3306 zdb should be able to issue reads in parallel

3321 'zpool reopen' command should be documented in the man page and help message Reviewed by: Adam Leventhal <[email protected]> Reviewed by: Matt Ahrens <[email protected]> Reviewed by: Christopher Siden <[email protected]> Approved by: Garrett D'Amore <[email protected]>
illumos · Nov 2, 2012 · 31d7e8f · 31d7e8f
1 parent 5c68564
commit 31d7e8f
Show file tree

Hide file tree

Showing 10 changed files with 287 additions and 66 deletions.
diff --git a/usr/src/cmd/zdb/zdb.c b/usr/src/cmd/zdb/zdb.c
@@ -86,6 +86,7 @@ extern void dump_intent_log(zilog_t *);
 uint64_t *zopt_object = NULL;
 int zopt_objects = 0;
 libzfs_handle_t *g_zfs;
+uint64_t max_inflight = 200;
 
 /*
  * These libumem hooks provide a reasonable set of defaults for the allocator's
@@ -108,13 +109,14 @@ usage(void)
 {
 	(void) fprintf(stderr,
 	    "Usage: %s [-CumdibcsDvhLXFPA] [-t txg] [-e [-p path...]] "
-	    "poolname [object...]\n"
-	    "       %s [-divPA] [-e -p path...] dataset [object...]\n"
-	    "       %s -m [-LXFPA] [-t txg] [-e [-p path...]] "
+	    "[-U config] [-M inflight I/Os] poolname [object...]\n"
+	    "       %s [-divPA] [-e -p path...] [-U config] dataset "
+	    "[object...]\n"
+	    "       %s -m [-LXFPA] [-t txg] [-e [-p path...]] [-U config] "
 	    "poolname [vdev [metaslab...]]\n"
 	    "       %s -R [-A] [-e [-p path...]] poolname "
 	    "vdev:offset:size[:flags]\n"
-	    "       %s -S [-PA] [-e [-p path...]] poolname\n"
+	    "       %s -S [-PA] [-e [-p path...]] [-U config] poolname\n"
 	    "       %s -l [-uA] device\n"
 	    "       %s -C [-A] [-U config]\n\n",
 	    cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname);
@@ -161,6 +163,8 @@ usage(void)
 	(void) fprintf(stderr, "        -P print numbers in parseable form\n");
 	(void) fprintf(stderr, "        -t <txg> -- highest txg to use when "
 	    "searching for uberblocks\n");
+	(void) fprintf(stderr, "        -M <number of inflight I/Os> -- "
+	    "specify the maximum number of checksumming I/Os [default is 200]");
 	(void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
 	    "to make only that option verbose\n");
 	(void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
@@ -2028,6 +2032,45 @@ zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
 	    bp, NULL, NULL, ZIO_FLAG_CANFAIL)), ==, 0);
 }
 
+static void
+zdb_blkptr_done(zio_t *zio)
+{
+	spa_t *spa = zio->io_spa;
+	blkptr_t *bp = zio->io_bp;
+	int ioerr = zio->io_error;
+	zdb_cb_t *zcb = zio->io_private;
+	zbookmark_t *zb = &zio->io_bookmark;
+
+	zio_data_buf_free(zio->io_data, zio->io_size);
+
+	mutex_enter(&spa->spa_scrub_lock);
+	spa->spa_scrub_inflight--;
+	cv_broadcast(&spa->spa_scrub_io_cv);
+
+	if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
+		char blkbuf[BP_SPRINTF_LEN];
+
+		zcb->zcb_haderrors = 1;
+		zcb->zcb_errors[ioerr]++;
+
+		if (dump_opt['b'] >= 2)
+			sprintf_blkptr(blkbuf, bp);
+		else
+			blkbuf[0] = '\0';
+
+		(void) printf("zdb_blkptr_cb: "
+		    "Got error %d reading "
+		    "<%llu, %llu, %lld, %llx> %s -- skipping\n",
+		    ioerr,
+		    (u_longlong_t)zb->zb_objset,
+		    (u_longlong_t)zb->zb_object,
+		    (u_longlong_t)zb->zb_level,
+		    (u_longlong_t)zb->zb_blkid,
+		    blkbuf);
+	}
+	mutex_exit(&spa->spa_scrub_lock);
+}
+
 /* ARGSUSED */
 static int
 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
@@ -2049,39 +2092,23 @@ zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
 	is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
 
 	if (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata)) {
-		int ioerr;
 		size_t size = BP_GET_PSIZE(bp);
-		void *data = malloc(size);
+		void *data = zio_data_buf_alloc(size);
 		int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
 
 		/* If it's an intent log block, failure is expected. */
 		if (zb->zb_level == ZB_ZIL_LEVEL)
 			flags |= ZIO_FLAG_SPECULATIVE;
 
-		ioerr = zio_wait(zio_read(NULL, spa, bp, data, size,
-		    NULL, NULL, ZIO_PRIORITY_ASYNC_READ, flags, zb));
-
-		free(data);
+		mutex_enter(&spa->spa_scrub_lock);
+		while (spa->spa_scrub_inflight > max_inflight)
+			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
+		spa->spa_scrub_inflight++;
+		mutex_exit(&spa->spa_scrub_lock);
 
-		if (ioerr && !(flags & ZIO_FLAG_SPECULATIVE)) {
-			zcb->zcb_haderrors = 1;
-			zcb->zcb_errors[ioerr]++;
+		zio_nowait(zio_read(NULL, spa, bp, data, size,
+		    zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
 
-			if (dump_opt['b'] >= 2)
-				sprintf_blkptr(blkbuf, bp);
-			else
-				blkbuf[0] = '\0';
-
-			(void) printf("zdb_blkptr_cb: "
-			    "Got error %d reading "
-			    "<%llu, %llu, %lld, %llx> %s -- skipping\n",
-			    ioerr,
-			    (u_longlong_t)zb->zb_objset,
-			    (u_longlong_t)zb->zb_object,
-			    (u_longlong_t)zb->zb_level,
-			    (u_longlong_t)zb->zb_blkid,
-			    blkbuf);
-		}
 	}
 
 	zcb->zcb_readfails = 0;
@@ -2283,6 +2310,18 @@ dump_block_stats(spa_t *spa)
 
 	zcb.zcb_haderrors |= traverse_pool(spa, 0, flags, zdb_blkptr_cb, &zcb);
 
+	/*
+	 * If we've traversed the data blocks then we need to wait for those
+	 * I/Os to complete. We leverage "The Godfather" zio to wait on
+	 * all async I/Os to complete.
+	 */
+	if (dump_opt['c']) {
+		(void) zio_wait(spa->spa_async_zio_root);
+		spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
+		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
+		    ZIO_FLAG_GODFATHER);
+	}
+
 	if (zcb.zcb_haderrors) {
 		(void) printf("\nError counts:\n\n");
 		(void) printf("\t%5s  %s\n", "errno", "count");
@@ -3040,7 +3079,7 @@ main(int argc, char **argv)
 
 	dprintf_setup(&argc, argv);
 
-	while ((c = getopt(argc, argv, "bcdhilmsuCDRSAFLXevp:t:U:P")) != -1) {
+	while ((c = getopt(argc, argv, "bcdhilmM:suCDRSAFLXevp:t:U:P")) != -1) {
 		switch (c) {
 		case 'b':
 		case 'c':
@@ -3069,6 +3108,15 @@ main(int argc, char **argv)
 		case 'v':
 			verbose++;
 			break;
+		case 'M':
+			max_inflight = strtoull(optarg, NULL, 0);
+			if (max_inflight == 0) {
+				(void) fprintf(stderr, "maximum number "
+				    "of inflight I/Os must be greater "
+				    "than 0\n");
+				usage();
+			}
+			break;
 		case 'p':
 			if (searchdirs == NULL) {
 				searchdirs = umem_alloc(sizeof (char *),

diff --git a/usr/src/cmd/zpool/zpool_main.c b/usr/src/cmd/zpool/zpool_main.c
@@ -238,7 +238,7 @@ get_usage(zpool_help_t idx) {
 	case HELP_REMOVE:
 		return (gettext("\tremove <pool> <device> ...\n"));
 	case HELP_REOPEN:
-		return (""); /* Undocumented command */
+		return (gettext("\treopen <pool>\n"));
 	case HELP_SCRUB:
 		return (gettext("\tscrub [-s] <pool> ...\n"));
 	case HELP_STATUS:
@@ -3550,22 +3550,37 @@ zpool_do_reguid(int argc, char **argv)
  * zpool reopen <pool>
  *
  * Reopen the pool so that the kernel can update the sizes of all vdevs.
- *
- * NOTE: This command is currently undocumented.  If the command is ever
- * exposed then the appropriate usage() messages will need to be made.
  */
 int
 zpool_do_reopen(int argc, char **argv)
 {
+	int c;
 	int ret = 0;
 	zpool_handle_t *zhp;
 	char *pool;
 
+	/* check options */
+	while ((c = getopt(argc, argv, "")) != -1) {
+		switch (c) {
+		case '?':
+			(void) fprintf(stderr, gettext("invalid option '%c'\n"),
+			    optopt);
+			usage(B_FALSE);
+		}
+	}
+
 	argc--;
 	argv++;
 
-	if (argc != 1)
-		return (2);
+	if (argc < 1) {
+		(void) fprintf(stderr, gettext("missing pool name\n"));
+		usage(B_FALSE);
+	}
+
+	if (argc > 1) {
+		(void) fprintf(stderr, gettext("too many arguments\n"));
+		usage(B_FALSE);
+	}
 
 	pool = argv[0];
 	if ((zhp = zpool_open_canfail(g_zfs, pool)) == NULL)

diff --git a/usr/src/lib/libzpool/common/kernel.c b/usr/src/lib/libzpool/common/kernel.c
@@ -1012,3 +1012,48 @@ zfs_onexit_cb_data(minor_t minor, uint64_t action_handle, void **data)
 {
 	return (0);
 }
+
+void
+bioinit(buf_t *bp)
+{
+	bzero(bp, sizeof (buf_t));
+}
+
+void
+biodone(buf_t *bp)
+{
+	if (bp->b_iodone != NULL) {
+		(*(bp->b_iodone))(bp);
+		return;
+	}
+	ASSERT((bp->b_flags & B_DONE) == 0);
+	bp->b_flags |= B_DONE;
+}
+
+void
+bioerror(buf_t *bp, int error)
+{
+	ASSERT(bp != NULL);
+	ASSERT(error >= 0);
+
+	if (error != 0) {
+		bp->b_flags |= B_ERROR;
+	} else {
+		bp->b_flags &= ~B_ERROR;
+	}
+	bp->b_error = error;
+}
+
+
+int
+geterror(struct buf *bp)
+{
+	int error = 0;
+
+	if (bp->b_flags & B_ERROR) {
+		error = bp->b_error;
+		if (!error)
+			error = EIO;
+	}
+	return (error);
+}
diff --git a/usr/src/lib/libzpool/common/sys/zfs_context.h b/usr/src/lib/libzpool/common/sys/zfs_context.h
@@ -603,6 +603,36 @@ extern cyclic_id_t cyclic_add(cyc_handler_t *, cyc_time_t *);
 extern void cyclic_remove(cyclic_id_t);
 extern int cyclic_reprogram(cyclic_id_t, hrtime_t);
 
+/*
+ * Buf structure
+ */
+#define	B_BUSY		0x0001
+#define	B_DONE		0x0002
+#define	B_ERROR		0x0004
+#define	B_READ		0x0040	/* read when I/O occurs */
+#define	B_WRITE		0x0100	/* non-read pseudo-flag */
+
+typedef struct buf {
+	int	b_flags;
+	size_t b_bcount;
+	union {
+		caddr_t b_addr;
+	} b_un;
+
+	lldaddr_t	_b_blkno;
+#define	b_lblkno	_b_blkno._f
+	size_t	b_resid;
+	size_t	b_bufsize;
+	int	(*b_iodone)(struct buf *);
+	int	b_error;
+	void	*b_private;
+} buf_t;
+
+extern void bioinit(buf_t *);
+extern void biodone(buf_t *);
+extern void bioerror(buf_t *, int);
+extern int geterror(buf_t *);
+
 #ifdef	__cplusplus
 }
 #endif

diff --git a/usr/src/man/man1m/zdb.1m b/usr/src/man/man1m/zdb.1m
@@ -11,6 +11,7 @@
 .\"
 .\"
 .\" Copyright 2012, Richard Lowe.
+.\" Copyright (c) 2012 by Delphix. All rights reserved.
 .\"
 .TH "ZDB" "1M" "February 15, 2012" "" ""
 
@@ -19,21 +20,23 @@
 
 .SH "SYNOPSIS"
 \fBzdb\fR [-CumdibcsDvhLXFPA] [-e [-p \fIpath\fR...]] [-t \fItxg\fR]
-    \fIpoolname\fR [\fIobject\fR ...]
+    [-U \fIcache\fR] [-M \fIinflight I/Os\fR] [\fIpoolname\fR
+    [\fIobject\fR ...]]
 
 .P
-\fBzdb\fR [-divPA] [-e [-p \fIpath\fR...]] \fIdataset\fR [\fIobject\fR ...]
+\fBzdb\fR [-divPA] [-e [-p \fIpath\fR...]] [-U \fIcache\fR]
+    \fIdataset\fR [\fIobject\fR ...]
 
 .P
-\fBzdb\fR -m [-LXFPA] [-t \fItxg\fR] [-e [-p \fIpath\fR...]] \fIpoolname\fR
-    [\fIvdev\fR [\fImetaslab\fR ...]]
+\fBzdb\fR -m [-LXFPA] [-t \fItxg\fR] [-e [-p \fIpath\fR...]] [-U \fIcache\fR]
+    \fIpoolname\fR [\fIvdev\fR [\fImetaslab\fR ...]]
 
 .P
-\fBzdb\fR -R [-A] [-e [-p \fIpath\fR...]] \fIpoolname\fR
+\fBzdb\fR -R [-A] [-e [-p \fIpath\fR...]] [-U \fIcache\fR] \fIpoolname\fR
     \fIvdev\fR:\fIoffset\fR:\fIsize\fR[:\fIflags\fR]
 
 .P
-\fBzdb\fR -S [-AP] [-e [-p \fIpath\fR...]] \fIpoolname\fR
+\fBzdb\fR -S [-AP] [-e [-p \fIpath\fR...]] [-U \fIcache\fR] \fIpoolname\fR
 
 .P
 \fBzdb\fR -l [-uA] \fIdevice\fR
@@ -354,6 +357,18 @@ Attempt to make an unreadable pool readable by trying progressively older
 transactions.
 .RE
 
+.sp
+.ne 2
+.na
+\fB-M \fIinflight I/Os\fR \fR
+.ad
+.sp .6
+.RS 4n
+Limit the number of outstanding checksum I/Os to the specified value. The
+default value is 200. This option affects the performance of the \fB-c\fR
+option.
+.RE
+
 .sp
 .ne 2
 .na
@@ -384,8 +399,7 @@ and their associated transaction numbers.
 .ad
 .sp .6
 .RS 4n
-Use a cache file other than \fB/etc/zfs/zpool.cache\fR. This option is only
-valid with \fB-C\fR
+Use a cache file other than \fB/etc/zfs/zpool.cache\fR.
 .RE
 
 .sp