xref: /illumos-gate/usr/src/uts/common/fs/ufs/ufs_vnops.c (revision fb3b6bbf2c922827ddf076fbf2172d96c58a71de)
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 (c) 1984, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2017 Joyent, Inc.
25  * Copyright (c) 2016 by Delphix. All rights reserved.
26  */
27 
28 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
29 /*	  All Rights Reserved  	*/
30 
31 /*
32  * Portions of this source code were derived from Berkeley 4.3 BSD
33  * under license from the Regents of the University of California.
34  */
35 
36 #include <sys/types.h>
37 #include <sys/t_lock.h>
38 #include <sys/ksynch.h>
39 #include <sys/param.h>
40 #include <sys/time.h>
41 #include <sys/systm.h>
42 #include <sys/sysmacros.h>
43 #include <sys/resource.h>
44 #include <sys/signal.h>
45 #include <sys/cred.h>
46 #include <sys/user.h>
47 #include <sys/buf.h>
48 #include <sys/vfs.h>
49 #include <sys/vfs_opreg.h>
50 #include <sys/vnode.h>
51 #include <sys/proc.h>
52 #include <sys/disp.h>
53 #include <sys/file.h>
54 #include <sys/fcntl.h>
55 #include <sys/flock.h>
56 #include <sys/atomic.h>
57 #include <sys/kmem.h>
58 #include <sys/uio.h>
59 #include <sys/dnlc.h>
60 #include <sys/conf.h>
61 #include <sys/mman.h>
62 #include <sys/pathname.h>
63 #include <sys/debug.h>
64 #include <sys/vmsystm.h>
65 #include <sys/cmn_err.h>
66 #include <sys/filio.h>
67 #include <sys/policy.h>
68 
69 #include <sys/fs/ufs_fs.h>
70 #include <sys/fs/ufs_lockfs.h>
71 #include <sys/fs/ufs_filio.h>
72 #include <sys/fs/ufs_inode.h>
73 #include <sys/fs/ufs_fsdir.h>
74 #include <sys/fs/ufs_quota.h>
75 #include <sys/fs/ufs_log.h>
76 #include <sys/fs/ufs_snap.h>
77 #include <sys/fs/ufs_trans.h>
78 #include <sys/fs/ufs_panic.h>
79 #include <sys/fs/ufs_bio.h>
80 #include <sys/dirent.h>		/* must be AFTER <sys/fs/fsdir.h>! */
81 #include <sys/errno.h>
82 #include <sys/fssnap_if.h>
83 #include <sys/unistd.h>
84 #include <sys/sunddi.h>
85 
86 #include <sys/filio.h>		/* _FIOIO */
87 
88 #include <vm/hat.h>
89 #include <vm/page.h>
90 #include <vm/pvn.h>
91 #include <vm/as.h>
92 #include <vm/seg.h>
93 #include <vm/seg_map.h>
94 #include <vm/seg_vn.h>
95 #include <vm/seg_kmem.h>
96 #include <vm/rm.h>
97 #include <sys/swap.h>
98 
99 #include <fs/fs_subr.h>
100 
101 #include <sys/fs/decomp.h>
102 
103 static struct instats ins;
104 
105 static 	int ufs_getpage_ra(struct vnode *, u_offset_t, struct seg *, caddr_t);
106 static	int ufs_getpage_miss(struct vnode *, u_offset_t, size_t, struct seg *,
107 		caddr_t, struct page **, size_t, enum seg_rw, int);
108 static	int ufs_open(struct vnode **, int, struct cred *, caller_context_t *);
109 static	int ufs_close(struct vnode *, int, int, offset_t, struct cred *,
110 		caller_context_t *);
111 static	int ufs_read(struct vnode *, struct uio *, int, struct cred *,
112 		struct caller_context *);
113 static	int ufs_write(struct vnode *, struct uio *, int, struct cred *,
114 		struct caller_context *);
115 static	int ufs_ioctl(struct vnode *, int, intptr_t, int, struct cred *,
116 		int *, caller_context_t *);
117 static	int ufs_getattr(struct vnode *, struct vattr *, int, struct cred *,
118 		caller_context_t *);
119 static	int ufs_setattr(struct vnode *, struct vattr *, int, struct cred *,
120 		caller_context_t *);
121 static	int ufs_access(struct vnode *, int, int, struct cred *,
122 		caller_context_t *);
123 static	int ufs_lookup(struct vnode *, char *, struct vnode **,
124 		struct pathname *, int, struct vnode *, struct cred *,
125 		caller_context_t *, int *, pathname_t *);
126 static	int ufs_create(struct vnode *, char *, struct vattr *, enum vcexcl,
127 		int, struct vnode **, struct cred *, int,
128 		caller_context_t *, vsecattr_t  *);
129 static	int ufs_remove(struct vnode *, char *, struct cred *,
130 		caller_context_t *, int);
131 static	int ufs_link(struct vnode *, struct vnode *, char *, struct cred *,
132 		caller_context_t *, int);
133 static	int ufs_rename(struct vnode *, char *, struct vnode *, char *,
134 		struct cred *, caller_context_t *, int);
135 static	int ufs_mkdir(struct vnode *, char *, struct vattr *, struct vnode **,
136 		struct cred *, caller_context_t *, int, vsecattr_t *);
137 static	int ufs_rmdir(struct vnode *, char *, struct vnode *, struct cred *,
138 		caller_context_t *, int);
139 static	int ufs_readdir(struct vnode *, struct uio *, struct cred *, int *,
140 		caller_context_t *, int);
141 static	int ufs_symlink(struct vnode *, char *, struct vattr *, char *,
142 		struct cred *, caller_context_t *, int);
143 static	int ufs_readlink(struct vnode *, struct uio *, struct cred *,
144 		caller_context_t *);
145 static	int ufs_fsync(struct vnode *, int, struct cred *, caller_context_t *);
146 static	void ufs_inactive(struct vnode *, struct cred *, caller_context_t *);
147 static	int ufs_fid(struct vnode *, struct fid *, caller_context_t *);
148 static	int ufs_rwlock(struct vnode *, int, caller_context_t *);
149 static	void ufs_rwunlock(struct vnode *, int, caller_context_t *);
150 static	int ufs_seek(struct vnode *, offset_t, offset_t *, caller_context_t *);
151 static	int ufs_frlock(struct vnode *, int, struct flock64 *, int, offset_t,
152 		struct flk_callback *, struct cred *,
153 		caller_context_t *);
154 static  int ufs_space(struct vnode *, int, struct flock64 *, int, offset_t,
155 		cred_t *, caller_context_t *);
156 static	int ufs_getpage(struct vnode *, offset_t, size_t, uint_t *,
157 		struct page **, size_t, struct seg *, caddr_t,
158 		enum seg_rw, struct cred *, caller_context_t *);
159 static	int ufs_putpage(struct vnode *, offset_t, size_t, int, struct cred *,
160 		caller_context_t *);
161 static	int ufs_putpages(struct vnode *, offset_t, size_t, int, struct cred *);
162 static	int ufs_map(struct vnode *, offset_t, struct as *, caddr_t *, size_t,
163 		uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
164 static	int ufs_addmap(struct vnode *, offset_t, struct as *, caddr_t,  size_t,
165 		uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
166 static	int ufs_delmap(struct vnode *, offset_t, struct as *, caddr_t,  size_t,
167 		uint_t, uint_t, uint_t, struct cred *, caller_context_t *);
168 static	int ufs_poll(vnode_t *, short, int, short *, struct pollhead **,
169 		caller_context_t *);
170 static	int ufs_dump(vnode_t *, caddr_t, offset_t, offset_t,
171     caller_context_t *);
172 static	int ufs_l_pathconf(struct vnode *, int, ulong_t *, struct cred *,
173 		caller_context_t *);
174 static	int ufs_pageio(struct vnode *, struct page *, u_offset_t, size_t, int,
175 		struct cred *, caller_context_t *);
176 static	int ufs_dumpctl(vnode_t *, int, offset_t *, caller_context_t *);
177 static	daddr32_t *save_dblks(struct inode *, struct ufsvfs *, daddr32_t *,
178 		daddr32_t *, int, int);
179 static	int ufs_getsecattr(struct vnode *, vsecattr_t *, int, struct cred *,
180 		caller_context_t *);
181 static	int ufs_setsecattr(struct vnode *, vsecattr_t *, int, struct cred *,
182 		caller_context_t *);
183 static	int ufs_priv_access(void *, int, struct cred *);
184 static	int ufs_eventlookup(struct vnode *, char *, struct cred *,
185     struct vnode **);
186 extern int as_map_locked(struct as *, caddr_t, size_t, int ((*)()), void *);
187 
188 /*
189  * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
190  *
191  * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
192  */
193 struct vnodeops *ufs_vnodeops;
194 
195 /* NOTE: "not blkd" below  means that the operation isn't blocked by lockfs */
196 const fs_operation_def_t ufs_vnodeops_template[] = {
197 	VOPNAME_OPEN,		{ .vop_open = ufs_open },	/* not blkd */
198 	VOPNAME_CLOSE,		{ .vop_close = ufs_close },	/* not blkd */
199 	VOPNAME_READ,		{ .vop_read = ufs_read },
200 	VOPNAME_WRITE,		{ .vop_write = ufs_write },
201 	VOPNAME_IOCTL,		{ .vop_ioctl = ufs_ioctl },
202 	VOPNAME_GETATTR,	{ .vop_getattr = ufs_getattr },
203 	VOPNAME_SETATTR,	{ .vop_setattr = ufs_setattr },
204 	VOPNAME_ACCESS,		{ .vop_access = ufs_access },
205 	VOPNAME_LOOKUP,		{ .vop_lookup = ufs_lookup },
206 	VOPNAME_CREATE,		{ .vop_create = ufs_create },
207 	VOPNAME_REMOVE,		{ .vop_remove = ufs_remove },
208 	VOPNAME_LINK,		{ .vop_link = ufs_link },
209 	VOPNAME_RENAME,		{ .vop_rename = ufs_rename },
210 	VOPNAME_MKDIR,		{ .vop_mkdir = ufs_mkdir },
211 	VOPNAME_RMDIR,		{ .vop_rmdir = ufs_rmdir },
212 	VOPNAME_READDIR,	{ .vop_readdir = ufs_readdir },
213 	VOPNAME_SYMLINK,	{ .vop_symlink = ufs_symlink },
214 	VOPNAME_READLINK,	{ .vop_readlink = ufs_readlink },
215 	VOPNAME_FSYNC,		{ .vop_fsync = ufs_fsync },
216 	VOPNAME_INACTIVE,	{ .vop_inactive = ufs_inactive }, /* not blkd */
217 	VOPNAME_FID,		{ .vop_fid = ufs_fid },
218 	VOPNAME_RWLOCK,		{ .vop_rwlock = ufs_rwlock },	/* not blkd */
219 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = ufs_rwunlock }, /* not blkd */
220 	VOPNAME_SEEK,		{ .vop_seek = ufs_seek },
221 	VOPNAME_FRLOCK,		{ .vop_frlock = ufs_frlock },
222 	VOPNAME_SPACE,		{ .vop_space = ufs_space },
223 	VOPNAME_GETPAGE,	{ .vop_getpage = ufs_getpage },
224 	VOPNAME_PUTPAGE,	{ .vop_putpage = ufs_putpage },
225 	VOPNAME_MAP,		{ .vop_map = ufs_map },
226 	VOPNAME_ADDMAP,		{ .vop_addmap = ufs_addmap },	/* not blkd */
227 	VOPNAME_DELMAP,		{ .vop_delmap = ufs_delmap },	/* not blkd */
228 	VOPNAME_POLL,		{ .vop_poll = ufs_poll },	/* not blkd */
229 	VOPNAME_DUMP,		{ .vop_dump = ufs_dump },
230 	VOPNAME_PATHCONF,	{ .vop_pathconf = ufs_l_pathconf },
231 	VOPNAME_PAGEIO,		{ .vop_pageio = ufs_pageio },
232 	VOPNAME_DUMPCTL,	{ .vop_dumpctl = ufs_dumpctl },
233 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = ufs_getsecattr },
234 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = ufs_setsecattr },
235 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
236 	NULL,			NULL
237 };
238 
239 #define	MAX_BACKFILE_COUNT	9999
240 
241 /*
242  * Created by ufs_dumpctl() to store a file's disk block info into memory.
243  * Used by ufs_dump() to dump data to disk directly.
244  */
245 struct dump {
246 	struct inode	*ip;		/* the file we contain */
247 	daddr_t		fsbs;		/* number of blocks stored */
248 	struct timeval32 time;		/* time stamp for the struct */
249 	daddr32_t 	dblk[1];	/* place holder for block info */
250 };
251 
252 static struct dump *dump_info = NULL;
253 
254 /*
255  * Previously there was no special action required for ordinary files.
256  * (Devices are handled through the device file system.)
257  * Now we support Large Files and Large File API requires open to
258  * fail if file is large.
259  * We could take care to prevent data corruption
260  * by doing an atomic check of size and truncate if file is opened with
261  * FTRUNC flag set but traditionally this is being done by the vfs/vnode
262  * layers. So taking care of truncation here is a change in the existing
263  * semantics of VOP_OPEN and therefore we chose not to implement any thing
264  * here. The check for the size of the file > 2GB is being done at the
265  * vfs layer in routine vn_open().
266  */
267 
268 /* ARGSUSED */
269 static int
270 ufs_open(struct vnode **vpp, int flag, struct cred *cr, caller_context_t *ct)
271 {
272 	return (0);
273 }
274 
275 /*ARGSUSED*/
276 static int
277 ufs_close(struct vnode *vp, int flag, int count, offset_t offset,
278     struct cred *cr, caller_context_t *ct)
279 {
280 	cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
281 	cleanshares(vp, ttoproc(curthread)->p_pid);
282 
283 	/*
284 	 * Push partially filled cluster at last close.
285 	 * ``last close'' is approximated because the dnlc
286 	 * may have a hold on the vnode.
287 	 * Checking for VBAD here will also act as a forced umount check.
288 	 */
289 	if (vp->v_count <= 2 && vp->v_type != VBAD) {
290 		struct inode *ip = VTOI(vp);
291 		if (ip->i_delaylen) {
292 			ins.in_poc.value.ul++;
293 			(void) ufs_putpages(vp, ip->i_delayoff, ip->i_delaylen,
294 			    B_ASYNC | B_FREE, cr);
295 			ip->i_delaylen = 0;
296 		}
297 	}
298 
299 	return (0);
300 }
301 
302 /*ARGSUSED*/
303 static int
304 ufs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cr,
305     struct caller_context *ct)
306 {
307 	struct inode *ip = VTOI(vp);
308 	struct ufsvfs *ufsvfsp;
309 	struct ulockfs *ulp = NULL;
310 	int error = 0;
311 	int intrans = 0;
312 
313 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
314 
315 	/*
316 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
317 	 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
318 	 */
319 	if (MANDLOCK(vp, ip->i_mode)) {
320 		/*
321 		 * ufs_getattr ends up being called by chklock
322 		 */
323 		error = chklock(vp, FREAD, uiop->uio_loffset,
324 		    uiop->uio_resid, uiop->uio_fmode, ct);
325 		if (error)
326 			goto out;
327 	}
328 
329 	ufsvfsp = ip->i_ufsvfs;
330 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READ_MASK);
331 	if (error)
332 		goto out;
333 
334 	/*
335 	 * In the case that a directory is opened for reading as a file
336 	 * (eg "cat .") with the  O_RSYNC, O_SYNC and O_DSYNC flags set.
337 	 * The locking order had to be changed to avoid a deadlock with
338 	 * an update taking place on that directory at the same time.
339 	 */
340 	if ((ip->i_mode & IFMT) == IFDIR) {
341 
342 		rw_enter(&ip->i_contents, RW_READER);
343 		error = rdip(ip, uiop, ioflag, cr);
344 		rw_exit(&ip->i_contents);
345 
346 		if (error) {
347 			if (ulp)
348 				ufs_lockfs_end(ulp);
349 			goto out;
350 		}
351 
352 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
353 		    TRANS_ISTRANS(ufsvfsp)) {
354 			rw_exit(&ip->i_rwlock);
355 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
356 			    error);
357 			ASSERT(!error);
358 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
359 			    TOP_READ_SIZE);
360 			rw_enter(&ip->i_rwlock, RW_READER);
361 		}
362 	} else {
363 		/*
364 		 * Only transact reads to files opened for sync-read and
365 		 * sync-write on a file system that is not write locked.
366 		 *
367 		 * The ``not write locked'' check prevents problems with
368 		 * enabling/disabling logging on a busy file system.  E.g.,
369 		 * logging exists at the beginning of the read but does not
370 		 * at the end.
371 		 *
372 		 */
373 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
374 		    TRANS_ISTRANS(ufsvfsp)) {
375 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
376 			    error);
377 			ASSERT(!error);
378 			intrans = 1;
379 		}
380 
381 		rw_enter(&ip->i_contents, RW_READER);
382 		error = rdip(ip, uiop, ioflag, cr);
383 		rw_exit(&ip->i_contents);
384 
385 		if (intrans) {
386 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
387 			    TOP_READ_SIZE);
388 		}
389 	}
390 
391 	if (ulp) {
392 		ufs_lockfs_end(ulp);
393 	}
394 out:
395 
396 	return (error);
397 }
398 
399 extern	int	ufs_HW;		/* high water mark */
400 extern	int	ufs_LW;		/* low water mark */
401 int	ufs_WRITES = 1;		/* XXX - enable/disable */
402 int	ufs_throttles = 0;	/* throttling count */
403 int	ufs_allow_shared_writes = 1;	/* directio shared writes */
404 
405 static int
406 ufs_check_rewrite(struct inode *ip, struct uio *uiop, int ioflag)
407 {
408 	int	shared_write;
409 
410 	/*
411 	 * If the FDSYNC flag is set then ignore the global
412 	 * ufs_allow_shared_writes in this case.
413 	 */
414 	shared_write = (ioflag & FDSYNC) | ufs_allow_shared_writes;
415 
416 	/*
417 	 * Filter to determine if this request is suitable as a
418 	 * concurrent rewrite. This write must not allocate blocks
419 	 * by extending the file or filling in holes. No use trying
420 	 * through FSYNC descriptors as the inode will be synchronously
421 	 * updated after the write. The uio structure has not yet been
422 	 * checked for sanity, so assume nothing.
423 	 */
424 	return (((ip->i_mode & IFMT) == IFREG) && !(ioflag & FAPPEND) &&
425 	    (uiop->uio_loffset >= (offset_t)0) &&
426 	    (uiop->uio_loffset < ip->i_size) && (uiop->uio_resid > 0) &&
427 	    ((ip->i_size - uiop->uio_loffset) >= uiop->uio_resid) &&
428 	    !(ioflag & FSYNC) && !bmap_has_holes(ip) &&
429 	    shared_write);
430 }
431 
432 /*ARGSUSED*/
433 static int
434 ufs_write(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cr,
435     caller_context_t *ct)
436 {
437 	struct inode *ip = VTOI(vp);
438 	struct ufsvfs *ufsvfsp;
439 	struct ulockfs *ulp;
440 	int retry = 1;
441 	int error, resv, resid = 0;
442 	int directio_status;
443 	int exclusive;
444 	int rewriteflg;
445 	long start_resid = uiop->uio_resid;
446 
447 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
448 
449 retry_mandlock:
450 	/*
451 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
452 	 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
453 	 * Check for forced unmounts normally done in ufs_lockfs_begin().
454 	 */
455 	if ((ufsvfsp = ip->i_ufsvfs) == NULL) {
456 		error = EIO;
457 		goto out;
458 	}
459 	if (MANDLOCK(vp, ip->i_mode)) {
460 
461 		ASSERT(RW_WRITE_HELD(&ip->i_rwlock));
462 
463 		/*
464 		 * ufs_getattr ends up being called by chklock
465 		 */
466 		error = chklock(vp, FWRITE, uiop->uio_loffset,
467 		    uiop->uio_resid, uiop->uio_fmode, ct);
468 		if (error)
469 			goto out;
470 	}
471 
472 	/* i_rwlock can change in chklock */
473 	exclusive = rw_write_held(&ip->i_rwlock);
474 	rewriteflg = ufs_check_rewrite(ip, uiop, ioflag);
475 
476 	/*
477 	 * Check for fast-path special case of directio re-writes.
478 	 */
479 	if ((ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) &&
480 	    !exclusive && rewriteflg) {
481 
482 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
483 		if (error)
484 			goto out;
485 
486 		rw_enter(&ip->i_contents, RW_READER);
487 		error = ufs_directio_write(ip, uiop, ioflag, 1, cr,
488 		    &directio_status);
489 		if (directio_status == DIRECTIO_SUCCESS) {
490 			uint_t i_flag_save;
491 
492 			if (start_resid != uiop->uio_resid)
493 				error = 0;
494 			/*
495 			 * Special treatment of access times for re-writes.
496 			 * If IMOD is not already set, then convert it
497 			 * to IMODACC for this operation. This defers
498 			 * entering a delta into the log until the inode
499 			 * is flushed. This mimics what is done for read
500 			 * operations and inode access time.
501 			 */
502 			mutex_enter(&ip->i_tlock);
503 			i_flag_save = ip->i_flag;
504 			ip->i_flag |= IUPD | ICHG;
505 			ip->i_seq++;
506 			ITIMES_NOLOCK(ip);
507 			if ((i_flag_save & IMOD) == 0) {
508 				ip->i_flag &= ~IMOD;
509 				ip->i_flag |= IMODACC;
510 			}
511 			mutex_exit(&ip->i_tlock);
512 			rw_exit(&ip->i_contents);
513 			if (ulp)
514 				ufs_lockfs_end(ulp);
515 			goto out;
516 		}
517 		rw_exit(&ip->i_contents);
518 		if (ulp)
519 			ufs_lockfs_end(ulp);
520 	}
521 
522 	if (!exclusive && !rw_tryupgrade(&ip->i_rwlock)) {
523 		rw_exit(&ip->i_rwlock);
524 		rw_enter(&ip->i_rwlock, RW_WRITER);
525 		/*
526 		 * Mandatory locking could have been enabled
527 		 * after dropping the i_rwlock.
528 		 */
529 		if (MANDLOCK(vp, ip->i_mode))
530 			goto retry_mandlock;
531 	}
532 
533 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
534 	if (error)
535 		goto out;
536 
537 	/*
538 	 * Amount of log space needed for this write
539 	 */
540 	if (!rewriteflg || !(ioflag & FDSYNC))
541 		TRANS_WRITE_RESV(ip, uiop, ulp, &resv, &resid);
542 
543 	/*
544 	 * Throttle writes.
545 	 */
546 	if (ufs_WRITES && (ip->i_writes > ufs_HW)) {
547 		mutex_enter(&ip->i_tlock);
548 		while (ip->i_writes > ufs_HW) {
549 			ufs_throttles++;
550 			cv_wait(&ip->i_wrcv, &ip->i_tlock);
551 		}
552 		mutex_exit(&ip->i_tlock);
553 	}
554 
555 	/*
556 	 * Enter Transaction
557 	 *
558 	 * If the write is a rewrite there is no need to open a transaction
559 	 * if the FDSYNC flag is set and not the FSYNC.  In this case just
560 	 * set the IMODACC flag to modify do the update at a later time
561 	 * thus avoiding the overhead of the logging transaction that is
562 	 * not required.
563 	 */
564 	if (ioflag & (FSYNC|FDSYNC)) {
565 		if (ulp) {
566 			if (rewriteflg) {
567 				uint_t i_flag_save;
568 
569 				rw_enter(&ip->i_contents, RW_READER);
570 				mutex_enter(&ip->i_tlock);
571 				i_flag_save = ip->i_flag;
572 				ip->i_flag |= IUPD | ICHG;
573 				ip->i_seq++;
574 				ITIMES_NOLOCK(ip);
575 				if ((i_flag_save & IMOD) == 0) {
576 					ip->i_flag &= ~IMOD;
577 					ip->i_flag |= IMODACC;
578 				}
579 				mutex_exit(&ip->i_tlock);
580 				rw_exit(&ip->i_contents);
581 			} else {
582 				int terr = 0;
583 				TRANS_BEGIN_SYNC(ufsvfsp, TOP_WRITE_SYNC, resv,
584 				    terr);
585 				ASSERT(!terr);
586 			}
587 		}
588 	} else {
589 		if (ulp)
590 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_WRITE, resv);
591 	}
592 
593 	/*
594 	 * Write the file
595 	 */
596 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
597 	rw_enter(&ip->i_contents, RW_WRITER);
598 	if ((ioflag & FAPPEND) != 0 && (ip->i_mode & IFMT) == IFREG) {
599 		/*
600 		 * In append mode start at end of file.
601 		 */
602 		uiop->uio_loffset = ip->i_size;
603 	}
604 
605 	/*
606 	 * Mild optimisation, don't call ufs_trans_write() unless we have to
607 	 * Also, suppress file system full messages if we will retry.
608 	 */
609 	if (retry)
610 		ip->i_flag |= IQUIET;
611 	if (resid) {
612 		TRANS_WRITE(ip, uiop, ioflag, error, ulp, cr, resv, resid);
613 	} else {
614 		error = wrip(ip, uiop, ioflag, cr);
615 	}
616 	ip->i_flag &= ~IQUIET;
617 
618 	rw_exit(&ip->i_contents);
619 	rw_exit(&ufsvfsp->vfs_dqrwlock);
620 
621 	/*
622 	 * Leave Transaction
623 	 */
624 	if (ulp) {
625 		if (ioflag & (FSYNC|FDSYNC)) {
626 			if (!rewriteflg) {
627 				int terr = 0;
628 
629 				TRANS_END_SYNC(ufsvfsp, terr, TOP_WRITE_SYNC,
630 				    resv);
631 				if (error == 0)
632 					error = terr;
633 			}
634 		} else {
635 			TRANS_END_ASYNC(ufsvfsp, TOP_WRITE, resv);
636 		}
637 		ufs_lockfs_end(ulp);
638 	}
639 out:
640 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
641 		/*
642 		 * Any blocks tied up in pending deletes?
643 		 */
644 		ufs_delete_drain_wait(ufsvfsp, 1);
645 		retry = 0;
646 		goto retry_mandlock;
647 	}
648 
649 	if (error == ENOSPC && (start_resid != uiop->uio_resid))
650 		error = 0;
651 
652 	return (error);
653 }
654 
655 /*
656  * Don't cache write blocks to files with the sticky bit set.
657  * Used to keep swap files from blowing the page cache on a server.
658  */
659 int stickyhack = 1;
660 
661 /*
662  * Free behind hacks.  The pager is busted.
663  * XXX - need to pass the information down to writedone() in a flag like B_SEQ
664  * or B_FREE_IF_TIGHT_ON_MEMORY.
665  */
666 int	freebehind = 1;
667 int	smallfile = 0;
668 u_offset_t smallfile64 = 32 * 1024;
669 
670 /*
671  * While we should, in most cases, cache the pages for write, we
672  * may also want to cache the pages for read as long as they are
673  * frequently re-usable.
674  *
675  * If cache_read_ahead = 1, the pages for read will go to the tail
676  * of the cache list when they are released, otherwise go to the head.
677  */
678 int	cache_read_ahead = 0;
679 
680 /*
681  * Freebehind exists  so that as we read  large files  sequentially we
682  * don't consume most of memory with pages  from a few files. It takes
683  * longer to re-read from disk multiple small files as it does reading
684  * one large one sequentially.  As system  memory grows customers need
685  * to retain bigger chunks   of files in  memory.   The advent of  the
686  * cachelist opens up of the possibility freeing pages  to the head or
687  * tail of the list.
688  *
689  * Not freeing a page is a bet that the page will be read again before
690  * it's segmap slot is needed for something else. If we loose the bet,
691  * it means some  other thread is  burdened with the  page free we did
692  * not do. If we win we save a free and reclaim.
693  *
694  * Freeing it at the tail  vs the head of cachelist  is a bet that the
695  * page will survive until the next  read.  It's also saying that this
696  * page is more likely to  be re-used than a  page freed some time ago
697  * and never reclaimed.
698  *
699  * Freebehind maintains a  range of  file offset [smallfile1; smallfile2]
700  *
701  *            0 < offset < smallfile1 : pages are not freed.
702  *   smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
703  *   smallfile2 < offset              : pages freed to head of cachelist.
704  *
705  * The range  is  computed  at most  once  per second  and  depends on
706  * freemem  and  ncpus_online.  Both parameters  are   bounded to be
707  * >= smallfile && >= smallfile64.
708  *
709  * smallfile1 = (free memory / ncpu) / 1000
710  * smallfile2 = (free memory / ncpu) / 10
711  *
712  * A few examples values:
713  *
714  *       Free Mem (in Bytes) [smallfile1; smallfile2]  [smallfile1; smallfile2]
715  *                                 ncpus_online = 4          ncpus_online = 64
716  *       ------------------  -----------------------   -----------------------
717  *             1G                   [256K;  25M]               [32K; 1.5M]
718  *            10G                   [2.5M; 250M]              [156K; 15M]
719  *           100G                    [25M; 2.5G]              [1.5M; 150M]
720  *
721  */
722 
723 #define	SMALLFILE1_D 1000
724 #define	SMALLFILE2_D 10
725 static u_offset_t smallfile1 = 32 * 1024;
726 static u_offset_t smallfile2 = 32 * 1024;
727 static clock_t smallfile_update = 0; /* lbolt value of when to recompute */
728 uint_t smallfile1_d = SMALLFILE1_D;
729 uint_t smallfile2_d = SMALLFILE2_D;
730 
731 /*
732  * wrip does the real work of write requests for ufs.
733  */
734 int
735 wrip(struct inode *ip, struct uio *uio, int ioflag, struct cred *cr)
736 {
737 	rlim64_t limit = uio->uio_llimit;
738 	u_offset_t off;
739 	u_offset_t old_i_size;
740 	struct fs *fs;
741 	struct vnode *vp;
742 	struct ufsvfs *ufsvfsp;
743 	caddr_t base;
744 	long start_resid = uio->uio_resid;	/* save starting resid */
745 	long premove_resid;			/* resid before uiomove() */
746 	uint_t flags;
747 	int newpage;
748 	int iupdat_flag, directio_status;
749 	int n, on, mapon;
750 	int error, pagecreate;
751 	int do_dqrwlock;		/* drop/reacquire vfs_dqrwlock */
752 	int32_t	iblocks;
753 	int	new_iblocks;
754 
755 	/*
756 	 * ip->i_size is incremented before the uiomove
757 	 * is done on a write.  If the move fails (bad user
758 	 * address) reset ip->i_size.
759 	 * The better way would be to increment ip->i_size
760 	 * only if the uiomove succeeds.
761 	 */
762 	int i_size_changed = 0;
763 	o_mode_t type;
764 	int i_seq_needed = 0;
765 
766 	vp = ITOV(ip);
767 
768 	/*
769 	 * check for forced unmount - should not happen as
770 	 * the request passed the lockfs checks.
771 	 */
772 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
773 		return (EIO);
774 
775 	fs = ip->i_fs;
776 
777 	ASSERT(RW_WRITE_HELD(&ip->i_contents));
778 
779 	/* check for valid filetype */
780 	type = ip->i_mode & IFMT;
781 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
782 	    (type != IFLNK) && (type != IFSHAD)) {
783 		return (EIO);
784 	}
785 
786 	/*
787 	 * the actual limit of UFS file size
788 	 * is UFS_MAXOFFSET_T
789 	 */
790 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
791 		limit = MAXOFFSET_T;
792 
793 	if (uio->uio_loffset >= limit) {
794 		proc_t *p = ttoproc(curthread);
795 
796 		mutex_enter(&p->p_lock);
797 		(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
798 		    p, RCA_UNSAFE_SIGINFO);
799 		mutex_exit(&p->p_lock);
800 		return (EFBIG);
801 	}
802 
803 	/*
804 	 * if largefiles are disallowed, the limit is
805 	 * the pre-largefiles value of 2GB
806 	 */
807 	if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
808 		limit = MIN(UFS_MAXOFFSET_T, limit);
809 	else
810 		limit = MIN(MAXOFF32_T, limit);
811 
812 	if (uio->uio_loffset < (offset_t)0) {
813 		return (EINVAL);
814 	}
815 	if (uio->uio_resid == 0) {
816 		return (0);
817 	}
818 
819 	if (uio->uio_loffset >= limit)
820 		return (EFBIG);
821 
822 	ip->i_flag |= INOACC;	/* don't update ref time in getpage */
823 
824 	if (ioflag & (FSYNC|FDSYNC)) {
825 		ip->i_flag |= ISYNC;
826 		iupdat_flag = 1;
827 	}
828 	/*
829 	 * Try to go direct
830 	 */
831 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
832 		uio->uio_llimit = limit;
833 		error = ufs_directio_write(ip, uio, ioflag, 0, cr,
834 		    &directio_status);
835 		/*
836 		 * If ufs_directio wrote to the file or set the flags,
837 		 * we need to update i_seq, but it may be deferred.
838 		 */
839 		if (start_resid != uio->uio_resid ||
840 		    (ip->i_flag & (ICHG|IUPD))) {
841 			i_seq_needed = 1;
842 			ip->i_flag |= ISEQ;
843 		}
844 		if (directio_status == DIRECTIO_SUCCESS)
845 			goto out;
846 	}
847 
848 	/*
849 	 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
850 	 *
851 	 * o shadow inodes: vfs_dqrwlock is not held at all
852 	 * o quota updates: vfs_dqrwlock is read or write held
853 	 * o other updates: vfs_dqrwlock is read held
854 	 *
855 	 * The first case is the only one where we do not hold
856 	 * vfs_dqrwlock at all while entering wrip().
857 	 * We must make sure not to downgrade/drop vfs_dqrwlock if we
858 	 * have it as writer, i.e. if we are updating the quota inode.
859 	 * There is no potential deadlock scenario in this case as
860 	 * ufs_getpage() takes care of this and avoids reacquiring
861 	 * vfs_dqrwlock in that case.
862 	 *
863 	 * This check is done here since the above conditions do not change
864 	 * and we possibly loop below, so save a few cycles.
865 	 */
866 	if ((type == IFSHAD) ||
867 	    (rw_owner(&ufsvfsp->vfs_dqrwlock) == curthread)) {
868 		do_dqrwlock = 0;
869 	} else {
870 		do_dqrwlock = 1;
871 	}
872 
873 	/*
874 	 * Large Files: We cast MAXBMASK to offset_t
875 	 * inorder to mask out the higher bits. Since offset_t
876 	 * is a signed value, the high order bit set in MAXBMASK
877 	 * value makes it do the right thing by having all bits 1
878 	 * in the higher word. May be removed for _SOLARIS64_.
879 	 */
880 
881 	fs = ip->i_fs;
882 	do {
883 		u_offset_t uoff = uio->uio_loffset;
884 		off = uoff & (offset_t)MAXBMASK;
885 		mapon = (int)(uoff & (offset_t)MAXBOFFSET);
886 		on = (int)blkoff(fs, uoff);
887 		n = (int)MIN(fs->fs_bsize - on, uio->uio_resid);
888 		new_iblocks = 1;
889 
890 		if (type == IFREG && uoff + n >= limit) {
891 			if (uoff >= limit) {
892 				error = EFBIG;
893 				goto out;
894 			}
895 			/*
896 			 * since uoff + n >= limit,
897 			 * therefore n >= limit - uoff, and n is an int
898 			 * so it is safe to cast it to an int
899 			 */
900 			n = (int)(limit - (rlim64_t)uoff);
901 		}
902 		if (uoff + n > ip->i_size) {
903 			/*
904 			 * We are extending the length of the file.
905 			 * bmap is used so that we are sure that
906 			 * if we need to allocate new blocks, that it
907 			 * is done here before we up the file size.
908 			 */
909 			error = bmap_write(ip, uoff, (int)(on + n),
910 			    mapon == 0, NULL, cr);
911 			/*
912 			 * bmap_write never drops i_contents so if
913 			 * the flags are set it changed the file.
914 			 */
915 			if (ip->i_flag & (ICHG|IUPD)) {
916 				i_seq_needed = 1;
917 				ip->i_flag |= ISEQ;
918 			}
919 			if (error)
920 				break;
921 			/*
922 			 * There is a window of vulnerability here.
923 			 * The sequence of operations: allocate file
924 			 * system blocks, uiomove the data into pages,
925 			 * and then update the size of the file in the
926 			 * inode, must happen atomically.  However, due
927 			 * to current locking constraints, this can not
928 			 * be done.
929 			 */
930 			ASSERT(ip->i_writer == NULL);
931 			ip->i_writer = curthread;
932 			i_size_changed = 1;
933 			/*
934 			 * If we are writing from the beginning of
935 			 * the mapping, we can just create the
936 			 * pages without having to read them.
937 			 */
938 			pagecreate = (mapon == 0);
939 		} else if (n == MAXBSIZE) {
940 			/*
941 			 * Going to do a whole mappings worth,
942 			 * so we can just create the pages w/o
943 			 * having to read them in.  But before
944 			 * we do that, we need to make sure any
945 			 * needed blocks are allocated first.
946 			 */
947 			iblocks = ip->i_blocks;
948 			error = bmap_write(ip, uoff, (int)(on + n),
949 			    BI_ALLOC_ONLY, NULL, cr);
950 			/*
951 			 * bmap_write never drops i_contents so if
952 			 * the flags are set it changed the file.
953 			 */
954 			if (ip->i_flag & (ICHG|IUPD)) {
955 				i_seq_needed = 1;
956 				ip->i_flag |= ISEQ;
957 			}
958 			if (error)
959 				break;
960 			pagecreate = 1;
961 			/*
962 			 * check if the new created page needed the
963 			 * allocation of new disk blocks.
964 			 */
965 			if (iblocks == ip->i_blocks)
966 				new_iblocks = 0; /* no new blocks allocated */
967 		} else {
968 			pagecreate = 0;
969 			/*
970 			 * In sync mode flush the indirect blocks which
971 			 * may have been allocated and not written on
972 			 * disk. In above cases bmap_write will allocate
973 			 * in sync mode.
974 			 */
975 			if (ioflag & (FSYNC|FDSYNC)) {
976 				error = ufs_indirblk_sync(ip, uoff);
977 				if (error)
978 					break;
979 			}
980 		}
981 
982 		/*
983 		 * At this point we can enter ufs_getpage() in one
984 		 * of two ways:
985 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
986 		 *    forcefault parameter is true (pagecreate == 0)
987 		 * 2) uiomove() causes a page fault.
988 		 *
989 		 * We have to drop the contents lock to prevent the VM
990 		 * system from trying to reacquire it in ufs_getpage()
991 		 * should the uiomove cause a pagefault.
992 		 *
993 		 * We have to drop the reader vfs_dqrwlock here as well.
994 		 */
995 		rw_exit(&ip->i_contents);
996 		if (do_dqrwlock) {
997 			ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock));
998 			ASSERT(!(RW_WRITE_HELD(&ufsvfsp->vfs_dqrwlock)));
999 			rw_exit(&ufsvfsp->vfs_dqrwlock);
1000 		}
1001 
1002 		newpage = 0;
1003 		premove_resid = uio->uio_resid;
1004 
1005 		/*
1006 		 * Touch the page and fault it in if it is not in core
1007 		 * before segmap_getmapflt or vpm_data_copy can lock it.
1008 		 * This is to avoid the deadlock if the buffer is mapped
1009 		 * to the same file through mmap which we want to write.
1010 		 */
1011 		uio_prefaultpages((long)n, uio);
1012 
1013 		if (vpm_enable) {
1014 			/*
1015 			 * Copy data. If new pages are created, part of
1016 			 * the page that is not written will be initizliazed
1017 			 * with zeros.
1018 			 */
1019 			error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1020 			    uio, !pagecreate, &newpage, 0, S_WRITE);
1021 		} else {
1022 
1023 			base = segmap_getmapflt(segkmap, vp, (off + mapon),
1024 			    (uint_t)n, !pagecreate, S_WRITE);
1025 
1026 			/*
1027 			 * segmap_pagecreate() returns 1 if it calls
1028 			 * page_create_va() to allocate any pages.
1029 			 */
1030 
1031 			if (pagecreate)
1032 				newpage = segmap_pagecreate(segkmap, base,
1033 				    (size_t)n, 0);
1034 
1035 			error = uiomove(base + mapon, (long)n, UIO_WRITE, uio);
1036 		}
1037 
1038 		/*
1039 		 * If "newpage" is set, then a new page was created and it
1040 		 * does not contain valid data, so it needs to be initialized
1041 		 * at this point.
1042 		 * Otherwise the page contains old data, which was overwritten
1043 		 * partially or as a whole in uiomove.
1044 		 * If there is only one iovec structure within uio, then
1045 		 * on error uiomove will not be able to update uio->uio_loffset
1046 		 * and we would zero the whole page here!
1047 		 *
1048 		 * If uiomove fails because of an error, the old valid data
1049 		 * is kept instead of filling the rest of the page with zero's.
1050 		 */
1051 		if (!vpm_enable && newpage &&
1052 		    uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) {
1053 			/*
1054 			 * We created pages w/o initializing them completely,
1055 			 * thus we need to zero the part that wasn't set up.
1056 			 * This happens on most EOF write cases and if
1057 			 * we had some sort of error during the uiomove.
1058 			 */
1059 			int nzero, nmoved;
1060 
1061 			nmoved = (int)(uio->uio_loffset - (off + mapon));
1062 			ASSERT(nmoved >= 0 && nmoved <= n);
1063 			nzero = roundup(on + n, PAGESIZE) - nmoved;
1064 			ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE);
1065 			(void) kzero(base + mapon + nmoved, (uint_t)nzero);
1066 		}
1067 
1068 		/*
1069 		 * Unlock the pages allocated by page_create_va()
1070 		 * in segmap_pagecreate()
1071 		 */
1072 		if (!vpm_enable && newpage)
1073 			segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE);
1074 
1075 		/*
1076 		 * If the size of the file changed, then update the
1077 		 * size field in the inode now.  This can't be done
1078 		 * before the call to segmap_pageunlock or there is
1079 		 * a potential deadlock with callers to ufs_putpage().
1080 		 * They will be holding i_contents and trying to lock
1081 		 * a page, while this thread is holding a page locked
1082 		 * and trying to acquire i_contents.
1083 		 */
1084 		if (i_size_changed) {
1085 			rw_enter(&ip->i_contents, RW_WRITER);
1086 			old_i_size = ip->i_size;
1087 			UFS_SET_ISIZE(uoff + n, ip);
1088 			TRANS_INODE(ufsvfsp, ip);
1089 			/*
1090 			 * file has grown larger than 2GB. Set flag
1091 			 * in superblock to indicate this, if it
1092 			 * is not already set.
1093 			 */
1094 			if ((ip->i_size > MAXOFF32_T) &&
1095 			    !(fs->fs_flags & FSLARGEFILES)) {
1096 				ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES);
1097 				mutex_enter(&ufsvfsp->vfs_lock);
1098 				fs->fs_flags |= FSLARGEFILES;
1099 				ufs_sbwrite(ufsvfsp);
1100 				mutex_exit(&ufsvfsp->vfs_lock);
1101 			}
1102 			mutex_enter(&ip->i_tlock);
1103 			ip->i_writer = NULL;
1104 			cv_broadcast(&ip->i_wrcv);
1105 			mutex_exit(&ip->i_tlock);
1106 			rw_exit(&ip->i_contents);
1107 		}
1108 
1109 		if (error) {
1110 			/*
1111 			 * If we failed on a write, we may have already
1112 			 * allocated file blocks as well as pages.  It's
1113 			 * hard to undo the block allocation, but we must
1114 			 * be sure to invalidate any pages that may have
1115 			 * been allocated.
1116 			 *
1117 			 * If the page was created without initialization
1118 			 * then we must check if it should be possible
1119 			 * to destroy the new page and to keep the old data
1120 			 * on the disk.
1121 			 *
1122 			 * It is possible to destroy the page without
1123 			 * having to write back its contents only when
1124 			 * - the size of the file keeps unchanged
1125 			 * - bmap_write() did not allocate new disk blocks
1126 			 *   it is possible to create big files using "seek" and
1127 			 *   write to the end of the file. A "write" to a
1128 			 *   position before the end of the file would not
1129 			 *   change the size of the file but it would allocate
1130 			 *   new disk blocks.
1131 			 * - uiomove intended to overwrite the whole page.
1132 			 * - a new page was created (newpage == 1).
1133 			 */
1134 
1135 			if (i_size_changed == 0 && new_iblocks == 0 &&
1136 			    newpage) {
1137 
1138 				/* unwind what uiomove eventually last did */
1139 				uio->uio_resid = premove_resid;
1140 
1141 				/*
1142 				 * destroy the page, do not write ambiguous
1143 				 * data to the disk.
1144 				 */
1145 				flags = SM_DESTROY;
1146 			} else {
1147 				/*
1148 				 * write the page back to the disk, if dirty,
1149 				 * and remove the page from the cache.
1150 				 */
1151 				flags = SM_INVAL;
1152 			}
1153 
1154 			if (vpm_enable) {
1155 				/*
1156 				 *  Flush pages.
1157 				 */
1158 				(void) vpm_sync_pages(vp, off, n, flags);
1159 			} else {
1160 				(void) segmap_release(segkmap, base, flags);
1161 			}
1162 		} else {
1163 			flags = 0;
1164 			/*
1165 			 * Force write back for synchronous write cases.
1166 			 */
1167 			if ((ioflag & (FSYNC|FDSYNC)) || type == IFDIR) {
1168 				/*
1169 				 * If the sticky bit is set but the
1170 				 * execute bit is not set, we do a
1171 				 * synchronous write back and free
1172 				 * the page when done.  We set up swap
1173 				 * files to be handled this way to
1174 				 * prevent servers from keeping around
1175 				 * the client's swap pages too long.
1176 				 * XXX - there ought to be a better way.
1177 				 */
1178 				if (IS_SWAPVP(vp)) {
1179 					flags = SM_WRITE | SM_FREE |
1180 					    SM_DONTNEED;
1181 					iupdat_flag = 0;
1182 				} else {
1183 					flags = SM_WRITE;
1184 				}
1185 			} else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
1186 				/*
1187 				 * Have written a whole block.
1188 				 * Start an asynchronous write and
1189 				 * mark the buffer to indicate that
1190 				 * it won't be needed again soon.
1191 				 */
1192 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
1193 			}
1194 			if (vpm_enable) {
1195 				/*
1196 				 * Flush pages.
1197 				 */
1198 				error = vpm_sync_pages(vp, off, n, flags);
1199 			} else {
1200 				error = segmap_release(segkmap, base, flags);
1201 			}
1202 			/*
1203 			 * If the operation failed and is synchronous,
1204 			 * then we need to unwind what uiomove() last
1205 			 * did so we can potentially return an error to
1206 			 * the caller.  If this write operation was
1207 			 * done in two pieces and the first succeeded,
1208 			 * then we won't return an error for the second
1209 			 * piece that failed.  However, we only want to
1210 			 * return a resid value that reflects what was
1211 			 * really done.
1212 			 *
1213 			 * Failures for non-synchronous operations can
1214 			 * be ignored since the page subsystem will
1215 			 * retry the operation until it succeeds or the
1216 			 * file system is unmounted.
1217 			 */
1218 			if (error) {
1219 				if ((ioflag & (FSYNC | FDSYNC)) ||
1220 				    type == IFDIR) {
1221 					uio->uio_resid = premove_resid;
1222 				} else {
1223 					error = 0;
1224 				}
1225 			}
1226 		}
1227 
1228 		/*
1229 		 * Re-acquire contents lock.
1230 		 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1231 		 */
1232 		if (do_dqrwlock)
1233 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1234 		rw_enter(&ip->i_contents, RW_WRITER);
1235 
1236 		/*
1237 		 * If the uiomove() failed or if a synchronous
1238 		 * page push failed, fix up i_size.
1239 		 */
1240 		if (error) {
1241 			if (i_size_changed) {
1242 				/*
1243 				 * The uiomove failed, and we
1244 				 * allocated blocks,so get rid
1245 				 * of them.
1246 				 */
1247 				(void) ufs_itrunc(ip, old_i_size, 0, cr);
1248 			}
1249 		} else {
1250 			/*
1251 			 * XXX - Can this be out of the loop?
1252 			 */
1253 			ip->i_flag |= IUPD | ICHG;
1254 			/*
1255 			 * Only do one increase of i_seq for multiple
1256 			 * pieces.  Because we drop locks, record
1257 			 * the fact that we changed the timestamp and
1258 			 * are deferring the increase in case another thread
1259 			 * pushes our timestamp update.
1260 			 */
1261 			i_seq_needed = 1;
1262 			ip->i_flag |= ISEQ;
1263 			if (i_size_changed)
1264 				ip->i_flag |= IATTCHG;
1265 			if ((ip->i_mode & (IEXEC | (IEXEC >> 3) |
1266 			    (IEXEC >> 6))) != 0 &&
1267 			    (ip->i_mode & (ISUID | ISGID)) != 0 &&
1268 			    secpolicy_vnode_setid_retain(cr,
1269 			    (ip->i_mode & ISUID) != 0 && ip->i_uid == 0) != 0) {
1270 				/*
1271 				 * Clear Set-UID & Set-GID bits on
1272 				 * successful write if not privileged
1273 				 * and at least one of the execute bits
1274 				 * is set.  If we always clear Set-GID,
1275 				 * mandatory file and record locking is
1276 				 * unuseable.
1277 				 */
1278 				ip->i_mode &= ~(ISUID | ISGID);
1279 			}
1280 		}
1281 		/*
1282 		 * In the case the FDSYNC flag is set and this is a
1283 		 * "rewrite" we won't log a delta.
1284 		 * The FSYNC flag overrides all cases.
1285 		 */
1286 		if (!ufs_check_rewrite(ip, uio, ioflag) || !(ioflag & FDSYNC)) {
1287 			TRANS_INODE(ufsvfsp, ip);
1288 		}
1289 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1290 
1291 out:
1292 	/*
1293 	 * Make sure i_seq is increased at least once per write
1294 	 */
1295 	if (i_seq_needed) {
1296 		ip->i_seq++;
1297 		ip->i_flag &= ~ISEQ;	/* no longer deferred */
1298 	}
1299 
1300 	/*
1301 	 * Inode is updated according to this table -
1302 	 *
1303 	 *   FSYNC	  FDSYNC(posix.4)
1304 	 *   --------------------------
1305 	 *   always@	  IATTCHG|IBDWRITE
1306 	 *
1307 	 * @ - 	If we are doing synchronous write the only time we should
1308 	 *	not be sync'ing the ip here is if we have the stickyhack
1309 	 *	activated, the file is marked with the sticky bit and
1310 	 *	no exec bit, the file length has not been changed and
1311 	 *	no new blocks have been allocated during this write.
1312 	 */
1313 
1314 	if ((ip->i_flag & ISYNC) != 0) {
1315 		/*
1316 		 * we have eliminated nosync
1317 		 */
1318 		if ((ip->i_flag & (IATTCHG|IBDWRITE)) ||
1319 		    ((ioflag & FSYNC) && iupdat_flag)) {
1320 			ufs_iupdat(ip, 1);
1321 		}
1322 	}
1323 
1324 	/*
1325 	 * If we've already done a partial-write, terminate
1326 	 * the write but return no error unless the error is ENOSPC
1327 	 * because the caller can detect this and free resources and
1328 	 * try again.
1329 	 */
1330 	if ((start_resid != uio->uio_resid) && (error != ENOSPC))
1331 		error = 0;
1332 
1333 	ip->i_flag &= ~(INOACC | ISYNC);
1334 	ITIMES_NOLOCK(ip);
1335 	return (error);
1336 }
1337 
1338 /*
1339  * rdip does the real work of read requests for ufs.
1340  */
1341 int
1342 rdip(struct inode *ip, struct uio *uio, int ioflag, cred_t *cr)
1343 {
1344 	u_offset_t off;
1345 	caddr_t base;
1346 	struct fs *fs;
1347 	struct ufsvfs *ufsvfsp;
1348 	struct vnode *vp;
1349 	long oresid = uio->uio_resid;
1350 	u_offset_t n, on, mapon;
1351 	int error = 0;
1352 	int doupdate = 1;
1353 	uint_t flags;
1354 	int dofree, directio_status;
1355 	krw_t rwtype;
1356 	o_mode_t type;
1357 	clock_t	now;
1358 
1359 	vp = ITOV(ip);
1360 
1361 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
1362 
1363 	ufsvfsp = ip->i_ufsvfs;
1364 
1365 	if (ufsvfsp == NULL)
1366 		return (EIO);
1367 
1368 	fs = ufsvfsp->vfs_fs;
1369 
1370 	/* check for valid filetype */
1371 	type = ip->i_mode & IFMT;
1372 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
1373 	    (type != IFLNK) && (type != IFSHAD)) {
1374 		return (EIO);
1375 	}
1376 
1377 	if (uio->uio_loffset > UFS_MAXOFFSET_T) {
1378 		error = 0;
1379 		goto out;
1380 	}
1381 	if (uio->uio_loffset < (offset_t)0) {
1382 		return (EINVAL);
1383 	}
1384 	if (uio->uio_resid == 0) {
1385 		return (0);
1386 	}
1387 
1388 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (fs->fs_ronly == 0) &&
1389 	    (!ufsvfsp->vfs_noatime)) {
1390 		mutex_enter(&ip->i_tlock);
1391 		ip->i_flag |= IACC;
1392 		mutex_exit(&ip->i_tlock);
1393 	}
1394 	/*
1395 	 * Try to go direct
1396 	 */
1397 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
1398 		error = ufs_directio_read(ip, uio, cr, &directio_status);
1399 		if (directio_status == DIRECTIO_SUCCESS)
1400 			goto out;
1401 	}
1402 
1403 	rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER);
1404 
1405 	do {
1406 		offset_t diff;
1407 		u_offset_t uoff = uio->uio_loffset;
1408 		off = uoff & (offset_t)MAXBMASK;
1409 		mapon = (u_offset_t)(uoff & (offset_t)MAXBOFFSET);
1410 		on = (u_offset_t)blkoff(fs, uoff);
1411 		n = MIN((u_offset_t)fs->fs_bsize - on,
1412 		    (u_offset_t)uio->uio_resid);
1413 
1414 		diff = ip->i_size - uoff;
1415 
1416 		if (diff <= (offset_t)0) {
1417 			error = 0;
1418 			goto out;
1419 		}
1420 		if (diff < (offset_t)n)
1421 			n = (int)diff;
1422 
1423 		/*
1424 		 * We update smallfile2 and smallfile1 at most every second.
1425 		 */
1426 		now = ddi_get_lbolt();
1427 		if (now >= smallfile_update) {
1428 			uint64_t percpufreeb;
1429 			if (smallfile1_d == 0) smallfile1_d = SMALLFILE1_D;
1430 			if (smallfile2_d == 0) smallfile2_d = SMALLFILE2_D;
1431 			percpufreeb = ptob((uint64_t)freemem) / ncpus_online;
1432 			smallfile1 = percpufreeb / smallfile1_d;
1433 			smallfile2 = percpufreeb / smallfile2_d;
1434 			smallfile1 = MAX(smallfile1, smallfile);
1435 			smallfile1 = MAX(smallfile1, smallfile64);
1436 			smallfile2 = MAX(smallfile1, smallfile2);
1437 			smallfile_update = now + hz;
1438 		}
1439 
1440 		dofree = freebehind &&
1441 		    ip->i_nextr == (off & PAGEMASK) && off > smallfile1;
1442 
1443 		/*
1444 		 * At this point we can enter ufs_getpage() in one of two
1445 		 * ways:
1446 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
1447 		 *    forcefault parameter is true (value of 1 is passed)
1448 		 * 2) uiomove() causes a page fault.
1449 		 *
1450 		 * We cannot hold onto an i_contents reader lock without
1451 		 * risking deadlock in ufs_getpage() so drop a reader lock.
1452 		 * The ufs_getpage() dolock logic already allows for a
1453 		 * thread holding i_contents as writer to work properly
1454 		 * so we keep a writer lock.
1455 		 */
1456 		if (rwtype == RW_READER)
1457 			rw_exit(&ip->i_contents);
1458 
1459 		if (vpm_enable) {
1460 			/*
1461 			 * Copy data.
1462 			 */
1463 			error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1464 			    uio, 1, NULL, 0, S_READ);
1465 		} else {
1466 			base = segmap_getmapflt(segkmap, vp, (off + mapon),
1467 			    (uint_t)n, 1, S_READ);
1468 			error = uiomove(base + mapon, (long)n, UIO_READ, uio);
1469 		}
1470 
1471 		flags = 0;
1472 		if (!error) {
1473 			/*
1474 			 * If  reading sequential  we won't need  this
1475 			 * buffer again  soon.  For  offsets in  range
1476 			 * [smallfile1,  smallfile2] release the pages
1477 			 * at   the  tail  of the   cache list, larger
1478 			 * offsets are released at the head.
1479 			 */
1480 			if (dofree) {
1481 				flags = SM_FREE | SM_ASYNC;
1482 				if ((cache_read_ahead == 0) &&
1483 				    (off > smallfile2))
1484 					flags |=  SM_DONTNEED;
1485 			}
1486 			/*
1487 			 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1488 			 * we want to make sure that the page which has
1489 			 * been read, is written on disk if it is dirty.
1490 			 * And corresponding indirect blocks should also
1491 			 * be flushed out.
1492 			 */
1493 			if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) {
1494 				flags &= ~SM_ASYNC;
1495 				flags |= SM_WRITE;
1496 			}
1497 			if (vpm_enable) {
1498 				error = vpm_sync_pages(vp, off, n, flags);
1499 			} else {
1500 				error = segmap_release(segkmap, base, flags);
1501 			}
1502 		} else {
1503 			if (vpm_enable) {
1504 				(void) vpm_sync_pages(vp, off, n, flags);
1505 			} else {
1506 				(void) segmap_release(segkmap, base, flags);
1507 			}
1508 		}
1509 
1510 		if (rwtype == RW_READER)
1511 			rw_enter(&ip->i_contents, rwtype);
1512 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1513 out:
1514 	/*
1515 	 * Inode is updated according to this table if FRSYNC is set.
1516 	 *
1517 	 *   FSYNC	  FDSYNC(posix.4)
1518 	 *   --------------------------
1519 	 *   always	  IATTCHG|IBDWRITE
1520 	 */
1521 	/*
1522 	 * The inode is not updated if we're logging and the inode is a
1523 	 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1524 	 */
1525 	if (ioflag & FRSYNC) {
1526 		if (TRANS_ISTRANS(ufsvfsp) && ((ip->i_mode & IFMT) == IFDIR)) {
1527 			doupdate = 0;
1528 		}
1529 		if (doupdate) {
1530 			if ((ioflag & FSYNC) ||
1531 			    ((ioflag & FDSYNC) &&
1532 			    (ip->i_flag & (IATTCHG|IBDWRITE)))) {
1533 				ufs_iupdat(ip, 1);
1534 			}
1535 		}
1536 	}
1537 	/*
1538 	 * If we've already done a partial read, terminate
1539 	 * the read but return no error.
1540 	 */
1541 	if (oresid != uio->uio_resid)
1542 		error = 0;
1543 	ITIMES(ip);
1544 
1545 	return (error);
1546 }
1547 
1548 /* ARGSUSED */
1549 static int
1550 ufs_ioctl(
1551 	struct vnode	*vp,
1552 	int		cmd,
1553 	intptr_t	arg,
1554 	int		flag,
1555 	struct cred	*cr,
1556 	int		*rvalp,
1557 	caller_context_t *ct)
1558 {
1559 	struct lockfs	lockfs, lockfs_out;
1560 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
1561 	char		*comment, *original_comment;
1562 	struct fs	*fs;
1563 	struct ulockfs	*ulp;
1564 	offset_t	off;
1565 	extern int	maxphys;
1566 	int		error;
1567 	int		issync;
1568 	int		trans_size;
1569 
1570 
1571 	/*
1572 	 * forcibly unmounted
1573 	 */
1574 	if (ufsvfsp == NULL || vp->v_vfsp == NULL ||
1575 	    vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1576 		return (EIO);
1577 	fs = ufsvfsp->vfs_fs;
1578 
1579 	if (cmd == Q_QUOTACTL) {
1580 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_QUOTA_MASK);
1581 		if (error)
1582 			return (error);
1583 
1584 		if (ulp) {
1585 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_QUOTA,
1586 			    TOP_SETQUOTA_SIZE(fs));
1587 		}
1588 
1589 		error = quotactl(vp, arg, flag, cr);
1590 
1591 		if (ulp) {
1592 			TRANS_END_ASYNC(ufsvfsp, TOP_QUOTA,
1593 			    TOP_SETQUOTA_SIZE(fs));
1594 			ufs_lockfs_end(ulp);
1595 		}
1596 		return (error);
1597 	}
1598 
1599 	switch (cmd) {
1600 		case _FIOLFS:
1601 			/*
1602 			 * file system locking
1603 			 */
1604 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1605 				return (EPERM);
1606 
1607 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1608 				if (copyin((caddr_t)arg, &lockfs,
1609 				    sizeof (struct lockfs)))
1610 					return (EFAULT);
1611 			}
1612 #ifdef _SYSCALL32_IMPL
1613 			else {
1614 				struct lockfs32	lockfs32;
1615 				/* Translate ILP32 lockfs to LP64 lockfs */
1616 				if (copyin((caddr_t)arg, &lockfs32,
1617 				    sizeof (struct lockfs32)))
1618 					return (EFAULT);
1619 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1620 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1621 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1622 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1623 				lockfs.lf_comment =
1624 				    (caddr_t)(uintptr_t)lockfs32.lf_comment;
1625 			}
1626 #endif /* _SYSCALL32_IMPL */
1627 
1628 			if (lockfs.lf_comlen) {
1629 				if (lockfs.lf_comlen > LOCKFS_MAXCOMMENTLEN)
1630 					return (ENAMETOOLONG);
1631 				comment =
1632 				    kmem_alloc(lockfs.lf_comlen, KM_SLEEP);
1633 				if (copyin(lockfs.lf_comment, comment,
1634 				    lockfs.lf_comlen)) {
1635 					kmem_free(comment, lockfs.lf_comlen);
1636 					return (EFAULT);
1637 				}
1638 				original_comment = lockfs.lf_comment;
1639 				lockfs.lf_comment = comment;
1640 			}
1641 			if ((error = ufs_fiolfs(vp, &lockfs, 0)) == 0) {
1642 				lockfs.lf_comment = original_comment;
1643 
1644 				if ((flag & DATAMODEL_MASK) ==
1645 				    DATAMODEL_NATIVE) {
1646 					(void) copyout(&lockfs, (caddr_t)arg,
1647 					    sizeof (struct lockfs));
1648 				}
1649 #ifdef _SYSCALL32_IMPL
1650 				else {
1651 					struct lockfs32	lockfs32;
1652 					/* Translate LP64 to ILP32 lockfs */
1653 					lockfs32.lf_lock =
1654 					    (uint32_t)lockfs.lf_lock;
1655 					lockfs32.lf_flags =
1656 					    (uint32_t)lockfs.lf_flags;
1657 					lockfs32.lf_key =
1658 					    (uint32_t)lockfs.lf_key;
1659 					lockfs32.lf_comlen =
1660 					    (uint32_t)lockfs.lf_comlen;
1661 					lockfs32.lf_comment =
1662 					    (uint32_t)(uintptr_t)
1663 					    lockfs.lf_comment;
1664 					(void) copyout(&lockfs32, (caddr_t)arg,
1665 					    sizeof (struct lockfs32));
1666 				}
1667 #endif /* _SYSCALL32_IMPL */
1668 
1669 			} else {
1670 				if (lockfs.lf_comlen)
1671 					kmem_free(comment, lockfs.lf_comlen);
1672 			}
1673 			return (error);
1674 
1675 		case _FIOLFSS:
1676 			/*
1677 			 * get file system locking status
1678 			 */
1679 
1680 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1681 				if (copyin((caddr_t)arg, &lockfs,
1682 				    sizeof (struct lockfs)))
1683 					return (EFAULT);
1684 			}
1685 #ifdef _SYSCALL32_IMPL
1686 			else {
1687 				struct lockfs32	lockfs32;
1688 				/* Translate ILP32 lockfs to LP64 lockfs */
1689 				if (copyin((caddr_t)arg, &lockfs32,
1690 				    sizeof (struct lockfs32)))
1691 					return (EFAULT);
1692 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1693 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1694 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1695 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1696 				lockfs.lf_comment =
1697 				    (caddr_t)(uintptr_t)lockfs32.lf_comment;
1698 			}
1699 #endif /* _SYSCALL32_IMPL */
1700 
1701 			if (error =  ufs_fiolfss(vp, &lockfs_out))
1702 				return (error);
1703 			lockfs.lf_lock = lockfs_out.lf_lock;
1704 			lockfs.lf_key = lockfs_out.lf_key;
1705 			lockfs.lf_flags = lockfs_out.lf_flags;
1706 			lockfs.lf_comlen = MIN(lockfs.lf_comlen,
1707 			    lockfs_out.lf_comlen);
1708 
1709 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1710 				if (copyout(&lockfs, (caddr_t)arg,
1711 				    sizeof (struct lockfs)))
1712 					return (EFAULT);
1713 			}
1714 #ifdef _SYSCALL32_IMPL
1715 			else {
1716 				/* Translate LP64 to ILP32 lockfs */
1717 				struct lockfs32	lockfs32;
1718 				lockfs32.lf_lock = (uint32_t)lockfs.lf_lock;
1719 				lockfs32.lf_flags = (uint32_t)lockfs.lf_flags;
1720 				lockfs32.lf_key = (uint32_t)lockfs.lf_key;
1721 				lockfs32.lf_comlen = (uint32_t)lockfs.lf_comlen;
1722 				lockfs32.lf_comment =
1723 				    (uint32_t)(uintptr_t)lockfs.lf_comment;
1724 				if (copyout(&lockfs32, (caddr_t)arg,
1725 				    sizeof (struct lockfs32)))
1726 					return (EFAULT);
1727 			}
1728 #endif /* _SYSCALL32_IMPL */
1729 
1730 			if (lockfs.lf_comlen &&
1731 			    lockfs.lf_comment && lockfs_out.lf_comment)
1732 				if (copyout(lockfs_out.lf_comment,
1733 				    lockfs.lf_comment, lockfs.lf_comlen))
1734 					return (EFAULT);
1735 			return (0);
1736 
1737 		case _FIOSATIME:
1738 			/*
1739 			 * set access time
1740 			 */
1741 
1742 			/*
1743 			 * if mounted w/o atime, return quietly.
1744 			 * I briefly thought about returning ENOSYS, but
1745 			 * figured that most apps would consider this fatal
1746 			 * but the idea is to make this as seamless as poss.
1747 			 */
1748 			if (ufsvfsp->vfs_noatime)
1749 				return (0);
1750 
1751 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1752 			    ULOCKFS_SETATTR_MASK);
1753 			if (error)
1754 				return (error);
1755 
1756 			if (ulp) {
1757 				trans_size = (int)TOP_SETATTR_SIZE(VTOI(vp));
1758 				TRANS_BEGIN_CSYNC(ufsvfsp, issync,
1759 				    TOP_SETATTR, trans_size);
1760 			}
1761 
1762 			error = ufs_fiosatime(vp, (struct timeval *)arg,
1763 			    flag, cr);
1764 
1765 			if (ulp) {
1766 				TRANS_END_CSYNC(ufsvfsp, error, issync,
1767 				    TOP_SETATTR, trans_size);
1768 				ufs_lockfs_end(ulp);
1769 			}
1770 			return (error);
1771 
1772 		case _FIOSDIO:
1773 			/*
1774 			 * set delayed-io
1775 			 */
1776 			return (ufs_fiosdio(vp, (uint_t *)arg, flag, cr));
1777 
1778 		case _FIOGDIO:
1779 			/*
1780 			 * get delayed-io
1781 			 */
1782 			return (ufs_fiogdio(vp, (uint_t *)arg, flag, cr));
1783 
1784 		case _FIOIO:
1785 			/*
1786 			 * inode open
1787 			 */
1788 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1789 			    ULOCKFS_VGET_MASK);
1790 			if (error)
1791 				return (error);
1792 
1793 			error = ufs_fioio(vp, (struct fioio *)arg, flag, cr);
1794 
1795 			if (ulp) {
1796 				ufs_lockfs_end(ulp);
1797 			}
1798 			return (error);
1799 
1800 		case _FIOFFS:
1801 			/*
1802 			 * file system flush (push w/invalidate)
1803 			 */
1804 			if ((caddr_t)arg != NULL)
1805 				return (EINVAL);
1806 			return (ufs_fioffs(vp, NULL, cr));
1807 
1808 		case _FIOISBUSY:
1809 			/*
1810 			 * Contract-private interface for Legato
1811 			 * Purge this vnode from the DNLC and decide
1812 			 * if this vnode is busy (*arg == 1) or not
1813 			 * (*arg == 0)
1814 			 */
1815 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1816 				return (EPERM);
1817 			error = ufs_fioisbusy(vp, (int *)arg, cr);
1818 			return (error);
1819 
1820 		case _FIODIRECTIO:
1821 			return (ufs_fiodirectio(vp, (int)arg, cr));
1822 
1823 		case _FIOTUNE:
1824 			/*
1825 			 * Tune the file system (aka setting fs attributes)
1826 			 */
1827 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1828 			    ULOCKFS_SETATTR_MASK);
1829 			if (error)
1830 				return (error);
1831 
1832 			error = ufs_fiotune(vp, (struct fiotune *)arg, cr);
1833 
1834 			if (ulp)
1835 				ufs_lockfs_end(ulp);
1836 			return (error);
1837 
1838 		case _FIOLOGENABLE:
1839 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1840 				return (EPERM);
1841 			return (ufs_fiologenable(vp, (void *)arg, cr, flag));
1842 
1843 		case _FIOLOGDISABLE:
1844 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1845 				return (EPERM);
1846 			return (ufs_fiologdisable(vp, (void *)arg, cr, flag));
1847 
1848 		case _FIOISLOG:
1849 			return (ufs_fioislog(vp, (void *)arg, cr, flag));
1850 
1851 		case _FIOSNAPSHOTCREATE_MULTI:
1852 		{
1853 			struct fiosnapcreate_multi	fc, *fcp;
1854 			size_t	fcm_size;
1855 
1856 			if (copyin((void *)arg, &fc, sizeof (fc)))
1857 				return (EFAULT);
1858 			if (fc.backfilecount > MAX_BACKFILE_COUNT)
1859 				return (EINVAL);
1860 			fcm_size = sizeof (struct fiosnapcreate_multi) +
1861 			    (fc.backfilecount - 1) * sizeof (int);
1862 			fcp = (struct fiosnapcreate_multi *)
1863 			    kmem_alloc(fcm_size, KM_SLEEP);
1864 			if (copyin((void *)arg, fcp, fcm_size)) {
1865 				kmem_free(fcp, fcm_size);
1866 				return (EFAULT);
1867 			}
1868 			error = ufs_snap_create(vp, fcp, cr);
1869 			/*
1870 			 * Do copyout even if there is an error because
1871 			 * the details of error is stored in fcp.
1872 			 */
1873 			if (copyout(fcp, (void *)arg, fcm_size))
1874 				error = EFAULT;
1875 			kmem_free(fcp, fcm_size);
1876 			return (error);
1877 		}
1878 
1879 		case _FIOSNAPSHOTDELETE:
1880 		{
1881 			struct fiosnapdelete	fc;
1882 
1883 			if (copyin((void *)arg, &fc, sizeof (fc)))
1884 				return (EFAULT);
1885 			error = ufs_snap_delete(vp, &fc, cr);
1886 			if (!error && copyout(&fc, (void *)arg, sizeof (fc)))
1887 				error = EFAULT;
1888 			return (error);
1889 		}
1890 
1891 		case _FIOGETSUPERBLOCK:
1892 			if (copyout(fs, (void *)arg, SBSIZE))
1893 				return (EFAULT);
1894 			return (0);
1895 
1896 		case _FIOGETMAXPHYS:
1897 			if (copyout(&maxphys, (void *)arg, sizeof (maxphys)))
1898 				return (EFAULT);
1899 			return (0);
1900 
1901 		/*
1902 		 * The following 3 ioctls are for TSufs support
1903 		 * although could potentially be used elsewhere
1904 		 */
1905 		case _FIO_SET_LUFS_DEBUG:
1906 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1907 				return (EPERM);
1908 			lufs_debug = (uint32_t)arg;
1909 			return (0);
1910 
1911 		case _FIO_SET_LUFS_ERROR:
1912 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1913 				return (EPERM);
1914 			TRANS_SETERROR(ufsvfsp);
1915 			return (0);
1916 
1917 		case _FIO_GET_TOP_STATS:
1918 		{
1919 			fio_lufs_stats_t *ls;
1920 			ml_unit_t *ul = ufsvfsp->vfs_log;
1921 
1922 			ls = kmem_zalloc(sizeof (*ls), KM_SLEEP);
1923 			ls->ls_debug = ul->un_debug; /* return debug value */
1924 			/* Copy stucture if statistics are being kept */
1925 			if (ul->un_logmap->mtm_tops) {
1926 				ls->ls_topstats = *(ul->un_logmap->mtm_tops);
1927 			}
1928 			error = 0;
1929 			if (copyout(ls, (void *)arg, sizeof (*ls)))
1930 				error = EFAULT;
1931 			kmem_free(ls, sizeof (*ls));
1932 			return (error);
1933 		}
1934 
1935 		case _FIO_SEEK_DATA:
1936 		case _FIO_SEEK_HOLE:
1937 			if (ddi_copyin((void *)arg, &off, sizeof (off), flag))
1938 				return (EFAULT);
1939 			/* offset paramater is in/out */
1940 			error = ufs_fio_holey(vp, cmd, &off);
1941 			if (error)
1942 				return (error);
1943 			if (ddi_copyout(&off, (void *)arg, sizeof (off), flag))
1944 				return (EFAULT);
1945 			return (0);
1946 
1947 		case _FIO_COMPRESSED:
1948 		{
1949 			/*
1950 			 * This is a project private ufs ioctl() to mark
1951 			 * the inode as that belonging to a compressed
1952 			 * file. This is used to mark individual
1953 			 * compressed files in a miniroot archive.
1954 			 * The files compressed in this manner are
1955 			 * automatically decompressed by the dcfs filesystem
1956 			 * (via an interception in ufs_lookup - see decompvp())
1957 			 * which is layered on top of ufs on a system running
1958 			 * from the archive. See uts/common/fs/dcfs for details.
1959 			 * This ioctl only marks the file as compressed - the
1960 			 * actual compression is done by fiocompress (a
1961 			 * userland utility) which invokes this ioctl().
1962 			 */
1963 			struct inode *ip = VTOI(vp);
1964 
1965 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1966 			    ULOCKFS_SETATTR_MASK);
1967 			if (error)
1968 				return (error);
1969 
1970 			if (ulp) {
1971 				TRANS_BEGIN_ASYNC(ufsvfsp, TOP_IUPDAT,
1972 				    TOP_IUPDAT_SIZE(ip));
1973 			}
1974 
1975 			error = ufs_mark_compressed(vp);
1976 
1977 			if (ulp) {
1978 				TRANS_END_ASYNC(ufsvfsp, TOP_IUPDAT,
1979 				    TOP_IUPDAT_SIZE(ip));
1980 				ufs_lockfs_end(ulp);
1981 			}
1982 
1983 			return (error);
1984 
1985 		}
1986 
1987 		default:
1988 			return (ENOTTY);
1989 	}
1990 }
1991 
1992 
1993 /* ARGSUSED */
1994 static int
1995 ufs_getattr(struct vnode *vp, struct vattr *vap, int flags,
1996     struct cred *cr, caller_context_t *ct)
1997 {
1998 	struct inode *ip = VTOI(vp);
1999 	struct ufsvfs *ufsvfsp;
2000 	int err;
2001 
2002 	if (vap->va_mask == AT_SIZE) {
2003 		/*
2004 		 * for performance, if only the size is requested don't bother
2005 		 * with anything else.
2006 		 */
2007 		UFS_GET_ISIZE(&vap->va_size, ip);
2008 		return (0);
2009 	}
2010 
2011 	/*
2012 	 * inlined lockfs checks
2013 	 */
2014 	ufsvfsp = ip->i_ufsvfs;
2015 	if ((ufsvfsp == NULL) || ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs)) {
2016 		err = EIO;
2017 		goto out;
2018 	}
2019 
2020 	rw_enter(&ip->i_contents, RW_READER);
2021 	/*
2022 	 * Return all the attributes.  This should be refined so
2023 	 * that it only returns what's asked for.
2024 	 */
2025 
2026 	/*
2027 	 * Copy from inode table.
2028 	 */
2029 	vap->va_type = vp->v_type;
2030 	vap->va_mode = ip->i_mode & MODEMASK;
2031 	/*
2032 	 * If there is an ACL and there is a mask entry, then do the
2033 	 * extra work that completes the equivalent of an acltomode(3)
2034 	 * call.  According to POSIX P1003.1e, the acl mask should be
2035 	 * returned in the group permissions field.
2036 	 *
2037 	 * - start with the original permission and mode bits (from above)
2038 	 * - clear the group owner bits
2039 	 * - add in the mask bits.
2040 	 */
2041 	if (ip->i_ufs_acl && ip->i_ufs_acl->aclass.acl_ismask) {
2042 		vap->va_mode &= ~((VREAD | VWRITE | VEXEC) >> 3);
2043 		vap->va_mode |=
2044 		    (ip->i_ufs_acl->aclass.acl_maskbits & PERMMASK) << 3;
2045 	}
2046 	vap->va_uid = ip->i_uid;
2047 	vap->va_gid = ip->i_gid;
2048 	vap->va_fsid = ip->i_dev;
2049 	vap->va_nodeid = (ino64_t)ip->i_number;
2050 	vap->va_nlink = ip->i_nlink;
2051 	vap->va_size = ip->i_size;
2052 	if (vp->v_type == VCHR || vp->v_type == VBLK)
2053 		vap->va_rdev = ip->i_rdev;
2054 	else
2055 		vap->va_rdev = 0;	/* not a b/c spec. */
2056 	mutex_enter(&ip->i_tlock);
2057 	ITIMES_NOLOCK(ip);	/* mark correct time in inode */
2058 	vap->va_seq = ip->i_seq;
2059 	vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec;
2060 	vap->va_atime.tv_nsec = ip->i_atime.tv_usec*1000;
2061 	vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec;
2062 	vap->va_mtime.tv_nsec = ip->i_mtime.tv_usec*1000;
2063 	vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec;
2064 	vap->va_ctime.tv_nsec = ip->i_ctime.tv_usec*1000;
2065 	mutex_exit(&ip->i_tlock);
2066 
2067 	switch (ip->i_mode & IFMT) {
2068 
2069 	case IFBLK:
2070 		vap->va_blksize = MAXBSIZE;		/* was BLKDEV_IOSIZE */
2071 		break;
2072 
2073 	case IFCHR:
2074 		vap->va_blksize = MAXBSIZE;
2075 		break;
2076 
2077 	default:
2078 		vap->va_blksize = ip->i_fs->fs_bsize;
2079 		break;
2080 	}
2081 	vap->va_nblocks = (fsblkcnt64_t)ip->i_blocks;
2082 	rw_exit(&ip->i_contents);
2083 	err = 0;
2084 
2085 out:
2086 	return (err);
2087 }
2088 
2089 /*
2090  * Special wrapper to provide a callback for secpolicy_vnode_setattr().
2091  * The i_contents lock is already held by the caller and we need to
2092  * declare the inode as 'void *' argument.
2093  */
2094 static int
2095 ufs_priv_access(void *vip, int mode, struct cred *cr)
2096 {
2097 	struct inode *ip = vip;
2098 
2099 	return (ufs_iaccess(ip, mode, cr, 0));
2100 }
2101 
2102 /*ARGSUSED4*/
2103 static int
2104 ufs_setattr(struct vnode *vp, struct vattr *vap, int flags, struct cred *cr,
2105     caller_context_t *ct)
2106 {
2107 	struct inode *ip = VTOI(vp);
2108 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2109 	struct fs *fs;
2110 	struct ulockfs *ulp;
2111 	char *errmsg1;
2112 	char *errmsg2;
2113 	long blocks;
2114 	long int mask = vap->va_mask;
2115 	size_t len1, len2;
2116 	int issync;
2117 	int trans_size;
2118 	int dotrans;
2119 	int dorwlock;
2120 	int error;
2121 	int owner_change;
2122 	int dodqlock;
2123 	timestruc_t now;
2124 	vattr_t oldva;
2125 	int retry = 1;
2126 	int indeadlock;
2127 
2128 	/*
2129 	 * Cannot set these attributes.
2130 	 */
2131 	if ((mask & AT_NOSET) || (mask & AT_XVATTR))
2132 		return (EINVAL);
2133 
2134 	/*
2135 	 * check for forced unmount
2136 	 */
2137 	if (ufsvfsp == NULL)
2138 		return (EIO);
2139 
2140 	fs = ufsvfsp->vfs_fs;
2141 	if (fs->fs_ronly != 0)
2142 		return (EROFS);
2143 
2144 again:
2145 	errmsg1 = NULL;
2146 	errmsg2 = NULL;
2147 	dotrans = 0;
2148 	dorwlock = 0;
2149 	dodqlock = 0;
2150 
2151 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK);
2152 	if (error)
2153 		goto out;
2154 
2155 	/*
2156 	 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2157 	 * This follows the protocol for read()/write().
2158 	 */
2159 	if (vp->v_type != VDIR) {
2160 		/*
2161 		 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2162 		 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2163 		 * possible, retries the operation.
2164 		 */
2165 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_file);
2166 		if (indeadlock) {
2167 			if (ulp)
2168 				ufs_lockfs_end(ulp);
2169 			goto again;
2170 		}
2171 		dorwlock = 1;
2172 	}
2173 
2174 	/*
2175 	 * Truncate file.  Must have write permission and not be a directory.
2176 	 */
2177 	if (mask & AT_SIZE) {
2178 		rw_enter(&ip->i_contents, RW_WRITER);
2179 		if (vp->v_type == VDIR) {
2180 			error = EISDIR;
2181 			goto update_inode;
2182 		}
2183 		if (error = ufs_iaccess(ip, IWRITE, cr, 0))
2184 			goto update_inode;
2185 
2186 		rw_exit(&ip->i_contents);
2187 		error = TRANS_ITRUNC(ip, vap->va_size, 0, cr);
2188 		if (error) {
2189 			rw_enter(&ip->i_contents, RW_WRITER);
2190 			goto update_inode;
2191 		}
2192 
2193 		if (error == 0 && vap->va_size)
2194 			vnevent_truncate(vp, ct);
2195 	}
2196 
2197 	if (ulp) {
2198 		trans_size = (int)TOP_SETATTR_SIZE(ip);
2199 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size);
2200 		++dotrans;
2201 	}
2202 
2203 	/*
2204 	 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2205 	 * This follows the protocol established by
2206 	 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2207 	 */
2208 	if (vp->v_type == VDIR) {
2209 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_SETATTR,
2210 		    retry_dir);
2211 		if (indeadlock)
2212 			goto again;
2213 		dorwlock = 1;
2214 	}
2215 
2216 	/*
2217 	 * Grab quota lock if we are changing the file's owner.
2218 	 */
2219 	if (mask & AT_UID) {
2220 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2221 		dodqlock = 1;
2222 	}
2223 	rw_enter(&ip->i_contents, RW_WRITER);
2224 
2225 	oldva.va_mode = ip->i_mode;
2226 	oldva.va_uid = ip->i_uid;
2227 	oldva.va_gid = ip->i_gid;
2228 
2229 	vap->va_mask &= ~AT_SIZE;
2230 
2231 	error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2232 	    ufs_priv_access, ip);
2233 	if (error)
2234 		goto update_inode;
2235 
2236 	mask = vap->va_mask;
2237 
2238 	/*
2239 	 * Change file access modes.
2240 	 */
2241 	if (mask & AT_MODE) {
2242 		ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT);
2243 		TRANS_INODE(ufsvfsp, ip);
2244 		ip->i_flag |= ICHG;
2245 		if (stickyhack) {
2246 			mutex_enter(&vp->v_lock);
2247 			if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX)
2248 				vp->v_flag |= VSWAPLIKE;
2249 			else
2250 				vp->v_flag &= ~VSWAPLIKE;
2251 			mutex_exit(&vp->v_lock);
2252 		}
2253 	}
2254 	if (mask & (AT_UID|AT_GID)) {
2255 		if (mask & AT_UID) {
2256 			/*
2257 			 * Don't change ownership of the quota inode.
2258 			 */
2259 			if (ufsvfsp->vfs_qinod == ip) {
2260 				ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED);
2261 				error = EINVAL;
2262 				goto update_inode;
2263 			}
2264 
2265 			/*
2266 			 * No real ownership change.
2267 			 */
2268 			if (ip->i_uid == vap->va_uid) {
2269 				blocks = 0;
2270 				owner_change = 0;
2271 			}
2272 			/*
2273 			 * Remove the blocks and the file, from the old user's
2274 			 * quota.
2275 			 */
2276 			else {
2277 				blocks = ip->i_blocks;
2278 				owner_change = 1;
2279 
2280 				(void) chkdq(ip, -blocks, /* force */ 1, cr,
2281 				    (char **)NULL, (size_t *)NULL);
2282 				(void) chkiq(ufsvfsp, /* change */ -1, ip,
2283 				    (uid_t)ip->i_uid, /* force */ 1, cr,
2284 				    (char **)NULL, (size_t *)NULL);
2285 				dqrele(ip->i_dquot);
2286 			}
2287 
2288 			ip->i_uid = vap->va_uid;
2289 
2290 			/*
2291 			 * There is a real ownership change.
2292 			 */
2293 			if (owner_change) {
2294 				/*
2295 				 * Add the blocks and the file to the new
2296 				 * user's quota.
2297 				 */
2298 				ip->i_dquot = getinoquota(ip);
2299 				(void) chkdq(ip, blocks, /* force */ 1, cr,
2300 				    &errmsg1, &len1);
2301 				(void) chkiq(ufsvfsp, /* change */ 1,
2302 				    (struct inode *)NULL, (uid_t)ip->i_uid,
2303 				    /* force */ 1, cr, &errmsg2, &len2);
2304 			}
2305 		}
2306 		if (mask & AT_GID) {
2307 			ip->i_gid = vap->va_gid;
2308 		}
2309 		TRANS_INODE(ufsvfsp, ip);
2310 		ip->i_flag |= ICHG;
2311 	}
2312 	/*
2313 	 * Change file access or modified times.
2314 	 */
2315 	if (mask & (AT_ATIME|AT_MTIME)) {
2316 		/* Check that the time value is within ufs range */
2317 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2318 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2319 			error = EOVERFLOW;
2320 			goto update_inode;
2321 		}
2322 
2323 		/*
2324 		 * if the "noaccess" mount option is set and only atime
2325 		 * update is requested, do nothing. No error is returned.
2326 		 */
2327 		if ((ufsvfsp->vfs_noatime) &&
2328 		    ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME))
2329 			goto skip_atime;
2330 
2331 		if (mask & AT_ATIME) {
2332 			ip->i_atime.tv_sec = vap->va_atime.tv_sec;
2333 			ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000;
2334 			ip->i_flag &= ~IACC;
2335 		}
2336 		if (mask & AT_MTIME) {
2337 			ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
2338 			ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000;
2339 			gethrestime(&now);
2340 			if (now.tv_sec > TIME32_MAX) {
2341 				/*
2342 				 * In 2038, ctime sticks forever..
2343 				 */
2344 				ip->i_ctime.tv_sec = TIME32_MAX;
2345 				ip->i_ctime.tv_usec = 0;
2346 			} else {
2347 				ip->i_ctime.tv_sec = now.tv_sec;
2348 				ip->i_ctime.tv_usec = now.tv_nsec / 1000;
2349 			}
2350 			ip->i_flag &= ~(IUPD|ICHG);
2351 			ip->i_flag |= IMODTIME;
2352 		}
2353 		TRANS_INODE(ufsvfsp, ip);
2354 		ip->i_flag |= IMOD;
2355 	}
2356 
2357 skip_atime:
2358 	/*
2359 	 * The presence of a shadow inode may indicate an ACL, but does
2360 	 * not imply an ACL.  Future FSD types should be handled here too
2361 	 * and check for the presence of the attribute-specific data
2362 	 * before referencing it.
2363 	 */
2364 	if (ip->i_shadow) {
2365 		/*
2366 		 * XXX if ufs_iupdat is changed to sandbagged write fix
2367 		 * ufs_acl_setattr to push ip to keep acls consistent
2368 		 *
2369 		 * Suppress out of inodes messages if we will retry.
2370 		 */
2371 		if (retry)
2372 			ip->i_flag |= IQUIET;
2373 		error = ufs_acl_setattr(ip, vap, cr);
2374 		ip->i_flag &= ~IQUIET;
2375 	}
2376 
2377 update_inode:
2378 	/*
2379 	 * Setattr always increases the sequence number
2380 	 */
2381 	ip->i_seq++;
2382 
2383 	/*
2384 	 * if nfsd and not logging; push synchronously
2385 	 */
2386 	if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) {
2387 		ufs_iupdat(ip, 1);
2388 	} else {
2389 		ITIMES_NOLOCK(ip);
2390 	}
2391 
2392 	rw_exit(&ip->i_contents);
2393 	if (dodqlock) {
2394 		rw_exit(&ufsvfsp->vfs_dqrwlock);
2395 	}
2396 	if (dorwlock)
2397 		rw_exit(&ip->i_rwlock);
2398 
2399 	if (ulp) {
2400 		if (dotrans) {
2401 			int terr = 0;
2402 			TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR,
2403 			    trans_size);
2404 			if (error == 0)
2405 				error = terr;
2406 		}
2407 		ufs_lockfs_end(ulp);
2408 	}
2409 out:
2410 	/*
2411 	 * If out of inodes or blocks, see if we can free something
2412 	 * up from the delete queue.
2413 	 */
2414 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
2415 		ufs_delete_drain_wait(ufsvfsp, 1);
2416 		retry = 0;
2417 		if (errmsg1 != NULL)
2418 			kmem_free(errmsg1, len1);
2419 		if (errmsg2 != NULL)
2420 			kmem_free(errmsg2, len2);
2421 		goto again;
2422 	}
2423 	if (errmsg1 != NULL) {
2424 		uprintf(errmsg1);
2425 		kmem_free(errmsg1, len1);
2426 	}
2427 	if (errmsg2 != NULL) {
2428 		uprintf(errmsg2);
2429 		kmem_free(errmsg2, len2);
2430 	}
2431 	return (error);
2432 }
2433 
2434 /*ARGSUSED*/
2435 static int
2436 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr,
2437     caller_context_t *ct)
2438 {
2439 	struct inode *ip = VTOI(vp);
2440 
2441 	if (ip->i_ufsvfs == NULL)
2442 		return (EIO);
2443 
2444 	/*
2445 	 * The ufs_iaccess function wants to be called with
2446 	 * mode bits expressed as "ufs specific" bits.
2447 	 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2448 	 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2449 	 * But since they're the same we just pass the vnode mode
2450 	 * bit but just verify that assumption at compile time.
2451 	 */
2452 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2453 #error "ufs_access needs to map Vmodes to Imodes"
2454 #endif
2455 	return (ufs_iaccess(ip, mode, cr, 1));
2456 }
2457 
2458 /* ARGSUSED */
2459 static int
2460 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr,
2461     caller_context_t *ct)
2462 {
2463 	struct inode *ip = VTOI(vp);
2464 	struct ufsvfs *ufsvfsp;
2465 	struct ulockfs *ulp;
2466 	int error;
2467 	int fastsymlink;
2468 
2469 	if (vp->v_type != VLNK) {
2470 		error = EINVAL;
2471 		goto nolockout;
2472 	}
2473 
2474 	/*
2475 	 * If the symbolic link is empty there is nothing to read.
2476 	 * Fast-track these empty symbolic links
2477 	 */
2478 	if (ip->i_size == 0) {
2479 		error = 0;
2480 		goto nolockout;
2481 	}
2482 
2483 	ufsvfsp = ip->i_ufsvfs;
2484 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK);
2485 	if (error)
2486 		goto nolockout;
2487 	/*
2488 	 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2489 	 */
2490 again:
2491 	fastsymlink = 0;
2492 	if (ip->i_flag & IFASTSYMLNK) {
2493 		rw_enter(&ip->i_rwlock, RW_READER);
2494 		rw_enter(&ip->i_contents, RW_READER);
2495 		if (ip->i_flag & IFASTSYMLNK) {
2496 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
2497 			    (ip->i_fs->fs_ronly == 0) &&
2498 			    (!ufsvfsp->vfs_noatime)) {
2499 				mutex_enter(&ip->i_tlock);
2500 				ip->i_flag |= IACC;
2501 				mutex_exit(&ip->i_tlock);
2502 			}
2503 			error = uiomove((caddr_t)&ip->i_db[1],
2504 			    MIN(ip->i_size, uiop->uio_resid),
2505 			    UIO_READ, uiop);
2506 			ITIMES(ip);
2507 			++fastsymlink;
2508 		}
2509 		rw_exit(&ip->i_contents);
2510 		rw_exit(&ip->i_rwlock);
2511 	}
2512 	if (!fastsymlink) {
2513 		ssize_t size;	/* number of bytes read  */
2514 		caddr_t basep;	/* pointer to input data */
2515 		ino_t ino;
2516 		long  igen;
2517 		struct uio tuio;	/* temp uio struct */
2518 		struct uio *tuiop;
2519 		iovec_t tiov;		/* temp iovec struct */
2520 		char kbuf[FSL_SIZE];	/* buffer to hold fast symlink */
2521 		int tflag = 0;		/* flag to indicate temp vars used */
2522 
2523 		ino = ip->i_number;
2524 		igen = ip->i_gen;
2525 		size = uiop->uio_resid;
2526 		basep = uiop->uio_iov->iov_base;
2527 		tuiop = uiop;
2528 
2529 		rw_enter(&ip->i_rwlock, RW_WRITER);
2530 		rw_enter(&ip->i_contents, RW_WRITER);
2531 		if (ip->i_flag & IFASTSYMLNK) {
2532 			rw_exit(&ip->i_contents);
2533 			rw_exit(&ip->i_rwlock);
2534 			goto again;
2535 		}
2536 
2537 		/* can this be a fast symlink and is it a user buffer? */
2538 		if (ip->i_size <= FSL_SIZE &&
2539 		    (uiop->uio_segflg == UIO_USERSPACE ||
2540 		    uiop->uio_segflg == UIO_USERISPACE)) {
2541 
2542 			bzero(&tuio, sizeof (struct uio));
2543 			/*
2544 			 * setup a kernel buffer to read link into.  this
2545 			 * is to fix a race condition where the user buffer
2546 			 * got corrupted before copying it into the inode.
2547 			 */
2548 			size = ip->i_size;
2549 			tiov.iov_len = size;
2550 			tiov.iov_base = kbuf;
2551 			tuio.uio_iov = &tiov;
2552 			tuio.uio_iovcnt = 1;
2553 			tuio.uio_offset = uiop->uio_offset;
2554 			tuio.uio_segflg = UIO_SYSSPACE;
2555 			tuio.uio_fmode = uiop->uio_fmode;
2556 			tuio.uio_extflg = uiop->uio_extflg;
2557 			tuio.uio_limit = uiop->uio_limit;
2558 			tuio.uio_resid = size;
2559 
2560 			basep = tuio.uio_iov->iov_base;
2561 			tuiop = &tuio;
2562 			tflag = 1;
2563 		}
2564 
2565 		error = rdip(ip, tuiop, 0, cr);
2566 		if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) {
2567 			rw_exit(&ip->i_contents);
2568 			rw_exit(&ip->i_rwlock);
2569 			goto out;
2570 		}
2571 
2572 		if (tflag == 0)
2573 			size -= uiop->uio_resid;
2574 
2575 		if ((tflag == 0 && ip->i_size <= FSL_SIZE &&
2576 		    ip->i_size == size) || (tflag == 1 &&
2577 		    tuio.uio_resid == 0)) {
2578 			error = kcopy(basep, &ip->i_db[1], ip->i_size);
2579 			if (error == 0) {
2580 				ip->i_flag |= IFASTSYMLNK;
2581 				/*
2582 				 * free page
2583 				 */
2584 				(void) VOP_PUTPAGE(ITOV(ip),
2585 				    (offset_t)0, PAGESIZE,
2586 				    (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC),
2587 				    cr, ct);
2588 			} else {
2589 				int i;
2590 				/* error, clear garbage left behind */
2591 				for (i = 1; i < NDADDR; i++)
2592 					ip->i_db[i] = 0;
2593 				for (i = 0; i < NIADDR; i++)
2594 					ip->i_ib[i] = 0;
2595 			}
2596 		}
2597 		if (tflag == 1) {
2598 			/* now, copy it into the user buffer */
2599 			error = uiomove((caddr_t)kbuf,
2600 			    MIN(size, uiop->uio_resid),
2601 			    UIO_READ, uiop);
2602 		}
2603 		rw_exit(&ip->i_contents);
2604 		rw_exit(&ip->i_rwlock);
2605 	}
2606 out:
2607 	if (ulp) {
2608 		ufs_lockfs_end(ulp);
2609 	}
2610 nolockout:
2611 	return (error);
2612 }
2613 
2614 /* ARGSUSED */
2615 static int
2616 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr, caller_context_t *ct)
2617 {
2618 	struct inode *ip = VTOI(vp);
2619 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2620 	struct ulockfs *ulp;
2621 	int error;
2622 
2623 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK);
2624 	if (error)
2625 		return (error);
2626 
2627 	if (TRANS_ISTRANS(ufsvfsp)) {
2628 		/*
2629 		 * First push out any data pages
2630 		 */
2631 		if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2632 		    (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) {
2633 			error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0,
2634 			    0, CRED(), ct);
2635 			if (error)
2636 				goto out;
2637 		}
2638 
2639 		/*
2640 		 * Delta any delayed inode times updates
2641 		 * and push inode to log.
2642 		 * All other inode deltas will have already been delta'd
2643 		 * and will be pushed during the commit.
2644 		 */
2645 		if (!(syncflag & FDSYNC) &&
2646 		    ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) {
2647 			if (ulp) {
2648 				TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC,
2649 				    TOP_SYNCIP_SIZE);
2650 			}
2651 			rw_enter(&ip->i_contents, RW_READER);
2652 			mutex_enter(&ip->i_tlock);
2653 			ip->i_flag &= ~IMODTIME;
2654 			mutex_exit(&ip->i_tlock);
2655 			ufs_iupdat(ip, I_SYNC);
2656 			rw_exit(&ip->i_contents);
2657 			if (ulp) {
2658 				TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC,
2659 				    TOP_SYNCIP_SIZE);
2660 			}
2661 		}
2662 
2663 		/*
2664 		 * Commit the Moby transaction
2665 		 *
2666 		 * Deltas have already been made so we just need to
2667 		 * commit them with a synchronous transaction.
2668 		 * TRANS_BEGIN_SYNC() will return an error
2669 		 * if there are no deltas to commit, for an
2670 		 * empty transaction.
2671 		 */
2672 		if (ulp) {
2673 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE,
2674 			    error);
2675 			if (error) {
2676 				error = 0; /* commit wasn't needed */
2677 				goto out;
2678 			}
2679 			TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC,
2680 			    TOP_COMMIT_SIZE);
2681 		}
2682 	} else {	/* not logging */
2683 		if (!(IS_SWAPVP(vp)))
2684 			if (syncflag & FNODSYNC) {
2685 				/* Just update the inode only */
2686 				TRANS_IUPDAT(ip, 1);
2687 				error = 0;
2688 			} else if (syncflag & FDSYNC)
2689 				/* Do data-synchronous writes */
2690 				error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC);
2691 			else
2692 				/* Do synchronous writes */
2693 				error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC);
2694 
2695 		rw_enter(&ip->i_contents, RW_WRITER);
2696 		if (!error)
2697 			error = ufs_sync_indir(ip);
2698 		rw_exit(&ip->i_contents);
2699 	}
2700 out:
2701 	if (ulp) {
2702 		ufs_lockfs_end(ulp);
2703 	}
2704 	return (error);
2705 }
2706 
2707 /*ARGSUSED*/
2708 static void
2709 ufs_inactive(struct vnode *vp, struct cred *cr, caller_context_t *ct)
2710 {
2711 	ufs_iinactive(VTOI(vp));
2712 }
2713 
2714 /*
2715  * Unix file system operations having to do with directory manipulation.
2716  */
2717 int ufs_lookup_idle_count = 2;	/* Number of inodes to idle each time */
2718 /* ARGSUSED */
2719 static int
2720 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
2721     struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr,
2722     caller_context_t *ct, int *direntflags, pathname_t *realpnp)
2723 {
2724 	struct inode *ip;
2725 	struct inode *sip;
2726 	struct inode *xip;
2727 	struct ufsvfs *ufsvfsp;
2728 	struct ulockfs *ulp;
2729 	struct vnode *vp;
2730 	int error;
2731 
2732 	/*
2733 	 * Check flags for type of lookup (regular file or attribute file)
2734 	 */
2735 
2736 	ip = VTOI(dvp);
2737 
2738 	if (flags & LOOKUP_XATTR) {
2739 
2740 		/*
2741 		 * If not mounted with XATTR support then return EINVAL
2742 		 */
2743 
2744 		if (!(ip->i_ufsvfs->vfs_vfs->vfs_flag & VFS_XATTR))
2745 			return (EINVAL);
2746 		/*
2747 		 * We don't allow recursive attributes...
2748 		 * Maybe someday we will.
2749 		 */
2750 		if ((ip->i_cflags & IXATTR)) {
2751 			return (EINVAL);
2752 		}
2753 
2754 		if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) {
2755 			error = ufs_xattr_getattrdir(dvp, &sip, flags, cr);
2756 			if (error) {
2757 				*vpp = NULL;
2758 				goto out;
2759 			}
2760 
2761 			vp = ITOV(sip);
2762 			dnlc_update(dvp, XATTR_DIR_NAME, vp);
2763 		}
2764 
2765 		/*
2766 		 * Check accessibility of directory.
2767 		 */
2768 		if (vp == DNLC_NO_VNODE) {
2769 			VN_RELE(vp);
2770 			error = ENOENT;
2771 			goto out;
2772 		}
2773 		if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr, 1)) != 0) {
2774 			VN_RELE(vp);
2775 			goto out;
2776 		}
2777 
2778 		*vpp = vp;
2779 		return (0);
2780 	}
2781 
2782 	/*
2783 	 * Check for a null component, which we should treat as
2784 	 * looking at dvp from within it's parent, so we don't
2785 	 * need a call to ufs_iaccess(), as it has already been
2786 	 * done.
2787 	 */
2788 	if (nm[0] == 0) {
2789 		VN_HOLD(dvp);
2790 		error = 0;
2791 		*vpp = dvp;
2792 		goto out;
2793 	}
2794 
2795 	/*
2796 	 * Check for "." ie itself. this is a quick check and
2797 	 * avoids adding "." into the dnlc (which have been seen
2798 	 * to occupy >10% of the cache).
2799 	 */
2800 	if ((nm[0] == '.') && (nm[1] == 0)) {
2801 		/*
2802 		 * Don't return without checking accessibility
2803 		 * of the directory. We only need the lock if
2804 		 * we are going to return it.
2805 		 */
2806 		if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) == 0) {
2807 			VN_HOLD(dvp);
2808 			*vpp = dvp;
2809 		}
2810 		goto out;
2811 	}
2812 
2813 	/*
2814 	 * Fast path: Check the directory name lookup cache.
2815 	 */
2816 	if (vp = dnlc_lookup(dvp, nm)) {
2817 		/*
2818 		 * Check accessibility of directory.
2819 		 */
2820 		if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) != 0) {
2821 			VN_RELE(vp);
2822 			goto out;
2823 		}
2824 		if (vp == DNLC_NO_VNODE) {
2825 			VN_RELE(vp);
2826 			error = ENOENT;
2827 			goto out;
2828 		}
2829 		xip = VTOI(vp);
2830 		ulp = NULL;
2831 		goto fastpath;
2832 	}
2833 
2834 	/*
2835 	 * Keep the idle queue from getting too long by
2836 	 * idling two inodes before attempting to allocate another.
2837 	 *    This operation must be performed before entering
2838 	 *    lockfs or a transaction.
2839 	 */
2840 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2841 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2842 			ins.in_lidles.value.ul += ufs_lookup_idle_count;
2843 			ufs_idle_some(ufs_lookup_idle_count);
2844 		}
2845 
2846 retry_lookup:
2847 	/*
2848 	 * Check accessibility of directory.
2849 	 */
2850 	if (error = ufs_diraccess(ip, IEXEC, cr))
2851 		goto out;
2852 
2853 	ufsvfsp = ip->i_ufsvfs;
2854 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK);
2855 	if (error)
2856 		goto out;
2857 
2858 	error = ufs_dirlook(ip, nm, &xip, cr, 1, 0);
2859 
2860 fastpath:
2861 	if (error == 0) {
2862 		ip = xip;
2863 		*vpp = ITOV(ip);
2864 
2865 		/*
2866 		 * If vnode is a device return special vnode instead.
2867 		 */
2868 		if (IS_DEVVP(*vpp)) {
2869 			struct vnode *newvp;
2870 
2871 			newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
2872 			    cr);
2873 			VN_RELE(*vpp);
2874 			if (newvp == NULL)
2875 				error = ENOSYS;
2876 			else
2877 				*vpp = newvp;
2878 		} else if (ip->i_cflags & ICOMPRESS) {
2879 			struct vnode *newvp;
2880 
2881 			/*
2882 			 * Compressed file, substitute dcfs vnode
2883 			 */
2884 			newvp = decompvp(*vpp, cr, ct);
2885 			VN_RELE(*vpp);
2886 			if (newvp == NULL)
2887 				error = ENOSYS;
2888 			else
2889 				*vpp = newvp;
2890 		}
2891 	}
2892 	if (ulp) {
2893 		ufs_lockfs_end(ulp);
2894 	}
2895 
2896 	if (error == EAGAIN)
2897 		goto retry_lookup;
2898 
2899 out:
2900 	return (error);
2901 }
2902 
2903 /*ARGSUSED*/
2904 static int
2905 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl,
2906     int mode, struct vnode **vpp, struct cred *cr, int flag,
2907     caller_context_t *ct, vsecattr_t *vsecp)
2908 {
2909 	struct inode *ip;
2910 	struct inode *xip;
2911 	struct inode *dip;
2912 	struct vnode *xvp;
2913 	struct ufsvfs *ufsvfsp;
2914 	struct ulockfs *ulp;
2915 	int error;
2916 	int issync;
2917 	int truncflag;
2918 	int trans_size;
2919 	int noentry;
2920 	int defer_dip_seq_update = 0;	/* need to defer update of dip->i_seq */
2921 	int retry = 1;
2922 	int indeadlock;
2923 
2924 again:
2925 	ip = VTOI(dvp);
2926 	ufsvfsp = ip->i_ufsvfs;
2927 	truncflag = 0;
2928 
2929 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK);
2930 	if (error)
2931 		goto out;
2932 
2933 	if (ulp) {
2934 		trans_size = (int)TOP_CREATE_SIZE(ip);
2935 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size);
2936 	}
2937 
2938 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
2939 		vap->va_mode &= ~VSVTX;
2940 
2941 	if (*name == '\0') {
2942 		/*
2943 		 * Null component name refers to the directory itself.
2944 		 */
2945 		VN_HOLD(dvp);
2946 		/*
2947 		 * Even though this is an error case, we need to grab the
2948 		 * quota lock since the error handling code below is common.
2949 		 */
2950 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2951 		rw_enter(&ip->i_contents, RW_WRITER);
2952 		error = EEXIST;
2953 	} else {
2954 		xip = NULL;
2955 		noentry = 0;
2956 		/*
2957 		 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2958 		 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2959 		 * possible, retries the operation.
2960 		 */
2961 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_CREATE,
2962 		    retry_dir);
2963 		if (indeadlock)
2964 			goto again;
2965 
2966 		xvp = dnlc_lookup(dvp, name);
2967 		if (xvp == DNLC_NO_VNODE) {
2968 			noentry = 1;
2969 			VN_RELE(xvp);
2970 			xvp = NULL;
2971 		}
2972 		if (xvp) {
2973 			rw_exit(&ip->i_rwlock);
2974 			if (error = ufs_iaccess(ip, IEXEC, cr, 1)) {
2975 				VN_RELE(xvp);
2976 			} else {
2977 				error = EEXIST;
2978 				xip = VTOI(xvp);
2979 			}
2980 		} else {
2981 			/*
2982 			 * Suppress file system full message if we will retry
2983 			 */
2984 			error = ufs_direnter_cm(ip, name, DE_CREATE,
2985 			    vap, &xip, cr, (noentry | (retry ? IQUIET : 0)));
2986 			if (error == EAGAIN) {
2987 				if (ulp) {
2988 					TRANS_END_CSYNC(ufsvfsp, error, issync,
2989 					    TOP_CREATE, trans_size);
2990 					ufs_lockfs_end(ulp);
2991 				}
2992 				goto again;
2993 			}
2994 			rw_exit(&ip->i_rwlock);
2995 		}
2996 		ip = xip;
2997 		if (ip != NULL) {
2998 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2999 			rw_enter(&ip->i_contents, RW_WRITER);
3000 		}
3001 	}
3002 
3003 	/*
3004 	 * If the file already exists and this is a non-exclusive create,
3005 	 * check permissions and allow access for non-directories.
3006 	 * Read-only create of an existing directory is also allowed.
3007 	 * We fail an exclusive create of anything which already exists.
3008 	 */
3009 	if (error == EEXIST) {
3010 		dip = VTOI(dvp);
3011 		if (excl == NONEXCL) {
3012 			if ((((ip->i_mode & IFMT) == IFDIR) ||
3013 			    ((ip->i_mode & IFMT) == IFATTRDIR)) &&
3014 			    (mode & IWRITE))
3015 				error = EISDIR;
3016 			else if (mode)
3017 				error = ufs_iaccess(ip, mode, cr, 0);
3018 			else
3019 				error = 0;
3020 		}
3021 		if (error) {
3022 			rw_exit(&ip->i_contents);
3023 			rw_exit(&ufsvfsp->vfs_dqrwlock);
3024 			VN_RELE(ITOV(ip));
3025 			goto unlock;
3026 		}
3027 		/*
3028 		 * If the error EEXIST was set, then i_seq can not
3029 		 * have been updated. The sequence number interface
3030 		 * is defined such that a non-error VOP_CREATE must
3031 		 * increase the dir va_seq it by at least one. If we
3032 		 * have cleared the error, increase i_seq. Note that
3033 		 * we are increasing the dir i_seq and in rare cases
3034 		 * ip may actually be from the dvp, so we already have
3035 		 * the locks and it will not be subject to truncation.
3036 		 * In case we have to update i_seq of the parent
3037 		 * directory dip, we have to defer it till we have
3038 		 * released our locks on ip due to lock ordering requirements.
3039 		 */
3040 		if (ip != dip)
3041 			defer_dip_seq_update = 1;
3042 		else
3043 			ip->i_seq++;
3044 
3045 		if (((ip->i_mode & IFMT) == IFREG) &&
3046 		    (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
3047 			/*
3048 			 * Truncate regular files, if requested by caller.
3049 			 * Grab i_rwlock to make sure no one else is
3050 			 * currently writing to the file (we promised
3051 			 * bmap we would do this).
3052 			 * Must get the locks in the correct order.
3053 			 */
3054 			if (ip->i_size == 0) {
3055 				ip->i_flag |= ICHG | IUPD;
3056 				ip->i_seq++;
3057 				TRANS_INODE(ufsvfsp, ip);
3058 			} else {
3059 				/*
3060 				 * Large Files: Why this check here?
3061 				 * Though we do it in vn_create() we really
3062 				 * want to guarantee that we do not destroy
3063 				 * Large file data by atomically checking
3064 				 * the size while holding the contents
3065 				 * lock.
3066 				 */
3067 				if (flag && !(flag & FOFFMAX) &&
3068 				    ((ip->i_mode & IFMT) == IFREG) &&
3069 				    (ip->i_size > (offset_t)MAXOFF32_T)) {
3070 					rw_exit(&ip->i_contents);
3071 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3072 					error = EOVERFLOW;
3073 					goto unlock;
3074 				}
3075 				if (TRANS_ISTRANS(ufsvfsp))
3076 					truncflag++;
3077 				else {
3078 					rw_exit(&ip->i_contents);
3079 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3080 					ufs_tryirwlock_trans(&ip->i_rwlock,
3081 					    RW_WRITER, TOP_CREATE,
3082 					    retry_file);
3083 					if (indeadlock) {
3084 						VN_RELE(ITOV(ip));
3085 						goto again;
3086 					}
3087 					rw_enter(&ufsvfsp->vfs_dqrwlock,
3088 					    RW_READER);
3089 					rw_enter(&ip->i_contents, RW_WRITER);
3090 					(void) ufs_itrunc(ip, (u_offset_t)0, 0,
3091 					    cr);
3092 					rw_exit(&ip->i_rwlock);
3093 				}
3094 
3095 			}
3096 			if (error == 0) {
3097 				vnevent_create(ITOV(ip), ct);
3098 			}
3099 		}
3100 	}
3101 
3102 	if (error) {
3103 		if (ip != NULL) {
3104 			rw_exit(&ufsvfsp->vfs_dqrwlock);
3105 			rw_exit(&ip->i_contents);
3106 		}
3107 		goto unlock;
3108 	}
3109 
3110 	*vpp = ITOV(ip);
3111 	ITIMES(ip);
3112 	rw_exit(&ip->i_contents);
3113 	rw_exit(&ufsvfsp->vfs_dqrwlock);
3114 
3115 	/*
3116 	 * If vnode is a device return special vnode instead.
3117 	 */
3118 	if (!error && IS_DEVVP(*vpp)) {
3119 		struct vnode *newvp;
3120 
3121 		newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
3122 		VN_RELE(*vpp);
3123 		if (newvp == NULL) {
3124 			error = ENOSYS;
3125 			goto unlock;
3126 		}
3127 		truncflag = 0;
3128 		*vpp = newvp;
3129 	}
3130 unlock:
3131 
3132 	/*
3133 	 * Do the deferred update of the parent directory's sequence
3134 	 * number now.
3135 	 */
3136 	if (defer_dip_seq_update == 1) {
3137 		rw_enter(&dip->i_contents, RW_READER);
3138 		mutex_enter(&dip->i_tlock);
3139 		dip->i_seq++;
3140 		mutex_exit(&dip->i_tlock);
3141 		rw_exit(&dip->i_contents);
3142 	}
3143 
3144 	if (ulp) {
3145 		int terr = 0;
3146 
3147 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE,
3148 		    trans_size);
3149 
3150 		/*
3151 		 * If we haven't had a more interesting failure
3152 		 * already, then anything that might've happened
3153 		 * here should be reported.
3154 		 */
3155 		if (error == 0)
3156 			error = terr;
3157 	}
3158 
3159 	if (!error && truncflag) {
3160 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_trunc);
3161 		if (indeadlock) {
3162 			if (ulp)
3163 				ufs_lockfs_end(ulp);
3164 			VN_RELE(ITOV(ip));
3165 			goto again;
3166 		}
3167 		(void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr);
3168 		rw_exit(&ip->i_rwlock);
3169 	}
3170 
3171 	if (ulp)
3172 		ufs_lockfs_end(ulp);
3173 
3174 	/*
3175 	 * If no inodes available, try to free one up out of the
3176 	 * pending delete queue.
3177 	 */
3178 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3179 		ufs_delete_drain_wait(ufsvfsp, 1);
3180 		retry = 0;
3181 		goto again;
3182 	}
3183 
3184 out:
3185 	return (error);
3186 }
3187 
3188 extern int ufs_idle_max;
3189 /*ARGSUSED*/
3190 static int
3191 ufs_remove(struct vnode *vp, char *nm, struct cred *cr, caller_context_t *ct,
3192     int flags)
3193 {
3194 	struct inode *ip = VTOI(vp);
3195 	struct ufsvfs *ufsvfsp	= ip->i_ufsvfs;
3196 	struct ulockfs *ulp;
3197 	vnode_t *rmvp = NULL;	/* Vnode corresponding to name being removed */
3198 	int indeadlock;
3199 	int error;
3200 	int issync;
3201 	int trans_size;
3202 
3203 	/*
3204 	 * don't let the delete queue get too long
3205 	 */
3206 	if (ufsvfsp == NULL) {
3207 		error = EIO;
3208 		goto out;
3209 	}
3210 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3211 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3212 
3213 	error = ufs_eventlookup(vp, nm, cr, &rmvp);
3214 	if (rmvp != NULL) {
3215 		/* Only send the event if there were no errors */
3216 		if (error == 0)
3217 			vnevent_remove(rmvp, vp, nm, ct);
3218 		VN_RELE(rmvp);
3219 	}
3220 
3221 retry_remove:
3222 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK);
3223 	if (error)
3224 		goto out;
3225 
3226 	if (ulp)
3227 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE,
3228 		    trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp)));
3229 
3230 	/*
3231 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3232 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3233 	 * possible, retries the operation.
3234 	 */
3235 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_REMOVE, retry);
3236 	if (indeadlock)
3237 		goto retry_remove;
3238 	error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0,
3239 	    DR_REMOVE, cr);
3240 	rw_exit(&ip->i_rwlock);
3241 
3242 	if (ulp) {
3243 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size);
3244 		ufs_lockfs_end(ulp);
3245 	}
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);
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(struct vnode *sdvp, char *snm, struct vnode *tdvp, char *tnm,
3354     struct cred *cr, caller_context_t *ct, int flags)
3355 {
3356 	struct inode *sip = NULL;	/* source inode */
3357 	struct inode *ip = NULL;	/* check inode */
3358 	struct inode *sdp;		/* old (source) parent inode */
3359 	struct inode *tdp;		/* new (target) parent inode */
3360 	struct vnode *svp = NULL;	/* source vnode */
3361 	struct vnode *tvp = NULL;	/* target vnode, if it exists */
3362 	struct vnode *realvp;
3363 	struct ufsvfs *ufsvfsp;
3364 	struct ulockfs *ulp = NULL;
3365 	struct ufs_slot slot;
3366 	timestruc_t now;
3367 	int error;
3368 	int issync;
3369 	int trans_size;
3370 	krwlock_t *first_lock;
3371 	krwlock_t *second_lock;
3372 	krwlock_t *reverse_lock;
3373 	int serr, terr;
3374 
3375 	sdp = VTOI(sdvp);
3376 	slot.fbp = NULL;
3377 	ufsvfsp = sdp->i_ufsvfs;
3378 
3379 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3380 		tdvp = realvp;
3381 
3382 	/* Must do this before taking locks in case of DNLC miss */
3383 	terr = ufs_eventlookup(tdvp, tnm, cr, &tvp);
3384 	serr = ufs_eventlookup(sdvp, snm, cr, &svp);
3385 
3386 	if ((serr == 0) && ((terr == 0) || (terr == ENOENT))) {
3387 		if (tvp != NULL)
3388 			vnevent_pre_rename_dest(tvp, tdvp, tnm, ct);
3389 
3390 		/*
3391 		 * Notify the target directory of the rename event
3392 		 * if source and target directories are not the same.
3393 		 */
3394 		if (sdvp != tdvp)
3395 			vnevent_pre_rename_dest_dir(tdvp, svp, tnm, ct);
3396 
3397 		if (svp != NULL)
3398 			vnevent_pre_rename_src(svp, sdvp, snm, ct);
3399 	}
3400 
3401 	if (svp != NULL)
3402 		VN_RELE(svp);
3403 
3404 retry_rename:
3405 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
3406 	if (error)
3407 		goto unlock;
3408 
3409 	if (ulp)
3410 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
3411 		    trans_size = (int)TOP_RENAME_SIZE(sdp));
3412 
3413 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3414 		tdvp = realvp;
3415 
3416 	tdp = VTOI(tdvp);
3417 
3418 	/*
3419 	 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3420 	 */
3421 	if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) {
3422 		error = EINVAL;
3423 		goto unlock;
3424 	}
3425 
3426 	/*
3427 	 * Check accessibility of directory.
3428 	 */
3429 	if (error = ufs_diraccess(sdp, IEXEC, cr))
3430 		goto unlock;
3431 
3432 	/*
3433 	 * Look up inode of file we're supposed to rename.
3434 	 */
3435 	gethrestime(&now);
3436 	if (error = ufs_dirlook(sdp, snm, &sip, cr, 0, 0)) {
3437 		if (error == EAGAIN) {
3438 			if (ulp) {
3439 				TRANS_END_CSYNC(ufsvfsp, error, issync,
3440 				    TOP_RENAME, trans_size);
3441 				ufs_lockfs_end(ulp);
3442 			}
3443 			goto retry_rename;
3444 		}
3445 
3446 		goto unlock;
3447 	}
3448 
3449 	/*
3450 	 * Lock both the source and target directories (they may be
3451 	 * the same) to provide the atomicity semantics that was
3452 	 * previously provided by the per file system vfs_rename_lock
3453 	 *
3454 	 * with vfs_rename_lock removed to allow simultaneous renames
3455 	 * within a file system, ufs_dircheckpath can deadlock while
3456 	 * traversing back to ensure that source is not a parent directory
3457 	 * of target parent directory. This is because we get into
3458 	 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3459 	 * If the tdp and sdp of the simultaneous renames happen to be
3460 	 * in the path of each other, it can lead to a deadlock. This
3461 	 * can be avoided by getting the locks as RW_READER here and then
3462 	 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3463 	 *
3464 	 * We hold the target directory's i_rwlock after calling
3465 	 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3466 	 * VOP_RWLOCK is explicitly called by the filesystem independent code
3467 	 * before calling the file system operation. In these cases the order
3468 	 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3469 	 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3470 	 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3471 	 * synchronizing object which might lead to a deadlock. So we use
3472 	 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3473 	 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3474 	 * operation.
3475 	 */
3476 retry:
3477 	first_lock = &tdp->i_rwlock;
3478 	second_lock = &sdp->i_rwlock;
3479 retry_firstlock:
3480 	if (!rw_tryenter(first_lock, RW_READER)) {
3481 		/*
3482 		 * We didn't get the lock. Check if the SLOCK is set in the
3483 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3484 		 * and wait for SLOCK to be cleared.
3485 		 */
3486 
3487 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3488 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3489 			    trans_size);
3490 			ufs_lockfs_end(ulp);
3491 			goto retry_rename;
3492 
3493 		} else {
3494 			/*
3495 			 * SLOCK isn't set so this is a genuine synchronization
3496 			 * case. Let's try again after giving them a breather.
3497 			 */
3498 			delay(RETRY_LOCK_DELAY);
3499 			goto  retry_firstlock;
3500 		}
3501 	}
3502 	/*
3503 	 * Need to check if the tdp and sdp are same !!!
3504 	 */
3505 	if ((tdp != sdp) && (!rw_tryenter(second_lock, RW_READER))) {
3506 		/*
3507 		 * We didn't get the lock. Check if the SLOCK is set in the
3508 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3509 		 * and wait for SLOCK to be cleared.
3510 		 */
3511 
3512 		rw_exit(first_lock);
3513 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3514 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3515 			    trans_size);
3516 			ufs_lockfs_end(ulp);
3517 			goto retry_rename;
3518 
3519 		} else {
3520 			/*
3521 			 * So we couldn't get the second level peer lock *and*
3522 			 * the SLOCK bit isn't set. Too bad we can be
3523 			 * contentding with someone wanting these locks otherway
3524 			 * round. Reverse the locks in case there is a heavy
3525 			 * contention for the second level lock.
3526 			 */
3527 			reverse_lock = first_lock;
3528 			first_lock = second_lock;
3529 			second_lock = reverse_lock;
3530 			ufs_rename_retry_cnt++;
3531 			goto  retry_firstlock;
3532 		}
3533 	}
3534 
3535 	if (sip == tdp) {
3536 		error = EINVAL;
3537 		goto errout;
3538 	}
3539 	/*
3540 	 * Make sure we can delete the source entry.  This requires
3541 	 * write permission on the containing directory.
3542 	 * Check for sticky directories.
3543 	 */
3544 	rw_enter(&sdp->i_contents, RW_READER);
3545 	rw_enter(&sip->i_contents, RW_READER);
3546 	if ((error = ufs_iaccess(sdp, IWRITE, cr, 0)) != 0 ||
3547 	    (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) {
3548 		rw_exit(&sip->i_contents);
3549 		rw_exit(&sdp->i_contents);
3550 		goto errout;
3551 	}
3552 
3553 	/*
3554 	 * If this is a rename of a directory and the parent is
3555 	 * different (".." must be changed), then the source
3556 	 * directory must not be in the directory hierarchy
3557 	 * above the target, as this would orphan everything
3558 	 * below the source directory.  Also the user must have
3559 	 * write permission in the source so as to be able to
3560 	 * change "..".
3561 	 */
3562 	if ((((sip->i_mode & IFMT) == IFDIR) ||
3563 	    ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) {
3564 		ino_t	inum;
3565 
3566 		if (error = ufs_iaccess(sip, IWRITE, cr, 0)) {
3567 			rw_exit(&sip->i_contents);
3568 			rw_exit(&sdp->i_contents);
3569 			goto errout;
3570 		}
3571 		inum = sip->i_number;
3572 		rw_exit(&sip->i_contents);
3573 		rw_exit(&sdp->i_contents);
3574 		if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) {
3575 			/*
3576 			 * If we got EAGAIN ufs_dircheckpath detected a
3577 			 * potential deadlock and backed out. We need
3578 			 * to retry the operation since sdp and tdp have
3579 			 * to be released to avoid the deadlock.
3580 			 */
3581 			if (error == EAGAIN) {
3582 				rw_exit(&tdp->i_rwlock);
3583 				if (tdp != sdp)
3584 					rw_exit(&sdp->i_rwlock);
3585 				delay(ufs_rename_backoff_delay);
3586 				ufs_rename_dircheck_retry_cnt++;
3587 				goto retry;
3588 			}
3589 			goto errout;
3590 		}
3591 	} else {
3592 		rw_exit(&sip->i_contents);
3593 		rw_exit(&sdp->i_contents);
3594 	}
3595 
3596 
3597 	/*
3598 	 * Check for renaming '.' or '..' or alias of '.'
3599 	 */
3600 	if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) {
3601 		error = EINVAL;
3602 		goto errout;
3603 	}
3604 
3605 	/*
3606 	 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3607 	 * tries to traverse back the file tree with both tdp and sdp held
3608 	 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3609 	 * as RW_READERS  till ufs_dircheckpath is done.
3610 	 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3611 	 * to RW_WRITER.
3612 	 */
3613 	if (!rw_tryupgrade(&tdp->i_rwlock)) {
3614 		/*
3615 		 * The upgrade failed. We got to give away the lock
3616 		 * as to avoid deadlocking with someone else who is
3617 		 * waiting for writer lock. With the lock gone, we
3618 		 * cannot be sure the checks done above will hold
3619 		 * good when we eventually get them back as writer.
3620 		 * So if we can't upgrade we drop the locks and retry
3621 		 * everything again.
3622 		 */
3623 		rw_exit(&tdp->i_rwlock);
3624 		if (tdp != sdp)
3625 			rw_exit(&sdp->i_rwlock);
3626 		delay(ufs_rename_backoff_delay);
3627 		ufs_rename_upgrade_retry_cnt++;
3628 		goto retry;
3629 	}
3630 	if (tdp != sdp) {
3631 		if (!rw_tryupgrade(&sdp->i_rwlock)) {
3632 			/*
3633 			 * The upgrade failed. We got to give away the lock
3634 			 * as to avoid deadlocking with someone else who is
3635 			 * waiting for writer lock. With the lock gone, we
3636 			 * cannot be sure the checks done above will hold
3637 			 * good when we eventually get them back as writer.
3638 			 * So if we can't upgrade we drop the locks and retry
3639 			 * everything again.
3640 			 */
3641 			rw_exit(&tdp->i_rwlock);
3642 			rw_exit(&sdp->i_rwlock);
3643 			delay(ufs_rename_backoff_delay);
3644 			ufs_rename_upgrade_retry_cnt++;
3645 			goto retry;
3646 		}
3647 	}
3648 
3649 	/*
3650 	 * Now that all the locks are held check to make sure another thread
3651 	 * didn't slip in and take out the sip.
3652 	 */
3653 	slot.status = NONE;
3654 	if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec ||
3655 	    sip->i_ctime.tv_sec > now.tv_sec) {
3656 		rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER);
3657 		rw_enter(&sdp->i_contents, RW_WRITER);
3658 		error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot,
3659 		    &ip, cr, 0);
3660 		rw_exit(&sdp->i_contents);
3661 		rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock);
3662 		if (error) {
3663 			goto errout;
3664 		}
3665 		if (ip == NULL) {
3666 			error = ENOENT;
3667 			goto errout;
3668 		} else {
3669 			/*
3670 			 * If the inode was found need to drop the v_count
3671 			 * so as not to keep the filesystem from being
3672 			 * unmounted at a later time.
3673 			 */
3674 			VN_RELE(ITOV(ip));
3675 		}
3676 
3677 		/*
3678 		 * Release the slot.fbp that has the page mapped and
3679 		 * locked SE_SHARED, and could be used in in
3680 		 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3681 		 * on said page.
3682 		 */
3683 		if (slot.fbp) {
3684 			fbrelse(slot.fbp, S_OTHER);
3685 			slot.fbp = NULL;
3686 		}
3687 	}
3688 
3689 	/*
3690 	 * Link source to the target.
3691 	 */
3692 	if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr)) {
3693 		/*
3694 		 * ESAME isn't really an error; it indicates that the
3695 		 * operation should not be done because the source and target
3696 		 * are the same file, but that no error should be reported.
3697 		 */
3698 		if (error == ESAME)
3699 			error = 0;
3700 		goto errout;
3701 	}
3702 
3703 	if (error == 0 && tvp != NULL)
3704 		vnevent_rename_dest(tvp, tdvp, tnm, ct);
3705 
3706 	/*
3707 	 * Unlink the source.
3708 	 * Remove the source entry.  ufs_dirremove() checks that the entry
3709 	 * still reflects sip, and returns an error if it doesn't.
3710 	 * If the entry has changed just forget about it.  Release
3711 	 * the source inode.
3712 	 */
3713 	if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0,
3714 	    DR_RENAME, cr)) == ENOENT)
3715 		error = 0;
3716 
3717 	if (error == 0) {
3718 		vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
3719 		/*
3720 		 * Notify the target directory of the rename event
3721 		 * if source and target directories are not the same.
3722 		 */
3723 		if (sdvp != tdvp)
3724 			vnevent_rename_dest_dir(tdvp, ct);
3725 	}
3726 
3727 errout:
3728 	if (slot.fbp)
3729 		fbrelse(slot.fbp, S_OTHER);
3730 
3731 	rw_exit(&tdp->i_rwlock);
3732 	if (sdp != tdp) {
3733 		rw_exit(&sdp->i_rwlock);
3734 	}
3735 
3736 unlock:
3737 	if (tvp != NULL)
3738 		VN_RELE(tvp);
3739 	if (sip != NULL)
3740 		VN_RELE(ITOV(sip));
3741 
3742 	if (ulp) {
3743 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
3744 		ufs_lockfs_end(ulp);
3745 	}
3746 
3747 	return (error);
3748 }
3749 
3750 /*ARGSUSED*/
3751 static int
3752 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap,
3753     struct vnode **vpp, struct cred *cr, caller_context_t *ct, int flags,
3754     vsecattr_t *vsecp)
3755 {
3756 	struct inode *ip;
3757 	struct inode *xip;
3758 	struct ufsvfs *ufsvfsp;
3759 	struct ulockfs *ulp;
3760 	int error;
3761 	int issync;
3762 	int trans_size;
3763 	int indeadlock;
3764 	int retry = 1;
3765 
3766 	ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
3767 
3768 	/*
3769 	 * Can't make directory in attr hidden dir
3770 	 */
3771 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3772 		return (EINVAL);
3773 
3774 again:
3775 	ip = VTOI(dvp);
3776 	ufsvfsp = ip->i_ufsvfs;
3777 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK);
3778 	if (error)
3779 		goto out;
3780 	if (ulp)
3781 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR,
3782 		    trans_size = (int)TOP_MKDIR_SIZE(ip));
3783 
3784 	/*
3785 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3786 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3787 	 * possible, retries the operation.
3788 	 */
3789 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_MKDIR, retry);
3790 	if (indeadlock)
3791 		goto again;
3792 
3793 	error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr,
3794 	    (retry ? IQUIET : 0));
3795 	if (error == EAGAIN) {
3796 		if (ulp) {
3797 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_MKDIR,
3798 			    trans_size);
3799 			ufs_lockfs_end(ulp);
3800 		}
3801 		goto again;
3802 	}
3803 
3804 	rw_exit(&ip->i_rwlock);
3805 	if (error == 0) {
3806 		ip = xip;
3807 		*vpp = ITOV(ip);
3808 	} else if (error == EEXIST)
3809 		VN_RELE(ITOV(xip));
3810 
3811 	if (ulp) {
3812 		int terr = 0;
3813 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size);
3814 		ufs_lockfs_end(ulp);
3815 		if (error == 0)
3816 			error = terr;
3817 	}
3818 out:
3819 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3820 		ufs_delete_drain_wait(ufsvfsp, 1);
3821 		retry = 0;
3822 		goto again;
3823 	}
3824 
3825 	return (error);
3826 }
3827 
3828 /*ARGSUSED*/
3829 static int
3830 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr,
3831     caller_context_t *ct, int flags)
3832 {
3833 	struct inode *ip = VTOI(vp);
3834 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3835 	struct ulockfs *ulp;
3836 	vnode_t *rmvp = NULL;	/* Vnode of removed directory */
3837 	int error;
3838 	int issync;
3839 	int trans_size;
3840 	int indeadlock;
3841 
3842 	/*
3843 	 * don't let the delete queue get too long
3844 	 */
3845 	if (ufsvfsp == NULL) {
3846 		error = EIO;
3847 		goto out;
3848 	}
3849 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3850 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3851 
3852 	error = ufs_eventlookup(vp, nm, cr, &rmvp);
3853 	if (rmvp != NULL) {
3854 		/* Only send the event if there were no errors */
3855 		if (error == 0)
3856 			vnevent_rmdir(rmvp, vp, nm, ct);
3857 		VN_RELE(rmvp);
3858 	}
3859 
3860 retry_rmdir:
3861 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK);
3862 	if (error)
3863 		goto out;
3864 
3865 	if (ulp)
3866 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR,
3867 		    trans_size = TOP_RMDIR_SIZE);
3868 
3869 	/*
3870 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3871 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3872 	 * possible, retries the operation.
3873 	 */
3874 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_RMDIR, retry);
3875 	if (indeadlock)
3876 		goto retry_rmdir;
3877 	error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr);
3878 
3879 	rw_exit(&ip->i_rwlock);
3880 
3881 	if (ulp) {
3882 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR,
3883 		    trans_size);
3884 		ufs_lockfs_end(ulp);
3885 	}
3886 
3887 out:
3888 	return (error);
3889 }
3890 
3891 /* ARGSUSED */
3892 static int
3893 ufs_readdir(struct vnode *vp, struct uio *uiop, struct cred *cr, int *eofp,
3894     caller_context_t *ct, 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(struct vnode *dvp, char *linkname, struct vattr *vap, char *target,
4098     struct cred *cr, caller_context_t *ct, int flags)
4099 {
4100 	struct inode *ip, *dip = VTOI(dvp);
4101 	struct ufsvfs *ufsvfsp = dip->i_ufsvfs;
4102 	struct ulockfs *ulp;
4103 	int error;
4104 	int issync;
4105 	int trans_size;
4106 	int residual;
4107 	int ioflag;
4108 	int retry = 1;
4109 
4110 	/*
4111 	 * No symlinks in attrdirs at this time
4112 	 */
4113 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
4114 		return (EINVAL);
4115 
4116 again:
4117 	ip = (struct inode *)NULL;
4118 	vap->va_type = VLNK;
4119 	vap->va_rdev = 0;
4120 
4121 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK);
4122 	if (error)
4123 		goto out;
4124 
4125 	if (ulp)
4126 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK,
4127 		    trans_size = (int)TOP_SYMLINK_SIZE(dip));
4128 
4129 	/*
4130 	 * We must create the inode before the directory entry, to avoid
4131 	 * racing with readlink().  ufs_dirmakeinode requires that we
4132 	 * hold the quota lock as reader, and directory locks as writer.
4133 	 */
4134 
4135 	rw_enter(&dip->i_rwlock, RW_WRITER);
4136 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4137 	rw_enter(&dip->i_contents, RW_WRITER);
4138 
4139 	/*
4140 	 * Suppress any out of inodes messages if we will retry on
4141 	 * ENOSP
4142 	 */
4143 	if (retry)
4144 		dip->i_flag |= IQUIET;
4145 
4146 	error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr);
4147 
4148 	dip->i_flag &= ~IQUIET;
4149 
4150 	rw_exit(&dip->i_contents);
4151 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4152 	rw_exit(&dip->i_rwlock);
4153 
4154 	if (error)
4155 		goto unlock;
4156 
4157 	/*
4158 	 * OK.  The inode has been created.  Write out the data of the
4159 	 * symbolic link.  Since symbolic links are metadata, and should
4160 	 * remain consistent across a system crash, we need to force the
4161 	 * data out synchronously.
4162 	 *
4163 	 * (This is a change from the semantics in earlier releases, which
4164 	 * only created symbolic links synchronously if the semi-documented
4165 	 * 'syncdir' option was set, or if we were being invoked by the NFS
4166 	 * server, which requires symbolic links to be created synchronously.)
4167 	 *
4168 	 * We need to pass in a pointer for the residual length; otherwise
4169 	 * ufs_rdwri() will always return EIO if it can't write the data,
4170 	 * even if the error was really ENOSPC or EDQUOT.
4171 	 */
4172 
4173 	ioflag = FWRITE | FDSYNC;
4174 	residual = 0;
4175 
4176 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4177 	rw_enter(&ip->i_contents, RW_WRITER);
4178 
4179 	/*
4180 	 * Suppress file system full messages if we will retry
4181 	 */
4182 	if (retry)
4183 		ip->i_flag |= IQUIET;
4184 
4185 	error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target),
4186 	    (offset_t)0, UIO_SYSSPACE, &residual, cr);
4187 
4188 	ip->i_flag &= ~IQUIET;
4189 
4190 	if (error) {
4191 		rw_exit(&ip->i_contents);
4192 		rw_exit(&ufsvfsp->vfs_dqrwlock);
4193 		goto remove;
4194 	}
4195 
4196 	/*
4197 	 * If the link's data is small enough, we can cache it in the inode.
4198 	 * This is a "fast symbolic link".  We don't use the first direct
4199 	 * block because that's actually used to point at the symbolic link's
4200 	 * contents on disk; but we know that none of the other direct or
4201 	 * indirect blocks can be used because symbolic links are restricted
4202 	 * to be smaller than a file system block.
4203 	 */
4204 
4205 	ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip)));
4206 
4207 	if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) {
4208 		if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) {
4209 			ip->i_flag |= IFASTSYMLNK;
4210 		} else {
4211 			int i;
4212 			/* error, clear garbage left behind */
4213 			for (i = 1; i < NDADDR; i++)
4214 				ip->i_db[i] = 0;
4215 			for (i = 0; i < NIADDR; i++)
4216 				ip->i_ib[i] = 0;
4217 		}
4218 	}
4219 
4220 	rw_exit(&ip->i_contents);
4221 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4222 
4223 	/*
4224 	 * OK.  We've successfully created the symbolic link.  All that
4225 	 * remains is to insert it into the appropriate directory.
4226 	 */
4227 
4228 	rw_enter(&dip->i_rwlock, RW_WRITER);
4229 	error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr);
4230 	rw_exit(&dip->i_rwlock);
4231 
4232 	/*
4233 	 * Fall through into remove-on-error code.  We're either done, or we
4234 	 * need to remove the inode (if we couldn't insert it).
4235 	 */
4236 
4237 remove:
4238 	if (error && (ip != NULL)) {
4239 		rw_enter(&ip->i_contents, RW_WRITER);
4240 		ip->i_nlink--;
4241 		ip->i_flag |= ICHG;
4242 		ip->i_seq++;
4243 		ufs_setreclaim(ip);
4244 		rw_exit(&ip->i_contents);
4245 	}
4246 
4247 unlock:
4248 	if (ip != NULL)
4249 		VN_RELE(ITOV(ip));
4250 
4251 	if (ulp) {
4252 		int terr = 0;
4253 
4254 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK,
4255 		    trans_size);
4256 		ufs_lockfs_end(ulp);
4257 		if (error == 0)
4258 			error = terr;
4259 	}
4260 
4261 	/*
4262 	 * We may have failed due to lack of an inode or of a block to
4263 	 * store the target in.  Try flushing the delete queue to free
4264 	 * logically-available things up and try again.
4265 	 */
4266 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
4267 		ufs_delete_drain_wait(ufsvfsp, 1);
4268 		retry = 0;
4269 		goto again;
4270 	}
4271 
4272 out:
4273 	return (error);
4274 }
4275 
4276 /*
4277  * Ufs specific routine used to do ufs io.
4278  */
4279 int
4280 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base,
4281     ssize_t len, offset_t offset, enum uio_seg seg, int *aresid,
4282     struct cred *cr)
4283 {
4284 	struct uio auio;
4285 	struct iovec aiov;
4286 	int error;
4287 
4288 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
4289 
4290 	bzero((caddr_t)&auio, sizeof (uio_t));
4291 	bzero((caddr_t)&aiov, sizeof (iovec_t));
4292 
4293 	aiov.iov_base = base;
4294 	aiov.iov_len = len;
4295 	auio.uio_iov = &aiov;
4296 	auio.uio_iovcnt = 1;
4297 	auio.uio_loffset = offset;
4298 	auio.uio_segflg = (short)seg;
4299 	auio.uio_resid = len;
4300 
4301 	if (rw == UIO_WRITE) {
4302 		auio.uio_fmode = FWRITE;
4303 		auio.uio_extflg = UIO_COPY_DEFAULT;
4304 		auio.uio_llimit = curproc->p_fsz_ctl;
4305 		error = wrip(ip, &auio, ioflag, cr);
4306 	} else {
4307 		auio.uio_fmode = FREAD;
4308 		auio.uio_extflg = UIO_COPY_CACHED;
4309 		auio.uio_llimit = MAXOFFSET_T;
4310 		error = rdip(ip, &auio, ioflag, cr);
4311 	}
4312 
4313 	if (aresid) {
4314 		*aresid = auio.uio_resid;
4315 	} else if (auio.uio_resid) {
4316 		error = EIO;
4317 	}
4318 	return (error);
4319 }
4320 
4321 /*ARGSUSED*/
4322 static int
4323 ufs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
4324 {
4325 	struct ufid *ufid;
4326 	struct inode *ip = VTOI(vp);
4327 
4328 	if (ip->i_ufsvfs == NULL)
4329 		return (EIO);
4330 
4331 	if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) {
4332 		fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t);
4333 		return (ENOSPC);
4334 	}
4335 
4336 	ufid = (struct ufid *)fidp;
4337 	bzero((char *)ufid, sizeof (struct ufid));
4338 	ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t);
4339 	ufid->ufid_ino = ip->i_number;
4340 	ufid->ufid_gen = ip->i_gen;
4341 
4342 	return (0);
4343 }
4344 
4345 /* ARGSUSED2 */
4346 static int
4347 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4348 {
4349 	struct inode	*ip = VTOI(vp);
4350 	struct ufsvfs	*ufsvfsp;
4351 	int		forcedirectio;
4352 
4353 	/*
4354 	 * Read case is easy.
4355 	 */
4356 	if (!write_lock) {
4357 		rw_enter(&ip->i_rwlock, RW_READER);
4358 		return (V_WRITELOCK_FALSE);
4359 	}
4360 
4361 	/*
4362 	 * Caller has requested a writer lock, but that inhibits any
4363 	 * concurrency in the VOPs that follow. Acquire the lock shared
4364 	 * and defer exclusive access until it is known to be needed in
4365 	 * other VOP handlers. Some cases can be determined here.
4366 	 */
4367 
4368 	/*
4369 	 * If directio is not set, there is no chance of concurrency,
4370 	 * so just acquire the lock exclusive. Beware of a forced
4371 	 * unmount before looking at the mount option.
4372 	 */
4373 	ufsvfsp = ip->i_ufsvfs;
4374 	forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0;
4375 	if (!(ip->i_flag & IDIRECTIO || forcedirectio) ||
4376 	    !ufs_allow_shared_writes) {
4377 		rw_enter(&ip->i_rwlock, RW_WRITER);
4378 		return (V_WRITELOCK_TRUE);
4379 	}
4380 
4381 	/*
4382 	 * Mandatory locking forces acquiring i_rwlock exclusive.
4383 	 */
4384 	if (MANDLOCK(vp, ip->i_mode)) {
4385 		rw_enter(&ip->i_rwlock, RW_WRITER);
4386 		return (V_WRITELOCK_TRUE);
4387 	}
4388 
4389 	/*
4390 	 * Acquire the lock shared in case a concurrent write follows.
4391 	 * Mandatory locking could have become enabled before the lock
4392 	 * was acquired. Re-check and upgrade if needed.
4393 	 */
4394 	rw_enter(&ip->i_rwlock, RW_READER);
4395 	if (MANDLOCK(vp, ip->i_mode)) {
4396 		rw_exit(&ip->i_rwlock);
4397 		rw_enter(&ip->i_rwlock, RW_WRITER);
4398 		return (V_WRITELOCK_TRUE);
4399 	}
4400 	return (V_WRITELOCK_FALSE);
4401 }
4402 
4403 /*ARGSUSED*/
4404 static void
4405 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4406 {
4407 	struct inode	*ip = VTOI(vp);
4408 
4409 	rw_exit(&ip->i_rwlock);
4410 }
4411 
4412 /* ARGSUSED */
4413 static int
4414 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
4415 {
4416 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4417 }
4418 
4419 /* ARGSUSED */
4420 static int
4421 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4422     offset_t offset, struct flk_callback *flk_cbp, struct cred *cr,
4423     caller_context_t *ct)
4424 {
4425 	struct inode *ip = VTOI(vp);
4426 
4427 	if (ip->i_ufsvfs == NULL)
4428 		return (EIO);
4429 
4430 	/*
4431 	 * If file is being mapped, disallow frlock.
4432 	 * XXX I am not holding tlock while checking i_mapcnt because the
4433 	 * current locking strategy drops all locks before calling fs_frlock.
4434 	 * So, mapcnt could change before we enter fs_frlock making is
4435 	 * meaningless to have held tlock in the first place.
4436 	 */
4437 	if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode))
4438 		return (EAGAIN);
4439 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4440 }
4441 
4442 /* ARGSUSED */
4443 static int
4444 ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4445     offset_t offset, cred_t *cr, caller_context_t *ct)
4446 {
4447 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
4448 	struct ulockfs *ulp;
4449 	int error;
4450 
4451 	if ((error = convoff(vp, bfp, 0, offset)) == 0) {
4452 		if (cmd == F_FREESP) {
4453 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4454 			    ULOCKFS_SPACE_MASK);
4455 			if (error)
4456 				return (error);
4457 			error = ufs_freesp(vp, bfp, flag, cr);
4458 
4459 			if (error == 0 && bfp->l_start == 0)
4460 				vnevent_truncate(vp, ct);
4461 		} else if (cmd == F_ALLOCSP) {
4462 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4463 			    ULOCKFS_FALLOCATE_MASK);
4464 			if (error)
4465 				return (error);
4466 			error = ufs_allocsp(vp, bfp, cr);
4467 		} else
4468 			return (EINVAL); /* Command not handled here */
4469 
4470 		if (ulp)
4471 			ufs_lockfs_end(ulp);
4472 
4473 	}
4474 	return (error);
4475 }
4476 
4477 /*
4478  * Used to determine if read ahead should be done. Also used to
4479  * to determine when write back occurs.
4480  */
4481 #define	CLUSTSZ(ip)		((ip)->i_ufsvfs->vfs_ioclustsz)
4482 
4483 /*
4484  * A faster version of ufs_getpage.
4485  *
4486  * We optimize by inlining the pvn_getpages iterator, eliminating
4487  * calls to bmap_read if file doesn't have UFS holes, and avoiding
4488  * the overhead of page_exists().
4489  *
4490  * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4491  * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4492  * victimizes performance when a file with UFS holes is faulted
4493  * first in the S_READ mode, and then in the S_WRITE mode. We will get
4494  * two MMU faults in this case.
4495  *
4496  * XXX - the inode fields which control the sequential mode are not
4497  *	 protected by any mutex. The read ahead will act wild if
4498  *	 multiple processes will access the file concurrently and
4499  *	 some of them in sequential mode. One particulary bad case
4500  *	 is if another thread will change the value of i_nextrio between
4501  *	 the time this thread tests the i_nextrio value and then reads it
4502  *	 again to use it as the offset for the read ahead.
4503  */
4504 /*ARGSUSED*/
4505 static int
4506 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
4507     page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr,
4508     enum seg_rw rw, struct cred *cr, caller_context_t *ct)
4509 {
4510 	u_offset_t	uoff = (u_offset_t)off; /* type conversion */
4511 	u_offset_t	pgoff;
4512 	u_offset_t	eoff;
4513 	struct inode 	*ip = VTOI(vp);
4514 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
4515 	struct fs 	*fs;
4516 	struct ulockfs	*ulp;
4517 	page_t		**pl;
4518 	caddr_t		pgaddr;
4519 	krw_t		rwtype;
4520 	int 		err;
4521 	int		has_holes;
4522 	int		beyond_eof;
4523 	int		seqmode;
4524 	int		pgsize = PAGESIZE;
4525 	int		dolock;
4526 	int		do_qlock;
4527 	int		trans_size;
4528 
4529 	ASSERT((uoff & PAGEOFFSET) == 0);
4530 
4531 	if (protp)
4532 		*protp = PROT_ALL;
4533 
4534 	/*
4535 	 * Obey the lockfs protocol
4536 	 */
4537 	err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg,
4538 	    rw == S_READ || rw == S_EXEC, protp);
4539 	if (err)
4540 		goto out;
4541 
4542 	fs = ufsvfsp->vfs_fs;
4543 
4544 	if (ulp && (rw == S_CREATE || rw == S_WRITE) &&
4545 	    !(vp->v_flag & VISSWAP)) {
4546 		/*
4547 		 * Try to start a transaction, will return if blocking is
4548 		 * expected to occur and the address space is not the
4549 		 * kernel address space.
4550 		 */
4551 		trans_size = TOP_GETPAGE_SIZE(ip);
4552 		if (seg->s_as != &kas) {
4553 			TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE,
4554 			    trans_size, err)
4555 			if (err == EWOULDBLOCK) {
4556 				/*
4557 				 * Use EDEADLK here because the VM code
4558 				 * can normally never see this error.
4559 				 */
4560 				err = EDEADLK;
4561 				ufs_lockfs_end(ulp);
4562 				goto out;
4563 			}
4564 		} else {
4565 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4566 		}
4567 	}
4568 
4569 	if (vp->v_flag & VNOMAP) {
4570 		err = ENOSYS;
4571 		goto unlock;
4572 	}
4573 
4574 	seqmode = ip->i_nextr == uoff && rw != S_CREATE;
4575 
4576 	rwtype = RW_READER;		/* start as a reader */
4577 	dolock = (rw_owner(&ip->i_contents) != curthread);
4578 	/*
4579 	 * If this thread owns the lock, i.e., this thread grabbed it
4580 	 * as writer somewhere above, then we don't need to grab the
4581 	 * lock as reader in this routine.
4582 	 */
4583 	do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread);
4584 
4585 retrylock:
4586 	if (dolock) {
4587 		/*
4588 		 * Grab the quota lock if we need to call
4589 		 * bmap_write() below (with i_contents as writer).
4590 		 */
4591 		if (do_qlock && rwtype == RW_WRITER)
4592 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4593 		rw_enter(&ip->i_contents, rwtype);
4594 	}
4595 
4596 	/*
4597 	 * We may be getting called as a side effect of a bmap using
4598 	 * fbread() when the blocks might be being allocated and the
4599 	 * size has not yet been up'ed.  In this case we want to be
4600 	 * able to return zero pages if we get back UFS_HOLE from
4601 	 * calling bmap for a non write case here.  We also might have
4602 	 * to read some frags from the disk into a page if we are
4603 	 * extending the number of frags for a given lbn in bmap().
4604 	 * Large Files: The read of i_size here is atomic because
4605 	 * i_contents is held here. If dolock is zero, the lock
4606 	 * is held in bmap routines.
4607 	 */
4608 	beyond_eof = uoff + len >
4609 	    P2ROUNDUP_TYPED(ip->i_size, PAGESIZE, u_offset_t);
4610 	if (beyond_eof && seg != segkmap) {
4611 		if (dolock) {
4612 			rw_exit(&ip->i_contents);
4613 			if (do_qlock && rwtype == RW_WRITER)
4614 				rw_exit(&ufsvfsp->vfs_dqrwlock);
4615 		}
4616 		err = EFAULT;
4617 		goto unlock;
4618 	}
4619 
4620 	/*
4621 	 * Must hold i_contents lock throughout the call to pvn_getpages
4622 	 * since locked pages are returned from each call to ufs_getapage.
4623 	 * Must *not* return locked pages and then try for contents lock
4624 	 * due to lock ordering requirements (inode > page)
4625 	 */
4626 
4627 	has_holes = bmap_has_holes(ip);
4628 
4629 	if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) {
4630 		int	blk_size;
4631 		u_offset_t offset;
4632 
4633 		/*
4634 		 * We must acquire the RW_WRITER lock in order to
4635 		 * call bmap_write().
4636 		 */
4637 		if (dolock && rwtype == RW_READER) {
4638 			rwtype = RW_WRITER;
4639 
4640 			/*
4641 			 * Grab the quota lock before
4642 			 * upgrading i_contents, but if we can't grab it
4643 			 * don't wait here due to lock order:
4644 			 * vfs_dqrwlock > i_contents.
4645 			 */
4646 			if (do_qlock &&
4647 			    rw_tryenter(&ufsvfsp->vfs_dqrwlock, RW_READER)
4648 			    == 0) {
4649 				rw_exit(&ip->i_contents);
4650 				goto retrylock;
4651 			}
4652 			if (!rw_tryupgrade(&ip->i_contents)) {
4653 				rw_exit(&ip->i_contents);
4654 				if (do_qlock)
4655 					rw_exit(&ufsvfsp->vfs_dqrwlock);
4656 				goto retrylock;
4657 			}
4658 		}
4659 
4660 		/*
4661 		 * May be allocating disk blocks for holes here as
4662 		 * a result of mmap faults. write(2) does the bmap_write
4663 		 * in rdip/wrip, not here. We are not dealing with frags
4664 		 * in this case.
4665 		 */
4666 		/*
4667 		 * Large Files: We cast fs_bmask field to offset_t
4668 		 * just as we do for MAXBMASK because uoff is a 64-bit
4669 		 * data type. fs_bmask will still be a 32-bit type
4670 		 * as we cannot change any ondisk data structures.
4671 		 */
4672 
4673 		offset = uoff & (offset_t)fs->fs_bmask;
4674 		while (offset < uoff + len) {
4675 			blk_size = (int)blksize(fs, ip, lblkno(fs, offset));
4676 			err = bmap_write(ip, offset, blk_size,
4677 			    BI_NORMAL, NULL, cr);
4678 			if (ip->i_flag & (ICHG|IUPD))
4679 				ip->i_seq++;
4680 			if (err)
4681 				goto update_inode;
4682 			offset += blk_size; /* XXX - make this contig */
4683 		}
4684 	}
4685 
4686 	/*
4687 	 * Can be a reader from now on.
4688 	 */
4689 	if (dolock && rwtype == RW_WRITER) {
4690 		rw_downgrade(&ip->i_contents);
4691 		/*
4692 		 * We can release vfs_dqrwlock early so do it, but make
4693 		 * sure we don't try to release it again at the bottom.
4694 		 */
4695 		if (do_qlock) {
4696 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4697 			do_qlock = 0;
4698 		}
4699 	}
4700 
4701 	/*
4702 	 * We remove PROT_WRITE in cases when the file has UFS holes
4703 	 * because we don't  want to call bmap_read() to check each
4704 	 * page if it is backed with a disk block.
4705 	 */
4706 	if (protp && has_holes && rw != S_WRITE && rw != S_CREATE)
4707 		*protp &= ~PROT_WRITE;
4708 
4709 	err = 0;
4710 
4711 	/*
4712 	 * The loop looks up pages in the range [off, off + len).
4713 	 * For each page, we first check if we should initiate an asynchronous
4714 	 * read ahead before we call page_lookup (we may sleep in page_lookup
4715 	 * for a previously initiated disk read).
4716 	 */
4717 	eoff = (uoff + len);
4718 	for (pgoff = uoff, pgaddr = addr, pl = plarr;
4719 	    pgoff < eoff; /* empty */) {
4720 		page_t	*pp;
4721 		u_offset_t	nextrio;
4722 		se_t	se;
4723 		int retval;
4724 
4725 		se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED);
4726 
4727 		/* Handle async getpage (faultahead) */
4728 		if (plarr == NULL) {
4729 			ip->i_nextrio = pgoff;
4730 			(void) ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4731 			pgoff += pgsize;
4732 			pgaddr += pgsize;
4733 			continue;
4734 		}
4735 		/*
4736 		 * Check if we should initiate read ahead of next cluster.
4737 		 * We call page_exists only when we need to confirm that
4738 		 * we have the current page before we initiate the read ahead.
4739 		 */
4740 		nextrio = ip->i_nextrio;
4741 		if (seqmode &&
4742 		    pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
4743 		    nextrio < ip->i_size && page_exists(vp, pgoff)) {
4744 			retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4745 			/*
4746 			 * We always read ahead the next cluster of data
4747 			 * starting from i_nextrio. If the page (vp,nextrio)
4748 			 * is actually in core at this point, the routine
4749 			 * ufs_getpage_ra() will stop pre-fetching data
4750 			 * until we read that page in a synchronized manner
4751 			 * through ufs_getpage_miss(). So, we should increase
4752 			 * i_nextrio if the page (vp, nextrio) exists.
4753 			 */
4754 			if ((retval == 0) && page_exists(vp, nextrio)) {
4755 				ip->i_nextrio = nextrio + pgsize;
4756 			}
4757 		}
4758 
4759 		if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
4760 			/*
4761 			 * We found the page in the page cache.
4762 			 */
4763 			*pl++ = pp;
4764 			pgoff += pgsize;
4765 			pgaddr += pgsize;
4766 			len -= pgsize;
4767 			plsz -= pgsize;
4768 		} else  {
4769 			/*
4770 			 * We have to create the page, or read it from disk.
4771 			 */
4772 			if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr,
4773 			    pl, plsz, rw, seqmode))
4774 				goto error;
4775 
4776 			while (*pl != NULL) {
4777 				pl++;
4778 				pgoff += pgsize;
4779 				pgaddr += pgsize;
4780 				len -= pgsize;
4781 				plsz -= pgsize;
4782 			}
4783 		}
4784 	}
4785 
4786 	/*
4787 	 * Return pages up to plsz if they are in the page cache.
4788 	 * We cannot return pages if there is a chance that they are
4789 	 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4790 	 */
4791 	if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
4792 
4793 		ASSERT((protp == NULL) ||
4794 		    !(has_holes && (*protp & PROT_WRITE)));
4795 
4796 		eoff = pgoff + plsz;
4797 		while (pgoff < eoff) {
4798 			page_t		*pp;
4799 
4800 			if ((pp = page_lookup_nowait(vp, pgoff,
4801 			    SE_SHARED)) == NULL)
4802 				break;
4803 
4804 			*pl++ = pp;
4805 			pgoff += pgsize;
4806 			plsz -= pgsize;
4807 		}
4808 	}
4809 
4810 	if (plarr)
4811 		*pl = NULL;			/* Terminate page list */
4812 	ip->i_nextr = pgoff;
4813 
4814 error:
4815 	if (err && plarr) {
4816 		/*
4817 		 * Release any pages we have locked.
4818 		 */
4819 		while (pl > &plarr[0])
4820 			page_unlock(*--pl);
4821 
4822 		plarr[0] = NULL;
4823 	}
4824 
4825 update_inode:
4826 	/*
4827 	 * If the inode is not already marked for IACC (in rdip() for read)
4828 	 * and the inode is not marked for no access time update (in wrip()
4829 	 * for write) then update the inode access time and mod time now.
4830 	 */
4831 	if ((ip->i_flag & (IACC | INOACC)) == 0) {
4832 		if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) {
4833 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
4834 			    (fs->fs_ronly == 0) &&
4835 			    (!ufsvfsp->vfs_noatime)) {
4836 				mutex_enter(&ip->i_tlock);
4837 				ip->i_flag |= IACC;
4838 				ITIMES_NOLOCK(ip);
4839 				mutex_exit(&ip->i_tlock);
4840 			}
4841 		}
4842 	}
4843 
4844 	if (dolock) {
4845 		rw_exit(&ip->i_contents);
4846 		if (do_qlock && rwtype == RW_WRITER)
4847 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4848 	}
4849 
4850 unlock:
4851 	if (ulp) {
4852 		if ((rw == S_CREATE || rw == S_WRITE) &&
4853 		    !(vp->v_flag & VISSWAP)) {
4854 			TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4855 		}
4856 		ufs_lockfs_end(ulp);
4857 	}
4858 out:
4859 	return (err);
4860 }
4861 
4862 /*
4863  * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4864  * cache. The page is either read from the disk, or it's created.
4865  * A page is created (without disk read) if rw == S_CREATE, or if
4866  * the page is not backed with a real disk block (UFS hole).
4867  */
4868 /* ARGSUSED */
4869 static int
4870 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg,
4871     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq)
4872 {
4873 	struct inode	*ip = VTOI(vp);
4874 	page_t		*pp;
4875 	daddr_t		bn;
4876 	size_t		io_len;
4877 	int		crpage = 0;
4878 	int		err;
4879 	int		contig;
4880 	int		bsize = ip->i_fs->fs_bsize;
4881 
4882 	/*
4883 	 * Figure out whether the page can be created, or must be
4884 	 * must be read from the disk.
4885 	 */
4886 	if (rw == S_CREATE)
4887 		crpage = 1;
4888 	else {
4889 		contig = 0;
4890 		if (err = bmap_read(ip, off, &bn, &contig))
4891 			return (err);
4892 
4893 		crpage = (bn == UFS_HOLE);
4894 
4895 		/*
4896 		 * If its also a fallocated block that hasn't been written to
4897 		 * yet, we will treat it just like a UFS_HOLE and create
4898 		 * a zero page for it
4899 		 */
4900 		if (ISFALLOCBLK(ip, bn))
4901 			crpage = 1;
4902 	}
4903 
4904 	if (crpage) {
4905 		if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg,
4906 		    addr)) == NULL) {
4907 			return (ufs_fault(vp,
4908 			    "ufs_getpage_miss: page_create == NULL"));
4909 		}
4910 
4911 		if (rw != S_CREATE)
4912 			pagezero(pp, 0, PAGESIZE);
4913 
4914 		io_len = PAGESIZE;
4915 	} else {
4916 		u_offset_t	io_off;
4917 		uint_t	xlen;
4918 		struct buf	*bp;
4919 		ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
4920 
4921 		/*
4922 		 * If access is not in sequential order, we read from disk
4923 		 * in bsize units.
4924 		 *
4925 		 * We limit the size of the transfer to bsize if we are reading
4926 		 * from the beginning of the file. Note in this situation we
4927 		 * will hedge our bets and initiate an async read ahead of
4928 		 * the second block.
4929 		 */
4930 		if (!seq || off == 0)
4931 			contig = MIN(contig, bsize);
4932 
4933 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4934 		    &io_len, off, contig, 0);
4935 
4936 		/*
4937 		 * Some other thread has entered the page.
4938 		 * ufs_getpage will retry page_lookup.
4939 		 */
4940 		if (pp == NULL) {
4941 			pl[0] = NULL;
4942 			return (0);
4943 		}
4944 
4945 		/*
4946 		 * Zero part of the page which we are not
4947 		 * going to read from the disk.
4948 		 */
4949 		xlen = io_len & PAGEOFFSET;
4950 		if (xlen != 0)
4951 			pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4952 
4953 		bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ);
4954 		bp->b_edev = ip->i_dev;
4955 		bp->b_dev = cmpdev(ip->i_dev);
4956 		bp->b_blkno = bn;
4957 		bp->b_un.b_addr = (caddr_t)0;
4958 		bp->b_file = ip->i_vnode;
4959 		bp->b_offset = off;
4960 
4961 		if (ufsvfsp->vfs_log) {
4962 			lufs_read_strategy(ufsvfsp->vfs_log, bp);
4963 		} else if (ufsvfsp->vfs_snapshot) {
4964 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4965 		} else {
4966 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
4967 			ub.ub_getpages.value.ul++;
4968 			(void) bdev_strategy(bp);
4969 			lwp_stat_update(LWP_STAT_INBLK, 1);
4970 		}
4971 
4972 		ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK);
4973 
4974 		/*
4975 		 * If the file access is sequential, initiate read ahead
4976 		 * of the next cluster.
4977 		 */
4978 		if (seq && ip->i_nextrio < ip->i_size)
4979 			(void) ufs_getpage_ra(vp, off, seg, addr);
4980 		err = biowait(bp);
4981 		pageio_done(bp);
4982 
4983 		if (err) {
4984 			pvn_read_done(pp, B_ERROR);
4985 			return (err);
4986 		}
4987 	}
4988 
4989 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4990 	return (0);
4991 }
4992 
4993 /*
4994  * Read ahead a cluster from the disk. Returns the length in bytes.
4995  */
4996 static int
4997 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr)
4998 {
4999 	struct inode	*ip = VTOI(vp);
5000 	page_t		*pp;
5001 	u_offset_t	io_off = ip->i_nextrio;
5002 	ufsvfs_t	*ufsvfsp;
5003 	caddr_t		addr2 = addr + (io_off - off);
5004 	struct buf	*bp;
5005 	daddr_t		bn;
5006 	size_t		io_len;
5007 	int		err;
5008 	int		contig;
5009 	int		xlen;
5010 	int		bsize = ip->i_fs->fs_bsize;
5011 
5012 	/*
5013 	 * If the directio advisory is in effect on this file,
5014 	 * then do not do buffered read ahead. Read ahead makes
5015 	 * it more difficult on threads using directio as they
5016 	 * will be forced to flush the pages from this vnode.
5017 	 */
5018 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5019 		return (0);
5020 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio)
5021 		return (0);
5022 
5023 	/*
5024 	 * Is this test needed?
5025 	 */
5026 	if (addr2 >= seg->s_base + seg->s_size)
5027 		return (0);
5028 
5029 	contig = 0;
5030 	err = bmap_read(ip, io_off, &bn, &contig);
5031 	/*
5032 	 * If its a UFS_HOLE or a fallocated block, do not perform
5033 	 * any read ahead's since there probably is nothing to read ahead
5034 	 */
5035 	if (err || bn == UFS_HOLE || ISFALLOCBLK(ip, bn))
5036 		return (0);
5037 
5038 	/*
5039 	 * Limit the transfer size to bsize if this is the 2nd block.
5040 	 */
5041 	if (io_off == (u_offset_t)bsize)
5042 		contig = MIN(contig, bsize);
5043 
5044 	if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off,
5045 	    &io_len, io_off, contig, 1)) == NULL)
5046 		return (0);
5047 
5048 	/*
5049 	 * Zero part of page which we are not going to read from disk
5050 	 */
5051 	if ((xlen = (io_len & PAGEOFFSET)) > 0)
5052 		pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
5053 
5054 	ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK;
5055 
5056 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC);
5057 	bp->b_edev = ip->i_dev;
5058 	bp->b_dev = cmpdev(ip->i_dev);
5059 	bp->b_blkno = bn;
5060 	bp->b_un.b_addr = (caddr_t)0;
5061 	bp->b_file = ip->i_vnode;
5062 	bp->b_offset = off;
5063 
5064 	if (ufsvfsp->vfs_log) {
5065 		lufs_read_strategy(ufsvfsp->vfs_log, bp);
5066 	} else if (ufsvfsp->vfs_snapshot) {
5067 		fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5068 	} else {
5069 		ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5070 		ub.ub_getras.value.ul++;
5071 		(void) bdev_strategy(bp);
5072 		lwp_stat_update(LWP_STAT_INBLK, 1);
5073 	}
5074 
5075 	return (io_len);
5076 }
5077 
5078 int	ufs_delay = 1;
5079 /*
5080  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5081  *
5082  * LMXXX - the inode really ought to contain a pointer to one of these
5083  * async args.  Stuff gunk in there and just hand the whole mess off.
5084  * This would replace i_delaylen, i_delayoff.
5085  */
5086 /*ARGSUSED*/
5087 static int
5088 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
5089     struct cred *cr, caller_context_t *ct)
5090 {
5091 	struct inode *ip = VTOI(vp);
5092 	int err = 0;
5093 
5094 	if (vp->v_count == 0) {
5095 		return (ufs_fault(vp, "ufs_putpage: bad v_count == 0"));
5096 	}
5097 
5098 	/*
5099 	 * XXX - Why should this check be made here?
5100 	 */
5101 	if (vp->v_flag & VNOMAP) {
5102 		err = ENOSYS;
5103 		goto errout;
5104 	}
5105 
5106 	if (ip->i_ufsvfs == NULL) {
5107 		err = EIO;
5108 		goto errout;
5109 	}
5110 
5111 	if (flags & B_ASYNC) {
5112 		if (ufs_delay && len &&
5113 		    (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
5114 			mutex_enter(&ip->i_tlock);
5115 			/*
5116 			 * If nobody stalled, start a new cluster.
5117 			 */
5118 			if (ip->i_delaylen == 0) {
5119 				ip->i_delayoff = off;
5120 				ip->i_delaylen = len;
5121 				mutex_exit(&ip->i_tlock);
5122 				goto errout;
5123 			}
5124 			/*
5125 			 * If we have a full cluster or they are not contig,
5126 			 * then push last cluster and start over.
5127 			 */
5128 			if (ip->i_delaylen >= CLUSTSZ(ip) ||
5129 			    ip->i_delayoff + ip->i_delaylen != off) {
5130 				u_offset_t doff;
5131 				size_t dlen;
5132 
5133 				doff = ip->i_delayoff;
5134 				dlen = ip->i_delaylen;
5135 				ip->i_delayoff = off;
5136 				ip->i_delaylen = len;
5137 				mutex_exit(&ip->i_tlock);
5138 				err = ufs_putpages(vp, doff, dlen,
5139 				    flags, cr);
5140 				/* LMXXX - flags are new val, not old */
5141 				goto errout;
5142 			}
5143 			/*
5144 			 * There is something there, it's not full, and
5145 			 * it is contig.
5146 			 */
5147 			ip->i_delaylen += len;
5148 			mutex_exit(&ip->i_tlock);
5149 			goto errout;
5150 		}
5151 		/*
5152 		 * Must have weird flags or we are not clustering.
5153 		 */
5154 	}
5155 
5156 	err = ufs_putpages(vp, off, len, flags, cr);
5157 
5158 errout:
5159 	return (err);
5160 }
5161 
5162 /*
5163  * If len == 0, do from off to EOF.
5164  *
5165  * The normal cases should be len == 0 & off == 0 (entire vp list),
5166  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5167  * (from pageout).
5168  */
5169 /*ARGSUSED*/
5170 static int
5171 ufs_putpages(struct vnode *vp, offset_t off, size_t len, int flags,
5172     struct cred *cr)
5173 {
5174 	u_offset_t io_off;
5175 	u_offset_t eoff;
5176 	struct inode *ip = VTOI(vp);
5177 	page_t *pp;
5178 	size_t io_len;
5179 	int err = 0;
5180 	int dolock;
5181 
5182 	if (vp->v_count == 0)
5183 		return (ufs_fault(vp, "ufs_putpages: v_count == 0"));
5184 	/*
5185 	 * Acquire the readers/write inode lock before locking
5186 	 * any pages in this inode.
5187 	 * The inode lock is held during i/o.
5188 	 */
5189 	if (len == 0) {
5190 		mutex_enter(&ip->i_tlock);
5191 		ip->i_delayoff = ip->i_delaylen = 0;
5192 		mutex_exit(&ip->i_tlock);
5193 	}
5194 	dolock = (rw_owner(&ip->i_contents) != curthread);
5195 	if (dolock) {
5196 		/*
5197 		 * Must synchronize this thread and any possible thread
5198 		 * operating in the window of vulnerability in wrip().
5199 		 * It is dangerous to allow both a thread doing a putpage
5200 		 * and a thread writing, so serialize them.  The exception
5201 		 * is when the thread in wrip() does something which causes
5202 		 * a putpage operation.  Then, the thread must be allowed
5203 		 * to continue.  It may encounter a bmap_read problem in
5204 		 * ufs_putapage, but that is handled in ufs_putapage.
5205 		 * Allow async writers to proceed, we don't want to block
5206 		 * the pageout daemon.
5207 		 */
5208 		if (ip->i_writer == curthread)
5209 			rw_enter(&ip->i_contents, RW_READER);
5210 		else {
5211 			for (;;) {
5212 				rw_enter(&ip->i_contents, RW_READER);
5213 				mutex_enter(&ip->i_tlock);
5214 				/*
5215 				 * If there is no thread in the critical
5216 				 * section of wrip(), then proceed.
5217 				 * Otherwise, wait until there isn't one.
5218 				 */
5219 				if (ip->i_writer == NULL) {
5220 					mutex_exit(&ip->i_tlock);
5221 					break;
5222 				}
5223 				rw_exit(&ip->i_contents);
5224 				/*
5225 				 * Bounce async writers when we have a writer
5226 				 * working on this file so we don't deadlock
5227 				 * the pageout daemon.
5228 				 */
5229 				if (flags & B_ASYNC) {
5230 					mutex_exit(&ip->i_tlock);
5231 					return (0);
5232 				}
5233 				cv_wait(&ip->i_wrcv, &ip->i_tlock);
5234 				mutex_exit(&ip->i_tlock);
5235 			}
5236 		}
5237 	}
5238 
5239 	if (!vn_has_cached_data(vp)) {
5240 		if (dolock)
5241 			rw_exit(&ip->i_contents);
5242 		return (0);
5243 	}
5244 
5245 	if (len == 0) {
5246 		/*
5247 		 * Search the entire vp list for pages >= off.
5248 		 */
5249 		err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage,
5250 		    flags, cr);
5251 	} else {
5252 		/*
5253 		 * Loop over all offsets in the range looking for
5254 		 * pages to deal with.
5255 		 */
5256 		if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0)
5257 			eoff = MIN(off + len, eoff);
5258 		else
5259 			eoff = off + len;
5260 
5261 		for (io_off = off; io_off < eoff; io_off += io_len) {
5262 			/*
5263 			 * If we are not invalidating, synchronously
5264 			 * freeing or writing pages, use the routine
5265 			 * page_lookup_nowait() to prevent reclaiming
5266 			 * them from the free list.
5267 			 */
5268 			if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
5269 				pp = page_lookup(vp, io_off,
5270 				    (flags & (B_INVAL | B_FREE)) ?
5271 				    SE_EXCL : SE_SHARED);
5272 			} else {
5273 				pp = page_lookup_nowait(vp, io_off,
5274 				    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
5275 			}
5276 
5277 			if (pp == NULL || pvn_getdirty(pp, flags) == 0)
5278 				io_len = PAGESIZE;
5279 			else {
5280 				u_offset_t *io_offp = &io_off;
5281 
5282 				err = ufs_putapage(vp, pp, io_offp, &io_len,
5283 				    flags, cr);
5284 				if (err != 0)
5285 					break;
5286 				/*
5287 				 * "io_off" and "io_len" are returned as
5288 				 * the range of pages we actually wrote.
5289 				 * This allows us to skip ahead more quickly
5290 				 * since several pages may've been dealt
5291 				 * with by this iteration of the loop.
5292 				 */
5293 			}
5294 		}
5295 	}
5296 	if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
5297 		/*
5298 		 * We have just sync'ed back all the pages on
5299 		 * the inode, turn off the IMODTIME flag.
5300 		 */
5301 		mutex_enter(&ip->i_tlock);
5302 		ip->i_flag &= ~IMODTIME;
5303 		mutex_exit(&ip->i_tlock);
5304 	}
5305 	if (dolock)
5306 		rw_exit(&ip->i_contents);
5307 	return (err);
5308 }
5309 
5310 static void
5311 ufs_iodone(buf_t *bp)
5312 {
5313 	struct inode *ip;
5314 
5315 	ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
5316 
5317 	bp->b_iodone = NULL;
5318 
5319 	ip = VTOI(bp->b_pages->p_vnode);
5320 
5321 	mutex_enter(&ip->i_tlock);
5322 	if (ip->i_writes >= ufs_LW) {
5323 		if ((ip->i_writes -= bp->b_bcount) <= ufs_LW)
5324 			if (ufs_WRITES)
5325 				cv_broadcast(&ip->i_wrcv); /* wake all up */
5326 	} else {
5327 		ip->i_writes -= bp->b_bcount;
5328 	}
5329 
5330 	mutex_exit(&ip->i_tlock);
5331 	iodone(bp);
5332 }
5333 
5334 /*
5335  * Write out a single page, possibly klustering adjacent
5336  * dirty pages.  The inode lock must be held.
5337  *
5338  * LMXXX - bsize < pagesize not done.
5339  */
5340 /*ARGSUSED*/
5341 int
5342 ufs_putapage(struct vnode *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
5343     int flags, struct cred *cr)
5344 {
5345 	u_offset_t io_off;
5346 	u_offset_t off;
5347 	struct inode *ip = VTOI(vp);
5348 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
5349 	struct fs *fs;
5350 	struct buf *bp;
5351 	size_t io_len;
5352 	daddr_t bn;
5353 	int err;
5354 	int contig;
5355 	int dotrans;
5356 
5357 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
5358 
5359 	if (ufsvfsp == NULL) {
5360 		err = EIO;
5361 		goto out_trace;
5362 	}
5363 
5364 	fs = ip->i_fs;
5365 	ASSERT(fs->fs_ronly == 0);
5366 
5367 	/*
5368 	 * If the modified time on the inode has not already been
5369 	 * set elsewhere (e.g. for write/setattr) we set the time now.
5370 	 * This gives us approximate modified times for mmap'ed files
5371 	 * which are modified via stores in the user address space.
5372 	 */
5373 	if ((ip->i_flag & IMODTIME) == 0) {
5374 		mutex_enter(&ip->i_tlock);
5375 		ip->i_flag |= IUPD;
5376 		ip->i_seq++;
5377 		ITIMES_NOLOCK(ip);
5378 		mutex_exit(&ip->i_tlock);
5379 	}
5380 
5381 	/*
5382 	 * Align the request to a block boundry (for old file systems),
5383 	 * and go ask bmap() how contiguous things are for this file.
5384 	 */
5385 	off = pp->p_offset & (offset_t)fs->fs_bmask;	/* block align it */
5386 	contig = 0;
5387 	err = bmap_read(ip, off, &bn, &contig);
5388 	if (err)
5389 		goto out;
5390 	if (bn == UFS_HOLE) {			/* putpage never allocates */
5391 		/*
5392 		 * logging device is in error mode; simply return EIO
5393 		 */
5394 		if (TRANS_ISERROR(ufsvfsp)) {
5395 			err = EIO;
5396 			goto out;
5397 		}
5398 		/*
5399 		 * Oops, the thread in the window in wrip() did some
5400 		 * sort of operation which caused a putpage in the bad
5401 		 * range.  In this case, just return an error which will
5402 		 * cause the software modified bit on the page to set
5403 		 * and the page will get written out again later.
5404 		 */
5405 		if (ip->i_writer == curthread) {
5406 			err = EIO;
5407 			goto out;
5408 		}
5409 		/*
5410 		 * If the pager is trying to push a page in the bad range
5411 		 * just tell it to try again later when things are better.
5412 		 */
5413 		if (flags & B_ASYNC) {
5414 			err = EAGAIN;
5415 			goto out;
5416 		}
5417 		err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE");
5418 		goto out;
5419 	}
5420 
5421 	/*
5422 	 * If it is an fallocate'd block, reverse the negativity since
5423 	 * we are now writing to it
5424 	 */
5425 	if (ISFALLOCBLK(ip, bn)) {
5426 		err = bmap_set_bn(vp, off, dbtofsb(fs, -bn));
5427 		if (err)
5428 			goto out;
5429 
5430 		bn = -bn;
5431 	}
5432 
5433 	/*
5434 	 * Take the length (of contiguous bytes) passed back from bmap()
5435 	 * and _try_ and get a set of pages covering that extent.
5436 	 */
5437 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags);
5438 
5439 	/*
5440 	 * May have run out of memory and not clustered backwards.
5441 	 * off		p_offset
5442 	 * [  pp - 1  ][   pp   ]
5443 	 * [	block		]
5444 	 * We told bmap off, so we have to adjust the bn accordingly.
5445 	 */
5446 	if (io_off > off) {
5447 		bn += btod(io_off - off);
5448 		contig -= (io_off - off);
5449 	}
5450 
5451 	/*
5452 	 * bmap was carefull to tell us the right size so use that.
5453 	 * There might be unallocated frags at the end.
5454 	 * LMXXX - bzero the end of the page?  We must be writing after EOF.
5455 	 */
5456 	if (io_len > contig) {
5457 		ASSERT(io_len - contig < fs->fs_bsize);
5458 		io_len -= (io_len - contig);
5459 	}
5460 
5461 	/*
5462 	 * Handle the case where we are writing the last page after EOF.
5463 	 *
5464 	 * XXX - just a patch for i-mt3.
5465 	 */
5466 	if (io_len == 0) {
5467 		ASSERT(pp->p_offset >=
5468 		    (u_offset_t)(roundup(ip->i_size, PAGESIZE)));
5469 		io_len = PAGESIZE;
5470 	}
5471 
5472 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags);
5473 
5474 	ULOCKFS_SET_MOD(ITOUL(ip));
5475 
5476 	bp->b_edev = ip->i_dev;
5477 	bp->b_dev = cmpdev(ip->i_dev);
5478 	bp->b_blkno = bn;
5479 	bp->b_un.b_addr = (caddr_t)0;
5480 	bp->b_file = ip->i_vnode;
5481 
5482 	/*
5483 	 * File contents of shadow or quota inodes are metadata, and updates
5484 	 * to these need to be put into a logging transaction. All direct
5485 	 * callers in UFS do that, but fsflush can come here _before_ the
5486 	 * normal codepath. An example would be updating ACL information, for
5487 	 * which the normal codepath would be:
5488 	 *	ufs_si_store()
5489 	 *	ufs_rdwri()
5490 	 *	wrip()
5491 	 *	segmap_release()
5492 	 *	VOP_PUTPAGE()
5493 	 * Here, fsflush can pick up the dirty page before segmap_release()
5494 	 * forces it out. If that happens, there's no transaction.
5495 	 * We therefore need to test whether a transaction exists, and if not
5496 	 * create one - for fsflush.
5497 	 */
5498 	dotrans =
5499 	    (((ip->i_mode & IFMT) == IFSHAD || ufsvfsp->vfs_qinod == ip) &&
5500 	    ((curthread->t_flag & T_DONTBLOCK) == 0) &&
5501 	    (TRANS_ISTRANS(ufsvfsp)));
5502 
5503 	if (dotrans) {
5504 		curthread->t_flag |= T_DONTBLOCK;
5505 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5506 	}
5507 	if (TRANS_ISTRANS(ufsvfsp)) {
5508 		if ((ip->i_mode & IFMT) == IFSHAD) {
5509 			TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD);
5510 		} else if (ufsvfsp->vfs_qinod == ip) {
5511 			TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR,
5512 			    0, 0);
5513 		}
5514 	}
5515 	if (dotrans) {
5516 		TRANS_END_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5517 		curthread->t_flag &= ~T_DONTBLOCK;
5518 	}
5519 
5520 	/* write throttle */
5521 
5522 	ASSERT(bp->b_iodone == NULL);
5523 	bp->b_iodone = (int (*)())ufs_iodone;
5524 	mutex_enter(&ip->i_tlock);
5525 	ip->i_writes += bp->b_bcount;
5526 	mutex_exit(&ip->i_tlock);
5527 
5528 	if (bp->b_flags & B_ASYNC) {
5529 		if (ufsvfsp->vfs_log) {
5530 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5531 		} else if (ufsvfsp->vfs_snapshot) {
5532 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5533 		} else {
5534 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5535 			ub.ub_putasyncs.value.ul++;
5536 			(void) bdev_strategy(bp);
5537 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5538 		}
5539 	} else {
5540 		if (ufsvfsp->vfs_log) {
5541 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5542 		} else if (ufsvfsp->vfs_snapshot) {
5543 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5544 		} else {
5545 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5546 			ub.ub_putsyncs.value.ul++;
5547 			(void) bdev_strategy(bp);
5548 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5549 		}
5550 		err = biowait(bp);
5551 		pageio_done(bp);
5552 		pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
5553 	}
5554 
5555 	pp = NULL;
5556 
5557 out:
5558 	if (err != 0 && pp != NULL)
5559 		pvn_write_done(pp, B_ERROR | B_WRITE | flags);
5560 
5561 	if (offp)
5562 		*offp = io_off;
5563 	if (lenp)
5564 		*lenp = io_len;
5565 out_trace:
5566 	return (err);
5567 }
5568 
5569 uint64_t ufs_map_alock_retry_cnt;
5570 uint64_t ufs_map_lockfs_retry_cnt;
5571 
5572 /* ARGSUSED */
5573 static int
5574 ufs_map(struct vnode *vp, offset_t off, struct as *as, caddr_t *addrp,
5575     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, struct cred *cr,
5576     caller_context_t *ct)
5577 {
5578 	struct segvn_crargs vn_a;
5579 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5580 	struct ulockfs *ulp;
5581 	int error, sig;
5582 	k_sigset_t smask;
5583 	caddr_t hint = *addrp;
5584 
5585 	if (vp->v_flag & VNOMAP) {
5586 		error = ENOSYS;
5587 		goto out;
5588 	}
5589 
5590 	if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) {
5591 		error = ENXIO;
5592 		goto out;
5593 	}
5594 
5595 	if (vp->v_type != VREG) {
5596 		error = ENODEV;
5597 		goto out;
5598 	}
5599 
5600 retry_map:
5601 	*addrp = hint;
5602 	/*
5603 	 * If file is being locked, disallow mapping.
5604 	 */
5605 	if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) {
5606 		error = EAGAIN;
5607 		goto out;
5608 	}
5609 
5610 	as_rangelock(as);
5611 	/*
5612 	 * Note that if we are retrying (because ufs_lockfs_trybegin failed in
5613 	 * the previous attempt), some other thread could have grabbed
5614 	 * the same VA range if MAP_FIXED is set. In that case, choose_addr
5615 	 * would unmap the valid VA range, that is ok.
5616 	 */
5617 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5618 	if (error != 0) {
5619 		as_rangeunlock(as);
5620 		goto out;
5621 	}
5622 
5623 	/*
5624 	 * a_lock has to be acquired before entering the lockfs protocol
5625 	 * because that is the order in which pagefault works. Also we cannot
5626 	 * block on a_lock here because this waiting writer will prevent
5627 	 * further readers like ufs_read from progressing and could cause
5628 	 * deadlock between ufs_read/ufs_map/pagefault when a quiesce is
5629 	 * pending.
5630 	 */
5631 	while (!AS_LOCK_TRYENTER(as, RW_WRITER)) {
5632 		ufs_map_alock_retry_cnt++;
5633 		delay(RETRY_LOCK_DELAY);
5634 	}
5635 
5636 	/*
5637 	 * We can't hold as->a_lock and wait for lockfs to succeed because
5638 	 * the proc tools might hang on a_lock, so call ufs_lockfs_trybegin()
5639 	 * instead.
5640 	 */
5641 	if (error = ufs_lockfs_trybegin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK)) {
5642 		/*
5643 		 * ufs_lockfs_trybegin() did not succeed. It is safer to give up
5644 		 * as->a_lock and wait for ulp->ul_fs_lock status to change.
5645 		 */
5646 		ufs_map_lockfs_retry_cnt++;
5647 		AS_LOCK_EXIT(as);
5648 		as_rangeunlock(as);
5649 		if (error == EIO)
5650 			goto out;
5651 
5652 		mutex_enter(&ulp->ul_lock);
5653 		while (ulp->ul_fs_lock & ULOCKFS_MAP_MASK) {
5654 			if (ULOCKFS_IS_SLOCK(ulp) || ufsvfsp->vfs_nointr) {
5655 				cv_wait(&ulp->ul_cv, &ulp->ul_lock);
5656 			} else {
5657 				sigintr(&smask, 1);
5658 				sig = cv_wait_sig(&ulp->ul_cv, &ulp->ul_lock);
5659 				sigunintr(&smask);
5660 				if (((ulp->ul_fs_lock & ULOCKFS_MAP_MASK) &&
5661 				    !sig) || ufsvfsp->vfs_dontblock) {
5662 					mutex_exit(&ulp->ul_lock);
5663 					return (EINTR);
5664 				}
5665 			}
5666 		}
5667 		mutex_exit(&ulp->ul_lock);
5668 		goto retry_map;
5669 	}
5670 
5671 	vn_a.vp = vp;
5672 	vn_a.offset = (u_offset_t)off;
5673 	vn_a.type = flags & MAP_TYPE;
5674 	vn_a.prot = prot;
5675 	vn_a.maxprot = maxprot;
5676 	vn_a.cred = cr;
5677 	vn_a.amp = NULL;
5678 	vn_a.flags = flags & ~MAP_TYPE;
5679 	vn_a.szc = 0;
5680 	vn_a.lgrp_mem_policy_flags = 0;
5681 
5682 	error = as_map_locked(as, *addrp, len, segvn_create, &vn_a);
5683 	if (ulp)
5684 		ufs_lockfs_end(ulp);
5685 	as_rangeunlock(as);
5686 out:
5687 	return (error);
5688 }
5689 
5690 /* ARGSUSED */
5691 static int
5692 ufs_addmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5693     size_t len, uchar_t  prot, uchar_t  maxprot, uint_t    flags,
5694     struct cred *cr, caller_context_t *ct)
5695 {
5696 	struct inode *ip = VTOI(vp);
5697 
5698 	if (vp->v_flag & VNOMAP) {
5699 		return (ENOSYS);
5700 	}
5701 
5702 	mutex_enter(&ip->i_tlock);
5703 	ip->i_mapcnt += btopr(len);
5704 	mutex_exit(&ip->i_tlock);
5705 	return (0);
5706 }
5707 
5708 /*ARGSUSED*/
5709 static int
5710 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5711     size_t len, uint_t prot,  uint_t maxprot,  uint_t flags, struct cred *cr,
5712     caller_context_t *ct)
5713 {
5714 	struct inode *ip = VTOI(vp);
5715 
5716 	if (vp->v_flag & VNOMAP) {
5717 		return (ENOSYS);
5718 	}
5719 
5720 	mutex_enter(&ip->i_tlock);
5721 	ip->i_mapcnt -= btopr(len); 	/* Count released mappings */
5722 	ASSERT(ip->i_mapcnt >= 0);
5723 	mutex_exit(&ip->i_tlock);
5724 	return (0);
5725 }
5726 /*
5727  * Return the answer requested to poll() for non-device files
5728  */
5729 struct pollhead ufs_pollhd;
5730 
5731 /* ARGSUSED */
5732 int
5733 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp,
5734     caller_context_t *ct)
5735 {
5736 	struct ufsvfs	*ufsvfsp;
5737 
5738 	/*
5739 	 * Regular files reject edge-triggered pollers.
5740 	 * See the comment in fs_poll() for a more detailed explanation.
5741 	 */
5742 	if (ev & POLLET) {
5743 		return (EPERM);
5744 	}
5745 
5746 	*revp = 0;
5747 	ufsvfsp = VTOI(vp)->i_ufsvfs;
5748 
5749 	if (!ufsvfsp) {
5750 		*revp = POLLHUP;
5751 		goto out;
5752 	}
5753 
5754 	if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) ||
5755 	    ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) {
5756 		*revp |= POLLERR;
5757 
5758 	} else {
5759 		if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly &&
5760 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5761 			*revp |= POLLOUT;
5762 
5763 		if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly &&
5764 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5765 			*revp |= POLLWRBAND;
5766 
5767 		if (ev & POLLIN)
5768 			*revp |= POLLIN;
5769 
5770 		if (ev & POLLRDNORM)
5771 			*revp |= POLLRDNORM;
5772 
5773 		if (ev & POLLRDBAND)
5774 			*revp |= POLLRDBAND;
5775 	}
5776 
5777 	if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP)))
5778 		*revp |= POLLPRI;
5779 out:
5780 	if (*revp == 0 && ! any) {
5781 		*phpp = &ufs_pollhd;
5782 	}
5783 
5784 	return (0);
5785 }
5786 
5787 /* ARGSUSED */
5788 static int
5789 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr,
5790     caller_context_t *ct)
5791 {
5792 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
5793 	struct ulockfs	*ulp = NULL;
5794 	struct inode 	*sip = NULL;
5795 	int		error;
5796 	struct inode 	*ip = VTOI(vp);
5797 	int		issync;
5798 
5799 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK);
5800 	if (error)
5801 		return (error);
5802 
5803 	switch (cmd) {
5804 		/*
5805 		 * Have to handle _PC_NAME_MAX here, because the normal way
5806 		 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5807 		 * results in a lock ordering reversal between
5808 		 * ufs_lockfs_{begin,end}() and
5809 		 * ufs_thread_{suspend,continue}().
5810 		 *
5811 		 * Keep in sync with ufs_statvfs().
5812 		 */
5813 	case _PC_NAME_MAX:
5814 		*valp = MAXNAMLEN;
5815 		break;
5816 
5817 	case _PC_FILESIZEBITS:
5818 		if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
5819 			*valp = UFS_FILESIZE_BITS;
5820 		else
5821 			*valp = 32;
5822 		break;
5823 
5824 	case _PC_XATTR_EXISTS:
5825 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
5826 
5827 			error =
5828 			    ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR, cr);
5829 			if (error ==  0 && sip != NULL) {
5830 				/* Start transaction */
5831 				if (ulp) {
5832 					TRANS_BEGIN_CSYNC(ufsvfsp, issync,
5833 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5834 				}
5835 				/*
5836 				 * Is directory empty
5837 				 */
5838 				rw_enter(&sip->i_rwlock, RW_WRITER);
5839 				rw_enter(&sip->i_contents, RW_WRITER);
5840 				if (ufs_xattrdirempty(sip,
5841 				    sip->i_number, CRED())) {
5842 					rw_enter(&ip->i_contents, RW_WRITER);
5843 					ufs_unhook_shadow(ip, sip);
5844 					rw_exit(&ip->i_contents);
5845 
5846 					*valp = 0;
5847 
5848 				} else
5849 					*valp = 1;
5850 				rw_exit(&sip->i_contents);
5851 				rw_exit(&sip->i_rwlock);
5852 				if (ulp) {
5853 					TRANS_END_CSYNC(ufsvfsp, error, issync,
5854 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5855 				}
5856 				VN_RELE(ITOV(sip));
5857 			} else if (error == ENOENT) {
5858 				*valp = 0;
5859 				error = 0;
5860 			}
5861 		} else {
5862 			error = fs_pathconf(vp, cmd, valp, cr, ct);
5863 		}
5864 		break;
5865 
5866 	case _PC_ACL_ENABLED:
5867 		*valp = _ACL_ACLENT_ENABLED;
5868 		break;
5869 
5870 	case _PC_MIN_HOLE_SIZE:
5871 		*valp = (ulong_t)ip->i_fs->fs_bsize;
5872 		break;
5873 
5874 	case _PC_SATTR_ENABLED:
5875 	case _PC_SATTR_EXISTS:
5876 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5877 		    (vp->v_type == VREG || vp->v_type == VDIR);
5878 		break;
5879 
5880 	case _PC_TIMESTAMP_RESOLUTION:
5881 		/*
5882 		 * UFS keeps only microsecond timestamp resolution.
5883 		 * This is historical and will probably never change.
5884 		 */
5885 		*valp = 1000L;
5886 		break;
5887 
5888 	default:
5889 		error = fs_pathconf(vp, cmd, valp, cr, ct);
5890 		break;
5891 	}
5892 
5893 	if (ulp != NULL) {
5894 		ufs_lockfs_end(ulp);
5895 	}
5896 	return (error);
5897 }
5898 
5899 int ufs_pageio_writes, ufs_pageio_reads;
5900 
5901 /*ARGSUSED*/
5902 static int
5903 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len,
5904     int flags, struct cred *cr, caller_context_t *ct)
5905 {
5906 	struct inode *ip = VTOI(vp);
5907 	struct ufsvfs *ufsvfsp;
5908 	page_t *npp = NULL, *opp = NULL, *cpp = pp;
5909 	struct buf *bp;
5910 	daddr_t bn;
5911 	size_t done_len = 0, cur_len = 0;
5912 	int err = 0;
5913 	int contig = 0;
5914 	int dolock;
5915 	int vmpss = 0;
5916 	struct ulockfs *ulp;
5917 
5918 	if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp &&
5919 	    vp->v_mpssdata != NULL) {
5920 		vmpss = 1;
5921 	}
5922 
5923 	dolock = (rw_owner(&ip->i_contents) != curthread);
5924 	/*
5925 	 * We need a better check.  Ideally, we would use another
5926 	 * vnodeops so that hlocked and forcibly unmounted file
5927 	 * systems would return EIO where appropriate and w/o the
5928 	 * need for these checks.
5929 	 */
5930 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5931 		return (EIO);
5932 
5933 	/*
5934 	 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5935 	 * ul_lock must be taken before locking pages so we can't use it here
5936 	 * if pp is non NULL because segvn already locked pages
5937 	 * SE_EXCL. Instead we rely on the fact that a forced umount or
5938 	 * applying a filesystem lock via ufs_fiolfs() will block in the
5939 	 * implicit call to ufs_flush() until we unlock the pages after the
5940 	 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5941 	 * above 0 until they are done. We have to be careful not to increment
5942 	 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5943 	 *
5944 	 * If pp is NULL use ul_lock to make sure we don't increment
5945 	 * ul_vnops_cnt after forceful unmount hlocks the file system.
5946 	 */
5947 	if (vmpss || pp == NULL) {
5948 		ulp = &ufsvfsp->vfs_ulockfs;
5949 		if (pp == NULL)
5950 			mutex_enter(&ulp->ul_lock);
5951 		if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) {
5952 			if (pp == NULL) {
5953 				mutex_exit(&ulp->ul_lock);
5954 			}
5955 			return (vmpss ? EIO : EINVAL);
5956 		}
5957 		atomic_inc_ulong(&ulp->ul_vnops_cnt);
5958 		if (pp == NULL)
5959 			mutex_exit(&ulp->ul_lock);
5960 		if (ufs_quiesce_pend) {
5961 			if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
5962 				cv_broadcast(&ulp->ul_cv);
5963 			return (vmpss ? EIO : EINVAL);
5964 		}
5965 	}
5966 
5967 	if (dolock) {
5968 		/*
5969 		 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
5970 		 * handle a fault against a segment that maps vnode pages with
5971 		 * large mappings.  Segvn creates pages and holds them locked
5972 		 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
5973 		 * use rw_tryenter() to avoid a potential deadlock since in
5974 		 * lock order i_contents needs to be taken first.
5975 		 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
5976 		 */
5977 		if (!vmpss) {
5978 			rw_enter(&ip->i_contents, RW_READER);
5979 		} else if (!rw_tryenter(&ip->i_contents, RW_READER)) {
5980 			if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
5981 				cv_broadcast(&ulp->ul_cv);
5982 			return (EDEADLK);
5983 		}
5984 	}
5985 
5986 	/*
5987 	 * Return an error to segvn because the pagefault request is beyond
5988 	 * PAGESIZE rounded EOF.
5989 	 */
5990 	if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) {
5991 		if (dolock)
5992 			rw_exit(&ip->i_contents);
5993 		if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
5994 			cv_broadcast(&ulp->ul_cv);
5995 		return (EFAULT);
5996 	}
5997 
5998 	if (pp == NULL) {
5999 		if (bmap_has_holes(ip)) {
6000 			err = ENOSYS;
6001 		} else {
6002 			err = EINVAL;
6003 		}
6004 		if (dolock)
6005 			rw_exit(&ip->i_contents);
6006 		if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6007 			cv_broadcast(&ulp->ul_cv);
6008 		return (err);
6009 	}
6010 
6011 	/*
6012 	 * Break the io request into chunks, one for each contiguous
6013 	 * stretch of disk blocks in the target file.
6014 	 */
6015 	while (done_len < io_len) {
6016 		ASSERT(cpp);
6017 		contig = 0;
6018 		if (err = bmap_read(ip, (u_offset_t)(io_off + done_len),
6019 		    &bn, &contig))
6020 			break;
6021 
6022 		if (bn == UFS_HOLE) {	/* No holey swapfiles */
6023 			if (vmpss) {
6024 				err = EFAULT;
6025 				break;
6026 			}
6027 			err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE");
6028 			break;
6029 		}
6030 
6031 		cur_len = MIN(io_len - done_len, contig);
6032 		/*
6033 		 * Zero out a page beyond EOF, when the last block of
6034 		 * a file is a UFS fragment so that ufs_pageio() can be used
6035 		 * instead of ufs_getpage() to handle faults against
6036 		 * segvn segments that use large pages.
6037 		 */
6038 		page_list_break(&cpp, &npp, btopr(cur_len));
6039 		if ((flags & B_READ) && (cur_len & PAGEOFFSET)) {
6040 			size_t xlen = cur_len & PAGEOFFSET;
6041 			pagezero(cpp->p_prev, xlen, PAGESIZE - xlen);
6042 		}
6043 
6044 		bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
6045 		ASSERT(bp != NULL);
6046 
6047 		bp->b_edev = ip->i_dev;
6048 		bp->b_dev = cmpdev(ip->i_dev);
6049 		bp->b_blkno = bn;
6050 		bp->b_un.b_addr = (caddr_t)0;
6051 		bp->b_file = ip->i_vnode;
6052 
6053 		ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
6054 		ub.ub_pageios.value.ul++;
6055 		if (ufsvfsp->vfs_snapshot)
6056 			fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp);
6057 		else
6058 			(void) bdev_strategy(bp);
6059 
6060 		if (flags & B_READ)
6061 			ufs_pageio_reads++;
6062 		else
6063 			ufs_pageio_writes++;
6064 		if (flags & B_READ)
6065 			lwp_stat_update(LWP_STAT_INBLK, 1);
6066 		else
6067 			lwp_stat_update(LWP_STAT_OUBLK, 1);
6068 		/*
6069 		 * If the request is not B_ASYNC, wait for i/o to complete
6070 		 * and re-assemble the page list to return to the caller.
6071 		 * If it is B_ASYNC we leave the page list in pieces and
6072 		 * cleanup() will dispose of them.
6073 		 */
6074 		if ((flags & B_ASYNC) == 0) {
6075 			err = biowait(bp);
6076 			pageio_done(bp);
6077 			if (err)
6078 				break;
6079 			page_list_concat(&opp, &cpp);
6080 		}
6081 		cpp = npp;
6082 		npp = NULL;
6083 		if (flags & B_READ)
6084 			cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t);
6085 		done_len += cur_len;
6086 	}
6087 	ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len));
6088 	if (err) {
6089 		if (flags & B_ASYNC) {
6090 			/* Cleanup unprocessed parts of list */
6091 			page_list_concat(&cpp, &npp);
6092 			if (flags & B_READ)
6093 				pvn_read_done(cpp, B_ERROR);
6094 			else
6095 				pvn_write_done(cpp, B_ERROR);
6096 		} else {
6097 			/* Re-assemble list and let caller clean up */
6098 			page_list_concat(&opp, &cpp);
6099 			page_list_concat(&opp, &npp);
6100 		}
6101 	}
6102 
6103 	if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) &&
6104 	    ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) {
6105 		mutex_enter(&ip->i_tlock);
6106 		ip->i_flag |= IACC;
6107 		ITIMES_NOLOCK(ip);
6108 		mutex_exit(&ip->i_tlock);
6109 	}
6110 
6111 	if (dolock)
6112 		rw_exit(&ip->i_contents);
6113 	if (vmpss && !atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6114 		cv_broadcast(&ulp->ul_cv);
6115 	return (err);
6116 }
6117 
6118 /*
6119  * Called when the kernel is in a frozen state to dump data
6120  * directly to the device. It uses a private dump data structure,
6121  * set up by dump_ctl, to locate the correct disk block to which to dump.
6122  */
6123 /*ARGSUSED*/
6124 static int
6125 ufs_dump(vnode_t *vp, caddr_t addr, offset_t ldbn, offset_t dblks,
6126     caller_context_t *ct)
6127 {
6128 	u_offset_t	file_size;
6129 	struct inode    *ip = VTOI(vp);
6130 	struct fs	*fs = ip->i_fs;
6131 	daddr_t		dbn, lfsbn;
6132 	int		disk_blks = fs->fs_bsize >> DEV_BSHIFT;
6133 	int		error = 0;
6134 	int		ndbs, nfsbs;
6135 
6136 	/*
6137 	 * forced unmount case
6138 	 */
6139 	if (ip->i_ufsvfs == NULL)
6140 		return (EIO);
6141 	/*
6142 	 * Validate the inode that it has not been modified since
6143 	 * the dump structure is allocated.
6144 	 */
6145 	mutex_enter(&ip->i_tlock);
6146 	if ((dump_info == NULL) ||
6147 	    (dump_info->ip != ip) ||
6148 	    (dump_info->time.tv_sec != ip->i_mtime.tv_sec) ||
6149 	    (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) {
6150 		mutex_exit(&ip->i_tlock);
6151 		return (-1);
6152 	}
6153 	mutex_exit(&ip->i_tlock);
6154 
6155 	/*
6156 	 * See that the file has room for this write
6157 	 */
6158 	UFS_GET_ISIZE(&file_size, ip);
6159 
6160 	if (ldbtob(ldbn + dblks) > file_size)
6161 		return (ENOSPC);
6162 
6163 	/*
6164 	 * Find the physical disk block numbers from the dump
6165 	 * private data structure directly and write out the data
6166 	 * in contiguous block lumps
6167 	 */
6168 	while (dblks > 0 && !error) {
6169 		lfsbn = (daddr_t)lblkno(fs, ldbtob(ldbn));
6170 		dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks;
6171 		nfsbs = 1;
6172 		ndbs = disk_blks - ldbn % disk_blks;
6173 		while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn +
6174 		    nfsbs]) == dbn + ndbs) {
6175 			nfsbs++;
6176 			ndbs += disk_blks;
6177 		}
6178 		if (ndbs > dblks)
6179 			ndbs = dblks;
6180 		error = bdev_dump(ip->i_dev, addr, dbn, ndbs);
6181 		addr += ldbtob((offset_t)ndbs);
6182 		dblks -= ndbs;
6183 		ldbn += ndbs;
6184 	}
6185 	return (error);
6186 
6187 }
6188 
6189 /*
6190  * Prepare the file system before and after the dump operation.
6191  *
6192  * action = DUMP_ALLOC:
6193  * Preparation before dump, allocate dump private data structure
6194  * to hold all the direct and indirect block info for dump.
6195  *
6196  * action = DUMP_FREE:
6197  * Clean up after dump, deallocate the dump private data structure.
6198  *
6199  * action = DUMP_SCAN:
6200  * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6201  * if found, the starting file-relative DEV_BSIZE lbn is written
6202  * to *bklp; that lbn is intended for use with VOP_DUMP()
6203  */
6204 /*ARGSUSED*/
6205 static int
6206 ufs_dumpctl(vnode_t *vp, int action, offset_t *blkp, caller_context_t *ct)
6207 {
6208 	struct inode	*ip = VTOI(vp);
6209 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
6210 	struct fs	*fs;
6211 	daddr32_t	*dblk, *storeblk;
6212 	daddr32_t	*nextblk, *endblk;
6213 	struct buf	*bp;
6214 	int		i, entry, entries;
6215 	int		n, ncontig;
6216 
6217 	/*
6218 	 * check for forced unmount
6219 	 */
6220 	if (ufsvfsp == NULL)
6221 		return (EIO);
6222 
6223 	if (action == DUMP_ALLOC) {
6224 		/*
6225 		 * alloc and record dump_info
6226 		 */
6227 		if (dump_info != NULL)
6228 			return (EINVAL);
6229 
6230 		ASSERT(vp->v_type == VREG);
6231 		fs = ufsvfsp->vfs_fs;
6232 
6233 		rw_enter(&ip->i_contents, RW_READER);
6234 
6235 		if (bmap_has_holes(ip)) {
6236 			rw_exit(&ip->i_contents);
6237 			return (EFAULT);
6238 		}
6239 
6240 		/*
6241 		 * calculate and allocate space needed according to i_size
6242 		 */
6243 		entries = (int)lblkno(fs, blkroundup(fs, ip->i_size));
6244 		dump_info = kmem_alloc(sizeof (struct dump) +
6245 		    (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP);
6246 		if (dump_info == NULL) {
6247 			rw_exit(&ip->i_contents);
6248 			return (ENOMEM);
6249 		}
6250 
6251 		/* Start saving the info */
6252 		dump_info->fsbs = entries;
6253 		dump_info->ip = ip;
6254 		storeblk = &dump_info->dblk[0];
6255 
6256 		/* Direct Blocks */
6257 		for (entry = 0; entry < NDADDR && entry < entries; entry++)
6258 			*storeblk++ = ip->i_db[entry];
6259 
6260 		/* Indirect Blocks */
6261 		for (i = 0; i < NIADDR; i++) {
6262 			int error = 0;
6263 
6264 			bp = UFS_BREAD(ufsvfsp,
6265 			    ip->i_dev, fsbtodb(fs, ip->i_ib[i]), fs->fs_bsize);
6266 			if (bp->b_flags & B_ERROR)
6267 				error = EIO;
6268 			else {
6269 				dblk = bp->b_un.b_daddr;
6270 				if ((storeblk = save_dblks(ip, ufsvfsp,
6271 				    storeblk, dblk, i, entries)) == NULL)
6272 					error = EIO;
6273 			}
6274 
6275 			brelse(bp);
6276 
6277 			if (error != 0) {
6278 				kmem_free(dump_info, sizeof (struct dump) +
6279 				    (entries - 1) * sizeof (daddr32_t));
6280 				rw_exit(&ip->i_contents);
6281 				dump_info = NULL;
6282 				return (error);
6283 			}
6284 		}
6285 		/* and time stamp the information */
6286 		mutex_enter(&ip->i_tlock);
6287 		dump_info->time = ip->i_mtime;
6288 		mutex_exit(&ip->i_tlock);
6289 
6290 		rw_exit(&ip->i_contents);
6291 	} else if (action == DUMP_FREE) {
6292 		/*
6293 		 * free dump_info
6294 		 */
6295 		if (dump_info == NULL)
6296 			return (EINVAL);
6297 		entries = dump_info->fsbs - 1;
6298 		kmem_free(dump_info, sizeof (struct dump) +
6299 		    entries * sizeof (daddr32_t));
6300 		dump_info = NULL;
6301 	} else if (action == DUMP_SCAN) {
6302 		/*
6303 		 * scan dump_info
6304 		 */
6305 		if (dump_info == NULL)
6306 			return (EINVAL);
6307 
6308 		dblk = dump_info->dblk;
6309 		nextblk = dblk + 1;
6310 		endblk = dblk + dump_info->fsbs - 1;
6311 		fs = ufsvfsp->vfs_fs;
6312 		ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT);
6313 
6314 		/*
6315 		 * scan dblk[] entries; contig fs space is found when:
6316 		 * ((current blkno + frags per block) == next blkno)
6317 		 */
6318 		n = 0;
6319 		while (n < ncontig && dblk < endblk) {
6320 			if ((*dblk + fs->fs_frag) == *nextblk)
6321 				n++;
6322 			else
6323 				n = 0;
6324 			dblk++;
6325 			nextblk++;
6326 		}
6327 
6328 		/*
6329 		 * index is where size bytes of contig space begins;
6330 		 * conversion from index to the file's DEV_BSIZE lbn
6331 		 * is equivalent to:  (index * fs_bsize) / DEV_BSIZE
6332 		 */
6333 		if (n == ncontig) {
6334 			i = (dblk - dump_info->dblk) - ncontig;
6335 			*blkp = i << (fs->fs_bshift - DEV_BSHIFT);
6336 		} else
6337 			return (EFAULT);
6338 	}
6339 	return (0);
6340 }
6341 
6342 /*
6343  * Recursive helper function for ufs_dumpctl().  It follows the indirect file
6344  * system  blocks until it reaches the the disk block addresses, which are
6345  * then stored into the given buffer, storeblk.
6346  */
6347 static daddr32_t *
6348 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp,  daddr32_t *storeblk,
6349     daddr32_t *dblk, int level, int entries)
6350 {
6351 	struct fs	*fs = ufsvfsp->vfs_fs;
6352 	struct buf	*bp;
6353 	int		i;
6354 
6355 	if (level == 0) {
6356 		for (i = 0; i < NINDIR(fs); i++) {
6357 			if (storeblk - dump_info->dblk >= entries)
6358 				break;
6359 			*storeblk++ = dblk[i];
6360 		}
6361 		return (storeblk);
6362 	}
6363 	for (i = 0; i < NINDIR(fs); i++) {
6364 		if (storeblk - dump_info->dblk >= entries)
6365 			break;
6366 		bp = UFS_BREAD(ufsvfsp,
6367 		    ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize);
6368 		if (bp->b_flags & B_ERROR) {
6369 			brelse(bp);
6370 			return (NULL);
6371 		}
6372 		storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr,
6373 		    level - 1, entries);
6374 		brelse(bp);
6375 
6376 		if (storeblk == NULL)
6377 			return (NULL);
6378 	}
6379 	return (storeblk);
6380 }
6381 
6382 /* ARGSUSED */
6383 static int
6384 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag,
6385     struct cred *cr, caller_context_t *ct)
6386 {
6387 	struct inode	*ip = VTOI(vp);
6388 	struct ulockfs	*ulp;
6389 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
6390 	ulong_t		vsa_mask = vsap->vsa_mask;
6391 	int		err = EINVAL;
6392 
6393 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6394 
6395 	/*
6396 	 * Only grab locks if needed - they're not needed to check vsa_mask
6397 	 * or if the mask contains no acl flags.
6398 	 */
6399 	if (vsa_mask != 0) {
6400 		if (err = ufs_lockfs_begin(ufsvfsp, &ulp,
6401 		    ULOCKFS_GETATTR_MASK))
6402 			return (err);
6403 
6404 		rw_enter(&ip->i_contents, RW_READER);
6405 		err = ufs_acl_get(ip, vsap, flag, cr);
6406 		rw_exit(&ip->i_contents);
6407 
6408 		if (ulp)
6409 			ufs_lockfs_end(ulp);
6410 	}
6411 	return (err);
6412 }
6413 
6414 /* ARGSUSED */
6415 static int
6416 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr,
6417     caller_context_t *ct)
6418 {
6419 	struct inode	*ip = VTOI(vp);
6420 	struct ulockfs	*ulp = NULL;
6421 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
6422 	ulong_t		vsa_mask = vsap->vsa_mask;
6423 	int		err;
6424 	int		haverwlock = 1;
6425 	int		trans_size;
6426 	int		donetrans = 0;
6427 	int		retry = 1;
6428 
6429 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
6430 
6431 	/* Abort now if the request is either empty or invalid. */
6432 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6433 	if ((vsa_mask == 0) ||
6434 	    ((vsap->vsa_aclentp == NULL) &&
6435 	    (vsap->vsa_dfaclentp == NULL))) {
6436 		err = EINVAL;
6437 		goto out;
6438 	}
6439 
6440 	/*
6441 	 * Following convention, if this is a directory then we acquire the
6442 	 * inode's i_rwlock after starting a UFS logging transaction;
6443 	 * otherwise, we acquire it beforehand. Since we were called (and
6444 	 * must therefore return) with the lock held, we will have to drop it,
6445 	 * and later reacquire it, if operating on a directory.
6446 	 */
6447 	if (vp->v_type == VDIR) {
6448 		rw_exit(&ip->i_rwlock);
6449 		haverwlock = 0;
6450 	} else {
6451 		/* Upgrade the lock if required. */
6452 		if (!rw_write_held(&ip->i_rwlock)) {
6453 			rw_exit(&ip->i_rwlock);
6454 			rw_enter(&ip->i_rwlock, RW_WRITER);
6455 		}
6456 	}
6457 
6458 again:
6459 	ASSERT(!(vp->v_type == VDIR && haverwlock));
6460 	if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) {
6461 		ulp = NULL;
6462 		retry = 0;
6463 		goto out;
6464 	}
6465 
6466 	/*
6467 	 * Check that the file system supports this operation. Note that
6468 	 * ufs_lockfs_begin() will have checked that the file system had
6469 	 * not been forcibly unmounted.
6470 	 */
6471 	if (ufsvfsp->vfs_fs->fs_ronly) {
6472 		err = EROFS;
6473 		goto out;
6474 	}
6475 	if (ufsvfsp->vfs_nosetsec) {
6476 		err = ENOSYS;
6477 		goto out;
6478 	}
6479 
6480 	if (ulp) {
6481 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR,
6482 		    trans_size = TOP_SETSECATTR_SIZE(VTOI(vp)));
6483 		donetrans = 1;
6484 	}
6485 
6486 	if (vp->v_type == VDIR) {
6487 		rw_enter(&ip->i_rwlock, RW_WRITER);
6488 		haverwlock = 1;
6489 	}
6490 
6491 	ASSERT(haverwlock);
6492 
6493 	/* Do the actual work. */
6494 	rw_enter(&ip->i_contents, RW_WRITER);
6495 	/*
6496 	 * Suppress out of inodes messages if we will retry.
6497 	 */
6498 	if (retry)
6499 		ip->i_flag |= IQUIET;
6500 	err = ufs_acl_set(ip, vsap, flag, cr);
6501 	ip->i_flag &= ~IQUIET;
6502 	rw_exit(&ip->i_contents);
6503 
6504 out:
6505 	if (ulp) {
6506 		if (donetrans) {
6507 			/*
6508 			 * top_end_async() can eventually call
6509 			 * top_end_sync(), which can block. We must
6510 			 * therefore observe the lock-ordering protocol
6511 			 * here as well.
6512 			 */
6513 			if (vp->v_type == VDIR) {
6514 				rw_exit(&ip->i_rwlock);
6515 				haverwlock = 0;
6516 			}
6517 			TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size);
6518 		}
6519 		ufs_lockfs_end(ulp);
6520 	}
6521 	/*
6522 	 * If no inodes available, try scaring a logically-
6523 	 * free one out of the delete queue to someplace
6524 	 * that we can find it.
6525 	 */
6526 	if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
6527 		ufs_delete_drain_wait(ufsvfsp, 1);
6528 		retry = 0;
6529 		if (vp->v_type == VDIR && haverwlock) {
6530 			rw_exit(&ip->i_rwlock);
6531 			haverwlock = 0;
6532 		}
6533 		goto again;
6534 	}
6535 	/*
6536 	 * If we need to reacquire the lock then it is safe to do so
6537 	 * as a reader. This is because ufs_rwunlock(), which will be
6538 	 * called by our caller after we return, does not differentiate
6539 	 * between shared and exclusive locks.
6540 	 */
6541 	if (!haverwlock) {
6542 		ASSERT(vp->v_type == VDIR);
6543 		rw_enter(&ip->i_rwlock, RW_READER);
6544 	}
6545 
6546 	return (err);
6547 }
6548 
6549 /*
6550  * Locate the vnode to be used for an event notification. As this will
6551  * be called prior to the name space change perform basic verification
6552  * that the change will be allowed.
6553  */
6554 
6555 static int
6556 ufs_eventlookup(struct vnode *dvp, char *nm, struct cred *cr,
6557     struct vnode **vpp)
6558 {
6559 	int	namlen;
6560 	int	error;
6561 	struct vnode	*vp;
6562 	struct inode	*ip;
6563 	struct inode	*xip;
6564 	struct ufsvfs	*ufsvfsp;
6565 	struct ulockfs	*ulp;
6566 
6567 	ip = VTOI(dvp);
6568 	*vpp = NULL;
6569 
6570 	if ((namlen = strlen(nm)) == 0)
6571 		return (EINVAL);
6572 
6573 	if (nm[0] == '.') {
6574 		if (namlen == 1)
6575 			return (EINVAL);
6576 		else if ((namlen == 2) && nm[1] == '.') {
6577 			return (EEXIST);
6578 		}
6579 	}
6580 
6581 	/*
6582 	 * Check accessibility and write access of parent directory as we
6583 	 * only want to post the event if we're able to make a change.
6584 	 */
6585 	if (error = ufs_diraccess(ip, IEXEC|IWRITE, cr))
6586 		return (error);
6587 
6588 	if (vp = dnlc_lookup(dvp, nm)) {
6589 		if (vp == DNLC_NO_VNODE) {
6590 			VN_RELE(vp);
6591 			return (ENOENT);
6592 		}
6593 
6594 		*vpp = vp;
6595 		return (0);
6596 	}
6597 
6598 	/*
6599 	 * Keep the idle queue from getting too long by idling two
6600 	 * inodes before attempting to allocate another.
6601 	 * This operation must be performed before entering lockfs
6602 	 * or a transaction.
6603 	 */
6604 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
6605 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
6606 			ins.in_lidles.value.ul += ufs_lookup_idle_count;
6607 			ufs_idle_some(ufs_lookup_idle_count);
6608 		}
6609 
6610 	ufsvfsp = ip->i_ufsvfs;
6611 
6612 retry_lookup:
6613 	if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK))
6614 		return (error);
6615 
6616 	if ((error = ufs_dirlook(ip, nm, &xip, cr, 1, 1)) == 0) {
6617 		vp = ITOV(xip);
6618 		*vpp = vp;
6619 	}
6620 
6621 	if (ulp) {
6622 		ufs_lockfs_end(ulp);
6623 	}
6624 
6625 	if (error == EAGAIN)
6626 		goto retry_lookup;
6627 
6628 	return (error);
6629 }
6630