?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
spi: sprd: adi: Add a reset reason for watchdog mode
[linux/fpc-iii.git] / drivers / block / loop.c
blob44c9985f352abd0cfb4ddda003302e09f76ce4bd
1 /*
2 * linux/drivers/block/loop.c
3 *
4 * Written by Theodore Ts'o, 3/29/93
5 *
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
8 *
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
11 *
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
14 *
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
16 *
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
18 *
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
20 *
21 * Loadable modules and other fixes by AK, 1998
22 *
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
26 *
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
29 *
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
33 *
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
36 * Al Viro too.
37 * Jens Axboe <axboe@suse.de>, Nov 2000
38 *
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
41 *
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
45 *
46 * Still To Fix:
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49 *
50 */
51
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
55 #include <linux/fs.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
79 #include <linux/ioprio.h>
80 #include <linux/blk-cgroup.h>
81
82 #include "loop.h"
83
84 #include <linux/uaccess.h>
85
86 static DEFINE_IDR(loop_index_idr);
87 static DEFINE_MUTEX(loop_ctl_mutex);
88
89 static int max_part;
90 static int part_shift;
91
92 static int transfer_xor(struct loop_device *lo, int cmd,
93 struct page *raw_page, unsigned raw_off,
94 struct page *loop_page, unsigned loop_off,
95 int size, sector_t real_block)
96 {
97 char *raw_buf = kmap_atomic(raw_page) + raw_off;
98 char *loop_buf = kmap_atomic(loop_page) + loop_off;
99 char *in, *out, *key;
100 int i, keysize;
101
102 if (cmd == READ) {
103 in = raw_buf;
104 out = loop_buf;
105 } else {
106 in = loop_buf;
107 out = raw_buf;
108 }
109
110 key = lo->lo_encrypt_key;
111 keysize = lo->lo_encrypt_key_size;
112 for (i = 0; i < size; i++)
113 *out++ = *in++ ^ key[(i & 511) % keysize];
114
115 kunmap_atomic(loop_buf);
116 kunmap_atomic(raw_buf);
117 cond_resched();
118 return 0;
119 }
120
121 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
122 {
123 if (unlikely(info->lo_encrypt_key_size <= 0))
124 return -EINVAL;
125 return 0;
126 }
127
128 static struct loop_func_table none_funcs = {
129 .number = LO_CRYPT_NONE,
130 };
131
132 static struct loop_func_table xor_funcs = {
133 .number = LO_CRYPT_XOR,
134 .transfer = transfer_xor,
135 .init = xor_init
136 };
137
138 /* xfer_funcs[0] is special - its release function is never called */
139 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
140 &none_funcs,
141 &xor_funcs
142 };
143
144 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
145 {
146 loff_t loopsize;
147
148 /* Compute loopsize in bytes */
149 loopsize = i_size_read(file->f_mapping->host);
150 if (offset > 0)
151 loopsize -= offset;
152 /* offset is beyond i_size, weird but possible */
153 if (loopsize < 0)
154 return 0;
155
156 if (sizelimit > 0 && sizelimit < loopsize)
157 loopsize = sizelimit;
158 /*
159 * Unfortunately, if we want to do I/O on the device,
160 * the number of 512-byte sectors has to fit into a sector_t.
161 */
162 return loopsize >> 9;
163 }
164
165 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
166 {
167 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
168 }
169
170 static void __loop_update_dio(struct loop_device *lo, bool dio)
171 {
172 struct file *file = lo->lo_backing_file;
173 struct address_space *mapping = file->f_mapping;
174 struct inode *inode = mapping->host;
175 unsigned short sb_bsize = 0;
176 unsigned dio_align = 0;
177 bool use_dio;
178
179 if (inode->i_sb->s_bdev) {
180 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
181 dio_align = sb_bsize - 1;
182 }
183
184 /*
185 * We support direct I/O only if lo_offset is aligned with the
186 * logical I/O size of backing device, and the logical block
187 * size of loop is bigger than the backing device's and the loop
188 * needn't transform transfer.
189 *
190 * TODO: the above condition may be loosed in the future, and
191 * direct I/O may be switched runtime at that time because most
192 * of requests in sane applications should be PAGE_SIZE aligned
193 */
194 if (dio) {
195 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
196 !(lo->lo_offset & dio_align) &&
197 mapping->a_ops->direct_IO &&
198 !lo->transfer)
199 use_dio = true;
200 else
201 use_dio = false;
202 } else {
203 use_dio = false;
204 }
205
206 if (lo->use_dio == use_dio)
207 return;
208
209 /* flush dirty pages before changing direct IO */
210 vfs_fsync(file, 0);
211
212 /*
213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 * will get updated by ioctl(LOOP_GET_STATUS)
216 */
217 blk_mq_freeze_queue(lo->lo_queue);
218 lo->use_dio = use_dio;
219 if (use_dio) {
220 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
221 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
222 } else {
223 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
224 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
225 }
226 blk_mq_unfreeze_queue(lo->lo_queue);
227 }
228
229 static int
230 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
231 {
232 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
233 sector_t x = (sector_t)size;
234 struct block_device *bdev = lo->lo_device;
235
236 if (unlikely((loff_t)x != size))
237 return -EFBIG;
238 if (lo->lo_offset != offset)
239 lo->lo_offset = offset;
240 if (lo->lo_sizelimit != sizelimit)
241 lo->lo_sizelimit = sizelimit;
242 set_capacity(lo->lo_disk, x);
243 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
244 /* let user-space know about the new size */
245 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
246 return 0;
247 }
248
249 static inline int
250 lo_do_transfer(struct loop_device *lo, int cmd,
251 struct page *rpage, unsigned roffs,
252 struct page *lpage, unsigned loffs,
253 int size, sector_t rblock)
254 {
255 int ret;
256
257 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
258 if (likely(!ret))
259 return 0;
260
261 printk_ratelimited(KERN_ERR
262 "loop: Transfer error at byte offset %llu, length %i.\n",
263 (unsigned long long)rblock << 9, size);
264 return ret;
265 }
266
267 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
268 {
269 struct iov_iter i;
270 ssize_t bw;
271
272 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
273
274 file_start_write(file);
275 bw = vfs_iter_write(file, &i, ppos, 0);
276 file_end_write(file);
277
278 if (likely(bw == bvec->bv_len))
279 return 0;
280
281 printk_ratelimited(KERN_ERR
282 "loop: Write error at byte offset %llu, length %i.\n",
283 (unsigned long long)*ppos, bvec->bv_len);
284 if (bw >= 0)
285 bw = -EIO;
286 return bw;
287 }
288
289 static int lo_write_simple(struct loop_device *lo, struct request *rq,
290 loff_t pos)
291 {
292 struct bio_vec bvec;
293 struct req_iterator iter;
294 int ret = 0;
295
296 rq_for_each_segment(bvec, rq, iter) {
297 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
298 if (ret < 0)
299 break;
300 cond_resched();
301 }
302
303 return ret;
304 }
305
306 /*
307 * This is the slow, transforming version that needs to double buffer the
308 * data as it cannot do the transformations in place without having direct
309 * access to the destination pages of the backing file.
310 */
311 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
312 loff_t pos)
313 {
314 struct bio_vec bvec, b;
315 struct req_iterator iter;
316 struct page *page;
317 int ret = 0;
318
319 page = alloc_page(GFP_NOIO);
320 if (unlikely(!page))
321 return -ENOMEM;
322
323 rq_for_each_segment(bvec, rq, iter) {
324 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
325 bvec.bv_offset, bvec.bv_len, pos >> 9);
326 if (unlikely(ret))
327 break;
328
329 b.bv_page = page;
330 b.bv_offset = 0;
331 b.bv_len = bvec.bv_len;
332 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
333 if (ret < 0)
334 break;
335 }
336
337 __free_page(page);
338 return ret;
339 }
340
341 static int lo_read_simple(struct loop_device *lo, struct request *rq,
342 loff_t pos)
343 {
344 struct bio_vec bvec;
345 struct req_iterator iter;
346 struct iov_iter i;
347 ssize_t len;
348
349 rq_for_each_segment(bvec, rq, iter) {
350 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
351 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
352 if (len < 0)
353 return len;
354
355 flush_dcache_page(bvec.bv_page);
356
357 if (len != bvec.bv_len) {
358 struct bio *bio;
359
360 __rq_for_each_bio(bio, rq)
361 zero_fill_bio(bio);
362 break;
363 }
364 cond_resched();
365 }
366
367 return 0;
368 }
369
370 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
371 loff_t pos)
372 {
373 struct bio_vec bvec, b;
374 struct req_iterator iter;
375 struct iov_iter i;
376 struct page *page;
377 ssize_t len;
378 int ret = 0;
379
380 page = alloc_page(GFP_NOIO);
381 if (unlikely(!page))
382 return -ENOMEM;
383
384 rq_for_each_segment(bvec, rq, iter) {
385 loff_t offset = pos;
386
387 b.bv_page = page;
388 b.bv_offset = 0;
389 b.bv_len = bvec.bv_len;
390
391 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
392 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
393 if (len < 0) {
394 ret = len;
395 goto out_free_page;
396 }
397
398 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
399 bvec.bv_offset, len, offset >> 9);
400 if (ret)
401 goto out_free_page;
402
403 flush_dcache_page(bvec.bv_page);
404
405 if (len != bvec.bv_len) {
406 struct bio *bio;
407
408 __rq_for_each_bio(bio, rq)
409 zero_fill_bio(bio);
410 break;
411 }
412 }
413
414 ret = 0;
415 out_free_page:
416 __free_page(page);
417 return ret;
418 }
419
420 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
421 {
422 /*
423 * We use punch hole to reclaim the free space used by the
424 * image a.k.a. discard. However we do not support discard if
425 * encryption is enabled, because it may give an attacker
426 * useful information.
427 */
428 struct file *file = lo->lo_backing_file;
429 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
430 int ret;
431
432 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
433 ret = -EOPNOTSUPP;
434 goto out;
435 }
436
437 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
438 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
439 ret = -EIO;
440 out:
441 return ret;
442 }
443
444 static int lo_req_flush(struct loop_device *lo, struct request *rq)
445 {
446 struct file *file = lo->lo_backing_file;
447 int ret = vfs_fsync(file, 0);
448 if (unlikely(ret && ret != -EINVAL))
449 ret = -EIO;
450
451 return ret;
452 }
453
454 static void lo_complete_rq(struct request *rq)
455 {
456 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
457 blk_status_t ret = BLK_STS_OK;
458
459 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
460 req_op(rq) != REQ_OP_READ) {
461 if (cmd->ret < 0)
462 ret = BLK_STS_IOERR;
463 goto end_io;
464 }
465
466 /*
467 * Short READ - if we got some data, advance our request and
468 * retry it. If we got no data, end the rest with EIO.
469 */
470 if (cmd->ret) {
471 blk_update_request(rq, BLK_STS_OK, cmd->ret);
472 cmd->ret = 0;
473 blk_mq_requeue_request(rq, true);
474 } else {
475 if (cmd->use_aio) {
476 struct bio *bio = rq->bio;
477
478 while (bio) {
479 zero_fill_bio(bio);
480 bio = bio->bi_next;
481 }
482 }
483 ret = BLK_STS_IOERR;
484 end_io:
485 blk_mq_end_request(rq, ret);
486 }
487 }
488
489 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
490 {
491 struct request *rq = blk_mq_rq_from_pdu(cmd);
492
493 if (!atomic_dec_and_test(&cmd->ref))
494 return;
495 kfree(cmd->bvec);
496 cmd->bvec = NULL;
497 blk_mq_complete_request(rq);
498 }
499
500 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
501 {
502 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
503
504 if (cmd->css)
505 css_put(cmd->css);
506 cmd->ret = ret;
507 lo_rw_aio_do_completion(cmd);
508 }
509
510 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
511 loff_t pos, bool rw)
512 {
513 struct iov_iter iter;
514 struct req_iterator rq_iter;
515 struct bio_vec *bvec;
516 struct request *rq = blk_mq_rq_from_pdu(cmd);
517 struct bio *bio = rq->bio;
518 struct file *file = lo->lo_backing_file;
519 struct bio_vec tmp;
520 unsigned int offset;
521 int nr_bvec = 0;
522 int ret;
523
524 rq_for_each_bvec(tmp, rq, rq_iter)
525 nr_bvec++;
526
527 if (rq->bio != rq->biotail) {
528
529 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
530 GFP_NOIO);
531 if (!bvec)
532 return -EIO;
533 cmd->bvec = bvec;
534
535 /*
536 * The bios of the request may be started from the middle of
537 * the 'bvec' because of bio splitting, so we can't directly
538 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
539 * API will take care of all details for us.
540 */
541 rq_for_each_bvec(tmp, rq, rq_iter) {
542 *bvec = tmp;
543 bvec++;
544 }
545 bvec = cmd->bvec;
546 offset = 0;
547 } else {
548 /*
549 * Same here, this bio may be started from the middle of the
550 * 'bvec' because of bio splitting, so offset from the bvec
551 * must be passed to iov iterator
552 */
553 offset = bio->bi_iter.bi_bvec_done;
554 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
555 }
556 atomic_set(&cmd->ref, 2);
557
558 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
559 iter.iov_offset = offset;
560
561 cmd->iocb.ki_pos = pos;
562 cmd->iocb.ki_filp = file;
563 cmd->iocb.ki_complete = lo_rw_aio_complete;
564 cmd->iocb.ki_flags = IOCB_DIRECT;
565 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
566 if (cmd->css)
567 kthread_associate_blkcg(cmd->css);
568
569 if (rw == WRITE)
570 ret = call_write_iter(file, &cmd->iocb, &iter);
571 else
572 ret = call_read_iter(file, &cmd->iocb, &iter);
573
574 lo_rw_aio_do_completion(cmd);
575 kthread_associate_blkcg(NULL);
576
577 if (ret != -EIOCBQUEUED)
578 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
579 return 0;
580 }
581
582 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
583 {
584 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
585 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
586
587 /*
588 * lo_write_simple and lo_read_simple should have been covered
589 * by io submit style function like lo_rw_aio(), one blocker
590 * is that lo_read_simple() need to call flush_dcache_page after
591 * the page is written from kernel, and it isn't easy to handle
592 * this in io submit style function which submits all segments
593 * of the req at one time. And direct read IO doesn't need to
594 * run flush_dcache_page().
595 */
596 switch (req_op(rq)) {
597 case REQ_OP_FLUSH:
598 return lo_req_flush(lo, rq);
599 case REQ_OP_DISCARD:
600 case REQ_OP_WRITE_ZEROES:
601 return lo_discard(lo, rq, pos);
602 case REQ_OP_WRITE:
603 if (lo->transfer)
604 return lo_write_transfer(lo, rq, pos);
605 else if (cmd->use_aio)
606 return lo_rw_aio(lo, cmd, pos, WRITE);
607 else
608 return lo_write_simple(lo, rq, pos);
609 case REQ_OP_READ:
610 if (lo->transfer)
611 return lo_read_transfer(lo, rq, pos);
612 else if (cmd->use_aio)
613 return lo_rw_aio(lo, cmd, pos, READ);
614 else
615 return lo_read_simple(lo, rq, pos);
616 default:
617 WARN_ON_ONCE(1);
618 return -EIO;
619 }
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 rc = blkdev_reread_part(bdev);
634 if (rc)
635 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
636 __func__, lo->lo_number, lo->lo_file_name, rc);
637 }
638
639 static inline int is_loop_device(struct file *file)
640 {
641 struct inode *i = file->f_mapping->host;
642
643 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
644 }
645
646 static int loop_validate_file(struct file *file, struct block_device *bdev)
647 {
648 struct inode *inode = file->f_mapping->host;
649 struct file *f = file;
650
651 /* Avoid recursion */
652 while (is_loop_device(f)) {
653 struct loop_device *l;
654
655 if (f->f_mapping->host->i_bdev == bdev)
656 return -EBADF;
657
658 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
659 if (l->lo_state != Lo_bound) {
660 return -EINVAL;
661 }
662 f = l->lo_backing_file;
663 }
664 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
665 return -EINVAL;
666 return 0;
667 }
668
669 /*
670 * loop_change_fd switched the backing store of a loopback device to
671 * a new file. This is useful for operating system installers to free up
672 * the original file and in High Availability environments to switch to
673 * an alternative location for the content in case of server meltdown.
674 * This can only work if the loop device is used read-only, and if the
675 * new backing store is the same size and type as the old backing store.
676 */
677 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
678 unsigned int arg)
679 {
680 struct file *file = NULL, *old_file;
681 int error;
682 bool partscan;
683
684 error = mutex_lock_killable(&loop_ctl_mutex);
685 if (error)
686 return error;
687 error = -ENXIO;
688 if (lo->lo_state != Lo_bound)
689 goto out_err;
690
691 /* the loop device has to be read-only */
692 error = -EINVAL;
693 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
694 goto out_err;
695
696 error = -EBADF;
697 file = fget(arg);
698 if (!file)
699 goto out_err;
700
701 error = loop_validate_file(file, bdev);
702 if (error)
703 goto out_err;
704
705 old_file = lo->lo_backing_file;
706
707 error = -EINVAL;
708
709 /* size of the new backing store needs to be the same */
710 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
711 goto out_err;
712
713 /* and ... switch */
714 blk_mq_freeze_queue(lo->lo_queue);
715 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
716 lo->lo_backing_file = file;
717 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
718 mapping_set_gfp_mask(file->f_mapping,
719 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
720 loop_update_dio(lo);
721 blk_mq_unfreeze_queue(lo->lo_queue);
722 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
723 mutex_unlock(&loop_ctl_mutex);
724 /*
725 * We must drop file reference outside of loop_ctl_mutex as dropping
726 * the file ref can take bd_mutex which creates circular locking
727 * dependency.
728 */
729 fput(old_file);
730 if (partscan)
731 loop_reread_partitions(lo, bdev);
732 return 0;
733
734 out_err:
735 mutex_unlock(&loop_ctl_mutex);
736 if (file)
737 fput(file);
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 void loop_update_rotational(struct loop_device *lo)
904 {
905 struct file *file = lo->lo_backing_file;
906 struct inode *file_inode = file->f_mapping->host;
907 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
908 struct request_queue *q = lo->lo_queue;
909 bool nonrot = true;
910
911 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
912 if (file_bdev)
913 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
914
915 if (nonrot)
916 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
917 else
918 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
919 }
920
921 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
922 struct block_device *bdev, unsigned int arg)
923 {
924 struct file *file;
925 struct inode *inode;
926 struct address_space *mapping;
927 int lo_flags = 0;
928 int error;
929 loff_t size;
930 bool partscan;
931
932 /* This is safe, since we have a reference from open(). */
933 __module_get(THIS_MODULE);
934
935 error = -EBADF;
936 file = fget(arg);
937 if (!file)
938 goto out;
939
940 /*
941 * If we don't hold exclusive handle for the device, upgrade to it
942 * here to avoid changing device under exclusive owner.
943 */
944 if (!(mode & FMODE_EXCL)) {
945 bdgrab(bdev);
946 error = blkdev_get(bdev, mode | FMODE_EXCL, loop_set_fd);
947 if (error)
948 goto out_putf;
949 }
950
951 error = mutex_lock_killable(&loop_ctl_mutex);
952 if (error)
953 goto out_bdev;
954
955 error = -EBUSY;
956 if (lo->lo_state != Lo_unbound)
957 goto out_unlock;
958
959 error = loop_validate_file(file, bdev);
960 if (error)
961 goto out_unlock;
962
963 mapping = file->f_mapping;
964 inode = mapping->host;
965
966 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
967 !file->f_op->write_iter)
968 lo_flags |= LO_FLAGS_READ_ONLY;
969
970 error = -EFBIG;
971 size = get_loop_size(lo, file);
972 if ((loff_t)(sector_t)size != size)
973 goto out_unlock;
974 error = loop_prepare_queue(lo);
975 if (error)
976 goto out_unlock;
977
978 error = 0;
979
980 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
981
982 lo->use_dio = false;
983 lo->lo_device = bdev;
984 lo->lo_flags = lo_flags;
985 lo->lo_backing_file = file;
986 lo->transfer = NULL;
987 lo->ioctl = NULL;
988 lo->lo_sizelimit = 0;
989 lo->old_gfp_mask = mapping_gfp_mask(mapping);
990 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
991
992 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
993 blk_queue_write_cache(lo->lo_queue, true, false);
994
995 loop_update_rotational(lo);
996 loop_update_dio(lo);
997 set_capacity(lo->lo_disk, size);
998 bd_set_size(bdev, size << 9);
999 loop_sysfs_init(lo);
1000 /* let user-space know about the new size */
1001 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1002
1003 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1004 block_size(inode->i_bdev) : PAGE_SIZE);
1005
1006 lo->lo_state = Lo_bound;
1007 if (part_shift)
1008 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1009 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1010
1011 /* Grab the block_device to prevent its destruction after we
1012 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1013 */
1014 bdgrab(bdev);
1015 mutex_unlock(&loop_ctl_mutex);
1016 if (partscan)
1017 loop_reread_partitions(lo, bdev);
1018 if (!(mode & FMODE_EXCL))
1019 blkdev_put(bdev, mode | FMODE_EXCL);
1020 return 0;
1021
1022 out_unlock:
1023 mutex_unlock(&loop_ctl_mutex);
1024 out_bdev:
1025 if (!(mode & FMODE_EXCL))
1026 blkdev_put(bdev, mode | FMODE_EXCL);
1027 out_putf:
1028 fput(file);
1029 out:
1030 /* This is safe: open() is still holding a reference. */
1031 module_put(THIS_MODULE);
1032 return error;
1033 }
1034
1035 static int
1036 loop_release_xfer(struct loop_device *lo)
1037 {
1038 int err = 0;
1039 struct loop_func_table *xfer = lo->lo_encryption;
1040
1041 if (xfer) {
1042 if (xfer->release)
1043 err = xfer->release(lo);
1044 lo->transfer = NULL;
1045 lo->lo_encryption = NULL;
1046 module_put(xfer->owner);
1047 }
1048 return err;
1049 }
1050
1051 static int
1052 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1053 const struct loop_info64 *i)
1054 {
1055 int err = 0;
1056
1057 if (xfer) {
1058 struct module *owner = xfer->owner;
1059
1060 if (!try_module_get(owner))
1061 return -EINVAL;
1062 if (xfer->init)
1063 err = xfer->init(lo, i);
1064 if (err)
1065 module_put(owner);
1066 else
1067 lo->lo_encryption = xfer;
1068 }
1069 return err;
1070 }
1071
1072 static int __loop_clr_fd(struct loop_device *lo, bool release)
1073 {
1074 struct file *filp = NULL;
1075 gfp_t gfp = lo->old_gfp_mask;
1076 struct block_device *bdev = lo->lo_device;
1077 int err = 0;
1078 bool partscan = false;
1079 int lo_number;
1080
1081 mutex_lock(&loop_ctl_mutex);
1082 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1083 err = -ENXIO;
1084 goto out_unlock;
1085 }
1086
1087 filp = lo->lo_backing_file;
1088 if (filp == NULL) {
1089 err = -EINVAL;
1090 goto out_unlock;
1091 }
1092
1093 /* freeze request queue during the transition */
1094 blk_mq_freeze_queue(lo->lo_queue);
1095
1096 spin_lock_irq(&lo->lo_lock);
1097 lo->lo_backing_file = NULL;
1098 spin_unlock_irq(&lo->lo_lock);
1099
1100 loop_release_xfer(lo);
1101 lo->transfer = NULL;
1102 lo->ioctl = NULL;
1103 lo->lo_device = NULL;
1104 lo->lo_encryption = NULL;
1105 lo->lo_offset = 0;
1106 lo->lo_sizelimit = 0;
1107 lo->lo_encrypt_key_size = 0;
1108 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1109 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1110 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1111 blk_queue_logical_block_size(lo->lo_queue, 512);
1112 blk_queue_physical_block_size(lo->lo_queue, 512);
1113 blk_queue_io_min(lo->lo_queue, 512);
1114 if (bdev) {
1115 bdput(bdev);
1116 invalidate_bdev(bdev);
1117 bdev->bd_inode->i_mapping->wb_err = 0;
1118 }
1119 set_capacity(lo->lo_disk, 0);
1120 loop_sysfs_exit(lo);
1121 if (bdev) {
1122 bd_set_size(bdev, 0);
1123 /* let user-space know about this change */
1124 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1125 }
1126 mapping_set_gfp_mask(filp->f_mapping, gfp);
1127 /* This is safe: open() is still holding a reference. */
1128 module_put(THIS_MODULE);
1129 blk_mq_unfreeze_queue(lo->lo_queue);
1130
1131 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1132 lo_number = lo->lo_number;
1133 loop_unprepare_queue(lo);
1134 out_unlock:
1135 mutex_unlock(&loop_ctl_mutex);
1136 if (partscan) {
1137 /*
1138 * bd_mutex has been held already in release path, so don't
1139 * acquire it if this function is called in such case.
1140 *
1141 * If the reread partition isn't from release path, lo_refcnt
1142 * must be at least one and it can only become zero when the
1143 * current holder is released.
1144 */
1145 if (release)
1146 err = __blkdev_reread_part(bdev);
1147 else
1148 err = blkdev_reread_part(bdev);
1149 if (err)
1150 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1151 __func__, lo_number, err);
1152 /* Device is gone, no point in returning error */
1153 err = 0;
1154 }
1155
1156 /*
1157 * lo->lo_state is set to Lo_unbound here after above partscan has
1158 * finished.
1159 *
1160 * There cannot be anybody else entering __loop_clr_fd() as
1161 * lo->lo_backing_file is already cleared and Lo_rundown state
1162 * protects us from all the other places trying to change the 'lo'
1163 * device.
1164 */
1165 mutex_lock(&loop_ctl_mutex);
1166 lo->lo_flags = 0;
1167 if (!part_shift)
1168 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1169 lo->lo_state = Lo_unbound;
1170 mutex_unlock(&loop_ctl_mutex);
1171
1172 /*
1173 * Need not hold loop_ctl_mutex to fput backing file.
1174 * Calling fput holding loop_ctl_mutex triggers a circular
1175 * lock dependency possibility warning as fput can take
1176 * bd_mutex which is usually taken before loop_ctl_mutex.
1177 */
1178 if (filp)
1179 fput(filp);
1180 return err;
1181 }
1182
1183 static int loop_clr_fd(struct loop_device *lo)
1184 {
1185 int err;
1186
1187 err = mutex_lock_killable(&loop_ctl_mutex);
1188 if (err)
1189 return err;
1190 if (lo->lo_state != Lo_bound) {
1191 mutex_unlock(&loop_ctl_mutex);
1192 return -ENXIO;
1193 }
1194 /*
1195 * If we've explicitly asked to tear down the loop device,
1196 * and it has an elevated reference count, set it for auto-teardown when
1197 * the last reference goes away. This stops $!~#$@ udev from
1198 * preventing teardown because it decided that it needs to run blkid on
1199 * the loopback device whenever they appear. xfstests is notorious for
1200 * failing tests because blkid via udev races with a losetup
1201 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1202 * command to fail with EBUSY.
1203 */
1204 if (atomic_read(&lo->lo_refcnt) > 1) {
1205 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1206 mutex_unlock(&loop_ctl_mutex);
1207 return 0;
1208 }
1209 lo->lo_state = Lo_rundown;
1210 mutex_unlock(&loop_ctl_mutex);
1211
1212 return __loop_clr_fd(lo, false);
1213 }
1214
1215 static int
1216 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1217 {
1218 int err;
1219 struct loop_func_table *xfer;
1220 kuid_t uid = current_uid();
1221 struct block_device *bdev;
1222 bool partscan = false;
1223
1224 err = mutex_lock_killable(&loop_ctl_mutex);
1225 if (err)
1226 return err;
1227 if (lo->lo_encrypt_key_size &&
1228 !uid_eq(lo->lo_key_owner, uid) &&
1229 !capable(CAP_SYS_ADMIN)) {
1230 err = -EPERM;
1231 goto out_unlock;
1232 }
1233 if (lo->lo_state != Lo_bound) {
1234 err = -ENXIO;
1235 goto out_unlock;
1236 }
1237 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1238 err = -EINVAL;
1239 goto out_unlock;
1240 }
1241
1242 if (lo->lo_offset != info->lo_offset ||
1243 lo->lo_sizelimit != info->lo_sizelimit) {
1244 sync_blockdev(lo->lo_device);
1245 kill_bdev(lo->lo_device);
1246 }
1247
1248 /* I/O need to be drained during transfer transition */
1249 blk_mq_freeze_queue(lo->lo_queue);
1250
1251 err = loop_release_xfer(lo);
1252 if (err)
1253 goto out_unfreeze;
1254
1255 if (info->lo_encrypt_type) {
1256 unsigned int type = info->lo_encrypt_type;
1257
1258 if (type >= MAX_LO_CRYPT) {
1259 err = -EINVAL;
1260 goto out_unfreeze;
1261 }
1262 xfer = xfer_funcs[type];
1263 if (xfer == NULL) {
1264 err = -EINVAL;
1265 goto out_unfreeze;
1266 }
1267 } else
1268 xfer = NULL;
1269
1270 err = loop_init_xfer(lo, xfer, info);
1271 if (err)
1272 goto out_unfreeze;
1273
1274 if (lo->lo_offset != info->lo_offset ||
1275 lo->lo_sizelimit != info->lo_sizelimit) {
1276 /* kill_bdev should have truncated all the pages */
1277 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1278 err = -EAGAIN;
1279 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1280 __func__, lo->lo_number, lo->lo_file_name,
1281 lo->lo_device->bd_inode->i_mapping->nrpages);
1282 goto out_unfreeze;
1283 }
1284 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1285 err = -EFBIG;
1286 goto out_unfreeze;
1287 }
1288 }
1289
1290 loop_config_discard(lo);
1291
1292 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1293 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1294 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1295 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1296
1297 if (!xfer)
1298 xfer = &none_funcs;
1299 lo->transfer = xfer->transfer;
1300 lo->ioctl = xfer->ioctl;
1301
1302 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1303 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1304 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1305
1306 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1307 lo->lo_init[0] = info->lo_init[0];
1308 lo->lo_init[1] = info->lo_init[1];
1309 if (info->lo_encrypt_key_size) {
1310 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1311 info->lo_encrypt_key_size);
1312 lo->lo_key_owner = uid;
1313 }
1314
1315 /* update dio if lo_offset or transfer is changed */
1316 __loop_update_dio(lo, lo->use_dio);
1317
1318 out_unfreeze:
1319 blk_mq_unfreeze_queue(lo->lo_queue);
1320
1321 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1322 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1323 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1324 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1325 bdev = lo->lo_device;
1326 partscan = true;
1327 }
1328 out_unlock:
1329 mutex_unlock(&loop_ctl_mutex);
1330 if (partscan)
1331 loop_reread_partitions(lo, bdev);
1332
1333 return err;
1334 }
1335
1336 static int
1337 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1338 {
1339 struct path path;
1340 struct kstat stat;
1341 int ret;
1342
1343 ret = mutex_lock_killable(&loop_ctl_mutex);
1344 if (ret)
1345 return ret;
1346 if (lo->lo_state != Lo_bound) {
1347 mutex_unlock(&loop_ctl_mutex);
1348 return -ENXIO;
1349 }
1350
1351 memset(info, 0, sizeof(*info));
1352 info->lo_number = lo->lo_number;
1353 info->lo_offset = lo->lo_offset;
1354 info->lo_sizelimit = lo->lo_sizelimit;
1355 info->lo_flags = lo->lo_flags;
1356 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1357 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1358 info->lo_encrypt_type =
1359 lo->lo_encryption ? lo->lo_encryption->number : 0;
1360 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1361 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1362 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1363 lo->lo_encrypt_key_size);
1364 }
1365
1366 /* Drop loop_ctl_mutex while we call into the filesystem. */
1367 path = lo->lo_backing_file->f_path;
1368 path_get(&path);
1369 mutex_unlock(&loop_ctl_mutex);
1370 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1371 if (!ret) {
1372 info->lo_device = huge_encode_dev(stat.dev);
1373 info->lo_inode = stat.ino;
1374 info->lo_rdevice = huge_encode_dev(stat.rdev);
1375 }
1376 path_put(&path);
1377 return ret;
1378 }
1379
1380 static void
1381 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1382 {
1383 memset(info64, 0, sizeof(*info64));
1384 info64->lo_number = info->lo_number;
1385 info64->lo_device = info->lo_device;
1386 info64->lo_inode = info->lo_inode;
1387 info64->lo_rdevice = info->lo_rdevice;
1388 info64->lo_offset = info->lo_offset;
1389 info64->lo_sizelimit = 0;
1390 info64->lo_encrypt_type = info->lo_encrypt_type;
1391 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1392 info64->lo_flags = info->lo_flags;
1393 info64->lo_init[0] = info->lo_init[0];
1394 info64->lo_init[1] = info->lo_init[1];
1395 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1396 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1397 else
1398 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1399 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1400 }
1401
1402 static int
1403 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1404 {
1405 memset(info, 0, sizeof(*info));
1406 info->lo_number = info64->lo_number;
1407 info->lo_device = info64->lo_device;
1408 info->lo_inode = info64->lo_inode;
1409 info->lo_rdevice = info64->lo_rdevice;
1410 info->lo_offset = info64->lo_offset;
1411 info->lo_encrypt_type = info64->lo_encrypt_type;
1412 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1413 info->lo_flags = info64->lo_flags;
1414 info->lo_init[0] = info64->lo_init[0];
1415 info->lo_init[1] = info64->lo_init[1];
1416 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1417 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1418 else
1419 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1420 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1421
1422 /* error in case values were truncated */
1423 if (info->lo_device != info64->lo_device ||
1424 info->lo_rdevice != info64->lo_rdevice ||
1425 info->lo_inode != info64->lo_inode ||
1426 info->lo_offset != info64->lo_offset)
1427 return -EOVERFLOW;
1428
1429 return 0;
1430 }
1431
1432 static int
1433 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1434 {
1435 struct loop_info info;
1436 struct loop_info64 info64;
1437
1438 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1439 return -EFAULT;
1440 loop_info64_from_old(&info, &info64);
1441 return loop_set_status(lo, &info64);
1442 }
1443
1444 static int
1445 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1446 {
1447 struct loop_info64 info64;
1448
1449 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1450 return -EFAULT;
1451 return loop_set_status(lo, &info64);
1452 }
1453
1454 static int
1455 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1456 struct loop_info info;
1457 struct loop_info64 info64;
1458 int err;
1459
1460 if (!arg)
1461 return -EINVAL;
1462 err = loop_get_status(lo, &info64);
1463 if (!err)
1464 err = loop_info64_to_old(&info64, &info);
1465 if (!err && copy_to_user(arg, &info, sizeof(info)))
1466 err = -EFAULT;
1467
1468 return err;
1469 }
1470
1471 static int
1472 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1473 struct loop_info64 info64;
1474 int err;
1475
1476 if (!arg)
1477 return -EINVAL;
1478 err = loop_get_status(lo, &info64);
1479 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1480 err = -EFAULT;
1481
1482 return err;
1483 }
1484
1485 static int loop_set_capacity(struct loop_device *lo)
1486 {
1487 if (unlikely(lo->lo_state != Lo_bound))
1488 return -ENXIO;
1489
1490 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1491 }
1492
1493 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1494 {
1495 int error = -ENXIO;
1496 if (lo->lo_state != Lo_bound)
1497 goto out;
1498
1499 __loop_update_dio(lo, !!arg);
1500 if (lo->use_dio == !!arg)
1501 return 0;
1502 error = -EINVAL;
1503 out:
1504 return error;
1505 }
1506
1507 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1508 {
1509 int err = 0;
1510
1511 if (lo->lo_state != Lo_bound)
1512 return -ENXIO;
1513
1514 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1515 return -EINVAL;
1516
1517 if (lo->lo_queue->limits.logical_block_size != arg) {
1518 sync_blockdev(lo->lo_device);
1519 kill_bdev(lo->lo_device);
1520 }
1521
1522 blk_mq_freeze_queue(lo->lo_queue);
1523
1524 /* kill_bdev should have truncated all the pages */
1525 if (lo->lo_queue->limits.logical_block_size != arg &&
1526 lo->lo_device->bd_inode->i_mapping->nrpages) {
1527 err = -EAGAIN;
1528 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1529 __func__, lo->lo_number, lo->lo_file_name,
1530 lo->lo_device->bd_inode->i_mapping->nrpages);
1531 goto out_unfreeze;
1532 }
1533
1534 blk_queue_logical_block_size(lo->lo_queue, arg);
1535 blk_queue_physical_block_size(lo->lo_queue, arg);
1536 blk_queue_io_min(lo->lo_queue, arg);
1537 loop_update_dio(lo);
1538 out_unfreeze:
1539 blk_mq_unfreeze_queue(lo->lo_queue);
1540
1541 return err;
1542 }
1543
1544 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1545 unsigned long arg)
1546 {
1547 int err;
1548
1549 err = mutex_lock_killable(&loop_ctl_mutex);
1550 if (err)
1551 return err;
1552 switch (cmd) {
1553 case LOOP_SET_CAPACITY:
1554 err = loop_set_capacity(lo);
1555 break;
1556 case LOOP_SET_DIRECT_IO:
1557 err = loop_set_dio(lo, arg);
1558 break;
1559 case LOOP_SET_BLOCK_SIZE:
1560 err = loop_set_block_size(lo, arg);
1561 break;
1562 default:
1563 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1564 }
1565 mutex_unlock(&loop_ctl_mutex);
1566 return err;
1567 }
1568
1569 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1570 unsigned int cmd, unsigned long arg)
1571 {
1572 struct loop_device *lo = bdev->bd_disk->private_data;
1573 int err;
1574
1575 switch (cmd) {
1576 case LOOP_SET_FD:
1577 return loop_set_fd(lo, mode, bdev, arg);
1578 case LOOP_CHANGE_FD:
1579 return loop_change_fd(lo, bdev, arg);
1580 case LOOP_CLR_FD:
1581 return loop_clr_fd(lo);
1582 case LOOP_SET_STATUS:
1583 err = -EPERM;
1584 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1585 err = loop_set_status_old(lo,
1586 (struct loop_info __user *)arg);
1587 }
1588 break;
1589 case LOOP_GET_STATUS:
1590 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1591 case LOOP_SET_STATUS64:
1592 err = -EPERM;
1593 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1594 err = loop_set_status64(lo,
1595 (struct loop_info64 __user *) arg);
1596 }
1597 break;
1598 case LOOP_GET_STATUS64:
1599 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1600 case LOOP_SET_CAPACITY:
1601 case LOOP_SET_DIRECT_IO:
1602 case LOOP_SET_BLOCK_SIZE:
1603 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1604 return -EPERM;
1605 /* Fall through */
1606 default:
1607 err = lo_simple_ioctl(lo, cmd, arg);
1608 break;
1609 }
1610
1611 return err;
1612 }
1613
1614 #ifdef CONFIG_COMPAT
1615 struct compat_loop_info {
1616 compat_int_t lo_number; /* ioctl r/o */
1617 compat_dev_t lo_device; /* ioctl r/o */
1618 compat_ulong_t lo_inode; /* ioctl r/o */
1619 compat_dev_t lo_rdevice; /* ioctl r/o */
1620 compat_int_t lo_offset;
1621 compat_int_t lo_encrypt_type;
1622 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1623 compat_int_t lo_flags; /* ioctl r/o */
1624 char lo_name[LO_NAME_SIZE];
1625 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1626 compat_ulong_t lo_init[2];
1627 char reserved[4];
1628 };
1629
1630 /*
1631 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1632 * - noinlined to reduce stack space usage in main part of driver
1633 */
1634 static noinline int
1635 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1636 struct loop_info64 *info64)
1637 {
1638 struct compat_loop_info info;
1639
1640 if (copy_from_user(&info, arg, sizeof(info)))
1641 return -EFAULT;
1642
1643 memset(info64, 0, sizeof(*info64));
1644 info64->lo_number = info.lo_number;
1645 info64->lo_device = info.lo_device;
1646 info64->lo_inode = info.lo_inode;
1647 info64->lo_rdevice = info.lo_rdevice;
1648 info64->lo_offset = info.lo_offset;
1649 info64->lo_sizelimit = 0;
1650 info64->lo_encrypt_type = info.lo_encrypt_type;
1651 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1652 info64->lo_flags = info.lo_flags;
1653 info64->lo_init[0] = info.lo_init[0];
1654 info64->lo_init[1] = info.lo_init[1];
1655 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1656 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1657 else
1658 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1659 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1660 return 0;
1661 }
1662
1663 /*
1664 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1665 * - noinlined to reduce stack space usage in main part of driver
1666 */
1667 static noinline int
1668 loop_info64_to_compat(const struct loop_info64 *info64,
1669 struct compat_loop_info __user *arg)
1670 {
1671 struct compat_loop_info info;
1672
1673 memset(&info, 0, sizeof(info));
1674 info.lo_number = info64->lo_number;
1675 info.lo_device = info64->lo_device;
1676 info.lo_inode = info64->lo_inode;
1677 info.lo_rdevice = info64->lo_rdevice;
1678 info.lo_offset = info64->lo_offset;
1679 info.lo_encrypt_type = info64->lo_encrypt_type;
1680 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1681 info.lo_flags = info64->lo_flags;
1682 info.lo_init[0] = info64->lo_init[0];
1683 info.lo_init[1] = info64->lo_init[1];
1684 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1685 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1686 else
1687 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1688 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1689
1690 /* error in case values were truncated */
1691 if (info.lo_device != info64->lo_device ||
1692 info.lo_rdevice != info64->lo_rdevice ||
1693 info.lo_inode != info64->lo_inode ||
1694 info.lo_offset != info64->lo_offset ||
1695 info.lo_init[0] != info64->lo_init[0] ||
1696 info.lo_init[1] != info64->lo_init[1])
1697 return -EOVERFLOW;
1698
1699 if (copy_to_user(arg, &info, sizeof(info)))
1700 return -EFAULT;
1701 return 0;
1702 }
1703
1704 static int
1705 loop_set_status_compat(struct loop_device *lo,
1706 const struct compat_loop_info __user *arg)
1707 {
1708 struct loop_info64 info64;
1709 int ret;
1710
1711 ret = loop_info64_from_compat(arg, &info64);
1712 if (ret < 0)
1713 return ret;
1714 return loop_set_status(lo, &info64);
1715 }
1716
1717 static int
1718 loop_get_status_compat(struct loop_device *lo,
1719 struct compat_loop_info __user *arg)
1720 {
1721 struct loop_info64 info64;
1722 int err;
1723
1724 if (!arg)
1725 return -EINVAL;
1726 err = loop_get_status(lo, &info64);
1727 if (!err)
1728 err = loop_info64_to_compat(&info64, arg);
1729 return err;
1730 }
1731
1732 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1733 unsigned int cmd, unsigned long arg)
1734 {
1735 struct loop_device *lo = bdev->bd_disk->private_data;
1736 int err;
1737
1738 switch(cmd) {
1739 case LOOP_SET_STATUS:
1740 err = loop_set_status_compat(lo,
1741 (const struct compat_loop_info __user *)arg);
1742 break;
1743 case LOOP_GET_STATUS:
1744 err = loop_get_status_compat(lo,
1745 (struct compat_loop_info __user *)arg);
1746 break;
1747 case LOOP_SET_CAPACITY:
1748 case LOOP_CLR_FD:
1749 case LOOP_GET_STATUS64:
1750 case LOOP_SET_STATUS64:
1751 arg = (unsigned long) compat_ptr(arg);
1752 /* fall through */
1753 case LOOP_SET_FD:
1754 case LOOP_CHANGE_FD:
1755 case LOOP_SET_BLOCK_SIZE:
1756 err = lo_ioctl(bdev, mode, cmd, arg);
1757 break;
1758 default:
1759 err = -ENOIOCTLCMD;
1760 break;
1761 }
1762 return err;
1763 }
1764 #endif
1765
1766 static int lo_open(struct block_device *bdev, fmode_t mode)
1767 {
1768 struct loop_device *lo;
1769 int err;
1770
1771 err = mutex_lock_killable(&loop_ctl_mutex);
1772 if (err)
1773 return err;
1774 lo = bdev->bd_disk->private_data;
1775 if (!lo) {
1776 err = -ENXIO;
1777 goto out;
1778 }
1779
1780 atomic_inc(&lo->lo_refcnt);
1781 out:
1782 mutex_unlock(&loop_ctl_mutex);
1783 return err;
1784 }
1785
1786 static void lo_release(struct gendisk *disk, fmode_t mode)
1787 {
1788 struct loop_device *lo;
1789
1790 mutex_lock(&loop_ctl_mutex);
1791 lo = disk->private_data;
1792 if (atomic_dec_return(&lo->lo_refcnt))
1793 goto out_unlock;
1794
1795 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1796 if (lo->lo_state != Lo_bound)
1797 goto out_unlock;
1798 lo->lo_state = Lo_rundown;
1799 mutex_unlock(&loop_ctl_mutex);
1800 /*
1801 * In autoclear mode, stop the loop thread
1802 * and remove configuration after last close.
1803 */
1804 __loop_clr_fd(lo, true);
1805 return;
1806 } else if (lo->lo_state == Lo_bound) {
1807 /*
1808 * Otherwise keep thread (if running) and config,
1809 * but flush possible ongoing bios in thread.
1810 */
1811 blk_mq_freeze_queue(lo->lo_queue);
1812 blk_mq_unfreeze_queue(lo->lo_queue);
1813 }
1814
1815 out_unlock:
1816 mutex_unlock(&loop_ctl_mutex);
1817 }
1818
1819 static const struct block_device_operations lo_fops = {
1820 .owner = THIS_MODULE,
1821 .open = lo_open,
1822 .release = lo_release,
1823 .ioctl = lo_ioctl,
1824 #ifdef CONFIG_COMPAT
1825 .compat_ioctl = lo_compat_ioctl,
1826 #endif
1827 };
1828
1829 /*
1830 * And now the modules code and kernel interface.
1831 */
1832 static int max_loop;
1833 module_param(max_loop, int, 0444);
1834 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1835 module_param(max_part, int, 0444);
1836 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1837 MODULE_LICENSE("GPL");
1838 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1839
1840 int loop_register_transfer(struct loop_func_table *funcs)
1841 {
1842 unsigned int n = funcs->number;
1843
1844 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1845 return -EINVAL;
1846 xfer_funcs[n] = funcs;
1847 return 0;
1848 }
1849
1850 static int unregister_transfer_cb(int id, void *ptr, void *data)
1851 {
1852 struct loop_device *lo = ptr;
1853 struct loop_func_table *xfer = data;
1854
1855 mutex_lock(&loop_ctl_mutex);
1856 if (lo->lo_encryption == xfer)
1857 loop_release_xfer(lo);
1858 mutex_unlock(&loop_ctl_mutex);
1859 return 0;
1860 }
1861
1862 int loop_unregister_transfer(int number)
1863 {
1864 unsigned int n = number;
1865 struct loop_func_table *xfer;
1866
1867 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1868 return -EINVAL;
1869
1870 xfer_funcs[n] = NULL;
1871 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1872 return 0;
1873 }
1874
1875 EXPORT_SYMBOL(loop_register_transfer);
1876 EXPORT_SYMBOL(loop_unregister_transfer);
1877
1878 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1879 const struct blk_mq_queue_data *bd)
1880 {
1881 struct request *rq = bd->rq;
1882 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1883 struct loop_device *lo = rq->q->queuedata;
1884
1885 blk_mq_start_request(rq);
1886
1887 if (lo->lo_state != Lo_bound)
1888 return BLK_STS_IOERR;
1889
1890 switch (req_op(rq)) {
1891 case REQ_OP_FLUSH:
1892 case REQ_OP_DISCARD:
1893 case REQ_OP_WRITE_ZEROES:
1894 cmd->use_aio = false;
1895 break;
1896 default:
1897 cmd->use_aio = lo->use_dio;
1898 break;
1899 }
1900
1901 /* always use the first bio's css */
1902 #ifdef CONFIG_BLK_CGROUP
1903 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1904 cmd->css = &bio_blkcg(rq->bio)->css;
1905 css_get(cmd->css);
1906 } else
1907 #endif
1908 cmd->css = NULL;
1909 kthread_queue_work(&lo->worker, &cmd->work);
1910
1911 return BLK_STS_OK;
1912 }
1913
1914 static void loop_handle_cmd(struct loop_cmd *cmd)
1915 {
1916 struct request *rq = blk_mq_rq_from_pdu(cmd);
1917 const bool write = op_is_write(req_op(rq));
1918 struct loop_device *lo = rq->q->queuedata;
1919 int ret = 0;
1920
1921 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1922 ret = -EIO;
1923 goto failed;
1924 }
1925
1926 ret = do_req_filebacked(lo, rq);
1927 failed:
1928 /* complete non-aio request */
1929 if (!cmd->use_aio || ret) {
1930 cmd->ret = ret ? -EIO : 0;
1931 blk_mq_complete_request(rq);
1932 }
1933 }
1934
1935 static void loop_queue_work(struct kthread_work *work)
1936 {
1937 struct loop_cmd *cmd =
1938 container_of(work, struct loop_cmd, work);
1939
1940 loop_handle_cmd(cmd);
1941 }
1942
1943 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1944 unsigned int hctx_idx, unsigned int numa_node)
1945 {
1946 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1947
1948 kthread_init_work(&cmd->work, loop_queue_work);
1949 return 0;
1950 }
1951
1952 static const struct blk_mq_ops loop_mq_ops = {
1953 .queue_rq = loop_queue_rq,
1954 .init_request = loop_init_request,
1955 .complete = lo_complete_rq,
1956 };
1957
1958 static int loop_add(struct loop_device **l, int i)
1959 {
1960 struct loop_device *lo;
1961 struct gendisk *disk;
1962 int err;
1963
1964 err = -ENOMEM;
1965 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1966 if (!lo)
1967 goto out;
1968
1969 lo->lo_state = Lo_unbound;
1970
1971 /* allocate id, if @id >= 0, we're requesting that specific id */
1972 if (i >= 0) {
1973 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1974 if (err == -ENOSPC)
1975 err = -EEXIST;
1976 } else {
1977 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1978 }
1979 if (err < 0)
1980 goto out_free_dev;
1981 i = err;
1982
1983 err = -ENOMEM;
1984 lo->tag_set.ops = &loop_mq_ops;
1985 lo->tag_set.nr_hw_queues = 1;
1986 lo->tag_set.queue_depth = 128;
1987 lo->tag_set.numa_node = NUMA_NO_NODE;
1988 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1989 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1990 lo->tag_set.driver_data = lo;
1991
1992 err = blk_mq_alloc_tag_set(&lo->tag_set);
1993 if (err)
1994 goto out_free_idr;
1995
1996 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1997 if (IS_ERR(lo->lo_queue)) {
1998 err = PTR_ERR(lo->lo_queue);
1999 goto out_cleanup_tags;
2000 }
2001 lo->lo_queue->queuedata = lo;
2002
2003 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2004
2005 /*
2006 * By default, we do buffer IO, so it doesn't make sense to enable
2007 * merge because the I/O submitted to backing file is handled page by
2008 * page. For directio mode, merge does help to dispatch bigger request
2009 * to underlayer disk. We will enable merge once directio is enabled.
2010 */
2011 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2012
2013 err = -ENOMEM;
2014 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2015 if (!disk)
2016 goto out_free_queue;
2017
2018 /*
2019 * Disable partition scanning by default. The in-kernel partition
2020 * scanning can be requested individually per-device during its
2021 * setup. Userspace can always add and remove partitions from all
2022 * devices. The needed partition minors are allocated from the
2023 * extended minor space, the main loop device numbers will continue
2024 * to match the loop minors, regardless of the number of partitions
2025 * used.
2026 *
2027 * If max_part is given, partition scanning is globally enabled for
2028 * all loop devices. The minors for the main loop devices will be
2029 * multiples of max_part.
2030 *
2031 * Note: Global-for-all-devices, set-only-at-init, read-only module
2032 * parameteters like 'max_loop' and 'max_part' make things needlessly
2033 * complicated, are too static, inflexible and may surprise
2034 * userspace tools. Parameters like this in general should be avoided.
2035 */
2036 if (!part_shift)
2037 disk->flags |= GENHD_FL_NO_PART_SCAN;
2038 disk->flags |= GENHD_FL_EXT_DEVT;
2039 atomic_set(&lo->lo_refcnt, 0);
2040 lo->lo_number = i;
2041 spin_lock_init(&lo->lo_lock);
2042 disk->major = LOOP_MAJOR;
2043 disk->first_minor = i << part_shift;
2044 disk->fops = &lo_fops;
2045 disk->private_data = lo;
2046 disk->queue = lo->lo_queue;
2047 sprintf(disk->disk_name, "loop%d", i);
2048 add_disk(disk);
2049 *l = lo;
2050 return lo->lo_number;
2051
2052 out_free_queue:
2053 blk_cleanup_queue(lo->lo_queue);
2054 out_cleanup_tags:
2055 blk_mq_free_tag_set(&lo->tag_set);
2056 out_free_idr:
2057 idr_remove(&loop_index_idr, i);
2058 out_free_dev:
2059 kfree(lo);
2060 out:
2061 return err;
2062 }
2063
2064 static void loop_remove(struct loop_device *lo)
2065 {
2066 del_gendisk(lo->lo_disk);
2067 blk_cleanup_queue(lo->lo_queue);
2068 blk_mq_free_tag_set(&lo->tag_set);
2069 put_disk(lo->lo_disk);
2070 kfree(lo);
2071 }
2072
2073 static int find_free_cb(int id, void *ptr, void *data)
2074 {
2075 struct loop_device *lo = ptr;
2076 struct loop_device **l = data;
2077
2078 if (lo->lo_state == Lo_unbound) {
2079 *l = lo;
2080 return 1;
2081 }
2082 return 0;
2083 }
2084
2085 static int loop_lookup(struct loop_device **l, int i)
2086 {
2087 struct loop_device *lo;
2088 int ret = -ENODEV;
2089
2090 if (i < 0) {
2091 int err;
2092
2093 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2094 if (err == 1) {
2095 *l = lo;
2096 ret = lo->lo_number;
2097 }
2098 goto out;
2099 }
2100
2101 /* lookup and return a specific i */
2102 lo = idr_find(&loop_index_idr, i);
2103 if (lo) {
2104 *l = lo;
2105 ret = lo->lo_number;
2106 }
2107 out:
2108 return ret;
2109 }
2110
2111 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2112 {
2113 struct loop_device *lo;
2114 struct kobject *kobj;
2115 int err;
2116
2117 mutex_lock(&loop_ctl_mutex);
2118 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2119 if (err < 0)
2120 err = loop_add(&lo, MINOR(dev) >> part_shift);
2121 if (err < 0)
2122 kobj = NULL;
2123 else
2124 kobj = get_disk_and_module(lo->lo_disk);
2125 mutex_unlock(&loop_ctl_mutex);
2126
2127 *part = 0;
2128 return kobj;
2129 }
2130
2131 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2132 unsigned long parm)
2133 {
2134 struct loop_device *lo;
2135 int ret;
2136
2137 ret = mutex_lock_killable(&loop_ctl_mutex);
2138 if (ret)
2139 return ret;
2140
2141 ret = -ENOSYS;
2142 switch (cmd) {
2143 case LOOP_CTL_ADD:
2144 ret = loop_lookup(&lo, parm);
2145 if (ret >= 0) {
2146 ret = -EEXIST;
2147 break;
2148 }
2149 ret = loop_add(&lo, parm);
2150 break;
2151 case LOOP_CTL_REMOVE:
2152 ret = loop_lookup(&lo, parm);
2153 if (ret < 0)
2154 break;
2155 if (lo->lo_state != Lo_unbound) {
2156 ret = -EBUSY;
2157 break;
2158 }
2159 if (atomic_read(&lo->lo_refcnt) > 0) {
2160 ret = -EBUSY;
2161 break;
2162 }
2163 lo->lo_disk->private_data = NULL;
2164 idr_remove(&loop_index_idr, lo->lo_number);
2165 loop_remove(lo);
2166 break;
2167 case LOOP_CTL_GET_FREE:
2168 ret = loop_lookup(&lo, -1);
2169 if (ret >= 0)
2170 break;
2171 ret = loop_add(&lo, -1);
2172 }
2173 mutex_unlock(&loop_ctl_mutex);
2174
2175 return ret;
2176 }
2177
2178 static const struct file_operations loop_ctl_fops = {
2179 .open = nonseekable_open,
2180 .unlocked_ioctl = loop_control_ioctl,
2181 .compat_ioctl = loop_control_ioctl,
2182 .owner = THIS_MODULE,
2183 .llseek = noop_llseek,
2184 };
2185
2186 static struct miscdevice loop_misc = {
2187 .minor = LOOP_CTRL_MINOR,
2188 .name = "loop-control",
2189 .fops = &loop_ctl_fops,
2190 };
2191
2192 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2193 MODULE_ALIAS("devname:loop-control");
2194
2195 static int __init loop_init(void)
2196 {
2197 int i, nr;
2198 unsigned long range;
2199 struct loop_device *lo;
2200 int err;
2201
2202 part_shift = 0;
2203 if (max_part > 0) {
2204 part_shift = fls(max_part);
2205
2206 /*
2207 * Adjust max_part according to part_shift as it is exported
2208 * to user space so that user can decide correct minor number
2209 * if [s]he want to create more devices.
2210 *
2211 * Note that -1 is required because partition 0 is reserved
2212 * for the whole disk.
2213 */
2214 max_part = (1UL << part_shift) - 1;
2215 }
2216
2217 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2218 err = -EINVAL;
2219 goto err_out;
2220 }
2221
2222 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2223 err = -EINVAL;
2224 goto err_out;
2225 }
2226
2227 /*
2228 * If max_loop is specified, create that many devices upfront.
2229 * This also becomes a hard limit. If max_loop is not specified,
2230 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2231 * init time. Loop devices can be requested on-demand with the
2232 * /dev/loop-control interface, or be instantiated by accessing
2233 * a 'dead' device node.
2234 */
2235 if (max_loop) {
2236 nr = max_loop;
2237 range = max_loop << part_shift;
2238 } else {
2239 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2240 range = 1UL << MINORBITS;
2241 }
2242
2243 err = misc_register(&loop_misc);
2244 if (err < 0)
2245 goto err_out;
2246
2247
2248 if (register_blkdev(LOOP_MAJOR, "loop")) {
2249 err = -EIO;
2250 goto misc_out;
2251 }
2252
2253 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2254 THIS_MODULE, loop_probe, NULL, NULL);
2255
2256 /* pre-create number of devices given by config or max_loop */
2257 mutex_lock(&loop_ctl_mutex);
2258 for (i = 0; i < nr; i++)
2259 loop_add(&lo, i);
2260 mutex_unlock(&loop_ctl_mutex);
2261
2262 printk(KERN_INFO "loop: module loaded\n");
2263 return 0;
2264
2265 misc_out:
2266 misc_deregister(&loop_misc);
2267 err_out:
2268 return err;
2269 }
2270
2271 static int loop_exit_cb(int id, void *ptr, void *data)
2272 {
2273 struct loop_device *lo = ptr;
2274
2275 loop_remove(lo);
2276 return 0;
2277 }
2278
2279 static void __exit loop_exit(void)
2280 {
2281 unsigned long range;
2282
2283 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2284
2285 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2286 idr_destroy(&loop_index_idr);
2287
2288 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2289 unregister_blkdev(LOOP_MAJOR, "loop");
2290
2291 misc_deregister(&loop_misc);
2292 }
2293
2294 module_init(loop_init);
2295 module_exit(loop_exit);
2296
2297 #ifndef MODULE
2298 static int __init max_loop_setup(char *str)
2299 {
2300 max_loop = simple_strtol(str, NULL, 0);
2301 return 1;
2302 }
2303
2304 __setup("max_loop=", max_loop_setup);
2305 #endif
Linux with added FPC-III support