projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
drivers: power: report battery voltage in AOSP compatible format
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / block / loop.c
... / ...
CommitLineData
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/loop.h>
67#include <linux/compat.h>
68#include <linux/suspend.h>
69#include <linux/freezer.h>
70#include <linux/mutex.h>
71#include <linux/writeback.h>
72#include <linux/completion.h>
73#include <linux/highmem.h>
74#include <linux/kthread.h>
75#include <linux/splice.h>
76#include <linux/sysfs.h>
77#include <linux/miscdevice.h>
78#include <linux/falloc.h>
79
80#include <asm/uaccess.h>
81
82static DEFINE_IDR(loop_index_idr);
83static DEFINE_MUTEX(loop_index_mutex);
84
85static int max_part;
86static int part_shift;
87
88/*
89 * Transfer functions
90 */
91static int transfer_none(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
95{
96 char *raw_buf = kmap_atomic(raw_page) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page) + loop_off;
98
99 if (cmd == READ)
100 memcpy(loop_buf, raw_buf, size);
101 else
102 memcpy(raw_buf, loop_buf, size);
103
104 kunmap_atomic(loop_buf);
105 kunmap_atomic(raw_buf);
106 cond_resched();
107 return 0;
108}
109
110static int transfer_xor(struct loop_device *lo, int cmd,
111 struct page *raw_page, unsigned raw_off,
112 struct page *loop_page, unsigned loop_off,
113 int size, sector_t real_block)
114{
115 char *raw_buf = kmap_atomic(raw_page) + raw_off;
116 char *loop_buf = kmap_atomic(loop_page) + loop_off;
117 char *in, *out, *key;
118 int i, keysize;
119
120 if (cmd == READ) {
121 in = raw_buf;
122 out = loop_buf;
123 } else {
124 in = loop_buf;
125 out = raw_buf;
126 }
127
128 key = lo->lo_encrypt_key;
129 keysize = lo->lo_encrypt_key_size;
130 for (i = 0; i < size; i++)
131 *out++ = *in++ ^ key[(i & 511) % keysize];
132
133 kunmap_atomic(loop_buf);
134 kunmap_atomic(raw_buf);
135 cond_resched();
136 return 0;
137}
138
139static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140{
141 if (unlikely(info->lo_encrypt_key_size <= 0))
142 return -EINVAL;
143 return 0;
144}
145
146static struct loop_func_table none_funcs = {
147 .number = LO_CRYPT_NONE,
148 .transfer = transfer_none,
149};
150
151static struct loop_func_table xor_funcs = {
152 .number = LO_CRYPT_XOR,
153 .transfer = transfer_xor,
154 .init = xor_init
155};
156
157/* xfer_funcs[0] is special - its release function is never called */
158static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
159 &none_funcs,
160 &xor_funcs
161};
162
163static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
164{
165 loff_t loopsize;
166
167 /* Compute loopsize in bytes */
168 loopsize = i_size_read(file->f_mapping->host);
169 if (offset > 0)
170 loopsize -= offset;
171 /* offset is beyond i_size, weird but possible */
172 if (loopsize < 0)
173 return 0;
174
175 if (sizelimit > 0 && sizelimit < loopsize)
176 loopsize = sizelimit;
177 /*
178 * Unfortunately, if we want to do I/O on the device,
179 * the number of 512-byte sectors has to fit into a sector_t.
180 */
181 return loopsize >> 9;
182}
183
184static loff_t get_loop_size(struct loop_device *lo, struct file *file)
185{
186 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
187}
188
189static int
190figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
191{
192 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
193 sector_t x = (sector_t)size;
194 struct block_device *bdev = lo->lo_device;
195
196 if (unlikely((loff_t)x != size))
197 return -EFBIG;
198 if (lo->lo_offset != offset)
199 lo->lo_offset = offset;
200 if (lo->lo_sizelimit != sizelimit)
201 lo->lo_sizelimit = sizelimit;
202 set_capacity(lo->lo_disk, x);
203 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
204 /* let user-space know about the new size */
205 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
206 return 0;
207}
208
209static inline int
210lo_do_transfer(struct loop_device *lo, int cmd,
211 struct page *rpage, unsigned roffs,
212 struct page *lpage, unsigned loffs,
213 int size, sector_t rblock)
214{
215 if (unlikely(!lo->transfer))
216 return 0;
217
218 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
219}
220
221/**
222 * __do_lo_send_write - helper for writing data to a loop device
223 *
224 * This helper just factors out common code between do_lo_send_direct_write()
225 * and do_lo_send_write().
226 */
227static int __do_lo_send_write(struct file *file,
228 u8 *buf, const int len, loff_t pos)
229{
230 ssize_t bw;
231 mm_segment_t old_fs = get_fs();
232
233 file_start_write(file);
234 set_fs(get_ds());
235 bw = file->f_op->write(file, buf, len, &pos);
236 set_fs(old_fs);
237 file_end_write(file);
238 if (likely(bw == len))
239 return 0;
240 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
241 (unsigned long long)pos, len);
242 if (bw >= 0)
243 bw = -EIO;
244 return bw;
245}
246
247/**
248 * do_lo_send_direct_write - helper for writing data to a loop device
249 *
250 * This is the fast, non-transforming version that does not need double
251 * buffering.
252 */
253static int do_lo_send_direct_write(struct loop_device *lo,
254 struct bio_vec *bvec, loff_t pos, struct page *page)
255{
256 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
257 kmap(bvec->bv_page) + bvec->bv_offset,
258 bvec->bv_len, pos);
259 kunmap(bvec->bv_page);
260 cond_resched();
261 return bw;
262}
263
264/**
265 * do_lo_send_write - helper for writing data to a loop device
266 *
267 * This is the slow, transforming version that needs to double buffer the
268 * data as it cannot do the transformations in place without having direct
269 * access to the destination pages of the backing file.
270 */
271static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
272 loff_t pos, struct page *page)
273{
274 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
275 bvec->bv_offset, bvec->bv_len, pos >> 9);
276 if (likely(!ret))
277 return __do_lo_send_write(lo->lo_backing_file,
278 page_address(page), bvec->bv_len,
279 pos);
280 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
281 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
282 if (ret > 0)
283 ret = -EIO;
284 return ret;
285}
286
287static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
288{
289 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
290 struct page *page);
291 struct bio_vec *bvec;
292 struct page *page = NULL;
293 int i, ret = 0;
294
295 if (lo->transfer != transfer_none) {
296 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
297 if (unlikely(!page))
298 goto fail;
299 kmap(page);
300 do_lo_send = do_lo_send_write;
301 } else {
302 do_lo_send = do_lo_send_direct_write;
303 }
304
305 bio_for_each_segment(bvec, bio, i) {
306 ret = do_lo_send(lo, bvec, pos, page);
307 if (ret < 0)
308 break;
309 pos += bvec->bv_len;
310 }
311 if (page) {
312 kunmap(page);
313 __free_page(page);
314 }
315out:
316 return ret;
317fail:
318 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
319 ret = -ENOMEM;
320 goto out;
321}
322
323struct lo_read_data {
324 struct loop_device *lo;
325 struct page *page;
326 unsigned offset;
327 int bsize;
328};
329
330static int
331lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
332 struct splice_desc *sd)
333{
334 struct lo_read_data *p = sd->u.data;
335 struct loop_device *lo = p->lo;
336 struct page *page = buf->page;
337 sector_t IV;
338 int size;
339
340 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
341 (buf->offset >> 9);
342 size = sd->len;
343 if (size > p->bsize)
344 size = p->bsize;
345
346 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
347 printk(KERN_ERR "loop: transfer error block %ld\n",
348 page->index);
349 size = -EINVAL;
350 }
351
352 flush_dcache_page(p->page);
353
354 if (size > 0)
355 p->offset += size;
356
357 return size;
358}
359
360static int
361lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
362{
363 return __splice_from_pipe(pipe, sd, lo_splice_actor);
364}
365
366static ssize_t
367do_lo_receive(struct loop_device *lo,
368 struct bio_vec *bvec, int bsize, loff_t pos)
369{
370 struct lo_read_data cookie;
371 struct splice_desc sd;
372 struct file *file;
373 ssize_t retval;
374
375 cookie.lo = lo;
376 cookie.page = bvec->bv_page;
377 cookie.offset = bvec->bv_offset;
378 cookie.bsize = bsize;
379
380 sd.len = 0;
381 sd.total_len = bvec->bv_len;
382 sd.flags = 0;
383 sd.pos = pos;
384 sd.u.data = &cookie;
385
386 file = lo->lo_backing_file;
387 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
388
389 return retval;
390}
391
392static int
393lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
394{
395 struct bio_vec *bvec;
396 ssize_t s;
397 int i;
398
399 bio_for_each_segment(bvec, bio, i) {
400 s = do_lo_receive(lo, bvec, bsize, pos);
401 if (s < 0)
402 return s;
403
404 if (s != bvec->bv_len) {
405 zero_fill_bio(bio);
406 break;
407 }
408 pos += bvec->bv_len;
409 }
410 return 0;
411}
412
413static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
414{
415 loff_t pos;
416 int ret;
417
418 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
419
420 if (bio_rw(bio) == WRITE) {
421 struct file *file = lo->lo_backing_file;
422
423 if (bio->bi_rw & REQ_FLUSH) {
424 ret = vfs_fsync(file, 0);
425 if (unlikely(ret && ret != -EINVAL)) {
426 ret = -EIO;
427 goto out;
428 }
429 }
430
431 /*
432 * We use punch hole to reclaim the free space used by the
433 * image a.k.a. discard. However we do not support discard if
434 * encryption is enabled, because it may give an attacker
435 * useful information.
436 */
437 if (bio->bi_rw & REQ_DISCARD) {
438 struct file *file = lo->lo_backing_file;
439 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
440
441 if ((!file->f_op->fallocate) ||
442 lo->lo_encrypt_key_size) {
443 ret = -EOPNOTSUPP;
444 goto out;
445 }
446 ret = file->f_op->fallocate(file, mode, pos,
447 bio->bi_size);
448 if (unlikely(ret && ret != -EINVAL &&
449 ret != -EOPNOTSUPP))
450 ret = -EIO;
451 goto out;
452 }
453
454 ret = lo_send(lo, bio, pos);
455
456 if ((bio->bi_rw & REQ_FUA) && !ret) {
457 ret = vfs_fsync(file, 0);
458 if (unlikely(ret && ret != -EINVAL))
459 ret = -EIO;
460 }
461 } else
462 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
463
464out:
465 return ret;
466}
467
468/*
469 * Add bio to back of pending list
470 */
471static void loop_add_bio(struct loop_device *lo, struct bio *bio)
472{
473 lo->lo_bio_count++;
474 bio_list_add(&lo->lo_bio_list, bio);
475}
476
477/*
478 * Grab first pending buffer
479 */
480static struct bio *loop_get_bio(struct loop_device *lo)
481{
482 lo->lo_bio_count--;
483 return bio_list_pop(&lo->lo_bio_list);
484}
485
486static void loop_make_request(struct request_queue *q, struct bio *old_bio)
487{
488 struct loop_device *lo = q->queuedata;
489 int rw = bio_rw(old_bio);
490
491 if (rw == READA)
492 rw = READ;
493
494 BUG_ON(!lo || (rw != READ && rw != WRITE));
495
496 spin_lock_irq(&lo->lo_lock);
497 if (lo->lo_state != Lo_bound)
498 goto out;
499 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
500 goto out;
501 if (lo->lo_bio_count >= q->nr_congestion_on)
502 wait_event_lock_irq(lo->lo_req_wait,
503 lo->lo_bio_count < q->nr_congestion_off,
504 lo->lo_lock);
505 loop_add_bio(lo, old_bio);
506 wake_up(&lo->lo_event);
507 spin_unlock_irq(&lo->lo_lock);
508 return;
509
510out:
511 spin_unlock_irq(&lo->lo_lock);
512 bio_io_error(old_bio);
513}
514
515struct switch_request {
516 struct file *file;
517 struct completion wait;
518};
519
520static void do_loop_switch(struct loop_device *, struct switch_request *);
521
522static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
523{
524 if (unlikely(!bio->bi_bdev)) {
525 do_loop_switch(lo, bio->bi_private);
526 bio_put(bio);
527 } else {
528 int ret = do_bio_filebacked(lo, bio);
529 bio_endio(bio, ret);
530 }
531}
532
533/*
534 * worker thread that handles reads/writes to file backed loop devices,
535 * to avoid blocking in our make_request_fn. it also does loop decrypting
536 * on reads for block backed loop, as that is too heavy to do from
537 * b_end_io context where irqs may be disabled.
538 *
539 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
540 * calling kthread_stop(). Therefore once kthread_should_stop() is
541 * true, make_request will not place any more requests. Therefore
542 * once kthread_should_stop() is true and lo_bio is NULL, we are
543 * done with the loop.
544 */
545static int loop_thread(void *data)
546{
547 struct loop_device *lo = data;
548 struct bio *bio;
549
550 set_user_nice(current, -20);
551
552 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
553
554 wait_event_interruptible(lo->lo_event,
555 !bio_list_empty(&lo->lo_bio_list) ||
556 kthread_should_stop());
557
558 if (bio_list_empty(&lo->lo_bio_list))
559 continue;
560 spin_lock_irq(&lo->lo_lock);
561 bio = loop_get_bio(lo);
562 if (lo->lo_bio_count < lo->lo_queue->nr_congestion_off)
563 wake_up(&lo->lo_req_wait);
564 spin_unlock_irq(&lo->lo_lock);
565
566 BUG_ON(!bio);
567 loop_handle_bio(lo, bio);
568 }
569
570 return 0;
571}
572
573/*
574 * loop_switch performs the hard work of switching a backing store.
575 * First it needs to flush existing IO, it does this by sending a magic
576 * BIO down the pipe. The completion of this BIO does the actual switch.
577 */
578static int loop_switch(struct loop_device *lo, struct file *file)
579{
580 struct switch_request w;
581 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
582 if (!bio)
583 return -ENOMEM;
584 init_completion(&w.wait);
585 w.file = file;
586 bio->bi_private = &w;
587 bio->bi_bdev = NULL;
588 loop_make_request(lo->lo_queue, bio);
589 wait_for_completion(&w.wait);
590 return 0;
591}
592
593/*
594 * Helper to flush the IOs in loop, but keeping loop thread running
595 */
596static int loop_flush(struct loop_device *lo)
597{
598 /* loop not yet configured, no running thread, nothing to flush */
599 if (!lo->lo_thread)
600 return 0;
601
602 return loop_switch(lo, NULL);
603}
604
605/*
606 * Do the actual switch; called from the BIO completion routine
607 */
608static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
609{
610 struct file *file = p->file;
611 struct file *old_file = lo->lo_backing_file;
612 struct address_space *mapping;
613
614 /* if no new file, only flush of queued bios requested */
615 if (!file)
616 goto out;
617
618 mapping = file->f_mapping;
619 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
620 lo->lo_backing_file = file;
621 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
622 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
623 lo->old_gfp_mask = mapping_gfp_mask(mapping);
624 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
625out:
626 complete(&p->wait);
627}
628
629
630/*
631 * loop_change_fd switched the backing store of a loopback device to
632 * a new file. This is useful for operating system installers to free up
633 * the original file and in High Availability environments to switch to
634 * an alternative location for the content in case of server meltdown.
635 * This can only work if the loop device is used read-only, and if the
636 * new backing store is the same size and type as the old backing store.
637 */
638static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
639 unsigned int arg)
640{
641 struct file *file, *old_file;
642 struct inode *inode;
643 int error;
644
645 error = -ENXIO;
646 if (lo->lo_state != Lo_bound)
647 goto out;
648
649 /* the loop device has to be read-only */
650 error = -EINVAL;
651 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
652 goto out;
653
654 error = -EBADF;
655 file = fget(arg);
656 if (!file)
657 goto out;
658
659 inode = file->f_mapping->host;
660 old_file = lo->lo_backing_file;
661
662 error = -EINVAL;
663
664 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
665 goto out_putf;
666
667 /* size of the new backing store needs to be the same */
668 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
669 goto out_putf;
670
671 /* and ... switch */
672 error = loop_switch(lo, file);
673 if (error)
674 goto out_putf;
675
676 fput(old_file);
677 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
678 ioctl_by_bdev(bdev, BLKRRPART, 0);
679 return 0;
680
681 out_putf:
682 fput(file);
683 out:
684 return error;
685}
686
687static inline int is_loop_device(struct file *file)
688{
689 struct inode *i = file->f_mapping->host;
690
691 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
692}
693
694/* loop sysfs attributes */
695
696static ssize_t loop_attr_show(struct device *dev, char *page,
697 ssize_t (*callback)(struct loop_device *, char *))
698{
699 struct gendisk *disk = dev_to_disk(dev);
700 struct loop_device *lo = disk->private_data;
701
702 return callback(lo, page);
703}
704
705#define LOOP_ATTR_RO(_name) \
706static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
707static ssize_t loop_attr_do_show_##_name(struct device *d, \
708 struct device_attribute *attr, char *b) \
709{ \
710 return loop_attr_show(d, b, loop_attr_##_name##_show); \
711} \
712static struct device_attribute loop_attr_##_name = \
713 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
714
715static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
716{
717 ssize_t ret;
718 char *p = NULL;
719
720 spin_lock_irq(&lo->lo_lock);
721 if (lo->lo_backing_file)
722 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
723 spin_unlock_irq(&lo->lo_lock);
724
725 if (IS_ERR_OR_NULL(p))
726 ret = PTR_ERR(p);
727 else {
728 ret = strlen(p);
729 memmove(buf, p, ret);
730 buf[ret++] = '\n';
731 buf[ret] = 0;
732 }
733
734 return ret;
735}
736
737static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
738{
739 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
740}
741
742static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
743{
744 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
745}
746
747static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
748{
749 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
750
751 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
752}
753
754static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
755{
756 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
757
758 return sprintf(buf, "%s\n", partscan ? "1" : "0");
759}
760
761LOOP_ATTR_RO(backing_file);
762LOOP_ATTR_RO(offset);
763LOOP_ATTR_RO(sizelimit);
764LOOP_ATTR_RO(autoclear);
765LOOP_ATTR_RO(partscan);
766
767static struct attribute *loop_attrs[] = {
768 &loop_attr_backing_file.attr,
769 &loop_attr_offset.attr,
770 &loop_attr_sizelimit.attr,
771 &loop_attr_autoclear.attr,
772 &loop_attr_partscan.attr,
773 NULL,
774};
775
776static struct attribute_group loop_attribute_group = {
777 .name = "loop",
778 .attrs= loop_attrs,
779};
780
781static int loop_sysfs_init(struct loop_device *lo)
782{
783 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
784 &loop_attribute_group);
785}
786
787static void loop_sysfs_exit(struct loop_device *lo)
788{
789 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
790 &loop_attribute_group);
791}
792
793static void loop_config_discard(struct loop_device *lo)
794{
795 struct file *file = lo->lo_backing_file;
796 struct inode *inode = file->f_mapping->host;
797 struct request_queue *q = lo->lo_queue;
798
799 /*
800 * We use punch hole to reclaim the free space used by the
801 * image a.k.a. discard. However we do support discard if
802 * encryption is enabled, because it may give an attacker
803 * useful information.
804 */
805 if ((!file->f_op->fallocate) ||
806 lo->lo_encrypt_key_size) {
807 q->limits.discard_granularity = 0;
808 q->limits.discard_alignment = 0;
809 q->limits.max_discard_sectors = 0;
810 q->limits.discard_zeroes_data = 0;
811 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
812 return;
813 }
814
815 q->limits.discard_granularity = inode->i_sb->s_blocksize;
816 q->limits.discard_alignment = 0;
817 q->limits.max_discard_sectors = UINT_MAX >> 9;
818 q->limits.discard_zeroes_data = 1;
819 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
820}
821
822static int loop_set_fd(struct loop_device *lo, fmode_t mode,
823 struct block_device *bdev, unsigned int arg)
824{
825 struct file *file, *f;
826 struct inode *inode;
827 struct address_space *mapping;
828 unsigned lo_blocksize;
829 int lo_flags = 0;
830 int error;
831 loff_t size;
832
833 /* This is safe, since we have a reference from open(). */
834 __module_get(THIS_MODULE);
835
836 error = -EBADF;
837 file = fget(arg);
838 if (!file)
839 goto out;
840
841 error = -EBUSY;
842 if (lo->lo_state != Lo_unbound)
843 goto out_putf;
844
845 /* Avoid recursion */
846 f = file;
847 while (is_loop_device(f)) {
848 struct loop_device *l;
849
850 if (f->f_mapping->host->i_bdev == bdev)
851 goto out_putf;
852
853 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
854 if (l->lo_state == Lo_unbound) {
855 error = -EINVAL;
856 goto out_putf;
857 }
858 f = l->lo_backing_file;
859 }
860
861 mapping = file->f_mapping;
862 inode = mapping->host;
863
864 error = -EINVAL;
865 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
866 goto out_putf;
867
868 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
869 !file->f_op->write)
870 lo_flags |= LO_FLAGS_READ_ONLY;
871
872 lo_blocksize = S_ISBLK(inode->i_mode) ?
873 inode->i_bdev->bd_block_size : PAGE_SIZE;
874
875 error = -EFBIG;
876 size = get_loop_size(lo, file);
877 if ((loff_t)(sector_t)size != size)
878 goto out_putf;
879
880 error = 0;
881
882 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
883
884 lo->lo_blocksize = lo_blocksize;
885 lo->lo_device = bdev;
886 lo->lo_flags = lo_flags;
887 lo->lo_backing_file = file;
888 lo->transfer = transfer_none;
889 lo->ioctl = NULL;
890 lo->lo_sizelimit = 0;
891 lo->lo_bio_count = 0;
892 lo->old_gfp_mask = mapping_gfp_mask(mapping);
893 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
894
895 bio_list_init(&lo->lo_bio_list);
896
897 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
898 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
899
900 set_capacity(lo->lo_disk, size);
901 bd_set_size(bdev, size << 9);
902 loop_sysfs_init(lo);
903 /* let user-space know about the new size */
904 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
905
906 set_blocksize(bdev, lo_blocksize);
907
908 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
909 lo->lo_number);
910 if (IS_ERR(lo->lo_thread)) {
911 error = PTR_ERR(lo->lo_thread);
912 goto out_clr;
913 }
914 lo->lo_state = Lo_bound;
915 wake_up_process(lo->lo_thread);
916 if (part_shift)
917 lo->lo_flags |= LO_FLAGS_PARTSCAN;
918 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
919 ioctl_by_bdev(bdev, BLKRRPART, 0);
920
921 /* Grab the block_device to prevent its destruction after we
922 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
923 */
924 bdgrab(bdev);
925 return 0;
926
927out_clr:
928 loop_sysfs_exit(lo);
929 lo->lo_thread = NULL;
930 lo->lo_device = NULL;
931 lo->lo_backing_file = NULL;
932 lo->lo_flags = 0;
933 set_capacity(lo->lo_disk, 0);
934 invalidate_bdev(bdev);
935 bd_set_size(bdev, 0);
936 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
937 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
938 lo->lo_state = Lo_unbound;
939 out_putf:
940 fput(file);
941 out:
942 /* This is safe: open() is still holding a reference. */
943 module_put(THIS_MODULE);
944 return error;
945}
946
947static int
948loop_release_xfer(struct loop_device *lo)
949{
950 int err = 0;
951 struct loop_func_table *xfer = lo->lo_encryption;
952
953 if (xfer) {
954 if (xfer->release)
955 err = xfer->release(lo);
956 lo->transfer = NULL;
957 lo->lo_encryption = NULL;
958 module_put(xfer->owner);
959 }
960 return err;
961}
962
963static int
964loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
965 const struct loop_info64 *i)
966{
967 int err = 0;
968
969 if (xfer) {
970 struct module *owner = xfer->owner;
971
972 if (!try_module_get(owner))
973 return -EINVAL;
974 if (xfer->init)
975 err = xfer->init(lo, i);
976 if (err)
977 module_put(owner);
978 else
979 lo->lo_encryption = xfer;
980 }
981 return err;
982}
983
984static int loop_clr_fd(struct loop_device *lo)
985{
986 struct file *filp = lo->lo_backing_file;
987 gfp_t gfp = lo->old_gfp_mask;
988 struct block_device *bdev = lo->lo_device;
989
990 if (lo->lo_state != Lo_bound)
991 return -ENXIO;
992
993 /*
994 * If we've explicitly asked to tear down the loop device,
995 * and it has an elevated reference count, set it for auto-teardown when
996 * the last reference goes away. This stops $!~#$@ udev from
997 * preventing teardown because it decided that it needs to run blkid on
998 * the loopback device whenever they appear. xfstests is notorious for
999 * failing tests because blkid via udev races with a losetup
1000 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1001 * command to fail with EBUSY.
1002 */
1003 if (lo->lo_refcnt > 1) {
1004 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1005 mutex_unlock(&lo->lo_ctl_mutex);
1006 return 0;
1007 }
1008
1009 if (filp == NULL)
1010 return -EINVAL;
1011
1012 spin_lock_irq(&lo->lo_lock);
1013 lo->lo_state = Lo_rundown;
1014 spin_unlock_irq(&lo->lo_lock);
1015
1016 kthread_stop(lo->lo_thread);
1017
1018 spin_lock_irq(&lo->lo_lock);
1019 lo->lo_backing_file = NULL;
1020 spin_unlock_irq(&lo->lo_lock);
1021
1022 loop_release_xfer(lo);
1023 lo->transfer = NULL;
1024 lo->ioctl = NULL;
1025 lo->lo_device = NULL;
1026 lo->lo_encryption = NULL;
1027 lo->lo_offset = 0;
1028 lo->lo_sizelimit = 0;
1029 lo->lo_encrypt_key_size = 0;
1030 lo->lo_thread = NULL;
1031 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1032 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1033 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1034 if (bdev) {
1035 bdput(bdev);
1036 invalidate_bdev(bdev);
1037 }
1038 set_capacity(lo->lo_disk, 0);
1039 loop_sysfs_exit(lo);
1040 if (bdev) {
1041 bd_set_size(bdev, 0);
1042 /* let user-space know about this change */
1043 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1044 }
1045 mapping_set_gfp_mask(filp->f_mapping, gfp);
1046 lo->lo_state = Lo_unbound;
1047 /* This is safe: open() is still holding a reference. */
1048 module_put(THIS_MODULE);
1049 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1050 ioctl_by_bdev(bdev, BLKRRPART, 0);
1051 lo->lo_flags = 0;
1052 if (!part_shift)
1053 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1054 mutex_unlock(&lo->lo_ctl_mutex);
1055 /*
1056 * Need not hold lo_ctl_mutex to fput backing file.
1057 * Calling fput holding lo_ctl_mutex triggers a circular
1058 * lock dependency possibility warning as fput can take
1059 * bd_mutex which is usually taken before lo_ctl_mutex.
1060 */
1061 fput(filp);
1062 return 0;
1063}
1064
1065static int
1066loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1067{
1068 int err;
1069 struct loop_func_table *xfer;
1070 kuid_t uid = current_uid();
1071
1072 if (lo->lo_encrypt_key_size &&
1073 !uid_eq(lo->lo_key_owner, uid) &&
1074 !capable(CAP_SYS_ADMIN))
1075 return -EPERM;
1076 if (lo->lo_state != Lo_bound)
1077 return -ENXIO;
1078 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1079 return -EINVAL;
1080
1081 err = loop_release_xfer(lo);
1082 if (err)
1083 return err;
1084
1085 if (info->lo_encrypt_type) {
1086 unsigned int type = info->lo_encrypt_type;
1087
1088 if (type >= MAX_LO_CRYPT)
1089 return -EINVAL;
1090 xfer = xfer_funcs[type];
1091 if (xfer == NULL)
1092 return -EINVAL;
1093 } else
1094 xfer = NULL;
1095
1096 err = loop_init_xfer(lo, xfer, info);
1097 if (err)
1098 return err;
1099
1100 if (lo->lo_offset != info->lo_offset ||
1101 lo->lo_sizelimit != info->lo_sizelimit)
1102 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
1103 return -EFBIG;
1104
1105 loop_config_discard(lo);
1106
1107 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1108 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1109 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1110 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1111
1112 if (!xfer)
1113 xfer = &none_funcs;
1114 lo->transfer = xfer->transfer;
1115 lo->ioctl = xfer->ioctl;
1116
1117 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1118 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1119 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1120
1121 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1122 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1123 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1124 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1125 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1126 }
1127
1128 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1129 lo->lo_init[0] = info->lo_init[0];
1130 lo->lo_init[1] = info->lo_init[1];
1131 if (info->lo_encrypt_key_size) {
1132 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1133 info->lo_encrypt_key_size);
1134 lo->lo_key_owner = uid;
1135 }
1136
1137 return 0;
1138}
1139
1140static int
1141loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1142{
1143 struct file *file = lo->lo_backing_file;
1144 struct kstat stat;
1145 int error;
1146
1147 if (lo->lo_state != Lo_bound)
1148 return -ENXIO;
1149 error = vfs_getattr(&file->f_path, &stat);
1150 if (error)
1151 return error;
1152 memset(info, 0, sizeof(*info));
1153 info->lo_number = lo->lo_number;
1154 info->lo_device = huge_encode_dev(stat.dev);
1155 info->lo_inode = stat.ino;
1156 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1157 info->lo_offset = lo->lo_offset;
1158 info->lo_sizelimit = lo->lo_sizelimit;
1159 info->lo_flags = lo->lo_flags;
1160 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1161 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1162 info->lo_encrypt_type =
1163 lo->lo_encryption ? lo->lo_encryption->number : 0;
1164 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1165 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1166 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1167 lo->lo_encrypt_key_size);
1168 }
1169 return 0;
1170}
1171
1172static void
1173loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1174{
1175 memset(info64, 0, sizeof(*info64));
1176 info64->lo_number = info->lo_number;
1177 info64->lo_device = info->lo_device;
1178 info64->lo_inode = info->lo_inode;
1179 info64->lo_rdevice = info->lo_rdevice;
1180 info64->lo_offset = info->lo_offset;
1181 info64->lo_sizelimit = 0;
1182 info64->lo_encrypt_type = info->lo_encrypt_type;
1183 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1184 info64->lo_flags = info->lo_flags;
1185 info64->lo_init[0] = info->lo_init[0];
1186 info64->lo_init[1] = info->lo_init[1];
1187 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1188 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1189 else
1190 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1191 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1192}
1193
1194static int
1195loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1196{
1197 memset(info, 0, sizeof(*info));
1198 info->lo_number = info64->lo_number;
1199 info->lo_device = info64->lo_device;
1200 info->lo_inode = info64->lo_inode;
1201 info->lo_rdevice = info64->lo_rdevice;
1202 info->lo_offset = info64->lo_offset;
1203 info->lo_encrypt_type = info64->lo_encrypt_type;
1204 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1205 info->lo_flags = info64->lo_flags;
1206 info->lo_init[0] = info64->lo_init[0];
1207 info->lo_init[1] = info64->lo_init[1];
1208 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1209 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1210 else
1211 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1212 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1213
1214 /* error in case values were truncated */
1215 if (info->lo_device != info64->lo_device ||
1216 info->lo_rdevice != info64->lo_rdevice ||
1217 info->lo_inode != info64->lo_inode ||
1218 info->lo_offset != info64->lo_offset)
1219 return -EOVERFLOW;
1220
1221 return 0;
1222}
1223
1224static int
1225loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1226{
1227 struct loop_info info;
1228 struct loop_info64 info64;
1229
1230 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1231 return -EFAULT;
1232 loop_info64_from_old(&info, &info64);
1233 return loop_set_status(lo, &info64);
1234}
1235
1236static int
1237loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1238{
1239 struct loop_info64 info64;
1240
1241 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1242 return -EFAULT;
1243 return loop_set_status(lo, &info64);
1244}
1245
1246static int
1247loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1248 struct loop_info info;
1249 struct loop_info64 info64;
1250 int err = 0;
1251
1252 if (!arg)
1253 err = -EINVAL;
1254 if (!err)
1255 err = loop_get_status(lo, &info64);
1256 if (!err)
1257 err = loop_info64_to_old(&info64, &info);
1258 if (!err && copy_to_user(arg, &info, sizeof(info)))
1259 err = -EFAULT;
1260
1261 return err;
1262}
1263
1264static int
1265loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1266 struct loop_info64 info64;
1267 int err = 0;
1268
1269 if (!arg)
1270 err = -EINVAL;
1271 if (!err)
1272 err = loop_get_status(lo, &info64);
1273 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1274 err = -EFAULT;
1275
1276 return err;
1277}
1278
1279static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1280{
1281 if (unlikely(lo->lo_state != Lo_bound))
1282 return -ENXIO;
1283
1284 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1285}
1286
1287static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1288 unsigned int cmd, unsigned long arg)
1289{
1290 struct loop_device *lo = bdev->bd_disk->private_data;
1291 int err;
1292
1293 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1294 switch (cmd) {
1295 case LOOP_SET_FD:
1296 err = loop_set_fd(lo, mode, bdev, arg);
1297 break;
1298 case LOOP_CHANGE_FD:
1299 err = loop_change_fd(lo, bdev, arg);
1300 break;
1301 case LOOP_CLR_FD:
1302 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1303 err = loop_clr_fd(lo);
1304 if (!err)
1305 goto out_unlocked;
1306 break;
1307 case LOOP_SET_STATUS:
1308 err = -EPERM;
1309 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1310 err = loop_set_status_old(lo,
1311 (struct loop_info __user *)arg);
1312 break;
1313 case LOOP_GET_STATUS:
1314 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1315 break;
1316 case LOOP_SET_STATUS64:
1317 err = -EPERM;
1318 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1319 err = loop_set_status64(lo,
1320 (struct loop_info64 __user *) arg);
1321 break;
1322 case LOOP_GET_STATUS64:
1323 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1324 break;
1325 case LOOP_SET_CAPACITY:
1326 err = -EPERM;
1327 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1328 err = loop_set_capacity(lo, bdev);
1329 break;
1330 default:
1331 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1332 }
1333 mutex_unlock(&lo->lo_ctl_mutex);
1334
1335out_unlocked:
1336 return err;
1337}
1338
1339#ifdef CONFIG_COMPAT
1340struct compat_loop_info {
1341 compat_int_t lo_number; /* ioctl r/o */
1342 compat_dev_t lo_device; /* ioctl r/o */
1343 compat_ulong_t lo_inode; /* ioctl r/o */
1344 compat_dev_t lo_rdevice; /* ioctl r/o */
1345 compat_int_t lo_offset;
1346 compat_int_t lo_encrypt_type;
1347 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1348 compat_int_t lo_flags; /* ioctl r/o */
1349 char lo_name[LO_NAME_SIZE];
1350 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1351 compat_ulong_t lo_init[2];
1352 char reserved[4];
1353};
1354
1355/*
1356 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1357 * - noinlined to reduce stack space usage in main part of driver
1358 */
1359static noinline int
1360loop_info64_from_compat(const struct compat_loop_info __user *arg,
1361 struct loop_info64 *info64)
1362{
1363 struct compat_loop_info info;
1364
1365 if (copy_from_user(&info, arg, sizeof(info)))
1366 return -EFAULT;
1367
1368 memset(info64, 0, sizeof(*info64));
1369 info64->lo_number = info.lo_number;
1370 info64->lo_device = info.lo_device;
1371 info64->lo_inode = info.lo_inode;
1372 info64->lo_rdevice = info.lo_rdevice;
1373 info64->lo_offset = info.lo_offset;
1374 info64->lo_sizelimit = 0;
1375 info64->lo_encrypt_type = info.lo_encrypt_type;
1376 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1377 info64->lo_flags = info.lo_flags;
1378 info64->lo_init[0] = info.lo_init[0];
1379 info64->lo_init[1] = info.lo_init[1];
1380 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1381 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1382 else
1383 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1384 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1385 return 0;
1386}
1387
1388/*
1389 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1390 * - noinlined to reduce stack space usage in main part of driver
1391 */
1392static noinline int
1393loop_info64_to_compat(const struct loop_info64 *info64,
1394 struct compat_loop_info __user *arg)
1395{
1396 struct compat_loop_info info;
1397
1398 memset(&info, 0, sizeof(info));
1399 info.lo_number = info64->lo_number;
1400 info.lo_device = info64->lo_device;
1401 info.lo_inode = info64->lo_inode;
1402 info.lo_rdevice = info64->lo_rdevice;
1403 info.lo_offset = info64->lo_offset;
1404 info.lo_encrypt_type = info64->lo_encrypt_type;
1405 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1406 info.lo_flags = info64->lo_flags;
1407 info.lo_init[0] = info64->lo_init[0];
1408 info.lo_init[1] = info64->lo_init[1];
1409 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1410 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1411 else
1412 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1413 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1414
1415 /* error in case values were truncated */
1416 if (info.lo_device != info64->lo_device ||
1417 info.lo_rdevice != info64->lo_rdevice ||
1418 info.lo_inode != info64->lo_inode ||
1419 info.lo_offset != info64->lo_offset ||
1420 info.lo_init[0] != info64->lo_init[0] ||
1421 info.lo_init[1] != info64->lo_init[1])
1422 return -EOVERFLOW;
1423
1424 if (copy_to_user(arg, &info, sizeof(info)))
1425 return -EFAULT;
1426 return 0;
1427}
1428
1429static int
1430loop_set_status_compat(struct loop_device *lo,
1431 const struct compat_loop_info __user *arg)
1432{
1433 struct loop_info64 info64;
1434 int ret;
1435
1436 ret = loop_info64_from_compat(arg, &info64);
1437 if (ret < 0)
1438 return ret;
1439 return loop_set_status(lo, &info64);
1440}
1441
1442static int
1443loop_get_status_compat(struct loop_device *lo,
1444 struct compat_loop_info __user *arg)
1445{
1446 struct loop_info64 info64;
1447 int err = 0;
1448
1449 if (!arg)
1450 err = -EINVAL;
1451 if (!err)
1452 err = loop_get_status(lo, &info64);
1453 if (!err)
1454 err = loop_info64_to_compat(&info64, arg);
1455 return err;
1456}
1457
1458static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1459 unsigned int cmd, unsigned long arg)
1460{
1461 struct loop_device *lo = bdev->bd_disk->private_data;
1462 int err;
1463
1464 switch(cmd) {
1465 case LOOP_SET_STATUS:
1466 mutex_lock(&lo->lo_ctl_mutex);
1467 err = loop_set_status_compat(
1468 lo, (const struct compat_loop_info __user *) arg);
1469 mutex_unlock(&lo->lo_ctl_mutex);
1470 break;
1471 case LOOP_GET_STATUS:
1472 mutex_lock(&lo->lo_ctl_mutex);
1473 err = loop_get_status_compat(
1474 lo, (struct compat_loop_info __user *) arg);
1475 mutex_unlock(&lo->lo_ctl_mutex);
1476 break;
1477 case LOOP_SET_CAPACITY:
1478 case LOOP_CLR_FD:
1479 case LOOP_GET_STATUS64:
1480 case LOOP_SET_STATUS64:
1481 arg = (unsigned long) compat_ptr(arg);
1482 case LOOP_SET_FD:
1483 case LOOP_CHANGE_FD:
1484 err = lo_ioctl(bdev, mode, cmd, arg);
1485 break;
1486 default:
1487 err = -ENOIOCTLCMD;
1488 break;
1489 }
1490 return err;
1491}
1492#endif
1493
1494static int lo_open(struct block_device *bdev, fmode_t mode)
1495{
1496 struct loop_device *lo;
1497 int err = 0;
1498
1499 mutex_lock(&loop_index_mutex);
1500 lo = bdev->bd_disk->private_data;
1501 if (!lo) {
1502 err = -ENXIO;
1503 goto out;
1504 }
1505
1506 mutex_lock(&lo->lo_ctl_mutex);
1507 lo->lo_refcnt++;
1508 mutex_unlock(&lo->lo_ctl_mutex);
1509out:
1510 mutex_unlock(&loop_index_mutex);
1511 return err;
1512}
1513
1514static void lo_release(struct gendisk *disk, fmode_t mode)
1515{
1516 struct loop_device *lo = disk->private_data;
1517 int err;
1518
1519 mutex_lock(&lo->lo_ctl_mutex);
1520
1521 if (--lo->lo_refcnt)
1522 goto out;
1523
1524 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1525 /*
1526 * In autoclear mode, stop the loop thread
1527 * and remove configuration after last close.
1528 */
1529 err = loop_clr_fd(lo);
1530 if (!err)
1531 return;
1532 } else {
1533 /*
1534 * Otherwise keep thread (if running) and config,
1535 * but flush possible ongoing bios in thread.
1536 */
1537 loop_flush(lo);
1538 }
1539
1540out:
1541 mutex_unlock(&lo->lo_ctl_mutex);
1542}
1543
1544static const struct block_device_operations lo_fops = {
1545 .owner = THIS_MODULE,
1546 .open = lo_open,
1547 .release = lo_release,
1548 .ioctl = lo_ioctl,
1549#ifdef CONFIG_COMPAT
1550 .compat_ioctl = lo_compat_ioctl,
1551#endif
1552};
1553
1554/*
1555 * And now the modules code and kernel interface.
1556 */
1557static int max_loop;
1558module_param(max_loop, int, S_IRUGO);
1559MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1560module_param(max_part, int, S_IRUGO);
1561MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1562MODULE_LICENSE("GPL");
1563MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1564
1565int loop_register_transfer(struct loop_func_table *funcs)
1566{
1567 unsigned int n = funcs->number;
1568
1569 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1570 return -EINVAL;
1571 xfer_funcs[n] = funcs;
1572 return 0;
1573}
1574
1575static int unregister_transfer_cb(int id, void *ptr, void *data)
1576{
1577 struct loop_device *lo = ptr;
1578 struct loop_func_table *xfer = data;
1579
1580 mutex_lock(&lo->lo_ctl_mutex);
1581 if (lo->lo_encryption == xfer)
1582 loop_release_xfer(lo);
1583 mutex_unlock(&lo->lo_ctl_mutex);
1584 return 0;
1585}
1586
1587int loop_unregister_transfer(int number)
1588{
1589 unsigned int n = number;
1590 struct loop_func_table *xfer;
1591
1592 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1593 return -EINVAL;
1594
1595 xfer_funcs[n] = NULL;
1596 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1597 return 0;
1598}
1599
1600EXPORT_SYMBOL(loop_register_transfer);
1601EXPORT_SYMBOL(loop_unregister_transfer);
1602
1603static int loop_add(struct loop_device **l, int i)
1604{
1605 struct loop_device *lo;
1606 struct gendisk *disk;
1607 int err;
1608
1609 err = -ENOMEM;
1610 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1611 if (!lo)
1612 goto out;
1613
1614 lo->lo_state = Lo_unbound;
1615
1616 /* allocate id, if @id >= 0, we're requesting that specific id */
1617 if (i >= 0) {
1618 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1619 if (err == -ENOSPC)
1620 err = -EEXIST;
1621 } else {
1622 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1623 }
1624 if (err < 0)
1625 goto out_free_dev;
1626 i = err;
1627
1628 err = -ENOMEM;
1629 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1630 if (!lo->lo_queue)
1631 goto out_free_idr;
1632
1633 /*
1634 * set queue make_request_fn
1635 */
1636 blk_queue_make_request(lo->lo_queue, loop_make_request);
1637 lo->lo_queue->queuedata = lo;
1638
1639 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1640 if (!disk)
1641 goto out_free_queue;
1642
1643 /*
1644 * Disable partition scanning by default. The in-kernel partition
1645 * scanning can be requested individually per-device during its
1646 * setup. Userspace can always add and remove partitions from all
1647 * devices. The needed partition minors are allocated from the
1648 * extended minor space, the main loop device numbers will continue
1649 * to match the loop minors, regardless of the number of partitions
1650 * used.
1651 *
1652 * If max_part is given, partition scanning is globally enabled for
1653 * all loop devices. The minors for the main loop devices will be
1654 * multiples of max_part.
1655 *
1656 * Note: Global-for-all-devices, set-only-at-init, read-only module
1657 * parameteters like 'max_loop' and 'max_part' make things needlessly
1658 * complicated, are too static, inflexible and may surprise
1659 * userspace tools. Parameters like this in general should be avoided.
1660 */
1661 if (!part_shift)
1662 disk->flags |= GENHD_FL_NO_PART_SCAN;
1663 disk->flags |= GENHD_FL_EXT_DEVT;
1664 mutex_init(&lo->lo_ctl_mutex);
1665 lo->lo_number = i;
1666 lo->lo_thread = NULL;
1667 init_waitqueue_head(&lo->lo_event);
1668 init_waitqueue_head(&lo->lo_req_wait);
1669 spin_lock_init(&lo->lo_lock);
1670 disk->major = LOOP_MAJOR;
1671 disk->first_minor = i << part_shift;
1672 disk->fops = &lo_fops;
1673 disk->private_data = lo;
1674 disk->queue = lo->lo_queue;
1675 sprintf(disk->disk_name, "loop%d", i);
1676 add_disk(disk);
1677 *l = lo;
1678 return lo->lo_number;
1679
1680out_free_queue:
1681 blk_cleanup_queue(lo->lo_queue);
1682out_free_idr:
1683 idr_remove(&loop_index_idr, i);
1684out_free_dev:
1685 kfree(lo);
1686out:
1687 return err;
1688}
1689
1690static void loop_remove(struct loop_device *lo)
1691{
1692 del_gendisk(lo->lo_disk);
1693 blk_cleanup_queue(lo->lo_queue);
1694 put_disk(lo->lo_disk);
1695 kfree(lo);
1696}
1697
1698static int find_free_cb(int id, void *ptr, void *data)
1699{
1700 struct loop_device *lo = ptr;
1701 struct loop_device **l = data;
1702
1703 if (lo->lo_state == Lo_unbound) {
1704 *l = lo;
1705 return 1;
1706 }
1707 return 0;
1708}
1709
1710static int loop_lookup(struct loop_device **l, int i)
1711{
1712 struct loop_device *lo;
1713 int ret = -ENODEV;
1714
1715 if (i < 0) {
1716 int err;
1717
1718 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1719 if (err == 1) {
1720 *l = lo;
1721 ret = lo->lo_number;
1722 }
1723 goto out;
1724 }
1725
1726 /* lookup and return a specific i */
1727 lo = idr_find(&loop_index_idr, i);
1728 if (lo) {
1729 *l = lo;
1730 ret = lo->lo_number;
1731 }
1732out:
1733 return ret;
1734}
1735
1736static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1737{
1738 struct loop_device *lo;
1739 struct kobject *kobj;
1740 int err;
1741
1742 mutex_lock(&loop_index_mutex);
1743 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1744 if (err < 0)
1745 err = loop_add(&lo, MINOR(dev) >> part_shift);
1746 if (err < 0)
1747 kobj = NULL;
1748 else
1749 kobj = get_disk(lo->lo_disk);
1750 mutex_unlock(&loop_index_mutex);
1751
1752 *part = 0;
1753 return kobj;
1754}
1755
1756static long loop_control_ioctl(struct file *file, unsigned int cmd,
1757 unsigned long parm)
1758{
1759 struct loop_device *lo;
1760 int ret = -ENOSYS;
1761
1762 mutex_lock(&loop_index_mutex);
1763 switch (cmd) {
1764 case LOOP_CTL_ADD:
1765 ret = loop_lookup(&lo, parm);
1766 if (ret >= 0) {
1767 ret = -EEXIST;
1768 break;
1769 }
1770 ret = loop_add(&lo, parm);
1771 break;
1772 case LOOP_CTL_REMOVE:
1773 ret = loop_lookup(&lo, parm);
1774 if (ret < 0)
1775 break;
1776 mutex_lock(&lo->lo_ctl_mutex);
1777 if (lo->lo_state != Lo_unbound) {
1778 ret = -EBUSY;
1779 mutex_unlock(&lo->lo_ctl_mutex);
1780 break;
1781 }
1782 if (lo->lo_refcnt > 0) {
1783 ret = -EBUSY;
1784 mutex_unlock(&lo->lo_ctl_mutex);
1785 break;
1786 }
1787 lo->lo_disk->private_data = NULL;
1788 mutex_unlock(&lo->lo_ctl_mutex);
1789 idr_remove(&loop_index_idr, lo->lo_number);
1790 loop_remove(lo);
1791 break;
1792 case LOOP_CTL_GET_FREE:
1793 ret = loop_lookup(&lo, -1);
1794 if (ret >= 0)
1795 break;
1796 ret = loop_add(&lo, -1);
1797 }
1798 mutex_unlock(&loop_index_mutex);
1799
1800 return ret;
1801}
1802
1803static const struct file_operations loop_ctl_fops = {
1804 .open = nonseekable_open,
1805 .unlocked_ioctl = loop_control_ioctl,
1806 .compat_ioctl = loop_control_ioctl,
1807 .owner = THIS_MODULE,
1808 .llseek = noop_llseek,
1809};
1810
1811static struct miscdevice loop_misc = {
1812 .minor = LOOP_CTRL_MINOR,
1813 .name = "loop-control",
1814 .fops = &loop_ctl_fops,
1815};
1816
1817MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1818MODULE_ALIAS("devname:loop-control");
1819
1820static int __init loop_init(void)
1821{
1822 int i, nr;
1823 unsigned long range;
1824 struct loop_device *lo;
1825 int err;
1826
1827 err = misc_register(&loop_misc);
1828 if (err < 0)
1829 return err;
1830
1831 part_shift = 0;
1832 if (max_part > 0) {
1833 part_shift = fls(max_part);
1834
1835 /*
1836 * Adjust max_part according to part_shift as it is exported
1837 * to user space so that user can decide correct minor number
1838 * if [s]he want to create more devices.
1839 *
1840 * Note that -1 is required because partition 0 is reserved
1841 * for the whole disk.
1842 */
1843 max_part = (1UL << part_shift) - 1;
1844 }
1845
1846 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1847 err = -EINVAL;
1848 goto misc_out;
1849 }
1850
1851 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1852 err = -EINVAL;
1853 goto misc_out;
1854 }
1855
1856 /*
1857 * If max_loop is specified, create that many devices upfront.
1858 * This also becomes a hard limit. If max_loop is not specified,
1859 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1860 * init time. Loop devices can be requested on-demand with the
1861 * /dev/loop-control interface, or be instantiated by accessing
1862 * a 'dead' device node.
1863 */
1864 if (max_loop) {
1865 nr = max_loop;
1866 range = max_loop << part_shift;
1867 } else {
1868 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1869 range = 1UL << MINORBITS;
1870 }
1871
1872 if (register_blkdev(LOOP_MAJOR, "loop")) {
1873 err = -EIO;
1874 goto misc_out;
1875 }
1876
1877 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1878 THIS_MODULE, loop_probe, NULL, NULL);
1879
1880 /* pre-create number of devices given by config or max_loop */
1881 mutex_lock(&loop_index_mutex);
1882 for (i = 0; i < nr; i++)
1883 loop_add(&lo, i);
1884 mutex_unlock(&loop_index_mutex);
1885
1886 printk(KERN_INFO "loop: module loaded\n");
1887 return 0;
1888
1889misc_out:
1890 misc_deregister(&loop_misc);
1891 return err;
1892}
1893
1894static int loop_exit_cb(int id, void *ptr, void *data)
1895{
1896 struct loop_device *lo = ptr;
1897
1898 loop_remove(lo);
1899 return 0;
1900}
1901
1902static void __exit loop_exit(void)
1903{
1904 unsigned long range;
1905
1906 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1907
1908 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1909 idr_destroy(&loop_index_idr);
1910
1911 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1912 unregister_blkdev(LOOP_MAJOR, "loop");
1913
1914 misc_deregister(&loop_misc);
1915}
1916
1917module_init(loop_init);
1918module_exit(loop_exit);
1919
1920#ifndef MODULE
1921static int __init max_loop_setup(char *str)
1922{
1923 max_loop = simple_strtol(str, NULL, 0);
1924 return 1;
1925}
1926
1927__setup("max_loop=", max_loop_setup);
1928#endif
kernel source for telekom puls (alcatel/mediatek)