#include "mdmon.h"
#include <sys/select.h>
-
+#include <signal.h>
static char *array_states[] = {
"clear", "inactive", "suspended", "readonly", "read-auto",
return rv;
}
+static void signal_manager(void)
+{
+ kill(getpid(), SIGUSR1);
+}
/* Monitor a set of active md arrays - all of which share the
* same metadata - and respond to events that require
* detected by rd-N/state reporting "faulty"
* mark device as 'failed' in metadata, let the kernel release the
* device by writing '-blocked' to rd/state, and finally write 'remove' to
- * rd/state
+ * rd/state. Before a disk can be replaced it must be failed and removed
+ * from all container members, this will be preemptive for the other
+ * arrays... safe?
*
* sync completes
* sync_action was 'resync' and becomes 'idle' and resync_start becomes
deactivate = 1;
}
if (a->curr_state == write_pending) {
- a->container->ss->mark_dirty(a);
+ a->container->ss->mark_clean(a, 0);
a->next_state = active;
}
if (a->curr_state == active_idle) {
/* Set array to 'clean' FIRST, then
- * a->ss->mark_clean(a);
+ * a->ss->mark_clean(a, ~0ULL);
* just ignore for now.
*/
}
if (a->resync_start == ~0ULL)
a->next_state = read_auto; /* array is clean */
else {
- a->container->ss->mark_dirty(a);
+ a->container->ss->mark_clean(a, 0);
a->next_state = active;
}
}
if (a->curr_action == idle &&
a->prev_action == resync) {
- /* check resync_start to see if it is 'max' */
- get_resync_start(a);
- a->container->ss->mark_sync(a, a->resync_start);
+ /* A resync has finished. The endpoint is recorded in
+ * 'sync_start'. We don't update the metadata
+ * until the array goes inactive or readonly though.
+ * Just check if we need to fiddle spares.
+ */
check_degraded = 1;
}
return 1;
}
-static int wait_and_act(struct active_array *aa, int pfd, int nowait)
+static struct mdinfo *
+find_device(struct active_array *a, int major, int minor)
+{
+ struct mdinfo *mdi;
+
+ for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
+ if (mdi->disk.major == major && mdi->disk.minor == minor)
+ return mdi;
+
+ return NULL;
+}
+
+static void reconcile_failed(struct active_array *aa, struct mdinfo *failed)
+{
+ struct active_array *a;
+ struct mdinfo *victim;
+
+ for (a = aa; a; a = a->next) {
+ if (!a->container)
+ continue;
+ victim = find_device(a, failed->disk.major, failed->disk.minor);
+ if (!victim)
+ continue;
+
+ if (!(victim->curr_state & DS_FAULTY))
+ write_attr("faulty", victim->state_fd);
+ }
+}
+
+static int handle_remove_device(struct md_remove_device_cmd *cmd, struct active_array *aa)
+{
+ struct active_array *a;
+ struct mdinfo *victim;
+ int rv;
+
+ /* scan all arrays for the given device, if ->state_fd is closed (-1)
+ * in all cases then mark the disk as removed in the metadata.
+ * Otherwise reply that it is busy.
+ */
+
+ /* pass1 check that it is not in use anywhere */
+ /* note: we are safe from re-adds as long as the device exists in the
+ * container
+ */
+ for (a = aa; a; a = a->next) {
+ if (!a->container)
+ continue;
+ victim = find_device(a, major(cmd->rdev), minor(cmd->rdev));
+ if (!victim)
+ continue;
+ if (victim->state_fd > 0)
+ return -EBUSY;
+ }
+
+ /* pass2 schedule and process removal per array */
+ for (a = aa; a; a = a->next) {
+ if (!a->container)
+ continue;
+ victim = find_device(a, major(cmd->rdev), minor(cmd->rdev));
+ if (!victim)
+ continue;
+ victim->curr_state |= DS_REMOVE;
+ rv = read_and_act(a);
+ if (rv < 0)
+ return rv;
+ }
+
+ return 0;
+}
+
+static int handle_pipe(struct md_generic_cmd *cmd, struct active_array *aa)
+{
+ switch (cmd->action) {
+ case md_action_ping_monitor:
+ return 0;
+ case md_action_remove_device:
+ return handle_remove_device((void *) cmd, aa);
+ }
+
+ return -1;
+}
+
+static int wait_and_act(struct active_array **aap, int pfd,
+ int monfd, int nowait)
{
fd_set rfds;
int maxfd = 0;
- struct active_array *a;
+ struct active_array *a, **ap;
int rv;
+ struct mdinfo *mdi;
FD_ZERO(&rfds);
add_fd(&rfds, &maxfd, pfd);
- for (a = aa ; a ; a = a->next) {
- struct mdinfo *mdi;
-
- /* once an array has been deactivated only the manager
- * thread can make us care about it again
+ for (ap = aap ; *ap ;) {
+ a = *ap;
+ /* once an array has been deactivated we want to
+ * ask the manager to discard it.
*/
- if (!a->container)
+ if (!a->container) {
+ if (discard_this) {
+ ap = &(*ap)->next;
+ continue;
+ }
+ *ap = a->next;
+ a->next = NULL;
+ discard_this = a;
+ signal_manager();
continue;
+ }
add_fd(&rfds, &maxfd, a->info.state_fd);
add_fd(&rfds, &maxfd, a->action_fd);
for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
add_fd(&rfds, &maxfd, mdi->state_fd);
+
+ ap = &(*ap)->next;
}
if (!nowait) {
return rv;
if (FD_ISSET(pfd, &rfds)) {
- char buf[4];
- read(pfd, buf, 4);
- ; // FIXME read from the pipe
+ int err = -1;
+
+ if (read(pfd, &err, 1) > 0)
+ err = handle_pipe(active_cmd, *aap);
+ write(monfd, &err, 1);
}
}
- for (a = aa; a ; a = a->next) {
+ for (a = *aap; a ; a = a->next) {
if (a->replaces && !discard_this) {
struct active_array **ap;
for (ap = &a->next; *ap && *ap != a->replaces;
*ap = (*ap)->next;
discard_this = a->replaces;
a->replaces = NULL;
+ signal_manager();
}
if (a->container)
rv += read_and_act(a);
}
+
+ /* propagate failures across container members */
+ for (a = *aap; a ; a = a->next) {
+ if (!a->container)
+ continue;
+ for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
+ if (mdi->curr_state & DS_FAULTY)
+ reconcile_failed(*aap, mdi);
+ }
+
return rv;
}
int rv;
int first = 1;
do {
- rv = wait_and_act(container->arrays, container->pipe[0], first);
+ rv = wait_and_act(&container->arrays, container->mgr_pipe[0],
+ container->mon_pipe[1], first);
first = 0;
} while (rv >= 0);
}
projects / thirdparty / mdadm.git / blobdiff