- 06 Apr, 2023 31 commits
-
-
Ojaswin Mujoo authored
Currently, the kernel uses i_prealloc_list to hold all the inode preallocations. This is known to cause degradation in performance in workloads which perform large number of sparse writes on a single file. This is mainly because functions like ext4_mb_normalize_request() and ext4_mb_use_preallocated() iterate over this complete list, resulting in slowdowns when large number of PAs are present. Patch 27bc446e partially fixed this by enforcing a limit of 512 for the inode preallocation list and adding logic to continually trim the list if it grows above the threshold, however our testing revealed that a hardcoded value is not suitable for all kinds of workloads. To optimize this, add an rbtree to the inode and hold the inode preallocations in this rbtree. This will make iterating over inode PAs faster and scale much better than a linked list. Additionally, we also had to remove the LRU logic that was added during trimming of the list (in ext4_mb_release_context()) as it will add extra overhead in rbtree. The discards now happen in the lowest-logical-offset-first order. ** Locking notes ** With the introduction of rbtree to maintain inode PAs, we can't use RCU to walk the tree for searching since it can result in partial traversals which might miss some nodes(or entire subtrees) while discards happen in parallel (which happens under a lock). Hence this patch converts the ei->i_prealloc_lock spin_lock to rw_lock. Almost all the codepaths that read/modify the PA rbtrees are protected by the higher level inode->i_data_sem (except ext4_mb_discard_group_preallocations() and ext4_clear_inode()) IIUC, the only place we need lock protection is when one thread is reading "searching" the PA rbtree (earlier protected under rcu_read_lock()) and another is "deleting" the PAs in ext4_mb_discard_group_preallocations() function (which iterates all the PAs using the grp->bb_prealloc_list and deletes PAs from the tree without taking any inode lock (i_data_sem)). So, this patch converts all rcu_read_lock/unlock() paths for inode list PA to use read_lock() and all places where we were using ei->i_prealloc_lock spinlock will now be using write_lock(). Note that this makes the fast path (searching of the right PA e.g. ext4_mb_use_preallocated() or ext4_mb_normalize_request()), now use read_lock() instead of rcu_read_lock/unlock(). Ths also will now block due to slow discard path (ext4_mb_discard_group_preallocations()) which uses write_lock(). But this is not as bad as it looks. This is because - 1. The slow path only occurs when the normal allocation failed and we can say that we are low on disk space. One can argue this scenario won't be much frequent. 2. ext4_mb_discard_group_preallocations(), locks and unlocks the rwlock for deleting every individual PA. This gives enough opportunity for the fast path to acquire the read_lock for searching the PA inode list. Suggested-by:
Ritesh Harjani (IBM) <ritesh.list@gmail.com> Signed-off-by:
Ojaswin Mujoo <ojaswin@linux.ibm.com> Reviewed-by:
Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/4137bce8f6948fedd8bae134dabae24acfe699c6.1679731817.git.ojaswin@linux.ibm.comSigned-off-by:
Theodore Ts'o <tytso@mit.edu>
-
Ojaswin Mujoo authored
** Splitting pa->pa_inode_list ** Currently, we use the same pa->pa_inode_list to add a pa to either the inode preallocation list or the locality group preallocation list. For better clarity, split this list into a union of 2 list_heads and use either of the them based on the type of pa. ** Splitting pa->pa_obj_lock ** Currently, pa->pa_obj_lock is either assigned &ei->i_prealloc_lock for inode PAs or lg_prealloc_lock for lg PAs, and is then used to lock the lists containing these PAs. Make the distinction between the 2 PA types clear by changing this lock to a union of 2 locks. Explicitly use the pa_lock_node.inode_lock for inode PAs and pa_lock_node.lg_lock for lg PAs. This patch is required so that the locality group preallocation code remains the same as in upcoming patches we are going to make changes to inode preallocation code to move from list to rbtree based implementation. This patch also makes it easier to review the upcoming patches. There are no functional changes in this patch. Suggested-by:
-
Ojaswin Mujoo authored
When the length of best extent found is less than the length of goal extent we need to make sure that the best extent atleast covers the start of the original request. This is done by adjusting the ac_b_ex.fe_logical (logical start) of the extent. While doing so, the current logic sometimes results in the best extent's logical range overflowing the goal extent. Since this best extent is later added to the inode preallocation list, we have a possibility of introducing overlapping preallocations. This is discussed in detail here [1]. As per Jan's suggestion, to fix this, replace the existing logic with the below logic for adjusting best extent as it keeps fragmentation in check while ensuring logical range of best extent doesn't overflow out of goal extent: 1. Check if best extent can be kept at end of goal range and still cover original start. 2. Else, check if best extent can be kept at start of goal range and still cover original start. 3. Else, keep the best extent at start of original request. Also, add a few extra BUG_ONs that might help catch errors faster. [1] https://lore.kernel.org/r/Y+OGkVvzPN0RMv0O@li-bb2b2a4c-3307-11b2-a85c-8fa5c3a69313.ibm.comSuggested-by:
-
Ojaswin Mujoo authored
Abstract out the logic of fixing PA overlaps in ext4_mb_normalize_request to improve readability of code. This also makes it easier to make changes to the overlap logic in future. There are no functional changes in this patch Signed-off-by:
-
Ojaswin Mujoo authored
Abstract out the logic to double check for overlaps in normalize_pa to a separate function. Since there has been no reports in past where we have seen any overlaps which hits this bug_on(), in future we can consider calling this function under "#ifdef AGGRESSIVE_CHECK" only. There are no functional changes in this patch Signed-off-by:
-
Ojaswin Mujoo authored
Change some variable names to be more consistent and refactor some of the code to make it easier to read. There are no functional changes in this patch Signed-off-by:
-
Ojaswin Mujoo authored
This patch makes the following changes: * Rename ext4_mb_pa_free to ext4_mb_pa_put_free to better reflect its purpose * Add new ext4_mb_pa_free() which only handles freeing * Refactor ext4_mb_pa_callback() to use ext4_mb_pa_free() There are no functional changes in this patch Signed-off-by:
-
Ojaswin Mujoo authored
If we come across a PA that matches the logical offset but is unable to satisfy a non-extent file due to its physical start being higher than that supported by non extent files, then simply stop searching for another PA and break out of loop. This is because, since PAs don't overlap, we won't be able to find another inode PA which can satisfy the original request. Signed-off-by:
-
Theodore Ts'o authored
In cases where we have an obvious way of continuing, let's use WARN_RATELIMITED() instead of BUG_ON(). Signed-off-by: -
Kemeng Shi authored
Result of EXT4_SB(ac->ac_sb) is already stored in sbi at beginning of ext4_mb_normalize_request. Use sbi instead of EXT4_SB(ac->ac_sb) to remove unnecessary pointer dereference. Signed-off-by:
Kemeng Shi <shikemeng@huaweicloud.com> Reviewed-by:
-
Kemeng Shi authored
pa_plen -> pa_len pa_start -> pa_pstart Signed-off-by:
-
Kemeng Shi authored
We try to allocate a block from goal in ext4_mb_new_blocks_simple. We only need get blkoff in first group with goal and set blkoff to 0 for the rest groups. Signed-off-by:
-
Kemeng Shi authored
Just remove comment code in ext4_discard_preallocations. Signed-off-by:
-
Kemeng Shi authored
Call trace to assign ac_f_ex: ext4_mb_use_best_found ac->ac_f_ex = ac->ac_b_ex; ext4_mb_new_preallocation ext4_mb_new_group_pa ac->ac_f_ex = ac->ac_b_ex; ext4_mb_new_inode_pa ac->ac_f_ex = ac->ac_b_ex; Actually allocated blocks is already stored in ac_f_ex in ext4_mb_use_best_found, so there is no need to assign ac_f_ex in ext4_mb_new_group_pa and ext4_mb_new_inode_pa. Just remove repeat assignment to ac_f_ex in ext4_mb_new_group_pa and ext4_mb_new_inode_pa. Signed-off-by:
-
Kemeng Shi authored
When ext4_read_block_bitmap fails, we can return PTR_ERR(bitmap_bh) to remove unnecessary NULL check of bitmap_bh. Signed-off-by:
-
Kemeng Shi authored
count2 is always 1 in mb_debug, just remove unnecessary count2. Signed-off-by:
-
Kemeng Shi authored
We goto exit_meta_group_info only to return -ENOMEM. Return -ENOMEM directly instead of goto to remove this unnecessary tag. Signed-off-by:
-
Kemeng Shi authored
If any bex which meets bex->fe_len >= gex->fe_len is found, then it will always be used when complex scan of group that bex belongs to finishs. So there will not be any lock-unlock period. Signed-off-by:
-
Kemeng Shi authored
Only call trace of ext4_mb_check_limits is as following: ext4_mb_complex_scan_group ext4_mb_measure_extent ext4_mb_check_limits(ac, e4b, 0); ext4_mb_check_limits(ac, e4b, 1); If the first ac->ac_found > sbi->s_mb_max_to_scan check in ext4_mb_check_limits is met, we will set ac_status to AC_STATUS_BREAK and call ext4_mb_try_best_found to try to use ac->ac_b_ex. If ext4_mb_try_best_found successes, then block allocation finishs, the removed ac->ac_found > sbi->s_mb_min_to_scan check is not reachable. If ext4_mb_try_best_found fails, then we set EXT4_MB_HINT_FIRST and reset ac->ac_b_ex to retry block allocation. We will use any found free extent in ext4_mb_measure_extent before reach the removed ac->ac_found > sbi->s_mb_min_to_scan check. In summary, the removed ac->ac_found > sbi->s_mb_min_to_scan check is not reachable and we can remove that dead check. Signed-off-by:
-
Kemeng Shi authored
We always adjust first to even number and adjust last to odd number, so first == last will never happen. Remove this dead check. Signed-off-by:
-
Kemeng Shi authored
1. remove unnecessary ac check: We always go to out tag before ac is successfully allocated, then we can move out tag after free of ac and remove NULL check of ac. 2. remove unnecessary *errp check: We always go to errout tag if *errp is non-zero, then we can move errout tag into error handle if *errp is non-zero. Signed-off-by:
-
Kemeng Shi authored
e4b->bd_buddy_page is only set if we initialize ext4_buddy successfully. So e4b->bd_buddy_page is always NULL in error handle branch. Just remove the dead check. Signed-off-by:
-
Kemeng Shi authored
If we alloc array of buffer_head failed, there is no resource need to be freed and we can simpily return error. Signed-off-by:
-
Kemeng Shi authored
Return value static function ext4_mb_try_best_found and ext4_mb_free_metadata is not used. Just remove unused return value. Signed-off-by:
-
Kemeng Shi authored
Release bitmap buffer_head we got if error occurs. Besides, this patch remove unused assignment to err. Signed-off-by:
-
Kemeng Shi authored
If ext4_mb_mark_diskspace_used fails in ext4_mb_new_blocks, we may discard pa already in list. Protect pa with pa_lock to avoid race. Signed-off-by:
-
Kemeng Shi authored
As we don't correct pa_lstart here, so there is no need to subtract pa_lstart with consumed len. Signed-off-by:
-
Kemeng Shi authored
We will add pa_free to s_mb_preallocated when new ext4_prealloc_space is created. In ext4_mb_new_inode_pa, we will call ext4_mb_use_inode_pa before adding pa_free to s_mb_preallocated. However, ext4_mb_use_inode_pa will consume pa_free for block allocation which triggerred the creation of ext4_prealloc_space. Add pa_free to s_mb_preallocated before ext4_mb_use_inode_pa to correct calculation of s_mb_preallocated. There is no such problem in ext4_mb_new_group_pa as pa_free of group pa is consumed in ext4_mb_release_context instead of ext4_mb_use_group_pa. Signed-off-by:
-
Kemeng Shi authored
We always get ext4_group_desc with group + 1 and ext4_group_info with group to check if we need do initialize ext4_group_info for the group. Just get ext4_group_desc with group for ext4_group_info initialization check. Signed-off-by:
-
Kemeng Shi authored
If EXT4_MB_HINT_GOAL_ONLY is set, ext4_mb_regular_allocator will only allocate blocks from ext4_mb_find_by_goal. Allow to find by goal in ext4_mb_find_by_goal if EXT4_MB_HINT_GOAL_ONLY is set or allocation with EXT4_MB_HINT_GOAL_ONLY set will always fail. EXT4_MB_HINT_GOAL_ONLY is not used at all, so the problem is not found for now. Signed-off-by:
-
Kemeng Shi authored
We need to set ac_g_ex to notify the goal start used in ext4_mb_find_by_goal. Set ac_g_ex instead of ac_f_ex in ext4_mb_normalize_request. Besides we should assure goal start is in range [first_data_block, blocks_count) as ext4_mb_initialize_context does. [ Added a check to make sure size is less than ar->pright; otherwise we could end up passing an underflowed value of ar->pright - size to ext4_get_group_no_and_offset(), which will trigger a BUG_ON later on. - TYT ] Signed-off-by:
-
- 24 Mar, 2023 9 commits
-
-
Kemeng Shi authored
Remove unused group parameter in ext4_block_bitmap_csum_set. After this, group parameter in ext4_set_bitmap_checksums is also not used, just remove it too. Signed-off-by:
-
Kemeng Shi authored
Remove unused group parameter in ext4_block_bitmap_csum_verify. Signed-off-by:
-
Kemeng Shi authored
Remove unused group parameter in ext4_inode_bitmap_csum_set. Signed-off-by:
-
Kemeng Shi authored
Remove unused group parameter in ext4_inode_bitmap_csum_verify. Signed-off-by:
-
Kemeng Shi authored
As inode table blocks are contiguous, inode table blocks inside the block_group can be represented as range [itbl_cluster_start, itbl_cluster_last]. Then we can simply account inode table cluters and check cluster overlap with [itbl_cluster_start, itbl_cluster_last] instead of traverse each block of inode table. By the way, this patch fixes code style problem of comment for ext4_num_overhead_clusters. [ Merged fix-up patch which fixed potentially access to an uninitialzied stack variable. --TYT ] Signed-off-by:
kernel test robot <lkp@intel.com> Reported-by:
Dan Carpenter <error27@gmail.com> Signed-off-by:
-
Kemeng Shi authored
For case we initialize a bitmap bh, we will set bitmap bh verified. We can return immediately instead of goto verify to remove unnecessary work for trying to verify bitmap bh. Signed-off-by:
-
Kemeng Shi authored
Commit bdfb6ff4 ("ext4: mark group corrupt on group descriptor checksum") added flag to indicate corruption of group instead of marking all blocks used. Just remove stale comment. Signed-off-by:
-
Kemeng Shi authored
We only call ext4_bg_num_gdb_nometa if there is no meta_bg feature or group does not reach first_meta_bg, so group must not reside in meta group. Then group has a global gdt if superblock exists. Just remove confusing branch that meta_bg feature exists. Signed-off-by:
-
Kemeng Shi authored
ext4_num_base_meta_clusters is already aware of meta_bg feature and test if block_group is inside real meta block groups before calling ext4_bg_num_gdb. Then ext4_bg_num_gdb will check if block group is inside a real meta block groups again to decide either ext4_bg_num_gdb_meta or ext4_bg_num_gdb_nometa is needed. Call ext4_bg_num_gdb_meta or ext4_bg_num_gdb_nometa directly after we check if block_group is inside a meta block groups in ext4_num_base_meta_clusters to remove redundant check of meta block groups in ext4_bg_num_gdb. Signed-off-by:
-
