git/lockfile.c

403 lines
9.5 KiB
C
Raw Normal View History

/*
* Copyright (c) 2005, Junio C Hamano
*/
#include "cache.h"
#include "lockfile.h"
#include "sigchain.h"
static struct lock_file *volatile lock_file_list;
static void remove_lock_files(int skip_fclose)
{
pid_t me = getpid();
while (lock_file_list) {
if (lock_file_list->owner == me) {
/* fclose() is not safe to call in a signal handler */
if (skip_fclose)
lock_file_list->fp = NULL;
rollback_lock_file(lock_file_list);
}
lock_file_list = lock_file_list->next;
}
}
static void remove_lock_files_on_exit(void)
{
remove_lock_files(0);
}
static void remove_lock_files_on_signal(int signo)
{
remove_lock_files(1);
sigchain_pop(signo);
raise(signo);
}
/*
* path = absolute or relative path name
*
* Remove the last path name element from path (leaving the preceding
* "/", if any). If path is empty or the root directory ("/"), set
* path to the empty string.
*/
static void trim_last_path_component(struct strbuf *path)
{
int i = path->len;
/* back up past trailing slashes, if any */
while (i && path->buf[i - 1] == '/')
i--;
/*
* then go backwards until a slash, or the beginning of the
* string
*/
while (i && path->buf[i - 1] != '/')
i--;
strbuf_setlen(path, i);
}
/* We allow "recursive" symbolic links. Only within reason, though */
#define MAXDEPTH 5
/*
* path contains a path that might be a symlink.
*
* If path is a symlink, attempt to overwrite it with a path to the
* real file or directory (which may or may not exist), following a
* chain of symlinks if necessary. Otherwise, leave path unmodified.
*
* This is a best-effort routine. If an error occurs, path will
* either be left unmodified or will name a different symlink in a
* symlink chain that started with the original path.
*/
static void resolve_symlink(struct strbuf *path)
{
int depth = MAXDEPTH;
static struct strbuf link = STRBUF_INIT;
while (depth--) {
if (strbuf_readlink(&link, path->buf, path->len) < 0)
break;
if (is_absolute_path(link.buf))
/* absolute path simply replaces p */
strbuf_reset(path);
else
/*
* link is a relative path, so replace the
* last element of p with it.
*/
trim_last_path_component(path);
strbuf_addbuf(path, &link);
}
strbuf_reset(&link);
}
/* Make sure errno contains a meaningful value on error */
static int lock_file(struct lock_file *lk, const char *path, int flags)
{
size_t pathlen = strlen(path);
if (!lock_file_list) {
/* One-time initialization */
sigchain_push_common(remove_lock_files_on_signal);
atexit(remove_lock_files_on_exit);
}
lockfile: avoid transitory invalid states Because remove_lock_file() can be called any time by the signal handler, it is important that any lock_file objects that are in the lock_file_list are always in a valid state. And since lock_file objects are often reused (but are never removed from lock_file_list), that means we have to be careful whenever mutating a lock_file object to always keep it in a well-defined state. This was formerly not the case, because part of the state was encoded by setting lk->filename to the empty string vs. a valid filename. It is wrong to assume that this string can be updated atomically; for example, even strcpy(lk->filename, value) is unsafe. But the old code was even more reckless; for example, strcpy(lk->filename, path); if (!(flags & LOCK_NODEREF)) resolve_symlink(lk->filename, max_path_len); strcat(lk->filename, ".lock"); During the call to resolve_symlink(), lk->filename contained the name of the file that was being locked, not the name of the lockfile. If a signal were raised during that interval, then the signal handler would have deleted the valuable file! We could probably continue to use the filename field to encode the state by being careful to write characters 1..N-1 of the filename first, and then overwrite the NUL at filename[0] with the first character of the filename, but that would be awkward and error-prone. So, instead of using the filename field to determine whether the lock_file object is active, add a new field "lock_file::active" for this purpose. Be careful to set this field only when filename really contains the name of a file that should be deleted on cleanup. Helped-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2014-10-01 18:28:27 +08:00
if (lk->active)
die("BUG: cannot lock_file(\"%s\") using active struct lock_file",
path);
if (!lk->on_list) {
/* Initialize *lk and add it to lock_file_list: */
lk->fd = -1;
lk->fp = NULL;
lockfile: avoid transitory invalid states Because remove_lock_file() can be called any time by the signal handler, it is important that any lock_file objects that are in the lock_file_list are always in a valid state. And since lock_file objects are often reused (but are never removed from lock_file_list), that means we have to be careful whenever mutating a lock_file object to always keep it in a well-defined state. This was formerly not the case, because part of the state was encoded by setting lk->filename to the empty string vs. a valid filename. It is wrong to assume that this string can be updated atomically; for example, even strcpy(lk->filename, value) is unsafe. But the old code was even more reckless; for example, strcpy(lk->filename, path); if (!(flags & LOCK_NODEREF)) resolve_symlink(lk->filename, max_path_len); strcat(lk->filename, ".lock"); During the call to resolve_symlink(), lk->filename contained the name of the file that was being locked, not the name of the lockfile. If a signal were raised during that interval, then the signal handler would have deleted the valuable file! We could probably continue to use the filename field to encode the state by being careful to write characters 1..N-1 of the filename first, and then overwrite the NUL at filename[0] with the first character of the filename, but that would be awkward and error-prone. So, instead of using the filename field to determine whether the lock_file object is active, add a new field "lock_file::active" for this purpose. Be careful to set this field only when filename really contains the name of a file that should be deleted on cleanup. Helped-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2014-10-01 18:28:27 +08:00
lk->active = 0;
lk->owner = 0;
strbuf_init(&lk->filename, pathlen + LOCK_SUFFIX_LEN);
lk->next = lock_file_list;
lock_file_list = lk;
lk->on_list = 1;
} else if (lk->filename.len) {
/* This shouldn't happen, but better safe than sorry. */
die("BUG: lock_file(\"%s\") called with improperly-reset lock_file object",
path);
}
if (flags & LOCK_NO_DEREF) {
strbuf_add_absolute_path(&lk->filename, path);
} else {
struct strbuf resolved_path = STRBUF_INIT;
strbuf_add(&resolved_path, path, pathlen);
resolve_symlink(&resolved_path);
strbuf_add_absolute_path(&lk->filename, resolved_path.buf);
strbuf_release(&resolved_path);
}
strbuf_addstr(&lk->filename, LOCK_SUFFIX);
lk->fd = open(lk->filename.buf, O_RDWR | O_CREAT | O_EXCL, 0666);
if (lk->fd < 0) {
strbuf_reset(&lk->filename);
return -1;
}
lk->owner = getpid();
lockfile: avoid transitory invalid states Because remove_lock_file() can be called any time by the signal handler, it is important that any lock_file objects that are in the lock_file_list are always in a valid state. And since lock_file objects are often reused (but are never removed from lock_file_list), that means we have to be careful whenever mutating a lock_file object to always keep it in a well-defined state. This was formerly not the case, because part of the state was encoded by setting lk->filename to the empty string vs. a valid filename. It is wrong to assume that this string can be updated atomically; for example, even strcpy(lk->filename, value) is unsafe. But the old code was even more reckless; for example, strcpy(lk->filename, path); if (!(flags & LOCK_NODEREF)) resolve_symlink(lk->filename, max_path_len); strcat(lk->filename, ".lock"); During the call to resolve_symlink(), lk->filename contained the name of the file that was being locked, not the name of the lockfile. If a signal were raised during that interval, then the signal handler would have deleted the valuable file! We could probably continue to use the filename field to encode the state by being careful to write characters 1..N-1 of the filename first, and then overwrite the NUL at filename[0] with the first character of the filename, but that would be awkward and error-prone. So, instead of using the filename field to determine whether the lock_file object is active, add a new field "lock_file::active" for this purpose. Be careful to set this field only when filename really contains the name of a file that should be deleted on cleanup. Helped-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2014-10-01 18:28:27 +08:00
lk->active = 1;
if (adjust_shared_perm(lk->filename.buf)) {
int save_errno = errno;
error("cannot fix permission bits on %s", lk->filename.buf);
rollback_lock_file(lk);
errno = save_errno;
return -1;
}
return lk->fd;
}
/*
* Constants defining the gaps between attempts to lock a file. The
* first backoff period is approximately INITIAL_BACKOFF_MS
* milliseconds. The longest backoff period is approximately
* (BACKOFF_MAX_MULTIPLIER * INITIAL_BACKOFF_MS) milliseconds.
*/
#define INITIAL_BACKOFF_MS 1L
#define BACKOFF_MAX_MULTIPLIER 1000
/*
* Try locking path, retrying with quadratic backoff for at least
* timeout_ms milliseconds. If timeout_ms is 0, try locking the file
* exactly once. If timeout_ms is -1, try indefinitely.
*/
static int lock_file_timeout(struct lock_file *lk, const char *path,
int flags, long timeout_ms)
{
int n = 1;
int multiplier = 1;
long remaining_ms = 0;
static int random_initialized = 0;
if (timeout_ms == 0)
return lock_file(lk, path, flags);
if (!random_initialized) {
srand((unsigned int)getpid());
random_initialized = 1;
}
if (timeout_ms > 0)
remaining_ms = timeout_ms;
while (1) {
long backoff_ms, wait_ms;
int fd;
fd = lock_file(lk, path, flags);
if (fd >= 0)
return fd; /* success */
else if (errno != EEXIST)
return -1; /* failure other than lock held */
else if (timeout_ms > 0 && remaining_ms <= 0)
return -1; /* failure due to timeout */
backoff_ms = multiplier * INITIAL_BACKOFF_MS;
/* back off for between 0.75*backoff_ms and 1.25*backoff_ms */
wait_ms = (750 + rand() % 500) * backoff_ms / 1000;
sleep_millisec(wait_ms);
remaining_ms -= wait_ms;
/* Recursion: (n+1)^2 = n^2 + 2n + 1 */
multiplier += 2*n + 1;
if (multiplier > BACKOFF_MAX_MULTIPLIER)
multiplier = BACKOFF_MAX_MULTIPLIER;
else
n++;
}
}
void unable_to_lock_message(const char *path, int err, struct strbuf *buf)
{
if (err == EEXIST) {
strbuf_addf(buf, "Unable to create '%s.lock': %s.\n\n"
"If no other git process is currently running, this probably means a\n"
"git process crashed in this repository earlier. Make sure no other git\n"
"process is running and remove the file manually to continue.",
absolute_path(path), strerror(err));
} else
strbuf_addf(buf, "Unable to create '%s.lock': %s",
absolute_path(path), strerror(err));
}
NORETURN void unable_to_lock_die(const char *path, int err)
{
struct strbuf buf = STRBUF_INIT;
unable_to_lock_message(path, err, &buf);
die("%s", buf.buf);
}
/* This should return a meaningful errno on failure */
int hold_lock_file_for_update_timeout(struct lock_file *lk, const char *path,
int flags, long timeout_ms)
{
int fd = lock_file_timeout(lk, path, flags, timeout_ms);
if (fd < 0 && (flags & LOCK_DIE_ON_ERROR))
unable_to_lock_die(path, errno);
return fd;
}
int hold_lock_file_for_append(struct lock_file *lk, const char *path, int flags)
{
int fd, orig_fd;
fd = lock_file(lk, path, flags);
if (fd < 0) {
if (flags & LOCK_DIE_ON_ERROR)
unable_to_lock_die(path, errno);
return fd;
}
orig_fd = open(path, O_RDONLY);
if (orig_fd < 0) {
if (errno != ENOENT) {
int save_errno = errno;
if (flags & LOCK_DIE_ON_ERROR)
die("cannot open '%s' for copying", path);
rollback_lock_file(lk);
error("cannot open '%s' for copying", path);
errno = save_errno;
return -1;
}
} else if (copy_fd(orig_fd, fd)) {
int save_errno = errno;
if (flags & LOCK_DIE_ON_ERROR)
exit(128);
close(orig_fd);
rollback_lock_file(lk);
errno = save_errno;
return -1;
} else {
close(orig_fd);
}
return fd;
}
FILE *fdopen_lock_file(struct lock_file *lk, const char *mode)
{
if (!lk->active)
die("BUG: fdopen_lock_file() called for unlocked object");
if (lk->fp)
die("BUG: fdopen_lock_file() called twice for file '%s'", lk->filename.buf);
lk->fp = fdopen(lk->fd, mode);
return lk->fp;
}
char *get_locked_file_path(struct lock_file *lk)
{
if (!lk->active)
die("BUG: get_locked_file_path() called for unlocked object");
if (lk->filename.len <= LOCK_SUFFIX_LEN)
die("BUG: get_locked_file_path() called for malformed lock object");
return xmemdupz(lk->filename.buf, lk->filename.len - LOCK_SUFFIX_LEN);
}
int close_lock_file(struct lock_file *lk)
{
int fd = lk->fd;
FILE *fp = lk->fp;
int err;
if (fd < 0)
return 0;
lk->fd = -1;
if (fp) {
lk->fp = NULL;
/*
* Note: no short-circuiting here; we want to fclose()
* in any case!
*/
err = ferror(fp) | fclose(fp);
} else {
err = close(fd);
}
if (err) {
int save_errno = errno;
rollback_lock_file(lk);
errno = save_errno;
return -1;
}
return 0;
}
int reopen_lock_file(struct lock_file *lk)
{
if (0 <= lk->fd)
die(_("BUG: reopen a lockfile that is still open"));
lockfile: avoid transitory invalid states Because remove_lock_file() can be called any time by the signal handler, it is important that any lock_file objects that are in the lock_file_list are always in a valid state. And since lock_file objects are often reused (but are never removed from lock_file_list), that means we have to be careful whenever mutating a lock_file object to always keep it in a well-defined state. This was formerly not the case, because part of the state was encoded by setting lk->filename to the empty string vs. a valid filename. It is wrong to assume that this string can be updated atomically; for example, even strcpy(lk->filename, value) is unsafe. But the old code was even more reckless; for example, strcpy(lk->filename, path); if (!(flags & LOCK_NODEREF)) resolve_symlink(lk->filename, max_path_len); strcat(lk->filename, ".lock"); During the call to resolve_symlink(), lk->filename contained the name of the file that was being locked, not the name of the lockfile. If a signal were raised during that interval, then the signal handler would have deleted the valuable file! We could probably continue to use the filename field to encode the state by being careful to write characters 1..N-1 of the filename first, and then overwrite the NUL at filename[0] with the first character of the filename, but that would be awkward and error-prone. So, instead of using the filename field to determine whether the lock_file object is active, add a new field "lock_file::active" for this purpose. Be careful to set this field only when filename really contains the name of a file that should be deleted on cleanup. Helped-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2014-10-01 18:28:27 +08:00
if (!lk->active)
die(_("BUG: reopen a lockfile that has been committed"));
lk->fd = open(lk->filename.buf, O_WRONLY);
return lk->fd;
}
int commit_lock_file_to(struct lock_file *lk, const char *path)
{
lockfile: avoid transitory invalid states Because remove_lock_file() can be called any time by the signal handler, it is important that any lock_file objects that are in the lock_file_list are always in a valid state. And since lock_file objects are often reused (but are never removed from lock_file_list), that means we have to be careful whenever mutating a lock_file object to always keep it in a well-defined state. This was formerly not the case, because part of the state was encoded by setting lk->filename to the empty string vs. a valid filename. It is wrong to assume that this string can be updated atomically; for example, even strcpy(lk->filename, value) is unsafe. But the old code was even more reckless; for example, strcpy(lk->filename, path); if (!(flags & LOCK_NODEREF)) resolve_symlink(lk->filename, max_path_len); strcat(lk->filename, ".lock"); During the call to resolve_symlink(), lk->filename contained the name of the file that was being locked, not the name of the lockfile. If a signal were raised during that interval, then the signal handler would have deleted the valuable file! We could probably continue to use the filename field to encode the state by being careful to write characters 1..N-1 of the filename first, and then overwrite the NUL at filename[0] with the first character of the filename, but that would be awkward and error-prone. So, instead of using the filename field to determine whether the lock_file object is active, add a new field "lock_file::active" for this purpose. Be careful to set this field only when filename really contains the name of a file that should be deleted on cleanup. Helped-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2014-10-01 18:28:27 +08:00
if (!lk->active)
die("BUG: attempt to commit unlocked object to \"%s\"", path);
if (close_lock_file(lk))
return -1;
if (rename(lk->filename.buf, path)) {
int save_errno = errno;
rollback_lock_file(lk);
errno = save_errno;
return -1;
}
lockfile: avoid transitory invalid states Because remove_lock_file() can be called any time by the signal handler, it is important that any lock_file objects that are in the lock_file_list are always in a valid state. And since lock_file objects are often reused (but are never removed from lock_file_list), that means we have to be careful whenever mutating a lock_file object to always keep it in a well-defined state. This was formerly not the case, because part of the state was encoded by setting lk->filename to the empty string vs. a valid filename. It is wrong to assume that this string can be updated atomically; for example, even strcpy(lk->filename, value) is unsafe. But the old code was even more reckless; for example, strcpy(lk->filename, path); if (!(flags & LOCK_NODEREF)) resolve_symlink(lk->filename, max_path_len); strcat(lk->filename, ".lock"); During the call to resolve_symlink(), lk->filename contained the name of the file that was being locked, not the name of the lockfile. If a signal were raised during that interval, then the signal handler would have deleted the valuable file! We could probably continue to use the filename field to encode the state by being careful to write characters 1..N-1 of the filename first, and then overwrite the NUL at filename[0] with the first character of the filename, but that would be awkward and error-prone. So, instead of using the filename field to determine whether the lock_file object is active, add a new field "lock_file::active" for this purpose. Be careful to set this field only when filename really contains the name of a file that should be deleted on cleanup. Helped-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2014-10-01 18:28:27 +08:00
lk->active = 0;
strbuf_reset(&lk->filename);
return 0;
}
int commit_lock_file(struct lock_file *lk)
{
static struct strbuf result_file = STRBUF_INIT;
int err;
if (!lk->active)
die("BUG: attempt to commit unlocked object");
if (lk->filename.len <= LOCK_SUFFIX_LEN ||
strcmp(lk->filename.buf + lk->filename.len - LOCK_SUFFIX_LEN, LOCK_SUFFIX))
die("BUG: lockfile filename corrupt");
/* remove ".lock": */
strbuf_add(&result_file, lk->filename.buf,
lk->filename.len - LOCK_SUFFIX_LEN);
err = commit_lock_file_to(lk, result_file.buf);
strbuf_reset(&result_file);
return err;
}
void rollback_lock_file(struct lock_file *lk)
{
lockfile: avoid transitory invalid states Because remove_lock_file() can be called any time by the signal handler, it is important that any lock_file objects that are in the lock_file_list are always in a valid state. And since lock_file objects are often reused (but are never removed from lock_file_list), that means we have to be careful whenever mutating a lock_file object to always keep it in a well-defined state. This was formerly not the case, because part of the state was encoded by setting lk->filename to the empty string vs. a valid filename. It is wrong to assume that this string can be updated atomically; for example, even strcpy(lk->filename, value) is unsafe. But the old code was even more reckless; for example, strcpy(lk->filename, path); if (!(flags & LOCK_NODEREF)) resolve_symlink(lk->filename, max_path_len); strcat(lk->filename, ".lock"); During the call to resolve_symlink(), lk->filename contained the name of the file that was being locked, not the name of the lockfile. If a signal were raised during that interval, then the signal handler would have deleted the valuable file! We could probably continue to use the filename field to encode the state by being careful to write characters 1..N-1 of the filename first, and then overwrite the NUL at filename[0] with the first character of the filename, but that would be awkward and error-prone. So, instead of using the filename field to determine whether the lock_file object is active, add a new field "lock_file::active" for this purpose. Be careful to set this field only when filename really contains the name of a file that should be deleted on cleanup. Helped-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2014-10-01 18:28:27 +08:00
if (!lk->active)
return;
if (!close_lock_file(lk)) {
unlink_or_warn(lk->filename.buf);
lockfile: avoid transitory invalid states Because remove_lock_file() can be called any time by the signal handler, it is important that any lock_file objects that are in the lock_file_list are always in a valid state. And since lock_file objects are often reused (but are never removed from lock_file_list), that means we have to be careful whenever mutating a lock_file object to always keep it in a well-defined state. This was formerly not the case, because part of the state was encoded by setting lk->filename to the empty string vs. a valid filename. It is wrong to assume that this string can be updated atomically; for example, even strcpy(lk->filename, value) is unsafe. But the old code was even more reckless; for example, strcpy(lk->filename, path); if (!(flags & LOCK_NODEREF)) resolve_symlink(lk->filename, max_path_len); strcat(lk->filename, ".lock"); During the call to resolve_symlink(), lk->filename contained the name of the file that was being locked, not the name of the lockfile. If a signal were raised during that interval, then the signal handler would have deleted the valuable file! We could probably continue to use the filename field to encode the state by being careful to write characters 1..N-1 of the filename first, and then overwrite the NUL at filename[0] with the first character of the filename, but that would be awkward and error-prone. So, instead of using the filename field to determine whether the lock_file object is active, add a new field "lock_file::active" for this purpose. Be careful to set this field only when filename really contains the name of a file that should be deleted on cleanup. Helped-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2014-10-01 18:28:27 +08:00
lk->active = 0;
strbuf_reset(&lk->filename);
}
}