Christopher Collins created MYNEWT-160:
------------------------------------------
Summary: NFFS - Files with scattered data blocks can cause restore
failures
Key: MYNEWT-160
URL: https://issues.apache.org/jira/browse/MYNEWT-160
Project: Mynewt
Issue Type: Bug
Components: NFFS
Reporter: Christopher Collins
Assignee: Christopher Collins
Fix For: v0_8_0_beta2
If a file's data blocks are written to disk in an "unexpected" order, the
subsequent restore operation can fail in a few ways:
1. The restore fails with an FS_ECORRUPT error code.
2. The file is only partially restored; the end of the file will be
inaccessible.
This bug is rooted in the inconvenient data structure that NFFS uses for
chained data blocks (reverse singly-linked list). The data structure is
inconvenient, but it is used to minimize RAM and flash usage.
On disk, each data block indicates the ID of the preceding data block in the
parent file. In RAM, a file inode provides access to all the constituent data
blocks by maintaining a pointer to the last data block.
The bug is that the NFFS code does not always ensure that a file inode points
to its last data block. The code assumes that data blocks will be restored in
order of increasing file offset. If data blocks are restored in any other
order, this bug occurs.
The correct fix will be quite involved. I am going to implement a quick fix
for 0.8.0 which fixes the bug, but introduces some horrible inefficiencies when
large files are restored. I will schedule a proper fix for 0.9.0.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
