]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - Monitor.c
Add new mode: --incremental
[thirdparty/mdadm.git] / Monitor.c
index 306c43cfe8e118f4a2d05776475dafe1467d5d17..213e58d6facaf9b661af79433fa45c42e83ff098 100644 (file)
--- a/Monitor.c
+++ b/Monitor.c
@@ -1,7 +1,7 @@
 /*
  * mdadm - manage Linux "md" devices aka RAID arrays.
  *
- * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
+ * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
  *
  *
  *    This program is free software; you can redistribute it and/or modify
@@ -54,7 +54,7 @@ static char *percentalerts[] = {
 int Monitor(mddev_dev_t devlist,
            char *mailaddr, char *alert_cmd,
            int period, int daemonise, int scan, int oneshot,
-           int dosyslog, char *config, int test, char* pidfile)
+           int dosyslog, int test, char* pidfile)
 {
        /*
         * Every few seconds, scan every md device looking for changes
@@ -116,15 +116,15 @@ int Monitor(mddev_dev_t devlist,
        char *mailfrom = NULL;
 
        if (!mailaddr) {
-               mailaddr = conf_get_mailaddr(config);
+               mailaddr = conf_get_mailaddr();
                if (mailaddr && ! scan)
                        fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n",
                               mailaddr);
        }
-       mailfrom = conf_get_mailfrom(config);
+       mailfrom = conf_get_mailfrom();
 
        if (!alert_cmd) {
-               alert_cmd = conf_get_program(config);
+               alert_cmd = conf_get_program();
                if (alert_cmd && ! scan)
                        fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n",
                               alert_cmd);
@@ -163,7 +163,7 @@ int Monitor(mddev_dev_t devlist,
        }
 
        if (devlist == NULL) {
-               mddev_ident_t mdlist = conf_get_ident(config, NULL);
+               mddev_ident_t mdlist = conf_get_ident(NULL);
                for (; mdlist; mdlist=mdlist->next) {
                        struct state *st = malloc(sizeof *st);
                        if (st == NULL)
@@ -184,7 +184,7 @@ int Monitor(mddev_dev_t devlist,
        } else {
                mddev_dev_t dv;
                for (dv=devlist ; dv; dv=dv->next) {
-                       mddev_ident_t mdlist = conf_get_ident(config, dv->devname);
+                       mddev_ident_t mdlist = conf_get_ident(dv->devname);
                        struct state *st = malloc(sizeof *st);
                        if (st == NULL)
                                continue;
@@ -215,6 +215,7 @@ int Monitor(mddev_dev_t devlist,
                mdstat = mdstat_read(oneshot?0:1, 0);
 
                for (st=statelist; st; st=st->next) {
+                       struct { int state, major, minor; } info[MaxDisks];
                        mdu_array_info_t array;
                        struct mdstat_ent *mse = NULL, *mse2;
                        char *dev = st->devname;
@@ -303,12 +304,39 @@ int Monitor(mddev_dev_t devlist,
 
                        if (mse &&
                            mse->percent == -1 &&
-                           st->percent >= 0)
-                               alert("RebuildFinished", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
+                           st->percent >= 0) {
+                               /* Rebuild/sync/whatever just finished.
+                                * If there is a number in /mismatch_cnt,
+                                * we should report that.
+                                */
+                               struct sysarray *sra =
+                                      sysfs_read(-1, st->devnum, GET_MISMATCH);
+                               if (sra && sra->mismatch_cnt > 0) {
+                                       char cnt[40];
+                                       sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt);
+                                       alert("RebuildFinished", dev, cnt, mailaddr, mailfrom, alert_cmd, dosyslog);
+                               } else
+                                       alert("RebuildFinished", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
+                               if (sra)
+                                       free(sra);
+                       }
 
                        if (mse)
                                st->percent = mse->percent;
 
+
+                       for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks;
+                            i++) {
+                               mdu_disk_info_t disc;
+                               if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
+                                       info[i].state = disc.state;
+                                       info[i].major = disc.major;
+                                       info[i].minor = disc.minor;
+                               } else
+                                       info[i].major = info[i].minor = 0;
+                       }
+                       close(fd);
+
                        for (i=0; i<MaxDisks; i++) {
                                mdu_disk_info_t disc;
                                int newstate=0;
@@ -318,9 +346,12 @@ int Monitor(mddev_dev_t devlist,
                                if (i > array.raid_disks + array.nr_disks) {
                                        newstate = 0;
                                        disc.major = disc.minor = 0;
-                               } else if (ioctl(fd, GET_DISK_INFO, &disc)>= 0) {
-                                       newstate = disc.state;
-                                       dv = map_dev(disc.major, disc.minor, 1);
+                               } else if (info[i].major || info[i].minor) {
+                                       newstate = info[i].state;
+                                       dv = map_dev(info[i].major, info[i].minor, 1);
+                                       disc.state = newstate;
+                                       disc.major = info[i].major;
+                                       disc.minor = info[i].minor;
                                } else if (mse &&  mse->pattern && i < strlen(mse->pattern)) {
                                        switch(mse->pattern[i]) {
                                        case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
@@ -355,7 +386,6 @@ int Monitor(mddev_dev_t devlist,
                                st->devstate[i] = disc.state;
                                st->devid[i] = makedev(disc.major, disc.minor);
                        }
-                       close(fd);
                        st->active = array.active_disks;
                        st->working = array.working_disks;
                        st->spare = array.spare_disks;
@@ -383,6 +413,7 @@ int Monitor(mddev_dev_t devlist,
                                            ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
                                                /* no such array */
                                                if (fd >=0) close(fd);
+                                               put_md_name(st->devname);
                                                free(st->devname);
                                                free(st);
                                                continue;
@@ -493,6 +524,7 @@ static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mail
             strncmp(event, "Degrade", 7)==0)) {
                FILE *mp = popen(Sendmail, "w");
                if (mp) {
+                       FILE *mdstat;
                        char hname[256];
                        gethostname(hname, sizeof(hname));
                        signal(SIGPIPE, SIG_IGN);
@@ -508,10 +540,22 @@ static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mail
 
                        fprintf(mp, "A %s event had been detected on md device %s.\n\n", event, dev);
 
-                       if (disc)
+                       if (disc && disc[0] != ' ')
                                fprintf(mp, "It could be related to component device %s.\n\n", disc);
+                       if (disc && disc[0] == ' ')
+                               fprintf(mp, "Extra information:%s.\n\n", disc);
 
                        fprintf(mp, "Faithfully yours, etc.\n");
+
+                       mdstat = fopen("/proc/mdstat", "r");
+                       if (mdstat) {
+                               char buf[8192];
+                               int n;
+                               fprintf(mp, "\nP.S. The /proc/mdstat file currently contains the following:\n\n");
+                               while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
+                                       n=fwrite(buf, 1, n, mp); /* yes, i don't care about the result */
+                               fclose(mdstat);
+                       }
                        fclose(mp);
                }
 
@@ -541,3 +585,38 @@ static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mail
                        syslog(priority, "%s event detected on md device %s", event, dev);
        }
 }
+
+/* Not really Monitor but ... */
+int Wait(char *dev)
+{
+       struct stat stb;
+       int devnum;
+       int rv = 1;
+
+       if (stat(dev, &stb) != 0) {
+               fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
+                       strerror(errno));
+               return 2;
+       }
+       if (major(stb.st_rdev) == MD_MAJOR)
+               devnum = minor(stb.st_rdev);
+       else
+               devnum = -1-(minor(stb.st_rdev)/64);
+
+       while(1) {
+               struct mdstat_ent *ms = mdstat_read(1, 0);
+               struct mdstat_ent *e;
+
+               for (e=ms ; e; e=e->next)
+                       if (e->devnum == devnum)
+                               break;
+
+               if (!e || e->percent < 0) {
+                       free_mdstat(ms);
+                       return rv;
+               }
+               free(ms);
+               rv = 0;
+               mdstat_wait(5);
+       }
+}