fix access to 'sock_extended_err' when generating cmsg from expressions

[dcaratti]

when packetdrill builds a cmsg parsed from the script, it handles 'ee_origin', 'ee_type' and 'ee_code' as 32-bit numbers. These members of 'struct sock_extended_err' are 8-bit wide: because of this, TCP tests belonging to 'timestamping' and 'zerocopy' categories systematically fail on big endian architectures (such as s390x). For example, the value of 'ee_origin' is written in 'ee_pad':

(gdb) set args -D TFO_COOKIE=de4f234f0f433a55 -D CMSG_LEVEL_IP=SOL_IP -D CMSG_TYPE_RECVERR=IP_RECVERR basic.pkt
(gdb) r

[...]

24: system call: recvmsg
24: invoke call: recvmsg
waiting until 1668090937472983 -- now is 1668090937472985
expected: 0.000 actual: 0.000 (secs)

Thread 1 "packetdrill" hit Breakpoint 2, new_extended_err (expr=0x111e7a0, ee=0x1124d80, error=0x3ffffff9730)
at run_system_call.c:500
500 if (get_s32(expr->ee_errno, (s32 *)&ee->ee_errno, error))
(gdb) n
502 if (get_s32(expr->ee_origin, (s32 *)&ee->ee_origin, error))
(gdb) n
504 if (get_s32(expr->ee_type, (s32 *)&ee->ee_type, error))
(gdb) p *ee
$46 = {ee_errno = 0, ee_origin = 0 '\000', ee_type = 0 '\000', ee_code = 0 '\000', ee_pad = 5 '\005', ee_info = 0,
ee_data = 0}
(gdb) up
#1 0x000000000101d892 in cmsg_new (expr=0x111f740, msg=0x1124c60, error=0x3ffffff9730) at run_system_call.c:617
617 if (new_extended_err(ee_expr,
(gdb) p *(struct sock_extended_err *)data
$47 = {ee_errno = 0, ee_origin = 0 '\000', ee_type = 0 '\000', ee_code = 0 '\000', ee_pad = 5 '\005', ee_info = 0,
ee_data = 0}
(gdb)

fix it by providing a correct integer size for 'ee_origin', 'ee_type' and 'ee_code'.

Reported-by: Xiumei Mu xmu@redhat.com
Signed-off-by: Davide Caratti dcaratti@redhat.com

[nealcardwell]

Thanks for the fix!

For consistency/clarity, can you please make the summary/first-line of the commit message look like:

net-test: packetdrill: fix access to 'sock_extended_err' when generating cmsg from expressions

Thanks!

[nealcardwell]

The packetdrill tree uses Linux's scripts/checkpatch.pl for style validation, and it caught some issues (see below). I will fix those issues and the commit title issue, and merge if all our internal tests pass. Thanks!

ERROR: code indent should use tabs where possible
#77: FILE: gtests/net/packetdrill/run_system_call.c:206:

•             s8 *value, char **error)$
  

WARNING: please, no spaces at the start of a line
#77: FILE: gtests/net/packetdrill/run_system_call.c:206:

•             s8 *value, char **error)$
  

WARNING: please, no spaces at the start of a line
#79: FILE: gtests/net/packetdrill/run_system_call.c:208:

•   if (check_type(expression, EXPR_INTEGER, error))$
  

WARNING: suspect code indent for conditional statements (7, 15)
#79: FILE: gtests/net/packetdrill/run_system_call.c:208:

•   if (check_type(expression, EXPR_INTEGER, error))
  

•           return STATUS_ERR;
  

ERROR: code indent should use tabs where possible
#80: FILE: gtests/net/packetdrill/run_system_call.c:209:

•           return STATUS_ERR;$
  

WARNING: please, no spaces at the start of a line
#80: FILE: gtests/net/packetdrill/run_system_call.c:209:

•           return STATUS_ERR;$
  

WARNING: please, no spaces at the start of a line
#81: FILE: gtests/net/packetdrill/run_system_call.c:210:

•   if ((expression->value.num > UCHAR_MAX) ||$
  

WARNING: suspect code indent for conditional statements (7, 15)
#81: FILE: gtests/net/packetdrill/run_system_call.c:210:

•   if ((expression->value.num > UCHAR_MAX) ||
  

[...]

•           asprintf(error,
  

ERROR: code indent should use tabs where possible
#82: FILE: gtests/net/packetdrill/run_system_call.c:211:

•       (expression->value.num < CHAR_MIN)) {$
  

WARNING: please, no spaces at the start of a line
#82: FILE: gtests/net/packetdrill/run_system_call.c:211:

•       (expression->value.num < CHAR_MIN)) {$
  

ERROR: code indent should use tabs where possible
#83: FILE: gtests/net/packetdrill/run_system_call.c:212:

•           asprintf(error,$
  

WARNING: please, no spaces at the start of a line
#83: FILE: gtests/net/packetdrill/run_system_call.c:212:

•           asprintf(error,$
  

ERROR: code indent should use tabs where possible
#84: FILE: gtests/net/packetdrill/run_system_call.c:213:

•                    "Value out of range for 8-bit integer: %lld",$
  

WARNING: please, no spaces at the start of a line
#84: FILE: gtests/net/packetdrill/run_system_call.c:213:

•                    "Value out of range for 8-bit integer: %lld",$
  

ERROR: code indent should use tabs where possible
#85: FILE: gtests/net/packetdrill/run_system_call.c:214:

•                    expression->value.num);$
  

WARNING: please, no spaces at the start of a line
#85: FILE: gtests/net/packetdrill/run_system_call.c:214:

•                    expression->value.num);$
  

ERROR: code indent should use tabs where possible
#86: FILE: gtests/net/packetdrill/run_system_call.c:215:

•           return STATUS_ERR;$
  

WARNING: please, no spaces at the start of a line
#86: FILE: gtests/net/packetdrill/run_system_call.c:215:

•           return STATUS_ERR;$
  

WARNING: please, no spaces at the start of a line
#87: FILE: gtests/net/packetdrill/run_system_call.c:216:

•   }$
  

WARNING: please, no spaces at the start of a line
#88: FILE: gtests/net/packetdrill/run_system_call.c:217:

•   *value = expression->value.num;$
  

WARNING: please, no spaces at the start of a line
#89: FILE: gtests/net/packetdrill/run_system_call.c:218:

•   return STATUS_OK;$
  

total: 7 errors, 15 warnings, 40 lines checked

[dcaratti]

Thanks for the fix!

For consistency/clarity, can you please make the summary/first-line of the commit message look like:

net-test: packetdrill: fix access to 'sock_extended_err' when generating cmsg from expressions

Thanks!

hi Neal, I just fixed the commit message. Thanks for reviewing!

[dcaratti]

The packetdrill tree uses Linux's scripts/checkpatch.pl for style validation, and it caught some issues (see below).

Ouch, sorry for the copy/paste noise.

I will fix those issues and the commit title issue, and merge if all our internal tests pass. Thanks!

Ok; for your convenience, I refreshed the PR with fixed indentation. Thanks!

[nealcardwell]

Looks great. Thanks!