Daniel P. Berrangé <berra...@redhat.com> writes: > On Tue, Aug 25, 2020 at 05:14:21PM +0200, Markus Armbruster wrote: >> Daniel P. Berrangé <berra...@redhat.com> writes: >> >> > Instead of relying on the limited information from errno, we can now >> > also provide detailed error messages. >> >> The more detailed error messages are currently always ignored, but the >> next patches will fix that. >> >> > Signed-off-by: Daniel P. Berrangé <berra...@redhat.com> >> > --- >> > util/osdep.c | 21 +++++++++++++++++++-- >> > 1 file changed, 19 insertions(+), 2 deletions(-) >> > >> > diff --git a/util/osdep.c b/util/osdep.c >> > index 9ff92551e7..9c7118d3cb 100644 >> > --- a/util/osdep.c >> > +++ b/util/osdep.c >> > @@ -284,7 +284,7 @@ int qemu_lock_fd_test(int fd, int64_t start, int64_t >> > len, bool exclusive) >> > * Opens a file with FD_CLOEXEC set >> > */ >> > static int >> > -qemu_open_internal(const char *name, int flags, mode_t mode) >> > +qemu_open_internal(const char *name, int flags, mode_t mode, Error **errp) >> > { >> > int ret; >> > >> > @@ -298,24 +298,31 @@ qemu_open_internal(const char *name, int flags, >> > mode_t mode) >> > >> > fdset_id = qemu_parse_fdset(fdset_id_str); >> > if (fdset_id == -1) { >> > + error_setg(errp, "Could not parse fdset %s", name); >> > errno = EINVAL; >> > return -1; >> > } >> > >> > fd = monitor_fdset_get_fd(fdset_id, flags); >> > if (fd < 0) { >> > + error_setg_errno(errp, -fd, "Could not acquire FD for %s >> > flags %x", >> > + name, flags); >> > errno = -fd; >> > return -1; >> > } >> > >> > dupfd = qemu_dup_flags(fd, flags); >> > if (dupfd == -1) { >> > + error_setg_errno(errp, errno, "Could not dup FD for %s flags >> > %x", >> > + name, flags); >> > return -1; >> > } >> > >> > ret = monitor_fdset_dup_fd_add(fdset_id, dupfd); >> > if (ret == -1) { >> > close(dupfd); >> > + error_setg(errp, "Could not save FD for %s flags %x", >> > + name, flags); >> >> Can this happen? > > Well there's code in monitor_fdset_dup_fd_add that can return -1.
It fails when * @fdset_id contains @dupfd @dupfd is a fresh file descriptor. If @fdset_id already contains it, it's stale there. That would be a programming error. Recommend to assert. * @fdset_id is not in @mon_fdsets monitor_fdset_get_fd() fails the same way. monitor_fdset_dup_fd_add() can fail that way after monitor_fdset_get_fd() succeed only if the fd set went away between the two. Could that happen? Would it be safe? This is the only user of monitor_fdset_dup_fd_add(). Why not remove the awkward failure mode by making monitor_fdset_dup_fd_add() dup the fd and add? >> > errno = EINVAL; >> > return -1; >> > } >> > @@ -336,6 +343,16 @@ qemu_open_internal(const char *name, int flags, >> > mode_t mode) >> > } >> > #endif /* ! O_CLOEXEC */ >> > >> > + if (ret == -1) { >> > + const char *action = "open"; >> > + if (flags & O_CREAT) { >> > + action = "create"; >> > + } >> > + error_setg_errno(errp, errno, "Could not %s '%s' flags 0x%x", >> > + action, name, flags); >> >> Not a good user experience: >> >> Could not open '/etc/shadow' flags 0x0: Permission denied >> >> Better: >> >> Could not open '/etc/shadow' for reading: Permission denied >> >> Are you sure flags other than the access mode (O_RDONLY, O_WRONLY, >> O_RDWR) must be included in the error message? > > It was the flags other than access mode that I was thinking were > more important to log. I'm ambivalent htough, so can drop the > flags if it is thought to be overkill. Hexadecimal flags are borderline useless even for developers: to make sense of them, you have to grep -R /usr/include/. For mere mortals, they are confusing in addition to useless. >> If you must report flags in hexadecimal, then please reporting them more >> consistently. Right now you have >> >> for %s flags 0x%x >> '%s' flags %x >> >> Perhaps '%s' with flags 0x%x >> >> > + } >> > + >> > + >> > return ret; >> > } >> > >> > @@ -352,7 +369,7 @@ int qemu_open_old(const char *name, int flags, ...) >> > } >> > va_end(ap); >> > >> > - ret = qemu_open_internal(name, flags, mode); >> > + ret = qemu_open_internal(name, flags, mode, NULL); >> > >> > #ifdef O_DIRECT >> > if (ret == -1 && errno == EINVAL && (flags & O_DIRECT)) { >> >> > > Regards, > Daniel