hsfs - zygaena/hammerhead

root / base / usr / src / uts / common / fs / hsfs
hsfs Plain Text 7531 lines 190.9 KB
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
 */

/*
 * Directory operations for High Sierra filesystem
 */

#include <sys/types.h>
#include <sys/t_lock.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/cred.h>
#include <sys/user.h>
#include <sys/vfs.h>
#include <sys/stat.h>
#include <sys/vnode.h>
#include <sys/mode.h>
#include <sys/dnlc.h>
#include <sys/cmn_err.h>
#include <sys/fbuf.h>
#include <sys/kmem.h>
#include <sys/policy.h>
#include <sys/sunddi.h>
#include <vm/hat.h>
#include <vm/as.h>
#include <vm/pvn.h>
#include <vm/seg.h>
#include <vm/seg_map.h>
#include <vm/seg_kmem.h>
#include <vm/page.h>

#include <sys/fs/hsfs_spec.h>
#include <sys/fs/hsfs_isospec.h>
#include <sys/fs/hsfs_node.h>
#include <sys/fs/hsfs_impl.h>
#include <sys/fs/hsfs_susp.h>
#include <sys/fs/hsfs_rrip.h>

#include <sys/sysinfo.h>
#include <sys/sysmacros.h>
#include <sys/errno.h>
#include <sys/debug.h>
#include <fs/fs_subr.h>

/*
 * This macro expects a name that ends in '.' and returns TRUE if the
 * name is not "." or ".."
 */
#define	CAN_TRUNCATE_DOT(name, namelen)	\
		(namelen > 1 && (namelen > 2 || name[0] != '.'))

enum dirblock_result { FOUND_ENTRY, WENT_PAST, HIT_END };

/*
 * These values determine whether we will try to read a file or dir;
 * they may be patched via /etc/system to allow users to read
 * record-oriented files.
 */
int ide_prohibited = IDE_PROHIBITED;
int hde_prohibited = HDE_PROHIBITED;

/*
 * This variable determines if the HSFS code will use the
 * directory name lookup cache. The default is for the cache to be used.
 */
static int hsfs_use_dnlc = 1;

/*
 * This variable determines whether strict ISO-9660 directory ordering
 * is to be assumed.  If false (which it is by default), then when
 * searching a directory of an ISO-9660 disk, we do not expect the
 * entries to be sorted (as the spec requires), and so cannot terminate
 * the search early.  Unfortunately, some vendors are producing
 * non-compliant disks.  This variable exists to revert to the old
 * behavior in case someone relies on this. This option is expected to be
 * removed at some point in the future.
 *
 * Use "set hsfs:strict_iso9660_ordering = 1" in /etc/system to override.
 */
static int strict_iso9660_ordering = 0;

/*
 * This tunable allows us to ignore inode numbers from rrip-1.12.
 * In this case, we fall back to our default inode algorithm.
 */
int use_rrip_inodes = 1;

static void hs_hsnode_cache_reclaim(void *unused);
static void hs_addfreeb(struct hsfs *fsp, struct hsnode *hp);
static enum dirblock_result process_dirblock(struct fbuf *fbp, uint_t *offset,
	uint_t last_offset, char *nm, int nmlen, struct hsfs *fsp,
	struct hsnode *dhp, struct vnode *dvp, struct vnode **vpp,
	int *error);
static int strip_trailing(struct hsfs *fsp, char *nm, int len);
static int hs_namelen(struct hsfs *fsp, char *nm, int len);
static int uppercase_cp(char *from, char *to, int size);
static void hs_log_bogus_joliet_warning(void);
static int hs_iso_copy(char *from, char *to, int size);
static int32_t hs_ucs2_2_utf8(uint16_t c_16, uint8_t *s_8);
static int hs_utf8_trunc(uint8_t *str, int len);

/*
 * hs_access
 * Return 0 if the desired access may be granted.
 * Otherwise return error code.
 */
int
hs_access(struct vnode *vp, mode_t m, struct cred *cred)
{
	struct hsnode *hp;
	int	shift = 0;

	/*
	 * Write access cannot be granted for a read-only medium
	 */
	if ((m & VWRITE) && !IS_DEVVP(vp))
		return (EROFS);

	hp = VTOH(vp);

	/*
	 * XXX - For now, use volume protections.
	 *  Also, always grant EXEC access for directories
	 *  if READ access is granted.
	 */
	if ((vp->v_type == VDIR) && (m & VEXEC)) {
		m &= ~VEXEC;
		m |= VREAD;
	}

	if (crgetuid(cred) != hp->hs_dirent.uid) {
		shift += 3;
		if (!groupmember((uid_t)hp->hs_dirent.gid, cred))
			shift += 3;
	}
	return (secpolicy_vnode_access2(cred, vp, hp->hs_dirent.uid,
	    hp->hs_dirent.mode << shift, m));
}

#if ((HS_HASHSIZE & (HS_HASHSIZE - 1)) == 0)
#define	HS_HASH(l)	((uint_t)(l) & (HS_HASHSIZE - 1))
#else
#define	HS_HASH(l)	((uint_t)(l) % HS_HASHSIZE)
#endif
#define	HS_HPASH(hp)	HS_HASH((hp)->hs_nodeid)

/*
 * The tunable nhsnode is now a threshold for a dynamically allocated
 * pool of hsnodes, not the size of a statically allocated table.
 * When the number of hsnodes for a particular file system exceeds
 * nhsnode, the allocate and free logic will try to reduce the number
 * of allocated nodes by returning unreferenced nodes to the kmem_cache
 * instead of putting them on the file system's private free list.
 */
int nhsnode = HS_HSNODESPACE / sizeof (struct hsnode);

struct kmem_cache *hsnode_cache;  /* free hsnode cache */

/*
 * Initialize the cache of free hsnodes.
 */
void
hs_init_hsnode_cache(void)
{
	/*
	 * A kmem_cache is used for the hsnodes
	 * No constructor because hsnodes are initialised by bzeroing.
	 */
	hsnode_cache = kmem_cache_create("hsfs_hsnode_cache",
	    sizeof (struct hsnode), 0, NULL,
	    NULL, hs_hsnode_cache_reclaim, NULL, NULL, 0);
}

/*
 * Destroy the cache of free hsnodes.
 */
void
hs_fini_hsnode_cache(void)
{
	kmem_cache_destroy(hsnode_cache);
}

/*
 * System is short on memory, free up as much as possible
 */
/*ARGSUSED*/
static void
hs_hsnode_cache_reclaim(void *unused)
{
	struct hsfs *fsp;
	struct hsnode *hp;

	/*
	 * For each vfs in the hs_mounttab list
	 */
	mutex_enter(&hs_mounttab_lock);
	for (fsp = hs_mounttab; fsp != NULL; fsp = fsp->hsfs_next) {
		/*
		 * Purge the dnlc of all hsfs entries
		 */
		(void) dnlc_purge_vfsp(fsp->hsfs_vfs, 0);

		/*
		 * For each entry in the free chain
		 */
		rw_enter(&fsp->hsfs_hash_lock, RW_WRITER);
		mutex_enter(&fsp->hsfs_free_lock);
		for (hp = fsp->hsfs_free_f; hp != NULL; hp = fsp->hsfs_free_f) {
			/*
			 * Remove from chain
			 */
			fsp->hsfs_free_f = hp->hs_freef;
			if (fsp->hsfs_free_f != NULL) {
				fsp->hsfs_free_f->hs_freeb = NULL;
			} else {
				fsp->hsfs_free_b = NULL;
			}
			/*
			 * Free the node. Force it to be fully freed
			 * by setting the 3rd arg (nopage) to 1.
			 */
			hs_freenode(HTOV(hp), fsp, 1);
		}
		mutex_exit(&fsp->hsfs_free_lock);
		rw_exit(&fsp->hsfs_hash_lock);
	}
	mutex_exit(&hs_mounttab_lock);
}

/*
 * Add an hsnode to the end of the free list.
 */
static void
hs_addfreeb(struct hsfs *fsp, struct hsnode *hp)
{
	struct hsnode *ep;

	vn_invalid(HTOV(hp));
	mutex_enter(&fsp->hsfs_free_lock);
	ep = fsp->hsfs_free_b;
	fsp->hsfs_free_b = hp;		/* hp is the last entry in free list */
	hp->hs_freef = NULL;
	hp->hs_freeb = ep;		/* point at previous last entry */
	if (ep == NULL)
		fsp->hsfs_free_f = hp;	/* hp is only entry in free list */
	else
		ep->hs_freef = hp;	/* point previous last entry at hp */

	mutex_exit(&fsp->hsfs_free_lock);
}

/*
 * Get an hsnode from the front of the free list.
 * Must be called with write hsfs_hash_lock held.
 */
static struct hsnode *
hs_getfree(struct hsfs *fsp)
{
	struct hsnode *hp, **tp;

	ASSERT(RW_WRITE_HELD(&fsp->hsfs_hash_lock));

	/*
	 * If the number of currently-allocated hsnodes is less than
	 * the hsnode count threshold (nhsnode), or if there are no
	 * nodes on the file system's local free list (which acts as a
	 * cache), call kmem_cache_alloc to get a new hsnode from
	 * kernel memory.
	 */
	mutex_enter(&fsp->hsfs_free_lock);
	if ((fsp->hsfs_nohsnode < nhsnode) || (fsp->hsfs_free_f == NULL)) {
		mutex_exit(&fsp->hsfs_free_lock);
		hp = kmem_cache_alloc(hsnode_cache, KM_SLEEP);
		fsp->hsfs_nohsnode++;
		bzero((caddr_t)hp, sizeof (*hp));
		hp->hs_vnode = vn_alloc(KM_SLEEP);
		return (hp);
	}
	hp = fsp->hsfs_free_f;
	/* hp cannot be NULL, since we already checked this above */
	fsp->hsfs_free_f = hp->hs_freef;
	if (fsp->hsfs_free_f != NULL)
		fsp->hsfs_free_f->hs_freeb = NULL;
	else
		fsp->hsfs_free_b = NULL;
	mutex_exit(&fsp->hsfs_free_lock);

	for (tp = &fsp->hsfs_hash[HS_HPASH(hp)]; *tp != NULL;
	    tp = &(*tp)->hs_hash) {
		if (*tp == hp) {
			struct vnode *vp;

			vp = HTOV(hp);

			/*
			 * file is no longer referenced, destroy all old pages
			 */
			if (vn_has_cached_data(vp))
				/*
				 * pvn_vplist_dirty will abort all old pages
				 */
				(void) pvn_vplist_dirty(vp, (u_offset_t)0,
				    hsfs_putapage, B_INVAL,
				    (struct cred *)NULL);
			*tp = hp->hs_hash;
			break;
		}
	}
	if (hp->hs_dirent.sym_link != (char *)NULL) {
		kmem_free(hp->hs_dirent.sym_link,
		    (size_t)(hp->hs_dirent.ext_size + 1));
	}

	mutex_destroy(&hp->hs_contents_lock);
	{
		vnode_t	*vp;

		vp = hp->hs_vnode;
		bzero((caddr_t)hp, sizeof (*hp));
		hp->hs_vnode = vp;
		vn_reinit(vp);
	}
	return (hp);
}

/*
 * Remove an hsnode from the free list.
 */
static void
hs_remfree(struct hsfs *fsp, struct hsnode *hp)
{
	mutex_enter(&fsp->hsfs_free_lock);
	if (hp->hs_freef != NULL)
		hp->hs_freef->hs_freeb = hp->hs_freeb;
	else
		fsp->hsfs_free_b = hp->hs_freeb;
	if (hp->hs_freeb != NULL)
		hp->hs_freeb->hs_freef = hp->hs_freef;
	else
		fsp->hsfs_free_f = hp->hs_freef;
	mutex_exit(&fsp->hsfs_free_lock);
}

/*
 * Look for hsnode in hash list.
 * If the inode number is != HS_DUMMY_INO (16), then only the inode
 * number is used for the check.
 * If the inode number is == HS_DUMMY_INO, we additionally always
 * check the directory offset for the file to avoid caching the
 * meta data for all zero sized to the first zero sized file that
 * was touched.
 *
 * If found, reactivate it if inactive.
 *
 * Must be entered with hsfs_hash_lock held.
 */
struct vnode *
hs_findhash(ino64_t nodeid, uint_t lbn, uint_t off, struct vfs *vfsp)
{
	struct hsnode *tp;
	struct hsfs *fsp;

	fsp = VFS_TO_HSFS(vfsp);

	ASSERT(RW_LOCK_HELD(&fsp->hsfs_hash_lock));

	for (tp = fsp->hsfs_hash[HS_HASH(nodeid)]; tp != NULL;
	    tp = tp->hs_hash) {
		if (tp->hs_nodeid == nodeid) {
			struct vnode *vp;

			if (nodeid == HS_DUMMY_INO) {
				/*
				 * If this is the dummy inode number, look for
				 * matching dir_lbn and dir_off.
				 */
				for (; tp != NULL; tp = tp->hs_hash) {
					if (tp->hs_nodeid == nodeid &&
					    tp->hs_dir_lbn == lbn &&
					    tp->hs_dir_off == off)
						break;
				}
				if (tp == NULL)
					return (NULL);
			}

			mutex_enter(&tp->hs_contents_lock);
			vp = HTOV(tp);
			VN_HOLD(vp);
			if ((tp->hs_flags & HREF) == 0) {
				tp->hs_flags |= HREF;
				/*
				 * reactivating a free hsnode:
				 * remove from free list
				 */
				hs_remfree(fsp, tp);
			}
			mutex_exit(&tp->hs_contents_lock);
			return (vp);
		}
	}
	return (NULL);
}

static void
hs_addhash(struct hsfs *fsp, struct hsnode *hp)
{
	ulong_t hashno;

	ASSERT(RW_WRITE_HELD(&fsp->hsfs_hash_lock));

	hashno = HS_HPASH(hp);
	hp->hs_hash = fsp->hsfs_hash[hashno];
	fsp->hsfs_hash[hashno] = hp;
}

/*
 * Destroy all old pages and free the hsnodes
 * Return 1 if busy (a hsnode is still referenced).
 */
int
hs_synchash(struct vfs *vfsp)
{
	struct hsfs *fsp;
	int i;
	struct hsnode *hp, *nhp;
	int busy = 0;
	struct vnode *vp, *rvp;

	fsp = VFS_TO_HSFS(vfsp);
	rvp = fsp->hsfs_rootvp;
	/* make sure no one can come in */
	rw_enter(&fsp->hsfs_hash_lock, RW_WRITER);
	for (i = 0; i < HS_HASHSIZE; i++) {
		for (hp = fsp->hsfs_hash[i]; hp != NULL; hp = hp->hs_hash) {
			vp = HTOV(hp);
			if ((hp->hs_flags & HREF) && (vp != rvp ||
			    (vp == rvp && vp->v_count > 1))) {
				busy = 1;
				continue;
			}
			if (vn_has_cached_data(vp))
				(void) pvn_vplist_dirty(vp, (u_offset_t)0,
				    hsfs_putapage, B_INVAL,
				    (struct cred *)NULL);
		}
	}
	if (busy) {
		rw_exit(&fsp->hsfs_hash_lock);
		return (1);
	}

	/* now free the hsnodes */
	for (i = 0; i < HS_HASHSIZE; i++) {
		for (hp = fsp->hsfs_hash[i]; hp != NULL; hp = nhp) {
			nhp = hp->hs_hash;
			/*
			 * We know there are no pages associated with
			 * all the hsnodes (they've all been released
			 * above). So remove from free list and
			 * free the entry with nopage set.
			 */
			vp = HTOV(hp);
			if (vp != rvp) {
				hs_remfree(fsp, hp);
				hs_freenode(vp, fsp, 1);
			}
		}
	}

	ASSERT(fsp->hsfs_nohsnode == 1);
	rw_exit(&fsp->hsfs_hash_lock);
	/* release the root hsnode, this should free the final hsnode */
	VN_RELE(rvp);

	return (0);
}

/*
 * hs_makenode
 *
 * Construct an hsnode.
 * Caller specifies the directory entry, the block number and offset
 * of the directory entry, and the vfs pointer.
 * note: off is the sector offset, not lbn offset
 * if NULL is returned implies file system hsnode table full
 */
struct vnode *
hs_makenode(
	struct hs_direntry *dp,
	uint_t lbn,
	uint_t off,
	struct vfs *vfsp)
{
	struct hsnode *hp;
	struct vnode *vp;
	struct hs_volume *hvp;
	struct vnode *newvp;
	struct hsfs *fsp;
	ino64_t nodeid;

	fsp = VFS_TO_HSFS(vfsp);

	/*
	 * Construct the data that allows us to re-read the meta data without
	 * knowing the name of the file: in the case of a directory
	 * entry, this should point to the canonical dirent, the "."
	 * directory entry for the directory.  This dirent is pointed
	 * to by all directory entries for that dir (including the ".")
	 * entry itself.
	 * In the case of a file, simply point to the dirent for that
	 * file (there are hard links in Rock Ridge, so we need to use
	 * different data to contruct the node id).
	 */
	if (dp->type == VDIR) {
		lbn = dp->ext_lbn;
		off = 0;
	}

	/*
	 * Normalize lbn and off before creating a nodeid
	 * and before storing them in a hs_node structure
	 */
	hvp = &fsp->hsfs_vol;
	lbn += off >> hvp->lbn_shift;
	off &= hvp->lbn_maxoffset;
	/*
	 * If the media carries rrip-v1.12 or newer, and we trust the inodes
	 * from the rrip data (use_rrip_inodes != 0), use that data. If the
	 * media has been created by a recent mkisofs version, we may trust
	 * all numbers in the starting extent number; otherwise, we cannot
	 * do this for zero sized files and symlinks, because if we did we'd
	 * end up mapping all of them to the same node.
	 * We use HS_DUMMY_INO in this case and make sure that we will not
	 * map all files to the same meta data.
	 */
	if (dp->inode != 0 && use_rrip_inodes) {
		nodeid = dp->inode;
	} else if ((dp->ext_size == 0 || dp->sym_link != (char *)NULL) &&
	    (fsp->hsfs_flags & HSFSMNT_INODE) == 0) {
		nodeid = HS_DUMMY_INO;
	} else {
		nodeid = dp->ext_lbn;
	}

	/* look for hsnode in cache first */

	rw_enter(&fsp->hsfs_hash_lock, RW_READER);

	if ((vp = hs_findhash(nodeid, lbn, off, vfsp)) == NULL) {

		/*
		 * Not in cache.  However, someone else may have come
		 * to the same conclusion and just put one in.	Upgrade
		 * our lock to a write lock and look again.
		 */
		rw_exit(&fsp->hsfs_hash_lock);
		rw_enter(&fsp->hsfs_hash_lock, RW_WRITER);

		if ((vp = hs_findhash(nodeid, lbn, off, vfsp)) == NULL) {
			/*
			 * Now we are really sure that the hsnode is not
			 * in the cache.  Get one off freelist or else
			 * allocate one. Either way get a bzeroed hsnode.
			 */
			hp = hs_getfree(fsp);

			bcopy((caddr_t)dp, (caddr_t)&hp->hs_dirent,
			    sizeof (*dp));
			/*
			 * We've just copied this pointer into hs_dirent,
			 * and don't want 2 references to same symlink.
			 */
			dp->sym_link = (char *)NULL;

			/*
			 * No need to hold any lock because hsnode is not
			 * yet in the hash chain.
			 */
			mutex_init(&hp->hs_contents_lock, NULL, MUTEX_DEFAULT,
			    NULL);
			hp->hs_dir_lbn = lbn;
			hp->hs_dir_off = off;
			hp->hs_nodeid = nodeid;
			hp->hs_seq = 0;
			hp->hs_prev_offset = 0;
			hp->hs_num_contig = 0;
			hp->hs_ra_bytes = 0;
			hp->hs_flags = HREF;
			if (off > HS_SECTOR_SIZE)
				cmn_err(CE_WARN, "hs_makenode: bad offset");

			vp = HTOV(hp);
			vp->v_vfsp = vfsp;
			vp->v_type = dp->type;
			vp->v_rdev = dp->r_dev;
			vn_setops(vp, hsfs_vnodeops);
			vp->v_data = (caddr_t)hp;
			vn_exists(vp);
			/*
			 * if it's a device, call specvp
			 */
			if (IS_DEVVP(vp)) {
				rw_exit(&fsp->hsfs_hash_lock);
				newvp = specvp(vp, vp->v_rdev, vp->v_type,
				    CRED());
				if (newvp == NULL)
					cmn_err(CE_NOTE,
					    "hs_makenode: specvp failed");
				VN_RELE(vp);
				return (newvp);
			}

			hs_addhash(fsp, hp);

		}
	}

	if (dp->sym_link != (char *)NULL) {
		kmem_free(dp->sym_link, (size_t)(dp->ext_size + 1));
		dp->sym_link = (char *)NULL;
	}

	rw_exit(&fsp->hsfs_hash_lock);
	return (vp);
}

/*
 * hs_freenode
 *
 * Deactivate an hsnode.
 * Leave it on the hash list but put it on the free list.
 * If the vnode does not have any pages, release the hsnode to the
 * kmem_cache using kmem_cache_free, else put in back of the free list.
 *
 * This function can be called with the hsfs_free_lock held, but only
 * when the code is guaranteed to go through the path where the
 * node is freed entirely, and not the path where the node could go back
 * on the free list (and where the free lock would need to be acquired).
 */
void
hs_freenode(vnode_t *vp, struct hsfs *fsp, int nopage)
{
	struct hsnode **tp;
	struct hsnode *hp = VTOH(vp);

	ASSERT(RW_LOCK_HELD(&fsp->hsfs_hash_lock));

	if (nopage || (fsp->hsfs_nohsnode >= nhsnode)) {
		/* remove this node from the hash list, if it's there */
		for (tp = &fsp->hsfs_hash[HS_HPASH(hp)]; *tp != NULL;
		    tp = &(*tp)->hs_hash) {

			if (*tp == hp) {
				*tp = hp->hs_hash;
				break;
			}
		}

		if (hp->hs_dirent.sym_link != (char *)NULL) {
			kmem_free(hp->hs_dirent.sym_link,
			    (size_t)(hp->hs_dirent.ext_size + 1));
			hp->hs_dirent.sym_link = NULL;
		}
		if (vn_has_cached_data(vp)) {
			/* clean all old pages */
			(void) pvn_vplist_dirty(vp, (u_offset_t)0,
			    hsfs_putapage, B_INVAL, (struct cred *)NULL);
			/* XXX - can we remove pages by fiat like this??? */
			vp->v_pages = NULL;
		}
		mutex_destroy(&hp->hs_contents_lock);
		vn_invalid(vp);
		vn_free(vp);
		kmem_cache_free(hsnode_cache, hp);
		fsp->hsfs_nohsnode--;
		return;
	}
	hs_addfreeb(fsp, hp); /* add to back of free list */
}

/*
 * hs_remakenode
 *
 * Reconstruct a vnode given the location of its directory entry.
 * Caller specifies the the block number and offset
 * of the directory entry, and the vfs pointer.
 * Returns an error code or 0.
 */
int
hs_remakenode(uint_t lbn, uint_t off, struct vfs *vfsp,
    struct vnode **vpp)
{
	struct buf *secbp;
	struct hsfs *fsp;
	uint_t secno;
	uchar_t *dirp;
	struct hs_direntry hd;
	int error;

	/* Convert to sector and offset */
	fsp = VFS_TO_HSFS(vfsp);
	if (off > HS_SECTOR_SIZE) {
		cmn_err(CE_WARN, "hs_remakenode: bad offset");
		error = EINVAL;
		goto end;
	}
	secno = LBN_TO_SEC(lbn, vfsp);
	secbp = bread(fsp->hsfs_devvp->v_rdev, secno * 4, HS_SECTOR_SIZE);

	error = geterror(secbp);
	if (error != 0) {
		cmn_err(CE_NOTE, "hs_remakenode: bread: error=(%d)", error);
		goto end;
	}

	dirp = (uchar_t *)secbp->b_un.b_addr;
	error = hs_parsedir(fsp, &dirp[off], &hd, (char *)NULL, (int *)NULL,
	    HS_SECTOR_SIZE - off);
	if (!error) {
		*vpp = hs_makenode(&hd, lbn, off, vfsp);
		if (*vpp == NULL)
			error = ENFILE;
	}

end:
	brelse(secbp);
	return (error);
}


/*
 * hs_dirlook
 *
 * Look for a given name in a given directory.
 * If found, construct an hsnode for it.
 */
int
hs_dirlook(
	struct vnode	*dvp,
	char		*name,
	int		namlen,		/* length of 'name' */
	struct vnode	**vpp,
	struct cred	*cred)
{
	struct hsnode *dhp;
	struct hsfs	*fsp;
	int		error = 0;
	uint_t		offset;		/* real offset in directory */
	uint_t		last_offset;	/* last index in directory */
	char		*cmpname;	/* case-folded name */
	int		cmpname_size;	/* how much memory we allocate for it */
	int		cmpnamelen;
	int		adhoc_search;	/* did we start at begin of dir? */
	int		end;
	uint_t		hsoffset;
	struct fbuf	*fbp;
	int		bytes_wanted;
	int		dirsiz;
	int		is_rrip;

	if (dvp->v_type != VDIR)
		return (ENOTDIR);

	if (error = hs_access(dvp, (mode_t)VEXEC, cred))
		return (error);

	if (hsfs_use_dnlc && (*vpp = dnlc_lookup(dvp, name)))
		return (0);

	dhp = VTOH(dvp);
	fsp = VFS_TO_HSFS(dvp->v_vfsp);
	is_rrip = IS_RRIP_IMPLEMENTED(fsp);

	/*
	 * name == "^A" is illegal for ISO-9660 and Joliet as '..' is '\1' on
	 * disk. It is no problem for Rock Ridge as RR uses '.' and '..'.
	 * XXX It could be OK for Joliet also (because namelen == 1 is
	 * XXX impossible for UCS-2) but then we need a better compare algorith.
	 */
	if (!is_rrip && *name == '\1' && namlen == 1)
		return (EINVAL);

	cmpname_size = (int)(fsp->hsfs_namemax + 1);
	cmpname = kmem_alloc((size_t)cmpname_size, KM_SLEEP);

	if (namlen >= cmpname_size)
		namlen = cmpname_size - 1;
	/*
	 * For the purposes of comparing the name against dir entries,
	 * fold it to upper case.
	 */
	if (is_rrip) {
		(void) strlcpy(cmpname, name, cmpname_size);
		cmpnamelen = namlen;
	} else {
		/*
		 * If we don't consider a trailing dot as part of the filename,
		 * remove it from the specified name
		 */
		if ((fsp->hsfs_flags & HSFSMNT_NOTRAILDOT) &&
		    name[namlen-1] == '.' &&
		    CAN_TRUNCATE_DOT(name, namlen))
			name[--namlen] = '\0';
		if (fsp->hsfs_vol_type == HS_VOL_TYPE_ISO_V2 ||
		    fsp->hsfs_vol_type == HS_VOL_TYPE_JOLIET) {
			cmpnamelen = hs_iso_copy(name, cmpname, namlen);
		} else {
			cmpnamelen = hs_uppercase_copy(name, cmpname, namlen);
		}
	}

	/* make sure dirent is filled up with all info */
	if (dhp->hs_dirent.ext_size == 0)
		hs_filldirent(dvp, &dhp->hs_dirent);

	/*
	 * No lock is needed - hs_offset is used as starting
	 * point for searching the directory.
	 */
	offset = dhp->hs_offset;
	hsoffset = offset;
	adhoc_search = (offset != 0);

	end = dhp->hs_dirent.ext_size;
	dirsiz = end;

tryagain:

	while (offset < end) {
		bytes_wanted = MIN(MAXBSIZE, dirsiz - (offset & MAXBMASK));

		error = fbread(dvp, (offset_t)(offset & MAXBMASK),
		    (unsigned int)bytes_wanted, S_READ, &fbp);
		if (error)
			goto done;

		last_offset = (offset & MAXBMASK) + fbp->fb_count;

		switch (process_dirblock(fbp, &offset, last_offset,
		    cmpname, cmpnamelen, fsp, dhp, dvp, vpp, &error)) {
		case FOUND_ENTRY:
			/* found an entry, either correct or not */
			goto done;

		case WENT_PAST:
			/*
			 * If we get here we know we didn't find it on the
			 * first pass. If adhoc_search, then we started a
			 * bit into the dir, and need to wrap around and
			 * search the first entries.  If not, then we started
			 * at the beginning and didn't find it.
			 */
			if (adhoc_search) {
				offset = 0;
				end = hsoffset;
				adhoc_search = 0;
				goto tryagain;
			}
			error = ENOENT;
			goto done;

		case HIT_END:
			goto tryagain;
		}
	}
	/*
	 * End of all dir blocks, didn't find entry.
	 */
	if (adhoc_search) {
		offset = 0;
		end = hsoffset;
		adhoc_search = 0;
		goto tryagain;
	}
	error = ENOENT;
done:
	/*
	 * If we found the entry, add it to the DNLC
	 * If the entry is a device file (assuming we support Rock Ridge),
	 * we enter the device vnode to the cache since that is what
	 * is in *vpp.
	 * That is ok since the CD-ROM is read-only, so (dvp,name) will
	 * always point to the same device.
	 */
	if (hsfs_use_dnlc && !error)
		dnlc_enter(dvp, name, *vpp);

	kmem_free(cmpname, (size_t)cmpname_size);

	return (error);
}

/*
 * hs_parsedir
 *
 * Parse a Directory Record into an hs_direntry structure.
 * High Sierra and ISO directory are almost the same
 * except the flag and date
 */
int
hs_parsedir(
	struct hsfs		*fsp,
	uchar_t			*dirp,
	struct hs_direntry	*hdp,
	char			*dnp,
	int			*dnlen,
	int			last_offset)	/* last offset in dirp */
{
	char	*on_disk_name;
	int	on_disk_namelen;
	int	on_disk_dirlen;
	uchar_t	flags;
	int	namelen;
	int	error;
	int	name_change_flag = 0;	/* set if name was gotten in SUA */

	hdp->ext_lbn = HDE_EXT_LBN(dirp);
	hdp->ext_size = HDE_EXT_SIZE(dirp);
	hdp->xar_len = HDE_XAR_LEN(dirp);
	hdp->intlf_sz = HDE_INTRLV_SIZE(dirp);
	hdp->intlf_sk = HDE_INTRLV_SKIP(dirp);
	hdp->sym_link = (char *)NULL;

	if (fsp->hsfs_vol_type == HS_VOL_TYPE_HS) {
		flags = HDE_FLAGS(dirp);
		hs_parse_dirdate(HDE_cdate(dirp), &hdp->cdate);
		hs_parse_dirdate(HDE_cdate(dirp), &hdp->adate);
		hs_parse_dirdate(HDE_cdate(dirp), &hdp->mdate);
		if ((flags & hde_prohibited) == 0) {
			/*
			 * Skip files with the associated bit set.
			 */
			if (flags & HDE_ASSOCIATED)
				return (EAGAIN);
			hdp->type = VREG;
			hdp->mode = HFREG;
			hdp->nlink = 1;
		} else if ((flags & hde_prohibited) == HDE_DIRECTORY) {
			hdp->type = VDIR;
			hdp->mode = HFDIR;
			hdp->nlink = 2;
		} else {
			hs_log_bogus_disk_warning(fsp,
			    HSFS_ERR_UNSUP_TYPE, flags);
			return (EINVAL);
		}
		hdp->uid = fsp -> hsfs_vol.vol_uid;
		hdp->gid = fsp -> hsfs_vol.vol_gid;
		hdp->mode = hdp-> mode | (fsp -> hsfs_vol.vol_prot & 0777);
	} else if ((fsp->hsfs_vol_type == HS_VOL_TYPE_ISO) ||
	    (fsp->hsfs_vol_type == HS_VOL_TYPE_ISO_V2) ||
	    (fsp->hsfs_vol_type == HS_VOL_TYPE_JOLIET)) {

		flags = IDE_FLAGS(dirp);
		hs_parse_dirdate(IDE_cdate(dirp), &hdp->cdate);
		hs_parse_dirdate(IDE_cdate(dirp), &hdp->adate);
		hs_parse_dirdate(IDE_cdate(dirp), &hdp->mdate);

		if ((flags & ide_prohibited) == 0) {
			/*
			 * Skip files with the associated bit set.
			 */
			if (flags & IDE_ASSOCIATED)
				return (EAGAIN);
			hdp->type = VREG;
			hdp->mode = HFREG;
			hdp->nlink = 1;
		} else if ((flags & ide_prohibited) == IDE_DIRECTORY) {
			hdp->type = VDIR;
			hdp->mode = HFDIR;
			hdp->nlink = 2;
		} else {
			hs_log_bogus_disk_warning(fsp,
			    HSFS_ERR_UNSUP_TYPE, flags);
			return (EINVAL);
		}
		hdp->uid = fsp -> hsfs_vol.vol_uid;
		hdp->gid = fsp -> hsfs_vol.vol_gid;
		hdp->mode = hdp-> mode | (fsp -> hsfs_vol.vol_prot & 0777);
		hdp->inode = 0;		/* initialize with 0, then check rrip */

		/*
		 * Having this all filled in, let's see if we have any
		 * SUA susp to look at.
		 */
		if (IS_SUSP_IMPLEMENTED(fsp)) {
			error = parse_sua((uchar_t *)dnp, dnlen,
			    &name_change_flag, dirp, last_offset,
			    hdp, fsp, NULL, 0);
			if (error) {
				if (hdp->sym_link) {
					kmem_free(hdp->sym_link,
					    (size_t)(hdp->ext_size + 1));
					hdp->sym_link = (char *)NULL;
				}
				return (error);
			}
		}
	}
	hdp->xar_prot = (HDE_PROTECTION & flags) != 0;

#if dontskip
	if (hdp->xar_len > 0) {
		cmn_err(CE_NOTE, "hsfs: extended attributes not supported");
		return (EINVAL);
	}
#endif

	/* check interleaf size and skip factor */
	/* must both be zero or non-zero */
	if (hdp->intlf_sz + hdp->intlf_sk) {
		if ((hdp->intlf_sz == 0) || (hdp->intlf_sk == 0)) {
			cmn_err(CE_NOTE,
			    "hsfs: interleaf size or skip factor error");
			return (EINVAL);
		}
		if (hdp->ext_size == 0) {
			cmn_err(CE_NOTE,
			    "hsfs: interleaving specified on zero length file");
			return (EINVAL);
		}
	}

	if (HDE_VOL_SET(dirp) != 1) {
		if (fsp->hsfs_vol.vol_set_size != 1 &&
		    fsp->hsfs_vol.vol_set_size != HDE_VOL_SET(dirp)) {
			cmn_err(CE_NOTE, "hsfs: multivolume file?");
			return (EINVAL);
		}
	}

	/*
	 * If the name changed, then the NM field for RRIP was hit and
	 * we should not copy the name again, just return.
	 */
	if (NAME_HAS_CHANGED(name_change_flag))
		return (0);

	/*
	 * Fall back to the ISO name. Note that as in process_dirblock,
	 * the on-disk filename length must be validated against ISO
	 * limits - which, in case of RR present but no RR name found,
	 * are NOT identical to fsp->hsfs_namemax on this filesystem.
	 */
	on_disk_name = (char *)HDE_name(dirp);
	on_disk_namelen = (int)HDE_NAME_LEN(dirp);
	on_disk_dirlen = (int)HDE_DIR_LEN(dirp);

	if (on_disk_dirlen < HDE_ROOT_DIR_REC_SIZE ||
	    ((on_disk_dirlen > last_offset) ||
	    ((HDE_FDESIZE + on_disk_namelen) > on_disk_dirlen))) {
		hs_log_bogus_disk_warning(fsp,
		    HSFS_ERR_BAD_DIR_ENTRY, 0);
		return (EINVAL);
	}

	if (on_disk_namelen > fsp->hsfs_namelen &&
	    hs_namelen(fsp, on_disk_name, on_disk_namelen) >
	    fsp->hsfs_namelen) {
		hs_log_bogus_disk_warning(fsp,
		    fsp->hsfs_vol_type == HS_VOL_TYPE_JOLIET ?
		    HSFS_ERR_BAD_JOLIET_FILE_LEN :
		    HSFS_ERR_BAD_FILE_LEN, 0);
	}
	if (on_disk_namelen > ISO_NAMELEN_V2_MAX)
		on_disk_namelen = fsp->hsfs_namemax;	/* Paranoia */

	if (dnp != NULL) {
		if (fsp->hsfs_vol_type == HS_VOL_TYPE_JOLIET) {
			namelen = hs_jnamecopy(on_disk_name, dnp,
			    on_disk_namelen, fsp->hsfs_namemax,
			    fsp->hsfs_flags);
			/*
			 * A negative return value means that the file name
			 * has been truncated to fsp->hsfs_namemax.
			 */
			if (namelen < 0) {
				namelen = -namelen;
				hs_log_bogus_disk_warning(fsp,
				    HSFS_ERR_TRUNC_JOLIET_FILE_LEN, 0);
			}
		} else {
			/*
			 * HS_VOL_TYPE_ISO && HS_VOL_TYPE_ISO_V2
			 */
			namelen = hs_namecopy(on_disk_name, dnp,
			    on_disk_namelen, fsp->hsfs_flags);
		}
		if (namelen == 0)
			return (EINVAL);
		if ((fsp->hsfs_flags & HSFSMNT_NOTRAILDOT) &&
		    dnp[ namelen-1 ] == '.' && CAN_TRUNCATE_DOT(dnp, namelen))
			dnp[ --namelen ] = '\0';
	} else
		namelen = on_disk_namelen;
	if (dnlen != NULL)
		*dnlen = namelen;

	return (0);
}

/*
 * hs_namecopy
 *
 * Parse a file/directory name into UNIX form.
 * Delete trailing blanks, upper-to-lower case, add NULL terminator.
 * Returns the (possibly new) length.
 *
 * Called from hsfs_readdir() via hs_parsedir()
 */
int
hs_namecopy(char *from, char *to, int size, ulong_t flags)
{
	uint_t i;
	uchar_t c;
	int lastspace;
	int maplc;
	int trailspace;
	int version;

	/* special handling for '.' and '..' */
	if (size == 1) {
		if (*from == '\0') {
			*to++ = '.';
			*to = '\0';
			return (1);
		} else if (*from == '\1') {
			*to++ = '.';
			*to++ = '.';
			*to = '\0';
			return (2);
		}
	}

	maplc = (flags & HSFSMNT_NOMAPLCASE) == 0;
	trailspace = (flags & HSFSMNT_NOTRAILSPACE) == 0;
	version = (flags & HSFSMNT_NOVERSION) == 0;
	for (i = 0, lastspace = -1; i < size; i++) {
		c = from[i];
		if (c == ';' && version)
			break;
		if (c <= ' ' && !trailspace) {
			if (lastspace == -1)
				lastspace = i;
		} else
			lastspace = -1;
		if (maplc && (c >= 'A') && (c <= 'Z'))
			c += 'a' - 'A';
		to[i] = c;
	}
	if (lastspace != -1)
		i = lastspace;
	to[i] = '\0';
	return (i);
}

/*
 * hs_jnamecopy
 *
 * This is the Joliet variant of hs_namecopy()
 *
 * Parse a UCS-2 Joliet file/directory name into UNIX form.
 * Add NULL terminator.
 * Returns the new length.
 *
 * Called from hsfs_readdir() via hs_parsedir()
 */
int
hs_jnamecopy(char *from, char *to, int size, int maxsize, ulong_t flags)
{
	uint_t i;
	uint_t len;
	uint16_t c;
	int	amt;
	int	version;

	/* special handling for '.' and '..' */
	if (size == 1) {
		if (*from == '\0') {
			*to++ = '.';
			*to = '\0';
			return (1);
		} else if (*from == '\1') {
			*to++ = '.';
			*to++ = '.';
			*to = '\0';
			return (2);
		}
	}

	version = (flags & HSFSMNT_NOVERSION) == 0;
	for (i = 0, len = 0; i < size; i++) {
		c = (from[i++] & 0xFF) << 8;
		c |= from[i] & 0xFF;
		if (c == ';' && version)
			break;

		if (len > (maxsize-3)) {
			if (c < 0x80)
				amt = 1;
			else if (c < 0x800)
				amt = 2;
			else
				amt = 3;
			if ((len+amt) > maxsize) {
				to[len] = '\0';
				return (-len);
			}
		}
		amt = hs_ucs2_2_utf8(c, (uint8_t *)&to[len]);
		if (amt == 0) {
			hs_log_bogus_joliet_warning(); /* should never happen */
			return (0);
		}
		len += amt;
	}
	to[len] = '\0';
	return (len);
}

/*
 * map a filename to upper case;
 * return 1 if found lowercase character
 *
 * Called from process_dirblock()
 * via hsfs_lookup() -> hs_dirlook() -> process_dirblock()
 * to create an intermedia name from on disk file names for
 * comparing names.
 */
static int
uppercase_cp(char *from, char *to, int size)
{
	uint_t i;
	uchar_t c;
	uchar_t had_lc = 0;

	for (i = 0; i < size; i++) {
		c = *from++;
		if ((c >= 'a') && (c <= 'z')) {
			c -= ('a' - 'A');
			had_lc = 1;
		}
		*to++ = c;
	}
	return (had_lc);
}

/*
 * This is the Joliet variant of uppercase_cp()
 *
 * map a UCS-2 filename to UTF-8;
 * return new length
 *
 * Called from process_dirblock()
 * via hsfs_lookup() -> hs_dirlook() -> process_dirblock()
 * to create an intermedia name from on disk file names for
 * comparing names.
 */
int
hs_joliet_cp(char *from, char *to, int size)
{
	uint_t		i;
	uint16_t	c;
	int		len = 0;
	int		amt;

	/* special handling for '\0' and '\1' */
	if (size == 1) {
		*to = *from;
		return (1);
	}
	for (i = 0; i < size; i += 2) {
		c = (*from++ & 0xFF) << 8;
		c |= *from++ & 0xFF;

		amt = hs_ucs2_2_utf8(c, (uint8_t *)to);
		if (amt == 0) {
			hs_log_bogus_joliet_warning(); /* should never happen */
			return (0);
		}

		to  += amt;
		len += amt;
	}
	return (len);
}

static void
hs_log_bogus_joliet_warning(void)
{
	static int	warned = 0;

	if (warned)
		return;
	warned = 1;
	cmn_err(CE_CONT, "hsfs: Warning: "
	    "file name contains bad UCS-2 chacarter\n");
}


/*
 * hs_uppercase_copy
 *
 * Convert a UNIX-style name into its HSFS equivalent
 * replacing '.' and '..' with '\0' and '\1'.
 * Map to upper case.
 * Returns the (possibly new) length.
 *
 * Called from hs_dirlook() and rrip_namecopy()
 * to create an intermediate name from the callers name from hsfs_lookup()
 * XXX Is the call from rrip_namecopy() OK?
 */
int
hs_uppercase_copy(char *from, char *to, int size)
{
	uint_t i;
	uchar_t c;

	/* special handling for '.' and '..' */

	if (size == 1 && *from == '.') {
		*to = '\0';
		return (1);
	} else if (size == 2 && *from == '.' && *(from+1) == '.') {
		*to = '\1';
		return (1);
	}

	for (i = 0; i < size; i++) {
		c = *from++;
		if ((c >= 'a') && (c <= 'z'))
			c = c - 'a' + 'A';
		*to++ = c;
	}
	return (size);
}

/*
 * hs_iso_copy
 *
 * This is the Joliet/ISO-9660:1999 variant of hs_uppercase_copy()
 *
 * Convert a UTF-8 UNIX-style name into its UTF-8 Joliet/ISO equivalent
 * replacing '.' and '..' with '\0' and '\1'.
 * Returns the (possibly new) length.
 *
 * Called from hs_dirlook()
 * to create an intermediate name from the callers name from hsfs_lookup()
 */
static int
hs_iso_copy(char *from, char *to, int size)
{
	uint_t i;
	uchar_t c;

	/* special handling for '.' and '..' */

	if (size == 1 && *from == '.') {
		*to = '\0';
		return (1);
	} else if (size == 2 && *from == '.' && *(from+1) == '.') {
		*to = '\1';
		return (1);
	}

	for (i = 0; i < size; i++) {
		c = *from++;
		*to++ = c;
	}
	return (size);
}

void
hs_filldirent(struct vnode *vp, struct hs_direntry *hdp)
{
	struct buf *secbp;
	uint_t	secno;
	offset_t secoff;
	struct hsfs *fsp;
	uchar_t *secp;
	int	error;

	if (vp->v_type != VDIR) {
		cmn_err(CE_WARN, "hsfs_filldirent: vp (0x%p) not a directory",
		    (void *)vp);
		return;
	}

	fsp = VFS_TO_HSFS(vp ->v_vfsp);
	secno = LBN_TO_SEC(hdp->ext_lbn+hdp->xar_len, vp->v_vfsp);
	secoff = LBN_TO_BYTE(hdp->ext_lbn+hdp->xar_len, vp->v_vfsp) &
	    MAXHSOFFSET;
	secbp = bread(fsp->hsfs_devvp->v_rdev, secno * 4, HS_SECTOR_SIZE);
	error = geterror(secbp);
	if (error != 0) {
		cmn_err(CE_NOTE, "hs_filldirent: bread: error=(%d)", error);
		goto end;
	}

	secp = (uchar_t *)secbp->b_un.b_addr;

	/* quick check */
	if (hdp->ext_lbn != HDE_EXT_LBN(&secp[secoff])) {
		cmn_err(CE_NOTE, "hsfs_filldirent: dirent not match");
		/* keep on going */
	}
	(void) hs_parsedir(fsp, &secp[secoff], hdp, (char *)NULL,
	    (int *)NULL, HS_SECTOR_SIZE - secoff);

end:
	brelse(secbp);
}

/*
 * Look through a directory block for a matching entry.
 * Note: this routine does an fbrelse() on the buffer passed in.
 */
static enum dirblock_result
process_dirblock(
	struct fbuf	*fbp,		/* buffer containing dirblk */
	uint_t		*offset,	/* lower index */
	uint_t		last_offset,	/* upper index */
	char		*nm,		/* upcase nm to compare against */
	int		nmlen,		/* length of name */
	struct hsfs	*fsp,
	struct hsnode	*dhp,
	struct vnode	*dvp,
	struct vnode	**vpp,
	int		*error)		/* return value: errno */
{
	uchar_t		*blkp = (uchar_t *)fbp->fb_addr; /* dir block */
	char		*dname;		/* name in directory entry */
	int		dnamelen;	/* length of name */
	struct hs_direntry hd;
	int		hdlen;
	uchar_t		*dirp;		/* the directory entry */
	int		res;
	int		parsedir_res;
	int		is_rrip;
	size_t		rrip_name_size;
	int		rr_namelen = 0;
	char		*rrip_name_str = NULL;
	char		*rrip_tmp_name = NULL;
	enum dirblock_result err = 0;
	int 		did_fbrelse = 0;
	char		uppercase_name[JOLIET_NAMELEN_MAX*3 + 1]; /* 331 */

#define	PD_return(retval)	\
	{ err = retval; goto do_ret; }		/* return after cleanup */
#define	rel_offset(offset)	\
	((offset) & MAXBOFFSET)			/* index into cur blk */
#define	RESTORE_NM(tmp, orig)	\
	if (is_rrip && *(tmp) != '\0') \
		(void) strcpy((orig), (tmp))

	is_rrip = IS_RRIP_IMPLEMENTED(fsp);
	if (is_rrip) {
		rrip_name_size = RRIP_FILE_NAMELEN + 1;
		rrip_name_str = kmem_alloc(rrip_name_size, KM_SLEEP);
		rrip_tmp_name = kmem_alloc(rrip_name_size, KM_SLEEP);
		rrip_name_str[0] = '\0';
		rrip_tmp_name[0] = '\0';
	}

	while (*offset < last_offset) {

		/*
		 * Directory Entries cannot span sectors.
		 *
		 * Unused bytes at the end of each sector are zeroed
		 * according to ISO9660, but we cannot rely on this
		 * since both media failures and maliciously corrupted
		 * media may return arbitrary values.
		 * We therefore have to check for consistency:
		 * The size of a directory entry must be at least
		 * 34 bytes (the size of the directory entry metadata),
		 * or zero (indicating the end-of-sector condition).
		 * For a non-zero directory entry size of less than
		 * 34 Bytes, log a warning.
		 * In any case, skip the rest of this sector and
		 * continue with the next.
		 */
		hdlen = (int)((uchar_t)
		    HDE_DIR_LEN(&blkp[rel_offset(*offset)]));

		if (hdlen < HDE_ROOT_DIR_REC_SIZE ||
		    *offset + hdlen > last_offset) {
			/*
			 * Advance to the next sector boundary
			 */
			*offset = roundup(*offset + 1, HS_SECTOR_SIZE);
			if (hdlen)
				hs_log_bogus_disk_warning(fsp,
				    HSFS_ERR_TRAILING_JUNK, 0);
			continue;
		}

		bzero(&hd, sizeof (hd));

		/*
		 * Check the filename length in the ISO record for
		 * plausibility and reset it to a safe value, in case
		 * the name length byte is out of range. Since the ISO
		 * name will be used as fallback if the rockridge name
		 * is invalid/nonexistant, we must make sure not to
		 * blow the bounds and initialize dnamelen to a sensible
		 * value within the limits of ISO9660.
		 * In addition to that, the ISO filename is part of the
		 * directory entry. If the filename length is too large
		 * to fit, the record is invalid and we'll advance to
		 * the next.
		 */
		dirp = &blkp[rel_offset(*offset)];
		dname = (char *)HDE_name(dirp);
		dnamelen = (int)((uchar_t)HDE_NAME_LEN(dirp));
		/*
		 * If the directory entry extends beyond the end of the
		 * block, it must be invalid. Skip it.
		 */
		if (dnamelen > hdlen - HDE_FDESIZE) {
			hs_log_bogus_disk_warning(fsp,
			    HSFS_ERR_BAD_DIR_ENTRY, 0);
			goto skip_rec;
		} else if (dnamelen > fsp->hsfs_namelen &&
		    hs_namelen(fsp, dname, dnamelen) > fsp->hsfs_namelen) {
			hs_log_bogus_disk_warning(fsp,
			    fsp->hsfs_vol_type == HS_VOL_TYPE_JOLIET ?
			    HSFS_ERR_BAD_JOLIET_FILE_LEN :
			    HSFS_ERR_BAD_FILE_LEN, 0);
		}
		if (dnamelen > ISO_NAMELEN_V2_MAX)
			dnamelen = fsp->hsfs_namemax;	/* Paranoia */

		/*
		 * If the rock ridge is implemented, then we copy the name
		 * from the SUA area to rrip_name_str. If no Alternate
		 * name is found, then use the uppercase NM in the
		 * rrip_name_str char array.
		 */
		if (is_rrip) {

			rrip_name_str[0] = '\0';
			rr_namelen = rrip_namecopy(nm, &rrip_name_str[0],
			    &rrip_tmp_name[0], dirp, last_offset - *offset,
			    fsp, &hd);
			if (hd.sym_link) {
				kmem_free(hd.sym_link,
				    (size_t)(hd.ext_size+1));
				hd.sym_link = (char *)NULL;
			}

			if (rr_namelen != -1) {
				dname = (char *)&rrip_name_str[0];
				dnamelen = rr_namelen;
			}
		}

		if (!is_rrip || rr_namelen == -1) {
			/* use iso name instead */

			int i = -1;
			/*
			 * make sure that we get rid of ';' in the dname of
			 * an iso direntry, as we should have no knowledge
			 * of file versions.
			 *
			 * XXX This is done the wrong way: it does not take
			 * XXX care of the fact that the version string is
			 * XXX a decimal number in the range 1 to 32767.
			 */
			if ((fsp->hsfs_flags & HSFSMNT_NOVERSION) == 0) {
				if (fsp->hsfs_vol_type == HS_VOL_TYPE_JOLIET) {
					for (i = dnamelen - 1; i > 0; i -= 2) {
						if (dname[i] == ';' &&
						    dname[i-1] == '\0') {
							--i;
							break;
						}
					}
				} else {
					for (i = dnamelen - 1; i > 0; i--) {
						if (dname[i] == ';')
							break;
					}
				}
			}
			if (i > 0) {
				dnamelen = i;
			} else if (fsp->hsfs_vol_type != HS_VOL_TYPE_ISO_V2 &&
			    fsp->hsfs_vol_type != HS_VOL_TYPE_JOLIET) {
				dnamelen = strip_trailing(fsp, dname, dnamelen);
			}

			ASSERT(dnamelen < sizeof (uppercase_name));

			if (fsp->hsfs_vol_type == HS_VOL_TYPE_ISO_V2) {
				(void) strncpy(uppercase_name, dname, dnamelen);
			} else if (fsp->hsfs_vol_type == HS_VOL_TYPE_JOLIET) {
				dnamelen = hs_joliet_cp(dname, uppercase_name,
				    dnamelen);
			} else if (uppercase_cp(dname, uppercase_name,
			    dnamelen)) {
				hs_log_bogus_disk_warning(fsp,
				    HSFS_ERR_LOWER_CASE_NM, 0);
			}
			dname = uppercase_name;
			if (!is_rrip &&
			    (fsp->hsfs_flags & HSFSMNT_NOTRAILDOT) &&
			    dname[dnamelen - 1] == '.' &&
			    CAN_TRUNCATE_DOT(dname, dnamelen))
				dname[--dnamelen] = '\0';
		}

		/*
		 * Quickly screen for a non-matching entry, but not for RRIP.
		 * This test doesn't work for lowercase vs. uppercase names.
		 */

		/* if we saw a lower case name we can't do this test either */
		if (strict_iso9660_ordering && !is_rrip &&
		    !HSFS_HAVE_LOWER_CASE(fsp) && *nm < *dname) {
			RESTORE_NM(rrip_tmp_name, nm);
			PD_return(WENT_PAST)
		}

		if (*nm != *dname || nmlen != dnamelen)
			goto skip_rec;

		if ((res = bcmp(dname, nm, nmlen)) == 0) {
			/* name matches */
			parsedir_res = hs_parsedir(fsp, dirp, &hd,
			    (char *)NULL, (int *)NULL,
			    last_offset - *offset);
			if (!parsedir_res) {
				uint_t lbn;	/* logical block number */

				lbn = dhp->hs_dirent.ext_lbn +
				    dhp->hs_dirent.xar_len;
				/*
				 * Need to do an fbrelse() on the buffer,
				 * as hs_makenode() may try to acquire
				 * hs_hashlock, which may not be required
				 * while a page is locked.
				 */
				fbrelse(fbp, S_READ);
				did_fbrelse = 1;
				*vpp = hs_makenode(&hd, lbn, *offset,
				    dvp->v_vfsp);
				if (*vpp == NULL) {
					*error = ENFILE;
					RESTORE_NM(rrip_tmp_name, nm);
					PD_return(FOUND_ENTRY)
				}

				dhp->hs_offset = *offset;
				RESTORE_NM(rrip_tmp_name, nm);
				PD_return(FOUND_ENTRY)
			} else if (parsedir_res != EAGAIN) {
				/* improper dir entry */
				*error = parsedir_res;
				RESTORE_NM(rrip_tmp_name, nm);
				PD_return(FOUND_ENTRY)
			}
		} else if (strict_iso9660_ordering && !is_rrip &&
		    !HSFS_HAVE_LOWER_CASE(fsp) && res < 0) {
			/* name < dir entry */
			RESTORE_NM(rrip_tmp_name, nm);
			PD_return(WENT_PAST)
		}
		/*
		 * name > dir entry,
		 * look at next one.
		 */
skip_rec:
		*offset += hdlen;
		RESTORE_NM(rrip_tmp_name, nm);
	}
	PD_return(HIT_END)

do_ret:
	if (rrip_name_str)
		kmem_free(rrip_name_str, rrip_name_size);
	if (rrip_tmp_name)
		kmem_free(rrip_tmp_name, rrip_name_size);
	if (!did_fbrelse)
		fbrelse(fbp, S_READ);
	return (err);
#undef PD_return
#undef RESTORE_NM
}

/*
 * Strip trailing nulls or spaces from the name;
 * return adjusted length.  If we find such junk,
 * log a non-conformant disk message.
 */
static int
strip_trailing(struct hsfs *fsp, char *nm, int len)
{
	char *c;
	int trailing_junk = 0;

	for (c = nm + len - 1; c > nm; c--) {
		if (*c == ' ' || *c == '\0')
			trailing_junk = 1;
		else
			break;
	}

	if (trailing_junk)
		hs_log_bogus_disk_warning(fsp, HSFS_ERR_TRAILING_JUNK, 0);

	return ((int)(c - nm + 1));
}

static int
hs_namelen(struct hsfs *fsp, char *nm, int len)
{
	char	*p = nm + len;

	if (fsp->hsfs_vol_type == HS_VOL_TYPE_ISO_V2) {
		return (len);
	} else if (fsp->hsfs_vol_type == HS_VOL_TYPE_JOLIET) {
		uint16_t c;

		while (--p > &nm[1]) {
			c = *p;
			c |= *--p * 256;
			if (c == ';')
				return (p - nm);
			if (c < '0' || c > '9') {
				p++;
				return (p - nm);
			}
		}
	} else {
		char	c;

		while (--p > nm) {
			c = *p;
			if (c == ';')
				return (p - nm);
			if (c < '0' || c > '9') {
				p++;
				return (p - nm);
			}
		}
	}
	return (len);
}

/*
 * Take a UCS-2 character and convert
 * it into a utf8 character.
 * A 0 will be returned if the conversion fails
 *
 * See http://www.cl.cam.ac.uk/~mgk25/unicode.html#utf-8
 *
 * The code has been taken from udfs/udf_subr.c
 */
static uint8_t hs_first_byte_mark[7] =
			{ 0x00, 0x00, 0xC0, 0xE0, 0xF0, 0xF8, 0xFC };
static int32_t
hs_ucs2_2_utf8(uint16_t c_16, uint8_t *s_8)
{
	int32_t nc;
	uint32_t c_32;
	uint32_t byte_mask = 0xBF;
	uint32_t byte_mark = 0x80;

	/*
	 * Convert the 16-bit character to a 32-bit character
	 */
	c_32 = c_16;

	/*
	 * By here the 16-bit character is converted
	 * to a 32-bit wide character
	 */
	if (c_32 < 0x80) {
		nc = 1;
	} else if (c_32 < 0x800) {
		nc = 2;
	} else if (c_32 < 0x10000) {
		nc = 3;
	} else if (c_32 < 0x200000) {
		nc = 4;
	} else if (c_32 < 0x4000000) {
		nc = 5;
	} else if (c_32 <= 0x7FFFFFFF) {	/* avoid signed overflow */
		nc = 6;
	} else {
		nc = 0;
	}
	s_8 += nc;
	switch (nc) {
		case 6 :
			*(--s_8) = (c_32 | byte_mark)  & byte_mask;
			c_32 >>= 6;
			/* FALLTHROUGH */
		case 5 :
			*(--s_8) = (c_32 | byte_mark)  & byte_mask;
			c_32 >>= 6;
			/* FALLTHROUGH */
		case 4 :
			*(--s_8) = (c_32 | byte_mark)  & byte_mask;
			c_32 >>= 6;
			/* FALLTHROUGH */
		case 3 :
			*(--s_8) = (c_32 | byte_mark)  & byte_mask;
			c_32 >>= 6;
			/* FALLTHROUGH */
		case 2 :
			*(--s_8) = (c_32 | byte_mark)  & byte_mask;
			c_32 >>= 6;
			/* FALLTHROUGH */
		case 1 :
			*(--s_8) = c_32 | hs_first_byte_mark[nc];
	}
	return (nc);
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

/*
 * Rock Ridge extensions to the System Use Sharing protocol
 * for the High Sierra filesystem
 */

#include <sys/types.h>
#include <sys/t_lock.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kmem.h>
#include <sys/signal.h>
#include <sys/user.h>
#include <sys/proc.h>
#include <sys/disp.h>
#include <sys/buf.h>
#include <sys/pathname.h>
#include <sys/vfs.h>
#include <sys/vnode.h>
#include <sys/file.h>
#include <sys/uio.h>
#include <sys/conf.h>
#include <sys/stat.h>
#include <sys/mode.h>
#include <sys/mkdev.h>
#include <sys/ddi.h>

#include <vm/page.h>

#include <sys/fs/hsfs_spec.h>
#include <sys/fs/hsfs_isospec.h>
#include <sys/fs/hsfs_node.h>
#include <sys/fs/hsfs_impl.h>
#include <sys/fs/hsfs_susp.h>
#include <sys/fs/hsfs_rrip.h>

#include <sys/statvfs.h>
#include <sys/mount.h>
#include <sys/swap.h>
#include <sys/errno.h>
#include <sys/debug.h>
#include "fs/fs_subr.h"
#include <sys/cmn_err.h>

static void form_time(int, uchar_t *, struct timeval *);
static void name_parse(int, uchar_t *, size_t, uchar_t *, int *,
    ulong_t *, int);

/*
 *  Signature table for RRIP
 */
ext_signature_t  rrip_signature_table[ ] = {
	RRIP_CL,	rrip_child_link,
	RRIP_NM,	rrip_name,
	RRIP_PL,	rrip_parent_link,
	RRIP_PN,	rrip_dev_nodes,
	RRIP_PX,	rrip_file_attr,
	RRIP_RE,	rrip_reloc_dir,
	RRIP_RR,	rrip_rock_ridge,
	RRIP_SL,	rrip_sym_link,
	RRIP_TF,	rrip_file_time,
	(char *)NULL,   NULL
};


/*
 * rrip_dev_nodes()
 *
 * sig_handler() for RRIP signature "PN"
 *
 * This function parses out the major and minor numbers from the "PN
 * " SUF.
 */
uchar_t *
rrip_dev_nodes(sig_args_t *sig_args_p)
{
	uchar_t *pn_ptr = sig_args_p->SUF_ptr;
	major_t	major_dev = (major_t)RRIP_MAJOR(pn_ptr);
	minor_t	minor_dev = (minor_t)RRIP_MINOR(pn_ptr);

	sig_args_p->hdp->r_dev = makedevice(major_dev, minor_dev);

	return (pn_ptr + SUF_LEN(pn_ptr));
}

/*
 * rrip_file_attr()
 *
 * sig_handler() for RRIP signature "PX"
 *
 * This function parses out the file attributes of a file from the "PX"
 * SUF.  The attributes is finds are : st_mode, st_nlink, st_uid,
 * and st_gid.
 */
uchar_t *
rrip_file_attr(sig_args_t *sig_args_p)
{
	uchar_t *px_ptr = sig_args_p->SUF_ptr;
	struct hs_direntry *hdp    = sig_args_p->hdp;

	hdp->mode  = RRIP_MODE(px_ptr);
	hdp->nlink = RRIP_NLINK(px_ptr);
	hdp->uid   = RRIP_UID(px_ptr);
	hdp->gid   = RRIP_GID(px_ptr);

	if (SUF_LEN(px_ptr) >= RRIP_PX_SIZE)
		hdp->inode = (ino64_t)RRIP_INO(px_ptr);
	else
		hdp->inode = 0;

	hdp->type = IFTOVT(hdp->mode);

	return (px_ptr + SUF_LEN(px_ptr));
}

/*
 * rrip_file_time()
 *
 * support function for rrip_file_time()
 *
 * This function decides whether to parse the times in a long time form
 * (17 bytes) or a short time form (7 bytes).  These time formats are
 * defined in the ISO 9660 specification.
 */
static void
form_time(int time_length, uchar_t *file_time, struct timeval *tvp)
{
	if (time_length == ISO_DATE_LEN)
		hs_parse_longdate(file_time, tvp);
	else
		hs_parse_dirdate(file_time, tvp);

}

/*
 * rrip_file_time()
 *
 * sig_handler() for RRIP signature RRIP_TF
 *
 * This function parses out the file time attributes of a file from the
 * "TI" SUF.  The times it parses are : st_mtime, st_atime and st_ctime.
 *
 * The function form_time is a support function only used in this
 * function.
 */
uchar_t *
rrip_file_time(sig_args_t *sig_args_p)
{
	uchar_t *tf_ptr = sig_args_p->SUF_ptr;

	if (IS_TIME_BIT_SET(RRIP_TF_FLAGS(tf_ptr), RRIP_TF_ACCESS_BIT)) {
		form_time(RRIP_TF_TIME_LENGTH(tf_ptr),
		    RRIP_tf_access(tf_ptr),
		    &sig_args_p->hdp->adate);
	}

	if (IS_TIME_BIT_SET(RRIP_TF_FLAGS(tf_ptr), RRIP_TF_MODIFY_BIT)) {
		form_time(RRIP_TF_TIME_LENGTH(tf_ptr), RRIP_tf_modify(tf_ptr),
		    &sig_args_p->hdp->mdate);
	}

	if (IS_TIME_BIT_SET(RRIP_TF_FLAGS(tf_ptr), RRIP_TF_ATTRIBUTES_BIT)) {
		form_time(RRIP_TF_TIME_LENGTH(tf_ptr),
		    RRIP_tf_attributes(tf_ptr),
		    &sig_args_p->hdp->cdate);
	}

	return (tf_ptr + SUF_LEN(tf_ptr));
}



/*
 * name_parse()
 *
 * This is a generic fuction used for sym links and filenames.  The
 * flags passed to it effect the way the name/component field is parsed.
 *
 * The return value will be the NAME_CONTINUE or NAME_CHANGE value.
 *
 */
static void
name_parse(
	int		rrip_flags,	/* component/name flag */
	uchar_t		*SUA_string,	/* string from SUA */
	size_t		SUA_string_len, /* length of SUA string */
	uchar_t		*dst,		/* string to copy to */
	int		*dst_lenp,	/* ptr to cur. str len */
	ulong_t		*name_flags_p,	/* internal name flags */
	int		dst_size)	/* limit dest string to */
						/* this value */
{
	size_t	off;
	size_t	len;

	if (IS_NAME_BIT_SET(rrip_flags, RRIP_NAME_ROOT))
		dst[0] = 0;

	if (IS_NAME_BIT_SET(rrip_flags, RRIP_NAME_CURRENT)) {
		SUA_string = (uchar_t *)".";
		SUA_string_len = 1;
	}

	if (IS_NAME_BIT_SET(rrip_flags, RRIP_NAME_PARENT)) {
		SUA_string = (uchar_t *)"..";
		SUA_string_len = 2;
	}

	/*
	 * XXX
	 * For now, ignore the following flags and return.
	 * have to figure out how to get host name in kernel.
	 * Unsure if this even should be done.
	 */
	if (IS_NAME_BIT_SET(rrip_flags, RRIP_NAME_VOLROOT) ||
	    IS_NAME_BIT_SET(rrip_flags, RRIP_NAME_HOST)) {
		cmn_err(CE_NOTE,
			"VOLUME ROOT and NAME_HOST currently unsupported\n");
		return;
	}

	/*
	 * strlcat() has two nice properties:
	 * - the size of the output buffer includes the trailing '\0'
	 * - we pass "total size" not "remaining size"
	 * It'd be the ideal candidate for this codeblock - make it:
	 *
	 *	strlcat(dst, SUA_string,
	 *	    MIN(dstsize, strlen(dst) + SUA_string_len + 1));
	 *
	 * Unfortunately, strlcat() cannot deal with input strings
	 * that are not NULL-terminated - like SUA_string can be in
	 * our case here. So we can't use it :(
	 * Now strncat() doesn't work either - because it doesn't deal
	 * with strings for which the 'potential NULL-termination' isn't
	 * accessible - strncat(dst, NULL, 0) crashes although it copies
	 * nothing in any case. If the SUA ends on a mapping boundary,
	 * then telling strncat() to copy no more than the remaining bytes
	 * in the buffer is of no avail if there's no NULL byte in them.
	 *
	 * Hence - binary copy. What are all these str* funcs for ??
	 */
	dst_size--;	/* trailing '\0' */

	off = strlen((char *)dst);
	len = MIN(dst_size - off, SUA_string_len);
	bcopy((char *)SUA_string, (char *)(dst + off), len);
	dst[off + len] = '\0';
	*dst_lenp = strlen((char *)dst);

	if (IS_NAME_BIT_SET(rrip_flags, RRIP_NAME_CONTINUE))
		SET_NAME_BIT(*(name_flags_p), RRIP_NAME_CONTINUE);
	else
		SET_NAME_BIT(*(name_flags_p), RRIP_NAME_CHANGE);

}

/*
 * rrip_name()
 *
 * sig_handler() for RRIP signature RRIP_NM
 *
 * This function handles the name of the current file.  It is case
 * sensitive to whatever was put into the field and does NO
 * translation. It will take whatever characters were in the field.
 *
 * Because the flags effect the way the name is parsed the same way
 * that the sym_link component parsing is done, we will use the same
 * function to do the actual parsing.
 */
uchar_t  *
rrip_name(sig_args_t *sig_args_p)
{
	uchar_t *nm_ptr = sig_args_p->SUF_ptr;

	if ((sig_args_p->name_p == (uchar_t *)NULL) ||
	    (sig_args_p->name_len_p == (int *)NULL))
		goto end;
	/*
	 * If we have a "." or ".." directory, we should not look for
	 * an alternate name
	 */
	if (HDE_NAME_LEN(sig_args_p->dirp) == 1) {
		if (*((char *)HDE_name(sig_args_p->dirp)) == '\0') {
			(void) strcpy((char *)sig_args_p->name_p, ".");
			*sig_args_p->name_len_p = 1;
			goto end;
		} else if (*((char *)HDE_name(sig_args_p->dirp)) == '\1') {
			(void) strcpy((char *)sig_args_p->name_p, "..");
			*sig_args_p->name_len_p = 2;
			goto end;
		}
	}

	name_parse((int)RRIP_NAME_FLAGS(nm_ptr), RRIP_name(nm_ptr),
	    (size_t)RRIP_NAME_LEN(nm_ptr), sig_args_p->name_p,
	    sig_args_p->name_len_p, &(sig_args_p->name_flags),
	    MAXNAMELEN);

end:
	return (nm_ptr + SUF_LEN(nm_ptr));
}


/*
 * rrip_sym_link()
 *
 * sig_handler() for RRIP signature RRIP_SL
 *
 * creates a symlink buffer to simulate sym_links.
 */
uchar_t *
rrip_sym_link(sig_args_t *sig_args_p)
{
	uchar_t	*sl_ptr = sig_args_p->SUF_ptr;
	uchar_t	*comp_ptr;
	char 	*tmp_sym_link;
	struct hs_direntry *hdp = sig_args_p->hdp;
	int	sym_link_len;
	char	*sym_link;

	if (hdp->type != VLNK)
		goto end;

	/*
	 * If the sym link has already been created, don't recreate it
	 */
	if (IS_NAME_BIT_SET(sig_args_p->name_flags, RRIP_SYM_LINK_COMPLETE))
		goto end;

	sym_link = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);

	/*
	 * If there is an original string put it into sym_link[], otherwise
	 * initialize sym_link[] to the empty string.
	 */
	if (hdp->sym_link != (char *)NULL) {
		sym_link_len = (int)strlen(strcpy(sym_link, hdp->sym_link));
	} else {
		sym_link[0] = '\0';
		sym_link_len = 0;
	}

	/* for all components */
	for (comp_ptr = RRIP_sl_comp(sl_ptr);
	    comp_ptr < (sl_ptr + SUF_LEN(sl_ptr));
	    comp_ptr += RRIP_COMP_LEN(comp_ptr)) {

		name_parse((int)RRIP_COMP_FLAGS(comp_ptr),
		    RRIP_comp(comp_ptr),
		    (size_t)RRIP_COMP_NAME_LEN(comp_ptr), (uchar_t *)sym_link,
		    &sym_link_len, &(sig_args_p->name_flags),
		    MAXPATHLEN);

		/*
		 * If the component is continued don't put a '/' in
		 * the pathname, but do NULL terminate it.
		 */
		if (IS_NAME_BIT_SET(RRIP_COMP_FLAGS(comp_ptr),
		    RRIP_NAME_CONTINUE)) {
			sym_link[sym_link_len] = '\0';
		} else {
			sym_link[sym_link_len] = '/';
			sym_link[sym_link_len + 1] = '\0';

			/* add 1 to sym_link_len for '/' */
			sym_link_len++;
		}

	}

	/*
	 * If we reached the end of the symbolic link, take out the
	 * last slash, but don't change ROOT "/" to an empty string.
	 */
	if (!IS_NAME_BIT_SET(RRIP_SL_FLAGS(sl_ptr), RRIP_NAME_CONTINUE) &&
	    sym_link_len > 1 && sym_link[sym_link_len - 1] == '/')
		sym_link[--sym_link_len] = '\0';

	/*
	 * if no memory has been allocated, get some, otherwise, append
	 * to the current allocation
	 */

	tmp_sym_link = kmem_alloc(SYM_LINK_LEN(sym_link), KM_SLEEP);

	(void) strcpy(tmp_sym_link, sym_link);

	if (hdp->sym_link != (char *)NULL)
		kmem_free(hdp->sym_link, (size_t)(hdp->ext_size + 1));

	hdp->sym_link = (char *)&tmp_sym_link[0];
	/* the size of a symlink is its length */
	hdp->ext_size = (uint_t)strlen(tmp_sym_link);

	if (!IS_NAME_BIT_SET(RRIP_SL_FLAGS(sl_ptr), RRIP_NAME_CONTINUE)) {
		/* reached the end of the symbolic link */
		SET_NAME_BIT(sig_args_p->name_flags, RRIP_SYM_LINK_COMPLETE);
	}

	kmem_free(sym_link, MAXPATHLEN + 1);
end:
	return (sl_ptr + SUF_LEN(sl_ptr));
}

/*
 * rrip_namecopy()
 *
 * This function will copy the rrip name to the "to" buffer, if it
 * exists.
 *
 * XXX -  We should speed this up by implementing the search in
 * parse_sua().  It works right now, so I don't want to mess with it.
 */
int
rrip_namecopy(
	char 	*from,			/* name to copy */
	char 	*to,			/* string to copy "from" to */
	char  	*tmp_name,		/* temp storage for original name */
	uchar_t	*dirp,			/* directory entry pointer */
	uint_t	last_offset,		/* last index into current dir block */
	struct 	hsfs *fsp,		/* filesystem pointer */
	struct 	hs_direntry *hdp)	/* directory entry pointer to put */
					/* all that good info in */
{
	int	size = 0;
	int	change_flag = 0;
	int	ret_val;

	if ((to == (char *)NULL) ||
	    (from == (char *)NULL) ||
	    (dirp == (uchar_t *)NULL)) {
		return (0);
	}

	/* special handling for '.' and '..' */

	if (HDE_NAME_LEN(dirp) == 1) {
		if (*((char *)HDE_name(dirp)) == '\0') {
			(void) strcpy(to, ".");
			return (1);
		} else if (*((char *)HDE_name(dirp)) == '\1') {
			(void) strcpy(to, "..");
			return (2);
		}
	}


	ret_val = parse_sua((uchar_t *)to, &size, &change_flag,
	    dirp, last_offset, hdp, fsp, NULL, 0);

	if (IS_NAME_BIT_SET(change_flag, RRIP_NAME_CHANGE) && !ret_val)
		return (size);

	/*
	 * Well, the name was not found
	 *
	 * make rripname an upper case "nm" (to), so that
	 * we can compare it the current HDE_DIR_NAME()
	 * without nuking the original "nm", for future case
	 * sensitive name comparing
	 */
	(void) strcpy(tmp_name, from);		/* keep original */
	size = hs_uppercase_copy(tmp_name, from, (int)strlen(from));

	return (-1);
}



/*
 * rrip_reloc_dir()
 *
 * This function is fairly bogus.  All it does is cause a failure of
 * the hs_parsedir, so that no vnode will be made for it and
 * essentially, the directory will no longer be seen.  This is part
 * of the directory hierarchy mess, where the relocated directory will
 * be hidden as far as ISO 9660 is concerned.  When we hit the child
 * link "CL" SUF, then we will read the new directory.
 */
uchar_t *
rrip_reloc_dir(sig_args_t *sig_args_p)
{
	uchar_t *re_ptr = sig_args_p->SUF_ptr;

	sig_args_p->flags = RELOC_DIR;

	return (re_ptr + SUF_LEN(re_ptr));
}



/*
 * rrip_child_link()
 *
 * This is the child link part of the directory hierarchy stuff and
 * this does not really do much anyway.  All it does is read the
 * directory entry that the child link SUF contains.  All should be
 * fine then.
 */
uchar_t *
rrip_child_link(sig_args_t *sig_args_p)
{
	uchar_t *cl_ptr = sig_args_p->SUF_ptr;

	sig_args_p->hdp->ext_lbn = RRIP_CHILD_LBN(cl_ptr);

	hs_filldirent(sig_args_p->fsp->hsfs_rootvp, sig_args_p->hdp);

	sig_args_p->flags = 0;

	return (cl_ptr + SUF_LEN(cl_ptr));
}


/*
 * rrip_parent_link()
 *
 * This is the parent link part of the directory hierarchy stuff and
 * this does not really do much anyway.  All it does is read the
 * directory entry that the parent link SUF contains.  All should be
 * fine then.
 */
uchar_t *
rrip_parent_link(sig_args_t *sig_args_p)
{
	uchar_t *pl_ptr = sig_args_p->SUF_ptr;

	sig_args_p->hdp->ext_lbn = RRIP_PARENT_LBN(pl_ptr);

	hs_filldirent(sig_args_p->fsp->hsfs_rootvp, sig_args_p->hdp);

	sig_args_p->flags = 0;

	return (pl_ptr + SUF_LEN(pl_ptr));
}


/*
 * rrip_rock_ridge()
 *
 * This function is supposed to aid in speed of the filesystem.
 *
 * XXX - It is only here as a place holder so far.
 */
uchar_t *
rrip_rock_ridge(sig_args_t *sig_args_p)
{
	uchar_t *rr_ptr = sig_args_p->SUF_ptr;

	return (rr_ptr + SUF_LEN(rr_ptr));
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Miscellaneous support subroutines for High Sierra filesystem
 *
 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#include <sys/types.h>
#include <sys/param.h>
#include <sys/time.h>
#include <sys/cmn_err.h>
#include <sys/systm.h>
#include <sys/sysmacros.h>
#include <sys/buf.h>
#include <sys/conf.h>
#include <sys/user.h>
#include <sys/vfs.h>
#include <sys/vnode.h>
#include <sys/proc.h>
#include <sys/debug.h>
#include <sys/kmem.h>
#include <sys/uio.h>
#include <vm/hat.h>
#include <vm/as.h>
#include <vm/seg.h>
#include <vm/page.h>
#include <vm/pvn.h>
#include <vm/seg_map.h>
#include <sys/swap.h>
#include <vm/seg_kmem.h>

#include <sys/fs/hsfs_spec.h>
#include <sys/fs/hsfs_node.h>
#include <sys/fs/hsfs_impl.h>

#define	THE_EPOCH	1970
#define	END_OF_TIME	2099
extern int hsfs_lostpage;

#ifdef __STDC__
static time_t hs_date_to_gmtime(int year, int mon, int day, int gmtoff);
#else
static time_t hs_date_to_gmtime();
#endif

/*
 * Table used in logging non-fatal errors which should be recorded
 * once per mount.  Indexed by HSFS_ERR values (defined in hsfs_node.h).
 */
struct hsfs_error {
	char	*hdr_text;	/* msg prefix: general error type */
				/* must contain %s for mnt pt */
	char 	*err_text;	/* specific error message */
	uchar_t	multiple;	/* > 1 such error per fs possible? */
	uchar_t	n_printf_args;	/* if err_text printf-like, # addtl args */
} hsfs_error[] = {
	/* HSFS_ERR_TRAILING_JUNK */
	"hsfs: Warning: the file system mounted on %s "
		"does not conform to the ISO-9660 specification:",
	"trailing blanks or null characters in file or directory name.\n",
	1, 0,
	/* HSFS_ERR_LOWER_CASE_NM */
	"hsfs: Warning: the file system mounted on %s "
		"does not conform to the ISO-9660 specification:",
	"lower case characters in file or directory name.\n",
	1, 0,
	/* HSFS_ERR_BAD_ROOT_DIR */
	"hsfs: Warning: the file system mounted on %s "
		"does not conform to the ISO-9660 specification:",
	"invalid root directory.\n",
	0,  0,
	/* HSFS_ERR_UNSUP_TYPE */
	"hsfs: Warning: the file system mounted on %s "
		"contains a file or directory with an unsupported type:",
	" 0x%x.\n",
	1, 1,
	/* HSFS_ERR_BAD_FILE_LEN */
	"hsfs: Warning: file system mounted on %s "
		"does not conform to the ISO-9660 specification:",
	"file name length greater than max allowed\n",
	1, 0,
	/* HSFS_ERR_BAD_JOLIET_FILE_LEN */
	"hsfs: Warning: file system mounted on %s "
		"does not conform to the Joliet specification:",
	"file name length greater than max allowed\n",
	1, 0,
	/* HSFS_ERR_TRUNC_JOLIET_FILE_LEN */
	"hsfs: Warning: file system mounted on %s "
		"does not conform to the Joliet specification:",
	"file name length greater than MAXNAMELEN (truncated)\n",
	1, 0,
	/* HSFS_ERR_BAD_DIR_ENTRY */
	"hsfs: Warning: file system mounted on %s "
		"has inconsistent data:",
	"invalid directory or file name length (ignored)\n",
	1, 0,
	/* HSFS_ERR_NEG_SUA_LEN */
	"hsfs: Warning: file system mounted on %s "
		"has inconsistent Rock Ridge data:",
	"negative SUA len\n",
	1, 0,
	/* HSFS_ERR_BAD_SUA_LEN */
	"hsfs: Warning: file system mounted on %s "
		"has inconsistent Rock Ridge data:",
	"SUA len too big\n",
	1, 0,
};

/*
 * Local datatype for defining tables of (Offset, Name) pairs for
 * kstats.
 */
typedef struct {
	offset_t	index;
	char		*name;
} hsfs_ksindex_t;

static const hsfs_ksindex_t hsfs_kstats[] = {
	{ 0,		"mountpoint"		},
	{ 1,		"pages_lost"		},
	{ 2,		"physical_read_pages"	},
	{ 3,		"cache_read_pages"	},
	{ 4,		"readahead_pages"	},
	{ 5,		"coalesced_pages"	},
	{ 6,		"total_pages_requested"	},
	{-1,		NULL 			}
};

/*
 * hs_parse_dirdate
 *
 * Parse the short 'directory-format' date into a Unix timeval.
 * This is the date format used in Directory Entries.
 *
 * If the date is not representable, make something up.
 */
void
hs_parse_dirdate(dp, tvp)
	uchar_t *dp;
	struct timeval *tvp;
{
	int year, month, day, hour, minute, sec, gmtoff;

	year = HDE_DATE_YEAR(dp);
	month = HDE_DATE_MONTH(dp);
	day = HDE_DATE_DAY(dp);
	hour = HDE_DATE_HOUR(dp);
	minute = HDE_DATE_MIN(dp);
	sec = HDE_DATE_SEC(dp);
	gmtoff = HDE_DATE_GMTOFF(dp);

	tvp->tv_usec = 0;
	if (year < THE_EPOCH) {
		tvp->tv_sec = 0;
	} else {
		tvp->tv_sec = hs_date_to_gmtime(year, month, day, gmtoff);
		if (tvp->tv_sec != -1) {
			tvp->tv_sec += ((hour * 60) + minute) * 60 + sec;
		}
	}

	return;

}

/*
 * hs_parse_longdate
 *
 * Parse the long 'user-oriented' date into a Unix timeval.
 * This is the date format used in the Volume Descriptor.
 *
 * If the date is not representable, make something up.
 */
void
hs_parse_longdate(dp, tvp)
	uchar_t *dp;
	struct timeval *tvp;
{
	int year, month, day, hour, minute, sec, gmtoff;

	year = HSV_DATE_YEAR(dp);
	month = HSV_DATE_MONTH(dp);
	day = HSV_DATE_DAY(dp);
	hour = HSV_DATE_HOUR(dp);
	minute = HSV_DATE_MIN(dp);
	sec = HSV_DATE_SEC(dp);
	gmtoff = HSV_DATE_GMTOFF(dp);

	tvp->tv_usec = 0;
	if (year < THE_EPOCH) {
		tvp->tv_sec = 0;
	} else {
		tvp->tv_sec = hs_date_to_gmtime(year, month, day, gmtoff);
		if (tvp->tv_sec != -1) {
			tvp->tv_sec += ((hour * 60) + minute) * 60 + sec;
			tvp->tv_usec = HSV_DATE_HSEC(dp) * 10000;
		}
	}

}

/* cumulative number of seconds per month,  non-leap and leap-year versions */
static time_t cum_sec[] = {
	0x0, 0x28de80, 0x4dc880, 0x76a700, 0x9e3400, 0xc71280,
	0xee9f80, 0x1177e00, 0x1405c80, 0x167e980, 0x190c800, 0x1b85500
};
static time_t cum_sec_leap[] = {
	0x0, 0x28de80, 0x4f1a00, 0x77f880, 0x9f8580, 0xc86400,
	0xeff100, 0x118cf80, 0x141ae00, 0x1693b00, 0x1921980, 0x1b9a680
};
#define	SEC_PER_DAY	0x15180
#define	SEC_PER_YEAR	0x1e13380

/*
 * hs_date_to_gmtime
 *
 * Convert year(1970-2099)/month(1-12)/day(1-31) to seconds-since-1970/1/1.
 *
 * Returns -1 if the date is out of range.
 */
static time_t
hs_date_to_gmtime(year, mon, day, gmtoff)
	int year;
	int mon;
	int day;
	int gmtoff;
{
	time_t sum;
	time_t *cp;
	int y;

	if ((year < THE_EPOCH) || (year > END_OF_TIME) ||
	    (mon < 1) || (mon > 12) ||
	    (day < 1) || (day > 31))
		return (-1);

	/*
	 * Figure seconds until this year and correct for leap years.
	 * Note: 2000 is a leap year but not 2100.
	 */
	y = year - THE_EPOCH;
	sum = y * SEC_PER_YEAR;
	sum += ((y + 1) / 4) * SEC_PER_DAY;
	/*
	 * Point to the correct table for this year and
	 * add in seconds until this month.
	 */
	cp = ((y + 2) % 4) ? cum_sec : cum_sec_leap;
	sum += cp[mon - 1];
	/*
	 * Add in seconds until 0:00 of this day.
	 * (days-per-month validation is not done here)
	 */
	sum += (day - 1) * SEC_PER_DAY;
	sum -= (gmtoff * 15 * 60);
	return (sum);
}

/*
 * Indicate whether the directory is valid.
 */

int
hsfs_valid_dir(hd)
	struct hs_direntry *hd;
{
	/*
	 * check to see if this directory is not marked as a directory.
	 * check to see if data length is zero.
	 */

	if (hd->ext_size == 0)
		return (0);

	if (hd->type != VDIR)
		return (0);

	return (1);
}



/*
 * If we haven't complained about this error type yet, do.
 */
void
hs_log_bogus_disk_warning(fsp, errtype, data)
	struct hsfs	*fsp;
	int 		errtype;
	uint_t		data;
{

	if (fsp->hsfs_err_flags & (1 << errtype))
		return;		/* already complained */

	cmn_err(CE_NOTE, hsfs_error[errtype].hdr_text,
		fsp->hsfs_fsmnt);

	switch (hsfs_error[errtype].n_printf_args) {
	case 0:
		cmn_err(CE_CONT, hsfs_error[errtype].err_text);
		break;
	case 1:
		cmn_err(CE_CONT, hsfs_error[errtype].err_text, data);
		break;
	default:
		/* don't currently handle more than 1 arg */
		cmn_err(CE_CONT, "unknown problem; internal error.\n");
	}
	cmn_err(CE_CONT,
"Due to this error, the file system may not be correctly interpreted.\n");
	if (hsfs_error[errtype].multiple)
		cmn_err(CE_CONT,
"Other such errors in this file system will be silently ignored.\n\n");
	else
		cmn_err(CE_CONT, "\n");

	fsp->hsfs_err_flags |= (1 << errtype);
}

/*
 * Callback from kstat framework. Grab a snapshot of the current hsfs
 * counters and populate the kstats.
 */
static int
hsfs_kstats_update(kstat_t *ksp, int flag)
{
	struct hsfs *fsp;
	kstat_named_t *knp;
	uint64_t pages_lost;
	uint64_t physical_read_bytes;
	uint64_t cache_read_pages;
	uint64_t readahead_bytes;
	uint64_t coalesced_bytes;
	uint64_t total_pages_requested;

	if (flag != KSTAT_READ)
		return (EACCES);

	fsp = ksp->ks_private;
	knp = ksp->ks_data;

	mutex_enter(&(fsp->hqueue->strategy_lock));
	mutex_enter(&(fsp->hqueue->hsfs_queue_lock));

	cache_read_pages = fsp->cache_read_pages;
	pages_lost = hsfs_lostpage;
	physical_read_bytes = fsp->physical_read_bytes;
	readahead_bytes =  fsp->readahead_bytes;
	coalesced_bytes = fsp->coalesced_bytes;
	total_pages_requested = fsp->total_pages_requested;

	mutex_exit(&(fsp->hqueue->strategy_lock));
	mutex_exit(&(fsp->hqueue->hsfs_queue_lock));

	knp++;
	(knp++)->value.ui64 = pages_lost;
	(knp++)->value.ui64 = howmany(physical_read_bytes, PAGESIZE);
	(knp++)->value.ui64 = cache_read_pages;
	(knp++)->value.ui64 = howmany(readahead_bytes, PAGESIZE);
	(knp++)->value.ui64 = howmany(coalesced_bytes, PAGESIZE);
	(knp++)->value.ui64 = total_pages_requested;

	return (0);
}

/*
 * Initialize hsfs kstats, which are all name value pairs with
 * values being various counters.
 */
static kstat_t *
hsfs_setup_named_kstats(struct hsfs *fsp, int fsid, char *name,
    const hsfs_ksindex_t *ksip, int (*update)(kstat_t *, int))
{
	kstat_t *ksp;
	kstat_named_t *knp;
	char *np;
	char *mntpt = fsp->hsfs_fsmnt;
	size_t size;

	size = (sizeof (hsfs_kstats)) / (sizeof (hsfs_ksindex_t));
	ksp = kstat_create("hsfs_fs", fsid, name, "hsfs",
	    KSTAT_TYPE_NAMED, size-1, KSTAT_FLAG_VIRTUAL);
	if (ksp == NULL)
		return (NULL);

	ksp->ks_data = kmem_alloc(sizeof (kstat_named_t) * size, KM_SLEEP);
	ksp->ks_private = fsp;
	ksp->ks_update = update;
	ksp->ks_data_size += strlen(mntpt) + 1;
	knp = ksp->ks_data;
	kstat_named_init(knp, ksip->name, KSTAT_DATA_STRING);
	kstat_named_setstr(knp, mntpt);
	knp++;
	ksip++;

	for (; (np = ksip->name) != NULL; ++knp, ++ksip) {
		kstat_named_init(knp, np, KSTAT_DATA_UINT64);
	}
	kstat_install(ksp);

	return (ksp);
}

void
hsfs_init_kstats(struct hsfs *fsp, int fsid)
{
	fsp->hsfs_kstats = hsfs_setup_named_kstats(fsp, fsid, "hsfs_read_stats",
	    hsfs_kstats, hsfs_kstats_update);
}

void
hsfs_fini_kstats(struct hsfs *fsp)
{
	void *data;

	if (fsp->hsfs_kstats != NULL) {
		data = fsp->hsfs_kstats->ks_data;
		kstat_delete(fsp->hsfs_kstats);
		kmem_free(data, sizeof (kstat_named_t) *
		    (sizeof (hsfs_kstats)) / (sizeof (hsfs_ksindex_t)));
	}
	fsp->hsfs_kstats = NULL;
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License, Version 1.0 only
 * (the "License").  You may not use this file except in compliance
 * with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * ISO 9660 System Use Sharing Protocol extension filesystem specifications
 * Copyright (c) 1991,2000,2001 by Sun Microsystems, Inc.
 * All rights reserved.
 */

#include <sys/types.h>
#include <sys/t_lock.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysmacros.h>
#include <sys/kmem.h>
#include <sys/signal.h>
#include <sys/user.h>
#include <sys/proc.h>
#include <sys/disp.h>
#include <sys/buf.h>
#include <sys/pathname.h>
#include <sys/vfs.h>
#include <sys/vnode.h>
#include <sys/file.h>
#include <sys/uio.h>
#include <sys/conf.h>

#include <vm/page.h>

#include <sys/fs/hsfs_spec.h>
#include <sys/fs/hsfs_isospec.h>
#include <sys/fs/hsfs_node.h>
#include <sys/fs/hsfs_impl.h>
#include <sys/fs/hsfs_susp.h>
#include <sys/fs/hsfs_rrip.h>

#include <sys/statvfs.h>
#include <sys/mount.h>
#include <sys/swap.h>
#include <sys/errno.h>
#include <sys/debug.h>
#include "fs/fs_subr.h"
#include <sys/cmn_err.h>



/*
 * 	Common Signatures for all SUSP
 */
ext_signature_t  susp_signature_table[ ] = {
	SUSP_SP,	share_protocol,		/* must be first in table */
	SUSP_CE,	share_continue,		/* must be second in table */
	SUSP_PD,	share_padding,
	SUSP_ER,	share_ext_ref,
	SUSP_ST,	share_stop,
	(char *)NULL,	NULL
};

/*
 * These are global pointers referring to the above table, so the
 * positions must not change as marked on the right above.
 */
ext_signature_t	*susp_sp = &susp_signature_table[0];
ext_signature_t	*susp_ce = &susp_signature_table[1];


/*
 * ext name	version 	implemented	signature table
 *
 * the SUSP must be the first entry in the table.
 * the RRIP must be the second entry in the table. We need to be able
 * to check the RRIP bit being set, so we must know it's position.
 * RRIP_BIT is set to 2 in rrip.h
 */
extension_name_t   extension_name_table[]  =  {
	"SUSP_1991A",	SUSP_VERSION,		susp_signature_table, /* #1 */
	RRIP_ER_EXT_ID,	RRIP_EXT_VERSION, 	rrip_signature_table,
	(char *)NULL,	0,			(ext_signature_t *)NULL
};

extension_name_t	*susp_ext_name = extension_name_table;

/*
 * share_protocol()
 *
 * sig_handler() for SUSP signature "SP"
 *
 * This function looks for the "SP" signature field, which means that
 * the SUSP is supported on the current CD-ROM.  It looks for the word
 * 0xBEEF in the signature.  If that exists, the SUSP is implemented.
 * The function will then set the implemented bit in the "SUSP" entry
 * of the extention_name_table[]. If the bytes don't match, then we
 * 	return a big fat NULL and treat this as an ISO 9660 CD-ROM.
 */
uchar_t *
share_protocol(sig_args_t *sig_args_p)
{
	uchar_t *sp_ptr = sig_args_p->SUF_ptr;

	/* Let's check the check bytes */
	if (!CHECK_BYTES_OK(sp_ptr))
		return ((uchar_t *)NULL);

	/*
	 * 	Ah, we have the go ahead, so let's set the implemented bit
	 * of the SUSP in the extension_name_table[]
	 */

	if (SUSP_VERSION < SUF_VER(sp_ptr)) {
		cmn_err(CE_NOTE,
		    "System Use Sharing Protocol ver. %d:not supported\n",
		    (int)SUF_VER(sp_ptr));
		return ((uchar_t *)NULL);
	}

	SET_SUSP_BIT(sig_args_p->fsp);

	sig_args_p->fsp->hsfs_sua_off = SP_SUA_OFFSET(sp_ptr);

	return (sp_ptr + SUF_LEN(sp_ptr));
}

/*
 * share_ext_ref()
 *
 * sig_handler() for SUSP signature "ER"
 *
 * This function looks for the "ER" signature field, which lists an
 * extension that is implemented on the current CD-ROM.  The function
 * will then search through the extention_name_table[], looking for the
 * extension reference in this SUF.
 *
 * If the correct extension reference is found, and the version number
 * in the "ER" SUF is less than or equal to the version specified in
 * the extension_name_table, the implemented bit will be set to 1.
 *
 * If the version number in the "ER" field is greater than that in the
 * extension_name_table or no reference can be matched, the reference
 * will be skipped the function will return the next field.
 */
uchar_t *
share_ext_ref(sig_args_t *sig_args_p)
{
	uchar_t *er_ptr = sig_args_p->SUF_ptr;
	extension_name_t *extnp;
	int index;

	/*
	 * Find appropriate extension and signature table
	 */
	for (extnp = extension_name_table, index = 0;
	    extnp->extension_name != (char *)NULL;
	    extnp++, index++)  {
		if (strncmp(extnp->extension_name,
			    (char *)ER_ext_id(er_ptr),
			    ER_ID_LEN(er_ptr)) == 0) {
			SET_IMPL_BIT(sig_args_p->fsp, index);
		}
	}

	return (er_ptr + SUF_LEN(er_ptr));
}

/*
 * share_continue()
 *
 * sig_handler() for SUSP signature "CE"
 *
 * This function looks for the "CE" signature field. This means that
 * the SUA is continued in another block on the CD-ROM.  Because it is
 * not a requirement that this "CE" field come at the end of the SUA,
 * this function will only set up a structure containing the
 * information  needed to read the next SUA, somewhere on the disk.
 *
 * The end of the SUA is signaled by 2 NULL bytes, where the next
 * signature would have been.
 *
 * This one will be tough to implement.
 */
uchar_t *
share_continue(sig_args_t *sig_args_p)
{
	uchar_t *ce_ptr = sig_args_p->SUF_ptr;

	sig_args_p->cont_info_p->cont_lbn    = CE_BLK_LOC(ce_ptr);
	sig_args_p->cont_info_p->cont_offset = CE_OFFSET(ce_ptr);
	sig_args_p->cont_info_p->cont_len    = CE_CONT_LEN(ce_ptr);

	return (ce_ptr + SUF_LEN(ce_ptr));
}


/*
 * share_padding()
 *
 * sig_handler() for SUSP signature "PD"
 *
 * All this function is needed for is bypassing a certain number of
 * bytes.  So, we just advance past this field and we're set.
 */
uchar_t *
share_padding(sig_args_t *sig_args_p)
{
	uchar_t *pd_ptr = sig_args_p->SUF_ptr;

	return (pd_ptr +  SUF_LEN(pd_ptr));
}



/*
 * share_stop()
 *
 * sig_handler() for SUSP signature "ST"
 *
 * All this is used for is signaling the end of an SUA.
 * It fills the flag variable with the
 */
uchar_t *
share_stop(sig_args_t *sig_args_p)
{
	uchar_t *st_ptr = sig_args_p->SUF_ptr;

	sig_args_p->flags = END_OF_SUA;	/* stop parsing the SUA NOW!!!! */

	return (st_ptr +  SUF_LEN(st_ptr));
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * System Use Sharing protocol subroutines for High Sierra filesystem
 */
/*
 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#include <sys/types.h>
#include <sys/t_lock.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysmacros.h>
#include <sys/kmem.h>
#include <sys/signal.h>
#include <sys/user.h>
#include <sys/proc.h>
#include <sys/disp.h>
#include <sys/buf.h>
#include <sys/pathname.h>
#include <sys/vfs.h>
#include <sys/vnode.h>
#include <sys/file.h>
#include <sys/uio.h>
#include <sys/conf.h>

#include <vm/page.h>

#include <sys/fs/hsfs_spec.h>
#include <sys/fs/hsfs_isospec.h>
#include <sys/fs/hsfs_node.h>
#include <sys/fs/hsfs_impl.h>
#include <sys/fs/hsfs_susp.h>
#include <sys/fs/hsfs_rrip.h>

#include <sys/statvfs.h>
#include <sys/mount.h>
#include <sys/swap.h>
#include <sys/errno.h>
#include <sys/debug.h>
#include "fs/fs_subr.h"
#include <sys/cmn_err.h>

/* static declarations */
static void free_cont_area(uchar_t *);
static int get_cont_area(struct hsfs *, uchar_t **, cont_info_t *);
static int parse_signatures(sig_args_t *, int, uchar_t *, int);

/*
 * parse_sua()
 *
 * This is the main SUSP routine, that gets all the SUA areas and
 * continuations.	It calls parse_signatures() to actually interpret
 * the signature fields.
 *
 * XXX - need to implement signature searching to speed things up and
 *		which is needed for the api, which isn't done yet.
 */
int
parse_sua(
	uchar_t			*name_p,	/* location to copy name */
	int			*name_len_p,	/* location to put name len */
	int			*name_change_p,	/* flags to signal name chg */
	uchar_t			*dirp,		/* pointer to ISO dir entry */
	uint_t			last_offset,	/* last ind. in cur. dirblock */
	struct hs_direntry	*hdp,		/* loc to store dir info */
	struct hsfs		*fsp,		/* filesystem pointer */
	uchar_t			*search_sig,	/* signature to search for */
	int			search_num)	/* n^th sig to search for */
{
	uchar_t			*SUA_p = IDE_sys_use_area(dirp);
	int			SUA_len = IDE_SUA_LEN(dirp);
	uchar_t			*tmp_SUA_p = (SUA_p + fsp->hsfs_sua_off);
	int			tmp_SUA_len = (SUA_len - fsp->hsfs_sua_off);
	short			ret_val = -1;
	uchar_t			*cont_p = (uchar_t *)NULL;
	sig_args_t		sig_args;
	cont_info_t		cont_info;

	/*
	 * If there is no SUA, just return, no error
	 */

	if (SUA_len == 0)
		return (0);

	/*
	 * Underflow on the length field means there's a mismatch
	 * between sizes of SUA and ISO directory entry. This entry
	 * is corrupted, return an appropriate error.
	 */
	if (SUA_len < 0) {
		hs_log_bogus_disk_warning(fsp, HSFS_ERR_NEG_SUA_LEN, 0);
		return (SUA_EINVAL);
	}

	if ((tmp_SUA_p + tmp_SUA_len) > (dirp + last_offset)) {
		hs_log_bogus_disk_warning(fsp, HSFS_ERR_BAD_SUA_LEN, 0);
		return (SUA_EINVAL);
	}

	/*
	 * Make sure that the continuation lenth is zero, as that is
	 * the way to tell if we must grab another continuation area.
	 */
	bzero((char *)&cont_info, sizeof (cont_info));

	sig_args.dirp		= dirp;
	sig_args.name_p		= name_p;
	sig_args.name_len_p	= name_len_p;
	sig_args.SUF_ptr	= tmp_SUA_p;
	sig_args.hdp		= hdp;
	sig_args.fsp		= fsp;
	sig_args.cont_info_p	= &cont_info;
	sig_args.flags		= 0;
	sig_args.name_flags	= 0;

	/*
	 * Get ready to put in a new name.	 If no "NM" is found, then
	 * hs_namecopy will come to the rescue.  Make sure you don't
	 * have NULL names, also.
	 */
	if (name_p)
		*(name_p) = '\0';
	if (name_len_p)
		*(name_len_p) = 0;

	while (ret_val == -1) {
		switch (parse_signatures(&sig_args, tmp_SUA_len, search_sig,
		    search_num)) {
		case END_OF_SUA :
			if (cont_info.cont_len) {

				if (get_cont_area(fsp, &cont_p, &cont_info)) {
					ret_val = 1;
					goto clean_up;
				}

				sig_args.SUF_ptr =
				    cont_p + cont_info.cont_offset;

				tmp_SUA_len = cont_info.cont_len;
				cont_info.cont_len = 0;

				continue;
			}
			sig_args.flags = 0;	/* reset */
			ret_val = 0;		/* keep going */
			break;
		case SUA_NULL_POINTER:
			ret_val = SUA_NULL_POINTER;
			goto clean_up;
		case SUA_ENOMEM:
			ret_val = SUA_ENOMEM;
			goto clean_up;
		case SUA_EINVAL:
			ret_val = SUA_EINVAL;
			goto clean_up;
		case RELOC_DIR:
			ret_val = RELOC_DIR;
			goto clean_up;
		}
	}

	if (ret_val != 0)
		goto clean_up;

	if (IS_NAME_BIT_SET(sig_args.name_flags, RRIP_NAME_CHANGE))
		SET_NAME_BIT(*(name_change_p), RRIP_NAME_CHANGE);

clean_up:
	free_cont_area(cont_p);
	return (ret_val);

}

/*
 * parse_signatures()
 *
 * Find the correct handling function for the signature string that is
 * passed to this function.
 *
 * signature searching:
 *
 * The two arguments of search_sig and search_num are for finding the
 * search_num^th occurance of the signature search_sig.  This will come
 * in handy with searching for the "NM" field and is part of the api
 * for rrip (which really can be used for any extension).
 */
/*ARGSUSED*/
static int
parse_signatures(
	sig_args_t	*sig_args_p,
	int		SUA_len,
	uchar_t		*search_sig,	/* possible signature to search for */
	int		search_num)	/* n^th occurance of search_sig to */
					/*   search for */
{
	uchar_t			*sig_string = sig_args_p->SUF_ptr;
	extension_name_t	*extnp;
	ext_signature_t		*ext_sigp;
	int			impl_bit_num = 0;
	int			SUA_rem = SUA_len; /* SUA length */
					/* remaining to be parsed */

	/* This should never happen ... just so we don't panic, literally */
	if (sig_string == (uchar_t *)NULL)
		return (SUA_NULL_POINTER);

	if (SUA_len < 0)
		return (SUA_EINVAL);

	/*
	 * Until the end of SUA, search for the signatures
	 * (check for end of SUA (2 consecutive NULL bytes)) or the
	 * remaining  length of the SUA is <= 3.  The minimum signature
	 * field  is 4.
	 */

	while ((SUA_rem >= SUF_MIN_LEN) && (*sig_string != '\0') &&
		(*(sig_string + 1) != '\0')) {

		/*
		 * Find appropriate extension and signature table
		 */
		for (extnp = extension_name_table, impl_bit_num = 0;
		    extnp->extension_name != (char *)NULL;
		    extnp++, impl_bit_num++)  {

			/*
			 * look at an extension only if it is implemented
			 * on the CD-ROM
			 */
			if (!IS_IMPL_BIT_SET(sig_args_p->fsp, impl_bit_num))
				continue;

			/*
			 * Find the appropriate signature
			 */
			for (ext_sigp = extnp->signature_table;
			    ext_sigp->ext_signature != (char *)NULL;
			    ext_sigp++)  {

				if (strncmp((char *)sig_string,
					    ext_sigp->ext_signature,
					    SUF_SIG_LEN) == 0) {

					SUA_rem -= SUF_LEN(sig_string);
					if (SUA_rem < 0)
						return (END_OF_SUA);

					/*
					 * The SUA_len parameter specifies the
					 * length of the SUA that the kernel
					 * expects. There is also a length
					 * encoded in the SUA data. If they
					 * do not agree, bail out.
					 */
					if (SUA_len < SUF_LEN(sig_string)) {
						cmn_err(CE_NOTE,
					"parse_signatures: SUA length too big: "
					"expected=%d, found=%d",
						    SUA_len,
						    SUF_LEN(sig_string));
						return (SUA_EINVAL);
					}

					sig_args_p->SUF_ptr = sig_string;
					sig_string =
					    (ext_sigp->sig_handler)(sig_args_p);

					switch (sig_args_p->flags) {
					case END_OF_SUA :
						return (END_OF_SUA);
					case SUA_ENOMEM :
						return (SUA_ENOMEM);
					case SUA_EINVAL :
						return (SUA_EINVAL);
					case RELOC_DIR :
						return (RELOC_DIR);
					default :
#if NAME_SEARCH
						case NAME_CONTINUE :
							/* nothing for now */
						case NAME_CHANGE :
							/* nothing for now */
#endif
						break;
					}

					/* reset to be zero */

					sig_args_p->flags = 0;
					goto next_signature;
				}

				/* off to the next signature .... */

			} /* for ext_sigp */

		} /* for extnp	(extension parsing) */

		/*
		 * Opps, did not find this signature. We must
		 * advance on the the next signature in the SUA
		 * and pray to persumedly omniscient, omnipresent,
		 * almighty transcendental being(s) that the next
		 * record is in the susp format, or we get hosed.
		 */
		if (SUA_rem < SUF_MIN_LEN)
			return (END_OF_SUA);

		SUA_rem -= SUF_LEN(sig_string);
		sig_string += SUF_LEN(sig_string);

next_signature:
		/*
		 * Failsafe
		 */
		if (SUA_rem < SUF_MIN_LEN ||
		    sig_string == NULL || SUF_LEN(sig_string) <= 0) {
			return (END_OF_SUA);
		}

	} /* while */

	return (END_OF_SUA);
}

/*
 * hs_fill_root_dirent()
 *
 *
 * This function reads the root directory extent to get to the SUA of
 * the "." entry of the root directory.  It the checks to see if the
 * susp is implemented.
 */
void
hs_check_root_dirent(struct vnode *vp, struct hs_direntry *hdp)
{
	struct buf	*secbp;
	uchar_t		*root_ptr;
	uchar_t		*secp;
	uint_t		secno;
	offset_t	secoff;
	sig_args_t	sig_args;
	struct hsfs	*fsp;
	int		error;

	if (vp->v_type != VDIR) {
		cmn_err(CE_NOTE,
		    "hs_check_root_dirent: vp (0x%p) not a directory",
		    (void *)vp);
		return;
	}

	bzero((caddr_t)&sig_args, sizeof (sig_args));

	fsp = VFS_TO_HSFS(vp->v_vfsp);
	secno = LBN_TO_SEC(hdp->ext_lbn+hdp->xar_len, vp->v_vfsp);
	secoff = LBN_TO_BYTE(hdp->ext_lbn+hdp->xar_len, vp->v_vfsp) &
	    MAXHSOFFSET;
	secbp = bread(fsp->hsfs_devvp->v_rdev, secno * 4, HS_SECTOR_SIZE);
	error = geterror(secbp);

	if (error != 0) {
		cmn_err(CE_NOTE,
		    "hs_check_root_dirent: bread: error=(%d)", error);
		goto end;
	}

	secp = (uchar_t *)secbp->b_un.b_addr;
	root_ptr = &secp[secoff];

	/* quick check */
	if (hdp->ext_lbn != HDE_EXT_LBN(root_ptr)) {
		cmn_err(CE_NOTE, "hs_check_root_dirent: dirent not match\n");
		/* keep on going */
	}

	/*
	 * Here, we know that the "." entry is the first in the sector
	 * just read (ISO 9660).  Let's now check for the sharing
	 * protocol and set call the susp sig_handler() if we should.
	 * Then we run through the hs_parsedir() function to catch all
	 * the other possibilities of SUSP fields and continuations.
	 *
	 * If there is no SUA area, just return, and assume ISO.
	 *
	 * If the SUA area length is invalid (negative, due to a mismatch
	 * between dirent size and SUA size), return and hope for the best.
	 */

	if (IDE_SUA_LEN(root_ptr) <= 0)
		goto end;

	if (strncmp(SUSP_SP, (char *)IDE_sys_use_area(root_ptr),
	    SUF_SIG_LEN) == 0) {
		/*
		 * We have a match of the sharing signature, so let's
		 * call the sig_handler to do what is necessary. We can
		 * ignore the return value, as implemented bits are set.
		 */
		sig_args.SUF_ptr = IDE_sys_use_area(root_ptr);
		sig_args.fsp	 = fsp;

		if ((susp_sp->sig_handler)(&sig_args) == (uchar_t *)NULL) {
			goto end;
		}
	} else {
		goto end;
	}

	/*
	 * If the "ER" signature in the root directory is past any non SU
	 * signature, the Rock Ridge signatures will be ignored. This happens
	 * e.g. for filesystems created by mkisofs. In this case,
	 * IS_RRIP_IMPLEMENTED(fsp) will return 0 when the "ER" signature is
	 * parsed. Unfortunately, the results of this run will be cached for
	 * the root vnode. The solution is to run hs_parsedir() a second time
	 * for the root directory.
	 */
	if (hs_parsedir(fsp, root_ptr, hdp, (char *)NULL, (int *)NULL,
	    HS_SECTOR_SIZE - secoff) == 0) {
		(void) hs_parsedir(fsp, root_ptr, hdp, (char *)NULL,
		    (int *)NULL, HS_SECTOR_SIZE - secoff);
	}

	/*
	 * If we did not get at least 1 extension, let's assume ISO and
	 * NULL out the implementation bits.
	 */
	if (fsp->hsfs_ext_impl <= 1L)
		fsp->hsfs_ext_impl = 0L;

end:
	brelse(secbp);
}


/*
 * get_cont_area()
 *
 * This function allocates a memory block, if necessary, and reads the
 * continuation area into the allocated space.
 *
 * Return value : 	0 if the read and allocation went OK.
 * 			1 if there was an error.
 */
static int
get_cont_area(struct hsfs *fsp, uchar_t **buf_pp, cont_info_t *cont_info_p)
{
	struct buf	*secbp;
	int		error;
	uint_t		secno;

	/*
	 * Guard against invalid continuation area records.
	 * Both cont_offset and cont_len must be no longer than
	 * HS_SECTOR_SIZE. If they are, return an error.
	 */
	if (cont_info_p->cont_offset > HS_SECTOR_SIZE ||
	    cont_info_p->cont_len > HS_SECTOR_SIZE) {
		cmn_err(CE_NOTE, "get_cont_area: invalid offset/length");
		return (1);
	}

	if (*buf_pp == (uchar_t *)NULL)
		*buf_pp = kmem_alloc((size_t)HS_SECTOR_SIZE, KM_SLEEP);

	secno = (uint_t)LBN_TO_SEC(cont_info_p->cont_lbn, fsp->hsfs_vfs);
	secbp = bread(fsp->hsfs_devvp->v_rdev, secno * 4, HS_SECTOR_SIZE);
	error = geterror(secbp);

	if (error != 0) {
		cmn_err(CE_NOTE, "get_cont_area: bread: error=(%d)", error);
		brelse(secbp);
		return (1);
	}

	/*
	 * This continuation area does not extend into the next sector
	 * so just copy the data to the buffer.
	 */
	if ((cont_info_p->cont_offset + cont_info_p->cont_len) <=
	    HS_SECTOR_SIZE) {
		bcopy(secbp->b_un.b_addr, (char *)*buf_pp, HS_SECTOR_SIZE);
	}
	/*
	 * This continuation area extends into the next sector so we
	 * need to do some dancing:
	 *
	 * - zero the return buffer so nothing random is returned
	 * - copy the partial data to the *beginning* of the return buffer
	 * - release the first sector's buffer
	 * - read the next sector
	 * - copy the remainder of the data to the return buffer
	 */
	else {
		uint_t	partial_size;

		bzero((char *)*buf_pp, HS_SECTOR_SIZE);
		partial_size = HS_SECTOR_SIZE - cont_info_p->cont_offset;
		bcopy(&secbp->b_un.b_addr[cont_info_p->cont_offset],
		    (char *)*buf_pp, partial_size);
		cont_info_p->cont_offset = 0;
		brelse(secbp);

		secbp = bread(fsp->hsfs_devvp->v_rdev, (secno + 1) * 4,
		    HS_SECTOR_SIZE);
		error = geterror(secbp);
		if (error != 0) {
			cmn_err(CE_NOTE, "get_cont_area: bread(2): error=(%d)",
			    error);
			brelse(secbp);
			return (1);
		}
		bcopy(secbp->b_un.b_addr, (char *)&(*buf_pp)[partial_size],
		    cont_info_p->cont_len - partial_size);
	}

	brelse(secbp);
	return (0);
}


/*
 * free_cont_area
 *
 * simple function to just free up memory, if it exists
 *
 */
static void
free_cont_area(uchar_t *cont_p)
{
	if (cont_p)
		(void) kmem_free((caddr_t)cont_p, (size_t)HS_SECTOR_SIZE);
	cont_p = (uchar_t *)NULL;
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2011 Bayard G. Bell. All rights reserved.
 * Copyright 2013 Joyent, Inc. All rights reserved.
 * Copyright (c) 2017 by Delphix. All rights reserved.
 */

/*
 * VFS operations for High Sierra filesystem
 */

#include <sys/types.h>
#include <sys/isa_defs.h>
#include <sys/t_lock.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysmacros.h>
#include <sys/kmem.h>
#include <sys/signal.h>
#include <sys/user.h>
#include <sys/proc.h>
#include <sys/disp.h>
#include <sys/buf.h>
#include <sys/pathname.h>
#include <sys/vfs.h>
#include <sys/vfs_opreg.h>
#include <sys/vnode.h>
#include <sys/file.h>
#include <sys/uio.h>
#include <sys/conf.h>
#include <sys/policy.h>

#include <vm/page.h>

#include <sys/fs/snode.h>
#include <sys/fs/hsfs_spec.h>
#include <sys/fs/hsfs_isospec.h>
#include <sys/fs/hsfs_node.h>
#include <sys/fs/hsfs_impl.h>
#include <sys/fs/hsfs_susp.h>
#include <sys/fs/hsfs_rrip.h>

#include <sys/statvfs.h>
#include <sys/mount.h>
#include <sys/mntent.h>
#include <sys/swap.h>
#include <sys/errno.h>
#include <sys/debug.h>
#include "fs/fs_subr.h"
#include <sys/cmn_err.h>
#include <sys/bootconf.h>

#include <sys/sdt.h>

/*
 * These are needed for the CDROMREADOFFSET Code
 */
#include <sys/cdio.h>
#include <sys/sunddi.h>

#define	HSFS_CLKSET

#include <sys/modctl.h>

/*
 * Options for mount.
 */
#define	HOPT_GLOBAL	MNTOPT_GLOBAL
#define	HOPT_NOGLOBAL	MNTOPT_NOGLOBAL
#define	HOPT_MAPLCASE	"maplcase"
#define	HOPT_NOMAPLCASE	"nomaplcase"
#define	HOPT_NOTRAILDOT	"notraildot"
#define	HOPT_TRAILDOT	"traildot"
#define	HOPT_NRR	"nrr"
#define	HOPT_RR		"rr"
#define	HOPT_JOLIET	"joliet"
#define	HOPT_NOJOLIET	"nojoliet"
#define	HOPT_JOLIETLONG	"jolietlong"
#define	HOPT_VERS2	"vers2"
#define	HOPT_NOVERS2	"novers2"
#define	HOPT_RO		MNTOPT_RO

static char *global_cancel[] = { HOPT_NOGLOBAL, NULL };
static char *noglobal_cancel[] = { HOPT_GLOBAL, NULL };
static char *mapl_cancel[] = { HOPT_NOMAPLCASE, NULL };
static char *nomapl_cancel[] = { HOPT_MAPLCASE, NULL };
static char *ro_cancel[] = { MNTOPT_RW, NULL };
static char *rr_cancel[] = { HOPT_NRR, NULL };
static char *nrr_cancel[] = { HOPT_RR, NULL };
static char *joliet_cancel[] = { HOPT_NOJOLIET, NULL };
static char *nojoliet_cancel[] = { HOPT_JOLIET, NULL };
static char *vers2_cancel[] = { HOPT_NOVERS2, NULL };
static char *novers2_cancel[] = { HOPT_VERS2, NULL };
static char *trail_cancel[] = { HOPT_NOTRAILDOT, NULL };
static char *notrail_cancel[] = { HOPT_TRAILDOT, NULL };

static mntopt_t hsfs_options[] = {
	{ HOPT_GLOBAL, global_cancel, NULL, 0, NULL },
	{ HOPT_NOGLOBAL, noglobal_cancel, NULL, MO_DEFAULT, NULL },
	{ HOPT_MAPLCASE, mapl_cancel, NULL, MO_DEFAULT, NULL },
	{ HOPT_NOMAPLCASE, nomapl_cancel, NULL, 0, NULL },
	{ HOPT_RO, ro_cancel, NULL, MO_DEFAULT, NULL },
	{ HOPT_RR, rr_cancel, NULL, MO_DEFAULT, NULL },
	{ HOPT_NRR, nrr_cancel, NULL, 0, NULL },
	{ HOPT_JOLIET, joliet_cancel, NULL, 0, NULL },
	{ HOPT_NOJOLIET, nojoliet_cancel, NULL, 0, NULL },
	{ HOPT_JOLIETLONG, NULL, NULL, 0, NULL },
	{ HOPT_VERS2, vers2_cancel, NULL, 0, NULL },
	{ HOPT_NOVERS2, novers2_cancel, NULL, 0, NULL },
	{ HOPT_TRAILDOT, trail_cancel, NULL, MO_DEFAULT, NULL },
	{ HOPT_NOTRAILDOT, notrail_cancel, NULL, 0, NULL },
	{ "sector", NULL, "0", MO_HASVALUE, NULL},
};

static mntopts_t hsfs_proto_opttbl = {
	sizeof (hsfs_options) / sizeof (mntopt_t),
	hsfs_options
};

/*
 * Indicates whether to enable the I/O scheduling and readahead logic
 * 1 - Enable, 0 - Do not Enable.
 * Debugging purposes.
 */
int do_schedio = 1;
static int hsfsfstype;
static int hsfsinit(int, char *);

static vfsdef_t vfw = {
	VFSDEF_VERSION,
	"hsfs",
	hsfsinit,
	/* We don't suppport remounting */
	VSW_HASPROTO|VSW_STATS|VSW_CANLOFI|VSW_MOUNTDEV,
	&hsfs_proto_opttbl
};

static struct modlfs modlfs = {
	&mod_fsops, "filesystem for HSFS", &vfw
};

static struct modlinkage modlinkage = {
	MODREV_1, (void *)&modlfs, NULL
};

extern void hsched_init_caches(void);
extern void hsched_fini_caches(void);


int
_init(void)
{
	return (mod_install(&modlinkage));
}

int
_fini(void)
{
	int	error;

	error = mod_remove(&modlinkage);

	DTRACE_PROBE1(mod_remove, int, error);

	if (error)
		return (error);

	mutex_destroy(&hs_mounttab_lock);

	/*
	 * Tear down the operations vectors
	 */
	(void) vfs_freevfsops_by_type(hsfsfstype);
	vn_freevnodeops(hsfs_vnodeops);

	hs_fini_hsnode_cache();
	hsched_fini_caches();
	return (0);
}

int
_info(struct modinfo *modinfop)
{
	return (mod_info(&modlinkage, modinfop));
}

#define	BDEVFLAG(dev)	((devopsp[getmajor(dev)])->devo_cb_ops->cb_flag)

kmutex_t hs_mounttab_lock;
struct hsfs *hs_mounttab = NULL;

/* default mode, uid, gid */
mode_t hsfs_default_mode = 0555;
uid_t hsfs_default_uid = 0;
gid_t hsfs_default_gid = 3;

extern void hsched_init(struct hsfs *fsp, int fsid,
					struct modlinkage *modlinkage);
extern void hsched_fini(struct hsfs_queue *hqueue);
extern void hsfs_init_kstats(struct hsfs *fsp, int fsid);
extern void hsfs_fini_kstats(struct hsfs *fsp);

static int hsfs_mount(struct vfs *vfsp, struct vnode *mvp,
	struct mounta *uap, struct cred *cr);
static int hsfs_unmount(struct vfs *vfsp, int, struct cred *cr);
static int hsfs_root(struct vfs *vfsp, struct vnode **vpp);
static int hsfs_statvfs(struct vfs *vfsp, struct statvfs64 *sbp);
static int hsfs_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp);
static int hsfs_mountroot(struct vfs *, enum whymountroot);

static int hs_mountfs(struct vfs *vfsp, dev_t dev, char *path,
	mode_t mode, int flags, struct cred *cr, int isroot);
static int hs_getrootvp(struct vfs *vfsp, struct hsfs *fsp, size_t pathsize);
static int hs_findhsvol(struct hsfs *fsp, struct vnode *vp,
	struct hs_volume *hvp);
static int hs_parsehsvol(struct hsfs *fsp, uchar_t *volp,
	struct hs_volume *hvp);
static int hs_findisovol(struct hsfs *fsp, struct vnode *vp,
	struct hs_volume *hvp,
	struct hs_volume *svp,
	struct hs_volume *jvp);
static int hs_joliet_level(uchar_t *volp);
static int hs_parseisovol(struct hsfs *fsp, uchar_t *volp,
	struct hs_volume *hvp);
static void hs_copylabel(struct hs_volume *, unsigned char *, int);
static int hs_getmdev(struct vfs *, char *fspec, int flags, dev_t *pdev,
	mode_t *mode, cred_t *cr);
static int hs_findvoldesc(dev_t rdev, int desc_sec);

static int
hsfsinit(int fstype, char *name)
{
	static const fs_operation_def_t hsfs_vfsops_template[] = {
		VFSNAME_MOUNT,		{ .vfs_mount = hsfs_mount },
		VFSNAME_UNMOUNT,	{ .vfs_unmount = hsfs_unmount },
		VFSNAME_ROOT,		{ .vfs_root = hsfs_root },
		VFSNAME_STATVFS,	{ .vfs_statvfs = hsfs_statvfs },
		VFSNAME_VGET,		{ .vfs_vget = hsfs_vget },
		VFSNAME_MOUNTROOT,	{ .vfs_mountroot = hsfs_mountroot },
		NULL,			NULL
	};
	int error;

	error = vfs_setfsops(fstype, hsfs_vfsops_template, NULL);
	if (error != 0) {
		cmn_err(CE_WARN, "hsfsinit: bad vfs ops template");
		return (error);
	}

	error = vn_make_ops(name, hsfs_vnodeops_template, &hsfs_vnodeops);
	if (error != 0) {
		(void) vfs_freevfsops_by_type(fstype);
		cmn_err(CE_WARN, "hsfsinit: bad vnode ops template");
		return (error);
	}

	hsfsfstype = fstype;
	mutex_init(&hs_mounttab_lock, NULL, MUTEX_DEFAULT, NULL);
	hs_init_hsnode_cache();
	hsched_init_caches();
	return (0);
}

/*ARGSUSED*/
static int
hsfs_mount(struct vfs *vfsp, struct vnode *mvp,
    struct mounta *uap, struct cred *cr)
{
	int		vnode_busy;
	dev_t		dev;
	struct pathname dpn;
	int		error;
	mode_t		mode;
	int		flags;	/* this will hold the mount specific data */

	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
		return (error);

	if (mvp->v_type != VDIR)
		return (ENOTDIR);

	/* mount option must be read only, else mount will be rejected */
	if (!(uap->flags & MS_RDONLY))
		return (EROFS);

	/*
	 * We already told the framework that we don't support remounting.
	 */
	ASSERT(!(uap->flags & MS_REMOUNT));

	mutex_enter(&mvp->v_lock);
	vnode_busy = (mvp->v_count != 1) || (mvp->v_flag & VROOT);
	mutex_exit(&mvp->v_lock);

	if ((uap->flags & MS_OVERLAY) == 0 && vnode_busy) {
		return (EBUSY);
	}

	/*
	 * Check for the options that actually affect things
	 * at our level.
	 */
	flags = 0;
	if (vfs_optionisset(vfsp, HOPT_NOMAPLCASE, NULL))
		flags |= HSFSMNT_NOMAPLCASE;
	if (vfs_optionisset(vfsp, HOPT_NOTRAILDOT, NULL))
		flags |= HSFSMNT_NOTRAILDOT;
	if (vfs_optionisset(vfsp, HOPT_NRR, NULL))
		flags |= HSFSMNT_NORRIP;
	if (vfs_optionisset(vfsp, HOPT_NOJOLIET, NULL))
		flags |= HSFSMNT_NOJOLIET;
	if (vfs_optionisset(vfsp, HOPT_JOLIETLONG, NULL))
		flags |= HSFSMNT_JOLIETLONG;
	if (vfs_optionisset(vfsp, HOPT_NOVERS2, NULL))
		flags |= HSFSMNT_NOVERS2;

	error = pn_get(uap->dir, (uap->flags & MS_SYSSPACE) ?
	    UIO_SYSSPACE : UIO_USERSPACE, &dpn);
	if (error)
		return (error);

	error = hs_getmdev(vfsp, uap->spec, uap->flags, &dev, &mode, cr);
	if (error != 0) {
		pn_free(&dpn);
		return (error);
	}

	/*
	 * If the device is a tape, return error
	 */
	if ((BDEVFLAG(dev) & D_TAPE) == D_TAPE)  {
		pn_free(&dpn);
		return (ENOTBLK);
	}

	/*
	 * Mount the filesystem.
	 */
	error = hs_mountfs(vfsp, dev, dpn.pn_path, mode, flags, cr, 0);
	pn_free(&dpn);
	return (error);
}

/*ARGSUSED*/
static int
hsfs_unmount(
	struct vfs *vfsp,
	int flag,
	struct cred *cr)
{
	struct hsfs **tspp;
	struct hsfs *fsp;

	if (secpolicy_fs_unmount(cr, vfsp) != 0)
		return (EPERM);

	/*
	 * forced unmount is not supported by this file system
	 * and thus, ENOTSUP is being returned.
	 */
	if (flag & MS_FORCE)
		return (ENOTSUP);

	fsp = VFS_TO_HSFS(vfsp);

	if (fsp->hsfs_rootvp->v_count != 1)
		return (EBUSY);

	/* destroy all old pages and hsnodes for this vfs */
	if (hs_synchash(vfsp))
		return (EBUSY);

	mutex_enter(&hs_mounttab_lock);
	for (tspp = &hs_mounttab; *tspp != NULL; tspp = &(*tspp)->hsfs_next) {
		if (*tspp == fsp)
			break;
	}
	if (*tspp == NULL) {
		mutex_exit(&hs_mounttab_lock);
		panic("hsfs_unmount: vfs not mounted?");
		/*NOTREACHED*/
	}

	*tspp = fsp->hsfs_next;

	mutex_exit(&hs_mounttab_lock);

	hsfs_fini_kstats(fsp);
	(void) VOP_CLOSE(fsp->hsfs_devvp, FREAD, 1, (offset_t)0, cr, NULL);
	VN_RELE(fsp->hsfs_devvp);
	/* free path table space */
	if (fsp->hsfs_ptbl != NULL)
		kmem_free(fsp->hsfs_ptbl, (size_t)fsp->hsfs_vol.ptbl_len);
	/* free path table index table */
	if (fsp->hsfs_ptbl_idx != NULL)
		kmem_free(fsp->hsfs_ptbl_idx, (size_t)
		    (fsp->hsfs_ptbl_idx_size * sizeof (struct ptable_idx)));

	/* free "mounted on" pathame */
	if (fsp->hsfs_fsmnt != NULL)
		kmem_free(fsp->hsfs_fsmnt, strlen(fsp->hsfs_fsmnt) + 1);

	hsched_fini(fsp->hqueue);
	kmem_free(fsp->hqueue, sizeof (struct hsfs_queue));

	mutex_destroy(&fsp->hsfs_free_lock);
	rw_destroy(&fsp->hsfs_hash_lock);

	kmem_free(fsp, sizeof (*fsp));
	return (0);
}

/*ARGSUSED*/
static int
hsfs_root(struct vfs *vfsp, struct vnode **vpp)
{
	*vpp = (VFS_TO_HSFS(vfsp))->hsfs_rootvp;
	VN_HOLD(*vpp);
	return (0);
}

/*ARGSUSED*/
static int
hsfs_statvfs(struct vfs *vfsp, struct statvfs64 *sbp)
{
	struct hsfs *fsp;
	dev32_t d32;

	fsp = VFS_TO_HSFS(vfsp);
	if (fsp->hsfs_magic != HSFS_MAGIC)
		return (EINVAL);
	bzero(sbp, sizeof (*sbp));
	sbp->f_bsize = vfsp->vfs_bsize;
	sbp->f_frsize = sbp->f_bsize; /* no fragment, same as block size */
	sbp->f_blocks = (fsblkcnt64_t)fsp->hsfs_vol.vol_size;

	sbp->f_bfree = (fsblkcnt64_t)0;
	sbp->f_bavail = (fsblkcnt64_t)0;
	sbp->f_files = (fsfilcnt64_t)-1;
	sbp->f_ffree = (fsfilcnt64_t)0;
	sbp->f_favail = (fsfilcnt64_t)0;
	(void) cmpldev(&d32, vfsp->vfs_dev);
	sbp->f_fsid = d32;
	(void) strcpy(sbp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
	sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
	sbp->f_namemax = fsp->hsfs_namemax;
	(void) strcpy(sbp->f_fstr, fsp->hsfs_vol.vol_id);

	return (0);
}

/*
 * Previously nodeid was declared as uint32_t. This has been changed
 * to conform better with the ISO9660 standard. The standard states that
 * a LBN can be a 32 bit number, as the MAKE_NODEID macro shifts this
 * LBN 11 places left (LBN_TO_BYTE) and then shifts the result 5 right
 * (divide by 32) we are left with the potential of an overflow if
 * confined to a 32 bit value.
 */

static int
hsfs_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp)
{
	struct hsfid *fid;
	struct hsfs *fsp;
	ino64_t nodeid;
	int error;

	fsp = (struct hsfs *)VFS_TO_HSFS(vfsp);
	fid = (struct hsfid *)fidp;

	/*
	 * Look for vnode on hashlist.
	 * If found, it's now active and the refcnt was incremented.
	 */

	rw_enter(&fsp->hsfs_hash_lock, RW_READER);

	nodeid = fid->hf_ino;

	if ((*vpp = hs_findhash(nodeid, fid->hf_dir_lbn,
	    (uint_t)fid->hf_dir_off, vfsp)) == NULL) {
		/*
		 * Not in cache, so we need to remake it.
		 * hs_remakenode() will read the directory entry
		 * and then check again to see if anyone else has
		 * put it in the cache.
		 */
		rw_exit(&fsp->hsfs_hash_lock);
		error = hs_remakenode(fid->hf_dir_lbn, (uint_t)fid->hf_dir_off,
		    vfsp, vpp);
		return (error);
	}
	rw_exit(&fsp->hsfs_hash_lock);
	return (0);
}


#define	CHECKSUM_SIZE				(64 * 1024)

/*
 * Compute a CD-ROM fsid by checksumming the first 64K of data on the CD
 * We use the 'fsp' argument to determine the location of the root
 * directory entry, and we start reading from there.
 */
static int
compute_cdrom_id(struct hsfs *fsp, vnode_t *devvp)
{
	uint_t		secno;
	struct hs_volume *hsvp = &fsp->hsfs_vol;
	struct buf	*bp;
	int		error;
	int		fsid;

	secno = hsvp->root_dir.ext_lbn >> hsvp->lbn_secshift;
	bp = bread(devvp->v_rdev, secno * 4, CHECKSUM_SIZE);
	error = geterror(bp);

	/*
	 * An error on read or a partial read means we asked
	 * for a nonexistant/corrupted piece of the device
	 * (including past-the-end of the media). Don't
	 * try to use the checksumming method then.
	 */
	if (!error && bp->b_bcount == CHECKSUM_SIZE) {
		int *ibuf = (int *)bp->b_un.b_addr;
		int i;

		fsid = 0;

		for (i = 0; i < CHECKSUM_SIZE / sizeof (int); i++)
			fsid ^= ibuf[ i ];
	} else {
		/*
		 * Fallback - use creation date
		 */
		fsid = hsvp->cre_date.tv_sec;
	}

	brelse(bp);

	return (fsid);
}


/*ARGSUSED*/
static int
hs_mountfs(
	struct vfs	*vfsp,
	dev_t		dev,
	char		*path,
	mode_t		mode,
	int		mount_flags,
	struct cred	*cr,
	int		isroot)
{
	struct vnode	*devvp;
	struct hsfs	*tsp;
	struct hsfs	*fsp = NULL;
	struct vattr	vap;
	struct hsnode	*hp;
	int		error;
	struct timeval	tv;
	int		fsid;
	int		use_rrip;
	int		use_vers2;
	int		use_joliet;
	int		has_rrip = 0;
	int		has_vers2 = 0;
	int		has_joliet = 0;
	int		force_rrip_off;
	int		force_vers2_off;
	int		force_joliet_off;
	size_t		pathbufsz = strlen(path) + 1;
	int		redo_rootvp;

	struct hs_volume *svp = NULL;	/* Supplemental VD for ISO-9660:1999 */
	struct hs_volume *jvp = NULL;	/* Joliet VD */

	/*
	 * The rules for which extension will be used are:
	 * 1. No specific mount options given:
	 *	- use rrip if available
	 *	- use ISO9660:1999 if available
	 *	- use joliet if available.
	 * 2. rrip/ISO9660:1999/joliet explicitly disabled via mount option:
	 *	- use next "lower" extension
	 * 3. joliet/ISO9660:1999/rrip explicitly requested via mount option:
	 *	- disable rrip support even if available
	 *	- disable IOS9660:1999 support even if available
	 *
	 * We need to adjust these flags as we discover the extensions
	 * present. See below. These are just the starting values.
	 */
	use_rrip = (mount_flags & HSFSMNT_NORRIP) == 0;
	use_vers2 = (mount_flags & HSFSMNT_NOVERS2) == 0;
	use_joliet = (mount_flags & HSFSMNT_NOJOLIET) == 0;

	/*
	 * Open the device
	 */
	devvp = makespecvp(dev, VBLK);
	ASSERT(devvp != 0);

	/*
	 * Open the target device (file) for read only.
	 */
	if (error = VOP_OPEN(&devvp, FREAD, cr, NULL)) {
		VN_RELE(devvp);
		return (error);
	}

	/*
	 * Refuse to go any further if this
	 * device is being used for swapping
	 */
	if (IS_SWAPVP(common_specvp(devvp))) {
		error = EBUSY;
		goto cleanup;
	}

	vap.va_mask = AT_SIZE;
	if ((error = VOP_GETATTR(devvp, &vap, ATTR_COMM, cr, NULL)) != 0) {
		cmn_err(CE_NOTE, "Cannot get attributes of the CD-ROM driver");
		goto cleanup;
	}

	/*
	 * Make sure we have a nonzero size partition.
	 * The current version of the SD driver will *not* fail the open
	 * of such a partition so we have to check for it here.
	 */
	if (vap.va_size == 0) {
		error = ENXIO;
		goto cleanup;
	}

	/*
	 * Init a new hsfs structure.
	 */
	fsp = kmem_zalloc(sizeof (*fsp), KM_SLEEP);
	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
	jvp = kmem_zalloc(sizeof (*jvp), KM_SLEEP);

	/* hardwire perms, uid, gid */
	fsp->hsfs_vol.vol_uid = hsfs_default_uid;
	fsp->hsfs_vol.vol_gid =  hsfs_default_gid;
	fsp->hsfs_vol.vol_prot = hsfs_default_mode;
	svp->vol_uid = hsfs_default_uid;
	svp->vol_gid =  hsfs_default_gid;
	svp->vol_prot = hsfs_default_mode;
	jvp->vol_uid = hsfs_default_uid;
	jvp->vol_gid =  hsfs_default_gid;
	jvp->vol_prot = hsfs_default_mode;

	/*
	 * Look for a Standard File Structure Volume Descriptor,
	 * of which there must be at least one.
	 * If found, check for volume size consistency.
	 *
	 * If svp->lbn_size is != 0, we did find a ISO-9660:1999 SVD
	 * If jvp->lbn_size is != 0, we did find a Joliet SVD.
	 */
	fsp->hsfs_namemax = ISO_FILE_NAMELEN;
	fsp->hsfs_namelen = ISO_FILE_NAMELEN;
	error = hs_findisovol(fsp, devvp, &fsp->hsfs_vol, svp, jvp);
	if (error == EINVAL) /* no iso 9660 - try high sierra ... */
		error = hs_findhsvol(fsp, devvp, &fsp->hsfs_vol);

	if (error)
		goto cleanup;

	DTRACE_PROBE4(findvol,
	    struct hsfs *, fsp,
	    struct hs_volume *, &fsp->hsfs_vol,
	    struct hs_volume *, svp,
	    struct hs_volume *, jvp);

	/*
	 * Generate a file system ID from the CD-ROM,
	 * and check it for uniqueness.
	 *
	 * What we are aiming for is some chance of integrity
	 * across disk change.  That is, if a client has an fhandle,
	 * it will be valid as long as the same disk is mounted.
	 */
	fsid = compute_cdrom_id(fsp, devvp);

	mutex_enter(&hs_mounttab_lock);

	if (fsid == 0 || fsid == -1) {
		uniqtime(&tv);
		fsid = tv.tv_sec;
	} else	/* make sure that the fsid is unique */
		for (tsp = hs_mounttab; tsp != NULL; tsp = tsp->hsfs_next) {
			if (fsid == tsp->hsfs_vfs->vfs_fsid.val[0]) {
				uniqtime(&tv);
				fsid = tv.tv_sec;
				break;
			}
		}

	fsp->hsfs_next = hs_mounttab;
	hs_mounttab = fsp;

	fsp->hsfs_devvp = devvp;
	fsp->hsfs_vfs = vfsp;
	fsp->hsfs_fsmnt = kmem_alloc(pathbufsz, KM_SLEEP);
	(void) strlcpy(fsp->hsfs_fsmnt, path, pathbufsz);

	mutex_init(&fsp->hsfs_free_lock, NULL, MUTEX_DEFAULT, NULL);
	rw_init(&fsp->hsfs_hash_lock, NULL, RW_DEFAULT, NULL);

	vfsp->vfs_data = (caddr_t)fsp;
	vfsp->vfs_dev = dev;
	vfsp->vfs_fstype = hsfsfstype;
	vfsp->vfs_bsize = fsp->hsfs_vol.lbn_size; /* %% */
	vfsp->vfs_fsid.val[0] = fsid;
	vfsp->vfs_fsid.val[1] =  hsfsfstype;

	if (!hs_getrootvp(vfsp, fsp, pathbufsz)) {
		DTRACE_PROBE1(rootvp__failed, struct hsfs *, fsp);
		error = EINVAL;
		goto cleanup;
	}
	DTRACE_PROBE1(rootvp, struct hsfs *, fsp);

	/*
	 * Attempt to discover a RR extension.
	 */
	if (use_rrip) {
		hp = VTOH(fsp->hsfs_rootvp);
		hs_check_root_dirent(fsp->hsfs_rootvp, &(hp->hs_dirent));
	}

	has_rrip = IS_RRIP_IMPLEMENTED(fsp);
	has_vers2 = (svp->lbn_size != 0);
	has_joliet = (jvp->lbn_size != 0);

	DTRACE_PROBE4(voltype__suggested, struct hsfs *, fsp,
	    int, use_rrip, int, use_vers2, int, use_joliet);

	DTRACE_PROBE4(voltype__actual, struct hsfs *, fsp,
	    int, has_rrip, int, has_vers2, int, has_joliet);

	DTRACE_PROBE4(findvol,
	    struct hsfs *, fsp,
	    struct hs_volume *, &fsp->hsfs_vol,
	    struct hs_volume *, svp,
	    struct hs_volume *, jvp);

	force_rrip_off = !use_rrip ||
	    (vfs_optionisset(vfsp, HOPT_JOLIET, NULL) && has_joliet) ||
	    (vfs_optionisset(vfsp, HOPT_VERS2, NULL) && has_vers2);

	force_vers2_off = !use_vers2 ||
	    (vfs_optionisset(vfsp, HOPT_JOLIET, NULL) && has_joliet);

	force_joliet_off = !use_joliet;

	DTRACE_PROBE4(voltype__force_off, struct hsfs *, fsp,
	    int, force_rrip_off, int, force_vers2_off, int, force_joliet_off);

	/*
	 * At the moment, we have references of all three possible
	 * extensions (RR, ISO9660:1999/v2 and Joliet) if present.
	 *
	 * The "active" volume descriptor is RRIP (or ISO9660:1988).
	 * We now switch to the user-requested one.
	 */
	redo_rootvp = 0;

	if (force_rrip_off || !has_rrip) {
		if (has_vers2 && !force_vers2_off) {
			VN_RELE(fsp->hsfs_rootvp);
			bcopy(svp, &fsp->hsfs_vol, sizeof (struct hs_volume));
			fsp->hsfs_vol_type = HS_VOL_TYPE_ISO_V2;
			vfsp->vfs_bsize = fsp->hsfs_vol.lbn_size;
			redo_rootvp = 1;
			has_joliet = 0;
		} else if (has_joliet && !force_joliet_off) {
			VN_RELE(fsp->hsfs_rootvp);
			bcopy(jvp, &fsp->hsfs_vol, sizeof (struct hs_volume));
			fsp->hsfs_vol_type = HS_VOL_TYPE_JOLIET;
			vfsp->vfs_bsize = fsp->hsfs_vol.lbn_size;
			redo_rootvp = 1;
			has_vers2 = 0;
		}
	}

	if (redo_rootvp) {
		/*
		 * Make sure not to use Rock Ridge.
		 */
		UNSET_IMPL_BIT(fsp, RRIP_BIT);
		UNSET_SUSP_BIT(fsp);
		has_rrip = 0;

		if (!hs_getrootvp(vfsp, fsp, pathbufsz)) {
			DTRACE_PROBE1(rootvp__failed, struct hsfs *, fsp);
			error = EINVAL;
			goto cleanup;
		}
		DTRACE_PROBE1(rootvp, struct hsfs *, fsp);
	}
	if (IS_RRIP_IMPLEMENTED(fsp)) {
		has_vers2 = 0;
		has_joliet = 0;
	}
	if (force_vers2_off)
		has_vers2 = 0;
	if (force_joliet_off)
		has_joliet = 0;
	DTRACE_PROBE4(voltype__taken, struct hsfs *, fsp,
	    int, has_rrip, int, has_vers2, int, has_joliet);

	/*
	 * mark root node as VROOT
	 */
	fsp->hsfs_rootvp->v_flag |= VROOT;

	/* Here we take care of some special case stuff for mountroot */
	if (isroot) {
		fsp->hsfs_rootvp->v_rdev = devvp->v_rdev;
		rootvp = fsp->hsfs_rootvp;
	}

	if (IS_RRIP_IMPLEMENTED(fsp)) {
		/*
		 * if RRIP, don't copy NOMAPLCASE or NOTRAILDOT to hsfs_flags
		 */
		mount_flags &= ~(HSFSMNT_NOMAPLCASE | HSFSMNT_NOTRAILDOT);

		fsp->hsfs_namemax = RRIP_FILE_NAMELEN;
		fsp->hsfs_namelen = RRIP_FILE_NAMELEN;

		ASSERT(vfs_optionisset(vfsp, HOPT_RR, NULL));
		vfs_clearmntopt(vfsp, HOPT_VERS2);
		vfs_clearmntopt(vfsp, HOPT_JOLIET);

	} else switch (fsp->hsfs_vol_type) {

	case HS_VOL_TYPE_HS:
	case HS_VOL_TYPE_ISO:
	default:
		/*
		 * if iso v1, don't allow trailing spaces in iso file names
		 */
		mount_flags |= HSFSMNT_NOTRAILSPACE;
		fsp->hsfs_namemax = ISO_NAMELEN_V2_MAX;
		fsp->hsfs_namelen = ISO_FILE_NAMELEN;
		vfs_clearmntopt(vfsp, HOPT_RR);
		vfs_clearmntopt(vfsp, HOPT_VERS2);
		vfs_clearmntopt(vfsp, HOPT_JOLIET);
		break;

	case HS_VOL_TYPE_ISO_V2:
		/*
		 * if iso v2, don't copy NOTRAILDOT to hsfs_flags
		 */
		mount_flags &= ~HSFSMNT_NOTRAILDOT;
		mount_flags |= HSFSMNT_NOMAPLCASE | HSFSMNT_NOVERSION;
		fsp->hsfs_namemax = ISO_NAMELEN_V2_MAX;
		fsp->hsfs_namelen = ISO_NAMELEN_V2;
		vfs_setmntopt(vfsp, HOPT_VERS2, NULL, 0);
		vfs_clearmntopt(vfsp, HOPT_RR);
		vfs_clearmntopt(vfsp, HOPT_JOLIET);
		break;

	case HS_VOL_TYPE_JOLIET:
		/*
		 * if Joliet, don't copy NOMAPLCASE or NOTRAILDOT to hsfs_flags
		 */
		mount_flags &= ~(HSFSMNT_NOMAPLCASE | HSFSMNT_NOTRAILDOT);
		mount_flags |= HSFSMNT_NOMAPLCASE;
		if (mount_flags & HSFSMNT_JOLIETLONG)
			fsp->hsfs_namemax = JOLIET_NAMELEN_MAX*3; /* UTF-8 */
		else
			fsp->hsfs_namemax = MAXNAMELEN-1;
		fsp->hsfs_namelen = JOLIET_NAMELEN*2;
		vfs_setmntopt(vfsp, HOPT_JOLIET, NULL, 0);
		vfs_clearmntopt(vfsp, HOPT_RR);
		vfs_clearmntopt(vfsp, HOPT_VERS2);
		break;
	}

	/*
	 * Add the HSFSMNT_INODE pseudo mount flag to the current mount flags.
	 */
	fsp->hsfs_flags = mount_flags | (fsp->hsfs_flags & HSFSMNT_INODE);

	/*
	 * Setup I/O Scheduling structures
	 */
	if (do_schedio) {
		fsp->hqueue = kmem_alloc(sizeof (struct hsfs_queue), KM_SLEEP);
		hsched_init(fsp, fsid, &modlinkage);
	}

	/*
	 * Setup kstats
	 */
	hsfs_init_kstats(fsp, fsid);

	DTRACE_PROBE1(mount__done, struct hsfs *, fsp);

	/*
	 * set the magic word
	 */
	fsp->hsfs_magic = HSFS_MAGIC;
	mutex_exit(&hs_mounttab_lock);

	kmem_free(svp, sizeof (*svp));
	kmem_free(jvp, sizeof (*jvp));

	return (0);

cleanup:
	(void) VOP_CLOSE(devvp, FREAD, 1, (offset_t)0, cr, NULL);
	VN_RELE(devvp);
	if (fsp)
		kmem_free(fsp, sizeof (*fsp));
	if (svp)
		kmem_free(svp, sizeof (*svp));
	if (jvp)
		kmem_free(jvp, sizeof (*jvp));
	return (error);
}

/*
 * Get the rootvp associated with fsp->hsfs_vol
 */
static int
hs_getrootvp(
	struct vfs	*vfsp,
	struct hsfs	*fsp,
	size_t		pathsize)
{
	struct hsnode	*hp;

	ASSERT(pathsize == strlen(fsp->hsfs_fsmnt) + 1);

	/*
	 * If the root directory does not appear to be
	 * valid, use what it points to as "." instead.
	 * Some Defense Mapping Agency disks are non-conformant
	 * in this way.
	 */
	if (!hsfs_valid_dir(&fsp->hsfs_vol.root_dir)) {
		hs_log_bogus_disk_warning(fsp, HSFS_ERR_BAD_ROOT_DIR, 0);
		if (hs_remakenode(fsp->hsfs_vol.root_dir.ext_lbn,
		    (uint_t)0, vfsp, &fsp->hsfs_rootvp)) {
			hs_mounttab = hs_mounttab->hsfs_next;
			mutex_destroy(&fsp->hsfs_free_lock);
			rw_destroy(&fsp->hsfs_hash_lock);
			kmem_free(fsp->hsfs_fsmnt, pathsize);
			mutex_exit(&hs_mounttab_lock);
			return (0);
		}
	} else {
		fsp->hsfs_rootvp = hs_makenode(&fsp->hsfs_vol.root_dir,
		    fsp->hsfs_vol.root_dir.ext_lbn, 0, vfsp);
	}

	/* XXX - ignore the path table for now */
	fsp->hsfs_ptbl = NULL;
	hp = VTOH(fsp->hsfs_rootvp);
	hp->hs_ptbl_idx = NULL;

	return (1);
}

/*
 * hs_findhsvol()
 *
 * Locate the Standard File Structure Volume Descriptor and
 * parse it into an hs_volume structure.
 *
 * XXX - May someday want to look for Coded Character Set FSVD, too.
 */
static int
hs_findhsvol(struct hsfs *fsp, struct vnode *vp, struct hs_volume *hvp)
{
	struct buf *secbp;
	int i;
	int n;
	uchar_t *volp;
	int error;
	uint_t secno;

	secno = hs_findvoldesc(vp->v_rdev, HS_VOLDESC_SEC);
	secbp = bread(vp->v_rdev, secno * 4, HS_SECTOR_SIZE);
	error = geterror(secbp);

	if (error != 0) {
		cmn_err(CE_NOTE, "hs_findhsvol: bread: error=(%d)", error);
		brelse(secbp);
		return (error);
	}

	volp = (uchar_t *)secbp->b_un.b_addr;

	/*
	 * To avoid that we read the whole medium in case that someone prepares
	 * a malicious "fs image", we read at most 32 blocks.
	 */
	for (n = 0; n < 32 &&
	    HSV_DESC_TYPE(volp) != VD_EOV; n++) {
		for (i = 0; i < HSV_ID_STRLEN; i++)
			if (HSV_STD_ID(volp)[i] != HSV_ID_STRING[i])
				goto cantfind;
		if (HSV_STD_VER(volp) != HSV_ID_VER)
			goto cantfind;
		switch (HSV_DESC_TYPE(volp)) {
		case VD_SFS:
			/* Standard File Structure */
			fsp->hsfs_vol_type = HS_VOL_TYPE_HS;
			error = hs_parsehsvol(fsp, volp, hvp);
			brelse(secbp);
			return (error);

		case VD_CCFS:
			/* Coded Character File Structure */
		case VD_BOOT:
		case VD_UNSPEC:
		case VD_EOV:
			break;
		}
		brelse(secbp);
		++secno;
		secbp = bread(vp->v_rdev, secno * 4, HS_SECTOR_SIZE);

		error = geterror(secbp);

		if (error != 0) {
			cmn_err(CE_NOTE, "hs_findhsvol: bread: error=(%d)",
			    error);
			brelse(secbp);
			return (error);
		}

		volp = (uchar_t *)secbp->b_un.b_addr;
	}
cantfind:
	brelse(secbp);
	return (EINVAL);
}

/*
 * hs_parsehsvol
 *
 * Parse the Standard File Structure Volume Descriptor into
 * an hs_volume structure.  We can't just bcopy it into the
 * structure because of byte-ordering problems.
 *
 */
static int
hs_parsehsvol(struct hsfs *fsp, uchar_t *volp, struct hs_volume *hvp)
{
	hvp->vol_size = HSV_VOL_SIZE(volp);
	hvp->lbn_size = HSV_BLK_SIZE(volp);
	if (hvp->lbn_size == 0) {
		cmn_err(CE_NOTE, "hs_parsehsvol: logical block size in the "
		    "SFSVD is zero");
		return (EINVAL);
	}
	hvp->lbn_shift = ffs((long)hvp->lbn_size) - 1;
	hvp->lbn_secshift =
	    ffs((long)howmany(HS_SECTOR_SIZE, (int)hvp->lbn_size)) - 1;
	hvp->lbn_maxoffset = hvp->lbn_size - 1;
	hs_parse_longdate(HSV_cre_date(volp), &hvp->cre_date);
	hs_parse_longdate(HSV_mod_date(volp), &hvp->mod_date);
	hvp->file_struct_ver = HSV_FILE_STRUCT_VER(volp);
	hvp->ptbl_len = HSV_PTBL_SIZE(volp);
	hvp->vol_set_size = (ushort_t)HSV_SET_SIZE(volp);
	hvp->vol_set_seq = (ushort_t)HSV_SET_SEQ(volp);
#if defined(_LITTLE_ENDIAN)
	hvp->ptbl_lbn = HSV_PTBL_MAN_LS(volp);
#else
	hvp->ptbl_lbn = HSV_PTBL_MAN_MS(volp);
#endif
	hs_copylabel(hvp, HSV_VOL_ID(volp), 0);

	/*
	 * Make sure that lbn_size is a power of two and otherwise valid.
	 */
	if (hvp->lbn_size & ~(1 << hvp->lbn_shift)) {
		cmn_err(CE_NOTE,
		    "hsfs: %d-byte logical block size not supported",
		    hvp->lbn_size);
		return (EINVAL);
	}
	return (hs_parsedir(fsp, HSV_ROOT_DIR(volp), &hvp->root_dir,
	    (char *)NULL, (int *)NULL, HDE_ROOT_DIR_REC_SIZE));
}

/*
 * hs_findisovol()
 *
 * Locate the Primary Volume Descriptor
 * parse it into an hs_volume structure.
 *
 * XXX - Partition not yet done
 *
 * Except for fsp->hsfs_vol_type, no fsp member may be modified.
 * fsp->hsfs_vol is modified indirectly via the *hvp argument.
 */
static int
hs_findisovol(struct hsfs *fsp, struct vnode *vp,
    struct hs_volume *hvp,
    struct hs_volume *svp,
    struct hs_volume *jvp)
{
	struct buf *secbp;
	int i;
	int n;
	uchar_t *volp;
	int error;
	uint_t secno;
	int foundpvd = 0;
	int foundsvd = 0;
	int foundjvd = 0;
	int pvd_sum = 0;

	secno = hs_findvoldesc(vp->v_rdev, ISO_VOLDESC_SEC);
	secbp = bread(vp->v_rdev, secno * 4, ISO_SECTOR_SIZE);
	error = geterror(secbp);

	if (error != 0) {
		cmn_err(CE_NOTE, "hs_findisovol: bread: error=(%d)", error);
		brelse(secbp);
		return (error);
	}

	volp = (uchar_t *)secbp->b_un.b_addr;

	/*
	 * To avoid that we read the whole medium in case that someone prepares
	 * a malicious "fs image", we read at most 32 blocks.
	 */
	for (n = 0; n < 32 && ISO_DESC_TYPE(volp) != ISO_VD_EOV; n++) {
		for (i = 0; i < ISO_ID_STRLEN; i++)
			if (ISO_STD_ID(volp)[i] != ISO_ID_STRING[i])
				goto cantfind;
		switch (ISO_DESC_TYPE(volp)) {
		case ISO_VD_PVD:
			/* Standard File Structure */
			if (ISO_STD_VER(volp) != ISO_ID_VER)
				goto cantfind;
			if (foundpvd != 1) {
				fsp->hsfs_vol_type = HS_VOL_TYPE_ISO;
				if (error = hs_parseisovol(fsp, volp, hvp)) {
					brelse(secbp);
					return (error);
				}
				foundpvd = 1;
				for (i = 0; i < ISO_SECTOR_SIZE; i++)
					pvd_sum += volp[i];
			}
			break;
		case ISO_VD_SVD:
			/* Supplementary Volume Descriptor */
			if (ISO_STD_VER(volp) == ISO_ID_VER2 &&
			    foundsvd != 1) {
				fsp->hsfs_vol_type = HS_VOL_TYPE_ISO;
				if (error = hs_parseisovol(fsp, volp, svp)) {
					brelse(secbp);
					return (error);
				}
				foundsvd = 1;
			}
			if (hs_joliet_level(volp) >= 1 && foundjvd != 1) {
				fsp->hsfs_vol_type = HS_VOL_TYPE_ISO;
				if (error = hs_parseisovol(fsp, volp, jvp)) {
					brelse(secbp);
					return (error);
				}
				foundjvd = 1;
			}
			break;
		case ISO_VD_BOOT:
			break;
		case ISO_VD_VPD:
			/* currently cannot handle partition */
			break;
		case VD_EOV:
			break;
		}
		brelse(secbp);
		++secno;
		secbp = bread(vp->v_rdev, secno * 4, HS_SECTOR_SIZE);
		error = geterror(secbp);

		if (error != 0) {
			cmn_err(CE_NOTE, "hs_findisovol: bread: error=(%d)",
			    error);
			brelse(secbp);
			return (error);
		}

		volp = (uchar_t *)secbp->b_un.b_addr;
	}
	for (n = 0; n < 16; n++) {
		brelse(secbp);
		++secno;
		secbp = bread(vp->v_rdev, secno * 4, HS_SECTOR_SIZE);
		error = geterror(secbp);

		if (error != 0) {
			cmn_err(CE_NOTE, "hs_findisovol: bread: error=(%d)",
			    error);
			brelse(secbp);
			return (error);
		}

		/*
		 * Check for the signature from mkisofs that grants that
		 * the current filesystem allows to use the extent lbn as
		 * inode number even in pure ISO9660 mode.
		 */
		volp = (uchar_t *)secbp->b_un.b_addr;
		if (strncmp((char *)volp, "MKI ", 4) == 0) {
			int	sum;

			sum  = volp[2045];
			sum *= 256;
			sum += volp[2046];
			sum *= 256;
			sum += volp[2047];
			if (sum == pvd_sum)
				fsp->hsfs_flags |= HSFSMNT_INODE;
			break;
		}
	}
	if (foundpvd) {
		brelse(secbp);
		return (0);
	}
cantfind:
	brelse(secbp);
	return (EINVAL);
}

/*
 * Return 0 if no Joliet is found
 * else return Joliet Level 1..3
 */
static int
hs_joliet_level(uchar_t *volp)
{
	if (ISO_std_ver(volp)[0] == ISO_ID_VER &&
	    ISO_svd_esc(volp)[0] == '%' &&
	    ISO_svd_esc(volp)[1] == '/') {

		switch (ISO_svd_esc(volp)[2]) {

		case '@':
			return (1);
		case 'C':
			return (2);
		case 'E':
			return (3);
		}
	}
	return (0);
}

/*
 * hs_parseisovol
 *
 * Parse the Primary Volume Descriptor into an hs_volume structure.
 *
 */
static int
hs_parseisovol(struct hsfs *fsp, uchar_t *volp, struct hs_volume *hvp)
{
	hvp->vol_size = ISO_VOL_SIZE(volp);
	hvp->lbn_size = ISO_BLK_SIZE(volp);
	if (hvp->lbn_size == 0) {
		cmn_err(CE_NOTE, "hs_parseisovol: logical block size in the "
		    "PVD is zero");
		return (EINVAL);
	}
	hvp->lbn_shift = ffs((long)hvp->lbn_size) - 1;
	hvp->lbn_secshift =
	    ffs((long)howmany(ISO_SECTOR_SIZE, (int)hvp->lbn_size)) - 1;
	hvp->lbn_maxoffset = hvp->lbn_size - 1;
	hs_parse_longdate(ISO_cre_date(volp), &hvp->cre_date);
	hs_parse_longdate(ISO_mod_date(volp), &hvp->mod_date);
	hvp->file_struct_ver = ISO_FILE_STRUCT_VER(volp);
	hvp->ptbl_len = ISO_PTBL_SIZE(volp);
	hvp->vol_set_size = (ushort_t)ISO_SET_SIZE(volp);
	hvp->vol_set_seq = (ushort_t)ISO_SET_SEQ(volp);
#if defined(_LITTLE_ENDIAN)
	hvp->ptbl_lbn = ISO_PTBL_MAN_LS(volp);
#else
	hvp->ptbl_lbn = ISO_PTBL_MAN_MS(volp);
#endif
	hs_copylabel(hvp, ISO_VOL_ID(volp), hs_joliet_level(volp) >= 1);

	/*
	 * Make sure that lbn_size is a power of two and otherwise valid.
	 */
	if (hvp->lbn_size & ~(1 << hvp->lbn_shift)) {
		cmn_err(CE_NOTE,
		    "hsfs: %d-byte logical block size not supported",
		    hvp->lbn_size);
		return (EINVAL);
	}
	return (hs_parsedir(fsp, ISO_ROOT_DIR(volp), &hvp->root_dir,
	    (char *)NULL, (int *)NULL, IDE_ROOT_DIR_REC_SIZE));
}

/*
 * Common code for mount and umount.
 * Check that the user's argument is a reasonable
 * thing on which to mount, and return the device number if so.
 */
static int
hs_getmdev(struct vfs *vfsp, char *fspec, int flags, dev_t *pdev, mode_t *mode,
    cred_t *cr)
{
	int error;
	struct vnode *svp = NULL;
	struct vnode *lvp = NULL;
	struct vnode *bvp;
	struct vattr vap;
	dev_t dev;
	enum uio_seg fromspace = (flags & MS_SYSSPACE) ?
	    UIO_SYSSPACE : UIO_USERSPACE;

	/*
	 * Look up the device/file to be mounted.
	 */
	error = lookupname(fspec, fromspace, FOLLOW, NULLVPP, &svp);
	if (error) {
		if (error == ENOENT)
			error = ENODEV;
		goto out;
	}

	error = vfs_get_lofi(vfsp, &lvp);

	if (error > 0) {
		if (error == ENOENT)
			error = ENODEV;
		goto out;
	} else if (error == 0) {
		bvp = lvp;
	} else {
		bvp = svp;

		if (bvp->v_type != VBLK) {
			error = ENOTBLK;
			goto out;
		}

		if ((error = secpolicy_spec_open(cr, bvp, FREAD)) != 0)
			goto out;
	}

	/*
	 * Can we read from the device/file ?
	 */
	if ((error = VOP_ACCESS(svp, VREAD, 0, cr, NULL)) != 0)
		goto out;

	vap.va_mask = AT_MODE;		/* get protection mode */
	(void) VOP_GETATTR(bvp, &vap, 0, CRED(), NULL);
	*mode = vap.va_mode;

	dev = *pdev = bvp->v_rdev;

	error = EBUSY;

	/*
	 * Ensure that this device isn't already mounted,
	 * unless this is a REMOUNT request or we are told to suppress
	 * mount checks.
	 */
	if ((flags & MS_NOCHECK) == 0) {
		if (vfs_devmounting(dev, vfsp))
			goto out;
		if (vfs_devismounted(dev) && !(flags & MS_REMOUNT))
			goto out;
	}

	if (getmajor(*pdev) >= devcnt) {
		error = ENXIO;
		goto out;
	}

	error = 0;
out:
	if (svp != NULL)
		VN_RELE(svp);
	if (lvp != NULL)
		VN_RELE(lvp);
	return (error);
}

static void
hs_copylabel(struct hs_volume *hvp, unsigned char *label, int isjoliet)
{
	char	lbuf[64];	/* hs_joliet_cp() creates 48 bytes at most */

	if (isjoliet) {
		/*
		 * hs_joliet_cp() will output 16..48 bytes.
		 * We need to clear 'lbuf' to avoid junk chars past byte 15.
		 */
		bzero(lbuf, sizeof (lbuf));
		(void) hs_joliet_cp((char *)label, lbuf, 32);
		label = (unsigned char *)lbuf;
	}
	/* cdrom volid is at most 32 bytes */
	bcopy(label, hvp->vol_id, 32);
	hvp->vol_id[31] = '\0';
}

/*
 * Mount root file system.
 * "why" is ROOT_INIT on initial call, ROOT_REMOUNT if called to
 * remount the root file system, and ROOT_UNMOUNT if called to
 * unmount the root (e.g., as part of a system shutdown).
 *
 * XXX - this may be partially machine-dependent; it, along with the VFS_SWAPVP
 * operation, goes along with auto-configuration.  A mechanism should be
 * provided by which machine-INdependent code in the kernel can say "get me the
 * right root file system" and "get me the right initial swap area", and have
 * that done in what may well be a machine-dependent fashion.
 * Unfortunately, it is also file-system-type dependent (NFS gets it via
 * bootparams calls, UFS gets it from various and sundry machine-dependent
 * mechanisms, as SPECFS does for swap).
 */
static int
hsfs_mountroot(struct vfs *vfsp, enum whymountroot why)
{
	int error;
	struct hsfs *fsp;
	struct hs_volume *fvolp;
	static int hsfsrootdone = 0;
	dev_t rootdev;
	mode_t mode = 0;

	if (why == ROOT_INIT) {
		if (hsfsrootdone++)
			return (EBUSY);
		rootdev = getrootdev();
		if (rootdev == (dev_t)NODEV)
			return (ENODEV);
		vfsp->vfs_dev = rootdev;
		vfsp->vfs_flag |= VFS_RDONLY;
	} else if (why == ROOT_REMOUNT) {
		cmn_err(CE_NOTE, "hsfs_mountroot: ROOT_REMOUNT");
		return (0);
	} else if (why == ROOT_UNMOUNT) {
		return (0);
	}
	error = vfs_lock(vfsp);
	if (error) {
		cmn_err(CE_NOTE, "hsfs_mountroot: couldn't get vfs_lock");
		return (error);
	}

	error = hs_mountfs(vfsp, rootdev, "/", mode, 1, CRED(), 1);
	/*
	 * XXX - assumes root device is not indirect, because we don't set
	 * rootvp.  Is rootvp used for anything?  If so, make another arg
	 * to mountfs.
	 */
	if (error) {
		vfs_unlock(vfsp);
		if (rootvp) {
			VN_RELE(rootvp);
			rootvp = (struct vnode *)0;
		}
		return (error);
	}
	if (why == ROOT_INIT)
		vfs_add((struct vnode *)0, vfsp,
		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
	vfs_unlock(vfsp);
	fsp = VFS_TO_HSFS(vfsp);
	fvolp = &fsp->hsfs_vol;
#ifdef HSFS_CLKSET
	if (fvolp->cre_date.tv_sec == 0) {
		cmn_err(CE_NOTE, "hsfs_mountroot: cre_date.tv_sec == 0");
		if (fvolp->mod_date.tv_sec == 0) {
			cmn_err(CE_NOTE,
			    "hsfs_mountroot: mod_date.tv_sec == 0");
			cmn_err(CE_NOTE, "hsfs_mountroot: clkset(-1L)");
			clkset(-1L);
		} else {
			clkset(fvolp->mod_date.tv_sec);
		}
	} else {
		clkset(fvolp->mod_date.tv_sec);
	}
#else	/* HSFS_CLKSET */
	clkset(-1L);
#endif	/* HSFS_CLKSET */
	return (0);
}

/*
 * hs_findvoldesc()
 *
 * Return the sector where the volume descriptor lives.  This is
 * a fixed value for "normal" cd-rom's, but can change for
 * multisession cd's.
 *
 * desc_sec is the same for high-sierra and iso 9660 formats, why
 * there are two different #defines used in the code for this is
 * beyond me.  These are standards, cast in concrete, right?
 * To be general, however, this function supports passing in different
 * values.
 */
static int
hs_findvoldesc(dev_t rdev, int desc_sec)
{
	int secno;
	int error;
	int rval;	/* ignored */

#ifdef CDROMREADOFFSET
	/*
	 * Issue the Read Offset ioctl directly to the
	 * device. Ignore any errors and set starting
	 * secno to the default, otherwise add the
	 * VOLDESC sector number to the offset.
	 */
	error = cdev_ioctl(rdev, CDROMREADOFFSET, (intptr_t)&secno,
	    FNATIVE|FKIOCTL|FREAD, CRED(), &rval);
	if (error) {
		secno = desc_sec;
	} else {
		secno += desc_sec;
	}
#else
	secno = desc_sec;
#endif

	return (secno);
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
 * Copyright (c) 2017 by Delphix. All rights reserved.
 */

/*
 * Vnode operations for the High Sierra filesystem
 */

#include <sys/types.h>
#include <sys/t_lock.h>
#include <sys/param.h>
#include <sys/time.h>
#include <sys/systm.h>
#include <sys/sysmacros.h>
#include <sys/resource.h>
#include <sys/signal.h>
#include <sys/cred.h>
#include <sys/user.h>
#include <sys/buf.h>
#include <sys/vfs.h>
#include <sys/vfs_opreg.h>
#include <sys/stat.h>
#include <sys/vnode.h>
#include <sys/mode.h>
#include <sys/proc.h>
#include <sys/disp.h>
#include <sys/file.h>
#include <sys/fcntl.h>
#include <sys/flock.h>
#include <sys/kmem.h>
#include <sys/uio.h>
#include <sys/conf.h>
#include <sys/errno.h>
#include <sys/mman.h>
#include <sys/pathname.h>
#include <sys/debug.h>
#include <sys/vmsystm.h>
#include <sys/cmn_err.h>
#include <sys/fbuf.h>
#include <sys/dirent.h>
#include <sys/errno.h>
#include <sys/dkio.h>
#include <sys/cmn_err.h>
#include <sys/atomic.h>

#include <vm/hat.h>
#include <vm/page.h>
#include <vm/pvn.h>
#include <vm/as.h>
#include <vm/seg.h>
#include <vm/seg_map.h>
#include <vm/seg_kmem.h>
#include <vm/seg_vn.h>
#include <vm/rm.h>
#include <vm/page.h>
#include <sys/swap.h>
#include <sys/avl.h>
#include <sys/sunldi.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/sdt.h>

/*
 * For struct modlinkage
 */
#include <sys/modctl.h>

#include <sys/fs/hsfs_spec.h>
#include <sys/fs/hsfs_node.h>
#include <sys/fs/hsfs_impl.h>
#include <sys/fs/hsfs_susp.h>
#include <sys/fs/hsfs_rrip.h>

#include <fs/fs_subr.h>

/* # of contiguous requests to detect sequential access pattern */
static int seq_contig_requests = 2;

/*
 * This is the max number os taskq threads that will be created
 * if required. Since we are using a Dynamic TaskQ by default only
 * one thread is created initially.
 *
 * NOTE: In the usual hsfs use case this per fs instance number
 * of taskq threads should not place any undue load on a system.
 * Even on an unusual system with say 100 CDROM drives, 800 threads
 * will not be created unless all the drives are loaded and all
 * of them are saturated with I/O at the same time! If there is at
 * all a complaint of system load due to such an unusual case it
 * should be easy enough to change to one per-machine Dynamic TaskQ
 * for all hsfs mounts with a nthreads of say 32.
 */
static int hsfs_taskq_nthreads = 8;	/* # of taskq threads per fs */

/* Min count of adjacent bufs that will avoid buf coalescing */
static int hsched_coalesce_min = 2;

/*
 * Kmem caches for heavily used small allocations. Using these kmem
 * caches provides a factor of 3 reduction in system time and greatly
 * aids overall throughput esp. on SPARC.
 */
struct kmem_cache *hio_cache;
struct kmem_cache *hio_info_cache;

/*
 * This tunable allows us to ignore inode numbers from rrip-1.12.
 * In this case, we fall back to our default inode algorithm.
 */
extern int use_rrip_inodes;

static int hsched_deadline_compare(const void *x1, const void *x2);
static int hsched_offset_compare(const void *x1, const void *x2);
static void hsched_enqueue_io(struct hsfs *fsp, struct hio *hsio, int ra);
int hsched_invoke_strategy(struct hsfs *fsp);

/* ARGSUSED */
static int
hsfs_fsync(vnode_t *cp, int syncflag, cred_t *cred, caller_context_t *ct)
{
	return (0);
}


/*ARGSUSED*/
static int
hsfs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cred,
    struct caller_context *ct)
{
	caddr_t base;
	offset_t diff;
	int error;
	struct hsnode *hp;
	uint_t filesize;

	hp = VTOH(vp);
	/*
	 * if vp is of type VDIR, make sure dirent
	 * is filled up with all info (because of ptbl)
	 */
	if (vp->v_type == VDIR) {
		if (hp->hs_dirent.ext_size == 0)
			hs_filldirent(vp, &hp->hs_dirent);
	}
	filesize = hp->hs_dirent.ext_size;

	/* Sanity checks. */
	if (uiop->uio_resid == 0 ||		/* No data wanted. */
	    uiop->uio_loffset > HS_MAXFILEOFF ||	/* Offset too big. */
	    uiop->uio_loffset >= filesize)	/* Past EOF. */
		return (0);

	do {
		/*
		 * We want to ask for only the "right" amount of data.
		 * In this case that means:-
		 *
		 * We can't get data from beyond our EOF. If asked,
		 * we will give a short read.
		 *
		 * segmap_getmapflt returns buffers of MAXBSIZE bytes.
		 * These buffers are always MAXBSIZE aligned.
		 * If our starting offset is not MAXBSIZE aligned,
		 * we can only ask for less than MAXBSIZE bytes.
		 *
		 * If our requested offset and length are such that
		 * they belong in different MAXBSIZE aligned slots
		 * then we'll be making more than one call on
		 * segmap_getmapflt.
		 *
		 * This diagram shows the variables we use and their
		 * relationships.
		 *
		 * |<-----MAXBSIZE----->|
		 * +--------------------------...+
		 * |.....mapon->|<--n-->|....*...|EOF
		 * +--------------------------...+
		 * uio_loffset->|
		 * uio_resid....|<---------->|
		 * diff.........|<-------------->|
		 *
		 * So, in this case our offset is not aligned
		 * and our request takes us outside of the
		 * MAXBSIZE window. We will break this up into
		 * two segmap_getmapflt calls.
		 */
		size_t nbytes;
		offset_t mapon;
		size_t n;
		uint_t flags;

		mapon = uiop->uio_loffset & MAXBOFFSET;
		diff = filesize - uiop->uio_loffset;
		nbytes = (size_t)MIN(MAXBSIZE - mapon, uiop->uio_resid);
		n = MIN(diff, nbytes);
		if (n <= 0) {
			/* EOF or request satisfied. */
			return (0);
		}

		base = segmap_getmapflt(segkmap, vp,
		    (u_offset_t)uiop->uio_loffset, n, 1, S_READ);

		error = uiomove(base + mapon, n, UIO_READ, uiop);

		if (error == 0) {
			/*
			 * if read a whole block, or read to eof,
			 *  won't need this buffer again soon.
			 */
			if (n + mapon == MAXBSIZE ||
			    uiop->uio_loffset == filesize)
				flags = SM_DONTNEED;
			else
				flags = 0;

			error = segmap_release(segkmap, base, flags);
		} else
			(void) segmap_release(segkmap, base, 0);
	} while (error == 0 && uiop->uio_resid > 0);

	return (error);
}

/*ARGSUSED2*/
static int
hsfs_getattr(struct vnode *vp, struct vattr *vap, int flags, struct cred *cred,
    caller_context_t *ct)
{
	struct hsnode *hp;
	struct vfs *vfsp;
	struct hsfs *fsp;

	hp = VTOH(vp);
	fsp = VFS_TO_HSFS(vp->v_vfsp);
	vfsp = vp->v_vfsp;

	if ((hp->hs_dirent.ext_size == 0) && (vp->v_type == VDIR)) {
		hs_filldirent(vp, &hp->hs_dirent);
	}
	vap->va_type = IFTOVT(hp->hs_dirent.mode);
	vap->va_mode = hp->hs_dirent.mode;
	vap->va_uid = hp->hs_dirent.uid;
	vap->va_gid = hp->hs_dirent.gid;

	vap->va_fsid = vfsp->vfs_dev;
	vap->va_nodeid = (ino64_t)hp->hs_nodeid;
	vap->va_nlink = hp->hs_dirent.nlink;
	vap->va_size =	(offset_t)hp->hs_dirent.ext_size;

	vap->va_atime.tv_sec = hp->hs_dirent.adate.tv_sec;
	vap->va_atime.tv_nsec = hp->hs_dirent.adate.tv_usec*1000;
	vap->va_mtime.tv_sec = hp->hs_dirent.mdate.tv_sec;
	vap->va_mtime.tv_nsec = hp->hs_dirent.mdate.tv_usec*1000;
	vap->va_ctime.tv_sec = hp->hs_dirent.cdate.tv_sec;
	vap->va_ctime.tv_nsec = hp->hs_dirent.cdate.tv_usec*1000;
	if (vp->v_type == VCHR || vp->v_type == VBLK)
		vap->va_rdev = hp->hs_dirent.r_dev;
	else
		vap->va_rdev = 0;
	vap->va_blksize = vfsp->vfs_bsize;
	/* no. of blocks = no. of data blocks + no. of xar blocks */
	vap->va_nblocks = (fsblkcnt64_t)howmany(vap->va_size + (u_longlong_t)
	    (hp->hs_dirent.xar_len << fsp->hsfs_vol.lbn_shift), DEV_BSIZE);
	vap->va_seq = hp->hs_seq;
	return (0);
}

/*ARGSUSED*/
static int
hsfs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cred,
    caller_context_t *ct)
{
	struct hsnode *hp;

	if (vp->v_type != VLNK)
		return (EINVAL);

	hp = VTOH(vp);

	if (hp->hs_dirent.sym_link == (char *)NULL)
		return (ENOENT);

	return (uiomove(hp->hs_dirent.sym_link,
	    (size_t)MIN(hp->hs_dirent.ext_size,
	    uiop->uio_resid), UIO_READ, uiop));
}

/*ARGSUSED*/
static void
hsfs_inactive(struct vnode *vp, struct cred *cred, caller_context_t *ct)
{
	struct hsnode *hp;
	struct hsfs *fsp;

	int nopage;

	hp = VTOH(vp);
	fsp = VFS_TO_HSFS(vp->v_vfsp);
	/*
	 * Note: acquiring and holding v_lock for quite a while
	 * here serializes on the vnode; this is unfortunate, but
	 * likely not to overly impact performance, as the underlying
	 * device (CDROM drive) is quite slow.
	 */
	rw_enter(&fsp->hsfs_hash_lock, RW_WRITER);
	mutex_enter(&hp->hs_contents_lock);
	mutex_enter(&vp->v_lock);

	if (vp->v_count < 1) {
		panic("hsfs_inactive: v_count < 1");
		/*NOTREACHED*/
	}

	VN_RELE_LOCKED(vp);
	if (vp->v_count > 0 || (hp->hs_flags & HREF) == 0) {
		mutex_exit(&vp->v_lock);
		mutex_exit(&hp->hs_contents_lock);
		rw_exit(&fsp->hsfs_hash_lock);
		return;
	}
	if (vp->v_count == 0) {
		/*
		 * Free the hsnode.
		 * If there are no pages associated with the
		 * hsnode, give it back to the kmem_cache,
		 * else put at the end of this file system's
		 * internal free list.
		 */
		nopage = !vn_has_cached_data(vp);
		hp->hs_flags = 0;
		/*
		 * exit these locks now, since hs_freenode may
		 * kmem_free the hsnode and embedded vnode
		 */
		mutex_exit(&vp->v_lock);
		mutex_exit(&hp->hs_contents_lock);
		hs_freenode(vp, fsp, nopage);
	} else {
		mutex_exit(&vp->v_lock);
		mutex_exit(&hp->hs_contents_lock);
	}
	rw_exit(&fsp->hsfs_hash_lock);
}


/*ARGSUSED*/
static int
hsfs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
    struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cred,
    caller_context_t *ct, int *direntflags, pathname_t *realpnp)
{
	int error;
	int namelen = (int)strlen(nm);

	if (*nm == '\0') {
		VN_HOLD(dvp);
		*vpp = dvp;
		return (0);
	}

	/*
	 * If we're looking for ourself, life is simple.
	 */
	if (namelen == 1 && *nm == '.') {
		if (error = hs_access(dvp, (mode_t)VEXEC, cred))
			return (error);
		VN_HOLD(dvp);
		*vpp = dvp;
		return (0);
	}

	return (hs_dirlook(dvp, nm, namelen, vpp, cred));
}


/*ARGSUSED*/
static int
hsfs_readdir(struct vnode *vp, struct uio *uiop, struct cred *cred, int *eofp,
    caller_context_t *ct, int flags)
{
	struct hsnode	*dhp;
	struct hsfs	*fsp;
	struct hs_direntry hd;
	struct dirent64	*nd;
	int		error;
	uint_t		offset;		/* real offset in directory */
	uint_t		dirsiz;		/* real size of directory */
	uchar_t		*blkp;
	int		hdlen;		/* length of hs directory entry */
	long		ndlen;		/* length of dirent entry */
	int		bytes_wanted;
	size_t		bufsize;	/* size of dirent buffer */
	char		*outbuf;	/* ptr to dirent buffer */
	char		*dname;
	int		dnamelen;
	size_t		dname_size;
	struct fbuf	*fbp;
	uint_t		last_offset;	/* last index into current dir block */
	ino64_t		dirino;	/* temporary storage before storing in dirent */
	off_t		diroff;

	dhp = VTOH(vp);
	fsp = VFS_TO_HSFS(vp->v_vfsp);
	if (dhp->hs_dirent.ext_size == 0)
		hs_filldirent(vp, &dhp->hs_dirent);
	dirsiz = dhp->hs_dirent.ext_size;
	if (uiop->uio_loffset >= dirsiz) {	/* at or beyond EOF */
		if (eofp)
			*eofp = 1;
		return (0);
	}
	ASSERT(uiop->uio_loffset <= HS_MAXFILEOFF);
	offset = uiop->uio_loffset;

	dname_size = fsp->hsfs_namemax + 1;	/* 1 for the ending NUL */
	dname = kmem_alloc(dname_size, KM_SLEEP);
	bufsize = uiop->uio_resid + sizeof (struct dirent64);

	outbuf = kmem_alloc(bufsize, KM_SLEEP);
	nd = (struct dirent64 *)outbuf;

	while (offset < dirsiz) {
		bytes_wanted = MIN(MAXBSIZE, dirsiz - (offset & MAXBMASK));

		error = fbread(vp, (offset_t)(offset & MAXBMASK),
		    (unsigned int)bytes_wanted, S_READ, &fbp);
		if (error)
			goto done;

		blkp = (uchar_t *)fbp->fb_addr;
		last_offset = (offset & MAXBMASK) + fbp->fb_count;

#define	rel_offset(offset) ((offset) & MAXBOFFSET)	/* index into blkp */

		while (offset < last_offset) {
			/*
			 * Very similar validation code is found in
			 * process_dirblock(), hsfs_node.c.
			 * For an explanation, see there.
			 * It may make sense for the future to
			 * "consolidate" the code in hs_parsedir(),
			 * process_dirblock() and hsfs_readdir() into
			 * a single utility function.
			 */
			hdlen = (int)((uchar_t)
			    HDE_DIR_LEN(&blkp[rel_offset(offset)]));
			if (hdlen < HDE_ROOT_DIR_REC_SIZE ||
			    offset + hdlen > last_offset) {
				/*
				 * advance to next sector boundary
				 */
				offset = roundup(offset + 1, HS_SECTOR_SIZE);
				if (hdlen)
					hs_log_bogus_disk_warning(fsp,
					    HSFS_ERR_TRAILING_JUNK, 0);

				continue;
			}

			bzero(&hd, sizeof (hd));

			/*
			 * Just ignore invalid directory entries.
			 * XXX - maybe hs_parsedir() will detect EXISTENCE bit
			 */
			if (!hs_parsedir(fsp, &blkp[rel_offset(offset)],
			    &hd, dname, &dnamelen, last_offset - offset)) {
				/*
				 * Determine if there is enough room
				 */
				ndlen = (long)DIRENT64_RECLEN((dnamelen));

				if ((ndlen + ((char *)nd - outbuf)) >
				    uiop->uio_resid) {
					fbrelse(fbp, S_READ);
					goto done; /* output buffer full */
				}

				diroff = offset + hdlen;
				/*
				 * If the media carries rrip-v1.12 or newer,
				 * and we trust the inodes from the rrip data
				 * (use_rrip_inodes != 0), use that data. If the
				 * media has been created by a recent mkisofs
				 * version, we may trust all numbers in the
				 * starting extent number; otherwise, we cannot
				 * do this for zero sized files and symlinks,
				 * because if we did we'd end up mapping all of
				 * them to the same node. We use HS_DUMMY_INO
				 * in this case and make sure that we will not
				 * map all files to the same meta data.
				 */
				if (hd.inode != 0 && use_rrip_inodes) {
					dirino = hd.inode;
				} else if ((hd.ext_size == 0 ||
				    hd.sym_link != (char *)NULL) &&
				    (fsp->hsfs_flags & HSFSMNT_INODE) == 0) {
					dirino = HS_DUMMY_INO;
				} else {
					dirino = hd.ext_lbn;
				}

				/* strncpy(9f) will zero uninitialized bytes */

				ASSERT(strlen(dname) + 1 <=
				    DIRENT64_NAMELEN(ndlen));
				(void) strncpy(nd->d_name, dname,
				    DIRENT64_NAMELEN(ndlen));
				nd->d_reclen = (ushort_t)ndlen;
				nd->d_off = (offset_t)diroff;
				nd->d_ino = dirino;
				nd = (struct dirent64 *)((char *)nd + ndlen);

				/*
				 * free up space allocated for symlink
				 */
				if (hd.sym_link != (char *)NULL) {
					kmem_free(hd.sym_link,
					    (size_t)(hd.ext_size+1));
					hd.sym_link = (char *)NULL;
				}
			}
			offset += hdlen;
		}
		fbrelse(fbp, S_READ);
	}

	/*
	 * Got here for one of the following reasons:
	 *	1) outbuf is full (error == 0)
	 *	2) end of directory reached (error == 0)
	 *	3) error reading directory sector (error != 0)
	 *	4) directory entry crosses sector boundary (error == 0)
	 *
	 * If any directory entries have been copied, don't report
	 * case 4.  Instead, return the valid directory entries.
	 *
	 * If no entries have been copied, report the error.
	 * If case 4, this will be indistiguishable from EOF.
	 */
done:
	ndlen = ((char *)nd - outbuf);
	if (ndlen != 0) {
		error = uiomove(outbuf, (size_t)ndlen, UIO_READ, uiop);
		uiop->uio_loffset = offset;
	}
	kmem_free(dname, dname_size);
	kmem_free(outbuf, bufsize);
	if (eofp && error == 0)
		*eofp = (uiop->uio_loffset >= dirsiz);
	return (error);
}

/*ARGSUSED2*/
static int
hsfs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
{
	struct hsnode *hp;
	struct hsfid *fid;

	if (fidp->fid_len < (sizeof (*fid) - sizeof (fid->hf_len))) {
		fidp->fid_len = sizeof (*fid) - sizeof (fid->hf_len);
		return (ENOSPC);
	}

	fid = (struct hsfid *)fidp;
	fid->hf_len = sizeof (*fid) - sizeof (fid->hf_len);
	hp = VTOH(vp);
	mutex_enter(&hp->hs_contents_lock);
	fid->hf_dir_lbn = hp->hs_dir_lbn;
	fid->hf_dir_off = (ushort_t)hp->hs_dir_off;
	fid->hf_ino = hp->hs_nodeid;
	mutex_exit(&hp->hs_contents_lock);
	return (0);
}

/*ARGSUSED*/
static int
hsfs_open(struct vnode **vpp, int flag, struct cred *cred, caller_context_t *ct)
{
	return (0);
}

/*ARGSUSED*/
static int
hsfs_close(struct vnode *vp, int flag, int count, offset_t offset,
    struct cred *cred, caller_context_t *ct)
{
	(void) cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
	cleanshares(vp, ttoproc(curthread)->p_pid);
	return (0);
}

/*ARGSUSED2*/
static int
hsfs_access(struct vnode *vp, int mode, int flags, cred_t *cred,
    caller_context_t *ct)
{
	return (hs_access(vp, (mode_t)mode, cred));
}

/*
 * the seek time of a CD-ROM is very slow, and data transfer
 * rate is even worse (max. 150K per sec).  The design
 * decision is to reduce access to cd-rom as much as possible,
 * and to transfer a sizable block (read-ahead) of data at a time.
 * UFS style of read ahead one block at a time is not appropriate,
 * and is not supported
 */

/*
 * KLUSTSIZE should be a multiple of PAGESIZE and <= MAXPHYS.
 */
#define	KLUSTSIZE	(56 * 1024)
/* we don't support read ahead */
int hsfs_lostpage;	/* no. of times we lost original page */

/*
 * Used to prevent biodone() from releasing buf resources that
 * we didn't allocate in quite the usual way.
 */
/*ARGSUSED*/
int
hsfs_iodone(struct buf *bp)
{
	sema_v(&bp->b_io);
	return (0);
}

/*
 * The taskq thread that invokes the scheduling function to ensure
 * that all readaheads are complete and cleans up the associated
 * memory and releases the page lock.
 */
void
hsfs_ra_task(void *arg)
{
	struct hio_info *info = arg;
	uint_t count;
	struct buf *wbuf;

	ASSERT(info->pp != NULL);

	for (count = 0; count < info->bufsused; count++) {
		wbuf = &(info->bufs[count]);

		DTRACE_PROBE1(hsfs_io_wait_ra, struct buf *, wbuf);
		while (sema_tryp(&(info->sema[count])) == 0) {
			if (hsched_invoke_strategy(info->fsp)) {
				sema_p(&(info->sema[count]));
				break;
			}
		}
		sema_destroy(&(info->sema[count]));
		DTRACE_PROBE1(hsfs_io_done_ra, struct buf *, wbuf);
		biofini(&(info->bufs[count]));
	}
	for (count = 0; count < info->bufsused; count++) {
		if (info->vas[count] != NULL) {
			ppmapout(info->vas[count]);
		}
	}
	kmem_free(info->vas, info->bufcnt * sizeof (caddr_t));
	kmem_free(info->bufs, info->bufcnt * sizeof (struct buf));
	kmem_free(info->sema, info->bufcnt * sizeof (ksema_t));

	pvn_read_done(info->pp, 0);
	kmem_cache_free(hio_info_cache, info);
}

/*
 * Submit asynchronous readahead requests to the I/O scheduler
 * depending on the number of pages to read ahead. These requests
 * are asynchronous to the calling thread but I/O requests issued
 * subsequently by other threads with higher LBNs must wait for
 * these readaheads to complete since we have a single ordered
 * I/O pipeline. Thus these readaheads are semi-asynchronous.
 * A TaskQ handles waiting for the readaheads to complete.
 *
 * This function is mostly a copy of hsfs_getapage but somewhat
 * simpler. A readahead request is aborted if page allocation
 * fails.
 */
/*ARGSUSED*/
static int
hsfs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg,
    caddr_t addr, struct hsnode *hp, struct hsfs *fsp, int xarsiz,
    offset_t bof, int chunk_lbn_count, int chunk_data_bytes)
{
	struct buf *bufs;
	caddr_t *vas;
	caddr_t va;
	struct page *pp, *searchp, *lastp;
	struct vnode *devvp;
	ulong_t	byte_offset;
	size_t	io_len_tmp;
	uint_t	io_off, io_len;
	uint_t	xlen;
	uint_t	filsiz;
	uint_t	secsize;
	uint_t	bufcnt;
	uint_t	bufsused;
	uint_t	count;
	uint_t	io_end;
	uint_t	which_chunk_lbn;
	uint_t	offset_lbn;
	uint_t	offset_extra;
	offset_t	offset_bytes;
	uint_t	remaining_bytes;
	uint_t	extension;
	int	remainder;	/* must be signed */
	diskaddr_t driver_block;
	u_offset_t io_off_tmp;
	ksema_t	*fio_done;
	struct hio_info *info;
	size_t len;

	ASSERT(fsp->hqueue != NULL);

	if (addr >= seg->s_base + seg->s_size) {
		return (-1);
	}

	devvp = fsp->hsfs_devvp;
	secsize = fsp->hsfs_vol.lbn_size;  /* bytes per logical block */

	/* file data size */
	filsiz = hp->hs_dirent.ext_size;

	if (off >= filsiz)
		return (0);

	extension = 0;
	pp = NULL;

	extension += hp->hs_ra_bytes;

	/*
	 * Some CD writers (e.g. Kodak Photo CD writers)
	 * create CDs in TAO mode and reserve tracks that
	 * are not completely written. Some sectors remain
	 * unreadable for this reason and give I/O errors.
	 * Also, there's no point in reading sectors
	 * we'll never look at.  So, if we're asked to go
	 * beyond the end of a file, truncate to the length
	 * of that file.
	 *
	 * Additionally, this behaviour is required by section
	 * 6.4.5 of ISO 9660:1988(E).
	 */
	len = MIN(extension ? extension : PAGESIZE, filsiz - off);

	/* A little paranoia */
	if (len <= 0)
		return (-1);

	/*
	 * After all that, make sure we're asking for things in units
	 * that bdev_strategy() will understand (see bug 4202551).
	 */
	len = roundup(len, DEV_BSIZE);

	pp = pvn_read_kluster(vp, off, seg, addr, &io_off_tmp,
	    &io_len_tmp, off, len, 1);

	if (pp == NULL) {
		hp->hs_num_contig = 0;
		hp->hs_ra_bytes = 0;
		hp->hs_prev_offset = 0;
		return (-1);
	}

	io_off = (uint_t)io_off_tmp;
	io_len = (uint_t)io_len_tmp;

	/* check for truncation */
	/*
	 * xxx Clean up and return EIO instead?
	 * xxx Ought to go to u_offset_t for everything, but we
	 * xxx call lots of things that want uint_t arguments.
	 */
	ASSERT(io_off == io_off_tmp);

	/*
	 * get enough buffers for worst-case scenario
	 * (i.e., no coalescing possible).
	 */
	bufcnt = (len + secsize - 1) / secsize;
	bufs = kmem_alloc(bufcnt * sizeof (struct buf), KM_SLEEP);
	vas = kmem_alloc(bufcnt * sizeof (caddr_t), KM_SLEEP);

	/*
	 * Allocate a array of semaphores since we are doing I/O
	 * scheduling.
	 */
	fio_done = kmem_alloc(bufcnt * sizeof (ksema_t), KM_SLEEP);

	/*
	 * If our filesize is not an integer multiple of PAGESIZE,
	 * we zero that part of the last page that's between EOF and
	 * the PAGESIZE boundary.
	 */
	xlen = io_len & PAGEOFFSET;
	if (xlen != 0)
		pagezero(pp->p_prev, xlen, PAGESIZE - xlen);

	DTRACE_PROBE2(hsfs_readahead, struct vnode *, vp, uint_t, io_len);

	va = NULL;
	lastp = NULL;
	searchp = pp;
	io_end = io_off + io_len;
	for (count = 0, byte_offset = io_off;
	    byte_offset < io_end;
	    count++) {
		ASSERT(count < bufcnt);

		bioinit(&bufs[count]);
		bufs[count].b_edev = devvp->v_rdev;
		bufs[count].b_dev = cmpdev(devvp->v_rdev);
		bufs[count].b_flags = B_NOCACHE|B_BUSY|B_READ;
		bufs[count].b_iodone = hsfs_iodone;
		bufs[count].b_vp = vp;
		bufs[count].b_file = vp;

		/* Compute disk address for interleaving. */

		/* considered without skips */
		which_chunk_lbn = byte_offset / chunk_data_bytes;

		/* factor in skips */
		offset_lbn = which_chunk_lbn * chunk_lbn_count;

		/* convert to physical byte offset for lbn */
		offset_bytes = LBN_TO_BYTE(offset_lbn, vp->v_vfsp);

		/* don't forget offset into lbn */
		offset_extra = byte_offset % chunk_data_bytes;

		/* get virtual block number for driver */
		driver_block = lbtodb(bof + xarsiz
		    + offset_bytes + offset_extra);

		if (lastp != searchp) {
			/* this branch taken first time through loop */
			va = vas[count] = ppmapin(searchp, PROT_WRITE,
			    (caddr_t)-1);
			/* ppmapin() guarantees not to return NULL */
		} else {
			vas[count] = NULL;
		}

		bufs[count].b_un.b_addr = va + byte_offset % PAGESIZE;
		bufs[count].b_offset =
		    (offset_t)(byte_offset - io_off + off);

		/*
		 * We specifically use the b_lblkno member here
		 * as even in the 32 bit world driver_block can
		 * get very large in line with the ISO9660 spec.
		 */

		bufs[count].b_lblkno = driver_block;

		remaining_bytes = ((which_chunk_lbn + 1) * chunk_data_bytes)
		    - byte_offset;

		/*
		 * remaining_bytes can't be zero, as we derived
		 * which_chunk_lbn directly from byte_offset.
		 */
		if ((remaining_bytes + byte_offset) < (off + len)) {
			/* coalesce-read the rest of the chunk */
			bufs[count].b_bcount = remaining_bytes;
		} else {
			/* get the final bits */
			bufs[count].b_bcount = off + len - byte_offset;
		}

		remainder = PAGESIZE - (byte_offset % PAGESIZE);
		if (bufs[count].b_bcount > remainder) {
			bufs[count].b_bcount = remainder;
		}

		bufs[count].b_bufsize = bufs[count].b_bcount;
		if (((offset_t)byte_offset + bufs[count].b_bcount) >
		    HS_MAXFILEOFF) {
			break;
		}
		byte_offset += bufs[count].b_bcount;

		/*
		 * We are scheduling I/O so we need to enqueue
		 * requests rather than calling bdev_strategy
		 * here. A later invocation of the scheduling
		 * function will take care of doing the actual
		 * I/O as it selects requests from the queue as
		 * per the scheduling logic.
		 */
		struct hio *hsio = kmem_cache_alloc(hio_cache,
		    KM_SLEEP);

		sema_init(&fio_done[count], 0, NULL,
		    SEMA_DEFAULT, NULL);
		hsio->bp = &bufs[count];
		hsio->sema = &fio_done[count];
		hsio->io_lblkno = bufs[count].b_lblkno;
		hsio->nblocks = howmany(hsio->bp->b_bcount,
		    DEV_BSIZE);

		/* used for deadline */
		hsio->io_timestamp = drv_hztousec(ddi_get_lbolt());

		/* for I/O coalescing */
		hsio->contig_chain = NULL;
		hsched_enqueue_io(fsp, hsio, 1);

		lwp_stat_update(LWP_STAT_INBLK, 1);
		lastp = searchp;
		if ((remainder - bufs[count].b_bcount) < 1) {
			searchp = searchp->p_next;
		}
	}

	bufsused = count;
	info = kmem_cache_alloc(hio_info_cache, KM_SLEEP);
	info->bufs = bufs;
	info->vas = vas;
	info->sema = fio_done;
	info->bufsused = bufsused;
	info->bufcnt = bufcnt;
	info->fsp = fsp;
	info->pp = pp;

	(void) taskq_dispatch(fsp->hqueue->ra_task,
	    hsfs_ra_task, info, KM_SLEEP);
	/*
	 * The I/O locked pages are unlocked in our taskq thread.
	 */
	return (0);
}

/*
 * Each file may have a different interleaving on disk.  This makes
 * things somewhat interesting.  The gist is that there are some
 * number of contiguous data sectors, followed by some other number
 * of contiguous skip sectors.  The sum of those two sets of sectors
 * defines the interleave size.  Unfortunately, it means that we generally
 * can't simply read N sectors starting at a given offset to satisfy
 * any given request.
 *
 * What we do is get the relevant memory pages via pvn_read_kluster(),
 * then stride through the interleaves, setting up a buf for each
 * sector that needs to be brought in.  Instead of kmem_alloc'ing
 * space for the sectors, though, we just point at the appropriate
 * spot in the relevant page for each of them.  This saves us a bunch
 * of copying.
 *
 * NOTICE: The code below in hsfs_getapage is mostly same as the code
 *         in hsfs_getpage_ra above (with some omissions). If you are
 *         making any change to this function, please also look at
 *         hsfs_getpage_ra.
 */
/*ARGSUSED*/
static int
hsfs_getapage(struct vnode *vp, u_offset_t off, size_t len, uint_t *protp,
    struct page *pl[], size_t plsz, struct seg *seg, caddr_t addr,
    enum seg_rw rw, struct cred *cred)
{
	struct hsnode *hp;
	struct hsfs *fsp;
	int	err;
	struct buf *bufs;
	caddr_t *vas;
	caddr_t va;
	struct page *pp, *searchp, *lastp;
	page_t	*pagefound;
	offset_t	bof;
	struct vnode *devvp;
	ulong_t	byte_offset;
	size_t	io_len_tmp;
	uint_t	io_off, io_len;
	uint_t	xlen;
	uint_t	filsiz;
	uint_t	secsize;
	uint_t	bufcnt;
	uint_t	bufsused;
	uint_t	count;
	uint_t	io_end;
	uint_t	which_chunk_lbn;
	uint_t	offset_lbn;
	uint_t	offset_extra;
	offset_t	offset_bytes;
	uint_t	remaining_bytes;
	uint_t	extension;
	int	remainder;	/* must be signed */
	int	chunk_lbn_count;
	int	chunk_data_bytes;
	int	xarsiz;
	diskaddr_t driver_block;
	u_offset_t io_off_tmp;
	ksema_t *fio_done;
	int	calcdone;

	/*
	 * We don't support asynchronous operation at the moment, so
	 * just pretend we did it.  If the pages are ever actually
	 * needed, they'll get brought in then.
	 */
	if (pl == NULL)
		return (0);

	hp = VTOH(vp);
	fsp = VFS_TO_HSFS(vp->v_vfsp);
	devvp = fsp->hsfs_devvp;
	secsize = fsp->hsfs_vol.lbn_size;  /* bytes per logical block */

	/* file data size */
	filsiz = hp->hs_dirent.ext_size;

	/* disk addr for start of file */
	bof = LBN_TO_BYTE((offset_t)hp->hs_dirent.ext_lbn, vp->v_vfsp);

	/* xarsiz byte must be skipped for data */
	xarsiz = hp->hs_dirent.xar_len << fsp->hsfs_vol.lbn_shift;

	/* how many logical blocks in an interleave (data+skip) */
	chunk_lbn_count = hp->hs_dirent.intlf_sz + hp->hs_dirent.intlf_sk;

	if (chunk_lbn_count == 0) {
		chunk_lbn_count = 1;
	}

	/*
	 * Convert interleaving size into bytes.  The zero case
	 * (no interleaving) optimization is handled as a side-
	 * effect of the read-ahead logic.
	 */
	if (hp->hs_dirent.intlf_sz == 0) {
		chunk_data_bytes = LBN_TO_BYTE(1, vp->v_vfsp);
		/*
		 * Optimization: If our pagesize is a multiple of LBN
		 * bytes, we can avoid breaking up a page into individual
		 * lbn-sized requests.
		 */
		if (PAGESIZE % chunk_data_bytes == 0) {
			chunk_lbn_count = BYTE_TO_LBN(PAGESIZE, vp->v_vfsp);
			chunk_data_bytes = PAGESIZE;
		}
	} else {
		chunk_data_bytes =
		    LBN_TO_BYTE(hp->hs_dirent.intlf_sz, vp->v_vfsp);
	}

reread:
	err = 0;
	pagefound = 0;
	calcdone = 0;

	/*
	 * Do some read-ahead.  This mostly saves us a bit of
	 * system cpu time more than anything else when doing
	 * sequential reads.  At some point, could do the
	 * read-ahead asynchronously which might gain us something
	 * on wall time, but it seems unlikely....
	 *
	 * We do the easy case here, which is to read through
	 * the end of the chunk, minus whatever's at the end that
	 * won't exactly fill a page.
	 */
	if (hp->hs_ra_bytes > 0 && chunk_data_bytes != PAGESIZE) {
		which_chunk_lbn = (off + len) / chunk_data_bytes;
		extension = ((which_chunk_lbn + 1) * chunk_data_bytes) - off;
		extension -= (extension % PAGESIZE);
	} else {
		extension = roundup(len, PAGESIZE);
	}

	atomic_inc_64(&fsp->total_pages_requested);

	pp = NULL;
again:
	/* search for page in buffer */
	if ((pagefound = page_exists(vp, off)) == 0) {
		/*
		 * Need to really do disk IO to get the page.
		 */
		if (!calcdone) {
			extension += hp->hs_ra_bytes;

			/*
			 * Some cd writers don't write sectors that aren't
			 * used. Also, there's no point in reading sectors
			 * we'll never look at.  So, if we're asked to go
			 * beyond the end of a file, truncate to the length
			 * of that file.
			 *
			 * Additionally, this behaviour is required by section
			 * 6.4.5 of ISO 9660:1988(E).
			 */
			len = MIN(extension ? extension : PAGESIZE,
			    filsiz - off);

			/* A little paranoia. */
			ASSERT(len > 0);

			/*
			 * After all that, make sure we're asking for things
			 * in units that bdev_strategy() will understand
			 * (see bug 4202551).
			 */
			len = roundup(len, DEV_BSIZE);
			calcdone = 1;
		}

		pp = pvn_read_kluster(vp, off, seg, addr, &io_off_tmp,
		    &io_len_tmp, off, len, 0);

		if (pp == NULL) {
			/*
			 * Pressure on memory, roll back readahead
			 */
			hp->hs_num_contig = 0;
			hp->hs_ra_bytes = 0;
			hp->hs_prev_offset = 0;
			goto again;
		}

		io_off = (uint_t)io_off_tmp;
		io_len = (uint_t)io_len_tmp;

		/* check for truncation */
		/*
		 * xxx Clean up and return EIO instead?
		 * xxx Ought to go to u_offset_t for everything, but we
		 * xxx call lots of things that want uint_t arguments.
		 */
		ASSERT(io_off == io_off_tmp);

		/*
		 * get enough buffers for worst-case scenario
		 * (i.e., no coalescing possible).
		 */
		bufcnt = (len + secsize - 1) / secsize;
		bufs = kmem_zalloc(bufcnt * sizeof (struct buf), KM_SLEEP);
		vas = kmem_alloc(bufcnt * sizeof (caddr_t), KM_SLEEP);

		/*
		 * Allocate a array of semaphores if we are doing I/O
		 * scheduling.
		 */
		if (fsp->hqueue != NULL)
			fio_done = kmem_alloc(bufcnt * sizeof (ksema_t),
			    KM_SLEEP);
		for (count = 0; count < bufcnt; count++) {
			bioinit(&bufs[count]);
			bufs[count].b_edev = devvp->v_rdev;
			bufs[count].b_dev = cmpdev(devvp->v_rdev);
			bufs[count].b_flags = B_NOCACHE|B_BUSY|B_READ;
			bufs[count].b_iodone = hsfs_iodone;
			bufs[count].b_vp = vp;
			bufs[count].b_file = vp;
		}

		/*
		 * If our filesize is not an integer multiple of PAGESIZE,
		 * we zero that part of the last page that's between EOF and
		 * the PAGESIZE boundary.
		 */
		xlen = io_len & PAGEOFFSET;
		if (xlen != 0)
			pagezero(pp->p_prev, xlen, PAGESIZE - xlen);

		va = NULL;
		lastp = NULL;
		searchp = pp;
		io_end = io_off + io_len;
		for (count = 0, byte_offset = io_off;
		    byte_offset < io_end; count++) {
			ASSERT(count < bufcnt);

			/* Compute disk address for interleaving. */

			/* considered without skips */
			which_chunk_lbn = byte_offset / chunk_data_bytes;

			/* factor in skips */
			offset_lbn = which_chunk_lbn * chunk_lbn_count;

			/* convert to physical byte offset for lbn */
			offset_bytes = LBN_TO_BYTE(offset_lbn, vp->v_vfsp);

			/* don't forget offset into lbn */
			offset_extra = byte_offset % chunk_data_bytes;

			/* get virtual block number for driver */
			driver_block =
			    lbtodb(bof + xarsiz + offset_bytes + offset_extra);

			if (lastp != searchp) {
				/* this branch taken first time through loop */
				va = vas[count] =
				    ppmapin(searchp, PROT_WRITE, (caddr_t)-1);
				/* ppmapin() guarantees not to return NULL */
			} else {
				vas[count] = NULL;
			}

			bufs[count].b_un.b_addr = va + byte_offset % PAGESIZE;
			bufs[count].b_offset =
			    (offset_t)(byte_offset - io_off + off);

			/*
			 * We specifically use the b_lblkno member here
			 * as even in the 32 bit world driver_block can
			 * get very large in line with the ISO9660 spec.
			 */

			bufs[count].b_lblkno = driver_block;

			remaining_bytes =
			    ((which_chunk_lbn + 1) * chunk_data_bytes)
			    - byte_offset;

			/*
			 * remaining_bytes can't be zero, as we derived
			 * which_chunk_lbn directly from byte_offset.
			 */
			if ((remaining_bytes + byte_offset) < (off + len)) {
				/* coalesce-read the rest of the chunk */
				bufs[count].b_bcount = remaining_bytes;
			} else {
				/* get the final bits */
				bufs[count].b_bcount = off + len - byte_offset;
			}

			/*
			 * It would be nice to do multiple pages'
			 * worth at once here when the opportunity
			 * arises, as that has been shown to improve
			 * our wall time.  However, to do that
			 * requires that we use the pageio subsystem,
			 * which doesn't mix well with what we're
			 * already using here.  We can't use pageio
			 * all the time, because that subsystem
			 * assumes that a page is stored in N
			 * contiguous blocks on the device.
			 * Interleaving violates that assumption.
			 *
			 * Update: This is now not so big a problem
			 * because of the I/O scheduler sitting below
			 * that can re-order and coalesce I/O requests.
			 */

			remainder = PAGESIZE - (byte_offset % PAGESIZE);
			if (bufs[count].b_bcount > remainder) {
				bufs[count].b_bcount = remainder;
			}

			bufs[count].b_bufsize = bufs[count].b_bcount;
			if (((offset_t)byte_offset + bufs[count].b_bcount) >
			    HS_MAXFILEOFF) {
				break;
			}
			byte_offset += bufs[count].b_bcount;

			if (fsp->hqueue == NULL) {
				(void) bdev_strategy(&bufs[count]);

			} else {
				/*
				 * We are scheduling I/O so we need to enqueue
				 * requests rather than calling bdev_strategy
				 * here. A later invocation of the scheduling
				 * function will take care of doing the actual
				 * I/O as it selects requests from the queue as
				 * per the scheduling logic.
				 */
				struct hio *hsio = kmem_cache_alloc(hio_cache,
				    KM_SLEEP);

				sema_init(&fio_done[count], 0, NULL,
				    SEMA_DEFAULT, NULL);
				hsio->bp = &bufs[count];
				hsio->sema = &fio_done[count];
				hsio->io_lblkno = bufs[count].b_lblkno;
				hsio->nblocks = howmany(hsio->bp->b_bcount,
				    DEV_BSIZE);

				/* used for deadline */
				hsio->io_timestamp =
				    drv_hztousec(ddi_get_lbolt());

				/* for I/O coalescing */
				hsio->contig_chain = NULL;
				hsched_enqueue_io(fsp, hsio, 0);
			}

			lwp_stat_update(LWP_STAT_INBLK, 1);
			lastp = searchp;
			if ((remainder - bufs[count].b_bcount) < 1) {
				searchp = searchp->p_next;
			}
		}

		bufsused = count;
		/* Now wait for everything to come in */
		if (fsp->hqueue == NULL) {
			for (count = 0; count < bufsused; count++) {
				if (err == 0) {
					err = biowait(&bufs[count]);
				} else
					(void) biowait(&bufs[count]);
			}
		} else {
			for (count = 0; count < bufsused; count++) {
				struct buf *wbuf;

				/*
				 * Invoke scheduling function till our buf
				 * is processed. In doing this it might
				 * process bufs enqueued by other threads
				 * which is good.
				 */
				wbuf = &bufs[count];
				DTRACE_PROBE1(hsfs_io_wait, struct buf *, wbuf);
				while (sema_tryp(&fio_done[count]) == 0) {
					/*
					 * hsched_invoke_strategy will return 1
					 * if the I/O queue is empty. This means
					 * that there is another thread who has
					 * issued our buf and is waiting. So we
					 * just block instead of spinning.
					 */
					if (hsched_invoke_strategy(fsp)) {
						sema_p(&fio_done[count]);
						break;
					}
				}
				sema_destroy(&fio_done[count]);
				DTRACE_PROBE1(hsfs_io_done, struct buf *, wbuf);

				if (err == 0) {
					err = geterror(wbuf);
				}
			}
			kmem_free(fio_done, bufcnt * sizeof (ksema_t));
		}

		/* Don't leak resources */
		for (count = 0; count < bufcnt; count++) {
			biofini(&bufs[count]);
			if (count < bufsused && vas[count] != NULL) {
				ppmapout(vas[count]);
			}
		}

		kmem_free(vas, bufcnt * sizeof (caddr_t));
		kmem_free(bufs, bufcnt * sizeof (struct buf));
	}

	if (err) {
		pvn_read_done(pp, B_ERROR);
		return (err);
	}

	/*
	 * Lock the requested page, and the one after it if possible.
	 * Don't bother if our caller hasn't given us a place to stash
	 * the page pointers, since otherwise we'd lock pages that would
	 * never get unlocked.
	 */
	if (pagefound) {
		int index;
		ulong_t soff;

		/*
		 * Make sure it's in memory before we say it's here.
		 */
		if ((pp = page_lookup(vp, off, SE_SHARED)) == NULL) {
			hsfs_lostpage++;
			goto reread;
		}

		pl[0] = pp;
		index = 1;
		atomic_inc_64(&fsp->cache_read_pages);

		/*
		 * Try to lock the next page, if it exists, without
		 * blocking.
		 */
		plsz -= PAGESIZE;
		/* LINTED (plsz is unsigned) */
		for (soff = off + PAGESIZE; plsz > 0;
		    soff += PAGESIZE, plsz -= PAGESIZE) {
			pp = page_lookup_nowait(vp, (u_offset_t)soff,
			    SE_SHARED);
			if (pp == NULL)
				break;
			pl[index++] = pp;
		}
		pl[index] = NULL;

		/*
		 * Schedule a semi-asynchronous readahead if we are
		 * accessing the last cached page for the current
		 * file.
		 *
		 * Doing this here means that readaheads will be
		 * issued only if cache-hits occur. This is an advantage
		 * since cache-hits would mean that readahead is giving
		 * the desired benefit. If cache-hits do not occur there
		 * is no point in reading ahead of time - the system
		 * is loaded anyway.
		 */
		if (fsp->hqueue != NULL &&
		    hp->hs_prev_offset - off == PAGESIZE &&
		    hp->hs_prev_offset < filsiz &&
		    hp->hs_ra_bytes > 0 &&
		    !page_exists(vp, hp->hs_prev_offset)) {
			(void) hsfs_getpage_ra(vp, hp->hs_prev_offset, seg,
			    addr + PAGESIZE, hp, fsp, xarsiz, bof,
			    chunk_lbn_count, chunk_data_bytes);
		}

		return (0);
	}

	if (pp != NULL) {
		pvn_plist_init(pp, pl, plsz, off, io_len, rw);
	}

	return (err);
}

/*ARGSUSED*/
static int
hsfs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
    struct page *pl[], size_t plsz, struct seg *seg, caddr_t addr,
    enum seg_rw rw, struct cred *cred, caller_context_t *ct)
{
	uint_t filsiz;
	struct hsfs *fsp;
	struct hsnode *hp;

	fsp = VFS_TO_HSFS(vp->v_vfsp);
	hp = VTOH(vp);

	/* does not support write */
	if (rw == S_WRITE) {
		return (EROFS);
	}

	if (vp->v_flag & VNOMAP) {
		return (ENOSYS);
	}

	ASSERT(off <= HS_MAXFILEOFF);

	/*
	 * Determine file data size for EOF check.
	 */
	filsiz = hp->hs_dirent.ext_size;
	if ((off + len) > (offset_t)(filsiz + PAGEOFFSET) && seg != segkmap)
		return (EFAULT);	/* beyond EOF */

	/*
	 * Async Read-ahead computation.
	 * This attempts to detect sequential access pattern and
	 * enables reading extra pages ahead of time.
	 */
	if (fsp->hqueue != NULL) {
		/*
		 * This check for sequential access also takes into
		 * account segmap weirdness when reading in chunks
		 * less than the segmap size of 8K.
		 */
		if (hp->hs_prev_offset == off || (off <
		    hp->hs_prev_offset && off + MAX(len, PAGESIZE)
		    >= hp->hs_prev_offset)) {
			if (hp->hs_num_contig <
			    (seq_contig_requests - 1)) {
				hp->hs_num_contig++;

			} else {
				/*
				 * We increase readahead quantum till
				 * a predefined max. max_readahead_bytes
				 * is a multiple of PAGESIZE.
				 */
				if (hp->hs_ra_bytes <
				    fsp->hqueue->max_ra_bytes) {
					hp->hs_ra_bytes += PAGESIZE;
				}
			}
		} else {
			/*
			 * Not contiguous so reduce read ahead counters.
			 */
			if (hp->hs_ra_bytes > 0)
				hp->hs_ra_bytes -= PAGESIZE;

			if (hp->hs_ra_bytes <= 0) {
				hp->hs_ra_bytes = 0;
				if (hp->hs_num_contig > 0)
					hp->hs_num_contig--;
			}
		}
		/*
		 * Length must be rounded up to page boundary.
		 * since we read in units of pages.
		 */
		hp->hs_prev_offset = off + roundup(len, PAGESIZE);
		DTRACE_PROBE1(hsfs_compute_ra, struct hsnode *, hp);
	}
	if (protp != NULL)
		*protp = PROT_ALL;

	return (pvn_getpages(hsfs_getapage, vp, off, len, protp, pl, plsz,
	    seg, addr, rw, cred));
}



/*
 * This function should never be called. We need to have it to pass
 * it as an argument to other functions.
 */
/*ARGSUSED*/
int
hsfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
    int flags, cred_t *cr)
{
	/* should never happen - just destroy it */
	cmn_err(CE_NOTE, "hsfs_putapage: dirty HSFS page");
	pvn_write_done(pp, B_ERROR | B_WRITE | B_INVAL | B_FORCE | flags);
	return (0);
}


/*
 * The only flags we support are B_INVAL, B_FREE and B_DONTNEED.
 * B_INVAL is set by:
 *
 *	1) the MC_SYNC command of memcntl(2) to support the MS_INVALIDATE flag.
 *	2) the MC_ADVISE command of memcntl(2) with the MADV_DONTNEED advice
 *	   which translates to an MC_SYNC with the MS_INVALIDATE flag.
 *
 * The B_FREE (as well as the B_DONTNEED) flag is set when the
 * MADV_SEQUENTIAL advice has been used. VOP_PUTPAGE is invoked
 * from SEGVN to release pages behind a pagefault.
 */
/*ARGSUSED*/
static int
hsfs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
    struct cred *cr, caller_context_t *ct)
{
	int error = 0;

	if (vp->v_count == 0) {
		panic("hsfs_putpage: bad v_count");
		/*NOTREACHED*/
	}

	if (vp->v_flag & VNOMAP)
		return (ENOSYS);

	ASSERT(off <= HS_MAXFILEOFF);

	if (!vn_has_cached_data(vp))	/* no pages mapped */
		return (0);

	if (len == 0) {		/* from 'off' to EOF */
		error = pvn_vplist_dirty(vp, off, hsfs_putapage, flags, cr);
	} else {
		offset_t end_off = off + len;
		offset_t file_size = VTOH(vp)->hs_dirent.ext_size;
		offset_t io_off;

		file_size = (file_size + PAGESIZE - 1) & PAGEMASK;
		if (end_off > file_size)
			end_off = file_size;

		for (io_off = off; io_off < end_off; io_off += PAGESIZE) {
			page_t *pp;

			/*
			 * We insist on getting the page only if we are
			 * about to invalidate, free or write it and
			 * the B_ASYNC flag is not set.
			 */
			if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
				pp = page_lookup(vp, io_off,
				    (flags & (B_INVAL | B_FREE)) ?
				    SE_EXCL : SE_SHARED);
			} else {
				pp = page_lookup_nowait(vp, io_off,
				    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
			}

			if (pp == NULL)
				continue;

			/*
			 * Normally pvn_getdirty() should return 0, which
			 * impies that it has done the job for us.
			 * The shouldn't-happen scenario is when it returns 1.
			 * This means that the page has been modified and
			 * needs to be put back.
			 * Since we can't write on a CD, we fake a failed
			 * I/O and force pvn_write_done() to destroy the page.
			 */
			if (pvn_getdirty(pp, flags) == 1) {
				cmn_err(CE_NOTE,
				    "hsfs_putpage: dirty HSFS page");
				pvn_write_done(pp, flags |
				    B_ERROR | B_WRITE | B_INVAL | B_FORCE);
			}
		}
	}
	return (error);
}


/*ARGSUSED*/
static int
hsfs_map(struct vnode *vp, offset_t off, struct as *as, caddr_t *addrp,
    size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, struct cred *cred,
    caller_context_t *ct)
{
	struct segvn_crargs vn_a;
	int error;

	/* VFS_RECORD(vp->v_vfsp, VS_MAP, VS_CALL); */

	if (vp->v_flag & VNOMAP)
		return (ENOSYS);

	if ((prot & PROT_WRITE) && (flags & MAP_SHARED))
		return (ENOSYS);

	if (off > HS_MAXFILEOFF || off < 0 ||
	    (off + len) < 0 || (off + len) > HS_MAXFILEOFF)
		return (ENXIO);

	if (vp->v_type != VREG) {
		return (ENODEV);
	}

	/*
	 * If file is being locked, disallow mapping.
	 */
	if (vn_has_mandatory_locks(vp, VTOH(vp)->hs_dirent.mode))
		return (EAGAIN);

	as_rangelock(as);
	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
	if (error != 0) {
		as_rangeunlock(as);
		return (error);
	}

	vn_a.vp = vp;
	vn_a.offset = off;
	vn_a.type = flags & MAP_TYPE;
	vn_a.prot = prot;
	vn_a.maxprot = maxprot;
	vn_a.flags = flags & ~MAP_TYPE;
	vn_a.cred = cred;
	vn_a.amp = NULL;
	vn_a.szc = 0;
	vn_a.lgrp_mem_policy_flags = 0;

	error = as_map(as, *addrp, len, segvn_create, &vn_a);
	as_rangeunlock(as);
	return (error);
}

/* ARGSUSED */
static int
hsfs_addmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
    size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, struct cred *cr,
    caller_context_t *ct)
{
	struct hsnode *hp;

	if (vp->v_flag & VNOMAP)
		return (ENOSYS);

	hp = VTOH(vp);
	mutex_enter(&hp->hs_contents_lock);
	hp->hs_mapcnt += btopr(len);
	mutex_exit(&hp->hs_contents_lock);
	return (0);
}

/*ARGSUSED*/
static int
hsfs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
    size_t len, uint_t prot, uint_t maxprot, uint_t flags, struct cred *cr,
    caller_context_t *ct)
{
	struct hsnode *hp;

	if (vp->v_flag & VNOMAP)
		return (ENOSYS);

	hp = VTOH(vp);
	mutex_enter(&hp->hs_contents_lock);
	hp->hs_mapcnt -= btopr(len);	/* Count released mappings */
	ASSERT(hp->hs_mapcnt >= 0);
	mutex_exit(&hp->hs_contents_lock);
	return (0);
}

/* ARGSUSED */
static int
hsfs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp,
    caller_context_t *ct)
{
	return (*noffp < 0 ? EINVAL : 0);
}

/* ARGSUSED */
static int
hsfs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
    offset_t offset, struct flk_callback *flk_cbp, cred_t *cr,
    caller_context_t *ct)
{
	struct hsnode *hp = VTOH(vp);

	/*
	 * If the file is being mapped, disallow fs_frlock.
	 * We are not holding the hs_contents_lock while checking
	 * hs_mapcnt because the current locking strategy drops all
	 * locks before calling fs_frlock.
	 * So, hs_mapcnt could change before we enter fs_frlock making
	 * it meaningless to have held hs_contents_lock in the first place.
	 */
	if (hp->hs_mapcnt > 0 && MANDLOCK(vp, hp->hs_dirent.mode))
		return (EAGAIN);

	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
}

static int
hsched_deadline_compare(const void *x1, const void *x2)
{
	const struct hio *h1 = x1;
	const struct hio *h2 = x2;

	if (h1->io_timestamp < h2->io_timestamp)
		return (-1);
	if (h1->io_timestamp > h2->io_timestamp)
		return (1);

	if (h1->io_lblkno < h2->io_lblkno)
		return (-1);
	if (h1->io_lblkno > h2->io_lblkno)
		return (1);

	if (h1 < h2)
		return (-1);
	if (h1 > h2)
		return (1);

	return (0);
}

static int
hsched_offset_compare(const void *x1, const void *x2)
{
	const struct hio *h1 = x1;
	const struct hio *h2 = x2;

	if (h1->io_lblkno < h2->io_lblkno)
		return (-1);
	if (h1->io_lblkno > h2->io_lblkno)
		return (1);

	if (h1 < h2)
		return (-1);
	if (h1 > h2)
		return (1);

	return (0);
}

void
hsched_init_caches(void)
{
	hio_cache = kmem_cache_create("hsfs_hio_cache",
	    sizeof (struct hio), 0, NULL,
	    NULL, NULL, NULL, NULL, 0);

	hio_info_cache = kmem_cache_create("hsfs_hio_info_cache",
	    sizeof (struct hio_info), 0, NULL,
	    NULL, NULL, NULL, NULL, 0);
}

void
hsched_fini_caches(void)
{
	kmem_cache_destroy(hio_cache);
	kmem_cache_destroy(hio_info_cache);
}

/*
 * Initialize I/O scheduling structures. This is called via hsfs_mount
 */
void
hsched_init(struct hsfs *fsp, int fsid, struct modlinkage *modlinkage)
{
	struct hsfs_queue *hqueue = fsp->hqueue;
	struct vnode *vp = fsp->hsfs_devvp;

	/* TaskQ name of the form: hsched_task_ + stringof(int) */
	char namebuf[23];
	int error, err;
	struct dk_cinfo info;
	ldi_handle_t lh;
	ldi_ident_t li;

	/*
	 * Default maxtransfer = 16k chunk
	 */
	hqueue->dev_maxtransfer = 16384;

	/*
	 * Try to fetch the maximum device transfer size. This is used to
	 * ensure that a coalesced block does not exceed the maxtransfer.
	 */
	err  = ldi_ident_from_mod(modlinkage, &li);
	if (err) {
		cmn_err(CE_NOTE, "hsched_init: Querying device failed");
		cmn_err(CE_NOTE, "hsched_init: ldi_ident_from_mod err=%d\n",
		    err);
		goto set_ra;
	}

	err = ldi_open_by_dev(&(vp->v_rdev), OTYP_CHR, FREAD, CRED(), &lh, li);
	ldi_ident_release(li);
	if (err) {
		cmn_err(CE_NOTE, "hsched_init: Querying device failed");
		cmn_err(CE_NOTE, "hsched_init: ldi_open err=%d\n", err);
		goto set_ra;
	}

	error = ldi_ioctl(lh, DKIOCINFO, (intptr_t)&info, FKIOCTL,
	    CRED(), &err);
	err = ldi_close(lh, FREAD, CRED());
	if (err) {
		cmn_err(CE_NOTE, "hsched_init: Querying device failed");
		cmn_err(CE_NOTE, "hsched_init: ldi_close err=%d\n", err);
	}

	if (error == 0) {
		hqueue->dev_maxtransfer = ldbtob(info.dki_maxtransfer);
	}

set_ra:
	/*
	 * Max size of data to read ahead for sequential access pattern.
	 * Conservative to avoid letting the underlying CD drive to spin
	 * down, in case the application is reading slowly.
	 * We read ahead upto a max of 4 pages.
	 */
	hqueue->max_ra_bytes = PAGESIZE * 8;

	mutex_init(&(hqueue->hsfs_queue_lock), NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&(hqueue->strategy_lock), NULL, MUTEX_DEFAULT, NULL);
	avl_create(&(hqueue->read_tree), hsched_offset_compare,
	    sizeof (struct hio), offsetof(struct hio, io_offset_node));
	avl_create(&(hqueue->deadline_tree), hsched_deadline_compare,
	    sizeof (struct hio), offsetof(struct hio, io_deadline_node));

	(void) snprintf(namebuf, sizeof (namebuf), "hsched_task_%d", fsid);
	hqueue->ra_task = taskq_create(namebuf, hsfs_taskq_nthreads,
	    minclsyspri + 2, 1, 104857600 / PAGESIZE, TASKQ_DYNAMIC);

	hqueue->next = NULL;
	hqueue->nbuf = kmem_zalloc(sizeof (struct buf), KM_SLEEP);
}

void
hsched_fini(struct hsfs_queue *hqueue)
{
	if (hqueue != NULL) {
		/*
		 * Remove the sentinel if there was one.
		 */
		if (hqueue->next != NULL) {
			avl_remove(&hqueue->read_tree, hqueue->next);
			kmem_cache_free(hio_cache, hqueue->next);
		}
		avl_destroy(&(hqueue->read_tree));
		avl_destroy(&(hqueue->deadline_tree));
		mutex_destroy(&(hqueue->hsfs_queue_lock));
		mutex_destroy(&(hqueue->strategy_lock));

		/*
		 * If there are any existing readahead threads running
		 * taskq_destroy will wait for them to finish.
		 */
		taskq_destroy(hqueue->ra_task);
		kmem_free(hqueue->nbuf, sizeof (struct buf));
	}
}

/*
 * Determine if two I/O requests are adjacent to each other so
 * that they can coalesced.
 */
#define	IS_ADJACENT(io, nio) \
	(((io)->io_lblkno + (io)->nblocks == (nio)->io_lblkno) && \
	(io)->bp->b_edev == (nio)->bp->b_edev)

/*
 * This performs the actual I/O scheduling logic. We use the Circular
 * Look algorithm here. Sort the I/O requests in ascending order of
 * logical block number and process them starting with the lowest
 * numbered block and progressing towards higher block numbers in the
 * queue. Once there are no more higher numbered blocks, start again
 * with the lowest one. This is good for CD/DVD as you keep moving
 * the head in one direction along the outward spiral track and avoid
 * too many seeks as much as possible. The re-ordering also allows
 * us to coalesce adjacent requests into one larger request.
 * This is thus essentially a 1-way Elevator with front merging.
 *
 * In addition each read request here has a deadline and will be
 * processed out of turn if the deadline (500ms) expires.
 *
 * This function is necessarily serialized via hqueue->strategy_lock.
 * This function sits just below hsfs_getapage and processes all read
 * requests orginating from that function.
 */
int
hsched_invoke_strategy(struct hsfs *fsp)
{
	struct hsfs_queue *hqueue;
	struct buf *nbuf;
	struct hio *fio, *nio, *tio, *prev, *last;
	size_t bsize, soffset, offset, data;
	int bioret, bufcount;
	struct vnode *fvp;
	ksema_t *io_done;
	caddr_t iodata;

	hqueue = fsp->hqueue;
	mutex_enter(&hqueue->strategy_lock);
	mutex_enter(&hqueue->hsfs_queue_lock);

	/*
	 * Check for Deadline expiration first
	 */
	fio = avl_first(&hqueue->deadline_tree);

	/*
	 * Paranoid check for empty I/O queue. Both deadline
	 * and read trees contain same data sorted in different
	 * ways. So empty deadline tree = empty read tree.
	 */
	if (fio == NULL) {
		/*
		 * Remove the sentinel if there was one.
		 */
		if (hqueue->next != NULL) {
			avl_remove(&hqueue->read_tree, hqueue->next);
			kmem_cache_free(hio_cache, hqueue->next);
			hqueue->next = NULL;
		}
		mutex_exit(&hqueue->hsfs_queue_lock);
		mutex_exit(&hqueue->strategy_lock);
		return (1);
	}

	if (drv_hztousec(ddi_get_lbolt()) - fio->io_timestamp
	    < HSFS_READ_DEADLINE) {
		/*
		 * Apply standard scheduling logic. This uses the
		 * C-LOOK approach. Process I/O requests in ascending
		 * order of logical block address till no subsequent
		 * higher numbered block request remains. Then start
		 * again from the lowest numbered block in the queue.
		 *
		 * We do this cheaply here by means of a sentinel.
		 * The last processed I/O structure from the previous
		 * invocation of this func, is left dangling in the
		 * read_tree so that we can easily scan to the next
		 * higher numbered request and remove the sentinel.
		 */
		fio = NULL;
		if (hqueue->next != NULL) {
			fio = AVL_NEXT(&hqueue->read_tree, hqueue->next);
			avl_remove(&hqueue->read_tree, hqueue->next);
			kmem_cache_free(hio_cache, hqueue->next);
			hqueue->next = NULL;
		}
		if (fio == NULL) {
			fio = avl_first(&hqueue->read_tree);
		}
	} else if (hqueue->next != NULL) {
		DTRACE_PROBE1(hsfs_deadline_expiry, struct hio *, fio);

		avl_remove(&hqueue->read_tree, hqueue->next);
		kmem_cache_free(hio_cache, hqueue->next);
		hqueue->next = NULL;
	}

	/*
	 * In addition we try to coalesce contiguous
	 * requests into one bigger request.
	 */
	bufcount = 1;
	bsize = ldbtob(fio->nblocks);
	fvp = fio->bp->b_file;
	nio = AVL_NEXT(&hqueue->read_tree, fio);
	tio = fio;
	while (nio != NULL && IS_ADJACENT(tio, nio) &&
	    bsize < hqueue->dev_maxtransfer) {
		avl_remove(&hqueue->deadline_tree, tio);
		avl_remove(&hqueue->read_tree, tio);
		tio->contig_chain = nio;
		bsize += ldbtob(nio->nblocks);
		prev = tio;
		tio = nio;

		/*
		 * This check is required to detect the case where
		 * we are merging adjacent buffers belonging to
		 * different files. fvp is used to set the b_file
		 * parameter in the coalesced buf. b_file is used
		 * by DTrace so we do not want DTrace to accrue
		 * requests to two different files to any one file.
		 */
		if (fvp && tio->bp->b_file != fvp) {
			fvp = NULL;
		}

		nio = AVL_NEXT(&hqueue->read_tree, nio);
		bufcount++;
	}

	/*
	 * tio is not removed from the read_tree as it serves as a sentinel
	 * to cheaply allow us to scan to the next higher numbered I/O
	 * request.
	 */
	hqueue->next = tio;
	avl_remove(&hqueue->deadline_tree, tio);
	mutex_exit(&hqueue->hsfs_queue_lock);
	DTRACE_PROBE3(hsfs_io_dequeued, struct hio *, fio, int, bufcount,
	    size_t, bsize);

	/*
	 * The benefit of coalescing occurs if the the savings in I/O outweighs
	 * the cost of doing the additional work below.
	 * It was observed that coalescing 2 buffers results in diminishing
	 * returns, so we do coalescing if we have >2 adjacent bufs.
	 */
	if (bufcount > hsched_coalesce_min) {
		/*
		 * We have coalesced blocks. First allocate mem and buf for
		 * the entire coalesced chunk.
		 * Since we are guaranteed single-threaded here we pre-allocate
		 * one buf at mount time and that is re-used every time. This
		 * is a synthesized buf structure that uses kmem_alloced chunk.
		 * Not quite a normal buf attached to pages.
		 */
		fsp->coalesced_bytes += bsize;
		nbuf = hqueue->nbuf;
		bioinit(nbuf);
		nbuf->b_edev = fio->bp->b_edev;
		nbuf->b_dev = fio->bp->b_dev;
		nbuf->b_flags = fio->bp->b_flags;
		nbuf->b_iodone = fio->bp->b_iodone;
		iodata = kmem_alloc(bsize, KM_SLEEP);
		nbuf->b_un.b_addr = iodata;
		nbuf->b_lblkno = fio->bp->b_lblkno;
		nbuf->b_vp = fvp;
		nbuf->b_file = fvp;
		nbuf->b_bcount = bsize;
		nbuf->b_bufsize = bsize;

		DTRACE_PROBE3(hsfs_coalesced_io_start, struct hio *, fio, int,
		    bufcount, size_t, bsize);

		/*
		 * Perform I/O for the coalesced block.
		 */
		(void) bdev_strategy(nbuf);

		/*
		 * Duplicate the last IO node to leave the sentinel alone.
		 * The sentinel is freed in the next invocation of this
		 * function.
		 */
		prev->contig_chain = kmem_cache_alloc(hio_cache, KM_SLEEP);
		prev->contig_chain->bp = tio->bp;
		prev->contig_chain->sema = tio->sema;
		tio = prev->contig_chain;
		tio->contig_chain = NULL;
		soffset = ldbtob(fio->bp->b_lblkno);
		nio = fio;

		bioret = biowait(nbuf);
		data = bsize - nbuf->b_resid;
		biofini(nbuf);
		mutex_exit(&hqueue->strategy_lock);

		/*
		 * We use the b_resid parameter to detect how much
		 * data was succesfully transferred. We will signal
		 * a success to all the fully retrieved actual bufs
		 * before coalescing, rest is signaled as error,
		 * if any.
		 */
		tio = nio;
		DTRACE_PROBE3(hsfs_coalesced_io_done, struct hio *, nio,
		    int, bioret, size_t, data);

		/*
		 * Copy data and signal success to all the bufs
		 * which can be fully satisfied from b_resid.
		 */
		while (nio != NULL && data >= nio->bp->b_bcount) {
			offset = ldbtob(nio->bp->b_lblkno) - soffset;
			bcopy(iodata + offset, nio->bp->b_un.b_addr,
			    nio->bp->b_bcount);
			data -= nio->bp->b_bcount;
			bioerror(nio->bp, 0);
			biodone(nio->bp);
			sema_v(nio->sema);
			tio = nio;
			nio = nio->contig_chain;
			kmem_cache_free(hio_cache, tio);
		}

		/*
		 * Signal error to all the leftover bufs (if any)
		 * after b_resid data is exhausted.
		 */
		while (nio != NULL) {
			nio->bp->b_resid = nio->bp->b_bcount - data;
			bzero(nio->bp->b_un.b_addr + data, nio->bp->b_resid);
			bioerror(nio->bp, bioret);
			biodone(nio->bp);
			sema_v(nio->sema);
			tio = nio;
			nio = nio->contig_chain;
			kmem_cache_free(hio_cache, tio);
			data = 0;
		}
		kmem_free(iodata, bsize);
	} else {

		nbuf = tio->bp;
		io_done = tio->sema;
		nio = fio;
		last = tio;

		while (nio != NULL) {
			(void) bdev_strategy(nio->bp);
			nio = nio->contig_chain;
		}
		nio = fio;
		mutex_exit(&hqueue->strategy_lock);

		while (nio != NULL) {
			if (nio == last) {
				(void) biowait(nbuf);
				sema_v(io_done);
				break;
				/* sentinel last not freed. See above. */
			} else {
				(void) biowait(nio->bp);
				sema_v(nio->sema);
			}
			tio = nio;
			nio = nio->contig_chain;
			kmem_cache_free(hio_cache, tio);
		}
	}
	return (0);
}

/*
 * Insert an I/O request in the I/O scheduler's pipeline
 * Using AVL tree makes it easy to reorder the I/O request
 * based on logical block number.
 */
static void
hsched_enqueue_io(struct hsfs *fsp, struct hio *hsio, int ra)
{
	struct hsfs_queue *hqueue = fsp->hqueue;

	mutex_enter(&hqueue->hsfs_queue_lock);

	fsp->physical_read_bytes += hsio->bp->b_bcount;
	if (ra)
		fsp->readahead_bytes += hsio->bp->b_bcount;

	avl_add(&hqueue->deadline_tree, hsio);
	avl_add(&hqueue->read_tree, hsio);

	DTRACE_PROBE3(hsfs_io_enqueued, struct hio *, hsio,
	    struct hsfs_queue *, hqueue, int, ra);

	mutex_exit(&hqueue->hsfs_queue_lock);
}

/* ARGSUSED */
static int
hsfs_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr,
    caller_context_t *ct)
{
	struct hsfs	*fsp;

	int		error = 0;

	switch (cmd) {

	case _PC_NAME_MAX:
		fsp = VFS_TO_HSFS(vp->v_vfsp);
		*valp = fsp->hsfs_namemax;
		break;

	case _PC_FILESIZEBITS:
		*valp = 33;	/* Without multi extent support: 4 GB - 2k */
		break;

	case _PC_TIMESTAMP_RESOLUTION:
		/*
		 * HSFS keeps, at best, 1/100 second timestamp resolution.
		 */
		*valp = 10000000L;
		break;

	default:
		error = fs_pathconf(vp, cmd, valp, cr, ct);
		break;
	}

	return (error);
}



const fs_operation_def_t hsfs_vnodeops_template[] = {
	VOPNAME_OPEN,		{ .vop_open = hsfs_open },
	VOPNAME_CLOSE,		{ .vop_close = hsfs_close },
	VOPNAME_READ,		{ .vop_read = hsfs_read },
	VOPNAME_GETATTR,	{ .vop_getattr = hsfs_getattr },
	VOPNAME_ACCESS,		{ .vop_access = hsfs_access },
	VOPNAME_LOOKUP,		{ .vop_lookup = hsfs_lookup },
	VOPNAME_READDIR,	{ .vop_readdir = hsfs_readdir },
	VOPNAME_READLINK,	{ .vop_readlink = hsfs_readlink },
	VOPNAME_FSYNC,		{ .vop_fsync = hsfs_fsync },
	VOPNAME_INACTIVE,	{ .vop_inactive = hsfs_inactive },
	VOPNAME_FID,		{ .vop_fid = hsfs_fid },
	VOPNAME_SEEK,		{ .vop_seek = hsfs_seek },
	VOPNAME_FRLOCK,		{ .vop_frlock = hsfs_frlock },
	VOPNAME_GETPAGE,	{ .vop_getpage = hsfs_getpage },
	VOPNAME_PUTPAGE,	{ .vop_putpage = hsfs_putpage },
	VOPNAME_MAP,		{ .vop_map = hsfs_map },
	VOPNAME_ADDMAP,		{ .vop_addmap = hsfs_addmap },
	VOPNAME_DELMAP,		{ .vop_delmap = hsfs_delmap },
	VOPNAME_PATHCONF,	{ .vop_pathconf = hsfs_pathconf },
	NULL,			NULL
};

struct vnodeops *hsfs_vnodeops;