?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: PPC: Book3S HV: Flush link stack on guest exit to host kernel
[linux/fpc-iii.git] / drivers / block / loop.c
blobf6f77eaa7217e0df51b51047ca3c734a001d0479
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 | PF_MEMALLOC_NOIO;
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 struct block_device *claimed_bdev = NULL;
928 int lo_flags = 0;
929 int error;
930 loff_t size;
931 bool partscan;
932
933 /* This is safe, since we have a reference from open(). */
934 __module_get(THIS_MODULE);
935
936 error = -EBADF;
937 file = fget(arg);
938 if (!file)
939 goto out;
940
941 /*
942 * If we don't hold exclusive handle for the device, upgrade to it
943 * here to avoid changing device under exclusive owner.
944 */
945 if (!(mode & FMODE_EXCL)) {
946 claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
947 if (IS_ERR(claimed_bdev)) {
948 error = PTR_ERR(claimed_bdev);
949 goto out_putf;
950 }
951 }
952
953 error = mutex_lock_killable(&loop_ctl_mutex);
954 if (error)
955 goto out_bdev;
956
957 error = -EBUSY;
958 if (lo->lo_state != Lo_unbound)
959 goto out_unlock;
960
961 error = loop_validate_file(file, bdev);
962 if (error)
963 goto out_unlock;
964
965 mapping = file->f_mapping;
966 inode = mapping->host;
967
968 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
969 !file->f_op->write_iter)
970 lo_flags |= LO_FLAGS_READ_ONLY;
971
972 error = -EFBIG;
973 size = get_loop_size(lo, file);
974 if ((loff_t)(sector_t)size != size)
975 goto out_unlock;
976 error = loop_prepare_queue(lo);
977 if (error)
978 goto out_unlock;
979
980 error = 0;
981
982 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
983
984 lo->use_dio = false;
985 lo->lo_device = bdev;
986 lo->lo_flags = lo_flags;
987 lo->lo_backing_file = file;
988 lo->transfer = NULL;
989 lo->ioctl = NULL;
990 lo->lo_sizelimit = 0;
991 lo->old_gfp_mask = mapping_gfp_mask(mapping);
992 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
993
994 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
995 blk_queue_write_cache(lo->lo_queue, true, false);
996
997 if (io_is_direct(lo->lo_backing_file) && inode->i_sb->s_bdev) {
998 /* In case of direct I/O, match underlying block size */
999 unsigned short bsize = bdev_logical_block_size(
1000 inode->i_sb->s_bdev);
1001
1002 blk_queue_logical_block_size(lo->lo_queue, bsize);
1003 blk_queue_physical_block_size(lo->lo_queue, bsize);
1004 blk_queue_io_min(lo->lo_queue, bsize);
1005 }
1006
1007 loop_update_rotational(lo);
1008 loop_update_dio(lo);
1009 set_capacity(lo->lo_disk, size);
1010 bd_set_size(bdev, size << 9);
1011 loop_sysfs_init(lo);
1012 /* let user-space know about the new size */
1013 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1014
1015 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1016 block_size(inode->i_bdev) : PAGE_SIZE);
1017
1018 lo->lo_state = Lo_bound;
1019 if (part_shift)
1020 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1021 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1022
1023 /* Grab the block_device to prevent its destruction after we
1024 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1025 */
1026 bdgrab(bdev);
1027 mutex_unlock(&loop_ctl_mutex);
1028 if (partscan)
1029 loop_reread_partitions(lo, bdev);
1030 if (claimed_bdev)
1031 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1032 return 0;
1033
1034 out_unlock:
1035 mutex_unlock(&loop_ctl_mutex);
1036 out_bdev:
1037 if (claimed_bdev)
1038 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1039 out_putf:
1040 fput(file);
1041 out:
1042 /* This is safe: open() is still holding a reference. */
1043 module_put(THIS_MODULE);
1044 return error;
1045 }
1046
1047 static int
1048 loop_release_xfer(struct loop_device *lo)
1049 {
1050 int err = 0;
1051 struct loop_func_table *xfer = lo->lo_encryption;
1052
1053 if (xfer) {
1054 if (xfer->release)
1055 err = xfer->release(lo);
1056 lo->transfer = NULL;
1057 lo->lo_encryption = NULL;
1058 module_put(xfer->owner);
1059 }
1060 return err;
1061 }
1062
1063 static int
1064 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1065 const struct loop_info64 *i)
1066 {
1067 int err = 0;
1068
1069 if (xfer) {
1070 struct module *owner = xfer->owner;
1071
1072 if (!try_module_get(owner))
1073 return -EINVAL;
1074 if (xfer->init)
1075 err = xfer->init(lo, i);
1076 if (err)
1077 module_put(owner);
1078 else
1079 lo->lo_encryption = xfer;
1080 }
1081 return err;
1082 }
1083
1084 static int __loop_clr_fd(struct loop_device *lo, bool release)
1085 {
1086 struct file *filp = NULL;
1087 gfp_t gfp = lo->old_gfp_mask;
1088 struct block_device *bdev = lo->lo_device;
1089 int err = 0;
1090 bool partscan = false;
1091 int lo_number;
1092
1093 mutex_lock(&loop_ctl_mutex);
1094 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1095 err = -ENXIO;
1096 goto out_unlock;
1097 }
1098
1099 filp = lo->lo_backing_file;
1100 if (filp == NULL) {
1101 err = -EINVAL;
1102 goto out_unlock;
1103 }
1104
1105 /* freeze request queue during the transition */
1106 blk_mq_freeze_queue(lo->lo_queue);
1107
1108 spin_lock_irq(&lo->lo_lock);
1109 lo->lo_backing_file = NULL;
1110 spin_unlock_irq(&lo->lo_lock);
1111
1112 loop_release_xfer(lo);
1113 lo->transfer = NULL;
1114 lo->ioctl = NULL;
1115 lo->lo_device = NULL;
1116 lo->lo_encryption = NULL;
1117 lo->lo_offset = 0;
1118 lo->lo_sizelimit = 0;
1119 lo->lo_encrypt_key_size = 0;
1120 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1121 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1122 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1123 blk_queue_logical_block_size(lo->lo_queue, 512);
1124 blk_queue_physical_block_size(lo->lo_queue, 512);
1125 blk_queue_io_min(lo->lo_queue, 512);
1126 if (bdev) {
1127 bdput(bdev);
1128 invalidate_bdev(bdev);
1129 bdev->bd_inode->i_mapping->wb_err = 0;
1130 }
1131 set_capacity(lo->lo_disk, 0);
1132 loop_sysfs_exit(lo);
1133 if (bdev) {
1134 bd_set_size(bdev, 0);
1135 /* let user-space know about this change */
1136 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1137 }
1138 mapping_set_gfp_mask(filp->f_mapping, gfp);
1139 /* This is safe: open() is still holding a reference. */
1140 module_put(THIS_MODULE);
1141 blk_mq_unfreeze_queue(lo->lo_queue);
1142
1143 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1144 lo_number = lo->lo_number;
1145 loop_unprepare_queue(lo);
1146 out_unlock:
1147 mutex_unlock(&loop_ctl_mutex);
1148 if (partscan) {
1149 /*
1150 * bd_mutex has been held already in release path, so don't
1151 * acquire it if this function is called in such case.
1152 *
1153 * If the reread partition isn't from release path, lo_refcnt
1154 * must be at least one and it can only become zero when the
1155 * current holder is released.
1156 */
1157 if (release)
1158 err = __blkdev_reread_part(bdev);
1159 else
1160 err = blkdev_reread_part(bdev);
1161 if (err)
1162 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1163 __func__, lo_number, err);
1164 /* Device is gone, no point in returning error */
1165 err = 0;
1166 }
1167
1168 /*
1169 * lo->lo_state is set to Lo_unbound here after above partscan has
1170 * finished.
1171 *
1172 * There cannot be anybody else entering __loop_clr_fd() as
1173 * lo->lo_backing_file is already cleared and Lo_rundown state
1174 * protects us from all the other places trying to change the 'lo'
1175 * device.
1176 */
1177 mutex_lock(&loop_ctl_mutex);
1178 lo->lo_flags = 0;
1179 if (!part_shift)
1180 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1181 lo->lo_state = Lo_unbound;
1182 mutex_unlock(&loop_ctl_mutex);
1183
1184 /*
1185 * Need not hold loop_ctl_mutex to fput backing file.
1186 * Calling fput holding loop_ctl_mutex triggers a circular
1187 * lock dependency possibility warning as fput can take
1188 * bd_mutex which is usually taken before loop_ctl_mutex.
1189 */
1190 if (filp)
1191 fput(filp);
1192 return err;
1193 }
1194
1195 static int loop_clr_fd(struct loop_device *lo)
1196 {
1197 int err;
1198
1199 err = mutex_lock_killable(&loop_ctl_mutex);
1200 if (err)
1201 return err;
1202 if (lo->lo_state != Lo_bound) {
1203 mutex_unlock(&loop_ctl_mutex);
1204 return -ENXIO;
1205 }
1206 /*
1207 * If we've explicitly asked to tear down the loop device,
1208 * and it has an elevated reference count, set it for auto-teardown when
1209 * the last reference goes away. This stops $!~#$@ udev from
1210 * preventing teardown because it decided that it needs to run blkid on
1211 * the loopback device whenever they appear. xfstests is notorious for
1212 * failing tests because blkid via udev races with a losetup
1213 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1214 * command to fail with EBUSY.
1215 */
1216 if (atomic_read(&lo->lo_refcnt) > 1) {
1217 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1218 mutex_unlock(&loop_ctl_mutex);
1219 return 0;
1220 }
1221 lo->lo_state = Lo_rundown;
1222 mutex_unlock(&loop_ctl_mutex);
1223
1224 return __loop_clr_fd(lo, false);
1225 }
1226
1227 static int
1228 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1229 {
1230 int err;
1231 struct loop_func_table *xfer;
1232 kuid_t uid = current_uid();
1233 struct block_device *bdev;
1234 bool partscan = false;
1235
1236 err = mutex_lock_killable(&loop_ctl_mutex);
1237 if (err)
1238 return err;
1239 if (lo->lo_encrypt_key_size &&
1240 !uid_eq(lo->lo_key_owner, uid) &&
1241 !capable(CAP_SYS_ADMIN)) {
1242 err = -EPERM;
1243 goto out_unlock;
1244 }
1245 if (lo->lo_state != Lo_bound) {
1246 err = -ENXIO;
1247 goto out_unlock;
1248 }
1249 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1250 err = -EINVAL;
1251 goto out_unlock;
1252 }
1253
1254 if (lo->lo_offset != info->lo_offset ||
1255 lo->lo_sizelimit != info->lo_sizelimit) {
1256 sync_blockdev(lo->lo_device);
1257 kill_bdev(lo->lo_device);
1258 }
1259
1260 /* I/O need to be drained during transfer transition */
1261 blk_mq_freeze_queue(lo->lo_queue);
1262
1263 err = loop_release_xfer(lo);
1264 if (err)
1265 goto out_unfreeze;
1266
1267 if (info->lo_encrypt_type) {
1268 unsigned int type = info->lo_encrypt_type;
1269
1270 if (type >= MAX_LO_CRYPT) {
1271 err = -EINVAL;
1272 goto out_unfreeze;
1273 }
1274 xfer = xfer_funcs[type];
1275 if (xfer == NULL) {
1276 err = -EINVAL;
1277 goto out_unfreeze;
1278 }
1279 } else
1280 xfer = NULL;
1281
1282 err = loop_init_xfer(lo, xfer, info);
1283 if (err)
1284 goto out_unfreeze;
1285
1286 if (lo->lo_offset != info->lo_offset ||
1287 lo->lo_sizelimit != info->lo_sizelimit) {
1288 /* kill_bdev should have truncated all the pages */
1289 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1290 err = -EAGAIN;
1291 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1292 __func__, lo->lo_number, lo->lo_file_name,
1293 lo->lo_device->bd_inode->i_mapping->nrpages);
1294 goto out_unfreeze;
1295 }
1296 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1297 err = -EFBIG;
1298 goto out_unfreeze;
1299 }
1300 }
1301
1302 loop_config_discard(lo);
1303
1304 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1305 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1306 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1307 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1308
1309 if (!xfer)
1310 xfer = &none_funcs;
1311 lo->transfer = xfer->transfer;
1312 lo->ioctl = xfer->ioctl;
1313
1314 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1315 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1316 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1317
1318 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1319 lo->lo_init[0] = info->lo_init[0];
1320 lo->lo_init[1] = info->lo_init[1];
1321 if (info->lo_encrypt_key_size) {
1322 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1323 info->lo_encrypt_key_size);
1324 lo->lo_key_owner = uid;
1325 }
1326
1327 /* update dio if lo_offset or transfer is changed */
1328 __loop_update_dio(lo, lo->use_dio);
1329
1330 out_unfreeze:
1331 blk_mq_unfreeze_queue(lo->lo_queue);
1332
1333 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1334 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1335 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1336 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1337 bdev = lo->lo_device;
1338 partscan = true;
1339 }
1340 out_unlock:
1341 mutex_unlock(&loop_ctl_mutex);
1342 if (partscan)
1343 loop_reread_partitions(lo, bdev);
1344
1345 return err;
1346 }
1347
1348 static int
1349 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1350 {
1351 struct path path;
1352 struct kstat stat;
1353 int ret;
1354
1355 ret = mutex_lock_killable(&loop_ctl_mutex);
1356 if (ret)
1357 return ret;
1358 if (lo->lo_state != Lo_bound) {
1359 mutex_unlock(&loop_ctl_mutex);
1360 return -ENXIO;
1361 }
1362
1363 memset(info, 0, sizeof(*info));
1364 info->lo_number = lo->lo_number;
1365 info->lo_offset = lo->lo_offset;
1366 info->lo_sizelimit = lo->lo_sizelimit;
1367 info->lo_flags = lo->lo_flags;
1368 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1369 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1370 info->lo_encrypt_type =
1371 lo->lo_encryption ? lo->lo_encryption->number : 0;
1372 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1373 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1374 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1375 lo->lo_encrypt_key_size);
1376 }
1377
1378 /* Drop loop_ctl_mutex while we call into the filesystem. */
1379 path = lo->lo_backing_file->f_path;
1380 path_get(&path);
1381 mutex_unlock(&loop_ctl_mutex);
1382 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1383 if (!ret) {
1384 info->lo_device = huge_encode_dev(stat.dev);
1385 info->lo_inode = stat.ino;
1386 info->lo_rdevice = huge_encode_dev(stat.rdev);
1387 }
1388 path_put(&path);
1389 return ret;
1390 }
1391
1392 static void
1393 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1394 {
1395 memset(info64, 0, sizeof(*info64));
1396 info64->lo_number = info->lo_number;
1397 info64->lo_device = info->lo_device;
1398 info64->lo_inode = info->lo_inode;
1399 info64->lo_rdevice = info->lo_rdevice;
1400 info64->lo_offset = info->lo_offset;
1401 info64->lo_sizelimit = 0;
1402 info64->lo_encrypt_type = info->lo_encrypt_type;
1403 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1404 info64->lo_flags = info->lo_flags;
1405 info64->lo_init[0] = info->lo_init[0];
1406 info64->lo_init[1] = info->lo_init[1];
1407 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1408 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1409 else
1410 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1411 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1412 }
1413
1414 static int
1415 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1416 {
1417 memset(info, 0, sizeof(*info));
1418 info->lo_number = info64->lo_number;
1419 info->lo_device = info64->lo_device;
1420 info->lo_inode = info64->lo_inode;
1421 info->lo_rdevice = info64->lo_rdevice;
1422 info->lo_offset = info64->lo_offset;
1423 info->lo_encrypt_type = info64->lo_encrypt_type;
1424 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1425 info->lo_flags = info64->lo_flags;
1426 info->lo_init[0] = info64->lo_init[0];
1427 info->lo_init[1] = info64->lo_init[1];
1428 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1429 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1430 else
1431 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1432 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1433
1434 /* error in case values were truncated */
1435 if (info->lo_device != info64->lo_device ||
1436 info->lo_rdevice != info64->lo_rdevice ||
1437 info->lo_inode != info64->lo_inode ||
1438 info->lo_offset != info64->lo_offset)
1439 return -EOVERFLOW;
1440
1441 return 0;
1442 }
1443
1444 static int
1445 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1446 {
1447 struct loop_info info;
1448 struct loop_info64 info64;
1449
1450 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1451 return -EFAULT;
1452 loop_info64_from_old(&info, &info64);
1453 return loop_set_status(lo, &info64);
1454 }
1455
1456 static int
1457 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1458 {
1459 struct loop_info64 info64;
1460
1461 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1462 return -EFAULT;
1463 return loop_set_status(lo, &info64);
1464 }
1465
1466 static int
1467 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1468 struct loop_info info;
1469 struct loop_info64 info64;
1470 int err;
1471
1472 if (!arg)
1473 return -EINVAL;
1474 err = loop_get_status(lo, &info64);
1475 if (!err)
1476 err = loop_info64_to_old(&info64, &info);
1477 if (!err && copy_to_user(arg, &info, sizeof(info)))
1478 err = -EFAULT;
1479
1480 return err;
1481 }
1482
1483 static int
1484 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1485 struct loop_info64 info64;
1486 int err;
1487
1488 if (!arg)
1489 return -EINVAL;
1490 err = loop_get_status(lo, &info64);
1491 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1492 err = -EFAULT;
1493
1494 return err;
1495 }
1496
1497 static int loop_set_capacity(struct loop_device *lo)
1498 {
1499 if (unlikely(lo->lo_state != Lo_bound))
1500 return -ENXIO;
1501
1502 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1503 }
1504
1505 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1506 {
1507 int error = -ENXIO;
1508 if (lo->lo_state != Lo_bound)
1509 goto out;
1510
1511 __loop_update_dio(lo, !!arg);
1512 if (lo->use_dio == !!arg)
1513 return 0;
1514 error = -EINVAL;
1515 out:
1516 return error;
1517 }
1518
1519 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1520 {
1521 int err = 0;
1522
1523 if (lo->lo_state != Lo_bound)
1524 return -ENXIO;
1525
1526 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1527 return -EINVAL;
1528
1529 if (lo->lo_queue->limits.logical_block_size != arg) {
1530 sync_blockdev(lo->lo_device);
1531 kill_bdev(lo->lo_device);
1532 }
1533
1534 blk_mq_freeze_queue(lo->lo_queue);
1535
1536 /* kill_bdev should have truncated all the pages */
1537 if (lo->lo_queue->limits.logical_block_size != arg &&
1538 lo->lo_device->bd_inode->i_mapping->nrpages) {
1539 err = -EAGAIN;
1540 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1541 __func__, lo->lo_number, lo->lo_file_name,
1542 lo->lo_device->bd_inode->i_mapping->nrpages);
1543 goto out_unfreeze;
1544 }
1545
1546 blk_queue_logical_block_size(lo->lo_queue, arg);
1547 blk_queue_physical_block_size(lo->lo_queue, arg);
1548 blk_queue_io_min(lo->lo_queue, arg);
1549 loop_update_dio(lo);
1550 out_unfreeze:
1551 blk_mq_unfreeze_queue(lo->lo_queue);
1552
1553 return err;
1554 }
1555
1556 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1557 unsigned long arg)
1558 {
1559 int err;
1560
1561 err = mutex_lock_killable(&loop_ctl_mutex);
1562 if (err)
1563 return err;
1564 switch (cmd) {
1565 case LOOP_SET_CAPACITY:
1566 err = loop_set_capacity(lo);
1567 break;
1568 case LOOP_SET_DIRECT_IO:
1569 err = loop_set_dio(lo, arg);
1570 break;
1571 case LOOP_SET_BLOCK_SIZE:
1572 err = loop_set_block_size(lo, arg);
1573 break;
1574 default:
1575 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1576 }
1577 mutex_unlock(&loop_ctl_mutex);
1578 return err;
1579 }
1580
1581 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1582 unsigned int cmd, unsigned long arg)
1583 {
1584 struct loop_device *lo = bdev->bd_disk->private_data;
1585 int err;
1586
1587 switch (cmd) {
1588 case LOOP_SET_FD:
1589 return loop_set_fd(lo, mode, bdev, arg);
1590 case LOOP_CHANGE_FD:
1591 return loop_change_fd(lo, bdev, arg);
1592 case LOOP_CLR_FD:
1593 return loop_clr_fd(lo);
1594 case LOOP_SET_STATUS:
1595 err = -EPERM;
1596 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1597 err = loop_set_status_old(lo,
1598 (struct loop_info __user *)arg);
1599 }
1600 break;
1601 case LOOP_GET_STATUS:
1602 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1603 case LOOP_SET_STATUS64:
1604 err = -EPERM;
1605 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1606 err = loop_set_status64(lo,
1607 (struct loop_info64 __user *) arg);
1608 }
1609 break;
1610 case LOOP_GET_STATUS64:
1611 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1612 case LOOP_SET_CAPACITY:
1613 case LOOP_SET_DIRECT_IO:
1614 case LOOP_SET_BLOCK_SIZE:
1615 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1616 return -EPERM;
1617 /* Fall through */
1618 default:
1619 err = lo_simple_ioctl(lo, cmd, arg);
1620 break;
1621 }
1622
1623 return err;
1624 }
1625
1626 #ifdef CONFIG_COMPAT
1627 struct compat_loop_info {
1628 compat_int_t lo_number; /* ioctl r/o */
1629 compat_dev_t lo_device; /* ioctl r/o */
1630 compat_ulong_t lo_inode; /* ioctl r/o */
1631 compat_dev_t lo_rdevice; /* ioctl r/o */
1632 compat_int_t lo_offset;
1633 compat_int_t lo_encrypt_type;
1634 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1635 compat_int_t lo_flags; /* ioctl r/o */
1636 char lo_name[LO_NAME_SIZE];
1637 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1638 compat_ulong_t lo_init[2];
1639 char reserved[4];
1640 };
1641
1642 /*
1643 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1644 * - noinlined to reduce stack space usage in main part of driver
1645 */
1646 static noinline int
1647 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1648 struct loop_info64 *info64)
1649 {
1650 struct compat_loop_info info;
1651
1652 if (copy_from_user(&info, arg, sizeof(info)))
1653 return -EFAULT;
1654
1655 memset(info64, 0, sizeof(*info64));
1656 info64->lo_number = info.lo_number;
1657 info64->lo_device = info.lo_device;
1658 info64->lo_inode = info.lo_inode;
1659 info64->lo_rdevice = info.lo_rdevice;
1660 info64->lo_offset = info.lo_offset;
1661 info64->lo_sizelimit = 0;
1662 info64->lo_encrypt_type = info.lo_encrypt_type;
1663 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1664 info64->lo_flags = info.lo_flags;
1665 info64->lo_init[0] = info.lo_init[0];
1666 info64->lo_init[1] = info.lo_init[1];
1667 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1668 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1669 else
1670 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1671 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1672 return 0;
1673 }
1674
1675 /*
1676 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1677 * - noinlined to reduce stack space usage in main part of driver
1678 */
1679 static noinline int
1680 loop_info64_to_compat(const struct loop_info64 *info64,
1681 struct compat_loop_info __user *arg)
1682 {
1683 struct compat_loop_info info;
1684
1685 memset(&info, 0, sizeof(info));
1686 info.lo_number = info64->lo_number;
1687 info.lo_device = info64->lo_device;
1688 info.lo_inode = info64->lo_inode;
1689 info.lo_rdevice = info64->lo_rdevice;
1690 info.lo_offset = info64->lo_offset;
1691 info.lo_encrypt_type = info64->lo_encrypt_type;
1692 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1693 info.lo_flags = info64->lo_flags;
1694 info.lo_init[0] = info64->lo_init[0];
1695 info.lo_init[1] = info64->lo_init[1];
1696 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1697 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1698 else
1699 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1700 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1701
1702 /* error in case values were truncated */
1703 if (info.lo_device != info64->lo_device ||
1704 info.lo_rdevice != info64->lo_rdevice ||
1705 info.lo_inode != info64->lo_inode ||
1706 info.lo_offset != info64->lo_offset ||
1707 info.lo_init[0] != info64->lo_init[0] ||
1708 info.lo_init[1] != info64->lo_init[1])
1709 return -EOVERFLOW;
1710
1711 if (copy_to_user(arg, &info, sizeof(info)))
1712 return -EFAULT;
1713 return 0;
1714 }
1715
1716 static int
1717 loop_set_status_compat(struct loop_device *lo,
1718 const struct compat_loop_info __user *arg)
1719 {
1720 struct loop_info64 info64;
1721 int ret;
1722
1723 ret = loop_info64_from_compat(arg, &info64);
1724 if (ret < 0)
1725 return ret;
1726 return loop_set_status(lo, &info64);
1727 }
1728
1729 static int
1730 loop_get_status_compat(struct loop_device *lo,
1731 struct compat_loop_info __user *arg)
1732 {
1733 struct loop_info64 info64;
1734 int err;
1735
1736 if (!arg)
1737 return -EINVAL;
1738 err = loop_get_status(lo, &info64);
1739 if (!err)
1740 err = loop_info64_to_compat(&info64, arg);
1741 return err;
1742 }
1743
1744 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1745 unsigned int cmd, unsigned long arg)
1746 {
1747 struct loop_device *lo = bdev->bd_disk->private_data;
1748 int err;
1749
1750 switch(cmd) {
1751 case LOOP_SET_STATUS:
1752 err = loop_set_status_compat(lo,
1753 (const struct compat_loop_info __user *)arg);
1754 break;
1755 case LOOP_GET_STATUS:
1756 err = loop_get_status_compat(lo,
1757 (struct compat_loop_info __user *)arg);
1758 break;
1759 case LOOP_SET_CAPACITY:
1760 case LOOP_CLR_FD:
1761 case LOOP_GET_STATUS64:
1762 case LOOP_SET_STATUS64:
1763 arg = (unsigned long) compat_ptr(arg);
1764 /* fall through */
1765 case LOOP_SET_FD:
1766 case LOOP_CHANGE_FD:
1767 case LOOP_SET_BLOCK_SIZE:
1768 case LOOP_SET_DIRECT_IO:
1769 err = lo_ioctl(bdev, mode, cmd, arg);
1770 break;
1771 default:
1772 err = -ENOIOCTLCMD;
1773 break;
1774 }
1775 return err;
1776 }
1777 #endif
1778
1779 static int lo_open(struct block_device *bdev, fmode_t mode)
1780 {
1781 struct loop_device *lo;
1782 int err;
1783
1784 err = mutex_lock_killable(&loop_ctl_mutex);
1785 if (err)
1786 return err;
1787 lo = bdev->bd_disk->private_data;
1788 if (!lo) {
1789 err = -ENXIO;
1790 goto out;
1791 }
1792
1793 atomic_inc(&lo->lo_refcnt);
1794 out:
1795 mutex_unlock(&loop_ctl_mutex);
1796 return err;
1797 }
1798
1799 static void lo_release(struct gendisk *disk, fmode_t mode)
1800 {
1801 struct loop_device *lo;
1802
1803 mutex_lock(&loop_ctl_mutex);
1804 lo = disk->private_data;
1805 if (atomic_dec_return(&lo->lo_refcnt))
1806 goto out_unlock;
1807
1808 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1809 if (lo->lo_state != Lo_bound)
1810 goto out_unlock;
1811 lo->lo_state = Lo_rundown;
1812 mutex_unlock(&loop_ctl_mutex);
1813 /*
1814 * In autoclear mode, stop the loop thread
1815 * and remove configuration after last close.
1816 */
1817 __loop_clr_fd(lo, true);
1818 return;
1819 } else if (lo->lo_state == Lo_bound) {
1820 /*
1821 * Otherwise keep thread (if running) and config,
1822 * but flush possible ongoing bios in thread.
1823 */
1824 blk_mq_freeze_queue(lo->lo_queue);
1825 blk_mq_unfreeze_queue(lo->lo_queue);
1826 }
1827
1828 out_unlock:
1829 mutex_unlock(&loop_ctl_mutex);
1830 }
1831
1832 static const struct block_device_operations lo_fops = {
1833 .owner = THIS_MODULE,
1834 .open = lo_open,
1835 .release = lo_release,
1836 .ioctl = lo_ioctl,
1837 #ifdef CONFIG_COMPAT
1838 .compat_ioctl = lo_compat_ioctl,
1839 #endif
1840 };
1841
1842 /*
1843 * And now the modules code and kernel interface.
1844 */
1845 static int max_loop;
1846 module_param(max_loop, int, 0444);
1847 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1848 module_param(max_part, int, 0444);
1849 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1850 MODULE_LICENSE("GPL");
1851 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1852
1853 int loop_register_transfer(struct loop_func_table *funcs)
1854 {
1855 unsigned int n = funcs->number;
1856
1857 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1858 return -EINVAL;
1859 xfer_funcs[n] = funcs;
1860 return 0;
1861 }
1862
1863 static int unregister_transfer_cb(int id, void *ptr, void *data)
1864 {
1865 struct loop_device *lo = ptr;
1866 struct loop_func_table *xfer = data;
1867
1868 mutex_lock(&loop_ctl_mutex);
1869 if (lo->lo_encryption == xfer)
1870 loop_release_xfer(lo);
1871 mutex_unlock(&loop_ctl_mutex);
1872 return 0;
1873 }
1874
1875 int loop_unregister_transfer(int number)
1876 {
1877 unsigned int n = number;
1878 struct loop_func_table *xfer;
1879
1880 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1881 return -EINVAL;
1882
1883 xfer_funcs[n] = NULL;
1884 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1885 return 0;
1886 }
1887
1888 EXPORT_SYMBOL(loop_register_transfer);
1889 EXPORT_SYMBOL(loop_unregister_transfer);
1890
1891 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1892 const struct blk_mq_queue_data *bd)
1893 {
1894 struct request *rq = bd->rq;
1895 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1896 struct loop_device *lo = rq->q->queuedata;
1897
1898 blk_mq_start_request(rq);
1899
1900 if (lo->lo_state != Lo_bound)
1901 return BLK_STS_IOERR;
1902
1903 switch (req_op(rq)) {
1904 case REQ_OP_FLUSH:
1905 case REQ_OP_DISCARD:
1906 case REQ_OP_WRITE_ZEROES:
1907 cmd->use_aio = false;
1908 break;
1909 default:
1910 cmd->use_aio = lo->use_dio;
1911 break;
1912 }
1913
1914 /* always use the first bio's css */
1915 #ifdef CONFIG_BLK_CGROUP
1916 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1917 cmd->css = &bio_blkcg(rq->bio)->css;
1918 css_get(cmd->css);
1919 } else
1920 #endif
1921 cmd->css = NULL;
1922 kthread_queue_work(&lo->worker, &cmd->work);
1923
1924 return BLK_STS_OK;
1925 }
1926
1927 static void loop_handle_cmd(struct loop_cmd *cmd)
1928 {
1929 struct request *rq = blk_mq_rq_from_pdu(cmd);
1930 const bool write = op_is_write(req_op(rq));
1931 struct loop_device *lo = rq->q->queuedata;
1932 int ret = 0;
1933
1934 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1935 ret = -EIO;
1936 goto failed;
1937 }
1938
1939 ret = do_req_filebacked(lo, rq);
1940 failed:
1941 /* complete non-aio request */
1942 if (!cmd->use_aio || ret) {
1943 cmd->ret = ret ? -EIO : 0;
1944 blk_mq_complete_request(rq);
1945 }
1946 }
1947
1948 static void loop_queue_work(struct kthread_work *work)
1949 {
1950 struct loop_cmd *cmd =
1951 container_of(work, struct loop_cmd, work);
1952
1953 loop_handle_cmd(cmd);
1954 }
1955
1956 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1957 unsigned int hctx_idx, unsigned int numa_node)
1958 {
1959 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1960
1961 kthread_init_work(&cmd->work, loop_queue_work);
1962 return 0;
1963 }
1964
1965 static const struct blk_mq_ops loop_mq_ops = {
1966 .queue_rq = loop_queue_rq,
1967 .init_request = loop_init_request,
1968 .complete = lo_complete_rq,
1969 };
1970
1971 static int loop_add(struct loop_device **l, int i)
1972 {
1973 struct loop_device *lo;
1974 struct gendisk *disk;
1975 int err;
1976
1977 err = -ENOMEM;
1978 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1979 if (!lo)
1980 goto out;
1981
1982 lo->lo_state = Lo_unbound;
1983
1984 /* allocate id, if @id >= 0, we're requesting that specific id */
1985 if (i >= 0) {
1986 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1987 if (err == -ENOSPC)
1988 err = -EEXIST;
1989 } else {
1990 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1991 }
1992 if (err < 0)
1993 goto out_free_dev;
1994 i = err;
1995
1996 err = -ENOMEM;
1997 lo->tag_set.ops = &loop_mq_ops;
1998 lo->tag_set.nr_hw_queues = 1;
1999 lo->tag_set.queue_depth = 128;
2000 lo->tag_set.numa_node = NUMA_NO_NODE;
2001 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2002 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2003 lo->tag_set.driver_data = lo;
2004
2005 err = blk_mq_alloc_tag_set(&lo->tag_set);
2006 if (err)
2007 goto out_free_idr;
2008
2009 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2010 if (IS_ERR(lo->lo_queue)) {
2011 err = PTR_ERR(lo->lo_queue);
2012 goto out_cleanup_tags;
2013 }
2014 lo->lo_queue->queuedata = lo;
2015
2016 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2017
2018 /*
2019 * By default, we do buffer IO, so it doesn't make sense to enable
2020 * merge because the I/O submitted to backing file is handled page by
2021 * page. For directio mode, merge does help to dispatch bigger request
2022 * to underlayer disk. We will enable merge once directio is enabled.
2023 */
2024 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2025
2026 err = -ENOMEM;
2027 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2028 if (!disk)
2029 goto out_free_queue;
2030
2031 /*
2032 * Disable partition scanning by default. The in-kernel partition
2033 * scanning can be requested individually per-device during its
2034 * setup. Userspace can always add and remove partitions from all
2035 * devices. The needed partition minors are allocated from the
2036 * extended minor space, the main loop device numbers will continue
2037 * to match the loop minors, regardless of the number of partitions
2038 * used.
2039 *
2040 * If max_part is given, partition scanning is globally enabled for
2041 * all loop devices. The minors for the main loop devices will be
2042 * multiples of max_part.
2043 *
2044 * Note: Global-for-all-devices, set-only-at-init, read-only module
2045 * parameteters like 'max_loop' and 'max_part' make things needlessly
2046 * complicated, are too static, inflexible and may surprise
2047 * userspace tools. Parameters like this in general should be avoided.
2048 */
2049 if (!part_shift)
2050 disk->flags |= GENHD_FL_NO_PART_SCAN;
2051 disk->flags |= GENHD_FL_EXT_DEVT;
2052 atomic_set(&lo->lo_refcnt, 0);
2053 lo->lo_number = i;
2054 spin_lock_init(&lo->lo_lock);
2055 disk->major = LOOP_MAJOR;
2056 disk->first_minor = i << part_shift;
2057 disk->fops = &lo_fops;
2058 disk->private_data = lo;
2059 disk->queue = lo->lo_queue;
2060 sprintf(disk->disk_name, "loop%d", i);
2061 add_disk(disk);
2062 *l = lo;
2063 return lo->lo_number;
2064
2065 out_free_queue:
2066 blk_cleanup_queue(lo->lo_queue);
2067 out_cleanup_tags:
2068 blk_mq_free_tag_set(&lo->tag_set);
2069 out_free_idr:
2070 idr_remove(&loop_index_idr, i);
2071 out_free_dev:
2072 kfree(lo);
2073 out:
2074 return err;
2075 }
2076
2077 static void loop_remove(struct loop_device *lo)
2078 {
2079 del_gendisk(lo->lo_disk);
2080 blk_cleanup_queue(lo->lo_queue);
2081 blk_mq_free_tag_set(&lo->tag_set);
2082 put_disk(lo->lo_disk);
2083 kfree(lo);
2084 }
2085
2086 static int find_free_cb(int id, void *ptr, void *data)
2087 {
2088 struct loop_device *lo = ptr;
2089 struct loop_device **l = data;
2090
2091 if (lo->lo_state == Lo_unbound) {
2092 *l = lo;
2093 return 1;
2094 }
2095 return 0;
2096 }
2097
2098 static int loop_lookup(struct loop_device **l, int i)
2099 {
2100 struct loop_device *lo;
2101 int ret = -ENODEV;
2102
2103 if (i < 0) {
2104 int err;
2105
2106 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2107 if (err == 1) {
2108 *l = lo;
2109 ret = lo->lo_number;
2110 }
2111 goto out;
2112 }
2113
2114 /* lookup and return a specific i */
2115 lo = idr_find(&loop_index_idr, i);
2116 if (lo) {
2117 *l = lo;
2118 ret = lo->lo_number;
2119 }
2120 out:
2121 return ret;
2122 }
2123
2124 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2125 {
2126 struct loop_device *lo;
2127 struct kobject *kobj;
2128 int err;
2129
2130 mutex_lock(&loop_ctl_mutex);
2131 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2132 if (err < 0)
2133 err = loop_add(&lo, MINOR(dev) >> part_shift);
2134 if (err < 0)
2135 kobj = NULL;
2136 else
2137 kobj = get_disk_and_module(lo->lo_disk);
2138 mutex_unlock(&loop_ctl_mutex);
2139
2140 *part = 0;
2141 return kobj;
2142 }
2143
2144 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2145 unsigned long parm)
2146 {
2147 struct loop_device *lo;
2148 int ret;
2149
2150 ret = mutex_lock_killable(&loop_ctl_mutex);
2151 if (ret)
2152 return ret;
2153
2154 ret = -ENOSYS;
2155 switch (cmd) {
2156 case LOOP_CTL_ADD:
2157 ret = loop_lookup(&lo, parm);
2158 if (ret >= 0) {
2159 ret = -EEXIST;
2160 break;
2161 }
2162 ret = loop_add(&lo, parm);
2163 break;
2164 case LOOP_CTL_REMOVE:
2165 ret = loop_lookup(&lo, parm);
2166 if (ret < 0)
2167 break;
2168 if (lo->lo_state != Lo_unbound) {
2169 ret = -EBUSY;
2170 break;
2171 }
2172 if (atomic_read(&lo->lo_refcnt) > 0) {
2173 ret = -EBUSY;
2174 break;
2175 }
2176 lo->lo_disk->private_data = NULL;
2177 idr_remove(&loop_index_idr, lo->lo_number);
2178 loop_remove(lo);
2179 break;
2180 case LOOP_CTL_GET_FREE:
2181 ret = loop_lookup(&lo, -1);
2182 if (ret >= 0)
2183 break;
2184 ret = loop_add(&lo, -1);
2185 }
2186 mutex_unlock(&loop_ctl_mutex);
2187
2188 return ret;
2189 }
2190
2191 static const struct file_operations loop_ctl_fops = {
2192 .open = nonseekable_open,
2193 .unlocked_ioctl = loop_control_ioctl,
2194 .compat_ioctl = loop_control_ioctl,
2195 .owner = THIS_MODULE,
2196 .llseek = noop_llseek,
2197 };
2198
2199 static struct miscdevice loop_misc = {
2200 .minor = LOOP_CTRL_MINOR,
2201 .name = "loop-control",
2202 .fops = &loop_ctl_fops,
2203 };
2204
2205 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2206 MODULE_ALIAS("devname:loop-control");
2207
2208 static int __init loop_init(void)
2209 {
2210 int i, nr;
2211 unsigned long range;
2212 struct loop_device *lo;
2213 int err;
2214
2215 part_shift = 0;
2216 if (max_part > 0) {
2217 part_shift = fls(max_part);
2218
2219 /*
2220 * Adjust max_part according to part_shift as it is exported
2221 * to user space so that user can decide correct minor number
2222 * if [s]he want to create more devices.
2223 *
2224 * Note that -1 is required because partition 0 is reserved
2225 * for the whole disk.
2226 */
2227 max_part = (1UL << part_shift) - 1;
2228 }
2229
2230 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2231 err = -EINVAL;
2232 goto err_out;
2233 }
2234
2235 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2236 err = -EINVAL;
2237 goto err_out;
2238 }
2239
2240 /*
2241 * If max_loop is specified, create that many devices upfront.
2242 * This also becomes a hard limit. If max_loop is not specified,
2243 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2244 * init time. Loop devices can be requested on-demand with the
2245 * /dev/loop-control interface, or be instantiated by accessing
2246 * a 'dead' device node.
2247 */
2248 if (max_loop) {
2249 nr = max_loop;
2250 range = max_loop << part_shift;
2251 } else {
2252 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2253 range = 1UL << MINORBITS;
2254 }
2255
2256 err = misc_register(&loop_misc);
2257 if (err < 0)
2258 goto err_out;
2259
2260
2261 if (register_blkdev(LOOP_MAJOR, "loop")) {
2262 err = -EIO;
2263 goto misc_out;
2264 }
2265
2266 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2267 THIS_MODULE, loop_probe, NULL, NULL);
2268
2269 /* pre-create number of devices given by config or max_loop */
2270 mutex_lock(&loop_ctl_mutex);
2271 for (i = 0; i < nr; i++)
2272 loop_add(&lo, i);
2273 mutex_unlock(&loop_ctl_mutex);
2274
2275 printk(KERN_INFO "loop: module loaded\n");
2276 return 0;
2277
2278 misc_out:
2279 misc_deregister(&loop_misc);
2280 err_out:
2281 return err;
2282 }
2283
2284 static int loop_exit_cb(int id, void *ptr, void *data)
2285 {
2286 struct loop_device *lo = ptr;
2287
2288 loop_remove(lo);
2289 return 0;
2290 }
2291
2292 static void __exit loop_exit(void)
2293 {
2294 unsigned long range;
2295
2296 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2297
2298 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2299 idr_destroy(&loop_index_idr);
2300
2301 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2302 unregister_blkdev(LOOP_MAJOR, "loop");
2303
2304 misc_deregister(&loop_misc);
2305 }
2306
2307 module_init(loop_init);
2308 module_exit(loop_exit);
2309
2310 #ifndef MODULE
2311 static int __init max_loop_setup(char *str)
2312 {
2313 max_loop = simple_strtol(str, NULL, 0);
2314 return 1;
2315 }
2316
2317 __setup("max_loop=", max_loop_setup);
2318 #endif
Linux with added FPC-III support