Get failed disk count from array state
authorTomasz Majchrzak <tomasz.majchrzak@intel.com>
Wed, 31 May 2017 10:46:57 +0000 (12:46 +0200)
committerJes Sorensen <jsorensen@fb.com>
Mon, 5 Jun 2017 15:11:36 +0000 (11:11 -0400)
Recent commit has changed the way failed disks are counted. It breaks
recovery for external metadata arrays as failed disks are not part of
the array and have no corresponding entries is sysfs (they are only
reported for containers) so degraded arrays show no failed disks.

Recent commit overwrites GET_DEGRADED result prior to GET_STATE and it
is not set again if GET_STATE has not been requested. As GET_STATE
provides the same information as GET_DEGRADED, the latter is not needed
anymore. Remove GET_DEGRADED option and replace it with GET_STATE
option.

Don't count number of failed disks looking at sysfs entries but
calculate it at the end. Do it only for arrays as containers report
no disks, just spares.

Signed-off-by: Tomasz Majchrzak <tomasz.majchrzak@intel.com>
Signed-off-by: Jes Sorensen <jsorensen@fb.com>
Incremental.c
Monitor.c
managemon.c
mdadm.h
raid6check.c
sysfs.c

index 30dc7a2..6cf2174 100644 (file)
@@ -886,16 +886,10 @@ static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
                }
                sra = sysfs_read(-1, mp->devnm,
                                 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
-                                GET_DEGRADED|GET_COMPONENT|GET_VERSION);
-               if (!sra) {
-                       /* Probably a container - no degraded info */
-                       sra = sysfs_read(-1, mp->devnm,
-                                        GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
-                                        GET_COMPONENT|GET_VERSION);
-                       if (sra)
-                               sra->array.failed_disks = -1;
-               }
-               if (!sra)
+                                GET_COMPONENT|GET_VERSION);
+               if (sra)
+                       sra->array.failed_disks = -1;
+               else
                        continue;
                if (st == NULL) {
                        int i;
index 725f47d..bef2f1b 100644 (file)
--- a/Monitor.c
+++ b/Monitor.c
@@ -485,8 +485,8 @@ static int check_array(struct state *st, struct mdstat_ent *mdstat,
        if (st->devnm[0] == 0)
                strcpy(st->devnm, fd2devnm(fd));
 
-       sra = sysfs_read(-1, st->devnm, GET_LEVEL | GET_DISKS | GET_DEGRADED |
-                        GET_MISMATCH | GET_DEVS | GET_STATE);
+       sra = sysfs_read(-1, st->devnm, GET_LEVEL | GET_DISKS | GET_MISMATCH |
+                        GET_DEVS | GET_STATE);
        if (!sra)
                goto disappeared;
 
index a8df666..68f0c2d 100644 (file)
@@ -685,8 +685,8 @@ static void manage_new(struct mdstat_ent *mdstat,
 
        mdi = sysfs_read(-1, mdstat->devnm,
                         GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT|
-                        GET_DEGRADED|GET_SAFEMODE|
-                        GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|GET_LAYOUT);
+                        GET_SAFEMODE|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
+                        GET_LAYOUT);
 
        if (!mdi)
                return;
diff --git a/mdadm.h b/mdadm.h
index ec0a39e..ee9b837 100644 (file)
--- a/mdadm.h
+++ b/mdadm.h
@@ -637,7 +637,6 @@ enum sysfs_read_flags {
        GET_MISMATCH    = (1 << 5),
        GET_VERSION     = (1 << 6),
        GET_DISKS       = (1 << 7),
-       GET_DEGRADED    = (1 << 8),
        GET_SAFEMODE    = (1 << 9),
        GET_BITMAP_LOCATION = (1 << 10),
 
index 551f835..a8e6005 100644 (file)
@@ -562,7 +562,7 @@ int main(int argc, char *argv[])
                          GET_LEVEL|
                          GET_LAYOUT|
                          GET_DISKS|
-                         GET_DEGRADED |
+                         GET_STATE |
                          GET_COMPONENT|
                          GET_CHUNK|
                          GET_DEVS|
diff --git a/sysfs.c b/sysfs.c
index e47f5e4..78d2b52 100644 (file)
--- a/sysfs.c
+++ b/sysfs.c
@@ -162,18 +162,12 @@ struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
                        goto abort;
                sra->array.layout = strtoul(buf, NULL, 0);
        }
-       if (options & GET_DISKS) {
+       if (options & (GET_DISKS|GET_STATE)) {
                strcpy(base, "raid_disks");
                if (load_sys(fname, buf, sizeof(buf)))
                        goto abort;
                sra->array.raid_disks = strtoul(buf, NULL, 0);
        }
-       if (options & GET_DEGRADED) {
-               strcpy(base, "degraded");
-               if (load_sys(fname, buf, sizeof(buf)))
-                       goto abort;
-               sra->array.failed_disks = strtoul(buf, NULL, 0);
-       }
        if (options & GET_COMPONENT) {
                strcpy(base, "component_size");
                if (load_sys(fname, buf, sizeof(buf)))
@@ -359,10 +353,9 @@ struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
                        strcpy(dbase, "state");
                        if (load_sys(fname, buf, sizeof(buf)))
                                goto abort;
-                       if (strstr(buf, "faulty")) {
+                       if (strstr(buf, "faulty"))
                                dev->disk.state |= (1<<MD_DISK_FAULTY);
-                               sra->array.failed_disks++;
-                       } else {
+                       else {
                                sra->array.working_disks++;
                                if (strstr(buf, "in_sync")) {
                                        dev->disk.state |= (1<<MD_DISK_SYNC);
@@ -379,6 +372,11 @@ struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
                        dev->errors = strtoul(buf, NULL, 0);
                }
        }
+
+       if ((options & GET_STATE) && sra->array.raid_disks)
+               sra->array.failed_disks = sra->array.raid_disks -
+                       sra->array.active_disks - sra->array.spare_disks;
+
        closedir(dir);
        return sra;