]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - Assemble.c
imsm: PPL support
[thirdparty/mdadm.git] / Assemble.c
index 06e122df6258a8e3c6a8c604182a4e512a979971..8e55b49fa4064e3cf8a50346477d75d5baf3b25b 100644 (file)
@@ -1,7 +1,7 @@
 /*
  * mdadm - manage Linux "md" devices aka RAID arrays.
  *
- * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
+ * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
  *
  *
  *    This program is free software; you can redistribute it and/or modify
@@ -250,7 +250,9 @@ static int select_devices(struct mddev_dev *devlist,
                                        pr_err("no recogniseable superblock on %s\n",
                                               devname);
                                tmpdev->used = 2;
-                       } else if (tst->ss->load_super(tst,dfd, NULL)) {
+                       } else if ((tst->ignore_hw_compat = 0),
+                                  tst->ss->load_super(tst, dfd,
+                                                      report_mismatch ? devname : NULL)) {
                                if (report_mismatch)
                                        pr_err("no RAID superblock on %s\n",
                                               devname);
@@ -597,18 +599,9 @@ static int load_devices(struct devs *devices, char *devmap,
                        int err;
                        fstat(mdfd, &stb2);
 
-                       if (strcmp(c->update, "uuid")==0 &&
-                           !ident->uuid_set) {
-                               int rfd;
-                               if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
-                                   read(rfd, ident->uuid, 16) != 16) {
-                                       *(__u32*)(ident->uuid) = random();
-                                       *(__u32*)(ident->uuid+1) = random();
-                                       *(__u32*)(ident->uuid+2) = random();
-                                       *(__u32*)(ident->uuid+3) = random();
-                               }
-                               if (rfd >= 0) close(rfd);
-                       }
+                       if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set)
+                               random_uuid((__u8 *)ident->uuid);
+
                        dfd = dev_open(devname,
                                       tmpdev->disposition == 'I'
                                       ? O_RDWR : (O_RDWR|O_EXCL));
@@ -635,6 +628,19 @@ static int load_devices(struct devs *devices, char *devmap,
 
                        if (strcmp(c->update, "byteorder") == 0)
                                err = 0;
+                       else if (strcmp(c->update, "home-cluster") == 0) {
+                               tst->cluster_name = c->homecluster;
+                               err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
+                       } else if (strcmp(c->update, "nodes") == 0) {
+                               tst->nodes = c->nodes;
+                               err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
+                       } else if (strcmp(c->update, "revert-reshape") == 0 &&
+                                  c->invalid_backup)
+                               err = tst->ss->update_super(tst, content,
+                                                           "revert-reshape-nobackup",
+                                                           devname, c->verbose,
+                                                           ident->uuid_set,
+                                                           c->homehost);
                        else
                                err = tst->ss->update_super(tst, content, c->update,
                                                            devname, c->verbose,
@@ -727,7 +733,7 @@ static int load_devices(struct devs *devices, char *devmap,
                        i = devcnt;
                else
                        i = devices[devcnt].i.disk.raid_disk;
-               if (i+1 == 0) {
+               if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
                        if (nextspare < content->array.raid_disks*2)
                                nextspare = content->array.raid_disks*2;
                        i = nextspare++;
@@ -905,7 +911,6 @@ static int force_array(struct mdinfo *content,
                avail[chosen_drive] = 1;
                okcnt++;
                tst->ss->free_super(tst);
-
                /* If there are any other drives of the same vintage,
                 * add them in as well.  We can't lose and we might gain
                 */
@@ -936,6 +941,7 @@ static int start_array(int mdfd,
                       unsigned int okcnt,
                       unsigned int sparecnt,
                       unsigned int rebuilding_cnt,
+                      unsigned int journalcnt,
                       struct context *c,
                       int clean, char *avail,
                       int start_partial_ok,
@@ -947,6 +953,15 @@ static int start_array(int mdfd,
        int i;
        unsigned int req_cnt;
 
+       if (content->journal_device_required && (content->journal_clean == 0)) {
+               if (!c->force) {
+                       pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
+                       return 1;
+               }
+               pr_err("Journal is missing or stale, starting array read only.\n");
+               c->readonly = 1;
+       }
+
        rv = set_array_info(mdfd, st, content);
        if (rv && !err_ok) {
                pr_err("failed to set array info for %s: %s\n",
@@ -1024,7 +1039,8 @@ static int start_array(int mdfd,
        if (content->array.level == LEVEL_CONTAINER) {
                if (c->verbose >= 0) {
                        pr_err("Container %s has been assembled with %d drive%s",
-                              mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
+                              mddev, okcnt+sparecnt+journalcnt,
+                              okcnt+sparecnt+journalcnt==1?"":"s");
                        if (okcnt < (unsigned)content->array.raid_disks)
                                fprintf(stderr, " (out of %d)",
                                        content->array.raid_disks);
@@ -1110,6 +1126,8 @@ static int start_array(int mdfd,
                                        fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
                                if (sparecnt)
                                        fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
+                               if (content->journal_clean)
+                                       fprintf(stderr, " and %d journal", journalcnt);
                                fprintf(stderr, ".\n");
                        }
                        if (content->reshape_active &&
@@ -1281,7 +1299,8 @@ int Assemble(struct supertype *st, char *mddev,
        int *best = NULL; /* indexed by raid_disk */
        int bestcnt = 0;
        int devcnt;
-       unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
+       unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
+       int journal_clean = 0;
        int i;
        int was_forced = 0;
        int most_recent = 0;
@@ -1522,6 +1541,7 @@ try_again:
        okcnt = 0;
        replcnt = 0;
        sparecnt=0;
+       journalcnt=0;
        rebuilding_cnt=0;
        for (i=0; i< bestcnt; i++) {
                int j = best[i];
@@ -1532,8 +1552,13 @@ try_again:
                /* note: we ignore error flags in multipath arrays
                 * as they don't make sense
                 */
-               if (content->array.level != LEVEL_MULTIPATH)
-                       if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
+               if (content->array.level != LEVEL_MULTIPATH) {
+                       if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
+                               if (content->journal_device_required)
+                                       journalcnt++;
+                               else    /* unexpected journal, mark as faulty */
+                                       devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
+                       } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
                                if (!(devices[j].i.disk.state
                                      & (1<<MD_DISK_FAULTY))) {
                                        devices[j].uptodate = 1;
@@ -1541,6 +1566,7 @@ try_again:
                                }
                                continue;
                        }
+               }
                /* If this device thinks that 'most_recent' has failed, then
                 * we must reject this device.
                 */
@@ -1564,6 +1590,8 @@ try_again:
                    devices[most_recent].i.events
                        ) {
                        devices[j].uptodate = 1;
+                       if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
+                               journal_clean = 1;
                        if (i < content->array.raid_disks * 2) {
                                if (devices[j].i.recovery_start == MaxSector ||
                                    (content->reshape_active &&
@@ -1575,7 +1603,7 @@ try_again:
                                                replcnt++;
                                } else
                                        rebuilding_cnt++;
-                       } else
+                       } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
                                sparecnt++;
                }
        }
@@ -1635,11 +1663,15 @@ try_again:
 #ifndef MDASSEMBLE
        sysfs_init(content, mdfd, NULL);
 #endif
+       /* after reload context, store journal_clean in context */
+       content->journal_clean = journal_clean;
        for (i=0; i<bestcnt; i++) {
                int j = best[i];
                unsigned int desired_state;
 
-               if (i >= content->array.raid_disks * 2)
+               if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
+                       desired_state = (1<<MD_DISK_JOURNAL);
+               else if (i >= content->array.raid_disks * 2)
                        desired_state = 0;
                else if (i & 1)
                        desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
@@ -1786,7 +1818,7 @@ try_again:
        rv = start_array(mdfd, mddev, content,
                         st, ident, best, bestcnt,
                         chosen_drive, devices, okcnt, sparecnt,
-                        rebuilding_cnt,
+                        rebuilding_cnt, journalcnt,
                         c,
                         clean, avail, start_partial_ok,
                         pre_exist != NULL,
@@ -1839,10 +1871,9 @@ int assemble_container_content(struct supertype *st, int mdfd,
        struct mdinfo *dev, *sra, *dev2;
        int working = 0, preexist = 0;
        int expansion = 0;
-       struct map_ent *map = NULL;
        int old_raid_disks;
        int start_reshape;
-       char *avail = NULL;
+       char *avail;
        int err;
 
        sysfs_init(content, mdfd, NULL);
@@ -1855,8 +1886,7 @@ int assemble_container_content(struct supertype *st, int mdfd,
                    content->text_version[0] == '/')
                        content->text_version[0] = '-';
                if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
-                       if (sra)
-                               sysfs_free(sra);
+                       sysfs_free(sra);
                        return 1;
                }
        }
@@ -1909,10 +1939,57 @@ int assemble_container_content(struct supertype *st, int mdfd,
                free(avail);
                return 1;/* Nothing new, don't try to start */
        }
-       map_update(&map, fd2devnm(mdfd),
-                  content->text_version,
+       map_update(NULL, fd2devnm(mdfd), content->text_version,
                   content->uuid, chosen_name);
 
+       if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
+           st->ss->validate_ppl) {
+               content->array.state |= 1;
+               err = 0;
+
+               for (dev = content->devs; dev; dev = dev->next) {
+                       int dfd;
+                       char *devpath;
+                       int ret;
+
+                       ret = st->ss->validate_ppl(st, content, dev);
+                       if (ret == 0)
+                               continue;
+
+                       if (ret < 0) {
+                               err = 1;
+                               break;
+                       }
+
+                       if (!c->force) {
+                               pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
+                                       chosen_name);
+                               content->array.state &= ~1;
+                               avail[dev->disk.raid_disk] = 0;
+                               break;
+                       }
+
+                       /* have --force - overwrite the invalid ppl */
+                       devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
+                       dfd = dev_open(devpath, O_RDWR);
+                       if (dfd < 0) {
+                               pr_err("Failed to open %s\n", devpath);
+                               err = 1;
+                               break;
+                       }
+
+                       err = st->ss->write_init_ppl(st, content, dfd);
+                       close(dfd);
+
+                       if (err)
+                               break;
+               }
+
+               if (err) {
+                       free(avail);
+                       return err;
+               }
+       }
 
        if (enough(content->array.level, content->array.raid_disks,
                   content->array.layout, content->array.state & 1, avail) == 0) {