/*
* mdmon - monitor external metadata arrays
*
- * Copyright (C) 2007-2008 Neil Brown <neilb@suse.de>
- * Copyright (C) 2007-2008 Intel Corporation
+ * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
+ * Copyright (C) 2007-2009 Intel Corporation
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
static void add_fd(fd_set *fds, int *maxfd, int fd)
{
+ struct stat st;
if (fd < 0)
return;
+ if (fstat(fd, &st) == -1) {
+ dprintf("%s: Invalid fd %d\n", __func__, fd);
+ return;
+ }
+ if (st.st_nlink == 0) {
+ dprintf("%s: fd %d was deleted\n", __func__, fd);
+ return;
+ }
if (fd > *maxfd)
*maxfd = fd;
FD_SET(fd, fds);
return n;
}
-int get_resync_start(struct active_array *a)
+static void read_resync_start(int fd, unsigned long long *v)
{
char buf[30];
int n;
- n = read_attr(buf, 30, a->resync_start_fd);
- if (n <= 0)
- return n;
+ n = read_attr(buf, 30, fd);
+ if (n <= 0) {
+ dprintf("%s: Failed to read resync_start (%d)\n",
+ __func__, fd);
+ return;
+ }
+ if (strncmp(buf, "none", 4) == 0)
+ *v = MaxSector;
+ else
+ *v = strtoull(buf, NULL, 10);
+}
+
+static unsigned long long read_sync_completed(int fd)
+{
+ unsigned long long val;
+ char buf[50];
+ int n;
+ char *ep;
- a->resync_start = strtoull(buf, NULL, 10);
+ n = read_attr(buf, 50, fd);
- return 1;
+ if (n <= 0)
+ return 0;
+ buf[n] = 0;
+ val = strtoull(buf, &ep, 0);
+ if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
+ return 0;
+ return val;
}
-
static enum array_state read_state(int fd)
{
char buf[20];
*
*/
+#define ARRAY_DIRTY 1
+#define ARRAY_BUSY 2
static int read_and_act(struct active_array *a)
{
+ unsigned long long sync_completed;
int check_degraded = 0;
+ int check_reshape = 0;
int deactivate = 0;
struct mdinfo *mdi;
- int dirty = 0;
+ int ret = 0;
+ int count = 0;
a->next_state = bad_word;
a->next_action = bad_action;
a->curr_state = read_state(a->info.state_fd);
a->curr_action = read_action(a->action_fd);
+ if (a->curr_state != clear)
+ /*
+ * In "clear" state, resync_start may wrongly be set to "0"
+ * when the kernel called md_clean but didn't remove the
+ * sysfs attributes yet
+ */
+ read_resync_start(a->resync_start_fd, &a->info.resync_start);
+ sync_completed = read_sync_completed(a->sync_completed_fd);
for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
mdi->next_state = 0;
- if (mdi->state_fd >= 0)
+ mdi->curr_state = 0;
+ if (mdi->state_fd >= 0) {
+ read_resync_start(mdi->recovery_fd,
+ &mdi->recovery_start);
mdi->curr_state = read_dev_state(mdi->state_fd);
+ }
}
- if (a->curr_state <= inactive &&
+ if (a->curr_state > inactive &&
+ a->prev_state == inactive) {
+ /* array has been started
+ * possible that container operation has to be completed
+ */
+ a->container->ss->set_array_state(a, 0);
+ }
+ if ((a->curr_state == bad_word || a->curr_state <= inactive) &&
a->prev_state > inactive) {
/* array has been stopped */
- get_resync_start(a);
a->container->ss->set_array_state(a, 1);
a->next_state = clear;
deactivate = 1;
}
if (a->curr_state == write_pending) {
- get_resync_start(a);
a->container->ss->set_array_state(a, 0);
a->next_state = active;
- dirty = 1;
+ ret |= ARRAY_DIRTY;
}
if (a->curr_state == active_idle) {
/* Set array to 'clean' FIRST, then mark clean
* in the metadata
*/
a->next_state = clean;
- dirty = 1;
+ ret |= ARRAY_DIRTY;
}
if (a->curr_state == clean) {
- get_resync_start(a);
a->container->ss->set_array_state(a, 1);
}
if (a->curr_state == active ||
- a->curr_state == suspended ||
- a->curr_state == bad_word)
- dirty = 1;
+ a->curr_state == suspended)
+ ret |= ARRAY_DIRTY;
if (a->curr_state == readonly) {
/* Well, I'm ready to handle things. If readonly
* wasn't requested, transition to read-auto.
/* explicit request for readonly array. Leave it alone */
;
} else {
- get_resync_start(a);
if (a->container->ss->set_array_state(a, 2))
a->next_state = read_auto; /* array is clean */
else {
a->next_state = active; /* Now active for recovery etc */
- dirty = 1;
+ ret |= ARRAY_DIRTY;
}
}
}
* until the array goes inactive or readonly though.
* Just check if we need to fiddle spares.
*/
- get_resync_start(a);
a->container->ss->set_array_state(a, a->curr_state <= clean);
check_degraded = 1;
}
mdi->curr_state);
if (! (mdi->curr_state & DS_INSYNC))
check_degraded = 1;
+ count++;
}
+ if (count != a->info.array.raid_disks)
+ check_degraded = 1;
}
+ if (!deactivate &&
+ a->curr_action == reshape &&
+ a->prev_action != reshape)
+ /* reshape was requested by mdadm. Need to see if
+ * new devices have been added. Manager does that
+ * when it sees check_reshape
+ */
+ check_reshape = 1;
+
/* Check for failures and if found:
* 1/ Record the failure in the metadata and unblock the device.
* FIXME update the kernel to stop notifying on failed drives when
a->container->ss->set_disk(a, mdi->disk.raid_disk,
mdi->curr_state);
check_degraded = 1;
- mdi->next_state |= DS_UNBLOCK;
+ if (mdi->curr_state & DS_BLOCKED)
+ mdi->next_state |= DS_UNBLOCK;
if (a->curr_state == read_auto) {
a->container->ss->set_array_state(a, 0);
a->next_state = active;
}
}
+ /* Check for recovery checkpoint notifications. We need to be a
+ * minimum distance away from the last checkpoint to prevent
+ * over checkpointing. Note reshape checkpointing is handled
+ * in the second branch.
+ */
+ if (sync_completed > a->last_checkpoint &&
+ sync_completed - a->last_checkpoint > a->info.component_size >> 4 &&
+ a->curr_action > reshape) {
+ /* A (non-reshape) sync_action has reached a checkpoint.
+ * Record the updated position in the metadata
+ */
+ a->last_checkpoint = sync_completed;
+ a->container->ss->set_array_state(a, a->curr_state <= clean);
+ } else if ((a->curr_action == idle && a->prev_action == reshape) ||
+ (a->curr_action == reshape
+ && sync_completed > a->last_checkpoint) ) {
+ /* Reshape has progressed or completed so we need to
+ * update the array state - and possibly the array size
+ */
+ if (sync_completed != 0)
+ a->last_checkpoint = sync_completed;
+ /* We might need to update last_checkpoint depending on
+ * the reason that reshape finished.
+ * if array reshape is really finished:
+ * set check point to the end, this allows
+ * set_array_state() to finalize reshape in metadata
+ * if reshape if broken: do not set checkpoint to the end
+ * this allows for reshape restart from checkpoint
+ */
+ if ((a->curr_action != reshape) &&
+ (a->prev_action == reshape)) {
+ char buf[40];
+ if ((sysfs_get_str(&a->info, NULL,
+ "reshape_position",
+ buf,
+ sizeof(buf)) >= 0) &&
+ strncmp(buf, "none", 4) == 0)
+ a->last_checkpoint = a->info.component_size;
+ }
+ a->container->ss->set_array_state(a, a->curr_state <= clean);
+ a->last_checkpoint = sync_completed;
+ }
+
+ if (sync_completed > a->last_checkpoint)
+ a->last_checkpoint = sync_completed;
+
a->container->ss->sync_metadata(a->container);
dprintf("%s(%d): state:%s action:%s next(", __func__, a->info.container_member,
array_states[a->curr_state], sync_actions[a->curr_action]);
if ((mdi->next_state & DS_REMOVE) && mdi->state_fd >= 0) {
int remove_result;
- /* the kernel may not be able to immediately remove the
- * disk, we can simply wait until the next event to try
- * again.
+ /* The kernel may not be able to immediately remove the
+ * disk. In that case we wait a little while and
+ * try again.
*/
remove_result = write_attr("remove", mdi->state_fd);
if (remove_result > 0) {
dprintf(" %d:removed", mdi->disk.raid_disk);
close(mdi->state_fd);
+ close(mdi->recovery_fd);
mdi->state_fd = -1;
- }
+ } else
+ ret |= ARRAY_BUSY;
}
if (mdi->next_state & DS_INSYNC) {
write_attr("+in_sync", mdi->state_fd);
mdi->next_state = 0;
}
- if (check_degraded) {
+ if (check_degraded || check_reshape) {
/* manager will do the actual check */
- a->check_degraded = 1;
+ if (check_degraded)
+ a->check_degraded = 1;
+ if (check_reshape)
+ a->check_reshape = 1;
signal_manager();
}
if (deactivate)
a->container = NULL;
- return dirty;
+ return ret;
}
static struct mdinfo *
struct mdinfo *victim;
for (a = aa; a; a = a->next) {
- if (!a->container)
+ if (!a->container || a->to_remove)
continue;
victim = find_device(a, failed->disk.major, failed->disk.minor);
if (!victim)
/* once an array has been deactivated we want to
* ask the manager to discard it.
*/
- if (!a->container) {
+ if (!a->container || a->to_remove) {
if (discard_this) {
ap = &(*ap)->next;
continue;
add_fd(&rfds, &maxfd, a->info.state_fd);
add_fd(&rfds, &maxfd, a->action_fd);
+ add_fd(&rfds, &maxfd, a->sync_completed_fd);
for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
add_fd(&rfds, &maxfd, mdi->state_fd);
* problem as there are no active arrays, there is
* nothing that we need to be ready to do.
*/
- int fd = open_dev_excl(container->devnum);
+ int fd;
+ if (sigterm)
+ fd = open_dev_excl(container->devnm);
+ else
+ fd = open_dev_flags(container->devnm, O_RDONLY|O_EXCL);
if (fd >= 0 || errno != EBUSY) {
/* OK, we are safe to leave */
if (sigterm && !dirty_arrays)
dprintf("caught sigterm, all clean... exiting\n");
else
dprintf("no arrays to monitor... exiting\n");
- remove_pidfile(container->devname);
+ if (!sigterm)
+ /* On SIGTERM, someone (the take-over mdmon) will
+ * clean up
+ */
+ remove_pidfile(container->devnm);
exit_now = 1;
signal_manager();
+ close(fd);
exit(0);
}
}
if (!nowait) {
sigset_t set;
+ struct timespec ts;
+ ts.tv_sec = 24*3600;
+ ts.tv_nsec = 0;
+ if (*aap == NULL || container->retry_soon) {
+ /* just waiting to get O_EXCL access */
+ ts.tv_sec = 0;
+ ts.tv_nsec = 20000000ULL;
+ }
sigprocmask(SIG_UNBLOCK, NULL, &set);
sigdelset(&set, SIGUSR1);
monitor_loop_cnt |= 1;
- rv = pselect(maxfd+1, &rfds, NULL, NULL, NULL, &set);
+ rv = pselect(maxfd+1, NULL, NULL, &rfds, &ts, &set);
monitor_loop_cnt += 1;
if (rv == -1 && errno == EINTR)
rv = 0;
#ifdef DEBUG
dprint_wake_reasons(&rfds);
#endif
-
+ container->retry_soon = 0;
}
if (update_queue) {
rv = 0;
dirty_arrays = 0;
for (a = *aap; a ; a = a->next) {
- int is_dirty;
if (a->replaces && !discard_this) {
struct active_array **ap;
/* FIXME check if device->state_fd need to be cleared?*/
signal_manager();
}
- if (a->container) {
- is_dirty = read_and_act(a);
+ if (a->container && !a->to_remove) {
+ int ret = read_and_act(a);
rv |= 1;
- dirty_arrays += is_dirty;
+ dirty_arrays += !!(ret & ARRAY_DIRTY);
/* when terminating stop manipulating the array after it
* is clean, but make sure read_and_act() is given a
* chance to handle 'active_idle'
*/
- if (sigterm && !is_dirty)
+ if (sigterm && !(ret & ARRAY_DIRTY))
a->container = NULL; /* stop touching this array */
+ if (ret & ARRAY_BUSY)
+ container->retry_soon = 1;
}
}
/* propagate failures across container members */
for (a = *aap; a ; a = a->next) {
- if (!a->container)
+ if (!a->container || a->to_remove)
continue;
for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
if (mdi->curr_state & DS_FAULTY)