4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * linux/drivers/block/loop.c
40 * Written by Theodore Ts'o, 3/29/93
42 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
43 * permitted under the GNU General Public License.
45 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
46 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
48 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
50 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
52 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
54 * Loadable modules and other fixes by AK, 1998
56 * Maximum number of loop devices now dynamic via max_loop module parameter.
57 * Russell Kroll <rkroll@exploits.org> 19990701
59 * Maximum number of loop devices when compiled-in now selectable by passing
60 * max_loop=<1-255> to the kernel on boot.
61 * Erik I. Bols?, <eriki@himolde.no>, Oct 31, 1999
63 * Completely rewrite request handling to be make_request_fn style and
64 * non blocking, pushing work to a helper thread. Lots of fixes from
66 * Jens Axboe <axboe@suse.de>, Nov 2000
68 * Support up to 256 loop devices
69 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
71 * Support for falling back on the write file operation when the address space
72 * operations prepare_write and/or commit_write are not available on the
74 * Anton Altaparmakov, 16 Feb 2005
77 * - Advisory locking is ignored here.
78 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
82 #include <linux/module.h>
84 #include <linux/sched.h>
85 #include <linux/kthread.h>
87 #include <linux/file.h>
88 #include <linux/stat.h>
89 #include <linux/errno.h>
90 #include <linux/major.h>
91 #include <linux/wait.h>
92 #include <linux/blkdev.h>
93 #include <linux/blkpg.h>
94 #include <linux/init.h>
95 #include <linux/swap.h>
96 #include <linux/slab.h>
97 #include <linux/suspend.h>
98 #include <linux/writeback.h>
99 #include <linux/buffer_head.h> /* for invalidate_bdev() */
100 #include <linux/completion.h>
101 #include <linux/highmem.h>
102 #include <linux/gfp.h>
103 #include <linux/swap.h>
104 #include <linux/pagevec.h>
106 #include <asm/uaccess.h>
108 #include <lustre_lib.h>
109 #include "llite_internal.h"
111 #define LLOOP_MAX_SEGMENTS LNET_MAX_IOV
113 /* Possible states of device */
120 struct lloop_device {
126 struct file *lo_backing_file;
127 struct block_device *lo_device;
128 unsigned lo_blocksize;
134 struct bio *lo_biotail;
136 struct semaphore lo_sem;
137 struct mutex lo_ctl_mutex;
139 wait_queue_head_t lo_bh_wait;
141 struct request_queue *lo_queue;
143 const struct lu_env *lo_env;
145 struct ll_dio_pages lo_pvec;
147 /* data to handle bio for lustre. */
148 struct lo_request_data {
149 struct page *lrd_pages[LLOOP_MAX_SEGMENTS];
150 loff_t lrd_offsets[LLOOP_MAX_SEGMENTS];
158 LO_FLAGS_READ_ONLY = 1,
161 static int lloop_major;
162 #define MAX_LOOP_DEFAULT 16
163 static int max_loop = MAX_LOOP_DEFAULT;
164 static struct lloop_device *loop_dev;
165 static struct gendisk **disks;
166 static struct mutex lloop_mutex;
167 static void *ll_iocontrol_magic = NULL;
169 static loff_t get_loop_size(struct lloop_device *lo, struct file *file)
171 loff_t size, offset, loopsize;
173 /* Compute loopsize in bytes */
174 size = i_size_read(file->f_mapping->host);
175 offset = lo->lo_offset;
176 loopsize = size - offset;
177 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
178 loopsize = lo->lo_sizelimit;
181 * Unfortunately, if we want to do I/O on the device,
182 * the number of 512-byte sectors has to fit into a sector_t.
184 return loopsize >> 9;
187 static int do_bio_lustrebacked(struct lloop_device *lo, struct bio *head)
189 const struct lu_env *env = lo->lo_env;
190 struct cl_io *io = &lo->lo_io;
191 struct dentry *de = file_dentry(lo->lo_backing_file);
192 struct inode *inode = de->d_inode;
193 struct cl_object *obj = ll_i2info(inode)->lli_clob;
197 struct bio_vec *bvec;
199 size_t page_count = 0;
203 struct ll_dio_pages *pvec = &lo->lo_pvec;
204 struct page **pages = pvec->ldp_pages;
205 loff_t *offsets = pvec->ldp_offsets;
207 truncate_inode_pages(inode->i_mapping, 0);
209 /* initialize the IO */
210 memset(io, 0, sizeof(*io));
212 ret = cl_io_init(env, io, CIT_MISC, obj);
214 return io->ci_result;
215 io->ci_lockreq = CILR_NEVER;
217 LASSERT(head != NULL);
219 for (bio = head; bio != NULL; bio = bio->bi_next) {
220 LASSERT(rw == bio->bi_rw);
222 #ifdef HAVE_BVEC_ITER
223 offset = (pgoff_t)(bio->bi_iter.bi_sector << 9) + lo->lo_offset;
225 offset = (pgoff_t)(bio->bi_sector << 9) + lo->lo_offset;
227 bio_for_each_segment_all(bvec, bio, iter) {
228 BUG_ON(bvec->bv_offset != 0);
229 BUG_ON(bvec->bv_len != PAGE_CACHE_SIZE);
231 pages[page_count] = bvec->bv_page;
232 offsets[page_count] = offset;
234 offset += bvec->bv_len;
236 LASSERT(page_count <= LLOOP_MAX_SEGMENTS);
239 ll_stats_ops_tally(ll_i2sbi(inode),
240 (rw == WRITE) ? LPROC_LL_BRW_WRITE : LPROC_LL_BRW_READ,
243 pvec->ldp_size = page_count << PAGE_CACHE_SHIFT;
244 pvec->ldp_nr = page_count;
246 /* FIXME: in ll_direct_rw_pages, it has to allocate many cl_page{}s to
247 * write those pages into OST. Even worse case is that more pages
248 * would be asked to write out to swap space, and then finally get here
250 * Unfortunately this is NOT easy to fix.
251 * Thoughts on solution:
252 * 0. Define a reserved pool for cl_pages, which could be a list of
253 * pre-allocated cl_pages;
254 * 1. Define a new operation in cl_object_operations{}, says clo_depth,
255 * which measures how many layers for this lustre object. Generally
256 * speaking, the depth would be 2, one for llite, and one for lovsub.
257 * However, for SNS, there will be more since we need additional page
259 * 2. Reserve the # of (page_count * depth) cl_pages from the reserved
260 * pool. Afterwards, the clio would allocate the pages from reserved
261 * pool, this guarantees we neeedn't allocate the cl_pages from
262 * generic cl_page slab cache.
263 * Of course, if there is NOT enough pages in the pool, we might
264 * be asked to write less pages once, this purely depends on
265 * implementation. Anyway, we should be careful to avoid deadlocking.
268 bytes = ll_direct_rw_pages(env, io, rw, inode, pvec);
271 return (bytes == pvec->ldp_size) ? 0 : (int)bytes;
275 * Add bio to back of pending list
277 static void loop_add_bio(struct lloop_device *lo, struct bio *bio)
281 spin_lock_irqsave(&lo->lo_lock, flags);
282 if (lo->lo_biotail) {
283 lo->lo_biotail->bi_next = bio;
284 lo->lo_biotail = bio;
286 lo->lo_bio = lo->lo_biotail = bio;
287 spin_unlock_irqrestore(&lo->lo_lock, flags);
289 atomic_inc(&lo->lo_pending);
290 if (waitqueue_active(&lo->lo_bh_wait))
291 wake_up(&lo->lo_bh_wait);
295 * Grab first pending buffer
297 static unsigned int loop_get_bio(struct lloop_device *lo, struct bio **req)
301 unsigned int count = 0;
302 unsigned int page_count = 0;
305 spin_lock_irq(&lo->lo_lock);
307 if (unlikely(first == NULL)) {
308 spin_unlock_irq(&lo->lo_lock);
312 /* TODO: need to split the bio, too bad. */
313 LASSERT(first->bi_vcnt <= LLOOP_MAX_SEGMENTS);
317 while (*bio && (*bio)->bi_rw == rw) {
318 #ifdef HAVE_BVEC_ITER
319 CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
320 (unsigned long long)(*bio)->bi_iter.bi_sector,
321 (*bio)->bi_iter.bi_size, page_count, (*bio)->bi_vcnt);
323 CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
324 (unsigned long long)(*bio)->bi_sector, (*bio)->bi_size,
325 page_count, (*bio)->bi_vcnt);
327 if (page_count + (*bio)->bi_vcnt > LLOOP_MAX_SEGMENTS)
330 page_count += (*bio)->bi_vcnt;
332 bio = &(*bio)->bi_next;
335 /* Some of bios can't be mergable. */
339 /* Hit the end of queue */
340 lo->lo_biotail = NULL;
344 spin_unlock_irq(&lo->lo_lock);
349 loop_make_request(struct request_queue *q, struct bio *old_bio)
351 struct lloop_device *lo = q->queuedata;
352 int rw = bio_rw(old_bio);
358 #ifdef HAVE_BVEC_ITER
359 CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
360 (unsigned long long)old_bio->bi_iter.bi_sector,
361 old_bio->bi_iter.bi_size);
363 CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
364 (unsigned long long)old_bio->bi_sector, old_bio->bi_size);
367 spin_lock_irq(&lo->lo_lock);
368 inactive = (lo->lo_state != LLOOP_BOUND);
369 spin_unlock_irq(&lo->lo_lock);
374 if (lo->lo_flags & LO_FLAGS_READ_ONLY)
376 } else if (rw == READA) {
378 } else if (rw != READ) {
379 CERROR("lloop: unknown command (%x)\n", rw);
382 loop_add_bio(lo, old_bio);
385 bio_io_error(old_bio);
389 #ifdef HAVE_REQUEST_QUEUE_UNPLUG_FN
391 * kick off io on the underlying address space
393 static void loop_unplug(struct request_queue *q)
395 struct lloop_device *lo = q->queuedata;
397 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
398 blk_run_address_space(lo->lo_backing_file->f_mapping);
402 static inline void loop_handle_bio(struct lloop_device *lo, struct bio *bio)
406 ret = do_bio_lustrebacked(lo, bio);
408 struct bio *tmp = bio->bi_next;
411 #ifdef HAVE_BIO_ENDIO_USES_ONE_ARG
421 static inline int loop_active(struct lloop_device *lo)
423 return atomic_read(&lo->lo_pending) ||
424 (lo->lo_state == LLOOP_RUNDOWN);
428 * worker thread that handles reads/writes to file backed loop devices,
429 * to avoid blocking in our make_request_fn.
431 static int loop_thread(void *data)
433 struct lloop_device *lo = data;
436 unsigned long times = 0;
437 unsigned long total_count = 0;
443 set_user_nice(current, -20);
445 lo->lo_state = LLOOP_BOUND;
447 env = cl_env_get(&refcheck);
449 GOTO(out, ret = PTR_ERR(env));
452 memset(&lo->lo_pvec, 0, sizeof(lo->lo_pvec));
453 lo->lo_pvec.ldp_pages = lo->lo_requests[0].lrd_pages;
454 lo->lo_pvec.ldp_offsets = lo->lo_requests[0].lrd_offsets;
457 * up sem, we are running
462 wait_event(lo->lo_bh_wait, loop_active(lo));
463 if (!atomic_read(&lo->lo_pending)) {
465 spin_lock_irq(&lo->lo_lock);
466 exiting = (lo->lo_state == LLOOP_RUNDOWN);
467 spin_unlock_irq(&lo->lo_lock);
473 count = loop_get_bio(lo, &bio);
475 CWARN("lloop(minor: %d): missing bio\n", lo->lo_number);
479 total_count += count;
480 if (total_count < count) { /* overflow */
486 if ((times & 127) == 0) {
487 CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n",
488 total_count, times, total_count / times);
491 LASSERT(bio != NULL);
492 LASSERT(count <= atomic_read(&lo->lo_pending));
493 loop_handle_bio(lo, bio);
494 atomic_sub(count, &lo->lo_pending);
496 cl_env_put(env, &refcheck);
503 static int loop_set_fd(struct lloop_device *lo, struct file *unused,
504 struct block_device *bdev, struct file *file)
507 struct address_space *mapping;
512 if (!try_module_get(THIS_MODULE))
516 if (lo->lo_state != LLOOP_UNBOUND)
519 mapping = file->f_mapping;
520 inode = mapping->host;
523 if (!S_ISREG(inode->i_mode) || inode->i_sb->s_magic != LL_SUPER_MAGIC)
526 if (!(file->f_mode & FMODE_WRITE))
527 lo_flags |= LO_FLAGS_READ_ONLY;
529 size = get_loop_size(lo, file);
531 if ((loff_t)(sector_t)size != size) {
536 /* remove all pages in cache so as dirty pages not to be existent. */
537 truncate_inode_pages(mapping, 0);
539 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
541 lo->lo_blocksize = PAGE_CACHE_SIZE;
542 lo->lo_device = bdev;
543 lo->lo_flags = lo_flags;
544 lo->lo_backing_file = file;
545 lo->lo_sizelimit = 0;
546 lo->old_gfp_mask = mapping_gfp_mask(mapping);
547 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
549 lo->lo_bio = lo->lo_biotail = NULL;
552 * set queue make_request_fn, and add limits based on lower level
555 blk_queue_make_request(lo->lo_queue, loop_make_request);
556 lo->lo_queue->queuedata = lo;
557 #ifdef HAVE_REQUEST_QUEUE_UNPLUG_FN
558 lo->lo_queue->unplug_fn = loop_unplug;
561 /* queue parameters */
562 blk_queue_max_hw_sectors(lo->lo_queue,
563 LLOOP_MAX_SEGMENTS << (PAGE_CACHE_SHIFT - 9));
564 blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS);
566 set_capacity(disks[lo->lo_number], size);
567 bd_set_size(bdev, size << 9);
569 set_blocksize(bdev, lo->lo_blocksize);
571 kthread_run(loop_thread, lo, "lloop%d", lo->lo_number);
576 /* This is safe: open() is still holding a reference. */
577 module_put(THIS_MODULE);
581 static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev,
584 struct file *filp = lo->lo_backing_file;
585 gfp_t gfp = lo->old_gfp_mask;
587 if (lo->lo_state != LLOOP_BOUND)
590 if (lo->lo_refcnt > count) /* we needed one fd for the ioctl */
596 spin_lock_irq(&lo->lo_lock);
597 lo->lo_state = LLOOP_RUNDOWN;
598 spin_unlock_irq(&lo->lo_lock);
599 wake_up(&lo->lo_bh_wait);
602 lo->lo_backing_file = NULL;
603 lo->lo_device = NULL;
605 lo->lo_sizelimit = 0;
607 invalidate_bdev(bdev);
608 set_capacity(disks[lo->lo_number], 0);
609 bd_set_size(bdev, 0);
610 mapping_set_gfp_mask(filp->f_mapping, gfp);
611 lo->lo_state = LLOOP_UNBOUND;
613 /* This is safe: open() is still holding a reference. */
614 module_put(THIS_MODULE);
618 static int lo_open(struct block_device *bdev, fmode_t mode)
620 struct lloop_device *lo = bdev->bd_disk->private_data;
622 mutex_lock(&lo->lo_ctl_mutex);
624 mutex_unlock(&lo->lo_ctl_mutex);
629 #ifdef HAVE_BLKDEV_RELEASE_RETURN_INT
634 lo_release(struct gendisk *disk, fmode_t mode)
636 struct lloop_device *lo = disk->private_data;
638 mutex_lock(&lo->lo_ctl_mutex);
640 mutex_unlock(&lo->lo_ctl_mutex);
641 #ifdef HAVE_BLKDEV_RELEASE_RETURN_INT
646 /* lloop device node's ioctl function. */
647 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
648 unsigned int cmd, unsigned long arg)
650 struct lloop_device *lo = bdev->bd_disk->private_data;
653 mutex_lock(&lloop_mutex);
655 case LL_IOC_LLOOP_DETACH: {
656 err = loop_clr_fd(lo, bdev, 2);
658 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
662 case LL_IOC_LLOOP_INFO: {
666 if (lo->lo_backing_file == NULL) {
670 inode = file_inode(lo->lo_backing_file);
671 if (inode != NULL && lo->lo_state == LLOOP_BOUND)
672 fid = ll_i2info(inode)->lli_fid;
676 if (copy_to_user((struct lu_fid __user *)arg,
686 mutex_unlock(&lloop_mutex);
691 static struct block_device_operations lo_fops = {
692 .owner = THIS_MODULE,
694 .release = lo_release,
698 /* dynamic iocontrol callback.
699 * This callback is registered in lloop_init and will be called by
702 * This is a llite regular file ioctl function. It takes the responsibility
703 * of attaching or detaching a file by a lloop's device numner.
705 static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
706 unsigned int cmd, unsigned long arg,
707 void *magic, int *rcp)
709 struct lloop_device *lo = NULL;
710 struct block_device *bdev = NULL;
714 if (magic != ll_iocontrol_magic)
718 GOTO(out1, err = -ENODEV);
720 CWARN("Enter llop_ioctl\n");
722 mutex_lock(&lloop_mutex);
724 case LL_IOC_LLOOP_ATTACH: {
725 struct inode *inode = file_inode(file);
726 struct lloop_device *lo_free = NULL;
729 for (i = 0; i < max_loop; i++, lo = NULL) {
731 if (lo->lo_state == LLOOP_UNBOUND) {
736 if (file_inode(lo->lo_backing_file) == inode)
740 GOTO(out, err = -EBUSY);
743 dev = MKDEV(lloop_major, lo->lo_number);
745 /* quit if the used pointer is writable */
746 if (put_user((long)old_encode_dev(dev), (long __user *)arg))
747 GOTO(out, err = -EFAULT);
749 bdev = blkdev_get_by_dev(dev, file->f_mode, NULL);
751 GOTO(out, err = PTR_ERR(bdev));
754 err = loop_set_fd(lo, NULL, bdev, file);
763 case LL_IOC_LLOOP_DETACH_BYDEV: {
766 dev = old_decode_dev(arg);
767 if (MAJOR(dev) != lloop_major)
768 GOTO(out, err = -EINVAL);
771 if (minor > max_loop - 1)
772 GOTO(out, err = -EINVAL);
774 lo = &loop_dev[minor];
775 if (lo->lo_state != LLOOP_BOUND)
776 GOTO(out, err = -EINVAL);
778 bdev = lo->lo_device;
779 err = loop_clr_fd(lo, bdev, 1);
781 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
792 mutex_unlock(&lloop_mutex);
799 static int __init lloop_init(void)
802 unsigned int cmdlist[] = {
804 LL_IOC_LLOOP_DETACH_BYDEV,
807 if (max_loop < 1 || max_loop > 256) {
808 max_loop = MAX_LOOP_DEFAULT;
809 CWARN("lloop: invalid max_loop (must be between"
810 " 1 and 256), using default (%u)\n", max_loop);
813 lloop_major = register_blkdev(0, "lloop");
817 CDEBUG(D_CONFIG, "registered lloop major %d with %u minors\n",
818 lloop_major, max_loop);
820 ll_iocontrol_magic = ll_iocontrol_register(lloop_ioctl, 2, cmdlist);
821 if (ll_iocontrol_magic == NULL)
824 OBD_ALLOC_WAIT(loop_dev, max_loop * sizeof(*loop_dev));
828 OBD_ALLOC_WAIT(disks, max_loop * sizeof(*disks));
832 for (i = 0; i < max_loop; i++) {
833 disks[i] = alloc_disk(1);
838 mutex_init(&lloop_mutex);
840 for (i = 0; i < max_loop; i++) {
841 struct lloop_device *lo = &loop_dev[i];
842 struct gendisk *disk = disks[i];
844 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
848 mutex_init(&lo->lo_ctl_mutex);
849 sema_init(&lo->lo_sem, 0);
850 init_waitqueue_head(&lo->lo_bh_wait);
852 spin_lock_init(&lo->lo_lock);
853 disk->major = lloop_major;
854 disk->first_minor = i;
855 disk->fops = &lo_fops;
856 sprintf(disk->disk_name, "lloop%d", i);
857 disk->private_data = lo;
858 disk->queue = lo->lo_queue;
861 /* We cannot fail after we call this, so another loop!*/
862 for (i = 0; i < max_loop; i++)
868 blk_cleanup_queue(loop_dev[i].lo_queue);
873 OBD_FREE(disks, max_loop * sizeof(*disks));
875 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
877 unregister_blkdev(lloop_major, "lloop");
878 ll_iocontrol_unregister(ll_iocontrol_magic);
879 CERROR("lloop: ran out of memory\n");
883 static void __exit lloop_exit(void)
887 ll_iocontrol_unregister(ll_iocontrol_magic);
888 for (i = 0; i < max_loop; i++) {
889 del_gendisk(disks[i]);
890 blk_cleanup_queue(loop_dev[i].lo_queue);
893 unregister_blkdev(lloop_major, "lloop");
895 OBD_FREE(disks, max_loop * sizeof(*disks));
896 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
899 module_param(max_loop, int, 0444);
900 MODULE_PARM_DESC(max_loop, "maximum of lloop_device");
901 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
902 MODULE_DESCRIPTION("Lustre virtual block device");
903 MODULE_VERSION(LUSTRE_VERSION_STRING);
904 MODULE_LICENSE("GPL");
906 module_init(lloop_init);
907 module_exit(lloop_exit);