#include "mdadm.h"
#include "mdmon.h"
-
+#include <sys/syscall.h>
#include <sys/select.h>
-
+#include <signal.h>
static char *array_states[] = {
"clear", "inactive", "suspended", "readonly", "read-auto",
return n;
}
-static int get_sync_pos(struct active_array *a)
-{
- char buf[30];
- int n;
- n = read_attr(buf, 30, a->sync_pos_fd);
- if (n <= 0)
- return n;
-
- if (strncmp(buf, "max", 3) == 0) {
- a->sync_pos = ~(unsigned long long)0;
- return 1;
- }
- a->sync_pos = strtoull(buf, NULL, 10);
- return 1;
-}
-
-static int get_resync_start(struct active_array *a)
+int get_resync_start(struct active_array *a)
{
char buf[30];
int n;
return rv;
}
+static void signal_manager(void)
+{
+ /* tgkill(getpid(), mon_tid, SIGUSR1); */
+ int pid = getpid();
+ syscall(SYS_tgkill, pid, mgr_tid, SIGUSR1);
+}
/* Monitor a set of active md arrays - all of which share the
* same metadata - and respond to events that require
* sync_action was 'resync' and becomes 'idle' and resync_start becomes
* MaxSector
* Notify metadata that sync is complete.
- * "Deal with Degraded"
*
* recovery completes
* sync_action changes from 'recover' to 'idle'
* Check each device state and mark metadata if 'faulty' or 'in_sync'.
- * "Deal with Degraded"
- *
- * 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 for a spare device
- * Add device as not in_sync and give a role
- * Update metadata.
- * Start recovery.
*
* deal with resync
* This only happens on finding a new array... mdadm will have set
static int read_and_act(struct active_array *a)
{
- int check_degraded;
+ int check_degraded = 0;
int deactivate = 0;
struct mdinfo *mdi;
a->curr_action = read_action(a->action_fd);
for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
mdi->next_state = 0;
- if (mdi->state_fd > 0)
+ if (mdi->state_fd >= 0)
mdi->curr_state = read_dev_state(mdi->state_fd);
}
if (a->curr_state <= inactive &&
a->prev_state > inactive) {
/* array has been stopped */
- get_sync_pos(a);
- a->container->ss->mark_clean(a, a->sync_pos);
+ a->container->ss->set_array_state(a, 1);
a->next_state = clear;
deactivate = 1;
}
if (a->curr_state == write_pending) {
- a->container->ss->mark_dirty(a);
+ get_resync_start(a);
+ a->container->ss->set_array_state(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.
*/
}
* readonly ???
*/
get_resync_start(a);
+// printf("Found a readonly array at %llu\n", a->resync_start);
if (a->resync_start == ~0ULL)
a->next_state = read_auto; /* array is clean */
else {
- a->container->ss->mark_dirty(a);
+ a->container->ss->set_array_state(a, 0);
a->next_state = active;
}
}
- if (a->curr_action == idle &&
+ if (!deactivate &&
+ a->curr_action == idle &&
a->prev_action == resync) {
- /* check resync_start to see if it is 'max' */
+ /* 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.
+ */
get_resync_start(a);
- a->container->ss->mark_sync(a, a->resync_start);
+ a->container->ss->set_array_state(a, 0);
check_degraded = 1;
}
- if (a->curr_action == idle &&
+ if (!deactivate &&
+ a->curr_action == idle &&
a->prev_action == recover) {
+ /* A recovery has finished. Some disks may be in sync now,
+ * and the array may no longer be degraded
+ */
for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
a->container->ss->set_disk(a, mdi->disk.raid_disk,
mdi->curr_state);
if (! (mdi->curr_state & DS_INSYNC))
check_degraded = 1;
}
+ a->container->ss->set_array_state(a, 0);
}
-
for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
if (mdi->curr_state & DS_FAULTY) {
a->container->ss->set_disk(a, mdi->disk.raid_disk,
}
}
- if (check_degraded) {
- // FIXME;
- }
-
- a->container->ss->sync_metadata(a);
+ a->container->ss->sync_metadata(a->container);
+ dprintf("%s: update[%d]: (", __func__, a->info.container_member);
/* Effect state changes in the array */
- if (a->next_state != bad_word)
+ if (a->next_state != bad_word) {
+ dprintf(" state:%s", array_states[a->next_state]);
write_attr(array_states[a->next_state], a->info.state_fd);
- if (a->next_action != bad_action)
+ }
+ if (a->next_action != bad_action) {
write_attr(sync_actions[a->next_action], a->action_fd);
+ dprintf(" action:%s", array_states[a->next_state]);
+ }
for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
- if (mdi->next_state == DS_REMOVE && mdi->state_fd > 0) {
- int remove_err;
+ if (mdi->next_state == DS_REMOVE && mdi->state_fd >= 0) {
+ int remove_result;
write_attr("-blocked", mdi->state_fd);
/* the kernel may not be able to immediately remove the
* disk, we can simply wait until the next event to try
* again.
*/
- remove_err = write_attr("remove", mdi->state_fd);
- if (!remove_err) {
+ dprintf(" %d:-blocked", mdi->disk.raid_disk);
+ remove_result = write_attr("remove", mdi->state_fd);
+ if (remove_result > 0) {
+ dprintf(" %d:removed", mdi->disk.raid_disk);
close(mdi->state_fd);
mdi->state_fd = -1;
}
}
- if (mdi->next_state & DS_INSYNC)
+ if (mdi->next_state & DS_INSYNC) {
write_attr("+in_sync", mdi->state_fd);
+ dprintf(" %d:+in_sync", mdi->disk.raid_disk);
+ }
}
+ dprintf(" )\n");
/* move curr_ to prev_ */
a->prev_state = a->curr_state;
mdi->next_state = 0;
}
+ if (check_degraded) {
+ /* manager will do the actual check */
+ a->check_degraded = 1;
+ signal_manager();
+ }
+
if (deactivate)
a->container = NULL;
}
}
-static int handle_remove_device(struct md_remove_device_cmd *cmd, struct active_array *aa)
+#ifdef DEBUG
+static void dprint_wake_reasons(fd_set *fds)
{
- struct active_array *a;
- struct mdinfo *victim;
+ int i;
+ char proc_path[256];
+ char link[256];
+ char *basename;
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.
- */
+ fprintf(stderr, "monitor: wake ( ");
+ for (i = 0; i < FD_SETSIZE; i++) {
+ if (FD_ISSET(i, fds)) {
+ sprintf(proc_path, "/proc/%d/fd/%d",
+ (int) getpid(), i);
- /* 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;
+ rv = readlink(proc_path, link, sizeof(link) - 1);
+ if (rv < 0) {
+ fprintf(stderr, "%d:unknown ", i);
+ continue;
+ }
+ link[rv] = '\0';
+ basename = strrchr(link, '/');
+ fprintf(stderr, "%d:%s ",
+ i, basename ? ++basename : link);
+ }
}
-
- return 0;
+ fprintf(stderr, ")\n");
}
+#endif
-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;
-}
+int monitor_loop_cnt;
-static int wait_and_act(struct active_array *aa, int pfd, int monfd, int nowait)
+static int wait_and_act(struct supertype *container, int nowait)
{
fd_set rfds;
int maxfd = 0;
- struct active_array *a;
+ struct active_array **aap = &container->arrays;
+ 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) {
- /* 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 (manager_ready && *aap == NULL) {
+ /* No interesting arrays. Lets see about exiting.
+ * Note that blocking at this point is not a problem
+ * as there are no active arrays, there is nothing that
+ * we need to be ready to do.
+ */
+ int fd = open(container->device_name, O_RDONLY|O_EXCL);
+ if (fd >= 0 || errno != EBUSY) {
+ /* OK, we are safe to leave */
+ dprintf("no arrays to monitor... exiting\n");
+ remove_pidfile(container->devname);
+ exit_now = 1;
+ signal_manager();
+ exit(0);
+ }
}
if (!nowait) {
- rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
+ sigset_t set;
+ sigprocmask(SIG_UNBLOCK, NULL, &set);
+ sigdelset(&set, SIGUSR1);
+ monitor_loop_cnt |= 1;
+ rv = pselect(maxfd+1, &rfds, NULL, NULL, NULL, &set);
+ monitor_loop_cnt += 1;
+ if (rv == -1 && errno == EINTR)
+ rv = 0;
+ #ifdef DEBUG
+ dprint_wake_reasons(&rfds);
+ #endif
+
+ }
- if (rv <= 0)
- return rv;
+ if (update_queue) {
+ struct metadata_update *this;
- if (FD_ISSET(pfd, &rfds)) {
- int err = -1;
+ for (this = update_queue; this ; this = this->next)
+ container->ss->process_update(container, this);
- if (read(pfd, &err, 1) > 0)
- err = handle_pipe(active_cmd, aa);
- write(monfd, &err, 1);
- }
+ update_queue_handled = update_queue;
+ update_queue = NULL;
+ signal_manager();
+ container->ss->sync_metadata(container);
}
- 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;
+ /* FIXME check if device->state_fd need to be cleared?*/
+ signal_manager();
}
if (a->container)
rv += read_and_act(a);
}
/* propagate failures across container members */
- for (a = aa; a ; a = a->next) {
+ 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(aa, mdi);
+ reconcile_failed(*aap, mdi);
}
return rv;
int rv;
int first = 1;
do {
- rv = wait_and_act(container->arrays, container->mgr_pipe[0],
- container->mon_pipe[1], first);
+ rv = wait_and_act(container, first);
first = 0;
} while (rv >= 0);
}
projects / thirdparty / mdadm.git / blobdiff