Restyle the sources and fix the audit's findings

1 file changed, 712 insertions, 637 deletions

diff --git a/source/cgit.c b/source/cgit.c
index 7cce3d3..5c11a93 100644
--- a/source/cgit.c
+++ b/source/cgit.c
@@ -1,43 +1,74 @@
-/* cgit.c: cgi for the git scm
- *
- * Copyright (C) 2006-2014 cgit Development Team <cgit@lists.zx2c4.com>
- *
- * Licensed under GNU General Public License v2
- *   (see LICENSE.txt for full license text)
+/*
+ * The entry point every request passes through. It reads cgitrc and the query
+ * string into the global ctx, works out how long the answer may be cached,
+ * and then hands over to the cache, which either serves a copy it already has
+ * or calls back here to render one. Rendering means authenticating the
+ * request, resolving the repository and the branch it names, and dispatching
+ * to the page handler cmd.c maps the page name to. Run from a shell rather
+ * than from a web server the same binary instead prints what it was built
+ * with, or scans a directory tree and writes out the repositories it found in
+ * cgitrc form.
  */
 
 #define USE_THE_REPOSITORY_VARIABLE
 
-#include "cgit.h"
 #include "cache.h"
+#include "cgit.h"
 #include "cmd.h"
-#include "configfile.h"
+#include "config.h"
+#include "filter.h"
 #include "html.h"
-#include "ui-shared.h"
-#include "ui-stats.h"
+#include "parsing.h"
+#include "scan-tree.h"
+#include "shared.h"
 #include "ui-blob.h"
+#include "ui-diff.h"
 #include "ui-empty.h"
-#include "ui-summary.h"
-#include "scan-tree.h"
+#include "ui-shared.h"
+#include "ui-snapshot.h"
+#include "ui-stats.h"
 
-/* We intentionally keep this rather small, instead of looping and
- * feeding it to the filter a couple bytes at a time. This way, the
- * filter itself does not need to handle any denial of service or
- * buffer bloat issues. If this winds up being too small, people
- * will complain on the mailing list, and we'll increase it as needed. */
+// Deliberately small. The whole body is read at once and handed to the auth
+// filter, so a filter never has to defend itself against a huge or a
+// dribbled-out POST.
 #define MAX_AUTHENTICATION_POST_BYTES 4096
+
+// The filter bar matches the search against every item a page lists, so this
+// bounds both the work one request can ask for and the cache key the query
+// lands in.
+#define MAX_SEARCH_LEN 512
+
+// Furthest into a listing a request may ask to start. A walk has to step over
+// every row it skips, so this bounds the work an offset alone can buy.
+#define MAX_QUERY_OFFSET 100000
+
+// cgit knows fewer than eight archive formats, so this stands in for every bit
+// a snapshots mask can carry.
+#define ALL_SNAPSHOT_FORMATS 0xFF
+
+// An Expires header has no way of saying never, so a page whose ttl is
+// negative claims ten years.
+#define NEVER_EXPIRES_SECONDS (10 * 365 * 24 * 60 * 60)
+
+/*
+ * The first branch found is the fallback, so a repository whose default branch
+ * does not exist still has something to show.
+ */
+struct refmatch {
+	char *wanted;
+	char *first;
+	int found;
+};
+
 const char *cgit_version = CGIT_VERSION;
 
 /*
- * Isolate git from the calling user's configuration. Ignore the system and
- * global config and attributes, so a snapshot cannot be broken by something
- * like a core.excludesfile pointing at a "~" path that git can no longer
- * expand once HOME is unset below.
- *
- * Called at the top of cmd_main rather than from a constructor attribute.
- * git-compat-util.h defines __attribute__ away on a compiler that does not
- * support it, which would leave this silently never running. Nothing git does
- * before cmd_main reads configuration, so an ordinary call is equivalent.
+ * Isolate git from the calling user's configuration, so a snapshot cannot be
+ * broken by something like a core.excludesfile pointing at a "~" path that git
+ * can no longer expand once HOME is unset below. Called from cmd_main rather
+ * than from a constructor attribute, because git-compat-util.h defines
+ * __attribute__ away on a compiler that does not support it, which would leave
+ * this silently never running.
  */
 static void isolate_git_environment(void)
 {
@@ -48,101 +79,385 @@ static void isolate_git_environment(void)
 	unsetenv("XDG_CONFIG_HOME");
 }
 
-static void add_mimetype(const char *name, const char *value)
+/*
+ * Turn a die from anywhere inside git into a rendered page, since a CGI that
+ * produced no output leaves the visitor with whatever the web server makes of
+ * it.
+ */
+static NORETURN void die_routine(const char *msg, va_list params)
 {
-	struct string_list_item *item;
+	// A page is rendered with stdout pointed at the cache file, so the
+	// message would otherwise be written there instead of to the visitor.
+	cache_abandon_fill();
+	cgit_vprint_error_page(400, "Bad request", msg, params);
+	exit(0);
+}
 
-	item = string_list_insert(&ctx.cfg.mimetypes, name);
-	item->util = xstrdup(value);
+static void prepare_context(void)
+{
+	memset(&ctx, 0, sizeof(ctx));
+	ctx.cfg.agefile = "info/web/last-modified";
+	ctx.cfg.cache_size = 0;
+	ctx.cfg.cache_root = CGIT_CACHE_ROOT;
+	ctx.cfg.cache_about_ttl = 15;
+	ctx.cfg.cache_snapshot_ttl = 5;
+	ctx.cfg.cache_repo_ttl = 5;
+	ctx.cfg.cache_root_ttl = 5;
+	ctx.cfg.cache_scanrc_ttl = 15;
+	ctx.cfg.cache_dynamic_ttl = 5;
+	ctx.cfg.cache_static_ttl = -1;
+	ctx.cfg.case_sensitive_sort = 1;
+	ctx.cfg.branch_sort = 0;
+	ctx.cfg.commit_sort = 0;
+	ctx.cfg.logo = "/cgit.png";
+	ctx.cfg.favicon = "/favicon.ico";
+	ctx.cfg.local_time = 0;
+	ctx.cfg.date_mode = date_mode_from_type(DATE_SHORT);
+	ctx.cfg.enable_relative_dates = 1;
+	ctx.cfg.enable_http_clone = 1;
+	ctx.cfg.enable_index_owner = 1;
+	ctx.cfg.enable_tree_linenumbers = 1;
+	ctx.cfg.enable_git_config = 0;
+	ctx.cfg.max_repo_count = 50;
+	ctx.cfg.max_commit_count = 50;
+	ctx.cfg.max_patch_count = 50;
+	ctx.cfg.max_diff_files = 200;
+	ctx.cfg.max_diff_lines = 1000;
+	ctx.cfg.max_msg_len = 80;
+	ctx.cfg.max_ref_count = 200;
+	ctx.cfg.max_repodesc_len = 80;
+	// Counted in kilobytes, so ten megabytes, which bounds the memory one
+	// request can be made to allocate for a blob.
+	ctx.cfg.max_blob_size = 10 * 1024;
+	ctx.cfg.max_stats = 0;
+	ctx.cfg.project_list = NULL;
+	ctx.cfg.renamelimit = -1;
+	ctx.cfg.remove_suffix = 0;
+	ctx.cfg.robots = "index, nofollow";
+	ctx.cfg.root_title = "Git repository browser";
+	ctx.cfg.root_desc = "a fast webinterface for the git dscm";
+	ctx.cfg.scan_hidden_path = 0;
+	ctx.cfg.script_name = CGIT_SCRIPT_NAME;
+	ctx.cfg.section = "";
+	ctx.cfg.repository_sort = "name";
+	ctx.cfg.section_sort = 0;
+	ctx.cfg.summary_branches = 10;
+	ctx.cfg.summary_log = 10;
+	ctx.cfg.summary_tags = 10;
+	ctx.cfg.max_atom_items = 10;
+	ctx.cfg.difftype = DIFF_UNIFIED;
+	ctx.env.cgit_config = getenv("CGIT_CONFIG");
+	ctx.env.http_host = getenv("HTTP_HOST");
+	ctx.env.https = getenv("HTTPS");
+	ctx.env.no_http = getenv("NO_HTTP");
+	ctx.env.path_info = getenv("PATH_INFO");
+	ctx.env.query_string = getenv("QUERY_STRING");
+	ctx.env.request_method = getenv("REQUEST_METHOD");
+	ctx.env.script_name = getenv("SCRIPT_NAME");
+	ctx.env.server_name = getenv("SERVER_NAME");
+	ctx.env.server_port = getenv("SERVER_PORT");
+	ctx.env.http_cookie = getenv("HTTP_COOKIE");
+	ctx.env.http_referer = getenv("HTTP_REFERER");
+	ctx.env.content_length = getenv("CONTENT_LENGTH") ?
+		strtoul(getenv("CONTENT_LENGTH"), NULL, 10) : 0;
+	ctx.env.authenticated = 0;
+	ctx.page.mimetype = "text/html";
+	ctx.page.charset = PAGE_ENCODING;
+	ctx.page.filename = NULL;
+	ctx.page.size = 0;
+	ctx.page.modified = time(NULL);
+	ctx.page.expires = ctx.page.modified;
+	ctx.page.etag = NULL;
+	string_list_init_dup(&ctx.cfg.mimetypes);
+	if (ctx.env.script_name)
+		ctx.cfg.script_name = xstrdup(ctx.env.script_name);
+	if (ctx.env.query_string)
+		ctx.qry.raw = xstrdup(ctx.env.query_string);
+	if (!ctx.env.cgit_config)
+		ctx.env.cgit_config = CGIT_CONFIG;
+}
+
+static void print_version(void)
+{
+	printf("CGit %s | https://github.com/brycekwon/cgit\n\nCompiled in features:\n", CGIT_VERSION);
+#ifdef NO_LUA
+	printf("[-] ");
+#else
+	printf("[+] ");
+#endif
+	printf("Lua scripting\n");
+#ifndef HAVE_LINUX_SENDFILE
+	printf("[-] ");
+#else
+	printf("[+] ");
+#endif
+	printf("Linux sendfile() usage\n");
 }
 
-static void process_cached_repolist(const char *path);
+static int cmp_repos(const void *a, const void *b)
+{
+	const struct cgit_repo *repo_a = a, *repo_b = b;
+	return strcmp(repo_a->url, repo_b->url);
+}
 
-void cgit_repo_config(struct cgit_repo *repo, const char *name, const char *value)
+static char *build_snapshot_setting(int mask)
+{
+	const struct cgit_snapshot_format *format;
+	struct strbuf result = STRBUF_INIT;
+
+	for (format = cgit_snapshot_formats; format->suffix; format++) {
+		if (cgit_snapshot_format_bit(format) & mask) {
+			if (result.len)
+				strbuf_addch(&result, ' ');
+			strbuf_addstr(&result, format->suffix);
+		}
+	}
+	return strbuf_detach(&result, NULL);
+}
+
+static void print_repo(FILE *f, struct cgit_repo *repo)
 {
-	const char *path;
 	struct string_list_item *item;
 
-	if (!strcmp(name, "name"))
-		repo->name = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "clone-url"))
-		repo->clone_url = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "desc"))
-		repo->desc = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "owner"))
-		repo->owner = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "homepage"))
-		repo->homepage = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "defbranch"))
-		repo->defbranch = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "extra-head-content"))
-		repo->extra_head_content = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "snapshots"))
-		repo->snapshots = ctx.cfg.snapshots & cgit_parse_snapshots_mask(value);
-	else if (!strcmp(name, "enable-blame"))
-		repo->enable_blame = atoi(value);
-	else if (!strcmp(name, "enable-commit-graph"))
-		repo->enable_commit_graph = atoi(value);
-	else if (!strcmp(name, "enable-follow-links"))
-		repo->enable_follow_links = atoi(value);
-	else if (!strcmp(name, "enable-log-filecount"))
-		repo->enable_log_filecount = atoi(value);
-	else if (!strcmp(name, "enable-log-linecount"))
-		repo->enable_log_linecount = atoi(value);
-	else if (!strcmp(name, "enable-remote-branches"))
-		repo->enable_remote_branches = atoi(value);
-	else if (!strcmp(name, "enable-subject-links"))
-		repo->enable_subject_links = atoi(value);
-	else if (!strcmp(name, "enable-html-serving"))
-		repo->enable_html_serving = atoi(value);
-	else if (!strcmp(name, "enable-stats"))
-		repo->enable_stats = atoi(value);
-	else if (!strcmp(name, "branch-sort")) {
-		if (!strcmp(value, "age"))
-			repo->branch_sort = 1;
-		if (!strcmp(value, "name"))
-			repo->branch_sort = 0;
-	} else if (!strcmp(name, "commit-sort")) {
-		if (!strcmp(value, "date"))
-			repo->commit_sort = 1;
-		if (!strcmp(value, "topo"))
-			repo->commit_sort = 2;
-	} else if (!strcmp(name, "max-stats"))
-		repo->max_stats = cgit_find_stats_period(value, NULL);
-	else if (!strcmp(name, "module-link"))
-		repo->module_link= cgit_strdup_first_line(value);
-	else if (skip_prefix(name, "module-link.", &path)) {
-		item = string_list_append(&repo->submodules, cgit_strdup_first_line(path));
-		item->util = cgit_strdup_first_line(value);
-	} else if (!strcmp(name, "section"))
-		repo->section = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "snapshot-prefix"))
-		repo->snapshot_prefix = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "readme") && value != NULL) {
-		if (repo->readme.items == ctx.cfg.readme.items)
-			memset(&repo->readme, 0, sizeof(repo->readme));
-		string_list_append(&repo->readme, cgit_strdup_first_line(value));
-	} else if (!strcmp(name, "logo") && value != NULL)
-		repo->logo = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "logo-link") && value != NULL)
-		repo->logo_link = cgit_strdup_first_line(value);
-	else if (!strcmp(name, "hide"))
-		repo->hide = atoi(value);
-	else if (!strcmp(name, "ignore"))
-		repo->ignore = atoi(value);
-	else if (ctx.cfg.enable_filter_overrides) {
-		if (!strcmp(name, "about-filter"))
-			repo->about_filter = cgit_new_filter(value, ABOUT);
-		else if (!strcmp(name, "commit-filter"))
-			repo->commit_filter = cgit_new_filter(value, COMMIT);
-		else if (!strcmp(name, "source-filter"))
-			repo->source_filter = cgit_new_filter(value, SOURCE);
-		else if (!strcmp(name, "email-filter"))
-			repo->email_filter = cgit_new_filter(value, EMAIL);
+	fprintf(f, "repo.url=%s\n", repo->url);
+	fprintf(f, "repo.name=%s\n", repo->name);
+	fprintf(f, "repo.path=%s\n", repo->path);
+	if (repo->owner)
+		fprintf(f, "repo.owner=%s\n", repo->owner);
+	if (repo->desc)
+		fprintf(f, "repo.desc=%s\n", repo->desc);
+	for_each_string_list_item(item, &repo->readme) {
+		if (item->util)
+			fprintf(f, "repo.readme=%s:%s\n", (char *)item->util, item->string);
+		else
+			fprintf(f, "repo.readme=%s\n", item->string);
+	}
+	if (repo->defbranch)
+		fprintf(f, "repo.defbranch=%s\n", repo->defbranch);
+	if (repo->extra_head_content)
+		fprintf(f, "repo.extra-head-content=%s\n", repo->extra_head_content);
+	if (repo->module_link)
+		fprintf(f, "repo.module-link=%s\n", repo->module_link);
+	if (repo->section)
+		fprintf(f, "repo.section=%s\n", repo->section);
+	if (repo->homepage)
+		fprintf(f, "repo.homepage=%s\n", repo->homepage);
+	if (repo->clone_url)
+		fprintf(f, "repo.clone-url=%s\n", repo->clone_url);
+	fprintf(f, "repo.enable-blame=%d\n", repo->enable_blame);
+	fprintf(f, "repo.enable-commit-graph=%d\n", repo->enable_commit_graph);
+	fprintf(f, "repo.enable-follow-links=%d\n", repo->enable_follow_links);
+	fprintf(f, "repo.enable-log-filecount=%d\n", repo->enable_log_filecount);
+	fprintf(f, "repo.enable-log-linecount=%d\n", repo->enable_log_linecount);
+	if (repo->about_filter && repo->about_filter != ctx.cfg.about_filter)
+		cgit_fprintf_filter(repo->about_filter, f, "repo.about-filter=");
+	if (repo->commit_filter && repo->commit_filter != ctx.cfg.commit_filter)
+		cgit_fprintf_filter(repo->commit_filter, f, "repo.commit-filter=");
+	if (repo->source_filter && repo->source_filter != ctx.cfg.source_filter)
+		cgit_fprintf_filter(repo->source_filter, f, "repo.source-filter=");
+	if (repo->email_filter && repo->email_filter != ctx.cfg.email_filter)
+		cgit_fprintf_filter(repo->email_filter, f, "repo.email-filter=");
+	if (repo->snapshots != ctx.cfg.snapshots) {
+		char *formats = build_snapshot_setting(repo->snapshots);
+		fprintf(f, "repo.snapshots=%s\n", formats ? formats : "");
+		free(formats);
+	}
+	if (repo->snapshot_prefix)
+		fprintf(f, "repo.snapshot-prefix=%s\n", repo->snapshot_prefix);
+	if (repo->enable_stats != ctx.cfg.enable_stats)
+		fprintf(f, "repo.enable-stats=%d\n", repo->enable_stats);
+	if (repo->max_stats != ctx.cfg.max_stats)
+		fprintf(f, "repo.max-stats=%s\n",
+			cgit_find_stats_periodname(repo->max_stats));
+	if (repo->logo)
+		fprintf(f, "repo.logo=%s\n", repo->logo);
+	if (repo->logo_link)
+		fprintf(f, "repo.logo-link=%s\n", repo->logo_link);
+	fprintf(f, "repo.enable-remote-branches=%d\n", repo->enable_remote_branches);
+	fprintf(f, "repo.enable-subject-links=%d\n", repo->enable_subject_links);
+	fprintf(f, "repo.enable-html-serving=%d\n", repo->enable_html_serving);
+	if (repo->branch_sort == 1)
+		fprintf(f, "repo.branch-sort=age\n");
+	if (repo->commit_sort) {
+		if (repo->commit_sort == 1)
+			fprintf(f, "repo.commit-sort=date\n");
+		else if (repo->commit_sort == 2)
+			fprintf(f, "repo.commit-sort=topo\n");
+	}
+	fprintf(f, "repo.hide=%d\n", repo->hide);
+	fprintf(f, "repo.ignore=%d\n", repo->ignore);
+	fprintf(f, "\n");
+}
+
+static void print_repolist(FILE *f, struct cgit_repolist *list, int start)
+{
+	int i;
+
+	for (i = start; i < list->count; i++)
+		print_repo(f, &list->repos[i]);
+}
+
+static void parse_args(int argc, const char **argv)
+{
+	int i;
+	const char *arg;
+	int scanned = 0;
+
+	for (i = 1; i < argc; i++) {
+		if (!strcmp(argv[i], "--version")) {
+			print_version();
+			exit(0);
+		}
+		if (skip_prefix(argv[i], "--cache=", &arg)) {
+			ctx.cfg.cache_root = xstrdup(arg);
+		} else if (!strcmp(argv[i], "--nohttp")) {
+			ctx.env.no_http = "1";
+		} else if (skip_prefix(argv[i], "--query=", &arg)) {
+			ctx.qry.raw = xstrdup(arg);
+		} else if (skip_prefix(argv[i], "--repo=", &arg)) {
+			ctx.qry.repo = xstrdup(arg);
+		} else if (skip_prefix(argv[i], "--page=", &arg)) {
+			ctx.qry.page = xstrdup(arg);
+		} else if (skip_prefix(argv[i], "--head=", &arg)) {
+			ctx.qry.head = xstrdup(arg);
+			ctx.qry.has_symref = 1;
+		} else if (skip_prefix(argv[i], "--oid=", &arg)) {
+			ctx.qry.oid = xstrdup(arg);
+			ctx.qry.has_oid = 1;
+		} else if (skip_prefix(argv[i], "--ofs=", &arg)) {
+			ctx.qry.ofs = atoi(arg);
+		} else if (skip_prefix(argv[i], "--scan-tree=", &arg) ||
+		           skip_prefix(argv[i], "--scan-path=", &arg)) {
+			// A repository's own snapshots setting is masked with
+			// the global one, which normally comes from cgitrc.
+			// That has not been read yet here, so an empty mask
+			// would discard whatever the repository asked for.
+			ctx.cfg.snapshots = ALL_SNAPSHOT_FORMATS;
+			scanned++;
+			scan_tree(arg);
+		}
+	}
+	if (scanned) {
+		qsort(cgit_repolist.repos, cgit_repolist.count,
+			sizeof(struct cgit_repo), cmp_repos);
+		print_repolist(stdout, &cgit_repolist, 0);
+		exit(0);
+	}
+}
+
+static int generate_cached_repolist(const char *path, const char *cached_rc)
+{
+	struct strbuf locked_rc = STRBUF_INIT;
+	int err = 0;
+	int first;
+	FILE *f;
+
+	strbuf_addf(&locked_rc, "%s.lock", cached_rc);
+	f = fopen(locked_rc.buf, "wx");
+	if (!f) {
+		// An existing lock file only means concurrent requests, which
+		// is not worth a line in the server log.
+		err = errno;
+		if (err != EEXIST)
+			fprintf(stderr, "[cgit] Error opening %s: %s (%d)\n",
+				locked_rc.buf, strerror(err), err);
+		goto out;
+	}
+	first = cgit_repolist.count;
+	if (ctx.cfg.project_list)
+		scan_projects(path, ctx.cfg.project_list);
+	else
+		scan_tree(path);
+	print_repolist(f, &cgit_repolist, first);
+	// Closed before the rename, because print_repolist writes through stdio
+	// and a rename over the live file would otherwise publish a repolist
+	// that stops wherever the buffer happened to end.
+	if (fclose(f)) {
+		err = errno;
+		fprintf(stderr, "[cgit] Error writing %s: %s (%d)\n",
+			locked_rc.buf, strerror(err), err);
+		unlink(locked_rc.buf);
+		goto out;
+	}
+	if (rename(locked_rc.buf, cached_rc)) {
+		err = errno;
+		fprintf(stderr, "[cgit] Error renaming %s to %s: %s (%d)\n",
+			locked_rc.buf, cached_rc, strerror(err), err);
+		unlink(locked_rc.buf);
+	}
+out:
+	strbuf_release(&locked_rc);
+	return err;
+}
+
+// A cached repolist is itself a config file, so these two call each other.
+static void apply_config(const char *name, const char *value);
+
+static void process_cached_repolist(const char *path)
+{
+	struct stat st;
+	struct strbuf cached_rc = STRBUF_INIT;
+	time_t age;
+	unsigned long hash;
+	int devnull;
+
+	hash = cache_hash_str(path);
+	if (ctx.cfg.project_list)
+		hash += cache_hash_str(ctx.cfg.project_list);
+	strbuf_addf(&cached_rc, "%s/rc-%8lx", ctx.cfg.cache_root, hash);
+
+	if (stat(cached_rc.buf, &st)) {
+		// Nothing is cached yet, so this request scans in its own
+		// process, leaving no copy behind when it cannot take the lock.
+		if (generate_cached_repolist(path, cached_rc.buf)) {
+			if (ctx.cfg.project_list)
+				scan_projects(path, ctx.cfg.project_list);
+			else
+				scan_tree(path);
+		}
+		goto out;
+	}
+
+	config_file_parse(cached_rc.buf, apply_config);
+
+	age = time(NULL) - st.st_mtime;
+	if (age <= (ctx.cfg.cache_scanrc_ttl * 60))
+		goto out;
+
+	// The list just parsed is stale but usable, so a child rebuilds it
+	// while this request answers from what it already has.
+	if (fork())
+		goto out;
+
+	// The child inherits the descriptors of the request, and the web server
+	// reads stdout until every holder of it is gone, so leaving them in
+	// place would keep the visitor waiting for the whole scan after their
+	// page was written. Anything the scan prints would land on that
+	// response as well.
+	devnull = open("/dev/null", O_RDWR);
+	if (devnull >= 0) {
+		dup2(devnull, STDIN_FILENO);
+		dup2(devnull, STDOUT_FILENO);
+		dup2(devnull, STDERR_FILENO);
+		if (devnull > STDERR_FILENO)
+			close(devnull);
 	}
+	// _exit rather than exit, so the handlers the request registered do
+	// not run a second time in the child.
+	_exit(generate_cached_repolist(path, cached_rc.buf));
+out:
+	strbuf_release(&cached_rc);
 }
 
-static void config_cb(const char *name, const char *value)
+static void add_mimetype(const char *name, const char *value)
+{
+	struct string_list_item *item;
+
+	item = string_list_insert(&ctx.cfg.mimetypes, name);
+	item->util = xstrdup(value);
+}
+
+static void apply_config(const char *name, const char *value)
 {
 	const char *arg;
 
@@ -282,7 +597,7 @@ static void config_cb(const char *name, const char *value)
 		ctx.cfg.max_patch_count = atoi(value);
 	else if (!strcmp(name, "project-list"))
 		ctx.cfg.project_list = cgit_strdup_first_line(cgit_expand_macros(value));
-	else if (!strcmp(name, "scan-path"))
+	else if (!strcmp(name, "scan-path")) {
 		if (ctx.cfg.cache_size)
 			process_cached_repolist(cgit_expand_macros(value));
 		else if (ctx.cfg.project_list)
@@ -290,7 +605,7 @@ static void config_cb(const char *name, const char *value)
 				      ctx.cfg.project_list);
 		else
 			scan_tree(cgit_expand_macros(value));
-	else if (!strcmp(name, "scan-hidden-path"))
+	} else if (!strcmp(name, "scan-hidden-path"))
 		ctx.cfg.scan_hidden_path = atoi(value);
 	else if (!strcmp(name, "section-from-path"))
 		ctx.cfg.section_from_path = atoi(value);
@@ -339,10 +654,27 @@ static void config_cb(const char *name, const char *value)
 	} else if (skip_prefix(name, "mimetype.", &arg))
 		add_mimetype(arg, value);
 	else if (!strcmp(name, "include"))
-		parse_configfile(cgit_expand_macros(value), config_cb);
+		config_file_parse(cgit_expand_macros(value), apply_config);
+}
+
+/*
+ * Read a whole number a request supplied, clamped into the range the caller
+ * accepts. strtol rather than atoi, because atoi has no defined behaviour once
+ * the digits overflow and every value here arrives straight from the query
+ * string.
+ */
+static int query_int(const char *value, int min, int max)
+{
+	long number = strtol(value, NULL, 10);
+
+	if (number < min)
+		return min;
+	if (number > max)
+		return max;
+	return number;
 }
 
-static void querystring_cb(const char *name, const char *value)
+static void apply_query_param(const char *name, const char *value)
 {
 	if (!value)
 		value = "";
@@ -353,19 +685,19 @@ static void querystring_cb(const char *name, const char *value)
 	} else if (!strcmp(name, "p")) {
 		ctx.qry.page = xstrdup(value);
 	} else if (!strcmp(name, "url")) {
-		if (*value == '/')
+		// Every leading slash goes, not just one. What is left is
+		// joined onto the virtual root, so a value like //example.com
+		// would otherwise survive as /example.com and make that join a
+		// scheme-relative link to another host.
+		while (*value == '/')
 			value++;
 		ctx.qry.url = xstrdup(value);
 		cgit_parse_url(value);
 	} else if (!strcmp(name, "qt")) {
 		ctx.qry.grep = xstrdup(value);
 	} else if (!strcmp(name, "q")) {
-		/* A query is matched against every repository, ref or commit
-		 * the page lists, so bound what one request can ask to be
-		 * compared. Nothing legible reaches this length, and the value
-		 * also lands in the cache key. */
-		if (strlen(value) > CGIT_MAX_SEARCH_LEN)
-			ctx.qry.search = xstrndup(value, CGIT_MAX_SEARCH_LEN);
+		if (strlen(value) > MAX_SEARCH_LEN)
+			ctx.qry.search = xstrndup(value, MAX_SEARCH_LEN);
 		else
 			ctx.qry.search = xstrdup(value);
 	} else if (!strcmp(name, "h")) {
@@ -378,165 +710,178 @@ static void querystring_cb(const char *name, const char *value)
 		ctx.qry.oid2 = xstrdup(value);
 		ctx.qry.has_oid = 1;
 	} else if (!strcmp(name, "ofs")) {
-		/* Bound the upper end so a crafted value cannot force a walk
-		 * over the whole history (and strtol avoids the atoi overflow).
-		 * Only clamp the upper end: ofs is overloaded, the stats page
-		 * submits -1 for "all authors", and the log skip loop already
-		 * floors negatives at zero. */
-		long ofs = strtol(value, NULL, 10);
-		if (ofs > 100000)
-			ofs = 100000;
-		else if (ofs < -1)
-			ofs = -1;
-		ctx.qry.ofs = ofs;
+		// Bounded above so a crafted value cannot force a walk over the
+		// whole history. Negatives stop at -1 rather than at zero,
+		// because the offset is overloaded, the stats page submits -1
+		// for all authors, and the log skip loop floors a negative
+		// itself.
+		ctx.qry.ofs = query_int(value, -1, MAX_QUERY_OFFSET);
 	} else if (!strcmp(name, "path")) {
 		ctx.qry.path = cgit_trim_end(value, '/');
 	} else if (!strcmp(name, "s")) {
 		ctx.qry.sort = xstrdup(value);
 	} else if (!strcmp(name, "showmsg")) {
-		ctx.qry.showmsg = atoi(value);
+		ctx.qry.showmsg = query_int(value, INT_MIN, INT_MAX);
 	} else if (!strcmp(name, "period")) {
 		ctx.qry.period = xstrdup(value);
 	} else if (!strcmp(name, "dt")) {
-		ctx.qry.difftype = atoi(value);
+		ctx.qry.difftype = query_int(value, INT_MIN, INT_MAX);
 		ctx.qry.has_difftype = 1;
 	} else if (!strcmp(name, "ss")) {
-		/* No longer generated, but there may be links out there. */
-		ctx.qry.difftype = atoi(value) ? DIFF_SSDIFF : DIFF_UNIFIED;
+		// No longer generated, but old links still carry it.
+		ctx.qry.difftype = query_int(value, INT_MIN, INT_MAX) ?
+			DIFF_SSDIFF : DIFF_UNIFIED;
 		ctx.qry.has_difftype = 1;
 	} else if (!strcmp(name, "all")) {
-		ctx.qry.show_all = atoi(value);
+		ctx.qry.show_all = query_int(value, INT_MIN, INT_MAX);
 	} else if (!strcmp(name, "context")) {
-		ctx.qry.context = atoi(value);
+		// Context lines are not counted against max-diff-lines, so an
+		// unbounded width turns a whole blob into context and renders
+		// it in full however small the change was.
+		ctx.qry.context = query_int(value, 0, MAX_DIFF_CONTEXT_LINES);
 	} else if (!strcmp(name, "ignorews")) {
-		ctx.qry.ignorews = atoi(value);
+		ctx.qry.ignorews = query_int(value, INT_MIN, INT_MAX);
 	} else if (!strcmp(name, "follow")) {
-		ctx.qry.follow = atoi(value);
+		ctx.qry.follow = query_int(value, INT_MIN, INT_MAX);
 	}
 }
 
-static void prepare_context(void)
+static void open_auth_filter(const char *action)
 {
-	memset(&ctx, 0, sizeof(ctx));
-	ctx.cfg.agefile = "info/web/last-modified";
-	ctx.cfg.cache_size = 0;
-	ctx.cfg.cache_root = CGIT_CACHE_ROOT;
-	ctx.cfg.cache_about_ttl = 15;
-	ctx.cfg.cache_snapshot_ttl = 5;
-	ctx.cfg.cache_repo_ttl = 5;
-	ctx.cfg.cache_root_ttl = 5;
-	ctx.cfg.cache_scanrc_ttl = 15;
-	ctx.cfg.cache_dynamic_ttl = 5;
-	ctx.cfg.cache_static_ttl = -1;
-	ctx.cfg.case_sensitive_sort = 1;
-	ctx.cfg.branch_sort = 0;
-	ctx.cfg.commit_sort = 0;
-	ctx.cfg.logo = "/cgit.png";
-	ctx.cfg.favicon = "/favicon.ico";
-	ctx.cfg.local_time = 0;
-	ctx.cfg.date_mode = date_mode_from_type(DATE_SHORT);
-	ctx.cfg.enable_relative_dates = 1;
-	ctx.cfg.enable_http_clone = 1;
-	ctx.cfg.enable_index_owner = 1;
-	ctx.cfg.enable_tree_linenumbers = 1;
-	ctx.cfg.enable_git_config = 0;
-	ctx.cfg.max_repo_count = 50;
-	ctx.cfg.max_commit_count = 50;
-	ctx.cfg.max_patch_count = 50;
-	ctx.cfg.max_diff_files = 200;  /* larger commits render stat only */
-	ctx.cfg.max_diff_lines = 1000; /* larger file diffs link out */
-	ctx.cfg.max_msg_len = 80;
-	ctx.cfg.max_ref_count = 200;   /* refs beyond this paginate */
-	ctx.cfg.max_repodesc_len = 80;
-	ctx.cfg.max_blob_size = 10 * 1024; /* 10 MB; bounds per-request memory */
-	ctx.cfg.max_stats = 0;
-	ctx.cfg.project_list = NULL;
-	ctx.cfg.renamelimit = -1;
-	ctx.cfg.remove_suffix = 0;
-	ctx.cfg.robots = "index, nofollow";
-	ctx.cfg.root_title = "Git repository browser";
-	ctx.cfg.root_desc = "a fast webinterface for the git dscm";
-	ctx.cfg.scan_hidden_path = 0;
-	ctx.cfg.script_name = CGIT_SCRIPT_NAME;
-	ctx.cfg.section = "";
-	ctx.cfg.repository_sort = "name";
-	ctx.cfg.section_sort = 0;
-	ctx.cfg.summary_branches = 10;
-	ctx.cfg.summary_log = 10;
-	ctx.cfg.summary_tags = 10;
-	ctx.cfg.max_atom_items = 10;
-	ctx.cfg.difftype = DIFF_UNIFIED;
-	ctx.env.cgit_config = getenv("CGIT_CONFIG");
-	ctx.env.http_host = getenv("HTTP_HOST");
-	ctx.env.https = getenv("HTTPS");
-	ctx.env.no_http = getenv("NO_HTTP");
-	ctx.env.path_info = getenv("PATH_INFO");
-	ctx.env.query_string = getenv("QUERY_STRING");
-	ctx.env.request_method = getenv("REQUEST_METHOD");
-	ctx.env.script_name = getenv("SCRIPT_NAME");
-	ctx.env.server_name = getenv("SERVER_NAME");
-	ctx.env.server_port = getenv("SERVER_PORT");
-	ctx.env.http_cookie = getenv("HTTP_COOKIE");
-	ctx.env.http_referer = getenv("HTTP_REFERER");
-	ctx.env.content_length = getenv("CONTENT_LENGTH") ? strtoul(getenv("CONTENT_LENGTH"), NULL, 10) : 0;
-	ctx.env.authenticated = 0;
-	ctx.page.mimetype = "text/html";
-	ctx.page.charset = PAGE_ENCODING;
-	ctx.page.filename = NULL;
-	ctx.page.size = 0;
-	ctx.page.modified = time(NULL);
-	ctx.page.expires = ctx.page.modified;
-	ctx.page.etag = NULL;
-	string_list_init_dup(&ctx.cfg.mimetypes);
-	if (ctx.env.script_name)
-		ctx.cfg.script_name = xstrdup(ctx.env.script_name);
-	if (ctx.env.query_string)
-		ctx.qry.raw = xstrdup(ctx.env.query_string);
-	if (!ctx.env.cgit_config)
-		ctx.env.cgit_config = CGIT_CONFIG;
+	cgit_open_filter(ctx.cfg.auth_filter, action,
+		ctx.env.http_cookie ? ctx.env.http_cookie : "",
+		ctx.env.request_method ? ctx.env.request_method : "",
+		ctx.env.query_string ? ctx.env.query_string : "",
+		ctx.env.http_referer ? ctx.env.http_referer : "",
+		ctx.env.path_info ? ctx.env.path_info : "",
+		ctx.env.http_host ? ctx.env.http_host : "",
+		ctx.env.https ? ctx.env.https : "",
+		ctx.qry.repo ? ctx.qry.repo : "",
+		ctx.qry.page ? ctx.qry.page : "",
+		cgit_currentfullurl(),
+		cgit_loginurl());
 }
 
-struct refmatch {
-	char *req_ref;
-	char *first_ref;
-	int match;
-};
+/*
+ * The filter answers the login POST itself, writing the status line and every
+ * header, so nothing here prints any and the process ends before this returns.
+ */
+static void authenticate_post(void)
+{
+	char buffer[MAX_AUTHENTICATION_POST_BYTES];
+	size_t len;
+	ssize_t got;
+
+	open_auth_filter("authenticate-post");
+	len = ctx.env.content_length;
+	if (len > MAX_AUTHENTICATION_POST_BYTES)
+		len = MAX_AUTHENTICATION_POST_BYTES;
+	if ((got = read(STDIN_FILENO, buffer, len)) < 0)
+		die_errno("Could not read POST from stdin");
+	if (write(STDOUT_FILENO, buffer, got) < 0)
+		die_errno("Could not write POST to stdout");
+	cgit_close_filter(ctx.cfg.auth_filter);
+	exit(0);
+}
+
+static void authenticate_cookie(void)
+{
+	if (!ctx.cfg.auth_filter) {
+		ctx.env.authenticated = 1;
+		return;
+	}
+
+	if (ctx.env.request_method && ctx.qry.page && !ctx.repo &&
+	    !strcmp(ctx.env.request_method, "POST") &&
+	    !strcmp(ctx.qry.page, "login")) {
+		authenticate_post();
+		return;
+	}
+
+	open_auth_filter("authenticate-cookie");
+	ctx.env.authenticated = cgit_close_filter(ctx.cfg.auth_filter);
+}
+
+// Every cache-*-ttl setting is written in minutes, and so is this.
+static int calc_ttl(void)
+{
+	if (!ctx.repo)
+		return ctx.cfg.cache_root_ttl;
+
+	if (!ctx.qry.page)
+		return ctx.cfg.cache_repo_ttl;
+
+	if (!strcmp(ctx.qry.page, "about"))
+		return ctx.cfg.cache_about_ttl;
+
+	if (!strcmp(ctx.qry.page, "snapshot"))
+		return ctx.cfg.cache_snapshot_ttl;
+
+	if (ctx.qry.has_oid)
+		return ctx.cfg.cache_static_ttl;
+
+	if (ctx.qry.has_symref)
+		return ctx.cfg.cache_dynamic_ttl;
+
+	return ctx.cfg.cache_repo_ttl;
+}
 
-static int find_current_ref(const struct reference *ref, void *cb_data)
+/*
+ * The scheme and the host are folded in because the absolute urls a page
+ * carries, its clone urls and atom links, are built from them, so a request
+ * arriving with a spoofed Host must not poison the page served to a visitor
+ * who came in on the real one.
+ */
+static void build_cache_key(struct strbuf *key)
 {
-	struct refmatch *info;
+	char *hosturl = cgit_hosturl();
+	const char *parts[] = {
+		cgit_httpscheme(),
+		hosturl ? hosturl : "",
+		ctx.env.path_info ? ctx.env.path_info : "",
+		ctx.env.query_string ? ctx.env.query_string : "",
+	};
+	size_t i;
 
-	info = (struct refmatch *)cb_data;
-	if (!strcmp(ref->name, info->req_ref))
-		info->match = 1;
-	if (!info->first_ref)
-		info->first_ref = xstrdup(ref->name);
-	return info->match;
+	// Each part is written behind its own length, so nothing a value
+	// contains can make two different requests spell one key. The path and
+	// the query come from the environment rather than the query string cgit
+	// rebuilds, since that rebuild folds the two together and would let the
+	// PATH_INFO and QUERY_STRING forms of one request share a slot.
+	for (i = 0; i < ARRAY_SIZE(parts); i++)
+		strbuf_addf(key, "%zu|%s", strlen(parts[i]), parts[i]);
+	free(hosturl);
 }
 
-static void free_refmatch_inner(struct refmatch *info)
+// Returning non-zero ends git's walk, so the search stops at the first hit.
+static int find_current_ref(const struct reference *ref, void *data)
 {
-	if (info->first_ref)
-		free(info->first_ref);
+	struct refmatch *match = data;
+
+	if (!strcmp(ref->name, match->wanted))
+		match->found = 1;
+	if (!match->first)
+		match->first = xstrdup(ref->name);
+	return match->found;
 }
 
 static char *find_default_branch(struct cgit_repo *repo)
 {
-	struct refmatch info;
+	struct refmatch match;
 	char *ref;
 
-	info.req_ref = repo->defbranch;
-	info.first_ref = NULL;
-	info.match = 0;
+	match.wanted = repo->defbranch;
+	match.first = NULL;
+	match.found = 0;
 	refs_for_each_branch_ref(get_main_ref_store(the_repository),
-				 find_current_ref, &info);
-	if (info.match)
-		ref = info.req_ref;
+				 find_current_ref, &match);
+	if (match.found)
+		ref = match.wanted;
 	else
-		ref = info.first_ref;
+		ref = match.first;
 	if (ref)
 		ref = xstrdup(ref);
-	free_refmatch_inner(&info);
+	free(match.first);
 
 	return ref;
 }
@@ -553,8 +898,12 @@ static char *guess_defbranch(void)
 	return xstrdup(refname);
 }
 
-/* The caller must free filename and ref after calling this. */
-static inline void parse_readme(const char *readme, char **filename, char **ref, struct cgit_repo *repo)
+/*
+ * Split one readme setting into the file it names and the ref that file is
+ * read from, leaving the ref NULL for a file on disk. The caller frees both.
+ */
+static void parse_readme(const char *readme, char **filename, char **ref,
+			 struct cgit_repo *repo)
 {
 	const char *colon;
 
@@ -564,11 +913,10 @@ static inline void parse_readme(const char *readme, char **filename, char **ref,
 	if (!readme || !readme[0])
 		return;
 
-	/* Check if the readme is tracked in the git repo. */
+	// A colon separates a ref from a path, so a setting carrying one names
+	// a file tracked in the repository rather than one on disk.
 	colon = strchr(readme, ':');
 	if (colon && strlen(colon) > 1) {
-		/* If it starts with a colon, we want to use head given
-		 * from query or the default branch */
 		if (colon == readme && ctx.qry.head)
 			*ref = xstrdup(ctx.qry.head);
 		else if (colon == readme && repo->defbranch)
@@ -578,12 +926,12 @@ static inline void parse_readme(const char *readme, char **filename, char **ref,
 		readme = colon + 1;
 	}
 
-	/* Prepend repo path to relative readme path unless tracked. */
 	if (!(*ref) && readme[0] != '/')
 		*filename = cgit_fmtalloc("%s/%s", repo->path, readme);
 	else
 		*filename = xstrdup(readme);
 }
+
 static void choose_readme(struct cgit_repo *repo)
 {
 	int found;
@@ -623,24 +971,26 @@ static void choose_readme(struct cgit_repo *repo)
 
 static void prepare_repo_env(int *nongit)
 {
-	/* The path to the git repository. */
 	setenv("GIT_DIR", ctx.repo->path, 1);
 
-	/* Setup the git directory and initialize the notes system. Both of these
-	 * load local configuration from the git repository, so we do them both while
-	 * the HOME variables are unset. */
+	// Both read configuration out of the repository, with the user's own
+	// git configuration already stripped by isolate_git_environment.
 	setup_git_directory_gently(the_repository, nongit);
 	load_display_notes(NULL);
 }
 
+/*
+ * Returns non-zero once it has written a complete response of its own, in
+ * which case the caller must not render a page over the top of it.
+ */
 static int prepare_repo_cmd(int nongit)
 {
 	struct object_id oid;
-	int rc;
+	int err;
 
 	if (nongit) {
 		const char *name = ctx.repo->name;
-		rc = errno;
+		err = errno;
 		ctx.page.title = cgit_fmtalloc("%s - %s", ctx.cfg.root_title,
 						"config error");
 		ctx.repo = NULL;
@@ -648,7 +998,7 @@ static int prepare_repo_cmd(int nongit)
 		cgit_print_docstart();
 		cgit_print_pageheader();
 		cgit_print_error("Failed to open %s: %s", name,
-				 rc ? strerror(rc) : "Not a valid git repository");
+				 err ? strerror(err) : "Not a valid git repository");
 		cgit_print_docend();
 		return 1;
 	}
@@ -690,71 +1040,13 @@ static int prepare_repo_cmd(int nongit)
 	return 0;
 }
 
-static inline void open_auth_filter(const char *function)
-{
-	cgit_open_filter(ctx.cfg.auth_filter, function,
-		ctx.env.http_cookie ? ctx.env.http_cookie : "",
-		ctx.env.request_method ? ctx.env.request_method : "",
-		ctx.env.query_string ? ctx.env.query_string : "",
-		ctx.env.http_referer ? ctx.env.http_referer : "",
-		ctx.env.path_info ? ctx.env.path_info : "",
-		ctx.env.http_host ? ctx.env.http_host : "",
-		ctx.env.https ? ctx.env.https : "",
-		ctx.qry.repo ? ctx.qry.repo : "",
-		ctx.qry.page ? ctx.qry.page : "",
-		cgit_currentfullurl(),
-		cgit_loginurl());
-}
-
-/* The filter is expected to spit out "Status: " and all headers. */
-static inline void authenticate_post(void)
-{
-	char buffer[MAX_AUTHENTICATION_POST_BYTES];
-	size_t len;
-	ssize_t got;
-
-	open_auth_filter("authenticate-post");
-	len = ctx.env.content_length;
-	if (len > MAX_AUTHENTICATION_POST_BYTES)
-		len = MAX_AUTHENTICATION_POST_BYTES;
-	if ((got = read(STDIN_FILENO, buffer, len)) < 0)
-		die_errno("Could not read POST from stdin");
-	if (write(STDOUT_FILENO, buffer, got) < 0)
-		die_errno("Could not write POST to stdout");
-	cgit_close_filter(ctx.cfg.auth_filter);
-	exit(0);
-}
-
-static inline void authenticate_cookie(void)
-{
-	/* If we don't have an auth_filter, consider all cookies valid, and thus return early. */
-	if (!ctx.cfg.auth_filter) {
-		ctx.env.authenticated = 1;
-		return;
-	}
-
-	/* If we're having something POST'd to /login, we're authenticating POST,
-	 * instead of the cookie, so call authenticate_post and bail out early.
-	 * This pattern here should match /?p=login with POST. */
-	if (ctx.env.request_method && ctx.qry.page && !ctx.repo && \
-	    !strcmp(ctx.env.request_method, "POST") && !strcmp(ctx.qry.page, "login")) {
-		authenticate_post();
-		return;
-	}
-
-	/* If we've made it this far, we're authenticating the cookie for real, so do that. */
-	open_auth_filter("authenticate-cookie");
-	ctx.env.authenticated = cgit_close_filter(ctx.cfg.auth_filter);
-}
-
 static void process_request(void)
 {
-	struct cgit_cmd *cmd;
+	const struct cgit_cmd *cmd;
 	int nongit = 0;
 
-	/* If we're not yet authenticated, no matter what page we're on,
-	 * display the authentication body from the auth_filter. This should
-	 * never be cached. */
+	// An unauthenticated request is answered with the filter's own body
+	// whatever page it asked for.
 	if (!ctx.env.authenticated) {
 		ctx.page.title = "Authentication Required";
 		cgit_print_http_headers();
@@ -788,10 +1080,6 @@ static void process_request(void)
 		return;
 	}
 
-	/* If cmd->want_vpath is set, assume ctx.qry.path contains a "virtual"
-	 * in-project path limit to be made available at ctx.qry.vpath.
-	 * Otherwise, no path limit is in effect (ctx.qry.vpath = NULL).
-	 */
 	ctx.qry.vpath = cmd->want_vpath ? ctx.qry.path : NULL;
 
 	if (ctx.repo && prepare_repo_cmd(nongit))
@@ -800,294 +1088,94 @@ static void process_request(void)
 	cmd->fn();
 }
 
-static int cmp_repos(const void *a, const void *b)
-{
-	const struct cgit_repo *ra = a, *rb = b;
-	return strcmp(ra->url, rb->url);
-}
-
-static char *build_snapshot_setting(int bitmap)
-{
-	const struct cgit_snapshot_format *f;
-	struct strbuf result = STRBUF_INIT;
-
-	for (f = cgit_snapshot_formats; f->suffix; f++) {
-		if (cgit_snapshot_format_bit(f) & bitmap) {
-			if (result.len)
-				strbuf_addch(&result, ' ');
-			strbuf_addstr(&result, f->suffix);
-		}
-	}
-	return strbuf_detach(&result, NULL);
-}
-
-static void print_repo(FILE *f, struct cgit_repo *repo)
+void cgit_repo_config(struct cgit_repo *repo, const char *name, const char *value)
 {
+	const char *path;
 	struct string_list_item *item;
-	fprintf(f, "repo.url=%s\n", repo->url);
-	fprintf(f, "repo.name=%s\n", repo->name);
-	fprintf(f, "repo.path=%s\n", repo->path);
-	if (repo->owner)
-		fprintf(f, "repo.owner=%s\n", repo->owner);
-	if (repo->desc)
-		fprintf(f, "repo.desc=%s\n", repo->desc);
-	for_each_string_list_item(item, &repo->readme) {
-		if (item->util)
-			fprintf(f, "repo.readme=%s:%s\n", (char *)item->util, item->string);
-		else
-			fprintf(f, "repo.readme=%s\n", item->string);
-	}
-	if (repo->defbranch)
-		fprintf(f, "repo.defbranch=%s\n", repo->defbranch);
-	if (repo->extra_head_content)
-		fprintf(f, "repo.extra-head-content=%s\n", repo->extra_head_content);
-	if (repo->module_link)
-		fprintf(f, "repo.module-link=%s\n", repo->module_link);
-	if (repo->section)
-		fprintf(f, "repo.section=%s\n", repo->section);
-	if (repo->homepage)
-		fprintf(f, "repo.homepage=%s\n", repo->homepage);
-	if (repo->clone_url)
-		fprintf(f, "repo.clone-url=%s\n", repo->clone_url);
-	fprintf(f, "repo.enable-blame=%d\n",
-	        repo->enable_blame);
-	fprintf(f, "repo.enable-commit-graph=%d\n",
-	        repo->enable_commit_graph);
-	fprintf(f, "repo.enable-follow-links=%d\n",
-		repo->enable_follow_links);
-	fprintf(f, "repo.enable-log-filecount=%d\n",
-	        repo->enable_log_filecount);
-	fprintf(f, "repo.enable-log-linecount=%d\n",
-	        repo->enable_log_linecount);
-	if (repo->about_filter && repo->about_filter != ctx.cfg.about_filter)
-		cgit_fprintf_filter(repo->about_filter, f, "repo.about-filter=");
-	if (repo->commit_filter && repo->commit_filter != ctx.cfg.commit_filter)
-		cgit_fprintf_filter(repo->commit_filter, f, "repo.commit-filter=");
-	if (repo->source_filter && repo->source_filter != ctx.cfg.source_filter)
-		cgit_fprintf_filter(repo->source_filter, f, "repo.source-filter=");
-	if (repo->email_filter && repo->email_filter != ctx.cfg.email_filter)
-		cgit_fprintf_filter(repo->email_filter, f, "repo.email-filter=");
-	if (repo->snapshots != ctx.cfg.snapshots) {
-		char *tmp = build_snapshot_setting(repo->snapshots);
-		fprintf(f, "repo.snapshots=%s\n", tmp ? tmp : "");
-		free(tmp);
-	}
-	if (repo->snapshot_prefix)
-		fprintf(f, "repo.snapshot-prefix=%s\n", repo->snapshot_prefix);
-	if (repo->enable_stats != ctx.cfg.enable_stats)
-		fprintf(f, "repo.enable-stats=%d\n", repo->enable_stats);
-	if (repo->max_stats != ctx.cfg.max_stats)
-		fprintf(f, "repo.max-stats=%s\n",
-		        cgit_find_stats_periodname(repo->max_stats));
-	if (repo->logo)
-		fprintf(f, "repo.logo=%s\n", repo->logo);
-	if (repo->logo_link)
-		fprintf(f, "repo.logo-link=%s\n", repo->logo_link);
-	fprintf(f, "repo.enable-remote-branches=%d\n", repo->enable_remote_branches);
-	fprintf(f, "repo.enable-subject-links=%d\n", repo->enable_subject_links);
-	fprintf(f, "repo.enable-html-serving=%d\n", repo->enable_html_serving);
-	if (repo->branch_sort == 1)
-		fprintf(f, "repo.branch-sort=age\n");
-	if (repo->commit_sort) {
-		if (repo->commit_sort == 1)
-			fprintf(f, "repo.commit-sort=date\n");
-		else if (repo->commit_sort == 2)
-			fprintf(f, "repo.commit-sort=topo\n");
-	}
-	fprintf(f, "repo.hide=%d\n", repo->hide);
-	fprintf(f, "repo.ignore=%d\n", repo->ignore);
-	fprintf(f, "\n");
-}
-
-static void print_repolist(FILE *f, struct cgit_repolist *list, int start)
-{
-	int i;
-
-	for (i = start; i < list->count; i++)
-		print_repo(f, &list->repos[i]);
-}
-
-/* Scan 'path' for git repositories, save the resulting repolist in 'cached_rc'
- * and return 0 on success.
- */
-static int generate_cached_repolist(const char *path, const char *cached_rc)
-{
-	struct strbuf locked_rc = STRBUF_INIT;
-	int result = 0;
-	int idx;
-	FILE *f;
-
-	strbuf_addf(&locked_rc, "%s.lock", cached_rc);
-	f = fopen(locked_rc.buf, "wx");
-	if (!f) {
-		/* Inform about the error unless the lockfile already existed,
-		 * since that only means we've got concurrent requests.
-		 */
-		result = errno;
-		if (result != EEXIST)
-			fprintf(stderr, "[cgit] Error opening %s: %s (%d)\n",
-				locked_rc.buf, strerror(result), result);
-		goto out;
-	}
-	idx = cgit_repolist.count;
-	if (ctx.cfg.project_list)
-		scan_projects(path, ctx.cfg.project_list);
-	else
-		scan_tree(path);
-	print_repolist(f, &cgit_repolist, idx);
-	if (rename(locked_rc.buf, cached_rc))
-		fprintf(stderr, "[cgit] Error renaming %s to %s: %s (%d)\n",
-			locked_rc.buf, cached_rc, strerror(errno), errno);
-	fclose(f);
-out:
-	strbuf_release(&locked_rc);
-	return result;
-}
-
-static void process_cached_repolist(const char *path)
-{
-	struct stat st;
-	struct strbuf cached_rc = STRBUF_INIT;
-	time_t age;
-	unsigned long hash;
-
-	hash = cache_hash_str(path);
-	if (ctx.cfg.project_list)
-		hash += cache_hash_str(ctx.cfg.project_list);
-	strbuf_addf(&cached_rc, "%s/rc-%8lx", ctx.cfg.cache_root, hash);
-
-	if (stat(cached_rc.buf, &st)) {
-		/* Nothing is cached, we need to scan without forking. And
-		 * if we fail to generate a cached repolist, we need to
-		 * invoke scan_tree manually.
-		 */
-		if (generate_cached_repolist(path, cached_rc.buf)) {
-			if (ctx.cfg.project_list)
-				scan_projects(path, ctx.cfg.project_list);
-			else
-				scan_tree(path);
-		}
-		goto out;
-	}
-
-	parse_configfile(cached_rc.buf, config_cb);
-
-	/* If the cached configfile hasn't expired, lets exit now */
-	age = time(NULL) - st.st_mtime;
-	if (age <= (ctx.cfg.cache_scanrc_ttl * 60))
-		goto out;
-
-	/* The cached repolist has been parsed, but it was old. So lets
-	 * rescan the specified path and generate a new cached repolist
-	 * in a child-process to avoid latency for the current request.
-	 */
-	if (fork())
-		goto out;
-
-	exit(generate_cached_repolist(path, cached_rc.buf));
-out:
-	strbuf_release(&cached_rc);
-}
-
-static void cgit_parse_args(int argc, const char **argv)
-{
-	int i;
-	const char *arg;
-	int scan = 0;
-
-	for (i = 1; i < argc; i++) {
-		if (!strcmp(argv[i], "--version")) {
-			printf("CGit %s | https://github.com/brycekwon/cgit\n\nCompiled in features:\n", CGIT_VERSION);
-#ifdef NO_LUA
-			printf("[-] ");
-#else
-			printf("[+] ");
-#endif
-			printf("Lua scripting\n");
-#ifndef HAVE_LINUX_SENDFILE
-			printf("[-] ");
-#else
-			printf("[+] ");
-#endif
-			printf("Linux sendfile() usage\n");
 
-			exit(0);
-		}
-		if (skip_prefix(argv[i], "--cache=", &arg)) {
-			ctx.cfg.cache_root = xstrdup(arg);
-		} else if (!strcmp(argv[i], "--nohttp")) {
-			ctx.env.no_http = "1";
-		} else if (skip_prefix(argv[i], "--query=", &arg)) {
-			ctx.qry.raw = xstrdup(arg);
-		} else if (skip_prefix(argv[i], "--repo=", &arg)) {
-			ctx.qry.repo = xstrdup(arg);
-		} else if (skip_prefix(argv[i], "--page=", &arg)) {
-			ctx.qry.page = xstrdup(arg);
-		} else if (skip_prefix(argv[i], "--head=", &arg)) {
-			ctx.qry.head = xstrdup(arg);
-			ctx.qry.has_symref = 1;
-		} else if (skip_prefix(argv[i], "--oid=", &arg)) {
-			ctx.qry.oid = xstrdup(arg);
-			ctx.qry.has_oid = 1;
-		} else if (skip_prefix(argv[i], "--ofs=", &arg)) {
-			ctx.qry.ofs = atoi(arg);
-		} else if (skip_prefix(argv[i], "--scan-tree=", &arg) ||
-		           skip_prefix(argv[i], "--scan-path=", &arg)) {
-			/*
-			 * HACK: The global snapshot bit mask defines the set
-			 * of allowed snapshot formats, but the config file
-			 * hasn't been parsed yet so the mask is currently 0.
-			 * By setting all bits high before scanning we make
-			 * sure that any in-repo cgitrc snapshot setting is
-			 * respected by scan_tree().
-			 *
-			 * NOTE: We assume that there aren't more than 8
-			 * different snapshot formats supported by cgit...
-			 */
-			ctx.cfg.snapshots = 0xFF;
-			scan++;
-			scan_tree(arg);
-		}
-	}
-	if (scan) {
-		qsort(cgit_repolist.repos, cgit_repolist.count,
-			sizeof(struct cgit_repo), cmp_repos);
-		print_repolist(stdout, &cgit_repolist, 0);
-		exit(0);
+	if (!strcmp(name, "name"))
+		repo->name = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "clone-url"))
+		repo->clone_url = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "desc"))
+		repo->desc = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "owner"))
+		repo->owner = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "homepage"))
+		repo->homepage = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "defbranch"))
+		repo->defbranch = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "extra-head-content"))
+		repo->extra_head_content = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "snapshots"))
+		repo->snapshots = ctx.cfg.snapshots & cgit_parse_snapshots_mask(value);
+	else if (!strcmp(name, "enable-blame"))
+		repo->enable_blame = atoi(value);
+	else if (!strcmp(name, "enable-commit-graph"))
+		repo->enable_commit_graph = atoi(value);
+	else if (!strcmp(name, "enable-follow-links"))
+		repo->enable_follow_links = atoi(value);
+	else if (!strcmp(name, "enable-log-filecount"))
+		repo->enable_log_filecount = atoi(value);
+	else if (!strcmp(name, "enable-log-linecount"))
+		repo->enable_log_linecount = atoi(value);
+	else if (!strcmp(name, "enable-remote-branches"))
+		repo->enable_remote_branches = atoi(value);
+	else if (!strcmp(name, "enable-subject-links"))
+		repo->enable_subject_links = atoi(value);
+	else if (!strcmp(name, "enable-html-serving"))
+		repo->enable_html_serving = atoi(value);
+	else if (!strcmp(name, "enable-stats"))
+		repo->enable_stats = atoi(value);
+	else if (!strcmp(name, "branch-sort")) {
+		if (!strcmp(value, "age"))
+			repo->branch_sort = 1;
+		if (!strcmp(value, "name"))
+			repo->branch_sort = 0;
+	} else if (!strcmp(name, "commit-sort")) {
+		if (!strcmp(value, "date"))
+			repo->commit_sort = 1;
+		if (!strcmp(value, "topo"))
+			repo->commit_sort = 2;
+	} else if (!strcmp(name, "max-stats"))
+		repo->max_stats = cgit_find_stats_period(value, NULL);
+	else if (!strcmp(name, "module-link"))
+		repo->module_link = cgit_strdup_first_line(value);
+	else if (skip_prefix(name, "module-link.", &path)) {
+		item = string_list_append(&repo->submodules,
+					  cgit_strdup_first_line(path));
+		item->util = cgit_strdup_first_line(value);
+	} else if (!strcmp(name, "section"))
+		repo->section = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "snapshot-prefix"))
+		repo->snapshot_prefix = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "readme") && value != NULL) {
+		if (repo->readme.items == ctx.cfg.readme.items)
+			memset(&repo->readme, 0, sizeof(repo->readme));
+		string_list_append(&repo->readme, cgit_strdup_first_line(value));
+	} else if (!strcmp(name, "logo") && value != NULL)
+		repo->logo = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "logo-link") && value != NULL)
+		repo->logo_link = cgit_strdup_first_line(value);
+	else if (!strcmp(name, "hide"))
+		repo->hide = atoi(value);
+	else if (!strcmp(name, "ignore"))
+		repo->ignore = atoi(value);
+	else if (ctx.cfg.enable_filter_overrides) {
+		if (!strcmp(name, "about-filter"))
+			repo->about_filter = cgit_new_filter(value, ABOUT);
+		else if (!strcmp(name, "commit-filter"))
+			repo->commit_filter = cgit_new_filter(value, COMMIT);
+		else if (!strcmp(name, "source-filter"))
+			repo->source_filter = cgit_new_filter(value, SOURCE);
+		else if (!strcmp(name, "email-filter"))
+			repo->email_filter = cgit_new_filter(value, EMAIL);
 	}
 }
 
-static int calc_ttl(void)
-{
-	if (!ctx.repo)
-		return ctx.cfg.cache_root_ttl;
-
-	if (!ctx.qry.page)
-		return ctx.cfg.cache_repo_ttl;
-
-	if (!strcmp(ctx.qry.page, "about"))
-		return ctx.cfg.cache_about_ttl;
-
-	if (!strcmp(ctx.qry.page, "snapshot"))
-		return ctx.cfg.cache_snapshot_ttl;
-
-	if (ctx.qry.has_oid)
-		return ctx.cfg.cache_static_ttl;
-
-	if (ctx.qry.has_symref)
-		return ctx.cfg.cache_dynamic_ttl;
-
-	return ctx.cfg.cache_repo_ttl;
-}
-
-static NORETURN void cgit_die_routine(const char *msg, va_list params)
-{
-	cgit_vprint_error_page(400, "Bad request", msg, params);
-	exit(0);
-}
-
 int cmd_main(int argc, const char **argv)
 {
+	struct strbuf cache_key = STRBUF_INIT;
 	const char *path;
 	int err, ttl;
 
@@ -1097,71 +1185,58 @@ int cmd_main(int argc, const char **argv)
 	// Registered second so it runs first, since exit is reached from error
 	// paths and from the HEAD shortcut with a page still buffered.
 	atexit(html_flush);
-	set_die_routine(cgit_die_routine);
+	set_die_routine(die_routine);
 
 	prepare_context();
 	cgit_repolist.length = 0;
 	cgit_repolist.count = 0;
 	cgit_repolist.repos = NULL;
 
-	cgit_parse_args(argc, argv);
-	parse_configfile(cgit_expand_macros(ctx.env.cgit_config), config_cb);
+	parse_args(argc, argv);
+	config_file_parse(cgit_expand_macros(ctx.env.cgit_config), apply_config);
 	ctx.repo = NULL;
-	http_parse_querystring(ctx.qry.raw, querystring_cb);
+	http_parse_querystring(ctx.qry.raw, apply_query_param);
 
-	/* If virtual-root isn't specified in cgitrc, lets pretend
-	 * that virtual-root equals SCRIPT_NAME, minus any possibly
-	 * trailing slashes.
-	 */
 	if (!ctx.cfg.virtual_root && ctx.cfg.script_name)
 		ctx.cfg.virtual_root = cgit_ensure_end(ctx.cfg.script_name, '/');
 
-	/* If no url parameter is specified on the querystring, lets
-	 * use PATH_INFO as url. This allows cgit to work with virtual
-	 * urls without the need for rewriterules in the webserver (as
-	 * long as PATH_INFO is included in the cache lookup key).
-	 */
+	// Falling back to PATH_INFO lets cgit serve virtual urls without a
+	// rewrite rule in the web server, and folding it into the raw query
+	// string keeps it part of the cache key.
 	path = ctx.env.path_info;
 	if (!ctx.qry.url && path) {
-		if (path[0] == '/')
+		// Stripped like the url parameter and for the same reason, so
+		// a request for //example.com cannot turn into a link off site.
+		while (*path == '/')
 			path++;
 		ctx.qry.url = xstrdup(path);
 		if (ctx.qry.raw) {
-			char *newqry = cgit_fmtalloc("%s?%s", path, ctx.qry.raw);
+			char *path_and_query = cgit_fmtalloc("%s?%s", path, ctx.qry.raw);
 			free(ctx.qry.raw);
-			ctx.qry.raw = newqry;
+			ctx.qry.raw = path_and_query;
 		} else
 			ctx.qry.raw = xstrdup(ctx.qry.url);
 		cgit_parse_url(ctx.qry.url);
 	}
 
-	/* Before we go any further, we set ctx.env.authenticated by checking to see
-	 * if the supplied cookie is valid. All cookies are valid if there is no
-	 * auth_filter. If there is an auth_filter, the filter decides. */
 	authenticate_cookie();
 
 	ttl = calc_ttl();
 	if (ttl < 0)
-		ctx.page.expires += 10 * 365 * 24 * 60 * 60; /* 10 years */
+		ctx.page.expires += NEVER_EXPIRES_SECONDS;
 	else
 		ctx.page.expires += ttl * 60;
-	if (!ctx.env.authenticated || (ctx.env.request_method && !strcmp(ctx.env.request_method, "HEAD")))
+	// An unauthenticated request gets a body meant for one visitor, and a
+	// HEAD request stops after the headers.
+	if (!ctx.env.authenticated ||
+	    (ctx.env.request_method && !strcmp(ctx.env.request_method, "HEAD")))
 		ctx.cfg.cache_size = 0;
-	/* Fold the scheme and host into the cache key. Absolute URLs in the
-	 * output (clone urls, atom links) are built from these, so a request
-	 * with a spoofed Host must not poison the cached page served to a
-	 * visitor arriving on a different host. */
-	{
-		struct strbuf cache_key = STRBUF_INIT;
-		char *hosturl = cgit_hosturl();
-		strbuf_addf(&cache_key, "%s%s|%s", cgit_httpscheme(),
-			    hosturl ? hosturl : "",
-			    ctx.qry.raw ? ctx.qry.raw : "");
-		free(hosturl);
-		err = cache_process(ctx.cfg.cache_size, ctx.cfg.cache_root,
-				    cache_key.buf, ttl, process_request);
-		strbuf_release(&cache_key);
-	}
+
+	build_cache_key(&cache_key);
+	err = cache_process(ctx.cfg.cache_size, ctx.cfg.cache_root,
+			    cache_key.buf, ttl, process_request);
+	strbuf_release(&cache_key);
+
 	cgit_cleanup_filters();
 	if (err)
 		cgit_print_error("Error processing page: %s (%d)",