/*
* 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,
{
struct mdinfo *d;
- for (d = aa->info.devs; d; d = d->next)
+ for (d = aa->info.devs; d; d = d->next) {
+ close(d->recovery_fd);
close(d->state_fd);
+ close(d->bb_fd);
+ close(d->ubb_fd);
+ }
- close(aa->action_fd);
- close(aa->info.state_fd);
- close(aa->resync_start_fd);
+ if (aa->action_fd >= 0)
+ close(aa->action_fd);
+ if (aa->info.state_fd >= 0)
+ close(aa->info.state_fd);
+ if (aa->resync_start_fd >= 0)
+ close(aa->resync_start_fd);
+ if (aa->metadata_fd >= 0)
+ close(aa->metadata_fd);
+ if (aa->sync_completed_fd >= 0)
+ close(aa->sync_completed_fd);
+ if (aa->safe_mode_delay_fd >= 0)
+ close(aa->safe_mode_delay_fd);
}
static void free_aa(struct active_array *aa)
/* 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);
+ dprintf("sys_name: %s\n", aa->info.sys_name);
if (!aa->container)
close_aa(aa);
while (aa->info.devs) {
static struct active_array *duplicate_aa(struct active_array *aa)
{
- struct active_array *newa = malloc(sizeof(*newa));
+ struct active_array *newa = xmalloc(sizeof(*newa));
struct mdinfo **dp1, **dp2;
*newa = *aa;
if ((*dp1)->state_fd < 0)
continue;
- d = malloc(sizeof(*d));
+ d = xmalloc(sizeof(*d));
*d = **dp1;
*dp2 = d;
dp2 = & d->next;
struct metadata_update *update_queue_handled = NULL;
struct metadata_update *update_queue_pending = NULL;
-void check_update_queue(struct supertype *container)
+static void free_updates(struct metadata_update **update)
{
- while (update_queue_handled) {
- struct metadata_update *this = update_queue_handled;
- update_queue_handled = this->next;
+ while (*update) {
+ struct metadata_update *this = *update;
+ void **space_list = this->space_list;
+
+ *update = this->next;
free(this->buf);
- if (this->space)
- free(this->space);
+ free(this->space);
+ while (space_list) {
+ void *space = space_list;
+ space_list = *space_list;
+ free(space);
+ }
free(this);
}
+}
+
+void check_update_queue(struct supertype *container)
+{
+ free_updates(&update_queue_handled);
+
if (update_queue == NULL &&
update_queue_pending) {
update_queue = update_queue_pending;
.state = 0,
};
- dprintf("%s: add %d:%d to container\n",
- __func__, sd->disk.major, sd->disk.minor);
+ dprintf("add %d:%d to container\n", sd->disk.major, sd->disk.minor);
+
+ sd->next = st->devs;
+ st->devs = sd;
sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
dfd = dev_open(nm, O_RDWR);
return;
st->update_tail = &update;
- st->ss->add_to_super(st, &dk, dfd, NULL);
+ st->ss->add_to_super(st, &dk, dfd, NULL, INVALID_SECTORS);
+ st->ss->write_init_super(st);
+ queue_metadata_update(update);
+ st->update_tail = NULL;
+}
+
+/*
+ * Create and queue update structure about the removed disks.
+ * The update is prepared by super type handler and passed to the monitor
+ * thread.
+ */
+static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd)
+{
+ struct metadata_update *update = NULL;
+ mdu_disk_info_t dk = {
+ .number = -1,
+ .major = sd->disk.major,
+ .minor = sd->disk.minor,
+ .raid_disk = -1,
+ .state = 0,
+ };
+ dprintf("remove %d:%d from container\n",
+ sd->disk.major, sd->disk.minor);
+
+ st->update_tail = &update;
+ st->ss->remove_from_super(st, &dk);
+ /* FIXME this write_init_super shouldn't be here.
+ * We have it after add_to_super to write to new device,
+ * but with 'remove' we don't ant to write to that device!
+ */
st->ss->write_init_super(st);
queue_metadata_update(update);
st->update_tail = NULL;
static void manage_container(struct mdstat_ent *mdstat,
struct supertype *container)
{
- /* The only thing of interest here is if a new device
- * has been added to the container. We add it to the
- * array ignoring any metadata on it.
+ /* Of interest here are:
+ * - if a new device has been added to the container, we
+ * add it to the array ignoring any metadata on it.
+ * - if a device has been removed from the container, we
+ * remove it from the device list and update the metadata.
* FIXME should we look for compatible metadata and take hints
* about spare assignment.... probably not.
*/
* To see what is removed and what is added.
* These need to be remove from, or added to, the array
*/
- mdi = sysfs_read(-1, mdstat->devnum, GET_DEVS);
+ mdi = sysfs_read(-1, mdstat->devnm, GET_DEVS);
if (!mdi) {
/* invalidate the current count so we can try again */
container->devcnt = -1;
if (!found) {
cd = *cdp;
*cdp = (*cdp)->next;
+ remove_disk_from_container(container, cd);
free(cd);
} else
cdp = &(*cdp)->next;
if (di->disk.major == cd->disk.major &&
di->disk.minor == cd->disk.minor)
break;
- if (!cd)
- add_disk_to_container(container, di);
+ if (!cd) {
+ struct mdinfo *newd = xmalloc(sizeof(*newd));
+
+ *newd = *di;
+ add_disk_to_container(container, newd);
+ }
}
sysfs_free(mdi);
container->devcnt = mdstat->devcnt;
}
}
+static int sysfs_open2(char *devnum, char *name, char *attr)
+{
+ int fd = sysfs_open(devnum, name, attr);
+ if (fd >= 0) {
+ /* seq_file in the kernel allocates buffer space
+ * on the first read. Do that now so 'monitor'
+ * never needs too.
+ */
+ char buf[200];
+ if (read(fd, buf, sizeof(buf)) < 0)
+ /* pretend not to ignore return value */
+ return fd;
+ }
+ return fd;
+}
+
+static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone,
+ struct active_array *aa)
+{
+ if (!disk || !clone)
+ return -1;
+
+ *disk = *clone;
+ disk->recovery_fd = sysfs_open2(aa->info.sys_name, disk->sys_name,
+ "recovery_start");
+ if (disk->recovery_fd < 0)
+ return -1;
+ disk->state_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, "state");
+ if (disk->state_fd < 0) {
+ close(disk->recovery_fd);
+ return -1;
+ }
+ disk->bb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name,
+ "bad_blocks");
+ if (disk->bb_fd < 0) {
+ close(disk->recovery_fd);
+ close(disk->state_fd);
+ return -1;
+ }
+ disk->ubb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name,
+ "unacknowledged_bad_blocks");
+ if (disk->ubb_fd < 0) {
+ close(disk->recovery_fd);
+ close(disk->state_fd);
+ close(disk->bb_fd);
+ return -1;
+ }
+ disk->prev_state = read_dev_state(disk->state_fd);
+ disk->curr_state = disk->prev_state;
+ disk->next = aa->info.devs;
+ aa->info.devs = disk;
+
+ return 0;
+}
+
static void manage_member(struct mdstat_ent *mdstat,
struct active_array *a)
{
* We do not need to look for device state changes here, that
* is dealt with by the monitor.
*
- * We just look for changes which suggest that a reshape is
- * being requested.
- * Unfortunately decreases in raid_disks don't show up in
- * mdstat until the reshape completes FIXME.
+ * If a reshape is being requested, monitor will have noticed
+ * that sync_action changed and will have set check_reshape.
+ * We just need to see if new devices have appeared. All metadata
+ * updates will already have been processed.
*
- * Actually, we also want to handle degraded arrays here by
+ * 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
+ char buf[64];
+ int frozen;
+ struct supertype *container = a->container;
+ struct mdinfo *mdi;
+
+ if (container == NULL)
+ /* Raced with something */
+ return;
+
+ if (mdstat->active) {
+ // FIXME
+ a->info.array.raid_disks = mdstat->raid_disks;
+ // MORE
+ }
+
+ mdi = sysfs_read(-1, mdstat->devnm,
+ GET_COMPONENT|GET_CONSISTENCY_POLICY);
+ if (mdi) {
+ a->info.component_size = mdi->component_size;
+ a->info.consistency_policy = mdi->consistency_policy;
+ sysfs_free(mdi);
+ }
- if (a->check_degraded) {
+ /* honor 'frozen' */
+ if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0)
+ frozen = buf[9] == '-';
+ else
+ frozen = 1; /* can't read metadata_version assume the worst */
+
+ /* If sync_action is not 'idle' then don't try recovery now */
+ if (!frozen &&
+ sysfs_get_str(&a->info, NULL, "sync_action",
+ buf, sizeof(buf)) > 0 && strncmp(buf, "idle", 4) != 0)
+ frozen = 1;
+
+ if (mdstat->level) {
+ int level = map_name(pers, mdstat->level);
+ if (level == 0 || level == LEVEL_LINEAR) {
+ a->to_remove = 1;
+ wakeup_monitor();
+ return;
+ }
+ else if (a->info.array.level != level && level > 0) {
+ struct active_array *newa = duplicate_aa(a);
+ if (newa) {
+ newa->info.array.level = level;
+ replace_array(container, a, newa);
+ a = newa;
+ }
+ }
+ }
+
+ /* we are after monitor kick,
+ * so container field can be cleared - check it again
+ */
+ if (a->container == NULL)
+ return;
+
+ if (sigterm && a->info.safe_mode_delay != 1 &&
+ a->safe_mode_delay_fd >= 0) {
+ long int new_delay = 1;
+ char delay[10];
+ ssize_t len;
+
+ len = snprintf(delay, sizeof(delay), "0.%03ld\n", new_delay);
+ if (write(a->safe_mode_delay_fd, delay, len) == len)
+ a->info.safe_mode_delay = new_delay;
+ }
+
+ /* We don't check the array while any update is pending, as it
+ * might container a change (such as a spare assignment) which
+ * could affect our decisions.
+ */
+ if (a->check_degraded && !frozen &&
+ update_queue == NULL && update_queue_pending == NULL) {
struct metadata_update *updates = NULL;
- struct mdinfo *newdev;
+ struct mdinfo *newdev = NULL;
struct active_array *newa;
+ struct mdinfo *d;
a->check_degraded = 0;
/* 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) < 0)
- continue;
- newd = malloc(sizeof(*newd));
- *newd = *d;
- newd->next = newa->info.devs;
- newa->info.devs = newd;
-
- 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;
+ newdev = container->ss->activate_spare(a, &updates);
+ if (!newdev)
+ return;
+
+ newa = duplicate_aa(a);
+ if (!newa)
+ goto out;
+ /* prevent the kernel from activating the disk(s) before we
+ * finish adding them
+ */
+ dprintf("freezing %s\n", a->info.sys_name);
+ sysfs_set_str(&a->info, NULL, "sync_action", "frozen");
+
+ /* 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;
+
+ newd = xmalloc(sizeof(*newd));
+ if (sysfs_add_disk(&newa->info, d, 0) < 0) {
+ free(newd);
+ continue;
+ }
+ disk_init_and_add(newd, d, newa);
+ }
+ queue_metadata_update(updates);
+ updates = NULL;
+ while (update_queue_pending || update_queue) {
+ check_update_queue(container);
+ usleep(15*1000);
+ }
+ replace_array(container, a, newa);
+ if (sysfs_set_str(&a->info, NULL,
+ "sync_action", "recover") == 0)
+ newa->prev_action = recover;
+ dprintf("recovery started on %s\n", a->info.sys_name);
+ out:
+ while (newdev) {
+ d = newdev->next;
+ free(newdev);
+ newdev = d;
+ }
+ free_updates(&updates);
+ }
+
+ if (a->check_reshape) {
+ /* mdadm might have added some devices to the array.
+ * We want to disk_init_and_add any such device to a
+ * duplicate_aa and replace a with that.
+ * mdstat doesn't have enough info so we sysfs_read
+ * and look for new stuff.
+ */
+ struct mdinfo *info, *d, *d2, *newd;
+ unsigned long long array_size;
+ struct active_array *newa = NULL;
+ a->check_reshape = 0;
+ info = sysfs_read(-1, mdstat->devnm,
+ GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE);
+ if (!info)
+ goto out2;
+ for (d = info->devs; d; d = d->next) {
+ if (d->disk.raid_disk < 0)
+ continue;
+ for (d2 = a->info.devs; d2; d2 = d2->next)
+ if (d2->disk.raid_disk ==
+ d->disk.raid_disk)
+ break;
+ if (d2)
+ /* already have this one */
+ continue;
+ if (!newa) {
+ newa = duplicate_aa(a);
+ if (!newa)
+ break;
}
- queue_metadata_update(updates);
- replace_array(a->container, a, newa);
- sysfs_set_str(&a->info, NULL, "sync_action", "recover");
+ newd = xmalloc(sizeof(*newd));
+ disk_init_and_add(newd, d, newa);
+ }
+ if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0 &&
+ a->info.custom_array_size > array_size*2) {
+ sysfs_set_num(info, NULL, "array_size",
+ a->info.custom_array_size/2);
}
+ out2:
+ sysfs_free(info);
+ if (newa)
+ replace_array(container, a, newa);
}
}
char *inst;
int i;
int failed = 0;
+ char buf[40];
/* check if array is ready to be monitored */
- if (!mdstat->active)
+ if (!mdstat->active || !mdstat->level)
+ return;
+ if (strcmp(mdstat->level, "raid0") == 0 ||
+ strcmp(mdstat->level, "linear") == 0)
return;
- mdi = sysfs_read(-1, mdstat->devnum,
+ mdi = sysfs_read(-1, mdstat->devnm,
GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT|
- GET_DEGRADED|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE);
+ GET_SAFEMODE|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
+ GET_LAYOUT|GET_DEVS_ALL);
- new = malloc(sizeof(*new));
-
- if (!new || !mdi) {
- if (mdi)
- sysfs_free(mdi);
- if (new)
- free(new);
+ if (!mdi)
return;
- }
- memset(new, 0, sizeof(*new));
+ new = xcalloc(1, sizeof(*new));
- new->devnum = mdstat->devnum;
- strcpy(new->info.sys_name, devnum2devname(new->devnum));
+ strcpy(new->info.sys_name, mdstat->devnm);
new->prev_state = new->curr_state = new->next_state = inactive;
new->prev_action= new->curr_action= new->next_action= idle;
new->container = container;
- inst = &mdstat->metadata_version[10+strlen(container->devname)+1];
+ inst = to_subarray(mdstat, container->devnm);
new->info.array = mdi->array;
new->info.component_size = mdi->component_size;
for (i = 0; i < new->info.array.raid_disks; i++) {
- struct mdinfo *newd = malloc(sizeof(*newd));
+ struct mdinfo *newd = xmalloc(sizeof(*newd));
for (di = mdi->devs; di; di = di->next)
if (i == di->disk.raid_disk)
break;
- if (di) {
- memcpy(newd, di, sizeof(*newd));
-
- newd->state_fd = sysfs_open(new->devnum,
- newd->sys_name,
- "state");
+ if (disk_init_and_add(newd, di, new) != 0) {
+ if (newd)
+ free(newd);
- newd->prev_state = read_dev_state(newd->state_fd);
- newd->curr_state = newd->prev_state;
- } else if (failed + 1 > new->info.array.failed_disks) {
- /* we cannot properly monitor without all working disks */
- new->container = NULL;
- break;
- } else {
failed++;
- free(newd);
- continue;
+ if (failed > new->info.array.failed_disks) {
+ /* we cannot properly monitor without all working disks */
+ new->container = NULL;
+ break;
+ }
}
- 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->metadata_fd = sysfs_open(new->devnum, NULL, "metadata_version");
- get_resync_start(new);
- dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst),
+ new->action_fd = sysfs_open2(new->info.sys_name, NULL, "sync_action");
+ new->info.state_fd = sysfs_open2(new->info.sys_name, NULL, "array_state");
+ new->resync_start_fd = sysfs_open2(new->info.sys_name, NULL, "resync_start");
+ new->metadata_fd = sysfs_open2(new->info.sys_name, NULL, "metadata_version");
+ new->sync_completed_fd = sysfs_open2(new->info.sys_name, NULL, "sync_completed");
+ new->safe_mode_delay_fd = sysfs_open2(new->info.sys_name, NULL,
+ "safe_mode_delay");
+
+ dprintf("inst: %s action: %d state: %d\n", inst,
new->action_fd, new->info.state_fd);
+ if (mdi->safe_mode_delay >= 50)
+ /* Normal start, mdadm set this. */
+ new->info.safe_mode_delay = mdi->safe_mode_delay;
+ else
+ /* Restart, just pick a number */
+ new->info.safe_mode_delay = 5000;
+ sysfs_set_safemode(&new->info, new->info.safe_mode_delay);
+
+ /* reshape_position is set by mdadm in sysfs
+ * read this information for new arrays only (empty victim)
+ */
+ if ((victim == NULL) &&
+ (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) &&
+ (strncmp(buf, "reshape", 7) == 0)) {
+ if (sysfs_get_ll(mdi, NULL, "reshape_position",
+ &new->last_checkpoint) != 0)
+ new->last_checkpoint = 0;
+ else {
+ int data_disks = mdi->array.raid_disks;
+ if (mdi->array.level == 4 || mdi->array.level == 5)
+ data_disks--;
+ if (mdi->array.level == 6)
+ data_disks -= 2;
+
+ new->last_checkpoint /= data_disks;
+ }
+ dprintf("mdmon: New monitored array is under reshape.\n"
+ " Last checkpoint is: %llu\n",
+ new->last_checkpoint);
+ }
+
sysfs_free(mdi);
/* if everything checks out tell the metadata handler we want to
* manage this instance
*/
if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) {
- fprintf(stderr, "mdmon: failed to monitor %s\n",
+ pr_err("failed to monitor %s\n",
mdstat->metadata_version);
new->container = NULL;
free_aa(new);
for ( ; mdstat ; mdstat = mdstat->next) {
struct active_array *a;
- if (mdstat->devnum == container->devnum) {
+ if (strcmp(mdstat->devnm, container->devnm) == 0) {
manage_container(mdstat, container);
continue;
}
- if (!is_container_member(mdstat, container->devname))
+ if (!is_container_member(mdstat, container->devnm))
/* Not for this array */
continue;
/* Looks like a member of this container */
for (a = container->arrays; a; a = a->next) {
- if (mdstat->devnum == a->devnum) {
- if (a->container)
+ if (strcmp(mdstat->devnm, a->info.sys_name) == 0) {
+ if (a->container && a->to_remove == 0)
manage_member(mdstat, a);
break;
}
}
- if (a == NULL || !a->container)
+ if ((a == NULL || !a->container) && !sigterm)
manage_new(mdstat, container, a);
}
}
if (msg->len == 0) { /* ping_monitor */
int cnt;
-
+
cnt = monitor_loop_cnt;
if (cnt & 1)
cnt += 2; /* wait until next pselect */
manage(mdstat, container);
free_mdstat(mdstat);
} else if (!sigterm) {
- mu = malloc(sizeof(*mu));
+ mu = xmalloc(sizeof(*mu));
mu->len = msg->len;
mu->buf = msg->buf;
msg->buf = NULL;
mu->space = NULL;
+ mu->space_list = NULL;
mu->next = NULL;
if (container->ss->prepare_update)
- container->ss->prepare_update(container, mu);
+ if (!container->ss->prepare_update(container, mu))
+ free_updates(&mu);
queue_metadata_update(mu);
}
}
/* read and validate the message */
if (receive_message(fd, &msg, tmo) == 0) {
handle_message(container, &msg);
- if (ack(fd, tmo) < 0)
+ if (msg.len == 0) {
+ /* ping reply with version */
+ msg.buf = Version;
+ msg.len = strlen(Version) + 1;
+ if (send_message(fd, &msg, tmo) < 0)
+ terminate = 1;
+ } else if (ack(fd, tmo) < 0)
terminate = 1;
} else
terminate = 1;
{
struct mdstat_ent *mdstat;
sigset_t set;
- int proc_fd;
sigprocmask(SIG_UNBLOCK, NULL, &set);
sigdelset(&set, SIGUSR1);
- sigdelset(&set, SIGHUP);
- sigdelset(&set, SIGALRM);
sigdelset(&set, SIGTERM);
- proc_fd = open("/proc/mounts", O_RDONLY);
do {
read_sock(container);
- if (container->sock < 0 || socket_hup_requested) {
- close(container->sock);
- container->sock = make_control_sock(container->devname);
- make_pidfile(container->devname, 0);
- socket_hup_requested = 0;
- }
- if (container->sock < 0)
- alarm(30);
-
free_mdstat(mdstat);
}
remove_old();
if (sigterm)
wakeup_monitor();
- if (update_queue == NULL) {
- if (container->sock < 0)
- mdstat_wait_fd(proc_fd, &set);
- else
- mdstat_wait_fd(container->sock, &set);
- } else
+ if (update_queue == NULL)
+ mdstat_wait_fd(container->sock, &set);
+ else
/* If an update is happening, just wait for signal */
pselect(0, NULL, NULL, NULL, NULL, &set);
} while(1);
projects / thirdparty / mdadm.git / blobdiff