Rejected reason: Red Hat Product Security determined this is not a security vulnerability. BusyBox internal TLS is intentionally minimal and documents in FEATUREWGETHTTPS that it does not verify peer identity or provide full TLS security; missing Finished verifydata checking is consistent with that documented design and does not cross a separate security boundary warranting a CVE.

A flaw was found in busybox. The tar applet defers creation of symlink and hardlink entries whose targets look unsafe (absolute paths, hardlinks, or targets containing '..') until after all archive members have been processed, in order to avoid a same-archive time-of-check-to-time-of-use issue. However, when those deferred links are finally created, the code performs the underlying symlink() or link() system call directly, without re-validating that the resolved destination parent remains inside the extraction directory. By chaining an immediately-created symlink with a deferred symlink whose target is exactly '..' (a value not caught by the separate prefix-based sanitizer applied to member names, which only strips leading '../' and embedded '/../' sequences), an attacker can make a path that appears to be inside the extraction directory resolve outside it once the deferred link is created. Combined with a deferred hardlink in the same archive, this creates a new file outside the extraction directory; reusing the same destination directory across two archives allows an existing external file to be deleted and replaced with attacker-controlled content, including attacker-controlled ownership and permissions for privileged extraction, since the second archive's replacement member is an ordinary file entry (not a hardlink) and is subject to busybox tar's normal chown()/chmod() restoration from the archive header. The one-archive hardlink primitive requires the internal source file and the external target to reside on the same filesystem; the two-archive primitive requires the extraction directory to be reused across two extraction invocations.

A flaw was found in busybox. The tar applet defers creation of symlink and hardlink entries whose targets look unsafe (absolute paths, hardlinks, or targets containing '..') until after all archive members have been processed, in order to avoid a same-archive time-of-check-to-time-of-use issue. However, when those deferred links are finally created, the code performs the underlying symlink() or link() system call directly, without re-validating that the resolved destination parent remains inside the extraction directory. By chaining an immediately-created symlink with a deferred symlink whose target is exactly '..' (a value not caught by the separate prefix-based sanitizer applied to member names, which only strips leading '../' and embedded '/../' sequences), an attacker can make a path that appears to be inside the extraction directory resolve outside it once the deferred link is created. Combined with a deferred hardlink in the same archive, this creates a new file outside the extraction directory; reusing the same destination directory across two archives allows an existing external file to be deleted and replaced with attacker-controlled content, including attacker-controlled ownership and permissions for privileged extraction, since the second archive's replacement member is an ordinary file entry (not a hardlink) and is subject to busybox tar's normal chown()/chmod() restoration from the archive header. The one-archive hardlink primitive requires the internal source file and the external target to reside on the same filesystem; the two-archive primitive requires the extraction directory to be reused across two extraction invocations.

This vulnerability allows remote attackers to create arbitrary files on affected installations of BusyBox. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file. The specific flaw exists within the libarchive component. The issue results from the lack of proper validation of a user-supplied path prior to using it in file operations. An attacker can leverage this vulnerability to create files in the context of the current user.

This vulnerability allows remote attackers to create arbitrary files on affected installations of BusyBox. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file. The specific flaw exists within the libarchive component. The issue results from the lack of proper validation of a user-supplied path prior to using it in file operations. An attacker can leverage this vulnerability to create files in the context of the current user.

BusyBox libarchive Symlink Directory Traversal Arbitrary File Creation Vulnerability. This vulnerability allows remote attackers to create arbitrary files on affected installations of BusyBox. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.

The specific flaw exists within the libarchive component. The issue results from the lack of proper validation of a user-supplied path prior to using it in file operations. An attacker can leverage this vulnerability to create files in the context of the current user. Was ZDI-CAN-27556.

A flaw was found in BusyBox's dpkg-deb archive extraction. The ar archive handler flushes deferred symlinks after each archive member, unlike the tar, cpio, and unzip handlers which flush once after the last member. A specially crafted .deb archive with two data.tar members can exploit this ordering: the first member declares a symlink pointing to an absolute path (e.g., '/'), which is deferred as unsafe but flushed after the first member. The second member declares a regular file whose path traverses the now-live symlink, resulting in a file being written outside the intended extraction directory. This could allow an attacker to write arbitrary files to arbitrary locations when a user extracts a crafted .deb package.

BusyBox provides lightweight Unix utilities as a single binary. A path traversal vulnerability exists in the dpkg-deb extraction path in archival/libarchive/getheaderar.c. The ar container handler flushes deferred symlinks (via createlinksfromlist() at line 133) inside the per-member loop, rather than once after all members are processed. The tar, cpio, and unzip front-ends correctly flush deferred symlinks after the main loop (tar.c:1269, cpio.c:578, unzip.c:1059), but the ar path does not follow this pattern.

The deferred-symlink mechanism (unsafesymlinktarget.c:13) exists to prevent directory traversal: symlinks with absolute targets or '..' components are deferred until after all archive members are extracted, so that no subsequent member can traverse them. By flushing per ar member, the ar path breaks this invariant: a symlink deferred from the first data.tar member is placed on disk before the second data.tar member is parsed.

A crafted .deb containing two uncompressed data.tar ar members can exploit this: the first declares a symlink whose target is the filesystem root, and the second declares a regular file whose path is resolved through that live symlink, writing a file outside the extraction root.

Constraints: only uncompressed data.tar members work (compressed members cause setupunziponfd() to replace srcfd with a pipe, making the subsequent lseek fail). filteracceptlistreassign() does not remove matched entries from the accept list, so duplicate member names are accepted. BBSUIDDROP is set for the dpkg-deb applet, preventing setuid escalation.

The reporter states the bug was introduced by commit 8c24af9dc (2018-05-22), affecting versions 1290 through current master. No fixing commit exists. Reproduced on Fedora busybox-1.37.0-7.fc44, Debian 13, and Ubuntu 24.04 LTS.

Reported by Shubham Raj, Causal Security. PSIRT ticket: PSIRTSUPT-23314.

A vulnerability exists in the BusyBox dpkg applet in archival/dpkg.c. The writestatusfile() function does not reset the fieldstart cursor between package stanzas. When two packages are removed in a single invocation and the first stanza is larger than the second, the stale cursor causes readpackagefield() to read past the second stanza's buffer.

Observable consequences include: silent data corruption (the second package stanza vanishes from the status file), potential heap information leakage, and crashes under ASan.

A vulnerability exists in the BusyBox TLS server implementation in networking/tls.c. The getclienthello() function reads p[0] (the compression-methods length byte) after consuming the cipher suite list without verifying that len > 0. A crafted ClientHello that ends immediately after the cipher suite list triggers a 1-byte out-of-bounds heap read.

This is a pre-authentication vulnerability. The one-byte OOB read can potentially leak a single byte of adjacent heap memory or cause a crash in memory-safety-hardened builds.

A vulnerability exists in the BusyBox TLS server implementation in networking/tls.c. The getclienthello() function reads p[0] (the compression-methods length byte) after consuming the cipher suite list without verifying that len > 0. A crafted ClientHello that ends immediately after the cipher suite list triggers a 1-byte out-of-bounds heap read.

This is a pre-authentication vulnerability. The one-byte OOB read can potentially leak a single byte of adjacent heap memory or cause a crash in memory-safety-hardened builds.

A vulnerability exists in the BusyBox passwd and group file parser in libpwdgrp/pwdgrp.c. The tokenize() function trims trailing whitespace via overlappingstrcpy() but continues to reference the stale end pointer, causing tokenizeend to be too large. This under-allocates in parsecommon(), and converttostruct() then writes grmem[] pointers past the end of the heap allocation.

Triggering the vulnerability requires injecting malformed entries with whitespace before commas into /etc/group, then invoking any applet that calls getgrnam().

A vulnerability exists in the BusyBox httpd applet's Basic Authentication password verification logic in networking/httpd.c. The code uses if (passwd[0] == '$' && isdigit(passwd[1])) to distinguish hashed passwords from plaintext. Yescrypt hashes begin with $y$, and because 'y' does not satisfy isdigit(), the detection logic fails. The code falls through to the plaintext strcmp branch, making the stored hash string itself the "password."

Supplying the literal hash string as the password results in HTTP 200, while the actual plaintext password is rejected with HTTP 401. Exploitation requires the attacker to first obtain the hash string from httpd.conf.

A vulnerability exists in the BusyBox httpd applet's Basic Authentication password verification logic in networking/httpd.c. The code uses if (passwd[0] == '$' && isdigit(passwd[1])) to distinguish hashed passwords from plaintext. Yescrypt hashes begin with $y$, and because 'y' does not satisfy isdigit(), the detection logic fails. The code falls through to the plaintext strcmp branch, making the stored hash string itself the "password."

Supplying the literal hash string as the password results in HTTP 200, while the actual plaintext password is rejected with HTTP 401. Exploitation requires the attacker to first obtain the hash string from httpd.conf.

A vulnerability exists in the BusyBox passwd and group file parser in libpwdgrp/pwdgrp.c. The tokenize() function trims trailing whitespace via overlappingstrcpy() but continues to reference the stale end pointer, causing tokenizeend to be too large. This under-allocates in parsecommon(), and converttostruct() then writes grmem[] pointers past the end of the heap allocation.

Triggering the vulnerability requires injecting malformed entries with whitespace before commas into /etc/group, then invoking any applet that calls getgrnam().

A vulnerability exists in the BusyBox mount applet's non-root option filter in util-linux/mount.c. The filter uses cmdoptflags & ~MSSILENT to test for dangerous options, but clearing-mask options (suid, dev, exec) AND bits off rather than OR bits on, evaluating to zero and passing the check unconditionally. This allows a non-root user to strip nosuid, nodev, and noexec restrictions from fstab-defined user mounts.

The finding was reproduced only after artificially setting the setuid bit on BusyBox. Only the filter bypass was demonstrated, not actual privilege escalation.

A vulnerability was found in the BusyBox dpkg implementation (archival/dpkg.c). The restart path in readpackagefield() incorrectly handles the case where a field has a name but an empty value followed by a NUL terminator, stepping one byte past the buffer and reading adjacent heap memory. This causes a bus error or segfault.

A vulnerability was found in the BusyBox dpkg implementation (archival/dpkg.c). The restart path in readpackagefield() incorrectly handles the case where a field has a name but an empty value followed by a NUL terminator, stepping one byte past the buffer and reading adjacent heap memory. This causes a bus error or segfault.

A vulnerability was found in the BusyBox romfs filesystem volume identification module (util-linux/volumeid/romfs.c). When parsing a romfs superblock, the code calls strlen() on attacker-controlled volume name metadata without any bounds checking. The resulting length is passed to memcpy() via volumeidsetlabelstring(), which copies the data into a fixed-size label field of approximately 65 bytes.

A crafted romfs filesystem image can contain a volume name of up to approximately 4080 bytes. When such an image is processed by blkid or findfs, the unbounded memcpy() writes far beyond the label buffer boundary, corrupting adjacent heap memory.

A vulnerability was found in the BusyBox romfs filesystem volume identification module (util-linux/volumeid/romfs.c). When parsing a romfs superblock, the code calls strlen() on attacker-controlled volume name metadata without any bounds checking. The resulting length is passed to memcpy() via volumeidsetlabelstring(), which copies the data into a fixed-size label field of approximately 65 bytes.

A crafted romfs filesystem image can contain a volume name of up to approximately 4080 bytes. When such an image is processed by blkid or findfs, the unbounded memcpy() writes far beyond the label buffer boundary, corrupting adjacent heap memory.

A vulnerability was found in the BusyBox httpd IP-based access control list (ACL) parser (networking/httpd.c). When a D: (deny) rule is configured with a syntactically invalid CIDR prefix length, such as /999, the parser does not validate the mask value. Instead of rejecting the malformed rule or defaulting to a restrictive interpretation, the resulting miscalculated network mask silently fails to match the intended IP address range.

This means that a deny rule intended to block an entire subnet may effectively deny no traffic at all. The silent failure mode violates the principle of failing securely and could remain undetected for extended periods.

A vulnerability was found in the BusyBox httpd IP-based access control list (ACL) parser (networking/httpd.c). When a D: (deny) rule is configured with a syntactically invalid CIDR prefix length, such as /999, the parser does not validate the mask value. Instead of rejecting the malformed rule or defaulting to a restrictive interpretation, the resulting miscalculated network mask silently fails to match the intended IP address range.

This means that a deny rule intended to block an entire subnet may effectively deny no traffic at all. The silent failure mode violates the principle of failing securely and could remain undetected for extended periods.

A unit confusion in BusyBox TLS Montgomery reduction buffer allocation causes a pre-authentication heap buffer overflow when processing a crafted ClientKeyExchange message.

A vulnerability was found in the BusyBox TLS implementation (networking/tlspstmmontgomeryreduce.c). A unit confusion error exists in the buffer allocation for the Montgomery reduction operation. The code calls xzalloc(2pa+1) where pa is measured in pstmdigit units (4 or 8 bytes each), but the allocation treats this value as a byte count. This results in an allocation approximately 4x to 8x smaller than required.

When a TLS client sends a crafted ClientKeyExchange message with an all-zeros payload, the RSA decryption path triggers the Montgomery reduction, which writes digit-sized elements beyond the allocated buffer boundary. This constitutes a pre-authentication out-of-bounds heap write.

The confirmed impact is a pre-authentication denial of service (crash). While the heap buffer overflow is theoretically exploitable for remote code execution, this was not demonstrated. On Fedora, system-level mitigations including ASLR, PIE, full RELRO, and SELinux confinement make practical code execution extremely unlikely.

A flaw was found in BusyBox httpd's reverse proxy feature. When the proxy is configured to forward requests matching a URL prefix to a backend server, only the first HTTP request on a TCP connection has its path rewritten. An attacker can send a pipelined second request that is relayed verbatim to the backend, bypassing the configured path restrictions. This allows accessing arbitrary paths on the backend server that were not intended to be exposed through the proxy.

Hi,

Anmol Bakshi brought the below with original Subject line saying "BusyBox dpkg applet: OS command injection -> root RCE (CWE-78)" to linux-distros on July 13. We quickly determined that no privilege boundary is crossed, so the coordinated disclosure request was withdrawn. However, we cannot just withdraw something sent to linux-distros (as long as it wasn't completely off-topic) without also bringing it to oss-security. In this case, we overlooked that until I started processing the linux-distros list statistics for July.

I include the linux-distros messages quoted below, sans the originally attached PDF file (which I didn't even look at).

Alexander

On Mon, Jul 13, 2026 at 08:44:40AM +0530, Anmol Bakshi wrote: Proposed coordinated public disclosure: 2026-07-27.

Summary -------

OS command injection in BusyBox's dpkg applet gives arbitrary command execution as the user running dpkg (root in typical use). Present through current BusyBox; confirmed on 1.36.1 and 1.37.0.git HEAD.

Vulnerability -------------

archival/dpkg.c (~line 1271): scriptpath = xasprintf("/var/lib/dpkg/info/%s.%s", packagename, scripttype); result = access(scriptpath, FOK) ? EXITSUCCESS : system(scriptpath);

The Package field from the .deb control file is passed unvalidated into the path handed to system() (/bin/sh -c). Shell metacharacters in the package name execute as command separators. The forward-slash filename restriction is bypassed with a base64-encoded payload the shell decodes at runtime, making injection unrestricted.

Impact ------

Arbitrary command execution as uid=0. Demonstrated: /etc/shadow exfiltration and an outbound network callback from a crafted .deb. CVSS 4.0: 9.3 (AV:L/AC:L/AT:N/PR:N/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H). CWE-78.

Affected scope --------------

Any build with CONFIGDPKG enabled that processes an attacker-supplied .deb. The applet is enabled in some distro builds (for example Debian's busybox deb config enables a minimal dpkg). Please assess your own configuration.

Suggested fix -------------

1. Replace system(scriptpath) with fork()/execl(scriptpath, scriptpath, (char)NULL) so no shell is invoked. 2. Validate package names against ^[a-z0-9][a-z0-9.+-]+$ to match GNU dpkg.

Coordination ------------

This is a coordinated disclosure. I am separately notifying Alpine Linux, OpenWrt, the Yocto Project, and Buildroot (none of which are on this list), all toward the same public date of 2026-07-27. BusyBox upstream has no working private security channel: the Bugzilla instance is offline and the security contact bounces. I sent a private report on 2026-04-21 with no response in ~3 months, so there is no upstream fix to coordinate. I have an open MITRE request from 2026-04-21 that has not been actioned; I would prefer a CVE ID be assigned via a CNA on this list. Full report and a self-contained reproduction script are attached. Please confirm receipt.

Regards, Anmol Bakshi https://anmolbakshi.com On Mon, Jul 13, 2026 at 08:32:59AM +0200, Solar Designer wrote: Hello Anmol,

Thank you for notifying linux-distros.

On Mon, Jul 13, 2026 at 08:44:40AM +0530, Anmol Bakshi wrote: Proposed coordinated public disclosure: 2026-07-27. I ack that this is a valid public disclosure date per this list's policy, but I note it's the latest possible - so please note that no extension beyond this date will be possible.

Please make sure you post specifically to the public oss-security list when making the issue public. Full report and a self-contained reproduction script are attached. Unfortunately, you sent these pre-encrypted to the list key, whereas the list only re-encrypts non-pre-encrypted attachments of full PGP/MIME messages as commonly produced by MUAs with PGP support. In fact, I think these attachments ended up double-encrypted when you sent the message, and only the outer layer got processed by the list software. So they arrived to list subscribers still encrypted to the list key as you prepared them, rather than re-encrypted to their keys, and thus are unreadable to them.

I suggest that you selectively send these to list members who ask, or you may resend the reproduction script as a simple attachment (don't pre-encrypt, but let the MIME message be encrypted) or in message body.

I doubt the PDF is worth resending, but I don't know what's in it. For oss-security, the message size limit (including MIME overhead) is 1 MB, so this large PDF wouldn't be suitable for posting to there anyway. Please confirm receipt. Confirmed as above. OS command injection in BusyBox's dpkg applet gives arbitrary command execution as the user running dpkg (root in typical use). I guess package installation via dpkg would invoke scripts found in the package as part of intended functionality, so there would be no additional risk from your finding - but perhaps there are query-only uses where your finding matters? You could want to elaborate on that.

I am not familiar with BusyBox's dpkg.

Replying in plaintext as I didn't bother searching for your public key, but I limited the quoting. (And this gets encrypted to list subscribers anyway.)

Alexander On Mon, Jul 13, 2026 at 12:57:10PM +0530, Anmol Bakshi wrote: Hi Alexander,

Thanks again for the detailed feedback - you're right, and I'm withdrawing this.

To close the loop on the impact question: runpackagescriptordie() is only reached during install / unpack / configure / remove, i.e. exactly the operations where a package's maintainer scripts are expected to run as root. BusyBox dpkg's only read-only operation (-l) doesn't touch that path. So the injected command runs with the same privilege a normal postinst already would; the unvalidated package name doesn't cross a boundary that an ordinary malicious maintainer script wouldn't.

I also considered the "trojan a real package, keep the same name to slip past review" angle, but it doesn't change that: dpkg -i on a local file does no signature or integrity check, so there's no verification for the name injection to bypass, and the operator has already chosen to run untrusted code as root regardless of what the scripts contain. No privilege boundary is crossed there either.

So I'm standing down the coordinated-disclosure request and won't pursue a CVE on this basis - no need for anyone on the list to hold anything for 2026-07-27. The one thing I still think is worthwhile is a hardening change to BusyBox (validate the package name as GNU dpkg does, and don't pass the path to a shell); I'll take that to BusyBox directly as a patch, outside this list.

Thanks for the time, and sorry for the noise.

Regards, सादर, Anmol Bakshi On Sat, Aug 22, 2026 at 09:46:57PM +0200, Solar Designer wrote: Hi Anmol,

I'm sorry everyone seems to have dropped the ball on this after your last message. I understand you may have thought you could simply withdraw a report, but others on linux-distros should have realized that's not the case.

On Mon, Jul 13, 2026 at 12:57:10PM +0530, Anmol Bakshi wrote: Thanks again for the detailed feedback - you're right, and I'm withdrawing this. Thanks, but currently you cannot just withdraw a report to linux-distros without bringing it to oss-security - all while acknowledging it was determined to be a non-security issue. We need this for transparency. If the ratio of non-security reports becomes much larger, we'll have to reconsider this approach - and announce that publicly - but for now that's the approach we use and the oss-security community may expect.

Are you going to bring this to oss-security ASAP? If not, I will.

Alexander

A vulnerability has been found in BusyBox up to 1.30.1. This vulnerability affects unknown code of the file networking/wget.c of the component FEATUREWGETTIMEOUT Handler. Such manipulation of the argument -T leads to null pointer dereference. Local access is required to approach this attack. The exploit has been disclosed to the public and may be used. The name of the patch is 83a40bf7a93c8ac093d33ab452222dd5b9eb57ff. A patch should be applied to remediate this issue.

A use-after-free in the awksub() function (editors/awk.c) of Busybox v1.38.0 allows attackers to cause a Denial of Service (DoS) via supplying a crafted AWK script.

A heap overflow in the evalcommand() function (shell/ash.c) of Busybox v1.38.0 allows attackers to cause a Denial of Service (DoS) via supplying a crafted input.

A heap overflow in the ifsbreakup() function (shell/ash.c) of Busybox v1.38.0 allows attackers to cause a Denial of Service (DoS) via supplying a crafted input.