On Feb 07 15:09:39, [email protected] wrote:
> On Tue, Feb 07, 2023 at 02:48:25PM +0100, Jan Stary wrote:
> > On Feb 06 22:56:02, [email protected] wrote:
> > > There is an ongoing discussion on audio/sox on oss-security:
> > >
> > > https://marc.info/?l=oss-security&m=167546008232629&w=2
> > >
> > > Steffen Nurpmeso ported the patches to apply against the commit
> > > we also use in our ports, that's what's included in the diff below.
> > >
> > > The patches look sensible to me although I haven't reviewed them
> > > thoroughly.
> > >
> > > It's probably a good idea to keep an eye on this discussion both for
> > > reviews of the patches and for possible developments of a new upstream
> > > repo containing them.
> >
> > I just asked upstream - let's wait a on whether the upstream maintainer
> > decides to include these in the upstream git (SF) that we build from;
> > I would prefer that to maintaining the patches (thank you Steffen!)
>
> For reference:
>
> https://marc.info/?l=sox-devel&m=167577672104072&w=2
>
> The patches were sent to oss-security *because* upstream failed to react:
>
> I am working on fixing known vulnerabilities in sox and since upstream
> seems mostly dead (no commits in more than a year, no replies to bug
> reports)
>
> Given this, waiting until upstream decides to unhibernate makes little
> sense to me.
After talkig to both Helmut and upstream,
I am convinced there is no point waiting
till upstream acts on this (or anything, really).
I have created merge requests upstream
for each of the patches individually.
Let's see ahat happens.
Going through the individual patches:
it seems SOX_EOF is the universal indicator of a returned error,
even though the err msgs use SOX_EHDR or SOX_EFMT.
But it seems the rest of the code doesn't really
distinguish the kind of returned error ...
Comments inline.
Jan
> diff --git a/src/aiff.c b/src/aiff.c
> index 3a152c588c..6de94f3276 100644
> --- a/src/aiff.c
> +++ b/src/aiff.c
> @@ -619,6 +619,11 @@ int lsx_aiffstartwrite(sox_format_t * ft)
> At 48 kHz, 16 bits stereo, this gives ~3 hours of audio.
> Sorry, the AIFF format does not provide for an indefinite
> number of samples. */
> + if (ft->signal.channels >= (0x7f000000 /
> (ft->encoding.bits_per_sample >> 3)))
> + {
> + lsx_fail_errno(ft, SOX_EOF, "too many channels for AIFF
> header");
> + return SOX_EOF;
> + }
> return(aiffwriteheader(ft, (uint64_t) 0x7f000000 /
> ((ft->encoding.bits_per_sample>>3)*ft->signal.channels)));
> }
This prevents sox from _writing_ an aiff file
with an absurd number of channels; the bug
https://sourceforge.net/p/sox/bugs/360/
seems related: should we also refuse to _read_
audio files with 1073807359 channels and the like?
>
> diff --git a/src/formats.c b/src/formats.c
> index 3fcf4382b6..5eda5e3612 100644
> --- a/src/formats.c
> +++ b/src/formats.c
> @@ -627,6 +627,7 @@ error:
> free(ft->priv);
> free(ft->filename);
> free(ft->filetype);
> + sox_delete_comments(&ft->oob.comments);
> free(ft);
> return NULL;
> }
This is in a failed open_read();
the same imho applies to a failed open_write();
updated patch attached.
However, once SoX fails an open_{read,write},
it exits soon afterwards anyway.
> diff --git a/src/formats_i.c b/src/formats_i.c
> index 7048040d1c..5f5ef979d4 100644
> --- a/src/formats_i.c
> +++ b/src/formats_i.c
> @@ -19,6 +19,7 @@
> */
>
> #include "sox_i.h"
> +#include <limits.h>
> #include <string.h>
> #include <sys/stat.h>
> #include <stdarg.h>
> @@ -60,13 +61,24 @@ int lsx_check_read_params(sox_format_t * ft, unsigned
> channels,
> if (ft->seekable)
> ft->data_start = lsx_tell(ft);
>
> - if (channels && ft->signal.channels && ft->signal.channels != channels)
> + if (channels && ft->signal.channels && ft->signal.channels != channels) {
> lsx_warn("`%s': overriding number of channels", ft->filename);
> - else ft->signal.channels = channels;
> + } else if (channels > SHRT_MAX) {
> + lsx_fail_errno(ft, EINVAL, "implausibly large number of channels");
> + return SOX_EOF;
> + } else {
> + ft->signal.channels = channels;
> + }
Ah, this is what correctly rejects the files at
https://sourceforge.net/p/sox/bugs/360/
> - if (rate && ft->signal.rate && ft->signal.rate != rate)
> + if (rate && ft->signal.rate && ft->signal.rate != rate) {
> lsx_warn("`%s': overriding sample rate", ft->filename);
> - else ft->signal.rate = rate;
> + /* Since NaN comparisons yield false, the negation rejects them. */
> + } else if (!(rate > 0)) {
> + lsx_fail_errno(ft, EINVAL, "invalid rate value");
> + return SOX_EOF;
> + } else {
> + ft->signal.rate = rate;
> + }
>
> if (encoding && ft->encoding.encoding && ft->encoding.encoding != encoding)
> lsx_warn("`%s': overriding encoding type", ft->filename);
> diff --git a/src/hcom.c b/src/hcom.c
> index 594c870606..94ed3dbdb0 100644
> --- a/src/hcom.c
> +++ b/src/hcom.c
> @@ -141,6 +141,11 @@ static int startread(sox_format_t * ft)
> return (SOX_EOF);
> }
> lsx_readw(ft, &dictsize);
> + if (dictsize == 0 || dictsize > 511)
> + {
> + lsx_fail_errno(ft, SOX_EHDR, "Implausible dictionary size in
> HCOM header");
> + return SOX_EOF;
Should this be SOX_EHDR instead?
> + }
>
> /* Translate to sox parameters */
> ft->encoding.encoding = SOX_ENCODING_HCOM;
> @@ -161,13 +166,18 @@ static int startread(sox_format_t * ft)
> p->dictionary[i].dict_rightson);
> if (!dictvalid(i, dictsize, p->dictionary[i].dict_leftson,
> p->dictionary[i].dict_rightson)) {
> + free(p->dictionary);
> + p->dictionary = NULL;
> lsx_fail_errno(ft, SOX_EHDR, "Invalid dictionary");
> return SOX_EOF;
> }
> }
> rc = lsx_skipbytes(ft, (size_t) 1); /* skip pad byte */
> - if (rc)
> + if (rc) {
> + free(p->dictionary);
> + p->dictionary = NULL;
> return rc;
> + }
>
> /* Initialized the decompression engine */
> p->checksum = checksum;
> @@ -249,6 +259,9 @@ static int stopread(sox_format_t * ft)
> {
> register priv_t *p = (priv_t *) ft->priv;
>
> + free(p->dictionary);
> + p->dictionary = NULL;
> +
> if (p->huffcount != 0)
> {
> lsx_fail_errno(ft,SOX_EFMT,"not all HCOM data read");
> @@ -259,8 +272,7 @@ static int stopread(sox_format_t * ft)
> lsx_fail_errno(ft,SOX_EFMT,"checksum error in HCOM data");
> return (SOX_EOF);
> }
> - free(p->dictionary);
> - p->dictionary = NULL;
> +
> return (SOX_SUCCESS);
> }
>
> diff --git a/src/sphere.c b/src/sphere.c
> index a3fd1c64c2..9544d16000 100644
> --- a/src/sphere.c
> +++ b/src/sphere.c
> @@ -63,7 +63,8 @@ static int start_read(sox_format_t * ft)
> return (SOX_EOF);
> }
>
> - header_size -= (strlen(buf) + 1);
> + bytes_read = strlen(buf);
> + header_size -= bytes_read >= header_size ? header_size : bytes_read + 1;
>
> while (strncmp(buf, "end_head", (size_t)8) != 0) {
> if (strncmp(buf, "sample_n_bytes", (size_t)14) == 0)
> @@ -105,7 +106,8 @@ static int start_read(sox_format_t * ft)
> return (SOX_EOF);
> }
>
> - header_size -= (strlen(buf) + 1);
> + bytes_read = strlen(buf);
> + header_size -= bytes_read >= header_size ? header_size : bytes_read + 1;
> }
>
> if (!bytes_per_sample)
> diff --git a/src/voc.c b/src/voc.c
> index a75639e94e..0ca07f9450 100644
> --- a/src/voc.c
> +++ b/src/voc.c
> @@ -625,6 +625,10 @@ static int getblock(sox_format_t * ft)
> v->rate = new_rate_32;
> ft->signal.rate = new_rate_32;
> lsx_readb(ft, &uc);
> + if (uc <= 1) {
> + lsx_fail_errno(ft, SOX_EFMT, "2 bits per word required");
> + return (SOX_EOF);
SOX_EFMT?
> + }
> v->size = uc;
> lsx_readb(ft, &uc);
> if (v->channels != -1 && uc != v->channels) {
> diff --git a/src/wav.c b/src/wav.c
> index 3f6beb4517..39e0c487e8 100644
> --- a/src/wav.c
> +++ b/src/wav.c
> @@ -654,6 +654,12 @@ static int wav_read_fmt(sox_format_t *ft, uint32_t len)
> if (err)
> return SOX_EOF;
>
> + if (wav->bitsPerSample == 0)
This avoids the division by zero,
but should it also check for <= sensible?
> + {
> + lsx_fail_errno(ft, SOX_EHDR, "WAV file bits per sample is zero");
> + return SOX_EOF;
Again, shouldn't this be SOX_EHDR?
> + }
> +
> /* non-PCM formats except alaw and mulaw formats have extended fmt chunk.
> * Check for those cases.
> */
> + if (p->huffcount != 0)
> + {
> + lsx_fail_errno(ft,SOX_EFMT,"not all HCOM data read");
> +@@ -259,8 +272,7 @@ static int stopread(sox_format_t * ft)
> + lsx_fail_errno(ft,SOX_EFMT,"checksum error in HCOM data");
> + return (SOX_EOF);
Maybe SOX_EOF is the universal indicator of an error to the caller
(which in itself is a bit confusing), but this is SOX_EFMT, right?
> Index: patches/patch-src_sphere_c
> ===================================================================
> RCS file: patches/patch-src_sphere_c
> diff -N patches/patch-src_sphere_c
> --- /dev/null 1 Jan 1970 00:00:00 -0000
> +++ patches/patch-src_sphere_c 6 Feb 2023 21:38:58 -0000
> @@ -0,0 +1,25 @@
> +https://marc.info/?l=oss-security&m=167571683504082&w=2
> +
> +Index: src/sphere.c
> +--- src/sphere.c.orig
> ++++ src/sphere.c
> +@@ -63,7 +63,8 @@ static int start_read(sox_format_t * ft)
> + return (SOX_EOF);
> + }
> +
> +- header_size -= (strlen(buf) + 1);
> ++ bytes_read = strlen(buf);
> ++ header_size -= bytes_read >= header_size ? header_size : bytes_read + 1;
> +
> + while (strncmp(buf, "end_head", (size_t)8) != 0) {
> + if (strncmp(buf, "sample_n_bytes", (size_t)14) == 0)
> +@@ -105,7 +106,8 @@ static int start_read(sox_format_t * ft)
> + return (SOX_EOF);
> + }
> +
> +- header_size -= (strlen(buf) + 1);
> ++ bytes_read = strlen(buf);
> ++ header_size -= bytes_read >= header_size ? header_size : bytes_read + 1;
> + }
> +
> + if (!bytes_per_sample)
> --- /dev/null 1 Jan 1970 00:00:00 -0000
> +++ patches/patch-src_voc_c 6 Feb 2023 21:38:58 -0000
> @@ -0,0 +1,16 @@
> +https://marc.info/?l=oss-security&m=167571683504082&w=2
> +
> +Index: src/voc.c
> +--- src/voc.c.orig
> ++++ src/voc.c
> +@@ -625,6 +625,10 @@ static int getblock(sox_format_t * ft)
> + v->rate = new_rate_32;
> + ft->signal.rate = new_rate_32;
> + lsx_readb(ft, &uc);
> ++ if (uc <= 1) {
> ++ lsx_fail_errno(ft, SOX_EFMT, "2 bits per word required");
> ++ return (SOX_EOF);
Again, SOX_EFMT?
> ++ }
> + v->size = uc;
> + lsx_readb(ft, &uc);
> + if (v->channels != -1 && uc != v->channels) {
> Index: patches/patch-src_wav_c
> ===================================================================
> RCS file: patches/patch-src_wav_c
> diff -N patches/patch-src_wav_c
> --- /dev/null 1 Jan 1970 00:00:00 -0000
> +++ patches/patch-src_wav_c 6 Feb 2023 21:38:58 -0000
> @@ -0,0 +1,18 @@
> +https://marc.info/?l=oss-security&m=167571683504082&w=2
> +
> +Index: src/wav.c
> +--- src/wav.c.orig
> ++++ src/wav.c
> +@@ -654,6 +654,12 @@ static int wav_read_fmt(sox_format_t *ft, uint32_t len
> + if (err)
> + return SOX_EOF;
> +
> ++ if (wav->bitsPerSample == 0)
> ++ {
> ++ lsx_fail_errno(ft, SOX_EHDR, "WAV file bits per sample is zero");
> ++ return SOX_EOF;
Also here: SOX_EHDR?
Jan
https://marc.info/?l=oss-security&m=167571683504082&w=2
Index: src/formats.c
--- src/formats.c.orig
+++ src/formats.c
@@ -360,7 +360,7 @@ static int sox_checkformat(sox_format_t * ft)
return SOX_SUCCESS;
}
-static sox_bool is_url(char const * text) /* detects only wget-supported URLs
*/
+static sox_bool is_url(char const * text)
{
return !(
strncasecmp(text, "http:" , (size_t)5) &&
@@ -442,7 +442,7 @@ static FILE * xfopen(char const * identifier, char con
else if (is_url(identifier)) {
FILE * f = NULL;
#ifdef HAVE_POPEN
- char const * const command_format = "wget --no-check-certificate -q -O-
\"%s\"";
+ char const * const command_format = "ftp -a -V -o - \"%s\"";
char * command = lsx_malloc(strlen(command_format) + strlen(identifier));
sprintf(command, command_format, identifier);
f = popen(command, POPEN_MODE);
@@ -627,6 +627,7 @@ error:
free(ft->priv);
free(ft->filename);
free(ft->filetype);
+ sox_delete_comments(&ft->oob.comments);
free(ft);
return NULL;
}
@@ -988,6 +989,7 @@ error:
free(ft->priv);
free(ft->filename);
free(ft->filetype);
+ sox_delete_comments(&ft->oob.comments);
free(ft);
return NULL;
}
