Memory leak in the seq_file implementation in the SCSI procfs interface (sg.c) in Linux kernel 2.6.13 and earlier allows local users to cause a denial of service (memory consumption) via certain repeated reads from the /proc/scsi/sg/devices file, which is not properly handled when the next() iterator returns NULL or an error.

Several local and remote vulnerabilities have been discovered in the Linux kernel that may lead to a denial of service or the execution of arbitrary code. The Common Vulnerabilities and Exposures project identifies the following problems :

  - CVE-2004-1017     Multiple overflows exist in the io_edgeport driver which     might be usable as a denial of service attack vector.

  - CVE-2005-0124     Bryan Fulton reported a bounds checking bug in the     coda_pioctl function which may allow local users to     execute arbitrary code or trigger a denial of service     attack.

  - CVE-2005-0449     An error in the skb_checksum_help() function from the     netfilter framework has been discovered that allows the     bypass of packet filter rules or a denial of service     attack.

  - CVE-2005-2457     Tim Yamin discovered that insufficient input validation     in the zisofs driver for compressed ISO file systems     allows a denial of service attack through maliciously     crafted ISO images.

  - CVE-2005-2490     A buffer overflow in the sendmsg() function allows local     users to execute arbitrary code.

  - CVE-2005-2555     Herbert Xu discovered that the setsockopt() function was     not restricted to users/processes with the CAP_NET_ADMIN     capability. This allows attackers to manipulate IPSEC     policies or initiate a denial of service attack. 

  - CVE-2005-2709     Al Viro discovered a race condition in the /proc     handling of network devices. A (local) attacker could     exploit the stale reference after interface shutdown to     cause a denial of service or possibly execute code in     kernel mode.

  - CVE-2005-2800     Jan Blunck discovered that repeated failed reads of     /proc/scsi/sg/devices leak memory, which allows a denial     of service attack.

  - CVE-2005-2973     Tetsuo Handa discovered that the udp_v6_get_port()     function from the IPv6 code can be forced into an     endless loop, which allows a denial of service attack.

  - CVE-2005-3044     Vasiliy Averin discovered that the reference counters     from sockfd_put() and fput() can be forced into     overlapping, which allows a denial of service attack     through a NULL pointer dereference.

  - CVE-2005-3053     Eric Dumazet discovered that the set_mempolicy() system     call accepts a negative value for its first argument,     which triggers a BUG() assert. This allows a denial of     service attack.

  - CVE-2005-3055     Harald Welte discovered that if a process issues a USB     Request Block (URB) to a device and terminates before     the URB completes, a stale pointer would be     dereferenced. This could be used to trigger a denial of     service attack.

  - CVE-2005-3180     Pavel Roskin discovered that the driver for Orinoco     wireless cards clears its buffers insufficiently. This     could leak sensitive information into user space.

  - CVE-2005-3181     Robert Derr discovered that the audit subsystem uses an     incorrect function to free memory, which allows a denial     of service attack.

  - CVE-2005-3257     Rudolf Polzer discovered that the kernel improperly     restricts access to the KDSKBSENT ioctl, which can     possibly lead to privilege escalation.

  - CVE-2005-3356     Doug Chapman discovered that the mq_open syscall can be     tricked into decrementing an internal counter twice,     which allows a denial of service attack through a kernel     panic.

  - CVE-2005-3358     Doug Chapman discovered that passing a zero bitmask to     the set_mempolicy() system call leads to a kernel panic,     which allows a denial of service attack.

  - CVE-2005-3783     The ptrace code using CLONE_THREAD didn't use the thread     group ID to determine whether the caller is attaching to     itself, which allows a denial of service attack.

  - CVE-2005-3784     The auto-reaping of child processes functionality     included ptraced-attached processes, which allows denial     of service through dangling references.

  - CVE-2005-3806     Yen Zheng discovered that the IPv6 flow label code     modified an incorrect variable, which could lead to     memory corruption and denial of service.

  - CVE-2005-3847     It was discovered that a threaded real-time process,     which is currently dumping core can be forced into a     dead-lock situation by sending it a SIGKILL signal,     which allows a denial of service attack. 

  - CVE-2005-3848     Ollie Wild discovered a memory leak in the     icmp_push_reply() function, which allows denial of     service through memory consumption.

  - CVE-2005-3857     Chris Wright discovered that excessive allocation of     broken file lock leases in the VFS layer can exhaust     memory and fill up the system logging, which allows     denial of service.

  - CVE-2005-3858     Patrick McHardy discovered a memory leak in the     ip6_input_finish() function from the IPv6 code, which     allows denial of service.

  - CVE-2005-4605     Karl Janmar discovered that a signedness error in the     procfs code can be exploited to read kernel memory,     which may disclose sensitive information.

  - CVE-2005-4618     Yi Ying discovered that sysctl does not properly enforce     the size of a buffer, which allows a denial of service     attack.

  - CVE-2006-0095     Stefan Rompf discovered that dm_crypt does not clear an     internal struct before freeing it, which might disclose     sensitive information.

  - CVE-2006-0096     It was discovered that the SDLA driver's capability     checks were too lax for firmware upgrades.

  - CVE-2006-0482     Ludovic Courtes discovered that get_compat_timespec()     performs insufficient input sanitizing, which allows a     local denial of service attack.

  - CVE-2006-1066     It was discovered that ptrace() on the ia64 architecture     allows a local denial of service attack, when preemption     is enabled.

Updated kernel packages that fix several security issues in the Red Hat Enterprise Linux 4 kernel are now available.

This security advisory has been rated as having important security impact by the Red Hat Security Response Team.

The Linux kernel handles the basic functions of the operating system.

These new kernel packages contain fixes for the security issues described below :

  - a flaw in network IGMP processing that a allowed a     remote user on the local network to cause a denial of     service (disabling of multicast reports) if the system     is running multicast applications (CVE-2002-2185,     moderate)

  - a flaw which allowed a local user to write to firmware     on read-only opened /dev/cdrom devices (CVE-2004-1190,     moderate)

  - a flaw in gzip/zlib handling internal to the kernel that     may allow a local user to cause a denial of service     (crash) (CVE-2005-2458, low)

  - a flaw in procfs handling during unloading of modules     that allowed a local user to cause a denial of service     or potentially gain privileges (CVE-2005-2709, moderate)

  - a flaw in the SCSI procfs interface that allowed a local     user to cause a denial of service (crash)     (CVE-2005-2800, moderate)

  - a flaw in 32-bit-compat handling of the TIOCGDEV ioctl     that allowed a local user to cause a denial of service     (crash) (CVE-2005-3044, important)

  - a race condition when threads share memory mapping that     allowed local users to cause a denial of service     (deadlock) (CVE-2005-3106, important)

  - a flaw when trying to mount a non-hfsplus filesystem     using hfsplus that allowed local users to cause a denial     of service (crash) (CVE-2005-3109, moderate)

  - a minor info leak with the get_thread_area() syscall     that allowed a local user to view uninitialized kernel     stack data (CVE-2005-3276, low)

  - a flaw in mq_open system call that allowed a local user     to cause a denial of service (crash) (CVE-2005-3356,     important)

  - a flaw in set_mempolicy that allowed a local user on     some 64-bit architectures to cause a denial of service     (crash) (CVE-2005-3358, important)

  - a flaw in the auto-reap of child processes that allowed     a local user to cause a denial of service (crash)     (CVE-2005-3784, important)

  - a flaw in the IPv6 flowlabel code that allowed a local     user to cause a denial of service (crash)     (CVE-2005-3806, important)

  - a flaw in network ICMP processing that allowed a local     user to cause a denial of service (memory exhaustion)     (CVE-2005-3848, important)

  - a flaw in file lease time-out handling that allowed a     local user to cause a denial of service (log file     overflow) (CVE-2005-3857, moderate)

  - a flaw in network IPv6 xfrm handling that allowed a     local user to cause a denial of service (memory     exhaustion) (CVE-2005-3858, important)

  - a flaw in procfs handling that allowed a local user to     read kernel memory (CVE-2005-4605, important)

All Red Hat Enterprise Linux 4 users are advised to upgrade their kernels to the packages associated with their machine architectures and configurations as listed in this erratum.

Oleg Nesterov discovered a local Denial of Service vulnerability in the timer handling. When a non group-leader thread called exec() to execute a different program while an itimer was pending, the timer expiry would signal the old group leader task, which did not exist any more. This caused a kernel panic. This vulnerability only affects Ubuntu 5.04. (CAN-2005-1913)

Al Viro discovered that the sendmsg() function did not sufficiently validate its input data. By calling sendmsg() and at the same time modifying the passed message in another thread, he could exploit this to execute arbitrary commands with kernel privileges. This only affects the amd64 bit platform. (CAN-2005-2490)

Al Viro discovered a vulnerability in the raw_sendmsg() function. By calling this function with specially crafted arguments, a local attacker could either read kernel memory contents (leading to information disclosure) or manipulate the hardware state by reading certain IO ports. This vulnerability only affects Ubuntu 5.04.
(CAN-2005-2492)

Jan Blunck discovered a Denial of Service vulnerability in the procfs interface of the SCSI driver. By repeatedly reading /proc/scsi/sg/devices, a local attacker could eventually exhaust kernel memory. (CAN-2005-2800)

A flaw was discovered in the handling of extended attributes on ext2 and ext3 file systems. Under certain condidions, this could prevent the enforcement of Access Control Lists, which eventually could lead to information disclosure, unauthorized program execution, or unauthorized data modification. This does not affect the standard Unix permissions. (CAN-2005-2801)

Chad Walstrom discovered a Denial of Service in the ipt_recent module, which can be used in netfilter (Firewall configuration). A remote attacker could exploit this to crash the kernel by sending certain packets (such as an SSH brute-force attack) to a host which uses the 'recent' module. (CAN-2005-2802).

Note that Tenable Network Security has extracted the preceding description block directly from the Ubuntu security advisory. Tenable has attempted to automatically clean and format it as much as possible without introducing additional issues.

Multiple vulnerabilities in the Linux 2.6 kernel have been discovered and corrected in this update :

An integer overflow in vc_resize (CVE-2004-1333).

A race condition in the sysfs_read_file and sysfs_write_file functions in 2.6.10 and earlier allows local users to read kernel memory and cause a DoS (crash) via large offsets in sysfs files (CVE-2004-2302).

An integer signedness error in scsi_ioctl.c (CVE-2005-0180).

Netfilter allows a local user to cause a DoS (memory consumption) via certain packet fragments that are reassembled twice, which causes a data structure to be allocated twice (CVE-2005-0210).

A DoS in pkt_ioctl in pktcdvc.c (CVE-2005-1589).

An array index overflow in the xfrm_sk_policy_insert function in xfrm_user.c allows local users to cause a DoS (oops or deadlock) and possibly execute arbitrary code (CVE-2005-2456).

The zisofs driver in versions prior to 2.6.12.5 allows local users and remove attackers to cause a DoS (crash) via a crafted compressed ISO filesystem (CVE-2005-2457).

inflate.c in the zlib routines in versions prior to 2.6.12.5 allow remove attackers to cause a DoS (crash) via a compressed file with 'improper tables' (CVE-2005-2458).

The huft_build function in inflate.c in the zlib routines in versions prior to 2.6.12.5 returns the wrong value, allowing remote attackers to cause a DoS (crash) via a certain compressed file that leads to a NULL pointer dereference (CVE-2005-2459).

A stack-based buffer overflow in the sendmsg function call in versions prior to 2.6.13.1 allow local users to execute arbitrary code by calling sendmsg and modifying the message contents in another thread (CVE-2005-2490).

vlan_dev.c in version 2.6.8 allows remote attackers to cause a DoS (oops from null dereference) via certain UDP packets that lead to a function call with the wrong argument (CVE-2005-2548).

The kernel does not properly restrict socket policy access to users with the CAP_NET_ADMIN capability, which could allow local users to conduct unauthorized activities via ipv4/ip_sockglue.c and ipv6/ipv6_sockglue.c (CVE-2005-2555).

A memory leak in the seq_file implementation in the SCSI procfs interface (sg.c) in 2.6.13 and earlier allows a local user to cause a DoS (memory consumption) via certain repeated reads from /proc/scsi/gs/devices file which is not properly handled when the next() interator returns NULL or an error (CVE-2005-2800).

xattr.c in the ext2 and ext3 file system code does not properly compare the name_index fields when sharing xattr blocks which could prevent ACLs from being applied (CVE-2005-2801).

The ipt_recent module in versions prior to 2.6.12 when running on 64bit processors allows remote attackers to cause a DoS (kernel panic) via certain attacks such as SSH brute force (CVE-2005-2872).

The ipt_recent module in versions prior to 2.6.12 does not properly perform certain tests when the jiffies value is greater than LONG_MAX, which can cause ipt_recent netfilter rules to block too early (CVE-2005-2873).

Multiple vulnerabilities in versions prior to 2.6.13.2 allow local users to cause a DoS (oops from null dereference) via fput in a 32bit ioctl on 64-bit x86 systems or sockfd_put in the 32-bit routing_ioctl function on 64-bit systems (CVE-2005-3044).

The sys_set_mempolicy function in mempolicy.c allows local users to cause a DoS via a negative first argument (CVE-2005-3053).

Versions 2.6.8 to 2.6.14-rc2 allow local users to cause a DoS (oops) via a userspace process that issues a USB Request Block (URB) to a USB device and terminates before the URB is finished, which leads to a stale pointer reference (CVE-2005-3055).

The Orinoco driver in 2.6.13 and earlier does not properly clear memory from a previously used packet whose length is increased, allowing remote attackers to obtain sensitive information (CVE-2005-3180).

Kernels 2.6.13 and earlier, when CONFIG_AUDITSYSCALL is enabled, use an incorrect function to free names_cache memory, preventing the memory from being tracked by AUDITSYSCALL code and leading to a memory leak (CVE-2005-3181).

The VT implementation in version 2.6.12 allows local users to use certain IOCTLs on terminals of other users and gain privileges (CVE-2005-3257).

Exec does not properly clear posix-timers in multi-threaded environments, which result in a resource leak and could allow a large number of multiple local users to cause a DoS by using more posix- timers than specified by the quota for a single user (CVE-2005-3271).

The rose_rt_ioctl function rose_route.c in versions prior to 2.6.12 does not properly verify the ndigis argument for a new route, allowing an attacker to trigger array out-of-bounds errors with a large number of digipeats (CVE-2005-3273).

A race condition in ip_vs_conn_flush in versions prior to 2.6.13, when running on SMP systems, allows local users to cause a DoS (null dereference) by causing a connection timer to expire while the connection table is being flushed before the appropriate lock is acquired (CVE-2005-3274).

The NAT code in versions prior to 2.6.13 incorrectly declares a variable to be static, allowing remote attackers to cause a DoS (memory corruption) by causing two packets for the same protocol to be NATed at the same time (CVE-2005-3275).

The sys_get_thread_area function in process.c in versions prior to 2.6.12.4 and 2.6.13 does not clear a data structure before copying it to userspace, which may allow a user process to obtain sensitive information (CVE-2005-3276).

The following non-security fixes are also applied :

Driver updates were made to the aic97xx and sata_sil modules.

Support was added for ATI ipx400 chipsets, for IDE and sound.

A build problem with icecream on the x86_64 platform was fixed.

The pin1 APIC timer on RS480-based motherboards was disabled.

The provided packages are patched to fix these vulnerabilities. All users are encouraged to upgrade to these updated kernels.

To update your kernel, please follow the directions located at :

http://www.mandriva.com/en/security/kernelupdate

The remote CentOS system is missing a security update which has been documented in Red Hat advisory RHSA-2006-0101.

ID	Name	Product	Family	Severity
22559	Debian DSA-1017-1 : kernel-source-2.6.8 - several vulnerabilities	Nessus	Debian Local Security Checks	critical
21977	CentOS 4 : kernel (CESA-2006:0101)	Nessus	CentOS Local Security Checks	high
20732	RHEL 4 : kernel (RHSA-2006:0101)	Nessus	Red Hat Local Security Checks	high
20588	Ubuntu 4.10 / 5.04 : linux-source-2.6.10, linux-source-2.6.8.1 vulnerabilities (USN-178-1)	Nessus	Ubuntu Local Security Checks	medium
20450	Mandrake Linux Security Advisory : kernel (MDKSA-2005:219)	Nessus	Mandriva Local Security Checks	high
801413	CentOS RHSA-2006-0101 Security Check	Log Correlation Engine	Generic	high

Detections

Analytics

CVE-2005-2800

medium

Tenable Plugins

critical

high

high

medium

high

high