Secure connection failed on Firefox
mdounin at mdounin.ru
Sat Oct 7 13:16:08 UTC 2017
On Sat, Oct 07, 2017 at 02:14:26AM -0400, joseph-pg wrote:
> Looks like it's the culprit. It happens at random times for a few seconds,
> so it's rather difficult to reproduce.
> Here's the output of the debug log when the error occured (I was wrong,
> weird things do happen here, eg. an entry with alert flag below):
> > 2017/10/07 11:28:05 [debug] 18764#15836: *15 cached open file:
> path/to/website/index.html, fd:560, c:2, e:0, u:16
> > 2017/10/07 11:28:05 [debug] 18764#15836: *15 http copy filter:
> > 2017/10/07 11:28:05 [debug] 18764#15836: *15 malloc:
> > 2017/10/07 11:28:05 [alert] 18764#15836: *15 ReadFile() read only 993 of
> 1013 from "path/to/website/index.html" while sending response to client,
> client: 127.0.0.1, server: my-internaldomain, request: "GET
> /?somequerystring HTTP/2.0", host: "my-internaldomain"
The message suggests that the file in question was non-atomically
modified while being served. It is expected that such a
modification will lead to a fatal error if nginx will be able to
detect the problem. If it won't, likely the client will get a
garbage with a mix of original and new contents of the file.
The only safe approach is to modify files atomically, that is,
create a new file and then use mv (the rename() syscall) to move
it atomically to the appropriate place. It might not be trivial
or even possible to do this correctly on Windows though.
Additionally, it looks like you are using open_file_cache. It is
actually a very bad idea if you modify files in-place, as it
greatly expands the race window between opening and stat()'ing the
file and serving its contents. Remove open_file_cache from the
configuration unless you are sure all file modifications are
More information about the nginx