17c478bd9Sstevel@tonic-gate /*
27c478bd9Sstevel@tonic-gate  * CDDL HEADER START
37c478bd9Sstevel@tonic-gate  *
47c478bd9Sstevel@tonic-gate  * The contents of this file are subject to the terms of the
59cbc422eSpeterte  * Common Development and Distribution License (the "License").
69cbc422eSpeterte  * You may not use this file except in compliance with the License.
77c478bd9Sstevel@tonic-gate  *
87c478bd9Sstevel@tonic-gate  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
97c478bd9Sstevel@tonic-gate  * or http://www.opensolaris.org/os/licensing.
107c478bd9Sstevel@tonic-gate  * See the License for the specific language governing permissions
117c478bd9Sstevel@tonic-gate  * and limitations under the License.
127c478bd9Sstevel@tonic-gate  *
137c478bd9Sstevel@tonic-gate  * When distributing Covered Code, include this CDDL HEADER in each
147c478bd9Sstevel@tonic-gate  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
157c478bd9Sstevel@tonic-gate  * If applicable, add the following below this CDDL HEADER, with the
167c478bd9Sstevel@tonic-gate  * fields enclosed by brackets "[]" replaced with your own identifying
177c478bd9Sstevel@tonic-gate  * information: Portions Copyright [yyyy] [name of copyright owner]
187c478bd9Sstevel@tonic-gate  *
197c478bd9Sstevel@tonic-gate  * CDDL HEADER END
207c478bd9Sstevel@tonic-gate  */
213b862e9aSRoger A. Faulkner 
227c478bd9Sstevel@tonic-gate /*
233b862e9aSRoger A. Faulkner  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
247c478bd9Sstevel@tonic-gate  * Use is subject to license terms.
2506e6833aSJosef 'Jeff' Sipek  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
26*ade42b55SSebastien Roy  * Copyright (c) 2017 by Delphix. All rights reserved.
277c478bd9Sstevel@tonic-gate  */
297c478bd9Sstevel@tonic-gate /*
307c478bd9Sstevel@tonic-gate  * Vnode operations for the High Sierra filesystem
317c478bd9Sstevel@tonic-gate  */
337c478bd9Sstevel@tonic-gate #include <sys/types.h>
347c478bd9Sstevel@tonic-gate #include <sys/t_lock.h>
357c478bd9Sstevel@tonic-gate #include <sys/param.h>
367c478bd9Sstevel@tonic-gate #include <sys/time.h>
377c478bd9Sstevel@tonic-gate #include <sys/systm.h>
387c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h>
397c478bd9Sstevel@tonic-gate #include <sys/resource.h>
407c478bd9Sstevel@tonic-gate #include <sys/signal.h>
417c478bd9Sstevel@tonic-gate #include <sys/cred.h>
427c478bd9Sstevel@tonic-gate #include <sys/user.h>
437c478bd9Sstevel@tonic-gate #include <sys/buf.h>
447c478bd9Sstevel@tonic-gate #include <sys/vfs.h>
45aa59c4cbSrsb #include <sys/vfs_opreg.h>
467c478bd9Sstevel@tonic-gate #include <sys/stat.h>
477c478bd9Sstevel@tonic-gate #include <sys/vnode.h>
487c478bd9Sstevel@tonic-gate #include <sys/mode.h>
497c478bd9Sstevel@tonic-gate #include <sys/proc.h>
507c478bd9Sstevel@tonic-gate #include <sys/disp.h>
517c478bd9Sstevel@tonic-gate #include <sys/file.h>
527c478bd9Sstevel@tonic-gate #include <sys/fcntl.h>
537c478bd9Sstevel@tonic-gate #include <sys/flock.h>
547c478bd9Sstevel@tonic-gate #include <sys/kmem.h>
557c478bd9Sstevel@tonic-gate #include <sys/uio.h>
567c478bd9Sstevel@tonic-gate #include <sys/conf.h>
577c478bd9Sstevel@tonic-gate #include <sys/errno.h>
587c478bd9Sstevel@tonic-gate #include <sys/mman.h>
597c478bd9Sstevel@tonic-gate #include <sys/pathname.h>
607c478bd9Sstevel@tonic-gate #include <sys/debug.h>
617c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h>
627c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h>
637c478bd9Sstevel@tonic-gate #include <sys/fbuf.h>
647c478bd9Sstevel@tonic-gate #include <sys/dirent.h>
657c478bd9Sstevel@tonic-gate #include <sys/errno.h>
6684b82766Smg #include <sys/dkio.h>
6784b82766Smg #include <sys/cmn_err.h>
6884b82766Smg #include <sys/atomic.h>
707c478bd9Sstevel@tonic-gate #include <vm/hat.h>
717c478bd9Sstevel@tonic-gate #include <vm/page.h>
727c478bd9Sstevel@tonic-gate #include <vm/pvn.h>
737c478bd9Sstevel@tonic-gate #include <vm/as.h>
747c478bd9Sstevel@tonic-gate #include <vm/seg.h>
757c478bd9Sstevel@tonic-gate #include <vm/seg_map.h>
767c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h>
777c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h>
787c478bd9Sstevel@tonic-gate #include <vm/rm.h>
797c478bd9Sstevel@tonic-gate #include <vm/page.h>
807c478bd9Sstevel@tonic-gate #include <sys/swap.h>
8184b82766Smg #include <sys/avl.h>
8284b82766Smg #include <sys/sunldi.h>
8384b82766Smg #include <sys/ddi.h>
8484b82766Smg #include <sys/sunddi.h>
8584b82766Smg #include <sys/sdt.h>
8784b82766Smg /*
8884b82766Smg  * For struct modlinkage
8984b82766Smg  */
9084b82766Smg #include <sys/modctl.h>
927c478bd9Sstevel@tonic-gate #include <sys/fs/hsfs_spec.h>
937c478bd9Sstevel@tonic-gate #include <sys/fs/hsfs_node.h>
947c478bd9Sstevel@tonic-gate #include <sys/fs/hsfs_impl.h>
957c478bd9Sstevel@tonic-gate #include <sys/fs/hsfs_susp.h>
967c478bd9Sstevel@tonic-gate #include <sys/fs/hsfs_rrip.h>
987c478bd9Sstevel@tonic-gate #include <fs/fs_subr.h>
10084b82766Smg /* # of contiguous requests to detect sequential access pattern */
10184b82766Smg static int seq_contig_requests = 2;
10384b82766Smg /*
10484b82766Smg  * This is the max number os taskq threads that will be created
10584b82766Smg  * if required. Since we are using a Dynamic TaskQ by default only
10684b82766Smg  * one thread is created initially.
10784b82766Smg  *
10884b82766Smg  * NOTE: In the usual hsfs use case this per fs instance number
10984b82766Smg  * of taskq threads should not place any undue load on a system.
11084b82766Smg  * Even on an unusual system with say 100 CDROM drives, 800 threads
11184b82766Smg  * will not be created unless all the drives are loaded and all
11284b82766Smg  * of them are saturated with I/O at the same time! If there is at
11384b82766Smg  * all a complaint of system load due to such an unusual case it
11484b82766Smg  * should be easy enough to change to one per-machine Dynamic TaskQ
11584b82766Smg  * for all hsfs mounts with a nthreads of say 32.
11684b82766Smg  */
11784b82766Smg static int hsfs_taskq_nthreads = 8;	/* # of taskq threads per fs */
11984b82766Smg /* Min count of adjacent bufs that will avoid buf coalescing */
12084b82766Smg static int hsched_coalesce_min = 2;
12284b82766Smg /*
12384b82766Smg  * Kmem caches for heavily used small allocations. Using these kmem
12484b82766Smg  * caches provides a factor of 3 reduction in system time and greatly
12584b82766Smg  * aids overall throughput esp. on SPARC.
12684b82766Smg  */
12784b82766Smg struct kmem_cache *hio_cache;
12884b82766Smg struct kmem_cache *hio_info_cache;
130d10b6702Sfrankho /*
131d10b6702Sfrankho  * This tunable allows us to ignore inode numbers from rrip-1.12.
132d10b6702Sfrankho  * In this case, we fall back to our default inode algorithm.
133d10b6702Sfrankho  */
134d10b6702Sfrankho extern int use_rrip_inodes;
13684b82766Smg static int hsched_deadline_compare(const void *x1, const void *x2);
13784b82766Smg static int hsched_offset_compare(const void *x1, const void *x2);
13884b82766Smg static void hsched_enqueue_io(struct hsfs *fsp, struct hio *hsio, int ra);
13984b82766Smg int hsched_invoke_strategy(struct hsfs *fsp);
1417c478bd9Sstevel@tonic-gate /* ARGSUSED */
1427c478bd9Sstevel@tonic-gate static int
hsfs_fsync(vnode_t * cp,int syncflag,cred_t * cred,caller_context_t * ct)143*ade42b55SSebastien Roy hsfs_fsync(vnode_t *cp, int syncflag, cred_t *cred, caller_context_t *ct)
1447c478bd9Sstevel@tonic-gate {
1457c478bd9Sstevel@tonic-gate 	return (0);
1467c478bd9Sstevel@tonic-gate }
1497c478bd9Sstevel@tonic-gate /*ARGSUSED*/
1507c478bd9Sstevel@tonic-gate static int
hsfs_read(struct vnode * vp,struct uio * uiop,int ioflag,struct cred * cred,struct caller_context * ct)151*ade42b55SSebastien Roy hsfs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cred,
152*ade42b55SSebastien Roy     struct caller_context *ct)
1537c478bd9Sstevel@tonic-gate {
1547c478bd9Sstevel@tonic-gate 	caddr_t base;
1558cd7c4fcSpeterte 	offset_t diff;
1567c478bd9Sstevel@tonic-gate 	int error;
1578cd7c4fcSpeterte 	struct hsnode *hp;
1588cd7c4fcSpeterte 	uint_t filesize;
1607c478bd9Sstevel@tonic-gate 	hp = VTOH(vp);
1617c478bd9Sstevel@tonic-gate 	/*
1627c478bd9Sstevel@tonic-gate 	 * if vp is of type VDIR, make sure dirent
1637c478bd9Sstevel@tonic-gate 	 * is filled up with all info (because of ptbl)
1647c478bd9Sstevel@tonic-gate 	 */
1657c478bd9Sstevel@tonic-gate 	if (vp->v_type == VDIR) {
1667c478bd9Sstevel@tonic-gate 		if (hp->hs_dirent.ext_size == 0)
1677c478bd9Sstevel@tonic-gate 			hs_filldirent(vp, &hp->hs_dirent);
1687c478bd9Sstevel@tonic-gate 	}
1697c478bd9Sstevel@tonic-gate 	filesize = hp->hs_dirent.ext_size;
1718cd7c4fcSpeterte 	/* Sanity checks. */
1728cd7c4fcSpeterte 	if (uiop->uio_resid == 0 ||		/* No data wanted. */
1739cbc422eSpeterte 	    uiop->uio_loffset > HS_MAXFILEOFF ||	/* Offset too big. */
1748cd7c4fcSpeterte 	    uiop->uio_loffset >= filesize)	/* Past EOF. */
1758cd7c4fcSpeterte 		return (0);
1777c478bd9Sstevel@tonic-gate 	do {
1788cd7c4fcSpeterte 		/*
1798cd7c4fcSpeterte 		 * We want to ask for only the "right" amount of data.
1808cd7c4fcSpeterte 		 * In this case that means:-
1818cd7c4fcSpeterte 		 *
1828cd7c4fcSpeterte 		 * We can't get data from beyond our EOF. If asked,
1838cd7c4fcSpeterte 		 * we will give a short read.
1848cd7c4fcSpeterte 		 *
1858cd7c4fcSpeterte 		 * segmap_getmapflt returns buffers of MAXBSIZE bytes.
1868cd7c4fcSpeterte 		 * These buffers are always MAXBSIZE aligned.
1878cd7c4fcSpeterte 		 * If our starting offset is not MAXBSIZE aligned,
1888cd7c4fcSpeterte 		 * we can only ask for less than MAXBSIZE bytes.
1898cd7c4fcSpeterte 		 *
1908cd7c4fcSpeterte 		 * If our requested offset and length are such that
1918cd7c4fcSpeterte 		 * they belong in different MAXBSIZE aligned slots
1928cd7c4fcSpeterte 		 * then we'll be making more than one call on
1938cd7c4fcSpeterte 		 * segmap_getmapflt.
1948cd7c4fcSpeterte 		 *
1958cd7c4fcSpeterte 		 * This diagram shows the variables we use and their
1968cd7c4fcSpeterte 		 * relationships.
1978cd7c4fcSpeterte 		 *
1988cd7c4fcSpeterte 		 * |<-----MAXBSIZE----->|
1998cd7c4fcSpeterte 		 * +--------------------------...+
2008cd7c4fcSpeterte 		 * |.....mapon->|<--n-->|....*...|EOF
2018cd7c4fcSpeterte 		 * +--------------------------...+
2028cd7c4fcSpeterte 		 * uio_loffset->|
2038cd7c4fcSpeterte 		 * uio_resid....|<---------->|
2048cd7c4fcSpeterte 		 * diff.........|<-------------->|
2058cd7c4fcSpeterte 		 *
2068cd7c4fcSpeterte 		 * So, in this case our offset is not aligned
2078cd7c4fcSpeterte 		 * and our request takes us outside of the
2088cd7c4fcSpeterte 		 * MAXBSIZE window. We will break this up into
2098cd7c4fcSpeterte 		 * two segmap_getmapflt calls.
2108cd7c4fcSpeterte 		 */
2118cd7c4fcSpeterte 		size_t nbytes;
2128cd7c4fcSpeterte 		offset_t mapon;
2138cd7c4fcSpeterte 		size_t n;
2148cd7c4fcSpeterte 		uint_t flags;
2168cd7c4fcSpeterte 		mapon = uiop->uio_loffset & MAXBOFFSET;
2178cd7c4fcSpeterte 		diff = filesize - uiop->uio_loffset;
2188cd7c4fcSpeterte 		nbytes = (size_t)MIN(MAXBSIZE - mapon, uiop->uio_resid);
2198cd7c4fcSpeterte 		n = MIN(diff, nbytes);
2208cd7c4fcSpeterte 		if (n <= 0) {
2218cd7c4fcSpeterte 			/* EOF or request satisfied. */
2228cd7c4fcSpeterte 			return (0);
2237c478bd9Sstevel@tonic-gate 		}
2258cd7c4fcSpeterte 		base = segmap_getmapflt(segkmap, vp,
2268cd7c4fcSpeterte 		    (u_offset_t)uiop->uio_loffset, n, 1, S_READ);
2288cd7c4fcSpeterte 		error = uiomove(base + mapon, n, UIO_READ, uiop);
2307c478bd9Sstevel@tonic-gate 		if (error == 0) {
2317c478bd9Sstevel@tonic-gate 			/*
2327c478bd9Sstevel@tonic-gate 			 * if read a whole block, or read to eof,
2337c478bd9Sstevel@tonic-gate 			 *  won't need this buffer again soon.
2347c478bd9Sstevel@tonic-gate 			 */
2358cd7c4fcSpeterte 			if (n + mapon == MAXBSIZE ||
2368cd7c4fcSpeterte 			    uiop->uio_loffset == filesize)
2377c478bd9Sstevel@tonic-gate 				flags = SM_DONTNEED;
2387c478bd9Sstevel@tonic-gate 			else
2397c478bd9Sstevel@tonic-gate 				flags = 0;
2417c478bd9Sstevel@tonic-gate 			error = segmap_release(segkmap, base, flags);
2427c478bd9Sstevel@tonic-gate 		} else
2437c478bd9Sstevel@tonic-gate 			(void) segmap_release(segkmap, base, 0);
2447c478bd9Sstevel@tonic-gate 	} while (error == 0 && uiop->uio_resid > 0);
2467c478bd9Sstevel@tonic-gate 	return (error);
2477c478bd9Sstevel@tonic-gate }
2497c478bd9Sstevel@tonic-gate /*ARGSUSED2*/
2507c478bd9Sstevel@tonic-gate static int
hsfs_getattr(struct vnode * vp,struct vattr * vap,int flags,struct cred * cred,caller_context_t * ct)251*ade42b55SSebastien Roy hsfs_getattr(struct vnode *vp, struct vattr *vap, int flags, struct cred *cred,
252*ade42b55SSebastien Roy     caller_context_t *ct)
2537c478bd9Sstevel@tonic-gate {
2547c478bd9Sstevel@tonic-gate 	struct hsnode *hp;
2557c478bd9Sstevel@tonic-gate 	struct vfs *vfsp;
2567c478bd9Sstevel@tonic-gate 	struct hsfs *fsp;
2587c478bd9Sstevel@tonic-gate 	hp = VTOH(vp);
2597c478bd9Sstevel@tonic-gate 	fsp = VFS_TO_HSFS(vp->v_vfsp);
2607c478bd9Sstevel@tonic-gate 	vfsp = vp->v_vfsp;
2627c478bd9Sstevel@tonic-gate 	if ((hp->hs_dirent.ext_size == 0) && (vp->v_type == VDIR)) {
2637c478bd9Sstevel@tonic-gate 		hs_filldirent(vp, &hp->hs_dirent);
2647c478bd9Sstevel@tonic-gate 	}
2657c478bd9Sstevel@tonic-gate 	vap->va_type = IFTOVT(hp->hs_dirent.mode);
2667c478bd9Sstevel@tonic-gate 	vap->va_mode = hp->hs_dirent.mode;
2677c478bd9Sstevel@tonic-gate 	vap->va_uid = hp->hs_dirent.uid;
2687c478bd9Sstevel@tonic-gate 	vap->va_gid = hp->hs_dirent.gid;
2707c478bd9Sstevel@tonic-gate 	vap->va_fsid = vfsp->vfs_dev;
2717c478bd9Sstevel@tonic-gate 	vap->va_nodeid = (ino64_t)hp->hs_nodeid;
2727c478bd9Sstevel@tonic-gate 	vap->va_nlink = hp->hs_dirent.nlink;
2737c478bd9Sstevel@tonic-gate 	vap->va_size =	(offset_t)hp->hs_dirent.ext_size;
2757c478bd9Sstevel@tonic-gate 	vap->va_atime.tv_sec = hp->hs_dirent.adate.tv_sec;
2767c478bd9Sstevel@tonic-gate 	vap->va_atime.tv_nsec = hp->hs_dirent.adate.tv_usec*1000;
2777c478bd9Sstevel@tonic-gate 	vap->va_mtime.tv_sec = hp->hs_dirent.mdate.tv_sec;
2787c478bd9Sstevel@tonic-gate 	vap->va_mtime.tv_nsec = hp->hs_dirent.mdate.tv_usec*1000;
2797c478bd9Sstevel@tonic-gate 	vap->va_ctime.tv_sec = hp->hs_dirent.cdate.tv_sec;
2807c478bd9Sstevel@tonic-gate 	vap->va_ctime.tv_nsec = hp->hs_dirent.cdate.tv_usec*1000;
2817c478bd9Sstevel@tonic-gate 	if (vp->v_type == VCHR || vp->v_type == VBLK)
2827c478bd9Sstevel@tonic-gate 		vap->va_rdev = hp->hs_dirent.r_dev;
2837c478bd9Sstevel@tonic-gate 	else
2847c478bd9Sstevel@tonic-gate 		vap->va_rdev = 0;
2857c478bd9Sstevel@tonic-gate 	vap->va_blksize = vfsp->vfs_bsize;
2867c478bd9Sstevel@tonic-gate 	/* no. of blocks = no. of data blocks + no. of xar blocks */
2877c478bd9Sstevel@tonic-gate 	vap->va_nblocks = (fsblkcnt64_t)howmany(vap->va_size + (u_longlong_t)
2887c478bd9Sstevel@tonic-gate 	    (hp->hs_dirent.xar_len << fsp->hsfs_vol.lbn_shift), DEV_BSIZE);
2897c478bd9Sstevel@tonic-gate 	vap->va_seq = hp->hs_seq;
2907c478bd9Sstevel@tonic-gate 	return (0);
2917c478bd9Sstevel@tonic-gate }
2937c478bd9Sstevel@tonic-gate /*ARGSUSED*/
2947c478bd9Sstevel@tonic-gate static int
hsfs_readlink(struct vnode * vp,struct uio * uiop,struct cred * cred,caller_context_t * ct)295*ade42b55SSebastien Roy hsfs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cred,
296*ade42b55SSebastien Roy     caller_context_t *ct)
2977c478bd9Sstevel@tonic-gate {
2987c478bd9Sstevel@tonic-gate 	struct hsnode *hp;
3007c478bd9Sstevel@tonic-gate 	if (vp->v_type != VLNK)
3017c478bd9Sstevel@tonic-gate 		return (EINVAL);
3037c478bd9Sstevel@tonic-gate 	hp = VTOH(vp);
3057c478bd9Sstevel@tonic-gate 	if (hp->hs_dirent.sym_link == (char *)NULL)
3067c478bd9Sstevel@tonic-gate 		return (ENOENT);
3087c478bd9Sstevel@tonic-gate 	return (uiomove(hp->hs_dirent.sym_link,
3097c478bd9Sstevel@tonic-gate 	    (size_t)MIN(hp->hs_dirent.ext_size,
3107c478bd9Sstevel@tonic-gate 	    uiop->uio_resid), UIO_READ, uiop));
3117c478bd9Sstevel@tonic-gate }
3137c478bd9Sstevel@tonic-gate /*ARGSUSED*/
3147c478bd9Sstevel@tonic-gate static void
hsfs_inactive(struct vnode * vp,struct cred * cred,caller_context_t * ct)315*ade42b55SSebastien Roy hsfs_inactive(struct vnode *vp, struct cred *cred, caller_context_t *ct)
3167c478bd9Sstevel@tonic-gate {
3177c478bd9Sstevel@tonic-gate 	struct hsnode *hp;
3187c478bd9Sstevel@tonic-gate 	struct hsfs *fsp;
3207c478bd9Sstevel@tonic-gate 	int nopage;
3227c478bd9Sstevel@tonic-gate 	hp = VTOH(vp);
3237c478bd9Sstevel@tonic-gate 	fsp = VFS_TO_HSFS(vp->v_vfsp);
3247c478bd9Sstevel@tonic-gate 	/*
3257c478bd9Sstevel@tonic-gate 	 * Note: acquiring and holding v_lock for quite a while
3267c478bd9Sstevel@tonic-gate 	 * here serializes on the vnode; this is unfortunate, but
3277c478bd9Sstevel@tonic-gate 	 * likely not to overly impact performance, as the underlying
3287c478bd9Sstevel@tonic-gate 	 * device (CDROM drive) is quite slow.
3297c478bd9Sstevel@tonic-gate 	 */
3307c478bd9Sstevel@tonic-gate 	rw_enter(&fsp->hsfs_hash_lock, RW_WRITER);
3317c478bd9Sstevel@tonic-gate 	mutex_enter(&hp->hs_contents_lock);
3327c478bd9Sstevel@tonic-gate 	mutex_enter(&vp->v_lock);
3347c478bd9Sstevel@tonic-gate 	if (vp->v_count < 1) {
3357c478bd9Sstevel@tonic-gate 		panic("hsfs_inactive: v_count < 1");
3367c478bd9Sstevel@tonic-gate 		/*NOTREACHED*/
3377c478bd9Sstevel@tonic-gate 	}
339*ade42b55SSebastien Roy 	VN_RELE_LOCKED(vp);
340*ade42b55SSebastien Roy 	if (vp->v_count > 0 || (hp->hs_flags & HREF) == 0) {
3417c478bd9Sstevel@tonic-gate 		mutex_exit(&vp->v_lock);
3427c478bd9Sstevel@tonic-gate 		mutex_exit(&hp->hs_contents_lock);
3437c478bd9Sstevel@tonic-gate 		rw_exit(&fsp->hsfs_hash_lock);
3447c478bd9Sstevel@tonic-gate 		return;
3457c478bd9Sstevel@tonic-gate 	}
3467c478bd9Sstevel@tonic-gate 	if (vp->v_count == 0) {
3477c478bd9Sstevel@tonic-gate 		/*
3487c478bd9Sstevel@tonic-gate 		 * Free the hsnode.
3497c478bd9Sstevel@tonic-gate 		 * If there are no pages associated with the
3507c478bd9Sstevel@tonic-gate 		 * hsnode, give it back to the kmem_cache,
3517c478bd9Sstevel@tonic-gate 		 * else put at the end of this file system's
3527c478bd9Sstevel@tonic-gate 		 * internal free list.
3537c478bd9Sstevel@tonic-gate 		 */
3547c478bd9Sstevel@tonic-gate 		nopage = !vn_has_cached_data(vp);
3557c478bd9Sstevel@tonic-gate 		hp->hs_flags = 0;
3567c478bd9Sstevel@tonic-gate 		/*
3577c478bd9Sstevel@tonic-gate 		 * exit these locks now, since hs_freenode may
3587c478bd9Sstevel@tonic-gate 		 * kmem_free the hsnode and embedded vnode
3597c478bd9Sstevel@tonic-gate 		 */
3607c478bd9Sstevel@tonic-gate 		mutex_exit(&vp->v_lock);
3617c478bd9Sstevel@tonic-gate 		mutex_exit(&hp->hs_contents_lock);
3627c478bd9Sstevel@tonic-gate 		hs_freenode(vp, fsp, nopage);
3637c478bd9Sstevel@tonic-gate 	} else {
3647c478bd9Sstevel@tonic-gate 		mutex_exit(&vp->v_lock);
3657c478bd9Sstevel@tonic-gate 		mutex_exit(&hp->hs_contents_lock);
3667c478bd9Sstevel@tonic-gate 	}
3677c478bd9Sstevel@tonic-gate 	rw_exit(&fsp->hsfs_hash_lock);
3687c478bd9Sstevel@tonic-gate }
3717c478bd9Sstevel@tonic-gate /*ARGSUSED*/
3727c478bd9Sstevel@tonic-gate 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)373*ade42b55SSebastien Roy hsfs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
374*ade42b55SSebastien Roy     struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cred,
375*ade42b55SSebastien Roy     caller_context_t *ct, int *direntflags, pathname_t *realpnp)
3767c478bd9Sstevel@tonic-gate {
3777c478bd9Sstevel@tonic-gate 	int error;
3787c478bd9Sstevel@tonic-gate 	int namelen = (int)strlen(nm);
3807c478bd9Sstevel@tonic-gate 	if (*nm == '\0') {
3817c478bd9Sstevel@tonic-gate 		VN_HOLD(dvp);
3827c478bd9Sstevel@tonic-gate 		*vpp = dvp;
3837c478bd9Sstevel@tonic-gate 		return (0);
3847c478bd9Sstevel@tonic-gate 	}
3867c478bd9Sstevel@tonic-gate 	/*
3877c478bd9Sstevel@tonic-gate 	 * If we're looking for ourself, life is simple.
3887c478bd9Sstevel@tonic-gate 	 */
3897c478bd9Sstevel@tonic-gate 	if (namelen == 1 && *nm == '.') {
3907c478bd9Sstevel@tonic-gate 		if (error = hs_access(dvp, (mode_t)VEXEC, cred))
3917c478bd9Sstevel@tonic-gate 			return (error);
3927c478bd9Sstevel@tonic-gate 		VN_HOLD(dvp);
3937c478bd9Sstevel@tonic-gate 		*vpp = dvp;
3947c478bd9Sstevel@tonic-gate 		return (0);
3957c478bd9Sstevel@tonic-gate 	}
3977c478bd9Sstevel@tonic-gate 	return (hs_dirlook(dvp, nm, namelen, vpp, cred));
3987c478bd9Sstevel@tonic-gate }
4017c478bd9Sstevel@tonic-gate /*ARGSUSED*/
4027c478bd9Sstevel@tonic-gate static int
hsfs_readdir(struct vnode * vp,struct uio * uiop,struct cred * cred,int * eofp,caller_context_t * ct,int flags)403*ade42b55SSebastien Roy hsfs_readdir(struct vnode *vp, struct uio *uiop, struct cred *cred, int *eofp,
404*ade42b55SSebastien Roy     caller_context_t *ct, int flags)
4057c478bd9Sstevel@tonic-gate {
4067c478bd9Sstevel@tonic-gate 	struct hsnode	*dhp;
4077c478bd9Sstevel@tonic-gate 	struct hsfs	*fsp;
4087c478bd9Sstevel@tonic-gate 	struct hs_direntry hd;
4097c478bd9Sstevel@tonic-gate 	struct dirent64	*nd;
4107c478bd9Sstevel@tonic-gate 	int		error;
4117c478bd9Sstevel@tonic-gate 	uint_t		offset;		/* real offset in directory */
4127c478bd9Sstevel@tonic-gate 	uint_t		dirsiz;		/* real size of directory */
4137c478bd9Sstevel@tonic-gate 	uchar_t		*blkp;
4147c478bd9Sstevel@tonic-gate 	int		hdlen;		/* length of hs directory entry */
4157c478bd9Sstevel@tonic-gate 	long		ndlen;		/* length of dirent entry */
4167c478bd9Sstevel@tonic-gate 	int		bytes_wanted;
4177c478bd9Sstevel@tonic-gate 	size_t		bufsize;	/* size of dirent buffer */
4187c478bd9Sstevel@tonic-gate 	char		*outbuf;	/* ptr to dirent buffer */
4197c478bd9Sstevel@tonic-gate 	char		*dname;
4207c478bd9Sstevel@tonic-gate 	int		dnamelen;
4217c478bd9Sstevel@tonic-gate 	size_t		dname_size;
4227c478bd9Sstevel@tonic-gate 	struct fbuf	*fbp;
4237c478bd9Sstevel@tonic-gate 	uint_t		last_offset;	/* last index into current dir block */
4247c478bd9Sstevel@tonic-gate 	ino64_t		dirino;	/* temporary storage before storing in dirent */
4257c478bd9Sstevel@tonic-gate 	off_t		diroff;
4277c478bd9Sstevel@tonic-gate 	dhp = VTOH(vp);
4287c478bd9Sstevel@tonic-gate 	fsp = VFS_TO_HSFS(vp->v_vfsp);
4297c478bd9Sstevel@tonic-gate 	if (dhp->hs_dirent.ext_size == 0)
4307c478bd9Sstevel@tonic-gate 		hs_filldirent(vp, &dhp->hs_dirent);
4317c478bd9Sstevel@tonic-gate 	dirsiz = dhp->hs_dirent.ext_size;
4327c478bd9Sstevel@tonic-gate 	if (uiop->uio_loffset >= dirsiz) {	/* at or beyond EOF */
4337c478bd9Sstevel@tonic-gate 		if (eofp)
4347c478bd9Sstevel@tonic-gate 			*eofp = 1;
4357c478bd9Sstevel@tonic-gate 		return (0);
4367c478bd9Sstevel@tonic-gate 	}
4379cbc422eSpeterte 	ASSERT(uiop->uio_loffset <= HS_MAXFILEOFF);
4389cbc422eSpeterte 	offset = uiop->uio_loffset;
4407c478bd9Sstevel@tonic-gate 	dname_size = fsp->hsfs_namemax + 1;	/* 1 for the ending NUL */
4417c478bd9Sstevel@tonic-gate 	dname = kmem_alloc(dname_size, KM_SLEEP);
4427c478bd9Sstevel@tonic-gate 	bufsize = uiop->uio_resid + sizeof (struct dirent64);
4447c478bd9Sstevel@tonic-gate 	outbuf = kmem_alloc(bufsize, KM_SLEEP);
4457c478bd9Sstevel@tonic-gate 	nd = (struct dirent64 *)outbuf;
4477c478bd9Sstevel@tonic-gate 	while (offset < dirsiz) {
448cf83459aSfrankho 		bytes_wanted = MIN(MAXBSIZE, dirsiz - (offset & MAXBMASK));
4507c478bd9Sstevel@tonic-gate 		error = fbread(vp, (offset_t)(offset & MAXBMASK),
451d10b6702Sfrankho 		    (unsigned int)bytes_wanted, S_READ, &fbp);
4527c478bd9Sstevel@tonic-gate 		if (error)
4537c478bd9Sstevel@tonic-gate 			goto done;
4557c478bd9Sstevel@tonic-gate 		blkp = (uchar_t *)fbp->fb_addr;
456cf83459aSfrankho 		last_offset = (offset & MAXBMASK) + fbp->fb_count;
4587c478bd9Sstevel@tonic-gate #define	rel_offset(offset) ((offset) & MAXBOFFSET)	/* index into blkp */
4607c478bd9Sstevel@tonic-gate 		while (offset < last_offset) {
4617c478bd9Sstevel@tonic-gate 			/*
462cf83459aSfrankho 			 * Very similar validation code is found in
463cf83459aSfrankho 			 * process_dirblock(), hsfs_node.c.
464cf83459aSfrankho 			 * For an explanation, see there.
465cf83459aSfrankho 			 * It may make sense for the future to
466cf83459aSfrankho 			 * "consolidate" the code in hs_parsedir(),
467cf83459aSfrankho 			 * process_dirblock() and hsfs_readdir() into
468cf83459aSfrankho 			 * a single utility function.
4697c478bd9Sstevel@tonic-gate 			 */
4707c478bd9Sstevel@tonic-gate 			hdlen = (int)((uchar_t)
471d10b6702Sfrankho 			    HDE_DIR_LEN(&blkp[rel_offset(offset)]));
472cf83459aSfrankho 			if (hdlen < HDE_ROOT_DIR_REC_SIZE ||
473cf83459aSfrankho 			    offset + hdlen > last_offset) {
4747c478bd9Sstevel@tonic-gate 				/*
475cf83459aSfrankho 				 * advance to next sector boundary
4767c478bd9Sstevel@tonic-gate 				 */
477cf83459aSfrankho 				offset = roundup(offset + 1, HS_SECTOR_SIZE);
478cf83459aSfrankho 				if (hdlen)
479cf83459aSfrankho 					hs_log_bogus_disk_warning(fsp,
480cf83459aSfrankho 					    HSFS_ERR_TRAILING_JUNK, 0);
482cf83459aSfrankho 				continue;
4837c478bd9Sstevel@tonic-gate 			}
4857c478bd9Sstevel@tonic-gate 			bzero(&hd, sizeof (hd));
4877c478bd9Sstevel@tonic-gate 			/*
4887c478bd9Sstevel@tonic-gate 			 * Just ignore invalid directory entries.
4897c478bd9Sstevel@tonic-gate 			 * XXX - maybe hs_parsedir() will detect EXISTENCE bit
4907c478bd9Sstevel@tonic-gate 			 */
4917c478bd9Sstevel@tonic-gate 			if (!hs_parsedir(fsp, &blkp[rel_offset(offset)],
492d10b6702Sfrankho 			    &hd, dname, &dnamelen, last_offset - offset)) {
4937c478bd9Sstevel@tonic-gate 				/*
4947c478bd9Sstevel@tonic-gate 				 * Determine if there is enough room
4957c478bd9Sstevel@tonic-gate 				 */
4967c478bd9Sstevel@tonic-gate 				ndlen = (long)DIRENT64_RECLEN((dnamelen));
4987c478bd9Sstevel@tonic-gate 				if ((ndlen + ((char *)nd - outbuf)) >
4997c478bd9Sstevel@tonic-gate 				    uiop->uio_resid) {
5007c478bd9Sstevel@tonic-gate 					fbrelse(fbp, S_READ);
5017c478bd9Sstevel@tonic-gate 					goto done; /* output buffer full */
5027c478bd9Sstevel@tonic-gate 				}
5047c478bd9Sstevel@tonic-gate 				diroff = offset + hdlen;
5057c478bd9Sstevel@tonic-gate 				/*
506d10b6702Sfrankho 				 * If the media carries rrip-v1.12 or newer,
507d10b6702Sfrankho 				 * and we trust the inodes from the rrip data
508d10b6702Sfrankho 				 * (use_rrip_inodes != 0), use that data. If the
509d10b6702Sfrankho 				 * media has been created by a recent mkisofs
510d10b6702Sfrankho 				 * version, we may trust all numbers in the
511d10b6702Sfrankho 				 * starting extent number; otherwise, we cannot
512d10b6702Sfrankho 				 * do this for zero sized files and symlinks,
513d10b6702Sfrankho 				 * because if we did we'd end up mapping all of
514d10b6702Sfrankho 				 * them to the same node. We use HS_DUMMY_INO
515d10b6702Sfrankho 				 * in this case and make sure that we will not
516d10b6702Sfrankho 				 * map all files to the same meta data.
5177c478bd9Sstevel@tonic-gate 				 */
518d10b6702Sfrankho 				if (hd.inode != 0 && use_rrip_inodes) {
519d10b6702Sfrankho 					dirino = hd.inode;
520d10b6702Sfrankho 				} else if ((hd.ext_size == 0 ||
521d10b6702Sfrankho 				    hd.sym_link != (char *)NULL) &&
522d10b6702Sfrankho 				    (fsp->hsfs_flags & HSFSMNT_INODE) == 0) {
523d10b6702Sfrankho 					dirino = HS_DUMMY_INO;
5247c478bd9Sstevel@tonic-gate 				} else {
525d10b6702Sfrankho 					dirino = hd.ext_lbn;
5267c478bd9Sstevel@tonic-gate 				}
5287c478bd9Sstevel@tonic-gate 				/* strncpy(9f) will zero uninitialized bytes */
5307c478bd9Sstevel@tonic-gate 				ASSERT(strlen(dname) + 1 <=
5317c478bd9Sstevel@tonic-gate 				    DIRENT64_NAMELEN(ndlen));
5327c478bd9Sstevel@tonic-gate 				(void) strncpy(nd->d_name, dname,
5337c478bd9Sstevel@tonic-gate 				    DIRENT64_NAMELEN(ndlen));
5347c478bd9Sstevel@tonic-gate 				nd->d_reclen = (ushort_t)ndlen;
5357c478bd9Sstevel@tonic-gate 				nd->d_off = (offset_t)diroff;
5367c478bd9Sstevel@tonic-gate 				nd->d_ino = dirino;
5377c478bd9Sstevel@tonic-gate 				nd = (struct dirent64 *)((char *)nd + ndlen);
5397c478bd9Sstevel@tonic-gate 				/*
5407c478bd9Sstevel@tonic-gate 				 * free up space allocated for symlink
5417c478bd9Sstevel@tonic-gate 				 */
5427c478bd9Sstevel@tonic-gate 				if (hd.sym_link != (char *)NULL) {
5437c478bd9Sstevel@tonic-gate 					kmem_free(hd.sym_link,
5447c478bd9Sstevel@tonic-gate 					    (size_t)(hd.ext_size+1));
5457c478bd9Sstevel@tonic-gate 					hd.sym_link = (char *)NULL;
5467c478bd9Sstevel@tonic-gate 				}
5477c478bd9Sstevel@tonic-gate 			}
5487c478bd9Sstevel@tonic-gate 			offset += hdlen;
5497c478bd9Sstevel@tonic-gate 		}
5507c478bd9Sstevel@tonic-gate 		fbrelse(fbp, S_READ);
5517c478bd9Sstevel@tonic-gate 	}
5537c478bd9Sstevel@tonic-gate 	/*
5547c478bd9Sstevel@tonic-gate 	 * Got here for one of the following reasons:
5557c478bd9Sstevel@tonic-gate 	 *	1) outbuf is full (error == 0)
5567c478bd9Sstevel@tonic-gate 	 *	2) end of directory reached (error == 0)
5577c478bd9Sstevel@tonic-gate 	 *	3) error reading directory sector (error != 0)
5587c478bd9Sstevel@tonic-gate 	 *	4) directory entry crosses sector boundary (error == 0)
5597c478bd9Sstevel@tonic-gate 	 *
5607c478bd9Sstevel@tonic-gate 	 * If any directory entries have been copied, don't report
5617c478bd9Sstevel@tonic-gate 	 * case 4.  Instead, return the valid directory entries.
5627c478bd9Sstevel@tonic-gate 	 *
5637c478bd9Sstevel@tonic-gate 	 * If no entries have been copied, report the error.
5647c478bd9Sstevel@tonic-gate 	 * If case 4, this will be indistiguishable from EOF.
5657c478bd9Sstevel@tonic-gate 	 */
5667c478bd9Sstevel@tonic-gate done:
5677c478bd9Sstevel@tonic-gate 	ndlen = ((char *)nd - outbuf);
5687c478bd9Sstevel@tonic-gate 	if (ndlen != 0) {
5697c478bd9Sstevel@tonic-gate 		error = uiomove(outbuf, (size_t)ndlen, UIO_READ, uiop);
5709cbc422eSpeterte 		uiop->uio_loffset = offset;
5717c478bd9Sstevel@tonic-gate 	}
5727c478bd9Sstevel@tonic-gate 	kmem_free(dname, dname_size);
5737c478bd9Sstevel@tonic-gate 	kmem_free(outbuf, bufsize);
5747c478bd9Sstevel@tonic-gate 	if (eofp && error == 0)
5759cbc422eSpeterte 		*eofp = (uiop->uio_loffset >= dirsiz);
5767c478bd9Sstevel@tonic-gate 	return (error);
5777c478bd9Sstevel@tonic-gate }
579da6c28aaSamw /*ARGSUSED2*/
5807c478bd9Sstevel@tonic-gate static int
hsfs_fid(struct vnode * vp,struct fid * fidp,caller_context_t * ct)581da6c28aaSamw hsfs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
5827c478bd9Sstevel@tonic-gate {
5837c478bd9Sstevel@tonic-gate 	struct hsnode *hp;
5847c478bd9Sstevel@tonic-gate 	struct hsfid *fid;
5867c478bd9Sstevel@tonic-gate 	if (fidp->fid_len < (sizeof (*fid) - sizeof (fid->hf_len))) {
5877c478bd9Sstevel@tonic-gate 		fidp->fid_len = sizeof (*fid) - sizeof (fid->hf_len);
5887c478bd9Sstevel@tonic-gate 		return (ENOSPC);
5897c478bd9Sstevel@tonic-gate 	}
5917c478bd9Sstevel@tonic-gate 	fid = (struct hsfid *)fidp;
5927c478bd9Sstevel@tonic-gate 	fid->hf_len = sizeof (*fid) - sizeof (fid->hf_len);
5937c478bd9Sstevel@tonic-gate 	hp = VTOH(vp);
5947c478bd9Sstevel@tonic-gate 	mutex_enter(&hp->hs_contents_lock);
5957c478bd9Sstevel@tonic-gate 	fid->hf_dir_lbn = hp->hs_dir_lbn;
5967c478bd9Sstevel@tonic-gate 	fid->hf_dir_off = (ushort_t)hp->hs_dir_off;
597d10b6702Sfrankho 	fid->hf_ino = hp->hs_nodeid;
5987c478bd9Sstevel@tonic-gate 	mutex_exit(&hp->hs_contents_lock);
5997c478bd9Sstevel@tonic-gate 	return (0);
6007c478bd9Sstevel@tonic-gate }
6027c478bd9Sstevel@tonic-gate /*ARGSUSED*/
6037c478bd9Sstevel@tonic-gate static int
hsfs_open(struct vnode ** vpp,int flag,struct cred * cred,caller_context_t * ct)604*ade42b55SSebastien Roy hsfs_open(struct vnode **vpp, int flag, struct cred *cred, caller_context_t *ct)
6057c478bd9Sstevel@tonic-gate {
6067c478bd9Sstevel@tonic-gate 	return (0);
6077c478bd9Sstevel@tonic-gate }
6097c478bd9Sstevel@tonic-gate /*ARGSUSED*/
6107c478bd9Sstevel@tonic-gate static int
hsfs_close(struct vnode * vp,int flag,int count,offset_t offset,struct cred * cred,caller_context_t * ct)611*ade42b55SSebastien Roy hsfs_close(struct vnode *vp, int flag, int count, offset_t offset,
612*ade42b55SSebastien Roy     struct cred *cred, caller_context_t *ct)
6137c478bd9Sstevel@tonic-gate {
6147c478bd9Sstevel@tonic-gate 	(void) cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
6157c478bd9Sstevel@tonic-gate 	cleanshares(vp, ttoproc(curthread)->p_pid);
6167c478bd9Sstevel@tonic-gate 	return (0);
6177c478bd9Sstevel@tonic-gate }
6197c478bd9Sstevel@tonic-gate /*ARGSUSED2*/
6207c478bd9Sstevel@tonic-gate static int
hsfs_access(struct vnode * vp,int mode,int flags,cred_t * cred,caller_context_t * ct)621*ade42b55SSebastien Roy hsfs_access(struct vnode *vp, int mode, int flags, cred_t *cred,
622*ade42b55SSebastien Roy     caller_context_t *ct)
6237c478bd9Sstevel@tonic-gate {
6247c478bd9Sstevel@tonic-gate 	return (hs_access(vp, (mode_t)mode, cred));
6257c478bd9Sstevel@tonic-gate }
6277c478bd9Sstevel@tonic-gate /*
6287c478bd9Sstevel@tonic-gate  * the seek time of a CD-ROM is very slow, and data transfer
6297c478bd9Sstevel@tonic-gate  * rate is even worse (max. 150K per sec).  The design
6307c478bd9Sstevel@tonic-gate  * decision is to reduce access to cd-rom as much as possible,
6317c478bd9Sstevel@tonic-gate  * and to transfer a sizable block (read-ahead) of data at a time.
6327c478bd9Sstevel@tonic-gate  * UFS style of read ahead one block at a time is not appropriate,
6337c478bd9Sstevel@tonic-gate  * and is not supported
6347c478bd9Sstevel@tonic-gate  */
6367c478bd9Sstevel@tonic-gate /*
6377c478bd9Sstevel@tonic-gate  * KLUSTSIZE should be a multiple of PAGESIZE and <= MAXPHYS.
6387c478bd9Sstevel@tonic-gate  */
6397c478bd9Sstevel@tonic-gate #define	KLUSTSIZE	(56 * 1024)
6407c478bd9Sstevel@tonic-gate /* we don't support read ahead */
6417c478bd9Sstevel@tonic-gate int hsfs_lostpage;	/* no. of times we lost original page */
6437c478bd9Sstevel@tonic-gate /*
6447c478bd9Sstevel@tonic-gate  * Used to prevent biodone() from releasing buf resources that
6457c478bd9Sstevel@tonic-gate  * we didn't allocate in quite the usual way.
6467c478bd9Sstevel@tonic-gate  */
6477c478bd9Sstevel@tonic-gate /*ARGSUSED*/
6487c478bd9Sstevel@tonic-gate int
hsfs_iodone(struct buf * bp)6497c478bd9Sstevel@tonic-gate hsfs_iodone(struct buf *bp)
6507c478bd9Sstevel@tonic-gate {
6517c478bd9Sstevel@tonic-gate 	sema_v(&bp->b_io);
6527c478bd9Sstevel@tonic-gate 	return (0);
6537c478bd9Sstevel@tonic-gate }
65584b82766Smg /*
65684b82766Smg  * The taskq thread that invokes the scheduling function to ensure
65784b82766Smg  * that all readaheads are complete and cleans up the associated
65884b82766Smg  * memory and releases the page lock.
65984b82766Smg  */
66084b82766Smg void
hsfs_ra_task(void * arg)66184b82766Smg hsfs_ra_task(void *arg)
66284b82766Smg {
66384b82766Smg 	struct hio_info *info = arg;
66484b82766Smg 	uint_t count;
66584b82766Smg 	struct buf *wbuf;
66784b82766Smg 	ASSERT(info->pp != NULL);
66984b82766Smg 	for (count = 0; count < info->bufsused; count++) {
67084b82766Smg 		wbuf = &(info->bufs[count]);
67284b82766Smg 		DTRACE_PROBE1(hsfs_io_wait_ra, struct buf *, wbuf);
67384b82766Smg 		while (sema_tryp(&(info->sema[count])) == 0) {
67484b82766Smg 			if (hsched_invoke_strategy(info->fsp)) {
67584b82766Smg 				sema_p(&(info->sema[count]));
67684b82766Smg 				break;
67784b82766Smg 			}
67884b82766Smg 		}
67984b82766Smg 		sema_destroy(&(info->sema[count]));
68084b82766Smg 		DTRACE_PROBE1(hsfs_io_done_ra, struct buf *, wbuf);
68184b82766Smg 		biofini(&(info->bufs[count]));
68284b82766Smg 	}
68384b82766Smg 	for (count = 0; count < info->bufsused; count++) {
68484b82766Smg 		if (info->vas[count] != NULL) {
68584b82766Smg 			ppmapout(info->vas[count]);
68684b82766Smg 		}
68784b82766Smg 	}
68884b82766Smg 	kmem_free(info->vas, info->bufcnt * sizeof (caddr_t));
68984b82766Smg 	kmem_free(info->bufs, info->bufcnt * sizeof (struct buf));
69084b82766Smg 	kmem_free(info->sema, info->bufcnt * sizeof (ksema_t));
69284b82766Smg 	pvn_read_done(info->pp, 0);
69384b82766Smg 	kmem_cache_free(hio_info_cache, info);
69484b82766Smg }
69684b82766Smg /*
69784b82766Smg  * Submit asynchronous readahead requests to the I/O scheduler
69884b82766Smg  * depending on the number of pages to read ahead. These requests
69984b82766Smg  * are asynchronous to the calling thread but I/O requests issued
70084b82766Smg  * subsequently by other threads with higher LBNs must wait for
70184b82766Smg  * these readaheads to complete since we have a single ordered
70284b82766Smg  * I/O pipeline. Thus these readaheads are semi-asynchronous.
70384b82766Smg  * A TaskQ handles waiting for the readaheads to complete.
70484b82766Smg  *
70584b82766Smg  * This function is mostly a copy of hsfs_getapage but somewhat
70684b82766Smg  * simpler. A readahead request is aborted if page allocation
70784b82766Smg  * fails.
70884b82766Smg  */
70984b82766Smg /*ARGSUSED*/
71084b82766Smg 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)711*ade42b55SSebastien Roy hsfs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg,
712*ade42b55SSebastien Roy     caddr_t addr, struct hsnode *hp, struct hsfs *fsp, int xarsiz,
713*ade42b55SSebastien Roy     offset_t bof, int chunk_lbn_count, int chunk_data_bytes)
71484b82766Smg {
71584b82766Smg 	struct buf *bufs;
71684b82766Smg 	caddr_t *vas;
71784b82766Smg 	caddr_t va;
71884b82766Smg 	struct page *pp, *searchp, *lastp;
71984b82766Smg 	struct vnode *devvp;
72084b82766Smg 	ulong_t	byte_offset;
72184b82766Smg 	size_t	io_len_tmp;
72284b82766Smg 	uint_t	io_off, io_len;
72384b82766Smg 	uint_t	xlen;
72484b82766Smg 	uint_t	filsiz;
72584b82766Smg 	uint_t	secsize;
72684b82766Smg 	uint_t	bufcnt;
72784b82766Smg 	uint_t	bufsused;
72884b82766Smg 	uint_t	count;
72984b82766Smg 	uint_t	io_end;
73084b82766Smg 	uint_t	which_chunk_lbn;
73184b82766Smg 	uint_t	offset_lbn;
73284b82766Smg 	uint_t	offset_extra;
73384b82766Smg 	offset_t	offset_bytes;
73484b82766Smg 	uint_t	remaining_bytes;
73584b82766Smg 	uint_t	extension;
73684b82766Smg 	int	remainder;	/* must be signed */
73784b82766Smg 	diskaddr_t driver_block;
73884b82766Smg 	u_offset_t io_off_tmp;
73984b82766Smg 	ksema_t	*fio_done;
74084b82766Smg 	struct hio_info *info;
74184b82766Smg 	size_t len;
74384b82766Smg 	ASSERT(fsp->hqueue != NULL);
74584b82766Smg 	if (addr >= seg->s_base + seg->s_size) {
74684b82766Smg 		return (-1);
74784b82766Smg 	}
74984b82766Smg 	devvp = fsp->hsfs_devvp;
75084b82766Smg 	secsize = fsp->hsfs_vol.lbn_size;  /* bytes per logical block */
75284b82766Smg 	/* file data size */
75384b82766Smg 	filsiz = hp->hs_dirent.ext_size;
75584b82766Smg 	if (off >= filsiz)
75684b82766Smg 		return (0);
75884b82766Smg 	extension = 0;
75984b82766Smg 	pp = NULL;
76184b82766Smg 	extension += hp->hs_ra_bytes;
76384b82766Smg 	/*
764f9ec9c5aSmg 	 * Some CD writers (e.g. Kodak Photo CD writers)
765f9ec9c5aSmg 	 * create CDs in TAO mode and reserve tracks that
766f9ec9c5aSmg 	 * are not completely written. Some sectors remain
767f9ec9c5aSmg 	 * unreadable for this reason and give I/O errors.
768f9ec9c5aSmg 	 * Also, there's no point in reading sectors
769f9ec9c5aSmg 	 * we'll never look at.  So, if we're asked to go
770f9ec9c5aSmg 	 * beyond the end of a file, truncate to the length
771f9ec9c5aSmg 	 * of that file.
77284b82766Smg 	 *
773f9ec9c5aSmg 	 * Additionally, this behaviour is required by section
774f9ec9c5aSmg 	 * 6.4.5 of ISO 9660:1988(E).
77584b82766Smg 	 */
77684b82766Smg 	len = MIN(extension ? extension : PAGESIZE, filsiz - off);
77884b82766Smg 	/* A little paranoia */
77984b82766Smg 	if (len <= 0)
78084b82766Smg 		return (-1);
78284b82766Smg 	/*
78384b82766Smg 	 * After all that, make sure we're asking for things in units
78484b82766Smg 	 * that bdev_strategy() will understand (see bug 4202551).
78584b82766Smg 	 */
78684b82766Smg 	len = roundup(len, DEV_BSIZE);
78884b82766Smg 	pp = pvn_read_kluster(vp, off, seg, addr, &io_off_tmp,
78984b82766Smg 	    &io_len_tmp, off, len, 1);
79184b82766Smg 	if (pp == NULL) {
79284b82766Smg 		hp->hs_num_contig = 0;
79384b82766Smg 		hp->hs_ra_bytes = 0;
79484b82766Smg 		hp->hs_prev_offset = 0;
79584b82766Smg 		return (-1);
79684b82766Smg 	}
79884b82766Smg 	io_off = (uint_t)io_off_tmp;
79984b82766Smg 	io_len = (uint_t)io_len_tmp;
80184b82766Smg 	/* check for truncation */
80284b82766Smg 	/*
80384b82766Smg 	 * xxx Clean up and return EIO instead?
80484b82766Smg 	 * xxx Ought to go to u_offset_t for everything, but we
80584b82766Smg 	 * xxx call lots of things that want uint_t arguments.
80684b82766Smg 	 */
80784b82766Smg 	ASSERT(io_off == io_off_tmp);
80984b82766Smg 	/*
81084b82766Smg 	 * get enough buffers for worst-case scenario
81184b82766Smg 	 * (i.e., no coalescing possible).
81284b82766Smg 	 */
81384b82766Smg 	bufcnt = (len + secsize - 1) / secsize;
81484b82766Smg 	bufs = kmem_alloc(bufcnt * sizeof (struct buf), KM_SLEEP);
81584b82766Smg 	vas = kmem_alloc(bufcnt * sizeof (caddr_t), KM_SLEEP);
81784b82766Smg 	/*
81884b82766Smg 	 * Allocate a array of semaphores since we are doing I/O
81984b82766Smg 	 * scheduling.
82084b82766Smg 	 */
82184b82766Smg 	fio_done = kmem_alloc(bufcnt * sizeof (ksema_t), KM_SLEEP);
82384b82766Smg 	/*
82484b82766Smg 	 * If our filesize is not an integer multiple of PAGESIZE,
82584b82766Smg 	 * we zero that part of the last page that's between EOF and
82684b82766Smg 	 * the PAGESIZE boundary.
82784b82766Smg 	 */
82884b82766Smg 	xlen = io_len & PAGEOFFSET;
82984b82766Smg 	if (xlen != 0)
83084b82766Smg 		pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
83284b82766Smg 	DTRACE_PROBE2(hsfs_readahead, struct vnode *, vp, uint_t, io_len);
83484b82766Smg 	va = NULL;
83584b82766Smg 	lastp = NULL;
83684b82766Smg 	searchp = pp;
83784b82766Smg 	io_end = io_off + io_len;
83884b82766Smg 	for (count = 0, byte_offset = io_off;
83984b82766Smg 	    byte_offset < io_end;
84084b82766Smg 	    count++) {
84184b82766Smg 		ASSERT(count < bufcnt);
84384b82766Smg 		bioinit(&bufs[count]);
84484b82766Smg 		bufs[count].b_edev = devvp->v_rdev;
84584b82766Smg 		bufs[count].b_dev = cmpdev(devvp->v_rdev);
84684b82766Smg 		bufs[count].b_flags = B_NOCACHE|B_BUSY|B_READ;
84784b82766Smg 		bufs[count].b_iodone = hsfs_iodone;
84884b82766Smg 		bufs[count].b_vp = vp;
84984b82766Smg 		bufs[count].b_file = vp;
85184b82766Smg 		/* Compute disk address for interleaving. */
85384b82766Smg 		/* considered without skips */
85484b82766Smg 		which_chunk_lbn = byte_offset / chunk_data_bytes;
85684b82766Smg 		/* factor in skips */
85784b82766Smg 		offset_lbn = which_chunk_lbn * chunk_lbn_count;
85984b82766Smg 		/* convert to physical byte offset for lbn */
86084b82766Smg 		offset_bytes = LBN_TO_BYTE(offset_lbn, vp->v_vfsp);
86284b82766Smg 		/* don't forget offset into lbn */
86384b82766Smg 		offset_extra = byte_offset % chunk_data_bytes;
86584b82766Smg 		/* get virtual block number for driver */
86684b82766Smg 		driver_block = lbtodb(bof + xarsiz
86784b82766Smg 		    + offset_bytes + offset_extra);
86984b82766Smg 		if (lastp != searchp) {
87084b82766Smg 			/* this branch taken first time through loop */
87184b82766Smg 			va = vas[count] = ppmapin(searchp, PROT_WRITE,
87284b82766Smg 			    (caddr_t)-1);
87384b82766Smg 			/* ppmapin() guarantees not to return NULL */
87484b82766Smg 		} else {
87584b82766Smg 			vas[count] = NULL;
87684b82766Smg 		}
87884b82766Smg 		bufs[count].b_un.b_addr = va + byte_offset % PAGESIZE;
87984b82766Smg 		bufs[count].b_offset =
88084b82766Smg 		    (offset_t)(byte_offset - io_off + off);
88284b82766Smg 		/*
88384b82766Smg 		 * We specifically use the b_lblkno member here
88484b82766Smg 		 * as even in the 32 bit world driver_block can
88584b82766Smg 		 * get very large in line with the ISO9660 spec.
88684b82766Smg 		 */
88884b82766Smg 		bufs[count].b_lblkno = driver_block;
89084b82766Smg 		remaining_bytes = ((which_chunk_lbn + 1) * chunk_data_bytes)
89184b82766Smg 		    - byte_offset;
89384b82766Smg 		/*
89484b82766Smg 		 * remaining_bytes can't be zero, as we derived
89584b82766Smg 		 * which_chunk_lbn directly from byte_offset.
89684b82766Smg 		 */
89784b82766Smg 		if ((remaining_bytes + byte_offset) < (off + len)) {
89884b82766Smg 			/* coalesce-read the rest of the chunk */
89984b82766Smg 			bufs[count].b_bcount = remaining_bytes;
90084b82766Smg 		} else {
90184b82766Smg 			/* get the final bits */
90284b82766Smg 			bufs[count].b_bcount = off + len - byte_offset;
90384b82766Smg 		}
90584b82766Smg 		remainder = PAGESIZE - (byte_offset % PAGESIZE);
90684b82766Smg 		if (bufs[count].b_bcount > remainder) {
90784b82766Smg 			bufs[count].b_bcount = remainder;
90884b82766Smg 		}
91084b82766Smg 		bufs[count].b_bufsize = bufs[count].b_bcount;
91184b82766Smg 		if (((offset_t)byte_offset + bufs[count].b_bcount) >
91284b82766Smg 		    HS_MAXFILEOFF) {
91384b82766Smg 			break;
91484b82766Smg 		}
91584b82766Smg 		byte_offset += bufs[count].b_bcount;
91784b82766Smg 		/*
91884b82766Smg 		 * We are scheduling I/O so we need to enqueue
91984b82766Smg 		 * requests rather than calling bdev_strategy
92084b82766Smg 		 * here. A later invocation of the scheduling
92184b82766Smg 		 * function will take care of doing the actual
92284b82766Smg 		 * I/O as it selects requests from the queue as
92384b82766Smg 		 * per the scheduling logic.
92484b82766Smg 		 */
92584b82766Smg 		struct hio *hsio = kmem_cache_alloc(hio_cache,
92684b82766Smg 		    KM_SLEEP);
92884b82766Smg 		sema_init(&fio_done[count], 0, NULL,
92984b82766Smg 		    SEMA_DEFAULT, NULL);
93084b82766Smg 		hsio->bp = &bufs[count];
93184b82766Smg 		hsio->sema = &fio_done[count];
93284b82766Smg 		hsio->io_lblkno = bufs[count].b_lblkno;
93384b82766Smg 		hsio->nblocks = howmany(hsio->bp->b_bcount,
93484b82766Smg 		    DEV_BSIZE);
93684b82766Smg 		/* used for deadline */
93784b82766Smg 		hsio->io_timestamp = drv_hztousec(ddi_get_lbolt());
93984b82766Smg 		/* for I/O coalescing */
94084b82766Smg 		hsio->contig_chain = NULL;
94184b82766Smg 		hsched_enqueue_io(fsp, hsio, 1);
94384b82766Smg 		lwp_stat_update(LWP_STAT_INBLK, 1);
94484b82766Smg 		lastp = searchp;
94584b82766Smg 		if ((remainder - bufs[count].b_bcount) < 1) {
94684b82766Smg 			searchp = searchp->p_next;
94784b82766Smg 		}
94884b82766Smg 	}
95084b82766Smg 	bufsused = count;
95184b82766Smg 	info = kmem_cache_alloc(hio_info_cache, KM_SLEEP);
95284b82766Smg 	info->bufs = bufs;
95384b82766Smg 	info->vas = vas;
95484b82766Smg 	info->sema = fio_done;
95584b82766Smg 	info->bufsused = bufsused;
95684b82766Smg 	info->bufcnt = bufcnt;
95784b82766Smg 	info->fsp = fsp;
95884b82766Smg 	info->pp = pp;
96084b82766Smg 	(void) taskq_dispatch(fsp->hqueue->ra_task,
96184b82766Smg 	    hsfs_ra_task, info, KM_SLEEP);
96284b82766Smg 	/*
96384b82766Smg 	 * The I/O locked pages are unlocked in our taskq thread.
96484b82766Smg 	 */
96584b82766Smg 	return (0);
96684b82766Smg }
9687c478bd9Sstevel@tonic-gate /*
9697c478bd9Sstevel@tonic-gate  * Each file may have a different interleaving on disk.  This makes
9707c478bd9Sstevel@tonic-gate  * things somewhat interesting.  The gist is that there are some
9717c478bd9Sstevel@tonic-gate  * number of contiguous data sectors, followed by some other number
9727c478bd9Sstevel@tonic-gate  * of contiguous skip sectors.  The sum of those two sets of sectors
9737c478bd9Sstevel@tonic-gate  * defines the interleave size.  Unfortunately, it means that we generally
9747c478bd9Sstevel@tonic-gate  * can't simply read N sectors starting at a given offset to satisfy
9757c478bd9Sstevel@tonic-gate  * any given request.
9767c478bd9Sstevel@tonic-gate  *
9777c478bd9Sstevel@tonic-gate  * What we do is get the relevant memory pages via pvn_read_kluster(),
9787c478bd9Sstevel@tonic-gate  * then stride through the interleaves, setting up a buf for each
9797c478bd9Sstevel@tonic-gate  * sector that needs to be brought in.  Instead of kmem_alloc'ing
9807c478bd9Sstevel@tonic-gate  * space for the sectors, though, we just point at the appropriate
9817c478bd9Sstevel@tonic-gate  * spot in the relevant page for each of them.  This saves us a bunch
9827c478bd9Sstevel@tonic-gate  * of copying.
98384b82766Smg  *
98484b82766Smg  * NOTICE: The code below in hsfs_getapage is mostly same as the code
98584b82766Smg  *         in hsfs_getpage_ra above (with some omissions). If you are
98684b82766Smg  *         making any change to this function, please also look at
98784b82766Smg  *         hsfs_getpage_ra.
9887c478bd9Sstevel@tonic-gate  */
9897c478bd9Sstevel@tonic-gate /*ARGSUSED*/
9907c478bd9Sstevel@tonic-gate 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)991*ade42b55SSebastien Roy hsfs_getapage(struct vnode *vp, u_offset_t off, size_t len, uint_t *protp,
992*ade42b55SSebastien Roy     struct page *pl[], size_t plsz, struct seg *seg, caddr_t addr,
993*ade42b55SSebastien Roy     enum seg_rw rw, struct cred *cred)
9947c478bd9Sstevel@tonic-gate {
9957c478bd9Sstevel@tonic-gate 	struct hsnode *hp;
9967c478bd9Sstevel@tonic-gate 	struct hsfs *fsp;
9977c478bd9Sstevel@tonic-gate 	int	err;
9987c478bd9Sstevel@tonic-gate 	struct buf *bufs;
9997c478bd9Sstevel@tonic-gate 	caddr_t *vas;
10007c478bd9Sstevel@tonic-gate 	caddr_t va;
10017c478bd9Sstevel@tonic-gate 	struct page *pp, *searchp, *lastp;
10027c478bd9Sstevel@tonic-gate 	page_t	*pagefound;
10037c478bd9Sstevel@tonic-gate 	offset_t	bof;
10047c478bd9Sstevel@tonic-gate 	struct vnode *devvp;
10057c478bd9Sstevel@tonic-gate 	ulong_t	byte_offset;
10067c478bd9Sstevel@tonic-gate 	size_t	io_len_tmp;
10077c478bd9Sstevel@tonic-gate 	uint_t	io_off, io_len;
10087c478bd9Sstevel@tonic-gate 	uint_t	xlen;
10097c478bd9Sstevel@tonic-gate 	uint_t	filsiz;
10107c478bd9Sstevel@tonic-gate 	uint_t	secsize;
10117c478bd9Sstevel@tonic-gate 	uint_t	bufcnt;
10127c478bd9Sstevel@tonic-gate 	uint_t	bufsused;
10137c478bd9Sstevel@tonic-gate 	uint_t	count;
10147c478bd9Sstevel@tonic-gate 	uint_t	io_end;
10157c478bd9Sstevel@tonic-gate 	uint_t	which_chunk_lbn;
10167c478bd9Sstevel@tonic-gate 	uint_t	offset_lbn;
10177c478bd9Sstevel@tonic-gate 	uint_t	offset_extra;
10187c478bd9Sstevel@tonic-gate 	offset_t	offset_bytes;
10197c478bd9Sstevel@tonic-gate 	uint_t	remaining_bytes;
10207c478bd9Sstevel@tonic-gate 	uint_t	extension;
10217c478bd9Sstevel@tonic-gate 	int	remainder;	/* must be signed */
10227c478bd9Sstevel@tonic-gate 	int	chunk_lbn_count;
10237c478bd9Sstevel@tonic-gate 	int	chunk_data_bytes;
10247c478bd9Sstevel@tonic-gate 	int	xarsiz;
10257c478bd9Sstevel@tonic-gate 	diskaddr_t driver_block;
10267c478bd9Sstevel@tonic-gate 	u_offset_t io_off_tmp;
102784b82766Smg 	ksema_t *fio_done;
102884b82766Smg 	int	calcdone;
10307c478bd9Sstevel@tonic-gate 	/*
10317c478bd9Sstevel@tonic-gate 	 * We don't support asynchronous operation at the moment, so
10327c478bd9Sstevel@tonic-gate 	 * just pretend we did it.  If the pages are ever actually
10337c478bd9Sstevel@tonic-gate 	 * needed, they'll get brought in then.
10347c478bd9Sstevel@tonic-gate 	 */
10357c478bd9Sstevel@tonic-gate 	if (pl == NULL)
10367c478bd9Sstevel@tonic-gate 		return (0);
10387c478bd9Sstevel@tonic-gate 	hp = VTOH(vp);
10397c478bd9Sstevel@tonic-gate 	fsp = VFS_TO_HSFS(vp->v_vfsp);
10407c478bd9Sstevel@tonic-gate 	devvp = fsp->hsfs_devvp;
10417c478bd9Sstevel@tonic-gate 	secsize = fsp->hsfs_vol.lbn_size;  /* bytes per logical block */
10437c478bd9Sstevel@tonic-gate 	/* file data size */
10447c478bd9Sstevel@tonic-gate 	filsiz = hp->hs_dirent.ext_size;
10467c478bd9Sstevel@tonic-gate 	/* disk addr for start of file */
10477c478bd9Sstevel@tonic-gate 	bof = LBN_TO_BYTE((offset_t)hp->hs_dirent.ext_lbn, vp->v_vfsp);
10497c478bd9Sstevel@tonic-gate 	/* xarsiz byte must be skipped for data */
10507c478bd9Sstevel@tonic-gate 	xarsiz = hp->hs_dirent.xar_len << fsp->hsfs_vol.lbn_shift;
10527c478bd9Sstevel@tonic-gate 	/* how many logical blocks in an interleave (data+skip) */
10537c478bd9Sstevel@tonic-gate 	chunk_lbn_count = hp->hs_dirent.intlf_sz + hp->hs_dirent.intlf_sk;
10557c478bd9Sstevel@tonic-gate 	if (chunk_lbn_count == 0) {
10567c478bd9Sstevel@tonic-gate 		chunk_lbn_count = 1;
10577c478bd9Sstevel@tonic-gate 	}
10597c478bd9Sstevel@tonic-gate 	/*
10607c478bd9Sstevel@tonic-gate 	 * Convert interleaving size into bytes.  The zero case
10617c478bd9Sstevel@tonic-gate 	 * (no interleaving) optimization is handled as a side-
10627c478bd9Sstevel@tonic-gate 	 * effect of the read-ahead logic.
10637c478bd9Sstevel@tonic-gate 	 */
10647c478bd9Sstevel@tonic-gate 	if (hp->hs_dirent.intlf_sz == 0) {
10657c478bd9Sstevel@tonic-gate 		chunk_data_bytes = LBN_TO_BYTE(1, vp->v_vfsp);
106684b82766Smg 		/*
106784b82766Smg 		 * Optimization: If our pagesize is a multiple of LBN
106884b82766Smg 		 * bytes, we can avoid breaking up a page into individual
106984b82766Smg 		 * lbn-sized requests.
107084b82766Smg 		 */
107184b82766Smg 		if (PAGESIZE % chunk_data_bytes == 0) {
107284b82766Smg 			chunk_lbn_count = BYTE_TO_LBN(PAGESIZE, vp->v_vfsp);
107384b82766Smg 			chunk_data_bytes = PAGESIZE;
107484b82766Smg 		}
10757c478bd9Sstevel@tonic-gate 	} else {
1076d10b6702Sfrankho 		chunk_data_bytes =
1077d10b6702Sfrankho 		    LBN_TO_BYTE(hp->hs_dirent.intlf_sz, vp->v_vfsp);
10787c478bd9Sstevel@tonic-gate 	}
10807c478bd9Sstevel@tonic-gate reread:
10817c478bd9Sstevel@tonic-gate 	err = 0;
10827c478bd9Sstevel@tonic-gate 	pagefound = 0;
108384b82766Smg 	calcdone = 0;
10857c478bd9Sstevel@tonic-gate 	/*
10867c478bd9Sstevel@tonic-gate 	 * Do some read-ahead.  This mostly saves us a bit of
10877c478bd9Sstevel@tonic-gate 	 * system cpu time more than anything else when doing
10887c478bd9Sstevel@tonic-gate 	 * sequential reads.  At some point, could do the
10897c478bd9Sstevel@tonic-gate 	 * read-ahead asynchronously which might gain us something
10907c478bd9Sstevel@tonic-gate 	 * on wall time, but it seems unlikely....