Skip to contentSkip to conversation

db: Fix timespec length assumptions

PATCH In review Khem Raj 2 participants 3 messages Feb 14, 2020 – Feb 17, 2020

lore
Share & tools
lore

Khem Raj

author
#0 lore raw

PATCH 1/1 db: Fix timespec length assumptions

This should fix the vdso crashes seen with kernel 5.0+

python3[11312] general protection fault ip:b7e966b0 sp:bf8175cc error:0 in libc.so[b7e8b000+6b000]

Cc Bruce Ashfield <[email protected]>
---
 ...001-clock-Do-not-define-own-timespec.patch | 45 +++++++++++++++++++
 meta/recipes-support/db/db_5.3.28.bb          |  1 +
 2 files changed, 46 insertions(+)
 create mode 100644 meta/recipes-support/db/db/0001-clock-Do-not-define-own-timespec.patch
meta/recipes-support/db/db/0001-clock-Do-not-define-own-timespec.patch new file +45 -0
new file mode 100644
index 0000000000..9d51a44a26
@@ -0,0 +1,45 @@
+From 96b303caf70a7635953c36e5bfb9ad6e75cb7637 Mon Sep 17 00:00:00 2001
+From: Khem Raj <[email protected]>
+Date: Fri, 14 Feb 2020 14:12:59 -0800
+Subject: [PATCH] clock: Do not define own timespec
+
+timespec is provided by libc and its best left to libc
+os_gettime takes a db_timespec and passed its address to clock_gettime
+which assumes that db_timespec and timespec are same but actually
+its 12-bytes here and libc has 16-bytes
+
+This can cause problems especially with 64bit time_t
+
+Upstream-Status: Pending
+Signed-off-by: Khem Raj <[email protected]>
+---
+ src/dbinc/clock.h | 17 +----------------
+ 1 file changed, 1 insertion(+), 16 deletions(-)
+
+--- a/src/dbinc/clock.h
++++ b/src/dbinc/clock.h
+@@ -44,22 +44,8 @@
+ extern "C" {
+ #endif
+ 
+-/*
+- * This declaration is POSIX-compatible.  Because there are lots of different
+- * time.h include file patterns out there, it's easier to declare our own name
+- * in all cases than to try and discover if a system has a struct timespec.
+- * For the same reason, and because we'd have to #include <sys/time.h> in db.h,
+- * we don't export any timespec structures in the DB API, even in places where
+- * it would make sense, like the replication statistics information.
+- */
+-typedef struct {
+-	time_t	tv_sec;				/* seconds */
+-#ifdef HAVE_MIXED_SIZE_ADDRESSING
+-	int32_t tv_nsec;
+-#else
+-	long	tv_nsec;			/* nanoseconds */
+-#endif
+-} db_timespec;
++#include <time.h>
++#define db_timespec struct timespec
+ 
+ /* Operations on timespecs */
+ #undef	timespecclear
meta/recipes-support/db/db_5.3.28.bb +1 -0
index a2969055a6..badb79f9ea 100644
@@ -26,6 +26,7 @@ SRC_URI += "file://fix-parallel-build.patch \
             file://0001-configure-Add-explicit-tag-options-to-libtool-invoca.patch \
             file://sequence-type.patch \
             file://0001-Fix-libc-compatibility-by-renaming-atomic_init-API.patch \
+            file://0001-clock-Do-not-define-own-timespec.patch \
            "
 # We are not interested in official latest 6.x versions;
 # let's track what debian is using.
Signature · 1 line
2.25.0
#1 lore raw
Quoted 2 lines · Khem Raj

This should fix the vdso crashes seen with kernel 5.0+

cool. I'll test this locally and follow up over the weekend.

Bruce

Quoted 77 lines

python3[11312] general protection fault ip:b7e966b0 sp:bf8175cc error:0 in libc.so[b7e8b000+6b000]

Cc Bruce Ashfield <[email protected]>
---
 ...001-clock-Do-not-define-own-timespec.patch | 45 +++++++++++++++++++
 meta/recipes-support/db/db_5.3.28.bb          |  1 +
 2 files changed, 46 insertions(+)
 create mode 100644 meta/recipes-support/db/db/0001-clock-Do-not-define-own-timespec.patch
meta/recipes-support/db/db/0001-clock-Do-not-define-own-timespec.patch new file +45 -0
new file mode 100644
index 0000000000..9d51a44a26
@@ -0,0 +1,45 @@
+From 96b303caf70a7635953c36e5bfb9ad6e75cb7637 Mon Sep 17 00:00:00 2001
+From: Khem Raj <[email protected]>
+Date: Fri, 14 Feb 2020 14:12:59 -0800
+Subject: [PATCH] clock: Do not define own timespec
+
+timespec is provided by libc and its best left to libc
+os_gettime takes a db_timespec and passed its address to clock_gettime
+which assumes that db_timespec and timespec are same but actually
+its 12-bytes here and libc has 16-bytes
+
+This can cause problems especially with 64bit time_t
+
+Upstream-Status: Pending
+Signed-off-by: Khem Raj <[email protected]>
+---
+ src/dbinc/clock.h | 17 +----------------
+ 1 file changed, 1 insertion(+), 16 deletions(-)
+
+--- a/src/dbinc/clock.h
++++ b/src/dbinc/clock.h
+@@ -44,22 +44,8 @@
+ extern "C" {
+ #endif
+
+-/*
+- * This declaration is POSIX-compatible.  Because there are lots of different
+- * time.h include file patterns out there, it's easier to declare our own name
+- * in all cases than to try and discover if a system has a struct timespec.
+- * For the same reason, and because we'd have to #include <sys/time.h> in db.h,
+- * we don't export any timespec structures in the DB API, even in places where
+- * it would make sense, like the replication statistics information.
+- */
+-typedef struct {
+-      time_t  tv_sec;                         /* seconds */
+-#ifdef HAVE_MIXED_SIZE_ADDRESSING
+-      int32_t tv_nsec;
+-#else
+-      long    tv_nsec;                        /* nanoseconds */
+-#endif
+-} db_timespec;
++#include <time.h>
++#define db_timespec struct timespec
+
+ /* Operations on timespecs */
+ #undef        timespecclear
meta/recipes-support/db/db_5.3.28.bb +1 -1
index a2969055a6..badb79f9ea 100644
@@ -26,6 +26,7 @@ SRC_URI += "file://fix-parallel-build.patch \
             file://0001-configure-Add-explicit-tag-options-to-libtool-invoca.patch \
             file://sequence-type.patch \
             file://0001-Fix-libc-compatibility-by-renaming-atomic_init-API.patch \
+            file://0001-clock-Do-not-define-own-timespec.patch \
            "
 # We are not interested in official latest 6.x versions;
 # let's track what debian is using.
--

2.25.0

Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II
#2 lore raw
Quoted 6 lines · Bruce Ashfield
Quoted 2 lines · Khem Raj

This should fix the vdso crashes seen with kernel 5.0+

cool. I'll test this locally and follow up over the weekend.

I can confirm that 32bit x86 + musl no longer traps for me with this patch in place.

I'll bump the 5.4 recipes again, retest and then send the patches to make them the default.

Bruce

Quoted 86 lines

Bruce

Quoted 75 lines

python3[11312] general protection fault ip:b7e966b0 sp:bf8175cc error:0 in libc.so[b7e8b000+6b000]

Cc Bruce Ashfield <[email protected]>
---
 ...001-clock-Do-not-define-own-timespec.patch | 45 +++++++++++++++++++
 meta/recipes-support/db/db_5.3.28.bb          |  1 +
 2 files changed, 46 insertions(+)
 create mode 100644 meta/recipes-support/db/db/0001-clock-Do-not-define-own-timespec.patch
meta/recipes-support/db/db/0001-clock-Do-not-define-own-timespec.patch new file +45 -0
new file mode 100644
index 0000000000..9d51a44a26
@@ -0,0 +1,45 @@
+From 96b303caf70a7635953c36e5bfb9ad6e75cb7637 Mon Sep 17 00:00:00 2001
+From: Khem Raj <[email protected]>
+Date: Fri, 14 Feb 2020 14:12:59 -0800
+Subject: [PATCH] clock: Do not define own timespec
+
+timespec is provided by libc and its best left to libc
+os_gettime takes a db_timespec and passed its address to clock_gettime
+which assumes that db_timespec and timespec are same but actually
+its 12-bytes here and libc has 16-bytes
+
+This can cause problems especially with 64bit time_t
+
+Upstream-Status: Pending
+Signed-off-by: Khem Raj <[email protected]>
+---
+ src/dbinc/clock.h | 17 +----------------
+ 1 file changed, 1 insertion(+), 16 deletions(-)
+
+--- a/src/dbinc/clock.h
++++ b/src/dbinc/clock.h
+@@ -44,22 +44,8 @@
+ extern "C" {
+ #endif
+
+-/*
+- * This declaration is POSIX-compatible.  Because there are lots of different
+- * time.h include file patterns out there, it's easier to declare our own name
+- * in all cases than to try and discover if a system has a struct timespec.
+- * For the same reason, and because we'd have to #include <sys/time.h> in db.h,
+- * we don't export any timespec structures in the DB API, even in places where
+- * it would make sense, like the replication statistics information.
+- */
+-typedef struct {
+-      time_t  tv_sec;                         /* seconds */
+-#ifdef HAVE_MIXED_SIZE_ADDRESSING
+-      int32_t tv_nsec;
+-#else
+-      long    tv_nsec;                        /* nanoseconds */
+-#endif
+-} db_timespec;
++#include <time.h>
++#define db_timespec struct timespec
+
+ /* Operations on timespecs */
+ #undef        timespecclear
meta/recipes-support/db/db_5.3.28.bb +1 -1
index a2969055a6..badb79f9ea 100644
@@ -26,6 +26,7 @@ SRC_URI += "file://fix-parallel-build.patch \
             file://0001-configure-Add-explicit-tag-options-to-libtool-invoca.patch \
             file://sequence-type.patch \
             file://0001-Fix-libc-compatibility-by-renaming-atomic_init-API.patch \
+            file://0001-clock-Do-not-define-own-timespec.patch \
            "
 # We are not interested in official latest 6.x versions;
 # let's track what debian is using.
--
Quoted 2 lines

2.25.0

--
- Thou shalt not follow the NULL pointer, for chaos and madness await thee at its end
- "Use the force Harry" - Gandalf, Star Trek II

Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II