When /dev/loop-control exists and *device is empty, the mounting fails due to concurrent contention. Code Execution Flow: try = xasprintf(LOOP_FORMAT, i); for (i = 0; i <= 0xfffff; i++) { // The value of "try" is not changed. ... lfd = rc = open(try, mode); ... rc = repeat_on_eagain(ioctl(lfd, BB_LOOP_GET_STATUS, &loopinfo)); // Because of race, the value of "rc" is 0. and the value of "try" is not changed ... close(lfd); }
add/remove: 0/0 grow/shrink: 1/0 up/down: 5/0 (5) Function old new delta set_loop 773 778 +5 Fixes: 4bc59a4cf ("mount: fix a race when a free loop device is snatched under us by another mount") Fiexe: 3b69ba799 ("mount,losetup: use /dev/loop-control is it exists") Signed-off-by: Xiaoming Ni <nixiaom...@huawei.com> --- libbb/loop.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/libbb/loop.c b/libbb/loop.c index cb8fa2442..845565d7b 100644 --- a/libbb/loop.c +++ b/libbb/loop.c @@ -218,8 +218,13 @@ int FAST_FUNC set_loop(char **device, const char *file, unsigned long long offse } /* failure, undo LOOP_SET_FD */ ioctl(lfd, LOOP_CLR_FD, 0); // actually, 0 param is unnecessary + } else { + if (rc == 0 && *device == NULL && try != dev) { + free(try); + close(lfd); + goto get_free_loopN; + } } - /* else: device is not free (rc == 0) or error other than ENXIO */ close_and_try_next_loopN: close(lfd); try_next_loopN: -- 2.27.0 _______________________________________________ busybox mailing list busybox@busybox.net http://lists.busybox.net/mailman/listinfo/busybox