+/*
+ * 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"
-
+#include <sys/syscall.h>
#include <sys/select.h>
-
+#include <signal.h>
static char *array_states[] = {
"clear", "inactive", "suspended", "readonly", "read-auto",
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;
}
-static int get_sync_pos(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->sync_pos_fd);
- if (n <= 0)
- return n;
-
- if (strncmp(buf, "max", 3) == 0) {
- a->sync_pos = ~(unsigned long long)0;
- return 1;
+ n = read_attr(buf, 30, fd);
+ if (n <= 0) {
+ dprintf("%s: Failed to read resync_start (%d)\n",
+ __func__, fd);
+ return;
}
- a->sync_pos = strtoull(buf, NULL, 10);
- return 1;
+ if (strncmp(buf, "none", 4) == 0)
+ *v = MaxSector;
+ else
+ *v = strtoull(buf, NULL, 10);
}
-static int get_resync_start(struct active_array *a)
+static unsigned long long read_sync_completed(int fd)
{
- char buf[30];
+ unsigned long long val;
+ char buf[50];
int n;
+ char *ep;
- n = read_attr(buf, 30, a->resync_start_fd);
- if (n <= 0)
- return n;
-
- a->resync_start = strtoull(buf, NULL, 10);
-
- return 1;
-}
+ n = read_attr(buf, 50, fd);
-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'))
+ if (n <= 0)
return 0;
- return 1;
-}
-
-static int match_word(const char *word, char **list)
-{
- int n;
- for (n=0; list[n]; n++)
- if (attr_match(word, list[n]))
- break;
- return n;
+ 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)
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
*
*/
+#define ARRAY_DIRTY 1
+#define ARRAY_BUSY 2
static int read_and_act(struct active_array *a)
{
- int check_degraded;
+ unsigned long long sync_completed;
+ int check_degraded = 0;
+ int check_reshape = 0;
int deactivate = 0;
struct mdinfo *mdi;
+ 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_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_clean(a, 0);
+ a->container->ss->set_array_state(a, 0);
a->next_state = active;
+ ret |= ARRAY_DIRTY;
}
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;
+ ret |= ARRAY_DIRTY;
}
-
+ if (a->curr_state == clean) {
+ a->container->ss->set_array_state(a, 1);
+ }
+ if (a->curr_state == active ||
+ a->curr_state == suspended)
+ ret |= ARRAY_DIRTY;
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 */
- else {
- a->container->ss->mark_clean(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 */
+ ret |= ARRAY_DIRTY;
+ }
}
}
- if (a->curr_action == idle &&
+ if (!deactivate &&
+ a->curr_action == idle &&
a->prev_action == resync) {
/* 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.
*/
+ a->container->ss->set_array_state(a, a->curr_state <= clean);
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;
+ 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
+ * the array is readonly and we have cleared 'blocked'
+ * 2/ Try to remove the device if the array is writable, or can be
+ * made writable.
+ */
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,
mdi->curr_state);
check_degraded = 1;
- mdi->next_state = DS_REMOVE;
+ 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;
+ }
+ if (a->curr_state > readonly)
+ mdi->next_state |= DS_REMOVE;
}
}
- if (check_degraded) {
- // FIXME;
+ /* 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;
}
- a->container->ss->sync_metadata(a);
+ 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]);
/* 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", sync_actions[a->next_action]);
+ }
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_UNBLOCK) {
+ dprintf(" %d:-blocked", mdi->disk.raid_disk);
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.
+ }
+
+ if ((mdi->next_state & DS_REMOVE) && mdi->state_fd >= 0) {
+ int remove_result;
+
+ /* The kernel may not be able to immediately remove the
+ * disk. In that case we wait a little while and
+ * try again.
*/
- remove_err = write_attr("remove", mdi->state_fd);
- if (!remove_err) {
+ 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)
+ 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 || check_reshape) {
+ /* manager will do the actual check */
+ if (check_degraded)
+ a->check_degraded = 1;
+ if (check_reshape)
+ a->check_reshape = 1;
+ signal_manager();
+ }
+
if (deactivate)
a->container = NULL;
- return 1;
+ 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)
}
}
-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.
- */
-
- /* 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;
- }
+ 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);
- /* 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;
+ static unsigned int dirty_arrays = ~0; /* start at some non-zero value */
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 || a->to_remove) {
+ 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);
+ 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 || (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;
+ 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");
+ 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) {
- rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
+ 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, 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 (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) {
+ rv = 0;
+ dirty_arrays = 0;
+ 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 && !a->to_remove) {
+ int ret = read_and_act(a);
+ rv |= 1;
+ 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 && !(ret & ARRAY_DIRTY))
+ a->container = NULL; /* stop touching this array */
+ if (ret & ARRAY_BUSY)
+ container->retry_soon = 1;
}
- if (a->container)
- rv += read_and_act(a);
}
/* propagate failures across container members */
- for (a = aa; a ; a = a->next) {
- if (!a->container)
+ for (a = *aap; a ; a = a->next) {
+ if (!a->container || a->to_remove)
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);
}