Re: [PATCH v2 5/6] proc: use down_read_killable mmap_sem for /proc/pid/map_files

From: Andrew Morton
Date: Thu Jun 13 2019 - 13:05:02 EST


On Wed, 12 Jun 2019 16:14:28 -0700 Andrei Vagin <avagin@xxxxxxxxx> wrote:

> On Sun, Jun 09, 2019 at 01:09:00PM +0300, Konstantin Khlebnikov wrote:
> > Do not stuck forever if something wrong.
> > Killable lock allows to cleanup stuck tasks and simplifies investigation.
>
> This patch breaks the CRIU project, because stat() returns EINTR instead
> of ENOENT:
>
> [root@fc24 criu]# stat /proc/self/map_files/0-0
> stat: cannot stat '/proc/self/map_files/0-0': Interrupted system call
>
> Here is one inline comment with the fix for this issue.
>
> > @@ -2107,7 +2113,10 @@ static struct dentry *proc_map_files_lookup(struct inode *dir,
> > if (!mm)
> > goto out_put_task;
> >
> > - down_read(&mm->mmap_sem);
> > + result = ERR_PTR(-EINTR);
> > + if (down_read_killable(&mm->mmap_sem))
> > + goto out_put_mm;
> > +
>
> result = ERR_PTR(-ENOENT);
>

yes, thanks.

--- a/fs/proc/base.c~proc-use-down_read_killable-mmap_sem-for-proc-pid-map_files-fix
+++ a/fs/proc/base.c
@@ -2117,6 +2117,7 @@ static struct dentry *proc_map_files_loo
if (down_read_killable(&mm->mmap_sem))
goto out_put_mm;

+ result = ERR_PTR(-ENOENT);
vma = find_exact_vma(mm, vm_start, vm_end);
if (!vma)
goto out_no_vma;



We started doing this trick of using

ret = -EFOO;
if (something)
goto out;

decades ago because it generated slightly better code. I rather doubt
whether that's still the case.

In fact this:

--- a/fs/proc/base.c~a
+++ a/fs/proc/base.c
@@ -2096,35 +2096,45 @@ static struct dentry *proc_map_files_loo
struct dentry *result;
struct mm_struct *mm;

- result = ERR_PTR(-ENOENT);
task = get_proc_task(dir);
- if (!task)
+ if (!task) {
+ result = ERR_PTR(-ENOENT);
goto out;
+ }

- result = ERR_PTR(-EACCES);
- if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
+ if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
+ result = ERR_PTR(-EACCES);
goto out_put_task;
+ }

- result = ERR_PTR(-ENOENT);
- if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
+ if (dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
+ result = ERR_PTR(-ENOENT);
goto out_put_task;
+ }

mm = get_task_mm(task);
- if (!mm)
+ if (!mm) {
+ result = ERR_PTR(-ENOENT);
goto out_put_task;
+ }

- result = ERR_PTR(-EINTR);
- if (down_read_killable(&mm->mmap_sem))
+ if (down_read_killable(&mm->mmap_sem)) {
+ result = ERR_PTR(-EINTR);
goto out_put_mm;
+ }

- result = ERR_PTR(-ENOENT);
vma = find_exact_vma(mm, vm_start, vm_end);
- if (!vma)
+ if (!vma) {
+ result = ERR_PTR(-ENOENT);
goto out_no_vma;
+ }

- if (vma->vm_file)
+ if (vma->vm_file) {
result = proc_map_files_instantiate(dentry, task,
(void *)(unsigned long)vma->vm_file->f_mode);
+ } else {
+ result = ERR_PTR(-ENOENT);
+ }

out_no_vma:
up_read(&mm->mmap_sem);

makes no change to the generated assembly with gcc-7.2.0.

And I do think that the above style is clearer and more reliable, as
this bug demonstrates.