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/loop.h> 67 #include <linux/compat.h> 68 #include <linux/suspend.h> 69 #include <linux/freezer.h> 70 #include <linux/mutex.h> 71 #include <linux/writeback.h> 72 #include <linux/completion.h> 73 #include <linux/highmem.h> 74 #include <linux/kthread.h> 75 #include <linux/splice.h> 76 #include <linux/sysfs.h> 77 #include <linux/miscdevice.h> 78 #include <linux/falloc.h> 79 80 #include <asm/uaccess.h> 81 82 static DEFINE_IDR(loop_index_idr); 83 static DEFINE_MUTEX(loop_index_mutex); 84 85 static int max_part; 86 static int part_shift; 87 88 /* 89 * Transfer functions 90 */ 91 static int transfer_none(struct loop_device *lo, int cmd, 92 struct page *raw_page, unsigned raw_off, 93 struct page *loop_page, unsigned loop_off, 94 int size, sector_t real_block) 95 { 96 char *raw_buf = kmap_atomic(raw_page) + raw_off; 97 char *loop_buf = kmap_atomic(loop_page) + loop_off; 98 99 if (cmd == READ) 100 memcpy(loop_buf, raw_buf, size); 101 else 102 memcpy(raw_buf, loop_buf, size); 103 104 kunmap_atomic(loop_buf); 105 kunmap_atomic(raw_buf); 106 cond_resched(); 107 return 0; 108 } 109 110 static int transfer_xor(struct loop_device *lo, int cmd, 111 struct page *raw_page, unsigned raw_off, 112 struct page *loop_page, unsigned loop_off, 113 int size, sector_t real_block) 114 { 115 char *raw_buf = kmap_atomic(raw_page) + raw_off; 116 char *loop_buf = kmap_atomic(loop_page) + loop_off; 117 char *in, *out, *key; 118 int i, keysize; 119 120 if (cmd == READ) { 121 in = raw_buf; 122 out = loop_buf; 123 } else { 124 in = loop_buf; 125 out = raw_buf; 126 } 127 128 key = lo->lo_encrypt_key; 129 keysize = lo->lo_encrypt_key_size; 130 for (i = 0; i < size; i++) 131 *out++ = *in++ ^ key[(i & 511) % keysize]; 132 133 kunmap_atomic(loop_buf); 134 kunmap_atomic(raw_buf); 135 cond_resched(); 136 return 0; 137 } 138 139 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 140 { 141 if (unlikely(info->lo_encrypt_key_size <= 0)) 142 return -EINVAL; 143 return 0; 144 } 145 146 static struct loop_func_table none_funcs = { 147 .number = LO_CRYPT_NONE, 148 .transfer = transfer_none, 149 }; 150 151 static struct loop_func_table xor_funcs = { 152 .number = LO_CRYPT_XOR, 153 .transfer = transfer_xor, 154 .init = xor_init 155 }; 156 157 /* xfer_funcs[0] is special - its release function is never called */ 158 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 159 &none_funcs, 160 &xor_funcs 161 }; 162 163 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 164 { 165 loff_t size, loopsize; 166 167 /* Compute loopsize in bytes */ 168 size = i_size_read(file->f_mapping->host); 169 loopsize = size - offset; 170 /* offset is beyond i_size, wierd but possible */ 171 if (loopsize < 0) 172 return 0; 173 174 if (sizelimit > 0 && sizelimit < loopsize) 175 loopsize = sizelimit; 176 /* 177 * Unfortunately, if we want to do I/O on the device, 178 * the number of 512-byte sectors has to fit into a sector_t. 179 */ 180 return loopsize >> 9; 181 } 182 183 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 184 { 185 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 186 } 187 188 static int 189 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 190 { 191 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 192 sector_t x = (sector_t)size; 193 struct block_device *bdev = lo->lo_device; 194 195 if (unlikely((loff_t)x != size)) 196 return -EFBIG; 197 if (lo->lo_offset != offset) 198 lo->lo_offset = offset; 199 if (lo->lo_sizelimit != sizelimit) 200 lo->lo_sizelimit = sizelimit; 201 set_capacity(lo->lo_disk, x); 202 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 203 /* let user-space know about the new size */ 204 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 205 return 0; 206 } 207 208 static inline int 209 lo_do_transfer(struct loop_device *lo, int cmd, 210 struct page *rpage, unsigned roffs, 211 struct page *lpage, unsigned loffs, 212 int size, sector_t rblock) 213 { 214 if (unlikely(!lo->transfer)) 215 return 0; 216 217 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 218 } 219 220 /** 221 * __do_lo_send_write - helper for writing data to a loop device 222 * 223 * This helper just factors out common code between do_lo_send_direct_write() 224 * and do_lo_send_write(). 225 */ 226 static int __do_lo_send_write(struct file *file, 227 u8 *buf, const int len, loff_t pos) 228 { 229 ssize_t bw; 230 mm_segment_t old_fs = get_fs(); 231 232 set_fs(get_ds()); 233 bw = file->f_op->write(file, buf, len, &pos); 234 set_fs(old_fs); 235 if (likely(bw == len)) 236 return 0; 237 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n", 238 (unsigned long long)pos, len); 239 if (bw >= 0) 240 bw = -EIO; 241 return bw; 242 } 243 244 /** 245 * do_lo_send_direct_write - helper for writing data to a loop device 246 * 247 * This is the fast, non-transforming version that does not need double 248 * buffering. 249 */ 250 static int do_lo_send_direct_write(struct loop_device *lo, 251 struct bio_vec *bvec, loff_t pos, struct page *page) 252 { 253 ssize_t bw = __do_lo_send_write(lo->lo_backing_file, 254 kmap(bvec->bv_page) + bvec->bv_offset, 255 bvec->bv_len, pos); 256 kunmap(bvec->bv_page); 257 cond_resched(); 258 return bw; 259 } 260 261 /** 262 * do_lo_send_write - helper for writing data to a loop device 263 * 264 * This is the slow, transforming version that needs to double buffer the 265 * data as it cannot do the transformations in place without having direct 266 * access to the destination pages of the backing file. 267 */ 268 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec, 269 loff_t pos, struct page *page) 270 { 271 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page, 272 bvec->bv_offset, bvec->bv_len, pos >> 9); 273 if (likely(!ret)) 274 return __do_lo_send_write(lo->lo_backing_file, 275 page_address(page), bvec->bv_len, 276 pos); 277 printk(KERN_ERR "loop: Transfer error at byte offset %llu, " 278 "length %i.\n", (unsigned long long)pos, bvec->bv_len); 279 if (ret > 0) 280 ret = -EIO; 281 return ret; 282 } 283 284 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos) 285 { 286 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t, 287 struct page *page); 288 struct bio_vec *bvec; 289 struct page *page = NULL; 290 int i, ret = 0; 291 292 if (lo->transfer != transfer_none) { 293 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); 294 if (unlikely(!page)) 295 goto fail; 296 kmap(page); 297 do_lo_send = do_lo_send_write; 298 } else { 299 do_lo_send = do_lo_send_direct_write; 300 } 301 302 bio_for_each_segment(bvec, bio, i) { 303 ret = do_lo_send(lo, bvec, pos, page); 304 if (ret < 0) 305 break; 306 pos += bvec->bv_len; 307 } 308 if (page) { 309 kunmap(page); 310 __free_page(page); 311 } 312 out: 313 return ret; 314 fail: 315 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n"); 316 ret = -ENOMEM; 317 goto out; 318 } 319 320 struct lo_read_data { 321 struct loop_device *lo; 322 struct page *page; 323 unsigned offset; 324 int bsize; 325 }; 326 327 static int 328 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 329 struct splice_desc *sd) 330 { 331 struct lo_read_data *p = sd->u.data; 332 struct loop_device *lo = p->lo; 333 struct page *page = buf->page; 334 sector_t IV; 335 int size; 336 337 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) + 338 (buf->offset >> 9); 339 size = sd->len; 340 if (size > p->bsize) 341 size = p->bsize; 342 343 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) { 344 printk(KERN_ERR "loop: transfer error block %ld\n", 345 page->index); 346 size = -EINVAL; 347 } 348 349 flush_dcache_page(p->page); 350 351 if (size > 0) 352 p->offset += size; 353 354 return size; 355 } 356 357 static int 358 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd) 359 { 360 return __splice_from_pipe(pipe, sd, lo_splice_actor); 361 } 362 363 static ssize_t 364 do_lo_receive(struct loop_device *lo, 365 struct bio_vec *bvec, int bsize, loff_t pos) 366 { 367 struct lo_read_data cookie; 368 struct splice_desc sd; 369 struct file *file; 370 ssize_t retval; 371 372 cookie.lo = lo; 373 cookie.page = bvec->bv_page; 374 cookie.offset = bvec->bv_offset; 375 cookie.bsize = bsize; 376 377 sd.len = 0; 378 sd.total_len = bvec->bv_len; 379 sd.flags = 0; 380 sd.pos = pos; 381 sd.u.data = &cookie; 382 383 file = lo->lo_backing_file; 384 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor); 385 386 return retval; 387 } 388 389 static int 390 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos) 391 { 392 struct bio_vec *bvec; 393 ssize_t s; 394 int i; 395 396 bio_for_each_segment(bvec, bio, i) { 397 s = do_lo_receive(lo, bvec, bsize, pos); 398 if (s < 0) 399 return s; 400 401 if (s != bvec->bv_len) { 402 zero_fill_bio(bio); 403 break; 404 } 405 pos += bvec->bv_len; 406 } 407 return 0; 408 } 409 410 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio) 411 { 412 loff_t pos; 413 int ret; 414 415 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset; 416 417 if (bio_rw(bio) == WRITE) { 418 struct file *file = lo->lo_backing_file; 419 420 if (bio->bi_rw & REQ_FLUSH) { 421 ret = vfs_fsync(file, 0); 422 if (unlikely(ret && ret != -EINVAL)) { 423 ret = -EIO; 424 goto out; 425 } 426 } 427 428 /* 429 * We use punch hole to reclaim the free space used by the 430 * image a.k.a. discard. However we do not support discard if 431 * encryption is enabled, because it may give an attacker 432 * useful information. 433 */ 434 if (bio->bi_rw & REQ_DISCARD) { 435 struct file *file = lo->lo_backing_file; 436 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 437 438 if ((!file->f_op->fallocate) || 439 lo->lo_encrypt_key_size) { 440 ret = -EOPNOTSUPP; 441 goto out; 442 } 443 ret = file->f_op->fallocate(file, mode, pos, 444 bio->bi_size); 445 if (unlikely(ret && ret != -EINVAL && 446 ret != -EOPNOTSUPP)) 447 ret = -EIO; 448 goto out; 449 } 450 451 ret = lo_send(lo, bio, pos); 452 453 if ((bio->bi_rw & REQ_FUA) && !ret) { 454 ret = vfs_fsync(file, 0); 455 if (unlikely(ret && ret != -EINVAL)) 456 ret = -EIO; 457 } 458 } else 459 ret = lo_receive(lo, bio, lo->lo_blocksize, pos); 460 461 out: 462 return ret; 463 } 464 465 /* 466 * Add bio to back of pending list 467 */ 468 static void loop_add_bio(struct loop_device *lo, struct bio *bio) 469 { 470 lo->lo_bio_count++; 471 bio_list_add(&lo->lo_bio_list, bio); 472 } 473 474 /* 475 * Grab first pending buffer 476 */ 477 static struct bio *loop_get_bio(struct loop_device *lo) 478 { 479 lo->lo_bio_count--; 480 return bio_list_pop(&lo->lo_bio_list); 481 } 482 483 static void loop_make_request(struct request_queue *q, struct bio *old_bio) 484 { 485 struct loop_device *lo = q->queuedata; 486 int rw = bio_rw(old_bio); 487 488 if (rw == READA) 489 rw = READ; 490 491 BUG_ON(!lo || (rw != READ && rw != WRITE)); 492 493 spin_lock_irq(&lo->lo_lock); 494 if (lo->lo_state != Lo_bound) 495 goto out; 496 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY))) 497 goto out; 498 if (lo->lo_bio_count >= q->nr_congestion_on) 499 wait_event_lock_irq(lo->lo_req_wait, 500 lo->lo_bio_count < q->nr_congestion_off, 501 lo->lo_lock); 502 loop_add_bio(lo, old_bio); 503 wake_up(&lo->lo_event); 504 spin_unlock_irq(&lo->lo_lock); 505 return; 506 507 out: 508 spin_unlock_irq(&lo->lo_lock); 509 bio_io_error(old_bio); 510 } 511 512 struct switch_request { 513 struct file *file; 514 struct completion wait; 515 }; 516 517 static void do_loop_switch(struct loop_device *, struct switch_request *); 518 519 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio) 520 { 521 if (unlikely(!bio->bi_bdev)) { 522 do_loop_switch(lo, bio->bi_private); 523 bio_put(bio); 524 } else { 525 int ret = do_bio_filebacked(lo, bio); 526 bio_endio(bio, ret); 527 } 528 } 529 530 /* 531 * worker thread that handles reads/writes to file backed loop devices, 532 * to avoid blocking in our make_request_fn. it also does loop decrypting 533 * on reads for block backed loop, as that is too heavy to do from 534 * b_end_io context where irqs may be disabled. 535 * 536 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before 537 * calling kthread_stop(). Therefore once kthread_should_stop() is 538 * true, make_request will not place any more requests. Therefore 539 * once kthread_should_stop() is true and lo_bio is NULL, we are 540 * done with the loop. 541 */ 542 static int loop_thread(void *data) 543 { 544 struct loop_device *lo = data; 545 struct bio *bio; 546 547 set_user_nice(current, -20); 548 549 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) { 550 551 wait_event_interruptible(lo->lo_event, 552 !bio_list_empty(&lo->lo_bio_list) || 553 kthread_should_stop()); 554 555 if (bio_list_empty(&lo->lo_bio_list)) 556 continue; 557 spin_lock_irq(&lo->lo_lock); 558 bio = loop_get_bio(lo); 559 if (lo->lo_bio_count < lo->lo_queue->nr_congestion_off) 560 wake_up(&lo->lo_req_wait); 561 spin_unlock_irq(&lo->lo_lock); 562 563 BUG_ON(!bio); 564 loop_handle_bio(lo, bio); 565 } 566 567 return 0; 568 } 569 570 /* 571 * loop_switch performs the hard work of switching a backing store. 572 * First it needs to flush existing IO, it does this by sending a magic 573 * BIO down the pipe. The completion of this BIO does the actual switch. 574 */ 575 static int loop_switch(struct loop_device *lo, struct file *file) 576 { 577 struct switch_request w; 578 struct bio *bio = bio_alloc(GFP_KERNEL, 0); 579 if (!bio) 580 return -ENOMEM; 581 init_completion(&w.wait); 582 w.file = file; 583 bio->bi_private = &w; 584 bio->bi_bdev = NULL; 585 loop_make_request(lo->lo_queue, bio); 586 wait_for_completion(&w.wait); 587 return 0; 588 } 589 590 /* 591 * Helper to flush the IOs in loop, but keeping loop thread running 592 */ 593 static int loop_flush(struct loop_device *lo) 594 { 595 /* loop not yet configured, no running thread, nothing to flush */ 596 if (!lo->lo_thread) 597 return 0; 598 599 return loop_switch(lo, NULL); 600 } 601 602 /* 603 * Do the actual switch; called from the BIO completion routine 604 */ 605 static void do_loop_switch(struct loop_device *lo, struct switch_request *p) 606 { 607 struct file *file = p->file; 608 struct file *old_file = lo->lo_backing_file; 609 struct address_space *mapping; 610 611 /* if no new file, only flush of queued bios requested */ 612 if (!file) 613 goto out; 614 615 mapping = file->f_mapping; 616 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 617 lo->lo_backing_file = file; 618 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? 619 mapping->host->i_bdev->bd_block_size : PAGE_SIZE; 620 lo->old_gfp_mask = mapping_gfp_mask(mapping); 621 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 622 out: 623 complete(&p->wait); 624 } 625 626 627 /* 628 * loop_change_fd switched the backing store of a loopback device to 629 * a new file. This is useful for operating system installers to free up 630 * the original file and in High Availability environments to switch to 631 * an alternative location for the content in case of server meltdown. 632 * This can only work if the loop device is used read-only, and if the 633 * new backing store is the same size and type as the old backing store. 634 */ 635 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 636 unsigned int arg) 637 { 638 struct file *file, *old_file; 639 struct inode *inode; 640 int error; 641 642 error = -ENXIO; 643 if (lo->lo_state != Lo_bound) 644 goto out; 645 646 /* the loop device has to be read-only */ 647 error = -EINVAL; 648 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 649 goto out; 650 651 error = -EBADF; 652 file = fget(arg); 653 if (!file) 654 goto out; 655 656 inode = file->f_mapping->host; 657 old_file = lo->lo_backing_file; 658 659 error = -EINVAL; 660 661 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 662 goto out_putf; 663 664 /* size of the new backing store needs to be the same */ 665 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 666 goto out_putf; 667 668 /* and ... switch */ 669 error = loop_switch(lo, file); 670 if (error) 671 goto out_putf; 672 673 fput(old_file); 674 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 675 ioctl_by_bdev(bdev, BLKRRPART, 0); 676 return 0; 677 678 out_putf: 679 fput(file); 680 out: 681 return error; 682 } 683 684 static inline int is_loop_device(struct file *file) 685 { 686 struct inode *i = file->f_mapping->host; 687 688 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 689 } 690 691 /* loop sysfs attributes */ 692 693 static ssize_t loop_attr_show(struct device *dev, char *page, 694 ssize_t (*callback)(struct loop_device *, char *)) 695 { 696 struct gendisk *disk = dev_to_disk(dev); 697 struct loop_device *lo = disk->private_data; 698 699 return callback(lo, page); 700 } 701 702 #define LOOP_ATTR_RO(_name) \ 703 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 704 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 705 struct device_attribute *attr, char *b) \ 706 { \ 707 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 708 } \ 709 static struct device_attribute loop_attr_##_name = \ 710 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); 711 712 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 713 { 714 ssize_t ret; 715 char *p = NULL; 716 717 spin_lock_irq(&lo->lo_lock); 718 if (lo->lo_backing_file) 719 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1); 720 spin_unlock_irq(&lo->lo_lock); 721 722 if (IS_ERR_OR_NULL(p)) 723 ret = PTR_ERR(p); 724 else { 725 ret = strlen(p); 726 memmove(buf, p, ret); 727 buf[ret++] = '\n'; 728 buf[ret] = 0; 729 } 730 731 return ret; 732 } 733 734 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 735 { 736 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 737 } 738 739 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 740 { 741 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 742 } 743 744 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 745 { 746 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 747 748 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 749 } 750 751 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 752 { 753 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 754 755 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 756 } 757 758 LOOP_ATTR_RO(backing_file); 759 LOOP_ATTR_RO(offset); 760 LOOP_ATTR_RO(sizelimit); 761 LOOP_ATTR_RO(autoclear); 762 LOOP_ATTR_RO(partscan); 763 764 static struct attribute *loop_attrs[] = { 765 &loop_attr_backing_file.attr, 766 &loop_attr_offset.attr, 767 &loop_attr_sizelimit.attr, 768 &loop_attr_autoclear.attr, 769 &loop_attr_partscan.attr, 770 NULL, 771 }; 772 773 static struct attribute_group loop_attribute_group = { 774 .name = "loop", 775 .attrs= loop_attrs, 776 }; 777 778 static int loop_sysfs_init(struct loop_device *lo) 779 { 780 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 781 &loop_attribute_group); 782 } 783 784 static void loop_sysfs_exit(struct loop_device *lo) 785 { 786 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 787 &loop_attribute_group); 788 } 789 790 static void loop_config_discard(struct loop_device *lo) 791 { 792 struct file *file = lo->lo_backing_file; 793 struct inode *inode = file->f_mapping->host; 794 struct request_queue *q = lo->lo_queue; 795 796 /* 797 * We use punch hole to reclaim the free space used by the 798 * image a.k.a. discard. However we do support discard if 799 * encryption is enabled, because it may give an attacker 800 * useful information. 801 */ 802 if ((!file->f_op->fallocate) || 803 lo->lo_encrypt_key_size) { 804 q->limits.discard_granularity = 0; 805 q->limits.discard_alignment = 0; 806 q->limits.max_discard_sectors = 0; 807 q->limits.discard_zeroes_data = 0; 808 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 809 return; 810 } 811 812 q->limits.discard_granularity = inode->i_sb->s_blocksize; 813 q->limits.discard_alignment = 0; 814 q->limits.max_discard_sectors = UINT_MAX >> 9; 815 q->limits.discard_zeroes_data = 1; 816 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 817 } 818 819 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 820 struct block_device *bdev, unsigned int arg) 821 { 822 struct file *file, *f; 823 struct inode *inode; 824 struct address_space *mapping; 825 unsigned lo_blocksize; 826 int lo_flags = 0; 827 int error; 828 loff_t size; 829 830 /* This is safe, since we have a reference from open(). */ 831 __module_get(THIS_MODULE); 832 833 error = -EBADF; 834 file = fget(arg); 835 if (!file) 836 goto out; 837 838 error = -EBUSY; 839 if (lo->lo_state != Lo_unbound) 840 goto out_putf; 841 842 /* Avoid recursion */ 843 f = file; 844 while (is_loop_device(f)) { 845 struct loop_device *l; 846 847 if (f->f_mapping->host->i_bdev == bdev) 848 goto out_putf; 849 850 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 851 if (l->lo_state == Lo_unbound) { 852 error = -EINVAL; 853 goto out_putf; 854 } 855 f = l->lo_backing_file; 856 } 857 858 mapping = file->f_mapping; 859 inode = mapping->host; 860 861 error = -EINVAL; 862 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 863 goto out_putf; 864 865 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 866 !file->f_op->write) 867 lo_flags |= LO_FLAGS_READ_ONLY; 868 869 lo_blocksize = S_ISBLK(inode->i_mode) ? 870 inode->i_bdev->bd_block_size : PAGE_SIZE; 871 872 error = -EFBIG; 873 size = get_loop_size(lo, file); 874 if ((loff_t)(sector_t)size != size) 875 goto out_putf; 876 877 error = 0; 878 879 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 880 881 lo->lo_blocksize = lo_blocksize; 882 lo->lo_device = bdev; 883 lo->lo_flags = lo_flags; 884 lo->lo_backing_file = file; 885 lo->transfer = transfer_none; 886 lo->ioctl = NULL; 887 lo->lo_sizelimit = 0; 888 lo->lo_bio_count = 0; 889 lo->old_gfp_mask = mapping_gfp_mask(mapping); 890 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 891 892 bio_list_init(&lo->lo_bio_list); 893 894 /* 895 * set queue make_request_fn, and add limits based on lower level 896 * device 897 */ 898 blk_queue_make_request(lo->lo_queue, loop_make_request); 899 lo->lo_queue->queuedata = lo; 900 901 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 902 blk_queue_flush(lo->lo_queue, REQ_FLUSH); 903 904 set_capacity(lo->lo_disk, size); 905 bd_set_size(bdev, size << 9); 906 loop_sysfs_init(lo); 907 /* let user-space know about the new size */ 908 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 909 910 set_blocksize(bdev, lo_blocksize); 911 912 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d", 913 lo->lo_number); 914 if (IS_ERR(lo->lo_thread)) { 915 error = PTR_ERR(lo->lo_thread); 916 goto out_clr; 917 } 918 lo->lo_state = Lo_bound; 919 wake_up_process(lo->lo_thread); 920 if (part_shift) 921 lo->lo_flags |= LO_FLAGS_PARTSCAN; 922 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 923 ioctl_by_bdev(bdev, BLKRRPART, 0); 924 return 0; 925 926 out_clr: 927 loop_sysfs_exit(lo); 928 lo->lo_thread = NULL; 929 lo->lo_device = NULL; 930 lo->lo_backing_file = NULL; 931 lo->lo_flags = 0; 932 set_capacity(lo->lo_disk, 0); 933 invalidate_bdev(bdev); 934 bd_set_size(bdev, 0); 935 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 936 mapping_set_gfp_mask(mapping, lo->old_gfp_mask); 937 lo->lo_state = Lo_unbound; 938 out_putf: 939 fput(file); 940 out: 941 /* This is safe: open() is still holding a reference. */ 942 module_put(THIS_MODULE); 943 return error; 944 } 945 946 static int 947 loop_release_xfer(struct loop_device *lo) 948 { 949 int err = 0; 950 struct loop_func_table *xfer = lo->lo_encryption; 951 952 if (xfer) { 953 if (xfer->release) 954 err = xfer->release(lo); 955 lo->transfer = NULL; 956 lo->lo_encryption = NULL; 957 module_put(xfer->owner); 958 } 959 return err; 960 } 961 962 static int 963 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 964 const struct loop_info64 *i) 965 { 966 int err = 0; 967 968 if (xfer) { 969 struct module *owner = xfer->owner; 970 971 if (!try_module_get(owner)) 972 return -EINVAL; 973 if (xfer->init) 974 err = xfer->init(lo, i); 975 if (err) 976 module_put(owner); 977 else 978 lo->lo_encryption = xfer; 979 } 980 return err; 981 } 982 983 static int loop_clr_fd(struct loop_device *lo) 984 { 985 struct file *filp = lo->lo_backing_file; 986 gfp_t gfp = lo->old_gfp_mask; 987 struct block_device *bdev = lo->lo_device; 988 989 if (lo->lo_state != Lo_bound) 990 return -ENXIO; 991 992 /* 993 * If we've explicitly asked to tear down the loop device, 994 * and it has an elevated reference count, set it for auto-teardown when 995 * the last reference goes away. This stops $!~#$@ udev from 996 * preventing teardown because it decided that it needs to run blkid on 997 * the loopback device whenever they appear. xfstests is notorious for 998 * failing tests because blkid via udev races with a losetup 999 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1000 * command to fail with EBUSY. 1001 */ 1002 if (lo->lo_refcnt > 1) { 1003 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1004 mutex_unlock(&lo->lo_ctl_mutex); 1005 return 0; 1006 } 1007 1008 if (filp == NULL) 1009 return -EINVAL; 1010 1011 spin_lock_irq(&lo->lo_lock); 1012 lo->lo_state = Lo_rundown; 1013 spin_unlock_irq(&lo->lo_lock); 1014 1015 kthread_stop(lo->lo_thread); 1016 1017 spin_lock_irq(&lo->lo_lock); 1018 lo->lo_backing_file = NULL; 1019 spin_unlock_irq(&lo->lo_lock); 1020 1021 loop_release_xfer(lo); 1022 lo->transfer = NULL; 1023 lo->ioctl = NULL; 1024 lo->lo_device = NULL; 1025 lo->lo_encryption = NULL; 1026 lo->lo_offset = 0; 1027 lo->lo_sizelimit = 0; 1028 lo->lo_encrypt_key_size = 0; 1029 lo->lo_thread = NULL; 1030 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1031 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1032 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1033 if (bdev) 1034 invalidate_bdev(bdev); 1035 set_capacity(lo->lo_disk, 0); 1036 loop_sysfs_exit(lo); 1037 if (bdev) { 1038 bd_set_size(bdev, 0); 1039 /* let user-space know about this change */ 1040 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1041 } 1042 mapping_set_gfp_mask(filp->f_mapping, gfp); 1043 lo->lo_state = Lo_unbound; 1044 /* This is safe: open() is still holding a reference. */ 1045 module_put(THIS_MODULE); 1046 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 1047 ioctl_by_bdev(bdev, BLKRRPART, 0); 1048 lo->lo_flags = 0; 1049 if (!part_shift) 1050 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1051 mutex_unlock(&lo->lo_ctl_mutex); 1052 /* 1053 * Need not hold lo_ctl_mutex to fput backing file. 1054 * Calling fput holding lo_ctl_mutex triggers a circular 1055 * lock dependency possibility warning as fput can take 1056 * bd_mutex which is usually taken before lo_ctl_mutex. 1057 */ 1058 fput(filp); 1059 return 0; 1060 } 1061 1062 static int 1063 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1064 { 1065 int err; 1066 struct loop_func_table *xfer; 1067 kuid_t uid = current_uid(); 1068 1069 if (lo->lo_encrypt_key_size && 1070 !uid_eq(lo->lo_key_owner, uid) && 1071 !capable(CAP_SYS_ADMIN)) 1072 return -EPERM; 1073 if (lo->lo_state != Lo_bound) 1074 return -ENXIO; 1075 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1076 return -EINVAL; 1077 1078 err = loop_release_xfer(lo); 1079 if (err) 1080 return err; 1081 1082 if (info->lo_encrypt_type) { 1083 unsigned int type = info->lo_encrypt_type; 1084 1085 if (type >= MAX_LO_CRYPT) 1086 return -EINVAL; 1087 xfer = xfer_funcs[type]; 1088 if (xfer == NULL) 1089 return -EINVAL; 1090 } else 1091 xfer = NULL; 1092 1093 err = loop_init_xfer(lo, xfer, info); 1094 if (err) 1095 return err; 1096 1097 if (lo->lo_offset != info->lo_offset || 1098 lo->lo_sizelimit != info->lo_sizelimit) 1099 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) 1100 return -EFBIG; 1101 1102 loop_config_discard(lo); 1103 1104 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1105 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1106 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1107 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1108 1109 if (!xfer) 1110 xfer = &none_funcs; 1111 lo->transfer = xfer->transfer; 1112 lo->ioctl = xfer->ioctl; 1113 1114 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1115 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1116 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1117 1118 if ((info->lo_flags & LO_FLAGS_PARTSCAN) && 1119 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1120 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1121 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1122 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0); 1123 } 1124 1125 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1126 lo->lo_init[0] = info->lo_init[0]; 1127 lo->lo_init[1] = info->lo_init[1]; 1128 if (info->lo_encrypt_key_size) { 1129 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1130 info->lo_encrypt_key_size); 1131 lo->lo_key_owner = uid; 1132 } 1133 1134 return 0; 1135 } 1136 1137 static int 1138 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1139 { 1140 struct file *file = lo->lo_backing_file; 1141 struct kstat stat; 1142 int error; 1143 1144 if (lo->lo_state != Lo_bound) 1145 return -ENXIO; 1146 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat); 1147 if (error) 1148 return error; 1149 memset(info, 0, sizeof(*info)); 1150 info->lo_number = lo->lo_number; 1151 info->lo_device = huge_encode_dev(stat.dev); 1152 info->lo_inode = stat.ino; 1153 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); 1154 info->lo_offset = lo->lo_offset; 1155 info->lo_sizelimit = lo->lo_sizelimit; 1156 info->lo_flags = lo->lo_flags; 1157 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1158 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1159 info->lo_encrypt_type = 1160 lo->lo_encryption ? lo->lo_encryption->number : 0; 1161 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1162 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1163 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1164 lo->lo_encrypt_key_size); 1165 } 1166 return 0; 1167 } 1168 1169 static void 1170 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1171 { 1172 memset(info64, 0, sizeof(*info64)); 1173 info64->lo_number = info->lo_number; 1174 info64->lo_device = info->lo_device; 1175 info64->lo_inode = info->lo_inode; 1176 info64->lo_rdevice = info->lo_rdevice; 1177 info64->lo_offset = info->lo_offset; 1178 info64->lo_sizelimit = 0; 1179 info64->lo_encrypt_type = info->lo_encrypt_type; 1180 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1181 info64->lo_flags = info->lo_flags; 1182 info64->lo_init[0] = info->lo_init[0]; 1183 info64->lo_init[1] = info->lo_init[1]; 1184 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1185 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1186 else 1187 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1188 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1189 } 1190 1191 static int 1192 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1193 { 1194 memset(info, 0, sizeof(*info)); 1195 info->lo_number = info64->lo_number; 1196 info->lo_device = info64->lo_device; 1197 info->lo_inode = info64->lo_inode; 1198 info->lo_rdevice = info64->lo_rdevice; 1199 info->lo_offset = info64->lo_offset; 1200 info->lo_encrypt_type = info64->lo_encrypt_type; 1201 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1202 info->lo_flags = info64->lo_flags; 1203 info->lo_init[0] = info64->lo_init[0]; 1204 info->lo_init[1] = info64->lo_init[1]; 1205 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1206 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1207 else 1208 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1209 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1210 1211 /* error in case values were truncated */ 1212 if (info->lo_device != info64->lo_device || 1213 info->lo_rdevice != info64->lo_rdevice || 1214 info->lo_inode != info64->lo_inode || 1215 info->lo_offset != info64->lo_offset) 1216 return -EOVERFLOW; 1217 1218 return 0; 1219 } 1220 1221 static int 1222 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1223 { 1224 struct loop_info info; 1225 struct loop_info64 info64; 1226 1227 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1228 return -EFAULT; 1229 loop_info64_from_old(&info, &info64); 1230 return loop_set_status(lo, &info64); 1231 } 1232 1233 static int 1234 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1235 { 1236 struct loop_info64 info64; 1237 1238 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1239 return -EFAULT; 1240 return loop_set_status(lo, &info64); 1241 } 1242 1243 static int 1244 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1245 struct loop_info info; 1246 struct loop_info64 info64; 1247 int err = 0; 1248 1249 if (!arg) 1250 err = -EINVAL; 1251 if (!err) 1252 err = loop_get_status(lo, &info64); 1253 if (!err) 1254 err = loop_info64_to_old(&info64, &info); 1255 if (!err && copy_to_user(arg, &info, sizeof(info))) 1256 err = -EFAULT; 1257 1258 return err; 1259 } 1260 1261 static int 1262 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1263 struct loop_info64 info64; 1264 int err = 0; 1265 1266 if (!arg) 1267 err = -EINVAL; 1268 if (!err) 1269 err = loop_get_status(lo, &info64); 1270 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1271 err = -EFAULT; 1272 1273 return err; 1274 } 1275 1276 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) 1277 { 1278 if (unlikely(lo->lo_state != Lo_bound)) 1279 return -ENXIO; 1280 1281 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1282 } 1283 1284 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1285 unsigned int cmd, unsigned long arg) 1286 { 1287 struct loop_device *lo = bdev->bd_disk->private_data; 1288 int err; 1289 1290 mutex_lock_nested(&lo->lo_ctl_mutex, 1); 1291 switch (cmd) { 1292 case LOOP_SET_FD: 1293 err = loop_set_fd(lo, mode, bdev, arg); 1294 break; 1295 case LOOP_CHANGE_FD: 1296 err = loop_change_fd(lo, bdev, arg); 1297 break; 1298 case LOOP_CLR_FD: 1299 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1300 err = loop_clr_fd(lo); 1301 if (!err) 1302 goto out_unlocked; 1303 break; 1304 case LOOP_SET_STATUS: 1305 err = -EPERM; 1306 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1307 err = loop_set_status_old(lo, 1308 (struct loop_info __user *)arg); 1309 break; 1310 case LOOP_GET_STATUS: 1311 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1312 break; 1313 case LOOP_SET_STATUS64: 1314 err = -EPERM; 1315 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1316 err = loop_set_status64(lo, 1317 (struct loop_info64 __user *) arg); 1318 break; 1319 case LOOP_GET_STATUS64: 1320 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1321 break; 1322 case LOOP_SET_CAPACITY: 1323 err = -EPERM; 1324 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1325 err = loop_set_capacity(lo, bdev); 1326 break; 1327 default: 1328 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1329 } 1330 mutex_unlock(&lo->lo_ctl_mutex); 1331 1332 out_unlocked: 1333 return err; 1334 } 1335 1336 #ifdef CONFIG_COMPAT 1337 struct compat_loop_info { 1338 compat_int_t lo_number; /* ioctl r/o */ 1339 compat_dev_t lo_device; /* ioctl r/o */ 1340 compat_ulong_t lo_inode; /* ioctl r/o */ 1341 compat_dev_t lo_rdevice; /* ioctl r/o */ 1342 compat_int_t lo_offset; 1343 compat_int_t lo_encrypt_type; 1344 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1345 compat_int_t lo_flags; /* ioctl r/o */ 1346 char lo_name[LO_NAME_SIZE]; 1347 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1348 compat_ulong_t lo_init[2]; 1349 char reserved[4]; 1350 }; 1351 1352 /* 1353 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1354 * - noinlined to reduce stack space usage in main part of driver 1355 */ 1356 static noinline int 1357 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1358 struct loop_info64 *info64) 1359 { 1360 struct compat_loop_info info; 1361 1362 if (copy_from_user(&info, arg, sizeof(info))) 1363 return -EFAULT; 1364 1365 memset(info64, 0, sizeof(*info64)); 1366 info64->lo_number = info.lo_number; 1367 info64->lo_device = info.lo_device; 1368 info64->lo_inode = info.lo_inode; 1369 info64->lo_rdevice = info.lo_rdevice; 1370 info64->lo_offset = info.lo_offset; 1371 info64->lo_sizelimit = 0; 1372 info64->lo_encrypt_type = info.lo_encrypt_type; 1373 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1374 info64->lo_flags = info.lo_flags; 1375 info64->lo_init[0] = info.lo_init[0]; 1376 info64->lo_init[1] = info.lo_init[1]; 1377 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1378 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1379 else 1380 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1381 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1382 return 0; 1383 } 1384 1385 /* 1386 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1387 * - noinlined to reduce stack space usage in main part of driver 1388 */ 1389 static noinline int 1390 loop_info64_to_compat(const struct loop_info64 *info64, 1391 struct compat_loop_info __user *arg) 1392 { 1393 struct compat_loop_info info; 1394 1395 memset(&info, 0, sizeof(info)); 1396 info.lo_number = info64->lo_number; 1397 info.lo_device = info64->lo_device; 1398 info.lo_inode = info64->lo_inode; 1399 info.lo_rdevice = info64->lo_rdevice; 1400 info.lo_offset = info64->lo_offset; 1401 info.lo_encrypt_type = info64->lo_encrypt_type; 1402 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1403 info.lo_flags = info64->lo_flags; 1404 info.lo_init[0] = info64->lo_init[0]; 1405 info.lo_init[1] = info64->lo_init[1]; 1406 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1407 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1408 else 1409 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1410 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1411 1412 /* error in case values were truncated */ 1413 if (info.lo_device != info64->lo_device || 1414 info.lo_rdevice != info64->lo_rdevice || 1415 info.lo_inode != info64->lo_inode || 1416 info.lo_offset != info64->lo_offset || 1417 info.lo_init[0] != info64->lo_init[0] || 1418 info.lo_init[1] != info64->lo_init[1]) 1419 return -EOVERFLOW; 1420 1421 if (copy_to_user(arg, &info, sizeof(info))) 1422 return -EFAULT; 1423 return 0; 1424 } 1425 1426 static int 1427 loop_set_status_compat(struct loop_device *lo, 1428 const struct compat_loop_info __user *arg) 1429 { 1430 struct loop_info64 info64; 1431 int ret; 1432 1433 ret = loop_info64_from_compat(arg, &info64); 1434 if (ret < 0) 1435 return ret; 1436 return loop_set_status(lo, &info64); 1437 } 1438 1439 static int 1440 loop_get_status_compat(struct loop_device *lo, 1441 struct compat_loop_info __user *arg) 1442 { 1443 struct loop_info64 info64; 1444 int err = 0; 1445 1446 if (!arg) 1447 err = -EINVAL; 1448 if (!err) 1449 err = loop_get_status(lo, &info64); 1450 if (!err) 1451 err = loop_info64_to_compat(&info64, arg); 1452 return err; 1453 } 1454 1455 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1456 unsigned int cmd, unsigned long arg) 1457 { 1458 struct loop_device *lo = bdev->bd_disk->private_data; 1459 int err; 1460 1461 switch(cmd) { 1462 case LOOP_SET_STATUS: 1463 mutex_lock(&lo->lo_ctl_mutex); 1464 err = loop_set_status_compat( 1465 lo, (const struct compat_loop_info __user *) arg); 1466 mutex_unlock(&lo->lo_ctl_mutex); 1467 break; 1468 case LOOP_GET_STATUS: 1469 mutex_lock(&lo->lo_ctl_mutex); 1470 err = loop_get_status_compat( 1471 lo, (struct compat_loop_info __user *) arg); 1472 mutex_unlock(&lo->lo_ctl_mutex); 1473 break; 1474 case LOOP_SET_CAPACITY: 1475 case LOOP_CLR_FD: 1476 case LOOP_GET_STATUS64: 1477 case LOOP_SET_STATUS64: 1478 arg = (unsigned long) compat_ptr(arg); 1479 case LOOP_SET_FD: 1480 case LOOP_CHANGE_FD: 1481 err = lo_ioctl(bdev, mode, cmd, arg); 1482 break; 1483 default: 1484 err = -ENOIOCTLCMD; 1485 break; 1486 } 1487 return err; 1488 } 1489 #endif 1490 1491 static int lo_open(struct block_device *bdev, fmode_t mode) 1492 { 1493 struct loop_device *lo; 1494 int err = 0; 1495 1496 mutex_lock(&loop_index_mutex); 1497 lo = bdev->bd_disk->private_data; 1498 if (!lo) { 1499 err = -ENXIO; 1500 goto out; 1501 } 1502 1503 mutex_lock(&lo->lo_ctl_mutex); 1504 lo->lo_refcnt++; 1505 mutex_unlock(&lo->lo_ctl_mutex); 1506 out: 1507 mutex_unlock(&loop_index_mutex); 1508 return err; 1509 } 1510 1511 static int lo_release(struct gendisk *disk, fmode_t mode) 1512 { 1513 struct loop_device *lo = disk->private_data; 1514 int err; 1515 1516 mutex_lock(&lo->lo_ctl_mutex); 1517 1518 if (--lo->lo_refcnt) 1519 goto out; 1520 1521 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1522 /* 1523 * In autoclear mode, stop the loop thread 1524 * and remove configuration after last close. 1525 */ 1526 err = loop_clr_fd(lo); 1527 if (!err) 1528 goto out_unlocked; 1529 } else { 1530 /* 1531 * Otherwise keep thread (if running) and config, 1532 * but flush possible ongoing bios in thread. 1533 */ 1534 loop_flush(lo); 1535 } 1536 1537 out: 1538 mutex_unlock(&lo->lo_ctl_mutex); 1539 out_unlocked: 1540 return 0; 1541 } 1542 1543 static const struct block_device_operations lo_fops = { 1544 .owner = THIS_MODULE, 1545 .open = lo_open, 1546 .release = lo_release, 1547 .ioctl = lo_ioctl, 1548 #ifdef CONFIG_COMPAT 1549 .compat_ioctl = lo_compat_ioctl, 1550 #endif 1551 }; 1552 1553 /* 1554 * And now the modules code and kernel interface. 1555 */ 1556 static int max_loop; 1557 module_param(max_loop, int, S_IRUGO); 1558 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1559 module_param(max_part, int, S_IRUGO); 1560 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1561 MODULE_LICENSE("GPL"); 1562 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1563 1564 int loop_register_transfer(struct loop_func_table *funcs) 1565 { 1566 unsigned int n = funcs->number; 1567 1568 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1569 return -EINVAL; 1570 xfer_funcs[n] = funcs; 1571 return 0; 1572 } 1573 1574 static int unregister_transfer_cb(int id, void *ptr, void *data) 1575 { 1576 struct loop_device *lo = ptr; 1577 struct loop_func_table *xfer = data; 1578 1579 mutex_lock(&lo->lo_ctl_mutex); 1580 if (lo->lo_encryption == xfer) 1581 loop_release_xfer(lo); 1582 mutex_unlock(&lo->lo_ctl_mutex); 1583 return 0; 1584 } 1585 1586 int loop_unregister_transfer(int number) 1587 { 1588 unsigned int n = number; 1589 struct loop_func_table *xfer; 1590 1591 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1592 return -EINVAL; 1593 1594 xfer_funcs[n] = NULL; 1595 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1596 return 0; 1597 } 1598 1599 EXPORT_SYMBOL(loop_register_transfer); 1600 EXPORT_SYMBOL(loop_unregister_transfer); 1601 1602 static int loop_add(struct loop_device **l, int i) 1603 { 1604 struct loop_device *lo; 1605 struct gendisk *disk; 1606 int err; 1607 1608 err = -ENOMEM; 1609 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1610 if (!lo) 1611 goto out; 1612 1613 if (!idr_pre_get(&loop_index_idr, GFP_KERNEL)) 1614 goto out_free_dev; 1615 1616 if (i >= 0) { 1617 int m; 1618 1619 /* create specific i in the index */ 1620 err = idr_get_new_above(&loop_index_idr, lo, i, &m); 1621 if (err >= 0 && i != m) { 1622 idr_remove(&loop_index_idr, m); 1623 err = -EEXIST; 1624 } 1625 } else if (i == -1) { 1626 int m; 1627 1628 /* get next free nr */ 1629 err = idr_get_new(&loop_index_idr, lo, &m); 1630 if (err >= 0) 1631 i = m; 1632 } else { 1633 err = -EINVAL; 1634 } 1635 if (err < 0) 1636 goto out_free_dev; 1637 1638 lo->lo_queue = blk_alloc_queue(GFP_KERNEL); 1639 if (!lo->lo_queue) 1640 goto out_free_dev; 1641 1642 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1643 if (!disk) 1644 goto out_free_queue; 1645 1646 /* 1647 * Disable partition scanning by default. The in-kernel partition 1648 * scanning can be requested individually per-device during its 1649 * setup. Userspace can always add and remove partitions from all 1650 * devices. The needed partition minors are allocated from the 1651 * extended minor space, the main loop device numbers will continue 1652 * to match the loop minors, regardless of the number of partitions 1653 * used. 1654 * 1655 * If max_part is given, partition scanning is globally enabled for 1656 * all loop devices. The minors for the main loop devices will be 1657 * multiples of max_part. 1658 * 1659 * Note: Global-for-all-devices, set-only-at-init, read-only module 1660 * parameteters like 'max_loop' and 'max_part' make things needlessly 1661 * complicated, are too static, inflexible and may surprise 1662 * userspace tools. Parameters like this in general should be avoided. 1663 */ 1664 if (!part_shift) 1665 disk->flags |= GENHD_FL_NO_PART_SCAN; 1666 disk->flags |= GENHD_FL_EXT_DEVT; 1667 mutex_init(&lo->lo_ctl_mutex); 1668 lo->lo_number = i; 1669 lo->lo_thread = NULL; 1670 init_waitqueue_head(&lo->lo_event); 1671 init_waitqueue_head(&lo->lo_req_wait); 1672 spin_lock_init(&lo->lo_lock); 1673 disk->major = LOOP_MAJOR; 1674 disk->first_minor = i << part_shift; 1675 disk->fops = &lo_fops; 1676 disk->private_data = lo; 1677 disk->queue = lo->lo_queue; 1678 sprintf(disk->disk_name, "loop%d", i); 1679 add_disk(disk); 1680 *l = lo; 1681 return lo->lo_number; 1682 1683 out_free_queue: 1684 blk_cleanup_queue(lo->lo_queue); 1685 out_free_dev: 1686 kfree(lo); 1687 out: 1688 return err; 1689 } 1690 1691 static void loop_remove(struct loop_device *lo) 1692 { 1693 del_gendisk(lo->lo_disk); 1694 blk_cleanup_queue(lo->lo_queue); 1695 put_disk(lo->lo_disk); 1696 kfree(lo); 1697 } 1698 1699 static int find_free_cb(int id, void *ptr, void *data) 1700 { 1701 struct loop_device *lo = ptr; 1702 struct loop_device **l = data; 1703 1704 if (lo->lo_state == Lo_unbound) { 1705 *l = lo; 1706 return 1; 1707 } 1708 return 0; 1709 } 1710 1711 static int loop_lookup(struct loop_device **l, int i) 1712 { 1713 struct loop_device *lo; 1714 int ret = -ENODEV; 1715 1716 if (i < 0) { 1717 int err; 1718 1719 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1720 if (err == 1) { 1721 *l = lo; 1722 ret = lo->lo_number; 1723 } 1724 goto out; 1725 } 1726 1727 /* lookup and return a specific i */ 1728 lo = idr_find(&loop_index_idr, i); 1729 if (lo) { 1730 *l = lo; 1731 ret = lo->lo_number; 1732 } 1733 out: 1734 return ret; 1735 } 1736 1737 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1738 { 1739 struct loop_device *lo; 1740 struct kobject *kobj; 1741 int err; 1742 1743 mutex_lock(&loop_index_mutex); 1744 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1745 if (err < 0) 1746 err = loop_add(&lo, MINOR(dev) >> part_shift); 1747 if (err < 0) 1748 kobj = ERR_PTR(err); 1749 else 1750 kobj = get_disk(lo->lo_disk); 1751 mutex_unlock(&loop_index_mutex); 1752 1753 *part = 0; 1754 return kobj; 1755 } 1756 1757 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1758 unsigned long parm) 1759 { 1760 struct loop_device *lo; 1761 int ret = -ENOSYS; 1762 1763 mutex_lock(&loop_index_mutex); 1764 switch (cmd) { 1765 case LOOP_CTL_ADD: 1766 ret = loop_lookup(&lo, parm); 1767 if (ret >= 0) { 1768 ret = -EEXIST; 1769 break; 1770 } 1771 ret = loop_add(&lo, parm); 1772 break; 1773 case LOOP_CTL_REMOVE: 1774 ret = loop_lookup(&lo, parm); 1775 if (ret < 0) 1776 break; 1777 mutex_lock(&lo->lo_ctl_mutex); 1778 if (lo->lo_state != Lo_unbound) { 1779 ret = -EBUSY; 1780 mutex_unlock(&lo->lo_ctl_mutex); 1781 break; 1782 } 1783 if (lo->lo_refcnt > 0) { 1784 ret = -EBUSY; 1785 mutex_unlock(&lo->lo_ctl_mutex); 1786 break; 1787 } 1788 lo->lo_disk->private_data = NULL; 1789 mutex_unlock(&lo->lo_ctl_mutex); 1790 idr_remove(&loop_index_idr, lo->lo_number); 1791 loop_remove(lo); 1792 break; 1793 case LOOP_CTL_GET_FREE: 1794 ret = loop_lookup(&lo, -1); 1795 if (ret >= 0) 1796 break; 1797 ret = loop_add(&lo, -1); 1798 } 1799 mutex_unlock(&loop_index_mutex); 1800 1801 return ret; 1802 } 1803 1804 static const struct file_operations loop_ctl_fops = { 1805 .open = nonseekable_open, 1806 .unlocked_ioctl = loop_control_ioctl, 1807 .compat_ioctl = loop_control_ioctl, 1808 .owner = THIS_MODULE, 1809 .llseek = noop_llseek, 1810 }; 1811 1812 static struct miscdevice loop_misc = { 1813 .minor = LOOP_CTRL_MINOR, 1814 .name = "loop-control", 1815 .fops = &loop_ctl_fops, 1816 }; 1817 1818 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 1819 MODULE_ALIAS("devname:loop-control"); 1820 1821 static int __init loop_init(void) 1822 { 1823 int i, nr; 1824 unsigned long range; 1825 struct loop_device *lo; 1826 int err; 1827 1828 err = misc_register(&loop_misc); 1829 if (err < 0) 1830 return err; 1831 1832 part_shift = 0; 1833 if (max_part > 0) { 1834 part_shift = fls(max_part); 1835 1836 /* 1837 * Adjust max_part according to part_shift as it is exported 1838 * to user space so that user can decide correct minor number 1839 * if [s]he want to create more devices. 1840 * 1841 * Note that -1 is required because partition 0 is reserved 1842 * for the whole disk. 1843 */ 1844 max_part = (1UL << part_shift) - 1; 1845 } 1846 1847 if ((1UL << part_shift) > DISK_MAX_PARTS) { 1848 err = -EINVAL; 1849 goto misc_out; 1850 } 1851 1852 if (max_loop > 1UL << (MINORBITS - part_shift)) { 1853 err = -EINVAL; 1854 goto misc_out; 1855 } 1856 1857 /* 1858 * If max_loop is specified, create that many devices upfront. 1859 * This also becomes a hard limit. If max_loop is not specified, 1860 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 1861 * init time. Loop devices can be requested on-demand with the 1862 * /dev/loop-control interface, or be instantiated by accessing 1863 * a 'dead' device node. 1864 */ 1865 if (max_loop) { 1866 nr = max_loop; 1867 range = max_loop << part_shift; 1868 } else { 1869 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 1870 range = 1UL << MINORBITS; 1871 } 1872 1873 if (register_blkdev(LOOP_MAJOR, "loop")) { 1874 err = -EIO; 1875 goto misc_out; 1876 } 1877 1878 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 1879 THIS_MODULE, loop_probe, NULL, NULL); 1880 1881 /* pre-create number of devices given by config or max_loop */ 1882 mutex_lock(&loop_index_mutex); 1883 for (i = 0; i < nr; i++) 1884 loop_add(&lo, i); 1885 mutex_unlock(&loop_index_mutex); 1886 1887 printk(KERN_INFO "loop: module loaded\n"); 1888 return 0; 1889 1890 misc_out: 1891 misc_deregister(&loop_misc); 1892 return err; 1893 } 1894 1895 static int loop_exit_cb(int id, void *ptr, void *data) 1896 { 1897 struct loop_device *lo = ptr; 1898 1899 loop_remove(lo); 1900 return 0; 1901 } 1902 1903 static void __exit loop_exit(void) 1904 { 1905 unsigned long range; 1906 1907 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 1908 1909 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 1910 idr_remove_all(&loop_index_idr); 1911 idr_destroy(&loop_index_idr); 1912 1913 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 1914 unregister_blkdev(LOOP_MAJOR, "loop"); 1915 1916 misc_deregister(&loop_misc); 1917 } 1918 1919 module_init(loop_init); 1920 module_exit(loop_exit); 1921 1922 #ifndef MODULE 1923 static int __init max_loop_setup(char *str) 1924 { 1925 max_loop = simple_strtol(str, NULL, 0); 1926 return 1; 1927 } 1928 1929 __setup("max_loop=", max_loop_setup); 1930 #endif 1931