Prerequisites
Last working version
10.1.3
Stopped working in version
10.1.4
Node.js version
24.21.0
Operating system
Linux
Operating system version (i.e. 20.04, 11.3, 10)
Ubuntu, kernel 6.8.0 (app runs in node:24-alpine; static root is an ext4 bind mount)
💥 Regression Report
Since 10.1.4 (the fix for GHSA-r799-r9gc-m956), getPathSpellingStatus() runs on every request served through the default wildcard: true route. For each path segment it calls readdir() on the parent directory and then entries.includes(segment). So every request costs time proportional to the number of entries in each directory along the path.
We serve product photos from one flat directory with ~815k files. After upgrading from 10.1.3 to 10.1.4, median latency for /static/product/photo/.webp in production went from ~3 ms to ~1.8 s (p90 ~4 s, p99 ~10 s). That includes 304 Not Modified responses and requests for non-existent files. Files in small directories under the same root are unaffected (~7 ms). A single stat() of the same file takes <1 ms on the host and inside the container, so the filesystem is not the bottleneck.
The spelling check only matters on case-insensitive filesystems. On case-sensitive ones (ext4, the default on Linux), a case alias cannot resolve to a different file, yet the full directory scan still runs on every request.
It is also an amplification vector: any unauthenticated, well-formed request under a large directory triggers a full readdir() of it.
Steps to Reproduce
// repro.mjs — node repro.mjs 10.1.3|10.1.4 (after installing that version)
import Fastify from 'fastify'
import fastifyStatic from '@fastify/static'
import { mkdirSync, writeFileSync } from 'node:fs'
import { join } from 'node:path'
import { tmpdir } from 'node:os'
const root = join(tmpdir(), 'fastify-static-large-dir')
const dir = join(root, 'photo')
mkdirSync(dir, { recursive: true })
for (let i = 0; i < 300_000; i++) writeFileSync(join(dir, f${i}.webp), '')
const app = Fastify()
await app.register(fastifyStatic, { root, prefix: '/static/' })
await app.ready()
const n = 20
const t = performance.now()
for (let i = 0; i < n; i++) {
const res = await app.inject(/static/photo/f${i * 1000 + 7}.webp)
if (res.statusCode !== 200) throw new Error(String(res.statusCode))
}
console.log(process.argv[2], 'avg ms/request:', ((performance.now() - t) / n).toFixed(1))
Results (macOS, 300k files, sequential requests):
| version |
avg per request |
| 10.1.3 |
1.9 ms |
| 10.1.4 |
419 ms |
Expected Behavior
Serving a file should not scale with the size of its directory, as in 10.1.3. Some possible approaches:
Detect once per root at startup whether the filesystem is case-insensitive (e.g. stat an upper-cased variant of a known entry), and skip the spelling check on case-sensitive filesystems.
Or cache the directory listing per directory, invalidated by mtime, and use a Set instead of Array#includes.
Or provide an option to opt out of the spelling check, for deployments on a known case-sensitive filesystem.
As a workaround we pinned @fastify/static@10.1.3.
Prerequisites
Last working version
10.1.3
Stopped working in version
10.1.4
Node.js version
24.21.0
Operating system
Linux
Operating system version (i.e. 20.04, 11.3, 10)
Ubuntu, kernel 6.8.0 (app runs in node:24-alpine; static root is an ext4 bind mount)
💥 Regression Report
Since 10.1.4 (the fix for GHSA-r799-r9gc-m956), getPathSpellingStatus() runs on every request served through the default wildcard: true route. For each path segment it calls readdir() on the parent directory and then entries.includes(segment). So every request costs time proportional to the number of entries in each directory along the path.
We serve product photos from one flat directory with ~815k files. After upgrading from 10.1.3 to 10.1.4, median latency for /static/product/photo/.webp in production went from ~3 ms to ~1.8 s (p90 ~4 s, p99 ~10 s). That includes 304 Not Modified responses and requests for non-existent files. Files in small directories under the same root are unaffected (~7 ms). A single stat() of the same file takes <1 ms on the host and inside the container, so the filesystem is not the bottleneck.
The spelling check only matters on case-insensitive filesystems. On case-sensitive ones (ext4, the default on Linux), a case alias cannot resolve to a different file, yet the full directory scan still runs on every request.
It is also an amplification vector: any unauthenticated, well-formed request under a large directory triggers a full readdir() of it.
Steps to Reproduce
Results (macOS, 300k files, sequential requests):
Expected Behavior
Serving a file should not scale with the size of its directory, as in 10.1.3. Some possible approaches:
Detect once per root at startup whether the filesystem is case-insensitive (e.g. stat an upper-cased variant of a known entry), and skip the spelling check on case-sensitive filesystems.
Or cache the directory listing per directory, invalidated by mtime, and use a Set instead of Array#includes.
Or provide an option to opt out of the spelling check, for deployments on a known case-sensitive filesystem.
As a workaround we pinned @fastify/static@10.1.3.