Re: mm: mkfs.ext4 invoked oom-killer on i386 - pagecache_get_page

[Johannes Weiner <hannes@cmpxchg.org>]

On Thu, May 21, 2020 at 02:44:44PM +0200, Michal Hocko wrote:
> On Thu 21-05-20 05:24:27, Hugh Dickins wrote:
> > On Thu, 21 May 2020, Michal Hocko wrote:
> > > On Thu 21-05-20 16:11:11, Naresh Kamboju wrote:
> > > > On Thu, 21 May 2020 at 15:25, Michal Hocko <mhocko@kernel.org> wrote:
> > > > >
> > > > > On Wed 20-05-20 20:09:06, Chris Down wrote:
> > > > > > Hi Naresh,
> > > > > >
> > > > > > Naresh Kamboju writes:
> > > > > > > As a part of investigation on this issue LKFT teammate Anders Roxell
> > > > > > > git bisected the problem and found bad commit(s) which caused this problem.
> > > > > > >
> > > > > > > The following two patches have been reverted on next-20200519 and retested the
> > > > > > > reproducible steps and confirmed the test case mkfs -t ext4 got PASS.
> > > > > > > ( invoked oom-killer is gone now)
> > > > > > >
> > > > > > > Revert "mm, memcg: avoid stale protection values when cgroup is above
> > > > > > > protection"
> > > > > > >    This reverts commit 23a53e1c02006120f89383270d46cbd040a70bc6.
> > > > > > >
> > > > > > > Revert "mm, memcg: decouple e{low,min} state mutations from protection
> > > > > > > checks"
> > > > > > >    This reverts commit 7b88906ab7399b58bb088c28befe50bcce076d82.
> > > > > >
> > > > > > Thanks Anders and Naresh for tracking this down and reverting.
> > > > > >
> > > > > > I'll take a look tomorrow. I don't see anything immediately obviously wrong
> > > > > > in either of those commits from a (very) cursory glance, but they should
> > > > > > only be taking effect if protections are set.
> > > > >
> > > > > Agreed. If memory.{low,min} is not used then the patch should be
> > > > > effectively a nop. Btw. do you see the problem when booting with
> > > > > cgroup_disable=memory kernel command line parameter?
> > > > 
> > > > With extra kernel command line parameters, cgroup_disable=memory
> > > > I have noticed a differ problem now.
> > > > 
> > > > + mkfs -t ext4 /dev/disk/by-id/ata-TOSHIBA_MG04ACA100N_Y8NRK0BPF6XF
> > > > mke2fs 1.43.8 (1-Jan-2018)
> > > > Creating filesystem with 244190646 4k blocks and 61054976 inodes
> > > > Filesystem UUID: 3bb1a285-2cb4-44b4-b6e8-62548f3ac620
> > > > Superblock backups stored on blocks:
> > > > 32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
> > > > 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968,
> > > > 102400000, 214990848
> > > > Allocating group tables:    0/7453                           done
> > > > Writing inode tables:    0/7453                           done
> > > > Creating journal (262144 blocks): [   35.502102] BUG: kernel NULL
> > > > pointer dereference, address: 000000c8
> > > > [   35.508372] #PF: supervisor read access in kernel mode
> > > > [   35.513506] #PF: error_code(0x0000) - not-present page
> > > > [   35.518638] *pde = 00000000
> > > > [   35.521514] Oops: 0000 [#1] SMP
> > > > [   35.524652] CPU: 0 PID: 145 Comm: kswapd0 Not tainted
> > > > 5.7.0-rc6-next-20200519+ #1
> > > > [   35.532121] Hardware name: Supermicro SYS-5019S-ML/X11SSH-F, BIOS
> > > > 2.2 05/23/2018
> > > > [   35.539507] EIP: mem_cgroup_get_nr_swap_pages+0x28/0x60
> > > 
> > > Could you get faddr2line for this offset?
> > 
> > No need for that, I can help with the "cgroup_disabled=memory" crash:
> > I've been happily running with the fixup below, but haven't got to
> > send it in yet (and wouldn't normally be reading mail at this time!)
> > because of busy chasing a couple of other bugs (not necessarily mm);
> > and maybe the fix would be better with explicit mem_cgroup_disabled()
> > test, or maybe that should be where cgroup_memory_noswap is decided -
> > up to Johannes.
> 
> Thanks Hugh. I can see what is the problem now. I was looking at the
> Linus' tree and we have a different code there
> 
> 	long nr_swap_pages = get_nr_swap_pages();
> 
>         if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
>                 return nr_swap_pages;
> 
> which would be impossible to crash so I was really wondering what is
> going on here. But there are other changes in the mmotm which I haven't
> reviewed yet. Looking at the next tree now it is a fallout from "mm:
> memcontrol: prepare swap controller setup for integration".
> 
> !memcg check slightly more cryptic than an explicit mem_cgroup_disabled
> but I would just leave it to Johannes as well.

Very much appreciate you guys tracking it down so quickly. Sorry about
the breakage.

I think mem_cgroup_disabled() checks are pretty good markers of public
entry points to the memcg API, so I'd prefer that even if a bit more
verbose. What do you think?

---
From cd373ec232942a9bc43ee5e7d2171352019a58fb Mon Sep 17 00:00:00 2001
From: Hugh Dickins <hughd@google.com>
Date: Thu, 21 May 2020 14:58:36 -0400
Subject: [PATCH] mm: memcontrol: prepare swap controller setup for integration
 fix

Fix crash with cgroup_disable=memory:

> > > > + mkfs -t ext4 /dev/disk/by-id/ata-TOSHIBA_MG04ACA100N_Y8NRK0BPF6XF
> > > > mke2fs 1.43.8 (1-Jan-2018)
> > > > Creating filesystem with 244190646 4k blocks and 61054976 inodes
> > > > Filesystem UUID: 3bb1a285-2cb4-44b4-b6e8-62548f3ac620
> > > > Superblock backups stored on blocks:
> > > > 32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
> > > > 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968,
> > > > 102400000, 214990848
> > > > Allocating group tables:    0/7453                           done
> > > > Writing inode tables:    0/7453                           done
> > > > Creating journal (262144 blocks): [   35.502102] BUG: kernel NULL
> > > > pointer dereference, address: 000000c8
> > > > [   35.508372] #PF: supervisor read access in kernel mode
> > > > [   35.513506] #PF: error_code(0x0000) - not-present page
> > > > [   35.518638] *pde = 00000000
> > > > [   35.521514] Oops: 0000 [#1] SMP
> > > > [   35.524652] CPU: 0 PID: 145 Comm: kswapd0 Not tainted
> > > > 5.7.0-rc6-next-20200519+ #1
> > > > [   35.532121] Hardware name: Supermicro SYS-5019S-ML/X11SSH-F, BIOS
> > > > 2.2 05/23/2018
> > > > [   35.539507] EIP: mem_cgroup_get_nr_swap_pages+0x28/0x60

do_memsw_account() used to be automatically false when the cgroup
controller was disabled. Now that it's replaced by
cgroup_memory_noswap, for which this isn't true, make the
mem_cgroup_disabled() checks explicit in the swap control API.

[hannes@cmpxchg.org: use mem_cgroup_disabled() in all API functions]
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Debugged-by: Hugh Dickins <hughd@google.com>
Debugged-by: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
---
 mm/memcontrol.c | 47 +++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 41 insertions(+), 6 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 3e000a316b59..850bca380562 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -6811,6 +6811,9 @@ void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
 	VM_BUG_ON_PAGE(PageLRU(page), page);
 	VM_BUG_ON_PAGE(page_count(page), page);
 
+	if (mem_cgroup_disabled())
+		return;
+
 	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
 		return;
 
@@ -6876,6 +6879,10 @@ int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
 	struct mem_cgroup *memcg;
 	unsigned short oldid;
 
+	if (mem_cgroup_disabled())
+		return 0;
+
+	/* Only cgroup2 has swap.max */
 	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
 		return 0;
 
@@ -6920,6 +6927,9 @@ void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
 	struct mem_cgroup *memcg;
 	unsigned short id;
 
+	if (mem_cgroup_disabled())
+		return;
+
 	id = swap_cgroup_record(entry, 0, nr_pages);
 	rcu_read_lock();
 	memcg = mem_cgroup_from_id(id);
@@ -6940,12 +6950,25 @@ long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
 {
 	long nr_swap_pages = get_nr_swap_pages();
 
-	if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
-		return nr_swap_pages;
+	if (mem_cgroup_disabled())
+		goto out;
+
+	/* Swap control disabled */
+	if (cgroup_memory_noswap)
+		goto out;
+
+	/*
+	 * Only cgroup2 has swap.max, cgroup1 does mem+sw accounting,
+	 * which does not place restrictions specifically on swap.
+	 */
+	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
+		goto out;
+
 	for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
 		nr_swap_pages = min_t(long, nr_swap_pages,
 				      READ_ONCE(memcg->swap.max) -
 				      page_counter_read(&memcg->swap));
+out:
 	return nr_swap_pages;
 }
 
@@ -6957,18 +6980,30 @@ bool mem_cgroup_swap_full(struct page *page)
 
 	if (vm_swap_full())
 		return true;
-	if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
-		return false;
+
+	if (mem_cgroup_disabled())
+		goto out;
+
+	/* Swap control disabled */
+	if (cgroup_memory_noswap)
+		goto out;
+
+	/*
+	 * Only cgroup2 has swap.max, cgroup1 does mem+sw accounting,
+	 * which does not place restrictions specifically on swap.
+	 */
+	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
+		goto out;
 
 	memcg = page->mem_cgroup;
 	if (!memcg)
-		return false;
+		goto out;
 
 	for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
 		if (page_counter_read(&memcg->swap) * 2 >=
 		    READ_ONCE(memcg->swap.max))
 			return true;
-
+out:
 	return false;
 }
 
-- 
2.26.2