]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - monitor.c
Make sure resync_start is initialised properly and maintained properly
[thirdparty/mdadm.git] / monitor.c
index 5a95149027da2f28244a5d8ff5b8728f7e39f7ef..314889657aaae88629241ab4ae69d80b4832284a 100644 (file)
--- a/monitor.c
+++ b/monitor.c
@@ -1,7 +1,7 @@
 
 #include "mdadm.h"
 #include "mdmon.h"
-
+#include <sys/syscall.h>
 #include <sys/select.h>
 #include <signal.h>
 
@@ -48,7 +48,7 @@ static int read_attr(char *buf, int len, int fd)
 }
 
 
-static int get_resync_start(struct active_array *a)
+int get_resync_start(struct active_array *a)
 {
        char buf[30];
        int n;
@@ -138,7 +138,9 @@ int read_dev_state(int fd)
 
 static void signal_manager(void)
 {
-       kill(getpid(), SIGUSR1);
+       /* 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
@@ -240,7 +242,7 @@ static int read_and_act(struct active_array *a)
                 * readonly ???
                 */
                get_resync_start(a);
-               printf("Found a readonly array at %llu\n", a->resync_start);
+//             printf("Found a readonly array at %llu\n", a->resync_start);
                if (a->resync_start == ~0ULL)
                        a->next_state = read_auto; /* array is clean */
                else {
@@ -249,7 +251,8 @@ static int read_and_act(struct active_array *a)
                }
        }
 
-       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
@@ -261,14 +264,19 @@ static int read_and_act(struct active_array *a)
                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;
                }
+               a->container->ss->set_array_state(a, 0);
        }
 
        for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
@@ -281,12 +289,17 @@ static int read_and_act(struct active_array *a)
        }
 
        a->container->ss->sync_metadata(a->container);
+       dprintf("%s: update[%d]: (", __func__, a->info.container_member);
 
        /* 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", array_states[a->next_state]);
+       }
        for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
                if (mdi->next_state == DS_REMOVE && mdi->state_fd >= 0) {
                        int remove_result;
@@ -296,15 +309,20 @@ static int read_and_act(struct active_array *a)
                         * disk, we can simply wait until the next event to try
                         * again.
                         */
+                       dprintf(" %d:-blocked", mdi->disk.raid_disk);
                        remove_result = write_attr("remove", mdi->state_fd);
                        if (remove_result > 0) {
+                               dprintf(" %d:removed", mdi->disk.raid_disk);
                                close(mdi->state_fd);
                                mdi->state_fd = -1;
                        }
                }
-               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;
@@ -357,61 +375,37 @@ static void reconcile_failed(struct active_array *aa, struct mdinfo *failed)
        }
 }
 
-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.
-        */
+       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);
 
-       /* 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;
-       }
-
-       /* 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;
-       }
-
-       return 0;
-}
-
-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);
+                       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 -1;
+       fprintf(stderr, ")\n");
 }
+#endif
 
-static int wait_and_act(struct supertype *container, int pfd,
-                       int monfd, int nowait)
+static int wait_and_act(struct supertype *container, int nowait)
 {
        fd_set rfds;
        int maxfd = 0;
@@ -422,7 +416,6 @@ static int wait_and_act(struct supertype *container, int pfd,
 
        FD_ZERO(&rfds);
 
-       add_fd(&rfds, &maxfd, pfd);
        for (ap = aap ; *ap ;) {
                a = *ap;
                /* once an array has been deactivated we want to
@@ -457,26 +450,25 @@ static int wait_and_act(struct supertype *container, int pfd,
                int fd = open(container->device_name, 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);
                        exit_now = 1;
                        signal_manager();
-                       remove_pidfile(container->devname);
                        exit(0);
                }
        }
 
        if (!nowait) {
-               rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
-
-               if (rv <= 0)
-                       return rv;
+               sigset_t set;
+               sigprocmask(SIG_UNBLOCK, NULL, &set);
+               sigdelset(&set, SIGUSR1);
+               rv = pselect(maxfd+1, &rfds, NULL, NULL, NULL, &set);
+               if (rv == -1 && errno == EINTR)
+                       rv = 0;
+               #ifdef DEBUG
+               dprint_wake_reasons(&rfds);
+               #endif
 
-               if (FD_ISSET(pfd, &rfds)) {
-                       int err = -1;
-
-                       if (read(pfd, &err, 1) > 0)
-                               err = handle_pipe(active_cmd, *aap);
-                       write(monfd, &err, 1);
-               }
        }
 
        if (update_queue) {
@@ -525,8 +517,7 @@ void do_monitor(struct supertype *container)
        int rv;
        int first = 1;
        do {
-               rv = wait_and_act(container, container->mgr_pipe[0],
-                                 container->mon_pipe[1], first);
+               rv = wait_and_act(container, first);
                first = 0;
        } while (rv >= 0);
 }