int
ufs_balloc_range(struct vnode *vp, off_t off, off_t len, kauth_cred_t cred,
int flags)
{
off_t neweof; /* file size after the operation */
off_t neweob; /* offset next to the last block after the operation */
off_t pagestart; /* starting offset of range covered by pgs */
off_t eob; /* offset next to allocated blocks */
struct uvm_object *uobj;
int i, delta, error, npages;
int bshift = vp->v_mount->mnt_fs_bshift;
int bsize = 1 << bshift;
int ppb = MAX(bsize >> PAGE_SHIFT, 1);
struct vm_page **pgs;
size_t pgssize;
UVMHIST_FUNC("ufs_balloc_range"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "vp %p off 0x%x len 0x%x u_size 0x%x",
vp, off, len, vp->v_size);
neweof = MAX(vp->v_size, off + len);
GOP_SIZE(vp, neweof, &neweob, 0);
error = 0;
uobj = &vp->v_uobj;
/*
* read or create pages covering the range of the allocation and
* keep them locked until the new block is allocated, so there
* will be no window where the old contents of the new block are
* visible to racing threads.
*/
pagestart = trunc_page(off) & ~(bsize - 1);
npages = MIN(ppb, (round_page(neweob) - pagestart) >> PAGE_SHIFT);
pgssize = npages * sizeof(struct vm_page *);
pgs = kmem_zalloc(pgssize, KM_SLEEP);
mutex_enter(&uobj->vmobjlock);
error = VOP_GETPAGES(vp, pagestart, pgs, &npages, 0,
VM_PROT_WRITE, 0,
PGO_SYNCIO|PGO_PASTEOF|PGO_NOBLOCKALLOC|PGO_NOTIMESTAMP);
if (error) {
goto out;
}
mutex_enter(&uobj->vmobjlock);
mutex_enter(&uvm_pageqlock);
for (i = 0; i < npages; i++) {
UVMHIST_LOG(ubchist, "got pgs[%d] %p", i, pgs[i],0,0);
KASSERT((pgs[i]->flags & PG_RELEASED) == 0);
pgs[i]->flags &= ~PG_CLEAN;
uvm_pageactivate(pgs[i]);
}
mutex_exit(&uvm_pageqlock);
mutex_exit(&uobj->vmobjlock);
/*
* adjust off to be block-aligned.
*/
delta = off & (bsize - 1);
off -= delta;
len += delta;
/*
* now allocate the range.
*/
genfs_node_wrlock(vp);
error = GOP_ALLOC(vp, off, len, flags, cred);
genfs_node_unlock(vp);
/*
* clear PG_RDONLY on any pages we are holding
* (since they now have backing store) and unbusy them.
*/
GOP_SIZE(vp, off + len, &eob, 0);
mutex_enter(&uobj->vmobjlock);
for (i = 0; i < npages; i++) {
if (error) {
pgs[i]->flags |= PG_RELEASED;
} else if (off <= pagestart + (i << PAGE_SHIFT) &&
pagestart + ((i + 1) << PAGE_SHIFT) <= eob) {
pgs[i]->flags &= ~PG_RDONLY;
}
}
if (error) {
mutex_enter(&uvm_pageqlock);
uvm_page_unbusy(pgs, npages);
mutex_exit(&uvm_pageqlock);
} else {
uvm_page_unbusy(pgs, npages);
}
mutex_exit(&uobj->vmobjlock);
out:
kmem_free(pgs, pgssize);
return error;
}
/*
* miscfs/genfs getpages routine. This is a fair bit simpler than the
* kernel counterpart since we're not being executed from a fault handler
* and generally don't need to care about PGO_LOCKED or other cruft.
* We do, however, need to care about page locking and we keep trying until
* we get all the pages within the range. The object locking protocol
* is the same as for the kernel: enter with the object lock held,
* return with it released.
*/
int
genfs_getpages(void *v)
{
struct vop_getpages_args /* {
struct vnode *a_vp;
voff_t a_offset;
struct vm_page **a_m;
int *a_count;
int a_centeridx;
vm_prot_t a_access_type;
int a_advice;
int a_flags;
} */ *ap = v;
struct vnode *vp = ap->a_vp;
struct uvm_object *uobj = (struct uvm_object *)vp;
struct vm_page *pg;
voff_t curoff, endoff;
off_t diskeof;
size_t bufsize, remain, bufoff, xfersize;
uint8_t *tmpbuf;
int bshift = vp->v_mount->mnt_fs_bshift;
int bsize = 1<<bshift;
int count = *ap->a_count;
int async;
int i, error;
/*
* Ignore async for now, the structure of this routine
* doesn't exactly allow for it ...
*/
async = 0;
if (ap->a_centeridx != 0)
panic("%s: centeridx != not supported", __func__);
if (ap->a_access_type & VM_PROT_WRITE)
vp->v_iflag |= VI_ONWORKLST;
curoff = ap->a_offset & ~PAGE_MASK;
for (i = 0; i < count; i++, curoff += PAGE_SIZE) {
retrylookup:
pg = uvm_pagelookup(uobj, curoff);
if (pg == NULL)
break;
/* page is busy? we need to wait until it's released */
if (pg->flags & PG_BUSY) {
pg->flags |= PG_WANTED;
UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0, "getpg",0);
mutex_enter(&uobj->vmobjlock);
goto retrylookup;
}
pg->flags |= PG_BUSY;
if (pg->flags & PG_FAKE)
break;
ap->a_m[i] = pg;
}
/* got everything? if so, just return */
if (i == count) {
mutex_exit(&uobj->vmobjlock);
return 0;
}
/*
* didn't? Ok, allocate backing pages. Start from the first
* one we missed.
*/
for (; i < count; i++, curoff += PAGE_SIZE) {
retrylookup2:
pg = uvm_pagelookup(uobj, curoff);
/* found? busy it and be happy */
if (pg) {
if (pg->flags & PG_BUSY) {
pg->flags = PG_WANTED;
UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
"getpg2", 0);
mutex_enter(&uobj->vmobjlock);
goto retrylookup2;
} else {
pg->flags |= PG_BUSY;
}
/* not found? make a new page */
} else {
pg = rumpvm_makepage(uobj, curoff);
}
ap->a_m[i] = pg;
}
/*
* We have done all the clerical work and have all pages busied.
* Release the vm object for other consumers.
*/
mutex_exit(&uobj->vmobjlock);
/*
* Now, we have all the pages here & busy. Transfer the range
* starting from the missing offset and transfer into the
* page buffers.
*/
GOP_SIZE(vp, vp->v_size, &diskeof, 0);
/* align to boundaries */
endoff = trunc_page(ap->a_offset) + (count << PAGE_SHIFT);
endoff = MIN(endoff, ((vp->v_writesize+bsize-1) & ~(bsize-1)));
curoff = ap->a_offset & ~(MAX(bsize,PAGE_SIZE)-1);
remain = endoff - curoff;
if (diskeof > curoff)
remain = MIN(remain, diskeof - curoff);
DPRINTF(("a_offset: %llx, startoff: 0x%llx, endoff 0x%llx\n",
(unsigned long long)ap->a_offset, (unsigned long long)curoff,
(unsigned long long)endoff));
/* read everything into a buffer */
bufsize = round_page(remain);
tmpbuf = kmem_zalloc(bufsize, KM_SLEEP);
for (bufoff = 0; remain; remain -= xfersize, bufoff+=xfersize) {
struct buf *bp;
struct vnode *devvp;
daddr_t lbn, bn;
int run;
lbn = (curoff + bufoff) >> bshift;
/* XXX: assume eof */
error = VOP_BMAP(vp, lbn, &devvp, &bn, &run);
if (error)
panic("%s: VOP_BMAP & lazy bum: %d", __func__, error);
DPRINTF(("lbn %d (off %d) -> bn %d run %d\n", (int)lbn,
(int)(curoff+bufoff), (int)bn, run));
xfersize = MIN(((lbn+1+run)<<bshift)-(curoff+bufoff), remain);
/* hole? */
if (bn == -1) {
memset(tmpbuf + bufoff, 0, xfersize);
continue;
}
bp = getiobuf(vp, true);
bp->b_data = tmpbuf + bufoff;
bp->b_bcount = xfersize;
bp->b_blkno = bn;
bp->b_lblkno = 0;
bp->b_flags = B_READ;
bp->b_cflags = BC_BUSY;
if (async) {
bp->b_flags |= B_ASYNC;
bp->b_iodone = uvm_aio_biodone;
}
VOP_STRATEGY(devvp, bp);
if (bp->b_error)
panic("%s: VOP_STRATEGY, lazy bum", __func__);
if (!async)
putiobuf(bp);
}
/* skip to beginning of pages we're interested in */
bufoff = 0;
while (round_page(curoff + bufoff) < trunc_page(ap->a_offset))
bufoff += PAGE_SIZE;
DPRINTF(("first page offset 0x%x\n", (int)(curoff + bufoff)));
for (i = 0; i < count; i++, bufoff += PAGE_SIZE) {
/* past our prime? */
if (curoff + bufoff >= endoff)
break;
pg = uvm_pagelookup(&vp->v_uobj, curoff + bufoff);
KASSERT(pg);
DPRINTF(("got page %p (off 0x%x)\n", pg, (int)(curoff+bufoff)));
if (pg->flags & PG_FAKE) {
memcpy((void *)pg->uanon, tmpbuf+bufoff, PAGE_SIZE);
pg->flags &= ~PG_FAKE;
pg->flags |= PG_CLEAN;
}
ap->a_m[i] = pg;
}
*ap->a_count = i;
kmem_free(tmpbuf, bufsize);
return 0;
}
/*
* Write a directory entry after a call to namei, using the parameters
* that ulfs_lookup left in nameidata and in the ulfs_lookup_results.
*
* DVP is the directory to be updated. It must be locked.
* ULR is the ulfs_lookup_results structure from the final lookup step.
* TVP is not used. (XXX: why is it here? remove it)
* DIRP is the new directory entry contents.
* CNP is the componentname from the final lookup step.
* NEWDIRBP is not used and (XXX) should be removed. The previous
* comment here said it was used by the now-removed softupdates code.
*
* The link count of the target inode is *not* incremented; the
* caller does that.
*
* If ulr->ulr_count is 0, ulfs_lookup did not find space to insert the
* directory entry. ulr_offset, which is the place to put the entry,
* should be on a block boundary (and should be at the end of the
* directory AFAIK) and a fresh block is allocated to put the new
* directory entry in.
*
* If ulr->ulr_count is not zero, ulfs_lookup found a slot to insert
* the entry into. This slot ranges from ulr_offset to ulr_offset +
* ulr_count. However, this slot may already be partially populated
* requiring compaction. See notes below.
*
* Furthermore, if ulr_count is not zero and ulr_endoff is not the
* same as i_size, the directory is truncated to size ulr_endoff.
*/
int
ulfs_direnter(struct vnode *dvp, const struct ulfs_lookup_results *ulr,
struct vnode *tvp, struct lfs_direct *dirp,
struct componentname *cnp, struct buf *newdirbp)
{
kauth_cred_t cr;
int newentrysize;
struct inode *dp;
struct buf *bp;
u_int dsize;
struct lfs_direct *ep, *nep;
int error, ret, lfs_blkoff, loc, spacefree;
char *dirbuf;
struct timespec ts;
struct ulfsmount *ump = VFSTOULFS(dvp->v_mount);
struct lfs *fs = ump->um_lfs;
const int needswap = ULFS_MPNEEDSWAP(fs);
int dirblksiz = fs->um_dirblksiz;
error = 0;
cr = cnp->cn_cred;
dp = VTOI(dvp);
newentrysize = LFS_DIRSIZ(0, dirp, 0);
if (ulr->ulr_count == 0) {
/*
* If ulr_count is 0, then namei could find no
* space in the directory. Here, ulr_offset will
* be on a directory block boundary and we will write the
* new entry into a fresh block.
*/
if (ulr->ulr_offset & (dirblksiz - 1))
panic("ulfs_direnter: newblk");
if ((error = lfs_balloc(dvp, (off_t)ulr->ulr_offset, dirblksiz,
cr, B_CLRBUF | B_SYNC, &bp)) != 0) {
return (error);
}
dp->i_size = ulr->ulr_offset + dirblksiz;
DIP_ASSIGN(dp, size, dp->i_size);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
uvm_vnp_setsize(dvp, dp->i_size);
dirp->d_reclen = ulfs_rw16(dirblksiz, needswap);
dirp->d_ino = ulfs_rw32(dirp->d_ino, needswap);
if (FSFMT(dvp)) {
#if (BYTE_ORDER == LITTLE_ENDIAN)
if (needswap == 0) {
#else
if (needswap != 0) {
#endif
u_char tmp = dirp->d_namlen;
dirp->d_namlen = dirp->d_type;
dirp->d_type = tmp;
}
}
lfs_blkoff = ulr->ulr_offset & (ump->um_mountp->mnt_stat.f_iosize - 1);
memcpy((char *)bp->b_data + lfs_blkoff, dirp, newentrysize);
#ifdef LFS_DIRHASH
if (dp->i_dirhash != NULL) {
ulfsdirhash_newblk(dp, ulr->ulr_offset);
ulfsdirhash_add(dp, dirp, ulr->ulr_offset);
ulfsdirhash_checkblock(dp, (char *)bp->b_data + lfs_blkoff,
ulr->ulr_offset);
}
#endif
error = VOP_BWRITE(bp->b_vp, bp);
vfs_timestamp(&ts);
ret = lfs_update(dvp, &ts, &ts, UPDATE_DIROP);
if (error == 0)
return (ret);
return (error);
}
/*
* If ulr_count is non-zero, then namei found space for the new
* entry in the range ulr_offset to ulr_offset + ulr_count
* in the directory. To use this space, we may have to compact
* the entries located there, by copying them together towards the
* beginning of the block, leaving the free space in one usable
* chunk at the end.
*/
/*
* Increase size of directory if entry eats into new space.
* This should never push the size past a new multiple of
* DIRBLKSIZ.
*
* N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
*/
if (ulr->ulr_offset + ulr->ulr_count > dp->i_size) {
#ifdef DIAGNOSTIC
printf("ulfs_direnter: reached 4.2-only block, "
"not supposed to happen\n");
#endif
dp->i_size = ulr->ulr_offset + ulr->ulr_count;
DIP_ASSIGN(dp, size, dp->i_size);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Get the block containing the space for the new directory entry.
*/
error = ulfs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp, true);
if (error) {
return (error);
}
/*
* Find space for the new entry. In the simple case, the entry at
* offset base will have the space. If it does not, then namei
* arranged that compacting the region ulr_offset to
* ulr_offset + ulr_count would yield the space.
*/
ep = (struct lfs_direct *)dirbuf;
dsize = (ep->d_ino != 0) ? LFS_DIRSIZ(FSFMT(dvp), ep, needswap) : 0;
spacefree = ulfs_rw16(ep->d_reclen, needswap) - dsize;
for (loc = ulfs_rw16(ep->d_reclen, needswap); loc < ulr->ulr_count; ) {
uint16_t reclen;
nep = (struct lfs_direct *)(dirbuf + loc);
/* Trim the existing slot (NB: dsize may be zero). */
ep->d_reclen = ulfs_rw16(dsize, needswap);
ep = (struct lfs_direct *)((char *)ep + dsize);
reclen = ulfs_rw16(nep->d_reclen, needswap);
loc += reclen;
if (nep->d_ino == 0) {
/*
* A mid-block unused entry. Such entries are
* never created by the kernel, but fsck_ffs
* can create them (and it doesn't fix them).
*
* Add up the free space, and initialise the
* relocated entry since we don't memcpy it.
*/
spacefree += reclen;
ep->d_ino = 0;
dsize = 0;
continue;
}
dsize = LFS_DIRSIZ(FSFMT(dvp), nep, needswap);
spacefree += reclen - dsize;
#ifdef LFS_DIRHASH
if (dp->i_dirhash != NULL)
ulfsdirhash_move(dp, nep,
ulr->ulr_offset + ((char *)nep - dirbuf),
ulr->ulr_offset + ((char *)ep - dirbuf));
#endif
memcpy((void *)ep, (void *)nep, dsize);
}
/*
* Here, `ep' points to a directory entry containing `dsize' in-use
* bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
* then the entry is completely unused (dsize == 0). The value
* of ep->d_reclen is always indeterminate.
*
* Update the pointer fields in the previous entry (if any),
* copy in the new entry, and write out the block.
*/
if (ep->d_ino == 0 ||
(ulfs_rw32(ep->d_ino, needswap) == ULFS_WINO &&
memcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
if (spacefree + dsize < newentrysize)
panic("ulfs_direnter: compact1");
dirp->d_reclen = spacefree + dsize;
} else {
if (spacefree < newentrysize)
panic("ulfs_direnter: compact2");
dirp->d_reclen = spacefree;
ep->d_reclen = ulfs_rw16(dsize, needswap);
ep = (struct lfs_direct *)((char *)ep + dsize);
}
dirp->d_reclen = ulfs_rw16(dirp->d_reclen, needswap);
dirp->d_ino = ulfs_rw32(dirp->d_ino, needswap);
if (FSFMT(dvp)) {
#if (BYTE_ORDER == LITTLE_ENDIAN)
if (needswap == 0) {
#else
if (needswap != 0) {
#endif
u_char tmp = dirp->d_namlen;
dirp->d_namlen = dirp->d_type;
dirp->d_type = tmp;
}
}
#ifdef LFS_DIRHASH
if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
dirp->d_reclen == spacefree))
ulfsdirhash_add(dp, dirp, ulr->ulr_offset + ((char *)ep - dirbuf));
#endif
memcpy((void *)ep, (void *)dirp, (u_int)newentrysize);
#ifdef LFS_DIRHASH
if (dp->i_dirhash != NULL)
ulfsdirhash_checkblock(dp, dirbuf -
(ulr->ulr_offset & (dirblksiz - 1)),
ulr->ulr_offset & ~(dirblksiz - 1));
#endif
error = VOP_BWRITE(bp->b_vp, bp);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* If all went well, and the directory can be shortened, proceed
* with the truncation. Note that we have to unlock the inode for
* the entry that we just entered, as the truncation may need to
* lock other inodes which can lead to deadlock if we also hold a
* lock on the newly entered node.
*/
if (error == 0 && ulr->ulr_endoff && ulr->ulr_endoff < dp->i_size) {
#ifdef LFS_DIRHASH
if (dp->i_dirhash != NULL)
ulfsdirhash_dirtrunc(dp, ulr->ulr_endoff);
#endif
(void) lfs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC, cr);
}
return (error);
}
/*
* Remove a directory entry after a call to namei, using the
* parameters that ulfs_lookup left in nameidata and in the
* ulfs_lookup_results.
*
* DVP is the directory to be updated. It must be locked.
* ULR is the ulfs_lookup_results structure from the final lookup step.
* IP, if not null, is the inode being unlinked.
* FLAGS may contain DOWHITEOUT.
* ISRMDIR is not used and (XXX) should be removed.
*
* If FLAGS contains DOWHITEOUT the entry is replaced with a whiteout
* instead of being cleared.
*
* ulr->ulr_offset contains the position of the directory entry
* to be removed.
*
* ulr->ulr_reclen contains the size of the directory entry to be
* removed.
*
* ulr->ulr_count contains the size of the *previous* directory
* entry. This allows finding it, for free space management. If
* ulr_count is 0, the target entry is at the beginning of the
* directory. (Does this ever happen? The first entry should be ".",
* which should only be removed at rmdir time. Does rmdir come here
* to clear out the "." and ".." entries? Perhaps, but I doubt it.)
*
* The space is marked free by adding it to the record length (not
* name length) of the preceding entry. If the first entry becomes
* free, it is marked free by setting the inode number to 0.
*
* The link count of IP is decremented. Note that this is not the
* inverse behavior of ulfs_direnter, which does not adjust link
* counts. Sigh.
*/
int
ulfs_dirremove(struct vnode *dvp, const struct ulfs_lookup_results *ulr,
struct inode *ip, int flags, int isrmdir)
{
struct inode *dp = VTOI(dvp);
struct lfs_direct *ep;
struct buf *bp;
int error;
const int needswap = ULFS_MPNEEDSWAP(dp->i_lfs);
if (flags & DOWHITEOUT) {
/*
* Whiteout entry: set d_ino to ULFS_WINO.
*/
error = ulfs_blkatoff(dvp, (off_t)ulr->ulr_offset, (void *)&ep,
&bp, true);
if (error)
return (error);
ep->d_ino = ulfs_rw32(ULFS_WINO, needswap);
ep->d_type = LFS_DT_WHT;
goto out;
}
if ((error = ulfs_blkatoff(dvp,
(off_t)(ulr->ulr_offset - ulr->ulr_count), (void *)&ep, &bp, true)) != 0)
return (error);
#ifdef LFS_DIRHASH
/*
* Remove the dirhash entry. This is complicated by the fact
* that `ep' is the previous entry when ulr_count != 0.
*/
if (dp->i_dirhash != NULL)
ulfsdirhash_remove(dp, (ulr->ulr_count == 0) ? ep :
(struct lfs_direct *)((char *)ep +
ulfs_rw16(ep->d_reclen, needswap)), ulr->ulr_offset);
#endif
if (ulr->ulr_count == 0) {
/*
* First entry in block: set d_ino to zero.
*/
ep->d_ino = 0;
} else {
/*
* Collapse new free space into previous entry.
*/
ep->d_reclen =
ulfs_rw16(ulfs_rw16(ep->d_reclen, needswap) + ulr->ulr_reclen,
needswap);
}
#ifdef LFS_DIRHASH
if (dp->i_dirhash != NULL) {
int dirblksiz = ip->i_lfs->um_dirblksiz;
ulfsdirhash_checkblock(dp, (char *)ep -
((ulr->ulr_offset - ulr->ulr_count) & (dirblksiz - 1)),
ulr->ulr_offset & ~(dirblksiz - 1));
}
#endif
out:
if (ip) {
ip->i_nlink--;
DIP_ASSIGN(ip, nlink, ip->i_nlink);
ip->i_flag |= IN_CHANGE;
}
/*
* XXX did it ever occur to anyone that it might be a good
* idea to restore ip->i_nlink if this fails? Or something?
* Currently on error return from this function the state of
* ip->i_nlink depends on what happened, and callers
* definitely do not take this into account.
*/
error = VOP_BWRITE(bp->b_vp, bp);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* If the last named reference to a snapshot goes away,
* drop its snapshot reference so that it will be reclaimed
* when last open reference goes away.
*/
if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 &&
ip->i_nlink == 0)
ulfs_snapgone(ip);
return (error);
}
/*
* Rewrite an existing directory entry to point at the inode supplied.
*
* DP is the directory to update.
* OFFSET is the position of the entry in question. It may come
* from ulr_offset of a ulfs_lookup_results.
* OIP is the old inode the directory previously pointed to.
* NEWINUM is the number of the new inode.
* NEWTYPE is the new value for the type field of the directory entry.
* (This is ignored if the fs doesn't support that.)
* ISRMDIR is not used and (XXX) should be removed.
* IFLAGS are added to DP's inode flags.
*
* The link count of OIP is decremented. Note that the link count of
* the new inode is *not* incremented. Yay for symmetry.
*/
int
ulfs_dirrewrite(struct inode *dp, off_t offset,
struct inode *oip, ino_t newinum, int newtype,
int isrmdir, int iflags)
{
struct buf *bp;
struct lfs_direct *ep;
struct vnode *vdp = ITOV(dp);
int error;
error = ulfs_blkatoff(vdp, offset, (void *)&ep, &bp, true);
if (error)
return (error);
ep->d_ino = ulfs_rw32(newinum, ULFS_IPNEEDSWAP(dp));
if (!FSFMT(vdp))
ep->d_type = newtype;
oip->i_nlink--;
DIP_ASSIGN(oip, nlink, oip->i_nlink);
oip->i_flag |= IN_CHANGE;
error = VOP_BWRITE(bp->b_vp, bp);
dp->i_flag |= iflags;
/*
* If the last named reference to a snapshot goes away,
* drop its snapshot reference so that it will be reclaimed
* when last open reference goes away.
*/
if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_nlink == 0)
ulfs_snapgone(oip);
return (error);
}
/*
* Check if a directory is empty or not.
* Inode supplied must be locked.
*
* Using a struct lfs_dirtemplate here is not precisely
* what we want, but better than using a struct lfs_direct.
*
* NB: does not handle corrupted directories.
*/
int
ulfs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
{
doff_t off;
struct lfs_dirtemplate dbuf;
struct lfs_direct *dp = (struct lfs_direct *)&dbuf;
int error, namlen;
size_t count;
const int needswap = ULFS_IPNEEDSWAP(ip);
#define MINDIRSIZ (sizeof (struct lfs_dirtemplate) / 2)
for (off = 0; off < ip->i_size;
off += ulfs_rw16(dp->d_reclen, needswap)) {
error = vn_rdwr(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ, off,
UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
/*
* Since we read MINDIRSIZ, residual must
* be 0 unless we're at end of file.
*/
if (error || count != 0)
return (0);
/* avoid infinite loops */
if (dp->d_reclen == 0)
return (0);
/* skip empty entries */
if (dp->d_ino == 0 || ulfs_rw32(dp->d_ino, needswap) == ULFS_WINO)
continue;
/* accept only "." and ".." */
#if (BYTE_ORDER == LITTLE_ENDIAN)
if (FSFMT(ITOV(ip)) && needswap == 0)
namlen = dp->d_type;
else
namlen = dp->d_namlen;
#else
if (FSFMT(ITOV(ip)) && needswap != 0)
namlen = dp->d_type;
else
namlen = dp->d_namlen;
#endif
if (namlen > 2)
return (0);
if (dp->d_name[0] != '.')
return (0);
/*
* At this point namlen must be 1 or 2.
* 1 implies ".", 2 implies ".." if second
* char is also "."
*/
if (namlen == 1 &&
ulfs_rw32(dp->d_ino, needswap) == ip->i_number)
continue;
if (dp->d_name[1] == '.' &&
ulfs_rw32(dp->d_ino, needswap) == parentino)
continue;
return (0);
}
return (1);
}
#define ULFS_DIRRABLKS 0
int ulfs_dirrablks = ULFS_DIRRABLKS;
/*
* ulfs_blkatoff: Return buffer with the contents of block "offset" from
* the beginning of directory "vp". If "res" is non-NULL, fill it in with
* a pointer to the remaining space in the directory. If the caller intends
* to modify the buffer returned, "modify" must be true.
*/
int
ulfs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp,
bool modify)
{
struct inode *ip __diagused;
struct buf *bp;
daddr_t lbn;
const int dirrablks = ulfs_dirrablks;
daddr_t *blks;
int *blksizes;
int run, error;
struct mount *mp = vp->v_mount;
const int bshift = mp->mnt_fs_bshift;
const int bsize = 1 << bshift;
off_t eof;
blks = kmem_alloc((1 + dirrablks) * sizeof(daddr_t), KM_SLEEP);
blksizes = kmem_alloc((1 + dirrablks) * sizeof(int), KM_SLEEP);
ip = VTOI(vp);
KASSERT(vp->v_size == ip->i_size);
GOP_SIZE(vp, vp->v_size, &eof, 0);
lbn = offset >> bshift;
for (run = 0; run <= dirrablks;) {
const off_t curoff = lbn << bshift;
const int size = MIN(eof - curoff, bsize);
if (size == 0) {
break;
}
KASSERT(curoff < eof);
blks[run] = lbn;
blksizes[run] = size;
lbn++;
run++;
if (size != bsize) {
break;
}
}
KASSERT(run >= 1);
error = breadn(vp, blks[0], blksizes[0], &blks[1], &blksizes[1],
run - 1, NOCRED, (modify ? B_MODIFY : 0), &bp);
if (error != 0) {
*bpp = NULL;
goto out;
}
if (res) {
*res = (char *)bp->b_data + (offset & (bsize - 1));
}
*bpp = bp;
out:
kmem_free(blks, (1 + dirrablks) * sizeof(daddr_t));
kmem_free(blksizes, (1 + dirrablks) * sizeof(int));
return error;
}
/*
* This is a slightly strangely structured routine. It always puts
* all the pages for a vnode. It starts by releasing pages which
* are clean and simultaneously looks up the smallest offset for a
* dirty page beloning to the object. If there is no smallest offset,
* all pages have been cleaned. Otherwise, it finds a contiguous range
* of dirty pages starting from the smallest offset and writes them out.
* After this the scan is restarted.
*/
int
genfs_do_putpages(struct vnode *vp, off_t startoff, off_t endoff, int flags,
struct vm_page **busypg)
{
char databuf[MAXPHYS];
struct uvm_object *uobj = &vp->v_uobj;
struct vm_page *pg, *pg_next;
voff_t smallest;
voff_t curoff, bufoff;
off_t eof;
size_t xfersize;
int bshift = vp->v_mount->mnt_fs_bshift;
int bsize = 1 << bshift;
#if 0
int async = (flags & PGO_SYNCIO) == 0;
#else
int async = 0;
#endif
restart:
/* check if all pages are clean */
smallest = -1;
for (pg = TAILQ_FIRST(&uobj->memq); pg; pg = pg_next) {
pg_next = TAILQ_NEXT(pg, listq.queue);
/*
* XXX: this is not correct at all. But it's based on
* assumptions we can make when accessing the pages
* only through the file system and not through the
* virtual memory subsystem. Well, at least I hope
* so ;)
*/
KASSERT((pg->flags & PG_BUSY) == 0);
/* If we can just dump the page, do so */
if (pg->flags & PG_CLEAN || flags & PGO_FREE) {
uvm_pagefree(pg);
continue;
}
if (pg->offset < smallest || smallest == -1)
smallest = pg->offset;
}
/* all done? */
if (TAILQ_EMPTY(&uobj->memq)) {
vp->v_iflag &= ~VI_ONWORKLST;
mutex_exit(&uobj->vmobjlock);
return 0;
}
/* we need to flush */
GOP_SIZE(vp, vp->v_writesize, &eof, 0);
for (curoff = smallest; curoff < eof; curoff += PAGE_SIZE) {
void *curva;
if (curoff - smallest >= MAXPHYS)
break;
pg = uvm_pagelookup(uobj, curoff);
if (pg == NULL)
break;
/* XXX: see comment about above KASSERT */
KASSERT((pg->flags & PG_BUSY) == 0);
curva = databuf + (curoff-smallest);
memcpy(curva, (void *)pg->uanon, PAGE_SIZE);
rumpvm_enterva((vaddr_t)curva, pg);
pg->flags |= PG_CLEAN;
}
KASSERT(curoff > smallest);
mutex_exit(&uobj->vmobjlock);
/* then we write */
for (bufoff = 0; bufoff < MIN(curoff-smallest,eof); bufoff+=xfersize) {
struct buf *bp;
struct vnode *devvp;
daddr_t bn, lbn;
int run, error;
lbn = (smallest + bufoff) >> bshift;
error = VOP_BMAP(vp, lbn, &devvp, &bn, &run);
if (error)
panic("%s: VOP_BMAP failed: %d", __func__, error);
xfersize = MIN(((lbn+1+run) << bshift) - (smallest+bufoff),
curoff - (smallest+bufoff));
/*
* We might run across blocks which aren't allocated yet.
* A reason might be e.g. the write operation being still
* in the kernel page cache while truncate has already
* enlarged the file. So just ignore those ranges.
*/
if (bn == -1)
continue;
bp = getiobuf(vp, true);
/* only write max what we are allowed to write */
bp->b_bcount = xfersize;
if (smallest + bufoff + xfersize > eof)
bp->b_bcount -= (smallest+bufoff+xfersize) - eof;
bp->b_bcount = (bp->b_bcount + DEV_BSIZE-1) & ~(DEV_BSIZE-1);
KASSERT(bp->b_bcount > 0);
KASSERT(smallest >= 0);
DPRINTF(("putpages writing from %x to %x (vp size %x)\n",
(int)(smallest + bufoff),
(int)(smallest + bufoff + bp->b_bcount),
(int)eof));
bp->b_bufsize = round_page(bp->b_bcount);
bp->b_lblkno = 0;
bp->b_blkno = bn + (((smallest+bufoff)&(bsize-1))>>DEV_BSHIFT);
bp->b_data = databuf + bufoff;
bp->b_flags = B_WRITE;
bp->b_cflags |= BC_BUSY;
if (async) {
bp->b_flags |= B_ASYNC;
bp->b_iodone = uvm_aio_biodone;
}
vp->v_numoutput++;
VOP_STRATEGY(devvp, bp);
if (bp->b_error)
panic("%s: VOP_STRATEGY lazy bum %d",
__func__, bp->b_error);
if (!async)
putiobuf(bp);
}
rumpvm_flushva();
mutex_enter(&uobj->vmobjlock);
goto restart;
}
int
ulfs_balloc_range(struct vnode *vp, off_t off, off_t len, kauth_cred_t cred,
int flags)
{
off_t neweof; /* file size after the operation */
off_t neweob; /* offset next to the last block after the operation */
off_t pagestart; /* starting offset of range covered by pgs */
off_t eob; /* offset next to allocated blocks */
struct uvm_object *uobj;
int i, delta, error, npages;
int bshift = vp->v_mount->mnt_fs_bshift;
int bsize = 1 << bshift;
int ppb = MAX(bsize >> PAGE_SHIFT, 1);
struct vm_page **pgs;
size_t pgssize;
UVMHIST_FUNC("ulfs_balloc_range"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "vp %p off 0x%x len 0x%x u_size 0x%x",
vp, off, len, vp->v_size);
neweof = MAX(vp->v_size, off + len);
GOP_SIZE(vp, neweof, &neweob, 0);
error = 0;
uobj = &vp->v_uobj;
/*
* read or create pages covering the range of the allocation and
* keep them locked until the new block is allocated, so there
* will be no window where the old contents of the new block are
* visible to racing threads.
*/
pagestart = trunc_page(off) & ~(bsize - 1);
npages = MIN(ppb, (round_page(neweob) - pagestart) >> PAGE_SHIFT);
pgssize = npages * sizeof(struct vm_page *);
pgs = kmem_zalloc(pgssize, KM_SLEEP);
/*
* adjust off to be block-aligned.
*/
delta = off & (bsize - 1);
off -= delta;
len += delta;
genfs_node_wrlock(vp);
mutex_enter(uobj->vmobjlock);
error = VOP_GETPAGES(vp, pagestart, pgs, &npages, 0,
VM_PROT_WRITE, 0, PGO_SYNCIO | PGO_PASTEOF | PGO_NOBLOCKALLOC |
PGO_NOTIMESTAMP | PGO_GLOCKHELD);
if (error) {
goto out;
}
/*
* now allocate the range.
*/
error = GOP_ALLOC(vp, off, len, flags, cred);
genfs_node_unlock(vp);
/*
* if the allocation succeeded, clear PG_CLEAN on all the pages
* and clear PG_RDONLY on any pages that are now fully backed
* by disk blocks. if the allocation failed, we do not invalidate
* the pages since they might have already existed and been dirty,
* in which case we need to keep them around. if we created the pages,
* they will be clean and read-only, and leaving such pages
* in the cache won't cause any problems.
*/
GOP_SIZE(vp, off + len, &eob, 0);
mutex_enter(uobj->vmobjlock);
mutex_enter(&uvm_pageqlock);
for (i = 0; i < npages; i++) {
KASSERT((pgs[i]->flags & PG_RELEASED) == 0);
if (!error) {
if (off <= pagestart + (i << PAGE_SHIFT) &&
pagestart + ((i + 1) << PAGE_SHIFT) <= eob) {
pgs[i]->flags &= ~PG_RDONLY;
}
pgs[i]->flags &= ~PG_CLEAN;
}
uvm_pageactivate(pgs[i]);
}
mutex_exit(&uvm_pageqlock);
uvm_page_unbusy(pgs, npages);
mutex_exit(uobj->vmobjlock);
out:
kmem_free(pgs, pgssize);
return error;
}
