SWWAF_WAF_BODY_LIMIT (default off, at most 1G) has the Core Rule Set read
form data and multipart up to the limit, the rest streaming on, and JSON
and XML no larger than it, with text/json and the application and text
types ending in +json or +xml. The part read is held for the app. A size
or time limit met while reading ends the request. Content-Encoding is
refused again on these kinds. A body Coraza cannot parse, or a multipart
body failing its strict checks, adds 5, as does a multipart body the limit
cuts in a part's headers before a colon or a line feed. Coraza is built
with no_fs_access, so writes no file. Rule 900300 moves to phase 2.
Judgement call: Content-Encoding is refused on a JSON or XML body too
large to read, as SPEC.md allows.
Model: opus-5-5
Coraza v3.8.1 runs the Core Rule Set 4.25.0 (coraza-coreruleset v4.25.0)
after the rule files, with the six changes and the default
SWWAF_WAF_DISABLED_RULES that SPEC.md gives; no body, no response. The
parameter names in the third and fourth changes are matched in any case,
as Coraza does. SWWAF_WAF_DISABLED_RULES refuses 900000 to 900999,
smallwebwaf's own rules. A request with more than 1000 query parameters
adds 5 (rule 900300). In block mode a match is refused with 403, an
offence counted toward the error burst; in detect mode it is let
through. Both log waf_rule_ids, waf_score and duration_waf, raise
waf_block, and count smallwebwaf_waf_matches_total.
Judgement call: waf_block is raised in block mode too.
Deviation: no engine-error path; with no body read, Coraza cannot fail.
Model: opus-5-5
Twelve tests in internal/proxy needed the GeoJS stand-in to be asked or
to answer within the default SWWAF_LOOKUP_TIMEOUT of one second on the
real clock. A hold-up of the test process past it abandoned the request
to the stand-in, or left the client unknown. Each now sets
SWWAF_LOOKUP_TIMEOUT to an hour, and the comment on startGeoJS asks the
same of later tests.
Judgement call: set in each test, not as a default in newProxy, where it
would change two tests that rely on the default second.
Model: opus-5-5
SWWAF_TRAP_PATHS: a request whose path, as a path rule sees it, is one
of them is a clear sign of attack, banned as a ban rule's match is; the
ban's notes give its trap_path. Checked after the rate limits, before the
rule files.
SWWAF_ERROR_BURST_THRESHOLD (default 30, or off): more refusals in a
minute after a block or ban rule or a trap path, or for a missing or
wrong token, ban the client as a broken limit does. Counted in
clients.json's minute_refusals; limit_hit error_burst, notes kind
refusals.
A token refusal is now the offence token_refused, and
smallwebwaf_offences_total counts every kind the history does.
Judgement call: the threshold is not lowered by a client's limit percentage.
Model: opus-5-5
SWWAF_CROWDSEC_LAPI_URL and SWWAF_CROWDSEC_LAPI_KEY name an engine whose
decision list, <url>/v1/decisions, is fetched every minute with the key in
X-Api-Key, following no redirect, and kept as a blocklist is: used while a
fetch fails, and across restarts through reputation.json. Ban decisions on an
Ip or a Range end at the fetch time plus their duration. A listed client's
request is refused and bans its netblock with the cause crowdsec until the
decision ends; bans.json, ban notes and metrics take the cause.
Judgement call: fetched every minute, not a setting.
Judgement call: a crowdsec ban never lengthens a limit ban.
Judgement call: a lifted crowdsec ban is remade while its decision lasts.
Model: opus-5-5
A ban's notes, in bans.json and in its alert, gain `reputation`: each
blocklist, DNSBL zone or AbuseIPDB that listed the client when the ban
was made, as its `source`, named and ordered as in the request log's
`reputation`, with AbuseIPDB's `score`. It is left out when none did.
README.md shows it in a bans.json example.
Notes now hold a list, so bans can no longer be compared with ==: the
tests compare them with reflect.DeepEqual.
Judgement call: the score is a pointer, so a score of 0, a hit while
SWWAF_ABUSEIPDB_MIN_SCORE is 0, is still written.
Model: opus-5-5
The IPv6 group that is one client, the size of the table of clients and
the level of the process's own lines become settings. clientGroup reads
the group length from them, so limits, bans, history, lookups, AbuseIPDB
scores and per-client anomaly counters all follow it; ratelimit.New
takes the table size; the process logger takes the level once the
settings are read, and request lines, written apart from it, are never
held back.
Judgement call: SWWAF_IPV6_GROUP_PREFIX accepts 32 to 128, the issue's example range.
Model: opus-5-5
With SWWAF_ABUSEIPDB_KEY set, a client whose history counts an offence
(a broken limit, a ban rule's match or a block rule's refusal, counted
by kind) is checked in the background, at most
SWWAF_ABUSEIPDB_DAILY_BUDGET checks a day, the count kept in
reputation.json. A client, an IPv4 address or an IPv6 /64, is checked by
the address it sent from, and its score serves all its addresses. A
score at or over SWWAF_ABUSEIPDB_MIN_SCORE is a hit for
SWWAF_REPUTATION_ACTION, logged as abuseipdb and alerted with its score.
A failure or the used-up budget gives no score and raises
source_failure. The key goes only in the Key header.
Judgement call: the budget's day is UTC; AbuseIPDB documents no reset time.
Judgement call: each check sent spends budget; a minute's pause after a failure.
Model: opus-5-5
Zones in SWWAF_DNSBL_ZONES are asked about each client in the background,
through SWWAF_DNSBL_RESOLVER or the host's resolver; no request waits.
Verdicts last SWWAF_REPUTATION_CACHE_TTL, kept in reputation.json.
SWWAF_REPUTATION_ACTION (limit:25) denies, limits or logs a listed
client; each zone listing it raises reputation_hit. A failed query gives
no verdict, raises source_failure, and pauses the zone a minute. A
zone's key, its first label under dq.spamhaus.net, is masked everywhere
but reputation.json. Zones compare without regard to case.
Judgement call: answers in 127.255.255.0/24 or outside 127.0.0.0/8 are failures.
Judgement call: the minute's pause after a failure; at most 1,000 queries at once.
Judgement call: one zone given with two keys stops the start as listed twice.
Rule suppressed: paralleltest on the DNSBL tests (Go's resolver shares state across synctest bubbles), funlen on the test of every logged setting.
Model: opus-5-5
SWWAF_BLOCKLIST_URLS names lists of addresses and netblocks, fetched every
SWWAF_BLOCKLIST_REFRESH (24h, never under 1h); an IPv4-mapped line stands
for its IPv4 address or netblock. reputation.json keeps each list's last
try, failed or not, even one cut off by a stop, which a restart waits on
as a running instance does, and its last good copy, whole, used while a
fetch fails. SWWAF_BLOCKLIST_ACTION denies, limits or only logs a listed
client; the log line names the lists, each raises reputation_hit, and a
failed fetch raises source_failure. SWWAF_ASN_LIMIT_PERCENT_URL is fetched
the same way and counts as SWWAF_ASN_LIMIT_PERCENT does, the lower winning.
Judgement call: a failed fetch is retried after the refresh, not sooner.
Not done: ban notes do not name the lists yet.
Model: opus-5-5
SWWAF_ANOMALY_CLIENT_*, _NET_*, _ASN_*, _TOTAL_* and SWWAF_WATCH_* with
SWWAF_WATCH_NETS: requests and bytes per minute and per hour, each off by
default; with all off, nothing is counted. Otherwise every request but the
health check is counted, allow-listed and exempt ones included; a count
over its threshold raises an anomaly alert, with a cooldown per scope. At
most 20,000 counters, kept in alerts.json. A per-AS-number threshold with
lookups off, or a malformed SWWAF_WATCH_NETS, stops the start. A cooldown
that has run out is dropped as the hour ends, whatever it held back; the
hour's summary gives its repeats.
Judgement call: refused requests are counted too.
Judgement call: per-client counters are kept in alerts.json, which SPEC.md does not list.
Judgement call: a request counts for an AS number only if the lookup answered before it ended.
Model: opus-5-5
SWWAF_ASN_LIMIT_PERCENT and SWWAF_COUNTRY_LIMIT_PERCENT give the clients
of the AS numbers and countries they list that percentage of every rate
and byte limit, rounded down; SWWAF_ASN_BYTES_PERCENT and
SWWAF_COUNTRY_BYTES_PERCENT take its place for the byte limits of those
they list; SWWAF_UNKNOWN_LIMIT_PERCENT (100) covers clients without a
country. The lowest applies. While one lowers a limit, a request waits
for its client's lookup, and SWWAF_LOOKUP_SOURCE=off stops the start. Log
lines give limit_percent and bytes_percent with their settings; ban
notes, and so alerts, give the broken limit's.
Judgement call: a client without a country is unknown, whatever its AS number.
Judgement call: bytes_percent and its setting are log fields SPEC does not name.
Rule suppressed: funlen on FromEnvironment, one line per setting.
Model: opus-5-5
SWWAF_BYTES_LIMIT_PER_MINUTE, _PER_HOUR and _PER_DAY (10G, 20G, 50G)
and SWWAF_BYTES_COUNT (both). A request's bytes are counted once its
answer has ended, for a request passed to the app that the rate limits
count; what a WebSocket carries each way, once it closes. Bytes over a
limit ban the client as a broken rate limit does, and cut nothing
short. clients.json keeps the byte buckets, the log line's counts carry
the byte totals, ban notes say what the limit is on, and the limit hits
metric is labelled by kind.
Judgement call: limit_hit names a byte window minute_bytes, hour_bytes
or day_bytes, as counts names the byte totals.
Judgement call: in observe mode, the bytes of a request enforce mode
would have refused are not counted.
Model: opus-5-5
script/tidy runs `go mod tidy` in a new tidy stage of the Dockerfile,
on the test phase's Go image, and writes go.mod and go.sum back into the
working tree with `docker build --output`. The test phase now runs
`go mod tidy -diff` before the tests and fails naming `make tidy`; it
comes first because a missing go.sum line otherwise fails the tests with
Go's own message.
Judgement call: script/tidy's build has no tag, unlike the other builds
in script/: it makes no image.
Model: opus-5-5
SWWAF_LOOKUP_SOURCE=file looks every client up in the file
SWWAF_LOOKUP_DB_PATH names, without GeoJS. file without the path, the
path with another source, or a file that cannot be read stops the start.
The file is read whole into memory, so overwriting it in place cannot
disturb a lookup, and read again 2 seconds after its last change; a
replacement that cannot be read is logged, counted and sent as a
file_error alert, and the old one stays in use. Metrics give when it
was read and the failed reads. Tests write their databases through
internal/lookup/lookuptest.
Deviation: go.mod and go.sum written by hand; go runs only through make.
Judgement call: the 2-second wait, as the rule files have.
Model: opus-5-5
GeoJS's geo.json is asked about every new visitor unless
SWWAF_LOOKUP_SOURCE is off. A request waits for its client's first
answer only while a country list or SWWAF_ADD_LOOKUP_HEADERS needs it;
otherwise the answer reaches the client's history and ban notes when it
comes. The AS number and name go beside the country in the request log,
history, ban notes, alerts and lookups.json, with metrics by AS number;
64512 counts as unknown. A client's own X-Client-ASN and
X-Client-Country never reach the app, whatever the setting says, and
make example-app sends no address to GeoJS.
Judgement call: AS numbers are written AS64496, as SPEC's settings write them.
Judgement call: SWWAF_LOOKUP_TIMEOUT is added, default 1s, and cannot be off.
Model: opus-5-5
Process log lines carry instance, as request lines do; the instance name
is read before the other settings, so the line saying a setting is
invalid carries it too. Every metric, Go's and the process's included,
carries the label instance, set once on the registry. README.md says so,
and that Prometheus keeps it as exported_instance unless the scrape sets
honor_labels. An instance name that is not valid UTF-8 stops the start,
as the metrics library panics on such a label.
Tests that read metrics expect the label; one helper replaces the alert
tests' loops that wait for them.
Judgement call: the label is named instance, as in the log lines and
alerts, although Prometheus gives each target a label of that name.
Model: opus-5-5
Each alert is posted as a message to the Slack incoming webhook
SWWAF_ALERT_SLACK_WEBHOOK_URL names, and published to the ntfy topic
SWWAF_ALERT_NTFY_URL names, with SWWAF_ALERT_NTFY_TOKEN as a bearer
token and a priority and tag by event. The cooldown and the hourly
limit stay shared; past them, each destination has its own bounded
queue and backoff, and its own sent, failed and dropped counts.
alerts.json keeps the alerts waiting by destination; one whose waiting
is still a list stops the start, saying what to change. A control
character in the ntfy token, or in the instance name ntfy is sent,
stops the start.
Judgement call: messages also give the detail's file, source, error and mode.
Judgement call: alerts_suppressed_total is the same for every destination.
Model: opus-5-5
SWWAF_ALERT_WEBHOOK_URL gets one JSON POST per alert, in SPEC.md's
schema, with SWWAF_ALERT_WEBHOOK_HEADERS: ban and permanent_ban, with
the ban's notes, in observe mode too, marked mode observe and worked
out only when the alert would be sent; source_failure for GeoJS;
file_error for a rule or state file with an error. SWWAF_ALERT_EVENTS
chooses; SWWAF_ALERT_COOLDOWN holds back repeats by netblock, file or
source; past SWWAF_ALERT_MAX_PER_HOUR the hour ends in one summary. A
bounded queue, retried with backoff, holds up no request; a 4xx other
than 408 and 429 gives the alert up. alerts.json keeps the queue, the
cooldowns and the hour. Nothing shows the URL's path or query.
Judgement call: the summary's event is summary, which SPEC.md omits.
Judgement call: an admin's ban raises no alert.
Model: opus-5-5
SWWAF_ADMIN_TOKEN, or its _FILE form, opens GET and POST
/_smallwebwaf/bans, DELETE /_smallwebwaf/bans/<client> and GET
/_smallwebwaf/clients/<ip>. Unset, they answer 404; a missing or wrong
token gets 401, in observe mode too. They go through every check, as
the metrics do. POST takes a netblock, not IPv4-mapped and without a
zone, or a client's address, a duration or permanent, and a reason, and
makes an admin ban even while another lasts. DELETE lifts every active
ban covering the address, kept and marked lifted. Bans come back as
bans.json entries; a client as clients.json holds it, with its bans.
Judgement call: answers leave out bans.json's version field.
Judgement call: DELETE takes an address, not a netblock.
Rule suppressed: gosec G304 on a test reading bans.json.
Model: opus-5-5
Every setting X may instead be given as a file that X_FILE names, read
once at start: its contents, less one trailing newline, are the value,
checked as X would be. X and X_FILE both set, or a file that cannot be
read, stops the start with a message naming the variable. The logged
settings name the file, and mask a token read from one.
SWWAF_LOG_REMOTE_TLS_CA_FILE, whose value is a file already, has no
_FILE form. The health check reads only SWWAF_LISTEN_ADDR and
SWWAF_UPSTREAM_URL, so no other setting or file can fail it.
Judgement call: an invalid value read from a file is named as X, not X_FILE.
Rule suppressed: gosec G304 on reading the named file, as for the CA file.
Model: opus-5-5
A bans.json entry without a cause gets the cause admin, written back so.
Bans whose cause is admin are never dropped and do not count toward
SWWAF_MAX_BANS, so setting a ban's cause to admin keeps it. Bans
smallwebwaf makes get a reason: the limit broken or the rule matched. A
lifted ban refuses nothing, is kept, and makes no later ban longer.
smallwebwaf_bans_made_total counts admin bans an edit adds while running;
earlier_bans counts admin in place of without_cause.
Judgement call: lifted lifts at once, whatever time it gives.
Judgement call: a lifted ban still counts in earlier_bans.
Known gap: a ban dropped from behind an admin's ban on its netblock leaves that netblock's later earlier_bans.
Model: opus-5-5
With SWWAF_LOG_REMOTE_URL set (syslog+udp, syslog+tcp or syslog+tls),
every line on stdout is also sent as the message of an RFC 5424 record,
octet-counted over TCP and TLS, from a bounded buffer that drops its
oldest line when full, so a slow or unreachable server holds up nothing.
Failed connections are retried with backoff; lines sent, dropped and
waiting are metrics. At a stop the lines still waiting get at most two
seconds. SWWAF_LOG_REMOTE_APP_NAME defaults to SWWAF_INSTANCE_NAME; while
sending, an app name RFC 5424 does not allow stops the start. Standard
library only: log/syslog writes only the older format.
Model: opus-5-5
Every *.rules file in SWWAF_RULES_DIR not named with a leading dot is
read at start, and again 2 seconds after the directory's last change.
Each request is checked against the rules after the rate limits: log
notes a match, block refuses with 403, ban refuses and bans the netblock
for SWWAF_ATTACK_BAN_DURATION, made permanent by its next request or
attack. path, query and uri are matched as the request line sent them;
header:Host and header:Transfer-Encoding are refused. Bans gain a cause.
The image ships 00-default.rules.
Judgement call: a header sent twice is matched with its values joined
by ", ".
Judgement call: SWWAF_MAX_BAN_DURATION does not cap a ban for an attack.
Not in this unit: offences for rule matches, with the error burst.
Model: opus-5-5
A request is neither counted nor refused by the request rate limits
when its path as sent, the path the app receives, not percent-decoded,
starts with one of the comma-separated prefixes in
SWWAF_RATE_LIMIT_EXEMPT_PATHS, so /%61ssets/x is not under /assets/. A
request whose decoded path contains .. or a backslash, or whose path as
sent holds an encoded slash, is never exempt, since an app may act on
it as a path outside every prefix, such as /assets/..%2Flogin as
/login. The static lists, bans and the country lists still apply, and
its log line has no counts. The setting is empty by default, and a
prefix that does not start with / stops the start. README.md documents
it.
Model: opus-5-5
Each request log line now has the fields "Request log" in SPEC.md lists
whose features are built: instance (SWWAF_INSTANCE_NAME), scheme,
request_id (a trusted proxy's X-Request-ID or a new one, sent on to the
app), forwarded_for, client_group, content_type, content_length, the
headers SWWAF_LOG_REQUEST_HEADERS names, has_authorization, has_cookie,
websocket, response_content_type, cache_control, location, counts and
the timings. Authorization, Cookie and Set-Cookie values are never
logged. An entry of SWWAF_LOG_REQUEST_HEADERS that is not a header name,
or is Host or Transfer-Encoding, stops the start.
Deviation: counts has request totals only.
Deviation: SWWAF_INSTANCE_NAME is on request lines only.
Model: opus-5-5
smallwebwaf watches SWWAF_STATE_DIR with fsnotify and takes in a saved
edit of a state file in place of what it held. It knows its own writes
by the SHA-256 of what it last read or wrote; each write first takes in
an edit made since. An edit that does not parse is renamed to
<name>.bad at the next write. Each edit taken in or set aside is logged
and counted. Every ban on a netblock is checked, and the next ban is
worked out from the one that ended last. README.md says how to add and
lift a ban.
Judgement call: a broken edit is set aside at the next write, since an
editor's file can be read half written.
Model: opus-5-5
SWWAF_MODE (default enforce) takes enforce or observe. In observe mode a
request that SWWAF_DENY_NETS, a ban, the country lists or a rate limit
would refuse is passed to the app, and its log line names that refusal
in would_action. The size and time limits and the 401 still apply. A
broken limit makes no ban; bans read from bans.json are kept but refuse
nothing, and Ledger.Find reads them without counting a refusal in their
notes.
Judgement call: in observe mode a broken limit does not reset the
client's counters, since the reset comes with the ban.
Judgement call: a request a ban would refuse keeps ban_expires.
Model: opus-5-5
GET /_smallwebwaf/metrics answers in the Prometheus text format for a
request carrying SWWAF_METRICS_TOKEN, 401 without it and 404 while it is
unset. Every request under /_smallwebwaf/ but the health check now goes
through the checks and is answered where it would be forwarded, 404 for
any path but the metrics, so none reaches the app. In the client's
history a 401 counts as refused, the metrics and the 404s as neither.
SWWAF_METRICS_TOP_N bounds the series by country, the rest counted as
other.
Deviation: go.mod and go.sum written by hand, as go runs only through
make.
Deviation: no metrics yet for state files read again after an edit or
edits set aside; that work is not merged.
Model: opus-5-5
The tests that have GeoJS asked now run in a synctest bubble, so a wait
lasts exactly as long as it should however slowly the test process runs:
a new client's wait is checked to be exactly one second, and its next
request exactly no wait. A request waiting on the network would stop the
bubble's clock, so the stand-in for GeoJS now answers in place of the
network, through a transport that a test-only file lets the tests set.
The test of the failure log uses an abandoned request instead of a
closed port.
Model: opus-5-5
smallwebwaf now copies its state to bans.json, clients.json and
lookups.json in SWWAF_STATE_DIR, as "Persistent state" in SPEC.md
describes, and reads them back at start, so a restart lifts no ban and
gives no client a fresh allowance. Each client gains a history, and a
ban's notes count the netblock's requests. bans.json is written
SWWAF_STATE_WRITE_DELAY after a ban, and every file every
SWWAF_STATE_COUNTER_INTERVAL and at the stop. A ban read back is masked
to its netblock and refuses every client in it. A file that does not
parse, an unknown version, an entry without a field it needs, or an
unwritable directory stops the start.
Deviation: no AS number or name, and no ban cause, reason or lifting yet.
Model: opus-5-5
A request over a rate limit is refused with SWWAF_BAN_RESPONSE and bans
the client's netblock: an hour at first, three times the last ban when
broken again within a day of its end, permanent past seven days. The
ban ledger in internal/bans is checked after the static lists and
before the lookup, and the requests it refuses are not counted. A ban
resets the client's counters and carries notes holding the request
that broke the limit, as SPEC.md now says. At most SWWAF_MAX_BANS are
held. SWWAF_BAN_RESPONSE also answers SWWAF_DENY_NETS and the country
lists.
Judgement call: the six ban settings cannot be off.
Judgement call: a permanent ban's ban_expires is "permanent".
Model: opus-5-5
SWWAF_CLIENT_REQUEST_HEADER_MAX_BYTES (default 32K) and
SWWAF_CLIENT_IDLE_TIMEOUT (default 120s) replace the two values the
proxy fixed. The idle time is read like the other durations, and can
be off.
Go's server reads 4K past the header limit it is given before it
refuses, so it is still given the setting less 4K. The header size
must be more than 4K and cannot be off; any other value stops the
start with a message that does not offer off.
SPEC.md and README.md say so. README.md lists both settings, no longer
calls them fixed, and names them as built.
Model: opus-5-5
The vendored files are fetched from sneak/prompts commit dd4027b. This
repository's own entries come after the canonical content, at the end of
each file: /bin in .dockerignore, the Go lines of .gitignore and [*.go]
in .editorconfig; the test-support deny list has no entries of its own.
The lint phase moves to golangci-lint v2.14.0. The build stage now takes
the version from git describe on the .git the build context carries,
unless VERSION is passed, and fails when .git is present but no version
comes out. The test phase drops -count=1, which the policy says it does
not need, and keeps its tmpfs build cache. One test calls Header.Get
with X-Real-IP, as canonicalheader asks.
Model: opus-5-5
Adds SWWAF_ALLOW_NETS, SWWAF_RATE_LIMIT_EXEMPT_NETS and SWWAF_DENY_NETS,
read like SWWAF_TRUSTED_PROXIES and empty by default, and checked against
the client's own address before its country is looked up. A client in
SWWAF_ALLOW_NETS skips the country lists and the rate limits and is not
looked up. One in SWWAF_DENY_NETS is refused with 403, logged as denied
and not counted. One in SWWAF_RATE_LIMIT_EXEMPT_NETS is neither counted
nor refused by the rate limits. SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES now
refuses a private, loopback or link-local client unless SWWAF_ALLOW_NETS
lists it.
Judgement call: an address in both SWWAF_ALLOW_NETS and SWWAF_DENY_NETS is let through.
Judgement call: the size and time limits still apply to SWWAF_ALLOW_NETS.
Model: opus-5-5
useradd --system warns when the uid it is given is above SYS_UID_MAX,
999 on Ubuntu; --key raises that limit for this one call, so the uid
stays 65532.
minsysusers, which runit's install runs to create its _runit-log user,
prints a Perl warning because runit's sysusers line leaves out the
shell. It reads /etc/sysusers.d/runit.conf in place of runit's file, so
the image writes a copy of that line there, naming the shell minsysusers
gives anyway; the user it creates is unchanged.
The runsvinit warning stays, since it needs a change to runsvinit: its
reaper and its own wait on runsvdir race for the same exited process.
Model: opus-5-5
The test phase spends most of its time compiling with the race
detector from an empty build cache; then come writing the test image
and the internal/proxy tests.
- Go's build cache is on a tmpfs in the test phase, so its 137 MB are
no longer written into the test image.
- TestUpgradedConnectionOutlastsTheTimeouts waits until just past
shortTimeout after the answer to the upgrade was read, by when every
timeout has started, rather than 7.5 s, so it ends with the other
timing tests.
- shortTimeout is written as waitLimit / 2, as its comment says it is.
Model: opus-5-5
The Dockerfile's last stage is now the image of "Deployment" in SPEC.md:
Ubuntu 26.04 with ca-certificates, nix-bin and runit from a dated
snapshot whose InRelease files are checked by hash, nixpkgs from its
release file checked by SHA-256, runsvinit built at a fixed commit, and
smallwebwaf as a runit service. smallwebwaf answers
/_smallwebwaf/healthz, and `smallwebwaf healthcheck`, which takes no
further argument, is the image's HEALTHCHECK. script/example-app builds
an app on the image and checks it end to end.
The Nix profile comes last on the PATH: first, busybox from nixpkgs
replaced runit's own runsvdir and sv. SPEC.md is corrected to match
what was built.
Model: opus-5-5
SWWAF_DENIED_COUNTRIES and SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES refuse a
request with 403 before its body is read or rate-limited, logged as
country_denied. internal/lookup asks GeoJS only while a list is set, 200
clients per request, one at a time, keeping answers 7 days. Failures, a
redirect or an answer leaving an address out included, are logged without
addresses; GeoJS is then left alone a second, doubling to five minutes.
Private, loopback and link-local clients are never sent.
Deviation, per the issue: no SWWAF_LOOKUP_SOURCE or SWWAF_LOOKUP_TIMEOUT; 403, not SWWAF_BAN_RESPONSE.
Deviation: GeoJS's country endpoint, not geo.json.
Judgement call: an IPv6 /64 is asked about by its first address; at most 10,000 clients wait.
Judgement call: config.go lists the ISO 3166-1 codes; no widely used library holds them.
Model: opus-5-5
smallwebwaf starts each timeout as the request arrives, before the step
a test needs first: the upgrade answered, the app's buffers full, the
first part of an answer passed on. A hold-up of the test process longer
than the 300 ms timeout ran it out before that step. No test can make
that step come first, and in the "waiting on the app" cases it cannot
see which side smallwebwaf was waiting on, so the timeout is now 5 s,
the hold-up wantTimedOut already allows. The timeout tests set when it
must not run out goes from 10 s to 1 m to stay clear of it. The upgrade
test waits 7.5 s past the upgrade.
Judgement call: one shared value; the proxy tests take about 8 s, not 2 s.
Model: opus-5-5
TestAnswers502WhenTheAppCannotBeReached closed a listener and pointed
smallwebwaf at its port, which another test running in parallel could
open in between, so the test sometimes got that server's answer instead
of a 502. It now uses 127.0.0.1:1: nothing listens there, and a test
listening on port 0 is always given a port from 32768 up, so no test
can take it. No other test reuses a closed listener's port.
Model: opus-5-5
In the cases where the app reads and the client stops sending halfway,
smallwebwaf waits on the client only once it has passed the first bytes
of the body to the app. A test process held up for the whole 300 ms
timeout before then rightly gets 504, as SPEC.md asks, so the test was
wrong to expect 408 every time.
The app in those cases now records whether it received any of the body.
Once the app has finished with the request, the case expects 408 and its
log line if it did, and 504 and its log line if not.
Model: opus-5-5
Each client, one IPv4 address or one IPv6 /64, is counted in two buckets
per window, the earlier weighted by how much of it the window covers; at
most 20,000 clients are kept, least recently seen dropped first. A
request over SWWAF_RATE_LIMIT_PER_MINUTE, _HOUR or _DAY (1000, 10000,
50000, or off) gets 429 before reaching the app. Refused requests count,
413s included. A clock set back over a second behind a bucket's start
restarts that window. The log line gains limit_hit and the action
rate_limited.
Deviation from SPEC.md, per the issue: the 20,000 bound and /64 are fixed.
Judgement call: golang-lru/v2 holds the table; httprate does not count refused requests.
Deviation: go.mod and go.sum hand-written; no make target tidies them.
Model: opus-5-5
The header size and the idle time stay fixed at their defaults through
milestone 2 and become settings in the first stage after it, which also
brings SWWAF_ALLOW_NETS and with it the allow-only country list's refusal
of private addresses. While a request body is on its way, a request
timeout answers 408 or 504 by the side smallwebwaf was waiting on, as
milestone 1's code does. The README says the state files, the GeoJS
answers among them, come in milestone 3 or later, and points at the
build order.
Model: opus-5-5
ca-certificates, nix-bin and runit come from a dated Ubuntu snapshot no
older than the pinned Ubuntu image. The Dockerfile names the SHA-256 hash
of each snapshot InRelease file apt uses, and the build checks them before
apt-get install, so every package is checked against hashed files. That
install uses the Go image's CA certificate file. ca-certificates is
installed by name. The image writes build-users-group = to
/etc/nix/nix.conf so root can build without a daemon. nixpkgs comes from
its release file on releases.nixos.org, checked by SHA-256, and takes about
500 MiB of disk. runsvinit is archived upstream and is built at a fixed
commit with a go.mod written for the build. The example run scripts put
their code in a main function.
Model: opus-5-5
SWWAF_LISTEN_ADDR may set another port: the health check takes its port
from it, and traefik's port label must name the same one. Its address
part stays empty (:9000), so smallwebwaf keeps listening on every
address, where traefik and the health check on 127.0.0.1 both reach it.
A token file is made on the host owned by uid 65532 with mode 0400 and
its directory mounted read-only; through upaas, that directory is one of
the app's volume mounts. The run script of smallwebwaf makes the state
directory and every file in it belong to the smallwebwaf user.
Model: opus-5-5
Milestone 1, the repo's first code. smallwebwaf passes each request to the app and the answer back unchanged, streaming bodies and WebSocket upgrades, within four timeouts (client and app, request and response) and two size limits, and writes one JSON line per request to stdout. Every setting has an SWWAF_ name and a default, and an invalid value stops the start. The repo gets the standard layout: script/ entrypoints, make targets that call them, a Dockerfile that runs the checks, and the Gitea workflow.
Disclosure: SPEC.md changed. Go's server reads the request line and headers before smallwebwaf sees the request, so slow headers are closed without an answer, and neither slow nor oversized headers get a log line.
Disclosure: standard library only.
Model: opus-5-5
clawbot
sneak