x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 304 insertions, 0 deletions

diff --git a/Documentation/trace/user_events.rst b/Documentation/trace/user_events.rst
new file mode 100644
index 000000000..1d5a7626e
--- /dev/null
+++ b/Documentation/trace/user_events.rst
@@ -0,0 +1,304 @@
+=========================================
+user_events: User-based Event Tracing
+=========================================
+
+:Author: Beau Belgrave
+
+Overview
+--------
+User based trace events allow user processes to create events and trace data
+that can be viewed via existing tools, such as ftrace and perf.
+To enable this feature, build your kernel with CONFIG_USER_EVENTS=y.
+
+Programs can view status of the events via
+/sys/kernel/tracing/user_events_status and can both register and write
+data out via /sys/kernel/tracing/user_events_data.
+
+Programs can also use /sys/kernel/tracing/dynamic_events to register and
+delete user based events via the u: prefix. The format of the command to
+dynamic_events is the same as the ioctl with the u: prefix applied. This
+requires CAP_PERFMON due to the event persisting, otherwise -EPERM is returned.
+
+Typically programs will register a set of events that they wish to expose to
+tools that can read trace_events (such as ftrace and perf). The registration
+process tells the kernel which address and bit to reflect if any tool has
+enabled the event and data should be written. The registration will give back
+a write index which describes the data when a write() or writev() is called
+on the /sys/kernel/tracing/user_events_data file.
+
+The structures referenced in this document are contained within the
+/include/uapi/linux/user_events.h file in the source tree.
+
+**NOTE:** *Both user_events_status and user_events_data are under the tracefs
+filesystem and may be mounted at different paths than above.*
+
+Registering
+-----------
+Registering within a user process is done via ioctl() out to the
+/sys/kernel/tracing/user_events_data file. The command to issue is
+DIAG_IOCSREG.
+
+This command takes a packed struct user_reg as an argument::
+
+  struct user_reg {
+        /* Input: Size of the user_reg structure being used */
+        __u32 size;
+
+        /* Input: Bit in enable address to use */
+        __u8 enable_bit;
+
+        /* Input: Enable size in bytes at address */
+        __u8 enable_size;
+
+        /* Input: Flags to use, if any */
+        __u16 flags;
+
+        /* Input: Address to update when enabled */
+        __u64 enable_addr;
+
+        /* Input: Pointer to string with event name, description and flags */
+        __u64 name_args;
+
+        /* Output: Index of the event to use when writing data */
+        __u32 write_index;
+  } __attribute__((__packed__));
+
+The struct user_reg requires all the above inputs to be set appropriately.
+
++ size: This must be set to sizeof(struct user_reg).
+
++ enable_bit: The bit to reflect the event status at the address specified by
+  enable_addr.
+
++ enable_size: The size of the value specified by enable_addr.
+  This must be 4 (32-bit) or 8 (64-bit). 64-bit values are only allowed to be
+  used on 64-bit kernels, however, 32-bit can be used on all kernels.
+
++ flags: The flags to use, if any.
+  Callers should first attempt to use flags and retry without flags to ensure
+  support for lower versions of the kernel. If a flag is not supported -EINVAL
+  is returned.
+
++ enable_addr: The address of the value to use to reflect event status. This
+  must be naturally aligned and write accessible within the user program.
+
++ name_args: The name and arguments to describe the event, see command format
+  for details.
+
+The following flags are currently supported.
+
++ USER_EVENT_REG_PERSIST: The event will not delete upon the last reference
+  closing. Callers may use this if an event should exist even after the
+  process closes or unregisters the event. Requires CAP_PERFMON otherwise
+  -EPERM is returned.
+
++ USER_EVENT_REG_MULTI_FORMAT: The event can contain multiple formats. This
+  allows programs to prevent themselves from being blocked when their event
+  format changes and they wish to use the same name. When this flag is used the
+  tracepoint name will be in the new format of "name.unique_id" vs the older
+  format of "name". A tracepoint will be created for each unique pair of name
+  and format. This means if several processes use the same name and format,
+  they will use the same tracepoint. If yet another process uses the same name,
+  but a different format than the other processes, it will use a different
+  tracepoint with a new unique id. Recording programs need to scan tracefs for
+  the various different formats of the event name they are interested in
+  recording. The system name of the tracepoint will also use "user_events_multi"
+  instead of "user_events". This prevents single-format event names conflicting
+  with any multi-format event names within tracefs. The unique_id is output as
+  a hex string. Recording programs should ensure the tracepoint name starts with
+  the event name they registered and has a suffix that starts with . and only
+  has hex characters. For example to find all versions of the event "test" you
+  can use the regex "^test\.[0-9a-fA-F]+$".
+
+Upon successful registration the following is set.
+
++ write_index: The index to use for this file descriptor that represents this
+  event when writing out data. The index is unique to this instance of the file
+  descriptor that was used for the registration. See writing data for details.
+
+User based events show up under tracefs like any other event under the
+subsystem named "user_events". This means tools that wish to attach to the
+events need to use /sys/kernel/tracing/events/user_events/[name]/enable
+or perf record -e user_events:[name] when attaching/recording.
+
+**NOTE:** The event subsystem name by default is "user_events". Callers should
+not assume it will always be "user_events". Operators reserve the right in the
+future to change the subsystem name per-process to accommodate event isolation.
+In addition if the USER_EVENT_REG_MULTI_FORMAT flag is used the tracepoint name
+will have a unique id appended to it and the system name will be
+"user_events_multi" as described above.
+
+Command Format
+^^^^^^^^^^^^^^
+The command string format is as follows::
+
+  name[:FLAG1[,FLAG2...]] [Field1[;Field2...]]
+
+Supported Flags
+^^^^^^^^^^^^^^^
+None yet
+
+Field Format
+^^^^^^^^^^^^
+::
+
+  type name [size]
+
+Basic types are supported (__data_loc, u32, u64, int, char, char[20], etc).
+User programs are encouraged to use clearly sized types like u32.
+
+**NOTE:** *Long is not supported since size can vary between user and kernel.*
+
+The size is only valid for types that start with a struct prefix.
+This allows user programs to describe custom structs out to tools, if required.
+
+For example, a struct in C that looks like this::
+
+  struct mytype {
+    char data[20];
+  };
+
+Would be represented by the following field::
+
+  struct mytype myname 20
+
+Deleting
+--------
+Deleting an event from within a user process is done via ioctl() out to the
+/sys/kernel/tracing/user_events_data file. The command to issue is
+DIAG_IOCSDEL.
+
+This command only requires a single string specifying the event to delete by
+its name. Delete will only succeed if there are no references left to the
+event (in both user and kernel space). User programs should use a separate file
+to request deletes than the one used for registration due to this.
+
+**NOTE:** By default events will auto-delete when there are no references left
+to the event. If programs do not want auto-delete, they must use the
+USER_EVENT_REG_PERSIST flag when registering the event. Once that flag is used
+the event exists until DIAG_IOCSDEL is invoked. Both register and delete of an
+event that persists requires CAP_PERFMON, otherwise -EPERM is returned. When
+there are multiple formats of the same event name, all events with the same
+name will be attempted to be deleted. If only a specific version is wanted to
+be deleted then the /sys/kernel/tracing/dynamic_events file should be used for
+that specific format of the event.
+
+Unregistering
+-------------
+If after registering an event it is no longer wanted to be updated then it can
+be disabled via ioctl() out to the /sys/kernel/tracing/user_events_data file.
+The command to issue is DIAG_IOCSUNREG. This is different than deleting, where
+deleting actually removes the event from the system. Unregistering simply tells
+the kernel your process is no longer interested in updates to the event.
+
+This command takes a packed struct user_unreg as an argument::
+
+  struct user_unreg {
+        /* Input: Size of the user_unreg structure being used */
+        __u32 size;
+
+        /* Input: Bit to unregister */
+        __u8 disable_bit;
+
+        /* Input: Reserved, set to 0 */
+        __u8 __reserved;
+
+        /* Input: Reserved, set to 0 */
+        __u16 __reserved2;
+
+        /* Input: Address to unregister */
+        __u64 disable_addr;
+  } __attribute__((__packed__));
+
+The struct user_unreg requires all the above inputs to be set appropriately.
+
++ size: This must be set to sizeof(struct user_unreg).
+
++ disable_bit: This must be set to the bit to disable (same bit that was
+  previously registered via enable_bit).
+
++ disable_addr: This must be set to the address to disable (same address that was
+  previously registered via enable_addr).
+
+**NOTE:** Events are automatically unregistered when execve() is invoked. During
+fork() the registered events will be retained and must be unregistered manually
+in each process if wanted.
+
+Status
+------
+When tools attach/record user based events the status of the event is updated
+in realtime. This allows user programs to only incur the cost of the write() or
+writev() calls when something is actively attached to the event.
+
+The kernel will update the specified bit that was registered for the event as
+tools attach/detach from the event. User programs simply check if the bit is set
+to see if something is attached or not.
+
+Administrators can easily check the status of all registered events by reading
+the user_events_status file directly via a terminal. The output is as follows::
+
+  Name [# Comments]
+  ...
+
+  Active: ActiveCount
+  Busy: BusyCount
+
+For example, on a system that has a single event the output looks like this::
+
+  test
+
+  Active: 1
+  Busy: 0
+
+If a user enables the user event via ftrace, the output would change to this::
+
+  test # Used by ftrace
+
+  Active: 1
+  Busy: 1
+
+Writing Data
+------------
+After registering an event the same fd that was used to register can be used
+to write an entry for that event. The write_index returned must be at the start
+of the data, then the remaining data is treated as the payload of the event.
+
+For example, if write_index returned was 1 and I wanted to write out an int
+payload of the event. Then the data would have to be 8 bytes (2 ints) in size,
+with the first 4 bytes being equal to 1 and the last 4 bytes being equal to the
+value I want as the payload.
+
+In memory this would look like this::
+
+  int index;
+  int payload;
+
+User programs might have well known structs that they wish to use to emit out
+as payloads. In those cases writev() can be used, with the first vector being
+the index and the following vector(s) being the actual event payload.
+
+For example, if I have a struct like this::
+
+  struct payload {
+        int src;
+        int dst;
+        int flags;
+  } __attribute__((__packed__));
+
+It's advised for user programs to do the following::
+
+  struct iovec io[2];
+  struct payload e;
+
+  io[0].iov_base = &write_index;
+  io[0].iov_len = sizeof(write_index);
+  io[1].iov_base = &e;
+  io[1].iov_len = sizeof(e);
+
+  writev(fd, (const struct iovec*)io, 2);
+
+**NOTE:** *The write_index is not emitted out into the trace being recorded.*
+
+Example Code
+------------
+See sample code in samples/user_events.