1 /* 2 * linux/drivers/block/loop.c 3 * 4 * Written by Theodore Ts'o, 3/29/93 5 * 6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is 7 * permitted under the GNU General Public License. 8 * 9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993 10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996 11 * 12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994 13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996 14 * 15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997 16 * 17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998 18 * 19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998 20 * 21 * Loadable modules and other fixes by AK, 1998 22 * 23 * Make real block number available to downstream transfer functions, enables 24 * CBC (and relatives) mode encryption requiring unique IVs per data block. 25 * Reed H. Petty, rhp@draper.net 26 * 27 * Maximum number of loop devices now dynamic via max_loop module parameter. 28 * Russell Kroll <rkroll@exploits.org> 19990701 29 * 30 * Maximum number of loop devices when compiled-in now selectable by passing 31 * max_loop=<1-255> to the kernel on boot. 32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999 33 * 34 * Completely rewrite request handling to be make_request_fn style and 35 * non blocking, pushing work to a helper thread. Lots of fixes from 36 * Al Viro too. 37 * Jens Axboe <axboe@suse.de>, Nov 2000 38 * 39 * Support up to 256 loop devices 40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002 41 * 42 * Support for falling back on the write file operation when the address space 43 * operations write_begin is not available on the backing filesystem. 44 * Anton Altaparmakov, 16 Feb 2005 45 * 46 * Still To Fix: 47 * - Advisory locking is ignored here. 48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN 49 * 50 */ 51 52 #include <linux/module.h> 53 #include <linux/moduleparam.h> 54 #include <linux/sched.h> 55 #include <linux/fs.h> 56 #include <linux/file.h> 57 #include <linux/stat.h> 58 #include <linux/errno.h> 59 #include <linux/major.h> 60 #include <linux/wait.h> 61 #include <linux/blkdev.h> 62 #include <linux/blkpg.h> 63 #include <linux/init.h> 64 #include <linux/swap.h> 65 #include <linux/slab.h> 66 #include <linux/compat.h> 67 #include <linux/suspend.h> 68 #include <linux/freezer.h> 69 #include <linux/mutex.h> 70 #include <linux/writeback.h> 71 #include <linux/completion.h> 72 #include <linux/highmem.h> 73 #include <linux/kthread.h> 74 #include <linux/splice.h> 75 #include <linux/sysfs.h> 76 #include <linux/miscdevice.h> 77 #include <linux/falloc.h> 78 #include <linux/uio.h> 79 #include "loop.h" 80 81 #include <linux/uaccess.h> 82 83 static DEFINE_IDR(loop_index_idr); 84 static DEFINE_MUTEX(loop_index_mutex); 85 86 static int max_part; 87 static int part_shift; 88 89 static int transfer_xor(struct loop_device *lo, int cmd, 90 struct page *raw_page, unsigned raw_off, 91 struct page *loop_page, unsigned loop_off, 92 int size, sector_t real_block) 93 { 94 char *raw_buf = kmap_atomic(raw_page) + raw_off; 95 char *loop_buf = kmap_atomic(loop_page) + loop_off; 96 char *in, *out, *key; 97 int i, keysize; 98 99 if (cmd == READ) { 100 in = raw_buf; 101 out = loop_buf; 102 } else { 103 in = loop_buf; 104 out = raw_buf; 105 } 106 107 key = lo->lo_encrypt_key; 108 keysize = lo->lo_encrypt_key_size; 109 for (i = 0; i < size; i++) 110 *out++ = *in++ ^ key[(i & 511) % keysize]; 111 112 kunmap_atomic(loop_buf); 113 kunmap_atomic(raw_buf); 114 cond_resched(); 115 return 0; 116 } 117 118 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 119 { 120 if (unlikely(info->lo_encrypt_key_size <= 0)) 121 return -EINVAL; 122 return 0; 123 } 124 125 static struct loop_func_table none_funcs = { 126 .number = LO_CRYPT_NONE, 127 }; 128 129 static struct loop_func_table xor_funcs = { 130 .number = LO_CRYPT_XOR, 131 .transfer = transfer_xor, 132 .init = xor_init 133 }; 134 135 /* xfer_funcs[0] is special - its release function is never called */ 136 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 137 &none_funcs, 138 &xor_funcs 139 }; 140 141 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 142 { 143 loff_t loopsize; 144 145 /* Compute loopsize in bytes */ 146 loopsize = i_size_read(file->f_mapping->host); 147 if (offset > 0) 148 loopsize -= offset; 149 /* offset is beyond i_size, weird but possible */ 150 if (loopsize < 0) 151 return 0; 152 153 if (sizelimit > 0 && sizelimit < loopsize) 154 loopsize = sizelimit; 155 /* 156 * Unfortunately, if we want to do I/O on the device, 157 * the number of 512-byte sectors has to fit into a sector_t. 158 */ 159 return loopsize >> 9; 160 } 161 162 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 163 { 164 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 165 } 166 167 static void __loop_update_dio(struct loop_device *lo, bool dio) 168 { 169 struct file *file = lo->lo_backing_file; 170 struct address_space *mapping = file->f_mapping; 171 struct inode *inode = mapping->host; 172 unsigned short sb_bsize = 0; 173 unsigned dio_align = 0; 174 bool use_dio; 175 176 if (inode->i_sb->s_bdev) { 177 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 178 dio_align = sb_bsize - 1; 179 } 180 181 /* 182 * We support direct I/O only if lo_offset is aligned with the 183 * logical I/O size of backing device, and the logical block 184 * size of loop is bigger than the backing device's and the loop 185 * needn't transform transfer. 186 * 187 * TODO: the above condition may be loosed in the future, and 188 * direct I/O may be switched runtime at that time because most 189 * of requests in sane applications should be PAGE_SIZE aligned 190 */ 191 if (dio) { 192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 193 !(lo->lo_offset & dio_align) && 194 mapping->a_ops->direct_IO && 195 !lo->transfer) 196 use_dio = true; 197 else 198 use_dio = false; 199 } else { 200 use_dio = false; 201 } 202 203 if (lo->use_dio == use_dio) 204 return; 205 206 /* flush dirty pages before changing direct IO */ 207 vfs_fsync(file, 0); 208 209 /* 210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 212 * will get updated by ioctl(LOOP_GET_STATUS) 213 */ 214 blk_mq_freeze_queue(lo->lo_queue); 215 lo->use_dio = use_dio; 216 if (use_dio) { 217 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 218 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 219 } else { 220 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 221 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 222 } 223 blk_mq_unfreeze_queue(lo->lo_queue); 224 } 225 226 static int 227 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 228 { 229 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 230 sector_t x = (sector_t)size; 231 struct block_device *bdev = lo->lo_device; 232 233 if (unlikely((loff_t)x != size)) 234 return -EFBIG; 235 if (lo->lo_offset != offset) 236 lo->lo_offset = offset; 237 if (lo->lo_sizelimit != sizelimit) 238 lo->lo_sizelimit = sizelimit; 239 set_capacity(lo->lo_disk, x); 240 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 241 /* let user-space know about the new size */ 242 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 243 return 0; 244 } 245 246 static inline int 247 lo_do_transfer(struct loop_device *lo, int cmd, 248 struct page *rpage, unsigned roffs, 249 struct page *lpage, unsigned loffs, 250 int size, sector_t rblock) 251 { 252 int ret; 253 254 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 255 if (likely(!ret)) 256 return 0; 257 258 printk_ratelimited(KERN_ERR 259 "loop: Transfer error at byte offset %llu, length %i.\n", 260 (unsigned long long)rblock << 9, size); 261 return ret; 262 } 263 264 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 265 { 266 struct iov_iter i; 267 ssize_t bw; 268 269 iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len); 270 271 file_start_write(file); 272 bw = vfs_iter_write(file, &i, ppos, 0); 273 file_end_write(file); 274 275 if (likely(bw == bvec->bv_len)) 276 return 0; 277 278 printk_ratelimited(KERN_ERR 279 "loop: Write error at byte offset %llu, length %i.\n", 280 (unsigned long long)*ppos, bvec->bv_len); 281 if (bw >= 0) 282 bw = -EIO; 283 return bw; 284 } 285 286 static int lo_write_simple(struct loop_device *lo, struct request *rq, 287 loff_t pos) 288 { 289 struct bio_vec bvec; 290 struct req_iterator iter; 291 int ret = 0; 292 293 rq_for_each_segment(bvec, rq, iter) { 294 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 295 if (ret < 0) 296 break; 297 cond_resched(); 298 } 299 300 return ret; 301 } 302 303 /* 304 * This is the slow, transforming version that needs to double buffer the 305 * data as it cannot do the transformations in place without having direct 306 * access to the destination pages of the backing file. 307 */ 308 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 309 loff_t pos) 310 { 311 struct bio_vec bvec, b; 312 struct req_iterator iter; 313 struct page *page; 314 int ret = 0; 315 316 page = alloc_page(GFP_NOIO); 317 if (unlikely(!page)) 318 return -ENOMEM; 319 320 rq_for_each_segment(bvec, rq, iter) { 321 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 322 bvec.bv_offset, bvec.bv_len, pos >> 9); 323 if (unlikely(ret)) 324 break; 325 326 b.bv_page = page; 327 b.bv_offset = 0; 328 b.bv_len = bvec.bv_len; 329 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 330 if (ret < 0) 331 break; 332 } 333 334 __free_page(page); 335 return ret; 336 } 337 338 static int lo_read_simple(struct loop_device *lo, struct request *rq, 339 loff_t pos) 340 { 341 struct bio_vec bvec; 342 struct req_iterator iter; 343 struct iov_iter i; 344 ssize_t len; 345 346 rq_for_each_segment(bvec, rq, iter) { 347 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); 348 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 349 if (len < 0) 350 return len; 351 352 flush_dcache_page(bvec.bv_page); 353 354 if (len != bvec.bv_len) { 355 struct bio *bio; 356 357 __rq_for_each_bio(bio, rq) 358 zero_fill_bio(bio); 359 break; 360 } 361 cond_resched(); 362 } 363 364 return 0; 365 } 366 367 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 368 loff_t pos) 369 { 370 struct bio_vec bvec, b; 371 struct req_iterator iter; 372 struct iov_iter i; 373 struct page *page; 374 ssize_t len; 375 int ret = 0; 376 377 page = alloc_page(GFP_NOIO); 378 if (unlikely(!page)) 379 return -ENOMEM; 380 381 rq_for_each_segment(bvec, rq, iter) { 382 loff_t offset = pos; 383 384 b.bv_page = page; 385 b.bv_offset = 0; 386 b.bv_len = bvec.bv_len; 387 388 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); 389 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 390 if (len < 0) { 391 ret = len; 392 goto out_free_page; 393 } 394 395 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 396 bvec.bv_offset, len, offset >> 9); 397 if (ret) 398 goto out_free_page; 399 400 flush_dcache_page(bvec.bv_page); 401 402 if (len != bvec.bv_len) { 403 struct bio *bio; 404 405 __rq_for_each_bio(bio, rq) 406 zero_fill_bio(bio); 407 break; 408 } 409 } 410 411 ret = 0; 412 out_free_page: 413 __free_page(page); 414 return ret; 415 } 416 417 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 418 { 419 /* 420 * We use punch hole to reclaim the free space used by the 421 * image a.k.a. discard. However we do not support discard if 422 * encryption is enabled, because it may give an attacker 423 * useful information. 424 */ 425 struct file *file = lo->lo_backing_file; 426 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 427 int ret; 428 429 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 430 ret = -EOPNOTSUPP; 431 goto out; 432 } 433 434 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 435 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 436 ret = -EIO; 437 out: 438 return ret; 439 } 440 441 static int lo_req_flush(struct loop_device *lo, struct request *rq) 442 { 443 struct file *file = lo->lo_backing_file; 444 int ret = vfs_fsync(file, 0); 445 if (unlikely(ret && ret != -EINVAL)) 446 ret = -EIO; 447 448 return ret; 449 } 450 451 static void lo_complete_rq(struct request *rq) 452 { 453 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 454 blk_status_t ret = BLK_STS_OK; 455 456 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 457 req_op(rq) != REQ_OP_READ) { 458 if (cmd->ret < 0) 459 ret = BLK_STS_IOERR; 460 goto end_io; 461 } 462 463 /* 464 * Short READ - if we got some data, advance our request and 465 * retry it. If we got no data, end the rest with EIO. 466 */ 467 if (cmd->ret) { 468 blk_update_request(rq, BLK_STS_OK, cmd->ret); 469 cmd->ret = 0; 470 blk_mq_requeue_request(rq, true); 471 } else { 472 if (cmd->use_aio) { 473 struct bio *bio = rq->bio; 474 475 while (bio) { 476 zero_fill_bio(bio); 477 bio = bio->bi_next; 478 } 479 } 480 ret = BLK_STS_IOERR; 481 end_io: 482 blk_mq_end_request(rq, ret); 483 } 484 } 485 486 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 487 { 488 struct request *rq = blk_mq_rq_from_pdu(cmd); 489 490 if (!atomic_dec_and_test(&cmd->ref)) 491 return; 492 kfree(cmd->bvec); 493 cmd->bvec = NULL; 494 blk_mq_complete_request(rq); 495 } 496 497 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 498 { 499 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 500 501 if (cmd->css) 502 css_put(cmd->css); 503 cmd->ret = ret; 504 lo_rw_aio_do_completion(cmd); 505 } 506 507 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 508 loff_t pos, bool rw) 509 { 510 struct iov_iter iter; 511 struct bio_vec *bvec; 512 struct request *rq = blk_mq_rq_from_pdu(cmd); 513 struct bio *bio = rq->bio; 514 struct file *file = lo->lo_backing_file; 515 unsigned int offset; 516 int segments = 0; 517 int ret; 518 519 if (rq->bio != rq->biotail) { 520 struct req_iterator iter; 521 struct bio_vec tmp; 522 523 __rq_for_each_bio(bio, rq) 524 segments += bio_segments(bio); 525 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO); 526 if (!bvec) 527 return -EIO; 528 cmd->bvec = bvec; 529 530 /* 531 * The bios of the request may be started from the middle of 532 * the 'bvec' because of bio splitting, so we can't directly 533 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment 534 * API will take care of all details for us. 535 */ 536 rq_for_each_segment(tmp, rq, iter) { 537 *bvec = tmp; 538 bvec++; 539 } 540 bvec = cmd->bvec; 541 offset = 0; 542 } else { 543 /* 544 * Same here, this bio may be started from the middle of the 545 * 'bvec' because of bio splitting, so offset from the bvec 546 * must be passed to iov iterator 547 */ 548 offset = bio->bi_iter.bi_bvec_done; 549 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 550 segments = bio_segments(bio); 551 } 552 atomic_set(&cmd->ref, 2); 553 554 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, 555 segments, blk_rq_bytes(rq)); 556 iter.iov_offset = offset; 557 558 cmd->iocb.ki_pos = pos; 559 cmd->iocb.ki_filp = file; 560 cmd->iocb.ki_complete = lo_rw_aio_complete; 561 cmd->iocb.ki_flags = IOCB_DIRECT; 562 if (cmd->css) 563 kthread_associate_blkcg(cmd->css); 564 565 if (rw == WRITE) 566 ret = call_write_iter(file, &cmd->iocb, &iter); 567 else 568 ret = call_read_iter(file, &cmd->iocb, &iter); 569 570 lo_rw_aio_do_completion(cmd); 571 kthread_associate_blkcg(NULL); 572 573 if (ret != -EIOCBQUEUED) 574 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 575 return 0; 576 } 577 578 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 579 { 580 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 581 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 582 583 /* 584 * lo_write_simple and lo_read_simple should have been covered 585 * by io submit style function like lo_rw_aio(), one blocker 586 * is that lo_read_simple() need to call flush_dcache_page after 587 * the page is written from kernel, and it isn't easy to handle 588 * this in io submit style function which submits all segments 589 * of the req at one time. And direct read IO doesn't need to 590 * run flush_dcache_page(). 591 */ 592 switch (req_op(rq)) { 593 case REQ_OP_FLUSH: 594 return lo_req_flush(lo, rq); 595 case REQ_OP_DISCARD: 596 case REQ_OP_WRITE_ZEROES: 597 return lo_discard(lo, rq, pos); 598 case REQ_OP_WRITE: 599 if (lo->transfer) 600 return lo_write_transfer(lo, rq, pos); 601 else if (cmd->use_aio) 602 return lo_rw_aio(lo, cmd, pos, WRITE); 603 else 604 return lo_write_simple(lo, rq, pos); 605 case REQ_OP_READ: 606 if (lo->transfer) 607 return lo_read_transfer(lo, rq, pos); 608 else if (cmd->use_aio) 609 return lo_rw_aio(lo, cmd, pos, READ); 610 else 611 return lo_read_simple(lo, rq, pos); 612 default: 613 WARN_ON_ONCE(1); 614 return -EIO; 615 break; 616 } 617 } 618 619 static inline void loop_update_dio(struct loop_device *lo) 620 { 621 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 622 lo->use_dio); 623 } 624 625 static void loop_reread_partitions(struct loop_device *lo, 626 struct block_device *bdev) 627 { 628 int rc; 629 630 /* 631 * bd_mutex has been held already in release path, so don't 632 * acquire it if this function is called in such case. 633 * 634 * If the reread partition isn't from release path, lo_refcnt 635 * must be at least one and it can only become zero when the 636 * current holder is released. 637 */ 638 if (!atomic_read(&lo->lo_refcnt)) 639 rc = __blkdev_reread_part(bdev); 640 else 641 rc = blkdev_reread_part(bdev); 642 if (rc) 643 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 644 __func__, lo->lo_number, lo->lo_file_name, rc); 645 } 646 647 static inline int is_loop_device(struct file *file) 648 { 649 struct inode *i = file->f_mapping->host; 650 651 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 652 } 653 654 static int loop_validate_file(struct file *file, struct block_device *bdev) 655 { 656 struct inode *inode = file->f_mapping->host; 657 struct file *f = file; 658 659 /* Avoid recursion */ 660 while (is_loop_device(f)) { 661 struct loop_device *l; 662 663 if (f->f_mapping->host->i_bdev == bdev) 664 return -EBADF; 665 666 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 667 if (l->lo_state == Lo_unbound) { 668 return -EINVAL; 669 } 670 f = l->lo_backing_file; 671 } 672 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 673 return -EINVAL; 674 return 0; 675 } 676 677 /* 678 * loop_change_fd switched the backing store of a loopback device to 679 * a new file. This is useful for operating system installers to free up 680 * the original file and in High Availability environments to switch to 681 * an alternative location for the content in case of server meltdown. 682 * This can only work if the loop device is used read-only, and if the 683 * new backing store is the same size and type as the old backing store. 684 */ 685 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 686 unsigned int arg) 687 { 688 struct file *file, *old_file; 689 struct inode *inode; 690 int error; 691 692 error = -ENXIO; 693 if (lo->lo_state != Lo_bound) 694 goto out; 695 696 /* the loop device has to be read-only */ 697 error = -EINVAL; 698 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 699 goto out; 700 701 error = -EBADF; 702 file = fget(arg); 703 if (!file) 704 goto out; 705 706 error = loop_validate_file(file, bdev); 707 if (error) 708 goto out_putf; 709 710 inode = file->f_mapping->host; 711 old_file = lo->lo_backing_file; 712 713 error = -EINVAL; 714 715 /* size of the new backing store needs to be the same */ 716 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 717 goto out_putf; 718 719 /* and ... switch */ 720 blk_mq_freeze_queue(lo->lo_queue); 721 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 722 lo->lo_backing_file = file; 723 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 724 mapping_set_gfp_mask(file->f_mapping, 725 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 726 loop_update_dio(lo); 727 blk_mq_unfreeze_queue(lo->lo_queue); 728 729 fput(old_file); 730 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 731 loop_reread_partitions(lo, bdev); 732 return 0; 733 734 out_putf: 735 fput(file); 736 out: 737 return error; 738 } 739 740 /* loop sysfs attributes */ 741 742 static ssize_t loop_attr_show(struct device *dev, char *page, 743 ssize_t (*callback)(struct loop_device *, char *)) 744 { 745 struct gendisk *disk = dev_to_disk(dev); 746 struct loop_device *lo = disk->private_data; 747 748 return callback(lo, page); 749 } 750 751 #define LOOP_ATTR_RO(_name) \ 752 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 753 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 754 struct device_attribute *attr, char *b) \ 755 { \ 756 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 757 } \ 758 static struct device_attribute loop_attr_##_name = \ 759 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 760 761 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 762 { 763 ssize_t ret; 764 char *p = NULL; 765 766 spin_lock_irq(&lo->lo_lock); 767 if (lo->lo_backing_file) 768 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 769 spin_unlock_irq(&lo->lo_lock); 770 771 if (IS_ERR_OR_NULL(p)) 772 ret = PTR_ERR(p); 773 else { 774 ret = strlen(p); 775 memmove(buf, p, ret); 776 buf[ret++] = '\n'; 777 buf[ret] = 0; 778 } 779 780 return ret; 781 } 782 783 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 784 { 785 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 786 } 787 788 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 789 { 790 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 791 } 792 793 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 794 { 795 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 796 797 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 798 } 799 800 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 801 { 802 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 803 804 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 805 } 806 807 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 808 { 809 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 810 811 return sprintf(buf, "%s\n", dio ? "1" : "0"); 812 } 813 814 LOOP_ATTR_RO(backing_file); 815 LOOP_ATTR_RO(offset); 816 LOOP_ATTR_RO(sizelimit); 817 LOOP_ATTR_RO(autoclear); 818 LOOP_ATTR_RO(partscan); 819 LOOP_ATTR_RO(dio); 820 821 static struct attribute *loop_attrs[] = { 822 &loop_attr_backing_file.attr, 823 &loop_attr_offset.attr, 824 &loop_attr_sizelimit.attr, 825 &loop_attr_autoclear.attr, 826 &loop_attr_partscan.attr, 827 &loop_attr_dio.attr, 828 NULL, 829 }; 830 831 static struct attribute_group loop_attribute_group = { 832 .name = "loop", 833 .attrs= loop_attrs, 834 }; 835 836 static void loop_sysfs_init(struct loop_device *lo) 837 { 838 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 839 &loop_attribute_group); 840 } 841 842 static void loop_sysfs_exit(struct loop_device *lo) 843 { 844 if (lo->sysfs_inited) 845 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 846 &loop_attribute_group); 847 } 848 849 static void loop_config_discard(struct loop_device *lo) 850 { 851 struct file *file = lo->lo_backing_file; 852 struct inode *inode = file->f_mapping->host; 853 struct request_queue *q = lo->lo_queue; 854 855 /* 856 * We use punch hole to reclaim the free space used by the 857 * image a.k.a. discard. However we do not support discard if 858 * encryption is enabled, because it may give an attacker 859 * useful information. 860 */ 861 if ((!file->f_op->fallocate) || 862 lo->lo_encrypt_key_size) { 863 q->limits.discard_granularity = 0; 864 q->limits.discard_alignment = 0; 865 blk_queue_max_discard_sectors(q, 0); 866 blk_queue_max_write_zeroes_sectors(q, 0); 867 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 868 return; 869 } 870 871 q->limits.discard_granularity = inode->i_sb->s_blocksize; 872 q->limits.discard_alignment = 0; 873 874 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 875 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 876 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 877 } 878 879 static void loop_unprepare_queue(struct loop_device *lo) 880 { 881 kthread_flush_worker(&lo->worker); 882 kthread_stop(lo->worker_task); 883 } 884 885 static int loop_kthread_worker_fn(void *worker_ptr) 886 { 887 current->flags |= PF_LESS_THROTTLE; 888 return kthread_worker_fn(worker_ptr); 889 } 890 891 static int loop_prepare_queue(struct loop_device *lo) 892 { 893 kthread_init_worker(&lo->worker); 894 lo->worker_task = kthread_run(loop_kthread_worker_fn, 895 &lo->worker, "loop%d", lo->lo_number); 896 if (IS_ERR(lo->worker_task)) 897 return -ENOMEM; 898 set_user_nice(lo->worker_task, MIN_NICE); 899 return 0; 900 } 901 902 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 903 struct block_device *bdev, unsigned int arg) 904 { 905 struct file *file; 906 struct inode *inode; 907 struct address_space *mapping; 908 int lo_flags = 0; 909 int error; 910 loff_t size; 911 912 /* This is safe, since we have a reference from open(). */ 913 __module_get(THIS_MODULE); 914 915 error = -EBADF; 916 file = fget(arg); 917 if (!file) 918 goto out; 919 920 error = -EBUSY; 921 if (lo->lo_state != Lo_unbound) 922 goto out_putf; 923 924 error = loop_validate_file(file, bdev); 925 if (error) 926 goto out_putf; 927 928 mapping = file->f_mapping; 929 inode = mapping->host; 930 931 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 932 !file->f_op->write_iter) 933 lo_flags |= LO_FLAGS_READ_ONLY; 934 935 error = -EFBIG; 936 size = get_loop_size(lo, file); 937 if ((loff_t)(sector_t)size != size) 938 goto out_putf; 939 error = loop_prepare_queue(lo); 940 if (error) 941 goto out_putf; 942 943 error = 0; 944 945 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 946 947 lo->use_dio = false; 948 lo->lo_device = bdev; 949 lo->lo_flags = lo_flags; 950 lo->lo_backing_file = file; 951 lo->transfer = NULL; 952 lo->ioctl = NULL; 953 lo->lo_sizelimit = 0; 954 lo->old_gfp_mask = mapping_gfp_mask(mapping); 955 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 956 957 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 958 blk_queue_write_cache(lo->lo_queue, true, false); 959 960 loop_update_dio(lo); 961 set_capacity(lo->lo_disk, size); 962 bd_set_size(bdev, size << 9); 963 loop_sysfs_init(lo); 964 /* let user-space know about the new size */ 965 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 966 967 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 968 block_size(inode->i_bdev) : PAGE_SIZE); 969 970 lo->lo_state = Lo_bound; 971 if (part_shift) 972 lo->lo_flags |= LO_FLAGS_PARTSCAN; 973 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 974 loop_reread_partitions(lo, bdev); 975 976 /* Grab the block_device to prevent its destruction after we 977 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). 978 */ 979 bdgrab(bdev); 980 return 0; 981 982 out_putf: 983 fput(file); 984 out: 985 /* This is safe: open() is still holding a reference. */ 986 module_put(THIS_MODULE); 987 return error; 988 } 989 990 static int 991 loop_release_xfer(struct loop_device *lo) 992 { 993 int err = 0; 994 struct loop_func_table *xfer = lo->lo_encryption; 995 996 if (xfer) { 997 if (xfer->release) 998 err = xfer->release(lo); 999 lo->transfer = NULL; 1000 lo->lo_encryption = NULL; 1001 module_put(xfer->owner); 1002 } 1003 return err; 1004 } 1005 1006 static int 1007 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1008 const struct loop_info64 *i) 1009 { 1010 int err = 0; 1011 1012 if (xfer) { 1013 struct module *owner = xfer->owner; 1014 1015 if (!try_module_get(owner)) 1016 return -EINVAL; 1017 if (xfer->init) 1018 err = xfer->init(lo, i); 1019 if (err) 1020 module_put(owner); 1021 else 1022 lo->lo_encryption = xfer; 1023 } 1024 return err; 1025 } 1026 1027 static int loop_clr_fd(struct loop_device *lo) 1028 { 1029 struct file *filp = lo->lo_backing_file; 1030 gfp_t gfp = lo->old_gfp_mask; 1031 struct block_device *bdev = lo->lo_device; 1032 1033 if (lo->lo_state != Lo_bound) 1034 return -ENXIO; 1035 1036 /* 1037 * If we've explicitly asked to tear down the loop device, 1038 * and it has an elevated reference count, set it for auto-teardown when 1039 * the last reference goes away. This stops $!~#$@ udev from 1040 * preventing teardown because it decided that it needs to run blkid on 1041 * the loopback device whenever they appear. xfstests is notorious for 1042 * failing tests because blkid via udev races with a losetup 1043 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1044 * command to fail with EBUSY. 1045 */ 1046 if (atomic_read(&lo->lo_refcnt) > 1) { 1047 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1048 mutex_unlock(&lo->lo_ctl_mutex); 1049 return 0; 1050 } 1051 1052 if (filp == NULL) 1053 return -EINVAL; 1054 1055 /* freeze request queue during the transition */ 1056 blk_mq_freeze_queue(lo->lo_queue); 1057 1058 spin_lock_irq(&lo->lo_lock); 1059 lo->lo_state = Lo_rundown; 1060 lo->lo_backing_file = NULL; 1061 spin_unlock_irq(&lo->lo_lock); 1062 1063 loop_release_xfer(lo); 1064 lo->transfer = NULL; 1065 lo->ioctl = NULL; 1066 lo->lo_device = NULL; 1067 lo->lo_encryption = NULL; 1068 lo->lo_offset = 0; 1069 lo->lo_sizelimit = 0; 1070 lo->lo_encrypt_key_size = 0; 1071 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1072 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1073 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1074 blk_queue_logical_block_size(lo->lo_queue, 512); 1075 blk_queue_physical_block_size(lo->lo_queue, 512); 1076 blk_queue_io_min(lo->lo_queue, 512); 1077 if (bdev) { 1078 bdput(bdev); 1079 invalidate_bdev(bdev); 1080 bdev->bd_inode->i_mapping->wb_err = 0; 1081 } 1082 set_capacity(lo->lo_disk, 0); 1083 loop_sysfs_exit(lo); 1084 if (bdev) { 1085 bd_set_size(bdev, 0); 1086 /* let user-space know about this change */ 1087 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1088 } 1089 mapping_set_gfp_mask(filp->f_mapping, gfp); 1090 lo->lo_state = Lo_unbound; 1091 /* This is safe: open() is still holding a reference. */ 1092 module_put(THIS_MODULE); 1093 blk_mq_unfreeze_queue(lo->lo_queue); 1094 1095 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 1096 loop_reread_partitions(lo, bdev); 1097 lo->lo_flags = 0; 1098 if (!part_shift) 1099 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1100 loop_unprepare_queue(lo); 1101 mutex_unlock(&lo->lo_ctl_mutex); 1102 /* 1103 * Need not hold lo_ctl_mutex to fput backing file. 1104 * Calling fput holding lo_ctl_mutex triggers a circular 1105 * lock dependency possibility warning as fput can take 1106 * bd_mutex which is usually taken before lo_ctl_mutex. 1107 */ 1108 fput(filp); 1109 return 0; 1110 } 1111 1112 static int 1113 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1114 { 1115 int err; 1116 struct loop_func_table *xfer; 1117 kuid_t uid = current_uid(); 1118 1119 if (lo->lo_encrypt_key_size && 1120 !uid_eq(lo->lo_key_owner, uid) && 1121 !capable(CAP_SYS_ADMIN)) 1122 return -EPERM; 1123 if (lo->lo_state != Lo_bound) 1124 return -ENXIO; 1125 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 1126 return -EINVAL; 1127 1128 /* I/O need to be drained during transfer transition */ 1129 blk_mq_freeze_queue(lo->lo_queue); 1130 1131 err = loop_release_xfer(lo); 1132 if (err) 1133 goto exit; 1134 1135 if (info->lo_encrypt_type) { 1136 unsigned int type = info->lo_encrypt_type; 1137 1138 if (type >= MAX_LO_CRYPT) { 1139 err = -EINVAL; 1140 goto exit; 1141 } 1142 xfer = xfer_funcs[type]; 1143 if (xfer == NULL) { 1144 err = -EINVAL; 1145 goto exit; 1146 } 1147 } else 1148 xfer = NULL; 1149 1150 err = loop_init_xfer(lo, xfer, info); 1151 if (err) 1152 goto exit; 1153 1154 if (lo->lo_offset != info->lo_offset || 1155 lo->lo_sizelimit != info->lo_sizelimit) { 1156 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1157 err = -EFBIG; 1158 goto exit; 1159 } 1160 } 1161 1162 loop_config_discard(lo); 1163 1164 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1165 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1166 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1167 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1168 1169 if (!xfer) 1170 xfer = &none_funcs; 1171 lo->transfer = xfer->transfer; 1172 lo->ioctl = xfer->ioctl; 1173 1174 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1175 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1176 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1177 1178 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1179 lo->lo_init[0] = info->lo_init[0]; 1180 lo->lo_init[1] = info->lo_init[1]; 1181 if (info->lo_encrypt_key_size) { 1182 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1183 info->lo_encrypt_key_size); 1184 lo->lo_key_owner = uid; 1185 } 1186 1187 /* update dio if lo_offset or transfer is changed */ 1188 __loop_update_dio(lo, lo->use_dio); 1189 1190 exit: 1191 blk_mq_unfreeze_queue(lo->lo_queue); 1192 1193 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1194 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1195 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1196 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1197 loop_reread_partitions(lo, lo->lo_device); 1198 } 1199 1200 return err; 1201 } 1202 1203 static int 1204 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1205 { 1206 struct file *file; 1207 struct kstat stat; 1208 int ret; 1209 1210 if (lo->lo_state != Lo_bound) { 1211 mutex_unlock(&lo->lo_ctl_mutex); 1212 return -ENXIO; 1213 } 1214 1215 memset(info, 0, sizeof(*info)); 1216 info->lo_number = lo->lo_number; 1217 info->lo_offset = lo->lo_offset; 1218 info->lo_sizelimit = lo->lo_sizelimit; 1219 info->lo_flags = lo->lo_flags; 1220 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1221 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1222 info->lo_encrypt_type = 1223 lo->lo_encryption ? lo->lo_encryption->number : 0; 1224 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1225 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1226 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1227 lo->lo_encrypt_key_size); 1228 } 1229 1230 /* Drop lo_ctl_mutex while we call into the filesystem. */ 1231 file = get_file(lo->lo_backing_file); 1232 mutex_unlock(&lo->lo_ctl_mutex); 1233 ret = vfs_getattr(&file->f_path, &stat, STATX_INO, 1234 AT_STATX_SYNC_AS_STAT); 1235 if (!ret) { 1236 info->lo_device = huge_encode_dev(stat.dev); 1237 info->lo_inode = stat.ino; 1238 info->lo_rdevice = huge_encode_dev(stat.rdev); 1239 } 1240 fput(file); 1241 return ret; 1242 } 1243 1244 static void 1245 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1246 { 1247 memset(info64, 0, sizeof(*info64)); 1248 info64->lo_number = info->lo_number; 1249 info64->lo_device = info->lo_device; 1250 info64->lo_inode = info->lo_inode; 1251 info64->lo_rdevice = info->lo_rdevice; 1252 info64->lo_offset = info->lo_offset; 1253 info64->lo_sizelimit = 0; 1254 info64->lo_encrypt_type = info->lo_encrypt_type; 1255 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1256 info64->lo_flags = info->lo_flags; 1257 info64->lo_init[0] = info->lo_init[0]; 1258 info64->lo_init[1] = info->lo_init[1]; 1259 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1260 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1261 else 1262 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1263 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1264 } 1265 1266 static int 1267 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1268 { 1269 memset(info, 0, sizeof(*info)); 1270 info->lo_number = info64->lo_number; 1271 info->lo_device = info64->lo_device; 1272 info->lo_inode = info64->lo_inode; 1273 info->lo_rdevice = info64->lo_rdevice; 1274 info->lo_offset = info64->lo_offset; 1275 info->lo_encrypt_type = info64->lo_encrypt_type; 1276 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1277 info->lo_flags = info64->lo_flags; 1278 info->lo_init[0] = info64->lo_init[0]; 1279 info->lo_init[1] = info64->lo_init[1]; 1280 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1281 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1282 else 1283 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1284 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1285 1286 /* error in case values were truncated */ 1287 if (info->lo_device != info64->lo_device || 1288 info->lo_rdevice != info64->lo_rdevice || 1289 info->lo_inode != info64->lo_inode || 1290 info->lo_offset != info64->lo_offset) 1291 return -EOVERFLOW; 1292 1293 return 0; 1294 } 1295 1296 static int 1297 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1298 { 1299 struct loop_info info; 1300 struct loop_info64 info64; 1301 1302 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1303 return -EFAULT; 1304 loop_info64_from_old(&info, &info64); 1305 return loop_set_status(lo, &info64); 1306 } 1307 1308 static int 1309 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1310 { 1311 struct loop_info64 info64; 1312 1313 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1314 return -EFAULT; 1315 return loop_set_status(lo, &info64); 1316 } 1317 1318 static int 1319 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1320 struct loop_info info; 1321 struct loop_info64 info64; 1322 int err; 1323 1324 if (!arg) { 1325 mutex_unlock(&lo->lo_ctl_mutex); 1326 return -EINVAL; 1327 } 1328 err = loop_get_status(lo, &info64); 1329 if (!err) 1330 err = loop_info64_to_old(&info64, &info); 1331 if (!err && copy_to_user(arg, &info, sizeof(info))) 1332 err = -EFAULT; 1333 1334 return err; 1335 } 1336 1337 static int 1338 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1339 struct loop_info64 info64; 1340 int err; 1341 1342 if (!arg) { 1343 mutex_unlock(&lo->lo_ctl_mutex); 1344 return -EINVAL; 1345 } 1346 err = loop_get_status(lo, &info64); 1347 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1348 err = -EFAULT; 1349 1350 return err; 1351 } 1352 1353 static int loop_set_capacity(struct loop_device *lo) 1354 { 1355 if (unlikely(lo->lo_state != Lo_bound)) 1356 return -ENXIO; 1357 1358 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1359 } 1360 1361 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1362 { 1363 int error = -ENXIO; 1364 if (lo->lo_state != Lo_bound) 1365 goto out; 1366 1367 __loop_update_dio(lo, !!arg); 1368 if (lo->use_dio == !!arg) 1369 return 0; 1370 error = -EINVAL; 1371 out: 1372 return error; 1373 } 1374 1375 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1376 { 1377 if (lo->lo_state != Lo_bound) 1378 return -ENXIO; 1379 1380 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1381 return -EINVAL; 1382 1383 blk_mq_freeze_queue(lo->lo_queue); 1384 1385 blk_queue_logical_block_size(lo->lo_queue, arg); 1386 blk_queue_physical_block_size(lo->lo_queue, arg); 1387 blk_queue_io_min(lo->lo_queue, arg); 1388 loop_update_dio(lo); 1389 1390 blk_mq_unfreeze_queue(lo->lo_queue); 1391 1392 return 0; 1393 } 1394 1395 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1396 unsigned int cmd, unsigned long arg) 1397 { 1398 struct loop_device *lo = bdev->bd_disk->private_data; 1399 int err; 1400 1401 err = mutex_lock_killable_nested(&lo->lo_ctl_mutex, 1); 1402 if (err) 1403 goto out_unlocked; 1404 1405 switch (cmd) { 1406 case LOOP_SET_FD: 1407 err = loop_set_fd(lo, mode, bdev, arg); 1408 break; 1409 case LOOP_CHANGE_FD: 1410 err = loop_change_fd(lo, bdev, arg); 1411 break; 1412 case LOOP_CLR_FD: 1413 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ 1414 err = loop_clr_fd(lo); 1415 if (!err) 1416 goto out_unlocked; 1417 break; 1418 case LOOP_SET_STATUS: 1419 err = -EPERM; 1420 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1421 err = loop_set_status_old(lo, 1422 (struct loop_info __user *)arg); 1423 break; 1424 case LOOP_GET_STATUS: 1425 err = loop_get_status_old(lo, (struct loop_info __user *) arg); 1426 /* loop_get_status() unlocks lo_ctl_mutex */ 1427 goto out_unlocked; 1428 case LOOP_SET_STATUS64: 1429 err = -EPERM; 1430 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1431 err = loop_set_status64(lo, 1432 (struct loop_info64 __user *) arg); 1433 break; 1434 case LOOP_GET_STATUS64: 1435 err = loop_get_status64(lo, (struct loop_info64 __user *) arg); 1436 /* loop_get_status() unlocks lo_ctl_mutex */ 1437 goto out_unlocked; 1438 case LOOP_SET_CAPACITY: 1439 err = -EPERM; 1440 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1441 err = loop_set_capacity(lo); 1442 break; 1443 case LOOP_SET_DIRECT_IO: 1444 err = -EPERM; 1445 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1446 err = loop_set_dio(lo, arg); 1447 break; 1448 case LOOP_SET_BLOCK_SIZE: 1449 err = -EPERM; 1450 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) 1451 err = loop_set_block_size(lo, arg); 1452 break; 1453 default: 1454 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1455 } 1456 mutex_unlock(&lo->lo_ctl_mutex); 1457 1458 out_unlocked: 1459 return err; 1460 } 1461 1462 #ifdef CONFIG_COMPAT 1463 struct compat_loop_info { 1464 compat_int_t lo_number; /* ioctl r/o */ 1465 compat_dev_t lo_device; /* ioctl r/o */ 1466 compat_ulong_t lo_inode; /* ioctl r/o */ 1467 compat_dev_t lo_rdevice; /* ioctl r/o */ 1468 compat_int_t lo_offset; 1469 compat_int_t lo_encrypt_type; 1470 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1471 compat_int_t lo_flags; /* ioctl r/o */ 1472 char lo_name[LO_NAME_SIZE]; 1473 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1474 compat_ulong_t lo_init[2]; 1475 char reserved[4]; 1476 }; 1477 1478 /* 1479 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1480 * - noinlined to reduce stack space usage in main part of driver 1481 */ 1482 static noinline int 1483 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1484 struct loop_info64 *info64) 1485 { 1486 struct compat_loop_info info; 1487 1488 if (copy_from_user(&info, arg, sizeof(info))) 1489 return -EFAULT; 1490 1491 memset(info64, 0, sizeof(*info64)); 1492 info64->lo_number = info.lo_number; 1493 info64->lo_device = info.lo_device; 1494 info64->lo_inode = info.lo_inode; 1495 info64->lo_rdevice = info.lo_rdevice; 1496 info64->lo_offset = info.lo_offset; 1497 info64->lo_sizelimit = 0; 1498 info64->lo_encrypt_type = info.lo_encrypt_type; 1499 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1500 info64->lo_flags = info.lo_flags; 1501 info64->lo_init[0] = info.lo_init[0]; 1502 info64->lo_init[1] = info.lo_init[1]; 1503 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1504 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1505 else 1506 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1507 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1508 return 0; 1509 } 1510 1511 /* 1512 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1513 * - noinlined to reduce stack space usage in main part of driver 1514 */ 1515 static noinline int 1516 loop_info64_to_compat(const struct loop_info64 *info64, 1517 struct compat_loop_info __user *arg) 1518 { 1519 struct compat_loop_info info; 1520 1521 memset(&info, 0, sizeof(info)); 1522 info.lo_number = info64->lo_number; 1523 info.lo_device = info64->lo_device; 1524 info.lo_inode = info64->lo_inode; 1525 info.lo_rdevice = info64->lo_rdevice; 1526 info.lo_offset = info64->lo_offset; 1527 info.lo_encrypt_type = info64->lo_encrypt_type; 1528 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1529 info.lo_flags = info64->lo_flags; 1530 info.lo_init[0] = info64->lo_init[0]; 1531 info.lo_init[1] = info64->lo_init[1]; 1532 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1533 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1534 else 1535 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1536 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1537 1538 /* error in case values were truncated */ 1539 if (info.lo_device != info64->lo_device || 1540 info.lo_rdevice != info64->lo_rdevice || 1541 info.lo_inode != info64->lo_inode || 1542 info.lo_offset != info64->lo_offset || 1543 info.lo_init[0] != info64->lo_init[0] || 1544 info.lo_init[1] != info64->lo_init[1]) 1545 return -EOVERFLOW; 1546 1547 if (copy_to_user(arg, &info, sizeof(info))) 1548 return -EFAULT; 1549 return 0; 1550 } 1551 1552 static int 1553 loop_set_status_compat(struct loop_device *lo, 1554 const struct compat_loop_info __user *arg) 1555 { 1556 struct loop_info64 info64; 1557 int ret; 1558 1559 ret = loop_info64_from_compat(arg, &info64); 1560 if (ret < 0) 1561 return ret; 1562 return loop_set_status(lo, &info64); 1563 } 1564 1565 static int 1566 loop_get_status_compat(struct loop_device *lo, 1567 struct compat_loop_info __user *arg) 1568 { 1569 struct loop_info64 info64; 1570 int err; 1571 1572 if (!arg) { 1573 mutex_unlock(&lo->lo_ctl_mutex); 1574 return -EINVAL; 1575 } 1576 err = loop_get_status(lo, &info64); 1577 if (!err) 1578 err = loop_info64_to_compat(&info64, arg); 1579 return err; 1580 } 1581 1582 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1583 unsigned int cmd, unsigned long arg) 1584 { 1585 struct loop_device *lo = bdev->bd_disk->private_data; 1586 int err; 1587 1588 switch(cmd) { 1589 case LOOP_SET_STATUS: 1590 err = mutex_lock_killable(&lo->lo_ctl_mutex); 1591 if (!err) { 1592 err = loop_set_status_compat(lo, 1593 (const struct compat_loop_info __user *)arg); 1594 mutex_unlock(&lo->lo_ctl_mutex); 1595 } 1596 break; 1597 case LOOP_GET_STATUS: 1598 err = mutex_lock_killable(&lo->lo_ctl_mutex); 1599 if (!err) { 1600 err = loop_get_status_compat(lo, 1601 (struct compat_loop_info __user *)arg); 1602 /* loop_get_status() unlocks lo_ctl_mutex */ 1603 } 1604 break; 1605 case LOOP_SET_CAPACITY: 1606 case LOOP_CLR_FD: 1607 case LOOP_GET_STATUS64: 1608 case LOOP_SET_STATUS64: 1609 arg = (unsigned long) compat_ptr(arg); 1610 case LOOP_SET_FD: 1611 case LOOP_CHANGE_FD: 1612 err = lo_ioctl(bdev, mode, cmd, arg); 1613 break; 1614 default: 1615 err = -ENOIOCTLCMD; 1616 break; 1617 } 1618 return err; 1619 } 1620 #endif 1621 1622 static int lo_open(struct block_device *bdev, fmode_t mode) 1623 { 1624 struct loop_device *lo; 1625 int err = 0; 1626 1627 mutex_lock(&loop_index_mutex); 1628 lo = bdev->bd_disk->private_data; 1629 if (!lo) { 1630 err = -ENXIO; 1631 goto out; 1632 } 1633 1634 atomic_inc(&lo->lo_refcnt); 1635 out: 1636 mutex_unlock(&loop_index_mutex); 1637 return err; 1638 } 1639 1640 static void __lo_release(struct loop_device *lo) 1641 { 1642 int err; 1643 1644 if (atomic_dec_return(&lo->lo_refcnt)) 1645 return; 1646 1647 mutex_lock(&lo->lo_ctl_mutex); 1648 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1649 /* 1650 * In autoclear mode, stop the loop thread 1651 * and remove configuration after last close. 1652 */ 1653 err = loop_clr_fd(lo); 1654 if (!err) 1655 return; 1656 } else if (lo->lo_state == Lo_bound) { 1657 /* 1658 * Otherwise keep thread (if running) and config, 1659 * but flush possible ongoing bios in thread. 1660 */ 1661 blk_mq_freeze_queue(lo->lo_queue); 1662 blk_mq_unfreeze_queue(lo->lo_queue); 1663 } 1664 1665 mutex_unlock(&lo->lo_ctl_mutex); 1666 } 1667 1668 static void lo_release(struct gendisk *disk, fmode_t mode) 1669 { 1670 mutex_lock(&loop_index_mutex); 1671 __lo_release(disk->private_data); 1672 mutex_unlock(&loop_index_mutex); 1673 } 1674 1675 static const struct block_device_operations lo_fops = { 1676 .owner = THIS_MODULE, 1677 .open = lo_open, 1678 .release = lo_release, 1679 .ioctl = lo_ioctl, 1680 #ifdef CONFIG_COMPAT 1681 .compat_ioctl = lo_compat_ioctl, 1682 #endif 1683 }; 1684 1685 /* 1686 * And now the modules code and kernel interface. 1687 */ 1688 static int max_loop; 1689 module_param(max_loop, int, 0444); 1690 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1691 module_param(max_part, int, 0444); 1692 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1693 MODULE_LICENSE("GPL"); 1694 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1695 1696 int loop_register_transfer(struct loop_func_table *funcs) 1697 { 1698 unsigned int n = funcs->number; 1699 1700 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1701 return -EINVAL; 1702 xfer_funcs[n] = funcs; 1703 return 0; 1704 } 1705 1706 static int unregister_transfer_cb(int id, void *ptr, void *data) 1707 { 1708 struct loop_device *lo = ptr; 1709 struct loop_func_table *xfer = data; 1710 1711 mutex_lock(&lo->lo_ctl_mutex); 1712 if (lo->lo_encryption == xfer) 1713 loop_release_xfer(lo); 1714 mutex_unlock(&lo->lo_ctl_mutex); 1715 return 0; 1716 } 1717 1718 int loop_unregister_transfer(int number) 1719 { 1720 unsigned int n = number; 1721 struct loop_func_table *xfer; 1722 1723 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1724 return -EINVAL; 1725 1726 xfer_funcs[n] = NULL; 1727 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1728 return 0; 1729 } 1730 1731 EXPORT_SYMBOL(loop_register_transfer); 1732 EXPORT_SYMBOL(loop_unregister_transfer); 1733 1734 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1735 const struct blk_mq_queue_data *bd) 1736 { 1737 struct request *rq = bd->rq; 1738 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1739 struct loop_device *lo = rq->q->queuedata; 1740 1741 blk_mq_start_request(rq); 1742 1743 if (lo->lo_state != Lo_bound) 1744 return BLK_STS_IOERR; 1745 1746 switch (req_op(rq)) { 1747 case REQ_OP_FLUSH: 1748 case REQ_OP_DISCARD: 1749 case REQ_OP_WRITE_ZEROES: 1750 cmd->use_aio = false; 1751 break; 1752 default: 1753 cmd->use_aio = lo->use_dio; 1754 break; 1755 } 1756 1757 /* always use the first bio's css */ 1758 #ifdef CONFIG_BLK_CGROUP 1759 if (cmd->use_aio && rq->bio && rq->bio->bi_css) { 1760 cmd->css = rq->bio->bi_css; 1761 css_get(cmd->css); 1762 } else 1763 #endif 1764 cmd->css = NULL; 1765 kthread_queue_work(&lo->worker, &cmd->work); 1766 1767 return BLK_STS_OK; 1768 } 1769 1770 static void loop_handle_cmd(struct loop_cmd *cmd) 1771 { 1772 struct request *rq = blk_mq_rq_from_pdu(cmd); 1773 const bool write = op_is_write(req_op(rq)); 1774 struct loop_device *lo = rq->q->queuedata; 1775 int ret = 0; 1776 1777 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1778 ret = -EIO; 1779 goto failed; 1780 } 1781 1782 ret = do_req_filebacked(lo, rq); 1783 failed: 1784 /* complete non-aio request */ 1785 if (!cmd->use_aio || ret) { 1786 cmd->ret = ret ? -EIO : 0; 1787 blk_mq_complete_request(rq); 1788 } 1789 } 1790 1791 static void loop_queue_work(struct kthread_work *work) 1792 { 1793 struct loop_cmd *cmd = 1794 container_of(work, struct loop_cmd, work); 1795 1796 loop_handle_cmd(cmd); 1797 } 1798 1799 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1800 unsigned int hctx_idx, unsigned int numa_node) 1801 { 1802 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1803 1804 kthread_init_work(&cmd->work, loop_queue_work); 1805 return 0; 1806 } 1807 1808 static const struct blk_mq_ops loop_mq_ops = { 1809 .queue_rq = loop_queue_rq, 1810 .init_request = loop_init_request, 1811 .complete = lo_complete_rq, 1812 }; 1813 1814 static int loop_add(struct loop_device **l, int i) 1815 { 1816 struct loop_device *lo; 1817 struct gendisk *disk; 1818 int err; 1819 1820 err = -ENOMEM; 1821 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1822 if (!lo) 1823 goto out; 1824 1825 lo->lo_state = Lo_unbound; 1826 1827 /* allocate id, if @id >= 0, we're requesting that specific id */ 1828 if (i >= 0) { 1829 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1830 if (err == -ENOSPC) 1831 err = -EEXIST; 1832 } else { 1833 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1834 } 1835 if (err < 0) 1836 goto out_free_dev; 1837 i = err; 1838 1839 err = -ENOMEM; 1840 lo->tag_set.ops = &loop_mq_ops; 1841 lo->tag_set.nr_hw_queues = 1; 1842 lo->tag_set.queue_depth = 128; 1843 lo->tag_set.numa_node = NUMA_NO_NODE; 1844 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1845 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1846 lo->tag_set.driver_data = lo; 1847 1848 err = blk_mq_alloc_tag_set(&lo->tag_set); 1849 if (err) 1850 goto out_free_idr; 1851 1852 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1853 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1854 err = PTR_ERR(lo->lo_queue); 1855 goto out_cleanup_tags; 1856 } 1857 lo->lo_queue->queuedata = lo; 1858 1859 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1860 1861 /* 1862 * By default, we do buffer IO, so it doesn't make sense to enable 1863 * merge because the I/O submitted to backing file is handled page by 1864 * page. For directio mode, merge does help to dispatch bigger request 1865 * to underlayer disk. We will enable merge once directio is enabled. 1866 */ 1867 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1868 1869 err = -ENOMEM; 1870 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1871 if (!disk) 1872 goto out_free_queue; 1873 1874 /* 1875 * Disable partition scanning by default. The in-kernel partition 1876 * scanning can be requested individually per-device during its 1877 * setup. Userspace can always add and remove partitions from all 1878 * devices. The needed partition minors are allocated from the 1879 * extended minor space, the main loop device numbers will continue 1880 * to match the loop minors, regardless of the number of partitions 1881 * used. 1882 * 1883 * If max_part is given, partition scanning is globally enabled for 1884 * all loop devices. The minors for the main loop devices will be 1885 * multiples of max_part. 1886 * 1887 * Note: Global-for-all-devices, set-only-at-init, read-only module 1888 * parameteters like 'max_loop' and 'max_part' make things needlessly 1889 * complicated, are too static, inflexible and may surprise 1890 * userspace tools. Parameters like this in general should be avoided. 1891 */ 1892 if (!part_shift) 1893 disk->flags |= GENHD_FL_NO_PART_SCAN; 1894 disk->flags |= GENHD_FL_EXT_DEVT; 1895 mutex_init(&lo->lo_ctl_mutex); 1896 atomic_set(&lo->lo_refcnt, 0); 1897 lo->lo_number = i; 1898 spin_lock_init(&lo->lo_lock); 1899 disk->major = LOOP_MAJOR; 1900 disk->first_minor = i << part_shift; 1901 disk->fops = &lo_fops; 1902 disk->private_data = lo; 1903 disk->queue = lo->lo_queue; 1904 sprintf(disk->disk_name, "loop%d", i); 1905 add_disk(disk); 1906 *l = lo; 1907 return lo->lo_number; 1908 1909 out_free_queue: 1910 blk_cleanup_queue(lo->lo_queue); 1911 out_cleanup_tags: 1912 blk_mq_free_tag_set(&lo->tag_set); 1913 out_free_idr: 1914 idr_remove(&loop_index_idr, i); 1915 out_free_dev: 1916 kfree(lo); 1917 out: 1918 return err; 1919 } 1920 1921 static void loop_remove(struct loop_device *lo) 1922 { 1923 del_gendisk(lo->lo_disk); 1924 blk_cleanup_queue(lo->lo_queue); 1925 blk_mq_free_tag_set(&lo->tag_set); 1926 put_disk(lo->lo_disk); 1927 kfree(lo); 1928 } 1929 1930 static int find_free_cb(int id, void *ptr, void *data) 1931 { 1932 struct loop_device *lo = ptr; 1933 struct loop_device **l = data; 1934 1935 if (lo->lo_state == Lo_unbound) { 1936 *l = lo; 1937 return 1; 1938 } 1939 return 0; 1940 } 1941 1942 static int loop_lookup(struct loop_device **l, int i) 1943 { 1944 struct loop_device *lo; 1945 int ret = -ENODEV; 1946 1947 if (i < 0) { 1948 int err; 1949 1950 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1951 if (err == 1) { 1952 *l = lo; 1953 ret = lo->lo_number; 1954 } 1955 goto out; 1956 } 1957 1958 /* lookup and return a specific i */ 1959 lo = idr_find(&loop_index_idr, i); 1960 if (lo) { 1961 *l = lo; 1962 ret = lo->lo_number; 1963 } 1964 out: 1965 return ret; 1966 } 1967 1968 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 1969 { 1970 struct loop_device *lo; 1971 struct kobject *kobj; 1972 int err; 1973 1974 mutex_lock(&loop_index_mutex); 1975 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 1976 if (err < 0) 1977 err = loop_add(&lo, MINOR(dev) >> part_shift); 1978 if (err < 0) 1979 kobj = NULL; 1980 else 1981 kobj = get_disk_and_module(lo->lo_disk); 1982 mutex_unlock(&loop_index_mutex); 1983 1984 *part = 0; 1985 return kobj; 1986 } 1987 1988 static long loop_control_ioctl(struct file *file, unsigned int cmd, 1989 unsigned long parm) 1990 { 1991 struct loop_device *lo; 1992 int ret = -ENOSYS; 1993 1994 mutex_lock(&loop_index_mutex); 1995 switch (cmd) { 1996 case LOOP_CTL_ADD: 1997 ret = loop_lookup(&lo, parm); 1998 if (ret >= 0) { 1999 ret = -EEXIST; 2000 break; 2001 } 2002 ret = loop_add(&lo, parm); 2003 break; 2004 case LOOP_CTL_REMOVE: 2005 ret = loop_lookup(&lo, parm); 2006 if (ret < 0) 2007 break; 2008 ret = mutex_lock_killable(&lo->lo_ctl_mutex); 2009 if (ret) 2010 break; 2011 if (lo->lo_state != Lo_unbound) { 2012 ret = -EBUSY; 2013 mutex_unlock(&lo->lo_ctl_mutex); 2014 break; 2015 } 2016 if (atomic_read(&lo->lo_refcnt) > 0) { 2017 ret = -EBUSY; 2018 mutex_unlock(&lo->lo_ctl_mutex); 2019 break; 2020 } 2021 lo->lo_disk->private_data = NULL; 2022 mutex_unlock(&lo->lo_ctl_mutex); 2023 idr_remove(&loop_index_idr, lo->lo_number); 2024 loop_remove(lo); 2025 break; 2026 case LOOP_CTL_GET_FREE: 2027 ret = loop_lookup(&lo, -1); 2028 if (ret >= 0) 2029 break; 2030 ret = loop_add(&lo, -1); 2031 } 2032 mutex_unlock(&loop_index_mutex); 2033 2034 return ret; 2035 } 2036 2037 static const struct file_operations loop_ctl_fops = { 2038 .open = nonseekable_open, 2039 .unlocked_ioctl = loop_control_ioctl, 2040 .compat_ioctl = loop_control_ioctl, 2041 .owner = THIS_MODULE, 2042 .llseek = noop_llseek, 2043 }; 2044 2045 static struct miscdevice loop_misc = { 2046 .minor = LOOP_CTRL_MINOR, 2047 .name = "loop-control", 2048 .fops = &loop_ctl_fops, 2049 }; 2050 2051 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2052 MODULE_ALIAS("devname:loop-control"); 2053 2054 static int __init loop_init(void) 2055 { 2056 int i, nr; 2057 unsigned long range; 2058 struct loop_device *lo; 2059 int err; 2060 2061 part_shift = 0; 2062 if (max_part > 0) { 2063 part_shift = fls(max_part); 2064 2065 /* 2066 * Adjust max_part according to part_shift as it is exported 2067 * to user space so that user can decide correct minor number 2068 * if [s]he want to create more devices. 2069 * 2070 * Note that -1 is required because partition 0 is reserved 2071 * for the whole disk. 2072 */ 2073 max_part = (1UL << part_shift) - 1; 2074 } 2075 2076 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2077 err = -EINVAL; 2078 goto err_out; 2079 } 2080 2081 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2082 err = -EINVAL; 2083 goto err_out; 2084 } 2085 2086 /* 2087 * If max_loop is specified, create that many devices upfront. 2088 * This also becomes a hard limit. If max_loop is not specified, 2089 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2090 * init time. Loop devices can be requested on-demand with the 2091 * /dev/loop-control interface, or be instantiated by accessing 2092 * a 'dead' device node. 2093 */ 2094 if (max_loop) { 2095 nr = max_loop; 2096 range = max_loop << part_shift; 2097 } else { 2098 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2099 range = 1UL << MINORBITS; 2100 } 2101 2102 err = misc_register(&loop_misc); 2103 if (err < 0) 2104 goto err_out; 2105 2106 2107 if (register_blkdev(LOOP_MAJOR, "loop")) { 2108 err = -EIO; 2109 goto misc_out; 2110 } 2111 2112 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2113 THIS_MODULE, loop_probe, NULL, NULL); 2114 2115 /* pre-create number of devices given by config or max_loop */ 2116 mutex_lock(&loop_index_mutex); 2117 for (i = 0; i < nr; i++) 2118 loop_add(&lo, i); 2119 mutex_unlock(&loop_index_mutex); 2120 2121 printk(KERN_INFO "loop: module loaded\n"); 2122 return 0; 2123 2124 misc_out: 2125 misc_deregister(&loop_misc); 2126 err_out: 2127 return err; 2128 } 2129 2130 static int loop_exit_cb(int id, void *ptr, void *data) 2131 { 2132 struct loop_device *lo = ptr; 2133 2134 loop_remove(lo); 2135 return 0; 2136 } 2137 2138 static void __exit loop_exit(void) 2139 { 2140 unsigned long range; 2141 2142 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2143 2144 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2145 idr_destroy(&loop_index_idr); 2146 2147 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2148 unregister_blkdev(LOOP_MAJOR, "loop"); 2149 2150 misc_deregister(&loop_misc); 2151 } 2152 2153 module_init(loop_init); 2154 module_exit(loop_exit); 2155 2156 #ifndef MODULE 2157 static int __init max_loop_setup(char *str) 2158 { 2159 max_loop = simple_strtol(str, NULL, 0); 2160 return 1; 2161 } 2162 2163 __setup("max_loop=", max_loop_setup); 2164 #endif 2165