svn commit: r348772 - head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs

Andriy Gapon avg at FreeBSD.org
Fri Jun 7 06:35:43 UTC 2019


Author: avg
Date: Fri Jun  7 06:35:42 2019
New Revision: 348772
URL: https://svnweb.freebsd.org/changeset/base/348772

Log:
  Restore ARC MFU/MRU pressure
  
  Before r305323 (MFV r302991: 6950 ARC should cache compressed data)
  arc_read() code did this for access to a ghost buffer:
   arc_adapt() (from arc_get_data_buf())
   arc_access(hdr, hash_lock)
  I.e., we first checked access to the MFU ghost/MRU ghost buffer and
  adapt MFU/MRU sizes (in arc_adapt()) and next move buffer from the ghost
  state to regular.
  
  After r305323 the sequence is different:
   arc_access(hdr, hash_lock);
   arc_hdr_alloc_pabd(hdr);
  I.e., we first move the buffer from the ghost state in arc_access() and
  then we check access to buffer in ghost state (in arc_hdr_alloc_pabd()
  -> arc_get_data_abd() -> arc_get_data_impl() -> arc_adapt()).  This is
  incorrect: arc_adapt() never see access to the ghost buffer because
  arc_access() already migrated the buffer from the ghost state to
  regular.
  
  So, the fix is to restore a call to arc_adapt() before arc_access() and
  to suppress the call to arc_adapt() after arc_access().
  
  Submitted by:	Slawa Olhovchenkov <slw at zxy.spb.ru>
  MFC after:	2 weeks
  Sponsored by:	Integros [integros.com]
  Differential Revision: https://reviews.freebsd.org/D19094

Modified:
  head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c

Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c
==============================================================================
--- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c	Fri Jun  7 05:03:03 2019	(r348771)
+++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c	Fri Jun  7 06:35:42 2019	(r348772)
@@ -1560,14 +1560,14 @@ static kmutex_t l2arc_feed_thr_lock;
 static kcondvar_t l2arc_feed_thr_cv;
 static uint8_t l2arc_thread_exit;
 
-static abd_t *arc_get_data_abd(arc_buf_hdr_t *, uint64_t, void *);
+static abd_t *arc_get_data_abd(arc_buf_hdr_t *, uint64_t, void *, boolean_t);
 static void *arc_get_data_buf(arc_buf_hdr_t *, uint64_t, void *);
-static void arc_get_data_impl(arc_buf_hdr_t *, uint64_t, void *);
+static void arc_get_data_impl(arc_buf_hdr_t *, uint64_t, void *, boolean_t);
 static void arc_free_data_abd(arc_buf_hdr_t *, abd_t *, uint64_t, void *);
 static void arc_free_data_buf(arc_buf_hdr_t *, void *, uint64_t, void *);
 static void arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag);
 static void arc_hdr_free_pabd(arc_buf_hdr_t *);
-static void arc_hdr_alloc_pabd(arc_buf_hdr_t *);
+static void arc_hdr_alloc_pabd(arc_buf_hdr_t *, boolean_t);
 static void arc_access(arc_buf_hdr_t *, kmutex_t *);
 static boolean_t arc_is_overflowing();
 static void arc_buf_watch(arc_buf_t *);
@@ -3353,14 +3353,14 @@ arc_buf_destroy_impl(arc_buf_t *buf)
 }
 
 static void
-arc_hdr_alloc_pabd(arc_buf_hdr_t *hdr)
+arc_hdr_alloc_pabd(arc_buf_hdr_t *hdr, boolean_t do_adapt)
 {
 	ASSERT3U(HDR_GET_LSIZE(hdr), >, 0);
 	ASSERT(HDR_HAS_L1HDR(hdr));
 	ASSERT(!HDR_SHARED_DATA(hdr));
 
 	ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
-	hdr->b_l1hdr.b_pabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr);
+	hdr->b_l1hdr.b_pabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr, do_adapt);
 	hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
 	ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
 
@@ -3424,7 +3424,7 @@ arc_hdr_alloc(uint64_t spa, int32_t psize, int32_t lsi
 	 * the compressed or uncompressed data depending on the block
 	 * it references and compressed arc enablement.
 	 */
-	arc_hdr_alloc_pabd(hdr);
+	arc_hdr_alloc_pabd(hdr, B_TRUE);
 	ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
 
 	return (hdr);
@@ -5145,11 +5145,11 @@ arc_is_overflowing(void)
 }
 
 static abd_t *
-arc_get_data_abd(arc_buf_hdr_t *hdr, uint64_t size, void *tag)
+arc_get_data_abd(arc_buf_hdr_t *hdr, uint64_t size, void *tag, boolean_t do_adapt)
 {
 	arc_buf_contents_t type = arc_buf_type(hdr);
 
-	arc_get_data_impl(hdr, size, tag);
+	arc_get_data_impl(hdr, size, tag, do_adapt);
 	if (type == ARC_BUFC_METADATA) {
 		return (abd_alloc(size, B_TRUE));
 	} else {
@@ -5163,7 +5163,7 @@ arc_get_data_buf(arc_buf_hdr_t *hdr, uint64_t size, vo
 {
 	arc_buf_contents_t type = arc_buf_type(hdr);
 
-	arc_get_data_impl(hdr, size, tag);
+	arc_get_data_impl(hdr, size, tag, B_TRUE);
 	if (type == ARC_BUFC_METADATA) {
 		return (zio_buf_alloc(size));
 	} else {
@@ -5179,12 +5179,13 @@ arc_get_data_buf(arc_buf_hdr_t *hdr, uint64_t size, vo
  * limit, we'll only signal the reclaim thread and continue on.
  */
 static void
-arc_get_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag)
+arc_get_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag, boolean_t do_adapt)
 {
 	arc_state_t *state = hdr->b_l1hdr.b_state;
 	arc_buf_contents_t type = arc_buf_type(hdr);
 
-	arc_adapt(size, state);
+	if (do_adapt)
+		arc_adapt(size, state);
 
 	/*
 	 * If arc_size is currently overflowing, and has grown past our
@@ -5935,8 +5936,9 @@ top:
 			 * do this after we've called arc_access() to
 			 * avoid hitting an assert in remove_reference().
 			 */
+			arc_adapt(arc_hdr_size(hdr), hdr->b_l1hdr.b_state);
 			arc_access(hdr, hash_lock);
-			arc_hdr_alloc_pabd(hdr);
+			arc_hdr_alloc_pabd(hdr, B_FALSE);
 		}
 		ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
 		size = arc_hdr_size(hdr);
@@ -6350,7 +6352,7 @@ arc_release(arc_buf_t *buf, void *tag)
 			if (arc_can_share(hdr, lastbuf)) {
 				arc_share_buf(hdr, lastbuf);
 			} else {
-				arc_hdr_alloc_pabd(hdr);
+				arc_hdr_alloc_pabd(hdr, B_TRUE);
 				abd_copy_from_buf(hdr->b_l1hdr.b_pabd,
 				    buf->b_data, psize);
 			}
@@ -6513,7 +6515,7 @@ arc_write_ready(zio_t *zio)
 	 * the data into it; otherwise, we share the data directly if we can.
 	 */
 	if (zfs_abd_scatter_enabled || !arc_can_share(hdr, buf)) {
-		arc_hdr_alloc_pabd(hdr);
+		arc_hdr_alloc_pabd(hdr, B_TRUE);
 
 		/*
 		 * Ideally, we would always copy the io_abd into b_pabd, but the


More information about the svn-src-all mailing list