On Fri, Nov 30, 2012 at 11:00 AM, Dieter Bloms vdr@bloms.de wrote:

--snip-- g++ -g -O3 -Wall -Werror=overloaded-virtual -Wno-parentheses -c -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -DVDR_USER="root" -DLIRC_DEVICE="/var/run/lirc/lircd" -D_GNU_SOURCE -DVIDEODIR="/remote/vdr/video" -DCONFDIR="/etc/vdr/config" -DCACHEDIR="" -DRESDIR="" -DPLUGINDIR="/usr/src/vdr-1.7.32/PLUGINS/lib" -DLOCDIR="/usr/share/vdr/locale" -I/usr/include/freetype2 -I/usr/src/v4l-dvb/linux/include epg.c epg.c: In member function 'bool tComponent::FromString(const char*)': epg.c:37:3: warning: format '%a' expects argument of type 'double', but argument 6 has type 'char*' [-Wformat] --snip--

And after start in the logfile

--snip-- Nov 30 09:48:05 vdrservernew local1.err vdr: [4094] dbloms: "1" "01" " deu" "0x8.799b08p-1052" --snip--

I think %a in my esyslog call isn't correct, but it is used this way with the sscanf call and when I changed it with %s I don't see any log entry.

So how should the esyslog call look like ?

%a to print a char* is not right. char* is interpreted as double (see the compiler warning) . If you want to print the pointer 'description', then use %p.

The following code crashes on my box too. I dont know why.

$ ./a.out errno: 0 n=4 Stream:1 Type:1 lang:'deu' desc addr:'0x40800000' Segmentation fault $

In gdb: Program received signal SIGSEGV, Segmentation fault. 0x0016e50b in _IO_vfprintf_internal (s=0x2844e0, format=0x8048711 "desc:'%s'\n", ap=0xbffff3f4 "") at vfprintf.c:1614 1614 vfprintf.c: No such file or directory. in vfprintf.c (gdb) bt #0 0x0016e50b in _IO_vfprintf_internal (s=0x2844e0, format=0x8048711 "desc:'%s'\n", ap=0xbffff3f4 "") at vfprintf.c:1614 #1 0x00175160 in __printf (format=0x8048711 "desc:'%s'\n") at printf.c:35 #2 0x080485aa in FromString (s=0x804871c "1 01 deu 4:3") at fromstring.c:14 #3 0x080485ee in main (argc=1, argv=0xbffff514) at fromstring.c:22

--- snip ---

#include <stdio.h> #include <errno.h>

void FromString(const char *s) { unsigned int Stream, Type; char language[8] = {0}; char* description=NULL; int n = sscanf(s, "%X %02X %7s %a[^\n]", &Stream, &Type, language, &description); // 7 = MAXLANGCODE2 - 1 printf("errno: %d\n", errno); printf("n=%d\n", n); printf("Stream:%u Type:%u lang:'%s'\n", Stream, Type, language); printf("desc addr:'%p'\n", description); printf("desc:'%s'\n", description); // XXX crashes here }

int main(int argc, char* argv[]) { if (argc == 2) FromString((const char*)argv[1]); else FromString("1 01 deu 4:3");

return 0; }

----- snip -----

On Fri, Nov 30, 2012 at 11:31 AM, Dieter Bloms vdr@bloms.de wrote:

Hi,

On Fri, Nov 30, sundararaj reel wrote:

...

%a to print a char* is not right. char* is interpreted as double (see the compiler warning) . If you want to print the pointer 'description', then use %p.

with "esyslog("dbloms: "%X" "%02X" "%7s" "%p"", Stream, Type, language, description);" I get the following log entry:

--snip-- Nov 30 10:27:24 vdrservernew local1.err vdr: [7118] dbloms: "1" "01" " deu" "0x40800000" --snip--

...

The following code crashes on my box too. I dont know why.

did you do it on uClibc system or a glibc one ?

Ran that code on ubuntu 10.04.

$ /lib/libc.so.6 GNU C Library (Ubuntu EGLIBC 2.11.1-0ubuntu7) stable release version 2.11.1, by Roland McGrath et al. Copyright (C) 2009 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. Compiled by GNU CC version 4.4.3. Compiled on a Linux >>2.6.24-27-server<< system on 2010-04-22. Available extensions: crypt add-on version 2.1 by Michael Glad and others GNU Libidn by Simon Josefsson Native POSIX Threads Library by Ulrich Drepper et al BIND-8.2.3-T5B For bug reporting instructions, please see: http://www.debian.org/Bugs/.

Hi Werner,

On Fri, Nov 30, Dr. Werner Fink wrote:

On Fri, Nov 30, 2012 at 11:00:52AM +0100, Dieter Bloms wrote:

...

I changed it this way:

--snip-- 32 bool tComponent::FromString(const char *s) 33 { 34 unsigned int Stream, Type; 35 description = NULL; 36 int n = sscanf(s, "%X %02X %7s %a[^\n]", &Stream, &Type, language, &description); // 7 = MAXLANGCODE2 - 1 37 esyslog("dbloms: "%X" "%02X" "%7s" "%a"", Stream, Type, language, description); 38 39 if (n != 4 || isempty(description)) { 40 free(description); 41 description = NULL; 42 } 43 stream = Stream; 44 type = Type; 45 return n >= 3; 46 } --snip--

IMHO the %a should become %s in esyslog() whereas in sscanf() the %a should become %as for the GNU extension of dynamically allocating string conversions.

Now I've the following code:

--snip-- 25 cString tComponent::ToString(void) 26 { 27 char buffer[256]; 28 snprintf(buffer, sizeof(buffer), "%X %02X %s %s", stream, type, language, description ? description : ""); 29 return buffer; 30 } 31 32 bool tComponent::FromString(const char *s) 33 { 34 unsigned int Stream, Type; 35 description = NULL; 36 int n = sscanf(s, "%X %02X %7s %as[^\n]", &Stream, &Type, language, &description); // 7 = MAXLANGCODE2 - 1 37 esyslog("dbloms: "%X" "%02X" "%7s" "%s"", Stream, Type, language, description); 38 39 if (n != 4 || isempty(description)) { 40 free(description); 41 description = NULL; 42 } 43 stream = Stream; 44 type = Type; 45 return n >= 3; 46 } --snip--

and get a core dump with this:

--snip-- vdrservernew:/tmp# gdb --core /tmp/core /usr/local/bin/vdr GNU gdb (GDB) 7.5 Copyright (C) 2012 Free Software Foundation, Inc. License GPLv3+: GNU GPL version 3 or later http://gnu.org/licenses/gpl.html This is free software: you are free to change and redistribute it. There is NO WARRANTY, to the extent permitted by law. Type "show copying" and "show warranty" for details. This GDB was configured as "x86_64-unknown-linux-gnu". For bug reporting instructions, please see: http://www.gnu.org/software/gdb/bugs/... Reading symbols from /usr/local/bin/vdr...done. [New LWP 8986] [New LWP 8985] [New LWP 8987] [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib/libthread_db.so.1". Core was generated by `/usr/local/bin/vdr --config=/etc/vdr --epgfile=/tmp/epg.data --grab=/dev/shm --'. Program terminated with signal 11, Segmentation fault. #0 0x00006b2476b263b3 in strnlen () from /lib/libc.so.0.9.32 (gdb) bt #0 0x00006b2476b263b3 in strnlen () from /lib/libc.so.0.9.32 #1 0x00006b2476b1ff56 in ?? () from /lib/libc.so.0.9.32 #2 0x00006b2476b1d5dc in vsnprintf () from /lib/libc.so.0.9.32 #3 0x00006b2476b14c29 in vsyslog () from /lib/libc.so.0.9.32 #4 0x000000000052e510 in syslog_with_tid (priority=3, format=0x551ab8 "dbloms: "%X" "%02X" "%7s" "%s"") at tools.c:40 #5 0x00000000004a1e4e in FromString (s=<optimized out>, this=0x3dabc20) at epg.c:37 #6 cComponents::SetComponent (this=<optimized out>, Index=<optimized out>, s=s@entry=0x3daaaa2 "1 01 deu 4:3") at epg.c:84 #7 0x00000000004a4163 in cEvent::Parse (this=0x3d9f4d0, s=<optimized out>) at epg.c:498 #8 0x00000000004e9f16 in cRecordingInfo::Read (this=0x3da8640, f=f@entry=0x3d9f840) at recording.c:468 #9 0x00000000004eb553 in cRecording::cRecording (this=0x3da05c0, FileName=0x3d8aa7c "Sex_and_the_City_2/2012-11-19.20.10.27-0.rec") at recording.c:723 #10 0x00000000004ecf21 in cRecordings::ScanVideoDir (this=0x7fe880 <Recordings>, DirName=0x3d9d4c0 "/remote/vdr/Sex_and_the_City_2", Foreground=false, LinkLevel=0) at recording.c:1165 #11 0x00000000004ed39c in cRecordings::ScanVideoDir (this=0x7fe880 <Recordings>, DirName=0x3d83a20 "/remote/vdr", Foreground=false, LinkLevel=0) at recording.c:1180 #12 0x00000000005269be in cThread::StartThread (Thread=0x7fe8a0 <Recordings+32>) at thread.c:262 #13 0x00006b247844a406 in start_thread () from /lib/libpthread.so.0.9.32 #14 0x00006b2478442885 in clone () from /lib/libpthread.so.0.9.32 #15 0x0000000000000000 in ?? () (gdb) --snip--

Hi,

On Fri, Nov 30, Dr. Werner Fink wrote:

Maybe some more checks are required as if n != 4 this could be that one of the variables are not used and therefore not initialized.

Before sscanf() this could be a `memset(language, 0, sizeof(language));' or more simple a `*language = 0;' depending on the type of language. Also the code

if (n != 4 || isempty(description)) { free(description); description = NULL; }

could become

if (n != 4 || isempty(description)) { if (description) free(description); description = NULL; }

as well as the line

esyslog("dbloms: "%X" "%02X" "%7s" "%s"", Stream, Type, language, description);

may look like

esyslog("dbloms: "%X" "%02X" "%7s" "%s"", Stream, Type, language, description ? description : "");

may code looks like:

--snip-- bool tComponent::FromString(const char *s) { unsigned int Stream, Type; description = NULL; int n = sscanf(s, "%X %02X %7s %as[^\n]", &Stream, &Type, language, &description); // 7 = MAXLANGCODE2 - 1 esyslog("dbloms: "%X" "%02X" "%7s" "%s"", Stream, Type, language, description ? description : "");

if (n != 4 || isempty(description)) { if (description) free(description); description = NULL; } stream = Stream; type = Type; return n >= 3; } --snip--

and I get a core dump like:

--snip-- vdrservernew:/usr/src/vdr-1.7.32# gdb --core /tmp/core /usr/local/bin/vdr GNU gdb (GDB) 7.5 Copyright (C) 2012 Free Software Foundation, Inc. License GPLv3+: GNU GPL version 3 or later http://gnu.org/licenses/gpl.html This is free software: you are free to change and redistribute it. There is NO WARRANTY, to the extent permitted by law. Type "show copying" and "show warranty" for details. This GDB was configured as "x86_64-unknown-linux-gnu". For bug reporting instructions, please see: http://www.gnu.org/software/gdb/bugs/... Reading symbols from /usr/local/bin/vdr...done. [New LWP 2829] [New LWP 2830] [New LWP 2828] [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib/libthread_db.so.1". Core was generated by `/usr/local/bin/vdr --config=/etc/vdr --epgfile=/tmp/epg.data --grab=/dev/shm --'. Program terminated with signal 11, Segmentation fault. #0 0x0000799a5fbbe3b3 in strnlen () from /lib/libc.so.0.9.32 (gdb) bt #0 0x0000799a5fbbe3b3 in strnlen () from /lib/libc.so.0.9.32 #1 0x0000799a5fbb7f56 in ?? () from /lib/libc.so.0.9.32 #2 0x0000799a5fbb55dc in vsnprintf () from /lib/libc.so.0.9.32 #3 0x0000799a5fbacc29 in vsyslog () from /lib/libc.so.0.9.32 #4 0x000000000052e570 in syslog_with_tid (priority=3, format=0x551b18 "dbloms: "%X" "%02X" "%7s" "%s"") at tools.c:40 #5 0x00000000004a1bbd in tComponent::FromString (this=0x22a24b0, s=<optimized out>) at epg.c:37 #6 0x00000000004a4197 in SetComponent (s=<optimized out>, Index=0, this=0x22ad550) at epg.c:85 #7 cEvent::Parse (this=<optimized out>, s=<optimized out>) at epg.c:499 #8 0x00000000004e9f76 in cRecordingInfo::Read (this=0x22a50e0, f=f@entry=0x22a1f00) at recording.c:468 #9 0x00000000004eb5b3 in cRecording::cRecording (this=0x22a2210, FileName=0x22a0fdc "Sex_and_the_City_2/2012-11-19.20.10.27-0.rec") at recording.c:723 #10 0x00000000004ecf81 in cRecordings::ScanVideoDir (this=0x7fe900 <Recordings>, DirName=0x229aca0 "/remote/vdr/Sex_and_the_City_2", Foreground=false, LinkLevel=0) at recording.c:1165 #11 0x00000000004ed3fc in cRecordings::ScanVideoDir (this=0x7fe900 <Recordings>, DirName=0x2285390 "/remote/vdr", Foreground=false, LinkLevel=0) at recording.c:1180 #12 0x0000000000526a1e in cThread::StartThread (Thread=0x7fe920 <Recordings+32>) at thread.c:262 #13 0x0000799a614e2406 in start_thread () from /lib/libpthread.so.0.9.32 #14 0x0000799a614da885 in clone () from /lib/libpthread.so.0.9.32 #15 0x0000000000000000 in ?? () (gdb) --snip--

Am 30.11.2012 11:32, schrieb Gerald Dachs:

Am 2012-11-30 10:17, schrieb Lars Hanisch:

...

Looks like the pointer returned by sscanf is not valid:

32: bool tComponent::FromString(const char *s) 33: { 34: unsigned int Stream, Type; 35: int n = sscanf(s, "%X %02X %7s %a[^\n]", &Stream, &Type, language, &description); // 7 = MAXLANGCODE2 - 1 36: if (n != 4 || isempty(description)) { 37: free(description); 38: description = NULL; 39: } 40: stream = Stream; 41: type = Type; 42: return n >= 3; 43: }

From man sscanf:

   The GNU C library supports a nonstandard extension that causes the library to
   dynamically allocate a string of sufficient size for input strings for the %s
   and %a[range] conversion specifiers.

This is the reason why it doesn't work with ulibc.

Then there should be a malloc or something similiar for description:

32: bool tComponent::FromString(const char *s) 33: { 34: unsigned int Stream, Type; description = malloc(strlen(s)); description[0] = 0; 35: int n = sscanf(s, "%X %02X %7s %a[^\n]", &Stream, &Type, language, &description); // 7 = MAXLANGCODE2 - 1 36: if (n != 4 || isempty(description)) { 37: free(description); 38: description = NULL; 39: } 40: stream = Stream; 41: type = Type; 42: return n >= 3; 43: }

A check for description != NULL before the free call is not needed.

But this is not the only place in the vdr code, where %a is used...

Lars.

Gerald

---

vdr mailing list vdr@linuxtv.org http://www.linuxtv.org/cgi-bin/mailman/listinfo/vdr

On 01.12.2012 21:14, Juergen Lock wrote:

In article 50BA09CC.2040104@flensrocker.de you write:

...

Am 30.11.2012 11:32, schrieb Gerald Dachs:

...

Am 2012-11-30 10:17, schrieb Lars Hanisch:

...

Looks like the pointer returned by sscanf is not valid:

32: bool tComponent::FromString(const char *s) 33: { 34: unsigned int Stream, Type; 35: int n = sscanf(s, "%X %02X %7s %a[^\n]", &Stream, &Type, language, &description); // 7 = MAXLANGCODE2 - 1 36: if (n != 4 || isempty(description)) { 37: free(description); 38: description = NULL; 39: } 40: stream = Stream; 41: type = Type; 42: return n >= 3; 43: }

From man sscanf:

    The GNU C library supports a nonstandard extension that causes the library to
    dynamically allocate a string of sufficient size for input strings for the %s
    and %a[range] conversion specifiers.

This is the reason why it doesn't work with ulibc.

Then there should be a malloc or something similiar for description:

32: bool tComponent::FromString(const char *s) 33: { 34: unsigned int Stream, Type; description = malloc(strlen(s)); description[0] = 0; 35: int n = sscanf(s, "%X %02X %7s %a[^\n]", &Stream, &Type, language, &description); // 7 = MAXLANGCODE2 - 1 36: if (n != 4 || isempty(description)) { 37: free(description); 38: description = NULL; 39: } 40: stream = Stream; 41: type = Type; 42: return n >= 3; 43: }

A check for description != NULL before the free call is not needed.

But this is not the only place in the vdr code, where %a is used...

Btw FreeBSD has the same problem (no *scanf %a in libc), so you could try taking the FreeBSD patches and replacing

#ifdef __FreeBSD__

by

#if defined(__FreeBSD__) || defined(__UCLIBC__)

in those parts handling %a here:

http://svnweb.freebsd.org/ports/head/multimedia/vdr/files/patch-vdr-1.7.28_F...

A few plugins are affected too: (of those we have in FreeBSD ports)

eepg:

http://svnweb.freebsd.org/ports/head/multimedia/vdr-plugin-eepg/files/patch-...

infosatepg:

http://svnweb.freebsd.org/ports/head/multimedia/vdr-plugin-infosatepg/files/...

ttxtsubs:

http://svnweb.freebsd.org/ports/head/multimedia/vdr-plugin-ttxtsubs/files/pa...

HTH, :) Juergen

PS: I think I never sent the FreeBSD patches to Klaus the last few times I updated them, atm FreeBSD ports is behind again (still at vdr 1.7.29), when I get around catching up to the latest version I can send the updated patches if you are interested, Klaus?

Well, in all honesty (and without meaning any offense or disrespect), I wouldn't like to have all these "#if ..." spread around the VDR source. For instance, things like a missing getline() function should not be patched right into the tools.c code, but rather be put into a separate file, maybe named oscompat.[hc]. In that file, depending on the OS in use, a local implementation of getline() could be implemented, and tools.c (like any other file that needs OS compatibility stuff) could just #include "oscompat.h" and the Makefile could link oscompat.o into the final executable.

There are also some patches that don't, at first glance, look like they would be BSD specific, like those in plugin.c. Are these just a different way of handling things (and would this also work on Linux)? Or is this whole shebang handled differently in BSD?

Well, finally for the root of this whole discussion: I really like the "%a" format character in sscanf(), because it is so simple and straightforward. I don't see why other libcs don't implement this, and I wouldn't want to have all that bulk in such areas. So I suggest you either get the libcs on those platforms to implement this (and thus solve the problem once and for all), or put your own version of sscanf() into the proposed oscompat.c and handle "%a" properly in there.

A few more 1.7.29 patches are in here:

http://svnweb.freebsd.org/ports/head/multimedia/vdr/files/

(I think I did send FreeBSD patches to those plugin maintainers that may still be active some time ago, haven't heard back from many tho.)

Klaus