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