From d5b81ced74afded85619ffbbe9c32ba9d82c9b1e Mon Sep 17 00:00:00 2001 From: Naohiro Aota Date: Wed, 31 Aug 2022 13:55:48 +0900 Subject: btrfs: zoned: fix API misuse of zone finish waiting The commit 2ce543f47843 ("btrfs: zoned: wait until zone is finished when allocation didn't progress") implemented a zone finish waiting mechanism to the write path of zoned mode. However, using wait_var_event()/wake_up_all() on fs_info->zone_finish_wait is wrong and wait_var_event() just hangs because no one ever wakes it up once it goes into sleep. Instead, we can simply use wait_on_bit_io() and clear_and_wake_up_bit() on fs_info->flags with a proper barrier installed. Fixes: 2ce543f47843 ("btrfs: zoned: wait until zone is finished when allocation didn't progress") CC: stable@vger.kernel.org # 5.16+ Signed-off-by: Naohiro Aota Signed-off-by: David Sterba --- fs/btrfs/zoned.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'fs/btrfs/zoned.c') diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c index b150b07ba1a7..fb25ba7fcb2a 100644 --- a/fs/btrfs/zoned.c +++ b/fs/btrfs/zoned.c @@ -2007,8 +2007,7 @@ static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_writ /* For active_bg_list */ btrfs_put_block_group(block_group); - clear_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags); - wake_up_all(&fs_info->zone_finish_wait); + clear_and_wake_up_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags); return 0; } -- cgit From cac5c44c48c9fb9cc31bea15ebd9ef0c6462314f Mon Sep 17 00:00:00 2001 From: Shin'ichiro Kawasaki Date: Fri, 26 Aug 2022 16:42:15 +0900 Subject: btrfs: zoned: set pseudo max append zone limit in zone emulation mode The commit 7d7672bc5d10 ("btrfs: convert count_max_extents() to use fs_info->max_extent_size") introduced a division by fs_info->max_extent_size. This max_extent_size is initialized with max zone append limit size of the device btrfs runs on. However, in zone emulation mode, the device is not zoned then its zone append limit is zero. This resulted in zero value of fs_info->max_extent_size and caused zero division error. Fix the error by setting non-zero pseudo value to max append zone limit in zone emulation mode. Set the pseudo value based on max_segments as suggested in the commit c2ae7b772ef4 ("btrfs: zoned: revive max_zone_append_bytes"). Fixes: 7d7672bc5d10 ("btrfs: convert count_max_extents() to use fs_info->max_extent_size") CC: stable@vger.kernel.org # 5.12+ Reviewed-by: Johannes Thumshirn Reviewed-by: Naohiro Aota Signed-off-by: Shin'ichiro Kawasaki Signed-off-by: David Sterba --- fs/btrfs/zoned.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) (limited to 'fs/btrfs/zoned.c') diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c index fb25ba7fcb2a..f576e30ceb33 100644 --- a/fs/btrfs/zoned.c +++ b/fs/btrfs/zoned.c @@ -421,10 +421,19 @@ int btrfs_get_dev_zone_info(struct btrfs_device *device, bool populate_cache) * since btrfs adds the pages one by one to a bio, and btrfs cannot * increase the metadata reservation even if it increases the number of * extents, it is safe to stick with the limit. + * + * With the zoned emulation, we can have non-zoned device on the zoned + * mode. In this case, we don't have a valid max zone append size. So, + * use max_segments * PAGE_SIZE as the pseudo max_zone_append_size. */ - zone_info->max_zone_append_size = - min_t(u64, (u64)bdev_max_zone_append_sectors(bdev) << SECTOR_SHIFT, - (u64)bdev_max_segments(bdev) << PAGE_SHIFT); + if (bdev_is_zoned(bdev)) { + zone_info->max_zone_append_size = min_t(u64, + (u64)bdev_max_zone_append_sectors(bdev) << SECTOR_SHIFT, + (u64)bdev_max_segments(bdev) << PAGE_SHIFT); + } else { + zone_info->max_zone_append_size = + (u64)bdev_max_segments(bdev) << PAGE_SHIFT; + } if (!IS_ALIGNED(nr_sectors, zone_sectors)) zone_info->nr_zones++; -- cgit From 6ca64ac2763149fb66c0b4bf12f5e0977a88e51d Mon Sep 17 00:00:00 2001 From: Johannes Thumshirn Date: Mon, 5 Sep 2022 05:38:24 -0700 Subject: btrfs: zoned: fix mounting with conventional zones Since commit 6a921de58992 ("btrfs: zoned: introduce space_info->active_total_bytes"), we're only counting the bytes of a block group on an active zone as usable for metadata writes. But on a SMR drive, we don't have active zones and short circuit some of the logic. This leads to an error on mount, because we cannot reserve space for metadata writes. Fix this by also setting the BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE bit in the block-group's runtime flag if the zone is a conventional zone. Fixes: 6a921de58992 ("btrfs: zoned: introduce space_info->active_total_bytes") Signed-off-by: Johannes Thumshirn Signed-off-by: David Sterba --- fs/btrfs/zoned.c | 81 ++++++++++++++++++++++++++++---------------------------- 1 file changed, 40 insertions(+), 41 deletions(-) (limited to 'fs/btrfs/zoned.c') diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c index f576e30ceb33..62e7007a7e46 100644 --- a/fs/btrfs/zoned.c +++ b/fs/btrfs/zoned.c @@ -1187,7 +1187,7 @@ int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size) * offset. */ static int calculate_alloc_pointer(struct btrfs_block_group *cache, - u64 *offset_ret) + u64 *offset_ret, bool new) { struct btrfs_fs_info *fs_info = cache->fs_info; struct btrfs_root *root; @@ -1197,6 +1197,21 @@ static int calculate_alloc_pointer(struct btrfs_block_group *cache, int ret; u64 length; + /* + * Avoid tree lookups for a new block group, there's no use for it. + * It must always be 0. + * + * Also, we have a lock chain of extent buffer lock -> chunk mutex. + * For new a block group, this function is called from + * btrfs_make_block_group() which is already taking the chunk mutex. + * Thus, we cannot call calculate_alloc_pointer() which takes extent + * buffer locks to avoid deadlock. + */ + if (new) { + *offset_ret = 0; + return 0; + } + path = btrfs_alloc_path(); if (!path) return -ENOMEM; @@ -1332,6 +1347,13 @@ int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) else num_conventional++; + /* + * Consider a zone as active if we can allow any number of + * active zones. + */ + if (!device->zone_info->max_active_zones) + __set_bit(i, active); + if (!is_sequential) { alloc_offsets[i] = WP_CONVENTIONAL; continue; @@ -1398,45 +1420,23 @@ int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) __set_bit(i, active); break; } - - /* - * Consider a zone as active if we can allow any number of - * active zones. - */ - if (!device->zone_info->max_active_zones) - __set_bit(i, active); } if (num_sequential > 0) cache->seq_zone = true; if (num_conventional > 0) { - /* - * Avoid calling calculate_alloc_pointer() for new BG. It - * is no use for new BG. It must be always 0. - * - * Also, we have a lock chain of extent buffer lock -> - * chunk mutex. For new BG, this function is called from - * btrfs_make_block_group() which is already taking the - * chunk mutex. Thus, we cannot call - * calculate_alloc_pointer() which takes extent buffer - * locks to avoid deadlock. - */ - /* Zone capacity is always zone size in emulation */ cache->zone_capacity = cache->length; - if (new) { - cache->alloc_offset = 0; - goto out; - } - ret = calculate_alloc_pointer(cache, &last_alloc); - if (ret || map->num_stripes == num_conventional) { - if (!ret) - cache->alloc_offset = last_alloc; - else - btrfs_err(fs_info, + ret = calculate_alloc_pointer(cache, &last_alloc, new); + if (ret) { + btrfs_err(fs_info, "zoned: failed to determine allocation offset of bg %llu", - cache->start); + cache->start); + goto out; + } else if (map->num_stripes == num_conventional) { + cache->alloc_offset = last_alloc; + cache->zone_is_active = 1; goto out; } } @@ -1504,13 +1504,6 @@ int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) goto out; } - if (cache->zone_is_active) { - btrfs_get_block_group(cache); - spin_lock(&fs_info->zone_active_bgs_lock); - list_add_tail(&cache->active_bg_list, &fs_info->zone_active_bgs); - spin_unlock(&fs_info->zone_active_bgs_lock); - } - out: if (cache->alloc_offset > fs_info->zone_size) { btrfs_err(fs_info, @@ -1535,10 +1528,16 @@ out: ret = -EIO; } - if (!ret) + if (!ret) { cache->meta_write_pointer = cache->alloc_offset + cache->start; - - if (ret) { + if (cache->zone_is_active) { + btrfs_get_block_group(cache); + spin_lock(&fs_info->zone_active_bgs_lock); + list_add_tail(&cache->active_bg_list, + &fs_info->zone_active_bgs); + spin_unlock(&fs_info->zone_active_bgs_lock); + } + } else { kfree(cache->physical_map); cache->physical_map = NULL; } -- cgit From 2dd7e7bc02829eded71be2342a93dc035f5223f9 Mon Sep 17 00:00:00 2001 From: Naohiro Aota Date: Fri, 9 Sep 2022 15:59:55 +0900 Subject: btrfs: zoned: wait for extent buffer IOs before finishing a zone Before sending REQ_OP_ZONE_FINISH to a zone, we need to ensure that ongoing IOs already finished. Or, we will see a "Zone Is Full" error for the IOs, as the ZONE_FINISH command makes the zone full. We ensure that with btrfs_wait_block_group_reservations() and btrfs_wait_ordered_roots() for a data block group. And, for a metadata block group, the comparison of alloc_offset vs meta_write_pointer mostly ensures IOs for the allocated region already sent. However, there still can be a little time frame where the IOs are sent but not yet completed. Introduce wait_eb_writebacks() to ensure such IOs are completed for a metadata block group. It walks the buffer_radix to find extent buffers in the block group and calls wait_on_extent_buffer_writeback() on them. Fixes: afba2bc036b0 ("btrfs: zoned: implement active zone tracking") CC: stable@vger.kernel.org # 5.19+ Signed-off-by: Naohiro Aota Signed-off-by: David Sterba --- fs/btrfs/zoned.c | 40 ++++++++++++++++++++++++++++++++++++++-- 1 file changed, 38 insertions(+), 2 deletions(-) (limited to 'fs/btrfs/zoned.c') diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c index 62e7007a7e46..73c6929f7be6 100644 --- a/fs/btrfs/zoned.c +++ b/fs/btrfs/zoned.c @@ -1918,10 +1918,44 @@ out_unlock: return ret; } +static void wait_eb_writebacks(struct btrfs_block_group *block_group) +{ + struct btrfs_fs_info *fs_info = block_group->fs_info; + const u64 end = block_group->start + block_group->length; + struct radix_tree_iter iter; + struct extent_buffer *eb; + void __rcu **slot; + + rcu_read_lock(); + radix_tree_for_each_slot(slot, &fs_info->buffer_radix, &iter, + block_group->start >> fs_info->sectorsize_bits) { + eb = radix_tree_deref_slot(slot); + if (!eb) + continue; + if (radix_tree_deref_retry(eb)) { + slot = radix_tree_iter_retry(&iter); + continue; + } + + if (eb->start < block_group->start) + continue; + if (eb->start >= end) + break; + + slot = radix_tree_iter_resume(slot, &iter); + rcu_read_unlock(); + wait_on_extent_buffer_writeback(eb); + rcu_read_lock(); + } + rcu_read_unlock(); +} + static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written) { struct btrfs_fs_info *fs_info = block_group->fs_info; struct map_lookup *map; + const bool is_metadata = (block_group->flags & + (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)); int ret = 0; int i; @@ -1932,8 +1966,7 @@ static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_writ } /* Check if we have unwritten allocated space */ - if ((block_group->flags & - (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)) && + if (is_metadata && block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) { spin_unlock(&block_group->lock); return -EAGAIN; @@ -1958,6 +1991,9 @@ static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_writ /* No need to wait for NOCOW writers. Zoned mode does not allow that */ btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start, block_group->length); + /* Wait for extent buffers to be written. */ + if (is_metadata) + wait_eb_writebacks(block_group); spin_lock(&block_group->lock); -- cgit