perf bench futex: Cache align the worker struct
[linux/fpc-iii.git] / drivers / block / loop.c
blobfa1b7a90ba11deecda3ba5853b545fc2fc0dc579
1 /*
2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
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
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
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
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
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
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
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
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
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
46 * Still To Fix:
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
55 #include <linux/fs.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
79 #include "loop.h"
81 #include <asm/uaccess.h>
83 static DEFINE_IDR(loop_index_idr);
84 static DEFINE_MUTEX(loop_index_mutex);
86 static int max_part;
87 static int part_shift;
89 static int transfer_xor(struct loop_device *lo, int cmd,
90 struct page *raw_page, unsigned raw_off,
91 struct page *loop_page, unsigned loop_off,
92 int size, sector_t real_block)
94 char *raw_buf = kmap_atomic(raw_page) + raw_off;
95 char *loop_buf = kmap_atomic(loop_page) + loop_off;
96 char *in, *out, *key;
97 int i, keysize;
99 if (cmd == READ) {
100 in = raw_buf;
101 out = loop_buf;
102 } else {
103 in = loop_buf;
104 out = raw_buf;
107 key = lo->lo_encrypt_key;
108 keysize = lo->lo_encrypt_key_size;
109 for (i = 0; i < size; i++)
110 *out++ = *in++ ^ key[(i & 511) % keysize];
112 kunmap_atomic(loop_buf);
113 kunmap_atomic(raw_buf);
114 cond_resched();
115 return 0;
118 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
120 if (unlikely(info->lo_encrypt_key_size <= 0))
121 return -EINVAL;
122 return 0;
125 static struct loop_func_table none_funcs = {
126 .number = LO_CRYPT_NONE,
129 static struct loop_func_table xor_funcs = {
130 .number = LO_CRYPT_XOR,
131 .transfer = transfer_xor,
132 .init = xor_init
135 /* xfer_funcs[0] is special - its release function is never called */
136 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
137 &none_funcs,
138 &xor_funcs
141 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
143 loff_t loopsize;
145 /* Compute loopsize in bytes */
146 loopsize = i_size_read(file->f_mapping->host);
147 if (offset > 0)
148 loopsize -= offset;
149 /* offset is beyond i_size, weird but possible */
150 if (loopsize < 0)
151 return 0;
153 if (sizelimit > 0 && sizelimit < loopsize)
154 loopsize = sizelimit;
156 * Unfortunately, if we want to do I/O on the device,
157 * the number of 512-byte sectors has to fit into a sector_t.
159 return loopsize >> 9;
162 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
167 static void __loop_update_dio(struct loop_device *lo, bool dio)
169 struct file *file = lo->lo_backing_file;
170 struct address_space *mapping = file->f_mapping;
171 struct inode *inode = mapping->host;
172 unsigned short sb_bsize = 0;
173 unsigned dio_align = 0;
174 bool use_dio;
176 if (inode->i_sb->s_bdev) {
177 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
178 dio_align = sb_bsize - 1;
182 * We support direct I/O only if lo_offset is aligned with the
183 * logical I/O size of backing device, and the logical block
184 * size of loop is bigger than the backing device's and the loop
185 * needn't transform transfer.
187 * TODO: the above condition may be loosed in the future, and
188 * direct I/O may be switched runtime at that time because most
189 * of requests in sane appplications should be PAGE_SIZE algined
191 if (dio) {
192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
193 !(lo->lo_offset & dio_align) &&
194 mapping->a_ops->direct_IO &&
195 !lo->transfer)
196 use_dio = true;
197 else
198 use_dio = false;
199 } else {
200 use_dio = false;
203 if (lo->use_dio == use_dio)
204 return;
206 /* flush dirty pages before changing direct IO */
207 vfs_fsync(file, 0);
210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 * will get updated by ioctl(LOOP_GET_STATUS)
214 blk_mq_freeze_queue(lo->lo_queue);
215 lo->use_dio = use_dio;
216 if (use_dio)
217 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
218 else
219 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
220 blk_mq_unfreeze_queue(lo->lo_queue);
223 static int
224 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
226 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
227 sector_t x = (sector_t)size;
228 struct block_device *bdev = lo->lo_device;
230 if (unlikely((loff_t)x != size))
231 return -EFBIG;
232 if (lo->lo_offset != offset)
233 lo->lo_offset = offset;
234 if (lo->lo_sizelimit != sizelimit)
235 lo->lo_sizelimit = sizelimit;
236 set_capacity(lo->lo_disk, x);
237 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
238 /* let user-space know about the new size */
239 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
240 return 0;
243 static inline int
244 lo_do_transfer(struct loop_device *lo, int cmd,
245 struct page *rpage, unsigned roffs,
246 struct page *lpage, unsigned loffs,
247 int size, sector_t rblock)
249 int ret;
251 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
252 if (likely(!ret))
253 return 0;
255 printk_ratelimited(KERN_ERR
256 "loop: Transfer error at byte offset %llu, length %i.\n",
257 (unsigned long long)rblock << 9, size);
258 return ret;
261 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
263 struct iov_iter i;
264 ssize_t bw;
266 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);
268 file_start_write(file);
269 bw = vfs_iter_write(file, &i, ppos);
270 file_end_write(file);
272 if (likely(bw == bvec->bv_len))
273 return 0;
275 printk_ratelimited(KERN_ERR
276 "loop: Write error at byte offset %llu, length %i.\n",
277 (unsigned long long)*ppos, bvec->bv_len);
278 if (bw >= 0)
279 bw = -EIO;
280 return bw;
283 static int lo_write_simple(struct loop_device *lo, struct request *rq,
284 loff_t pos)
286 struct bio_vec bvec;
287 struct req_iterator iter;
288 int ret = 0;
290 rq_for_each_segment(bvec, rq, iter) {
291 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
292 if (ret < 0)
293 break;
294 cond_resched();
297 return ret;
301 * This is the slow, transforming version that needs to double buffer the
302 * data as it cannot do the transformations in place without having direct
303 * access to the destination pages of the backing file.
305 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
306 loff_t pos)
308 struct bio_vec bvec, b;
309 struct req_iterator iter;
310 struct page *page;
311 int ret = 0;
313 page = alloc_page(GFP_NOIO);
314 if (unlikely(!page))
315 return -ENOMEM;
317 rq_for_each_segment(bvec, rq, iter) {
318 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
319 bvec.bv_offset, bvec.bv_len, pos >> 9);
320 if (unlikely(ret))
321 break;
323 b.bv_page = page;
324 b.bv_offset = 0;
325 b.bv_len = bvec.bv_len;
326 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
327 if (ret < 0)
328 break;
331 __free_page(page);
332 return ret;
335 static int lo_read_simple(struct loop_device *lo, struct request *rq,
336 loff_t pos)
338 struct bio_vec bvec;
339 struct req_iterator iter;
340 struct iov_iter i;
341 ssize_t len;
343 rq_for_each_segment(bvec, rq, iter) {
344 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
345 len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
346 if (len < 0)
347 return len;
349 flush_dcache_page(bvec.bv_page);
351 if (len != bvec.bv_len) {
352 struct bio *bio;
354 __rq_for_each_bio(bio, rq)
355 zero_fill_bio(bio);
356 break;
358 cond_resched();
361 return 0;
364 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
365 loff_t pos)
367 struct bio_vec bvec, b;
368 struct req_iterator iter;
369 struct iov_iter i;
370 struct page *page;
371 ssize_t len;
372 int ret = 0;
374 page = alloc_page(GFP_NOIO);
375 if (unlikely(!page))
376 return -ENOMEM;
378 rq_for_each_segment(bvec, rq, iter) {
379 loff_t offset = pos;
381 b.bv_page = page;
382 b.bv_offset = 0;
383 b.bv_len = bvec.bv_len;
385 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
386 len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
387 if (len < 0) {
388 ret = len;
389 goto out_free_page;
392 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
393 bvec.bv_offset, len, offset >> 9);
394 if (ret)
395 goto out_free_page;
397 flush_dcache_page(bvec.bv_page);
399 if (len != bvec.bv_len) {
400 struct bio *bio;
402 __rq_for_each_bio(bio, rq)
403 zero_fill_bio(bio);
404 break;
408 ret = 0;
409 out_free_page:
410 __free_page(page);
411 return ret;
414 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
417 * We use punch hole to reclaim the free space used by the
418 * image a.k.a. discard. However we do not support discard if
419 * encryption is enabled, because it may give an attacker
420 * useful information.
422 struct file *file = lo->lo_backing_file;
423 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
424 int ret;
426 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
427 ret = -EOPNOTSUPP;
428 goto out;
431 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
432 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
433 ret = -EIO;
434 out:
435 return ret;
438 static int lo_req_flush(struct loop_device *lo, struct request *rq)
440 struct file *file = lo->lo_backing_file;
441 int ret = vfs_fsync(file, 0);
442 if (unlikely(ret && ret != -EINVAL))
443 ret = -EIO;
445 return ret;
448 static inline void handle_partial_read(struct loop_cmd *cmd, long bytes)
450 if (bytes < 0 || op_is_write(req_op(cmd->rq)))
451 return;
453 if (unlikely(bytes < blk_rq_bytes(cmd->rq))) {
454 struct bio *bio = cmd->rq->bio;
456 bio_advance(bio, bytes);
457 zero_fill_bio(bio);
461 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
463 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
464 struct request *rq = cmd->rq;
466 handle_partial_read(cmd, ret);
468 if (ret > 0)
469 ret = 0;
470 else if (ret < 0)
471 ret = -EIO;
473 blk_mq_complete_request(rq, ret);
476 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
477 loff_t pos, bool rw)
479 struct iov_iter iter;
480 struct bio_vec *bvec;
481 struct bio *bio = cmd->rq->bio;
482 struct file *file = lo->lo_backing_file;
483 int ret;
485 /* nomerge for loop request queue */
486 WARN_ON(cmd->rq->bio != cmd->rq->biotail);
488 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
489 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
490 bio_segments(bio), blk_rq_bytes(cmd->rq));
492 * This bio may be started from the middle of the 'bvec'
493 * because of bio splitting, so offset from the bvec must
494 * be passed to iov iterator
496 iter.iov_offset = bio->bi_iter.bi_bvec_done;
498 cmd->iocb.ki_pos = pos;
499 cmd->iocb.ki_filp = file;
500 cmd->iocb.ki_complete = lo_rw_aio_complete;
501 cmd->iocb.ki_flags = IOCB_DIRECT;
503 if (rw == WRITE)
504 ret = file->f_op->write_iter(&cmd->iocb, &iter);
505 else
506 ret = file->f_op->read_iter(&cmd->iocb, &iter);
508 if (ret != -EIOCBQUEUED)
509 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
510 return 0;
513 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
515 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
516 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
519 * lo_write_simple and lo_read_simple should have been covered
520 * by io submit style function like lo_rw_aio(), one blocker
521 * is that lo_read_simple() need to call flush_dcache_page after
522 * the page is written from kernel, and it isn't easy to handle
523 * this in io submit style function which submits all segments
524 * of the req at one time. And direct read IO doesn't need to
525 * run flush_dcache_page().
527 switch (req_op(rq)) {
528 case REQ_OP_FLUSH:
529 return lo_req_flush(lo, rq);
530 case REQ_OP_DISCARD:
531 return lo_discard(lo, rq, pos);
532 case REQ_OP_WRITE:
533 if (lo->transfer)
534 return lo_write_transfer(lo, rq, pos);
535 else if (cmd->use_aio)
536 return lo_rw_aio(lo, cmd, pos, WRITE);
537 else
538 return lo_write_simple(lo, rq, pos);
539 case REQ_OP_READ:
540 if (lo->transfer)
541 return lo_read_transfer(lo, rq, pos);
542 else if (cmd->use_aio)
543 return lo_rw_aio(lo, cmd, pos, READ);
544 else
545 return lo_read_simple(lo, rq, pos);
546 default:
547 WARN_ON_ONCE(1);
548 return -EIO;
549 break;
553 struct switch_request {
554 struct file *file;
555 struct completion wait;
558 static inline void loop_update_dio(struct loop_device *lo)
560 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
561 lo->use_dio);
565 * Do the actual switch; called from the BIO completion routine
567 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
569 struct file *file = p->file;
570 struct file *old_file = lo->lo_backing_file;
571 struct address_space *mapping;
573 /* if no new file, only flush of queued bios requested */
574 if (!file)
575 return;
577 mapping = file->f_mapping;
578 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
579 lo->lo_backing_file = file;
580 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
581 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
582 lo->old_gfp_mask = mapping_gfp_mask(mapping);
583 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
584 loop_update_dio(lo);
588 * loop_switch performs the hard work of switching a backing store.
589 * First it needs to flush existing IO, it does this by sending a magic
590 * BIO down the pipe. The completion of this BIO does the actual switch.
592 static int loop_switch(struct loop_device *lo, struct file *file)
594 struct switch_request w;
596 w.file = file;
598 /* freeze queue and wait for completion of scheduled requests */
599 blk_mq_freeze_queue(lo->lo_queue);
601 /* do the switch action */
602 do_loop_switch(lo, &w);
604 /* unfreeze */
605 blk_mq_unfreeze_queue(lo->lo_queue);
607 return 0;
611 * Helper to flush the IOs in loop, but keeping loop thread running
613 static int loop_flush(struct loop_device *lo)
615 return loop_switch(lo, NULL);
618 static void loop_reread_partitions(struct loop_device *lo,
619 struct block_device *bdev)
621 int rc;
624 * bd_mutex has been held already in release path, so don't
625 * acquire it if this function is called in such case.
627 * If the reread partition isn't from release path, lo_refcnt
628 * must be at least one and it can only become zero when the
629 * current holder is released.
631 if (!atomic_read(&lo->lo_refcnt))
632 rc = __blkdev_reread_part(bdev);
633 else
634 rc = blkdev_reread_part(bdev);
635 if (rc)
636 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
637 __func__, lo->lo_number, lo->lo_file_name, rc);
641 * loop_change_fd switched the backing store of a loopback device to
642 * a new file. This is useful for operating system installers to free up
643 * the original file and in High Availability environments to switch to
644 * an alternative location for the content in case of server meltdown.
645 * This can only work if the loop device is used read-only, and if the
646 * new backing store is the same size and type as the old backing store.
648 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
649 unsigned int arg)
651 struct file *file, *old_file;
652 struct inode *inode;
653 int error;
655 error = -ENXIO;
656 if (lo->lo_state != Lo_bound)
657 goto out;
659 /* the loop device has to be read-only */
660 error = -EINVAL;
661 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
662 goto out;
664 error = -EBADF;
665 file = fget(arg);
666 if (!file)
667 goto out;
669 inode = file->f_mapping->host;
670 old_file = lo->lo_backing_file;
672 error = -EINVAL;
674 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
675 goto out_putf;
677 /* size of the new backing store needs to be the same */
678 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
679 goto out_putf;
681 /* and ... switch */
682 error = loop_switch(lo, file);
683 if (error)
684 goto out_putf;
686 fput(old_file);
687 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
688 loop_reread_partitions(lo, bdev);
689 return 0;
691 out_putf:
692 fput(file);
693 out:
694 return error;
697 static inline int is_loop_device(struct file *file)
699 struct inode *i = file->f_mapping->host;
701 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
704 /* loop sysfs attributes */
706 static ssize_t loop_attr_show(struct device *dev, char *page,
707 ssize_t (*callback)(struct loop_device *, char *))
709 struct gendisk *disk = dev_to_disk(dev);
710 struct loop_device *lo = disk->private_data;
712 return callback(lo, page);
715 #define LOOP_ATTR_RO(_name) \
716 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
717 static ssize_t loop_attr_do_show_##_name(struct device *d, \
718 struct device_attribute *attr, char *b) \
720 return loop_attr_show(d, b, loop_attr_##_name##_show); \
722 static struct device_attribute loop_attr_##_name = \
723 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
725 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
727 ssize_t ret;
728 char *p = NULL;
730 spin_lock_irq(&lo->lo_lock);
731 if (lo->lo_backing_file)
732 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
733 spin_unlock_irq(&lo->lo_lock);
735 if (IS_ERR_OR_NULL(p))
736 ret = PTR_ERR(p);
737 else {
738 ret = strlen(p);
739 memmove(buf, p, ret);
740 buf[ret++] = '\n';
741 buf[ret] = 0;
744 return ret;
747 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
749 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
752 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
754 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
757 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
759 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
761 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
764 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
766 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
768 return sprintf(buf, "%s\n", partscan ? "1" : "0");
771 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
773 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
775 return sprintf(buf, "%s\n", dio ? "1" : "0");
778 LOOP_ATTR_RO(backing_file);
779 LOOP_ATTR_RO(offset);
780 LOOP_ATTR_RO(sizelimit);
781 LOOP_ATTR_RO(autoclear);
782 LOOP_ATTR_RO(partscan);
783 LOOP_ATTR_RO(dio);
785 static struct attribute *loop_attrs[] = {
786 &loop_attr_backing_file.attr,
787 &loop_attr_offset.attr,
788 &loop_attr_sizelimit.attr,
789 &loop_attr_autoclear.attr,
790 &loop_attr_partscan.attr,
791 &loop_attr_dio.attr,
792 NULL,
795 static struct attribute_group loop_attribute_group = {
796 .name = "loop",
797 .attrs= loop_attrs,
800 static int loop_sysfs_init(struct loop_device *lo)
802 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
803 &loop_attribute_group);
806 static void loop_sysfs_exit(struct loop_device *lo)
808 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
809 &loop_attribute_group);
812 static void loop_config_discard(struct loop_device *lo)
814 struct file *file = lo->lo_backing_file;
815 struct inode *inode = file->f_mapping->host;
816 struct request_queue *q = lo->lo_queue;
819 * We use punch hole to reclaim the free space used by the
820 * image a.k.a. discard. However we do not support discard if
821 * encryption is enabled, because it may give an attacker
822 * useful information.
824 if ((!file->f_op->fallocate) ||
825 lo->lo_encrypt_key_size) {
826 q->limits.discard_granularity = 0;
827 q->limits.discard_alignment = 0;
828 blk_queue_max_discard_sectors(q, 0);
829 q->limits.discard_zeroes_data = 0;
830 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
831 return;
834 q->limits.discard_granularity = inode->i_sb->s_blocksize;
835 q->limits.discard_alignment = 0;
836 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
837 q->limits.discard_zeroes_data = 1;
838 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
841 static void loop_unprepare_queue(struct loop_device *lo)
843 kthread_flush_worker(&lo->worker);
844 kthread_stop(lo->worker_task);
847 static int loop_prepare_queue(struct loop_device *lo)
849 kthread_init_worker(&lo->worker);
850 lo->worker_task = kthread_run(kthread_worker_fn,
851 &lo->worker, "loop%d", lo->lo_number);
852 if (IS_ERR(lo->worker_task))
853 return -ENOMEM;
854 set_user_nice(lo->worker_task, MIN_NICE);
855 return 0;
858 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
859 struct block_device *bdev, unsigned int arg)
861 struct file *file, *f;
862 struct inode *inode;
863 struct address_space *mapping;
864 unsigned lo_blocksize;
865 int lo_flags = 0;
866 int error;
867 loff_t size;
869 /* This is safe, since we have a reference from open(). */
870 __module_get(THIS_MODULE);
872 error = -EBADF;
873 file = fget(arg);
874 if (!file)
875 goto out;
877 error = -EBUSY;
878 if (lo->lo_state != Lo_unbound)
879 goto out_putf;
881 /* Avoid recursion */
882 f = file;
883 while (is_loop_device(f)) {
884 struct loop_device *l;
886 if (f->f_mapping->host->i_bdev == bdev)
887 goto out_putf;
889 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
890 if (l->lo_state == Lo_unbound) {
891 error = -EINVAL;
892 goto out_putf;
894 f = l->lo_backing_file;
897 mapping = file->f_mapping;
898 inode = mapping->host;
900 error = -EINVAL;
901 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
902 goto out_putf;
904 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
905 !file->f_op->write_iter)
906 lo_flags |= LO_FLAGS_READ_ONLY;
908 lo_blocksize = S_ISBLK(inode->i_mode) ?
909 inode->i_bdev->bd_block_size : PAGE_SIZE;
911 error = -EFBIG;
912 size = get_loop_size(lo, file);
913 if ((loff_t)(sector_t)size != size)
914 goto out_putf;
915 error = loop_prepare_queue(lo);
916 if (error)
917 goto out_putf;
919 error = 0;
921 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
923 lo->use_dio = false;
924 lo->lo_blocksize = lo_blocksize;
925 lo->lo_device = bdev;
926 lo->lo_flags = lo_flags;
927 lo->lo_backing_file = file;
928 lo->transfer = NULL;
929 lo->ioctl = NULL;
930 lo->lo_sizelimit = 0;
931 lo->old_gfp_mask = mapping_gfp_mask(mapping);
932 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
934 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
935 blk_queue_write_cache(lo->lo_queue, true, false);
937 loop_update_dio(lo);
938 set_capacity(lo->lo_disk, size);
939 bd_set_size(bdev, size << 9);
940 loop_sysfs_init(lo);
941 /* let user-space know about the new size */
942 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
944 set_blocksize(bdev, lo_blocksize);
946 lo->lo_state = Lo_bound;
947 if (part_shift)
948 lo->lo_flags |= LO_FLAGS_PARTSCAN;
949 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
950 loop_reread_partitions(lo, bdev);
952 /* Grab the block_device to prevent its destruction after we
953 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
955 bdgrab(bdev);
956 return 0;
958 out_putf:
959 fput(file);
960 out:
961 /* This is safe: open() is still holding a reference. */
962 module_put(THIS_MODULE);
963 return error;
966 static int
967 loop_release_xfer(struct loop_device *lo)
969 int err = 0;
970 struct loop_func_table *xfer = lo->lo_encryption;
972 if (xfer) {
973 if (xfer->release)
974 err = xfer->release(lo);
975 lo->transfer = NULL;
976 lo->lo_encryption = NULL;
977 module_put(xfer->owner);
979 return err;
982 static int
983 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
984 const struct loop_info64 *i)
986 int err = 0;
988 if (xfer) {
989 struct module *owner = xfer->owner;
991 if (!try_module_get(owner))
992 return -EINVAL;
993 if (xfer->init)
994 err = xfer->init(lo, i);
995 if (err)
996 module_put(owner);
997 else
998 lo->lo_encryption = xfer;
1000 return err;
1003 static int loop_clr_fd(struct loop_device *lo)
1005 struct file *filp = lo->lo_backing_file;
1006 gfp_t gfp = lo->old_gfp_mask;
1007 struct block_device *bdev = lo->lo_device;
1009 if (lo->lo_state != Lo_bound)
1010 return -ENXIO;
1013 * If we've explicitly asked to tear down the loop device,
1014 * and it has an elevated reference count, set it for auto-teardown when
1015 * the last reference goes away. This stops $!~#$@ udev from
1016 * preventing teardown because it decided that it needs to run blkid on
1017 * the loopback device whenever they appear. xfstests is notorious for
1018 * failing tests because blkid via udev races with a losetup
1019 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1020 * command to fail with EBUSY.
1022 if (atomic_read(&lo->lo_refcnt) > 1) {
1023 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1024 mutex_unlock(&lo->lo_ctl_mutex);
1025 return 0;
1028 if (filp == NULL)
1029 return -EINVAL;
1031 /* freeze request queue during the transition */
1032 blk_mq_freeze_queue(lo->lo_queue);
1034 spin_lock_irq(&lo->lo_lock);
1035 lo->lo_state = Lo_rundown;
1036 lo->lo_backing_file = NULL;
1037 spin_unlock_irq(&lo->lo_lock);
1039 loop_release_xfer(lo);
1040 lo->transfer = NULL;
1041 lo->ioctl = NULL;
1042 lo->lo_device = NULL;
1043 lo->lo_encryption = NULL;
1044 lo->lo_offset = 0;
1045 lo->lo_sizelimit = 0;
1046 lo->lo_encrypt_key_size = 0;
1047 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1048 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1049 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1050 if (bdev) {
1051 bdput(bdev);
1052 invalidate_bdev(bdev);
1054 set_capacity(lo->lo_disk, 0);
1055 loop_sysfs_exit(lo);
1056 if (bdev) {
1057 bd_set_size(bdev, 0);
1058 /* let user-space know about this change */
1059 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1061 mapping_set_gfp_mask(filp->f_mapping, gfp);
1062 lo->lo_state = Lo_unbound;
1063 /* This is safe: open() is still holding a reference. */
1064 module_put(THIS_MODULE);
1065 blk_mq_unfreeze_queue(lo->lo_queue);
1067 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1068 loop_reread_partitions(lo, bdev);
1069 lo->lo_flags = 0;
1070 if (!part_shift)
1071 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1072 loop_unprepare_queue(lo);
1073 mutex_unlock(&lo->lo_ctl_mutex);
1075 * Need not hold lo_ctl_mutex to fput backing file.
1076 * Calling fput holding lo_ctl_mutex triggers a circular
1077 * lock dependency possibility warning as fput can take
1078 * bd_mutex which is usually taken before lo_ctl_mutex.
1080 fput(filp);
1081 return 0;
1084 static int
1085 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1087 int err;
1088 struct loop_func_table *xfer;
1089 kuid_t uid = current_uid();
1091 if (lo->lo_encrypt_key_size &&
1092 !uid_eq(lo->lo_key_owner, uid) &&
1093 !capable(CAP_SYS_ADMIN))
1094 return -EPERM;
1095 if (lo->lo_state != Lo_bound)
1096 return -ENXIO;
1097 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1098 return -EINVAL;
1100 err = loop_release_xfer(lo);
1101 if (err)
1102 return err;
1104 if (info->lo_encrypt_type) {
1105 unsigned int type = info->lo_encrypt_type;
1107 if (type >= MAX_LO_CRYPT)
1108 return -EINVAL;
1109 xfer = xfer_funcs[type];
1110 if (xfer == NULL)
1111 return -EINVAL;
1112 } else
1113 xfer = NULL;
1115 err = loop_init_xfer(lo, xfer, info);
1116 if (err)
1117 return err;
1119 if (lo->lo_offset != info->lo_offset ||
1120 lo->lo_sizelimit != info->lo_sizelimit)
1121 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
1122 return -EFBIG;
1124 loop_config_discard(lo);
1126 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1127 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1128 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1129 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1131 if (!xfer)
1132 xfer = &none_funcs;
1133 lo->transfer = xfer->transfer;
1134 lo->ioctl = xfer->ioctl;
1136 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1137 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1138 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1140 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1141 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1142 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1143 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1144 loop_reread_partitions(lo, lo->lo_device);
1147 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1148 lo->lo_init[0] = info->lo_init[0];
1149 lo->lo_init[1] = info->lo_init[1];
1150 if (info->lo_encrypt_key_size) {
1151 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1152 info->lo_encrypt_key_size);
1153 lo->lo_key_owner = uid;
1156 /* update dio if lo_offset or transfer is changed */
1157 __loop_update_dio(lo, lo->use_dio);
1159 return 0;
1162 static int
1163 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1165 struct file *file = lo->lo_backing_file;
1166 struct kstat stat;
1167 int error;
1169 if (lo->lo_state != Lo_bound)
1170 return -ENXIO;
1171 error = vfs_getattr(&file->f_path, &stat);
1172 if (error)
1173 return error;
1174 memset(info, 0, sizeof(*info));
1175 info->lo_number = lo->lo_number;
1176 info->lo_device = huge_encode_dev(stat.dev);
1177 info->lo_inode = stat.ino;
1178 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1179 info->lo_offset = lo->lo_offset;
1180 info->lo_sizelimit = lo->lo_sizelimit;
1181 info->lo_flags = lo->lo_flags;
1182 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1183 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1184 info->lo_encrypt_type =
1185 lo->lo_encryption ? lo->lo_encryption->number : 0;
1186 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1187 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1188 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1189 lo->lo_encrypt_key_size);
1191 return 0;
1194 static void
1195 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1197 memset(info64, 0, sizeof(*info64));
1198 info64->lo_number = info->lo_number;
1199 info64->lo_device = info->lo_device;
1200 info64->lo_inode = info->lo_inode;
1201 info64->lo_rdevice = info->lo_rdevice;
1202 info64->lo_offset = info->lo_offset;
1203 info64->lo_sizelimit = 0;
1204 info64->lo_encrypt_type = info->lo_encrypt_type;
1205 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1206 info64->lo_flags = info->lo_flags;
1207 info64->lo_init[0] = info->lo_init[0];
1208 info64->lo_init[1] = info->lo_init[1];
1209 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1210 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1211 else
1212 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1213 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1216 static int
1217 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1219 memset(info, 0, sizeof(*info));
1220 info->lo_number = info64->lo_number;
1221 info->lo_device = info64->lo_device;
1222 info->lo_inode = info64->lo_inode;
1223 info->lo_rdevice = info64->lo_rdevice;
1224 info->lo_offset = info64->lo_offset;
1225 info->lo_encrypt_type = info64->lo_encrypt_type;
1226 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1227 info->lo_flags = info64->lo_flags;
1228 info->lo_init[0] = info64->lo_init[0];
1229 info->lo_init[1] = info64->lo_init[1];
1230 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1231 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1232 else
1233 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1234 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1236 /* error in case values were truncated */
1237 if (info->lo_device != info64->lo_device ||
1238 info->lo_rdevice != info64->lo_rdevice ||
1239 info->lo_inode != info64->lo_inode ||
1240 info->lo_offset != info64->lo_offset)
1241 return -EOVERFLOW;
1243 return 0;
1246 static int
1247 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1249 struct loop_info info;
1250 struct loop_info64 info64;
1252 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1253 return -EFAULT;
1254 loop_info64_from_old(&info, &info64);
1255 return loop_set_status(lo, &info64);
1258 static int
1259 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1261 struct loop_info64 info64;
1263 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1264 return -EFAULT;
1265 return loop_set_status(lo, &info64);
1268 static int
1269 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1270 struct loop_info info;
1271 struct loop_info64 info64;
1272 int err = 0;
1274 if (!arg)
1275 err = -EINVAL;
1276 if (!err)
1277 err = loop_get_status(lo, &info64);
1278 if (!err)
1279 err = loop_info64_to_old(&info64, &info);
1280 if (!err && copy_to_user(arg, &info, sizeof(info)))
1281 err = -EFAULT;
1283 return err;
1286 static int
1287 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1288 struct loop_info64 info64;
1289 int err = 0;
1291 if (!arg)
1292 err = -EINVAL;
1293 if (!err)
1294 err = loop_get_status(lo, &info64);
1295 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1296 err = -EFAULT;
1298 return err;
1301 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1303 if (unlikely(lo->lo_state != Lo_bound))
1304 return -ENXIO;
1306 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1309 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1311 int error = -ENXIO;
1312 if (lo->lo_state != Lo_bound)
1313 goto out;
1315 __loop_update_dio(lo, !!arg);
1316 if (lo->use_dio == !!arg)
1317 return 0;
1318 error = -EINVAL;
1319 out:
1320 return error;
1323 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1324 unsigned int cmd, unsigned long arg)
1326 struct loop_device *lo = bdev->bd_disk->private_data;
1327 int err;
1329 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1330 switch (cmd) {
1331 case LOOP_SET_FD:
1332 err = loop_set_fd(lo, mode, bdev, arg);
1333 break;
1334 case LOOP_CHANGE_FD:
1335 err = loop_change_fd(lo, bdev, arg);
1336 break;
1337 case LOOP_CLR_FD:
1338 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1339 err = loop_clr_fd(lo);
1340 if (!err)
1341 goto out_unlocked;
1342 break;
1343 case LOOP_SET_STATUS:
1344 err = -EPERM;
1345 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1346 err = loop_set_status_old(lo,
1347 (struct loop_info __user *)arg);
1348 break;
1349 case LOOP_GET_STATUS:
1350 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1351 break;
1352 case LOOP_SET_STATUS64:
1353 err = -EPERM;
1354 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1355 err = loop_set_status64(lo,
1356 (struct loop_info64 __user *) arg);
1357 break;
1358 case LOOP_GET_STATUS64:
1359 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1360 break;
1361 case LOOP_SET_CAPACITY:
1362 err = -EPERM;
1363 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1364 err = loop_set_capacity(lo, bdev);
1365 break;
1366 case LOOP_SET_DIRECT_IO:
1367 err = -EPERM;
1368 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1369 err = loop_set_dio(lo, arg);
1370 break;
1371 default:
1372 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1374 mutex_unlock(&lo->lo_ctl_mutex);
1376 out_unlocked:
1377 return err;
1380 #ifdef CONFIG_COMPAT
1381 struct compat_loop_info {
1382 compat_int_t lo_number; /* ioctl r/o */
1383 compat_dev_t lo_device; /* ioctl r/o */
1384 compat_ulong_t lo_inode; /* ioctl r/o */
1385 compat_dev_t lo_rdevice; /* ioctl r/o */
1386 compat_int_t lo_offset;
1387 compat_int_t lo_encrypt_type;
1388 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1389 compat_int_t lo_flags; /* ioctl r/o */
1390 char lo_name[LO_NAME_SIZE];
1391 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1392 compat_ulong_t lo_init[2];
1393 char reserved[4];
1397 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1398 * - noinlined to reduce stack space usage in main part of driver
1400 static noinline int
1401 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1402 struct loop_info64 *info64)
1404 struct compat_loop_info info;
1406 if (copy_from_user(&info, arg, sizeof(info)))
1407 return -EFAULT;
1409 memset(info64, 0, sizeof(*info64));
1410 info64->lo_number = info.lo_number;
1411 info64->lo_device = info.lo_device;
1412 info64->lo_inode = info.lo_inode;
1413 info64->lo_rdevice = info.lo_rdevice;
1414 info64->lo_offset = info.lo_offset;
1415 info64->lo_sizelimit = 0;
1416 info64->lo_encrypt_type = info.lo_encrypt_type;
1417 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1418 info64->lo_flags = info.lo_flags;
1419 info64->lo_init[0] = info.lo_init[0];
1420 info64->lo_init[1] = info.lo_init[1];
1421 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1422 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1423 else
1424 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1425 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1426 return 0;
1430 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1431 * - noinlined to reduce stack space usage in main part of driver
1433 static noinline int
1434 loop_info64_to_compat(const struct loop_info64 *info64,
1435 struct compat_loop_info __user *arg)
1437 struct compat_loop_info info;
1439 memset(&info, 0, sizeof(info));
1440 info.lo_number = info64->lo_number;
1441 info.lo_device = info64->lo_device;
1442 info.lo_inode = info64->lo_inode;
1443 info.lo_rdevice = info64->lo_rdevice;
1444 info.lo_offset = info64->lo_offset;
1445 info.lo_encrypt_type = info64->lo_encrypt_type;
1446 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1447 info.lo_flags = info64->lo_flags;
1448 info.lo_init[0] = info64->lo_init[0];
1449 info.lo_init[1] = info64->lo_init[1];
1450 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1451 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1452 else
1453 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1454 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1456 /* error in case values were truncated */
1457 if (info.lo_device != info64->lo_device ||
1458 info.lo_rdevice != info64->lo_rdevice ||
1459 info.lo_inode != info64->lo_inode ||
1460 info.lo_offset != info64->lo_offset ||
1461 info.lo_init[0] != info64->lo_init[0] ||
1462 info.lo_init[1] != info64->lo_init[1])
1463 return -EOVERFLOW;
1465 if (copy_to_user(arg, &info, sizeof(info)))
1466 return -EFAULT;
1467 return 0;
1470 static int
1471 loop_set_status_compat(struct loop_device *lo,
1472 const struct compat_loop_info __user *arg)
1474 struct loop_info64 info64;
1475 int ret;
1477 ret = loop_info64_from_compat(arg, &info64);
1478 if (ret < 0)
1479 return ret;
1480 return loop_set_status(lo, &info64);
1483 static int
1484 loop_get_status_compat(struct loop_device *lo,
1485 struct compat_loop_info __user *arg)
1487 struct loop_info64 info64;
1488 int err = 0;
1490 if (!arg)
1491 err = -EINVAL;
1492 if (!err)
1493 err = loop_get_status(lo, &info64);
1494 if (!err)
1495 err = loop_info64_to_compat(&info64, arg);
1496 return err;
1499 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1500 unsigned int cmd, unsigned long arg)
1502 struct loop_device *lo = bdev->bd_disk->private_data;
1503 int err;
1505 switch(cmd) {
1506 case LOOP_SET_STATUS:
1507 mutex_lock(&lo->lo_ctl_mutex);
1508 err = loop_set_status_compat(
1509 lo, (const struct compat_loop_info __user *) arg);
1510 mutex_unlock(&lo->lo_ctl_mutex);
1511 break;
1512 case LOOP_GET_STATUS:
1513 mutex_lock(&lo->lo_ctl_mutex);
1514 err = loop_get_status_compat(
1515 lo, (struct compat_loop_info __user *) arg);
1516 mutex_unlock(&lo->lo_ctl_mutex);
1517 break;
1518 case LOOP_SET_CAPACITY:
1519 case LOOP_CLR_FD:
1520 case LOOP_GET_STATUS64:
1521 case LOOP_SET_STATUS64:
1522 arg = (unsigned long) compat_ptr(arg);
1523 case LOOP_SET_FD:
1524 case LOOP_CHANGE_FD:
1525 err = lo_ioctl(bdev, mode, cmd, arg);
1526 break;
1527 default:
1528 err = -ENOIOCTLCMD;
1529 break;
1531 return err;
1533 #endif
1535 static int lo_open(struct block_device *bdev, fmode_t mode)
1537 struct loop_device *lo;
1538 int err = 0;
1540 mutex_lock(&loop_index_mutex);
1541 lo = bdev->bd_disk->private_data;
1542 if (!lo) {
1543 err = -ENXIO;
1544 goto out;
1547 atomic_inc(&lo->lo_refcnt);
1548 out:
1549 mutex_unlock(&loop_index_mutex);
1550 return err;
1553 static void lo_release(struct gendisk *disk, fmode_t mode)
1555 struct loop_device *lo = disk->private_data;
1556 int err;
1558 if (atomic_dec_return(&lo->lo_refcnt))
1559 return;
1561 mutex_lock(&lo->lo_ctl_mutex);
1562 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1564 * In autoclear mode, stop the loop thread
1565 * and remove configuration after last close.
1567 err = loop_clr_fd(lo);
1568 if (!err)
1569 return;
1570 } else {
1572 * Otherwise keep thread (if running) and config,
1573 * but flush possible ongoing bios in thread.
1575 loop_flush(lo);
1578 mutex_unlock(&lo->lo_ctl_mutex);
1581 static const struct block_device_operations lo_fops = {
1582 .owner = THIS_MODULE,
1583 .open = lo_open,
1584 .release = lo_release,
1585 .ioctl = lo_ioctl,
1586 #ifdef CONFIG_COMPAT
1587 .compat_ioctl = lo_compat_ioctl,
1588 #endif
1592 * And now the modules code and kernel interface.
1594 static int max_loop;
1595 module_param(max_loop, int, S_IRUGO);
1596 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1597 module_param(max_part, int, S_IRUGO);
1598 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1599 MODULE_LICENSE("GPL");
1600 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1602 int loop_register_transfer(struct loop_func_table *funcs)
1604 unsigned int n = funcs->number;
1606 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1607 return -EINVAL;
1608 xfer_funcs[n] = funcs;
1609 return 0;
1612 static int unregister_transfer_cb(int id, void *ptr, void *data)
1614 struct loop_device *lo = ptr;
1615 struct loop_func_table *xfer = data;
1617 mutex_lock(&lo->lo_ctl_mutex);
1618 if (lo->lo_encryption == xfer)
1619 loop_release_xfer(lo);
1620 mutex_unlock(&lo->lo_ctl_mutex);
1621 return 0;
1624 int loop_unregister_transfer(int number)
1626 unsigned int n = number;
1627 struct loop_func_table *xfer;
1629 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1630 return -EINVAL;
1632 xfer_funcs[n] = NULL;
1633 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1634 return 0;
1637 EXPORT_SYMBOL(loop_register_transfer);
1638 EXPORT_SYMBOL(loop_unregister_transfer);
1640 static int loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1641 const struct blk_mq_queue_data *bd)
1643 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1644 struct loop_device *lo = cmd->rq->q->queuedata;
1646 blk_mq_start_request(bd->rq);
1648 if (lo->lo_state != Lo_bound)
1649 return -EIO;
1651 switch (req_op(cmd->rq)) {
1652 case REQ_OP_FLUSH:
1653 case REQ_OP_DISCARD:
1654 cmd->use_aio = false;
1655 break;
1656 default:
1657 cmd->use_aio = lo->use_dio;
1658 break;
1661 kthread_queue_work(&lo->worker, &cmd->work);
1663 return BLK_MQ_RQ_QUEUE_OK;
1666 static void loop_handle_cmd(struct loop_cmd *cmd)
1668 const bool write = op_is_write(req_op(cmd->rq));
1669 struct loop_device *lo = cmd->rq->q->queuedata;
1670 int ret = 0;
1672 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1673 ret = -EIO;
1674 goto failed;
1677 ret = do_req_filebacked(lo, cmd->rq);
1678 failed:
1679 /* complete non-aio request */
1680 if (!cmd->use_aio || ret)
1681 blk_mq_complete_request(cmd->rq, ret ? -EIO : 0);
1684 static void loop_queue_work(struct kthread_work *work)
1686 struct loop_cmd *cmd =
1687 container_of(work, struct loop_cmd, work);
1689 loop_handle_cmd(cmd);
1692 static int loop_init_request(void *data, struct request *rq,
1693 unsigned int hctx_idx, unsigned int request_idx,
1694 unsigned int numa_node)
1696 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1698 cmd->rq = rq;
1699 kthread_init_work(&cmd->work, loop_queue_work);
1701 return 0;
1704 static struct blk_mq_ops loop_mq_ops = {
1705 .queue_rq = loop_queue_rq,
1706 .init_request = loop_init_request,
1709 static int loop_add(struct loop_device **l, int i)
1711 struct loop_device *lo;
1712 struct gendisk *disk;
1713 int err;
1715 err = -ENOMEM;
1716 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1717 if (!lo)
1718 goto out;
1720 lo->lo_state = Lo_unbound;
1722 /* allocate id, if @id >= 0, we're requesting that specific id */
1723 if (i >= 0) {
1724 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1725 if (err == -ENOSPC)
1726 err = -EEXIST;
1727 } else {
1728 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1730 if (err < 0)
1731 goto out_free_dev;
1732 i = err;
1734 err = -ENOMEM;
1735 lo->tag_set.ops = &loop_mq_ops;
1736 lo->tag_set.nr_hw_queues = 1;
1737 lo->tag_set.queue_depth = 128;
1738 lo->tag_set.numa_node = NUMA_NO_NODE;
1739 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1740 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1741 lo->tag_set.driver_data = lo;
1743 err = blk_mq_alloc_tag_set(&lo->tag_set);
1744 if (err)
1745 goto out_free_idr;
1747 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1748 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1749 err = PTR_ERR(lo->lo_queue);
1750 goto out_cleanup_tags;
1752 lo->lo_queue->queuedata = lo;
1755 * It doesn't make sense to enable merge because the I/O
1756 * submitted to backing file is handled page by page.
1758 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1760 err = -ENOMEM;
1761 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1762 if (!disk)
1763 goto out_free_queue;
1766 * Disable partition scanning by default. The in-kernel partition
1767 * scanning can be requested individually per-device during its
1768 * setup. Userspace can always add and remove partitions from all
1769 * devices. The needed partition minors are allocated from the
1770 * extended minor space, the main loop device numbers will continue
1771 * to match the loop minors, regardless of the number of partitions
1772 * used.
1774 * If max_part is given, partition scanning is globally enabled for
1775 * all loop devices. The minors for the main loop devices will be
1776 * multiples of max_part.
1778 * Note: Global-for-all-devices, set-only-at-init, read-only module
1779 * parameteters like 'max_loop' and 'max_part' make things needlessly
1780 * complicated, are too static, inflexible and may surprise
1781 * userspace tools. Parameters like this in general should be avoided.
1783 if (!part_shift)
1784 disk->flags |= GENHD_FL_NO_PART_SCAN;
1785 disk->flags |= GENHD_FL_EXT_DEVT;
1786 mutex_init(&lo->lo_ctl_mutex);
1787 atomic_set(&lo->lo_refcnt, 0);
1788 lo->lo_number = i;
1789 spin_lock_init(&lo->lo_lock);
1790 disk->major = LOOP_MAJOR;
1791 disk->first_minor = i << part_shift;
1792 disk->fops = &lo_fops;
1793 disk->private_data = lo;
1794 disk->queue = lo->lo_queue;
1795 sprintf(disk->disk_name, "loop%d", i);
1796 add_disk(disk);
1797 *l = lo;
1798 return lo->lo_number;
1800 out_free_queue:
1801 blk_cleanup_queue(lo->lo_queue);
1802 out_cleanup_tags:
1803 blk_mq_free_tag_set(&lo->tag_set);
1804 out_free_idr:
1805 idr_remove(&loop_index_idr, i);
1806 out_free_dev:
1807 kfree(lo);
1808 out:
1809 return err;
1812 static void loop_remove(struct loop_device *lo)
1814 blk_cleanup_queue(lo->lo_queue);
1815 del_gendisk(lo->lo_disk);
1816 blk_mq_free_tag_set(&lo->tag_set);
1817 put_disk(lo->lo_disk);
1818 kfree(lo);
1821 static int find_free_cb(int id, void *ptr, void *data)
1823 struct loop_device *lo = ptr;
1824 struct loop_device **l = data;
1826 if (lo->lo_state == Lo_unbound) {
1827 *l = lo;
1828 return 1;
1830 return 0;
1833 static int loop_lookup(struct loop_device **l, int i)
1835 struct loop_device *lo;
1836 int ret = -ENODEV;
1838 if (i < 0) {
1839 int err;
1841 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1842 if (err == 1) {
1843 *l = lo;
1844 ret = lo->lo_number;
1846 goto out;
1849 /* lookup and return a specific i */
1850 lo = idr_find(&loop_index_idr, i);
1851 if (lo) {
1852 *l = lo;
1853 ret = lo->lo_number;
1855 out:
1856 return ret;
1859 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1861 struct loop_device *lo;
1862 struct kobject *kobj;
1863 int err;
1865 mutex_lock(&loop_index_mutex);
1866 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1867 if (err < 0)
1868 err = loop_add(&lo, MINOR(dev) >> part_shift);
1869 if (err < 0)
1870 kobj = NULL;
1871 else
1872 kobj = get_disk(lo->lo_disk);
1873 mutex_unlock(&loop_index_mutex);
1875 *part = 0;
1876 return kobj;
1879 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1880 unsigned long parm)
1882 struct loop_device *lo;
1883 int ret = -ENOSYS;
1885 mutex_lock(&loop_index_mutex);
1886 switch (cmd) {
1887 case LOOP_CTL_ADD:
1888 ret = loop_lookup(&lo, parm);
1889 if (ret >= 0) {
1890 ret = -EEXIST;
1891 break;
1893 ret = loop_add(&lo, parm);
1894 break;
1895 case LOOP_CTL_REMOVE:
1896 ret = loop_lookup(&lo, parm);
1897 if (ret < 0)
1898 break;
1899 mutex_lock(&lo->lo_ctl_mutex);
1900 if (lo->lo_state != Lo_unbound) {
1901 ret = -EBUSY;
1902 mutex_unlock(&lo->lo_ctl_mutex);
1903 break;
1905 if (atomic_read(&lo->lo_refcnt) > 0) {
1906 ret = -EBUSY;
1907 mutex_unlock(&lo->lo_ctl_mutex);
1908 break;
1910 lo->lo_disk->private_data = NULL;
1911 mutex_unlock(&lo->lo_ctl_mutex);
1912 idr_remove(&loop_index_idr, lo->lo_number);
1913 loop_remove(lo);
1914 break;
1915 case LOOP_CTL_GET_FREE:
1916 ret = loop_lookup(&lo, -1);
1917 if (ret >= 0)
1918 break;
1919 ret = loop_add(&lo, -1);
1921 mutex_unlock(&loop_index_mutex);
1923 return ret;
1926 static const struct file_operations loop_ctl_fops = {
1927 .open = nonseekable_open,
1928 .unlocked_ioctl = loop_control_ioctl,
1929 .compat_ioctl = loop_control_ioctl,
1930 .owner = THIS_MODULE,
1931 .llseek = noop_llseek,
1934 static struct miscdevice loop_misc = {
1935 .minor = LOOP_CTRL_MINOR,
1936 .name = "loop-control",
1937 .fops = &loop_ctl_fops,
1940 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1941 MODULE_ALIAS("devname:loop-control");
1943 static int __init loop_init(void)
1945 int i, nr;
1946 unsigned long range;
1947 struct loop_device *lo;
1948 int err;
1950 err = misc_register(&loop_misc);
1951 if (err < 0)
1952 return err;
1954 part_shift = 0;
1955 if (max_part > 0) {
1956 part_shift = fls(max_part);
1959 * Adjust max_part according to part_shift as it is exported
1960 * to user space so that user can decide correct minor number
1961 * if [s]he want to create more devices.
1963 * Note that -1 is required because partition 0 is reserved
1964 * for the whole disk.
1966 max_part = (1UL << part_shift) - 1;
1969 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1970 err = -EINVAL;
1971 goto misc_out;
1974 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1975 err = -EINVAL;
1976 goto misc_out;
1980 * If max_loop is specified, create that many devices upfront.
1981 * This also becomes a hard limit. If max_loop is not specified,
1982 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1983 * init time. Loop devices can be requested on-demand with the
1984 * /dev/loop-control interface, or be instantiated by accessing
1985 * a 'dead' device node.
1987 if (max_loop) {
1988 nr = max_loop;
1989 range = max_loop << part_shift;
1990 } else {
1991 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1992 range = 1UL << MINORBITS;
1995 if (register_blkdev(LOOP_MAJOR, "loop")) {
1996 err = -EIO;
1997 goto misc_out;
2000 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2001 THIS_MODULE, loop_probe, NULL, NULL);
2003 /* pre-create number of devices given by config or max_loop */
2004 mutex_lock(&loop_index_mutex);
2005 for (i = 0; i < nr; i++)
2006 loop_add(&lo, i);
2007 mutex_unlock(&loop_index_mutex);
2009 printk(KERN_INFO "loop: module loaded\n");
2010 return 0;
2012 misc_out:
2013 misc_deregister(&loop_misc);
2014 return err;
2017 static int loop_exit_cb(int id, void *ptr, void *data)
2019 struct loop_device *lo = ptr;
2021 loop_remove(lo);
2022 return 0;
2025 static void __exit loop_exit(void)
2027 unsigned long range;
2029 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2031 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2032 idr_destroy(&loop_index_idr);
2034 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2035 unregister_blkdev(LOOP_MAJOR, "loop");
2037 misc_deregister(&loop_misc);
2040 module_init(loop_init);
2041 module_exit(loop_exit);
2043 #ifndef MODULE
2044 static int __init max_loop_setup(char *str)
2046 max_loop = simple_strtol(str, NULL, 0);
2047 return 1;
2050 __setup("max_loop=", max_loop_setup);
2051 #endif