cluster-devel.redhat.com archive mirror
 help / color / mirror / Atom feed
From: Andrew Price <anprice@redhat.com>
To: cluster-devel.redhat.com
Subject: [Cluster-devel] [PATCH 03/14] libgfs2: Don't try to read more than IOV_MAX iovecs
Date: Thu,  3 Apr 2014 16:12:36 +0100	[thread overview]
Message-ID: <1396537967-12399-4-git-send-email-anprice@redhat.com> (raw)
In-Reply-To: <1396537967-12399-1-git-send-email-anprice@redhat.com>

Reading large collections of blocks, as gfs2_rgrp_read may do with large
rgrps and small block sizes, can cause preadv() to fail with EINVAL in
breadm(). This patch splits large reads into IOV_MAX chunks to avoid
reads failing in that way. It also includes a test to exercise these
changes.

Signed-off-by: Andrew Price <anprice@redhat.com>
---
 gfs2/libgfs2/buf.c | 57 ++++++++++++++++++++++++++++++++----------------------
 tests/mkfs.at      |  4 ++++
 2 files changed, 38 insertions(+), 23 deletions(-)

diff --git a/gfs2/libgfs2/buf.c b/gfs2/libgfs2/buf.c
index 6fcdd17..92cd393 100644
--- a/gfs2/libgfs2/buf.c
+++ b/gfs2/libgfs2/buf.c
@@ -15,6 +15,14 @@
 
 #include "libgfs2.h"
 
+#ifndef IOV_MAX
+  #ifdef UIO_MAXIOV
+    #define IOV_MAX UIO_MAXIOV
+  #else
+    #define IOV_MAX (1024)
+  #endif
+#endif
+
 struct gfs2_buffer_head *bget(struct gfs2_sbd *sdp, uint64_t num)
 {
 	struct gfs2_buffer_head *bh;
@@ -34,31 +42,34 @@ struct gfs2_buffer_head *bget(struct gfs2_sbd *sdp, uint64_t num)
 int __breadm(struct gfs2_sbd *sdp, struct gfs2_buffer_head **bhs, size_t n,
 	     uint64_t block, int line, const char *caller)
 {
-	struct iovec *iov = alloca(n * sizeof(struct iovec));
+	size_t v = (n < IOV_MAX) ? n : IOV_MAX;
+	struct iovec *iov = alloca(v * sizeof(struct iovec));
 	struct iovec *iovbase = iov;
-	uint64_t b = block;
-	size_t size = 0;
-	size_t i;
-	int ret;
-
-	for (i = 0; i < n; i++) {
-		bhs[i] = bget(sdp, b++);
-		if (bhs[i] == NULL)
-			return -1;
-		*iov++ = bhs[i]->iov;
-		size += bhs[i]->iov.iov_len;
+	size_t i = 0;
+
+	while (i < n) {
+		int j;
+		ssize_t ret;
+		ssize_t size = 0;
+
+		for (j = 0; (i + j < n) && (j < IOV_MAX); j++) {
+			bhs[i + j] = bget(sdp, block + i + j);
+			if (bhs[i + j] == NULL)
+				return -1;
+			iov[j] = bhs[i + j]->iov;
+			size += bhs[i + j]->iov.iov_len;
+		}
+
+		ret = preadv(sdp->device_fd, iovbase, j, (block + i) * sdp->bsize);
+		if (ret != size) {
+			fprintf(stderr, "bad read: %s from %s:%d: block %llu (0x%llx) "
+					"count: %d size: %zd ret: %zd\n", strerror(errno),
+					caller, line, (unsigned long long)block,
+					(unsigned long long)block, j, size, ret);
+			exit(-1);
+		}
+		i += j;
 	}
-
-	ret = preadv(sdp->device_fd, iovbase, n, block * sdp->bsize);
-
-	if (ret != size) {
-		fprintf(stderr, "bad read: %s from %s:%d: block "
-				"%llu (0x%llx)\n", strerror(errno),
-				caller, line, (unsigned long long)block,
-				(unsigned long long)block);
-		exit(-1);
-	}
-
 	return 0;
 }
 
diff --git a/tests/mkfs.at b/tests/mkfs.at
index 2616109..ff99bb1 100644
--- a/tests/mkfs.at
+++ b/tests/mkfs.at
@@ -54,6 +54,10 @@ AT_SETUP([Min. resource group size])
 GFS_FSCK_CHECK([$GFS_MKFS -p lock_nolock -r 32 $GFS_TGT])
 AT_CLEANUP
 
+AT_SETUP([Max. resource group size, min. block size])
+GFS_FSCK_CHECK([$GFS_MKFS -p lock_nolock -r 2048 -b 512 $GFS_TGT])
+AT_CLEANUP
+
 AT_SETUP([Max. journal size])
 GFS_FSCK_CHECK([$GFS_MKFS -p lock_nolock -J 1024 $GFS_TGT])
 AT_CLEANUP
-- 
1.8.5.3



  parent reply	other threads:[~2014-04-03 15:12 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-04-03 15:12 [Cluster-devel] [PATCH 00/14] gfs2_grow and libgfs2 rgrp API improvements Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 01/14] mkfs.gfs2: Make dev a member of mkfs_opts Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 02/14] libgfs2: Add lgfs2_space_for_data() Andrew Price
2014-04-03 15:12 ` Andrew Price [this message]
2014-04-03 15:12 ` [Cluster-devel] [PATCH 04/14] mkfs.gfs2: Fix the resource group layout strategy, again Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 05/14] libgfs2: Don't call gfs2_blk2rgrpd in gfs2_set_bitmap Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 06/14] libgfs2: Add abstractions for rgrp tree traversal Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 07/14] libgfs2: Split out the rindex calculation from lgfs2_rgrp_append Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 08/14] libgfs2: Consolidate rgrp_tree and bitstruct allocations Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 09/14] libgfs2: Add a lgfs2_rindex_read_fd() function Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 10/14] libgfs2: Const-ify the 'ri' argument to gfs2_rindex_out Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 11/14] libgfs2: Fix off-by-one in lgfs2_rgrps_plan Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 12/14] libgfs2: Stick to the (rgrp) plan in lgfs2_rindex_entry_new Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 13/14] gfs2_grow: Migrate to the new resource group API Andrew Price
2014-04-03 15:12 ` [Cluster-devel] [PATCH 14/14] gfs2_grow: Add stripe alignment Andrew Price
2014-04-07 15:15 ` [Cluster-devel] [PATCH 00/14] gfs2_grow and libgfs2 rgrp API improvements Bob Peterson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1396537967-12399-4-git-send-email-anprice@redhat.com \
    --to=anprice@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).