]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - monitor.c
Detail: clean up handing of the 'info' we load from superblock.
[thirdparty/mdadm.git] / monitor.c
index 15791f4e93ae79f073621b42ae8147fcb2149e31..59b4181954c24c0dc209e93c17c6ba391e2a4eab 100644 (file)
--- a/monitor.c
+++ b/monitor.c
@@ -1,8 +1,8 @@
 /*
  * 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,
@@ -66,20 +66,37 @@ static int read_attr(char *buf, int len, int fd)
        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;
+               return 0;
+       if (strncmp(buf, "none", 4) == 0)
+               return MaxSector;
+       else
+               return strtoull(buf, NULL, 10);
+}
 
-       a->resync_start = strtoull(buf, NULL, 10);
+static unsigned long long read_sync_completed(int fd)
+{
+       unsigned long long val;
+       char buf[50];
+       int n;
+       char *ep;
 
-       return 1;
-}
+       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)
 {
@@ -196,45 +213,53 @@ static void signal_manager(void)
 
 static int read_and_act(struct active_array *a)
 {
+       unsigned long long sync_completed;
        int check_degraded = 0;
        int deactivate = 0;
        struct mdinfo *mdi;
+       int dirty = 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)
+               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 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 mark clean
                 * in the metadata
                 */
                a->next_state = clean;
+               dirty = 1;
        }
        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;
        if (a->curr_state == readonly) {
                /* Well, I'm ready to handle things.  If readonly
                 * wasn't requested, transition to read-auto.
@@ -245,11 +270,12 @@ static int read_and_act(struct active_array *a)
                        /* 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
+                       else {
                                a->next_state = active; /* Now active for recovery etc */
+                               dirty = 1;
+                       }
                }
        }
 
@@ -261,7 +287,6 @@ static int read_and_act(struct active_array *a)
                 * 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;
        }
@@ -302,6 +327,22 @@ static int read_and_act(struct active_array *a)
                }
        }
 
+       /* 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 not
+        * handled here.
+        */
+       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 (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]);
@@ -361,7 +402,7 @@ static int read_and_act(struct active_array *a)
        if (deactivate)
                a->container = NULL;
 
-       return 1;
+       return dirty;
 }
 
 static struct mdinfo *
@@ -456,6 +497,7 @@ static int wait_and_act(struct supertype *container, int nowait)
 
                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);
 
@@ -469,14 +511,18 @@ static int wait_and_act(struct supertype *container, int nowait)
                 * 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 = open_dev_excl(container->devnum);
                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->devname);
                        exit_now = 1;
                        signal_manager();
                        exit(0);
@@ -488,7 +534,7 @@ static int wait_and_act(struct supertype *container, int nowait)
                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, NULL, &set);
                monitor_loop_cnt += 1;
                if (rv == -1 && errno == EINTR)
                        rv = 0;
@@ -527,32 +573,17 @@ static int wait_and_act(struct supertype *container, int nowait)
                        /* FIXME check if device->state_fd need to be cleared?*/
                        signal_manager();
                }
-               if (a->container)
-                       rv += read_and_act(a);
-               else
-                       continue;
-
-               /* when terminating stop manipulating the array after it is
-                * clean, but make sure read_and_act() is given a chance to
-                * handle 'active_idle'
-                */
-               switch (read_state(a->info.state_fd)) {
-                       case active:
-                       case active_idle:
-                       case suspended:
-                       case bad_word:
-                               is_dirty = 1;
-                               break;
-                       default:
-                               if (a->curr_state == active_idle)
-                                       is_dirty = 1;
-                               else
-                                       is_dirty = 0;
-                               break;
+               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 */
                }
-               dirty_arrays += is_dirty;
-               if (sigterm && !is_dirty)
-                       a->container = NULL; /* stop touching this array */
        }
 
        /* propagate failures across container members */