Leave SWWAF_RATE_LIMIT_EXEMPT_PATHS out of the request rate limits (closes #77)

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

README.md

@@ -13,20 +13,25 @@ JSON log line for every request.
 
 Status: the first two milestones are built
 (https://git.eeqj.de/sneak/smallwebwaf/issues/13 and
-https://git.eeqj.de/sneak/smallwebwaf/issues/14), and so are five parts of
-milestone 3: the static lists, the bans that broken rate limits lead to and the
-JSON state files, which come next in the build order, and the metrics endpoint
-and the header size and the idle time as settings, which come last in it.
-`smallwebwaf` passes each request to the app and the app's answer back,
+https://git.eeqj.de/sneak/smallwebwaf/issues/14), and so are eight parts of
+milestone 3: the static lists, the bans that broken rate limits lead to, the
+JSON state files with your edits taken in while it runs and the paths the rate
+limits do not count, which come next in the build order, `observe` mode and the
+rest of the request log's fields, which come a little later, and the metrics
+endpoint and the header size and the idle time as settings, which come last in
+it. `smallwebwaf` passes each request to the app and the app's answer back,
 unchanged, within its timeouts and size limits, works out each client's address,
-bans a client that sends too many requests, refuses a client that comes from a
-country you refuse or from a network you refuse, lets the networks you choose
-through, keeps its bans, each client's counters and history, and GeoJS's answers
-in JSON files across restarts, writes a JSON log line for every request, and
-serves Prometheus metrics to a scraper that holds the metrics token. It comes as
-the image the app's own image is built on. The rest of the design comes after
-that, in the order of the build order in [`SPEC.md`](SPEC.md). The survey of
-existing tools that led to the design is in [`EVALUATION.md`](EVALUATION.md).
+bans a client that sends too many requests, not counting those for the paths you
+choose, refuses a client that comes from a country you refuse or from a network
+you refuse, lets the networks you choose through, keeps its bans, each client's
+counters and history, and GeoJS's answers in JSON files across restarts, takes
+in your edits of those files while it runs, writes a JSON log line for every
+request, serves Prometheus metrics to a scraper that holds the metrics token,
+and in `observe` mode passes on the requests it would refuse, logging what it
+would have done with them. It comes as the image the app's own image is built
+on. The rest of the design comes after that, in the order of the build order in
+[`SPEC.md`](SPEC.md). The survey of existing tools that led to the design is in
+[`EVALUATION.md`](EVALUATION.md).
 
 ## Getting started
 
@@ -63,7 +68,9 @@ in `bin/state` unless `SWWAF_STATE_DIR` is set.
   address outside `SWWAF_TRUSTED_PROXIES` is the client; if every address in it
   is inside, the leftmost is, and with no header the peer is. The app sees what
   it would see from traefik directly: the same `Host`, the same
-  `X-Forwarded-Proto`, and `X-Forwarded-For` with the peer added at the end.
+  `X-Forwarded-Proto`, and `X-Forwarded-For` with the peer added at the end. It
+  also gets the request's id in `X-Request-ID`, the same id as in the request's
+  log line (see `request_id` in "Request log" below).
 - Enforces the timeouts and the size limits below. A limit passed before the
   response has started gets `smallwebwaf`'s own answer: `408` for a client too
   slow to send its request, `413` for a request body that is too large, `504`
@@ -77,12 +84,15 @@ in `bin/state` unless `SWWAF_STATE_DIR` is set.
 - Counts each client's requests over a minute, an hour and a day. A request that
   takes the client over one of the rate limits below is refused with
   `SWWAF_BAN_RESPONSE`, `403` by default, before anything reaches the app, and
-  bans the client. A client is one IPv4 address, or one IPv6 /64, since one
-  abuser usually holds a whole /64. Each window is counted in two fixed buckets,
-  the earlier one weighted by how much of it the window still covers. At most
-  20,000 clients are kept, the least recently seen dropped first, with their
-  history, and a restart gives no client a fresh allowance (see "State files"
-  below).
+  bans the client. A request whose path starts with one of
+  `SWWAF_RATE_LIMIT_EXEMPT_PATHS`, as that setting below describes, is neither
+  counted nor refused by the rate limits; the static lists, bans and the country
+  lists still apply to it. A client is one IPv4 address, or one IPv6 /64, since
+  one abuser usually holds a whole /64. Each window is counted in two fixed
+  buckets, the earlier one weighted by how much of it the window still covers.
+  At most 20,000 clients are kept, the least recently seen dropped first, with
+  their history, and a restart gives no client a fresh allowance (see "State
+  files" below).
 - Bans a client that breaks a rate limit, as "Bans" in [`SPEC.md`](SPEC.md)
   describes: the first ban lasts an hour, and a limit broken again within a day
   of a ban ending bans for three times as long as that ban, so 1, 3, 9, 27 and
@@ -98,9 +108,9 @@ in `bin/state` unless `SWWAF_STATE_DIR` is set.
   seen, how many of them the ban has refused, and how many bans the netblock had
   before. At most `SWWAF_MAX_BANS` bans are kept, past, active and permanent;
   past that, the earliest ban of the netblock that has gone longest without a
-  request is dropped first. `bans.json` shows the bans and their notes, and a
-  restart lifts none (see "State files" below); lifting a ban by editing it
-  comes with https://git.eeqj.de/sneak/smallwebwaf/issues/68.
+  request is dropped first. `bans.json` shows the bans and their notes, a
+  restart lifts none, and you add or lift a ban by editing it (see "State files"
+  below).
 - Refuses a request from a country you refuse with `SWWAF_BAN_RESPONSE`, as soon
   as the client's country is known and before its body is read; such a request
   is not counted for the rate limits. While one of the country lists below is
@@ -118,6 +128,18 @@ in `bin/state` unless `SWWAF_STATE_DIR` is set.
   `SWWAF_ALLOW_NETS` too is let through. A client in
   `SWWAF_RATE_LIMIT_EXEMPT_NETS` is neither counted nor refused by the rate
   limits; the country lists and bans still apply to it.
+- In `observe` mode, with `SWWAF_MODE=observe`, refuses none of the requests
+  that `SWWAF_DENY_NETS`, a ban, the country lists or a rate limit would refuse:
+  it passes them to the app, and their log lines name what `enforce` mode would
+  have done (see `would_action` in "Request log" below). The checks run, and
+  requests are counted, as in `enforce` mode, but a broken rate limit makes no
+  ban and does not set the client's counters back to zero, so each request over
+  the limit is logged as one that would be refused. The bans in `bans.json` are
+  kept, and refuse requests again when `smallwebwaf` next runs in `enforce`
+  mode, as long as they last. The timeouts and size limits still apply, since
+  they protect `smallwebwaf` and the app themselves, and a request for the
+  metrics without the token is still answered `401`. It is for trying a
+  configuration before enforcing it.
 - Answers `GET /_smallwebwaf/healthz` itself with `200` and `ok`, before any
   check and without asking the app, for the image's health check.
 - Answers `GET /_smallwebwaf/metrics` with its metrics (see "Metrics" below) for
@@ -139,6 +161,14 @@ it, and the effective settings are logged at start.
 - `SWWAF_LISTEN_ADDR` (default `:8080`): where `smallwebwaf` listens.
 - `SWWAF_UPSTREAM_URL` (default `http://127.0.0.1:8081`): the app, as `http` or
   `https`, a host and an optional port, and nothing more.
+- `SWWAF_INSTANCE_NAME` (default: the host's name, which docker sets to the
+  first 12 characters of the container's id unless the deployment names one):
+  the name each request log line gives as `instance`. Set it, for example to
+  `fsn1app1/gitea`, for a name that stays the same when a deploy replaces the
+  container, and that tells instances apart when several log to one place.
+- `SWWAF_MODE` (default `enforce`): `enforce`, or `observe` to pass on the
+  requests `smallwebwaf` would refuse and log what it would have done (see "What
+  it does so far" above).
 - `SWWAF_TRUSTED_PROXIES` (default `10.0.0.0/8,172.16.0.0/12,192.168.0.0/16`,
   the private address ranges): the netblocks whose `X-Forwarded-For` is
   believed. A list given replaces the default; set but empty, it trusts nothing.
@@ -174,6 +204,20 @@ it, and the effective settings are logged at start.
   requests a client may make in a minute, an hour and a day. The defaults are
   several times what one busy person produces, since a browser loading a heavy
   page makes a few hundred requests and several people often share one address.
+- `SWWAF_RATE_LIMIT_EXEMPT_PATHS` (default empty): path prefixes whose requests
+  the rate limits neither count nor refuse, such as `/assets/` for static
+  assets; each starts with `/`. A request whose path, percent-decoded, contains
+  `..` anywhere or a backslash, or whose path as sent holds an encoded slash
+  (`%2F` or `%2f`), is never exempt, since the app may act on it as a path
+  outside every prefix: `/assets/..%2Flogin` as `/login`. Any other request is
+  exempt when its path as sent, the path the app receives, before any query
+  string and not percent-decoded, starts with a prefix, character for character.
+  `/assets/` matches `/assets/app.js` and `/assets/`, but not `/assets`,
+  `/Assets/app.js`, `/%61ssets/app.js`, `/static/assets/app.js`,
+  `/static/../assets/app.js` or `/assets%2Fapp.js`. A character the client sends
+  percent-encoded, such as a space, is written percent-encoded in a prefix, as
+  in `/my%20files/`, and there are no wildcards: `*` is a character like any
+  other.
 - `SWWAF_DENIED_COUNTRIES` (default empty): countries whose clients are refused,
   for example `cn,ru,kp`.
 - `SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES` (default empty): when set, the only
@@ -203,6 +247,13 @@ it, and the effective settings are logged at start.
   `bans.json` is written, with every ban made in between.
 - `SWWAF_STATE_COUNTER_INTERVAL` (default `15m`): how often every state file is
   written.
+- `SWWAF_LOG_REQUEST_HEADERS` (default
+  `accept,accept-language,accept-encoding,content-type,origin,range`): the
+  request headers whose values the request log gives, in either case.
+  `Authorization`, `Cookie` and `Set-Cookie` are never logged, even when listed
+  (see "Request log" below). An entry naming `Host` or `Transfer-Encoding` stops
+  the start, since Go's HTTP server takes both out of the request; the request's
+  host is the field `host`.
 - `SWWAF_METRICS_TOKEN` (default unset): the token a scraper sends for the
   metrics, a long random value. While it is unset the metrics are off; one
   shorter than 32 characters stops the start. The settings logged at start show
@@ -232,18 +283,42 @@ GeoJS are kept, for 7 days each.
 refused ones included:
 
 ```
-{"type":"request","time":"2026-10-03T12:00:00.123Z","client_ip":"203.0.113.9","peer_ip":"172.18.0.2","country":"DE","method":"GET","host":"app.example","path":"/","query":"","protocol":"HTTP/1.1","status":200,"upstream_status":200,"request_bytes":0,"response_bytes":5120,"referer":"","user_agent":"curl/8.9.1","action":"forward","duration_total":3.217,"duration_upstream_total":3.104}
+{"type":"request","time":"2026-10-03T12:00:00.123Z","instance":"fsn1app1/gitea","client_ip":"203.0.113.9","method":"GET","scheme":"https","host":"app.example","path":"/","query":"","protocol":"HTTP/1.1","status":200,"request_bytes":0,"response_bytes":5120,"referer":"","user_agent":"curl/8.9.1","request_id":"7Q2NHZ4KJ3VXW5YB6R3MEFTD2A","peer_ip":"172.18.0.2","forwarded_for":"203.0.113.9","client_group":"203.0.113.9/32","country":"DE","request_headers":{"accept":"*/*"},"response_content_type":"text/html; charset=utf-8","upstream_status":200,"action":"forward","counts":{"minute":1,"hour":12,"day":40},"duration_total":3.217,"duration_checks":0.041,"duration_upstream_connect":0.052,"duration_upstream_first_byte":2.874,"duration_upstream_total":3.104}
 ```
 
-- `time` is when the request arrived, in UTC. `peer_ip` is the TCP peer,
-  normally traefik. `path` and `query` are as the client sent them.
+A field that does not apply to a request is left out of its line, apart from
+`type`, the fields from `time` to `user_agent`, `request_id`, `peer_ip`,
+`client_group`, `country`, `action` and `duration_total`, which every line has.
+
+- `time` is when the request arrived, in UTC. `instance` is
+  `SWWAF_INSTANCE_NAME`. `scheme` is the `X-Forwarded-Proto` a trusted proxy
+  sent, and otherwise `http`. `path` and `query` are as the client sent them.
+- `request_id` is the `X-Request-ID` a trusted proxy sent, or a new random one
+  of 26 letters and digits when it sent none, or when the peer is not a trusted
+  proxy. A request passed to the app takes it there in `X-Request-ID`.
+- `peer_ip` is the TCP peer, normally traefik. `forwarded_for` is the
+  `X-Forwarded-For` header as received, several lines of it joined with `, `.
+  `client_group` is the client as the rate limits count it: its IPv4 address as
+  a /32, or the /64 of its IPv6 address.
 - `country` is the client's country as GeoJS places it. It is empty with neither
   country list set, for a client in `SWWAF_ALLOW_NETS` or `SWWAF_DENY_NETS`, for
   a client on a private, loopback or link-local address, when GeoJS cannot place
-  the client or has not answered in time, and for a request refused because a
-  ban covers its client, even when the client's country is known.
+  the client or has not answered in time, and for a request whose client a ban
+  covers, even when the client's country is known.
+- `content_type` is the request's `Content-Type`, and `content_length` the
+  length the request announced for its body, which is left out for none or zero.
+- `request_headers` are the request's headers that `SWWAF_LOG_REQUEST_HEADERS`
+  names, by name in lower case, several lines of one joined with `, `.
+  `Authorization`, `Cookie` and `Set-Cookie` are never among them, whatever the
+  setting says: `has_authorization` and `has_cookie` are there instead, and
+  true, when the request has an `Authorization` or a `Cookie` header.
+- `websocket` is there, and true, when the app switched the connection to
+  another protocol, as it does for a WebSocket.
 - `status` is what the client was sent, `0` if nothing was; `upstream_status` is
   what the app answered, and is left out when the app did not answer.
+- `response_content_type`, `cache_control` and `location` are the
+  `Content-Type`, `Cache-Control` and `Location` headers of the answer: the
+  app's, as passed on, or those of `smallwebwaf`'s own answer.
 - `request_bytes` and `response_bytes` count body bytes.
 - `action` is `forward` for a request passed to the app, `denied` for one
   refused because its client is in `SWWAF_DENY_NETS`, `banned` for one refused
@@ -253,16 +328,42 @@ refused ones included:
   `timed_out` for one that ran out of time, `upstream_error` when the app could
   not be reached or its answer broke off, and `admin` for one `smallwebwaf`
   answered at its own endpoint.
+- `would_action` is there in `observe` mode for a request that
+  `SWWAF_DENY_NETS`, a ban, the country lists or a rate limit would have refused
+  in `enforce` mode, and names the action that refusal would have had: `denied`,
+  `banned`, `country_denied` or `rate_limited`. `action` then names what was
+  done: `forward` for a request passed to the app, and another action, such as
+  `too_large`, for one a size or time limit refused.
+- `counts` gives the client's requests in the minute, the hour and the day as
+  the rate limits count them, this request included: in each window, those in
+  the bucket under way and a share of those in the bucket before, so a count can
+  have a fraction. For a request that broke a limit, they are the counts that
+  broke it. It is left out for a request the rate limits do not count: the
+  health check, one from a client in `SWWAF_ALLOW_NETS` or
+  `SWWAF_RATE_LIMIT_EXEMPT_NETS`, one for a path that
+  `SWWAF_RATE_LIMIT_EXEMPT_PATHS` exempts, and one that `SWWAF_DENY_NETS`, a ban
+  or the country lists refuse, or would refuse in `observe` mode. The byte
+  totals come with the byte limits.
 - `limit_hit` is there for a request that broke a rate limit, and names the
   window whose limit it went over: `minute`, `hour` or `day`, the shortest if it
   went over several. `offence` is then `limit`.
 - `ban_expires` is there for a request that made a ban or was refused under one,
-  and gives when the ban ends, in the same form as `time`, or `permanent`.
+  or in `observe` mode would have been refused under one, and gives when the ban
+  ends, in the same form as `time`, or `permanent`.
 - `aborted` is there, and true, when the client went away early.
-- `duration_total` and `duration_upstream_total` are in milliseconds.
+- The timings are in milliseconds, to the microsecond. `duration_total` runs
+  from when the request's headers had been read to when its line is written, and
+  `duration_checks` over the same start to when the checks were done; the health
+  check runs none, and its line has no `duration_checks`.
+  `duration_upstream_connect`, `duration_upstream_first_byte` and
+  `duration_upstream_total` are there for a request passed to the app, and run
+  from when it was handed to the app: until there was a connection to it, new or
+  kept open from an earlier request, until the first byte of its answer arrived,
+  and until the end. The first two are left out when that never happened, as for
+  an app that cannot be reached.
 
-No body and no other header is logged. `smallwebwaf`'s own messages (start, the
-settings, stop, errors) share the stream as JSON lines marked
+No body is logged, and no header but those above. `smallwebwaf`'s own messages
+(start, the settings, stop, errors) share the stream as JSON lines marked
 `"type":"process"`.
 
 Go's HTTP server, on which `smallwebwaf` is built, reads a request's line and
@@ -311,9 +412,42 @@ without a field it needs, named with the entry's place in the file: a ban's
 `netblock`, `start` or `expires`, which is `null` for a permanent ban; a
 client's `client`, or the `start` of a window in which it has requests; an
 answer's `client`, `country`, which is `""` for a client GeoJS cannot place, or
-`answered`. An edit made while `smallwebwaf` runs is overwritten by its next
-write: taking it in comes with https://git.eeqj.de/sneak/smallwebwaf/issues/68.
-The AS number and AS name come with their lookup.
+`answered`. The AS number and AS name come with their lookup.
+
+While it runs, `smallwebwaf` watches `SWWAF_STATE_DIR` and takes in your edit of
+a state file as soon as you save it: what the file then holds replaces what
+`smallwebwaf` held for it, as if read at start. It tells its own writes from
+yours by comparing the file with what it last read or wrote, and before it
+writes a file it takes in any edit made since, so your edit is not overwritten;
+a change `smallwebwaf` made after you opened the file, such as a new ban, is
+lost when you save over it. An edit that would stop the start, because it does
+not parse, has another `version` or leaves out a field an entry needs, does not
+stop the running `smallwebwaf`: it keeps what it holds, and at the file's next
+write renames your file to `<name>.bad`, such as `bans.json.bad`, writes the
+file again from memory, and logs the file and where the error is. It waits for
+that write because an editor's file can be read before the editor has finished
+writing it. Mend the `.bad` file and move it back. A file you remove is written
+again at its next write.
+
+To ban a netblock, add an entry to `bans.json` with its `netblock`, its `start`
+and its `expires`, `null` for a ban that never ends; its `notes` may be left
+out. This `bans.json` bans `203.0.113.0/24` for good:
+
+```json
+{
+    "version": 1,
+    "bans": [
+        {
+            "netblock": "203.0.113.0/24",
+            "start": "2026-10-06T12:00:00Z",
+            "expires": null
+        }
+    ]
+}
+```
+
+To lift a ban, delete its entry. `smallwebwaf` then forgets the ban, so it does
+not make the netblock's next ban longer.
 
 ## Metrics
 
@@ -352,7 +486,10 @@ other request. No metric carries a client's address.
 - `smallwebwaf_state_file_writes_total`,
   `smallwebwaf_state_file_write_failures_total`,
   `smallwebwaf_state_file_last_write_timestamp_seconds` and
-  `smallwebwaf_state_file_size_bytes`, by `file`.
+  `smallwebwaf_state_file_size_bytes`, by `file`; and, by `file` too,
+  `smallwebwaf_state_file_edits_taken_in_total`: your edits taken in, and
+  `smallwebwaf_state_file_edits_set_aside_total`: those renamed to `<name>.bad`
+  because they would stop the start.
 - Go's own `go_` metrics and the process's `process_` metrics.
 
 The requests Go's HTTP server ends before `smallwebwaf` sees them (see "Request
@@ -460,9 +597,9 @@ goes through the candidates one by one.
   readable JSON files, written regularly and at every stop, so a restart loses
   nothing. Edit a file, or add a rule file, and the running `smallwebwaf` picks
   up the change. Nothing is read from disk while serving a request. The files
-  for the bans, the clients and the GeoJS answers are built (see "State files"
-  above); the others come with their features, and taking in an edit while
-  running comes with https://git.eeqj.de/sneak/smallwebwaf/issues/68.
+  for the bans, the clients and the GeoJS answers are built, with an edit taken
+  in while running (see "State files" above); the others come with their
+  features.
 - Health checks, the metrics, and listing, adding and lifting bans or asking why
   a given address was refused, all on the one port every request uses: under
   `/_smallwebwaf/` on the app's own address, through traefik like any other
@@ -581,17 +718,16 @@ the metrics, failure behaviour and the build order.
 So far `smallwebwaf` looks up only the country, only through GeoJS, and only
 while `SWWAF_DENIED_COUNTRIES` or `SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES` is set:
 then the address of every new visitor is sent to GeoJS, except a visitor in
-`SWWAF_ALLOW_NETS` or `SWWAF_DENY_NETS` and one refused because a ban covers its
-netblock, and with neither set, none is. An IPv6 visitor is asked about by the
-first address of its /64. A new visitor waits at most a second for its answer,
-and without one counts as coming from an unknown country until the answer
-arrives. The addresses waiting are asked about together, up to 200 in one
-request, one request at a time; at most 10,000 visitors wait, and one more
-counts as coming from an unknown country until there is room. While GeoJS fails,
-visitors with a kept answer are unaffected and new ones count as coming from an
-unknown country. GeoJS is then left alone for a second, twice as long after each
-further failure up to five minutes, and asked again by the next request that
-needs it.
+`SWWAF_ALLOW_NETS` or `SWWAF_DENY_NETS` and one whose netblock a ban covers, and
+with neither set, none is. An IPv6 visitor is asked about by the first address
+of its /64. A new visitor waits at most a second for its answer, and without one
+counts as coming from an unknown country until the answer arrives. The addresses
+waiting are asked about together, up to 200 in one request, one request at a
+time; at most 10,000 visitors wait, and one more counts as coming from an
+unknown country until there is room. While GeoJS fails, visitors with a kept
+answer are unaffected and new ones count as coming from an unknown country.
+GeoJS is then left alone for a second, twice as long after each further failure
+up to five minutes, and asked again by the next request that needs it.
 
 In the full design, `smallwebwaf` looks up the AS number and country of every
 client, for the request log, the metrics and the ban notes, and for the country
@@ -643,8 +779,10 @@ addresses are never sent to GeoJS.
   limits, and writes the request's log line. Its `check` method is where a
   request is refused before anything reaches the app: for `SWWAF_DENY_NETS`, for
   a ban, for the country lists, for a rate limit, which bans the client, and for
-  an announced body over the size limit. A request under `/_smallwebwaf/` that
-  `check` lets through is answered by `answerAdmin` instead of reaching the app.
+  an announced body over the size limit; in `observe` mode, only for the size
+  limit, with what it would have refused for noted in the log line. A request
+  under `/_smallwebwaf/` that `check` lets through is answered by `answerAdmin`
+  instead of reaching the app.
 - `internal/metrics`: the metrics, counted as the other parts tell it what
   happened, and served in the Prometheus text format.
 - `internal/bans`: the ban ledger: each netblock's bans with their notes, how
@@ -653,8 +791,8 @@ addresses are never sent to GeoJS.
   answers.
 - `internal/ratelimit`: the table of clients: counts each client's requests,
   tells when one takes it over a rate limit, and keeps each client's history.
-- `internal/state`: reads the state files at start, and writes them when they
-  are due and at the stop.
+- `internal/state`: reads the state files at start, takes in an admin's edit of
+  one while running, and writes them when they are due and at the stop.
 - `internal/requestlog`: the lines on stdout: the request log line and the
   process's own messages.
 - `Dockerfile`: the lint and test phases, then the image, whose last stage
@@ -666,8 +804,9 @@ addresses are never sent to GeoJS.
 Besides the Go standard library, `github.com/hashicorp/golang-lru/v2` keeps the
 table of clients to 20,000, the GeoJS answers to 100,000 and the banned
 netblocks to `SWWAF_MAX_BANS`, dropping the least recently seen, and
-`github.com/prometheus/client_golang` keeps the metrics and serves them. The
-country codes are the list in `internal/config/config.go`.
+`github.com/prometheus/client_golang` keeps the metrics and serves them, and
+`github.com/fsnotify/fsnotify` tells `smallwebwaf` when a state file is saved.
+The country codes are the list in `internal/config/config.go`.
 
 ## Entrypoints
 
@@ -710,10 +849,8 @@ so that they run in minimal containers.
 
 ## TODO
 
-- The rest of milestone 3, from taking in an admin's edits to the state files
-  (https://git.eeqj.de/sneak/smallwebwaf/issues/68) up to the rest of the
-  request log's fields, and the rest of the design, in the order of the build
-  order in [`SPEC.md`](SPEC.md).
+- The rest of the design, in the order of the build order in
+  [`SPEC.md`](SPEC.md).
 
 ## Documents
 

go.mod

@@ -3,6 +3,7 @@ module sneak.berlin/go/smallwebwaf
 go 1.26.0
 
 require (
+	github.com/fsnotify/fsnotify v1.10.1
 	github.com/hashicorp/golang-lru/v2 v2.0.7
 	github.com/prometheus/client_golang v1.24.1
 )

go.sum

@@ -4,6 +4,8 @@ github.com/cespare/xxhash/v2 v2.3.0 h1:UL815xU9SqsFlibzuggzjXhog7bL6oX9BbNZnL2UF
 github.com/cespare/xxhash/v2 v2.3.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/fsnotify/fsnotify v1.10.1 h1:b0/UzAf9yR5rhf3RPm9gf3ehBPpf0oZKIjtpKrx59Ho=
+github.com/fsnotify/fsnotify v1.10.1/go.mod h1:TLheqan6HD6GBK6PrDWyDPBaEV8LspOxvPSjC+bVfgo=
 github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
 github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
 github.com/hashicorp/golang-lru/v2 v2.0.7 h1:a+bsQ5rvGLjzHuww6tVxozPZFVghXaHOwFs4luLUK2k=

internal/bans/bans.go

@@ -26,9 +26,9 @@ const maxTextBytes = 256
 type Rules struct {
 	// LimitBanDuration is how long a first ban lasts.
 	LimitBanDuration time.Duration
-	// LimitBanRepeatWindow is how soon after the netblock's last ban
-	// ended a broken limit counts as a repeat, which bans for
-	// repeatFactor times as long as that ban.
+	// LimitBanRepeatWindow is how soon after the end of the netblock's
+	// ban that ended last a broken limit counts as a repeat, which bans
+	// for repeatFactor times as long as that ban.
 	LimitBanRepeatWindow time.Duration
 	// MaxBanDuration is the longest ban; a ban that would be longer is
 	// permanent instead.
@@ -107,8 +107,8 @@ type Ledger struct {
 	changed chan struct{}
 
 	mu sync.Mutex
-	// netblocks holds each banned netblock's bans, oldest first. Check
-	// makes each netblock it finds the most recently seen.
+	// netblocks holds each banned netblock's bans, oldest first. Check and
+	// Find make each netblock they find the most recently seen.
 	netblocks *simplelru.LRU[netip.Prefix, *[]Ban]
 	// held is how many bans netblocks holds, at most rules.MaxBans.
 	held int
@@ -144,41 +144,55 @@ func (l *Ledger) Changed() <-chan struct{} {
 	return l.changed
 }
 
-// Check is called for each request from client, at now. It reports
-// whether a ban on a netblock client is in is active, and returns that
-// ban, with the request counted among those it refused.
+// Check is called for a request from client, at now. It reports whether
+// a ban on a netblock client is in is active, and returns that ban, with
+// the request counted among those it refused.
 func (l *Ledger) Check(client netip.Addr, now time.Time) (Ban, bool) {
 	l.mu.Lock()
 	defer l.mu.Unlock()
 
-	lengths := l.v6Lengths
-	if client.Is4() {
-		lengths = l.v4Lengths
+	ban := l.active(client, now)
+	if ban == nil {
+		return Ban{}, false
 	}
 
-	for _, length := range lengths {
-		bans, found := l.netblocks.Get(netip.PrefixFrom(client, length).Masked())
-		if !found {
-			continue
-		}
+	ban.Notes.Requests++
+	ban.Notes.Refused++
 
-		// A ban is made only once the one before has ended, so only the
-		// last can be active.
-		last := &(*bans)[len(*bans)-1]
-		if last.ActiveAt(now) {
-			last.Notes.Requests++
-			last.Notes.Refused++
+	return *ban, true
+}
 
-			return *last, true
+// Find is Check without counting the request among those the ban
+// refused: in observe mode a ban refuses nothing.
+func (l *Ledger) Find(client netip.Addr, now time.Time) (Ban, bool) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	ban := l.active(client, now)
+	if ban == nil {
+		return Ban{}, false
+	}
+
+	return *ban, true
+}
+
+// activeBan returns the ban in bans, a netblock's bans oldest first, that
+// is active at now, or nil when none is. If several are, it returns the
+// one that started last. Every ban is looked at, since a ban an admin adds
+// to bans.json can start before the netblock's others and outlast them.
+func activeBan(bans []Ban, now time.Time) *Ban {
+	for i := len(bans) - 1; i >= 0; i-- {
+		if bans[i].ActiveAt(now) {
+			return &bans[i]
 		}
 	}
 
-	return Ban{}, false
+	return nil
 }
 
 // BanForLimit bans netblock at now for a broken limit, with notes, and
 // returns the ban. A first ban lasts LimitBanDuration. A ban made within
-// LimitBanRepeatWindow after the netblock's last ban ended lasts
+// LimitBanRepeatWindow after the netblock's ban that ended last lasts
 // repeatFactor times as long as that one. A ban that would be longer
 // than MaxBanDuration is permanent instead. If a ban on netblock is still
 // active, as when two of its requests break a limit at once, that ban is
@@ -192,12 +206,22 @@ func (l *Ledger) BanForLimit(netblock netip.Prefix, now time.Time, notes Notes)
 
 	bans, found := l.netblocks.Get(netblock)
 	if found {
-		last = &(*bans)[len(*bans)-1]
-		if last.ActiveAt(now) {
-			return *last
+		active := activeBan(*bans, now)
+		if active != nil {
+			return *active
 		}
 
-		notes.EarlierBans = last.Notes.EarlierBans + 1
+		// No ban is active, so each has an end. A ban an admin adds to
+		// bans.json can start after another and end before it, so the
+		// ban that ended last is looked for among them all.
+		ended := slices.MaxFunc(*bans, func(a, b Ban) int {
+			return a.Expires.Compare(b.Expires)
+		})
+		last = &ended
+
+		// The netblock's first ban held counts the bans it had before that
+		// one, since dropped to make room, and each ban held adds one.
+		notes.EarlierBans = (*bans)[0].Notes.EarlierBans + len(*bans)
 	}
 
 	notes.Request = notes.Request.cut()
@@ -282,21 +306,25 @@ func (l *Ledger) Snapshot() []Ban {
 	return held
 }
 
-// Load puts bans read from bans.json into a ledger that holds none yet,
-// in the order they started, so that a netblock whose last ban started
-// latest counts as the most recently seen. Each netblock is masked to its
-// length, so that 203.0.113.9/24 is 203.0.113.0/24, and each text in the
-// notes is cut to 256 bytes. Past MaxBans the earliest bans are dropped,
-// as when they are made.
+// Load puts bans read from bans.json into the ledger, in place of the
+// bans it holds, in the order they started, so that a netblock whose last
+// ban started latest counts as the most recently seen. Each netblock is
+// masked to its length, so that 203.0.113.9/24 is 203.0.113.0/24, and
+// each text in the notes is cut to 256 bytes. Past MaxBans the earliest
+// bans are dropped, as when they are made.
 func (l *Ledger) Load(bans []Ban) {
-	l.mu.Lock()
-	defer l.mu.Unlock()
-
 	bans = slices.Clone(bans)
 	slices.SortStableFunc(bans, func(a, b Ban) int {
 		return a.Start.Compare(b.Start)
 	})
 
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	l.netblocks.Purge()
+	l.held = 0
+	l.v4Lengths, l.v6Lengths = nil, nil
+
 	for _, ban := range bans {
 		ban.Netblock = ban.Netblock.Masked()
 		ban.Notes.Request = ban.Notes.Request.cut()
@@ -304,6 +332,29 @@ func (l *Ledger) Load(bans []Ban) {
 	}
 }
 
+// active returns the ban active at now on a netblock client is in, or
+// nil.
+func (l *Ledger) active(client netip.Addr, now time.Time) *Ban {
+	lengths := l.v6Lengths
+	if client.Is4() {
+		lengths = l.v4Lengths
+	}
+
+	for _, length := range lengths {
+		bans, found := l.netblocks.Get(netip.PrefixFrom(client, length).Masked())
+		if !found {
+			continue
+		}
+
+		ban := activeBan(*bans, now)
+		if ban != nil {
+			return ban
+		}
+	}
+
+	return nil
+}
+
 // add adds ban to its netblock's bans, after the last, and makes its
 // netblock the most recently seen. With MaxBans held, it drops one first.
 func (l *Ledger) add(ban Ban) {
@@ -333,8 +384,8 @@ func (l *Ledger) add(ban Ban) {
 }
 
 // expiry returns when a ban for a broken limit made at now ends, or zero
-// when it is permanent. last is the netblock's last ban, which has ended,
-// or nil when it has none.
+// when it is permanent. last is the netblock's ban that ended last, or nil
+// when it has none.
 func (l *Ledger) expiry(last *Ban, now time.Time) time.Time {
 	length := l.rules.LimitBanDuration
 

internal/bans/bans_test.go

@@ -162,6 +162,28 @@ func TestCheckRefusesWhileTheBanLastsAndCountsTheRefusals(t *testing.T) {
 	}
 }
 
+func TestFindCountsNothing(t *testing.T) {
+	t.Parallel()
+
+	ledger := bans.New(defaultRules())
+	netblock := netip.MustParsePrefix("203.0.113.9/32")
+	ban := ledger.BanForLimit(netblock, midnight(), bans.Notes{Requests: 5})
+
+	got, banned := ledger.Find(netblock.Addr(), ban.Expires.Add(-time.Nanosecond))
+	if !banned || got != ban {
+		t.Errorf("find during the ban gives %+v and %t, want %+v", got, banned, ban)
+	}
+
+	_, banned = ledger.Find(netblock.Addr(), ban.Expires)
+	if banned {
+		t.Error("the ban did not end")
+	}
+
+	if notes := ledger.Bans(netblock)[0].Notes; notes != ban.Notes {
+		t.Errorf("the notes are %+v, want them unchanged, %+v", notes, ban.Notes)
+	}
+}
+
 func TestMaxBansDropsTheEarliestBanOfTheNetblockSeenLongestAgo(t *testing.T) {
 	t.Parallel()
 

internal/bans/snapshot_test.go

@@ -130,6 +130,75 @@ func TestLoadedBanRefusesEveryClientInItsNetblock(t *testing.T) {
 	}
 }
 
+func TestPermanentBanStartedBeforeAnEndedOneRefuses(t *testing.T) {
+	t.Parallel()
+
+	// As when an admin adds a permanent ban to bans.json with a start
+	// before that of the netblock's ban that has ended.
+	netblock := netip.MustParsePrefix("203.0.113.0/24")
+	permanent := bans.Ban{Netblock: netblock, Start: midnight().Add(-time.Hour)}
+	ended := bans.Ban{
+		Netblock: netblock,
+		Start:    midnight(),
+		Expires:  midnight().Add(time.Hour),
+	}
+
+	ledger := bans.New(defaultRules())
+	ledger.Load([]bans.Ban{permanent, ended})
+
+	now := midnight().Add(2 * time.Hour)
+	client := netip.MustParseAddr("203.0.113.9")
+
+	ban, banned := ledger.Find(client, now)
+	if !banned || !ban.Permanent() {
+		t.Errorf("find gives %+v and %t, want the permanent ban", ban, banned)
+	}
+
+	ban, banned = ledger.Check(client, now)
+	if !banned || !ban.Permanent() {
+		t.Errorf("the client is refused: %t, under %+v, want under the permanent ban",
+			banned, ban)
+	}
+
+	// A limit broken now makes no shorter ban over the permanent one.
+	ban = ledger.BanForLimit(netblock, now, bans.Notes{})
+	if !ban.Permanent() || len(ledger.Bans(netblock)) != 2 {
+		t.Errorf("a broken limit returned %+v and left the netblock %d bans, "+
+			"want the permanent ban and 2", ban, len(ledger.Bans(netblock)))
+	}
+}
+
+func TestNextBanWorkedOutFromTheBanThatEndedLast(t *testing.T) {
+	t.Parallel()
+
+	// A 9-hour ban smallwebwaf made, the third in a row, and an admin's
+	// 1-hour ban added to bans.json over it, with no notes.
+	netblock := netip.MustParsePrefix("203.0.113.9/32")
+	nineHours := bans.Ban{
+		Netblock: netblock,
+		Start:    midnight(),
+		Expires:  midnight().Add(9 * time.Hour),
+		Notes:    bans.Notes{EarlierBans: 2},
+	}
+	admins := bans.Ban{
+		Netblock: netblock,
+		Start:    midnight().Add(time.Hour),
+		Expires:  midnight().Add(2 * time.Hour),
+	}
+
+	ledger := bans.New(defaultRules())
+	ledger.Load([]bans.Ban{nineHours, admins})
+
+	// Once both have ended, a limit broken within the repeat window bans
+	// for three times the 9 hours, and the notes count the two bans
+	// before the 9-hour one, it, and the admin's.
+	ban := ledger.BanForLimit(netblock, nineHours.Expires.Add(time.Hour), bans.Notes{})
+	if ban.Expires.Sub(ban.Start) != 27*time.Hour || ban.Notes.EarlierBans != 4 {
+		t.Errorf("the next ban lasts %s with %d earlier bans, want 27h and 4",
+			ban.Expires.Sub(ban.Start), ban.Notes.EarlierBans)
+	}
+}
+
 func TestLoadKeepsAtMostMaxBansDroppingTheEarliest(t *testing.T) {
 	t.Parallel()
 
@@ -151,6 +220,41 @@ func TestLoadKeepsAtMostMaxBansDroppingTheEarliest(t *testing.T) {
 	}
 }
 
+func TestLoadReplacesTheBansHeld(t *testing.T) {
+	t.Parallel()
+
+	// Room for three bans, so that the second load, were it added to the
+	// two bans held, would drop none of them to make room.
+	rules := defaultRules()
+	rules.MaxBans = 3
+	ledger := bans.New(rules)
+	kept := bans.Ban{Netblock: netip.MustParsePrefix("2001:db8::/64"), Start: midnight()}
+	ledger.Load([]bans.Ban{
+		{Netblock: netip.MustParsePrefix("203.0.113.0/24"), Start: midnight()},
+		kept,
+	})
+
+	// Loaded again without the first ban, as when an admin's edit of
+	// bans.json is taken in, that ban is lifted.
+	ledger.Load([]bans.Ban{kept})
+
+	_, banned := ledger.Check(netip.MustParseAddr("203.0.113.9"), midnight())
+	if banned {
+		t.Error("a ban left out of the second load still refuses")
+	}
+
+	// The ledger holds one ban, so it makes two more without dropping any.
+	first := ledger.BanForLimit(netip.MustParsePrefix("198.51.100.7/32"), midnight(),
+		bans.Notes{})
+	second := ledger.BanForLimit(netip.MustParsePrefix("198.51.100.8/32"), midnight(),
+		bans.Notes{})
+
+	want := []bans.Ban{first, second, kept}
+	if got := ledger.Snapshot(); !slices.Equal(got, want) {
+		t.Errorf("the ledger holds %+v, want %+v", got, want)
+	}
+}
+
 func TestLoadCutsTheTextsTo256Bytes(t *testing.T) {
 	t.Parallel()
 

internal/config/config.go

@@ -12,6 +12,7 @@ import (
 	"net/http"
 	"net/netip"
 	"net/url"
+	"os"
 	"path/filepath"
 	"slices"
 	"strconv"
@@ -27,6 +28,15 @@ type Config struct {
 	ListenAddr string
 	// UpstreamURL is the app (SWWAF_UPSTREAM_URL).
 	UpstreamURL *url.URL
+	// InstanceName is the name each request log line gives as instance
+	// (SWWAF_INSTANCE_NAME), by default the host's name, which docker sets
+	// to the first 12 characters of the container's id.
+	InstanceName string
+	// Observe is true in observe mode, when SWWAF_MODE is observe rather
+	// than enforce: a request that SWWAF_DENY_NETS, a ban, the country
+	// lists or a rate limit would refuse is passed to the app instead, and
+	// no ban is made.
+	Observe bool
 	// TrustedProxies are the netblocks whose X-Forwarded-For is
 	// believed (SWWAF_TRUSTED_PROXIES).
 	TrustedProxies []netip.Prefix
@@ -70,6 +80,10 @@ type Config struct {
 	RateLimitPerMinute int64
 	RateLimitPerHour   int64
 	RateLimitPerDay    int64
+	// RateLimitExemptPaths are the path prefixes whose requests the rate
+	// limits neither count nor refuse (SWWAF_RATE_LIMIT_EXEMPT_PATHS).
+	// Each starts with /.
+	RateLimitExemptPaths []string
 	// DeniedCountries are the countries whose clients are refused
 	// (SWWAF_DENIED_COUNTRIES). ExclusivelyAllowedCountries, when not
 	// empty, are the only countries whose clients are let through
@@ -104,6 +118,9 @@ type Config struct {
 	StateDir             string
 	StateWriteDelay      time.Duration
 	StateCounterInterval time.Duration
+	// LogRequestHeaders are the request headers whose values the request
+	// log gives, in lower case (SWWAF_LOG_REQUEST_HEADERS).
+	LogRequestHeaders []string
 	// MetricsToken is the bearer token a scraper sends for the metrics
 	// (SWWAF_METRICS_TOKEN), "" while it is unset and the metrics are off.
 	// MetricsTopN is how many countries get series of their own in the
@@ -150,6 +167,11 @@ var (
 			"such as http://127.0.0.1:8081")
 	errNotCountry = errors.New(
 		"is not a two-letter country code such as de or kp")
+	errNotHeaderName = errors.New(
+		"is not a header name such as accept-language")
+	errHeaderTakenOut = errors.New(
+		"is taken out of every request by Go's HTTP server, so it can never " +
+			"be logged")
 	errOnBothLists          = errors.New("is in SWWAF_DENIED_COUNTRIES too")
 	errNotOver4K            = errors.New("is not a size of more than 4K, such as 32K")
 	errNotDurationAboveZero = errors.New(
@@ -161,7 +183,10 @@ var (
 		"is not the length of an IPv4 netblock, from 0 to 32, such as 24")
 	errNotAbsolutePath = errors.New(
 		"is not an absolute path, such as /var/lib/smallwebwaf")
-	errShortToken = errors.New("is shorter than 32 characters")
+	errShortToken    = errors.New("is shorter than 32 characters")
+	errNotMode       = errors.New("is not enforce or observe")
+	errNotPathPrefix = errors.New(
+		"is not a path prefix starting with /, such as /assets/")
 )
 
 // FromEnvironment reads the settings with lookupEnv, normally
@@ -169,9 +194,12 @@ var (
 // that is set but invalid is an error that names it.
 func FromEnvironment(lookupEnv func(string) (string, bool)) (*Config, error) {
 	env := &environment{lookupEnv: lookupEnv}
+	hostname, _ := os.Hostname() // "" when the host has no name to give
 	cfg := &Config{
 		ListenAddr:           env.address("SWWAF_LISTEN_ADDR", ":8080"),
 		UpstreamURL:          env.appURL("SWWAF_UPSTREAM_URL", "http://127.0.0.1:8081"),
+		InstanceName:         env.value("SWWAF_INSTANCE_NAME", hostname),
+		Observe:              env.observe("SWWAF_MODE", "enforce"),
 		TrustedProxies:       env.netblocks("SWWAF_TRUSTED_PROXIES", privateRanges),
 		ClientRequestTimeout: env.duration("SWWAF_CLIENT_REQUEST_TIMEOUT", "60s"),
 		ClientRequestHeaderMaxBytes: env.headerSize(
@@ -188,6 +216,7 @@ func FromEnvironment(lookupEnv func(string) (string, bool)) (*Config, error) {
 		RateLimitPerMinute:      env.count("SWWAF_RATE_LIMIT_PER_MINUTE", "1000"),
 		RateLimitPerHour:        env.count("SWWAF_RATE_LIMIT_PER_HOUR", "10000"),
 		RateLimitPerDay:         env.count("SWWAF_RATE_LIMIT_PER_DAY", "50000"),
+		RateLimitExemptPaths:    env.pathPrefixes("SWWAF_RATE_LIMIT_EXEMPT_PATHS", ""),
 		DeniedCountries:         env.countries("SWWAF_DENIED_COUNTRIES", ""),
 		ExclusivelyAllowedCountries: env.countries(
 			"SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES", ""),
@@ -200,8 +229,10 @@ func FromEnvironment(lookupEnv func(string) (string, bool)) (*Config, error) {
 		StateDir:             env.absolutePath("SWWAF_STATE_DIR", "/var/lib/smallwebwaf"),
 		StateWriteDelay:      env.durationNotOff("SWWAF_STATE_WRITE_DELAY", "10s"),
 		StateCounterInterval: env.durationNotOff("SWWAF_STATE_COUNTER_INTERVAL", "15m"),
-		MetricsToken:         env.token("SWWAF_METRICS_TOKEN"),
-		MetricsTopN:          env.numberNotOff("SWWAF_METRICS_TOP_N", "50"),
+		LogRequestHeaders: env.headerNames("SWWAF_LOG_REQUEST_HEADERS",
+			"accept,accept-language,accept-encoding,content-type,origin,range"),
+		MetricsToken: env.token("SWWAF_METRICS_TOKEN"),
+		MetricsTopN:  env.numberNotOff("SWWAF_METRICS_TOP_N", "50"),
 	}
 
 	for _, country := range cfg.ExclusivelyAllowedCountries {
@@ -274,6 +305,17 @@ func (e *environment) appURL(name, defaultValue string) *url.URL {
 	return upstream
 }
 
+// observe reads the setting that is the mode, enforce or observe, and
+// reports whether it is observe.
+func (e *environment) observe(name, defaultValue string) bool {
+	mode := e.value(name, defaultValue)
+	if mode != "enforce" && mode != "observe" {
+		e.check(name, fmt.Errorf("%q %w", mode, errNotMode))
+	}
+
+	return mode == "observe"
+}
+
 // netblocks reads a setting that is a list of netblocks.
 func (e *environment) netblocks(name, defaultValue string) []netip.Prefix {
 	netblocks, err := parseNetblocks(e.value(name, defaultValue))
@@ -315,6 +357,14 @@ func (e *environment) count(name, defaultValue string) int64 {
 	return count
 }
 
+// pathPrefixes reads a setting that is a list of path prefixes.
+func (e *environment) pathPrefixes(name, defaultValue string) []string {
+	prefixes, err := parsePathPrefixes(e.value(name, defaultValue))
+	e.check(name, err)
+
+	return prefixes
+}
+
 // countries reads a setting that is a list of countries.
 func (e *environment) countries(name, defaultValue string) []string {
 	countries, err := parseCountries(e.value(name, defaultValue))
@@ -323,6 +373,15 @@ func (e *environment) countries(name, defaultValue string) []string {
 	return countries
 }
 
+// headerNames reads a setting that is a list of header names, and
+// returns them in lower case.
+func (e *environment) headerNames(name, defaultValue string) []string {
+	headers, err := parseHeaderNames(e.value(name, defaultValue))
+	e.check(name, err)
+
+	return headers
+}
+
 // durationNotOff reads a setting that is a duration and, unlike a
 // timeout, cannot be off.
 func (e *environment) durationNotOff(name, defaultValue string) time.Duration {
@@ -593,6 +652,23 @@ func parseNetblock(value string) (netip.Prefix, error) {
 	return netip.PrefixFrom(addr, addr.BitLen()), nil
 }
 
+// parsePathPrefixes reads a comma-separated list of path prefixes, each
+// starting with /.
+func parsePathPrefixes(value string) ([]string, error) {
+	prefixes, err := parseList(value)
+	if err != nil {
+		return nil, err
+	}
+
+	for _, prefix := range prefixes {
+		if !strings.HasPrefix(prefix, "/") {
+			return nil, fmt.Errorf("%q %w", prefix, errNotPathPrefix)
+		}
+	}
+
+	return prefixes, nil
+}
+
 // countryCodes are the two-letter codes ISO 3166-1 assigns today, and XK,
 // the code in common use for Kosovo. golang.org/x/text/language cannot
 // check them: it also takes withdrawn codes such as su, and reserved ones
@@ -649,6 +725,44 @@ func parseCountries(value string) ([]string, error) {
 	return countries, nil
 }
 
+// headerNameChars are the characters RFC 9110 allows in a header name:
+// letters, digits and these marks.
+const headerNameChars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz" +
+	"0123456789!#$%&'*+-.^_`|~"
+
+// parseHeaderNames reads a comma-separated list of header names in either
+// case, and returns them in lower case. Host and Transfer-Encoding are
+// refused: Go's HTTP server takes them out of the request's headers.
+func parseHeaderNames(value string) ([]string, error) {
+	items, err := parseList(value)
+	if err != nil {
+		return nil, err
+	}
+
+	headers := make([]string, 0, len(items))
+
+	for _, item := range items {
+		for _, char := range item {
+			if !strings.ContainsRune(headerNameChars, char) {
+				return nil, fmt.Errorf("%q %w", item, errNotHeaderName)
+			}
+		}
+
+		header := strings.ToLower(item)
+		switch header {
+		case "host":
+			return nil, fmt.Errorf("%q %w; the request's host is the field host",
+				item, errHeaderTakenOut)
+		case "transfer-encoding":
+			return nil, fmt.Errorf("%q %w", item, errHeaderTakenOut)
+		}
+
+		headers = append(headers, header)
+	}
+
+	return headers, nil
+}
+
 // parseListenAddr checks an address to listen on: an optional host and a
 // port number.
 func parseListenAddr(value string) (string, error) {

internal/config/config_test.go

@@ -6,6 +6,7 @@ import (
 	"log/slog"
 	"maps"
 	"net/netip"
+	"os"
 	"slices"
 	"strings"
 	"testing"
@@ -18,6 +19,7 @@ import (
 const (
 	listenAddr              = "SWWAF_LISTEN_ADDR"
 	upstreamURL             = "SWWAF_UPSTREAM_URL"
+	mode                    = "SWWAF_MODE"
 	trustedProxies          = "SWWAF_TRUSTED_PROXIES"
 	clientRequestTimeout    = "SWWAF_CLIENT_REQUEST_TIMEOUT"
 	clientHeaderMaxBytes    = "SWWAF_CLIENT_REQUEST_HEADER_MAX_BYTES"
@@ -33,6 +35,7 @@ const (
 	rateLimitPerMinute      = "SWWAF_RATE_LIMIT_PER_MINUTE"
 	rateLimitPerHour        = "SWWAF_RATE_LIMIT_PER_HOUR"
 	rateLimitPerDay         = "SWWAF_RATE_LIMIT_PER_DAY"
+	rateLimitExemptPaths    = "SWWAF_RATE_LIMIT_EXEMPT_PATHS"
 	deniedCountries         = "SWWAF_DENIED_COUNTRIES"
 	allowedCountries        = "SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES"
 	banResponse             = "SWWAF_BAN_RESPONSE"
@@ -46,8 +49,14 @@ const (
 	stateCounterInterval    = "SWWAF_STATE_COUNTER_INTERVAL"
 	metricsToken            = "SWWAF_METRICS_TOKEN" //nolint:gosec // the setting's name
 	metricsTopN             = "SWWAF_METRICS_TOP_N"
+	instanceName            = "SWWAF_INSTANCE_NAME"
+	logRequestHeaders       = "SWWAF_LOG_REQUEST_HEADERS"
 )
 
+// defaultLogRequestHeaders is the default of SWWAF_LOG_REQUEST_HEADERS.
+const defaultLogRequestHeaders = "accept,accept-language,accept-encoding," +
+	"content-type,origin,range"
+
 // token is a token of 32 characters, the shortest allowed.
 const token = "0123456789abcdef0123456789abcdef"
 
@@ -83,6 +92,7 @@ func TestDefaults(t *testing.T) {
 
 	wantSettings(t, cfg, config.Config{
 		ListenAddr:                  ":8080",
+		Observe:                     false,
 		ClientRequestTimeout:        time.Minute,
 		ClientRequestHeaderMaxBytes: 32 << 10,
 		ClientIdleTimeout:           2 * time.Minute,
@@ -118,6 +128,22 @@ func TestDefaults(t *testing.T) {
 	wantNetblocks(t, cfg.DenyNets)
 	wantCountries(t, deniedCountries, cfg.DeniedCountries)
 	wantCountries(t, allowedCountries, cfg.ExclusivelyAllowedCountries)
+
+	hostname, err := os.Hostname()
+	if err != nil || hostname == "" || cfg.InstanceName != hostname {
+		t.Errorf("%s is %q, want the host's name %q (%v)", instanceName,
+			cfg.InstanceName, hostname, err)
+	}
+
+	wantHeaders := strings.Split(defaultLogRequestHeaders, ",")
+	if !slices.Equal(cfg.LogRequestHeaders, wantHeaders) {
+		t.Errorf("%s gave %v, want %v", logRequestHeaders, cfg.LogRequestHeaders,
+			wantHeaders)
+	}
+
+	if len(cfg.RateLimitExemptPaths) != 0 {
+		t.Errorf("%s gave %v, want none", rateLimitExemptPaths, cfg.RateLimitExemptPaths)
+	}
 }
 
 func TestValuesAsSet(t *testing.T) {
@@ -126,6 +152,7 @@ func TestValuesAsSet(t *testing.T) {
 	cfg := fromEnvironment(t, environment{
 		listenAddr:              "127.0.0.1:9000",
 		upstreamURL:             "https://app.internal:8443/",
+		mode:                    "observe",
 		trustedProxies:          " 192.0.2.1, 10.1.2.3/8 ,2001:db8::/32",
 		clientRequestTimeout:    "90s",
 		clientHeaderMaxBytes:    "8K",
@@ -141,6 +168,7 @@ func TestValuesAsSet(t *testing.T) {
 		rateLimitPerMinute:      "60",
 		rateLimitPerHour:        "600",
 		rateLimitPerDay:         "6000",
+		rateLimitExemptPaths:    "/assets/, /favicon.ico",
 		deniedCountries:         "cn, RU,kp,Xk",
 		allowedCountries:        "de",
 		banResponse:             "429",
@@ -158,6 +186,7 @@ func TestValuesAsSet(t *testing.T) {
 
 	wantSettings(t, cfg, config.Config{
 		ListenAddr:                  "127.0.0.1:9000",
+		Observe:                     true,
 		ClientRequestTimeout:        90 * time.Second,
 		ClientRequestHeaderMaxBytes: 8 << 10,
 		ClientIdleTimeout:           5 * time.Minute,
@@ -192,6 +221,39 @@ func TestValuesAsSet(t *testing.T) {
 	wantNetblocks(t, cfg.DenyNets, "198.51.100.0/24")
 	wantCountries(t, deniedCountries, cfg.DeniedCountries, "CN", "RU", "KP", "XK")
 	wantCountries(t, allowedCountries, cfg.ExclusivelyAllowedCountries, "DE")
+
+	if !slices.Equal(cfg.RateLimitExemptPaths, []string{"/assets/", "/favicon.ico"}) {
+		t.Errorf("%s gave %v, want /assets/ and /favicon.ico",
+			rateLimitExemptPaths, cfg.RateLimitExemptPaths)
+	}
+}
+
+func TestPathPrefixNotStartingWithSlashStopsTheStart(t *testing.T) {
+	t.Parallel()
+
+	_, err := config.FromEnvironment(
+		environment{rateLimitExemptPaths: "/favicon.ico,assets/"}.lookupEnv)
+
+	want := rateLimitExemptPaths + `: "assets/" is not a path prefix ` +
+		`starting with /, such as /assets/`
+	if err == nil || err.Error() != want {
+		t.Errorf("error %v, want %s", err, want)
+	}
+}
+
+func TestInstanceNameAndLoggedHeadersAsSet(t *testing.T) {
+	t.Parallel()
+
+	cfg := fromEnvironment(t, environment{
+		instanceName:      "fsn1app1/gitea",
+		logRequestHeaders: " Accept , X-Custom",
+	})
+
+	if cfg.InstanceName != "fsn1app1/gitea" ||
+		!slices.Equal(cfg.LogRequestHeaders, []string{"accept", "x-custom"}) {
+		t.Errorf("%s is %q and %s gives %v", instanceName, cfg.InstanceName,
+			logRequestHeaders, cfg.LogRequestHeaders)
+	}
 }
 
 func TestCodeOnBothCountryListsStopsTheStart(t *testing.T) {
@@ -295,9 +357,7 @@ func TestInvalidValueStopsTheStart(t *testing.T) {
 	t.Parallel()
 
 	for _, tc := range []struct{ name, value string }{
-		{listenAddr, "8080"},
-		{listenAddr, ":http"},
-		{listenAddr, ":65536"},
+		{listenAddr, "8080"}, {listenAddr, ":http"}, {listenAddr, ":65536"},
 		{upstreamURL, "127.0.0.1:8081"},
 		{upstreamURL, "ftp://127.0.0.1:8081"},
 		{upstreamURL, "http://"},
@@ -307,6 +367,7 @@ func TestInvalidValueStopsTheStart(t *testing.T) {
 		{upstreamURL, "http://127.0.0.1:8081/app"},
 		{upstreamURL, "http://127.0.0.1:8081/?a=1"},
 		{upstreamURL, "http://user:secret@127.0.0.1:8081"},
+		{mode, "Observe"}, {mode, "block"}, {mode, ""},
 		{trustedProxies, "10.0.0.0/33"},
 		{trustedProxies, "traefik"},
 		{trustedProxies, "10.0.0.0/8,,192.168.0.0/16"},
@@ -314,8 +375,7 @@ func TestInvalidValueStopsTheStart(t *testing.T) {
 		{allowNets, "192.0.2.0/24,monitoring"},
 		{rateLimitExemptNets, "2001:db8::/129"},
 		{denyNets, "198.51.100.0/24,"},
-		{clientRequestTimeout, "60"},
-		{clientRequestTimeout, ""},
+		{clientRequestTimeout, "60"}, {clientRequestTimeout, ""},
 		{clientIdleTimeout, "0s"},
 		{clientIdleTimeout, "2 minutes"},
 		{clientResponseTimeout, "1y"},
@@ -332,8 +392,8 @@ func TestInvalidValueStopsTheStart(t *testing.T) {
 		{rateLimitPerMinute, "1K"},
 		{rateLimitPerHour, "0"},
 		{rateLimitPerHour, "1.5"},
-		{rateLimitPerDay, "-1"},
-		{rateLimitPerDay, "lots"},
+		{rateLimitPerDay, "-1"}, {rateLimitPerDay, "lots"},
+		{rateLimitExemptPaths, "/assets/,,/static/"},
 		{deniedCountries, "nk"},
 		{deniedCountries, "kp,,ir"},
 		{deniedCountries, "prk"},
@@ -356,6 +416,10 @@ func TestInvalidValueStopsTheStart(t *testing.T) {
 		{stateWriteDelay, off}, {stateWriteDelay, "0s"},
 		{stateCounterInterval, off}, {stateCounterInterval, "15"},
 		{metricsTopN, off}, {metricsTopN, "0"}, {metricsTopN, "-1"},
+		{logRequestHeaders, "accept,,origin"}, {logRequestHeaders, "accept;origin"},
+		{logRequestHeaders, "accept language"}, {logRequestHeaders, "x-foo:"},
+		{logRequestHeaders, "host"}, {logRequestHeaders, "accept,Host"},
+		{logRequestHeaders, "transfer-encoding"}, {logRequestHeaders, "TRANSFER-ENCODING"},
 	} {
 		t.Run(tc.name+"="+tc.value, func(t *testing.T) {
 			t.Parallel()
@@ -372,6 +436,27 @@ func TestInvalidValueStopsTheStart(t *testing.T) {
 	}
 }
 
+func TestHostOrTransferEncodingStopsTheStart(t *testing.T) {
+	t.Parallel()
+
+	// Only Host's message points to the field host.
+	for value, want := range map[string]string{
+		"Host": `"Host" is taken out of every request by Go's HTTP server, ` +
+			"so it can never be logged; the request's host is the field host",
+		"transfer-encoding": `"transfer-encoding" is taken out of every ` +
+			"request by Go's HTTP server, so it can never be logged",
+	} {
+		t.Run(value, func(t *testing.T) {
+			t.Parallel()
+
+			_, err := config.FromEnvironment(environment{logRequestHeaders: value}.lookupEnv)
+			if err == nil || err.Error() != logRequestHeaders+": "+want {
+				t.Errorf("error %v, want %s: %s", err, logRequestHeaders, want)
+			}
+		})
+	}
+}
+
 func TestShortTokenStopsTheStartWithoutShowingIt(t *testing.T) {
 	t.Parallel()
 
@@ -423,9 +508,12 @@ func TestLogsEachSettingWithItsValue(t *testing.T) {
 		t.Fatalf("decode %s: %v", out.Bytes(), err)
 	}
 
+	hostname, _ := os.Hostname()
+
 	want := map[string]string{
 		listenAddr:              ":8080",
 		upstreamURL:             "http://127.0.0.1:8081",
+		mode:                    "enforce",
 		trustedProxies:          "10.0.0.0/8,172.16.0.0/12,192.168.0.0/16",
 		clientRequestTimeout:    "45s",
 		clientHeaderMaxBytes:    "32K",
@@ -441,6 +529,7 @@ func TestLogsEachSettingWithItsValue(t *testing.T) {
 		rateLimitPerMinute:      "1000",
 		rateLimitPerHour:        "10000",
 		rateLimitPerDay:         "50000",
+		rateLimitExemptPaths:    "",
 		deniedCountries:         "",
 		allowedCountries:        "",
 		banResponse:             "403",
@@ -454,6 +543,8 @@ func TestLogsEachSettingWithItsValue(t *testing.T) {
 		stateCounterInterval:    "15m",
 		metricsToken:            "",
 		metricsTopN:             "50",
+		instanceName:            hostname,
+		logRequestHeaders:       defaultLogRequestHeaders,
 	}
 	if !maps.Equal(line.Settings, want) {
 		t.Errorf("logged settings\n%v\nwant\n%v", line.Settings, want)
@@ -465,6 +556,7 @@ func wantSettings(t *testing.T, got *config.Config, want config.Config) {
 	t.Helper()
 
 	if got.ListenAddr != want.ListenAddr ||
+		got.Observe != want.Observe ||
 		got.ClientRequestTimeout != want.ClientRequestTimeout ||
 		got.ClientRequestHeaderMaxBytes != want.ClientRequestHeaderMaxBytes ||
 		got.ClientIdleTimeout != want.ClientIdleTimeout ||

internal/lookup/lookup.go

@@ -197,19 +197,21 @@ func (g *GeoJS) Snapshot() []Answer {
 	return answers
 }
 
-// Load keeps answers read from lookups.json, in a GeoJS that keeps none
-// yet, in the order they were last used, so that the one used longest
+// Load keeps answers read from lookups.json, in place of the answers it
+// keeps, in the order they were last used, so that the one used longest
 // ago is dropped first. Answers GeoJS gave keepFor ago or more are
 // dropped.
 func (g *GeoJS) Load(answers []Answer) {
-	g.mu.Lock()
-	defer g.mu.Unlock()
-
 	answers = slices.Clone(answers)
 	slices.SortStableFunc(answers, func(a, b Answer) int {
 		return a.Used.Compare(b.Used)
 	})
 
+	g.mu.Lock()
+	defer g.mu.Unlock()
+
+	g.answers.Purge()
+
 	now := g.now()
 
 	for _, answer := range answers {

internal/metrics/metrics.go

@@ -44,6 +44,8 @@ type Metrics struct {
 	stateFileWriteFailures *prometheus.CounterVec
 	stateFileLastWrite     *prometheus.GaugeVec
 	stateFileSize          *prometheus.GaugeVec
+	stateFileEditsTakenIn  *prometheus.CounterVec
+	stateFileEditsSetAside *prometheus.CounterVec
 }
 
 // New returns the metrics, with the Go runtime's and the process's own.
@@ -105,6 +107,11 @@ func New(topN int) *Metrics {
 			"When each state file was last written, in seconds since 1970.", byFile),
 		stateFileSize: gaugeVec("smallwebwaf_state_file_size_bytes",
 			"The size of each state file, as it was last written.", byFile),
+		stateFileEditsTakenIn: counterVec("smallwebwaf_state_file_edits_taken_in_total",
+			"Edits of each state file taken in while running.", byFile),
+		stateFileEditsSetAside: counterVec("smallwebwaf_state_file_edits_set_aside_total",
+			"Edits of each state file renamed to <name>.bad because they did not parse.",
+			byFile),
 	}
 
 	m.handler = promhttp.HandlerFor(m.registry, promhttp.HandlerOpts{})
@@ -120,6 +127,7 @@ func New(topN int) *Metrics {
 		m.GeoJSRequests, m.GeoJSFailures, m.GeoJSUnanswered,
 		m.stateFileWrites, m.stateFileWriteFailures,
 		m.stateFileLastWrite, m.stateFileSize,
+		m.stateFileEditsTakenIn, m.stateFileEditsSetAside,
 	)
 
 	return m
@@ -230,6 +238,18 @@ func (m *Metrics) StateFileWritten(name string, size int, err error) {
 	m.stateFileSize.WithLabelValues(name).Set(float64(size))
 }
 
+// StateFileEditTakenIn counts an admin's edit of the state file name
+// taken in while smallwebwaf runs.
+func (m *Metrics) StateFileEditTakenIn(name string) {
+	m.stateFileEditsTakenIn.WithLabelValues(name).Inc()
+}
+
+// StateFileEditSetAside counts an admin's edit of the state file name
+// renamed to name.bad because it did not parse.
+func (m *Metrics) StateFileEditSetAside(name string) {
+	m.stateFileEditsSetAside.WithLabelValues(name).Inc()
+}
+
 // statusClass returns the class of status, such as 2xx, or none when no
 // status was sent.
 func statusClass(status int) string {

internal/proxy/bans.go

@@ -14,10 +14,15 @@ func (rq *request) banResponse(action string) *refusal {
 	return &refusal{status: rq.h.config.BanResponse, action: action}
 }
 
-// banned reports whether a ban on a netblock the client is in refuses
-// the request at now, and notes for the log line when that ban ends.
+// banned reports whether a ban on a netblock the client is in covers the
+// request at now, and notes for the log line when that ban ends.
 func (rq *request) banned(now time.Time) bool {
-	ban, banned := rq.h.ledger.Check(rq.client, now)
+	check := rq.h.ledger.Check
+	if rq.h.config.Observe {
+		check = rq.h.ledger.Find // in observe mode the ban refuses nothing
+	}
+
+	ban, banned := check(rq.client, now)
 	if banned {
 		rq.line.BanExpires = banExpires(ban)
 	}
@@ -25,17 +30,28 @@ func (rq *request) banned(now time.Time) bool {
 	return banned
 }
 
-// limitBroken counts the request for the rate limits at now, and reports
-// whether it takes the client over one. Such a request bans the client's
-// netblock, and sets the client's counters back to zero.
+// limitBroken counts the request for the rate limits at now, notes the
+// client's counts for the log line, and reports whether the request takes
+// the client over a limit. In enforce mode such a request bans the
+// client's netblock, and sets the client's counters back to zero; in
+// observe mode it does neither.
 func (rq *request) limitBroken(now time.Time) bool {
 	group := clientGroup(rq.client)
 
-	hit, over := rq.h.limiter.Count(group, now)
+	counts, hit, over := rq.h.limiter.Count(group, now)
+	rq.line.Counts = counts
+
 	if !over {
 		return false
 	}
 
+	rq.line.LimitHit = hit.Window
+	rq.line.Offence = requestlog.OffenceLimit
+
+	if rq.h.config.Observe {
+		return true
+	}
+
 	netblock := rq.netblock()
 	ban := rq.h.ledger.BanForLimit(netblock, now, bans.Notes{
 		Country: rq.line.Country,
@@ -54,9 +70,6 @@ func (rq *request) limitBroken(now time.Time) bool {
 		Requests: rq.h.limiter.Requests(netblock) + 1,
 	})
 	rq.h.limiter.Reset(group)
-
-	rq.line.LimitHit = hit.Window
-	rq.line.Offence = requestlog.OffenceLimit
 	rq.line.BanExpires = banExpires(ban)
 
 	return true

internal/proxy/client.go

@@ -1,6 +1,7 @@
 package proxy
 
 import (
+	"crypto/rand"
 	"net/http"
 	"net/netip"
 	"slices"
@@ -48,6 +49,33 @@ func clientAddress(
 	return client
 }
 
+// requestIDHeader carries the request's id, from traefik and to the app.
+const requestIDHeader = "X-Request-ID"
+
+// requestID is the request's id: the one a trusted proxy sent, or a new
+// random one. A peer outside the trusted proxies did not come through
+// traefik, so the id it sends is its own claim, and is replaced.
+func requestID(r *http.Request, peerTrusted bool) string {
+	id := r.Header.Get(requestIDHeader)
+	if !peerTrusted || id == "" {
+		id = rand.Text()
+	}
+
+	return id
+}
+
+// scheme is how the client reached traefik, as a trusted proxy says in
+// X-Forwarded-Proto, or otherwise http, the only scheme smallwebwaf
+// serves.
+func scheme(r *http.Request, peerTrusted bool) string {
+	proto := r.Header.Get("X-Forwarded-Proto")
+	if !peerTrusted || proto == "" {
+		return "http"
+	}
+
+	return proto
+}
+
 // ipv6GroupPrefix is the length of the IPv6 netblock that is one client.
 const ipv6GroupPrefix = 64
 

internal/proxy/client_test.go

@@ -14,10 +14,14 @@ const (
 	appHost = "app.example"
 	// client is the client's address, as a proxy names it.
 	client = "203.0.113.9"
-	// forwardedFor is the header that lists the client and its proxies.
-	forwardedFor = "X-Forwarded-For"
-	// secure is the scheme a client reached traefik with.
+	// forwardedFor is the header that lists the client and its proxies,
+	// and forwardedProto the one that gives the scheme the client used.
+	forwardedFor   = "X-Forwarded-For"
+	forwardedProto = "X-Forwarded-Proto"
+	// secure is the scheme a client reached traefik with, and plain the
+	// one smallwebwaf serves.
 	secure = "https"
+	plain  = "http"
 )
 
 // appHeaders is what the app tells about the headers it received.
@@ -65,13 +69,13 @@ func TestClientAddressAndForwardedHeaders(t *testing.T) {
 func clientAddressCases() []clientAddressCase {
 	trusted := map[string]string{trustedProxies: trustLocalhost}
 	forged := http.Header{
-		forwardedFor:        {client},
-		"X-Forwarded-Host":  {"forged.example"},
-		"X-Forwarded-Proto": {secure},
-		"X-Real-Ip":         {client},
+		forwardedFor:       {client},
+		"X-Forwarded-Host": {"forged.example"},
+		forwardedProto:     {secure},
+		"X-Real-Ip":        {client},
 	}
 	replaced := appHeaders{
-		ForwardedFor: localhost, ForwardedHost: appHost, ForwardedProto: "http",
+		ForwardedFor: localhost, ForwardedHost: appHost, ForwardedProto: plain,
 	}
 
 	return []clientAddressCase{{
@@ -87,10 +91,10 @@ func clientAddressCases() []clientAddressCase {
 			"outside the trusted proxies from the right",
 		env: trusted,
 		header: http.Header{
-			forwardedFor:        {"198.51.100.7, " + client + ", 10.0.0.2"},
-			"X-Forwarded-Host":  {appHost},
-			"X-Forwarded-Proto": {secure},
-			"X-Real-Ip":         {client},
+			forwardedFor:       {"198.51.100.7, " + client + ", 10.0.0.2"},
+			"X-Forwarded-Host": {appHost},
+			forwardedProto:     {secure},
+			"X-Real-Ip":        {client},
 		},
 		wantClient: client,
 		wantApp: appHeaders{
@@ -138,7 +142,7 @@ func requestWithHeaders(
 			Host:           r.Host,
 			ForwardedFor:   r.Header.Get(forwardedFor),
 			ForwardedHost:  r.Header.Get("X-Forwarded-Host"),
-			ForwardedProto: r.Header.Get("X-Forwarded-Proto"),
+			ForwardedProto: r.Header.Get(forwardedProto),
 			RealIP:         r.Header.Get("X-Real-IP"),
 		})
 	})

internal/proxy/health_test.go

@@ -21,14 +21,18 @@ func TestHealthEndpointIsAnsweredBeforeAnyCheck(t *testing.T) {
 	// the last one would have it refused.
 	addr, out := startProxy(t, app.URL, map[string]string{rateLimitPerMinute: "1"})
 
-	const healthChecks = 3
+	const (
+		healthChecks = 3
+		contentType  = "text/plain; charset=utf-8"
+	)
 
 	for range healthChecks {
 		got := get(t, addr, proxy.HealthPath)
 		wantStatus(t, got, http.StatusOK)
 
-		if string(got.body) != "ok\n" {
-			t.Errorf("health endpoint answered %q, want ok", got.body)
+		if string(got.body) != "ok\n" || got.header.Get("Content-Type") != contentType {
+			t.Errorf("health endpoint answered %q with Content-Type %q, want ok "+
+				"with %q", got.body, got.header.Get("Content-Type"), contentType)
 		}
 	}
 
@@ -37,6 +41,11 @@ func TestHealthEndpointIsAnsweredBeforeAnyCheck(t *testing.T) {
 	lines := out.requestLines(t, healthChecks+1)
 	for _, line := range lines[:healthChecks] {
 		wantLine(t, line, http.StatusOK, requestlog.ActionAdmin)
+
+		if line.ResponseContentType != contentType {
+			t.Errorf("health check's log line has response_content_type %q, "+
+				"want %q", line.ResponseContentType, contentType)
+		}
 	}
 
 	wantLine(t, lines[healthChecks], http.StatusOK, requestlog.ActionForward)

internal/proxy/observe_test.go

@@ -0,0 +1,202 @@
+package proxy_test
+
+import (
+	"bytes"
+	"io"
+	"net/http"
+	"net/netip"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"sneak.berlin/go/smallwebwaf/internal/bans"
+	"sneak.berlin/go/smallwebwaf/internal/proxy"
+	"sneak.berlin/go/smallwebwaf/internal/requestlog"
+)
+
+// observe is the value of SWWAF_MODE for observe mode.
+const observe = "observe"
+
+func TestObserveModeForwardsWhatEnforceModeRefuses(t *testing.T) {
+	t.Parallel()
+
+	const (
+		denied = "192.0.2.50" // in SWWAF_DENY_NETS
+		banned = otherClient  // under a ban read from bans.json
+	)
+
+	for _, tc := range []struct {
+		setting string // "" leaves SWWAF_MODE at its default
+		observe bool
+	}{
+		{"", false},
+		{"enforce", false},
+		{observe, true},
+	} {
+		t.Run(mode+"="+tc.setting, func(t *testing.T) {
+			t.Parallel()
+
+			geojsURL, _ := startGeoJS(t)
+			env := map[string]string{
+				rateLimitPerMinute: "1",
+				denyNets:           denied,
+				deniedCountries:    "kp",
+			}
+
+			if tc.setting != "" {
+				env[mode] = tc.setting
+			}
+
+			s, clk, server := startWithClock(t, geojsURL, env)
+			server.Ledger.Load([]bans.Ban{{
+				Netblock: netip.MustParsePrefix(banned + "/32"),
+				Start:    clk.Now(),
+				Expires:  clk.Now().Add(time.Hour),
+			}})
+
+			// fromDE's first request is within the limit of one a minute,
+			// and its second breaks it.
+			s.get(fromDE, http.StatusOK, requestlog.ActionForward)
+
+			for _, sent := range []struct{ from, refusal string }{
+				{denied, requestlog.ActionDenied},
+				{banned, requestlog.ActionBanned},
+				{fromKP, requestlog.ActionCountryDenied},
+				{fromDE, requestlog.ActionRateLimited},
+			} {
+				if !tc.observe {
+					line := s.get(sent.from, http.StatusForbidden, sent.refusal)
+					wantWouldAction(t, line, "")
+
+					continue
+				}
+
+				// Passed to the app, which answered it.
+				line := s.get(sent.from, http.StatusOK, requestlog.ActionForward)
+				wantWouldAction(t, line, sent.refusal)
+
+				if line.UpstreamStatus != http.StatusOK {
+					t.Errorf("log line has upstream_status %d, want 200",
+						line.UpstreamStatus)
+				}
+			}
+		})
+	}
+}
+
+func TestObserveModeMakesNoBanAndKeepsTheBansItHas(t *testing.T) {
+	t.Parallel()
+
+	s, clk, server := startWithClock(t, "", map[string]string{
+		mode:               observe,
+		rateLimitPerMinute: "1",
+	})
+	kept := bans.Ban{
+		Netblock: netip.MustParsePrefix(otherClient + "/32"),
+		Start:    clk.Now(),
+		Expires:  clk.Now().Add(time.Hour),
+	}
+	server.Ledger.Load([]bans.Ban{kept})
+
+	// No ban sets client's counters back to zero, so each request after
+	// the first breaks the limit of one a minute.
+	s.get(client, http.StatusOK, requestlog.ActionForward)
+
+	for range 2 {
+		line := s.get(client, http.StatusOK, requestlog.ActionForward)
+		wantWouldAction(t, line, requestlog.ActionRateLimited)
+
+		if line.LimitHit != minute || line.Offence != requestlog.OffenceLimit ||
+			line.BanExpires != "" {
+			t.Errorf("log line has limit_hit %q, offence %q and ban_expires %q, "+
+				"want minute, limit and none", line.LimitHit, line.Offence,
+				line.BanExpires)
+		}
+	}
+
+	// The ban read from bans.json refuses nothing, and so counts no
+	// refusal in its notes, but is kept.
+	line := s.get(otherClient, http.StatusOK, requestlog.ActionForward)
+	wantWouldAction(t, line, requestlog.ActionBanned)
+
+	if line.BanExpires != requestlog.FormatTime(kept.Expires) {
+		t.Errorf("log line has ban_expires %q, want %s", line.BanExpires,
+			requestlog.FormatTime(kept.Expires))
+	}
+
+	got := server.Ledger.Snapshot()
+	if len(got) != 1 || got[0] != kept {
+		t.Errorf("bans\n%+v\nwant only\n%+v", got, kept)
+	}
+}
+
+func TestObserveModeKeepsTheSizeLimitsAndTheToken(t *testing.T) {
+	t.Parallel()
+
+	const denied = "192.0.2.50" // in SWWAF_DENY_NETS
+
+	var calls atomic.Int32
+
+	app := startApp(t, func(w http.ResponseWriter, _ *http.Request) {
+		calls.Add(1)
+		answerWithSize(w, 2*sizeLimit, true)
+	})
+	addr, out := startProxy(t, app.URL, map[string]string{
+		mode:             observe,
+		trustedProxies:   trustLocalhost,
+		denyNets:         denied,
+		requestMaxBytes:  sizeLimitSetting,
+		responseMaxBytes: sizeLimitSetting,
+		metricsToken:     token,
+	})
+
+	// SWWAF_DENY_NETS would refuse each request; instead a size limit or
+	// the missing token does.
+	for i, tc := range []struct {
+		method, path string
+		body         io.Reader
+		status       int
+		action       string
+	}{
+		{
+			http.MethodPost, "/upload", bytes.NewReader(make([]byte, 2*sizeLimit)),
+			http.StatusRequestEntityTooLarge, requestlog.ActionTooLarge,
+		},
+		{
+			http.MethodGet, "/download", http.NoBody,
+			http.StatusBadGateway, requestlog.ActionTooLarge,
+		},
+		{
+			http.MethodGet, proxy.MetricsPath, http.NoBody,
+			http.StatusUnauthorized, requestlog.ActionAdmin,
+		},
+	} {
+		req := newRequest(t, tc.method, addr, tc.path, tc.body)
+		req.Header.Set(forwardedFor, denied)
+		wantStatus(t, do(t, req), tc.status)
+
+		line := out.requestLines(t, i+1)[i]
+		wantLine(t, line, tc.status, tc.action)
+		wantWouldAction(t, line, requestlog.ActionDenied)
+	}
+
+	// The upload was refused before it reached the app.
+	if calls.Load() != 1 {
+		t.Errorf("the app was called %d times, want once", calls.Load())
+	}
+}
+
+// wantWouldAction checks the request log line's would_action, and that a
+// line that should have none has no such field.
+func wantWouldAction(t *testing.T, line logLine, want string) {
+	t.Helper()
+
+	got, present := line.fields["would_action"]
+
+	switch {
+	case want == "" && present:
+		t.Errorf("log line has would_action %v, want none", got)
+	case want != "" && got != want:
+		t.Errorf("log line has would_action %v, want %s", got, want)
+	}
+}

internal/proxy/passthrough_test.go

@@ -6,6 +6,8 @@ import (
 	"errors"
 	"io"
 	"net/http"
+	"os"
+	"reflect"
 	"slices"
 	"strings"
 	"sync/atomic"
@@ -14,6 +16,7 @@ import (
 
 	"sneak.berlin/go/smallwebwaf/internal/config"
 	"sneak.berlin/go/smallwebwaf/internal/proxy"
+	"sneak.berlin/go/smallwebwaf/internal/ratelimit"
 	"sneak.berlin/go/smallwebwaf/internal/requestlog"
 )
 
@@ -115,27 +118,34 @@ func wantAnswer(t *testing.T, got answer, body []byte) {
 	}
 }
 
-// wantRequestFields checks the log line's fields about the request.
+// wantRequestFields checks the log line's fields about the request. Its
+// time, its id and its timings are checked only for being there.
 func wantRequestFields(t *testing.T, line logLine, host string, sent, received int) {
 	t.Helper()
 
-	want := requestlog.Line{
-		Type: "request", Time: line.Time, ClientIP: localhost, PeerIP: localhost,
-		Method: http.MethodPatch, Host: host, Path: rawPath, Query: rawQuery,
-		Protocol: "HTTP/1.1", Status: http.StatusTeapot,
-		UpstreamStatus: http.StatusTeapot, RequestBytes: int64(sent),
+	hostname, _ := os.Hostname()
+
+	want := withTimings(line, requestlog.Line{
+		Type: requestType, Time: line.Time, Instance: hostname,
+		ClientIP: localhost, Method: http.MethodPatch, Scheme: plain, Host: host,
+		Path: rawPath, Query: rawQuery, Protocol: protocol,
+		Status: http.StatusTeapot, RequestBytes: int64(sent),
 		ResponseBytes: int64(received), UserAgent: "test-agent",
-		Action: requestlog.ActionForward, DurationTotal: line.DurationTotal,
-		DurationUpstreamTotal: line.DurationUpstreamTotal,
-	}
-	if line.Line != want {
+		RequestID: line.RequestID, PeerIP: localhost, ClientGroup: localhost + "/32",
+		ContentLength: int64(sent), ResponseContentType: "text/plain; charset=utf-8",
+		UpstreamStatus: http.StatusTeapot, Action: requestlog.ActionForward,
+		Counts: ratelimit.Counts{Minute: 1, Hour: 1, Day: 1},
+	})
+	if !reflect.DeepEqual(line.Line, want) {
 		t.Errorf("log line\n%+v\nwant\n%+v", line.Line, want)
 	}
 
 	_, err := time.Parse(time.RFC3339, line.Time)
-	if err != nil || line.DurationTotal <= 0 || line.DurationUpstreamTotal <= 0 {
-		t.Errorf("log line has time %q and durations %v and %v",
-			line.Time, line.DurationTotal, line.DurationUpstreamTotal)
+	if err != nil || line.RequestID == "" || line.DurationTotal <= 0 ||
+		line.DurationUpstreamTotal == nil || *line.DurationUpstreamTotal <= 0 {
+		t.Errorf("log line has time %q, request_id %q and durations %v and %v",
+			line.Time, line.RequestID, line.DurationTotal,
+			line.fields["duration_upstream_total"])
 	}
 }
 
@@ -371,8 +381,13 @@ func TestAnswers502WhenTheAppCannotBeReached(t *testing.T) {
 	addr, out := startProxy(t, "http://"+localhost+":1", nil)
 
 	wantStatus(t, get(t, addr, "/"), http.StatusBadGateway)
-	wantLine(t, out.requestLine(t), http.StatusBadGateway,
-		requestlog.ActionUpstreamError)
+
+	line := out.requestLine(t)
+	wantLine(t, line, http.StatusBadGateway, requestlog.ActionUpstreamError)
+
+	// There never was a connection to the app, nor an answer from it.
+	wantTimings(t, line, "duration_total", "duration_checks",
+		"duration_upstream_total")
 
 	logged := slices.ContainsFunc(out.lines(t), func(line map[string]any) bool {
 		return line["type"] == "process" && line["msg"] == "request to the app failed"

internal/proxy/proxy.go

@@ -160,6 +160,9 @@ func (h *handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	// a health checker is never refused. It does not ask the app.
 	if r.Method == http.MethodGet && r.URL.Path == HealthPath {
 		rq.line.Action = requestlog.ActionAdmin
+		// Set here rather than left to Go's server, which would set it only
+		// after the log line has taken the response's headers.
+		rq.out.Header().Set("Content-Type", "text/plain; charset=utf-8")
 		_, _ = io.WriteString(rq.out, "ok\n")
 
 		return
@@ -169,6 +172,8 @@ func (h *handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	defer rq.addToHistory()
 
 	refused := rq.check(r.Context())
+	rq.checked = time.Now()
+
 	if refused != nil {
 		rq.answer(*refused)
 

internal/proxy/proxy_test.go

@@ -35,6 +35,10 @@ const (
 	// localhost is where every test server listens, and so the address
 	// smallwebwaf sees each test's requests come from.
 	localhost = "127.0.0.1"
+	// requestType is the type that marks a request log line.
+	requestType = "request"
+	// protocol is the protocol of every test's requests.
+	protocol = "HTTP/1.1"
 )
 
 // shortTimeoutSetting is shortTimeout as a setting's value.
@@ -50,6 +54,7 @@ const (
 	clientResponseTimeout   = "SWWAF_CLIENT_RESPONSE_TIMEOUT"
 	upstreamRequestTimeout  = "SWWAF_UPSTREAM_REQUEST_TIMEOUT"
 	upstreamResponseTimeout = "SWWAF_UPSTREAM_RESPONSE_TIMEOUT"
+	mode                    = "SWWAF_MODE"
 	requestMaxBytes         = "SWWAF_REQUEST_MAX_BYTES"
 	responseMaxBytes        = "SWWAF_RESPONSE_MAX_BYTES"
 	trustedProxies          = "SWWAF_TRUSTED_PROXIES"
@@ -58,6 +63,7 @@ const (
 	denyNets                = "SWWAF_DENY_NETS"
 	rateLimitPerMinute      = "SWWAF_RATE_LIMIT_PER_MINUTE"
 	rateLimitPerDay         = "SWWAF_RATE_LIMIT_PER_DAY"
+	rateLimitExemptPaths    = "SWWAF_RATE_LIMIT_EXEMPT_PATHS"
 	deniedCountries         = "SWWAF_DENIED_COUNTRIES"
 	allowedCountries        = "SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES"
 	banResponse             = "SWWAF_BAN_RESPONSE"
@@ -66,6 +72,8 @@ const (
 	maxBanDuration          = "SWWAF_MAX_BAN_DURATION"
 	maxBans                 = "SWWAF_MAX_BANS"
 	banScopeV4Prefix        = "SWWAF_BAN_SCOPE_V4_PREFIX"
+	instanceName            = "SWWAF_INSTANCE_NAME"
+	logRequestHeaders       = "SWWAF_LOG_REQUEST_HEADERS"
 )
 
 // output collects what smallwebwaf writes on stdout.
@@ -82,6 +90,14 @@ func (o *output) Write(p []byte) (int, error) {
 	return o.buf.Write(p)
 }
 
+// text returns everything written so far.
+func (o *output) text() string {
+	o.mu.Lock()
+	defer o.mu.Unlock()
+
+	return o.buf.String()
+}
+
 // lines returns every line written so far, decoded.
 func (o *output) lines(t *testing.T) []map[string]any {
 	t.Helper()
@@ -121,7 +137,7 @@ func (o *output) requestLines(t *testing.T, count int) []logLine {
 		var found []logLine
 
 		for _, fields := range o.lines(t) {
-			if fields["type"] == "request" {
+			if fields["type"] == requestType {
 				found = append(found, decodeLine(t, fields))
 			}
 		}

internal/proxy/ratelimits_test.go

@@ -5,6 +5,8 @@ import (
 	"sync/atomic"
 	"testing"
 
+	"sneak.berlin/go/smallwebwaf/internal/lookup"
+	"sneak.berlin/go/smallwebwaf/internal/ratelimit"
 	"sneak.berlin/go/smallwebwaf/internal/requestlog"
 )
 
@@ -69,3 +71,89 @@ func TestRateLimitRefusesBeforeTheApp(t *testing.T) {
 		t.Errorf("the app was called %d times, want 4", calls.Load())
 	}
 }
+
+func TestRateLimitExemptPathsAreNeitherCountedNorRefused(t *testing.T) {
+	t.Parallel()
+
+	const denied = "192.0.2.50" // in SWWAF_DENY_NETS
+
+	s, _, server := startWithClock(t, "", map[string]string{
+		rateLimitPerMinute:   "1",
+		rateLimitExemptPaths: "/assets/,/favicon.ico",
+		denyNets:             denied,
+		deniedCountries:      "kp",
+	})
+
+	// The answers are kept before the requests, so that none waits for
+	// GeoJS.
+	server.GeoJS.Load([]lookup.Answer{
+		keptAnswer(client, "DE"), keptAnswer(fromKP, "KP"),
+	})
+
+	// With a limit of one request a minute, the requests for paths under a
+	// prefix are not counted, so client's first request for / is within
+	// the limit; and once client has reached it, they are not refused.
+	s.request(client, "/assets/app.js", http.StatusOK, requestlog.ActionForward)
+	s.request(client, "/favicon.ico?v=2", http.StatusOK, requestlog.ActionForward)
+	s.get(client, http.StatusOK, requestlog.ActionForward)
+
+	line := s.request(client, "/assets/app.js", http.StatusOK, requestlog.ActionForward)
+	if line.LimitHit != "" || line.Counts != (ratelimit.Counts{}) {
+		t.Errorf("log line has limit_hit %q and counts %+v, want neither",
+			line.LimitHit, line.Counts)
+	}
+
+	// A path outside every prefix is counted: /assets is not under
+	// /assets/, and breaks the limit.
+	s.request(client, "/assets", http.StatusForbidden, requestlog.ActionRateLimited)
+
+	// A ban, SWWAF_DENY_NETS and the country lists still refuse a path
+	// under a prefix.
+	s.request(client, "/assets/app.js", http.StatusForbidden, requestlog.ActionBanned)
+	s.request(denied, "/assets/app.js", http.StatusForbidden, requestlog.ActionDenied)
+	s.request(fromKP, "/assets/app.js",
+		http.StatusForbidden, requestlog.ActionCountryDenied)
+}
+
+func TestRateLimitCountsPathsThatAreNotExempt(t *testing.T) {
+	t.Parallel()
+
+	for _, sent := range []string{
+		// A prefix matches only at the start of the path.
+		"/static/assets/app.js",
+		// A prefix matches the path as sent: a router that matches the
+		// path as received does not take /%61ssets/x for a path under
+		// /assets/.
+		"/%61ssets/x",
+		// .. once percent-decoded: an app may act on these as /login, the
+		// last as a path under /sneak/app/ or as /assets/x.
+		"/assets/../login",
+		"/assets/%2e%2e/login",
+		"/assets/..%2Flogin",
+		"/assets/..;/login",
+		"/sneak/app/src/branch/main/..%2F..%2F..%2F..%2F..%2F..%2Fassets/x",
+		// Not under /assets/ as sent: Go's router takes /assets%2Fx for one
+		// path segment, not a path under /assets/.
+		"/assets%2Fx",
+		"/assets%2fx",
+		// Under /assets/ as sent, but holding an encoded slash, in either
+		// case, or a backslash: never exempt, whatever the prefix.
+		"/assets/x%2Fy",
+		"/assets/x%2fy",
+		`/assets/x\y`,
+	} {
+		t.Run(sent, func(t *testing.T) {
+			t.Parallel()
+
+			s, _, _ := startWithClock(t, "", map[string]string{
+				rateLimitPerMinute:   "1",
+				rateLimitExemptPaths: "/assets/",
+			})
+
+			// Counted, the second request breaks the limit of one request
+			// a minute.
+			s.request(client, sent, http.StatusOK, requestlog.ActionForward)
+			s.request(client, sent, http.StatusForbidden, requestlog.ActionRateLimited)
+		})
+	}
+}

internal/proxy/request.go

@@ -7,7 +7,10 @@ import (
 	"net/http/httptrace"
 	"net/http/httputil"
 	"net/netip"
+	"net/url"
 	"os"
+	"slices"
+	"strings"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -46,7 +49,9 @@ type request struct {
 	peer        netip.Addr
 	peerTrusted bool
 	start       time.Time
-	// upstreamStart is when the request was handed to the app.
+	// checked is when the checks were done, and upstreamStart when the
+	// request was handed to the app.
+	checked       time.Time
 	upstreamStart time.Time
 	// cancel ends the request to the app.
 	cancel context.CancelFunc
@@ -56,26 +61,34 @@ type request struct {
 	complete bool
 
 	// mu guards what follows. The timeouts run on goroutines of their
-	// own, and the transport starts and stops them from its own; once
-	// timersStopped is set, none of them acts any more.
+	// own, and the transport starts and stops them, and notes the times
+	// below, from its own; once timersStopped is set, none of the timeouts
+	// acts any more.
 	mu                    sync.Mutex
 	timersStopped         bool
 	clientRequestTimer    *time.Timer
 	upstreamRequestTimer  *time.Timer
 	upstreamResponseTimer *time.Timer
-	// requestSent is when the app had been sent the whole request.
-	requestSent time.Time
+	// connected is when there was a connection to the app, requestSent
+	// when the app had been sent the whole request, and answerStarted
+	// when the first byte of its answer arrived.
+	connected     time.Time
+	requestSent   time.Time
+	answerStarted time.Time
 }
 
 // newRequest starts handling r: it notes the time, counts the request as
-// under way, and works out the client.
+// under way, works out the client, and starts the log line with what is
+// known of the request.
 func (h *handler) newRequest(w http.ResponseWriter, r *http.Request) *request {
 	h.metrics.RequestStarted()
 
 	start := time.Now()
 	peer := peerAddress(r)
 	trusted := h.config.TrustedProxies
-	client := clientAddress(peer, r.Header.Values("X-Forwarded-For"), trusted)
+	peerTrusted := isInside(peer, trusted)
+	forwardedFor := r.Header.Values("X-Forwarded-For")
+	client := clientAddress(peer, forwardedFor, trusted)
 
 	rq := &request{
 		h:           h,
@@ -84,22 +97,37 @@ func (h *handler) newRequest(w http.ResponseWriter, r *http.Request) *request {
 		out:         &responseWriter{ResponseWriter: w},
 		client:      client,
 		peer:        peer,
-		peerTrusted: isInside(peer, trusted),
+		peerTrusted: peerTrusted,
 		start:       start,
 		line: requestlog.Line{
-			Time:      requestlog.FormatTime(start),
-			ClientIP:  client.String(),
-			PeerIP:    peer.String(),
-			Method:    r.Method,
-			Host:      r.Host,
-			Path:      r.URL.EscapedPath(),
-			Query:     r.URL.RawQuery,
-			Protocol:  r.Proto,
-			Referer:   r.Referer(),
-			UserAgent: r.UserAgent(),
-			Action:    requestlog.ActionForward,
+			Time:             requestlog.FormatTime(start),
+			Instance:         h.config.InstanceName,
+			ClientIP:         client.String(),
+			Method:           r.Method,
+			Scheme:           scheme(r, peerTrusted),
+			Host:             r.Host,
+			Path:             r.URL.EscapedPath(),
+			Query:            r.URL.RawQuery,
+			Protocol:         r.Proto,
+			Referer:          r.Referer(),
+			UserAgent:        r.UserAgent(),
+			RequestID:        requestID(r, peerTrusted),
+			PeerIP:           peer.String(),
+			ForwardedFor:     strings.Join(forwardedFor, ", "),
+			ClientGroup:      clientGroup(client).String(),
+			ContentType:      r.Header.Get("Content-Type"),
+			RequestHeaders:   requestHeaders(r, h.config.LogRequestHeaders),
+			HasAuthorization: len(r.Header.Values("Authorization")) > 0,
+			HasCookie:        len(r.Header.Values("Cookie")) > 0,
+			Action:           requestlog.ActionForward,
 		},
 	}
+
+	// A length of -1 is a body whose length was not announced.
+	if r.ContentLength > 0 {
+		rq.line.ContentLength = r.ContentLength
+	}
+
 	if r.Body != http.NoBody {
 		rq.body = &requestBody{body: limitBody(r.Body, h.config.RequestMaxBytes), rq: rq}
 	}
@@ -107,40 +135,47 @@ func (h *handler) newRequest(w http.ResponseWriter, r *http.Request) *request {
 	return rq
 }
 
+// requestHeaders returns the headers of r that names lists, by name in
+// lower case, each with its values joined by ", ". Authorization, Cookie
+// and Set-Cookie are never among them, whatever names says.
+func requestHeaders(r *http.Request, names []string) map[string]string {
+	headers := map[string]string{}
+
+	for _, name := range names {
+		switch name {
+		case "authorization", "cookie", "set-cookie":
+			continue
+		}
+
+		values := r.Header.Values(name)
+		if len(values) > 0 {
+			headers[name] = strings.Join(values, ", ")
+		}
+	}
+
+	return headers
+}
+
 // check is the one place where a request can be refused once its client
 // is known, before its body is read or anything reaches the app. It
-// returns nil to let the request through. A client in SWWAF_ALLOW_NETS
-// skips every check but the size limit. For any other client,
-// SWWAF_DENY_NETS comes first, then a ban on its netblock, so that a
-// client either refuses is not looked up, and then the country lists; a
-// request any of them refuses is not counted for the rate limits. Then
-// come the rate limits, unless the client is in
-// SWWAF_RATE_LIMIT_EXEMPT_NETS, so that every other request is counted,
-// one refused for its size too. Every refusal but the size limit's is
-// answered with SWWAF_BAN_RESPONSE. ctx is the request's own context.
+// returns nil to let the request through. The checks of checkClient come
+// first, answered with SWWAF_BAN_RESPONSE, and then the size limit, so
+// that a request the rate limits count is counted even when it is
+// refused for its size. In observe mode a request checkClient refuses
+// goes on to the size limit like any other. ctx is the request's own
+// context.
 func (rq *request) check(ctx context.Context) *refusal {
-	cfg := rq.h.config
-	allowed := isInside(rq.client, cfg.AllowNets)
-	exempt := isInside(rq.client, cfg.RateLimitExemptNets)
-	now := rq.h.now()
+	action := rq.checkClient(ctx)
+	if action != "" {
+		if !rq.h.config.Observe {
+			return rq.banResponse(action)
+		}
 
-	if !allowed && isInside(rq.client, cfg.DenyNets) {
-		return rq.banResponse(requestlog.ActionDenied)
+		// The log line names what enforce mode would have done.
+		rq.line.WouldAction = action
 	}
 
-	if !allowed && rq.banned(now) {
-		return rq.banResponse(requestlog.ActionBanned)
-	}
-
-	if !allowed && rq.countryDenied(ctx) {
-		return rq.banResponse(requestlog.ActionCountryDenied)
-	}
-
-	if !allowed && !exempt && rq.limitBroken(now) {
-		return rq.banResponse(requestlog.ActionRateLimited)
-	}
-
-	maxBytes := cfg.RequestMaxBytes
+	maxBytes := rq.h.config.RequestMaxBytes
 	if maxBytes > 0 && rq.in.ContentLength > maxBytes {
 		return &refusal{
 			status: http.StatusRequestEntityTooLarge,
@@ -152,6 +187,69 @@ func (rq *request) check(ctx context.Context) *refusal {
 	return nil
 }
 
+// checkClient runs the checks on the request's client, and returns the
+// action of the first that refuses the request, or "" when none does. A
+// client in SWWAF_ALLOW_NETS skips them. For any other client,
+// SWWAF_DENY_NETS comes first, then a ban on its netblock, so that a
+// client either refuses is not looked up, and then the country lists; a
+// request any of them refuses is not counted for the rate limits. Then
+// come the rate limits, unless the client is in
+// SWWAF_RATE_LIMIT_EXEMPT_NETS or the request's path is exempt under
+// SWWAF_RATE_LIMIT_EXEMPT_PATHS, so that every other request is counted.
+// ctx is the request's own context.
+func (rq *request) checkClient(ctx context.Context) string {
+	cfg := rq.h.config
+	if isInside(rq.client, cfg.AllowNets) {
+		return ""
+	}
+
+	now := rq.h.now()
+
+	if isInside(rq.client, cfg.DenyNets) {
+		return requestlog.ActionDenied
+	}
+
+	if rq.banned(now) {
+		return requestlog.ActionBanned
+	}
+
+	if rq.countryDenied(ctx) {
+		return requestlog.ActionCountryDenied
+	}
+
+	exempt := isInside(rq.client, cfg.RateLimitExemptNets) ||
+		pathExempt(rq.in.URL, cfg.RateLimitExemptPaths)
+	if !exempt && rq.limitBroken(now) {
+		return requestlog.ActionRateLimited
+	}
+
+	return ""
+}
+
+// pathExempt reports whether the rate limits leave out a request for u
+// because of SWWAF_RATE_LIMIT_EXEMPT_PATHS: whether its path as sent, the
+// path the app receives, not percent-decoded, starts with one of
+// prefixes, so that /%61ssets/x is not under /assets/ for an app whose
+// router matches the path as received. A request whose decoded path
+// contains .. anywhere or a backslash, or whose path as sent holds an
+// encoded slash (%2F or %2f), never is, since an app may act on it as a
+// path outside every prefix: /assets/..%2Flogin as /login, or /assets%2Fx
+// as one path segment, as Go's router does.
+func pathExempt(u *url.URL, prefixes []string) bool {
+	decoded := u.Path
+	// EscapedPath is the path as the app receives it, not decoded.
+	sent := u.EscapedPath()
+
+	if strings.Contains(decoded, "..") || strings.Contains(decoded, `\`) ||
+		strings.Contains(strings.ToLower(sent), "%2f") {
+		return false
+	}
+
+	return slices.ContainsFunc(prefixes, func(prefix string) bool {
+		return strings.HasPrefix(sent, prefix)
+	})
+}
+
 // forward passes the request to the app and the app's answer back. ctx
 // is the request's own context.
 func (rq *request) forward(ctx context.Context) {
@@ -160,7 +258,9 @@ func (rq *request) forward(ctx context.Context) {
 
 	rq.cancel = cancel
 	ctx = httptrace.WithClientTrace(ctx, &httptrace.ClientTrace{
-		WroteRequest: rq.wroteRequest,
+		GotConn:              rq.gotConn,
+		WroteRequest:         rq.wroteRequest,
+		GotFirstResponseByte: rq.gotFirstResponseByte,
 	})
 
 	out := rq.in.WithContext(ctx)
@@ -183,7 +283,8 @@ func (rq *request) forward(ctx context.Context) {
 }
 
 // rewrite makes the request the app receives: the client's request,
-// unchanged, sent to SWWAF_UPSTREAM_URL, with the forwarded headers set.
+// unchanged, sent to SWWAF_UPSTREAM_URL, with the forwarded headers and
+// the request's id set.
 func (rq *request) rewrite(pr *httputil.ProxyRequest) {
 	upstream := rq.h.config.UpstreamURL
 	pr.Out.URL.Scheme = upstream.Scheme
@@ -192,6 +293,7 @@ func (rq *request) rewrite(pr *httputil.ProxyRequest) {
 	// the query as the client sent it.
 	pr.Out.URL.RawQuery = pr.In.URL.RawQuery
 	setForwardedHeaders(pr.In, pr.Out, rq.peer, rq.peerTrusted)
+	pr.Out.Header.Set(requestIDHeader, rq.line.RequestID)
 }
 
 // modifyResponse looks at the app's answer before ReverseProxy passes it
@@ -205,6 +307,7 @@ func (rq *request) modifyResponse(res *http.Response) error {
 		// connection it takes over, not through rq.out.
 		rq.stopTimers()
 		rq.out.status = res.StatusCode
+		rq.line.Websocket = true
 
 		return nil
 	}
@@ -300,6 +403,10 @@ func (rq *request) finish() {
 	line := &rq.line
 	line.Status = rq.out.status
 	line.ResponseBytes = rq.out.bytes
+	header := rq.out.Header()
+	line.ResponseContentType = header.Get("Content-Type")
+	line.CacheControl = header.Get("Cache-Control")
+	line.Location = header.Get("Location")
 
 	if rq.body != nil {
 		line.RequestBytes = rq.body.bytes.Load()
@@ -324,12 +431,18 @@ func (rq *request) finish() {
 	now := time.Now()
 	duration := now.Sub(rq.start)
 	line.DurationTotal = requestlog.Milliseconds(duration)
+	line.DurationChecks = timing(rq.start, rq.checked)
 
 	var upstreamDuration time.Duration
 
 	if !rq.upstreamStart.IsZero() {
 		upstreamDuration = now.Sub(rq.upstreamStart)
-		line.DurationUpstreamTotal = requestlog.Milliseconds(upstreamDuration)
+		line.DurationUpstreamTotal = new(requestlog.Milliseconds(upstreamDuration))
+
+		rq.mu.Lock()
+		line.DurationUpstreamConnect = timing(rq.upstreamStart, rq.connected)
+		line.DurationUpstreamFirstByte = timing(rq.upstreamStart, rq.answerStarted)
+		rq.mu.Unlock()
 	}
 
 	// Counted before the log line is written, so that the metrics count
@@ -342,6 +455,17 @@ func (rq *request) finish() {
 	}
 }
 
+// timing is the time from start to end in milliseconds, for one of the
+// log line's timings, or nil when end is zero: what it times never
+// happened.
+func timing(start, end time.Time) *float64 {
+	if end.IsZero() {
+		return nil
+	}
+
+	return new(requestlog.Milliseconds(end.Sub(start)))
+}
+
 // addToHistory adds the request, which has ended, to its client's
 // history.
 func (rq *request) addToHistory() {
@@ -451,6 +575,24 @@ func (rq *request) bodyReceived() {
 	stopTimer(rq.clientRequestTimer)
 }
 
+// gotConn is called once there is a connection to the app, a new one or
+// one kept open from an earlier request.
+func (rq *request) gotConn(httptrace.GotConnInfo) {
+	rq.mu.Lock()
+	defer rq.mu.Unlock()
+
+	rq.connected = time.Now()
+}
+
+// gotFirstResponseByte is called once the first byte of the app's answer
+// has arrived.
+func (rq *request) gotFirstResponseByte() {
+	rq.mu.Lock()
+	defer rq.mu.Unlock()
+
+	rq.answerStarted = time.Now()
+}
+
 // wroteRequest is called once the app has been sent the whole request:
 // the request timeouts end and SWWAF_UPSTREAM_RESPONSE_TIMEOUT starts.
 func (rq *request) wroteRequest(info httptrace.WroteRequestInfo) {

internal/proxy/requestlog_test.go

@@ -0,0 +1,368 @@
+package proxy_test
+
+import (
+	"io"
+	"maps"
+	"math"
+	"net/http"
+	"reflect"
+	"slices"
+	"strings"
+	"testing"
+	"time"
+
+	"sneak.berlin/go/smallwebwaf/internal/proxy"
+	"sneak.berlin/go/smallwebwaf/internal/ratelimit"
+	"sneak.berlin/go/smallwebwaf/internal/requestlog"
+)
+
+const (
+	// requestIDHeader carries the request's id.
+	requestIDHeader = "X-Request-ID"
+	// instance is the SWWAF_INSTANCE_NAME a test sets.
+	instance = "fsn1app1/gitea"
+	// ipv6Client is a client on IPv6, and ipv6Group the netblock the rate
+	// limits count it as.
+	ipv6Client = "2001:db8::7"
+	ipv6Group  = "2001:db8::/64"
+)
+
+func TestLogLineHasEachFieldWhereItApplies(t *testing.T) {
+	t.Parallel()
+
+	received := make(chan string, 2) // the request ids the app received
+	app := startApp(t, func(w http.ResponseWriter, r *http.Request) {
+		received <- r.Header.Get(requestIDHeader)
+
+		_, _ = io.Copy(io.Discard, r.Body)
+
+		if r.URL.Path != "/full" {
+			w.WriteHeader(http.StatusNoContent)
+
+			return
+		}
+
+		w.Header().Set("Content-Type", "text/html")
+		w.Header().Set("Cache-Control", "no-store")
+		w.Header().Set("Location", "/elsewhere")
+		w.WriteHeader(http.StatusFound)
+		_, _ = io.WriteString(w, "moved")
+	})
+	addr, out := startProxy(t, app.URL, map[string]string{
+		trustedProxies:      trustLocalhost,
+		rateLimitExemptNets: localhost,
+		instanceName:        instance,
+		logRequestHeaders:   "Accept,x-custom,Authorization,cookie,SET-COOKIE",
+	})
+
+	// This request comes from ipv6Client through a trusted proxy, with a
+	// body and each header the log line looks at, and is answered with a
+	// redirect.
+	conn := dial(t, addr)
+	send(t, conn, "POST /full HTTP/1.1\r\nHost: "+appHost+"\r\n"+
+		forwardedFor+": 198.51.100.7, "+ipv6Client+"\r\n"+
+		forwardedProto+": "+secure+"\r\n"+requestIDHeader+": from-traefik\r\n"+
+		"Content-Type: application/x-www-form-urlencoded\r\nContent-Length: 3\r\n"+
+		"Accept: text/html\r\nX-Custom: one\r\nX-Custom: two\r\n"+
+		"Authorization: Bearer secret-token\r\nCookie: session=secret-cookie\r\n"+
+		"Set-Cookie: secret-set-cookie\r\n\r\na=b")
+	wantStatus(t, readResponse(t, conn), http.StatusFound)
+
+	// A request's log line can come after its answer: each is waited for
+	// before the next request, so that the lines are in order.
+	full := out.requestLines(t, 1)[0]
+
+	// This one comes from 127.0.0.1, which the rate limits do not count,
+	// with a body of 4 bytes whose length it does not announce, so that its
+	// request_bytes is not its content_length, and no header the log line
+	// looks at, and is answered with 204 and no header.
+	conn = dial(t, addr)
+	send(t, conn, "POST /bare HTTP/1.1\r\nHost: "+appHost+"\r\n"+
+		"Transfer-Encoding: chunked\r\n\r\n4\r\nbody\r\n0\r\n\r\n")
+	wantStatus(t, readResponse(t, conn), http.StatusNoContent)
+
+	bare := out.requestLines(t, 2)[1]
+
+	wantFullLine(t, full)
+	wantBareLine(t, bare)
+
+	for _, line := range []logLine{full, bare} {
+		got := <-received
+		if got != line.RequestID {
+			t.Errorf("the app received request id %q, the log line has %q",
+				got, line.RequestID)
+		}
+	}
+
+	if strings.Contains(out.text(), "secret") {
+		t.Errorf("a value of Authorization, Cookie or Set-Cookie is logged:\n%s",
+			out.text())
+	}
+}
+
+// wantFullLine checks the log line of the request with every header the
+// line looks at. Its timings are checked by TestTimingsAreInOrder.
+func wantFullLine(t *testing.T, line logLine) {
+	t.Helper()
+
+	headers := map[string]string{"accept": "text/html", "x-custom": "one, two"}
+
+	want := withTimings(line, requestlog.Line{
+		Type: requestType, Time: line.Time, Instance: instance,
+		ClientIP: ipv6Client, Method: http.MethodPost, Scheme: secure,
+		Host: appHost, Path: "/full", Protocol: protocol,
+		Status: http.StatusFound, RequestBytes: 3, ResponseBytes: 5,
+		RequestID: "from-traefik", PeerIP: localhost,
+		ForwardedFor: "198.51.100.7, " + ipv6Client, ClientGroup: ipv6Group,
+		ContentType: "application/x-www-form-urlencoded", ContentLength: 3,
+		RequestHeaders: headers, HasAuthorization: true, HasCookie: true,
+		ResponseContentType: "text/html", UpstreamStatus: http.StatusFound,
+		CacheControl: "no-store", Location: "/elsewhere",
+		Action: requestlog.ActionForward,
+		Counts: ratelimit.Counts{Minute: 1, Hour: 1, Day: 1},
+	})
+	if !reflect.DeepEqual(line.Line, want) {
+		t.Errorf("log line\n%+v\nwant\n%+v", line.Line, want)
+	}
+}
+
+// wantBareLine checks the log line of the request with none of them, and
+// that the fields that do not apply to it are left out.
+func wantBareLine(t *testing.T, line logLine) {
+	t.Helper()
+
+	want := withTimings(line, requestlog.Line{
+		Type: requestType, Time: line.Time, Instance: instance,
+		ClientIP: localhost, Method: http.MethodPost, Scheme: plain,
+		Host: appHost, Path: "/bare", Protocol: protocol,
+		Status: http.StatusNoContent, RequestBytes: 4, RequestID: line.RequestID,
+		PeerIP: localhost, ClientGroup: localhost + "/32",
+		UpstreamStatus: http.StatusNoContent, Action: requestlog.ActionForward,
+	})
+	if !reflect.DeepEqual(line.Line, want) || line.RequestID == "" {
+		t.Errorf("log line\n%+v\nwant\n%+v, with a request id", line.Line, want)
+	}
+
+	for _, name := range []string{
+		"forwarded_for", "content_type", "content_length", "request_headers",
+		"has_authorization", "has_cookie", "websocket", "response_content_type",
+		"cache_control", "location", "counts",
+	} {
+		_, present := line.fields[name]
+		if present {
+			t.Errorf("log line has %s, which does not apply", name)
+		}
+	}
+}
+
+// withTimings returns want with the timings of line.
+func withTimings(line logLine, want requestlog.Line) requestlog.Line {
+	want.DurationTotal = line.DurationTotal
+	want.DurationChecks = line.DurationChecks
+	want.DurationUpstreamConnect = line.DurationUpstreamConnect
+	want.DurationUpstreamFirstByte = line.DurationUpstreamFirstByte
+	want.DurationUpstreamTotal = line.DurationUpstreamTotal
+
+	return want
+}
+
+func TestHasAuthorizationAndHasCookieEachComeFromTheirOwnHeader(t *testing.T) {
+	t.Parallel()
+
+	const hasAuthorization, hasCookie = "has_authorization", "has_cookie"
+
+	for _, tc := range []struct{ header, field, other string }{
+		{"Authorization", hasAuthorization, hasCookie},
+		{"Cookie", hasCookie, hasAuthorization},
+	} {
+		t.Run("only "+tc.header, func(t *testing.T) {
+			t.Parallel()
+
+			app := startApp(t, func(http.ResponseWriter, *http.Request) {})
+			addr, out := startProxy(t, app.URL, nil)
+
+			req := newRequest(t, http.MethodGet, addr, "/", http.NoBody)
+			req.Header.Set(tc.header, "secret")
+			wantStatus(t, do(t, req), http.StatusOK)
+
+			line := out.requestLine(t)
+
+			_, otherPresent := line.fields[tc.other]
+			if line.fields[tc.field] != true || otherPresent {
+				t.Errorf("log line has %s %v and %s %v, want true and none",
+					tc.field, line.fields[tc.field], tc.other, line.fields[tc.other])
+			}
+		})
+	}
+}
+
+func TestRequestIDAndSchemeComeOnlyFromATrustedProxy(t *testing.T) {
+	t.Parallel()
+
+	const sentID = "from-traefik"
+
+	sent := http.Header{requestIDHeader: {sentID}, forwardedProto: {secure}}
+	trusted := map[string]string{trustedProxies: trustLocalhost}
+
+	for _, tc := range []struct {
+		name   string
+		env    map[string]string
+		header http.Header
+		// wantID is the request id logged, "" for a new one.
+		wantID, wantScheme string
+	}{
+		{"a trusted proxy's are kept", trusted, sent, sentID, secure},
+		{"without them, the id is new and the scheme http", trusted, nil, "", plain},
+		{"another peer's are replaced", nil, sent, "", plain},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			received := make(chan string, 2)
+			app := startApp(t, func(_ http.ResponseWriter, r *http.Request) {
+				received <- r.Header.Get(requestIDHeader)
+			})
+			addr, out := startProxy(t, app.URL, tc.env)
+
+			// Two requests, so that two new ids can be told apart.
+			ids := make([]string, 0, 2)
+
+			for i := range 2 {
+				req := newRequest(t, http.MethodGet, addr, "/", http.NoBody)
+				maps.Copy(req.Header, tc.header)
+				wantStatus(t, do(t, req), http.StatusOK)
+
+				line := out.requestLines(t, i+1)[i]
+				ids = append(ids, line.RequestID)
+
+				got := <-received
+				if line.RequestID != got || line.Scheme != tc.wantScheme {
+					t.Errorf("log line has request_id %q and scheme %q, and the "+
+						"app received id %q; want the same id and scheme %q",
+						line.RequestID, line.Scheme, got, tc.wantScheme)
+				}
+			}
+
+			switch {
+			case tc.wantID != "" && (ids[0] != tc.wantID || ids[1] != tc.wantID):
+				t.Errorf("request ids %q, want %q", ids, tc.wantID)
+			case tc.wantID == "" && (slices.Contains(ids, sentID) ||
+				slices.Contains(ids, "") || ids[0] == ids[1]):
+				t.Errorf("request ids %q, want two new ones", ids)
+			}
+		})
+	}
+}
+
+func TestTimingsAreInOrder(t *testing.T) {
+	t.Parallel()
+
+	const denied = "192.0.2.50" // in SWWAF_DENY_NETS
+
+	app := startApp(t, func(w http.ResponseWriter, _ *http.Request) {
+		// The pauses set the times apart; a hold-up of the test only
+		// lengthens them.
+		time.Sleep(time.Millisecond)
+		w.WriteHeader(http.StatusOK)
+		_ = http.NewResponseController(w).Flush()
+
+		time.Sleep(time.Millisecond)
+
+		_, _ = io.WriteString(w, "done")
+	})
+	addr, out := startProxy(t, app.URL, map[string]string{
+		trustedProxies: trustLocalhost,
+		denyNets:       denied,
+	})
+
+	// Each log line is waited for before the next request, so that the
+	// lines are in order.
+	wantStatus(t, get(t, addr, "/"), http.StatusOK)
+	forwarded := out.requestLines(t, 1)[0]
+
+	req := newRequest(t, http.MethodGet, addr, "/", http.NoBody)
+	req.Header.Set(forwardedFor, denied)
+	wantStatus(t, do(t, req), http.StatusForbidden)
+	refused := out.requestLines(t, 2)[1]
+
+	wantStatus(t, get(t, addr, proxy.HealthPath), http.StatusOK)
+	health := out.requestLines(t, 3)[2]
+
+	// A request passed to the app has every timing; one refused, none of
+	// the app's; the health check, which runs no check, only the total.
+	wantTimings(t, forwarded, "duration_total", "duration_checks",
+		"duration_upstream_connect", "duration_upstream_first_byte",
+		"duration_upstream_total")
+	wantTimings(t, refused, "duration_total", "duration_checks")
+	wantTimings(t, health, "duration_total")
+
+	if t.Failed() {
+		return
+	}
+
+	// In whole microseconds, as they are logged, so that the sum below is
+	// exact.
+	total := microseconds(forwarded.DurationTotal)
+	checks := microseconds(*forwarded.DurationChecks)
+	connect := microseconds(*forwarded.DurationUpstreamConnect)
+	firstByte := microseconds(*forwarded.DurationUpstreamFirstByte)
+	upstream := microseconds(*forwarded.DurationUpstreamTotal)
+
+	// The checks end before the request is handed to the app, and the
+	// connection comes before the answer, which the app ends after a
+	// pause.
+	if checks+upstream > total || connect >= firstByte || firstByte >= upstream {
+		t.Errorf("timings in microseconds: total %d, checks %d, connect %d, "+
+			"first byte %d, upstream total %d", total, checks, connect, firstByte,
+			upstream)
+	}
+
+	if *refused.DurationChecks > refused.DurationTotal {
+		t.Errorf("refused request's checks took %v of %v milliseconds",
+			*refused.DurationChecks, refused.DurationTotal)
+	}
+}
+
+// wantTimings checks that the timings named are the only ones line has.
+func wantTimings(t *testing.T, line logLine, want ...string) {
+	t.Helper()
+
+	var got []string
+
+	for name := range line.fields {
+		if strings.HasPrefix(name, "duration_") {
+			got = append(got, name)
+		}
+	}
+
+	slices.Sort(got)
+	slices.Sort(want)
+
+	if !slices.Equal(got, want) {
+		t.Errorf("log line of %s has timings %v, want %v", line.Path, got, want)
+	}
+}
+
+// microseconds is a timing in whole microseconds.
+func microseconds(milliseconds float64) int64 {
+	return int64(math.Round(milliseconds * 1000))
+}
+
+func TestLogsAnUpgradedConnection(t *testing.T) {
+	t.Parallel()
+
+	app := startApp(t, echoAfterUpgrade)
+	addr, out := startProxy(t, app.URL, nil)
+
+	conn := dial(t, addr)
+	send(t, conn, "GET /socket HTTP/1.1\r\nHost: app\r\n"+
+		"Connection: Upgrade\r\nUpgrade: websocket\r\n\r\n")
+	wantStatus(t, readResponse(t, conn), http.StatusSwitchingProtocols)
+
+	_ = conn.Close()
+
+	line := out.requestLine(t)
+	if line.fields["websocket"] != true {
+		t.Errorf("log line has websocket %v, want true", line.fields["websocket"])
+	}
+}

internal/ratelimit/ratelimit.go

@@ -149,26 +149,39 @@ type Hit struct {
 	Requests float64
 }
 
+// Counts are a client's requests in the minute, the hour and the day that
+// end at a request, that request included.
+type Counts struct {
+	Minute float64 `json:"minute"`
+	Hour   float64 `json:"hour"`
+	Day    float64 `json:"day"`
+}
+
 // Count counts a request from client at now, in every window, whether or
-// not it is refused. It reports whether the request takes the client over
-// a limit, and the window whose limit it goes over, the shortest if it is
-// over several.
-func (l *Limiter) Count(client netip.Prefix, now time.Time) (Hit, bool) {
+// not it is refused, and returns the client's requests in each window. It
+// reports whether the request takes the client over a limit, and the
+// window whose limit it goes over, the shortest if it is over several.
+func (l *Limiter) Count(client netip.Prefix, now time.Time) (Counts, Hit, bool) {
 	l.mu.Lock()
 	defer l.mu.Unlock()
 
-	var hit Hit
+	var (
+		requests [3]float64
+		hit      Hit
+	)
 
 	for i, b := range l.get(client).buckets() {
 		w := l.windows[i]
 
-		requests := b.add(now, w.length)
-		if hit.Window == "" && w.limit > 0 && requests > float64(w.limit) {
-			hit = Hit{Window: w.name, Limit: w.limit, Requests: requests}
+		requests[i] = b.add(now, w.length)
+		if hit.Window == "" && w.limit > 0 && requests[i] > float64(w.limit) {
+			hit = Hit{Window: w.name, Limit: w.limit, Requests: requests[i]}
 		}
 	}
 
-	return hit, hit.Window != ""
+	counts := Counts{Minute: requests[0], Hour: requests[1], Day: requests[2]}
+
+	return counts, hit, hit.Window != ""
 }
 
 // Reset sets client's counts in every window back to zero. Its history
@@ -269,18 +282,21 @@ func (l *Limiter) Snapshot() []Client {
 	return clients
 }
 
-// Load puts clients read from clients.json into a table that holds none
-// yet, in the order they were last seen, so that the least recently seen
-// is dropped first. Buckets whose time has passed at now are emptied.
+// Load puts clients read from clients.json into the table, in place of
+// the clients it holds, in the order they were last seen, so that the
+// least recently seen is dropped first. Buckets whose time has passed at
+// now are emptied.
 func (l *Limiter) Load(clients []Client, now time.Time) {
-	l.mu.Lock()
-	defer l.mu.Unlock()
-
 	clients = slices.Clone(clients)
 	slices.SortStableFunc(clients, func(a, b Client) int {
 		return a.History.LastSeen.Compare(b.History.LastSeen)
 	})
 
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	l.clients.Purge()
+
 	for _, c := range clients {
 		for i, b := range c.buckets() {
 			// The window that ends at now covers neither bucket once it

internal/ratelimit/ratelimit_test.go

@@ -62,14 +62,14 @@ func TestHitGivesTheLimitAndTheRequestsCounted(t *testing.T) {
 	start := midnight()
 
 	for range limit {
-		_, over := limiter.Count(client, start)
+		_, _, over := limiter.Count(client, start)
 		if over {
 			t.Fatal("a request within the limit is over it")
 		}
 	}
 
 	// Over both limits; the minute's is named, with the four requests.
-	hit, over := limiter.Count(client, start)
+	_, hit, over := limiter.Count(client, start)
 
 	want := ratelimit.Hit{Window: minute, Limit: limit, Requests: limit + 1}
 	if !over || hit != want {
@@ -78,6 +78,29 @@ func TestHitGivesTheLimitAndTheRequestsCounted(t *testing.T) {
 	}
 }
 
+func TestCountGivesTheRequestsInEachWindow(t *testing.T) {
+	t.Parallel()
+
+	limiter := ratelimit.New(ratelimit.Limits{})
+	client := netip.MustParsePrefix("203.0.113.9/32")
+	start := midnight()
+
+	for range 3 {
+		limiter.Count(client, start)
+	}
+
+	// A quarter into the next hour, the minute has only this request. The
+	// hour still covers three quarters of the bucket before, with its three
+	// requests, which count 2.25, and this one: 3.25. The day covers all
+	// four.
+	counts, _, _ := limiter.Count(client, start.Add(time.Hour+time.Hour/4))
+
+	want := ratelimit.Counts{Minute: 1, Hour: 3.25, Day: 4}
+	if counts != want {
+		t.Errorf("counts %+v, want %+v", counts, want)
+	}
+}
+
 func TestResetSetsTheCountsBackToZero(t *testing.T) {
 	t.Parallel()
 
@@ -238,7 +261,7 @@ func wantCount(
 ) {
 	t.Helper()
 
-	hit, _ := limiter.Count(client, now)
+	_, hit, _ := limiter.Count(client, now)
 	if hit.Window != want {
 		t.Errorf("request from %s at %s is over %q, want %q",
 			client, now.Format(time.RFC3339), hit.Window, want)

internal/requestlog/requestlog.go

@@ -9,6 +9,8 @@ import (
 	"io"
 	"log/slog"
 	"time"
+
+	"sneak.berlin/go/smallwebwaf/internal/ratelimit"
 )
 
 // The action a request line names: what smallwebwaf did with the
@@ -45,28 +47,71 @@ const OffenceLimit = "limit"
 // timeLayout is RFC 3339 with milliseconds.
 const timeLayout = "2006-01-02T15:04:05.000Z07:00"
 
-// Line is one request's line in the request log. The field names are
-// those of the "Request log" section of SPEC.md.
+// Line is one request's line in the request log. The field names, and
+// their order, are those of the "Request log" section of SPEC.md. A field
+// that may not apply to a request is left out of its line when it does
+// not.
 //
 //nolint:tagliatelle // SPEC.md's request log names its fields in snake_case
 type Line struct {
-	Type           string `json:"type"`
-	Time           string `json:"time"`
-	ClientIP       string `json:"client_ip"`
-	PeerIP         string `json:"peer_ip"`
-	Country        string `json:"country"`
-	Method         string `json:"method"`
-	Host           string `json:"host"`
-	Path           string `json:"path"`
-	Query          string `json:"query"`
-	Protocol       string `json:"protocol"`
-	Status         int    `json:"status"`
-	UpstreamStatus int    `json:"upstream_status,omitempty"`
-	RequestBytes   int64  `json:"request_bytes"`
-	ResponseBytes  int64  `json:"response_bytes"`
-	Referer        string `json:"referer"`
-	UserAgent      string `json:"user_agent"`
-	Action         string `json:"action"`
+	Type string `json:"type"`
+
+	// The standard web log fields. Scheme is how the client reached
+	// smallwebwaf, or the trusted proxy in front of it.
+	Time          string `json:"time"`
+	Instance      string `json:"instance"`
+	ClientIP      string `json:"client_ip"`
+	Method        string `json:"method"`
+	Scheme        string `json:"scheme"`
+	Host          string `json:"host"`
+	Path          string `json:"path"`
+	Query         string `json:"query"`
+	Protocol      string `json:"protocol"`
+	Status        int    `json:"status"`
+	RequestBytes  int64  `json:"request_bytes"`
+	ResponseBytes int64  `json:"response_bytes"`
+	Referer       string `json:"referer"`
+	UserAgent     string `json:"user_agent"`
+
+	// Request detail. RequestID is the X-Request-ID a trusted proxy sent,
+	// or a new one, and is sent on to the app. ForwardedFor is the
+	// X-Forwarded-For header as received. ClientGroup is the netblock the
+	// client is counted as.
+	RequestID    string `json:"request_id"`
+	PeerIP       string `json:"peer_ip"`
+	ForwardedFor string `json:"forwarded_for,omitempty"`
+	ClientGroup  string `json:"client_group"`
+	Country      string `json:"country"`
+	ContentType  string `json:"content_type,omitempty"`
+	// ContentLength is the length of its body the request announced.
+	ContentLength int64 `json:"content_length,omitempty"`
+	// RequestHeaders are the headers SWWAF_LOG_REQUEST_HEADERS names that
+	// the request carried, by name in lower case.
+	RequestHeaders   map[string]string `json:"request_headers,omitempty"`
+	HasAuthorization bool              `json:"has_authorization,omitempty"`
+	HasCookie        bool              `json:"has_cookie,omitempty"`
+	// Websocket is true when the connection was upgraded, as for a
+	// WebSocket.
+	Websocket bool `json:"websocket,omitempty"`
+
+	// Response detail, from the headers of the answer: the app's, as
+	// passed on, or those of smallwebwaf's own. Aborted is true when the
+	// client went away early.
+	ResponseContentType string `json:"response_content_type,omitempty"`
+	UpstreamStatus      int    `json:"upstream_status,omitempty"`
+	CacheControl        string `json:"cache_control,omitempty"`
+	Location            string `json:"location,omitempty"`
+	Aborted             bool   `json:"aborted,omitempty"`
+
+	// The decision.
+	Action string `json:"action"`
+	// WouldAction is, in observe mode, the action enforce mode would have
+	// taken with a request it would have refused: ActionDenied,
+	// ActionBanned, ActionCountryDenied or ActionRateLimited.
+	WouldAction string `json:"would_action,omitempty"`
+	// Counts are the client's requests as the rate limits counted them
+	// with this one, for a request they counted.
+	Counts ratelimit.Counts `json:"counts,omitzero"`
 	// LimitHit is the window whose rate limit the request went over:
 	// minute, hour or day.
 	LimitHit string `json:"limit_hit,omitempty"`
@@ -75,11 +120,18 @@ type Line struct {
 	// BanExpires is when the ban the request made, or was refused under,
 	// ends: a time, or "permanent".
 	BanExpires string `json:"ban_expires,omitempty"`
-	// Aborted is true when the client went away early.
-	Aborted bool `json:"aborted,omitempty"`
-	// DurationTotal and DurationUpstreamTotal are in milliseconds.
-	DurationTotal         float64 `json:"duration_total"`
-	DurationUpstreamTotal float64 `json:"duration_upstream_total,omitempty"`
+
+	// The timings, in milliseconds. DurationChecks is the time until the
+	// checks were done. DurationUpstreamConnect, DurationUpstreamFirstByte
+	// and DurationUpstreamTotal run from when the request was handed to the
+	// app: until there was a connection to it, until the first byte of its
+	// answer arrived, and until the end. Each but DurationTotal is nil for
+	// a request that did not get that far.
+	DurationTotal             float64  `json:"duration_total"`
+	DurationChecks            *float64 `json:"duration_checks,omitempty"`
+	DurationUpstreamConnect   *float64 `json:"duration_upstream_connect,omitempty"`
+	DurationUpstreamFirstByte *float64 `json:"duration_upstream_first_byte,omitempty"`
+	DurationUpstreamTotal     *float64 `json:"duration_upstream_total,omitempty"`
 }
 
 // Write writes line to w as one JSON line marked "type":"request".

internal/requestlog/requestlog_test.go

@@ -50,7 +50,11 @@ func TestWriteWritesOneJSONLineMarkedRequest(t *testing.T) {
 	}
 
 	unset := []string{
-		"upstream_status", "limit_hit", "offence", "ban_expires", "aborted",
+		"forwarded_for", "content_type", "content_length", "request_headers",
+		"has_authorization", "has_cookie", "websocket", "response_content_type",
+		"upstream_status", "cache_control", "location", "aborted", "counts",
+		"limit_hit", "offence", "ban_expires", "duration_checks",
+		"duration_upstream_connect", "duration_upstream_first_byte",
 		"duration_upstream_total",
 	}
 	for _, name := range unset {

internal/smallwebwaf/smallwebwaf.go

@@ -113,9 +113,10 @@ func Run(ctx context.Context, params Params) int {
 	return serve(ctx, server.Server, listener, files, processLog)
 }
 
-// serve serves requests on listener, and writes the state files as they
-// are due, until ctx is done. Then it gives the requests in progress
-// shutdownTimeout to finish, and writes every state file.
+// serve serves requests on listener, writes the state files as they are
+// due, and takes in an admin's edits of them, until ctx is done. Then it
+// gives the requests in progress shutdownTimeout to finish, and writes
+// every state file.
 func serve(
 	ctx context.Context, server *http.Server, listener net.Listener,
 	files *state.Files, processLog *slog.Logger,
@@ -130,12 +131,18 @@ func serve(
 	defer stopWriting()
 
 	written := make(chan struct{})
+	watched := make(chan struct{})
 
 	go func() {
 		files.Run(writing)
 		close(written)
 	}()
 
+	go func() {
+		files.Watch(writing)
+		close(watched)
+	}()
+
 	select {
 	case err := <-served:
 		processLog.Error("serving failed", "error", err.Error())
@@ -165,13 +172,15 @@ func serve(
 		return 1
 	}
 
-	// Run's last write has ended, so nothing else writes the files. Every
-	// request has ended too, but for two kinds that Go's server does not
-	// wait for: one cut off because Shutdown timed out, and one whose
-	// connection switched protocols, such as a WebSocket. Such a request
-	// adds to its client's history only as it ends, which can be after
-	// this write, and then that request is missing from clients.json.
+	// Run and Watch have ended, so nothing else reads or writes the
+	// files. Every request has ended too, but for two kinds
+	// that Go's server does not wait for: one cut off because Shutdown
+	// timed out, and one whose connection switched protocols, such as a
+	// WebSocket. Such a request adds to its client's history only as it
+	// ends, which can be after this write, and then that request is
+	// missing from clients.json.
 	<-written
+	<-watched
 
 	err = files.WriteAll()
 	if err != nil {

internal/smallwebwaf/smallwebwaf_test.go

@@ -26,11 +26,14 @@ const (
 	// testVersion is the version the tests give smallwebwaf.
 	testVersion = "test"
 	// localhost is where the tests listen.
-	localhost       = "127.0.0.1"
-	listenAddr      = "SWWAF_LISTEN_ADDR"
-	upstreamURL     = "SWWAF_UPSTREAM_URL"
-	stateDir        = "SWWAF_STATE_DIR"
-	rateLimitPerDay = "SWWAF_RATE_LIMIT_PER_DAY"
+	localhost            = "127.0.0.1"
+	listenAddr           = "SWWAF_LISTEN_ADDR"
+	upstreamURL          = "SWWAF_UPSTREAM_URL"
+	trustedProxies       = "SWWAF_TRUSTED_PROXIES"
+	stateDir             = "SWWAF_STATE_DIR"
+	stateWriteDelay      = "SWWAF_STATE_WRITE_DELAY"
+	stateCounterInterval = "SWWAF_STATE_COUNTER_INTERVAL"
+	rateLimitPerDay      = "SWWAF_RATE_LIMIT_PER_DAY"
 	// greeting is what the tests' app answers.
 	greeting = "hello from the app"
 )
@@ -217,8 +220,8 @@ func TestStateKeptAcrossRestarts(t *testing.T) {
 		rateLimitPerDay: "2",
 		// Neither comes due in the test: the files are written as
 		// smallwebwaf stops.
-		"SWWAF_STATE_WRITE_DELAY":      "1h",
-		"SWWAF_STATE_COUNTER_INTERVAL": "1h",
+		stateWriteDelay:      "1h",
+		stateCounterInterval: "1h",
 	}
 
 	// The two requests a day allows, and a stop.
@@ -247,12 +250,12 @@ func TestBanRefusesItsNetblockAfterARestartWithAnotherScope(t *testing.T) {
 	const scope = "SWWAF_BAN_SCOPE_V4_PREFIX"
 
 	env := map[string]string{
-		listenAddr:              localhost + ":0",
-		upstreamURL:             startApp(t),
-		stateDir:                t.TempDir(),
-		"SWWAF_TRUSTED_PROXIES": localhost + "/32",
-		rateLimitPerDay:         "1",
-		scope:                   "24",
+		listenAddr:      localhost + ":0",
+		upstreamURL:     startApp(t),
+		stateDir:        t.TempDir(),
+		trustedProxies:  localhost + "/32",
+		rateLimitPerDay: "1",
+		scope:           "24",
 	}
 
 	// 203.0.113.9's second request breaks the day limit, and bans
@@ -281,6 +284,38 @@ func TestBanRefusesItsNetblockAfterARestartWithAnotherScope(t *testing.T) {
 	})
 }
 
+func TestBanAddedAndLiftedByEditingBansJSON(t *testing.T) {
+	t.Parallel()
+
+	const (
+		// bans.json as an admin writes it with a ban, permanent, on
+		// 203.0.113.0/24, and with none.
+		oneBan = `{"version": 1, "bans": [{"netblock": "203.0.113.0/24", ` +
+			`"start": "2026-10-06T00:00:00Z", "expires": null}]}`
+		noBan = `{"version": 1, "bans": []}`
+	)
+
+	dir := t.TempDir()
+	env := map[string]string{
+		listenAddr:     localhost + ":0",
+		upstreamURL:    startApp(t),
+		stateDir:       dir,
+		trustedProxies: localhost + "/32",
+		// No write comes due in the test, so only the watch on the
+		// directory can take the edits in.
+		stateWriteDelay:      "1h",
+		stateCounterInterval: "1h",
+	}
+
+	runUntilStopped(t, env, func(url string) {
+		path := filepath.Join(dir, "bans.json")
+
+		saveUntilAnswered(t, path, oneBan, url, "203.0.113.9", http.StatusForbidden)
+		wantStatus(t, url, "198.51.100.7", http.StatusOK)
+		saveUntilAnswered(t, path, noBan, url, "203.0.113.9", http.StatusOK)
+	})
+}
+
 func TestStateFileThatDoesNotParseStopsTheStart(t *testing.T) {
 	t.Parallel()
 
@@ -383,9 +418,10 @@ func wantStartingLine(t *testing.T, line map[string]any, appURL, dir string) {
 		listenAddr:                              localhost + ":0",
 		upstreamURL:                             appURL,
 		stateDir:                                dir,
-		"SWWAF_STATE_WRITE_DELAY":               "10s",
-		"SWWAF_STATE_COUNTER_INTERVAL":          "15m",
-		"SWWAF_TRUSTED_PROXIES":                 "10.0.0.0/8,172.16.0.0/12,192.168.0.0/16",
+		"SWWAF_MODE":                            "enforce",
+		stateWriteDelay:                         "10s",
+		stateCounterInterval:                    "15m",
+		trustedProxies:                          "10.0.0.0/8,172.16.0.0/12,192.168.0.0/16",
 		"SWWAF_CLIENT_REQUEST_TIMEOUT":          "60s",
 		"SWWAF_CLIENT_REQUEST_HEADER_MAX_BYTES": "32K",
 		"SWWAF_CLIENT_IDLE_TIMEOUT":             "120s",
@@ -400,6 +436,7 @@ func wantStartingLine(t *testing.T, line map[string]any, appURL, dir string) {
 		"SWWAF_RATE_LIMIT_PER_MINUTE":           "1000",
 		"SWWAF_RATE_LIMIT_PER_HOUR":             "10000",
 		rateLimitPerDay:                         "50000",
+		"SWWAF_RATE_LIMIT_EXEMPT_PATHS":         "",
 		"SWWAF_DENIED_COUNTRIES":                "",
 		"SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES":   "",
 		"SWWAF_BAN_RESPONSE":                    "403",
@@ -478,6 +515,40 @@ func wantRefused(t *testing.T, url string) {
 func wantStatus(t *testing.T, url, from string, status int) {
 	t.Helper()
 
+	got := statusFrom(t, url, from)
+	if got != status {
+		t.Errorf("request from %s: status %d, want %d", from, got, status)
+	}
+}
+
+// saveUntilAnswered writes content to the state file at path, as an
+// admin saves an edit of it, until a request to url from the client at
+// from is answered with status. The file is written again before each
+// request, since smallwebwaf may not watch its directory yet when it is
+// first written. It waits as long as that takes, so that a slow test
+// process cannot fail the test.
+func saveUntilAnswered(t *testing.T, path, content, url, from string, status int) {
+	t.Helper()
+
+	for {
+		err := os.WriteFile(path, []byte(content), 0o600)
+		if err != nil {
+			t.Fatalf("write %s: %v", path, err)
+		}
+
+		if statusFrom(t, url, from) == status {
+			return
+		}
+
+		time.Sleep(pollInterval)
+	}
+}
+
+// statusFrom returns the status a request to url from the client at
+// from, as X-Forwarded-For names it, is answered with.
+func statusFrom(t *testing.T, url, from string) int {
+	t.Helper()
+
 	req, err := http.NewRequestWithContext(t.Context(), http.MethodGet, url,
 		http.NoBody)
 	if err != nil {
@@ -496,7 +567,5 @@ func wantStatus(t *testing.T, url, from string, status int) {
 
 	_ = res.Body.Close()
 
-	if res.StatusCode != status {
-		t.Errorf("request from %s: status %d, want %d", from, res.StatusCode, status)
-	}
+	return res.StatusCode
 }

internal/state/state.go

@@ -1,14 +1,17 @@
 // Package state keeps smallwebwaf's state in JSON files in
 // SWWAF_STATE_DIR, as the "Persistent state" section of SPEC.md describes:
 // bans.json holds the bans, clients.json each client's counters and
-// history, and lookups.json GeoJS's answers. Load reads them at start, and
-// Run and WriteAll write them, each from a snapshot its part takes under
-// its own lock, so that no request waits on the disk.
+// history, and lookups.json GeoJS's answers. Load reads them at start,
+// Watch takes in an admin's edit of one while smallwebwaf runs, and Run
+// and WriteAll write them. The disk is read and written outside the
+// parts' locks, which are held only to take a snapshot or to put in what
+// a file holds, so that no request waits on the disk.
 package state
 
 import (
 	"bytes"
 	"context"
+	"crypto/sha256"
 	"encoding/json"
 	"errors"
 	"fmt"
@@ -17,8 +20,10 @@ import (
 	"net/netip"
 	"os"
 	"path/filepath"
+	"sync"
 	"time"
 
+	"github.com/fsnotify/fsnotify"
 	"sneak.berlin/go/smallwebwaf/internal/bans"
 	"sneak.berlin/go/smallwebwaf/internal/lookup"
 	"sneak.berlin/go/smallwebwaf/internal/metrics"
@@ -61,15 +66,26 @@ type Params struct {
 	// Now tells the time by which the counters' buckets run out, normally
 	// time.Now in UTC.
 	Now func() time.Time
-	// ProcessLog receives what was read, and the writes that fail.
+	// ProcessLog receives what was read and taken in, the edits set aside,
+	// and the writes that fail.
 	ProcessLog *slog.Logger
-	// Metrics count each file's writes.
+	// Metrics count each file's writes, and the edits taken in and set
+	// aside.
 	Metrics *metrics.Metrics
 }
 
 // Files are the state files of a running smallwebwaf.
 type Files struct {
 	params Params
+
+	// mu is held while a file is read for an edit, and while it is
+	// written, so that Watch and the writes take turns. No request takes
+	// it.
+	mu sync.Mutex
+	// sums are the SHA-256 sums of what each file held, by name, when
+	// smallwebwaf last read or wrote it. A file that holds anything else
+	// has been edited since.
+	sums map[string][sha256.Size]byte
 }
 
 // bansFile is bans.json, indented for an admin to read and edit.
@@ -120,41 +136,28 @@ func Load(params Params) (*Files, error) {
 		return nil, fmt.Errorf("SWWAF_STATE_DIR cannot be written: %w", err)
 	}
 
-	var (
-		bansIn    bansFile
-		clientsIn clientsFile
-		lookupsIn lookupsFile
-	)
+	f := &Files{params: params, sums: map[string][sha256.Size]byte{}}
 
-	err = errors.Join(
-		read(params.Dir, bansJSON, &bansIn),
-		read(params.Dir, clientsJSON, &clientsIn),
-		read(params.Dir, lookupsJSON, &lookupsIn),
-	)
+	bansRead, bansErr := f.read(bansJSON)
+	clientsRead, clientsErr := f.read(clientsJSON)
+	lookupsRead, lookupsErr := f.read(lookupsJSON)
+
+	err = errors.Join(bansErr, clientsErr, lookupsErr)
 	if err != nil {
 		return nil, err
 	}
 
-	held := make([]bans.Ban, 0, len(bansIn.Bans))
-	for _, entry := range bansIn.Bans {
-		held = append(held, entry.ban())
-	}
-
-	params.Ledger.Load(held)
-	params.Limiter.Load(clientsIn.Clients, params.Now())
-	params.GeoJS.Load(lookupsIn.Lookups)
-
 	params.ProcessLog.Info("read the state files", "directory", params.Dir,
-		"bans", len(bansIn.Bans), "clients", len(clientsIn.Clients),
-		"lookups", len(lookupsIn.Lookups))
+		"bans", bansRead, "clients", clientsRead, "lookups", lookupsRead)
 
-	return &Files{params: params}, nil
+	return f, nil
 }
 
 // Run writes bans.json WriteDelay after a ban is made, with every ban
 // made in between, and every file every CounterInterval, until ctx is
 // done. A write that fails is logged, and the file is written again at
-// its next write.
+// its next write. Each write takes in an admin's edit of its file first,
+// as writeFile describes.
 func (f *Files) Run(ctx context.Context) {
 	interval := time.NewTicker(f.params.CounterInterval)
 	defer interval.Stop()
@@ -172,7 +175,7 @@ func (f *Files) Run(ctx context.Context) {
 		case <-bansDue:
 			bansDue = nil
 
-			f.logFailure(f.writeBans())
+			f.logFailure(f.writeFile(bansJSON))
 		case <-interval.C:
 			f.logFailure(f.WriteAll())
 		}
@@ -182,7 +185,50 @@ func (f *Files) Run(ctx context.Context) {
 // WriteAll writes every state file, as smallwebwaf stops. A file that
 // fails does not keep the others from being written.
 func (f *Files) WriteAll() error {
-	return errors.Join(f.writeBans(), f.writeClients(), f.writeLookups())
+	return errors.Join(f.writeFile(bansJSON), f.writeFile(clientsJSON),
+		f.writeFile(lookupsJSON))
+}
+
+// Watch watches Dir until ctx is done, and takes in an admin's edit of a
+// state file as soon as it is saved: what the file holds replaces what
+// smallwebwaf held for it. An edit that does not parse is left for the
+// file's next write, which sets it aside, since a file can be read while
+// an editor is still writing it. If Dir cannot be watched, that is
+// logged, and an edit is taken in only before its file is written.
+func (f *Files) Watch(ctx context.Context) {
+	watcher, err := fsnotify.NewWatcher()
+	if err == nil {
+		defer func() {
+			_ = watcher.Close()
+		}()
+
+		err = watcher.Add(f.params.Dir)
+	}
+
+	if err != nil {
+		f.params.ProcessLog.Error("cannot watch the state files for edits",
+			"error", err.Error())
+
+		return
+	}
+
+	f.params.ProcessLog.Info("watching the state files for edits",
+		"directory", f.params.Dir)
+
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case event := <-watcher.Events:
+			switch name := filepath.Base(event.Name); name {
+			case bansJSON, clientsJSON, lookupsJSON:
+				f.fileChanged(name)
+			}
+		case err = <-watcher.Errors:
+			f.params.ProcessLog.Warn("watching the state files failed",
+				"error", err.Error())
+		}
+	}
 }
 
 // logFailure logs a write that failed.
@@ -193,52 +239,209 @@ func (f *Files) logFailure(err error) {
 	}
 }
 
-// writeBans writes bans.json.
-func (f *Files) writeBans() error {
-	held := f.params.Ledger.Snapshot()
+// fileChanged takes in what the state file name holds, as Watch sees it
+// change, if that is an edit made since smallwebwaf last read or wrote
+// the file. A file that cannot be read or does not parse is left for its
+// next write.
+func (f *Files) fileChanged(name string) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
 
-	file := bansFile{Version: version, Bans: make([]banEntry, 0, len(held))}
-	for _, ban := range held {
-		file.Bans = append(file.Bans, newBanEntry(ban))
+	data, changed, err := f.readChanged(name)
+	if err != nil || !changed {
+		return
 	}
 
-	data, err := json.MarshalIndent(file, "", "    ")
-	if err != nil {
-		return fmt.Errorf("encode %s: %w", bansJSON, err)
-	}
-
-	return f.writeCounted(bansJSON, append(data, '\n'))
+	_ = f.takeInEdit(name, data)
 }
 
-// writeClients writes clients.json.
-func (f *Files) writeClients() error {
-	data, err := encodeOnePerLine("clients", f.params.Limiter.Snapshot())
+// takeInEdit takes in data, an edit of the state file name, as takeIn
+// does, and counts and logs it. Every edit taken in while smallwebwaf
+// runs, by Watch or by a write, is taken in here. An edit that does not
+// parse is neither counted nor logged, and takeIn's error returned.
+func (f *Files) takeInEdit(name string, data []byte) error {
+	_, err := f.takeIn(name, data)
 	if err != nil {
-		return fmt.Errorf("encode %s: %w", clientsJSON, err)
+		return err
 	}
 
-	return f.writeCounted(clientsJSON, data)
+	// Counted before it is logged, so that the count is there once the
+	// log line is.
+	f.params.Metrics.StateFileEditTakenIn(name)
+	f.params.ProcessLog.Info("took in an edit of a state file",
+		"file", filepath.Join(f.params.Dir, name))
+
+	return nil
 }
 
-// writeLookups writes lookups.json.
-func (f *Files) writeLookups() error {
-	data, err := encodeOnePerLine("lookups", f.params.GeoJS.Snapshot())
-	if err != nil {
-		return fmt.Errorf("encode %s: %w", lookupsJSON, err)
+// read takes in the state file name at start, and returns how many
+// entries it holds. A missing file holds none.
+func (f *Files) read(name string) (int, error) {
+	data, changed, err := f.readChanged(name)
+	if err != nil || !changed {
+		return 0, err
 	}
 
-	return f.writeCounted(lookupsJSON, data)
+	return f.takeIn(name, data)
 }
 
-// writeCounted writes data to the state file name, as write does, and
-// counts the write in the metrics.
-func (f *Files) writeCounted(name string, data []byte) error {
-	err := write(f.params.Dir, name, data)
+// readChanged returns what the state file name holds, and whether that
+// has changed since smallwebwaf last read or wrote the file, as it has
+// for a file smallwebwaf never read or wrote. A missing file has not
+// changed: it is written again at its next write.
+func (f *Files) readChanged(name string) ([]byte, bool, error) {
+	path := filepath.Join(f.params.Dir, name)
+
+	data, err := os.ReadFile(path) //nolint:gosec // a state file, in SWWAF_STATE_DIR
+	if errors.Is(err, fs.ErrNotExist) {
+		return nil, false, nil
+	}
+
+	if err != nil {
+		return nil, false, err
+	}
+
+	return data, sha256.Sum256(data) != f.sums[name], nil
+}
+
+// takeIn parses data, what the state file name holds, puts it into the
+// part that keeps that state, in place of what the part held, and returns
+// how many entries the file holds. An error names the file and, where the
+// JSON decoder tells it, the line and column, or else the entry.
+func (f *Files) takeIn(name string, data []byte) (int, error) {
+	path := filepath.Join(f.params.Dir, name)
+
+	var entries int
+
+	switch name {
+	case bansJSON:
+		var file bansFile
+
+		err := parse(path, data, &file)
+		if err != nil {
+			return 0, err
+		}
+
+		held := make([]bans.Ban, 0, len(file.Bans))
+		for _, entry := range file.Bans {
+			held = append(held, entry.ban())
+		}
+
+		f.params.Ledger.Load(held)
+		entries = len(held)
+	case clientsJSON:
+		var file clientsFile
+
+		err := parse(path, data, &file)
+		if err != nil {
+			return 0, err
+		}
+
+		f.params.Limiter.Load(file.Clients, f.params.Now())
+		entries = len(file.Clients)
+	case lookupsJSON:
+		var file lookupsFile
+
+		err := parse(path, data, &file)
+		if err != nil {
+			return 0, err
+		}
+
+		f.params.GeoJS.Load(file.Lookups)
+		entries = len(file.Lookups)
+	}
+
+	f.sums[name] = sha256.Sum256(data)
+
+	return entries, nil
+}
+
+// writeFile writes the state file name from what smallwebwaf holds. An
+// edit made since smallwebwaf last read or wrote the file is taken in
+// first, so that it is not overwritten, or set aside if it does not
+// parse. A file that cannot be read, or an edit that cannot be set
+// aside, is left as it is, and the write given up. Every write is counted
+// in the metrics, and one that fails or is given up as a failure.
+func (f *Files) writeFile(name string) error {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+
+	data, changed, err := f.readChanged(name)
+	if err == nil && changed {
+		err = f.takeInEdit(name, data)
+		if err != nil {
+			err = f.setAside(name, err)
+		}
+	}
+
+	if err == nil {
+		data, err = f.encode(name)
+		if err != nil {
+			err = fmt.Errorf("encode %s: %w", name, err)
+		}
+	}
+
+	if err == nil {
+		err = write(f.params.Dir, name, data)
+	}
+
+	if err == nil {
+		// The file holds data from here on, even if the directory sync
+		// fails, so that its next read does not take it for an admin's
+		// edit.
+		f.sums[name] = sha256.Sum256(data)
+		err = syncDirectory(f.params.Dir)
+	}
+
 	f.params.Metrics.StateFileWritten(name, len(data), err)
 
 	return err
 }
 
+// setAside renames the state file name, an edit that does not parse with
+// parseErr, to name.bad, for the admin to mend, and logs it with where in
+// the file the error is. If the rename fails, the edit is left as it is,
+// and the error returned is parseErr joined with the rename's.
+func (f *Files) setAside(name string, parseErr error) error {
+	path := filepath.Join(f.params.Dir, name)
+
+	err := os.Rename(path, path+".bad")
+	if err != nil {
+		return errors.Join(parseErr, err)
+	}
+
+	f.params.ProcessLog.Error("set aside an edit of a state file that does not parse",
+		"file", path+".bad", "error", parseErr.Error())
+	f.params.Metrics.StateFileEditSetAside(name)
+
+	return nil
+}
+
+// encode returns the state file name as smallwebwaf writes it, from a
+// snapshot of the part that keeps that state.
+func (f *Files) encode(name string) ([]byte, error) {
+	switch name {
+	case bansJSON:
+		held := f.params.Ledger.Snapshot()
+
+		file := bansFile{Version: version, Bans: make([]banEntry, 0, len(held))}
+		for _, ban := range held {
+			file.Bans = append(file.Bans, newBanEntry(ban))
+		}
+
+		data, err := json.MarshalIndent(file, "", "    ")
+		if err != nil {
+			return nil, err
+		}
+
+		return append(data, '\n'), nil
+	case clientsJSON:
+		return encodeOnePerLine("clients", f.params.Limiter.Snapshot())
+	default: // lookups.json
+		return encodeOnePerLine("lookups", f.params.GeoJS.Snapshot())
+	}
+}
+
 // newBanEntry returns ban as bans.json holds it.
 func newBanEntry(ban bans.Ban) banEntry {
 	entry := banEntry{Netblock: ban.Netblock, Start: ban.Start, Notes: ban.Notes}
@@ -389,28 +592,16 @@ func checkWritable(dir string) error {
 	return errors.Join(file.Close(), os.Remove(file.Name()))
 }
 
-// read reads the state file name in dir into file, a pointer to that
-// file's struct, and checks its entries. A missing file leaves file as it
-// is.
-func read(dir, name string, file stateFile) error {
-	path := filepath.Join(dir, name)
-
-	data, err := os.ReadFile(path) //nolint:gosec // a state file, in SWWAF_STATE_DIR
-	if errors.Is(err, fs.ErrNotExist) {
-		return nil
-	}
-
-	if err != nil {
-		return err
-	}
-
+// parse reads data, what the state file at path holds, into file, a
+// pointer to that file's struct, and checks its entries.
+func parse(path string, data []byte, file stateFile) error {
 	// The version is read first, so that a file of another version is
 	// refused for that, and not for an entry this version cannot read.
 	var header struct {
 		Version int `json:"version"`
 	}
 
-	err = json.Unmarshal(data, &header)
+	err := json.Unmarshal(data, &header)
 	if err == nil && header.Version != version {
 		err = fmt.Errorf("%w %d, where this smallwebwaf reads version %d",
 			errVersion, header.Version, version)
@@ -463,7 +654,7 @@ func position(data []byte, err error) string {
 // write writes data to the file name in dir so that a crash at any
 // moment leaves either the old file or the new one, whole: data goes to a
 // temporary file in the same directory, which is synced and renamed over
-// name, and then the directory is synced, so that the rename lasts.
+// name. syncDirectory must follow, so that the rename lasts.
 func write(dir, name string, data []byte) error {
 	path := filepath.Join(dir, name)
 	temporary := path + ".tmp"
@@ -475,10 +666,13 @@ func write(dir, name string, data []byte) error {
 
 	if err != nil {
 		_ = os.Remove(temporary)
-
-		return err
 	}
 
+	return err
+}
+
+// syncDirectory syncs dir to the disk, so that a rename in it lasts.
+func syncDirectory(dir string) error {
 	directory, err := os.Open(dir) //nolint:gosec // SWWAF_STATE_DIR itself
 	if err != nil {
 		return err

internal/state/state_test.go

@@ -3,7 +3,10 @@ package state_test
 import (
 	"context"
 	"encoding/json"
+	"io/fs"
 	"log/slog"
+	"maps"
+	"net"
 	"net/http"
 	"net/http/httptest"
 	"net/netip"
@@ -28,6 +31,13 @@ const (
 	bansJSON    = "bans.json"
 	clientsJSON = "clients.json"
 	lookupsJSON = "lookups.json"
+	// What the process log says once Watch watches the directory, and as
+	// it takes in an edit.
+	watching = "watching the state files for edits"
+	tookIn   = "took in an edit of a state file"
+	// maxLogLines is how many lines of the process log wait for a test to
+	// read them.
+	maxLogLines = 64
 )
 
 // permanentBansJSON is bans.json holding permanentBan.
@@ -290,7 +300,7 @@ func TestUnwritableDirectoryStopsTheStart(t *testing.T) {
 	}
 }
 
-// The two tests below run Run in a synctest bubble, where time is a clock
+// The three tests below run Run in a synctest bubble, where time is a clock
 // of the test's own: time.Sleep moves it on at once, and synctest.Wait
 // returns once Run waits for its next write, so that every write due by
 // then is on disk.
@@ -302,7 +312,7 @@ func TestBansWrittenOnceWriteDelayAfterABan(t *testing.T) {
 		dir := t.TempDir()
 		params := newParams(dir)
 		params.WriteDelay = 10 * time.Second
-		run(t, load(t, params))
+		run(t, load(t, params).Run)
 
 		// A second ban, made while the first waits to be written, puts the
 		// write off no further, and is written with it.
@@ -347,7 +357,7 @@ func TestEveryFileWrittenEveryCounterInterval(t *testing.T) {
 		dir := t.TempDir()
 		params := newParams(dir)
 		params.CounterInterval = time.Minute
-		run(t, load(t, params))
+		run(t, load(t, params).Run)
 
 		// The files are removed once written, so that each interval shows
 		// them written again.
@@ -364,6 +374,44 @@ func TestEveryFileWrittenEveryCounterInterval(t *testing.T) {
 	})
 }
 
+func TestEditJustBeforeAScheduledWriteSurvivesIt(t *testing.T) {
+	t.Parallel()
+
+	synctest.Test(t, func(t *testing.T) {
+		dir := t.TempDir()
+		params := newParams(dir)
+		params.WriteDelay = 10 * time.Second
+		run(t, load(t, params).Run)
+
+		// A ban, and bans.json written with it.
+		params.Ledger.BanForLimit(netip.MustParsePrefix("203.0.113.9/32"),
+			midnight(), bans.Notes{})
+		time.Sleep(params.WriteDelay)
+		synctest.Wait()
+
+		// A second ban is to be written WriteDelay later. Just before
+		// then, an admin saves bans.json with the first ban lifted and
+		// another added.
+		params.Ledger.BanForLimit(netip.MustParsePrefix("203.0.113.10/32"),
+			midnight(), bans.Notes{})
+		time.Sleep(params.WriteDelay - time.Nanosecond)
+		synctest.Wait()
+		edit(t, dir, bansJSON, permanentBansJSON)
+
+		// The write takes the edit in first, and writes it back. The second
+		// ban, made after the admin opened the file, is lost, as "Edits
+		// while running" in SPEC.md says.
+		time.Sleep(time.Nanosecond)
+		synctest.Wait()
+
+		if got := readFile(t, filepath.Join(dir, bansJSON)); got != permanentBansJSON {
+			t.Errorf("bans.json holds\n%s\nwant the edit", got)
+		}
+
+		wantEqual(t, bansJSON, params.Ledger.Snapshot(), []bans.Ban{permanentBan()})
+	})
+}
+
 func TestFailedWriteLeavesTheFileAsItWas(t *testing.T) {
 	t.Parallel()
 
@@ -459,24 +507,359 @@ func TestWritesAreCountedInTheMetrics(t *testing.T) {
 		float64(len(permanentBansJSON)))
 }
 
-func TestFailedRenameLeavesNoTemporaryFile(t *testing.T) {
+func TestFileThatCannotBeReadIsNotWrittenOver(t *testing.T) {
 	t.Parallel()
 
 	dir := t.TempDir()
-	files := load(t, newParams(dir))
+	path := filepath.Join(dir, bansJSON)
+	params := newParams(dir)
+	files := load(t, params)
 
-	// A directory named bans.json cannot be renamed over.
-	err := os.Mkdir(filepath.Join(dir, bansJSON), 0o700)
+	// bans.json is a socket, which cannot be opened as a file, even by
+	// root, as the tests run in Docker, but which a rename could replace.
+	// Whether it holds an edit cannot be told, so it is left as it is.
+	socket, err := (&net.ListenConfig{}).Listen(t.Context(), "unix", path)
+	if err != nil {
+		t.Fatalf("listen: %v", err)
+	}
+
+	defer func() {
+		_ = socket.Close()
+	}()
+
+	err = files.WriteAll()
+	if err == nil {
+		t.Error("writing with bans.json unreadable did not fail")
+	}
+
+	info, err := os.Lstat(path)
+	if err != nil || info.Mode().Type() != fs.ModeSocket {
+		t.Errorf("bans.json is now %v (%v), want the socket", info, err)
+	}
+
+	wantFiles(t, dir, bansJSON, clientsJSON, lookupsJSON)
+	wantWriteFailed(t, params, bansJSON)
+}
+
+func TestBrokenEditThatCannotBeSetAsideIsNotWrittenOver(t *testing.T) {
+	t.Parallel()
+
+	const broken = `{"version": 1, "bans": [`
+
+	dir := t.TempDir()
+	path := filepath.Join(dir, bansJSON)
+	params := newParams(dir)
+	files := load(t, params)
+
+	// A directory named bans.json.bad cannot be renamed over, so the
+	// broken edit cannot be set aside, and is left as it is.
+	edit(t, dir, bansJSON, broken)
+
+	err := os.Mkdir(path+".bad", 0o700)
 	if err != nil {
 		t.Fatalf("mkdir: %v", err)
 	}
 
 	err = files.WriteAll()
 	if err == nil {
-		t.Error("writing over a directory did not fail")
+		t.Error("writing with bans.json.bad in the way did not fail")
 	}
 
+	if got := readFile(t, path); got != broken {
+		t.Errorf("bans.json holds\n%s\nwant the edit", got)
+	}
+
+	wantWriteFailed(t, params, bansJSON)
+}
+
+func TestEditOfEachFileTakenIn(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	params := newParams(dir)
+	lines := logInto(&params)
+	fill(params)
+	files := load(t, params)
+
+	err := files.WriteAll()
+	if err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	watch(t, files, lines)
+
+	// Each edit holds one entry, for a client the parts did not hold, and
+	// takes the place of everything the part held.
+	client := netip.MustParsePrefix("198.51.100.7/32")
+
+	edit(t, dir, bansJSON, `{"version": 1, "bans": [{"netblock": "198.51.100.7/32", `+
+		`"start": "2026-10-06T00:00:00Z", "expires": null}]}`)
+	wantTakenIn(t, lines, dir, bansJSON)
+	wantEqual(t, bansJSON, params.Ledger.Snapshot(),
+		[]bans.Ban{{Netblock: client, Start: midnight()}})
+
+	edit(t, dir, clientsJSON, `{"version": 1, "clients": [`+
+		`{"client": "198.51.100.7/32", "history": {"requests": 7}}]}`)
+	wantTakenIn(t, lines, dir, clientsJSON)
+	wantEqual(t, clientsJSON, params.Limiter.Snapshot(),
+		[]ratelimit.Client{{Client: client, History: ratelimit.History{Requests: 7}}})
+
+	edit(t, dir, lookupsJSON, `{"version": 1, "lookups": [{"client": "198.51.100.7/32", `+
+		`"country": "FR", "answered": "2026-10-06T00:00:00Z"}]}`)
+	wantTakenIn(t, lines, dir, lookupsJSON)
+	wantEqual(t, lookupsJSON, params.GeoJS.Snapshot(),
+		[]lookup.Answer{{Client: client, Country: "FR", Answered: midnight()}})
+}
+
+func TestOwnWritesAreNotTakenIn(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	params := newParams(dir)
+	lines := logInto(&params)
+	fill(params)
+	files := load(t, params)
+	watch(t, files, lines)
+
+	// Every file is written while watched, and then lookups.json edited:
+	// the first edit taken in is that one.
+	err := files.WriteAll()
+	if err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	edit(t, dir, lookupsJSON, `{"version": 1, "lookups": []}`)
+	wantTakenIn(t, lines, dir, lookupsJSON)
+}
+
+func TestFileRenamedOverAStateFileTakenIn(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	path := filepath.Join(dir, bansJSON)
+	params := newParams(dir)
+	lines := logInto(&params)
+	files := load(t, params)
+
+	err := files.WriteAll()
+	if err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	watch(t, files, lines)
+
+	// The admin mends bans.json.bad and moves it back, as editors that
+	// save by renaming do with a file of their own: nothing is written
+	// into bans.json itself. An edit of clients.json after it must be
+	// taken in second.
+	edit(t, dir, bansJSON+".bad", permanentBansJSON)
+
+	err = os.Rename(path+".bad", path)
+	if err != nil {
+		t.Fatalf("rename: %v", err)
+	}
+
+	edit(t, dir, clientsJSON, `{"version": 1, "clients": []}`)
+
+	wantTakenIn(t, lines, dir, bansJSON)
+	wantEqual(t, bansJSON, params.Ledger.Snapshot(), []bans.Ban{permanentBan()})
+}
+
+func TestBanAddedAndLiftedThroughBansJSON(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	params := newParams(dir)
+	lines := logInto(&params)
+	watch(t, load(t, params), lines)
+
+	client := netip.MustParseAddr("203.0.113.9")
+
+	// An entry added, as an admin writes it, bans its netblock.
+	edit(t, dir, bansJSON, `{"version": 1, "bans": [{"netblock": "203.0.113.0/24", `+
+		`"start": "2026-10-06T00:00:00Z", "expires": null}]}`)
+	wantTakenIn(t, lines, dir, bansJSON)
+
+	_, banned := params.Ledger.Check(client, midnight())
+	if !banned {
+		t.Error("the ban added to bans.json does not refuse")
+	}
+
+	// The entry removed lifts the ban.
+	edit(t, dir, bansJSON, `{"version": 1, "bans": []}`)
+	wantTakenIn(t, lines, dir, bansJSON)
+
+	_, banned = params.Ledger.Check(client, midnight())
+	if banned {
+		t.Error("the ban removed from bans.json still refuses")
+	}
+}
+
+func TestBrokenEditSetAsideAtTheNextWrite(t *testing.T) {
+	t.Parallel()
+
+	// It ends a ban's entry with a comma.
+	const broken = "{\n    \"version\": 1,\n    \"bans\": [\n" +
+		"        {\"netblock\": \"203.0.113.9/32\",}\n    ]\n}\n"
+
+	dir := t.TempDir()
+	path := filepath.Join(dir, bansJSON)
+	params := newParams(dir)
+	lines := logInto(&params)
+	params.Ledger.Load([]bans.Ban{permanentBan()})
+	files := load(t, params)
+
+	err := files.WriteAll()
+	if err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	watch(t, files, lines)
+
+	// While smallwebwaf runs, the broken edit is left as it is: an edit
+	// of clients.json, made after it and taken in, shows that it has been
+	// seen.
+	edit(t, dir, bansJSON, broken)
+	edit(t, dir, clientsJSON, `{"version": 1, "clients": []}`)
+	wantTakenIn(t, lines, dir, clientsJSON)
 	wantFiles(t, dir, bansJSON, clientsJSON, lookupsJSON)
+
+	// The next write sets it aside, logged with where the error is, and
+	// writes bans.json again from what smallwebwaf still holds.
+	err = files.WriteAll()
+	if err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	line := lines.waitFor(t, "set aside an edit of a state file that does not parse")
+	message, _ := line["error"].(string)
+
+	if line["file"] != path+".bad" ||
+		!strings.HasPrefix(message, path+", line 4, column 39: ") {
+		t.Errorf("set aside with %v", line)
+	}
+
+	wantFiles(t, dir, bansJSON, bansJSON+".bad", clientsJSON, lookupsJSON)
+
+	if got := readFile(t, path+".bad"); got != broken {
+		t.Errorf("bans.json.bad holds\n%s\nwant the edit", got)
+	}
+
+	if got := readFile(t, path); got != permanentBansJSON {
+		t.Errorf("bans.json holds\n%s\nwant\n%s", got, permanentBansJSON)
+	}
+}
+
+func TestEditsTakenInAreCountedInTheMetrics(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	params := newParams(dir)
+	lines := logInto(&params)
+	files := load(t, params)
+
+	// One edit is taken in by the write of its file, before Watch runs,
+	// and one by Watch.
+	edit(t, dir, bansJSON, `{"version": 1, "bans": []}`)
+
+	err := files.WriteAll()
+	if err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	watch(t, files, lines)
+	edit(t, dir, bansJSON, permanentBansJSON)
+	wantTakenIn(t, lines, dir, bansJSON)
+
+	wantMetric(t, scrape(t, params),
+		`smallwebwaf_state_file_edits_taken_in_total{file="bans.json"}`, 2)
+}
+
+func TestEditTakenInByAWriteIsLoggedAsWatchLogsIt(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	params := newParams(dir)
+	lines := logInto(&params)
+	files := load(t, params)
+
+	// An edit taken in by Watch, which is then stopped.
+	ctx, stop := context.WithCancel(t.Context())
+	stopped := make(chan struct{})
+
+	go func() {
+		files.Watch(ctx)
+		close(stopped)
+	}()
+
+	lines.waitFor(t, watching)
+	edit(t, dir, bansJSON, `{"version": 1, "bans": []}`)
+	byWatch := lines.waitFor(t, tookIn)
+
+	stop()
+	<-stopped
+
+	// An edit taken in by the write of its file. Nothing logs after the
+	// write, so the log is closed, and a write that does not log the edit
+	// fails the test at once instead of waiting for the line.
+	edit(t, dir, bansJSON, permanentBansJSON)
+
+	err := files.WriteAll()
+	if err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	close(lines)
+
+	byWrite := lines.waitFor(t, tookIn)
+
+	// The two lines differ only in their time.
+	delete(byWatch, "time")
+	delete(byWrite, "time")
+
+	if !maps.Equal(byWrite, byWatch) {
+		t.Errorf("the write logged %v, where Watch logged %v", byWrite, byWatch)
+	}
+}
+
+func TestEditsSetAsideAreCountedInTheMetrics(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	params := newParams(dir)
+	files := load(t, params)
+
+	edit(t, dir, bansJSON, `{"version": 1, "bans": [`)
+
+	err := files.WriteAll()
+	if err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	wantMetric(t, scrape(t, params),
+		`smallwebwaf_state_file_edits_set_aside_total{file="bans.json"}`, 1)
+}
+
+func TestDirectoryThatCannotBeWatchedIsLogged(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	params := newParams(dir)
+	lines := logInto(&params)
+	files := load(t, params)
+
+	err := os.Remove(dir)
+	if err != nil {
+		t.Fatalf("remove: %v", err)
+	}
+
+	// Watch returns at once.
+	files.Watch(t.Context())
+
+	line := lines.waitFor(t, "cannot watch the state files for edits")
+	if line["level"] != "ERROR" {
+		t.Errorf("logged as %v", line)
+	}
 }
 
 // midnight is the time of the tests' clock.
@@ -573,15 +956,16 @@ func load(t *testing.T, params state.Params) *state.Files {
 	return files
 }
 
-// run runs files' writes until the test ends.
-func run(t *testing.T, files *state.Files) {
+// run runs task, the Run or the Watch of state files, until the test
+// ends.
+func run(t *testing.T, task func(context.Context)) {
 	t.Helper()
 
 	ctx, stop := context.WithCancel(t.Context())
 	stopped := make(chan struct{})
 
 	go func() {
-		files.Run(ctx)
+		task(ctx)
 		close(stopped)
 	}()
 
@@ -591,6 +975,80 @@ func run(t *testing.T, files *state.Files) {
 	})
 }
 
+// watch runs files' Watch until the test ends, and waits until it
+// watches the directory.
+func watch(t *testing.T, files *state.Files, lines processLog) {
+	t.Helper()
+
+	run(t, files.Watch)
+	lines.waitFor(t, watching)
+}
+
+// processLog receives the lines of a process log, each a JSON object, for
+// a test to wait for.
+type processLog chan string
+
+// logInto has params' process log write its lines into a new processLog,
+// and returns that.
+func logInto(params *state.Params) processLog {
+	lines := make(processLog, maxLogLines)
+	params.ProcessLog = slog.New(slog.NewJSONHandler(lines, nil))
+
+	return lines
+}
+
+// Write receives a line of the process log.
+func (l processLog) Write(line []byte) (int, error) {
+	l <- string(line)
+
+	return len(line), nil
+}
+
+// waitFor returns the next line of the process log whose message is msg,
+// passing over the lines before it, or nil if the log is closed first. It
+// waits as long as that takes, so that a slow test process cannot fail
+// the test.
+func (l processLog) waitFor(t *testing.T, msg string) map[string]any {
+	t.Helper()
+
+	for line := range l {
+		var fields map[string]any
+
+		err := json.Unmarshal([]byte(line), &fields)
+		if err != nil {
+			t.Fatalf("process log line %q is not JSON: %v", line, err)
+		}
+
+		if fields["msg"] == msg {
+			return fields
+		}
+	}
+
+	return nil
+}
+
+// wantTakenIn waits for the next edit taken in, and checks that it is of
+// the state file name in dir.
+func wantTakenIn(t *testing.T, lines processLog, dir, name string) {
+	t.Helper()
+
+	line := lines.waitFor(t, tookIn)
+	if line["file"] != filepath.Join(dir, name) {
+		t.Fatalf("took in %v, want an edit of %s", line, name)
+	}
+}
+
+// edit writes content to the state file name in dir, as an admin saves an
+// edit of it.
+func edit(t *testing.T, dir, name, content string) {
+	t.Helper()
+
+	err := os.WriteFile(filepath.Join(dir, name), []byte(content), 0o600)
+	if err != nil {
+		t.Fatalf("write %s: %v", name, err)
+	}
+}
+
 // wantEqual checks that the entries read back from file are those
 // written.
 func wantEqual[E comparable](t *testing.T, file string, got, want []E) {
@@ -735,6 +1193,18 @@ func metric(t *testing.T, text, series string) float64 {
 	return 0
 }
 
+// wantWriteFailed checks that the metrics of params count one write of the
+// state file name, and that it failed.
+func wantWriteFailed(t *testing.T, params state.Params, name string) {
+	t.Helper()
+
+	got := scrape(t, params)
+	file := `{file="` + name + `"}`
+
+	wantMetric(t, got, "smallwebwaf_state_file_writes_total"+file, 1)
+	wantMetric(t, got, "smallwebwaf_state_file_write_failures_total"+file, 1)
+}
+
 // wantMetric checks the value of series in text, the metrics, as metric
 // reads it.
 func wantMetric(t *testing.T, text, series string, want float64) {

internal/state/write_internal_test.go

@@ -0,0 +1,35 @@
+package state
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// The test is on write itself: a state file is read before it is
+// written, and a directory in its place fails that read first.
+func TestFailedRenameLeavesNoTemporaryFile(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+
+	// A directory named bans.json cannot be renamed over.
+	err := os.Mkdir(filepath.Join(dir, bansJSON), 0o700)
+	if err != nil {
+		t.Fatalf("mkdir: %v", err)
+	}
+
+	err = write(dir, bansJSON, []byte("{}\n"))
+	if err == nil {
+		t.Error("writing over a directory did not fail")
+	}
+
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		t.Fatalf("read %s: %v", dir, err)
+	}
+
+	if len(entries) != 1 || entries[0].Name() != bansJSON {
+		t.Errorf("%s holds %v, want only bans.json", dir, entries)
+	}
+}