From 53b381b3abeb86f12787a6c40fee9b2f71edc23b Mon Sep 17 00:00:00 2001 From: David Woodhouse Date: Tue, 29 Jan 2013 18:40:14 -0500 Subject: Btrfs: RAID5 and RAID6 This builds on David Woodhouse's original Btrfs raid5/6 implementation. The code has changed quite a bit, blame Chris Mason for any bugs. Read/modify/write is done after the higher levels of the filesystem have prepared a given bio. This means the higher layers are not responsible for building full stripes, and they don't need to query for the topology of the extents that may get allocated during delayed allocation runs. It also means different files can easily share the same stripe. But, it does expose us to incorrect parity if we crash or lose power while doing a read/modify/write cycle. This will be addressed in a later commit. Scrub is unable to repair crc errors on raid5/6 chunks. Discard does not work on raid5/6 (yet) The stripe size is fixed at 64KiB per disk. This will be tunable in a later commit. Signed-off-by: Chris Mason --- fs/btrfs/extent-tree.c | 88 +++++++++++++++++++++++++++++++++----------------- 1 file changed, 59 insertions(+), 29 deletions(-) (limited to 'fs/btrfs/extent-tree.c') diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index d133edfcd449..3345f68fc64b 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -31,6 +31,7 @@ #include "print-tree.h" #include "transaction.h" #include "volumes.h" +#include "raid56.h" #include "locking.h" #include "free-space-cache.h" #include "math.h" @@ -1852,6 +1853,8 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, *actual_bytes = discarded_bytes; + if (ret == -EOPNOTSUPP) + ret = 0; return ret; } @@ -3276,6 +3279,7 @@ u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) u64 num_devices = root->fs_info->fs_devices->rw_devices + root->fs_info->fs_devices->missing_devices; u64 target; + u64 tmp; /* * see if restripe for this chunk_type is in progress, if so @@ -3292,30 +3296,32 @@ u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) } spin_unlock(&root->fs_info->balance_lock); + /* First, mask out the RAID levels which aren't possible */ if (num_devices == 1) - flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); + flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 | + BTRFS_BLOCK_GROUP_RAID5); + if (num_devices < 3) + flags &= ~BTRFS_BLOCK_GROUP_RAID6; if (num_devices < 4) flags &= ~BTRFS_BLOCK_GROUP_RAID10; - if ((flags & BTRFS_BLOCK_GROUP_DUP) && - (flags & (BTRFS_BLOCK_GROUP_RAID1 | - BTRFS_BLOCK_GROUP_RAID10))) { - flags &= ~BTRFS_BLOCK_GROUP_DUP; - } - - if ((flags & BTRFS_BLOCK_GROUP_RAID1) && - (flags & BTRFS_BLOCK_GROUP_RAID10)) { - flags &= ~BTRFS_BLOCK_GROUP_RAID1; - } + tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | + BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 | + BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10); + flags &= ~tmp; - if ((flags & BTRFS_BLOCK_GROUP_RAID0) && - ((flags & BTRFS_BLOCK_GROUP_RAID1) | - (flags & BTRFS_BLOCK_GROUP_RAID10) | - (flags & BTRFS_BLOCK_GROUP_DUP))) { - flags &= ~BTRFS_BLOCK_GROUP_RAID0; - } + if (tmp & BTRFS_BLOCK_GROUP_RAID6) + tmp = BTRFS_BLOCK_GROUP_RAID6; + else if (tmp & BTRFS_BLOCK_GROUP_RAID5) + tmp = BTRFS_BLOCK_GROUP_RAID5; + else if (tmp & BTRFS_BLOCK_GROUP_RAID10) + tmp = BTRFS_BLOCK_GROUP_RAID10; + else if (tmp & BTRFS_BLOCK_GROUP_RAID1) + tmp = BTRFS_BLOCK_GROUP_RAID1; + else if (tmp & BTRFS_BLOCK_GROUP_RAID0) + tmp = BTRFS_BLOCK_GROUP_RAID0; - return extended_to_chunk(flags); + return extended_to_chunk(flags | tmp); } static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) @@ -3333,6 +3339,7 @@ static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) { u64 flags; + u64 ret; if (data) flags = BTRFS_BLOCK_GROUP_DATA; @@ -3341,7 +3348,8 @@ u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) else flags = BTRFS_BLOCK_GROUP_METADATA; - return get_alloc_profile(root, flags); + ret = get_alloc_profile(root, flags); + return ret; } /* @@ -3516,8 +3524,10 @@ static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) { u64 num_dev; - if (type & BTRFS_BLOCK_GROUP_RAID10 || - type & BTRFS_BLOCK_GROUP_RAID0) + if (type & (BTRFS_BLOCK_GROUP_RAID10 | + BTRFS_BLOCK_GROUP_RAID0 | + BTRFS_BLOCK_GROUP_RAID5 | + BTRFS_BLOCK_GROUP_RAID6)) num_dev = root->fs_info->fs_devices->rw_devices; else if (type & BTRFS_BLOCK_GROUP_RAID1) num_dev = 2; @@ -3667,7 +3677,9 @@ static int can_overcommit(struct btrfs_root *root, /* * If we have dup, raid1 or raid10 then only half of the free - * space is actually useable. + * space is actually useable. For raid56, the space info used + * doesn't include the parity drive, so we don't have to + * change the math */ if (profile & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | @@ -5455,10 +5467,14 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, return ret; } -static u64 stripe_align(struct btrfs_root *root, u64 val) +static u64 stripe_align(struct btrfs_root *root, + struct btrfs_block_group_cache *cache, + u64 val, u64 num_bytes) { - u64 mask = ((u64)root->stripesize - 1); - u64 ret = (val + mask) & ~mask; + u64 mask; + u64 ret; + mask = ((u64)root->stripesize - 1); + ret = (val + mask) & ~mask; return ret; } @@ -5519,9 +5535,12 @@ int __get_raid_index(u64 flags) index = 2; else if (flags & BTRFS_BLOCK_GROUP_RAID0) index = 3; + else if (flags & BTRFS_BLOCK_GROUP_RAID5) + index = 5; + else if (flags & BTRFS_BLOCK_GROUP_RAID6) + index = 6; else - index = 4; - + index = 4; /* BTRFS_BLOCK_GROUP_SINGLE */ return index; } @@ -5665,6 +5684,8 @@ search: if (!block_group_bits(block_group, data)) { u64 extra = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | + BTRFS_BLOCK_GROUP_RAID5 | + BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10; /* @@ -5835,7 +5856,8 @@ unclustered_alloc: goto loop; } checks: - search_start = stripe_align(root, offset); + search_start = stripe_align(root, used_block_group, + offset, num_bytes); /* move on to the next group */ if (search_start + num_bytes > @@ -7203,6 +7225,7 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) root->fs_info->fs_devices->missing_devices; stripped = BTRFS_BLOCK_GROUP_RAID0 | + BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; if (num_devices == 1) { @@ -7754,7 +7777,9 @@ int btrfs_read_block_groups(struct btrfs_root *root) btrfs_release_path(path); cache->flags = btrfs_block_group_flags(&cache->item); cache->sectorsize = root->sectorsize; - + cache->full_stripe_len = btrfs_full_stripe_len(root, + &root->fs_info->mapping_tree, + found_key.objectid); btrfs_init_free_space_ctl(cache); /* @@ -7808,6 +7833,8 @@ int btrfs_read_block_groups(struct btrfs_root *root) if (!(get_alloc_profile(root, space_info->flags) & (BTRFS_BLOCK_GROUP_RAID10 | BTRFS_BLOCK_GROUP_RAID1 | + BTRFS_BLOCK_GROUP_RAID5 | + BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_DUP))) continue; /* @@ -7883,6 +7910,9 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; cache->sectorsize = root->sectorsize; cache->fs_info = root->fs_info; + cache->full_stripe_len = btrfs_full_stripe_len(root, + &root->fs_info->mapping_tree, + chunk_offset); atomic_set(&cache->count, 1); spin_lock_init(&cache->lock); -- cgit v1.2.3 From 8de972b4faa4be9b2a3c53103e18d86092a5da45 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 4 Jan 2013 15:39:43 -0500 Subject: Btrfs: fix cluster alignment for mount -o ssd With the new raid56 code, we want to make sure we're properly aligning our allocation clusters with -o ssd Signed-off-by: Chris Mason --- fs/btrfs/extent-tree.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'fs/btrfs/extent-tree.c') diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 3345f68fc64b..f13402104c96 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -5715,6 +5715,7 @@ have_block_group: * lets look there */ if (last_ptr) { + unsigned long aligned_cluster; /* * the refill lock keeps out other * people trying to start a new cluster @@ -5781,11 +5782,15 @@ refill_cluster: goto unclustered_alloc; } + aligned_cluster = max_t(unsigned long, + empty_cluster + empty_size, + block_group->full_stripe_len); + /* allocate a cluster in this block group */ ret = btrfs_find_space_cluster(trans, root, block_group, last_ptr, search_start, num_bytes, - empty_cluster + empty_size); + aligned_cluster); if (ret == 0) { /* * now pull our allocation out of this -- cgit v1.2.3 From bb721703aa551e98dc5c7fb259cf90343408baf2 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Tue, 29 Jan 2013 18:44:12 -0500 Subject: Btrfs: reduce CPU contention while waiting for delayed extent operations We batch up operations to the extent allocation tree, which allows us to deal with the recursive nature of using the extent allocation tree to allocate extents to the extent allocation tree. It also provides a mechanism to sort and collect extent operations, which makes it much more efficient to record extents that are close together. The delayed extent operations must all be finished before the running transaction commits, so we have code to make sure and run a few of the batched operations when closing our transaction handles. This creates a great deal of contention for the locks in the delayed extent operation tree, and also contention for the lock on the extent allocation tree itself. All the extra contention just slows down the operations and doesn't get things done any faster. This commit changes things to use a wait queue instead. As procs want to run the delayed operations, one of them races in and gets permission to hit the tree, and the others step back and wait for progress to be made. Signed-off-by: Chris Mason --- fs/btrfs/extent-tree.c | 60 ++++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 56 insertions(+), 4 deletions(-) (limited to 'fs/btrfs/extent-tree.c') diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index f13402104c96..87b0e856b6d0 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -2438,6 +2438,16 @@ int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, return ret; } +static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq, + int count) +{ + int val = atomic_read(&delayed_refs->ref_seq); + + if (val < seq || val >= seq + count) + return 1; + return 0; +} + /* * this starts processing the delayed reference count updates and * extent insertions we have queued up so far. count can be @@ -2472,6 +2482,44 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, delayed_refs = &trans->transaction->delayed_refs; INIT_LIST_HEAD(&cluster); + if (count == 0) { + count = delayed_refs->num_entries * 2; + run_most = 1; + } + + if (!run_all && !run_most) { + int old; + int seq = atomic_read(&delayed_refs->ref_seq); + +progress: + old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); + if (old) { + DEFINE_WAIT(__wait); + if (delayed_refs->num_entries < 16348) + return 0; + + prepare_to_wait(&delayed_refs->wait, &__wait, + TASK_UNINTERRUPTIBLE); + + old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); + if (old) { + schedule(); + finish_wait(&delayed_refs->wait, &__wait); + + if (!refs_newer(delayed_refs, seq, 256)) + goto progress; + else + return 0; + } else { + finish_wait(&delayed_refs->wait, &__wait); + goto again; + } + } + + } else { + atomic_inc(&delayed_refs->procs_running_refs); + } + again: loops = 0; spin_lock(&delayed_refs->lock); @@ -2480,10 +2528,6 @@ again: delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); #endif - if (count == 0) { - count = delayed_refs->num_entries * 2; - run_most = 1; - } while (1) { if (!(run_all || run_most) && delayed_refs->num_heads_ready < 64) @@ -2505,9 +2549,12 @@ again: if (ret < 0) { spin_unlock(&delayed_refs->lock); btrfs_abort_transaction(trans, root, ret); + atomic_dec(&delayed_refs->procs_running_refs); return ret; } + atomic_add(ret, &delayed_refs->ref_seq); + count -= min_t(unsigned long, ret, count); if (count == 0) @@ -2576,6 +2623,11 @@ again: goto again; } out: + atomic_dec(&delayed_refs->procs_running_refs); + smp_mb(); + if (waitqueue_active(&delayed_refs->wait)) + wake_up(&delayed_refs->wait); + spin_unlock(&delayed_refs->lock); assert_qgroups_uptodate(trans); return 0; -- cgit v1.2.3