Segmentation Fault in unescape_bytea_9x (?)

[smondet]

I've been getting seemingly “random” segmentation faults, I tried to run my server in GDB and it gave this:

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffe6d27700 (LWP 708)]
unescape_bytea_9x (v_from=<value optimized out>) at lib/postgresql_stubs.c:966
966     lib/postgresql_stubs.c: No such file or directory.
        in lib/postgresql_stubs.c

The segmentation faults only happen with the PG backend of https://github.com/smondet/trakeva, exactly the same server works fine the Sqlite backend.

The PG server is on version 9.3.

[smondet]

I forgot to paste the backtrace:

(gdb) bt full
#0  unescape_bytea_9x (v_from=<value optimized out>) at lib/postgresql_stubs.c:966
        v_res = 51368496
        res = <value optimized out>
        n_hex_pairs = <value optimized out>
        end = 0x7fffb7ffeff8 <Address 0x7fffb7ffeff8 out of bounds>
#1  PQunescapeBytea9x_stub (v_from=<value optimized out>) at lib/postgresql_stubs.c:979
        s = <value optimized out>
#2  0x00000000005ebcc6 in camlPostgresql__unescape_bytea_1238 () at lib/postgresql.ml:531
No locals.
#3  0x00000000005e3b3c in camlTrakeva_postgresql__fun_1654 () at src/lib_postgresql/trakeva_postgresql.ml:130
No locals.
#4  0x00000000006d4513 in camlNonstd__loop_1313 ()
No symbol table info available.
#5  0x00000000006d4513 in camlNonstd__loop_1313 ()
No symbol table info available.
#6  0x00000000005e46f2 in camlTrakeva_postgresql__exec_one_1297 () at src/lib_postgresql/trakeva_postgresql.ml:127
No locals.
#7  0x00000000006d52d8 in camlNonstd__count_map_1166 ()
No symbol table info available.
#8  0x00000000005e5518 in camlTrakeva_postgresql__exec_sql_exn_1286 () at src/lib_postgresql/trakeva_postgresql.ml:96
No locals.
#9  0x00000000005e57b3 in camlTrakeva_postgresql__get_exn_1338 () at src/lib_postgresql/trakeva_postgresql.ml:167
No locals.
#10 0x00000000005e4155 in camlTrakeva_postgresql__fun_1604 () at src/lib_postgresql/trakeva_postgresql.ml:51
No locals.
#11 0x00000000005f86e1 in camlLwt_preemptive__task_1061 () at src/preemptive/lwt_preemptive.ml:161
No locals.
#12 0x00000000005f89aa in camlLwt_preemptive__worker_loop_1030 () at src/preemptive/lwt_preemptive.ml:80
No locals.
#13 0x00000000006e3f79 in camlThread__fun_1084 () at thread.ml:37
No locals.
#14 0x000000000075d532 in caml_start_program ()
No symbol table info available.
#15 0x0000000000ecc580 in ?? ()
No symbol table info available.
#16 0x0000000000000000 in ?? ()
No symbol table info available.

[mmottl]

Thanks, I'm looking at it right now and have already found the (subtle!) bug. Basically, there are two ways to enter the unescaping routine: given a string that comes from the PostgreSQL C-library, and given a string that comes from OCaml. The former string will never move in memory, the latter can if you allocate in the OCaml heap. I'll upload a patch soon.

[mmottl]

@smondet could you please test the latest development snapshot? It's completely untested, but was carefully read and compiles without errors 😉. I guess you will have an easier time than me determining whether the problem is fixed now.

[smondet]

@mmottl I've been using the master branch today, haven't noticed any seg-fault so far !!
Thanks!

[mmottl]

@smondet Sounds good, thanks for testing! I'll hold off until the end of the week. If you don't report any segfaults by then, I'll make a new release.

[mmottl]

The new release containing the fix has been uploaded to OPAM and should be available there shortly.