skbuff: skb_segment, Update nfrags after calling zero copy functions
Subsystems (4)
PATCH 1/1 skbuff: skb_segment, Update nfrags after calling zero copy functions
We have seen kernel panic with stacktrace below. This is 5.15.123 LTS kernel running a qemu VM with virtio network interface and vhost=on. When enabling packet corruption, with command below, we see the kernel panic.
tc qdisc add dev eth2 root netem corrupt 0.7065335155074846%
[ 193.894380] BUG: kernel NULL pointer dereference, address: 00000000000000bc [ 193.894635] #PF: supervisor read access in kernel mode [ 193.894828] #PF: error_code(0x0000) - not-present page [ 193.895027] PGD 0 P4D 0 [ 193.895140] Oops: 0000 [#1] SMP [ 193.895273] CPU: 13 PID: 18164 Comm: vh-net-17428 Kdump: loaded Tainted: G O 5.15.123+ #26 [ 193.895602] Hardware name: [ 193.903919] RIP: 0010:skb_segment+0xb0e/0x12f0 [ 193.908176] Code: 45 a8 50 e8 54 46 be ff 44 8b 5d 80 41 01 c7 48 83 c4 18 44 89 7d b4 44 3b 5d b0 0f 8c f0 00 00 00 4d 85 e4 0f 84 65 07 00 00 <45> 8b b4 24 bc 00 00 00 48 8b b5 70 ff ff ff 4d 03 b4 24 c0 00 00 [ 193.921099] RSP: 0018:ffffc9002bf2f770 EFLAGS: 00010282 [ 193.925552] RAX: 00000000000000ee RBX: ffff88baab5092c8 RCX: 0000000000000003 [ 193.934308] RDX: 0000000000000000 RSI: 00000000fff7ffff RDI: 0000000000000001 [ 193.943281] RBP: ffffc9002bf2f850 R08: 0000000000000000 R09: c0000000fff7ffff [ 193.952658] R10: 0000000000000029 R11: ffffc9002bf2f310 R12: 0000000000000000 [ 193.962423] R13: ffff88abc2291e00 R14: ffff88abc2291d00 R15: ffff88abc2290600 [ 193.972593] FS: 0000000000000000(0000) GS:ffff88c07f840000(0000) knlGS:0000000000000000 [ 193.983302] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 193.988891] CR2: 00000000000000bc CR3: 0000001dccb4b006 CR4: 00000000001726e0 [ 193.999925] MSR 198h IA32 perf status 0x00001ba000001a00 [ 194.005488] MSR 19Ch IA32 thermal status 0x0000000088370000 [ 194.010983] MSR 1B1h IA32 package thermal status 0x0000000088330000 [ 194.021892] Call Trace: [ 194.027422] <TASK> [ 194.032838] ? __die_body+0x1a/0x60 [ 194.038172] ? page_fault_oops+0x12d/0x4d0 [ 194.043395] ? skb_segment+0xb0e/0x12f0 [ 194.048501] ? search_bpf_extables+0x59/0x60 [ 194.053547] ? fixup_exception+0x1d/0x250 [ 194.058537] ? exc_page_fault+0x67/0x140 [ 194.063382] ? asm_exc_page_fault+0x1f/0x30 [ 194.068171] ? skb_segment+0xb0e/0x12f0 [ 194.072861] tcp_gso_segment+0x107/0x540 [ 194.077507] ? sk_common_release+0xe0/0xe0 [ 194.082031] inet_gso_segment+0x15c/0x3d0 [ 194.086441] ? __skb_get_hash_symmetric+0x190/0x190 [ 194.090783] skb_mac_gso_segment+0x9f/0x110 [ 194.095016] __skb_gso_segment+0xc1/0x190 [ 194.099124] ? netif_skb_features+0xb5/0x280 [ 194.103131] netem_enqueue+0x290/0xb10 [sch_netem] [ 194.107071] dev_qdisc_enqueue+0x16/0x70 [ 194.110884] __dev_queue_xmit+0x63b/0xb30 [ 194.114569] ? inet_gso_segment+0x15c/0x3d0 [ 194.118160] ? bond_start_xmit+0x159/0x380 [bonding] [ 194.121670] bond_start_xmit+0x159/0x380 [bonding] [ 194.125101] ? skb_mac_gso_segment+0xa7/0x110 [ 194.128506] dev_hard_start_xmit+0xc3/0x1e0 [ 194.131787] __dev_queue_xmit+0x8a0/0xb30 [ 194.134977] ? macvlan_start_xmit+0x4f/0x100 [macvlan] [ 194.138225] macvlan_start_xmit+0x4f/0x100 [macvlan] [ 194.141477] dev_hard_start_xmit+0xc3/0x1e0 [ 194.144622] sch_direct_xmit+0xe3/0x280 [ 194.147748] __dev_queue_xmit+0x54a/0xb30 [ 194.150924] ? tap_get_user+0x2a8/0x9c0 [tap] [ 194.154131] tap_get_user+0x2a8/0x9c0 [tap] [ 194.157358] tap_sendmsg+0x52/0x8e0 [tap] [ 194.160565] ? get_tx_bufs+0x42/0x1d0 [vhost_net] [ 194.163815] ? get_tx_bufs+0x16a/0x1d0 [vhost_net] [ 194.167049] handle_tx_zerocopy+0x14e/0x4c0 [vhost_net] [ 194.170351] ? add_range+0x11/0x30 [ 194.173631] handle_tx+0xcd/0xe0 [vhost_net] [ 194.176959] vhost_worker+0x76/0xb0 [vhost] [ 194.180299] ? vhost_flush_work+0x10/0x10 [vhost] [ 194.183667] kthread+0x118/0x140 [ 194.187007] ? set_kthread_struct+0x40/0x40 [ 194.190358] ret_from_fork+0x1f/0x30 [ 194.193670] </TASK>
I have narrowed the issue down to these lines in skb_segment()
4247 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4248 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4249 goto err;
It is possible for skb_orphan_frags() or skb_zerocopy_clone() to update nr_frags as both functions may call skb_copy_ubufs(). If nr_frags gets updated, the local copy in nfrags might end up stale and cause this panic. In particular it is possible the while loop below hits i >= nrfrags prematurely and tries to move to next frag_skb by using list_skb. If list_skb is NULL then we hit the panic above.
The naive way to fix this is to update nfrags as shown in this patch. This way we get the correct number of fragments and while loop can find all fragments without needing to move to next skbuff.
I wanted to share this with the list and see what people think before submitting a patch. Specially that I have not tested this on master branch.
--- net/core/skbuff.c | 4 ++++ 1 file changed, 4 insertions(+)
net/core/skbuff.c +4 -0
@@ -4567,6 +4567,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) ||skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC )) goto err; +/* Update nfrags in case skb_copy_ubufs() updates nr_frags */ + nfrags =skb_shinfo (frag_skb)->nr_frags;while (pos < offset + len) {if (i >= nfrags) {
@@ -4587,6 +4589,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) goto err; +/* Update nfrags in case skb_copy_ubufs() updates nr_frags */ + nfrags =skb_shinfo (frag_skb)->nr_frags; list_skb = list_skb->next; }
Signature · 1 line
2.17.1
Quoted 115 lines · Mohamed Khalfella
We have seen kernel panic with stacktrace below. This is 5.15.123 LTS kernel running a qemu VM with virtio network interface and vhost=on. When enabling packet corruption, with command below, we see the kernel panic.
tc qdisc add dev eth2 root netem corrupt 0.7065335155074846%
[ 193.894380] BUG: kernel NULL pointer dereference, address: 00000000000000bc [ 193.894635] #PF: supervisor read access in kernel mode [ 193.894828] #PF: error_code(0x0000) - not-present page [ 193.895027] PGD 0 P4D 0 [ 193.895140] Oops: 0000 [#1] SMP [ 193.895273] CPU: 13 PID: 18164 Comm: vh-net-17428 Kdump: loaded Tainted: G O 5.15.123+ #26 [ 193.895602] Hardware name: [ 193.903919] RIP: 0010:skb_segment+0xb0e/0x12f0 [ 193.908176] Code: 45 a8 50 e8 54 46 be ff 44 8b 5d 80 41 01 c7 48 83 c4 18 44 89 7d b4 44 3b 5d b0 0f 8c f0 00 00 00 4d 85 e4 0f 84 65 07 00 00 <45> 8b b4 24 bc 00 00 00 48 8b b5 70 ff ff ff 4d 03 b4 24 c0 00 00 [ 193.921099] RSP: 0018:ffffc9002bf2f770 EFLAGS: 00010282 [ 193.925552] RAX: 00000000000000ee RBX: ffff88baab5092c8 RCX: 0000000000000003 [ 193.934308] RDX: 0000000000000000 RSI: 00000000fff7ffff RDI: 0000000000000001 [ 193.943281] RBP: ffffc9002bf2f850 R08: 0000000000000000 R09: c0000000fff7ffff [ 193.952658] R10: 0000000000000029 R11: ffffc9002bf2f310 R12: 0000000000000000 [ 193.962423] R13: ffff88abc2291e00 R14: ffff88abc2291d00 R15: ffff88abc2290600 [ 193.972593] FS: 0000000000000000(0000) GS:ffff88c07f840000(0000) knlGS:0000000000000000 [ 193.983302] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 193.988891] CR2: 00000000000000bc CR3: 0000001dccb4b006 CR4: 00000000001726e0 [ 193.999925] MSR 198h IA32 perf status 0x00001ba000001a00 [ 194.005488] MSR 19Ch IA32 thermal status 0x0000000088370000 [ 194.010983] MSR 1B1h IA32 package thermal status 0x0000000088330000 [ 194.021892] Call Trace: [ 194.027422] <TASK> [ 194.032838] ? __die_body+0x1a/0x60 [ 194.038172] ? page_fault_oops+0x12d/0x4d0 [ 194.043395] ? skb_segment+0xb0e/0x12f0 [ 194.048501] ? search_bpf_extables+0x59/0x60 [ 194.053547] ? fixup_exception+0x1d/0x250 [ 194.058537] ? exc_page_fault+0x67/0x140 [ 194.063382] ? asm_exc_page_fault+0x1f/0x30 [ 194.068171] ? skb_segment+0xb0e/0x12f0 [ 194.072861] tcp_gso_segment+0x107/0x540 [ 194.077507] ? sk_common_release+0xe0/0xe0 [ 194.082031] inet_gso_segment+0x15c/0x3d0 [ 194.086441] ? __skb_get_hash_symmetric+0x190/0x190 [ 194.090783] skb_mac_gso_segment+0x9f/0x110 [ 194.095016] __skb_gso_segment+0xc1/0x190 [ 194.099124] ? netif_skb_features+0xb5/0x280 [ 194.103131] netem_enqueue+0x290/0xb10 [sch_netem] [ 194.107071] dev_qdisc_enqueue+0x16/0x70 [ 194.110884] __dev_queue_xmit+0x63b/0xb30 [ 194.114569] ? inet_gso_segment+0x15c/0x3d0 [ 194.118160] ? bond_start_xmit+0x159/0x380 [bonding] [ 194.121670] bond_start_xmit+0x159/0x380 [bonding] [ 194.125101] ? skb_mac_gso_segment+0xa7/0x110 [ 194.128506] dev_hard_start_xmit+0xc3/0x1e0 [ 194.131787] __dev_queue_xmit+0x8a0/0xb30 [ 194.134977] ? macvlan_start_xmit+0x4f/0x100 [macvlan] [ 194.138225] macvlan_start_xmit+0x4f/0x100 [macvlan] [ 194.141477] dev_hard_start_xmit+0xc3/0x1e0 [ 194.144622] sch_direct_xmit+0xe3/0x280 [ 194.147748] __dev_queue_xmit+0x54a/0xb30 [ 194.150924] ? tap_get_user+0x2a8/0x9c0 [tap] [ 194.154131] tap_get_user+0x2a8/0x9c0 [tap] [ 194.157358] tap_sendmsg+0x52/0x8e0 [tap] [ 194.160565] ? get_tx_bufs+0x42/0x1d0 [vhost_net] [ 194.163815] ? get_tx_bufs+0x16a/0x1d0 [vhost_net] [ 194.167049] handle_tx_zerocopy+0x14e/0x4c0 [vhost_net] [ 194.170351] ? add_range+0x11/0x30 [ 194.173631] handle_tx+0xcd/0xe0 [vhost_net] [ 194.176959] vhost_worker+0x76/0xb0 [vhost] [ 194.180299] ? vhost_flush_work+0x10/0x10 [vhost] [ 194.183667] kthread+0x118/0x140 [ 194.187007] ? set_kthread_struct+0x40/0x40 [ 194.190358] ret_from_fork+0x1f/0x30 [ 194.193670] </TASK>I have narrowed the issue down to these lines in
skb_segment()4247 if (
skb_orphan_frags(frag_skb, GFP_ATOMIC)||
4248skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4249 goto err;It is possible for
skb_orphan_frags()orskb_zerocopy_clone()to updatenr_fragsas both functions may callskb_copy_ubufs(). Ifnr_fragsgets updated, the local copy innfragsmight end up stale and cause this panic. In particular it is possible the while loop below hitsi >= nrfragsprematurely and tries to move to nextfrag_skbby usinglist_skb. Iflist_skbis NULL then we hit the panic above.The naive way to fix this is to update
nfragsas shown in this patch. This way we get the correct number of fragments and while loop can find all fragments without needing to move to next skbuff.I wanted to share this with the list and see what people think before submitting a patch. Specially that I have not tested this on master branch.
Fixes bf5c25d60861 ("skbuff: in skb_segment, call zerocopy functions once per nskb")Signed-off-by Mohamed Khalfella <[email protected]>--- net/core/skbuff.c | 4 ++++ 1 file changed, 4 insertions(+)net/core/skbuff.c +2
-0@@ -4567,6 +4567,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) ||skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC )) goto err; +/* Update nfrags in case skb_copy_ubufs() updates nr_frags */ + nfrags =skb_shinfo (frag_skb)->nr_frags;while (pos < offset + len) {if (i >= nfrags) {
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4596
Quoted 8 lines
net/core/skbuff.c +2
-0@@ -4587,6 +4589,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) goto err; +/* Update nfrags in case skb_copy_ubufs() updates nr_frags */ + nfrags =skb_shinfo (frag_skb)->nr_frags; list_skb = list_skb->next; }
In principle this makes sense to me. skb_copy_ubufs will create new frags, which do not map 1:1 on to the original frags. It originally allocated the number of order 0 pages required to store __skb_pagelen(skb). Recently it switched to using high order pages.
The two calls to skb_orphan_frags and skb_zerocopy clone are called for each new segment skb. But only on skb_orphan_frags will it possibly call skb_copy_ubufs, an then always for the first call of the input skb (head_skb or a list_skb). From commit "skbuff: in skb_segment, call zerocopy functions once per nskb":
"When segmenting skbs with user frags, either the user frags must be
replaced with private copies and uarg released, or the uarg must have its refcount increased for each new skb.
skb_orphan_frags does the first, except for cases that can handle reference counting. skb_zerocopy_clone then does the second."
skb_copy_ubufs only called on the first check for an input simplifies. A possible realloc in the middle of processing a series of frags would be much harder.
Small point: nfrags is not the only state that needs to be refreshed after a fags realloc, also frag.
Thanks for the report. I'm traveling likely without internet until the weekend. Apologies if it takes a while for me to follow up.
Quoted 2 lines · willemjdebruijn
Small point: nfrags is not the only state that needs to be refreshed after a fags realloc, also frag.
I am new to this code. Can you help me understand why frag needs to be updated too? My reading of this code is that frag points to frags array in shared info. As long as shared info pointer remain the same frag pointer should remain valid.
Am I missing something?
Quoted 3 lines
Thanks for the report. I'm traveling likely without internet until the weekend. Apologies if it takes a while for me to follow up.
No problem. Thanks for the quick response!
Quoted 10 lines · Mohamed Khalfella
Quoted 2 lines · willemjdebruijn
Small point: nfrags is not the only state that needs to be refreshed after a fags realloc, also frag.
I am new to this code. Can you help me understand why frag needs to be updated too? My reading of this code is that frag points to frags array in shared info. As long as shared info pointer remain the same frag pointer should remain valid.
skb_copy_ubufs() could actually call skb_unclone() and thus skb->head could be re-allocated.
I guess that if you run your patch (and a repro of the bug ?) with KASAN enabled kernel, you should see a possible use-after-free ?
To force the skb_unclone() path, having a tcpdump catching all packets would be enough I think.
Quoted 5 lines
Am I missing something?
Quoted 3 lines
Thanks for the report. I'm traveling likely without internet until the weekend. Apologies if it takes a while for me to follow up.
No problem. Thanks for the quick response!
Quoted 13 lines · Eric Dumazet
Quoted 10 lines · Mohamed Khalfella
Quoted 2 lines · willemjdebruijn
Small point: nfrags is not the only state that needs to be refreshed after a fags realloc, also frag.
I am new to this code. Can you help me understand why frag needs to be updated too? My reading of this code is that frag points to frags array in shared info. As long as shared info pointer remain the same frag pointer should remain valid.
Quoted 9 lines
skb_copy_ubufs()could actually callskb_unclone()and thusskb->headcould be re-allocated.I guess that if you run your patch (and a repro of the bug ?) with KASAN enabled kernel, you should see a possible use-after-free ?
To force the
skb_unclone()path, having a tcpdump catching all packets would be enough I think.
Okay, I see it now. I have not tested this patch with tcpdump capturing packets at the same time. Also, during my testing I have not seen the value of skb->head changnig. Now you are mentioning it it, I will make sure to test with tcpdump running and see skb->head changing. Thank you for pointing that out.
For frag, I guess something like frag = &skb_shinfo(list_skb)->frags[i]; should do the job. I have not tested it though. I will need to do more testing before posting updated patch.
Quoted 31 lines · Mohamed Khalfella
Quoted 13 lines · Eric Dumazet
Quoted 10 lines · Mohamed Khalfella
Quoted 2 lines · willemjdebruijn
Small point: nfrags is not the only state that needs to be refreshed after a fags realloc, also frag.
I am new to this code. Can you help me understand why frag needs to be updated too? My reading of this code is that frag points to frags array in shared info. As long as shared info pointer remain the same frag pointer should remain valid.
Quoted 9 lines
skb_copy_ubufs()could actually callskb_unclone()and thusskb->headcould be re-allocated.I guess that if you run your patch (and a repro of the bug ?) with KASAN enabled kernel, you should see a possible use-after-free ?
To force the
skb_unclone()path, having a tcpdump catching all packets would be enough I think.Okay, I see it now. I have not tested this patch with tcpdump capturing packets at the same time. Also, during my testing I have not seen the value of
skb->headchangnig. Now you are mentioning it it, I will make sure to test with tcpdump running and seeskb->headchanging. Thank you for pointing that out.
Quoted 3 lines
For frag, I guess something like frag = &skb_shinfo(list_skb)->frags[i]; should do the job. I have not tested it though. I will need to do more testing before posting updated patch.
Another way to test this path for certain (without tcpdump having to race) is to add a temporary/debug patch like this one:
net/core/skbuff.c +0 -0
100644
net/core/skbuff.c +2 -1
@@ -1749,7 +1749,8 @@ int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)int i, order, psize, new_frags;u32 d_off; -if (skb_shared (skb) ||skb_unclone (skb, gfp_mask)) +if (skb_shared (skb) || +pskb_expand_head (skb,0 ,0 , gfp_mask))return -EINVAL ;if (!num_frags)
Note that this might catch other bugs :/
Quoted 19 lines · Eric Dumazet
Another way to test this path for certain (without tcpdump having to race) is to add a temporary/debug patch like this one:
net/core/skbuff.c +0
-0100644
net/core/skbuff.c +2
-1@@ -1749,7 +1749,8 @@ int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)int i, order, psize, new_frags;u32 d_off; -if (skb_shared (skb) ||skb_unclone (skb, gfp_mask)) +if (skb_shared (skb) || +pskb_expand_head (skb,0 ,0 , gfp_mask))return -EINVAL ;if (!num_frags)
Quoted 1 line
Note that this might catch other bugs :/
I was not able to make it allocate a new frags by running tcpdump while reproing the problem. However, I was able to do it with your patch.
Quoted 19 lines · Mohamed Khalfella
Quoted 17 lines · Eric Dumazet
Another way to test this path for certain (without tcpdump having to race) is to add a temporary/debug patch like this one:
net/core/skbuff.c +0
-0100644
net/core/skbuff.c +2
-1@@ -1749,7 +1749,8 @@ int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)int i, order, psize, new_frags;u32 d_off; -if (skb_shared (skb) ||skb_unclone (skb, gfp_mask)) +if (skb_shared (skb) || +pskb_expand_head (skb,0 ,0 , gfp_mask))return -EINVAL ;
Quoted 6 lines
Quoted 3 lines
net/core/skbuff.c +0
-0if (!num_frags)Note that this might catch other bugs :/
I was not able to make it allocate a new frags by running tcpdump while reproing the problem. However, I was able to do it with your patch.
I am glad this worked, and looking forward to a v2 of your patch, thanks !
PATCH v2 1/1 skbuff: skb_segment, Call zero copy functions before using skbuff frags
In reply to willemjdebruijnCommit bf5c25d60861 ("skbuff: in skb_segment, call zerocopy functions once per nskb") added the call to zero copy functions in skb_segment(). The change introduced a bug in skb_segment() because skb_orphan_frags() may possibly change the number of fragments or allocate new fragments altogether leaving nrfrags and frag to point to the old values. This can cause a panic with stacktrace like the one below.
[ 193.894380] BUG: kernel NULL pointer dereference, address: 00000000000000bc [ 193.895273] CPU: 13 PID: 18164 Comm: vh-net-17428 Kdump: loaded Tainted: G O 5.15.123+ #26 [ 193.903919] RIP: 0010:skb_segment+0xb0e/0x12f0 [ 194.021892] Call Trace: [ 194.027422] <TASK> [ 194.072861] tcp_gso_segment+0x107/0x540 [ 194.082031] inet_gso_segment+0x15c/0x3d0 [ 194.090783] skb_mac_gso_segment+0x9f/0x110 [ 194.095016] __skb_gso_segment+0xc1/0x190 [ 194.103131] netem_enqueue+0x290/0xb10 [sch_netem] [ 194.107071] dev_qdisc_enqueue+0x16/0x70 [ 194.110884] __dev_queue_xmit+0x63b/0xb30 [ 194.121670] bond_start_xmit+0x159/0x380 [bonding] [ 194.128506] dev_hard_start_xmit+0xc3/0x1e0 [ 194.131787] __dev_queue_xmit+0x8a0/0xb30 [ 194.138225] macvlan_start_xmit+0x4f/0x100 [macvlan] [ 194.141477] dev_hard_start_xmit+0xc3/0x1e0 [ 194.144622] sch_direct_xmit+0xe3/0x280 [ 194.147748] __dev_queue_xmit+0x54a/0xb30 [ 194.154131] tap_get_user+0x2a8/0x9c0 [tap] [ 194.157358] tap_sendmsg+0x52/0x8e0 [tap] [ 194.167049] handle_tx_zerocopy+0x14e/0x4c0 [vhost_net] [ 194.173631] handle_tx+0xcd/0xe0 [vhost_net] [ 194.176959] vhost_worker+0x76/0xb0 [vhost] [ 194.183667] kthread+0x118/0x140 [ 194.190358] ret_from_fork+0x1f/0x30 [ 194.193670] </TASK>
In this case calling skb_orphan_frags() updated nr_frags leaving nrfrags local variable in skb_segment() stale. This resulted in the code hitting i >= nrfrags prematurely and trying to move to next frag_skb using list_skb pointer, which was NULL, and caused kernel panic. Move the call to zero copy functions before using frags and nr_frags.
--- net/core/skbuff.c | 34 ++++++++++++++++++++-------------- 1 file changed, 20 insertions(+), 14 deletions(-)
net/core/skbuff.c +20 -14
@@ -4354,21 +4354,20 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,struct sk_buff *segs =NULL ;struct sk_buff *tail =NULL ;struct sk_buff *list_skb =skb_shinfo (head_skb)->frag_list; -skb_frag_t *frag =skb_shinfo (head_skb)->frags;unsigned int mss =skb_shinfo (head_skb)->gso_size;unsigned int doffset = head_skb->data -skb_mac_header (head_skb); -struct sk_buff *frag_skb = head_skb;unsigned int offset = doffset;unsigned int tnl_hlen =skb_tnl_header_len (head_skb);unsigned int partial_segs =0 ;unsigned int headroom;unsigned int len = head_skb->len; +struct sk_buff *frag_skb; +skb_frag_t *frag;__be16 proto;bool csum, sg; -int nfrags =skb_shinfo (head_skb)->nr_frags;int err = -ENOMEM ;int i =0 ; -int pos; +int nfrags, pos;if ((skb_shinfo (head_skb)->gso_type &SKB_GSO_DODGY ) && mss !=GSO_BY_FRAGS && mss !=skb_headlen (head_skb)) {
@@ -4445,6 +4444,13 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, headroom =skb_headroom (head_skb); pos =skb_headlen (head_skb); +if (skb_orphan_frags (head_skb,GFP_ATOMIC )) +return ERR_PTR (-ENOMEM ); + + nfrags =skb_shinfo (head_skb)->nr_frags; + frag =skb_shinfo (head_skb)->frags; + frag_skb = head_skb; + do { struct sk_buff *nskb; skb_frag_t *nskb_frag;
@@ -4465,6 +4471,10 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, (skb_headlen(list_skb) == len || sg)) {BUG_ON (skb_headlen (list_skb) > len); + nskb =skb_clone (list_skb,GFP_ATOMIC ); +if (unlikely (!nskb)) + goto err; + i =0 ; nfrags =skb_shinfo (list_skb)->nr_frags; frag =skb_shinfo (list_skb)->frags;
@@ -4483,12 +4493,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, frag++; } - nskb =skb_clone (list_skb,GFP_ATOMIC ); list_skb = list_skb->next; -if (unlikely (!nskb)) - goto err; -if (unlikely (pskb_trim (nskb, len))) {kfree_skb (nskb); goto err;
@@ -4564,12 +4570,16 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,skb_shinfo (nskb)->flags |=skb_shinfo (head_skb)->flags &SKBFL_SHARED_FRAG ; -if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) || -skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC )) +if (skb_zerocopy_clone (nskb, list_skb,GFP_ATOMIC )) goto err;while (pos < offset + len) {if (i >= nfrags) { +if (skb_orphan_frags (list_skb,GFP_ATOMIC ) || +skb_zerocopy_clone (nskb, list_skb, +GFP_ATOMIC )) + goto err; + i =0 ; nfrags =skb_shinfo (list_skb)->nr_frags; frag =skb_shinfo (list_skb)->frags;
@@ -4583,10 +4593,6 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, i--; frag--; } -if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) || -skb_zerocopy_clone (nskb, frag_skb, -GFP_ATOMIC )) - goto err; list_skb = list_skb->next; }
Signature · 1 line
2.17.1
Re: [PATCH v2] skbuff: skb_segment, Call zero copy functions before using skbuff frags
Quoted 94 lines · Mohamed Khalfella
Commit
bf5c25d60861("skbuff: inskb_segment, call zerocopy functions once per nskb") added the call to zero copy functions inskb_segment(). The change introduced a bug inskb_segment()becauseskb_orphan_frags()may possibly change the number of fragments or allocate new fragments altogether leaving nrfrags and frag to point to the old values. This can cause a panic with stacktrace like the one below.[ 193.894380] BUG: kernel NULL pointer dereference, address: 00000000000000bc [ 193.895273] CPU: 13 PID: 18164 Comm: vh-net-17428 Kdump: loaded Tainted: G O 5.15.123+ #26 [ 193.903919] RIP: 0010:skb_segment+0xb0e/0x12f0 [ 194.021892] Call Trace: [ 194.027422] <TASK> [ 194.072861] tcp_gso_segment+0x107/0x540 [ 194.082031] inet_gso_segment+0x15c/0x3d0 [ 194.090783] skb_mac_gso_segment+0x9f/0x110 [ 194.095016] __skb_gso_segment+0xc1/0x190 [ 194.103131] netem_enqueue+0x290/0xb10 [sch_netem] [ 194.107071] dev_qdisc_enqueue+0x16/0x70 [ 194.110884] __dev_queue_xmit+0x63b/0xb30 [ 194.121670] bond_start_xmit+0x159/0x380 [bonding] [ 194.128506] dev_hard_start_xmit+0xc3/0x1e0 [ 194.131787] __dev_queue_xmit+0x8a0/0xb30 [ 194.138225] macvlan_start_xmit+0x4f/0x100 [macvlan] [ 194.141477] dev_hard_start_xmit+0xc3/0x1e0 [ 194.144622] sch_direct_xmit+0xe3/0x280 [ 194.147748] __dev_queue_xmit+0x54a/0xb30 [ 194.154131] tap_get_user+0x2a8/0x9c0 [tap] [ 194.157358] tap_sendmsg+0x52/0x8e0 [tap] [ 194.167049] handle_tx_zerocopy+0x14e/0x4c0 [vhost_net] [ 194.173631] handle_tx+0xcd/0xe0 [vhost_net] [ 194.176959] vhost_worker+0x76/0xb0 [vhost] [ 194.183667] kthread+0x118/0x140 [ 194.190358] ret_from_fork+0x1f/0x30 [ 194.193670] </TASK>In this case calling
skb_orphan_frags()updatednr_fragsleaving nrfrags local variable inskb_segment()stale. This resulted in the code hitting i >= nrfrags prematurely and trying to move to nextfrag_skbusinglist_skbpointer, which was NULL, and caused kernel panic. Move the call to zero copy functions before using frags andnr_frags.Fixes bf5c25d60861 ("skbuff: in skb_segment, call zerocopy functions once per nskb")Signed-off-by Mohamed Khalfella <[email protected]>Reported-by Amit Goyal <[email protected]>--- net/core/skbuff.c | 34 ++++++++++++++++++++-------------- 1 file changed, 20 insertions(+), 14 deletions(-)net/core/skbuff.c +10
-4@@ -4354,21 +4354,20 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,struct sk_buff *segs =NULL ;struct sk_buff *tail =NULL ;struct sk_buff *list_skb =skb_shinfo (head_skb)->frag_list; -skb_frag_t *frag =skb_shinfo (head_skb)->frags;unsigned int mss =skb_shinfo (head_skb)->gso_size;unsigned int doffset = head_skb->data -skb_mac_header (head_skb); -struct sk_buff *frag_skb = head_skb;unsigned int offset = doffset;unsigned int tnl_hlen =skb_tnl_header_len (head_skb);unsigned int partial_segs =0 ;unsigned int headroom;unsigned int len = head_skb->len; +struct sk_buff *frag_skb; +skb_frag_t *frag;__be16 proto;bool csum, sg; -int nfrags =skb_shinfo (head_skb)->nr_frags;int err = -ENOMEM ;int i =0 ; -int pos; +int nfrags, pos;if ((skb_shinfo (head_skb)->gso_type &SKB_GSO_DODGY ) && mss !=GSO_BY_FRAGS && mss !=skb_headlen (head_skb)) {@@ -4445,6 +4444,13 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, headroom =skb_headroom (head_skb); pos =skb_headlen (head_skb); +if (skb_orphan_frags (head_skb,GFP_ATOMIC )) +return ERR_PTR (-ENOMEM ); + + nfrags =skb_shinfo (head_skb)->nr_frags; + frag =skb_shinfo (head_skb)->frags; + frag_skb = head_skb; + do { struct sk_buff *nskb; skb_frag_t *nskb_frag;
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4477
Quoted 7 lines
net/core/skbuff.c +4
-0@@ -4465,6 +4471,10 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, (skb_headlen(list_skb) == len || sg)) {BUG_ON (skb_headlen (list_skb) > len); + nskb =skb_clone (list_skb,GFP_ATOMIC ); +if (unlikely (!nskb)) + goto err; +
This patch is quite complex to review, so I am asking if this part was really needed ?
<1> : You moved here <2> and <3>
If this is not strictly needed, please keep the code as is to ease code review...
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4486
Quoted 8 lines
net/core/skbuff.c +0
-1i =0 ; nfrags =skb_shinfo (list_skb)->nr_frags; frag =skb_shinfo (list_skb)->frags;@@ -4483,12 +4493,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, frag++; } - nskb =skb_clone (list_skb,GFP_ATOMIC );
<2>
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4491
Quoted 5 lines
net/core/skbuff.c +0
-3list_skb = list_skb->next; -if (unlikely (!nskb)) - goto err; -
<3>
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4573
Quoted 10 lines
net/core/skbuff.c +1
-2if (unlikely (pskb_trim (nskb, len))) {kfree_skb (nskb); goto err;@@ -4564,12 +4570,16 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,skb_shinfo (nskb)->flags |=skb_shinfo (head_skb)->flags &SKBFL_SHARED_FRAG ; -if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) || -skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC )) +if (skb_zerocopy_clone (nskb, list_skb,GFP_ATOMIC ))
Why using list_skb here instead of frag_skb ?
Again, I have to look at the whole thing to understand why you did this.
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4582
Quoted 9 lines
net/core/skbuff.c +5
-0goto err;while (pos < offset + len) {if (i >= nfrags) { +if (skb_orphan_frags (list_skb,GFP_ATOMIC ) || +skb_zerocopy_clone (nskb, list_skb, +GFP_ATOMIC )) + goto err; +
This part is fine.
Quoted 15 lines
net/core/skbuff.c +0
-5i =0 ; nfrags =skb_shinfo (list_skb)->nr_frags; frag =skb_shinfo (list_skb)->frags;@@ -4583,10 +4593,6 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, i--; frag--; } -if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) || -skb_zerocopy_clone (nskb, frag_skb, -GFP_ATOMIC )) - goto err; list_skb = list_skb->next; } --
Quoted 2 lines
2.17.1
Thanks.
Re: [PATCH v2] skbuff: skb_segment, Call zero copy functions before using skbuff frags
Quoted 9 lines · Eric Dumazet
Quoted 7 lines · Mohamed Khalfella
do { struct sk_buff *nskb; skb_frag_t *nskb_frag;net/core/skbuff.c +0
-0@@ -4465,6 +4471,10 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, (skb_headlen(list_skb) == len || sg)) {BUG_ON (skb_headlen (list_skb) > len);
Quoted 7 lines
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4477
Quoted 4 lines
net/core/skbuff.c +4
-0+ nskb =skb_clone (list_skb,GFP_ATOMIC ); +if (unlikely (!nskb)) + goto err; +This patch is quite complex to review, so I am asking if this part was really needed ?
Unfortunately the patch is complex because I try to avoid calling skb_orphan_frags() in the middle of processing these frags. Otherwise it would be much harder to implement because as reallocated frags do not map 1:1 with existing frags as Willem mentioned.
Quoted 1 line
<1> : You moved here <2> and <3>
<2> was moved here because skb_clone() calls skb_orphan_frags(). By moving this up we do not need to call skb_orphan_frags() for list_skb and we can start to use nr_frags and frags without worrying their value is going to change.
<3> was moved here because <2> was moved here. Fail fast if we can not clone list_skb.
Quoted 30 lines
If this is not strictly needed, please keep the code as is to ease code review...
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4486
Quoted 8 lines
net/core/skbuff.c +0
-1i =0 ; nfrags =skb_shinfo (list_skb)->nr_frags; frag =skb_shinfo (list_skb)->frags;@@ -4483,12 +4493,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, frag++; } - nskb =skb_clone (list_skb,GFP_ATOMIC );<2>
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4491
Quoted 5 lines
net/core/skbuff.c +0
-3list_skb = list_skb->next; -if (unlikely (!nskb)) - goto err; -<3>
Quoted 7 lines
net/core/skbuff.c +0
-0if (unlikely (pskb_trim (nskb, len))) {kfree_skb (nskb); goto err;@@ -4564,12 +4570,16 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,skb_shinfo (nskb)->flags |=skb_shinfo (head_skb)->flags &SKBFL_SHARED_FRAG ;
Quoted 6 lines
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4573
Quoted 3 lines
net/core/skbuff.c +1
-2-if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) || -skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC )) +if (skb_zerocopy_clone (nskb, list_skb,GFP_ATOMIC ))Why using
list_skbhere instead offrag_skb?
Again, I have to look at the whole thing to understand why you did this.
Oops, this is a mistake. It should be frag_skb. Will fix it run the test one more time and post v3.
Re: [PATCH v2] skbuff: skb_segment, Call zero copy functions before using skbuff frags
Quoted 19 lines · Mohamed Khalfella
Quoted 9 lines · Eric Dumazet
Quoted 7 lines · Mohamed Khalfella
do { struct sk_buff *nskb; skb_frag_t *nskb_frag;net/core/skbuff.c +0
-0@@ -4465,6 +4471,10 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, (skb_headlen(list_skb) == len || sg)) {BUG_ON (skb_headlen (list_skb) > len);Quoted 7 lines
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4477
Quoted 4 lines
net/core/skbuff.c +4
-0+ nskb =skb_clone (list_skb,GFP_ATOMIC ); +if (unlikely (!nskb)) + goto err; +This patch is quite complex to review, so I am asking if this part was really needed ?
Quoted 8 lines
Unfortunately the patch is complex because I try to avoid calling
skb_orphan_frags()in the middle of processing these frags. Otherwise it would be much harder to implement because as reallocated frags do not map 1:1 with existing frags as Willem mentioned.Quoted 1 line
<1> : You moved here <2> and <3>
<2> was moved here because
skb_clone()callsskb_orphan_frags(). By
Oh right, I think we should amend skb_clone() documentation, it is slightly wrong.
( I will take care of this change)
Quoted 46 lines
moving this up we do not need to call
skb_orphan_frags()forlist_skband we can start to usenr_fragsand frags without worrying their value is going to change.<3> was moved here because <2> was moved here. Fail fast if we can not clone
list_skb.Quoted 30 lines
If this is not strictly needed, please keep the code as is to ease code review...
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4486
Quoted 8 lines
net/core/skbuff.c +0
-1i =0 ; nfrags =skb_shinfo (list_skb)->nr_frags; frag =skb_shinfo (list_skb)->frags;@@ -4483,12 +4493,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, frag++; } - nskb =skb_clone (list_skb,GFP_ATOMIC );<2>
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4491
Quoted 5 lines
net/core/skbuff.c +0
-3list_skb = list_skb->next; -if (unlikely (!nskb)) - goto err; -<3>
Quoted 7 lines
net/core/skbuff.c +0
-0if (unlikely (pskb_trim (nskb, len))) {kfree_skb (nskb); goto err;@@ -4564,12 +4570,16 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,skb_shinfo (nskb)->flags |=skb_shinfo (head_skb)->flags &SKBFL_SHARED_FRAG ;Quoted 6 lines
net/core/skbuff.c struct sk_buff *skb_segment(struct sk_buff *head_skb, · line 4573
Quoted 3 lines
net/core/skbuff.c +1
-2-if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) || -skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC )) +if (skb_zerocopy_clone (nskb, list_skb,GFP_ATOMIC ))Why using
list_skbhere instead offrag_skb?
Again, I have to look at the whole thing to understand why you did this.Oops, this is a mistake. It should be
frag_skb. Will fix it run the test one more time and post v3.
PATCH v3 1/1 skbuff: skb_segment, Call zero copy functions before using skbuff frags
In reply to willemjdebruijnCommit bf5c25d60861 ("skbuff: in skb_segment, call zerocopy functions once per nskb") added the call to zero copy functions in skb_segment(). The change introduced a bug in skb_segment() because skb_orphan_frags() may possibly change the number of fragments or allocate new fragments altogether leaving nrfrags and frag to point to the old values. This can cause a panic with stacktrace like the one below.
[ 193.894380] BUG: kernel NULL pointer dereference, address: 00000000000000bc [ 193.895273] CPU: 13 PID: 18164 Comm: vh-net-17428 Kdump: loaded Tainted: G O 5.15.123+ #26 [ 193.903919] RIP: 0010:skb_segment+0xb0e/0x12f0 [ 194.021892] Call Trace: [ 194.027422] <TASK> [ 194.072861] tcp_gso_segment+0x107/0x540 [ 194.082031] inet_gso_segment+0x15c/0x3d0 [ 194.090783] skb_mac_gso_segment+0x9f/0x110 [ 194.095016] __skb_gso_segment+0xc1/0x190 [ 194.103131] netem_enqueue+0x290/0xb10 [sch_netem] [ 194.107071] dev_qdisc_enqueue+0x16/0x70 [ 194.110884] __dev_queue_xmit+0x63b/0xb30 [ 194.121670] bond_start_xmit+0x159/0x380 [bonding] [ 194.128506] dev_hard_start_xmit+0xc3/0x1e0 [ 194.131787] __dev_queue_xmit+0x8a0/0xb30 [ 194.138225] macvlan_start_xmit+0x4f/0x100 [macvlan] [ 194.141477] dev_hard_start_xmit+0xc3/0x1e0 [ 194.144622] sch_direct_xmit+0xe3/0x280 [ 194.147748] __dev_queue_xmit+0x54a/0xb30 [ 194.154131] tap_get_user+0x2a8/0x9c0 [tap] [ 194.157358] tap_sendmsg+0x52/0x8e0 [tap] [ 194.167049] handle_tx_zerocopy+0x14e/0x4c0 [vhost_net] [ 194.173631] handle_tx+0xcd/0xe0 [vhost_net] [ 194.176959] vhost_worker+0x76/0xb0 [vhost] [ 194.183667] kthread+0x118/0x140 [ 194.190358] ret_from_fork+0x1f/0x30 [ 194.193670] </TASK>
In this case calling skb_orphan_frags() updated nr_frags leaving nrfrags local variable in skb_segment() stale. This resulted in the code hitting i >= nrfrags prematurely and trying to move to next frag_skb using list_skb pointer, which was NULL, and caused kernel panic. Move the call to zero copy functions before using frags and nr_frags.
--- net/core/skbuff.c | 34 ++++++++++++++++++++-------------- 1 file changed, 20 insertions(+), 14 deletions(-)
net/core/skbuff.c +20 -14
@@ -4354,21 +4354,20 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,struct sk_buff *segs =NULL ;struct sk_buff *tail =NULL ;struct sk_buff *list_skb =skb_shinfo (head_skb)->frag_list; -skb_frag_t *frag =skb_shinfo (head_skb)->frags;unsigned int mss =skb_shinfo (head_skb)->gso_size;unsigned int doffset = head_skb->data -skb_mac_header (head_skb); -struct sk_buff *frag_skb = head_skb;unsigned int offset = doffset;unsigned int tnl_hlen =skb_tnl_header_len (head_skb);unsigned int partial_segs =0 ;unsigned int headroom;unsigned int len = head_skb->len; +struct sk_buff *frag_skb; +skb_frag_t *frag;__be16 proto;bool csum, sg; -int nfrags =skb_shinfo (head_skb)->nr_frags;int err = -ENOMEM ;int i =0 ; -int pos; +int nfrags, pos;if ((skb_shinfo (head_skb)->gso_type &SKB_GSO_DODGY ) && mss !=GSO_BY_FRAGS && mss !=skb_headlen (head_skb)) {
@@ -4445,6 +4444,13 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, headroom =skb_headroom (head_skb); pos =skb_headlen (head_skb); +if (skb_orphan_frags (head_skb,GFP_ATOMIC )) +return ERR_PTR (-ENOMEM ); + + nfrags =skb_shinfo (head_skb)->nr_frags; + frag =skb_shinfo (head_skb)->frags; + frag_skb = head_skb; + do { struct sk_buff *nskb; skb_frag_t *nskb_frag;
@@ -4465,6 +4471,10 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, (skb_headlen(list_skb) == len || sg)) {BUG_ON (skb_headlen (list_skb) > len); + nskb =skb_clone (list_skb,GFP_ATOMIC ); +if (unlikely (!nskb)) + goto err; + i =0 ; nfrags =skb_shinfo (list_skb)->nr_frags; frag =skb_shinfo (list_skb)->frags;
@@ -4483,12 +4493,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, frag++; } - nskb =skb_clone (list_skb,GFP_ATOMIC ); list_skb = list_skb->next; -if (unlikely (!nskb)) - goto err; -if (unlikely (pskb_trim (nskb, len))) {kfree_skb (nskb); goto err;
@@ -4564,12 +4570,16 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,skb_shinfo (nskb)->flags |=skb_shinfo (head_skb)->flags &SKBFL_SHARED_FRAG ; -if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) || -skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC )) +if (skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC )) goto err;while (pos < offset + len) {if (i >= nfrags) { +if (skb_orphan_frags (list_skb,GFP_ATOMIC ) || +skb_zerocopy_clone (nskb, list_skb, +GFP_ATOMIC )) + goto err; + i =0 ; nfrags =skb_shinfo (list_skb)->nr_frags; frag =skb_shinfo (list_skb)->frags;
@@ -4583,10 +4593,6 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb, i--; frag--; } -if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) || -skb_zerocopy_clone (nskb, frag_skb, -GFP_ATOMIC )) - goto err; list_skb = list_skb->next; }
Signature · 1 line
2.17.1
Re: [PATCH v3] skbuff: skb_segment, Call zero copy functions before using skbuff frags
Quoted 16 lines · Mohamed Khalfella
Commit
bf5c25d60861("skbuff: inskb_segment, call zerocopy functions once per nskb") added the call to zero copy functions inskb_segment(). The change introduced a bug inskb_segment()becauseskb_orphan_frags()may possibly change the number of fragments or allocate new fragments altogether leaving nrfrags and frag to point to the old values. This can cause a panic with stacktrace like the one below.In this case calling
skb_orphan_frags()updatednr_fragsleaving nrfrags local variable inskb_segment()stale. This resulted in the code hitting i >= nrfrags prematurely and trying to move to nextfrag_skbusinglist_skbpointer, which was NULL, and caused kernel panic. Move the call to zero copy functions before using frags andnr_frags.
Quoted 5 lines
Fixes bf5c25d60861 ("skbuff: in skb_segment, call zerocopy functions once per nskb")Signed-off-by Mohamed Khalfella <[email protected]>Reported-by Amit Goyal <[email protected]>---
Re: [PATCH v3] skbuff: skb_segment, Call zero copy functions before using skbuff frags
In reply to Mohamed KhalfellaThis patch was applied to netdev/net.git (main) by David S. Miller <[email protected]>:
Hello:
This patch was applied to netdev/net.git (main)
by David S. Miller <[email protected]>:
Quoted 8 lines · On Thu, 31 Aug 2023 02:17:02 -0600 you wrote:
Commit
bf5c25d60861("skbuff: inskb_segment, call zerocopy functions once per nskb") added the call to zero copy functions inskb_segment(). The change introduced a bug inskb_segment()becauseskb_orphan_frags()may possibly change the number of fragments or allocate new fragments altogether leaving nrfrags and frag to point to the old values. This can cause a panic with stacktrace like the one below.[...]
Here is the summary with links:
- [v3] skbuff: skb_segment, Call zero copy functions before using skbuff frags
https://git.kernel.org/netdev/net/c/2ea35288c83b
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html