]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - Assemble.c
Use O_EXCL when opening component devices to be assembled into an array
[thirdparty/mdadm.git] / Assemble.c
index a2ff502390089664ca4a242ddf1b4676879652a1..ace5717c310adb7dc932b21dd30bc67756025df9 100644 (file)
@@ -98,22 +98,26 @@ int Assemble(char *mddev, int mdfd,
        mdp_super_t first_super, super;
        struct {
                char *devname;
-               int major, minor;
-               int oldmajor, oldminor;
+               unsigned int major, minor;
+               unsigned int oldmajor, oldminor;
                long long events;
                time_t utime;
                int uptodate;
+               int state;
                int raid_disk;
        } *devices;
-       int *best; /* indexed by raid_disk */
-       int devcnt = 0, okcnt, sparecnt;
-       int i;
+       int *best = NULL; /* indexed by raid_disk */
+       unsigned int bestcnt = 0;
+       int devcnt = 0;
+       unsigned int okcnt, sparecnt;
+       unsigned int req_cnt;
+       unsigned int i;
        int most_recent = 0;
        int chosen_drive;
        int change = 0;
        int inargv = 0;
        int start_partial_ok = force || devlist==NULL;
-       int num_devs;
+       unsigned int num_devs;
        mddev_dev_t tmpdev;
        
        vers = md_get_version(mdfd);
@@ -183,10 +187,13 @@ int Assemble(char *mddev, int mdfd,
                devlist = devlist->next;
 
                if (ident->devices &&
-                   !match_oneof(ident->devices, devname))
+                   !match_oneof(ident->devices, devname)) {
+                       if (inargv || verbose)
+                               fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
                        continue;
+               }
                
-               dfd = open(devname, O_RDONLY, 0);
+               dfd = open(devname, O_RDONLY|O_EXCL, 0);
                if (dfd < 0) {
                        if (inargv || verbose)
                                fprintf(stderr, Name ": cannot open device %s: %s\n",
@@ -218,21 +225,21 @@ int Assemble(char *mddev, int mdfd,
                                        devname);
                        continue;
                }
-               if (ident->super_minor >= 0 &&
+               if (ident->super_minor != UnSet &&
                    (!havesuper || ident->super_minor != super.md_minor)) {
                        if (inargv || verbose)
                                fprintf(stderr, Name ": %s has wrong super-minor.\n",
                                        devname);
                        continue;
                }
-               if (ident->level != -10 &&
-                   (!havesuper|| ident->level != super.level)) {
+               if (ident->level != UnSet &&
+                   (!havesuper|| ident->level != (int)super.level)) {
                        if (inargv || verbose)
                                fprintf(stderr, Name ": %s has wrong raid level.\n",
                                        devname);
                        continue;
                }
-               if (ident->raid_disks != -1 &&
+               if (ident->raid_disks != UnSet &&
                    (!havesuper || ident->raid_disks!= super.raid_disks)) {
                        if (inargv || verbose)
                                fprintf(stderr, Name ": %s requires wrong number of drives.\n",
@@ -281,13 +288,45 @@ int Assemble(char *mddev, int mdfd,
                        if (strcmp(update, "super-minor") ==0) {
                                struct stat stb2;
                                fstat(mdfd, &stb2);
-                               super.md_minor = MINOR(stb2.st_rdev);
+                               super.md_minor = minor(stb2.st_rdev);
                                if (verbose)
                                        fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
                                                devname, super.md_minor);
                        }
+                       if (strcmp(update, "summaries") == 0) {
+                               /* set nr_disks, active_disks, working_disks,
+                                * failed_disks, spare_disks based on disks[] 
+                                * array in superblock.
+                                * Also make sure extra slots aren't 'failed'
+                                */
+                               super.nr_disks = super.active_disks =
+                                       super.working_disks = super.failed_disks =
+                                       super.spare_disks = 0;
+                               for (i=0; i < MD_SB_DISKS ; i++) 
+                                       if (super.disks[i].major ||
+                                           super.disks[i].minor) {
+                                               int state = super.disks[i].state;
+                                               if (state & (1<<MD_DISK_REMOVED))
+                                                       continue;
+                                               super.nr_disks++;
+                                               if (state & (1<<MD_DISK_ACTIVE))
+                                                       super.active_disks++;
+                                               if (state & (1<<MD_DISK_FAULTY))
+                                                       super.failed_disks++;
+                                               else
+                                                       super.working_disks++;
+                                               if (state == 0)
+                                                       super.spare_disks++;
+                                       } else if (i >= super.raid_disks && super.disks[i].number == 0)
+                                               super.disks[i].state = 0;
+                       }
+                       if (strcmp(update, "resync") == 0) {
+                               /* make sure resync happens */
+                               super.state &= ~(1<<MD_SB_CLEAN);
+                               super.recovery_cp = 0;
+                       }
                        super.sb_csum = calc_sb_csum(&super);
-                       dfd = open(devname, O_RDWR, 0);
+                       dfd = open(devname, O_RDWR|O_EXCL, 0);
                        if (dfd < 0) 
                                fprintf(stderr, Name ": Cannot open %s for superblock update\n",
                                        devname);
@@ -302,29 +341,43 @@ int Assemble(char *mddev, int mdfd,
                        fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
                                devname, mddev, super.this_disk.raid_disk);
                devices[devcnt].devname = devname;
-               devices[devcnt].major = MAJOR(stb.st_rdev);
-               devices[devcnt].minor = MINOR(stb.st_rdev);
+               devices[devcnt].major = major(stb.st_rdev);
+               devices[devcnt].minor = minor(stb.st_rdev);
                devices[devcnt].oldmajor = super.this_disk.major;
                devices[devcnt].oldminor = super.this_disk.minor;
                devices[devcnt].events = md_event(&super);
                devices[devcnt].utime = super.utime;
                devices[devcnt].raid_disk = super.this_disk.raid_disk;
                devices[devcnt].uptodate = 0;
+               devices[devcnt].state = super.this_disk.state;
                if (most_recent < devcnt) {
                        if (devices[devcnt].events
                            > devices[most_recent].events)
                                most_recent = devcnt;
                }
-               if (super.level == -4) 
+               if ((int)super.level == -4) 
                        /* with multipath, the raid_disk from the superblock is meaningless */
                        i = devcnt;
                else
                        i = devices[devcnt].raid_disk;
-               if (i>=0 && i < num_devs)
+               if (i < 10000) {
+                       if (i >= bestcnt) {
+                               unsigned int newbestcnt = i+10;
+                               int *newbest = malloc(sizeof(int)*newbestcnt);
+                               unsigned int c;
+                               for (c=0; c < newbestcnt; c++)
+                                       if (c < bestcnt)
+                                               newbest[c] = best[c];
+                                       else
+                                               newbest[c] = -1;
+                               if (best)free(best);
+                               best = newbest;
+                               bestcnt = newbestcnt;
+                       }
                        if (best[i] == -1
                            || devices[best[i]].events < devices[devcnt].events)
                                best[i] = devcnt;
-
+               }
                devcnt++;
        }
 
@@ -338,10 +391,19 @@ int Assemble(char *mddev, int mdfd,
         */
        okcnt = 0;
        sparecnt=0;
-       for (i=0; i< num_devs ;i++) {
+       for (i=0; i< bestcnt ;i++) {
                int j = best[i];
                int event_margin = !force;
                if (j < 0) continue;
+               /* note: we ignore error flags in multipath arrays
+                * as they don't make sense
+                */
+               if ((int)first_super.level != -4)
+                       if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
+                               if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
+                                       sparecnt++;
+                               continue;
+                       }
                if (devices[j].events+event_margin >=
                    devices[most_recent].events) {
                        devices[j].uptodate = 1;
@@ -358,7 +420,7 @@ int Assemble(char *mddev, int mdfd,
                 */
                int fd;
                chosen_drive = -1;
-               for (i=0; i<first_super.raid_disks; i++) {
+               for (i=0; i<first_super.raid_disks && i < bestcnt; i++) {
                        int j = best[i];
                        if (j>=0 &&
                            !devices[j].uptodate &&
@@ -373,7 +435,7 @@ int Assemble(char *mddev, int mdfd,
                        devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
                        (int)(devices[chosen_drive].events),
                        (int)(devices[most_recent].events));
-               fd = open(devices[chosen_drive].devname, O_RDWR);
+               fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
                if (fd < 0) {
                        fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
                                devices[chosen_drive].devname);
@@ -391,7 +453,7 @@ int Assemble(char *mddev, int mdfd,
                super.events_lo = (devices[most_recent].events)&0xFFFFFFFF;
                if (super.level == 5 || super.level == 4) {
                        /* need to force clean */
-                       super.state = 0;
+                       super.state = (1<<MD_SB_CLEAN);
                }
                super.sb_csum = calc_sb_csum(&super);
 /*DRYRUN*/     if (store_super(fd, &super)) {
@@ -414,7 +476,7 @@ int Assemble(char *mddev, int mdfd,
         * superblock.
         */
        chosen_drive = -1;
-       for (i=0; chosen_drive < 0 && i<num_devs; i++) {
+       for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
                int j = best[i];
                int fd;
                if (j<0)
@@ -422,7 +484,7 @@ int Assemble(char *mddev, int mdfd,
                if (!devices[j].uptodate)
                        continue;
                chosen_drive = j;
-               if ((fd=open(devices[j].devname, O_RDONLY))< 0) {
+               if ((fd=open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
                        fprintf(stderr, Name ": Cannot open %s: %s\n",
                                devices[j].devname, strerror(errno));
                        return 1;
@@ -436,9 +498,9 @@ int Assemble(char *mddev, int mdfd,
                close(fd);
        }
 
-       for (i=0; i<num_devs; i++) {
+       for (i=0; i<bestcnt; i++) {
                int j = best[i];
-               int desired_state;
+               unsigned int desired_state;
 
                if (i < super.raid_disks)
                        desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
@@ -484,12 +546,17 @@ This doesnt work yet
                                i, mddev);
                }
        }
+       if (force && (super.level == 4 || super.level == 5) && 
+           okcnt == super.raid_disks-1) {
+               super.state = (1<< MD_SB_CLEAN);
+               change |= 2;
+       }
 
        if ((force && (change & 2))
            || (old_linux && (change & 1))) {
                int fd;
                super.sb_csum = calc_sb_csum(&super);
-               fd = open(devices[chosen_drive].devname, O_RDWR);
+               fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
                if (fd < 0) {
                        fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
                                devices[chosen_drive].devname);
@@ -505,6 +572,17 @@ This doesnt work yet
                change = 0;
        }
 
+       /* count number of in-sync devices according to the superblock.
+        * We must have this number to start the array without -s or -R
+        */
+       req_cnt = 0;
+       for (i=0; i<MD_SB_DISKS; i++)
+               if ((first_super.disks[i].state & (1<<MD_DISK_SYNC)) &&
+                   (first_super.disks[i].state & (1<<MD_DISK_ACTIVE)) &&
+                   !(first_super.disks[i].state & (1<<MD_DISK_FAULTY)))
+                       req_cnt ++;
+                                                                           
+
        /* Almost ready to actually *do* something */
        if (!old_linux) {
                if (ioctl(mdfd, SET_ARRAY_INFO, NULL) != 0) {
@@ -513,16 +591,16 @@ This doesnt work yet
                        return 1;
                }
                /* First, add the raid disks, but add the chosen one last */
-               for (i=0; i<= num_devs; i++) {
+               for (i=0; i<= bestcnt; i++) {
                        int j;
-                       if (i < num_devs) {
+                       if (i < bestcnt) {
                                j = best[i];
                                if (j == chosen_drive)
                                        continue;
                        } else
                                j = chosen_drive;
 
-                       if (j >= 0 && devices[j].uptodate) {
+                       if (j >= 0 /* && devices[j].uptodate */) {
                                mdu_disk_info_t disk;
                                memset(&disk, 0, sizeof(disk));
                                disk.major = devices[j].major;
@@ -532,7 +610,7 @@ This doesnt work yet
                                                devices[j].devname,
                                                mddev,
                                                strerror(errno));
-                                       if (i < first_super.raid_disks)
+                                       if (i < first_super.raid_disks || i == bestcnt)
                                                okcnt--;
                                        else
                                                sparecnt--;
@@ -546,12 +624,14 @@ This doesnt work yet
                
                if (runstop == 1 ||
                    (runstop == 0 && 
-                    ( first_super.raid_disks == okcnt
-                      || (start_partial_ok && enough(first_super.level, first_super.raid_disks, okcnt)))
-                           )) {
+                    ( enough(first_super.level, first_super.raid_disks, okcnt) &&
+                      (okcnt >= req_cnt || start_partial_ok)
+                            ))) {
                        if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
                                fprintf(stderr, Name ": %s has been started with %d drive%s",
                                        mddev, okcnt, okcnt==1?"":"s");
+                               if (okcnt < first_super.raid_disks) 
+                                       fprintf(stderr, " (out of %d)", first_super.raid_disks);
                                if (sparecnt)
                                        fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
                                fprintf(stderr, ".\n");
@@ -566,8 +646,18 @@ This doesnt work yet
                                mddev, okcnt, okcnt==1?"":"s");
                        return 0;
                }
-               fprintf(stderr, Name ": %s assembled from %d drive%s - not enough to start it (use --run to insist).\n",
-                       mddev, okcnt, okcnt==1?"":"s");
+               fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
+               if (sparecnt)
+                       fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
+               if (!enough(first_super.level, first_super.raid_disks, okcnt))
+                       fprintf(stderr, " - not enough to start the array.\n");
+               else {
+                       if (req_cnt == first_super.raid_disks)
+                               fprintf(stderr, " - need all %d to start it", req_cnt);
+                       else
+                               fprintf(stderr, " - need %d of %d to start", req_cnt, first_super.raid_disks);
+                       fprintf(stderr, " (use --run to insist).\n");
+               }
                return 1;
        } else {
                /* The "chosen_drive" is a good choice, and if necessary, the superblock has
@@ -575,7 +665,7 @@ This doesnt work yet
                 * so we can just start the array
                 */
                unsigned long dev;
-               dev = MKDEV(devices[chosen_drive].major,
+               dev = makedev(devices[chosen_drive].major,
                            devices[chosen_drive].minor);
                if (ioctl(mdfd, START_ARRAY, dev)) {
                    fprintf(stderr, Name ": Cannot start array: %s\n",