| Message ID | 20171212102453.15258-1-john@phrozen.org |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [LEDE-DEV] base-files: add comment to uci-defaults.sh script | expand |
On 12/12/2017 11:24 AM, John Crispin wrote: > > > board_config_flush() { > + # when calling json_dump with a redirect, the shell will truncate the target > + # file and not update it until a close/flush is called. this results in a short > + # window where the file is 0 bytes. mitigate the issue by generating the file > + # in 2 steps As Roman correctly noted in his patch, this is not the case. The file is moved between different filesystems (tmpfs to actual root), which is never guaranteed to be atomic; internally, rename() can't be used, so mv just falls back to truncate+write as well. The mv makes the race window shorter, but does not completely mitigate it. Either the comment should correctly reflect the situation, or we should actually fix this by creating the temporary file on the root fs. Regards, Matthias > json_dump -i > /tmp/.board.json > mv /tmp/.board.json ${CFG} > } >
diff --git a/package/base-files/files/lib/functions/uci-defaults.sh b/package/base-files/files/lib/functions/uci-defaults.sh index 82ac1d2a4d..fd37088208 100755 --- a/package/base-files/files/lib/functions/uci-defaults.sh +++ b/package/base-files/files/lib/functions/uci-defaults.sh @@ -609,6 +609,10 @@ board_config_update() { } board_config_flush() { + # when calling json_dump with a redirect, the shell will truncate the target + # file and not update it until a close/flush is called. this results in a short + # window where the file is 0 bytes. mitigate the issue by generating the file + # in 2 steps json_dump -i > /tmp/.board.json mv /tmp/.board.json ${CFG} }
add comment explaining why we generate board.json in 2 steps Signed-off-by: John Crispin <john@phrozen.org> --- package/base-files/files/lib/functions/uci-defaults.sh | 4 ++++ 1 file changed, 4 insertions(+)