From: Michal Hocko <mhocko@kernel.org>
To: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: linux-mm@kvack.org, akpm@linux-foundation.org,
david@fromorbit.com, tytso@mit.edu, clm@fb.com, jack@suse.cz,
linux-kernel@vger.kernel.org, zohar@linux.vnet.ibm.com
Subject: Re: [PATCH 1/2] mm: add PF_MEMALLOC_NOFS
Date: Wed, 27 Apr 2016 22:05:30 +0200 [thread overview]
Message-ID: <20160427200530.GB22544@dhcp22.suse.cz> (raw)
In-Reply-To: <201604272344.JHJ05701.JFSQtLHFOOMVOF@I-love.SAKURA.ne.jp>
On Wed 27-04-16 23:44:35, Tetsuo Handa wrote:
> Michal Hocko wrote:
> > On Wed 27-04-16 19:53:21, Tetsuo Handa wrote:
> > [...]
> > > > Let's hope that filesystems will drop direct GFP_NOFS (resp. ~__GFP_FS)
> > > > usage as much and possible and only use a properly documented
> > > > memalloc_nofs_{save,restore} checkpoints where they are appropriate.
> > >
> > > Is the story simple enough to monotonically replace GFP_NOFS/GFP_NOIO
> > > with GFP_KERNEL after memalloc_no{fs,io}_{save,restore} are inserted?
> > > We sometimes delegate some operations to somebody else. Don't we need to
> > > convey PF_MEMALLOC_NOFS/PF_MEMALLOC_NOIO flags to APIs which interact with
> > > other threads?
> >
> > We can add an api to do that if that is really needed.
> >
>
> I'm not familiar with integrity subsystem.
> But if call traces shown below is possible and evm_verify_hmac() is called from
> genuine GFP_NOFS context, we are currently using GFP_KERNEL incorrectly.
> Therefore, inserting memalloc_nofs_{save,restore} would avoid possible memory
> reclaim deadlock by __GFP_FS.
I am not familiar with this code as well but you are definitely right
that scope GFP_NOFS would be better. I have a suspicious that NOFS is
used here improperly and it just copies the same gfp mask used for
all allocations in the same file without any good reason.
This would be a question for Mimi. Could you clarify please?
> ----------
> static enum integrity_status evm_verify_hmac(struct dentry *dentry, const char *xattr_name, char *xattr_value, size_t xattr_value_len, struct integrity_iint_cache *iint) {
> rc = vfs_getxattr_alloc(dentry, XATTR_NAME_EVM, (char **)&xattr_data, 0, GFP_NOFS); /***** GFP_NOFS is used here. *****/
> rc = integrity_digsig_verify(INTEGRITY_KEYRING_EVM, (const char *)xattr_data, xattr_len, calc.digest, sizeof(calc.digest)) {
> keyring[id] = request_key(&key_type_keyring, keyring_name[id], NULL) {
> key = request_key_and_link(type, description, callout_info, callout_len, NULL, NULL, KEY_ALLOC_IN_QUOTA) {
> key = construct_key_and_link(&ctx, callout_info, callout_len, aux, dest_keyring, flags) {
> ret = construct_alloc_key(ctx, dest_keyring, flags, user, &key) {
> key = key_alloc(ctx->index_key.type, ctx->index_key.description, ctx->cred->fsuid, ctx->cred->fsgid, ctx->cred, perm, flags) {
> key = kmem_cache_zalloc(key_jar, GFP_KERNEL); /***** Needs to use GFP_NOFS here if above GFP_NOFS usage is correct. *****/
> }
> }
> ret = construct_key(key, callout_info, callout_len, aux, dest_keyring) {
> cons = kmalloc(sizeof(*cons), GFP_KERNEL); /***** Ditto. *****/
> actor = call_sbin_request_key;
> ret = actor(cons, "create", aux) {
> ret = call_usermodehelper_keys(argv[0], argv, envp, keyring, UMH_WAIT_PROC) {
> info = call_usermodehelper_setup(path, argv, envp, GFP_KERNEL, umh_keys_init, umh_keys_cleanup, session_keyring); /***** Ditto. *****/
> return call_usermodehelper_exec(info, wait) {
> queue_work(system_unbound_wq, &sub_info->work); /***** Queuing a GFP_NOFS work item here if above GFP_NOFS usage is correct. *****/
> wait_for_completion(&done); /***** But kworker uses GFP_KERNEL to create process for executing userspace program. *****/
> }
> }
> }
> }
> }
> }
> }
> }
> }
> ----------
>
> But there is a path where evm_verify_hmac() calls usermode helper.
> If evm_verify_hmac() calls usermode helper from genuine GFP_NOFS context,
> we will be still failing to tell kworker to use GFP_NOFS.
This would be a terrible thing to do. Because ...
> More problematic thing might be that we queue both GFP_KERNEL work item
> and GFP_NOFS work item into the same work queue. This means that the
> kworker will try __GFP_FS reclaim if current GFP_KERNEL work item
> and be blocked on a fs lock held by next GFP_NOFS work item. Then, simply
> conveying PF_MEMALLOC_NOFS/PF_MEMALLOC_NOIO flags to other threads is not
> sufficient, and we need to create separate workqueues (and respective
> consumers) for GFP_KERNEL work items and GFP_NOFS work items?
... of this very reason. If some GFP_NOFS code path relies on kworkers
and wait for the work synchronously then it really has to make sure that
the WQ has a rescuer and there are no __GFP_FS allocation requeuests
enqueued on the same WQ.
> (Or we have no such problem because khelper_wq was replaced with
> system_unbound_wq ?)
I do not think so. system_unbound_wq still depends to have some workers
and that might be not true under memory pressure.
Thanks!
--
Michal Hocko
SUSE Labs
next prev parent reply other threads:[~2016-04-27 20:05 UTC|newest]
Thread overview: 38+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-04-26 11:56 [PATCH 0/2] scop GFP_NOFS api Michal Hocko
2016-04-26 11:56 ` [PATCH 1/2] mm: add PF_MEMALLOC_NOFS Michal Hocko
2016-04-26 23:07 ` Dave Chinner
2016-04-27 7:51 ` Michal Hocko
2016-04-27 10:53 ` Tetsuo Handa
2016-04-27 11:15 ` Michal Hocko
2016-04-27 14:44 ` Tetsuo Handa
2016-04-27 20:05 ` Michal Hocko [this message]
2016-04-27 11:54 ` [PATCH 1.1/2] xfs: abstract PF_FSTRANS to PF_MEMALLOC_NOFS Michal Hocko
2016-04-27 11:54 ` [PATCH 1.2/2] mm: introduce memalloc_nofs_{save,restore} API Michal Hocko
2016-04-27 13:07 ` Michal Hocko
2016-04-27 20:09 ` Michal Hocko
2016-04-27 20:30 ` Michal Hocko
2016-04-27 21:14 ` Michal Hocko
2016-04-27 17:41 ` [PATCH 1.1/2] xfs: abstract PF_FSTRANS to PF_MEMALLOC_NOFS Andreas Dilger
2016-04-27 19:43 ` Michal Hocko
2016-04-26 11:56 ` [PATCH 2/2] mm, debug: report when GFP_NO{FS,IO} is used explicitly from memalloc_no{fs,io}_{save,restore} context Michal Hocko
2016-04-26 22:58 ` Dave Chinner
2016-04-27 8:03 ` Michal Hocko
2016-04-27 22:55 ` Dave Chinner
2016-04-28 8:17 ` Michal Hocko
2016-04-28 21:51 ` Dave Chinner
2016-04-29 12:12 ` Michal Hocko
2016-04-29 23:40 ` Dave Chinner
2016-05-03 15:38 ` Michal Hocko
2016-05-04 0:07 ` Dave Chinner
2016-04-29 5:35 ` [PATCH 0/2] scop GFP_NOFS api NeilBrown
2016-04-29 10:20 ` [Cluster-devel] " Steven Whitehouse
2016-04-30 21:17 ` NeilBrown
2016-04-29 12:04 ` Michal Hocko
2016-04-30 0:24 ` Dave Chinner
2016-04-30 21:55 ` NeilBrown
2016-05-03 15:13 ` Michal Hocko
2016-05-03 23:26 ` NeilBrown
2016-04-30 0:11 ` Dave Chinner
2016-04-30 22:19 ` NeilBrown
2016-05-04 1:00 ` Dave Chinner
2016-05-06 3:20 ` NeilBrown
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20160427200530.GB22544@dhcp22.suse.cz \
--to=mhocko@kernel.org \
--cc=akpm@linux-foundation.org \
--cc=clm@fb.com \
--cc=david@fromorbit.com \
--cc=jack@suse.cz \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=penguin-kernel@I-love.SAKURA.ne.jp \
--cc=tytso@mit.edu \
--cc=zohar@linux.vnet.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).