xref: /illumos-gate/usr/src/uts/common/fs/ufs/ufs_vnops.c (revision d21cedec4ed074c3a6feb4a8a007a9cb83cca060)
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 2015, 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(
2105 	struct vnode *vp,
2106 	struct vattr *vap,
2107 	int flags,
2108 	struct cred *cr,
2109 	caller_context_t *ct)
2110 {
2111 	struct inode *ip = VTOI(vp);
2112 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2113 	struct fs *fs;
2114 	struct ulockfs *ulp;
2115 	char *errmsg1;
2116 	char *errmsg2;
2117 	long blocks;
2118 	long int mask = vap->va_mask;
2119 	size_t len1, len2;
2120 	int issync;
2121 	int trans_size;
2122 	int dotrans;
2123 	int dorwlock;
2124 	int error;
2125 	int owner_change;
2126 	int dodqlock;
2127 	timestruc_t now;
2128 	vattr_t oldva;
2129 	int retry = 1;
2130 	int indeadlock;
2131 
2132 	/*
2133 	 * Cannot set these attributes.
2134 	 */
2135 	if ((mask & AT_NOSET) || (mask & AT_XVATTR))
2136 		return (EINVAL);
2137 
2138 	/*
2139 	 * check for forced unmount
2140 	 */
2141 	if (ufsvfsp == NULL)
2142 		return (EIO);
2143 
2144 	fs = ufsvfsp->vfs_fs;
2145 	if (fs->fs_ronly != 0)
2146 		return (EROFS);
2147 
2148 again:
2149 	errmsg1 = NULL;
2150 	errmsg2 = NULL;
2151 	dotrans = 0;
2152 	dorwlock = 0;
2153 	dodqlock = 0;
2154 
2155 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK);
2156 	if (error)
2157 		goto out;
2158 
2159 	/*
2160 	 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2161 	 * This follows the protocol for read()/write().
2162 	 */
2163 	if (vp->v_type != VDIR) {
2164 		/*
2165 		 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2166 		 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2167 		 * possible, retries the operation.
2168 		 */
2169 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_file);
2170 		if (indeadlock) {
2171 			if (ulp)
2172 				ufs_lockfs_end(ulp);
2173 			goto again;
2174 		}
2175 		dorwlock = 1;
2176 	}
2177 
2178 	/*
2179 	 * Truncate file.  Must have write permission and not be a directory.
2180 	 */
2181 	if (mask & AT_SIZE) {
2182 		rw_enter(&ip->i_contents, RW_WRITER);
2183 		if (vp->v_type == VDIR) {
2184 			error = EISDIR;
2185 			goto update_inode;
2186 		}
2187 		if (error = ufs_iaccess(ip, IWRITE, cr, 0))
2188 			goto update_inode;
2189 
2190 		rw_exit(&ip->i_contents);
2191 		error = TRANS_ITRUNC(ip, vap->va_size, 0, cr);
2192 		if (error) {
2193 			rw_enter(&ip->i_contents, RW_WRITER);
2194 			goto update_inode;
2195 		}
2196 
2197 		if (error == 0 && vap->va_size)
2198 			vnevent_truncate(vp, ct);
2199 	}
2200 
2201 	if (ulp) {
2202 		trans_size = (int)TOP_SETATTR_SIZE(ip);
2203 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size);
2204 		++dotrans;
2205 	}
2206 
2207 	/*
2208 	 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2209 	 * This follows the protocol established by
2210 	 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2211 	 */
2212 	if (vp->v_type == VDIR) {
2213 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_SETATTR,
2214 		    retry_dir);
2215 		if (indeadlock)
2216 			goto again;
2217 		dorwlock = 1;
2218 	}
2219 
2220 	/*
2221 	 * Grab quota lock if we are changing the file's owner.
2222 	 */
2223 	if (mask & AT_UID) {
2224 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2225 		dodqlock = 1;
2226 	}
2227 	rw_enter(&ip->i_contents, RW_WRITER);
2228 
2229 	oldva.va_mode = ip->i_mode;
2230 	oldva.va_uid = ip->i_uid;
2231 	oldva.va_gid = ip->i_gid;
2232 
2233 	vap->va_mask &= ~AT_SIZE;
2234 
2235 	error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2236 	    ufs_priv_access, ip);
2237 	if (error)
2238 		goto update_inode;
2239 
2240 	mask = vap->va_mask;
2241 
2242 	/*
2243 	 * Change file access modes.
2244 	 */
2245 	if (mask & AT_MODE) {
2246 		ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT);
2247 		TRANS_INODE(ufsvfsp, ip);
2248 		ip->i_flag |= ICHG;
2249 		if (stickyhack) {
2250 			mutex_enter(&vp->v_lock);
2251 			if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX)
2252 				vp->v_flag |= VSWAPLIKE;
2253 			else
2254 				vp->v_flag &= ~VSWAPLIKE;
2255 			mutex_exit(&vp->v_lock);
2256 		}
2257 	}
2258 	if (mask & (AT_UID|AT_GID)) {
2259 		if (mask & AT_UID) {
2260 			/*
2261 			 * Don't change ownership of the quota inode.
2262 			 */
2263 			if (ufsvfsp->vfs_qinod == ip) {
2264 				ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED);
2265 				error = EINVAL;
2266 				goto update_inode;
2267 			}
2268 
2269 			/*
2270 			 * No real ownership change.
2271 			 */
2272 			if (ip->i_uid == vap->va_uid) {
2273 				blocks = 0;
2274 				owner_change = 0;
2275 			}
2276 			/*
2277 			 * Remove the blocks and the file, from the old user's
2278 			 * quota.
2279 			 */
2280 			else {
2281 				blocks = ip->i_blocks;
2282 				owner_change = 1;
2283 
2284 				(void) chkdq(ip, -blocks, /* force */ 1, cr,
2285 				    (char **)NULL, (size_t *)NULL);
2286 				(void) chkiq(ufsvfsp, /* change */ -1, ip,
2287 				    (uid_t)ip->i_uid, /* force */ 1, cr,
2288 				    (char **)NULL, (size_t *)NULL);
2289 				dqrele(ip->i_dquot);
2290 			}
2291 
2292 			ip->i_uid = vap->va_uid;
2293 
2294 			/*
2295 			 * There is a real ownership change.
2296 			 */
2297 			if (owner_change) {
2298 				/*
2299 				 * Add the blocks and the file to the new
2300 				 * user's quota.
2301 				 */
2302 				ip->i_dquot = getinoquota(ip);
2303 				(void) chkdq(ip, blocks, /* force */ 1, cr,
2304 				    &errmsg1, &len1);
2305 				(void) chkiq(ufsvfsp, /* change */ 1,
2306 				    (struct inode *)NULL, (uid_t)ip->i_uid,
2307 				    /* force */ 1, cr, &errmsg2, &len2);
2308 			}
2309 		}
2310 		if (mask & AT_GID) {
2311 			ip->i_gid = vap->va_gid;
2312 		}
2313 		TRANS_INODE(ufsvfsp, ip);
2314 		ip->i_flag |= ICHG;
2315 	}
2316 	/*
2317 	 * Change file access or modified times.
2318 	 */
2319 	if (mask & (AT_ATIME|AT_MTIME)) {
2320 		/* Check that the time value is within ufs range */
2321 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2322 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2323 			error = EOVERFLOW;
2324 			goto update_inode;
2325 		}
2326 
2327 		/*
2328 		 * if the "noaccess" mount option is set and only atime
2329 		 * update is requested, do nothing. No error is returned.
2330 		 */
2331 		if ((ufsvfsp->vfs_noatime) &&
2332 		    ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME))
2333 			goto skip_atime;
2334 
2335 		if (mask & AT_ATIME) {
2336 			ip->i_atime.tv_sec = vap->va_atime.tv_sec;
2337 			ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000;
2338 			ip->i_flag &= ~IACC;
2339 		}
2340 		if (mask & AT_MTIME) {
2341 			ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
2342 			ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000;
2343 			gethrestime(&now);
2344 			if (now.tv_sec > TIME32_MAX) {
2345 				/*
2346 				 * In 2038, ctime sticks forever..
2347 				 */
2348 				ip->i_ctime.tv_sec = TIME32_MAX;
2349 				ip->i_ctime.tv_usec = 0;
2350 			} else {
2351 				ip->i_ctime.tv_sec = now.tv_sec;
2352 				ip->i_ctime.tv_usec = now.tv_nsec / 1000;
2353 			}
2354 			ip->i_flag &= ~(IUPD|ICHG);
2355 			ip->i_flag |= IMODTIME;
2356 		}
2357 		TRANS_INODE(ufsvfsp, ip);
2358 		ip->i_flag |= IMOD;
2359 	}
2360 
2361 skip_atime:
2362 	/*
2363 	 * The presence of a shadow inode may indicate an ACL, but does
2364 	 * not imply an ACL.  Future FSD types should be handled here too
2365 	 * and check for the presence of the attribute-specific data
2366 	 * before referencing it.
2367 	 */
2368 	if (ip->i_shadow) {
2369 		/*
2370 		 * XXX if ufs_iupdat is changed to sandbagged write fix
2371 		 * ufs_acl_setattr to push ip to keep acls consistent
2372 		 *
2373 		 * Suppress out of inodes messages if we will retry.
2374 		 */
2375 		if (retry)
2376 			ip->i_flag |= IQUIET;
2377 		error = ufs_acl_setattr(ip, vap, cr);
2378 		ip->i_flag &= ~IQUIET;
2379 	}
2380 
2381 update_inode:
2382 	/*
2383 	 * Setattr always increases the sequence number
2384 	 */
2385 	ip->i_seq++;
2386 
2387 	/*
2388 	 * if nfsd and not logging; push synchronously
2389 	 */
2390 	if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) {
2391 		ufs_iupdat(ip, 1);
2392 	} else {
2393 		ITIMES_NOLOCK(ip);
2394 	}
2395 
2396 	rw_exit(&ip->i_contents);
2397 	if (dodqlock) {
2398 		rw_exit(&ufsvfsp->vfs_dqrwlock);
2399 	}
2400 	if (dorwlock)
2401 		rw_exit(&ip->i_rwlock);
2402 
2403 	if (ulp) {
2404 		if (dotrans) {
2405 			int terr = 0;
2406 			TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR,
2407 			    trans_size);
2408 			if (error == 0)
2409 				error = terr;
2410 		}
2411 		ufs_lockfs_end(ulp);
2412 	}
2413 out:
2414 	/*
2415 	 * If out of inodes or blocks, see if we can free something
2416 	 * up from the delete queue.
2417 	 */
2418 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
2419 		ufs_delete_drain_wait(ufsvfsp, 1);
2420 		retry = 0;
2421 		if (errmsg1 != NULL)
2422 			kmem_free(errmsg1, len1);
2423 		if (errmsg2 != NULL)
2424 			kmem_free(errmsg2, len2);
2425 		goto again;
2426 	}
2427 	if (errmsg1 != NULL) {
2428 		uprintf(errmsg1);
2429 		kmem_free(errmsg1, len1);
2430 	}
2431 	if (errmsg2 != NULL) {
2432 		uprintf(errmsg2);
2433 		kmem_free(errmsg2, len2);
2434 	}
2435 	return (error);
2436 }
2437 
2438 /*ARGSUSED*/
2439 static int
2440 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr,
2441 	caller_context_t *ct)
2442 {
2443 	struct inode *ip = VTOI(vp);
2444 
2445 	if (ip->i_ufsvfs == NULL)
2446 		return (EIO);
2447 
2448 	/*
2449 	 * The ufs_iaccess function wants to be called with
2450 	 * mode bits expressed as "ufs specific" bits.
2451 	 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2452 	 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2453 	 * But since they're the same we just pass the vnode mode
2454 	 * bit but just verify that assumption at compile time.
2455 	 */
2456 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2457 #error "ufs_access needs to map Vmodes to Imodes"
2458 #endif
2459 	return (ufs_iaccess(ip, mode, cr, 1));
2460 }
2461 
2462 /* ARGSUSED */
2463 static int
2464 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr,
2465 	caller_context_t *ct)
2466 {
2467 	struct inode *ip = VTOI(vp);
2468 	struct ufsvfs *ufsvfsp;
2469 	struct ulockfs *ulp;
2470 	int error;
2471 	int fastsymlink;
2472 
2473 	if (vp->v_type != VLNK) {
2474 		error = EINVAL;
2475 		goto nolockout;
2476 	}
2477 
2478 	/*
2479 	 * If the symbolic link is empty there is nothing to read.
2480 	 * Fast-track these empty symbolic links
2481 	 */
2482 	if (ip->i_size == 0) {
2483 		error = 0;
2484 		goto nolockout;
2485 	}
2486 
2487 	ufsvfsp = ip->i_ufsvfs;
2488 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK);
2489 	if (error)
2490 		goto nolockout;
2491 	/*
2492 	 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2493 	 */
2494 again:
2495 	fastsymlink = 0;
2496 	if (ip->i_flag & IFASTSYMLNK) {
2497 		rw_enter(&ip->i_rwlock, RW_READER);
2498 		rw_enter(&ip->i_contents, RW_READER);
2499 		if (ip->i_flag & IFASTSYMLNK) {
2500 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
2501 			    (ip->i_fs->fs_ronly == 0) &&
2502 			    (!ufsvfsp->vfs_noatime)) {
2503 				mutex_enter(&ip->i_tlock);
2504 				ip->i_flag |= IACC;
2505 				mutex_exit(&ip->i_tlock);
2506 			}
2507 			error = uiomove((caddr_t)&ip->i_db[1],
2508 			    MIN(ip->i_size, uiop->uio_resid),
2509 			    UIO_READ, uiop);
2510 			ITIMES(ip);
2511 			++fastsymlink;
2512 		}
2513 		rw_exit(&ip->i_contents);
2514 		rw_exit(&ip->i_rwlock);
2515 	}
2516 	if (!fastsymlink) {
2517 		ssize_t size;	/* number of bytes read  */
2518 		caddr_t basep;	/* pointer to input data */
2519 		ino_t ino;
2520 		long  igen;
2521 		struct uio tuio;	/* temp uio struct */
2522 		struct uio *tuiop;
2523 		iovec_t tiov;		/* temp iovec struct */
2524 		char kbuf[FSL_SIZE];	/* buffer to hold fast symlink */
2525 		int tflag = 0;		/* flag to indicate temp vars used */
2526 
2527 		ino = ip->i_number;
2528 		igen = ip->i_gen;
2529 		size = uiop->uio_resid;
2530 		basep = uiop->uio_iov->iov_base;
2531 		tuiop = uiop;
2532 
2533 		rw_enter(&ip->i_rwlock, RW_WRITER);
2534 		rw_enter(&ip->i_contents, RW_WRITER);
2535 		if (ip->i_flag & IFASTSYMLNK) {
2536 			rw_exit(&ip->i_contents);
2537 			rw_exit(&ip->i_rwlock);
2538 			goto again;
2539 		}
2540 
2541 		/* can this be a fast symlink and is it a user buffer? */
2542 		if (ip->i_size <= FSL_SIZE &&
2543 		    (uiop->uio_segflg == UIO_USERSPACE ||
2544 		    uiop->uio_segflg == UIO_USERISPACE)) {
2545 
2546 			bzero(&tuio, sizeof (struct uio));
2547 			/*
2548 			 * setup a kernel buffer to read link into.  this
2549 			 * is to fix a race condition where the user buffer
2550 			 * got corrupted before copying it into the inode.
2551 			 */
2552 			size = ip->i_size;
2553 			tiov.iov_len = size;
2554 			tiov.iov_base = kbuf;
2555 			tuio.uio_iov = &tiov;
2556 			tuio.uio_iovcnt = 1;
2557 			tuio.uio_offset = uiop->uio_offset;
2558 			tuio.uio_segflg = UIO_SYSSPACE;
2559 			tuio.uio_fmode = uiop->uio_fmode;
2560 			tuio.uio_extflg = uiop->uio_extflg;
2561 			tuio.uio_limit = uiop->uio_limit;
2562 			tuio.uio_resid = size;
2563 
2564 			basep = tuio.uio_iov->iov_base;
2565 			tuiop = &tuio;
2566 			tflag = 1;
2567 		}
2568 
2569 		error = rdip(ip, tuiop, 0, cr);
2570 		if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) {
2571 			rw_exit(&ip->i_contents);
2572 			rw_exit(&ip->i_rwlock);
2573 			goto out;
2574 		}
2575 
2576 		if (tflag == 0)
2577 			size -= uiop->uio_resid;
2578 
2579 		if ((tflag == 0 && ip->i_size <= FSL_SIZE &&
2580 		    ip->i_size == size) || (tflag == 1 &&
2581 		    tuio.uio_resid == 0)) {
2582 			error = kcopy(basep, &ip->i_db[1], ip->i_size);
2583 			if (error == 0) {
2584 				ip->i_flag |= IFASTSYMLNK;
2585 				/*
2586 				 * free page
2587 				 */
2588 				(void) VOP_PUTPAGE(ITOV(ip),
2589 				    (offset_t)0, PAGESIZE,
2590 				    (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC),
2591 				    cr, ct);
2592 			} else {
2593 				int i;
2594 				/* error, clear garbage left behind */
2595 				for (i = 1; i < NDADDR; i++)
2596 					ip->i_db[i] = 0;
2597 				for (i = 0; i < NIADDR; i++)
2598 					ip->i_ib[i] = 0;
2599 			}
2600 		}
2601 		if (tflag == 1) {
2602 			/* now, copy it into the user buffer */
2603 			error = uiomove((caddr_t)kbuf,
2604 			    MIN(size, uiop->uio_resid),
2605 			    UIO_READ, uiop);
2606 		}
2607 		rw_exit(&ip->i_contents);
2608 		rw_exit(&ip->i_rwlock);
2609 	}
2610 out:
2611 	if (ulp) {
2612 		ufs_lockfs_end(ulp);
2613 	}
2614 nolockout:
2615 	return (error);
2616 }
2617 
2618 /* ARGSUSED */
2619 static int
2620 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr,
2621 	caller_context_t *ct)
2622 {
2623 	struct inode *ip = VTOI(vp);
2624 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2625 	struct ulockfs *ulp;
2626 	int error;
2627 
2628 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK);
2629 	if (error)
2630 		return (error);
2631 
2632 	if (TRANS_ISTRANS(ufsvfsp)) {
2633 		/*
2634 		 * First push out any data pages
2635 		 */
2636 		if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2637 		    (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) {
2638 			error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0,
2639 			    0, CRED(), ct);
2640 			if (error)
2641 				goto out;
2642 		}
2643 
2644 		/*
2645 		 * Delta any delayed inode times updates
2646 		 * and push inode to log.
2647 		 * All other inode deltas will have already been delta'd
2648 		 * and will be pushed during the commit.
2649 		 */
2650 		if (!(syncflag & FDSYNC) &&
2651 		    ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) {
2652 			if (ulp) {
2653 				TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC,
2654 				    TOP_SYNCIP_SIZE);
2655 			}
2656 			rw_enter(&ip->i_contents, RW_READER);
2657 			mutex_enter(&ip->i_tlock);
2658 			ip->i_flag &= ~IMODTIME;
2659 			mutex_exit(&ip->i_tlock);
2660 			ufs_iupdat(ip, I_SYNC);
2661 			rw_exit(&ip->i_contents);
2662 			if (ulp) {
2663 				TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC,
2664 				    TOP_SYNCIP_SIZE);
2665 			}
2666 		}
2667 
2668 		/*
2669 		 * Commit the Moby transaction
2670 		 *
2671 		 * Deltas have already been made so we just need to
2672 		 * commit them with a synchronous transaction.
2673 		 * TRANS_BEGIN_SYNC() will return an error
2674 		 * if there are no deltas to commit, for an
2675 		 * empty transaction.
2676 		 */
2677 		if (ulp) {
2678 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE,
2679 			    error);
2680 			if (error) {
2681 				error = 0; /* commit wasn't needed */
2682 				goto out;
2683 			}
2684 			TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC,
2685 			    TOP_COMMIT_SIZE);
2686 		}
2687 	} else {	/* not logging */
2688 		if (!(IS_SWAPVP(vp)))
2689 			if (syncflag & FNODSYNC) {
2690 				/* Just update the inode only */
2691 				TRANS_IUPDAT(ip, 1);
2692 				error = 0;
2693 			} else if (syncflag & FDSYNC)
2694 				/* Do data-synchronous writes */
2695 				error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC);
2696 			else
2697 				/* Do synchronous writes */
2698 				error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC);
2699 
2700 		rw_enter(&ip->i_contents, RW_WRITER);
2701 		if (!error)
2702 			error = ufs_sync_indir(ip);
2703 		rw_exit(&ip->i_contents);
2704 	}
2705 out:
2706 	if (ulp) {
2707 		ufs_lockfs_end(ulp);
2708 	}
2709 	return (error);
2710 }
2711 
2712 /*ARGSUSED*/
2713 static void
2714 ufs_inactive(struct vnode *vp, struct cred *cr, caller_context_t *ct)
2715 {
2716 	ufs_iinactive(VTOI(vp));
2717 }
2718 
2719 /*
2720  * Unix file system operations having to do with directory manipulation.
2721  */
2722 int ufs_lookup_idle_count = 2;	/* Number of inodes to idle each time */
2723 /* ARGSUSED */
2724 static int
2725 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
2726 	struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr,
2727 	caller_context_t *ct, int *direntflags, pathname_t *realpnp)
2728 {
2729 	struct inode *ip;
2730 	struct inode *sip;
2731 	struct inode *xip;
2732 	struct ufsvfs *ufsvfsp;
2733 	struct ulockfs *ulp;
2734 	struct vnode *vp;
2735 	int error;
2736 
2737 	/*
2738 	 * Check flags for type of lookup (regular file or attribute file)
2739 	 */
2740 
2741 	ip = VTOI(dvp);
2742 
2743 	if (flags & LOOKUP_XATTR) {
2744 
2745 		/*
2746 		 * If not mounted with XATTR support then return EINVAL
2747 		 */
2748 
2749 		if (!(ip->i_ufsvfs->vfs_vfs->vfs_flag & VFS_XATTR))
2750 			return (EINVAL);
2751 		/*
2752 		 * We don't allow recursive attributes...
2753 		 * Maybe someday we will.
2754 		 */
2755 		if ((ip->i_cflags & IXATTR)) {
2756 			return (EINVAL);
2757 		}
2758 
2759 		if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) {
2760 			error = ufs_xattr_getattrdir(dvp, &sip, flags, cr);
2761 			if (error) {
2762 				*vpp = NULL;
2763 				goto out;
2764 			}
2765 
2766 			vp = ITOV(sip);
2767 			dnlc_update(dvp, XATTR_DIR_NAME, vp);
2768 		}
2769 
2770 		/*
2771 		 * Check accessibility of directory.
2772 		 */
2773 		if (vp == DNLC_NO_VNODE) {
2774 			VN_RELE(vp);
2775 			error = ENOENT;
2776 			goto out;
2777 		}
2778 		if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr, 1)) != 0) {
2779 			VN_RELE(vp);
2780 			goto out;
2781 		}
2782 
2783 		*vpp = vp;
2784 		return (0);
2785 	}
2786 
2787 	/*
2788 	 * Check for a null component, which we should treat as
2789 	 * looking at dvp from within it's parent, so we don't
2790 	 * need a call to ufs_iaccess(), as it has already been
2791 	 * done.
2792 	 */
2793 	if (nm[0] == 0) {
2794 		VN_HOLD(dvp);
2795 		error = 0;
2796 		*vpp = dvp;
2797 		goto out;
2798 	}
2799 
2800 	/*
2801 	 * Check for "." ie itself. this is a quick check and
2802 	 * avoids adding "." into the dnlc (which have been seen
2803 	 * to occupy >10% of the cache).
2804 	 */
2805 	if ((nm[0] == '.') && (nm[1] == 0)) {
2806 		/*
2807 		 * Don't return without checking accessibility
2808 		 * of the directory. We only need the lock if
2809 		 * we are going to return it.
2810 		 */
2811 		if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) == 0) {
2812 			VN_HOLD(dvp);
2813 			*vpp = dvp;
2814 		}
2815 		goto out;
2816 	}
2817 
2818 	/*
2819 	 * Fast path: Check the directory name lookup cache.
2820 	 */
2821 	if (vp = dnlc_lookup(dvp, nm)) {
2822 		/*
2823 		 * Check accessibility of directory.
2824 		 */
2825 		if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) != 0) {
2826 			VN_RELE(vp);
2827 			goto out;
2828 		}
2829 		if (vp == DNLC_NO_VNODE) {
2830 			VN_RELE(vp);
2831 			error = ENOENT;
2832 			goto out;
2833 		}
2834 		xip = VTOI(vp);
2835 		ulp = NULL;
2836 		goto fastpath;
2837 	}
2838 
2839 	/*
2840 	 * Keep the idle queue from getting too long by
2841 	 * idling two inodes before attempting to allocate another.
2842 	 *    This operation must be performed before entering
2843 	 *    lockfs or a transaction.
2844 	 */
2845 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2846 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2847 			ins.in_lidles.value.ul += ufs_lookup_idle_count;
2848 			ufs_idle_some(ufs_lookup_idle_count);
2849 		}
2850 
2851 retry_lookup:
2852 	/*
2853 	 * Check accessibility of directory.
2854 	 */
2855 	if (error = ufs_diraccess(ip, IEXEC, cr))
2856 		goto out;
2857 
2858 	ufsvfsp = ip->i_ufsvfs;
2859 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK);
2860 	if (error)
2861 		goto out;
2862 
2863 	error = ufs_dirlook(ip, nm, &xip, cr, 1, 0);
2864 
2865 fastpath:
2866 	if (error == 0) {
2867 		ip = xip;
2868 		*vpp = ITOV(ip);
2869 
2870 		/*
2871 		 * If vnode is a device return special vnode instead.
2872 		 */
2873 		if (IS_DEVVP(*vpp)) {
2874 			struct vnode *newvp;
2875 
2876 			newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
2877 			    cr);
2878 			VN_RELE(*vpp);
2879 			if (newvp == NULL)
2880 				error = ENOSYS;
2881 			else
2882 				*vpp = newvp;
2883 		} else if (ip->i_cflags & ICOMPRESS) {
2884 			struct vnode *newvp;
2885 
2886 			/*
2887 			 * Compressed file, substitute dcfs vnode
2888 			 */
2889 			newvp = decompvp(*vpp, cr, ct);
2890 			VN_RELE(*vpp);
2891 			if (newvp == NULL)
2892 				error = ENOSYS;
2893 			else
2894 				*vpp = newvp;
2895 		}
2896 	}
2897 	if (ulp) {
2898 		ufs_lockfs_end(ulp);
2899 	}
2900 
2901 	if (error == EAGAIN)
2902 		goto retry_lookup;
2903 
2904 out:
2905 	return (error);
2906 }
2907 
2908 /*ARGSUSED*/
2909 static int
2910 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl,
2911 	int mode, struct vnode **vpp, struct cred *cr, int flag,
2912 	caller_context_t *ct, vsecattr_t *vsecp)
2913 {
2914 	struct inode *ip;
2915 	struct inode *xip;
2916 	struct inode *dip;
2917 	struct vnode *xvp;
2918 	struct ufsvfs *ufsvfsp;
2919 	struct ulockfs *ulp;
2920 	int error;
2921 	int issync;
2922 	int truncflag;
2923 	int trans_size;
2924 	int noentry;
2925 	int defer_dip_seq_update = 0;	/* need to defer update of dip->i_seq */
2926 	int retry = 1;
2927 	int indeadlock;
2928 
2929 again:
2930 	ip = VTOI(dvp);
2931 	ufsvfsp = ip->i_ufsvfs;
2932 	truncflag = 0;
2933 
2934 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK);
2935 	if (error)
2936 		goto out;
2937 
2938 	if (ulp) {
2939 		trans_size = (int)TOP_CREATE_SIZE(ip);
2940 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size);
2941 	}
2942 
2943 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
2944 		vap->va_mode &= ~VSVTX;
2945 
2946 	if (*name == '\0') {
2947 		/*
2948 		 * Null component name refers to the directory itself.
2949 		 */
2950 		VN_HOLD(dvp);
2951 		/*
2952 		 * Even though this is an error case, we need to grab the
2953 		 * quota lock since the error handling code below is common.
2954 		 */
2955 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2956 		rw_enter(&ip->i_contents, RW_WRITER);
2957 		error = EEXIST;
2958 	} else {
2959 		xip = NULL;
2960 		noentry = 0;
2961 		/*
2962 		 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2963 		 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2964 		 * possible, retries the operation.
2965 		 */
2966 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_CREATE,
2967 		    retry_dir);
2968 		if (indeadlock)
2969 			goto again;
2970 
2971 		xvp = dnlc_lookup(dvp, name);
2972 		if (xvp == DNLC_NO_VNODE) {
2973 			noentry = 1;
2974 			VN_RELE(xvp);
2975 			xvp = NULL;
2976 		}
2977 		if (xvp) {
2978 			rw_exit(&ip->i_rwlock);
2979 			if (error = ufs_iaccess(ip, IEXEC, cr, 1)) {
2980 				VN_RELE(xvp);
2981 			} else {
2982 				error = EEXIST;
2983 				xip = VTOI(xvp);
2984 			}
2985 		} else {
2986 			/*
2987 			 * Suppress file system full message if we will retry
2988 			 */
2989 			error = ufs_direnter_cm(ip, name, DE_CREATE,
2990 			    vap, &xip, cr, (noentry | (retry ? IQUIET : 0)));
2991 			if (error == EAGAIN) {
2992 				if (ulp) {
2993 					TRANS_END_CSYNC(ufsvfsp, error, issync,
2994 					    TOP_CREATE, trans_size);
2995 					ufs_lockfs_end(ulp);
2996 				}
2997 				goto again;
2998 			}
2999 			rw_exit(&ip->i_rwlock);
3000 		}
3001 		ip = xip;
3002 		if (ip != NULL) {
3003 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
3004 			rw_enter(&ip->i_contents, RW_WRITER);
3005 		}
3006 	}
3007 
3008 	/*
3009 	 * If the file already exists and this is a non-exclusive create,
3010 	 * check permissions and allow access for non-directories.
3011 	 * Read-only create of an existing directory is also allowed.
3012 	 * We fail an exclusive create of anything which already exists.
3013 	 */
3014 	if (error == EEXIST) {
3015 		dip = VTOI(dvp);
3016 		if (excl == NONEXCL) {
3017 			if ((((ip->i_mode & IFMT) == IFDIR) ||
3018 			    ((ip->i_mode & IFMT) == IFATTRDIR)) &&
3019 			    (mode & IWRITE))
3020 				error = EISDIR;
3021 			else if (mode)
3022 				error = ufs_iaccess(ip, mode, cr, 0);
3023 			else
3024 				error = 0;
3025 		}
3026 		if (error) {
3027 			rw_exit(&ip->i_contents);
3028 			rw_exit(&ufsvfsp->vfs_dqrwlock);
3029 			VN_RELE(ITOV(ip));
3030 			goto unlock;
3031 		}
3032 		/*
3033 		 * If the error EEXIST was set, then i_seq can not
3034 		 * have been updated. The sequence number interface
3035 		 * is defined such that a non-error VOP_CREATE must
3036 		 * increase the dir va_seq it by at least one. If we
3037 		 * have cleared the error, increase i_seq. Note that
3038 		 * we are increasing the dir i_seq and in rare cases
3039 		 * ip may actually be from the dvp, so we already have
3040 		 * the locks and it will not be subject to truncation.
3041 		 * In case we have to update i_seq of the parent
3042 		 * directory dip, we have to defer it till we have
3043 		 * released our locks on ip due to lock ordering requirements.
3044 		 */
3045 		if (ip != dip)
3046 			defer_dip_seq_update = 1;
3047 		else
3048 			ip->i_seq++;
3049 
3050 		if (((ip->i_mode & IFMT) == IFREG) &&
3051 		    (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
3052 			/*
3053 			 * Truncate regular files, if requested by caller.
3054 			 * Grab i_rwlock to make sure no one else is
3055 			 * currently writing to the file (we promised
3056 			 * bmap we would do this).
3057 			 * Must get the locks in the correct order.
3058 			 */
3059 			if (ip->i_size == 0) {
3060 				ip->i_flag |= ICHG | IUPD;
3061 				ip->i_seq++;
3062 				TRANS_INODE(ufsvfsp, ip);
3063 			} else {
3064 				/*
3065 				 * Large Files: Why this check here?
3066 				 * Though we do it in vn_create() we really
3067 				 * want to guarantee that we do not destroy
3068 				 * Large file data by atomically checking
3069 				 * the size while holding the contents
3070 				 * lock.
3071 				 */
3072 				if (flag && !(flag & FOFFMAX) &&
3073 				    ((ip->i_mode & IFMT) == IFREG) &&
3074 				    (ip->i_size > (offset_t)MAXOFF32_T)) {
3075 					rw_exit(&ip->i_contents);
3076 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3077 					error = EOVERFLOW;
3078 					goto unlock;
3079 				}
3080 				if (TRANS_ISTRANS(ufsvfsp))
3081 					truncflag++;
3082 				else {
3083 					rw_exit(&ip->i_contents);
3084 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3085 					ufs_tryirwlock_trans(&ip->i_rwlock,
3086 					    RW_WRITER, TOP_CREATE,
3087 					    retry_file);
3088 					if (indeadlock) {
3089 						VN_RELE(ITOV(ip));
3090 						goto again;
3091 					}
3092 					rw_enter(&ufsvfsp->vfs_dqrwlock,
3093 					    RW_READER);
3094 					rw_enter(&ip->i_contents, RW_WRITER);
3095 					(void) ufs_itrunc(ip, (u_offset_t)0, 0,
3096 					    cr);
3097 					rw_exit(&ip->i_rwlock);
3098 				}
3099 
3100 			}
3101 			if (error == 0) {
3102 				vnevent_create(ITOV(ip), ct);
3103 			}
3104 		}
3105 	}
3106 
3107 	if (error) {
3108 		if (ip != NULL) {
3109 			rw_exit(&ufsvfsp->vfs_dqrwlock);
3110 			rw_exit(&ip->i_contents);
3111 		}
3112 		goto unlock;
3113 	}
3114 
3115 	*vpp = ITOV(ip);
3116 	ITIMES(ip);
3117 	rw_exit(&ip->i_contents);
3118 	rw_exit(&ufsvfsp->vfs_dqrwlock);
3119 
3120 	/*
3121 	 * If vnode is a device return special vnode instead.
3122 	 */
3123 	if (!error && IS_DEVVP(*vpp)) {
3124 		struct vnode *newvp;
3125 
3126 		newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
3127 		VN_RELE(*vpp);
3128 		if (newvp == NULL) {
3129 			error = ENOSYS;
3130 			goto unlock;
3131 		}
3132 		truncflag = 0;
3133 		*vpp = newvp;
3134 	}
3135 unlock:
3136 
3137 	/*
3138 	 * Do the deferred update of the parent directory's sequence
3139 	 * number now.
3140 	 */
3141 	if (defer_dip_seq_update == 1) {
3142 		rw_enter(&dip->i_contents, RW_READER);
3143 		mutex_enter(&dip->i_tlock);
3144 		dip->i_seq++;
3145 		mutex_exit(&dip->i_tlock);
3146 		rw_exit(&dip->i_contents);
3147 	}
3148 
3149 	if (ulp) {
3150 		int terr = 0;
3151 
3152 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE,
3153 		    trans_size);
3154 
3155 		/*
3156 		 * If we haven't had a more interesting failure
3157 		 * already, then anything that might've happened
3158 		 * here should be reported.
3159 		 */
3160 		if (error == 0)
3161 			error = terr;
3162 	}
3163 
3164 	if (!error && truncflag) {
3165 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_trunc);
3166 		if (indeadlock) {
3167 			if (ulp)
3168 				ufs_lockfs_end(ulp);
3169 			VN_RELE(ITOV(ip));
3170 			goto again;
3171 		}
3172 		(void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr);
3173 		rw_exit(&ip->i_rwlock);
3174 	}
3175 
3176 	if (ulp)
3177 		ufs_lockfs_end(ulp);
3178 
3179 	/*
3180 	 * If no inodes available, try to free one up out of the
3181 	 * pending delete queue.
3182 	 */
3183 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3184 		ufs_delete_drain_wait(ufsvfsp, 1);
3185 		retry = 0;
3186 		goto again;
3187 	}
3188 
3189 out:
3190 	return (error);
3191 }
3192 
3193 extern int ufs_idle_max;
3194 /*ARGSUSED*/
3195 static int
3196 ufs_remove(struct vnode *vp, char *nm, struct cred *cr,
3197 	caller_context_t *ct, int flags)
3198 {
3199 	struct inode *ip = VTOI(vp);
3200 	struct ufsvfs *ufsvfsp	= ip->i_ufsvfs;
3201 	struct ulockfs *ulp;
3202 	vnode_t *rmvp = NULL;	/* Vnode corresponding to name being removed */
3203 	int indeadlock;
3204 	int error;
3205 	int issync;
3206 	int trans_size;
3207 
3208 	/*
3209 	 * don't let the delete queue get too long
3210 	 */
3211 	if (ufsvfsp == NULL) {
3212 		error = EIO;
3213 		goto out;
3214 	}
3215 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3216 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3217 
3218 	error = ufs_eventlookup(vp, nm, cr, &rmvp);
3219 	if (rmvp != NULL) {
3220 		/* Only send the event if there were no errors */
3221 		if (error == 0)
3222 			vnevent_remove(rmvp, vp, nm, ct);
3223 		VN_RELE(rmvp);
3224 	}
3225 
3226 retry_remove:
3227 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK);
3228 	if (error)
3229 		goto out;
3230 
3231 	if (ulp)
3232 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE,
3233 		    trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp)));
3234 
3235 	/*
3236 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3237 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3238 	 * possible, retries the operation.
3239 	 */
3240 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_REMOVE, retry);
3241 	if (indeadlock)
3242 		goto retry_remove;
3243 	error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0,
3244 	    DR_REMOVE, cr);
3245 	rw_exit(&ip->i_rwlock);
3246 
3247 	if (ulp) {
3248 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size);
3249 		ufs_lockfs_end(ulp);
3250 	}
3251 
3252 out:
3253 	return (error);
3254 }
3255 
3256 /*
3257  * Link a file or a directory.  Only privileged processes are allowed to
3258  * make links to directories.
3259  */
3260 /*ARGSUSED*/
3261 static int
3262 ufs_link(struct vnode *tdvp, struct vnode *svp, char *tnm, struct cred *cr,
3263 	caller_context_t *ct, int flags)
3264 {
3265 	struct inode *sip;
3266 	struct inode *tdp = VTOI(tdvp);
3267 	struct ufsvfs *ufsvfsp = tdp->i_ufsvfs;
3268 	struct ulockfs *ulp;
3269 	struct vnode *realvp;
3270 	int error;
3271 	int issync;
3272 	int trans_size;
3273 	int isdev;
3274 	int indeadlock;
3275 
3276 retry_link:
3277 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LINK_MASK);
3278 	if (error)
3279 		goto out;
3280 
3281 	if (ulp)
3282 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_LINK,
3283 		    trans_size = (int)TOP_LINK_SIZE(VTOI(tdvp)));
3284 
3285 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3286 		svp = realvp;
3287 
3288 	/*
3289 	 * Make sure link for extended attributes is valid
3290 	 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3291 	 *
3292 	 * Make certain we don't attempt to look at a device node as
3293 	 * a ufs inode.
3294 	 */
3295 
3296 	isdev = IS_DEVVP(svp);
3297 	if (((isdev == 0) && ((VTOI(svp)->i_cflags & IXATTR) == 0) &&
3298 	    ((tdp->i_mode & IFMT) == IFATTRDIR)) ||
3299 	    ((isdev == 0) && (VTOI(svp)->i_cflags & IXATTR) &&
3300 	    ((tdp->i_mode & IFMT) == IFDIR))) {
3301 		error = EINVAL;
3302 		goto unlock;
3303 	}
3304 
3305 	sip = VTOI(svp);
3306 	if ((svp->v_type == VDIR &&
3307 	    secpolicy_fs_linkdir(cr, ufsvfsp->vfs_vfs) != 0) ||
3308 	    (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)) {
3309 		error = EPERM;
3310 		goto unlock;
3311 	}
3312 
3313 	/*
3314 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3315 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3316 	 * possible, retries the operation.
3317 	 */
3318 	ufs_tryirwlock_trans(&tdp->i_rwlock, RW_WRITER, TOP_LINK, retry);
3319 	if (indeadlock)
3320 		goto retry_link;
3321 	error = ufs_direnter_lr(tdp, tnm, DE_LINK, (struct inode *)0,
3322 	    sip, cr);
3323 	rw_exit(&tdp->i_rwlock);
3324 
3325 unlock:
3326 	if (ulp) {
3327 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_LINK, trans_size);
3328 		ufs_lockfs_end(ulp);
3329 	}
3330 
3331 	if (!error) {
3332 		vnevent_link(svp, ct);
3333 	}
3334 out:
3335 	return (error);
3336 }
3337 
3338 uint64_t ufs_rename_retry_cnt;
3339 uint64_t ufs_rename_upgrade_retry_cnt;
3340 uint64_t ufs_rename_dircheck_retry_cnt;
3341 clock_t	 ufs_rename_backoff_delay = 1;
3342 
3343 /*
3344  * Rename a file or directory.
3345  * We are given the vnode and entry string of the source and the
3346  * vnode and entry string of the place we want to move the source
3347  * to (the target). The essential operation is:
3348  *	unlink(target);
3349  *	link(source, target);
3350  *	unlink(source);
3351  * but "atomically".  Can't do full commit without saving state in
3352  * the inode on disk, which isn't feasible at this time.  Best we
3353  * can do is always guarantee that the TARGET exists.
3354  */
3355 
3356 /*ARGSUSED*/
3357 static int
3358 ufs_rename(
3359 	struct vnode *sdvp,		/* old (source) parent vnode */
3360 	char *snm,			/* old (source) entry name */
3361 	struct vnode *tdvp,		/* new (target) parent vnode */
3362 	char *tnm,			/* new (target) entry name */
3363 	struct cred *cr,
3364 	caller_context_t *ct,
3365 	int flags)
3366 {
3367 	struct inode *sip = NULL;	/* source inode */
3368 	struct inode *ip = NULL;	/* check inode */
3369 	struct inode *sdp;		/* old (source) parent inode */
3370 	struct inode *tdp;		/* new (target) parent inode */
3371 	struct vnode *svp = NULL;	/* source vnode */
3372 	struct vnode *tvp = NULL;	/* target vnode, if it exists */
3373 	struct vnode *realvp;
3374 	struct ufsvfs *ufsvfsp;
3375 	struct ulockfs *ulp = NULL;
3376 	struct ufs_slot slot;
3377 	timestruc_t now;
3378 	int error;
3379 	int issync;
3380 	int trans_size;
3381 	krwlock_t *first_lock;
3382 	krwlock_t *second_lock;
3383 	krwlock_t *reverse_lock;
3384 	int serr, terr;
3385 
3386 	sdp = VTOI(sdvp);
3387 	slot.fbp = NULL;
3388 	ufsvfsp = sdp->i_ufsvfs;
3389 
3390 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3391 		tdvp = realvp;
3392 
3393 	/* Must do this before taking locks in case of DNLC miss */
3394 	terr = ufs_eventlookup(tdvp, tnm, cr, &tvp);
3395 	serr = ufs_eventlookup(sdvp, snm, cr, &svp);
3396 
3397 	if ((serr == 0) && ((terr == 0) || (terr == ENOENT))) {
3398 		if (tvp != NULL)
3399 			vnevent_pre_rename_dest(tvp, tdvp, tnm, ct);
3400 
3401 		/*
3402 		 * Notify the target directory of the rename event
3403 		 * if source and target directories are not the same.
3404 		 */
3405 		if (sdvp != tdvp)
3406 			vnevent_pre_rename_dest_dir(tdvp, svp, tnm, ct);
3407 
3408 		if (svp != NULL)
3409 			vnevent_pre_rename_src(svp, sdvp, snm, ct);
3410 	}
3411 
3412 	if (svp != NULL)
3413 		VN_RELE(svp);
3414 
3415 retry_rename:
3416 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
3417 	if (error)
3418 		goto unlock;
3419 
3420 	if (ulp)
3421 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
3422 		    trans_size = (int)TOP_RENAME_SIZE(sdp));
3423 
3424 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3425 		tdvp = realvp;
3426 
3427 	tdp = VTOI(tdvp);
3428 
3429 	/*
3430 	 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3431 	 */
3432 	if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) {
3433 		error = EINVAL;
3434 		goto unlock;
3435 	}
3436 
3437 	/*
3438 	 * Check accessibility of directory.
3439 	 */
3440 	if (error = ufs_diraccess(sdp, IEXEC, cr))
3441 		goto unlock;
3442 
3443 	/*
3444 	 * Look up inode of file we're supposed to rename.
3445 	 */
3446 	gethrestime(&now);
3447 	if (error = ufs_dirlook(sdp, snm, &sip, cr, 0, 0)) {
3448 		if (error == EAGAIN) {
3449 			if (ulp) {
3450 				TRANS_END_CSYNC(ufsvfsp, error, issync,
3451 				    TOP_RENAME, trans_size);
3452 				ufs_lockfs_end(ulp);
3453 			}
3454 			goto retry_rename;
3455 		}
3456 
3457 		goto unlock;
3458 	}
3459 
3460 	/*
3461 	 * Lock both the source and target directories (they may be
3462 	 * the same) to provide the atomicity semantics that was
3463 	 * previously provided by the per file system vfs_rename_lock
3464 	 *
3465 	 * with vfs_rename_lock removed to allow simultaneous renames
3466 	 * within a file system, ufs_dircheckpath can deadlock while
3467 	 * traversing back to ensure that source is not a parent directory
3468 	 * of target parent directory. This is because we get into
3469 	 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3470 	 * If the tdp and sdp of the simultaneous renames happen to be
3471 	 * in the path of each other, it can lead to a deadlock. This
3472 	 * can be avoided by getting the locks as RW_READER here and then
3473 	 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3474 	 *
3475 	 * We hold the target directory's i_rwlock after calling
3476 	 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3477 	 * VOP_RWLOCK is explicitly called by the filesystem independent code
3478 	 * before calling the file system operation. In these cases the order
3479 	 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3480 	 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3481 	 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3482 	 * synchronizing object which might lead to a deadlock. So we use
3483 	 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3484 	 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3485 	 * operation.
3486 	 */
3487 retry:
3488 	first_lock = &tdp->i_rwlock;
3489 	second_lock = &sdp->i_rwlock;
3490 retry_firstlock:
3491 	if (!rw_tryenter(first_lock, RW_READER)) {
3492 		/*
3493 		 * We didn't get the lock. Check if the SLOCK is set in the
3494 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3495 		 * and wait for SLOCK to be cleared.
3496 		 */
3497 
3498 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3499 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3500 			    trans_size);
3501 			ufs_lockfs_end(ulp);
3502 			goto retry_rename;
3503 
3504 		} else {
3505 			/*
3506 			 * SLOCK isn't set so this is a genuine synchronization
3507 			 * case. Let's try again after giving them a breather.
3508 			 */
3509 			delay(RETRY_LOCK_DELAY);
3510 			goto  retry_firstlock;
3511 		}
3512 	}
3513 	/*
3514 	 * Need to check if the tdp and sdp are same !!!
3515 	 */
3516 	if ((tdp != sdp) && (!rw_tryenter(second_lock, RW_READER))) {
3517 		/*
3518 		 * We didn't get the lock. Check if the SLOCK is set in the
3519 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3520 		 * and wait for SLOCK to be cleared.
3521 		 */
3522 
3523 		rw_exit(first_lock);
3524 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3525 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3526 			    trans_size);
3527 			ufs_lockfs_end(ulp);
3528 			goto retry_rename;
3529 
3530 		} else {
3531 			/*
3532 			 * So we couldn't get the second level peer lock *and*
3533 			 * the SLOCK bit isn't set. Too bad we can be
3534 			 * contentding with someone wanting these locks otherway
3535 			 * round. Reverse the locks in case there is a heavy
3536 			 * contention for the second level lock.
3537 			 */
3538 			reverse_lock = first_lock;
3539 			first_lock = second_lock;
3540 			second_lock = reverse_lock;
3541 			ufs_rename_retry_cnt++;
3542 			goto  retry_firstlock;
3543 		}
3544 	}
3545 
3546 	if (sip == tdp) {
3547 		error = EINVAL;
3548 		goto errout;
3549 	}
3550 	/*
3551 	 * Make sure we can delete the source entry.  This requires
3552 	 * write permission on the containing directory.
3553 	 * Check for sticky directories.
3554 	 */
3555 	rw_enter(&sdp->i_contents, RW_READER);
3556 	rw_enter(&sip->i_contents, RW_READER);
3557 	if ((error = ufs_iaccess(sdp, IWRITE, cr, 0)) != 0 ||
3558 	    (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) {
3559 		rw_exit(&sip->i_contents);
3560 		rw_exit(&sdp->i_contents);
3561 		goto errout;
3562 	}
3563 
3564 	/*
3565 	 * If this is a rename of a directory and the parent is
3566 	 * different (".." must be changed), then the source
3567 	 * directory must not be in the directory hierarchy
3568 	 * above the target, as this would orphan everything
3569 	 * below the source directory.  Also the user must have
3570 	 * write permission in the source so as to be able to
3571 	 * change "..".
3572 	 */
3573 	if ((((sip->i_mode & IFMT) == IFDIR) ||
3574 	    ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) {
3575 		ino_t	inum;
3576 
3577 		if (error = ufs_iaccess(sip, IWRITE, cr, 0)) {
3578 			rw_exit(&sip->i_contents);
3579 			rw_exit(&sdp->i_contents);
3580 			goto errout;
3581 		}
3582 		inum = sip->i_number;
3583 		rw_exit(&sip->i_contents);
3584 		rw_exit(&sdp->i_contents);
3585 		if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) {
3586 			/*
3587 			 * If we got EAGAIN ufs_dircheckpath detected a
3588 			 * potential deadlock and backed out. We need
3589 			 * to retry the operation since sdp and tdp have
3590 			 * to be released to avoid the deadlock.
3591 			 */
3592 			if (error == EAGAIN) {
3593 				rw_exit(&tdp->i_rwlock);
3594 				if (tdp != sdp)
3595 					rw_exit(&sdp->i_rwlock);
3596 				delay(ufs_rename_backoff_delay);
3597 				ufs_rename_dircheck_retry_cnt++;
3598 				goto retry;
3599 			}
3600 			goto errout;
3601 		}
3602 	} else {
3603 		rw_exit(&sip->i_contents);
3604 		rw_exit(&sdp->i_contents);
3605 	}
3606 
3607 
3608 	/*
3609 	 * Check for renaming '.' or '..' or alias of '.'
3610 	 */
3611 	if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) {
3612 		error = EINVAL;
3613 		goto errout;
3614 	}
3615 
3616 	/*
3617 	 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3618 	 * tries to traverse back the file tree with both tdp and sdp held
3619 	 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3620 	 * as RW_READERS  till ufs_dircheckpath is done.
3621 	 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3622 	 * to RW_WRITER.
3623 	 */
3624 	if (!rw_tryupgrade(&tdp->i_rwlock)) {
3625 		/*
3626 		 * The upgrade failed. We got to give away the lock
3627 		 * as to avoid deadlocking with someone else who is
3628 		 * waiting for writer lock. With the lock gone, we
3629 		 * cannot be sure the checks done above will hold
3630 		 * good when we eventually get them back as writer.
3631 		 * So if we can't upgrade we drop the locks and retry
3632 		 * everything again.
3633 		 */
3634 		rw_exit(&tdp->i_rwlock);
3635 		if (tdp != sdp)
3636 			rw_exit(&sdp->i_rwlock);
3637 		delay(ufs_rename_backoff_delay);
3638 		ufs_rename_upgrade_retry_cnt++;
3639 		goto retry;
3640 	}
3641 	if (tdp != sdp) {
3642 		if (!rw_tryupgrade(&sdp->i_rwlock)) {
3643 			/*
3644 			 * The upgrade failed. We got to give away the lock
3645 			 * as to avoid deadlocking with someone else who is
3646 			 * waiting for writer lock. With the lock gone, we
3647 			 * cannot be sure the checks done above will hold
3648 			 * good when we eventually get them back as writer.
3649 			 * So if we can't upgrade we drop the locks and retry
3650 			 * everything again.
3651 			 */
3652 			rw_exit(&tdp->i_rwlock);
3653 			rw_exit(&sdp->i_rwlock);
3654 			delay(ufs_rename_backoff_delay);
3655 			ufs_rename_upgrade_retry_cnt++;
3656 			goto retry;
3657 		}
3658 	}
3659 
3660 	/*
3661 	 * Now that all the locks are held check to make sure another thread
3662 	 * didn't slip in and take out the sip.
3663 	 */
3664 	slot.status = NONE;
3665 	if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec ||
3666 	    sip->i_ctime.tv_sec > now.tv_sec) {
3667 		rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER);
3668 		rw_enter(&sdp->i_contents, RW_WRITER);
3669 		error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot,
3670 		    &ip, cr, 0);
3671 		rw_exit(&sdp->i_contents);
3672 		rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock);
3673 		if (error) {
3674 			goto errout;
3675 		}
3676 		if (ip == NULL) {
3677 			error = ENOENT;
3678 			goto errout;
3679 		} else {
3680 			/*
3681 			 * If the inode was found need to drop the v_count
3682 			 * so as not to keep the filesystem from being
3683 			 * unmounted at a later time.
3684 			 */
3685 			VN_RELE(ITOV(ip));
3686 		}
3687 
3688 		/*
3689 		 * Release the slot.fbp that has the page mapped and
3690 		 * locked SE_SHARED, and could be used in in
3691 		 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3692 		 * on said page.
3693 		 */
3694 		if (slot.fbp) {
3695 			fbrelse(slot.fbp, S_OTHER);
3696 			slot.fbp = NULL;
3697 		}
3698 	}
3699 
3700 	/*
3701 	 * Link source to the target.
3702 	 */
3703 	if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr)) {
3704 		/*
3705 		 * ESAME isn't really an error; it indicates that the
3706 		 * operation should not be done because the source and target
3707 		 * are the same file, but that no error should be reported.
3708 		 */
3709 		if (error == ESAME)
3710 			error = 0;
3711 		goto errout;
3712 	}
3713 
3714 	if (error == 0 && tvp != NULL)
3715 		vnevent_rename_dest(tvp, tdvp, tnm, ct);
3716 
3717 	/*
3718 	 * Unlink the source.
3719 	 * Remove the source entry.  ufs_dirremove() checks that the entry
3720 	 * still reflects sip, and returns an error if it doesn't.
3721 	 * If the entry has changed just forget about it.  Release
3722 	 * the source inode.
3723 	 */
3724 	if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0,
3725 	    DR_RENAME, cr)) == ENOENT)
3726 		error = 0;
3727 
3728 	if (error == 0) {
3729 		vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
3730 		/*
3731 		 * Notify the target directory of the rename event
3732 		 * if source and target directories are not the same.
3733 		 */
3734 		if (sdvp != tdvp)
3735 			vnevent_rename_dest_dir(tdvp, ct);
3736 	}
3737 
3738 errout:
3739 	if (slot.fbp)
3740 		fbrelse(slot.fbp, S_OTHER);
3741 
3742 	rw_exit(&tdp->i_rwlock);
3743 	if (sdp != tdp) {
3744 		rw_exit(&sdp->i_rwlock);
3745 	}
3746 
3747 unlock:
3748 	if (tvp != NULL)
3749 		VN_RELE(tvp);
3750 	if (sip != NULL)
3751 		VN_RELE(ITOV(sip));
3752 
3753 	if (ulp) {
3754 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
3755 		ufs_lockfs_end(ulp);
3756 	}
3757 
3758 	return (error);
3759 }
3760 
3761 /*ARGSUSED*/
3762 static int
3763 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap,
3764 	struct vnode **vpp, struct cred *cr, caller_context_t *ct, int flags,
3765 	vsecattr_t *vsecp)
3766 {
3767 	struct inode *ip;
3768 	struct inode *xip;
3769 	struct ufsvfs *ufsvfsp;
3770 	struct ulockfs *ulp;
3771 	int error;
3772 	int issync;
3773 	int trans_size;
3774 	int indeadlock;
3775 	int retry = 1;
3776 
3777 	ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
3778 
3779 	/*
3780 	 * Can't make directory in attr hidden dir
3781 	 */
3782 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3783 		return (EINVAL);
3784 
3785 again:
3786 	ip = VTOI(dvp);
3787 	ufsvfsp = ip->i_ufsvfs;
3788 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK);
3789 	if (error)
3790 		goto out;
3791 	if (ulp)
3792 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR,
3793 		    trans_size = (int)TOP_MKDIR_SIZE(ip));
3794 
3795 	/*
3796 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3797 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3798 	 * possible, retries the operation.
3799 	 */
3800 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_MKDIR, retry);
3801 	if (indeadlock)
3802 		goto again;
3803 
3804 	error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr,
3805 	    (retry ? IQUIET : 0));
3806 	if (error == EAGAIN) {
3807 		if (ulp) {
3808 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_MKDIR,
3809 			    trans_size);
3810 			ufs_lockfs_end(ulp);
3811 		}
3812 		goto again;
3813 	}
3814 
3815 	rw_exit(&ip->i_rwlock);
3816 	if (error == 0) {
3817 		ip = xip;
3818 		*vpp = ITOV(ip);
3819 	} else if (error == EEXIST)
3820 		VN_RELE(ITOV(xip));
3821 
3822 	if (ulp) {
3823 		int terr = 0;
3824 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size);
3825 		ufs_lockfs_end(ulp);
3826 		if (error == 0)
3827 			error = terr;
3828 	}
3829 out:
3830 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3831 		ufs_delete_drain_wait(ufsvfsp, 1);
3832 		retry = 0;
3833 		goto again;
3834 	}
3835 
3836 	return (error);
3837 }
3838 
3839 /*ARGSUSED*/
3840 static int
3841 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr,
3842 	caller_context_t *ct, int flags)
3843 {
3844 	struct inode *ip = VTOI(vp);
3845 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3846 	struct ulockfs *ulp;
3847 	vnode_t *rmvp = NULL;	/* Vnode of removed directory */
3848 	int error;
3849 	int issync;
3850 	int trans_size;
3851 	int indeadlock;
3852 
3853 	/*
3854 	 * don't let the delete queue get too long
3855 	 */
3856 	if (ufsvfsp == NULL) {
3857 		error = EIO;
3858 		goto out;
3859 	}
3860 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3861 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3862 
3863 	error = ufs_eventlookup(vp, nm, cr, &rmvp);
3864 	if (rmvp != NULL) {
3865 		/* Only send the event if there were no errors */
3866 		if (error == 0)
3867 			vnevent_rmdir(rmvp, vp, nm, ct);
3868 		VN_RELE(rmvp);
3869 	}
3870 
3871 retry_rmdir:
3872 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK);
3873 	if (error)
3874 		goto out;
3875 
3876 	if (ulp)
3877 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR,
3878 		    trans_size = TOP_RMDIR_SIZE);
3879 
3880 	/*
3881 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3882 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3883 	 * possible, retries the operation.
3884 	 */
3885 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_RMDIR, retry);
3886 	if (indeadlock)
3887 		goto retry_rmdir;
3888 	error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr);
3889 
3890 	rw_exit(&ip->i_rwlock);
3891 
3892 	if (ulp) {
3893 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR,
3894 		    trans_size);
3895 		ufs_lockfs_end(ulp);
3896 	}
3897 
3898 out:
3899 	return (error);
3900 }
3901 
3902 /* ARGSUSED */
3903 static int
3904 ufs_readdir(
3905 	struct vnode *vp,
3906 	struct uio *uiop,
3907 	struct cred *cr,
3908 	int *eofp,
3909 	caller_context_t *ct,
3910 	int flags)
3911 {
3912 	struct iovec *iovp;
3913 	struct inode *ip;
3914 	struct direct *idp;
3915 	struct dirent64 *odp;
3916 	struct fbuf *fbp;
3917 	struct ufsvfs *ufsvfsp;
3918 	struct ulockfs *ulp;
3919 	caddr_t outbuf;
3920 	size_t bufsize;
3921 	uint_t offset;
3922 	uint_t bytes_wanted, total_bytes_wanted;
3923 	int incount = 0;
3924 	int outcount = 0;
3925 	int error;
3926 
3927 	ip = VTOI(vp);
3928 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
3929 
3930 	if (uiop->uio_loffset >= MAXOFF32_T) {
3931 		if (eofp)
3932 			*eofp = 1;
3933 		return (0);
3934 	}
3935 
3936 	/*
3937 	 * Check if we have been called with a valid iov_len
3938 	 * and bail out if not, otherwise we may potentially loop
3939 	 * forever further down.
3940 	 */
3941 	if (uiop->uio_iov->iov_len <= 0) {
3942 		error = EINVAL;
3943 		goto out;
3944 	}
3945 
3946 	/*
3947 	 * Large Files: When we come here we are guaranteed that
3948 	 * uio_offset can be used safely. The high word is zero.
3949 	 */
3950 
3951 	ufsvfsp = ip->i_ufsvfs;
3952 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READDIR_MASK);
3953 	if (error)
3954 		goto out;
3955 
3956 	iovp = uiop->uio_iov;
3957 	total_bytes_wanted = iovp->iov_len;
3958 
3959 	/* Large Files: directory files should not be "large" */
3960 
3961 	ASSERT(ip->i_size <= MAXOFF32_T);
3962 
3963 	/* Force offset to be valid (to guard against bogus lseek() values) */
3964 	offset = (uint_t)uiop->uio_offset & ~(DIRBLKSIZ - 1);
3965 
3966 	/* Quit if at end of file or link count of zero (posix) */
3967 	if (offset >= (uint_t)ip->i_size || ip->i_nlink <= 0) {
3968 		if (eofp)
3969 			*eofp = 1;
3970 		error = 0;
3971 		goto unlock;
3972 	}
3973 
3974 	/*
3975 	 * Get space to change directory entries into fs independent format.
3976 	 * Do fast alloc for the most commonly used-request size (filesystem
3977 	 * block size).
3978 	 */
3979 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) {
3980 		bufsize = total_bytes_wanted;
3981 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
3982 		odp = (struct dirent64 *)outbuf;
3983 	} else {
3984 		bufsize = total_bytes_wanted;
3985 		odp = (struct dirent64 *)iovp->iov_base;
3986 	}
3987 
3988 nextblk:
3989 	bytes_wanted = total_bytes_wanted;
3990 
3991 	/* Truncate request to file size */
3992 	if (offset + bytes_wanted > (int)ip->i_size)
3993 		bytes_wanted = (int)(ip->i_size - offset);
3994 
3995 	/* Comply with MAXBSIZE boundary restrictions of fbread() */
3996 	if ((offset & MAXBOFFSET) + bytes_wanted > MAXBSIZE)
3997 		bytes_wanted = MAXBSIZE - (offset & MAXBOFFSET);
3998 
3999 	/*
4000 	 * Read in the next chunk.
4001 	 * We are still holding the i_rwlock.
4002 	 */
4003 	error = fbread(vp, (offset_t)offset, bytes_wanted, S_OTHER, &fbp);
4004 
4005 	if (error)
4006 		goto update_inode;
4007 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (ip->i_fs->fs_ronly == 0) &&
4008 	    (!ufsvfsp->vfs_noatime)) {
4009 		ip->i_flag |= IACC;
4010 	}
4011 	incount = 0;
4012 	idp = (struct direct *)fbp->fb_addr;
4013 	if (idp->d_ino == 0 && idp->d_reclen == 0 && idp->d_namlen == 0) {
4014 		cmn_err(CE_WARN, "ufs_readdir: bad dir, inumber = %llu, "
4015 		    "fs = %s\n",
4016 		    (u_longlong_t)ip->i_number, ufsvfsp->vfs_fs->fs_fsmnt);
4017 		fbrelse(fbp, S_OTHER);
4018 		error = ENXIO;
4019 		goto update_inode;
4020 	}
4021 	/* Transform to file-system independent format */
4022 	while (incount < bytes_wanted) {
4023 		/*
4024 		 * If the current directory entry is mangled, then skip
4025 		 * to the next block.  It would be nice to set the FSBAD
4026 		 * flag in the super-block so that a fsck is forced on
4027 		 * next reboot, but locking is a problem.
4028 		 */
4029 		if (idp->d_reclen & 0x3) {
4030 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4031 			break;
4032 		}
4033 
4034 		/* Skip to requested offset and skip empty entries */
4035 		if (idp->d_ino != 0 && offset >= (uint_t)uiop->uio_offset) {
4036 			ushort_t this_reclen =
4037 			    DIRENT64_RECLEN(idp->d_namlen);
4038 			/* Buffer too small for any entries */
4039 			if (!outcount && this_reclen > bufsize) {
4040 				fbrelse(fbp, S_OTHER);
4041 				error = EINVAL;
4042 				goto update_inode;
4043 			}
4044 			/* If would overrun the buffer, quit */
4045 			if (outcount + this_reclen > bufsize) {
4046 				break;
4047 			}
4048 			/* Take this entry */
4049 			odp->d_ino = (ino64_t)idp->d_ino;
4050 			odp->d_reclen = (ushort_t)this_reclen;
4051 			odp->d_off = (offset_t)(offset + idp->d_reclen);
4052 
4053 			/* use strncpy(9f) to zero out uninitialized bytes */
4054 
4055 			ASSERT(strlen(idp->d_name) + 1 <=
4056 			    DIRENT64_NAMELEN(this_reclen));
4057 			(void) strncpy(odp->d_name, idp->d_name,
4058 			    DIRENT64_NAMELEN(this_reclen));
4059 			outcount += odp->d_reclen;
4060 			odp = (struct dirent64 *)
4061 			    ((intptr_t)odp + odp->d_reclen);
4062 			ASSERT(outcount <= bufsize);
4063 		}
4064 		if (idp->d_reclen) {
4065 			incount += idp->d_reclen;
4066 			offset += idp->d_reclen;
4067 			idp = (struct direct *)((intptr_t)idp + idp->d_reclen);
4068 		} else {
4069 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4070 			break;
4071 		}
4072 	}
4073 	/* Release the chunk */
4074 	fbrelse(fbp, S_OTHER);
4075 
4076 	/* Read whole block, but got no entries, read another if not eof */
4077 
4078 	/*
4079 	 * Large Files: casting i_size to int here is not a problem
4080 	 * because directory sizes are always less than MAXOFF32_T.
4081 	 * See assertion above.
4082 	 */
4083 
4084 	if (offset < (int)ip->i_size && !outcount)
4085 		goto nextblk;
4086 
4087 	/* Copy out the entry data */
4088 	if (uiop->uio_segflg == UIO_SYSSPACE && uiop->uio_iovcnt == 1) {
4089 		iovp->iov_base += outcount;
4090 		iovp->iov_len -= outcount;
4091 		uiop->uio_resid -= outcount;
4092 		uiop->uio_offset = offset;
4093 	} else if ((error = uiomove(outbuf, (long)outcount, UIO_READ,
4094 	    uiop)) == 0)
4095 		uiop->uio_offset = offset;
4096 update_inode:
4097 	ITIMES(ip);
4098 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1)
4099 		kmem_free(outbuf, bufsize);
4100 
4101 	if (eofp && error == 0)
4102 		*eofp = (uiop->uio_offset >= (int)ip->i_size);
4103 unlock:
4104 	if (ulp) {
4105 		ufs_lockfs_end(ulp);
4106 	}
4107 out:
4108 	return (error);
4109 }
4110 
4111 /*ARGSUSED*/
4112 static int
4113 ufs_symlink(
4114 	struct vnode *dvp,		/* ptr to parent dir vnode */
4115 	char *linkname,			/* name of symbolic link */
4116 	struct vattr *vap,		/* attributes */
4117 	char *target,			/* target path */
4118 	struct cred *cr,		/* user credentials */
4119 	caller_context_t *ct,
4120 	int flags)
4121 {
4122 	struct inode *ip, *dip = VTOI(dvp);
4123 	struct ufsvfs *ufsvfsp = dip->i_ufsvfs;
4124 	struct ulockfs *ulp;
4125 	int error;
4126 	int issync;
4127 	int trans_size;
4128 	int residual;
4129 	int ioflag;
4130 	int retry = 1;
4131 
4132 	/*
4133 	 * No symlinks in attrdirs at this time
4134 	 */
4135 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
4136 		return (EINVAL);
4137 
4138 again:
4139 	ip = (struct inode *)NULL;
4140 	vap->va_type = VLNK;
4141 	vap->va_rdev = 0;
4142 
4143 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK);
4144 	if (error)
4145 		goto out;
4146 
4147 	if (ulp)
4148 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK,
4149 		    trans_size = (int)TOP_SYMLINK_SIZE(dip));
4150 
4151 	/*
4152 	 * We must create the inode before the directory entry, to avoid
4153 	 * racing with readlink().  ufs_dirmakeinode requires that we
4154 	 * hold the quota lock as reader, and directory locks as writer.
4155 	 */
4156 
4157 	rw_enter(&dip->i_rwlock, RW_WRITER);
4158 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4159 	rw_enter(&dip->i_contents, RW_WRITER);
4160 
4161 	/*
4162 	 * Suppress any out of inodes messages if we will retry on
4163 	 * ENOSP
4164 	 */
4165 	if (retry)
4166 		dip->i_flag |= IQUIET;
4167 
4168 	error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr);
4169 
4170 	dip->i_flag &= ~IQUIET;
4171 
4172 	rw_exit(&dip->i_contents);
4173 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4174 	rw_exit(&dip->i_rwlock);
4175 
4176 	if (error)
4177 		goto unlock;
4178 
4179 	/*
4180 	 * OK.  The inode has been created.  Write out the data of the
4181 	 * symbolic link.  Since symbolic links are metadata, and should
4182 	 * remain consistent across a system crash, we need to force the
4183 	 * data out synchronously.
4184 	 *
4185 	 * (This is a change from the semantics in earlier releases, which
4186 	 * only created symbolic links synchronously if the semi-documented
4187 	 * 'syncdir' option was set, or if we were being invoked by the NFS
4188 	 * server, which requires symbolic links to be created synchronously.)
4189 	 *
4190 	 * We need to pass in a pointer for the residual length; otherwise
4191 	 * ufs_rdwri() will always return EIO if it can't write the data,
4192 	 * even if the error was really ENOSPC or EDQUOT.
4193 	 */
4194 
4195 	ioflag = FWRITE | FDSYNC;
4196 	residual = 0;
4197 
4198 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4199 	rw_enter(&ip->i_contents, RW_WRITER);
4200 
4201 	/*
4202 	 * Suppress file system full messages if we will retry
4203 	 */
4204 	if (retry)
4205 		ip->i_flag |= IQUIET;
4206 
4207 	error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target),
4208 	    (offset_t)0, UIO_SYSSPACE, &residual, cr);
4209 
4210 	ip->i_flag &= ~IQUIET;
4211 
4212 	if (error) {
4213 		rw_exit(&ip->i_contents);
4214 		rw_exit(&ufsvfsp->vfs_dqrwlock);
4215 		goto remove;
4216 	}
4217 
4218 	/*
4219 	 * If the link's data is small enough, we can cache it in the inode.
4220 	 * This is a "fast symbolic link".  We don't use the first direct
4221 	 * block because that's actually used to point at the symbolic link's
4222 	 * contents on disk; but we know that none of the other direct or
4223 	 * indirect blocks can be used because symbolic links are restricted
4224 	 * to be smaller than a file system block.
4225 	 */
4226 
4227 	ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip)));
4228 
4229 	if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) {
4230 		if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) {
4231 			ip->i_flag |= IFASTSYMLNK;
4232 		} else {
4233 			int i;
4234 			/* error, clear garbage left behind */
4235 			for (i = 1; i < NDADDR; i++)
4236 				ip->i_db[i] = 0;
4237 			for (i = 0; i < NIADDR; i++)
4238 				ip->i_ib[i] = 0;
4239 		}
4240 	}
4241 
4242 	rw_exit(&ip->i_contents);
4243 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4244 
4245 	/*
4246 	 * OK.  We've successfully created the symbolic link.  All that
4247 	 * remains is to insert it into the appropriate directory.
4248 	 */
4249 
4250 	rw_enter(&dip->i_rwlock, RW_WRITER);
4251 	error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr);
4252 	rw_exit(&dip->i_rwlock);
4253 
4254 	/*
4255 	 * Fall through into remove-on-error code.  We're either done, or we
4256 	 * need to remove the inode (if we couldn't insert it).
4257 	 */
4258 
4259 remove:
4260 	if (error && (ip != NULL)) {
4261 		rw_enter(&ip->i_contents, RW_WRITER);
4262 		ip->i_nlink--;
4263 		ip->i_flag |= ICHG;
4264 		ip->i_seq++;
4265 		ufs_setreclaim(ip);
4266 		rw_exit(&ip->i_contents);
4267 	}
4268 
4269 unlock:
4270 	if (ip != NULL)
4271 		VN_RELE(ITOV(ip));
4272 
4273 	if (ulp) {
4274 		int terr = 0;
4275 
4276 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK,
4277 		    trans_size);
4278 		ufs_lockfs_end(ulp);
4279 		if (error == 0)
4280 			error = terr;
4281 	}
4282 
4283 	/*
4284 	 * We may have failed due to lack of an inode or of a block to
4285 	 * store the target in.  Try flushing the delete queue to free
4286 	 * logically-available things up and try again.
4287 	 */
4288 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
4289 		ufs_delete_drain_wait(ufsvfsp, 1);
4290 		retry = 0;
4291 		goto again;
4292 	}
4293 
4294 out:
4295 	return (error);
4296 }
4297 
4298 /*
4299  * Ufs specific routine used to do ufs io.
4300  */
4301 int
4302 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base,
4303 	ssize_t len, offset_t offset, enum uio_seg seg, int *aresid,
4304 	struct cred *cr)
4305 {
4306 	struct uio auio;
4307 	struct iovec aiov;
4308 	int error;
4309 
4310 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
4311 
4312 	bzero((caddr_t)&auio, sizeof (uio_t));
4313 	bzero((caddr_t)&aiov, sizeof (iovec_t));
4314 
4315 	aiov.iov_base = base;
4316 	aiov.iov_len = len;
4317 	auio.uio_iov = &aiov;
4318 	auio.uio_iovcnt = 1;
4319 	auio.uio_loffset = offset;
4320 	auio.uio_segflg = (short)seg;
4321 	auio.uio_resid = len;
4322 
4323 	if (rw == UIO_WRITE) {
4324 		auio.uio_fmode = FWRITE;
4325 		auio.uio_extflg = UIO_COPY_DEFAULT;
4326 		auio.uio_llimit = curproc->p_fsz_ctl;
4327 		error = wrip(ip, &auio, ioflag, cr);
4328 	} else {
4329 		auio.uio_fmode = FREAD;
4330 		auio.uio_extflg = UIO_COPY_CACHED;
4331 		auio.uio_llimit = MAXOFFSET_T;
4332 		error = rdip(ip, &auio, ioflag, cr);
4333 	}
4334 
4335 	if (aresid) {
4336 		*aresid = auio.uio_resid;
4337 	} else if (auio.uio_resid) {
4338 		error = EIO;
4339 	}
4340 	return (error);
4341 }
4342 
4343 /*ARGSUSED*/
4344 static int
4345 ufs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
4346 {
4347 	struct ufid *ufid;
4348 	struct inode *ip = VTOI(vp);
4349 
4350 	if (ip->i_ufsvfs == NULL)
4351 		return (EIO);
4352 
4353 	if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) {
4354 		fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t);
4355 		return (ENOSPC);
4356 	}
4357 
4358 	ufid = (struct ufid *)fidp;
4359 	bzero((char *)ufid, sizeof (struct ufid));
4360 	ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t);
4361 	ufid->ufid_ino = ip->i_number;
4362 	ufid->ufid_gen = ip->i_gen;
4363 
4364 	return (0);
4365 }
4366 
4367 /* ARGSUSED2 */
4368 static int
4369 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4370 {
4371 	struct inode	*ip = VTOI(vp);
4372 	struct ufsvfs	*ufsvfsp;
4373 	int		forcedirectio;
4374 
4375 	/*
4376 	 * Read case is easy.
4377 	 */
4378 	if (!write_lock) {
4379 		rw_enter(&ip->i_rwlock, RW_READER);
4380 		return (V_WRITELOCK_FALSE);
4381 	}
4382 
4383 	/*
4384 	 * Caller has requested a writer lock, but that inhibits any
4385 	 * concurrency in the VOPs that follow. Acquire the lock shared
4386 	 * and defer exclusive access until it is known to be needed in
4387 	 * other VOP handlers. Some cases can be determined here.
4388 	 */
4389 
4390 	/*
4391 	 * If directio is not set, there is no chance of concurrency,
4392 	 * so just acquire the lock exclusive. Beware of a forced
4393 	 * unmount before looking at the mount option.
4394 	 */
4395 	ufsvfsp = ip->i_ufsvfs;
4396 	forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0;
4397 	if (!(ip->i_flag & IDIRECTIO || forcedirectio) ||
4398 	    !ufs_allow_shared_writes) {
4399 		rw_enter(&ip->i_rwlock, RW_WRITER);
4400 		return (V_WRITELOCK_TRUE);
4401 	}
4402 
4403 	/*
4404 	 * Mandatory locking forces acquiring i_rwlock exclusive.
4405 	 */
4406 	if (MANDLOCK(vp, ip->i_mode)) {
4407 		rw_enter(&ip->i_rwlock, RW_WRITER);
4408 		return (V_WRITELOCK_TRUE);
4409 	}
4410 
4411 	/*
4412 	 * Acquire the lock shared in case a concurrent write follows.
4413 	 * Mandatory locking could have become enabled before the lock
4414 	 * was acquired. Re-check and upgrade if needed.
4415 	 */
4416 	rw_enter(&ip->i_rwlock, RW_READER);
4417 	if (MANDLOCK(vp, ip->i_mode)) {
4418 		rw_exit(&ip->i_rwlock);
4419 		rw_enter(&ip->i_rwlock, RW_WRITER);
4420 		return (V_WRITELOCK_TRUE);
4421 	}
4422 	return (V_WRITELOCK_FALSE);
4423 }
4424 
4425 /*ARGSUSED*/
4426 static void
4427 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4428 {
4429 	struct inode	*ip = VTOI(vp);
4430 
4431 	rw_exit(&ip->i_rwlock);
4432 }
4433 
4434 /* ARGSUSED */
4435 static int
4436 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp,
4437 	caller_context_t *ct)
4438 {
4439 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4440 }
4441 
4442 /* ARGSUSED */
4443 static int
4444 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4445 	offset_t offset, struct flk_callback *flk_cbp, struct cred *cr,
4446 	caller_context_t *ct)
4447 {
4448 	struct inode *ip = VTOI(vp);
4449 
4450 	if (ip->i_ufsvfs == NULL)
4451 		return (EIO);
4452 
4453 	/*
4454 	 * If file is being mapped, disallow frlock.
4455 	 * XXX I am not holding tlock while checking i_mapcnt because the
4456 	 * current locking strategy drops all locks before calling fs_frlock.
4457 	 * So, mapcnt could change before we enter fs_frlock making is
4458 	 * meaningless to have held tlock in the first place.
4459 	 */
4460 	if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode))
4461 		return (EAGAIN);
4462 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4463 }
4464 
4465 /* ARGSUSED */
4466 static int
4467 ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4468 	offset_t offset, cred_t *cr, caller_context_t *ct)
4469 {
4470 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
4471 	struct ulockfs *ulp;
4472 	int error;
4473 
4474 	if ((error = convoff(vp, bfp, 0, offset)) == 0) {
4475 		if (cmd == F_FREESP) {
4476 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4477 			    ULOCKFS_SPACE_MASK);
4478 			if (error)
4479 				return (error);
4480 			error = ufs_freesp(vp, bfp, flag, cr);
4481 
4482 			if (error == 0 && bfp->l_start == 0)
4483 				vnevent_truncate(vp, ct);
4484 		} else if (cmd == F_ALLOCSP) {
4485 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4486 			    ULOCKFS_FALLOCATE_MASK);
4487 			if (error)
4488 				return (error);
4489 			error = ufs_allocsp(vp, bfp, cr);
4490 		} else
4491 			return (EINVAL); /* Command not handled here */
4492 
4493 		if (ulp)
4494 			ufs_lockfs_end(ulp);
4495 
4496 	}
4497 	return (error);
4498 }
4499 
4500 /*
4501  * Used to determine if read ahead should be done. Also used to
4502  * to determine when write back occurs.
4503  */
4504 #define	CLUSTSZ(ip)		((ip)->i_ufsvfs->vfs_ioclustsz)
4505 
4506 /*
4507  * A faster version of ufs_getpage.
4508  *
4509  * We optimize by inlining the pvn_getpages iterator, eliminating
4510  * calls to bmap_read if file doesn't have UFS holes, and avoiding
4511  * the overhead of page_exists().
4512  *
4513  * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4514  * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4515  * victimizes performance when a file with UFS holes is faulted
4516  * first in the S_READ mode, and then in the S_WRITE mode. We will get
4517  * two MMU faults in this case.
4518  *
4519  * XXX - the inode fields which control the sequential mode are not
4520  *	 protected by any mutex. The read ahead will act wild if
4521  *	 multiple processes will access the file concurrently and
4522  *	 some of them in sequential mode. One particulary bad case
4523  *	 is if another thread will change the value of i_nextrio between
4524  *	 the time this thread tests the i_nextrio value and then reads it
4525  *	 again to use it as the offset for the read ahead.
4526  */
4527 /*ARGSUSED*/
4528 static int
4529 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
4530 	page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr,
4531 	enum seg_rw rw, struct cred *cr, caller_context_t *ct)
4532 {
4533 	u_offset_t	uoff = (u_offset_t)off; /* type conversion */
4534 	u_offset_t	pgoff;
4535 	u_offset_t	eoff;
4536 	struct inode 	*ip = VTOI(vp);
4537 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
4538 	struct fs 	*fs;
4539 	struct ulockfs	*ulp;
4540 	page_t		**pl;
4541 	caddr_t		pgaddr;
4542 	krw_t		rwtype;
4543 	int 		err;
4544 	int		has_holes;
4545 	int		beyond_eof;
4546 	int		seqmode;
4547 	int		pgsize = PAGESIZE;
4548 	int		dolock;
4549 	int		do_qlock;
4550 	int		trans_size;
4551 
4552 	ASSERT((uoff & PAGEOFFSET) == 0);
4553 
4554 	if (protp)
4555 		*protp = PROT_ALL;
4556 
4557 	/*
4558 	 * Obey the lockfs protocol
4559 	 */
4560 	err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg,
4561 	    rw == S_READ || rw == S_EXEC, protp);
4562 	if (err)
4563 		goto out;
4564 
4565 	fs = ufsvfsp->vfs_fs;
4566 
4567 	if (ulp && (rw == S_CREATE || rw == S_WRITE) &&
4568 	    !(vp->v_flag & VISSWAP)) {
4569 		/*
4570 		 * Try to start a transaction, will return if blocking is
4571 		 * expected to occur and the address space is not the
4572 		 * kernel address space.
4573 		 */
4574 		trans_size = TOP_GETPAGE_SIZE(ip);
4575 		if (seg->s_as != &kas) {
4576 			TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE,
4577 			    trans_size, err)
4578 			if (err == EWOULDBLOCK) {
4579 				/*
4580 				 * Use EDEADLK here because the VM code
4581 				 * can normally never see this error.
4582 				 */
4583 				err = EDEADLK;
4584 				ufs_lockfs_end(ulp);
4585 				goto out;
4586 			}
4587 		} else {
4588 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4589 		}
4590 	}
4591 
4592 	if (vp->v_flag & VNOMAP) {
4593 		err = ENOSYS;
4594 		goto unlock;
4595 	}
4596 
4597 	seqmode = ip->i_nextr == uoff && rw != S_CREATE;
4598 
4599 	rwtype = RW_READER;		/* start as a reader */
4600 	dolock = (rw_owner(&ip->i_contents) != curthread);
4601 	/*
4602 	 * If this thread owns the lock, i.e., this thread grabbed it
4603 	 * as writer somewhere above, then we don't need to grab the
4604 	 * lock as reader in this routine.
4605 	 */
4606 	do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread);
4607 
4608 retrylock:
4609 	if (dolock) {
4610 		/*
4611 		 * Grab the quota lock if we need to call
4612 		 * bmap_write() below (with i_contents as writer).
4613 		 */
4614 		if (do_qlock && rwtype == RW_WRITER)
4615 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4616 		rw_enter(&ip->i_contents, rwtype);
4617 	}
4618 
4619 	/*
4620 	 * We may be getting called as a side effect of a bmap using
4621 	 * fbread() when the blocks might be being allocated and the
4622 	 * size has not yet been up'ed.  In this case we want to be
4623 	 * able to return zero pages if we get back UFS_HOLE from
4624 	 * calling bmap for a non write case here.  We also might have
4625 	 * to read some frags from the disk into a page if we are
4626 	 * extending the number of frags for a given lbn in bmap().
4627 	 * Large Files: The read of i_size here is atomic because
4628 	 * i_contents is held here. If dolock is zero, the lock
4629 	 * is held in bmap routines.
4630 	 */
4631 	beyond_eof = uoff + len >
4632 	    P2ROUNDUP_TYPED(ip->i_size, PAGESIZE, u_offset_t);
4633 	if (beyond_eof && seg != segkmap) {
4634 		if (dolock) {
4635 			rw_exit(&ip->i_contents);
4636 			if (do_qlock && rwtype == RW_WRITER)
4637 				rw_exit(&ufsvfsp->vfs_dqrwlock);
4638 		}
4639 		err = EFAULT;
4640 		goto unlock;
4641 	}
4642 
4643 	/*
4644 	 * Must hold i_contents lock throughout the call to pvn_getpages
4645 	 * since locked pages are returned from each call to ufs_getapage.
4646 	 * Must *not* return locked pages and then try for contents lock
4647 	 * due to lock ordering requirements (inode > page)
4648 	 */
4649 
4650 	has_holes = bmap_has_holes(ip);
4651 
4652 	if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) {
4653 		int	blk_size;
4654 		u_offset_t offset;
4655 
4656 		/*
4657 		 * We must acquire the RW_WRITER lock in order to
4658 		 * call bmap_write().
4659 		 */
4660 		if (dolock && rwtype == RW_READER) {
4661 			rwtype = RW_WRITER;
4662 
4663 			/*
4664 			 * Grab the quota lock before
4665 			 * upgrading i_contents, but if we can't grab it
4666 			 * don't wait here due to lock order:
4667 			 * vfs_dqrwlock > i_contents.
4668 			 */
4669 			if (do_qlock &&
4670 			    rw_tryenter(&ufsvfsp->vfs_dqrwlock, RW_READER)
4671 			    == 0) {
4672 				rw_exit(&ip->i_contents);
4673 				goto retrylock;
4674 			}
4675 			if (!rw_tryupgrade(&ip->i_contents)) {
4676 				rw_exit(&ip->i_contents);
4677 				if (do_qlock)
4678 					rw_exit(&ufsvfsp->vfs_dqrwlock);
4679 				goto retrylock;
4680 			}
4681 		}
4682 
4683 		/*
4684 		 * May be allocating disk blocks for holes here as
4685 		 * a result of mmap faults. write(2) does the bmap_write
4686 		 * in rdip/wrip, not here. We are not dealing with frags
4687 		 * in this case.
4688 		 */
4689 		/*
4690 		 * Large Files: We cast fs_bmask field to offset_t
4691 		 * just as we do for MAXBMASK because uoff is a 64-bit
4692 		 * data type. fs_bmask will still be a 32-bit type
4693 		 * as we cannot change any ondisk data structures.
4694 		 */
4695 
4696 		offset = uoff & (offset_t)fs->fs_bmask;
4697 		while (offset < uoff + len) {
4698 			blk_size = (int)blksize(fs, ip, lblkno(fs, offset));
4699 			err = bmap_write(ip, offset, blk_size,
4700 			    BI_NORMAL, NULL, cr);
4701 			if (ip->i_flag & (ICHG|IUPD))
4702 				ip->i_seq++;
4703 			if (err)
4704 				goto update_inode;
4705 			offset += blk_size; /* XXX - make this contig */
4706 		}
4707 	}
4708 
4709 	/*
4710 	 * Can be a reader from now on.
4711 	 */
4712 	if (dolock && rwtype == RW_WRITER) {
4713 		rw_downgrade(&ip->i_contents);
4714 		/*
4715 		 * We can release vfs_dqrwlock early so do it, but make
4716 		 * sure we don't try to release it again at the bottom.
4717 		 */
4718 		if (do_qlock) {
4719 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4720 			do_qlock = 0;
4721 		}
4722 	}
4723 
4724 	/*
4725 	 * We remove PROT_WRITE in cases when the file has UFS holes
4726 	 * because we don't  want to call bmap_read() to check each
4727 	 * page if it is backed with a disk block.
4728 	 */
4729 	if (protp && has_holes && rw != S_WRITE && rw != S_CREATE)
4730 		*protp &= ~PROT_WRITE;
4731 
4732 	err = 0;
4733 
4734 	/*
4735 	 * The loop looks up pages in the range [off, off + len).
4736 	 * For each page, we first check if we should initiate an asynchronous
4737 	 * read ahead before we call page_lookup (we may sleep in page_lookup
4738 	 * for a previously initiated disk read).
4739 	 */
4740 	eoff = (uoff + len);
4741 	for (pgoff = uoff, pgaddr = addr, pl = plarr;
4742 	    pgoff < eoff; /* empty */) {
4743 		page_t	*pp;
4744 		u_offset_t	nextrio;
4745 		se_t	se;
4746 		int retval;
4747 
4748 		se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED);
4749 
4750 		/* Handle async getpage (faultahead) */
4751 		if (plarr == NULL) {
4752 			ip->i_nextrio = pgoff;
4753 			(void) ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4754 			pgoff += pgsize;
4755 			pgaddr += pgsize;
4756 			continue;
4757 		}
4758 		/*
4759 		 * Check if we should initiate read ahead of next cluster.
4760 		 * We call page_exists only when we need to confirm that
4761 		 * we have the current page before we initiate the read ahead.
4762 		 */
4763 		nextrio = ip->i_nextrio;
4764 		if (seqmode &&
4765 		    pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
4766 		    nextrio < ip->i_size && page_exists(vp, pgoff)) {
4767 			retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4768 			/*
4769 			 * We always read ahead the next cluster of data
4770 			 * starting from i_nextrio. If the page (vp,nextrio)
4771 			 * is actually in core at this point, the routine
4772 			 * ufs_getpage_ra() will stop pre-fetching data
4773 			 * until we read that page in a synchronized manner
4774 			 * through ufs_getpage_miss(). So, we should increase
4775 			 * i_nextrio if the page (vp, nextrio) exists.
4776 			 */
4777 			if ((retval == 0) && page_exists(vp, nextrio)) {
4778 				ip->i_nextrio = nextrio + pgsize;
4779 			}
4780 		}
4781 
4782 		if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
4783 			/*
4784 			 * We found the page in the page cache.
4785 			 */
4786 			*pl++ = pp;
4787 			pgoff += pgsize;
4788 			pgaddr += pgsize;
4789 			len -= pgsize;
4790 			plsz -= pgsize;
4791 		} else  {
4792 			/*
4793 			 * We have to create the page, or read it from disk.
4794 			 */
4795 			if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr,
4796 			    pl, plsz, rw, seqmode))
4797 				goto error;
4798 
4799 			while (*pl != NULL) {
4800 				pl++;
4801 				pgoff += pgsize;
4802 				pgaddr += pgsize;
4803 				len -= pgsize;
4804 				plsz -= pgsize;
4805 			}
4806 		}
4807 	}
4808 
4809 	/*
4810 	 * Return pages up to plsz if they are in the page cache.
4811 	 * We cannot return pages if there is a chance that they are
4812 	 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4813 	 */
4814 	if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
4815 
4816 		ASSERT((protp == NULL) ||
4817 		    !(has_holes && (*protp & PROT_WRITE)));
4818 
4819 		eoff = pgoff + plsz;
4820 		while (pgoff < eoff) {
4821 			page_t		*pp;
4822 
4823 			if ((pp = page_lookup_nowait(vp, pgoff,
4824 			    SE_SHARED)) == NULL)
4825 				break;
4826 
4827 			*pl++ = pp;
4828 			pgoff += pgsize;
4829 			plsz -= pgsize;
4830 		}
4831 	}
4832 
4833 	if (plarr)
4834 		*pl = NULL;			/* Terminate page list */
4835 	ip->i_nextr = pgoff;
4836 
4837 error:
4838 	if (err && plarr) {
4839 		/*
4840 		 * Release any pages we have locked.
4841 		 */
4842 		while (pl > &plarr[0])
4843 			page_unlock(*--pl);
4844 
4845 		plarr[0] = NULL;
4846 	}
4847 
4848 update_inode:
4849 	/*
4850 	 * If the inode is not already marked for IACC (in rdip() for read)
4851 	 * and the inode is not marked for no access time update (in wrip()
4852 	 * for write) then update the inode access time and mod time now.
4853 	 */
4854 	if ((ip->i_flag & (IACC | INOACC)) == 0) {
4855 		if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) {
4856 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
4857 			    (fs->fs_ronly == 0) &&
4858 			    (!ufsvfsp->vfs_noatime)) {
4859 				mutex_enter(&ip->i_tlock);
4860 				ip->i_flag |= IACC;
4861 				ITIMES_NOLOCK(ip);
4862 				mutex_exit(&ip->i_tlock);
4863 			}
4864 		}
4865 	}
4866 
4867 	if (dolock) {
4868 		rw_exit(&ip->i_contents);
4869 		if (do_qlock && rwtype == RW_WRITER)
4870 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4871 	}
4872 
4873 unlock:
4874 	if (ulp) {
4875 		if ((rw == S_CREATE || rw == S_WRITE) &&
4876 		    !(vp->v_flag & VISSWAP)) {
4877 			TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4878 		}
4879 		ufs_lockfs_end(ulp);
4880 	}
4881 out:
4882 	return (err);
4883 }
4884 
4885 /*
4886  * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4887  * cache. The page is either read from the disk, or it's created.
4888  * A page is created (without disk read) if rw == S_CREATE, or if
4889  * the page is not backed with a real disk block (UFS hole).
4890  */
4891 /* ARGSUSED */
4892 static int
4893 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg,
4894 	caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq)
4895 {
4896 	struct inode	*ip = VTOI(vp);
4897 	page_t		*pp;
4898 	daddr_t		bn;
4899 	size_t		io_len;
4900 	int		crpage = 0;
4901 	int		err;
4902 	int		contig;
4903 	int		bsize = ip->i_fs->fs_bsize;
4904 
4905 	/*
4906 	 * Figure out whether the page can be created, or must be
4907 	 * must be read from the disk.
4908 	 */
4909 	if (rw == S_CREATE)
4910 		crpage = 1;
4911 	else {
4912 		contig = 0;
4913 		if (err = bmap_read(ip, off, &bn, &contig))
4914 			return (err);
4915 
4916 		crpage = (bn == UFS_HOLE);
4917 
4918 		/*
4919 		 * If its also a fallocated block that hasn't been written to
4920 		 * yet, we will treat it just like a UFS_HOLE and create
4921 		 * a zero page for it
4922 		 */
4923 		if (ISFALLOCBLK(ip, bn))
4924 			crpage = 1;
4925 	}
4926 
4927 	if (crpage) {
4928 		if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg,
4929 		    addr)) == NULL) {
4930 			return (ufs_fault(vp,
4931 			    "ufs_getpage_miss: page_create == NULL"));
4932 		}
4933 
4934 		if (rw != S_CREATE)
4935 			pagezero(pp, 0, PAGESIZE);
4936 
4937 		io_len = PAGESIZE;
4938 	} else {
4939 		u_offset_t	io_off;
4940 		uint_t	xlen;
4941 		struct buf	*bp;
4942 		ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
4943 
4944 		/*
4945 		 * If access is not in sequential order, we read from disk
4946 		 * in bsize units.
4947 		 *
4948 		 * We limit the size of the transfer to bsize if we are reading
4949 		 * from the beginning of the file. Note in this situation we
4950 		 * will hedge our bets and initiate an async read ahead of
4951 		 * the second block.
4952 		 */
4953 		if (!seq || off == 0)
4954 			contig = MIN(contig, bsize);
4955 
4956 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4957 		    &io_len, off, contig, 0);
4958 
4959 		/*
4960 		 * Some other thread has entered the page.
4961 		 * ufs_getpage will retry page_lookup.
4962 		 */
4963 		if (pp == NULL) {
4964 			pl[0] = NULL;
4965 			return (0);
4966 		}
4967 
4968 		/*
4969 		 * Zero part of the page which we are not
4970 		 * going to read from the disk.
4971 		 */
4972 		xlen = io_len & PAGEOFFSET;
4973 		if (xlen != 0)
4974 			pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4975 
4976 		bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ);
4977 		bp->b_edev = ip->i_dev;
4978 		bp->b_dev = cmpdev(ip->i_dev);
4979 		bp->b_blkno = bn;
4980 		bp->b_un.b_addr = (caddr_t)0;
4981 		bp->b_file = ip->i_vnode;
4982 		bp->b_offset = off;
4983 
4984 		if (ufsvfsp->vfs_log) {
4985 			lufs_read_strategy(ufsvfsp->vfs_log, bp);
4986 		} else if (ufsvfsp->vfs_snapshot) {
4987 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4988 		} else {
4989 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
4990 			ub.ub_getpages.value.ul++;
4991 			(void) bdev_strategy(bp);
4992 			lwp_stat_update(LWP_STAT_INBLK, 1);
4993 		}
4994 
4995 		ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK);
4996 
4997 		/*
4998 		 * If the file access is sequential, initiate read ahead
4999 		 * of the next cluster.
5000 		 */
5001 		if (seq && ip->i_nextrio < ip->i_size)
5002 			(void) ufs_getpage_ra(vp, off, seg, addr);
5003 		err = biowait(bp);
5004 		pageio_done(bp);
5005 
5006 		if (err) {
5007 			pvn_read_done(pp, B_ERROR);
5008 			return (err);
5009 		}
5010 	}
5011 
5012 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
5013 	return (0);
5014 }
5015 
5016 /*
5017  * Read ahead a cluster from the disk. Returns the length in bytes.
5018  */
5019 static int
5020 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr)
5021 {
5022 	struct inode	*ip = VTOI(vp);
5023 	page_t		*pp;
5024 	u_offset_t	io_off = ip->i_nextrio;
5025 	ufsvfs_t	*ufsvfsp;
5026 	caddr_t		addr2 = addr + (io_off - off);
5027 	struct buf	*bp;
5028 	daddr_t		bn;
5029 	size_t		io_len;
5030 	int		err;
5031 	int		contig;
5032 	int		xlen;
5033 	int		bsize = ip->i_fs->fs_bsize;
5034 
5035 	/*
5036 	 * If the directio advisory is in effect on this file,
5037 	 * then do not do buffered read ahead. Read ahead makes
5038 	 * it more difficult on threads using directio as they
5039 	 * will be forced to flush the pages from this vnode.
5040 	 */
5041 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5042 		return (0);
5043 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio)
5044 		return (0);
5045 
5046 	/*
5047 	 * Is this test needed?
5048 	 */
5049 	if (addr2 >= seg->s_base + seg->s_size)
5050 		return (0);
5051 
5052 	contig = 0;
5053 	err = bmap_read(ip, io_off, &bn, &contig);
5054 	/*
5055 	 * If its a UFS_HOLE or a fallocated block, do not perform
5056 	 * any read ahead's since there probably is nothing to read ahead
5057 	 */
5058 	if (err || bn == UFS_HOLE || ISFALLOCBLK(ip, bn))
5059 		return (0);
5060 
5061 	/*
5062 	 * Limit the transfer size to bsize if this is the 2nd block.
5063 	 */
5064 	if (io_off == (u_offset_t)bsize)
5065 		contig = MIN(contig, bsize);
5066 
5067 	if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off,
5068 	    &io_len, io_off, contig, 1)) == NULL)
5069 		return (0);
5070 
5071 	/*
5072 	 * Zero part of page which we are not going to read from disk
5073 	 */
5074 	if ((xlen = (io_len & PAGEOFFSET)) > 0)
5075 		pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
5076 
5077 	ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK;
5078 
5079 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC);
5080 	bp->b_edev = ip->i_dev;
5081 	bp->b_dev = cmpdev(ip->i_dev);
5082 	bp->b_blkno = bn;
5083 	bp->b_un.b_addr = (caddr_t)0;
5084 	bp->b_file = ip->i_vnode;
5085 	bp->b_offset = off;
5086 
5087 	if (ufsvfsp->vfs_log) {
5088 		lufs_read_strategy(ufsvfsp->vfs_log, bp);
5089 	} else if (ufsvfsp->vfs_snapshot) {
5090 		fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5091 	} else {
5092 		ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5093 		ub.ub_getras.value.ul++;
5094 		(void) bdev_strategy(bp);
5095 		lwp_stat_update(LWP_STAT_INBLK, 1);
5096 	}
5097 
5098 	return (io_len);
5099 }
5100 
5101 int	ufs_delay = 1;
5102 /*
5103  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5104  *
5105  * LMXXX - the inode really ought to contain a pointer to one of these
5106  * async args.  Stuff gunk in there and just hand the whole mess off.
5107  * This would replace i_delaylen, i_delayoff.
5108  */
5109 /*ARGSUSED*/
5110 static int
5111 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
5112 	struct cred *cr, caller_context_t *ct)
5113 {
5114 	struct inode *ip = VTOI(vp);
5115 	int err = 0;
5116 
5117 	if (vp->v_count == 0) {
5118 		return (ufs_fault(vp, "ufs_putpage: bad v_count == 0"));
5119 	}
5120 
5121 	/*
5122 	 * XXX - Why should this check be made here?
5123 	 */
5124 	if (vp->v_flag & VNOMAP) {
5125 		err = ENOSYS;
5126 		goto errout;
5127 	}
5128 
5129 	if (ip->i_ufsvfs == NULL) {
5130 		err = EIO;
5131 		goto errout;
5132 	}
5133 
5134 	if (flags & B_ASYNC) {
5135 		if (ufs_delay && len &&
5136 		    (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
5137 			mutex_enter(&ip->i_tlock);
5138 			/*
5139 			 * If nobody stalled, start a new cluster.
5140 			 */
5141 			if (ip->i_delaylen == 0) {
5142 				ip->i_delayoff = off;
5143 				ip->i_delaylen = len;
5144 				mutex_exit(&ip->i_tlock);
5145 				goto errout;
5146 			}
5147 			/*
5148 			 * If we have a full cluster or they are not contig,
5149 			 * then push last cluster and start over.
5150 			 */
5151 			if (ip->i_delaylen >= CLUSTSZ(ip) ||
5152 			    ip->i_delayoff + ip->i_delaylen != off) {
5153 				u_offset_t doff;
5154 				size_t dlen;
5155 
5156 				doff = ip->i_delayoff;
5157 				dlen = ip->i_delaylen;
5158 				ip->i_delayoff = off;
5159 				ip->i_delaylen = len;
5160 				mutex_exit(&ip->i_tlock);
5161 				err = ufs_putpages(vp, doff, dlen,
5162 				    flags, cr);
5163 				/* LMXXX - flags are new val, not old */
5164 				goto errout;
5165 			}
5166 			/*
5167 			 * There is something there, it's not full, and
5168 			 * it is contig.
5169 			 */
5170 			ip->i_delaylen += len;
5171 			mutex_exit(&ip->i_tlock);
5172 			goto errout;
5173 		}
5174 		/*
5175 		 * Must have weird flags or we are not clustering.
5176 		 */
5177 	}
5178 
5179 	err = ufs_putpages(vp, off, len, flags, cr);
5180 
5181 errout:
5182 	return (err);
5183 }
5184 
5185 /*
5186  * If len == 0, do from off to EOF.
5187  *
5188  * The normal cases should be len == 0 & off == 0 (entire vp list),
5189  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5190  * (from pageout).
5191  */
5192 /*ARGSUSED*/
5193 static int
5194 ufs_putpages(
5195 	struct vnode *vp,
5196 	offset_t off,
5197 	size_t len,
5198 	int flags,
5199 	struct cred *cr)
5200 {
5201 	u_offset_t io_off;
5202 	u_offset_t eoff;
5203 	struct inode *ip = VTOI(vp);
5204 	page_t *pp;
5205 	size_t io_len;
5206 	int err = 0;
5207 	int dolock;
5208 
5209 	if (vp->v_count == 0)
5210 		return (ufs_fault(vp, "ufs_putpages: v_count == 0"));
5211 	/*
5212 	 * Acquire the readers/write inode lock before locking
5213 	 * any pages in this inode.
5214 	 * The inode lock is held during i/o.
5215 	 */
5216 	if (len == 0) {
5217 		mutex_enter(&ip->i_tlock);
5218 		ip->i_delayoff = ip->i_delaylen = 0;
5219 		mutex_exit(&ip->i_tlock);
5220 	}
5221 	dolock = (rw_owner(&ip->i_contents) != curthread);
5222 	if (dolock) {
5223 		/*
5224 		 * Must synchronize this thread and any possible thread
5225 		 * operating in the window of vulnerability in wrip().
5226 		 * It is dangerous to allow both a thread doing a putpage
5227 		 * and a thread writing, so serialize them.  The exception
5228 		 * is when the thread in wrip() does something which causes
5229 		 * a putpage operation.  Then, the thread must be allowed
5230 		 * to continue.  It may encounter a bmap_read problem in
5231 		 * ufs_putapage, but that is handled in ufs_putapage.
5232 		 * Allow async writers to proceed, we don't want to block
5233 		 * the pageout daemon.
5234 		 */
5235 		if (ip->i_writer == curthread)
5236 			rw_enter(&ip->i_contents, RW_READER);
5237 		else {
5238 			for (;;) {
5239 				rw_enter(&ip->i_contents, RW_READER);
5240 				mutex_enter(&ip->i_tlock);
5241 				/*
5242 				 * If there is no thread in the critical
5243 				 * section of wrip(), then proceed.
5244 				 * Otherwise, wait until there isn't one.
5245 				 */
5246 				if (ip->i_writer == NULL) {
5247 					mutex_exit(&ip->i_tlock);
5248 					break;
5249 				}
5250 				rw_exit(&ip->i_contents);
5251 				/*
5252 				 * Bounce async writers when we have a writer
5253 				 * working on this file so we don't deadlock
5254 				 * the pageout daemon.
5255 				 */
5256 				if (flags & B_ASYNC) {
5257 					mutex_exit(&ip->i_tlock);
5258 					return (0);
5259 				}
5260 				cv_wait(&ip->i_wrcv, &ip->i_tlock);
5261 				mutex_exit(&ip->i_tlock);
5262 			}
5263 		}
5264 	}
5265 
5266 	if (!vn_has_cached_data(vp)) {
5267 		if (dolock)
5268 			rw_exit(&ip->i_contents);
5269 		return (0);
5270 	}
5271 
5272 	if (len == 0) {
5273 		/*
5274 		 * Search the entire vp list for pages >= off.
5275 		 */
5276 		err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage,
5277 		    flags, cr);
5278 	} else {
5279 		/*
5280 		 * Loop over all offsets in the range looking for
5281 		 * pages to deal with.
5282 		 */
5283 		if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0)
5284 			eoff = MIN(off + len, eoff);
5285 		else
5286 			eoff = off + len;
5287 
5288 		for (io_off = off; io_off < eoff; io_off += io_len) {
5289 			/*
5290 			 * If we are not invalidating, synchronously
5291 			 * freeing or writing pages, use the routine
5292 			 * page_lookup_nowait() to prevent reclaiming
5293 			 * them from the free list.
5294 			 */
5295 			if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
5296 				pp = page_lookup(vp, io_off,
5297 				    (flags & (B_INVAL | B_FREE)) ?
5298 				    SE_EXCL : SE_SHARED);
5299 			} else {
5300 				pp = page_lookup_nowait(vp, io_off,
5301 				    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
5302 			}
5303 
5304 			if (pp == NULL || pvn_getdirty(pp, flags) == 0)
5305 				io_len = PAGESIZE;
5306 			else {
5307 				u_offset_t *io_offp = &io_off;
5308 
5309 				err = ufs_putapage(vp, pp, io_offp, &io_len,
5310 				    flags, cr);
5311 				if (err != 0)
5312 					break;
5313 				/*
5314 				 * "io_off" and "io_len" are returned as
5315 				 * the range of pages we actually wrote.
5316 				 * This allows us to skip ahead more quickly
5317 				 * since several pages may've been dealt
5318 				 * with by this iteration of the loop.
5319 				 */
5320 			}
5321 		}
5322 	}
5323 	if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
5324 		/*
5325 		 * We have just sync'ed back all the pages on
5326 		 * the inode, turn off the IMODTIME flag.
5327 		 */
5328 		mutex_enter(&ip->i_tlock);
5329 		ip->i_flag &= ~IMODTIME;
5330 		mutex_exit(&ip->i_tlock);
5331 	}
5332 	if (dolock)
5333 		rw_exit(&ip->i_contents);
5334 	return (err);
5335 }
5336 
5337 static void
5338 ufs_iodone(buf_t *bp)
5339 {
5340 	struct inode *ip;
5341 
5342 	ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
5343 
5344 	bp->b_iodone = NULL;
5345 
5346 	ip = VTOI(bp->b_pages->p_vnode);
5347 
5348 	mutex_enter(&ip->i_tlock);
5349 	if (ip->i_writes >= ufs_LW) {
5350 		if ((ip->i_writes -= bp->b_bcount) <= ufs_LW)
5351 			if (ufs_WRITES)
5352 				cv_broadcast(&ip->i_wrcv); /* wake all up */
5353 	} else {
5354 		ip->i_writes -= bp->b_bcount;
5355 	}
5356 
5357 	mutex_exit(&ip->i_tlock);
5358 	iodone(bp);
5359 }
5360 
5361 /*
5362  * Write out a single page, possibly klustering adjacent
5363  * dirty pages.  The inode lock must be held.
5364  *
5365  * LMXXX - bsize < pagesize not done.
5366  */
5367 /*ARGSUSED*/
5368 int
5369 ufs_putapage(
5370 	struct vnode *vp,
5371 	page_t *pp,
5372 	u_offset_t *offp,
5373 	size_t *lenp,		/* return values */
5374 	int flags,
5375 	struct cred *cr)
5376 {
5377 	u_offset_t io_off;
5378 	u_offset_t off;
5379 	struct inode *ip = VTOI(vp);
5380 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
5381 	struct fs *fs;
5382 	struct buf *bp;
5383 	size_t io_len;
5384 	daddr_t bn;
5385 	int err;
5386 	int contig;
5387 	int dotrans;
5388 
5389 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
5390 
5391 	if (ufsvfsp == NULL) {
5392 		err = EIO;
5393 		goto out_trace;
5394 	}
5395 
5396 	fs = ip->i_fs;
5397 	ASSERT(fs->fs_ronly == 0);
5398 
5399 	/*
5400 	 * If the modified time on the inode has not already been
5401 	 * set elsewhere (e.g. for write/setattr) we set the time now.
5402 	 * This gives us approximate modified times for mmap'ed files
5403 	 * which are modified via stores in the user address space.
5404 	 */
5405 	if ((ip->i_flag & IMODTIME) == 0) {
5406 		mutex_enter(&ip->i_tlock);
5407 		ip->i_flag |= IUPD;
5408 		ip->i_seq++;
5409 		ITIMES_NOLOCK(ip);
5410 		mutex_exit(&ip->i_tlock);
5411 	}
5412 
5413 	/*
5414 	 * Align the request to a block boundry (for old file systems),
5415 	 * and go ask bmap() how contiguous things are for this file.
5416 	 */
5417 	off = pp->p_offset & (offset_t)fs->fs_bmask;	/* block align it */
5418 	contig = 0;
5419 	err = bmap_read(ip, off, &bn, &contig);
5420 	if (err)
5421 		goto out;
5422 	if (bn == UFS_HOLE) {			/* putpage never allocates */
5423 		/*
5424 		 * logging device is in error mode; simply return EIO
5425 		 */
5426 		if (TRANS_ISERROR(ufsvfsp)) {
5427 			err = EIO;
5428 			goto out;
5429 		}
5430 		/*
5431 		 * Oops, the thread in the window in wrip() did some
5432 		 * sort of operation which caused a putpage in the bad
5433 		 * range.  In this case, just return an error which will
5434 		 * cause the software modified bit on the page to set
5435 		 * and the page will get written out again later.
5436 		 */
5437 		if (ip->i_writer == curthread) {
5438 			err = EIO;
5439 			goto out;
5440 		}
5441 		/*
5442 		 * If the pager is trying to push a page in the bad range
5443 		 * just tell it to try again later when things are better.
5444 		 */
5445 		if (flags & B_ASYNC) {
5446 			err = EAGAIN;
5447 			goto out;
5448 		}
5449 		err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE");
5450 		goto out;
5451 	}
5452 
5453 	/*
5454 	 * If it is an fallocate'd block, reverse the negativity since
5455 	 * we are now writing to it
5456 	 */
5457 	if (ISFALLOCBLK(ip, bn)) {
5458 		err = bmap_set_bn(vp, off, dbtofsb(fs, -bn));
5459 		if (err)
5460 			goto out;
5461 
5462 		bn = -bn;
5463 	}
5464 
5465 	/*
5466 	 * Take the length (of contiguous bytes) passed back from bmap()
5467 	 * and _try_ and get a set of pages covering that extent.
5468 	 */
5469 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags);
5470 
5471 	/*
5472 	 * May have run out of memory and not clustered backwards.
5473 	 * off		p_offset
5474 	 * [  pp - 1  ][   pp   ]
5475 	 * [	block		]
5476 	 * We told bmap off, so we have to adjust the bn accordingly.
5477 	 */
5478 	if (io_off > off) {
5479 		bn += btod(io_off - off);
5480 		contig -= (io_off - off);
5481 	}
5482 
5483 	/*
5484 	 * bmap was carefull to tell us the right size so use that.
5485 	 * There might be unallocated frags at the end.
5486 	 * LMXXX - bzero the end of the page?  We must be writing after EOF.
5487 	 */
5488 	if (io_len > contig) {
5489 		ASSERT(io_len - contig < fs->fs_bsize);
5490 		io_len -= (io_len - contig);
5491 	}
5492 
5493 	/*
5494 	 * Handle the case where we are writing the last page after EOF.
5495 	 *
5496 	 * XXX - just a patch for i-mt3.
5497 	 */
5498 	if (io_len == 0) {
5499 		ASSERT(pp->p_offset >=
5500 		    (u_offset_t)(roundup(ip->i_size, PAGESIZE)));
5501 		io_len = PAGESIZE;
5502 	}
5503 
5504 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags);
5505 
5506 	ULOCKFS_SET_MOD(ITOUL(ip));
5507 
5508 	bp->b_edev = ip->i_dev;
5509 	bp->b_dev = cmpdev(ip->i_dev);
5510 	bp->b_blkno = bn;
5511 	bp->b_un.b_addr = (caddr_t)0;
5512 	bp->b_file = ip->i_vnode;
5513 
5514 	/*
5515 	 * File contents of shadow or quota inodes are metadata, and updates
5516 	 * to these need to be put into a logging transaction. All direct
5517 	 * callers in UFS do that, but fsflush can come here _before_ the
5518 	 * normal codepath. An example would be updating ACL information, for
5519 	 * which the normal codepath would be:
5520 	 *	ufs_si_store()
5521 	 *	ufs_rdwri()
5522 	 *	wrip()
5523 	 *	segmap_release()
5524 	 *	VOP_PUTPAGE()
5525 	 * Here, fsflush can pick up the dirty page before segmap_release()
5526 	 * forces it out. If that happens, there's no transaction.
5527 	 * We therefore need to test whether a transaction exists, and if not
5528 	 * create one - for fsflush.
5529 	 */
5530 	dotrans =
5531 	    (((ip->i_mode & IFMT) == IFSHAD || ufsvfsp->vfs_qinod == ip) &&
5532 	    ((curthread->t_flag & T_DONTBLOCK) == 0) &&
5533 	    (TRANS_ISTRANS(ufsvfsp)));
5534 
5535 	if (dotrans) {
5536 		curthread->t_flag |= T_DONTBLOCK;
5537 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5538 	}
5539 	if (TRANS_ISTRANS(ufsvfsp)) {
5540 		if ((ip->i_mode & IFMT) == IFSHAD) {
5541 			TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD);
5542 		} else if (ufsvfsp->vfs_qinod == ip) {
5543 			TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR,
5544 			    0, 0);
5545 		}
5546 	}
5547 	if (dotrans) {
5548 		TRANS_END_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5549 		curthread->t_flag &= ~T_DONTBLOCK;
5550 	}
5551 
5552 	/* write throttle */
5553 
5554 	ASSERT(bp->b_iodone == NULL);
5555 	bp->b_iodone = (int (*)())ufs_iodone;
5556 	mutex_enter(&ip->i_tlock);
5557 	ip->i_writes += bp->b_bcount;
5558 	mutex_exit(&ip->i_tlock);
5559 
5560 	if (bp->b_flags & B_ASYNC) {
5561 		if (ufsvfsp->vfs_log) {
5562 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5563 		} else if (ufsvfsp->vfs_snapshot) {
5564 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5565 		} else {
5566 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5567 			ub.ub_putasyncs.value.ul++;
5568 			(void) bdev_strategy(bp);
5569 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5570 		}
5571 	} else {
5572 		if (ufsvfsp->vfs_log) {
5573 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5574 		} else if (ufsvfsp->vfs_snapshot) {
5575 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5576 		} else {
5577 			ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5578 			ub.ub_putsyncs.value.ul++;
5579 			(void) bdev_strategy(bp);
5580 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5581 		}
5582 		err = biowait(bp);
5583 		pageio_done(bp);
5584 		pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
5585 	}
5586 
5587 	pp = NULL;
5588 
5589 out:
5590 	if (err != 0 && pp != NULL)
5591 		pvn_write_done(pp, B_ERROR | B_WRITE | flags);
5592 
5593 	if (offp)
5594 		*offp = io_off;
5595 	if (lenp)
5596 		*lenp = io_len;
5597 out_trace:
5598 	return (err);
5599 }
5600 
5601 uint64_t ufs_map_alock_retry_cnt;
5602 uint64_t ufs_map_lockfs_retry_cnt;
5603 
5604 /* ARGSUSED */
5605 static int
5606 ufs_map(struct vnode *vp,
5607 	offset_t off,
5608 	struct as *as,
5609 	caddr_t *addrp,
5610 	size_t len,
5611 	uchar_t prot,
5612 	uchar_t maxprot,
5613 	uint_t flags,
5614 	struct cred *cr,
5615 	caller_context_t *ct)
5616 {
5617 	struct segvn_crargs vn_a;
5618 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5619 	struct ulockfs *ulp;
5620 	int error, sig;
5621 	k_sigset_t smask;
5622 	caddr_t hint = *addrp;
5623 
5624 	if (vp->v_flag & VNOMAP) {
5625 		error = ENOSYS;
5626 		goto out;
5627 	}
5628 
5629 	if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) {
5630 		error = ENXIO;
5631 		goto out;
5632 	}
5633 
5634 	if (vp->v_type != VREG) {
5635 		error = ENODEV;
5636 		goto out;
5637 	}
5638 
5639 retry_map:
5640 	*addrp = hint;
5641 	/*
5642 	 * If file is being locked, disallow mapping.
5643 	 */
5644 	if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) {
5645 		error = EAGAIN;
5646 		goto out;
5647 	}
5648 
5649 	as_rangelock(as);
5650 	/*
5651 	 * Note that if we are retrying (because ufs_lockfs_trybegin failed in
5652 	 * the previous attempt), some other thread could have grabbed
5653 	 * the same VA range if MAP_FIXED is set. In that case, choose_addr
5654 	 * would unmap the valid VA range, that is ok.
5655 	 */
5656 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5657 	if (error != 0) {
5658 		as_rangeunlock(as);
5659 		goto out;
5660 	}
5661 
5662 	/*
5663 	 * a_lock has to be acquired before entering the lockfs protocol
5664 	 * because that is the order in which pagefault works. Also we cannot
5665 	 * block on a_lock here because this waiting writer will prevent
5666 	 * further readers like ufs_read from progressing and could cause
5667 	 * deadlock between ufs_read/ufs_map/pagefault when a quiesce is
5668 	 * pending.
5669 	 */
5670 	while (!AS_LOCK_TRYENTER(as, RW_WRITER)) {
5671 		ufs_map_alock_retry_cnt++;
5672 		delay(RETRY_LOCK_DELAY);
5673 	}
5674 
5675 	/*
5676 	 * We can't hold as->a_lock and wait for lockfs to succeed because
5677 	 * the proc tools might hang on a_lock, so call ufs_lockfs_trybegin()
5678 	 * instead.
5679 	 */
5680 	if (error = ufs_lockfs_trybegin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK)) {
5681 		/*
5682 		 * ufs_lockfs_trybegin() did not succeed. It is safer to give up
5683 		 * as->a_lock and wait for ulp->ul_fs_lock status to change.
5684 		 */
5685 		ufs_map_lockfs_retry_cnt++;
5686 		AS_LOCK_EXIT(as);
5687 		as_rangeunlock(as);
5688 		if (error == EIO)
5689 			goto out;
5690 
5691 		mutex_enter(&ulp->ul_lock);
5692 		while (ulp->ul_fs_lock & ULOCKFS_MAP_MASK) {
5693 			if (ULOCKFS_IS_SLOCK(ulp) || ufsvfsp->vfs_nointr) {
5694 				cv_wait(&ulp->ul_cv, &ulp->ul_lock);
5695 			} else {
5696 				sigintr(&smask, 1);
5697 				sig = cv_wait_sig(&ulp->ul_cv, &ulp->ul_lock);
5698 				sigunintr(&smask);
5699 				if (((ulp->ul_fs_lock & ULOCKFS_MAP_MASK) &&
5700 				    !sig) || ufsvfsp->vfs_dontblock) {
5701 					mutex_exit(&ulp->ul_lock);
5702 					return (EINTR);
5703 				}
5704 			}
5705 		}
5706 		mutex_exit(&ulp->ul_lock);
5707 		goto retry_map;
5708 	}
5709 
5710 	vn_a.vp = vp;
5711 	vn_a.offset = (u_offset_t)off;
5712 	vn_a.type = flags & MAP_TYPE;
5713 	vn_a.prot = prot;
5714 	vn_a.maxprot = maxprot;
5715 	vn_a.cred = cr;
5716 	vn_a.amp = NULL;
5717 	vn_a.flags = flags & ~MAP_TYPE;
5718 	vn_a.szc = 0;
5719 	vn_a.lgrp_mem_policy_flags = 0;
5720 
5721 	error = as_map_locked(as, *addrp, len, segvn_create, &vn_a);
5722 	if (ulp)
5723 		ufs_lockfs_end(ulp);
5724 	as_rangeunlock(as);
5725 out:
5726 	return (error);
5727 }
5728 
5729 /* ARGSUSED */
5730 static int
5731 ufs_addmap(struct vnode *vp,
5732 	offset_t off,
5733 	struct as *as,
5734 	caddr_t addr,
5735 	size_t	len,
5736 	uchar_t  prot,
5737 	uchar_t  maxprot,
5738 	uint_t    flags,
5739 	struct cred *cr,
5740 	caller_context_t *ct)
5741 {
5742 	struct inode *ip = VTOI(vp);
5743 
5744 	if (vp->v_flag & VNOMAP) {
5745 		return (ENOSYS);
5746 	}
5747 
5748 	mutex_enter(&ip->i_tlock);
5749 	ip->i_mapcnt += btopr(len);
5750 	mutex_exit(&ip->i_tlock);
5751 	return (0);
5752 }
5753 
5754 /*ARGSUSED*/
5755 static int
5756 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5757 	size_t len, uint_t prot,  uint_t maxprot,  uint_t flags,
5758 	struct cred *cr, caller_context_t *ct)
5759 {
5760 	struct inode *ip = VTOI(vp);
5761 
5762 	if (vp->v_flag & VNOMAP) {
5763 		return (ENOSYS);
5764 	}
5765 
5766 	mutex_enter(&ip->i_tlock);
5767 	ip->i_mapcnt -= btopr(len); 	/* Count released mappings */
5768 	ASSERT(ip->i_mapcnt >= 0);
5769 	mutex_exit(&ip->i_tlock);
5770 	return (0);
5771 }
5772 /*
5773  * Return the answer requested to poll() for non-device files
5774  */
5775 struct pollhead ufs_pollhd;
5776 
5777 /* ARGSUSED */
5778 int
5779 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp,
5780 	caller_context_t *ct)
5781 {
5782 	struct ufsvfs	*ufsvfsp;
5783 
5784 	*revp = 0;
5785 	ufsvfsp = VTOI(vp)->i_ufsvfs;
5786 
5787 	if (!ufsvfsp) {
5788 		*revp = POLLHUP;
5789 		goto out;
5790 	}
5791 
5792 	if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) ||
5793 	    ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) {
5794 		*revp |= POLLERR;
5795 
5796 	} else {
5797 		if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly &&
5798 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5799 			*revp |= POLLOUT;
5800 
5801 		if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly &&
5802 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5803 			*revp |= POLLWRBAND;
5804 
5805 		if (ev & POLLIN)
5806 			*revp |= POLLIN;
5807 
5808 		if (ev & POLLRDNORM)
5809 			*revp |= POLLRDNORM;
5810 
5811 		if (ev & POLLRDBAND)
5812 			*revp |= POLLRDBAND;
5813 	}
5814 
5815 	if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP)))
5816 		*revp |= POLLPRI;
5817 out:
5818 	*phpp = !any && !*revp ? &ufs_pollhd : (struct pollhead *)NULL;
5819 
5820 	return (0);
5821 }
5822 
5823 /* ARGSUSED */
5824 static int
5825 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr,
5826 	caller_context_t *ct)
5827 {
5828 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
5829 	struct ulockfs	*ulp = NULL;
5830 	struct inode 	*sip = NULL;
5831 	int		error;
5832 	struct inode 	*ip = VTOI(vp);
5833 	int		issync;
5834 
5835 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK);
5836 	if (error)
5837 		return (error);
5838 
5839 	switch (cmd) {
5840 		/*
5841 		 * Have to handle _PC_NAME_MAX here, because the normal way
5842 		 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5843 		 * results in a lock ordering reversal between
5844 		 * ufs_lockfs_{begin,end}() and
5845 		 * ufs_thread_{suspend,continue}().
5846 		 *
5847 		 * Keep in sync with ufs_statvfs().
5848 		 */
5849 	case _PC_NAME_MAX:
5850 		*valp = MAXNAMLEN;
5851 		break;
5852 
5853 	case _PC_FILESIZEBITS:
5854 		if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
5855 			*valp = UFS_FILESIZE_BITS;
5856 		else
5857 			*valp = 32;
5858 		break;
5859 
5860 	case _PC_XATTR_EXISTS:
5861 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
5862 
5863 			error =
5864 			    ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR, cr);
5865 			if (error ==  0 && sip != NULL) {
5866 				/* Start transaction */
5867 				if (ulp) {
5868 					TRANS_BEGIN_CSYNC(ufsvfsp, issync,
5869 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5870 				}
5871 				/*
5872 				 * Is directory empty
5873 				 */
5874 				rw_enter(&sip->i_rwlock, RW_WRITER);
5875 				rw_enter(&sip->i_contents, RW_WRITER);
5876 				if (ufs_xattrdirempty(sip,
5877 				    sip->i_number, CRED())) {
5878 					rw_enter(&ip->i_contents, RW_WRITER);
5879 					ufs_unhook_shadow(ip, sip);
5880 					rw_exit(&ip->i_contents);
5881 
5882 					*valp = 0;
5883 
5884 				} else
5885 					*valp = 1;
5886 				rw_exit(&sip->i_contents);
5887 				rw_exit(&sip->i_rwlock);
5888 				if (ulp) {
5889 					TRANS_END_CSYNC(ufsvfsp, error, issync,
5890 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5891 				}
5892 				VN_RELE(ITOV(sip));
5893 			} else if (error == ENOENT) {
5894 				*valp = 0;
5895 				error = 0;
5896 			}
5897 		} else {
5898 			error = fs_pathconf(vp, cmd, valp, cr, ct);
5899 		}
5900 		break;
5901 
5902 	case _PC_ACL_ENABLED:
5903 		*valp = _ACL_ACLENT_ENABLED;
5904 		break;
5905 
5906 	case _PC_MIN_HOLE_SIZE:
5907 		*valp = (ulong_t)ip->i_fs->fs_bsize;
5908 		break;
5909 
5910 	case _PC_SATTR_ENABLED:
5911 	case _PC_SATTR_EXISTS:
5912 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5913 		    (vp->v_type == VREG || vp->v_type == VDIR);
5914 		break;
5915 
5916 	case _PC_TIMESTAMP_RESOLUTION:
5917 		/*
5918 		 * UFS keeps only microsecond timestamp resolution.
5919 		 * This is historical and will probably never change.
5920 		 */
5921 		*valp = 1000L;
5922 		break;
5923 
5924 	default:
5925 		error = fs_pathconf(vp, cmd, valp, cr, ct);
5926 		break;
5927 	}
5928 
5929 	if (ulp != NULL) {
5930 		ufs_lockfs_end(ulp);
5931 	}
5932 	return (error);
5933 }
5934 
5935 int ufs_pageio_writes, ufs_pageio_reads;
5936 
5937 /*ARGSUSED*/
5938 static int
5939 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len,
5940 	int flags, struct cred *cr, caller_context_t *ct)
5941 {
5942 	struct inode *ip = VTOI(vp);
5943 	struct ufsvfs *ufsvfsp;
5944 	page_t *npp = NULL, *opp = NULL, *cpp = pp;
5945 	struct buf *bp;
5946 	daddr_t bn;
5947 	size_t done_len = 0, cur_len = 0;
5948 	int err = 0;
5949 	int contig = 0;
5950 	int dolock;
5951 	int vmpss = 0;
5952 	struct ulockfs *ulp;
5953 
5954 	if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp &&
5955 	    vp->v_mpssdata != NULL) {
5956 		vmpss = 1;
5957 	}
5958 
5959 	dolock = (rw_owner(&ip->i_contents) != curthread);
5960 	/*
5961 	 * We need a better check.  Ideally, we would use another
5962 	 * vnodeops so that hlocked and forcibly unmounted file
5963 	 * systems would return EIO where appropriate and w/o the
5964 	 * need for these checks.
5965 	 */
5966 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5967 		return (EIO);
5968 
5969 	/*
5970 	 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5971 	 * ul_lock must be taken before locking pages so we can't use it here
5972 	 * if pp is non NULL because segvn already locked pages
5973 	 * SE_EXCL. Instead we rely on the fact that a forced umount or
5974 	 * applying a filesystem lock via ufs_fiolfs() will block in the
5975 	 * implicit call to ufs_flush() until we unlock the pages after the
5976 	 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5977 	 * above 0 until they are done. We have to be careful not to increment
5978 	 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5979 	 *
5980 	 * If pp is NULL use ul_lock to make sure we don't increment
5981 	 * ul_vnops_cnt after forceful unmount hlocks the file system.
5982 	 */
5983 	if (vmpss || pp == NULL) {
5984 		ulp = &ufsvfsp->vfs_ulockfs;
5985 		if (pp == NULL)
5986 			mutex_enter(&ulp->ul_lock);
5987 		if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) {
5988 			if (pp == NULL) {
5989 				mutex_exit(&ulp->ul_lock);
5990 			}
5991 			return (vmpss ? EIO : EINVAL);
5992 		}
5993 		atomic_inc_ulong(&ulp->ul_vnops_cnt);
5994 		if (pp == NULL)
5995 			mutex_exit(&ulp->ul_lock);
5996 		if (ufs_quiesce_pend) {
5997 			if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
5998 				cv_broadcast(&ulp->ul_cv);
5999 			return (vmpss ? EIO : EINVAL);
6000 		}
6001 	}
6002 
6003 	if (dolock) {
6004 		/*
6005 		 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
6006 		 * handle a fault against a segment that maps vnode pages with
6007 		 * large mappings.  Segvn creates pages and holds them locked
6008 		 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
6009 		 * use rw_tryenter() to avoid a potential deadlock since in
6010 		 * lock order i_contents needs to be taken first.
6011 		 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
6012 		 */
6013 		if (!vmpss) {
6014 			rw_enter(&ip->i_contents, RW_READER);
6015 		} else if (!rw_tryenter(&ip->i_contents, RW_READER)) {
6016 			if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6017 				cv_broadcast(&ulp->ul_cv);
6018 			return (EDEADLK);
6019 		}
6020 	}
6021 
6022 	/*
6023 	 * Return an error to segvn because the pagefault request is beyond
6024 	 * PAGESIZE rounded EOF.
6025 	 */
6026 	if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) {
6027 		if (dolock)
6028 			rw_exit(&ip->i_contents);
6029 		if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6030 			cv_broadcast(&ulp->ul_cv);
6031 		return (EFAULT);
6032 	}
6033 
6034 	if (pp == NULL) {
6035 		if (bmap_has_holes(ip)) {
6036 			err = ENOSYS;
6037 		} else {
6038 			err = EINVAL;
6039 		}
6040 		if (dolock)
6041 			rw_exit(&ip->i_contents);
6042 		if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6043 			cv_broadcast(&ulp->ul_cv);
6044 		return (err);
6045 	}
6046 
6047 	/*
6048 	 * Break the io request into chunks, one for each contiguous
6049 	 * stretch of disk blocks in the target file.
6050 	 */
6051 	while (done_len < io_len) {
6052 		ASSERT(cpp);
6053 		contig = 0;
6054 		if (err = bmap_read(ip, (u_offset_t)(io_off + done_len),
6055 		    &bn, &contig))
6056 			break;
6057 
6058 		if (bn == UFS_HOLE) {	/* No holey swapfiles */
6059 			if (vmpss) {
6060 				err = EFAULT;
6061 				break;
6062 			}
6063 			err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE");
6064 			break;
6065 		}
6066 
6067 		cur_len = MIN(io_len - done_len, contig);
6068 		/*
6069 		 * Zero out a page beyond EOF, when the last block of
6070 		 * a file is a UFS fragment so that ufs_pageio() can be used
6071 		 * instead of ufs_getpage() to handle faults against
6072 		 * segvn segments that use large pages.
6073 		 */
6074 		page_list_break(&cpp, &npp, btopr(cur_len));
6075 		if ((flags & B_READ) && (cur_len & PAGEOFFSET)) {
6076 			size_t xlen = cur_len & PAGEOFFSET;
6077 			pagezero(cpp->p_prev, xlen, PAGESIZE - xlen);
6078 		}
6079 
6080 		bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
6081 		ASSERT(bp != NULL);
6082 
6083 		bp->b_edev = ip->i_dev;
6084 		bp->b_dev = cmpdev(ip->i_dev);
6085 		bp->b_blkno = bn;
6086 		bp->b_un.b_addr = (caddr_t)0;
6087 		bp->b_file = ip->i_vnode;
6088 
6089 		ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
6090 		ub.ub_pageios.value.ul++;
6091 		if (ufsvfsp->vfs_snapshot)
6092 			fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp);
6093 		else
6094 			(void) bdev_strategy(bp);
6095 
6096 		if (flags & B_READ)
6097 			ufs_pageio_reads++;
6098 		else
6099 			ufs_pageio_writes++;
6100 		if (flags & B_READ)
6101 			lwp_stat_update(LWP_STAT_INBLK, 1);
6102 		else
6103 			lwp_stat_update(LWP_STAT_OUBLK, 1);
6104 		/*
6105 		 * If the request is not B_ASYNC, wait for i/o to complete
6106 		 * and re-assemble the page list to return to the caller.
6107 		 * If it is B_ASYNC we leave the page list in pieces and
6108 		 * cleanup() will dispose of them.
6109 		 */
6110 		if ((flags & B_ASYNC) == 0) {
6111 			err = biowait(bp);
6112 			pageio_done(bp);
6113 			if (err)
6114 				break;
6115 			page_list_concat(&opp, &cpp);
6116 		}
6117 		cpp = npp;
6118 		npp = NULL;
6119 		if (flags & B_READ)
6120 			cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t);
6121 		done_len += cur_len;
6122 	}
6123 	ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len));
6124 	if (err) {
6125 		if (flags & B_ASYNC) {
6126 			/* Cleanup unprocessed parts of list */
6127 			page_list_concat(&cpp, &npp);
6128 			if (flags & B_READ)
6129 				pvn_read_done(cpp, B_ERROR);
6130 			else
6131 				pvn_write_done(cpp, B_ERROR);
6132 		} else {
6133 			/* Re-assemble list and let caller clean up */
6134 			page_list_concat(&opp, &cpp);
6135 			page_list_concat(&opp, &npp);
6136 		}
6137 	}
6138 
6139 	if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) &&
6140 	    ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) {
6141 		mutex_enter(&ip->i_tlock);
6142 		ip->i_flag |= IACC;
6143 		ITIMES_NOLOCK(ip);
6144 		mutex_exit(&ip->i_tlock);
6145 	}
6146 
6147 	if (dolock)
6148 		rw_exit(&ip->i_contents);
6149 	if (vmpss && !atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6150 		cv_broadcast(&ulp->ul_cv);
6151 	return (err);
6152 }
6153 
6154 /*
6155  * Called when the kernel is in a frozen state to dump data
6156  * directly to the device. It uses a private dump data structure,
6157  * set up by dump_ctl, to locate the correct disk block to which to dump.
6158  */
6159 /*ARGSUSED*/
6160 static int
6161 ufs_dump(vnode_t *vp, caddr_t addr, offset_t ldbn, offset_t dblks,
6162     caller_context_t *ct)
6163 {
6164 	u_offset_t	file_size;
6165 	struct inode    *ip = VTOI(vp);
6166 	struct fs	*fs = ip->i_fs;
6167 	daddr_t		dbn, lfsbn;
6168 	int		disk_blks = fs->fs_bsize >> DEV_BSHIFT;
6169 	int		error = 0;
6170 	int		ndbs, nfsbs;
6171 
6172 	/*
6173 	 * forced unmount case
6174 	 */
6175 	if (ip->i_ufsvfs == NULL)
6176 		return (EIO);
6177 	/*
6178 	 * Validate the inode that it has not been modified since
6179 	 * the dump structure is allocated.
6180 	 */
6181 	mutex_enter(&ip->i_tlock);
6182 	if ((dump_info == NULL) ||
6183 	    (dump_info->ip != ip) ||
6184 	    (dump_info->time.tv_sec != ip->i_mtime.tv_sec) ||
6185 	    (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) {
6186 		mutex_exit(&ip->i_tlock);
6187 		return (-1);
6188 	}
6189 	mutex_exit(&ip->i_tlock);
6190 
6191 	/*
6192 	 * See that the file has room for this write
6193 	 */
6194 	UFS_GET_ISIZE(&file_size, ip);
6195 
6196 	if (ldbtob(ldbn + dblks) > file_size)
6197 		return (ENOSPC);
6198 
6199 	/*
6200 	 * Find the physical disk block numbers from the dump
6201 	 * private data structure directly and write out the data
6202 	 * in contiguous block lumps
6203 	 */
6204 	while (dblks > 0 && !error) {
6205 		lfsbn = (daddr_t)lblkno(fs, ldbtob(ldbn));
6206 		dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks;
6207 		nfsbs = 1;
6208 		ndbs = disk_blks - ldbn % disk_blks;
6209 		while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn +
6210 		    nfsbs]) == dbn + ndbs) {
6211 			nfsbs++;
6212 			ndbs += disk_blks;
6213 		}
6214 		if (ndbs > dblks)
6215 			ndbs = dblks;
6216 		error = bdev_dump(ip->i_dev, addr, dbn, ndbs);
6217 		addr += ldbtob((offset_t)ndbs);
6218 		dblks -= ndbs;
6219 		ldbn += ndbs;
6220 	}
6221 	return (error);
6222 
6223 }
6224 
6225 /*
6226  * Prepare the file system before and after the dump operation.
6227  *
6228  * action = DUMP_ALLOC:
6229  * Preparation before dump, allocate dump private data structure
6230  * to hold all the direct and indirect block info for dump.
6231  *
6232  * action = DUMP_FREE:
6233  * Clean up after dump, deallocate the dump private data structure.
6234  *
6235  * action = DUMP_SCAN:
6236  * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6237  * if found, the starting file-relative DEV_BSIZE lbn is written
6238  * to *bklp; that lbn is intended for use with VOP_DUMP()
6239  */
6240 /*ARGSUSED*/
6241 static int
6242 ufs_dumpctl(vnode_t *vp, int action, offset_t *blkp, caller_context_t *ct)
6243 {
6244 	struct inode	*ip = VTOI(vp);
6245 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
6246 	struct fs	*fs;
6247 	daddr32_t	*dblk, *storeblk;
6248 	daddr32_t	*nextblk, *endblk;
6249 	struct buf	*bp;
6250 	int		i, entry, entries;
6251 	int		n, ncontig;
6252 
6253 	/*
6254 	 * check for forced unmount
6255 	 */
6256 	if (ufsvfsp == NULL)
6257 		return (EIO);
6258 
6259 	if (action == DUMP_ALLOC) {
6260 		/*
6261 		 * alloc and record dump_info
6262 		 */
6263 		if (dump_info != NULL)
6264 			return (EINVAL);
6265 
6266 		ASSERT(vp->v_type == VREG);
6267 		fs = ufsvfsp->vfs_fs;
6268 
6269 		rw_enter(&ip->i_contents, RW_READER);
6270 
6271 		if (bmap_has_holes(ip)) {
6272 			rw_exit(&ip->i_contents);
6273 			return (EFAULT);
6274 		}
6275 
6276 		/*
6277 		 * calculate and allocate space needed according to i_size
6278 		 */
6279 		entries = (int)lblkno(fs, blkroundup(fs, ip->i_size));
6280 		dump_info = kmem_alloc(sizeof (struct dump) +
6281 		    (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP);
6282 		if (dump_info == NULL) {
6283 			rw_exit(&ip->i_contents);
6284 			return (ENOMEM);
6285 		}
6286 
6287 		/* Start saving the info */
6288 		dump_info->fsbs = entries;
6289 		dump_info->ip = ip;
6290 		storeblk = &dump_info->dblk[0];
6291 
6292 		/* Direct Blocks */
6293 		for (entry = 0; entry < NDADDR && entry < entries; entry++)
6294 			*storeblk++ = ip->i_db[entry];
6295 
6296 		/* Indirect Blocks */
6297 		for (i = 0; i < NIADDR; i++) {
6298 			int error = 0;
6299 
6300 			bp = UFS_BREAD(ufsvfsp,
6301 			    ip->i_dev, fsbtodb(fs, ip->i_ib[i]), fs->fs_bsize);
6302 			if (bp->b_flags & B_ERROR)
6303 				error = EIO;
6304 			else {
6305 				dblk = bp->b_un.b_daddr;
6306 				if ((storeblk = save_dblks(ip, ufsvfsp,
6307 				    storeblk, dblk, i, entries)) == NULL)
6308 					error = EIO;
6309 			}
6310 
6311 			brelse(bp);
6312 
6313 			if (error != 0) {
6314 				kmem_free(dump_info, sizeof (struct dump) +
6315 				    (entries - 1) * sizeof (daddr32_t));
6316 				rw_exit(&ip->i_contents);
6317 				dump_info = NULL;
6318 				return (error);
6319 			}
6320 		}
6321 		/* and time stamp the information */
6322 		mutex_enter(&ip->i_tlock);
6323 		dump_info->time = ip->i_mtime;
6324 		mutex_exit(&ip->i_tlock);
6325 
6326 		rw_exit(&ip->i_contents);
6327 	} else if (action == DUMP_FREE) {
6328 		/*
6329 		 * free dump_info
6330 		 */
6331 		if (dump_info == NULL)
6332 			return (EINVAL);
6333 		entries = dump_info->fsbs - 1;
6334 		kmem_free(dump_info, sizeof (struct dump) +
6335 		    entries * sizeof (daddr32_t));
6336 		dump_info = NULL;
6337 	} else if (action == DUMP_SCAN) {
6338 		/*
6339 		 * scan dump_info
6340 		 */
6341 		if (dump_info == NULL)
6342 			return (EINVAL);
6343 
6344 		dblk = dump_info->dblk;
6345 		nextblk = dblk + 1;
6346 		endblk = dblk + dump_info->fsbs - 1;
6347 		fs = ufsvfsp->vfs_fs;
6348 		ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT);
6349 
6350 		/*
6351 		 * scan dblk[] entries; contig fs space is found when:
6352 		 * ((current blkno + frags per block) == next blkno)
6353 		 */
6354 		n = 0;
6355 		while (n < ncontig && dblk < endblk) {
6356 			if ((*dblk + fs->fs_frag) == *nextblk)
6357 				n++;
6358 			else
6359 				n = 0;
6360 			dblk++;
6361 			nextblk++;
6362 		}
6363 
6364 		/*
6365 		 * index is where size bytes of contig space begins;
6366 		 * conversion from index to the file's DEV_BSIZE lbn
6367 		 * is equivalent to:  (index * fs_bsize) / DEV_BSIZE
6368 		 */
6369 		if (n == ncontig) {
6370 			i = (dblk - dump_info->dblk) - ncontig;
6371 			*blkp = i << (fs->fs_bshift - DEV_BSHIFT);
6372 		} else
6373 			return (EFAULT);
6374 	}
6375 	return (0);
6376 }
6377 
6378 /*
6379  * Recursive helper function for ufs_dumpctl().  It follows the indirect file
6380  * system  blocks until it reaches the the disk block addresses, which are
6381  * then stored into the given buffer, storeblk.
6382  */
6383 static daddr32_t *
6384 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp,  daddr32_t *storeblk,
6385     daddr32_t *dblk, int level, int entries)
6386 {
6387 	struct fs	*fs = ufsvfsp->vfs_fs;
6388 	struct buf	*bp;
6389 	int		i;
6390 
6391 	if (level == 0) {
6392 		for (i = 0; i < NINDIR(fs); i++) {
6393 			if (storeblk - dump_info->dblk >= entries)
6394 				break;
6395 			*storeblk++ = dblk[i];
6396 		}
6397 		return (storeblk);
6398 	}
6399 	for (i = 0; i < NINDIR(fs); i++) {
6400 		if (storeblk - dump_info->dblk >= entries)
6401 			break;
6402 		bp = UFS_BREAD(ufsvfsp,
6403 		    ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize);
6404 		if (bp->b_flags & B_ERROR) {
6405 			brelse(bp);
6406 			return (NULL);
6407 		}
6408 		storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr,
6409 		    level - 1, entries);
6410 		brelse(bp);
6411 
6412 		if (storeblk == NULL)
6413 			return (NULL);
6414 	}
6415 	return (storeblk);
6416 }
6417 
6418 /* ARGSUSED */
6419 static int
6420 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag,
6421 	struct cred *cr, caller_context_t *ct)
6422 {
6423 	struct inode	*ip = VTOI(vp);
6424 	struct ulockfs	*ulp;
6425 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
6426 	ulong_t		vsa_mask = vsap->vsa_mask;
6427 	int		err = EINVAL;
6428 
6429 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6430 
6431 	/*
6432 	 * Only grab locks if needed - they're not needed to check vsa_mask
6433 	 * or if the mask contains no acl flags.
6434 	 */
6435 	if (vsa_mask != 0) {
6436 		if (err = ufs_lockfs_begin(ufsvfsp, &ulp,
6437 		    ULOCKFS_GETATTR_MASK))
6438 			return (err);
6439 
6440 		rw_enter(&ip->i_contents, RW_READER);
6441 		err = ufs_acl_get(ip, vsap, flag, cr);
6442 		rw_exit(&ip->i_contents);
6443 
6444 		if (ulp)
6445 			ufs_lockfs_end(ulp);
6446 	}
6447 	return (err);
6448 }
6449 
6450 /* ARGSUSED */
6451 static int
6452 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr,
6453 	caller_context_t *ct)
6454 {
6455 	struct inode	*ip = VTOI(vp);
6456 	struct ulockfs	*ulp = NULL;
6457 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
6458 	ulong_t		vsa_mask = vsap->vsa_mask;
6459 	int		err;
6460 	int		haverwlock = 1;
6461 	int		trans_size;
6462 	int		donetrans = 0;
6463 	int		retry = 1;
6464 
6465 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
6466 
6467 	/* Abort now if the request is either empty or invalid. */
6468 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6469 	if ((vsa_mask == 0) ||
6470 	    ((vsap->vsa_aclentp == NULL) &&
6471 	    (vsap->vsa_dfaclentp == NULL))) {
6472 		err = EINVAL;
6473 		goto out;
6474 	}
6475 
6476 	/*
6477 	 * Following convention, if this is a directory then we acquire the
6478 	 * inode's i_rwlock after starting a UFS logging transaction;
6479 	 * otherwise, we acquire it beforehand. Since we were called (and
6480 	 * must therefore return) with the lock held, we will have to drop it,
6481 	 * and later reacquire it, if operating on a directory.
6482 	 */
6483 	if (vp->v_type == VDIR) {
6484 		rw_exit(&ip->i_rwlock);
6485 		haverwlock = 0;
6486 	} else {
6487 		/* Upgrade the lock if required. */
6488 		if (!rw_write_held(&ip->i_rwlock)) {
6489 			rw_exit(&ip->i_rwlock);
6490 			rw_enter(&ip->i_rwlock, RW_WRITER);
6491 		}
6492 	}
6493 
6494 again:
6495 	ASSERT(!(vp->v_type == VDIR && haverwlock));
6496 	if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) {
6497 		ulp = NULL;
6498 		retry = 0;
6499 		goto out;
6500 	}
6501 
6502 	/*
6503 	 * Check that the file system supports this operation. Note that
6504 	 * ufs_lockfs_begin() will have checked that the file system had
6505 	 * not been forcibly unmounted.
6506 	 */
6507 	if (ufsvfsp->vfs_fs->fs_ronly) {
6508 		err = EROFS;
6509 		goto out;
6510 	}
6511 	if (ufsvfsp->vfs_nosetsec) {
6512 		err = ENOSYS;
6513 		goto out;
6514 	}
6515 
6516 	if (ulp) {
6517 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR,
6518 		    trans_size = TOP_SETSECATTR_SIZE(VTOI(vp)));
6519 		donetrans = 1;
6520 	}
6521 
6522 	if (vp->v_type == VDIR) {
6523 		rw_enter(&ip->i_rwlock, RW_WRITER);
6524 		haverwlock = 1;
6525 	}
6526 
6527 	ASSERT(haverwlock);
6528 
6529 	/* Do the actual work. */
6530 	rw_enter(&ip->i_contents, RW_WRITER);
6531 	/*
6532 	 * Suppress out of inodes messages if we will retry.
6533 	 */
6534 	if (retry)
6535 		ip->i_flag |= IQUIET;
6536 	err = ufs_acl_set(ip, vsap, flag, cr);
6537 	ip->i_flag &= ~IQUIET;
6538 	rw_exit(&ip->i_contents);
6539 
6540 out:
6541 	if (ulp) {
6542 		if (donetrans) {
6543 			/*
6544 			 * top_end_async() can eventually call
6545 			 * top_end_sync(), which can block. We must
6546 			 * therefore observe the lock-ordering protocol
6547 			 * here as well.
6548 			 */
6549 			if (vp->v_type == VDIR) {
6550 				rw_exit(&ip->i_rwlock);
6551 				haverwlock = 0;
6552 			}
6553 			TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size);
6554 		}
6555 		ufs_lockfs_end(ulp);
6556 	}
6557 	/*
6558 	 * If no inodes available, try scaring a logically-
6559 	 * free one out of the delete queue to someplace
6560 	 * that we can find it.
6561 	 */
6562 	if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
6563 		ufs_delete_drain_wait(ufsvfsp, 1);
6564 		retry = 0;
6565 		if (vp->v_type == VDIR && haverwlock) {
6566 			rw_exit(&ip->i_rwlock);
6567 			haverwlock = 0;
6568 		}
6569 		goto again;
6570 	}
6571 	/*
6572 	 * If we need to reacquire the lock then it is safe to do so
6573 	 * as a reader. This is because ufs_rwunlock(), which will be
6574 	 * called by our caller after we return, does not differentiate
6575 	 * between shared and exclusive locks.
6576 	 */
6577 	if (!haverwlock) {
6578 		ASSERT(vp->v_type == VDIR);
6579 		rw_enter(&ip->i_rwlock, RW_READER);
6580 	}
6581 
6582 	return (err);
6583 }
6584 
6585 /*
6586  * Locate the vnode to be used for an event notification. As this will
6587  * be called prior to the name space change perform basic verification
6588  * that the change will be allowed.
6589  */
6590 
6591 static int
6592 ufs_eventlookup(struct vnode *dvp, char *nm, struct cred *cr,
6593     struct vnode **vpp)
6594 {
6595 	int	namlen;
6596 	int	error;
6597 	struct vnode	*vp;
6598 	struct inode	*ip;
6599 	struct inode	*xip;
6600 	struct ufsvfs	*ufsvfsp;
6601 	struct ulockfs	*ulp;
6602 
6603 	ip = VTOI(dvp);
6604 	*vpp = NULL;
6605 
6606 	if ((namlen = strlen(nm)) == 0)
6607 		return (EINVAL);
6608 
6609 	if (nm[0] == '.') {
6610 		if (namlen == 1)
6611 			return (EINVAL);
6612 		else if ((namlen == 2) && nm[1] == '.') {
6613 			return (EEXIST);
6614 		}
6615 	}
6616 
6617 	/*
6618 	 * Check accessibility and write access of parent directory as we
6619 	 * only want to post the event if we're able to make a change.
6620 	 */
6621 	if (error = ufs_diraccess(ip, IEXEC|IWRITE, cr))
6622 		return (error);
6623 
6624 	if (vp = dnlc_lookup(dvp, nm)) {
6625 		if (vp == DNLC_NO_VNODE) {
6626 			VN_RELE(vp);
6627 			return (ENOENT);
6628 		}
6629 
6630 		*vpp = vp;
6631 		return (0);
6632 	}
6633 
6634 	/*
6635 	 * Keep the idle queue from getting too long by idling two
6636 	 * inodes before attempting to allocate another.
6637 	 * This operation must be performed before entering lockfs
6638 	 * or a transaction.
6639 	 */
6640 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
6641 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
6642 			ins.in_lidles.value.ul += ufs_lookup_idle_count;
6643 			ufs_idle_some(ufs_lookup_idle_count);
6644 		}
6645 
6646 	ufsvfsp = ip->i_ufsvfs;
6647 
6648 retry_lookup:
6649 	if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK))
6650 		return (error);
6651 
6652 	if ((error = ufs_dirlook(ip, nm, &xip, cr, 1, 1)) == 0) {
6653 		vp = ITOV(xip);
6654 		*vpp = vp;
6655 	}
6656 
6657 	if (ulp) {
6658 		ufs_lockfs_end(ulp);
6659 	}
6660 
6661 	if (error == EAGAIN)
6662 		goto retry_lookup;
6663 
6664 	return (error);
6665 }
6666