Stop the one-line install handing a standard user a way in

The bootstrap command staged its script into C:\Windows\Temp and then ran it
elevated. That directory grants Users create-file with CREATOR OWNER full
control, so somebody without admin rights can pre-plant the exact path the
elevated run is about to execute. It now stages under Program Files, creates the
directory with an explicit DACL when it is missing, and refuses to run a staged
file that is not owned by an admin principal. A failed hardening is loud rather
than silent: the run stops instead of carrying on with a file it could not
protect.

The token moved out of the query string into a header. In a query string it
lands verbatim in the reverse proxy and IIS logs, which is a credential written
to a file nobody thinks of as a secret store. The query form is still accepted,
because bays provisioned before this change are holding install commands that
use it, and it logs that it was used - without logging the token - so a site can
see who still needs a re-issued command before turning it off with
plugin.endpoints.bootstrapquerytoken.

Two smaller things this also closes:

A request with no token at all answered 400 naming the header, which told an
anonymous caller the state of that switch and contradicted the module's own
promise that anything unusable answers 404. No token is now 404. The 400 is kept
for a token genuinely presented in the query form while the query form is off,
which is a caller who needs to know their command is stale.

The rollout documentation was republishing the vulnerable one-liner verbatim, so
a tech following the runbook reinstated exactly what this fixes. Both documents
now carry the current command.

Also corrects the operator-facing copy for the token, which named a query
argument and a header that do not exist: it is ?t= and X-Endpoint-Bootstrap-Token,
per bootstrap.py's own constants. Anybody following the old text got a 404 with
no reason given.

Both switches added in this range are now declared with the other plugin
settings and reachable from their settings page. Read through Setting.get but
never declared, they were rows that did not exist until someone wrote one by
hand, which on an air-gapped box means the feature was unreachable.
This commit is contained in:
cproudlock
2026-09-18 13:30:08 -04:00
parent 3f80425f30
commit 987f5c0d0f
7 changed files with 934 additions and 81 deletions

View File

@@ -14,14 +14,25 @@ Routes
POST /siteagent/bootstrap/tokens admin, audited; secret once
GET /siteagent/bootstrap/tokens admin; never a secret
POST /siteagent/bootstrap/tokens/<id>/revoke admin, audited
GET /siteagent/bootstrap.ps1?t=... token only; not a use
GET /siteagent/bootstrap/package?t=... token only; one use
GET /siteagent/bootstrap.ps1 token only; not a use
GET /siteagent/bootstrap/package token only; one use
The two token-authenticated routes answer 404 for anything not currently
usable - unknown, expired, revoked or exhausted - so the URL confirms nothing
about which tokens exist. Both are per-IP rate limited with the same fixed
window shape the login and payload-download limiters use, and both write an
audit row carrying the caller's IP.
The two token-authenticated routes take the token in the
X-Endpoint-Bootstrap-Token header, and still accept the deprecated ?t= form so
a command already pasted on a bay keeps working. They answer 404 for anything
not currently usable - unknown, malformed, expired, revoked, exhausted, or no
token presented at all - so the URL confirms nothing about which tokens exist,
nor about whether the query form is still accepted here. The one 400 is a real
query token offered at a site that has turned the query form off.
The admin UI's Download .ps1 link has to use the query form (a browser link
cannot send a header), so its URL carries a marker that tells it apart from a
bay still running a command pasted before this change. Only the bay logs the
deprecation warning, which is what the rollout counts un-migrated bays with.
Both routes are per-IP rate limited with the same fixed window shape the login
and payload-download limiters use, and both write an audit row carrying the
caller's IP.
"""
import hashlib
@@ -60,12 +71,60 @@ LIST_LIMIT = 100
SCRIPT_PATH = '/api/endpoints/siteagent/bootstrap.ps1'
PACKAGE_PATH = '/api/endpoints/siteagent/bootstrap/package'
# Where the one-line command drops the script before running it. A literal, not
# $env:TEMP: the line has to survive being pasted into an already-running
# Where the one-line command stages the script before running it. A literal,
# not $env:TEMP: the line has to survive being pasted into an already-running
# PowerShell, which expands every $ inside the outer double quotes before the
# child process sees them. The install needs elevation anyway, so this path is
# writable by definition.
COMMAND_SCRIPT_PATH = r'C:\Windows\Temp\shopdb-bootstrap.ps1'
# child process sees them.
#
# NOT C:\Windows\Temp. That directory grants Users create-file and gives
# CREATOR OWNER full control of what they create, so any standard user on the
# bay can pre-plant this exact path and wait for a tech to paste the command,
# which then runs the planted content elevated. Program Files is admin-only on
# a stock install, the command re-asserts an admin-only DACL on the directory
# it creates, and it refuses to run a staged file owned by anyone else.
COMMAND_STAGE_DIR = r'C:\Program Files\ShopDB\bootstrap'
COMMAND_SCRIPT_PATH = COMMAND_STAGE_DIR + r'\shopdb-bootstrap.ps1'
# Well-known SIDs, not names: the bays are domain-joined and a renamed or
# localized Administrators group would quietly break both the grant and the
# ownership test.
ADMINISTRATORS_SID = 'S-1-5-32-544'
SYSTEM_SID = 'S-1-5-18'
# The token the two public routes accept. A header, not ?t=: a query string
# lands verbatim in the reverse proxy and gunicorn access logs, and a live
# install token has no business sitting in a log file.
BOOTSTRAP_TOKEN_HEADER = 'X-Endpoint-Bootstrap-Token'
QUERY_TOKEN_ARG = 't'
# Where the token on this request came from. Kept apart because the rollout
# counts un-migrated bays by the deprecation warning SOURCE_QUERY logs: an
# admin clicking Download .ps1 in the browser must not land in that count, and
# a request carrying no token at all must not answer differently from one
# carrying a token nobody issued.
SOURCE_HEADER = 'header'
SOURCE_QUERY = 'query'
SOURCE_UI_LINK = 'uilink'
SOURCE_NONE = 'none'
# The marker on the admin Download .ps1 link. A hint, NOT authentication: a
# caller can add it by hand, and nothing is granted by it. What matters is the
# other direction - a one-line command generated before this change cannot
# carry it, so a request without it really is a bay to go and re-issue.
UI_LINK_ARG = 'via'
UI_LINK_VALUE = 'download'
# DEPRECATED, kept on purpose: every command already pasted on a bay, and the
# UI's "Download .ps1" link (a browser link cannot send a header), carry the
# token in the query string. Accepting it stays the default so a mid-rollout
# bay keeps working; a site that has re-issued its live commands turns it off.
# Turning it off also disables the download link.
#
# Declared in the endpoints plugin's get_settings_defaults() (so the row seeds
# boolean, not string) and flipped from Settings > Endpoints > Build site agent
# > One-line install, "Accept the token in the URL". A switch with no seed and
# no control would be inert on an air-gapped box.
QUERY_TOKEN_SETTING = 'plugin.endpoints.bootstrapquerytoken'
# Fixed-window per-IP limiter, same shape and cache extension as the login
# limiter in shopdb.core.api.auth and the GE-Enforce payload limiter. A tech
@@ -86,6 +145,22 @@ def _naive_utcnow():
def _caller_ip():
"""The caller address for the audit rows and the deprecation warning.
Read the value before trusting it to identify a bay. client_ip() returns
remote_addr and nothing else. On the shipped Windows deploy IIS launches
waitress on loopback through HttpPlatformHandler, and deploy/windows/
web.config ships the X-Forwarded-For rewrite rule COMMENTED OUT - IIS sets
no such header on its own, so waitress's --trusted-proxy=127.0.0.1
--trusted-proxy-count=1 has nothing to rewrite REMOTE_ADDR from and leaves
it as the loopback peer. TRUSTED_PROXY_COUNT also defaults to 0, so
ProxyFix is not installed either. Every row and every warning then reads
127.0.0.1, and the un-migrated-bay list identifies no bay.
It is a real address only where the site enabled that rewrite rule (the
installer's -ClientIpSource direct), or set TRUSTED_PROXY_COUNT for a
front proxy that sets the header itself.
"""
return client_ip(request) or 'unknown'
@@ -142,11 +217,31 @@ def bootstrap_base_url():
return 'https://{0}{1}'.format(host, root)
def _base(baseurl=None):
resolved = baseurl if baseurl is not None else bootstrap_base_url()
return resolved.rstrip('/')
def bootstrap_script_url(secret, baseurl=None):
"""The token URL that serves the script: downloaded by the command, and
offered in the UI as a plain file a tech can save and run instead."""
base = (baseurl if baseurl is not None else bootstrap_base_url()).rstrip('/')
return '{0}{1}?t={2}'.format(base, SCRIPT_PATH, secret)
"""The UI's "Download .ps1" link: a plain file a tech can save and run.
Still the query-string form, because an <a download> cannot set a header.
The one-line command no longer uses this shape - see bootstrap_command.
It carries UI_LINK_ARG so the server can tell this click apart from a bay
running a pre-change command. Without that the admin's own Download link
fires the deprecation warning the rollout uses to FIND un-migrated bays,
and the count becomes noise the moment an admin mints a token.
"""
return '{0}{1}?{2}={3}&{4}={5}'.format(_base(baseurl), SCRIPT_PATH,
QUERY_TOKEN_ARG, secret,
UI_LINK_ARG, UI_LINK_VALUE)
def bootstrap_fetch_url(path, baseurl=None):
"""A token-free URL for the machine-to-machine fetches. The token rides in
BOOTSTRAP_TOKEN_HEADER so it never reaches an access log."""
return '{0}{1}'.format(_base(baseurl), path)
def bootstrap_command(secret, baseurl=None):
@@ -158,28 +253,145 @@ def bootstrap_command(secret, baseurl=None):
Single quotes inside and no $ anywhere, so the one line means the same
whether it is pasted into cmd.exe or into a PowerShell prompt. The download
and the run are wrapped in try/finally so the dropped file goes away even
and the run are wrapped in try/finally so the staged file goes away even
when the script exits non-zero or the download itself fails.
Staging is the part worth being careful about, so the line creates
COMMAND_STAGE_DIR with an admin-only DACL, clears any leftover file, and
refuses (exit 5) to run a staged script that some other principal owns -
which is what a pre-planted file looks like. Well-known SIDs throughout,
and the owner may also be the elevated caller, since a file created by an
administrator is owned by that account rather than by the group.
The hardening is also CHECKED (exit 6). icacls is a native command, so it
reports failure only in an exit code: no exception is raised, and
$ErrorActionPreference never applies to it. Unchecked, an icacls that
cannot write the DACL leaves the directory inheriting whatever the parent
grants and the line downloads into it anyway - the silent version of the
hole this staging move was made to close. The same check also covers the
New-Item before it, because icacls on a directory that was never created
exits non-zero too.
Read as `Get-Variable LASTEXITCODE -ValueOnly`, not `$LASTEXITCODE`: the
whole line must stay free of `$`, or a paste into an already-running
PowerShell expands it in the PARENT and the child is handed `-ne 0` with
the parent's last exit code baked in.
"""
return ('powershell -ExecutionPolicy Bypass -c "'
"try {{ Invoke-WebRequest -UseBasicParsing '{0}' -OutFile '{1}'; "
"& '{1}' }} finally {{ Remove-Item '{1}' -Force "
'-ErrorAction SilentlyContinue }}"'
).format(bootstrap_script_url(secret, baseurl),
COMMAND_SCRIPT_PATH)
return (
'powershell -ExecutionPolicy Bypass -c "'
"try {{ New-Item -ItemType Directory -Force -Path '{stagedir}' "
'| Out-Null; '
"icacls '{stagedir}' /inheritance:r /grant '*{admin}:(OI)(CI)F' "
"'*{system}:(OI)(CI)F' | Out-Null; "
'if ((Get-Variable LASTEXITCODE -ValueOnly) -ne 0) '
"{{ Write-Host 'Refusing to run: could not restrict "
"{stagedir} to administrators (icacls failed). Fix the directory "
"or delete it, then run the command again.' -ForegroundColor Red; "
'exit 6 }}; '
"Remove-Item '{path}' -Force -ErrorAction SilentlyContinue; "
"Invoke-WebRequest -UseBasicParsing '{url}' "
"-Headers @{{'{header}'='{secret}'}} -OutFile '{path}'; "
"if ((Get-Acl '{path}').GetOwner("
'[Security.Principal.SecurityIdentifier]).Value -notin '
"@('{admin}','{system}',"
'[Security.Principal.WindowsIdentity]::GetCurrent().User.Value)) '
"{{ Write-Host 'Refusing to run: the staged bootstrap script is not "
"owned by an administrator.' -ForegroundColor Red; exit 5 }}; "
"& '{path}' }} finally {{ Remove-Item '{path}' -Force "
'-ErrorAction SilentlyContinue }}"'
).format(stagedir=COMMAND_STAGE_DIR,
path=COMMAND_SCRIPT_PATH,
url=bootstrap_fetch_url(SCRIPT_PATH, baseurl),
header=BOOTSTRAP_TOKEN_HEADER,
secret=secret,
admin=ADMINISTRATORS_SID,
system=SYSTEM_SID)
def _hash_token(secret):
return hashlib.sha256(secret.encode('utf-8')).hexdigest()
def _resolve_token():
"""The usable token named by ?t=, else None.
def _presented_token():
"""(secret, source) for the token this request carries.
Header first, query string second. The query form is deprecated but still
the only one a browser download link can use.
SOURCE_NONE when nothing was presented, which is NOT the same as an empty
query token: reporting a bare GET as 'query' made it answer 400 naming the
header, telling an anonymous caller whether this site had turned the query
form off, and contradicting the 404-for-everything-unusable rule at the top
of this module.
"""
header = (request.headers.get(BOOTSTRAP_TOKEN_HEADER) or '').strip()
if header:
return header, SOURCE_HEADER
query = (request.args.get(QUERY_TOKEN_ARG) or '').strip()
if not query:
return '', SOURCE_NONE
if request.args.get(UI_LINK_ARG) == UI_LINK_VALUE:
return query, SOURCE_UI_LINK
return query, SOURCE_QUERY
def _is_query_source(source):
"""Both query-string forms, the bay's and the admin download link's. The
setting is about a token in an access log, and the admin link puts one
there just as surely as a pasted command does."""
return source in (SOURCE_QUERY, SOURCE_UI_LINK)
def _query_token_allowed():
"""Defaults to True: today's pasted commands and the UI download link both
use the query form, and turning that off mid-rollout strands a bay."""
return bool(Setting.get(QUERY_TOKEN_SETTING, True))
def _query_token_refused():
"""Answered before any lookup, so it confirms nothing about any token."""
return error_response(
ErrorCodes.VALIDATION_ERROR,
'Send the bootstrap token in the {0} header.'.format(
BOOTSTRAP_TOKEN_HEADER),
http_code=400)
def _note_token_source(source):
"""Audit suffix for a token that really resolved, and, for a query-string
token, one log line naming the caller and WHICH query path it was.
Only the bay warns. That warning is the rollout's un-migrated-bay counter
(see PER-PC-CREDENTIALS-ROLLOUT.md), so the admin's own Download .ps1 link
logs at info instead - otherwise every mint-and-test by an admin adds a
bay that does not exist.
Never the token itself, in either line: keeping it out of the logs is the
whole point of the header form.
"""
if source == SOURCE_UI_LINK:
current_app.logger.info(
'bootstrap script fetched by %s from the admin Download .ps1 '
'link; query-string token, but the admin browser path, NOT an '
'un-migrated bay', _caller_ip())
return ' (query-string token, admin download link)'
if source == SOURCE_QUERY:
current_app.logger.warning(
'bootstrap token presented in the query string by %s with no '
'admin download marker, so this is an un-migrated bay running a '
'command pasted before the header form; that form is deprecated '
'and slated for removal, re-issue the one-line install command',
_caller_ip())
return ' (query-string token, deprecated)'
return ''
def _resolve_token(secret):
"""The usable token with this secret, else None.
Unknown, malformed, expired, revoked and exhausted all resolve to None so
the caller can answer 404 for every one of them.
"""
secret = (request.args.get('t') or '').strip()
if not _TOKEN_PATTERN.match(secret):
return None
row = EndpointBootstrapToken.query.filter_by(
@@ -218,6 +430,7 @@ _SCRIPT_TEMPLATE = '''# ShopDB endpoint agent bootstrap
$ErrorActionPreference = "Stop"
$BootstrapPackageUrl = {packageurl}
$BootstrapPackageHeaders = @{{ {tokenheader} = {token} }}
$BootstrapServiceName = "ShopDbEndpointAgent"
$BootstrapInstallDir = "C:\\Program Files\\ShopDB\\EndpointAgent"
$BootstrapAgentSha256 = {agentsha}
@@ -308,7 +521,7 @@ try {{
New-Item -ItemType Directory -Force -Path $bootstrapTempDir | Out-Null
Write-Host "[1/4] Downloading the agent package ..."
Invoke-WebRequest -Uri $BootstrapPackageUrl -OutFile $bootstrapZipPath -UseBasicParsing
Invoke-WebRequest -Uri $BootstrapPackageUrl -Headers $BootstrapPackageHeaders -OutFile $bootstrapZipPath -UseBasicParsing
Write-Host "[2/4] Expanding and verifying ..."
Expand-Archive -Path $bootstrapZipPath -DestinationPath $bootstrapExtractDir -Force
@@ -394,11 +607,12 @@ exit 0
def build_bootstrap_script(secret, baseurl=None):
"""The PowerShell text served for one token, with the URL and hashes baked in."""
base = (baseurl if baseurl is not None else bootstrap_base_url()).rstrip('/')
base = _base(baseurl)
hashes = siteagent.staged_binary_hashes()
return _SCRIPT_TEMPLATE.format(
packageurl=json.dumps('{0}{1}?t={2}'.format(base, PACKAGE_PATH,
secret)),
packageurl=json.dumps(bootstrap_fetch_url(PACKAGE_PATH, base)),
tokenheader=json.dumps(BOOTSTRAP_TOKEN_HEADER),
token=json.dumps(secret),
agentsha=json.dumps(hashes.get('agent', '')),
traysha=json.dumps(hashes.get('tray', '')),
forcevariable=FORCE_VARIABLE,
@@ -500,7 +714,8 @@ def revoke_bootstrap_token(tokenid):
# ---------------------------------------------------------------------------
# Token-authenticated: no login, the token in ?t= is the whole authorization
# Token-authenticated: no login, the token is the whole authorization. It
# rides in the X-Endpoint-Bootstrap-Token header; ?t= is still accepted.
# ---------------------------------------------------------------------------
@endpoints_bp.route('/siteagent/bootstrap.ps1', methods=['GET'])
@@ -516,12 +731,15 @@ def serve_bootstrap_script():
"""
if _ratelimited():
return _rate_limit_response()
row = _resolve_token()
secret, source = _presented_token()
if _is_query_source(source) and not _query_token_allowed():
return _query_token_refused()
row = _resolve_token(secret)
if row is None:
return _not_found()
secret = (request.args.get('t') or '').strip()
_audit('requested', row, 'bootstrap script served')
_audit('requested', row,
'bootstrap script served' + _note_token_source(source))
db.session.commit()
return Response(build_bootstrap_script(secret), mimetype='text/plain')
@@ -536,7 +754,10 @@ def download_bootstrap_package():
"""
if _ratelimited():
return _rate_limit_response()
row = _resolve_token()
secret, source = _presented_token()
if _is_query_source(source) and not _query_token_allowed():
return _query_token_refused()
row = _resolve_token(secret)
if row is None:
return _not_found()
@@ -544,8 +765,8 @@ def download_bootstrap_package():
row.usecount = (row.usecount or 0) + 1
row.lastusedat = now
row.lastusedip = _caller_ip()[:45]
_audit('downloaded', row, 'bootstrap package, use {0} of {1}'.format(
row.usecount, row.maxuses))
_audit('downloaded', row, 'bootstrap package, use {0} of {1}{2}'.format(
row.usecount, row.maxuses, _note_token_source(source)))
db.session.commit()
buffer, warning = siteagent.build_keyless_package(bootstrap_base_url())