Messages in this thread Patch in this message | | | Subject | Re: [PATCH 3/4 v2] fs: proc: use down_read_killable() in environ_read() | From | Yang Shi <> | Date | Tue, 27 Feb 2018 08:59:06 -0800 |
| |
On 2/26/18 11:15 PM, Alexey Dobriyan wrote: > On Tue, Feb 27, 2018 at 08:25:50AM +0800, Yang Shi wrote: >> Like reading /proc/*/cmdline, it is possible to be blocked for long time >> when reading /proc/*/environ when manipulating large mapping at the mean >> time. The environ reading process will be waiting for mmap_sem become >> available for a long time then it may cause the reading task hung. >> >> Convert down_read() and access_remote_vm() to killable version. >> >> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> >> Suggested-by: Alexey Dobriyan <adobriyan@gmail.com> > Ehh, bloody tags.
I mean fix reading /proc/*/environ part.
> I didn't suggest _killable() variants, they're quite ugly because API > multiplies. access_remote_vm() could be converted to down_read_killable().
There might be other places that need non-killable access_remote_vm(), like patch 4/4.
And, it sounds keeping access_remote_vm() semantic intact may prevent from confusion. The most people may get used to assume access_remote_vm() behaves like access_process_vm() except not inc mm reference count.
Thanks, Yang
> >> --- a/fs/proc/base.c >> +++ b/fs/proc/base.c >> @@ -933,7 +933,9 @@ static ssize_t environ_read(struct file *file, char __user *buf, >> if (!mmget_not_zero(mm)) >> goto free; >> >> - down_read(&mm->mmap_sem); >> + ret = down_read_killable(&mm->mmap_sem); >> + if (ret) >> + goto out_mmput; >> env_start = mm->env_start; >> env_end = mm->env_end; >> up_read(&mm->mmap_sem); >> @@ -950,7 +952,8 @@ static ssize_t environ_read(struct file *file, char __user *buf, >> max_len = min_t(size_t, PAGE_SIZE, count); >> this_len = min(max_len, this_len); >> >> - retval = access_remote_vm(mm, (env_start + src), page, this_len, 0); >> + retval = access_remote_vm_killable(mm, (env_start + src), >> + page, this_len, 0);
| |