* Static linking of musl with code compiled using GNU header files @ 2014-03-14 15:47 David Grothe 2014-03-14 16:09 ` Luca Barbato ` (2 more replies) 0 siblings, 3 replies; 11+ messages in thread From: David Grothe @ 2014-03-14 15:47 UTC (permalink / raw) To: musl; +Cc: Support at Gcom [-- Attachment #1: Type: text/plain, Size: 1463 bytes --] Hello, I have a very large code base that I have been compiling on Linux using the standard GNU C compiler [gcc (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3]. I have been using shared object libraries, but for reasons of software support I would now like to link all my commands (a couple of dozen) and daemons using static libraries so that the code files are self-contained and can be copied, along with a core file, to any server back in my shop for analysis. With dynamic libraries I have to have exactly the same version of libc installed on the machine that I use to examine the core file as were present on the machine that generated the core file, or else gdb will not produce a stack back trace with file and line number information. So much for the background. I really don't want to port my code base to using the musl header files. I want to keep compiling with the GNU headers. When I do this and link my-huge-program.o with musl libc.a I get the following list of unresolved externals: U __divdi3 w __fini_array_end w __fini_array_start U __moddi3 U __sysv_signal U __udivdi3 U __umoddi3 U __vfprintf_chk U __vsnprintf_chk U __vsprintf_chk U __sysv_signal So, I am wondering if the musl library could at some point provide these routines to enable users to do what I am trying to do. Any possibility of that? Thanks, Dave [-- Attachment #2: Type: text/html, Size: 2426 bytes --] ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 15:47 Static linking of musl with code compiled using GNU header files David Grothe @ 2014-03-14 16:09 ` Luca Barbato 2014-03-14 16:29 ` Szabolcs Nagy 2014-03-14 16:47 ` Rich Felker 2 siblings, 0 replies; 11+ messages in thread From: Luca Barbato @ 2014-03-14 16:09 UTC (permalink / raw) To: musl On 14/03/14 16:47, David Grothe wrote: > I really don't want to port my code base to using the musl header > files. There shouldn't be anything to port. > I want to keep compiling with the GNU headers. When I do this > and link my-huge-program.o with musl libc.a I get the following list of > unresolved externals: > > U __divdi3 > w __fini_array_end > w __fini_array_start > U __moddi3 > U __sysv_signal > U __udivdi3 > U __umoddi3 > U __vfprintf_chk > U __vsnprintf_chk > U __vsprintf_chk > U __sysv_signal > > So, I am wondering if the musl library could at some point provide these > routines to enable users to do what I am trying to do. Not sure, but those internal symbols are close to implementation details... lu ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 15:47 Static linking of musl with code compiled using GNU header files David Grothe 2014-03-14 16:09 ` Luca Barbato @ 2014-03-14 16:29 ` Szabolcs Nagy 2014-03-14 18:52 ` David Grothe 2014-03-14 16:47 ` Rich Felker 2 siblings, 1 reply; 11+ messages in thread From: Szabolcs Nagy @ 2014-03-14 16:29 UTC (permalink / raw) To: musl; +Cc: Support at Gcom * David Grothe <dave@gcom.com> [2014-03-14 10:47:31 -0500]: > I have a very large code base that I have been compiling on Linux > using the standard GNU C compiler [gcc (Ubuntu/Linaro > 4.6.3-1ubuntu5) 4.6.3]. I have been using shared object libraries, > but for reasons of software support I would now like to link all my > commands (a couple of dozen) and daemons using static libraries so > that the code files are self-contained and can be copied, along with > a core file, to any server back in my shop for analysis. With > dynamic libraries I have to have exactly the same version of libc > installed on the machine that I use to examine the core file as were > present on the machine that generated the core file, or else gdb > will not produce a stack back trace with file and line number > information. So much for the background. > > I really don't want to port my code base to using the musl header > files. I want to keep compiling with the GNU headers. When I do compiling with the gnu headers is broken and it depends on the cflags used > this and link my-huge-program.o with musl libc.a I get the following > list of unresolved externals: > > U __divdi3 comes from libgcc.a, if it's missing you have a toolchain issue > w __fini_array_end > w __fini_array_start i think musl supports init/fini arrays (see src/exit/exit.c) > U __moddi3 libgcc > U __sysv_signal you may want to replace it with signal > U __udivdi3 > U __umoddi3 libgcc > U __vfprintf_chk > U __vsnprintf_chk > U __vsprintf_chk there are many _chk functions for _FORTIFY_SOURCE, musl may provide these eventually, until then you can add your own chk.o with dummy implementations (possibly with the safety checks i omit here): int __vfprintf_chk(FILE *f, int flag, const char *fmt, va_list ap) { return vfprintf(f, fmt, ap); } int __vsnprintf_chk(char *s, size_t n, int flag, size_t size, const char *fmt, va_list ap) { return vsnprintf(s, n, fmt, ap); } int __vsprintf_chk(char *s, int flag, size_t size, const char *fmt, va_list ap) { return vsprintf(s, fmt, ap); } > U __sysv_signal use signal > So, I am wondering if the musl library could at some point provide > these routines to enable users to do what I am trying to do. compiling with glibc headers and then linking to musl cannot be supported in general, because of ABI compat issues (eg glibc headers define PTHREAD_*_INITIALIZER macros that hardcode glibc internal ABI at compile time that does not match musl) if you are sure you don't have such ABI breakage (see glibc vs musl differences on the wiki) then you may get away by adding a glibc-compat.o to your musl build > > Any possibility of that? > > Thanks, > Dave ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 16:29 ` Szabolcs Nagy @ 2014-03-14 18:52 ` David Grothe 2014-03-14 19:25 ` Kurt H Maier 2014-03-14 21:04 ` David Grothe 0 siblings, 2 replies; 11+ messages in thread From: David Grothe @ 2014-03-14 18:52 UTC (permalink / raw) To: musl; +Cc: Support at Gcom [-- Attachment #1: Type: text/plain, Size: 3748 bytes --] Thanks for the suggestions. My musl build did not include a libgcc: linuxsvr:dave:musl-0.9.15> find . -name '*libgcc*' linuxsvr:dave:musl-0.9.15> It is correct that something in the GNU headers changed "signal" into "sysv_signal" without my knowledge. My code base is several million lines of code and I have many other projects to do that are higher priority than porting to another set of header files. It would be a few days worth of effort and I just have other things to do right now. That said I do have a reason for wanting static linking, so maybe I will find the time to do the port some time. (I tried just aiming my build at the musl include directory and it did not "just work".) I can act on the suggestions made and see how that helps. But what about libgcc? Thanks, Dave On 3/14/2014 11:29 AM, Szabolcs Nagy wrote: > * David Grothe <dave@gcom.com> [2014-03-14 10:47:31 -0500]: >> I have a very large code base that I have been compiling on Linux >> using the standard GNU C compiler [gcc (Ubuntu/Linaro >> 4.6.3-1ubuntu5) 4.6.3]. I have been using shared object libraries, >> but for reasons of software support I would now like to link all my >> commands (a couple of dozen) and daemons using static libraries so >> that the code files are self-contained and can be copied, along with >> a core file, to any server back in my shop for analysis. With >> dynamic libraries I have to have exactly the same version of libc >> installed on the machine that I use to examine the core file as were >> present on the machine that generated the core file, or else gdb >> will not produce a stack back trace with file and line number >> information. So much for the background. >> >> I really don't want to port my code base to using the musl header >> files. I want to keep compiling with the GNU headers. When I do > compiling with the gnu headers is broken and > it depends on the cflags used > >> this and link my-huge-program.o with musl libc.a I get the following >> list of unresolved externals: >> >> U __divdi3 > comes from libgcc.a, if it's missing you have a toolchain issue > >> w __fini_array_end >> w __fini_array_start > i think musl supports init/fini arrays > (see src/exit/exit.c) > >> U __moddi3 > libgcc > >> U __sysv_signal > you may want to replace it with signal > >> U __udivdi3 >> U __umoddi3 > libgcc > >> U __vfprintf_chk >> U __vsnprintf_chk >> U __vsprintf_chk > there are many _chk functions for _FORTIFY_SOURCE, musl may provide > these eventually, until then you can add your own chk.o with dummy > implementations (possibly with the safety checks i omit here): > > int __vfprintf_chk(FILE *f, int flag, const char *fmt, va_list ap) > { > return vfprintf(f, fmt, ap); > } > int __vsnprintf_chk(char *s, size_t n, int flag, size_t size, const char *fmt, va_list ap) > { > return vsnprintf(s, n, fmt, ap); > } > int __vsprintf_chk(char *s, int flag, size_t size, const char *fmt, va_list ap) > { > return vsprintf(s, fmt, ap); > } > >> U __sysv_signal > use signal > >> So, I am wondering if the musl library could at some point provide >> these routines to enable users to do what I am trying to do. > compiling with glibc headers and then linking to musl > cannot be supported in general, because of ABI compat issues > > (eg glibc headers define PTHREAD_*_INITIALIZER macros that hardcode > glibc internal ABI at compile time that does not match musl) > > if you are sure you don't have such ABI breakage (see glibc > vs musl differences on the wiki) then you may get away by > adding a glibc-compat.o to your musl build > >> Any possibility of that? >> >> Thanks, >> Dave [-- Attachment #2: Type: text/html, Size: 5279 bytes --] ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 18:52 ` David Grothe @ 2014-03-14 19:25 ` Kurt H Maier 2014-03-14 19:35 ` David Grothe 2014-03-14 21:04 ` David Grothe 1 sibling, 1 reply; 11+ messages in thread From: Kurt H Maier @ 2014-03-14 19:25 UTC (permalink / raw) To: musl Quoting David Grothe <dave@gcom.com>: > My musl build did not include a libgcc: libgcc comes with gcc. http://gcc.gnu.org/onlinedocs/gccint/Libgcc.html khm ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 19:25 ` Kurt H Maier @ 2014-03-14 19:35 ` David Grothe 0 siblings, 0 replies; 11+ messages in thread From: David Grothe @ 2014-03-14 19:35 UTC (permalink / raw) To: musl [-- Attachment #1: Type: text/plain, Size: 248 bytes --] Got it. Thanks. -- Dave On 3/14/2014 2:25 PM, Kurt H Maier wrote: > Quoting David Grothe <dave@gcom.com>: > >> My musl build did not include a libgcc: > > libgcc comes with gcc. > > http://gcc.gnu.org/onlinedocs/gccint/Libgcc.html > > khm > > > [-- Attachment #2: Type: text/html, Size: 1011 bytes --] ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 18:52 ` David Grothe 2014-03-14 19:25 ` Kurt H Maier @ 2014-03-14 21:04 ` David Grothe 2014-03-14 21:37 ` John Spencer 2014-03-15 0:22 ` Rich Felker 1 sibling, 2 replies; 11+ messages in thread From: David Grothe @ 2014-03-14 21:04 UTC (permalink / raw) To: musl; +Cc: Support at Gcom [-- Attachment #1: Type: text/plain, Size: 7695 bytes --] I built a shim module that defined all the undefined "__" routines that showed up in my link. Then all my programs linked successfully. But when I went to run one of my daemon processes it got a segv in the malloc code, as follows. Program terminated with signal 11, Segmentation fault. #0 0x0811cd5d in unbin (c=0x9b53898, i=8) at src/malloc/malloc.c:242 #1 0x0811d266 in malloc (n=112) at src/malloc/malloc.c:371 #2 0x0804b3ce in ssd_malloc_fcn (nbytes=16, file=0x81348e6 "../pi.c", linenr=2398) at ../pi.c:632 #3 0x0804b597 in ssd_zalloc_fcn (nbytes=12, file=0x81348e6 "../pi.c", linenr=2398) at ../pi.c:687 #4 0x0804b5e2 in ssd_calloc_fcn (n_memb=1, memb_size=12, file=0x81348e6 "../pi.c", linenr=2398) at ../pi.c:696 #5 0x0804ef18 in ss_setup_code_path (size=1024) at ../pi.c:2398 #6 0x080548be in register_connections () at ../pi.c:5074 #7 0x0805a2b8 in main (argc=2, argv=0xbfae15f4) at ../pi.c:7393 (gdb) p *c $1 = {psize = 17, csize = 144, next = 0x81a3990, prev = 0x1} (gdb) p mal $2 = {brk = 163028992, heap = 0x9b53008, binmap = 35184372089088, bins = {{lock = {0, 0}, head = 0x0, tail = 0x0}, {lock = {0, 0}, head = 0x81a3920, tail = 0x81a3920}, {lock = {0, 0}, head = 0x81a3930, tail = 0x81a3930}, {lock = {0, 0}, head = 0x81a3940, tail = 0x81a3940}, {lock = {0, 0}, head = 0x0, tail = 0x0}, {lock = {0, 0}, head = 0x81a3960, tail = 0x81a3960}, { lock = {0, 0}, head = 0x81a3970, tail = 0x81a3970}, {lock = {0, 0}, head = 0x81a3980, tail = 0x81a3980}, {lock = {0, 0}, head = 0x9b53898, tail = 0x9b53898}, {lock = {0, 0}, head = 0x81a39a0, tail = 0x81a39a0}, {lock = {0, 0}, head = 0x81a39b0, tail = 0x81a39b0}, {lock = {0, 0}, head = 0x81a39c0, tail = 0x81a39c0}, {lock = {0, 0}, head = 0x81a39d0, tail = 0x81a39d0}, {lock = {0, 0}, head = 0x81a39e0, tail = 0x81a39e0}, {lock = {0, 0}, head = 0x81a39f0, tail = 0x81a39f0}, {lock = {0, 0}, head = 0x81a3a00, tail = 0x81a3a00}, {lock = {0, 0}, head = 0x81a3a10, tail = 0x81a3a10}, {lock = {0, 0}, head = 0x0, tail = 0x0}, {lock = {0, 0}, head = 0x81a3a30, tail = 0x81a3a30}, {lock = {0, 0}, head = 0x81a3a40, tail = 0x81a3a40}, {lock = {0, 0}, head = 0x81a3a50, tail = 0x81a3a50}, {lock = {0, 0}, head = 0x81a3a60, tail = 0x81a3a60}, {lock = {0, 0}, head = 0x0, tail = 0x0}, {lock = {0, 0}, head = 0x81a3a80, tail = 0x81a3a80}, {lock = {0, 0}, head = 0x81a3a90, tail = 0x81a3a90}, {lock = {0, 0}, head = 0x81a3aa0, tail = 0x81a3aa0}, {lock = {0, 0}, head = 0x0, tail = 0x0}, {lock = {0, 0}, head = 0x81a3ac0, tail = 0x81a3ac0}, {lock = {0, 0}, head = 0x81a3ad0, tail = 0x81a3ad0}, {lock = {0, 0}, head = 0x81a3ae0, tail = 0x81a3ae0}, {lock = {0, 0}, head = 0x81a3af0, tail = 0x81a3af0}, {lock = {0, 0}, head = 0x0, tail = 0x0}, {lock = {0, 0}, head = 0x81a3b10, tail = 0x81a3b10}, {lock = {0, 0}, head = 0x81a3b20, tail = 0x81a3b20}, {lock = {0, 0}, head = 0x81a3b30, tail = 0x81a3b30}, {lock = {0, 0}, head = 0x81a3b40, tail = 0x81a3b40}, {lock = {0, 0}, head = 0x81a3b50, tail = 0x81a3b50}, {lock = {0, 0}, head = 0x81a3b60, tail = 0x81a3b60}, {lock = {0, 0}, head = 0x81a3b70, tail = 0x81a3b70}, {lock = {0, 0}, head = 0x81a3b80, tail = 0x81a3b80}, {lock = {0, 0}, head = 0x81a3b90, tail = 0x81a3b90}, {lock = {0, 0}, head = 0x81a3ba0, tail = 0x81a3ba0}, {lock = {0, 0}, head = 0x81a3bb0, tail = 0x81a3bb0}, {lock = {0, 0}, head = 0x81a3bc0, tail = 0x81a3bc0}, {lock = {0, 0}, head = 0x81a3bd0, tail = 0x81a3bd0}, {lock = {0, 0}, head = 0x9b78888, tail = 0x9b78888}, {lock = {0, 0}, head = 0x81a3bf0, tail = 0x81a3bf0}, {lock = {0, 0}, head = 0x0, tail = 0x0} <repeats 17 times>}, brk_lock = {0, 0}, free_lock = {0, 0}} I can supply more details as needed. Thanks, Dave On 3/14/2014 1:52 PM, David Grothe wrote: > Thanks for the suggestions. > > My musl build did not include a libgcc: > > linuxsvr:dave:musl-0.9.15> find . -name '*libgcc*' > linuxsvr:dave:musl-0.9.15> > > It is correct that something in the GNU headers changed "signal" into > "sysv_signal" without my knowledge. > > My code base is several million lines of code and I have many other > projects to do that are higher priority than porting to another set of > header files. It would be a few days worth of effort and I just have > other things to do right now. > > That said I do have a reason for wanting static linking, so maybe I > will find the time to do the port some time. (I tried just aiming my > build at the musl include directory and it did not "just work".) > > I can act on the suggestions made and see how that helps. But what > about libgcc? > > Thanks, > Dave > > On 3/14/2014 11:29 AM, Szabolcs Nagy wrote: >> * David Grothe<dave@gcom.com> [2014-03-14 10:47:31 -0500]: >>> I have a very large code base that I have been compiling on Linux >>> using the standard GNU C compiler [gcc (Ubuntu/Linaro >>> 4.6.3-1ubuntu5) 4.6.3]. I have been using shared object libraries, >>> but for reasons of software support I would now like to link all my >>> commands (a couple of dozen) and daemons using static libraries so >>> that the code files are self-contained and can be copied, along with >>> a core file, to any server back in my shop for analysis. With >>> dynamic libraries I have to have exactly the same version of libc >>> installed on the machine that I use to examine the core file as were >>> present on the machine that generated the core file, or else gdb >>> will not produce a stack back trace with file and line number >>> information. So much for the background. >>> >>> I really don't want to port my code base to using the musl header >>> files. I want to keep compiling with the GNU headers. When I do >> compiling with the gnu headers is broken and >> it depends on the cflags used >> >>> this and link my-huge-program.o with musl libc.a I get the following >>> list of unresolved externals: >>> >>> U __divdi3 >> comes from libgcc.a, if it's missing you have a toolchain issue >> >>> w __fini_array_end >>> w __fini_array_start >> i think musl supports init/fini arrays >> (see src/exit/exit.c) >> >>> U __moddi3 >> libgcc >> >>> U __sysv_signal >> you may want to replace it with signal >> >>> U __udivdi3 >>> U __umoddi3 >> libgcc >> >>> U __vfprintf_chk >>> U __vsnprintf_chk >>> U __vsprintf_chk >> there are many _chk functions for _FORTIFY_SOURCE, musl may provide >> these eventually, until then you can add your own chk.o with dummy >> implementations (possibly with the safety checks i omit here): >> >> int __vfprintf_chk(FILE *f, int flag, const char *fmt, va_list ap) >> { >> return vfprintf(f, fmt, ap); >> } >> int __vsnprintf_chk(char *s, size_t n, int flag, size_t size, const char *fmt, va_list ap) >> { >> return vsnprintf(s, n, fmt, ap); >> } >> int __vsprintf_chk(char *s, int flag, size_t size, const char *fmt, va_list ap) >> { >> return vsprintf(s, fmt, ap); >> } >> >>> U __sysv_signal >> use signal >> >>> So, I am wondering if the musl library could at some point provide >>> these routines to enable users to do what I am trying to do. >> compiling with glibc headers and then linking to musl >> cannot be supported in general, because of ABI compat issues >> >> (eg glibc headers define PTHREAD_*_INITIALIZER macros that hardcode >> glibc internal ABI at compile time that does not match musl) >> >> if you are sure you don't have such ABI breakage (see glibc >> vs musl differences on the wiki) then you may get away by >> adding a glibc-compat.o to your musl build >> >>> Any possibility of that? >>> >>> Thanks, >>> Dave > [-- Attachment #2: Type: text/html, Size: 10449 bytes --] ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 21:04 ` David Grothe @ 2014-03-14 21:37 ` John Spencer 2014-03-15 0:09 ` Rich Felker 2014-03-15 0:22 ` Rich Felker 1 sibling, 1 reply; 11+ messages in thread From: John Spencer @ 2014-03-14 21:37 UTC (permalink / raw) To: musl David Grothe wrote: > I built a shim module that defined all the undefined "__" routines that > showed up in my link. Then all my programs linked successfully. But > when I went to run one of my daemon processes it got a segv in the > malloc code, as follows. on which platform is this ? the linaro bit in your toolchain suggests that it is ARM. is that correct ? and which version of musl ? limited ABI compat is only there for x86 platforms. as for your problem below, it's possible that something else calls sbrk() messing up musl's allocator. you should check strace output to see if sbrk(0) is called more than once. also make sure that nothing pulls in glibc's libc.so. btw did you check your code against the ABI checklist that was pointed out earlier ? to me, it's not very surprising that your broken usage breaks and invokes UB somewhere.. > > Program terminated with signal 11, Segmentation fault. > #0 0x0811cd5d in unbin (c=0x9b53898, i=8) at src/malloc/malloc.c:242 > #1 0x0811d266 in malloc (n=112) at src/malloc/malloc.c:371 > #2 0x0804b3ce in ssd_malloc_fcn (nbytes=16, file=0x81348e6 "../pi.c", > linenr=2398) at ../pi.c:632 > #3 0x0804b597 in ssd_zalloc_fcn (nbytes=12, file=0x81348e6 "../pi.c", > linenr=2398) at ../pi.c:687 > #4 0x0804b5e2 in ssd_calloc_fcn (n_memb=1, memb_size=12, file=0x81348e6 ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 21:37 ` John Spencer @ 2014-03-15 0:09 ` Rich Felker 0 siblings, 0 replies; 11+ messages in thread From: Rich Felker @ 2014-03-15 0:09 UTC (permalink / raw) To: musl On Fri, Mar 14, 2014 at 10:37:08PM +0100, John Spencer wrote: > David Grothe wrote: > >I built a shim module that defined all the undefined "__" routines > >that showed up in my link. Then all my programs linked > >successfully. But when I went to run one of my daemon processes > >it got a segv in the malloc code, as follows. > > on which platform is this ? the linaro bit in your toolchain > suggests that it is ARM. is that correct ? and which version of musl > ? > limited ABI compat is only there for x86 platforms. While it's untested on anything but x86, the ABI compatibility should be comparable on arm, mips, microblaze, and sh. OTOH powerpc is known to be ABI-incompatible for multiple reasons (which basically amount to the glibc powerpc ABI being really bad) and x86_64 has some minor incompatibilities due to glibc bugs (regoff_t being the wrong size) that we'll eventually work around by making the dynamic linker detect glibc-linked callers and redirect calls to regexec to a fixup wrapper (but I don't see an easy way to do the same for static linking). > as for your problem below, it's possible that something else calls > sbrk() messing up musl's allocator. > you should check strace output to see if sbrk(0) is called more than once. > also make sure that nothing pulls in glibc's libc.so. In latest musl, sbrk is dummied out, so it's probably unlikely that this is the issue. > btw did you check your code against the ABI checklist that was > pointed out earlier ? From nsz's email? If so, I'm not sure that's quite a "checklist". But it's important to be aware that trying to rely on the "ABI compat" will potentially hide problems where your program is using a glibc feature that musl does not provide. (This would likely be caught at compile-time if you were using the musl headers, e.g. due to missing macro constants.) Rich ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 21:04 ` David Grothe 2014-03-14 21:37 ` John Spencer @ 2014-03-15 0:22 ` Rich Felker 1 sibling, 0 replies; 11+ messages in thread From: Rich Felker @ 2014-03-15 0:22 UTC (permalink / raw) To: musl; +Cc: Support at Gcom On Fri, Mar 14, 2014 at 04:04:50PM -0500, David Grothe wrote: > I built a shim module that defined all the undefined "__" routines > that showed up in my link. Then all my programs linked > successfully. But when I went to run one of my daemon processes it > got a segv in the malloc code, as follows. > > Program terminated with signal 11, Segmentation fault. > #0 0x0811cd5d in unbin (c=0x9b53898, i=8) at src/malloc/malloc.c:242 > #1 0x0811d266 in malloc (n=112) at src/malloc/malloc.c:371 > #2 0x0804b3ce in ssd_malloc_fcn (nbytes=16, file=0x81348e6 > "../pi.c", linenr=2398) at ../pi.c:632 > #3 0x0804b597 in ssd_zalloc_fcn (nbytes=12, file=0x81348e6 > "../pi.c", linenr=2398) at ../pi.c:687 > #4 0x0804b5e2 in ssd_calloc_fcn (n_memb=1, memb_size=12, > file=0x81348e6 "../pi.c", linenr=2398) at ../pi.c:696 > #5 0x0804ef18 in ss_setup_code_path (size=1024) at ../pi.c:2398 > #6 0x080548be in register_connections () at ../pi.c:5074 > #7 0x0805a2b8 in main (argc=2, argv=0xbfae15f4) at ../pi.c:7393 > (gdb) p *c > $1 = {psize = 17, csize = 144, next = 0x81a3990, prev = 0x1} The crashing line is: c->prev->next = c->next; Based on this and your gdb print of *c, it looks like the chunk malloc is trying to pull from the bin has had its contents (where it stores its membership in the linked list of free chunks) clobbered, most likely by your program. This is probably a use-after-free error. At the very least, c->prev has been clobbered; it's also possible that c->next was clobbered. You could try printing *c->next to see if it looks like a valid chunk header (i can tell you if you send it to the list). Looking for the code that called free((void *)0x9b538a0) might be a good way to track this down. Rich ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Static linking of musl with code compiled using GNU header files 2014-03-14 15:47 Static linking of musl with code compiled using GNU header files David Grothe 2014-03-14 16:09 ` Luca Barbato 2014-03-14 16:29 ` Szabolcs Nagy @ 2014-03-14 16:47 ` Rich Felker 2 siblings, 0 replies; 11+ messages in thread From: Rich Felker @ 2014-03-14 16:47 UTC (permalink / raw) To: musl; +Cc: Support at Gcom On Fri, Mar 14, 2014 at 10:47:31AM -0500, David Grothe wrote: > Hello, > > I have a very large code base that I have been compiling on Linux > using the standard GNU C compiler [gcc (Ubuntu/Linaro > 4.6.3-1ubuntu5) 4.6.3]. I have been using shared object libraries, > but for reasons of software support I would now like to link all my > commands (a couple of dozen) and daemons using static libraries so > that the code files are self-contained and can be copied, along with > a core file, to any server back in my shop for analysis. With > dynamic libraries I have to have exactly the same version of libc > installed on the machine that I use to examine the core file as were > present on the machine that generated the core file, or else gdb > will not produce a stack back trace with file and line number > information. So much for the background. > > I really don't want to port my code base to using the musl header > files. I want to keep compiling with the GNU headers. When I do > this and link my-huge-program.o with musl libc.a I get the following > list of unresolved externals: > > U __divdi3 > w __fini_array_end > w __fini_array_start > U __moddi3 > U __sysv_signal > U __udivdi3 > U __umoddi3 > U __vfprintf_chk > U __vsnprintf_chk > U __vsprintf_chk > U __sysv_signal The presence of __divdi3, __moddi3, __udivdi3, and __umoddi3 in this list indicates that you're missing libgcc.a. If you're using -nostdlib, you need to manually add libgcc back to the linker command line. __fini_array_start and __fini_array_end are provided by the linker and are not necessary unless your code has global destructors that the compiler is implementing via fini_array (this is why they're weak). The rest are __sysv_signal and __*_chk. The former looks suspicious: I really doubt you _want_ to be using the sysv version of signal(); it probably got pulled in by glibc's headers due to bad feature test macros or something. As for the latter, these come from _FORTIFY_SOURCE which musl does not yet support. > So, I am wondering if the musl library could at some point provide > these routines to enable users to do what I am trying to do. > > Any possibility of that? Likely for at least some of them, but not right away. And there are at least a few features (e.g. pthread cancellation) that will never work this way. BTW is there a reason you want to use glibc's headers with musl? If your program is having lots of build errors with musl's, it's probably indicative of problems you should fix; some of these problems may become problems with future glibc versions too. Rich ^ permalink raw reply [flat|nested] 11+ messages in thread
end of thread, other threads:[~2014-03-15 0:22 UTC | newest] Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2014-03-14 15:47 Static linking of musl with code compiled using GNU header files David Grothe 2014-03-14 16:09 ` Luca Barbato 2014-03-14 16:29 ` Szabolcs Nagy 2014-03-14 18:52 ` David Grothe 2014-03-14 19:25 ` Kurt H Maier 2014-03-14 19:35 ` David Grothe 2014-03-14 21:04 ` David Grothe 2014-03-14 21:37 ` John Spencer 2014-03-15 0:09 ` Rich Felker 2014-03-15 0:22 ` Rich Felker 2014-03-14 16:47 ` Rich Felker
Code repositories for project(s) associated with this public inbox https://git.vuxu.org/mirror/musl/ This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).