/*
* Attempt to expand the size of a directory
*/
static int
expanddir(struct uvnode *vp, union lfs_dinode *dp, char *name)
{
daddr_t lastbn;
struct ubuf *bp;
char *cp, firstblk[LFS_DIRBLKSIZ];
lastbn = lfs_lblkno(fs, lfs_dino_getsize(fs, dp));
if (lastbn >= ULFS_NDADDR - 1 || lfs_dino_getdb(fs, dp, lastbn) == 0 ||
lfs_dino_getsize(fs, dp) == 0)
return (0);
lfs_dino_setdb(fs, dp, lastbn + 1, lfs_dino_getdb(fs, dp, lastbn));
lfs_dino_setdb(fs, dp, lastbn, 0);
bp = getblk(vp, lastbn, lfs_sb_getbsize(fs));
VOP_BWRITE(bp);
lfs_dino_setsize(fs, dp,
lfs_dino_getsize(fs, dp) + lfs_sb_getbsize(fs));
lfs_dino_setblocks(fs, dp,
lfs_dino_getblocks(fs, dp) + lfs_btofsb(fs, lfs_sb_getbsize(fs)));
bread(vp, lfs_dino_getdb(fs, dp, lastbn + 1),
(long) lfs_dblksize(fs, dp, lastbn + 1), 0, &bp);
if (bp->b_flags & B_ERROR)
goto bad;
memcpy(firstblk, bp->b_data, LFS_DIRBLKSIZ);
bread(vp, lastbn, lfs_sb_getbsize(fs), 0, &bp);
if (bp->b_flags & B_ERROR)
goto bad;
memcpy(bp->b_data, firstblk, LFS_DIRBLKSIZ);
for (cp = &bp->b_data[LFS_DIRBLKSIZ];
cp < &bp->b_data[lfs_sb_getbsize(fs)];
cp += LFS_DIRBLKSIZ)
zerodirblk(cp);
VOP_BWRITE(bp);
bread(vp, lfs_dino_getdb(fs, dp, lastbn + 1),
(long) lfs_dblksize(fs, dp, lastbn + 1), 0, &bp);
if (bp->b_flags & B_ERROR)
goto bad;
zerodirblk(bp->b_data);
pwarn("NO SPACE LEFT IN %s", name);
if (preen)
printf(" (EXPANDED)\n");
else if (reply("EXPAND") == 0)
goto bad;
VOP_BWRITE(bp);
inodirty(VTOI(vp));
return (1);
bad:
lfs_dino_setdb(fs, dp, lastbn, lfs_dino_getdb(fs, dp, lastbn + 1));
lfs_dino_setdb(fs, dp, lastbn + 1, 0);
lfs_dino_setsize(fs, dp,
lfs_dino_getsize(fs, dp) - lfs_sb_getbsize(fs));
lfs_dino_setblocks(fs, dp,
lfs_dino_getblocks(fs, dp) - lfs_btofsb(fs, lfs_sb_getbsize(fs)));
return (0);
}
static void
lfs_do_deregister(struct lfs *fs, struct inode *ip, struct lbnentry *lbp)
{
ASSERT_MAYBE_SEGLOCK(fs);
mutex_enter(&lfs_lock);
--ip->i_lfs_nbtree;
SPLAY_REMOVE(lfs_splay, &ip->i_lfs_lbtree, lbp);
if (fs->lfs_favail > lfs_btofsb(fs, (1 << lfs_sb_getbshift(fs))))
fs->lfs_favail -= lfs_btofsb(fs, (1 << lfs_sb_getbshift(fs)));
fs->lfs_pages -= lfs_sb_getbsize(fs) >> PAGE_SHIFT;
if (locked_fakequeue_count > 0)
--locked_fakequeue_count;
lfs_subsys_pages -= lfs_sb_getbsize(fs) >> PAGE_SHIFT;
mutex_exit(&lfs_lock);
pool_put(&lfs_lbnentry_pool, lbp);
}
/*
* Record this lbn as being "write pending". We used to have this information
* on the buffer headers, but since pages don't have buffer headers we
* record it here instead.
*/
void
lfs_register_block(struct vnode *vp, daddr_t lbn)
{
struct lfs *fs;
struct inode *ip;
struct lbnentry *lbp;
ip = VTOI(vp);
/* Don't count metadata */
if (lbn < 0 || vp->v_type != VREG || ip->i_number == LFS_IFILE_INUM)
return;
fs = ip->i_lfs;
ASSERT_NO_SEGLOCK(fs);
/* If no space, wait for the cleaner */
lfs_availwait(fs, lfs_btofsb(fs, 1 << lfs_sb_getbshift(fs)));
lbp = (struct lbnentry *)pool_get(&lfs_lbnentry_pool, PR_WAITOK);
lbp->lbn = lbn;
mutex_enter(&lfs_lock);
if (SPLAY_INSERT(lfs_splay, &ip->i_lfs_lbtree, lbp) != NULL) {
mutex_exit(&lfs_lock);
/* Already there */
pool_put(&lfs_lbnentry_pool, lbp);
return;
}
++ip->i_lfs_nbtree;
fs->lfs_favail += lfs_btofsb(fs, (1 << lfs_sb_getbshift(fs)));
fs->lfs_pages += lfs_sb_getbsize(fs) >> PAGE_SHIFT;
++locked_fakequeue_count;
lfs_subsys_pages += lfs_sb_getbsize(fs) >> PAGE_SHIFT;
mutex_exit(&lfs_lock);
}
/*
* 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
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);
}
/*
* 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;
}
/*
* Release blocks associated with the inode ip and stored in the indirect
* block bn. Blocks are free'd in LIFO order up to (but not including)
* lastbn. If level is greater than SINGLE, the block is an indirect block
* and recursive calls to indirtrunc must be used to cleanse other indirect
* blocks.
*
* NB: triple indirect blocks are untested.
*/
static int
lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
daddr_t lastbn, int level, daddr_t *countp,
daddr_t *rcountp, long *lastsegp, size_t *bcp)
{
int i;
struct buf *bp;
struct lfs *fs = ip->i_lfs;
int32_t *bap; /* XXX ondisk32 */
struct vnode *vp;
daddr_t nb, nlbn, last;
int32_t *copy = NULL; /* XXX ondisk32 */
daddr_t blkcount, rblkcount, factor;
int nblocks;
daddr_t blocksreleased = 0, real_released = 0;
int error = 0, allerror = 0;
ASSERT_SEGLOCK(fs);
/*
* Calculate index in current block of last
* block to be kept. -1 indicates the entire
* block so we need not calculate the index.
*/
factor = 1;
for (i = SINGLE; i < level; i++)
factor *= LFS_NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
/*
* Get buffer of block pointers, zero those entries corresponding
* to blocks to be free'd, and update on disk copy first. Since
* double(triple) indirect before single(double) indirect, calls
* to bmap on these blocks will fail. However, we already have
* the on disk address, so we have to set the b_blkno field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lbn, lfs_sb_getbsize(fs), 0, 0);
if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
/* Braces must be here in case trace evaluates to nothing. */
trace(TR_BREADHIT, pack(vp, lfs_sb_getbsize(fs)), lbn);
} else {
trace(TR_BREADMISS, pack(vp, lfs_sb_getbsize(fs)), lbn);
curlwp->l_ru.ru_inblock++; /* pay for read */
bp->b_flags |= B_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("lfs_indirtrunc: bad buffer size");
bp->b_blkno = LFS_FSBTODB(fs, dbn);
VOP_STRATEGY(vp, bp);
error = biowait(bp);
}
if (error) {
brelse(bp, 0);
*countp = *rcountp = 0;
return (error);
}
bap = (int32_t *)bp->b_data; /* XXX ondisk32 */
if (lastbn >= 0) {
copy = lfs_malloc(fs, lfs_sb_getbsize(fs), LFS_NB_IBLOCK);
memcpy((void *)copy, (void *)bap, lfs_sb_getbsize(fs));
memset((void *)&bap[last + 1], 0,
/* XXX ondisk32 */
(u_int)(LFS_NINDIR(fs) - (last + 1)) * sizeof (int32_t));
error = VOP_BWRITE(bp->b_vp, bp);
if (error)
allerror = error;
bap = copy;
}
/*
* Recursively free totally unused blocks.
*/
for (i = LFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = bap[i];
if (nb == 0)
continue;
if (level > SINGLE) {
error = lfs_indirtrunc(ip, nlbn, nb,
(daddr_t)-1, level - 1,
&blkcount, &rblkcount,
lastsegp, bcp);
if (error)
allerror = error;
blocksreleased += blkcount;
real_released += rblkcount;
}
lfs_blkfree(fs, ip, nb, lfs_sb_getbsize(fs), lastsegp, bcp);
if (bap[i] > 0)
real_released += nblocks;
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = bap[i];
if (nb != 0) {
error = lfs_indirtrunc(ip, nlbn, nb,
last, level - 1, &blkcount,
&rblkcount, lastsegp, bcp);
if (error)
allerror = error;
real_released += rblkcount;
blocksreleased += blkcount;
}
}
if (copy != NULL) {
lfs_free(fs, copy, LFS_NB_IBLOCK);
} else {
mutex_enter(&bufcache_lock);
if (bp->b_oflags & BO_DELWRI) {
LFS_UNLOCK_BUF(bp);
lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
wakeup(&fs->lfs_availsleep);
}
brelsel(bp, BC_INVAL);
mutex_exit(&bufcache_lock);
}
*countp = blocksreleased;
*rcountp = real_released;
return (allerror);
}
int
lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
{
daddr_t lastblock;
struct inode *oip = VTOI(ovp);
daddr_t bn, lbn, lastiblock[ULFS_NIADDR], indir_lbn[ULFS_NIADDR];
/* XXX ondisk32 */
int32_t newblks[ULFS_NDADDR + ULFS_NIADDR];
struct lfs *fs;
struct buf *bp;
int offset, size, level;
daddr_t count, rcount;
daddr_t blocksreleased = 0, real_released = 0;
int i, nblocks;
int aflags, error, allerror = 0;
off_t osize;
long lastseg;
size_t bc;
int obufsize, odb;
int usepc;
if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
KASSERT(oip->i_size == 0);
return 0;
}
if (length < 0)
return (EINVAL);
/*
* Just return and not update modification times.
*/
if (oip->i_size == length) {
/* still do a uvm_vnp_setsize() as writesize may be larger */
uvm_vnp_setsize(ovp, length);
return (0);
}
fs = oip->i_lfs;
if (ovp->v_type == VLNK &&
(oip->i_size < fs->um_maxsymlinklen ||
(fs->um_maxsymlinklen == 0 &&
oip->i_ffs1_blocks == 0))) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("lfs_truncate: partial truncate of symlink");
#endif
memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
oip->i_size = oip->i_ffs1_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (lfs_update(ovp, NULL, NULL, 0));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (lfs_update(ovp, NULL, NULL, 0));
}
lfs_imtime(fs);
osize = oip->i_size;
usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
ASSERT_NO_SEGLOCK(fs);
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
if (length > fs->um_maxfilesize)
return (EFBIG);
aflags = B_CLRBUF;
if (ioflag & IO_SYNC)
aflags |= B_SYNC;
if (usepc) {
if (lfs_lblkno(fs, osize) < ULFS_NDADDR &&
lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) &&
lfs_blkroundup(fs, osize) != osize) {
off_t eob;
eob = lfs_blkroundup(fs, osize);
uvm_vnp_setwritesize(ovp, eob);
error = ulfs_balloc_range(ovp, osize,
eob - osize, cred, aflags);
if (error) {
(void) lfs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
if (ioflag & IO_SYNC) {
mutex_enter(ovp->v_interlock);
VOP_PUTPAGES(ovp,
trunc_page(osize & lfs_sb_getbmask(fs)),
round_page(eob),
PGO_CLEANIT | PGO_SYNCIO);
}
}
uvm_vnp_setwritesize(ovp, length);
error = ulfs_balloc_range(ovp, length - 1, 1, cred,
aflags);
if (error) {
(void) lfs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
uvm_vnp_setsize(ovp, length);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
KASSERT(ovp->v_size == oip->i_size);
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
return (lfs_update(ovp, NULL, NULL, 0));
} else {
error = lfs_reserve(fs, ovp, NULL,
lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
if (error)
return (error);
error = lfs_balloc(ovp, length - 1, 1, cred,
aflags, &bp);
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
if (error)
return (error);
oip->i_ffs1_size = oip->i_size = length;
uvm_vnp_setsize(ovp, length);
(void) VOP_BWRITE(bp->b_vp, bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
return (lfs_update(ovp, NULL, NULL, 0));
}
}
if ((error = lfs_reserve(fs, ovp, NULL,
lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 0)
return (error);
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundary, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever becomes accessible again because
* of subsequent file growth. Directories however are not
* zero'ed as they should grow back initialized to empty.
*/
offset = lfs_blkoff(fs, length);
lastseg = -1;
bc = 0;
if (ovp != fs->lfs_ivnode)
lfs_seglock(fs, SEGM_PROT);
if (offset == 0) {
oip->i_size = oip->i_ffs1_size = length;
} else if (!usepc) {
lbn = lfs_lblkno(fs, length);
aflags = B_CLRBUF;
if (ioflag & IO_SYNC)
aflags |= B_SYNC;
error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
obufsize = bp->b_bufsize;
odb = lfs_btofsb(fs, bp->b_bcount);
oip->i_size = oip->i_ffs1_size = length;
size = lfs_blksize(fs, oip, lbn);
if (ovp->v_type != VDIR)
memset((char *)bp->b_data + offset, 0,
(u_int)(size - offset));
allocbuf(bp, size, 1);
if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
mutex_enter(&lfs_lock);
locked_queue_bytes -= obufsize - bp->b_bufsize;
mutex_exit(&lfs_lock);
}
if (bp->b_oflags & BO_DELWRI) {
lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size));
/* XXX shouldn't this wake up on lfs_availsleep? */
}
(void) VOP_BWRITE(bp->b_vp, bp);
} else { /* vp->v_type == VREG && length < osize && offset != 0 */
/*
* When truncating a regular file down to a non-block-aligned
* size, we must zero the part of last block which is past
* the new EOF. We must synchronously flush the zeroed pages
* to disk since the new pages will be invalidated as soon
* as we inform the VM system of the new, smaller size.
* We must do this before acquiring the GLOCK, since fetching
* the pages will acquire the GLOCK internally.
* So there is a window where another thread could see a whole
* zeroed page past EOF, but that's life.
*/
daddr_t xlbn;
voff_t eoz;
aflags = ioflag & IO_SYNC ? B_SYNC : 0;
error = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags);
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
xlbn = lfs_lblkno(fs, length);
size = lfs_blksize(fs, oip, xlbn);
eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize);
ubc_zerorange(&ovp->v_uobj, length, eoz - length,
UBC_UNMAP_FLAG(ovp));
if (round_page(eoz) > round_page(length)) {
mutex_enter(ovp->v_interlock);
error = VOP_PUTPAGES(ovp, round_page(length),
round_page(eoz),
PGO_CLEANIT | PGO_DEACTIVATE |
((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
}
}
genfs_node_wrlock(ovp);
oip->i_size = oip->i_ffs1_size = length;
uvm_vnp_setsize(ovp, length);
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
if (length > QUAD_MAX - lfs_sb_getbsize(fs))
lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs));
else
lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1;
lastiblock[SINGLE] = lastblock - ULFS_NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs);
nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
/*
* Record changed file and block pointers before we start
* freeing blocks. lastiblock values are also normalized to -1
* for calls to lfs_indirtrunc below.
*/
memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
newblks[ULFS_NDADDR+level] = 0;
lastiblock[level] = -1;
}
for (i = ULFS_NDADDR - 1; i > lastblock; i--)
newblks[i] = 0;
oip->i_size = oip->i_ffs1_size = osize;
error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
if (error && !allerror)
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -ULFS_NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ffs1_ib[level];
if (bn != 0) {
error = lfs_indirtrunc(oip, indir_lbn[level],
bn, lastiblock[level],
level, &count, &rcount,
&lastseg, &bc);
if (error)
allerror = error;
real_released += rcount;
blocksreleased += count;
if (lastiblock[level] < 0) {
if (oip->i_ffs1_ib[level] > 0)
real_released += nblocks;
blocksreleased += nblocks;
oip->i_ffs1_ib[level] = 0;
lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs),
&lastseg, &bc);
lfs_deregister_block(ovp, bn);
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = ULFS_NDADDR - 1; i > lastblock; i--) {
long bsize, obsize;
bn = oip->i_ffs1_db[i];
if (bn == 0)
continue;
bsize = lfs_blksize(fs, oip, i);
if (oip->i_ffs1_db[i] > 0) {
/* Check for fragment size changes */
obsize = oip->i_lfs_fragsize[i];
real_released += lfs_btofsb(fs, obsize);
oip->i_lfs_fragsize[i] = 0;
} else
obsize = 0;
blocksreleased += lfs_btofsb(fs, bsize);
oip->i_ffs1_db[i] = 0;
lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
lfs_deregister_block(ovp, bn);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_ffs1_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
#if 0
long olddspace;
#endif
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = lfs_blksize(fs, oip, lastblock);
#if 0
olddspace = oip->i_lfs_fragsize[lastblock];
#endif
oip->i_size = oip->i_ffs1_size = length;
newspace = lfs_blksize(fs, oip, lastblock);
if (newspace == 0)
panic("itrunc: newspace");
if (oldspace - newspace > 0) {
blocksreleased += lfs_btofsb(fs, oldspace - newspace);
}
#if 0
if (bn > 0 && olddspace - newspace > 0) {
/* No segment accounting here, just vnode */
real_released += lfs_btofsb(fs, olddspace - newspace);
}
#endif
}
done:
/* Finish segment accounting corrections */
lfs_update_seguse(fs, oip, lastseg, bc);
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if ((newblks[ULFS_NDADDR + level] == 0) !=
((oip->i_ffs1_ib[level]) == 0)) {
panic("lfs itrunc1");
}
for (i = 0; i < ULFS_NDADDR; i++)
if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
panic("lfs itrunc2");
}
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("lfs itrunc3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = oip->i_ffs1_size = length;
oip->i_lfs_effnblks -= blocksreleased;
oip->i_ffs1_blocks -= real_released;
mutex_enter(&lfs_lock);
lfs_sb_addbfree(fs, blocksreleased);
mutex_exit(&lfs_lock);
#ifdef DIAGNOSTIC
if (oip->i_size == 0 &&
(oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
printf("lfs_truncate: truncate to 0 but %d blks/%jd effblks\n",
oip->i_ffs1_blocks, (intmax_t)oip->i_lfs_effnblks);
panic("lfs_truncate: persistent blocks");
}
#endif
/*
* If we truncated to zero, take us off the paging queue.
*/
mutex_enter(&lfs_lock);
if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
oip->i_flags &= ~IN_PAGING;
TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
}
mutex_exit(&lfs_lock);
oip->i_flag |= IN_CHANGE;
#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
(void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
genfs_node_unlock(ovp);
errout:
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
if (ovp != fs->lfs_ivnode)
lfs_segunlock(fs);
return (allerror ? allerror : error);
}
