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