Subject: [to-be-updated] ocfs2-implement-delayed-dropping-of-last-dquot-reference.patch removed from -mm tree To: jack@xxxxxxx,jlbec@xxxxxxxxxxxx,mfasheh@xxxxxxx,rgoldwyn@xxxxxxxx,srinivas.eeda@xxxxxxxxxx,mm-commits@xxxxxxxxxxxxxxx From: akpm@xxxxxxxxxxxxxxxxxxxx Date: Wed, 26 Feb 2014 15:46:22 -0800 The patch titled Subject: ocfs2: implement delayed dropping of last dquot reference has been removed from the -mm tree. Its filename was ocfs2-implement-delayed-dropping-of-last-dquot-reference.patch This patch was dropped because an updated version will be merged ------------------------------------------------------ From: Jan Kara <jack@xxxxxxx> Subject: ocfs2: implement delayed dropping of last dquot reference We cannot drop last dquot reference from downconvert thread as that creates the following deadlock: NODE 1 NODE2 holds dentry lock for 'foo' holds inode lock for GLOBAL_BITMAP_SYSTEM_INODE dquot_initialize(bar) ocfs2_dquot_acquire() ocfs2_inode_lock(USER_QUOTA_SYSTEM_INODE) ... downconvert thread (triggered from another node or a different process from NODE2) ocfs2_dentry_post_unlock() ... iput(foo) ocfs2_evict_inode(foo) ocfs2_clear_inode(foo) dquot_drop(inode) ... ocfs2_dquot_release() ocfs2_inode_lock(USER_QUOTA_SYSTEM_INODE) - blocks finds we need more space in quota file ... ocfs2_extend_no_holes() ocfs2_inode_lock(GLOBAL_BITMAP_SYSTEM_INODE) - deadlocks waiting for downconvert thread We solve the problem by postponing dropping of the last dquot reference to a workqueue if it happens from the downconvert thread. Signed-off-by: Jan Kara <jack@xxxxxxx> Reviewed-by: Srinivas Eeda <srinivas.eeda@xxxxxxxxxx> Cc: Goldwyn Rodrigues <rgoldwyn@xxxxxxxx> Reviewed-by: Mark Fasheh <mfasheh@xxxxxxx> Cc: Joel Becker <jlbec@xxxxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- fs/ocfs2/ocfs2.h | 5 +++++ fs/ocfs2/quota.h | 2 ++ fs/ocfs2/quota_global.c | 35 +++++++++++++++++++++++++++++++++++ fs/ocfs2/super.c | 8 ++++++++ 4 files changed, 50 insertions(+) diff -puN fs/ocfs2/ocfs2.h~ocfs2-implement-delayed-dropping-of-last-dquot-reference fs/ocfs2/ocfs2.h --- a/fs/ocfs2/ocfs2.h~ocfs2-implement-delayed-dropping-of-last-dquot-reference +++ a/fs/ocfs2/ocfs2.h @@ -30,6 +30,7 @@ #include <linux/sched.h> #include <linux/wait.h> #include <linux/list.h> +#include <linux/llist.h> #include <linux/rbtree.h> #include <linux/workqueue.h> #include <linux/kref.h> @@ -419,6 +420,10 @@ struct ocfs2_super struct ocfs2_dentry_lock *dentry_lock_list; struct work_struct dentry_lock_work; + /* List of dquot structures to drop last reference to */ + struct llist_head dquot_drop_list; + struct work_struct dquot_drop_work; + wait_queue_head_t osb_mount_event; /* Truncate log info */ diff -puN fs/ocfs2/quota.h~ocfs2-implement-delayed-dropping-of-last-dquot-reference fs/ocfs2/quota.h --- a/fs/ocfs2/quota.h~ocfs2-implement-delayed-dropping-of-last-dquot-reference +++ a/fs/ocfs2/quota.h @@ -28,6 +28,7 @@ struct ocfs2_dquot { unsigned int dq_use_count; /* Number of nodes having reference to this entry in global quota file */ s64 dq_origspace; /* Last globally synced space usage */ s64 dq_originodes; /* Last globally synced inode usage */ + struct llist_node list; /* Member of list of dquots to drop */ }; /* Description of one chunk to recover in memory */ @@ -110,6 +111,7 @@ int ocfs2_read_quota_phys_block(struct i int ocfs2_create_local_dquot(struct dquot *dquot); int ocfs2_local_release_dquot(handle_t *handle, struct dquot *dquot); int ocfs2_local_write_dquot(struct dquot *dquot); +void ocfs2_drop_dquot_refs(struct work_struct *work); extern const struct dquot_operations ocfs2_quota_operations; extern struct quota_format_type ocfs2_quota_format; diff -puN fs/ocfs2/quota_global.c~ocfs2-implement-delayed-dropping-of-last-dquot-reference fs/ocfs2/quota_global.c --- a/fs/ocfs2/quota_global.c~ocfs2-implement-delayed-dropping-of-last-dquot-reference +++ a/fs/ocfs2/quota_global.c @@ -10,6 +10,7 @@ #include <linux/jiffies.h> #include <linux/writeback.h> #include <linux/workqueue.h> +#include <linux/llist.h> #include <cluster/masklog.h> @@ -679,6 +680,27 @@ static int ocfs2_calc_qdel_credits(struc OCFS2_INODE_UPDATE_CREDITS; } +void ocfs2_drop_dquot_refs(struct work_struct *work) +{ + struct ocfs2_super *osb = container_of(work, struct ocfs2_super, + dquot_drop_work); + struct llist_node *list; + struct ocfs2_dquot *odquot, *next_odquot; + + list = llist_del_all(&osb->dquot_drop_list); + llist_for_each_entry_safe(odquot, next_odquot, list, list) { + /* Drop the reference we acquired in ocfs2_dquot_release() */ + dqput(&odquot->dq_dquot); + } +} + +/* + * Called when the last reference to dquot is dropped. If we are called from + * downconvert thread, we cannot do all the handling here because grabbing + * quota lock could deadlock (the node holding the quota lock could need some + * other cluster lock to proceed but with blocked downconvert thread we cannot + * release any lock). + */ static int ocfs2_release_dquot(struct dquot *dquot) { handle_t *handle; @@ -694,6 +716,19 @@ static int ocfs2_release_dquot(struct dq /* Check whether we are not racing with some other dqget() */ if (atomic_read(&dquot->dq_count) > 1) goto out; + /* Running from downconvert thread? Postpone quota processing to wq */ + if (current == osb->dc_task) { + /* + * Grab our own reference to dquot and queue it for delayed + * dropping. Quota code rechecks after calling + * ->release_dquot() and won't free dquot structure. + */ + dqgrab(dquot); + /* First entry on list -> queue work */ + if (llist_add(&OCFS2_DQUOT(dquot)->list, &osb->dquot_drop_list)) + queue_work(ocfs2_wq, &osb->dquot_drop_work); + goto out; + } status = ocfs2_lock_global_qf(oinfo, 1); if (status < 0) goto out; diff -puN fs/ocfs2/super.c~ocfs2-implement-delayed-dropping-of-last-dquot-reference fs/ocfs2/super.c --- a/fs/ocfs2/super.c~ocfs2-implement-delayed-dropping-of-last-dquot-reference +++ a/fs/ocfs2/super.c @@ -1941,6 +1941,11 @@ static void ocfs2_dismount_volume(struct ocfs2_disable_quotas(osb); + /* All dquots should be freed by now */ + WARN_ON(!llist_empty(&osb->dquot_drop_list)); + /* Wait for worker to be done with the work structure in osb */ + cancel_work_sync(&osb->dquot_drop_work); + ocfs2_shutdown_local_alloc(osb); /* This will disable recovery and flush any recovery work. */ @@ -2276,6 +2281,9 @@ static int ocfs2_initialize_super(struct INIT_WORK(&osb->dentry_lock_work, ocfs2_drop_dl_inodes); osb->dentry_lock_list = NULL; + INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs); + init_llist_head(&osb->dquot_drop_list); + /* get some pseudo constants for clustersize bits */ osb->s_clustersize_bits = le32_to_cpu(di->id2.i_super.s_clustersize_bits); _ Patches currently in -mm which might be from jack@xxxxxxx are ocfs2-fix-quota-file-corruption.patch backing_dev-fix-hung-task-on-sync.patch kthread-ensure-locality-of-task_struct-allocations.patch ocfs2-avoid-blocking-in-ocfs2_mark_lockres_freeing-in-downconvert-thread.patch ocfs2-revert-iput-deferring-code-in-ocfs2_drop_dentry_lock.patch mm-vmstat-fix-up-zone-state-accounting.patch fs-cachefiles-use-add_to_page_cache_lru.patch lib-radix-tree-radix_tree_delete_item.patch mm-shmem-save-one-radix-tree-lookup-when-truncating-swapped-pages.patch mm-filemap-move-radix-tree-hole-searching-here.patch mm-fs-prepare-for-non-page-entries-in-page-cache-radix-trees.patch mm-fs-store-shadow-entries-in-page-cache.patch mm-thrash-detection-based-file-cache-sizing.patch lib-radix_tree-tree-node-interface.patch mm-keep-page-cache-radix-tree-nodes-in-check.patch mm-readaheadc-fix-readahead-failure-for-memoryless-numa-nodes-and-limit-readahead-pages.patch printk-remove-duplicated-check-for-log-level.patch printk-remove-obsolete-check-for-log-level-c.patch printk-add-comment-about-tricky-check-for-text-buffer-size.patch printk-use-also-the-last-bytes-in-the-ring-buffer.patch printk-do-not-compute-the-size-of-the-message-twice.patch linux-next.patch mm-add-strictlimit-knob-v2.patch -- To unsubscribe from this list: send the line "unsubscribe mm-commits" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html