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 <linux/ioprio.h> 80 #include <linux/blk-cgroup.h> 81 82 #include "loop.h" 83 84 #include <linux/uaccess.h> 85 86 static DEFINE_IDR(loop_index_idr); 87 static DEFINE_MUTEX(loop_ctl_mutex); 88 89 static int max_part; 90 static int part_shift; 91 92 static int transfer_xor(struct loop_device *lo, int cmd, 93 struct page *raw_page, unsigned raw_off, 94 struct page *loop_page, unsigned loop_off, 95 int size, sector_t real_block) 96 { 97 char *raw_buf = kmap_atomic(raw_page) + raw_off; 98 char *loop_buf = kmap_atomic(loop_page) + loop_off; 99 char *in, *out, *key; 100 int i, keysize; 101 102 if (cmd == READ) { 103 in = raw_buf; 104 out = loop_buf; 105 } else { 106 in = loop_buf; 107 out = raw_buf; 108 } 109 110 key = lo->lo_encrypt_key; 111 keysize = lo->lo_encrypt_key_size; 112 for (i = 0; i < size; i++) 113 *out++ = *in++ ^ key[(i & 511) % keysize]; 114 115 kunmap_atomic(loop_buf); 116 kunmap_atomic(raw_buf); 117 cond_resched(); 118 return 0; 119 } 120 121 static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 122 { 123 if (unlikely(info->lo_encrypt_key_size <= 0)) 124 return -EINVAL; 125 return 0; 126 } 127 128 static struct loop_func_table none_funcs = { 129 .number = LO_CRYPT_NONE, 130 }; 131 132 static struct loop_func_table xor_funcs = { 133 .number = LO_CRYPT_XOR, 134 .transfer = transfer_xor, 135 .init = xor_init 136 }; 137 138 /* xfer_funcs[0] is special - its release function is never called */ 139 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 140 &none_funcs, 141 &xor_funcs 142 }; 143 144 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 145 { 146 loff_t loopsize; 147 148 /* Compute loopsize in bytes */ 149 loopsize = i_size_read(file->f_mapping->host); 150 if (offset > 0) 151 loopsize -= offset; 152 /* offset is beyond i_size, weird but possible */ 153 if (loopsize < 0) 154 return 0; 155 156 if (sizelimit > 0 && sizelimit < loopsize) 157 loopsize = sizelimit; 158 /* 159 * Unfortunately, if we want to do I/O on the device, 160 * the number of 512-byte sectors has to fit into a sector_t. 161 */ 162 return loopsize >> 9; 163 } 164 165 static loff_t get_loop_size(struct loop_device *lo, struct file *file) 166 { 167 return get_size(lo->lo_offset, lo->lo_sizelimit, file); 168 } 169 170 static void __loop_update_dio(struct loop_device *lo, bool dio) 171 { 172 struct file *file = lo->lo_backing_file; 173 struct address_space *mapping = file->f_mapping; 174 struct inode *inode = mapping->host; 175 unsigned short sb_bsize = 0; 176 unsigned dio_align = 0; 177 bool use_dio; 178 179 if (inode->i_sb->s_bdev) { 180 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 181 dio_align = sb_bsize - 1; 182 } 183 184 /* 185 * We support direct I/O only if lo_offset is aligned with the 186 * logical I/O size of backing device, and the logical block 187 * size of loop is bigger than the backing device's and the loop 188 * needn't transform transfer. 189 * 190 * TODO: the above condition may be loosed in the future, and 191 * direct I/O may be switched runtime at that time because most 192 * of requests in sane applications should be PAGE_SIZE aligned 193 */ 194 if (dio) { 195 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 196 !(lo->lo_offset & dio_align) && 197 mapping->a_ops->direct_IO && 198 !lo->transfer) 199 use_dio = true; 200 else 201 use_dio = false; 202 } else { 203 use_dio = false; 204 } 205 206 if (lo->use_dio == use_dio) 207 return; 208 209 /* flush dirty pages before changing direct IO */ 210 vfs_fsync(file, 0); 211 212 /* 213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 215 * will get updated by ioctl(LOOP_GET_STATUS) 216 */ 217 blk_mq_freeze_queue(lo->lo_queue); 218 lo->use_dio = use_dio; 219 if (use_dio) { 220 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 221 lo->lo_flags |= LO_FLAGS_DIRECT_IO; 222 } else { 223 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 224 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 225 } 226 blk_mq_unfreeze_queue(lo->lo_queue); 227 } 228 229 static int 230 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 231 { 232 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 233 sector_t x = (sector_t)size; 234 struct block_device *bdev = lo->lo_device; 235 236 if (unlikely((loff_t)x != size)) 237 return -EFBIG; 238 if (lo->lo_offset != offset) 239 lo->lo_offset = offset; 240 if (lo->lo_sizelimit != sizelimit) 241 lo->lo_sizelimit = sizelimit; 242 set_capacity(lo->lo_disk, x); 243 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); 244 /* let user-space know about the new size */ 245 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 246 return 0; 247 } 248 249 static inline int 250 lo_do_transfer(struct loop_device *lo, int cmd, 251 struct page *rpage, unsigned roffs, 252 struct page *lpage, unsigned loffs, 253 int size, sector_t rblock) 254 { 255 int ret; 256 257 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 258 if (likely(!ret)) 259 return 0; 260 261 printk_ratelimited(KERN_ERR 262 "loop: Transfer error at byte offset %llu, length %i.\n", 263 (unsigned long long)rblock << 9, size); 264 return ret; 265 } 266 267 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 268 { 269 struct iov_iter i; 270 ssize_t bw; 271 272 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 273 274 file_start_write(file); 275 bw = vfs_iter_write(file, &i, ppos, 0); 276 file_end_write(file); 277 278 if (likely(bw == bvec->bv_len)) 279 return 0; 280 281 printk_ratelimited(KERN_ERR 282 "loop: Write error at byte offset %llu, length %i.\n", 283 (unsigned long long)*ppos, bvec->bv_len); 284 if (bw >= 0) 285 bw = -EIO; 286 return bw; 287 } 288 289 static int lo_write_simple(struct loop_device *lo, struct request *rq, 290 loff_t pos) 291 { 292 struct bio_vec bvec; 293 struct req_iterator iter; 294 int ret = 0; 295 296 rq_for_each_segment(bvec, rq, iter) { 297 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 298 if (ret < 0) 299 break; 300 cond_resched(); 301 } 302 303 return ret; 304 } 305 306 /* 307 * This is the slow, transforming version that needs to double buffer the 308 * data as it cannot do the transformations in place without having direct 309 * access to the destination pages of the backing file. 310 */ 311 static int lo_write_transfer(struct loop_device *lo, struct request *rq, 312 loff_t pos) 313 { 314 struct bio_vec bvec, b; 315 struct req_iterator iter; 316 struct page *page; 317 int ret = 0; 318 319 page = alloc_page(GFP_NOIO); 320 if (unlikely(!page)) 321 return -ENOMEM; 322 323 rq_for_each_segment(bvec, rq, iter) { 324 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 325 bvec.bv_offset, bvec.bv_len, pos >> 9); 326 if (unlikely(ret)) 327 break; 328 329 b.bv_page = page; 330 b.bv_offset = 0; 331 b.bv_len = bvec.bv_len; 332 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 333 if (ret < 0) 334 break; 335 } 336 337 __free_page(page); 338 return ret; 339 } 340 341 static int lo_read_simple(struct loop_device *lo, struct request *rq, 342 loff_t pos) 343 { 344 struct bio_vec bvec; 345 struct req_iterator iter; 346 struct iov_iter i; 347 ssize_t len; 348 349 rq_for_each_segment(bvec, rq, iter) { 350 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 351 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 352 if (len < 0) 353 return len; 354 355 flush_dcache_page(bvec.bv_page); 356 357 if (len != bvec.bv_len) { 358 struct bio *bio; 359 360 __rq_for_each_bio(bio, rq) 361 zero_fill_bio(bio); 362 break; 363 } 364 cond_resched(); 365 } 366 367 return 0; 368 } 369 370 static int lo_read_transfer(struct loop_device *lo, struct request *rq, 371 loff_t pos) 372 { 373 struct bio_vec bvec, b; 374 struct req_iterator iter; 375 struct iov_iter i; 376 struct page *page; 377 ssize_t len; 378 int ret = 0; 379 380 page = alloc_page(GFP_NOIO); 381 if (unlikely(!page)) 382 return -ENOMEM; 383 384 rq_for_each_segment(bvec, rq, iter) { 385 loff_t offset = pos; 386 387 b.bv_page = page; 388 b.bv_offset = 0; 389 b.bv_len = bvec.bv_len; 390 391 iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 392 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 393 if (len < 0) { 394 ret = len; 395 goto out_free_page; 396 } 397 398 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 399 bvec.bv_offset, len, offset >> 9); 400 if (ret) 401 goto out_free_page; 402 403 flush_dcache_page(bvec.bv_page); 404 405 if (len != bvec.bv_len) { 406 struct bio *bio; 407 408 __rq_for_each_bio(bio, rq) 409 zero_fill_bio(bio); 410 break; 411 } 412 } 413 414 ret = 0; 415 out_free_page: 416 __free_page(page); 417 return ret; 418 } 419 420 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) 421 { 422 /* 423 * We use punch hole to reclaim the free space used by the 424 * image a.k.a. discard. However we do not support discard if 425 * encryption is enabled, because it may give an attacker 426 * useful information. 427 */ 428 struct file *file = lo->lo_backing_file; 429 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; 430 int ret; 431 432 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { 433 ret = -EOPNOTSUPP; 434 goto out; 435 } 436 437 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 438 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 439 ret = -EIO; 440 out: 441 return ret; 442 } 443 444 static int lo_req_flush(struct loop_device *lo, struct request *rq) 445 { 446 struct file *file = lo->lo_backing_file; 447 int ret = vfs_fsync(file, 0); 448 if (unlikely(ret && ret != -EINVAL)) 449 ret = -EIO; 450 451 return ret; 452 } 453 454 static void lo_complete_rq(struct request *rq) 455 { 456 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 457 blk_status_t ret = BLK_STS_OK; 458 459 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 460 req_op(rq) != REQ_OP_READ) { 461 if (cmd->ret < 0) 462 ret = BLK_STS_IOERR; 463 goto end_io; 464 } 465 466 /* 467 * Short READ - if we got some data, advance our request and 468 * retry it. If we got no data, end the rest with EIO. 469 */ 470 if (cmd->ret) { 471 blk_update_request(rq, BLK_STS_OK, cmd->ret); 472 cmd->ret = 0; 473 blk_mq_requeue_request(rq, true); 474 } else { 475 if (cmd->use_aio) { 476 struct bio *bio = rq->bio; 477 478 while (bio) { 479 zero_fill_bio(bio); 480 bio = bio->bi_next; 481 } 482 } 483 ret = BLK_STS_IOERR; 484 end_io: 485 blk_mq_end_request(rq, ret); 486 } 487 } 488 489 static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 490 { 491 struct request *rq = blk_mq_rq_from_pdu(cmd); 492 493 if (!atomic_dec_and_test(&cmd->ref)) 494 return; 495 kfree(cmd->bvec); 496 cmd->bvec = NULL; 497 blk_mq_complete_request(rq); 498 } 499 500 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 501 { 502 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 503 504 if (cmd->css) 505 css_put(cmd->css); 506 cmd->ret = ret; 507 lo_rw_aio_do_completion(cmd); 508 } 509 510 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 511 loff_t pos, bool rw) 512 { 513 struct iov_iter iter; 514 struct bio_vec *bvec; 515 struct request *rq = blk_mq_rq_from_pdu(cmd); 516 struct bio *bio = rq->bio; 517 struct file *file = lo->lo_backing_file; 518 unsigned int offset; 519 int segments = 0; 520 int ret; 521 522 if (rq->bio != rq->biotail) { 523 struct req_iterator iter; 524 struct bio_vec tmp; 525 526 __rq_for_each_bio(bio, rq) 527 segments += bio_segments(bio); 528 bvec = kmalloc_array(segments, sizeof(struct bio_vec), 529 GFP_NOIO); 530 if (!bvec) 531 return -EIO; 532 cmd->bvec = bvec; 533 534 /* 535 * The bios of the request may be started from the middle of 536 * the 'bvec' because of bio splitting, so we can't directly 537 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment 538 * API will take care of all details for us. 539 */ 540 rq_for_each_segment(tmp, rq, iter) { 541 *bvec = tmp; 542 bvec++; 543 } 544 bvec = cmd->bvec; 545 offset = 0; 546 } else { 547 /* 548 * Same here, this bio may be started from the middle of the 549 * 'bvec' because of bio splitting, so offset from the bvec 550 * must be passed to iov iterator 551 */ 552 offset = bio->bi_iter.bi_bvec_done; 553 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 554 segments = bio_segments(bio); 555 } 556 atomic_set(&cmd->ref, 2); 557 558 iov_iter_bvec(&iter, rw, bvec, segments, blk_rq_bytes(rq)); 559 iter.iov_offset = offset; 560 561 cmd->iocb.ki_pos = pos; 562 cmd->iocb.ki_filp = file; 563 cmd->iocb.ki_complete = lo_rw_aio_complete; 564 cmd->iocb.ki_flags = IOCB_DIRECT; 565 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 566 if (cmd->css) 567 kthread_associate_blkcg(cmd->css); 568 569 if (rw == WRITE) 570 ret = call_write_iter(file, &cmd->iocb, &iter); 571 else 572 ret = call_read_iter(file, &cmd->iocb, &iter); 573 574 lo_rw_aio_do_completion(cmd); 575 kthread_associate_blkcg(NULL); 576 577 if (ret != -EIOCBQUEUED) 578 cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 579 return 0; 580 } 581 582 static int do_req_filebacked(struct loop_device *lo, struct request *rq) 583 { 584 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 585 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 586 587 /* 588 * lo_write_simple and lo_read_simple should have been covered 589 * by io submit style function like lo_rw_aio(), one blocker 590 * is that lo_read_simple() need to call flush_dcache_page after 591 * the page is written from kernel, and it isn't easy to handle 592 * this in io submit style function which submits all segments 593 * of the req at one time. And direct read IO doesn't need to 594 * run flush_dcache_page(). 595 */ 596 switch (req_op(rq)) { 597 case REQ_OP_FLUSH: 598 return lo_req_flush(lo, rq); 599 case REQ_OP_DISCARD: 600 case REQ_OP_WRITE_ZEROES: 601 return lo_discard(lo, rq, pos); 602 case REQ_OP_WRITE: 603 if (lo->transfer) 604 return lo_write_transfer(lo, rq, pos); 605 else if (cmd->use_aio) 606 return lo_rw_aio(lo, cmd, pos, WRITE); 607 else 608 return lo_write_simple(lo, rq, pos); 609 case REQ_OP_READ: 610 if (lo->transfer) 611 return lo_read_transfer(lo, rq, pos); 612 else if (cmd->use_aio) 613 return lo_rw_aio(lo, cmd, pos, READ); 614 else 615 return lo_read_simple(lo, rq, pos); 616 default: 617 WARN_ON_ONCE(1); 618 return -EIO; 619 break; 620 } 621 } 622 623 static inline void loop_update_dio(struct loop_device *lo) 624 { 625 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 626 lo->use_dio); 627 } 628 629 static void loop_reread_partitions(struct loop_device *lo, 630 struct block_device *bdev) 631 { 632 int rc; 633 634 rc = blkdev_reread_part(bdev); 635 if (rc) 636 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 637 __func__, lo->lo_number, lo->lo_file_name, rc); 638 } 639 640 static inline int is_loop_device(struct file *file) 641 { 642 struct inode *i = file->f_mapping->host; 643 644 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 645 } 646 647 static int loop_validate_file(struct file *file, struct block_device *bdev) 648 { 649 struct inode *inode = file->f_mapping->host; 650 struct file *f = file; 651 652 /* Avoid recursion */ 653 while (is_loop_device(f)) { 654 struct loop_device *l; 655 656 if (f->f_mapping->host->i_bdev == bdev) 657 return -EBADF; 658 659 l = f->f_mapping->host->i_bdev->bd_disk->private_data; 660 if (l->lo_state == Lo_unbound) { 661 return -EINVAL; 662 } 663 f = l->lo_backing_file; 664 } 665 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 666 return -EINVAL; 667 return 0; 668 } 669 670 /* 671 * loop_change_fd switched the backing store of a loopback device to 672 * a new file. This is useful for operating system installers to free up 673 * the original file and in High Availability environments to switch to 674 * an alternative location for the content in case of server meltdown. 675 * This can only work if the loop device is used read-only, and if the 676 * new backing store is the same size and type as the old backing store. 677 */ 678 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 679 unsigned int arg) 680 { 681 struct file *file = NULL, *old_file; 682 int error; 683 bool partscan; 684 685 error = mutex_lock_killable(&loop_ctl_mutex); 686 if (error) 687 return error; 688 error = -ENXIO; 689 if (lo->lo_state != Lo_bound) 690 goto out_err; 691 692 /* the loop device has to be read-only */ 693 error = -EINVAL; 694 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 695 goto out_err; 696 697 error = -EBADF; 698 file = fget(arg); 699 if (!file) 700 goto out_err; 701 702 error = loop_validate_file(file, bdev); 703 if (error) 704 goto out_err; 705 706 old_file = lo->lo_backing_file; 707 708 error = -EINVAL; 709 710 /* size of the new backing store needs to be the same */ 711 if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 712 goto out_err; 713 714 /* and ... switch */ 715 blk_mq_freeze_queue(lo->lo_queue); 716 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 717 lo->lo_backing_file = file; 718 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 719 mapping_set_gfp_mask(file->f_mapping, 720 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 721 loop_update_dio(lo); 722 blk_mq_unfreeze_queue(lo->lo_queue); 723 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 724 mutex_unlock(&loop_ctl_mutex); 725 /* 726 * We must drop file reference outside of loop_ctl_mutex as dropping 727 * the file ref can take bd_mutex which creates circular locking 728 * dependency. 729 */ 730 fput(old_file); 731 if (partscan) 732 loop_reread_partitions(lo, bdev); 733 return 0; 734 735 out_err: 736 mutex_unlock(&loop_ctl_mutex); 737 if (file) 738 fput(file); 739 return error; 740 } 741 742 /* loop sysfs attributes */ 743 744 static ssize_t loop_attr_show(struct device *dev, char *page, 745 ssize_t (*callback)(struct loop_device *, char *)) 746 { 747 struct gendisk *disk = dev_to_disk(dev); 748 struct loop_device *lo = disk->private_data; 749 750 return callback(lo, page); 751 } 752 753 #define LOOP_ATTR_RO(_name) \ 754 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 755 static ssize_t loop_attr_do_show_##_name(struct device *d, \ 756 struct device_attribute *attr, char *b) \ 757 { \ 758 return loop_attr_show(d, b, loop_attr_##_name##_show); \ 759 } \ 760 static struct device_attribute loop_attr_##_name = \ 761 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 762 763 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 764 { 765 ssize_t ret; 766 char *p = NULL; 767 768 spin_lock_irq(&lo->lo_lock); 769 if (lo->lo_backing_file) 770 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 771 spin_unlock_irq(&lo->lo_lock); 772 773 if (IS_ERR_OR_NULL(p)) 774 ret = PTR_ERR(p); 775 else { 776 ret = strlen(p); 777 memmove(buf, p, ret); 778 buf[ret++] = '\n'; 779 buf[ret] = 0; 780 } 781 782 return ret; 783 } 784 785 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 786 { 787 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 788 } 789 790 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 791 { 792 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 793 } 794 795 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 796 { 797 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 798 799 return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 800 } 801 802 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 803 { 804 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 805 806 return sprintf(buf, "%s\n", partscan ? "1" : "0"); 807 } 808 809 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 810 { 811 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 812 813 return sprintf(buf, "%s\n", dio ? "1" : "0"); 814 } 815 816 LOOP_ATTR_RO(backing_file); 817 LOOP_ATTR_RO(offset); 818 LOOP_ATTR_RO(sizelimit); 819 LOOP_ATTR_RO(autoclear); 820 LOOP_ATTR_RO(partscan); 821 LOOP_ATTR_RO(dio); 822 823 static struct attribute *loop_attrs[] = { 824 &loop_attr_backing_file.attr, 825 &loop_attr_offset.attr, 826 &loop_attr_sizelimit.attr, 827 &loop_attr_autoclear.attr, 828 &loop_attr_partscan.attr, 829 &loop_attr_dio.attr, 830 NULL, 831 }; 832 833 static struct attribute_group loop_attribute_group = { 834 .name = "loop", 835 .attrs= loop_attrs, 836 }; 837 838 static void loop_sysfs_init(struct loop_device *lo) 839 { 840 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 841 &loop_attribute_group); 842 } 843 844 static void loop_sysfs_exit(struct loop_device *lo) 845 { 846 if (lo->sysfs_inited) 847 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 848 &loop_attribute_group); 849 } 850 851 static void loop_config_discard(struct loop_device *lo) 852 { 853 struct file *file = lo->lo_backing_file; 854 struct inode *inode = file->f_mapping->host; 855 struct request_queue *q = lo->lo_queue; 856 857 /* 858 * We use punch hole to reclaim the free space used by the 859 * image a.k.a. discard. However we do not support discard if 860 * encryption is enabled, because it may give an attacker 861 * useful information. 862 */ 863 if ((!file->f_op->fallocate) || 864 lo->lo_encrypt_key_size) { 865 q->limits.discard_granularity = 0; 866 q->limits.discard_alignment = 0; 867 blk_queue_max_discard_sectors(q, 0); 868 blk_queue_max_write_zeroes_sectors(q, 0); 869 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 870 return; 871 } 872 873 q->limits.discard_granularity = inode->i_sb->s_blocksize; 874 q->limits.discard_alignment = 0; 875 876 blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 877 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 878 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 879 } 880 881 static void loop_unprepare_queue(struct loop_device *lo) 882 { 883 kthread_flush_worker(&lo->worker); 884 kthread_stop(lo->worker_task); 885 } 886 887 static int loop_kthread_worker_fn(void *worker_ptr) 888 { 889 current->flags |= PF_LESS_THROTTLE; 890 return kthread_worker_fn(worker_ptr); 891 } 892 893 static int loop_prepare_queue(struct loop_device *lo) 894 { 895 kthread_init_worker(&lo->worker); 896 lo->worker_task = kthread_run(loop_kthread_worker_fn, 897 &lo->worker, "loop%d", lo->lo_number); 898 if (IS_ERR(lo->worker_task)) 899 return -ENOMEM; 900 set_user_nice(lo->worker_task, MIN_NICE); 901 return 0; 902 } 903 904 static int loop_set_fd(struct loop_device *lo, fmode_t mode, 905 struct block_device *bdev, unsigned int arg) 906 { 907 struct file *file; 908 struct inode *inode; 909 struct address_space *mapping; 910 int lo_flags = 0; 911 int error; 912 loff_t size; 913 bool partscan; 914 915 /* This is safe, since we have a reference from open(). */ 916 __module_get(THIS_MODULE); 917 918 error = -EBADF; 919 file = fget(arg); 920 if (!file) 921 goto out; 922 923 error = mutex_lock_killable(&loop_ctl_mutex); 924 if (error) 925 goto out_putf; 926 927 error = -EBUSY; 928 if (lo->lo_state != Lo_unbound) 929 goto out_unlock; 930 931 error = loop_validate_file(file, bdev); 932 if (error) 933 goto out_unlock; 934 935 mapping = file->f_mapping; 936 inode = mapping->host; 937 938 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 939 !file->f_op->write_iter) 940 lo_flags |= LO_FLAGS_READ_ONLY; 941 942 error = -EFBIG; 943 size = get_loop_size(lo, file); 944 if ((loff_t)(sector_t)size != size) 945 goto out_unlock; 946 error = loop_prepare_queue(lo); 947 if (error) 948 goto out_unlock; 949 950 error = 0; 951 952 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 953 954 lo->use_dio = false; 955 lo->lo_device = bdev; 956 lo->lo_flags = lo_flags; 957 lo->lo_backing_file = file; 958 lo->transfer = NULL; 959 lo->ioctl = NULL; 960 lo->lo_sizelimit = 0; 961 lo->old_gfp_mask = mapping_gfp_mask(mapping); 962 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 963 964 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 965 blk_queue_write_cache(lo->lo_queue, true, false); 966 967 loop_update_dio(lo); 968 set_capacity(lo->lo_disk, size); 969 bd_set_size(bdev, size << 9); 970 loop_sysfs_init(lo); 971 /* let user-space know about the new size */ 972 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 973 974 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 975 block_size(inode->i_bdev) : PAGE_SIZE); 976 977 lo->lo_state = Lo_bound; 978 if (part_shift) 979 lo->lo_flags |= LO_FLAGS_PARTSCAN; 980 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 981 982 /* Grab the block_device to prevent its destruction after we 983 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 984 */ 985 bdgrab(bdev); 986 mutex_unlock(&loop_ctl_mutex); 987 if (partscan) 988 loop_reread_partitions(lo, bdev); 989 return 0; 990 991 out_unlock: 992 mutex_unlock(&loop_ctl_mutex); 993 out_putf: 994 fput(file); 995 out: 996 /* This is safe: open() is still holding a reference. */ 997 module_put(THIS_MODULE); 998 return error; 999 } 1000 1001 static int 1002 loop_release_xfer(struct loop_device *lo) 1003 { 1004 int err = 0; 1005 struct loop_func_table *xfer = lo->lo_encryption; 1006 1007 if (xfer) { 1008 if (xfer->release) 1009 err = xfer->release(lo); 1010 lo->transfer = NULL; 1011 lo->lo_encryption = NULL; 1012 module_put(xfer->owner); 1013 } 1014 return err; 1015 } 1016 1017 static int 1018 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1019 const struct loop_info64 *i) 1020 { 1021 int err = 0; 1022 1023 if (xfer) { 1024 struct module *owner = xfer->owner; 1025 1026 if (!try_module_get(owner)) 1027 return -EINVAL; 1028 if (xfer->init) 1029 err = xfer->init(lo, i); 1030 if (err) 1031 module_put(owner); 1032 else 1033 lo->lo_encryption = xfer; 1034 } 1035 return err; 1036 } 1037 1038 static int __loop_clr_fd(struct loop_device *lo, bool release) 1039 { 1040 struct file *filp = NULL; 1041 gfp_t gfp = lo->old_gfp_mask; 1042 struct block_device *bdev = lo->lo_device; 1043 int err = 0; 1044 bool partscan = false; 1045 int lo_number; 1046 1047 mutex_lock(&loop_ctl_mutex); 1048 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1049 err = -ENXIO; 1050 goto out_unlock; 1051 } 1052 1053 filp = lo->lo_backing_file; 1054 if (filp == NULL) { 1055 err = -EINVAL; 1056 goto out_unlock; 1057 } 1058 1059 /* freeze request queue during the transition */ 1060 blk_mq_freeze_queue(lo->lo_queue); 1061 1062 spin_lock_irq(&lo->lo_lock); 1063 lo->lo_backing_file = NULL; 1064 spin_unlock_irq(&lo->lo_lock); 1065 1066 loop_release_xfer(lo); 1067 lo->transfer = NULL; 1068 lo->ioctl = NULL; 1069 lo->lo_device = NULL; 1070 lo->lo_encryption = NULL; 1071 lo->lo_offset = 0; 1072 lo->lo_sizelimit = 0; 1073 lo->lo_encrypt_key_size = 0; 1074 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1075 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1076 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1077 blk_queue_logical_block_size(lo->lo_queue, 512); 1078 blk_queue_physical_block_size(lo->lo_queue, 512); 1079 blk_queue_io_min(lo->lo_queue, 512); 1080 if (bdev) { 1081 bdput(bdev); 1082 invalidate_bdev(bdev); 1083 bdev->bd_inode->i_mapping->wb_err = 0; 1084 } 1085 set_capacity(lo->lo_disk, 0); 1086 loop_sysfs_exit(lo); 1087 if (bdev) { 1088 bd_set_size(bdev, 0); 1089 /* let user-space know about this change */ 1090 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1091 } 1092 mapping_set_gfp_mask(filp->f_mapping, gfp); 1093 lo->lo_state = Lo_unbound; 1094 /* This is safe: open() is still holding a reference. */ 1095 module_put(THIS_MODULE); 1096 blk_mq_unfreeze_queue(lo->lo_queue); 1097 1098 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 1099 lo_number = lo->lo_number; 1100 lo->lo_flags = 0; 1101 if (!part_shift) 1102 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1103 loop_unprepare_queue(lo); 1104 out_unlock: 1105 mutex_unlock(&loop_ctl_mutex); 1106 if (partscan) { 1107 /* 1108 * bd_mutex has been held already in release path, so don't 1109 * acquire it if this function is called in such case. 1110 * 1111 * If the reread partition isn't from release path, lo_refcnt 1112 * must be at least one and it can only become zero when the 1113 * current holder is released. 1114 */ 1115 if (release) 1116 err = __blkdev_reread_part(bdev); 1117 else 1118 err = blkdev_reread_part(bdev); 1119 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 1120 __func__, lo_number, err); 1121 /* Device is gone, no point in returning error */ 1122 err = 0; 1123 } 1124 /* 1125 * Need not hold loop_ctl_mutex to fput backing file. 1126 * Calling fput holding loop_ctl_mutex triggers a circular 1127 * lock dependency possibility warning as fput can take 1128 * bd_mutex which is usually taken before loop_ctl_mutex. 1129 */ 1130 if (filp) 1131 fput(filp); 1132 return err; 1133 } 1134 1135 static int loop_clr_fd(struct loop_device *lo) 1136 { 1137 int err; 1138 1139 err = mutex_lock_killable(&loop_ctl_mutex); 1140 if (err) 1141 return err; 1142 if (lo->lo_state != Lo_bound) { 1143 mutex_unlock(&loop_ctl_mutex); 1144 return -ENXIO; 1145 } 1146 /* 1147 * If we've explicitly asked to tear down the loop device, 1148 * and it has an elevated reference count, set it for auto-teardown when 1149 * the last reference goes away. This stops $!~#$@ udev from 1150 * preventing teardown because it decided that it needs to run blkid on 1151 * the loopback device whenever they appear. xfstests is notorious for 1152 * failing tests because blkid via udev races with a losetup 1153 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1154 * command to fail with EBUSY. 1155 */ 1156 if (atomic_read(&lo->lo_refcnt) > 1) { 1157 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1158 mutex_unlock(&loop_ctl_mutex); 1159 return 0; 1160 } 1161 lo->lo_state = Lo_rundown; 1162 mutex_unlock(&loop_ctl_mutex); 1163 1164 return __loop_clr_fd(lo, false); 1165 } 1166 1167 static int 1168 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1169 { 1170 int err; 1171 struct loop_func_table *xfer; 1172 kuid_t uid = current_uid(); 1173 struct block_device *bdev; 1174 bool partscan = false; 1175 1176 err = mutex_lock_killable(&loop_ctl_mutex); 1177 if (err) 1178 return err; 1179 if (lo->lo_encrypt_key_size && 1180 !uid_eq(lo->lo_key_owner, uid) && 1181 !capable(CAP_SYS_ADMIN)) { 1182 err = -EPERM; 1183 goto out_unlock; 1184 } 1185 if (lo->lo_state != Lo_bound) { 1186 err = -ENXIO; 1187 goto out_unlock; 1188 } 1189 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) { 1190 err = -EINVAL; 1191 goto out_unlock; 1192 } 1193 1194 /* I/O need to be drained during transfer transition */ 1195 blk_mq_freeze_queue(lo->lo_queue); 1196 1197 err = loop_release_xfer(lo); 1198 if (err) 1199 goto out_unfreeze; 1200 1201 if (info->lo_encrypt_type) { 1202 unsigned int type = info->lo_encrypt_type; 1203 1204 if (type >= MAX_LO_CRYPT) { 1205 err = -EINVAL; 1206 goto out_unfreeze; 1207 } 1208 xfer = xfer_funcs[type]; 1209 if (xfer == NULL) { 1210 err = -EINVAL; 1211 goto out_unfreeze; 1212 } 1213 } else 1214 xfer = NULL; 1215 1216 err = loop_init_xfer(lo, xfer, info); 1217 if (err) 1218 goto out_unfreeze; 1219 1220 if (lo->lo_offset != info->lo_offset || 1221 lo->lo_sizelimit != info->lo_sizelimit) { 1222 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1223 err = -EFBIG; 1224 goto out_unfreeze; 1225 } 1226 } 1227 1228 loop_config_discard(lo); 1229 1230 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1231 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1232 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1233 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1234 1235 if (!xfer) 1236 xfer = &none_funcs; 1237 lo->transfer = xfer->transfer; 1238 lo->ioctl = xfer->ioctl; 1239 1240 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1241 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1242 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1243 1244 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1245 lo->lo_init[0] = info->lo_init[0]; 1246 lo->lo_init[1] = info->lo_init[1]; 1247 if (info->lo_encrypt_key_size) { 1248 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1249 info->lo_encrypt_key_size); 1250 lo->lo_key_owner = uid; 1251 } 1252 1253 /* update dio if lo_offset or transfer is changed */ 1254 __loop_update_dio(lo, lo->use_dio); 1255 1256 out_unfreeze: 1257 blk_mq_unfreeze_queue(lo->lo_queue); 1258 1259 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1260 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1261 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1262 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1263 bdev = lo->lo_device; 1264 partscan = true; 1265 } 1266 out_unlock: 1267 mutex_unlock(&loop_ctl_mutex); 1268 if (partscan) 1269 loop_reread_partitions(lo, bdev); 1270 1271 return err; 1272 } 1273 1274 static int 1275 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1276 { 1277 struct path path; 1278 struct kstat stat; 1279 int ret; 1280 1281 ret = mutex_lock_killable(&loop_ctl_mutex); 1282 if (ret) 1283 return ret; 1284 if (lo->lo_state != Lo_bound) { 1285 mutex_unlock(&loop_ctl_mutex); 1286 return -ENXIO; 1287 } 1288 1289 memset(info, 0, sizeof(*info)); 1290 info->lo_number = lo->lo_number; 1291 info->lo_offset = lo->lo_offset; 1292 info->lo_sizelimit = lo->lo_sizelimit; 1293 info->lo_flags = lo->lo_flags; 1294 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1295 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1296 info->lo_encrypt_type = 1297 lo->lo_encryption ? lo->lo_encryption->number : 0; 1298 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1299 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1300 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1301 lo->lo_encrypt_key_size); 1302 } 1303 1304 /* Drop loop_ctl_mutex while we call into the filesystem. */ 1305 path = lo->lo_backing_file->f_path; 1306 path_get(&path); 1307 mutex_unlock(&loop_ctl_mutex); 1308 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1309 if (!ret) { 1310 info->lo_device = huge_encode_dev(stat.dev); 1311 info->lo_inode = stat.ino; 1312 info->lo_rdevice = huge_encode_dev(stat.rdev); 1313 } 1314 path_put(&path); 1315 return ret; 1316 } 1317 1318 static void 1319 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1320 { 1321 memset(info64, 0, sizeof(*info64)); 1322 info64->lo_number = info->lo_number; 1323 info64->lo_device = info->lo_device; 1324 info64->lo_inode = info->lo_inode; 1325 info64->lo_rdevice = info->lo_rdevice; 1326 info64->lo_offset = info->lo_offset; 1327 info64->lo_sizelimit = 0; 1328 info64->lo_encrypt_type = info->lo_encrypt_type; 1329 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1330 info64->lo_flags = info->lo_flags; 1331 info64->lo_init[0] = info->lo_init[0]; 1332 info64->lo_init[1] = info->lo_init[1]; 1333 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1334 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1335 else 1336 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1337 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1338 } 1339 1340 static int 1341 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1342 { 1343 memset(info, 0, sizeof(*info)); 1344 info->lo_number = info64->lo_number; 1345 info->lo_device = info64->lo_device; 1346 info->lo_inode = info64->lo_inode; 1347 info->lo_rdevice = info64->lo_rdevice; 1348 info->lo_offset = info64->lo_offset; 1349 info->lo_encrypt_type = info64->lo_encrypt_type; 1350 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1351 info->lo_flags = info64->lo_flags; 1352 info->lo_init[0] = info64->lo_init[0]; 1353 info->lo_init[1] = info64->lo_init[1]; 1354 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1355 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1356 else 1357 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1358 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1359 1360 /* error in case values were truncated */ 1361 if (info->lo_device != info64->lo_device || 1362 info->lo_rdevice != info64->lo_rdevice || 1363 info->lo_inode != info64->lo_inode || 1364 info->lo_offset != info64->lo_offset) 1365 return -EOVERFLOW; 1366 1367 return 0; 1368 } 1369 1370 static int 1371 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1372 { 1373 struct loop_info info; 1374 struct loop_info64 info64; 1375 1376 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1377 return -EFAULT; 1378 loop_info64_from_old(&info, &info64); 1379 return loop_set_status(lo, &info64); 1380 } 1381 1382 static int 1383 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1384 { 1385 struct loop_info64 info64; 1386 1387 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1388 return -EFAULT; 1389 return loop_set_status(lo, &info64); 1390 } 1391 1392 static int 1393 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1394 struct loop_info info; 1395 struct loop_info64 info64; 1396 int err; 1397 1398 if (!arg) 1399 return -EINVAL; 1400 err = loop_get_status(lo, &info64); 1401 if (!err) 1402 err = loop_info64_to_old(&info64, &info); 1403 if (!err && copy_to_user(arg, &info, sizeof(info))) 1404 err = -EFAULT; 1405 1406 return err; 1407 } 1408 1409 static int 1410 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1411 struct loop_info64 info64; 1412 int err; 1413 1414 if (!arg) 1415 return -EINVAL; 1416 err = loop_get_status(lo, &info64); 1417 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1418 err = -EFAULT; 1419 1420 return err; 1421 } 1422 1423 static int loop_set_capacity(struct loop_device *lo) 1424 { 1425 if (unlikely(lo->lo_state != Lo_bound)) 1426 return -ENXIO; 1427 1428 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1429 } 1430 1431 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1432 { 1433 int error = -ENXIO; 1434 if (lo->lo_state != Lo_bound) 1435 goto out; 1436 1437 __loop_update_dio(lo, !!arg); 1438 if (lo->use_dio == !!arg) 1439 return 0; 1440 error = -EINVAL; 1441 out: 1442 return error; 1443 } 1444 1445 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1446 { 1447 if (lo->lo_state != Lo_bound) 1448 return -ENXIO; 1449 1450 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1451 return -EINVAL; 1452 1453 blk_mq_freeze_queue(lo->lo_queue); 1454 1455 blk_queue_logical_block_size(lo->lo_queue, arg); 1456 blk_queue_physical_block_size(lo->lo_queue, arg); 1457 blk_queue_io_min(lo->lo_queue, arg); 1458 loop_update_dio(lo); 1459 1460 blk_mq_unfreeze_queue(lo->lo_queue); 1461 1462 return 0; 1463 } 1464 1465 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1466 unsigned long arg) 1467 { 1468 int err; 1469 1470 err = mutex_lock_killable(&loop_ctl_mutex); 1471 if (err) 1472 return err; 1473 switch (cmd) { 1474 case LOOP_SET_CAPACITY: 1475 err = loop_set_capacity(lo); 1476 break; 1477 case LOOP_SET_DIRECT_IO: 1478 err = loop_set_dio(lo, arg); 1479 break; 1480 case LOOP_SET_BLOCK_SIZE: 1481 err = loop_set_block_size(lo, arg); 1482 break; 1483 default: 1484 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1485 } 1486 mutex_unlock(&loop_ctl_mutex); 1487 return err; 1488 } 1489 1490 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1491 unsigned int cmd, unsigned long arg) 1492 { 1493 struct loop_device *lo = bdev->bd_disk->private_data; 1494 int err; 1495 1496 switch (cmd) { 1497 case LOOP_SET_FD: 1498 return loop_set_fd(lo, mode, bdev, arg); 1499 case LOOP_CHANGE_FD: 1500 return loop_change_fd(lo, bdev, arg); 1501 case LOOP_CLR_FD: 1502 return loop_clr_fd(lo); 1503 case LOOP_SET_STATUS: 1504 err = -EPERM; 1505 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1506 err = loop_set_status_old(lo, 1507 (struct loop_info __user *)arg); 1508 } 1509 break; 1510 case LOOP_GET_STATUS: 1511 return loop_get_status_old(lo, (struct loop_info __user *) arg); 1512 case LOOP_SET_STATUS64: 1513 err = -EPERM; 1514 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1515 err = loop_set_status64(lo, 1516 (struct loop_info64 __user *) arg); 1517 } 1518 break; 1519 case LOOP_GET_STATUS64: 1520 return loop_get_status64(lo, (struct loop_info64 __user *) arg); 1521 case LOOP_SET_CAPACITY: 1522 case LOOP_SET_DIRECT_IO: 1523 case LOOP_SET_BLOCK_SIZE: 1524 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1525 return -EPERM; 1526 /* Fall through */ 1527 default: 1528 err = lo_simple_ioctl(lo, cmd, arg); 1529 break; 1530 } 1531 1532 return err; 1533 } 1534 1535 #ifdef CONFIG_COMPAT 1536 struct compat_loop_info { 1537 compat_int_t lo_number; /* ioctl r/o */ 1538 compat_dev_t lo_device; /* ioctl r/o */ 1539 compat_ulong_t lo_inode; /* ioctl r/o */ 1540 compat_dev_t lo_rdevice; /* ioctl r/o */ 1541 compat_int_t lo_offset; 1542 compat_int_t lo_encrypt_type; 1543 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1544 compat_int_t lo_flags; /* ioctl r/o */ 1545 char lo_name[LO_NAME_SIZE]; 1546 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1547 compat_ulong_t lo_init[2]; 1548 char reserved[4]; 1549 }; 1550 1551 /* 1552 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1553 * - noinlined to reduce stack space usage in main part of driver 1554 */ 1555 static noinline int 1556 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1557 struct loop_info64 *info64) 1558 { 1559 struct compat_loop_info info; 1560 1561 if (copy_from_user(&info, arg, sizeof(info))) 1562 return -EFAULT; 1563 1564 memset(info64, 0, sizeof(*info64)); 1565 info64->lo_number = info.lo_number; 1566 info64->lo_device = info.lo_device; 1567 info64->lo_inode = info.lo_inode; 1568 info64->lo_rdevice = info.lo_rdevice; 1569 info64->lo_offset = info.lo_offset; 1570 info64->lo_sizelimit = 0; 1571 info64->lo_encrypt_type = info.lo_encrypt_type; 1572 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1573 info64->lo_flags = info.lo_flags; 1574 info64->lo_init[0] = info.lo_init[0]; 1575 info64->lo_init[1] = info.lo_init[1]; 1576 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1577 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1578 else 1579 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1580 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1581 return 0; 1582 } 1583 1584 /* 1585 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1586 * - noinlined to reduce stack space usage in main part of driver 1587 */ 1588 static noinline int 1589 loop_info64_to_compat(const struct loop_info64 *info64, 1590 struct compat_loop_info __user *arg) 1591 { 1592 struct compat_loop_info info; 1593 1594 memset(&info, 0, sizeof(info)); 1595 info.lo_number = info64->lo_number; 1596 info.lo_device = info64->lo_device; 1597 info.lo_inode = info64->lo_inode; 1598 info.lo_rdevice = info64->lo_rdevice; 1599 info.lo_offset = info64->lo_offset; 1600 info.lo_encrypt_type = info64->lo_encrypt_type; 1601 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1602 info.lo_flags = info64->lo_flags; 1603 info.lo_init[0] = info64->lo_init[0]; 1604 info.lo_init[1] = info64->lo_init[1]; 1605 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1606 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1607 else 1608 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1609 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1610 1611 /* error in case values were truncated */ 1612 if (info.lo_device != info64->lo_device || 1613 info.lo_rdevice != info64->lo_rdevice || 1614 info.lo_inode != info64->lo_inode || 1615 info.lo_offset != info64->lo_offset || 1616 info.lo_init[0] != info64->lo_init[0] || 1617 info.lo_init[1] != info64->lo_init[1]) 1618 return -EOVERFLOW; 1619 1620 if (copy_to_user(arg, &info, sizeof(info))) 1621 return -EFAULT; 1622 return 0; 1623 } 1624 1625 static int 1626 loop_set_status_compat(struct loop_device *lo, 1627 const struct compat_loop_info __user *arg) 1628 { 1629 struct loop_info64 info64; 1630 int ret; 1631 1632 ret = loop_info64_from_compat(arg, &info64); 1633 if (ret < 0) 1634 return ret; 1635 return loop_set_status(lo, &info64); 1636 } 1637 1638 static int 1639 loop_get_status_compat(struct loop_device *lo, 1640 struct compat_loop_info __user *arg) 1641 { 1642 struct loop_info64 info64; 1643 int err; 1644 1645 if (!arg) 1646 return -EINVAL; 1647 err = loop_get_status(lo, &info64); 1648 if (!err) 1649 err = loop_info64_to_compat(&info64, arg); 1650 return err; 1651 } 1652 1653 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1654 unsigned int cmd, unsigned long arg) 1655 { 1656 struct loop_device *lo = bdev->bd_disk->private_data; 1657 int err; 1658 1659 switch(cmd) { 1660 case LOOP_SET_STATUS: 1661 err = loop_set_status_compat(lo, 1662 (const struct compat_loop_info __user *)arg); 1663 break; 1664 case LOOP_GET_STATUS: 1665 err = loop_get_status_compat(lo, 1666 (struct compat_loop_info __user *)arg); 1667 break; 1668 case LOOP_SET_CAPACITY: 1669 case LOOP_CLR_FD: 1670 case LOOP_GET_STATUS64: 1671 case LOOP_SET_STATUS64: 1672 arg = (unsigned long) compat_ptr(arg); 1673 /* fall through */ 1674 case LOOP_SET_FD: 1675 case LOOP_CHANGE_FD: 1676 case LOOP_SET_BLOCK_SIZE: 1677 err = lo_ioctl(bdev, mode, cmd, arg); 1678 break; 1679 default: 1680 err = -ENOIOCTLCMD; 1681 break; 1682 } 1683 return err; 1684 } 1685 #endif 1686 1687 static int lo_open(struct block_device *bdev, fmode_t mode) 1688 { 1689 struct loop_device *lo; 1690 int err; 1691 1692 err = mutex_lock_killable(&loop_ctl_mutex); 1693 if (err) 1694 return err; 1695 lo = bdev->bd_disk->private_data; 1696 if (!lo) { 1697 err = -ENXIO; 1698 goto out; 1699 } 1700 1701 atomic_inc(&lo->lo_refcnt); 1702 out: 1703 mutex_unlock(&loop_ctl_mutex); 1704 return err; 1705 } 1706 1707 static void lo_release(struct gendisk *disk, fmode_t mode) 1708 { 1709 struct loop_device *lo; 1710 1711 mutex_lock(&loop_ctl_mutex); 1712 lo = disk->private_data; 1713 if (atomic_dec_return(&lo->lo_refcnt)) 1714 goto out_unlock; 1715 1716 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1717 if (lo->lo_state != Lo_bound) 1718 goto out_unlock; 1719 lo->lo_state = Lo_rundown; 1720 mutex_unlock(&loop_ctl_mutex); 1721 /* 1722 * In autoclear mode, stop the loop thread 1723 * and remove configuration after last close. 1724 */ 1725 __loop_clr_fd(lo, true); 1726 return; 1727 } else if (lo->lo_state == Lo_bound) { 1728 /* 1729 * Otherwise keep thread (if running) and config, 1730 * but flush possible ongoing bios in thread. 1731 */ 1732 blk_mq_freeze_queue(lo->lo_queue); 1733 blk_mq_unfreeze_queue(lo->lo_queue); 1734 } 1735 1736 out_unlock: 1737 mutex_unlock(&loop_ctl_mutex); 1738 } 1739 1740 static const struct block_device_operations lo_fops = { 1741 .owner = THIS_MODULE, 1742 .open = lo_open, 1743 .release = lo_release, 1744 .ioctl = lo_ioctl, 1745 #ifdef CONFIG_COMPAT 1746 .compat_ioctl = lo_compat_ioctl, 1747 #endif 1748 }; 1749 1750 /* 1751 * And now the modules code and kernel interface. 1752 */ 1753 static int max_loop; 1754 module_param(max_loop, int, 0444); 1755 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1756 module_param(max_part, int, 0444); 1757 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1758 MODULE_LICENSE("GPL"); 1759 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1760 1761 int loop_register_transfer(struct loop_func_table *funcs) 1762 { 1763 unsigned int n = funcs->number; 1764 1765 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1766 return -EINVAL; 1767 xfer_funcs[n] = funcs; 1768 return 0; 1769 } 1770 1771 static int unregister_transfer_cb(int id, void *ptr, void *data) 1772 { 1773 struct loop_device *lo = ptr; 1774 struct loop_func_table *xfer = data; 1775 1776 mutex_lock(&loop_ctl_mutex); 1777 if (lo->lo_encryption == xfer) 1778 loop_release_xfer(lo); 1779 mutex_unlock(&loop_ctl_mutex); 1780 return 0; 1781 } 1782 1783 int loop_unregister_transfer(int number) 1784 { 1785 unsigned int n = number; 1786 struct loop_func_table *xfer; 1787 1788 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1789 return -EINVAL; 1790 1791 xfer_funcs[n] = NULL; 1792 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1793 return 0; 1794 } 1795 1796 EXPORT_SYMBOL(loop_register_transfer); 1797 EXPORT_SYMBOL(loop_unregister_transfer); 1798 1799 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1800 const struct blk_mq_queue_data *bd) 1801 { 1802 struct request *rq = bd->rq; 1803 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1804 struct loop_device *lo = rq->q->queuedata; 1805 1806 blk_mq_start_request(rq); 1807 1808 if (lo->lo_state != Lo_bound) 1809 return BLK_STS_IOERR; 1810 1811 switch (req_op(rq)) { 1812 case REQ_OP_FLUSH: 1813 case REQ_OP_DISCARD: 1814 case REQ_OP_WRITE_ZEROES: 1815 cmd->use_aio = false; 1816 break; 1817 default: 1818 cmd->use_aio = lo->use_dio; 1819 break; 1820 } 1821 1822 /* always use the first bio's css */ 1823 #ifdef CONFIG_BLK_CGROUP 1824 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 1825 cmd->css = &bio_blkcg(rq->bio)->css; 1826 css_get(cmd->css); 1827 } else 1828 #endif 1829 cmd->css = NULL; 1830 kthread_queue_work(&lo->worker, &cmd->work); 1831 1832 return BLK_STS_OK; 1833 } 1834 1835 static void loop_handle_cmd(struct loop_cmd *cmd) 1836 { 1837 struct request *rq = blk_mq_rq_from_pdu(cmd); 1838 const bool write = op_is_write(req_op(rq)); 1839 struct loop_device *lo = rq->q->queuedata; 1840 int ret = 0; 1841 1842 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1843 ret = -EIO; 1844 goto failed; 1845 } 1846 1847 ret = do_req_filebacked(lo, rq); 1848 failed: 1849 /* complete non-aio request */ 1850 if (!cmd->use_aio || ret) { 1851 cmd->ret = ret ? -EIO : 0; 1852 blk_mq_complete_request(rq); 1853 } 1854 } 1855 1856 static void loop_queue_work(struct kthread_work *work) 1857 { 1858 struct loop_cmd *cmd = 1859 container_of(work, struct loop_cmd, work); 1860 1861 loop_handle_cmd(cmd); 1862 } 1863 1864 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1865 unsigned int hctx_idx, unsigned int numa_node) 1866 { 1867 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1868 1869 kthread_init_work(&cmd->work, loop_queue_work); 1870 return 0; 1871 } 1872 1873 static const struct blk_mq_ops loop_mq_ops = { 1874 .queue_rq = loop_queue_rq, 1875 .init_request = loop_init_request, 1876 .complete = lo_complete_rq, 1877 }; 1878 1879 static int loop_add(struct loop_device **l, int i) 1880 { 1881 struct loop_device *lo; 1882 struct gendisk *disk; 1883 int err; 1884 1885 err = -ENOMEM; 1886 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1887 if (!lo) 1888 goto out; 1889 1890 lo->lo_state = Lo_unbound; 1891 1892 /* allocate id, if @id >= 0, we're requesting that specific id */ 1893 if (i >= 0) { 1894 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1895 if (err == -ENOSPC) 1896 err = -EEXIST; 1897 } else { 1898 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1899 } 1900 if (err < 0) 1901 goto out_free_dev; 1902 i = err; 1903 1904 err = -ENOMEM; 1905 lo->tag_set.ops = &loop_mq_ops; 1906 lo->tag_set.nr_hw_queues = 1; 1907 lo->tag_set.queue_depth = 128; 1908 lo->tag_set.numa_node = NUMA_NO_NODE; 1909 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1910 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1911 lo->tag_set.driver_data = lo; 1912 1913 err = blk_mq_alloc_tag_set(&lo->tag_set); 1914 if (err) 1915 goto out_free_idr; 1916 1917 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1918 if (IS_ERR(lo->lo_queue)) { 1919 err = PTR_ERR(lo->lo_queue); 1920 goto out_cleanup_tags; 1921 } 1922 lo->lo_queue->queuedata = lo; 1923 1924 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1925 1926 /* 1927 * By default, we do buffer IO, so it doesn't make sense to enable 1928 * merge because the I/O submitted to backing file is handled page by 1929 * page. For directio mode, merge does help to dispatch bigger request 1930 * to underlayer disk. We will enable merge once directio is enabled. 1931 */ 1932 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1933 1934 err = -ENOMEM; 1935 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1936 if (!disk) 1937 goto out_free_queue; 1938 1939 /* 1940 * Disable partition scanning by default. The in-kernel partition 1941 * scanning can be requested individually per-device during its 1942 * setup. Userspace can always add and remove partitions from all 1943 * devices. The needed partition minors are allocated from the 1944 * extended minor space, the main loop device numbers will continue 1945 * to match the loop minors, regardless of the number of partitions 1946 * used. 1947 * 1948 * If max_part is given, partition scanning is globally enabled for 1949 * all loop devices. The minors for the main loop devices will be 1950 * multiples of max_part. 1951 * 1952 * Note: Global-for-all-devices, set-only-at-init, read-only module 1953 * parameteters like 'max_loop' and 'max_part' make things needlessly 1954 * complicated, are too static, inflexible and may surprise 1955 * userspace tools. Parameters like this in general should be avoided. 1956 */ 1957 if (!part_shift) 1958 disk->flags |= GENHD_FL_NO_PART_SCAN; 1959 disk->flags |= GENHD_FL_EXT_DEVT; 1960 atomic_set(&lo->lo_refcnt, 0); 1961 lo->lo_number = i; 1962 spin_lock_init(&lo->lo_lock); 1963 disk->major = LOOP_MAJOR; 1964 disk->first_minor = i << part_shift; 1965 disk->fops = &lo_fops; 1966 disk->private_data = lo; 1967 disk->queue = lo->lo_queue; 1968 sprintf(disk->disk_name, "loop%d", i); 1969 add_disk(disk); 1970 *l = lo; 1971 return lo->lo_number; 1972 1973 out_free_queue: 1974 blk_cleanup_queue(lo->lo_queue); 1975 out_cleanup_tags: 1976 blk_mq_free_tag_set(&lo->tag_set); 1977 out_free_idr: 1978 idr_remove(&loop_index_idr, i); 1979 out_free_dev: 1980 kfree(lo); 1981 out: 1982 return err; 1983 } 1984 1985 static void loop_remove(struct loop_device *lo) 1986 { 1987 del_gendisk(lo->lo_disk); 1988 blk_cleanup_queue(lo->lo_queue); 1989 blk_mq_free_tag_set(&lo->tag_set); 1990 put_disk(lo->lo_disk); 1991 kfree(lo); 1992 } 1993 1994 static int find_free_cb(int id, void *ptr, void *data) 1995 { 1996 struct loop_device *lo = ptr; 1997 struct loop_device **l = data; 1998 1999 if (lo->lo_state == Lo_unbound) { 2000 *l = lo; 2001 return 1; 2002 } 2003 return 0; 2004 } 2005 2006 static int loop_lookup(struct loop_device **l, int i) 2007 { 2008 struct loop_device *lo; 2009 int ret = -ENODEV; 2010 2011 if (i < 0) { 2012 int err; 2013 2014 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 2015 if (err == 1) { 2016 *l = lo; 2017 ret = lo->lo_number; 2018 } 2019 goto out; 2020 } 2021 2022 /* lookup and return a specific i */ 2023 lo = idr_find(&loop_index_idr, i); 2024 if (lo) { 2025 *l = lo; 2026 ret = lo->lo_number; 2027 } 2028 out: 2029 return ret; 2030 } 2031 2032 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 2033 { 2034 struct loop_device *lo; 2035 struct kobject *kobj; 2036 int err; 2037 2038 mutex_lock(&loop_ctl_mutex); 2039 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 2040 if (err < 0) 2041 err = loop_add(&lo, MINOR(dev) >> part_shift); 2042 if (err < 0) 2043 kobj = NULL; 2044 else 2045 kobj = get_disk_and_module(lo->lo_disk); 2046 mutex_unlock(&loop_ctl_mutex); 2047 2048 *part = 0; 2049 return kobj; 2050 } 2051 2052 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2053 unsigned long parm) 2054 { 2055 struct loop_device *lo; 2056 int ret; 2057 2058 ret = mutex_lock_killable(&loop_ctl_mutex); 2059 if (ret) 2060 return ret; 2061 2062 ret = -ENOSYS; 2063 switch (cmd) { 2064 case LOOP_CTL_ADD: 2065 ret = loop_lookup(&lo, parm); 2066 if (ret >= 0) { 2067 ret = -EEXIST; 2068 break; 2069 } 2070 ret = loop_add(&lo, parm); 2071 break; 2072 case LOOP_CTL_REMOVE: 2073 ret = loop_lookup(&lo, parm); 2074 if (ret < 0) 2075 break; 2076 if (lo->lo_state != Lo_unbound) { 2077 ret = -EBUSY; 2078 break; 2079 } 2080 if (atomic_read(&lo->lo_refcnt) > 0) { 2081 ret = -EBUSY; 2082 break; 2083 } 2084 lo->lo_disk->private_data = NULL; 2085 idr_remove(&loop_index_idr, lo->lo_number); 2086 loop_remove(lo); 2087 break; 2088 case LOOP_CTL_GET_FREE: 2089 ret = loop_lookup(&lo, -1); 2090 if (ret >= 0) 2091 break; 2092 ret = loop_add(&lo, -1); 2093 } 2094 mutex_unlock(&loop_ctl_mutex); 2095 2096 return ret; 2097 } 2098 2099 static const struct file_operations loop_ctl_fops = { 2100 .open = nonseekable_open, 2101 .unlocked_ioctl = loop_control_ioctl, 2102 .compat_ioctl = loop_control_ioctl, 2103 .owner = THIS_MODULE, 2104 .llseek = noop_llseek, 2105 }; 2106 2107 static struct miscdevice loop_misc = { 2108 .minor = LOOP_CTRL_MINOR, 2109 .name = "loop-control", 2110 .fops = &loop_ctl_fops, 2111 }; 2112 2113 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2114 MODULE_ALIAS("devname:loop-control"); 2115 2116 static int __init loop_init(void) 2117 { 2118 int i, nr; 2119 unsigned long range; 2120 struct loop_device *lo; 2121 int err; 2122 2123 part_shift = 0; 2124 if (max_part > 0) { 2125 part_shift = fls(max_part); 2126 2127 /* 2128 * Adjust max_part according to part_shift as it is exported 2129 * to user space so that user can decide correct minor number 2130 * if [s]he want to create more devices. 2131 * 2132 * Note that -1 is required because partition 0 is reserved 2133 * for the whole disk. 2134 */ 2135 max_part = (1UL << part_shift) - 1; 2136 } 2137 2138 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2139 err = -EINVAL; 2140 goto err_out; 2141 } 2142 2143 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2144 err = -EINVAL; 2145 goto err_out; 2146 } 2147 2148 /* 2149 * If max_loop is specified, create that many devices upfront. 2150 * This also becomes a hard limit. If max_loop is not specified, 2151 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2152 * init time. Loop devices can be requested on-demand with the 2153 * /dev/loop-control interface, or be instantiated by accessing 2154 * a 'dead' device node. 2155 */ 2156 if (max_loop) { 2157 nr = max_loop; 2158 range = max_loop << part_shift; 2159 } else { 2160 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2161 range = 1UL << MINORBITS; 2162 } 2163 2164 err = misc_register(&loop_misc); 2165 if (err < 0) 2166 goto err_out; 2167 2168 2169 if (register_blkdev(LOOP_MAJOR, "loop")) { 2170 err = -EIO; 2171 goto misc_out; 2172 } 2173 2174 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2175 THIS_MODULE, loop_probe, NULL, NULL); 2176 2177 /* pre-create number of devices given by config or max_loop */ 2178 mutex_lock(&loop_ctl_mutex); 2179 for (i = 0; i < nr; i++) 2180 loop_add(&lo, i); 2181 mutex_unlock(&loop_ctl_mutex); 2182 2183 printk(KERN_INFO "loop: module loaded\n"); 2184 return 0; 2185 2186 misc_out: 2187 misc_deregister(&loop_misc); 2188 err_out: 2189 return err; 2190 } 2191 2192 static int loop_exit_cb(int id, void *ptr, void *data) 2193 { 2194 struct loop_device *lo = ptr; 2195 2196 loop_remove(lo); 2197 return 0; 2198 } 2199 2200 static void __exit loop_exit(void) 2201 { 2202 unsigned long range; 2203 2204 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2205 2206 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2207 idr_destroy(&loop_index_idr); 2208 2209 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2210 unregister_blkdev(LOOP_MAJOR, "loop"); 2211 2212 misc_deregister(&loop_misc); 2213 } 2214 2215 module_init(loop_init); 2216 module_exit(loop_exit); 2217 2218 #ifndef MODULE 2219 static int __init max_loop_setup(char *str) 2220 { 2221 max_loop = simple_strtol(str, NULL, 0); 2222 return 1; 2223 } 2224 2225 __setup("max_loop=", max_loop_setup); 2226 #endif 2227