SSV-27159: Fix ZIL zil_commit()/zil_suspend() race causing BSOD

include/os/windows/spl/sys/types.h

@@ -27,6 +27,10 @@
 #ifndef _SPL_TYPES_H
 #define	_SPL_TYPES_H
 
+#if !defined(zfs_fallthrough)
+#define	zfs_fallthrough		((void)0)
+#endif
+
 // use ntintsafe.h ?
 typedef enum { B_FALSE = 0, B_TRUE = 1 }	boolean_t;
 typedef short				pri_t;

include/sys/zil_impl.h

@@ -44,7 +44,7 @@ extern "C" {
  * must be held.
  *
  * After the lwb is "opened", it can transition into the "issued" state
- * via zil_lwb_write_issue(). Again, the zilog's "zl_issuer_lock" must
+ * via zil_lwb_write_close(). Again, the zilog's "zl_issuer_lock" must
  * be held when making this transition.
  *
  * After the lwb's write zio completes, it transitions into the "write
@@ -93,20 +93,23 @@ typedef struct lwb {
 	blkptr_t	lwb_blk;	/* on disk address of this log blk */
 	boolean_t	lwb_fastwrite;	/* is blk marked for fastwrite? */
 	boolean_t	lwb_slog;	/* lwb_blk is on SLOG device */
+	boolean_t	lwb_indirect;	/* do not postpone zil_lwb_commit() */
 	int		lwb_nused;	/* # used bytes in buffer */
+	int		lwb_nfilled;	/* # filled bytes in buffer */
 	int		lwb_sz;		/* size of block and buffer */
 	lwb_state_t	lwb_state;	/* the state of this lwb */
 	char		*lwb_buf;	/* log write buffer */
 	zio_t		*lwb_write_zio;	/* zio for the lwb buffer */
 	zio_t		*lwb_root_zio;	/* root zio for lwb write and flushes */
-	dmu_tx_t	*lwb_tx;	/* tx for log block allocation */
+	hrtime_t	lwb_issued_timestamp; /* when was the lwb issued? */
+	uint64_t	lwb_issued_txg;	/* the txg when the write is issued */
 	uint64_t	lwb_max_txg;	/* highest txg in this lwb */
 	list_node_t	lwb_node;	/* zilog->zl_lwb_list linkage */
+	list_node_t	lwb_issue_node;	/* linkage of lwbs ready for issue */
 	list_t		lwb_itxs;	/* list of itx's */
 	list_t		lwb_waiters;	/* list of zil_commit_waiter's */
 	avl_tree_t	lwb_vdev_tree;	/* vdevs to flush after lwb write */
 	kmutex_t	lwb_vdev_lock;	/* protects lwb_vdev_tree */
-	hrtime_t	lwb_issued_timestamp; /* when was the lwb issued? */
 } lwb_t;
 
 /*
@@ -183,7 +186,6 @@ struct zilog {
 	uint64_t	zl_destroy_txg;	/* txg of last zil_destroy() */
 	uint64_t	zl_replayed_seq[TXG_SIZE]; /* last replayed rec seq */
 	uint64_t	zl_replaying_seq; /* current replay seq number */
-	krwlock_t	zl_suspend_lock;	/* protects suspend count */
 	uint32_t	zl_suspend;	/* log suspend count */
 	kcondvar_t	zl_cv_suspend;	/* log suspend completion */
 	uint8_t		zl_suspending;	/* log is currently suspending */
@@ -210,6 +212,12 @@ struct zilog {
 	uint_t		zl_prev_rotor;	/* rotor for zl_prev[] */
 	txg_node_t	zl_dirty_link;	/* protected by dp_dirty_zilogs list */
 	uint64_t	zl_dirty_max_txg; /* highest txg used to dirty zilog */
+
+	kmutex_t	zl_lwb_io_lock; /* protect following members */
+	uint64_t	zl_lwb_inflight[TXG_SIZE]; /* io issued, but not done */
+	kcondvar_t	zl_lwb_io_cv;	/* signal when the flush is done */
+	uint64_t	zl_lwb_max_issued_txg; /* max txg when lwb io issued */
+
 	/*
 	 * Max block size for this ZIL.  Note that this can not be changed
 	 * while the ZIL is in use because consumers (ZPL/zvol) need to take

lib/libspl/include/sys/feature_tests.h

@@ -33,4 +33,14 @@
 #define	__NORETURN	__attribute__((__noreturn__))
 #endif
 
+#if !defined(zfs_fallthrough)
+#if defined(_MSC_VER)
+#define	zfs_fallthrough		((void)0)
+#elif defined(HAVE_IMPLICIT_FALLTHROUGH)
+#define	zfs_fallthrough		__attribute__((__fallthrough__))
+#else
+#define	zfs_fallthrough		((void)0)
+#endif
+#endif
+
 #endif

man/man4/zfs.4

@@ -2058,6 +2058,13 @@ On very fragmented pools, lowering this
 .Pq typically to Sy 36kB
 can improve performance.
 .
+.It Sy zil_min_commit_timeout Ns = Ns Sy 5000 Pq u64
+This sets the minimum delay in nanoseconds ZIL care to delay block commit,
+waiting for more records.
+If ZIL writes are too fast, kernel may not be able sleep for so short interval,
+increasing log latency above allowed by
+.Sy zfs_commit_timeout_pct .
+.
 .It Sy zil_nocacheflush Ns = Ns Sy 0 Ns | Ns 1 Pq int
 Disable the cache flush commands that are normally sent to disk by
 the ZIL after an LWB write has completed.