ca40a59d67
in a bad state. There is a PR upstream to vendor Docker to have this fix as well, but it probably won't get in until 1.11.2. bnc#964673 (https://github.com/docker/docker/pull/22765) + bnc964673-boltdb-metadata-recovery.patch OBS-URL: https://build.opensuse.org/package/show/Virtualization:containers/docker?expand=0&rev=105
96 lines
3.2 KiB
Diff
96 lines
3.2 KiB
Diff
From 8f0e47cee034cdc08ca515d98a6733130908fc26 Mon Sep 17 00:00:00 2001
|
|
From: Aleksa Sarai <asarai@suse.de>
|
|
Date: Mon, 16 May 2016 23:53:46 +1000
|
|
Subject: [PATCH] db: fix recovery from unsynced metadata
|
|
|
|
Bolt stores the two latest transactions' metadata, but previously did
|
|
not recover from validation failures in the latest by using the second
|
|
latest. Fix this by correctly handling validation failures in db.go, as
|
|
well as returning the metadata with highest txid which is also valid in
|
|
DB.meta().
|
|
|
|
Signed-off-by: Aleksa Sarai <asarai@suse.de>
|
|
---
|
|
vendor/src/github.com/boltdb/bolt/db.go | 49 +++++++++++++++++++++++++--------
|
|
1 file changed, 38 insertions(+), 11 deletions(-)
|
|
|
|
diff --git a/vendor/src/github.com/boltdb/bolt/db.go b/vendor/src/github.com/boltdb/bolt/db.go
|
|
index 501d36aac24a..f713485ffab6 100644
|
|
--- a/vendor/src/github.com/boltdb/bolt/db.go
|
|
+++ b/vendor/src/github.com/boltdb/bolt/db.go
|
|
@@ -200,9 +200,15 @@ func Open(path string, mode os.FileMode, options *Options) (*DB, error) {
|
|
if _, err := db.file.ReadAt(buf[:], 0); err == nil {
|
|
m := db.pageInBuffer(buf[:], 0).meta()
|
|
if err := m.validate(); err != nil {
|
|
- return nil, err
|
|
+ // If we can't read the page size, we can assume it's the same
|
|
+ // as the OS -- since that's how the page size was chosen in the
|
|
+ // first place.
|
|
+ // XXX: Does this cause issues with opening a database on a
|
|
+ // different OS than the one it was created on?
|
|
+ db.pageSize = os.Getpagesize()
|
|
+ } else {
|
|
+ db.pageSize = int(m.pageSize)
|
|
}
|
|
- db.pageSize = int(m.pageSize)
|
|
}
|
|
}
|
|
|
|
@@ -262,12 +268,13 @@ func (db *DB) mmap(minsz int) error {
|
|
db.meta0 = db.page(0).meta()
|
|
db.meta1 = db.page(1).meta()
|
|
|
|
- // Validate the meta pages.
|
|
- if err := db.meta0.validate(); err != nil {
|
|
- return err
|
|
- }
|
|
- if err := db.meta1.validate(); err != nil {
|
|
- return err
|
|
+ // Validate the meta pages. We only return an error if both meta pages fail
|
|
+ // validation, since meta0 failing validation means that it wasn't saved
|
|
+ // properly -- but we can recover using meta1. And vice-versa.
|
|
+ err0 := db.meta0.validate()
|
|
+ err1 := db.meta1.validate()
|
|
+ if err0 != nil && err1 != nil {
|
|
+ return fmt.Errorf("meta0(%v) meta1(%v)", err0, err1)
|
|
}
|
|
|
|
return nil
|
|
@@ -778,10 +785,30 @@ func (db *DB) pageInBuffer(b []byte, id pgid) *page {
|
|
|
|
// meta retrieves the current meta page reference.
|
|
func (db *DB) meta() *meta {
|
|
- if db.meta0.txid > db.meta1.txid {
|
|
- return db.meta0
|
|
+ // We have to return the meta with the highest txid which doesn't fail
|
|
+ // validation. Otherwise, we can cause errors when in fact the database is
|
|
+ // in a consistent state. metaA is the one with the higher txid.
|
|
+ metaA := db.meta0
|
|
+ metaB := db.meta1
|
|
+ if db.meta1.txid > db.meta0.txid {
|
|
+ metaA = db.meta1
|
|
+ metaB = db.meta0
|
|
}
|
|
- return db.meta1
|
|
+
|
|
+ errA := metaA.validate()
|
|
+ errB := metaB.validate()
|
|
+
|
|
+ if errA == nil {
|
|
+ return metaA
|
|
+ }
|
|
+
|
|
+ if errB == nil {
|
|
+ return metaB
|
|
+ }
|
|
+
|
|
+ // This should never be reached, because both meta1 and meta0 were validated
|
|
+ // on mmap() and we do fsync() on every write.
|
|
+ panic("both meta0 and meta1 could not be validated in DB.meta()!")
|
|
}
|
|
|
|
// allocate returns a contiguous block of memory starting at a given page.
|
|
--
|
|
2.8.2
|
|
|