package/apache: atomic creation of pid file.
PATCH 1/1 package/apache: atomic creation of pid file.
The original pattern for creating the pid file was:open_create(pid_file)write(pid_file, pid)close(pid_file)
But if a power outage occurs between open_create and write, the file will be empty and httpd will refuse to start afterwards unless the corrupt pid file is removed.
This patch uses the pattern:open_create(temp_pid_file)write(temp_pid_file)close(temp_pid_file)rename(temp_pid_file, pid_file)
which is guaranteed to be atomic, provided that temp_pid_file and pid_file are located in the same file system, which this patch does by creating a temporary file name with the pattern:
pid_file_name + random_suffix
This patch has been submitted upstream, but did receive no answer at all:
https://bz.apache.org/bugzilla/show_bug.cgi?id=63140
--- package/apache/0003-corrupt-pid-file.patch | 51 ++++++++++++++++++++++ 1 file changed, 51 insertions(+) create mode 100644 package/apache/0003-corrupt-pid-file.patch
package/apache/0003-corrupt-pid-file.patch new file +51 -0
@@ -0,0 +1,51 @@ +diff -ur httpd-2.4.35.ref/server/log.c httpd-2.4.35/server/log.c +--- httpd-2.4.35.ref/server/log.c 2018-06-25 15:24:16.000000000 +0200 ++++ httpd-2.4.35/server/log.c 2019-02-04 10:11:22.313157132 +0100 +@@ -1598,6 +1598,8 @@ + pid_t mypid; + apr_status_t rv; + const char *fname; ++ char *temp_fname; ++ apr_fileperms_t perms; + + if (!filename) { + return; +@@ -1609,6 +1611,10 @@ + ap_server_conf, APLOGNO(00097) "Invalid PID file path %s, ignoring.", filename); + return; + } ++ temp_fname = apr_psprintf(p, "%s.XXXXXX", filename); ++ if (!temp_fname) { ++ return; ++ } + + mypid = getpid(); + if (mypid != saved_pid +@@ -1626,19 +1632,23 @@ + fname); + } + +- if ((rv = apr_file_open(&pid_file, fname, +- APR_WRITE | APR_CREATE | APR_TRUNCATE, +- APR_UREAD | APR_UWRITE | APR_GREAD | APR_WREAD, p)) ++ if ((rv = apr_file_mktemp(&pid_file, temp_fname, ++ APR_WRITE | APR_CREATE | APR_TRUNCATE, p)) + != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ERR, rv, NULL, APLOGNO(00099) +- "could not create %s", fname); ++ "could not create %s", temp_fname); + ap_log_error(APLOG_MARK, APLOG_ERR, 0, NULL, APLOGNO(00100) + "%s: could not log pid to file %s", + ap_server_argv0, fname); + exit(1); + } ++ ++ perms = APR_UREAD | APR_UWRITE | APR_GREAD | APR_WREAD; ++ apr_file_perms_set(temp_fname, perms); ++ + apr_file_printf(pid_file, "%" APR_PID_T_FMT APR_EOL_STR, mypid); + apr_file_close(pid_file); ++ apr_file_rename(temp_fname, fname, p); + saved_pid = mypid; + } +
Signature · 1 line
2.20.1
PATCH 1/1 package/apache: atomic creation of pid file.
Quoted 19 lines · On Fri, 20 Sep 2019 15:04:36 +0000 Nicolas Carrier…
The original pattern for creating the pid file was:
open_create(pid_file)write(pid_file, pid)close(pid_file)But if a power outage occurs between
open_createand write, the file will be empty and httpd will refuse to start afterwards unless the corrupt pid file is removed.This patch uses the pattern:
open_create(temp_pid_file)write(temp_pid_file)close(temp_pid_file)rename(temp_pid_file, pid_file)
which is guaranteed to be atomic, provided thattemp_pid_fileandpid_fileare located in the same file system, which this patch does by creating a temporary file name with the pattern:
pid_file_name+random_suffix
Quoted 4 lines
This patch has been submitted upstream, but did receive no answer at all:
https://bz.apache.org/bugzilla/show_bug.cgi?id=63140Signed-off-by Nicolas Carrier <[email protected]>
We're not really fond of having patches like this that haven't been reviewed/blessed by upstream. But I see you submitted since months ago and got no feedback.
package/apache/0003-corrupt-pid-file.patch
Quoted 11 lines
--- package/apache/0003-corrupt-pid-file.patch | 51 ++++++++++++++++++++++ 1 file changed, 51 insertions(+) create mode 100644 package/apache/0003-corrupt-pid-file.patchpackage/apache/0003-corrupt-pid-file.patch new file +0
-0@@ -0,0 +1,51 @@
In any case, we need a description + Signed-off-by line in the patch itself, and since Apache is available in a Git repo
(https://github.com/apache/httpd), we want the patch to be generated by git format-patch.
Thanks!
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com