lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:   Wed, 13 May 2020 09:31:41 +0200
From:   David Hildenbrand <david@...hat.com>
To:     Qian Cai <cai@....pw>, Vitaly Wool <vitaly.wool@...sulko.com>
Cc:     Michal Hocko <mhocko@...e.com>, Linux-MM <linux-mm@...ck.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Minchan Kim <minchan@...nel.org>,
        Seth Jennings <sjenning@...hat.com>,
        Dan Streetman <ddstreet@...e.org>
Subject: Re: zswap z3fold + memory offline = infinite loop

On 13.05.20 02:36, Qian Cai wrote:
> Put zswap z3fold pages into the memory and then offline those memory would trigger an infinite loop here in
> 
> __offline_pages() --> do_migrate_range() because there is no error handling,

That's usually not the case, that's why we - nowadays - basically never
see endless loops anymore, except when some code is doing something
wrong with movable pages. The TODO below is somewhat misleading.

> 
> 			if (pfn) {
> 				/*
> 				 * TODO: fatal migration failures should bail
> 				 * out
> 				 */
> 				do_migrate_range(pfn, end_pfn);
> 
> There, isolate_movable_page() will always return -EBUSY  because,
> 
> 	if (!mapping->a_ops->isolate_page(page, mode))
> 		goto out_no_isolated;
> 
> i.e., z3fold_page_isolate() will always return false because,
> 
> zhdr->mapped_count == 2

Whenever we have such endless loops when offlining it's either because
1. Somebody puts unmovable data into the movable zone
2. Somebody marks unmovable data as movable
3. Somebody turns movable data into unmovable data although the
pageblock is isolated

After start_isolate_page_range() we are sure that we only have movable
data in the isolated page range. The pageblocks are isolated, so e.g.,
no new allocations can end up on that memory.

Offlining code can therefore assume that all memory is in fact movable
and can be moved eventually. It might not always work on the first
attempt, that's why we retry.

We don't expect permanent migration failure, because that would mean the
page is unmovable.

I can see that mm/z3fold.c uses  __SetPageMovable(). So the question is,
why are these pages permanently unmovable.


> 
> It should be easy to reproduce. Otherwise, use this one,
> 
> https://github.com/cailca/linux-mm/blob/master/random.c
> 
> and then watch the console burning with those,
> 
> [12661.793667][T566417] failed to isolate pfn 1045b2
> [12661.793745][T566417] page:c00c000004116c80 refcount:2 mapcount:0 mapping:00000000999f9672 index:0x0
> [12661.793865][T566417] mapping->a_ops:z3fold_aops
> [12661.793919][T566417] flags: 0x3fffc000000000()
> [12661.793969][T566417] raw: 003fffc000000000 c00c000003cef388 c00c000006b0da08 c000001275b87f6a
> [12661.794071][T566417] raw: 0000000000000000 0000000000000000 00000002ffffffff 0000000000000000
> [12661.794158][T566417] page dumped because: isolation failed
> [12661.794226][T566417] page_owner tracks the page as allocated
> [12661.794292][T566417] page last allocated via order 0, migratetype Unmovable, gfp_mask 0x12800(GFP_NOWAIT|__GFP_NOWARN|__GFP_NORETRY)
> [12661.794463][T566417]  prep_new_page+0x3d0/0x450
> [12661.794508][T566417]  get_page_from_freelist+0x1bb8/0x27c0
> [12661.794575][T566417]  __alloc_pages_slowpath.constprop.60+0x240/0x15a0
> [12661.794654][T566417]  __alloc_pages_nodemask+0x520/0x650
> [12661.794715][T566417]  alloc_pages_current+0xbc/0x140
> [12661.794772][T566417]  z3fold_zpool_malloc+0x6cc/0xe20
> [12661.794826][T566417]  zpool_malloc+0x34/0x50
> [12661.794888][T566417]  zswap_frontswap_store+0x60c/0xe20
> [12661.794942][T566417]  __frontswap_store+0x128/0x330
> [12661.794995][T566417]  swap_writepage+0x58/0x110
> [12661.795048][T566417]  pageout+0x16c/0xa40
> [12661.795092][T566417]  shrink_page_list+0x1ab4/0x2490
> [12661.795155][T566417]  shrink_inactive_list+0x25c/0x710
> [12661.795206][T566417]  shrink_lruvec+0x444/0x1260
> [12661.795274][T566417]  shrink_node+0x288/0x9a0
> [12661.795330][T566417]  do_try_to_free_pages+0x158/0x640
> [12661.795383][T566417] page last free stack trace:
> [12661.795437][T566417]  free_pcp_prepare+0x52c/0x590
> [12661.795493][T566417]  free_unref_page+0x38/0xf0
> [12662.156109][T566417]  free_z3fold_page+0x58/0x120
> [12662.156131][T566417]  free_pages_work+0x148/0x1c0
> [12662.156195][T566417]  process_one_work+0x310/0x900
> [12662.156257][T566417]  worker_thread+0x78/0x530
> [12662.156306][T566417]  kthread+0x1c4/0x1d0
> [12662.156354][T566417]  ret_from_kernel_thread+0x5c/0x74
> 


-- 
Thanks,

David / dhildenb

Powered by blists - more mailing lists