* - When a device is added to the container, we add it to the metadata
* as a spare.
*
- * - assist with activating spares by opening relevant sysfs file.
+ * - Deal with degraded array
+ * We only do this when first noticing the array is degraded.
+ * This can be when we first see the array, when sync completes or
+ * when recovery completes.
+ *
+ * Check if number of failed devices suggests recovery is needed, and
+ * skip if not.
+ * Ask metadata to allocate a spare device
+ * Add device as not in_sync and give a role
+ * Update metadata.
+ * Open sysfs files and pass to monitor.
+ * Make sure that monitor Starts recovery....
*
* - Pass on metadata updates from external programs such as
* mdadm creating a new array.
#include "mdadm.h"
#include "mdmon.h"
#include <sys/socket.h>
+#include <signal.h>
+
+static void close_aa(struct active_array *aa)
+{
+ struct mdinfo *d;
+ for (d = aa->info.devs; d; d = d->next)
+ close(d->state_fd);
+
+ close(aa->action_fd);
+ close(aa->info.state_fd);
+ close(aa->resync_start_fd);
+}
static void free_aa(struct active_array *aa)
{
- /* Note that this doesn't close fds, as they may be in used
- * by a clone. Use close_aa for that.
+ /* Note that this doesn't close fds if they are being used
+ * by a clone. ->container will be set for a clone
*/
+ dprintf("%s: devnum: %d\n", __func__, aa->devnum);
+ if (!aa->container)
+ close_aa(aa);
while (aa->info.devs) {
struct mdinfo *d = aa->info.devs;
aa->info.devs = d->next;
free(aa);
}
+static struct active_array *duplicate_aa(struct active_array *aa)
+{
+ struct active_array *newa = malloc(sizeof(*newa));
+ struct mdinfo **dp1, **dp2;
+
+ *newa = *aa;
+ newa->next = NULL;
+ newa->replaces = NULL;
+ newa->info.next = NULL;
+
+ dp2 = &newa->info.devs;
+
+ for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) {
+ struct mdinfo *d;
+ if ((*dp1)->state_fd < 0)
+ continue;
+
+ d = malloc(sizeof(*d));
+ *d = **dp1;
+ *dp2 = d;
+ dp2 = & d->next;
+ }
+ *dp2 = NULL;
+
+ return newa;
+}
+
+static void write_wakeup(struct supertype *c)
+{
+ static struct md_generic_cmd cmd = { .action = md_action_ping_monitor };
+ int err;
+
+ active_cmd = &cmd;
+
+ /* send the monitor thread a pointer to the ping action */
+ write(c->mgr_pipe[1], &err, 1);
+ read(c->mon_pipe[0], &err, 1);
+}
+
+static void remove_old(void)
+{
+ if (discard_this) {
+ discard_this->next = NULL;
+ free_aa(discard_this);
+ if (pending_discard == discard_this)
+ pending_discard = NULL;
+ discard_this = NULL;
+ }
+}
+
static void replace_array(struct supertype *container,
struct active_array *old,
struct active_array *new)
* and put it on 'discard_this'. We take it from there
* and discard it.
*/
-
+ remove_old();
while (pending_discard) {
+ write_wakeup(container);
while (discard_this == NULL)
sleep(1);
- if (discard_this != pending_discard)
- abort();
- discard_this->next = NULL;
- free_aa(discard_this);
- discard_this = NULL;
- pending_discard = NULL;
+ remove_old();
}
pending_discard = old;
new->replaces = old;
new->next = container->arrays;
container->arrays = new;
+ write_wakeup(container);
+}
+
+struct metadata_update *update_queue = NULL;
+struct metadata_update *update_queue_handled = NULL;
+struct metadata_update *update_queue_pending = NULL;
+
+void check_update_queue(struct supertype *container)
+{
+ while (update_queue_handled) {
+ struct metadata_update *this = update_queue_handled;
+ update_queue_handled = this->next;
+// free(this->buf);
+ free(this);
+ }
+ if (update_queue == NULL &&
+ update_queue_pending) {
+ update_queue = update_queue_pending;
+ update_queue_pending = NULL;
+ write_wakeup(container);
+ }
+}
+
+static void queue_metadata_update(struct metadata_update *mu)
+{
+ struct metadata_update **qp;
+
+ qp = &update_queue_pending;
+ while (*qp)
+ qp = & ((*qp)->next);
+ *qp = mu;
}
+void wait_update_handled(void)
+{
+ /* Wait for any pending update to be handled by monitor.
+ * i.e. wait until update_queue is NULL
+ */
+ while (update_queue)
+ usleep(100 * 1000);
+}
static void manage_container(struct mdstat_ent *mdstat,
struct supertype *container)
* array ignoring any metadata on it.
* FIXME should we look for compatible metadata and take hints
* about spare assignment.... probably not.
- *
*/
if (mdstat->devcnt != container->devcnt) {
/* read /sys/block/NAME/md/dev-??/block/dev to find out
* being requested.
* Unfortunately decreases in raid_disks don't show up in
* mdstat until the reshape completes FIXME.
+ *
+ * Actually, we also want to handle degraded arrays here by
+ * trying to find and assign a spare.
+ * We do that whenever the monitor tells us too.
*/
// FIXME
a->info.array.raid_disks = mdstat->raid_disks;
a->info.array.chunk_size = mdstat->chunk_size;
// MORE
-}
+ if (a->check_degraded) {
+ struct metadata_update *updates = NULL;
+ struct mdinfo *newdev;
+ struct active_array *newa;
+ wait_update_handled();
+ a->check_degraded = 0;
-static void write_wakeup(struct supertype *c)
-{
- write(c->pipe[1], "PING", 4);
+ /* The array may not be degraded, this is just a good time
+ * to check.
+ */
+ newdev = a->container->ss->activate_spare(a, &updates);
+ if (newdev) {
+ struct mdinfo *d;
+ /* Cool, we can add a device or several. */
+ newa = duplicate_aa(a);
+ /* suspend recovery - maybe not needed */
+
+ /* Add device to array and set offset/size/slot.
+ * and open files for each newdev */
+ for (d = newdev; d ; d = d->next) {
+ struct mdinfo *newd;
+ if (sysfs_add_disk(&newa->info, d))
+ continue;
+ newd = newa->info.devs;
+ newd->state_fd = sysfs_open(a->devnum,
+ newd->sys_name,
+ "state");
+ newd->prev_state
+ = read_dev_state(newd->state_fd);
+ newd->curr_state = newd->prev_state;
+ }
+ queue_metadata_update(updates);
+ replace_array(a->container, a, newa);
+ sysfs_set_str(&a->info, NULL, "sync_action", "repair");
+ }
+ }
}
static void manage_new(struct mdstat_ent *mdstat,
- struct supertype *container)
+ struct supertype *container,
+ struct active_array *victim)
{
/* A new array has appeared in this container.
* Hopefully it is already recorded in the metadata.
struct active_array *new;
struct mdinfo *mdi, *di;
- char *n;
- int inst;
+ char *inst;
int i;
new = malloc(sizeof(*new));
+ memset(new, 0, sizeof(*new));
+
new->devnum = mdstat->devnum;
+ strcpy(new->info.sys_name, devnum2devname(new->devnum));
new->prev_state = new->curr_state = new->next_state = inactive;
new->prev_action= new->curr_action= new->next_action= idle;
new->container = container;
- n = &mdstat->metadata_version[10+strlen(container->devname)+1];
- inst = atoi(n);
- if (inst < 0)
- abort();//FIXME
+ inst = &mdstat->metadata_version[10+strlen(container->devname)+1];
mdi = sysfs_read(-1, new->devnum,
GET_LEVEL|GET_CHUNK|GET_DISKS|
* Mark it to be ignored by setting container to NULL
*/
new->container = NULL;
- replace_array(container, NULL, new);
+ replace_array(container, victim, new);
return;
}
if (di) {
memcpy(newd, di, sizeof(*newd));
- sprintf(newd->sys_name, "rd%d", i);
-
newd->state_fd = sysfs_open(new->devnum,
newd->sys_name,
"state");
newd->prev_state = read_dev_state(newd->state_fd);
- newd->curr_state = newd->curr_state;
+ newd->curr_state = newd->prev_state;
} else {
newd->state_fd = -1;
+ newd->disk.raid_disk = i;
+ newd->prev_state = DS_REMOVE;
+ newd->curr_state = DS_REMOVE;
}
+ sprintf(newd->sys_name, "rd%d", i);
newd->next = new->info.devs;
new->info.devs = newd;
}
new->action_fd = sysfs_open(new->devnum, NULL, "sync_action");
new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state");
new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start");
- new->sync_pos_fd = sysfs_open(new->devnum, NULL, "sync_completed");
- new->sync_pos = 0;
+ new->resync_start = 0;
+ dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst),
+ new->action_fd, new->info.state_fd);
+ sysfs_free(mdi);
// finds and compares.
if (container->ss->open_new(container, new, inst) < 0) {
// FIXME close all those files
new->container = NULL;
- replace_array(container, NULL, new);
+ replace_array(container, victim, new);
return;
}
- replace_array(container, NULL, new);
- write_wakeup(container);
+ replace_array(container, victim, new);
return;
}
-void manage(struct mdstat_ent *mdstat, struct active_array *aa,
- struct supertype *container)
+void manage(struct mdstat_ent *mdstat, struct supertype *container)
{
/* We have just read mdstat and need to compare it with
* the known active arrays.
/* Not for this array */
continue;
/* Looks like a member of this container */
- for (a = aa; a; a = a->next) {
+ for (a = container->arrays; a; a = a->next) {
if (mdstat->devnum == a->devnum) {
if (a->container)
manage_member(mdstat, a);
break;
}
}
- if (a == NULL)
- manage_new(mdstat, container);
+ if (a == NULL || !a->container)
+ manage_new(mdstat, container, a);
}
}
-void read_sock(int pfd)
+static int handle_message(struct supertype *container, struct md_message *msg)
+{
+ int err;
+ struct md_generic_cmd *cmd = msg->buf;
+
+ if (!cmd)
+ return 0;
+
+ switch (cmd->action) {
+ case md_action_remove_device:
+
+ /* forward to the monitor */
+ active_cmd = cmd;
+ write(container->mgr_pipe[1], &err, 1);
+ read(container->mon_pipe[0], &err, 1);
+ return err;
+
+ default:
+ return -1;
+ }
+}
+
+void read_sock(struct supertype *container)
{
int fd;
+ struct md_message msg;
+ int terminate = 0;
+ long fl;
+ int tmo = 3; /* 3 second timeout before hanging up the socket */
- // FIXME set non-blocking
- fd = accept(pfd, NULL, NULL);
+ fd = accept(container->sock, NULL, NULL);
if (fd < 0)
return;
- // FIXME do something useful
+
+ fl = fcntl(fd, F_GETFL, 0);
+ fl |= O_NONBLOCK;
+ fcntl(fd, F_SETFL, fl);
+
+ do {
+ int err;
+
+ msg.buf = NULL;
+
+ /* read and validate the message */
+ if (receive_message(fd, &msg, tmo) == 0) {
+ err = handle_message(container, &msg);
+ if (!err)
+ ack(fd, msg.seq, tmo);
+ else
+ nack(fd, err, tmo);
+ } else {
+ terminate = 1;
+ nack(fd, -1, tmo);
+ }
+
+ if (msg.buf)
+ free(msg.buf);
+ } while (!terminate);
+
close(fd);
}
+
+static int woke = 0;
+void wake_me(int sig)
+{
+ woke = 1;
+}
+
+int exit_now = 0;
+int manager_ready = 0;
void do_manager(struct supertype *container)
{
struct mdstat_ent *mdstat;
+ sigset_t block, orig;
+
+ sigemptyset(&block);
+ sigaddset(&block, SIGUSR1);
+
+ signal(SIGUSR1, wake_me);
do {
+ woke = 0;
+
+ if (exit_now)
+ exit(0);
+
mdstat = mdstat_read(1, 0);
- manage(mdstat, array_list, container);
+ manage(mdstat, container);
+
+ read_sock(container);
+
+ free_mdstat(mdstat);
+
+ remove_old();
- read_sock(container->sock);
+ check_update_queue(container);
- mdstat_wait_fd(container->sock);
+ manager_ready = 1;
+ sigprocmask(SIG_SETMASK, &block, &orig);
+ if (woke == 0)
+ mdstat_wait_fd(container->sock, &orig);
+ sigprocmask(SIG_SETMASK, &orig, NULL);
} while(1);
}