fatal error when rolling back transaction with NoFreelistSync

[mappu]

Hi!

I have a user who repeatedly gets this kind of crash:

unexpected fault address 0x7bf31040
fatal error: fault
[signal 0xc0000005 code=0x0 addr=0x7bf31040 pc=0x95e4b3]

goroutine 87 [running]:
runtime.throw(0xb6ac3a, 0x5)
	C:/Go/src/runtime/panic.go:616 +0x6b fp=0x11c69a98 sp=0x11c69a8c pc=0x42928b
runtime.sigpanic()
	C:/Go/src/runtime/signal_windows.go:170 +0x11d fp=0x11c69aac sp=0x11c69a98 pc=0x43ac3d
backupapp/vendor/go.etcd.io/bbolt.(*DB).meta(0x13783900, 0x1410a060)
	vendor/go.etcd.io/bbolt/db.go:901 +0x23 fp=0x11c69ac4 sp=0x11c69aac pc=0x95e4b3
backupapp/vendor/go.etcd.io/bbolt.(*DB).hasSyncedFreelist(0x13783900, 0xc96)
	vendor/go.etcd.io/bbolt/db.go:323 +0x21 fp=0x11c69ad0 sp=0x11c69ac4 pc=0x95c441
backupapp/vendor/go.etcd.io/bbolt.(*Tx).rollback(0x13cdd700)
	vendor/go.etcd.io/bbolt/tx.go:279 +0x69 fp=0x11c69ae8 sp=0x11c69ad0 pc=0x9676e9
backupapp/vendor/go.etcd.io/bbolt.(*Tx).Commit(0x13cdd700, 0x0, 0x0)
	vendor/go.etcd.io/bbolt/tx.go:161 +0x490 fp=0x11c69b7c sp=0x11c69ae8 pc=0x967270
backupapp/vendor/go.etcd.io/bbolt.(*DB).Update(0x13783900, 0x11c69bb0, 0x0, 0x0)
	vendor/go.etcd.io/bbolt/db.go:701 +0xc5 fp=0x11c69ba0 sp=0x11c69b7c pc=0x95dab5

Database is created fresh, in-memory, for this process only, using these bolt.Options:

• Timeout: 3 * time.Second
• NoFreelistSync: true
• FreelistType: bolt.FreelistMapType

Using bbolt v1.3.3, Go 1.10, Windows i386 (for Server 2008 RTM)

At this commit, db.go:901 is looking at db.meta1.txid / db.meta0.txid to order metaA / metaB properly.

It seems like a similar kind of problem to #152 but that particular one should be fixed in v1.3.3 already.

[xiang90]

can you provide a script to reproduce this issue?

[ahrtr]

Fixed in #362.

FYI. The fix will be included in v1.3.7.