xref: /titanic_41/usr/src/uts/common/fs/ufs/ufs_vnops.c (revision 96d9f183facd90dbbc2268c9a51689be0b6a0b46)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
28 /*	  All Rights Reserved  	*/
29 
30 /*
31  * Portions of this source code were derived from Berkeley 4.3 BSD
32  * under license from the Regents of the University of California.
33  */
34 
35 #include <sys/types.h>
36 #include <sys/t_lock.h>
37 #include <sys/ksynch.h>
38 #include <sys/param.h>
39 #include <sys/time.h>
40 #include <sys/systm.h>
41 #include <sys/sysmacros.h>
42 #include <sys/resource.h>
43 #include <sys/signal.h>
44 #include <sys/cred.h>
45 #include <sys/user.h>
46 #include <sys/buf.h>
47 #include <sys/vfs.h>
48 #include <sys/vfs_opreg.h>
49 #include <sys/vnode.h>
50 #include <sys/proc.h>
51 #include <sys/disp.h>
52 #include <sys/file.h>
53 #include <sys/fcntl.h>
54 #include <sys/flock.h>
55 #include <sys/atomic.h>
56 #include <sys/kmem.h>
57 #include <sys/uio.h>
58 #include <sys/dnlc.h>
59 #include <sys/conf.h>
60 #include <sys/mman.h>
61 #include <sys/pathname.h>
62 #include <sys/debug.h>
63 #include <sys/vmsystm.h>
64 #include <sys/cmn_err.h>
65 #include <sys/filio.h>
66 #include <sys/policy.h>
67 
68 #include <sys/fs/ufs_fs.h>
69 #include <sys/fs/ufs_lockfs.h>
70 #include <sys/fs/ufs_filio.h>
71 #include <sys/fs/ufs_inode.h>
72 #include <sys/fs/ufs_fsdir.h>
73 #include <sys/fs/ufs_quota.h>
74 #include <sys/fs/ufs_log.h>
75 #include <sys/fs/ufs_snap.h>
76 #include <sys/fs/ufs_trans.h>
77 #include <sys/fs/ufs_panic.h>
78 #include <sys/fs/ufs_bio.h>
79 #include <sys/dirent.h>		/* must be AFTER <sys/fs/fsdir.h>! */
80 #include <sys/errno.h>
81 #include <sys/fssnap_if.h>
82 #include <sys/unistd.h>
83 #include <sys/sunddi.h>
84 
85 #include <sys/filio.h>		/* _FIOIO */
86 
87 #include <vm/hat.h>
88 #include <vm/page.h>
89 #include <vm/pvn.h>
90 #include <vm/as.h>
91 #include <vm/seg.h>
92 #include <vm/seg_map.h>
93 #include <vm/seg_vn.h>
94 #include <vm/seg_kmem.h>
95 #include <vm/rm.h>
96 #include <sys/swap.h>
97 
98 #include <fs/fs_subr.h>
99 
100 #include <sys/fs/decomp.h>
101 
102 static struct instats ins;
103 
104 static 	int ufs_getpage_ra(struct vnode *, u_offset_t, struct seg *, caddr_t);
105 static	int ufs_getpage_miss(struct vnode *, u_offset_t, size_t, struct seg *,
106 		caddr_t, struct page **, size_t, enum seg_rw, int);
107 static	int ufs_open(struct vnode **, int, struct cred *, caller_context_t *);
108 static	int ufs_close(struct vnode *, int, int, offset_t, struct cred *,
109 		caller_context_t *);
110 static	int ufs_read(struct vnode *, struct uio *, int, struct cred *,
111 		struct caller_context *);
112 static	int ufs_write(struct vnode *, struct uio *, int, struct cred *,
113 		struct caller_context *);
114 static	int ufs_ioctl(struct vnode *, int, intptr_t, int, struct cred *,
115 		int *, caller_context_t *);
116 static	int ufs_getattr(struct vnode *, struct vattr *, int, struct cred *,
117 		caller_context_t *);
118 static	int ufs_setattr(struct vnode *, struct vattr *, int, struct cred *,
119 		caller_context_t *);
120 static	int ufs_access(struct vnode *, int, int, struct cred *,
121 		caller_context_t *);
122 static	int ufs_lookup(struct vnode *, char *, struct vnode **,
123 		struct pathname *, int, struct vnode *, struct cred *,
124 		caller_context_t *, int *, pathname_t *);
125 static	int ufs_create(struct vnode *, char *, struct vattr *, enum vcexcl,
126 		int, struct vnode **, struct cred *, int,
127 		caller_context_t *, vsecattr_t  *);
128 static	int ufs_remove(struct vnode *, char *, struct cred *,
129 		caller_context_t *, int);
130 static	int ufs_link(struct vnode *, struct vnode *, char *, struct cred *,
131 		caller_context_t *, int);
132 static	int ufs_rename(struct vnode *, char *, struct vnode *, char *,
133 		struct cred *, caller_context_t *, int);
134 static	int ufs_mkdir(struct vnode *, char *, struct vattr *, struct vnode **,
135 		struct cred *, caller_context_t *, int, vsecattr_t *);
136 static	int ufs_rmdir(struct vnode *, char *, struct vnode *, struct cred *,
137 		caller_context_t *, int);
138 static	int ufs_readdir(struct vnode *, struct uio *, struct cred *, int *,
139 		caller_context_t *, int);
140 static	int ufs_symlink(struct vnode *, char *, struct vattr *, char *,
141 		struct cred *, caller_context_t *, int);
142 static	int ufs_readlink(struct vnode *, struct uio *, struct cred *,
143 		caller_context_t *);
144 static	int ufs_fsync(struct vnode *, int, struct cred *, caller_context_t *);
145 static	void ufs_inactive(struct vnode *, struct cred *, caller_context_t *);
146 static	int ufs_fid(struct vnode *, struct fid *, caller_context_t *);
147 static	int ufs_rwlock(struct vnode *, int, caller_context_t *);
148 static	void ufs_rwunlock(struct vnode *, int, caller_context_t *);
149 static	int ufs_seek(struct vnode *, offset_t, offset_t *, caller_context_t *);
150 static	int ufs_frlock(struct vnode *, int, struct flock64 *, int, offset_t,
151 		struct flk_callback *, struct cred *,
152 		caller_context_t *);
153 static  int ufs_space(struct vnode *, int, struct flock64 *, int, offset_t,
154 		cred_t *, caller_context_t *);
155 static	int ufs_getpage(struct vnode *, offset_t, size_t, uint_t *,
156 		struct page **, size_t, struct seg *, caddr_t,
157 		enum seg_rw, struct cred *, caller_context_t *);
158 static	int ufs_putpage(struct vnode *, offset_t, size_t, int, struct cred *,
159 		caller_context_t *);
160 static	int ufs_putpages(struct vnode *, offset_t, size_t, int, struct cred *);
161 static	int ufs_map(struct vnode *, offset_t, struct as *, caddr_t *, size_t,
162 		uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
163 static	int ufs_addmap(struct vnode *, offset_t, struct as *, caddr_t,  size_t,
164 		uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
165 static	int ufs_delmap(struct vnode *, offset_t, struct as *, caddr_t,  size_t,
166 		uint_t, uint_t, uint_t, struct cred *, caller_context_t *);
167 static	int ufs_poll(vnode_t *, short, int, short *, struct pollhead **,
168 		caller_context_t *);
169 static	int ufs_dump(vnode_t *, caddr_t, offset_t, offset_t,
170     caller_context_t *);
171 static	int ufs_l_pathconf(struct vnode *, int, ulong_t *, struct cred *,
172 		caller_context_t *);
173 static	int ufs_pageio(struct vnode *, struct page *, u_offset_t, size_t, int,
174 		struct cred *, caller_context_t *);
175 static	int ufs_dumpctl(vnode_t *, int, offset_t *, caller_context_t *);
176 static	daddr32_t *save_dblks(struct inode *, struct ufsvfs *, daddr32_t *,
177 		daddr32_t *, int, int);
178 static	int ufs_getsecattr(struct vnode *, vsecattr_t *, int, struct cred *,
179 		caller_context_t *);
180 static	int ufs_setsecattr(struct vnode *, vsecattr_t *, int, struct cred *,
181 		caller_context_t *);
182 static	int ufs_priv_access(void *, int, struct cred *);
183 extern int as_map_locked(struct as *, caddr_t, size_t, int ((*)()), void *);
184 
185 /*
186  * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
187  *
188  * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
189  */
190 struct vnodeops *ufs_vnodeops;
191 
192 /* NOTE: "not blkd" below  means that the operation isn't blocked by lockfs */
193 const fs_operation_def_t ufs_vnodeops_template[] = {
194 	VOPNAME_OPEN,		{ .vop_open = ufs_open },	/* not blkd */
195 	VOPNAME_CLOSE,		{ .vop_close = ufs_close },	/* not blkd */
196 	VOPNAME_READ,		{ .vop_read = ufs_read },
197 	VOPNAME_WRITE,		{ .vop_write = ufs_write },
198 	VOPNAME_IOCTL,		{ .vop_ioctl = ufs_ioctl },
199 	VOPNAME_GETATTR,	{ .vop_getattr = ufs_getattr },
200 	VOPNAME_SETATTR,	{ .vop_setattr = ufs_setattr },
201 	VOPNAME_ACCESS,		{ .vop_access = ufs_access },
202 	VOPNAME_LOOKUP,		{ .vop_lookup = ufs_lookup },
203 	VOPNAME_CREATE,		{ .vop_create = ufs_create },
204 	VOPNAME_REMOVE,		{ .vop_remove = ufs_remove },
205 	VOPNAME_LINK,		{ .vop_link = ufs_link },
206 	VOPNAME_RENAME,		{ .vop_rename = ufs_rename },
207 	VOPNAME_MKDIR,		{ .vop_mkdir = ufs_mkdir },
208 	VOPNAME_RMDIR,		{ .vop_rmdir = ufs_rmdir },
209 	VOPNAME_READDIR,	{ .vop_readdir = ufs_readdir },
210 	VOPNAME_SYMLINK,	{ .vop_symlink = ufs_symlink },
211 	VOPNAME_READLINK,	{ .vop_readlink = ufs_readlink },
212 	VOPNAME_FSYNC,		{ .vop_fsync = ufs_fsync },
213 	VOPNAME_INACTIVE,	{ .vop_inactive = ufs_inactive }, /* not blkd */
214 	VOPNAME_FID,		{ .vop_fid = ufs_fid },
215 	VOPNAME_RWLOCK,		{ .vop_rwlock = ufs_rwlock },	/* not blkd */
216 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = ufs_rwunlock }, /* not blkd */
217 	VOPNAME_SEEK,		{ .vop_seek = ufs_seek },
218 	VOPNAME_FRLOCK,		{ .vop_frlock = ufs_frlock },
219 	VOPNAME_SPACE,		{ .vop_space = ufs_space },
220 	VOPNAME_GETPAGE,	{ .vop_getpage = ufs_getpage },
221 	VOPNAME_PUTPAGE,	{ .vop_putpage = ufs_putpage },
222 	VOPNAME_MAP,		{ .vop_map = ufs_map },
223 	VOPNAME_ADDMAP,		{ .vop_addmap = ufs_addmap },	/* not blkd */
224 	VOPNAME_DELMAP,		{ .vop_delmap = ufs_delmap },	/* not blkd */
225 	VOPNAME_POLL,		{ .vop_poll = ufs_poll },	/* not blkd */
226 	VOPNAME_DUMP,		{ .vop_dump = ufs_dump },
227 	VOPNAME_PATHCONF,	{ .vop_pathconf = ufs_l_pathconf },
228 	VOPNAME_PAGEIO,		{ .vop_pageio = ufs_pageio },
229 	VOPNAME_DUMPCTL,	{ .vop_dumpctl = ufs_dumpctl },
230 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = ufs_getsecattr },
231 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = ufs_setsecattr },
232 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
233 	NULL,			NULL
234 };
235 
236 #define	MAX_BACKFILE_COUNT	9999
237 
238 /*
239  * Created by ufs_dumpctl() to store a file's disk block info into memory.
240  * Used by ufs_dump() to dump data to disk directly.
241  */
242 struct dump {
243 	struct inode	*ip;		/* the file we contain */
244 	daddr_t		fsbs;		/* number of blocks stored */
245 	struct timeval32 time;		/* time stamp for the struct */
246 	daddr32_t 	dblk[1];	/* place holder for block info */
247 };
248 
249 static struct dump *dump_info = NULL;
250 
251 /*
252  * Previously there was no special action required for ordinary files.
253  * (Devices are handled through the device file system.)
254  * Now we support Large Files and Large File API requires open to
255  * fail if file is large.
256  * We could take care to prevent data corruption
257  * by doing an atomic check of size and truncate if file is opened with
258  * FTRUNC flag set but traditionally this is being done by the vfs/vnode
259  * layers. So taking care of truncation here is a change in the existing
260  * semantics of VOP_OPEN and therefore we chose not to implement any thing
261  * here. The check for the size of the file > 2GB is being done at the
262  * vfs layer in routine vn_open().
263  */
264 
265 /* ARGSUSED */
266 static int
267 ufs_open(struct vnode **vpp, int flag, struct cred *cr, caller_context_t *ct)
268 {
269 	return (0);
270 }
271 
272 /*ARGSUSED*/
273 static int
274 ufs_close(struct vnode *vp, int flag, int count, offset_t offset,
275 	struct cred *cr, caller_context_t *ct)
276 {
277 	cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
278 	cleanshares(vp, ttoproc(curthread)->p_pid);
279 
280 	/*
281 	 * Push partially filled cluster at last close.
282 	 * ``last close'' is approximated because the dnlc
283 	 * may have a hold on the vnode.
284 	 * Checking for VBAD here will also act as a forced umount check.
285 	 */
286 	if (vp->v_count <= 2 && vp->v_type != VBAD) {
287 		struct inode *ip = VTOI(vp);
288 		if (ip->i_delaylen) {
289 			ins.in_poc.value.ul++;
290 			(void) ufs_putpages(vp, ip->i_delayoff, ip->i_delaylen,
291 			    B_ASYNC | B_FREE, cr);
292 			ip->i_delaylen = 0;
293 		}
294 	}
295 
296 	return (0);
297 }
298 
299 /*ARGSUSED*/
300 static int
301 ufs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cr,
302 	struct caller_context *ct)
303 {
304 	struct inode *ip = VTOI(vp);
305 	struct ufsvfs *ufsvfsp;
306 	struct ulockfs *ulp = NULL;
307 	int error = 0;
308 	int intrans = 0;
309 
310 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
311 
312 	/*
313 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
314 	 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
315 	 */
316 	if (MANDLOCK(vp, ip->i_mode)) {
317 		/*
318 		 * ufs_getattr ends up being called by chklock
319 		 */
320 		error = chklock(vp, FREAD, uiop->uio_loffset,
321 		    uiop->uio_resid, uiop->uio_fmode, ct);
322 		if (error)
323 			goto out;
324 	}
325 
326 	ufsvfsp = ip->i_ufsvfs;
327 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READ_MASK);
328 	if (error)
329 		goto out;
330 
331 	/*
332 	 * In the case that a directory is opened for reading as a file
333 	 * (eg "cat .") with the  O_RSYNC, O_SYNC and O_DSYNC flags set.
334 	 * The locking order had to be changed to avoid a deadlock with
335 	 * an update taking place on that directory at the same time.
336 	 */
337 	if ((ip->i_mode & IFMT) == IFDIR) {
338 
339 		rw_enter(&ip->i_contents, RW_READER);
340 		error = rdip(ip, uiop, ioflag, cr);
341 		rw_exit(&ip->i_contents);
342 
343 		if (error) {
344 			if (ulp)
345 				ufs_lockfs_end(ulp);
346 			goto out;
347 		}
348 
349 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
350 		    TRANS_ISTRANS(ufsvfsp)) {
351 			rw_exit(&ip->i_rwlock);
352 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
353 			    error);
354 			ASSERT(!error);
355 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
356 			    TOP_READ_SIZE);
357 			rw_enter(&ip->i_rwlock, RW_READER);
358 		}
359 	} else {
360 		/*
361 		 * Only transact reads to files opened for sync-read and
362 		 * sync-write on a file system that is not write locked.
363 		 *
364 		 * The ``not write locked'' check prevents problems with
365 		 * enabling/disabling logging on a busy file system.  E.g.,
366 		 * logging exists at the beginning of the read but does not
367 		 * at the end.
368 		 *
369 		 */
370 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
371 		    TRANS_ISTRANS(ufsvfsp)) {
372 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
373 			    error);
374 			ASSERT(!error);
375 			intrans = 1;
376 		}
377 
378 		rw_enter(&ip->i_contents, RW_READER);
379 		error = rdip(ip, uiop, ioflag, cr);
380 		rw_exit(&ip->i_contents);
381 
382 		if (intrans) {
383 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
384 			    TOP_READ_SIZE);
385 		}
386 	}
387 
388 	if (ulp) {
389 		ufs_lockfs_end(ulp);
390 	}
391 out:
392 
393 	return (error);
394 }
395 
396 extern	int	ufs_HW;		/* high water mark */
397 extern	int	ufs_LW;		/* low water mark */
398 int	ufs_WRITES = 1;		/* XXX - enable/disable */
399 int	ufs_throttles = 0;	/* throttling count */
400 int	ufs_allow_shared_writes = 1;	/* directio shared writes */
401 
402 static int
403 ufs_check_rewrite(struct inode *ip, struct uio *uiop, int ioflag)
404 {
405 	int	shared_write;
406 
407 	/*
408 	 * If the FDSYNC flag is set then ignore the global
409 	 * ufs_allow_shared_writes in this case.
410 	 */
411 	shared_write = (ioflag & FDSYNC) | ufs_allow_shared_writes;
412 
413 	/*
414 	 * Filter to determine if this request is suitable as a
415 	 * concurrent rewrite. This write must not allocate blocks
416 	 * by extending the file or filling in holes. No use trying
417 	 * through FSYNC descriptors as the inode will be synchronously
418 	 * updated after the write. The uio structure has not yet been
419 	 * checked for sanity, so assume nothing.
420 	 */
421 	return (((ip->i_mode & IFMT) == IFREG) && !(ioflag & FAPPEND) &&
422 	    (uiop->uio_loffset >= (offset_t)0) &&
423 	    (uiop->uio_loffset < ip->i_size) && (uiop->uio_resid > 0) &&
424 	    ((ip->i_size - uiop->uio_loffset) >= uiop->uio_resid) &&
425 	    !(ioflag & FSYNC) && !bmap_has_holes(ip) &&
426 	    shared_write);
427 }
428 
429 /*ARGSUSED*/
430 static int
431 ufs_write(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cr,
432 	caller_context_t *ct)
433 {
434 	struct inode *ip = VTOI(vp);
435 	struct ufsvfs *ufsvfsp;
436 	struct ulockfs *ulp;
437 	int retry = 1;
438 	int error, resv, resid = 0;
439 	int directio_status;
440 	int exclusive;
441 	int rewriteflg;
442 	long start_resid = uiop->uio_resid;
443 
444 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
445 
446 retry_mandlock:
447 	/*
448 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
449 	 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
450 	 * Check for forced unmounts normally done in ufs_lockfs_begin().
451 	 */
452 	if ((ufsvfsp = ip->i_ufsvfs) == NULL) {
453 		error = EIO;
454 		goto out;
455 	}
456 	if (MANDLOCK(vp, ip->i_mode)) {
457 
458 		ASSERT(RW_WRITE_HELD(&ip->i_rwlock));
459 
460 		/*
461 		 * ufs_getattr ends up being called by chklock
462 		 */
463 		error = chklock(vp, FWRITE, uiop->uio_loffset,
464 		    uiop->uio_resid, uiop->uio_fmode, ct);
465 		if (error)
466 			goto out;
467 	}
468 
469 	/* i_rwlock can change in chklock */
470 	exclusive = rw_write_held(&ip->i_rwlock);
471 	rewriteflg = ufs_check_rewrite(ip, uiop, ioflag);
472 
473 	/*
474 	 * Check for fast-path special case of directio re-writes.
475 	 */
476 	if ((ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) &&
477 	    !exclusive && rewriteflg) {
478 
479 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
480 		if (error)
481 			goto out;
482 
483 		rw_enter(&ip->i_contents, RW_READER);
484 		error = ufs_directio_write(ip, uiop, ioflag, 1, cr,
485 		    &directio_status);
486 		if (directio_status == DIRECTIO_SUCCESS) {
487 			uint_t i_flag_save;
488 
489 			if (start_resid != uiop->uio_resid)
490 				error = 0;
491 			/*
492 			 * Special treatment of access times for re-writes.
493 			 * If IMOD is not already set, then convert it
494 			 * to IMODACC for this operation. This defers
495 			 * entering a delta into the log until the inode
496 			 * is flushed. This mimics what is done for read
497 			 * operations and inode access time.
498 			 */
499 			mutex_enter(&ip->i_tlock);
500 			i_flag_save = ip->i_flag;
501 			ip->i_flag |= IUPD | ICHG;
502 			ip->i_seq++;
503 			ITIMES_NOLOCK(ip);
504 			if ((i_flag_save & IMOD) == 0) {
505 				ip->i_flag &= ~IMOD;
506 				ip->i_flag |= IMODACC;
507 			}
508 			mutex_exit(&ip->i_tlock);
509 			rw_exit(&ip->i_contents);
510 			if (ulp)
511 				ufs_lockfs_end(ulp);
512 			goto out;
513 		}
514 		rw_exit(&ip->i_contents);
515 		if (ulp)
516 			ufs_lockfs_end(ulp);
517 	}
518 
519 	if (!exclusive && !rw_tryupgrade(&ip->i_rwlock)) {
520 		rw_exit(&ip->i_rwlock);
521 		rw_enter(&ip->i_rwlock, RW_WRITER);
522 		/*
523 		 * Mandatory locking could have been enabled
524 		 * after dropping the i_rwlock.
525 		 */
526 		if (MANDLOCK(vp, ip->i_mode))
527 			goto retry_mandlock;
528 	}
529 
530 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
531 	if (error)
532 		goto out;
533 
534 	/*
535 	 * Amount of log space needed for this write
536 	 */
537 	if (!rewriteflg || !(ioflag & FDSYNC))
538 		TRANS_WRITE_RESV(ip, uiop, ulp, &resv, &resid);
539 
540 	/*
541 	 * Throttle writes.
542 	 */
543 	if (ufs_WRITES && (ip->i_writes > ufs_HW)) {
544 		mutex_enter(&ip->i_tlock);
545 		while (ip->i_writes > ufs_HW) {
546 			ufs_throttles++;
547 			cv_wait(&ip->i_wrcv, &ip->i_tlock);
548 		}
549 		mutex_exit(&ip->i_tlock);
550 	}
551 
552 	/*
553 	 * Enter Transaction
554 	 *
555 	 * If the write is a rewrite there is no need to open a transaction
556 	 * if the FDSYNC flag is set and not the FSYNC.  In this case just
557 	 * set the IMODACC flag to modify do the update at a later time
558 	 * thus avoiding the overhead of the logging transaction that is
559 	 * not required.
560 	 */
561 	if (ioflag & (FSYNC|FDSYNC)) {
562 		if (ulp) {
563 			if (rewriteflg) {
564 				uint_t i_flag_save;
565 
566 				rw_enter(&ip->i_contents, RW_READER);
567 				mutex_enter(&ip->i_tlock);
568 				i_flag_save = ip->i_flag;
569 				ip->i_flag |= IUPD | ICHG;
570 				ip->i_seq++;
571 				ITIMES_NOLOCK(ip);
572 				if ((i_flag_save & IMOD) == 0) {
573 					ip->i_flag &= ~IMOD;
574 					ip->i_flag |= IMODACC;
575 				}
576 				mutex_exit(&ip->i_tlock);
577 				rw_exit(&ip->i_contents);
578 			} else {
579 				int terr = 0;
580 				TRANS_BEGIN_SYNC(ufsvfsp, TOP_WRITE_SYNC, resv,
581 				    terr);
582 				ASSERT(!terr);
583 			}
584 		}
585 	} else {
586 		if (ulp)
587 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_WRITE, resv);
588 	}
589 
590 	/*
591 	 * Write the file
592 	 */
593 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
594 	rw_enter(&ip->i_contents, RW_WRITER);
595 	if ((ioflag & FAPPEND) != 0 && (ip->i_mode & IFMT) == IFREG) {
596 		/*
597 		 * In append mode start at end of file.
598 		 */
599 		uiop->uio_loffset = ip->i_size;
600 	}
601 
602 	/*
603 	 * Mild optimisation, don't call ufs_trans_write() unless we have to
604 	 * Also, suppress file system full messages if we will retry.
605 	 */
606 	if (retry)
607 		ip->i_flag |= IQUIET;
608 	if (resid) {
609 		TRANS_WRITE(ip, uiop, ioflag, error, ulp, cr, resv, resid);
610 	} else {
611 		error = wrip(ip, uiop, ioflag, cr);
612 	}
613 	ip->i_flag &= ~IQUIET;
614 
615 	rw_exit(&ip->i_contents);
616 	rw_exit(&ufsvfsp->vfs_dqrwlock);
617 
618 	/*
619 	 * Leave Transaction
620 	 */
621 	if (ulp) {
622 		if (ioflag & (FSYNC|FDSYNC)) {
623 			if (!rewriteflg) {
624 				int terr = 0;
625 
626 				TRANS_END_SYNC(ufsvfsp, terr, TOP_WRITE_SYNC,
627 				    resv);
628 				if (error == 0)
629 					error = terr;
630 			}
631 		} else {
632 			TRANS_END_ASYNC(ufsvfsp, TOP_WRITE, resv);
633 		}
634 		ufs_lockfs_end(ulp);
635 	}
636 out:
637 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
638 		/*
639 		 * Any blocks tied up in pending deletes?
640 		 */
641 		ufs_delete_drain_wait(ufsvfsp, 1);
642 		retry = 0;
643 		goto retry_mandlock;
644 	}
645 
646 	if (error == ENOSPC && (start_resid != uiop->uio_resid))
647 		error = 0;
648 
649 	return (error);
650 }
651 
652 /*
653  * Don't cache write blocks to files with the sticky bit set.
654  * Used to keep swap files from blowing the page cache on a server.
655  */
656 int stickyhack = 1;
657 
658 /*
659  * Free behind hacks.  The pager is busted.
660  * XXX - need to pass the information down to writedone() in a flag like B_SEQ
661  * or B_FREE_IF_TIGHT_ON_MEMORY.
662  */
663 int	freebehind = 1;
664 int	smallfile = 0;
665 u_offset_t smallfile64 = 32 * 1024;
666 
667 /*
668  * While we should, in most cases, cache the pages for write, we
669  * may also want to cache the pages for read as long as they are
670  * frequently re-usable.
671  *
672  * If cache_read_ahead = 1, the pages for read will go to the tail
673  * of the cache list when they are released, otherwise go to the head.
674  */
675 int	cache_read_ahead = 0;
676 
677 /*
678  * Freebehind exists  so that as we read  large files  sequentially we
679  * don't consume most of memory with pages  from a few files. It takes
680  * longer to re-read from disk multiple small files as it does reading
681  * one large one sequentially.  As system  memory grows customers need
682  * to retain bigger chunks   of files in  memory.   The advent of  the
683  * cachelist opens up of the possibility freeing pages  to the head or
684  * tail of the list.
685  *
686  * Not freeing a page is a bet that the page will be read again before
687  * it's segmap slot is needed for something else. If we loose the bet,
688  * it means some  other thread is  burdened with the  page free we did
689  * not do. If we win we save a free and reclaim.
690  *
691  * Freeing it at the tail  vs the head of cachelist  is a bet that the
692  * page will survive until the next  read.  It's also saying that this
693  * page is more likely to  be re-used than a  page freed some time ago
694  * and never reclaimed.
695  *
696  * Freebehind maintains a  range of  file offset [smallfile1; smallfile2]
697  *
698  *            0 < offset < smallfile1 : pages are not freed.
699  *   smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
700  *   smallfile2 < offset              : pages freed to head of cachelist.
701  *
702  * The range  is  computed  at most  once  per second  and  depends on
703  * freemem  and  ncpus_online.  Both parameters  are   bounded to be
704  * >= smallfile && >= smallfile64.
705  *
706  * smallfile1 = (free memory / ncpu) / 1000
707  * smallfile2 = (free memory / ncpu) / 10
708  *
709  * A few examples values:
710  *
711  *       Free Mem (in Bytes) [smallfile1; smallfile2]  [smallfile1; smallfile2]
712  *                                 ncpus_online = 4          ncpus_online = 64
713  *       ------------------  -----------------------   -----------------------
714  *             1G                   [256K;  25M]               [32K; 1.5M]
715  *            10G                   [2.5M; 250M]              [156K; 15M]
716  *           100G                    [25M; 2.5G]              [1.5M; 150M]
717  *
718  */
719 
720 #define	SMALLFILE1_D 1000
721 #define	SMALLFILE2_D 10
722 static u_offset_t smallfile1 = 32 * 1024;
723 static u_offset_t smallfile2 = 32 * 1024;
724 static clock_t smallfile_update = 0; /* lbolt value of when to recompute */
725 uint_t smallfile1_d = SMALLFILE1_D;
726 uint_t smallfile2_d = SMALLFILE2_D;
727 
728 /*
729  * wrip does the real work of write requests for ufs.
730  */
731 int
732 wrip(struct inode *ip, struct uio *uio, int ioflag, struct cred *cr)
733 {
734 	rlim64_t limit = uio->uio_llimit;
735 	u_offset_t off;
736 	u_offset_t old_i_size;
737 	struct fs *fs;
738 	struct vnode *vp;
739 	struct ufsvfs *ufsvfsp;
740 	caddr_t base;
741 	long start_resid = uio->uio_resid;	/* save starting resid */
742 	long premove_resid;			/* resid before uiomove() */
743 	uint_t flags;
744 	int newpage;
745 	int iupdat_flag, directio_status;
746 	int n, on, mapon;
747 	int error, pagecreate;
748 	int do_dqrwlock;		/* drop/reacquire vfs_dqrwlock */
749 	int32_t	iblocks;
750 	int	new_iblocks;
751 
752 	/*
753 	 * ip->i_size is incremented before the uiomove
754 	 * is done on a write.  If the move fails (bad user
755 	 * address) reset ip->i_size.
756 	 * The better way would be to increment ip->i_size
757 	 * only if the uiomove succeeds.
758 	 */
759 	int i_size_changed = 0;
760 	o_mode_t type;
761 	int i_seq_needed = 0;
762 
763 	vp = ITOV(ip);
764 
765 	/*
766 	 * check for forced unmount - should not happen as
767 	 * the request passed the lockfs checks.
768 	 */
769 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
770 		return (EIO);
771 
772 	fs = ip->i_fs;
773 
774 	ASSERT(RW_WRITE_HELD(&ip->i_contents));
775 
776 	/* check for valid filetype */
777 	type = ip->i_mode & IFMT;
778 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
779 	    (type != IFLNK) && (type != IFSHAD)) {
780 		return (EIO);
781 	}
782 
783 	/*
784 	 * the actual limit of UFS file size
785 	 * is UFS_MAXOFFSET_T
786 	 */
787 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
788 		limit = MAXOFFSET_T;
789 
790 	if (uio->uio_loffset >= limit) {
791 		proc_t *p = ttoproc(curthread);
792 
793 		mutex_enter(&p->p_lock);
794 		(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
795 		    p, RCA_UNSAFE_SIGINFO);
796 		mutex_exit(&p->p_lock);
797 		return (EFBIG);
798 	}
799 
800 	/*
801 	 * if largefiles are disallowed, the limit is
802 	 * the pre-largefiles value of 2GB
803 	 */
804 	if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
805 		limit = MIN(UFS_MAXOFFSET_T, limit);
806 	else
807 		limit = MIN(MAXOFF32_T, limit);
808 
809 	if (uio->uio_loffset < (offset_t)0) {
810 		return (EINVAL);
811 	}
812 	if (uio->uio_resid == 0) {
813 		return (0);
814 	}
815 
816 	if (uio->uio_loffset >= limit)
817 		return (EFBIG);
818 
819 	ip->i_flag |= INOACC;	/* don't update ref time in getpage */
820 
821 	if (ioflag & (FSYNC|FDSYNC)) {
822 		ip->i_flag |= ISYNC;
823 		iupdat_flag = 1;
824 	}
825 	/*
826 	 * Try to go direct
827 	 */
828 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
829 		uio->uio_llimit = limit;
830 		error = ufs_directio_write(ip, uio, ioflag, 0, cr,
831 		    &directio_status);
832 		/*
833 		 * If ufs_directio wrote to the file or set the flags,
834 		 * we need to update i_seq, but it may be deferred.
835 		 */
836 		if (start_resid != uio->uio_resid ||
837 		    (ip->i_flag & (ICHG|IUPD))) {
838 			i_seq_needed = 1;
839 			ip->i_flag |= ISEQ;
840 		}
841 		if (directio_status == DIRECTIO_SUCCESS)
842 			goto out;
843 	}
844 
845 	/*
846 	 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
847 	 *
848 	 * o shadow inodes: vfs_dqrwlock is not held at all
849 	 * o quota updates: vfs_dqrwlock is read or write held
850 	 * o other updates: vfs_dqrwlock is read held
851 	 *
852 	 * The first case is the only one where we do not hold
853 	 * vfs_dqrwlock at all while entering wrip().
854 	 * We must make sure not to downgrade/drop vfs_dqrwlock if we
855 	 * have it as writer, i.e. if we are updating the quota inode.
856 	 * There is no potential deadlock scenario in this case as
857 	 * ufs_getpage() takes care of this and avoids reacquiring
858 	 * vfs_dqrwlock in that case.
859 	 *
860 	 * This check is done here since the above conditions do not change
861 	 * and we possibly loop below, so save a few cycles.
862 	 */
863 	if ((type == IFSHAD) ||
864 	    (rw_owner(&ufsvfsp->vfs_dqrwlock) == curthread)) {
865 		do_dqrwlock = 0;
866 	} else {
867 		do_dqrwlock = 1;
868 	}
869 
870 	/*
871 	 * Large Files: We cast MAXBMASK to offset_t
872 	 * inorder to mask out the higher bits. Since offset_t
873 	 * is a signed value, the high order bit set in MAXBMASK
874 	 * value makes it do the right thing by having all bits 1
875 	 * in the higher word. May be removed for _SOLARIS64_.
876 	 */
877 
878 	fs = ip->i_fs;
879 	do {
880 		u_offset_t uoff = uio->uio_loffset;
881 		off = uoff & (offset_t)MAXBMASK;
882 		mapon = (int)(uoff & (offset_t)MAXBOFFSET);
883 		on = (int)blkoff(fs, uoff);
884 		n = (int)MIN(fs->fs_bsize - on, uio->uio_resid);
885 		new_iblocks = 1;
886 
887 		if (type == IFREG && uoff + n >= limit) {
888 			if (uoff >= limit) {
889 				error = EFBIG;
890 				goto out;
891 			}
892 			/*
893 			 * since uoff + n >= limit,
894 			 * therefore n >= limit - uoff, and n is an int
895 			 * so it is safe to cast it to an int
896 			 */
897 			n = (int)(limit - (rlim64_t)uoff);
898 		}
899 		if (uoff + n > ip->i_size) {
900 			/*
901 			 * We are extending the length of the file.
902 			 * bmap is used so that we are sure that
903 			 * if we need to allocate new blocks, that it
904 			 * is done here before we up the file size.
905 			 */
906 			error = bmap_write(ip, uoff, (int)(on + n),
907 			    mapon == 0, NULL, cr);
908 			/*
909 			 * bmap_write never drops i_contents so if
910 			 * the flags are set it changed the file.
911 			 */
912 			if (ip->i_flag & (ICHG|IUPD)) {
913 				i_seq_needed = 1;
914 				ip->i_flag |= ISEQ;
915 			}
916 			if (error)
917 				break;
918 			/*
919 			 * There is a window of vulnerability here.
920 			 * The sequence of operations: allocate file
921 			 * system blocks, uiomove the data into pages,
922 			 * and then update the size of the file in the
923 			 * inode, must happen atomically.  However, due
924 			 * to current locking constraints, this can not
925 			 * be done.
926 			 */
927 			ASSERT(ip->i_writer == NULL);
928 			ip->i_writer = curthread;
929 			i_size_changed = 1;
930 			/*
931 			 * If we are writing from the beginning of
932 			 * the mapping, we can just create the
933 			 * pages without having to read them.
934 			 */
935 			pagecreate = (mapon == 0);
936 		} else if (n == MAXBSIZE) {
937 			/*
938 			 * Going to do a whole mappings worth,
939 			 * so we can just create the pages w/o
940 			 * having to read them in.  But before
941 			 * we do that, we need to make sure any
942 			 * needed blocks are allocated first.
943 			 */
944 			iblocks = ip->i_blocks;
945 			error = bmap_write(ip, uoff, (int)(on + n),
946 			    BI_ALLOC_ONLY, NULL, cr);
947 			/*
948 			 * bmap_write never drops i_contents so if
949 			 * the flags are set it changed the file.
950 			 */
951 			if (ip->i_flag & (ICHG|IUPD)) {
952 				i_seq_needed = 1;
953 				ip->i_flag |= ISEQ;
954 			}
955 			if (error)
956 				break;
957 			pagecreate = 1;
958 			/*
959 			 * check if the new created page needed the
960 			 * allocation of new disk blocks.
961 			 */
962 			if (iblocks == ip->i_blocks)
963 				new_iblocks = 0; /* no new blocks allocated */
964 		} else {
965 			pagecreate = 0;
966 			/*
967 			 * In sync mode flush the indirect blocks which
968 			 * may have been allocated and not written on
969 			 * disk. In above cases bmap_write will allocate
970 			 * in sync mode.
971 			 */
972 			if (ioflag & (FSYNC|FDSYNC)) {
973 				error = ufs_indirblk_sync(ip, uoff);
974 				if (error)
975 					break;
976 			}
977 		}
978 
979 		/*
980 		 * At this point we can enter ufs_getpage() in one
981 		 * of two ways:
982 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
983 		 *    forcefault parameter is true (pagecreate == 0)
984 		 * 2) uiomove() causes a page fault.
985 		 *
986 		 * We have to drop the contents lock to prevent the VM
987 		 * system from trying to reacquire it in ufs_getpage()
988 		 * should the uiomove cause a pagefault.
989 		 *
990 		 * We have to drop the reader vfs_dqrwlock here as well.
991 		 */
992 		rw_exit(&ip->i_contents);
993 		if (do_dqrwlock) {
994 			ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock));
995 			ASSERT(!(RW_WRITE_HELD(&ufsvfsp->vfs_dqrwlock)));
996 			rw_exit(&ufsvfsp->vfs_dqrwlock);
997 		}
998 
999 		newpage = 0;
1000 		premove_resid = uio->uio_resid;
1001 
1002 		/*
1003 		 * Touch the page and fault it in if it is not in core
1004 		 * before segmap_getmapflt or vpm_data_copy can lock it.
1005 		 * This is to avoid the deadlock if the buffer is mapped
1006 		 * to the same file through mmap which we want to write.
1007 		 */
1008 		uio_prefaultpages((long)n, uio);
1009 
1010 		if (vpm_enable) {
1011 			/*
1012 			 * Copy data. If new pages are created, part of
1013 			 * the page that is not written will be initizliazed
1014 			 * with zeros.
1015 			 */
1016 			error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1017 			    uio, !pagecreate, &newpage, 0, S_WRITE);
1018 		} else {
1019 
1020 			base = segmap_getmapflt(segkmap, vp, (off + mapon),
1021 			    (uint_t)n, !pagecreate, S_WRITE);
1022 
1023 			/*
1024 			 * segmap_pagecreate() returns 1 if it calls
1025 			 * page_create_va() to allocate any pages.
1026 			 */
1027 
1028 			if (pagecreate)
1029 				newpage = segmap_pagecreate(segkmap, base,
1030 				    (size_t)n, 0);
1031 
1032 			error = uiomove(base + mapon, (long)n, UIO_WRITE, uio);
1033 		}
1034 
1035 		/*
1036 		 * If "newpage" is set, then a new page was created and it
1037 		 * does not contain valid data, so it needs to be initialized
1038 		 * at this point.
1039 		 * Otherwise the page contains old data, which was overwritten
1040 		 * partially or as a whole in uiomove.
1041 		 * If there is only one iovec structure within uio, then
1042 		 * on error uiomove will not be able to update uio->uio_loffset
1043 		 * and we would zero the whole page here!
1044 		 *
1045 		 * If uiomove fails because of an error, the old valid data
1046 		 * is kept instead of filling the rest of the page with zero's.
1047 		 */
1048 		if (!vpm_enable && newpage &&
1049 		    uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) {
1050 			/*
1051 			 * We created pages w/o initializing them completely,
1052 			 * thus we need to zero the part that wasn't set up.
1053 			 * This happens on most EOF write cases and if
1054 			 * we had some sort of error during the uiomove.
1055 			 */
1056 			int nzero, nmoved;
1057 
1058 			nmoved = (int)(uio->uio_loffset - (off + mapon));
1059 			ASSERT(nmoved >= 0 && nmoved <= n);
1060 			nzero = roundup(on + n, PAGESIZE) - nmoved;
1061 			ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE);
1062 			(void) kzero(base + mapon + nmoved, (uint_t)nzero);
1063 		}
1064 
1065 		/*
1066 		 * Unlock the pages allocated by page_create_va()
1067 		 * in segmap_pagecreate()
1068 		 */
1069 		if (!vpm_enable && newpage)
1070 			segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE);
1071 
1072 		/*
1073 		 * If the size of the file changed, then update the
1074 		 * size field in the inode now.  This can't be done
1075 		 * before the call to segmap_pageunlock or there is
1076 		 * a potential deadlock with callers to ufs_putpage().
1077 		 * They will be holding i_contents and trying to lock
1078 		 * a page, while this thread is holding a page locked
1079 		 * and trying to acquire i_contents.
1080 		 */
1081 		if (i_size_changed) {
1082 			rw_enter(&ip->i_contents, RW_WRITER);
1083 			old_i_size = ip->i_size;
1084 			UFS_SET_ISIZE(uoff + n, ip);
1085 			TRANS_INODE(ufsvfsp, ip);
1086 			/*
1087 			 * file has grown larger than 2GB. Set flag
1088 			 * in superblock to indicate this, if it
1089 			 * is not already set.
1090 			 */
1091 			if ((ip->i_size > MAXOFF32_T) &&
1092 			    !(fs->fs_flags & FSLARGEFILES)) {
1093 				ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES);
1094 				mutex_enter(&ufsvfsp->vfs_lock);
1095 				fs->fs_flags |= FSLARGEFILES;
1096 				ufs_sbwrite(ufsvfsp);
1097 				mutex_exit(&ufsvfsp->vfs_lock);
1098 			}
1099 			mutex_enter(&ip->i_tlock);
1100 			ip->i_writer = NULL;
1101 			cv_broadcast(&ip->i_wrcv);
1102 			mutex_exit(&ip->i_tlock);
1103 			rw_exit(&ip->i_contents);
1104 		}
1105 
1106 		if (error) {
1107 			/*
1108 			 * If we failed on a write, we may have already
1109 			 * allocated file blocks as well as pages.  It's
1110 			 * hard to undo the block allocation, but we must
1111 			 * be sure to invalidate any pages that may have
1112 			 * been allocated.
1113 			 *
1114 			 * If the page was created without initialization
1115 			 * then we must check if it should be possible
1116 			 * to destroy the new page and to keep the old data
1117 			 * on the disk.
1118 			 *
1119 			 * It is possible to destroy the page without
1120 			 * having to write back its contents only when
1121 			 * - the size of the file keeps unchanged
1122 			 * - bmap_write() did not allocate new disk blocks
1123 			 *   it is possible to create big files using "seek" and
1124 			 *   write to the end of the file. A "write" to a
1125 			 *   position before the end of the file would not
1126 			 *   change the size of the file but it would allocate
1127 			 *   new disk blocks.
1128 			 * - uiomove intended to overwrite the whole page.
1129 			 * - a new page was created (newpage == 1).
1130 			 */
1131 
1132 			if (i_size_changed == 0 && new_iblocks == 0 &&
1133 			    newpage) {
1134 
1135 				/* unwind what uiomove eventually last did */
1136 				uio->uio_resid = premove_resid;
1137 
1138 				/*
1139 				 * destroy the page, do not write ambiguous
1140 				 * data to the disk.
1141 				 */
1142 				flags = SM_DESTROY;
1143 			} else {
1144 				/*
1145 				 * write the page back to the disk, if dirty,
1146 				 * and remove the page from the cache.
1147 				 */
1148 				flags = SM_INVAL;
1149 			}
1150 
1151 			if (vpm_enable) {
1152 				/*
1153 				 *  Flush pages.
1154 				 */
1155 				(void) vpm_sync_pages(vp, off, n, flags);
1156 			} else {
1157 				(void) segmap_release(segkmap, base, flags);
1158 			}
1159 		} else {
1160 			flags = 0;
1161 			/*
1162 			 * Force write back for synchronous write cases.
1163 			 */
1164 			if ((ioflag & (FSYNC|FDSYNC)) || type == IFDIR) {
1165 				/*
1166 				 * If the sticky bit is set but the
1167 				 * execute bit is not set, we do a
1168 				 * synchronous write back and free
1169 				 * the page when done.  We set up swap
1170 				 * files to be handled this way to
1171 				 * prevent servers from keeping around
1172 				 * the client's swap pages too long.
1173 				 * XXX - there ought to be a better way.
1174 				 */
1175 				if (IS_SWAPVP(vp)) {
1176 					flags = SM_WRITE | SM_FREE |
1177 					    SM_DONTNEED;
1178 					iupdat_flag = 0;
1179 				} else {
1180 					flags = SM_WRITE;
1181 				}
1182 			} else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
1183 				/*
1184 				 * Have written a whole block.
1185 				 * Start an asynchronous write and
1186 				 * mark the buffer to indicate that
1187 				 * it won't be needed again soon.
1188 				 */
1189 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
1190 			}
1191 			if (vpm_enable) {
1192 				/*
1193 				 * Flush pages.
1194 				 */
1195 				error = vpm_sync_pages(vp, off, n, flags);
1196 			} else {
1197 				error = segmap_release(segkmap, base, flags);
1198 			}
1199 			/*
1200 			 * If the operation failed and is synchronous,
1201 			 * then we need to unwind what uiomove() last
1202 			 * did so we can potentially return an error to
1203 			 * the caller.  If this write operation was
1204 			 * done in two pieces and the first succeeded,
1205 			 * then we won't return an error for the second
1206 			 * piece that failed.  However, we only want to
1207 			 * return a resid value that reflects what was
1208 			 * really done.
1209 			 *
1210 			 * Failures for non-synchronous operations can
1211 			 * be ignored since the page subsystem will
1212 			 * retry the operation until it succeeds or the
1213 			 * file system is unmounted.
1214 			 */
1215 			if (error) {
1216 				if ((ioflag & (FSYNC | FDSYNC)) ||
1217 				    type == IFDIR) {
1218 					uio->uio_resid = premove_resid;
1219 				} else {
1220 					error = 0;
1221 				}
1222 			}
1223 		}
1224 
1225 		/*
1226 		 * Re-acquire contents lock.
1227 		 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1228 		 */
1229 		if (do_dqrwlock)
1230 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1231 		rw_enter(&ip->i_contents, RW_WRITER);
1232 
1233 		/*
1234 		 * If the uiomove() failed or if a synchronous
1235 		 * page push failed, fix up i_size.
1236 		 */
1237 		if (error) {
1238 			if (i_size_changed) {
1239 				/*
1240 				 * The uiomove failed, and we
1241 				 * allocated blocks,so get rid
1242 				 * of them.
1243 				 */
1244 				(void) ufs_itrunc(ip, old_i_size, 0, cr);
1245 			}
1246 		} else {
1247 			/*
1248 			 * XXX - Can this be out of the loop?
1249 			 */
1250 			ip->i_flag |= IUPD | ICHG;
1251 			/*
1252 			 * Only do one increase of i_seq for multiple
1253 			 * pieces.  Because we drop locks, record
1254 			 * the fact that we changed the timestamp and
1255 			 * are deferring the increase in case another thread
1256 			 * pushes our timestamp update.
1257 			 */
1258 			i_seq_needed = 1;
1259 			ip->i_flag |= ISEQ;
1260 			if (i_size_changed)
1261 				ip->i_flag |= IATTCHG;
1262 			if ((ip->i_mode & (IEXEC | (IEXEC >> 3) |
1263 			    (IEXEC >> 6))) != 0 &&
1264 			    (ip->i_mode & (ISUID | ISGID)) != 0 &&
1265 			    secpolicy_vnode_setid_retain(cr,
1266 			    (ip->i_mode & ISUID) != 0 && ip->i_uid == 0) != 0) {
1267 				/*
1268 				 * Clear Set-UID & Set-GID bits on
1269 				 * successful write if not privileged
1270 				 * and at least one of the execute bits
1271 				 * is set.  If we always clear Set-GID,
1272 				 * mandatory file and record locking is
1273 				 * unuseable.
1274 				 */
1275 				ip->i_mode &= ~(ISUID | ISGID);
1276 			}
1277 		}
1278 		/*
1279 		 * In the case the FDSYNC flag is set and this is a
1280 		 * "rewrite" we won't log a delta.
1281 		 * The FSYNC flag overrides all cases.
1282 		 */
1283 		if (!ufs_check_rewrite(ip, uio, ioflag) || !(ioflag & FDSYNC)) {
1284 			TRANS_INODE(ufsvfsp, ip);
1285 		}
1286 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1287 
1288 out:
1289 	/*
1290 	 * Make sure i_seq is increased at least once per write
1291 	 */
1292 	if (i_seq_needed) {
1293 		ip->i_seq++;
1294 		ip->i_flag &= ~ISEQ;	/* no longer deferred */
1295 	}
1296 
1297 	/*
1298 	 * Inode is updated according to this table -
1299 	 *
1300 	 *   FSYNC	  FDSYNC(posix.4)
1301 	 *   --------------------------
1302 	 *   always@	  IATTCHG|IBDWRITE
1303 	 *
1304 	 * @ - 	If we are doing synchronous write the only time we should
1305 	 *	not be sync'ing the ip here is if we have the stickyhack
1306 	 *	activated, the file is marked with the sticky bit and
1307 	 *	no exec bit, the file length has not been changed and
1308 	 *	no new blocks have been allocated during this write.
1309 	 */
1310 
1311 	if ((ip->i_flag & ISYNC) != 0) {
1312 		/*
1313 		 * we have eliminated nosync
1314 		 */
1315 		if ((ip->i_flag & (IATTCHG|IBDWRITE)) ||
1316 		    ((ioflag & FSYNC) && iupdat_flag)) {
1317 			ufs_iupdat(ip, 1);
1318 		}
1319 	}
1320 
1321 	/*
1322 	 * If we've already done a partial-write, terminate
1323 	 * the write but return no error unless the error is ENOSPC
1324 	 * because the caller can detect this and free resources and
1325 	 * try again.
1326 	 */
1327 	if ((start_resid != uio->uio_resid) && (error != ENOSPC))
1328 		error = 0;
1329 
1330 	ip->i_flag &= ~(INOACC | ISYNC);
1331 	ITIMES_NOLOCK(ip);
1332 	return (error);
1333 }
1334 
1335 /*
1336  * rdip does the real work of read requests for ufs.
1337  */
1338 int
1339 rdip(struct inode *ip, struct uio *uio, int ioflag, cred_t *cr)
1340 {
1341 	u_offset_t off;
1342 	caddr_t base;
1343 	struct fs *fs;
1344 	struct ufsvfs *ufsvfsp;
1345 	struct vnode *vp;
1346 	long oresid = uio->uio_resid;
1347 	u_offset_t n, on, mapon;
1348 	int error = 0;
1349 	int doupdate = 1;
1350 	uint_t flags;
1351 	int dofree, directio_status;
1352 	krw_t rwtype;
1353 	o_mode_t type;
1354 	clock_t	now;
1355 
1356 	vp = ITOV(ip);
1357 
1358 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
1359 
1360 	ufsvfsp = ip->i_ufsvfs;
1361 
1362 	if (ufsvfsp == NULL)
1363 		return (EIO);
1364 
1365 	fs = ufsvfsp->vfs_fs;
1366 
1367 	/* check for valid filetype */
1368 	type = ip->i_mode & IFMT;
1369 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
1370 	    (type != IFLNK) && (type != IFSHAD)) {
1371 		return (EIO);
1372 	}
1373 
1374 	if (uio->uio_loffset > UFS_MAXOFFSET_T) {
1375 		error = 0;
1376 		goto out;
1377 	}
1378 	if (uio->uio_loffset < (offset_t)0) {
1379 		return (EINVAL);
1380 	}
1381 	if (uio->uio_resid == 0) {
1382 		return (0);
1383 	}
1384 
1385 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (fs->fs_ronly == 0) &&
1386 	    (!ufsvfsp->vfs_noatime)) {
1387 		mutex_enter(&ip->i_tlock);
1388 		ip->i_flag |= IACC;
1389 		mutex_exit(&ip->i_tlock);
1390 	}
1391 	/*
1392 	 * Try to go direct
1393 	 */
1394 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
1395 		error = ufs_directio_read(ip, uio, cr, &directio_status);
1396 		if (directio_status == DIRECTIO_SUCCESS)
1397 			goto out;
1398 	}
1399 
1400 	rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER);
1401 
1402 	do {
1403 		offset_t diff;
1404 		u_offset_t uoff = uio->uio_loffset;
1405 		off = uoff & (offset_t)MAXBMASK;
1406 		mapon = (u_offset_t)(uoff & (offset_t)MAXBOFFSET);
1407 		on = (u_offset_t)blkoff(fs, uoff);
1408 		n = MIN((u_offset_t)fs->fs_bsize - on,
1409 		    (u_offset_t)uio->uio_resid);
1410 
1411 		diff = ip->i_size - uoff;
1412 
1413 		if (diff <= (offset_t)0) {
1414 			error = 0;
1415 			goto out;
1416 		}
1417 		if (diff < (offset_t)n)
1418 			n = (int)diff;
1419 
1420 		/*
1421 		 * We update smallfile2 and smallfile1 at most every second.
1422 		 */
1423 		now = ddi_get_lbolt();
1424 		if (now >= smallfile_update) {
1425 			uint64_t percpufreeb;
1426 			if (smallfile1_d == 0) smallfile1_d = SMALLFILE1_D;
1427 			if (smallfile2_d == 0) smallfile2_d = SMALLFILE2_D;
1428 			percpufreeb = ptob((uint64_t)freemem) / ncpus_online;
1429 			smallfile1 = percpufreeb / smallfile1_d;
1430 			smallfile2 = percpufreeb / smallfile2_d;
1431 			smallfile1 = MAX(smallfile1, smallfile);
1432 			smallfile1 = MAX(smallfile1, smallfile64);
1433 			smallfile2 = MAX(smallfile1, smallfile2);
1434 			smallfile_update = now + hz;
1435 		}
1436 
1437 		dofree = freebehind &&
1438 		    ip->i_nextr == (off & PAGEMASK) && off > smallfile1;
1439 
1440 		/*
1441 		 * At this point we can enter ufs_getpage() in one of two
1442 		 * ways:
1443 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
1444 		 *    forcefault parameter is true (value of 1 is passed)
1445 		 * 2) uiomove() causes a page fault.
1446 		 *
1447 		 * We cannot hold onto an i_contents reader lock without
1448 		 * risking deadlock in ufs_getpage() so drop a reader lock.
1449 		 * The ufs_getpage() dolock logic already allows for a
1450 		 * thread holding i_contents as writer to work properly
1451 		 * so we keep a writer lock.
1452 		 */
1453 		if (rwtype == RW_READER)
1454 			rw_exit(&ip->i_contents);
1455 
1456 		if (vpm_enable) {
1457 			/*
1458 			 * Copy data.
1459 			 */
1460 			error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1461 			    uio, 1, NULL, 0, S_READ);
1462 		} else {
1463 			base = segmap_getmapflt(segkmap, vp, (off + mapon),
1464 			    (uint_t)n, 1, S_READ);
1465 			error = uiomove(base + mapon, (long)n, UIO_READ, uio);
1466 		}
1467 
1468 		flags = 0;
1469 		if (!error) {
1470 			/*
1471 			 * If  reading sequential  we won't need  this
1472 			 * buffer again  soon.  For  offsets in  range
1473 			 * [smallfile1,  smallfile2] release the pages
1474 			 * at   the  tail  of the   cache list, larger
1475 			 * offsets are released at the head.
1476 			 */
1477 			if (dofree) {
1478 				flags = SM_FREE | SM_ASYNC;
1479 				if ((cache_read_ahead == 0) &&
1480 				    (off > smallfile2))
1481 					flags |=  SM_DONTNEED;
1482 			}
1483 			/*
1484 			 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1485 			 * we want to make sure that the page which has
1486 			 * been read, is written on disk if it is dirty.
1487 			 * And corresponding indirect blocks should also
1488 			 * be flushed out.
1489 			 */
1490 			if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) {
1491 				flags &= ~SM_ASYNC;
1492 				flags |= SM_WRITE;
1493 			}
1494 			if (vpm_enable) {
1495 				error = vpm_sync_pages(vp, off, n, flags);
1496 			} else {
1497 				error = segmap_release(segkmap, base, flags);
1498 			}
1499 		} else {
1500 			if (vpm_enable) {
1501 				(void) vpm_sync_pages(vp, off, n, flags);
1502 			} else {
1503 				(void) segmap_release(segkmap, base, flags);
1504 			}
1505 		}
1506 
1507 		if (rwtype == RW_READER)
1508 			rw_enter(&ip->i_contents, rwtype);
1509 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1510 out:
1511 	/*
1512 	 * Inode is updated according to this table if FRSYNC is set.
1513 	 *
1514 	 *   FSYNC	  FDSYNC(posix.4)
1515 	 *   --------------------------
1516 	 *   always	  IATTCHG|IBDWRITE
1517 	 */
1518 	/*
1519 	 * The inode is not updated if we're logging and the inode is a
1520 	 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1521 	 */
1522 	if (ioflag & FRSYNC) {
1523 		if (TRANS_ISTRANS(ufsvfsp) && ((ip->i_mode & IFMT) == IFDIR)) {
1524 			doupdate = 0;
1525 		}
1526 		if (doupdate) {
1527 			if ((ioflag & FSYNC) ||
1528 			    ((ioflag & FDSYNC) &&
1529 			    (ip->i_flag & (IATTCHG|IBDWRITE)))) {
1530 				ufs_iupdat(ip, 1);
1531 			}
1532 		}
1533 	}
1534 	/*
1535 	 * If we've already done a partial read, terminate
1536 	 * the read but return no error.
1537 	 */
1538 	if (oresid != uio->uio_resid)
1539 		error = 0;
1540 	ITIMES(ip);
1541 
1542 	return (error);
1543 }
1544 
1545 /* ARGSUSED */
1546 static int
1547 ufs_ioctl(
1548 	struct vnode	*vp,
1549 	int		cmd,
1550 	intptr_t	arg,
1551 	int		flag,
1552 	struct cred	*cr,
1553 	int		*rvalp,
1554 	caller_context_t *ct)
1555 {
1556 	struct lockfs	lockfs, lockfs_out;
1557 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
1558 	char		*comment, *original_comment;
1559 	struct fs	*fs;
1560 	struct ulockfs	*ulp;
1561 	offset_t	off;
1562 	extern int	maxphys;
1563 	int		error;
1564 	int		issync;
1565 	int		trans_size;
1566 
1567 
1568 	/*
1569 	 * forcibly unmounted
1570 	 */
1571 	if (ufsvfsp == NULL || vp->v_vfsp == NULL ||
1572 	    vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1573 		return (EIO);
1574 	fs = ufsvfsp->vfs_fs;
1575 
1576 	if (cmd == Q_QUOTACTL) {
1577 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_QUOTA_MASK);
1578 		if (error)
1579 			return (error);
1580 
1581 		if (ulp) {
1582 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_QUOTA,
1583 			    TOP_SETQUOTA_SIZE(fs));
1584 		}
1585 
1586 		error = quotactl(vp, arg, flag, cr);
1587 
1588 		if (ulp) {
1589 			TRANS_END_ASYNC(ufsvfsp, TOP_QUOTA,
1590 			    TOP_SETQUOTA_SIZE(fs));
1591 			ufs_lockfs_end(ulp);
1592 		}
1593 		return (error);
1594 	}
1595 
1596 	switch (cmd) {
1597 		case _FIOLFS:
1598 			/*
1599 			 * file system locking
1600 			 */
1601 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1602 				return (EPERM);
1603 
1604 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1605 				if (copyin((caddr_t)arg, &lockfs,
1606 				    sizeof (struct lockfs)))
1607 					return (EFAULT);
1608 			}
1609 #ifdef _SYSCALL32_IMPL
1610 			else {
1611 				struct lockfs32	lockfs32;
1612 				/* Translate ILP32 lockfs to LP64 lockfs */
1613 				if (copyin((caddr_t)arg, &lockfs32,
1614 				    sizeof (struct lockfs32)))
1615 					return (EFAULT);
1616 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1617 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1618 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1619 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1620 				lockfs.lf_comment =
1621 				    (caddr_t)(uintptr_t)lockfs32.lf_comment;
1622 			}
1623 #endif /* _SYSCALL32_IMPL */
1624 
1625 			if (lockfs.lf_comlen) {
1626 				if (lockfs.lf_comlen > LOCKFS_MAXCOMMENTLEN)
1627 					return (ENAMETOOLONG);
1628 				comment =
1629 				    kmem_alloc(lockfs.lf_comlen, KM_SLEEP);
1630 				if (copyin(lockfs.lf_comment, comment,
1631 				    lockfs.lf_comlen)) {
1632 					kmem_free(comment, lockfs.lf_comlen);
1633 					return (EFAULT);
1634 				}
1635 				original_comment = lockfs.lf_comment;
1636 				lockfs.lf_comment = comment;
1637 			}
1638 			if ((error = ufs_fiolfs(vp, &lockfs, 0)) == 0) {
1639 				lockfs.lf_comment = original_comment;
1640 
1641 				if ((flag & DATAMODEL_MASK) ==
1642 				    DATAMODEL_NATIVE) {
1643 					(void) copyout(&lockfs, (caddr_t)arg,
1644 					    sizeof (struct lockfs));
1645 				}
1646 #ifdef _SYSCALL32_IMPL
1647 				else {
1648 					struct lockfs32	lockfs32;
1649 					/* Translate LP64 to ILP32 lockfs */
1650 					lockfs32.lf_lock =
1651 					    (uint32_t)lockfs.lf_lock;
1652 					lockfs32.lf_flags =
1653 					    (uint32_t)lockfs.lf_flags;
1654 					lockfs32.lf_key =
1655 					    (uint32_t)lockfs.lf_key;
1656 					lockfs32.lf_comlen =
1657 					    (uint32_t)lockfs.lf_comlen;
1658 					lockfs32.lf_comment =
1659 					    (uint32_t)(uintptr_t)
1660 					    lockfs.lf_comment;
1661 					(void) copyout(&lockfs32, (caddr_t)arg,
1662 					    sizeof (struct lockfs32));
1663 				}
1664 #endif /* _SYSCALL32_IMPL */
1665 
1666 			} else {
1667 				if (lockfs.lf_comlen)
1668 					kmem_free(comment, lockfs.lf_comlen);
1669 			}
1670 			return (error);
1671 
1672 		case _FIOLFSS:
1673 			/*
1674 			 * get file system locking status
1675 			 */
1676 
1677 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1678 				if (copyin((caddr_t)arg, &lockfs,
1679 				    sizeof (struct lockfs)))
1680 					return (EFAULT);
1681 			}
1682 #ifdef _SYSCALL32_IMPL
1683 			else {
1684 				struct lockfs32	lockfs32;
1685 				/* Translate ILP32 lockfs to LP64 lockfs */
1686 				if (copyin((caddr_t)arg, &lockfs32,
1687 				    sizeof (struct lockfs32)))
1688 					return (EFAULT);
1689 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1690 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1691 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1692 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1693 				lockfs.lf_comment =
1694 				    (caddr_t)(uintptr_t)lockfs32.lf_comment;
1695 			}
1696 #endif /* _SYSCALL32_IMPL */
1697 
1698 			if (error =  ufs_fiolfss(vp, &lockfs_out))
1699 				return (error);
1700 			lockfs.lf_lock = lockfs_out.lf_lock;
1701 			lockfs.lf_key = lockfs_out.lf_key;
1702 			lockfs.lf_flags = lockfs_out.lf_flags;
1703 			lockfs.lf_comlen = MIN(lockfs.lf_comlen,
1704 			    lockfs_out.lf_comlen);
1705 
1706 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1707 				if (copyout(&lockfs, (caddr_t)arg,
1708 				    sizeof (struct lockfs)))
1709 					return (EFAULT);
1710 			}
1711 #ifdef _SYSCALL32_IMPL
1712 			else {
1713 				/* Translate LP64 to ILP32 lockfs */
1714 				struct lockfs32	lockfs32;
1715 				lockfs32.lf_lock = (uint32_t)lockfs.lf_lock;
1716 				lockfs32.lf_flags = (uint32_t)lockfs.lf_flags;
1717 				lockfs32.lf_key = (uint32_t)lockfs.lf_key;
1718 				lockfs32.lf_comlen = (uint32_t)lockfs.lf_comlen;
1719 				lockfs32.lf_comment =
1720 				    (uint32_t)(uintptr_t)lockfs.lf_comment;
1721 				if (copyout(&lockfs32, (caddr_t)arg,
1722 				    sizeof (struct lockfs32)))
1723 					return (EFAULT);
1724 			}
1725 #endif /* _SYSCALL32_IMPL */
1726 
1727 			if (lockfs.lf_comlen &&
1728 			    lockfs.lf_comment && lockfs_out.lf_comment)
1729 				if (copyout(lockfs_out.lf_comment,
1730 				    lockfs.lf_comment, lockfs.lf_comlen))
1731 					return (EFAULT);
1732 			return (0);
1733 
1734 		case _FIOSATIME:
1735 			/*
1736 			 * set access time
1737 			 */
1738 
1739 			/*
1740 			 * if mounted w/o atime, return quietly.
1741 			 * I briefly thought about returning ENOSYS, but
1742 			 * figured that most apps would consider this fatal
1743 			 * but the idea is to make this as seamless as poss.
1744 			 */
1745 			if (ufsvfsp->vfs_noatime)
1746 				return (0);
1747 
1748 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1749 			    ULOCKFS_SETATTR_MASK);
1750 			if (error)
1751 				return (error);
1752 
1753 			if (ulp) {
1754 				trans_size = (int)TOP_SETATTR_SIZE(VTOI(vp));
1755 				TRANS_BEGIN_CSYNC(ufsvfsp, issync,
1756 				    TOP_SETATTR, trans_size);
1757 			}
1758 
1759 			error = ufs_fiosatime(vp, (struct timeval *)arg,
1760 			    flag, cr);
1761 
1762 			if (ulp) {
1763 				TRANS_END_CSYNC(ufsvfsp, error, issync,
1764 				    TOP_SETATTR, trans_size);
1765 				ufs_lockfs_end(ulp);
1766 			}
1767 			return (error);
1768 
1769 		case _FIOSDIO:
1770 			/*
1771 			 * set delayed-io
1772 			 */
1773 			return (ufs_fiosdio(vp, (uint_t *)arg, flag, cr));
1774 
1775 		case _FIOGDIO:
1776 			/*
1777 			 * get delayed-io
1778 			 */
1779 			return (ufs_fiogdio(vp, (uint_t *)arg, flag, cr));
1780 
1781 		case _FIOIO:
1782 			/*
1783 			 * inode open
1784 			 */
1785 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1786 			    ULOCKFS_VGET_MASK);
1787 			if (error)
1788 				return (error);
1789 
1790 			error = ufs_fioio(vp, (struct fioio *)arg, flag, cr);
1791 
1792 			if (ulp) {
1793 				ufs_lockfs_end(ulp);
1794 			}
1795 			return (error);
1796 
1797 		case _FIOFFS:
1798 			/*
1799 			 * file system flush (push w/invalidate)
1800 			 */
1801 			if ((caddr_t)arg != NULL)
1802 				return (EINVAL);
1803 			return (ufs_fioffs(vp, NULL, cr));
1804 
1805 		case _FIOISBUSY:
1806 			/*
1807 			 * Contract-private interface for Legato
1808 			 * Purge this vnode from the DNLC and decide
1809 			 * if this vnode is busy (*arg == 1) or not
1810 			 * (*arg == 0)
1811 			 */
1812 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1813 				return (EPERM);
1814 			error = ufs_fioisbusy(vp, (int *)arg, cr);
1815 			return (error);
1816 
1817 		case _FIODIRECTIO:
1818 			return (ufs_fiodirectio(vp, (int)arg, cr));
1819 
1820 		case _FIOTUNE:
1821 			/*
1822 			 * Tune the file system (aka setting fs attributes)
1823 			 */
1824 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1825 			    ULOCKFS_SETATTR_MASK);
1826 			if (error)
1827 				return (error);
1828 
1829 			error = ufs_fiotune(vp, (struct fiotune *)arg, cr);
1830 
1831 			if (ulp)
1832 				ufs_lockfs_end(ulp);
1833 			return (error);
1834 
1835 		case _FIOLOGENABLE:
1836 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1837 				return (EPERM);
1838 			return (ufs_fiologenable(vp, (void *)arg, cr, flag));
1839 
1840 		case _FIOLOGDISABLE:
1841 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1842 				return (EPERM);
1843 			return (ufs_fiologdisable(vp, (void *)arg, cr, flag));
1844 
1845 		case _FIOISLOG:
1846 			return (ufs_fioislog(vp, (void *)arg, cr, flag));
1847 
1848 		case _FIOSNAPSHOTCREATE_MULTI:
1849 		{
1850 			struct fiosnapcreate_multi	fc, *fcp;
1851 			size_t	fcm_size;
1852 
1853 			if (copyin((void *)arg, &fc, sizeof (fc)))
1854 				return (EFAULT);
1855 			if (fc.backfilecount > MAX_BACKFILE_COUNT)
1856 				return (EINVAL);
1857 			fcm_size = sizeof (struct fiosnapcreate_multi) +
1858 			    (fc.backfilecount - 1) * sizeof (int);
1859 			fcp = (struct fiosnapcreate_multi *)
1860 			    kmem_alloc(fcm_size, KM_SLEEP);
1861 			if (copyin((void *)arg, fcp, fcm_size)) {
1862 				kmem_free(fcp, fcm_size);
1863 				return (EFAULT);
1864 			}
1865 			error = ufs_snap_create(vp, fcp, cr);
1866 			/*
1867 			 * Do copyout even if there is an error because
1868 			 * the details of error is stored in fcp.
1869 			 */
1870 			if (copyout(fcp, (void *)arg, fcm_size))
1871 				error = EFAULT;
1872 			kmem_free(fcp, fcm_size);
1873 			return (error);
1874 		}
1875 
1876 		case _FIOSNAPSHOTDELETE:
1877 		{
1878 			struct fiosnapdelete	fc;
1879 
1880 			if (copyin((void *)arg, &fc, sizeof (fc)))
1881 				return (EFAULT);
1882 			error = ufs_snap_delete(vp, &fc, cr);
1883 			if (!error && copyout(&fc, (void *)arg, sizeof (fc)))
1884 				error = EFAULT;
1885 			return (error);
1886 		}
1887 
1888 		case _FIOGETSUPERBLOCK:
1889 			if (copyout(fs, (void *)arg, SBSIZE))
1890 				return (EFAULT);
1891 			return (0);
1892 
1893 		case _FIOGETMAXPHYS:
1894 			if (copyout(&maxphys, (void *)arg, sizeof (maxphys)))
1895 				return (EFAULT);
1896 			return (0);
1897 
1898 		/*
1899 		 * The following 3 ioctls are for TSufs support
1900 		 * although could potentially be used elsewhere
1901 		 */
1902 		case _FIO_SET_LUFS_DEBUG:
1903 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1904 				return (EPERM);
1905 			lufs_debug = (uint32_t)arg;
1906 			return (0);
1907 
1908 		case _FIO_SET_LUFS_ERROR:
1909 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1910 				return (EPERM);
1911 			TRANS_SETERROR(ufsvfsp);
1912 			return (0);
1913 
1914 		case _FIO_GET_TOP_STATS:
1915 		{
1916 			fio_lufs_stats_t *ls;
1917 			ml_unit_t *ul = ufsvfsp->vfs_log;
1918 
1919 			ls = kmem_zalloc(sizeof (*ls), KM_SLEEP);
1920 			ls->ls_debug = ul->un_debug; /* return debug value */
1921 			/* Copy stucture if statistics are being kept */
1922 			if (ul->un_logmap->mtm_tops) {
1923 				ls->ls_topstats = *(ul->un_logmap->mtm_tops);
1924 			}
1925 			error = 0;
1926 			if (copyout(ls, (void *)arg, sizeof (*ls)))
1927 				error = EFAULT;
1928 			kmem_free(ls, sizeof (*ls));
1929 			return (error);
1930 		}
1931 
1932 		case _FIO_SEEK_DATA:
1933 		case _FIO_SEEK_HOLE:
1934 			if (ddi_copyin((void *)arg, &off, sizeof (off), flag))
1935 				return (EFAULT);
1936 			/* offset paramater is in/out */
1937 			error = ufs_fio_holey(vp, cmd, &off);
1938 			if (error)
1939 				return (error);
1940 			if (ddi_copyout(&off, (void *)arg, sizeof (off), flag))
1941 				return (EFAULT);
1942 			return (0);
1943 
1944 		case _FIO_COMPRESSED:
1945 		{
1946 			/*
1947 			 * This is a project private ufs ioctl() to mark
1948 			 * the inode as that belonging to a compressed
1949 			 * file. This is used to mark individual
1950 			 * compressed files in a miniroot archive.
1951 			 * The files compressed in this manner are
1952 			 * automatically decompressed by the dcfs filesystem
1953 			 * (via an interception in ufs_lookup - see decompvp())
1954 			 * which is layered on top of ufs on a system running
1955 			 * from the archive. See uts/common/fs/dcfs for details.
1956 			 * This ioctl only marks the file as compressed - the
1957 			 * actual compression is done by fiocompress (a
1958 			 * userland utility) which invokes this ioctl().
1959 			 */
1960 			struct inode *ip = VTOI(vp);
1961 
1962 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1963 			    ULOCKFS_SETATTR_MASK);
1964 			if (error)
1965 				return (error);
1966 
1967 			if (ulp) {
1968 				TRANS_BEGIN_ASYNC(ufsvfsp, TOP_IUPDAT,
1969 				    TOP_IUPDAT_SIZE(ip));
1970 			}
1971 
1972 			error = ufs_mark_compressed(vp);
1973 
1974 			if (ulp) {
1975 				TRANS_END_ASYNC(ufsvfsp, TOP_IUPDAT,
1976 				    TOP_IUPDAT_SIZE(ip));
1977 				ufs_lockfs_end(ulp);
1978 			}
1979 
1980 			return (error);
1981 
1982 		}
1983 
1984 		default:
1985 			return (ENOTTY);
1986 	}
1987 }
1988 
1989 
1990 /* ARGSUSED */
1991 static int
1992 ufs_getattr(struct vnode *vp, struct vattr *vap, int flags,
1993 	struct cred *cr, caller_context_t *ct)
1994 {
1995 	struct inode *ip = VTOI(vp);
1996 	struct ufsvfs *ufsvfsp;
1997 	int err;
1998 
1999 	if (vap->va_mask == AT_SIZE) {
2000 		/*
2001 		 * for performance, if only the size is requested don't bother
2002 		 * with anything else.
2003 		 */
2004 		UFS_GET_ISIZE(&vap->va_size, ip);
2005 		return (0);
2006 	}
2007 
2008 	/*
2009 	 * inlined lockfs checks
2010 	 */
2011 	ufsvfsp = ip->i_ufsvfs;
2012 	if ((ufsvfsp == NULL) || ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs)) {
2013 		err = EIO;
2014 		goto out;
2015 	}
2016 
2017 	rw_enter(&ip->i_contents, RW_READER);
2018 	/*
2019 	 * Return all the attributes.  This should be refined so
2020 	 * that it only returns what's asked for.
2021 	 */
2022 
2023 	/*
2024 	 * Copy from inode table.
2025 	 */
2026 	vap->va_type = vp->v_type;
2027 	vap->va_mode = ip->i_mode & MODEMASK;
2028 	/*
2029 	 * If there is an ACL and there is a mask entry, then do the
2030 	 * extra work that completes the equivalent of an acltomode(3)
2031 	 * call.  According to POSIX P1003.1e, the acl mask should be
2032 	 * returned in the group permissions field.
2033 	 *
2034 	 * - start with the original permission and mode bits (from above)
2035 	 * - clear the group owner bits
2036 	 * - add in the mask bits.
2037 	 */
2038 	if (ip->i_ufs_acl && ip->i_ufs_acl->aclass.acl_ismask) {
2039 		vap->va_mode &= ~((VREAD | VWRITE | VEXEC) >> 3);
2040 		vap->va_mode |=
2041 		    (ip->i_ufs_acl->aclass.acl_maskbits & PERMMASK) << 3;
2042 	}
2043 	vap->va_uid = ip->i_uid;
2044 	vap->va_gid = ip->i_gid;
2045 	vap->va_fsid = ip->i_dev;
2046 	vap->va_nodeid = (ino64_t)ip->i_number;
2047 	vap->va_nlink = ip->i_nlink;
2048 	vap->va_size = ip->i_size;
2049 	if (vp->v_type == VCHR || vp->v_type == VBLK)
2050 		vap->va_rdev = ip->i_rdev;
2051 	else
2052 		vap->va_rdev = 0;	/* not a b/c spec. */
2053 	mutex_enter(&ip->i_tlock);
2054 	ITIMES_NOLOCK(ip);	/* mark correct time in inode */
2055 	vap->va_seq = ip->i_seq;
2056 	vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec;
2057 	vap->va_atime.tv_nsec = ip->i_atime.tv_usec*1000;
2058 	vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec;
2059 	vap->va_mtime.tv_nsec = ip->i_mtime.tv_usec*1000;
2060 	vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec;
2061 	vap->va_ctime.tv_nsec = ip->i_ctime.tv_usec*1000;
2062 	mutex_exit(&ip->i_tlock);
2063 
2064 	switch (ip->i_mode & IFMT) {
2065 
2066 	case IFBLK:
2067 		vap->va_blksize = MAXBSIZE;		/* was BLKDEV_IOSIZE */
2068 		break;
2069 
2070 	case IFCHR:
2071 		vap->va_blksize = MAXBSIZE;
2072 		break;
2073 
2074 	default:
2075 		vap->va_blksize = ip->i_fs->fs_bsize;
2076 		break;
2077 	}
2078 	vap->va_nblocks = (fsblkcnt64_t)ip->i_blocks;
2079 	rw_exit(&ip->i_contents);
2080 	err = 0;
2081 
2082 out:
2083 	return (err);
2084 }
2085 
2086 /*
2087  * Special wrapper to provide a callback for secpolicy_vnode_setattr().
2088  * The i_contents lock is already held by the caller and we need to
2089  * declare the inode as 'void *' argument.
2090  */
2091 static int
2092 ufs_priv_access(void *vip, int mode, struct cred *cr)
2093 {
2094 	struct inode *ip = vip;
2095 
2096 	return (ufs_iaccess(ip, mode, cr, 0));
2097 }
2098 
2099 /*ARGSUSED4*/
2100 static int
2101 ufs_setattr(
2102 	struct vnode *vp,
2103 	struct vattr *vap,
2104 	int flags,
2105 	struct cred *cr,
2106 	caller_context_t *ct)
2107 {
2108 	struct inode *ip = VTOI(vp);
2109 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2110 	struct fs *fs;
2111 	struct ulockfs *ulp;
2112 	char *errmsg1;
2113 	char *errmsg2;
2114 	long blocks;
2115 	long int mask = vap->va_mask;
2116 	size_t len1, len2;
2117 	int issync;
2118 	int trans_size;
2119 	int dotrans;
2120 	int dorwlock;
2121 	int error;
2122 	int owner_change;
2123 	int dodqlock;
2124 	timestruc_t now;
2125 	vattr_t oldva;
2126 	int retry = 1;
2127 	int indeadlock;
2128 
2129 	/*
2130 	 * Cannot set these attributes.
2131 	 */
2132 	if ((mask & AT_NOSET) || (mask & AT_XVATTR))
2133 		return (EINVAL);
2134 
2135 	/*
2136 	 * check for forced unmount
2137 	 */
2138 	if (ufsvfsp == NULL)
2139 		return (EIO);
2140 
2141 	fs = ufsvfsp->vfs_fs;
2142 	if (fs->fs_ronly != 0)
2143 		return (EROFS);
2144 
2145 again:
2146 	errmsg1 = NULL;
2147 	errmsg2 = NULL;
2148 	dotrans = 0;
2149 	dorwlock = 0;
2150 	dodqlock = 0;
2151 
2152 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK);
2153 	if (error)
2154 		goto out;
2155 
2156 	/*
2157 	 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2158 	 * This follows the protocol for read()/write().
2159 	 */
2160 	if (vp->v_type != VDIR) {
2161 		/*
2162 		 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2163 		 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2164 		 * possible, retries the operation.
2165 		 */
2166 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_file);
2167 		if (indeadlock) {
2168 			if (ulp)
2169 				ufs_lockfs_end(ulp);
2170 			goto again;
2171 		}
2172 		dorwlock = 1;
2173 	}
2174 
2175 	/*
2176 	 * Truncate file.  Must have write permission and not be a directory.
2177 	 */
2178 	if (mask & AT_SIZE) {
2179 		rw_enter(&ip->i_contents, RW_WRITER);
2180 		if (vp->v_type == VDIR) {
2181 			error = EISDIR;
2182 			goto update_inode;
2183 		}
2184 		if (error = ufs_iaccess(ip, IWRITE, cr, 0))
2185 			goto update_inode;
2186 
2187 		rw_exit(&ip->i_contents);
2188 		error = TRANS_ITRUNC(ip, vap->va_size, 0, cr);
2189 		if (error) {
2190 			rw_enter(&ip->i_contents, RW_WRITER);
2191 			goto update_inode;
2192 		}
2193 	}
2194 
2195 	if (ulp) {
2196 		trans_size = (int)TOP_SETATTR_SIZE(ip);
2197 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size);
2198 		++dotrans;
2199 	}
2200 
2201 	/*
2202 	 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2203 	 * This follows the protocol established by
2204 	 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2205 	 */
2206 	if (vp->v_type == VDIR) {
2207 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_SETATTR,
2208 		    retry_dir);
2209 		if (indeadlock)
2210 			goto again;
2211 		dorwlock = 1;
2212 	}
2213 
2214 	/*
2215 	 * Grab quota lock if we are changing the file's owner.
2216 	 */
2217 	if (mask & AT_UID) {
2218 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2219 		dodqlock = 1;
2220 	}
2221 	rw_enter(&ip->i_contents, RW_WRITER);
2222 
2223 	oldva.va_mode = ip->i_mode;
2224 	oldva.va_uid = ip->i_uid;
2225 	oldva.va_gid = ip->i_gid;
2226 
2227 	vap->va_mask &= ~AT_SIZE;
2228 
2229 	error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2230 	    ufs_priv_access, ip);
2231 	if (error)
2232 		goto update_inode;
2233 
2234 	mask = vap->va_mask;
2235 
2236 	/*
2237 	 * Change file access modes.
2238 	 */
2239 	if (mask & AT_MODE) {
2240 		ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT);
2241 		TRANS_INODE(ufsvfsp, ip);
2242 		ip->i_flag |= ICHG;
2243 		if (stickyhack) {
2244 			mutex_enter(&vp->v_lock);
2245 			if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX)
2246 				vp->v_flag |= VSWAPLIKE;
2247 			else
2248 				vp->v_flag &= ~VSWAPLIKE;
2249 			mutex_exit(&vp->v_lock);
2250 		}
2251 	}
2252 	if (mask & (AT_UID|AT_GID)) {
2253 		if (mask & AT_UID) {
2254 			/*
2255 			 * Don't change ownership of the quota inode.
2256 			 */
2257 			if (ufsvfsp->vfs_qinod == ip) {
2258 				ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED);
2259 				error = EINVAL;
2260 				goto update_inode;
2261 			}
2262 
2263 			/*
2264 			 * No real ownership change.
2265 			 */
2266 			if (ip->i_uid == vap->va_uid) {
2267 				blocks = 0;
2268 				owner_change = 0;
2269 			}
2270 			/*
2271 			 * Remove the blocks and the file, from the old user's
2272 			 * quota.
2273 			 */
2274 			else {
2275 				blocks = ip->i_blocks;
2276 				owner_change = 1;
2277 
2278 				(void) chkdq(ip, -blocks, /* force */ 1, cr,
2279 				    (char **)NULL, (size_t *)NULL);
2280 				(void) chkiq(ufsvfsp, /* change */ -1, ip,
2281 				    (uid_t)ip->i_uid, /* force */ 1, cr,
2282 				    (char **)NULL, (size_t *)NULL);
2283 				dqrele(ip->i_dquot);
2284 			}
2285 
2286 			ip->i_uid = vap->va_uid;
2287 
2288 			/*
2289 			 * There is a real ownership change.
2290 			 */
2291 			if (owner_change) {
2292 				/*
2293 				 * Add the blocks and the file to the new
2294 				 * user's quota.
2295 				 */
2296 				ip->i_dquot = getinoquota(ip);
2297 				(void) chkdq(ip, blocks, /* force */ 1, cr,
2298 				    &errmsg1, &len1);
2299 				(void) chkiq(ufsvfsp, /* change */ 1,
2300 				    (struct inode *)NULL, (uid_t)ip->i_uid,
2301 				    /* force */ 1, cr, &errmsg2, &len2);
2302 			}
2303 		}
2304 		if (mask & AT_GID) {
2305 			ip->i_gid = vap->va_gid;
2306 		}
2307 		TRANS_INODE(ufsvfsp, ip);
2308 		ip->i_flag |= ICHG;
2309 	}
2310 	/*
2311 	 * Change file access or modified times.
2312 	 */
2313 	if (mask & (AT_ATIME|AT_MTIME)) {
2314 		/* Check that the time value is within ufs range */
2315 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2316 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2317 			error = EOVERFLOW;
2318 			goto update_inode;
2319 		}
2320 
2321 		/*
2322 		 * if the "noaccess" mount option is set and only atime
2323 		 * update is requested, do nothing. No error is returned.
2324 		 */
2325 		if ((ufsvfsp->vfs_noatime) &&
2326 		    ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME))
2327 			goto skip_atime;
2328 
2329 		if (mask & AT_ATIME) {
2330 			ip->i_atime.tv_sec = vap->va_atime.tv_sec;
2331 			ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000;
2332 			ip->i_flag &= ~IACC;
2333 		}
2334 		if (mask & AT_MTIME) {
2335 			ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
2336 			ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000;
2337 			gethrestime(&now);
2338 			if (now.tv_sec > TIME32_MAX) {
2339 				/*
2340 				 * In 2038, ctime sticks forever..
2341 				 */
2342 				ip->i_ctime.tv_sec = TIME32_MAX;
2343 				ip->i_ctime.tv_usec = 0;
2344 			} else {
2345 				ip->i_ctime.tv_sec = now.tv_sec;
2346 				ip->i_ctime.tv_usec = now.tv_nsec / 1000;
2347 			}
2348 			ip->i_flag &= ~(IUPD|ICHG);
2349 			ip->i_flag |= IMODTIME;
2350 		}
2351 		TRANS_INODE(ufsvfsp, ip);
2352 		ip->i_flag |= IMOD;
2353 	}
2354 
2355 skip_atime:
2356 	/*
2357 	 * The presence of a shadow inode may indicate an ACL, but does
2358 	 * not imply an ACL.  Future FSD types should be handled here too
2359 	 * and check for the presence of the attribute-specific data
2360 	 * before referencing it.
2361 	 */
2362 	if (ip->i_shadow) {
2363 		/*
2364 		 * XXX if ufs_iupdat is changed to sandbagged write fix
2365 		 * ufs_acl_setattr to push ip to keep acls consistent
2366 		 *
2367 		 * Suppress out of inodes messages if we will retry.
2368 		 */
2369 		if (retry)
2370 			ip->i_flag |= IQUIET;
2371 		error = ufs_acl_setattr(ip, vap, cr);
2372 		ip->i_flag &= ~IQUIET;
2373 	}
2374 
2375 update_inode:
2376 	/*
2377 	 * Setattr always increases the sequence number
2378 	 */
2379 	ip->i_seq++;
2380 
2381 	/*
2382 	 * if nfsd and not logging; push synchronously
2383 	 */
2384 	if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) {
2385 		ufs_iupdat(ip, 1);
2386 	} else {
2387 		ITIMES_NOLOCK(ip);
2388 	}
2389 
2390 	rw_exit(&ip->i_contents);
2391 	if (dodqlock) {
2392 		rw_exit(&ufsvfsp->vfs_dqrwlock);
2393 	}
2394 	if (dorwlock)
2395 		rw_exit(&ip->i_rwlock);
2396 
2397 	if (ulp) {
2398 		if (dotrans) {
2399 			int terr = 0;
2400 			TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR,
2401 			    trans_size);
2402 			if (error == 0)
2403 				error = terr;
2404 		}
2405 		ufs_lockfs_end(ulp);
2406 	}
2407 out:
2408 	/*
2409 	 * If out of inodes or blocks, see if we can free something
2410 	 * up from the delete queue.
2411 	 */
2412 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
2413 		ufs_delete_drain_wait(ufsvfsp, 1);
2414 		retry = 0;
2415 		if (errmsg1 != NULL)
2416 			kmem_free(errmsg1, len1);
2417 		if (errmsg2 != NULL)
2418 			kmem_free(errmsg2, len2);
2419 		goto again;
2420 	}
2421 	if (errmsg1 != NULL) {
2422 		uprintf(errmsg1);
2423 		kmem_free(errmsg1, len1);
2424 	}
2425 	if (errmsg2 != NULL) {
2426 		uprintf(errmsg2);
2427 		kmem_free(errmsg2, len2);
2428 	}
2429 	return (error);
2430 }
2431 
2432 /*ARGSUSED*/
2433 static int
2434 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr,
2435 	caller_context_t *ct)
2436 {
2437 	struct inode *ip = VTOI(vp);
2438 
2439 	if (ip->i_ufsvfs == NULL)
2440 		return (EIO);
2441 
2442 	/*
2443 	 * The ufs_iaccess function wants to be called with
2444 	 * mode bits expressed as "ufs specific" bits.
2445 	 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2446 	 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2447 	 * But since they're the same we just pass the vnode mode
2448 	 * bit but just verify that assumption at compile time.
2449 	 */
2450 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2451 #error "ufs_access needs to map Vmodes to Imodes"
2452 #endif
2453 	return (ufs_iaccess(ip, mode, cr, 1));
2454 }
2455 
2456 /* ARGSUSED */
2457 static int
2458 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr,
2459 	caller_context_t *ct)
2460 {
2461 	struct inode *ip = VTOI(vp);
2462 	struct ufsvfs *ufsvfsp;
2463 	struct ulockfs *ulp;
2464 	int error;
2465 	int fastsymlink;
2466 
2467 	if (vp->v_type != VLNK) {
2468 		error = EINVAL;
2469 		goto nolockout;
2470 	}
2471 
2472 	/*
2473 	 * If the symbolic link is empty there is nothing to read.
2474 	 * Fast-track these empty symbolic links
2475 	 */
2476 	if (ip->i_size == 0) {
2477 		error = 0;
2478 		goto nolockout;
2479 	}
2480 
2481 	ufsvfsp = ip->i_ufsvfs;
2482 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK);
2483 	if (error)
2484 		goto nolockout;
2485 	/*
2486 	 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2487 	 */
2488 again:
2489 	fastsymlink = 0;
2490 	if (ip->i_flag & IFASTSYMLNK) {
2491 		rw_enter(&ip->i_rwlock, RW_READER);
2492 		rw_enter(&ip->i_contents, RW_READER);
2493 		if (ip->i_flag & IFASTSYMLNK) {
2494 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
2495 			    (ip->i_fs->fs_ronly == 0) &&
2496 			    (!ufsvfsp->vfs_noatime)) {
2497 				mutex_enter(&ip->i_tlock);
2498 				ip->i_flag |= IACC;
2499 				mutex_exit(&ip->i_tlock);
2500 			}
2501 			error = uiomove((caddr_t)&ip->i_db[1],
2502 			    MIN(ip->i_size, uiop->uio_resid),
2503 			    UIO_READ, uiop);
2504 			ITIMES(ip);
2505 			++fastsymlink;
2506 		}
2507 		rw_exit(&ip->i_contents);
2508 		rw_exit(&ip->i_rwlock);
2509 	}
2510 	if (!fastsymlink) {
2511 		ssize_t size;	/* number of bytes read  */
2512 		caddr_t basep;	/* pointer to input data */
2513 		ino_t ino;
2514 		long  igen;
2515 		struct uio tuio;	/* temp uio struct */
2516 		struct uio *tuiop;
2517 		iovec_t tiov;		/* temp iovec struct */
2518 		char kbuf[FSL_SIZE];	/* buffer to hold fast symlink */
2519 		int tflag = 0;		/* flag to indicate temp vars used */
2520 
2521 		ino = ip->i_number;
2522 		igen = ip->i_gen;
2523 		size = uiop->uio_resid;
2524 		basep = uiop->uio_iov->iov_base;
2525 		tuiop = uiop;
2526 
2527 		rw_enter(&ip->i_rwlock, RW_WRITER);
2528 		rw_enter(&ip->i_contents, RW_WRITER);
2529 		if (ip->i_flag & IFASTSYMLNK) {
2530 			rw_exit(&ip->i_contents);
2531 			rw_exit(&ip->i_rwlock);
2532 			goto again;
2533 		}
2534 
2535 		/* can this be a fast symlink and is it a user buffer? */
2536 		if (ip->i_size <= FSL_SIZE &&
2537 		    (uiop->uio_segflg == UIO_USERSPACE ||
2538 		    uiop->uio_segflg == UIO_USERISPACE)) {
2539 
2540 			bzero(&tuio, sizeof (struct uio));
2541 			/*
2542 			 * setup a kernel buffer to read link into.  this
2543 			 * is to fix a race condition where the user buffer
2544 			 * got corrupted before copying it into the inode.
2545 			 */
2546 			size = ip->i_size;
2547 			tiov.iov_len = size;
2548 			tiov.iov_base = kbuf;
2549 			tuio.uio_iov = &tiov;
2550 			tuio.uio_iovcnt = 1;
2551 			tuio.uio_offset = uiop->uio_offset;
2552 			tuio.uio_segflg = UIO_SYSSPACE;
2553 			tuio.uio_fmode = uiop->uio_fmode;
2554 			tuio.uio_extflg = uiop->uio_extflg;
2555 			tuio.uio_limit = uiop->uio_limit;
2556 			tuio.uio_resid = size;
2557 
2558 			basep = tuio.uio_iov->iov_base;
2559 			tuiop = &tuio;
2560 			tflag = 1;
2561 		}
2562 
2563 		error = rdip(ip, tuiop, 0, cr);
2564 		if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) {
2565 			rw_exit(&ip->i_contents);
2566 			rw_exit(&ip->i_rwlock);
2567 			goto out;
2568 		}
2569 
2570 		if (tflag == 0)
2571 			size -= uiop->uio_resid;
2572 
2573 		if ((tflag == 0 && ip->i_size <= FSL_SIZE &&
2574 		    ip->i_size == size) || (tflag == 1 &&
2575 		    tuio.uio_resid == 0)) {
2576 			error = kcopy(basep, &ip->i_db[1], ip->i_size);
2577 			if (error == 0) {
2578 				ip->i_flag |= IFASTSYMLNK;
2579 				/*
2580 				 * free page
2581 				 */
2582 				(void) VOP_PUTPAGE(ITOV(ip),
2583 				    (offset_t)0, PAGESIZE,
2584 				    (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC),
2585 				    cr, ct);
2586 			} else {
2587 				int i;
2588 				/* error, clear garbage left behind */
2589 				for (i = 1; i < NDADDR; i++)
2590 					ip->i_db[i] = 0;
2591 				for (i = 0; i < NIADDR; i++)
2592 					ip->i_ib[i] = 0;
2593 			}
2594 		}
2595 		if (tflag == 1) {
2596 			/* now, copy it into the user buffer */
2597 			error = uiomove((caddr_t)kbuf,
2598 			    MIN(size, uiop->uio_resid),
2599 			    UIO_READ, uiop);
2600 		}
2601 		rw_exit(&ip->i_contents);
2602 		rw_exit(&ip->i_rwlock);
2603 	}
2604 out:
2605 	if (ulp) {
2606 		ufs_lockfs_end(ulp);
2607 	}
2608 nolockout:
2609 	return (error);
2610 }
2611 
2612 /* ARGSUSED */
2613 static int
2614 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr,
2615 	caller_context_t *ct)
2616 {
2617 	struct inode *ip = VTOI(vp);
2618 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2619 	struct ulockfs *ulp;
2620 	int error;
2621 
2622 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK);
2623 	if (error)
2624 		return (error);
2625 
2626 	if (TRANS_ISTRANS(ufsvfsp)) {
2627 		/*
2628 		 * First push out any data pages
2629 		 */
2630 		if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2631 		    (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) {
2632 			error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0,
2633 			    0, CRED(), ct);
2634 			if (error)
2635 				goto out;
2636 		}
2637 
2638 		/*
2639 		 * Delta any delayed inode times updates
2640 		 * and push inode to log.
2641 		 * All other inode deltas will have already been delta'd
2642 		 * and will be pushed during the commit.
2643 		 */
2644 		if (!(syncflag & FDSYNC) &&
2645 		    ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) {
2646 			if (ulp) {
2647 				TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC,
2648 				    TOP_SYNCIP_SIZE);
2649 			}
2650 			rw_enter(&ip->i_contents, RW_READER);
2651 			mutex_enter(&ip->i_tlock);
2652 			ip->i_flag &= ~IMODTIME;
2653 			mutex_exit(&ip->i_tlock);
2654 			ufs_iupdat(ip, I_SYNC);
2655 			rw_exit(&ip->i_contents);
2656 			if (ulp) {
2657 				TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC,
2658 				    TOP_SYNCIP_SIZE);
2659 			}
2660 		}
2661 
2662 		/*
2663 		 * Commit the Moby transaction
2664 		 *
2665 		 * Deltas have already been made so we just need to
2666 		 * commit them with a synchronous transaction.
2667 		 * TRANS_BEGIN_SYNC() will return an error
2668 		 * if there are no deltas to commit, for an
2669 		 * empty transaction.
2670 		 */
2671 		if (ulp) {
2672 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE,
2673 			    error);
2674 			if (error) {
2675 				error = 0; /* commit wasn't needed */
2676 				goto out;
2677 			}
2678 			TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC,
2679 			    TOP_COMMIT_SIZE);
2680 		}
2681 	} else {	/* not logging */
2682 		if (!(IS_SWAPVP(vp)))
2683 			if (syncflag & FNODSYNC) {
2684 				/* Just update the inode only */
2685 				TRANS_IUPDAT(ip, 1);
2686 				error = 0;
2687 			} else if (syncflag & FDSYNC)
2688 				/* Do data-synchronous writes */
2689 				error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC);
2690 			else
2691 				/* Do synchronous writes */
2692 				error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC);
2693 
2694 		rw_enter(&ip->i_contents, RW_WRITER);
2695 		if (!error)
2696 			error = ufs_sync_indir(ip);
2697 		rw_exit(&ip->i_contents);
2698 	}
2699 out:
2700 	if (ulp) {
2701 		ufs_lockfs_end(ulp);
2702 	}
2703 	return (error);
2704 }
2705 
2706 /*ARGSUSED*/
2707 static void
2708 ufs_inactive(struct vnode *vp, struct cred *cr, caller_context_t *ct)
2709 {
2710 	ufs_iinactive(VTOI(vp));
2711 }
2712 
2713 /*
2714  * Unix file system operations having to do with directory manipulation.
2715  */
2716 int ufs_lookup_idle_count = 2;	/* Number of inodes to idle each time */
2717 /* ARGSUSED */
2718 static int
2719 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
2720 	struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr,
2721 	caller_context_t *ct, int *direntflags, pathname_t *realpnp)
2722 {
2723 	struct inode *ip;
2724 	struct inode *sip;
2725 	struct inode *xip;
2726 	struct ufsvfs *ufsvfsp;
2727 	struct ulockfs *ulp;
2728 	struct vnode *vp;
2729 	int error;
2730 
2731 	/*
2732 	 * Check flags for type of lookup (regular file or attribute file)
2733 	 */
2734 
2735 	ip = VTOI(dvp);
2736 
2737 	if (flags & LOOKUP_XATTR) {
2738 
2739 		/*
2740 		 * If not mounted with XATTR support then return EINVAL
2741 		 */
2742 
2743 		if (!(ip->i_ufsvfs->vfs_vfs->vfs_flag & VFS_XATTR))
2744 			return (EINVAL);
2745 		/*
2746 		 * We don't allow recursive attributes...
2747 		 * Maybe someday we will.
2748 		 */
2749 		if ((ip->i_cflags & IXATTR)) {
2750 			return (EINVAL);
2751 		}
2752 
2753 		if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) {
2754 			error = ufs_xattr_getattrdir(dvp, &sip, flags, cr);
2755 			if (error) {
2756 				*vpp = NULL;
2757 				goto out;
2758 			}
2759 
2760 			vp = ITOV(sip);
2761 			dnlc_update(dvp, XATTR_DIR_NAME, vp);
2762 		}
2763 
2764 		/*
2765 		 * Check accessibility of directory.
2766 		 */
2767 		if (vp == DNLC_NO_VNODE) {
2768 			VN_RELE(vp);
2769 			error = ENOENT;
2770 			goto out;
2771 		}
2772 		if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr, 1)) != 0) {
2773 			VN_RELE(vp);
2774 			goto out;
2775 		}
2776 
2777 		*vpp = vp;
2778 		return (0);
2779 	}
2780 
2781 	/*
2782 	 * Check for a null component, which we should treat as
2783 	 * looking at dvp from within it's parent, so we don't
2784 	 * need a call to ufs_iaccess(), as it has already been
2785 	 * done.
2786 	 */
2787 	if (nm[0] == 0) {
2788 		VN_HOLD(dvp);
2789 		error = 0;
2790 		*vpp = dvp;
2791 		goto out;
2792 	}
2793 
2794 	/*
2795 	 * Check for "." ie itself. this is a quick check and
2796 	 * avoids adding "." into the dnlc (which have been seen
2797 	 * to occupy >10% of the cache).
2798 	 */
2799 	if ((nm[0] == '.') && (nm[1] == 0)) {
2800 		/*
2801 		 * Don't return without checking accessibility
2802 		 * of the directory. We only need the lock if
2803 		 * we are going to return it.
2804 		 */
2805 		if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) == 0) {
2806 			VN_HOLD(dvp);
2807 			*vpp = dvp;
2808 		}
2809 		goto out;
2810 	}
2811 
2812 	/*
2813 	 * Fast path: Check the directory name lookup cache.
2814 	 */
2815 	if (vp = dnlc_lookup(dvp, nm)) {
2816 		/*
2817 		 * Check accessibility of directory.
2818 		 */
2819 		if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) != 0) {
2820 			VN_RELE(vp);
2821 			goto out;
2822 		}
2823 		if (vp == DNLC_NO_VNODE) {
2824 			VN_RELE(vp);
2825 			error = ENOENT;
2826 			goto out;
2827 		}
2828 		xip = VTOI(vp);
2829 		ulp = NULL;
2830 		goto fastpath;
2831 	}
2832 
2833 	/*
2834 	 * Keep the idle queue from getting too long by
2835 	 * idling two inodes before attempting to allocate another.
2836 	 *    This operation must be performed before entering
2837 	 *    lockfs or a transaction.
2838 	 */
2839 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2840 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2841 			ins.in_lidles.value.ul += ufs_lookup_idle_count;
2842 			ufs_idle_some(ufs_lookup_idle_count);
2843 		}
2844 
2845 retry_lookup:
2846 	/*
2847 	 * Check accessibility of directory.
2848 	 */
2849 	if (error = ufs_diraccess(ip, IEXEC, cr))
2850 		goto out;
2851 
2852 	ufsvfsp = ip->i_ufsvfs;
2853 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK);
2854 	if (error)
2855 		goto out;
2856 
2857 	error = ufs_dirlook(ip, nm, &xip, cr, 1);
2858 
2859 fastpath:
2860 	if (error == 0) {
2861 		ip = xip;
2862 		*vpp = ITOV(ip);
2863 
2864 		/*
2865 		 * If vnode is a device return special vnode instead.
2866 		 */
2867 		if (IS_DEVVP(*vpp)) {
2868 			struct vnode *newvp;
2869 
2870 			newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
2871 			    cr);
2872 			VN_RELE(*vpp);
2873 			if (newvp == NULL)
2874 				error = ENOSYS;
2875 			else
2876 				*vpp = newvp;
2877 		} else if (ip->i_cflags & ICOMPRESS) {
2878 			struct vnode *newvp;
2879 
2880 			/*
2881 			 * Compressed file, substitute dcfs vnode
2882 			 */
2883 			newvp = decompvp(*vpp, cr, ct);
2884 			VN_RELE(*vpp);
2885 			if (newvp == NULL)
2886 				error = ENOSYS;
2887 			else
2888 				*vpp = newvp;
2889 		}
2890 	}
2891 	if (ulp) {
2892 		ufs_lockfs_end(ulp);
2893 	}
2894 
2895 	if (error == EAGAIN)
2896 		goto retry_lookup;
2897 
2898 out:
2899 	return (error);
2900 }
2901 
2902 /*ARGSUSED*/
2903 static int
2904 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl,
2905 	int mode, struct vnode **vpp, struct cred *cr, int flag,
2906 	caller_context_t *ct, vsecattr_t *vsecp)
2907 {
2908 	struct inode *ip;
2909 	struct inode *xip;
2910 	struct inode *dip;
2911 	struct vnode *xvp;
2912 	struct ufsvfs *ufsvfsp;
2913 	struct ulockfs *ulp;
2914 	int error;
2915 	int issync;
2916 	int truncflag;
2917 	int trans_size;
2918 	int noentry;
2919 	int defer_dip_seq_update = 0;	/* need to defer update of dip->i_seq */
2920 	int retry = 1;
2921 	int indeadlock;
2922 
2923 again:
2924 	ip = VTOI(dvp);
2925 	ufsvfsp = ip->i_ufsvfs;
2926 	truncflag = 0;
2927 
2928 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK);
2929 	if (error)
2930 		goto out;
2931 
2932 	if (ulp) {
2933 		trans_size = (int)TOP_CREATE_SIZE(ip);
2934 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size);
2935 	}
2936 
2937 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
2938 		vap->va_mode &= ~VSVTX;
2939 
2940 	if (*name == '\0') {
2941 		/*
2942 		 * Null component name refers to the directory itself.
2943 		 */
2944 		VN_HOLD(dvp);
2945 		/*
2946 		 * Even though this is an error case, we need to grab the
2947 		 * quota lock since the error handling code below is common.
2948 		 */
2949 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2950 		rw_enter(&ip->i_contents, RW_WRITER);
2951 		error = EEXIST;
2952 	} else {
2953 		xip = NULL;
2954 		noentry = 0;
2955 		/*
2956 		 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2957 		 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2958 		 * possible, retries the operation.
2959 		 */
2960 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_CREATE,
2961 		    retry_dir);
2962 		if (indeadlock)
2963 			goto again;
2964 
2965 		xvp = dnlc_lookup(dvp, name);
2966 		if (xvp == DNLC_NO_VNODE) {
2967 			noentry = 1;
2968 			VN_RELE(xvp);
2969 			xvp = NULL;
2970 		}
2971 		if (xvp) {
2972 			rw_exit(&ip->i_rwlock);
2973 			if (error = ufs_iaccess(ip, IEXEC, cr, 1)) {
2974 				VN_RELE(xvp);
2975 			} else {
2976 				error = EEXIST;
2977 				xip = VTOI(xvp);
2978 			}
2979 		} else {
2980 			/*
2981 			 * Suppress file system full message if we will retry
2982 			 */
2983 			error = ufs_direnter_cm(ip, name, DE_CREATE,
2984 			    vap, &xip, cr, (noentry | (retry ? IQUIET : 0)));
2985 			if (error == EAGAIN) {
2986 				if (ulp) {
2987 					TRANS_END_CSYNC(ufsvfsp, error, issync,
2988 					    TOP_CREATE, trans_size);
2989 					ufs_lockfs_end(ulp);
2990 				}
2991 				goto again;
2992 			}
2993 			rw_exit(&ip->i_rwlock);
2994 		}
2995 		ip = xip;
2996 		if (ip != NULL) {
2997 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2998 			rw_enter(&ip->i_contents, RW_WRITER);
2999 		}
3000 	}
3001 
3002 	/*
3003 	 * If the file already exists and this is a non-exclusive create,
3004 	 * check permissions and allow access for non-directories.
3005 	 * Read-only create of an existing directory is also allowed.
3006 	 * We fail an exclusive create of anything which already exists.
3007 	 */
3008 	if (error == EEXIST) {
3009 		dip = VTOI(dvp);
3010 		if (excl == NONEXCL) {
3011 			if ((((ip->i_mode & IFMT) == IFDIR) ||
3012 			    ((ip->i_mode & IFMT) == IFATTRDIR)) &&
3013 			    (mode & IWRITE))
3014 				error = EISDIR;
3015 			else if (mode)
3016 				error = ufs_iaccess(ip, mode, cr, 0);
3017 			else
3018 				error = 0;
3019 		}
3020 		if (error) {
3021 			rw_exit(&ip->i_contents);
3022 			rw_exit(&ufsvfsp->vfs_dqrwlock);
3023 			VN_RELE(ITOV(ip));
3024 			goto unlock;
3025 		}
3026 		/*
3027 		 * If the error EEXIST was set, then i_seq can not
3028 		 * have been updated. The sequence number interface
3029 		 * is defined such that a non-error VOP_CREATE must
3030 		 * increase the dir va_seq it by at least one. If we
3031 		 * have cleared the error, increase i_seq. Note that
3032 		 * we are increasing the dir i_seq and in rare cases
3033 		 * ip may actually be from the dvp, so we already have
3034 		 * the locks and it will not be subject to truncation.
3035 		 * In case we have to update i_seq of the parent
3036 		 * directory dip, we have to defer it till we have
3037 		 * released our locks on ip due to lock ordering requirements.
3038 		 */
3039 		if (ip != dip)
3040 			defer_dip_seq_update = 1;
3041 		else
3042 			ip->i_seq++;
3043 
3044 		if (((ip->i_mode & IFMT) == IFREG) &&
3045 		    (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
3046 			/*
3047 			 * Truncate regular files, if requested by caller.
3048 			 * Grab i_rwlock to make sure no one else is
3049 			 * currently writing to the file (we promised
3050 			 * bmap we would do this).
3051 			 * Must get the locks in the correct order.
3052 			 */
3053 			if (ip->i_size == 0) {
3054 				ip->i_flag |= ICHG | IUPD;
3055 				ip->i_seq++;
3056 				TRANS_INODE(ufsvfsp, ip);
3057 			} else {
3058 				/*
3059 				 * Large Files: Why this check here?
3060 				 * Though we do it in vn_create() we really
3061 				 * want to guarantee that we do not destroy
3062 				 * Large file data by atomically checking
3063 				 * the size while holding the contents
3064 				 * lock.
3065 				 */
3066 				if (flag && !(flag & FOFFMAX) &&
3067 				    ((ip->i_mode & IFMT) == IFREG) &&
3068 				    (ip->i_size > (offset_t)MAXOFF32_T)) {
3069 					rw_exit(&ip->i_contents);
3070 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3071 					error = EOVERFLOW;
3072 					goto unlock;
3073 				}
3074 				if (TRANS_ISTRANS(ufsvfsp))
3075 					truncflag++;
3076 				else {
3077 					rw_exit(&ip->i_contents);
3078 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3079 					ufs_tryirwlock_trans(&ip->i_rwlock,
3080 					    RW_WRITER, TOP_CREATE,
3081 					    retry_file);
3082 					if (indeadlock) {
3083 						VN_RELE(ITOV(ip));
3084 						goto again;
3085 					}
3086 					rw_enter(&ufsvfsp->vfs_dqrwlock,
3087 					    RW_READER);
3088 					rw_enter(&ip->i_contents, RW_WRITER);
3089 					(void) ufs_itrunc(ip, (u_offset_t)0, 0,
3090 					    cr);
3091 					rw_exit(&ip->i_rwlock);
3092 				}
3093 
3094 			}
3095 			if (error == 0) {
3096 				vnevent_create(ITOV(ip), ct);
3097 			}
3098 		}
3099 	}
3100 
3101 	if (error) {
3102 		if (ip != NULL) {
3103 			rw_exit(&ufsvfsp->vfs_dqrwlock);
3104 			rw_exit(&ip->i_contents);
3105 		}
3106 		goto unlock;
3107 	}
3108 
3109 	*vpp = ITOV(ip);
3110 	ITIMES(ip);
3111 	rw_exit(&ip->i_contents);
3112 	rw_exit(&ufsvfsp->vfs_dqrwlock);
3113 
3114 	/*
3115 	 * If vnode is a device return special vnode instead.
3116 	 */
3117 	if (!error && IS_DEVVP(*vpp)) {
3118 		struct vnode *newvp;
3119 
3120 		newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
3121 		VN_RELE(*vpp);
3122 		if (newvp == NULL) {
3123 			error = ENOSYS;
3124 			goto unlock;
3125 		}
3126 		truncflag = 0;
3127 		*vpp = newvp;
3128 	}
3129 unlock:
3130 
3131 	/*
3132 	 * Do the deferred update of the parent directory's sequence
3133 	 * number now.
3134 	 */
3135 	if (defer_dip_seq_update == 1) {
3136 		rw_enter(&dip->i_contents, RW_READER);
3137 		mutex_enter(&dip->i_tlock);
3138 		dip->i_seq++;
3139 		mutex_exit(&dip->i_tlock);
3140 		rw_exit(&dip->i_contents);
3141 	}
3142 
3143 	if (ulp) {
3144 		int terr = 0;
3145 
3146 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE,
3147 		    trans_size);
3148 
3149 		/*
3150 		 * If we haven't had a more interesting failure
3151 		 * already, then anything that might've happened
3152 		 * here should be reported.
3153 		 */
3154 		if (error == 0)
3155 			error = terr;
3156 	}
3157 
3158 	if (!error && truncflag) {
3159 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_trunc);
3160 		if (indeadlock) {
3161 			if (ulp)
3162 				ufs_lockfs_end(ulp);
3163 			VN_RELE(ITOV(ip));
3164 			goto again;
3165 		}
3166 		(void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr);
3167 		rw_exit(&ip->i_rwlock);
3168 	}
3169 
3170 	if (ulp)
3171 		ufs_lockfs_end(ulp);
3172 
3173 	/*
3174 	 * If no inodes available, try to free one up out of the
3175 	 * pending delete queue.
3176 	 */
3177 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3178 		ufs_delete_drain_wait(ufsvfsp, 1);
3179 		retry = 0;
3180 		goto again;
3181 	}
3182 
3183 out:
3184 	return (error);
3185 }
3186 
3187 extern int ufs_idle_max;
3188 /*ARGSUSED*/
3189 static int
3190 ufs_remove(struct vnode *vp, char *nm, struct cred *cr,
3191 	caller_context_t *ct, int flags)
3192 {
3193 	struct inode *ip = VTOI(vp);
3194 	struct ufsvfs *ufsvfsp	= ip->i_ufsvfs;
3195 	struct ulockfs *ulp;
3196 	vnode_t *rmvp = NULL;	/* Vnode corresponding to name being removed */
3197 	int indeadlock;
3198 	int error;
3199 	int issync;
3200 	int trans_size;
3201 
3202 	/*
3203 	 * don't let the delete queue get too long
3204 	 */
3205 	if (ufsvfsp == NULL) {
3206 		error = EIO;
3207 		goto out;
3208 	}
3209 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3210 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3211 
3212 retry_remove:
3213 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK);
3214 	if (error)
3215 		goto out;
3216 
3217 	if (ulp)
3218 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE,
3219 		    trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp)));
3220 
3221 	/*
3222 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3223 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3224 	 * possible, retries the operation.
3225 	 */
3226 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_REMOVE, retry);
3227 	if (indeadlock)
3228 		goto retry_remove;
3229 	error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0,
3230 	    DR_REMOVE, cr, &rmvp);
3231 	rw_exit(&ip->i_rwlock);
3232 
3233 	if (ulp) {
3234 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size);
3235 		ufs_lockfs_end(ulp);
3236 	}
3237 
3238 	/*
3239 	 * This must be called after the remove transaction is closed.
3240 	 */
3241 	if (rmvp != NULL) {
3242 		/* Only send the event if there were no errors */
3243 		if (error == 0)
3244 			vnevent_remove(rmvp, vp, nm, ct);
3245 		VN_RELE(rmvp);
3246 	}
3247 out:
3248 	return (error);
3249 }
3250 
3251 /*
3252  * Link a file or a directory.  Only privileged processes are allowed to
3253  * make links to directories.
3254  */
3255 /*ARGSUSED*/
3256 static int
3257 ufs_link(struct vnode *tdvp, struct vnode *svp, char *tnm, struct cred *cr,
3258 	caller_context_t *ct, int flags)
3259 {
3260 	struct inode *sip;
3261 	struct inode *tdp = VTOI(tdvp);
3262 	struct ufsvfs *ufsvfsp = tdp->i_ufsvfs;
3263 	struct ulockfs *ulp;
3264 	struct vnode *realvp;
3265 	int error;
3266 	int issync;
3267 	int trans_size;
3268 	int isdev;
3269 	int indeadlock;
3270 
3271 retry_link:
3272 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LINK_MASK);
3273 	if (error)
3274 		goto out;
3275 
3276 	if (ulp)
3277 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_LINK,
3278 		    trans_size = (int)TOP_LINK_SIZE(VTOI(tdvp)));
3279 
3280 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3281 		svp = realvp;
3282 
3283 	/*
3284 	 * Make sure link for extended attributes is valid
3285 	 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3286 	 *
3287 	 * Make certain we don't attempt to look at a device node as
3288 	 * a ufs inode.
3289 	 */
3290 
3291 	isdev = IS_DEVVP(svp);
3292 	if (((isdev == 0) && ((VTOI(svp)->i_cflags & IXATTR) == 0) &&
3293 	    ((tdp->i_mode & IFMT) == IFATTRDIR)) ||
3294 	    ((isdev == 0) && (VTOI(svp)->i_cflags & IXATTR) &&
3295 	    ((tdp->i_mode & IFMT) == IFDIR))) {
3296 		error = EINVAL;
3297 		goto unlock;
3298 	}
3299 
3300 	sip = VTOI(svp);
3301 	if ((svp->v_type == VDIR &&
3302 	    secpolicy_fs_linkdir(cr, ufsvfsp->vfs_vfs) != 0) ||
3303 	    (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)) {
3304 		error = EPERM;
3305 		goto unlock;
3306 	}
3307 
3308 	/*
3309 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3310 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3311 	 * possible, retries the operation.
3312 	 */
3313 	ufs_tryirwlock_trans(&tdp->i_rwlock, RW_WRITER, TOP_LINK, retry);
3314 	if (indeadlock)
3315 		goto retry_link;
3316 	error = ufs_direnter_lr(tdp, tnm, DE_LINK, (struct inode *)0,
3317 	    sip, cr, NULL);
3318 	rw_exit(&tdp->i_rwlock);
3319 
3320 unlock:
3321 	if (ulp) {
3322 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_LINK, trans_size);
3323 		ufs_lockfs_end(ulp);
3324 	}
3325 
3326 	if (!error) {
3327 		vnevent_link(svp, ct);
3328 	}
3329 out:
3330 	return (error);
3331 }
3332 
3333 uint64_t ufs_rename_retry_cnt;
3334 uint64_t ufs_rename_upgrade_retry_cnt;
3335 uint64_t ufs_rename_dircheck_retry_cnt;
3336 clock_t	 ufs_rename_backoff_delay = 1;
3337 
3338 /*
3339  * Rename a file or directory.
3340  * We are given the vnode and entry string of the source and the
3341  * vnode and entry string of the place we want to move the source
3342  * to (the target). The essential operation is:
3343  *	unlink(target);
3344  *	link(source, target);
3345  *	unlink(source);
3346  * but "atomically".  Can't do full commit without saving state in
3347  * the inode on disk, which isn't feasible at this time.  Best we
3348  * can do is always guarantee that the TARGET exists.
3349  */
3350 
3351 /*ARGSUSED*/
3352 static int
3353 ufs_rename(
3354 	struct vnode *sdvp,		/* old (source) parent vnode */
3355 	char *snm,			/* old (source) entry name */
3356 	struct vnode *tdvp,		/* new (target) parent vnode */
3357 	char *tnm,			/* new (target) entry name */
3358 	struct cred *cr,
3359 	caller_context_t *ct,
3360 	int flags)
3361 {
3362 	struct inode *sip = NULL;	/* source inode */
3363 	struct inode *ip = NULL;	/* check inode */
3364 	struct inode *sdp;		/* old (source) parent inode */
3365 	struct inode *tdp;		/* new (target) parent inode */
3366 	struct vnode *tvp = NULL;	/* target vnode, if it exists */
3367 	struct vnode *realvp;
3368 	struct ufsvfs *ufsvfsp;
3369 	struct ulockfs *ulp;
3370 	struct ufs_slot slot;
3371 	timestruc_t now;
3372 	int error;
3373 	int issync;
3374 	int trans_size;
3375 	krwlock_t *first_lock;
3376 	krwlock_t *second_lock;
3377 	krwlock_t *reverse_lock;
3378 
3379 	sdp = VTOI(sdvp);
3380 	slot.fbp = NULL;
3381 	ufsvfsp = sdp->i_ufsvfs;
3382 retry_rename:
3383 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
3384 	if (error)
3385 		goto out;
3386 
3387 	if (ulp)
3388 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
3389 		    trans_size = (int)TOP_RENAME_SIZE(sdp));
3390 
3391 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3392 		tdvp = realvp;
3393 
3394 	tdp = VTOI(tdvp);
3395 
3396 	/*
3397 	 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3398 	 */
3399 	if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) {
3400 		error = EINVAL;
3401 		goto unlock;
3402 	}
3403 
3404 	/*
3405 	 * Check accessibility of directory.
3406 	 */
3407 	if (error = ufs_diraccess(sdp, IEXEC, cr))
3408 		goto unlock;
3409 
3410 	/*
3411 	 * Look up inode of file we're supposed to rename.
3412 	 */
3413 	gethrestime(&now);
3414 	if (error = ufs_dirlook(sdp, snm, &sip, cr, 0)) {
3415 		if (error == EAGAIN) {
3416 			if (ulp) {
3417 				TRANS_END_CSYNC(ufsvfsp, error, issync,
3418 				    TOP_RENAME, trans_size);
3419 				ufs_lockfs_end(ulp);
3420 			}
3421 			goto retry_rename;
3422 		}
3423 
3424 		goto unlock;
3425 	}
3426 
3427 	/*
3428 	 * Lock both the source and target directories (they may be
3429 	 * the same) to provide the atomicity semantics that was
3430 	 * previously provided by the per file system vfs_rename_lock
3431 	 *
3432 	 * with vfs_rename_lock removed to allow simultaneous renames
3433 	 * within a file system, ufs_dircheckpath can deadlock while
3434 	 * traversing back to ensure that source is not a parent directory
3435 	 * of target parent directory. This is because we get into
3436 	 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3437 	 * If the tdp and sdp of the simultaneous renames happen to be
3438 	 * in the path of each other, it can lead to a deadlock. This
3439 	 * can be avoided by getting the locks as RW_READER here and then
3440 	 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3441 	 *
3442 	 * We hold the target directory's i_rwlock after calling
3443 	 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3444 	 * VOP_RWLOCK is explicitly called by the filesystem independent code
3445 	 * before calling the file system operation. In these cases the order
3446 	 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3447 	 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3448 	 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3449 	 * synchronizing object which might lead to a deadlock. So we use
3450 	 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3451 	 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3452 	 * operation.
3453 	 */
3454 retry:
3455 	first_lock = &tdp->i_rwlock;
3456 	second_lock = &sdp->i_rwlock;
3457 retry_firstlock:
3458 	if (!rw_tryenter(first_lock, RW_READER)) {
3459 		/*
3460 		 * We didn't get the lock. Check if the SLOCK is set in the
3461 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3462 		 * and wait for SLOCK to be cleared.
3463 		 */
3464 
3465 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3466 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3467 			    trans_size);
3468 			ufs_lockfs_end(ulp);
3469 			goto retry_rename;
3470 
3471 		} else {
3472 			/*
3473 			 * SLOCK isn't set so this is a genuine synchronization
3474 			 * case. Let's try again after giving them a breather.
3475 			 */
3476 			delay(RETRY_LOCK_DELAY);
3477 			goto  retry_firstlock;
3478 		}
3479 	}
3480 	/*
3481 	 * Need to check if the tdp and sdp are same !!!
3482 	 */
3483 	if ((tdp != sdp) && (!rw_tryenter(second_lock, RW_READER))) {
3484 		/*
3485 		 * We didn't get the lock. Check if the SLOCK is set in the
3486 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3487 		 * and wait for SLOCK to be cleared.
3488 		 */
3489 
3490 		rw_exit(first_lock);
3491 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3492 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3493 			    trans_size);
3494 			ufs_lockfs_end(ulp);
3495 			goto retry_rename;
3496 
3497 		} else {
3498 			/*
3499 			 * So we couldn't get the second level peer lock *and*
3500 			 * the SLOCK bit isn't set. Too bad we can be
3501 			 * contentding with someone wanting these locks otherway
3502 			 * round. Reverse the locks in case there is a heavy
3503 			 * contention for the second level lock.
3504 			 */
3505 			reverse_lock = first_lock;
3506 			first_lock = second_lock;
3507 			second_lock = reverse_lock;
3508 			ufs_rename_retry_cnt++;
3509 			goto  retry_firstlock;
3510 		}
3511 	}
3512 
3513 	if (sip == tdp) {
3514 		error = EINVAL;
3515 		goto errout;
3516 	}
3517 	/*
3518 	 * Make sure we can delete the source entry.  This requires
3519 	 * write permission on the containing directory.
3520 	 * Check for sticky directories.
3521 	 */
3522 	rw_enter(&sdp->i_contents, RW_READER);
3523 	rw_enter(&sip->i_contents, RW_READER);
3524 	if ((error = ufs_iaccess(sdp, IWRITE, cr, 0)) != 0 ||
3525 	    (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) {
3526 		rw_exit(&sip->i_contents);
3527 		rw_exit(&sdp->i_contents);
3528 		goto errout;
3529 	}
3530 
3531 	/*
3532 	 * If this is a rename of a directory and the parent is
3533 	 * different (".." must be changed), then the source
3534 	 * directory must not be in the directory hierarchy
3535 	 * above the target, as this would orphan everything
3536 	 * below the source directory.  Also the user must have
3537 	 * write permission in the source so as to be able to
3538 	 * change "..".
3539 	 */
3540 	if ((((sip->i_mode & IFMT) == IFDIR) ||
3541 	    ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) {
3542 		ino_t	inum;
3543 
3544 		if (error = ufs_iaccess(sip, IWRITE, cr, 0)) {
3545 			rw_exit(&sip->i_contents);
3546 			rw_exit(&sdp->i_contents);
3547 			goto errout;
3548 		}
3549 		inum = sip->i_number;
3550 		rw_exit(&sip->i_contents);
3551 		rw_exit(&sdp->i_contents);
3552 		if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) {
3553 			/*
3554 			 * If we got EAGAIN ufs_dircheckpath detected a
3555 			 * potential deadlock and backed out. We need
3556 			 * to retry the operation since sdp and tdp have
3557 			 * to be released to avoid the deadlock.
3558 			 */
3559 			if (error == EAGAIN) {
3560 				rw_exit(&tdp->i_rwlock);
3561 				if (tdp != sdp)
3562 					rw_exit(&sdp->i_rwlock);
3563 				delay(ufs_rename_backoff_delay);
3564 				ufs_rename_dircheck_retry_cnt++;
3565 				goto retry;
3566 			}
3567 			goto errout;
3568 		}
3569 	} else {
3570 		rw_exit(&sip->i_contents);
3571 		rw_exit(&sdp->i_contents);
3572 	}
3573 
3574 
3575 	/*
3576 	 * Check for renaming '.' or '..' or alias of '.'
3577 	 */
3578 	if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) {
3579 		error = EINVAL;
3580 		goto errout;
3581 	}
3582 
3583 	/*
3584 	 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3585 	 * tries to traverse back the file tree with both tdp and sdp held
3586 	 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3587 	 * as RW_READERS  till ufs_dircheckpath is done.
3588 	 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3589 	 * to RW_WRITER.
3590 	 */
3591 	if (!rw_tryupgrade(&tdp->i_rwlock)) {
3592 		/*
3593 		 * The upgrade failed. We got to give away the lock
3594 		 * as to avoid deadlocking with someone else who is
3595 		 * waiting for writer lock. With the lock gone, we
3596 		 * cannot be sure the checks done above will hold
3597 		 * good when we eventually get them back as writer.
3598 		 * So if we can't upgrade we drop the locks and retry
3599 		 * everything again.
3600 		 */
3601 		rw_exit(&tdp->i_rwlock);
3602 		if (tdp != sdp)
3603 			rw_exit(&sdp->i_rwlock);
3604 		delay(ufs_rename_backoff_delay);
3605 		ufs_rename_upgrade_retry_cnt++;
3606 		goto retry;
3607 	}
3608 	if (tdp != sdp) {
3609 		if (!rw_tryupgrade(&sdp->i_rwlock)) {
3610 			/*
3611 			 * The upgrade failed. We got to give away the lock
3612 			 * as to avoid deadlocking with someone else who is
3613 			 * waiting for writer lock. With the lock gone, we
3614 			 * cannot be sure the checks done above will hold
3615 			 * good when we eventually get them back as writer.
3616 			 * So if we can't upgrade we drop the locks and retry
3617 			 * everything again.
3618 			 */
3619 			rw_exit(&tdp->i_rwlock);
3620 			rw_exit(&sdp->i_rwlock);
3621 			delay(ufs_rename_backoff_delay);
3622 			ufs_rename_upgrade_retry_cnt++;
3623 			goto retry;
3624 		}
3625 	}
3626 
3627 	/*
3628 	 * Now that all the locks are held check to make sure another thread
3629 	 * didn't slip in and take out the sip.
3630 	 */
3631 	slot.status = NONE;
3632 	if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec ||
3633 	    sip->i_ctime.tv_sec > now.tv_sec) {
3634 		rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER);
3635 		rw_enter(&sdp->i_contents, RW_WRITER);
3636 		error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot,
3637 		    &ip, cr, 0);
3638 		rw_exit(&sdp->i_contents);
3639 		rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock);
3640 		if (error) {
3641 			goto errout;
3642 		}
3643 		if (ip == NULL) {
3644 			error = ENOENT;
3645 			goto errout;
3646 		} else {
3647 			/*
3648 			 * If the inode was found need to drop the v_count
3649 			 * so as not to keep the filesystem from being
3650 			 * unmounted at a later time.
3651 			 */
3652 			VN_RELE(ITOV(ip));
3653 		}
3654 
3655 		/*
3656 		 * Release the slot.fbp that has the page mapped and
3657 		 * locked SE_SHARED, and could be used in in
3658 		 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3659 		 * on said page.
3660 		 */
3661 		if (slot.fbp) {
3662 			fbrelse(slot.fbp, S_OTHER);
3663 			slot.fbp = NULL;
3664 		}
3665 	}
3666 
3667 	/*
3668 	 * Link source to the target.  If a target exists, return its
3669 	 * vnode pointer in tvp.  We'll release it after sending the
3670 	 * vnevent.
3671 	 */
3672 	if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr, &tvp)) {
3673 		/*
3674 		 * ESAME isn't really an error; it indicates that the
3675 		 * operation should not be done because the source and target
3676 		 * are the same file, but that no error should be reported.
3677 		 */
3678 		if (error == ESAME)
3679 			error = 0;
3680 		goto errout;
3681 	}
3682 
3683 	/*
3684 	 * Unlink the source.
3685 	 * Remove the source entry.  ufs_dirremove() checks that the entry
3686 	 * still reflects sip, and returns an error if it doesn't.
3687 	 * If the entry has changed just forget about it.  Release
3688 	 * the source inode.
3689 	 */
3690 	if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0,
3691 	    DR_RENAME, cr, NULL)) == ENOENT)
3692 		error = 0;
3693 
3694 errout:
3695 	if (slot.fbp)
3696 		fbrelse(slot.fbp, S_OTHER);
3697 
3698 	rw_exit(&tdp->i_rwlock);
3699 	if (sdp != tdp) {
3700 		rw_exit(&sdp->i_rwlock);
3701 	}
3702 
3703 unlock:
3704 	if (ulp) {
3705 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
3706 		ufs_lockfs_end(ulp);
3707 	}
3708 
3709 	/*
3710 	 * If no errors, send the appropriate events on the source
3711 	 * and destination (a.k.a, target) vnodes, if they exist.
3712 	 * This has to be done after the rename transaction has closed.
3713 	 */
3714 	if (error == 0) {
3715 		if (tvp != NULL)
3716 			vnevent_rename_dest(tvp, tdvp, tnm, ct);
3717 
3718 		/*
3719 		 * Notify the target directory of the rename event
3720 		 * if source and target directories are not same.
3721 		 */
3722 		if (sdvp != tdvp)
3723 			vnevent_rename_dest_dir(tdvp, ct);
3724 
3725 		/*
3726 		 * Note that if ufs_direnter_lr() returned ESAME then
3727 		 * this event will still be sent.  This isn't expected
3728 		 * to be a problem for anticipated usage by consumers.
3729 		 */
3730 		if (sip != NULL)
3731 			vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
3732 	}
3733 
3734 	if (tvp != NULL)
3735 		VN_RELE(tvp);
3736 
3737 	if (sip != NULL)
3738 		VN_RELE(ITOV(sip));
3739 
3740 out:
3741 	return (error);
3742 }
3743 
3744 /*ARGSUSED*/
3745 static int
3746 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap,
3747 	struct vnode **vpp, struct cred *cr, caller_context_t *ct, int flags,
3748 	vsecattr_t *vsecp)
3749 {
3750 	struct inode *ip;
3751 	struct inode *xip;
3752 	struct ufsvfs *ufsvfsp;
3753 	struct ulockfs *ulp;
3754 	int error;
3755 	int issync;
3756 	int trans_size;
3757 	int indeadlock;
3758 	int retry = 1;
3759 
3760 	ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
3761 
3762 	/*
3763 	 * Can't make directory in attr hidden dir
3764 	 */
3765 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3766 		return (EINVAL);
3767 
3768 again:
3769 	ip = VTOI(dvp);
3770 	ufsvfsp = ip->i_ufsvfs;
3771 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK);
3772 	if (error)
3773 		goto out;
3774 	if (ulp)
3775 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR,
3776 		    trans_size = (int)TOP_MKDIR_SIZE(ip));
3777 
3778 	/*
3779 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3780 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3781 	 * possible, retries the operation.
3782 	 */
3783 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_MKDIR, retry);
3784 	if (indeadlock)
3785 		goto again;
3786 
3787 	error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr,
3788 	    (retry ? IQUIET : 0));
3789 	if (error == EAGAIN) {
3790 		if (ulp) {
3791 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_MKDIR,
3792 			    trans_size);
3793 			ufs_lockfs_end(ulp);
3794 		}
3795 		goto again;
3796 	}
3797 
3798 	rw_exit(&ip->i_rwlock);
3799 	if (error == 0) {
3800 		ip = xip;
3801 		*vpp = ITOV(ip);
3802 	} else if (error == EEXIST)
3803 		VN_RELE(ITOV(xip));
3804 
3805 	if (ulp) {
3806 		int terr = 0;
3807 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size);
3808 		ufs_lockfs_end(ulp);
3809 		if (error == 0)
3810 			error = terr;
3811 	}
3812 out:
3813 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3814 		ufs_delete_drain_wait(ufsvfsp, 1);
3815 		retry = 0;
3816 		goto again;
3817 	}
3818 
3819 	return (error);
3820 }
3821 
3822 /*ARGSUSED*/
3823 static int
3824 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr,
3825 	caller_context_t *ct, int flags)
3826 {
3827 	struct inode *ip = VTOI(vp);
3828 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3829 	struct ulockfs *ulp;
3830 	vnode_t *rmvp = NULL;	/* Vnode of removed directory */
3831 	int error;
3832 	int issync;
3833 	int trans_size;
3834 	int indeadlock;
3835 
3836 	/*
3837 	 * don't let the delete queue get too long
3838 	 */
3839 	if (ufsvfsp == NULL) {
3840 		error = EIO;
3841 		goto out;
3842 	}
3843 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3844 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3845 
3846 retry_rmdir:
3847 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK);
3848 	if (error)
3849 		goto out;
3850 
3851 	if (ulp)
3852 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR,
3853 		    trans_size = TOP_RMDIR_SIZE);
3854 
3855 	/*
3856 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3857 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3858 	 * possible, retries the operation.
3859 	 */
3860 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_RMDIR, retry);
3861 	if (indeadlock)
3862 		goto retry_rmdir;
3863 	error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr,
3864 	    &rmvp);
3865 	rw_exit(&ip->i_rwlock);
3866 
3867 	if (ulp) {
3868 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR,
3869 		    trans_size);
3870 		ufs_lockfs_end(ulp);
3871 	}
3872 
3873 	/*
3874 	 * This must be done AFTER the rmdir transaction has closed.
3875 	 */
3876 	if (rmvp != NULL) {
3877 		/* Only send the event if there were no errors */
3878 		if (error == 0)
3879 			vnevent_rmdir(rmvp, vp, nm, ct);
3880 		VN_RELE(rmvp);
3881 	}
3882 out:
3883 	return (error);
3884 }
3885 
3886 /* ARGSUSED */
3887 static int
3888 ufs_readdir(
3889 	struct vnode *vp,
3890 	struct uio *uiop,
3891 	struct cred *cr,
3892 	int *eofp,
3893 	caller_context_t *ct,
3894 	int flags)
3895 {
3896 	struct iovec *iovp;
3897 	struct inode *ip;
3898 	struct direct *idp;
3899 	struct dirent64 *odp;
3900 	struct fbuf *fbp;
3901 	struct ufsvfs *ufsvfsp;
3902 	struct ulockfs *ulp;
3903 	caddr_t outbuf;
3904 	size_t bufsize;
3905 	uint_t offset;
3906 	uint_t bytes_wanted, total_bytes_wanted;
3907 	int incount = 0;
3908 	int outcount = 0;
3909 	int error;
3910 
3911 	ip = VTOI(vp);
3912 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
3913 
3914 	if (uiop->uio_loffset >= MAXOFF32_T) {
3915 		if (eofp)
3916 			*eofp = 1;
3917 		return (0);
3918 	}
3919 
3920 	/*
3921 	 * Check if we have been called with a valid iov_len
3922 	 * and bail out if not, otherwise we may potentially loop
3923 	 * forever further down.
3924 	 */
3925 	if (uiop->uio_iov->iov_len <= 0) {
3926 		error = EINVAL;
3927 		goto out;
3928 	}
3929 
3930 	/*
3931 	 * Large Files: When we come here we are guaranteed that
3932 	 * uio_offset can be used safely. The high word is zero.
3933 	 */
3934 
3935 	ufsvfsp = ip->i_ufsvfs;
3936 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READDIR_MASK);
3937 	if (error)
3938 		goto out;
3939 
3940 	iovp = uiop->uio_iov;
3941 	total_bytes_wanted = iovp->iov_len;
3942 
3943 	/* Large Files: directory files should not be "large" */
3944 
3945 	ASSERT(ip->i_size <= MAXOFF32_T);
3946 
3947 	/* Force offset to be valid (to guard against bogus lseek() values) */
3948 	offset = (uint_t)uiop->uio_offset & ~(DIRBLKSIZ - 1);
3949 
3950 	/* Quit if at end of file or link count of zero (posix) */
3951 	if (offset >= (uint_t)ip->i_size || ip->i_nlink <= 0) {
3952 		if (eofp)
3953 			*eofp = 1;
3954 		error = 0;
3955 		goto unlock;
3956 	}
3957 
3958 	/*
3959 	 * Get space to change directory entries into fs independent format.
3960 	 * Do fast alloc for the most commonly used-request size (filesystem
3961 	 * block size).
3962 	 */
3963 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) {
3964 		bufsize = total_bytes_wanted;
3965 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
3966 		odp = (struct dirent64 *)outbuf;
3967 	} else {
3968 		bufsize = total_bytes_wanted;
3969 		odp = (struct dirent64 *)iovp->iov_base;
3970 	}
3971 
3972 nextblk:
3973 	bytes_wanted = total_bytes_wanted;
3974 
3975 	/* Truncate request to file size */
3976 	if (offset + bytes_wanted > (int)ip->i_size)
3977 		bytes_wanted = (int)(ip->i_size - offset);
3978 
3979 	/* Comply with MAXBSIZE boundary restrictions of fbread() */
3980 	if ((offset & MAXBOFFSET) + bytes_wanted > MAXBSIZE)
3981 		bytes_wanted = MAXBSIZE - (offset & MAXBOFFSET);
3982 
3983 	/*
3984 	 * Read in the next chunk.
3985 	 * We are still holding the i_rwlock.
3986 	 */
3987 	error = fbread(vp, (offset_t)offset, bytes_wanted, S_OTHER, &fbp);
3988 
3989 	if (error)
3990 		goto update_inode;
3991 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (ip->i_fs->fs_ronly == 0) &&
3992 	    (!ufsvfsp->vfs_noatime)) {
3993 		ip->i_flag |= IACC;
3994 	}
3995 	incount = 0;
3996 	idp = (struct direct *)fbp->fb_addr;
3997 	if (idp->d_ino == 0 && idp->d_reclen == 0 && idp->d_namlen == 0) {
3998 		cmn_err(CE_WARN, "ufs_readdir: bad dir, inumber = %llu, "
3999 		    "fs = %s\n",
4000 		    (u_longlong_t)ip->i_number, ufsvfsp->vfs_fs->fs_fsmnt);
4001 		fbrelse(fbp, S_OTHER);
4002 		error = ENXIO;
4003 		goto update_inode;
4004 	}
4005 	/* Transform to file-system independent format */
4006 	while (incount < bytes_wanted) {
4007 		/*
4008 		 * If the current directory entry is mangled, then skip
4009 		 * to the next block.  It would be nice to set the FSBAD
4010 		 * flag in the super-block so that a fsck is forced on
4011 		 * next reboot, but locking is a problem.
4012 		 */
4013 		if (idp->d_reclen & 0x3) {
4014 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4015 			break;
4016 		}
4017 
4018 		/* Skip to requested offset and skip empty entries */
4019 		if (idp->d_ino != 0 && offset >= (uint_t)uiop->uio_offset) {
4020 			ushort_t this_reclen =
4021 			    DIRENT64_RECLEN(idp->d_namlen);
4022 			/* Buffer too small for any entries */
4023 			if (!outcount && this_reclen > bufsize) {
4024 				fbrelse(fbp, S_OTHER);
4025 				error = EINVAL;
4026 				goto update_inode;
4027 			}
4028 			/* If would overrun the buffer, quit */
4029 			if (outcount + this_reclen > bufsize) {
4030 				break;
4031 			}
4032 			/* Take this entry */
4033 			odp->d_ino = (ino64_t)idp->d_ino;
4034 			odp->d_reclen = (ushort_t)this_reclen;
4035 			odp->d_off = (offset_t)(offset + idp->d_reclen);
4036 
4037 			/* use strncpy(9f) to zero out uninitialized bytes */
4038 
4039 			ASSERT(strlen(idp->d_name) + 1 <=
4040 			    DIRENT64_NAMELEN(this_reclen));
4041 			(void) strncpy(odp->d_name, idp->d_name,
4042 			    DIRENT64_NAMELEN(this_reclen));
4043 			outcount += odp->d_reclen;
4044 			odp = (struct dirent64 *)
4045 			    ((intptr_t)odp + odp->d_reclen);
4046 			ASSERT(outcount <= bufsize);
4047 		}
4048 		if (idp->d_reclen) {
4049 			incount += idp->d_reclen;
4050 			offset += idp->d_reclen;
4051 			idp = (struct direct *)((intptr_t)idp + idp->d_reclen);
4052 		} else {
4053 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4054 			break;
4055 		}
4056 	}
4057 	/* Release the chunk */
4058 	fbrelse(fbp, S_OTHER);
4059 
4060 	/* Read whole block, but got no entries, read another if not eof */
4061 
4062 	/*
4063 	 * Large Files: casting i_size to int here is not a problem
4064 	 * because directory sizes are always less than MAXOFF32_T.
4065 	 * See assertion above.
4066 	 */
4067 
4068 	if (offset < (int)ip->i_size && !outcount)
4069 		goto nextblk;
4070 
4071 	/* Copy out the entry data */
4072 	if (uiop->uio_segflg == UIO_SYSSPACE && uiop->uio_iovcnt == 1) {
4073 		iovp->iov_base += outcount;
4074 		iovp->iov_len -= outcount;
4075 		uiop->uio_resid -= outcount;
4076 		uiop->uio_offset = offset;
4077 	} else if ((error = uiomove(outbuf, (long)outcount, UIO_READ,
4078 	    uiop)) == 0)
4079 		uiop->uio_offset = offset;
4080 update_inode:
4081 	ITIMES(ip);
4082 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1)
4083 		kmem_free(outbuf, bufsize);
4084 
4085 	if (eofp && error == 0)
4086 		*eofp = (uiop->uio_offset >= (int)ip->i_size);
4087 unlock:
4088 	if (ulp) {
4089 		ufs_lockfs_end(ulp);
4090 	}
4091 out:
4092 	return (error);
4093 }
4094 
4095 /*ARGSUSED*/
4096 static int
4097 ufs_symlink(
4098 	struct vnode *dvp,		/* ptr to parent dir vnode */
4099 	char *linkname,			/* name of symbolic link */
4100 	struct vattr *vap,		/* attributes */
4101 	char *target,			/* target path */
4102 	struct cred *cr,		/* user credentials */
4103 	caller_context_t *ct,
4104 	int flags)
4105 {
4106 	struct inode *ip, *dip = VTOI(dvp);
4107 	struct ufsvfs *ufsvfsp = dip->i_ufsvfs;
4108 	struct ulockfs *ulp;
4109 	int error;
4110 	int issync;
4111 	int trans_size;
4112 	int residual;
4113 	int ioflag;
4114 	int retry = 1;
4115 
4116 	/*
4117 	 * No symlinks in attrdirs at this time
4118 	 */
4119 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
4120 		return (EINVAL);
4121 
4122 again:
4123 	ip = (struct inode *)NULL;
4124 	vap->va_type = VLNK;
4125 	vap->va_rdev = 0;
4126 
4127 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK);
4128 	if (error)
4129 		goto out;
4130 
4131 	if (ulp)
4132 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK,
4133 		    trans_size = (int)TOP_SYMLINK_SIZE(dip));
4134 
4135 	/*
4136 	 * We must create the inode before the directory entry, to avoid
4137 	 * racing with readlink().  ufs_dirmakeinode requires that we
4138 	 * hold the quota lock as reader, and directory locks as writer.
4139 	 */
4140 
4141 	rw_enter(&dip->i_rwlock, RW_WRITER);
4142 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4143 	rw_enter(&dip->i_contents, RW_WRITER);
4144 
4145 	/*
4146 	 * Suppress any out of inodes messages if we will retry on
4147 	 * ENOSP
4148 	 */
4149 	if (retry)
4150 		dip->i_flag |= IQUIET;
4151 
4152 	error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr);
4153 
4154 	dip->i_flag &= ~IQUIET;
4155 
4156 	rw_exit(&dip->i_contents);
4157 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4158 	rw_exit(&dip->i_rwlock);
4159 
4160 	if (error)
4161 		goto unlock;
4162 
4163 	/*
4164 	 * OK.  The inode has been created.  Write out the data of the
4165 	 * symbolic link.  Since symbolic links are metadata, and should
4166 	 * remain consistent across a system crash, we need to force the
4167 	 * data out synchronously.
4168 	 *
4169 	 * (This is a change from the semantics in earlier releases, which
4170 	 * only created symbolic links synchronously if the semi-documented
4171 	 * 'syncdir' option was set, or if we were being invoked by the NFS
4172 	 * server, which requires symbolic links to be created synchronously.)
4173 	 *
4174 	 * We need to pass in a pointer for the residual length; otherwise
4175 	 * ufs_rdwri() will always return EIO if it can't write the data,
4176 	 * even if the error was really ENOSPC or EDQUOT.
4177 	 */
4178 
4179 	ioflag = FWRITE | FDSYNC;
4180 	residual = 0;
4181 
4182 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4183 	rw_enter(&ip->i_contents, RW_WRITER);
4184 
4185 	/*
4186 	 * Suppress file system full messages if we will retry
4187 	 */
4188 	if (retry)
4189 		ip->i_flag |= IQUIET;
4190 
4191 	error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target),
4192 	    (offset_t)0, UIO_SYSSPACE, &residual, cr);
4193 
4194 	ip->i_flag &= ~IQUIET;
4195 
4196 	if (error) {
4197 		rw_exit(&ip->i_contents);
4198 		rw_exit(&ufsvfsp->vfs_dqrwlock);
4199 		goto remove;
4200 	}
4201 
4202 	/*
4203 	 * If the link's data is small enough, we can cache it in the inode.
4204 	 * This is a "fast symbolic link".  We don't use the first direct
4205 	 * block because that's actually used to point at the symbolic link's
4206 	 * contents on disk; but we know that none of the other direct or
4207 	 * indirect blocks can be used because symbolic links are restricted
4208 	 * to be smaller than a file system block.
4209 	 */
4210 
4211 	ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip)));
4212 
4213 	if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) {
4214 		if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) {
4215 			ip->i_flag |= IFASTSYMLNK;
4216 		} else {
4217 			int i;
4218 			/* error, clear garbage left behind */
4219 			for (i = 1; i < NDADDR; i++)
4220 				ip->i_db[i] = 0;
4221 			for (i = 0; i < NIADDR; i++)
4222 				ip->i_ib[i] = 0;
4223 		}
4224 	}
4225 
4226 	rw_exit(&ip->i_contents);
4227 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4228 
4229 	/*
4230 	 * OK.  We've successfully created the symbolic link.  All that
4231 	 * remains is to insert it into the appropriate directory.
4232 	 */
4233 
4234 	rw_enter(&dip->i_rwlock, RW_WRITER);
4235 	error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr, NULL);
4236 	rw_exit(&dip->i_rwlock);
4237 
4238 	/*
4239 	 * Fall through into remove-on-error code.  We're either done, or we
4240 	 * need to remove the inode (if we couldn't insert it).
4241 	 */
4242 
4243 remove:
4244 	if (error && (ip != NULL)) {
4245 		rw_enter(&ip->i_contents, RW_WRITER);
4246 		ip->i_nlink--;
4247 		ip->i_flag |= ICHG;
4248 		ip->i_seq++;
4249 		ufs_setreclaim(ip);
4250 		rw_exit(&ip->i_contents);
4251 	}
4252 
4253 unlock:
4254 	if (ip != NULL)
4255 		VN_RELE(ITOV(ip));
4256 
4257 	if (ulp) {
4258 		int terr = 0;
4259 
4260 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK,
4261 		    trans_size);
4262 		ufs_lockfs_end(ulp);
4263 		if (error == 0)
4264 			error = terr;
4265 	}
4266 
4267 	/*
4268 	 * We may have failed due to lack of an inode or of a block to
4269 	 * store the target in.  Try flushing the delete queue to free
4270 	 * logically-available things up and try again.
4271 	 */
4272 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
4273 		ufs_delete_drain_wait(ufsvfsp, 1);
4274 		retry = 0;
4275 		goto again;
4276 	}
4277 
4278 out:
4279 	return (error);
4280 }
4281 
4282 /*
4283  * Ufs specific routine used to do ufs io.
4284  */
4285 int
4286 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base,
4287 	ssize_t len, offset_t offset, enum uio_seg seg, int *aresid,
4288 	struct cred *cr)
4289 {
4290 	struct uio auio;
4291 	struct iovec aiov;
4292 	int error;
4293 
4294 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
4295 
4296 	bzero((caddr_t)&auio, sizeof (uio_t));
4297 	bzero((caddr_t)&aiov, sizeof (iovec_t));
4298 
4299 	aiov.iov_base = base;
4300 	aiov.iov_len = len;
4301 	auio.uio_iov = &aiov;
4302 	auio.uio_iovcnt = 1;
4303 	auio.uio_loffset = offset;
4304 	auio.uio_segflg = (short)seg;
4305 	auio.uio_resid = len;
4306 
4307 	if (rw == UIO_WRITE) {
4308 		auio.uio_fmode = FWRITE;
4309 		auio.uio_extflg = UIO_COPY_DEFAULT;
4310 		auio.uio_llimit = curproc->p_fsz_ctl;
4311 		error = wrip(ip, &auio, ioflag, cr);
4312 	} else {
4313 		auio.uio_fmode = FREAD;
4314 		auio.uio_extflg = UIO_COPY_CACHED;
4315 		auio.uio_llimit = MAXOFFSET_T;
4316 		error = rdip(ip, &auio, ioflag, cr);
4317 	}
4318 
4319 	if (aresid) {
4320 		*aresid = auio.uio_resid;
4321 	} else if (auio.uio_resid) {
4322 		error = EIO;
4323 	}
4324 	return (error);
4325 }
4326 
4327 /*ARGSUSED*/
4328 static int
4329 ufs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
4330 {
4331 	struct ufid *ufid;
4332 	struct inode *ip = VTOI(vp);
4333 
4334 	if (ip->i_ufsvfs == NULL)
4335 		return (EIO);
4336 
4337 	if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) {
4338 		fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t);
4339 		return (ENOSPC);
4340 	}
4341 
4342 	ufid = (struct ufid *)fidp;
4343 	bzero((char *)ufid, sizeof (struct ufid));
4344 	ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t);
4345 	ufid->ufid_ino = ip->i_number;
4346 	ufid->ufid_gen = ip->i_gen;
4347 
4348 	return (0);
4349 }
4350 
4351 /* ARGSUSED2 */
4352 static int
4353 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4354 {
4355 	struct inode	*ip = VTOI(vp);
4356 	struct ufsvfs	*ufsvfsp;
4357 	int		forcedirectio;
4358 
4359 	/*
4360 	 * Read case is easy.
4361 	 */
4362 	if (!write_lock) {
4363 		rw_enter(&ip->i_rwlock, RW_READER);
4364 		return (V_WRITELOCK_FALSE);
4365 	}
4366 
4367 	/*
4368 	 * Caller has requested a writer lock, but that inhibits any
4369 	 * concurrency in the VOPs that follow. Acquire the lock shared
4370 	 * and defer exclusive access until it is known to be needed in
4371 	 * other VOP handlers. Some cases can be determined here.
4372 	 */
4373 
4374 	/*
4375 	 * If directio is not set, there is no chance of concurrency,
4376 	 * so just acquire the lock exclusive. Beware of a forced
4377 	 * unmount before looking at the mount option.
4378 	 */
4379 	ufsvfsp = ip->i_ufsvfs;
4380 	forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0;
4381 	if (!(ip->i_flag & IDIRECTIO || forcedirectio) ||
4382 	    !ufs_allow_shared_writes) {
4383 		rw_enter(&ip->i_rwlock, RW_WRITER);
4384 		return (V_WRITELOCK_TRUE);
4385 	}
4386 
4387 	/*
4388 	 * Mandatory locking forces acquiring i_rwlock exclusive.
4389 	 */
4390 	if (MANDLOCK(vp, ip->i_mode)) {
4391 		rw_enter(&ip->i_rwlock, RW_WRITER);
4392 		return (V_WRITELOCK_TRUE);
4393 	}
4394 
4395 	/*
4396 	 * Acquire the lock shared in case a concurrent write follows.
4397 	 * Mandatory locking could have become enabled before the lock
4398 	 * was acquired. Re-check and upgrade if needed.
4399 	 */
4400 	rw_enter(&ip->i_rwlock, RW_READER);
4401 	if (MANDLOCK(vp, ip->i_mode)) {
4402 		rw_exit(&ip->i_rwlock);
4403 		rw_enter(&ip->i_rwlock, RW_WRITER);
4404 		return (V_WRITELOCK_TRUE);
4405 	}
4406 	return (V_WRITELOCK_FALSE);
4407 }
4408 
4409 /*ARGSUSED*/
4410 static void
4411 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4412 {
4413 	struct inode	*ip = VTOI(vp);
4414 
4415 	rw_exit(&ip->i_rwlock);
4416 }
4417 
4418 /* ARGSUSED */
4419 static int
4420 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp,
4421 	caller_context_t *ct)
4422 {
4423 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4424 }
4425 
4426 /* ARGSUSED */
4427 static int
4428 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4429 	offset_t offset, struct flk_callback *flk_cbp, struct cred *cr,
4430 	caller_context_t *ct)
4431 {
4432 	struct inode *ip = VTOI(vp);
4433 
4434 	if (ip->i_ufsvfs == NULL)
4435 		return (EIO);
4436 
4437 	/*
4438 	 * If file is being mapped, disallow frlock.
4439 	 * XXX I am not holding tlock while checking i_mapcnt because the
4440 	 * current locking strategy drops all locks before calling fs_frlock.
4441 	 * So, mapcnt could change before we enter fs_frlock making is
4442 	 * meaningless to have held tlock in the first place.
4443 	 */
4444 	if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode))
4445 		return (EAGAIN);
4446 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4447 }
4448 
4449 /* ARGSUSED */
4450 static int
4451 ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4452 	offset_t offset, cred_t *cr, caller_context_t *ct)
4453 {
4454 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
4455 	struct ulockfs *ulp;
4456 	int error;
4457 
4458 	if ((error = convoff(vp, bfp, 0, offset)) == 0) {
4459 		if (cmd == F_FREESP) {
4460 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4461 			    ULOCKFS_SPACE_MASK);
4462 			if (error)
4463 				return (error);
4464 			error = ufs_freesp(vp, bfp, flag, cr);
4465 		} else if (cmd == F_ALLOCSP) {
4466 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4467 			    ULOCKFS_FALLOCATE_MASK);
4468 			if (error)
4469 				return (error);
4470 			error = ufs_allocsp(vp, bfp, cr);
4471 		} else
4472 			return (EINVAL); /* Command not handled here */
4473 
4474 		if (ulp)
4475 			ufs_lockfs_end(ulp);
4476 
4477 	}
4478 	return (error);
4479 }
4480 
4481 /*
4482  * Used to determine if read ahead should be done. Also used to
4483  * to determine when write back occurs.
4484  */
4485 #define	CLUSTSZ(ip)		((ip)->i_ufsvfs->vfs_ioclustsz)
4486 
4487 /*
4488  * A faster version of ufs_getpage.
4489  *
4490  * We optimize by inlining the pvn_getpages iterator, eliminating
4491  * calls to bmap_read if file doesn't have UFS holes, and avoiding
4492  * the overhead of page_exists().
4493  *
4494  * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4495  * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4496  * victimizes performance when a file with UFS holes is faulted
4497  * first in the S_READ mode, and then in the S_WRITE mode. We will get
4498  * two MMU faults in this case.
4499  *
4500  * XXX - the inode fields which control the sequential mode are not
4501  *	 protected by any mutex. The read ahead will act wild if
4502  *	 multiple processes will access the file concurrently and
4503  *	 some of them in sequential mode. One particulary bad case
4504  *	 is if another thread will change the value of i_nextrio between
4505  *	 the time this thread tests the i_nextrio value and then reads it
4506  *	 again to use it as the offset for the read ahead.
4507  */
4508 /*ARGSUSED*/
4509 static int
4510 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
4511 	page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr,
4512 	enum seg_rw rw, struct cred *cr, caller_context_t *ct)
4513 {
4514 	u_offset_t	uoff = (u_offset_t)off; /* type conversion */
4515 	u_offset_t	pgoff;
4516 	u_offset_t	eoff;
4517 	struct inode 	*ip = VTOI(vp);
4518 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
4519 	struct fs 	*fs;
4520 	struct ulockfs	*ulp;
4521 	page_t		**pl;
4522 	caddr_t		pgaddr;
4523 	krw_t		rwtype;
4524 	int 		err;
4525 	int		has_holes;
4526 	int		beyond_eof;
4527 	int		seqmode;
4528 	int		pgsize = PAGESIZE;
4529 	int		dolock;
4530 	int		do_qlock;
4531 	int		trans_size;
4532 
4533 	ASSERT((uoff & PAGEOFFSET) == 0);
4534 
4535 	if (protp)
4536 		*protp = PROT_ALL;
4537 
4538 	/*
4539 	 * Obey the lockfs protocol
4540 	 */
4541 	err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg,
4542 	    rw == S_READ || rw == S_EXEC, protp);
4543 	if (err)
4544 		goto out;
4545 
4546 	fs = ufsvfsp->vfs_fs;
4547 
4548 	if (ulp && (rw == S_CREATE || rw == S_WRITE) &&
4549 	    !(vp->v_flag & VISSWAP)) {
4550 		/*
4551 		 * Try to start a transaction, will return if blocking is
4552 		 * expected to occur and the address space is not the
4553 		 * kernel address space.
4554 		 */
4555 		trans_size = TOP_GETPAGE_SIZE(ip);
4556 		if (seg->s_as != &kas) {
4557 			TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE,
4558 			    trans_size, err)
4559 			if (err == EWOULDBLOCK) {
4560 				/*
4561 				 * Use EDEADLK here because the VM code
4562 				 * can normally never see this error.
4563 				 */
4564 				err = EDEADLK;
4565 				ufs_lockfs_end(ulp);
4566 				goto out;
4567 			}
4568 		} else {
4569 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4570 		}
4571 	}
4572 
4573 	if (vp->v_flag & VNOMAP) {
4574 		err = ENOSYS;
4575 		goto unlock;
4576 	}
4577 
4578 	seqmode = ip->i_nextr == uoff && rw != S_CREATE;
4579 
4580 	rwtype = RW_READER;		/* start as a reader */
4581 	dolock = (rw_owner(&ip->i_contents) != curthread);
4582 	/*
4583 	 * If this thread owns the lock, i.e., this thread grabbed it
4584 	 * as writer somewhere above, then we don't need to grab the
4585 	 * lock as reader in this routine.
4586 	 */
4587 	do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread);
4588 
4589 retrylock:
4590 	if (dolock) {
4591 		/*
4592 		 * Grab the quota lock if we need to call
4593 		 * bmap_write() below (with i_contents as writer).
4594 		 */
4595 		if (do_qlock && rwtype == RW_WRITER)
4596 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4597 		rw_enter(&ip->i_contents, rwtype);
4598 	}
4599 
4600 	/*
4601 	 * We may be getting called as a side effect of a bmap using
4602 	 * fbread() when the blocks might be being allocated and the
4603 	 * size has not yet been up'ed.  In this case we want to be
4604 	 * able to return zero pages if we get back UFS_HOLE from
4605 	 * calling bmap for a non write case here.  We also might have
4606 	 * to read some frags from the disk into a page if we are
4607 	 * extending the number of frags for a given lbn in bmap().
4608 	 * Large Files: The read of i_size here is atomic because
4609 	 * i_contents is held here. If dolock is zero, the lock
4610 	 * is held in bmap routines.
4611 	 */
4612 	beyond_eof = uoff + len >
4613 	    P2ROUNDUP_TYPED(ip->i_size, PAGESIZE, u_offset_t);
4614 	if (beyond_eof && seg != segkmap) {
4615 		if (dolock) {
4616 			rw_exit(&ip->i_contents);
4617 			if (do_qlock && rwtype == RW_WRITER)
4618 				rw_exit(&ufsvfsp->vfs_dqrwlock);
4619 		}
4620 		err = EFAULT;
4621 		goto unlock;
4622 	}
4623 
4624 	/*
4625 	 * Must hold i_contents lock throughout the call to pvn_getpages
4626 	 * since locked pages are returned from each call to ufs_getapage.
4627 	 * Must *not* return locked pages and then try for contents lock
4628 	 * due to lock ordering requirements (inode > page)
4629 	 */
4630 
4631 	has_holes = bmap_has_holes(ip);
4632 
4633 	if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) {
4634 		int	blk_size;
4635 		u_offset_t offset;
4636 
4637 		/*
4638 		 * We must acquire the RW_WRITER lock in order to
4639 		 * call bmap_write().
4640 		 */
4641 		if (dolock && rwtype == RW_READER) {
4642 			rwtype = RW_WRITER;
4643 
4644 			/*
4645 			 * Grab the quota lock before
4646 			 * upgrading i_contents, but if we can't grab it
4647 			 * don't wait here due to lock order:
4648 			 * vfs_dqrwlock > i_contents.
4649 			 */
4650 			if (do_qlock &&
4651 			    rw_tryenter(&ufsvfsp->vfs_dqrwlock, RW_READER)
4652 			    == 0) {
4653 				rw_exit(&ip->i_contents);
4654 				goto retrylock;
4655 			}
4656 			if (!rw_tryupgrade(&ip->i_contents)) {
4657 				rw_exit(&ip->i_contents);
4658 				if (do_qlock)
4659 					rw_exit(&ufsvfsp->vfs_dqrwlock);
4660 				goto retrylock;
4661 			}
4662 		}
4663 
4664 		/*
4665 		 * May be allocating disk blocks for holes here as
4666 		 * a result of mmap faults. write(2) does the bmap_write
4667 		 * in rdip/wrip, not here. We are not dealing with frags
4668 		 * in this case.
4669 		 */
4670 		/*
4671 		 * Large Files: We cast fs_bmask field to offset_t
4672 		 * just as we do for MAXBMASK because uoff is a 64-bit
4673 		 * data type. fs_bmask will still be a 32-bit type
4674 		 * as we cannot change any ondisk data structures.
4675 		 */
4676 
4677 		offset = uoff & (offset_t)fs->fs_bmask;
4678 		while (offset < uoff + len) {
4679 			blk_size = (int)blksize(fs, ip, lblkno(fs, offset));
4680 			err = bmap_write(ip, offset, blk_size,
4681 			    BI_NORMAL, NULL, cr);
4682 			if (ip->i_flag & (ICHG|IUPD))
4683 				ip->i_seq++;
4684 			if (err)
4685 				goto update_inode;
4686 			offset += blk_size; /* XXX - make this contig */
4687 		}
4688 	}
4689 
4690 	/*
4691 	 * Can be a reader from now on.
4692 	 */
4693 	if (dolock && rwtype == RW_WRITER) {
4694 		rw_downgrade(&ip->i_contents);
4695 		/*
4696 		 * We can release vfs_dqrwlock early so do it, but make
4697 		 * sure we don't try to release it again at the bottom.
4698 		 */
4699 		if (do_qlock) {
4700 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4701 			do_qlock = 0;
4702 		}
4703 	}
4704 
4705 	/*
4706 	 * We remove PROT_WRITE in cases when the file has UFS holes
4707 	 * because we don't  want to call bmap_read() to check each
4708 	 * page if it is backed with a disk block.
4709 	 */
4710 	if (protp && has_holes && rw != S_WRITE && rw != S_CREATE)
4711 		*protp &= ~PROT_WRITE;
4712 
4713 	err = 0;
4714 
4715 	/*
4716 	 * The loop looks up pages in the range [off, off + len).
4717 	 * For each page, we first check if we should initiate an asynchronous
4718 	 * read ahead before we call page_lookup (we may sleep in page_lookup
4719 	 * for a previously initiated disk read).
4720 	 */
4721 	eoff = (uoff + len);
4722 	for (pgoff = uoff, pgaddr = addr, pl = plarr;
4723 	    pgoff < eoff; /* empty */) {
4724 		page_t	*pp;
4725 		u_offset_t	nextrio;
4726 		se_t	se;
4727 		int retval;
4728 
4729 		se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED);
4730 
4731 		/* Handle async getpage (faultahead) */
4732 		if (plarr == NULL) {
4733 			ip->i_nextrio = pgoff;
4734 			(void) ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4735 			pgoff += pgsize;
4736 			pgaddr += pgsize;
4737 			continue;
4738 		}
4739 		/*
4740 		 * Check if we should initiate read ahead of next cluster.
4741 		 * We call page_exists only when we need to confirm that
4742 		 * we have the current page before we initiate the read ahead.
4743 		 */
4744 		nextrio = ip->i_nextrio;
4745 		if (seqmode &&
4746 		    pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
4747 		    nextrio < ip->i_size && page_exists(vp, pgoff)) {
4748 			retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4749 			/*
4750 			 * We always read ahead the next cluster of data
4751 			 * starting from i_nextrio. If the page (vp,nextrio)
4752 			 * is actually in core at this point, the routine
4753 			 * ufs_getpage_ra() will stop pre-fetching data
4754 			 * until we read that page in a synchronized manner
4755 			 * through ufs_getpage_miss(). So, we should increase
4756 			 * i_nextrio if the page (vp, nextrio) exists.
4757 			 */
4758 			if ((retval == 0) && page_exists(vp, nextrio)) {
4759 				ip->i_nextrio = nextrio + pgsize;
4760 			}
4761 		}
4762 
4763 		if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
4764 			/*
4765 			 * We found the page in the page cache.
4766 			 */
4767 			*pl++ = pp;
4768 			pgoff += pgsize;
4769 			pgaddr += pgsize;
4770 			len -= pgsize;
4771 			plsz -= pgsize;
4772 		} else  {
4773 			/*
4774 			 * We have to create the page, or read it from disk.
4775 			 */
4776 			if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr,
4777 			    pl, plsz, rw, seqmode))
4778 				goto error;
4779 
4780 			while (*pl != NULL) {
4781 				pl++;
4782 				pgoff += pgsize;
4783 				pgaddr += pgsize;
4784 				len -= pgsize;
4785 				plsz -= pgsize;
4786 			}
4787 		}
4788 	}
4789 
4790 	/*
4791 	 * Return pages up to plsz if they are in the page cache.
4792 	 * We cannot return pages if there is a chance that they are
4793 	 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4794 	 */
4795 	if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
4796 
4797 		ASSERT((protp == NULL) ||
4798 		    !(has_holes && (*protp & PROT_WRITE)));
4799 
4800 		eoff = pgoff + plsz;
4801 		while (pgoff < eoff) {
4802 			page_t		*pp;
4803 
4804 			if ((pp = page_lookup_nowait(vp, pgoff,
4805 			    SE_SHARED)) == NULL)
4806 				break;
4807 
4808 			*pl++ = pp;
4809 			pgoff += pgsize;
4810 			plsz -= pgsize;
4811 		}
4812 	}
4813 
4814 	if (plarr)
4815 		*pl = NULL;			/* Terminate page list */
4816 	ip->i_nextr = pgoff;
4817 
4818 error:
4819 	if (err && plarr) {
4820 		/*
4821 		 * Release any pages we have locked.
4822 		 */
4823 		while (pl > &plarr[0])
4824 			page_unlock(*--pl);
4825 
4826 		plarr[0] = NULL;
4827 	}
4828 
4829 update_inode:
4830 	/*
4831 	 * If the inode is not already marked for IACC (in rdip() for read)
4832 	 * and the inode is not marked for no access time update (in wrip()
4833 	 * for write) then update the inode access time and mod time now.
4834 	 */
4835 	if ((ip->i_flag & (IACC | INOACC)) == 0) {
4836 		if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) {
4837 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
4838 			    (fs->fs_ronly == 0) &&
4839 			    (!ufsvfsp->vfs_noatime)) {
4840 				mutex_enter(&ip->i_tlock);
4841 				ip->i_flag |= IACC;
4842 				ITIMES_NOLOCK(ip);
4843 				mutex_exit(&ip->i_tlock);
4844 			}
4845 		}
4846 	}
4847 
4848 	if (dolock) {
4849 		rw_exit(&ip->i_contents);
4850 		if (do_qlock && rwtype == RW_WRITER)
4851 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4852 	}
4853 
4854 unlock:
4855 	if (ulp) {
4856 		if ((rw == S_CREATE || rw == S_WRITE) &&
4857 		    !(vp->v_flag & VISSWAP)) {
4858 			TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4859 		}
4860 		ufs_lockfs_end(ulp);
4861 	}
4862 out:
4863 	return (err);
4864 }
4865 
4866 /*
4867  * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4868  * cache. The page is either read from the disk, or it's created.
4869  * A page is created (without disk read) if rw == S_CREATE, or if
4870  * the page is not backed with a real disk block (UFS hole).
4871  */
4872 /* ARGSUSED */
4873 static int
4874 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg,
4875 	caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq)
4876 {
4877 	struct inode	*ip = VTOI(vp);
4878 	page_t		*pp;
4879 	daddr_t		bn;
4880 	size_t		io_len;
4881 	int		crpage = 0;
4882 	int		err;
4883 	int		contig;
4884 	int		bsize = ip->i_fs->fs_bsize;
4885 
4886 	/*
4887 	 * Figure out whether the page can be created, or must be
4888 	 * must be read from the disk.
4889 	 */
4890 	if (rw == S_CREATE)
4891 		crpage = 1;
4892 	else {
4893 		contig = 0;
4894 		if (err = bmap_read(ip, off, &bn, &contig))
4895 			return (err);
4896 
4897 		crpage = (bn == UFS_HOLE);
4898 
4899 		/*
4900 		 * If its also a fallocated block that hasn't been written to
4901 		 * yet, we will treat it just like a UFS_HOLE and create
4902 		 * a zero page for it
4903 		 */
4904 		if (ISFALLOCBLK(ip, bn))
4905 			crpage = 1;
4906 	}
4907 
4908 	if (crpage) {
4909 		if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg,
4910 		    addr)) == NULL) {
4911 			return (ufs_fault(vp,
4912 			    "ufs_getpage_miss: page_create == NULL"));
4913 		}
4914 
4915 		if (rw != S_CREATE)
4916 			pagezero(pp, 0, PAGESIZE);
4917 
4918 		io_len = PAGESIZE;
4919 	} else {
4920 		u_offset_t	io_off;
4921 		uint_t	xlen;
4922 		struct buf	*bp;
4923 		ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
4924 
4925 		/*
4926 		 * If access is not in sequential order, we read from disk
4927 		 * in bsize units.
4928 		 *
4929 		 * We limit the size of the transfer to bsize if we are reading
4930 		 * from the beginning of the file. Note in this situation we
4931 		 * will hedge our bets and initiate an async read ahead of
4932 		 * the second block.
4933 		 */
4934 		if (!seq || off == 0)
4935 			contig = MIN(contig, bsize);
4936 
4937 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4938 		    &io_len, off, contig, 0);
4939 
4940 		/*
4941 		 * Some other thread has entered the page.
4942 		 * ufs_getpage will retry page_lookup.
4943 		 */
4944 		if (pp == NULL) {
4945 			pl[0] = NULL;
4946 			return (0);
4947 		}
4948 
4949 		/*
4950 		 * Zero part of the page which we are not
4951 		 * going to read from the disk.
4952 		 */
4953 		xlen = io_len & PAGEOFFSET;
4954 		if (xlen != 0)
4955 			pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4956 
4957 		bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ);
4958 		bp->b_edev = ip->i_dev;
4959 		bp->b_dev = cmpdev(ip->i_dev);
4960 		bp->b_blkno = bn;
4961 		bp->b_un.b_addr = (caddr_t)0;
4962 		bp->b_file = ip->i_vnode;
4963 		bp->b_offset = off;
4964 
4965 		if (ufsvfsp->vfs_log) {
4966 			lufs_read_strategy(ufsvfsp->vfs_log, bp);
4967 		} else if (ufsvfsp->vfs_snapshot) {
4968 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4969 		} else {
4970 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
4971 			ub.ub_getpages.value.ul++;
4972 			(void) bdev_strategy(bp);
4973 			lwp_stat_update(LWP_STAT_INBLK, 1);
4974 		}
4975 
4976 		ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK);
4977 
4978 		/*
4979 		 * If the file access is sequential, initiate read ahead
4980 		 * of the next cluster.
4981 		 */
4982 		if (seq && ip->i_nextrio < ip->i_size)
4983 			(void) ufs_getpage_ra(vp, off, seg, addr);
4984 		err = biowait(bp);
4985 		pageio_done(bp);
4986 
4987 		if (err) {
4988 			pvn_read_done(pp, B_ERROR);
4989 			return (err);
4990 		}
4991 	}
4992 
4993 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4994 	return (0);
4995 }
4996 
4997 /*
4998  * Read ahead a cluster from the disk. Returns the length in bytes.
4999  */
5000 static int
5001 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr)
5002 {
5003 	struct inode	*ip = VTOI(vp);
5004 	page_t		*pp;
5005 	u_offset_t	io_off = ip->i_nextrio;
5006 	ufsvfs_t	*ufsvfsp;
5007 	caddr_t		addr2 = addr + (io_off - off);
5008 	struct buf	*bp;
5009 	daddr_t		bn;
5010 	size_t		io_len;
5011 	int		err;
5012 	int		contig;
5013 	int		xlen;
5014 	int		bsize = ip->i_fs->fs_bsize;
5015 
5016 	/*
5017 	 * If the directio advisory is in effect on this file,
5018 	 * then do not do buffered read ahead. Read ahead makes
5019 	 * it more difficult on threads using directio as they
5020 	 * will be forced to flush the pages from this vnode.
5021 	 */
5022 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5023 		return (0);
5024 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio)
5025 		return (0);
5026 
5027 	/*
5028 	 * Is this test needed?
5029 	 */
5030 	if (addr2 >= seg->s_base + seg->s_size)
5031 		return (0);
5032 
5033 	contig = 0;
5034 	err = bmap_read(ip, io_off, &bn, &contig);
5035 	/*
5036 	 * If its a UFS_HOLE or a fallocated block, do not perform
5037 	 * any read ahead's since there probably is nothing to read ahead
5038 	 */
5039 	if (err || bn == UFS_HOLE || ISFALLOCBLK(ip, bn))
5040 		return (0);
5041 
5042 	/*
5043 	 * Limit the transfer size to bsize if this is the 2nd block.
5044 	 */
5045 	if (io_off == (u_offset_t)bsize)
5046 		contig = MIN(contig, bsize);
5047 
5048 	if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off,
5049 	    &io_len, io_off, contig, 1)) == NULL)
5050 		return (0);
5051 
5052 	/*
5053 	 * Zero part of page which we are not going to read from disk
5054 	 */
5055 	if ((xlen = (io_len & PAGEOFFSET)) > 0)
5056 		pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
5057 
5058 	ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK;
5059 
5060 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC);
5061 	bp->b_edev = ip->i_dev;
5062 	bp->b_dev = cmpdev(ip->i_dev);
5063 	bp->b_blkno = bn;
5064 	bp->b_un.b_addr = (caddr_t)0;
5065 	bp->b_file = ip->i_vnode;
5066 	bp->b_offset = off;
5067 
5068 	if (ufsvfsp->vfs_log) {
5069 		lufs_read_strategy(ufsvfsp->vfs_log, bp);
5070 	} else if (ufsvfsp->vfs_snapshot) {
5071 		fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5072 	} else {
5073 		ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5074 		ub.ub_getras.value.ul++;
5075 		(void) bdev_strategy(bp);
5076 		lwp_stat_update(LWP_STAT_INBLK, 1);
5077 	}
5078 
5079 	return (io_len);
5080 }
5081 
5082 int	ufs_delay = 1;
5083 /*
5084  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5085  *
5086  * LMXXX - the inode really ought to contain a pointer to one of these
5087  * async args.  Stuff gunk in there and just hand the whole mess off.
5088  * This would replace i_delaylen, i_delayoff.
5089  */
5090 /*ARGSUSED*/
5091 static int
5092 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
5093 	struct cred *cr, caller_context_t *ct)
5094 {
5095 	struct inode *ip = VTOI(vp);
5096 	int err = 0;
5097 
5098 	if (vp->v_count == 0) {
5099 		return (ufs_fault(vp, "ufs_putpage: bad v_count == 0"));
5100 	}
5101 
5102 	/*
5103 	 * XXX - Why should this check be made here?
5104 	 */
5105 	if (vp->v_flag & VNOMAP) {
5106 		err = ENOSYS;
5107 		goto errout;
5108 	}
5109 
5110 	if (ip->i_ufsvfs == NULL) {
5111 		err = EIO;
5112 		goto errout;
5113 	}
5114 
5115 	if (flags & B_ASYNC) {
5116 		if (ufs_delay && len &&
5117 		    (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
5118 			mutex_enter(&ip->i_tlock);
5119 			/*
5120 			 * If nobody stalled, start a new cluster.
5121 			 */
5122 			if (ip->i_delaylen == 0) {
5123 				ip->i_delayoff = off;
5124 				ip->i_delaylen = len;
5125 				mutex_exit(&ip->i_tlock);
5126 				goto errout;
5127 			}
5128 			/*
5129 			 * If we have a full cluster or they are not contig,
5130 			 * then push last cluster and start over.
5131 			 */
5132 			if (ip->i_delaylen >= CLUSTSZ(ip) ||
5133 			    ip->i_delayoff + ip->i_delaylen != off) {
5134 				u_offset_t doff;
5135 				size_t dlen;
5136 
5137 				doff = ip->i_delayoff;
5138 				dlen = ip->i_delaylen;
5139 				ip->i_delayoff = off;
5140 				ip->i_delaylen = len;
5141 				mutex_exit(&ip->i_tlock);
5142 				err = ufs_putpages(vp, doff, dlen,
5143 				    flags, cr);
5144 				/* LMXXX - flags are new val, not old */
5145 				goto errout;
5146 			}
5147 			/*
5148 			 * There is something there, it's not full, and
5149 			 * it is contig.
5150 			 */
5151 			ip->i_delaylen += len;
5152 			mutex_exit(&ip->i_tlock);
5153 			goto errout;
5154 		}
5155 		/*
5156 		 * Must have weird flags or we are not clustering.
5157 		 */
5158 	}
5159 
5160 	err = ufs_putpages(vp, off, len, flags, cr);
5161 
5162 errout:
5163 	return (err);
5164 }
5165 
5166 /*
5167  * If len == 0, do from off to EOF.
5168  *
5169  * The normal cases should be len == 0 & off == 0 (entire vp list),
5170  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5171  * (from pageout).
5172  */
5173 /*ARGSUSED*/
5174 static int
5175 ufs_putpages(
5176 	struct vnode *vp,
5177 	offset_t off,
5178 	size_t len,
5179 	int flags,
5180 	struct cred *cr)
5181 {
5182 	u_offset_t io_off;
5183 	u_offset_t eoff;
5184 	struct inode *ip = VTOI(vp);
5185 	page_t *pp;
5186 	size_t io_len;
5187 	int err = 0;
5188 	int dolock;
5189 
5190 	if (vp->v_count == 0)
5191 		return (ufs_fault(vp, "ufs_putpages: v_count == 0"));
5192 	/*
5193 	 * Acquire the readers/write inode lock before locking
5194 	 * any pages in this inode.
5195 	 * The inode lock is held during i/o.
5196 	 */
5197 	if (len == 0) {
5198 		mutex_enter(&ip->i_tlock);
5199 		ip->i_delayoff = ip->i_delaylen = 0;
5200 		mutex_exit(&ip->i_tlock);
5201 	}
5202 	dolock = (rw_owner(&ip->i_contents) != curthread);
5203 	if (dolock) {
5204 		/*
5205 		 * Must synchronize this thread and any possible thread
5206 		 * operating in the window of vulnerability in wrip().
5207 		 * It is dangerous to allow both a thread doing a putpage
5208 		 * and a thread writing, so serialize them.  The exception
5209 		 * is when the thread in wrip() does something which causes
5210 		 * a putpage operation.  Then, the thread must be allowed
5211 		 * to continue.  It may encounter a bmap_read problem in
5212 		 * ufs_putapage, but that is handled in ufs_putapage.
5213 		 * Allow async writers to proceed, we don't want to block
5214 		 * the pageout daemon.
5215 		 */
5216 		if (ip->i_writer == curthread)
5217 			rw_enter(&ip->i_contents, RW_READER);
5218 		else {
5219 			for (;;) {
5220 				rw_enter(&ip->i_contents, RW_READER);
5221 				mutex_enter(&ip->i_tlock);
5222 				/*
5223 				 * If there is no thread in the critical
5224 				 * section of wrip(), then proceed.
5225 				 * Otherwise, wait until there isn't one.
5226 				 */
5227 				if (ip->i_writer == NULL) {
5228 					mutex_exit(&ip->i_tlock);
5229 					break;
5230 				}
5231 				rw_exit(&ip->i_contents);
5232 				/*
5233 				 * Bounce async writers when we have a writer
5234 				 * working on this file so we don't deadlock
5235 				 * the pageout daemon.
5236 				 */
5237 				if (flags & B_ASYNC) {
5238 					mutex_exit(&ip->i_tlock);
5239 					return (0);
5240 				}
5241 				cv_wait(&ip->i_wrcv, &ip->i_tlock);
5242 				mutex_exit(&ip->i_tlock);
5243 			}
5244 		}
5245 	}
5246 
5247 	if (!vn_has_cached_data(vp)) {
5248 		if (dolock)
5249 			rw_exit(&ip->i_contents);
5250 		return (0);
5251 	}
5252 
5253 	if (len == 0) {
5254 		/*
5255 		 * Search the entire vp list for pages >= off.
5256 		 */
5257 		err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage,
5258 		    flags, cr);
5259 	} else {
5260 		/*
5261 		 * Loop over all offsets in the range looking for
5262 		 * pages to deal with.
5263 		 */
5264 		if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0)
5265 			eoff = MIN(off + len, eoff);
5266 		else
5267 			eoff = off + len;
5268 
5269 		for (io_off = off; io_off < eoff; io_off += io_len) {
5270 			/*
5271 			 * If we are not invalidating, synchronously
5272 			 * freeing or writing pages, use the routine
5273 			 * page_lookup_nowait() to prevent reclaiming
5274 			 * them from the free list.
5275 			 */
5276 			if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
5277 				pp = page_lookup(vp, io_off,
5278 				    (flags & (B_INVAL | B_FREE)) ?
5279 				    SE_EXCL : SE_SHARED);
5280 			} else {
5281 				pp = page_lookup_nowait(vp, io_off,
5282 				    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
5283 			}
5284 
5285 			if (pp == NULL || pvn_getdirty(pp, flags) == 0)
5286 				io_len = PAGESIZE;
5287 			else {
5288 				u_offset_t *io_offp = &io_off;
5289 
5290 				err = ufs_putapage(vp, pp, io_offp, &io_len,
5291 				    flags, cr);
5292 				if (err != 0)
5293 					break;
5294 				/*
5295 				 * "io_off" and "io_len" are returned as
5296 				 * the range of pages we actually wrote.
5297 				 * This allows us to skip ahead more quickly
5298 				 * since several pages may've been dealt
5299 				 * with by this iteration of the loop.
5300 				 */
5301 			}
5302 		}
5303 	}
5304 	if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
5305 		/*
5306 		 * We have just sync'ed back all the pages on
5307 		 * the inode, turn off the IMODTIME flag.
5308 		 */
5309 		mutex_enter(&ip->i_tlock);
5310 		ip->i_flag &= ~IMODTIME;
5311 		mutex_exit(&ip->i_tlock);
5312 	}
5313 	if (dolock)
5314 		rw_exit(&ip->i_contents);
5315 	return (err);
5316 }
5317 
5318 static void
5319 ufs_iodone(buf_t *bp)
5320 {
5321 	struct inode *ip;
5322 
5323 	ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
5324 
5325 	bp->b_iodone = NULL;
5326 
5327 	ip = VTOI(bp->b_pages->p_vnode);
5328 
5329 	mutex_enter(&ip->i_tlock);
5330 	if (ip->i_writes >= ufs_LW) {
5331 		if ((ip->i_writes -= bp->b_bcount) <= ufs_LW)
5332 			if (ufs_WRITES)
5333 				cv_broadcast(&ip->i_wrcv); /* wake all up */
5334 	} else {
5335 		ip->i_writes -= bp->b_bcount;
5336 	}
5337 
5338 	mutex_exit(&ip->i_tlock);
5339 	iodone(bp);
5340 }
5341 
5342 /*
5343  * Write out a single page, possibly klustering adjacent
5344  * dirty pages.  The inode lock must be held.
5345  *
5346  * LMXXX - bsize < pagesize not done.
5347  */
5348 /*ARGSUSED*/
5349 int
5350 ufs_putapage(
5351 	struct vnode *vp,
5352 	page_t *pp,
5353 	u_offset_t *offp,
5354 	size_t *lenp,		/* return values */
5355 	int flags,
5356 	struct cred *cr)
5357 {
5358 	u_offset_t io_off;
5359 	u_offset_t off;
5360 	struct inode *ip = VTOI(vp);
5361 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
5362 	struct fs *fs;
5363 	struct buf *bp;
5364 	size_t io_len;
5365 	daddr_t bn;
5366 	int err;
5367 	int contig;
5368 	int dotrans;
5369 
5370 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
5371 
5372 	if (ufsvfsp == NULL) {
5373 		err = EIO;
5374 		goto out_trace;
5375 	}
5376 
5377 	fs = ip->i_fs;
5378 	ASSERT(fs->fs_ronly == 0);
5379 
5380 	/*
5381 	 * If the modified time on the inode has not already been
5382 	 * set elsewhere (e.g. for write/setattr) we set the time now.
5383 	 * This gives us approximate modified times for mmap'ed files
5384 	 * which are modified via stores in the user address space.
5385 	 */
5386 	if ((ip->i_flag & IMODTIME) == 0) {
5387 		mutex_enter(&ip->i_tlock);
5388 		ip->i_flag |= IUPD;
5389 		ip->i_seq++;
5390 		ITIMES_NOLOCK(ip);
5391 		mutex_exit(&ip->i_tlock);
5392 	}
5393 
5394 	/*
5395 	 * Align the request to a block boundry (for old file systems),
5396 	 * and go ask bmap() how contiguous things are for this file.
5397 	 */
5398 	off = pp->p_offset & (offset_t)fs->fs_bmask;	/* block align it */
5399 	contig = 0;
5400 	err = bmap_read(ip, off, &bn, &contig);
5401 	if (err)
5402 		goto out;
5403 	if (bn == UFS_HOLE) {			/* putpage never allocates */
5404 		/*
5405 		 * logging device is in error mode; simply return EIO
5406 		 */
5407 		if (TRANS_ISERROR(ufsvfsp)) {
5408 			err = EIO;
5409 			goto out;
5410 		}
5411 		/*
5412 		 * Oops, the thread in the window in wrip() did some
5413 		 * sort of operation which caused a putpage in the bad
5414 		 * range.  In this case, just return an error which will
5415 		 * cause the software modified bit on the page to set
5416 		 * and the page will get written out again later.
5417 		 */
5418 		if (ip->i_writer == curthread) {
5419 			err = EIO;
5420 			goto out;
5421 		}
5422 		/*
5423 		 * If the pager is trying to push a page in the bad range
5424 		 * just tell him to try again later when things are better.
5425 		 */
5426 		if (flags & B_ASYNC) {
5427 			err = EAGAIN;
5428 			goto out;
5429 		}
5430 		err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE");
5431 		goto out;
5432 	}
5433 
5434 	/*
5435 	 * If it is an fallocate'd block, reverse the negativity since
5436 	 * we are now writing to it
5437 	 */
5438 	if (ISFALLOCBLK(ip, bn)) {
5439 		err = bmap_set_bn(vp, off, dbtofsb(fs, -bn));
5440 		if (err)
5441 			goto out;
5442 
5443 		bn = -bn;
5444 	}
5445 
5446 	/*
5447 	 * Take the length (of contiguous bytes) passed back from bmap()
5448 	 * and _try_ and get a set of pages covering that extent.
5449 	 */
5450 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags);
5451 
5452 	/*
5453 	 * May have run out of memory and not clustered backwards.
5454 	 * off		p_offset
5455 	 * [  pp - 1  ][   pp   ]
5456 	 * [	block		]
5457 	 * We told bmap off, so we have to adjust the bn accordingly.
5458 	 */
5459 	if (io_off > off) {
5460 		bn += btod(io_off - off);
5461 		contig -= (io_off - off);
5462 	}
5463 
5464 	/*
5465 	 * bmap was carefull to tell us the right size so use that.
5466 	 * There might be unallocated frags at the end.
5467 	 * LMXXX - bzero the end of the page?  We must be writing after EOF.
5468 	 */
5469 	if (io_len > contig) {
5470 		ASSERT(io_len - contig < fs->fs_bsize);
5471 		io_len -= (io_len - contig);
5472 	}
5473 
5474 	/*
5475 	 * Handle the case where we are writing the last page after EOF.
5476 	 *
5477 	 * XXX - just a patch for i-mt3.
5478 	 */
5479 	if (io_len == 0) {
5480 		ASSERT(pp->p_offset >=
5481 		    (u_offset_t)(roundup(ip->i_size, PAGESIZE)));
5482 		io_len = PAGESIZE;
5483 	}
5484 
5485 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags);
5486 
5487 	ULOCKFS_SET_MOD(ITOUL(ip));
5488 
5489 	bp->b_edev = ip->i_dev;
5490 	bp->b_dev = cmpdev(ip->i_dev);
5491 	bp->b_blkno = bn;
5492 	bp->b_un.b_addr = (caddr_t)0;
5493 	bp->b_file = ip->i_vnode;
5494 
5495 	/*
5496 	 * File contents of shadow or quota inodes are metadata, and updates
5497 	 * to these need to be put into a logging transaction. All direct
5498 	 * callers in UFS do that, but fsflush can come here _before_ the
5499 	 * normal codepath. An example would be updating ACL information, for
5500 	 * which the normal codepath would be:
5501 	 *	ufs_si_store()
5502 	 *	ufs_rdwri()
5503 	 *	wrip()
5504 	 *	segmap_release()
5505 	 *	VOP_PUTPAGE()
5506 	 * Here, fsflush can pick up the dirty page before segmap_release()
5507 	 * forces it out. If that happens, there's no transaction.
5508 	 * We therefore need to test whether a transaction exists, and if not
5509 	 * create one - for fsflush.
5510 	 */
5511 	dotrans =
5512 	    (((ip->i_mode & IFMT) == IFSHAD || ufsvfsp->vfs_qinod == ip) &&
5513 	    ((curthread->t_flag & T_DONTBLOCK) == 0) &&
5514 	    (TRANS_ISTRANS(ufsvfsp)));
5515 
5516 	if (dotrans) {
5517 		curthread->t_flag |= T_DONTBLOCK;
5518 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5519 	}
5520 	if (TRANS_ISTRANS(ufsvfsp)) {
5521 		if ((ip->i_mode & IFMT) == IFSHAD) {
5522 			TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD);
5523 		} else if (ufsvfsp->vfs_qinod == ip) {
5524 			TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR,
5525 			    0, 0);
5526 		}
5527 	}
5528 	if (dotrans) {
5529 		TRANS_END_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5530 		curthread->t_flag &= ~T_DONTBLOCK;
5531 	}
5532 
5533 	/* write throttle */
5534 
5535 	ASSERT(bp->b_iodone == NULL);
5536 	bp->b_iodone = (int (*)())ufs_iodone;
5537 	mutex_enter(&ip->i_tlock);
5538 	ip->i_writes += bp->b_bcount;
5539 	mutex_exit(&ip->i_tlock);
5540 
5541 	if (bp->b_flags & B_ASYNC) {
5542 		if (ufsvfsp->vfs_log) {
5543 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5544 		} else if (ufsvfsp->vfs_snapshot) {
5545 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5546 		} else {
5547 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5548 			ub.ub_putasyncs.value.ul++;
5549 			(void) bdev_strategy(bp);
5550 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5551 		}
5552 	} else {
5553 		if (ufsvfsp->vfs_log) {
5554 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5555 		} else if (ufsvfsp->vfs_snapshot) {
5556 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5557 		} else {
5558 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5559 			ub.ub_putsyncs.value.ul++;
5560 			(void) bdev_strategy(bp);
5561 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5562 		}
5563 		err = biowait(bp);
5564 		pageio_done(bp);
5565 		pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
5566 	}
5567 
5568 	pp = NULL;
5569 
5570 out:
5571 	if (err != 0 && pp != NULL)
5572 		pvn_write_done(pp, B_ERROR | B_WRITE | flags);
5573 
5574 	if (offp)
5575 		*offp = io_off;
5576 	if (lenp)
5577 		*lenp = io_len;
5578 out_trace:
5579 	return (err);
5580 }
5581 
5582 uint64_t ufs_map_alock_retry_cnt;
5583 uint64_t ufs_map_lockfs_retry_cnt;
5584 
5585 /* ARGSUSED */
5586 static int
5587 ufs_map(struct vnode *vp,
5588 	offset_t off,
5589 	struct as *as,
5590 	caddr_t *addrp,
5591 	size_t len,
5592 	uchar_t prot,
5593 	uchar_t maxprot,
5594 	uint_t flags,
5595 	struct cred *cr,
5596 	caller_context_t *ct)
5597 {
5598 	struct segvn_crargs vn_a;
5599 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5600 	struct ulockfs *ulp;
5601 	int error, sig;
5602 	k_sigset_t smask;
5603 	caddr_t hint = *addrp;
5604 
5605 	if (vp->v_flag & VNOMAP) {
5606 		error = ENOSYS;
5607 		goto out;
5608 	}
5609 
5610 	if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) {
5611 		error = ENXIO;
5612 		goto out;
5613 	}
5614 
5615 	if (vp->v_type != VREG) {
5616 		error = ENODEV;
5617 		goto out;
5618 	}
5619 
5620 retry_map:
5621 	*addrp = hint;
5622 	/*
5623 	 * If file is being locked, disallow mapping.
5624 	 */
5625 	if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) {
5626 		error = EAGAIN;
5627 		goto out;
5628 	}
5629 
5630 	as_rangelock(as);
5631 	/*
5632 	 * Note that if we are retrying (because ufs_lockfs_trybegin failed in
5633 	 * the previous attempt), some other thread could have grabbed
5634 	 * the same VA range if MAP_FIXED is set. In that case, choose_addr
5635 	 * would unmap the valid VA range, that is ok.
5636 	 */
5637 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5638 	if (error != 0) {
5639 		as_rangeunlock(as);
5640 		goto out;
5641 	}
5642 
5643 	/*
5644 	 * a_lock has to be acquired before entering the lockfs protocol
5645 	 * because that is the order in which pagefault works. Also we cannot
5646 	 * block on a_lock here because this waiting writer will prevent
5647 	 * further readers like ufs_read from progressing and could cause
5648 	 * deadlock between ufs_read/ufs_map/pagefault when a quiesce is
5649 	 * pending.
5650 	 */
5651 	while (!AS_LOCK_TRYENTER(as, &as->a_lock, RW_WRITER)) {
5652 		ufs_map_alock_retry_cnt++;
5653 		delay(RETRY_LOCK_DELAY);
5654 	}
5655 
5656 	/*
5657 	 * We can't hold as->a_lock and wait for lockfs to succeed because
5658 	 * the proc tools might hang on a_lock, so call ufs_lockfs_trybegin()
5659 	 * instead.
5660 	 */
5661 	if (error = ufs_lockfs_trybegin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK)) {
5662 		/*
5663 		 * ufs_lockfs_trybegin() did not succeed. It is safer to give up
5664 		 * as->a_lock and wait for ulp->ul_fs_lock status to change.
5665 		 */
5666 		ufs_map_lockfs_retry_cnt++;
5667 		AS_LOCK_EXIT(as, &as->a_lock);
5668 		as_rangeunlock(as);
5669 		if (error == EIO)
5670 			goto out;
5671 
5672 		mutex_enter(&ulp->ul_lock);
5673 		while (ulp->ul_fs_lock & ULOCKFS_MAP_MASK) {
5674 			if (ULOCKFS_IS_SLOCK(ulp) || ufsvfsp->vfs_nointr) {
5675 				cv_wait(&ulp->ul_cv, &ulp->ul_lock);
5676 			} else {
5677 				sigintr(&smask, 1);
5678 				sig = cv_wait_sig(&ulp->ul_cv, &ulp->ul_lock);
5679 				sigunintr(&smask);
5680 				if (((ulp->ul_fs_lock & ULOCKFS_MAP_MASK) &&
5681 				    !sig) || ufsvfsp->vfs_dontblock) {
5682 					mutex_exit(&ulp->ul_lock);
5683 					return (EINTR);
5684 				}
5685 			}
5686 		}
5687 		mutex_exit(&ulp->ul_lock);
5688 		goto retry_map;
5689 	}
5690 
5691 	vn_a.vp = vp;
5692 	vn_a.offset = (u_offset_t)off;
5693 	vn_a.type = flags & MAP_TYPE;
5694 	vn_a.prot = prot;
5695 	vn_a.maxprot = maxprot;
5696 	vn_a.cred = cr;
5697 	vn_a.amp = NULL;
5698 	vn_a.flags = flags & ~MAP_TYPE;
5699 	vn_a.szc = 0;
5700 	vn_a.lgrp_mem_policy_flags = 0;
5701 
5702 	error = as_map_locked(as, *addrp, len, segvn_create, &vn_a);
5703 	if (ulp)
5704 		ufs_lockfs_end(ulp);
5705 	as_rangeunlock(as);
5706 out:
5707 	return (error);
5708 }
5709 
5710 /* ARGSUSED */
5711 static int
5712 ufs_addmap(struct vnode *vp,
5713 	offset_t off,
5714 	struct as *as,
5715 	caddr_t addr,
5716 	size_t	len,
5717 	uchar_t  prot,
5718 	uchar_t  maxprot,
5719 	uint_t    flags,
5720 	struct cred *cr,
5721 	caller_context_t *ct)
5722 {
5723 	struct inode *ip = VTOI(vp);
5724 
5725 	if (vp->v_flag & VNOMAP) {
5726 		return (ENOSYS);
5727 	}
5728 
5729 	mutex_enter(&ip->i_tlock);
5730 	ip->i_mapcnt += btopr(len);
5731 	mutex_exit(&ip->i_tlock);
5732 	return (0);
5733 }
5734 
5735 /*ARGSUSED*/
5736 static int
5737 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5738 	size_t len, uint_t prot,  uint_t maxprot,  uint_t flags,
5739 	struct cred *cr, caller_context_t *ct)
5740 {
5741 	struct inode *ip = VTOI(vp);
5742 
5743 	if (vp->v_flag & VNOMAP) {
5744 		return (ENOSYS);
5745 	}
5746 
5747 	mutex_enter(&ip->i_tlock);
5748 	ip->i_mapcnt -= btopr(len); 	/* Count released mappings */
5749 	ASSERT(ip->i_mapcnt >= 0);
5750 	mutex_exit(&ip->i_tlock);
5751 	return (0);
5752 }
5753 /*
5754  * Return the answer requested to poll() for non-device files
5755  */
5756 struct pollhead ufs_pollhd;
5757 
5758 /* ARGSUSED */
5759 int
5760 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp,
5761 	caller_context_t *ct)
5762 {
5763 	struct ufsvfs	*ufsvfsp;
5764 
5765 	*revp = 0;
5766 	ufsvfsp = VTOI(vp)->i_ufsvfs;
5767 
5768 	if (!ufsvfsp) {
5769 		*revp = POLLHUP;
5770 		goto out;
5771 	}
5772 
5773 	if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) ||
5774 	    ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) {
5775 		*revp |= POLLERR;
5776 
5777 	} else {
5778 		if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly &&
5779 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5780 			*revp |= POLLOUT;
5781 
5782 		if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly &&
5783 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5784 			*revp |= POLLWRBAND;
5785 
5786 		if (ev & POLLIN)
5787 			*revp |= POLLIN;
5788 
5789 		if (ev & POLLRDNORM)
5790 			*revp |= POLLRDNORM;
5791 
5792 		if (ev & POLLRDBAND)
5793 			*revp |= POLLRDBAND;
5794 	}
5795 
5796 	if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP)))
5797 		*revp |= POLLPRI;
5798 out:
5799 	*phpp = !any && !*revp ? &ufs_pollhd : (struct pollhead *)NULL;
5800 
5801 	return (0);
5802 }
5803 
5804 /* ARGSUSED */
5805 static int
5806 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr,
5807 	caller_context_t *ct)
5808 {
5809 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
5810 	struct ulockfs	*ulp = NULL;
5811 	struct inode 	*sip = NULL;
5812 	int		error;
5813 	struct inode 	*ip = VTOI(vp);
5814 	int		issync;
5815 
5816 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK);
5817 	if (error)
5818 		return (error);
5819 
5820 	switch (cmd) {
5821 		/*
5822 		 * Have to handle _PC_NAME_MAX here, because the normal way
5823 		 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5824 		 * results in a lock ordering reversal between
5825 		 * ufs_lockfs_{begin,end}() and
5826 		 * ufs_thread_{suspend,continue}().
5827 		 *
5828 		 * Keep in sync with ufs_statvfs().
5829 		 */
5830 	case _PC_NAME_MAX:
5831 		*valp = MAXNAMLEN;
5832 		break;
5833 
5834 	case _PC_FILESIZEBITS:
5835 		if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
5836 			*valp = UFS_FILESIZE_BITS;
5837 		else
5838 			*valp = 32;
5839 		break;
5840 
5841 	case _PC_XATTR_EXISTS:
5842 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
5843 
5844 			error =
5845 			    ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR, cr);
5846 			if (error ==  0 && sip != NULL) {
5847 				/* Start transaction */
5848 				if (ulp) {
5849 					TRANS_BEGIN_CSYNC(ufsvfsp, issync,
5850 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5851 				}
5852 				/*
5853 				 * Is directory empty
5854 				 */
5855 				rw_enter(&sip->i_rwlock, RW_WRITER);
5856 				rw_enter(&sip->i_contents, RW_WRITER);
5857 				if (ufs_xattrdirempty(sip,
5858 				    sip->i_number, CRED())) {
5859 					rw_enter(&ip->i_contents, RW_WRITER);
5860 					ufs_unhook_shadow(ip, sip);
5861 					rw_exit(&ip->i_contents);
5862 
5863 					*valp = 0;
5864 
5865 				} else
5866 					*valp = 1;
5867 				rw_exit(&sip->i_contents);
5868 				rw_exit(&sip->i_rwlock);
5869 				if (ulp) {
5870 					TRANS_END_CSYNC(ufsvfsp, error, issync,
5871 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5872 				}
5873 				VN_RELE(ITOV(sip));
5874 			} else if (error == ENOENT) {
5875 				*valp = 0;
5876 				error = 0;
5877 			}
5878 		} else {
5879 			error = fs_pathconf(vp, cmd, valp, cr, ct);
5880 		}
5881 		break;
5882 
5883 	case _PC_ACL_ENABLED:
5884 		*valp = _ACL_ACLENT_ENABLED;
5885 		break;
5886 
5887 	case _PC_MIN_HOLE_SIZE:
5888 		*valp = (ulong_t)ip->i_fs->fs_bsize;
5889 		break;
5890 
5891 	case _PC_SATTR_ENABLED:
5892 	case _PC_SATTR_EXISTS:
5893 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5894 		    (vp->v_type == VREG || vp->v_type == VDIR);
5895 		break;
5896 
5897 	case _PC_TIMESTAMP_RESOLUTION:
5898 		/*
5899 		 * UFS keeps only microsecond timestamp resolution.
5900 		 * This is historical and will probably never change.
5901 		 */
5902 		*valp = 1000L;
5903 		break;
5904 
5905 	default:
5906 		error = fs_pathconf(vp, cmd, valp, cr, ct);
5907 		break;
5908 	}
5909 
5910 	if (ulp != NULL) {
5911 		ufs_lockfs_end(ulp);
5912 	}
5913 	return (error);
5914 }
5915 
5916 int ufs_pageio_writes, ufs_pageio_reads;
5917 
5918 /*ARGSUSED*/
5919 static int
5920 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len,
5921 	int flags, struct cred *cr, caller_context_t *ct)
5922 {
5923 	struct inode *ip = VTOI(vp);
5924 	struct ufsvfs *ufsvfsp;
5925 	page_t *npp = NULL, *opp = NULL, *cpp = pp;
5926 	struct buf *bp;
5927 	daddr_t bn;
5928 	size_t done_len = 0, cur_len = 0;
5929 	int err = 0;
5930 	int contig = 0;
5931 	int dolock;
5932 	int vmpss = 0;
5933 	struct ulockfs *ulp;
5934 
5935 	if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp &&
5936 	    vp->v_mpssdata != NULL) {
5937 		vmpss = 1;
5938 	}
5939 
5940 	dolock = (rw_owner(&ip->i_contents) != curthread);
5941 	/*
5942 	 * We need a better check.  Ideally, we would use another
5943 	 * vnodeops so that hlocked and forcibly unmounted file
5944 	 * systems would return EIO where appropriate and w/o the
5945 	 * need for these checks.
5946 	 */
5947 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5948 		return (EIO);
5949 
5950 	/*
5951 	 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5952 	 * ul_lock must be taken before locking pages so we can't use it here
5953 	 * if pp is non NULL because segvn already locked pages
5954 	 * SE_EXCL. Instead we rely on the fact that a forced umount or
5955 	 * applying a filesystem lock via ufs_fiolfs() will block in the
5956 	 * implicit call to ufs_flush() until we unlock the pages after the
5957 	 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5958 	 * above 0 until they are done. We have to be careful not to increment
5959 	 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5960 	 *
5961 	 * If pp is NULL use ul_lock to make sure we don't increment
5962 	 * ul_vnops_cnt after forceful unmount hlocks the file system.
5963 	 */
5964 	if (vmpss || pp == NULL) {
5965 		ulp = &ufsvfsp->vfs_ulockfs;
5966 		if (pp == NULL)
5967 			mutex_enter(&ulp->ul_lock);
5968 		if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) {
5969 			if (pp == NULL) {
5970 				mutex_exit(&ulp->ul_lock);
5971 			}
5972 			return (vmpss ? EIO : EINVAL);
5973 		}
5974 		atomic_add_long(&ulp->ul_vnops_cnt, 1);
5975 		if (pp == NULL)
5976 			mutex_exit(&ulp->ul_lock);
5977 		if (ufs_quiesce_pend) {
5978 			if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5979 				cv_broadcast(&ulp->ul_cv);
5980 			return (vmpss ? EIO : EINVAL);
5981 		}
5982 	}
5983 
5984 	if (dolock) {
5985 		/*
5986 		 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
5987 		 * handle a fault against a segment that maps vnode pages with
5988 		 * large mappings.  Segvn creates pages and holds them locked
5989 		 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
5990 		 * use rw_tryenter() to avoid a potential deadlock since in
5991 		 * lock order i_contents needs to be taken first.
5992 		 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
5993 		 */
5994 		if (!vmpss) {
5995 			rw_enter(&ip->i_contents, RW_READER);
5996 		} else if (!rw_tryenter(&ip->i_contents, RW_READER)) {
5997 			if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5998 				cv_broadcast(&ulp->ul_cv);
5999 			return (EDEADLK);
6000 		}
6001 	}
6002 
6003 	/*
6004 	 * Return an error to segvn because the pagefault request is beyond
6005 	 * PAGESIZE rounded EOF.
6006 	 */
6007 	if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) {
6008 		if (dolock)
6009 			rw_exit(&ip->i_contents);
6010 		if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
6011 			cv_broadcast(&ulp->ul_cv);
6012 		return (EFAULT);
6013 	}
6014 
6015 	if (pp == NULL) {
6016 		if (bmap_has_holes(ip)) {
6017 			err = ENOSYS;
6018 		} else {
6019 			err = EINVAL;
6020 		}
6021 		if (dolock)
6022 			rw_exit(&ip->i_contents);
6023 		if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
6024 			cv_broadcast(&ulp->ul_cv);
6025 		return (err);
6026 	}
6027 
6028 	/*
6029 	 * Break the io request into chunks, one for each contiguous
6030 	 * stretch of disk blocks in the target file.
6031 	 */
6032 	while (done_len < io_len) {
6033 		ASSERT(cpp);
6034 		contig = 0;
6035 		if (err = bmap_read(ip, (u_offset_t)(io_off + done_len),
6036 		    &bn, &contig))
6037 			break;
6038 
6039 		if (bn == UFS_HOLE) {	/* No holey swapfiles */
6040 			if (vmpss) {
6041 				err = EFAULT;
6042 				break;
6043 			}
6044 			err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE");
6045 			break;
6046 		}
6047 
6048 		cur_len = MIN(io_len - done_len, contig);
6049 		/*
6050 		 * Zero out a page beyond EOF, when the last block of
6051 		 * a file is a UFS fragment so that ufs_pageio() can be used
6052 		 * instead of ufs_getpage() to handle faults against
6053 		 * segvn segments that use large pages.
6054 		 */
6055 		page_list_break(&cpp, &npp, btopr(cur_len));
6056 		if ((flags & B_READ) && (cur_len & PAGEOFFSET)) {
6057 			size_t xlen = cur_len & PAGEOFFSET;
6058 			pagezero(cpp->p_prev, xlen, PAGESIZE - xlen);
6059 		}
6060 
6061 		bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
6062 		ASSERT(bp != NULL);
6063 
6064 		bp->b_edev = ip->i_dev;
6065 		bp->b_dev = cmpdev(ip->i_dev);
6066 		bp->b_blkno = bn;
6067 		bp->b_un.b_addr = (caddr_t)0;
6068 		bp->b_file = ip->i_vnode;
6069 
6070 		ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
6071 		ub.ub_pageios.value.ul++;
6072 		if (ufsvfsp->vfs_snapshot)
6073 			fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp);
6074 		else
6075 			(void) bdev_strategy(bp);
6076 
6077 		if (flags & B_READ)
6078 			ufs_pageio_reads++;
6079 		else
6080 			ufs_pageio_writes++;
6081 		if (flags & B_READ)
6082 			lwp_stat_update(LWP_STAT_INBLK, 1);
6083 		else
6084 			lwp_stat_update(LWP_STAT_OUBLK, 1);
6085 		/*
6086 		 * If the request is not B_ASYNC, wait for i/o to complete
6087 		 * and re-assemble the page list to return to the caller.
6088 		 * If it is B_ASYNC we leave the page list in pieces and
6089 		 * cleanup() will dispose of them.
6090 		 */
6091 		if ((flags & B_ASYNC) == 0) {
6092 			err = biowait(bp);
6093 			pageio_done(bp);
6094 			if (err)
6095 				break;
6096 			page_list_concat(&opp, &cpp);
6097 		}
6098 		cpp = npp;
6099 		npp = NULL;
6100 		if (flags & B_READ)
6101 			cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t);
6102 		done_len += cur_len;
6103 	}
6104 	ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len));
6105 	if (err) {
6106 		if (flags & B_ASYNC) {
6107 			/* Cleanup unprocessed parts of list */
6108 			page_list_concat(&cpp, &npp);
6109 			if (flags & B_READ)
6110 				pvn_read_done(cpp, B_ERROR);
6111 			else
6112 				pvn_write_done(cpp, B_ERROR);
6113 		} else {
6114 			/* Re-assemble list and let caller clean up */
6115 			page_list_concat(&opp, &cpp);
6116 			page_list_concat(&opp, &npp);
6117 		}
6118 	}
6119 
6120 	if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) &&
6121 	    ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) {
6122 		mutex_enter(&ip->i_tlock);
6123 		ip->i_flag |= IACC;
6124 		ITIMES_NOLOCK(ip);
6125 		mutex_exit(&ip->i_tlock);
6126 	}
6127 
6128 	if (dolock)
6129 		rw_exit(&ip->i_contents);
6130 	if (vmpss && !atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
6131 		cv_broadcast(&ulp->ul_cv);
6132 	return (err);
6133 }
6134 
6135 /*
6136  * Called when the kernel is in a frozen state to dump data
6137  * directly to the device. It uses a private dump data structure,
6138  * set up by dump_ctl, to locate the correct disk block to which to dump.
6139  */
6140 /*ARGSUSED*/
6141 static int
6142 ufs_dump(vnode_t *vp, caddr_t addr, offset_t ldbn, offset_t dblks,
6143     caller_context_t *ct)
6144 {
6145 	u_offset_t	file_size;
6146 	struct inode    *ip = VTOI(vp);
6147 	struct fs	*fs = ip->i_fs;
6148 	daddr_t		dbn, lfsbn;
6149 	int		disk_blks = fs->fs_bsize >> DEV_BSHIFT;
6150 	int		error = 0;
6151 	int		ndbs, nfsbs;
6152 
6153 	/*
6154 	 * forced unmount case
6155 	 */
6156 	if (ip->i_ufsvfs == NULL)
6157 		return (EIO);
6158 	/*
6159 	 * Validate the inode that it has not been modified since
6160 	 * the dump structure is allocated.
6161 	 */
6162 	mutex_enter(&ip->i_tlock);
6163 	if ((dump_info == NULL) ||
6164 	    (dump_info->ip != ip) ||
6165 	    (dump_info->time.tv_sec != ip->i_mtime.tv_sec) ||
6166 	    (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) {
6167 		mutex_exit(&ip->i_tlock);
6168 		return (-1);
6169 	}
6170 	mutex_exit(&ip->i_tlock);
6171 
6172 	/*
6173 	 * See that the file has room for this write
6174 	 */
6175 	UFS_GET_ISIZE(&file_size, ip);
6176 
6177 	if (ldbtob(ldbn + dblks) > file_size)
6178 		return (ENOSPC);
6179 
6180 	/*
6181 	 * Find the physical disk block numbers from the dump
6182 	 * private data structure directly and write out the data
6183 	 * in contiguous block lumps
6184 	 */
6185 	while (dblks > 0 && !error) {
6186 		lfsbn = (daddr_t)lblkno(fs, ldbtob(ldbn));
6187 		dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks;
6188 		nfsbs = 1;
6189 		ndbs = disk_blks - ldbn % disk_blks;
6190 		while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn +
6191 		    nfsbs]) == dbn + ndbs) {
6192 			nfsbs++;
6193 			ndbs += disk_blks;
6194 		}
6195 		if (ndbs > dblks)
6196 			ndbs = dblks;
6197 		error = bdev_dump(ip->i_dev, addr, dbn, ndbs);
6198 		addr += ldbtob((offset_t)ndbs);
6199 		dblks -= ndbs;
6200 		ldbn += ndbs;
6201 	}
6202 	return (error);
6203 
6204 }
6205 
6206 /*
6207  * Prepare the file system before and after the dump operation.
6208  *
6209  * action = DUMP_ALLOC:
6210  * Preparation before dump, allocate dump private data structure
6211  * to hold all the direct and indirect block info for dump.
6212  *
6213  * action = DUMP_FREE:
6214  * Clean up after dump, deallocate the dump private data structure.
6215  *
6216  * action = DUMP_SCAN:
6217  * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6218  * if found, the starting file-relative DEV_BSIZE lbn is written
6219  * to *bklp; that lbn is intended for use with VOP_DUMP()
6220  */
6221 /*ARGSUSED*/
6222 static int
6223 ufs_dumpctl(vnode_t *vp, int action, offset_t *blkp, caller_context_t *ct)
6224 {
6225 	struct inode	*ip = VTOI(vp);
6226 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
6227 	struct fs	*fs;
6228 	daddr32_t	*dblk, *storeblk;
6229 	daddr32_t	*nextblk, *endblk;
6230 	struct buf	*bp;
6231 	int		i, entry, entries;
6232 	int		n, ncontig;
6233 
6234 	/*
6235 	 * check for forced unmount
6236 	 */
6237 	if (ufsvfsp == NULL)
6238 		return (EIO);
6239 
6240 	if (action == DUMP_ALLOC) {
6241 		/*
6242 		 * alloc and record dump_info
6243 		 */
6244 		if (dump_info != NULL)
6245 			return (EINVAL);
6246 
6247 		ASSERT(vp->v_type == VREG);
6248 		fs = ufsvfsp->vfs_fs;
6249 
6250 		rw_enter(&ip->i_contents, RW_READER);
6251 
6252 		if (bmap_has_holes(ip)) {
6253 			rw_exit(&ip->i_contents);
6254 			return (EFAULT);
6255 		}
6256 
6257 		/*
6258 		 * calculate and allocate space needed according to i_size
6259 		 */
6260 		entries = (int)lblkno(fs, blkroundup(fs, ip->i_size));
6261 		dump_info = kmem_alloc(sizeof (struct dump) +
6262 		    (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP);
6263 		if (dump_info == NULL) {
6264 			rw_exit(&ip->i_contents);
6265 			return (ENOMEM);
6266 		}
6267 
6268 		/* Start saving the info */
6269 		dump_info->fsbs = entries;
6270 		dump_info->ip = ip;
6271 		storeblk = &dump_info->dblk[0];
6272 
6273 		/* Direct Blocks */
6274 		for (entry = 0; entry < NDADDR && entry < entries; entry++)
6275 			*storeblk++ = ip->i_db[entry];
6276 
6277 		/* Indirect Blocks */
6278 		for (i = 0; i < NIADDR; i++) {
6279 			int error = 0;
6280 
6281 			bp = UFS_BREAD(ufsvfsp,
6282 			    ip->i_dev, fsbtodb(fs, ip->i_ib[i]), fs->fs_bsize);
6283 			if (bp->b_flags & B_ERROR)
6284 				error = EIO;
6285 			else {
6286 				dblk = bp->b_un.b_daddr;
6287 				if ((storeblk = save_dblks(ip, ufsvfsp,
6288 				    storeblk, dblk, i, entries)) == NULL)
6289 					error = EIO;
6290 			}
6291 
6292 			brelse(bp);
6293 
6294 			if (error != 0) {
6295 				kmem_free(dump_info, sizeof (struct dump) +
6296 				    (entries - 1) * sizeof (daddr32_t));
6297 				rw_exit(&ip->i_contents);
6298 				dump_info = NULL;
6299 				return (error);
6300 			}
6301 		}
6302 		/* and time stamp the information */
6303 		mutex_enter(&ip->i_tlock);
6304 		dump_info->time = ip->i_mtime;
6305 		mutex_exit(&ip->i_tlock);
6306 
6307 		rw_exit(&ip->i_contents);
6308 	} else if (action == DUMP_FREE) {
6309 		/*
6310 		 * free dump_info
6311 		 */
6312 		if (dump_info == NULL)
6313 			return (EINVAL);
6314 		entries = dump_info->fsbs - 1;
6315 		kmem_free(dump_info, sizeof (struct dump) +
6316 		    entries * sizeof (daddr32_t));
6317 		dump_info = NULL;
6318 	} else if (action == DUMP_SCAN) {
6319 		/*
6320 		 * scan dump_info
6321 		 */
6322 		if (dump_info == NULL)
6323 			return (EINVAL);
6324 
6325 		dblk = dump_info->dblk;
6326 		nextblk = dblk + 1;
6327 		endblk = dblk + dump_info->fsbs - 1;
6328 		fs = ufsvfsp->vfs_fs;
6329 		ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT);
6330 
6331 		/*
6332 		 * scan dblk[] entries; contig fs space is found when:
6333 		 * ((current blkno + frags per block) == next blkno)
6334 		 */
6335 		n = 0;
6336 		while (n < ncontig && dblk < endblk) {
6337 			if ((*dblk + fs->fs_frag) == *nextblk)
6338 				n++;
6339 			else
6340 				n = 0;
6341 			dblk++;
6342 			nextblk++;
6343 		}
6344 
6345 		/*
6346 		 * index is where size bytes of contig space begins;
6347 		 * conversion from index to the file's DEV_BSIZE lbn
6348 		 * is equivalent to:  (index * fs_bsize) / DEV_BSIZE
6349 		 */
6350 		if (n == ncontig) {
6351 			i = (dblk - dump_info->dblk) - ncontig;
6352 			*blkp = i << (fs->fs_bshift - DEV_BSHIFT);
6353 		} else
6354 			return (EFAULT);
6355 	}
6356 	return (0);
6357 }
6358 
6359 /*
6360  * Recursive helper function for ufs_dumpctl().  It follows the indirect file
6361  * system  blocks until it reaches the the disk block addresses, which are
6362  * then stored into the given buffer, storeblk.
6363  */
6364 static daddr32_t *
6365 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp,  daddr32_t *storeblk,
6366     daddr32_t *dblk, int level, int entries)
6367 {
6368 	struct fs	*fs = ufsvfsp->vfs_fs;
6369 	struct buf	*bp;
6370 	int		i;
6371 
6372 	if (level == 0) {
6373 		for (i = 0; i < NINDIR(fs); i++) {
6374 			if (storeblk - dump_info->dblk >= entries)
6375 				break;
6376 			*storeblk++ = dblk[i];
6377 		}
6378 		return (storeblk);
6379 	}
6380 	for (i = 0; i < NINDIR(fs); i++) {
6381 		if (storeblk - dump_info->dblk >= entries)
6382 			break;
6383 		bp = UFS_BREAD(ufsvfsp,
6384 		    ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize);
6385 		if (bp->b_flags & B_ERROR) {
6386 			brelse(bp);
6387 			return (NULL);
6388 		}
6389 		storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr,
6390 		    level - 1, entries);
6391 		brelse(bp);
6392 
6393 		if (storeblk == NULL)
6394 			return (NULL);
6395 	}
6396 	return (storeblk);
6397 }
6398 
6399 /* ARGSUSED */
6400 static int
6401 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag,
6402 	struct cred *cr, caller_context_t *ct)
6403 {
6404 	struct inode	*ip = VTOI(vp);
6405 	struct ulockfs	*ulp;
6406 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
6407 	ulong_t		vsa_mask = vsap->vsa_mask;
6408 	int		err = EINVAL;
6409 
6410 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6411 
6412 	/*
6413 	 * Only grab locks if needed - they're not needed to check vsa_mask
6414 	 * or if the mask contains no acl flags.
6415 	 */
6416 	if (vsa_mask != 0) {
6417 		if (err = ufs_lockfs_begin(ufsvfsp, &ulp,
6418 		    ULOCKFS_GETATTR_MASK))
6419 			return (err);
6420 
6421 		rw_enter(&ip->i_contents, RW_READER);
6422 		err = ufs_acl_get(ip, vsap, flag, cr);
6423 		rw_exit(&ip->i_contents);
6424 
6425 		if (ulp)
6426 			ufs_lockfs_end(ulp);
6427 	}
6428 	return (err);
6429 }
6430 
6431 /* ARGSUSED */
6432 static int
6433 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr,
6434 	caller_context_t *ct)
6435 {
6436 	struct inode	*ip = VTOI(vp);
6437 	struct ulockfs	*ulp = NULL;
6438 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
6439 	ulong_t		vsa_mask = vsap->vsa_mask;
6440 	int		err;
6441 	int		haverwlock = 1;
6442 	int		trans_size;
6443 	int		donetrans = 0;
6444 	int		retry = 1;
6445 
6446 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
6447 
6448 	/* Abort now if the request is either empty or invalid. */
6449 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6450 	if ((vsa_mask == 0) ||
6451 	    ((vsap->vsa_aclentp == NULL) &&
6452 	    (vsap->vsa_dfaclentp == NULL))) {
6453 		err = EINVAL;
6454 		goto out;
6455 	}
6456 
6457 	/*
6458 	 * Following convention, if this is a directory then we acquire the
6459 	 * inode's i_rwlock after starting a UFS logging transaction;
6460 	 * otherwise, we acquire it beforehand. Since we were called (and
6461 	 * must therefore return) with the lock held, we will have to drop it,
6462 	 * and later reacquire it, if operating on a directory.
6463 	 */
6464 	if (vp->v_type == VDIR) {
6465 		rw_exit(&ip->i_rwlock);
6466 		haverwlock = 0;
6467 	} else {
6468 		/* Upgrade the lock if required. */
6469 		if (!rw_write_held(&ip->i_rwlock)) {
6470 			rw_exit(&ip->i_rwlock);
6471 			rw_enter(&ip->i_rwlock, RW_WRITER);
6472 		}
6473 	}
6474 
6475 again:
6476 	ASSERT(!(vp->v_type == VDIR && haverwlock));
6477 	if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) {
6478 		ulp = NULL;
6479 		retry = 0;
6480 		goto out;
6481 	}
6482 
6483 	/*
6484 	 * Check that the file system supports this operation. Note that
6485 	 * ufs_lockfs_begin() will have checked that the file system had
6486 	 * not been forcibly unmounted.
6487 	 */
6488 	if (ufsvfsp->vfs_fs->fs_ronly) {
6489 		err = EROFS;
6490 		goto out;
6491 	}
6492 	if (ufsvfsp->vfs_nosetsec) {
6493 		err = ENOSYS;
6494 		goto out;
6495 	}
6496 
6497 	if (ulp) {
6498 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR,
6499 		    trans_size = TOP_SETSECATTR_SIZE(VTOI(vp)));
6500 		donetrans = 1;
6501 	}
6502 
6503 	if (vp->v_type == VDIR) {
6504 		rw_enter(&ip->i_rwlock, RW_WRITER);
6505 		haverwlock = 1;
6506 	}
6507 
6508 	ASSERT(haverwlock);
6509 
6510 	/* Do the actual work. */
6511 	rw_enter(&ip->i_contents, RW_WRITER);
6512 	/*
6513 	 * Suppress out of inodes messages if we will retry.
6514 	 */
6515 	if (retry)
6516 		ip->i_flag |= IQUIET;
6517 	err = ufs_acl_set(ip, vsap, flag, cr);
6518 	ip->i_flag &= ~IQUIET;
6519 	rw_exit(&ip->i_contents);
6520 
6521 out:
6522 	if (ulp) {
6523 		if (donetrans) {
6524 			/*
6525 			 * top_end_async() can eventually call
6526 			 * top_end_sync(), which can block. We must
6527 			 * therefore observe the lock-ordering protocol
6528 			 * here as well.
6529 			 */
6530 			if (vp->v_type == VDIR) {
6531 				rw_exit(&ip->i_rwlock);
6532 				haverwlock = 0;
6533 			}
6534 			TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size);
6535 		}
6536 		ufs_lockfs_end(ulp);
6537 	}
6538 	/*
6539 	 * If no inodes available, try scaring a logically-
6540 	 * free one out of the delete queue to someplace
6541 	 * that we can find it.
6542 	 */
6543 	if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
6544 		ufs_delete_drain_wait(ufsvfsp, 1);
6545 		retry = 0;
6546 		if (vp->v_type == VDIR && haverwlock) {
6547 			rw_exit(&ip->i_rwlock);
6548 			haverwlock = 0;
6549 		}
6550 		goto again;
6551 	}
6552 	/*
6553 	 * If we need to reacquire the lock then it is safe to do so
6554 	 * as a reader. This is because ufs_rwunlock(), which will be
6555 	 * called by our caller after we return, does not differentiate
6556 	 * between shared and exclusive locks.
6557 	 */
6558 	if (!haverwlock) {
6559 		ASSERT(vp->v_type == VDIR);
6560 		rw_enter(&ip->i_rwlock, RW_READER);
6561 	}
6562 
6563 	return (err);
6564 }
6565