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 = -EBUSY; 922 if (lo->lo_state != Lo_unbound) 923 goto out_putf; 924 925 error = loop_validate_file(file, bdev); 926 if (error) 927 goto out_putf; 928 929 mapping = file->f_mapping; 930 inode = mapping->host; 931 932 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 933 !file->f_op->write_iter) 934 lo_flags |= LO_FLAGS_READ_ONLY; 935 936 error = -EFBIG; 937 size = get_loop_size(lo, file); 938 if ((loff_t)(sector_t)size != size) 939 goto out_putf; 940 error = loop_prepare_queue(lo); 941 if (error) 942 goto out_putf; 943 944 error = 0; 945 946 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 947 948 lo->use_dio = false; 949 lo->lo_device = bdev; 950 lo->lo_flags = lo_flags; 951 lo->lo_backing_file = file; 952 lo->transfer = NULL; 953 lo->ioctl = NULL; 954 lo->lo_sizelimit = 0; 955 lo->old_gfp_mask = mapping_gfp_mask(mapping); 956 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 957 958 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 959 blk_queue_write_cache(lo->lo_queue, true, false); 960 961 loop_update_dio(lo); 962 set_capacity(lo->lo_disk, size); 963 bd_set_size(bdev, size << 9); 964 loop_sysfs_init(lo); 965 /* let user-space know about the new size */ 966 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 967 968 set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 969 block_size(inode->i_bdev) : PAGE_SIZE); 970 971 lo->lo_state = Lo_bound; 972 if (part_shift) 973 lo->lo_flags |= LO_FLAGS_PARTSCAN; 974 if (lo->lo_flags & LO_FLAGS_PARTSCAN) 975 loop_reread_partitions(lo, bdev); 976 977 /* Grab the block_device to prevent its destruction after we 978 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 979 */ 980 bdgrab(bdev); 981 return 0; 982 983 out_putf: 984 fput(file); 985 out: 986 /* This is safe: open() is still holding a reference. */ 987 module_put(THIS_MODULE); 988 return error; 989 } 990 991 static int 992 loop_release_xfer(struct loop_device *lo) 993 { 994 int err = 0; 995 struct loop_func_table *xfer = lo->lo_encryption; 996 997 if (xfer) { 998 if (xfer->release) 999 err = xfer->release(lo); 1000 lo->transfer = NULL; 1001 lo->lo_encryption = NULL; 1002 module_put(xfer->owner); 1003 } 1004 return err; 1005 } 1006 1007 static int 1008 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 1009 const struct loop_info64 *i) 1010 { 1011 int err = 0; 1012 1013 if (xfer) { 1014 struct module *owner = xfer->owner; 1015 1016 if (!try_module_get(owner)) 1017 return -EINVAL; 1018 if (xfer->init) 1019 err = xfer->init(lo, i); 1020 if (err) 1021 module_put(owner); 1022 else 1023 lo->lo_encryption = xfer; 1024 } 1025 return err; 1026 } 1027 1028 static int __loop_clr_fd(struct loop_device *lo) 1029 { 1030 struct file *filp = NULL; 1031 gfp_t gfp = lo->old_gfp_mask; 1032 struct block_device *bdev = lo->lo_device; 1033 int err = 0; 1034 1035 mutex_lock(&loop_ctl_mutex); 1036 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 1037 err = -ENXIO; 1038 goto out_unlock; 1039 } 1040 1041 filp = lo->lo_backing_file; 1042 if (filp == NULL) { 1043 err = -EINVAL; 1044 goto out_unlock; 1045 } 1046 1047 /* freeze request queue during the transition */ 1048 blk_mq_freeze_queue(lo->lo_queue); 1049 1050 spin_lock_irq(&lo->lo_lock); 1051 lo->lo_backing_file = NULL; 1052 spin_unlock_irq(&lo->lo_lock); 1053 1054 loop_release_xfer(lo); 1055 lo->transfer = NULL; 1056 lo->ioctl = NULL; 1057 lo->lo_device = NULL; 1058 lo->lo_encryption = NULL; 1059 lo->lo_offset = 0; 1060 lo->lo_sizelimit = 0; 1061 lo->lo_encrypt_key_size = 0; 1062 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 1063 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 1064 memset(lo->lo_file_name, 0, LO_NAME_SIZE); 1065 blk_queue_logical_block_size(lo->lo_queue, 512); 1066 blk_queue_physical_block_size(lo->lo_queue, 512); 1067 blk_queue_io_min(lo->lo_queue, 512); 1068 if (bdev) { 1069 bdput(bdev); 1070 invalidate_bdev(bdev); 1071 bdev->bd_inode->i_mapping->wb_err = 0; 1072 } 1073 set_capacity(lo->lo_disk, 0); 1074 loop_sysfs_exit(lo); 1075 if (bdev) { 1076 bd_set_size(bdev, 0); 1077 /* let user-space know about this change */ 1078 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1079 } 1080 mapping_set_gfp_mask(filp->f_mapping, gfp); 1081 lo->lo_state = Lo_unbound; 1082 /* This is safe: open() is still holding a reference. */ 1083 module_put(THIS_MODULE); 1084 blk_mq_unfreeze_queue(lo->lo_queue); 1085 1086 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) 1087 loop_reread_partitions(lo, bdev); 1088 lo->lo_flags = 0; 1089 if (!part_shift) 1090 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1091 loop_unprepare_queue(lo); 1092 out_unlock: 1093 mutex_unlock(&loop_ctl_mutex); 1094 /* 1095 * Need not hold loop_ctl_mutex to fput backing file. 1096 * Calling fput holding loop_ctl_mutex triggers a circular 1097 * lock dependency possibility warning as fput can take 1098 * bd_mutex which is usually taken before loop_ctl_mutex. 1099 */ 1100 if (filp) 1101 fput(filp); 1102 return err; 1103 } 1104 1105 static int loop_clr_fd(struct loop_device *lo) 1106 { 1107 int err; 1108 1109 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1110 if (err) 1111 return err; 1112 if (lo->lo_state != Lo_bound) { 1113 mutex_unlock(&loop_ctl_mutex); 1114 return -ENXIO; 1115 } 1116 /* 1117 * If we've explicitly asked to tear down the loop device, 1118 * and it has an elevated reference count, set it for auto-teardown when 1119 * the last reference goes away. This stops $!~#$@ udev from 1120 * preventing teardown because it decided that it needs to run blkid on 1121 * the loopback device whenever they appear. xfstests is notorious for 1122 * failing tests because blkid via udev races with a losetup 1123 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1124 * command to fail with EBUSY. 1125 */ 1126 if (atomic_read(&lo->lo_refcnt) > 1) { 1127 lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1128 mutex_unlock(&loop_ctl_mutex); 1129 return 0; 1130 } 1131 lo->lo_state = Lo_rundown; 1132 mutex_unlock(&loop_ctl_mutex); 1133 1134 return __loop_clr_fd(lo); 1135 } 1136 1137 static int 1138 loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 1139 { 1140 int err; 1141 struct loop_func_table *xfer; 1142 kuid_t uid = current_uid(); 1143 1144 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1145 if (err) 1146 return err; 1147 if (lo->lo_encrypt_key_size && 1148 !uid_eq(lo->lo_key_owner, uid) && 1149 !capable(CAP_SYS_ADMIN)) { 1150 err = -EPERM; 1151 goto out_unlock; 1152 } 1153 if (lo->lo_state != Lo_bound) { 1154 err = -ENXIO; 1155 goto out_unlock; 1156 } 1157 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) { 1158 err = -EINVAL; 1159 goto out_unlock; 1160 } 1161 1162 /* I/O need to be drained during transfer transition */ 1163 blk_mq_freeze_queue(lo->lo_queue); 1164 1165 err = loop_release_xfer(lo); 1166 if (err) 1167 goto out_unfreeze; 1168 1169 if (info->lo_encrypt_type) { 1170 unsigned int type = info->lo_encrypt_type; 1171 1172 if (type >= MAX_LO_CRYPT) { 1173 err = -EINVAL; 1174 goto out_unfreeze; 1175 } 1176 xfer = xfer_funcs[type]; 1177 if (xfer == NULL) { 1178 err = -EINVAL; 1179 goto out_unfreeze; 1180 } 1181 } else 1182 xfer = NULL; 1183 1184 err = loop_init_xfer(lo, xfer, info); 1185 if (err) 1186 goto out_unfreeze; 1187 1188 if (lo->lo_offset != info->lo_offset || 1189 lo->lo_sizelimit != info->lo_sizelimit) { 1190 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { 1191 err = -EFBIG; 1192 goto out_unfreeze; 1193 } 1194 } 1195 1196 loop_config_discard(lo); 1197 1198 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 1199 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 1200 lo->lo_file_name[LO_NAME_SIZE-1] = 0; 1201 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 1202 1203 if (!xfer) 1204 xfer = &none_funcs; 1205 lo->transfer = xfer->transfer; 1206 lo->ioctl = xfer->ioctl; 1207 1208 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 1209 (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 1210 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 1211 1212 lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 1213 lo->lo_init[0] = info->lo_init[0]; 1214 lo->lo_init[1] = info->lo_init[1]; 1215 if (info->lo_encrypt_key_size) { 1216 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 1217 info->lo_encrypt_key_size); 1218 lo->lo_key_owner = uid; 1219 } 1220 1221 /* update dio if lo_offset or transfer is changed */ 1222 __loop_update_dio(lo, lo->use_dio); 1223 1224 out_unfreeze: 1225 blk_mq_unfreeze_queue(lo->lo_queue); 1226 1227 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1228 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1229 lo->lo_flags |= LO_FLAGS_PARTSCAN; 1230 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 1231 loop_reread_partitions(lo, lo->lo_device); 1232 } 1233 out_unlock: 1234 mutex_unlock(&loop_ctl_mutex); 1235 1236 return err; 1237 } 1238 1239 static int 1240 loop_get_status(struct loop_device *lo, struct loop_info64 *info) 1241 { 1242 struct path path; 1243 struct kstat stat; 1244 int ret; 1245 1246 ret = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1247 if (ret) 1248 return ret; 1249 if (lo->lo_state != Lo_bound) { 1250 mutex_unlock(&loop_ctl_mutex); 1251 return -ENXIO; 1252 } 1253 1254 memset(info, 0, sizeof(*info)); 1255 info->lo_number = lo->lo_number; 1256 info->lo_offset = lo->lo_offset; 1257 info->lo_sizelimit = lo->lo_sizelimit; 1258 info->lo_flags = lo->lo_flags; 1259 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 1260 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 1261 info->lo_encrypt_type = 1262 lo->lo_encryption ? lo->lo_encryption->number : 0; 1263 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 1264 info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 1265 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 1266 lo->lo_encrypt_key_size); 1267 } 1268 1269 /* Drop loop_ctl_mutex while we call into the filesystem. */ 1270 path = lo->lo_backing_file->f_path; 1271 path_get(&path); 1272 mutex_unlock(&loop_ctl_mutex); 1273 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 1274 if (!ret) { 1275 info->lo_device = huge_encode_dev(stat.dev); 1276 info->lo_inode = stat.ino; 1277 info->lo_rdevice = huge_encode_dev(stat.rdev); 1278 } 1279 path_put(&path); 1280 return ret; 1281 } 1282 1283 static void 1284 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 1285 { 1286 memset(info64, 0, sizeof(*info64)); 1287 info64->lo_number = info->lo_number; 1288 info64->lo_device = info->lo_device; 1289 info64->lo_inode = info->lo_inode; 1290 info64->lo_rdevice = info->lo_rdevice; 1291 info64->lo_offset = info->lo_offset; 1292 info64->lo_sizelimit = 0; 1293 info64->lo_encrypt_type = info->lo_encrypt_type; 1294 info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 1295 info64->lo_flags = info->lo_flags; 1296 info64->lo_init[0] = info->lo_init[0]; 1297 info64->lo_init[1] = info->lo_init[1]; 1298 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1299 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 1300 else 1301 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 1302 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 1303 } 1304 1305 static int 1306 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 1307 { 1308 memset(info, 0, sizeof(*info)); 1309 info->lo_number = info64->lo_number; 1310 info->lo_device = info64->lo_device; 1311 info->lo_inode = info64->lo_inode; 1312 info->lo_rdevice = info64->lo_rdevice; 1313 info->lo_offset = info64->lo_offset; 1314 info->lo_encrypt_type = info64->lo_encrypt_type; 1315 info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 1316 info->lo_flags = info64->lo_flags; 1317 info->lo_init[0] = info64->lo_init[0]; 1318 info->lo_init[1] = info64->lo_init[1]; 1319 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1320 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1321 else 1322 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 1323 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1324 1325 /* error in case values were truncated */ 1326 if (info->lo_device != info64->lo_device || 1327 info->lo_rdevice != info64->lo_rdevice || 1328 info->lo_inode != info64->lo_inode || 1329 info->lo_offset != info64->lo_offset) 1330 return -EOVERFLOW; 1331 1332 return 0; 1333 } 1334 1335 static int 1336 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 1337 { 1338 struct loop_info info; 1339 struct loop_info64 info64; 1340 1341 if (copy_from_user(&info, arg, sizeof (struct loop_info))) 1342 return -EFAULT; 1343 loop_info64_from_old(&info, &info64); 1344 return loop_set_status(lo, &info64); 1345 } 1346 1347 static int 1348 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 1349 { 1350 struct loop_info64 info64; 1351 1352 if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 1353 return -EFAULT; 1354 return loop_set_status(lo, &info64); 1355 } 1356 1357 static int 1358 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 1359 struct loop_info info; 1360 struct loop_info64 info64; 1361 int err; 1362 1363 if (!arg) 1364 return -EINVAL; 1365 err = loop_get_status(lo, &info64); 1366 if (!err) 1367 err = loop_info64_to_old(&info64, &info); 1368 if (!err && copy_to_user(arg, &info, sizeof(info))) 1369 err = -EFAULT; 1370 1371 return err; 1372 } 1373 1374 static int 1375 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 1376 struct loop_info64 info64; 1377 int err; 1378 1379 if (!arg) 1380 return -EINVAL; 1381 err = loop_get_status(lo, &info64); 1382 if (!err && copy_to_user(arg, &info64, sizeof(info64))) 1383 err = -EFAULT; 1384 1385 return err; 1386 } 1387 1388 static int loop_set_capacity(struct loop_device *lo) 1389 { 1390 if (unlikely(lo->lo_state != Lo_bound)) 1391 return -ENXIO; 1392 1393 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1394 } 1395 1396 static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1397 { 1398 int error = -ENXIO; 1399 if (lo->lo_state != Lo_bound) 1400 goto out; 1401 1402 __loop_update_dio(lo, !!arg); 1403 if (lo->use_dio == !!arg) 1404 return 0; 1405 error = -EINVAL; 1406 out: 1407 return error; 1408 } 1409 1410 static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 1411 { 1412 if (lo->lo_state != Lo_bound) 1413 return -ENXIO; 1414 1415 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 1416 return -EINVAL; 1417 1418 blk_mq_freeze_queue(lo->lo_queue); 1419 1420 blk_queue_logical_block_size(lo->lo_queue, arg); 1421 blk_queue_physical_block_size(lo->lo_queue, arg); 1422 blk_queue_io_min(lo->lo_queue, arg); 1423 loop_update_dio(lo); 1424 1425 blk_mq_unfreeze_queue(lo->lo_queue); 1426 1427 return 0; 1428 } 1429 1430 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1431 unsigned long arg) 1432 { 1433 int err; 1434 1435 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1436 if (err) 1437 return err; 1438 switch (cmd) { 1439 case LOOP_SET_CAPACITY: 1440 err = loop_set_capacity(lo); 1441 break; 1442 case LOOP_SET_DIRECT_IO: 1443 err = loop_set_dio(lo, arg); 1444 break; 1445 case LOOP_SET_BLOCK_SIZE: 1446 err = loop_set_block_size(lo, arg); 1447 break; 1448 default: 1449 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 1450 } 1451 mutex_unlock(&loop_ctl_mutex); 1452 return err; 1453 } 1454 1455 static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1456 unsigned int cmd, unsigned long arg) 1457 { 1458 struct loop_device *lo = bdev->bd_disk->private_data; 1459 int err; 1460 1461 switch (cmd) { 1462 case LOOP_SET_FD: 1463 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1464 if (err) 1465 return err; 1466 err = loop_set_fd(lo, mode, bdev, arg); 1467 mutex_unlock(&loop_ctl_mutex); 1468 break; 1469 case LOOP_CHANGE_FD: 1470 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1); 1471 if (err) 1472 return err; 1473 err = loop_change_fd(lo, bdev, arg); 1474 mutex_unlock(&loop_ctl_mutex); 1475 break; 1476 case LOOP_CLR_FD: 1477 return loop_clr_fd(lo); 1478 case LOOP_SET_STATUS: 1479 err = -EPERM; 1480 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1481 err = loop_set_status_old(lo, 1482 (struct loop_info __user *)arg); 1483 } 1484 break; 1485 case LOOP_GET_STATUS: 1486 return loop_get_status_old(lo, (struct loop_info __user *) arg); 1487 case LOOP_SET_STATUS64: 1488 err = -EPERM; 1489 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1490 err = loop_set_status64(lo, 1491 (struct loop_info64 __user *) arg); 1492 } 1493 break; 1494 case LOOP_GET_STATUS64: 1495 return loop_get_status64(lo, (struct loop_info64 __user *) arg); 1496 case LOOP_SET_CAPACITY: 1497 case LOOP_SET_DIRECT_IO: 1498 case LOOP_SET_BLOCK_SIZE: 1499 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1500 return -EPERM; 1501 /* Fall through */ 1502 default: 1503 err = lo_simple_ioctl(lo, cmd, arg); 1504 break; 1505 } 1506 1507 return err; 1508 } 1509 1510 #ifdef CONFIG_COMPAT 1511 struct compat_loop_info { 1512 compat_int_t lo_number; /* ioctl r/o */ 1513 compat_dev_t lo_device; /* ioctl r/o */ 1514 compat_ulong_t lo_inode; /* ioctl r/o */ 1515 compat_dev_t lo_rdevice; /* ioctl r/o */ 1516 compat_int_t lo_offset; 1517 compat_int_t lo_encrypt_type; 1518 compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1519 compat_int_t lo_flags; /* ioctl r/o */ 1520 char lo_name[LO_NAME_SIZE]; 1521 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1522 compat_ulong_t lo_init[2]; 1523 char reserved[4]; 1524 }; 1525 1526 /* 1527 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1528 * - noinlined to reduce stack space usage in main part of driver 1529 */ 1530 static noinline int 1531 loop_info64_from_compat(const struct compat_loop_info __user *arg, 1532 struct loop_info64 *info64) 1533 { 1534 struct compat_loop_info info; 1535 1536 if (copy_from_user(&info, arg, sizeof(info))) 1537 return -EFAULT; 1538 1539 memset(info64, 0, sizeof(*info64)); 1540 info64->lo_number = info.lo_number; 1541 info64->lo_device = info.lo_device; 1542 info64->lo_inode = info.lo_inode; 1543 info64->lo_rdevice = info.lo_rdevice; 1544 info64->lo_offset = info.lo_offset; 1545 info64->lo_sizelimit = 0; 1546 info64->lo_encrypt_type = info.lo_encrypt_type; 1547 info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1548 info64->lo_flags = info.lo_flags; 1549 info64->lo_init[0] = info.lo_init[0]; 1550 info64->lo_init[1] = info.lo_init[1]; 1551 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1552 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1553 else 1554 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1555 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1556 return 0; 1557 } 1558 1559 /* 1560 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1561 * - noinlined to reduce stack space usage in main part of driver 1562 */ 1563 static noinline int 1564 loop_info64_to_compat(const struct loop_info64 *info64, 1565 struct compat_loop_info __user *arg) 1566 { 1567 struct compat_loop_info info; 1568 1569 memset(&info, 0, sizeof(info)); 1570 info.lo_number = info64->lo_number; 1571 info.lo_device = info64->lo_device; 1572 info.lo_inode = info64->lo_inode; 1573 info.lo_rdevice = info64->lo_rdevice; 1574 info.lo_offset = info64->lo_offset; 1575 info.lo_encrypt_type = info64->lo_encrypt_type; 1576 info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1577 info.lo_flags = info64->lo_flags; 1578 info.lo_init[0] = info64->lo_init[0]; 1579 info.lo_init[1] = info64->lo_init[1]; 1580 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1581 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1582 else 1583 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1584 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1585 1586 /* error in case values were truncated */ 1587 if (info.lo_device != info64->lo_device || 1588 info.lo_rdevice != info64->lo_rdevice || 1589 info.lo_inode != info64->lo_inode || 1590 info.lo_offset != info64->lo_offset || 1591 info.lo_init[0] != info64->lo_init[0] || 1592 info.lo_init[1] != info64->lo_init[1]) 1593 return -EOVERFLOW; 1594 1595 if (copy_to_user(arg, &info, sizeof(info))) 1596 return -EFAULT; 1597 return 0; 1598 } 1599 1600 static int 1601 loop_set_status_compat(struct loop_device *lo, 1602 const struct compat_loop_info __user *arg) 1603 { 1604 struct loop_info64 info64; 1605 int ret; 1606 1607 ret = loop_info64_from_compat(arg, &info64); 1608 if (ret < 0) 1609 return ret; 1610 return loop_set_status(lo, &info64); 1611 } 1612 1613 static int 1614 loop_get_status_compat(struct loop_device *lo, 1615 struct compat_loop_info __user *arg) 1616 { 1617 struct loop_info64 info64; 1618 int err; 1619 1620 if (!arg) 1621 return -EINVAL; 1622 err = loop_get_status(lo, &info64); 1623 if (!err) 1624 err = loop_info64_to_compat(&info64, arg); 1625 return err; 1626 } 1627 1628 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1629 unsigned int cmd, unsigned long arg) 1630 { 1631 struct loop_device *lo = bdev->bd_disk->private_data; 1632 int err; 1633 1634 switch(cmd) { 1635 case LOOP_SET_STATUS: 1636 err = loop_set_status_compat(lo, 1637 (const struct compat_loop_info __user *)arg); 1638 break; 1639 case LOOP_GET_STATUS: 1640 err = loop_get_status_compat(lo, 1641 (struct compat_loop_info __user *)arg); 1642 break; 1643 case LOOP_SET_CAPACITY: 1644 case LOOP_CLR_FD: 1645 case LOOP_GET_STATUS64: 1646 case LOOP_SET_STATUS64: 1647 arg = (unsigned long) compat_ptr(arg); 1648 /* fall through */ 1649 case LOOP_SET_FD: 1650 case LOOP_CHANGE_FD: 1651 case LOOP_SET_BLOCK_SIZE: 1652 err = lo_ioctl(bdev, mode, cmd, arg); 1653 break; 1654 default: 1655 err = -ENOIOCTLCMD; 1656 break; 1657 } 1658 return err; 1659 } 1660 #endif 1661 1662 static int lo_open(struct block_device *bdev, fmode_t mode) 1663 { 1664 struct loop_device *lo; 1665 int err; 1666 1667 err = mutex_lock_killable(&loop_ctl_mutex); 1668 if (err) 1669 return err; 1670 lo = bdev->bd_disk->private_data; 1671 if (!lo) { 1672 err = -ENXIO; 1673 goto out; 1674 } 1675 1676 atomic_inc(&lo->lo_refcnt); 1677 out: 1678 mutex_unlock(&loop_ctl_mutex); 1679 return err; 1680 } 1681 1682 static void lo_release(struct gendisk *disk, fmode_t mode) 1683 { 1684 struct loop_device *lo; 1685 1686 mutex_lock(&loop_ctl_mutex); 1687 lo = disk->private_data; 1688 if (atomic_dec_return(&lo->lo_refcnt)) 1689 goto out_unlock; 1690 1691 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1692 if (lo->lo_state != Lo_bound) 1693 goto out_unlock; 1694 lo->lo_state = Lo_rundown; 1695 mutex_unlock(&loop_ctl_mutex); 1696 /* 1697 * In autoclear mode, stop the loop thread 1698 * and remove configuration after last close. 1699 */ 1700 __loop_clr_fd(lo); 1701 return; 1702 } else if (lo->lo_state == Lo_bound) { 1703 /* 1704 * Otherwise keep thread (if running) and config, 1705 * but flush possible ongoing bios in thread. 1706 */ 1707 blk_mq_freeze_queue(lo->lo_queue); 1708 blk_mq_unfreeze_queue(lo->lo_queue); 1709 } 1710 1711 out_unlock: 1712 mutex_unlock(&loop_ctl_mutex); 1713 } 1714 1715 static const struct block_device_operations lo_fops = { 1716 .owner = THIS_MODULE, 1717 .open = lo_open, 1718 .release = lo_release, 1719 .ioctl = lo_ioctl, 1720 #ifdef CONFIG_COMPAT 1721 .compat_ioctl = lo_compat_ioctl, 1722 #endif 1723 }; 1724 1725 /* 1726 * And now the modules code and kernel interface. 1727 */ 1728 static int max_loop; 1729 module_param(max_loop, int, 0444); 1730 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 1731 module_param(max_part, int, 0444); 1732 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1733 MODULE_LICENSE("GPL"); 1734 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 1735 1736 int loop_register_transfer(struct loop_func_table *funcs) 1737 { 1738 unsigned int n = funcs->number; 1739 1740 if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 1741 return -EINVAL; 1742 xfer_funcs[n] = funcs; 1743 return 0; 1744 } 1745 1746 static int unregister_transfer_cb(int id, void *ptr, void *data) 1747 { 1748 struct loop_device *lo = ptr; 1749 struct loop_func_table *xfer = data; 1750 1751 mutex_lock(&loop_ctl_mutex); 1752 if (lo->lo_encryption == xfer) 1753 loop_release_xfer(lo); 1754 mutex_unlock(&loop_ctl_mutex); 1755 return 0; 1756 } 1757 1758 int loop_unregister_transfer(int number) 1759 { 1760 unsigned int n = number; 1761 struct loop_func_table *xfer; 1762 1763 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 1764 return -EINVAL; 1765 1766 xfer_funcs[n] = NULL; 1767 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 1768 return 0; 1769 } 1770 1771 EXPORT_SYMBOL(loop_register_transfer); 1772 EXPORT_SYMBOL(loop_unregister_transfer); 1773 1774 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1775 const struct blk_mq_queue_data *bd) 1776 { 1777 struct request *rq = bd->rq; 1778 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1779 struct loop_device *lo = rq->q->queuedata; 1780 1781 blk_mq_start_request(rq); 1782 1783 if (lo->lo_state != Lo_bound) 1784 return BLK_STS_IOERR; 1785 1786 switch (req_op(rq)) { 1787 case REQ_OP_FLUSH: 1788 case REQ_OP_DISCARD: 1789 case REQ_OP_WRITE_ZEROES: 1790 cmd->use_aio = false; 1791 break; 1792 default: 1793 cmd->use_aio = lo->use_dio; 1794 break; 1795 } 1796 1797 /* always use the first bio's css */ 1798 #ifdef CONFIG_BLK_CGROUP 1799 if (cmd->use_aio && rq->bio && rq->bio->bi_css) { 1800 cmd->css = rq->bio->bi_css; 1801 css_get(cmd->css); 1802 } else 1803 #endif 1804 cmd->css = NULL; 1805 kthread_queue_work(&lo->worker, &cmd->work); 1806 1807 return BLK_STS_OK; 1808 } 1809 1810 static void loop_handle_cmd(struct loop_cmd *cmd) 1811 { 1812 struct request *rq = blk_mq_rq_from_pdu(cmd); 1813 const bool write = op_is_write(req_op(rq)); 1814 struct loop_device *lo = rq->q->queuedata; 1815 int ret = 0; 1816 1817 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1818 ret = -EIO; 1819 goto failed; 1820 } 1821 1822 ret = do_req_filebacked(lo, rq); 1823 failed: 1824 /* complete non-aio request */ 1825 if (!cmd->use_aio || ret) { 1826 cmd->ret = ret ? -EIO : 0; 1827 blk_mq_complete_request(rq); 1828 } 1829 } 1830 1831 static void loop_queue_work(struct kthread_work *work) 1832 { 1833 struct loop_cmd *cmd = 1834 container_of(work, struct loop_cmd, work); 1835 1836 loop_handle_cmd(cmd); 1837 } 1838 1839 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 1840 unsigned int hctx_idx, unsigned int numa_node) 1841 { 1842 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 1843 1844 kthread_init_work(&cmd->work, loop_queue_work); 1845 return 0; 1846 } 1847 1848 static const struct blk_mq_ops loop_mq_ops = { 1849 .queue_rq = loop_queue_rq, 1850 .init_request = loop_init_request, 1851 .complete = lo_complete_rq, 1852 }; 1853 1854 static int loop_add(struct loop_device **l, int i) 1855 { 1856 struct loop_device *lo; 1857 struct gendisk *disk; 1858 int err; 1859 1860 err = -ENOMEM; 1861 lo = kzalloc(sizeof(*lo), GFP_KERNEL); 1862 if (!lo) 1863 goto out; 1864 1865 lo->lo_state = Lo_unbound; 1866 1867 /* allocate id, if @id >= 0, we're requesting that specific id */ 1868 if (i >= 0) { 1869 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1870 if (err == -ENOSPC) 1871 err = -EEXIST; 1872 } else { 1873 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 1874 } 1875 if (err < 0) 1876 goto out_free_dev; 1877 i = err; 1878 1879 err = -ENOMEM; 1880 lo->tag_set.ops = &loop_mq_ops; 1881 lo->tag_set.nr_hw_queues = 1; 1882 lo->tag_set.queue_depth = 128; 1883 lo->tag_set.numa_node = NUMA_NO_NODE; 1884 lo->tag_set.cmd_size = sizeof(struct loop_cmd); 1885 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; 1886 lo->tag_set.driver_data = lo; 1887 1888 err = blk_mq_alloc_tag_set(&lo->tag_set); 1889 if (err) 1890 goto out_free_idr; 1891 1892 lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 1893 if (IS_ERR_OR_NULL(lo->lo_queue)) { 1894 err = PTR_ERR(lo->lo_queue); 1895 goto out_cleanup_tags; 1896 } 1897 lo->lo_queue->queuedata = lo; 1898 1899 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 1900 1901 /* 1902 * By default, we do buffer IO, so it doesn't make sense to enable 1903 * merge because the I/O submitted to backing file is handled page by 1904 * page. For directio mode, merge does help to dispatch bigger request 1905 * to underlayer disk. We will enable merge once directio is enabled. 1906 */ 1907 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 1908 1909 err = -ENOMEM; 1910 disk = lo->lo_disk = alloc_disk(1 << part_shift); 1911 if (!disk) 1912 goto out_free_queue; 1913 1914 /* 1915 * Disable partition scanning by default. The in-kernel partition 1916 * scanning can be requested individually per-device during its 1917 * setup. Userspace can always add and remove partitions from all 1918 * devices. The needed partition minors are allocated from the 1919 * extended minor space, the main loop device numbers will continue 1920 * to match the loop minors, regardless of the number of partitions 1921 * used. 1922 * 1923 * If max_part is given, partition scanning is globally enabled for 1924 * all loop devices. The minors for the main loop devices will be 1925 * multiples of max_part. 1926 * 1927 * Note: Global-for-all-devices, set-only-at-init, read-only module 1928 * parameteters like 'max_loop' and 'max_part' make things needlessly 1929 * complicated, are too static, inflexible and may surprise 1930 * userspace tools. Parameters like this in general should be avoided. 1931 */ 1932 if (!part_shift) 1933 disk->flags |= GENHD_FL_NO_PART_SCAN; 1934 disk->flags |= GENHD_FL_EXT_DEVT; 1935 atomic_set(&lo->lo_refcnt, 0); 1936 lo->lo_number = i; 1937 spin_lock_init(&lo->lo_lock); 1938 disk->major = LOOP_MAJOR; 1939 disk->first_minor = i << part_shift; 1940 disk->fops = &lo_fops; 1941 disk->private_data = lo; 1942 disk->queue = lo->lo_queue; 1943 sprintf(disk->disk_name, "loop%d", i); 1944 add_disk(disk); 1945 *l = lo; 1946 return lo->lo_number; 1947 1948 out_free_queue: 1949 blk_cleanup_queue(lo->lo_queue); 1950 out_cleanup_tags: 1951 blk_mq_free_tag_set(&lo->tag_set); 1952 out_free_idr: 1953 idr_remove(&loop_index_idr, i); 1954 out_free_dev: 1955 kfree(lo); 1956 out: 1957 return err; 1958 } 1959 1960 static void loop_remove(struct loop_device *lo) 1961 { 1962 del_gendisk(lo->lo_disk); 1963 blk_cleanup_queue(lo->lo_queue); 1964 blk_mq_free_tag_set(&lo->tag_set); 1965 put_disk(lo->lo_disk); 1966 kfree(lo); 1967 } 1968 1969 static int find_free_cb(int id, void *ptr, void *data) 1970 { 1971 struct loop_device *lo = ptr; 1972 struct loop_device **l = data; 1973 1974 if (lo->lo_state == Lo_unbound) { 1975 *l = lo; 1976 return 1; 1977 } 1978 return 0; 1979 } 1980 1981 static int loop_lookup(struct loop_device **l, int i) 1982 { 1983 struct loop_device *lo; 1984 int ret = -ENODEV; 1985 1986 if (i < 0) { 1987 int err; 1988 1989 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 1990 if (err == 1) { 1991 *l = lo; 1992 ret = lo->lo_number; 1993 } 1994 goto out; 1995 } 1996 1997 /* lookup and return a specific i */ 1998 lo = idr_find(&loop_index_idr, i); 1999 if (lo) { 2000 *l = lo; 2001 ret = lo->lo_number; 2002 } 2003 out: 2004 return ret; 2005 } 2006 2007 static struct kobject *loop_probe(dev_t dev, int *part, void *data) 2008 { 2009 struct loop_device *lo; 2010 struct kobject *kobj; 2011 int err; 2012 2013 mutex_lock(&loop_ctl_mutex); 2014 err = loop_lookup(&lo, MINOR(dev) >> part_shift); 2015 if (err < 0) 2016 err = loop_add(&lo, MINOR(dev) >> part_shift); 2017 if (err < 0) 2018 kobj = NULL; 2019 else 2020 kobj = get_disk_and_module(lo->lo_disk); 2021 mutex_unlock(&loop_ctl_mutex); 2022 2023 *part = 0; 2024 return kobj; 2025 } 2026 2027 static long loop_control_ioctl(struct file *file, unsigned int cmd, 2028 unsigned long parm) 2029 { 2030 struct loop_device *lo; 2031 int ret; 2032 2033 ret = mutex_lock_killable(&loop_ctl_mutex); 2034 if (ret) 2035 return ret; 2036 2037 ret = -ENOSYS; 2038 switch (cmd) { 2039 case LOOP_CTL_ADD: 2040 ret = loop_lookup(&lo, parm); 2041 if (ret >= 0) { 2042 ret = -EEXIST; 2043 break; 2044 } 2045 ret = loop_add(&lo, parm); 2046 break; 2047 case LOOP_CTL_REMOVE: 2048 ret = loop_lookup(&lo, parm); 2049 if (ret < 0) 2050 break; 2051 if (lo->lo_state != Lo_unbound) { 2052 ret = -EBUSY; 2053 mutex_unlock(&loop_ctl_mutex); 2054 break; 2055 } 2056 if (atomic_read(&lo->lo_refcnt) > 0) { 2057 ret = -EBUSY; 2058 mutex_unlock(&loop_ctl_mutex); 2059 break; 2060 } 2061 lo->lo_disk->private_data = NULL; 2062 idr_remove(&loop_index_idr, lo->lo_number); 2063 loop_remove(lo); 2064 break; 2065 case LOOP_CTL_GET_FREE: 2066 ret = loop_lookup(&lo, -1); 2067 if (ret >= 0) 2068 break; 2069 ret = loop_add(&lo, -1); 2070 } 2071 mutex_unlock(&loop_ctl_mutex); 2072 2073 return ret; 2074 } 2075 2076 static const struct file_operations loop_ctl_fops = { 2077 .open = nonseekable_open, 2078 .unlocked_ioctl = loop_control_ioctl, 2079 .compat_ioctl = loop_control_ioctl, 2080 .owner = THIS_MODULE, 2081 .llseek = noop_llseek, 2082 }; 2083 2084 static struct miscdevice loop_misc = { 2085 .minor = LOOP_CTRL_MINOR, 2086 .name = "loop-control", 2087 .fops = &loop_ctl_fops, 2088 }; 2089 2090 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2091 MODULE_ALIAS("devname:loop-control"); 2092 2093 static int __init loop_init(void) 2094 { 2095 int i, nr; 2096 unsigned long range; 2097 struct loop_device *lo; 2098 int err; 2099 2100 part_shift = 0; 2101 if (max_part > 0) { 2102 part_shift = fls(max_part); 2103 2104 /* 2105 * Adjust max_part according to part_shift as it is exported 2106 * to user space so that user can decide correct minor number 2107 * if [s]he want to create more devices. 2108 * 2109 * Note that -1 is required because partition 0 is reserved 2110 * for the whole disk. 2111 */ 2112 max_part = (1UL << part_shift) - 1; 2113 } 2114 2115 if ((1UL << part_shift) > DISK_MAX_PARTS) { 2116 err = -EINVAL; 2117 goto err_out; 2118 } 2119 2120 if (max_loop > 1UL << (MINORBITS - part_shift)) { 2121 err = -EINVAL; 2122 goto err_out; 2123 } 2124 2125 /* 2126 * If max_loop is specified, create that many devices upfront. 2127 * This also becomes a hard limit. If max_loop is not specified, 2128 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2129 * init time. Loop devices can be requested on-demand with the 2130 * /dev/loop-control interface, or be instantiated by accessing 2131 * a 'dead' device node. 2132 */ 2133 if (max_loop) { 2134 nr = max_loop; 2135 range = max_loop << part_shift; 2136 } else { 2137 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2138 range = 1UL << MINORBITS; 2139 } 2140 2141 err = misc_register(&loop_misc); 2142 if (err < 0) 2143 goto err_out; 2144 2145 2146 if (register_blkdev(LOOP_MAJOR, "loop")) { 2147 err = -EIO; 2148 goto misc_out; 2149 } 2150 2151 blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2152 THIS_MODULE, loop_probe, NULL, NULL); 2153 2154 /* pre-create number of devices given by config or max_loop */ 2155 mutex_lock(&loop_ctl_mutex); 2156 for (i = 0; i < nr; i++) 2157 loop_add(&lo, i); 2158 mutex_unlock(&loop_ctl_mutex); 2159 2160 printk(KERN_INFO "loop: module loaded\n"); 2161 return 0; 2162 2163 misc_out: 2164 misc_deregister(&loop_misc); 2165 err_out: 2166 return err; 2167 } 2168 2169 static int loop_exit_cb(int id, void *ptr, void *data) 2170 { 2171 struct loop_device *lo = ptr; 2172 2173 loop_remove(lo); 2174 return 0; 2175 } 2176 2177 static void __exit loop_exit(void) 2178 { 2179 unsigned long range; 2180 2181 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2182 2183 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 2184 idr_destroy(&loop_index_idr); 2185 2186 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 2187 unregister_blkdev(LOOP_MAJOR, "loop"); 2188 2189 misc_deregister(&loop_misc); 2190 } 2191 2192 module_init(loop_init); 2193 module_exit(loop_exit); 2194 2195 #ifndef MODULE 2196 static int __init max_loop_setup(char *str) 2197 { 2198 max_loop = simple_strtol(str, NULL, 0); 2199 return 1; 2200 } 2201 2202 __setup("max_loop=", max_loop_setup); 2203 #endif 2204