On Wed, Jun 09, 2021 at 08:50:14PM +0200, David Sterba wrote:

On Wed, Jun 09, 2021 at 11:24:21AM -0700, Omar Sandoval wrote:

quoted

On Fri, Jun 04, 2021 at 03:20:58PM +0200, David Sterba wrote:

quoted

The device stats can be read by ioctl, wrapped by command 'btrfs device
stats'. Provide another source where to read the information in
/sys/fs/btrfs/FSID/devinfo/DEVID/stats . The format is a list of
'key value' pairs one per line, which is common in other stat files.
The names are the same as used in other device stat outputs.

The stats are all in one file as it's the snapshot of all available
stats. The 'one value per file' is not very suitable here. The stats
should be valid right after the stats item is read from disk, shortly
after initializing the device, but in any case also print the validity
status.

Signed-off-by: David Sterba <dsterba@suse.com>
---
 fs/btrfs/sysfs.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/fs/btrfs/sysfs.c b/fs/btrfs/sysfs.c
index 4b508938e728..3d4c806c4f73 100644
--- a/fs/btrfs/sysfs.c
+++ b/fs/btrfs/sysfs.c

@@ -1495,11 +1495,39 @@ static ssize_t btrfs_devinfo_writeable_show(struct kobject *kobj,
 }
 BTRFS_ATTR(devid, writeable, btrfs_devinfo_writeable_show);
 
+static ssize_t btrfs_devinfo_stats_show(struct kobject *kobj,
+		struct kobj_attribute *a, char *buf)
+{
+	struct btrfs_device *device = container_of(kobj, struct btrfs_device,
+						   devid_kobj);
+
+	/*
+	 * Print all at once so we get a snapshot of all values from the same
+	 * time. Keep them in sync and in order of definition of
+	 * btrfs_dev_stat_values.
+	 */
+	return scnprintf(buf, PAGE_SIZE,
+		"stats_valid %d\n",
+		"write_errs %d\n"
+		"read_errs %d\n"
+		"flush_errs %d\n"
+		"corruption_errs %d\n"
+		"generation_errs %d\n",
+		!!(device->dev_stats_valid),

The ioctl returns ENODEV is !dev_stats_valid, maybe this file should do
the same? It seems a little awkward to have a flag that means that the
rest of the file is meaningless.

You mean returning -ENODEV when reading the stats file? Or return 0 but
the contents is something like 'stats invalid' or similar.

I'd vote for returning -ENODEV when reading the stats file, but I think
either one is fine.