kprobes string reading broken on s390
Hi Christoph,
with the latest linux-next i noticed that some tests in the ftrace test suites are failing on s390, namely:
[FAIL] Kprobe event symbol argument
[FAIL] Kprobe event with comm arguments
The following doesn't work anymore:
cd /sys/kernel/tracing
echo 'p:testprobe _do_fork comm=$comm ' >kprobe_events echo 1 >/sys/kernel/tracing/events/kprobes/testprobe/enable cat /sys/kernel/tracing/trace
it will just show
test.sh-519 [012] .... 18.580625: testprobe: (_do_fork+0x0/0x3c8) comm=(fault)
Looking at d411a9c4e95a ("tracing/kprobes: handle mixed kernel/userspace probes better") i see that there are two helpers for reading strings:
fetch_store_string_user() -> read string from user space fetch_store_string() -> read string from kernel space(?)
but in the end both are using strncpy_from_user_nofault(), but i would think that fetch_store_string() should use strncpy_from_kernel_nofault(). However, i'm not sure about the exact semantics of fetch_store_string(), as there where a lot of wrong assumptions in the past, especially since on x86 you usually don't fail if you use the same function for accessing kernel and userspace although it's technically wrong.
Regards,
Sven
commit 81408eab8fcc79dc0871a95462b13176d3446f5e
Author: Sven Schnelle <[email protected]> Date: Fri Jun 5 13:01:24 2020 +0200
kprobes: use strncpy_from_kernel_nofault() in fetch_store_string()
kernel/trace/trace_kprobe.c +1 -1
@@ -1278,7 +1278,7 @@ fetch_store_string(unsigned long addr, void *dest, void *base)* Try to get string again, since the string can be changed while * probing. */ - ret =strncpy_from_user_nofault (__dest, (void *)addr, maxlen); + ret =strncpy_from_kernel_nofault (__dest, (void *)addr, maxlen);if (ret >=0 ) *(u32 *)dest =make_data_loc (ret, __dest - base);
Yes, this looks correct. You probably want to write a small changelog and add a Fixes tag, though.
Quoted 48 lines · Sven Schnelle
Hi Christoph,
with the latest linux-next i noticed that some tests in the ftrace test suites are failing on s390, namely:
[FAIL] Kprobe event symbol argument
[FAIL] Kprobe event with comm argumentsThe following doesn't work anymore:
cd
/sys/kernel/tracing
echo 'p:testprobe_do_forkcomm=$comm ' >kprobe_events echo 1 >/sys/kernel/tracing/events/kprobes/testprobe/enable cat/sys/kernel/tracing/traceit will just show
test.sh-519 [012] .... 18.580625: testprobe: (
_do_fork+0x0/0x3c8) comm=(fault)Looking at
d411a9c4e95a("tracing/kprobes: handle mixed kernel/userspace probes better") i see that there are two helpers for reading strings:
fetch_store_string_user()-> read string from user spacefetch_store_string()-> read string from kernel space(?)but in the end both are using
strncpy_from_user_nofault(), but i would think thatfetch_store_string()should usestrncpy_from_kernel_nofault(). However, i'm not sure about the exact semantics offetch_store_string(), as there where a lot of wrong assumptions in the past, especially since on x86 you usually don't fail if you use the same function for accessing kernel and userspace although it's technically wrong.Regards,
Svencommit
81408eab8fcc79dc0871a95462b13176d3446f5eAuthor: Sven Schnelle <[email protected]> Date: Fri Jun 5 13:01:24 2020 +0200kprobes: use
strncpy_from_kernel_nofault()infetch_store_string()Signed-off-by Sven Schnelle <[email protected]>kernel/trace/trace_kprobe.c +0
-0
kernel/trace/trace_kprobe.c fetch_store_string(unsigned long addr, void *dest, void *base) · line 1284
Quoted 8 lines
kernel/trace/trace_kprobe.c +1
-1@@ -1278,7 +1278,7 @@ fetch_store_string(unsigned long addr, void *dest, void *base)* Try to get string again, since the string can be changed while * probing. */ - ret =strncpy_from_user_nofault (__dest, (void *)addr, maxlen); + ret =strncpy_from_kernel_nofault (__dest, (void *)addr, maxlen);if (ret >=0 ) *(u32 *)dest =make_data_loc (ret, __dest - base);
---end quoted text---
Masami Hiramatsu
maintainerHi Sven,
Quoted 35 lines · On Fri, 5 Jun 2020 15:25:41 +0200 Christoph Hellwig <[email protected]> wrote:
Yes, this looks correct. You probably want to write a small changelog and add a Fixes tag, though.
Quoted 31 lines · Sven Schnelle
Hi Christoph,
with the latest linux-next i noticed that some tests in the ftrace test suites are failing on s390, namely:
[FAIL] Kprobe event symbol argument
[FAIL] Kprobe event with comm argumentsThe following doesn't work anymore:
cd
/sys/kernel/tracing
echo 'p:testprobe_do_forkcomm=$comm ' >kprobe_events echo 1 >/sys/kernel/tracing/events/kprobes/testprobe/enable cat/sys/kernel/tracing/traceit will just show
test.sh-519 [012] .... 18.580625: testprobe: (
_do_fork+0x0/0x3c8) comm=(fault)Looking at
d411a9c4e95a("tracing/kprobes: handle mixed kernel/userspace probes better") i see that there are two helpers for reading strings:
fetch_store_string_user()-> read string from user spacefetch_store_string()-> read string from kernel space(?)but in the end both are using
strncpy_from_user_nofault(), but i would think thatfetch_store_string()should usestrncpy_from_kernel_nofault(). However, i'm not sure about the exact semantics offetch_store_string(), as there where a lot of wrong assumptions in the past, especially since on x86 you usually don't fail if you use the same function for accessing kernel and userspace although it's technically wrong.
Thanks for fixing!
This report can be a good changelog. Please resend it with Fixed tag as Christoph said.
Christoph, it seems your series in the akpm tree, so this also should be sent to Andrew?
Thank you,
Quoted 26 lines
Quoted 17 lines
Regards,
Svencommit
81408eab8fcc79dc0871a95462b13176d3446f5eAuthor: Sven Schnelle <[email protected]> Date: Fri Jun 5 13:01:24 2020 +0200kprobes: use
strncpy_from_kernel_nofault()infetch_store_string()Signed-off-by Sven Schnelle <[email protected]>kernel/trace/trace_kprobe.c +0
-0kernel/trace/trace_kprobe.c fetch_store_string(unsigned long addr, void *dest, void *base) · line 1284
Quoted 8 lines
kernel/trace/trace_kprobe.c +1
-1@@ -1278,7 +1278,7 @@ fetch_store_string(unsigned long addr, void *dest, void *base)* Try to get string again, since the string can be changed while * probing. */ - ret =strncpy_from_user_nofault (__dest, (void *)addr, maxlen); + ret =strncpy_from_kernel_nofault (__dest, (void *)addr, maxlen);if (ret >=0 ) *(u32 *)dest =make_data_loc (ret, __dest - base);---end quoted text---
Signature · 1 line
Masami Hiramatsu <[email protected]>
Masami,
Quoted 41 lines · Masami Hiramatsu
Hi Sven,
Quoted 35 lines · On Fri, 5 Jun 2020 15:25:41 +0200 Christoph Hellwig <[email protected]> wrote:
Yes, this looks correct. You probably want to write a small changelog and add a Fixes tag, though.
Quoted 31 lines · Sven Schnelle
Hi Christoph,
with the latest linux-next i noticed that some tests in the ftrace test suites are failing on s390, namely:
[FAIL] Kprobe event symbol argument
[FAIL] Kprobe event with comm argumentsThe following doesn't work anymore:
cd
/sys/kernel/tracing
echo 'p:testprobe_do_forkcomm=$comm ' >kprobe_events echo 1 >/sys/kernel/tracing/events/kprobes/testprobe/enable cat/sys/kernel/tracing/traceit will just show
test.sh-519 [012] .... 18.580625: testprobe: (
_do_fork+0x0/0x3c8) comm=(fault)Looking at
d411a9c4e95a("tracing/kprobes: handle mixed kernel/userspace probes better") i see that there are two helpers for reading strings:
fetch_store_string_user()-> read string from user spacefetch_store_string()-> read string from kernel space(?)but in the end both are using
strncpy_from_user_nofault(), but i would think thatfetch_store_string()should usestrncpy_from_kernel_nofault(). However, i'm not sure about the exact semantics offetch_store_string(), as there where a lot of wrong assumptions in the past, especially since on x86 you usually don't fail if you use the same function for accessing kernel and userspace although it's technically wrong.
Quoted 3 lines
Thanks for fixing!
This report can be a good changelog. Please resend it with Fixed tag as Christoph said.
Which SHA1 should the Fixes tag carry? The one from linux-next?
Regards
Sven
Christoph Hellwig
Acked-byQuoted 5 lines · Sven Schnelle
Quoted 3 lines
Thanks for fixing!
This report can be a good changelog. Please resend it with Fixed tag as Christoph said.Which SHA1 should the Fixes tag carry? The one from linux-next?
Oh, I thought it had hit mainline already. If this is just in -mm there is no need for a fixes tag, just send it to Andrew with the acks from Masami and me.