[jgarzik]

No description provided.

[Copilot]

Note

Copilot was unable to complete this review at the selected tier due to model policy or plan restrictions.

🟡 Changes recommended

The change still has moderate unresolved issues in core logic—an O(n²) path-walk loop in fs/namei.rs and a potentially panicking under-allocation in io/poll.rs—that need to be addressed before it can be safely approved.

2 open issues

• Path walker scans components twice causing quadratic time · New
• PollContext capacity underestimates poll registrations · New

What changed in this PR

This pull request focuses on reducing allocation churn and allocator work in hot paths by pre-sizing buffers (with_capacity) where sizes are known/bounded, and by removing unnecessary intermediate allocations across netlink, procfs, directory iteration, pathname walking, and poll/select setup.

Changes:

• Pre-size transient output buffers for netlink dumps and multiple /proc/* generators to reduce reallocation overhead.
• Remove an intermediate allocation in netlink string attribute encoding by writing attributes directly into the message buffer.
• Rework path component iteration to avoid allocating/collecting components; add small unit tests.
• Introduce/use a fallible PollContext::with_capacity to size poll/select context storage more predictably.
• Update style guidance on allocation sizing (especially avoiding allocator work under locks); remove the standalone sizing note doc.

Outstanding review comments (must be addressed):

• fs/namei.rs: Current component loop calls next_comp(rest) twice per component, turning path walking into O(n²) on many-component paths; keep a lookahead to scan once. (severity: moderate)
• fs/proc/mounts.rs: show_mountinfo() clones mountpoint Vec<u8> per mount via vis.iter().cloned(); iterate by reference and only clone what’s needed (e.g., Arc<Mount>) to avoid undermining pre-sizing. (severity: moderate)
• io/poll.rs: PollContext::with_capacity assumes 1 registration per fd, but some fds register multiple wait queues; entries.push(...) can exceed capacity and trigger an infallible grow/panic despite the ENOMEM intent—budget for multiple registrations per fd. (severity: moderate)

File	Summary
net/​netlink/​mod.rs	Pre-allocates dump buffer capacity for netlink multipart replies/dumps.
net/​netlink/​attr.rs	Avoids temporary allocation when emitting NUL-terminated string attributes by encoding directly into the message buffer.
mm/​slab.rs	Pre-sizes slabinfo() output string based on estimated output size/line count.
io/​poll.rs	Adds fallible PollContext::with_capacity and adjusts registration behavior; capacity assumptions flagged in review comments.
fs/​proc/​mounts.rs	Pre-sizes /proc/<pid>/{mounts,mountinfo} output buffers based on visible mount count; cloning behavior flagged in review comments.
fs/​poll.rs	Uses PollContext::with_capacity sized from expected registrations for poll/select.
fs/​namei.rs	Switches to allocation-free, in-place path component iteration; adds helper unit tests; loop complexity flagged in review comments.
fs/​dir.rs	Pre-sizes getdents buffer to avoid allocations while directory locks are held.
doc/​STYLE.md	Documents allocation sizing guidance, especially around avoiding allocator work under locks/atomic contexts.
doc/​res/​hk-with-capacity.md	Removes the standalone sizing write-up (content consolidated into style guidance).

This pull request requires changes.

_{Review effort: Balanced}