git/builtin/init-db.c

591 lines
16 KiB
C
Raw Normal View History

/*
* GIT - The information manager from hell
*
* Copyright (C) Linus Torvalds, 2005
*/
#include "cache.h"
#include "config.h"
#include "refs.h"
#include "builtin.h"
#include "exec-cmd.h"
#include "parse-options.h"
#ifndef DEFAULT_GIT_TEMPLATE_DIR
#define DEFAULT_GIT_TEMPLATE_DIR "/usr/share/git-core/templates"
#endif
#ifdef NO_TRUSTABLE_FILEMODE
#define TEST_FILEMODE 0
#else
#define TEST_FILEMODE 1
#endif
static int init_is_bare_repository = 0;
static int init_shared_repository = -1;
static const char *init_db_template_dir;
static void copy_templates_1(struct strbuf *path, struct strbuf *template_path,
DIR *dir)
{
size_t path_baselen = path->len;
size_t template_baselen = template_path->len;
struct dirent *de;
/* Note: if ".git/hooks" file exists in the repository being
* re-initialized, /etc/core-git/templates/hooks/update would
* cause "git init" to fail here. I think this is sane but
* it means that the set of templates we ship by default, along
* with the way the namespace under .git/ is organized, should
* be really carefully chosen.
*/
safe_create_dir(path->buf, 1);
while ((de = readdir(dir)) != NULL) {
struct stat st_git, st_template;
int exists = 0;
strbuf_setlen(path, path_baselen);
strbuf_setlen(template_path, template_baselen);
if (de->d_name[0] == '.')
continue;
strbuf_addstr(path, de->d_name);
strbuf_addstr(template_path, de->d_name);
if (lstat(path->buf, &st_git)) {
if (errno != ENOENT)
die_errno(_("cannot stat '%s'"), path->buf);
}
else
exists = 1;
if (lstat(template_path->buf, &st_template))
die_errno(_("cannot stat template '%s'"), template_path->buf);
if (S_ISDIR(st_template.st_mode)) {
DIR *subdir = opendir(template_path->buf);
if (!subdir)
die_errno(_("cannot opendir '%s'"), template_path->buf);
strbuf_addch(path, '/');
strbuf_addch(template_path, '/');
copy_templates_1(path, template_path, subdir);
closedir(subdir);
}
else if (exists)
continue;
else if (S_ISLNK(st_template.st_mode)) {
struct strbuf lnk = STRBUF_INIT;
if (strbuf_readlink(&lnk, template_path->buf,
st_template.st_size) < 0)
die_errno(_("cannot readlink '%s'"), template_path->buf);
if (symlink(lnk.buf, path->buf))
die_errno(_("cannot symlink '%s' '%s'"),
lnk.buf, path->buf);
strbuf_release(&lnk);
}
else if (S_ISREG(st_template.st_mode)) {
if (copy_file(path->buf, template_path->buf, st_template.st_mode))
die_errno(_("cannot copy '%s' to '%s'"),
template_path->buf, path->buf);
}
else
error(_("ignoring template %s"), template_path->buf);
}
}
static void copy_templates(const char *template_dir)
{
struct strbuf path = STRBUF_INIT;
struct strbuf template_path = STRBUF_INIT;
size_t template_len;
struct repository_format template_format;
struct strbuf err = STRBUF_INIT;
DIR *dir;
char *to_free = NULL;
if (!template_dir)
template_dir = getenv(TEMPLATE_DIR_ENVIRONMENT);
if (!template_dir)
template_dir = init_db_template_dir;
if (!template_dir)
template_dir = to_free = system_path(DEFAULT_GIT_TEMPLATE_DIR);
if (!template_dir[0]) {
free(to_free);
return;
}
strbuf_addstr(&template_path, template_dir);
strbuf_complete(&template_path, '/');
template_len = template_path.len;
dir = opendir(template_path.buf);
if (!dir) {
warning(_("templates not found in %s"), template_dir);
goto free_return;
}
/* Make sure that template is from the correct vintage */
strbuf_addstr(&template_path, "config");
read_repository_format(&template_format, template_path.buf);
strbuf_setlen(&template_path, template_len);
/*
* No mention of version at all is OK, but anything else should be
* verified.
*/
if (template_format.version >= 0 &&
verify_repository_format(&template_format, &err) < 0) {
warning(_("not copying templates from '%s': %s"),
template_dir, err.buf);
strbuf_release(&err);
goto close_free_return;
}
strbuf_addstr(&path, get_git_common_dir());
strbuf_complete(&path, '/');
copy_templates_1(&path, &template_path, dir);
close_free_return:
closedir(dir);
free_return:
free(to_free);
strbuf_release(&path);
strbuf_release(&template_path);
}
static int git_init_db_config(const char *k, const char *v, void *cb)
{
if (!strcmp(k, "init.templatedir"))
return git_config_pathname(&init_db_template_dir, k, v);
return 0;
}
/*
* If the git_dir is not directly inside the working tree, then git will not
* find it by default, and we need to set the worktree explicitly.
*/
static int needs_work_tree_config(const char *git_dir, const char *work_tree)
{
if (!strcmp(work_tree, "/") && !strcmp(git_dir, "/.git"))
return 0;
if (skip_prefix(git_dir, work_tree, &git_dir) &&
!strcmp(git_dir, "/.git"))
return 0;
return 1;
}
static int create_default_files(const char *template_path,
const char *original_git_dir)
{
struct stat st1;
struct strbuf buf = STRBUF_INIT;
char *path;
char repo_version_string[10];
char junk[2];
int reinit;
int filemode;
struct strbuf err = STRBUF_INIT;
/* Just look for `init.templatedir` */
git_config(git_init_db_config, NULL);
/*
* First copy the templates -- we might have the default
* config file there, in which case we would want to read
* from it after installing.
*
* Before reading that config, we also need to clear out any cached
* values (since we've just potentially changed what's available on
* disk).
*/
copy_templates(template_path);
git_config_clear();
reset_shared_repository();
git_config(git_default_config, NULL);
/*
* We must make sure command-line options continue to override any
* values we might have just re-read from the config.
*/
is_bare_repository_cfg = init_is_bare_repository;
if (init_shared_repository != -1)
set_shared_repository(init_shared_repository);
/*
* We would have created the above under user's umask -- under
* shared-repository settings, we would need to fix them up.
*/
if (get_shared_repository()) {
adjust_shared_perm(get_git_dir());
}
/*
* We need to create a "refs" dir in any case so that older
* versions of git can tell that this is a repository.
*/
safe_create_dir(git_path("refs"), 1);
adjust_shared_perm(git_path("refs"));
if (refs_init_db(&err))
die("failed to set up refs db: %s", err.buf);
/*
* Create the default symlink from ".git/HEAD" to the "master"
* branch, if it does not exist yet.
*/
path = git_path_buf(&buf, "HEAD");
reinit = (!access(path, R_OK)
|| readlink(path, junk, sizeof(junk)-1) != -1);
if (!reinit) {
if (create_symref("HEAD", "refs/heads/master", NULL) < 0)
exit(1);
}
/* This forces creation of new config file */
xsnprintf(repo_version_string, sizeof(repo_version_string),
"%d", GIT_REPO_VERSION);
git_config_set("core.repositoryformatversion", repo_version_string);
/* Check filemode trustability */
path = git_path_buf(&buf, "config");
filemode = TEST_FILEMODE;
if (TEST_FILEMODE && !lstat(path, &st1)) {
struct stat st2;
filemode = (!chmod(path, st1.st_mode ^ S_IXUSR) &&
!lstat(path, &st2) &&
st1.st_mode != st2.st_mode &&
!chmod(path, st1.st_mode));
if (filemode && !reinit && (st1.st_mode & S_IXUSR))
filemode = 0;
}
git_config_set("core.filemode", filemode ? "true" : "false");
Clean up work-tree handling The old version of work-tree support was an unholy mess, barely readable, and not to the point. For example, why do you have to provide a worktree, when it is not used? As in "git status". Now it works. Another riddle was: if you can have work trees inside the git dir, why are some programs complaining that they need a work tree? IOW it is allowed to call $ git --git-dir=../ --work-tree=. bla when you really want to. In this case, you are both in the git directory and in the working tree. So, programs have to actually test for the right thing, namely if they are inside a working tree, and not if they are inside a git directory. Also, GIT_DIR=../.git should behave the same as if no GIT_DIR was specified, unless there is a repository in the current working directory. It does now. The logic to determine if a repository is bare, or has a work tree (tertium non datur), is this: --work-tree=bla overrides GIT_WORK_TREE, which overrides core.bare = true, which overrides core.worktree, which overrides GIT_DIR/.. when GIT_DIR ends in /.git, which overrides the directory in which .git/ was found. In related news, a long standing bug was fixed: when in .git/bla/x.git/, which is a bare repository, git formerly assumed ../.. to be the appropriate git dir. This problem was reported by Shawn Pearce to have caused much pain, where a colleague mistakenly ran "git init" in "/" a long time ago, and bare repositories just would not work. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2007-08-01 08:30:14 +08:00
if (is_bare_repository())
git_config_set("core.bare", "true");
else {
Clean up work-tree handling The old version of work-tree support was an unholy mess, barely readable, and not to the point. For example, why do you have to provide a worktree, when it is not used? As in "git status". Now it works. Another riddle was: if you can have work trees inside the git dir, why are some programs complaining that they need a work tree? IOW it is allowed to call $ git --git-dir=../ --work-tree=. bla when you really want to. In this case, you are both in the git directory and in the working tree. So, programs have to actually test for the right thing, namely if they are inside a working tree, and not if they are inside a git directory. Also, GIT_DIR=../.git should behave the same as if no GIT_DIR was specified, unless there is a repository in the current working directory. It does now. The logic to determine if a repository is bare, or has a work tree (tertium non datur), is this: --work-tree=bla overrides GIT_WORK_TREE, which overrides core.bare = true, which overrides core.worktree, which overrides GIT_DIR/.. when GIT_DIR ends in /.git, which overrides the directory in which .git/ was found. In related news, a long standing bug was fixed: when in .git/bla/x.git/, which is a bare repository, git formerly assumed ../.. to be the appropriate git dir. This problem was reported by Shawn Pearce to have caused much pain, where a colleague mistakenly ran "git init" in "/" a long time ago, and bare repositories just would not work. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2007-08-01 08:30:14 +08:00
const char *work_tree = get_git_work_tree();
git_config_set("core.bare", "false");
/* allow template config file to override the default */
if (log_all_ref_updates == LOG_REFS_UNSET)
git_config_set("core.logallrefupdates", "true");
if (needs_work_tree_config(original_git_dir, work_tree))
git_config_set("core.worktree", work_tree);
}
if (!reinit) {
/* Check if symlink is supported in the work tree */
path = git_path_buf(&buf, "tXXXXXX");
if (!close(xmkstemp(path)) &&
!unlink(path) &&
!symlink("testing", path) &&
!lstat(path, &st1) &&
S_ISLNK(st1.st_mode))
unlink(path); /* good */
else
git_config_set("core.symlinks", "false");
/* Check if the filesystem is case-insensitive */
path = git_path_buf(&buf, "CoNfIg");
if (!access(path, F_OK))
git_config_set("core.ignorecase", "true");
probe_utf8_pathname_composition();
}
strbuf_release(&buf);
return reinit;
}
static void create_object_directory(void)
{
struct strbuf path = STRBUF_INIT;
size_t baselen;
strbuf_addstr(&path, get_object_directory());
baselen = path.len;
safe_create_dir(path.buf, 1);
strbuf_setlen(&path, baselen);
strbuf_addstr(&path, "/pack");
safe_create_dir(path.buf, 1);
strbuf_setlen(&path, baselen);
strbuf_addstr(&path, "/info");
safe_create_dir(path.buf, 1);
strbuf_release(&path);
}
static void separate_git_dir(const char *git_dir, const char *git_link)
{
struct stat st;
if (!stat(git_link, &st)) {
const char *src;
if (S_ISREG(st.st_mode))
src = read_gitfile(git_link);
else if (S_ISDIR(st.st_mode))
src = git_link;
else
die(_("unable to handle file type %d"), (int)st.st_mode);
if (rename(src, git_dir))
die_errno(_("unable to move %s to %s"), src, git_dir);
}
write_file(git_link, "gitdir: %s", git_dir);
}
int init_db(const char *git_dir, const char *real_git_dir,
const char *template_dir, unsigned int flags)
{
int reinit;
int exist_ok = flags & INIT_DB_EXIST_OK;
char *original_git_dir = real_pathdup(git_dir, 1);
if (real_git_dir) {
struct stat st;
if (!exist_ok && !stat(git_dir, &st))
die(_("%s already exists"), git_dir);
if (!exist_ok && !stat(real_git_dir, &st))
die(_("%s already exists"), real_git_dir);
set_git_dir(real_path(real_git_dir));
git_dir = get_git_dir();
separate_git_dir(git_dir, original_git_dir);
}
else {
set_git_dir(real_path(git_dir));
git_dir = get_git_dir();
}
startup_info->have_repository = 1;
safe_create_dir(git_dir, 0);
init_is_bare_repository = is_bare_repository();
/* Check to see if the repository version is right.
* Note that a newly created repository does not have
* config file, so this will not fail. What we are catching
* is an attempt to reinitialize new repository with an old tool.
*/
check_repository_format();
reinit = create_default_files(template_dir, original_git_dir);
create_object_directory();
if (get_shared_repository()) {
char buf[10];
/* We do not spell "group" and such, so that
* the configuration can be read by older version
* of git. Note, we use octal numbers for new share modes,
* and compatibility values for PERM_GROUP and
* PERM_EVERYBODY.
*/
if (get_shared_repository() < 0)
/* force to the mode value */
xsnprintf(buf, sizeof(buf), "0%o", -get_shared_repository());
else if (get_shared_repository() == PERM_GROUP)
xsnprintf(buf, sizeof(buf), "%d", OLD_PERM_GROUP);
else if (get_shared_repository() == PERM_EVERYBODY)
xsnprintf(buf, sizeof(buf), "%d", OLD_PERM_EVERYBODY);
else
BUG("invalid value for shared_repository");
git_config_set("core.sharedrepository", buf);
git_config_set("receive.denyNonFastforwards", "true");
}
if (!(flags & INIT_DB_QUIET)) {
int len = strlen(git_dir);
if (reinit)
printf(get_shared_repository()
? _("Reinitialized existing shared Git repository in %s%s\n")
: _("Reinitialized existing Git repository in %s%s\n"),
git_dir, len && git_dir[len-1] != '/' ? "/" : "");
else
printf(get_shared_repository()
? _("Initialized empty shared Git repository in %s%s\n")
: _("Initialized empty Git repository in %s%s\n"),
git_dir, len && git_dir[len-1] != '/' ? "/" : "");
}
free(original_git_dir);
return 0;
}
static int guess_repository_type(const char *git_dir)
{
const char *slash;
char *cwd;
int cwd_is_git_dir;
/*
* "GIT_DIR=. git init" is always bare.
* "GIT_DIR=`pwd` git init" too.
*/
if (!strcmp(".", git_dir))
return 1;
cwd = xgetcwd();
cwd_is_git_dir = !strcmp(git_dir, cwd);
free(cwd);
if (cwd_is_git_dir)
return 1;
/*
* "GIT_DIR=.git or GIT_DIR=something/.git is usually not.
*/
if (!strcmp(git_dir, ".git"))
return 0;
slash = strrchr(git_dir, '/');
if (slash && !strcmp(slash, "/.git"))
return 0;
/*
* Otherwise it is often bare. At this point
* we are just guessing.
*/
return 1;
}
static int shared_callback(const struct option *opt, const char *arg, int unset)
{
assert NOARG/NONEG behavior of parse-options callbacks When we define a parse-options callback, the flags we put in the option struct must match what the callback expects. For example, a callback which does not handle the "unset" parameter should only be used with PARSE_OPT_NONEG. But since the callback and the option struct are not defined next to each other, it's easy to get this wrong (as earlier patches in this series show). Fortunately, the compiler can help us here: compiling with -Wunused-parameters can show us which callbacks ignore their "unset" parameters (and likewise, ones that ignore "arg" expect to be triggered with PARSE_OPT_NOARG). But after we've inspected a callback and determined that all of its callers use the right flags, what do we do next? We'd like to silence the compiler warning, but do so in a way that will catch any wrong calls in the future. We can do that by actually checking those variables and asserting that they match our expectations. Because this is such a common pattern, we'll introduce some helper macros. The resulting messages aren't as descriptive as we could make them, but the file/line information from BUG() is enough to identify the problem (and anyway, the point is that these should never be seen). Each of the annotated callbacks in this patch triggers -Wunused-parameters, and was manually inspected to make sure all callers use the correct options (so none of these BUGs should be triggerable). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-11-05 14:45:42 +08:00
BUG_ON_OPT_NEG(unset);
*((int *) opt->value) = (arg) ? git_config_perm("arg", arg) : PERM_GROUP;
return 0;
}
static const char *const init_db_usage[] = {
N_("git init [-q | --quiet] [--bare] [--template=<template-directory>] [--shared[=<permissions>]] [<directory>]"),
NULL
};
/*
* If you want to, you can share the DB area with any number of branches.
* That has advantages: you can save space by sharing all the SHA1 objects.
* On the other hand, it might just make lookup slower and messier. You
* be the judge. The default case is to have one DB per managed directory.
*/
int cmd_init_db(int argc, const char **argv, const char *prefix)
{
const char *git_dir;
const char *real_git_dir = NULL;
const char *work_tree;
const char *template_dir = NULL;
unsigned int flags = 0;
const struct option init_db_options[] = {
OPT_STRING(0, "template", &template_dir, N_("template-directory"),
N_("directory from which templates will be used")),
OPT_SET_INT(0, "bare", &is_bare_repository_cfg,
N_("create a bare repository"), 1),
{ OPTION_CALLBACK, 0, "shared", &init_shared_repository,
N_("permissions"),
N_("specify that the git repository is to be shared amongst several users"),
PARSE_OPT_OPTARG | PARSE_OPT_NONEG, shared_callback, 0},
OPT_BIT('q', "quiet", &flags, N_("be quiet"), INIT_DB_QUIET),
OPT_STRING(0, "separate-git-dir", &real_git_dir, N_("gitdir"),
N_("separate git dir from working tree")),
OPT_END()
};
argc = parse_options(argc, argv, prefix, init_db_options, init_db_usage, 0);
if (real_git_dir && !is_absolute_path(real_git_dir))
real_git_dir = real_pathdup(real_git_dir, 1);
if (argc == 1) {
int mkdir_tried = 0;
retry:
if (chdir(argv[0]) < 0) {
if (!mkdir_tried) {
int saved;
/*
* At this point we haven't read any configuration,
* and we know shared_repository should always be 0;
* but just in case we play safe.
*/
saved = get_shared_repository();
set_shared_repository(0);
switch (safe_create_leading_directories_const(argv[0])) {
case SCLD_OK:
case SCLD_PERMS:
break;
case SCLD_EXISTS:
errno = EEXIST;
/* fallthru */
default:
die_errno(_("cannot mkdir %s"), argv[0]);
break;
}
set_shared_repository(saved);
if (mkdir(argv[0], 0777) < 0)
die_errno(_("cannot mkdir %s"), argv[0]);
mkdir_tried = 1;
goto retry;
}
die_errno(_("cannot chdir to %s"), argv[0]);
}
} else if (0 < argc) {
usage(init_db_usage[0]);
}
if (is_bare_repository_cfg == 1) {
char *cwd = xgetcwd();
setenv(GIT_DIR_ENVIRONMENT, cwd, argc > 0);
free(cwd);
}
if (init_shared_repository != -1)
set_shared_repository(init_shared_repository);
/*
* GIT_WORK_TREE makes sense only in conjunction with GIT_DIR
* without --bare. Catch the error early.
*/
git_dir = xstrdup_or_null(getenv(GIT_DIR_ENVIRONMENT));
work_tree = xstrdup_or_null(getenv(GIT_WORK_TREE_ENVIRONMENT));
if ((!git_dir || is_bare_repository_cfg == 1) && work_tree)
die(_("%s (or --work-tree=<directory>) not allowed without "
"specifying %s (or --git-dir=<directory>)"),
GIT_WORK_TREE_ENVIRONMENT,
GIT_DIR_ENVIRONMENT);
/*
* Set up the default .git directory contents
*/
if (!git_dir)
git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
if (is_bare_repository_cfg < 0)
is_bare_repository_cfg = guess_repository_type(git_dir);
if (!is_bare_repository_cfg) {
const char *git_dir_parent = strrchr(git_dir, '/');
if (git_dir_parent) {
char *rel = xstrndup(git_dir, git_dir_parent - git_dir);
git_work_tree_cfg = real_pathdup(rel, 1);
free(rel);
}
if (!git_work_tree_cfg)
git_work_tree_cfg = xgetcwd();
if (work_tree)
set_git_work_tree(work_tree);
else
set_git_work_tree(git_work_tree_cfg);
if (access(get_git_work_tree(), X_OK))
die_errno (_("Cannot access work tree '%s'"),
get_git_work_tree());
}
else {
if (work_tree)
set_git_work_tree(work_tree);
}
add UNLEAK annotation for reducing leak false positives It's a common pattern in git commands to allocate some memory that should last for the lifetime of the program and then not bother to free it, relying on the OS to throw it away. This keeps the code simple, and it's fast (we don't waste time traversing structures or calling free at the end of the program). But it also triggers warnings from memory-leak checkers like valgrind or LSAN. They know that the memory was still allocated at program exit, but they don't know _when_ the leaked memory stopped being useful. If it was early in the program, then it's probably a real and important leak. But if it was used right up until program exit, it's not an interesting leak and we'd like to suppress it so that we can see the real leaks. This patch introduces an UNLEAK() macro that lets us do so. To understand its design, let's first look at some of the alternatives. Unfortunately the suppression systems offered by leak-checking tools don't quite do what we want. A leak-checker basically knows two things: 1. Which blocks were allocated via malloc, and the callstack during the allocation. 2. Which blocks were left un-freed at the end of the program (and which are unreachable, but more on that later). Their suppressions work by mentioning the function or callstack of a particular allocation, and marking it as OK to leak. So imagine you have code like this: int cmd_foo(...) { /* this allocates some memory */ char *p = some_function(); printf("%s", p); return 0; } You can say "ignore allocations from some_function(), they're not leaks". But that's not right. That function may be called elsewhere, too, and we would potentially want to know about those leaks. So you can say "ignore the callstack when main calls some_function". That works, but your annotations are brittle. In this case it's only two functions, but you can imagine that the actual allocation is much deeper. If any of the intermediate code changes, you have to update the suppression. What we _really_ want to say is that "the value assigned to p at the end of the function is not a real leak". But leak-checkers can't understand that; they don't know about "p" in the first place. However, we can do something a little bit tricky if we make some assumptions about how leak-checkers work. They generally don't just report all un-freed blocks. That would report even globals which are still accessible when the leak-check is run. Instead they take some set of memory (like BSS) as a root and mark it as "reachable". Then they scan the reachable blocks for anything that looks like a pointer to a malloc'd block, and consider that block reachable. And then they scan those blocks, and so on, transitively marking anything reachable from a global as "not leaked" (or at least leaked in a different category). So we can mark the value of "p" as reachable by putting it into a variable with program lifetime. One way to do that is to just mark "p" as static. But that actually affects the run-time behavior if the function is called twice (you aren't likely to call main() twice, but some of our cmd_*() functions are called from other commands). Instead, we can trick the leak-checker by putting the value into _any_ reachable bytes. This patch keeps a global linked-list of bytes copied from "unleaked" variables. That list is reachable even at program exit, which confers recursive reachability on whatever values we unleak. In other words, you can do: int cmd_foo(...) { char *p = some_function(); printf("%s", p); UNLEAK(p); return 0; } to annotate "p" and suppress the leak report. But wait, couldn't we just say "free(p)"? In this toy example, yes. But UNLEAK()'s byte-copying strategy has several advantages over actually freeing the memory: 1. It's recursive across structures. In many cases our "p" is not just a pointer, but a complex struct whose fields may have been allocated by a sub-function. And in some cases (e.g., dir_struct) we don't even have a function which knows how to free all of the struct members. By marking the struct itself as reachable, that confers reachability on any pointers it contains (including those found in embedded structs, or reachable by walking heap blocks recursively. 2. It works on cases where we're not sure if the value is allocated or not. For example: char *p = argc > 1 ? argv[1] : some_function(); It's safe to use UNLEAK(p) here, because it's not freeing any memory. In the case that we're pointing to argv here, the reachability checker will just ignore our bytes. 3. Likewise, it works even if the variable has _already_ been freed. We're just copying the pointer bytes. If the block has been freed, the leak-checker will skip over those bytes as uninteresting. 4. Because it's not actually freeing memory, you can UNLEAK() before we are finished accessing the variable. This is helpful in cases like this: char *p = some_function(); return another_function(p); Writing this with free() requires: int ret; char *p = some_function(); ret = another_function(p); free(p); return ret; But with unleak we can just write: char *p = some_function(); UNLEAK(p); return another_function(p); This patch adds the UNLEAK() macro and enables it automatically when Git is compiled with SANITIZE=leak. In normal builds it's a noop, so we pay no runtime cost. It also adds some UNLEAK() annotations to show off how the feature works. On top of other recent leak fixes, these are enough to get t0000 and t0001 to pass when compiled with LSAN. Note the case in commit.c which actually converts a strbuf_release() into an UNLEAK. This code was already non-leaky, but the free didn't do anything useful, since we're exiting. Converting it to an annotation means that non-leak-checking builds pay no runtime cost. The cost is minimal enough that it's probably not worth going on a crusade to convert these kinds of frees to UNLEAKS. I did it here for consistency with the "sb" leak (though it would have been equally correct to go the other way, and turn them both into strbuf_release() calls). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 14:38:41 +08:00
UNLEAK(real_git_dir);
UNLEAK(git_dir);
UNLEAK(work_tree);
add UNLEAK annotation for reducing leak false positives It's a common pattern in git commands to allocate some memory that should last for the lifetime of the program and then not bother to free it, relying on the OS to throw it away. This keeps the code simple, and it's fast (we don't waste time traversing structures or calling free at the end of the program). But it also triggers warnings from memory-leak checkers like valgrind or LSAN. They know that the memory was still allocated at program exit, but they don't know _when_ the leaked memory stopped being useful. If it was early in the program, then it's probably a real and important leak. But if it was used right up until program exit, it's not an interesting leak and we'd like to suppress it so that we can see the real leaks. This patch introduces an UNLEAK() macro that lets us do so. To understand its design, let's first look at some of the alternatives. Unfortunately the suppression systems offered by leak-checking tools don't quite do what we want. A leak-checker basically knows two things: 1. Which blocks were allocated via malloc, and the callstack during the allocation. 2. Which blocks were left un-freed at the end of the program (and which are unreachable, but more on that later). Their suppressions work by mentioning the function or callstack of a particular allocation, and marking it as OK to leak. So imagine you have code like this: int cmd_foo(...) { /* this allocates some memory */ char *p = some_function(); printf("%s", p); return 0; } You can say "ignore allocations from some_function(), they're not leaks". But that's not right. That function may be called elsewhere, too, and we would potentially want to know about those leaks. So you can say "ignore the callstack when main calls some_function". That works, but your annotations are brittle. In this case it's only two functions, but you can imagine that the actual allocation is much deeper. If any of the intermediate code changes, you have to update the suppression. What we _really_ want to say is that "the value assigned to p at the end of the function is not a real leak". But leak-checkers can't understand that; they don't know about "p" in the first place. However, we can do something a little bit tricky if we make some assumptions about how leak-checkers work. They generally don't just report all un-freed blocks. That would report even globals which are still accessible when the leak-check is run. Instead they take some set of memory (like BSS) as a root and mark it as "reachable". Then they scan the reachable blocks for anything that looks like a pointer to a malloc'd block, and consider that block reachable. And then they scan those blocks, and so on, transitively marking anything reachable from a global as "not leaked" (or at least leaked in a different category). So we can mark the value of "p" as reachable by putting it into a variable with program lifetime. One way to do that is to just mark "p" as static. But that actually affects the run-time behavior if the function is called twice (you aren't likely to call main() twice, but some of our cmd_*() functions are called from other commands). Instead, we can trick the leak-checker by putting the value into _any_ reachable bytes. This patch keeps a global linked-list of bytes copied from "unleaked" variables. That list is reachable even at program exit, which confers recursive reachability on whatever values we unleak. In other words, you can do: int cmd_foo(...) { char *p = some_function(); printf("%s", p); UNLEAK(p); return 0; } to annotate "p" and suppress the leak report. But wait, couldn't we just say "free(p)"? In this toy example, yes. But UNLEAK()'s byte-copying strategy has several advantages over actually freeing the memory: 1. It's recursive across structures. In many cases our "p" is not just a pointer, but a complex struct whose fields may have been allocated by a sub-function. And in some cases (e.g., dir_struct) we don't even have a function which knows how to free all of the struct members. By marking the struct itself as reachable, that confers reachability on any pointers it contains (including those found in embedded structs, or reachable by walking heap blocks recursively. 2. It works on cases where we're not sure if the value is allocated or not. For example: char *p = argc > 1 ? argv[1] : some_function(); It's safe to use UNLEAK(p) here, because it's not freeing any memory. In the case that we're pointing to argv here, the reachability checker will just ignore our bytes. 3. Likewise, it works even if the variable has _already_ been freed. We're just copying the pointer bytes. If the block has been freed, the leak-checker will skip over those bytes as uninteresting. 4. Because it's not actually freeing memory, you can UNLEAK() before we are finished accessing the variable. This is helpful in cases like this: char *p = some_function(); return another_function(p); Writing this with free() requires: int ret; char *p = some_function(); ret = another_function(p); free(p); return ret; But with unleak we can just write: char *p = some_function(); UNLEAK(p); return another_function(p); This patch adds the UNLEAK() macro and enables it automatically when Git is compiled with SANITIZE=leak. In normal builds it's a noop, so we pay no runtime cost. It also adds some UNLEAK() annotations to show off how the feature works. On top of other recent leak fixes, these are enough to get t0000 and t0001 to pass when compiled with LSAN. Note the case in commit.c which actually converts a strbuf_release() into an UNLEAK. This code was already non-leaky, but the free didn't do anything useful, since we're exiting. Converting it to an annotation means that non-leak-checking builds pay no runtime cost. The cost is minimal enough that it's probably not worth going on a crusade to convert these kinds of frees to UNLEAKS. I did it here for consistency with the "sb" leak (though it would have been equally correct to go the other way, and turn them both into strbuf_release() calls). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-08 14:38:41 +08:00
flags |= INIT_DB_EXIST_OK;
return init_db(git_dir, real_git_dir, template_dir, flags);
}