/* Search a block for a specific dinode. */
union lfs_dinode *
lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp)
{
union lfs_dinode *ldip;
unsigned num, i;
ASSERT_NO_SEGLOCK(fs);
/*
* Read the inode block backwards, since later versions of the
* inode will supercede earlier ones. Though it is unlikely, it is
* possible that the same inode will appear in the same inode block.
*/
num = LFS_INOPB(fs);
for (i = num; i-- > 0; ) {
ldip = DINO_IN_BLOCK(fs, bp->b_data, i);
if (lfs_dino_getinumber(fs, ldip) == ino)
return (ldip);
}
printf("searched %u entries for %ju\n", num, (uintmax_t)ino);
printf("offset is 0x%jx (seg %d)\n", (uintmax_t)lfs_sb_getoffset(fs),
lfs_dtosn(fs, lfs_sb_getoffset(fs)));
printf("block is 0x%jx (seg %d)\n",
(uintmax_t)LFS_DBTOFSB(fs, bp->b_blkno),
lfs_dtosn(fs, LFS_DBTOFSB(fs, bp->b_blkno)));
return NULL;
}
/* Update segment and avail usage information when removing a block. */
static int
lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
size_t bsize, long *lastseg, size_t *num)
{
long seg;
int error = 0;
ASSERT_SEGLOCK(fs);
bsize = lfs_fragroundup(fs, bsize);
if (daddr > 0) {
if (*lastseg != (seg = lfs_dtosn(fs, daddr))) {
error = lfs_update_seguse(fs, ip, *lastseg, *num);
*num = bsize;
*lastseg = seg;
} else
*num += bsize;
}
return error;
}
/*
* Actually mark the segment clean.
* Must be called with the segment lock held.
*/
int
lfs_do_segclean(struct lfs *fs, unsigned long segnum)
{
extern int lfs_dostats;
struct buf *bp;
CLEANERINFO *cip;
SEGUSE *sup;
if (lfs_dtosn(fs, fs->lfs_curseg) == segnum) {
return (EBUSY);
}
LFS_SEGENTRY(sup, fs, segnum, bp);
if (sup->su_nbytes) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" %d live bytes\n", segnum, sup->su_nbytes));
brelse(bp, 0);
return (EBUSY);
}
if (sup->su_flags & SEGUSE_ACTIVE) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" segment is active\n", segnum));
brelse(bp, 0);
return (EBUSY);
}
if (!(sup->su_flags & SEGUSE_DIRTY)) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" segment is already clean\n", segnum));
brelse(bp, 0);
return (EALREADY);
}
fs->lfs_avail += lfs_segtod(fs, 1);
if (sup->su_flags & SEGUSE_SUPERBLOCK)
fs->lfs_avail -= lfs_btofsb(fs, LFS_SBPAD);
if (fs->lfs_version > 1 && segnum == 0 &&
fs->lfs_start < lfs_btofsb(fs, LFS_LABELPAD))
fs->lfs_avail -= lfs_btofsb(fs, LFS_LABELPAD) - fs->lfs_start;
mutex_enter(&lfs_lock);
fs->lfs_bfree += sup->su_nsums * lfs_btofsb(fs, fs->lfs_sumsize) +
lfs_btofsb(fs, sup->su_ninos * fs->lfs_ibsize);
fs->lfs_dmeta -= sup->su_nsums * lfs_btofsb(fs, fs->lfs_sumsize) +
lfs_btofsb(fs, sup->su_ninos * fs->lfs_ibsize);
if (fs->lfs_dmeta < 0)
fs->lfs_dmeta = 0;
mutex_exit(&lfs_lock);
sup->su_flags &= ~SEGUSE_DIRTY;
LFS_WRITESEGENTRY(sup, fs, segnum, bp);
LFS_CLEANERINFO(cip, fs, bp);
++cip->clean;
--cip->dirty;
fs->lfs_nclean = cip->clean;
cip->bfree = fs->lfs_bfree;
mutex_enter(&lfs_lock);
cip->avail = fs->lfs_avail - fs->lfs_ravail - fs->lfs_favail;
wakeup(&fs->lfs_avail);
mutex_exit(&lfs_lock);
(void) LFS_BWRITE_LOG(bp);
if (lfs_dostats)
++lfs_stats.segs_reclaimed;
return (0);
}
int
lfs_markv(struct proc *p, fsid_t *fsidp, BLOCK_INFO *blkiov,
int blkcnt)
{
BLOCK_INFO *blkp;
IFILE *ifp;
struct buf *bp;
struct inode *ip = NULL;
struct lfs *fs;
struct mount *mntp;
struct vnode *vp = NULL;
ino_t lastino;
daddr_t b_daddr, v_daddr;
int cnt, error;
int do_again = 0;
int numrefed = 0;
ino_t maxino;
size_t obsize;
/* number of blocks/inodes that we have already bwrite'ed */
int nblkwritten, ninowritten;
if ((mntp = vfs_getvfs(fsidp)) == NULL)
return (ENOENT);
fs = VFSTOULFS(mntp)->um_lfs;
if (fs->lfs_ronly)
return EROFS;
maxino = (lfs_fragstoblks(fs, VTOI(fs->lfs_ivnode)->i_ffs1_blocks) -
fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb;
cnt = blkcnt;
if ((error = vfs_busy(mntp, NULL)) != 0)
return (error);
/*
* This seglock is just to prevent the fact that we might have to sleep
* from allowing the possibility that our blocks might become
* invalid.
*
* It is also important to note here that unless we specify SEGM_CKP,
* any Ifile blocks that we might be asked to clean will never get
* to the disk.
*/
lfs_seglock(fs, SEGM_CLEAN | SEGM_CKP | SEGM_SYNC);
/* Mark blocks/inodes dirty. */
error = 0;
/* these were inside the initialization for the for loop */
v_daddr = LFS_UNUSED_DADDR;
lastino = LFS_UNUSED_INUM;
nblkwritten = ninowritten = 0;
for (blkp = blkiov; cnt--; ++blkp)
{
/* Bounds-check incoming data, avoid panic for failed VGET */
if (blkp->bi_inode <= 0 || blkp->bi_inode >= maxino) {
error = EINVAL;
goto err3;
}
/*
* Get the IFILE entry (only once) and see if the file still
* exists.
*/
if (lastino != blkp->bi_inode) {
/*
* Finish the old file, if there was one. The presence
* of a usable vnode in vp is signaled by a valid v_daddr.
*/
if (v_daddr != LFS_UNUSED_DADDR) {
lfs_vunref(vp);
numrefed--;
}
/*
* Start a new file
*/
lastino = blkp->bi_inode;
if (blkp->bi_inode == LFS_IFILE_INUM)
v_daddr = fs->lfs_idaddr;
else {
LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
/* XXX fix for force write */
v_daddr = ifp->if_daddr;
brelse(bp, 0);
}
if (v_daddr == LFS_UNUSED_DADDR)
continue;
/* Get the vnode/inode. */
error = lfs_fastvget(mntp, blkp->bi_inode, v_daddr,
&vp,
(blkp->bi_lbn == LFS_UNUSED_LBN
? blkp->bi_bp
: NULL));
if (!error) {
numrefed++;
}
if (error) {
DLOG((DLOG_CLEAN, "lfs_markv: lfs_fastvget"
" failed with %d (ino %d, segment %d)\n",
error, blkp->bi_inode,
lfs_dtosn(fs, blkp->bi_daddr)));
/*
* If we got EAGAIN, that means that the
* Inode was locked. This is
* recoverable: just clean the rest of
* this segment, and let the cleaner try
* again with another. (When the
* cleaner runs again, this segment will
* sort high on the list, since it is
* now almost entirely empty.) But, we
* still set v_daddr = LFS_UNUSED_ADDR
* so as not to test this over and over
* again.
*/
if (error == EAGAIN) {
error = 0;
do_again++;
}
#ifdef DIAGNOSTIC
else if (error != ENOENT)
panic("lfs_markv VFS_VGET FAILED");
#endif
/* lastino = LFS_UNUSED_INUM; */
v_daddr = LFS_UNUSED_DADDR;
vp = NULL;
ip = NULL;
continue;
}
ip = VTOI(vp);
ninowritten++;
} else if (v_daddr == LFS_UNUSED_DADDR) {
/*
* This can only happen if the vnode is dead (or
* in any case we can't get it...e.g., it is
* inlocked). Keep going.
*/
continue;
}
/* Past this point we are guaranteed that vp, ip are valid. */
/* Can't clean VU_DIROP directories in case of truncation */
/* XXX - maybe we should mark removed dirs specially? */
if (vp->v_type == VDIR && (vp->v_uflag & VU_DIROP)) {
do_again++;
continue;
}
/* If this BLOCK_INFO didn't contain a block, keep going. */
if (blkp->bi_lbn == LFS_UNUSED_LBN) {
/* XXX need to make sure that the inode gets written in this case */
/* XXX but only write the inode if it's the right one */
if (blkp->bi_inode != LFS_IFILE_INUM) {
LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
if (ifp->if_daddr == blkp->bi_daddr) {
mutex_enter(&lfs_lock);
LFS_SET_UINO(ip, IN_CLEANING);
mutex_exit(&lfs_lock);
}
brelse(bp, 0);
}
continue;
}
b_daddr = 0;
if (VOP_BMAP(vp, blkp->bi_lbn, NULL, &b_daddr, NULL) ||
LFS_DBTOFSB(fs, b_daddr) != blkp->bi_daddr)
{
if (lfs_dtosn(fs, LFS_DBTOFSB(fs, b_daddr)) ==
lfs_dtosn(fs, blkp->bi_daddr))
{
DLOG((DLOG_CLEAN, "lfs_markv: wrong da same seg: %llx vs %llx\n",
(long long)blkp->bi_daddr, (long long)LFS_DBTOFSB(fs, b_daddr)));
}
do_again++;
continue;
}
/*
* Check block sizes. The blocks being cleaned come from
* disk, so they should have the same size as their on-disk
* counterparts.
*/
if (blkp->bi_lbn >= 0)
obsize = lfs_blksize(fs, ip, blkp->bi_lbn);
else
obsize = fs->lfs_bsize;
/* Check for fragment size change */
if (blkp->bi_lbn >= 0 && blkp->bi_lbn < ULFS_NDADDR) {
obsize = ip->i_lfs_fragsize[blkp->bi_lbn];
}
if (obsize != blkp->bi_size) {
DLOG((DLOG_CLEAN, "lfs_markv: ino %d lbn %lld wrong"
" size (%ld != %d), try again\n",
blkp->bi_inode, (long long)blkp->bi_lbn,
(long) obsize, blkp->bi_size));
do_again++;
continue;
}
/*
* If we get to here, then we are keeping the block. If
* it is an indirect block, we want to actually put it
* in the buffer cache so that it can be updated in the
* finish_meta section. If it's not, we need to
* allocate a fake buffer so that writeseg can perform
* the copyin and write the buffer.
*/
if (ip->i_number != LFS_IFILE_INUM && blkp->bi_lbn >= 0) {
/* Data Block */
bp = lfs_fakebuf(fs, vp, blkp->bi_lbn,
blkp->bi_size, blkp->bi_bp);
/* Pretend we used bread() to get it */
bp->b_blkno = LFS_FSBTODB(fs, blkp->bi_daddr);
} else {
/* Indirect block or ifile */
if (blkp->bi_size != fs->lfs_bsize &&
ip->i_number != LFS_IFILE_INUM)
panic("lfs_markv: partial indirect block?"
" size=%d\n", blkp->bi_size);
bp = getblk(vp, blkp->bi_lbn, blkp->bi_size, 0, 0);
if (!(bp->b_oflags & (BO_DONE|BO_DELWRI))) {
/*
* The block in question was not found
* in the cache; i.e., the block that
* getblk() returned is empty. So, we
* can (and should) copy in the
* contents, because we've already
* determined that this was the right
* version of this block on disk.
*
* And, it can't have changed underneath
* us, because we have the segment lock.
*/
error = copyin(blkp->bi_bp, bp->b_data, blkp->bi_size);
if (error)
goto err2;
}
}
if ((error = lfs_bwrite_ext(bp, BW_CLEAN)) != 0)
goto err2;
nblkwritten++;
/*
* XXX should account indirect blocks and ifile pages as well
*/
if (nblkwritten + lfs_lblkno(fs, ninowritten * sizeof (struct ulfs1_dinode))
> LFS_MARKV_MAX_BLOCKS) {
DLOG((DLOG_CLEAN, "lfs_markv: writing %d blks %d inos\n",
nblkwritten, ninowritten));
lfs_segwrite(mntp, SEGM_CLEAN);
nblkwritten = ninowritten = 0;
}
}
/*
* Finish the old file, if there was one
*/
if (v_daddr != LFS_UNUSED_DADDR) {
lfs_vunref(vp);
numrefed--;
}
#ifdef DIAGNOSTIC
if (numrefed != 0)
panic("lfs_markv: numrefed=%d", numrefed);
#endif
DLOG((DLOG_CLEAN, "lfs_markv: writing %d blks %d inos (check point)\n",
nblkwritten, ninowritten));
/*
* The last write has to be SEGM_SYNC, because of calling semantics.
* It also has to be SEGM_CKP, because otherwise we could write
* over the newly cleaned data contained in a checkpoint, and then
* we'd be unhappy at recovery time.
*/
lfs_segwrite(mntp, SEGM_CLEAN | SEGM_CKP | SEGM_SYNC);
lfs_segunlock(fs);
vfs_unbusy(mntp, false, NULL);
if (error)
return (error);
else if (do_again)
return EAGAIN;
return 0;
err2:
DLOG((DLOG_CLEAN, "lfs_markv err2\n"));
/*
* XXX we're here because copyin() failed.
* XXX it means that we can't trust the cleanerd. too bad.
* XXX how can we recover from this?
*/
err3:
/*
* XXX should do segwrite here anyway?
*/
if (v_daddr != LFS_UNUSED_DADDR) {
lfs_vunref(vp);
--numrefed;
}
lfs_segunlock(fs);
vfs_unbusy(mntp, false, NULL);
#ifdef DIAGNOSTIC
if (numrefed != 0)
panic("lfs_markv: numrefed=%d", numrefed);
#endif
return (error);
}
int
setup(const char *dev)
{
long bmapsize;
struct stat statb;
int doskipclean;
u_int64_t maxfilesize;
int open_flags;
struct uvnode *ivp;
struct ubuf *bp;
int i, isdirty;
long sn, curseg;
SEGUSE *sup;
havesb = 0;
doskipclean = skipclean;
if (stat(dev, &statb) < 0) {
pfatal("Can't stat %s: %s\n", dev, strerror(errno));
return (0);
}
if (!S_ISCHR(statb.st_mode) && skipclean) {
pfatal("%s is not a character device", dev);
if (reply("CONTINUE") == 0)
return (0);
}
if (nflag)
open_flags = O_RDONLY;
else
open_flags = O_RDWR;
if ((fsreadfd = open(dev, open_flags)) < 0) {
pfatal("Can't open %s: %s\n", dev, strerror(errno));
return (0);
}
if (nflag) {
if (preen)
pfatal("NO WRITE ACCESS");
printf("** %s (NO WRITE)\n", dev);
quiet = 0;
} else if (!preen && !quiet)
printf("** %s\n", dev);
fsmodified = 0;
lfdir = 0;
/* Initialize time in case we have to write */
time(&write_time);
bufinit(0); /* XXX we could make a better guess */
fs = lfs_init(fsreadfd, bflag, idaddr, 0, debug);
if (fs == NULL) {
if (preen)
printf("%s: ", cdevname());
errexit("BAD SUPER BLOCK OR IFILE INODE NOT FOUND");
}
/* Resize buffer cache now that we have a superblock to guess from. */
bufrehash((fs->lfs_segtabsz + maxino / fs->lfs_ifpb) << 4);
if (fs->lfs_pflags & LFS_PF_CLEAN) {
if (doskipclean) {
if (!quiet)
pwarn("%sile system is clean; not checking\n",
preen ? "f" : "** F");
return (-1);
}
if (!preen)
pwarn("** File system is already clean\n");
}
if (idaddr) {
daddr_t tdaddr;
SEGSUM *sp;
FINFO *fp;
int bc;
if (debug)
pwarn("adjusting offset, serial for -i 0x%lx\n",
(unsigned long)idaddr);
tdaddr = lfs_sntod(fs, lfs_dtosn(fs, idaddr));
if (lfs_sntod(fs, lfs_dtosn(fs, tdaddr)) == tdaddr) {
if (tdaddr == fs->lfs_start)
tdaddr += lfs_btofsb(fs, LFS_LABELPAD);
for (i = 0; i < LFS_MAXNUMSB; i++) {
if (fs->lfs_sboffs[i] == tdaddr)
tdaddr += lfs_btofsb(fs, LFS_SBPAD);
if (fs->lfs_sboffs[i] > tdaddr)
break;
}
}
fs->lfs_offset = tdaddr;
if (debug)
pwarn("begin with offset/serial 0x%x/%d\n",
(int)fs->lfs_offset, (int)fs->lfs_serial);
while (tdaddr < idaddr) {
bread(fs->lfs_devvp, LFS_FSBTODB(fs, tdaddr),
fs->lfs_sumsize,
NULL, 0, &bp);
sp = (SEGSUM *)bp->b_data;
if (sp->ss_sumsum != cksum(&sp->ss_datasum,
fs->lfs_sumsize -
sizeof(sp->ss_sumsum))) {
brelse(bp, 0);
if (debug)
printf("bad cksum at %x\n",
(unsigned)tdaddr);
break;
}
fp = (FINFO *)(sp + 1);
bc = howmany(sp->ss_ninos, LFS_INOPB(fs)) <<
(fs->lfs_version > 1 ? fs->lfs_ffshift :
fs->lfs_bshift);
for (i = 0; i < sp->ss_nfinfo; i++) {
bc += fp->fi_lastlength + ((fp->fi_nblocks - 1)
<< fs->lfs_bshift);
fp = (FINFO *)(fp->fi_blocks + fp->fi_nblocks);
}
tdaddr += lfs_btofsb(fs, bc) + 1;
fs->lfs_offset = tdaddr;
fs->lfs_serial = sp->ss_serial + 1;
brelse(bp, 0);
}
/*
* Set curseg, nextseg appropriately -- inlined from
* lfs_newseg()
*/
curseg = lfs_dtosn(fs, fs->lfs_offset);
fs->lfs_curseg = lfs_sntod(fs, curseg);
for (sn = curseg + fs->lfs_interleave;;) {
sn = (sn + 1) % fs->lfs_nseg;
if (sn == curseg)
errx(1, "init: no clean segments");
LFS_SEGENTRY(sup, fs, sn, bp);
isdirty = sup->su_flags & SEGUSE_DIRTY;
brelse(bp, 0);
if (!isdirty)
break;
}
/* Skip superblock if necessary */
for (i = 0; i < LFS_MAXNUMSB; i++)
if (fs->lfs_offset == fs->lfs_sboffs[i])
fs->lfs_offset += lfs_btofsb(fs, LFS_SBPAD);
++fs->lfs_nactive;
fs->lfs_nextseg = lfs_sntod(fs, sn);
if (debug) {
pwarn("offset = 0x%" PRIx32 ", serial = %" PRId64 "\n",
fs->lfs_offset, fs->lfs_serial);
pwarn("curseg = %" PRIx32 ", nextseg = %" PRIx32 "\n",
fs->lfs_curseg, fs->lfs_nextseg);
}
if (!nflag && !skipclean) {
fs->lfs_idaddr = idaddr;
fsmodified = 1;
sbdirty();
}
}
if (debug) {
pwarn("idaddr = 0x%lx\n", idaddr ? (unsigned long)idaddr :
(unsigned long)fs->lfs_idaddr);
pwarn("dev_bsize = %lu\n", dev_bsize);
pwarn("lfs_bsize = %lu\n", (unsigned long) fs->lfs_bsize);
pwarn("lfs_fsize = %lu\n", (unsigned long) fs->lfs_fsize);
pwarn("lfs_frag = %lu\n", (unsigned long) fs->lfs_frag);
pwarn("lfs_inopb = %lu\n", (unsigned long) fs->lfs_inopb);
}
if (fs->lfs_version == 1)
maxfsblock = fs->lfs_size * (fs->lfs_bsize / dev_bsize);
else
maxfsblock = fs->lfs_size;
maxfilesize = calcmaxfilesize(fs->lfs_bshift);
if (/* fs->lfs_minfree < 0 || */ fs->lfs_minfree > 99) {
pfatal("IMPOSSIBLE MINFREE=%d IN SUPERBLOCK",
fs->lfs_minfree);
if (reply("SET TO DEFAULT") == 1) {
fs->lfs_minfree = 10;
sbdirty();
}
}
if (fs->lfs_bmask != fs->lfs_bsize - 1) {
pwarn("INCORRECT BMASK=0x%x IN SUPERBLOCK (SHOULD BE 0x%x)",
(unsigned int) fs->lfs_bmask,
(unsigned int) fs->lfs_bsize - 1);
fs->lfs_bmask = fs->lfs_bsize - 1;
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
sbdirty();
}
}
if (fs->lfs_ffmask != fs->lfs_fsize - 1) {
pwarn("INCORRECT FFMASK=%" PRId64 " IN SUPERBLOCK",
fs->lfs_ffmask);
fs->lfs_ffmask = fs->lfs_fsize - 1;
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
sbdirty();
}
}
if (fs->lfs_fbmask != (1 << fs->lfs_fbshift) - 1) {
pwarn("INCORRECT FBMASK=%" PRId64 " IN SUPERBLOCK",
fs->lfs_fbmask);
fs->lfs_fbmask = (1 << fs->lfs_fbshift) - 1;
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
sbdirty();
}
}
if (fs->lfs_maxfilesize != maxfilesize) {
pwarn(
"INCORRECT MAXFILESIZE=%llu IN SUPERBLOCK (SHOULD BE %llu WITH BSHIFT %d)",
(unsigned long long) fs->lfs_maxfilesize,
(unsigned long long) maxfilesize, (int)fs->lfs_bshift);
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
fs->lfs_maxfilesize = maxfilesize;
sbdirty();
}
}
if (fs->lfs_maxsymlinklen != ULFS1_MAXSYMLINKLEN) {
pwarn("INCORRECT MAXSYMLINKLEN=%d IN SUPERBLOCK",
fs->lfs_maxsymlinklen);
fs->lfs_maxsymlinklen = ULFS1_MAXSYMLINKLEN;
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
sbdirty();
}
}
/*
* Read in the Ifile; we'll be using it a lot.
* XXX If the Ifile is corrupted we are in bad shape. We need to
* XXX run through the segment headers of the entire disk to
* XXX reconstruct the inode table, then pretend all segments are
* XXX dirty while we do the rest.
*/
ivp = fs->lfs_ivnode;
maxino = ((VTOI(ivp)->i_ffs1_size - (fs->lfs_cleansz + fs->lfs_segtabsz)
* fs->lfs_bsize) / fs->lfs_bsize) * fs->lfs_ifpb;
if (debug)
pwarn("maxino = %llu\n", (unsigned long long)maxino);
for (i = 0; i < VTOI(ivp)->i_ffs1_size; i += fs->lfs_bsize) {
bread(ivp, i >> fs->lfs_bshift, fs->lfs_bsize, NOCRED, 0, &bp);
/* XXX check B_ERROR */
brelse(bp, 0);
}
/*
* allocate and initialize the necessary maps
*/
din_table = ecalloc(maxino, sizeof(*din_table));
seg_table = ecalloc(fs->lfs_nseg, sizeof(SEGUSE));
/* Get segment flags */
for (i = 0; i < fs->lfs_nseg; i++) {
LFS_SEGENTRY(sup, fs, i, bp);
seg_table[i].su_flags = sup->su_flags & ~SEGUSE_ACTIVE;
if (preen)
seg_table[i].su_nbytes = sup->su_nbytes;
brelse(bp, 0);
}
/* Initialize Ifile entry */
din_table[fs->lfs_ifile] = fs->lfs_idaddr;
seg_table[lfs_dtosn(fs, fs->lfs_idaddr)].su_nbytes += LFS_DINODE1_SIZE;
#ifndef VERBOSE_BLOCKMAP
bmapsize = roundup(howmany(maxfsblock, NBBY), sizeof(int16_t));
blockmap = ecalloc(bmapsize, sizeof(char));
#else
bmapsize = maxfsblock * sizeof(ino_t);
blockmap = ecalloc(maxfsblock, sizeof(ino_t));
#endif
statemap = ecalloc(maxino, sizeof(char));
typemap = ecalloc(maxino, sizeof(char));
lncntp = ecalloc(maxino, sizeof(int16_t));
if (preen) {
n_files = fs->lfs_nfiles;
n_blks = fs->lfs_dsize - fs->lfs_bfree;
numdirs = maxino;
inplast = 0;
listmax = numdirs + 10;
inpsort = ecalloc(listmax, sizeof(struct inoinfo *));
inphead = ecalloc(numdirs, sizeof(struct inoinfo *));
}
return (1);
ckfini(0);
return (0);
}
int
setup(const char *dev)
{
#ifndef VERBOSE_BLOCKMAP
long bmapsize;
#endif
struct stat statb;
int doskipclean;
u_int64_t maxfilesize;
int open_flags;
struct uvnode *ivp;
struct ubuf *bp;
int i, isdirty;
long sn, curseg;
SEGUSE *sup;
size_t sumstart;
havesb = 0;
doskipclean = skipclean;
if (stat(dev, &statb) < 0) {
pfatal("Can't stat %s: %s\n", dev, strerror(errno));
return (0);
}
if (!S_ISCHR(statb.st_mode) && skipclean) {
pfatal("%s is not a character device", dev);
if (reply("CONTINUE") == 0)
return (0);
}
if (nflag)
open_flags = O_RDONLY;
else
open_flags = O_RDWR;
if ((fsreadfd = open(dev, open_flags)) < 0) {
pfatal("Can't open %s: %s\n", dev, strerror(errno));
return (0);
}
if (nflag) {
if (preen)
pfatal("NO WRITE ACCESS");
printf("** %s (NO WRITE)\n", dev);
quiet = 0;
} else if (!preen && !quiet)
printf("** %s\n", dev);
fsmodified = 0;
lfdir = 0;
/* Initialize time in case we have to write */
time(&write_time);
bufinit(0); /* XXX we could make a better guess */
fs = lfs_init(fsreadfd, bflag, idaddr, 0, debug);
if (fs == NULL) {
if (preen)
printf("%s: ", cdevname());
errexit("BAD SUPER BLOCK OR IFILE INODE NOT FOUND");
}
/* Resize buffer cache now that we have a superblock to guess from. */
bufrehash((lfs_sb_getsegtabsz(fs) + maxino / lfs_sb_getifpb(fs)) << 4);
if (lfs_sb_getpflags(fs) & LFS_PF_CLEAN) {
if (doskipclean) {
if (!quiet)
pwarn("%sile system is clean; not checking\n",
preen ? "f" : "** F");
return (-1);
}
if (!preen)
pwarn("** File system is already clean\n");
}
if (idaddr) {
daddr_t tdaddr;
SEGSUM *sp;
FINFO *fp;
int bc;
if (debug)
pwarn("adjusting offset, serial for -i 0x%jx\n",
(uintmax_t)idaddr);
tdaddr = lfs_sntod(fs, lfs_dtosn(fs, idaddr));
if (lfs_sntod(fs, lfs_dtosn(fs, tdaddr)) == tdaddr) {
if (tdaddr == lfs_sb_gets0addr(fs))
tdaddr += lfs_btofsb(fs, LFS_LABELPAD);
for (i = 0; i < LFS_MAXNUMSB; i++) {
if (lfs_sb_getsboff(fs, i) == tdaddr)
tdaddr += lfs_btofsb(fs, LFS_SBPAD);
if (lfs_sb_getsboff(fs, i) > tdaddr)
break;
}
}
lfs_sb_setoffset(fs, tdaddr);
if (debug)
pwarn("begin with offset/serial 0x%jx/%jd\n",
(uintmax_t)lfs_sb_getoffset(fs),
(intmax_t)lfs_sb_getserial(fs));
while (tdaddr < idaddr) {
bread(fs->lfs_devvp, LFS_FSBTODB(fs, tdaddr),
lfs_sb_getsumsize(fs),
0, &bp);
sp = (SEGSUM *)bp->b_data;
sumstart = lfs_ss_getsumstart(fs);
if (lfs_ss_getsumsum(fs, sp) !=
cksum((char *)sp + sumstart,
lfs_sb_getsumsize(fs) - sumstart)) {
brelse(bp, 0);
if (debug)
printf("bad cksum at %jx\n",
(uintmax_t)tdaddr);
break;
}
fp = SEGSUM_FINFOBASE(fs, sp);
bc = howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)) <<
(lfs_sb_getversion(fs) > 1 ? lfs_sb_getffshift(fs) :
lfs_sb_getbshift(fs));
for (i = 0; i < lfs_ss_getnfinfo(fs, sp); i++) {
bc += lfs_fi_getlastlength(fs, fp) + ((lfs_fi_getnblocks(fs, fp) - 1)
<< lfs_sb_getbshift(fs));
fp = NEXT_FINFO(fs, fp);
}
tdaddr += lfs_btofsb(fs, bc) + 1;
lfs_sb_setoffset(fs, tdaddr);
lfs_sb_setserial(fs, lfs_ss_getserial(fs, sp) + 1);
brelse(bp, 0);
}
/*
* Set curseg, nextseg appropriately -- inlined from
* lfs_newseg()
*/
curseg = lfs_dtosn(fs, lfs_sb_getoffset(fs));
lfs_sb_setcurseg(fs, lfs_sntod(fs, curseg));
for (sn = curseg + lfs_sb_getinterleave(fs);;) {
sn = (sn + 1) % lfs_sb_getnseg(fs);
if (sn == curseg)
errx(1, "init: no clean segments");
LFS_SEGENTRY(sup, fs, sn, bp);
isdirty = sup->su_flags & SEGUSE_DIRTY;
brelse(bp, 0);
if (!isdirty)
break;
}
/* Skip superblock if necessary */
for (i = 0; i < LFS_MAXNUMSB; i++)
if (lfs_sb_getoffset(fs) == lfs_sb_getsboff(fs, i))
lfs_sb_addoffset(fs, lfs_btofsb(fs, LFS_SBPAD));
++fs->lfs_nactive;
lfs_sb_setnextseg(fs, lfs_sntod(fs, sn));
if (debug) {
pwarn("offset = 0x%" PRIx64 ", serial = %" PRIu64 "\n",
lfs_sb_getoffset(fs), lfs_sb_getserial(fs));
pwarn("curseg = %" PRIx64 ", nextseg = %" PRIx64 "\n",
lfs_sb_getcurseg(fs), lfs_sb_getnextseg(fs));
}
if (!nflag && !skipclean) {
lfs_sb_setidaddr(fs, idaddr);
fsmodified = 1;
sbdirty();
}
}
if (debug) {
pwarn("idaddr = 0x%jx\n", idaddr ? (uintmax_t)idaddr :
(uintmax_t)lfs_sb_getidaddr(fs));
pwarn("dev_bsize = %lu\n", dev_bsize);
pwarn("lfs_bsize = %lu\n", (unsigned long) lfs_sb_getbsize(fs));
pwarn("lfs_fsize = %lu\n", (unsigned long) lfs_sb_getfsize(fs));
pwarn("lfs_frag = %lu\n", (unsigned long) lfs_sb_getfrag(fs));
pwarn("lfs_inopb = %lu\n", (unsigned long) lfs_sb_getinopb(fs));
}
if (lfs_sb_getversion(fs) == 1)
maxfsblock = lfs_blkstofrags(fs, lfs_sb_getsize(fs));
else
maxfsblock = lfs_sb_getsize(fs);
maxfilesize = calcmaxfilesize(lfs_sb_getbshift(fs));
if (/* lfs_sb_getminfree(fs) < 0 || */ lfs_sb_getminfree(fs) > 99) {
pfatal("IMPOSSIBLE MINFREE=%u IN SUPERBLOCK",
lfs_sb_getminfree(fs));
if (reply("SET TO DEFAULT") == 1) {
lfs_sb_setminfree(fs, 10);
sbdirty();
}
}
if (lfs_sb_getbmask(fs) != lfs_sb_getbsize(fs) - 1) {
pwarn("INCORRECT BMASK=0x%jx IN SUPERBLOCK (SHOULD BE 0x%x)",
(uintmax_t)lfs_sb_getbmask(fs),
lfs_sb_getbsize(fs) - 1);
lfs_sb_setbmask(fs, lfs_sb_getbsize(fs) - 1);
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
sbdirty();
}
}
if (lfs_sb_getffmask(fs) != lfs_sb_getfsize(fs) - 1) {
pwarn("INCORRECT FFMASK=0x%jx IN SUPERBLOCK (SHOULD BE 0x%x)",
(uintmax_t)lfs_sb_getffmask(fs),
lfs_sb_getfsize(fs) - 1);
lfs_sb_setffmask(fs, lfs_sb_getfsize(fs) - 1);
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
sbdirty();
}
}
if (lfs_sb_getfbmask(fs) != (1U << lfs_sb_getfbshift(fs)) - 1) {
pwarn("INCORRECT FBMASK=0x%jx IN SUPERBLOCK (SHOULD BE 0x%x)",
(uintmax_t)lfs_sb_getfbmask(fs),
(1U << lfs_sb_getfbshift(fs)) - 1);
lfs_sb_setfbmask(fs, (1U << lfs_sb_getfbshift(fs)) - 1);
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
sbdirty();
}
}
if (lfs_sb_getmaxfilesize(fs) != maxfilesize) {
pwarn(
"INCORRECT MAXFILESIZE=%ju IN SUPERBLOCK (SHOULD BE %ju WITH BSHIFT %u)",
(uintmax_t) lfs_sb_getmaxfilesize(fs),
(uintmax_t) maxfilesize, lfs_sb_getbshift(fs));
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
lfs_sb_setmaxfilesize(fs, maxfilesize);
sbdirty();
}
}
if (lfs_sb_getmaxsymlinklen(fs) != LFS_MAXSYMLINKLEN(fs)) {
pwarn("INCORRECT MAXSYMLINKLEN=%d IN SUPERBLOCK (SHOULD BE %zu)",
lfs_sb_getmaxsymlinklen(fs), LFS_MAXSYMLINKLEN(fs));
lfs_sb_setmaxsymlinklen(fs, LFS_MAXSYMLINKLEN(fs));
if (preen)
printf(" (FIXED)\n");
if (preen || reply("FIX") == 1) {
sbdirty();
}
}
/*
* Read in the Ifile; we'll be using it a lot.
* XXX If the Ifile is corrupted we are in bad shape. We need to
* XXX run through the segment headers of the entire disk to
* XXX reconstruct the inode table, then pretend all segments are
* XXX dirty while we do the rest.
*/
ivp = fs->lfs_ivnode;
maxino = ((lfs_dino_getsize(fs, VTOI(ivp)->i_din) - (lfs_sb_getcleansz(fs) + lfs_sb_getsegtabsz(fs))
* lfs_sb_getbsize(fs)) / lfs_sb_getbsize(fs)) * lfs_sb_getifpb(fs);
if (debug)
pwarn("maxino = %llu\n", (unsigned long long)maxino);
for (i = 0; i < lfs_dino_getsize(fs, VTOI(ivp)->i_din); i += lfs_sb_getbsize(fs)) {
bread(ivp, i >> lfs_sb_getbshift(fs), lfs_sb_getbsize(fs), 0, &bp);
/* XXX check B_ERROR */
brelse(bp, 0);
}
/*
* allocate and initialize the necessary maps
*/
din_table = ecalloc(maxino, sizeof(*din_table));
seg_table = ecalloc(lfs_sb_getnseg(fs), sizeof(SEGUSE));
/* Get segment flags */
for (i = 0; i < lfs_sb_getnseg(fs); i++) {
LFS_SEGENTRY(sup, fs, i, bp);
seg_table[i].su_flags = sup->su_flags & ~SEGUSE_ACTIVE;
if (preen)
seg_table[i].su_nbytes = sup->su_nbytes;
brelse(bp, 0);
}
/* Initialize Ifile entry */
din_table[LFS_IFILE_INUM] = lfs_sb_getidaddr(fs);
seg_table[lfs_dtosn(fs, lfs_sb_getidaddr(fs))].su_nbytes += DINOSIZE(fs);
#ifndef VERBOSE_BLOCKMAP
bmapsize = roundup(howmany(maxfsblock, NBBY), sizeof(int16_t));
blockmap = ecalloc(bmapsize, sizeof(char));
#else
blockmap = ecalloc(maxfsblock, sizeof(ino_t));
#endif
statemap = ecalloc(maxino, sizeof(char));
typemap = ecalloc(maxino, sizeof(char));
lncntp = ecalloc(maxino, sizeof(int16_t));
if (preen) {
n_files = lfs_sb_getnfiles(fs);
n_blks = lfs_sb_getdsize(fs) - lfs_sb_getbfree(fs);
numdirs = maxino;
inplast = 0;
listmax = numdirs + 10;
inpsort = ecalloc(listmax, sizeof(struct inoinfo *));
inphead = ecalloc(numdirs, sizeof(struct inoinfo *));
}
return (1);
ckfini(0);
return (0);
}
/*
* Actually mark the segment clean.
* Must be called with the segment lock held.
*/
int
lfs_do_segclean(struct lfs *fs, unsigned long segnum)
{
extern int lfs_dostats;
struct buf *bp;
CLEANERINFO *cip;
SEGUSE *sup;
if (lfs_dtosn(fs, lfs_sb_getcurseg(fs)) == segnum) {
return (EBUSY);
}
LFS_SEGENTRY(sup, fs, segnum, bp);
if (sup->su_nbytes) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" %d live bytes\n", segnum, sup->su_nbytes));
brelse(bp, 0);
return (EBUSY);
}
if (sup->su_flags & SEGUSE_ACTIVE) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" segment is active\n", segnum));
brelse(bp, 0);
return (EBUSY);
}
if (!(sup->su_flags & SEGUSE_DIRTY)) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" segment is already clean\n", segnum));
brelse(bp, 0);
return (EALREADY);
}
lfs_sb_addavail(fs, lfs_segtod(fs, 1));
if (sup->su_flags & SEGUSE_SUPERBLOCK)
lfs_sb_subavail(fs, lfs_btofsb(fs, LFS_SBPAD));
if (lfs_sb_getversion(fs) > 1 && segnum == 0 &&
lfs_sb_gets0addr(fs) < lfs_btofsb(fs, LFS_LABELPAD))
lfs_sb_subavail(fs, lfs_btofsb(fs, LFS_LABELPAD) - lfs_sb_gets0addr(fs));
mutex_enter(&lfs_lock);
lfs_sb_addbfree(fs, sup->su_nsums * lfs_btofsb(fs, lfs_sb_getsumsize(fs)) +
lfs_btofsb(fs, sup->su_ninos * lfs_sb_getibsize(fs)));
lfs_sb_subdmeta(fs, sup->su_nsums * lfs_btofsb(fs, lfs_sb_getsumsize(fs)) +
lfs_btofsb(fs, sup->su_ninos * lfs_sb_getibsize(fs)));
if (lfs_sb_getdmeta(fs) < 0)
lfs_sb_setdmeta(fs, 0);
mutex_exit(&lfs_lock);
sup->su_flags &= ~SEGUSE_DIRTY;
LFS_WRITESEGENTRY(sup, fs, segnum, bp);
LFS_CLEANERINFO(cip, fs, bp);
lfs_ci_shiftdirtytoclean(fs, cip, 1);
lfs_sb_setnclean(fs, lfs_ci_getclean(fs, cip));
mutex_enter(&lfs_lock);
lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs));
lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs)
- fs->lfs_ravail - fs->lfs_favail);
wakeup(&fs->lfs_availsleep);
mutex_exit(&lfs_lock);
(void) LFS_BWRITE_LOG(bp);
if (lfs_dostats)
++lfs_stats.segs_reclaimed;
return (0);
}
static int
update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred,
struct lwp *l)
{
struct vnode *devvp, *vp;
struct inode *ip;
struct ulfs1_dinode *dip;
struct buf *dbp, *ibp;
int error;
daddr_t daddr;
IFILE *ifp;
SEGUSE *sup;
devvp = VTOI(fs->lfs_ivnode)->i_devvp;
/*
* Get the inode, update times and perms.
* DO NOT update disk blocks, we do that separately.
*/
error = bread(devvp, LFS_FSBTODB(fs, offset), fs->lfs_ibsize,
cred, 0, &dbp);
if (error) {
DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error));
return error;
}
dip = ((struct ulfs1_dinode *)(dbp->b_data)) + LFS_INOPB(fs);
while (--dip >= (struct ulfs1_dinode *)dbp->b_data) {
if (dip->di_inumber > LFS_IFILE_INUM) {
error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
l, &vp);
if (error) {
DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc"
" returned %d\n", error));
continue;
}
ip = VTOI(vp);
if (dip->di_size != ip->i_size)
lfs_truncate(vp, dip->di_size, 0, NOCRED);
/* Get mode, link count, size, and times */
memcpy(ip->i_din.ffs1_din, dip,
offsetof(struct ulfs1_dinode, di_db[0]));
/* Then the rest, except di_blocks */
ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
ip->i_mode = ip->i_ffs1_mode;
ip->i_nlink = ip->i_ffs1_nlink;
ip->i_size = ip->i_ffs1_size;
LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
/* Re-initialize to get type right */
ulfs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
&vp);
vput(vp);
/* Record change in location */
LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
daddr = ifp->if_daddr;
ifp->if_daddr = LFS_DBTOFSB(fs, dbp->b_blkno);
error = LFS_BWRITE_LOG(ibp); /* Ifile */
/* And do segment accounting */
if (lfs_dtosn(fs, daddr) != lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno))) {
if (daddr > 0) {
LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, daddr),
ibp);
sup->su_nbytes -= sizeof (struct ulfs1_dinode);
LFS_WRITESEGENTRY(sup, fs,
lfs_dtosn(fs, daddr),
ibp);
}
LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno)),
ibp);
sup->su_nbytes += sizeof (struct ulfs1_dinode);
LFS_WRITESEGENTRY(sup, fs,
lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno)),
ibp);
}
}
}
/*
* Load the appropriate indirect block, and change the appropriate pointer.
* Mark the block dirty. Do segment and avail accounting.
*/
static int
update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn,
daddr_t ndaddr, size_t size, struct lwp *l)
{
int error;
struct vnode *vp;
struct inode *ip;
#ifdef DEBUG
daddr_t odaddr;
struct indir a[ULFS_NIADDR];
int num;
int i;
#endif /* DEBUG */
struct buf *bp;
SEGUSE *sup;
KASSERT(lbn >= 0); /* no indirect blocks */
if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) {
DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc"
" returned %d\n", ino, error));
return error;
}
if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size,
NOCRED, 0, &bp)) != 0) {
vput(vp);
return (error);
}
/* No need to write, the block is already on disk */
if (bp->b_oflags & BO_DELWRI) {
LFS_UNLOCK_BUF(bp);
fs->lfs_avail += lfs_btofsb(fs, bp->b_bcount);
}
brelse(bp, BC_INVAL);
/*
* Extend the file, if it is not large enough already.
* XXX this is not exactly right, we don't know how much of the
* XXX last block is actually used. We hope that an inode will
* XXX appear later to give the correct size.
*/
ip = VTOI(vp);
if (ip->i_size <= (lbn << fs->lfs_bshift)) {
u_int64_t newsize;
if (lbn < ULFS_NDADDR)
newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) +
(size - fs->lfs_fsize) + 1;
else
newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1;
if (ip->i_size < newsize) {
ip->i_size = newsize;
/*
* tell vm our new size for the case the inode won't
* appear later.
*/
uvm_vnp_setsize(vp, newsize);
}
}
lfs_update_single(fs, NULL, vp, lbn, ndaddr, size);
LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, ndaddr), bp);
sup->su_nbytes += size;
LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, ndaddr), bp);
/* differences here should be due to UNWRITTEN indirect blocks. */
KASSERT((lfs_lblkno(fs, ip->i_size) > ULFS_NDADDR &&
ip->i_lfs_effnblks == ip->i_ffs1_blocks) ||
ip->i_lfs_effnblks >= ip->i_ffs1_blocks);
#ifdef DEBUG
/* Now look again to make sure it worked */
ulfs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
for (i = num; i > 0; i--) {
if (!a[i].in_exists)
panic("update_meta: absent %d lv indirect block", i);
}
if (LFS_DBTOFSB(fs, odaddr) != ndaddr)
DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %"
PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr));
#endif /* DEBUG */
vput(vp);
return 0;
}
