ZIL: Second attempt to reduce scope of zl_issuer_lock.

The previous patch #14841 appeared to have significant flaw, causing
deadlocks if zl_get_data callback got blocked waiting for TXG sync.  I
already handled some of such cases in the original patch, but issue
 #14982 shown cases that were impossible to solve in that design.

This patch fixes the problem by postponing log blocks allocation till
the very end, just before the zios issue, leaving nothing blocking after
that point to cause deadlocks.  Before that point though any sleeps are
now allowed, not causing sync thread blockage.  This require slightly
more complicated lwb state machine to allocate blocks and issue zios
in proper order.  But with removal of special early issue workarounds
the new code is much cleaner now, and should even be more efficient.

Since this patch uses null zios between write, I've found that null
zios do not wait for logical children ready status in zio_ready(),
that makes parent write to proceed prematurely, producing incorrect
log blocks.  Added ZIO_CHILD_LOGICAL_BIT to zio_wait_for_children()
fixes it.

Signed-off-by:	Alexander Motin <[email protected]>
Sponsored by:	iXsystems, Inc.

cmd/ztest.c

@@ -2412,7 +2412,6 @@ ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
 	int error;
 
 	ASSERT3P(lwb, !=, NULL);
-	ASSERT3P(zio, !=, NULL);
 	ASSERT3U(size, !=, 0);
 
 	ztest_object_lock(zd, object, RL_READER);
@@ -2446,6 +2445,7 @@ ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
 		    DMU_READ_NO_PREFETCH);
 		ASSERT0(error);
 	} else {
+		ASSERT3P(zio, !=, NULL);
 		size = doi.doi_data_block_size;
 		if (ISP2(size)) {
 			offset = P2ALIGN(offset, size);

include/sys/zil_impl.h

@@ -38,14 +38,22 @@ extern "C" {
 /*
  * Possible states for a given lwb structure.
  *
- * An lwb will start out in the "closed" state, and then transition to
- * the "opened" state via a call to zil_lwb_write_open(). When
- * transitioning from "closed" to "opened" the zilog's "zl_issuer_lock"
- * must be held.
+ * An lwb will start out in the "new" state, and transition to the "opened"
+ * state via a call to zil_lwb_write_open() on first itx assignment.  When
+ * transitioning from "new" to "opened" the zilog's "zl_issuer_lock" must be
+ * held.
  *
- * After the lwb is "opened", it can transition into the "issued" state
- * via zil_lwb_write_close(). Again, the zilog's "zl_issuer_lock" must
- * be held when making this transition.
+ * After the lwb is "opened", it can be assigned number of itxs and transition
+ * into the "closed" state via zil_lwb_write_close() when full or on timeout.
+ * When transitioning from "opened" to "closed" the zilog's "zl_issuer_lock"
+ * must be held.  New lwb allocation also takes "zl_lock" to protect the list.
+ *
+ * After the lwb is "closed", it can transition into the "ready" state via
+ * zil_lwb_write_issue().  "zl_lock" must be held when making this transition.
+ * Since it is done by the same thread, "zl_issuer_lock" is not needed.
+ *
+ * When lwb in "ready" state receives its block pointer, it can transition to
+ * "issued". "zl_lock" must be held when making this transition.
  *
  * After the lwb's write zio completes, it transitions into the "write
  * done" state via zil_lwb_write_done(); and then into the "flush done"
@@ -62,17 +70,20 @@ extern "C" {
  *
  * Additionally, correctness when reading an lwb's state is often
  * achieved by exploiting the fact that these state transitions occur in
- * this specific order; i.e. "closed" to "opened" to "issued" to "done".
+ * this specific order; i.e. "new" to "opened" to "closed" to "ready" to
+ * "issued" to "write_done" and finally "flush_done".
  *
- * Thus, if an lwb is in the "closed" or "opened" state, holding the
+ * Thus, if an lwb is in the "new" or "opened" state, holding the
  * "zl_issuer_lock" will prevent a concurrent thread from transitioning
- * that lwb to the "issued" state. Likewise, if an lwb is already in the
- * "issued" state, holding the "zl_lock" will prevent a concurrent
- * thread from transitioning that lwb to the "write done" state.
+ * that lwb to the "closed" state. Likewise, if an lwb is already in the
+ * "ready" state, holding the "zl_lock" will prevent a concurrent thread
+ * from transitioning that lwb to the "issued" state.
  */
 typedef enum {
-    LWB_STATE_CLOSED,
+    LWB_STATE_NEW,
     LWB_STATE_OPENED,
+    LWB_STATE_CLOSED,
+    LWB_STATE_READY,
     LWB_STATE_ISSUED,
     LWB_STATE_WRITE_DONE,
     LWB_STATE_FLUSH_DONE,
@@ -91,17 +102,21 @@ typedef enum {
 typedef struct lwb {
 	zilog_t		*lwb_zilog;	/* back pointer to log struct */
 	blkptr_t	lwb_blk;	/* on disk address of this log blk */
+	boolean_t	lwb_slim;	/* log block has slim format */
 	boolean_t	lwb_slog;	/* lwb_blk is on SLOG device */
-	boolean_t	lwb_indirect;	/* do not postpone zil_lwb_commit() */
+	int		lwb_error;	/* log block allocation error */
+	int		lwb_nmax;	/* max bytes in the buffer */
 	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_child_zio;	/* parent zio for children */
 	zio_t		*lwb_write_zio;	/* zio for the lwb buffer */
 	zio_t		*lwb_root_zio;	/* root zio for lwb write and flushes */
 	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_alloc_txg;	/* the txg when lwb_blk is allocated */
 	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 */

module/zfs/zfs_vnops.c

@@ -839,7 +839,6 @@ zfs_get_data(void *arg, uint64_t gen, lr_write_t *lr, char *buf,
 	uint64_t zp_gen;
 
 	ASSERT3P(lwb, !=, NULL);
-	ASSERT3P(zio, !=, NULL);
 	ASSERT3U(size, !=, 0);
 
 	/*
@@ -889,6 +888,7 @@ zfs_get_data(void *arg, uint64_t gen, lr_write_t *lr, char *buf,
 		}
 		ASSERT(error == 0 || error == ENOENT);
 	} else { /* indirect write */
+		ASSERT3P(zio, !=, NULL);
 		/*
 		 * Have to lock the whole block to ensure when it's
 		 * written out and its checksum is being calculated