target/linux/generic/backport-5.10/633-v6.3-skbuff-Fix-a-race-between-coalescing-and-releasing-S.patch

   1 From: Liang Chen <liangchen.linux@gmail.com>
   2 Date: Thu, 13 Apr 2023 17:03:53 +0800
   3 Subject: [PATCH] skbuff: Fix a race between coalescing and releasing SKBs
   4
   5 Commit 1effe8ca4e34 ("skbuff: fix coalescing for page_pool fragment
   6 recycling") allowed coalescing to proceed with non page pool page and page
   7 pool page when @from is cloned, i.e.
   8
   9 to->pp_recycle    --> false
  10 from->pp_recycle  --> true
  11 skb_cloned(from)  --> true
  12
  13 However, it actually requires skb_cloned(@from) to hold true until
  14 coalescing finishes in this situation. If the other cloned SKB is
  15 released while the merging is in process, from_shinfo->nr_frags will be
  16 set to 0 toward the end of the function, causing the increment of frag
  17 page _refcount to be unexpectedly skipped resulting in inconsistent
  18 reference counts. Later when SKB(@to) is released, it frees the page
  19 directly even though the page pool page is still in use, leading to
  20 use-after-free or double-free errors. So it should be prohibited.
  21
  22 The double-free error message below prompted us to investigate:
  23 BUG: Bad page state in process swapper/1  pfn:0e0d1
  24 page:00000000c6548b28 refcount:-1 mapcount:0 mapping:0000000000000000
  25 index:0x2 pfn:0xe0d1
  26 flags: 0xfffffc0000000(node=0|zone=1|lastcpupid=0x1fffff)
  27 raw: 000fffffc0000000 0000000000000000 ffffffff00000101 0000000000000000
  28 raw: 0000000000000002 0000000000000000 ffffffffffffffff 0000000000000000
  29 page dumped because: nonzero _refcount
  30
  31 CPU: 1 PID: 0 Comm: swapper/1 Tainted: G            E      6.2.0+
  32 Call Trace:
  33  <IRQ>
  34 dump_stack_lvl+0x32/0x50
  35 bad_page+0x69/0xf0
  36 free_pcp_prepare+0x260/0x2f0
  37 free_unref_page+0x20/0x1c0
  38 skb_release_data+0x10b/0x1a0
  39 napi_consume_skb+0x56/0x150
  40 net_rx_action+0xf0/0x350
  41 ? __napi_schedule+0x79/0x90
  42 __do_softirq+0xc8/0x2b1
  43 __irq_exit_rcu+0xb9/0xf0
  44 common_interrupt+0x82/0xa0
  45 </IRQ>
  46 <TASK>
  47 asm_common_interrupt+0x22/0x40
  48 RIP: 0010:default_idle+0xb/0x20
  49
  50 Fixes: 53e0961da1c7 ("page_pool: add frag page recycling support in page pool")
  51 Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
  52 Reviewed-by: Eric Dumazet <edumazet@google.com>
  53 Link: https://lore.kernel.org/r/20230413090353.14448-1-liangchen.linux@gmail.com
  54 Signed-off-by: Jakub Kicinski <kuba@kernel.org>
  55 ---
  56
  57 --- a/net/core/skbuff.c
  58 +++ b/net/core/skbuff.c
  59 @@ -5208,18 +5208,18 @@ bool skb_try_coalesce(struct sk_buff *to
  60         if (skb_cloned(to))
  61                 return false;
  62
  63 -       /* In general, avoid mixing slab allocated and page_pool allocated
  64 -        * pages within the same SKB. However when @to is not pp_recycle and
  65 -        * @from is cloned, we can transition frag pages from page_pool to
  66 -        * reference counted.
  67 -        *
  68 -        * On the other hand, don't allow coalescing two pp_recycle SKBs if
  69 -        * @from is cloned, in case the SKB is using page_pool fragment
  70 +       /* In general, avoid mixing page_pool and non-page_pool allocated
  71 +        * pages within the same SKB. Additionally avoid dealing with clones
  72 +        * with page_pool pages, in case the SKB is using page_pool fragment
  73          * references (PP_FLAG_PAGE_FRAG). Since we only take full page
  74          * references for cloned SKBs at the moment that would result in
  75          * inconsistent reference counts.
  76 +        * In theory we could take full references if @from is cloned and
  77 +        * !@to->pp_recycle but its tricky (due to potential race with
  78 +        * the clone disappearing) and rare, so not worth dealing with.
  79          */
  80 -       if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from)))
  81 +       if (to->pp_recycle != from->pp_recycle ||
  82 +           (from->pp_recycle && skb_cloned(from)))
  83                 return false;
  84
  85         if (len <= skb_tailroom(to)) {