548 549 return callback(lo, page); 550 } 551 552 #define LOOP_ATTR_RO(_name) \ 553 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 554 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 555 struct device_attribute *attr, char *b) \ 556 { \ 557 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 558 } \ 559 static struct device_attribute loop_attr_##_name = \ 560 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); 561 562 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 563 { 564 ssize_t ret; 565 char *p = NULL; 566 567 spin_lock_irq(&lo->lo_lock); 568 if (lo->lo_backing_file) 569 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1); 570 spin_unlock_irq(&lo->lo_lock); 571 572 if (IS_ERR_OR_NULL(p)) 573 ret = PTR_ERR(p); 574 else { 575 ret = strlen(p); 576 memmove(buf, p, ret); 577 buf[ret++] = '\n'; 578 buf[ret] = 0; 579 } 580 581 return ret; 582 } 583 584 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 585 { 586 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 587 } 588 589 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 590 { 591 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 592 } 593 594 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 595 { 596 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 597 598 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 599 } 600 601 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 602 { 603 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 604 605 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 606 } 607 608 LOOP_ATTR_RO(backing_file); 609 LOOP_ATTR_RO(offset); 610 LOOP_ATTR_RO(sizelimit); 611 LOOP_ATTR_RO(autoclear); 612 LOOP_ATTR_RO(partscan); 613 614 static struct attribute *loop_attrs[] = { 615 &loop_attr_backing_file.attr, 616 &loop_attr_offset.attr, 617 &loop_attr_sizelimit.attr, 618 &loop_attr_autoclear.attr, 619 &loop_attr_partscan.attr, 620 NULL, 621 }; 622 623 static struct attribute_group loop_attribute_group = { 624 .name = "loop", 625 .attrs= loop_attrs, 626 }; 627 628 static int loop_sysfs_init(struct loop_device *lo) 629 { 630 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 631 &loop_attribute_group); 632 } 633 634 static void loop_sysfs_exit(struct loop_device *lo) 635 { 636 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 637 &loop_attribute_group); 638 } 639 640 static void loop_config_discard(struct loop_device *lo) 641 { 642 struct file *file = lo->lo_backing_file; 643 struct inode *inode = file->f_mapping->host; 644 struct request_queue *q = lo->lo_queue; 645 646 /* 647 * We use punch hole to reclaim the free space used by the 648 * image a.k.a. discard. However we do not support discard if 649 * encryption is enabled, because it may give an attacker 650 * useful information. 651 */ 652 if ((!file->f_op->fallocate) || 653 lo->lo_encrypt_key_size) { 654 q->limits.discard_granularity = 0; 655 q->limits.discard_alignment = 0; 656 q->limits.max_discard_sectors = 0; 657 q->limits.discard_zeroes_data = 0; 658 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 659 return; 660 } 661 662 q->limits.discard_granularity = inode->i_sb->s_blocksize; 663 q->limits.discard_alignment = 0; 664 q->limits.max_discard_sectors = UINT_MAX >> 9; 665 q->limits.discard_zeroes_data = 1; 666 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 667 } 668 669 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 670 struct block_device *bdev, unsigned int arg) 671 { 672 struct file *file, *f; 673 struct inode *inode; 674 struct address_space *mapping; 675 unsigned lo_blocksize; 676 int lo_flags = 0; 677 int error; 678 loff_t size; 679 680 /* This is safe, since we have a reference from open(). */ 681 __module_get(THIS_MODULE); 682 683 error = -EBADF; 684 file = fget(arg); 685 if (!file) 686 goto out; 687 688 error = -EBUSY; 689 if (lo->lo_state != Lo_unbound) 690 goto out_putf; 691 692 /* Avoid recursion */ 693 f = file; 694 while (is_loop_device(f)) { 695 struct loop_device *l; 696 697 if (f->f_mapping->host->i_bdev == bdev) 698 goto out_putf; 699 700 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 701 if (l->lo_state == Lo_unbound) { 702 error = -EINVAL; 703 goto out_putf; 704 } 705 f = l->lo_backing_file; 706 } 707 708 mapping = file->f_mapping; 709 inode = mapping->host; 710 711 error = -EINVAL; 712 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 713 goto out_putf; 714 715 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 716 !file->f_op->write_iter) 717 lo_flags |= LO_FLAGS_READ_ONLY; 718 719 lo_blocksize = S_ISBLK(inode->i_mode) ? 720 inode->i_bdev->bd_block_size : PAGE_SIZE; 721 722 error = -EFBIG; 723 size = get_loop_size(lo, file); 724 if ((loff_t)(sector_t)size != size) 725 goto out_putf; 726 error = -ENOMEM; 727 lo->wq = alloc_workqueue("kloopd%d", 728 WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 0, 729 lo->lo_number); 730 if (!lo->wq) 731 goto out_putf; 732 733 error = 0; 734 735 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 736 737 lo->lo_blocksize = lo_blocksize; 738 lo->lo_device = bdev; 739 lo->lo_flags = lo_flags; 740 lo->lo_backing_file = file; 741 lo->transfer = NULL; 742 lo->ioctl = NULL; 743 lo->lo_sizelimit = 0; 744 lo->old_gfp_mask = mapping_gfp_mask(mapping); 745 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 746 747 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 748 blk_queue_flush(lo->lo_queue, REQ_FLUSH); 749 750 set_capacity(lo->lo_disk, size); 751 bd_set_size(bdev, size << 9); 752 loop_sysfs_init(lo); 753 /* let user-space know about the new size */ 754 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 755 756 set_blocksize(bdev, lo_blocksize); 757 758 lo->lo_state = Lo_bound; 759 if (part_shift) 760 lo->lo_flags |= LO_FLAGS_PARTSCAN; 761 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 762 ioctl_by_bdev(bdev, BLKRRPART, 0); 763 764 /* Grab the block_device to prevent its destruction after we 765 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 766 */ 767 bdgrab(bdev); 768 return 0; 769 770 out_putf: 771 fput(file); 772 out: 773 /* This is safe: open() is still holding a reference. */ 774 module_put(THIS_MODULE); 775 return error; 776 } 777 778 static int 779 loop_release_xfer(struct loop_device *lo) 780 { 781 int err = 0; 782 struct loop_func_table *xfer = lo->lo_encryption; 783 784 if (xfer) { 785 if (xfer->release) 786 err = xfer->release(lo); 787 lo->transfer = NULL; 788 lo->lo_encryption = NULL; 789 module_put(xfer->owner); 790 } 791 return err; 792 } 793 794 static int 795 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 796 const struct loop_info64 *i) 797 { 798 int err = 0; 799 800 if (xfer) { 801 struct module *owner = xfer->owner; 802 803 if (!try_module_get(owner)) 804 return -EINVAL; 805 if (xfer->init) 806 err = xfer->init(lo, i); 807 if (err) 808 module_put(owner); 809 else 810 lo->lo_encryption = xfer; 811 } 812 return err; 813 } 814 815 static int loop_clr_fd(struct loop_device *lo) 816 { 817 struct file *filp = lo->lo_backing_file; 818 gfp_t gfp = lo->old_gfp_mask; 819 struct block_device *bdev = lo->lo_device; 820 821 if (lo->lo_state != Lo_bound) 822 return -ENXIO; 823 824 /* 825 * If we've explicitly asked to tear down the loop device, 826 * and it has an elevated reference count, set it for auto-teardown when 827 * the last reference goes away. This stops $!~#$@ udev from 828 * preventing teardown because it decided that it needs to run blkid on 829 * the loopback device whenever they appear. xfstests is notorious for 830 * failing tests because blkid via udev races with a losetup 831 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 832 * command to fail with EBUSY. 833 */ 834 if (lo->lo_refcnt > 1) { 835 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 836 mutex_unlock(&lo->lo_ctl_mutex); 837 return 0; 838 } 839 840 if (filp == NULL) 841 return -EINVAL; 842 843 spin_lock_irq(&lo->lo_lock); 844 lo->lo_state = Lo_rundown; 845 lo->lo_backing_file = NULL; 846 spin_unlock_irq(&lo->lo_lock); 847 848 loop_release_xfer(lo); 849 lo->transfer = NULL; 850 lo->ioctl = NULL; 851 lo->lo_device = NULL; 852 lo->lo_encryption = NULL; 853 lo->lo_offset = 0; 854 lo->lo_sizelimit = 0; 855 lo->lo_encrypt_key_size = 0; 856 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 857 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 858 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 859 if (bdev) { 860 bdput(bdev); 861 invalidate_bdev(bdev); 862 } 863 set_capacity(lo->lo_disk, 0); 864 loop_sysfs_exit(lo); 865 if (bdev) { 866 bd_set_size(bdev, 0); 867 /* let user-space know about this change */ 868 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 869 } 870 mapping_set_gfp_mask(filp->f_mapping, gfp); 871 lo->lo_state = Lo_unbound; 872 /* This is safe: open() is still holding a reference. */ 873 module_put(THIS_MODULE); 874 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 875 ioctl_by_bdev(bdev, BLKRRPART, 0); 876 lo->lo_flags = 0; 877 if (!part_shift) 878 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 879 destroy_workqueue(lo->wq); 880 lo->wq = NULL; 881 mutex_unlock(&lo->lo_ctl_mutex); 882 /* 883 * Need not hold lo_ctl_mutex to fput backing file. 884 * Calling fput holding lo_ctl_mutex triggers a circular 885 * lock dependency possibility warning as fput can take 886 * bd_mutex which is usually taken before lo_ctl_mutex. 887 */ 888 fput(filp); 889 return 0; 890 } 891 892 static int 893 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 894 { 895 int err; 896 struct loop_func_table *xfer; 897 kuid_t uid = current_uid(); 898 899 if (lo->lo_encrypt_key_size && 900 !uid_eq(lo->lo_key_owner, uid) && 901 !capable(CAP_SYS_ADMIN)) 902 return -EPERM; 903 if (lo->lo_state != Lo_bound) 904 return -ENXIO; 905 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 906 return -EINVAL; 907 908 err = loop_release_xfer(lo); 909 if (err) 910 return err; 911 912 if (info->lo_encrypt_type) { 913 unsigned int type = info->lo_encrypt_type; 914 915 if (type >= MAX_LO_CRYPT) 916 return -EINVAL; 917 xfer = xfer_funcs[type]; 918 if (xfer == NULL) 919 return -EINVAL; 920 } else 921 xfer = NULL; 922 923 err = loop_init_xfer(lo, xfer, info); 924 if (err) 925 return err; 926 927 if (lo->lo_offset != info->lo_offset || 928 lo->lo_sizelimit != info->lo_sizelimit) 929 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) 930 return -EFBIG; 931 932 loop_config_discard(lo); 933 934 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 935 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 936 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 937 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 938 939 if (!xfer) 940 xfer = &none_funcs; 941 lo->transfer = xfer->transfer; 942 lo->ioctl = xfer->ioctl; 943 944 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 945 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 946 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 947 948 if ((info->lo_flags & LO_FLAGS_PARTSCAN) && 949 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 950 lo->lo_flags |= LO_FLAGS_PARTSCAN; 951 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 952 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0); 953 } 954 955 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 956 lo->lo_init[0] = info->lo_init[0]; 957 lo->lo_init[1] = info->lo_init[1]; 958 if (info->lo_encrypt_key_size) { 959 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 960 info->lo_encrypt_key_size); 961 lo->lo_key_owner = uid; 962 } 963 964 return 0; 965 } 966 967 static int 968 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 969 { 970 struct file *file = lo->lo_backing_file; 971 struct kstat stat; 972 int error; 973 974 if (lo->lo_state != Lo_bound) 975 return -ENXIO; 976 error = vfs_getattr(&file->f_path, &stat); 977 if (error) 978 return error; 979 memset(info, 0, sizeof(*info)); 980 info->lo_number = lo->lo_number; 981 info->lo_device = huge_encode_dev(stat.dev); 982 info->lo_inode = stat.ino; 983 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); 984 info->lo_offset = lo->lo_offset; 985 info->lo_sizelimit = lo->lo_sizelimit; 986 info->lo_flags = lo->lo_flags; 987 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 988 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 989 info->lo_encrypt_type = 990 lo->lo_encryption ? lo->lo_encryption->number : 0; 991 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 992 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 993 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 994 lo->lo_encrypt_key_size); 995 } 996 return 0; 997 } 998 999 static void 1000 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1001 { 1002 memset(info64, 0, sizeof(*info64)); 1003 info64->lo_number = info->lo_number; 1004 info64->lo_device = info->lo_device; 1005 info64->lo_inode = info->lo_inode; 1006 info64->lo_rdevice = info->lo_rdevice; 1007 info64->lo_offset = info->lo_offset; 1008 info64->lo_sizelimit = 0; 1009 info64->lo_encrypt_type = info->lo_encrypt_type; 1010 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1011 info64->lo_flags = info->lo_flags; 1012 info64->lo_init[0] = info->lo_init[0]; 1013 info64->lo_init[1] = info->lo_init[1]; 1014 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1015 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1016 else 1017 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1018 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1019 } 1020 1021 static int 1022 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1023 { 1024 memset(info, 0, sizeof(*info)); 1025 info->lo_number = info64->lo_number; 1026 info->lo_device = info64->lo_device; 1027 info->lo_inode = info64->lo_inode; 1028 info->lo_rdevice = info64->lo_rdevice; 1029 info->lo_offset = info64->lo_offset; 1030 info->lo_encrypt_type = info64->lo_encrypt_type; 1031 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1032 info->lo_flags = info64->lo_flags; 1033 info->lo_init[0] = info64->lo_init[0]; 1034 info->lo_init[1] = info64->lo_init[1]; 1035 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1036 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1037 else 1038 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1039 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1040 1041 /* error in case values were truncated */ 1042 if (info->lo_device != info64->lo_device || 1043 info->lo_rdevice != info64->lo_rdevice || 1044 info->lo_inode != info64->lo_inode || 1045 info->lo_offset != info64->lo_offset) 1046 return -EOVERFLOW; 1047 1048 return 0; 1049 } 1050 1051 static int 1052 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1053 { 1054 struct loop_info info; 1055 struct loop_info64 info64; 1056 1057 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1058 return -EFAULT; 1059 loop_info64_from_old(&info, &info64); 1060 return loop_set_status(lo, &info64); 1061 } 1062 1063 static int 1064 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1065 { 1066 struct loop_info64 info64; 1067 1068 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1069 return -EFAULT; 1070 return loop_set_status(lo, &info64); 1071 } 1072 1073 static int 1074 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1075 struct loop_info info; 1076 struct loop_info64 info64; 1077 int err = 0; 1078 1079 if (!arg) 1080 err = -EINVAL; 1081 if (!err) 1082 err = loop_get_status(lo, &info64); 1083 if (!err) 1084 err = loop_info64_to_old(&info64, &info); 1085 if (!err && copy_to_user(arg, &info, sizeof(info))) 1086 err = -EFAULT; 1087 1088 return err; 1089 } 1090 1091 static int 1092 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1093 struct loop_info64 info64; 1094 int err = 0; 1095 1096 if (!arg) 1097 err = -EINVAL; 1098 if (!err) 1099 err = loop_get_status(lo, &info64); 1100 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1101 err = -EFAULT; 1102 1103 return err; 1104 } 1105 1106 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) 1107 { 1108 if (unlikely(lo->lo_state != Lo_bound)) 1109 return -ENXIO; 1110 1111 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1112 } 1113 1114 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1115 unsigned int cmd, unsigned long arg) 1116 { 1117 struct loop_device *lo = bdev->bd_disk->private_data; 1118 int err; 1119 1120 mutex_lock_nested(&lo->lo_ctl_mutex, 1); 1121 switch (cmd) { 1122 case LOOP_SET_FD: 1123 err = loop_set_fd(lo, mode, bdev, arg); 1124 break; 1125 case LOOP_CHANGE_FD: 1126 err = loop_change_fd(lo, bdev, arg); 1127 break; 1128 case LOOP_CLR_FD: 1129 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1130 err = loop_clr_fd(lo); 1131 if (!err) 1132 goto out_unlocked; 1133 break; 1134 case LOOP_SET_STATUS: 1135 err = -EPERM; 1136 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1137 err = loop_set_status_old(lo, 1138 (struct loop_info __user *)arg); 1139 break; 1140 case LOOP_GET_STATUS: 1141 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1142 break; 1143 case LOOP_SET_STATUS64: 1144 err = -EPERM; 1145 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1146 err = loop_set_status64(lo, 1147 (struct loop_info64 __user *) arg); 1148 break; 1149 case LOOP_GET_STATUS64: 1150 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1151 break; 1152 case LOOP_SET_CAPACITY: 1153 err = -EPERM; 1154 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1155 err = loop_set_capacity(lo, bdev); 1156 break; 1157 default: 1158 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1159 } 1160 mutex_unlock(&lo->lo_ctl_mutex); 1161 1162 out_unlocked: 1163 return err; 1164 } 1165 1166 #ifdef CONFIG_COMPAT 1167 struct compat_loop_info { 1168 compat_int_t lo_number; /* ioctl r/o */ 1169 compat_dev_t lo_device; /* ioctl r/o */ 1170 compat_ulong_t lo_inode; /* ioctl r/o */ 1171 compat_dev_t lo_rdevice; /* ioctl r/o */ 1172 compat_int_t lo_offset; 1173 compat_int_t lo_encrypt_type; 1174 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1175 compat_int_t lo_flags; /* ioctl r/o */ 1176 char lo_name[LO_NAME_SIZE]; 1177 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1178 compat_ulong_t lo_init[2]; 1179 char reserved[4]; 1180 }; 1181 1182 /* 1183 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1184 * - noinlined to reduce stack space usage in main part of driver 1185 */ 1186 static noinline int 1187 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1188 struct loop_info64 *info64) 1189 { 1190 struct compat_loop_info info; 1191 1192 if (copy_from_user(&info, arg, sizeof(info))) 1193 return -EFAULT; 1194 1195 memset(info64, 0, sizeof(*info64)); 1196 info64->lo_number = info.lo_number; 1197 info64->lo_device = info.lo_device; 1198 info64->lo_inode = info.lo_inode; 1199 info64->lo_rdevice = info.lo_rdevice; 1200 info64->lo_offset = info.lo_offset; 1201 info64->lo_sizelimit = 0; 1202 info64->lo_encrypt_type = info.lo_encrypt_type; 1203 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1204 info64->lo_flags = info.lo_flags; 1205 info64->lo_init[0] = info.lo_init[0]; 1206 info64->lo_init[1] = info.lo_init[1]; 1207 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1208 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1209 else 1210 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1211 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1212 return 0; 1213 } 1214 1215 /* 1216 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1217 * - noinlined to reduce stack space usage in main part of driver 1218 */ 1219 static noinline int 1220 loop_info64_to_compat(const struct loop_info64 *info64, 1221 struct compat_loop_info __user *arg) 1222 { 1223 struct compat_loop_info info; 1224 1225 memset(&info, 0, sizeof(info)); 1226 info.lo_number = info64->lo_number; 1227 info.lo_device = info64->lo_device; 1228 info.lo_inode = info64->lo_inode; 1229 info.lo_rdevice = info64->lo_rdevice; 1230 info.lo_offset = info64->lo_offset; 1231 info.lo_encrypt_type = info64->lo_encrypt_type; 1232 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1233 info.lo_flags = info64->lo_flags; 1234 info.lo_init[0] = info64->lo_init[0]; 1235 info.lo_init[1] = info64->lo_init[1]; 1236 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1237 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1238 else 1239 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1240 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1241 1242 /* error in case values were truncated */ 1243 if (info.lo_device != info64->lo_device || 1244 info.lo_rdevice != info64->lo_rdevice || 1245 info.lo_inode != info64->lo_inode || 1246 info.lo_offset != info64->lo_offset || 1247 info.lo_init[0] != info64->lo_init[0] || 1248 info.lo_init[1] != info64->lo_init[1]) 1249 return -EOVERFLOW; 1250 1251 if (copy_to_user(arg, &info, sizeof(info))) 1252 return -EFAULT; 1253 return 0; 1254 } 1255 1256 static int 1257 loop_set_status_compat(struct loop_device *lo, 1258 const struct compat_loop_info __user *arg) 1259 { 1260 struct loop_info64 info64; 1261 int ret; 1262 1263 ret = loop_info64_from_compat(arg, &info64); 1264 if (ret < 0) 1265 return ret; 1266 return loop_set_status(lo, &info64); 1267 } 1268 1269 static int 1270 loop_get_status_compat(struct loop_device *lo, 1271 struct compat_loop_info __user *arg) 1272 { 1273 struct loop_info64 info64; 1274 int err = 0; 1275 1276 if (!arg) 1277 err = -EINVAL; 1278 if (!err) 1279 err = loop_get_status(lo, &info64); 1280 if (!err) 1281 err = loop_info64_to_compat(&info64, arg); 1282 return err; 1283 } 1284 1285 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1286 unsigned int cmd, unsigned long arg) 1287 { 1288 struct loop_device *lo = bdev->bd_disk->private_data; 1289 int err; 1290 1291 switch(cmd) { 1292 case LOOP_SET_STATUS: 1293 mutex_lock(&lo->lo_ctl_mutex); 1294 err = loop_set_status_compat( 1295 lo, (const struct compat_loop_info __user *) arg); 1296 mutex_unlock(&lo->lo_ctl_mutex); 1297 break; 1298 case LOOP_GET_STATUS: 1299 mutex_lock(&lo->lo_ctl_mutex); 1300 err = loop_get_status_compat( 1301 lo, (struct compat_loop_info __user *) arg); 1302 mutex_unlock(&lo->lo_ctl_mutex); 1303 break; 1304 case LOOP_SET_CAPACITY: 1305 case LOOP_CLR_FD: 1306 case LOOP_GET_STATUS64: 1307 case LOOP_SET_STATUS64: 1308 arg = (unsigned long) compat_ptr(arg); 1309 case LOOP_SET_FD: 1310 case LOOP_CHANGE_FD: 1311 err = lo_ioctl(bdev, mode, cmd, arg); 1312 break; 1313 default: 1314 err = -ENOIOCTLCMD; 1315 break; 1316 } 1317 return err; 1318 } 1319 #endif 1320 1321 static int lo_open(struct block_device *bdev, fmode_t mode) 1322 { 1323 struct loop_device *lo; 1324 int err = 0; 1325 1326 mutex_lock(&loop_index_mutex); 1327 lo = bdev->bd_disk->private_data; 1328 if (!lo) { 1329 err = -ENXIO; 1330 goto out; 1331 } 1332 1333 mutex_lock(&lo->lo_ctl_mutex); 1334 lo->lo_refcnt++; 1335 mutex_unlock(&lo->lo_ctl_mutex); 1336 out: 1337 mutex_unlock(&loop_index_mutex); 1338 return err; 1339 } 1340 1341 static void lo_release(struct gendisk *disk, fmode_t mode) 1342 { 1343 struct loop_device *lo = disk->private_data; 1344 int err; 1345 1346 mutex_lock(&lo->lo_ctl_mutex); 1347 1348 if (--lo->lo_refcnt) 1349 goto out; 1350 1351 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1352 /* 1353 * In autoclear mode, stop the loop thread 1354 * and remove configuration after last close. 1355 */ 1356 err = loop_clr_fd(lo); 1357 if (!err) 1358 return; 1359 } else { 1360 /* 1361 * Otherwise keep thread (if running) and config, 1362 * but flush possible ongoing bios in thread. 1363 */ 1364 loop_flush(lo); 1365 } 1366 1367 out: 1368 mutex_unlock(&lo->lo_ctl_mutex); 1369 } 1370 1371 static const struct block_device_operations lo_fops = { 1372 .owner = THIS_MODULE, 1373 .open = lo_open, 1374 .release = lo_release, 1375 .ioctl = lo_ioctl, 1376 #ifdef CONFIG_COMPAT 1377 .compat_ioctl = lo_compat_ioctl, 1378 #endif 1379 }; 1380 1381 /* 1382 * And now the modules code and kernel interface. 1383 */ 1384 static int max_loop; 1385 module_param(max_loop, int, S_IRUGO); 1386 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1387 module_param(max_part, int, S_IRUGO); 1388 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1389 MODULE_LICENSE("GPL"); 1390 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1391 1392 int loop_register_transfer(struct loop_func_table *funcs) 1393 { 1394 unsigned int n = funcs->number; 1395 1396 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1397 return -EINVAL; 1398 xfer_funcs[n] = funcs; 1399 return 0; 1400 } 1401 1402 static int unregister_transfer_cb(int id, void *ptr, void *data) 1403 { 1404 struct loop_device *lo = ptr; 1405 struct loop_func_table *xfer = data; 1406 1407 mutex_lock(&lo->lo_ctl_mutex); 1408 if (lo->lo_encryption == xfer) 1409 loop_release_xfer(lo); 1410 mutex_unlock(&lo->lo_ctl_mutex); 1411 return 0; 1412 } 1413 1414 int loop_unregister_transfer(int number) 1415 { 1416 unsigned int n = number; 1417 struct loop_func_table *xfer; 1418 1419 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1420 return -EINVAL; 1421 1422 xfer_funcs[n] = NULL; 1423 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1424 return 0; 1425 } 1426 1427 EXPORT_SYMBOL(loop_register_transfer); 1428 EXPORT_SYMBOL(loop_unregister_transfer); 1429 1430 static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1431 const struct blk_mq_queue_data *bd) 1432 { 1433 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); 1434 struct loop_device *lo = cmd->rq->q->queuedata; 1435 1436 blk_mq_start_request(bd->rq); 1437 1438 if (lo->lo_state != Lo_bound) 1439 return -EIO; 1440 1441 if (cmd->rq->cmd_flags & REQ_WRITE) { 1442 struct loop_device *lo = cmd->rq->q->queuedata; 1443 bool need_sched = true; 1444 1445 spin_lock_irq(&lo->lo_lock); 1446 if (lo->write_started) 1447 need_sched = false; 1448 else 1449 lo->write_started = true; 1450 list_add_tail(&cmd->list, &lo->write_cmd_head); 1451 spin_unlock_irq(&lo->lo_lock); 1452 1453 if (need_sched) 1454 queue_work(lo->wq, &lo->write_work); 1455 } else { 1456 queue_work(lo->wq, &cmd->read_work); 1457 } 1458 1459 return BLK_MQ_RQ_QUEUE_OK; 1460 } 1461 1462 static void loop_handle_cmd(struct loop_cmd *cmd) 1463 { 1464 const bool write = cmd->rq->cmd_flags & REQ_WRITE; 1465 struct loop_device *lo = cmd->rq->q->queuedata; 1466 int ret = -EIO; 1467 1468 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) 1469 goto failed; 1470 1471 ret = do_req_filebacked(lo, cmd->rq); 1472 1473 failed: 1474 if (ret) 1475 cmd->rq->errors = -EIO; 1476 blk_mq_complete_request(cmd->rq); 1477 } 1478 1479 static void loop_queue_write_work(struct work_struct *work) 1480 { 1481 struct loop_device *lo = 1482 container_of(work, struct loop_device, write_work); 1483 LIST_HEAD(cmd_list); 1484 1485 spin_lock_irq(&lo->lo_lock); 1486 repeat: 1487 list_splice_init(&lo->write_cmd_head, &cmd_list); 1488 spin_unlock_irq(&lo->lo_lock); 1489 1490 while (!list_empty(&cmd_list)) { 1491 struct loop_cmd *cmd = list_first_entry(&cmd_list, 1492 struct loop_cmd, list); 1493 list_del_init(&cmd->list); 1494 loop_handle_cmd(cmd); 1495 } 1496 1497 spin_lock_irq(&lo->lo_lock); 1498 if (!list_empty(&lo->write_cmd_head)) 1499 goto repeat; 1500 lo->write_started = false; 1501 spin_unlock_irq(&lo->lo_lock); 1502 } 1503 1504 static void loop_queue_read_work(struct work_struct *work) 1505 { 1506 struct loop_cmd *cmd = 1507 container_of(work, struct loop_cmd, read_work); 1508 1509 loop_handle_cmd(cmd); 1510 } 1511 1512 static int loop_init_request(void *data, struct request *rq, 1513 unsigned int hctx_idx, unsigned int request_idx, 1514 unsigned int numa_node) 1515 { 1516 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1517 1518 cmd->rq = rq; 1519 INIT_WORK(&cmd->read_work, loop_queue_read_work); 1520 1521 return 0; 1522 } 1523 1524 static struct blk_mq_ops loop_mq_ops = { 1525 .queue_rq = loop_queue_rq, 1526 .map_queue = blk_mq_map_queue, 1527 .init_request = loop_init_request, 1528 }; 1529 1530 static int loop_add(struct loop_device **l, int i) 1531 { 1532 struct loop_device *lo; 1533 struct gendisk *disk; 1534 int err; 1535 1536 err = -ENOMEM; 1537 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1538 if (!lo) 1539 goto out; 1540 1541 lo->lo_state = Lo_unbound; 1542 1543 /* allocate id, if @id >= 0, we're requesting that specific id */ 1544 if (i >= 0) { 1545 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1546 if (err == -ENOSPC) 1547 err = -EEXIST; 1548 } else { 1549 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1550 } 1551 if (err < 0) 1552 goto out_free_dev; 1553 i = err; 1554 1555 err = -ENOMEM; 1556 lo->tag_set.ops = &loop_mq_ops; 1557 lo->tag_set.nr_hw_queues = 1; 1558 lo->tag_set.queue_depth = 128; 1559 lo->tag_set.numa_node = NUMA_NO_NODE; 1560 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1561 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1562 lo->tag_set.driver_data = lo; 1563 1564 err = blk_mq_alloc_tag_set(&lo->tag_set); 1565 if (err) 1566 goto out_free_idr; 1567 1568 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1569 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1570 err = PTR_ERR(lo->lo_queue); 1571 goto out_cleanup_tags; 1572 } 1573 lo->lo_queue->queuedata = lo; 1574 1575 INIT_LIST_HEAD(&lo->write_cmd_head); 1576 INIT_WORK(&lo->write_work, loop_queue_write_work); 1577 1578 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1579 if (!disk) 1580 goto out_free_queue; 1581 1582 /* 1583 * Disable partition scanning by default. The in-kernel partition 1584 * scanning can be requested individually per-device during its 1585 * setup. Userspace can always add and remove partitions from all 1586 * devices. The needed partition minors are allocated from the 1587 * extended minor space, the main loop device numbers will continue 1588 * to match the loop minors, regardless of the number of partitions 1589 * used. 1590 * 1591 * If max_part is given, partition scanning is globally enabled for 1592 * all loop devices. The minors for the main loop devices will be 1593 * multiples of max_part. 1594 * 1595 * Note: Global-for-all-devices, set-only-at-init, read-only module 1596 * parameteters like 'max_loop' and 'max_part' make things needlessly 1597 * complicated, are too static, inflexible and may surprise 1598 * userspace tools. Parameters like this in general should be avoided. 1599 */ 1600 if (!part_shift) 1601 disk->flags |= GENHD_FL_NO_PART_SCAN; 1602 disk->flags |= GENHD_FL_EXT_DEVT; 1603 mutex_init(&lo->lo_ctl_mutex); 1604 lo->lo_number = i; 1605 spin_lock_init(&lo->lo_lock); 1606 disk->major = LOOP_MAJOR; 1607 disk->first_minor = i << part_shift; 1608 disk->fops = &lo_fops; 1609 disk->private_data = lo; 1610 disk->queue = lo->lo_queue; 1611 sprintf(disk->disk_name, "loop%d", i); 1612 add_disk(disk); 1613 *l = lo; 1614 return lo->lo_number; 1615 1616 out_free_queue: 1617 blk_cleanup_queue(lo->lo_queue); 1618 out_cleanup_tags: 1619 blk_mq_free_tag_set(&lo->tag_set); 1620 out_free_idr: 1621 idr_remove(&loop_index_idr, i); 1622 out_free_dev: 1623 kfree(lo); 1624 out: 1625 return err; 1626 } 1627 1628 static void loop_remove(struct loop_device *lo) 1629 { 1630 del_gendisk(lo->lo_disk); 1631 blk_cleanup_queue(lo->lo_queue); 1632 blk_mq_free_tag_set(&lo->tag_set); 1633 put_disk(lo->lo_disk); 1634 kfree(lo); 1635 } 1636 1637 static int find_free_cb(int id, void *ptr, void *data) 1638 { 1639 struct loop_device *lo = ptr; 1640 struct loop_device **l = data; 1641 1642 if (lo->lo_state == Lo_unbound) { 1643 *l = lo; 1644 return 1; 1645 } 1646 return 0; 1647 } 1648 1649 static int loop_lookup(struct loop_device **l, int i) 1650 { 1651 struct loop_device *lo; 1652 int ret = -ENODEV; 1653 1654 if (i < 0) { 1655 int err; 1656 1657 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1658 if (err == 1) { 1659 *l = lo; 1660 ret = lo->lo_number; 1661 } 1662 goto out; 1663 } 1664 1665 /* lookup and return a specific i */ 1666 lo = idr_find(&loop_index_idr, i); 1667 if (lo) { 1668 *l = lo; 1669 ret = lo->lo_number; 1670 } 1671 out: 1672 return ret; 1673 } 1674 1675 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1676 { 1677 struct loop_device *lo; 1678 struct kobject *kobj; 1679 int err; 1680 1681 mutex_lock(&loop_index_mutex); 1682 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1683 if (err < 0) 1684 err = loop_add(&lo, MINOR(dev) >> part_shift); 1685 if (err < 0) 1686 kobj = NULL; 1687 else 1688 kobj = get_disk(lo->lo_disk); 1689 mutex_unlock(&loop_index_mutex); 1690 1691 *part = 0; 1692 return kobj; 1693 } 1694 1695 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1696 unsigned long parm) 1697 { 1698 struct loop_device *lo; 1699 int ret = -ENOSYS; 1700 1701 mutex_lock(&loop_index_mutex); 1702 switch (cmd) { 1703 case LOOP_CTL_ADD: 1704 ret = loop_lookup(&lo, parm); 1705 if (ret >= 0) { 1706 ret = -EEXIST; 1707 break; 1708 } 1709 ret = loop_add(&lo, parm); 1710 break; 1711 case LOOP_CTL_REMOVE: 1712 ret = loop_lookup(&lo, parm); 1713 if (ret < 0) 1714 break; 1715 mutex_lock(&lo->lo_ctl_mutex); 1716 if (lo->lo_state != Lo_unbound) { 1717 ret = -EBUSY; 1718 mutex_unlock(&lo->lo_ctl_mutex); 1719 break; 1720 } 1721 if (lo->lo_refcnt > 0) { 1722 ret = -EBUSY; 1723 mutex_unlock(&lo->lo_ctl_mutex); 1724 break; 1725 } 1726 lo->lo_disk->private_data = NULL; 1727 mutex_unlock(&lo->lo_ctl_mutex); 1728 idr_remove(&loop_index_idr, lo->lo_number); 1729 loop_remove(lo); 1730 break; 1731 case LOOP_CTL_GET_FREE: 1732 ret = loop_lookup(&lo, -1); 1733 if (ret >= 0) 1734 break; 1735 ret = loop_add(&lo, -1); 1736 } 1737 mutex_unlock(&loop_index_mutex); 1738 1739 return ret; 1740 } 1741 1742 static const struct file_operations loop_ctl_fops = { 1743 .open = nonseekable_open, 1744 .unlocked_ioctl = loop_control_ioctl, 1745 .compat_ioctl = loop_control_ioctl, 1746 .owner = THIS_MODULE, 1747 .llseek = noop_llseek, 1748 }; 1749 1750 static struct miscdevice loop_misc = { 1751 .minor = LOOP_CTRL_MINOR, 1752 .name = "loop-control", 1753 .fops = &loop_ctl_fops, 1754 }; 1755 1756 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 1757 MODULE_ALIAS("devname:loop-control"); 1758 1759 static int __init loop_init(void) 1760 { 1761 int i, nr; 1762 unsigned long range; 1763 struct loop_device *lo; 1764 int err; 1765 1766 err = misc_register(&loop_misc); 1767 if (err < 0) 1768 return err; 1769 1770 part_shift = 0; 1771 if (max_part > 0) { 1772 part_shift = fls(max_part); 1773 1774 /* 1775 * Adjust max_part according to part_shift as it is exported 1776 * to user space so that user can decide correct minor number 1777 * if [s]he want to create more devices. 1778 * 1779 * Note that -1 is required because partition 0 is reserved 1780 * for the whole disk. 1781 */ 1782 max_part = (1UL << part_shift) - 1; 1783 } 1784 1785 if ((1UL << part_shift) > DISK_MAX_PARTS) { 1786 err = -EINVAL; 1787 goto misc_out; 1788 } 1789 1790 if (max_loop > 1UL << (MINORBITS - part_shift)) { 1791 err = -EINVAL; 1792 goto misc_out; 1793 } 1794 1795 /* 1796 * If max_loop is specified, create that many devices upfront. 1797 * This also becomes a hard limit. If max_loop is not specified, 1798 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 1799 * init time. Loop devices can be requested on-demand with the 1800 * /dev/loop-control interface, or be instantiated by accessing 1801 * a 'dead' device node. 1802 */ 1803 if (max_loop) { 1804 nr = max_loop; 1805 range = max_loop << part_shift; 1806 } else { 1807 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 1808 range = 1UL << MINORBITS; 1809 } 1810 1811 if (register_blkdev(LOOP_MAJOR, "loop")) { 1812 err = -EIO; 1813 goto misc_out; 1814 } 1815 1816 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 1817 THIS_MODULE, loop_probe, NULL, NULL); 1818 1819 /* pre-create number of devices given by config or max_loop */ 1820 mutex_lock(&loop_index_mutex); 1821 for (i = 0; i < nr; i++) 1822 loop_add(&lo, i); 1823 mutex_unlock(&loop_index_mutex); 1824 1825 printk(KERN_INFO "loop: module loaded\n"); 1826 return 0; 1827 1828 misc_out: 1829 misc_deregister(&loop_misc); 1830 return err; 1831 } 1832 1833 static int loop_exit_cb(int id, void *ptr, void *data) 1834 { 1835 struct loop_device *lo = ptr; 1836 1837 loop_remove(lo); 1838 return 0; 1839 } 1840 1841 static void __exit loop_exit(void) 1842 { 1843 unsigned long range; 1844 1845 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 1846 1847 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 1848 idr_destroy(&loop_index_idr); 1849 1850 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 1851 unregister_blkdev(LOOP_MAJOR, "loop"); 1852 1853 misc_deregister(&loop_misc); 1854 } 1855 1856 module_init(loop_init); 1857 module_exit(loop_exit); 1858 1859 #ifndef MODULE 1860 static int __init max_loop_setup(char *str) 1861 { 1862 max_loop = simple_strtol(str, NULL, 0); 1863 return 1; 1864 } 1865 1866 __setup("max_loop=", max_loop_setup); 1867 #endif 1868