/*
* save away a lump, and return it's score.
* doesn't store duplicates, but checks that the data is really the same.
*/
int
writelump(Packet *p, u8int *score, int type, u32int creator, uint ms)
{
Lump *u;
int ok;
/*
qlock(&stats.lock);
stats.lumpwrites++;
qunlock(&stats.lock);
*/
packetsha1(p, score);
if(packetsize(p) == 0 || writestodevnull==1){
packetfree(p);
return 0;
}
u = lookuplump(score, type);
if(u->data != nil){
ok = 0;
if(packetcmp(p, u->data) != 0){
uchar nscore[VtScoreSize];
packetsha1(u->data, nscore);
if(scorecmp(u->score, score) != 0)
seterr(EStrange, "lookuplump returned bad score %V not %V", u->score, score);
else if(scorecmp(u->score, nscore) != 0)
seterr(EStrange, "lookuplump returned bad data %V not %V", nscore, u->score);
else
seterr(EStrange, "score collision %V", score);
ok = -1;
}
packetfree(p);
putlump(u);
return ok;
}
if(writestodevnull==2){
packetfree(p);
return 0;
}
if(queuewrites)
return queuewrite(u, p, creator, ms);
ok = writeqlump(u, p, creator, ms);
putlump(u);
return ok;
}
/*
* make an Arena, and initialize it based upon the disk header and trailer.
*/
Arena*
initarena(Part *part, uint64_t base, uint64_t size, uint32_t blocksize)
{
Arena *arena;
arena = MKZ(Arena);
arena->part = part;
arena->blocksize = blocksize;
arena->clumpmax = arena->blocksize / ClumpInfoSize;
arena->base = base + blocksize;
arena->size = size - 2 * blocksize;
if(loadarena(arena) < 0){
seterr(ECorrupt, "arena header or trailer corrupted");
freearena(arena);
return nil;
}
if(okarena(arena) < 0){
freearena(arena);
return nil;
}
if(arena->diskstats.sealed && scorecmp(zeroscore, arena->score)==0)
sealarena(arena);
return arena;
}
void
xmlscore(Hio *hout, u8int *v, char *tag)
{
if(scorecmp(zeroscore, v) == 0)
return;
hprint(hout, " %s=\"%V\"", tag, v);
}
static vlong
findintoc(HConnect *c, Arena *arena, uchar *score)
{
uchar *blk;
int i;
vlong off;
vlong coff;
ClumpInfo ci;
blk = vtmalloc(arena->blocksize);
off = arena->base + arena->size;
coff = 0;
for(i=0; i<arena->memstats.clumps; i++){
if(i%arena->clumpmax == 0){
off -= arena->blocksize;
if(readpart(arena->part, off, blk, arena->blocksize) != arena->blocksize){
if(c)
hprint(&c->hout, "<i>clump info directory at %#llx: %r</i>\n<br>\n",
off);
break;
}
}
unpackclumpinfo(&ci, blk+(i%arena->clumpmax)*ClumpInfoSize);
if(scorecmp(ci.score, score) == 0){
vtfree(blk);
return coff;
}
coff += ClumpSize + ci.size;
}
vtfree(blk);
return -1;
}
static void
darena(Hio *hout, Arena *arena)
{
hprint(hout, "arena='%s' on %s at [%lld,%lld)\n\tversion=%d created=%d modified=%d",
arena->name, arena->part->name, arena->base, arena->base + arena->size + 2 * arena->blocksize,
arena->version, arena->ctime, arena->wtime);
if(arena->memstats.sealed)
hprint(hout, " mem=sealed");
if(arena->diskstats.sealed)
hprint(hout, " disk=sealed");
if(arena->inqueue)
hprint(hout, " inqueue");
hprint(hout, "\n");
if(scorecmp(zeroscore, arena->score) != 0)
hprint(hout, "\tscore=%V\n", arena->score);
hprint(hout, "\twritten: clumps=%d compressed clumps=%d data=%,lld compressed data=%,lld storage=%,lld\n",
arena->memstats.clumps, arena->memstats.cclumps, arena->memstats.uncsize,
arena->memstats.used - arena->memstats.clumps * ClumpSize,
arena->memstats.used + arena->memstats.clumps * ClumpInfoSize);
hprint(hout, "\tindexed: clumps=%d compressed clumps=%d data=%,lld compressed data=%,lld storage=%,lld\n",
arena->diskstats.clumps, arena->diskstats.cclumps, arena->diskstats.uncsize,
arena->diskstats.used - arena->diskstats.clumps * ClumpSize,
arena->diskstats.used + arena->diskstats.clumps * ClumpInfoSize);
}
int
clumpinfoeq(ClumpInfo *c, ClumpInfo *d)
{
return c->type == d->type
&& c->size == d->size
&& c->uncsize == d->uncsize
&& scorecmp(c->score, d->score)==0;
}
/*
* Some of this logic is duplicated in hdisk.c
*/
Packet*
readlump(u8int *score, int type, u32int size, int *cached)
{
Lump *u;
Packet *p;
IAddr ia;
u32int n;
trace(TraceLump, "readlump enter");
/*
qlock(&stats.lock);
stats.lumpreads++;
qunlock(&stats.lock);
*/
if(scorecmp(score, zeroscore) == 0)
return packetalloc();
u = lookuplump(score, type);
if(u->data != nil){
trace(TraceLump, "readlump lookuplump hit");
if(cached)
*cached = 1;
n = packetsize(u->data);
if(n > size){
seterr(EOk, "read too small: asked for %d need at least %d", size, n);
putlump(u);
return nil;
}
p = packetdup(u->data, 0, n);
putlump(u);
return p;
}
if(cached)
*cached = 0;
if(lookupscore(score, type, &ia) < 0){
/* ZZZ place to check for someone trying to guess scores */
seterr(EOk, "no block with score %V/%d exists", score, type);
putlump(u);
return nil;
}
if(ia.size > size){
seterr(EOk, "read too small 1: asked for %d need at least %d", size, ia.size);
putlump(u);
return nil;
}
trace(TraceLump, "readlump readilump");
p = readilump(u, &ia, score);
putlump(u);
trace(TraceLump, "readlump exit");
return p;
}
static IEntry*
ihashlookup(IHash *ih, u8int score[VtScoreSize], int type)
{
u32int h;
IEntry *ie;
h = hashbits(score, ih->bits);
for(ie=ih->table[h]; ie; ie=ie->nexthash)
if((type == -1 || type == ie->ia.type) && scorecmp(score, ie->score) == 0)
return ie;
return nil;
}
static IEntry*
iesort(IEntry *ie)
{
int cmp;
IEntry **l;
IEntry *ie1, *ie2, *sorted;
if(ie == nil || ie->nextdirty == nil)
return ie;
/* split the lists */
ie1 = ie;
ie2 = ie;
if(ie2)
ie2 = ie2->nextdirty;
if(ie2)
ie2 = ie2->nextdirty;
while(ie1 && ie2){
ie1 = ie1->nextdirty;
ie2 = ie2->nextdirty;
if(ie2)
ie2 = ie2->nextdirty;
}
if(ie1){
ie2 = ie1->nextdirty;
ie1->nextdirty = nil;
}
/* sort the lists */
ie1 = iesort(ie);
ie2 = iesort(ie2);
/* merge the lists */
sorted = nil;
l = &sorted;
cmp = 0;
while(ie1 || ie2){
if(ie1 && ie2)
cmp = scorecmp(ie1->score, ie2->score);
if(ie1==nil || (ie2 && cmp > 0)){
*l = ie2;
l = &ie2->nextdirty;
ie2 = ie2->nextdirty;
}else{
*l = ie1;
l = &ie1->nextdirty;
ie1 = ie1->nextdirty;
}
}
*l = nil;
return sorted;
}
struct map *findscore(u8int *score)
{
int ix;
ix = hashbits(score, hashb);
//fprint(2, "find for %V is %d, maps[].data %p\n", score, ix, maps[ix].data);
while (maps[ix].data) {
//fprint(2, "Check: %d, %V\n", ix, maps[ix].score);
//fprint(2, "scorecmp(%V,%V, %d\n", maps[ix].score, score,scorecmp(maps[ix].score, score) );
if (scorecmp(maps[ix].score, score) == 0)
return &maps[ix];
ix++;
}
return nil;
}
static Packet*
readilump(Lump *u, IAddr *ia, u8int *score)
{
Arena *arena;
ZBlock *zb;
Packet *p, *pp;
Clump cl;
u64int aa;
u8int sc[VtScoreSize];
trace(TraceLump, "readilump enter");
arena = amapitoa(mainindex, ia->addr, &aa);
if(arena == nil){
trace(TraceLump, "readilump amapitoa failed");
return nil;
}
trace(TraceLump, "readilump loadclump");
zb = loadclump(arena, aa, ia->blocks, &cl, sc, paranoid);
if(zb == nil){
trace(TraceLump, "readilump loadclump failed");
return nil;
}
if(ia->size != cl.info.uncsize){
seterr(EInconsist, "index and clump size mismatch");
freezblock(zb);
return nil;
}
if(ia->type != cl.info.type){
seterr(EInconsist, "index and clump type mismatch");
freezblock(zb);
return nil;
}
if(scorecmp(score, sc) != 0){
seterr(ECrash, "score mismatch");
freezblock(zb);
return nil;
}
trace(TraceLump, "readilump success");
p = zblock2packet(zb, cl.info.uncsize);
freezblock(zb);
pp = packetdup(p, 0, packetsize(p));
trace(TraceLump, "readilump insertlump");
insertlump(u, pp);
trace(TraceLump, "readilump exit");
return p;
}
int
findscore(Arena *arena, uchar *score)
{
IEntry ie;
ClumpInfo *ci, *cis;
u64int a;
u32int clump;
int i, n, found;
//ZZZ remove fprint?
if(arena->memstats.clumps)
fprint(2, "reading directory for arena=%s with %d entries\n",
arena->name, arena->memstats.clumps);
cis = MKN(ClumpInfo, ClumpChunks);
found = 0;
a = 0;
memset(&ie, 0, sizeof(IEntry));
for(clump = 0; clump < arena->memstats.clumps; clump += n){
n = ClumpChunks;
if(n > arena->memstats.clumps - clump)
n = arena->memstats.clumps - clump;
if(readclumpinfos(arena, clump, cis, n) != n){
seterr(EOk, "arena directory read failed: %r");
break;
}
for(i = 0; i < n; i++){
ci = &cis[i];
if(scorecmp(score, ci->score)==0){
fprint(2, "found at clump=%d with type=%d size=%d csize=%d position=%lld\n",
clump + i, ci->type, ci->uncsize, ci->size, a);
found++;
}
a += ci->size + ClumpSize;
}
}
free(cis);
return found;
}
int
writeqlump(Lump *u, Packet *p, int creator, uint ms)
{
ZBlock *flat;
Packet *old;
IAddr ia;
int ok;
if(lookupscore(u->score, u->type, &ia) == 0){
if(verifywrites == 0){
/* assume the data is here! */
packetfree(p);
ms = msec() - ms;
addstat2(StatRpcWriteOld, 1, StatRpcWriteOldTime, ms);
return 0;
}
/*
* if the read fails,
* assume it was corrupted data and store the block again
*/
old = readilump(u, &ia, u->score);
if(old != nil){
ok = 0;
if(packetcmp(p, old) != 0){
uchar nscore[VtScoreSize];
packetsha1(old, nscore);
if(scorecmp(u->score, nscore) != 0)
seterr(EStrange, "readilump returned bad data %V not %V", nscore, u->score);
else
seterr(EStrange, "score collision %V", u->score);
ok = -1;
}
packetfree(p);
packetfree(old);
ms = msec() - ms;
addstat2(StatRpcWriteOld, 1, StatRpcWriteOldTime, ms);
return ok;
}
logerr(EAdmin, "writelump: read %V failed, rewriting: %r\n", u->score);
}
flat = packet2zblock(p, packetsize(p));
ok = storeclump(mainindex, flat, u->score, u->type, creator, &ia);
freezblock(flat);
if(ok == 0)
insertlump(u, p);
else
packetfree(p);
if(syncwrites){
flushdcache();
flushicache();
flushdcache();
}
ms = msec() - ms;
addstat2(StatRpcWriteNew, 1, StatRpcWriteNewTime, ms);
return ok;
}
static int
diskarenaclump(HConnect *c, Arena *arena, vlong off, char *scorestr)
{
uchar *blk, *blk2;
Clump cl;
char err[ERRMAX];
uchar xscore[VtScoreSize], score[VtScoreSize];
Unwhack uw;
int n;
if(scorestr){
if(vtparsescore(scorestr, nil, score) < 0){
hprint(&c->hout, "bad score %s: %r\n", scorestr);
return -1;
}
if(off < 0){
off = findintoc(c, arena, score);
if(off < 0){
hprint(&c->hout, "score %V not found in arena %s\n", score, arena->name);
return -1;
}
hprint(&c->hout, "score %V at %#llx\n", score, off);
}
}else
memset(score, 0, sizeof score);
if(off < 0){
hprint(&c->hout, "bad offset %#llx\n", off);
return -1;
}
off += arena->base;
blk = vtmalloc(ClumpSize + VtMaxLumpSize);
if(readpart(arena->part, off, blk, ClumpSize + VtMaxLumpSize) != ClumpSize + VtMaxLumpSize){
hprint(&c->hout, "reading at %#llx: %r\n", off);
vtfree(blk);
return -1;
}
if(unpackclump(&cl, blk, arena->clumpmagic) < 0){
hprint(&c->hout, "unpackclump: %r\n<br>");
rerrstr(err, sizeof err);
if(strstr(err, "magic")){
hprint(&c->hout, "trying again with magic=%#ux<br>\n", U32GET(blk));
if(unpackclump(&cl, blk, U32GET(blk)) < 0){
hprint(&c->hout, "unpackclump: %r\n<br>\n");
goto error;
}
}else
goto error;
}
hprint(&c->hout, "<pre>type=%d size=%d uncsize=%d score=%V\n", cl.info.type, cl.info.size, cl.info.uncsize, cl.info.score);
hprint(&c->hout, "encoding=%d creator=%d time=%d %s</pre>\n", cl.encoding, cl.creator, cl.time, fmttime(err, cl.time));
if(cl.info.type == VtCorruptType)
hprint(&c->hout, "clump is marked corrupt<br>\n");
if(cl.info.size >= VtMaxLumpSize){
hprint(&c->hout, "clump too big\n");
goto error;
}
switch(cl.encoding){
case ClumpECompress:
blk2 = vtmalloc(VtMaxLumpSize);
unwhackinit(&uw);
n = unwhack(&uw, blk2, cl.info.uncsize, blk+ClumpSize, cl.info.size);
if(n < 0){
hprint(&c->hout, "decompression failed\n");
vtfree(blk2);
goto error;
}
if(n != cl.info.uncsize){
hprint(&c->hout, "got wrong amount: %d wanted %d\n", n, cl.info.uncsize);
// hhex(blk2, n);
vtfree(blk2);
goto error;
}
scoremem(xscore, blk2, cl.info.uncsize);
vtfree(blk2);
break;
case ClumpENone:
scoremem(xscore, blk+ClumpSize, cl.info.size);
break;
}
hprint(&c->hout, "score=%V<br>\n", xscore);
if(scorestr && scorecmp(score, xscore) != 0)
hprint(&c->hout, "score does NOT match expected %V\n", score);
vtfree(blk);
return 0;
error:
// hhex(blk, ClumpSize + VtMaxLumpSize);
vtfree(blk);
return -1;
}
void
sumarena(Arena *arena)
{
ZBlock *b;
DigestState s;
uint64_t a, e;
uint32_t bs;
int t;
uint8_t score[VtScoreSize];
bs = MaxIoSize;
if(bs < arena->blocksize)
bs = arena->blocksize;
/*
* read & sum all blocks except the last one
*/
flushdcache();
memset(&s, 0, sizeof s);
b = alloczblock(bs, 0, arena->part->blocksize);
e = arena->base + arena->size;
for(a = arena->base - arena->blocksize; a + arena->blocksize <= e; a += bs){
disksched();
while((t=arenasumsleeptime) == SleepForever){
sleep(1000);
disksched();
}
sleep(t);
if(a + bs > e)
bs = arena->blocksize;
if(readpart(arena->part, a, b->data, bs) < 0)
goto ReadErr;
addstat(StatSumRead, 1);
addstat(StatSumReadBytes, bs);
sha1(b->data, bs, nil, &s);
}
/*
* the last one is special, since it may already have the checksum included
*/
bs = arena->blocksize;
if(readpart(arena->part, e, b->data, bs) < 0){
ReadErr:
logerr(EOk, "sumarena can't sum %s, read at %lld failed: %r", arena->name, a);
freezblock(b);
return;
}
addstat(StatSumRead, 1);
addstat(StatSumReadBytes, bs);
sha1(b->data, bs-VtScoreSize, nil, &s);
sha1(zeroscore, VtScoreSize, nil, &s);
sha1(nil, 0, score, &s);
/*
* check for no checksum or the same
*/
if(scorecmp(score, &b->data[bs - VtScoreSize]) != 0
&& scorecmp(zeroscore, &b->data[bs - VtScoreSize]) != 0)
logerr(EOk, "overwriting mismatched checksums for arena=%s, found=%V calculated=%V",
arena->name, &b->data[bs - VtScoreSize], score);
freezblock(b);
qlock(&arena->lock);
scorecp(arena->score, score);
wbarena(arena);
qunlock(&arena->lock);
}
static void
rdarena(Arena *arena, u64int offset)
{
int i;
u64int a, aa, e;
uchar score[VtScoreSize];
Clump cl;
ClumpInfo ci;
ZBlock *lump;
ZClump zcl;
fprint(2, "wrarena: copying %s to venti\n", arena->name);
printarena(2, arena);
a = arena->base;
e = arena->base + arena->size;
if(offset != ~(u64int)0) {
if(offset >= e - a)
sysfatal("bad offset %#llx >= %#llx", offset, e - a);
aa = offset;
} else
aa = 0;
i = 0;
for(a = 0; maxwrites != 0 && i < arena->memstats.clumps;
a += ClumpSize + ci.size){
if(readclumpinfo(arena, i++, &ci) < 0)
break;
if(a < aa || ci.type == VtCorruptType){
if(ci.type == VtCorruptType)
fprint(2, "%s: corrupt clump read at %#llx: +%d\n",
argv0, a, ClumpSize+ci.size);
continue;
}
lump = loadclump(arena, a, 0, &cl, score, 0);
if(lump == nil) {
fprint(2, "clump %#llx failed to read: %r\n", a);
continue;
}
if(!fast && cl.info.type != VtCorruptType) {
scoremem(score, lump->data, cl.info.uncsize);
if(scorecmp(cl.info.score, score) != 0) {
fprint(2, "clump %#llx has mismatched score\n",
a);
break;
}
if(vttypevalid(cl.info.type) < 0) {
fprint(2, "clump %#llx has bad type %d\n",
a, cl.info.type);
break;
}
}
if(z && cl.info.type != VtCorruptType){
zcl.cl = cl;
zcl.lump = lump;
zcl.aa = a;
send(c, &zcl);
}else
freezblock(lump);
if(maxwrites > 0)
--maxwrites;
}
if(a > aa)
aa = a;
if(haveaoffset)
print("end offset %#llx\n", aa);
}
static void
debugread(HConnect *c, u8int *score)
{
int type;
Lump *u;
IAddr ia;
IEntry ie;
int i;
Arena *arena;
u64int aa;
ZBlock *zb;
Clump cl;
vlong off;
u8int sc[VtScoreSize];
if(scorecmp(score, zeroscore) == 0){
hprint(&c->hout, "zero score\n");
return;
}
hprint(&c->hout, "<h2>index search %V</h2><pre>\n", score);
if(icachelookup(score, -1, &ia) < 0)
hprint(&c->hout, " icache: not found\n");
else
hprint(&c->hout, " icache: addr=%#llx size=%d type=%d blocks=%d\n",
ia.addr, ia.size, ia.type, ia.blocks);
if(loadientry(mainindex, score, -1, &ie) < 0)
hprint(&c->hout, " idisk: not found\n");
else
hprint(&c->hout, " idisk: addr=%#llx size=%d type=%d blocks=%d\n",
ie.ia.addr, ie.ia.size, ie.ia.type, ie.ia.blocks);
hprint(&c->hout, "</pre><h2>lookup %V</h2>\n", score);
hprint(&c->hout, "<pre>\n");
for(type=0; type < VtMaxType; type++){
hprint(&c->hout, "%V type %d:", score, type);
u = lookuplump(score, type);
if(u->data != nil)
hprint(&c->hout, " +cache");
else
hprint(&c->hout, " -cache");
putlump(u);
if(lookupscore(score, type, &ia) < 0){
hprint(&c->hout, " -lookup\n");
continue;
}
hprint(&c->hout, "\n lookupscore: addr=%#llx size=%d blocks=%d\n",
ia.addr, ia.size, ia.blocks);
arena = amapitoa(mainindex, ia.addr, &aa);
if(arena == nil){
hprint(&c->hout, " amapitoa failed: %r\n");
continue;
}
hprint(&c->hout, " amapitoa: aa=%#llx arena="
"<a href=\"/disk?disk=%s&type=a&arena=%s&score=%V\">%s</a>\n",
aa, arena->part->name, arena->name, score, arena->name);
zb = loadclump(arena, aa, ia.blocks, &cl, sc, 1);
if(zb == nil){
hprint(&c->hout, " loadclump failed: %r\n");
continue;
}
hprint(&c->hout, " loadclump: uncsize=%d type=%d score=%V\n",
cl.info.uncsize, cl.info.type, sc);
if(ia.size != cl.info.uncsize || ia.type != cl.info.type || scorecmp(score, sc) != 0){
hprint(&c->hout, " clump info mismatch\n");
continue;
}
}
if(hargstr(c, "brute", "")[0] == 'y'){
hprint(&c->hout, "</pre>\n");
hprint(&c->hout, "<h2>brute force arena search %V</h2>\n", score);
hprint(&c->hout, "<pre>\n");
for(i=0; i<mainindex->narenas; i++){
arena = mainindex->arenas[i];
hprint(&c->hout, "%s...\n", arena->name);
hflush(&c->hout);
off = findintoc(nil, arena, score);
if(off >= 0)
hprint(&c->hout, "%s %#llx (%#llx)\n", arena->name, off, mainindex->amap[i].start + off);
}
}
hprint(&c->hout, "</pre>\n");
}