exec: remove redundant save asides of old pid/vpid
Subsystems (3)
PATCH 1/1 exec: remove redundant save asides of old pid/vpid
Problem: Old pid and vpid are redundantly saved aside before starting to parse the binary, with the comment claiming that it is required since load_binary changes it, though from inspection in the source, load_binary does not change the pid and this wouldn't make sense since execve does not create any new process, quote from man page of execve:
"there is no new process; many attributes of the calling process remain unchanged (in particular, its PID)."
Solution: Remove the saving aside of both and later on use them directly from the current object, instead of via the saved aside objects.
---
Side-note: Tested this solution with a defconfig x86_64 and an initramfs with Busybox and confirmed to work fine.
fs/exec.c | 12 +++--------- 1 file changed, 3 insertions(+), 9 deletions(-)
fs/exec.c +3 -9
@@ -1789,15 +1789,8 @@ static int search_binary_handler(struct linux_binprm *bprm)/* binfmt handlers will call back into begin_new_exec() on success. */ static int exec_binprm (struct linux_binprm *bprm) { -pid_t old_pid, old_vpid;int ret, depth; -/* Need to fetch pid before load_binary changes it */ - old_pid = current->pid; -rcu_read_lock (); - old_vpid =task_pid_nr_ns (current,task_active_pid_ns (current->parent)); -rcu_read_unlock (); -/* This allows 4 levels of binfmt rewrites before failing hard. */ for (depth =0 ;; depth++) {struct file *exec;
@@ -1826,8 +1819,9 @@ static int exec_binprm(struct linux_binprm *bprm) }audit_bprm (bprm); -trace_sched_process_exec (current, old_pid, bprm); -ptrace_event (PTRACE_EVENT_EXEC , old_vpid); +trace_sched_process_exec (current, current->pid, bprm); +ptrace_event (PTRACE_EVENT_EXEC , +task_pid_nr_ns (current,task_active_pid_ns (current->parent)));proc_exec_connector (current);return 0 ; }
Signature · 1 line
2.39.5
Kees Cook
maintainerQuoted 7 lines · Nir Lichtman
Problem: Old pid and vpid are redundantly saved aside before starting to parse the binary, with the comment claiming that it is required since
load_binarychanges it, though from inspection in the source,load_binarydoes not change the pid and this wouldn't make sense since execve does not create any new process, quote from man page of execve:
"there is no new process; many attributes of the calling process remain unchanged (in particular, its PID)."
See commit bb188d7e64de ("ptrace: make former thread ID available via PTRACE_GETEVENTMSG after PTRACE_EVENT_EXEC stop")
This is for making sense of a concurrent exec made by a multi threaded process. Specifically see de_thread(), where the pid *can* change:
/*
* At this point all other threads have exited, all we have to
* do is to wait for the thread group leader to become inactive,
* and to assume its PID:
*/
The described problem in the commit hasn't changed, so this code needs to stay as-is. Or perhaps the comment could be improved?
-Kees
Quoted 45 lines
Solution: Remove the saving aside of both and later on use them directly from the current object, instead of via the saved aside objects.
Signed-off-by Nir Lichtman <[email protected]>---
Side-note: Tested this solution with a defconfig
x86_64and an initramfs with Busybox and confirmed to work fine.
fs/exec.c| 12 +++---------1 file changed, 3 insertions(+), 9 deletions(-)fs/exec.c +3
-9@@ -1789,15 +1789,8 @@ static int search_binary_handler(struct linux_binprm *bprm)/* binfmt handlers will call back into begin_new_exec() on success. */ static int exec_binprm (struct linux_binprm *bprm) { -pid_t old_pid, old_vpid;int ret, depth; -/* Need to fetch pid before load_binary changes it */ - old_pid = current->pid; -rcu_read_lock (); - old_vpid =task_pid_nr_ns (current,task_active_pid_ns (current->parent)); -rcu_read_unlock (); -/* This allows 4 levels of binfmt rewrites before failing hard. */ for (depth =0 ;; depth++) {struct file *exec;@@ -1826,8 +1819,9 @@ static int exec_binprm(struct linux_binprm *bprm) }audit_bprm (bprm); -trace_sched_process_exec (current, old_pid, bprm); -ptrace_event (PTRACE_EVENT_EXEC , old_vpid); +trace_sched_process_exec (current, current->pid, bprm); +ptrace_event (PTRACE_EVENT_EXEC , +task_pid_nr_ns (current,task_active_pid_ns (current->parent)));proc_exec_connector (current);return 0 ; }
Signature · 1 line
Kees Cook
Quoted 15 lines · Kees Cook
Quoted 7 lines · Nir Lichtman
Problem: Old pid and vpid are redundantly saved aside before starting to parse the binary, with the comment claiming that it is required since
load_binarychanges it, though from inspection in the source,load_binarydoes not change the pid and this wouldn't make sense since execve does not create any new process, quote from man page of execve:
"there is no new process; many attributes of the calling process remain unchanged (in particular, its PID)."See commit
bb188d7e64de("ptrace: make former thread ID available viaPTRACE_GETEVENTMSGafterPTRACE_EVENT_EXECstop")This is for making sense of a concurrent exec made by a multi threaded process. Specifically see
de_thread(), where the pid *can* change:
Quoted 7 lines
/*
* At this point all other threads have exited, all we have to
* do is to wait for the thread group leader to become inactive,
* and to assume its PID:
*/The described problem in the commit hasn't changed, so this code needs to stay as-is. Or perhaps the comment could be improved?
Thanks for answering, interesting, I'll take a deeper look.
Nir