kernel: backport fix for a page pool related race condition
authorFelix Fietkau <nbd@nbd.name>
Thu, 13 Apr 2023 19:28:07 +0000 (21:28 +0200)
committerFelix Fietkau <nbd@nbd.name>
Fri, 14 Apr 2023 08:27:09 +0000 (10:27 +0200)
Signed-off-by: Felix Fietkau <nbd@nbd.name>
target/linux/generic/backport-5.10/633-v6.3-skbuff-Fix-a-race-between-coalescing-and-releasing-S.patch [new file with mode: 0644]
target/linux/generic/backport-5.15/612-v6.3-skbuff-Fix-a-race-between-coalescing-and-releasing-S.patch [new file with mode: 0644]

diff --git a/target/linux/generic/backport-5.10/633-v6.3-skbuff-Fix-a-race-between-coalescing-and-releasing-S.patch b/target/linux/generic/backport-5.10/633-v6.3-skbuff-Fix-a-race-between-coalescing-and-releasing-S.patch
new file mode 100644 (file)
index 0000000..2ad315d
--- /dev/null
@@ -0,0 +1,85 @@
+From: Liang Chen <liangchen.linux@gmail.com>
+Date: Thu, 13 Apr 2023 17:03:53 +0800
+Subject: [PATCH] skbuff: Fix a race between coalescing and releasing SKBs
+
+Commit 1effe8ca4e34 ("skbuff: fix coalescing for page_pool fragment
+recycling") allowed coalescing to proceed with non page pool page and page
+pool page when @from is cloned, i.e.
+
+to->pp_recycle    --> false
+from->pp_recycle  --> true
+skb_cloned(from)  --> true
+
+However, it actually requires skb_cloned(@from) to hold true until
+coalescing finishes in this situation. If the other cloned SKB is
+released while the merging is in process, from_shinfo->nr_frags will be
+set to 0 toward the end of the function, causing the increment of frag
+page _refcount to be unexpectedly skipped resulting in inconsistent
+reference counts. Later when SKB(@to) is released, it frees the page
+directly even though the page pool page is still in use, leading to
+use-after-free or double-free errors. So it should be prohibited.
+
+The double-free error message below prompted us to investigate:
+BUG: Bad page state in process swapper/1  pfn:0e0d1
+page:00000000c6548b28 refcount:-1 mapcount:0 mapping:0000000000000000
+index:0x2 pfn:0xe0d1
+flags: 0xfffffc0000000(node=0|zone=1|lastcpupid=0x1fffff)
+raw: 000fffffc0000000 0000000000000000 ffffffff00000101 0000000000000000
+raw: 0000000000000002 0000000000000000 ffffffffffffffff 0000000000000000
+page dumped because: nonzero _refcount
+
+CPU: 1 PID: 0 Comm: swapper/1 Tainted: G            E      6.2.0+
+Call Trace:
+ <IRQ>
+dump_stack_lvl+0x32/0x50
+bad_page+0x69/0xf0
+free_pcp_prepare+0x260/0x2f0
+free_unref_page+0x20/0x1c0
+skb_release_data+0x10b/0x1a0
+napi_consume_skb+0x56/0x150
+net_rx_action+0xf0/0x350
+? __napi_schedule+0x79/0x90
+__do_softirq+0xc8/0x2b1
+__irq_exit_rcu+0xb9/0xf0
+common_interrupt+0x82/0xa0
+</IRQ>
+<TASK>
+asm_common_interrupt+0x22/0x40
+RIP: 0010:default_idle+0xb/0x20
+
+Fixes: 53e0961da1c7 ("page_pool: add frag page recycling support in page pool")
+Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
+Reviewed-by: Eric Dumazet <edumazet@google.com>
+Link: https://lore.kernel.org/r/20230413090353.14448-1-liangchen.linux@gmail.com
+Signed-off-by: Jakub Kicinski <kuba@kernel.org>
+---
+
+--- a/net/core/skbuff.c
++++ b/net/core/skbuff.c
+@@ -5208,18 +5208,18 @@ bool skb_try_coalesce(struct sk_buff *to
+       if (skb_cloned(to))
+               return false;
+-      /* In general, avoid mixing slab allocated and page_pool allocated
+-       * pages within the same SKB. However when @to is not pp_recycle and
+-       * @from is cloned, we can transition frag pages from page_pool to
+-       * reference counted.
+-       *
+-       * On the other hand, don't allow coalescing two pp_recycle SKBs if
+-       * @from is cloned, in case the SKB is using page_pool fragment
++      /* In general, avoid mixing page_pool and non-page_pool allocated
++       * pages within the same SKB. Additionally avoid dealing with clones
++       * with page_pool pages, in case the SKB is using page_pool fragment
+        * references (PP_FLAG_PAGE_FRAG). Since we only take full page
+        * references for cloned SKBs at the moment that would result in
+        * inconsistent reference counts.
++       * In theory we could take full references if @from is cloned and
++       * !@to->pp_recycle but its tricky (due to potential race with
++       * the clone disappearing) and rare, so not worth dealing with.
+        */
+-      if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from)))
++      if (to->pp_recycle != from->pp_recycle ||
++          (from->pp_recycle && skb_cloned(from)))
+               return false;
+       if (len <= skb_tailroom(to)) {
diff --git a/target/linux/generic/backport-5.15/612-v6.3-skbuff-Fix-a-race-between-coalescing-and-releasing-S.patch b/target/linux/generic/backport-5.15/612-v6.3-skbuff-Fix-a-race-between-coalescing-and-releasing-S.patch
new file mode 100644 (file)
index 0000000..7f1b1ee
--- /dev/null
@@ -0,0 +1,85 @@
+From: Liang Chen <liangchen.linux@gmail.com>
+Date: Thu, 13 Apr 2023 17:03:53 +0800
+Subject: [PATCH] skbuff: Fix a race between coalescing and releasing SKBs
+
+Commit 1effe8ca4e34 ("skbuff: fix coalescing for page_pool fragment
+recycling") allowed coalescing to proceed with non page pool page and page
+pool page when @from is cloned, i.e.
+
+to->pp_recycle    --> false
+from->pp_recycle  --> true
+skb_cloned(from)  --> true
+
+However, it actually requires skb_cloned(@from) to hold true until
+coalescing finishes in this situation. If the other cloned SKB is
+released while the merging is in process, from_shinfo->nr_frags will be
+set to 0 toward the end of the function, causing the increment of frag
+page _refcount to be unexpectedly skipped resulting in inconsistent
+reference counts. Later when SKB(@to) is released, it frees the page
+directly even though the page pool page is still in use, leading to
+use-after-free or double-free errors. So it should be prohibited.
+
+The double-free error message below prompted us to investigate:
+BUG: Bad page state in process swapper/1  pfn:0e0d1
+page:00000000c6548b28 refcount:-1 mapcount:0 mapping:0000000000000000
+index:0x2 pfn:0xe0d1
+flags: 0xfffffc0000000(node=0|zone=1|lastcpupid=0x1fffff)
+raw: 000fffffc0000000 0000000000000000 ffffffff00000101 0000000000000000
+raw: 0000000000000002 0000000000000000 ffffffffffffffff 0000000000000000
+page dumped because: nonzero _refcount
+
+CPU: 1 PID: 0 Comm: swapper/1 Tainted: G            E      6.2.0+
+Call Trace:
+ <IRQ>
+dump_stack_lvl+0x32/0x50
+bad_page+0x69/0xf0
+free_pcp_prepare+0x260/0x2f0
+free_unref_page+0x20/0x1c0
+skb_release_data+0x10b/0x1a0
+napi_consume_skb+0x56/0x150
+net_rx_action+0xf0/0x350
+? __napi_schedule+0x79/0x90
+__do_softirq+0xc8/0x2b1
+__irq_exit_rcu+0xb9/0xf0
+common_interrupt+0x82/0xa0
+</IRQ>
+<TASK>
+asm_common_interrupt+0x22/0x40
+RIP: 0010:default_idle+0xb/0x20
+
+Fixes: 53e0961da1c7 ("page_pool: add frag page recycling support in page pool")
+Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
+Reviewed-by: Eric Dumazet <edumazet@google.com>
+Link: https://lore.kernel.org/r/20230413090353.14448-1-liangchen.linux@gmail.com
+Signed-off-by: Jakub Kicinski <kuba@kernel.org>
+---
+
+--- a/net/core/skbuff.c
++++ b/net/core/skbuff.c
+@@ -5397,18 +5397,18 @@ bool skb_try_coalesce(struct sk_buff *to
+       if (skb_cloned(to))
+               return false;
+-      /* In general, avoid mixing slab allocated and page_pool allocated
+-       * pages within the same SKB. However when @to is not pp_recycle and
+-       * @from is cloned, we can transition frag pages from page_pool to
+-       * reference counted.
+-       *
+-       * On the other hand, don't allow coalescing two pp_recycle SKBs if
+-       * @from is cloned, in case the SKB is using page_pool fragment
++      /* In general, avoid mixing page_pool and non-page_pool allocated
++       * pages within the same SKB. Additionally avoid dealing with clones
++       * with page_pool pages, in case the SKB is using page_pool fragment
+        * references (PP_FLAG_PAGE_FRAG). Since we only take full page
+        * references for cloned SKBs at the moment that would result in
+        * inconsistent reference counts.
++       * In theory we could take full references if @from is cloned and
++       * !@to->pp_recycle but its tricky (due to potential race with
++       * the clone disappearing) and rare, so not worth dealing with.
+        */
+-      if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from)))
++      if (to->pp_recycle != from->pp_recycle ||
++          (from->pp_recycle && skb_cloned(from)))
+               return false;
+       if (len <= skb_tailroom(to)) {