xref: /linux/drivers/md/dm-log-writes.c (revision f4ad317aedf836e7bb67108a119bfec7f3d97ef1)
10e9cebe7SJosef Bacik /*
20e9cebe7SJosef Bacik  * Copyright (C) 2014 Facebook. All rights reserved.
30e9cebe7SJosef Bacik  *
40e9cebe7SJosef Bacik  * This file is released under the GPL.
50e9cebe7SJosef Bacik  */
60e9cebe7SJosef Bacik 
70e9cebe7SJosef Bacik #include <linux/device-mapper.h>
80e9cebe7SJosef Bacik 
90e9cebe7SJosef Bacik #include <linux/module.h>
100e9cebe7SJosef Bacik #include <linux/init.h>
110e9cebe7SJosef Bacik #include <linux/blkdev.h>
120e9cebe7SJosef Bacik #include <linux/bio.h>
130e9cebe7SJosef Bacik #include <linux/slab.h>
140e9cebe7SJosef Bacik #include <linux/kthread.h>
150e9cebe7SJosef Bacik #include <linux/freezer.h>
160e9cebe7SJosef Bacik 
170e9cebe7SJosef Bacik #define DM_MSG_PREFIX "log-writes"
180e9cebe7SJosef Bacik 
190e9cebe7SJosef Bacik /*
200e9cebe7SJosef Bacik  * This target will sequentially log all writes to the target device onto the
210e9cebe7SJosef Bacik  * log device.  This is helpful for replaying writes to check for fs consistency
220e9cebe7SJosef Bacik  * at all times.  This target provides a mechanism to mark specific events to
230e9cebe7SJosef Bacik  * check data at a later time.  So for example you would:
240e9cebe7SJosef Bacik  *
250e9cebe7SJosef Bacik  * write data
260e9cebe7SJosef Bacik  * fsync
270e9cebe7SJosef Bacik  * dmsetup message /dev/whatever mark mymark
280e9cebe7SJosef Bacik  * unmount /mnt/test
290e9cebe7SJosef Bacik  *
300e9cebe7SJosef Bacik  * Then replay the log up to mymark and check the contents of the replay to
310e9cebe7SJosef Bacik  * verify it matches what was written.
320e9cebe7SJosef Bacik  *
330e9cebe7SJosef Bacik  * We log writes only after they have been flushed, this makes the log describe
340e9cebe7SJosef Bacik  * close to the order in which the data hits the actual disk, not its cache.  So
350e9cebe7SJosef Bacik  * for example the following sequence (W means write, C means complete)
360e9cebe7SJosef Bacik  *
370e9cebe7SJosef Bacik  * Wa,Wb,Wc,Cc,Ca,FLUSH,FUAd,Cb,CFLUSH,CFUAd
380e9cebe7SJosef Bacik  *
390e9cebe7SJosef Bacik  * Would result in the log looking like this:
400e9cebe7SJosef Bacik  *
410e9cebe7SJosef Bacik  * c,a,flush,fuad,b,<other writes>,<next flush>
420e9cebe7SJosef Bacik  *
430e9cebe7SJosef Bacik  * This is meant to help expose problems where file systems do not properly wait
440e9cebe7SJosef Bacik  * on data being written before invoking a FLUSH.  FUA bypasses cache so once it
450e9cebe7SJosef Bacik  * completes it is added to the log as it should be on disk.
460e9cebe7SJosef Bacik  *
470e9cebe7SJosef Bacik  * We treat DISCARDs as if they don't bypass cache so that they are logged in
480e9cebe7SJosef Bacik  * order of completion along with the normal writes.  If we didn't do it this
490e9cebe7SJosef Bacik  * way we would process all the discards first and then write all the data, when
500e9cebe7SJosef Bacik  * in fact we want to do the data and the discard in the order that they
510e9cebe7SJosef Bacik  * completed.
520e9cebe7SJosef Bacik  */
530e9cebe7SJosef Bacik #define LOG_FLUSH_FLAG (1 << 0)
540e9cebe7SJosef Bacik #define LOG_FUA_FLAG (1 << 1)
550e9cebe7SJosef Bacik #define LOG_DISCARD_FLAG (1 << 2)
560e9cebe7SJosef Bacik #define LOG_MARK_FLAG (1 << 3)
570e9cebe7SJosef Bacik 
58*f4ad317aSGeert Uytterhoeven #define WRITE_LOG_VERSION 1ULL
59*f4ad317aSGeert Uytterhoeven #define WRITE_LOG_MAGIC 0x6a736677736872ULL
600e9cebe7SJosef Bacik 
610e9cebe7SJosef Bacik /*
620e9cebe7SJosef Bacik  * The disk format for this is braindead simple.
630e9cebe7SJosef Bacik  *
640e9cebe7SJosef Bacik  * At byte 0 we have our super, followed by the following sequence for
650e9cebe7SJosef Bacik  * nr_entries:
660e9cebe7SJosef Bacik  *
670e9cebe7SJosef Bacik  * [   1 sector    ][  entry->nr_sectors ]
680e9cebe7SJosef Bacik  * [log_write_entry][    data written    ]
690e9cebe7SJosef Bacik  *
700e9cebe7SJosef Bacik  * The log_write_entry takes up a full sector so we can have arbitrary length
710e9cebe7SJosef Bacik  * marks and it leaves us room for extra content in the future.
720e9cebe7SJosef Bacik  */
730e9cebe7SJosef Bacik 
740e9cebe7SJosef Bacik /*
750e9cebe7SJosef Bacik  * Basic info about the log for userspace.
760e9cebe7SJosef Bacik  */
770e9cebe7SJosef Bacik struct log_write_super {
780e9cebe7SJosef Bacik 	__le64 magic;
790e9cebe7SJosef Bacik 	__le64 version;
800e9cebe7SJosef Bacik 	__le64 nr_entries;
810e9cebe7SJosef Bacik 	__le32 sectorsize;
820e9cebe7SJosef Bacik };
830e9cebe7SJosef Bacik 
840e9cebe7SJosef Bacik /*
850e9cebe7SJosef Bacik  * sector - the sector we wrote.
860e9cebe7SJosef Bacik  * nr_sectors - the number of sectors we wrote.
870e9cebe7SJosef Bacik  * flags - flags for this log entry.
880e9cebe7SJosef Bacik  * data_len - the size of the data in this log entry, this is for private log
890e9cebe7SJosef Bacik  * entry stuff, the MARK data provided by userspace for example.
900e9cebe7SJosef Bacik  */
910e9cebe7SJosef Bacik struct log_write_entry {
920e9cebe7SJosef Bacik 	__le64 sector;
930e9cebe7SJosef Bacik 	__le64 nr_sectors;
940e9cebe7SJosef Bacik 	__le64 flags;
950e9cebe7SJosef Bacik 	__le64 data_len;
960e9cebe7SJosef Bacik };
970e9cebe7SJosef Bacik 
980e9cebe7SJosef Bacik struct log_writes_c {
990e9cebe7SJosef Bacik 	struct dm_dev *dev;
1000e9cebe7SJosef Bacik 	struct dm_dev *logdev;
1010e9cebe7SJosef Bacik 	u64 logged_entries;
1020e9cebe7SJosef Bacik 	u32 sectorsize;
1030e9cebe7SJosef Bacik 	atomic_t io_blocks;
1040e9cebe7SJosef Bacik 	atomic_t pending_blocks;
1050e9cebe7SJosef Bacik 	sector_t next_sector;
1060e9cebe7SJosef Bacik 	sector_t end_sector;
1070e9cebe7SJosef Bacik 	bool logging_enabled;
1080e9cebe7SJosef Bacik 	bool device_supports_discard;
1090e9cebe7SJosef Bacik 	spinlock_t blocks_lock;
1100e9cebe7SJosef Bacik 	struct list_head unflushed_blocks;
1110e9cebe7SJosef Bacik 	struct list_head logging_blocks;
1120e9cebe7SJosef Bacik 	wait_queue_head_t wait;
1130e9cebe7SJosef Bacik 	struct task_struct *log_kthread;
1140e9cebe7SJosef Bacik };
1150e9cebe7SJosef Bacik 
1160e9cebe7SJosef Bacik struct pending_block {
1170e9cebe7SJosef Bacik 	int vec_cnt;
1180e9cebe7SJosef Bacik 	u64 flags;
1190e9cebe7SJosef Bacik 	sector_t sector;
1200e9cebe7SJosef Bacik 	sector_t nr_sectors;
1210e9cebe7SJosef Bacik 	char *data;
1220e9cebe7SJosef Bacik 	u32 datalen;
1230e9cebe7SJosef Bacik 	struct list_head list;
1240e9cebe7SJosef Bacik 	struct bio_vec vecs[0];
1250e9cebe7SJosef Bacik };
1260e9cebe7SJosef Bacik 
1270e9cebe7SJosef Bacik struct per_bio_data {
1280e9cebe7SJosef Bacik 	struct pending_block *block;
1290e9cebe7SJosef Bacik };
1300e9cebe7SJosef Bacik 
1310e9cebe7SJosef Bacik static void put_pending_block(struct log_writes_c *lc)
1320e9cebe7SJosef Bacik {
1330e9cebe7SJosef Bacik 	if (atomic_dec_and_test(&lc->pending_blocks)) {
1340e9cebe7SJosef Bacik 		smp_mb__after_atomic();
1350e9cebe7SJosef Bacik 		if (waitqueue_active(&lc->wait))
1360e9cebe7SJosef Bacik 			wake_up(&lc->wait);
1370e9cebe7SJosef Bacik 	}
1380e9cebe7SJosef Bacik }
1390e9cebe7SJosef Bacik 
1400e9cebe7SJosef Bacik static void put_io_block(struct log_writes_c *lc)
1410e9cebe7SJosef Bacik {
1420e9cebe7SJosef Bacik 	if (atomic_dec_and_test(&lc->io_blocks)) {
1430e9cebe7SJosef Bacik 		smp_mb__after_atomic();
1440e9cebe7SJosef Bacik 		if (waitqueue_active(&lc->wait))
1450e9cebe7SJosef Bacik 			wake_up(&lc->wait);
1460e9cebe7SJosef Bacik 	}
1470e9cebe7SJosef Bacik }
1480e9cebe7SJosef Bacik 
1490e9cebe7SJosef Bacik static void log_end_io(struct bio *bio, int err)
1500e9cebe7SJosef Bacik {
1510e9cebe7SJosef Bacik 	struct log_writes_c *lc = bio->bi_private;
1520e9cebe7SJosef Bacik 	struct bio_vec *bvec;
1530e9cebe7SJosef Bacik 	int i;
1540e9cebe7SJosef Bacik 
1550e9cebe7SJosef Bacik 	if (err) {
1560e9cebe7SJosef Bacik 		unsigned long flags;
1570e9cebe7SJosef Bacik 
1580e9cebe7SJosef Bacik 		DMERR("Error writing log block, error=%d", err);
1590e9cebe7SJosef Bacik 		spin_lock_irqsave(&lc->blocks_lock, flags);
1600e9cebe7SJosef Bacik 		lc->logging_enabled = false;
1610e9cebe7SJosef Bacik 		spin_unlock_irqrestore(&lc->blocks_lock, flags);
1620e9cebe7SJosef Bacik 	}
1630e9cebe7SJosef Bacik 
1640e9cebe7SJosef Bacik 	bio_for_each_segment_all(bvec, bio, i)
1650e9cebe7SJosef Bacik 		__free_page(bvec->bv_page);
1660e9cebe7SJosef Bacik 
1670e9cebe7SJosef Bacik 	put_io_block(lc);
1680e9cebe7SJosef Bacik 	bio_put(bio);
1690e9cebe7SJosef Bacik }
1700e9cebe7SJosef Bacik 
1710e9cebe7SJosef Bacik /*
1720e9cebe7SJosef Bacik  * Meant to be called if there is an error, it will free all the pages
1730e9cebe7SJosef Bacik  * associated with the block.
1740e9cebe7SJosef Bacik  */
1750e9cebe7SJosef Bacik static void free_pending_block(struct log_writes_c *lc,
1760e9cebe7SJosef Bacik 			       struct pending_block *block)
1770e9cebe7SJosef Bacik {
1780e9cebe7SJosef Bacik 	int i;
1790e9cebe7SJosef Bacik 
1800e9cebe7SJosef Bacik 	for (i = 0; i < block->vec_cnt; i++) {
1810e9cebe7SJosef Bacik 		if (block->vecs[i].bv_page)
1820e9cebe7SJosef Bacik 			__free_page(block->vecs[i].bv_page);
1830e9cebe7SJosef Bacik 	}
1840e9cebe7SJosef Bacik 	kfree(block->data);
1850e9cebe7SJosef Bacik 	kfree(block);
1860e9cebe7SJosef Bacik 	put_pending_block(lc);
1870e9cebe7SJosef Bacik }
1880e9cebe7SJosef Bacik 
1890e9cebe7SJosef Bacik static int write_metadata(struct log_writes_c *lc, void *entry,
1900e9cebe7SJosef Bacik 			  size_t entrylen, void *data, size_t datalen,
1910e9cebe7SJosef Bacik 			  sector_t sector)
1920e9cebe7SJosef Bacik {
1930e9cebe7SJosef Bacik 	struct bio *bio;
1940e9cebe7SJosef Bacik 	struct page *page;
1950e9cebe7SJosef Bacik 	void *ptr;
1960e9cebe7SJosef Bacik 	size_t ret;
1970e9cebe7SJosef Bacik 
1980e9cebe7SJosef Bacik 	bio = bio_alloc(GFP_KERNEL, 1);
1990e9cebe7SJosef Bacik 	if (!bio) {
2000e9cebe7SJosef Bacik 		DMERR("Couldn't alloc log bio");
2010e9cebe7SJosef Bacik 		goto error;
2020e9cebe7SJosef Bacik 	}
2030e9cebe7SJosef Bacik 	bio->bi_iter.bi_size = 0;
2040e9cebe7SJosef Bacik 	bio->bi_iter.bi_sector = sector;
2050e9cebe7SJosef Bacik 	bio->bi_bdev = lc->logdev->bdev;
2060e9cebe7SJosef Bacik 	bio->bi_end_io = log_end_io;
2070e9cebe7SJosef Bacik 	bio->bi_private = lc;
2080e9cebe7SJosef Bacik 	set_bit(BIO_UPTODATE, &bio->bi_flags);
2090e9cebe7SJosef Bacik 
2100e9cebe7SJosef Bacik 	page = alloc_page(GFP_KERNEL);
2110e9cebe7SJosef Bacik 	if (!page) {
2120e9cebe7SJosef Bacik 		DMERR("Couldn't alloc log page");
2130e9cebe7SJosef Bacik 		bio_put(bio);
2140e9cebe7SJosef Bacik 		goto error;
2150e9cebe7SJosef Bacik 	}
2160e9cebe7SJosef Bacik 
2170e9cebe7SJosef Bacik 	ptr = kmap_atomic(page);
2180e9cebe7SJosef Bacik 	memcpy(ptr, entry, entrylen);
2190e9cebe7SJosef Bacik 	if (datalen)
2200e9cebe7SJosef Bacik 		memcpy(ptr + entrylen, data, datalen);
2210e9cebe7SJosef Bacik 	memset(ptr + entrylen + datalen, 0,
2220e9cebe7SJosef Bacik 	       lc->sectorsize - entrylen - datalen);
2230e9cebe7SJosef Bacik 	kunmap_atomic(ptr);
2240e9cebe7SJosef Bacik 
2250e9cebe7SJosef Bacik 	ret = bio_add_page(bio, page, lc->sectorsize, 0);
2260e9cebe7SJosef Bacik 	if (ret != lc->sectorsize) {
2270e9cebe7SJosef Bacik 		DMERR("Couldn't add page to the log block");
2280e9cebe7SJosef Bacik 		goto error_bio;
2290e9cebe7SJosef Bacik 	}
2300e9cebe7SJosef Bacik 	submit_bio(WRITE, bio);
2310e9cebe7SJosef Bacik 	return 0;
2320e9cebe7SJosef Bacik error_bio:
2330e9cebe7SJosef Bacik 	bio_put(bio);
2340e9cebe7SJosef Bacik 	__free_page(page);
2350e9cebe7SJosef Bacik error:
2360e9cebe7SJosef Bacik 	put_io_block(lc);
2370e9cebe7SJosef Bacik 	return -1;
2380e9cebe7SJosef Bacik }
2390e9cebe7SJosef Bacik 
2400e9cebe7SJosef Bacik static int log_one_block(struct log_writes_c *lc,
2410e9cebe7SJosef Bacik 			 struct pending_block *block, sector_t sector)
2420e9cebe7SJosef Bacik {
2430e9cebe7SJosef Bacik 	struct bio *bio;
2440e9cebe7SJosef Bacik 	struct log_write_entry entry;
2450e9cebe7SJosef Bacik 	size_t ret;
2460e9cebe7SJosef Bacik 	int i;
2470e9cebe7SJosef Bacik 
2480e9cebe7SJosef Bacik 	entry.sector = cpu_to_le64(block->sector);
2490e9cebe7SJosef Bacik 	entry.nr_sectors = cpu_to_le64(block->nr_sectors);
2500e9cebe7SJosef Bacik 	entry.flags = cpu_to_le64(block->flags);
2510e9cebe7SJosef Bacik 	entry.data_len = cpu_to_le64(block->datalen);
2520e9cebe7SJosef Bacik 	if (write_metadata(lc, &entry, sizeof(entry), block->data,
2530e9cebe7SJosef Bacik 			   block->datalen, sector)) {
2540e9cebe7SJosef Bacik 		free_pending_block(lc, block);
2550e9cebe7SJosef Bacik 		return -1;
2560e9cebe7SJosef Bacik 	}
2570e9cebe7SJosef Bacik 
2580e9cebe7SJosef Bacik 	if (!block->vec_cnt)
2590e9cebe7SJosef Bacik 		goto out;
2600e9cebe7SJosef Bacik 	sector++;
2610e9cebe7SJosef Bacik 
2620e9cebe7SJosef Bacik 	bio = bio_alloc(GFP_KERNEL, block->vec_cnt);
2630e9cebe7SJosef Bacik 	if (!bio) {
2640e9cebe7SJosef Bacik 		DMERR("Couldn't alloc log bio");
2650e9cebe7SJosef Bacik 		goto error;
2660e9cebe7SJosef Bacik 	}
2670e9cebe7SJosef Bacik 	atomic_inc(&lc->io_blocks);
2680e9cebe7SJosef Bacik 	bio->bi_iter.bi_size = 0;
2690e9cebe7SJosef Bacik 	bio->bi_iter.bi_sector = sector;
2700e9cebe7SJosef Bacik 	bio->bi_bdev = lc->logdev->bdev;
2710e9cebe7SJosef Bacik 	bio->bi_end_io = log_end_io;
2720e9cebe7SJosef Bacik 	bio->bi_private = lc;
2730e9cebe7SJosef Bacik 	set_bit(BIO_UPTODATE, &bio->bi_flags);
2740e9cebe7SJosef Bacik 
2750e9cebe7SJosef Bacik 	for (i = 0; i < block->vec_cnt; i++) {
2760e9cebe7SJosef Bacik 		/*
2770e9cebe7SJosef Bacik 		 * The page offset is always 0 because we allocate a new page
2780e9cebe7SJosef Bacik 		 * for every bvec in the original bio for simplicity sake.
2790e9cebe7SJosef Bacik 		 */
2800e9cebe7SJosef Bacik 		ret = bio_add_page(bio, block->vecs[i].bv_page,
2810e9cebe7SJosef Bacik 				   block->vecs[i].bv_len, 0);
2820e9cebe7SJosef Bacik 		if (ret != block->vecs[i].bv_len) {
2830e9cebe7SJosef Bacik 			atomic_inc(&lc->io_blocks);
2840e9cebe7SJosef Bacik 			submit_bio(WRITE, bio);
2850e9cebe7SJosef Bacik 			bio = bio_alloc(GFP_KERNEL, block->vec_cnt - i);
2860e9cebe7SJosef Bacik 			if (!bio) {
2870e9cebe7SJosef Bacik 				DMERR("Couldn't alloc log bio");
2880e9cebe7SJosef Bacik 				goto error;
2890e9cebe7SJosef Bacik 			}
2900e9cebe7SJosef Bacik 			bio->bi_iter.bi_size = 0;
2910e9cebe7SJosef Bacik 			bio->bi_iter.bi_sector = sector;
2920e9cebe7SJosef Bacik 			bio->bi_bdev = lc->logdev->bdev;
2930e9cebe7SJosef Bacik 			bio->bi_end_io = log_end_io;
2940e9cebe7SJosef Bacik 			bio->bi_private = lc;
2950e9cebe7SJosef Bacik 			set_bit(BIO_UPTODATE, &bio->bi_flags);
2960e9cebe7SJosef Bacik 
2970e9cebe7SJosef Bacik 			ret = bio_add_page(bio, block->vecs[i].bv_page,
2980e9cebe7SJosef Bacik 					   block->vecs[i].bv_len, 0);
2990e9cebe7SJosef Bacik 			if (ret != block->vecs[i].bv_len) {
3000e9cebe7SJosef Bacik 				DMERR("Couldn't add page on new bio?");
3010e9cebe7SJosef Bacik 				bio_put(bio);
3020e9cebe7SJosef Bacik 				goto error;
3030e9cebe7SJosef Bacik 			}
3040e9cebe7SJosef Bacik 		}
3050e9cebe7SJosef Bacik 		sector += block->vecs[i].bv_len >> SECTOR_SHIFT;
3060e9cebe7SJosef Bacik 	}
3070e9cebe7SJosef Bacik 	submit_bio(WRITE, bio);
3080e9cebe7SJosef Bacik out:
3090e9cebe7SJosef Bacik 	kfree(block->data);
3100e9cebe7SJosef Bacik 	kfree(block);
3110e9cebe7SJosef Bacik 	put_pending_block(lc);
3120e9cebe7SJosef Bacik 	return 0;
3130e9cebe7SJosef Bacik error:
3140e9cebe7SJosef Bacik 	free_pending_block(lc, block);
3150e9cebe7SJosef Bacik 	put_io_block(lc);
3160e9cebe7SJosef Bacik 	return -1;
3170e9cebe7SJosef Bacik }
3180e9cebe7SJosef Bacik 
3190e9cebe7SJosef Bacik static int log_super(struct log_writes_c *lc)
3200e9cebe7SJosef Bacik {
3210e9cebe7SJosef Bacik 	struct log_write_super super;
3220e9cebe7SJosef Bacik 
3230e9cebe7SJosef Bacik 	super.magic = cpu_to_le64(WRITE_LOG_MAGIC);
3240e9cebe7SJosef Bacik 	super.version = cpu_to_le64(WRITE_LOG_VERSION);
3250e9cebe7SJosef Bacik 	super.nr_entries = cpu_to_le64(lc->logged_entries);
3260e9cebe7SJosef Bacik 	super.sectorsize = cpu_to_le32(lc->sectorsize);
3270e9cebe7SJosef Bacik 
3280e9cebe7SJosef Bacik 	if (write_metadata(lc, &super, sizeof(super), NULL, 0, 0)) {
3290e9cebe7SJosef Bacik 		DMERR("Couldn't write super");
3300e9cebe7SJosef Bacik 		return -1;
3310e9cebe7SJosef Bacik 	}
3320e9cebe7SJosef Bacik 
3330e9cebe7SJosef Bacik 	return 0;
3340e9cebe7SJosef Bacik }
3350e9cebe7SJosef Bacik 
3360e9cebe7SJosef Bacik static inline sector_t logdev_last_sector(struct log_writes_c *lc)
3370e9cebe7SJosef Bacik {
3380e9cebe7SJosef Bacik 	return i_size_read(lc->logdev->bdev->bd_inode) >> SECTOR_SHIFT;
3390e9cebe7SJosef Bacik }
3400e9cebe7SJosef Bacik 
3410e9cebe7SJosef Bacik static int log_writes_kthread(void *arg)
3420e9cebe7SJosef Bacik {
3430e9cebe7SJosef Bacik 	struct log_writes_c *lc = (struct log_writes_c *)arg;
3440e9cebe7SJosef Bacik 	sector_t sector = 0;
3450e9cebe7SJosef Bacik 
3460e9cebe7SJosef Bacik 	while (!kthread_should_stop()) {
3470e9cebe7SJosef Bacik 		bool super = false;
3480e9cebe7SJosef Bacik 		bool logging_enabled;
3490e9cebe7SJosef Bacik 		struct pending_block *block = NULL;
3500e9cebe7SJosef Bacik 		int ret;
3510e9cebe7SJosef Bacik 
3520e9cebe7SJosef Bacik 		spin_lock_irq(&lc->blocks_lock);
3530e9cebe7SJosef Bacik 		if (!list_empty(&lc->logging_blocks)) {
3540e9cebe7SJosef Bacik 			block = list_first_entry(&lc->logging_blocks,
3550e9cebe7SJosef Bacik 						 struct pending_block, list);
3560e9cebe7SJosef Bacik 			list_del_init(&block->list);
3570e9cebe7SJosef Bacik 			if (!lc->logging_enabled)
3580e9cebe7SJosef Bacik 				goto next;
3590e9cebe7SJosef Bacik 
3600e9cebe7SJosef Bacik 			sector = lc->next_sector;
3610e9cebe7SJosef Bacik 			if (block->flags & LOG_DISCARD_FLAG)
3620e9cebe7SJosef Bacik 				lc->next_sector++;
3630e9cebe7SJosef Bacik 			else
3640e9cebe7SJosef Bacik 				lc->next_sector += block->nr_sectors + 1;
3650e9cebe7SJosef Bacik 
3660e9cebe7SJosef Bacik 			/*
3670e9cebe7SJosef Bacik 			 * Apparently the size of the device may not be known
3680e9cebe7SJosef Bacik 			 * right away, so handle this properly.
3690e9cebe7SJosef Bacik 			 */
3700e9cebe7SJosef Bacik 			if (!lc->end_sector)
3710e9cebe7SJosef Bacik 				lc->end_sector = logdev_last_sector(lc);
3720e9cebe7SJosef Bacik 			if (lc->end_sector &&
3730e9cebe7SJosef Bacik 			    lc->next_sector >= lc->end_sector) {
3740e9cebe7SJosef Bacik 				DMERR("Ran out of space on the logdev");
3750e9cebe7SJosef Bacik 				lc->logging_enabled = false;
3760e9cebe7SJosef Bacik 				goto next;
3770e9cebe7SJosef Bacik 			}
3780e9cebe7SJosef Bacik 			lc->logged_entries++;
3790e9cebe7SJosef Bacik 			atomic_inc(&lc->io_blocks);
3800e9cebe7SJosef Bacik 
3810e9cebe7SJosef Bacik 			super = (block->flags & (LOG_FUA_FLAG | LOG_MARK_FLAG));
3820e9cebe7SJosef Bacik 			if (super)
3830e9cebe7SJosef Bacik 				atomic_inc(&lc->io_blocks);
3840e9cebe7SJosef Bacik 		}
3850e9cebe7SJosef Bacik next:
3860e9cebe7SJosef Bacik 		logging_enabled = lc->logging_enabled;
3870e9cebe7SJosef Bacik 		spin_unlock_irq(&lc->blocks_lock);
3880e9cebe7SJosef Bacik 		if (block) {
3890e9cebe7SJosef Bacik 			if (logging_enabled) {
3900e9cebe7SJosef Bacik 				ret = log_one_block(lc, block, sector);
3910e9cebe7SJosef Bacik 				if (!ret && super)
3920e9cebe7SJosef Bacik 					ret = log_super(lc);
3930e9cebe7SJosef Bacik 				if (ret) {
3940e9cebe7SJosef Bacik 					spin_lock_irq(&lc->blocks_lock);
3950e9cebe7SJosef Bacik 					lc->logging_enabled = false;
3960e9cebe7SJosef Bacik 					spin_unlock_irq(&lc->blocks_lock);
3970e9cebe7SJosef Bacik 				}
3980e9cebe7SJosef Bacik 			} else
3990e9cebe7SJosef Bacik 				free_pending_block(lc, block);
4000e9cebe7SJosef Bacik 			continue;
4010e9cebe7SJosef Bacik 		}
4020e9cebe7SJosef Bacik 
4030e9cebe7SJosef Bacik 		if (!try_to_freeze()) {
4040e9cebe7SJosef Bacik 			set_current_state(TASK_INTERRUPTIBLE);
4050e9cebe7SJosef Bacik 			if (!kthread_should_stop() &&
4060e9cebe7SJosef Bacik 			    !atomic_read(&lc->pending_blocks))
4070e9cebe7SJosef Bacik 				schedule();
4080e9cebe7SJosef Bacik 			__set_current_state(TASK_RUNNING);
4090e9cebe7SJosef Bacik 		}
4100e9cebe7SJosef Bacik 	}
4110e9cebe7SJosef Bacik 	return 0;
4120e9cebe7SJosef Bacik }
4130e9cebe7SJosef Bacik 
4140e9cebe7SJosef Bacik /*
4150e9cebe7SJosef Bacik  * Construct a log-writes mapping:
4160e9cebe7SJosef Bacik  * log-writes <dev_path> <log_dev_path>
4170e9cebe7SJosef Bacik  */
4180e9cebe7SJosef Bacik static int log_writes_ctr(struct dm_target *ti, unsigned int argc, char **argv)
4190e9cebe7SJosef Bacik {
4200e9cebe7SJosef Bacik 	struct log_writes_c *lc;
4210e9cebe7SJosef Bacik 	struct dm_arg_set as;
4220e9cebe7SJosef Bacik 	const char *devname, *logdevname;
4230e9cebe7SJosef Bacik 
4240e9cebe7SJosef Bacik 	as.argc = argc;
4250e9cebe7SJosef Bacik 	as.argv = argv;
4260e9cebe7SJosef Bacik 
4270e9cebe7SJosef Bacik 	if (argc < 2) {
4280e9cebe7SJosef Bacik 		ti->error = "Invalid argument count";
4290e9cebe7SJosef Bacik 		return -EINVAL;
4300e9cebe7SJosef Bacik 	}
4310e9cebe7SJosef Bacik 
4320e9cebe7SJosef Bacik 	lc = kzalloc(sizeof(struct log_writes_c), GFP_KERNEL);
4330e9cebe7SJosef Bacik 	if (!lc) {
4340e9cebe7SJosef Bacik 		ti->error = "Cannot allocate context";
4350e9cebe7SJosef Bacik 		return -ENOMEM;
4360e9cebe7SJosef Bacik 	}
4370e9cebe7SJosef Bacik 	spin_lock_init(&lc->blocks_lock);
4380e9cebe7SJosef Bacik 	INIT_LIST_HEAD(&lc->unflushed_blocks);
4390e9cebe7SJosef Bacik 	INIT_LIST_HEAD(&lc->logging_blocks);
4400e9cebe7SJosef Bacik 	init_waitqueue_head(&lc->wait);
4410e9cebe7SJosef Bacik 	lc->sectorsize = 1 << SECTOR_SHIFT;
4420e9cebe7SJosef Bacik 	atomic_set(&lc->io_blocks, 0);
4430e9cebe7SJosef Bacik 	atomic_set(&lc->pending_blocks, 0);
4440e9cebe7SJosef Bacik 
4450e9cebe7SJosef Bacik 	devname = dm_shift_arg(&as);
4460e9cebe7SJosef Bacik 	if (dm_get_device(ti, devname, dm_table_get_mode(ti->table), &lc->dev)) {
4470e9cebe7SJosef Bacik 		ti->error = "Device lookup failed";
4480e9cebe7SJosef Bacik 		goto bad;
4490e9cebe7SJosef Bacik 	}
4500e9cebe7SJosef Bacik 
4510e9cebe7SJosef Bacik 	logdevname = dm_shift_arg(&as);
4520e9cebe7SJosef Bacik 	if (dm_get_device(ti, logdevname, dm_table_get_mode(ti->table), &lc->logdev)) {
4530e9cebe7SJosef Bacik 		ti->error = "Log device lookup failed";
4540e9cebe7SJosef Bacik 		dm_put_device(ti, lc->dev);
4550e9cebe7SJosef Bacik 		goto bad;
4560e9cebe7SJosef Bacik 	}
4570e9cebe7SJosef Bacik 
4580e9cebe7SJosef Bacik 	lc->log_kthread = kthread_run(log_writes_kthread, lc, "log-write");
4590e9cebe7SJosef Bacik 	if (!lc->log_kthread) {
4600e9cebe7SJosef Bacik 		ti->error = "Couldn't alloc kthread";
4610e9cebe7SJosef Bacik 		dm_put_device(ti, lc->dev);
4620e9cebe7SJosef Bacik 		dm_put_device(ti, lc->logdev);
4630e9cebe7SJosef Bacik 		goto bad;
4640e9cebe7SJosef Bacik 	}
4650e9cebe7SJosef Bacik 
4660e9cebe7SJosef Bacik 	/* We put the super at sector 0, start logging at sector 1 */
4670e9cebe7SJosef Bacik 	lc->next_sector = 1;
4680e9cebe7SJosef Bacik 	lc->logging_enabled = true;
4690e9cebe7SJosef Bacik 	lc->end_sector = logdev_last_sector(lc);
4700e9cebe7SJosef Bacik 	lc->device_supports_discard = true;
4710e9cebe7SJosef Bacik 
4720e9cebe7SJosef Bacik 	ti->num_flush_bios = 1;
4730e9cebe7SJosef Bacik 	ti->flush_supported = true;
4740e9cebe7SJosef Bacik 	ti->num_discard_bios = 1;
4750e9cebe7SJosef Bacik 	ti->discards_supported = true;
4760e9cebe7SJosef Bacik 	ti->per_bio_data_size = sizeof(struct per_bio_data);
4770e9cebe7SJosef Bacik 	ti->private = lc;
4780e9cebe7SJosef Bacik 	return 0;
4790e9cebe7SJosef Bacik 
4800e9cebe7SJosef Bacik bad:
4810e9cebe7SJosef Bacik 	kfree(lc);
4820e9cebe7SJosef Bacik 	return -EINVAL;
4830e9cebe7SJosef Bacik }
4840e9cebe7SJosef Bacik 
4850e9cebe7SJosef Bacik static int log_mark(struct log_writes_c *lc, char *data)
4860e9cebe7SJosef Bacik {
4870e9cebe7SJosef Bacik 	struct pending_block *block;
4880e9cebe7SJosef Bacik 	size_t maxsize = lc->sectorsize - sizeof(struct log_write_entry);
4890e9cebe7SJosef Bacik 
4900e9cebe7SJosef Bacik 	block = kzalloc(sizeof(struct pending_block), GFP_KERNEL);
4910e9cebe7SJosef Bacik 	if (!block) {
4920e9cebe7SJosef Bacik 		DMERR("Error allocating pending block");
4930e9cebe7SJosef Bacik 		return -ENOMEM;
4940e9cebe7SJosef Bacik 	}
4950e9cebe7SJosef Bacik 
4960e9cebe7SJosef Bacik 	block->data = kstrndup(data, maxsize, GFP_KERNEL);
4970e9cebe7SJosef Bacik 	if (!block->data) {
4980e9cebe7SJosef Bacik 		DMERR("Error copying mark data");
4990e9cebe7SJosef Bacik 		kfree(block);
5000e9cebe7SJosef Bacik 		return -ENOMEM;
5010e9cebe7SJosef Bacik 	}
5020e9cebe7SJosef Bacik 	atomic_inc(&lc->pending_blocks);
5030e9cebe7SJosef Bacik 	block->datalen = strlen(block->data);
5040e9cebe7SJosef Bacik 	block->flags |= LOG_MARK_FLAG;
5050e9cebe7SJosef Bacik 	spin_lock_irq(&lc->blocks_lock);
5060e9cebe7SJosef Bacik 	list_add_tail(&block->list, &lc->logging_blocks);
5070e9cebe7SJosef Bacik 	spin_unlock_irq(&lc->blocks_lock);
5080e9cebe7SJosef Bacik 	wake_up_process(lc->log_kthread);
5090e9cebe7SJosef Bacik 	return 0;
5100e9cebe7SJosef Bacik }
5110e9cebe7SJosef Bacik 
5120e9cebe7SJosef Bacik static void log_writes_dtr(struct dm_target *ti)
5130e9cebe7SJosef Bacik {
5140e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
5150e9cebe7SJosef Bacik 
5160e9cebe7SJosef Bacik 	spin_lock_irq(&lc->blocks_lock);
5170e9cebe7SJosef Bacik 	list_splice_init(&lc->unflushed_blocks, &lc->logging_blocks);
5180e9cebe7SJosef Bacik 	spin_unlock_irq(&lc->blocks_lock);
5190e9cebe7SJosef Bacik 
5200e9cebe7SJosef Bacik 	/*
5210e9cebe7SJosef Bacik 	 * This is just nice to have since it'll update the super to include the
5220e9cebe7SJosef Bacik 	 * unflushed blocks, if it fails we don't really care.
5230e9cebe7SJosef Bacik 	 */
5240e9cebe7SJosef Bacik 	log_mark(lc, "dm-log-writes-end");
5250e9cebe7SJosef Bacik 	wake_up_process(lc->log_kthread);
5260e9cebe7SJosef Bacik 	wait_event(lc->wait, !atomic_read(&lc->io_blocks) &&
5270e9cebe7SJosef Bacik 		   !atomic_read(&lc->pending_blocks));
5280e9cebe7SJosef Bacik 	kthread_stop(lc->log_kthread);
5290e9cebe7SJosef Bacik 
5300e9cebe7SJosef Bacik 	WARN_ON(!list_empty(&lc->logging_blocks));
5310e9cebe7SJosef Bacik 	WARN_ON(!list_empty(&lc->unflushed_blocks));
5320e9cebe7SJosef Bacik 	dm_put_device(ti, lc->dev);
5330e9cebe7SJosef Bacik 	dm_put_device(ti, lc->logdev);
5340e9cebe7SJosef Bacik 	kfree(lc);
5350e9cebe7SJosef Bacik }
5360e9cebe7SJosef Bacik 
5370e9cebe7SJosef Bacik static void normal_map_bio(struct dm_target *ti, struct bio *bio)
5380e9cebe7SJosef Bacik {
5390e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
5400e9cebe7SJosef Bacik 
5410e9cebe7SJosef Bacik 	bio->bi_bdev = lc->dev->bdev;
5420e9cebe7SJosef Bacik }
5430e9cebe7SJosef Bacik 
5440e9cebe7SJosef Bacik static int log_writes_map(struct dm_target *ti, struct bio *bio)
5450e9cebe7SJosef Bacik {
5460e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
5470e9cebe7SJosef Bacik 	struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
5480e9cebe7SJosef Bacik 	struct pending_block *block;
5490e9cebe7SJosef Bacik 	struct bvec_iter iter;
5500e9cebe7SJosef Bacik 	struct bio_vec bv;
5510e9cebe7SJosef Bacik 	size_t alloc_size;
5520e9cebe7SJosef Bacik 	int i = 0;
5530e9cebe7SJosef Bacik 	bool flush_bio = (bio->bi_rw & REQ_FLUSH);
5540e9cebe7SJosef Bacik 	bool fua_bio = (bio->bi_rw & REQ_FUA);
5550e9cebe7SJosef Bacik 	bool discard_bio = (bio->bi_rw & REQ_DISCARD);
5560e9cebe7SJosef Bacik 
5570e9cebe7SJosef Bacik 	pb->block = NULL;
5580e9cebe7SJosef Bacik 
5590e9cebe7SJosef Bacik 	/* Don't bother doing anything if logging has been disabled */
5600e9cebe7SJosef Bacik 	if (!lc->logging_enabled)
5610e9cebe7SJosef Bacik 		goto map_bio;
5620e9cebe7SJosef Bacik 
5630e9cebe7SJosef Bacik 	/*
5640e9cebe7SJosef Bacik 	 * Map reads as normal.
5650e9cebe7SJosef Bacik 	 */
5660e9cebe7SJosef Bacik 	if (bio_data_dir(bio) == READ)
5670e9cebe7SJosef Bacik 		goto map_bio;
5680e9cebe7SJosef Bacik 
5690e9cebe7SJosef Bacik 	/* No sectors and not a flush?  Don't care */
5700e9cebe7SJosef Bacik 	if (!bio_sectors(bio) && !flush_bio)
5710e9cebe7SJosef Bacik 		goto map_bio;
5720e9cebe7SJosef Bacik 
5730e9cebe7SJosef Bacik 	/*
5740e9cebe7SJosef Bacik 	 * Discards will have bi_size set but there's no actual data, so just
5750e9cebe7SJosef Bacik 	 * allocate the size of the pending block.
5760e9cebe7SJosef Bacik 	 */
5770e9cebe7SJosef Bacik 	if (discard_bio)
5780e9cebe7SJosef Bacik 		alloc_size = sizeof(struct pending_block);
5790e9cebe7SJosef Bacik 	else
5800e9cebe7SJosef Bacik 		alloc_size = sizeof(struct pending_block) + sizeof(struct bio_vec) * bio_segments(bio);
5810e9cebe7SJosef Bacik 
5820e9cebe7SJosef Bacik 	block = kzalloc(alloc_size, GFP_NOIO);
5830e9cebe7SJosef Bacik 	if (!block) {
5840e9cebe7SJosef Bacik 		DMERR("Error allocating pending block");
5850e9cebe7SJosef Bacik 		spin_lock_irq(&lc->blocks_lock);
5860e9cebe7SJosef Bacik 		lc->logging_enabled = false;
5870e9cebe7SJosef Bacik 		spin_unlock_irq(&lc->blocks_lock);
5880e9cebe7SJosef Bacik 		return -ENOMEM;
5890e9cebe7SJosef Bacik 	}
5900e9cebe7SJosef Bacik 	INIT_LIST_HEAD(&block->list);
5910e9cebe7SJosef Bacik 	pb->block = block;
5920e9cebe7SJosef Bacik 	atomic_inc(&lc->pending_blocks);
5930e9cebe7SJosef Bacik 
5940e9cebe7SJosef Bacik 	if (flush_bio)
5950e9cebe7SJosef Bacik 		block->flags |= LOG_FLUSH_FLAG;
5960e9cebe7SJosef Bacik 	if (fua_bio)
5970e9cebe7SJosef Bacik 		block->flags |= LOG_FUA_FLAG;
5980e9cebe7SJosef Bacik 	if (discard_bio)
5990e9cebe7SJosef Bacik 		block->flags |= LOG_DISCARD_FLAG;
6000e9cebe7SJosef Bacik 
6010e9cebe7SJosef Bacik 	block->sector = bio->bi_iter.bi_sector;
6020e9cebe7SJosef Bacik 	block->nr_sectors = bio_sectors(bio);
6030e9cebe7SJosef Bacik 
6040e9cebe7SJosef Bacik 	/* We don't need the data, just submit */
6050e9cebe7SJosef Bacik 	if (discard_bio) {
6060e9cebe7SJosef Bacik 		WARN_ON(flush_bio || fua_bio);
6070e9cebe7SJosef Bacik 		if (lc->device_supports_discard)
6080e9cebe7SJosef Bacik 			goto map_bio;
6090e9cebe7SJosef Bacik 		bio_endio(bio, 0);
6100e9cebe7SJosef Bacik 		return DM_MAPIO_SUBMITTED;
6110e9cebe7SJosef Bacik 	}
6120e9cebe7SJosef Bacik 
6130e9cebe7SJosef Bacik 	/* Flush bio, splice the unflushed blocks onto this list and submit */
6140e9cebe7SJosef Bacik 	if (flush_bio && !bio_sectors(bio)) {
6150e9cebe7SJosef Bacik 		spin_lock_irq(&lc->blocks_lock);
6160e9cebe7SJosef Bacik 		list_splice_init(&lc->unflushed_blocks, &block->list);
6170e9cebe7SJosef Bacik 		spin_unlock_irq(&lc->blocks_lock);
6180e9cebe7SJosef Bacik 		goto map_bio;
6190e9cebe7SJosef Bacik 	}
6200e9cebe7SJosef Bacik 
6210e9cebe7SJosef Bacik 	/*
6220e9cebe7SJosef Bacik 	 * We will write this bio somewhere else way later so we need to copy
6230e9cebe7SJosef Bacik 	 * the actual contents into new pages so we know the data will always be
6240e9cebe7SJosef Bacik 	 * there.
6250e9cebe7SJosef Bacik 	 *
6260e9cebe7SJosef Bacik 	 * We do this because this could be a bio from O_DIRECT in which case we
6270e9cebe7SJosef Bacik 	 * can't just hold onto the page until some later point, we have to
6280e9cebe7SJosef Bacik 	 * manually copy the contents.
6290e9cebe7SJosef Bacik 	 */
6300e9cebe7SJosef Bacik 	bio_for_each_segment(bv, bio, iter) {
6310e9cebe7SJosef Bacik 		struct page *page;
6320e9cebe7SJosef Bacik 		void *src, *dst;
6330e9cebe7SJosef Bacik 
6340e9cebe7SJosef Bacik 		page = alloc_page(GFP_NOIO);
6350e9cebe7SJosef Bacik 		if (!page) {
6360e9cebe7SJosef Bacik 			DMERR("Error allocing page");
6370e9cebe7SJosef Bacik 			free_pending_block(lc, block);
6380e9cebe7SJosef Bacik 			spin_lock_irq(&lc->blocks_lock);
6390e9cebe7SJosef Bacik 			lc->logging_enabled = false;
6400e9cebe7SJosef Bacik 			spin_unlock_irq(&lc->blocks_lock);
6410e9cebe7SJosef Bacik 			return -ENOMEM;
6420e9cebe7SJosef Bacik 		}
6430e9cebe7SJosef Bacik 
6440e9cebe7SJosef Bacik 		src = kmap_atomic(bv.bv_page);
6450e9cebe7SJosef Bacik 		dst = kmap_atomic(page);
6460e9cebe7SJosef Bacik 		memcpy(dst, src + bv.bv_offset, bv.bv_len);
6470e9cebe7SJosef Bacik 		kunmap_atomic(dst);
6480e9cebe7SJosef Bacik 		kunmap_atomic(src);
6490e9cebe7SJosef Bacik 		block->vecs[i].bv_page = page;
6500e9cebe7SJosef Bacik 		block->vecs[i].bv_len = bv.bv_len;
6510e9cebe7SJosef Bacik 		block->vec_cnt++;
6520e9cebe7SJosef Bacik 		i++;
6530e9cebe7SJosef Bacik 	}
6540e9cebe7SJosef Bacik 
6550e9cebe7SJosef Bacik 	/* Had a flush with data in it, weird */
6560e9cebe7SJosef Bacik 	if (flush_bio) {
6570e9cebe7SJosef Bacik 		spin_lock_irq(&lc->blocks_lock);
6580e9cebe7SJosef Bacik 		list_splice_init(&lc->unflushed_blocks, &block->list);
6590e9cebe7SJosef Bacik 		spin_unlock_irq(&lc->blocks_lock);
6600e9cebe7SJosef Bacik 	}
6610e9cebe7SJosef Bacik map_bio:
6620e9cebe7SJosef Bacik 	normal_map_bio(ti, bio);
6630e9cebe7SJosef Bacik 	return DM_MAPIO_REMAPPED;
6640e9cebe7SJosef Bacik }
6650e9cebe7SJosef Bacik 
6660e9cebe7SJosef Bacik static int normal_end_io(struct dm_target *ti, struct bio *bio, int error)
6670e9cebe7SJosef Bacik {
6680e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
6690e9cebe7SJosef Bacik 	struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
6700e9cebe7SJosef Bacik 
6710e9cebe7SJosef Bacik 	if (bio_data_dir(bio) == WRITE && pb->block) {
6720e9cebe7SJosef Bacik 		struct pending_block *block = pb->block;
6730e9cebe7SJosef Bacik 		unsigned long flags;
6740e9cebe7SJosef Bacik 
6750e9cebe7SJosef Bacik 		spin_lock_irqsave(&lc->blocks_lock, flags);
6760e9cebe7SJosef Bacik 		if (block->flags & LOG_FLUSH_FLAG) {
6770e9cebe7SJosef Bacik 			list_splice_tail_init(&block->list, &lc->logging_blocks);
6780e9cebe7SJosef Bacik 			list_add_tail(&block->list, &lc->logging_blocks);
6790e9cebe7SJosef Bacik 			wake_up_process(lc->log_kthread);
6800e9cebe7SJosef Bacik 		} else if (block->flags & LOG_FUA_FLAG) {
6810e9cebe7SJosef Bacik 			list_add_tail(&block->list, &lc->logging_blocks);
6820e9cebe7SJosef Bacik 			wake_up_process(lc->log_kthread);
6830e9cebe7SJosef Bacik 		} else
6840e9cebe7SJosef Bacik 			list_add_tail(&block->list, &lc->unflushed_blocks);
6850e9cebe7SJosef Bacik 		spin_unlock_irqrestore(&lc->blocks_lock, flags);
6860e9cebe7SJosef Bacik 	}
6870e9cebe7SJosef Bacik 
6880e9cebe7SJosef Bacik 	return error;
6890e9cebe7SJosef Bacik }
6900e9cebe7SJosef Bacik 
6910e9cebe7SJosef Bacik /*
6920e9cebe7SJosef Bacik  * INFO format: <logged entries> <highest allocated sector>
6930e9cebe7SJosef Bacik  */
6940e9cebe7SJosef Bacik static void log_writes_status(struct dm_target *ti, status_type_t type,
6950e9cebe7SJosef Bacik 			      unsigned status_flags, char *result,
6960e9cebe7SJosef Bacik 			      unsigned maxlen)
6970e9cebe7SJosef Bacik {
6980e9cebe7SJosef Bacik 	unsigned sz = 0;
6990e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
7000e9cebe7SJosef Bacik 
7010e9cebe7SJosef Bacik 	switch (type) {
7020e9cebe7SJosef Bacik 	case STATUSTYPE_INFO:
7030e9cebe7SJosef Bacik 		DMEMIT("%llu %llu", lc->logged_entries,
7040e9cebe7SJosef Bacik 		       (unsigned long long)lc->next_sector - 1);
7050e9cebe7SJosef Bacik 		if (!lc->logging_enabled)
7060e9cebe7SJosef Bacik 			DMEMIT(" logging_disabled");
7070e9cebe7SJosef Bacik 		break;
7080e9cebe7SJosef Bacik 
7090e9cebe7SJosef Bacik 	case STATUSTYPE_TABLE:
7100e9cebe7SJosef Bacik 		DMEMIT("%s %s", lc->dev->name, lc->logdev->name);
7110e9cebe7SJosef Bacik 		break;
7120e9cebe7SJosef Bacik 	}
7130e9cebe7SJosef Bacik }
7140e9cebe7SJosef Bacik 
7150e9cebe7SJosef Bacik static int log_writes_ioctl(struct dm_target *ti, unsigned int cmd,
7160e9cebe7SJosef Bacik 			    unsigned long arg)
7170e9cebe7SJosef Bacik {
7180e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
7190e9cebe7SJosef Bacik 	struct dm_dev *dev = lc->dev;
7200e9cebe7SJosef Bacik 	int r = 0;
7210e9cebe7SJosef Bacik 
7220e9cebe7SJosef Bacik 	/*
7230e9cebe7SJosef Bacik 	 * Only pass ioctls through if the device sizes match exactly.
7240e9cebe7SJosef Bacik 	 */
7250e9cebe7SJosef Bacik 	if (ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT)
7260e9cebe7SJosef Bacik 		r = scsi_verify_blk_ioctl(NULL, cmd);
7270e9cebe7SJosef Bacik 
7280e9cebe7SJosef Bacik 	return r ? : __blkdev_driver_ioctl(dev->bdev, dev->mode, cmd, arg);
7290e9cebe7SJosef Bacik }
7300e9cebe7SJosef Bacik 
7310e9cebe7SJosef Bacik static int log_writes_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
7320e9cebe7SJosef Bacik 			    struct bio_vec *biovec, int max_size)
7330e9cebe7SJosef Bacik {
7340e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
7350e9cebe7SJosef Bacik 	struct request_queue *q = bdev_get_queue(lc->dev->bdev);
7360e9cebe7SJosef Bacik 
7370e9cebe7SJosef Bacik 	if (!q->merge_bvec_fn)
7380e9cebe7SJosef Bacik 		return max_size;
7390e9cebe7SJosef Bacik 
7400e9cebe7SJosef Bacik 	bvm->bi_bdev = lc->dev->bdev;
7410e9cebe7SJosef Bacik 	bvm->bi_sector = dm_target_offset(ti, bvm->bi_sector);
7420e9cebe7SJosef Bacik 
7430e9cebe7SJosef Bacik 	return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
7440e9cebe7SJosef Bacik }
7450e9cebe7SJosef Bacik 
7460e9cebe7SJosef Bacik static int log_writes_iterate_devices(struct dm_target *ti,
7470e9cebe7SJosef Bacik 				      iterate_devices_callout_fn fn,
7480e9cebe7SJosef Bacik 				      void *data)
7490e9cebe7SJosef Bacik {
7500e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
7510e9cebe7SJosef Bacik 
7520e9cebe7SJosef Bacik 	return fn(ti, lc->dev, 0, ti->len, data);
7530e9cebe7SJosef Bacik }
7540e9cebe7SJosef Bacik 
7550e9cebe7SJosef Bacik /*
7560e9cebe7SJosef Bacik  * Messages supported:
7570e9cebe7SJosef Bacik  *   mark <mark data> - specify the marked data.
7580e9cebe7SJosef Bacik  */
7590e9cebe7SJosef Bacik static int log_writes_message(struct dm_target *ti, unsigned argc, char **argv)
7600e9cebe7SJosef Bacik {
7610e9cebe7SJosef Bacik 	int r = -EINVAL;
7620e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
7630e9cebe7SJosef Bacik 
7640e9cebe7SJosef Bacik 	if (argc != 2) {
7650e9cebe7SJosef Bacik 		DMWARN("Invalid log-writes message arguments, expect 2 arguments, got %d", argc);
7660e9cebe7SJosef Bacik 		return r;
7670e9cebe7SJosef Bacik 	}
7680e9cebe7SJosef Bacik 
7690e9cebe7SJosef Bacik 	if (!strcasecmp(argv[0], "mark"))
7700e9cebe7SJosef Bacik 		r = log_mark(lc, argv[1]);
7710e9cebe7SJosef Bacik 	else
7720e9cebe7SJosef Bacik 		DMWARN("Unrecognised log writes target message received: %s", argv[0]);
7730e9cebe7SJosef Bacik 
7740e9cebe7SJosef Bacik 	return r;
7750e9cebe7SJosef Bacik }
7760e9cebe7SJosef Bacik 
7770e9cebe7SJosef Bacik static void log_writes_io_hints(struct dm_target *ti, struct queue_limits *limits)
7780e9cebe7SJosef Bacik {
7790e9cebe7SJosef Bacik 	struct log_writes_c *lc = ti->private;
7800e9cebe7SJosef Bacik 	struct request_queue *q = bdev_get_queue(lc->dev->bdev);
7810e9cebe7SJosef Bacik 
7820e9cebe7SJosef Bacik 	if (!q || !blk_queue_discard(q)) {
7830e9cebe7SJosef Bacik 		lc->device_supports_discard = false;
7840e9cebe7SJosef Bacik 		limits->discard_granularity = 1 << SECTOR_SHIFT;
7850e9cebe7SJosef Bacik 		limits->max_discard_sectors = (UINT_MAX >> SECTOR_SHIFT);
7860e9cebe7SJosef Bacik 	}
7870e9cebe7SJosef Bacik }
7880e9cebe7SJosef Bacik 
7890e9cebe7SJosef Bacik static struct target_type log_writes_target = {
7900e9cebe7SJosef Bacik 	.name   = "log-writes",
7910e9cebe7SJosef Bacik 	.version = {1, 0, 0},
7920e9cebe7SJosef Bacik 	.module = THIS_MODULE,
7930e9cebe7SJosef Bacik 	.ctr    = log_writes_ctr,
7940e9cebe7SJosef Bacik 	.dtr    = log_writes_dtr,
7950e9cebe7SJosef Bacik 	.map    = log_writes_map,
7960e9cebe7SJosef Bacik 	.end_io = normal_end_io,
7970e9cebe7SJosef Bacik 	.status = log_writes_status,
7980e9cebe7SJosef Bacik 	.ioctl	= log_writes_ioctl,
7990e9cebe7SJosef Bacik 	.merge	= log_writes_merge,
8000e9cebe7SJosef Bacik 	.message = log_writes_message,
8010e9cebe7SJosef Bacik 	.iterate_devices = log_writes_iterate_devices,
8020e9cebe7SJosef Bacik 	.io_hints = log_writes_io_hints,
8030e9cebe7SJosef Bacik };
8040e9cebe7SJosef Bacik 
8050e9cebe7SJosef Bacik static int __init dm_log_writes_init(void)
8060e9cebe7SJosef Bacik {
8070e9cebe7SJosef Bacik 	int r = dm_register_target(&log_writes_target);
8080e9cebe7SJosef Bacik 
8090e9cebe7SJosef Bacik 	if (r < 0)
8100e9cebe7SJosef Bacik 		DMERR("register failed %d", r);
8110e9cebe7SJosef Bacik 
8120e9cebe7SJosef Bacik 	return r;
8130e9cebe7SJosef Bacik }
8140e9cebe7SJosef Bacik 
8150e9cebe7SJosef Bacik static void __exit dm_log_writes_exit(void)
8160e9cebe7SJosef Bacik {
8170e9cebe7SJosef Bacik 	dm_unregister_target(&log_writes_target);
8180e9cebe7SJosef Bacik }
8190e9cebe7SJosef Bacik 
8200e9cebe7SJosef Bacik module_init(dm_log_writes_init);
8210e9cebe7SJosef Bacik module_exit(dm_log_writes_exit);
8220e9cebe7SJosef Bacik 
8230e9cebe7SJosef Bacik MODULE_DESCRIPTION(DM_NAME " log writes target");
8240e9cebe7SJosef Bacik MODULE_AUTHOR("Josef Bacik <jbacik@fb.com>");
8250e9cebe7SJosef Bacik MODULE_LICENSE("GPL");
826