vldb-check-fix-more-kinds-of-corruption-20081007

LICENSE IPL10

work harder to ferret out corruption; still reruns itself so if 2 hash chains
are comingled they get split back out correctly without having to flush state
continuously

diff --git a/src/vlserver/vldb_check.c b/src/vlserver/vldb_check.c
index 1c6567d89703ef1c999fc092b2c3143dd5ecc985..05602d009889eb0dcd30fbf306d99bc0a514cb66 100644 (file)
--- a/src/vlserver/vldb_check.c
+++ b/src/vlserver/vldb_check.c
@@ -26,6 +26,16 @@
 #define REFBK 0x4000            /* linked from something (BK) */
 #define REFN  0x8000            /* linked from something (name) */
 
+#define MULTRW 0x10000         /* multiply-chained (RW) */
+#define MULTRO 0x20000         /* multiply-chained (RO) */
+#define MULTBK 0x40000         /* multiply-chained (BK) */
+#define MULTN  0x80000         /* multiply-chained (name) */
+
+#define MISRWH 0x100000          /* mischained (RW) */
+#define MISROH 0x200000          /* mischained (RO) */
+#define MISBKH 0x400000          /* mischained (BK) */
+#define MISNH  0x800000          /* mischained (name) */
+
 #define vldbread(x,y,z) vldbio(x,y,z,0)
 #define vldbwrite(x,y,z) vldbio(x,y,z,1)
 
@@ -62,6 +72,8 @@ int fd;
 int listentries, listservers, listheader, listuheader, verbose;
 
 int fix = 0;
+int fixed = 0;
+int passes = 0;
 
 struct er {
     long addr;
@@ -404,7 +416,7 @@ writeentry(afs_int32 addr, struct nvlentry *vlentryp)
 
     if (verbose)
        printf("Writing back entry at addr %u\n", addr);
-
+    fixed++;
     for (i = 0; i < MAXTYPES; i++)
        vlentryp->volumeId[i] = htonl(vlentryp->volumeId[i]);
     vlentryp->flags = htonl(vlentryp->flags);
@@ -663,10 +675,10 @@ FollowNameHash(struct vlheader *header)
                printf
                    ("Name Hash %d: Bad entry '%s': Already in the name hash\n",
                     i, vlentry.name);
+               record[rindex].type |= MULTN;
                break;
            }
            record[rindex].type |= NH;
-
            record[rindex].type |= REFN;
 
            chainlength++;
@@ -677,6 +689,7 @@ FollowNameHash(struct vlheader *header)
                printf
                    ("Name Hash %d: Bad entry '%s': Incorrect name hash chain (should be in %d)\n",
                     i, vlentry.name, NameHash(vlentry.name));
+               record[rindex].type |= MULTN;
            }
        }
        if (chainlength > longest)
@@ -703,7 +716,7 @@ FollowIdHash(struct vlheader *header)
     int count = 0, longest = 0, shortest = -1, chainlength;
     struct nvlentry vlentry;
     afs_uint32 addr;
-    afs_int32 i, j, hash, type, rindex, ref;
+    afs_int32 i, j, hash, type, rindex, ref, badref, badhash;
 
     /* Now follow the RW, RO, and BK Hash Tables */
     if (verbose)
@@ -711,6 +724,8 @@ FollowIdHash(struct vlheader *header)
     for (i = 0; i < MAXTYPES; i++) {
        hash = ((i == 0) ? RWH : ((i == 1) ? ROH : BKH));
        ref = ((i == 0) ? REFRW : ((i == 1) ? REFRO : REFBK));
+       badref = ((i == 0) ? MULTRW : ((i == 1) ? MULTRO : MULTBK));
+       badhash = ((i == 0) ? MULTRW : ((i == 1) ? MULTRO : MULTBK));
        count = longest = 0;
        shortest = -1;
 
@@ -734,8 +749,9 @@ FollowIdHash(struct vlheader *header)
                }
                if (record[rindex].type & hash) {
                    printf
-                       ("%s Id Hash %d: Bad entry '%s': Already in the the hash table\n",
+                       ("%s Id Hash %d: Bad entry '%s': Already in the hash table\n",
                         vtype(i), j, vlentry.name);
+                   record[rindex].type |= badref;
                    break;
                }
                record[rindex].type |= hash;
@@ -750,6 +766,8 @@ FollowIdHash(struct vlheader *header)
                        ("%s Id Hash %d: Bad entry '%s': Incorrect Id hash chain (should be in %d)\n",
                         vtype(i), j, vlentry.name,
                         IdHash(vlentry.volumeId[i]));
+                   record[rindex].type |= badhash;
+                   printf("%d: %x\n", rindex, record[rindex].type);
                }
            }
 
@@ -1026,10 +1044,10 @@ CheckIpAddrs(struct vlheader *header)
 
 void 
 FixBad(afs_uint32 idx, afs_uint32 addr, afs_uint32 type, afs_uint32 tmp, 
-       struct nvlentry *vlentry) {
+       struct nvlentry *vlentry, afs_uint32 hash) {
     SetHashEnd(addr, type, tmp);
-    printf("linked unlinked chain %u (index %d) to end of chain\n", 
-          tmp, ADDR(tmp));
+    printf("linked unlinked chain %u (index %d) to end of chain %d for %s hash\n", 
+          tmp, ADDR(tmp), hash, type==NH?"Name":(type==RWH?"RW":(type==ROH?"RO":"BK")));
 }
 
 int
@@ -1038,7 +1056,7 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
     char *dbfile;
     afs_int32 maxentries, type, tmp;
     struct vlheader header;
-    struct nvlentry vlentry;
+    struct nvlentry vlentry, vlentry2;
     int i, j, help = 0;
 
     dbfile = as->parms[0].items->data; /* -database */
@@ -1049,6 +1067,7 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
     verbose = (as->parms[5].items ? 1 : 0);    /* -verbose  */
     fix = (as->parms[6].items ? 1 : 0);        /* -fix  */
 
+ restart:
     /* open the vldb database file */
     fd = open(dbfile, (fix > 0)?O_RDWR:O_RDONLY, 0);
     if (fd < 0) {
@@ -1085,7 +1104,7 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
     /* Follow the chain of free entries */
     FollowFreeChain(&header);
 
-    /* Now check the record we have been keeping for inconsistancies
+    /* Now check the record we have been keeping for inconsistencies
      * For valid vlentries, also check that the server we point to is 
      * valid (the serveraddrs array).
      */
@@ -1094,7 +1113,7 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
     for (i = 0; i < maxentries; i++) {
        int nextp;
        int reft;
-       int hash;
+       int hash, nexthash = 0;
        int *nextpp;
        char *which;
 
@@ -1154,6 +1173,58 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
                foundbad = 1;
            }
 
+           if (record[ADDR(vlentry.nextNameHash)].type & MULTN) {
+               nextp = ADDR(vlentry.nextNameHash);
+               reft = REFN;
+               hash = NameHash(vlentry.name);
+               nextpp = &vlentry.nextNameHash;
+               which = "name";
+               sprintf(volidbuf, "");
+               readentry(nextp, &vlentry2, &type);
+               nexthash = NameHash(vlentry2.name);
+               if (hash != nexthash)
+                   foundbad = 1;
+           }
+
+           if ((record[ADDR(vlentry.nextIdHash[0])].type & MULTRW)) {
+               nextp = ADDR(vlentry.nextIdHash[0]);
+               reft = REFRW;
+               hash = IdHash(vlentry.volumeId[0]);
+               nextpp = &(vlentry.nextIdHash[0]);
+               which = "RW";
+               sprintf(volidbuf, "id %u ", vlentry.volumeId[0]);
+               readentry(nextp, &vlentry2, &type);
+               nexthash = IdHash(vlentry2.volumeId[0]);
+               if (hash != nexthash)
+                   foundbad = 1;
+           }
+
+           if ((record[ADDR(vlentry.nextIdHash[1])].type & MULTRO)) {
+               nextp = ADDR(vlentry.nextIdHash[1]);
+               reft = REFRO;
+               hash = IdHash(vlentry.volumeId[1]);
+               nextpp = &(vlentry.nextIdHash[1]);
+               which = "RO";
+               sprintf(volidbuf, "id %u ", vlentry.volumeId[1]);
+               readentry(nextp, &vlentry2, &type);
+               nexthash = IdHash(vlentry2.volumeId[1]);
+               if (hash != nexthash)
+                   foundbad = 1;
+           }
+
+           if ((record[ADDR(vlentry.nextIdHash[2])].type & MULTBK)) {
+               nextp = ADDR(vlentry.nextIdHash[2]);
+               reft = REFBK;
+               hash = IdHash(vlentry.volumeId[2]);
+               nextpp = &(vlentry.nextIdHash[2]);
+               which = "BK";
+               sprintf(volidbuf, "id %u ", vlentry.volumeId[2]);
+               readentry(nextp, &vlentry2, &type);
+               nexthash = IdHash(vlentry2.volumeId[2]);
+               if (hash != nexthash)
+                   foundbad = 1;
+           }
+
            if (foundbad) {
                printf("%d: Volume '%s' %snot found in %s hash %d", i, 
                       vlentry.name, volidbuf, which, hash);
@@ -1217,19 +1288,84 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
        }
     }
 
+    if (verbose)
+       printf("Verify each chain head\n");
+
+    {
+       afs_uint32 addr;
+       int hash;
+
+       for (j = 0; j < HASHSIZE; j++) {
+           for (addr = header.VolnameHash[j]; j < HASHSIZE; j++) {
+               if (record[ADDR(addr)].type & MULTN) {
+                   hash = NameHash(vlentry.name);
+                   if (hash != j) {
+                       header.VolnameHash[j] = vlentry.nextNameHash;
+                       vlentry.nextNameHash = 0;
+                       if (fix)
+                           writeentry(record[i].addr, &vlentry);
+                   }
+               }
+           }
+       }
+       for (i = 0; i <= 2; i++) {
+           for (j = 0; j < HASHSIZE; j++) {
+               addr = header.VolidHash[i][j]; 
+               if (i == 0 && (record[ADDR(addr)].type & MULTRW)) {
+                   hash = IdHash(vlentry.volumeId[i]);
+                   if (hash != j) {
+                       header.VolidHash[i][j] = vlentry.nextIdHash[i];
+                       vlentry.nextIdHash[i] = 0;
+                       if (fix)
+                           writeentry(record[i].addr, &vlentry);
+                   }
+               }
+
+               if (i == 1 && (record[ADDR(addr)].type & MULTRO)) {
+                   hash = IdHash(vlentry.volumeId[i]);
+                   if (hash != j) {
+                       header.VolidHash[i][j] = vlentry.nextIdHash[i];
+                       vlentry.nextIdHash[i] = 0;
+                       if (fix)
+                           writeentry(record[i].addr, &vlentry);
+                   }
+               }
+
+               if (i == 2 && (record[ADDR(addr)].type & MULTBK)) {
+                   hash = IdHash(vlentry.volumeId[i]);
+                   if (hash != j) {
+                       header.VolidHash[i][j] = vlentry.nextIdHash[i];
+                       vlentry.nextIdHash[i] = 0;
+                       if (fix)
+                           writeentry(record[i].addr, &vlentry);
+                   }
+               }
+           }
+       }
+    }
     /* By the time we get here, unchained entries are really unchained */
     printf("Scanning %u entries for possible repairs\n", maxentries);
     for (i = 0; i < maxentries; i++) {
+       int *nextpp;
        if (record[i].type & VL) {
            readentry(record[i].addr, &vlentry, &type);
-           if (!(record[i].type & REFN) && (strlen(vlentry.name)>0)) {
+           if (!(record[i].type & REFN)) {
                printf("%d: Record %u (type 0x%x) not in a name chain\n", i, 
                       record[i].addr, record[i].type);
-               if (fix) {
-                   if (header.VolnameHash[NameHash(vlentry.name)] == 0)
-                       header.VolnameHash[NameHash(vlentry.name)] = record[i].addr;
-                   else
-                       FixBad(i, header.VolnameHash[NameHash(vlentry.name)], NH, record[i].addr, &vlentry);
+               if (strlen(vlentry.name)>0) {
+                   if (fix) {
+                       if (header.VolnameHash[NameHash(vlentry.name)] == 0)
+                           header.VolnameHash[NameHash(vlentry.name)] = record[i].addr;
+                       else
+                           FixBad(i, header.VolnameHash[NameHash(vlentry.name)], NH, record[i].addr, &vlentry, NameHash(vlentry.name));
+                   }
+               } else {
+                   nextpp = &vlentry.nextNameHash;
+                   if (fix && *nextpp) {
+                       printf(", unchaining");
+                       *nextpp = 0;
+                       writeentry(record[i].addr, &vlentry);
+                   }
                }
            }
            if (vlentry.volumeId[0] && !(record[i].type & REFRW)) {
@@ -1239,7 +1375,7 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
                    if (header.VolidHash[0][IdHash(vlentry.volumeId[0])] == 0)
                        header.VolidHash[0][IdHash(vlentry.volumeId[0])] = record[i].addr;
                    else
-                       FixBad(i, header.VolidHash[0][IdHash(vlentry.volumeId[0])], RWH, record[i].addr, &vlentry);
+                       FixBad(i, header.VolidHash[0][IdHash(vlentry.volumeId[0])], RWH, record[i].addr, &vlentry, IdHash(vlentry.volumeId[0]));
                }
            }
            if (vlentry.volumeId[1] && !(record[i].type & REFRO)) {
@@ -1249,7 +1385,7 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
                    if (header.VolidHash[1][IdHash(vlentry.volumeId[1])] == 0)
                        header.VolidHash[1][IdHash(vlentry.volumeId[1])] = record[i].addr;
                    else
-                       FixBad(i, header.VolidHash[1][IdHash(vlentry.volumeId[1])], ROH, record[i].addr, &vlentry);
+                       FixBad(i, header.VolidHash[1][IdHash(vlentry.volumeId[1])], ROH, record[i].addr, &vlentry, IdHash(vlentry.volumeId[1]));
                }
            }
            if (vlentry.volumeId[2] && !(record[i].type & REFBK)) {
@@ -1259,7 +1395,7 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
                    if (header.VolidHash[2][IdHash(vlentry.volumeId[2])] == 0)
                        header.VolidHash[2][IdHash(vlentry.volumeId[2])] = record[i].addr;
                    else
-                       FixBad(i, header.VolidHash[2][IdHash(vlentry.volumeId[2])], BKH, record[i].addr, &vlentry);
+                       FixBad(i, header.VolidHash[2][IdHash(vlentry.volumeId[2])], BKH, record[i].addr, &vlentry, IdHash(vlentry.volumeId[2]));
                }
 
            }
@@ -1268,6 +1404,16 @@ WorkerBee(struct cmd_syndesc *as, void *arock)
     if (fix) 
        writeheader(&header);
 
+    close(fd);
+
+    if (fixed) {
+      fixed=0;
+      passes++;
+      if (passes < 20)
+       goto restart;
+      else
+       return 1;
+    }
     return 0;
 }