Speed up WB_SYNC_NONE when a WB_SYNC_ALL occurs simultaneously

Page writebacks with WB_SYNC_NONE can take several seconds to complete 
since they wait for the transaction group to close before being 
committed. This is usually not a problem since the caller does not 
need to wait. However, if we're simultaneously doing a writeback 
with WB_SYNC_ALL (e.g via msync), the latter can block for several 
seconds (up to zfs_txg_timeout) due to the active WB_SYNC_NONE 
writeback since it needs to wait for the transaction to complete 
and the PG_writeback bit to be cleared.

This commit deals with 2 cases:

- No page writeback is active. A WB_SYNC_ALL page writeback starts 
  and even completes. But when it's about to check if the PG_writeback 
  bit has been cleared, another writeback with WB_SYNC_NONE starts. 
  The sync page writeback ends up waiting for the non-sync page 
  writeback to complete.

- A page writeback with WB_SYNC_NONE is already active when a 
  WB_SYNC_ALL writeback starts. The WB_SYNC_ALL writeback ends up 
  waiting for the WB_SYNC_NONE writeback.

The fix works by carefully keeping track of active sync/non-sync 
writebacks and committing when beneficial.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Shaan Nobee <sniper111@gmail.com>
Closes #12662
Closes #12790
This commit is contained in:
Shaan Nobee
2022-05-04 00:23:26 +04:00
committed by GitHub
parent a64d757aa4
commit 411f4a018d
17 changed files with 351 additions and 20 deletions
+10 -4
View File
@@ -58,6 +58,8 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
__field(uint64_t, z_size)
__field(uint64_t, z_pflags)
__field(uint32_t, z_sync_cnt)
__field(uint32_t, z_sync_writes_cnt)
__field(uint32_t, z_async_writes_cnt)
__field(mode_t, z_mode)
__field(boolean_t, z_is_sa)
__field(boolean_t, z_is_mapped)
@@ -91,6 +93,8 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
__entry->z_size = zn->z_size;
__entry->z_pflags = zn->z_pflags;
__entry->z_sync_cnt = zn->z_sync_cnt;
__entry->z_sync_writes_cnt = zn->z_sync_writes_cnt;
__entry->z_async_writes_cnt = zn->z_async_writes_cnt;
__entry->z_mode = zn->z_mode;
__entry->z_is_sa = zn->z_is_sa;
__entry->z_is_mapped = zn->z_is_mapped;
@@ -116,16 +120,18 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
TP_printk("zn { id %llu unlinked %u atime_dirty %u "
"zn_prefetch %u blksz %u seq %u "
"mapcnt %llu size %llu pflags %llu "
"sync_cnt %u mode 0x%x is_sa %d "
"is_mapped %d is_ctldir %d is_stale %d inode { "
"sync_cnt %u sync_writes_cnt %u async_writes_cnt %u "
"mode 0x%x is_sa %d is_mapped %d "
"is_ctldir %d is_stale %d inode { "
"uid %u gid %u ino %lu nlink %u size %lli "
"blkbits %u bytes %u mode 0x%x generation %x } } "
"ace { type %u flags %u access_mask %u } mask_matched %u",
__entry->z_id, __entry->z_unlinked, __entry->z_atime_dirty,
__entry->z_zn_prefetch, __entry->z_blksz,
__entry->z_seq, __entry->z_mapcnt, __entry->z_size,
__entry->z_pflags, __entry->z_sync_cnt, __entry->z_mode,
__entry->z_is_sa, __entry->z_is_mapped,
__entry->z_pflags, __entry->z_sync_cnt,
__entry->z_sync_writes_cnt, __entry->z_async_writes_cnt,
__entry->z_mode, __entry->z_is_sa, __entry->z_is_mapped,
__entry->z_is_ctldir, __entry->z_is_stale, __entry->i_uid,
__entry->i_gid, __entry->i_ino, __entry->i_nlink,
__entry->i_size, __entry->i_blkbits,
+1 -1
View File
@@ -70,7 +70,7 @@ extern int zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag,
extern int zfs_fid(struct inode *ip, fid_t *fidp);
extern int zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages);
extern int zfs_putpage(struct inode *ip, struct page *pp,
struct writeback_control *wbc);
struct writeback_control *wbc, boolean_t for_sync);
extern int zfs_dirty_inode(struct inode *ip, int flags);
extern int zfs_map(struct inode *ip, offset_t off, caddr_t *addrp,
size_t len, unsigned long vm_flags);