Tracepoints for the VFS?

Adding tracepoints to some kernel subsystems has been controversial—or disallowed—due to concerns about the user-space ABI that they might create. The virtual filesystem (VFS) layer has long been one of the subsystems that has not allowed any tracepoints, but that may be changing. At the 2025 Linux Storage, Filesystem, Memory Management, and BPF Summit (LSFMM+BPF), Ted Ts'o led a discussion about whether the ABI concerns are outweighed by the utility of tracepoints for the VFS.

由于担心可能会引入用户空间 ABI，一些内核子系统中添加 tracepoint 一直存在争议，甚至被禁止。虚拟文件系统（VFS）层长期以来就是一个不允许添加 tracepoint 的子系统，但这种情况可能正在发生变化。在 2025 年的 Linux 存储、文件系统、内存管理和 BPF 峰会（LSFMM+BPF）上，Ted Ts’o 主持了一场讨论，探讨 VFS 是否应该添加 tracepoint，以及其带来的实用性是否能胜过对 ABI 的担忧。

Ts'o began by noting that Al Viro, who has opposed VFS tracepoints over the years, was not present, but that VFS co-maintainer Christian Brauner was in attendance to give his opinions on the matter. Historically, there have been concerns about placing tracepoints at various places in the VFS, Ts'o said, such as for system calls like open() and rename(). One concern is about tracepoints in hot paths affecting performance, but he thinks that could be worked around by keeping the tracepoints at the system-call level. Another is that the tracepoints "might potentially constrain our implementation" because of the user-space interface question, with the powertop incident often cited as an example of the problem.

Ts’o 说，长期反对在 VFS 中添加 tracepoint 的 Al Viro 并未出席会议，但 VFS 的共同维护者 Christian Brauner 到场，可以就此问题发表看法。他指出，历史上人们曾担忧在 VFS 的多个位置添加