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/blk-mq.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 struct workqueue_struct *loop_wq; 90 91 /* 92 * Transfer functions 93 */ 94 static int transfer_none(struct loop_device *lo, int cmd, 95 struct page *raw_page, unsigned raw_off, 96 struct page *loop_page, unsigned loop_off, 97 int size, sector_t real_block) 98 { 99 char *raw_buf = kmap_atomic(raw_page) + raw_off; 100 char *loop_buf = kmap_atomic(loop_page) + loop_off; 101 102 if (cmd == READ) 103 memcpy(loop_buf, raw_buf, size); 104 else 105 memcpy(raw_buf, loop_buf, size); 106 107 kunmap_atomic(loop_buf); 108 kunmap_atomic(raw_buf); 109 cond_resched(); 110 return 0; 111 } 112 113 static int transfer_xor(struct loop_device *lo, int cmd, 114 struct page *raw_page, unsigned raw_off, 115 struct page *loop_page, unsigned loop_off, 116 int size, sector_t real_block) 117 { 118 char *raw_buf = kmap_atomic(raw_page) + raw_off; 119 char *loop_buf = kmap_atomic(loop_page) + loop_off; 120 char *in, *out, *key; 121 int i, keysize; 122 123 if (cmd == READ) { 124 in = raw_buf; 125 out = loop_buf; 126 } else { 127 in = loop_buf; 128 out = raw_buf; 129 } 130 131 key = lo->lo_encrypt_key; 132 keysize = lo->lo_encrypt_key_size; 133 for (i = 0; i < size; i++) 134 *out++ = *in++ ^ key[(i & 511) % keysize]; 135 136 kunmap_atomic(loop_buf); 137 kunmap_atomic(raw_buf); 138 cond_resched(); 139 return 0; 140 } 141 142 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 143 { 144 if (unlikely(info->lo_encrypt_key_size <= 0)) 145 return -EINVAL; 146 return 0; 147 } 148 149 static struct loop_func_table none_funcs = { 150 .number = LO_CRYPT_NONE, 151 .transfer = transfer_none, 152 }; 153 154 static struct loop_func_table xor_funcs = { 155 .number = LO_CRYPT_XOR, 156 .transfer = transfer_xor, 157 .init = xor_init 158 }; 159 160 /* xfer_funcs[0] is special - its release function is never called */ 161 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 162 &none_funcs, 163 &xor_funcs 164 }; 165 166 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 167 { 168 loff_t loopsize; 169 170 /* Compute loopsize in bytes */ 171 loopsize = i_size_read(file->f_mapping->host); 172 if (offset > 0) 173 loopsize -= offset; 174 /* offset is beyond i_size, weird but possible */ 175 if (loopsize < 0) 176 return 0; 177 178 if (sizelimit > 0 && sizelimit < loopsize) 179 loopsize = sizelimit; 180 /* 181 * Unfortunately, if we want to do I/O on the device, 182 * the number of 512-byte sectors has to fit into a sector_t. 183 */ 184 return loopsize >> 9; 185 } 186 187 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 188 { 189 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 190 } 191 192 static int 193 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 194 { 195 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 196 sector_t x = (sector_t)size; 197 struct block_device *bdev = lo->lo_device; 198 199 if (unlikely((loff_t)x != size)) 200 return -EFBIG; 201 if (lo->lo_offset != offset) 202 lo->lo_offset = offset; 203 if (lo->lo_sizelimit != sizelimit) 204 lo->lo_sizelimit = sizelimit; 205 set_capacity(lo->lo_disk, x); 206 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 207 /* let user-space know about the new size */ 208 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 209 return 0; 210 } 211 212 static inline int 213 lo_do_transfer(struct loop_device *lo, int cmd, 214 struct page *rpage, unsigned roffs, 215 struct page *lpage, unsigned loffs, 216 int size, sector_t rblock) 217 { 218 if (unlikely(!lo->transfer)) 219 return 0; 220 221 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 222 } 223 224 /** 225 * __do_lo_send_write - helper for writing data to a loop device 226 * 227 * This helper just factors out common code between do_lo_send_direct_write() 228 * and do_lo_send_write(). 229 */ 230 static int __do_lo_send_write(struct file *file, 231 u8 *buf, const int len, loff_t pos) 232 { 233 ssize_t bw; 234 mm_segment_t old_fs = get_fs(); 235 236 file_start_write(file); 237 set_fs(get_ds()); 238 bw = file->f_op->write(file, buf, len, &pos); 239 set_fs(old_fs); 240 file_end_write(file); 241 if (likely(bw == len)) 242 return 0; 243 printk_ratelimited(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n", 244 (unsigned long long)pos, len); 245 if (bw >= 0) 246 bw = -EIO; 247 return bw; 248 } 249 250 /** 251 * do_lo_send_direct_write - helper for writing data to a loop device 252 * 253 * This is the fast, non-transforming version that does not need double 254 * buffering. 255 */ 256 static int do_lo_send_direct_write(struct loop_device *lo, 257 struct bio_vec *bvec, loff_t pos, struct page *page) 258 { 259 ssize_t bw = __do_lo_send_write(lo->lo_backing_file, 260 kmap(bvec->bv_page) + bvec->bv_offset, 261 bvec->bv_len, pos); 262 kunmap(bvec->bv_page); 263 cond_resched(); 264 return bw; 265 } 266 267 /** 268 * do_lo_send_write - helper for writing data to a loop device 269 * 270 * This is the slow, transforming version that needs to double buffer the 271 * data as it cannot do the transformations in place without having direct 272 * access to the destination pages of the backing file. 273 */ 274 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec, 275 loff_t pos, struct page *page) 276 { 277 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page, 278 bvec->bv_offset, bvec->bv_len, pos >> 9); 279 if (likely(!ret)) 280 return __do_lo_send_write(lo->lo_backing_file, 281 page_address(page), bvec->bv_len, 282 pos); 283 printk_ratelimited(KERN_ERR "loop: Transfer error at byte offset %llu, " 284 "length %i.\n", (unsigned long long)pos, bvec->bv_len); 285 if (ret > 0) 286 ret = -EIO; 287 return ret; 288 } 289 290 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos) 291 { 292 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t, 293 struct page *page); 294 struct bio_vec bvec; 295 struct bvec_iter iter; 296 struct page *page = NULL; 297 int ret = 0; 298 299 if (lo->transfer != transfer_none) { 300 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); 301 if (unlikely(!page)) 302 goto fail; 303 kmap(page); 304 do_lo_send = do_lo_send_write; 305 } else { 306 do_lo_send = do_lo_send_direct_write; 307 } 308 309 bio_for_each_segment(bvec, bio, iter) { 310 ret = do_lo_send(lo, &bvec, pos, page); 311 if (ret < 0) 312 break; 313 pos += bvec.bv_len; 314 } 315 if (page) { 316 kunmap(page); 317 __free_page(page); 318 } 319 out: 320 return ret; 321 fail: 322 printk_ratelimited(KERN_ERR "loop: Failed to allocate temporary page for write.\n"); 323 ret = -ENOMEM; 324 goto out; 325 } 326 327 struct lo_read_data { 328 struct loop_device *lo; 329 struct page *page; 330 unsigned offset; 331 int bsize; 332 }; 333 334 static int 335 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 336 struct splice_desc *sd) 337 { 338 struct lo_read_data *p = sd->u.data; 339 struct loop_device *lo = p->lo; 340 struct page *page = buf->page; 341 sector_t IV; 342 int size; 343 344 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) + 345 (buf->offset >> 9); 346 size = sd->len; 347 if (size > p->bsize) 348 size = p->bsize; 349 350 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) { 351 printk_ratelimited(KERN_ERR "loop: transfer error block %ld\n", 352 page->index); 353 size = -EINVAL; 354 } 355 356 flush_dcache_page(p->page); 357 358 if (size > 0) 359 p->offset += size; 360 361 return size; 362 } 363 364 static int 365 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd) 366 { 367 return __splice_from_pipe(pipe, sd, lo_splice_actor); 368 } 369 370 static ssize_t 371 do_lo_receive(struct loop_device *lo, 372 struct bio_vec *bvec, int bsize, loff_t pos) 373 { 374 struct lo_read_data cookie; 375 struct splice_desc sd; 376 struct file *file; 377 ssize_t retval; 378 379 cookie.lo = lo; 380 cookie.page = bvec->bv_page; 381 cookie.offset = bvec->bv_offset; 382 cookie.bsize = bsize; 383 384 sd.len = 0; 385 sd.total_len = bvec->bv_len; 386 sd.flags = 0; 387 sd.pos = pos; 388 sd.u.data = &cookie; 389 390 file = lo->lo_backing_file; 391 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor); 392 393 return retval; 394 } 395 396 static int 397 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos) 398 { 399 struct bio_vec bvec; 400 struct bvec_iter iter; 401 ssize_t s; 402 403 bio_for_each_segment(bvec, bio, iter) { 404 s = do_lo_receive(lo, &bvec, bsize, pos); 405 if (s < 0) 406 return s; 407 408 if (s != bvec.bv_len) { 409 zero_fill_bio(bio); 410 break; 411 } 412 pos += bvec.bv_len; 413 } 414 return 0; 415 } 416 417 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio) 418 { 419 loff_t pos; 420 int ret; 421 422 pos = ((loff_t) bio->bi_iter.bi_sector << 9) + lo->lo_offset; 423 424 if (bio_rw(bio) == WRITE) { 425 struct file *file = lo->lo_backing_file; 426 427 if (bio->bi_rw & REQ_FLUSH) { 428 ret = vfs_fsync(file, 0); 429 if (unlikely(ret && ret != -EINVAL)) { 430 ret = -EIO; 431 goto out; 432 } 433 } 434 435 /* 436 * We use punch hole to reclaim the free space used by the 437 * image a.k.a. discard. However we do not support discard if 438 * encryption is enabled, because it may give an attacker 439 * useful information. 440 */ 441 if (bio->bi_rw & REQ_DISCARD) { 442 struct file *file = lo->lo_backing_file; 443 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 444 445 if ((!file->f_op->fallocate) || 446 lo->lo_encrypt_key_size) { 447 ret = -EOPNOTSUPP; 448 goto out; 449 } 450 ret = file->f_op->fallocate(file, mode, pos, 451 bio->bi_iter.bi_size); 452 if (unlikely(ret && ret != -EINVAL && 453 ret != -EOPNOTSUPP)) 454 ret = -EIO; 455 goto out; 456 } 457 458 ret = lo_send(lo, bio, pos); 459 460 if ((bio->bi_rw & REQ_FUA) && !ret) { 461 ret = vfs_fsync(file, 0); 462 if (unlikely(ret && ret != -EINVAL)) 463 ret = -EIO; 464 } 465 } else 466 ret = lo_receive(lo, bio, lo->lo_blocksize, pos); 467 468 out: 469 return ret; 470 } 471 472 struct switch_request { 473 struct file *file; 474 struct completion wait; 475 }; 476 477 static inline int loop_handle_bio(struct loop_device *lo, struct bio *bio) 478 { 479 return do_bio_filebacked(lo, bio); 480 } 481 482 /* 483 * Do the actual switch; called from the BIO completion routine 484 */ 485 static void do_loop_switch(struct loop_device *lo, struct switch_request *p) 486 { 487 struct file *file = p->file; 488 struct file *old_file = lo->lo_backing_file; 489 struct address_space *mapping; 490 491 /* if no new file, only flush of queued bios requested */ 492 if (!file) 493 return; 494 495 mapping = file->f_mapping; 496 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 497 lo->lo_backing_file = file; 498 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? 499 mapping->host->i_bdev->bd_block_size : PAGE_SIZE; 500 lo->old_gfp_mask = mapping_gfp_mask(mapping); 501 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 502 } 503 504 /* 505 * loop_switch performs the hard work of switching a backing store. 506 * First it needs to flush existing IO, it does this by sending a magic 507 * BIO down the pipe. The completion of this BIO does the actual switch. 508 */ 509 static int loop_switch(struct loop_device *lo, struct file *file) 510 { 511 struct switch_request w; 512 513 w.file = file; 514 515 /* freeze queue and wait for completion of scheduled requests */ 516 blk_mq_freeze_queue(lo->lo_queue); 517 518 /* do the switch action */ 519 do_loop_switch(lo, &w); 520 521 /* unfreeze */ 522 blk_mq_unfreeze_queue(lo->lo_queue); 523 524 return 0; 525 } 526 527 /* 528 * Helper to flush the IOs in loop, but keeping loop thread running 529 */ 530 static int loop_flush(struct loop_device *lo) 531 { 532 return loop_switch(lo, NULL); 533 } 534 535 /* 536 * loop_change_fd switched the backing store of a loopback device to 537 * a new file. This is useful for operating system installers to free up 538 * the original file and in High Availability environments to switch to 539 * an alternative location for the content in case of server meltdown. 540 * This can only work if the loop device is used read-only, and if the 541 * new backing store is the same size and type as the old backing store. 542 */ 543 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 544 unsigned int arg) 545 { 546 struct file *file, *old_file; 547 struct inode *inode; 548 int error; 549 550 error = -ENXIO; 551 if (lo->lo_state != Lo_bound) 552 goto out; 553 554 /* the loop device has to be read-only */ 555 error = -EINVAL; 556 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 557 goto out; 558 559 error = -EBADF; 560 file = fget(arg); 561 if (!file) 562 goto out; 563 564 inode = file->f_mapping->host; 565 old_file = lo->lo_backing_file; 566 567 error = -EINVAL; 568 569 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 570 goto out_putf; 571 572 /* size of the new backing store needs to be the same */ 573 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 574 goto out_putf; 575 576 /* and ... switch */ 577 error = loop_switch(lo, file); 578 if (error) 579 goto out_putf; 580 581 fput(old_file); 582 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 583 ioctl_by_bdev(bdev, BLKRRPART, 0); 584 return 0; 585 586 out_putf: 587 fput(file); 588 out: 589 return error; 590 } 591 592 static inline int is_loop_device(struct file *file) 593 { 594 struct inode *i = file->f_mapping->host; 595 596 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 597 } 598 599 /* loop sysfs attributes */ 600 601 static ssize_t loop_attr_show(struct device *dev, char *page, 602 ssize_t (*callback)(struct loop_device *, char *)) 603 { 604 struct gendisk *disk = dev_to_disk(dev); 605 struct loop_device *lo = disk->private_data; 606 607 return callback(lo, page); 608 } 609 610 #define LOOP_ATTR_RO(_name) \ 611 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 612 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 613 struct device_attribute *attr, char *b) \ 614 { \ 615 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 616 } \ 617 static struct device_attribute loop_attr_##_name = \ 618 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); 619 620 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 621 { 622 ssize_t ret; 623 char *p = NULL; 624 625 spin_lock_irq(&lo->lo_lock); 626 if (lo->lo_backing_file) 627 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1); 628 spin_unlock_irq(&lo->lo_lock); 629 630 if (IS_ERR_OR_NULL(p)) 631 ret = PTR_ERR(p); 632 else { 633 ret = strlen(p); 634 memmove(buf, p, ret); 635 buf[ret++] = '\n'; 636 buf[ret] = 0; 637 } 638 639 return ret; 640 } 641 642 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 643 { 644 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 645 } 646 647 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 648 { 649 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 650 } 651 652 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 653 { 654 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 655 656 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 657 } 658 659 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 660 { 661 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 662 663 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 664 } 665 666 LOOP_ATTR_RO(backing_file); 667 LOOP_ATTR_RO(offset); 668 LOOP_ATTR_RO(sizelimit); 669 LOOP_ATTR_RO(autoclear); 670 LOOP_ATTR_RO(partscan); 671 672 static struct attribute *loop_attrs[] = { 673 &loop_attr_backing_file.attr, 674 &loop_attr_offset.attr, 675 &loop_attr_sizelimit.attr, 676 &loop_attr_autoclear.attr, 677 &loop_attr_partscan.attr, 678 NULL, 679 }; 680 681 static struct attribute_group loop_attribute_group = { 682 .name = "loop", 683 .attrs= loop_attrs, 684 }; 685 686 static int loop_sysfs_init(struct loop_device *lo) 687 { 688 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 689 &loop_attribute_group); 690 } 691 692 static void loop_sysfs_exit(struct loop_device *lo) 693 { 694 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 695 &loop_attribute_group); 696 } 697 698 static void loop_config_discard(struct loop_device *lo) 699 { 700 struct file *file = lo->lo_backing_file; 701 struct inode *inode = file->f_mapping->host; 702 struct request_queue *q = lo->lo_queue; 703 704 /* 705 * We use punch hole to reclaim the free space used by the 706 * image a.k.a. discard. However we do not support discard if 707 * encryption is enabled, because it may give an attacker 708 * useful information. 709 */ 710 if ((!file->f_op->fallocate) || 711 lo->lo_encrypt_key_size) { 712 q->limits.discard_granularity = 0; 713 q->limits.discard_alignment = 0; 714 q->limits.max_discard_sectors = 0; 715 q->limits.discard_zeroes_data = 0; 716 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 717 return; 718 } 719 720 q->limits.discard_granularity = inode->i_sb->s_blocksize; 721 q->limits.discard_alignment = 0; 722 q->limits.max_discard_sectors = UINT_MAX >> 9; 723 q->limits.discard_zeroes_data = 1; 724 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 725 } 726 727 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 728 struct block_device *bdev, unsigned int arg) 729 { 730 struct file *file, *f; 731 struct inode *inode; 732 struct address_space *mapping; 733 unsigned lo_blocksize; 734 int lo_flags = 0; 735 int error; 736 loff_t size; 737 738 /* This is safe, since we have a reference from open(). */ 739 __module_get(THIS_MODULE); 740 741 error = -EBADF; 742 file = fget(arg); 743 if (!file) 744 goto out; 745 746 error = -EBUSY; 747 if (lo->lo_state != Lo_unbound) 748 goto out_putf; 749 750 /* Avoid recursion */ 751 f = file; 752 while (is_loop_device(f)) { 753 struct loop_device *l; 754 755 if (f->f_mapping->host->i_bdev == bdev) 756 goto out_putf; 757 758 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 759 if (l->lo_state == Lo_unbound) { 760 error = -EINVAL; 761 goto out_putf; 762 } 763 f = l->lo_backing_file; 764 } 765 766 mapping = file->f_mapping; 767 inode = mapping->host; 768 769 error = -EINVAL; 770 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 771 goto out_putf; 772 773 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 774 !file->f_op->write) 775 lo_flags |= LO_FLAGS_READ_ONLY; 776 777 lo_blocksize = S_ISBLK(inode->i_mode) ? 778 inode->i_bdev->bd_block_size : PAGE_SIZE; 779 780 error = -EFBIG; 781 size = get_loop_size(lo, file); 782 if ((loff_t)(sector_t)size != size) 783 goto out_putf; 784 785 error = 0; 786 787 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 788 789 lo->lo_blocksize = lo_blocksize; 790 lo->lo_device = bdev; 791 lo->lo_flags = lo_flags; 792 lo->lo_backing_file = file; 793 lo->transfer = transfer_none; 794 lo->ioctl = NULL; 795 lo->lo_sizelimit = 0; 796 lo->old_gfp_mask = mapping_gfp_mask(mapping); 797 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 798 799 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 800 blk_queue_flush(lo->lo_queue, REQ_FLUSH); 801 802 set_capacity(lo->lo_disk, size); 803 bd_set_size(bdev, size << 9); 804 loop_sysfs_init(lo); 805 /* let user-space know about the new size */ 806 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 807 808 set_blocksize(bdev, lo_blocksize); 809 810 lo->lo_state = Lo_bound; 811 if (part_shift) 812 lo->lo_flags |= LO_FLAGS_PARTSCAN; 813 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 814 ioctl_by_bdev(bdev, BLKRRPART, 0); 815 816 /* Grab the block_device to prevent its destruction after we 817 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 818 */ 819 bdgrab(bdev); 820 return 0; 821 822 out_putf: 823 fput(file); 824 out: 825 /* This is safe: open() is still holding a reference. */ 826 module_put(THIS_MODULE); 827 return error; 828 } 829 830 static int 831 loop_release_xfer(struct loop_device *lo) 832 { 833 int err = 0; 834 struct loop_func_table *xfer = lo->lo_encryption; 835 836 if (xfer) { 837 if (xfer->release) 838 err = xfer->release(lo); 839 lo->transfer = NULL; 840 lo->lo_encryption = NULL; 841 module_put(xfer->owner); 842 } 843 return err; 844 } 845 846 static int 847 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 848 const struct loop_info64 *i) 849 { 850 int err = 0; 851 852 if (xfer) { 853 struct module *owner = xfer->owner; 854 855 if (!try_module_get(owner)) 856 return -EINVAL; 857 if (xfer->init) 858 err = xfer->init(lo, i); 859 if (err) 860 module_put(owner); 861 else 862 lo->lo_encryption = xfer; 863 } 864 return err; 865 } 866 867 static int loop_clr_fd(struct loop_device *lo) 868 { 869 struct file *filp = lo->lo_backing_file; 870 gfp_t gfp = lo->old_gfp_mask; 871 struct block_device *bdev = lo->lo_device; 872 873 if (lo->lo_state != Lo_bound) 874 return -ENXIO; 875 876 /* 877 * If we've explicitly asked to tear down the loop device, 878 * and it has an elevated reference count, set it for auto-teardown when 879 * the last reference goes away. This stops $!~#$@ udev from 880 * preventing teardown because it decided that it needs to run blkid on 881 * the loopback device whenever they appear. xfstests is notorious for 882 * failing tests because blkid via udev races with a losetup 883 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 884 * command to fail with EBUSY. 885 */ 886 if (lo->lo_refcnt > 1) { 887 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 888 mutex_unlock(&lo->lo_ctl_mutex); 889 return 0; 890 } 891 892 if (filp == NULL) 893 return -EINVAL; 894 895 spin_lock_irq(&lo->lo_lock); 896 lo->lo_state = Lo_rundown; 897 lo->lo_backing_file = NULL; 898 spin_unlock_irq(&lo->lo_lock); 899 900 loop_release_xfer(lo); 901 lo->transfer = NULL; 902 lo->ioctl = NULL; 903 lo->lo_device = NULL; 904 lo->lo_encryption = NULL; 905 lo->lo_offset = 0; 906 lo->lo_sizelimit = 0; 907 lo->lo_encrypt_key_size = 0; 908 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 909 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 910 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 911 if (bdev) { 912 bdput(bdev); 913 invalidate_bdev(bdev); 914 } 915 set_capacity(lo->lo_disk, 0); 916 loop_sysfs_exit(lo); 917 if (bdev) { 918 bd_set_size(bdev, 0); 919 /* let user-space know about this change */ 920 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 921 } 922 mapping_set_gfp_mask(filp->f_mapping, gfp); 923 lo->lo_state = Lo_unbound; 924 /* This is safe: open() is still holding a reference. */ 925 module_put(THIS_MODULE); 926 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 927 ioctl_by_bdev(bdev, BLKRRPART, 0); 928 lo->lo_flags = 0; 929 if (!part_shift) 930 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 931 mutex_unlock(&lo->lo_ctl_mutex); 932 /* 933 * Need not hold lo_ctl_mutex to fput backing file. 934 * Calling fput holding lo_ctl_mutex triggers a circular 935 * lock dependency possibility warning as fput can take 936 * bd_mutex which is usually taken before lo_ctl_mutex. 937 */ 938 fput(filp); 939 return 0; 940 } 941 942 static int 943 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 944 { 945 int err; 946 struct loop_func_table *xfer; 947 kuid_t uid = current_uid(); 948 949 if (lo->lo_encrypt_key_size && 950 !uid_eq(lo->lo_key_owner, uid) && 951 !capable(CAP_SYS_ADMIN)) 952 return -EPERM; 953 if (lo->lo_state != Lo_bound) 954 return -ENXIO; 955 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 956 return -EINVAL; 957 958 err = loop_release_xfer(lo); 959 if (err) 960 return err; 961 962 if (info->lo_encrypt_type) { 963 unsigned int type = info->lo_encrypt_type; 964 965 if (type >= MAX_LO_CRYPT) 966 return -EINVAL; 967 xfer = xfer_funcs[type]; 968 if (xfer == NULL) 969 return -EINVAL; 970 } else 971 xfer = NULL; 972 973 err = loop_init_xfer(lo, xfer, info); 974 if (err) 975 return err; 976 977 if (lo->lo_offset != info->lo_offset || 978 lo->lo_sizelimit != info->lo_sizelimit) 979 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) 980 return -EFBIG; 981 982 loop_config_discard(lo); 983 984 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 985 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 986 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 987 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 988 989 if (!xfer) 990 xfer = &none_funcs; 991 lo->transfer = xfer->transfer; 992 lo->ioctl = xfer->ioctl; 993 994 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 995 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 996 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 997 998 if ((info->lo_flags & LO_FLAGS_PARTSCAN) && 999 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1000 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1001 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1002 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0); 1003 } 1004 1005 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1006 lo->lo_init[0] = info->lo_init[0]; 1007 lo->lo_init[1] = info->lo_init[1]; 1008 if (info->lo_encrypt_key_size) { 1009 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1010 info->lo_encrypt_key_size); 1011 lo->lo_key_owner = uid; 1012 } 1013 1014 return 0; 1015 } 1016 1017 static int 1018 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1019 { 1020 struct file *file = lo->lo_backing_file; 1021 struct kstat stat; 1022 int error; 1023 1024 if (lo->lo_state != Lo_bound) 1025 return -ENXIO; 1026 error = vfs_getattr(&file->f_path, &stat); 1027 if (error) 1028 return error; 1029 memset(info, 0, sizeof(*info)); 1030 info->lo_number = lo->lo_number; 1031 info->lo_device = huge_encode_dev(stat.dev); 1032 info->lo_inode = stat.ino; 1033 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); 1034 info->lo_offset = lo->lo_offset; 1035 info->lo_sizelimit = lo->lo_sizelimit; 1036 info->lo_flags = lo->lo_flags; 1037 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1038 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1039 info->lo_encrypt_type = 1040 lo->lo_encryption ? lo->lo_encryption->number : 0; 1041 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1042 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1043 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1044 lo->lo_encrypt_key_size); 1045 } 1046 return 0; 1047 } 1048 1049 static void 1050 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1051 { 1052 memset(info64, 0, sizeof(*info64)); 1053 info64->lo_number = info->lo_number; 1054 info64->lo_device = info->lo_device; 1055 info64->lo_inode = info->lo_inode; 1056 info64->lo_rdevice = info->lo_rdevice; 1057 info64->lo_offset = info->lo_offset; 1058 info64->lo_sizelimit = 0; 1059 info64->lo_encrypt_type = info->lo_encrypt_type; 1060 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1061 info64->lo_flags = info->lo_flags; 1062 info64->lo_init[0] = info->lo_init[0]; 1063 info64->lo_init[1] = info->lo_init[1]; 1064 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1065 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1066 else 1067 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1068 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1069 } 1070 1071 static int 1072 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1073 { 1074 memset(info, 0, sizeof(*info)); 1075 info->lo_number = info64->lo_number; 1076 info->lo_device = info64->lo_device; 1077 info->lo_inode = info64->lo_inode; 1078 info->lo_rdevice = info64->lo_rdevice; 1079 info->lo_offset = info64->lo_offset; 1080 info->lo_encrypt_type = info64->lo_encrypt_type; 1081 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1082 info->lo_flags = info64->lo_flags; 1083 info->lo_init[0] = info64->lo_init[0]; 1084 info->lo_init[1] = info64->lo_init[1]; 1085 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1086 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1087 else 1088 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1089 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1090 1091 /* error in case values were truncated */ 1092 if (info->lo_device != info64->lo_device || 1093 info->lo_rdevice != info64->lo_rdevice || 1094 info->lo_inode != info64->lo_inode || 1095 info->lo_offset != info64->lo_offset) 1096 return -EOVERFLOW; 1097 1098 return 0; 1099 } 1100 1101 static int 1102 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1103 { 1104 struct loop_info info; 1105 struct loop_info64 info64; 1106 1107 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1108 return -EFAULT; 1109 loop_info64_from_old(&info, &info64); 1110 return loop_set_status(lo, &info64); 1111 } 1112 1113 static int 1114 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1115 { 1116 struct loop_info64 info64; 1117 1118 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1119 return -EFAULT; 1120 return loop_set_status(lo, &info64); 1121 } 1122 1123 static int 1124 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1125 struct loop_info info; 1126 struct loop_info64 info64; 1127 int err = 0; 1128 1129 if (!arg) 1130 err = -EINVAL; 1131 if (!err) 1132 err = loop_get_status(lo, &info64); 1133 if (!err) 1134 err = loop_info64_to_old(&info64, &info); 1135 if (!err && copy_to_user(arg, &info, sizeof(info))) 1136 err = -EFAULT; 1137 1138 return err; 1139 } 1140 1141 static int 1142 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1143 struct loop_info64 info64; 1144 int err = 0; 1145 1146 if (!arg) 1147 err = -EINVAL; 1148 if (!err) 1149 err = loop_get_status(lo, &info64); 1150 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1151 err = -EFAULT; 1152 1153 return err; 1154 } 1155 1156 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) 1157 { 1158 if (unlikely(lo->lo_state != Lo_bound)) 1159 return -ENXIO; 1160 1161 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1162 } 1163 1164 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1165 unsigned int cmd, unsigned long arg) 1166 { 1167 struct loop_device *lo = bdev->bd_disk->private_data; 1168 int err; 1169 1170 mutex_lock_nested(&lo->lo_ctl_mutex, 1); 1171 switch (cmd) { 1172 case LOOP_SET_FD: 1173 err = loop_set_fd(lo, mode, bdev, arg); 1174 break; 1175 case LOOP_CHANGE_FD: 1176 err = loop_change_fd(lo, bdev, arg); 1177 break; 1178 case LOOP_CLR_FD: 1179 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1180 err = loop_clr_fd(lo); 1181 if (!err) 1182 goto out_unlocked; 1183 break; 1184 case LOOP_SET_STATUS: 1185 err = -EPERM; 1186 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1187 err = loop_set_status_old(lo, 1188 (struct loop_info __user *)arg); 1189 break; 1190 case LOOP_GET_STATUS: 1191 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1192 break; 1193 case LOOP_SET_STATUS64: 1194 err = -EPERM; 1195 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1196 err = loop_set_status64(lo, 1197 (struct loop_info64 __user *) arg); 1198 break; 1199 case LOOP_GET_STATUS64: 1200 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1201 break; 1202 case LOOP_SET_CAPACITY: 1203 err = -EPERM; 1204 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1205 err = loop_set_capacity(lo, bdev); 1206 break; 1207 default: 1208 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1209 } 1210 mutex_unlock(&lo->lo_ctl_mutex); 1211 1212 out_unlocked: 1213 return err; 1214 } 1215 1216 #ifdef CONFIG_COMPAT 1217 struct compat_loop_info { 1218 compat_int_t lo_number; /* ioctl r/o */ 1219 compat_dev_t lo_device; /* ioctl r/o */ 1220 compat_ulong_t lo_inode; /* ioctl r/o */ 1221 compat_dev_t lo_rdevice; /* ioctl r/o */ 1222 compat_int_t lo_offset; 1223 compat_int_t lo_encrypt_type; 1224 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1225 compat_int_t lo_flags; /* ioctl r/o */ 1226 char lo_name[LO_NAME_SIZE]; 1227 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1228 compat_ulong_t lo_init[2]; 1229 char reserved[4]; 1230 }; 1231 1232 /* 1233 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1234 * - noinlined to reduce stack space usage in main part of driver 1235 */ 1236 static noinline int 1237 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1238 struct loop_info64 *info64) 1239 { 1240 struct compat_loop_info info; 1241 1242 if (copy_from_user(&info, arg, sizeof(info))) 1243 return -EFAULT; 1244 1245 memset(info64, 0, sizeof(*info64)); 1246 info64->lo_number = info.lo_number; 1247 info64->lo_device = info.lo_device; 1248 info64->lo_inode = info.lo_inode; 1249 info64->lo_rdevice = info.lo_rdevice; 1250 info64->lo_offset = info.lo_offset; 1251 info64->lo_sizelimit = 0; 1252 info64->lo_encrypt_type = info.lo_encrypt_type; 1253 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1254 info64->lo_flags = info.lo_flags; 1255 info64->lo_init[0] = info.lo_init[0]; 1256 info64->lo_init[1] = info.lo_init[1]; 1257 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1258 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1259 else 1260 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1261 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1262 return 0; 1263 } 1264 1265 /* 1266 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1267 * - noinlined to reduce stack space usage in main part of driver 1268 */ 1269 static noinline int 1270 loop_info64_to_compat(const struct loop_info64 *info64, 1271 struct compat_loop_info __user *arg) 1272 { 1273 struct compat_loop_info info; 1274 1275 memset(&info, 0, sizeof(info)); 1276 info.lo_number = info64->lo_number; 1277 info.lo_device = info64->lo_device; 1278 info.lo_inode = info64->lo_inode; 1279 info.lo_rdevice = info64->lo_rdevice; 1280 info.lo_offset = info64->lo_offset; 1281 info.lo_encrypt_type = info64->lo_encrypt_type; 1282 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1283 info.lo_flags = info64->lo_flags; 1284 info.lo_init[0] = info64->lo_init[0]; 1285 info.lo_init[1] = info64->lo_init[1]; 1286 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1287 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1288 else 1289 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1290 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1291 1292 /* error in case values were truncated */ 1293 if (info.lo_device != info64->lo_device || 1294 info.lo_rdevice != info64->lo_rdevice || 1295 info.lo_inode != info64->lo_inode || 1296 info.lo_offset != info64->lo_offset || 1297 info.lo_init[0] != info64->lo_init[0] || 1298 info.lo_init[1] != info64->lo_init[1]) 1299 return -EOVERFLOW; 1300 1301 if (copy_to_user(arg, &info, sizeof(info))) 1302 return -EFAULT; 1303 return 0; 1304 } 1305 1306 static int 1307 loop_set_status_compat(struct loop_device *lo, 1308 const struct compat_loop_info __user *arg) 1309 { 1310 struct loop_info64 info64; 1311 int ret; 1312 1313 ret = loop_info64_from_compat(arg, &info64); 1314 if (ret < 0) 1315 return ret; 1316 return loop_set_status(lo, &info64); 1317 } 1318 1319 static int 1320 loop_get_status_compat(struct loop_device *lo, 1321 struct compat_loop_info __user *arg) 1322 { 1323 struct loop_info64 info64; 1324 int err = 0; 1325 1326 if (!arg) 1327 err = -EINVAL; 1328 if (!err) 1329 err = loop_get_status(lo, &info64); 1330 if (!err) 1331 err = loop_info64_to_compat(&info64, arg); 1332 return err; 1333 } 1334 1335 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1336 unsigned int cmd, unsigned long arg) 1337 { 1338 struct loop_device *lo = bdev->bd_disk->private_data; 1339 int err; 1340 1341 switch(cmd) { 1342 case LOOP_SET_STATUS: 1343 mutex_lock(&lo->lo_ctl_mutex); 1344 err = loop_set_status_compat( 1345 lo, (const struct compat_loop_info __user *) arg); 1346 mutex_unlock(&lo->lo_ctl_mutex); 1347 break; 1348 case LOOP_GET_STATUS: 1349 mutex_lock(&lo->lo_ctl_mutex); 1350 err = loop_get_status_compat( 1351 lo, (struct compat_loop_info __user *) arg); 1352 mutex_unlock(&lo->lo_ctl_mutex); 1353 break; 1354 case LOOP_SET_CAPACITY: 1355 case LOOP_CLR_FD: 1356 case LOOP_GET_STATUS64: 1357 case LOOP_SET_STATUS64: 1358 arg = (unsigned long) compat_ptr(arg); 1359 case LOOP_SET_FD: 1360 case LOOP_CHANGE_FD: 1361 err = lo_ioctl(bdev, mode, cmd, arg); 1362 break; 1363 default: 1364 err = -ENOIOCTLCMD; 1365 break; 1366 } 1367 return err; 1368 } 1369 #endif 1370 1371 static int lo_open(struct block_device *bdev, fmode_t mode) 1372 { 1373 struct loop_device *lo; 1374 int err = 0; 1375 1376 mutex_lock(&loop_index_mutex); 1377 lo = bdev->bd_disk->private_data; 1378 if (!lo) { 1379 err = -ENXIO; 1380 goto out; 1381 } 1382 1383 mutex_lock(&lo->lo_ctl_mutex); 1384 lo->lo_refcnt++; 1385 mutex_unlock(&lo->lo_ctl_mutex); 1386 out: 1387 mutex_unlock(&loop_index_mutex); 1388 return err; 1389 } 1390 1391 static void lo_release(struct gendisk *disk, fmode_t mode) 1392 { 1393 struct loop_device *lo = disk->private_data; 1394 int err; 1395 1396 mutex_lock(&lo->lo_ctl_mutex); 1397 1398 if (--lo->lo_refcnt) 1399 goto out; 1400 1401 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1402 /* 1403 * In autoclear mode, stop the loop thread 1404 * and remove configuration after last close. 1405 */ 1406 err = loop_clr_fd(lo); 1407 if (!err) 1408 return; 1409 } else { 1410 /* 1411 * Otherwise keep thread (if running) and config, 1412 * but flush possible ongoing bios in thread. 1413 */ 1414 loop_flush(lo); 1415 } 1416 1417 out: 1418 mutex_unlock(&lo->lo_ctl_mutex); 1419 } 1420 1421 static const struct block_device_operations lo_fops = { 1422 .owner = THIS_MODULE, 1423 .open = lo_open, 1424 .release = lo_release, 1425 .ioctl = lo_ioctl, 1426 #ifdef CONFIG_COMPAT 1427 .compat_ioctl = lo_compat_ioctl, 1428 #endif 1429 }; 1430 1431 /* 1432 * And now the modules code and kernel interface. 1433 */ 1434 static int max_loop; 1435 module_param(max_loop, int, S_IRUGO); 1436 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1437 module_param(max_part, int, S_IRUGO); 1438 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1439 MODULE_LICENSE("GPL"); 1440 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1441 1442 int loop_register_transfer(struct loop_func_table *funcs) 1443 { 1444 unsigned int n = funcs->number; 1445 1446 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1447 return -EINVAL; 1448 xfer_funcs[n] = funcs; 1449 return 0; 1450 } 1451 1452 static int unregister_transfer_cb(int id, void *ptr, void *data) 1453 { 1454 struct loop_device *lo = ptr; 1455 struct loop_func_table *xfer = data; 1456 1457 mutex_lock(&lo->lo_ctl_mutex); 1458 if (lo->lo_encryption == xfer) 1459 loop_release_xfer(lo); 1460 mutex_unlock(&lo->lo_ctl_mutex); 1461 return 0; 1462 } 1463 1464 int loop_unregister_transfer(int number) 1465 { 1466 unsigned int n = number; 1467 struct loop_func_table *xfer; 1468 1469 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1470 return -EINVAL; 1471 1472 xfer_funcs[n] = NULL; 1473 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1474 return 0; 1475 } 1476 1477 EXPORT_SYMBOL(loop_register_transfer); 1478 EXPORT_SYMBOL(loop_unregister_transfer); 1479 1480 static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1481 const struct blk_mq_queue_data *bd) 1482 { 1483 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); 1484 1485 blk_mq_start_request(bd->rq); 1486 1487 if (cmd->rq->cmd_flags & REQ_WRITE) { 1488 struct loop_device *lo = cmd->rq->q->queuedata; 1489 bool need_sched = true; 1490 1491 spin_lock_irq(&lo->lo_lock); 1492 if (lo->write_started) 1493 need_sched = false; 1494 else 1495 lo->write_started = true; 1496 list_add_tail(&cmd->list, &lo->write_cmd_head); 1497 spin_unlock_irq(&lo->lo_lock); 1498 1499 if (need_sched) 1500 queue_work(loop_wq, &lo->write_work); 1501 } else { 1502 queue_work(loop_wq, &cmd->read_work); 1503 } 1504 1505 return BLK_MQ_RQ_QUEUE_OK; 1506 } 1507 1508 static void loop_handle_cmd(struct loop_cmd *cmd) 1509 { 1510 const bool write = cmd->rq->cmd_flags & REQ_WRITE; 1511 struct loop_device *lo = cmd->rq->q->queuedata; 1512 int ret = -EIO; 1513 struct bio *bio; 1514 1515 if (lo->lo_state != Lo_bound) 1516 goto failed; 1517 1518 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) 1519 goto failed; 1520 1521 ret = 0; 1522 __rq_for_each_bio(bio, cmd->rq) 1523 ret |= loop_handle_bio(lo, bio); 1524 1525 failed: 1526 if (ret) 1527 cmd->rq->errors = -EIO; 1528 blk_mq_complete_request(cmd->rq); 1529 } 1530 1531 static void loop_queue_write_work(struct work_struct *work) 1532 { 1533 struct loop_device *lo = 1534 container_of(work, struct loop_device, write_work); 1535 LIST_HEAD(cmd_list); 1536 1537 spin_lock_irq(&lo->lo_lock); 1538 repeat: 1539 list_splice_init(&lo->write_cmd_head, &cmd_list); 1540 spin_unlock_irq(&lo->lo_lock); 1541 1542 while (!list_empty(&cmd_list)) { 1543 struct loop_cmd *cmd = list_first_entry(&cmd_list, 1544 struct loop_cmd, list); 1545 list_del_init(&cmd->list); 1546 loop_handle_cmd(cmd); 1547 } 1548 1549 spin_lock_irq(&lo->lo_lock); 1550 if (!list_empty(&lo->write_cmd_head)) 1551 goto repeat; 1552 lo->write_started = false; 1553 spin_unlock_irq(&lo->lo_lock); 1554 } 1555 1556 static void loop_queue_read_work(struct work_struct *work) 1557 { 1558 struct loop_cmd *cmd = 1559 container_of(work, struct loop_cmd, read_work); 1560 1561 loop_handle_cmd(cmd); 1562 } 1563 1564 static int loop_init_request(void *data, struct request *rq, 1565 unsigned int hctx_idx, unsigned int request_idx, 1566 unsigned int numa_node) 1567 { 1568 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1569 1570 cmd->rq = rq; 1571 INIT_WORK(&cmd->read_work, loop_queue_read_work); 1572 1573 return 0; 1574 } 1575 1576 static struct blk_mq_ops loop_mq_ops = { 1577 .queue_rq = loop_queue_rq, 1578 .map_queue = blk_mq_map_queue, 1579 .init_request = loop_init_request, 1580 }; 1581 1582 static int loop_add(struct loop_device **l, int i) 1583 { 1584 struct loop_device *lo; 1585 struct gendisk *disk; 1586 int err; 1587 1588 err = -ENOMEM; 1589 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1590 if (!lo) 1591 goto out; 1592 1593 lo->lo_state = Lo_unbound; 1594 1595 /* allocate id, if @id >= 0, we're requesting that specific id */ 1596 if (i >= 0) { 1597 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1598 if (err == -ENOSPC) 1599 err = -EEXIST; 1600 } else { 1601 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1602 } 1603 if (err < 0) 1604 goto out_free_dev; 1605 i = err; 1606 1607 err = -ENOMEM; 1608 lo->tag_set.ops = &loop_mq_ops; 1609 lo->tag_set.nr_hw_queues = 1; 1610 lo->tag_set.queue_depth = 128; 1611 lo->tag_set.numa_node = NUMA_NO_NODE; 1612 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1613 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1614 lo->tag_set.driver_data = lo; 1615 1616 err = blk_mq_alloc_tag_set(&lo->tag_set); 1617 if (err) 1618 goto out_free_idr; 1619 1620 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1621 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1622 err = PTR_ERR(lo->lo_queue); 1623 goto out_cleanup_tags; 1624 } 1625 lo->lo_queue->queuedata = lo; 1626 1627 INIT_LIST_HEAD(&lo->write_cmd_head); 1628 INIT_WORK(&lo->write_work, loop_queue_write_work); 1629 1630 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1631 if (!disk) 1632 goto out_free_queue; 1633 1634 /* 1635 * Disable partition scanning by default. The in-kernel partition 1636 * scanning can be requested individually per-device during its 1637 * setup. Userspace can always add and remove partitions from all 1638 * devices. The needed partition minors are allocated from the 1639 * extended minor space, the main loop device numbers will continue 1640 * to match the loop minors, regardless of the number of partitions 1641 * used. 1642 * 1643 * If max_part is given, partition scanning is globally enabled for 1644 * all loop devices. The minors for the main loop devices will be 1645 * multiples of max_part. 1646 * 1647 * Note: Global-for-all-devices, set-only-at-init, read-only module 1648 * parameteters like 'max_loop' and 'max_part' make things needlessly 1649 * complicated, are too static, inflexible and may surprise 1650 * userspace tools. Parameters like this in general should be avoided. 1651 */ 1652 if (!part_shift) 1653 disk->flags |= GENHD_FL_NO_PART_SCAN; 1654 disk->flags |= GENHD_FL_EXT_DEVT; 1655 mutex_init(&lo->lo_ctl_mutex); 1656 lo->lo_number = i; 1657 spin_lock_init(&lo->lo_lock); 1658 disk->major = LOOP_MAJOR; 1659 disk->first_minor = i << part_shift; 1660 disk->fops = &lo_fops; 1661 disk->private_data = lo; 1662 disk->queue = lo->lo_queue; 1663 sprintf(disk->disk_name, "loop%d", i); 1664 add_disk(disk); 1665 *l = lo; 1666 return lo->lo_number; 1667 1668 out_free_queue: 1669 blk_cleanup_queue(lo->lo_queue); 1670 out_cleanup_tags: 1671 blk_mq_free_tag_set(&lo->tag_set); 1672 out_free_idr: 1673 idr_remove(&loop_index_idr, i); 1674 out_free_dev: 1675 kfree(lo); 1676 out: 1677 return err; 1678 } 1679 1680 static void loop_remove(struct loop_device *lo) 1681 { 1682 del_gendisk(lo->lo_disk); 1683 blk_cleanup_queue(lo->lo_queue); 1684 blk_mq_free_tag_set(&lo->tag_set); 1685 put_disk(lo->lo_disk); 1686 kfree(lo); 1687 } 1688 1689 static int find_free_cb(int id, void *ptr, void *data) 1690 { 1691 struct loop_device *lo = ptr; 1692 struct loop_device **l = data; 1693 1694 if (lo->lo_state == Lo_unbound) { 1695 *l = lo; 1696 return 1; 1697 } 1698 return 0; 1699 } 1700 1701 static int loop_lookup(struct loop_device **l, int i) 1702 { 1703 struct loop_device *lo; 1704 int ret = -ENODEV; 1705 1706 if (i < 0) { 1707 int err; 1708 1709 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1710 if (err == 1) { 1711 *l = lo; 1712 ret = lo->lo_number; 1713 } 1714 goto out; 1715 } 1716 1717 /* lookup and return a specific i */ 1718 lo = idr_find(&loop_index_idr, i); 1719 if (lo) { 1720 *l = lo; 1721 ret = lo->lo_number; 1722 } 1723 out: 1724 return ret; 1725 } 1726 1727 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1728 { 1729 struct loop_device *lo; 1730 struct kobject *kobj; 1731 int err; 1732 1733 mutex_lock(&loop_index_mutex); 1734 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1735 if (err < 0) 1736 err = loop_add(&lo, MINOR(dev) >> part_shift); 1737 if (err < 0) 1738 kobj = NULL; 1739 else 1740 kobj = get_disk(lo->lo_disk); 1741 mutex_unlock(&loop_index_mutex); 1742 1743 *part = 0; 1744 return kobj; 1745 } 1746 1747 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1748 unsigned long parm) 1749 { 1750 struct loop_device *lo; 1751 int ret = -ENOSYS; 1752 1753 mutex_lock(&loop_index_mutex); 1754 switch (cmd) { 1755 case LOOP_CTL_ADD: 1756 ret = loop_lookup(&lo, parm); 1757 if (ret >= 0) { 1758 ret = -EEXIST; 1759 break; 1760 } 1761 ret = loop_add(&lo, parm); 1762 break; 1763 case LOOP_CTL_REMOVE: 1764 ret = loop_lookup(&lo, parm); 1765 if (ret < 0) 1766 break; 1767 mutex_lock(&lo->lo_ctl_mutex); 1768 if (lo->lo_state != Lo_unbound) { 1769 ret = -EBUSY; 1770 mutex_unlock(&lo->lo_ctl_mutex); 1771 break; 1772 } 1773 if (lo->lo_refcnt > 0) { 1774 ret = -EBUSY; 1775 mutex_unlock(&lo->lo_ctl_mutex); 1776 break; 1777 } 1778 lo->lo_disk->private_data = NULL; 1779 mutex_unlock(&lo->lo_ctl_mutex); 1780 idr_remove(&loop_index_idr, lo->lo_number); 1781 loop_remove(lo); 1782 break; 1783 case LOOP_CTL_GET_FREE: 1784 ret = loop_lookup(&lo, -1); 1785 if (ret >= 0) 1786 break; 1787 ret = loop_add(&lo, -1); 1788 } 1789 mutex_unlock(&loop_index_mutex); 1790 1791 return ret; 1792 } 1793 1794 static const struct file_operations loop_ctl_fops = { 1795 .open = nonseekable_open, 1796 .unlocked_ioctl = loop_control_ioctl, 1797 .compat_ioctl = loop_control_ioctl, 1798 .owner = THIS_MODULE, 1799 .llseek = noop_llseek, 1800 }; 1801 1802 static struct miscdevice loop_misc = { 1803 .minor = LOOP_CTRL_MINOR, 1804 .name = "loop-control", 1805 .fops = &loop_ctl_fops, 1806 }; 1807 1808 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 1809 MODULE_ALIAS("devname:loop-control"); 1810 1811 static int __init loop_init(void) 1812 { 1813 int i, nr; 1814 unsigned long range; 1815 struct loop_device *lo; 1816 int err; 1817 1818 err = misc_register(&loop_misc); 1819 if (err < 0) 1820 return err; 1821 1822 part_shift = 0; 1823 if (max_part > 0) { 1824 part_shift = fls(max_part); 1825 1826 /* 1827 * Adjust max_part according to part_shift as it is exported 1828 * to user space so that user can decide correct minor number 1829 * if [s]he want to create more devices. 1830 * 1831 * Note that -1 is required because partition 0 is reserved 1832 * for the whole disk. 1833 */ 1834 max_part = (1UL << part_shift) - 1; 1835 } 1836 1837 if ((1UL << part_shift) > DISK_MAX_PARTS) { 1838 err = -EINVAL; 1839 goto misc_out; 1840 } 1841 1842 if (max_loop > 1UL << (MINORBITS - part_shift)) { 1843 err = -EINVAL; 1844 goto misc_out; 1845 } 1846 1847 /* 1848 * If max_loop is specified, create that many devices upfront. 1849 * This also becomes a hard limit. If max_loop is not specified, 1850 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 1851 * init time. Loop devices can be requested on-demand with the 1852 * /dev/loop-control interface, or be instantiated by accessing 1853 * a 'dead' device node. 1854 */ 1855 if (max_loop) { 1856 nr = max_loop; 1857 range = max_loop << part_shift; 1858 } else { 1859 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 1860 range = 1UL << MINORBITS; 1861 } 1862 1863 if (register_blkdev(LOOP_MAJOR, "loop")) { 1864 err = -EIO; 1865 goto misc_out; 1866 } 1867 1868 loop_wq = alloc_workqueue("kloopd", 1869 WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 0); 1870 if (!loop_wq) { 1871 err = -ENOMEM; 1872 goto misc_out; 1873 } 1874 1875 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 1876 THIS_MODULE, loop_probe, NULL, NULL); 1877 1878 /* pre-create number of devices given by config or max_loop */ 1879 mutex_lock(&loop_index_mutex); 1880 for (i = 0; i < nr; i++) 1881 loop_add(&lo, i); 1882 mutex_unlock(&loop_index_mutex); 1883 1884 printk(KERN_INFO "loop: module loaded\n"); 1885 return 0; 1886 1887 misc_out: 1888 misc_deregister(&loop_misc); 1889 return err; 1890 } 1891 1892 static int loop_exit_cb(int id, void *ptr, void *data) 1893 { 1894 struct loop_device *lo = ptr; 1895 1896 loop_remove(lo); 1897 return 0; 1898 } 1899 1900 static void __exit loop_exit(void) 1901 { 1902 unsigned long range; 1903 1904 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 1905 1906 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 1907 idr_destroy(&loop_index_idr); 1908 1909 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 1910 unregister_blkdev(LOOP_MAJOR, "loop"); 1911 1912 destroy_workqueue(loop_wq); 1913 1914 misc_deregister(&loop_misc); 1915 } 1916 1917 module_init(loop_init); 1918 module_exit(loop_exit); 1919 1920 #ifndef MODULE 1921 static int __init max_loop_setup(char *str) 1922 { 1923 max_loop = simple_strtol(str, NULL, 0); 1924 return 1; 1925 } 1926 1927 __setup("max_loop=", max_loop_setup); 1928 #endif 1929