X-Git-Url: https://git.camperquake.de/gitweb.cgi?a=blobdiff_plain;f=module%2Fzfs%2Ftxg.c;h=5702a61f04fe8dbd65f3294a3899ba6d7e6372ac;hb=87d98efe9ea7dd1e464f38fc5e39f0d08191d3be;hp=f9f24dd0a9459a5f8de66f4375d526841d978546;hpb=090ff0929ec7a2e2e65efaaddb9981d15964f7d9;p=zfs.git

diff --git a/module/zfs/txg.c b/module/zfs/txg.c
index f9f24dd..5702a61 100644
--- a/module/zfs/txg.c
+++ b/module/zfs/txg.c
@@ -49,7 +49,7 @@ txg_init(dsl_pool_t *dp, uint64_t txg)
 	int c;
 	bzero(tx, sizeof (tx_state_t));
 
-	tx->tx_cpu = kmem_zalloc(max_ncpus * sizeof (tx_cpu_t), KM_SLEEP);
+	tx->tx_cpu = vmem_zalloc(max_ncpus * sizeof (tx_cpu_t), KM_SLEEP);
 
 	for (c = 0; c < max_ncpus; c++) {
 		int i;
@@ -107,7 +107,7 @@ txg_fini(dsl_pool_t *dp)
 	if (tx->tx_commit_cb_taskq != NULL)
 		taskq_destroy(tx->tx_commit_cb_taskq);
 
-	kmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t));
+	vmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t));
 
 	bzero(tx, sizeof (tx_state_t));
 }
@@ -166,10 +166,10 @@ txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, uint64_t time)
 	CALLB_CPR_SAFE_BEGIN(cpr);
 
 	if (time)
-		(void) cv_timedwait(cv, &tx->tx_sync_lock,
+		(void) cv_timedwait_interruptible(cv, &tx->tx_sync_lock,
 		    ddi_get_lbolt() + time);
 	else
-		cv_wait(cv, &tx->tx_sync_lock);
+		cv_wait_interruptible(cv, &tx->tx_sync_lock);
 
 	CALLB_CPR_SAFE_END(cpr, &tx->tx_sync_lock);
 }
@@ -218,9 +218,19 @@ uint64_t
 txg_hold_open(dsl_pool_t *dp, txg_handle_t *th)
 {
 	tx_state_t *tx = &dp->dp_tx;
-	tx_cpu_t *tc = &tx->tx_cpu[CPU_SEQID];
+	tx_cpu_t *tc;
 	uint64_t txg;
 
+	/*
+	 * It appears the processor id is simply used as a "random"
+	 * number to index into the array, and there isn't any other
+	 * significance to the chosen tx_cpu. Because.. Why not use
+	 * the current cpu to index into the array?
+	 */
+	kpreempt_disable();
+	tc = &tx->tx_cpu[CPU_SEQID];
+	kpreempt_enable();
+
 	mutex_enter(&tc->tc_lock);
 
 	txg = tx->tx_open_txg;
@@ -339,7 +349,7 @@ txg_dispatch_callbacks(dsl_pool_t *dp, uint64_t txg)
 			    TASKQ_THREADS_CPU_PCT | TASKQ_PREPOPULATE);
 		}
 
-		cb_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
+		cb_list = kmem_alloc(sizeof (list_t), KM_PUSHPAGE);
 		list_create(cb_list, sizeof (dmu_tx_callback_t),
 		    offsetof(dmu_tx_callback_t, dcb_node));
 
@@ -350,6 +360,20 @@ txg_dispatch_callbacks(dsl_pool_t *dp, uint64_t txg)
 	}
 }
 
+/*
+ * Wait for pending commit callbacks of already-synced transactions to finish
+ * processing.
+ * Calling this function from within a commit callback will deadlock.
+ */
+void
+txg_wait_callbacks(dsl_pool_t *dp)
+{
+	tx_state_t *tx = &dp->dp_tx;
+
+	if (tx->tx_commit_cb_taskq != NULL)
+		taskq_wait(tx->tx_commit_cb_taskq);
+}
+
 static void
 txg_sync_thread(dsl_pool_t *dp)
 {
@@ -358,13 +382,24 @@ txg_sync_thread(dsl_pool_t *dp)
 	callb_cpr_t cpr;
 	uint64_t start, delta;
 
+#ifdef _KERNEL
+	/*
+	 * Annotate this process with a flag that indicates that it is
+	 * unsafe to use KM_SLEEP during memory allocations due to the
+	 * potential for a deadlock.  KM_PUSHPAGE should be used instead.
+	 */
+	current->flags |= PF_NOFS;
+#endif /* _KERNEL */
+
 	txg_thread_enter(tx, &cpr);
 
 	start = delta = 0;
 	for (;;) {
-		uint64_t timer, timeout = zfs_txg_timeout * hz;
+		uint64_t timer, timeout;
 		uint64_t txg;
 
+		timeout = zfs_txg_timeout * hz;
+
 		/*
 		 * We sync when we're scanning, there's someone waiting
 		 * on us, or the quiesce thread has handed off a txg to
@@ -479,7 +514,7 @@ void
 txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks)
 {
 	tx_state_t *tx = &dp->dp_tx;
-	int timeout = ddi_get_lbolt() + ticks;
+	clock_t timeout = ddi_get_lbolt() + ticks;
 
 	/* don't delay if this txg could transition to quiesing immediately */
 	if (tx->tx_open_txg > txg ||
@@ -497,6 +532,8 @@ txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks)
 		(void) cv_timedwait(&tx->tx_quiesce_more_cv, &tx->tx_sync_lock,
 		    timeout);
 
+	DMU_TX_STAT_BUMP(dmu_tx_delay);
+
 	mutex_exit(&tx->tx_sync_lock);
 }
 
@@ -722,3 +759,23 @@ txg_list_next(txg_list_t *tl, void *p, uint64_t txg)
 
 	return (tn == NULL ? NULL : (char *)tn - tl->tl_offset);
 }
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+EXPORT_SYMBOL(txg_init);
+EXPORT_SYMBOL(txg_fini);
+EXPORT_SYMBOL(txg_sync_start);
+EXPORT_SYMBOL(txg_sync_stop);
+EXPORT_SYMBOL(txg_hold_open);
+EXPORT_SYMBOL(txg_rele_to_quiesce);
+EXPORT_SYMBOL(txg_rele_to_sync);
+EXPORT_SYMBOL(txg_register_callbacks);
+EXPORT_SYMBOL(txg_delay);
+EXPORT_SYMBOL(txg_wait_synced);
+EXPORT_SYMBOL(txg_wait_open);
+EXPORT_SYMBOL(txg_wait_callbacks);
+EXPORT_SYMBOL(txg_stalled);
+EXPORT_SYMBOL(txg_sync_waiting);
+
+module_param(zfs_txg_timeout, int, 0644);
+MODULE_PARM_DESC(zfs_txg_timeout, "Max seconds worth of delta per txg");
+#endif