From: Qu Wenruo <wqu@suse.com>
To: Johannes Thumshirn <johannes.thumshirn@wdc.com>,
linux-btrfs@vger.kernel.org
Cc: David Sterba <dsterba@suse.com>,
Naohiro Aota <naohiro.aota@wdc.com>,
Damien Le Moal <dlemoal@kernel.org>,
Mark Korenberg <socketpair@gmail.com>,
Dongjiang Zhu <zhudongjiang@fygo.io>
Subject: Re: [PATCH 4/4] btrfs-progs: check: detect and repair exhausted zoned active zones
Date: Wed, 18 Nov 2026 18:10:59 +1030 [thread overview]
Message-ID: <5739cbcc-bb71-4e9f-95c3-bd2e18d44e53@suse.com> (raw)
In-Reply-To: <20261117084948.3848762-5-johannes.thumshirn@wdc.com>
在 2026/11/17 19:19, Johannes Thumshirn 写道:
> Kernels before "btrfs: zoned: don't release active zone reservations
> for conventional zones" can leave all active zones to data. The first
> transaction commit after mount then can't activate a metadata
> block-group and aborts with -EAGAIN.
>
> Report when the active zones in use exceed the kernel's limit minus its
> metadata reservation. With --repair, finish the fullest single profile
> data zones until it fits again, plus one zone for the relocation
> block-group.
>
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> ---
> check/main.c | 187 +++++++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 187 insertions(+)
>
> diff --git a/check/main.c b/check/main.c
> index 717d576f..f065555f 100644
> --- a/check/main.c
> +++ b/check/main.c
> @@ -9203,6 +9203,188 @@ static int check_zoned_active_zones(void)
>
> return ret;
> }
> +
> +static bool zone_is_active(const struct blk_zone *zone)
> +{
> + if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
> + return false;
> +
> + switch (zone->cond) {
> + case BLK_ZONE_COND_IMP_OPEN:
> + case BLK_ZONE_COND_EXP_OPEN:
> + case BLK_ZONE_COND_CLOSED:
> + case BLK_ZONE_COND_ACTIVE:
> + return true;
> + default:
> + return false;
> + }
> +}
> +
> +struct finish_candidate {
> + u32 zno;
> + u64 left;
> +};
> +
> +static int cmp_finish_candidate(const void *a, const void *b)
> +{
> + const struct finish_candidate *ca = a, *cb = b;
> +
> + return ca->left < cb->left ? -1 : ca->left > cb->left;
> +}
> +
> +/*
> + * The kernel keeps active zones in reserve for a metadata, a tree-log and a
> + * system block-group, and only activates metadata block-groups when writing
> + * them. If more zones are active on disk than the kernel's limit minus that
> + * reservation, the first transaction commit after mount can't activate a
> + * metadata block-group and aborts with -EAGAIN. Older kernels could leave
> + * filesystems in that state.
> + *
> + * Mirror the kernel's limit and reservation, see btrfs_get_max_active_zones()
> + * and btrfs_check_active_zone_reservation(), and finish the zones of the
> + * fullest single profile data block-groups until it fits again. Leave one
> + * zone more, for the data relocation block-group activated on mount.
> + *
> + * Return 0 if fine or repaired, 1 if a problem was found, <0 on error.
> + */
> +static int check_dev_active_reservation(struct btrfs_device *dev, int reserved)
> +{
> + struct btrfs_root *dev_root = gfs_info->dev_root;
> + struct btrfs_zoned_device_info *zinfo = dev->zone_info;
> + struct finish_candidate *cand;
> + BTRFS_PATH_AUTO_RELEASE(path);
> + struct btrfs_key key;
> + int max_active, nactive = 0, ncand = 0;
> + int shift = ilog2(zinfo->zone_size);
> + int left, need;
> + int ret;
> +
> + max_active = btrfs_get_max_active_zones(dev);
> + for (int i = 0; i < zinfo->nr_zones; i++)
> + nactive += zone_is_active(&zinfo->zones[i]);
> + /* The kernel disables active zone tracking in this case. */
> + if (nactive > max_active && !zinfo->max_active_zones)
> + return 0;
> +
> + cand = calloc(zinfo->nr_zones, sizeof(*cand));
> + if (!cand)
> + return -ENOMEM;
> +
> + key.objectid = dev->devid;
> + key.type = BTRFS_DEV_EXTENT_KEY;
> + key.offset = 0;
> + ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
> + if (ret < 0)
> + goto out;
> +
> + while (1) {
> + struct btrfs_block_group *bg;
> + struct btrfs_dev_extent *devext;
> + u64 pstart, plen, chunk;
> +
> + if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
> + ret = btrfs_next_leaf(dev_root, &path);
> + if (ret < 0)
> + goto out;
> + if (ret > 0)
> + break;
> + }
> +
> + btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
> + if (key.objectid != dev->devid || key.type != BTRFS_DEV_EXTENT_KEY)
> + break;
> +
> + devext = btrfs_item_ptr(path.nodes[0], path.slots[0],
> + struct btrfs_dev_extent);
> + pstart = key.offset;
> + plen = btrfs_dev_extent_length(path.nodes[0], devext);
> + chunk = btrfs_dev_extent_chunk_offset(path.nodes[0], devext);
> + bg = btrfs_lookup_block_group(gfs_info, chunk);
> + path.slots[0]++;
> + if (!bg)
> + continue;
> +
> + for (u64 zno = pstart >> shift;
> + zno <= (pstart + plen - 1) >> shift && zno < zinfo->nr_zones;
> + zno++) {
> + struct blk_zone *zone = &zinfo->zones[zno];
> +
> + if (!zone_is_active(zone))
> + continue;
> + if (bg->flags & (BTRFS_BLOCK_GROUP_METADATA |
> + BTRFS_BLOCK_GROUP_SYSTEM)) {
> + reserved--;
> + } else if (!(bg->flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
> + cand[ncand].zno = zno;
> + cand[ncand].left = (zone->start + zone->capacity -
> + zone->wp) << SECTOR_SHIFT;
> + ncand++;
> + }
> + }
> + }
> +
> + ret = 0;
> + reserved = max(reserved, 0);
> + left = max_active - nactive;
> + if (left >= reserved)
> + goto out;
> + need = reserved + 1 - left;
> +
> + ret = 1;
> + error(
> +"zoned: devid %llu has %d of %d active zones in use and %d reserved for metadata, the first commit after mount can fail",
> + dev->devid, nactive, max_active, reserved);
> + if (!opt_check_repair)
> + goto out;
> +
> + if (ncand < need) {
> + error("zoned: only %d data zones on devid %llu can be finished, need %d",
> + ncand, dev->devid, need);
> + goto out;
> + }
> +
> + qsort(cand, ncand, sizeof(*cand), cmp_finish_candidate);
> + for (int i = 0; i < need; i++) {
> + struct blk_zone *zone = &zinfo->zones[cand[i].zno];
> +
> + ret = btrfs_finish_dev_zone(dev->fd, zone);
Shoudln't zinfo->active_zones or active_zones_left also to be updated
here or by the btrfs_finish_dev_zone() itself?
> + if (ret) {
> + errno = -ret;
> + error("zoned: failed to finish zone %u on devid %llu: %m",
> + cand[i].zno, dev->devid);
> + goto out;
> + }
> + printf("Finished zone %u on devid %llu, %llu bytes left unused\n",
> + cand[i].zno, dev->devid, cand[i].left);
> + }
> +out:
> + free(cand);
> + return ret;
> +}
> +
> +static int check_zoned_active_reservation(void)
> +{
> + struct btrfs_device *dev;
> + int reserved;
> + int ret = 0;
> +
> + reserved = (gfs_info->avail_metadata_alloc_bits & BTRFS_BLOCK_GROUP_DUP) ? 4 : 2;
> + reserved += (gfs_info->avail_system_alloc_bits & BTRFS_BLOCK_GROUP_DUP) ? 2 : 1;
> +
> + list_for_each_entry(dev, &gfs_info->fs_devices->devices, dev_list) {
> + int err;
> +
> + if (!dev->zone_info || dev->zone_info->model != ZONED_HOST_MANAGED)
> + continue;
> +
> + err = check_dev_active_reservation(dev, reserved);
> + if (err < 0)
> + return err;
> + ret |= err;
> + }
> +
> + return ret;
> +}
> #endif
>
> /*
> @@ -9219,6 +9401,7 @@ static int check_zoned_active_zones(void)
> static int check_zoned(void)
> {
> int ret = 0;
> + int err __maybe_unused;
>
> if (!btrfs_is_zoned(gfs_info))
> return 0;
> @@ -9227,6 +9410,10 @@ static int check_zoned(void)
> ret = check_zoned_active_zones();
> if (ret < 0)
> return ret;
> + err = check_zoned_active_reservation();
> + if (err < 0)
> + return err;
> + ret |= err;
> #endif
> return ret;
> }
prev parent reply other threads:[~2026-11-18 7:41 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-17 8:49 [PATCH 0/4] btrfs-progs: zoned: detect and repair FS with exhausted active zones Johannes Thumshirn
2026-11-17 8:49 ` [PATCH 1/4] btrfs-progs: zoned: use cached zone reports if available Johannes Thumshirn
2026-11-17 11:59 ` Damien Le Moal
2026-11-17 8:49 ` [PATCH 2/4] btrfs-progs: zoned: add btrfs_finish_dev_zone() Johannes Thumshirn
2026-11-17 11:59 ` Damien Le Moal
2026-11-17 8:49 ` [PATCH 3/4] btrfs-progs: zoned: add btrfs_get_max_active_zones() Johannes Thumshirn
2026-11-17 12:03 ` Damien Le Moal
2026-11-17 8:49 ` [PATCH 4/4] btrfs-progs: check: detect and repair exhausted zoned active zones Johannes Thumshirn
2026-11-18 7:40 ` Qu Wenruo [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=5739cbcc-bb71-4e9f-95c3-bd2e18d44e53@suse.com \
--to=wqu@suse.com \
--cc=dlemoal@kernel.org \
--cc=dsterba@suse.com \
--cc=johannes.thumshirn@wdc.com \
--cc=linux-btrfs@vger.kernel.org \
--cc=naohiro.aota@wdc.com \
--cc=socketpair@gmail.com \
--cc=zhudongjiang@fygo.io \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.