index : cgit

/*
 * The helpers that no single page owns, along with the process-wide state
 * every page reads. Repositories are registered here as cgitrc is parsed, and
 * git's diff machinery is wrapped so a caller is handed whole filepairs or
 * whole lines instead of xdiff buffers.
 */

#define USE_THE_REPOSITORY_VARIABLE

#include "cgit.h"
#include "shared.h"

#define MACRO_EXPANSION_BUFSIZE (1024 * 8)

#define MAX_FIRST_LINE_READ (1024 * 64)

// Git writes this line into the description file of every repository it
// creates, so a repository still carrying it gets cgit's own default instead.
static const char *default_git_desc =
	"Unnamed repository; edit this file 'description' to name the repository.";

struct cgit_repolist cgit_repolist;
struct cgit_context ctx;
char cgit_default_repo_desc[] = "[no description]";

struct cgit_repo *cgit_add_repo(const char *url)
{
	struct cgit_repo *repo;

	ALLOC_GROW(cgit_repolist.repos, cgit_repolist.count + 1, cgit_repolist.length);
	repo = &cgit_repolist.repos[cgit_repolist.count++];
	memset(repo, 0, sizeof(struct cgit_repo));
	repo->url = cgit_trim_end(url, '/');
	if (repo->url) {
		*strchrnul(repo->url, '\n') = '\0';
	} else {
		// Nothing can address a repository with no url, and every lookup
		// compares the url, so it is kept out of the way.
		fprintf(stderr, "[cgit] Ignoring repository with an empty url\n");
		repo->url = xstrdup("");
		repo->ignore = 1;
	}
	repo->name = repo->url;
	repo->desc = cgit_default_repo_desc;
	// A cached scan keeps these copies, so build_scan_key in cgit.c has to
	// name every setting copied here.
	repo->section = ctx.cfg.section;
	repo->snapshots = ctx.cfg.snapshots;
	repo->enable_blame = ctx.cfg.enable_blame;
	repo->enable_commit_graph = ctx.cfg.enable_commit_graph;
	repo->enable_follow_links = ctx.cfg.enable_follow_links;
	repo->enable_gitmodules_links = ctx.cfg.enable_gitmodules_links;
	repo->enable_log_filecount = ctx.cfg.enable_log_filecount;
	repo->enable_log_linecount = ctx.cfg.enable_log_linecount;
	repo->enable_mailmap = ctx.cfg.enable_mailmap;
	repo->enable_remote_branches = ctx.cfg.enable_remote_branches;
	repo->enable_subject_links = ctx.cfg.enable_subject_links;
	repo->enable_commit_trailers = ctx.cfg.enable_commit_trailers;
	repo->enable_html_serving = ctx.cfg.enable_html_serving;
	repo->max_stats = ctx.cfg.max_stats;
	repo->branch_sort = ctx.cfg.branch_sort;
	repo->commit_sort = ctx.cfg.commit_sort;
	repo->module_link = ctx.cfg.module_link;
	repo->mtime = -1;
	repo->about_filter = ctx.cfg.about_filter;
	repo->commit_filter = ctx.cfg.commit_filter;
	repo->source_filter = ctx.cfg.source_filter;
	repo->email_filter = ctx.cfg.email_filter;
	repo->trailer_filter = ctx.cfg.trailer_filter;
	repo->clone_url = ctx.cfg.clone_url;
	repo->submodules.strdup_strings = 1;
	return repo;
}

struct cgit_repo *cgit_get_repoinfo(const char *url)
{
	int i;
	struct cgit_repo *repo;

	for (i = 0; i < cgit_repolist.count; i++) {
		repo = &cgit_repolist.repos[i];
		if (repo->ignore)
			continue;
		if (!strcmp(repo->url, url))
			return repo;
	}
	return NULL;
}

void cgit_diff_tree_cb(struct diff_queue_struct *q, struct diff_options *options, void *data)
{
	filepair_fn fn = *(filepair_fn *)data;
	const char *wanted = options->single_follow;
	int i;

	for (i = 0; i < q->nr; i++) {
		// An unmerged path has no single pair of blobs to show.
		if (q->queue[i]->status == 'U')
			continue;
		// A diff aimed at one path ran over the whole commit.
		if (wanted && strcmp(q->queue[i]->two->path, wanted))
			continue;
		fn(q->queue[i]);
	}
}

static int load_mmfile(mmfile_t *file, const struct object_id *oid)
{
	enum object_type type;
	size_t size;

	// A null oid is the absent side of an add or a delete, and diffs as an
	// empty file rather than as a failure.
	if (is_null_oid(oid)) {
		file->ptr = (char *)"";
		file->size = 0;
		return 1;
	}

	file->ptr = odb_read_object(the_repository->objects, oid, &type, &size);
	if (!file->ptr)
		return 0;
	file->size = size;
	return 1;
}

/*
 * The test is on the oid rather than the size, because load_mmfile uses a
 * literal only for a null oid, while a real but empty blob owns its buffer.
 */
static void release_mmfile(mmfile_t *file, const struct object_id *oid)
{
	if (!is_null_oid(oid))
		free(file->ptr);
}

/*
 * Xdiff hands over one line per call, in pieces that are the prefix, the
 * record and, for a record with no newline, the marker saying so. They are
 * joined here before the line reaches fn.
 */
static int emit_line(void *priv, mmbuffer_t *mb, int nbuf)
{
	linediff_fn emit = *(linediff_fn *)priv;
	struct strbuf line = STRBUF_INIT;
	int i;

	for (i = 0; i < nbuf; i++)
		strbuf_add(&line, mb[i].ptr, mb[i].size);
	emit(line.buf, line.len);
	strbuf_release(&line);
	return 0;
}

int cgit_diff_files(const struct object_id *old_oid, const struct object_id *new_oid,
	size_t *old_size, size_t *new_size, int *binary, int context, int ignorews,
	linediff_fn fn)
{
	mmfile_t old_file, new_file;
	xpparam_t diff_params;
	xdemitconf_t emit_params;
	xdemitcb_t emit_cb;
	size_t old_bytes = 0, new_bytes = 0;

	// Read the object headers first so an oversized blob is never inflated
	// into memory just to be diffed. Reporting it as binary suppresses
	// inlining the same way max-blob-size does in the other views.
	if (
		!is_null_oid(old_oid) &&
		odb_read_object_info(the_repository->objects, old_oid, &old_bytes) < 0
	)
		return 1;
	if (
		!is_null_oid(new_oid) &&
		odb_read_object_info(the_repository->objects, new_oid, &new_bytes) < 0
	)
		return 1;

	*old_size = old_bytes;
	*new_size = new_bytes;

	if (cgit_blob_too_large(old_bytes) || cgit_blob_too_large(new_bytes)) {
		*binary = 1;
		return 0;
	}

	if (!load_mmfile(&old_file, old_oid))
		return 1;
	if (!load_mmfile(&new_file, new_oid)) {
		release_mmfile(&old_file, old_oid);
		return 1;
	}

	if (
		buffer_is_binary(old_file.ptr, old_file.size) ||
		buffer_is_binary(new_file.ptr, new_file.size)
	) {
		*binary = 1;
		release_mmfile(&old_file, old_oid);
		release_mmfile(&new_file, new_oid);
		return 0;
	}

	memset(&diff_params, 0, sizeof(diff_params));
	memset(&emit_params, 0, sizeof(emit_params));
	memset(&emit_cb, 0, sizeof(emit_cb));
	// Without XDF_NEED_MINIMAL, as git diffs by default: on a file whose
	// lines were reordered, a minimal diff costs time quadratic in its
	// length.
	if (ignorews)
		diff_params.flags |= XDF_IGNORE_WHITESPACE;
	emit_params.ctxlen = context > 0 ? context : DEFAULT_DIFF_CONTEXT;
	emit_params.flags = XDL_EMIT_FUNCNAMES;
	emit_cb.out_line = emit_line;
	emit_cb.priv = &fn;
	xdl_diff(&old_file, &new_file, &diff_params, &emit_params, &emit_cb);
	release_mmfile(&old_file, old_oid);
	release_mmfile(&new_file, new_oid);
	return 0;
}

static void set_literal_item(struct pathspec_item *item, const char *path)
{
	int len = path ? strlen(path) : 0;

	// nowildcard_len matching len makes git treat the path as literal
	// rather than as a glob.
	*item = (struct pathspec_item){ .match = path, .len = len, .nowildcard_len = len };
}

void cgit_diff_limit_path(struct diff_options *opt, const char *prefix)
{
	cgit_literal_pathspec(&opt->pathspec, xmalloc(sizeof(struct pathspec_item)), xstrdup(prefix));
}

void cgit_diff_limit_rename(struct diff_options *opt, const char *path, const char *source)
{
	struct pathspec_item *items = xcalloc(2, sizeof(*items));

	set_literal_item(&items[0], xstrdup(path));
	set_literal_item(&items[1], xstrdup(source));
	opt->pathspec.nr = 2;
	opt->pathspec.items = items;
	opt->detect_rename = DIFF_DETECT_RENAME;
}

/*
 * Whether path names something other than a directory in the new tree that
 * the old tree lacks, which is where a rename lands.
 */
static int path_is_new(const struct object_id *old_oid, const struct object_id *new_oid,
	const char *path)
{
	struct object_id oid;
	unsigned short mode;

	if (!*path || !old_oid || is_null_oid(old_oid))
		return 0;
	if (get_tree_entry(the_repository, new_oid, path, &oid, &mode) || S_ISDIR(mode))
		return 0;
	return get_tree_entry(the_repository, old_oid, path, &oid, &mode) != 0;
}

void cgit_diff_tree(const struct object_id *old_oid, const struct object_id *new_oid,
	filepair_fn fn, const char *prefix, int ignorews)
{
	struct diff_options opt;

	repo_diff_setup(the_repository, &opt);
	opt.output_format = DIFF_FORMAT_CALLBACK;
	opt.detect_rename = 1;
	opt.rename_limit = ctx.cfg.renamelimit;
	opt.flags.recursive = 1;
	if (ignorews)
		DIFF_XDL_SET(&opt, IGNORE_WHITESPACE);
	opt.format_callback = cgit_diff_tree_cb;
	opt.format_callback_data = &fn;
	// The source of a rename lies outside a limit naming its destination,
	// so a new path is diffed across the whole commit instead, with
	// rename detection aimed at that path alone.
	if (prefix && path_is_new(old_oid, new_oid, prefix))
		opt.single_follow = prefix;
	else if (prefix)
		cgit_diff_limit_path(&opt, prefix);
	diff_setup_done(&opt);

	if (old_oid && !is_null_oid(old_oid))
		diff_tree_oid(old_oid, new_oid, "", &opt);
	else
		diff_root_tree_oid(new_oid, "", &opt);
	diffcore_std(&opt);
	diff_flush(&opt);
}

// The old name cgit_rename_source's diff finds, which a filepair_fn has no
// other way to hand back.
static char *rename_source;

static void note_rename_source(struct diff_filepair *pair)
{
	if (pair->status == DIFF_STATUS_RENAMED && !rename_source)
		rename_source = xstrdup(pair->one->path);
}

char *cgit_rename_source(const struct object_id *old_oid, const struct object_id *new_oid,
	const char *path)
{
	char *source;

	if (!path_is_new(old_oid, new_oid, path))
		return NULL;
	cgit_diff_tree(old_oid, new_oid, note_rename_source, path, 0);
	source = rename_source;
	rename_source = NULL;
	return source;
}

void cgit_diff_commit(struct commit *commit, filepair_fn fn, const char *prefix)
{
	const struct object_id *old_oid = NULL;

	if (commit->parents)
		old_oid = &commit->parents->item->object.oid;
	cgit_diff_tree(old_oid, &commit->object.oid, fn, prefix, ctx.qry.ignorews);
}

int cgit_blob_too_large(size_t size)
{
	return ctx.cfg.max_blob_size && size / 1024 > (size_t)ctx.cfg.max_blob_size;
}

void cgit_literal_pathspec(struct pathspec *ps, struct pathspec_item *item, const char *path)
{
	set_literal_item(item, path);
	ps->nr = path ? 1 : 0;
	ps->items = item;
}

char *cgit_trim_end(const char *str, char c)
{
	size_t len;

	if (!str)
		return NULL;
	len = strlen(str);
	while (len > 0 && str[len - 1] == c)
		len--;
	if (len == 0)
		return NULL;
	return xstrndup(str, len);
}

int cgit_read_first_line(const char *path, char **buf)
{
	int fd, err;
	ssize_t got;
	size_t want;
	struct stat st;

	fd = open(path, O_RDONLY);
	if (fd == -1)
		return errno;
	if (fstat(fd, &st)) {
		err = errno;
		close(fd);
		return err;
	}
	if (!S_ISREG(st.st_mode)) {
		close(fd);
		return EISDIR;
	}
	// Only the first line is wanted, and a description file in a scanned
	// repository is whatever a pusher wrote, so the read is bounded.
	want = st.st_size < MAX_FIRST_LINE_READ ? st.st_size : MAX_FIRST_LINE_READ;
	*buf = xmalloc(want + 1);
	got = read_in_full(fd, *buf, want);
	if (got < 0) {
		err = errno;
		free(*buf);
		*buf = NULL;
		close(fd);
		return err;
	}
	(*buf)[got] = '\0';
	*strchrnul(*buf, '\n') = '\0';
	close(fd);
	return 0;
}

char *cgit_strdup_first_line(const char *text)
{
	char *line = xstrdup(text);

	*strchrnul(line, '\n') = '\0';
	return line;
}

void cgit_read_repo_desc(struct cgit_repo *repo)
{
	struct strbuf path = STRBUF_INIT;
	char *desc = NULL;

	if (repo->desc != cgit_default_repo_desc || !repo->path || repo->desc_read)
		return;
	repo->desc_read = 1;
	strbuf_addf(&path, "%s/description", repo->path);
	if (!cgit_read_first_line(path.buf, &desc) && strcmp(desc, default_git_desc))
		repo->desc = desc;
	else
		free(desc);
	strbuf_release(&path);
}

/*
 * The lead-byte ranges fold in the overlong, surrogate and out-of-range
 * cases, so 0 is the only error signal a caller needs.
 */
size_t cgit_utf8_seq_len(const unsigned char *p, size_t left)
{
	size_t len, i;

	if (p[0] >= 0xc2 && p[0] <= 0xdf)
		len = 2;
	else if (p[0] >= 0xe0 && p[0] <= 0xef)
		len = 3;
	else if (p[0] >= 0xf0 && p[0] <= 0xf4)
		len = 4;
	else
		return 0;
	if (left < len)
		return 0;
	for (i = 1; i < len; i++)
		if ((p[i] & 0xc0) != 0x80)
			return 0;
	if (p[0] == 0xe0 && p[1] < 0xa0)
		return 0;
	if (p[0] == 0xed && p[1] > 0x9f)
		return 0;
	if (p[0] == 0xf0 && p[1] < 0x90)
		return 0;
	if (p[0] == 0xf4 && p[1] > 0x8f)
		return 0;
	return len;
}

void cgit_utf8_sanitize(char *text)
{
	unsigned char *p = (unsigned char *)text;
	size_t left = text ? strlen(text) : 0;

	while (left) {
		size_t seq = *p < 0x80 ? 1 : cgit_utf8_seq_len(p, left);

		if (!seq) {
			*p = '?';
			seq = 1;
		}
		p += seq;
		left -= seq;
	}
}

/*
 * Takes an unsigned char because a byte over 0x7f is negative where char is
 * signed, which the ctype tests are not defined for.
 */
static int is_token_char(unsigned char c)
{
	return isalnum(c) || c == '_';
}

/*
 * Replace the variable name held at name with its value from the environment,
 * writing at most room bytes including the terminator, and return where the
 * text now ends.
 */
static char *expand_macro(char *name, size_t room)
{
	char *value;
	size_t len;

	len = 0;
	value = getenv(name);
	if (value) {
		len = strlen(value) + 1;
		if (len > room)
			len = room;
		strlcpy(name, value, len);
		--len;
	}
	return name + len;
}

char *cgit_expand_macros(const char *text)
{
	static char result[MACRO_EXPANSION_BUFSIZE];
	char *limit = result + MACRO_EXPANSION_BUFSIZE - 1;
	char *out, *start;

	out = result;
	start = NULL;
	while (out < limit && text && *text) {
		*out = *text;
		if (start) {
			if (!is_token_char(*text)) {
				if (out - start > 0) {
					*out = '\0';
					out = expand_macro(start, limit - start) - 1;
					// Step back so the byte that ended the
					// token is written again and may open a
					// new one. A bare dollar expanded nothing
					// and its follower is already in place.
					text--;
				}
				start = NULL;
			}
			out++;
			text++;
			continue;
		}
		if (*text == '$') {
			start = out;
			text++;
			continue;
		}
		out++;
		text++;
	}
	*out = '\0';
	if (start && out - start > 0) {
		out = expand_macro(start, limit - start);
		*out = '\0';
	}
	return result;
}

int cgit_readme_is_tracked(const char *readme)
{
	const char *colon = strchr(readme, ':');

	return colon && colon[1];
}

/*
 * Git's wider syntax is refused, since :/pattern and rev^{/pattern} walk the
 * whole history for a match, and a leading dash reads as an option further
 * down.
 */
int cgit_valid_rev(const char *rev)
{
	const char *p = rev;

	if (!rev || !*rev || *rev == '-')
		return 0;
	while (isxdigit((unsigned char)*p))
		p++;
	if (!*p)
		return 1;
	return !check_refname_format(rev, REFNAME_ALLOW_ONELEVEL);
}

int cgit_notes_ref_usable(const char *notes_ref)
{
	struct object_id oid;

	if (repo_get_oid_treeish(the_repository, notes_ref, &oid))
		return 0;
	return !!parse_tree_indirect(&oid);
}