* Re: [bug report] printk: ringbuffer: Add KUnit test
2025-06-26 6:59 ` Thomas Weißschuh
@ 2025-07-01 15:23 ` Dan Carpenter
2025-07-01 15:56 ` Petr Mladek
2025-07-02 20:22 ` Nathan Chancellor
2 siblings, 0 replies; 6+ messages in thread
From: Dan Carpenter @ 2025-07-01 15:23 UTC (permalink / raw)
To: Thomas Weißschuh, Nathan Chancellor
Cc: John Ogness, Kees Cook, linux-hardening, Petr Mladek
Let me add Nathan to the CC list. He knows a lot about __counted_by().
The dat->text is __counted_by() dat->len so the line:
if (dat->text[dat->len] != '\0')
return false;
Should have generated a warning of some sort, right? I'm including
the whole thread even though I assume you are huge fan of b4.
regards,
dan carpenter
On Thu, Jun 26, 2025 at 08:59:52AM +0200, Thomas Weißschuh wrote:
> On Wed, Jun 25, 2025 at 10:22:19AM -0500, Dan Carpenter wrote:
> > Hello Thomas Weißschuh,
> >
> > The patch 5ea2bcdfbf46: "printk: ringbuffer: Add KUnit test" from Jun
> > 12, 2025, leads to the following static checker warning:
> >
> > kernel/printk/printk_ringbuffer_kunit_test.c:91 prbtest_check_data()
> > (unpublished script worries this an off by one)
> >
> > kernel/printk/printk_ringbuffer_kunit_test.c
> > 83 static bool prbtest_check_data(const struct prbtest_rbdata *dat)
> > 84 {
> > 85 unsigned int len;
> > 86
> > 87 /* Sane length? */
> > 88 if (dat->len < 1 || dat->len > MAX_RBDATA_TEXT_SIZE)
> > 89 return false;
> > 90
> > --> 91 if (dat->text[dat->len] != '\0')
> > 92 return false;
> > 93
> >
> > My question is that the prbtest_rbdata structure is declared like this:
> >
> > 53 /* test data structure */
> > 54 struct prbtest_rbdata {
> > 55 unsigned int len;
> > 56 char text[] __counted_by(len);
> > 57 };
> >
> > The size of text is not really counted by len, it's "MAX_RBDATA_TEXT_SIZE
> > + 1". The condition "if (dat->text[dat->len] != '\0')" is reading one
> > element beyond the __counted_by() value so something should complain if
> > we enable all the debugging, right?
>
> You are right, we are reading past the __counted_by().
> But I don't get any complains with CONFIG_FORTIFY_SOURCE=y and CONFIG_KASAN=y
> on either clang or gcc.
> We could remove the __counted_by, but I assume somebody will try to add it back
> at some point.
> Or we account for the terminator in dat->len:
>
> diff --git a/kernel/printk/printk_ringbuffer_kunit_test.c b/kernel/printk/printk_ringbuffer_kunit_test.c
> index ef4a2beea57a..106f4c7ffc86 100644
> --- a/kernel/printk/printk_ringbuffer_kunit_test.c
> +++ b/kernel/printk/printk_ringbuffer_kunit_test.c
> @@ -85,14 +85,15 @@ static bool prbtest_check_data(const struct prbtest_rbdata *dat)
> unsigned int len;
>
> /* Sane length? */
> - if (dat->len < 1 || dat->len > MAX_RBDATA_TEXT_SIZE)
> + if (dat->len < 2 || dat->len > MAX_RBDATA_TEXT_SIZE + 1)
> return false;
>
> - if (dat->text[dat->len] != '\0')
> + len = dat->len - 1;
> +
> + if (dat->text[len] != '\0')
> return false;
>
> /* String repeats with the same character? */
> - len = dat->len;
> while (len--) {
> if (dat->text[len] != dat->text[0])
> return false;
> @@ -114,10 +115,9 @@ static int prbtest_writer(void *data)
> kunit_info(tr->test_data->test, "start thread %03lu (writer)\n", tr->num);
>
> for (;;) {
> - /* ensure at least 1 character */
> - text_size = get_random_u32_inclusive(1, MAX_RBDATA_TEXT_SIZE);
> - /* +1 for terminator. */
> - record_size = sizeof(struct prbtest_rbdata) + text_size + 1;
> + /* ensure at least 1 character, +1 for terminator */
> + text_size = get_random_u32_inclusive(1, MAX_RBDATA_TEXT_SIZE) + 1;
> + record_size = sizeof(struct prbtest_rbdata) + text_size;
> WARN_ON_ONCE(record_size > MAX_PRB_RECORD_SIZE);
>
> /* specify the text sizes for reservation */
> @@ -142,7 +142,7 @@ static int prbtest_writer(void *data)
> dat = (struct prbtest_rbdata *)r.text_buf;
> dat->len = text_size;
> memset(dat->text, text_id, text_size);
> - dat->text[text_size] = 0;
> + dat->text[text_size - 1] = '\0';
>
> prb_commit(&e);
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [bug report] printk: ringbuffer: Add KUnit test
2025-06-26 6:59 ` Thomas Weißschuh
2025-07-01 15:23 ` Dan Carpenter
@ 2025-07-01 15:56 ` Petr Mladek
2025-07-02 20:22 ` Nathan Chancellor
2 siblings, 0 replies; 6+ messages in thread
From: Petr Mladek @ 2025-07-01 15:56 UTC (permalink / raw)
To: Thomas Weißschuh
Cc: Dan Carpenter, John Ogness, Kees Cook, linux-hardening
On Thu 2025-06-26 08:59:52, Thomas Weißschuh wrote:
> On Wed, Jun 25, 2025 at 10:22:19AM -0500, Dan Carpenter wrote:
> > Hello Thomas Weißschuh,
> >
> > The patch 5ea2bcdfbf46: "printk: ringbuffer: Add KUnit test" from Jun
> > 12, 2025, leads to the following static checker warning:
> >
> > kernel/printk/printk_ringbuffer_kunit_test.c:91 prbtest_check_data()
> > (unpublished script worries this an off by one)
> >
> > kernel/printk/printk_ringbuffer_kunit_test.c
> > 83 static bool prbtest_check_data(const struct prbtest_rbdata *dat)
> > 84 {
> > 85 unsigned int len;
> > 86
> > 87 /* Sane length? */
> > 88 if (dat->len < 1 || dat->len > MAX_RBDATA_TEXT_SIZE)
> > 89 return false;
> > 90
> > --> 91 if (dat->text[dat->len] != '\0')
> > 92 return false;
> > 93
> >
> > My question is that the prbtest_rbdata structure is declared like this:
> >
> > 53 /* test data structure */
> > 54 struct prbtest_rbdata {
> > 55 unsigned int len;
> > 56 char text[] __counted_by(len);
> > 57 };
> >
> > The size of text is not really counted by len, it's "MAX_RBDATA_TEXT_SIZE
> > + 1". The condition "if (dat->text[dat->len] != '\0')" is reading one
> > element beyond the __counted_by() value so something should complain if
> > we enable all the debugging, right?
>
> You are right, we are reading past the __counted_by().
> But I don't get any complains with CONFIG_FORTIFY_SOURCE=y and CONFIG_KASAN=y
> on either clang or gcc.
> We could remove the __counted_by, but I assume somebody will try to add it back
> at some point.
> Or we account for the terminator in dat->len:
It means that the value will be the size occupied by the string
including the trailing '\0'.
It means that we need to rename it, for example, len -> size.
Because using "len" for size is confusing and error prone.
See below.
> diff --git a/kernel/printk/printk_ringbuffer_kunit_test.c b/kernel/printk/printk_ringbuffer_kunit_test.c
> index ef4a2beea57a..106f4c7ffc86 100644
> --- a/kernel/printk/printk_ringbuffer_kunit_test.c
> +++ b/kernel/printk/printk_ringbuffer_kunit_test.c
> @@ -85,14 +85,15 @@ static bool prbtest_check_data(const struct prbtest_rbdata *dat)
> unsigned int len;
>
> /* Sane length? */
> - if (dat->len < 1 || dat->len > MAX_RBDATA_TEXT_SIZE)
> + if (dat->len < 2 || dat->len > MAX_RBDATA_TEXT_SIZE + 1)
> return false;
>
> - if (dat->text[dat->len] != '\0')
> + len = dat->len - 1;
This is one example, where it just looks just ugly.
> +
> + if (dat->text[len] != '\0')
> return false;
>
> /* String repeats with the same character? */
> - len = dat->len;
> while (len--) {
> if (dat->text[len] != dat->text[0])
> return false;
> @@ -114,10 +115,9 @@ static int prbtest_writer(void *data)
> kunit_info(tr->test_data->test, "start thread %03lu (writer)\n", tr->num);
>
> for (;;) {
> - /* ensure at least 1 character */
> - text_size = get_random_u32_inclusive(1, MAX_RBDATA_TEXT_SIZE);
> - /* +1 for terminator. */
> - record_size = sizeof(struct prbtest_rbdata) + text_size + 1;
> + /* ensure at least 1 character, +1 for terminator */
> + text_size = get_random_u32_inclusive(1, MAX_RBDATA_TEXT_SIZE) + 1;
This is where the naming goes beyond sanity. We allow to break the
limit by setting "text_size" to MAX_RBDATA_TEXT_SIZE + 1.
It is because MAX_RBDATA_TEXT_SIZE is used to limit the length of
the string (without the trailing '\0'). Huh.
> + record_size = sizeof(struct prbtest_rbdata) + text_size;
> WARN_ON_ONCE(record_size > MAX_PRB_RECORD_SIZE);
>
> /* specify the text sizes for reservation */
> @@ -142,7 +142,7 @@ static int prbtest_writer(void *data)
> dat = (struct prbtest_rbdata *)r.text_buf;
> dat->len = text_size;
> memset(dat->text, text_id, text_size);
> - dat->text[text_size] = 0;
> + dat->text[text_size - 1] = '\0';
>
> prb_commit(&e);
This patch forgot to update prbtest_fail_record(). It limits the
printed string by dat->len. But the value newly counts the trailing
'\0'.
OK, I am going to send a patch with sane names where all this
should be fixed.
Best Regards,
Petr
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [bug report] printk: ringbuffer: Add KUnit test
2025-06-26 6:59 ` Thomas Weißschuh
2025-07-01 15:23 ` Dan Carpenter
2025-07-01 15:56 ` Petr Mladek
@ 2025-07-02 20:22 ` Nathan Chancellor
2025-07-02 20:33 ` Dan Carpenter
2 siblings, 1 reply; 6+ messages in thread
From: Nathan Chancellor @ 2025-07-02 20:22 UTC (permalink / raw)
To: Thomas Weißschuh
Cc: Dan Carpenter, John Ogness, Kees Cook, linux-hardening,
Petr Mladek
On Thu, Jun 26, 2025 at 08:59:52AM +0200, Thomas Weißschuh wrote:
> On Wed, Jun 25, 2025 at 10:22:19AM -0500, Dan Carpenter wrote:
> > Hello Thomas Weißschuh,
> >
> > The patch 5ea2bcdfbf46: "printk: ringbuffer: Add KUnit test" from Jun
> > 12, 2025, leads to the following static checker warning:
> >
> > kernel/printk/printk_ringbuffer_kunit_test.c:91 prbtest_check_data()
> > (unpublished script worries this an off by one)
> >
> > kernel/printk/printk_ringbuffer_kunit_test.c
> > 83 static bool prbtest_check_data(const struct prbtest_rbdata *dat)
> > 84 {
> > 85 unsigned int len;
> > 86
> > 87 /* Sane length? */
> > 88 if (dat->len < 1 || dat->len > MAX_RBDATA_TEXT_SIZE)
> > 89 return false;
> > 90
> > --> 91 if (dat->text[dat->len] != '\0')
> > 92 return false;
> > 93
> >
> > My question is that the prbtest_rbdata structure is declared like this:
> >
> > 53 /* test data structure */
> > 54 struct prbtest_rbdata {
> > 55 unsigned int len;
> > 56 char text[] __counted_by(len);
> > 57 };
> >
> > The size of text is not really counted by len, it's "MAX_RBDATA_TEXT_SIZE
> > + 1". The condition "if (dat->text[dat->len] != '\0')" is reading one
> > element beyond the __counted_by() value so something should complain if
> > we enable all the debugging, right?
>
> You are right, we are reading past the __counted_by().
> But I don't get any complains with CONFIG_FORTIFY_SOURCE=y and CONFIG_KASAN=y
> on either clang or gcc.
You will need CONFIG_UBSAN_BOUNDS to see warnings from __counted_by()
but they are present if I enable it with this test:
[ 0.507904] ------------[ cut here ]------------
[ 0.507904] UBSAN: array-index-out-of-bounds in kernel/printk/printk_ringbuffer_kunit_test.c:132:4
[ 0.507907] index 108 is out of range for type 'char[] __counted_by(len)' (aka 'char[]')
[ 0.507909] CPU: 3 UID: 0 PID: 392 Comm: prbtest writer Tainted: G N 6.16.0-rc4-next-20250702 #1 PREEMPT(voluntary)
[ 0.507912] Tainted: [N]=TEST
[ 0.507913] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.17.0-0-gb52ca86e094d-prebuilt.qemu.org 04/01/2014
[ 0.507914] Call Trace:
[ 0.507916] <TASK>
[ 0.507918] dump_stack_lvl+0x84/0xc0
[ 0.507922] ubsan_epilogue+0x5/0x30
[ 0.507925] __ubsan_handle_out_of_bounds+0xa2/0xc0
[ 0.507928] prbtest_writer+0x16a/0x210
[ 0.507932] ? __pfx_prbtest_writer+0x10/0x10
[ 0.507934] kthread+0x21a/0x260
[ 0.507937] ? __pfx_kthread+0x10/0x10
[ 0.507938] ret_from_fork+0x77/0xd0
[ 0.507940] ? __pfx_kthread+0x10/0x10
[ 0.507941] ret_from_fork_asm+0x1a/0x30
[ 0.507945] </TASK>
[ 0.507945] ---[ end trace ]---
[ 0.516113] ------------[ cut here ]------------
[ 0.516115] UBSAN: array-index-out-of-bounds in kernel/printk/printk_ringbuffer_kunit_test.c:91:6
[ 0.516117] # test_readerwriter: lib/ubsan.c:228: array-index-out-of-bounds in kernel/printk/printk_ringbuffer_kunit_test.c
[ 0.516119] index 126 is out of range for type 'char const[] __counted_by(len)' (aka 'const char[]')
[ 0.516122] CPU: 0 UID: 0 PID: 389 Comm: kunit_try_catch Tainted: G N 6.16.0-rc4-next-20250702 #1 PREEMPT(voluntary)
[ 0.516125] Tainted: [N]=TEST
[ 0.516126] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.17.0-0-gb52ca86e094d-prebuilt.qemu.org 04/01/2014
[ 0.516127] Call Trace:
[ 0.516129] <TASK>
[ 0.516131] dump_stack_lvl+0x84/0xc0
[ 0.516136] ubsan_epilogue+0x5/0x30
[ 0.516138] __ubsan_handle_out_of_bounds+0xa2/0xc0
[ 0.516142] prbtest_reader+0x2ff/0x480
[ 0.516145] ? srso_alias_return_thunk+0x5/0xfbef5
[ 0.516147] ? set_next_entity+0x5e/0x140
[ 0.516150] ? __pfx_prbtest_wakeup_callback+0x10/0x10
[ 0.516155] ? set_cpus_allowed_ptr+0x83/0xb0
[ 0.516158] test_readerwriter+0x297/0x3c0
[ 0.516161] kunit_try_run_case+0x8c/0x190
[ 0.516164] kunit_generic_run_threadfn_adapter+0x1a/0x40
[ 0.516166] ? __pfx_kunit_generic_run_threadfn_adapter+0x10/0x10
[ 0.516167] kthread+0x21a/0x260
[ 0.516169] ? __pfx_kthread+0x10/0x10
[ 0.516170] ret_from_fork+0x77/0xd0
[ 0.516173] ? __pfx_kthread+0x10/0x10
[ 0.516174] ret_from_fork_asm+0x1a/0x30
[ 0.516177] </TASK>
[ 0.516178] ---[ end trace ]---
I see Petr sent an updated patch for this, which does resolve the
warning in my testing.
https://lore.kernel.org/20250702095157.110916-4-pmladek@suse.com/
Cheers,
Nathan
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [bug report] printk: ringbuffer: Add KUnit test
2025-07-02 20:22 ` Nathan Chancellor
@ 2025-07-02 20:33 ` Dan Carpenter
0 siblings, 0 replies; 6+ messages in thread
From: Dan Carpenter @ 2025-07-02 20:33 UTC (permalink / raw)
To: Nathan Chancellor
Cc: Thomas Weißschuh, John Ogness, Kees Cook, linux-hardening,
Petr Mladek
On Wed, Jul 02, 2025 at 01:22:03PM -0700, Nathan Chancellor wrote:
> On Thu, Jun 26, 2025 at 08:59:52AM +0200, Thomas Weißschuh wrote:
> > On Wed, Jun 25, 2025 at 10:22:19AM -0500, Dan Carpenter wrote:
> > > Hello Thomas Weißschuh,
> > >
> > > The patch 5ea2bcdfbf46: "printk: ringbuffer: Add KUnit test" from Jun
> > > 12, 2025, leads to the following static checker warning:
> > >
> > > kernel/printk/printk_ringbuffer_kunit_test.c:91 prbtest_check_data()
> > > (unpublished script worries this an off by one)
> > >
> > > kernel/printk/printk_ringbuffer_kunit_test.c
> > > 83 static bool prbtest_check_data(const struct prbtest_rbdata *dat)
> > > 84 {
> > > 85 unsigned int len;
> > > 86
> > > 87 /* Sane length? */
> > > 88 if (dat->len < 1 || dat->len > MAX_RBDATA_TEXT_SIZE)
> > > 89 return false;
> > > 90
> > > --> 91 if (dat->text[dat->len] != '\0')
> > > 92 return false;
> > > 93
> > >
> > > My question is that the prbtest_rbdata structure is declared like this:
> > >
> > > 53 /* test data structure */
> > > 54 struct prbtest_rbdata {
> > > 55 unsigned int len;
> > > 56 char text[] __counted_by(len);
> > > 57 };
> > >
> > > The size of text is not really counted by len, it's "MAX_RBDATA_TEXT_SIZE
> > > + 1". The condition "if (dat->text[dat->len] != '\0')" is reading one
> > > element beyond the __counted_by() value so something should complain if
> > > we enable all the debugging, right?
> >
> > You are right, we are reading past the __counted_by().
> > But I don't get any complains with CONFIG_FORTIFY_SOURCE=y and CONFIG_KASAN=y
> > on either clang or gcc.
>
> You will need CONFIG_UBSAN_BOUNDS to see warnings from __counted_by()
> but they are present if I enable it with this test:
>
> [ 0.507904] ------------[ cut here ]------------
> [ 0.507904] UBSAN: array-index-out-of-bounds in kernel/printk/printk_ringbuffer_kunit_test.c:132:4
Thanks, Nathan. You are the best.
regards,
dan carpenter
^ permalink raw reply [flat|nested] 6+ messages in thread