fs.c

来自「这是一个同样来自贝尔实验室的和UNIX有着渊源的操作系统, 其简洁的设计和实现易」· C语言 代码 · 共 1,093 行 · 第 1/2 页

	 * about b.  Specifically, we bump the file system's epoch and
	 * then rewalk the path from the root down to and including b.
	 * This will copy-on-write as we walk, so now the state will be:
	 *
	 *	b	Snap w/ epoch e, holds ptrs to sb and mb.
	 *	new-b	Active w/ epoch e+1, holds ptrs to sb and mb.
	 *	sb	Active w/ epoch e.
	 *	mb	Active w/ epoch e.
	 *
	 * In this state, it's perfectly okay to make more pointers to sb and mb.
	 */
	if(!bumpEpoch(fs, 0) || !fileWalkSources(src))
		goto Err;

	/*
	 * Sync to disk.  I'm not sure this is necessary, but better safe than sorry.
	 */
	cacheFlush(fs->cache, 1);

	/*
	 * Create the directory where we will store the copy of src.
	 */
	dst = fileOpenSnapshot(fs, dstpath, doarchive);
	if(dst == nil)
		goto Err;

	/*
	 * Actually make the copy by setting dst's source and msource
	 * to be src's.
	 */
	if(!fileSnapshot(dst, src, fs->ehi-1, doarchive))
		goto Err;

	fileDecRef(src);
	fileDecRef(dst);
	src = nil;
	dst = nil;

	/*
	 * Make another copy of the file system.  This one is for the
	 * archiver, so that the file system we archive has the recently
	 * added snapshot both in /active and in /archive/yyyy/mmdd[.#].
	 */
	if(doarchive){
		if(!saveQid(fs))
			goto Err;
		if(!bumpEpoch(fs, 1))
			goto Err;
	}

	vtUnlock(fs->elk);

	/* BUG? can fs->arch fall out from under us here? */
	if(doarchive && fs->arch)
		archKick(fs->arch);

	return 1;

Err:
	fprint(2, "fsSnapshot: %R\n");
	if(src)
		fileDecRef(src);
	if(dst)
		fileDecRef(dst);
	vtUnlock(fs->elk);
	return 0;
}

int
fsVac(Fs *fs, char *name, uchar score[VtScoreSize])
{
	int r;
	DirEntry de;
	Entry e, ee;
	File *f;

	vtRLock(fs->elk);
	f = fileOpen(fs, name);
	if(f == nil){
		vtRUnlock(fs->elk);
		return 0;
	}

	if(!fileGetSources(f, &e, &ee) || !fileGetDir(f, &de)){
		fileDecRef(f);
		vtRUnlock(fs->elk);
		return 0;
	}
	fileDecRef(f);

	r = mkVac(fs->z, fs->blockSize, &e, &ee, &de, score);
	vtRUnlock(fs->elk);
	return r;
}

static int
vtWriteBlock(VtSession *z, uchar *buf, uint n, uint type, uchar score[VtScoreSize])
{
	if(!vtWrite(z, score, type, buf, n))
		return 0;
	if(!vtSha1Check(score, buf, n))
		return 0;
	return 1;
}

int
mkVac(VtSession *z, uint blockSize, Entry *pe, Entry *pee, DirEntry *pde, uchar score[VtScoreSize])
{
	uchar buf[8192];
	int i;
	uchar *p;
	uint n;
	DirEntry de;
	Entry e, ee, eee;
	MetaBlock mb;
	MetaEntry me;
	VtRoot root;

	e = *pe;
	ee = *pee;
	de = *pde;

	if(globalToLocal(e.score) != NilBlock
	|| (ee.flags&VtEntryActive && globalToLocal(ee.score) != NilBlock)){
		vtSetError("can only vac paths already stored on venti");
		return 0;
	}

	/*
	 * Build metadata source for root.
	 */
	n = deSize(&de);
	if(n+MetaHeaderSize+MetaIndexSize > sizeof buf){
		vtSetError("DirEntry too big");
		return 0;
	}
	memset(buf, 0, sizeof buf);
	mbInit(&mb, buf, n+MetaHeaderSize+MetaIndexSize, 1);
	p = mbAlloc(&mb, n);
	if(p == nil)
		abort();
	mbSearch(&mb, de.elem, &i, &me);
	assert(me.p == nil);
	me.p = p;
	me.size = n;
	dePack(&de, &me);
	mbInsert(&mb, i, &me);
	mbPack(&mb);

	eee.size = n+MetaHeaderSize+MetaIndexSize;
	if(!vtWriteBlock(z, buf, eee.size, VtDataType, eee.score))
		return 0;
	eee.psize = 8192;
	eee.dsize = 8192;
	eee.depth = 0;
	eee.flags = VtEntryActive;

	/*
	 * Build root source with three entries in it.
	 */
	entryPack(&e, buf, 0);
	entryPack(&ee, buf, 1);
	entryPack(&eee, buf, 2);

	n = VtEntrySize*3;
	memset(&root, 0, sizeof root);
	if(!vtWriteBlock(z, buf, n, VtDirType, root.score))
		return 0;

	/*
	 * Save root.
	 */
	root.version = VtRootVersion;
	strecpy(root.type, root.type+sizeof root.type, "vac");
	strecpy(root.name, root.name+sizeof root.name, de.elem);
	root.blockSize = blockSize;
	vtRootPack(&root, buf);
	if(!vtWriteBlock(z, buf, VtRootSize, VtRootType, score))
		return 0;

	return 1;
}

int
fsSync(Fs *fs)
{
	vtLock(fs->elk);
	fileMetaFlush(fs->file, 1);
	cacheFlush(fs->cache, 1);
	vtUnlock(fs->elk);
	return 1;
}

int
fsHalt(Fs *fs)
{
	vtLock(fs->elk);
	fs->halted = 1;
	fileMetaFlush(fs->file, 1);
	cacheFlush(fs->cache, 1);
	return 1;
}

int
fsUnhalt(Fs *fs)
{
	if(!fs->halted)
		return 0;
	fs->halted = 0;
	vtUnlock(fs->elk);
	return 1;
}

int
fsNextQid(Fs *fs, u64int *qid)
{
	Block *b;
	Super super;

	if((b = superGet(fs->cache, &super)) == nil)
		return 0;

	*qid = super.qid++;

	/*
	 * It's okay if the super block doesn't go to disk immediately,
	 * since fileMetaAlloc will record a dependency between the
	 * block holding this qid and the super block.  See file.c:/^fileMetaAlloc.
	 */
	superWrite(b, &super, 0);
	blockPut(b);
	return 1;
}

static void
fsMetaFlush(void *a)
{
	int rv;
	Fs *fs = a;

	vtRLock(fs->elk);
	rv = fileMetaFlush(fs->file, 1);
	vtRUnlock(fs->elk);
	if(rv > 0)
		cacheFlush(fs->cache, 0);
}

static int
fsEsearch1(File *f, char *path, u32int savetime, u32int *plo)
{
	int n, r;
	DirEntry de;
	DirEntryEnum *dee;
	File *ff;
	Entry e, ee;
	char *t;

	dee = deeOpen(f);
	if(dee == nil)
		return 0;

	n = 0;
	for(;;){
		r = deeRead(dee, &de);
		if(r <= 0)
			break;
		if(de.mode & ModeSnapshot){
			if((ff = fileWalk(f, de.elem)) != nil){
				if(fileGetSources(ff, &e, &ee))
					if(de.mtime >= savetime && e.snap != 0)
						if(e.snap < *plo)
							*plo = e.snap;
				fileDecRef(ff);
			}
		}
		else if(de.mode & ModeDir){
			if((ff = fileWalk(f, de.elem)) != nil){
				t = smprint("%s/%s", path, de.elem);
				n += fsEsearch1(ff, t, savetime, plo);
				vtMemFree(t);
				fileDecRef(ff);
			}
		}
		deCleanup(&de);
		if(r < 0)
			break;
	}
	deeClose(dee);

	return n;
}

static int
fsEsearch(Fs *fs, char *path, u32int savetime, u32int *plo)
{
	int n;
	File *f;
	DirEntry de;

	f = fileOpen(fs, path);
	if(f == nil)
		return 0;
	if(!fileGetDir(f, &de)){
		fileDecRef(f);
		return 0;
	}
	if((de.mode & ModeDir) == 0){
		fileDecRef(f);
		deCleanup(&de);
		return 0;
	}
	deCleanup(&de);
	n = fsEsearch1(f, path, savetime, plo);
	fileDecRef(f);
	return n;
}

void
fsSnapshotCleanup(Fs *fs, u32int age)
{
	u32int lo;

	/*
	 * Find the best low epoch we can use,
	 * given that we need to save all the unventied archives
	 * and all the snapshots younger than age.
	 */
	vtRLock(fs->elk);
	lo = fs->ehi;
	fsEsearch(fs, "/archive", 0, &lo);
	fsEsearch(fs, "/snapshot", time(0)-age*60, &lo);
	vtRUnlock(fs->elk);

	fsEpochLow(fs, lo);
	fsSnapshotRemove(fs);
}

/* remove all snapshots that have expired */
/* return number of directory entries remaining */
static int
fsRsearch1(File *f, char *s)
{
	int n, r;
	DirEntry de;
	DirEntryEnum *dee;
	File *ff;
	char *t;

	dee = deeOpen(f);
	if(dee == nil)
		return 0;

	n = 0;
	for(;;){
		r = deeRead(dee, &de);
		if(r <= 0)
			break;
		n++;
		if(de.mode & ModeSnapshot){
			if((ff = fileWalk(f, de.elem)) != nil)
				fileDecRef(ff);
			else if(strcmp(vtGetError(), ESnapOld) == 0){
				if(fileClri(f, de.elem, "adm"))
					n--;
			}
		}
		else if(de.mode & ModeDir){
			if((ff = fileWalk(f, de.elem)) != nil){
				t = smprint("%s/%s", s, de.elem);
				if(fsRsearch1(ff, t) == 0)
					if(fileRemove(ff, "adm"))
						n--;
				vtMemFree(t);
				fileDecRef(ff);
			}
		}
		deCleanup(&de);
		if(r < 0)
			break;
	}
	deeClose(dee);

	return n;
}

static int
fsRsearch(Fs *fs, char *path)
{
	File *f;
	DirEntry de;

	f = fileOpen(fs, path);
	if(f == nil)
		return 0;
	if(!fileGetDir(f, &de)){
		fileDecRef(f);
		return 0;
	}
	if((de.mode & ModeDir) == 0){
		fileDecRef(f);
		deCleanup(&de);
		return 0;
	}
	deCleanup(&de);
	fsRsearch1(f, path);
	fileDecRef(f);
	return 1;
}

void
fsSnapshotRemove(Fs *fs)
{
	vtRLock(fs->elk);
	fsRsearch(fs, "/snapshot");
	vtRUnlock(fs->elk);
}

struct Snap
{
	Fs *fs;
	Periodic *tick;
	VtLock *lk;
	uint snapMinutes;
	uint archMinute;
	uint snapLife;
	u32int lastSnap;
	u32int lastArch;
	u32int lastCleanup;
	uint ignore;
};

static void
snapEvent(void *v)
{
	Snap *s;
	u32int now, min;
	Tm tm;
	int need;

	s = v;

	now = time(0)/60;
	vtLock(s->lk);

	/*
	 * Snapshots happen every snapMinutes minutes.
	 * If we miss a snapshot (for example, because we
	 * were down), we wait for the next one.
	 */
	if(s->snapMinutes != ~0 && s->snapMinutes != 0
	&& now%s->snapMinutes==0 && now != s->lastSnap){
		if(!fsSnapshot(s->fs, nil, nil, 0))
			fprint(2, "fsSnapshot snap: %R\n");
		s->lastSnap = now;
	}

	/*
	 * Archival snapshots happen at archMinute.
	 * If we miss an archive (for example, because we
	 * were down), we do it as soon as possible.
	 */
	tm = *localtime(now*60);
	min = tm.hour*60+tm.min;
	if(s->archMinute != ~0){
		need = 0;
		if(min == s->archMinute && now != s->lastArch)
			need = 1;
		if(s->lastArch == 0){
			s->lastArch = 1;
			if(fsNeedArch(s->fs, s->archMinute))
				need = 1;
		}
		if(need){
			fsSnapshot(s->fs, nil, nil, 1);
			s->lastArch = now;
		}
	}

	/*
	 * Snapshot cleanup happens every snaplife or every day.
	 */
	if(s->snapLife != ~0
	&& (s->lastCleanup+s->snapLife < now || s->lastCleanup+24*60 < now)){
		fsSnapshotCleanup(s->fs, s->snapLife);
		s->lastCleanup = now;
	}
	vtUnlock(s->lk);
}

static Snap*
snapInit(Fs *fs)
{
	Snap *s;

	s = vtMemAllocZ(sizeof(Snap));
	s->fs = fs;
	s->tick = periodicAlloc(snapEvent, s, 10*1000);
	s->lk = vtLockAlloc();
	s->snapMinutes = -1;
	s->archMinute = -1;
	s->snapLife = -1;
	s->ignore = 5*2;	/* wait five minutes for clock to stabilize */
	return s;
}

void
snapGetTimes(Snap *s, u32int *arch, u32int *snap, u32int *snaplen)
{
	if(s == nil){
		*snap = -1;
		*arch = -1;
		*snaplen = -1;
		return;
	}

	vtLock(s->lk);
	*snap = s->snapMinutes;
	*arch = s->archMinute;
	*snaplen = s->snapLife;
	vtUnlock(s->lk);
}

void
snapSetTimes(Snap *s, u32int arch, u32int snap, u32int snaplen)
{
	if(s == nil)
		return;

	vtLock(s->lk);
	s->snapMinutes = snap;
	s->archMinute = arch;
	s->snapLife = snaplen;
	vtUnlock(s->lk);
}

static void
snapClose(Snap *s)
{
	if(s == nil)
		return;

	periodicKill(s->tick);
	vtMemFree(s);
}