* [PATCH v2 0/2] tpm: Track the response read position internally
@ 2026-11-20 12:16 Rustam Mustafin
2026-11-20 12:16 ` [PATCH v2 1/2] " Rustam Mustafin
2026-11-20 12:16 ` [PATCH v2 2/2] selftests: tpm2: test that io_uring reads ignore the file offset Rustam Mustafin
0 siblings, 2 replies; 4+ messages in thread
From: Rustam Mustafin @ 2026-11-20 12:16 UTC (permalink / raw)
To: Peter Huewe, Jarkko Sakkinen
Cc: Jason Gunthorpe, Shuah Khan, Jaewon Yang, linux-integrity,
linux-kselftest, linux-kernel, Rustam Mustafin
tpm_common_read() uses *off as an index into priv->data_buffer.
nonseekable_open() makes pread() fail, but an IORING_OP_READ request
still reaches tpm_common_read() through loop_rw_iter() with *off set
to sqe->off, and a plain read() racing a write() reaches it with a
stale f_pos. Both allow out-of-bounds reads and zeroing of kernel
memory.
Patch 1 moves the read position into struct file_priv, under
buffer_mutex, and ignores *off.
Patch 2 adds a kselftest: an io_uring read with sqe->off inside the
response buffer must return the whole response from its start.
Blocking io_uring, suggested in the v1 review, is no longer needed for
safety with patch 1; I discuss it in the v1 thread.
Tested in a QEMU guest with swtpm and KASAN: the tpm2 selftests pass
with patch 1, and the new one fails without it.
Claude Opus 5.5 was used in preparing this series.
Changes in v2:
- commit message reworked per review, read()/write() race trace added
- add a kselftest
v1: https://lore.kernel.org/all/20261117104838.2650999-1-rmustafin@cloudlinux.com/
Rustam Mustafin (2):
tpm: Track the response read position internally
selftests: tpm2: test that io_uring reads ignore the file offset
drivers/char/tpm/tpm-dev-common.c | 11 +-
drivers/char/tpm/tpm-dev.h | 1 +
tools/testing/selftests/tpm2/.gitignore | 1 +
tools/testing/selftests/tpm2/Makefile | 4 +
tools/testing/selftests/tpm2/tpm_uring.c | 131 +++++++++++++++++++++++
5 files changed, 143 insertions(+), 5 deletions(-)
create mode 100644 tools/testing/selftests/tpm2/tpm_uring.c
base-commit: 666539eecdf4db623f5e3c3e55cf9f2dfad24aa4
--
2.52.0
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v2 1/2] tpm: Track the response read position internally
2026-11-20 12:16 [PATCH v2 0/2] tpm: Track the response read position internally Rustam Mustafin
@ 2026-11-20 12:16 ` Rustam Mustafin
2026-11-20 12:27 ` Jarkko Sakkinen
2026-11-20 12:16 ` [PATCH v2 2/2] selftests: tpm2: test that io_uring reads ignore the file offset Rustam Mustafin
1 sibling, 1 reply; 4+ messages in thread
From: Rustam Mustafin @ 2026-11-20 12:16 UTC (permalink / raw)
To: Peter Huewe, Jarkko Sakkinen
Cc: Jason Gunthorpe, Shuah Khan, Jaewon Yang, linux-integrity,
linux-kselftest, linux-kernel, Rustam Mustafin, stable
tpm_common_read() indexes data_buffer with the caller supplied *off.
A read issued through io_uring reaches the driver with that offset:
loop_rw_iter() passes &ki_pos to ->read(). This allows out-of-bounds
reads and zeroing of kernel memory:
BUG: KASAN: slab-out-of-bounds in _copy_to_user+0x4b/0x70
Read of size 16 by task iou-wrk-148/149
Call Trace:
_copy_to_user+0x4b/0x70
tpm_common_read+0x16e/0x2c0
loop_rw_iter+0x235/0x5a0
__io_read+0xc6e/0x11a0
io_read+0x8a/0x180
A plain read() racing a write() reaches the same access: read()
samples f_pos before taking buffer_mutex, so it can consume a new
response from the offset left by the previous one:
BUG: KASAN: slab-out-of-bounds in _copy_to_user+0x4b/0x70
Read of size 75 by task trigger/151
Call Trace:
_copy_to_user+0x4b/0x70
tpm_common_read+0x16e/0x2c0
vfs_read+0x189/0xc00
ksys_read+0xf9/0x1d0
Replace *off with a read_pos member of struct file_priv. It is only
accessed under buffer_mutex, so the offset passed by the caller no
longer matters.
Fixes: 9488585b21be ("tpm: add support for partial reads")
Cc: stable@vger.kernel.org
Signed-off-by: Rustam Mustafin <rmustafin@cloudlinux.com>
---
drivers/char/tpm/tpm-dev-common.c | 11 ++++++-----
drivers/char/tpm/tpm-dev.h | 1 +
2 files changed, 7 insertions(+), 5 deletions(-)
diff --git a/drivers/char/tpm/tpm-dev-common.c b/drivers/char/tpm/tpm-dev-common.c
index ee91f456f975..553fc6aca94d 100644
--- a/drivers/char/tpm/tpm-dev-common.c
+++ b/drivers/char/tpm/tpm-dev-common.c
@@ -107,6 +107,7 @@ static void tpm_timeout_work(struct work_struct *work)
mutex_lock(&priv->buffer_mutex);
priv->response_read = true;
priv->response_length = 0;
+ priv->read_pos = 0;
memset(priv->data_buffer, 0, sizeof(priv->data_buffer));
mutex_unlock(&priv->buffer_mutex);
wake_up_interruptible(&priv->async_wait);
@@ -149,21 +150,21 @@ ssize_t tpm_common_read(struct file *file, char __user *buf,
goto out;
}
- rc = copy_to_user(buf, priv->data_buffer + *off, ret_size);
+ rc = copy_to_user(buf, priv->data_buffer + priv->read_pos, ret_size);
if (rc) {
memset(priv->data_buffer, 0, TPM_BUFSIZE);
priv->response_length = 0;
ret_size = -EFAULT;
} else {
- memset(priv->data_buffer + *off, 0, ret_size);
+ memset(priv->data_buffer + priv->read_pos, 0, ret_size);
priv->response_length -= ret_size;
- *off += ret_size;
+ priv->read_pos += ret_size;
}
}
out:
if (!priv->response_length) {
- *off = 0;
+ priv->read_pos = 0;
timer_delete_sync(&priv->user_read_timer);
flush = true;
}
@@ -214,7 +215,7 @@ ssize_t tpm_common_write(struct file *file, const char __user *buf,
priv->response_length = 0;
priv->response_read = false;
- *off = 0;
+ priv->read_pos = 0;
/*
* If in nonblocking mode schedule an async job to send
diff --git a/drivers/char/tpm/tpm-dev.h b/drivers/char/tpm/tpm-dev.h
index f3742bcc73e3..34efb1ffa491 100644
--- a/drivers/char/tpm/tpm-dev.h
+++ b/drivers/char/tpm/tpm-dev.h
@@ -15,6 +15,7 @@ struct file_priv {
struct work_struct async_work;
wait_queue_head_t async_wait;
ssize_t response_length;
+ size_t read_pos;
bool response_read;
bool command_enqueued;
--
2.52.0
^ permalink raw reply related [flat|nested] 4+ messages in thread
* [PATCH v2 2/2] selftests: tpm2: test that io_uring reads ignore the file offset
2026-11-20 12:16 [PATCH v2 0/2] tpm: Track the response read position internally Rustam Mustafin
2026-11-20 12:16 ` [PATCH v2 1/2] " Rustam Mustafin
@ 2026-11-20 12:16 ` Rustam Mustafin
1 sibling, 0 replies; 4+ messages in thread
From: Rustam Mustafin @ 2026-11-20 12:16 UTC (permalink / raw)
To: Peter Huewe, Jarkko Sakkinen
Cc: Jason Gunthorpe, Shuah Khan, Jaewon Yang, linux-integrity,
linux-kselftest, linux-kernel, Rustam Mustafin
Test that an IORING_OP_READ request on /dev/tpm0 and /dev/tpmrm0
returns the whole pending response from its start, although sqe->off
points past the response header. The offset stays inside
priv->data_buffer, so an unpatched kernel fails the test on the
returned header without an out-of-bounds access.
Signed-off-by: Rustam Mustafin <rmustafin@cloudlinux.com>
---
tools/testing/selftests/tpm2/.gitignore | 1 +
tools/testing/selftests/tpm2/Makefile | 4 +
tools/testing/selftests/tpm2/tpm_uring.c | 131 +++++++++++++++++++++++
3 files changed, 136 insertions(+)
create mode 100644 tools/testing/selftests/tpm2/tpm_uring.c
diff --git a/tools/testing/selftests/tpm2/.gitignore b/tools/testing/selftests/tpm2/.gitignore
index 6d6165c5e35d..d2324bd0eca2 100644
--- a/tools/testing/selftests/tpm2/.gitignore
+++ b/tools/testing/selftests/tpm2/.gitignore
@@ -1,3 +1,4 @@
# SPDX-License-Identifier: GPL-2.0-only
AsyncTest.log
SpaceTest.log
+tpm_uring
diff --git a/tools/testing/selftests/tpm2/Makefile b/tools/testing/selftests/tpm2/Makefile
index a9bf9459fb25..67f346f4cdef 100644
--- a/tools/testing/selftests/tpm2/Makefile
+++ b/tools/testing/selftests/tpm2/Makefile
@@ -1,4 +1,8 @@
# SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
+CFLAGS += $(KHDR_INCLUDES) -idirafter ../../../include -Wall
+
+TEST_GEN_PROGS := tpm_uring
+
include ../lib.mk
TEST_PROGS := test_smoke.sh test_space.sh test_async.sh
diff --git a/tools/testing/selftests/tpm2/tpm_uring.c b/tools/testing/selftests/tpm2/tpm_uring.c
new file mode 100644
index 000000000000..9caa3cba5bf0
--- /dev/null
+++ b/tools/testing/selftests/tpm2/tpm_uring.c
@@ -0,0 +1,131 @@
+// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
+/*
+ * An io_uring read chooses its own file offset. The TPM character devices
+ * must ignore it and return the response from its start.
+ */
+
+#include <endian.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <poll.h>
+#include <signal.h>
+#include <io_uring/mini_liburing.h>
+#include "../kselftest.h"
+
+#define RING_ENTRIES 4
+#define BUF_LEN 4096
+#define POLL_TIMEOUT_MS 10000
+#define TPM_VERSION_FILE "/sys/class/tpm/tpm0/tpm_version_major"
+struct tpm_header {
+ uint16_t tag;
+ uint32_t size;
+ uint32_t return_code;
+} __attribute__((packed));
+
+/* TPM2_GetCapability(TPM_CAP_TPM_PROPERTIES, TPM_PT_FIXED, 1) */
+static const unsigned char cmd[] = {
+ 0x80, 0x01, /* TPM_ST_NO_SESSIONS */
+ 0x00, 0x00, 0x00, 0x16, /* command size */
+ 0x00, 0x00, 0x01, 0x7a, /* TPM_CC_GetCapability */
+ 0x00, 0x00, 0x00, 0x06, /* TPM_CAP_TPM_PROPERTIES */
+ 0x00, 0x00, 0x01, 0x00, /* TPM_PT_FIXED */
+ 0x00, 0x00, 0x00, 0x01, /* property count */
+};
+
+static void test_read(struct io_uring *ring, const char *path)
+{
+ static char buf[BUF_LEN];
+ const struct tpm_header *hdr = (const void *)buf;
+ struct pollfd pfd = { .events = POLLIN };
+ struct io_uring_sqe *sqe;
+ struct io_uring_cqe *cqe;
+ int fd, ret, res;
+
+ fd = open(path, O_RDWR);
+ if (fd < 0) {
+ if (errno == ENOENT)
+ ksft_test_result_skip("%s: not present\n", path);
+ else
+ ksft_test_result_fail("%s: open: %s\n", path,
+ strerror(errno));
+ return;
+ }
+
+ if (write(fd, cmd, sizeof(cmd)) != sizeof(cmd)) {
+ ksft_test_result_fail("%s: write: %s\n", path, strerror(errno));
+ goto out;
+ }
+
+ pfd.fd = fd;
+ if (poll(&pfd, 1, POLL_TIMEOUT_MS) != 1) {
+ ksft_test_result_fail("%s: no response to the command\n", path);
+ goto out;
+ }
+
+ sqe = io_uring_get_sqe(ring);
+ memset(sqe, 0, sizeof(*sqe));
+ sqe->opcode = IORING_OP_READ;
+ sqe->fd = fd;
+ /*
+ * Inside the response buffer: an unpatched kernel returns the response
+ * without its header instead of reading out of bounds
+ */
+ sqe->off = sizeof(*hdr);
+ sqe->addr = (unsigned long)buf;
+ sqe->len = BUF_LEN;
+ ret = io_uring_submit(ring);
+ if (ret != 1) {
+ ksft_test_result_fail("%s: io_uring_submit: %d\n", path, ret);
+ goto out;
+ }
+ ret = io_uring_wait_cqe(ring, &cqe);
+ if (ret) {
+ ksft_test_result_fail("%s: io_uring_wait_cqe: %d\n", path, ret);
+ goto out;
+ }
+ res = cqe->res;
+ io_uring_cqe_seen(ring);
+
+ /* the offset is ignored: the whole response, from its start */
+ ksft_test_result(res >= (int)sizeof(*hdr) &&
+ be32toh(hdr->size) == res && !hdr->return_code,
+ "%s: io_uring read %d\n", path, res);
+out:
+ close(fd);
+}
+
+/* the command is TPM2 only; test_smoke.sh skips the same way */
+static int tpm_version_major(void)
+{
+ FILE *f = fopen(TPM_VERSION_FILE, "r");
+ int major = 0;
+
+ if (f) {
+ if (fscanf(f, "%d", &major) != 1)
+ major = 0;
+ fclose(f);
+ }
+ return major;
+}
+
+int main(void)
+{
+ struct io_uring ring;
+
+ ksft_print_header();
+ if (tpm_version_major() != 2)
+ ksft_exit_skip("no TPM 2.0 device\n");
+ if (io_uring_queue_init(RING_ENTRIES, &ring, 0)) {
+ /* not built in, or disabled by kernel.io_uring_disabled */
+ if (errno == ENOSYS || errno == EPERM)
+ ksft_exit_skip("io_uring_setup: %s\n", strerror(errno));
+ ksft_exit_fail_msg("io_uring_queue_init: %s\n", strerror(errno));
+ }
+ ksft_set_plan(2);
+
+ test_read(&ring, "/dev/tpm0");
+ test_read(&ring, "/dev/tpmrm0");
+
+ io_uring_queue_exit(&ring);
+ ksft_finished();
+}
--
2.52.0
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH v2 1/2] tpm: Track the response read position internally
2026-11-20 12:16 ` [PATCH v2 1/2] " Rustam Mustafin
@ 2026-11-20 12:27 ` Jarkko Sakkinen
0 siblings, 0 replies; 4+ messages in thread
From: Jarkko Sakkinen @ 2026-11-20 12:27 UTC (permalink / raw)
To: Rustam Mustafin
Cc: Peter Huewe, Jason Gunthorpe, Shuah Khan, Jaewon Yang,
linux-integrity, linux-kselftest, linux-kernel, stable
On Fri, Nov 20, 2026 at 07:16:13PM +0700, Rustam Mustafin wrote:
> tpm_common_read() indexes data_buffer with the caller supplied *off.
> A read issued through io_uring reaches the driver with that offset:
> loop_rw_iter() passes &ki_pos to ->read(). This allows out-of-bounds
> reads and zeroing of kernel memory:
>
> BUG: KASAN: slab-out-of-bounds in _copy_to_user+0x4b/0x70
> Read of size 16 by task iou-wrk-148/149
> Call Trace:
> _copy_to_user+0x4b/0x70
> tpm_common_read+0x16e/0x2c0
> loop_rw_iter+0x235/0x5a0
> __io_read+0xc6e/0x11a0
> io_read+0x8a/0x180
>
> A plain read() racing a write() reaches the same access: read()
> samples f_pos before taking buffer_mutex, so it can consume a new
> response from the offset left by the previous one:
>
> BUG: KASAN: slab-out-of-bounds in _copy_to_user+0x4b/0x70
> Read of size 75 by task trigger/151
> Call Trace:
> _copy_to_user+0x4b/0x70
> tpm_common_read+0x16e/0x2c0
> vfs_read+0x189/0xc00
> ksys_read+0xf9/0x1d0
>
> Replace *off with a read_pos member of struct file_priv. It is only
> accessed under buffer_mutex, so the offset passed by the caller no
> longer matters.
>
> Fixes: 9488585b21be ("tpm: add support for partial reads")
> Cc: stable@vger.kernel.org
> Signed-off-by: Rustam Mustafin <rmustafin@cloudlinux.com>
> ---
So... why does not this block uring? You are implementing a feature
as a bug fix, or so that it appears.
> drivers/char/tpm/tpm-dev-common.c | 11 ++++++-----
> drivers/char/tpm/tpm-dev.h | 1 +
> 2 files changed, 7 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/char/tpm/tpm-dev-common.c b/drivers/char/tpm/tpm-dev-common.c
> index ee91f456f975..553fc6aca94d 100644
> --- a/drivers/char/tpm/tpm-dev-common.c
> +++ b/drivers/char/tpm/tpm-dev-common.c
> @@ -107,6 +107,7 @@ static void tpm_timeout_work(struct work_struct *work)
> mutex_lock(&priv->buffer_mutex);
> priv->response_read = true;
> priv->response_length = 0;
> + priv->read_pos = 0;
> memset(priv->data_buffer, 0, sizeof(priv->data_buffer));
> mutex_unlock(&priv->buffer_mutex);
> wake_up_interruptible(&priv->async_wait);
> @@ -149,21 +150,21 @@ ssize_t tpm_common_read(struct file *file, char __user *buf,
> goto out;
> }
>
> - rc = copy_to_user(buf, priv->data_buffer + *off, ret_size);
> + rc = copy_to_user(buf, priv->data_buffer + priv->read_pos, ret_size);
> if (rc) {
> memset(priv->data_buffer, 0, TPM_BUFSIZE);
> priv->response_length = 0;
> ret_size = -EFAULT;
> } else {
> - memset(priv->data_buffer + *off, 0, ret_size);
> + memset(priv->data_buffer + priv->read_pos, 0, ret_size);
> priv->response_length -= ret_size;
> - *off += ret_size;
> + priv->read_pos += ret_size;
> }
> }
>
> out:
> if (!priv->response_length) {
> - *off = 0;
> + priv->read_pos = 0;
> timer_delete_sync(&priv->user_read_timer);
> flush = true;
> }
> @@ -214,7 +215,7 @@ ssize_t tpm_common_write(struct file *file, const char __user *buf,
>
> priv->response_length = 0;
> priv->response_read = false;
> - *off = 0;
> + priv->read_pos = 0;
>
> /*
> * If in nonblocking mode schedule an async job to send
> diff --git a/drivers/char/tpm/tpm-dev.h b/drivers/char/tpm/tpm-dev.h
> index f3742bcc73e3..34efb1ffa491 100644
> --- a/drivers/char/tpm/tpm-dev.h
> +++ b/drivers/char/tpm/tpm-dev.h
> @@ -15,6 +15,7 @@ struct file_priv {
> struct work_struct async_work;
> wait_queue_head_t async_wait;
> ssize_t response_length;
> + size_t read_pos;
> bool response_read;
> bool command_enqueued;
>
> --
> 2.52.0
>
Br, Jarkko
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-11-20 12:27 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-20 12:16 [PATCH v2 0/2] tpm: Track the response read position internally Rustam Mustafin
2026-11-20 12:16 ` [PATCH v2 1/2] " Rustam Mustafin
2026-11-20 12:27 ` Jarkko Sakkinen
2026-11-20 12:16 ` [PATCH v2 2/2] selftests: tpm2: test that io_uring reads ignore the file offset Rustam Mustafin
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).