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, 2012, 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>
86 #include <linux/file.h>
87 #include <linux/stat.h>
88 #include <linux/errno.h>
89 #include <linux/major.h>
90 #include <linux/wait.h>
91 #include <linux/blkdev.h>
92 #include <linux/blkpg.h>
93 #include <linux/init.h>
94 #include <linux/swap.h>
95 #include <linux/slab.h>
96 #include <linux/suspend.h>
97 #include <linux/writeback.h>
98 #include <linux/buffer_head.h> /* for invalidate_bdev() */
99 #include <linux/completion.h>
100 #include <linux/highmem.h>
101 #include <linux/gfp.h>
102 #include <linux/pagevec.h>
104 #include <asm/uaccess.h>
106 #include "../include/lustre_lib.h"
107 #include "../include/lustre_lite.h"
108 #include "llite_internal.h"
110 #define LLOOP_MAX_SEGMENTS LNET_MAX_IOV
112 /* Possible states of device */
119 struct lloop_device {
125 struct file *lo_backing_file;
126 struct block_device *lo_device;
127 unsigned lo_blocksize;
133 struct bio *lo_biotail;
135 struct semaphore lo_sem;
136 struct mutex lo_ctl_mutex;
138 wait_queue_head_t lo_bh_wait;
140 struct request_queue *lo_queue;
142 const struct lu_env *lo_env;
144 struct ll_dio_pages lo_pvec;
146 /* data to handle bio for lustre. */
147 struct lo_request_data {
148 struct page *lrd_pages[LLOOP_MAX_SEGMENTS];
149 loff_t lrd_offsets[LLOOP_MAX_SEGMENTS];
157 LO_FLAGS_READ_ONLY = 1,
160 static int lloop_major;
161 #define MAX_LOOP_DEFAULT 16
162 static int max_loop = MAX_LOOP_DEFAULT;
163 static struct lloop_device *loop_dev;
164 static struct gendisk **disks;
165 static struct mutex lloop_mutex;
166 static void *ll_iocontrol_magic = NULL;
168 static loff_t get_loop_size(struct lloop_device *lo, struct file *file)
170 loff_t size, offset, loopsize;
172 /* Compute loopsize in bytes */
173 size = i_size_read(file->f_mapping->host);
174 offset = lo->lo_offset;
175 loopsize = size - offset;
176 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
177 loopsize = lo->lo_sizelimit;
180 * Unfortunately, if we want to do I/O on the device,
181 * the number of 512-byte sectors has to fit into a sector_t.
183 return loopsize >> 9;
186 static int do_bio_lustrebacked(struct lloop_device *lo, struct bio *head)
188 const struct lu_env *env = lo->lo_env;
189 struct cl_io *io = &lo->lo_io;
190 struct inode *inode = lo->lo_backing_file->f_dentry->d_inode;
191 struct cl_object *obj = ll_i2info(inode)->lli_clob;
197 struct bvec_iter iter;
201 struct ll_dio_pages *pvec = &lo->lo_pvec;
202 struct page **pages = pvec->ldp_pages;
203 loff_t *offsets = pvec->ldp_offsets;
205 truncate_inode_pages(inode->i_mapping, 0);
207 /* initialize the IO */
208 memset(io, 0, sizeof(*io));
210 ret = cl_io_init(env, io, CIT_MISC, obj);
212 return io->ci_result;
213 io->ci_lockreq = CILR_NEVER;
215 LASSERT(head != NULL);
217 for (bio = head; bio != NULL; bio = bio->bi_next) {
218 LASSERT(rw == bio->bi_rw);
220 offset = (pgoff_t)(bio->bi_iter.bi_sector << 9) + lo->lo_offset;
221 bio_for_each_segment(bvec, bio, iter) {
222 BUG_ON(bvec.bv_offset != 0);
223 BUG_ON(bvec.bv_len != PAGE_CACHE_SIZE);
225 pages[page_count] = bvec.bv_page;
226 offsets[page_count] = offset;
228 offset += bvec.bv_len;
230 LASSERT(page_count <= LLOOP_MAX_SEGMENTS);
233 ll_stats_ops_tally(ll_i2sbi(inode),
234 (rw == WRITE) ? LPROC_LL_BRW_WRITE : LPROC_LL_BRW_READ,
237 pvec->ldp_size = page_count << PAGE_CACHE_SHIFT;
238 pvec->ldp_nr = page_count;
240 /* FIXME: in ll_direct_rw_pages, it has to allocate many cl_page{}s to
241 * write those pages into OST. Even worse case is that more pages
242 * would be asked to write out to swap space, and then finally get here
244 * Unfortunately this is NOT easy to fix.
245 * Thoughts on solution:
246 * 0. Define a reserved pool for cl_pages, which could be a list of
247 * pre-allocated cl_pages;
248 * 1. Define a new operation in cl_object_operations{}, says clo_depth,
249 * which measures how many layers for this lustre object. Generally
250 * speaking, the depth would be 2, one for llite, and one for lovsub.
251 * However, for SNS, there will be more since we need additional page
253 * 2. Reserve the # of (page_count * depth) cl_pages from the reserved
254 * pool. Afterwards, the clio would allocate the pages from reserved
255 * pool, this guarantees we needn't allocate the cl_pages from
256 * generic cl_page slab cache.
257 * Of course, if there is NOT enough pages in the pool, we might
258 * be asked to write less pages once, this purely depends on
259 * implementation. Anyway, we should be careful to avoid deadlocking.
261 mutex_lock(&inode->i_mutex);
262 bytes = ll_direct_rw_pages(env, io, rw, inode, pvec);
263 mutex_unlock(&inode->i_mutex);
265 return (bytes == pvec->ldp_size) ? 0 : (int)bytes;
269 * Add bio to back of pending list
271 static void loop_add_bio(struct lloop_device *lo, struct bio *bio)
275 spin_lock_irqsave(&lo->lo_lock, flags);
276 if (lo->lo_biotail) {
277 lo->lo_biotail->bi_next = bio;
278 lo->lo_biotail = bio;
280 lo->lo_bio = lo->lo_biotail = bio;
281 spin_unlock_irqrestore(&lo->lo_lock, flags);
283 atomic_inc(&lo->lo_pending);
284 if (waitqueue_active(&lo->lo_bh_wait))
285 wake_up(&lo->lo_bh_wait);
289 * Grab first pending buffer
291 static unsigned int loop_get_bio(struct lloop_device *lo, struct bio **req)
295 unsigned int count = 0;
296 unsigned int page_count = 0;
299 spin_lock_irq(&lo->lo_lock);
301 if (unlikely(first == NULL)) {
302 spin_unlock_irq(&lo->lo_lock);
306 /* TODO: need to split the bio, too bad. */
307 LASSERT(first->bi_vcnt <= LLOOP_MAX_SEGMENTS);
311 while (*bio && (*bio)->bi_rw == rw) {
312 CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
313 (unsigned long long)(*bio)->bi_iter.bi_sector,
314 (*bio)->bi_iter.bi_size,
315 page_count, (*bio)->bi_vcnt);
316 if (page_count + (*bio)->bi_vcnt > LLOOP_MAX_SEGMENTS)
320 page_count += (*bio)->bi_vcnt;
322 bio = &(*bio)->bi_next;
325 /* Some of bios can't be mergeable. */
329 /* Hit the end of queue */
330 lo->lo_biotail = NULL;
334 spin_unlock_irq(&lo->lo_lock);
338 static void loop_make_request(struct request_queue *q, struct bio *old_bio)
340 struct lloop_device *lo = q->queuedata;
341 int rw = bio_rw(old_bio);
347 CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
348 (unsigned long long)old_bio->bi_iter.bi_sector,
349 old_bio->bi_iter.bi_size);
351 spin_lock_irq(&lo->lo_lock);
352 inactive = (lo->lo_state != LLOOP_BOUND);
353 spin_unlock_irq(&lo->lo_lock);
358 if (lo->lo_flags & LO_FLAGS_READ_ONLY)
360 } else if (rw == READA) {
362 } else if (rw != READ) {
363 CERROR("lloop: unknown command (%x)\n", rw);
366 loop_add_bio(lo, old_bio);
369 cfs_bio_io_error(old_bio, old_bio->bi_iter.bi_size);
373 static inline void loop_handle_bio(struct lloop_device *lo, struct bio *bio)
376 ret = do_bio_lustrebacked(lo, bio);
378 struct bio *tmp = bio->bi_next;
380 cfs_bio_endio(bio, bio->bi_iter.bi_size, ret);
385 static inline int loop_active(struct lloop_device *lo)
387 return atomic_read(&lo->lo_pending) ||
388 (lo->lo_state == LLOOP_RUNDOWN);
392 * worker thread that handles reads/writes to file backed loop devices,
393 * to avoid blocking in our make_request_fn.
395 static int loop_thread(void *data)
397 struct lloop_device *lo = data;
400 unsigned long times = 0;
401 unsigned long total_count = 0;
407 set_user_nice(current, MIN_NICE);
409 lo->lo_state = LLOOP_BOUND;
411 env = cl_env_get(&refcheck);
418 memset(&lo->lo_pvec, 0, sizeof(lo->lo_pvec));
419 lo->lo_pvec.ldp_pages = lo->lo_requests[0].lrd_pages;
420 lo->lo_pvec.ldp_offsets = lo->lo_requests[0].lrd_offsets;
423 * up sem, we are running
428 wait_event(lo->lo_bh_wait, loop_active(lo));
429 if (!atomic_read(&lo->lo_pending)) {
431 spin_lock_irq(&lo->lo_lock);
432 exiting = (lo->lo_state == LLOOP_RUNDOWN);
433 spin_unlock_irq(&lo->lo_lock);
439 count = loop_get_bio(lo, &bio);
441 CWARN("lloop(minor: %d): missing bio\n", lo->lo_number);
445 total_count += count;
446 if (total_count < count) { /* overflow */
452 if ((times & 127) == 0) {
453 CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n",
454 total_count, times, total_count / times);
457 LASSERT(bio != NULL);
458 LASSERT(count <= atomic_read(&lo->lo_pending));
459 loop_handle_bio(lo, bio);
460 atomic_sub(count, &lo->lo_pending);
462 cl_env_put(env, &refcheck);
469 static int loop_set_fd(struct lloop_device *lo, struct file *unused,
470 struct block_device *bdev, struct file *file)
473 struct address_space *mapping;
478 if (!try_module_get(THIS_MODULE))
482 if (lo->lo_state != LLOOP_UNBOUND)
485 mapping = file->f_mapping;
486 inode = mapping->host;
489 if (!S_ISREG(inode->i_mode) || inode->i_sb->s_magic != LL_SUPER_MAGIC)
492 if (!(file->f_mode & FMODE_WRITE))
493 lo_flags |= LO_FLAGS_READ_ONLY;
495 size = get_loop_size(lo, file);
497 if ((loff_t)(sector_t)size != size) {
502 /* remove all pages in cache so as dirty pages not to be existent. */
503 truncate_inode_pages(mapping, 0);
505 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
507 lo->lo_blocksize = PAGE_CACHE_SIZE;
508 lo->lo_device = bdev;
509 lo->lo_flags = lo_flags;
510 lo->lo_backing_file = file;
511 lo->lo_sizelimit = 0;
512 lo->old_gfp_mask = mapping_gfp_mask(mapping);
513 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
515 lo->lo_bio = lo->lo_biotail = NULL;
518 * set queue make_request_fn, and add limits based on lower level
521 blk_queue_make_request(lo->lo_queue, loop_make_request);
522 lo->lo_queue->queuedata = lo;
524 /* queue parameters */
525 CLASSERT(PAGE_CACHE_SIZE < (1 << (sizeof(unsigned short) * 8)));
526 blk_queue_logical_block_size(lo->lo_queue,
527 (unsigned short)PAGE_CACHE_SIZE);
528 blk_queue_max_hw_sectors(lo->lo_queue,
529 LLOOP_MAX_SEGMENTS << (PAGE_CACHE_SHIFT - 9));
530 blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS);
532 set_capacity(disks[lo->lo_number], size);
533 bd_set_size(bdev, size << 9);
535 set_blocksize(bdev, lo->lo_blocksize);
537 kthread_run(loop_thread, lo, "lloop%d", lo->lo_number);
542 /* This is safe: open() is still holding a reference. */
543 module_put(THIS_MODULE);
547 static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev,
550 struct file *filp = lo->lo_backing_file;
551 gfp_t gfp = lo->old_gfp_mask;
553 if (lo->lo_state != LLOOP_BOUND)
556 if (lo->lo_refcnt > count) /* we needed one fd for the ioctl */
562 spin_lock_irq(&lo->lo_lock);
563 lo->lo_state = LLOOP_RUNDOWN;
564 spin_unlock_irq(&lo->lo_lock);
565 wake_up(&lo->lo_bh_wait);
568 lo->lo_backing_file = NULL;
569 lo->lo_device = NULL;
571 lo->lo_sizelimit = 0;
573 invalidate_bdev(bdev);
574 set_capacity(disks[lo->lo_number], 0);
575 bd_set_size(bdev, 0);
576 mapping_set_gfp_mask(filp->f_mapping, gfp);
577 lo->lo_state = LLOOP_UNBOUND;
579 /* This is safe: open() is still holding a reference. */
580 module_put(THIS_MODULE);
584 static int lo_open(struct block_device *bdev, fmode_t mode)
586 struct lloop_device *lo = bdev->bd_disk->private_data;
588 mutex_lock(&lo->lo_ctl_mutex);
590 mutex_unlock(&lo->lo_ctl_mutex);
595 static void lo_release(struct gendisk *disk, fmode_t mode)
597 struct lloop_device *lo = disk->private_data;
599 mutex_lock(&lo->lo_ctl_mutex);
601 mutex_unlock(&lo->lo_ctl_mutex);
604 /* lloop device node's ioctl function. */
605 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
606 unsigned int cmd, unsigned long arg)
608 struct lloop_device *lo = bdev->bd_disk->private_data;
609 struct inode *inode = NULL;
612 mutex_lock(&lloop_mutex);
614 case LL_IOC_LLOOP_DETACH: {
615 err = loop_clr_fd(lo, bdev, 2);
617 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
621 case LL_IOC_LLOOP_INFO: {
624 if (lo->lo_backing_file == NULL) {
629 inode = lo->lo_backing_file->f_dentry->d_inode;
630 if (lo->lo_state == LLOOP_BOUND)
631 fid = ll_i2info(inode)->lli_fid;
635 if (copy_to_user((struct lu_fid *)arg, &fid, sizeof(fid)))
644 mutex_unlock(&lloop_mutex);
649 static struct block_device_operations lo_fops = {
650 .owner = THIS_MODULE,
652 .release = lo_release,
656 /* dynamic iocontrol callback.
657 * This callback is registered in lloop_init and will be called by
660 * This is a llite regular file ioctl function. It takes the responsibility
661 * of attaching or detaching a file by a lloop's device number.
663 static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
664 unsigned int cmd, unsigned long arg,
665 void *magic, int *rcp)
667 struct lloop_device *lo = NULL;
668 struct block_device *bdev = NULL;
672 if (magic != ll_iocontrol_magic)
680 CWARN("Enter llop_ioctl\n");
682 mutex_lock(&lloop_mutex);
684 case LL_IOC_LLOOP_ATTACH: {
685 struct lloop_device *lo_free = NULL;
688 for (i = 0; i < max_loop; i++, lo = NULL) {
690 if (lo->lo_state == LLOOP_UNBOUND) {
695 if (lo->lo_backing_file->f_dentry->d_inode ==
696 file->f_dentry->d_inode)
699 if (lo || !lo_free) {
705 dev = MKDEV(lloop_major, lo->lo_number);
707 /* quit if the used pointer is writable */
708 if (put_user((long)old_encode_dev(dev), (long *)arg)) {
713 bdev = blkdev_get_by_dev(dev, file->f_mode, NULL);
720 err = loop_set_fd(lo, NULL, bdev, file);
729 case LL_IOC_LLOOP_DETACH_BYDEV: {
732 dev = old_decode_dev(arg);
733 if (MAJOR(dev) != lloop_major) {
739 if (minor > max_loop - 1) {
744 lo = &loop_dev[minor];
745 if (lo->lo_state != LLOOP_BOUND) {
750 bdev = lo->lo_device;
751 err = loop_clr_fd(lo, bdev, 1);
753 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
764 mutex_unlock(&lloop_mutex);
771 static int __init lloop_init(void)
774 unsigned int cmdlist[] = {
776 LL_IOC_LLOOP_DETACH_BYDEV,
779 if (max_loop < 1 || max_loop > 256) {
780 max_loop = MAX_LOOP_DEFAULT;
781 CWARN("lloop: invalid max_loop (must be between"
782 " 1 and 256), using default (%u)\n", max_loop);
785 lloop_major = register_blkdev(0, "lloop");
789 CDEBUG(D_CONFIG, "registered lloop major %d with %u minors\n",
790 lloop_major, max_loop);
792 ll_iocontrol_magic = ll_iocontrol_register(lloop_ioctl, 2, cmdlist);
793 if (ll_iocontrol_magic == NULL)
796 loop_dev = kzalloc(max_loop * sizeof(*loop_dev), GFP_KERNEL);
800 disks = kzalloc(max_loop * sizeof(*disks), GFP_KERNEL);
804 for (i = 0; i < max_loop; i++) {
805 disks[i] = alloc_disk(1);
810 mutex_init(&lloop_mutex);
812 for (i = 0; i < max_loop; i++) {
813 struct lloop_device *lo = &loop_dev[i];
814 struct gendisk *disk = disks[i];
816 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
820 mutex_init(&lo->lo_ctl_mutex);
821 sema_init(&lo->lo_sem, 0);
822 init_waitqueue_head(&lo->lo_bh_wait);
824 spin_lock_init(&lo->lo_lock);
825 disk->major = lloop_major;
826 disk->first_minor = i;
827 disk->fops = &lo_fops;
828 sprintf(disk->disk_name, "lloop%d", i);
829 disk->private_data = lo;
830 disk->queue = lo->lo_queue;
833 /* We cannot fail after we call this, so another loop!*/
834 for (i = 0; i < max_loop; i++)
840 blk_cleanup_queue(loop_dev[i].lo_queue);
845 OBD_FREE(disks, max_loop * sizeof(*disks));
847 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
849 unregister_blkdev(lloop_major, "lloop");
850 ll_iocontrol_unregister(ll_iocontrol_magic);
851 CERROR("lloop: ran out of memory\n");
855 static void lloop_exit(void)
859 ll_iocontrol_unregister(ll_iocontrol_magic);
860 for (i = 0; i < max_loop; i++) {
861 del_gendisk(disks[i]);
862 blk_cleanup_queue(loop_dev[i].lo_queue);
866 unregister_blkdev(lloop_major, "lloop");
868 OBD_FREE(disks, max_loop * sizeof(*disks));
869 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
872 module_init(lloop_init);
873 module_exit(lloop_exit);
875 module_param(max_loop, int, 0444);
876 MODULE_PARM_DESC(max_loop, "maximum of lloop_device");
877 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
878 MODULE_DESCRIPTION("Lustre virtual block device");
879 MODULE_LICENSE("GPL");