ring-buffer: Just update the subbuffers when changing their allocation order
authorSteven Rostedt (Google) <rostedt@goodmis.org>
Tue, 19 Dec 2023 18:54:26 +0000 (13:54 -0500)
committerSteven Rostedt (Google) <rostedt@goodmis.org>
Thu, 21 Dec 2023 16:02:52 +0000 (11:02 -0500)
commit8e7b58c27b3c567316a51079b375b846f9223bba
tree0741788e8611ed2602f9bbd6feb4bf5e67adde0b
parent353cc219372935805218e05a7754a059ab104641
ring-buffer: Just update the subbuffers when changing their allocation order

The ring_buffer_subbuf_order_set() was creating ring_buffer_per_cpu
cpu_buffers with the new subbuffers with the updated order, and if they
all successfully were created, then they the ring_buffer's per_cpu buffers
would be freed and replaced by them.

The problem is that the freed per_cpu buffers contains state that would be
lost. Running the following commands:

1. # echo 3 > /sys/kernel/tracing/buffer_subbuf_order
2. # echo 0 > /sys/kernel/tracing/tracing_cpumask
3. # echo 1 > /sys/kernel/tracing/snapshot
4. # echo ff > /sys/kernel/tracing/tracing_cpumask
5. # echo test > /sys/kernel/tracing/trace_marker

Would result in:

 -bash: echo: write error: Bad file descriptor

That's because the state of the per_cpu buffers of the snapshot buffer is
lost when the order is changed (the order of a freed snapshot buffer goes
to 0 to save memory, and when the snapshot buffer is allocated again, it
goes back to what the main buffer is).

In operation 2, the snapshot buffers were set to "disable" (as all the
ring buffers CPUs were disabled).

In operation 3, the snapshot is allocated and a call to
ring_buffer_subbuf_order_set() replaced the per_cpu buffers losing the
"record_disable" count.

When it was enabled again, the atomic_dec(&cpu_buffer->record_disable) was
decrementing a zero, setting it to -1. Writing 1 into the snapshot would
swap the snapshot buffer with the main buffer, so now the main buffer is
"disabled", and nothing can write to the ring buffer anymore.

Instead of creating new per_cpu buffers and losing the state of the old
buffers, basically do what the resize does and just allocate new subbuf
pages into the new_pages link list of the per_cpu buffer and if they all
succeed, then replace the old sub buffers with the new ones. This keeps
the per_cpu buffer descriptor in tact and by doing so, keeps its state.

Link: https://lore.kernel.org/linux-trace-kernel/20231219185630.944104939@goodmis.org
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Tzvetomir Stoyanov <tz.stoyanov@gmail.com>
Cc: Vincent Donnefort <vdonnefort@google.com>
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Fixes: f9b94daa542a ("ring-buffer: Set new size of the ring buffer sub page")
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
kernel/trace/ring_buffer.c