static int virtualblocks(struct fs *super, union dinode *dp)
{
off_t nblk, sz;
sz = DIP(super, dp, di_size);
#ifdef COMPAT
if (lblkno(super,sz) >= NDADDR) {
nblk = blkroundup(super,sz);
if (sz == nblk)
nblk += super->fs_bsize;
}
return sz / 1024;
#else /* COMPAT */
if (lblkno(super,sz) >= NDADDR) {
nblk = blkroundup(super,sz);
sz = lblkno(super,nblk);
sz = (sz - NDADDR + NINDIR(super) - 1) / NINDIR(super);
while (sz > 0) {
nblk += sz * super->fs_bsize;
/* sz - 1 rounded up */
sz = (sz - 1 + NINDIR(super) - 1) / NINDIR(super);
}
} else
nblk = fragroundup(super,sz);
return nblk / 512;
#endif /* COMPAT */
}
static int
virtualblocks(struct fs *super, union dinode *dp)
{
off_t nblk, sz;
sz = DIP(super, dp, di_size);
#ifdef COMPAT
if (lblkno(super, sz) >= NDADDR) {
nblk = blkroundup(super, sz);
if (sz == nblk)
nblk += super->fs_bsize;
}
return sz / 1024;
#else /* COMPAT */
if (lblkno(super, sz) >= NDADDR) {
nblk = blkroundup(super, sz);
sz = lblkno(super, nblk);
sz = howmany(sz - NDADDR, NINDIR(super));
while (sz > 0) {
nblk += sz * super->fs_bsize;
/* One block on this level is in the inode itself */
sz = howmany(sz - 1, NINDIR(super));
}
} else
nblk = fragroundup(super, sz);
return nblk / DEV_BSIZE;
#endif /* COMPAT */
}
void
setinodebuf(ufs1_ino_t inum)
{
if (inum % sblock.fs_ipg != 0)
errx(EEXIT, "bad inode number %d to setinodebuf", inum);
startinum = 0;
nextino = inum;
lastinum = inum;
readcnt = 0;
if (inodebuf != NULL)
return;
inobufsize = blkroundup(&sblock, INOBUFSIZE);
fullcnt = inobufsize / sizeof(struct ufs1_dinode);
readpercg = sblock.fs_ipg / fullcnt;
partialcnt = sblock.fs_ipg % fullcnt;
partialsize = partialcnt * sizeof(struct ufs1_dinode);
if (partialcnt != 0) {
readpercg++;
} else {
partialcnt = fullcnt;
partialsize = inobufsize;
}
if ((inodebuf = (struct ufs1_dinode *)malloc((unsigned)inobufsize)) == NULL)
errx(EEXIT, "cannot allocate space for inode buffer");
}
/*
* Prepare to scan a set of inodes.
*/
void
setinodebuf(ino_t inum)
{
if (inum % sblock.fs_ipg != 0)
errx(1, "bad inode number %ju to setinodebuf", (uintmax_t)inum);
lastvalidinum = inum + sblock.fs_ipg - 1;
nextino = inum;
lastinum = inum;
readcnt = 0;
if (inodebuf != NULL)
return;
inobufsize = blkroundup(&sblock, INOBUFSIZE);
fullcnt = inobufsize / ((sblock.fs_magic == FS_UFS1_MAGIC) ?
sizeof(struct ufs1_dinode) : sizeof(struct ufs2_dinode));
readpercg = sblock.fs_ipg / fullcnt;
partialcnt = sblock.fs_ipg % fullcnt;
partialsize = partialcnt * ((sblock.fs_magic == FS_UFS1_MAGIC) ?
sizeof(struct ufs1_dinode) : sizeof(struct ufs2_dinode));
if (partialcnt != 0) {
readpercg++;
} else {
partialcnt = fullcnt;
partialsize = inobufsize;
}
if ((inodebuf = malloc((unsigned)inobufsize)) == NULL)
errx(1, "cannot allocate space for inode buffer");
}
void
ffs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
{
struct inode *ip = VTOI(vp);
struct fs *fs = ip->i_fs;
daddr_t olbn, nlbn;
olbn = lblkno(fs, ip->i_size);
nlbn = lblkno(fs, size);
if (nlbn < NDADDR && olbn <= nlbn) {
*eobp = fragroundup(fs, size);
} else {
*eobp = blkroundup(fs, size);
}
}
/*
* Prepare to scan a set of inodes.
*/
void
resetinodebuf(void)
{
nextino = 0;
lastinum = 0;
readcnt = 0;
inobufsize = blkroundup(&sblock, INOBUFSIZE);
fullcnt = inobufsize / sizeof(struct ufs1_dinode);
readpercg = sblock.fs_ipg / fullcnt;
partialcnt = sblock.fs_ipg % fullcnt;
partialsize = partialcnt * sizeof(struct ufs1_dinode);
if (partialcnt != 0) {
readpercg++;
} else {
partialcnt = fullcnt;
partialsize = inobufsize;
}
if (inodebuf == NULL &&
(inodebuf = malloc((u_int)inobufsize)) == NULL)
errx(1, "malloc failed");
while (nextino < ROOTINO)
getnextinode(nextino);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
{
daddr_t lastblock;
struct inode *oip = VTOI(ovp);
daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
daddr_t blks[NDADDR + NIADDR];
struct fs *fs;
int offset, pgoffset, level;
int64_t count, blocksreleased = 0;
int i, aflag, nblocks;
int error, allerror = 0;
off_t osize;
int sync;
struct ufsmount *ump = oip->i_ump;
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);
if (ovp->v_type == VLNK &&
(oip->i_size < ump->um_maxsymlinklen ||
(ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) {
KDASSERT(length == 0);
memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
oip->i_size = 0;
DIP_ASSIGN(oip, size, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, NULL, NULL, 0));
}
if (oip->i_size == length) {
/* still do a uvm_vnp_setsize() as writesize may be larger */
uvm_vnp_setsize(ovp, length);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, NULL, NULL, 0));
}
fs = oip->i_fs;
if (length > ump->um_maxfilesize)
return (EFBIG);
if ((oip->i_flags & SF_SNAPSHOT) != 0)
ffs_snapremove(ovp);
osize = oip->i_size;
aflag = ioflag & IO_SYNC ? B_SYNC : 0;
/*
* 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 (lblkno(fs, osize) < NDADDR &&
lblkno(fs, osize) != lblkno(fs, length) &&
blkroundup(fs, osize) != osize) {
off_t eob;
eob = blkroundup(fs, osize);
uvm_vnp_setwritesize(ovp, eob);
error = ufs_balloc_range(ovp, osize, eob - osize,
cred, aflag);
if (error) {
(void) ffs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
if (ioflag & IO_SYNC) {
mutex_enter(ovp->v_interlock);
VOP_PUTPAGES(ovp,
trunc_page(osize & fs->fs_bmask),
round_page(eob), PGO_CLEANIT | PGO_SYNCIO |
PGO_JOURNALLOCKED);
}
}
uvm_vnp_setwritesize(ovp, length);
error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag);
if (error) {
(void) ffs_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);
return (ffs_update(ovp, NULL, NULL, 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.
*/
offset = blkoff(fs, length);
pgoffset = length & PAGE_MASK;
if (ovp->v_type == VREG && (pgoffset != 0 || offset != 0) &&
osize > length) {
daddr_t lbn;
voff_t eoz;
int size;
if (offset != 0) {
error = ufs_balloc_range(ovp, length - 1, 1, cred,
aflag);
if (error)
return error;
}
lbn = lblkno(fs, length);
size = blksize(fs, oip, lbn);
eoz = MIN(MAX(lblktosize(fs, lbn) + size, round_page(pgoffset)),
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 | PGO_JOURNALLOCKED |
((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
if (error)
return error;
}
}
genfs_node_wrlock(ovp);
oip->i_size = length;
DIP_ASSIGN(oip, 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.
*/
lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ffs_indirtrunc below.
*/
sync = 0;
for (level = TRIPLE; level >= SINGLE; level--) {
blks[NDADDR + level] = DIP(oip, ib[level]);
if (lastiblock[level] < 0 && blks[NDADDR + level] != 0) {
sync = 1;
DIP_ASSIGN(oip, ib[level], 0);
lastiblock[level] = -1;
}
}
for (i = 0; i < NDADDR; i++) {
blks[i] = DIP(oip, db[i]);
if (i > lastblock && blks[i] != 0) {
sync = 1;
DIP_ASSIGN(oip, db[i], 0);
}
}
oip->i_flag |= IN_CHANGE | IN_UPDATE;
if (sync) {
error = ffs_update(ovp, NULL, NULL, UPDATE_WAIT);
if (error && !allerror)
allerror = error;
}
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
for (i = 0; i < NDADDR; i++) {
bn = DIP(oip, db[i]);
DIP_ASSIGN(oip, db[i], blks[i]);
blks[i] = bn;
}
for (i = 0; i < NIADDR; i++) {
bn = DIP(oip, ib[i]);
DIP_ASSIGN(oip, ib[i], blks[NDADDR + i]);
blks[NDADDR + i] = bn;
}
oip->i_size = osize;
DIP_ASSIGN(oip, size, osize);
error = vtruncbuf(ovp, lastblock + 1, 0, 0);
if (error && !allerror)
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
if (oip->i_ump->um_fstype == UFS1)
bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
else
bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
if (bn != 0) {
error = ffs_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
DIP_ASSIGN(oip, ib[level], 0);
if (oip->i_ump->um_mountp->mnt_wapbl) {
UFS_WAPBL_REGISTER_DEALLOCATION(
oip->i_ump->um_mountp,
fsbtodb(fs, bn), fs->fs_bsize);
} else
ffs_blkfree(fs, oip->i_devvp, bn,
fs->fs_bsize, oip->i_number);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
if (oip->i_ump->um_fstype == UFS1)
bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
else
bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
if (bn == 0)
continue;
DIP_ASSIGN(oip, db[i], 0);
bsize = blksize(fs, oip, i);
if ((oip->i_ump->um_mountp->mnt_wapbl) &&
(ovp->v_type != VREG)) {
UFS_WAPBL_REGISTER_DEALLOCATION(oip->i_ump->um_mountp,
fsbtodb(fs, bn), bsize);
} else
ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
if (oip->i_ump->um_fstype == UFS1)
bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
else
bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
DIP_ASSIGN(oip, size, length);
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("itrunc: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
if ((oip->i_ump->um_mountp->mnt_wapbl) &&
(ovp->v_type != VREG)) {
UFS_WAPBL_REGISTER_DEALLOCATION(
oip->i_ump->um_mountp, fsbtodb(fs, bn),
oldspace - newspace);
} else
ffs_blkfree(fs, oip->i_devvp, bn,
oldspace - newspace, oip->i_number);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (blks[NDADDR + level] != DIP(oip, ib[level]))
panic("itrunc1");
for (i = 0; i < NDADDR; i++)
if (blks[i] != DIP(oip, db[i]))
panic("itrunc2");
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("itrunc3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = length;
DIP_ASSIGN(oip, size, length);
DIP_ADD(oip, blocks, -blocksreleased);
genfs_node_unlock(ovp);
oip->i_flag |= IN_CHANGE;
UFS_WAPBL_UPDATE(ovp, NULL, NULL, 0);
#if defined(QUOTA) || defined(QUOTA2)
(void) chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
return (allerror);
}
/*
* Enable logging
*/
int
lqfs_enable(struct vnode *vp, struct fiolog *flp, cred_t *cr)
{
int error;
inode_t *ip = VTOI(vp);
qfsvfs_t *qfsvfsp = ip->i_qfsvfs;
fs_lqfs_common_t *fs = VFS_FS_PTR(qfsvfsp);
ml_unit_t *ul;
#ifdef LQFS_TODO_LOCKFS
int reclaim = 0;
struct lockfs lf;
struct ulockfs *ulp;
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
vfs_t *vfsp = qfsvfsp->vfs_vfs;
uint64_t tmp_nbytes_actual;
char fsclean;
sam_sblk_t *sblk = qfsvfsp->mi.m_sbp;
/*
* File system is not capable of logging.
*/
if (!LQFS_CAPABLE(qfsvfsp)) {
flp->error = FIOLOG_ENOTSUP;
error = 0;
goto out;
}
if (!SAM_MAGIC_V2A_OR_HIGHER(&sblk->info.sb)) {
cmn_err(CE_WARN, "SAM-QFS: %s: Not enabling logging, "
" file system is not version 2A.", qfsvfsp->mt.fi_name);
cmn_err(CE_WARN, "\tUpgrade file system with samfsck -u "
"first.");
flp->error = FIOLOG_ENOTSUP;
error = 0;
goto out;
}
if (LQFS_GET_LOGBNO(fs)) {
error = lqfs_log_validate(qfsvfsp, flp, cr);
}
/*
* Check if logging is already enabled
*/
if (LQFS_GET_LOGP(qfsvfsp)) {
flp->error = FIOLOG_ETRANS;
/* for root ensure logging option is set */
vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
error = 0;
goto out;
}
/*
* Come back here to recheck if we had to disable the log.
*/
recheck:
error = 0;
flp->error = FIOLOG_ENONE;
/*
* Adjust requested log size
*/
flp->nbytes_actual = flp->nbytes_requested;
if (flp->nbytes_actual == 0) {
tmp_nbytes_actual =
(((uint64_t)FS_SIZE(fs)) / ldl_divisor) << FS_FSHIFT(fs);
flp->nbytes_actual = (uint_t)MIN(tmp_nbytes_actual, INT_MAX);
}
flp->nbytes_actual = MAX(flp->nbytes_actual, ldl_minlogsize);
flp->nbytes_actual = MIN(flp->nbytes_actual, ldl_maxlogsize);
flp->nbytes_actual = blkroundup(fs, flp->nbytes_actual);
/*
* logging is enabled and the log is the right size; done
*/
ul = LQFS_GET_LOGP(qfsvfsp);
if (ul && LQFS_GET_LOGBNO(fs) &&
(flp->nbytes_actual == ul->un_requestsize)) {
vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
error = 0;
goto out;
}
/*
* Readonly file system
*/
if (FS_RDONLY(fs)) {
flp->error = FIOLOG_EROFS;
error = 0;
goto out;
}
#ifdef LQFS_TODO_LOCKFS
/*
* File system must be write locked to enable logging
*/
error = qfs_fiolfss(vp, &lf);
if (error) {
goto out;
}
if (!LOCKFS_IS_ULOCK(&lf)) {
flp->error = FIOLOG_EULOCK;
error = 0;
goto out;
}
lf.lf_lock = LOCKFS_WLOCK;
lf.lf_flags = 0;
lf.lf_comment = NULL;
error = qfs_fiolfs(vp, &lf, 1);
if (error) {
flp->error = FIOLOG_EWLOCK;
error = 0;
goto out;
}
#else
/* QFS doesn't really support lockfs. */
#endif /* LQFS_TODO_LOCKFS */
/*
* Grab appropriate locks to synchronize with the rest
* of the system
*/
vfs_lock_wait(vfsp);
#ifdef LQFS_TODO_LOCKFS
ulp = &ufsvfsp->vfs_ulockfs;
mutex_enter(&ulp->ul_lock);
#else
/* QFS doesn't really support lockfs. */
#endif /* LQFS_TODO_LOCKFS */
/*
* File system must be fairly consistent to enable logging
*/
fsclean = LQFS_GET_FS_CLEAN(fs);
if (fsclean != FSLOG &&
fsclean != FSACTIVE &&
fsclean != FSSTABLE &&
fsclean != FSCLEAN) {
flp->error = FIOLOG_ECLEAN;
goto unlockout;
}
#ifdef LUFS
/*
* A write-locked file system is only active if there are
* open deleted files; so remember to set FS_RECLAIM later.
*/
if (LQFS_GET_FS_CLEAN(fs) == FSACTIVE) {
reclaim = FS_RECLAIM;
}
#else
/* QFS doesn't have a reclaim file thread. */
#endif /* LUFS */
/*
* Logging is already enabled; must be changing the log's size
*/
if (LQFS_GET_LOGBNO(fs) && LQFS_GET_LOGP(qfsvfsp)) {
#ifdef LQFS_TODO_LOCKFS
/*
* Before we can disable logging, we must give up our
* lock. As a consequence of unlocking and disabling the
* log, the fs structure may change. Because of this, when
* disabling is complete, we will go back to recheck to
* repeat all of the checks that we performed to get to
* this point. Disabling sets fs->fs_logbno to 0, so this
* will not put us into an infinite loop.
*/
mutex_exit(&ulp->ul_lock);
#else
/* QFS doesn't really support lockfs. */
#endif /* LQFS_TODO_LOCKFS */
vfs_unlock(vfsp);
#ifdef LQFS_TODO_LOCKFS
lf.lf_lock = LOCKFS_ULOCK;
lf.lf_flags = 0;
error = qfs_fiolfs(vp, &lf, 1);
if (error) {
flp->error = FIOLOG_ENOULOCK;
error = 0;
goto out;
}
#else
/* QFS doesn't really support lockfs. */
#endif /* LQFS_TODO_LOCKFS */
error = lqfs_disable(vp, flp);
if (error || (flp->error != FIOLOG_ENONE)) {
error = 0;
goto out;
}
goto recheck;
}
error = lqfs_alloc(qfsvfsp, flp, cr);
if (error) {
goto errout;
}
#ifdef LUFS
#else
if ((error = lqfs_log_validate(qfsvfsp, flp, cr)) != 0) {
goto errout;
}
#endif /* LUFS */
/*
* Create all of the incore structs
*/
error = lqfs_snarf(qfsvfsp, fs, 0);
if (error) {
goto errout;
}
/*
* DON'T ``GOTO ERROUT'' PAST THIS POINT
*/
/*
* Pretend we were just mounted with logging enabled
* freeze and drain the file system of readers
* Get the ops vector
* If debug, record metadata locations with log subsystem
* Start the delete thread
* Start the reclaim thread, if necessary
* Thaw readers
*/
vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
TRANS_DOMATAMAP(qfsvfsp);
TRANS_MATA_MOUNT(qfsvfsp);
TRANS_MATA_SI(qfsvfsp, fs);
#ifdef LUFS
qfs_thread_start(&qfsvfsp->vfs_delete, qfs_thread_delete, vfsp);
if (fs->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) {
fs->fs_reclaim &= ~FS_RECLAIM;
fs->fs_reclaim |= FS_RECLAIMING;
qfs_thread_start(&qfsvfsp->vfs_reclaim,
qfs_thread_reclaim, vfsp);
} else {
fs->fs_reclaim |= reclaim;
}
#else
/* QFS doesn't have file reclaim nor i-node delete threads. */
#endif /* LUFS */
#ifdef LUFS
mutex_exit(&ulp->ul_lock);
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LUFS */
vfs_unlock(vfsp);
#ifdef LQFS_TODO_LOCKFS
/*
* Unlock the file system
*/
lf.lf_lock = LOCKFS_ULOCK;
lf.lf_flags = 0;
error = qfs_fiolfs(vp, &lf, 1);
if (error) {
flp->error = FIOLOG_ENOULOCK;
error = 0;
goto out;
}
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
/*
* There's nothing in the log yet (we've just allocated it)
* so directly write out the super block.
* Note, we have to force this sb out to disk
* (not just to the log) so that if we crash we know we are logging
*/
VFS_LOCK_MUTEX_ENTER(qfsvfsp);
LQFS_SET_FS_CLEAN(fs, FSLOG);
LQFS_SET_FS_ROLLED(fs, FS_NEED_ROLL); /* Mark the fs as unrolled */
#ifdef LUFS
QFS_BWRITE2(NULL, qfsvfsp->vfs_bufp);
#else
sam_update_sblk(qfsvfsp, 0, 0, TRUE);
#endif /* LUFS */
VFS_LOCK_MUTEX_EXIT(qfsvfsp);
error = 0;
goto out;
errout:
/*
* Aquire the qfs_scan_lock before de-linking the mtm data
* structure so that we keep qfs_sync() and qfs_update() away
* when they execute the ufs_scan_inodes() run while we're in
* progress of enabling/disabling logging.
*/
mutex_enter(&qfs_scan_lock);
(void) lqfs_unsnarf(qfsvfsp);
mutex_exit(&qfs_scan_lock);
(void) lqfs_free(qfsvfsp);
unlockout:
#ifdef LQFS_TODO_LOCKFS
mutex_exit(&ulp->ul_lock);
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
vfs_unlock(vfsp);
#ifdef LQFS_TODO_LOCKFS
lf.lf_lock = LOCKFS_ULOCK;
lf.lf_flags = 0;
(void) qfs_fiolfs(vp, &lf, 1);
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
out:
mutex_enter(&ip->mp->ms.m_waitwr_mutex);
ip->mp->mt.fi_status |= FS_LOGSTATE_KNOWN;
mutex_exit(&ip->mp->ms.m_waitwr_mutex);
return (error);
}
/*
* do a simple estimate of the space needed to hold the statefile
* taking compression into account, but be fairly conservative
* so we have a better chance of completing; when dump fails,
* the retry cost is fairly high.
*
* Do disk blocks allocation for the state file if no space has
* been allocated yet. Since the state file will not be removed,
* allocation should only be done once.
*/
static int
cpr_statefile_ok(vnode_t *vp, int alloc_retry)
{
extern size_t cpr_bitmap_size;
struct inode *ip = VTOI(vp);
const int UCOMP_RATE = 20; /* comp. ratio*10 for user pages */
u_longlong_t size, isize, ksize, raw_data;
char *str, *est_fmt;
size_t space;
int error;
/*
* number of pages short for swapping.
*/
STAT->cs_nosw_pages = k_anoninfo.ani_mem_resv;
if (STAT->cs_nosw_pages < 0)
STAT->cs_nosw_pages = 0;
str = "cpr_statefile_ok:";
CPR_DEBUG(CPR_DEBUG9, "Phys swap: max=%lu resv=%lu\n",
k_anoninfo.ani_max, k_anoninfo.ani_phys_resv);
CPR_DEBUG(CPR_DEBUG9, "Mem swap: max=%ld resv=%lu\n",
MAX(availrmem - swapfs_minfree, 0),
k_anoninfo.ani_mem_resv);
CPR_DEBUG(CPR_DEBUG9, "Total available swap: %ld\n",
CURRENT_TOTAL_AVAILABLE_SWAP);
/*
* try increasing filesize by 15%
*/
if (alloc_retry) {
/*
* block device doesn't get any bigger
*/
if (vp->v_type == VBLK) {
if (cpr_debug & (CPR_DEBUG1 | CPR_DEBUG6))
prom_printf(
"Retry statefile on special file\n");
return (ENOMEM);
} else {
rw_enter(&ip->i_contents, RW_READER);
size = (ip->i_size * SIZE_RATE) / INTEGRAL;
rw_exit(&ip->i_contents);
}
if (cpr_debug & (CPR_DEBUG1 | CPR_DEBUG6))
prom_printf("Retry statefile size = %lld\n", size);
} else {
u_longlong_t cpd_size;
pgcnt_t npages, nback;
int ndvram;
ndvram = 0;
(void) callb_execute_class(CB_CL_CPR_FB,
(int)(uintptr_t)&ndvram);
if (cpr_debug & (CPR_DEBUG1 | CPR_DEBUG6))
prom_printf("ndvram size = %d\n", ndvram);
/*
* estimate 1 cpd_t for every (CPR_MAXCONTIG / 2) pages
*/
npages = cpr_count_kpages(REGULAR_BITMAP, cpr_nobit);
cpd_size = sizeof (cpd_t) * (npages / (CPR_MAXCONTIG / 2));
raw_data = cpd_size + cpr_bitmap_size;
ksize = ndvram + mmu_ptob(npages);
est_fmt = "%s estimated size with "
"%scompression %lld, ksize %lld\n";
nback = mmu_ptob(STAT->cs_nosw_pages);
if (CPR->c_flags & C_COMPRESSING) {
size = ((ksize * COMPRESS_PERCENT) / INTEGRAL) +
raw_data + ((nback * 10) / UCOMP_RATE);
CPR_DEBUG(CPR_DEBUG1, est_fmt, str, "", size, ksize);
} else {
size = ksize + raw_data + nback;
CPR_DEBUG(CPR_DEBUG1, est_fmt, str, "no ",
size, ksize);
}
}
/*
* All this is much simpler for a block device
*/
if (vp->v_type == VBLK) {
space = cpr_get_devsize(vp->v_rdev);
if (cpr_debug & (CPR_DEBUG1 | CPR_DEBUG6))
prom_printf("statefile dev size %lu\n", space);
/*
* Export the estimated filesize info, this value will be
* compared before dumping out the statefile in the case of
* no compression.
*/
STAT->cs_est_statefsz = size;
if (cpr_debug & (CPR_DEBUG1 | CPR_DEBUG6))
prom_printf("%s Estimated statefile size %llu, "
"space %lu\n", str, size, space);
if (size > space) {
cpr_err(CE_CONT, "Statefile partition too small.");
return (ENOMEM);
}
return (0);
} else {
if (CPR->c_alloc_cnt++ > C_MAX_ALLOC_RETRY) {
cpr_err(CE_CONT, "Statefile allocation retry failed\n");
return (ENOMEM);
}
/*
* Estimate space needed for the state file.
*
* State file size in bytes:
* kernel size + non-cache pte seg +
* bitmap size + cpr state file headers size
* (round up to fs->fs_bsize)
*/
size = blkroundup(ip->i_fs, size);
/*
* Export the estimated filesize info, this value will be
* compared before dumping out the statefile in the case of
* no compression.
*/
STAT->cs_est_statefsz = size;
error = cpr_grow_statefile(vp, size);
if (cpr_debug & (CPR_DEBUG1 | CPR_DEBUG6)) {
rw_enter(&ip->i_contents, RW_READER);
isize = ip->i_size;
rw_exit(&ip->i_contents);
prom_printf("%s Estimated statefile size %lld, "
"i_size %lld\n", str, size, isize);
}
return (error);
}
}
