[PATCH v4 0/3] Btrfs: add IO error device stats

[PATCH v4 0/3] Btrfs: add IO error device stats

[Stefan Behrens]

Changes v1-v2:
- Remove restriction that BTRFS_IOC_GET_DEVICE_STATS is a privileged
  operation
- Cast u64 to unsigned long long for printf()

Changes v2-v3:
- Rebased on Chris' current master

Changes v3-v4:
- Add padding at end of ioctl structure

The goal is to detect when drives start to get an increased error rate,
when drives should be replaced soon. Therefore statistic counters are
added that count IO errors (read, write and flush). Additionally, the
software detected errors like checksum errors and corrupted blocks are
counted.

An ioctl interface is added to get the device statistic counters.
A second ioctl is added to atomically get and reset these counters.

The device statistics are written into the device tree with each
transaction commit. Only modified statistics are written.
When a filesystem is mounted, the device statistics for each involved
device are read from the device tree and used to initialize the
counters.

A patch for the btrfs-progs world will also be sent.

Stefan Behrens (3):
  Btrfs: add device counters for detected IO and checksum errors
  Btrfs: add ioctl to get and reset the device stats
  Btrfs: read device stats on mount, write modified ones during commit

 fs/btrfs/ctree.h       |   51 ++++++++
 fs/btrfs/disk-io.c     |   25 +++-
 fs/btrfs/extent_io.c   |   27 +++-
 fs/btrfs/ioctl.c       |   26 ++++
 fs/btrfs/ioctl.h       |   28 ++++
 fs/btrfs/print-tree.c  |    3 +
 fs/btrfs/scrub.c       |   72 ++++++++---
 fs/btrfs/transaction.c |    4 +
 fs/btrfs/volumes.c     |  335 +++++++++++++++++++++++++++++++++++++++++++++++-
 fs/btrfs/volumes.h     |   43 +++++++
 10 files changed, 589 insertions(+), 25 deletions(-)

-- 
1.7.10.2

[PATCH v4 1/3] Btrfs: add device counters for detected IO and checksum errors

[Stefan Behrens]

The goal is to detect when drives start to get an increased error rate,
when drives should be replaced soon. Therefore statistic counters are
added that count IO errors (read, write and flush). Additionally, the
software detected errors like checksum errors and corrupted blocks are
counted.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
---
 fs/btrfs/disk-io.c   |   18 ++++++++++---
 fs/btrfs/extent_io.c |   27 +++++++++++++++++--
 fs/btrfs/scrub.c     |   72 +++++++++++++++++++++++++++++++++++++++-----------
 fs/btrfs/volumes.c   |   61 +++++++++++++++++++++++++++++++++++++++---
 fs/btrfs/volumes.h   |   21 +++++++++++++++
 5 files changed, 174 insertions(+), 25 deletions(-)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index a7ffc88..e123629 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -2556,18 +2556,21 @@ recovery_tree_root:
 
 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
 {
-	char b[BDEVNAME_SIZE];
-
 	if (uptodate) {
 		set_buffer_uptodate(bh);
 	} else {
+		struct btrfs_device *device = (struct btrfs_device *)
+			bh->b_private;
+
 		printk_ratelimited(KERN_WARNING "lost page write due to "
-					"I/O error on %s\n",
-				       bdevname(bh->b_bdev, b));
+				   "I/O error on %s\n", device->name);
 		/* note, we dont' set_buffer_write_io_error because we have
 		 * our own ways of dealing with the IO errors
 		 */
 		clear_buffer_uptodate(bh);
+		btrfs_device_stat_inc(&device->cnt_write_io_errs);
+		device->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(device);
 	}
 	unlock_buffer(bh);
 	put_bh(bh);
@@ -2682,6 +2685,7 @@ static int write_dev_supers(struct btrfs_device *device,
 			set_buffer_uptodate(bh);
 			lock_buffer(bh);
 			bh->b_end_io = btrfs_end_buffer_write_sync;
+			bh->b_private = device;
 		}
 
 		/*
@@ -2740,6 +2744,12 @@ static int write_dev_flush(struct btrfs_device *device, int wait)
 		}
 		if (!bio_flagged(bio, BIO_UPTODATE)) {
 			ret = -EIO;
+			if (!bio_flagged(bio, BIO_EOPNOTSUPP)) {
+				btrfs_device_stat_inc(
+					&device->cnt_flush_io_errs);
+				device->device_stats_dirty = 1;
+				btrfs_device_stat_print_on_error(device);
+			}
 		}
 
 		/* drop the reference from the wait == 0 run */
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 2fb52c2..6cd9a55 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -1923,6 +1923,9 @@ int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
 	if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 		/* try to remap that extent elsewhere? */
 		bio_put(bio);
+		btrfs_device_stat_inc(&dev->cnt_write_io_errs);
+		dev->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(dev);
 		return -EIO;
 	}
 
@@ -2347,10 +2350,30 @@ static void end_bio_extent_readpage(struct bio *bio, int err)
 		if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
 			ret = tree->ops->readpage_end_io_hook(page, start, end,
 							      state, mirror);
-			if (ret)
+			if (ret) {
+				/* no IO indicated but software detected errors
+				 * in the block, either checksum errors or
+				 * issues with the contents */
+				int failed_mirror = (int)(uintptr_t)
+						    bio->bi_bdev;
+				struct btrfs_root *root =
+					BTRFS_I(page->mapping->host)->root;
+				struct btrfs_device *device;
+
 				uptodate = 0;
-			else
+				device = btrfs_find_device_for_logical(
+						root, start,
+						(int)failed_mirror);
+				if (device) {
+					btrfs_device_stat_inc(
+						&device->cnt_corruption_errs);
+					device->device_stats_dirty = 1;
+					btrfs_device_stat_print_on_error(
+						device);
+				}
+			} else {
 				clean_io_failure(start, page);
+			}
 		}
 
 		if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c
index 7e487be..2795c94 100644
--- a/fs/btrfs/scrub.c
+++ b/fs/btrfs/scrub.c
@@ -50,7 +50,7 @@ struct scrub_dev;
 struct scrub_page {
 	struct scrub_block	*sblock;
 	struct page		*page;
-	struct block_device	*bdev;
+	struct btrfs_device	*dev;
 	u64			flags;  /* extent flags */
 	u64			generation;
 	u64			logical;
@@ -86,6 +86,7 @@ struct scrub_block {
 		unsigned int	header_error:1;
 		unsigned int	checksum_error:1;
 		unsigned int	no_io_error_seen:1;
+		unsigned int	generation_error:1; /* also sets header_error */
 	};
 };
 
@@ -675,6 +676,9 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
 		sdev->stat.read_errors++;
 		sdev->stat.uncorrectable_errors++;
 		spin_unlock(&sdev->stat_lock);
+		btrfs_device_stat_inc(&sdev->dev->cnt_read_io_errs);
+		sdev->dev->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(sdev->dev);
 		goto out;
 	}
 
@@ -686,6 +690,9 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
 		sdev->stat.read_errors++;
 		sdev->stat.uncorrectable_errors++;
 		spin_unlock(&sdev->stat_lock);
+		btrfs_device_stat_inc(&sdev->dev->cnt_read_io_errs);
+		sdev->dev->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(sdev->dev);
 		goto out;
 	}
 	BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
@@ -699,6 +706,9 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
 		sdev->stat.read_errors++;
 		sdev->stat.uncorrectable_errors++;
 		spin_unlock(&sdev->stat_lock);
+		btrfs_device_stat_inc(&sdev->dev->cnt_read_io_errs);
+		sdev->dev->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(sdev->dev);
 		goto out;
 	}
 
@@ -725,12 +735,18 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
 		spin_unlock(&sdev->stat_lock);
 		if (__ratelimit(&_rs))
 			scrub_print_warning("i/o error", sblock_to_check);
+		btrfs_device_stat_inc(&sdev->dev->cnt_read_io_errs);
+		sdev->dev->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(sdev->dev);
 	} else if (sblock_bad->checksum_error) {
 		spin_lock(&sdev->stat_lock);
 		sdev->stat.csum_errors++;
 		spin_unlock(&sdev->stat_lock);
 		if (__ratelimit(&_rs))
 			scrub_print_warning("checksum error", sblock_to_check);
+		btrfs_device_stat_inc(&sdev->dev->cnt_corruption_errs);
+		sdev->dev->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(sdev->dev);
 	} else if (sblock_bad->header_error) {
 		spin_lock(&sdev->stat_lock);
 		sdev->stat.verify_errors++;
@@ -738,6 +754,12 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
 		if (__ratelimit(&_rs))
 			scrub_print_warning("checksum/header error",
 					    sblock_to_check);
+		if (sblock_bad->generation_error)
+			btrfs_device_stat_inc(&sdev->dev->cnt_generation_errs);
+		else
+			btrfs_device_stat_inc(&sdev->dev->cnt_corruption_errs);
+		sdev->dev->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(sdev->dev);
 	}
 
 	if (sdev->readonly)
@@ -998,8 +1020,8 @@ static int scrub_setup_recheck_block(struct scrub_dev *sdev,
 			page = sblock->pagev + page_index;
 			page->logical = logical;
 			page->physical = bbio->stripes[mirror_index].physical;
-			/* for missing devices, bdev is NULL */
-			page->bdev = bbio->stripes[mirror_index].dev->bdev;
+			/* for missing devices, dev->bdev is NULL */
+			page->dev = bbio->stripes[mirror_index].dev;
 			page->mirror_num = mirror_index + 1;
 			page->page = alloc_page(GFP_NOFS);
 			if (!page->page) {
@@ -1043,7 +1065,7 @@ static int scrub_recheck_block(struct btrfs_fs_info *fs_info,
 		struct scrub_page *page = sblock->pagev + page_num;
 		DECLARE_COMPLETION_ONSTACK(complete);
 
-		if (page->bdev == NULL) {
+		if (page->dev->bdev == NULL) {
 			page->io_error = 1;
 			sblock->no_io_error_seen = 0;
 			continue;
@@ -1053,7 +1075,7 @@ static int scrub_recheck_block(struct btrfs_fs_info *fs_info,
 		bio = bio_alloc(GFP_NOFS, 1);
 		if (!bio)
 			return -EIO;
-		bio->bi_bdev = page->bdev;
+		bio->bi_bdev = page->dev->bdev;
 		bio->bi_sector = page->physical >> 9;
 		bio->bi_end_io = scrub_complete_bio_end_io;
 		bio->bi_private = &complete;
@@ -1102,11 +1124,14 @@ static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
 		h = (struct btrfs_header *)mapped_buffer;
 
 		if (sblock->pagev[0].logical != le64_to_cpu(h->bytenr) ||
-		    generation != le64_to_cpu(h->generation) ||
 		    memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) ||
 		    memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
-			   BTRFS_UUID_SIZE))
+			   BTRFS_UUID_SIZE)) {
 			sblock->header_error = 1;
+		} else if (generation != le64_to_cpu(h->generation)) {
+			sblock->header_error = 1;
+			sblock->generation_error = 1;
+		}
 		csum = h->csum;
 	} else {
 		if (!have_csum)
@@ -1183,7 +1208,7 @@ static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
 		bio = bio_alloc(GFP_NOFS, 1);
 		if (!bio)
 			return -EIO;
-		bio->bi_bdev = page_bad->bdev;
+		bio->bi_bdev = page_bad->dev->bdev;
 		bio->bi_sector = page_bad->physical >> 9;
 		bio->bi_end_io = scrub_complete_bio_end_io;
 		bio->bi_private = &complete;
@@ -1197,6 +1222,14 @@ static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
 
 		/* this will also unplug the queue */
 		wait_for_completion(&complete);
+		if (!bio_flagged(bio, BIO_UPTODATE)) {
+			btrfs_device_stat_inc(
+				&page_bad->dev->cnt_write_io_errs);
+			page_bad->dev->device_stats_dirty = 1;
+			btrfs_device_stat_print_on_error(page_bad->dev);
+			bio_put(bio);
+			return -EIO;
+		}
 		bio_put(bio);
 	}
 
@@ -1353,7 +1386,8 @@ static int scrub_checksum_super(struct scrub_block *sblock)
 	u64 mapped_size;
 	void *p;
 	u32 crc = ~(u32)0;
-	int fail = 0;
+	int fail_gen = 0;
+	int fail_cor = 0;
 	u64 len;
 	int index;
 
@@ -1364,13 +1398,13 @@ static int scrub_checksum_super(struct scrub_block *sblock)
 	memcpy(on_disk_csum, s->csum, sdev->csum_size);
 
 	if (sblock->pagev[0].logical != le64_to_cpu(s->bytenr))
-		++fail;
+		++fail_cor;
 
 	if (sblock->pagev[0].generation != le64_to_cpu(s->generation))
-		++fail;
+		++fail_gen;
 
 	if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
-		++fail;
+		++fail_cor;
 
 	len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
 	mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
@@ -1395,9 +1429,9 @@ static int scrub_checksum_super(struct scrub_block *sblock)
 
 	btrfs_csum_final(crc, calculated_csum);
 	if (memcmp(calculated_csum, on_disk_csum, sdev->csum_size))
-		++fail;
+		++fail_cor;
 
-	if (fail) {
+	if (fail_cor + fail_gen) {
 		/*
 		 * if we find an error in a super block, we just report it.
 		 * They will get written with the next transaction commit
@@ -1406,9 +1440,15 @@ static int scrub_checksum_super(struct scrub_block *sblock)
 		spin_lock(&sdev->stat_lock);
 		++sdev->stat.super_errors;
 		spin_unlock(&sdev->stat_lock);
+		if (fail_cor)
+			btrfs_device_stat_inc(&sdev->dev->cnt_corruption_errs);
+		else
+			btrfs_device_stat_inc(&sdev->dev->cnt_generation_errs);
+		sdev->dev->device_stats_dirty = 1;
+		btrfs_device_stat_print_on_error(sdev->dev);
 	}
 
-	return fail;
+	return fail_cor + fail_gen;
 }
 
 static void scrub_block_get(struct scrub_block *sblock)
@@ -1552,7 +1592,7 @@ static int scrub_pages(struct scrub_dev *sdev, u64 logical, u64 len,
 			return -ENOMEM;
 		}
 		spage->sblock = sblock;
-		spage->bdev = sdev->dev->bdev;
+		spage->dev = sdev->dev;
 		spage->flags = flags;
 		spage->generation = gen;
 		spage->logical = logical;
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 1411b99..c458c74 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -23,6 +23,7 @@
 #include <linux/random.h>
 #include <linux/iocontext.h>
 #include <linux/capability.h>
+#include <linux/ratelimit.h>
 #include <linux/kthread.h>
 #include <asm/div64.h>
 #include "compat.h"
@@ -4003,11 +4004,28 @@ int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
 
 static void btrfs_end_bio(struct bio *bio, int err)
 {
-	struct btrfs_bio *bbio = bio->bi_private;
+	struct btrfs_bio *bbio = (struct btrfs_bio *)
+		(((uintptr_t)bio->bi_private) & ~((uintptr_t)3));
 	int is_orig_bio = 0;
 
-	if (err)
+	if (err) {
 		atomic_inc(&bbio->error);
+		if (err == -EIO || err == -EREMOTEIO) {
+			unsigned int dev_nr = ((uintptr_t)bio->bi_private) & 3;
+			struct btrfs_device *dev;
+
+			BUG_ON(dev_nr >= bbio->num_stripes);
+			dev = bbio->stripes[dev_nr].dev;
+			if (bio->bi_rw & WRITE)
+				btrfs_device_stat_inc(&dev->cnt_write_io_errs);
+			else
+				btrfs_device_stat_inc(&dev->cnt_read_io_errs);
+			if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
+				btrfs_device_stat_inc(&dev->cnt_flush_io_errs);
+			dev->device_stats_dirty = 1;
+			btrfs_device_stat_print_on_error(dev);
+		}
+	}
 
 	if (bio == bbio->orig_bio)
 		is_orig_bio = 1;
@@ -4148,7 +4166,9 @@ int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
 		} else {
 			bio = first_bio;
 		}
-		bio->bi_private = bbio;
+		BUG_ON((((uintptr_t)bbio) & 3) != 0);
+		BUG_ON(dev_nr > 3);
+		bio->bi_private = (void *)(((uintptr_t)bbio) | dev_nr);
 		bio->bi_end_io = btrfs_end_bio;
 		bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
 		dev = bbio->stripes[dev_nr].dev;
@@ -4509,6 +4529,28 @@ int btrfs_read_sys_array(struct btrfs_root *root)
 	return ret;
 }
 
+struct btrfs_device *btrfs_find_device_for_logical(struct btrfs_root *root,
+						   u64 logical, int mirror_num)
+{
+	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
+	int ret;
+	u64 map_length = 0;
+	struct btrfs_bio *bbio = NULL;
+	struct btrfs_device *device;
+
+	BUG_ON(mirror_num == 0);
+	ret = btrfs_map_block(map_tree, WRITE, logical, &map_length, &bbio,
+			      mirror_num);
+	if (ret) {
+		BUG_ON(bbio != NULL);
+		return NULL;
+	}
+	BUG_ON(mirror_num != bbio->mirror_num);
+	device = bbio->stripes[mirror_num - 1].dev;
+	kfree(bbio);
+	return device;
+}
+
 int btrfs_read_chunk_tree(struct btrfs_root *root)
 {
 	struct btrfs_path *path;
@@ -4583,3 +4625,16 @@ error:
 	btrfs_free_path(path);
 	return ret;
 }
+
+void btrfs_device_stat_print_on_error(struct btrfs_device *device)
+{
+	printk_ratelimited(KERN_ERR
+			   "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
+			   device->name,
+			   btrfs_device_stat_read(&device->cnt_write_io_errs),
+			   btrfs_device_stat_read(&device->cnt_read_io_errs),
+			   btrfs_device_stat_read(&device->cnt_flush_io_errs),
+			   btrfs_device_stat_read(&device->cnt_corruption_errs),
+			   btrfs_device_stat_read(
+				&device->cnt_generation_errs));
+}
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index bb6b03f..08afa6c 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -106,6 +106,14 @@ struct btrfs_device {
 	struct completion flush_wait;
 	int nobarriers;
 
+	/* disk I/O failure stats. For detailed description refer to
+	 * struct btrfs_device_stats_item in ctree.h */
+	int device_stats_dirty; /* counters need to be written to disk */
+	atomic_t cnt_write_io_errs;
+	atomic_t cnt_read_io_errs;
+	atomic_t cnt_flush_io_errs;
+	atomic_t cnt_corruption_errs;
+	atomic_t cnt_generation_errs;
 };
 
 struct btrfs_fs_devices {
@@ -281,4 +289,17 @@ int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset);
 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 			 u64 *start, u64 *max_avail);
+struct btrfs_device *btrfs_find_device_for_logical(struct btrfs_root *root,
+						   u64 logical, int mirror_num);
+void btrfs_device_stat_print_on_error(struct btrfs_device *device);
+
+static inline void btrfs_device_stat_inc(atomic_t *cnt)
+{
+	atomic_inc(cnt);
+}
+
+static inline int btrfs_device_stat_read(atomic_t *cnt)
+{
+	return atomic_read(cnt);
+}
 #endif
-- 
1.7.10.2

[PATCH v4 2/3] Btrfs: add ioctl to get and reset the device stats

[Stefan Behrens]

An ioctl interface is added to get the device statistic counters.
A second ioctl is added to atomically get and reset these counters.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
---
 fs/btrfs/ioctl.c   |   26 ++++++++++++++++++++
 fs/btrfs/ioctl.h   |   28 +++++++++++++++++++++
 fs/btrfs/volumes.c |   69 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/volumes.h |   13 ++++++++++
 4 files changed, 136 insertions(+)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 14f8e1f..19d2244 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -3042,6 +3042,28 @@ static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
 	return ret;
 }
 
+static long btrfs_ioctl_get_device_stats(struct btrfs_root *root,
+					 void __user *arg, int reset_after_read)
+{
+	struct btrfs_ioctl_get_device_stats *sa;
+	int ret;
+
+	if (reset_after_read && !capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	sa = memdup_user(arg, sizeof(*sa));
+	if (IS_ERR(sa))
+		return PTR_ERR(sa);
+
+	ret = btrfs_get_device_stats(root, sa, reset_after_read);
+
+	if (copy_to_user(arg, sa, sizeof(*sa)))
+		ret = -EFAULT;
+
+	kfree(sa);
+	return ret;
+}
+
 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
 {
 	int ret = 0;
@@ -3424,6 +3446,10 @@ long btrfs_ioctl(struct file *file, unsigned int
 		return btrfs_ioctl_balance_ctl(root, arg);
 	case BTRFS_IOC_BALANCE_PROGRESS:
 		return btrfs_ioctl_balance_progress(root, argp);
+	case BTRFS_IOC_GET_DEVICE_STATS:
+		return btrfs_ioctl_get_device_stats(root, argp, 0);
+	case BTRFS_IOC_GET_AND_RESET_DEVICE_STATS:
+		return btrfs_ioctl_get_device_stats(root, argp, 1);
 	}
 
 	return -ENOTTY;
diff --git a/fs/btrfs/ioctl.h b/fs/btrfs/ioctl.h
index 086e6bd..f1c1196 100644
--- a/fs/btrfs/ioctl.h
+++ b/fs/btrfs/ioctl.h
@@ -266,6 +266,30 @@ struct btrfs_ioctl_logical_ino_args {
 	__u64				inodes;
 };
 
+#define BTRFS_IOCTL_GET_DEVICE_STATS_MAX_NR_ITEMS	5
+struct btrfs_ioctl_get_device_stats {
+	__u64 devid;				/* in */
+	__u64 nr_items;				/* in/out */
+
+	/* out values: */
+
+	/* disk I/O failure stats */
+	__u64 cnt_write_io_errs; /* EIO or EREMOTEIO from lower layers */
+	__u64 cnt_read_io_errs; /* EIO or EREMOTEIO from lower layers */
+	__u64 cnt_flush_io_errs; /* EIO or EREMOTEIO from lower layers */
+
+	/* stats for indirect indications for I/O failures */
+	__u64 cnt_corruption_errs; /* checksum error, bytenr error or
+				    * contents is illegal: this is an
+				    * indication that the block was damaged
+				    * during read or write, or written to
+				    * wrong location or read from wrong
+				    * location */
+	__u64 cnt_generation_errs; /* an indication that blocks have not
+				    * been written */
+	__u64 unused[121]; /* pad to 1k */
+};
+
 #define BTRFS_IOC_SNAP_CREATE _IOW(BTRFS_IOCTL_MAGIC, 1, \
 				   struct btrfs_ioctl_vol_args)
 #define BTRFS_IOC_DEFRAG _IOW(BTRFS_IOCTL_MAGIC, 2, \
@@ -330,5 +354,9 @@ struct btrfs_ioctl_logical_ino_args {
 					struct btrfs_ioctl_ino_path_args)
 #define BTRFS_IOC_LOGICAL_INO _IOWR(BTRFS_IOCTL_MAGIC, 36, \
 					struct btrfs_ioctl_ino_path_args)
+#define BTRFS_IOC_GET_DEVICE_STATS _IOWR(BTRFS_IOCTL_MAGIC, 52, \
+					 struct btrfs_ioctl_get_device_stats)
+#define BTRFS_IOC_GET_AND_RESET_DEVICE_STATS _IOWR(BTRFS_IOCTL_MAGIC, 53, \
+					 struct btrfs_ioctl_get_device_stats)
 
 #endif
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index c458c74..5f5a6ce 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -4638,3 +4638,72 @@ void btrfs_device_stat_print_on_error(struct btrfs_device *device)
 			   btrfs_device_stat_read(
 				&device->cnt_generation_errs));
 }
+
+int btrfs_get_device_stats(struct btrfs_root *root,
+			   struct btrfs_ioctl_get_device_stats *stats,
+			   int reset_after_read)
+{
+	struct btrfs_device *dev;
+	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
+
+	mutex_lock(&fs_devices->device_list_mutex);
+	dev = btrfs_find_device(root, stats->devid, NULL, NULL);
+	mutex_unlock(&fs_devices->device_list_mutex);
+
+	if (!dev) {
+		printk(KERN_WARNING
+		       "btrfs: get device_stats failed, device not found\n");
+		return -ENODEV;
+	} else if (reset_after_read) {
+		if (stats->nr_items >= 1)
+			stats->cnt_write_io_errs =
+				btrfs_device_stat_read_and_reset(
+					&dev->cnt_write_io_errs);
+		else
+			btrfs_device_stat_reset(&dev->cnt_write_io_errs);
+		if (stats->nr_items >= 2)
+			stats->cnt_read_io_errs =
+				btrfs_device_stat_read_and_reset(
+					&dev->cnt_read_io_errs);
+		else
+			btrfs_device_stat_reset(&dev->cnt_read_io_errs);
+		if (stats->nr_items >= 3)
+			stats->cnt_flush_io_errs =
+				btrfs_device_stat_read_and_reset(
+					&dev->cnt_flush_io_errs);
+		else
+			btrfs_device_stat_reset(&dev->cnt_flush_io_errs);
+		if (stats->nr_items >= 4)
+			stats->cnt_corruption_errs =
+				btrfs_device_stat_read_and_reset(
+					&dev->cnt_corruption_errs);
+		else
+			btrfs_device_stat_reset(&dev->cnt_corruption_errs);
+		if (stats->nr_items >= 5)
+			stats->cnt_generation_errs =
+				btrfs_device_stat_read_and_reset(
+					&dev->cnt_generation_errs);
+		else
+			btrfs_device_stat_reset(&dev->cnt_generation_errs);
+		dev->device_stats_dirty = 1;
+	} else {
+		if (stats->nr_items >= 1)
+			stats->cnt_write_io_errs = btrfs_device_stat_read(
+					&dev->cnt_write_io_errs);
+		if (stats->nr_items >= 2)
+			stats->cnt_read_io_errs = btrfs_device_stat_read(
+					&dev->cnt_read_io_errs);
+		if (stats->nr_items >= 3)
+			stats->cnt_flush_io_errs = btrfs_device_stat_read(
+					&dev->cnt_flush_io_errs);
+		if (stats->nr_items >= 4)
+			stats->cnt_corruption_errs = btrfs_device_stat_read(
+					&dev->cnt_corruption_errs);
+		if (stats->nr_items >= 5)
+			stats->cnt_generation_errs = btrfs_device_stat_read(
+					&dev->cnt_generation_errs);
+	}
+	if (stats->nr_items > 5)
+		stats->nr_items = 5;
+	return 0;
+}
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index 08afa6c..e0b31f1 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -292,6 +292,9 @@ int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 struct btrfs_device *btrfs_find_device_for_logical(struct btrfs_root *root,
 						   u64 logical, int mirror_num);
 void btrfs_device_stat_print_on_error(struct btrfs_device *device);
+int btrfs_get_device_stats(struct btrfs_root *root,
+			   struct btrfs_ioctl_get_device_stats *stats,
+			   int reset_after_read);
 
 static inline void btrfs_device_stat_inc(atomic_t *cnt)
 {
@@ -302,4 +305,14 @@ static inline int btrfs_device_stat_read(atomic_t *cnt)
 {
 	return atomic_read(cnt);
 }
+
+static inline int btrfs_device_stat_read_and_reset(atomic_t *cnt)
+{
+	return atomic_xchg(cnt, 0);
+}
+
+static inline void btrfs_device_stat_reset(atomic_t *cnt)
+{
+	atomic_set(cnt, 0);
+}
 #endif
-- 
1.7.10.2

[PATCH v4 3/3] Btrfs: read device stats on mount, write modified ones during commit

[Stefan Behrens]

The device statistics are written into the device tree with each
transaction commit. Only modified statistics are written.
When a filesystem is mounted, the device statistics for each involved
device are read from the device tree and used to initialize the
counters.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
---
 fs/btrfs/ctree.h       |   51 ++++++++++++
 fs/btrfs/disk-io.c     |    7 ++
 fs/btrfs/print-tree.c  |    3 +
 fs/btrfs/transaction.c |    4 +
 fs/btrfs/volumes.c     |  205 ++++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/volumes.h     |    9 +++
 6 files changed, 279 insertions(+)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index ec42a24..1dd7651 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -823,6 +823,26 @@ struct btrfs_csum_item {
 	u8 csum;
 } __attribute__ ((__packed__));
 
+struct btrfs_device_stats_item {
+	/*
+	 * grow this item struct at the end for future enhancements and keep
+	 * the existing values unchanged
+	 */
+	__le64 cnt_write_io_errs; /* EIO or EREMOTEIO from lower layers */
+	__le64 cnt_read_io_errs; /* EIO or EREMOTEIO from lower layers */
+	__le64 cnt_flush_io_errs; /* EIO or EREMOTEIO from lower layers */
+
+	/* stats for indirect indications for I/O failures */
+	__le64 cnt_corruption_errs; /* checksum error, bytenr error or
+				     * contents is illegal: this is an
+				     * indication that the block was damaged
+				     * during read or write, or written to
+				     * wrong location or read from wrong
+				     * location */
+	__le64 cnt_generation_errs; /* an indication that blocks have not
+				     * been written */
+} __attribute__ ((__packed__));
+
 /* different types of block groups (and chunks) */
 #define BTRFS_BLOCK_GROUP_DATA		(1ULL << 0)
 #define BTRFS_BLOCK_GROUP_SYSTEM	(1ULL << 1)
@@ -1508,6 +1528,12 @@ struct btrfs_ioctl_defrag_range_args {
 #define BTRFS_BALANCE_ITEM_KEY	248
 
 /*
+ * Persistantly stores the io stats in the device tree.
+ * One key for all stats, (0, BTRFS_DEVICE_STATS_KEY, devid).
+ */
+#define BTRFS_DEVICE_STATS_KEY	249
+
+/*
  * string items are for debugging.  They just store a short string of
  * data in the FS
  */
@@ -2415,6 +2441,31 @@ static inline u32 btrfs_file_extent_inline_item_len(struct extent_buffer *eb,
 	return btrfs_item_size(eb, e) - offset;
 }
 
+/* btrfs_device_stats_item */
+BTRFS_SETGET_FUNCS(device_stats_cnt_write_io_errs,
+		   struct btrfs_device_stats_item, cnt_write_io_errs, 64);
+BTRFS_SETGET_FUNCS(device_stats_cnt_read_io_errs,
+		   struct btrfs_device_stats_item, cnt_read_io_errs, 64);
+BTRFS_SETGET_FUNCS(device_stats_cnt_flush_io_errs,
+		   struct btrfs_device_stats_item, cnt_flush_io_errs, 64);
+BTRFS_SETGET_FUNCS(device_stats_cnt_corruption_errs,
+		   struct btrfs_device_stats_item, cnt_corruption_errs, 64);
+BTRFS_SETGET_FUNCS(device_stats_cnt_generation_errs,
+		   struct btrfs_device_stats_item, cnt_generation_errs, 64);
+
+BTRFS_SETGET_STACK_FUNCS(stack_device_stats_cnt_write_io_errs,
+			 struct btrfs_device_stats_item, cnt_write_io_errs, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_device_stats_cnt_read_io_errs,
+			 struct btrfs_device_stats_item, cnt_read_io_errs, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_device_stats_cnt_flush_io_errs,
+			 struct btrfs_device_stats_item, cnt_flush_io_errs, 64);
+BTRFS_SETGET_STACK_FUNCS(stack_device_stats_cnt_corruption_errs,
+			 struct btrfs_device_stats_item, cnt_corruption_errs,
+			 64);
+BTRFS_SETGET_STACK_FUNCS(stack_device_stats_cnt_generation_errs,
+			 struct btrfs_device_stats_item, cnt_generation_errs,
+			 64);
+
 static inline struct btrfs_fs_info *btrfs_sb(struct super_block *sb)
 {
 	return sb->s_fs_info;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index e123629..7ba08f7 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -2353,6 +2353,13 @@ retry_root_backup:
 	fs_info->generation = generation;
 	fs_info->last_trans_committed = generation;
 
+	ret = btrfs_init_device_stats(fs_info);
+	if (ret) {
+		printk(KERN_ERR "btrfs: failed to init device_stats: %d\n",
+		       ret);
+		goto fail_block_groups;
+	}
+
 	ret = btrfs_init_space_info(fs_info);
 	if (ret) {
 		printk(KERN_ERR "Failed to initial space info: %d\n", ret);
diff --git a/fs/btrfs/print-tree.c b/fs/btrfs/print-tree.c
index f38e452..a9e45e4 100644
--- a/fs/btrfs/print-tree.c
+++ b/fs/btrfs/print-tree.c
@@ -294,6 +294,9 @@ void btrfs_print_leaf(struct btrfs_root *root, struct extent_buffer *l)
 			       btrfs_dev_extent_chunk_offset(l, dev_extent),
 			       (unsigned long long)
 			       btrfs_dev_extent_length(l, dev_extent));
+		case BTRFS_DEVICE_STATS_KEY:
+			printk(KERN_INFO "\t\tdevice stats\n");
+			break;
 		};
 	}
 }
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 3642225..1722af0 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -28,6 +28,7 @@
 #include "locking.h"
 #include "tree-log.h"
 #include "inode-map.h"
+#include "volumes.h"
 
 #define BTRFS_ROOT_TRANS_TAG 0
 
@@ -758,6 +759,9 @@ static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
 	if (ret)
 		return ret;
 
+	ret = btrfs_run_device_stats(trans, root->fs_info);
+	BUG_ON(ret);
+
 	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
 		next = fs_info->dirty_cowonly_roots.next;
 		list_del_init(next);
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 5f5a6ce..d0edead 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -40,6 +40,8 @@ static int init_first_rw_device(struct btrfs_trans_handle *trans,
 				struct btrfs_root *root,
 				struct btrfs_device *device);
 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
+static void __btrfs_reset_device_stats(struct btrfs_device *dev);
+static void btrfs_device_stat_print_on_load(struct btrfs_device *device);
 
 static DEFINE_MUTEX(uuid_mutex);
 static LIST_HEAD(fs_uuids);
@@ -362,6 +364,7 @@ static noinline int device_list_add(const char *path,
 			return -ENOMEM;
 		}
 		device->devid = devid;
+		device->device_stats_valid = 0;
 		device->work.func = pending_bios_fn;
 		memcpy(device->uuid, disk_super->dev_item.uuid,
 		       BTRFS_UUID_SIZE);
@@ -4626,8 +4629,194 @@ error:
 	return ret;
 }
 
+static void __btrfs_reset_device_stats(struct btrfs_device *device)
+{
+	btrfs_device_stat_reset(&device->cnt_write_io_errs);
+	btrfs_device_stat_reset(&device->cnt_read_io_errs);
+	btrfs_device_stat_reset(&device->cnt_flush_io_errs);
+	btrfs_device_stat_reset(&device->cnt_corruption_errs);
+	btrfs_device_stat_reset(&device->cnt_generation_errs);
+}
+
+int btrfs_init_device_stats(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct btrfs_root *dev_root = fs_info->dev_root;
+	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
+	struct extent_buffer *eb;
+	int slot;
+	int ret = 0;
+	struct btrfs_device *device;
+	struct btrfs_path *path = NULL;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	mutex_lock(&fs_devices->device_list_mutex);
+	list_for_each_entry(device, &fs_devices->devices, dev_list) {
+		int item_size;
+		struct btrfs_device_stats_item *ptr;
+
+		key.objectid = 0;
+		key.type = BTRFS_DEVICE_STATS_KEY;
+		key.offset = device->devid;
+		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
+		if (ret) {
+			printk(KERN_WARNING "btrfs: no device_stats entry found for device %s (devid %llu) (OK on first mount after mkfs)\n",
+			       device->name, (unsigned long long)device->devid);
+			__btrfs_reset_device_stats(device);
+			device->device_stats_valid = 1;
+			device->device_stats_dirty = 1;
+			btrfs_release_path(path);
+			continue;
+		}
+		slot = path->slots[0];
+		eb = path->nodes[0];
+		btrfs_item_key_to_cpu(eb, &found_key, slot);
+		item_size = btrfs_item_size_nr(eb, slot);
+
+		ptr = btrfs_item_ptr(eb, slot,
+				     struct btrfs_device_stats_item);
+
+		if (item_size >= 1 * sizeof(__le64))
+			btrfs_device_stat_set(&device->cnt_write_io_errs,
+				btrfs_device_stats_cnt_write_io_errs(eb, ptr));
+		else
+			btrfs_device_stat_reset(&device->cnt_write_io_errs);
+		if (item_size >= 2 * sizeof(__le64))
+			btrfs_device_stat_set(&device->cnt_read_io_errs,
+				btrfs_device_stats_cnt_read_io_errs(eb, ptr));
+		else
+			btrfs_device_stat_reset(&device->cnt_read_io_errs);
+		if (item_size >= 3 * sizeof(__le64))
+			btrfs_device_stat_set(&device->cnt_flush_io_errs,
+				btrfs_device_stats_cnt_flush_io_errs(eb, ptr));
+		else
+			btrfs_device_stat_reset(&device->cnt_flush_io_errs);
+		if (item_size >= 4 * sizeof(__le64))
+			btrfs_device_stat_set(&device->cnt_corruption_errs,
+				btrfs_device_stats_cnt_corruption_errs(eb,
+								       ptr));
+		else
+			btrfs_device_stat_reset(&device->cnt_corruption_errs);
+		if (item_size >= 5 * sizeof(__le64))
+			btrfs_device_stat_set(&device->cnt_generation_errs,
+				btrfs_device_stats_cnt_generation_errs(eb,
+								       ptr));
+		else
+			btrfs_device_stat_reset(&device->cnt_generation_errs);
+
+		btrfs_device_stat_print_on_load(device);
+		device->device_stats_valid = 1;
+		btrfs_release_path(path);
+	}
+	mutex_unlock(&fs_devices->device_list_mutex);
+
+out:
+	btrfs_free_path(path);
+	return ret < 0 ? ret : 0;
+}
+
+static int update_device_stat_item(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *dev_root,
+				   struct btrfs_device *device)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct extent_buffer *eb;
+	struct btrfs_device_stats_item *ptr;
+	int ret;
+
+	key.objectid = 0;
+	key.type = BTRFS_DEVICE_STATS_KEY;
+	key.offset = device->devid;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_search_slot(trans, dev_root, &key, path, 0, 1);
+	if (ret < 0) {
+		printk(KERN_WARNING "btrfs: error %d while searching for device_stats item for device %s!\n",
+		       ret, device->name);
+		goto out;
+	}
+
+	if (ret == 0 &&
+	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
+		/* need to delete old one and insert a new one */
+		ret = btrfs_del_item(trans, dev_root, path);
+		if (ret != 0) {
+			printk(KERN_WARNING "btrfs: delete too small device_stats item for device %s failed %d!\n",
+			       device->name, ret);
+			goto out;
+		}
+		ret = 1;
+	}
+
+	if (ret == 1) {
+		/* need to insert a new item */
+		btrfs_release_path(path);
+		ret = btrfs_insert_empty_item(trans, dev_root, path,
+					      &key, sizeof(*ptr));
+		if (ret < 0) {
+			printk(KERN_WARNING "btrfs: insert device_stats item for device %s failed %d!\n",
+			       device->name, ret);
+			goto out;
+		}
+	}
+
+	eb = path->nodes[0];
+	ptr = btrfs_item_ptr(eb, path->slots[0],
+			     struct btrfs_device_stats_item);
+	btrfs_set_device_stats_cnt_write_io_errs(eb, ptr,
+		btrfs_device_stat_read(&device->cnt_write_io_errs));
+	btrfs_set_device_stats_cnt_read_io_errs(eb, ptr,
+		btrfs_device_stat_read(&device->cnt_read_io_errs));
+	btrfs_set_device_stats_cnt_flush_io_errs(eb, ptr,
+		btrfs_device_stat_read(&device->cnt_flush_io_errs));
+	btrfs_set_device_stats_cnt_corruption_errs(eb, ptr,
+		btrfs_device_stat_read(&device->cnt_corruption_errs));
+	btrfs_set_device_stats_cnt_generation_errs(eb, ptr,
+		btrfs_device_stat_read(&device->cnt_generation_errs));
+	btrfs_mark_buffer_dirty(eb);
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * called from commit_transaction. Writes all changed device stats to disk.
+ */
+int btrfs_run_device_stats(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_root *dev_root = fs_info->dev_root;
+	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
+	struct btrfs_device *device;
+	int ret = 0;
+
+	mutex_lock(&fs_devices->device_list_mutex);
+	list_for_each_entry(device, &fs_devices->devices, dev_list) {
+		if (!device->device_stats_valid || !device->device_stats_dirty)
+			continue;
+
+		ret = update_device_stat_item(trans, dev_root, device);
+		if (!ret)
+			device->device_stats_dirty = 0;
+	}
+	mutex_unlock(&fs_devices->device_list_mutex);
+
+	return ret;
+}
+
 void btrfs_device_stat_print_on_error(struct btrfs_device *device)
 {
+	if (!device->device_stats_valid)
+		return;
 	printk_ratelimited(KERN_ERR
 			   "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
 			   device->name,
@@ -4639,6 +4828,18 @@ void btrfs_device_stat_print_on_error(struct btrfs_device *device)
 				&device->cnt_generation_errs));
 }
 
+static void btrfs_device_stat_print_on_load(struct btrfs_device *device)
+{
+	printk(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u,"
+	       " corrupt %u, gen %u\n",
+	       device->name,
+	       btrfs_device_stat_read(&device->cnt_write_io_errs),
+	       btrfs_device_stat_read(&device->cnt_read_io_errs),
+	       btrfs_device_stat_read(&device->cnt_flush_io_errs),
+	       btrfs_device_stat_read(&device->cnt_corruption_errs),
+	       btrfs_device_stat_read(&device->cnt_generation_errs));
+}
+
 int btrfs_get_device_stats(struct btrfs_root *root,
 			   struct btrfs_ioctl_get_device_stats *stats,
 			   int reset_after_read)
@@ -4654,6 +4855,10 @@ int btrfs_get_device_stats(struct btrfs_root *root,
 		printk(KERN_WARNING
 		       "btrfs: get device_stats failed, device not found\n");
 		return -ENODEV;
+	} else if (!dev->device_stats_valid) {
+		printk(KERN_WARNING
+		       "btrfs: get device_stats failed, not yet valid\n");
+		return -ENODEV;
 	} else if (reset_after_read) {
 		if (stats->nr_items >= 1)
 			stats->cnt_write_io_errs =
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index e0b31f1..3134662 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -108,6 +108,7 @@ struct btrfs_device {
 
 	/* disk I/O failure stats. For detailed description refer to
 	 * struct btrfs_device_stats_item in ctree.h */
+	int device_stats_valid;
 	int device_stats_dirty; /* counters need to be written to disk */
 	atomic_t cnt_write_io_errs;
 	atomic_t cnt_read_io_errs;
@@ -291,6 +292,9 @@ int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 			 u64 *start, u64 *max_avail);
 struct btrfs_device *btrfs_find_device_for_logical(struct btrfs_root *root,
 						   u64 logical, int mirror_num);
+int btrfs_init_device_stats(struct btrfs_fs_info *fs_info);
+int btrfs_run_device_stats(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info);
 void btrfs_device_stat_print_on_error(struct btrfs_device *device);
 int btrfs_get_device_stats(struct btrfs_root *root,
 			   struct btrfs_ioctl_get_device_stats *stats,
@@ -315,4 +319,9 @@ static inline void btrfs_device_stat_reset(atomic_t *cnt)
 {
 	atomic_set(cnt, 0);
 }
+
+static inline void btrfs_device_stat_set(atomic_t *cnt, unsigned long val)
+{
+	atomic_set(cnt, val);
+}
 #endif
-- 
1.7.10.2

Re: [PATCH v4 3/3] Btrfs: read device stats on mount, write modified ones during commit

[Liu Bo]

On 05/22/2012 06:53 PM, Stefan Behrens wrote:

> The device statistics are written into the device tree with each
> transaction commit. Only modified statistics are written.
> When a filesystem is mounted, the device statistics for each involved
> device are read from the device tree and used to initialize the
> counters.
> 
> Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
> ---
>  fs/btrfs/ctree.h       |   51 ++++++++++++
>  fs/btrfs/disk-io.c     |    7 ++
>  fs/btrfs/print-tree.c  |    3 +
>  fs/btrfs/transaction.c |    4 +
>  fs/btrfs/volumes.c     |  205 ++++++++++++++++++++++++++++++++++++++++++++++++
>  fs/btrfs/volumes.h     |    9 +++
>  6 files changed, 279 insertions(+)
> 
[...]
> +static int update_device_stat_item(struct btrfs_trans_handle *trans,
> +				   struct btrfs_root *dev_root,
> +				   struct btrfs_device *device)
> +{
> +	struct btrfs_path *path;
> +	struct btrfs_key key;
> +	struct extent_buffer *eb;
> +	struct btrfs_device_stats_item *ptr;
> +	int ret;
> +
> +	key.objectid = 0;
> +	key.type = BTRFS_DEVICE_STATS_KEY;
> +	key.offset = device->devid;
> +
> +	path = btrfs_alloc_path();
> +	BUG_ON(!path);
> +	ret = btrfs_search_slot(trans, dev_root, &key, path, 0, 1);


Since we may delete this item, I prefer cow: -1,

btrfs_search_slot(trans, dev_root, &key, path, 0, -1);

thanks,
liubo

> +	if (ret < 0) {
> +		printk(KERN_WARNING "btrfs: error %d while searching for device_stats item for device %s!\n",
> +		       ret, device->name);
> +		goto out;
> +	}
> +
> +	if (ret == 0 &&
> +	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
> +		/* need to delete old one and insert a new one */
> +		ret = btrfs_del_item(trans, dev_root, path);
> +		if (ret != 0) {
> +			printk(KERN_WARNING "btrfs: delete too small device_stats item for device %s failed %d!\n",
> +			       device->name, ret);
> +			goto out;
> +		}
> +		ret = 1;
> +	}
> +
> +	if (ret == 1) {
> +		/* need to insert a new item */
> +		btrfs_release_path(path);
> +		ret = btrfs_insert_empty_item(trans, dev_root, path,
> +					      &key, sizeof(*ptr));
> +		if (ret < 0) {
> +			printk(KERN_WARNING "btrfs: insert device_stats item for device %s failed %d!\n",
> +			       device->name, ret);
> +			goto out;
> +		}
> +	}
> +
> +	eb = path->nodes[0];
> +	ptr = btrfs_item_ptr(eb, path->slots[0],
> +			     struct btrfs_device_stats_item);
> +	btrfs_set_device_stats_cnt_write_io_errs(eb, ptr,
> +		btrfs_device_stat_read(&device->cnt_write_io_errs));
> +	btrfs_set_device_stats_cnt_read_io_errs(eb, ptr,
> +		btrfs_device_stat_read(&device->cnt_read_io_errs));
> +	btrfs_set_device_stats_cnt_flush_io_errs(eb, ptr,
> +		btrfs_device_stat_read(&device->cnt_flush_io_errs));
> +	btrfs_set_device_stats_cnt_corruption_errs(eb, ptr,
> +		btrfs_device_stat_read(&device->cnt_corruption_errs));
> +	btrfs_set_device_stats_cnt_generation_errs(eb, ptr,
> +		btrfs_device_stat_read(&device->cnt_generation_errs));
> +	btrfs_mark_buffer_dirty(eb);
> +
> +out:
> +	btrfs_free_path(path);
> +	return ret;
> +}
> +
> +/*
> + * called from commit_transaction. Writes all changed device stats to disk.
> + */
> +int btrfs_run_device_stats(struct btrfs_trans_handle *trans,
> +			   struct btrfs_fs_info *fs_info)
> +{
> +	struct btrfs_root *dev_root = fs_info->dev_root;
> +	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
> +	struct btrfs_device *device;
> +	int ret = 0;
> +
> +	mutex_lock(&fs_devices->device_list_mutex);
> +	list_for_each_entry(device, &fs_devices->devices, dev_list) {
> +		if (!device->device_stats_valid || !device->device_stats_dirty)
> +			continue;
> +
> +		ret = update_device_stat_item(trans, dev_root, device);
> +		if (!ret)
> +			device->device_stats_dirty = 0;
> +	}
> +	mutex_unlock(&fs_devices->device_list_mutex);
> +
> +	return ret;
> +}
> +
>  void btrfs_device_stat_print_on_error(struct btrfs_device *device)
>  {
> +	if (!device->device_stats_valid)
> +		return;
>  	printk_ratelimited(KERN_ERR
>  			   "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
>  			   device->name,
> @@ -4639,6 +4828,18 @@ void btrfs_device_stat_print_on_error(struct btrfs_device *device)
>  				&device->cnt_generation_errs));
>  }
>  
> +static void btrfs_device_stat_print_on_load(struct btrfs_device *device)
> +{
> +	printk(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u,"
> +	       " corrupt %u, gen %u\n",
> +	       device->name,
> +	       btrfs_device_stat_read(&device->cnt_write_io_errs),
> +	       btrfs_device_stat_read(&device->cnt_read_io_errs),
> +	       btrfs_device_stat_read(&device->cnt_flush_io_errs),
> +	       btrfs_device_stat_read(&device->cnt_corruption_errs),
> +	       btrfs_device_stat_read(&device->cnt_generation_errs));
> +}
> +
>  int btrfs_get_device_stats(struct btrfs_root *root,
>  			   struct btrfs_ioctl_get_device_stats *stats,
>  			   int reset_after_read)
> @@ -4654,6 +4855,10 @@ int btrfs_get_device_stats(struct btrfs_root *root,
>  		printk(KERN_WARNING
>  		       "btrfs: get device_stats failed, device not found\n");
>  		return -ENODEV;
> +	} else if (!dev->device_stats_valid) {
> +		printk(KERN_WARNING
> +		       "btrfs: get device_stats failed, not yet valid\n");
> +		return -ENODEV;
>  	} else if (reset_after_read) {
>  		if (stats->nr_items >= 1)
>  			stats->cnt_write_io_errs =
> diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
> index e0b31f1..3134662 100644
> --- a/fs/btrfs/volumes.h
> +++ b/fs/btrfs/volumes.h
> @@ -108,6 +108,7 @@ struct btrfs_device {
>  
>  	/* disk I/O failure stats. For detailed description refer to
>  	 * struct btrfs_device_stats_item in ctree.h */
> +	int device_stats_valid;
>  	int device_stats_dirty; /* counters need to be written to disk */
>  	atomic_t cnt_write_io_errs;
>  	atomic_t cnt_read_io_errs;
> @@ -291,6 +292,9 @@ int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
>  			 u64 *start, u64 *max_avail);
>  struct btrfs_device *btrfs_find_device_for_logical(struct btrfs_root *root,
>  						   u64 logical, int mirror_num);
> +int btrfs_init_device_stats(struct btrfs_fs_info *fs_info);
> +int btrfs_run_device_stats(struct btrfs_trans_handle *trans,
> +			   struct btrfs_fs_info *fs_info);
>  void btrfs_device_stat_print_on_error(struct btrfs_device *device);
>  int btrfs_get_device_stats(struct btrfs_root *root,
>  			   struct btrfs_ioctl_get_device_stats *stats,
> @@ -315,4 +319,9 @@ static inline void btrfs_device_stat_reset(atomic_t *cnt)
>  {
>  	atomic_set(cnt, 0);
>  }
> +
> +static inline void btrfs_device_stat_set(atomic_t *cnt, unsigned long val)
> +{
> +	atomic_set(cnt, val);
> +}
>  #endif

Re: [PATCH v4 1/3] Btrfs: add device counters for detected IO and checksum errors

[David Sterba]

On Tue, May 22, 2012 at 12:53:47PM +0200, Stefan Behrens wrote:
>  		 */
>  		clear_buffer_uptodate(bh);
> +		btrfs_device_stat_inc(&device->cnt_write_io_errs);
> +		device->device_stats_dirty = 1;
> +		btrfs_device_stat_print_on_error(device);

this sequence repeats several times (with a slight change), IMHO it asks
for a helper.

void stat_inc(... *device, atomic_inc *cnt) {
	atomic_inc(cntr);
	device->device_stats_dirty = 1;
	btrfs_device_stat_print_on_error(device);
}

usage:

stat_inc(device, &device->cnt_write_io_errs);

I think it's unavoidable to leave out the 'device' parameter without a
macro trickery, but as there are only 2 params, it don't it's needed.


>  static void btrfs_end_bio(struct bio *bio, int err)
>  {
> -	struct btrfs_bio *bbio = bio->bi_private;
> +	struct btrfs_bio *bbio = (struct btrfs_bio *)
> +		(((uintptr_t)bio->bi_private) & ~((uintptr_t)3));

I suggest do use a helper for all the "var & 3" places, and comment why
is it so and what are the limitations. I can see from the code below
that it encodes number of devices and fails for > 3.

>  	int is_orig_bio = 0;
>  
> -	if (err)
> +	if (err) {
>  		atomic_inc(&bbio->error);
> +		if (err == -EIO || err == -EREMOTEIO) {
> +			unsigned int dev_nr = ((uintptr_t)bio->bi_private) & 3;

eg.

unsigned int dev_nr = bio_private_to_nr_dev(bio);

> @@ -4148,7 +4166,9 @@ int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
>  		} else {
>  			bio = first_bio;
>  		}
> -		bio->bi_private = bbio;
> +		BUG_ON((((uintptr_t)bbio) & 3) != 0);
> +		BUG_ON(dev_nr > 3);
> +		bio->bi_private = (void *)(((uintptr_t)bbio) | dev_nr);
>  		bio->bi_end_io = btrfs_end_bio;
>  		bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
>  		dev = bbio->stripes[dev_nr].dev;


otherwise ok,
david

Re: [PATCH v4 2/3] Btrfs: add ioctl to get and reset the device stats

[David Sterba]

On Tue, May 22, 2012 at 12:53:48PM +0200, Stefan Behrens wrote:
> An ioctl interface is added to get the device statistic counters.
> A second ioctl is added to atomically get and reset these counters.
> 
> Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
> ---
>  fs/btrfs/ioctl.c   |   26 ++++++++++++++++++++
>  fs/btrfs/ioctl.h   |   28 +++++++++++++++++++++
>  fs/btrfs/volumes.c |   69 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>  fs/btrfs/volumes.h |   13 ++++++++++
>  4 files changed, 136 insertions(+)
> 
> diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
> index 14f8e1f..19d2244 100644
> --- a/fs/btrfs/ioctl.c
> +++ b/fs/btrfs/ioctl.c
> @@ -3424,6 +3446,10 @@ long btrfs_ioctl(struct file *file, unsigned int
>  		return btrfs_ioctl_balance_ctl(root, arg);
>  	case BTRFS_IOC_BALANCE_PROGRESS:
>  		return btrfs_ioctl_balance_progress(root, argp);
> +	case BTRFS_IOC_GET_DEVICE_STATS:
> +		return btrfs_ioctl_get_device_stats(root, argp, 0);
> +	case BTRFS_IOC_GET_AND_RESET_DEVICE_STATS:
> +		return btrfs_ioctl_get_device_stats(root, argp, 1);
>  	}
>  
>  	return -ENOTTY;
> diff --git a/fs/btrfs/ioctl.h b/fs/btrfs/ioctl.h
> index 086e6bd..f1c1196 100644
> --- a/fs/btrfs/ioctl.h
> +++ b/fs/btrfs/ioctl.h
> @@ -266,6 +266,30 @@ struct btrfs_ioctl_logical_ino_args {
>  	__u64				inodes;
>  };
>  
> +#define BTRFS_IOCTL_GET_DEVICE_STATS_MAX_NR_ITEMS	5

The check at the end of btrfs_get_device_stats() should use this number:

> +       if (stats->nr_items > 5)
> +               stats->nr_items = 5;
> +       return 0;
> +}


> +struct btrfs_ioctl_get_device_stats {
> +	__u64 devid;				/* in */
> +	__u64 nr_items;				/* in/out */
> +
> +	/* out values: */
> +
> +	/* disk I/O failure stats */
> +	__u64 cnt_write_io_errs; /* EIO or EREMOTEIO from lower layers */
> +	__u64 cnt_read_io_errs; /* EIO or EREMOTEIO from lower layers */
> +	__u64 cnt_flush_io_errs; /* EIO or EREMOTEIO from lower layers */
> +
> +	/* stats for indirect indications for I/O failures */
> +	__u64 cnt_corruption_errs; /* checksum error, bytenr error or
> +				    * contents is illegal: this is an
> +				    * indication that the block was damaged
> +				    * during read or write, or written to
> +				    * wrong location or read from wrong
> +				    * location */
> +	__u64 cnt_generation_errs; /* an indication that blocks have not
> +				    * been written */
> +	__u64 unused[121]; /* pad to 1k */
> +};

Padding has landed, and thanks for the explanation in the V3, I see how
the compatibility works.

> +
>  #define BTRFS_IOC_SNAP_CREATE _IOW(BTRFS_IOCTL_MAGIC, 1, \
>  				   struct btrfs_ioctl_vol_args)
>  #define BTRFS_IOC_DEFRAG _IOW(BTRFS_IOCTL_MAGIC, 2, \
> @@ -330,5 +354,9 @@ struct btrfs_ioctl_logical_ino_args {
>  					struct btrfs_ioctl_ino_path_args)
>  #define BTRFS_IOC_LOGICAL_INO _IOWR(BTRFS_IOCTL_MAGIC, 36, \
>  					struct btrfs_ioctl_ino_path_args)
> +#define BTRFS_IOC_GET_DEVICE_STATS _IOWR(BTRFS_IOCTL_MAGIC, 52, \
> +					 struct btrfs_ioctl_get_device_stats)
> +#define BTRFS_IOC_GET_AND_RESET_DEVICE_STATS _IOWR(BTRFS_IOCTL_MAGIC, 53, \
> +					 struct btrfs_ioctl_get_device_stats)

I think that 'reset device' should go into the structure, besides the
ioctl number there is no difference.

Then the permission check will look like:

> @@ -3042,6 +3042,28 @@ static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
>  	return ret;
>  }
>  
> +static long btrfs_ioctl_get_device_stats(struct btrfs_root *root,
> +					 void __user *arg, int reset_after_read)

					 void __user *arg)

> +{
> +	struct btrfs_ioctl_get_device_stats *sa;
> +	int ret;
> +
> +	sa = memdup_user(arg, sizeof(*sa));
> +	if (IS_ERR(sa))
> +		return PTR_ERR(sa);
> +
> +	if (reset_after_read && !capable(CAP_SYS_ADMIN))

	if (sa->reset_after_read && !capable(CAP_SYS_ADMIN)) {
		kfree(sa);

> +		return -EPERM;

	}
> +
> +	ret = btrfs_get_device_stats(root, sa, reset_after_read);
> +
> +	if (copy_to_user(arg, sa, sizeof(*sa)))
> +		ret = -EFAULT;
> +
> +	kfree(sa);
> +	return ret;
> +}

So the EINVAL could come earlier than EPERM, but I don't think it's a
problem.


david

Re: [PATCH v4 3/3] Btrfs: read device stats on mount, write modified ones during commit

[Stefan Behrens]

On Wed, 23 May 2012 09:29:12 +0800, Liu Bo wrote:
> On 05/22/2012 06:53 PM, Stefan Behrens wrote:
> 
>> The device statistics are written into the device tree with each
>> transaction commit. Only modified statistics are written.
>> When a filesystem is mounted, the device statistics for each involved
>> device are read from the device tree and used to initialize the
>> counters.
>>
>> Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
>> ---
>>  fs/btrfs/ctree.h       |   51 ++++++++++++
>>  fs/btrfs/disk-io.c     |    7 ++
>>  fs/btrfs/print-tree.c  |    3 +
>>  fs/btrfs/transaction.c |    4 +
>>  fs/btrfs/volumes.c     |  205 ++++++++++++++++++++++++++++++++++++++++++++++++
>>  fs/btrfs/volumes.h     |    9 +++
>>  6 files changed, 279 insertions(+)
>>
> [...]
>> +static int update_device_stat_item(struct btrfs_trans_handle *trans,
>> +				   struct btrfs_root *dev_root,
>> +				   struct btrfs_device *device)
>> +{
>> +	struct btrfs_path *path;
>> +	struct btrfs_key key;
>> +	struct extent_buffer *eb;
>> +	struct btrfs_device_stats_item *ptr;
>> +	int ret;
>> +
>> +	key.objectid = 0;
>> +	key.type = BTRFS_DEVICE_STATS_KEY;
>> +	key.offset = device->devid;
>> +
>> +	path = btrfs_alloc_path();
>> +	BUG_ON(!path);
>> +	ret = btrfs_search_slot(trans, dev_root, &key, path, 0, 1);
> 
> 
> Since we may delete this item, I prefer cow: -1,
> 
> btrfs_search_slot(trans, dev_root, &key, path, 0, -1);
> 
> thanks,
> liubo

I am sure you mean to set ins_len (the 5th parameter) to -1 and you are
right. Thanks for this finding! I will send a v5.


>> +	if (ret < 0) {
>> +		printk(KERN_WARNING "btrfs: error %d while searching for device_stats item for device %s!\n",
>> +		       ret, device->name);
>> +		goto out;
>> +	}
>> +
>> +	if (ret == 0 &&
>> +	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
>> +		/* need to delete old one and insert a new one */
>> +		ret = btrfs_del_item(trans, dev_root, path);
>> +		if (ret != 0) {
>> +			printk(KERN_WARNING "btrfs: delete too small device_stats item for device %s failed %d!\n",
>> +			       device->name, ret);
>> +			goto out;
>> +		}
>> +		ret = 1;
>> +	}
>> +
>> +	if (ret == 1) {
>> +		/* need to insert a new item */
>> +		btrfs_release_path(path);
>> +		ret = btrfs_insert_empty_item(trans, dev_root, path,
>> +					      &key, sizeof(*ptr));
>> +		if (ret < 0) {
>> +			printk(KERN_WARNING "btrfs: insert device_stats item for device %s failed %d!\n",
>> +			       device->name, ret);
>> +			goto out;
>> +		}
>> +	}
>> +
>> +	eb = path->nodes[0];
>> +	ptr = btrfs_item_ptr(eb, path->slots[0],
>> +			     struct btrfs_device_stats_item);
>> +	btrfs_set_device_stats_cnt_write_io_errs(eb, ptr,
>> +		btrfs_device_stat_read(&device->cnt_write_io_errs));
>> +	btrfs_set_device_stats_cnt_read_io_errs(eb, ptr,
>> +		btrfs_device_stat_read(&device->cnt_read_io_errs));
>> +	btrfs_set_device_stats_cnt_flush_io_errs(eb, ptr,
>> +		btrfs_device_stat_read(&device->cnt_flush_io_errs));
>> +	btrfs_set_device_stats_cnt_corruption_errs(eb, ptr,
>> +		btrfs_device_stat_read(&device->cnt_corruption_errs));
>> +	btrfs_set_device_stats_cnt_generation_errs(eb, ptr,
>> +		btrfs_device_stat_read(&device->cnt_generation_errs));
>> +	btrfs_mark_buffer_dirty(eb);
>> +
>> +out:
>> +	btrfs_free_path(path);
>> +	return ret;
>> +}
[...]