+/*
+ * mdmon - monitor external metadata arrays
+ *
+ * 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,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
#include "mdadm.h"
#include "mdmon.h"
return n;
}
-
-int get_resync_start(struct active_array *a)
+static unsigned long long read_resync_start(int fd)
{
char buf[30];
int n;
- n = read_attr(buf, 30, a->resync_start_fd);
+ n = read_attr(buf, 30, fd);
if (n <= 0)
- return n;
-
- a->resync_start = strtoull(buf, NULL, 10);
-
- return 1;
-}
-
-static int attr_match(const char *attr, const char *str)
-{
- /* See if attr, read from a sysfs file, matches
- * str. They must either be the same, or attr can
- * have a trailing newline or comma
- */
- while (*attr && *str && *attr == *str) {
- attr++;
- str++;
- }
-
- if (*str || (*attr && *attr != ',' && *attr != '\n'))
return 0;
- return 1;
+ if (strncmp(buf, "none", 4) == 0)
+ return MaxSector;
+ else
+ return strtoull(buf, NULL, 10);
}
-static int match_word(const char *word, char **list)
+static unsigned long long read_sync_completed(int fd)
{
+ unsigned long long val;
+ char buf[50];
int n;
- for (n=0; list[n]; n++)
- if (attr_match(word, list[n]))
- break;
- return n;
+ char *ep;
+
+ n = read_attr(buf, 50, fd);
+
+ 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)
if (n <= 0)
return bad_word;
- return (enum array_state) match_word(buf, array_states);
+ return (enum array_state) sysfs_match_word(buf, array_states);
}
static enum sync_action read_action( int fd)
if (n <= 0)
return bad_action;
- return (enum sync_action) match_word(buf, sync_actions);
+ return (enum sync_action) sysfs_match_word(buf, sync_actions);
}
int read_dev_state(int fd)
cp = buf;
while (cp) {
- if (attr_match(cp, "faulty"))
+ if (sysfs_attr_match(cp, "faulty"))
rv |= DS_FAULTY;
- if (attr_match(cp, "in_sync"))
+ if (sysfs_attr_match(cp, "in_sync"))
rv |= DS_INSYNC;
- if (attr_match(cp, "write_mostly"))
+ if (sysfs_attr_match(cp, "write_mostly"))
rv |= DS_WRITE_MOSTLY;
- if (attr_match(cp, "spare"))
+ if (sysfs_attr_match(cp, "spare"))
rv |= DS_SPARE;
- if (attr_match(cp, "blocked"))
+ if (sysfs_attr_match(cp, "blocked"))
rv |= DS_BLOCKED;
cp = strchr(cp, ',');
if (cp)
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 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);
+ a->info.resync_start = read_resync_start(a->resync_start_fd);
+ 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) {
+ mdi->recovery_start = read_resync_start(mdi->recovery_fd);
mdi->curr_state = read_dev_state(mdi->state_fd);
+ }
}
+ 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 <= 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;
}
if (a->curr_state == active_idle) {
- /* Set array to 'clean' FIRST, then
- * a->ss->mark_clean(a, ~0ULL);
- * just ignore for now.
+ /* Set array to 'clean' FIRST, then mark clean
+ * in the metadata
*/
+ a->next_state = clean;
+ dirty = 1;
}
-
+ if (a->curr_state == clean) {
+ a->container->ss->set_array_state(a, 1);
+ }
+ if (a->curr_state == active ||
+ a->curr_state == suspended ||
+ a->curr_state == bad_word)
+ dirty = 1;
if (a->curr_state == readonly) {
- /* Well, I'm ready to handle things, so
- * read-auto is OK. FIXME what if we really want
- * readonly ???
+ /* Well, I'm ready to handle things. If readonly
+ * wasn't requested, transition to read-auto.
*/
- get_resync_start(a);
- if (a->resync_start == ~0ULL) {
- a->next_state = read_auto; /* array is clean */
- /* give the metadata a chance to force active if
- * we have some recovery to do. metadata sets
- * resync_start to !MaxSector in this case
- */
- a->container->ss->set_array_state(a, 1);
- }
- if (a->resync_start != ~0ULL) {
- a->container->ss->set_array_state(a, 0);
- a->next_state = active;
+ char buf[64];
+ read_attr(buf, sizeof(buf), a->metadata_fd);
+ if (strncmp(buf, "external:-", 10) == 0) {
+ /* explicit request for readonly array. Leave it alone */
+ ;
+ } else {
+ 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;
+ }
}
}
* 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, 0);
+ 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
}
}
+ /* 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: update[%d]: (", __func__, a->info.container_member);
+ dprintf("%s(%d): state:%s action:%s next(", __func__, a->info.container_member,
+ array_states[a->curr_state], sync_actions[a->curr_action]);
/* Effect state changes in the array */
if (a->next_state != bad_word) {
}
if (a->next_action != bad_action) {
write_attr(sync_actions[a->next_action], a->action_fd);
- dprintf(" action:%s", array_states[a->next_state]);
+ dprintf(" action:%s", sync_actions[a->next_action]);
}
for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
if (mdi->next_state & DS_UNBLOCK) {
if (remove_result > 0) {
dprintf(" %d:removed", mdi->disk.raid_disk);
close(mdi->state_fd);
+ close(mdi->recovery_fd);
mdi->state_fd = -1;
}
}
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 1;
+ return dirty;
}
static struct mdinfo *
struct active_array *a, **ap;
int rv;
struct mdinfo *mdi;
+ static unsigned int dirty_arrays = ~0; /* start at some non-zero value */
FD_ZERO(&rfds);
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);
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.
+ if (manager_ready && (*aap == NULL || (sigterm && !dirty_arrays))) {
+ /* No interesting arrays, or we have been told to
+ * terminate and everything is clean. 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);
+ int fd;
+ if (sigterm)
+ fd = open_dev_excl(container->devnum);
+ else
+ fd = open_dev_flags(container->devnum, 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);
+ if (sigterm && !dirty_arrays)
+ dprintf("caught sigterm, all clean... exiting\n");
+ else
+ dprintf("no arrays to monitor... exiting\n");
+ if (!sigterm)
+ /* On SIGTERM, someone (the take-over mdmon) will
+ * clean up
+ */
+ remove_pidfile(container->devname);
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) {
+ /* 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;
container->ss->sync_metadata(container);
}
+ rv = 0;
+ dirty_arrays = 0;
for (a = *aap; a ; a = a->next) {
+ int is_dirty;
+
if (a->replaces && !discard_this) {
struct active_array **ap;
for (ap = &a->next; *ap && *ap != a->replaces;
/* FIXME check if device->state_fd need to be cleared?*/
signal_manager();
}
- if (a->container)
- rv += read_and_act(a);
+ if (a->container) {
+ is_dirty = read_and_act(a);
+ rv |= 1;
+ dirty_arrays += is_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)
+ a->container = NULL; /* stop touching this array */
+ }
}
/* propagate failures across container members */