yeahuh / rpms / qemu-kvm

Forked from rpms/qemu-kvm 2 years ago
Clone

Blame SOURCES/kvm-block-backup-fix-fleecing-scheme-use-serialized-writ.patch

26ba25
From d9a55a5815a040032f85c20020b118dda54bba43 Mon Sep 17 00:00:00 2001
26ba25
From: John Snow <jsnow@redhat.com>
26ba25
Date: Wed, 18 Jul 2018 22:54:58 +0200
26ba25
Subject: [PATCH 240/268] block/backup: fix fleecing scheme: use serialized
26ba25
 writes
26ba25
26ba25
RH-Author: John Snow <jsnow@redhat.com>
26ba25
Message-id: <20180718225511.14878-23-jsnow@redhat.com>
26ba25
Patchwork-id: 81396
26ba25
O-Subject: [RHEL-7.6 qemu-kvm-rhev PATCH 22/35] block/backup: fix fleecing scheme: use serialized writes
26ba25
Bugzilla: 1207657
26ba25
RH-Acked-by: Eric Blake <eblake@redhat.com>
26ba25
RH-Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
26ba25
RH-Acked-by: Fam Zheng <famz@redhat.com>
26ba25
26ba25
From: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
26ba25
26ba25
Fleecing scheme works as follows: we want a kind of temporary snapshot
26ba25
of active drive A. We create temporary image B, with B->backing = A.
26ba25
Then we start backup(sync=none) from A to B. From this point, B reads
26ba25
as point-in-time snapshot of A (A continues to be active drive,
26ba25
accepting guest IO).
26ba25
26ba25
This scheme needs some additional synchronization between reads from B
26ba25
and backup COW operations, otherwise, the following situation is
26ba25
theoretically possible:
26ba25
26ba25
(assume B is qcow2, client is NBD client, reading from B)
26ba25
26ba25
1. client starts reading and take qcow2 mutex in qcow2_co_preadv, and
26ba25
   goes up to l2 table loading (assume cache miss)
26ba25
26ba25
2) guest write => backup COW => qcow2 write =>
26ba25
   try to take qcow2 mutex => waiting
26ba25
26ba25
3. l2 table loaded, we see that cluster is UNALLOCATED, go to
26ba25
   "case QCOW2_CLUSTER_UNALLOCATED" and unlock mutex before
26ba25
   bdrv_co_preadv(bs->backing, ...)
26ba25
26ba25
4) aha, mutex unlocked, backup COW continues, and we finally finish
26ba25
   guest write and change cluster in our active disk A
26ba25
26ba25
5. actually, do bdrv_co_preadv(bs->backing, ...) and read
26ba25
   _new updated_ data.
26ba25
26ba25
To avoid this, let's make backup writes serializing, to not intersect
26ba25
with reads from B.
26ba25
26ba25
Note: we expand range of handled cases from (sync=none and
26ba25
B->backing = A) to just (A in backing chain of B), to finally allow
26ba25
safe reading from B during backup for all cases when A in backing chain
26ba25
of B, i.e. B formally looks like point-in-time snapshot of A.
26ba25
26ba25
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
26ba25
Reviewed-by: Fam Zheng <famz@redhat.com>
26ba25
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
26ba25
(cherry picked from commit f8d59dfb40bbc6f5aeea57c8aac1e68c1d2454ee)
26ba25
Signed-off-by: John Snow <jsnow@redhat.com>
26ba25
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
26ba25
---
26ba25
 block/backup.c | 20 ++++++++++++++------
26ba25
 1 file changed, 14 insertions(+), 6 deletions(-)
26ba25
26ba25
diff --git a/block/backup.c b/block/backup.c
26ba25
index 369155a..4ba1a6a 100644
26ba25
--- a/block/backup.c
26ba25
+++ b/block/backup.c
26ba25
@@ -47,6 +47,8 @@ typedef struct BackupBlockJob {
26ba25
     HBitmap *copy_bitmap;
26ba25
     bool use_copy_range;
26ba25
     int64_t copy_range_size;
26ba25
+
26ba25
+    bool serialize_target_writes;
26ba25
 } BackupBlockJob;
26ba25
 
26ba25
 static const BlockJobDriver backup_job_driver;
26ba25
@@ -102,6 +104,8 @@ static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job,
26ba25
     QEMUIOVector qiov;
26ba25
     BlockBackend *blk = job->common.blk;
26ba25
     int nbytes;
26ba25
+    int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
26ba25
+    int write_flags = job->serialize_target_writes ? BDRV_REQ_SERIALISING : 0;
26ba25
 
26ba25
     hbitmap_reset(job->copy_bitmap, start / job->cluster_size, 1);
26ba25
     nbytes = MIN(job->cluster_size, job->len - start);
26ba25
@@ -112,8 +116,7 @@ static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job,
26ba25
     iov.iov_len = nbytes;
26ba25
     qemu_iovec_init_external(&qiov, &iov, 1);
26ba25
 
26ba25
-    ret = blk_co_preadv(blk, start, qiov.size, &qiov,
26ba25
-                        is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0);
26ba25
+    ret = blk_co_preadv(blk, start, qiov.size, &qiov, read_flags);
26ba25
     if (ret < 0) {
26ba25
         trace_backup_do_cow_read_fail(job, start, ret);
26ba25
         if (error_is_read) {
26ba25
@@ -124,11 +127,11 @@ static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job,
26ba25
 
26ba25
     if (qemu_iovec_is_zero(&qiov)) {
26ba25
         ret = blk_co_pwrite_zeroes(job->target, start,
26ba25
-                                   qiov.size, BDRV_REQ_MAY_UNMAP);
26ba25
+                                   qiov.size, write_flags | BDRV_REQ_MAY_UNMAP);
26ba25
     } else {
26ba25
         ret = blk_co_pwritev(job->target, start,
26ba25
-                             qiov.size, &qiov,
26ba25
-                             job->compress ? BDRV_REQ_WRITE_COMPRESSED : 0);
26ba25
+                             qiov.size, &qiov, write_flags |
26ba25
+                             (job->compress ? BDRV_REQ_WRITE_COMPRESSED : 0));
26ba25
     }
26ba25
     if (ret < 0) {
26ba25
         trace_backup_do_cow_write_fail(job, start, ret);
26ba25
@@ -156,6 +159,8 @@ static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job,
26ba25
     int nr_clusters;
26ba25
     BlockBackend *blk = job->common.blk;
26ba25
     int nbytes;
26ba25
+    int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
26ba25
+    int write_flags = job->serialize_target_writes ? BDRV_REQ_SERIALISING : 0;
26ba25
 
26ba25
     assert(QEMU_IS_ALIGNED(job->copy_range_size, job->cluster_size));
26ba25
     nbytes = MIN(job->copy_range_size, end - start);
26ba25
@@ -163,7 +168,7 @@ static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job,
26ba25
     hbitmap_reset(job->copy_bitmap, start / job->cluster_size,
26ba25
                   nr_clusters);
26ba25
     ret = blk_co_copy_range(blk, start, job->target, start, nbytes,
26ba25
-                            is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0, 0);
26ba25
+                            read_flags, write_flags);
26ba25
     if (ret < 0) {
26ba25
         trace_backup_do_cow_copy_range_fail(job, start, ret);
26ba25
         hbitmap_set(job->copy_bitmap, start / job->cluster_size,
26ba25
@@ -701,6 +706,9 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
26ba25
                        sync_bitmap : NULL;
26ba25
     job->compress = compress;
26ba25
 
26ba25
+    /* Detect image-fleecing (and similar) schemes */
26ba25
+    job->serialize_target_writes = bdrv_chain_contains(target, bs);
26ba25
+
26ba25
     /* If there is no backing file on the target, we cannot rely on COW if our
26ba25
      * backup cluster size is smaller than the target cluster size. Even for
26ba25
      * targets with a backing file, try to avoid COW if possible. */
26ba25
-- 
26ba25
1.8.3.1
26ba25