Looking further into this segfault at settings-parser.c:setting_copy():1519 *dest_size = *src_size; *src_size points to type size_t (typedef unsigned long), a 4-byte aligned value consistent with a 32-bit build. This is mismatched with declared type (gdb) whatis src_size type = const uoff_t * (gdb) whatis uoff_t type = unsigned long long (gdb) p sizeof(uoff_t) $1 = 8 resulting in the segfault when *src_size is dereferened. The implied condition of this code segment is typeof(uoff_t)==typeof(size_t) which is clearly not the case. I'm not sure how/if uoff_t is defined, but configure reports checking for uoff_t... no checking type of off_t... long long The latter is weird, because if I compile and run using the same compiler flags #include <stdio.h> int main(void) { printf("%d %d\n",sizeof(long long),sizeof(off_t)); } the output is "8 4". Joseph Tam <jtam.home at gmail.com>
Ah, okay, I see. submission_max_mail_size should be defined as uoff_t instead of size_t in struct submission_settings and struct submission_settings.> On 20 Jul 2019, at 1.47, Joseph Tam via dovecot <dovecot at dovecot.org> wrote: > > > Looking further into this segfault at > > settings-parser.c:setting_copy():1519 > *dest_size = *src_size; > > *src_size points to type size_t (typedef unsigned long), a 4-byte > aligned value consistent with a 32-bit build. This is mismatched with > declared type > > (gdb) whatis src_size > type = const uoff_t * > (gdb) whatis uoff_t > type = unsigned long long > (gdb) p sizeof(uoff_t) > $1 = 8 > > resulting in the segfault when *src_size is dereferened. The implied > condition of this code segment is typeof(uoff_t)==typeof(size_t) which > is clearly not the case. > > I'm not sure how/if uoff_t is defined, but configure reports > > checking for uoff_t... no > checking type of off_t... long long > > The latter is weird, because if I compile and run using the same compiler flags > > #include <stdio.h> > int main(void) { printf("%d %d\n",sizeof(long long),sizeof(off_t)); } > > the output is "8 4". > > Joseph Tam <jtam.home at gmail.com>
On Mon, 22 Jul 2019, Timo Sirainen wrote:> Ah, okay, I see. submission_max_mail_size should be defined as uoff_t > instead of size_t in struct submission_settings and struct > submission_settings.Thanks! This appears to be the correct diagnosis as this patch (for 2.3.7) got rid of the segfaults ===============================================================================--- a/src/submission-login/submission-login-settings.h Mon Jul 22 14:37:26 2019 +++ b/src/submission-login/submission-login-settings.h Mon Jul 22 14:38:16 2019 @@ -7,3 +7,3 @@ /* submission: */ - size_t submission_max_mail_size; + uoff_t submission_max_mail_size; const char *submission_backend_capabilities; --- a/src/submission/submission-settings.h Mon Jul 22 14:37:17 2019 +++ b/src/submission/submission-settings.h Mon Jul 22 14:38:06 2019 @@ -22,3 +22,3 @@ /* submission: */ - size_t submission_max_mail_size; + uoff_t submission_max_mail_size; unsigned int submission_max_recipients; =============================================================================== Joseph Tam <jtam.home at gmail.com>
Seemingly Similar Threads
- Dovecot 2.3.6 on Solaris10: build issues, segfaults
- Dovecot 2.3.6 on Solaris10: build issues, segfaults
- Dovecot 2.3.6 on Solaris10: build issues, segfaults
- Dovecot 2.3.6 on Solaris10: build issues, segfaults
- [PATCH v9 71/84] KVM: introspection: add KVMI_VCPU_SET_XSAVE