TCP stack should make sure it owns skbs before mangling them.
We had various crashes using bnx2x, and it turned out gso_size
was cleared right before bnx2x driver was populating TC descriptor
of the _previous_ packet send. TCP stack can sometime retransmit
packets that are still in Qdisc.
Of course we could make bnx2x driver more robust (using
ACCESS_ONCE(shinfo->gso_size) for example), but the bug is TCP stack.
We have identified two points where skb_unclone() was needed.
This patch adds a WARN_ON_ONCE() to warn us if we missed another
fix of this kind.
Kudos to Neal for finding the root cause of this bug. Its visible
using small MSS.
Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Neal Cardwell <ncardwell@google.com> Cc: Yuchung Cheng <ycheng@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Eric Dumazet [Wed, 23 Nov 2011 20:49:31 +0000 (15:49 -0500)]
ipv6: tcp: fix panic in SYN processing
commit 72a3effaf633bc ([NET]: Size listen hash tables using backlog
hint) added a bug allowing inet6_synq_hash() to return an out of bound
array index, because of u16 overflow.
Bug can happen if system admins set net.core.somaxconn &
net.ipv4.tcp_max_syn_backlog sysctls to values greater than 65536
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit c16a98ed91597b40b22b540c6517103497ef8e74) Signed-off-by: Willy Tarreau <w@1wt.eu>
64z is missing rhel6 commit 3af031a395c0 ("[crypto] algboss: Hold ref
count on larval") which is causing cosmetic fuzz, because crypto_alg_get
was move from crypto/api.c to crypto/internal.h.
crypto_larval_lookup should only return a larval if it created one.
Any larval created by another entity must be processed through
crypto_larval_wait before being returned.
Otherwise this will lead to a larval being killed twice, which
will most likely lead to a crash.
Many drivers need to validate the characteristics of their HID report
during initialization to avoid misusing the reports. This adds a common
helper to perform validation of the report exisitng, the field existing,
and the expected number of values within the field.
Signed-off-by: Kees Cook <keescook@chromium.org> Cc: stable@vger.kernel.org Reviewed-by: Benjamin Tissoires <benjamin.tissoires@redhat.com> Signed-off-by: Jiri Kosina <jkosina@suse.cz>
[jmm: backported to 2.6.32]
[wt: dev_err() in 2.6.32 instead of hid_err()] Signed-off-by: Willy Tarreau <w@1wt.eu>
A HID device could send a malicious output report that would cause the
lg, lg3, and lg4 HID drivers to write beyond the output report allocation
during an event, causing a heap overflow:
[ 325.245240] usb 1-1: New USB device found, idVendor=046d, idProduct=c287
...
[ 414.518960] BUG kmalloc-4096 (Not tainted): Redzone overwritten
Additionally, while lg2 did correctly validate the report details, it was
cleaned up and shortened.
A HID device could send a malicious output report that would cause the
pantherlord HID driver to write beyond the output report allocation
during initialization, causing a heap overflow:
[ 310.939483] usb 1-1: New USB device found, idVendor=0e8f, idProduct=0003
...
[ 315.980774] BUG kmalloc-192 (Tainted: G W ): Redzone overwritten
The zeroplus HID driver was not checking the size of allocated values
in fields it used. A HID device could send a malicious output report
that would cause the driver to write beyond the output report allocation
during initialization, causing a heap overflow:
[ 1442.728680] usb 1-1: New USB device found, idVendor=0c12, idProduct=0005
...
[ 1466.243173] BUG kmalloc-192 (Tainted: G W ): Redzone overwritten
The "Report ID" field of a HID report is used to build indexes of
reports. The kernel's index of these is limited to 256 entries, so any
malicious device that sets a Report ID greater than 255 will trigger
memory corruption on the host:
[ 1347.156239] BUG: unable to handle kernel paging request at ffff88094958a878
[ 1347.156261] IP: [<ffffffff813e4da0>] hid_register_report+0x2a/0x8b
The module parameter "fwpostfix" is userspace controllable, unfiltered,
and is used to define the firmware filename. b43_do_request_fw() populates
ctx->errors[] on error, containing the firmware filename. b43err()
parses its arguments as a format string. For systems with b43 hardware,
this could lead to a uid-0 to ring-0 escalation.
CVE-2013-2852
Signed-off-by: Kees Cook <keescook@chromium.org> Cc: stable@vger.kernel.org Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
Disk names may contain arbitrary strings, so they must not be
interpreted as format strings. It seems that only md allows arbitrary
strings to be used for disk names, but this could allow for a local
memory corruption from uid 0 into ring 0.
key_notify_sa_flush() and key_notify_policy_flush() miss to initialize
the sadb_msg_reserved member of the broadcasted message and thereby
leak 2 bytes of heap memory to listeners. Fix that.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
It's possible to use AF_INET6 sockets and to connect to an IPv4
destination. After this, socket dst cache is a pointer to a rtable,
not rt6_info.
ip6_sk_dst_check() should check the socket dst cache is IPv6, or else
various corruptions/crashes can happen.
Dave Jones can reproduce immediate crash with
trinity -q -l off -n -c sendmsg -c connect
With help from Hannes Frederic Sowa
Reported-by: Dave Jones <davej@redhat.com> Reported-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: Eric Dumazet <edumazet@google.com> Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
When SCTP is done processing a duplicate cookie chunk, it tries
to delete a newly created association. For that, it has to set
the right association for the side-effect processing to work.
However, when it uses the SCTP_CMD_NEW_ASOC command, that performs
more work then really needed (like hashing the associationa and
assigning it an id) and there is no point to do that only to
delete the association as a next step. In fact, it also creates
an impossible condition where an association may be found by
the getsockopt() call, and that association is empty. This
causes a crash in some sctp getsockopts.
The solution is rather simple. We simply use SCTP_CMD_SET_ASOC
command that doesn't have all the overhead and does exactly
what we need.
Reported-by: Karl Heiss <kheiss@gmail.com> Tested-by: Karl Heiss <kheiss@gmail.com> CC: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: Vlad Yasevich <vyasevich@gmail.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Attempt to reduce the number of IP packets emitted in response to single
SCTP packet (2e3216cd) introduced a complication - if a packet contains
two COOKIE_ECHO chunks and nothing else then SCTP state machine corks the
socket while processing first COOKIE_ECHO and then loses the association
and forgets to uncork the socket. To deal with the issue add new SCTP
command which can be used to set association explictly. Use this new
command when processing second COOKIE_ECHO chunk to restore the context
for SCTP state machine.
Signed-off-by: Max Matveev <makc@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[dannf: backported to Debian's 2.6.32] Signed-off-by: Willy Tarreau <w@1wt.eu>
In line 2908 we can find the copy_to_user function:
2908 if (!ret && copy_to_user(arg, cgc->buffer, blocksize))
The cgc->buffer is never cleaned and initialized before this function.
If ret = 0 with the previous basic block, it's possible to display some
memory bytes in kernel space from userspace.
When we read a block from the disk it normally fills the ->buffer but if
the drive is malfunctioning there is a chance that it would only be
partially filled. The result is an leak information to userspace.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Cc: Jens Axboe <axboe@kernel.dk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
The pciinfo struct has a two byte hole after ->dev_fn so stack
information could be leaked to the user.
This was assigned CVE-2013-2147.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Acked-by: Mike Miller <mike.miller@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
The arg64 struct has a hole after ->buf_size which isn't cleared. Or if
any of the calls to copy_from_user() fail then that would cause an
information leak as well.
This was assigned CVE-2013-2147.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Acked-by: Mike Miller <mike.miller@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
kernel/kmod.c: check for NULL in call_usermodehelper_exec()
If /proc/sys/kernel/core_pattern contains only "|", a NULL pointer
dereference happens upon core dump because argv_split("") returns
argv[0] == NULL.
This bug was once fixed by commit 264b83c07a84 ("usermodehelper: check
subprocess_info->path != NULL") but was by error reintroduced by commit 7f57cfa4e2aa ("usermodehelper: kill the sub_info->path[0] check").
This bug seems to exist since 2.6.19 (the version which core dump to
pipe was added). Depending on kernel version and config, some side
effect might happen immediately after this oops (e.g. kernel panic with
2.6.32-358.18.1.el6).
The `insn_bits` handler `ni_65xx_dio_insn_bits()` has a `for` loop that
currently writes (optionally) and reads back up to 5 "ports" consisting
of 8 channels each. It reads up to 32 1-bit channels but can only read
and write a whole port at once - it needs to handle up to 5 ports as the
first channel it reads might not be aligned on a port boundary. It
breaks out of the loop early if the next port it handles is beyond the
final port on the card. It also breaks out early on the 5th port in the
loop if the first channel was aligned. Unfortunately, it doesn't check
that the current port it is dealing with belongs to the comedi subdevice
the `insn_bits` handler is acting on. That's a bug.
Redo the `for` loop to terminate after the final port belonging to the
subdevice, changing the loop variable in the process to simplify things
a bit. The `for` loop could now try and handle more than 5 ports if the
subdevice has more than 40 channels, but the test `if (bitshift >= 32)`
ensures it will break out early after 4 or 5 ports (depending on whether
the first channel is aligned on a port boundary). (`bitshift` will be
between -7 and 7 inclusive on the first iteration, increasing by 8 for
each subsequent operation.)
Signed-off-by: Ian Abbott <abbotti@mev.co.uk> Signed-off-by: Willy Tarreau <w@1wt.eu>
Julian Anastasov [Sun, 17 Oct 2010 13:14:31 +0000 (16:14 +0300)]
ipvs: fix CHECKSUM_PARTIAL for TCP, UDP
Fix CHECKSUM_PARTIAL handling. Tested for IPv4 TCP,
UDP not tested because it needs network card with HW CSUM support.
May be fixes problem where IPVS can not be used in virtual boxes.
Problem appears with DNAT to local address when the local stack
sends reply in CHECKSUM_PARTIAL mode.
Fix tcp_dnat_handler and udp_dnat_handler to provide
vaddr and daddr in right order (old and new IP) when calling
tcp_partial_csum_update/udp_partial_csum_update (CHECKSUM_PARTIAL).
Signed-off-by: Julian Anastasov <ja@ssi.bg> Signed-off-by: Simon Horman <horms@verge.net.au>
(cherry picked from commit 5bc9068e9d962ca6b8bec3f0eb6f60ab4dee1d04) Signed-off-by: Willy Tarreau <w@1wt.eu>
Willy Tarreau [Thu, 13 Jun 2013 17:36:35 +0000 (19:36 +0200)]
x86, ptrace: fix build breakage with gcc 4.7 (second try)
syscall_trace_enter() and syscall_trace_leave() are only called from
within asm code and do not need to be declared in the .c at all.
Removing their reference fixes the build issue that was happening
with gcc 4.7.
Both Sven-Haegar Koch and Christoph Biedl confirmed this patch
addresses their respective build issues.
As reported by Sven-Haegar Koch, this patch breaks make headers_check :
CHECK include (0 files)
CHECK include/asm (54 files)
/home/haegar/src/2.6.32/linux/usr/include/asm/ptrace.h:5: included file 'linux/linkage.h' is not exported
Ben Greear [Thu, 6 Jun 2013 21:29:49 +0000 (14:29 -0700)]
Fix lockup related to stop_machine being stuck in __do_softirq.
The stop machine logic can lock up if all but one of the migration
threads make it through the disable-irq step and the one remaining
thread gets stuck in __do_softirq. The reason __do_softirq can hang is
that it has a bail-out based on jiffies timeout, but in the lockup case,
jiffies itself is not incremented.
To work around this, re-add the max_restart counter in __do_irq and stop
processing irqs after 10 restarts.
Thanks to Tejun Heo and Rusty Russell and others for helping me track
this down.
This was introduced in 3.9 by commit c10d73671ad3 ("softirq: reduce
latencies").
It may be worth looking into ath9k to see if it has issues with its irq
handler at a later date.
The hang stack traces look something like this:
------------[ cut here ]------------
WARNING: at kernel/watchdog.c:245 watchdog_overflow_callback+0x9c/0xa7()
Watchdog detected hard LOCKUP on cpu 2
Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc]
Pid: 23, comm: migration/2 Tainted: G C 3.9.4+ #11
Call Trace:
<NMI> warn_slowpath_common+0x85/0x9f
warn_slowpath_fmt+0x46/0x48
watchdog_overflow_callback+0x9c/0xa7
__perf_event_overflow+0x137/0x1cb
perf_event_overflow+0x14/0x16
intel_pmu_handle_irq+0x2dc/0x359
perf_event_nmi_handler+0x19/0x1b
nmi_handle+0x7f/0xc2
do_nmi+0xbc/0x304
end_repeat_nmi+0x1e/0x2e
<<EOE>>
cpu_stopper_thread+0xae/0x162
smpboot_thread_fn+0x258/0x260
kthread+0xc7/0xcf
ret_from_fork+0x7c/0xb0
---[ end trace 4947dfa9b0a4cec3 ]---
BUG: soft lockup - CPU#1 stuck for 22s! [migration/1:17]
Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc]
irq event stamp: 835637905
hardirqs last enabled at (835637904): __do_softirq+0x9f/0x257
hardirqs last disabled at (835637905): apic_timer_interrupt+0x6d/0x80
softirqs last enabled at (5654720): __do_softirq+0x1ff/0x257
softirqs last disabled at (5654725): irq_exit+0x5f/0xbb
CPU 1
Pid: 17, comm: migration/1 Tainted: G WC 3.9.4+ #11 To be filled by O.E.M. To be filled by O.E.M./To be filled by O.E.M.
RIP: tasklet_hi_action+0xf0/0xf0
Process migration/1
Call Trace:
<IRQ>
__do_softirq+0x117/0x257
irq_exit+0x5f/0xbb
smp_apic_timer_interrupt+0x8a/0x98
apic_timer_interrupt+0x72/0x80
<EOI>
printk+0x4d/0x4f
stop_machine_cpu_stop+0x22c/0x274
cpu_stopper_thread+0xae/0x162
smpboot_thread_fn+0x258/0x260
kthread+0xc7/0xcf
ret_from_fork+0x7c/0xb0
Signed-off-by: Ben Greear <greearb@candelatech.com> Acked-by: Tejun Heo <tj@kernel.org> Acked-by: Pekka Riikonen <priikone@iki.fi> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: stable@kernel.org Cc: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
(cherry picked from commit 34376a50fb1fa095b9d0636fa41ed2e73125f214) Signed-off-by: Willy Tarreau <w@1wt.eu>
Thomas Bork [Sun, 16 Jun 2013 17:47:23 +0000 (19:47 +0200)]
scsi: fix missing include linux/types.h in scsi_netlink.h
Thomas Bork reported that commit c6203cd ("scsi: use __uX
types for headers exported to user space") caused a regression
as now he's getting this warning :
> /usr/src/linux-2.6.32-eisfair-1/usr/include/scsi/scsi_netlink.h:108:
> found __[us]{8,16,32,64} type without #include <linux/types.h>
This issue was addressed later by commit 10db4e1 ("headers:
include linux/types.h where appropriate"), so let's just pick the
relevant part from it.
Cc: Thomas Bork <tom@eisfair.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Willy Tarreau [Fri, 7 Jun 2013 05:11:37 +0000 (07:11 +0200)]
x86, ptrace: fix build breakage with gcc 4.7
Christoph Biedl reported that 2.6.32 does not build with gcc 4.7 on
i386 :
CC arch/x86/kernel/ptrace.o
arch/x86/kernel/ptrace.c:1472:17: error: conflicting types for 'syscall_trace_enter'
In file included from /«PKGBUILDDIR»/arch/x86/include/asm/vm86.h:130:0,
from /«PKGBUILDDIR»/arch/x86/include/asm/processor.h:10,
from /«PKGBUILDDIR»/arch/x86/include/asm/thread_info.h:22,
from include/linux/thread_info.h:56,
from include/linux/preempt.h:9,
from include/linux/spinlock.h:50,
from include/linux/seqlock.h:29,
from include/linux/time.h:8,
from include/linux/timex.h:56,
from include/linux/sched.h:56,
from arch/x86/kernel/ptrace.c:11:
/«PKGBUILDDIR»/arch/x86/include/asm/ptrace.h:145:13: note: previous declaration of 'syscall_trace_enter' was here
arch/x86/kernel/ptrace.c:1517:17: error: conflicting types for 'syscall_trace_leave'
In file included from /«PKGBUILDDIR»/arch/x86/include/asm/vm86.h:130:0,
from /«PKGBUILDDIR»/arch/x86/include/asm/processor.h:10,
from /«PKGBUILDDIR»/arch/x86/include/asm/thread_info.h:22,
from include/linux/thread_info.h:56,
from include/linux/preempt.h:9,
from include/linux/spinlock.h:50,
from include/linux/seqlock.h:29,
from include/linux/time.h:8,
from include/linux/timex.h:56,
from include/linux/sched.h:56,
from arch/x86/kernel/ptrace.c:11:
/«PKGBUILDDIR»/arch/x86/include/asm/ptrace.h:146:13: note: previous declaration of 'syscall_trace_leave' was here
make[4]: *** [arch/x86/kernel/ptrace.o] Error 1
make[3]: *** [arch/x86/kernel] Error 2
make[3]: *** Waiting for unfinished jobs....
He also found that this issue did not appear in more recent kernels since
this asmregparm disappeared in 3.0-rc1 with commit 1b4ac2a935 that was
applied after some UM changes that we don't necessarily want in 2.6.32.
Thus, the cleanest fix for older kernels is to make the declaration in
ptrace.h match the one in ptrace.c by specifying asmregparm on these
functions. They're only called from asm which explains why it used to
work despite the inconsistency in the declaration.
Reported-by: Christoph Biedl <linux-kernel.bfrz@manchmal.in-ulm.de> Tested-by: Christoph Biedl <linux-kernel.bfrz@manchmal.in-ulm.de> Signed-off-by: Willy Tarreau <w@1wt.eu>
The code in set_orig_addr() does not initialize all of the members of
struct sockaddr_tipc when filling the sockaddr info -- namely the union
is only partly filled. This will make recv_msg() and recv_stream() --
the only users of this function -- leak kernel stack memory as the
msg_name member is a local variable in net/socket.c.
Additionally to that both recv_msg() and recv_stream() fail to update
the msg_namelen member to 0 while otherwise returning with 0, i.e.
"success". This is the case for, e.g., non-blocking sockets. This will
lead to a 128 byte kernel stack leak in net/socket.c.
Fix the first issue by initializing the memory of the union with
memset(0). Fix the second one by setting msg_namelen to 0 early as it
will be updated later if we're going to fill the msg_name member.
Cc: Jon Maloy <jon.maloy@ericsson.com> Cc: Allan Stephens <allan.stephens@windriver.com> Signed-off-by: Mathias Krause <minipli@googlemail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[dannf: backported to Debian's 2.6.32] Signed-off-by: Willy Tarreau <w@1wt.eu>
The current code does not fill the msg_name member in case it is set.
It also does not set the msg_namelen member to 0 and therefore makes
net/socket.c leak the local, uninitialized sockaddr_storage variable
to userland -- 128 bytes of kernel stack memory.
Fix that by simply setting msg_namelen to 0 as obviously nobody cared
about irda_recvmsg_dgram() not filling the msg_name in case it was
set.
Cc: Samuel Ortiz <samuel@sortiz.org> Signed-off-by: Mathias Krause <minipli@googlemail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[dannf: adjusted to apply to Debian's 2.6.32] Signed-off-by: Willy Tarreau <w@1wt.eu>
The code in rose_recvmsg() does not initialize all of the members of
struct sockaddr_rose/full_sockaddr_rose when filling the sockaddr info.
Nor does it initialize the padding bytes of the structure inserted by
the compiler for alignment. This will lead to leaking uninitialized
kernel stack bytes in net/socket.c.
Fix the issue by initializing the memory used for sockaddr info with
memset(0).
Signed-off-by: Mathias Krause <minipli@googlemail.com> Cc: Ralf Baechle <ralf@linux-mips.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Jay Fenlason (fenlason@redhat.com) found a bug,
that recvfrom() on an RDS socket can return the contents of random kernel
memory to userspace if it was called with a address length larger than
sizeof(struct sockaddr_in).
rds_recvmsg() also fails to set the addr_len paramater properly before
returning, but that's just a bug.
There are also a number of cases wher recvfrom() can return an entirely bogus
address. Anything in rds_recvmsg() that returns a non-negative value but does
not go through the "sin = (struct sockaddr_in *)msg->msg_name;" code path
at the end of the while(1) loop will return up to 128 bytes of kernel memory
to userspace.
And I write two test programs to reproduce this bug, you will see that in
rds_server, fromAddr will be overwritten and the following sock_fd will be
destroyed.
Yes, it is the programmer's fault to set msg_namelen incorrectly, but it is
better to make the kernel copy the real length of address to user space in
such case.
How to run the test programs ?
I test them on 32bit x86 system, 3.5.0-rc7.
4 you will see something like:
server is waiting to receive data...
old socket fd=3
server received data from client:data from client
msg.msg_namelen=32
new socket fd=-1067277685
sendmsg()
: Bad file descriptor
printf("server is waiting to receive data...\n");
msg.msg_name = &fromAddr;
/*
* I add 16 to sizeof(fromAddr), ie 32,
* and pay attention to the definition of fromAddr,
* recvmsg() will overwrite sock_fd,
* since kernel will copy 32 bytes to userspace.
*
* If you just use sizeof(fromAddr), it works fine.
* */
msg.msg_namelen = sizeof(fromAddr) + 16;
/* msg.msg_namelen = sizeof(fromAddr); */
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
msg.msg_iov->iov_base = recvBuffer;
msg.msg_iov->iov_len = 128;
msg.msg_control = 0;
msg.msg_controllen = 0;
msg.msg_flags = 0;
while (1) {
printf("old socket fd=%d\n", sock_fd);
if (recvmsg(sock_fd, &msg, 0) == -1) {
perror("recvmsg() error\n");
close(sock_fd);
exit(1);
}
printf("server received data from client:%s\n", recvBuffer);
printf("msg.msg_namelen=%d\n", msg.msg_namelen);
printf("new socket fd=%d\n", sock_fd);
strcat(recvBuffer, "--data from server");
if (sendmsg(sock_fd, &msg, 0) == -1) {
perror("sendmsg()\n");
close(sock_fd);
exit(1);
}
}
close(sock_fd);
return 0;
}
Signed-off-by: Weiping Pan <wpan@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[dannf: Adjusted to apply to Debian's 2.6.32] Signed-off-by: Willy Tarreau <w@1wt.eu>
For stream sockets the code misses to update the msg_namelen member
to 0 and therefore makes net/socket.c leak the local, uninitialized
sockaddr_storage variable to userland -- 128 bytes of kernel stack
memory. The msg_namelen update is also missing for datagram sockets
in case the socket is shutting down during receive.
Fix both issues by setting msg_namelen to 0 early. It will be
updated later if we're going to fill the msg_name member.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
The LLC code wrongly returns 0, i.e. "success", when the socket is
zapped. Together with the uninitialized uaddrlen pointer argument from
sys_getsockname this leads to an arbitrary memory leak of up to 128
bytes kernel stack via the getsockname() syscall.
Return an error instead when the socket is zapped to prevent the info
leak. Also remove the unnecessary memset(0). We don't directly write to
the memory pointed by uaddr but memcpy() a local structure at the end of
the function that is properly initialized.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
The current code does not fill the msg_name member in case it is set.
It also does not set the msg_namelen member to 0 and therefore makes
net/socket.c leak the local, uninitialized sockaddr_storage variable
to userland -- 128 bytes of kernel stack memory.
Fix that by simply setting msg_namelen to 0 as obviously nobody cared
about iucv_sock_recvmsg() not filling the msg_name in case it was set.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Cc: Ursula Braun <ursula.braun@de.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
When msg_namelen is non-zero the sockaddr info gets filled out, as
requested, but the code fails to initialize the padding bytes of struct
sockaddr_ax25 inserted by the compiler for alignment. Additionally the
msg_namelen value is updated to sizeof(struct full_sockaddr_ax25) but is
not always filled up to this size.
Both issues lead to the fact that the code will leak uninitialized
kernel stack bytes in net/socket.c.
Fix both issues by initializing the memory with memset(0).
Signed-off-by: Mathias Krause <minipli@googlemail.com> Cc: Ralf Baechle <ralf@linux-mips.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
The ATM code fails to initialize the two padding bytes of struct
sockaddr_atmpvc inserted for alignment. Add an explicit memset(0)
before filling the structure to avoid the info leak.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 2.6.32: adjust context, indentation] Signed-off-by: Willy Tarreau <w@1wt.eu>
The ATM code fails to initialize the two padding bytes of struct
sockaddr_atmpvc inserted for alignment. Add an explicit memset(0)
before filling the structure to avoid the info leak.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 2.6.32: adjust context] Signed-off-by: Willy Tarreau <w@1wt.eu>
The current code does not fill the msg_name member in case it is set.
It also does not set the msg_namelen member to 0 and therefore makes
net/socket.c leak the local, uninitialized sockaddr_storage variable
to userland -- 128 bytes of kernel stack memory.
Fix that by simply setting msg_namelen to 0 as obviously nobody cared
about vcc_recvmsg() not filling the msg_name in case it was set.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
If at least one of CONFIG_IP_VS_PROTO_TCP or CONFIG_IP_VS_PROTO_UDP is
not set, __ip_vs_get_timeouts() does not fully initialize the structure
that gets copied to userland and that for leaks up to 12 bytes of kernel
stack. Add an explicit memset(0) before passing the structure to
__ip_vs_get_timeouts() to avoid the info leak.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Cc: Wensong Zhang <wensong@linux-vs.org> Cc: Simon Horman <horms@verge.net.au> Cc: Julian Anastasov <ja@ssi.bg> Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 2.6.32: adjust context] Signed-off-by: Willy Tarreau <w@1wt.eu>
Cleaning up the IPv6 MTU checking in the IPVS xmit code, by using
a common helper function __mtu_check_toobig_v6().
The MTU check for tunnel mode can also use this helper as
ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr) is qual to
skb->len. And the 'mtu' variable have been adjusted before
calling helper.
Notice, this also fixes a bug, as the the MTU check in ip_vs_dr_xmit_v6()
were missing a check for skb_is_gso().
This bug e.g. caused issues for KVM IPVS setups, where different
Segmentation Offloading techniques are utilized, between guests,
via the virtio driver. This resulted in very bad performance,
due to the ICMPv6 "too big" messages didn't affect the sender.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
(cherry picked from commit 590e3f79a21edd2e9857ac3ced25ba6b2a491ef8) Signed-off-by: Willy Tarreau <w@1wt.eu>
It was reported that the Linux kernel sometimes logs:
klogd: [2629147.402413] kernel BUG at net / netfilter /
nf_conntrack_proto_tcp.c: 447!
klogd: [1072212.887368] kernel BUG at net / netfilter /
nf_conntrack_proto_tcp.c: 392
ipv4_get_l4proto() in nf_conntrack_l3proto_ipv4.c and tcp_error() in
nf_conntrack_proto_tcp.c should catch malformed packets, so the errors
at the indicated lines - TCP options parsing - should not happen.
However, tcp_error() relies on the "dataoff" offset to the TCP header,
calculated by ipv4_get_l4proto(). But ipv4_get_l4proto() does not check
bogus ihl values in IPv4 packets, which then can slip through tcp_error()
and get caught at the TCP options parsing routines.
The patch fixes ipv4_get_l4proto() by invalidating packets with bogus
ihl value.
The patch closes netfilter bugzilla id 771.
Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Acked-by: David Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Fernando Gont reported current IPv6 fragment identification generation
was not secure, because using a very predictable system-wide generator,
allowing various attacks.
IPv4 uses inetpeer cache to address this problem and to get good
performance. We'll use this mechanism when IPv6 inetpeer is stable
enough in linux-3.1
For the time being, we use jhash on destination address to provide less
predictable identifications. Also remove a spinlock and use cmpxchg() to
get better SMP performance.
Reported-by: Fernando Gont <fernando@gont.com.ar> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
[bwh: Backport further to 2.6.32] Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Willy Tarreau <w@1wt.eu>
RFC5722 prohibits reassembling fragments when some data overlaps.
Bug spotted by Zhang Zuotao <zuotao.zhang@6wind.com>.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[dannf: backported to Debian's 2.6.32] Signed-off-by: Willy Tarreau <w@1wt.eu>
For sensitive data like keying material, it is common practice to zero
out keys before returning the memory back to the allocator. Thus, use
kzfree instead of kfree.
Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Acked-by: Vlad Yasevich <vyasevich@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
On sctp_endpoint_destroy, previously used sensitive keying material
should be zeroed out before the memory is returned, as we already do
with e.g. auth keys when released.
Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Acked-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
In sctp_setsockopt_auth_key, we create a temporary copy of the user
passed shared auth key for the endpoint or association and after
internal setup, we free it right away. Since it's sensitive data, we
should zero out the key before returning the memory back to the
allocator. Thus, use kzfree instead of kfree, just as we do in
sctp_auth_key_put().
Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Trinity (the syscall fuzzer) discovered a memory leak in SCTP,
reproducible e.g. with the sendto() syscall by passing invalid
user space pointer in the second argument:
The dcb netlink interface leaks stack memory in various places:
* perm_addr[] buffer is only filled at max with 12 of the 32 bytes but
copied completely,
* no in-kernel driver fills all fields of an IEEE 802.1Qaz subcommand,
so we're leaking up to 58 bytes for ieee_ets structs, up to 136 bytes
for ieee_pfc structs, etc.,
* the same is true for CEE -- no in-kernel driver fills the whole
struct,
Prevent all of the above stack info leaks by properly initializing the
buffers/structures involved.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 2.6.32: no support for IEEE or CEE commands, so only
deal with perm_addr] Signed-off-by: Willy Tarreau <w@1wt.eu>
As reported by Jan, and others over the past few years, there is a
race condition caused by unix_release setting the sock->sk pointer
to NULL before properly marking the socket as dead/orphaned. This
can cause a problem with the LSM hook security_unix_may_send() if
there is another socket attempting to write to this partially
released socket in between when sock->sk is set to NULL and it is
marked as dead/orphaned. This patch fixes this by only setting
sock->sk to NULL after the socket has been marked as dead; I also
take the opportunity to make unix_release_sock() a void function
as it only ever returned 0/success.
Dave, I think this one should go on the -stable pile.
Special thanks to Jan for coming up with a reproducer for this
problem.
Reported-by: Jan Stancek <jan.stancek@gmail.com> Signed-off-by: Paul Moore <pmoore@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
commit 35f9c09fe9c72e (tcp: tcp_sendpages() should call tcp_push() once)
added an internal flag : MSG_SENDPAGE_NOTLAST meant to be set on all
frags but the last one for a splice() call.
The condition used to set the flag in pipe_to_sendpage() relied on
splice() user passing the exact number of bytes present in the pipe,
or a smaller one.
But some programs pass an arbitrary high value, and the test fails.
The effect of this bug is a lack of tcp_push() at the end of a
splice(pipe -> socket) call, and possibly very slow or erratic TCP
sessions.
We should both test sd->total_len and fact that another fragment
is in the pipe (pipe->nrbufs > 1)
Many thanks to Willy for providing very clear bug report, bisection
and test programs.
Reported-by: Willy Tarreau <w@1wt.eu> Bisected-by: Willy Tarreau <w@1wt.eu> Tested-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
vmsplice()/splice(pipe, socket) call do_tcp_sendpages() one page at a
time, adding at most 4096 bytes to an skb. (assuming PAGE_SIZE=4096)
The call to tcp_push() at the end of do_tcp_sendpages() forces an
immediate xmit when pipe is not already filled, and tso_fragment() try
to split these skb to MSS multiples.
4096 bytes are usually split in a skb with 2 MSS, and a remaining
sub-mss skb (assuming MTU=1500)
This makes slow start suboptimal because many small frames are sent to
qdisc/driver layers instead of big ones (constrained by cwnd and packets
in flight of course)
In fact, applications using sendmsg() (adding an additional memory copy)
instead of vmsplice()/splice()/sendfile() are a bit faster because of
this anomaly, especially if serving small files in environments with
large initial [c]wnd.
Call tcp_push() only if MSG_MORE is not set in the flags parameter.
This bit is automatically provided by splice() internals but for the
last page, or on all pages if user specified SPLICE_F_MORE splice()
flag.
In some workloads, this can reduce number of sent logical packets by an
order of magnitude, making zero-copy TCP actually faster than
one-copy :)
Reported-by: Tom Herbert <therbert@google.com> Cc: Nandita Dukkipati <nanditad@google.com> Cc: Neal Cardwell <ncardwell@google.com> Cc: Tom Herbert <therbert@google.com> Cc: Yuchung Cheng <ycheng@google.com> Cc: H.K. Jerry Chu <hkchu@google.com> Cc: Maciej Żenczykowski <maze@google.com> Cc: Mahesh Bandewar <maheshb@google.com> Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
We lack proper synchronization to manipulate inet->opt ip_options
Problem is ip_make_skb() calls ip_setup_cork() and
ip_setup_cork() possibly makes a copy of ipc->opt (struct ip_options),
without any protection against another thread manipulating inet->opt.
Another thread can change inet->opt pointer and free old one under us.
Use RCU to protect inet->opt (changed to inet->inet_opt).
Instead of handling atomic refcounts, just copy ip_options when
necessary, to avoid cache line dirtying.
We cant insert an rcu_head in struct ip_options since its included in
skb->cb[], so this patch is large because I had to introduce a new
ip_options_rcu structure.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
[dannf/bwh: backported to Debian's 2.6.32] Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Willy Tarreau <w@1wt.eu>
The implementation of dev_ifconf() for the compat ioctl interface uses
an intermediate ifc structure allocated in userland for the duration of
the syscall. Though, it fails to initialize the padding bytes inserted
for alignment and that for leaks four bytes of kernel stack. Add an
explicit memset(0) before filling the structure to avoid the info leak.
Signed-off-by: Mathias Krause <minipli@googlemail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 2.6.32: adjust filename, context] Signed-off-by: Willy Tarreau <w@1wt.eu>
Its possible to use RAW sockets to get a crash in
tcp_set_keepalive() / sk_reset_timer()
Fix is to make sure socket is a SOCK_STREAM one.
Reported-by: Dave Jones <davej@redhat.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Reading TCP stats when using TCP Illinois congestion control algorithm
can cause a divide by zero kernel oops.
The division by zero occur in tcp_illinois_info() at:
do_div(t, ca->cnt_rtt);
where ca->cnt_rtt can become zero (when rtt_reset is called)
Steps to Reproduce:
1. Register tcp_illinois:
# sysctl -w net.ipv4.tcp_congestion_control=illinois
2. Monitor internal TCP information via command "ss -i"
# watch -d ss -i
3. Establish new TCP conn to machine
Either it fails at the initial conn, or else it needs to wait
for a loss or a reset.
This is only related to reading stats. The function avg_delay() also
performs the same divide, but is guarded with a (ca->cnt_rtt > 0) at its
calling point in update_params(). Thus, simply fix tcp_illinois_info().
Function tcp_illinois_info() / get_info() is called without
socket lock. Thus, eliminate any race condition on ca->cnt_rtt
by using a local stack variable. Simply reuse info.tcpv_rttcnt,
as its already set to ca->cnt_rtt.
Function avg_delay() is not affected by this race condition, as
its called with the socket lock.
Cc: Petr Matousek <pmatouse@redhat.com> Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Stephen Hemminger <shemminger@vyatta.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
- if (val != -1 && (val < 1 || val>255))
+ if (val != -1 && (val < 0 || val > 255))
Since there is no reason provided to allow ttl=0, change it back.
Reported-by: nitin padalia <padalia.nitin@gmail.com> Cc: nitin padalia <padalia.nitin@gmail.com> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Stefan Hasko <hasko.stevo@gmail.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Xfrm_dst keeps a reference to ipv4 rtable entries on each
cached bundle. The only way to renew xfrm_dst when the underlying
route has changed, is to implement dst_check for this. This is
what ipv6 side does too.
The problems started after 87c1e12b5eeb7b30b4b41291bef8e0b41fc3dde9
("ipsec: Fix bogus bundle flowi") which fixed a bug causing xfrm_dst
to not get reused, until that all lookups always generated new
xfrm_dst with new route reference and path mtu worked. But after the
fix, the old routes started to get reused even after they were expired
causing pmtu to break (well it would occationally work if the rtable
gc had run recently and marked the route obsolete causing dst_check to
get called).
Signed-off-by: Timo Teras <timo.teras@iki.fi> Acked-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is based on the above, with the addition of verifying blackhole
routes in the same manner.
Fixing the issue with blackhole routes as it was accomplished in mainline
would require pulling in a lot more code, and people were not interested
in pulling in all of the dependencies given the much higher risk of trying
to select the right subset of changes to include. The addition of the
single line of "dst->obsolete = -1;" in ipv4_dst_blackhole() was much
easier to verify, and is in the spirit of the patch in question.
This is the minimal set of changes to fix the bug in question.
A test case is available here :
http://marc.info/?l=linux-netdev&m=135015076708950&w=2
Signed-off-by: Benjamin LaHaise <bcrl@kvack.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Hillf Danton [Fri, 10 Dec 2010 18:54:11 +0000 (18:54 +0000)]
bonding: Fix slave selection bug.
The returned slave is incorrect, if the net device under check is not
charged yet by the master.
Signed-off-by: Hillf Danton <dhillf@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit af3e5bd5f650163c2e12297f572910a1af1b8236) Signed-off-by: Willy Tarreau <w@1wt.eu>
Spanning Tree Protocol packets should have always been marked as
control packets, this causes them to get queued in the high prirority
FIFO. As Radia Perlman mentioned in her LCA talk, STP dies if bridge
gets overloaded and can't communicate. This is a long-standing bug back
to the first versions of Linux bridge.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit 547b4e718115eea74087e28d7fa70aec619200db) Signed-off-by: Willy Tarreau <w@1wt.eu>
Here's a quote of the comment about the BUG macro from asm-generic/bug.h:
Don't use BUG() or BUG_ON() unless there's really no way out; one
example might be detecting data structure corruption in the middle
of an operation that can't be backed out of. If the (sub)system
can somehow continue operating, perhaps with reduced functionality,
it's probably not BUG-worthy.
If you're tempted to BUG(), think again: is completely giving up
really the *only* solution? There are usually better options, where
users don't need to reboot ASAP and can mostly shut down cleanly.
In our case, the status flag of a ring buffer slot is managed from both sides,
the kernel space and the user space. This means that even though the kernel
side might work as expected, the user space screws up and changes this flag
right between the send(2) is triggered when the flag is changed to
TP_STATUS_SENDING and a given skb is destructed after some time. Then, this
will hit the BUG macro. As David suggested, the best solution is to simply
remove this statement since it cannot be used for kernel side internal
consistency checks. I've tested it and the system still behaves /stable/ in
this case, so in accordance with the above comment, we should rather remove it.
Signed-off-by: Daniel Borkmann <daniel.borkmann@tik.ee.ethz.ch> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Eric Dumazet [Thu, 10 Jan 2013 23:26:34 +0000 (15:26 -0800)]
softirq: reduce latencies
In various network workloads, __do_softirq() latencies can be up
to 20 ms if HZ=1000, and 200 ms if HZ=100.
This is because we iterate 10 times in the softirq dispatcher,
and some actions can consume a lot of cycles.
This patch changes the fallback to ksoftirqd condition to :
- A time limit of 2 ms.
- need_resched() being set on current task
When one of this condition is met, we wakeup ksoftirqd for further
softirq processing if we still have pending softirqs.
Using need_resched() as the only condition can trigger RCU stalls,
as we can keep BH disabled for too long.
I ran several benchmarks and got no significant difference in
throughput, but a very significant reduction of latencies (one order
of magnitude) :
In following bench, 200 antagonist "netperf -t TCP_RR" are started in
background, using all available cpus.
Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC
IRQ (hard+soft)
Before patch :
RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET
to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind
RT_LATENCY=550110.424
MIN_LATENCY=146858
MAX_LATENCY=997109
P50_LATENCY=305000
P90_LATENCY=550000
P99_LATENCY=710000
MEAN_LATENCY=376989.12
STDDEV_LATENCY=184046.92
After patch :
RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET
to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind
RT_LATENCY=40545.492
MIN_LATENCY=9834
MAX_LATENCY=78366
P50_LATENCY=33583
P90_LATENCY=59000
P99_LATENCY=69000
MEAN_LATENCY=38364.67
STDDEV_LATENCY=12865.26
Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Miller <davem@davemloft.net> Cc: Tom Herbert <therbert@google.com> Cc: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit c10d73671ad30f54692f7f69f0e09e75d3a8926a) Signed-off-by: Willy Tarreau <w@1wt.eu>
Eric Dumazet [Tue, 5 Mar 2013 07:15:13 +0000 (07:15 +0000)]
net: reduce net_rx_action() latency to 2 HZ
We should use time_after_eq() to get maximum latency of two ticks,
instead of three.
Bug added in commit 24f8b2385 (net: increase receive packet quantum)
Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit d1f41b67ff7735193bc8b418b98ac99a448833e2) Signed-off-by: Willy Tarreau <w@1wt.eu>
When testing NFSv4 at RHEL6 with kernel 2.6.32, I got a kernel panic
at NFS client's __rpc_create_common function.
The panic place is:
rpc_mkpipe
__rpc_lookup_create() <=== find pipefile *idmap*
__rpc_mkpipe() <=== pipefile is *idmap*
__rpc_create_common()
****** BUG_ON(!d_unhashed(dentry)); ****** *panic*
The test is wrong: we can find ourselves with a hashed negative dentry here
if the idmapper tried to look up the file before we got round to creating
it.
Just replace the BUG_ON() with a d_drop(dentry).
[2.6.32 background info from Jonathan below]
> Hi Willy et al,
>
> Please consider
>
> beb0f0a9fba1 kernel panic when mount NFSv4, 2010-12-20
>
> for application to kernel.org's 2.6.32.y and 2.6.34.y trees. The
> patch was applied upstream during the 2.6.38 merge window, so newer
> kernels don't need it.
>
> (Context: <http://bugs.debian.org/695872>.) Tom Downes (cc-ed)
> experienced the bug on a Debian kernel close to 2.6.32.58 and
> confirmed that the patch doesn't seem to hurt.
>
> The patch is part of Fedora 13's 2.6.34-based and Fedora 14's
> 2.6.35-based kernels[1]. It was also included in the RHEL kernel at
> some point between 2.6.32-71.29.1.el6 and 2.6.32-131.0.15.el6[2].
>
> Thoughts of all kinds welcome, as always.
>
> Regards,
> Jonathan
>
> [1] https://bugzilla.redhat.com/673207
> [2] https://oss.oracle.com/git/?p=redpatch.git;a=commit;h=8028cccdc4b1
Reported-by: Mi Jinlong <mijinlong@cn.fujitsu.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from commit beb0f0a9fba1fa98b378329a9a5b0a73f25097ae) Cc: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
Doing this would reliably fail with -EBUSY for me:
# mount /dev/sdb2 /mnt/scratch; umount /mnt/scratch; mkfs.btrfs -f /dev/sdb2
...
unable to open /dev/sdb2: Device or resource busy
because mkfs.btrfs tries to open the device O_EXCL, and somebody still has it.
Using systemtap to track bdev gets & puts shows a kworker thread doing a
blkdev put after mkfs attempts a get; this is left over from the unmount
path:
so unmount might complete before __free_device fires & does its blkdev_put.
Adding an rcu_barrier() to btrfs_close_devices() causes unmount to wait
until all blkdev_put()s are done, and the device is truly free once
unmount completes.
Signed-off-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Josef Bacik <jbacik@fusionio.com> Signed-off-by: Chris Mason <chris.mason@fusionio.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
The utf8s_to_utf16s conversion routine needs to be improved. Unlike
its utf16s_to_utf8s sibling, it doesn't accept arguments specifying
the maximum length of the output buffer or the endianness of its
16-bit output.
This patch (as1501) adds the two missing arguments, and adjusts the
only two places in the kernel where the function is called. A
follow-on patch will add a third caller that does utilize the new
capabilities.
The two conversion routines are still annoyingly inconsistent in the
way they handle invalid byte combinations. But that's a subject for a
different patch.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu> CC: Clemens Ladisch <clemens@ladisch.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
[bwh: Bakckported to 2.6.32: drop Hyper-V change] Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Willy Tarreau <w@1wt.eu>
When it goes to error through line 144, the memory allocated to *devname is
not freed, and the caller doesn't free it either in line 250. So we free the
memroy of *devname in function cifs_compose_mount_options() when it goes to
error.
Signed-off-by: Cong Ding <dinggnu@gmail.com> Reviewed-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <smfrench@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
For large UDF filesystems with 512-byte blocks the number of necessary
bitmap blocks is larger than 2^16 so s_nr_groups in udf_bitmap overflows
(the number will overflow for filesystems larger than 128 GB with
512-byte blocks). That results in ENOSPC errors despite the filesystem
has plenty of free space.
Fix the problem by changing s_nr_groups' type to 'int'. That is enough
even for filesystems 2^32 blocks (UDF maximum) and 512-byte blocksize.
Reported-and-tested-by: v10lator@myway.de Signed-off-by: Jan Kara <jack@suse.cz> Cc: Jim Trigg <jtrigg@spamcop.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
When trying to mount a file system which does not contain a journal,
but which does have a orphan list containing an inode which needs to
be truncated, the mount call with hang forever in
ext4_orphan_cleanup() because ext4_orphan_del() will return
immediately without removing the inode from the orphan list, leading
to an uninterruptible loop in kernel code which will busy out one of
the CPU's on the system.
This can be trivially reproduced by trying to mount the file system
found in tests/f_orphan_extents_inode/image.gz from the e2fsprogs
source tree. If a malicious user were to put this on a USB stick, and
mount it on a Linux desktop which has automatic mounts enabled, this
could be considered a potential denial of service attack. (Not a big
deal in practice, but professional paranoids worry about such things,
and have even been known to allocate CVE numbers for such problems.)
Instead of checking whether the handle is valid, we check if journal
is enabled. This avoids taking the s_orphan_lock mutex in all cases
when there is no journal in use, including the error paths where
ext4_orphan_del() is called with a handle set to NULL.
Jamie Iles [Thu, 21 Feb 2013 10:18:51 +0000 (10:18 +0000)]
CVE-2012-4508 kernel: ext4: AIO vs fallocate stale data exposure
CVE-2012-4508 kernel: ext4: AIO vs fallocate stale data exposure
[dannf: backported to Debian's 2.6.32]
According to Ben :
> The original upstream commits were c278531d39f3158bfee93dc67da0b77e09776de2,
> 60d4616f3dc63371b3dc367e5e88fd4b4f037f65 and (most importantly)
> dee1f973ca341c266229faa5a1a5bb268bed3531 by Dmitry Monakhov
> <dmonakhov@openvz.org>. They were backported into the RHEL 6 kernel by
> Lukas Czerner, according to its changelog. Dann got this version from
> Oracle's redpatch repository, where, if I understand rightly, Jamie Iles
> attempted to regenerate Lukas's patch(es).
In the case where we are allocating for a non-extent file,
we must limit the groups we allocate from to those below
2^32 blocks, and ext4_mb_regular_allocator() attempts to
do this initially by putting a cap on ngroups for the
subsequent search loop.
However, the initial target group comes in from the
allocation context (ac), and it may already be beyond
the artificially limited ngroups. In this case,
the limit
if (group == ngroups)
group = 0;
at the top of the loop is never true, and the loop will
run away.
Catch this case inside the loop and reset the search to
start at group 0.
Commit c278531d39 added a warning when ext4_flush_unwritten_io() is
called without i_mutex being taken. It had previously not been taken
during orphan cleanup since races weren't possible at that point in
the mount process, but as a result of this c278531d39, we will now see
a kernel WARN_ON in this case. Take the i_mutex in
ext4_orphan_cleanup() to suppress this warning.
Reported-by: Alexander Beregalov <a.beregalov@gmail.com> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu> Reviewed-by: Zheng Liu <wenqing.lz@taobao.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Code tracking when transaction needs to be committed on fdatasync(2) forgets
to handle a situation when only inode's i_size is changed. Thus in such
situations fdatasync(2) doesn't force transaction with new i_size to disk
and that can result in wrong i_size after a crash.
Fix the issue by updating inode's i_datasync_tid whenever its size is
updated.
Reported-by: Kristian Nielsen <knielsen@knielsen-hq.org> Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
ext4_special_inode_operations have their own ifdef CONFIG_EXT4_FS_XATTR
to mask those methods. And ext4_iget also always sets it, so there is
an inconsistency.
Proper block swap for inodes with full journaling enabled is
truly non obvious task. In order to be on a safe side let's
explicitly disable it for now.
Kazuya Mio reported that he was able to hit BUG_ON(next == lblock)
in ext4_ext_put_gap_in_cache() while creating a sparse file in extent
format and fill the tail of file up to its end. We will hit the BUG_ON
when we write the last block (2^32-1) into the sparse file.
The root cause of the problem lies in the fact that we specifically set
s_maxbytes so that block at s_maxbytes fit into on-disk extent format,
which is 32 bit long. However, we are not storing start and end block
number, but rather start block number and length in blocks. It means
that in order to cover extent from 0 to EXT_MAX_BLOCK we need
EXT_MAX_BLOCK+1 to fit into len (because we counting block 0 as well) -
and it does not.
The only way to fix it without changing the meaning of the struct
ext4_extent members is, as Kazuya Mio suggested, to lower s_maxbytes
by one fs block so we can cover the whole extent we can get by the
on-disk extent format.
Also in many places EXT_MAX_BLOCK is used as length instead of maximum
logical block number as the name suggests, it is all a bit messy. So
this commit renames it to EXT_MAX_BLOCKS and change its usage in some
places to actually be maximum number of blocks in the extent.
The bug which this commit fixes can be reproduced as follows:
Jan Kara [Tue, 3 May 2011 15:05:55 +0000 (11:05 -0400)]
ext4: Fix fs corruption when make_indexed_dir() fails
When make_indexed_dir() fails (e.g. because of ENOSPC) after it has
allocated block for index tree root, we did not properly mark all
changed buffers dirty. This lead to only some of these buffers being
written out and thus effectively corrupting the directory.
Fix the issue by marking all changed data dirty even in the error
failure case.
Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
(cherry picked from commit 7ad8e4e6ae2a7c95445ee1715b1714106fb95037) Signed-off-by: Willy Tarreau <w@1wt.eu>
Commit 09e05d48 introduced a wait for transaction commit into
journal_unmap_buffer() in the case we are truncating a buffer undergoing commit
in the page stradding i_size on a filesystem with blocksize < pagesize. Sadly
we forgot to drop buffer lock before waiting for transaction commit and thus
deadlock is possible when kjournald wants to lock the buffer.
Fix the problem by dropping the buffer lock before waiting for transaction
commit. Since we are still holding page lock (and that is OK), buffer cannot
disappear under us.
Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Jan Kara [Wed, 11 Jul 2012 21:16:25 +0000 (23:16 +0200)]
jbd: Fix assertion failure in commit code due to lacking transaction credits
ext3 users of data=journal mode with blocksize < pagesize were occasionally
hitting assertion failure in journal_commit_transaction() checking whether the
transaction has at least as many credits reserved as buffers attached. The
core of the problem is that when a file gets truncated, buffers that still need
checkpointing or that are attached to the committing transaction are left with
buffer_mapped set. When this happens to buffers beyond i_size attached to a
page stradding i_size, subsequent write extending the file will see these
buffers and as they are mapped (but underlying blocks were freed) things go
awry from here.
The assertion failure just coincidentally (and in this case luckily as we would
start corrupting filesystem) triggers due to journal_head not being properly
cleaned up as well.
Under some rare circumstances this bug could even hit data=ordered mode users.
There the assertion won't trigger and we would end up corrupting the
filesystem.
We fix the problem by unmapping buffers if possible (in lots of cases we just
need a buffer attached to a transaction as a place holder but it must not be
written out anyway). And in one case, we just have to bite the bullet and wait
for transaction commit to finish.
Reviewed-by: Josef Bacik <jbacik@fusionio.com> Signed-off-by: Jan Kara <jack@suse.cz>
(cherry picked from commit 09e05d4805e6c524c1af74e524e5d0528bb3fef3) Signed-off-by: Willy Tarreau <w@1wt.eu>
Jan Kara [Tue, 16 Feb 2010 19:37:12 +0000 (20:37 +0100)]
jbd: Delay discarding buffers in journal_unmap_buffer
Delay discarding buffers in journal_unmap_buffer until
we know that "add to orphan" operation has definitely been
committed, otherwise the log space of committing transation
may be freed and reused before truncate get committed, updates
may get lost if crash happens.
This patch is a backport of JBD2 fix by dingdinghua <dingdinghua@nrchpc.ac.cn>.
The tmpfs remount logic preserves filesystem mempolicy if the mpol=M
option is not specified in the remount request. A new policy can be
specified if mpol=M is given.
Before this patch remounting an mpol bound tmpfs without specifying
mpol= mount option in the remount request would set the filesystem's
mempolicy object to a freed mempolicy object.
To reproduce the problem boot a DEBUG_PAGEALLOC kernel and run:
# mkdir /tmp/x
# mount -t tmpfs -o size=100M,mpol=interleave nodev /tmp/x
Non-debug kernels will not crash immediately because referencing the
dangling mpol will not cause a fault. Instead the filesystem will
reference a freed mempolicy object, which will cause unpredictable
behavior.
The problem boils down to a dropped mpol reference below if
shmem_parse_options() does not allocate a new mpol:
The warning check for duplicate sysfs entries can cause a buffer overflow
when printing the warning, as strcat() doesn't check buffer sizes.
Use strlcat() instead.
Since strlcat() doesn't return a pointer to the passed buffer, unlike
strcat(), I had to convert the nested concatenation in sysfs_add_one() to
an admittedly more obscure comma operator construct, to avoid emitting code
for the concatenation if CONFIG_BUG is disabled.
projects / thirdparty / kernel / stable.git / log
