mm: add Set,ClearPageSwapCache stubs
[linux-2.6/mini2440.git] / drivers / block / loop.c
blobedbaac6c05739ab183e69f94c1dc230acea27176
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/smp_lock.h>
65 #include <linux/swap.h>
66 #include <linux/slab.h>
67 #include <linux/loop.h>
68 #include <linux/compat.h>
69 #include <linux/suspend.h>
70 #include <linux/freezer.h>
71 #include <linux/writeback.h>
72 #include <linux/buffer_head.h> /* for invalidate_bdev() */
73 #include <linux/completion.h>
74 #include <linux/highmem.h>
75 #include <linux/gfp.h>
76 #include <linux/kthread.h>
77 #include <linux/splice.h>
79 #include <asm/uaccess.h>
81 static LIST_HEAD(loop_devices);
82 static DEFINE_MUTEX(loop_devices_mutex);
84 static int max_part;
85 static int part_shift;
88 * Transfer functions
90 static int transfer_none(struct loop_device *lo, int cmd,
91 struct page *raw_page, unsigned raw_off,
92 struct page *loop_page, unsigned loop_off,
93 int size, sector_t real_block)
95 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
96 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
98 if (cmd == READ)
99 memcpy(loop_buf, raw_buf, size);
100 else
101 memcpy(raw_buf, loop_buf, size);
103 kunmap_atomic(raw_buf, KM_USER0);
104 kunmap_atomic(loop_buf, KM_USER1);
105 cond_resched();
106 return 0;
109 static int transfer_xor(struct loop_device *lo, int cmd,
110 struct page *raw_page, unsigned raw_off,
111 struct page *loop_page, unsigned loop_off,
112 int size, sector_t real_block)
114 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
115 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
116 char *in, *out, *key;
117 int i, keysize;
119 if (cmd == READ) {
120 in = raw_buf;
121 out = loop_buf;
122 } else {
123 in = loop_buf;
124 out = raw_buf;
127 key = lo->lo_encrypt_key;
128 keysize = lo->lo_encrypt_key_size;
129 for (i = 0; i < size; i++)
130 *out++ = *in++ ^ key[(i & 511) % keysize];
132 kunmap_atomic(raw_buf, KM_USER0);
133 kunmap_atomic(loop_buf, KM_USER1);
134 cond_resched();
135 return 0;
138 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140 if (unlikely(info->lo_encrypt_key_size <= 0))
141 return -EINVAL;
142 return 0;
145 static struct loop_func_table none_funcs = {
146 .number = LO_CRYPT_NONE,
147 .transfer = transfer_none,
150 static struct loop_func_table xor_funcs = {
151 .number = LO_CRYPT_XOR,
152 .transfer = transfer_xor,
153 .init = xor_init
156 /* xfer_funcs[0] is special - its release function is never called */
157 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
158 &none_funcs,
159 &xor_funcs
162 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164 loff_t size, offset, loopsize;
166 /* Compute loopsize in bytes */
167 size = i_size_read(file->f_mapping->host);
168 offset = lo->lo_offset;
169 loopsize = size - offset;
170 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
171 loopsize = lo->lo_sizelimit;
174 * Unfortunately, if we want to do I/O on the device,
175 * the number of 512-byte sectors has to fit into a sector_t.
177 return loopsize >> 9;
180 static int
181 figure_loop_size(struct loop_device *lo)
183 loff_t size = get_loop_size(lo, lo->lo_backing_file);
184 sector_t x = (sector_t)size;
186 if (unlikely((loff_t)x != size))
187 return -EFBIG;
189 set_capacity(lo->lo_disk, x);
190 return 0;
193 static inline int
194 lo_do_transfer(struct loop_device *lo, int cmd,
195 struct page *rpage, unsigned roffs,
196 struct page *lpage, unsigned loffs,
197 int size, sector_t rblock)
199 if (unlikely(!lo->transfer))
200 return 0;
202 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
206 * do_lo_send_aops - helper for writing data to a loop device
208 * This is the fast version for backing filesystems which implement the address
209 * space operations write_begin and write_end.
211 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
212 loff_t pos, struct page *unused)
214 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
215 struct address_space *mapping = file->f_mapping;
216 pgoff_t index;
217 unsigned offset, bv_offs;
218 int len, ret;
220 mutex_lock(&mapping->host->i_mutex);
221 index = pos >> PAGE_CACHE_SHIFT;
222 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
223 bv_offs = bvec->bv_offset;
224 len = bvec->bv_len;
225 while (len > 0) {
226 sector_t IV;
227 unsigned size, copied;
228 int transfer_result;
229 struct page *page;
230 void *fsdata;
232 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
233 size = PAGE_CACHE_SIZE - offset;
234 if (size > len)
235 size = len;
237 ret = pagecache_write_begin(file, mapping, pos, size, 0,
238 &page, &fsdata);
239 if (ret)
240 goto fail;
242 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
243 bvec->bv_page, bv_offs, size, IV);
244 copied = size;
245 if (unlikely(transfer_result))
246 copied = 0;
248 ret = pagecache_write_end(file, mapping, pos, size, copied,
249 page, fsdata);
250 if (ret < 0 || ret != copied)
251 goto fail;
253 if (unlikely(transfer_result))
254 goto fail;
256 bv_offs += copied;
257 len -= copied;
258 offset = 0;
259 index++;
260 pos += copied;
262 ret = 0;
263 out:
264 mutex_unlock(&mapping->host->i_mutex);
265 return ret;
266 fail:
267 ret = -1;
268 goto out;
272 * __do_lo_send_write - helper for writing data to a loop device
274 * This helper just factors out common code between do_lo_send_direct_write()
275 * and do_lo_send_write().
277 static int __do_lo_send_write(struct file *file,
278 u8 *buf, const int len, loff_t pos)
280 ssize_t bw;
281 mm_segment_t old_fs = get_fs();
283 set_fs(get_ds());
284 bw = file->f_op->write(file, buf, len, &pos);
285 set_fs(old_fs);
286 if (likely(bw == len))
287 return 0;
288 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
289 (unsigned long long)pos, len);
290 if (bw >= 0)
291 bw = -EIO;
292 return bw;
296 * do_lo_send_direct_write - helper for writing data to a loop device
298 * This is the fast, non-transforming version for backing filesystems which do
299 * not implement the address space operations write_begin and write_end.
300 * It uses the write file operation which should be present on all writeable
301 * filesystems.
303 static int do_lo_send_direct_write(struct loop_device *lo,
304 struct bio_vec *bvec, loff_t pos, struct page *page)
306 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
307 kmap(bvec->bv_page) + bvec->bv_offset,
308 bvec->bv_len, pos);
309 kunmap(bvec->bv_page);
310 cond_resched();
311 return bw;
315 * do_lo_send_write - helper for writing data to a loop device
317 * This is the slow, transforming version for filesystems which do not
318 * implement the address space operations write_begin and write_end. It
319 * uses the write file operation which should be present on all writeable
320 * filesystems.
322 * Using fops->write is slower than using aops->{prepare,commit}_write in the
323 * transforming case because we need to double buffer the data as we cannot do
324 * the transformations in place as we do not have direct access to the
325 * destination pages of the backing file.
327 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
328 loff_t pos, struct page *page)
330 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
331 bvec->bv_offset, bvec->bv_len, pos >> 9);
332 if (likely(!ret))
333 return __do_lo_send_write(lo->lo_backing_file,
334 page_address(page), bvec->bv_len,
335 pos);
336 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
337 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
338 if (ret > 0)
339 ret = -EIO;
340 return ret;
343 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
345 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
346 struct page *page);
347 struct bio_vec *bvec;
348 struct page *page = NULL;
349 int i, ret = 0;
351 do_lo_send = do_lo_send_aops;
352 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
353 do_lo_send = do_lo_send_direct_write;
354 if (lo->transfer != transfer_none) {
355 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
356 if (unlikely(!page))
357 goto fail;
358 kmap(page);
359 do_lo_send = do_lo_send_write;
362 bio_for_each_segment(bvec, bio, i) {
363 ret = do_lo_send(lo, bvec, pos, page);
364 if (ret < 0)
365 break;
366 pos += bvec->bv_len;
368 if (page) {
369 kunmap(page);
370 __free_page(page);
372 out:
373 return ret;
374 fail:
375 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
376 ret = -ENOMEM;
377 goto out;
380 struct lo_read_data {
381 struct loop_device *lo;
382 struct page *page;
383 unsigned offset;
384 int bsize;
387 static int
388 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
389 struct splice_desc *sd)
391 struct lo_read_data *p = sd->u.data;
392 struct loop_device *lo = p->lo;
393 struct page *page = buf->page;
394 sector_t IV;
395 size_t size;
396 int ret;
398 ret = buf->ops->confirm(pipe, buf);
399 if (unlikely(ret))
400 return ret;
402 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
403 (buf->offset >> 9);
404 size = sd->len;
405 if (size > p->bsize)
406 size = p->bsize;
408 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
409 printk(KERN_ERR "loop: transfer error block %ld\n",
410 page->index);
411 size = -EINVAL;
414 flush_dcache_page(p->page);
416 if (size > 0)
417 p->offset += size;
419 return size;
422 static int
423 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
425 return __splice_from_pipe(pipe, sd, lo_splice_actor);
428 static int
429 do_lo_receive(struct loop_device *lo,
430 struct bio_vec *bvec, int bsize, loff_t pos)
432 struct lo_read_data cookie;
433 struct splice_desc sd;
434 struct file *file;
435 long retval;
437 cookie.lo = lo;
438 cookie.page = bvec->bv_page;
439 cookie.offset = bvec->bv_offset;
440 cookie.bsize = bsize;
442 sd.len = 0;
443 sd.total_len = bvec->bv_len;
444 sd.flags = 0;
445 sd.pos = pos;
446 sd.u.data = &cookie;
448 file = lo->lo_backing_file;
449 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
451 if (retval < 0)
452 return retval;
454 return 0;
457 static int
458 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
460 struct bio_vec *bvec;
461 int i, ret = 0;
463 bio_for_each_segment(bvec, bio, i) {
464 ret = do_lo_receive(lo, bvec, bsize, pos);
465 if (ret < 0)
466 break;
467 pos += bvec->bv_len;
469 return ret;
472 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
474 loff_t pos;
475 int ret;
477 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
478 if (bio_rw(bio) == WRITE)
479 ret = lo_send(lo, bio, pos);
480 else
481 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
482 return ret;
486 * Add bio to back of pending list
488 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
490 if (lo->lo_biotail) {
491 lo->lo_biotail->bi_next = bio;
492 lo->lo_biotail = bio;
493 } else
494 lo->lo_bio = lo->lo_biotail = bio;
498 * Grab first pending buffer
500 static struct bio *loop_get_bio(struct loop_device *lo)
502 struct bio *bio;
504 if ((bio = lo->lo_bio)) {
505 if (bio == lo->lo_biotail)
506 lo->lo_biotail = NULL;
507 lo->lo_bio = bio->bi_next;
508 bio->bi_next = NULL;
511 return bio;
514 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
516 struct loop_device *lo = q->queuedata;
517 int rw = bio_rw(old_bio);
519 if (rw == READA)
520 rw = READ;
522 BUG_ON(!lo || (rw != READ && rw != WRITE));
524 spin_lock_irq(&lo->lo_lock);
525 if (lo->lo_state != Lo_bound)
526 goto out;
527 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
528 goto out;
529 loop_add_bio(lo, old_bio);
530 wake_up(&lo->lo_event);
531 spin_unlock_irq(&lo->lo_lock);
532 return 0;
534 out:
535 spin_unlock_irq(&lo->lo_lock);
536 bio_io_error(old_bio);
537 return 0;
541 * kick off io on the underlying address space
543 static void loop_unplug(struct request_queue *q)
545 struct loop_device *lo = q->queuedata;
547 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
548 blk_run_address_space(lo->lo_backing_file->f_mapping);
551 struct switch_request {
552 struct file *file;
553 struct completion wait;
556 static void do_loop_switch(struct loop_device *, struct switch_request *);
558 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
560 if (unlikely(!bio->bi_bdev)) {
561 do_loop_switch(lo, bio->bi_private);
562 bio_put(bio);
563 } else {
564 int ret = do_bio_filebacked(lo, bio);
565 bio_endio(bio, ret);
570 * worker thread that handles reads/writes to file backed loop devices,
571 * to avoid blocking in our make_request_fn. it also does loop decrypting
572 * on reads for block backed loop, as that is too heavy to do from
573 * b_end_io context where irqs may be disabled.
575 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
576 * calling kthread_stop(). Therefore once kthread_should_stop() is
577 * true, make_request will not place any more requests. Therefore
578 * once kthread_should_stop() is true and lo_bio is NULL, we are
579 * done with the loop.
581 static int loop_thread(void *data)
583 struct loop_device *lo = data;
584 struct bio *bio;
586 set_user_nice(current, -20);
588 while (!kthread_should_stop() || lo->lo_bio) {
590 wait_event_interruptible(lo->lo_event,
591 lo->lo_bio || kthread_should_stop());
593 if (!lo->lo_bio)
594 continue;
595 spin_lock_irq(&lo->lo_lock);
596 bio = loop_get_bio(lo);
597 spin_unlock_irq(&lo->lo_lock);
599 BUG_ON(!bio);
600 loop_handle_bio(lo, bio);
603 return 0;
607 * loop_switch performs the hard work of switching a backing store.
608 * First it needs to flush existing IO, it does this by sending a magic
609 * BIO down the pipe. The completion of this BIO does the actual switch.
611 static int loop_switch(struct loop_device *lo, struct file *file)
613 struct switch_request w;
614 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
615 if (!bio)
616 return -ENOMEM;
617 init_completion(&w.wait);
618 w.file = file;
619 bio->bi_private = &w;
620 bio->bi_bdev = NULL;
621 loop_make_request(lo->lo_queue, bio);
622 wait_for_completion(&w.wait);
623 return 0;
627 * Helper to flush the IOs in loop, but keeping loop thread running
629 static int loop_flush(struct loop_device *lo)
631 /* loop not yet configured, no running thread, nothing to flush */
632 if (!lo->lo_thread)
633 return 0;
635 return loop_switch(lo, NULL);
639 * Do the actual switch; called from the BIO completion routine
641 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
643 struct file *file = p->file;
644 struct file *old_file = lo->lo_backing_file;
645 struct address_space *mapping;
647 /* if no new file, only flush of queued bios requested */
648 if (!file)
649 goto out;
651 mapping = file->f_mapping;
652 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
653 lo->lo_backing_file = file;
654 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
655 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
656 lo->old_gfp_mask = mapping_gfp_mask(mapping);
657 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
658 out:
659 complete(&p->wait);
664 * loop_change_fd switched the backing store of a loopback device to
665 * a new file. This is useful for operating system installers to free up
666 * the original file and in High Availability environments to switch to
667 * an alternative location for the content in case of server meltdown.
668 * This can only work if the loop device is used read-only, and if the
669 * new backing store is the same size and type as the old backing store.
671 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
672 unsigned int arg)
674 struct file *file, *old_file;
675 struct inode *inode;
676 int error;
678 error = -ENXIO;
679 if (lo->lo_state != Lo_bound)
680 goto out;
682 /* the loop device has to be read-only */
683 error = -EINVAL;
684 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
685 goto out;
687 error = -EBADF;
688 file = fget(arg);
689 if (!file)
690 goto out;
692 inode = file->f_mapping->host;
693 old_file = lo->lo_backing_file;
695 error = -EINVAL;
697 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
698 goto out_putf;
700 /* new backing store needs to support loop (eg splice_read) */
701 if (!inode->i_fop->splice_read)
702 goto out_putf;
704 /* size of the new backing store needs to be the same */
705 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
706 goto out_putf;
708 /* and ... switch */
709 error = loop_switch(lo, file);
710 if (error)
711 goto out_putf;
713 fput(old_file);
714 if (max_part > 0)
715 ioctl_by_bdev(bdev, BLKRRPART, 0);
716 return 0;
718 out_putf:
719 fput(file);
720 out:
721 return error;
724 static inline int is_loop_device(struct file *file)
726 struct inode *i = file->f_mapping->host;
728 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
731 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
732 struct block_device *bdev, unsigned int arg)
734 struct file *file, *f;
735 struct inode *inode;
736 struct address_space *mapping;
737 unsigned lo_blocksize;
738 int lo_flags = 0;
739 int error;
740 loff_t size;
742 /* This is safe, since we have a reference from open(). */
743 __module_get(THIS_MODULE);
745 error = -EBADF;
746 file = fget(arg);
747 if (!file)
748 goto out;
750 error = -EBUSY;
751 if (lo->lo_state != Lo_unbound)
752 goto out_putf;
754 /* Avoid recursion */
755 f = file;
756 while (is_loop_device(f)) {
757 struct loop_device *l;
759 if (f->f_mapping->host->i_bdev == bdev)
760 goto out_putf;
762 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
763 if (l->lo_state == Lo_unbound) {
764 error = -EINVAL;
765 goto out_putf;
767 f = l->lo_backing_file;
770 mapping = file->f_mapping;
771 inode = mapping->host;
773 if (!(file->f_mode & FMODE_WRITE))
774 lo_flags |= LO_FLAGS_READ_ONLY;
776 error = -EINVAL;
777 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
778 const struct address_space_operations *aops = mapping->a_ops;
780 * If we can't read - sorry. If we only can't write - well,
781 * it's going to be read-only.
783 if (!file->f_op->splice_read)
784 goto out_putf;
785 if (aops->write_begin)
786 lo_flags |= LO_FLAGS_USE_AOPS;
787 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
788 lo_flags |= LO_FLAGS_READ_ONLY;
790 lo_blocksize = S_ISBLK(inode->i_mode) ?
791 inode->i_bdev->bd_block_size : PAGE_SIZE;
793 error = 0;
794 } else {
795 goto out_putf;
798 size = get_loop_size(lo, file);
800 if ((loff_t)(sector_t)size != size) {
801 error = -EFBIG;
802 goto out_putf;
805 if (!(mode & FMODE_WRITE))
806 lo_flags |= LO_FLAGS_READ_ONLY;
808 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
810 lo->lo_blocksize = lo_blocksize;
811 lo->lo_device = bdev;
812 lo->lo_flags = lo_flags;
813 lo->lo_backing_file = file;
814 lo->transfer = transfer_none;
815 lo->ioctl = NULL;
816 lo->lo_sizelimit = 0;
817 lo->old_gfp_mask = mapping_gfp_mask(mapping);
818 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
820 lo->lo_bio = lo->lo_biotail = NULL;
823 * set queue make_request_fn, and add limits based on lower level
824 * device
826 blk_queue_make_request(lo->lo_queue, loop_make_request);
827 lo->lo_queue->queuedata = lo;
828 lo->lo_queue->unplug_fn = loop_unplug;
830 set_capacity(lo->lo_disk, size);
831 bd_set_size(bdev, size << 9);
833 set_blocksize(bdev, lo_blocksize);
835 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
836 lo->lo_number);
837 if (IS_ERR(lo->lo_thread)) {
838 error = PTR_ERR(lo->lo_thread);
839 goto out_clr;
841 lo->lo_state = Lo_bound;
842 wake_up_process(lo->lo_thread);
843 if (max_part > 0)
844 ioctl_by_bdev(bdev, BLKRRPART, 0);
845 return 0;
847 out_clr:
848 lo->lo_thread = NULL;
849 lo->lo_device = NULL;
850 lo->lo_backing_file = NULL;
851 lo->lo_flags = 0;
852 set_capacity(lo->lo_disk, 0);
853 invalidate_bdev(bdev);
854 bd_set_size(bdev, 0);
855 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
856 lo->lo_state = Lo_unbound;
857 out_putf:
858 fput(file);
859 out:
860 /* This is safe: open() is still holding a reference. */
861 module_put(THIS_MODULE);
862 return error;
865 static int
866 loop_release_xfer(struct loop_device *lo)
868 int err = 0;
869 struct loop_func_table *xfer = lo->lo_encryption;
871 if (xfer) {
872 if (xfer->release)
873 err = xfer->release(lo);
874 lo->transfer = NULL;
875 lo->lo_encryption = NULL;
876 module_put(xfer->owner);
878 return err;
881 static int
882 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
883 const struct loop_info64 *i)
885 int err = 0;
887 if (xfer) {
888 struct module *owner = xfer->owner;
890 if (!try_module_get(owner))
891 return -EINVAL;
892 if (xfer->init)
893 err = xfer->init(lo, i);
894 if (err)
895 module_put(owner);
896 else
897 lo->lo_encryption = xfer;
899 return err;
902 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
904 struct file *filp = lo->lo_backing_file;
905 gfp_t gfp = lo->old_gfp_mask;
907 if (lo->lo_state != Lo_bound)
908 return -ENXIO;
910 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
911 return -EBUSY;
913 if (filp == NULL)
914 return -EINVAL;
916 spin_lock_irq(&lo->lo_lock);
917 lo->lo_state = Lo_rundown;
918 spin_unlock_irq(&lo->lo_lock);
920 kthread_stop(lo->lo_thread);
922 lo->lo_queue->unplug_fn = NULL;
923 lo->lo_backing_file = NULL;
925 loop_release_xfer(lo);
926 lo->transfer = NULL;
927 lo->ioctl = NULL;
928 lo->lo_device = NULL;
929 lo->lo_encryption = NULL;
930 lo->lo_offset = 0;
931 lo->lo_sizelimit = 0;
932 lo->lo_encrypt_key_size = 0;
933 lo->lo_flags = 0;
934 lo->lo_thread = NULL;
935 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
936 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
937 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
938 if (bdev)
939 invalidate_bdev(bdev);
940 set_capacity(lo->lo_disk, 0);
941 if (bdev)
942 bd_set_size(bdev, 0);
943 mapping_set_gfp_mask(filp->f_mapping, gfp);
944 lo->lo_state = Lo_unbound;
945 fput(filp);
946 /* This is safe: open() is still holding a reference. */
947 module_put(THIS_MODULE);
948 if (max_part > 0)
949 ioctl_by_bdev(bdev, BLKRRPART, 0);
950 return 0;
953 static int
954 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
956 int err;
957 struct loop_func_table *xfer;
958 uid_t uid = current_uid();
960 if (lo->lo_encrypt_key_size &&
961 lo->lo_key_owner != uid &&
962 !capable(CAP_SYS_ADMIN))
963 return -EPERM;
964 if (lo->lo_state != Lo_bound)
965 return -ENXIO;
966 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
967 return -EINVAL;
969 err = loop_release_xfer(lo);
970 if (err)
971 return err;
973 if (info->lo_encrypt_type) {
974 unsigned int type = info->lo_encrypt_type;
976 if (type >= MAX_LO_CRYPT)
977 return -EINVAL;
978 xfer = xfer_funcs[type];
979 if (xfer == NULL)
980 return -EINVAL;
981 } else
982 xfer = NULL;
984 err = loop_init_xfer(lo, xfer, info);
985 if (err)
986 return err;
988 if (lo->lo_offset != info->lo_offset ||
989 lo->lo_sizelimit != info->lo_sizelimit) {
990 lo->lo_offset = info->lo_offset;
991 lo->lo_sizelimit = info->lo_sizelimit;
992 if (figure_loop_size(lo))
993 return -EFBIG;
996 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
997 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
998 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
999 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1001 if (!xfer)
1002 xfer = &none_funcs;
1003 lo->transfer = xfer->transfer;
1004 lo->ioctl = xfer->ioctl;
1006 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1007 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1008 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1010 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1011 lo->lo_init[0] = info->lo_init[0];
1012 lo->lo_init[1] = info->lo_init[1];
1013 if (info->lo_encrypt_key_size) {
1014 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1015 info->lo_encrypt_key_size);
1016 lo->lo_key_owner = uid;
1019 return 0;
1022 static int
1023 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1025 struct file *file = lo->lo_backing_file;
1026 struct kstat stat;
1027 int error;
1029 if (lo->lo_state != Lo_bound)
1030 return -ENXIO;
1031 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1032 if (error)
1033 return error;
1034 memset(info, 0, sizeof(*info));
1035 info->lo_number = lo->lo_number;
1036 info->lo_device = huge_encode_dev(stat.dev);
1037 info->lo_inode = stat.ino;
1038 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1039 info->lo_offset = lo->lo_offset;
1040 info->lo_sizelimit = lo->lo_sizelimit;
1041 info->lo_flags = lo->lo_flags;
1042 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1043 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1044 info->lo_encrypt_type =
1045 lo->lo_encryption ? lo->lo_encryption->number : 0;
1046 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1047 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1048 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1049 lo->lo_encrypt_key_size);
1051 return 0;
1054 static void
1055 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1057 memset(info64, 0, sizeof(*info64));
1058 info64->lo_number = info->lo_number;
1059 info64->lo_device = info->lo_device;
1060 info64->lo_inode = info->lo_inode;
1061 info64->lo_rdevice = info->lo_rdevice;
1062 info64->lo_offset = info->lo_offset;
1063 info64->lo_sizelimit = 0;
1064 info64->lo_encrypt_type = info->lo_encrypt_type;
1065 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1066 info64->lo_flags = info->lo_flags;
1067 info64->lo_init[0] = info->lo_init[0];
1068 info64->lo_init[1] = info->lo_init[1];
1069 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1070 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1071 else
1072 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1073 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1076 static int
1077 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1079 memset(info, 0, sizeof(*info));
1080 info->lo_number = info64->lo_number;
1081 info->lo_device = info64->lo_device;
1082 info->lo_inode = info64->lo_inode;
1083 info->lo_rdevice = info64->lo_rdevice;
1084 info->lo_offset = info64->lo_offset;
1085 info->lo_encrypt_type = info64->lo_encrypt_type;
1086 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1087 info->lo_flags = info64->lo_flags;
1088 info->lo_init[0] = info64->lo_init[0];
1089 info->lo_init[1] = info64->lo_init[1];
1090 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1091 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1092 else
1093 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1094 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1096 /* error in case values were truncated */
1097 if (info->lo_device != info64->lo_device ||
1098 info->lo_rdevice != info64->lo_rdevice ||
1099 info->lo_inode != info64->lo_inode ||
1100 info->lo_offset != info64->lo_offset)
1101 return -EOVERFLOW;
1103 return 0;
1106 static int
1107 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1109 struct loop_info info;
1110 struct loop_info64 info64;
1112 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1113 return -EFAULT;
1114 loop_info64_from_old(&info, &info64);
1115 return loop_set_status(lo, &info64);
1118 static int
1119 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1121 struct loop_info64 info64;
1123 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1124 return -EFAULT;
1125 return loop_set_status(lo, &info64);
1128 static int
1129 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1130 struct loop_info info;
1131 struct loop_info64 info64;
1132 int err = 0;
1134 if (!arg)
1135 err = -EINVAL;
1136 if (!err)
1137 err = loop_get_status(lo, &info64);
1138 if (!err)
1139 err = loop_info64_to_old(&info64, &info);
1140 if (!err && copy_to_user(arg, &info, sizeof(info)))
1141 err = -EFAULT;
1143 return err;
1146 static int
1147 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1148 struct loop_info64 info64;
1149 int err = 0;
1151 if (!arg)
1152 err = -EINVAL;
1153 if (!err)
1154 err = loop_get_status(lo, &info64);
1155 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1156 err = -EFAULT;
1158 return err;
1161 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1162 unsigned int cmd, unsigned long arg)
1164 struct loop_device *lo = bdev->bd_disk->private_data;
1165 int err;
1167 mutex_lock(&lo->lo_ctl_mutex);
1168 switch (cmd) {
1169 case LOOP_SET_FD:
1170 err = loop_set_fd(lo, mode, bdev, arg);
1171 break;
1172 case LOOP_CHANGE_FD:
1173 err = loop_change_fd(lo, bdev, arg);
1174 break;
1175 case LOOP_CLR_FD:
1176 err = loop_clr_fd(lo, bdev);
1177 break;
1178 case LOOP_SET_STATUS:
1179 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1180 break;
1181 case LOOP_GET_STATUS:
1182 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1183 break;
1184 case LOOP_SET_STATUS64:
1185 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1186 break;
1187 case LOOP_GET_STATUS64:
1188 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1189 break;
1190 default:
1191 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1193 mutex_unlock(&lo->lo_ctl_mutex);
1194 return err;
1197 #ifdef CONFIG_COMPAT
1198 struct compat_loop_info {
1199 compat_int_t lo_number; /* ioctl r/o */
1200 compat_dev_t lo_device; /* ioctl r/o */
1201 compat_ulong_t lo_inode; /* ioctl r/o */
1202 compat_dev_t lo_rdevice; /* ioctl r/o */
1203 compat_int_t lo_offset;
1204 compat_int_t lo_encrypt_type;
1205 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1206 compat_int_t lo_flags; /* ioctl r/o */
1207 char lo_name[LO_NAME_SIZE];
1208 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1209 compat_ulong_t lo_init[2];
1210 char reserved[4];
1214 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1215 * - noinlined to reduce stack space usage in main part of driver
1217 static noinline int
1218 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1219 struct loop_info64 *info64)
1221 struct compat_loop_info info;
1223 if (copy_from_user(&info, arg, sizeof(info)))
1224 return -EFAULT;
1226 memset(info64, 0, sizeof(*info64));
1227 info64->lo_number = info.lo_number;
1228 info64->lo_device = info.lo_device;
1229 info64->lo_inode = info.lo_inode;
1230 info64->lo_rdevice = info.lo_rdevice;
1231 info64->lo_offset = info.lo_offset;
1232 info64->lo_sizelimit = 0;
1233 info64->lo_encrypt_type = info.lo_encrypt_type;
1234 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1235 info64->lo_flags = info.lo_flags;
1236 info64->lo_init[0] = info.lo_init[0];
1237 info64->lo_init[1] = info.lo_init[1];
1238 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1239 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1240 else
1241 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1242 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1243 return 0;
1247 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1248 * - noinlined to reduce stack space usage in main part of driver
1250 static noinline int
1251 loop_info64_to_compat(const struct loop_info64 *info64,
1252 struct compat_loop_info __user *arg)
1254 struct compat_loop_info info;
1256 memset(&info, 0, sizeof(info));
1257 info.lo_number = info64->lo_number;
1258 info.lo_device = info64->lo_device;
1259 info.lo_inode = info64->lo_inode;
1260 info.lo_rdevice = info64->lo_rdevice;
1261 info.lo_offset = info64->lo_offset;
1262 info.lo_encrypt_type = info64->lo_encrypt_type;
1263 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1264 info.lo_flags = info64->lo_flags;
1265 info.lo_init[0] = info64->lo_init[0];
1266 info.lo_init[1] = info64->lo_init[1];
1267 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1268 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1269 else
1270 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1271 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1273 /* error in case values were truncated */
1274 if (info.lo_device != info64->lo_device ||
1275 info.lo_rdevice != info64->lo_rdevice ||
1276 info.lo_inode != info64->lo_inode ||
1277 info.lo_offset != info64->lo_offset ||
1278 info.lo_init[0] != info64->lo_init[0] ||
1279 info.lo_init[1] != info64->lo_init[1])
1280 return -EOVERFLOW;
1282 if (copy_to_user(arg, &info, sizeof(info)))
1283 return -EFAULT;
1284 return 0;
1287 static int
1288 loop_set_status_compat(struct loop_device *lo,
1289 const struct compat_loop_info __user *arg)
1291 struct loop_info64 info64;
1292 int ret;
1294 ret = loop_info64_from_compat(arg, &info64);
1295 if (ret < 0)
1296 return ret;
1297 return loop_set_status(lo, &info64);
1300 static int
1301 loop_get_status_compat(struct loop_device *lo,
1302 struct compat_loop_info __user *arg)
1304 struct loop_info64 info64;
1305 int err = 0;
1307 if (!arg)
1308 err = -EINVAL;
1309 if (!err)
1310 err = loop_get_status(lo, &info64);
1311 if (!err)
1312 err = loop_info64_to_compat(&info64, arg);
1313 return err;
1316 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1317 unsigned int cmd, unsigned long arg)
1319 struct loop_device *lo = bdev->bd_disk->private_data;
1320 int err;
1322 switch(cmd) {
1323 case LOOP_SET_STATUS:
1324 mutex_lock(&lo->lo_ctl_mutex);
1325 err = loop_set_status_compat(
1326 lo, (const struct compat_loop_info __user *) arg);
1327 mutex_unlock(&lo->lo_ctl_mutex);
1328 break;
1329 case LOOP_GET_STATUS:
1330 mutex_lock(&lo->lo_ctl_mutex);
1331 err = loop_get_status_compat(
1332 lo, (struct compat_loop_info __user *) arg);
1333 mutex_unlock(&lo->lo_ctl_mutex);
1334 break;
1335 case LOOP_CLR_FD:
1336 case LOOP_GET_STATUS64:
1337 case LOOP_SET_STATUS64:
1338 arg = (unsigned long) compat_ptr(arg);
1339 case LOOP_SET_FD:
1340 case LOOP_CHANGE_FD:
1341 err = lo_ioctl(bdev, mode, cmd, arg);
1342 break;
1343 default:
1344 err = -ENOIOCTLCMD;
1345 break;
1347 return err;
1349 #endif
1351 static int lo_open(struct block_device *bdev, fmode_t mode)
1353 struct loop_device *lo = bdev->bd_disk->private_data;
1355 mutex_lock(&lo->lo_ctl_mutex);
1356 lo->lo_refcnt++;
1357 mutex_unlock(&lo->lo_ctl_mutex);
1359 return 0;
1362 static int lo_release(struct gendisk *disk, fmode_t mode)
1364 struct loop_device *lo = disk->private_data;
1366 mutex_lock(&lo->lo_ctl_mutex);
1368 if (--lo->lo_refcnt)
1369 goto out;
1371 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1373 * In autoclear mode, stop the loop thread
1374 * and remove configuration after last close.
1376 loop_clr_fd(lo, NULL);
1377 } else {
1379 * Otherwise keep thread (if running) and config,
1380 * but flush possible ongoing bios in thread.
1382 loop_flush(lo);
1385 out:
1386 mutex_unlock(&lo->lo_ctl_mutex);
1388 return 0;
1391 static struct block_device_operations lo_fops = {
1392 .owner = THIS_MODULE,
1393 .open = lo_open,
1394 .release = lo_release,
1395 .ioctl = lo_ioctl,
1396 #ifdef CONFIG_COMPAT
1397 .compat_ioctl = lo_compat_ioctl,
1398 #endif
1402 * And now the modules code and kernel interface.
1404 static int max_loop;
1405 module_param(max_loop, int, 0);
1406 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1407 module_param(max_part, int, 0);
1408 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1409 MODULE_LICENSE("GPL");
1410 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1412 int loop_register_transfer(struct loop_func_table *funcs)
1414 unsigned int n = funcs->number;
1416 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1417 return -EINVAL;
1418 xfer_funcs[n] = funcs;
1419 return 0;
1422 int loop_unregister_transfer(int number)
1424 unsigned int n = number;
1425 struct loop_device *lo;
1426 struct loop_func_table *xfer;
1428 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1429 return -EINVAL;
1431 xfer_funcs[n] = NULL;
1433 list_for_each_entry(lo, &loop_devices, lo_list) {
1434 mutex_lock(&lo->lo_ctl_mutex);
1436 if (lo->lo_encryption == xfer)
1437 loop_release_xfer(lo);
1439 mutex_unlock(&lo->lo_ctl_mutex);
1442 return 0;
1445 EXPORT_SYMBOL(loop_register_transfer);
1446 EXPORT_SYMBOL(loop_unregister_transfer);
1448 static struct loop_device *loop_alloc(int i)
1450 struct loop_device *lo;
1451 struct gendisk *disk;
1453 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1454 if (!lo)
1455 goto out;
1457 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1458 if (!lo->lo_queue)
1459 goto out_free_dev;
1461 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1462 if (!disk)
1463 goto out_free_queue;
1465 mutex_init(&lo->lo_ctl_mutex);
1466 lo->lo_number = i;
1467 lo->lo_thread = NULL;
1468 init_waitqueue_head(&lo->lo_event);
1469 spin_lock_init(&lo->lo_lock);
1470 disk->major = LOOP_MAJOR;
1471 disk->first_minor = i << part_shift;
1472 disk->fops = &lo_fops;
1473 disk->private_data = lo;
1474 disk->queue = lo->lo_queue;
1475 sprintf(disk->disk_name, "loop%d", i);
1476 return lo;
1478 out_free_queue:
1479 blk_cleanup_queue(lo->lo_queue);
1480 out_free_dev:
1481 kfree(lo);
1482 out:
1483 return NULL;
1486 static void loop_free(struct loop_device *lo)
1488 blk_cleanup_queue(lo->lo_queue);
1489 put_disk(lo->lo_disk);
1490 list_del(&lo->lo_list);
1491 kfree(lo);
1494 static struct loop_device *loop_init_one(int i)
1496 struct loop_device *lo;
1498 list_for_each_entry(lo, &loop_devices, lo_list) {
1499 if (lo->lo_number == i)
1500 return lo;
1503 lo = loop_alloc(i);
1504 if (lo) {
1505 add_disk(lo->lo_disk);
1506 list_add_tail(&lo->lo_list, &loop_devices);
1508 return lo;
1511 static void loop_del_one(struct loop_device *lo)
1513 del_gendisk(lo->lo_disk);
1514 loop_free(lo);
1517 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1519 struct loop_device *lo;
1520 struct kobject *kobj;
1522 mutex_lock(&loop_devices_mutex);
1523 lo = loop_init_one(dev & MINORMASK);
1524 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1525 mutex_unlock(&loop_devices_mutex);
1527 *part = 0;
1528 return kobj;
1531 static int __init loop_init(void)
1533 int i, nr;
1534 unsigned long range;
1535 struct loop_device *lo, *next;
1538 * loop module now has a feature to instantiate underlying device
1539 * structure on-demand, provided that there is an access dev node.
1540 * However, this will not work well with user space tool that doesn't
1541 * know about such "feature". In order to not break any existing
1542 * tool, we do the following:
1544 * (1) if max_loop is specified, create that many upfront, and this
1545 * also becomes a hard limit.
1546 * (2) if max_loop is not specified, create 8 loop device on module
1547 * load, user can further extend loop device by create dev node
1548 * themselves and have kernel automatically instantiate actual
1549 * device on-demand.
1552 part_shift = 0;
1553 if (max_part > 0)
1554 part_shift = fls(max_part);
1556 if (max_loop > 1UL << (MINORBITS - part_shift))
1557 return -EINVAL;
1559 if (max_loop) {
1560 nr = max_loop;
1561 range = max_loop;
1562 } else {
1563 nr = 8;
1564 range = 1UL << (MINORBITS - part_shift);
1567 if (register_blkdev(LOOP_MAJOR, "loop"))
1568 return -EIO;
1570 for (i = 0; i < nr; i++) {
1571 lo = loop_alloc(i);
1572 if (!lo)
1573 goto Enomem;
1574 list_add_tail(&lo->lo_list, &loop_devices);
1577 /* point of no return */
1579 list_for_each_entry(lo, &loop_devices, lo_list)
1580 add_disk(lo->lo_disk);
1582 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1583 THIS_MODULE, loop_probe, NULL, NULL);
1585 printk(KERN_INFO "loop: module loaded\n");
1586 return 0;
1588 Enomem:
1589 printk(KERN_INFO "loop: out of memory\n");
1591 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1592 loop_free(lo);
1594 unregister_blkdev(LOOP_MAJOR, "loop");
1595 return -ENOMEM;
1598 static void __exit loop_exit(void)
1600 unsigned long range;
1601 struct loop_device *lo, *next;
1603 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1605 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1606 loop_del_one(lo);
1608 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1609 unregister_blkdev(LOOP_MAJOR, "loop");
1612 module_init(loop_init);
1613 module_exit(loop_exit);
1615 #ifndef MODULE
1616 static int __init max_loop_setup(char *str)
1618 max_loop = simple_strtol(str, NULL, 0);
1619 return 1;
1622 __setup("max_loop=", max_loop_setup);
1623 #endif