Re-vendor the canonical files from sneak/prompts at 0653579 (closes #221)

REPO_POLICIES.md, script/cibuild, script/docker, script/lint and
script/test are their copies at that commit. In every Dockerfile stage
the system package install and the Go module download now come before
ARG CACHEBUST, so check builds keep those layers until the base image,
the install line, go.mod or go.sum change; every step that lints, tests
or compiles still reruns. golangci-lint run also mounts golangci-lint's
cache. The stages install their packages with apt-get and apk directly
instead of through script/bootstrap --cgo. The test phase installs no
recommended packages and names libheif's AVIF encoder,
libheif-plugin-aomenc, which the tests use.

Model: opus-5-5

Dockerfile

@@ -1,12 +1,16 @@
-# Every stage declares ARG CACHEBUST first: script/ passes a new value
+# In every stage ARG CACHEBUST comes right after the steps that install
-# on every build, which reruns every RUN step after it. The steps that
+# system packages and download Go's modules, or right after FROM in a
-# compile mount Go's build cache, which survives from build to build.
+# stage with neither, and every step that lints, tests or compiles
+# comes after it. script/ passes a new value on every build, which
+# reruns every RUN step after it. Docker keeps the layers above it
+# until the base image, their own lines or go.mod and go.sum change.
+# The steps that compile mount Go's build cache, and golangci-lint run
+# also mounts golangci-lint's; both survive from build to build.
 
 # Lint phase. script/lint builds it alone. The linter is run directly:
 # `make lint` and script/lint are themselves a docker build.
 # golangci/golangci-lint:v2.14.0, 2026-10-15
 FROM golangci/golangci-lint@sha256:ad862ba6b3798cbe0fd9fd7408d498fd74fbd2623a92406b2fd3898faf0bf98f AS lint
-ARG CACHEBUST
 
 # The linter compiles every package, and govips needs the libvips
 # headers for that. REPO_POLICIES.md has the lint phase install them
@@ -18,25 +22,29 @@ RUN apt-get update \
 WORKDIR /src
 COPY go.mod go.sum ./
 RUN go mod download
+ARG CACHEBUST
 COPY . .
 RUN --mount=type=cache,target=/root/.cache/go-build \
+    --mount=type=cache,target=/root/.cache/golangci-lint \
     golangci-lint run --config .golangci.yml ./...
 
 # Test phase. script/test builds it alone. -race needs cgo and so a C
 # compiler, which the Debian Go image ships and the alpine one does not.
 # golang:1.25.4-trixie, 2026-10-15
 FROM golang@sha256:a02d35efc036053fdf0da8c15919276bf777a80cbfda6a35c5e9f087e652adfc AS test
-ARG CACHEBUST
+
+# govips compiles against the libvips and libheif headers, and the tests
+# need libvips' JPEG XL support; libvips-dev brings all three. The tests
+# also encode AVIF, which needs libheif's AVIF encoder,
+# libheif-plugin-aomenc: Debian only recommends it, so it is named here.
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends libvips-dev libheif-plugin-aomenc \
+    && rm -rf /var/lib/apt/lists/*
 
 WORKDIR /src
-
-# script/bootstrap --cgo installs the build dependencies (the libvips
-# and libheif headers, and libvips' JPEG XL support, which the tests
-# need) and downloads the Go modules.
-COPY script/ ./script/
 COPY go.mod go.sum ./
-RUN script/bootstrap --cgo
+RUN go mod download
-
+ARG CACHEBUST
 COPY . .
 
 # Without -v first; on a failure, again with -v for the details, and
@@ -52,22 +60,23 @@ RUN --mount=type=cache,target=/root/.cache/go-build \
       go test -count=1 -timeout 90s -race -parallel 4 -v ./...; exit 1; }
 
 # Build stage. Nothing is wanted from the two phases above: these copies
-# make BuildKit build them first, so this stage runs only when lint and
+# make BuildKit build them first, so nothing after them runs unless lint
-# test passed.
+# and test passed.
 # golang:1.25.4-alpine3.22, 2026-02-25; the runtime stage uses Alpine 3.22 too
 FROM golang:1.25.4-alpine3.22@sha256:d3f0cf7723f3429e3f9ed846243970b20a2de7bae6a5b66fc5914e228d831bbb AS builder
+
+# git for the version, and for govips a C compiler, pkg-config and the
+# libvips and libheif headers
+RUN apk add --no-cache git build-base pkgconfig vips-dev libheif-dev
+
+WORKDIR /src
+COPY go.mod go.sum ./
+RUN go mod download
 ARG CACHEBUST
 
 COPY --from=lint /src/go.sum /dev/null
 COPY --from=test /src/go.sum /dev/null
 
-WORKDIR /src
-
-# Build dependencies, git among them, and Go modules, as in the test
-# phase
-COPY script/ ./script/
-COPY go.mod go.sum ./
-RUN script/bootstrap --cgo
 # A tar-stream context keeps the sender's file owners, which git refuses.
 RUN git config --system --add safe.directory /src
 
@@ -102,7 +111,6 @@ RUN --mount=type=cache,target=/root/.cache/go-build \
 # the libvips and musl it was built with.
 # alpine:3.22, 2026-10-08
 FROM alpine:3.22@sha256:5291449c3df73caf6ed85e649dec1b9e818b39a5d8c871e97afc13e9cd5e8fa8
-ARG CACHEBUST
 
 # Install runtime dependencies only. vips-jxl is libvips' JPEG XL
 # support, without which pixad does not start.
@@ -113,6 +121,7 @@ RUN apk add --no-cache \
     ca-certificates \
     tzdata \
     su-exec
+ARG CACHEBUST
 
 # Copy binary from builder
 COPY --from=builder /pixad /usr/local/bin/pixad

README.md

@@ -102,8 +102,8 @@ it uses libvips through CGO. pixad does not start unless libvips has its JPEG XL
 support, which on Alpine is the `vips-jxl` package and which the nix and brew
 packages of libvips include, as do the apt ones from Debian 12 and Ubuntu 24.04
 on. Building pixa also needs the development files of libvips and libheif,
-`pkg-config` and a C compiler. `script/bootstrap --cgo` installs all of these,
+`pkg-config` and a C compiler. `script/bootstrap --cgo` installs all of these;
-as the `Dockerfile` does where it compiles pixa. Plain `script/bootstrap`, which
+the `Dockerfile` installs its own packages. Plain `script/bootstrap`, which
 `script/setup` and `script/cibuild` run, installs git, make and Go, and Node,
 Yarn and the prettier pinned in `yarn.lock` for formatting the markdown, but
 none of the C libraries: the checks compile pixa in Docker. Docker itself must
@@ -199,7 +199,11 @@ into place, so a crash leaves no file under its own name with part of its data.
 When pixa starts, and every five minutes after, it removes each cached source or
 transformed image whose file is empty, or not the size the database records for
 it, with a warning naming the file; the next request for it fetches the source
-or converts the image again.
+or converts the image again. Looking for those files reads the database once for
+every cached file, so pixa skips each five-minute look that would start less
+than ten times as long after the last look ended as that look took. After a look
+that took one minute, the next one starts 10 to 15 minutes after that look
+ended.
 
 pixa's target is 1-5k r/s, which has not been measured at that rate (see Load
 Test). Toward it, pixa keeps in memory the content types of the 10,000
@@ -316,6 +320,35 @@ once, is refused with 400.
   both optional (values under Signature Specification). Both are part of what is
   cached, so each value of either is a separate cached image.
 
+A source URL's query follows its path after `%3F`, the encoded `?`:
+`/v1/image/<host>/<path>%3F<query>/<size>.<format>`. pixa decodes the
+percent-escapes in the image URL's path once, then reads the host up to the
+first `/`, the size and format after the last `/`, and between them the source
+path up to the first `?` and the source query after it. So the `?` before the
+query, any `?` in the query, and every `%` in the source path or query must be
+percent-encoded, as `%3F` and `%25`. Any other character may be written as
+itself or percent-encoded: `/`, `;`, `,`, `=` and `&` work either way. Neither
+the source path nor the source query may hold a `#`, and an image URL that
+writes one as `%23` is refused with 400. For example,
+`https://cdn.example.com/img?url=https://origin.example.com/my%20cat.jpg` at
+800x600 WebP is
+`/v1/image/cdn.example.com/img%3Furl=https://origin.example.com/my%2520cat.jpg/800x600.webp`,
+signed as shown under Signature Specification.
+
+The source path is the path with its percent-escapes decoded, and pixa fetches
+it percent-encoded again: every character other than letters, digits, `-`, `.`,
+`_`, `~`, `/`, `$`, `&`, `+`, `,`, `:`, `;`, `=` and `@` is sent as a
+percent-escape. So `my%20cat.jpg` in the image URL is the source path
+`/my cat.jpg`, fetched as `/my%20cat.jpg`; `my%2520cat.jpg` is the source path
+`/my%20cat.jpg`, fetched as `/my%2520cat.jpg`; and `100%25.png` is fetched as
+`/100%25.png`. The source query is the source URL's query as that URL writes it,
+its percent-escapes included, and is sent as it is, except that the characters a
+browser percent-encodes in a query (a space, `"`, `'`, `<`, `>`, control
+characters and characters outside ASCII) and a `%` that does not begin a
+percent-escape are percent-encoded. pixa signs and caches the query in that
+form, so `img%3Fname=my%20cat` and `img%3Fname=my%2520cat` in the image URL are
+the same source query, `name=my%20cat`.
+
 A size with one side `0` has that side scaled to keep the source's proportions,
 and is refused with 400 when that side would come out above 8192, the most
 either side may be; so is a size with `fit=outside`, which scales the image to
@@ -430,7 +463,13 @@ nor change what it asks for.
    sent with an empty format, or none, also makes a JPEG XL URL. Width and
    height both empty or `0` keep the original size; if only one of them is empty
    or `0`, that side is scaled to keep the image's proportions. A URL whose
-   scaled side would come out above 8192 answers 400.
+   scaled side would come out above 8192 answers 400, and so does a source URL
+   whose path holds a `..` segment. So does a source URL that the image URL
+   would fetch written another way, and the page then shows the URL pixa would
+   fetch: one with a user name and password, which pixa does not send, or one
+   whose path holds `%2F`, fetched as `/`, or `(`, fetched as `%28`, as pixa
+   keeps the source path decoded and fetches it percent-encoded again (see
+   Routes).
 3. The page shows the URL, `https://<host>/v1/e/<token>/img.<format>`, and when
    it expires. `<host>` is the host the page was opened on, and the URL starts
    with `http` instead while `debug` is on. The name after the token is ignored
@@ -438,9 +477,11 @@ nor change what it asks for.
    `jxl` for a form with no format.
 
 The token holds the source's host, path and query and the size, format, quality,
-fit and expiry, encrypted with a key derived from `signing_key`. The source
+fit and expiry, encrypted with a key derived from `signing_key`. The host is
-URL's scheme is not kept: the image is fetched like any other (see Routes), and
+kept in lower case and an empty path as `/`, as for `/v1/image/`, so
-the blocked networks still apply.
+`https://CDN.Example.com` and `https://cdn.example.com/` are one source image,
+fetched and cached once. The source URL's scheme is not kept: the image is
+fetched like any other (see Routes), and the blocked networks still apply.
 
 How long the URL lasts is chosen on the page, from 1 minute to 1 year, or never.
 The expiry is fixed in the token when the URL is made and cannot be changed or
@@ -492,13 +533,20 @@ Where:
 - `host` — source host as the URL writes it up to the first `/`, in lower case,
   keeping an IPv6 address's brackets and any `:` and port after it (e.g.
   `cdn.example.com`, `cdn.example.com:8443`, `[2001:db8::1]:8443`)
-- `path` — source path (e.g. `/photos/cat.jpg`), with each `%` written as `%25`
+- `path` — source path, with its percent-escapes decoded as under Routes (e.g.
-  and each `:` as `%3A`, so `/photos/a:b.jpg` is signed as `/photos/a%3Ab.jpg`
+  `/photos/cat.jpg`, or `/photos/my cat.jpg` for `/photos/my%20cat.jpg`), with
-- `query` — source query string, empty string if none, with `%` and `:` written
+  each `%` written as `%25` and each `:` as `%3A`, so `/photos/a:b.jpg` is
-  as in `path`; neither then holds a `:`, so path `/a:b` with no query and path
+  signed as `/photos/a%3Ab.jpg`
-  `/a` with query `b:` are signed differently
+- `query` — source query string as under Routes, empty string if none, with `%`
-- `width` — requested width in pixels, `0` for original
+  and `:` written as in `path`; neither then holds a `:`, so path `/a:b` with no
-- `height` — requested height in pixels, `0` for original
+  query and path `/a` with query `b:` are signed differently
+- `width` — requested width in pixels, or `0`: the width is then worked out from
+  the source's proportions and the requested height, rounded down but at least
+  1, or is the source's own width when the height is `0` too (the `orig` size)
+- `height` — requested height in pixels, or `0`: the height is then worked out
+  from the source's proportions and the requested width, rounded down but at
+  least 1, or is the source's own height when the width is `0` too (the `orig`
+  size)
 - `format` — output format, one of those listed under Routes, with `original`
   signed as `orig`, `jpg` as `jpeg`, and no format as `jxl`, so a URL with no
   format has the signature of the same URL ending in `.jxl`; `auto` is signed as
@@ -511,7 +559,10 @@ Where:
   with 400
 - `fit` — the URL's `fit` query parameter (cover, contain, fill, inside,
   outside), or `cover` when the URL has no `fit`; a request whose `fit` is
-  anything else, an empty `fit=` included, is refused with 400
+  anything else, an empty `fit=` included, is refused with 400. `outside` scales
+  the image to cover the requested size, keeping its proportions, so it can come
+  out larger than that size on one side: an 8192x1 source at `100x0`, which is
+  100x1 by its proportions, comes out at 8192x1
 
 The URL's `sig` is the HMAC-SHA256 result in base64url (the URL-safe alphabet of
 RFC 4648) with the trailing `=` padding kept, 44 characters in all. pixa
@@ -535,6 +586,17 @@ For the same image at quality 40 with fit `contain`, the input ends in
 and the URL is
 `/v1/image/cdn.example.com/photos/cat.jpg/800x600.webp?sig=5IwXUx6vf7yefhaUvFzgXZvG2o0Df4RJxPTK3pKq5VU=&exp=1704067200&q=40&fit=contain`.
 
+For a source URL with a query,
+`https://cdn.example.com/img?url=https://origin.example.com/my%20cat.jpg`, at
+800x600 WebP with the same key and expiration and the default quality and fit:
+
+1. Build input, with the query's `:` written as `%3A` and its `%` as `%25`:
+   `cdn.example.com:/img:url=https%3A//origin.example.com/my%2520cat.jpg:800:600:webp:1704067200:85:cover`
+2. The signature is `37aFa9GSVCZM1kRY9SzMOk2fOg_M7Wt5gKNR4xM-FYA=`
+3. URL, with the `?` before the query written as `%3F` and the query's `%` as
+   `%25` (see Routes):
+   `/v1/image/cdn.example.com/img%3Furl=https://origin.example.com/my%2520cat.jpg/800x600.webp?sig=37aFa9GSVCZM1kRY9SzMOk2fOg_M7Wt5gKNR4xM-FYA=&exp=1704067200`
+
 **Allowlist patterns:**
 
 - **Exact match**: `cdn.example.com` — matches only that host
@@ -550,7 +612,11 @@ pixa lower-cases the host of a `/v1/image/` URL before anything else uses it, so
 `CDN.Example.com` is matched, fetched, cached and counted against
 `upstream_connections_per_host` as `cdn.example.com`. Such a URL whose host has
 a port, as in `cdn.example.com:8443`, or a `:` with no port, needs a signature
-even when the host is allowlisted. A host holding `@` is refused with 400.
+even when the host is allowlisted. The host must be a host name or an IPv4
+address (letters, digits, hyphens, underscores and dots) or an IPv6 address in
+brackets, either one optionally followed by `:` and a port. Any other host, such
+as one holding `@`, or a `?`, `#`, `%`, `\` or space written percent-encoded, is
+refused with 400.
 
 ### Configuration
 
@@ -632,7 +698,12 @@ Key settings in more detail:
   and generator pages. Default: empty
 - `blocked_networks` — list of CIDR ranges to refuse for SSRF protection, added
   to the always-enforced built-in ranges (loopback, private, link-local, CGNAT,
-  benchmark, NAT64, and the like); an invalid CIDR aborts startup
+  benchmark, NAT64, and the like); an invalid CIDR aborts startup. An IPv6
+  address in the IPv4-mapped form (`::ffff:a.b.c.d`) is matched only as the IPv4
+  address it holds, so an IPv6 range such as `::/0` refuses neither it nor a
+  plain IPv4 address. An entry written in that form is read as the IPv4 range it
+  covers, so `::ffff:10.0.0.0/104` refuses what `10.0.0.0/8` refuses; such an
+  entry shorter than `/96` covers more than IPv4 addresses and aborts startup
 - `trusted_proxies` — list of CIDR ranges of the reverse proxies in front of
   pixa. `X-Forwarded-For` is believed only when the direct peer falls inside one
   of these ranges; the logged and login-recorded client address is then the
@@ -642,16 +713,22 @@ Key settings in more detail:
   omitted key defaults to the RFC 1918 private ranges (`10.0.0.0/8`,
   `172.16.0.0/12`, `192.168.0.0/16`), since pixa is deployed behind a proxy on a
   private network; an explicitly empty list (`[]`) trusts no one, and an
-  explicit list replaces the default. An invalid CIDR aborts startup. Set this
+  explicit list replaces the default. An invalid CIDR aborts startup. An IPv6
-  to the address pixa sees for requests that come through your proxy, such as
+  address in the IPv4-mapped form (`::ffff:a.b.c.d`) is matched only as the IPv4
-  `172.17.0.1/32`, when the defaults do not cover it, or to trust nothing else
+  address it holds, so an IPv6 range such as `::/0` trusts neither it nor a
-  (see the login limit and `image_requests_per_minute` under Routes). For a
+  plain IPv4 address. An entry written in that form is read as the IPv4 range it
-  proxy on the Docker host that connects to pixa over `127.0.0.1`, that address
+  covers, so `::ffff:10.0.0.0/104` trusts what `10.0.0.0/8` trusts; such an
-  is the gateway of the container's Docker network (`172.17.0.1` on the default
+  entry shorter than `/96` covers more than IPv4 addresses and aborts startup.
-  bridge), not the proxy's own address; a proxy that connects through another of
+  Set this to the address pixa sees for requests that come through your proxy,
-  the host's addresses is seen with that address. To be sure which address it
+  such as `172.17.0.1/32`, when the defaults do not cover it, or to trust
-  is, set this to `[]` (or `PIXA_TRUSTED_PROXIES` to empty), send a request
+  nothing else (see the login limit and `image_requests_per_minute` under
-  through the proxy, and read `remoteIP` in pixa's request log line for it
+  Routes). For a proxy on the Docker host that connects to pixa over
+  `127.0.0.1`, that address is the gateway of the container's Docker network
+  (`172.17.0.1` on the default bridge), not the proxy's own address; a proxy
+  that connects through another of the host's addresses is seen with that
+  address. To be sure which address it is, set this to `[]` (or
+  `PIXA_TRUSTED_PROXIES` to empty), send a request through the proxy, and read
+  `remoteIP` in pixa's request log line for it
 - `upstream_fetch_timeout` — time allowed for one fetch from an upstream host,
   as a duration such as `30s` (the default) or `2m`
 - `upstream_max_response_size` — largest upstream response accepted, in bytes;
@@ -663,11 +740,14 @@ Key settings in more detail:
 - `signing_key` — HMAC secret for URL signatures
 - `db_url` — the SQLite database to open; omitted, it is
   `file:<state_dir>/state.sqlite3?_pragma=journal_mode(WAL)`, which keeps the
-  database in WAL mode. pixa opens one connection to it, so its own reads and
+  database in WAL mode. `state_dir` is percent-encoded in it, so a `%`, `?` or
-  writes run one at a time. pixa adds `_pragma=busy_timeout(5000)` to any
+  `#` in the name is written as `%25`, `%3F` or `%23`. For a `state_dir` of `/`
-  `db_url`, so a write that finds another program writing to the file waits up
+  or one starting with `//` it starts `file:////`, so SQLite does not read the
-  to five seconds for it instead of failing. WAL mode comes only from the URL:
+  first name in the path as a host. pixa opens one connection to it, so its own
-  keep `_pragma=journal_mode(WAL)` in one you set
+  reads and writes run one at a time. pixa adds `_pragma=busy_timeout(5000)` to
+  any `db_url`, so a write that finds another program writing to the file waits
+  up to five seconds for it instead of failing. WAL mode comes only from the
+  URL: keep `_pragma=journal_mode(WAL)` in one you set
 - `cache_max_bytes` — disk cache size limit in bytes; `0` disables the disk
   cache entirely; omitted defaults to 75% of the sum of the free space on the
   filesystem containing `<state_dir>/cache/` and the bytes of source and
@@ -757,19 +837,25 @@ provide:
   `script/precommit`
 
 Every `docker build` in these scripts passes a new value of the `CACHEBUST`
-build argument, which every stage of the `Dockerfile` declares first, so every
+build argument, which every stage of the `Dockerfile` declares, so every `RUN`
-`RUN` step, the lint and test phases included, runs on every build instead of
+step after that line, the lint and test steps included, runs on every build
-coming from the build cache. Go's build cache is kept from build to build: the
+instead of coming from the build cache. Above the line are the steps that
-steps that compile mount it, so a package whose inputs are unchanged is not
+install system packages and download the Go modules; Docker keeps their layers
-compiled again, and both `go test` runs pass `-count=1` so that no stored test
+until the base image, the step itself, `go.mod` or `go.sum` changes. Go's build
-result is replayed. The `test` phase runs on the Debian Go image, with Debian's
+cache is kept from build to build: the steps that compile mount it, so a package
-libvips and libheif; the image pixa ships is built and runs on Alpine.
+whose inputs are unchanged is not compiled again, and both `go test` runs pass
-`script/check`, `script/cibuild`, `script/docker`, `script/lint`, `script/test`,
+`-count=1` so that no stored test result is replayed. The lint step also mounts
-`script/setup` and `script/install-precommit` are the standard copies from
+golangci-lint's cache, which reuses a package's findings only when its files,
-`sneak/prompts`, kept identical to them. `script/fmt` and `script/fmt-check` are
+those of the packages it imports, `go.mod`, the linters and their settings in
-the standard copies with pixa's `gofmt` step kept before prettier. prettier
+`.golangci.yml`, and the golangci-lint version are all unchanged. The `test`
-formats the markdown only, not the HTML templates, as it cannot parse a Go
+phase runs on the Debian Go image, with Debian's libvips and libheif; the image
-template action inside a tag.
+pixa ships is built and runs on Alpine. `script/check`, `script/cibuild`,
+`script/docker`, `script/lint`, `script/test`, `script/setup` and
+`script/install-precommit` are the standard copies from `sneak/prompts`, kept
+identical to them. `script/fmt` and `script/fmt-check` are the standard copies
+with pixa's `gofmt` step kept before prettier. prettier formats the markdown
+only, not the HTML templates, as it cannot parse a Go template action inside a
+tag.
 
 ## Load Test
 

REPO_POLICIES.md

@@ -1,6 +1,6 @@
 ---
 title: Repository Policies
-last_modified: 2026-10-15
+last_modified: 2026-10-18
 ---
 
 This document covers repository structure, tooling, and workflow standards. Code
@@ -159,8 +159,11 @@ style conventions are in separate documents:
     one's findings, and a host-global lock in `$TMPDIR` makes concurrent runs
     exit non-zero with `parallel golangci-lint is running` — a status a caller
     cannot tell from real findings. Both have produced wrong verdicts in this
-    org, in both directions. A container has its own golangci-lint cache, its
+    org, in both directions. A container has its own `TMPDIR`, so the lock is
-    own `TMPDIR` and a digest-pinned binary, so neither is reachable.
+    out of reach. The lint phase does keep golangci-lint's cache between check
+    builds (below), but it lints every checkout at `/src`, and a finding is
+    taken from that cache only for the same files, settings and golangci-lint
+    version.
 
 - **Any build that runs checks is given a new `CACHEBUST` value.** Docker
   invalidates a `COPY` layer only when the copied content changes, so on an
@@ -170,15 +173,21 @@ style conventions are in separate documents:
   id, so no two builds share a value, not even the three in one
   `script/cibuild`. `script/lint`, `script/test`, `script/cibuild` and
   `script/docker` are the four, and there is no fifth — `script/check` runs the
-  two gate phases and `script/fmt-check`, and builds no image of its own. Every
+  two gate phases and `script/fmt-check`, and builds no image of its own. In
-  stage of the `Dockerfile` declares `ARG CACHEBUST` as its first line after
+  every stage of the `Dockerfile`, `ARG CACHEBUST` comes right after the steps
-  `FROM`, and a new value reruns every `RUN` step after that line; `COPY` steps
+  that install system packages and download Go's modules, or right after `FROM`
-  are compared by content and need nothing. A stage without the `ARG` line is
+  in a stage with neither, and every step that lints, tests or compiles comes
-  served from cache on an unchanged tree, so a repository re-vendors the scripts
+  after it. A step that installs anything else as well, such as a
-  and adds the `ARG` lines in one commit. The scripts do not pass `--no-cache`:
+  `script/bootstrap` that also installs node packages, comes after it too. A new
-  it reruns every step too, but it also empties every cache mount, and Go's
+  value reruns every `RUN` step after that line; `COPY` steps are compared by
-  build cache, which the canonical Go `Dockerfile` below mounts so that compiled
+  content and need nothing. Docker keeps the layers above the line until the
-  packages survive between check builds, would start empty every time. A bare
+  base image, their own lines or the files they copy change. A stage without the
+  `ARG` line is served from cache on an unchanged tree, so a repository
+  re-vendors the scripts and adds the `ARG` lines in one commit. The scripts do
+  not pass `--no-cache`: it reruns every step too, but it also builds the layers
+  above the line again and empties every cache mount, so Go's build cache and
+  golangci-lint's, which the canonical Go `Dockerfile` below mounts so that they
+  survive between check builds, would start empty every time. A bare
   `docker build .` is not evidence that anything ran: a sub-second build
   reporting success is a cache hit, not a result. Never invalidate by pruning —
   `docker builder prune` and friends destroy a build cache shared with every
@@ -194,19 +203,25 @@ style conventions are in separate documents:
   `Dockerfile`:
 
     ```dockerfile
-    # Every stage declares ARG CACHEBUST first: script/ passes a new value
+    # In every stage ARG CACHEBUST comes right after the steps that install
-    # on every build, which reruns every RUN step after it. The steps that
+    # system packages and download Go's modules, or right after FROM in a
-    # compile mount Go's build cache, which survives from build to build.
+    # stage with neither, and every step that lints, tests or compiles
+    # comes after it. script/ passes a new value on every build, which
+    # reruns every RUN step after it. Docker keeps the layers above it
+    # until the base image, their own lines or go.mod and go.sum change.
+    # The steps that compile mount Go's build cache, and golangci-lint run
+    # also mounts golangci-lint's; both survive from build to build.
 
     # Lint phase
     # golangci/golangci-lint:v2.x.x, YYYY-MM-DD
     FROM golangci/golangci-lint@sha256:... AS lint
-    ARG CACHEBUST
     WORKDIR /src
     COPY go.mod go.sum ./
     RUN go mod download
+    ARG CACHEBUST
     COPY . .
     RUN --mount=type=cache,target=/root/.cache/go-build \
+        --mount=type=cache,target=/root/.cache/golangci-lint \
         golangci-lint run --config .golangci.yml ./...
 
     # Test phase. -race needs cgo and so a C compiler, which the Debian Go
@@ -214,10 +229,10 @@ style conventions are in separate documents:
     # cache also holds test results, which go test would otherwise replay.
     # golang:1.x, YYYY-MM-DD
     FROM golang@sha256:... AS test
-    ARG CACHEBUST
     WORKDIR /src
     COPY go.mod go.sum ./
     RUN go mod download
+    ARG CACHEBUST
     COPY . .
     RUN --mount=type=cache,target=/root/.cache/go-build \
         go test -count=1 -timeout 90s -race -cover ./... || \
@@ -225,19 +240,19 @@ style conventions are in separate documents:
           go test -count=1 -timeout 90s -race -v ./...; exit 1; }
 
     # Build stage. Nothing is wanted from either phase above; the copies
-    # are what make BuildKit build them first, so this stage cannot run
+    # are what make BuildKit build them first, so nothing after them can
-    # unless lint and test passed.
+    # run unless lint and test passed.
     # golang:1.x-alpine, YYYY-MM-DD
     FROM golang@sha256:... AS builder
-    ARG CACHEBUST
-    COPY --from=lint /src/go.sum /dev/null
-    COPY --from=test /src/go.sum /dev/null
     RUN apk add --no-cache git
-    # A tar-stream context keeps the sender's file owners, which git refuses.
-    RUN git config --system --add safe.directory /src
     WORKDIR /src
     COPY go.mod go.sum ./
     RUN go mod download
+    ARG CACHEBUST
+    COPY --from=lint /src/go.sum /dev/null
+    COPY --from=test /src/go.sum /dev/null
+    # A tar-stream context keeps the sender's file owners, which git refuses.
+    RUN git config --system --add safe.directory /src
     COPY . .
 
     # The VERSION build arg when one is given, otherwise
@@ -274,16 +289,26 @@ style conventions are in separate documents:
     - Keep the runtime stage last, and if you add a stage after it, give it the
       same two copies. A plain `docker build .` builds the last stage's chain
       and nothing else.
-    - Only Go's build cache is mounted (`/root/.cache/go-build`, where Go keeps
+    - Check builds keep these, and none of them can hand back a stale result:
-      it for root), and only on the steps that compile: `golangci-lint run`,
+        - The layers above `ARG CACHEBUST`: the system packages and Go's
-      both `go test` runs and `go build`. It keeps compiled packages between
+          modules. Docker installs the packages again when the base image or the
-      check builds, and Go reuses one only when all its inputs are unchanged, so
+          install line changes, and downloads the modules again when `go.mod` or
-      it saves the compile without skipping a check. Nothing else is kept:
+          `go.sum` changes. Go checks every module against `go.sum` as it
-      golangci-lint's own cache would replay stored findings in place of a lint
+          downloads it, and no later build can change a kept layer. The modules
-      run, and Go builds from whatever files a kept module cache holds without
+          are kept in a layer rather than a cache mount because a cache mount is
-      checking them again (only `go mod verify` does), so a file changed there
+          shared by every build on the host, and Go builds from whatever files
-      would go unnoticed. Go's build cache also holds test results, which is why
+          it holds without checking them again (only `go mod verify` does), so a
-      both `go test` runs carry `-count=1`.
+          file changed there would go unnoticed.
+        - Go's build cache (`/root/.cache/go-build`, where Go keeps it for
+          root), mounted on the steps that compile: `golangci-lint run`, both
+          `go test` runs and `go build`. Go reuses a compiled package only when
+          all its inputs are unchanged. The cache also holds test results, which
+          is why both `go test` runs carry `-count=1`.
+        - golangci-lint's cache (`/root/.cache/golangci-lint`, where the pinned
+          image keeps it), mounted on `golangci-lint run`. golangci-lint reuses
+          a package's findings only when the package's files and those of every
+          package it imports, `go.mod`, the linters and their settings in
+          `.golangci.yml`, and the golangci-lint version are all unchanged.
     - If the project uses `//go:embed` directives that reference build artifacts
       (e.g. a web frontend compiled in a separate stage), the lint phase must
       create placeholder files so the embed directives resolve. Example:
@@ -292,12 +317,20 @@ style conventions are in separate documents:
       in the lint phase. The `golangci/golangci-lint` image is Debian-based and
       has no `apk`, so install with `apt-get` under the Debian package name
       (`libvips-dev`, where alpine says `vips-dev`), and delete the package
-      lists in the same `RUN`, so the layer does not keep them:
+      lists in the same `RUN`, so the layer does not keep them. The test phase,
+      on the Debian Go image, installs the libraries its tests need the same
+      way. Either install goes right after `FROM`, above `ARG CACHEBUST`, so
+      Docker keeps it until the base image or the install line changes:
 
         ```dockerfile
+        FROM golangci/golangci-lint@sha256:... AS lint
         RUN apt-get update \
             && apt-get install -y --no-install-recommends libvips-dev \
             && rm -rf /var/lib/apt/lists/*
+        WORKDIR /src
+        COPY go.mod go.sum ./
+        RUN go mod download
+        ARG CACHEBUST
         ```
 
     - `.dockerignore` lets `.git` into the build context. It keeps out every git

TODO.md

@@ -30,6 +30,128 @@ P2: security: per-origin rate limiting
 
 # Completed Steps
 
+- 2026-10-18 re-vendored from `sneak/prompts` commit `0653579` (closes #221):
+  each `Dockerfile` stage installs its system packages and downloads the Go
+  modules above `ARG CACHEBUST`, so check builds keep them, and the lint step
+  keeps golangci-lint's cache.
+- 2026-10-19 a `trusted_proxies` entry in the IPv4-mapped form trusted nothing;
+  it is now read as the IPv4 range it covers by the same code that reads
+  `blocked_networks`, and such an entry shorter than `/96` aborts startup
+  (closes #303).
+- 2026-10-18 the next reconciliation pass waits at least ten times as long as
+  the last one took, so passes on a large cache no longer run back to back
+  (closes #250).
+- 2026-10-18 a `blocked_networks` entry in the IPv4-mapped form refused nothing;
+  `::ffff:10.0.0.0/104` is now read as `10.0.0.0/8`, and such an entry shorter
+  than `/96` aborts startup (closes #301).
+- 2026-10-16 P1 security bug: the check that keeps pixa from fetching internal
+  addresses refuses the local-use NAT64 range (`64:ff9b:1::/48`, RFC 8215) as it
+  refuses `64:ff9b::/96`, and refuses an IPv6 address in the IPv4-compatible
+  form (`::/96`) when `blocked_networks` contains either the address as written
+  or the IPv4 address it holds (closes #299). A source host such as
+  `[64:ff9b:1::a00:1]` was fetched, which on a network that runs NAT64 on that
+  range reaches the IPv4 address it holds, here `10.0.0.1`; and with
+  `203.0.113.0/24` in `blocked_networks`, `[::ffff:203.0.113.5]` was refused but
+  `[::203.0.113.5]` was fetched. Both now answer 403. The mapped form
+  (`::ffff:a.b.c.d`) is still matched only as the IPv4 address it holds: Go
+  keeps a plain IPv4 address in that form, and an IPv6 entry such as `::/0` must
+  not refuse every IPv4 address.
+- 2026-10-16 P1 security bug: the check that keeps pixa from fetching internal
+  addresses judges an IPv6 address in the IPv4-compatible form (`::/96`), such
+  as `::127.0.0.1` or `::10.0.0.1`, by the IPv4 address in its last four bytes
+  (closes #297). It read only the mapped form (`::ffff:a.b.c.d`) as IPv4, so a
+  source host such as `[::127.0.0.1]` was checked as a public IPv6 address and
+  fetched, which on a network that routes it reaches the internal IPv4 host; it
+  is now refused with 403. The NAT64 form (`64:ff9b::/96`) was already refused
+  as a whole, as one of the built-in blocked ranges, and still is.
+- 2026-10-16 a source host that is an IPv6 address in brackets with no port,
+  such as `[2001:db8::a]`, is fetched like one with a port, through `/v1/image/`
+  and through an encrypted URL (closes #295). The fetcher's check of the source
+  URL kept the brackets when no port followed them, could not look the host up,
+  and the request answered 500. It now checks the address without brackets, so a
+  private one is refused with 403, as with a port.
+- 2026-10-16 a `state_dir` whose name holds `%`, `?` or `#` keeps its database
+  at `state.sqlite3` in that directory (closes #290): the `db_url` derived from
+  it put the name into a `file:` URL as it was, so SQLite read `%41` as `A`, a
+  `?` or `#` ended the path early, and the journal mode parameter was lost. The
+  derived `db_url` is now built with Go's `net/url`, which writes those
+  characters as `%25`, `%3F` and `%23`. For a `state_dir` of `/` or one starting
+  with `//` it starts `file:////`, so SQLite no longer reads the first name in
+  the path as a host.
+- 2026-10-16 a size with one side `0` keeps the side computed from the source's
+  proportions at 1 or more (closes #283): it was rounded down, so for a thin
+  source it came out at 0, and libvips then made the size square. An 8192x1
+  source at `100x0` came out at 100x100, and at `1300x0` with `fit=cover` made
+  libvips ask for about 39 GB at once; an 8192x2 source at `100x0` answered 500.
+  `fit=contain` and `fit=inside` work both sides out again from the smaller of
+  the two scales and `fit=outside` from the larger, and each keeps both sides at
+  1 or more too, so a 1x8192 source at `0x100` with `fit=contain`, and a 49x8192
+  source at `0x100` or a 49x49 source at `1x1` with `fit=outside`, no longer
+  answer 500. An 8192x1 source at `1300x0` now comes out at 1300x1, and the
+  process running it alone peaked at about 34 MiB resident. With `fit=outside`
+  the image is scaled to cover the size, so that source at `100x0` comes out at
+  8192x1. `README.md` now says this under Signature Specification, and says how
+  a `0` side is worked out where it said `0` meant the original.
+- 2026-10-16 the generator page lower-cases the source host and writes an empty
+  source path as `/` before it makes the encrypted URL, as `/v1/image/` does
+  (closes #293). A host written in upper and in lower case, or a source URL with
+  no path and with the path `/`, made two cache keys and two upstream requests
+  for one image.
+- 2026-10-16 the source URL a request fetches is built with `net/url` from the
+  source host, path and query, not joined as text (closes #288). The source
+  path, decoded, is fetched percent-encoded again, so `100%25.png` fetches
+  `/100%25.png` where it answered 401, and `my%20cat.png` and `my%2520cat.png`
+  fetch different paths where both fetched `/my%20cat.png` under two cache keys.
+  The source query stays as the source URL writes it; the characters a browser
+  percent-encodes in a query, such as `'` or a space, which was sent raw and
+  answered 502, are percent-encoded when the image URL is read, so the query is
+  signed and cached as it is sent. The generator page answers 400, showing the
+  URL pixa would fetch, for a source URL its image URL would fetch written
+  another way, such as with `%2F` in the path, whatever else the path holds, or
+  with a user name and password. It also answers 400 for a source path holding a
+  `..` segment, as `/v1/image/` refuses one.
+- 2026-10-15 a thin source requested with `fit=cover` no longer makes libvips
+  build a huge image (closes #282): libvips' thumbnail scaled the whole source
+  to cover the requested size and copied that into memory before it cut out the
+  centre, so an 8192x64 source at 8192x8192 made it allocate about 24 GiB, and
+  an 8192x1 source answered 500. The image processor now first cuts the centre
+  of the source, widened to whole source pixels, and thumbnail scales and cuts
+  only that, by the same amount on both sides as before. What it copies into
+  memory is then larger than the output by less than one scaled source pixel on
+  each side. With the runtime image's libvips and JPEG output at 8192x8192,
+  resident memory peaked at about 242 MiB for the 8192x64 source and 431 MiB for
+  the 8192x1 one. The output can sit a fraction of an output pixel from where it
+  was.
+- 2026-10-15 a `/v1/image/` URL whose source path or source query holds a `#`,
+  written as `%23`, is refused with 400 (closes #286): the source URL built from
+  it ended at the `#` when fetched, so each text after it fetched the same
+  upstream image, without a signature for an allowlisted host, and cached it
+  under keys of its own. `README.md` says neither may hold a `#`.
+- 2026-10-15 the host of a `/v1/image/` URL must be a host name or an IPv4
+  address (letters, digits, hyphens, underscores and dots) or an IPv6 address in
+  brackets, either one optionally followed by `:` and a port, and any other host
+  is refused with 400 (closes #275): only `@` was refused, so a `?` written as
+  `%3F` gave the host `cdn.example.com?x=1`, which Go's URL parser reads as
+  `cdn.example.com` with a query. Such a URL passed the allowlist, and each text
+  after the `?` fetched the host's root path and cached the answer under keys of
+  its own. A `#`, `%`, `\` or space written percent-encoded is refused the same
+  way, and so is an IPv6 address without brackets, which needed a signature
+  before. Underscores are kept because `allowlist_hosts` accepts host names
+  holding them.
+- 2026-10-15 the README says how a source URL with a query is written into an
+  image URL, and the image route accepts it (closes #255): the query follows the
+  source path after `%3F`. pixa now reads the image URL's path decoded once, so
+  the `?` before the query, any `?` in it and every `%` in the source path or
+  query must be percent-encoded, and any other character may be. The route used
+  to parse chi's wildcard, which holds the path as the client wrote it whenever
+  that is not Go's own encoding of it; a `/`, `;` or `,` in the query written as
+  `%2F`, `%3B` or `%2C`, as `GenerateSignedURL` wrote them, kept the `%3F` from
+  being read as `?`, and the request failed with 401. `GenerateSignedURL` now
+  writes Go's own encoding of the path. The README has an example with a query
+  and its signature. The path traversal check no longer reads the source query,
+  and keeps a `%` that does not start a percent-escape as it is instead of
+  refusing the URL, so a source path or query holding `%25` is accepted, while a
+  double-encoded `..` next to such a `%` is still refused.
 - 2026-10-15 `max_concurrent_processing` defaults to half the CPUs pixa sees,
   rounded down, and at least 1 (closes #242): it defaulted to one image per CPU,
   and a container with no CPU limit sees every CPU of the host, so on a 48-CPU

internal/config/blocked_networks_internal_test.go

@@ -88,3 +88,56 @@ func TestBlockedNetworksInvalidAbortsStartup(t *testing.T) {
 		},
 	})
 }
+
+// TestBlockedNetworksMappedEntryIsIPv4Range checks that an entry written in
+// the IPv4-mapped form (::ffff:a.b.c.d) is kept as the IPv4 range it covers,
+// since the fetcher matches an IPv4-mapped address as the IPv4 address it
+// holds. /96, the shortest such entry, covers every IPv4 address.
+func TestBlockedNetworksMappedEntryIsIPv4Range(t *testing.T) {
+	t.Parallel()
+
+	yamlContent := signingKeyLine + `blocked_networks:
+  - "::ffff:10.0.0.0/104"
+  - "::ffff:0:0/96"
+`
+
+	c, err := configFromYAML(t, yamlContent)
+	if err != nil {
+		t.Fatalf("IPv4-mapped blocked_networks entries should load, got error: %v", err)
+	}
+
+	want := []string{"10.0.0.0/8", "0.0.0.0/0"}
+	if len(c.BlockedNetworks) != len(want) {
+		t.Fatalf("BlockedNetworks = %v, want %d entries", c.BlockedNetworks, len(want))
+	}
+
+	for i, w := range want {
+		if got := c.BlockedNetworks[i].String(); got != w {
+			t.Errorf("BlockedNetworks[%d] = %q, want %q", i, got, w)
+		}
+	}
+}
+
+// TestBlockedNetworksMappedEntryShorterThan96AbortsStartup checks that an
+// entry in the IPv4-mapped form shorter than /96, which covers more than the
+// IPv4 addresses, aborts startup with an error naming the key and the value.
+func TestBlockedNetworksMappedEntryShorterThan96AbortsStartup(t *testing.T) {
+	t.Parallel()
+
+	runAbortCases(t, []abortCase{
+		{
+			name: "mapped-shorter-than-96",
+			yaml: signingKeyLine + `blocked_networks:
+  - "::ffff:0.0.0.0/80"
+`,
+			wantErrSubstrings: []string{keyBlockedNetworks, "::ffff:0.0.0.0/80"},
+		},
+		{
+			name: "mapped-95",
+			yaml: signingKeyLine + `blocked_networks:
+  - "::ffff:0:0/95"
+`,
+			wantErrSubstrings: []string{keyBlockedNetworks},
+		},
+	})
+}

internal/config/config.go

@@ -91,6 +91,7 @@ var (
 	errNotABoolean       = errors.New("not a boolean")
 	errNotAStringList    = errors.New("not a list of strings")
 	errNotAValidCIDR     = errors.New("not a valid CIDR network")
+	errMappedTooShort    = errors.New("in the IPv4-mapped form but shorter than /96")
 	errNotAMetricsMap    = errors.New("not a map of metrics settings")
 	errEmptyListEntry    = errors.New("list contains an empty entry")
 	errEmptyEntry        = errors.New("contains an empty entry")
@@ -338,8 +339,7 @@ func newFromSmartConfig(sc *smartconfig.Config) (*Config, error) {
 				settingName(keyDBURL), errValueEmpty)
 		}
 
-		// The driver sets the journal mode only through a _pragma parameter.
+		c.DBURL = defaultDBURL(c.StateDir)
-		c.DBURL = fmt.Sprintf("file:%s/state.sqlite3?_pragma=journal_mode(WAL)", c.StateDir)
 	}
 
 	if loader.err != nil {
@@ -354,6 +354,30 @@ func newFromSmartConfig(sc *smartconfig.Config) (*Config, error) {
 	return c, nil
 }
 
+// defaultDBURL returns the db_url used when db_url is omitted: the file
+// state.sqlite3 in stateDir, in WAL mode, which the driver sets only
+// through a _pragma parameter. String() percent-encodes any %, ? or # in
+// the directory name, which SQLite decodes back, so the URL names exactly
+// this file. OmitHost writes file:/var/lib/pixa/... instead of
+// file:///var/lib/pixa/..., and keeps a relative stateDir from being read
+// as a host name. The path written into the URL is stateDir followed by
+// /state.sqlite3; when it starts with //, as it does for a stateDir of /
+// or one starting with //, the empty host is kept, so the URL starts
+// file://// and SQLite does not read the first name in the path as a
+// host.
+func defaultDBURL(stateDir string) string {
+	dbPath := stateDir + "/state.sqlite3"
+
+	dbURL := url.URL{
+		Scheme:   "file",
+		OmitHost: !strings.HasPrefix(dbPath, "//"),
+		Path:     dbPath,
+		RawQuery: "_pragma=journal_mode(WAL)",
+	}
+
+	return dbURL.String()
+}
+
 // validateKnownKeys rejects configuration files containing keys the
 // application does not understand, so typos fail at startup instead of
 // being silently ignored, and rejects keys that are explicitly set to
@@ -1225,6 +1249,13 @@ func defaultTrustedProxies() []netip.Prefix {
 // wrong type, an empty entry, a non-string entry, or an unparseable CIDR
 // aborts startup naming the key and the offending value; the default
 // (an empty list) applies only to an omitted key.
+//
+// An entry written in the IPv4-mapped form (::ffff:a.b.c.d) is kept as the
+// IPv4 range it covers: ::ffff:10.0.0.0/104 is kept as 10.0.0.0/8. Both
+// blocked_networks and trusted_proxies are matched against an IPv4-mapped
+// address as the IPv4 address it holds, so the entry as written would match
+// nothing. Such an entry shorter than /96 covers more than the IPv4
+// addresses and aborts startup.
 func parseCIDRList(sc *smartconfig.Config, key string) ([]netip.Prefix, error) {
 	raw, ok := lookupValue(sc, key)
 	if !ok {
@@ -1240,6 +1271,10 @@ func parseCIDRList(sc *smartconfig.Config, key string) ([]netip.Prefix, error) {
 		return nil, err
 	}
 
+	// An IPv4-mapped address is the 96 bits of ::ffff:0:0/96 followed by
+	// the 32 bits of the IPv4 address it holds.
+	const mappedPrefixBits = 96
+
 	prefixes := make([]netip.Prefix, 0, len(entries))
 
 	for _, entry := range entries {
@@ -1249,6 +1284,16 @@ func parseCIDRList(sc *smartconfig.Config, key string) ([]netip.Prefix, error) {
 				settingName(key), entry, errNotAValidCIDR)
 		}
 
+		if prefix.Addr().Is4In6() && prefix.Bits() < mappedPrefixBits {
+			return nil, fmt.Errorf("%s: value %q is %w",
+				settingName(key), entry, errMappedTooShort)
+		}
+
+		if prefix.Addr().Is4In6() {
+			prefix = netip.PrefixFrom(prefix.Addr().Unmap(),
+				prefix.Bits()-mappedPrefixBits)
+		}
+
 		prefixes = append(prefixes, prefix)
 	}
 

internal/config/config_validation_internal_test.go

@@ -135,6 +135,104 @@ func TestDefaultDBURLOpensTheDatabaseInWALMode(t *testing.T) {
 	}
 }
 
+// TestDerivedDBURLOpensStateDirHoldingURLCharacters opens the db_url
+// derived from a state_dir whose name holds %41, ? and #, and checks that
+// the database is state.sqlite3 in that directory, in WAL mode. Put into
+// a file: URL as it is, the name would lose its meaning: %41 is read as
+// A, and the path ends at the ?, which also loses the journal mode.
+func TestDerivedDBURLOpensStateDirHoldingURLCharacters(t *testing.T) {
+	t.Parallel()
+
+	stateDir := filepath.Join(t.TempDir(), "a%41b?c#d")
+
+	err := os.Mkdir(stateDir, 0o750)
+	if err != nil {
+		t.Fatalf("failed to create %q: %v", stateDir, err)
+	}
+
+	// Quoted so YAML takes the whole name as it is.
+	c, err := configFromYAML(t, signingKeyLine+"state_dir: '"+stateDir+"'\n")
+	if err != nil {
+		t.Fatalf("config with state_dir %q should be valid, got: %v", stateDir, err)
+	}
+
+	t.Logf("state_dir %q gives db_url %q", stateDir, c.DBURL)
+
+	checkDBURLOpensFileInWALMode(t, c.DBURL, filepath.Join(stateDir, "state.sqlite3"))
+}
+
+// TestDerivedDBURLOpensStateDirStartingWithTwoSlashes opens the db_url
+// derived from a state_dir that starts with //, and checks that the
+// database is state.sqlite3 in that directory, in WAL mode. Written right
+// after file:, such a name gives a URL starting file://, and SQLite reads
+// the first directory name as a host and refuses to open it.
+func TestDerivedDBURLOpensStateDirStartingWithTwoSlashes(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	stateDir := "/" + dir
+
+	c, err := configFromYAML(t, signingKeyLine+"state_dir: "+stateDir+"\n")
+	if err != nil {
+		t.Fatalf("config with state_dir %q should be valid, got: %v", stateDir, err)
+	}
+
+	t.Logf("state_dir %q gives db_url %q", stateDir, c.DBURL)
+
+	checkDBURLOpensFileInWALMode(t, c.DBURL, filepath.Join(dir, "state.sqlite3"))
+}
+
+// TestDerivedDBURLForRootStateDir checks the db_url derived from a
+// state_dir of /. The path written into the URL is then
+// //state.sqlite3, and written right after file: it gives a URL starting
+// file://, where SQLite reads state.sqlite3 as a host and refuses to open
+// it. The test only checks the URL, since it cannot write to /.
+func TestDerivedDBURLForRootStateDir(t *testing.T) {
+	t.Parallel()
+
+	c, err := configFromYAML(t, signingKeyLine+"state_dir: /\n")
+	if err != nil {
+		t.Fatalf("config with state_dir / should be valid, got: %v", err)
+	}
+
+	t.Logf("state_dir / gives db_url %q", c.DBURL)
+
+	wantDBURL := "file:////state.sqlite3?_pragma=journal_mode(WAL)"
+	if c.DBURL != wantDBURL {
+		t.Errorf("DBURL = %q, want %q", c.DBURL, wantDBURL)
+	}
+}
+
+// checkDBURLOpensFileInWALMode opens dbURL with the SQLite driver pixad
+// uses, and checks that the database is in WAL mode and is the file
+// dbFile.
+func checkDBURLOpensFileInWALMode(t *testing.T, dbURL, dbFile string) {
+	t.Helper()
+
+	db, err := sql.Open("sqlite", dbURL)
+	if err != nil {
+		t.Fatalf("failed to open %q: %v", dbURL, err)
+	}
+
+	t.Cleanup(func() { _ = db.Close() })
+
+	var journalMode string
+
+	err = db.QueryRowContext(t.Context(), "PRAGMA journal_mode").Scan(&journalMode)
+	if err != nil {
+		t.Fatalf("failed to read the journal mode of %q: %v", dbURL, err)
+	}
+
+	if journalMode != "wal" {
+		t.Errorf("journal mode of %q = %q, want wal", dbURL, journalMode)
+	}
+
+	_, err = os.Stat(dbFile)
+	if err != nil {
+		t.Errorf("opening %q did not create %q: %v", dbURL, dbFile, err)
+	}
+}
+
 func TestExplicitValidValuesAreUsed(t *testing.T) {
 	t.Parallel()
 

internal/config/trusted_proxies_internal_test.go

@@ -83,3 +83,41 @@ func TestTrustedProxiesInvalidAbortsStartup(t *testing.T) {
 		},
 	})
 }
+
+// TestTrustedProxiesMappedEntryIsIPv4Range checks that an entry written in
+// the IPv4-mapped form (::ffff:a.b.c.d) is kept as the IPv4 range it covers,
+// since the client address is matched as the IPv4 address it holds. /96, the
+// shortest such entry, covers every IPv4 address.
+func TestTrustedProxiesMappedEntryIsIPv4Range(t *testing.T) {
+	t.Parallel()
+
+	c, err := configFromYAML(t, signingKeyLine+
+		`trusted_proxies: ["::ffff:10.0.0.0/104", "::ffff:0:0/96"]`+"\n")
+	if err != nil {
+		t.Fatalf("IPv4-mapped trusted_proxies entries should load: %v", err)
+	}
+
+	got := make([]string, len(c.TrustedProxies))
+	for i, p := range c.TrustedProxies {
+		got[i] = p.String()
+	}
+
+	if joined := strings.Join(got, ","); joined != "10.0.0.0/8,0.0.0.0/0" {
+		t.Errorf("TrustedProxies = %v, want 10.0.0.0/8 and 0.0.0.0/0", got)
+	}
+}
+
+// TestTrustedProxiesMappedEntryShorterThan96AbortsStartup checks that an
+// entry in the IPv4-mapped form shorter than /96, which covers more than the
+// IPv4 addresses, aborts startup with an error naming the key and the value.
+func TestTrustedProxiesMappedEntryShorterThan96AbortsStartup(t *testing.T) {
+	t.Parallel()
+
+	runAbortCases(t, []abortCase{
+		{
+			name:              "mapped-95",
+			yaml:              signingKeyLine + `trusted_proxies: ["::ffff:0:0/95"]` + "\n",
+			wantErrSubstrings: []string{keyTrustedProxies, "::ffff:0:0/95"},
+		},
+	})
+}

internal/handlers/auth.go

@@ -9,7 +9,9 @@ import (
 	"math"
 	"net/http"
 	"net/url"
+	"slices"
 	"strconv"
+	"strings"
 	"time"
 
 	"sneak.berlin/go/pixa/internal/clientip"
@@ -125,7 +127,7 @@ func (s *Handlers) HandleGenerateURL() http.HandlerFunc {
 		sourceURL := r.FormValue("url")
 
 		// Validate source URL
-		parsed, err := url.Parse(sourceURL)
+		parsed, err := parseSourceURL(sourceURL)
 		if err != nil || parsed.Host == "" {
 			s.renderGeneratorWithForm(w, r, "Invalid source URL", r.Form,
 				http.StatusBadRequest)
@@ -133,6 +135,16 @@ func (s *Handlers) HandleGenerateURL() http.HandlerFunc {
 			return
 		}
 
+		// /v1/image/ refuses a ".." segment in the source path, and so does
+		// the generator. The path is decoded, so %2e%2e counts too
+		if slices.Contains(strings.Split(parsed.Path, "/"), "..") {
+			s.renderGeneratorWithForm(w, r,
+				"Source URL path cannot hold a .. segment", r.Form,
+				http.StatusBadRequest)
+
+			return
+		}
+
 		payload, expiresAt, ttl, err := buildGeneratePayload(parsed, r.Form)
 		if err != nil {
 			s.renderGeneratorWithForm(w, r, err.Error(), r.Form,
@@ -141,6 +153,19 @@ func (s *Handlers) HandleGenerateURL() http.HandlerFunc {
 			return
 		}
 
+		// The image URL must fetch the source URL as it was written
+		sourceRequest := payload.ToImageRequest()
+		sourceRequest.AllowHTTP = s.config.AllowHTTP
+		fetchURL := sourceRequest.SourceURL()
+
+		if !fetchedAsWritten(parsed, fetchURL) {
+			s.renderGeneratorWithForm(w, r, "Source URL cannot be fetched as "+
+				"written: pixa would fetch "+fetchURL,
+				r.Form, http.StatusBadRequest)
+
+			return
+		}
+
 		// Generate encrypted token
 		token, err := s.encGen.Generate(payload)
 		if err != nil {
@@ -173,6 +198,48 @@ func (s *Handlers) HandleGenerateURL() http.HandlerFunc {
 	}
 }
 
+// parseSourceURL parses the source URL entered on the generator page with its
+// host lower-cased and an empty path written as "/", as /v1/image/ reads a
+// source URL, so that each spelling of one source URL makes one cache key and
+// one upstream request.
+func parseSourceURL(sourceURL string) (*url.URL, error) {
+	parsed, err := url.Parse(sourceURL)
+	if err != nil {
+		return nil, err
+	}
+
+	parsed.Host = strings.ToLower(parsed.Host)
+
+	if parsed.Path == "" {
+		parsed.Path = "/"
+	}
+
+	return parsed, nil
+}
+
+// fetchedAsWritten reports whether fetchURL, the URL pixa fetches for an
+// image URL made from the source URL source, is source as it was written,
+// apart from the scheme, which the token does not keep. It is not when source
+// holds a user name and password, which the token does not keep either, or a
+// "?" with no query after it. The token keeps the source path decoded and
+// pixa fetches it percent-encoded again, so a path written another way, such
+// as with %2F, is fetched as another path. url.URL keeps the path as written
+// in RawPath when it is not the decoded path percent-encoded again.
+// EscapedPath ignores RawPath when it also holds a character net/url encodes,
+// such as a space, so RawPath is compared itself.
+func fetchedAsWritten(source *url.URL, fetchURL string) bool {
+	writtenPath := source.RawPath
+	if writtenPath == "" {
+		writtenPath = source.EscapedPath()
+	}
+
+	fetched, err := url.Parse(fetchURL)
+
+	return err == nil && source.User == nil && !source.ForceQuery &&
+		fetched.EscapedPath() == writtenPath &&
+		fetched.RawQuery == source.RawQuery
+}
+
 // buildGeneratePayload parses the numeric form fields and assembles the
 // encrypted URL payload. ttl=0 means never expires (ExpiresAt stays 0). A
 // non-numeric or out-of-range width, height, quality or ttl, or an

internal/handlers/image.go

@@ -9,7 +9,6 @@ import (
 	"strconv"
 	"time"
 
-	"github.com/go-chi/chi/v5"
 	"github.com/go-chi/chi/v5/middleware"
 	"sneak.berlin/go/pixa/internal/encurl"
 	"sneak.berlin/go/pixa/internal/httpfetcher"
@@ -80,20 +79,21 @@ func (s *Handlers) HandleRobotsTxt() http.HandlerFunc {
 	}
 }
 
-// parseImageRequest parses the wildcard path and query parameters into
+// parseImageRequest parses the image URL's path and query parameters into
 // an ImageRequest. On invalid input it writes an error response and
 // returns false.
 func (s *Handlers) parseImageRequest(
 	w http.ResponseWriter, r *http.Request,
 ) (*imgcache.ImageRequest, bool) {
-	// Get the wildcard path from chi
+	// The path is parsed decoded once, as r.URL.Path holds it, and not from
-	pathParam := chi.URLParam(r, "*")
+	// chi's wildcard: chi routes on the path as the client wrote it whenever
-
+	// that is not Go's own encoding of it, so its wildcard would still hold
-	// Parse the URL path
+	// the %3F before a source query when, say, a "/" in the query is written
-	parsed, err := imgcache.ParseImagePath(pathParam)
+	// as %2F
+	parsed, err := imgcache.ParseImageURL(r.URL.Path)
 	if err != nil {
 		s.log.Warn("failed to parse image URL",
-			"path", pathParam,
+			"path", r.URL.Path,
 			"error", err,
 		)
 		s.respondError(w, "invalid image URL: "+err.Error(), http.StatusBadRequest)

internal/handlers/image_errors_internal_test.go

@@ -264,3 +264,127 @@ func TestHandleImage_AllowlistOrSignature(t *testing.T) {
 		})
 	}
 }
+
+// TestHandleImage_InvalidHost_Returns400 sends image URLs whose host no URL
+// host can be, and checks that each is refused with 400 naming the host as
+// decoded: allowlistedHost with a character after it that a URL host cannot
+// hold, percent-encoded, an IPv6 address without brackets, and brackets
+// around something that is not an IPv6 address. A "?" or "#" after
+// allowlistedHost would otherwise end the host when the source URL is read,
+// so the request would pass the allowlist and be fetched and cached under a
+// host that is not the one fetched.
+func TestHandleImage_InvalidHost_Returns400(t *testing.T) {
+	t.Parallel()
+
+	photos := newPhotoFetcher(t, allowlistedHost)
+
+	tests := []struct {
+		name        string
+		writtenHost string
+		decodedHost string
+	}{
+		{"? written as %3F", allowlistedHost + "%3Fx=1", allowlistedHost + "?x=1"},
+		{"# written as %23", allowlistedHost + "%23x", allowlistedHost + "#x"},
+		{"% written as %25", allowlistedHost + "%253F", allowlistedHost + "%3F"},
+		{`\ written as %5C`, allowlistedHost + "%5Cx", allowlistedHost + `\x`},
+		{"space written as %20", allowlistedHost + "%20", allowlistedHost + " "},
+		{"IPv6 address without brackets", "2001:db8::1", "2001:db8::1"},
+		{"IPv4 address in brackets", "[203.0.113.5]", "[203.0.113.5]"},
+		{"no address in brackets", "[abc]", "[abc]"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			rec := sendGet(t, newImageRoute(t, photos), photoURL(tt.writtenHost))
+			checkErrorBody(t, rec, http.StatusBadRequest,
+				fmt.Sprintf("invalid image URL: invalid source host: %q",
+					tt.decodedHost))
+		})
+	}
+}
+
+// TestHandleImage_HostForms_Served checks that the image route serves, with
+// a valid signature, the image at photoPath on a host name, an IPv4 address,
+// an IPv6 address in brackets, and a host followed by ":" with or without a
+// port.
+func TestHandleImage_HostForms_Served(t *testing.T) {
+	t.Parallel()
+
+	hosts := []string{
+		allowlistedHost,
+		"img_1.example.com",
+		"203.0.113.5",
+		"[2001:db8::1]",
+		allowlistedHost + ":8443",
+		"[2001:db8::1]:8443",
+		allowlistedHost + ":",
+	}
+
+	photos := newPhotoFetcher(t, hosts...)
+	exp := time.Now().Add(time.Hour)
+
+	for _, host := range hosts {
+		t.Run(host, func(t *testing.T) {
+			t.Parallel()
+
+			sig := photoSignature(testSigningKey, host, exp)
+
+			rec := sendGet(t, newImageRoute(t, photos),
+				photoURLWithSig(host, sig, exp))
+			if rec.Code != http.StatusOK {
+				t.Errorf("status = %d, want %d: %s",
+					rec.Code, http.StatusOK, rec.Body)
+			}
+		})
+	}
+}
+
+// TestHandleImage_HashInSourcePathOrQuery_Returns400 sends image URLs whose
+// source path or source query holds a "#", written as %23, and checks that
+// each is refused with 400, the one signed for that source path included. The
+// source URL would otherwise end at the "#" when it is fetched, so each text
+// after it would fetch the image at photoPath and cache it under keys of its
+// own.
+func TestHandleImage_HashInSourcePathOrQuery_Returns400(t *testing.T) {
+	t.Parallel()
+
+	photos := newPhotoFetcher(t, allowlistedHost, signedHost)
+	exp := time.Now().Add(time.Hour)
+
+	sig := signature.New(testSigningKey).Sign(&signature.Request{
+		SourceHost: signedHost,
+		SourcePath: photoPath + "#x",
+		Width:      50,
+		Height:     50,
+		Format:     string(imgcache.FormatJPEG),
+		Quality:    85,
+		FitMode:    string(imgcache.FitCover),
+		Expires:    exp,
+	})
+
+	tests := []struct {
+		name   string
+		target string
+	}{
+		{"# in the source path",
+			"/v1/image/" + allowlistedHost + photoPath + "%23x/50x50.jpeg"},
+		{"# in the source query",
+			"/v1/image/" + allowlistedHost + photoPath + "%3Fv=1%23x/50x50.jpeg"},
+		{"# in the source path, signed",
+			fmt.Sprintf("/v1/image/%s%s%%23x/50x50.jpeg?sig=%s&exp=%d",
+				signedHost, photoPath, sig, exp.Unix())},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			rec := sendGet(t, newImageRoute(t, photos), tt.target)
+			checkErrorBody(t, rec, http.StatusBadRequest,
+				`invalid image URL: invalid image path: `+
+					`the source path or query holds a "#"`)
+		})
+	}
+}

internal/handlers/image_signature_internal_test.go

@@ -169,3 +169,166 @@ func TestHandleImage_InvalidExp_Returns400(t *testing.T) {
 		})
 	}
 }
+
+// TestHandleImage_GeneratedSignedURLWithSourceQueryVerifies sends URLs built
+// by the service's signed-URL generator for source URLs with a query to the
+// route, through a chi router: chi routes on the path as the client wrote it
+// whenever that is not Go's own encoding of it. A 502 means the signature
+// passed, so the route read the source path and query that were signed.
+func TestHandleImage_GeneratedSignedURLWithSourceQueryVerifies(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name  string
+		query string
+	}{
+		{"slash and colon", "url=https://origin.example.com/cat.jpg"},
+		{"semicolon", "a=1;b=2"},
+		{"comma", "crop=0,0,50,50"},
+		{"equals and ampersand", "a=1&b=2"},
+		{"percent-escape", "name=my%20cat"},
+		{"question mark", "next=/a?b=1"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			fix := setupTestHandler(t)
+
+			signedURL, err := fix.service.GenerateSignedURL("", &imgcache.ImageRequest{
+				SourceHost:  signedHost,
+				SourcePath:  photoPath,
+				SourceQuery: tt.query,
+				Size:        imgcache.Size{Width: 50, Height: 50},
+				Format:      imgcache.FormatJPEG,
+			}, time.Hour)
+			if err != nil {
+				t.Fatalf("GenerateSignedURL() error = %v", err)
+			}
+
+			status := getImage(t, fix, signedURL)
+			if status != http.StatusBadGateway {
+				t.Errorf("status = %d, want %d", status, http.StatusBadGateway)
+			}
+		})
+	}
+}
+
+// TestHandleImage_SourceQueryEncodedAnyWay signs one source URL with a query
+// and writes it into the image URL in ways the README allows: the "?" before
+// the query and each "%" in it percent-encoded, and every other character as
+// itself or percent-encoded. A 502 means the signature passed.
+func TestHandleImage_SourceQueryEncodedAnyWay(t *testing.T) {
+	t.Parallel()
+
+	expires := time.Now().Add(time.Hour)
+	signer := signature.New("test-signing-key-must-be-32-chars")
+	sig := signer.Sign(&signature.Request{
+		SourceHost:  signedHost,
+		SourcePath:  photoPath,
+		SourceQuery: "url=https://origin.example.com/my%20cat.jpg",
+		Width:       50,
+		Height:      50,
+		Format:      string(imgcache.FormatJPEG),
+		Quality:     85,
+		FitMode:     string(imgcache.FitCover),
+		Expires:     expires,
+	})
+
+	tests := []struct {
+		name         string
+		writtenQuery string
+	}{
+		{"only ? and % encoded",
+			"%3Furl=https://origin.example.com/my%2520cat.jpg"},
+		{"every reserved character encoded",
+			"%3Furl%3Dhttps%3A%2F%2Forigin.example.com%2Fmy%2520cat.jpg"},
+		{"lower-case hex digits",
+			"%3furl=https://origin.example.com/my%2520cat.jpg"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			target := fmt.Sprintf("/v1/image/%s%s%s/50x50.jpeg?sig=%s&exp=%d",
+				signedHost, photoPath, tt.writtenQuery, sig, expires.Unix())
+
+			status := getImage(t, setupTestHandler(t), target)
+			if status != http.StatusBadGateway {
+				t.Errorf("status = %d, want %d", status, http.StatusBadGateway)
+			}
+		})
+	}
+}
+
+// TestHandleImage_SourceURLHoldingPercent25Verifies signs a source path and a
+// source query that each hold %25, the percent-escape of a literal "%", and
+// sends each to the route written as the README says, with that "%" written as
+// %25, and as GenerateSignedURL writes it. A 502 means the signature passed.
+func TestHandleImage_SourceURLHoldingPercent25Verifies(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name        string
+		sourcePath  string
+		sourceQuery string
+		// writtenPathAndQuery is the source path and query as the README says
+		// to write them into the image URL
+		writtenPathAndQuery string
+	}{
+		{"path", "/sale/50%25off.jpg", "", "/sale/50%2525off.jpg"},
+		{"query", photoPath, "discount=50%25", photoPath + "%3Fdiscount=50%2525"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name+" written by the README rule", func(t *testing.T) {
+			t.Parallel()
+
+			expires := time.Now().Add(time.Hour)
+			signer := signature.New("test-signing-key-must-be-32-chars")
+			sig := signer.Sign(&signature.Request{
+				SourceHost:  signedHost,
+				SourcePath:  tt.sourcePath,
+				SourceQuery: tt.sourceQuery,
+				Width:       50,
+				Height:      50,
+				Format:      string(imgcache.FormatJPEG),
+				Quality:     85,
+				FitMode:     string(imgcache.FitCover),
+				Expires:     expires,
+			})
+
+			target := fmt.Sprintf("/v1/image/%s%s/50x50.jpeg?sig=%s&exp=%d",
+				signedHost, tt.writtenPathAndQuery, sig, expires.Unix())
+
+			status := getImage(t, setupTestHandler(t), target)
+			if status != http.StatusBadGateway {
+				t.Errorf("status = %d, want %d", status, http.StatusBadGateway)
+			}
+		})
+
+		t.Run(tt.name+" written by GenerateSignedURL", func(t *testing.T) {
+			t.Parallel()
+
+			fix := setupTestHandler(t)
+
+			signedURL, err := fix.service.GenerateSignedURL("", &imgcache.ImageRequest{
+				SourceHost:  signedHost,
+				SourcePath:  tt.sourcePath,
+				SourceQuery: tt.sourceQuery,
+				Size:        imgcache.Size{Width: 50, Height: 50},
+				Format:      imgcache.FormatJPEG,
+			}, time.Hour)
+			if err != nil {
+				t.Fatalf("GenerateSignedURL() error = %v", err)
+			}
+
+			status := getImage(t, fix, signedURL)
+			if status != http.StatusBadGateway {
+				t.Errorf("status = %d, want %d", status, http.StatusBadGateway)
+			}
+		})
+	}
+}

internal/httpfetcher/httpfetcher.go

@@ -59,15 +59,17 @@ const (
 // builtinBlockedPrefixes are internal or special-use ranges that Go's
 // net.IP predicates (IsPrivate, IsLinkLocalUnicast, and the like) do not
 // already cover. They are always blocked, in addition to any
-// operator-supplied networks. IPv4-mapped IPv6 addresses are unmapped
+// operator-supplied networks. IPv4-mapped and IPv4-compatible IPv6
-// before matching, so these IPv4 ranges are caught in both forms.
+// addresses are read as the IPv4 address they hold before matching, so
+// these IPv4 ranges are caught in all three forms.
 //
 //nolint:gochecknoglobals // immutable built-in blocklist
 var builtinBlockedPrefixes = []netip.Prefix{
-	netip.MustParsePrefix("100.64.0.0/10"), // RFC 6598 CGNAT / carrier-grade NAT
+	netip.MustParsePrefix("100.64.0.0/10"),  // RFC 6598 CGNAT / carrier-grade NAT
-	netip.MustParsePrefix("192.0.0.0/24"),  // RFC 6890 IETF protocol assignments
+	netip.MustParsePrefix("192.0.0.0/24"),   // RFC 6890 IETF protocol assignments
-	netip.MustParsePrefix("198.18.0.0/15"), // RFC 2544 benchmarking range
+	netip.MustParsePrefix("198.18.0.0/15"),  // RFC 2544 benchmarking range
-	netip.MustParsePrefix("64:ff9b::/96"),  // RFC 6052 NAT64 (maps onto IPv4)
+	netip.MustParsePrefix("64:ff9b::/96"),   // RFC 6052 NAT64 (maps onto IPv4)
+	netip.MustParsePrefix("64:ff9b:1::/48"), // RFC 8215 local-use NAT64 (maps onto IPv4)
 }
 
 // Fetcher errors.
@@ -473,16 +475,16 @@ func validateURL(ctx context.Context, rawURL string, allowHTTP bool) error {
 		return ErrUnsupportedScheme
 	}
 
-	// Parse to extract host
+	parsedURL, err := neturl.Parse(rawURL)
-	host := extractHost(rawURL)
+	if err != nil {
-	if host == "" {
+		return fmt.Errorf("%w: %w", ErrInvalidHost, err)
-		return ErrInvalidHost
 	}
 
-	// Remove port if present
+	// Hostname drops any port, and the brackets of an IPv6 address whether
-	h, _, err := net.SplitHostPort(host)
+	// or not a port follows them
-	if err == nil {
+	host := parsedURL.Hostname()
-		host = h
+	if host == "" {
+		return ErrInvalidHost
 	}
 
 	// Block obvious localhost patterns
@@ -543,6 +545,9 @@ func isPrivateIP(ip net.IP) bool {
 		return true
 	}
 
+	// An IPv4-compatible IPv6 address is judged as the IPv4 address it holds
+	ip = ipv4CompatibleToIPv4(ip)
+
 	// Check for loopback
 	if ip.IsLoopback() {
 		return true
@@ -594,7 +599,12 @@ func isPrivateIP(ip net.IP) bool {
 }
 
 // isBlockedIP reports whether ip is refused, either by the built-in
-// internal-range check or by one of the operator-supplied prefixes.
+// internal-range check or by one of the operator-supplied prefixes. An
+// IPv4-compatible address (in ::/96) is refused when a prefix contains the
+// address as written or the IPv4 address it holds. An IPv4-mapped address (in
+// ::ffff:0:0/96) is matched only as the IPv4 address it holds: Go keeps a
+// plain IPv4 address in that form too, and an IPv6 prefix such as ::/0 must
+// not refuse every IPv4 address.
 func isBlockedIP(ip net.IP, blocked []netip.Prefix) bool {
 	if isPrivateIP(ip) {
 		return true
@@ -607,11 +617,28 @@ func isBlockedIP(ip net.IP, blocked []netip.Prefix) bool {
 
 	addr = addr.Unmap()
 
+	// ip read as IPv4 where it is or holds an IPv4 address, otherwise the
+	// zero netip.Addr, which no prefix contains. Matching asIPv4 as well as
+	// addr refuses more only for an IPv4-compatible address.
+	asIPv4, _ := netip.AddrFromSlice(ipv4CompatibleToIPv4(ip).To4())
+
 	return slices.ContainsFunc(blocked, func(prefix netip.Prefix) bool {
-		return prefix.Contains(addr)
+		return prefix.Contains(addr) || prefix.Contains(asIPv4)
 	})
 }
 
+// ipv4CompatibleToIPv4 returns the IPv4 address that an IPv4-compatible IPv6
+// address (in ::/96, such as ::127.0.0.1) holds in its last four bytes, and
+// any other address unchanged.
+func ipv4CompatibleToIPv4(ip net.IP) net.IP {
+	addr, ok := netip.AddrFromSlice(ip)
+	if ok && netip.MustParsePrefix("::/96").Contains(addr) {
+		return ip[12:]
+	}
+
+	return ip
+}
+
 // ssrfSafeDialer validates IP addresses against the built-in blocked ranges
 // before connecting. New wraps dialSSRFSafe with the operator-supplied
 // blocklist; this entry point enforces the built-in ranges alone.

internal/httpfetcher/httpfetcher_internal_test.go

@@ -186,6 +186,44 @@ func TestIsPrivateIP(t *testing.T) {
 	}
 }
 
+// TestIsPrivateIPJudgesEmbeddedIPv4Address checks IPv6 addresses that hold an
+// IPv4 address in their last four bytes: the IPv4-compatible form (::/96) and
+// the NAT64 form (64:ff9b::/96). One holding a loopback, private or
+// link-local IPv4 address is refused as that address is, and one holding a
+// public IPv4 address in the IPv4-compatible form passes.
+func TestIsPrivateIPJudgesEmbeddedIPv4Address(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		ip   string
+		want bool
+	}{
+		{"::127.0.0.1", true},       // loopback
+		{"::10.0.0.1", true},        // private
+		{"::169.254.169.254", true}, // link-local
+		{"64:ff9b::7f00:1", true},   // NAT64 form of 127.0.0.1
+		{"::8.8.8.8", false},        // public
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.ip, func(t *testing.T) {
+			t.Parallel()
+
+			ip := net.ParseIP(tc.ip)
+			if ip == nil {
+				t.Fatalf("failed to parse IP %q", tc.ip)
+			}
+
+			got := isPrivateIP(ip)
+			t.Logf("isPrivateIP(%s) = %v", ip, got)
+
+			if got != tc.want {
+				t.Errorf("isPrivateIP(%q) = %v, want %v", tc.ip, got, tc.want)
+			}
+		})
+	}
+}
+
 func TestValidateURL_RejectsNonHTTPS(t *testing.T) {
 	t.Parallel()
 

internal/imageprocessor/computed_side_internal_test.go

@@ -50,3 +50,69 @@ func TestImageProcessor_RefusesComputedSideAboveMaximum(t *testing.T) {
 		})
 	}
 }
+
+// TestImageProcessor_ComputedSideIsAtLeastOne verifies that, for a size with
+// one side 0, the side computed from the input's proportions is 1 where it
+// would round down to 0, and that Process makes that size in every fit mode
+// but FitOutside, which scales the input to cover the size and so can make
+// it longer. Each case checks the size targetDimensions gives before it
+// calls Process: with that side at 0, an 8192x1 input at 1300x0 makes
+// libvips ask for about 39 GB with FitCover.
+func TestImageProcessor_ComputedSideIsAtLeastOne(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name                    string
+		inputWidth, inputHeight int
+		size                    Size
+		wantWidth, wantHeight   int
+	}{
+		{"8192x1 at 100x0", 8192, 1, Size{Width: 100}, 100, 1},
+		{"8192x1 at 1300x0", 8192, 1, Size{Width: 1300}, 1300, 1},
+		{"8192x2 at 100x0", 8192, 2, Size{Width: 100}, 100, 1},
+		{"1x8192 at 0x100", 1, 8192, Size{Height: 100}, 1, 100},
+		{"1x8192 at 0x1300", 1, 8192, Size{Height: 1300}, 1, 1300},
+		{"2x8192 at 0x100", 2, 8192, Size{Height: 100}, 1, 100},
+	}
+
+	fits := []FitMode{FitCover, FitContain, FitFill, FitInside, FitOutside}
+
+	for _, tt := range tests {
+		for _, fit := range fits {
+			t.Run(tt.name+" "+string(fit), func(t *testing.T) {
+				t.Parallel()
+
+				width, height, err := targetDimensions(tt.size,
+					tt.inputWidth, tt.inputHeight)
+				if err != nil || width != tt.wantWidth || height != tt.wantHeight {
+					t.Fatalf("targetDimensions() = %dx%d, %v, want %dx%d",
+						width, height, err, tt.wantWidth, tt.wantHeight)
+				}
+
+				input := createTestJPEG(t, tt.inputWidth, tt.inputHeight)
+
+				result, err := New(Params{}).Process(t.Context(),
+					bytes.NewReader(input),
+					&Request{Size: tt.size, Format: FormatJPEG, FitMode: fit})
+				if err != nil {
+					t.Fatalf("Process() error = %v", err)
+				}
+
+				defer func() { _ = result.Content.Close() }()
+
+				t.Logf("Process() made %dx%d", result.Width, result.Height)
+
+				ok := result.Width == tt.wantWidth && result.Height == tt.wantHeight
+				if fit == FitOutside {
+					ok = result.Width >= tt.wantWidth && result.Height >= tt.wantHeight
+				}
+
+				if !ok {
+					t.Errorf("Process() made %dx%d, want %dx%d (at least, for %s)",
+						result.Width, result.Height, tt.wantWidth, tt.wantHeight,
+						FitOutside)
+				}
+			})
+		}
+	}
+}

internal/imageprocessor/fit_cover_internal_test.go

@@ -0,0 +1,268 @@
+package imageprocessor
+
+import (
+	"bytes"
+	"image"
+	"image/color"
+	"image/png"
+	"math"
+	"testing"
+)
+
+// TestImageProcessor_FitCoverServesThinSource verifies that FitCover serves
+// a thin source at the requested size. An 8192x1 source scaled to cover
+// 8192x8192 is 8192 times as long as the output on its long side. The centre
+// of an 8192x1 source with the proportions of 1x8192 is 1/8192 of a pixel
+// wide. Each case is also turned on its side.
+func TestImageProcessor_FitCoverServesThinSource(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name                    string
+		inputWidth, inputHeight int
+		size                    Size
+	}{
+		{"8192x1 at 8192x8192", 8192, 1, Size{Width: 8192, Height: 8192}},
+		{"8192x1 at 1x8192", 8192, 1, Size{Width: 1, Height: 8192}},
+		{"1x8192 at 8192x8192", 1, 8192, Size{Width: 8192, Height: 8192}},
+		{"1x8192 at 8192x1", 1, 8192, Size{Width: 8192, Height: 1}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			input := createTestJPEG(t, tt.inputWidth, tt.inputHeight)
+
+			result, err := New(Params{}).Process(t.Context(),
+				bytes.NewReader(input),
+				&Request{Size: tt.size, Format: FormatJPEG, FitMode: FitCover})
+			if err != nil {
+				t.Fatalf("Process() error = %v", err)
+			}
+
+			defer func() { _ = result.Content.Close() }()
+
+			t.Logf("Process() made %dx%d", result.Width, result.Height)
+
+			if result.Width != tt.size.Width || result.Height != tt.size.Height {
+				t.Errorf("Process() made %dx%d, want %dx%d",
+					result.Width, result.Height, tt.size.Width, tt.size.Height)
+			}
+		})
+	}
+}
+
+// createBandsPNG creates a PNG of three equal bands, red, green and blue,
+// side by side when it is wider than tall and top to bottom otherwise.
+func createBandsPNG(t *testing.T, width, height int) []byte {
+	t.Helper()
+
+	bands := []color.RGBA{
+		{R: 255, A: 255},
+		{G: 255, A: 255},
+		{B: 255, A: 255},
+	}
+
+	img := image.NewRGBA(image.Rect(0, 0, width, height))
+	for y := range height {
+		for x := range width {
+			band := 3 * y / height
+			if width > height {
+				band = 3 * x / width
+			}
+
+			img.Set(x, y, bands[band])
+		}
+	}
+
+	return encodePNG(t, img)
+}
+
+// TestImageProcessor_FitCoverCropsTheCentre verifies that FitCover keeps the
+// centre of the source: a 300x100 source of red, green and blue bands side
+// by side, or a 100x300 one with them top to bottom, comes out at 50x50 as
+// its green band.
+func TestImageProcessor_FitCoverCropsTheCentre(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name                    string
+		inputWidth, inputHeight int
+	}{
+		{"bands side by side", 300, 100},
+		{"bands top to bottom", 100, 300},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			input := createBandsPNG(t, tt.inputWidth, tt.inputHeight)
+
+			result, err := New(Params{}).Process(t.Context(),
+				bytes.NewReader(input),
+				&Request{
+					Size:    Size{Width: 50, Height: 50},
+					Format:  FormatPNG,
+					FitMode: FitCover,
+				})
+			if err != nil {
+				t.Fatalf("Process() error = %v", err)
+			}
+
+			defer func() { _ = result.Content.Close() }()
+
+			output, err := png.Decode(result.Content)
+			if err != nil {
+				t.Fatalf("failed to decode output: %v", err)
+			}
+
+			bounds := output.Bounds()
+			t.Logf("Process() made %dx%d", bounds.Dx(), bounds.Dy())
+
+			if bounds.Dx() != 50 || bounds.Dy() != 50 {
+				t.Fatalf("Process() made %dx%d, want 50x50",
+					bounds.Dx(), bounds.Dy())
+			}
+
+			// Resampling may blend a little of the bands beside the
+			// green one into the output's edges.
+			for y := bounds.Min.Y; y < bounds.Max.Y; y++ {
+				for x := bounds.Min.X; x < bounds.Max.X; x++ {
+					r, g, b, _ := output.At(x, y).RGBA()
+					if r>>8 > 64 || g>>8 < 192 || b>>8 > 64 {
+						t.Fatalf("output pixel %d,%d is %d,%d,%d, want green",
+							x, y, r>>8, g>>8, b>>8)
+					}
+				}
+			}
+		})
+	}
+}
+
+// The greys of createSquarePNG, and the grey halfway between them.
+const (
+	backgroundGrey = 192
+	squareGrey     = 64
+	halfwayGrey    = (backgroundGrey + squareGrey) / 2
+)
+
+// createSquarePNG creates a grey PNG side pixels square with a darker grey
+// square squareSide pixels wide in its centre.
+func createSquarePNG(t *testing.T, side, squareSide int) []byte {
+	t.Helper()
+
+	first := (side - squareSide) / 2
+	last := first + squareSide - 1
+
+	img := image.NewGray(image.Rect(0, 0, side, side))
+	for y := range side {
+		for x := range side {
+			grey := uint8(backgroundGrey)
+			if x >= first && x <= last && y >= first && y <= last {
+				grey = squareGrey
+			}
+
+			img.SetGray(x, y, color.Gray{Y: grey})
+		}
+	}
+
+	return encodePNG(t, img)
+}
+
+// squareLength returns how long the square is along greys, the greys of a
+// row or column of the output that crosses it: the distance between the
+// first and the last point where greys passes halfwayGrey, or 0 when there
+// is no such point. Resampling blurs the square's edges over several
+// pixels, and each point is placed between the two pixels on either side
+// of it in proportion to their greys, so the length is measured to a
+// fraction of a pixel.
+func squareLength(greys []float64) float64 {
+	first, last := -1.0, -1.0
+
+	for i := 1; i < len(greys); i++ {
+		before, after := greys[i-1], greys[i]
+		if (before > halfwayGrey) == (after > halfwayGrey) {
+			continue
+		}
+
+		point := float64(i-1) + (before-halfwayGrey)/(before-after)
+		if first < 0 {
+			first = point
+		}
+
+		last = point
+	}
+
+	return last - first
+}
+
+// TestImageProcessor_FitCoverKeepsProportions verifies that FitCover scales
+// both sides of the source by the same amount: a square in the centre of
+// the source comes out square. Its width and height are measured along the
+// output's middle row and column.
+func TestImageProcessor_FitCoverKeepsProportions(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name             string
+		side, squareSide int
+		size             Size
+	}{
+		{"16x16 at 100x60", 16, 8, Size{Width: 100, Height: 60}},
+		{"16x16 at 60x100", 16, 8, Size{Width: 60, Height: 100}},
+		{"10x10 at 100x35", 10, 2, Size{Width: 100, Height: 35}},
+		{"10x10 at 35x100", 10, 2, Size{Width: 35, Height: 100}},
+		{"64x64 at 300x100", 64, 16, Size{Width: 300, Height: 100}},
+		{"64x64 at 100x300", 64, 16, Size{Width: 100, Height: 300}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			input := createSquarePNG(t, tt.side, tt.squareSide)
+
+			result, err := New(Params{}).Process(t.Context(),
+				bytes.NewReader(input),
+				&Request{Size: tt.size, Format: FormatPNG, FitMode: FitCover})
+			if err != nil {
+				t.Fatalf("Process() error = %v", err)
+			}
+
+			defer func() { _ = result.Content.Close() }()
+
+			output, err := png.Decode(result.Content)
+			if err != nil {
+				t.Fatalf("failed to decode output: %v", err)
+			}
+
+			bounds := output.Bounds()
+			if bounds.Dx() != tt.size.Width || bounds.Dy() != tt.size.Height {
+				t.Fatalf("Process() made %dx%d, want %dx%d",
+					bounds.Dx(), bounds.Dy(), tt.size.Width, tt.size.Height)
+			}
+
+			row := make([]float64, bounds.Dx())
+			for x := range row {
+				r, _, _, _ := output.At(x, bounds.Dy()/2).RGBA()
+				row[x] = float64(r >> 8)
+			}
+
+			column := make([]float64, bounds.Dy())
+			for y := range column {
+				r, _, _, _ := output.At(bounds.Dx()/2, y).RGBA()
+				column[y] = float64(r >> 8)
+			}
+
+			width, height := squareLength(row), squareLength(column)
+			t.Logf("square is %.2f wide and %.2f high", width, height)
+
+			if width == 0 || math.Abs(width-height) > 0.5 {
+				t.Errorf("square is %.2f wide and %.2f high, want it square",
+					width, height)
+			}
+		})
+	}
+}

internal/imageprocessor/fit_outside_internal_test.go

@@ -60,3 +60,45 @@ func TestImageProcessor_RefusesFitOutsideAboveMaximum(t *testing.T) {
 		})
 	}
 }
+
+// TestImageProcessor_FitOutsideSideIsAtLeastOne verifies that FitOutside
+// keeps each side at 1 or more when it works both sides out again from the
+// larger scale. A side of 49 scaled to 1 is 49 times 1/49, which in floating
+// point is just under 1 and so rounded down to 0, a width libvips refuses.
+func TestImageProcessor_FitOutsideSideIsAtLeastOne(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name                    string
+		inputWidth, inputHeight int
+		size                    Size
+		wantWidth, wantHeight   int
+	}{
+		{"49x8192 at 0x100", 49, 8192, Size{Height: 100}, 1, 167},
+		{"49x49 at 1x1", 49, 49, Size{Width: 1, Height: 1}, 1, 1},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			input := createTestJPEG(t, tt.inputWidth, tt.inputHeight)
+
+			result, err := New(Params{}).Process(t.Context(),
+				bytes.NewReader(input),
+				&Request{Size: tt.size, Format: FormatJPEG, FitMode: FitOutside})
+			if err != nil {
+				t.Fatalf("Process() error = %v", err)
+			}
+
+			defer func() { _ = result.Content.Close() }()
+
+			t.Logf("Process() made %dx%d", result.Width, result.Height)
+
+			if result.Width != tt.wantWidth || result.Height != tt.wantHeight {
+				t.Errorf("Process() made %dx%d, want %dx%d",
+					result.Width, result.Height, tt.wantWidth, tt.wantHeight)
+			}
+		})
+	}
+}

internal/imageprocessor/imageprocessor.go

@@ -311,8 +311,10 @@ func (p *ImageProcessor) Process(
 
 // targetDimensions calculates the output dimensions for a requested size,
 // scaling proportionally when only one dimension is given and keeping the
-// original dimensions when both are zero. A side scaled proportionally can
+// original dimensions when both are zero. A side scaled proportionally is
-// come out above MaxOutputDimension, which is ErrOutputTooLarge.
+// rounded down but kept at 1 or more, as for a thin image it would round
+// down to 0, a side of no pixels. It can come out above MaxOutputDimension,
+// which is ErrOutputTooLarge.
 func targetDimensions(size Size, origWidth, origHeight int) (int, int, error) {
 	width, height := size.Width, size.Height
 
@@ -322,10 +324,10 @@ func targetDimensions(size Size, origWidth, origHeight int) (int, int, error) {
 		return origWidth, origHeight, nil
 	case width == 0:
 		// Only height specified: calculate width proportionally
-		width = origWidth * height / origHeight
+		width = max(1, origWidth*height/origHeight)
 	case height == 0:
 		// Only width specified: calculate height proportionally
-		height = origHeight * width / origWidth
+		height = max(1, origHeight*width/origWidth)
 	}
 
 	if width > MaxOutputDimension || height > MaxOutputDimension {
@@ -475,17 +477,20 @@ func (p *ImageProcessor) resize(
 ) error {
 	switch fit {
 	case FitCover, "":
-		// Resize and crop to fill exact dimensions (default)
+		// Crop the centre and resize to fill exact dimensions (default)
-		return img.Thumbnail(width, height, vips.InterestingCentre)
+		return cropCentreAndResize(img, width, height)
 
 	case FitContain:
-		// Resize to fit within dimensions, maintaining aspect ratio
+		// Resize to fit within dimensions, maintaining aspect ratio. Scaled
+		// by the smaller scale, the short side of a thin image can round
+		// down to 0, which libvips refuses as a width, so each side is at
+		// least 1.
 		imgW, imgH := img.Width(), img.Height()
 		scaleW := float64(width) / float64(imgW)
 		scaleH := float64(height) / float64(imgH)
 		scale := min(scaleW, scaleH)
-		newW := int(float64(imgW) * scale)
+		newW := max(1, int(float64(imgW)*scale))
-		newH := int(float64(imgH) * scale)
+		newH := max(1, int(float64(imgH)*scale))
 
 		return img.Thumbnail(newW, newH, vips.InterestingNone)
 
@@ -503,19 +508,22 @@ func (p *ImageProcessor) resize(
 		scaleW := float64(width) / float64(imgW)
 		scaleH := float64(height) / float64(imgH)
 		scale := min(scaleW, scaleH)
-		newW := int(float64(imgW) * scale)
+		newW := max(1, int(float64(imgW)*scale))
-		newH := int(float64(imgH) * scale)
+		newH := max(1, int(float64(imgH)*scale))
 
 		return img.Thumbnail(newW, newH, vips.InterestingNone)
 
 	case FitOutside:
-		// Resize so smallest dimension fits, may exceed target on other dimension
+		// Resize so smallest dimension fits, may exceed target on other
+		// dimension. A side scaled to 1 can come out just under 1 in
+		// floating point, as 49 times 1/49 does, and round down to 0, which
+		// libvips refuses as a width, so each side is at least 1.
 		imgW, imgH := img.Width(), img.Height()
 		scaleW := float64(width) / float64(imgW)
 		scaleH := float64(height) / float64(imgH)
 		scale := max(scaleW, scaleH)
-		newW := int(float64(imgW) * scale)
+		newW := max(1, int(float64(imgW)*scale))
-		newH := int(float64(imgH) * scale)
+		newH := max(1, int(float64(imgH)*scale))
 
 		// The dimension that exceeds the target is held to the maximum too
 		if newW > MaxOutputDimension || newH > MaxOutputDimension {
@@ -530,6 +538,38 @@ func (p *ImageProcessor) resize(
 	}
 }
 
+// cropCentreAndResize scales img by the same amount on both sides, the
+// least that makes it cover width x height, and cuts the centre width x
+// height out of it, as libvips' thumbnail does with InterestingCentre.
+// thumbnail alone scales the whole image and copies the result into memory
+// before it cuts, and for a thin image that is far larger than the image or
+// the output: 8192x64 at 8192x8192 is 1048576x8192, about 24 GiB. So the
+// centre is cut from img first, widened to whole pixels of img, which adds
+// less than one pixel of img on each side, and thumbnail scales and cuts
+// only that.
+func cropCentreAndResize(img *vips.ImageRef, width, height int) error {
+	imgW, imgH := img.Width(), img.Height()
+
+	// An image wider than the requested proportions keeps its height and
+	// loses width down to imgH*width/height, any other keeps its width and
+	// loses height down to imgW*height/width. Half of what is lost comes
+	// off each side, rounded down to whole pixels, so the part kept holds
+	// all of the centre.
+	left, top := 0, 0
+	if imgW*height > imgH*width {
+		left = (imgW*height - imgH*width) / (2 * height) //nolint:mnd // each side
+	} else {
+		top = (imgH*width - imgW*height) / (2 * width) //nolint:mnd // each side
+	}
+
+	err := img.ExtractArea(left, top, imgW-2*left, imgH-2*top)
+	if err != nil {
+		return err
+	}
+
+	return img.Thumbnail(width, height, vips.InterestingCentre)
+}
+
 const defaultQuality = 85
 
 // encode encodes an image to the specified format. govips gives pixa no way

internal/imgcache/eviction.go

@@ -18,6 +18,14 @@ import (
 // write-pressure wakeups triggered by stores.
 const DefaultEvictionInterval = 5 * time.Minute
 
+// reconciliationWaitFactor is how many times as long as a
+// reconciliation pass took the eviction loop waits at least, after that
+// pass ends, before it starts the next one. A pass reads a row and stats a
+// file for every cached file, so its time grows with the cache; waiting
+// ten times as long keeps reconciliation to at most a tenth of the time,
+// however large the cache is.
+const reconciliationWaitFactor = 10
+
 // evictionBatchSize is how many LRU candidates of each class (variants
 // and source blobs) one eviction pass fetches from the database.
 const evictionBatchSize = 100
@@ -438,10 +446,12 @@ func (c *Cache) notifyWritePressure() {
 }
 
 // StartEviction launches the background eviction goroutine, which
-// reconciles the database accounting with the cache directories and
 // removes the expired negative cache rows at startup and again on every
-// periodic tick thereafter, and evicts to the configured limit on the
+// periodic tick thereafter, reconciles the database accounting with the
-// given periodic interval and on write-pressure notifications. On a
+// cache directories at startup and again on the first periodic tick
+// that comes at least ten times as long after a reconciliation pass
+// ended as that pass took, and evicts to the configured limit on the given
+// periodic interval and on write-pressure notifications. On a
 // disabled cache it only removes the expired negative cache rows, as
 // the negative cache is used either way. It is a no-op when already
 // started. The goroutine outlives the caller, so it runs with its own
@@ -482,7 +492,7 @@ func (c *Cache) StopEviction(ctx context.Context) error {
 func (c *Cache) evictionLoop(ctx context.Context, interval time.Duration) {
 	defer close(c.evictionDone)
 
-	c.runReconciliationPass(ctx)
+	nextReconciliation := c.runReconciliationPass(ctx)
 	c.runCleanExpired(ctx)
 	c.runEvictionPass(ctx)
 
@@ -494,16 +504,20 @@ func (c *Cache) evictionLoop(ctx context.Context, interval time.Duration) {
 		case <-ctx.Done():
 			return
 		case <-ticker.C:
-			// Reconciliation walks the cache directories, so it only
+			// Reconciliation walks the cache directories and reads a
-			// runs on the periodic ticker rather than on every
+			// row for every file, so it runs only on the periodic
-			// write-pressure wakeup, keeping it off the per-store hot
+			// ticker rather than on every write-pressure wakeup, and
-			// path. Reusing the eviction interval itself (rather than a
+			// only on the first tick at or after nextReconciliation,
-			// separate, longer one) is a deliberate choice: it is the
+			// which is ten times as long after the last pass ended as
-			// simplest option that still bounds how long a store's
+			// that pass took. Passes then never run back to back,
-			// best-effort accounting insert can stay silently
+			// however long one takes. A file whose best-effort
-			// unaccounted for to one interval, on a process that is
+			// accounting insert failed in a store stays unaccounted for
-			// already running this loop regardless.
+			// until the next pass, up to about one eviction interval
-			c.runReconciliationPass(ctx)
+			// plus ten times the last pass's time.
+			if !time.Now().Before(nextReconciliation) {
+				nextReconciliation = c.runReconciliationPass(ctx)
+			}
+
 			c.runCleanExpired(ctx)
 		case <-c.evictionPressure:
 		}
@@ -527,17 +541,25 @@ func (c *Cache) runEvictionPass(ctx context.Context) {
 }
 
 // runReconciliationPass runs one reconciliation pass, logging failures
-// instead of propagating them (the loop must keep running). It does
+// instead of propagating them (the loop must keep running), and returns
-// nothing once ctx is cancelled.
+// the earliest time the next pass may start: reconciliationWaitFactor
-func (c *Cache) runReconciliationPass(ctx context.Context) {
+// times as long after this pass ended as it took. It does nothing once
+// ctx is cancelled.
+func (c *Cache) runReconciliationPass(ctx context.Context) time.Time {
 	if ctx.Err() != nil {
-		return
+		return time.Now()
 	}
 
+	started := time.Now()
+
 	err := c.reconcileAccounting(ctx)
 	if err != nil {
 		c.log.Warn("cache accounting reconciliation failed", "error", err)
 	}
+
+	ended := time.Now()
+
+	return ended.Add(reconciliationWaitFactor * ended.Sub(started))
 }
 
 // runCleanExpired runs CleanExpired, logging a failure instead of
@@ -556,7 +578,8 @@ func (c *Cache) runCleanExpired(ctx context.Context) {
 
 // reconcileAccounting synchronizes the database size accounting with
 // the actual contents of the cache directories. It runs at startup and
-// again on every periodic eviction tick thereafter, off the request
+// then on the first periodic eviction tick that comes at least ten times
+// as long after the last pass ended as that pass took, off the request
 // hot path: it adopts variant files that predate the accounting table
 // (or whose accounting insert failed, e.g. StoreVariant's best-effort
 // insert under transient DB contention), drops accounting rows whose
@@ -566,9 +589,10 @@ func (c *Cache) runCleanExpired(ctx context.Context) {
 // files left behind by crashed writes, and last checks the total cache
 // usage against the tables. It reads the tables a page at a time.
 // Running it periodically, not just once, bounds how long such drift can
-// accumulate unaccounted for on a long-running process to one eviction
+// accumulate unaccounted for on a long-running process to about one
-// interval. Once ctx is cancelled, it stops at the next file or row and
+// eviction interval plus ten times the last pass's time.
-// returns ctx's error.
+// Once ctx is cancelled, it stops at the next file or row and returns
+// ctx's error.
 func (c *Cache) reconcileAccounting(ctx context.Context) error {
 	if c.disabled {
 		return nil

internal/imgcache/eviction_internal_test.go

@@ -918,6 +918,89 @@ func TestPeriodicReconciliationAdoptsFileThatAppearsAfterStartup(t *testing.T) {
 	}
 }
 
+// TestReconciliationPassesAreSpacedByTenTimesTheirTime makes the hook
+// called after every read of the reconciliation pass sleep 10 ms, runs
+// the eviction loop on a 10 ms interval, and records when each of those
+// calls starts and ends. A pass starts before its first call and ends
+// after its last, so the time from the first call's start to the last
+// call's end is at most the pass's time. The next pass's first call must
+// start at least ten times that long after the last call ended.
+func TestReconciliationPassesAreSpacedByTenTimesTheirTime(t *testing.T) {
+	t.Parallel()
+
+	cache, _ := newEvictionTestCache(t, 1<<30)
+
+	// A pass made here, before the loop starts, counts the reads one
+	// pass makes on this empty cache, so the reads recorded below can be
+	// split into passes.
+	readsPerPass := 0
+	cache.reconciliationReadTestHook = func(int) { readsPerPass++ }
+
+	err := cache.reconcileAccounting(t.Context())
+	if err != nil {
+		t.Fatalf("reconcileAccounting failed: %v", err)
+	}
+
+	t.Logf("one reconciliation pass makes %d reads", readsPerPass)
+
+	if readsPerPass == 0 {
+		t.Fatal("a reconciliation pass made no read")
+	}
+
+	// The pass at startup and two after it.
+	const passes = 3
+
+	type read struct{ start, end time.Time }
+
+	// The channel holds every read the test looks at, so none of those
+	// is dropped; reads after them are, and never block the loop.
+	reads := make(chan read, passes*readsPerPass)
+
+	cache.reconciliationReadTestHook = func(int) {
+		start := time.Now()
+
+		time.Sleep(10 * time.Millisecond)
+
+		select {
+		case reads <- read{start: start, end: time.Now()}:
+		default:
+		}
+	}
+
+	cache.StartEviction(10 * time.Millisecond)
+	defer func() { _ = cache.StopEviction(t.Context()) }()
+
+	recorded := make([]read, 0, passes*readsPerPass)
+	timeout := time.After(15 * time.Second)
+
+	for len(recorded) < passes*readsPerPass {
+		select {
+		case r := <-reads:
+			recorded = append(recorded, r)
+		case <-timeout:
+			t.Fatalf("saw %d reconciliation reads in 15 s, want %d (%d passes)",
+				len(recorded), passes*readsPerPass, passes)
+		}
+	}
+
+	for pass := range passes - 1 {
+		first := recorded[pass*readsPerPass]
+		last := recorded[(pass+1)*readsPerPass-1]
+		next := recorded[(pass+1)*readsPerPass]
+
+		passTime := last.end.Sub(first.start)
+		wait := next.start.Sub(last.end)
+
+		t.Logf("pass %d took at least %v, and the next pass started %v after it",
+			pass+1, passTime, wait)
+
+		if wait < 10*passTime {
+			t.Errorf("pass %d took at least %v, and the next pass started %v "+
+				"after it, want at least %v", pass+1, passTime, wait, 10*passTime)
+		}
+	}
+}
+
 // TestStopEvictionInterruptsPassInProgress holds the test database's
 // only connection, so the startup reconciliation pass waits for it, and
 // checks that StopEviction stops that pass instead of waiting for the

internal/imgcache/imgcache.go

@@ -5,6 +5,7 @@ import (
 	"context"
 	"errors"
 	"io"
+	"net/url"
 	"time"
 )
 
@@ -86,9 +87,11 @@ func ValidateImageRequest(req *ImageRequest) error {
 type ImageRequest struct {
 	// SourceHost is the origin host (e.g., "cdn.example.com")
 	SourceHost string
-	// SourcePath is the path on the origin (e.g., "/photos/cat.jpg")
+	// SourcePath is the path on the origin with its percent-escapes decoded,
+	// as url.URL's Path holds it (e.g., "/photos/my cat.jpg")
 	SourcePath string
-	// SourceQuery is the optional query string for the origin URL
+	// SourceQuery is the optional query string for the origin URL as the URL
+	// writes it, percent-escapes included, as url.URL's RawQuery holds it
 	SourceQuery string
 	// Size is the requested output dimensions
 	Size Size
@@ -107,20 +110,25 @@ type ImageRequest struct {
 	AllowHTTP bool
 }
 
-// SourceURL returns the full upstream URL to fetch.
+// SourceURL returns the full upstream URL to fetch: the source path
-// Uses http:// scheme when AllowHTTP is true, otherwise https://.
+// percent-encoded again, as net/url writes a path, so each source path is one
+// upstream path, and the source query as it is, with the characters
+// escapeQuery encodes percent-encoded. Uses http:// scheme when AllowHTTP is
+// true, otherwise https://.
 func (r *ImageRequest) SourceURL() string {
 	scheme := "https"
 	if r.AllowHTTP {
 		scheme = "http"
 	}
 
-	url := scheme + "://" + r.SourceHost + r.SourcePath
+	sourceURL := url.URL{
-	if r.SourceQuery != "" {
+		Scheme:   scheme,
-		url += "?" + r.SourceQuery
+		Host:     r.SourceHost,
+		Path:     r.SourcePath,
+		RawQuery: escapeQuery(r.SourceQuery),
 	}
 
-	return url
+	return sourceURL.String()
 }
 
 // ImageResponse represents a processed image ready to serve

internal/imgcache/urlparser.go

@@ -3,6 +3,7 @@ package imgcache
 import (
 	"errors"
 	"fmt"
+	"net/netip"
 	"net/url"
 	"regexp"
 	"strconv"
@@ -28,7 +29,9 @@ var (
 const MaxDimension = imageprocessor.MaxOutputDimension
 
 // ValidateDimension checks one requested width or height; name ("width" or
-// "height") appears in the error. 0 means "original size" and is valid.
+// "height") appears in the error. 0 is valid: that side is then worked out from
+// the source's proportions and the other requested side, rounded down but at
+// least 1, or is the source's own side when the other side is 0 too.
 func ValidateDimension(name string, value int) error {
 	if value < 0 {
 		return fmt.Errorf("%w: %s is negative", ErrInvalidSize, name)
@@ -47,6 +50,24 @@ func ValidateDimension(name string, value int) error {
 var sizeFormatRegex = regexp.MustCompile(
 	`^(\d+)x(\d+)(?:\.(\w+))?$|^(orig)(?:\.(\w+))?$`)
 
+// percentEscapeRegex matches one percent-escape, such as "%2e"
+var percentEscapeRegex = regexp.MustCompile(`%[0-9A-Fa-f]{2}`)
+
+// queryEscapeRegex matches a percent-escape, or one character that a browser
+// percent-encodes in the query of an http or https URL: one outside the
+// printable ASCII characters "!" to "~", such as a space, a control character
+// or a character outside ASCII, or one of `"#'<>`. A "%" that does not begin a
+// percent-escape matches too.
+var queryEscapeRegex = regexp.MustCompile(`%[0-9A-Fa-f]{2}|[^!-~]|["#'<>%]`)
+
+// hostRegex matches the forms the host of an image URL may take: a host name
+// or an IPv4 address (letters, digits, "-", "_" and "."), or an IPv6 address
+// in brackets (hex digits, ":" and "."), either one optionally followed by ":"
+// and a port, which may be empty. Its one group holds the text in brackets,
+// which parseImageComponents also checks is an IPv6 address.
+var hostRegex = regexp.MustCompile(
+	`^(?:[A-Za-z0-9_.-]+|\[([0-9A-Fa-f:.]+)\])(?::[0-9]*)?$`)
+
 // ParsedURL contains the parsed components of an image proxy URL.
 type ParsedURL struct {
 	// Host is the source origin host (e.g., "cdn.example.com")
@@ -61,9 +82,8 @@ type ParsedURL struct {
 	Format ImageFormat
 }
 
-// ParseImagePath parses the path captured by chi's wildcard:
+// ParseImagePath parses an image URL's path after /v1/image/, decoded once:
-// <host>/<path>/<size>.<format>, or <host>/<path>/<size> for JPEG XL
+// <host>/<path>/<size>.<format>, or <host>/<path>/<size> for JPEG XL.
-// This is the primary entry point when using chi routing.
 // Examples:
 //   - cdn.example.com/photos/cat.jpg/800x600.webp
 //   - cdn.example.com/photos/cat.jpg/0x0.jpeg
@@ -79,9 +99,10 @@ func ParseImagePath(path string) (*ParsedURL, error) {
 	return parseImageComponents(path)
 }
 
-// ParseImageURL parses a full URL path like /v1/image/<host>/<path>/<size>.<format>,
+// ParseImageURL parses an image URL's path, decoded once as r.URL.Path holds
-// or /v1/image/<host>/<path>/<size> for JPEG XL
+// it: /v1/image/<host>/<path>/<size>.<format>, or /v1/image/<host>/<path>/<size>
-// Use ParseImagePath instead when working with chi's wildcard capture.
+// for JPEG XL. A source query follows the source path after a "?", which the
+// URL writes as %3F.
 func ParseImageURL(urlPath string) (*ParsedURL, error) {
 	// Remove the /v1/image/ prefix
 	const prefix = "/v1/image/"
@@ -100,12 +121,6 @@ func ParseImageURL(urlPath string) (*ParsedURL, error) {
 // parseImageComponents parses <host>/<path>/<size>.<format>, or
 // <host>/<path>/<size> for JPEG XL.
 func parseImageComponents(remainder string) (*ParsedURL, error) {
-	// Check for path traversal before any other processing
-	err := checkPathTraversal(remainder)
-	if err != nil {
-		return nil, err
-	}
-
 	// Find the last path segment, which holds "size" or "size.format"
 	lastSlash := strings.LastIndex(remainder, "/")
 	if lastSlash == -1 {
@@ -144,10 +159,20 @@ func parseImageComponents(remainder string) (*ParsedURL, error) {
 		return nil, ErrMissingHost
 	}
 
-	// A URL reads what comes before "@" as a user name and password, and the
+	// Only the forms hostRegex describes are accepted. A "?", "#" or "@" in
-	// rest as the host
+	// the host would make Go's URL parser, which the allowlist check and the
-	if strings.Contains(host, "@") {
+	// fetcher use, read a different host from the one allowlisted and cached
-		return nil, fmt.Errorf("%w: %q holds @", ErrInvalidHost, host)
+	hostMatch := hostRegex.FindStringSubmatch(host)
+	if hostMatch == nil {
+		return nil, fmt.Errorf("%w: %q", ErrInvalidHost, host)
+	}
+
+	// Go's URL parser refuses brackets around anything but an IPv6 address
+	if inBrackets := hostMatch[1]; inBrackets != "" {
+		addr, err := netip.ParseAddr(inBrackets)
+		if err != nil || !addr.Is6() {
+			return nil, fmt.Errorf("%w: %q", ErrInvalidHost, host)
+		}
 	}
 
 	// Host names do not depend on letter case, so the allowlist, the upstream
@@ -155,12 +180,30 @@ func parseImageComponents(remainder string) (*ParsedURL, error) {
 	// host
 	host = strings.ToLower(host)
 
-	// Extract query string if present in path
+	// Go's URL parser, which the fetcher uses, reads a "#" as the start of a
+	// fragment and drops the rest, so each text after it would be another
+	// spelling, cached under its own keys, of one upstream request
+	if strings.Contains(path, "#") {
+		return nil, fmt.Errorf(
+			`%w: the source path or query holds a "#"`, ErrInvalidPath)
+	}
+
+	// Extract query string if present in path. It is the source URL's query
+	// as that URL writes it. A character a URL's query cannot hold as it is,
+	// such as a space written %20 in the image URL, is percent-encoded here,
+	// so the query is signed and cached as it is sent upstream
 	if qIndex := strings.Index(path, "?"); qIndex != -1 {
-		query = path[qIndex+1:]
+		query = escapeQuery(path[qIndex+1:])
 		path = path[:qIndex]
 	}
 
+	// The source query names no file on the upstream, so only the host and
+	// the source path are checked for path traversal
+	err = checkPathTraversal(host + path)
+	if err != nil {
+		return nil, err
+	}
+
 	// Ensure path starts with /
 	if !strings.HasPrefix(path, "/") {
 		path = "/" + path
@@ -186,12 +229,7 @@ func checkPathTraversal(path string) error {
 	// Decode multiple times to catch double-encoding
 	decoded := path
 	for range 3 {
-		newDecoded, err := url.PathUnescape(decoded)
+		newDecoded := decodePercentEscapes(decoded)
-		if err != nil {
-			// Malformed encoding is suspicious
-			return ErrPathTraversal
-		}
-
 		if newDecoded == decoded {
 			break
 		}
@@ -232,6 +270,32 @@ func checkPathTraversal(path string) error {
 	return nil
 }
 
+// decodePercentEscapes decodes every percent-escape in s, such as %2e, and
+// keeps a "%" that does not start one as it is, where url.PathUnescape would
+// refuse the whole string. A source path holding %25, a literal "%", has such
+// a "%" after one more decoding.
+func decodePercentEscapes(s string) string {
+	return percentEscapeRegex.ReplaceAllStringFunc(s, func(escape string) string {
+		decoded, _ := url.PathUnescape(escape)
+
+		return decoded
+	})
+}
+
+// escapeQuery returns a source query as a URL's query holds it: each
+// character that queryEscapeRegex names, such as a space, percent-encoded,
+// and the rest, the query's percent-escapes included, kept as they are. The
+// result is unchanged by a second escapeQuery.
+func escapeQuery(query string) string {
+	return queryEscapeRegex.ReplaceAllStringFunc(query, func(match string) string {
+		if percentEscapeRegex.MatchString(match) {
+			return match
+		}
+
+		return url.PathEscape(match)
+	})
+}
+
 // parseSizeFormat parses strings like "800x600.webp", "orig.png" or "800x600"
 func parseSizeFormat(s string) (Size, ImageFormat, error) {
 	matches := sizeFormatRegex.FindStringSubmatch(s)

internal/imgcache/urlparser_internal_test.go

@@ -407,6 +407,71 @@ func TestParseImagePath_PathTraversal(t *testing.T) {
 	}
 }
 
+// TestParseImageURL_SourceURLHoldingPercent25 verifies that a source path and
+// a source query holding %25, the percent-escape of a literal "%", are
+// accepted as they are, although decoding them once more leaves a "%" that
+// does not start a percent-escape.
+func TestParseImageURL_SourceURLHoldingPercent25(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name  string
+		input string
+		want  *ParsedURL
+	}{
+		{
+			name:  "path",
+			input: "/v1/image/cdn.example.com/sale/50%25off.jpg/800x600.webp",
+			want: &ParsedURL{
+				Host: testHostCDN, Path: "/sale/50%25off.jpg",
+				Size: Size{Width: 800, Height: 600}, Format: FormatWebP,
+			},
+		},
+		{
+			name:  "query",
+			input: "/v1/image/cdn.example.com/photos/cat.jpg?discount=50%25/800x600.webp",
+			want: &ParsedURL{
+				Host: testHostCDN, Path: testPathCat, Query: "discount=50%25",
+				Size: Size{Width: 800, Height: 600}, Format: FormatWebP,
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			got, err := ParseImageURL(tt.input)
+			if err != nil {
+				t.Fatalf("ParseImageURL() unexpected error = %v", err)
+			}
+
+			assertParsedURL(t, got, tt.want)
+		})
+	}
+}
+
+// TestParseImageURL_DoubleEncodedTraversalNextToPercent verifies that a
+// double-encoded ".." is still refused when the path also holds a "%" that
+// does not start a percent-escape, either as written or after one more
+// decoding.
+func TestParseImageURL_DoubleEncodedTraversalNextToPercent(t *testing.T) {
+	t.Parallel()
+
+	for _, input := range []string{
+		"/v1/image/cdn.example.com/photos/%252e%252e/secret/50%25off.jpg/800x600.jpeg",
+		"/v1/image/cdn.example.com/photos/%252e%252e/secret/100%/800x600.jpeg",
+	} {
+		_, err := ParseImageURL(input)
+		t.Logf("ParseImageURL(%q) error = %v", input, err)
+
+		if !errors.Is(err, ErrPathTraversal) {
+			t.Errorf("ParseImageURL(%q) error = %v, want ErrPathTraversal",
+				input, err)
+		}
+	}
+}
+
 // errorIs checks if err matches target (handles wrapped errors).
 func errorIs(err, target error) bool {
 	if errors.Is(err, target) {

internal/server/image_proxy_internal_test.go

@@ -14,10 +14,12 @@ import (
 	"net"
 	"net/http"
 	"net/http/httptest"
+	"net/netip"
 	"os"
 	"path/filepath"
 	"sync/atomic"
 	"testing"
+	"time"
 
 	"go.uber.org/fx"
 	"go.uber.org/fx/fxtest"
@@ -183,6 +185,57 @@ func startImageProxy(
 	return s, db.DB(), stateDir
 }
 
+// startImageProxyWithBlockedNetworks starts the components pixad's fx app
+// builds, from a config with a fresh state directory and blockedNetworks as
+// blocked_networks. Unlike startImageProxy it provides no upstream fetcher, so
+// the handlers build the one pixad uses, which checks every address it
+// connects to against the built-in blocked ranges and blocked_networks. It
+// returns the server with its routes.
+func startImageProxyWithBlockedNetworks(
+	t *testing.T, blockedNetworks []netip.Prefix,
+) *Server {
+	t.Helper()
+
+	stateDir := t.TempDir()
+	cfg := &config.Config{
+		SigningKey:      testSigningKey,
+		StateDir:        stateDir,
+		DBURL:           "file:" + filepath.Join(stateDir, "state.sqlite3"),
+		BlockedNetworks: blockedNetworks,
+		// A fetch that gets past the checks and finds nothing at the address
+		// fails within this time at the latest.
+		UpstreamFetchTimeout: 2 * time.Second,
+		// With no connection slots, the fetch would fail before connecting.
+		UpstreamConnections: config.DefaultUpstreamConnections,
+		DownstreamTimeout:   config.DefaultDownstreamTimeout,
+	}
+
+	var (
+		h  *handlers.Handlers
+		mw *middleware.Middleware
+	)
+
+	app := fxtest.New(t,
+		fx.Supply(cfg),
+		fx.Provide(
+			globals.New,
+			logger.New,
+			database.New,
+			healthcheck.New,
+			handlers.New,
+			middleware.New,
+		),
+		fx.Populate(&h, &mw),
+	)
+	app.RequireStart()
+	t.Cleanup(app.RequireStop)
+
+	s := &Server{config: cfg, mw: mw, h: h}
+	s.SetupRoutes()
+
+	return s
+}
+
 // getImage sends a GET for target to s and fails unless it answers 200.
 func getImage(t *testing.T, s *Server, target string) *httptest.ResponseRecorder {
 	t.Helper()

internal/server/source_url_internal_test.go

@@ -0,0 +1,665 @@
+package server
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"net/netip"
+	"net/url"
+	"regexp"
+	"slices"
+	"strconv"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"sneak.berlin/go/pixa/internal/encurl"
+	"sneak.berlin/go/pixa/internal/imgcache"
+	"sneak.berlin/go/pixa/internal/session"
+	"sneak.berlin/go/pixa/internal/signature"
+)
+
+// percentSignPath is a source path holding a "%", which a URL writes %25.
+const percentSignPath = "/100%25.png"
+
+// generateFormPattern extracts the CSRF token rendered into the generator
+// page's form.
+var generateFormPattern = regexp.MustCompile(
+	`(?s)action="/generate".*?name="gorilla\.csrf\.Token" value="([^"]+)"`)
+
+// generatedURLPattern extracts the path of the encrypted URL the generator
+// page shows.
+var generatedURLPattern = regexp.MustCompile(`value="[^"]*(/v1/e/[^"]+)"`)
+
+// recordingUpstream is a test upstream that answers every request with a PNG
+// and records the request target, the path and query as sent, of each request
+// it receives.
+type recordingUpstream struct {
+	*httptest.Server
+
+	mu       sync.Mutex
+	received []string
+}
+
+// startRecordingUpstream starts a recordingUpstream, which is closed when the
+// test ends.
+func startRecordingUpstream(t *testing.T) *recordingUpstream {
+	t.Helper()
+
+	source := encodeTestPNG(t, 64, 48)
+	upstream := &recordingUpstream{}
+
+	upstream.Server = httptest.NewServer(http.HandlerFunc(
+		func(w http.ResponseWriter, r *http.Request) {
+			upstream.mu.Lock()
+			upstream.received = append(upstream.received, r.RequestURI)
+			upstream.mu.Unlock()
+
+			w.Header().Set("Content-Type", "image/png")
+			_, _ = w.Write(source)
+		}))
+	t.Cleanup(upstream.Close)
+
+	return upstream
+}
+
+// requestTargets returns the request targets the upstream has received, in
+// the order it received them.
+func (u *recordingUpstream) requestTargets() []string {
+	u.mu.Lock()
+	defer u.mu.Unlock()
+
+	return slices.Clone(u.received)
+}
+
+// pageOrigin is the origin of the login and generator pages below, opened
+// over HTTPS; httptest gives every request the host example.com.
+const pageOrigin = "https://example.com"
+
+// openRoot sends GET / to s over HTTPS, with cookie when it is not nil, and
+// returns the page: the login page, or the generator page for a login session.
+func openRoot(
+	t *testing.T, s *Server, cookie *http.Cookie,
+) *httptest.ResponseRecorder {
+	t.Helper()
+
+	req := httptest.NewRequestWithContext(t.Context(), http.MethodGet,
+		pageOrigin+"/", nil)
+	if cookie != nil {
+		req.AddCookie(cookie)
+	}
+
+	page := httptest.NewRecorder()
+	s.ServeHTTP(page, req)
+
+	return page
+}
+
+// submitForm posts form to path on s as a browser submits it from page over
+// HTTPS: with the CSRF token tokenPattern finds in page, the cookies page set,
+// cookie when it is not nil, and a Referer naming the page, which pixa needs
+// while debug is off, as it is in startImageProxy's config.
+func submitForm(
+	t *testing.T, s *Server, page *httptest.ResponseRecorder,
+	tokenPattern *regexp.Regexp, path string, form url.Values,
+	cookie *http.Cookie,
+) *httptest.ResponseRecorder {
+	t.Helper()
+
+	match := tokenPattern.FindStringSubmatch(page.Body.String())
+	if match == nil {
+		t.Fatalf("the page has no form for %s with a CSRF token", path)
+	}
+
+	form.Set("gorilla.csrf.Token", match[1])
+
+	req := httptest.NewRequestWithContext(t.Context(), http.MethodPost,
+		pageOrigin+path, strings.NewReader(form.Encode()))
+	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+	req.Header.Set("Referer", pageOrigin+"/")
+
+	for _, c := range page.Result().Cookies() {
+		req.AddCookie(c)
+	}
+
+	if cookie != nil {
+		req.AddCookie(cookie)
+	}
+
+	rec := httptest.NewRecorder()
+	s.ServeHTTP(rec, req)
+
+	return rec
+}
+
+// postGenerate logs in to s with the signing key, submits the generator page's
+// form with sourceURL, and returns the response.
+func postGenerate(
+	t *testing.T, s *Server, sourceURL string,
+) *httptest.ResponseRecorder {
+	t.Helper()
+
+	login := submitForm(t, s, openRoot(t, s, nil), csrfFieldPattern, "/",
+		url.Values{"key": {testSigningKey}}, nil)
+
+	var sessionCookie *http.Cookie
+
+	for _, c := range login.Result().Cookies() {
+		if c.Name == session.CookieName {
+			sessionCookie = c
+		}
+	}
+
+	if sessionCookie == nil {
+		t.Fatalf("logging in answered %d and set no session cookie", login.Code)
+	}
+
+	rec := submitForm(t, s, openRoot(t, s, sessionCookie), generateFormPattern,
+		"/generate", url.Values{"url": {sourceURL}}, sessionCookie)
+	t.Logf("POST /generate with url %s: %d", sourceURL, rec.Code)
+
+	return rec
+}
+
+// TestImageURLFetchesItsSourcePathAndQuery requests image URLs whose source
+// query holds a space, written percent-encoded, or a "'", and whose source
+// path holds a "%", written percent-encoded, and checks the one request the
+// upstream receives: the source path percent-encoded again, and the source
+// query with the space written %20 and the "'" written %27, as a browser
+// writes them in a query.
+func TestImageURLFetchesItsSourcePathAndQuery(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name string
+		// pathAndQuery is the source path and query as the image URL writes
+		// them
+		pathAndQuery string
+		// wantRequest is the path and query the upstream must receive
+		wantRequest string
+	}{
+		{"space in the query", "/photo.png%3Fname=my%20cat",
+			"/photo.png?name=my%20cat"},
+		{"apostrophe in the query", "/photo.png%3Fname=it's",
+			"/photo.png?name=it%27s"},
+		{"percent sign in the path", percentSignPath, percentSignPath},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			upstream := startRecordingUpstream(t)
+			s, _, _ := startImageProxy(t, upstream.Server)
+
+			getImage(t, s, "/v1/image/"+upstreamHost+tt.pathAndQuery+"/orig.png")
+
+			got := upstream.requestTargets()
+			t.Logf("upstream received %q", got)
+
+			if !slices.Equal(got, []string{tt.wantRequest}) {
+				t.Errorf("upstream received %q, want [%q]", got, tt.wantRequest)
+			}
+		})
+	}
+}
+
+// TestSourcePathsHoldingSpaceAndPercent20AreFetchedApart requests twice each
+// of the source paths /my cat.png and /my%20cat.png, which the image URL
+// writes my%20cat.png and my%2520cat.png. Each is fetched once, as a request
+// of its own, and its second request is served from the cache.
+func TestSourcePathsHoldingSpaceAndPercent20AreFetchedApart(t *testing.T) {
+	t.Parallel()
+
+	upstream := startRecordingUpstream(t)
+	s, _, _ := startImageProxy(t, upstream.Server)
+
+	imageURLs := []string{
+		"/v1/image/" + upstreamHost + "/my%20cat.png/orig.png",
+		"/v1/image/" + upstreamHost + "/my%2520cat.png/orig.png",
+	}
+
+	for _, imageURL := range imageURLs {
+		getImage(t, s, imageURL)
+	}
+
+	for _, imageURL := range imageURLs {
+		cacheStatus := getImage(t, s, imageURL).Header().Get("X-Pixa-Cache")
+		if cacheStatus != string(imgcache.CacheHit) {
+			t.Errorf("second GET %s: X-Pixa-Cache = %q, want HIT",
+				imageURL, cacheStatus)
+		}
+	}
+
+	got := upstream.requestTargets()
+	t.Logf("upstream received %q", got)
+
+	want := []string{"/my%20cat.png", "/my%2520cat.png"}
+	if !slices.Equal(got, want) {
+		t.Errorf("upstream received %q, want %q", got, want)
+	}
+}
+
+// TestSourceQueryWrittenTwoWaysIsOneUpstreamRequest requests the source query
+// name=my%20cat written as the README says, name=my%2520cat, and with the
+// space written %20, name=my%20cat. Both are one source query: the upstream
+// receives one request, and the second image URL is served from the cache.
+func TestSourceQueryWrittenTwoWaysIsOneUpstreamRequest(t *testing.T) {
+	t.Parallel()
+
+	upstream := startRecordingUpstream(t)
+	s, _, _ := startImageProxy(t, upstream.Server)
+
+	getImage(t, s, "/v1/image/"+upstreamHost+"/photo.png%3Fname=my%2520cat/orig.png")
+
+	second := getImage(t, s,
+		"/v1/image/"+upstreamHost+"/photo.png%3Fname=my%20cat/orig.png")
+	if got := second.Header().Get("X-Pixa-Cache"); got != string(imgcache.CacheHit) {
+		t.Errorf("second X-Pixa-Cache = %q, want HIT", got)
+	}
+
+	got := upstream.requestTargets()
+	t.Logf("upstream received %q", got)
+
+	want := []string{"/photo.png?name=my%20cat"}
+	if !slices.Equal(got, want) {
+		t.Errorf("upstream received %q, want %q", got, want)
+	}
+}
+
+// TestGeneratedURLFetchesTheSourceURLAsEntered makes an encrypted URL on the
+// generator page for source URLs holding %20, %25 and %23 in the path or the
+// query, requests it, and checks that the upstream receives the path and
+// query exactly as entered.
+func TestGeneratedURLFetchesTheSourceURLAsEntered(t *testing.T) {
+	t.Parallel()
+
+	// The names hold no "%": each test's state directory is named after it
+	// and goes into the database URL, which decodes percent-escapes
+	tests := []struct {
+		name         string
+		pathAndQuery string
+	}{
+		{"space in the path", "/my%20cat.png"},
+		{"percent sign in the path", percentSignPath},
+		{"number sign in the path", "/a%23b.png"},
+		{"all three in the query", "/photo.png?name=my%20cat%25%23"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			upstream := startRecordingUpstream(t)
+			s, _, _ := startImageProxy(t, upstream.Server)
+
+			rec := postGenerate(t, s, "https://"+upstreamHost+tt.pathAndQuery)
+
+			match := generatedURLPattern.FindStringSubmatch(rec.Body.String())
+			if rec.Code != http.StatusOK || match == nil {
+				t.Fatalf("POST /generate: %d with no encrypted URL; body %s",
+					rec.Code, rec.Body.String())
+			}
+
+			getImage(t, s, match[1])
+
+			got := upstream.requestTargets()
+			t.Logf("upstream received %q", got)
+
+			if !slices.Equal(got, []string{tt.pathAndQuery}) {
+				t.Errorf("upstream received %q, want [%q]", got, tt.pathAndQuery)
+			}
+		})
+	}
+}
+
+// TestGeneratorMakesOneImageURLForTwoSpellingsOfASourceURL enters two
+// spellings of one source URL on the generator page, requests the encrypted
+// URL made for each, and checks that the upstream receives one request and
+// that the second encrypted URL is served from the cache. One pair writes the
+// host in upper and in lower case, which /v1/image/ lower-cases. The host is
+// an IPv6 address, whose hex digits have letter case, because the fetcher
+// would look a host name up in DNS. It has a port, as the fetcher refuses an
+// IPv6 address written without one. The other pair is a source URL with no
+// path and with the path "/", both of which fetch "/".
+func TestGeneratorMakesOneImageURLForTwoSpellingsOfASourceURL(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name string
+		// sourceURLs are the two spellings, after "https://"
+		sourceURLs []string
+		// wantRequest is the path the upstream must receive
+		wantRequest string
+	}{
+		{"host in upper and in lower case",
+			[]string{"[2001:DB8::A]:8443/a.png", "[2001:db8::a]:8443/a.png"},
+			"/a.png"},
+		{"no path and the path written as a slash",
+			[]string{upstreamHost, upstreamHost + "/"}, "/"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			upstream := startRecordingUpstream(t)
+			s, _, _ := startImageProxy(t, upstream.Server)
+
+			var imageURLs []string
+
+			for _, sourceURL := range tt.sourceURLs {
+				rec := postGenerate(t, s, "https://"+sourceURL)
+
+				match := generatedURLPattern.FindStringSubmatch(rec.Body.String())
+				if rec.Code != http.StatusOK || match == nil {
+					t.Fatalf("POST /generate: %d with no encrypted URL; body %s",
+						rec.Code, rec.Body.String())
+				}
+
+				imageURLs = append(imageURLs, match[1])
+			}
+
+			getImage(t, s, imageURLs[0])
+
+			second := getImage(t, s, imageURLs[1])
+			if got := second.Header().Get("X-Pixa-Cache"); got != string(imgcache.CacheHit) {
+				t.Errorf("second X-Pixa-Cache = %q, want HIT", got)
+			}
+
+			got := upstream.requestTargets()
+			t.Logf("upstream received %q", got)
+
+			if !slices.Equal(got, []string{tt.wantRequest}) {
+				t.Errorf("upstream received %q, want [%q]", got, tt.wantRequest)
+			}
+		})
+	}
+}
+
+// TestGeneratorRefusesSourceURLItCannotFetchAsWritten submits source URLs the
+// generator page must make no encrypted URL for, and checks that it answers
+// 400 and shows why. pixa keeps the source path decoded and fetches it
+// percent-encoded again, so a path holding %2F would be fetched with a "/"
+// there, whatever else the path holds. pixa sends no user name or password,
+// and no "?" with no query after it. For these the page shows the URL pixa
+// would fetch. A source path holding a ".." segment is refused, as
+// /v1/image/ refuses it.
+func TestGeneratorRefusesSourceURLItCannotFetchAsWritten(t *testing.T) {
+	t.Parallel()
+
+	// The test config sets allow_http, so pixa would fetch over plain HTTP
+	const fetchedFrom = "http://" + upstreamHost
+
+	tests := []struct {
+		name string
+		// sourceURL is the source URL after "https://"
+		sourceURL string
+		// wantShown is what the page must show: the URL pixa would fetch, or
+		// why it makes no encrypted URL
+		wantShown string
+	}{
+		{"encoded slash", upstreamHost + "/a%2Fb.png",
+			fetchedFrom + "/a/b.png"},
+		{"encoded slash and a space", upstreamHost + "/a%2Fb c.png",
+			fetchedFrom + "/a/b%20c.png"},
+		{"encoded slash and a bar", upstreamHost + "/a%2Fb|c.png",
+			fetchedFrom + "/a/b%7Cc.png"},
+		{"encoded slash and a letter outside ASCII", upstreamHost + "/café%2Fx.png",
+			fetchedFrom + "/caf%C3%A9/x.png"},
+		{"user name and password", "user:pass@" + upstreamHost + "/a.png",
+			fetchedFrom + "/a.png"},
+		{"question mark with no query", upstreamHost + "/a.png?",
+			fetchedFrom + "/a.png"},
+		{"two-dot segment", upstreamHost + "/a/../b.png",
+			"cannot hold a .. segment"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			s, _, _ := startImageProxy(t, startRecordingUpstream(t).Server)
+
+			rec := postGenerate(t, s, "https://"+tt.sourceURL)
+			body := rec.Body.String()
+
+			if rec.Code != http.StatusBadRequest {
+				t.Errorf("status = %d, want %d", rec.Code, http.StatusBadRequest)
+			}
+
+			if strings.Contains(body, "/v1/e/") {
+				t.Error("the page shows an encrypted URL")
+			}
+
+			if !strings.Contains(body, tt.wantShown) {
+				t.Errorf("the page does not show %s; body %s", tt.wantShown, body)
+			}
+		})
+	}
+}
+
+// ipv6SourcePath is the source path of the image URLs signedImageURL and
+// encryptedImageURL make.
+const ipv6SourcePath = "/ipv6.png"
+
+// TestImageRoutesFetchIPv6SourceHostWithOrWithoutPort requests ipv6SourcePath
+// on an IPv6 source host, written with and without a port, through a
+// /v1/image/ URL with a signature and through an encrypted URL. The public
+// address 2001:db8::a, a documentation address (RFC 3849) the fetcher accepts
+// as public, answers 200, and the upstream receives the request for
+// ipv6SourcePath. The private address fd00::1 answers 403, and the upstream
+// receives no request: the test's fetcher connects every address to the test
+// upstream, so only the fetcher's check of the source URL can refuse it.
+func TestImageRoutesFetchIPv6SourceHostWithOrWithoutPort(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name       string
+		sourceHost string
+		wantStatus int
+		// wantRequests are the request targets the upstream must receive
+		wantRequests []string
+	}{
+		{"public address without a port", "[2001:db8::a]",
+			http.StatusOK, []string{ipv6SourcePath}},
+		{"public address with a port", "[2001:db8::a]:8443",
+			http.StatusOK, []string{ipv6SourcePath}},
+		{"private address without a port", "[fd00::1]",
+			http.StatusForbidden, nil},
+		{"private address with a port", "[fd00::1]:8443",
+			http.StatusForbidden, nil},
+	}
+
+	for _, tt := range tests {
+		imageURLs := map[string]string{
+			"image URL with a signature": signedImageURL(t, tt.sourceHost),
+			"encrypted URL":              encryptedImageURL(t, tt.sourceHost),
+		}
+
+		for route, imageURL := range imageURLs {
+			t.Run(tt.name+" "+route, func(t *testing.T) {
+				t.Parallel()
+
+				upstream := startRecordingUpstream(t)
+				s, _, _ := startImageProxy(t, upstream.Server)
+
+				rec := httptest.NewRecorder()
+				s.ServeHTTP(rec, httptest.NewRequestWithContext(
+					t.Context(), http.MethodGet, imageURL, nil))
+				t.Logf("GET %s: %d", imageURL, rec.Code)
+
+				if rec.Code != tt.wantStatus {
+					t.Errorf("status = %d, want %d; body %s",
+						rec.Code, tt.wantStatus, rec.Body.String())
+				}
+
+				got := upstream.requestTargets()
+				t.Logf("upstream received %q", got)
+
+				if !slices.Equal(got, tt.wantRequests) {
+					t.Errorf("upstream received %q, want %q", got, tt.wantRequests)
+				}
+			})
+		}
+	}
+}
+
+// signedImageURL returns the /v1/image/ URL of ipv6SourcePath on sourceHost at
+// its original size as PNG, with a signature made with the test signing key,
+// which a host not on the allowlist needs. The URL has no q or fit, so it is
+// signed with the quality 85 and the fit cover.
+func signedImageURL(t *testing.T, sourceHost string) string {
+	t.Helper()
+
+	path, sig, exp := signature.New(testSigningKey).GenerateSignedURL(
+		&signature.Request{
+			SourceHost: sourceHost,
+			SourcePath: ipv6SourcePath,
+			Format:     string(imgcache.FormatPNG),
+			Quality:    85,
+			FitMode:    string(imgcache.FitCover),
+		}, time.Hour)
+
+	return path + "?sig=" + sig + "&exp=" + strconv.FormatInt(exp, 10)
+}
+
+// encryptedImageURL returns the encrypted URL of ipv6SourcePath on sourceHost
+// at its original size as PNG, made with the test signing key.
+func encryptedImageURL(t *testing.T, sourceHost string) string {
+	t.Helper()
+
+	generator, err := encurl.NewGenerator(testSigningKey)
+	if err != nil {
+		t.Fatalf("encurl.NewGenerator: %v", err)
+	}
+
+	token, err := generator.Generate(&encurl.Payload{
+		SourceHost: sourceHost, SourcePath: ipv6SourcePath,
+		Format: imgcache.FormatPNG,
+	})
+	if err != nil {
+		t.Fatalf("encurl Generate: %v", err)
+	}
+
+	return "/v1/e/" + token + "/img.png"
+}
+
+// TestImageRoutesRefuseIPv6SourceHostHoldingInternalIPv4Address requests
+// ipv6SourcePath on IPv6 source hosts that hold a loopback or private IPv4
+// address in their last four bytes, in the IPv4-compatible form (::/96) or the
+// NAT64 form (64:ff9b::/96), through a /v1/image/ URL with a signature and
+// through an encrypted URL. Each answers 403, and the upstream receives no
+// request: the test's fetcher connects every address to the test upstream, so
+// only the fetcher's check of the source URL can refuse it.
+func TestImageRoutesRefuseIPv6SourceHostHoldingInternalIPv4Address(t *testing.T) {
+	t.Parallel()
+
+	upstream := startRecordingUpstream(t)
+	s, _, _ := startImageProxy(t, upstream.Server)
+
+	sourceHosts := []string{
+		"[::127.0.0.1]",     // IPv4-compatible form of a loopback address
+		"[::10.0.0.1]",      // IPv4-compatible form of a private address
+		"[64:ff9b::7f00:1]", // NAT64 form of 127.0.0.1
+	}
+
+	for _, sourceHost := range sourceHosts {
+		imageURLs := []string{
+			signedImageURL(t, sourceHost),
+			encryptedImageURL(t, sourceHost),
+		}
+
+		for _, imageURL := range imageURLs {
+			rec := httptest.NewRecorder()
+			s.ServeHTTP(rec, httptest.NewRequestWithContext(
+				t.Context(), http.MethodGet, imageURL, nil))
+			t.Logf("GET %s (source host %s): %d", imageURL, sourceHost, rec.Code)
+
+			if rec.Code != http.StatusForbidden {
+				t.Errorf("source host %s: status = %d, want %d; body %s",
+					sourceHost, rec.Code, http.StatusForbidden, rec.Body.String())
+			}
+		}
+	}
+
+	if got := upstream.requestTargets(); len(got) != 0 {
+		t.Errorf("upstream received %q, want no request", got)
+	}
+}
+
+// TestImageRoutesCheckIPv4AddressInIPv6SourceHost requests ipv6SourcePath on
+// source hosts that hold an IPv4 address, through a /v1/image/ URL with a
+// signature and through an encrypted URL, with the fetcher pixad uses.
+// [64:ff9b:1::a00:1], in the local-use NAT64 range (64:ff9b:1::/48, RFC 8215),
+// answers 403. [::203.0.113.5] holds the public address 203.0.113.5 in the
+// IPv4-compatible form (::/96) and answers 403 when blocked_networks lists
+// 203.0.113.0/24 or ::/96. [::ffff:203.0.113.5] holds it in the IPv4-mapped
+// form (::ffff:0:0/96) and answers 403 when blocked_networks lists
+// 203.0.113.0/24. Go keeps the plain address 203.0.113.5 in the mapped form
+// too, and an IPv6 range must not refuse it: with ::/0 listed it is not
+// refused. Neither is [::203.0.113.5] with blocked_networks empty. Where a
+// host is not refused, the fetcher tries to connect to it, finds nothing
+// there, and the request fails with another status.
+func TestImageRoutesCheckIPv4AddressInIPv6SourceHost(t *testing.T) {
+	t.Parallel()
+
+	const (
+		ipv4CompatibleHost = "[::203.0.113.5]"
+		ipv4MappedHost     = "[::ffff:203.0.113.5]"
+	)
+
+	tests := []struct {
+		name            string
+		sourceHost      string
+		blockedNetworks []netip.Prefix
+		wantForbidden   bool
+	}{
+		{"local-use NAT64 form of 10.0.0.1", "[64:ff9b:1::a00:1]", nil, true},
+		{"IPv4-compatible form of an address in blocked_networks",
+			ipv4CompatibleHost,
+			[]netip.Prefix{netip.MustParsePrefix("203.0.113.0/24")}, true},
+		{"IPv4-compatible form with ::/96 in blocked_networks",
+			ipv4CompatibleHost,
+			[]netip.Prefix{netip.MustParsePrefix("::/96")}, true},
+		{"IPv4-mapped form of an address in blocked_networks",
+			ipv4MappedHost,
+			[]netip.Prefix{netip.MustParsePrefix("203.0.113.0/24")}, true},
+		{"plain IPv4 address with ::/0 in blocked_networks",
+			"203.0.113.5",
+			[]netip.Prefix{netip.MustParsePrefix("::/0")}, false},
+		{"IPv4-compatible form of an address with blocked_networks empty",
+			ipv4CompatibleHost, nil, false},
+	}
+
+	for _, tt := range tests {
+		imageURLs := map[string]string{
+			"image URL with a signature": signedImageURL(t, tt.sourceHost),
+			"encrypted URL":              encryptedImageURL(t, tt.sourceHost),
+		}
+
+		for route, imageURL := range imageURLs {
+			t.Run(tt.name+" "+route, func(t *testing.T) {
+				t.Parallel()
+
+				s := startImageProxyWithBlockedNetworks(t, tt.blockedNetworks)
+
+				rec := httptest.NewRecorder()
+				s.ServeHTTP(rec, httptest.NewRequestWithContext(
+					t.Context(), http.MethodGet, imageURL, nil))
+				t.Logf("GET %s: %d", imageURL, rec.Code)
+
+				forbidden := rec.Code == http.StatusForbidden
+				if tt.wantForbidden && !forbidden {
+					t.Errorf("status = %d, want %d; body %s",
+						rec.Code, http.StatusForbidden, rec.Body.String())
+				}
+
+				if !tt.wantForbidden && forbidden {
+					t.Errorf("status = %d, want any other status; body %s",
+						rec.Code, rec.Body.String())
+				}
+			})
+		}
+	}
+}

internal/signature/signature.go

@@ -126,31 +126,20 @@ func (s *Signer) GenerateSignedURL(
 		sizeStr = fmt.Sprintf("%dx%d", req.Width, req.Height)
 	}
 
-	// Build the path.
+	// Build the path. A source query follows the source path after a "?".
-	// When a source query is present, it is embedded as a path segment
+	// The image route reads the path decoded once, so the whole path is
-	// (e.g. /host/path?query/size.fmt) so that the URL parser can extract
+	// written in Go's own percent-encoding of a URL path, which writes "?"
-	// it from the last-slash split. The "?" inside a path segment is
+	// as %3F and "%" as %25, and leaves "/", ";", ",", "=" and "&" as they
-	// percent-encoded by clients but chi delivers it decoded, which is
+	// are.
-	// exactly what the URL parser expects.
+	sourcePathAndQuery := req.SourcePath
-	var path string
 	if req.SourceQuery != "" {
-		path = fmt.Sprintf("/v1/image/%s%s%%3F%s/%s.%s",
+		sourcePathAndQuery += "?" + req.SourceQuery
-			req.SourceHost,
-			req.SourcePath,
-			url.PathEscape(req.SourceQuery),
-			sizeStr,
-			req.Format,
-		)
-	} else {
-		path = fmt.Sprintf("/v1/image/%s%s/%s.%s",
-			req.SourceHost,
-			req.SourcePath,
-			sizeStr,
-			req.Format,
-		)
 	}
 
-	return path, sig, exp
+	unescapedPath := fmt.Sprintf("/v1/image/%s%s/%s.%s",
+		req.SourceHost, sourcePathAndQuery, sizeStr, req.Format)
+
+	return (&url.URL{Path: unescapedPath}).EscapedPath(), sig, exp
 }
 
 // buildSignatureData creates the string to be signed.

internal/signature/signature_test.go

@@ -475,6 +475,44 @@ func TestGenerateSignedURL_WithoutQueryString(t *testing.T) {
 	}
 }
 
+// TestGenerateSignedURL_ReadmeQueryExample checks the README's example of a
+// source URL with a query, the image at
+// https://cdn.example.com/img?url=https://origin.example.com/my%20cat.jpg: its
+// signature with the README's key, and the path GenerateSignedURL writes for
+// it, with the "?" written as %3F and the "%" as %25.
+func TestGenerateSignedURL_ReadmeQueryExample(t *testing.T) {
+	t.Parallel()
+
+	signer := signature.New("example-signing-key-for-documentation")
+
+	req := signature.Request{
+		SourceHost:  testHost,
+		SourcePath:  "/img",
+		SourceQuery: "url=https://origin.example.com/my%20cat.jpg",
+		Width:       800,
+		Height:      600,
+		Format:      testFormatWebP,
+		Quality:     85,
+		FitMode:     testFitCover,
+		Expires:     time.Unix(1704067200, 0),
+	}
+
+	const wantSig = "37aFa9GSVCZM1kRY9SzMOk2fOg_M7Wt5gKNR4xM-FYA="
+
+	gotSig := signer.Sign(&req)
+	if gotSig != wantSig {
+		t.Errorf("Sign() = %q, want %q", gotSig, wantSig)
+	}
+
+	const wantPath = "/v1/image/cdn.example.com" +
+		"/img%3Furl=https://origin.example.com/my%2520cat.jpg/800x600.webp"
+
+	gotPath, _, _ := signer.GenerateSignedURL(&req, time.Hour)
+	if gotPath != wantPath {
+		t.Errorf("GenerateSignedURL() path = %q, want %q", gotPath, wantPath)
+	}
+}
+
 func TestParseParams(t *testing.T) {
 	t.Parallel()
 

script/bootstrap

@@ -19,9 +19,8 @@
 #                            CGO image libraries (pkg-config, vips
 #                            with its JPEG XL support, libheif) for
 #                            the govips bindings instead of Node: to
-#                            compile pixa, in the Dockerfile's test
+#                            compile pixa outside Docker, as
-#                            phase and build stage, which format
+#                            `make build` does
-#                            nothing
 set -eu
 
 ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"

script/cibuild

@@ -2,11 +2,12 @@
 # script/cibuild: run the CI build. It bootstraps first: a CI runner
 # checks out and runs this and nothing else, and script/fmt-check runs
 # the formatter on the host, which a pristine checkout cannot do.
-# A new CACHEBUST value reruns every RUN step, since a cached gate phase
+# A new CACHEBUST value reruns every RUN step after the ARG CACHEBUST
-# is a check that did not run; unlike --no-cache, it keeps Go's build cache.
+# line, since a cached gate phase is a check that did not run; unlike
+# --no-cache, it keeps the layers above that line and the cache mounts.
 # This relies on every stage of the Dockerfile declaring ARG CACHEBUST
-# first; a stage without that line is served from cache on an unchanged
+# before every step that lints, tests or compiles; a stage without that
-# tree, with no warning.
+# line is served from cache on an unchanged tree, with no warning.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"

script/docker

@@ -1,11 +1,12 @@
 #!/bin/sh
 # script/docker: build the Docker image tagged with the project name.
 # Identical in all repos; the tag comes from script/projectname.
-# A new CACHEBUST value reruns every RUN step, since a cached gate phase
+# A new CACHEBUST value reruns every RUN step after the ARG CACHEBUST
-# is a check that did not run; unlike --no-cache, it keeps Go's build cache.
+# line, since a cached gate phase is a check that did not run; unlike
+# --no-cache, it keeps the layers above that line and the cache mounts.
 # This relies on every stage of the Dockerfile declaring ARG CACHEBUST
-# first; a stage without that line is served from cache on an unchanged
+# before every step that lints, tests or compiles; a stage without that
-# tree, with no warning.
+# line is served from cache on an unchanged tree, with no warning.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"

script/lint

@@ -5,11 +5,13 @@
 # make its answer untrustworthy.
 #
 # The phase is not the last stage in the file, so it is built only when
-# --target names it. A new CACHEBUST value reruns every RUN step, since
+# --target names it. A new CACHEBUST value reruns every RUN step after
-# a cached lint layer is a lint that did not run; unlike --no-cache, it
+# the ARG CACHEBUST line, since a cached lint layer is a lint that did
-# keeps Go's build cache. This relies on every stage of the Dockerfile
+# not run; unlike --no-cache, it keeps the layers above that line and
-# declaring ARG CACHEBUST first; a stage without that line is served
+# the cache mounts. This relies on every stage of the Dockerfile
-# from cache on an unchanged tree, with no warning.
+# declaring ARG CACHEBUST before every step that lints, tests or
+# compiles; a stage without that line is served from cache on an
+# unchanged tree, with no warning.
 # --output type=cacheonly writes no image, since nothing uses one.
 set -eu
 

script/test

@@ -3,11 +3,12 @@
 # and this builds that phase alone, on the same terms as script/lint:
 # --target because a phase that is not the last stage is built only when
 # named, and a new CACHEBUST value because a cached test layer is a test
-# that did not run; unlike --no-cache, it keeps Go's build cache. This
+# that did not run; unlike --no-cache, it keeps the layers above the ARG
-# relies on every stage of the Dockerfile declaring ARG CACHEBUST first;
+# CACHEBUST line and the cache mounts. This relies on every stage of the
-# a stage without that line is served from cache on an unchanged tree,
+# Dockerfile declaring ARG CACHEBUST before every step that lints, tests
-# with no warning. --output type=cacheonly writes no image, since
+# or compiles; a stage without that line is served from cache on an
-# nothing uses one.
+# unchanged tree, with no warning. --output type=cacheonly writes no
+# image, since nothing uses one.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"