On 1/20/2018 6:49 PM, Ken Brown wrote: > On 1/20/2018 7:23 AM, Ken Brown wrote: >> On 1/19/2018 10:27 PM, Ken Brown wrote: >>> On 1/18/2018 6:28 PM, Ken Brown wrote: >>>> On 1/18/2018 4:30 PM, Yaakov Selkowitz wrote: >>>>> On 2018-01-18 08:35, Ken Brown wrote: >>>>>> On 1/17/2018 5:29 PM, Ken Brown wrote: >>>>>>> Do we need a new gcc release to go along with the recent ssp >>>>>>> changes? >>>>>> >>>>>> The following commit message seems to answer my question: >>>>>> >>>>>> Note that this does require building gcc with >>>>>> --disable-libssp and >>>>>> gcc_cv_libc_provides_ssp=yes. >>>>> >>>>> Correct. >>>>> >>>>>> Are there plans to coordinate the release of Cygwin 2.10.0 with a new >>>>>> gcc release? In the meantime, I guess package maintainers have to >>>>>> build >>>>>> with -U_FORTIFY_SOURCE in order to test building with Cygwin >>>>>> 2.10.0. Or >>>>>> am I missing something? >>>>> >>>>> -D_FORTIFY_SOURCE is not the default, so simply omitting it is >>>>> sufficient. >>>> >>>> I was talking about building projects in which _FORTIFY_SOURCE is >>>> defined by default. That happens, for instance, in the gnulib >>>> subdirectory of the emacs tree, so it may affect other projects that >>>> use gnulib also. >>>> >>>>> You could also just delete >>>>> /usr/lib/gcc/*-pc-cygwin/6.4.0/include/ssp, since we won't need it >>>>> anymore and it wasn't even being used properly in the first place. >>>> >>>> That's a simpler workaround than what I was doing. Thanks. >>> >>> Here's another issue that's come up with _FORTIFY_SOURCE. One of the >>> emacs source files, fileio.c, makes use of a pointer to readlinkat. >>> [More precisely, the file uses an external function foo() with a >>> parameter 'bar' that's a pointer to a function; foo is called in >>> fileio.c with bar = readlinkat.] >>> >>> When _FORTIFY_SOURCE > 0, this leads to an "undefined reference to >>> `__ssp_protected_readlinkat'" linking error. Does this sound like >>> something that will be fixed with the new gcc release? >>> >>> I realize I haven't given you full details, but it might be a few >>> days until I have a chance to extract an STC for this issue, so I >>> thought I'd give it a shot. >>> >>> If you can't answer the question based on the information above, I'll >>> make an STC as soon as I can. >> >> I got to this sooner than expected: >> >> $ cat ssp_test.c >> #define _FORTIFY_SOURCE 1 >> #include <unistd.h> >> void foo (ssize_t (*preadlinkat) (int, char const *, char *, size_t)); >> >> void baz () >> { >> foo (readlinkat); >> } >> >> $ gcc -c -O1 ssp_test.c >> >> $ objdump -x ssp_test.o | grep readlinkat >> 6 .rdata$.refptr.__ssp_protected_readlinkat 00000010 >> 0000000000000000 0000000000000000 00000180 2**4 >> [...]
The following patch seems to fix the problem: --- ssp.h~ 2018-01-22 09:18:18.000000000 -0500 +++ ssp.h 2018-01-24 13:44:55.856635800 -0500 @@ -41,7 +41,7 @@ #endif #define __ssp_real(fun) __ssp_real_(fun) -#define __ssp_inline extern __inline__ __attribute__((__always_inline__, __gnu_inline__)) +#define __ssp_inline extern __inline__ __attribute__((__always_inline__)) #define __ssp_bos(ptr) __builtin_object_size(ptr, __SSP_FORTIFY_LEVEL > 1) #define __ssp_bos0(ptr) __builtin_object_size(ptr, 0) I arrived at this by comparing Cygwin's ssp.h with NetBSD's, on which Cygwin's was based, and I noticed that NetBSD didn't use __gnu_inline__. Yaakov, is there a reason that Cygwin needs __gnu_inline__? It apparently prevents fortified functions from being used as function pointers. Using my test case again, here's what happens with and without __gnu_inline__: With: $ gcc -O1 -c ssp_test.c && objdump -x ssp_test.o | grep readlinkat 6 .rdata$.refptr.__ssp_protected_readlinkat 00000010 0000000000000000 0000000000000000 00000180 2**4 CONTENTS, ALLOC, LOAD, RELOC, READONLY, DATA, LINK_ONCE_DISCARD (COMDAT .refptr.__ssp_protected_readlinkat 18) [ 4](sec 7)(fl 0x00)(ty 0)(scl 3) (nx 1) 0x0000000000000000 .rdata$.refptr.__ssp_protected_readlinkat [ 18](sec 7)(fl 0x00)(ty 0)(scl 2) (nx 0) 0x0000000000000000 .refptr.__ssp_protected_readlinkat [ 19](sec 0)(fl 0x00)(ty 0)(scl 2) (nx 0) 0x0000000000000000 __ssp_protected_readlinkat 0000000000000007 R_X86_64_PC32 __ssp_protected_readlinkat RELOCATION RECORDS FOR [.rdata$.refptr.__ssp_protected_readlinkat]: 0000000000000000 R_X86_64_64 __ssp_protected_readlinkat Without: $ gcc -O1 -c ssp_test.c && objdump -x ssp_test.o | grep readlinkat [ 2](sec 1)(fl 0x00)(ty 20)(scl 2) (nx 1) 0x0000000000000000 __ssp_protected_readlinkat [ 27](sec 0)(fl 0x00)(ty 0)(scl 2) (nx 0) 0x0000000000000000 readlinkat 0000000000000005 R_X86_64_PC32 readlinkat Ken -- Problem reports: http://cygwin.com/problems.html FAQ: http://cygwin.com/faq/ Documentation: http://cygwin.com/docs.html Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple