Enable Data Execution Protection and Address Space Layout Randomization on Windows
cover Enable Data Execution Protection and Address Space Layout Randomization on Windows
These two techniques make it harder to come up with exploits, by reducing what is commonly called the "attack surface" in security circles: by making the addresses less predictable, and by making it harder to inject data that is then (mis-)interpreted as code, this hardens Git's executables on Windows.
These patches have been carried in Git for Windows for over 3 years, and should therefore be considered battle-tested.
İsmail Dönmez (2):
mingw: do not let ld strip relocations
mingw: enable DEP and ASLR
config.mak.uname | 8 ++++++++ 1 file changed, 8 insertions(+)
base-commit: 39ffebd23b1ef6830bf86043ef0b5c069d9299a9
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-134%2Fdscho%2Faslr-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-134/dscho/aslr-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/134
Signature · 1 line
gitgitgadget
PATCH 2/2 mingw: enable DEP and ASLR
From: =?UTF-8?q?=C4=B0smail=20D=C3=B6nmez?= <[email protected]>
Enable DEP (Data Execution Prevention) and ASLR (Address Space Layout Randomization) support. This applies to both 32bit and 64bit builds and makes it substantially harder to exploit security holes in Git by offering a much more unpredictable attack surface.
ASLR interferes with GDB's ability to set breakpoints. A similar issue holds true when compiling with -O2 (in which case single-stepping is messed up because GDB cannot map the code back to the original source code properly). Therefore we simply enable ASLR only when an optimization flag is present in the CFLAGS, using it as an indicator that the developer does not want to debug in GDB anyway.
--- config.mak.uname | 6 ++++++ 1 file changed, 6 insertions(+)
config.mak.uname +6 -0
@@ -570,6 +570,12 @@ else ifeq ($(shell expr "$(uname_R)" : '2\.'),2) # MSys2 prefix = /usr/ + # Enable DEP + BASIC_LDFLAGS += -Wl,--nxcompat + # Enable ASLR (unless debugging) + ifneq (,$(findstring -O,$(CFLAGS))) + BASIC_LDFLAGS += -Wl,--dynamicbase + endif ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686
Signature · 1 line
gitgitgadget
PATCH 1/2 mingw: do not let ld strip relocations
In reply to Johannes Schindelin via GitGitGadgetFrom: =?UTF-8?q?=C4=B0smail=20D=C3=B6nmez?= <[email protected]>
This is the first step for enabling ASLR (Address Space Layout Randomization) support. We want to enable ASLR for better protection against exploiting security holes in Git: it makes it harder to attack software by making code addresses unpredictable.
The problem fixed by this commit is that ld.exe seems to be stripping relocations which in turn will break ASLR support. We just make sure it's not stripping the main executable entry.
--- config.mak.uname | 2 ++ 1 file changed, 2 insertions(+)
config.mak.uname +2 -0
@@ -573,10 +573,12 @@ else ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686 + BASIC_LDFLAGS += -Wl,--pic-executable,-e,_mainCRTStartup endif ifeq (MINGW64,$(MSYSTEM)) prefix = /mingw64 HOST_CPU = x86_64 + BASIC_LDFLAGS += -Wl,--pic-executable,-e,mainCRTStartup else COMPAT_CFLAGS += -D_USE_32BIT_TIME_T BASIC_LDFLAGS += -Wl,--large-address-aware
Signature · 1 line
gitgitgadget
Re: [PATCH 2/2] mingw: enable DEP and ASLR
In reply to İsmail Dönmez via GitGitGadget[had to add Dscho as recipient manually, mind you]
Am 29.04.19 um 23:56 schrieb İsmail Dönmez via GitGitGadget:
Quoted 28 lines
From: =?UTF-8?q?=C4=B0smail=20D=C3=B6nmez?= <[email protected]>
Enable DEP (Data Execution Prevention) and ASLR (Address Space Layout Randomization) support. This applies to both 32bit and 64bit builds and makes it substantially harder to exploit security holes in Git by offering a much more unpredictable attack surface.
ASLR interferes with GDB's ability to set breakpoints. A similar issue holds true when compiling with -O2 (in which case single-stepping is messed up because GDB cannot map the code back to the original source code properly). Therefore we simply enable ASLR only when an optimization flag is present in the CFLAGS, using it as an indicator that the developer does not want to debug in GDB anyway.
Signed-off-by İsmail Dönmez <[email protected]>Signed-off-by Johannes Schindelin <[email protected]>--- config.mak.uname | 6 ++++++ 1 file changed, 6 insertions(+)config.mak.uname +0
-0@@ -570,6 +570,12 @@ else ifeq ($(shell expr "$(uname_R)" : '2\.'),2) # MSys2
config.mak.uname else · line 581
Quoted 10 lines
config.mak.uname +6
-0prefix = /usr/ + # Enable DEP + BASIC_LDFLAGS += -Wl,--nxcompat + # Enable ASLR (unless debugging) + ifneq (,$(findstring -O,$(CFLAGS))) + BASIC_LDFLAGS += -Wl,--dynamicbase + endif ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686
I'm a bit concerned that this breaks my debug sessions where I use -O0. But I'll test without -O0 before I really complain.
-- Hannes
Re: [PATCH 2/2] mingw: enable DEP and ASLR
Hi Hannes,
Quoted 1 line · Johannes Sixt
[had to add Dscho as recipient manually, mind you]
I usually pick up responses to GitGitGadget patch series even if I am not on explicit Cc: (but it might take a couple of days when I am too busy elsewhere to read the Git mailing list).
Quoted 34 lines
Am 29.04.19 um 23:56 schrieb İsmail Dönmez via GitGitGadget:
Quoted 33 lines
From: =?UTF-8?q?=C4=B0smail=20D=C3=B6nmez?= <[email protected]>
Enable DEP (Data Execution Prevention) and ASLR (Address Space Layout Randomization) support. This applies to both 32bit and 64bit builds and makes it substantially harder to exploit security holes in Git by offering a much more unpredictable attack surface.
ASLR interferes with GDB's ability to set breakpoints. A similar issue holds true when compiling with -O2 (in which case single-stepping is messed up because GDB cannot map the code back to the original source code properly). Therefore we simply enable ASLR only when an optimization flag is present in the CFLAGS, using it as an indicator that the developer does not want to debug in GDB anyway.
Signed-off-by İsmail Dönmez <[email protected]>Signed-off-by Johannes Schindelin <[email protected]>--- config.mak.uname | 6 ++++++ 1 file changed, 6 insertions(+)config.mak.uname +5
-0@@ -570,6 +570,12 @@ else ifeq ($(shell expr "$(uname_R)" : '2\.'),2) # MSys2 prefix = /usr/ + # Enable DEP + BASIC_LDFLAGS += -Wl,--nxcompat + # Enable ASLR (unless debugging) + ifneq (,$(findstring -O,$(CFLAGS))) + BASIC_LDFLAGS += -Wl,--dynamicbase
Quoted 8 lines
config.mak.uname else · line 581
Quoted 5 lines
config.mak.uname +1
-0+ endif ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686I'm a bit concerned that this breaks my debug sessions where I use -O0. But I'll test without -O0 before I really complain.
Weird. Jameson Miller also mentioned this very concern in an internal review.
I guess I'll do something like
ifneq (,$(findstring -O,$(filter-out -O0,$(CFLAGS))))
Does that work for you?
Ciao,
Dscho
Re: [PATCH 2/2] mingw: enable DEP and ASLR
Am 01.05.19 um 00:41 schrieb Johannes Schindelin:
Quoted 24 lines · Johannes Sixt
Quoted 14 lines
Am 29.04.19 um 23:56 schrieb İsmail Dönmez via GitGitGadget:
Quoted 13 lines
config.mak.uname +5
-0@@ -570,6 +570,12 @@ else ifeq ($(shell expr "$(uname_R)" : '2\.'),2) # MSys2 prefix = /usr/ + # Enable DEP + BASIC_LDFLAGS += -Wl,--nxcompat + # Enable ASLR (unless debugging) + ifneq (,$(findstring -O,$(CFLAGS))) + BASIC_LDFLAGS += -Wl,--dynamicbaseQuoted 8 lines
config.mak.uname else · line 581
Quoted 5 lines
config.mak.uname +1
-0+ endif ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686I'm a bit concerned that this breaks my debug sessions where I use -O0. But I'll test without -O0 before I really complain.
Quoted 8 lines
Weird. Jameson Miller also mentioned this very concern in an internal review.
I guess I'll do something like
ifneq (,$(findstring -O,$(filter-out -O0,$(CFLAGS))))Does that work for you?
That could work. I'm a bit distracted at the moment, so it may take some time until I can test.
-- Hannes
Re: [PATCH 2/2] mingw: enable DEP and ASLR
In reply to Johannes SchindelinHi Johannes,
Le 01/05/2019 à 00:41, Johannes Schindelin a écrit :
Quoted 22 lines
Hi Hannes,
Quoted 1 line · Johannes Sixt
[had to add Dscho as recipient manually, mind you]
I usually pick up responses to GitGitGadget patch series even if I am not on explicit Cc: (but it might take a couple of days when I am too busy elsewhere to read the Git mailing list).
Quoted 12 lines
Am 29.04.19 um 23:56 schrieb İsmail Dönmez via GitGitGadget:
Quoted 11 lines
From: =?UTF-8?q?=C4=B0smail=20D=C3=B6nmez?= <[email protected]>
Enable DEP (Data Execution Prevention) and ASLR (Address Space Layout Randomization) support. This applies to both 32bit and 64bit builds and makes it substantially harder to exploit security holes in Git by offering a much more unpredictable attack surface.
ASLR interferes with GDB's ability to set breakpoints. A similar issue holds true when compiling with -O2 (in which case single-stepping is messed up because GDB cannot map the code back to the original source code properly). Therefore we simply enable ASLR only when an
I don’t know if it stands true when combined with something like -ggdb3, but I may be very wrong. Feel free to correct me.
Quoted 31 lines
Quoted 22 lines
Quoted 22 lines
optimization flag is present in the CFLAGS, using it as an indicator that the developer does not want to debug in GDB anyway.
Signed-off-by İsmail Dönmez <[email protected]>Signed-off-by Johannes Schindelin <[email protected]>--- config.mak.uname | 6 ++++++ 1 file changed, 6 insertions(+)config.mak.uname +5
-0@@ -570,6 +570,12 @@ else ifeq ($(shell expr "$(uname_R)" : '2\.'),2) # MSys2 prefix = /usr/ + # Enable DEP + BASIC_LDFLAGS += -Wl,--nxcompat + # Enable ASLR (unless debugging) + ifneq (,$(findstring -O,$(CFLAGS))) + BASIC_LDFLAGS += -Wl,--dynamicbaseQuoted 8 lines
config.mak.uname else · line 581
Quoted 5 lines
config.mak.uname +1
-0+ endif ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686I'm a bit concerned that this breaks my debug sessions where I use -O0. But I'll test without -O0 before I really complain.
Quoted 7 lines
Weird. Jameson Miller also mentioned this very concern in an internal review.
I guess I'll do something like
ifneq (,$(findstring -O,$(filter-out -O0,$(CFLAGS))))
-Og also exists to debug[0], even if it’s far less known. Perhaps it’s better to check for -g (and its variants[1]) as the user clearly states their intent to debug the resulting binary, rather than checking for special cases.
Quoted 5 lines
Does that work for you?
Ciao,
Dscho
[0] https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-Og
[1] https://gcc.gnu.org/onlinedocs/gcc/Debugging-Options.html
Cheers,
Alban
Re: [PATCH 2/2] mingw: enable DEP and ASLR
In reply to Johannes SchindelinQuoted 11 lines · Johannes Schindelin
Quoted 2 lines
I'm a bit concerned that this breaks my debug sessions where I use -O0. But I'll test without -O0 before I really complain.
Weird. Jameson Miller also mentioned this very concern in an internal review.
I guess I'll do something like
ifneq (,$(findstring -O,$(filter-out -O0,$(CFLAGS))))Does that work for you?
I wonder if this points to this patch touching the wrong level. These compiler flags are a thing that _some_ builds want (i.e., production builds where people care most about security and not about debugging), but not necessarily all.
I'd have expected this to be tweakable by a Makefile knob (either a specific knob, or just the caller setting the right CFLAGS etc), and then for the builds of Git for Windows to turn those knobs when making a package to distribute.
Our internal package builds at GitHub all have this in their config.mak (for Linux, of course):
CFLAGS += -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=1
CFLAGS += -fstack-protector-strong
CFLAGS += -fpie
LDFLAGS += -z relro -z now
LDFLAGS += -pie
and I wouldn't be surprised if other binary distributors (like the Debian package) do something similar.
-Peff
Re: [PATCH 2/2] mingw: enable DEP and ASLR
Hi,
Quoted 16 lines · Jeff King
I wonder if this points to this patch touching the wrong level. These compiler flags are a thing that _some_ builds want (i.e., production builds where people care most about security and not about debugging), but not necessarily all.
I'd have expected this to be tweakable by a
Makefileknob (either a specific knob, or just the caller setting the right CFLAGS etc), and then for the builds of Git for Windows to turn those knobs when making a package to distribute.Our internal package builds at GitHub all have this in their config.mak (for Linux, of course):
CFLAGS += -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=1
CFLAGS += -fstack-protector-strong
Quoted 6 lines
CFLAGS += -fpie
LDFLAGS += -z relro -z now
LDFLAGS += -pieand I wouldn't be surprised if other binary distributors (like the Debian package) do something similar.
Yes, the Debian package uses
CFLAGS := -Wall \ $(shell dpkg-buildflags --get CFLAGS) \ $(shell dpkg-buildflags --get CPPFLAGS)
and then passes CFLAGS='$(CFLAGS)' to "make".
That means we're using
-g -O2 -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2
Dscho's suggestion for the Windows build sounds fine to me (if checking for -Og, too). Maybe it would make sense to factor out a makefile variable for this, that could be used for builds on other platforms, too. That way, the autodetection can be in one place, and there is a standard way to override it when the user wants something else.
Thanks,
Jonathan
Re: [PATCH 2/2] mingw: enable DEP and ASLR
In reply to Alban GruinQuoted 4 lines · Alban Gruin
-Og also exists to debug[0], even if it’s far less known. Perhaps it’s better to check for -g (and its variants[1]) as the user clearly states their intent to debug the resulting binary, rather than checking for special cases.
I can't speak for the Windows folks, but Debian frequently builds with -O2 -g and strips the debugging symbols into a separate package that can be installed in case of a crash. So -g need not be an indication of non-production use.
Signature · 2 lines
brian m. carlson: Houston, Texas, US OpenPGP: https://keybase.io/bk2204
Re: [PATCH 2/2] mingw: enable DEP and ASLR
In reply to Jonathan NiederHi Jonathan & Peff,
Quoted 38 lines · Jonathan Nieder
Quoted 16 lines · Jeff King
I wonder if this points to this patch touching the wrong level. These compiler flags are a thing that _some_ builds want (i.e., production builds where people care most about security and not about debugging), but not necessarily all.
I'd have expected this to be tweakable by a
Makefileknob (either a specific knob, or just the caller setting the right CFLAGS etc), and then for the builds of Git for Windows to turn those knobs when making a package to distribute.Our internal package builds at GitHub all have this in their config.mak (for Linux, of course):
CFLAGS += -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=1
CFLAGS += -fstack-protector-strongQuoted 6 lines
CFLAGS += -fpie
LDFLAGS += -z relro -z now
LDFLAGS += -pieand I wouldn't be surprised if other binary distributors (like the Debian package) do something similar.
Yes, the Debian package uses
CFLAGS := -Wall \ $(shell dpkg-buildflags --get CFLAGS) \ $(shell dpkg-buildflags --get CPPFLAGS)and then passes CFLAGS='$(CFLAGS)' to "make".
That means we're using
-g -O2 -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2
Quoted 6 lines
Dscho's suggestion for the Windows build sounds fine to me (if checking for -Og, too). Maybe it would make sense to factor out a makefile variable for this, that could be used for builds on other platforms, too. That way, the autodetection can be in one place, and there is a standard way to override it when the user wants something else.
Indeed, if I was to add a generic "are we building for production?" function, this would be incorrect.
But this is not the case here, we are doing something very specific, Windows-only here, and for the sole reason to keep debuggability (for which the presence of the -g option indeed would not be a good
indicator: in Git for Windows, we build .pdb files so that stackdumps can be more meaningful, but we do not want to have full debug information in those executables).
In the long run, I think we need to become more explicit about this, by adding a "FOR_PRODUCTION" flag. It's really no good if we use implementation details such as CFLAGS to deduce intent.
That's for another patch series, though, as it is pretty clear-cut here:
If you build with optimization flags using Git for Windows' SDK, you cannot use gdb for single-stepping, likewise if you use ASLR, so we can totally piggyback the latter onto the former.
Ciao,
Dscho
v2 cover Enable Data Execution Protection and Address Space Layout Randomization on Windows
In reply to Johannes Schindelin via GitGitGadgetThese two techniques make it harder to come up with exploits, by reducing what is commonly called the "attack surface" in security circles: by making the addresses less predictable, and by making it harder to inject data that is then (mis-)interpreted as code, this hardens Git's executables on Windows.
These patches have been carried in Git for Windows for over 3 years, and should therefore be considered battle-tested.
Changes since v1:
* When determining whether we build with optimization, -O0 and -Og are explicitly ignored.
İsmail Dönmez (2):
mingw: do not let ld strip relocations
mingw: enable DEP and ASLR
config.mak.uname | 8 ++++++++ 1 file changed, 8 insertions(+)
base-commit: 83232e38648b51abbcbdb56c94632b6906cc85a6
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-134%2Fdscho%2Faslr-v2
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-134/dscho/aslr-v2
Pull-Request: https://github.com/gitgitgadget/git/pull/134
Range-diff vs v1:
1: e6acdba586 = 1: 828913e96c mingw: do not let ld strip relocations
2: e142c1396e ! 2: 9f1da73829 mingw: enable DEP and ASLR
@@ -21,13 +21,13 @@
--- a/config.mak.uname
+++ b/config.mak.uname
@@
- ifeq ($(shell expr "$(uname_R)" : '2\.'),2)
+ ifneq ($(shell expr "$(uname_R)" : '1\.'),2)
# MSys2
prefix = /usr/
+ # Enable DEP
+ BASIC_LDFLAGS += -Wl,--nxcompat
+ # Enable ASLR (unless debugging)
-+ ifneq (,$(findstring -O,$(CFLAGS)))
++ ifneq (,$(findstring -O,$(filter-out -O0 -Og,$(CFLAGS))))
+ BASIC_LDFLAGS += -Wl,--dynamicbase
+ endif
ifeq (MINGW32,$(MSYSTEM))Signature · 1 line
gitgitgadget
PATCH v2 2/2 mingw: enable DEP and ASLR
From: =?UTF-8?q?=C4=B0smail=20D=C3=B6nmez?= <[email protected]>
Enable DEP (Data Execution Prevention) and ASLR (Address Space Layout Randomization) support. This applies to both 32bit and 64bit builds and makes it substantially harder to exploit security holes in Git by offering a much more unpredictable attack surface.
ASLR interferes with GDB's ability to set breakpoints. A similar issue holds true when compiling with -O2 (in which case single-stepping is messed up because GDB cannot map the code back to the original source code properly). Therefore we simply enable ASLR only when an optimization flag is present in the CFLAGS, using it as an indicator that the developer does not want to debug in GDB anyway.
--- config.mak.uname | 6 ++++++ 1 file changed, 6 insertions(+)
config.mak.uname +6 -0
@@ -573,6 +573,12 @@ else ifneq ($(shell expr "$(uname_R)" : '1\.'),2) # MSys2 prefix = /usr/ + # Enable DEP + BASIC_LDFLAGS += -Wl,--nxcompat + # Enable ASLR (unless debugging) + ifneq (,$(findstring -O,$(filter-out -O0 -Og,$(CFLAGS)))) + BASIC_LDFLAGS += -Wl,--dynamicbase + endif ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686
Signature · 1 line
gitgitgadget
PATCH v2 1/2 mingw: do not let ld strip relocations
In reply to Johannes Schindelin via GitGitGadgetFrom: =?UTF-8?q?=C4=B0smail=20D=C3=B6nmez?= <[email protected]>
This is the first step for enabling ASLR (Address Space Layout Randomization) support. We want to enable ASLR for better protection against exploiting security holes in Git: it makes it harder to attack software by making code addresses unpredictable.
The problem fixed by this commit is that ld.exe seems to be stripping relocations which in turn will break ASLR support. We just make sure it's not stripping the main executable entry.
--- config.mak.uname | 2 ++ 1 file changed, 2 insertions(+)
config.mak.uname +2 -0
@@ -576,10 +576,12 @@ else ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686 + BASIC_LDFLAGS += -Wl,--pic-executable,-e,_mainCRTStartup endif ifeq (MINGW64,$(MSYSTEM)) prefix = /mingw64 HOST_CPU = x86_64 + BASIC_LDFLAGS += -Wl,--pic-executable,-e,mainCRTStartup else COMPAT_CFLAGS += -D_USE_32BIT_TIME_T BASIC_LDFLAGS += -Wl,--large-address-aware
Signature · 1 line
gitgitgadget
Re: [PATCH 2/2] mingw: enable DEP and ASLR
In reply to Alban GruinHi Alban,
Quoted 18 lines · Alban Gruin
Le 01/05/2019 à 00:41, Johannes Schindelin a écrit :
Quoted 17 lines · Johannes Sixt
Quoted 2 lines
[had to add Dscho as recipient manually, mind you]
I usually pick up responses to GitGitGadget patch series even if I am not on explicit Cc: (but it might take a couple of days when I am too busy elsewhere to read the Git mailing list).
Quoted 8 lines
Am 29.04.19 um 23:56 schrieb İsmail Dönmez via GitGitGadget:
Quoted 7 lines
From: =?UTF-8?q?=C4=B0smail=20D=C3=B6nmez?= <[email protected]>
Enable DEP (Data Execution Prevention) and ASLR (Address Space Layout Randomization) support. This applies to both 32bit and 64bit builds and makes it substantially harder to exploit security holes in Git by offering a much more unpredictable attack surface.
Quoted 7 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
ASLR interferes with GDB's ability to set breakpoints. A similar issue holds true when compiling with -O2 (in which case single-stepping is messed up because GDB cannot map the code back to the original source code properly). Therefore we simply enable ASLR only when an
I don’t know if it stands true when combined with something like -ggdb3, but I may be very wrong. Feel free to correct me.
Possibly, but that makes my job here harder, so I won't even try right now ;-)
Quoted 34 lines
Quoted 34 lines
Quoted 22 lines
Quoted 22 lines
optimization flag is present in the CFLAGS, using it as an indicator that the developer does not want to debug in GDB anyway.
Signed-off-by İsmail Dönmez <[email protected]>Signed-off-by Johannes Schindelin <[email protected]>--- config.mak.uname | 6 ++++++ 1 file changed, 6 insertions(+)config.mak.uname +5
-0@@ -570,6 +570,12 @@ else ifeq ($(shell expr "$(uname_R)" : '2\.'),2) # MSys2 prefix = /usr/ + # Enable DEP + BASIC_LDFLAGS += -Wl,--nxcompat + # Enable ASLR (unless debugging) + ifneq (,$(findstring -O,$(CFLAGS))) + BASIC_LDFLAGS += -Wl,--dynamicbaseQuoted 8 lines
config.mak.uname else · line 581
Quoted 5 lines
config.mak.uname +1
-0+ endif ifeq (MINGW32,$(MSYSTEM)) prefix = /mingw32 HOST_CPU = i686I'm a bit concerned that this breaks my debug sessions where I use -O0. But I'll test without -O0 before I really complain.
Weird. Jameson Miller also mentioned this very concern in an internal review.
Quoted 6 lines
Quoted 4 lines
I guess I'll do something like
ifneq (,$(findstring -O,$(filter-out -O0,$(CFLAGS))))-Og also exists to debug[0], even if it’s far less known.
Good point.
Quoted 3 lines
Perhaps it’s better to check for -g (and its variants[1]) as the user clearly states their intent to debug the resulting binary, rather than checking for special cases.
I don't think we can use that, as we specifically build Git for Windows with optimization *and* with debug symbols (and then use cv2pdb to extract those debug symbols into external .pdb files for use with advanced post-mortem tools, i.e. we do *not* need to single-step).
Thanks,
Dscho
Re: [PATCH 2/2] mingw: enable DEP and ASLR
In reply to brian m. carlsonHi Brian,
Quoted 10 lines · brian m. carlson
Quoted 4 lines · Alban Gruin
-Og also exists to debug[0], even if it’s far less known. Perhaps it’s better to check for -g (and its variants[1]) as the user clearly states their intent to debug the resulting binary, rather than checking for special cases.
I can't speak for the Windows folks, but Debian frequently builds with -O2 -g and strips the debugging symbols into a separate package that can be installed in case of a crash. So -g need not be an indication of non-production use.
Precisely, Git for Windows imitates this strategy.
Thanks,
Dscho