xref: /titanic_41/usr/src/uts/common/fs/ufs/ufs_vnops.c (revision 4bd2082ff2d009263265d7de938de336894b6009)
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  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * Portions of this source code were derived from Berkeley 4.3 BSD
31  * under license from the Regents of the University of California.
32  */
33 
34 #pragma ident	"%Z%%M%	%I%	%E% SMI"
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/vtrace.h>
67 #include <sys/filio.h>
68 #include <sys/policy.h>
69 
70 #include <sys/fs/ufs_fs.h>
71 #include <sys/fs/ufs_lockfs.h>
72 #include <sys/fs/ufs_filio.h>
73 #include <sys/fs/ufs_inode.h>
74 #include <sys/fs/ufs_fsdir.h>
75 #include <sys/fs/ufs_quota.h>
76 #include <sys/fs/ufs_log.h>
77 #include <sys/fs/ufs_snap.h>
78 #include <sys/fs/ufs_trans.h>
79 #include <sys/fs/ufs_panic.h>
80 #include <sys/fs/ufs_bio.h>
81 #include <sys/dirent.h>		/* must be AFTER <sys/fs/fsdir.h>! */
82 #include <sys/errno.h>
83 #include <sys/fssnap_if.h>
84 #include <sys/unistd.h>
85 #include <sys/sunddi.h>
86 
87 #include <sys/filio.h>		/* _FIOIO */
88 
89 #include <vm/hat.h>
90 #include <vm/page.h>
91 #include <vm/pvn.h>
92 #include <vm/as.h>
93 #include <vm/seg.h>
94 #include <vm/seg_map.h>
95 #include <vm/seg_vn.h>
96 #include <vm/seg_kmem.h>
97 #include <vm/rm.h>
98 #include <sys/swap.h>
99 
100 #include <fs/fs_subr.h>
101 
102 static struct instats ins;
103 
104 static 	int ufs_getpage_ra(struct vnode *, u_offset_t, struct seg *, caddr_t);
105 static	int ufs_getpage_miss(struct vnode *, u_offset_t, size_t, struct seg *,
106 		caddr_t, struct page **, size_t, enum seg_rw, int);
107 static	int ufs_open(struct vnode **, int, struct cred *);
108 static	int ufs_close(struct vnode *, int, int, offset_t, struct cred *);
109 static	int ufs_read(struct vnode *, struct uio *, int, struct cred *,
110 			struct caller_context *);
111 static	int ufs_write(struct vnode *, struct uio *, int, struct cred *,
112 			struct caller_context *);
113 static	int ufs_ioctl(struct vnode *, int, intptr_t, int, struct cred *, int *);
114 static	int ufs_getattr(struct vnode *, struct vattr *, int, struct cred *);
115 static	int ufs_setattr(struct vnode *, struct vattr *, int, struct cred *,
116 			caller_context_t *);
117 static	int ufs_access(struct vnode *, int, int, struct cred *);
118 static	int ufs_lookup(struct vnode *, char *, struct vnode **,
119 		struct pathname *, int, struct vnode *, struct cred *);
120 static	int ufs_create(struct vnode *, char *, struct vattr *, enum vcexcl,
121 			int, struct vnode **, struct cred *, int);
122 static	int ufs_remove(struct vnode *, char *, struct cred *);
123 static	int ufs_link(struct vnode *, struct vnode *, char *, struct cred *);
124 static	int ufs_rename(struct vnode *, char *, struct vnode *, char *,
125 			struct cred *);
126 static	int ufs_mkdir(struct vnode *, char *, struct vattr *, struct vnode **,
127 			struct cred *);
128 static	int ufs_rmdir(struct vnode *, char *, struct vnode *, struct cred *);
129 static	int ufs_readdir(struct vnode *, struct uio *, struct cred *, int *);
130 static	int ufs_symlink(struct vnode *, char *, struct vattr *, char *,
131 			struct cred *);
132 static	int ufs_readlink(struct vnode *, struct uio *, struct cred *);
133 static	int ufs_fsync(struct vnode *, int, struct cred *);
134 static	void ufs_inactive(struct vnode *, struct cred *);
135 static	int ufs_fid(struct vnode *, struct fid *);
136 static	int ufs_rwlock(struct vnode *, int, caller_context_t *);
137 static	void ufs_rwunlock(struct vnode *, int, caller_context_t *);
138 static	int ufs_seek(struct vnode *, offset_t, offset_t *);
139 static	int ufs_frlock(struct vnode *, int, struct flock64 *, int, offset_t,
140 			struct flk_callback *, struct cred *);
141 static  int ufs_space(struct vnode *, int, struct flock64 *, int, offset_t,
142 		cred_t *, caller_context_t *);
143 static	int ufs_getpage(struct vnode *, offset_t, size_t, uint_t *,
144 		struct page **, size_t, struct seg *, caddr_t,
145 		enum seg_rw, struct cred *);
146 static	int ufs_putpage(struct vnode *, offset_t, size_t, int, struct cred *);
147 static	int ufs_putpages(struct vnode *, offset_t, size_t, int, struct cred *);
148 static	int ufs_map(struct vnode *, offset_t, struct as *, caddr_t *, size_t,
149 			uchar_t, uchar_t, uint_t, struct cred *);
150 static	int ufs_addmap(struct vnode *, offset_t, struct as *, caddr_t,  size_t,
151 			uchar_t, uchar_t, uint_t, struct cred *);
152 static	int ufs_delmap(struct vnode *, offset_t, struct as *, caddr_t,  size_t,
153 			uint_t, uint_t, uint_t, struct cred *);
154 static	int ufs_poll(vnode_t *, short, int, short *, struct pollhead **);
155 static	int ufs_dump(vnode_t *, caddr_t, int, int);
156 static	int ufs_l_pathconf(struct vnode *, int, ulong_t *, struct cred *);
157 static	int ufs_pageio(struct vnode *, struct page *, u_offset_t, size_t, int,
158 			struct cred *);
159 static	int ufs_dump(vnode_t *, caddr_t, int, int);
160 static	int ufs_dumpctl(vnode_t *, int, int *);
161 static	daddr32_t *save_dblks(struct inode *, struct ufsvfs *, daddr32_t *,
162 			daddr32_t *, int, int);
163 static	int ufs_getsecattr(struct vnode *, vsecattr_t *, int, struct cred *);
164 static	int ufs_setsecattr(struct vnode *, vsecattr_t *, int, struct cred *);
165 
166 extern int as_map_locked(struct as *, caddr_t, size_t, int ((*)()), void *);
167 
168 /*
169  * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
170  *
171  * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
172  */
173 struct vnodeops *ufs_vnodeops;
174 
175 /* NOTE: "not blkd" below  means that the operation isn't blocked by lockfs */
176 const fs_operation_def_t ufs_vnodeops_template[] = {
177 	VOPNAME_OPEN,		{ .vop_open = ufs_open },	/* not blkd */
178 	VOPNAME_CLOSE,		{ .vop_close = ufs_close },	/* not blkd */
179 	VOPNAME_READ,		{ .vop_read = ufs_read },
180 	VOPNAME_WRITE,		{ .vop_write = ufs_write },
181 	VOPNAME_IOCTL,		{ .vop_ioctl = ufs_ioctl },
182 	VOPNAME_GETATTR,	{ .vop_getattr = ufs_getattr },
183 	VOPNAME_SETATTR,	{ .vop_setattr = ufs_setattr },
184 	VOPNAME_ACCESS,		{ .vop_access = ufs_access },
185 	VOPNAME_LOOKUP,		{ .vop_lookup = ufs_lookup },
186 	VOPNAME_CREATE,		{ .vop_create = ufs_create },
187 	VOPNAME_REMOVE,		{ .vop_remove = ufs_remove },
188 	VOPNAME_LINK,		{ .vop_link = ufs_link },
189 	VOPNAME_RENAME,		{ .vop_rename = ufs_rename },
190 	VOPNAME_MKDIR,		{ .vop_mkdir = ufs_mkdir },
191 	VOPNAME_RMDIR,		{ .vop_rmdir = ufs_rmdir },
192 	VOPNAME_READDIR,	{ .vop_readdir = ufs_readdir },
193 	VOPNAME_SYMLINK,	{ .vop_symlink = ufs_symlink },
194 	VOPNAME_READLINK,	{ .vop_readlink = ufs_readlink },
195 	VOPNAME_FSYNC,		{ .vop_fsync = ufs_fsync },
196 	VOPNAME_INACTIVE,	{ .vop_inactive = ufs_inactive }, /* not blkd */
197 	VOPNAME_FID,		{ .vop_fid = ufs_fid },
198 	VOPNAME_RWLOCK,		{ .vop_rwlock = ufs_rwlock },	/* not blkd */
199 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = ufs_rwunlock }, /* not blkd */
200 	VOPNAME_SEEK,		{ .vop_seek = ufs_seek },
201 	VOPNAME_FRLOCK,		{ .vop_frlock = ufs_frlock },
202 	VOPNAME_SPACE,		{ .vop_space = ufs_space },
203 	VOPNAME_GETPAGE,	{ .vop_getpage = ufs_getpage },
204 	VOPNAME_PUTPAGE,	{ .vop_putpage = ufs_putpage },
205 	VOPNAME_MAP,		{ .vop_map = ufs_map },
206 	VOPNAME_ADDMAP,		{ .vop_addmap = ufs_addmap },	/* not blkd */
207 	VOPNAME_DELMAP,		{ .vop_delmap = ufs_delmap },	/* not blkd */
208 	VOPNAME_POLL,		{ .vop_poll = ufs_poll },	/* not blkd */
209 	VOPNAME_DUMP,		{ .vop_dump = ufs_dump },
210 	VOPNAME_PATHCONF,	{ .vop_pathconf = ufs_l_pathconf },
211 	VOPNAME_PAGEIO,		{ .vop_pageio = ufs_pageio },
212 	VOPNAME_DUMPCTL,	{ .vop_dumpctl = ufs_dumpctl },
213 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = ufs_getsecattr },
214 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = ufs_setsecattr },
215 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
216 	NULL,			NULL
217 };
218 
219 #define	MAX_BACKFILE_COUNT	9999
220 
221 /*
222  * Created by ufs_dumpctl() to store a file's disk block info into memory.
223  * Used by ufs_dump() to dump data to disk directly.
224  */
225 struct dump {
226 	struct inode	*ip;		/* the file we contain */
227 	daddr_t		fsbs;		/* number of blocks stored */
228 	struct timeval32 time;		/* time stamp for the struct */
229 	daddr32_t 	dblk[1];	/* place holder for block info */
230 };
231 
232 static struct dump *dump_info = NULL;
233 
234 /*
235  * Previously there was no special action required for ordinary files.
236  * (Devices are handled through the device file system.)
237  * Now we support Large Files and Large File API requires open to
238  * fail if file is large.
239  * We could take care to prevent data corruption
240  * by doing an atomic check of size and truncate if file is opened with
241  * FTRUNC flag set but traditionally this is being done by the vfs/vnode
242  * layers. So taking care of truncation here is a change in the existing
243  * semantics of VOP_OPEN and therefore we chose not to implement any thing
244  * here. The check for the size of the file > 2GB is being done at the
245  * vfs layer in routine vn_open().
246  */
247 
248 /* ARGSUSED */
249 static int
250 ufs_open(struct vnode **vpp, int flag, struct cred *cr)
251 {
252 	TRACE_1(TR_FAC_UFS, TR_UFS_OPEN, "ufs_open:vpp %p", vpp);
253 	return (0);
254 }
255 
256 /*ARGSUSED*/
257 static int
258 ufs_close(struct vnode *vp, int flag, int count, offset_t offset,
259 	struct cred *cr)
260 {
261 	TRACE_1(TR_FAC_UFS, TR_UFS_CLOSE, "ufs_close:vp %p", vp);
262 
263 	cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
264 	cleanshares(vp, ttoproc(curthread)->p_pid);
265 
266 	/*
267 	 * Push partially filled cluster at last close.
268 	 * ``last close'' is approximated because the dnlc
269 	 * may have a hold on the vnode.
270 	 * Checking for VBAD here will also act as a forced umount check.
271 	 */
272 	if (vp->v_count <= 2 && vp->v_type != VBAD) {
273 		struct inode *ip = VTOI(vp);
274 		if (ip->i_delaylen) {
275 			ins.in_poc.value.ul++;
276 			(void) ufs_putpages(vp, ip->i_delayoff, ip->i_delaylen,
277 					B_ASYNC | B_FREE, cr);
278 			ip->i_delaylen = 0;
279 		}
280 	}
281 
282 	return (0);
283 }
284 
285 /*ARGSUSED*/
286 static int
287 ufs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cr,
288 	struct caller_context *ct)
289 {
290 	struct inode *ip = VTOI(vp);
291 	struct ufsvfs *ufsvfsp;
292 	struct ulockfs *ulp = NULL;
293 	int error = 0;
294 	int intrans = 0;
295 
296 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
297 	TRACE_3(TR_FAC_UFS, TR_UFS_READ_START,
298 		"ufs_read_start:vp %p uiop %p ioflag %x",
299 		vp, uiop, ioflag);
300 
301 	/*
302 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
303 	 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
304 	 */
305 	if (MANDLOCK(vp, ip->i_mode)) {
306 		/*
307 		 * ufs_getattr ends up being called by chklock
308 		 */
309 		error = chklock(vp, FREAD, uiop->uio_loffset,
310 				uiop->uio_resid, uiop->uio_fmode, ct);
311 		if (error)
312 			goto out;
313 	}
314 
315 	ufsvfsp = ip->i_ufsvfs;
316 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READ_MASK);
317 	if (error)
318 		goto out;
319 
320 	/*
321 	 * In the case that a directory is opened for reading as a file
322 	 * (eg "cat .") with the  O_RSYNC, O_SYNC and O_DSYNC flags set.
323 	 * The locking order had to be changed to avoid a deadlock with
324 	 * an update taking place on that directory at the same time.
325 	 */
326 	if ((ip->i_mode & IFMT) == IFDIR) {
327 
328 		rw_enter(&ip->i_contents, RW_READER);
329 		error = rdip(ip, uiop, ioflag, cr);
330 		rw_exit(&ip->i_contents);
331 
332 		if (error) {
333 			if (ulp)
334 				ufs_lockfs_end(ulp);
335 			goto out;
336 		}
337 
338 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
339 		    TRANS_ISTRANS(ufsvfsp)) {
340 			rw_exit(&ip->i_rwlock);
341 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
342 			    error);
343 			ASSERT(!error);
344 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
345 			    TOP_READ_SIZE);
346 			rw_enter(&ip->i_rwlock, RW_READER);
347 		}
348 	} else {
349 		/*
350 		 * Only transact reads to files opened for sync-read and
351 		 * sync-write on a file system that is not write locked.
352 		 *
353 		 * The ``not write locked'' check prevents problems with
354 		 * enabling/disabling logging on a busy file system.  E.g.,
355 		 * logging exists at the beginning of the read but does not
356 		 * at the end.
357 		 *
358 		 */
359 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
360 		    TRANS_ISTRANS(ufsvfsp)) {
361 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
362 			    error);
363 			ASSERT(!error);
364 			intrans = 1;
365 		}
366 
367 		rw_enter(&ip->i_contents, RW_READER);
368 		error = rdip(ip, uiop, ioflag, cr);
369 		rw_exit(&ip->i_contents);
370 
371 		if (intrans) {
372 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
373 			    TOP_READ_SIZE);
374 		}
375 	}
376 
377 	if (ulp) {
378 		ufs_lockfs_end(ulp);
379 	}
380 out:
381 
382 	TRACE_2(TR_FAC_UFS, TR_UFS_READ_END,
383 		"ufs_read_end:vp %p error %d", vp, error);
384 	return (error);
385 }
386 
387 extern	int	ufs_HW;		/* high water mark */
388 extern	int	ufs_LW;		/* low water mark */
389 int	ufs_WRITES = 1;		/* XXX - enable/disable */
390 int	ufs_throttles = 0;	/* throttling count */
391 int	ufs_allow_shared_writes = 1;	/* directio shared writes */
392 
393 static int
394 ufs_check_rewrite(struct inode *ip, struct uio *uiop, int ioflag)
395 {
396 	int	shared_write;
397 
398 	/*
399 	 * If the FDSYNC flag is set then ignore the global
400 	 * ufs_allow_shared_writes in this case.
401 	 */
402 	shared_write = (ioflag & FDSYNC) | ufs_allow_shared_writes;
403 
404 	/*
405 	 * Filter to determine if this request is suitable as a
406 	 * concurrent rewrite. This write must not allocate blocks
407 	 * by extending the file or filling in holes. No use trying
408 	 * through FSYNC descriptors as the inode will be synchronously
409 	 * updated after the write. The uio structure has not yet been
410 	 * checked for sanity, so assume nothing.
411 	 */
412 	return (((ip->i_mode & IFMT) == IFREG) && !(ioflag & FAPPEND) &&
413 		(uiop->uio_loffset >= (offset_t)0) &&
414 		(uiop->uio_loffset < ip->i_size) && (uiop->uio_resid > 0) &&
415 		((ip->i_size - uiop->uio_loffset) >= uiop->uio_resid) &&
416 		!(ioflag & FSYNC) && !bmap_has_holes(ip) &&
417 		shared_write);
418 }
419 
420 /*ARGSUSED*/
421 static int
422 ufs_write(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cr,
423 	caller_context_t *ct)
424 {
425 	struct inode *ip = VTOI(vp);
426 	struct ufsvfs *ufsvfsp;
427 	struct ulockfs *ulp;
428 	int retry = 1;
429 	int error, resv, resid = 0;
430 	int directio_status;
431 	int exclusive;
432 	int rewriteflg;
433 	long start_resid = uiop->uio_resid;
434 
435 	TRACE_3(TR_FAC_UFS, TR_UFS_WRITE_START,
436 		"ufs_write_start:vp %p uiop %p ioflag %x",
437 		vp, uiop, ioflag);
438 
439 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
440 
441 retry_mandlock:
442 	/*
443 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
444 	 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
445 	 * Check for forced unmounts normally done in ufs_lockfs_begin().
446 	 */
447 	if ((ufsvfsp = ip->i_ufsvfs) == NULL) {
448 		error = EIO;
449 		goto out;
450 	}
451 	if (MANDLOCK(vp, ip->i_mode)) {
452 
453 		ASSERT(RW_WRITE_HELD(&ip->i_rwlock));
454 
455 		/*
456 		 * ufs_getattr ends up being called by chklock
457 		 */
458 		error = chklock(vp, FWRITE, uiop->uio_loffset,
459 				uiop->uio_resid, uiop->uio_fmode, ct);
460 		if (error)
461 			goto out;
462 	}
463 
464 	/* i_rwlock can change in chklock */
465 	exclusive = rw_write_held(&ip->i_rwlock);
466 	rewriteflg = ufs_check_rewrite(ip, uiop, ioflag);
467 
468 	/*
469 	 * Check for fast-path special case of directio re-writes.
470 	 */
471 	if ((ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) &&
472 	    !exclusive && rewriteflg) {
473 
474 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
475 		if (error)
476 			goto out;
477 
478 		rw_enter(&ip->i_contents, RW_READER);
479 		error = ufs_directio_write(ip, uiop, ioflag, 1, cr,
480 			&directio_status);
481 		if (directio_status == DIRECTIO_SUCCESS) {
482 			uint_t i_flag_save;
483 
484 			if (start_resid != uiop->uio_resid)
485 				error = 0;
486 			/*
487 			 * Special treatment of access times for re-writes.
488 			 * If IMOD is not already set, then convert it
489 			 * to IMODACC for this operation. This defers
490 			 * entering a delta into the log until the inode
491 			 * is flushed. This mimics what is done for read
492 			 * operations and inode access time.
493 			 */
494 			mutex_enter(&ip->i_tlock);
495 			i_flag_save = ip->i_flag;
496 			ip->i_flag |= IUPD | ICHG;
497 			ip->i_seq++;
498 			ITIMES_NOLOCK(ip);
499 			if ((i_flag_save & IMOD) == 0) {
500 				ip->i_flag &= ~IMOD;
501 				ip->i_flag |= IMODACC;
502 			}
503 			mutex_exit(&ip->i_tlock);
504 			rw_exit(&ip->i_contents);
505 			if (ulp)
506 				ufs_lockfs_end(ulp);
507 			goto out;
508 		}
509 		rw_exit(&ip->i_contents);
510 		if (ulp)
511 			ufs_lockfs_end(ulp);
512 	}
513 
514 	if (!exclusive && !rw_tryupgrade(&ip->i_rwlock)) {
515 		rw_exit(&ip->i_rwlock);
516 		rw_enter(&ip->i_rwlock, RW_WRITER);
517 		/*
518 		 * Mandatory locking could have been enabled
519 		 * after dropping the i_rwlock.
520 		 */
521 		if (MANDLOCK(vp, ip->i_mode))
522 			goto retry_mandlock;
523 	}
524 
525 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
526 	if (error)
527 		goto out;
528 
529 	/*
530 	 * Amount of log space needed for this write
531 	 */
532 	if (!rewriteflg || !(ioflag & FDSYNC))
533 		TRANS_WRITE_RESV(ip, uiop, ulp, &resv, &resid);
534 
535 	/*
536 	 * Throttle writes.
537 	 */
538 	if (ufs_WRITES && (ip->i_writes > ufs_HW)) {
539 		mutex_enter(&ip->i_tlock);
540 		while (ip->i_writes > ufs_HW) {
541 			ufs_throttles++;
542 			cv_wait(&ip->i_wrcv, &ip->i_tlock);
543 		}
544 		mutex_exit(&ip->i_tlock);
545 	}
546 
547 	/*
548 	 * Enter Transaction
549 	 *
550 	 * If the write is a rewrite there is no need to open a transaction
551 	 * if the FDSYNC flag is set and not the FSYNC.  In this case just
552 	 * set the IMODACC flag to modify do the update at a later time
553 	 * thus avoiding the overhead of the logging transaction that is
554 	 * not required.
555 	 */
556 	if (ioflag & (FSYNC|FDSYNC)) {
557 		if (ulp) {
558 			if (rewriteflg) {
559 				uint_t i_flag_save;
560 
561 				rw_enter(&ip->i_contents, RW_READER);
562 				mutex_enter(&ip->i_tlock);
563 				i_flag_save = ip->i_flag;
564 				ip->i_flag |= IUPD | ICHG;
565 				ip->i_seq++;
566 				ITIMES_NOLOCK(ip);
567 				if ((i_flag_save & IMOD) == 0) {
568 					ip->i_flag &= ~IMOD;
569 					ip->i_flag |= IMODACC;
570 				}
571 				mutex_exit(&ip->i_tlock);
572 				rw_exit(&ip->i_contents);
573 			} else {
574 				int terr = 0;
575 				TRANS_BEGIN_SYNC(ufsvfsp, TOP_WRITE_SYNC, resv,
576 				    terr);
577 				ASSERT(!terr);
578 			}
579 		}
580 	} else {
581 		if (ulp)
582 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_WRITE, resv);
583 	}
584 
585 	/*
586 	 * Write the file
587 	 */
588 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
589 	rw_enter(&ip->i_contents, RW_WRITER);
590 	if ((ioflag & FAPPEND) != 0 && (ip->i_mode & IFMT) == IFREG) {
591 		/*
592 		 * In append mode start at end of file.
593 		 */
594 		uiop->uio_loffset = ip->i_size;
595 	}
596 
597 	/*
598 	 * Mild optimisation, don't call ufs_trans_write() unless we have to
599 	 * Also, suppress file system full messages if we will retry.
600 	 */
601 	if (retry)
602 		ip->i_flag |= IQUIET;
603 	if (resid) {
604 		TRANS_WRITE(ip, uiop, ioflag, error, ulp, cr, resv, resid);
605 	} else {
606 		error = wrip(ip, uiop, ioflag, cr);
607 	}
608 	ip->i_flag &= ~IQUIET;
609 
610 	rw_exit(&ip->i_contents);
611 	rw_exit(&ufsvfsp->vfs_dqrwlock);
612 
613 	/*
614 	 * Leave Transaction
615 	 */
616 	if (ulp) {
617 		if (ioflag & (FSYNC|FDSYNC)) {
618 			if (!rewriteflg) {
619 				int terr = 0;
620 
621 				TRANS_END_SYNC(ufsvfsp, terr, TOP_WRITE_SYNC,
622 					resv);
623 				if (error == 0)
624 					error = terr;
625 			}
626 		} else {
627 			TRANS_END_ASYNC(ufsvfsp, TOP_WRITE, resv);
628 		}
629 		ufs_lockfs_end(ulp);
630 	}
631 out:
632 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
633 		/*
634 		 * Any blocks tied up in pending deletes?
635 		 */
636 		ufs_delete_drain_wait(ufsvfsp, 1);
637 		retry = 0;
638 		goto retry_mandlock;
639 	}
640 
641 	if (error == ENOSPC && (start_resid != uiop->uio_resid))
642 		error = 0;
643 
644 	TRACE_2(TR_FAC_UFS, TR_UFS_WRITE_END,
645 		"ufs_write_end:vp %p error %d", vp, error);
646 	return (error);
647 }
648 
649 /*
650  * Don't cache write blocks to files with the sticky bit set.
651  * Used to keep swap files from blowing the page cache on a server.
652  */
653 int stickyhack = 1;
654 
655 /*
656  * Free behind hacks.  The pager is busted.
657  * XXX - need to pass the information down to writedone() in a flag like B_SEQ
658  * or B_FREE_IF_TIGHT_ON_MEMORY.
659  */
660 int	freebehind = 1;
661 int	smallfile = 0;
662 u_offset_t smallfile64 = 32 * 1024;
663 
664 /*
665  * While we should, in most cases, cache the pages for write, we
666  * may also want to cache the pages for read as long as they are
667  * frequently re-usable.
668  *
669  * If cache_read_ahead = 1, the pages for read will go to the tail
670  * of the cache list when they are released, otherwise go to the head.
671  */
672 int	cache_read_ahead = 0;
673 
674 /*
675  * Freebehind exists  so that as we read  large files  sequentially we
676  * don't consume most of memory with pages  from a few files. It takes
677  * longer to re-read from disk multiple small files as it does reading
678  * one large one sequentially.  As system  memory grows customers need
679  * to retain bigger chunks   of files in  memory.   The advent of  the
680  * cachelist opens up of the possibility freeing pages  to the head or
681  * tail of the list.
682  *
683  * Not freeing a page is a bet that the page will be read again before
684  * it's segmap slot is needed for something else. If we loose the bet,
685  * it means some  other thread is  burdened with the  page free we did
686  * not do. If we win we save a free and reclaim.
687  *
688  * Freeing it at the tail  vs the head of cachelist  is a bet that the
689  * page will survive until the next  read.  It's also saying that this
690  * page is more likely to  be re-used than a  page freed some time ago
691  * and never reclaimed.
692  *
693  * Freebehind maintains a  range of  file offset [smallfile1; smallfile2]
694  *
695  *            0 < offset < smallfile1 : pages are not freed.
696  *   smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
697  *   smallfile2 < offset              : pages freed to head of cachelist.
698  *
699  * The range  is  computed  at most  once  per second  and  depends on
700  * freemem  and  ncpus_online.  Both parameters  are   bounded to be
701  * >= smallfile && >= smallfile64.
702  *
703  * smallfile1 = (free memory / ncpu) / 1000
704  * smallfile2 = (free memory / ncpu) / 10
705  *
706  * A few examples values:
707  *
708  *       Free Mem (in Bytes) [smallfile1; smallfile2]  [smallfile1; smallfile2]
709  *                                 ncpus_online = 4          ncpus_online = 64
710  *       ------------------  -----------------------   -----------------------
711  *             1G                   [256K;  25M]               [32K; 1.5M]
712  *            10G                   [2.5M; 250M]              [156K; 15M]
713  *           100G                    [25M; 2.5G]              [1.5M; 150M]
714  *
715  */
716 
717 #define	SMALLFILE1_D 1000
718 #define	SMALLFILE2_D 10
719 static u_offset_t smallfile1 = 32 * 1024;
720 static u_offset_t smallfile2 = 32 * 1024;
721 static clock_t smallfile_update = 0; /* lbolt value of when to recompute */
722 uint_t smallfile1_d = SMALLFILE1_D;
723 uint_t smallfile2_d = SMALLFILE2_D;
724 
725 /*
726  * wrip does the real work of write requests for ufs.
727  */
728 int
729 wrip(struct inode *ip, struct uio *uio, int ioflag, struct cred *cr)
730 {
731 	rlim64_t limit = uio->uio_llimit;
732 	u_offset_t off;
733 	u_offset_t old_i_size;
734 	struct fs *fs;
735 	struct vnode *vp;
736 	struct ufsvfs *ufsvfsp;
737 	caddr_t base;
738 	long start_resid = uio->uio_resid;	/* save starting resid */
739 	long premove_resid;			/* resid before uiomove() */
740 	uint_t flags;
741 	int newpage;
742 	int iupdat_flag, directio_status;
743 	int n, on, mapon;
744 	int error, pagecreate;
745 	int do_dqrwlock;		/* drop/reacquire vfs_dqrwlock */
746 	int32_t	iblocks;
747 	int	new_iblocks;
748 
749 	/*
750 	 * ip->i_size is incremented before the uiomove
751 	 * is done on a write.  If the move fails (bad user
752 	 * address) reset ip->i_size.
753 	 * The better way would be to increment ip->i_size
754 	 * only if the uiomove succeeds.
755 	 */
756 	int i_size_changed = 0;
757 	o_mode_t type;
758 	int i_seq_needed = 0;
759 
760 	vp = ITOV(ip);
761 
762 	/*
763 	 * check for forced unmount - should not happen as
764 	 * the request passed the lockfs checks.
765 	 */
766 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
767 		return (EIO);
768 
769 	fs = ip->i_fs;
770 
771 	TRACE_1(TR_FAC_UFS, TR_UFS_RWIP_START,
772 		"ufs_wrip_start:vp %p", vp);
773 
774 	ASSERT(RW_WRITE_HELD(&ip->i_contents));
775 
776 	/* check for valid filetype */
777 	type = ip->i_mode & IFMT;
778 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
779 	    (type != IFLNK) && (type != IFSHAD)) {
780 		return (EIO);
781 	}
782 
783 	/*
784 	 * the actual limit of UFS file size
785 	 * is UFS_MAXOFFSET_T
786 	 */
787 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
788 		limit = MAXOFFSET_T;
789 
790 	if (uio->uio_loffset >= limit) {
791 		proc_t *p = ttoproc(curthread);
792 
793 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
794 			"ufs_wrip_end:vp %p error %d", vp, EINVAL);
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 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
814 			"ufs_wrip_end:vp %p error %d", vp, EINVAL);
815 		return (EINVAL);
816 	}
817 	if (uio->uio_resid == 0) {
818 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
819 			"ufs_wrip_end:vp %p error %d", vp, 0);
820 		return (0);
821 	}
822 
823 	if (uio->uio_loffset >= limit)
824 		return (EFBIG);
825 
826 	ip->i_flag |= INOACC;	/* don't update ref time in getpage */
827 
828 	if (ioflag & (FSYNC|FDSYNC)) {
829 		ip->i_flag |= ISYNC;
830 		iupdat_flag = 1;
831 	}
832 	/*
833 	 * Try to go direct
834 	 */
835 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
836 		uio->uio_llimit = limit;
837 		error = ufs_directio_write(ip, uio, ioflag, 0, cr,
838 			&directio_status);
839 		/*
840 		 * If ufs_directio wrote to the file or set the flags,
841 		 * we need to update i_seq, but it may be deferred.
842 		 */
843 		if (start_resid != uio->uio_resid ||
844 					(ip->i_flag & (ICHG|IUPD))) {
845 			i_seq_needed = 1;
846 			ip->i_flag |= ISEQ;
847 		}
848 		if (directio_status == DIRECTIO_SUCCESS)
849 			goto out;
850 	}
851 
852 	/*
853 	 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
854 	 *
855 	 * o shadow inodes: vfs_dqrwlock is not held at all
856 	 * o quota updates: vfs_dqrwlock is read or write held
857 	 * o other updates: vfs_dqrwlock is read held
858 	 *
859 	 * The first case is the only one where we do not hold
860 	 * vfs_dqrwlock at all while entering wrip().
861 	 * We must make sure not to downgrade/drop vfs_dqrwlock if we
862 	 * have it as writer, i.e. if we are updating the quota inode.
863 	 * There is no potential deadlock scenario in this case as
864 	 * ufs_getpage() takes care of this and avoids reacquiring
865 	 * vfs_dqrwlock in that case.
866 	 *
867 	 * This check is done here since the above conditions do not change
868 	 * and we possibly loop below, so save a few cycles.
869 	 */
870 	if ((type == IFSHAD) ||
871 		(rw_owner(&ufsvfsp->vfs_dqrwlock) == curthread)) {
872 			do_dqrwlock = 0;
873 	} else {
874 		do_dqrwlock = 1;
875 	}
876 
877 	/*
878 	 * Large Files: We cast MAXBMASK to offset_t
879 	 * inorder to mask out the higher bits. Since offset_t
880 	 * is a signed value, the high order bit set in MAXBMASK
881 	 * value makes it do the right thing by having all bits 1
882 	 * in the higher word. May be removed for _SOLARIS64_.
883 	 */
884 
885 	fs = ip->i_fs;
886 	do {
887 		u_offset_t uoff = uio->uio_loffset;
888 		off = uoff & (offset_t)MAXBMASK;
889 		mapon = (int)(uoff & (offset_t)MAXBOFFSET);
890 		on = (int)blkoff(fs, uoff);
891 		n = (int)MIN(fs->fs_bsize - on, uio->uio_resid);
892 		new_iblocks = 1;
893 
894 		if (type == IFREG && uoff + n >= limit) {
895 			if (uoff >= limit) {
896 				error = EFBIG;
897 				goto out;
898 			}
899 			/*
900 			 * since uoff + n >= limit,
901 			 * therefore n >= limit - uoff, and n is an int
902 			 * so it is safe to cast it to an int
903 			 */
904 			n = (int)(limit - (rlim64_t)uoff);
905 		}
906 		if (uoff + n > ip->i_size) {
907 			/*
908 			 * We are extending the length of the file.
909 			 * bmap is used so that we are sure that
910 			 * if we need to allocate new blocks, that it
911 			 * is done here before we up the file size.
912 			 */
913 			error = bmap_write(ip, uoff, (int)(on + n),
914 			    mapon == 0, NULL, cr);
915 			/*
916 			 * bmap_write never drops i_contents so if
917 			 * the flags are set it changed the file.
918 			 */
919 			if (ip->i_flag & (ICHG|IUPD)) {
920 				i_seq_needed = 1;
921 				ip->i_flag |= ISEQ;
922 			}
923 			if (error)
924 				break;
925 			/*
926 			 * There is a window of vulnerability here.
927 			 * The sequence of operations: allocate file
928 			 * system blocks, uiomove the data into pages,
929 			 * and then update the size of the file in the
930 			 * inode, must happen atomically.  However, due
931 			 * to current locking constraints, this can not
932 			 * be done.
933 			 */
934 			ASSERT(ip->i_writer == NULL);
935 			ip->i_writer = curthread;
936 			i_size_changed = 1;
937 			/*
938 			 * If we are writing from the beginning of
939 			 * the mapping, we can just create the
940 			 * pages without having to read them.
941 			 */
942 			pagecreate = (mapon == 0);
943 		} else if (n == MAXBSIZE) {
944 			/*
945 			 * Going to do a whole mappings worth,
946 			 * so we can just create the pages w/o
947 			 * having to read them in.  But before
948 			 * we do that, we need to make sure any
949 			 * needed blocks are allocated first.
950 			 */
951 			iblocks = ip->i_blocks;
952 			error = bmap_write(ip, uoff, (int)(on + n),
953 			    BI_ALLOC_ONLY, NULL, cr);
954 			/*
955 			 * bmap_write never drops i_contents so if
956 			 * the flags are set it changed the file.
957 			 */
958 			if (ip->i_flag & (ICHG|IUPD)) {
959 				i_seq_needed = 1;
960 				ip->i_flag |= ISEQ;
961 			}
962 			if (error)
963 				break;
964 			pagecreate = 1;
965 			/*
966 			 * check if the new created page needed the
967 			 * allocation of new disk blocks.
968 			 */
969 			if (iblocks == ip->i_blocks)
970 				new_iblocks = 0; /* no new blocks allocated */
971 		} else {
972 			pagecreate = 0;
973 			/*
974 			 * In sync mode flush the indirect blocks which
975 			 * may have been allocated and not written on
976 			 * disk. In above cases bmap_write will allocate
977 			 * in sync mode.
978 			 */
979 			if (ioflag & (FSYNC|FDSYNC)) {
980 				error = ufs_indirblk_sync(ip, uoff);
981 				if (error)
982 					break;
983 			}
984 		}
985 
986 		/*
987 		 * At this point we can enter ufs_getpage() in one
988 		 * of two ways:
989 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
990 		 *    forcefault parameter is true (pagecreate == 0)
991 		 * 2) uiomove() causes a page fault.
992 		 *
993 		 * We have to drop the contents lock to prevent the VM
994 		 * system from trying to reaquire it in ufs_getpage()
995 		 * should the uiomove cause a pagefault.
996 		 *
997 		 * We have to drop the reader vfs_dqrwlock here as well.
998 		 */
999 		rw_exit(&ip->i_contents);
1000 		if (do_dqrwlock) {
1001 			ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock));
1002 			ASSERT(!(RW_WRITE_HELD(&ufsvfsp->vfs_dqrwlock)));
1003 			rw_exit(&ufsvfsp->vfs_dqrwlock);
1004 		}
1005 
1006 		newpage = 0;
1007 		premove_resid = uio->uio_resid;
1008 		if (vpm_enable) {
1009 			/*
1010 			 * Copy data. If new pages are created, part of
1011 			 * the page that is not written will be initizliazed
1012 			 * with zeros.
1013 			 */
1014 			error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1015 				uio, !pagecreate, &newpage, 0, S_WRITE);
1016 		} else {
1017 
1018 			base = segmap_getmapflt(segkmap, vp, (off + mapon),
1019 					(uint_t)n, !pagecreate, S_WRITE);
1020 
1021 			/*
1022 			 * segmap_pagecreate() returns 1 if it calls
1023 			 * page_create_va() to allocate any pages.
1024 			 */
1025 
1026 			if (pagecreate)
1027 				newpage = segmap_pagecreate(segkmap, base,
1028 				    (size_t)n, 0);
1029 
1030 			error = uiomove(base + mapon, (long)n, UIO_WRITE, uio);
1031 		}
1032 
1033 		/*
1034 		 * If "newpage" is set, then a new page was created and it
1035 		 * does not contain valid data, so it needs to be initialized
1036 		 * at this point.
1037 		 * Otherwise the page contains old data, which was overwritten
1038 		 * partially or as a whole in uiomove.
1039 		 * If there is only one iovec structure within uio, then
1040 		 * on error uiomove will not be able to update uio->uio_loffset
1041 		 * and we would zero the whole page here!
1042 		 *
1043 		 * If uiomove fails because of an error, the old valid data
1044 		 * is kept instead of filling the rest of the page with zero's.
1045 		 */
1046 		if (!vpm_enable && newpage &&
1047 		    uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) {
1048 			/*
1049 			 * We created pages w/o initializing them completely,
1050 			 * thus we need to zero the part that wasn't set up.
1051 			 * This happens on most EOF write cases and if
1052 			 * we had some sort of error during the uiomove.
1053 			 */
1054 			int nzero, nmoved;
1055 
1056 			nmoved = (int)(uio->uio_loffset - (off + mapon));
1057 			ASSERT(nmoved >= 0 && nmoved <= n);
1058 			nzero = roundup(on + n, PAGESIZE) - nmoved;
1059 			ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE);
1060 			(void) kzero(base + mapon + nmoved, (uint_t)nzero);
1061 		}
1062 
1063 		/*
1064 		 * Unlock the pages allocated by page_create_va()
1065 		 * in segmap_pagecreate()
1066 		 */
1067 		if (!vpm_enable && newpage)
1068 			segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE);
1069 
1070 		/*
1071 		 * If the size of the file changed, then update the
1072 		 * size field in the inode now.  This can't be done
1073 		 * before the call to segmap_pageunlock or there is
1074 		 * a potential deadlock with callers to ufs_putpage().
1075 		 * They will be holding i_contents and trying to lock
1076 		 * a page, while this thread is holding a page locked
1077 		 * and trying to acquire i_contents.
1078 		 */
1079 		if (i_size_changed) {
1080 			rw_enter(&ip->i_contents, RW_WRITER);
1081 			old_i_size = ip->i_size;
1082 			UFS_SET_ISIZE(uoff + n, ip);
1083 			TRANS_INODE(ufsvfsp, ip);
1084 			/*
1085 			 * file has grown larger than 2GB. Set flag
1086 			 * in superblock to indicate this, if it
1087 			 * is not already set.
1088 			 */
1089 			if ((ip->i_size > MAXOFF32_T) &&
1090 			    !(fs->fs_flags & FSLARGEFILES)) {
1091 				ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES);
1092 				mutex_enter(&ufsvfsp->vfs_lock);
1093 				fs->fs_flags |= FSLARGEFILES;
1094 				ufs_sbwrite(ufsvfsp);
1095 				mutex_exit(&ufsvfsp->vfs_lock);
1096 			}
1097 			mutex_enter(&ip->i_tlock);
1098 			ip->i_writer = NULL;
1099 			cv_broadcast(&ip->i_wrcv);
1100 			mutex_exit(&ip->i_tlock);
1101 			rw_exit(&ip->i_contents);
1102 		}
1103 
1104 		if (error) {
1105 			/*
1106 			 * If we failed on a write, we may have already
1107 			 * allocated file blocks as well as pages.  It's
1108 			 * hard to undo the block allocation, but we must
1109 			 * be sure to invalidate any pages that may have
1110 			 * been allocated.
1111 			 *
1112 			 * If the page was created without initialization
1113 			 * then we must check if it should be possible
1114 			 * to destroy the new page and to keep the old data
1115 			 * on the disk.
1116 			 *
1117 			 * It is possible to destroy the page without
1118 			 * having to write back its contents only when
1119 			 * - the size of the file keeps unchanged
1120 			 * - bmap_write() did not allocate new disk blocks
1121 			 *   it is possible to create big files using "seek" and
1122 			 *   write to the end of the file. A "write" to a
1123 			 *   position before the end of the file would not
1124 			 *   change the size of the file but it would allocate
1125 			 *   new disk blocks.
1126 			 * - uiomove intended to overwrite the whole page.
1127 			 * - a new page was created (newpage == 1).
1128 			 */
1129 
1130 			if (i_size_changed == 0 && new_iblocks == 0 &&
1131 			    newpage) {
1132 
1133 				/* unwind what uiomove eventually last did */
1134 				uio->uio_resid = premove_resid;
1135 
1136 				/*
1137 				 * destroy the page, do not write ambiguous
1138 				 * data to the disk.
1139 				 */
1140 				flags = SM_DESTROY;
1141 			} else {
1142 				/*
1143 				 * write the page back to the disk, if dirty,
1144 				 * and remove the page from the cache.
1145 				 */
1146 				flags = SM_INVAL;
1147 			}
1148 
1149 			if (vpm_enable) {
1150 				/*
1151 				 *  Flush pages.
1152 				 */
1153 				(void) vpm_sync_pages(vp, off, n, flags);
1154 			} else {
1155 				(void) segmap_release(segkmap, base, flags);
1156 			}
1157 		} else {
1158 			flags = 0;
1159 			/*
1160 			 * Force write back for synchronous write cases.
1161 			 */
1162 			if ((ioflag & (FSYNC|FDSYNC)) || type == IFDIR) {
1163 				/*
1164 				 * If the sticky bit is set but the
1165 				 * execute bit is not set, we do a
1166 				 * synchronous write back and free
1167 				 * the page when done.  We set up swap
1168 				 * files to be handled this way to
1169 				 * prevent servers from keeping around
1170 				 * the client's swap pages too long.
1171 				 * XXX - there ought to be a better way.
1172 				 */
1173 				if (IS_SWAPVP(vp)) {
1174 					flags = SM_WRITE | SM_FREE |
1175 					    SM_DONTNEED;
1176 					iupdat_flag = 0;
1177 				} else {
1178 					flags = SM_WRITE;
1179 				}
1180 			} else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
1181 				/*
1182 				 * Have written a whole block.
1183 				 * Start an asynchronous write and
1184 				 * mark the buffer to indicate that
1185 				 * it won't be needed again soon.
1186 				 */
1187 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
1188 			}
1189 			if (vpm_enable) {
1190 				/*
1191 				 * Flush pages.
1192 				 */
1193 				error = vpm_sync_pages(vp, off, n, flags);
1194 			} else {
1195 				error = segmap_release(segkmap, base, flags);
1196 			}
1197 			/*
1198 			 * If the operation failed and is synchronous,
1199 			 * then we need to unwind what uiomove() last
1200 			 * did so we can potentially return an error to
1201 			 * the caller.  If this write operation was
1202 			 * done in two pieces and the first succeeded,
1203 			 * then we won't return an error for the second
1204 			 * piece that failed.  However, we only want to
1205 			 * return a resid value that reflects what was
1206 			 * really done.
1207 			 *
1208 			 * Failures for non-synchronous operations can
1209 			 * be ignored since the page subsystem will
1210 			 * retry the operation until it succeeds or the
1211 			 * file system is unmounted.
1212 			 */
1213 			if (error) {
1214 				if ((ioflag & (FSYNC | FDSYNC)) ||
1215 				    type == IFDIR) {
1216 					uio->uio_resid = premove_resid;
1217 				} else {
1218 					error = 0;
1219 				}
1220 			}
1221 		}
1222 
1223 		/*
1224 		 * Re-acquire contents lock.
1225 		 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1226 		 */
1227 		if (do_dqrwlock)
1228 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1229 		rw_enter(&ip->i_contents, RW_WRITER);
1230 
1231 		/*
1232 		 * If the uiomove() failed or if a synchronous
1233 		 * page push failed, fix up i_size.
1234 		 */
1235 		if (error) {
1236 			if (i_size_changed) {
1237 				/*
1238 				 * The uiomove failed, and we
1239 				 * allocated blocks,so get rid
1240 				 * of them.
1241 				 */
1242 				(void) ufs_itrunc(ip, old_i_size, 0, cr);
1243 			}
1244 		} else {
1245 			/*
1246 			 * XXX - Can this be out of the loop?
1247 			 */
1248 			ip->i_flag |= IUPD | ICHG;
1249 			/*
1250 			 * Only do one increase of i_seq for multiple
1251 			 * pieces.  Because we drop locks, record
1252 			 * the fact that we changed the timestamp and
1253 			 * are deferring the increase in case another thread
1254 			 * pushes our timestamp update.
1255 			 */
1256 			i_seq_needed = 1;
1257 			ip->i_flag |= ISEQ;
1258 			if (i_size_changed)
1259 				ip->i_flag |= IATTCHG;
1260 			if ((ip->i_mode & (IEXEC | (IEXEC >> 3) |
1261 			    (IEXEC >> 6))) != 0 &&
1262 			    (ip->i_mode & (ISUID | ISGID)) != 0 &&
1263 			    secpolicy_vnode_setid_retain(cr,
1264 			    (ip->i_mode & ISUID) != 0 && ip->i_uid == 0) != 0) {
1265 				/*
1266 				 * Clear Set-UID & Set-GID bits on
1267 				 * successful write if not privileged
1268 				 * and at least one of the execute bits
1269 				 * is set.  If we always clear Set-GID,
1270 				 * mandatory file and record locking is
1271 				 * unuseable.
1272 				 */
1273 				ip->i_mode &= ~(ISUID | ISGID);
1274 			}
1275 		}
1276 		/*
1277 		 * In the case the FDSYNC flag is set and this is a
1278 		 * "rewrite" we won't log a delta.
1279 		 * The FSYNC flag overrides all cases.
1280 		 */
1281 		if (!ufs_check_rewrite(ip, uio, ioflag) || !(ioflag & FDSYNC)) {
1282 			TRANS_INODE(ufsvfsp, ip);
1283 		}
1284 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1285 
1286 out:
1287 	/*
1288 	 * Make sure i_seq is increased at least once per write
1289 	 */
1290 	if (i_seq_needed) {
1291 		ip->i_seq++;
1292 		ip->i_flag &= ~ISEQ;	/* no longer deferred */
1293 	}
1294 
1295 	/*
1296 	 * Inode is updated according to this table -
1297 	 *
1298 	 *   FSYNC	  FDSYNC(posix.4)
1299 	 *   --------------------------
1300 	 *   always@	  IATTCHG|IBDWRITE
1301 	 *
1302 	 * @ - 	If we are doing synchronous write the only time we should
1303 	 *	not be sync'ing the ip here is if we have the stickyhack
1304 	 *	activated, the file is marked with the sticky bit and
1305 	 *	no exec bit, the file length has not been changed and
1306 	 *	no new blocks have been allocated during this write.
1307 	 */
1308 
1309 	if ((ip->i_flag & ISYNC) != 0) {
1310 		/*
1311 		 * we have eliminated nosync
1312 		 */
1313 		if ((ip->i_flag & (IATTCHG|IBDWRITE)) ||
1314 			((ioflag & FSYNC) && iupdat_flag)) {
1315 			ufs_iupdat(ip, 1);
1316 		}
1317 	}
1318 
1319 	/*
1320 	 * If we've already done a partial-write, terminate
1321 	 * the write but return no error unless the error is ENOSPC
1322 	 * because the caller can detect this and free resources and
1323 	 * try again.
1324 	 */
1325 	if ((start_resid != uio->uio_resid) && (error != ENOSPC))
1326 		error = 0;
1327 
1328 	ip->i_flag &= ~(INOACC | ISYNC);
1329 	ITIMES_NOLOCK(ip);
1330 	TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1331 		"ufs_wrip_end:vp %p error %d", vp, error);
1332 	return (error);
1333 }
1334 
1335 /*
1336  * rdip does the real work of read requests for ufs.
1337  */
1338 int
1339 rdip(struct inode *ip, struct uio *uio, int ioflag, cred_t *cr)
1340 {
1341 	u_offset_t off;
1342 	caddr_t base;
1343 	struct fs *fs;
1344 	struct ufsvfs *ufsvfsp;
1345 	struct vnode *vp;
1346 	long oresid = uio->uio_resid;
1347 	u_offset_t n, on, mapon;
1348 	int error = 0;
1349 	int doupdate = 1;
1350 	uint_t flags;
1351 	int dofree, directio_status;
1352 	krw_t rwtype;
1353 	o_mode_t type;
1354 
1355 	vp = ITOV(ip);
1356 
1357 	TRACE_1(TR_FAC_UFS, TR_UFS_RWIP_START,
1358 		"ufs_rdip_start:vp %p", vp);
1359 
1360 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
1361 
1362 	ufsvfsp = ip->i_ufsvfs;
1363 
1364 	if (ufsvfsp == NULL)
1365 		return (EIO);
1366 
1367 	fs = ufsvfsp->vfs_fs;
1368 
1369 	/* check for valid filetype */
1370 	type = ip->i_mode & IFMT;
1371 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
1372 	    (type != IFLNK) && (type != IFSHAD)) {
1373 		return (EIO);
1374 	}
1375 
1376 	if (uio->uio_loffset > UFS_MAXOFFSET_T) {
1377 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1378 			"ufs_rdip_end:vp %p error %d", vp, EINVAL);
1379 		error = 0;
1380 		goto out;
1381 	}
1382 	if (uio->uio_loffset < (offset_t)0) {
1383 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1384 			"ufs_rdip_end:vp %p error %d", vp, EINVAL);
1385 		return (EINVAL);
1386 	}
1387 	if (uio->uio_resid == 0) {
1388 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1389 			"ufs_rdip_end:vp %p error %d", vp, 0);
1390 		return (0);
1391 	}
1392 
1393 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (fs->fs_ronly == 0) &&
1394 		(!ufsvfsp->vfs_noatime)) {
1395 		mutex_enter(&ip->i_tlock);
1396 		ip->i_flag |= IACC;
1397 		mutex_exit(&ip->i_tlock);
1398 	}
1399 	/*
1400 	 * Try to go direct
1401 	 */
1402 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
1403 		error = ufs_directio_read(ip, uio, cr, &directio_status);
1404 		if (directio_status == DIRECTIO_SUCCESS)
1405 			goto out;
1406 	}
1407 
1408 	rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER);
1409 
1410 	do {
1411 		offset_t diff;
1412 		u_offset_t uoff = uio->uio_loffset;
1413 		off = uoff & (offset_t)MAXBMASK;
1414 		mapon = (u_offset_t)(uoff & (offset_t)MAXBOFFSET);
1415 		on = (u_offset_t)blkoff(fs, uoff);
1416 		n = MIN((u_offset_t)fs->fs_bsize - on,
1417 			(u_offset_t)uio->uio_resid);
1418 
1419 		diff = ip->i_size - uoff;
1420 
1421 		if (diff <= (offset_t)0) {
1422 			error = 0;
1423 			goto out;
1424 		}
1425 		if (diff < (offset_t)n)
1426 			n = (int)diff;
1427 
1428 		/*
1429 		 * We update smallfile2 and smallfile1 at most every second.
1430 		 */
1431 		if (lbolt >= smallfile_update) {
1432 			uint64_t percpufreeb;
1433 			if (smallfile1_d == 0) smallfile1_d = SMALLFILE1_D;
1434 			if (smallfile2_d == 0) smallfile2_d = SMALLFILE2_D;
1435 			percpufreeb = ptob((uint64_t)freemem) / ncpus_online;
1436 			smallfile1 = percpufreeb / smallfile1_d;
1437 			smallfile2 = percpufreeb / smallfile2_d;
1438 			smallfile1 = MAX(smallfile1, smallfile);
1439 			smallfile1 = MAX(smallfile1, smallfile64);
1440 			smallfile2 = MAX(smallfile1, smallfile2);
1441 			smallfile_update = lbolt + hz;
1442 		}
1443 
1444 		dofree = freebehind &&
1445 		    ip->i_nextr == (off & PAGEMASK) && off > smallfile1;
1446 
1447 		/*
1448 		 * At this point we can enter ufs_getpage() in one of two
1449 		 * ways:
1450 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
1451 		 *    forcefault parameter is true (value of 1 is passed)
1452 		 * 2) uiomove() causes a page fault.
1453 		 *
1454 		 * We cannot hold onto an i_contents reader lock without
1455 		 * risking deadlock in ufs_getpage() so drop a reader lock.
1456 		 * The ufs_getpage() dolock logic already allows for a
1457 		 * thread holding i_contents as writer to work properly
1458 		 * so we keep a writer lock.
1459 		 */
1460 		if (rwtype == RW_READER)
1461 			rw_exit(&ip->i_contents);
1462 
1463 		if (vpm_enable) {
1464 			/*
1465 			 * Copy data.
1466 			 */
1467 			error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1468 				uio, 1, NULL, 0, S_READ);
1469 		} else {
1470 			base = segmap_getmapflt(segkmap, vp, (off + mapon),
1471 					(uint_t)n, 1, S_READ);
1472 			error = uiomove(base + mapon, (long)n, UIO_READ, uio);
1473 		}
1474 
1475 		flags = 0;
1476 		if (!error) {
1477 			/*
1478 			 * If  reading sequential  we won't need  this
1479 			 * buffer again  soon.  For  offsets in  range
1480 			 * [smallfile1,  smallfile2] release the pages
1481 			 * at   the  tail  of the   cache list, larger
1482 			 * offsets are released at the head.
1483 			 */
1484 			if (dofree) {
1485 				flags = SM_FREE | SM_ASYNC;
1486 				if ((cache_read_ahead == 0) &&
1487 					(off > smallfile2))
1488 					flags |=  SM_DONTNEED;
1489 			}
1490 			/*
1491 			 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1492 			 * we want to make sure that the page which has
1493 			 * been read, is written on disk if it is dirty.
1494 			 * And corresponding indirect blocks should also
1495 			 * be flushed out.
1496 			 */
1497 			if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) {
1498 				flags &= ~SM_ASYNC;
1499 				flags |= SM_WRITE;
1500 			}
1501 			if (vpm_enable) {
1502 				error = vpm_sync_pages(vp, off, n, flags);
1503 			} else {
1504 				error = segmap_release(segkmap, base, flags);
1505 			}
1506 		} else {
1507 			if (vpm_enable) {
1508 				(void) vpm_sync_pages(vp, off, n, flags);
1509 			} else {
1510 				(void) segmap_release(segkmap, base, flags);
1511 			}
1512 		}
1513 
1514 		if (rwtype == RW_READER)
1515 			rw_enter(&ip->i_contents, rwtype);
1516 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1517 out:
1518 	/*
1519 	 * Inode is updated according to this table if FRSYNC is set.
1520 	 *
1521 	 *   FSYNC	  FDSYNC(posix.4)
1522 	 *   --------------------------
1523 	 *   always	  IATTCHG|IBDWRITE
1524 	 */
1525 	/*
1526 	 * The inode is not updated if we're logging and the inode is a
1527 	 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1528 	 */
1529 	if (ioflag & FRSYNC) {
1530 		if (TRANS_ISTRANS(ufsvfsp) && ((ip->i_mode & IFMT) == IFDIR)) {
1531 				doupdate = 0;
1532 		}
1533 		if (doupdate) {
1534 			if ((ioflag & FSYNC) ||
1535 			    ((ioflag & FDSYNC) &&
1536 			    (ip->i_flag & (IATTCHG|IBDWRITE)))) {
1537 				ufs_iupdat(ip, 1);
1538 			}
1539 		}
1540 	}
1541 	/*
1542 	 * If we've already done a partial read, terminate
1543 	 * the read but return no error.
1544 	 */
1545 	if (oresid != uio->uio_resid)
1546 		error = 0;
1547 	ITIMES(ip);
1548 
1549 	TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1550 		"ufs_rdip_end:vp %p error %d", vp, error);
1551 	return (error);
1552 }
1553 
1554 /* ARGSUSED */
1555 static int
1556 ufs_ioctl(
1557 	struct vnode	*vp,
1558 	int		cmd,
1559 	intptr_t	arg,
1560 	int		flag,
1561 	struct cred	*cr,
1562 	int		*rvalp)
1563 {
1564 	struct lockfs	lockfs, lockfs_out;
1565 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
1566 	char		*comment, *original_comment;
1567 	struct fs	*fs;
1568 	struct ulockfs	*ulp;
1569 	offset_t	off;
1570 	extern int	maxphys;
1571 	int		error;
1572 	int		issync;
1573 	int		trans_size;
1574 
1575 
1576 	/*
1577 	 * forcibly unmounted
1578 	 */
1579 	if (ufsvfsp == NULL) {
1580 		return (EIO);
1581 	}
1582 
1583 	fs = ufsvfsp->vfs_fs;
1584 
1585 	if (cmd == Q_QUOTACTL) {
1586 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_QUOTA_MASK);
1587 		if (error)
1588 			return (error);
1589 
1590 		if (ulp) {
1591 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_QUOTA,
1592 						TOP_SETQUOTA_SIZE(fs));
1593 		}
1594 
1595 		error = quotactl(vp, arg, flag, cr);
1596 
1597 		if (ulp) {
1598 			TRANS_END_ASYNC(ufsvfsp, TOP_QUOTA,
1599 						TOP_SETQUOTA_SIZE(fs));
1600 			ufs_lockfs_end(ulp);
1601 		}
1602 		return (error);
1603 	}
1604 
1605 	switch (cmd) {
1606 		case _FIOLFS:
1607 			/*
1608 			 * file system locking
1609 			 */
1610 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1611 				return (EPERM);
1612 
1613 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1614 				if (copyin((caddr_t)arg, &lockfs,
1615 						sizeof (struct lockfs)))
1616 					return (EFAULT);
1617 			}
1618 #ifdef _SYSCALL32_IMPL
1619 			else {
1620 				struct lockfs32	lockfs32;
1621 				/* Translate ILP32 lockfs to LP64 lockfs */
1622 				if (copyin((caddr_t)arg, &lockfs32,
1623 				    sizeof (struct lockfs32)))
1624 					return (EFAULT);
1625 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1626 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1627 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1628 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1629 				lockfs.lf_comment =
1630 					(caddr_t)(uintptr_t)lockfs32.lf_comment;
1631 			}
1632 #endif /* _SYSCALL32_IMPL */
1633 
1634 			if (lockfs.lf_comlen) {
1635 				if (lockfs.lf_comlen > LOCKFS_MAXCOMMENTLEN)
1636 					return (ENAMETOOLONG);
1637 				comment = kmem_alloc(lockfs.lf_comlen,
1638 						KM_SLEEP);
1639 				if (copyin(lockfs.lf_comment, comment,
1640 					lockfs.lf_comlen)) {
1641 					kmem_free(comment, lockfs.lf_comlen);
1642 					return (EFAULT);
1643 				}
1644 				original_comment = lockfs.lf_comment;
1645 				lockfs.lf_comment = comment;
1646 			}
1647 			if ((error = ufs_fiolfs(vp, &lockfs, 0)) == 0) {
1648 				lockfs.lf_comment = original_comment;
1649 
1650 				if ((flag & DATAMODEL_MASK) ==
1651 				    DATAMODEL_NATIVE) {
1652 					(void) copyout(&lockfs, (caddr_t)arg,
1653 					    sizeof (struct lockfs));
1654 				}
1655 #ifdef _SYSCALL32_IMPL
1656 				else {
1657 					struct lockfs32	lockfs32;
1658 					/* Translate LP64 to ILP32 lockfs */
1659 					lockfs32.lf_lock =
1660 					    (uint32_t)lockfs.lf_lock;
1661 					lockfs32.lf_flags =
1662 					    (uint32_t)lockfs.lf_flags;
1663 					lockfs32.lf_key =
1664 					    (uint32_t)lockfs.lf_key;
1665 					lockfs32.lf_comlen =
1666 					    (uint32_t)lockfs.lf_comlen;
1667 					lockfs32.lf_comment =
1668 					(uint32_t)(uintptr_t)lockfs.lf_comment;
1669 					(void) copyout(&lockfs32, (caddr_t)arg,
1670 					    sizeof (struct lockfs32));
1671 				}
1672 #endif /* _SYSCALL32_IMPL */
1673 
1674 			} else {
1675 				if (lockfs.lf_comlen)
1676 					kmem_free(comment, lockfs.lf_comlen);
1677 			}
1678 			return (error);
1679 
1680 		case _FIOLFSS:
1681 			/*
1682 			 * get file system locking status
1683 			 */
1684 
1685 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1686 				if (copyin((caddr_t)arg, &lockfs,
1687 						sizeof (struct lockfs)))
1688 					return (EFAULT);
1689 			}
1690 #ifdef _SYSCALL32_IMPL
1691 			else {
1692 				struct lockfs32	lockfs32;
1693 				/* Translate ILP32 lockfs to LP64 lockfs */
1694 				if (copyin((caddr_t)arg, &lockfs32,
1695 						sizeof (struct lockfs32)))
1696 					return (EFAULT);
1697 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1698 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1699 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1700 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1701 				lockfs.lf_comment =
1702 					(caddr_t)(uintptr_t)lockfs32.lf_comment;
1703 			}
1704 #endif /* _SYSCALL32_IMPL */
1705 
1706 			if (error =  ufs_fiolfss(vp, &lockfs_out))
1707 				return (error);
1708 			lockfs.lf_lock = lockfs_out.lf_lock;
1709 			lockfs.lf_key = lockfs_out.lf_key;
1710 			lockfs.lf_flags = lockfs_out.lf_flags;
1711 			lockfs.lf_comlen = MIN(lockfs.lf_comlen,
1712 				lockfs_out.lf_comlen);
1713 
1714 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1715 				if (copyout(&lockfs, (caddr_t)arg,
1716 						sizeof (struct lockfs)))
1717 					return (EFAULT);
1718 			}
1719 #ifdef _SYSCALL32_IMPL
1720 			else {
1721 				/* Translate LP64 to ILP32 lockfs */
1722 				struct lockfs32	lockfs32;
1723 				lockfs32.lf_lock = (uint32_t)lockfs.lf_lock;
1724 				lockfs32.lf_flags = (uint32_t)lockfs.lf_flags;
1725 				lockfs32.lf_key = (uint32_t)lockfs.lf_key;
1726 				lockfs32.lf_comlen = (uint32_t)lockfs.lf_comlen;
1727 				lockfs32.lf_comment =
1728 					(uint32_t)(uintptr_t)lockfs.lf_comment;
1729 				if (copyout(&lockfs32, (caddr_t)arg,
1730 					    sizeof (struct lockfs32)))
1731 					return (EFAULT);
1732 			}
1733 #endif /* _SYSCALL32_IMPL */
1734 
1735 			if (lockfs.lf_comlen &&
1736 			    lockfs.lf_comment && lockfs_out.lf_comment)
1737 				if (copyout(lockfs_out.lf_comment,
1738 					lockfs.lf_comment,
1739 					lockfs.lf_comlen))
1740 					return (EFAULT);
1741 			return (0);
1742 
1743 		case _FIOSATIME:
1744 			/*
1745 			 * set access time
1746 			 */
1747 
1748 			/*
1749 			 * if mounted w/o atime, return quietly.
1750 			 * I briefly thought about returning ENOSYS, but
1751 			 * figured that most apps would consider this fatal
1752 			 * but the idea is to make this as seamless as poss.
1753 			 */
1754 			if (ufsvfsp->vfs_noatime)
1755 				return (0);
1756 
1757 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1758 					ULOCKFS_SETATTR_MASK);
1759 			if (error)
1760 				return (error);
1761 
1762 			if (ulp) {
1763 				trans_size = (int)TOP_SETATTR_SIZE(VTOI(vp));
1764 				TRANS_BEGIN_CSYNC(ufsvfsp, issync,
1765 						TOP_SETATTR, trans_size);
1766 			}
1767 
1768 			error = ufs_fiosatime(vp, (struct timeval *)arg,
1769 					flag, cr);
1770 
1771 			if (ulp) {
1772 				TRANS_END_CSYNC(ufsvfsp, error, issync,
1773 						TOP_SETATTR, trans_size);
1774 				ufs_lockfs_end(ulp);
1775 			}
1776 			return (error);
1777 
1778 		case _FIOSDIO:
1779 			/*
1780 			 * set delayed-io
1781 			 */
1782 			return (ufs_fiosdio(vp, (uint_t *)arg, flag, cr));
1783 
1784 		case _FIOGDIO:
1785 			/*
1786 			 * get delayed-io
1787 			 */
1788 			return (ufs_fiogdio(vp, (uint_t *)arg, flag, cr));
1789 
1790 		case _FIOIO:
1791 			/*
1792 			 * inode open
1793 			 */
1794 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1795 					ULOCKFS_VGET_MASK);
1796 			if (error)
1797 				return (error);
1798 
1799 			error = ufs_fioio(vp, (struct fioio *)arg, flag, cr);
1800 
1801 			if (ulp) {
1802 				ufs_lockfs_end(ulp);
1803 			}
1804 			return (error);
1805 
1806 		case _FIOFFS:
1807 			/*
1808 			 * file system flush (push w/invalidate)
1809 			 */
1810 			if ((caddr_t)arg != NULL)
1811 				return (EINVAL);
1812 			return (ufs_fioffs(vp, NULL, cr));
1813 
1814 		case _FIOISBUSY:
1815 			/*
1816 			 * Contract-private interface for Legato
1817 			 * Purge this vnode from the DNLC and decide
1818 			 * if this vnode is busy (*arg == 1) or not
1819 			 * (*arg == 0)
1820 			 */
1821 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1822 				return (EPERM);
1823 			error = ufs_fioisbusy(vp, (int *)arg, cr);
1824 			return (error);
1825 
1826 		case _FIODIRECTIO:
1827 			return (ufs_fiodirectio(vp, (int)arg, cr));
1828 
1829 		case _FIOTUNE:
1830 			/*
1831 			 * Tune the file system (aka setting fs attributes)
1832 			 */
1833 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1834 					ULOCKFS_SETATTR_MASK);
1835 			if (error)
1836 				return (error);
1837 
1838 			error = ufs_fiotune(vp, (struct fiotune *)arg, cr);
1839 
1840 			if (ulp)
1841 				ufs_lockfs_end(ulp);
1842 			return (error);
1843 
1844 		case _FIOLOGENABLE:
1845 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1846 				return (EPERM);
1847 			return (ufs_fiologenable(vp, (void *)arg, cr, flag));
1848 
1849 		case _FIOLOGDISABLE:
1850 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1851 				return (EPERM);
1852 			return (ufs_fiologdisable(vp, (void *)arg, cr, flag));
1853 
1854 		case _FIOISLOG:
1855 			return (ufs_fioislog(vp, (void *)arg, cr, flag));
1856 
1857 		case _FIOSNAPSHOTCREATE_MULTI:
1858 		{
1859 			struct fiosnapcreate_multi	fc, *fcp;
1860 			size_t	fcm_size;
1861 
1862 			if (copyin((void *)arg, &fc, sizeof (fc)))
1863 				return (EFAULT);
1864 			if (fc.backfilecount > MAX_BACKFILE_COUNT)
1865 				return (EINVAL);
1866 			fcm_size = sizeof (struct fiosnapcreate_multi) +
1867 			    (fc.backfilecount - 1) * sizeof (int);
1868 			fcp = (struct fiosnapcreate_multi *)
1869 			    kmem_alloc(fcm_size, KM_SLEEP);
1870 			if (copyin((void *)arg, fcp, fcm_size)) {
1871 				kmem_free(fcp, fcm_size);
1872 				return (EFAULT);
1873 			}
1874 			error = ufs_snap_create(vp, fcp, cr);
1875 			if (!error && copyout(fcp, (void *)arg, fcm_size))
1876 				error = EFAULT;
1877 			kmem_free(fcp, fcm_size);
1878 			return (error);
1879 		}
1880 
1881 		case _FIOSNAPSHOTDELETE:
1882 		{
1883 			struct fiosnapdelete	fc;
1884 
1885 			if (copyin((void *)arg, &fc, sizeof (fc)))
1886 				return (EFAULT);
1887 			error = ufs_snap_delete(vp, &fc, cr);
1888 			if (!error && copyout(&fc, (void *)arg, sizeof (fc)))
1889 				error = EFAULT;
1890 			return (error);
1891 		}
1892 
1893 		case _FIOGETSUPERBLOCK:
1894 			if (copyout(fs, (void *)arg, SBSIZE))
1895 				return (EFAULT);
1896 			return (0);
1897 
1898 		case _FIOGETMAXPHYS:
1899 			if (copyout(&maxphys, (void *)arg, sizeof (maxphys)))
1900 				return (EFAULT);
1901 			return (0);
1902 
1903 		/*
1904 		 * The following 3 ioctls are for TSufs support
1905 		 * although could potentially be used elsewhere
1906 		 */
1907 		case _FIO_SET_LUFS_DEBUG:
1908 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1909 				return (EPERM);
1910 			lufs_debug = (uint32_t)arg;
1911 			return (0);
1912 
1913 		case _FIO_SET_LUFS_ERROR:
1914 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1915 				return (EPERM);
1916 			TRANS_SETERROR(ufsvfsp);
1917 			return (0);
1918 
1919 		case _FIO_GET_TOP_STATS:
1920 		{
1921 			fio_lufs_stats_t *ls;
1922 			ml_unit_t *ul = ufsvfsp->vfs_log;
1923 
1924 			ls = kmem_zalloc(sizeof (*ls), KM_SLEEP);
1925 			ls->ls_debug = ul->un_debug; /* return debug value */
1926 			/* Copy stucture if statistics are being kept */
1927 			if (ul->un_logmap->mtm_tops) {
1928 				ls->ls_topstats = *(ul->un_logmap->mtm_tops);
1929 			}
1930 			error = 0;
1931 			if (copyout(ls, (void *)arg, sizeof (*ls)))
1932 				error = EFAULT;
1933 			kmem_free(ls, sizeof (*ls));
1934 			return (error);
1935 		}
1936 
1937 		case _FIO_SEEK_DATA:
1938 		case _FIO_SEEK_HOLE:
1939 			if (ddi_copyin((void *)arg, &off, sizeof (off), flag))
1940 				return (EFAULT);
1941 			/* offset paramater is in/out */
1942 			error = ufs_fio_holey(vp, cmd, &off);
1943 			if (error)
1944 				return (error);
1945 			if (ddi_copyout(&off, (void *)arg, sizeof (off), flag))
1946 				return (EFAULT);
1947 			return (0);
1948 
1949 		default:
1950 			return (ENOTTY);
1951 	}
1952 }
1953 
1954 /* ARGSUSED */
1955 static int
1956 ufs_getattr(struct vnode *vp, struct vattr *vap, int flags,
1957 	struct cred *cr)
1958 {
1959 	struct inode *ip = VTOI(vp);
1960 	struct ufsvfs *ufsvfsp;
1961 	int err;
1962 
1963 	TRACE_2(TR_FAC_UFS, TR_UFS_GETATTR_START,
1964 		"ufs_getattr_start:vp %p flags %x", vp, flags);
1965 
1966 	if (vap->va_mask == AT_SIZE) {
1967 		/*
1968 		 * for performance, if only the size is requested don't bother
1969 		 * with anything else.
1970 		 */
1971 		UFS_GET_ISIZE(&vap->va_size, ip);
1972 		TRACE_1(TR_FAC_UFS, TR_UFS_GETATTR_END,
1973 			"ufs_getattr_end:vp %p", vp);
1974 		return (0);
1975 	}
1976 
1977 	/*
1978 	 * inlined lockfs checks
1979 	 */
1980 	ufsvfsp = ip->i_ufsvfs;
1981 	if ((ufsvfsp == NULL) || ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs)) {
1982 		err = EIO;
1983 		goto out;
1984 	}
1985 
1986 	rw_enter(&ip->i_contents, RW_READER);
1987 	/*
1988 	 * Return all the attributes.  This should be refined so
1989 	 * that it only returns what's asked for.
1990 	 */
1991 
1992 	/*
1993 	 * Copy from inode table.
1994 	 */
1995 	vap->va_type = vp->v_type;
1996 	vap->va_mode = ip->i_mode & MODEMASK;
1997 	/*
1998 	 * If there is an ACL and there is a mask entry, then do the
1999 	 * extra work that completes the equivalent of an acltomode(3)
2000 	 * call.  According to POSIX P1003.1e, the acl mask should be
2001 	 * returned in the group permissions field.
2002 	 *
2003 	 * - start with the original permission and mode bits (from above)
2004 	 * - clear the group owner bits
2005 	 * - add in the mask bits.
2006 	 */
2007 	if (ip->i_ufs_acl && ip->i_ufs_acl->aclass.acl_ismask) {
2008 		vap->va_mode &= ~((VREAD | VWRITE | VEXEC) >> 3);
2009 		vap->va_mode |=
2010 		    (ip->i_ufs_acl->aclass.acl_maskbits & PERMMASK) << 3;
2011 	}
2012 	vap->va_uid = ip->i_uid;
2013 	vap->va_gid = ip->i_gid;
2014 	vap->va_fsid = ip->i_dev;
2015 	vap->va_nodeid = (ino64_t)ip->i_number;
2016 	vap->va_nlink = ip->i_nlink;
2017 	vap->va_size = ip->i_size;
2018 	if (vp->v_type == VCHR || vp->v_type == VBLK)
2019 		vap->va_rdev = ip->i_rdev;
2020 	else
2021 		vap->va_rdev = 0;	/* not a b/c spec. */
2022 	mutex_enter(&ip->i_tlock);
2023 	ITIMES_NOLOCK(ip);	/* mark correct time in inode */
2024 	vap->va_seq = ip->i_seq;
2025 	vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec;
2026 	vap->va_atime.tv_nsec = ip->i_atime.tv_usec*1000;
2027 	vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec;
2028 	vap->va_mtime.tv_nsec = ip->i_mtime.tv_usec*1000;
2029 	vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec;
2030 	vap->va_ctime.tv_nsec = ip->i_ctime.tv_usec*1000;
2031 	mutex_exit(&ip->i_tlock);
2032 
2033 	switch (ip->i_mode & IFMT) {
2034 
2035 	case IFBLK:
2036 		vap->va_blksize = MAXBSIZE;		/* was BLKDEV_IOSIZE */
2037 		break;
2038 
2039 	case IFCHR:
2040 		vap->va_blksize = MAXBSIZE;
2041 		break;
2042 
2043 	default:
2044 		vap->va_blksize = ip->i_fs->fs_bsize;
2045 		break;
2046 	}
2047 	vap->va_nblocks = (fsblkcnt64_t)ip->i_blocks;
2048 	rw_exit(&ip->i_contents);
2049 	err = 0;
2050 
2051 out:
2052 	TRACE_1(TR_FAC_UFS, TR_UFS_GETATTR_END, "ufs_getattr_end:vp %p", vp);
2053 
2054 	return (err);
2055 }
2056 
2057 /*ARGSUSED4*/
2058 static int
2059 ufs_setattr(
2060 	struct vnode *vp,
2061 	struct vattr *vap,
2062 	int flags,
2063 	struct cred *cr,
2064 	caller_context_t *ct)
2065 {
2066 	struct inode *ip = VTOI(vp);
2067 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2068 	struct fs *fs;
2069 	struct ulockfs *ulp;
2070 	char *errmsg1;
2071 	char *errmsg2;
2072 	long blocks;
2073 	long int mask = vap->va_mask;
2074 	size_t len1, len2;
2075 	int issync;
2076 	int trans_size;
2077 	int dotrans;
2078 	int dorwlock;
2079 	int error;
2080 	int owner_change;
2081 	int dodqlock;
2082 	timestruc_t now;
2083 	vattr_t oldva;
2084 	int retry = 1;
2085 	int indeadlock;
2086 
2087 	TRACE_2(TR_FAC_UFS, TR_UFS_SETATTR_START,
2088 		"ufs_setattr_start:vp %p flags %x", vp, flags);
2089 
2090 	/*
2091 	 * Cannot set these attributes.
2092 	 */
2093 	if (mask & AT_NOSET) {
2094 		error = EINVAL;
2095 		goto out;
2096 	}
2097 
2098 	/*
2099 	 * check for forced unmount
2100 	 */
2101 	if (ufsvfsp == NULL)
2102 		return (EIO);
2103 
2104 	fs = ufsvfsp->vfs_fs;
2105 	if (fs->fs_ronly != 0)
2106 		return (EROFS);
2107 
2108 again:
2109 	errmsg1 = NULL;
2110 	errmsg2 = NULL;
2111 	dotrans = 0;
2112 	dorwlock = 0;
2113 	dodqlock = 0;
2114 
2115 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK);
2116 	if (error)
2117 		goto out;
2118 
2119 	/*
2120 	 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2121 	 * This follows the protocol for read()/write().
2122 	 */
2123 	if (vp->v_type != VDIR) {
2124 		/*
2125 		 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2126 		 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2127 		 * possible, retries the operation.
2128 		 */
2129 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_file);
2130 		if (indeadlock) {
2131 			if (ulp)
2132 				ufs_lockfs_end(ulp);
2133 			goto again;
2134 		}
2135 		dorwlock = 1;
2136 	}
2137 
2138 	/*
2139 	 * Truncate file.  Must have write permission and not be a directory.
2140 	 */
2141 	if (mask & AT_SIZE) {
2142 		rw_enter(&ip->i_contents, RW_WRITER);
2143 		if (vp->v_type == VDIR) {
2144 			error = EISDIR;
2145 			goto update_inode;
2146 		}
2147 		if (error = ufs_iaccess(ip, IWRITE, cr))
2148 			goto update_inode;
2149 
2150 		rw_exit(&ip->i_contents);
2151 		error = TRANS_ITRUNC(ip, vap->va_size, 0, cr);
2152 		if (error) {
2153 			rw_enter(&ip->i_contents, RW_WRITER);
2154 			goto update_inode;
2155 		}
2156 	}
2157 
2158 	if (ulp) {
2159 		trans_size = (int)TOP_SETATTR_SIZE(ip);
2160 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size);
2161 		++dotrans;
2162 	}
2163 
2164 	/*
2165 	 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2166 	 * This follows the protocol established by
2167 	 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2168 	 */
2169 	if (vp->v_type == VDIR) {
2170 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_SETATTR,
2171 					retry_dir);
2172 		if (indeadlock)
2173 			goto again;
2174 		dorwlock = 1;
2175 	}
2176 
2177 	/*
2178 	 * Grab quota lock if we are changing the file's owner.
2179 	 */
2180 	if (mask & AT_UID) {
2181 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2182 		dodqlock = 1;
2183 	}
2184 	rw_enter(&ip->i_contents, RW_WRITER);
2185 
2186 	oldva.va_mode = ip->i_mode;
2187 	oldva.va_uid = ip->i_uid;
2188 	oldva.va_gid = ip->i_gid;
2189 
2190 	vap->va_mask &= ~AT_SIZE;
2191 	/*
2192 	 * ufs_iaccess is "close enough"; that's because it doesn't
2193 	 * map the defines.
2194 	 */
2195 	error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2196 				ufs_iaccess, ip);
2197 	if (error)
2198 		goto update_inode;
2199 
2200 	mask = vap->va_mask;
2201 
2202 	/*
2203 	 * Change file access modes.
2204 	 */
2205 	if (mask & AT_MODE) {
2206 		ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT);
2207 		TRANS_INODE(ufsvfsp, ip);
2208 		ip->i_flag |= ICHG;
2209 		if (stickyhack) {
2210 			mutex_enter(&vp->v_lock);
2211 			if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX)
2212 				vp->v_flag |= VSWAPLIKE;
2213 			else
2214 				vp->v_flag &= ~VSWAPLIKE;
2215 			mutex_exit(&vp->v_lock);
2216 		}
2217 	}
2218 	if (mask & (AT_UID|AT_GID)) {
2219 		if (mask & AT_UID) {
2220 			/*
2221 			 * Don't change ownership of the quota inode.
2222 			 */
2223 			if (ufsvfsp->vfs_qinod == ip) {
2224 				ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED);
2225 				error = EINVAL;
2226 				goto update_inode;
2227 			}
2228 
2229 			/*
2230 			 * No real ownership change.
2231 			 */
2232 			if (ip->i_uid == vap->va_uid) {
2233 				blocks = 0;
2234 				owner_change = 0;
2235 			}
2236 			/*
2237 			 * Remove the blocks and the file, from the old user's
2238 			 * quota.
2239 			 */
2240 			else {
2241 				blocks = ip->i_blocks;
2242 				owner_change = 1;
2243 
2244 				(void) chkdq(ip, -blocks, /* force */ 1, cr,
2245 						(char **)NULL, (size_t *)NULL);
2246 				(void) chkiq(ufsvfsp, /* change */ -1, ip,
2247 						(uid_t)ip->i_uid,
2248 						/* force */ 1, cr,
2249 						(char **)NULL, (size_t *)NULL);
2250 				dqrele(ip->i_dquot);
2251 			}
2252 
2253 			ip->i_uid = vap->va_uid;
2254 
2255 			/*
2256 			 * There is a real ownership change.
2257 			 */
2258 			if (owner_change) {
2259 				/*
2260 				 * Add the blocks and the file to the new
2261 				 * user's quota.
2262 				 */
2263 				ip->i_dquot = getinoquota(ip);
2264 				(void) chkdq(ip, blocks, /* force */ 1, cr,
2265 						&errmsg1, &len1);
2266 				(void) chkiq(ufsvfsp, /* change */ 1,
2267 						(struct inode *)NULL,
2268 						(uid_t)ip->i_uid,
2269 						/* force */ 1, cr,
2270 						&errmsg2, &len2);
2271 			}
2272 		}
2273 		if (mask & AT_GID) {
2274 			ip->i_gid = vap->va_gid;
2275 		}
2276 		TRANS_INODE(ufsvfsp, ip);
2277 		ip->i_flag |= ICHG;
2278 	}
2279 	/*
2280 	 * Change file access or modified times.
2281 	 */
2282 	if (mask & (AT_ATIME|AT_MTIME)) {
2283 		/* Check that the time value is within ufs range */
2284 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2285 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2286 			error = EOVERFLOW;
2287 			goto update_inode;
2288 		}
2289 
2290 		/*
2291 		 * if the "noaccess" mount option is set and only atime
2292 		 * update is requested, do nothing. No error is returned.
2293 		 */
2294 		if ((ufsvfsp->vfs_noatime) &&
2295 		    ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME))
2296 			goto skip_atime;
2297 
2298 		if (mask & AT_ATIME) {
2299 			ip->i_atime.tv_sec = vap->va_atime.tv_sec;
2300 			ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000;
2301 			ip->i_flag &= ~IACC;
2302 		}
2303 		if (mask & AT_MTIME) {
2304 			ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
2305 			ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000;
2306 			gethrestime(&now);
2307 			if (now.tv_sec > TIME32_MAX) {
2308 				/*
2309 				 * In 2038, ctime sticks forever..
2310 				 */
2311 				ip->i_ctime.tv_sec = TIME32_MAX;
2312 				ip->i_ctime.tv_usec = 0;
2313 			} else {
2314 				ip->i_ctime.tv_sec = now.tv_sec;
2315 				ip->i_ctime.tv_usec = now.tv_nsec / 1000;
2316 			}
2317 			ip->i_flag &= ~(IUPD|ICHG);
2318 			ip->i_flag |= IMODTIME;
2319 		}
2320 		TRANS_INODE(ufsvfsp, ip);
2321 		ip->i_flag |= IMOD;
2322 	}
2323 
2324 skip_atime:
2325 	/*
2326 	 * The presence of a shadow inode may indicate an ACL, but does
2327 	 * not imply an ACL.  Future FSD types should be handled here too
2328 	 * and check for the presence of the attribute-specific data
2329 	 * before referencing it.
2330 	 */
2331 	if (ip->i_shadow) {
2332 		/*
2333 		 * XXX if ufs_iupdat is changed to sandbagged write fix
2334 		 * ufs_acl_setattr to push ip to keep acls consistent
2335 		 *
2336 		 * Suppress out of inodes messages if we will retry.
2337 		 */
2338 		if (retry)
2339 			ip->i_flag |= IQUIET;
2340 		error = ufs_acl_setattr(ip, vap, cr);
2341 		ip->i_flag &= ~IQUIET;
2342 	}
2343 
2344 update_inode:
2345 	/*
2346 	 * Setattr always increases the sequence number
2347 	 */
2348 	ip->i_seq++;
2349 
2350 	/*
2351 	 * if nfsd and not logging; push synchronously
2352 	 */
2353 	if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) {
2354 		ufs_iupdat(ip, 1);
2355 	} else {
2356 		ITIMES_NOLOCK(ip);
2357 	}
2358 
2359 	rw_exit(&ip->i_contents);
2360 	if (dodqlock) {
2361 		rw_exit(&ufsvfsp->vfs_dqrwlock);
2362 	}
2363 	if (dorwlock)
2364 		rw_exit(&ip->i_rwlock);
2365 
2366 	if (ulp) {
2367 		if (dotrans) {
2368 			int terr = 0;
2369 			TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR,
2370 			    trans_size);
2371 			if (error == 0)
2372 				error = terr;
2373 		}
2374 		ufs_lockfs_end(ulp);
2375 	}
2376 out:
2377 	/*
2378 	 * If out of inodes or blocks, see if we can free something
2379 	 * up from the delete queue.
2380 	 */
2381 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
2382 		ufs_delete_drain_wait(ufsvfsp, 1);
2383 		retry = 0;
2384 		if (errmsg1 != NULL)
2385 			kmem_free(errmsg1, len1);
2386 		if (errmsg2 != NULL)
2387 			kmem_free(errmsg2, len2);
2388 		goto again;
2389 	}
2390 	TRACE_2(TR_FAC_UFS, TR_UFS_SETATTR_END,
2391 		"ufs_setattr_end:vp %p error %d", vp, error);
2392 	if (errmsg1 != NULL) {
2393 		uprintf(errmsg1);
2394 		kmem_free(errmsg1, len1);
2395 	}
2396 	if (errmsg2 != NULL) {
2397 		uprintf(errmsg2);
2398 		kmem_free(errmsg2, len2);
2399 	}
2400 	return (error);
2401 }
2402 
2403 /*ARGSUSED*/
2404 static int
2405 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr)
2406 {
2407 	struct inode *ip = VTOI(vp);
2408 	int error;
2409 
2410 	TRACE_3(TR_FAC_UFS, TR_UFS_ACCESS_START,
2411 		"ufs_access_start:vp %p mode %x flags %x", vp, mode, flags);
2412 
2413 	if (ip->i_ufsvfs == NULL)
2414 		return (EIO);
2415 
2416 	rw_enter(&ip->i_contents, RW_READER);
2417 
2418 	/*
2419 	 * The ufs_iaccess function wants to be called with
2420 	 * mode bits expressed as "ufs specific" bits.
2421 	 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2422 	 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2423 	 * But since they're the same we just pass the vnode mode
2424 	 * bit but just verify that assumption at compile time.
2425 	 */
2426 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2427 #error "ufs_access needs to map Vmodes to Imodes"
2428 #endif
2429 	error = ufs_iaccess(ip, mode, cr);
2430 
2431 	rw_exit(&ip->i_contents);
2432 
2433 	TRACE_2(TR_FAC_UFS, TR_UFS_ACCESS_END,
2434 		"ufs_access_end:vp %p error %d", vp, error);
2435 	return (error);
2436 }
2437 
2438 /* ARGSUSED */
2439 static int
2440 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr)
2441 {
2442 	struct inode *ip = VTOI(vp);
2443 	struct ufsvfs *ufsvfsp;
2444 	struct ulockfs *ulp;
2445 	int error;
2446 	int fastsymlink;
2447 
2448 	TRACE_2(TR_FAC_UFS, TR_UFS_READLINK_START,
2449 		"ufs_readlink_start:vp %p uiop %p", uiop, vp);
2450 
2451 	if (vp->v_type != VLNK) {
2452 		error = EINVAL;
2453 		goto nolockout;
2454 	}
2455 
2456 	/*
2457 	 * If the symbolic link is empty there is nothing to read.
2458 	 * Fast-track these empty symbolic links
2459 	 */
2460 	if (ip->i_size == 0) {
2461 		error = 0;
2462 		goto nolockout;
2463 	}
2464 
2465 	ufsvfsp = ip->i_ufsvfs;
2466 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK);
2467 	if (error)
2468 		goto nolockout;
2469 	/*
2470 	 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2471 	 */
2472 again:
2473 	fastsymlink = 0;
2474 	if (ip->i_flag & IFASTSYMLNK) {
2475 		rw_enter(&ip->i_rwlock, RW_READER);
2476 		rw_enter(&ip->i_contents, RW_READER);
2477 		if (ip->i_flag & IFASTSYMLNK) {
2478 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
2479 			    (ip->i_fs->fs_ronly == 0) &&
2480 			    (!ufsvfsp->vfs_noatime)) {
2481 				mutex_enter(&ip->i_tlock);
2482 				ip->i_flag |= IACC;
2483 				mutex_exit(&ip->i_tlock);
2484 			}
2485 			error = uiomove((caddr_t)&ip->i_db[1],
2486 				MIN(ip->i_size, uiop->uio_resid),
2487 				UIO_READ, uiop);
2488 			ITIMES(ip);
2489 			++fastsymlink;
2490 		}
2491 		rw_exit(&ip->i_contents);
2492 		rw_exit(&ip->i_rwlock);
2493 	}
2494 	if (!fastsymlink) {
2495 		ssize_t size;	/* number of bytes read  */
2496 		caddr_t basep;	/* pointer to input data */
2497 		ino_t ino;
2498 		long  igen;
2499 		struct uio tuio;	/* temp uio struct */
2500 		struct uio *tuiop;
2501 		iovec_t tiov;		/* temp iovec struct */
2502 		char kbuf[FSL_SIZE];	/* buffer to hold fast symlink */
2503 		int tflag = 0;		/* flag to indicate temp vars used */
2504 
2505 		ino = ip->i_number;
2506 		igen = ip->i_gen;
2507 		size = uiop->uio_resid;
2508 		basep = uiop->uio_iov->iov_base;
2509 		tuiop = uiop;
2510 
2511 		rw_enter(&ip->i_rwlock, RW_WRITER);
2512 		rw_enter(&ip->i_contents, RW_WRITER);
2513 		if (ip->i_flag & IFASTSYMLNK) {
2514 			rw_exit(&ip->i_contents);
2515 			rw_exit(&ip->i_rwlock);
2516 			goto again;
2517 		}
2518 
2519 		/* can this be a fast symlink and is it a user buffer? */
2520 		if (ip->i_size <= FSL_SIZE &&
2521 		    (uiop->uio_segflg == UIO_USERSPACE ||
2522 		    uiop->uio_segflg == UIO_USERISPACE)) {
2523 
2524 			bzero(&tuio, sizeof (struct uio));
2525 			/*
2526 			 * setup a kernel buffer to read link into.  this
2527 			 * is to fix a race condition where the user buffer
2528 			 * got corrupted before copying it into the inode.
2529 			 */
2530 			size = ip->i_size;
2531 			tiov.iov_len = size;
2532 			tiov.iov_base = kbuf;
2533 			tuio.uio_iov = &tiov;
2534 			tuio.uio_iovcnt = 1;
2535 			tuio.uio_offset = uiop->uio_offset;
2536 			tuio.uio_segflg = UIO_SYSSPACE;
2537 			tuio.uio_fmode = uiop->uio_fmode;
2538 			tuio.uio_extflg = uiop->uio_extflg;
2539 			tuio.uio_limit = uiop->uio_limit;
2540 			tuio.uio_resid = size;
2541 
2542 			basep = tuio.uio_iov->iov_base;
2543 			tuiop = &tuio;
2544 			tflag = 1;
2545 		}
2546 
2547 		error = rdip(ip, tuiop, 0, cr);
2548 		if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) {
2549 			rw_exit(&ip->i_contents);
2550 			rw_exit(&ip->i_rwlock);
2551 			goto out;
2552 		}
2553 
2554 		if (tflag == 0)
2555 			size -= uiop->uio_resid;
2556 
2557 		if ((tflag == 0 && ip->i_size <= FSL_SIZE &&
2558 		    ip->i_size == size) || (tflag == 1 &&
2559 		    tuio.uio_resid == 0)) {
2560 			error = kcopy(basep, &ip->i_db[1], ip->i_size);
2561 			if (error == 0) {
2562 				ip->i_flag |= IFASTSYMLNK;
2563 				/*
2564 				 * free page
2565 				 */
2566 				(void) VOP_PUTPAGE(ITOV(ip),
2567 				    (offset_t)0, PAGESIZE,
2568 				    (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC),
2569 				    cr);
2570 			} else {
2571 				int i;
2572 				/* error, clear garbage left behind */
2573 				for (i = 1; i < NDADDR; i++)
2574 					ip->i_db[i] = 0;
2575 				for (i = 0; i < NIADDR; i++)
2576 					ip->i_ib[i] = 0;
2577 			}
2578 		}
2579 		if (tflag == 1) {
2580 			/* now, copy it into the user buffer */
2581 			error = uiomove((caddr_t)kbuf,
2582 				MIN(size, uiop->uio_resid),
2583 				UIO_READ, uiop);
2584 		}
2585 		rw_exit(&ip->i_contents);
2586 		rw_exit(&ip->i_rwlock);
2587 	}
2588 out:
2589 	if (ulp) {
2590 		ufs_lockfs_end(ulp);
2591 	}
2592 nolockout:
2593 	TRACE_2(TR_FAC_UFS, TR_UFS_READLINK_END,
2594 		"ufs_readlink_end:vp %p error %d", vp, error);
2595 
2596 	return (error);
2597 }
2598 
2599 /* ARGSUSED */
2600 static int
2601 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr)
2602 {
2603 	struct inode *ip = VTOI(vp);
2604 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2605 	struct ulockfs *ulp;
2606 	int error;
2607 
2608 	TRACE_1(TR_FAC_UFS, TR_UFS_FSYNC_START,
2609 		"ufs_fsync_start:vp %p", vp);
2610 
2611 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK);
2612 	if (error)
2613 		return (error);
2614 
2615 	if (TRANS_ISTRANS(ufsvfsp)) {
2616 		/*
2617 		 * First push out any data pages
2618 		 */
2619 		if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2620 		    (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) {
2621 			error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0,
2622 			    0, CRED());
2623 			if (error)
2624 				goto out;
2625 		}
2626 
2627 		/*
2628 		 * Delta any delayed inode times updates
2629 		 * and push inode to log.
2630 		 * All other inode deltas will have already been delta'd
2631 		 * and will be pushed during the commit.
2632 		 */
2633 		if (!(syncflag & FDSYNC) &&
2634 		    ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) {
2635 			if (ulp) {
2636 				TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC,
2637 				    TOP_SYNCIP_SIZE);
2638 			}
2639 			rw_enter(&ip->i_contents, RW_READER);
2640 			mutex_enter(&ip->i_tlock);
2641 			ip->i_flag &= ~IMODTIME;
2642 			mutex_exit(&ip->i_tlock);
2643 			ufs_iupdat(ip, I_SYNC);
2644 			rw_exit(&ip->i_contents);
2645 			if (ulp) {
2646 				TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC,
2647 				    TOP_SYNCIP_SIZE);
2648 			}
2649 		}
2650 
2651 		/*
2652 		 * Commit the Moby transaction
2653 		 *
2654 		 * Deltas have already been made so we just need to
2655 		 * commit them with a synchronous transaction.
2656 		 * TRANS_BEGIN_SYNC() will return an error
2657 		 * if there are no deltas to commit, for an
2658 		 * empty transaction.
2659 		 */
2660 		if (ulp) {
2661 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE,
2662 			    error);
2663 			if (error) {
2664 				error = 0; /* commit wasn't needed */
2665 				goto out;
2666 			}
2667 			TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC,
2668 			    TOP_COMMIT_SIZE);
2669 		}
2670 	} else {	/* not logging */
2671 		if (!(IS_SWAPVP(vp)))
2672 			if (syncflag & FNODSYNC) {
2673 				/* Just update the inode only */
2674 				TRANS_IUPDAT(ip, 1);
2675 				error = 0;
2676 			} else if (syncflag & FDSYNC)
2677 				/* Do data-synchronous writes */
2678 				error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC);
2679 			else
2680 				/* Do synchronous writes */
2681 				error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC);
2682 
2683 		rw_enter(&ip->i_contents, RW_WRITER);
2684 		if (!error)
2685 			error = ufs_sync_indir(ip);
2686 		rw_exit(&ip->i_contents);
2687 	}
2688 out:
2689 	if (ulp) {
2690 		ufs_lockfs_end(ulp);
2691 	}
2692 	TRACE_2(TR_FAC_UFS, TR_UFS_FSYNC_END,
2693 		"ufs_fsync_end:vp %p error %d", vp, error);
2694 	return (error);
2695 }
2696 
2697 /*ARGSUSED*/
2698 static void
2699 ufs_inactive(struct vnode *vp, struct cred *cr)
2700 {
2701 	ufs_iinactive(VTOI(vp));
2702 }
2703 
2704 /*
2705  * Unix file system operations having to do with directory manipulation.
2706  */
2707 int ufs_lookup_idle_count = 2;	/* Number of inodes to idle each time */
2708 /* ARGSUSED */
2709 static int
2710 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
2711 	struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr)
2712 {
2713 	struct inode *ip;
2714 	struct inode *sip;
2715 	struct inode *xip;
2716 	struct ufsvfs *ufsvfsp;
2717 	struct ulockfs *ulp;
2718 	struct vnode *vp;
2719 	int error;
2720 
2721 	TRACE_2(TR_FAC_UFS, TR_UFS_LOOKUP_START,
2722 		"ufs_lookup_start:dvp %p name %s", dvp, nm);
2723 
2724 
2725 	/*
2726 	 * Check flags for type of lookup (regular file or attribute file)
2727 	 */
2728 
2729 	ip = VTOI(dvp);
2730 
2731 	if (flags & LOOKUP_XATTR) {
2732 
2733 		/*
2734 		 * We don't allow recursive attributes...
2735 		 * Maybe someday we will.
2736 		 */
2737 		if ((ip->i_cflags & IXATTR)) {
2738 			return (EINVAL);
2739 		}
2740 
2741 		if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) {
2742 			error = ufs_xattr_getattrdir(dvp, &sip, flags, cr);
2743 			if (error) {
2744 				*vpp = NULL;
2745 				goto out;
2746 			}
2747 
2748 			vp = ITOV(sip);
2749 			dnlc_update(dvp, XATTR_DIR_NAME, vp);
2750 		}
2751 
2752 		/*
2753 		 * Check accessibility of directory.
2754 		 */
2755 		if (vp == DNLC_NO_VNODE) {
2756 			VN_RELE(vp);
2757 			error = ENOENT;
2758 			goto out;
2759 		}
2760 		if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr)) != 0) {
2761 			VN_RELE(vp);
2762 			goto out;
2763 		}
2764 
2765 		*vpp = vp;
2766 		return (0);
2767 	}
2768 
2769 	/*
2770 	 * Check for a null component, which we should treat as
2771 	 * looking at dvp from within it's parent, so we don't
2772 	 * need a call to ufs_iaccess(), as it has already been
2773 	 * done.
2774 	 */
2775 	if (nm[0] == 0) {
2776 		VN_HOLD(dvp);
2777 		error = 0;
2778 		*vpp = dvp;
2779 		goto out;
2780 	}
2781 
2782 	/*
2783 	 * Check for "." ie itself. this is a quick check and
2784 	 * avoids adding "." into the dnlc (which have been seen
2785 	 * to occupy >10% of the cache).
2786 	 */
2787 	if ((nm[0] == '.') && (nm[1] == 0)) {
2788 		/*
2789 		 * Don't return without checking accessibility
2790 		 * of the directory. We only need the lock if
2791 		 * we are going to return it.
2792 		 */
2793 		if ((error = ufs_iaccess(ip, IEXEC, cr)) == 0) {
2794 			VN_HOLD(dvp);
2795 			*vpp = dvp;
2796 		}
2797 		goto out;
2798 	}
2799 
2800 	/*
2801 	 * Fast path: Check the directory name lookup cache.
2802 	 */
2803 	if (vp = dnlc_lookup(dvp, nm)) {
2804 		/*
2805 		 * Check accessibility of directory.
2806 		 */
2807 		if ((error = ufs_iaccess(ip, IEXEC, cr)) != 0) {
2808 			VN_RELE(vp);
2809 			goto out;
2810 		}
2811 		if (vp == DNLC_NO_VNODE) {
2812 			VN_RELE(vp);
2813 			error = ENOENT;
2814 			goto out;
2815 		}
2816 		xip = VTOI(vp);
2817 		ulp = NULL;
2818 		goto fastpath;
2819 	}
2820 
2821 	/*
2822 	 * Keep the idle queue from getting too long by
2823 	 * idling two inodes before attempting to allocate another.
2824 	 *    This operation must be performed before entering
2825 	 *    lockfs or a transaction.
2826 	 */
2827 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2828 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2829 			ins.in_lidles.value.ul += ufs_lookup_idle_count;
2830 			ufs_idle_some(ufs_lookup_idle_count);
2831 		}
2832 
2833 retry_lookup:
2834 	ufsvfsp = ip->i_ufsvfs;
2835 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK);
2836 	if (error)
2837 		goto out;
2838 
2839 	error = ufs_dirlook(ip, nm, &xip, cr, 1);
2840 
2841 fastpath:
2842 	if (error == 0) {
2843 		ip = xip;
2844 		*vpp = ITOV(ip);
2845 
2846 		/*
2847 		 * If vnode is a device return special vnode instead.
2848 		 */
2849 		if (IS_DEVVP(*vpp)) {
2850 			struct vnode *newvp;
2851 
2852 			newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
2853 			    cr);
2854 			VN_RELE(*vpp);
2855 			if (newvp == NULL)
2856 				error = ENOSYS;
2857 			else
2858 				*vpp = newvp;
2859 		}
2860 	}
2861 	if (ulp) {
2862 		ufs_lockfs_end(ulp);
2863 	}
2864 
2865 	if (error == EAGAIN)
2866 		goto retry_lookup;
2867 
2868 out:
2869 	TRACE_3(TR_FAC_UFS, TR_UFS_LOOKUP_END,
2870 		"ufs_lookup_end:dvp %p name %s error %d", vpp, nm, error);
2871 	return (error);
2872 }
2873 
2874 static int
2875 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl,
2876 	int mode, struct vnode **vpp, struct cred *cr, int flag)
2877 {
2878 	struct inode *ip;
2879 	struct inode *xip;
2880 	struct inode *dip;
2881 	struct vnode *xvp;
2882 	struct ufsvfs *ufsvfsp;
2883 	struct ulockfs *ulp;
2884 	int error;
2885 	int issync;
2886 	int truncflag;
2887 	int trans_size;
2888 	int noentry;
2889 	int defer_dip_seq_update = 0;	/* need to defer update of dip->i_seq */
2890 	int retry = 1;
2891 	int indeadlock;
2892 
2893 	TRACE_1(TR_FAC_UFS, TR_UFS_CREATE_START,
2894 		"ufs_create_start:dvp %p", dvp);
2895 
2896 again:
2897 	ip = VTOI(dvp);
2898 	ufsvfsp = ip->i_ufsvfs;
2899 	truncflag = 0;
2900 
2901 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK);
2902 	if (error)
2903 		goto out;
2904 
2905 	if (ulp) {
2906 		trans_size = (int)TOP_CREATE_SIZE(ip);
2907 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size);
2908 	}
2909 
2910 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
2911 		vap->va_mode &= ~VSVTX;
2912 
2913 	if (*name == '\0') {
2914 		/*
2915 		 * Null component name refers to the directory itself.
2916 		 */
2917 		VN_HOLD(dvp);
2918 		/*
2919 		 * Even though this is an error case, we need to grab the
2920 		 * quota lock since the error handling code below is common.
2921 		 */
2922 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2923 		rw_enter(&ip->i_contents, RW_WRITER);
2924 		error = EEXIST;
2925 	} else {
2926 		xip = NULL;
2927 		noentry = 0;
2928 		/*
2929 		 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2930 		 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2931 		 * possible, retries the operation.
2932 		 */
2933 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_CREATE,
2934 					retry_dir);
2935 		if (indeadlock)
2936 			goto again;
2937 
2938 		xvp = dnlc_lookup(dvp, name);
2939 		if (xvp == DNLC_NO_VNODE) {
2940 			noentry = 1;
2941 			VN_RELE(xvp);
2942 			xvp = NULL;
2943 		}
2944 		if (xvp) {
2945 			rw_exit(&ip->i_rwlock);
2946 			if (error = ufs_iaccess(ip, IEXEC, cr)) {
2947 				VN_RELE(xvp);
2948 			} else {
2949 				error = EEXIST;
2950 				xip = VTOI(xvp);
2951 			}
2952 		} else {
2953 			/*
2954 			 * Suppress file system full message if we will retry
2955 			 */
2956 			error = ufs_direnter_cm(ip, name, DE_CREATE,
2957 				vap, &xip, cr,
2958 				(noentry | (retry ? IQUIET : 0)));
2959 			if (error == EAGAIN) {
2960 				if (ulp) {
2961 					TRANS_END_CSYNC(ufsvfsp, error, issync,
2962 						TOP_CREATE, trans_size);
2963 					ufs_lockfs_end(ulp);
2964 				}
2965 				goto again;
2966 			}
2967 			rw_exit(&ip->i_rwlock);
2968 		}
2969 		ip = xip;
2970 		if (ip != NULL) {
2971 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2972 			rw_enter(&ip->i_contents, RW_WRITER);
2973 		}
2974 	}
2975 
2976 	/*
2977 	 * If the file already exists and this is a non-exclusive create,
2978 	 * check permissions and allow access for non-directories.
2979 	 * Read-only create of an existing directory is also allowed.
2980 	 * We fail an exclusive create of anything which already exists.
2981 	 */
2982 	if (error == EEXIST) {
2983 		dip = VTOI(dvp);
2984 		if (excl == NONEXCL) {
2985 			if ((((ip->i_mode & IFMT) == IFDIR) ||
2986 			    ((ip->i_mode & IFMT) == IFATTRDIR)) &&
2987 			    (mode & IWRITE))
2988 				error = EISDIR;
2989 			else if (mode)
2990 				error = ufs_iaccess(ip, mode, cr);
2991 			else
2992 				error = 0;
2993 		}
2994 		if (error) {
2995 			rw_exit(&ip->i_contents);
2996 			rw_exit(&ufsvfsp->vfs_dqrwlock);
2997 			VN_RELE(ITOV(ip));
2998 			goto unlock;
2999 		}
3000 		/*
3001 		 * If the error EEXIST was set, then i_seq can not
3002 		 * have been updated. The sequence number interface
3003 		 * is defined such that a non-error VOP_CREATE must
3004 		 * increase the dir va_seq it by at least one. If we
3005 		 * have cleared the error, increase i_seq. Note that
3006 		 * we are increasing the dir i_seq and in rare cases
3007 		 * ip may actually be from the dvp, so we already have
3008 		 * the locks and it will not be subject to truncation.
3009 		 * In case we have to update i_seq of the parent
3010 		 * directory dip, we have to defer it till we have
3011 		 * released our locks on ip due to lock ordering requirements.
3012 		 */
3013 		if (ip != dip)
3014 			defer_dip_seq_update = 1;
3015 		else
3016 			ip->i_seq++;
3017 
3018 		if (((ip->i_mode & IFMT) == IFREG) &&
3019 		    (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
3020 			/*
3021 			 * Truncate regular files, if requested by caller.
3022 			 * Grab i_rwlock to make sure no one else is
3023 			 * currently writing to the file (we promised
3024 			 * bmap we would do this).
3025 			 * Must get the locks in the correct order.
3026 			 */
3027 			if (ip->i_size == 0) {
3028 				ip->i_flag |= ICHG | IUPD;
3029 				ip->i_seq++;
3030 				TRANS_INODE(ufsvfsp, ip);
3031 			} else {
3032 				/*
3033 				 * Large Files: Why this check here?
3034 				 * Though we do it in vn_create() we really
3035 				 * want to guarantee that we do not destroy
3036 				 * Large file data by atomically checking
3037 				 * the size while holding the contents
3038 				 * lock.
3039 				 */
3040 				if (flag && !(flag & FOFFMAX) &&
3041 				    ((ip->i_mode & IFMT) == IFREG) &&
3042 				    (ip->i_size > (offset_t)MAXOFF32_T)) {
3043 					rw_exit(&ip->i_contents);
3044 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3045 					error = EOVERFLOW;
3046 					goto unlock;
3047 				}
3048 				if (TRANS_ISTRANS(ufsvfsp))
3049 					truncflag++;
3050 				else {
3051 					rw_exit(&ip->i_contents);
3052 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3053 					ufs_tryirwlock_trans(&ip->i_rwlock,
3054 							RW_WRITER, TOP_CREATE,
3055 							retry_file);
3056 					if (indeadlock) {
3057 						VN_RELE(ITOV(ip));
3058 						goto again;
3059 					}
3060 					rw_enter(&ufsvfsp->vfs_dqrwlock,
3061 							RW_READER);
3062 					rw_enter(&ip->i_contents, RW_WRITER);
3063 					(void) ufs_itrunc(ip, (u_offset_t)0, 0,
3064 								cr);
3065 					rw_exit(&ip->i_rwlock);
3066 				}
3067 			}
3068 		}
3069 	}
3070 
3071 	if (error) {
3072 		if (ip != NULL) {
3073 			rw_exit(&ufsvfsp->vfs_dqrwlock);
3074 			rw_exit(&ip->i_contents);
3075 		}
3076 		goto unlock;
3077 	}
3078 
3079 	*vpp = ITOV(ip);
3080 	ITIMES(ip);
3081 	rw_exit(&ip->i_contents);
3082 	rw_exit(&ufsvfsp->vfs_dqrwlock);
3083 
3084 	/*
3085 	 * If vnode is a device return special vnode instead.
3086 	 */
3087 	if (!error && IS_DEVVP(*vpp)) {
3088 		struct vnode *newvp;
3089 
3090 		newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
3091 		VN_RELE(*vpp);
3092 		if (newvp == NULL) {
3093 			error = ENOSYS;
3094 			goto unlock;
3095 		}
3096 		truncflag = 0;
3097 		*vpp = newvp;
3098 	}
3099 unlock:
3100 
3101 	/*
3102 	 * Do the deferred update of the parent directory's sequence
3103 	 * number now.
3104 	 */
3105 	if (defer_dip_seq_update == 1) {
3106 		rw_enter(&dip->i_contents, RW_READER);
3107 		mutex_enter(&dip->i_tlock);
3108 		dip->i_seq++;
3109 		mutex_exit(&dip->i_tlock);
3110 		rw_exit(&dip->i_contents);
3111 	}
3112 
3113 	if (ulp) {
3114 		int terr = 0;
3115 
3116 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE,
3117 		    trans_size);
3118 
3119 		/*
3120 		 * If we haven't had a more interesting failure
3121 		 * already, then anything that might've happened
3122 		 * here should be reported.
3123 		 */
3124 		if (error == 0)
3125 			error = terr;
3126 	}
3127 
3128 	if (!error && truncflag) {
3129 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_trunc);
3130 		if (indeadlock) {
3131 			if (ulp)
3132 				ufs_lockfs_end(ulp);
3133 			VN_RELE(ITOV(ip));
3134 			goto again;
3135 		}
3136 		(void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr);
3137 		rw_exit(&ip->i_rwlock);
3138 	}
3139 
3140 	if (ulp)
3141 		ufs_lockfs_end(ulp);
3142 
3143 	/*
3144 	 * If no inodes available, try to free one up out of the
3145 	 * pending delete queue.
3146 	 */
3147 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3148 		ufs_delete_drain_wait(ufsvfsp, 1);
3149 		retry = 0;
3150 		goto again;
3151 	}
3152 
3153 out:
3154 	TRACE_3(TR_FAC_UFS, TR_UFS_CREATE_END,
3155 		"ufs_create_end:dvp %p name %s error %d", vpp, name, error);
3156 	return (error);
3157 }
3158 
3159 extern int ufs_idle_max;
3160 /*ARGSUSED*/
3161 static int
3162 ufs_remove(struct vnode *vp, char *nm, struct cred *cr)
3163 {
3164 	struct inode *ip = VTOI(vp);
3165 	struct ufsvfs *ufsvfsp	= ip->i_ufsvfs;
3166 	struct ulockfs *ulp;
3167 	vnode_t *rmvp = NULL;	/* Vnode corresponding to name being removed */
3168 	int indeadlock;
3169 	int error;
3170 	int issync;
3171 	int trans_size;
3172 
3173 	TRACE_1(TR_FAC_UFS, TR_UFS_REMOVE_START,
3174 		"ufs_remove_start:vp %p", vp);
3175 
3176 	/*
3177 	 * don't let the delete queue get too long
3178 	 */
3179 	if (ufsvfsp == NULL) {
3180 		error = EIO;
3181 		goto out;
3182 	}
3183 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3184 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3185 
3186 retry_remove:
3187 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK);
3188 	if (error)
3189 		goto out;
3190 
3191 	if (ulp)
3192 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE,
3193 		    trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp)));
3194 
3195 	/*
3196 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3197 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3198 	 * possible, retries the operation.
3199 	 */
3200 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_REMOVE, retry);
3201 	if (indeadlock)
3202 		goto retry_remove;
3203 	error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0,
3204 	    DR_REMOVE, cr, &rmvp);
3205 	rw_exit(&ip->i_rwlock);
3206 
3207 	if (ulp) {
3208 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size);
3209 		ufs_lockfs_end(ulp);
3210 	}
3211 
3212 	/*
3213 	 * This must be called after the remove transaction is closed.
3214 	 */
3215 	if (rmvp != NULL) {
3216 		/* Only send the event if there were no errors */
3217 		if (error == 0)
3218 			vnevent_remove(rmvp);
3219 		VN_RELE(rmvp);
3220 	}
3221 out:
3222 	TRACE_3(TR_FAC_UFS, TR_UFS_REMOVE_END,
3223 		"ufs_remove_end:vp %p name %s error %d", vp, nm, error);
3224 	return (error);
3225 }
3226 
3227 /*
3228  * Link a file or a directory.  Only privileged processes are allowed to
3229  * make links to directories.
3230  */
3231 static int
3232 ufs_link(struct vnode *tdvp, struct vnode *svp, char *tnm, struct cred *cr)
3233 {
3234 	struct inode *sip;
3235 	struct inode *tdp = VTOI(tdvp);
3236 	struct ufsvfs *ufsvfsp = tdp->i_ufsvfs;
3237 	struct ulockfs *ulp;
3238 	struct vnode *realvp;
3239 	int error;
3240 	int issync;
3241 	int trans_size;
3242 	int isdev;
3243 	int indeadlock;
3244 
3245 	TRACE_1(TR_FAC_UFS, TR_UFS_LINK_START,
3246 		"ufs_link_start:tdvp %p", tdvp);
3247 
3248 retry_link:
3249 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LINK_MASK);
3250 	if (error)
3251 		goto out;
3252 
3253 	if (ulp)
3254 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_LINK,
3255 		    trans_size = (int)TOP_LINK_SIZE(VTOI(tdvp)));
3256 
3257 	if (VOP_REALVP(svp, &realvp) == 0)
3258 		svp = realvp;
3259 
3260 	/*
3261 	 * Make sure link for extended attributes is valid
3262 	 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3263 	 *
3264 	 * Make certain we don't attempt to look at a device node as
3265 	 * a ufs inode.
3266 	 */
3267 
3268 	isdev = IS_DEVVP(svp);
3269 	if (((isdev == 0) && ((VTOI(svp)->i_cflags & IXATTR) == 0) &&
3270 	    ((tdp->i_mode & IFMT) == IFATTRDIR)) ||
3271 	    ((isdev == 0) && (VTOI(svp)->i_cflags & IXATTR) &&
3272 	    ((tdp->i_mode & IFMT) == IFDIR))) {
3273 		error = EINVAL;
3274 		goto unlock;
3275 	}
3276 
3277 	sip = VTOI(svp);
3278 	if ((svp->v_type == VDIR &&
3279 	    secpolicy_fs_linkdir(cr, ufsvfsp->vfs_vfs) != 0) ||
3280 	    (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)) {
3281 		error = EPERM;
3282 		goto unlock;
3283 	}
3284 
3285 	/*
3286 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3287 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3288 	 * possible, retries the operation.
3289 	 */
3290 	ufs_tryirwlock_trans(&tdp->i_rwlock, RW_WRITER, TOP_LINK, retry);
3291 	if (indeadlock)
3292 		goto retry_link;
3293 	error = ufs_direnter_lr(tdp, tnm, DE_LINK, (struct inode *)0,
3294 	    sip, cr, NULL);
3295 	rw_exit(&tdp->i_rwlock);
3296 
3297 unlock:
3298 	if (ulp) {
3299 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_LINK, trans_size);
3300 		ufs_lockfs_end(ulp);
3301 	}
3302 out:
3303 	TRACE_2(TR_FAC_UFS, TR_UFS_LINK_END,
3304 		"ufs_link_end:tdvp %p error %d", tdvp, error);
3305 	return (error);
3306 }
3307 
3308 uint64_t ufs_rename_retry_cnt;
3309 uint64_t ufs_rename_upgrade_retry_cnt;
3310 uint64_t ufs_rename_dircheck_retry_cnt;
3311 clock_t	 ufs_rename_backoff_delay = 1;
3312 
3313 /*
3314  * Rename a file or directory.
3315  * We are given the vnode and entry string of the source and the
3316  * vnode and entry string of the place we want to move the source
3317  * to (the target). The essential operation is:
3318  *	unlink(target);
3319  *	link(source, target);
3320  *	unlink(source);
3321  * but "atomically".  Can't do full commit without saving state in
3322  * the inode on disk, which isn't feasible at this time.  Best we
3323  * can do is always guarantee that the TARGET exists.
3324  */
3325 
3326 /*ARGSUSED*/
3327 static int
3328 ufs_rename(
3329 	struct vnode *sdvp,		/* old (source) parent vnode */
3330 	char *snm,			/* old (source) entry name */
3331 	struct vnode *tdvp,		/* new (target) parent vnode */
3332 	char *tnm,			/* new (target) entry name */
3333 	struct cred *cr)
3334 {
3335 	struct inode *sip = NULL;	/* source inode */
3336 	struct inode *ip = NULL;	/* check inode */
3337 	struct inode *sdp;		/* old (source) parent inode */
3338 	struct inode *tdp;		/* new (target) parent inode */
3339 	struct vnode *tvp = NULL;	/* target vnode, if it exists */
3340 	struct vnode *realvp;
3341 	struct ufsvfs *ufsvfsp;
3342 	struct ulockfs *ulp;
3343 	struct ufs_slot slot;
3344 	timestruc_t now;
3345 	int error;
3346 	int issync;
3347 	int trans_size;
3348 	krwlock_t *first_lock;
3349 	krwlock_t *second_lock;
3350 	krwlock_t *reverse_lock;
3351 
3352 	TRACE_1(TR_FAC_UFS, TR_UFS_RENAME_START,
3353 		"ufs_rename_start:sdvp %p", sdvp);
3354 
3355 
3356 	sdp = VTOI(sdvp);
3357 	slot.fbp = NULL;
3358 	ufsvfsp = sdp->i_ufsvfs;
3359 retry_rename:
3360 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
3361 	if (error)
3362 		goto out;
3363 
3364 	if (ulp)
3365 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
3366 		    trans_size = (int)TOP_RENAME_SIZE(sdp));
3367 
3368 	if (VOP_REALVP(tdvp, &realvp) == 0)
3369 		tdvp = realvp;
3370 
3371 	tdp = VTOI(tdvp);
3372 
3373 
3374 	/*
3375 	 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3376 	 */
3377 	if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) {
3378 		error = EINVAL;
3379 		goto unlock;
3380 	}
3381 
3382 	/*
3383 	 * Look up inode of file we're supposed to rename.
3384 	 */
3385 	gethrestime(&now);
3386 	if (error = ufs_dirlook(sdp, snm, &sip, cr, 0)) {
3387 		if (error == EAGAIN) {
3388 			if (ulp) {
3389 				TRANS_END_CSYNC(ufsvfsp, error, issync,
3390 					TOP_RENAME, trans_size);
3391 				ufs_lockfs_end(ulp);
3392 			}
3393 			goto retry_rename;
3394 		}
3395 
3396 		goto unlock;
3397 	}
3398 
3399 	/*
3400 	 * Lock both the source and target directories (they may be
3401 	 * the same) to provide the atomicity semantics that was
3402 	 * previously provided by the per file system vfs_rename_lock
3403 	 *
3404 	 * with vfs_rename_lock removed to allow simultaneous renames
3405 	 * within a file system, ufs_dircheckpath can deadlock while
3406 	 * traversing back to ensure that source is not a parent directory
3407 	 * of target parent directory. This is because we get into
3408 	 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3409 	 * If the tdp and sdp of the simultaneous renames happen to be
3410 	 * in the path of each other, it can lead to a deadlock. This
3411 	 * can be avoided by getting the locks as RW_READER here and then
3412 	 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3413 	 *
3414 	 * We hold the target directory's i_rwlock after calling
3415 	 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3416 	 * VOP_RWLOCK is explicitly called by the filesystem independent code
3417 	 * before calling the file system operation. In these cases the order
3418 	 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3419 	 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3420 	 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3421 	 * synchronizing object which might lead to a deadlock. So we use
3422 	 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3423 	 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3424 	 * operation.
3425 	 */
3426 retry:
3427 	first_lock = &tdp->i_rwlock;
3428 	second_lock = &sdp->i_rwlock;
3429 retry_firstlock:
3430 	if (!rw_tryenter(first_lock, RW_READER)) {
3431 		/*
3432 		 * We didn't get the lock. Check if the SLOCK is set in the
3433 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3434 		 * and wait for SLOCK to be cleared.
3435 		 */
3436 
3437 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3438 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3439 					trans_size);
3440 			ufs_lockfs_end(ulp);
3441 			goto retry_rename;
3442 
3443 		} else {
3444 			/*
3445 			 * SLOCK isn't set so this is a genuine synchronization
3446 			 * case. Let's try again after giving them a breather.
3447 			 */
3448 			delay(RETRY_LOCK_DELAY);
3449 			goto  retry_firstlock;
3450 		}
3451 	}
3452 	/*
3453 	 * Need to check if the tdp and sdp are same !!!
3454 	 */
3455 	if ((tdp != sdp) && (!rw_tryenter(second_lock, RW_READER))) {
3456 		/*
3457 		 * We didn't get the lock. Check if the SLOCK is set in the
3458 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3459 		 * and wait for SLOCK to be cleared.
3460 		 */
3461 
3462 		rw_exit(first_lock);
3463 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3464 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3465 					trans_size);
3466 			ufs_lockfs_end(ulp);
3467 			goto retry_rename;
3468 
3469 		} else {
3470 			/*
3471 			 * So we couldn't get the second level peer lock *and*
3472 			 * the SLOCK bit isn't set. Too bad we can be
3473 			 * contentding with someone wanting these locks otherway
3474 			 * round. Reverse the locks in case there is a heavy
3475 			 * contention for the second level lock.
3476 			 */
3477 			reverse_lock = first_lock;
3478 			first_lock = second_lock;
3479 			second_lock = reverse_lock;
3480 			ufs_rename_retry_cnt++;
3481 			goto  retry_firstlock;
3482 		}
3483 	}
3484 
3485 	if (sip == tdp) {
3486 		error = EINVAL;
3487 		goto errout;
3488 	}
3489 	/*
3490 	 * Make sure we can delete the source entry.  This requires
3491 	 * write permission on the containing directory.
3492 	 * Check for sticky directories.
3493 	 */
3494 	rw_enter(&sdp->i_contents, RW_READER);
3495 	rw_enter(&sip->i_contents, RW_READER);
3496 	if ((error = ufs_iaccess(sdp, IWRITE, cr)) != 0 ||
3497 	    (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) {
3498 		rw_exit(&sip->i_contents);
3499 		rw_exit(&sdp->i_contents);
3500 		goto errout;
3501 	}
3502 
3503 	/*
3504 	 * If this is a rename of a directory and the parent is
3505 	 * different (".." must be changed), then the source
3506 	 * directory must not be in the directory hierarchy
3507 	 * above the target, as this would orphan everything
3508 	 * below the source directory.  Also the user must have
3509 	 * write permission in the source so as to be able to
3510 	 * change "..".
3511 	 */
3512 	if ((((sip->i_mode & IFMT) == IFDIR) ||
3513 	    ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) {
3514 		ino_t	inum;
3515 
3516 		if ((error = ufs_iaccess(sip, IWRITE, cr))) {
3517 			rw_exit(&sip->i_contents);
3518 			rw_exit(&sdp->i_contents);
3519 			goto errout;
3520 		}
3521 		inum = sip->i_number;
3522 		rw_exit(&sip->i_contents);
3523 		rw_exit(&sdp->i_contents);
3524 		if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) {
3525 			/*
3526 			 * If we got EAGAIN ufs_dircheckpath detected a
3527 			 * potential deadlock and backed out. We need
3528 			 * to retry the operation since sdp and tdp have
3529 			 * to be released to avoid the deadlock.
3530 			 */
3531 			if (error == EAGAIN) {
3532 				rw_exit(&tdp->i_rwlock);
3533 				if (tdp != sdp)
3534 					rw_exit(&sdp->i_rwlock);
3535 				delay(ufs_rename_backoff_delay);
3536 				ufs_rename_dircheck_retry_cnt++;
3537 				goto retry;
3538 			}
3539 			goto errout;
3540 		}
3541 	} else {
3542 		rw_exit(&sip->i_contents);
3543 		rw_exit(&sdp->i_contents);
3544 	}
3545 
3546 
3547 	/*
3548 	 * Check for renaming '.' or '..' or alias of '.'
3549 	 */
3550 	if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) {
3551 		error = EINVAL;
3552 		goto errout;
3553 	}
3554 
3555 	/*
3556 	 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3557 	 * tries to traverse back the file tree with both tdp and sdp held
3558 	 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3559 	 * as RW_READERS  till ufs_dircheckpath is done.
3560 	 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3561 	 * to RW_WRITER.
3562 	 */
3563 	if (!rw_tryupgrade(&tdp->i_rwlock)) {
3564 		/*
3565 		 * The upgrade failed. We got to give away the lock
3566 		 * as to avoid deadlocking with someone else who is
3567 		 * waiting for writer lock. With the lock gone, we
3568 		 * cannot be sure the checks done above will hold
3569 		 * good when we eventually get them back as writer.
3570 		 * So if we can't upgrade we drop the locks and retry
3571 		 * everything again.
3572 		 */
3573 		rw_exit(&tdp->i_rwlock);
3574 		if (tdp != sdp)
3575 			rw_exit(&sdp->i_rwlock);
3576 		delay(ufs_rename_backoff_delay);
3577 		ufs_rename_upgrade_retry_cnt++;
3578 		goto retry;
3579 	}
3580 	if (tdp != sdp) {
3581 		if (!rw_tryupgrade(&sdp->i_rwlock)) {
3582 			/*
3583 			 * The upgrade failed. We got to give away the lock
3584 			 * as to avoid deadlocking with someone else who is
3585 			 * waiting for writer lock. With the lock gone, we
3586 			 * cannot be sure the checks done above will hold
3587 			 * good when we eventually get them back as writer.
3588 			 * So if we can't upgrade we drop the locks and retry
3589 			 * everything again.
3590 			 */
3591 			rw_exit(&tdp->i_rwlock);
3592 			rw_exit(&sdp->i_rwlock);
3593 			delay(ufs_rename_backoff_delay);
3594 			ufs_rename_upgrade_retry_cnt++;
3595 			goto retry;
3596 		}
3597 	}
3598 
3599 	/*
3600 	 * Now that all the locks are held check to make sure another thread
3601 	 * didn't slip in and take out the sip.
3602 	 */
3603 	slot.status = NONE;
3604 	if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec ||
3605 	    sip->i_ctime.tv_sec > now.tv_sec) {
3606 		rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER);
3607 		rw_enter(&sdp->i_contents, RW_WRITER);
3608 		error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot,
3609 		    &ip, cr, 0);
3610 		rw_exit(&sdp->i_contents);
3611 		rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock);
3612 		if (error) {
3613 			goto errout;
3614 		}
3615 		if (ip == NULL) {
3616 			error = ENOENT;
3617 			goto errout;
3618 		} else {
3619 			/*
3620 			 * If the inode was found need to drop the v_count
3621 			 * so as not to keep the filesystem from being
3622 			 * unmounted at a later time.
3623 			 */
3624 			VN_RELE(ITOV(ip));
3625 		}
3626 
3627 		/*
3628 		 * Release the slot.fbp that has the page mapped and
3629 		 * locked SE_SHARED, and could be used in in
3630 		 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3631 		 * on said page.
3632 		 */
3633 		if (slot.fbp) {
3634 			fbrelse(slot.fbp, S_OTHER);
3635 			slot.fbp = NULL;
3636 		}
3637 	}
3638 
3639 	/*
3640 	 * Link source to the target.  If a target exists, return its
3641 	 * vnode pointer in tvp.  We'll release it after sending the
3642 	 * vnevent.
3643 	 */
3644 	if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr, &tvp)) {
3645 		/*
3646 		 * ESAME isn't really an error; it indicates that the
3647 		 * operation should not be done because the source and target
3648 		 * are the same file, but that no error should be reported.
3649 		 */
3650 		if (error == ESAME)
3651 			error = 0;
3652 		goto errout;
3653 	}
3654 
3655 	/*
3656 	 * Unlink the source.
3657 	 * Remove the source entry.  ufs_dirremove() checks that the entry
3658 	 * still reflects sip, and returns an error if it doesn't.
3659 	 * If the entry has changed just forget about it.  Release
3660 	 * the source inode.
3661 	 */
3662 	if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0,
3663 	    DR_RENAME, cr, NULL)) == ENOENT)
3664 		error = 0;
3665 
3666 errout:
3667 	if (slot.fbp)
3668 		fbrelse(slot.fbp, S_OTHER);
3669 
3670 	rw_exit(&tdp->i_rwlock);
3671 	if (sdp != tdp) {
3672 		rw_exit(&sdp->i_rwlock);
3673 	}
3674 
3675 unlock:
3676 	if (ulp) {
3677 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
3678 		ufs_lockfs_end(ulp);
3679 	}
3680 
3681 	/*
3682 	 * If no errors, send the appropriate events on the source
3683 	 * and destination (a.k.a, target) vnodes, if they exist.
3684 	 * This has to be done after the rename transaction has closed.
3685 	 */
3686 	if (error == 0) {
3687 		if (tvp != NULL)
3688 			vnevent_rename_dest(tvp);
3689 		/*
3690 		 * Note that if ufs_direnter_lr() returned ESAME then
3691 		 * this event will still be sent.  This isn't expected
3692 		 * to be a problem for anticipated usage by consumers.
3693 		 */
3694 		if (sip != NULL)
3695 			vnevent_rename_src(ITOV(sip));
3696 	}
3697 
3698 	if (tvp != NULL)
3699 		VN_RELE(tvp);
3700 
3701 	if (sip != NULL)
3702 		VN_RELE(ITOV(sip));
3703 
3704 out:
3705 	TRACE_5(TR_FAC_UFS, TR_UFS_RENAME_END,
3706 		"ufs_rename_end:sdvp %p snm %s tdvp %p tnm %s error %d",
3707 			sdvp, snm, tdvp, tnm, error);
3708 	return (error);
3709 }
3710 
3711 /*ARGSUSED*/
3712 static int
3713 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap,
3714 	struct vnode **vpp, struct cred *cr)
3715 {
3716 	struct inode *ip;
3717 	struct inode *xip;
3718 	struct ufsvfs *ufsvfsp;
3719 	struct ulockfs *ulp;
3720 	int error;
3721 	int issync;
3722 	int trans_size;
3723 	int indeadlock;
3724 	int retry = 1;
3725 
3726 	ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
3727 
3728 	TRACE_1(TR_FAC_UFS, TR_UFS_MKDIR_START,
3729 		"ufs_mkdir_start:dvp %p", dvp);
3730 
3731 	/*
3732 	 * Can't make directory in attr hidden dir
3733 	 */
3734 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3735 		return (EINVAL);
3736 
3737 again:
3738 	ip = VTOI(dvp);
3739 	ufsvfsp = ip->i_ufsvfs;
3740 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK);
3741 	if (error)
3742 		goto out;
3743 	if (ulp)
3744 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR,
3745 		    trans_size = (int)TOP_MKDIR_SIZE(ip));
3746 
3747 	/*
3748 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3749 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3750 	 * possible, retries the operation.
3751 	 */
3752 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_MKDIR, retry);
3753 	if (indeadlock)
3754 		goto again;
3755 
3756 	error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr,
3757 		(retry ? IQUIET : 0));
3758 	if (error == EAGAIN) {
3759 		if (ulp) {
3760 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_MKDIR,
3761 					trans_size);
3762 			ufs_lockfs_end(ulp);
3763 		}
3764 		goto again;
3765 	}
3766 
3767 	rw_exit(&ip->i_rwlock);
3768 	if (error == 0) {
3769 		ip = xip;
3770 		*vpp = ITOV(ip);
3771 	} else if (error == EEXIST)
3772 		VN_RELE(ITOV(xip));
3773 
3774 	if (ulp) {
3775 		int terr = 0;
3776 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size);
3777 		ufs_lockfs_end(ulp);
3778 		if (error == 0)
3779 			error = terr;
3780 	}
3781 out:
3782 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3783 		ufs_delete_drain_wait(ufsvfsp, 1);
3784 		retry = 0;
3785 		goto again;
3786 	}
3787 
3788 	TRACE_2(TR_FAC_UFS, TR_UFS_MKDIR_END,
3789 		"ufs_mkdir_end:dvp %p error %d", dvp, error);
3790 	return (error);
3791 }
3792 
3793 /*ARGSUSED*/
3794 static int
3795 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr)
3796 {
3797 	struct inode *ip = VTOI(vp);
3798 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3799 	struct ulockfs *ulp;
3800 	vnode_t *rmvp = NULL;	/* Vnode of removed directory */
3801 	int error;
3802 	int issync;
3803 	int trans_size;
3804 	int indeadlock;
3805 
3806 	TRACE_1(TR_FAC_UFS, TR_UFS_RMDIR_START,
3807 		"ufs_rmdir_start:vp %p", vp);
3808 
3809 	/*
3810 	 * don't let the delete queue get too long
3811 	 */
3812 	if (ufsvfsp == NULL) {
3813 		error = EIO;
3814 		goto out;
3815 	}
3816 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3817 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3818 
3819 retry_rmdir:
3820 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK);
3821 	if (error)
3822 		goto out;
3823 
3824 	if (ulp)
3825 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR,
3826 					trans_size = TOP_RMDIR_SIZE);
3827 
3828 	/*
3829 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3830 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3831 	 * possible, retries the operation.
3832 	 */
3833 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_RMDIR, retry);
3834 	if (indeadlock)
3835 		goto retry_rmdir;
3836 	error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr,
3837 									&rmvp);
3838 	rw_exit(&ip->i_rwlock);
3839 
3840 	if (ulp) {
3841 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR,
3842 				trans_size);
3843 		ufs_lockfs_end(ulp);
3844 	}
3845 
3846 	/*
3847 	 * This must be done AFTER the rmdir transaction has closed.
3848 	 */
3849 	if (rmvp != NULL) {
3850 		/* Only send the event if there were no errors */
3851 		if (error == 0)
3852 			vnevent_rmdir(rmvp);
3853 		VN_RELE(rmvp);
3854 	}
3855 out:
3856 	TRACE_2(TR_FAC_UFS, TR_UFS_RMDIR_END,
3857 		"ufs_rmdir_end:vp %p error %d", vp, error);
3858 
3859 	return (error);
3860 }
3861 
3862 /* ARGSUSED */
3863 static int
3864 ufs_readdir(
3865 	struct vnode *vp,
3866 	struct uio *uiop,
3867 	struct cred *cr,
3868 	int *eofp)
3869 {
3870 	struct iovec *iovp;
3871 	struct inode *ip;
3872 	struct direct *idp;
3873 	struct dirent64 *odp;
3874 	struct fbuf *fbp;
3875 	struct ufsvfs *ufsvfsp;
3876 	struct ulockfs *ulp;
3877 	caddr_t outbuf;
3878 	size_t bufsize;
3879 	uint_t offset;
3880 	uint_t bytes_wanted, total_bytes_wanted;
3881 	int incount = 0;
3882 	int outcount = 0;
3883 	int error;
3884 
3885 	ip = VTOI(vp);
3886 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
3887 
3888 	TRACE_2(TR_FAC_UFS, TR_UFS_READDIR_START,
3889 		"ufs_readdir_start:vp %p uiop %p", vp, uiop);
3890 
3891 	if (uiop->uio_loffset >= MAXOFF32_T) {
3892 		if (eofp)
3893 			*eofp = 1;
3894 		return (0);
3895 	}
3896 
3897 	/*
3898 	 * Check if we have been called with a valid iov_len
3899 	 * and bail out if not, otherwise we may potentially loop
3900 	 * forever further down.
3901 	 */
3902 	if (uiop->uio_iov->iov_len <= 0) {
3903 		error = EINVAL;
3904 		goto out;
3905 	}
3906 
3907 	/*
3908 	 * Large Files: When we come here we are guaranteed that
3909 	 * uio_offset can be used safely. The high word is zero.
3910 	 */
3911 
3912 	ufsvfsp = ip->i_ufsvfs;
3913 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READDIR_MASK);
3914 	if (error)
3915 		goto out;
3916 
3917 	iovp = uiop->uio_iov;
3918 	total_bytes_wanted = iovp->iov_len;
3919 
3920 	/* Large Files: directory files should not be "large" */
3921 
3922 	ASSERT(ip->i_size <= MAXOFF32_T);
3923 
3924 	/* Force offset to be valid (to guard against bogus lseek() values) */
3925 	offset = (uint_t)uiop->uio_offset & ~(DIRBLKSIZ - 1);
3926 
3927 	/* Quit if at end of file or link count of zero (posix) */
3928 	if (offset >= (uint_t)ip->i_size || ip->i_nlink <= 0) {
3929 		if (eofp)
3930 			*eofp = 1;
3931 		error = 0;
3932 		goto unlock;
3933 	}
3934 
3935 	/*
3936 	 * Get space to change directory entries into fs independent format.
3937 	 * Do fast alloc for the most commonly used-request size (filesystem
3938 	 * block size).
3939 	 */
3940 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) {
3941 		bufsize = total_bytes_wanted;
3942 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
3943 		odp = (struct dirent64 *)outbuf;
3944 	} else {
3945 		bufsize = total_bytes_wanted;
3946 		odp = (struct dirent64 *)iovp->iov_base;
3947 	}
3948 
3949 nextblk:
3950 	bytes_wanted = total_bytes_wanted;
3951 
3952 	/* Truncate request to file size */
3953 	if (offset + bytes_wanted > (int)ip->i_size)
3954 		bytes_wanted = (int)(ip->i_size - offset);
3955 
3956 	/* Comply with MAXBSIZE boundary restrictions of fbread() */
3957 	if ((offset & MAXBOFFSET) + bytes_wanted > MAXBSIZE)
3958 		bytes_wanted = MAXBSIZE - (offset & MAXBOFFSET);
3959 
3960 	/*
3961 	 * Read in the next chunk.
3962 	 * We are still holding the i_rwlock.
3963 	 */
3964 	error = fbread(vp, (offset_t)offset, bytes_wanted, S_OTHER, &fbp);
3965 
3966 	if (error)
3967 		goto update_inode;
3968 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (ip->i_fs->fs_ronly == 0) &&
3969 	    (!ufsvfsp->vfs_noatime)) {
3970 		ip->i_flag |= IACC;
3971 	}
3972 	incount = 0;
3973 	idp = (struct direct *)fbp->fb_addr;
3974 	if (idp->d_ino == 0 && idp->d_reclen == 0 &&
3975 		idp->d_namlen == 0) {
3976 		cmn_err(CE_WARN, "ufs_readdir: bad dir, inumber = %llu, "
3977 			"fs = %s\n",
3978 			(u_longlong_t)ip->i_number, ufsvfsp->vfs_fs->fs_fsmnt);
3979 		fbrelse(fbp, S_OTHER);
3980 		error = ENXIO;
3981 		goto update_inode;
3982 	}
3983 	/* Transform to file-system independent format */
3984 	while (incount < bytes_wanted) {
3985 		/*
3986 		 * If the current directory entry is mangled, then skip
3987 		 * to the next block.  It would be nice to set the FSBAD
3988 		 * flag in the super-block so that a fsck is forced on
3989 		 * next reboot, but locking is a problem.
3990 		 */
3991 		if (idp->d_reclen & 0x3) {
3992 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
3993 			break;
3994 		}
3995 
3996 		/* Skip to requested offset and skip empty entries */
3997 		if (idp->d_ino != 0 && offset >= (uint_t)uiop->uio_offset) {
3998 			ushort_t this_reclen =
3999 			    DIRENT64_RECLEN(idp->d_namlen);
4000 			/* Buffer too small for any entries */
4001 			if (!outcount && this_reclen > bufsize) {
4002 				fbrelse(fbp, S_OTHER);
4003 				error = EINVAL;
4004 				goto update_inode;
4005 			}
4006 			/* If would overrun the buffer, quit */
4007 			if (outcount + this_reclen > bufsize) {
4008 				break;
4009 			}
4010 			/* Take this entry */
4011 			odp->d_ino = (ino64_t)idp->d_ino;
4012 			odp->d_reclen = (ushort_t)this_reclen;
4013 			odp->d_off = (offset_t)(offset + idp->d_reclen);
4014 
4015 			/* use strncpy(9f) to zero out uninitialized bytes */
4016 
4017 			ASSERT(strlen(idp->d_name) + 1 <=
4018 			    DIRENT64_NAMELEN(this_reclen));
4019 			(void) strncpy(odp->d_name, idp->d_name,
4020 			    DIRENT64_NAMELEN(this_reclen));
4021 			outcount += odp->d_reclen;
4022 			odp = (struct dirent64 *)((intptr_t)odp +
4023 				    odp->d_reclen);
4024 			ASSERT(outcount <= bufsize);
4025 		}
4026 		if (idp->d_reclen) {
4027 			incount += idp->d_reclen;
4028 			offset += idp->d_reclen;
4029 			idp = (struct direct *)((intptr_t)idp + idp->d_reclen);
4030 		} else {
4031 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4032 			break;
4033 		}
4034 	}
4035 	/* Release the chunk */
4036 	fbrelse(fbp, S_OTHER);
4037 
4038 	/* Read whole block, but got no entries, read another if not eof */
4039 
4040 	/*
4041 	 * Large Files: casting i_size to int here is not a problem
4042 	 * because directory sizes are always less than MAXOFF32_T.
4043 	 * See assertion above.
4044 	 */
4045 
4046 	if (offset < (int)ip->i_size && !outcount)
4047 		goto nextblk;
4048 
4049 	/* Copy out the entry data */
4050 	if (uiop->uio_segflg == UIO_SYSSPACE && uiop->uio_iovcnt == 1) {
4051 		iovp->iov_base += outcount;
4052 		iovp->iov_len -= outcount;
4053 		uiop->uio_resid -= outcount;
4054 		uiop->uio_offset = offset;
4055 	} else if ((error = uiomove(outbuf, (long)outcount, UIO_READ,
4056 				    uiop)) == 0)
4057 		uiop->uio_offset = offset;
4058 update_inode:
4059 	ITIMES(ip);
4060 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1)
4061 		kmem_free(outbuf, bufsize);
4062 
4063 	if (eofp && error == 0)
4064 		*eofp = (uiop->uio_offset >= (int)ip->i_size);
4065 unlock:
4066 	if (ulp) {
4067 		ufs_lockfs_end(ulp);
4068 	}
4069 out:
4070 	TRACE_2(TR_FAC_UFS, TR_UFS_READDIR_END,
4071 		"ufs_readdir_end:vp %p error %d", vp, error);
4072 	return (error);
4073 }
4074 
4075 /*ARGSUSED*/
4076 static int
4077 ufs_symlink(
4078 	struct vnode *dvp,		/* ptr to parent dir vnode */
4079 	char *linkname,			/* name of symbolic link */
4080 	struct vattr *vap,		/* attributes */
4081 	char *target,			/* target path */
4082 	struct cred *cr)		/* user credentials */
4083 {
4084 	struct inode *ip, *dip = VTOI(dvp);
4085 	struct ufsvfs *ufsvfsp = dip->i_ufsvfs;
4086 	struct ulockfs *ulp;
4087 	int error;
4088 	int issync;
4089 	int trans_size;
4090 	int residual;
4091 	int ioflag;
4092 	int retry = 1;
4093 
4094 	TRACE_1(TR_FAC_UFS, TR_UFS_SYMLINK_START,
4095 		"ufs_symlink_start:dvp %p", dvp);
4096 
4097 	/*
4098 	 * No symlinks in attrdirs at this time
4099 	 */
4100 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
4101 		return (EINVAL);
4102 
4103 again:
4104 	ip = (struct inode *)NULL;
4105 	vap->va_type = VLNK;
4106 	vap->va_rdev = 0;
4107 
4108 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK);
4109 	if (error)
4110 		goto out;
4111 
4112 	if (ulp)
4113 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK,
4114 		    trans_size = (int)TOP_SYMLINK_SIZE(dip));
4115 
4116 	/*
4117 	 * We must create the inode before the directory entry, to avoid
4118 	 * racing with readlink().  ufs_dirmakeinode requires that we
4119 	 * hold the quota lock as reader, and directory locks as writer.
4120 	 */
4121 
4122 	rw_enter(&dip->i_rwlock, RW_WRITER);
4123 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4124 	rw_enter(&dip->i_contents, RW_WRITER);
4125 
4126 	/*
4127 	 * Suppress any out of inodes messages if we will retry on
4128 	 * ENOSP
4129 	 */
4130 	if (retry)
4131 		dip->i_flag |= IQUIET;
4132 
4133 	error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr);
4134 
4135 	dip->i_flag &= ~IQUIET;
4136 
4137 	rw_exit(&dip->i_contents);
4138 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4139 	rw_exit(&dip->i_rwlock);
4140 
4141 	if (error)
4142 		goto unlock;
4143 
4144 	/*
4145 	 * OK.  The inode has been created.  Write out the data of the
4146 	 * symbolic link.  Since symbolic links are metadata, and should
4147 	 * remain consistent across a system crash, we need to force the
4148 	 * data out synchronously.
4149 	 *
4150 	 * (This is a change from the semantics in earlier releases, which
4151 	 * only created symbolic links synchronously if the semi-documented
4152 	 * 'syncdir' option was set, or if we were being invoked by the NFS
4153 	 * server, which requires symbolic links to be created synchronously.)
4154 	 *
4155 	 * We need to pass in a pointer for the residual length; otherwise
4156 	 * ufs_rdwri() will always return EIO if it can't write the data,
4157 	 * even if the error was really ENOSPC or EDQUOT.
4158 	 */
4159 
4160 	ioflag = FWRITE | FDSYNC;
4161 	residual = 0;
4162 
4163 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4164 	rw_enter(&ip->i_contents, RW_WRITER);
4165 
4166 	/*
4167 	 * Suppress file system full messages if we will retry
4168 	 */
4169 	if (retry)
4170 		ip->i_flag |= IQUIET;
4171 
4172 	error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target),
4173 	    (offset_t)0, UIO_SYSSPACE, &residual, cr);
4174 
4175 	ip->i_flag &= ~IQUIET;
4176 
4177 	if (error) {
4178 		rw_exit(&ip->i_contents);
4179 		rw_exit(&ufsvfsp->vfs_dqrwlock);
4180 		goto remove;
4181 	}
4182 
4183 	/*
4184 	 * If the link's data is small enough, we can cache it in the inode.
4185 	 * This is a "fast symbolic link".  We don't use the first direct
4186 	 * block because that's actually used to point at the symbolic link's
4187 	 * contents on disk; but we know that none of the other direct or
4188 	 * indirect blocks can be used because symbolic links are restricted
4189 	 * to be smaller than a file system block.
4190 	 */
4191 
4192 	ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip)));
4193 
4194 	if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) {
4195 		if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) {
4196 			ip->i_flag |= IFASTSYMLNK;
4197 		} else {
4198 			int i;
4199 			/* error, clear garbage left behind */
4200 			for (i = 1; i < NDADDR; i++)
4201 				ip->i_db[i] = 0;
4202 			for (i = 0; i < NIADDR; i++)
4203 				ip->i_ib[i] = 0;
4204 		}
4205 	}
4206 
4207 	rw_exit(&ip->i_contents);
4208 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4209 
4210 	/*
4211 	 * OK.  We've successfully created the symbolic link.  All that
4212 	 * remains is to insert it into the appropriate directory.
4213 	 */
4214 
4215 	rw_enter(&dip->i_rwlock, RW_WRITER);
4216 	error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr, NULL);
4217 	rw_exit(&dip->i_rwlock);
4218 
4219 	/*
4220 	 * Fall through into remove-on-error code.  We're either done, or we
4221 	 * need to remove the inode (if we couldn't insert it).
4222 	 */
4223 
4224 remove:
4225 	if (error && (ip != NULL)) {
4226 		rw_enter(&ip->i_contents, RW_WRITER);
4227 		ip->i_nlink--;
4228 		ip->i_flag |= ICHG;
4229 		ip->i_seq++;
4230 		ufs_setreclaim(ip);
4231 		rw_exit(&ip->i_contents);
4232 	}
4233 
4234 unlock:
4235 	if (ip != NULL)
4236 		VN_RELE(ITOV(ip));
4237 
4238 	if (ulp) {
4239 		int terr = 0;
4240 
4241 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK,
4242 				trans_size);
4243 		ufs_lockfs_end(ulp);
4244 		if (error == 0)
4245 			error = terr;
4246 	}
4247 
4248 	/*
4249 	 * We may have failed due to lack of an inode or of a block to
4250 	 * store the target in.  Try flushing the delete queue to free
4251 	 * logically-available things up and try again.
4252 	 */
4253 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
4254 		ufs_delete_drain_wait(ufsvfsp, 1);
4255 		retry = 0;
4256 		goto again;
4257 	}
4258 
4259 out:
4260 	TRACE_2(TR_FAC_UFS, TR_UFS_SYMLINK_END,
4261 		"ufs_symlink_end:dvp %p error %d", dvp, error);
4262 	return (error);
4263 }
4264 
4265 /*
4266  * Ufs specific routine used to do ufs io.
4267  */
4268 int
4269 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base,
4270 	ssize_t len, offset_t offset, enum uio_seg seg, int *aresid,
4271 	struct cred *cr)
4272 {
4273 	struct uio auio;
4274 	struct iovec aiov;
4275 	int error;
4276 
4277 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
4278 
4279 	bzero((caddr_t)&auio, sizeof (uio_t));
4280 	bzero((caddr_t)&aiov, sizeof (iovec_t));
4281 
4282 	aiov.iov_base = base;
4283 	aiov.iov_len = len;
4284 	auio.uio_iov = &aiov;
4285 	auio.uio_iovcnt = 1;
4286 	auio.uio_loffset = offset;
4287 	auio.uio_segflg = (short)seg;
4288 	auio.uio_resid = len;
4289 
4290 	if (rw == UIO_WRITE) {
4291 		auio.uio_fmode = FWRITE;
4292 		auio.uio_extflg = UIO_COPY_DEFAULT;
4293 		auio.uio_llimit = curproc->p_fsz_ctl;
4294 		error = wrip(ip, &auio, ioflag, cr);
4295 	} else {
4296 		auio.uio_fmode = FREAD;
4297 		auio.uio_extflg = UIO_COPY_CACHED;
4298 		auio.uio_llimit = MAXOFFSET_T;
4299 		error = rdip(ip, &auio, ioflag, cr);
4300 	}
4301 
4302 	if (aresid) {
4303 		*aresid = auio.uio_resid;
4304 	} else if (auio.uio_resid) {
4305 		error = EIO;
4306 	}
4307 	return (error);
4308 }
4309 
4310 static int
4311 ufs_fid(vp, fidp)
4312 	struct vnode *vp;
4313 	struct fid *fidp;
4314 {
4315 	struct ufid *ufid;
4316 	struct inode *ip = VTOI(vp);
4317 
4318 	if (ip->i_ufsvfs == NULL)
4319 		return (EIO);
4320 
4321 	if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) {
4322 		fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t);
4323 		return (ENOSPC);
4324 	}
4325 
4326 	ufid = (struct ufid *)fidp;
4327 	bzero((char *)ufid, sizeof (struct ufid));
4328 	ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t);
4329 	ufid->ufid_ino = ip->i_number;
4330 	ufid->ufid_gen = ip->i_gen;
4331 
4332 	return (0);
4333 }
4334 
4335 /* ARGSUSED2 */
4336 static int
4337 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4338 {
4339 	struct inode	*ip = VTOI(vp);
4340 	struct ufsvfs	*ufsvfsp;
4341 	int		forcedirectio;
4342 
4343 	/*
4344 	 * Read case is easy.
4345 	 */
4346 	if (!write_lock) {
4347 		rw_enter(&ip->i_rwlock, RW_READER);
4348 		return (V_WRITELOCK_FALSE);
4349 	}
4350 
4351 	/*
4352 	 * Caller has requested a writer lock, but that inhibits any
4353 	 * concurrency in the VOPs that follow. Acquire the lock shared
4354 	 * and defer exclusive access until it is known to be needed in
4355 	 * other VOP handlers. Some cases can be determined here.
4356 	 */
4357 
4358 	/*
4359 	 * If directio is not set, there is no chance of concurrency,
4360 	 * so just acquire the lock exclusive. Beware of a forced
4361 	 * unmount before looking at the mount option.
4362 	 */
4363 	ufsvfsp = ip->i_ufsvfs;
4364 	forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0;
4365 	if (!(ip->i_flag & IDIRECTIO || forcedirectio) ||
4366 	    !ufs_allow_shared_writes) {
4367 		rw_enter(&ip->i_rwlock, RW_WRITER);
4368 		return (V_WRITELOCK_TRUE);
4369 	}
4370 
4371 	/*
4372 	 * Mandatory locking forces acquiring i_rwlock exclusive.
4373 	 */
4374 	if (MANDLOCK(vp, ip->i_mode)) {
4375 		rw_enter(&ip->i_rwlock, RW_WRITER);
4376 		return (V_WRITELOCK_TRUE);
4377 	}
4378 
4379 	/*
4380 	 * Acquire the lock shared in case a concurrent write follows.
4381 	 * Mandatory locking could have become enabled before the lock
4382 	 * was acquired. Re-check and upgrade if needed.
4383 	 */
4384 	rw_enter(&ip->i_rwlock, RW_READER);
4385 	if (MANDLOCK(vp, ip->i_mode)) {
4386 		rw_exit(&ip->i_rwlock);
4387 		rw_enter(&ip->i_rwlock, RW_WRITER);
4388 		return (V_WRITELOCK_TRUE);
4389 	}
4390 	return (V_WRITELOCK_FALSE);
4391 }
4392 
4393 /*ARGSUSED*/
4394 static void
4395 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4396 {
4397 	struct inode	*ip = VTOI(vp);
4398 
4399 	rw_exit(&ip->i_rwlock);
4400 }
4401 
4402 /* ARGSUSED */
4403 static int
4404 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp)
4405 {
4406 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4407 }
4408 
4409 /* ARGSUSED */
4410 static int
4411 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4412 	offset_t offset, struct flk_callback *flk_cbp, struct cred *cr)
4413 {
4414 	struct inode *ip = VTOI(vp);
4415 
4416 	if (ip->i_ufsvfs == NULL)
4417 		return (EIO);
4418 
4419 	/*
4420 	 * If file is being mapped, disallow frlock.
4421 	 * XXX I am not holding tlock while checking i_mapcnt because the
4422 	 * current locking strategy drops all locks before calling fs_frlock.
4423 	 * So, mapcnt could change before we enter fs_frlock making is
4424 	 * meaningless to have held tlock in the first place.
4425 	 */
4426 	if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode))
4427 		return (EAGAIN);
4428 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr));
4429 }
4430 
4431 /* ARGSUSED */
4432 static int
4433 ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4434 	offset_t offset, cred_t *cr, caller_context_t *ct)
4435 {
4436 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
4437 	struct ulockfs *ulp;
4438 	int error;
4439 
4440 	if ((error = convoff(vp, bfp, 0, offset)) == 0) {
4441 		if (cmd == F_FREESP) {
4442 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4443 			    ULOCKFS_SPACE_MASK);
4444 			if (error)
4445 				return (error);
4446 			error = ufs_freesp(vp, bfp, flag, cr);
4447 		} else if (cmd == F_ALLOCSP) {
4448 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4449 			    ULOCKFS_FALLOCATE_MASK);
4450 			if (error)
4451 				return (error);
4452 			error = ufs_allocsp(vp, bfp, cr);
4453 		} else
4454 			return (EINVAL); /* Command not handled here */
4455 
4456 		if (ulp)
4457 			ufs_lockfs_end(ulp);
4458 
4459 	}
4460 	return (error);
4461 }
4462 
4463 /*
4464  * Used to determine if read ahead should be done. Also used to
4465  * to determine when write back occurs.
4466  */
4467 #define	CLUSTSZ(ip)		((ip)->i_ufsvfs->vfs_ioclustsz)
4468 
4469 /*
4470  * A faster version of ufs_getpage.
4471  *
4472  * We optimize by inlining the pvn_getpages iterator, eliminating
4473  * calls to bmap_read if file doesn't have UFS holes, and avoiding
4474  * the overhead of page_exists().
4475  *
4476  * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4477  * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4478  * victimizes performance when a file with UFS holes is faulted
4479  * first in the S_READ mode, and then in the S_WRITE mode. We will get
4480  * two MMU faults in this case.
4481  *
4482  * XXX - the inode fields which control the sequential mode are not
4483  *	 protected by any mutex. The read ahead will act wild if
4484  *	 multiple processes will access the file concurrently and
4485  *	 some of them in sequential mode. One particulary bad case
4486  *	 is if another thread will change the value of i_nextrio between
4487  *	 the time this thread tests the i_nextrio value and then reads it
4488  *	 again to use it as the offset for the read ahead.
4489  */
4490 static int
4491 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
4492 	page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr,
4493 	enum seg_rw rw, struct cred *cr)
4494 {
4495 	u_offset_t	uoff = (u_offset_t)off; /* type conversion */
4496 	u_offset_t	pgoff;
4497 	u_offset_t	eoff;
4498 	struct inode 	*ip = VTOI(vp);
4499 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
4500 	struct fs 	*fs;
4501 	struct ulockfs	*ulp;
4502 	page_t		**pl;
4503 	caddr_t		pgaddr;
4504 	krw_t		rwtype;
4505 	int 		err;
4506 	int		has_holes;
4507 	int		beyond_eof;
4508 	int		seqmode;
4509 	int		pgsize = PAGESIZE;
4510 	int		dolock;
4511 	int		do_qlock;
4512 	int		trans_size;
4513 
4514 	TRACE_1(TR_FAC_UFS, TR_UFS_GETPAGE_START,
4515 		"ufs_getpage_start:vp %p", vp);
4516 
4517 	ASSERT((uoff & PAGEOFFSET) == 0);
4518 
4519 	if (protp)
4520 		*protp = PROT_ALL;
4521 
4522 	/*
4523 	 * Obey the lockfs protocol
4524 	 */
4525 	err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg,
4526 			rw == S_READ || rw == S_EXEC, protp);
4527 	if (err)
4528 		goto out;
4529 
4530 	fs = ufsvfsp->vfs_fs;
4531 
4532 	if (ulp && (rw == S_CREATE || rw == S_WRITE) &&
4533 	    !(vp->v_flag & VISSWAP)) {
4534 		/*
4535 		 * Try to start a transaction, will return if blocking is
4536 		 * expected to occur and the address space is not the
4537 		 * kernel address space.
4538 		 */
4539 		trans_size = TOP_GETPAGE_SIZE(ip);
4540 		if (seg->s_as != &kas) {
4541 			TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE,
4542 				trans_size, err)
4543 			if (err == EWOULDBLOCK) {
4544 				/*
4545 				 * Use EDEADLK here because the VM code
4546 				 * can normally never see this error.
4547 				 */
4548 				err = EDEADLK;
4549 				ufs_lockfs_end(ulp);
4550 				goto out;
4551 			}
4552 		} else {
4553 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4554 		}
4555 	}
4556 
4557 	if (vp->v_flag & VNOMAP) {
4558 		err = ENOSYS;
4559 		goto unlock;
4560 	}
4561 
4562 	seqmode = ip->i_nextr == uoff && rw != S_CREATE;
4563 
4564 	rwtype = RW_READER;		/* start as a reader */
4565 	dolock = (rw_owner(&ip->i_contents) != curthread);
4566 	/*
4567 	 * If this thread owns the lock, i.e., this thread grabbed it
4568 	 * as writer somewhere above, then we don't need to grab the
4569 	 * lock as reader in this routine.
4570 	 */
4571 	do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread);
4572 
4573 retrylock:
4574 	if (dolock) {
4575 		/*
4576 		 * Grab the quota lock if we need to call
4577 		 * bmap_write() below (with i_contents as writer).
4578 		 */
4579 		if (do_qlock && rwtype == RW_WRITER)
4580 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4581 		rw_enter(&ip->i_contents, rwtype);
4582 	}
4583 
4584 	/*
4585 	 * We may be getting called as a side effect of a bmap using
4586 	 * fbread() when the blocks might be being allocated and the
4587 	 * size has not yet been up'ed.  In this case we want to be
4588 	 * able to return zero pages if we get back UFS_HOLE from
4589 	 * calling bmap for a non write case here.  We also might have
4590 	 * to read some frags from the disk into a page if we are
4591 	 * extending the number of frags for a given lbn in bmap().
4592 	 * Large Files: The read of i_size here is atomic because
4593 	 * i_contents is held here. If dolock is zero, the lock
4594 	 * is held in bmap routines.
4595 	 */
4596 	beyond_eof = uoff + len > ip->i_size + PAGEOFFSET;
4597 	if (beyond_eof && seg != segkmap) {
4598 		if (dolock) {
4599 			rw_exit(&ip->i_contents);
4600 			if (do_qlock && rwtype == RW_WRITER)
4601 				rw_exit(&ufsvfsp->vfs_dqrwlock);
4602 		}
4603 		err = EFAULT;
4604 		goto unlock;
4605 	}
4606 
4607 	/*
4608 	 * Must hold i_contents lock throughout the call to pvn_getpages
4609 	 * since locked pages are returned from each call to ufs_getapage.
4610 	 * Must *not* return locked pages and then try for contents lock
4611 	 * due to lock ordering requirements (inode > page)
4612 	 */
4613 
4614 	has_holes = bmap_has_holes(ip);
4615 
4616 	if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) {
4617 		int	blk_size;
4618 		u_offset_t offset;
4619 
4620 		/*
4621 		 * We must acquire the RW_WRITER lock in order to
4622 		 * call bmap_write().
4623 		 */
4624 		if (dolock && rwtype == RW_READER) {
4625 			rwtype = RW_WRITER;
4626 
4627 			/*
4628 			 * Grab the quota lock before
4629 			 * upgrading i_contents, but if we can't grab it
4630 			 * don't wait here due to lock order:
4631 			 * vfs_dqrwlock > i_contents.
4632 			 */
4633 			if (do_qlock && rw_tryenter(&ufsvfsp->vfs_dqrwlock,
4634 							RW_READER) == 0) {
4635 				rw_exit(&ip->i_contents);
4636 				goto retrylock;
4637 			}
4638 			if (!rw_tryupgrade(&ip->i_contents)) {
4639 				rw_exit(&ip->i_contents);
4640 				if (do_qlock)
4641 					rw_exit(&ufsvfsp->vfs_dqrwlock);
4642 				goto retrylock;
4643 			}
4644 		}
4645 
4646 		/*
4647 		 * May be allocating disk blocks for holes here as
4648 		 * a result of mmap faults. write(2) does the bmap_write
4649 		 * in rdip/wrip, not here. We are not dealing with frags
4650 		 * in this case.
4651 		 */
4652 		/*
4653 		 * Large Files: We cast fs_bmask field to offset_t
4654 		 * just as we do for MAXBMASK because uoff is a 64-bit
4655 		 * data type. fs_bmask will still be a 32-bit type
4656 		 * as we cannot change any ondisk data structures.
4657 		 */
4658 
4659 		offset = uoff & (offset_t)fs->fs_bmask;
4660 		while (offset < uoff + len) {
4661 			blk_size = (int)blksize(fs, ip, lblkno(fs, offset));
4662 			err = bmap_write(ip, offset, blk_size,
4663 			    BI_NORMAL, NULL, cr);
4664 			if (ip->i_flag & (ICHG|IUPD))
4665 				ip->i_seq++;
4666 			if (err)
4667 				goto update_inode;
4668 			offset += blk_size; /* XXX - make this contig */
4669 		}
4670 	}
4671 
4672 	/*
4673 	 * Can be a reader from now on.
4674 	 */
4675 	if (dolock && rwtype == RW_WRITER) {
4676 		rw_downgrade(&ip->i_contents);
4677 		/*
4678 		 * We can release vfs_dqrwlock early so do it, but make
4679 		 * sure we don't try to release it again at the bottom.
4680 		 */
4681 		if (do_qlock) {
4682 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4683 			do_qlock = 0;
4684 		}
4685 	}
4686 
4687 	/*
4688 	 * We remove PROT_WRITE in cases when the file has UFS holes
4689 	 * because we don't  want to call bmap_read() to check each
4690 	 * page if it is backed with a disk block.
4691 	 */
4692 	if (protp && has_holes && rw != S_WRITE && rw != S_CREATE)
4693 		*protp &= ~PROT_WRITE;
4694 
4695 	err = 0;
4696 
4697 	/*
4698 	 * The loop looks up pages in the range [off, off + len).
4699 	 * For each page, we first check if we should initiate an asynchronous
4700 	 * read ahead before we call page_lookup (we may sleep in page_lookup
4701 	 * for a previously initiated disk read).
4702 	 */
4703 	eoff = (uoff + len);
4704 	for (pgoff = uoff, pgaddr = addr, pl = plarr;
4705 	    pgoff < eoff; /* empty */) {
4706 		page_t	*pp;
4707 		u_offset_t	nextrio;
4708 		se_t	se;
4709 		int retval;
4710 
4711 		se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED);
4712 
4713 		/* Handle async getpage (faultahead) */
4714 		if (plarr == NULL) {
4715 			ip->i_nextrio = pgoff;
4716 			(void) ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4717 			pgoff += pgsize;
4718 			pgaddr += pgsize;
4719 			continue;
4720 		}
4721 		/*
4722 		 * Check if we should initiate read ahead of next cluster.
4723 		 * We call page_exists only when we need to confirm that
4724 		 * we have the current page before we initiate the read ahead.
4725 		 */
4726 		nextrio = ip->i_nextrio;
4727 		if (seqmode &&
4728 		    pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
4729 		    nextrio < ip->i_size && page_exists(vp, pgoff)) {
4730 			retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4731 			/*
4732 			 * We always read ahead the next cluster of data
4733 			 * starting from i_nextrio. If the page (vp,nextrio)
4734 			 * is actually in core at this point, the routine
4735 			 * ufs_getpage_ra() will stop pre-fetching data
4736 			 * until we read that page in a synchronized manner
4737 			 * through ufs_getpage_miss(). So, we should increase
4738 			 * i_nextrio if the page (vp, nextrio) exists.
4739 			 */
4740 			if ((retval == 0) && page_exists(vp, nextrio)) {
4741 				ip->i_nextrio = nextrio + pgsize;
4742 			}
4743 		}
4744 
4745 		if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
4746 			/*
4747 			 * We found the page in the page cache.
4748 			 */
4749 			*pl++ = pp;
4750 			pgoff += pgsize;
4751 			pgaddr += pgsize;
4752 			len -= pgsize;
4753 			plsz -= pgsize;
4754 		} else  {
4755 			/*
4756 			 * We have to create the page, or read it from disk.
4757 			 */
4758 			if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr,
4759 			    pl, plsz, rw, seqmode))
4760 				goto error;
4761 
4762 			while (*pl != NULL) {
4763 				pl++;
4764 				pgoff += pgsize;
4765 				pgaddr += pgsize;
4766 				len -= pgsize;
4767 				plsz -= pgsize;
4768 			}
4769 		}
4770 	}
4771 
4772 	/*
4773 	 * Return pages up to plsz if they are in the page cache.
4774 	 * We cannot return pages if there is a chance that they are
4775 	 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4776 	 */
4777 	if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
4778 
4779 		ASSERT((protp == NULL) ||
4780 			!(has_holes && (*protp & PROT_WRITE)));
4781 
4782 		eoff = pgoff + plsz;
4783 		while (pgoff < eoff) {
4784 			page_t		*pp;
4785 
4786 			if ((pp = page_lookup_nowait(vp, pgoff,
4787 			    SE_SHARED)) == NULL)
4788 				break;
4789 
4790 			*pl++ = pp;
4791 			pgoff += pgsize;
4792 			plsz -= pgsize;
4793 		}
4794 	}
4795 
4796 	if (plarr)
4797 		*pl = NULL;			/* Terminate page list */
4798 	ip->i_nextr = pgoff;
4799 
4800 error:
4801 	if (err && plarr) {
4802 		/*
4803 		 * Release any pages we have locked.
4804 		 */
4805 		while (pl > &plarr[0])
4806 			page_unlock(*--pl);
4807 
4808 		plarr[0] = NULL;
4809 	}
4810 
4811 update_inode:
4812 	/*
4813 	 * If the inode is not already marked for IACC (in rdip() for read)
4814 	 * and the inode is not marked for no access time update (in wrip()
4815 	 * for write) then update the inode access time and mod time now.
4816 	 */
4817 	if ((ip->i_flag & (IACC | INOACC)) == 0) {
4818 		if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) {
4819 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
4820 			    (fs->fs_ronly == 0) &&
4821 			    (!ufsvfsp->vfs_noatime)) {
4822 				mutex_enter(&ip->i_tlock);
4823 				ip->i_flag |= IACC;
4824 				ITIMES_NOLOCK(ip);
4825 				mutex_exit(&ip->i_tlock);
4826 			}
4827 		}
4828 	}
4829 
4830 	if (dolock) {
4831 		rw_exit(&ip->i_contents);
4832 		if (do_qlock && rwtype == RW_WRITER)
4833 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4834 	}
4835 
4836 unlock:
4837 	if (ulp) {
4838 		if ((rw == S_CREATE || rw == S_WRITE) &&
4839 		    !(vp->v_flag & VISSWAP)) {
4840 			TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4841 		}
4842 		ufs_lockfs_end(ulp);
4843 	}
4844 out:
4845 	TRACE_2(TR_FAC_UFS, TR_UFS_GETPAGE_END,
4846 		"ufs_getpage_end:vp %p error %d", vp, err);
4847 	return (err);
4848 }
4849 
4850 /*
4851  * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4852  * cache. The page is either read from the disk, or it's created.
4853  * A page is created (without disk read) if rw == S_CREATE, or if
4854  * the page is not backed with a real disk block (UFS hole).
4855  */
4856 /* ARGSUSED */
4857 static int
4858 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg,
4859 	caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq)
4860 {
4861 	struct inode	*ip = VTOI(vp);
4862 	page_t		*pp;
4863 	daddr_t		bn;
4864 	size_t		io_len;
4865 	int		crpage = 0;
4866 	int		err;
4867 	int		contig;
4868 	int		bsize = ip->i_fs->fs_bsize;
4869 
4870 	/*
4871 	 * Figure out whether the page can be created, or must be
4872 	 * must be read from the disk.
4873 	 */
4874 	if (rw == S_CREATE)
4875 		crpage = 1;
4876 	else {
4877 		contig = 0;
4878 		if (err = bmap_read(ip, off, &bn, &contig))
4879 			return (err);
4880 
4881 		crpage = (bn == UFS_HOLE);
4882 
4883 		/*
4884 		 * If its also a fallocated block that hasn't been written to
4885 		 * yet, we will treat it just like a UFS_HOLE and create
4886 		 * a zero page for it
4887 		 */
4888 		if (ISFALLOCBLK(ip, bn))
4889 			crpage = 1;
4890 	}
4891 
4892 	if (crpage) {
4893 		if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg,
4894 		    addr)) == NULL) {
4895 			return (ufs_fault(vp,
4896 				    "ufs_getpage_miss: page_create == NULL"));
4897 		}
4898 
4899 		if (rw != S_CREATE)
4900 			pagezero(pp, 0, PAGESIZE);
4901 
4902 		io_len = PAGESIZE;
4903 	} else {
4904 		u_offset_t	io_off;
4905 		uint_t	xlen;
4906 		struct buf	*bp;
4907 		ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
4908 
4909 		/*
4910 		 * If access is not in sequential order, we read from disk
4911 		 * in bsize units.
4912 		 *
4913 		 * We limit the size of the transfer to bsize if we are reading
4914 		 * from the beginning of the file. Note in this situation we
4915 		 * will hedge our bets and initiate an async read ahead of
4916 		 * the second block.
4917 		 */
4918 		if (!seq || off == 0)
4919 			contig = MIN(contig, bsize);
4920 
4921 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4922 		    &io_len, off, contig, 0);
4923 
4924 		/*
4925 		 * Some other thread has entered the page.
4926 		 * ufs_getpage will retry page_lookup.
4927 		 */
4928 		if (pp == NULL) {
4929 			pl[0] = NULL;
4930 			return (0);
4931 		}
4932 
4933 		/*
4934 		 * Zero part of the page which we are not
4935 		 * going to read from the disk.
4936 		 */
4937 		xlen = io_len & PAGEOFFSET;
4938 		if (xlen != 0)
4939 			pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4940 
4941 		bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ);
4942 		bp->b_edev = ip->i_dev;
4943 		bp->b_dev = cmpdev(ip->i_dev);
4944 		bp->b_blkno = bn;
4945 		bp->b_un.b_addr = (caddr_t)0;
4946 		bp->b_file = ip->i_vnode;
4947 		bp->b_offset = off;
4948 
4949 		if (ufsvfsp->vfs_log) {
4950 			lufs_read_strategy(ufsvfsp->vfs_log, bp);
4951 		} else if (ufsvfsp->vfs_snapshot) {
4952 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4953 		} else {
4954 			ufsvfsp->vfs_iotstamp = lbolt;
4955 			ub.ub_getpages.value.ul++;
4956 			(void) bdev_strategy(bp);
4957 			lwp_stat_update(LWP_STAT_INBLK, 1);
4958 		}
4959 
4960 		ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK);
4961 
4962 		/*
4963 		 * If the file access is sequential, initiate read ahead
4964 		 * of the next cluster.
4965 		 */
4966 		if (seq && ip->i_nextrio < ip->i_size)
4967 			(void) ufs_getpage_ra(vp, off, seg, addr);
4968 		err = biowait(bp);
4969 		pageio_done(bp);
4970 
4971 		if (err) {
4972 			pvn_read_done(pp, B_ERROR);
4973 			return (err);
4974 		}
4975 	}
4976 
4977 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4978 	return (0);
4979 }
4980 
4981 /*
4982  * Read ahead a cluster from the disk. Returns the length in bytes.
4983  */
4984 static int
4985 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr)
4986 {
4987 	struct inode	*ip = VTOI(vp);
4988 	page_t		*pp;
4989 	u_offset_t	io_off = ip->i_nextrio;
4990 	ufsvfs_t	*ufsvfsp;
4991 	caddr_t		addr2 = addr + (io_off - off);
4992 	struct buf	*bp;
4993 	daddr_t		bn;
4994 	size_t		io_len;
4995 	int		err;
4996 	int		contig;
4997 	int		xlen;
4998 	int		bsize = ip->i_fs->fs_bsize;
4999 
5000 	/*
5001 	 * If the directio advisory is in effect on this file,
5002 	 * then do not do buffered read ahead. Read ahead makes
5003 	 * it more difficult on threads using directio as they
5004 	 * will be forced to flush the pages from this vnode.
5005 	 */
5006 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5007 		return (0);
5008 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio)
5009 		return (0);
5010 
5011 	/*
5012 	 * Is this test needed?
5013 	 */
5014 	if (addr2 >= seg->s_base + seg->s_size)
5015 		return (0);
5016 
5017 	contig = 0;
5018 	err = bmap_read(ip, io_off, &bn, &contig);
5019 	/*
5020 	 * If its a UFS_HOLE or a fallocated block, do not perform
5021 	 * any read ahead's since there probably is nothing to read ahead
5022 	 */
5023 	if (err || bn == UFS_HOLE || ISFALLOCBLK(ip, bn))
5024 		return (0);
5025 
5026 	/*
5027 	 * Limit the transfer size to bsize if this is the 2nd block.
5028 	 */
5029 	if (io_off == (u_offset_t)bsize)
5030 		contig = MIN(contig, bsize);
5031 
5032 	if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off,
5033 	    &io_len, io_off, contig, 1)) == NULL)
5034 		return (0);
5035 
5036 	/*
5037 	 * Zero part of page which we are not going to read from disk
5038 	 */
5039 	if ((xlen = (io_len & PAGEOFFSET)) > 0)
5040 		pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
5041 
5042 	ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK;
5043 
5044 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC);
5045 	bp->b_edev = ip->i_dev;
5046 	bp->b_dev = cmpdev(ip->i_dev);
5047 	bp->b_blkno = bn;
5048 	bp->b_un.b_addr = (caddr_t)0;
5049 	bp->b_file = ip->i_vnode;
5050 	bp->b_offset = off;
5051 
5052 	if (ufsvfsp->vfs_log) {
5053 		lufs_read_strategy(ufsvfsp->vfs_log, bp);
5054 	} else if (ufsvfsp->vfs_snapshot) {
5055 		fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5056 	} else {
5057 		ufsvfsp->vfs_iotstamp = lbolt;
5058 		ub.ub_getras.value.ul++;
5059 		(void) bdev_strategy(bp);
5060 		lwp_stat_update(LWP_STAT_INBLK, 1);
5061 	}
5062 
5063 	return (io_len);
5064 }
5065 
5066 int	ufs_delay = 1;
5067 /*
5068  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5069  *
5070  * LMXXX - the inode really ought to contain a pointer to one of these
5071  * async args.  Stuff gunk in there and just hand the whole mess off.
5072  * This would replace i_delaylen, i_delayoff.
5073  */
5074 /*ARGSUSED*/
5075 static int
5076 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
5077 	struct cred *cr)
5078 {
5079 	struct inode *ip = VTOI(vp);
5080 	int err = 0;
5081 
5082 	if (vp->v_count == 0) {
5083 		return (ufs_fault(vp, "ufs_putpage: bad v_count == 0"));
5084 	}
5085 
5086 	TRACE_1(TR_FAC_UFS, TR_UFS_PUTPAGE_START,
5087 		"ufs_putpage_start:vp %p", vp);
5088 
5089 	/*
5090 	 * XXX - Why should this check be made here?
5091 	 */
5092 	if (vp->v_flag & VNOMAP) {
5093 		err = ENOSYS;
5094 		goto errout;
5095 	}
5096 
5097 	if (ip->i_ufsvfs == NULL) {
5098 		err = EIO;
5099 		goto errout;
5100 	}
5101 
5102 	if (flags & B_ASYNC) {
5103 		if (ufs_delay && len &&
5104 		    (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
5105 			mutex_enter(&ip->i_tlock);
5106 			/*
5107 			 * If nobody stalled, start a new cluster.
5108 			 */
5109 			if (ip->i_delaylen == 0) {
5110 				ip->i_delayoff = off;
5111 				ip->i_delaylen = len;
5112 				mutex_exit(&ip->i_tlock);
5113 				goto errout;
5114 			}
5115 			/*
5116 			 * If we have a full cluster or they are not contig,
5117 			 * then push last cluster and start over.
5118 			 */
5119 			if (ip->i_delaylen >= CLUSTSZ(ip) ||
5120 			    ip->i_delayoff + ip->i_delaylen != off) {
5121 				u_offset_t doff;
5122 				size_t dlen;
5123 
5124 				doff = ip->i_delayoff;
5125 				dlen = ip->i_delaylen;
5126 				ip->i_delayoff = off;
5127 				ip->i_delaylen = len;
5128 				mutex_exit(&ip->i_tlock);
5129 				err = ufs_putpages(vp, doff, dlen,
5130 				    flags, cr);
5131 				/* LMXXX - flags are new val, not old */
5132 				goto errout;
5133 			}
5134 			/*
5135 			 * There is something there, it's not full, and
5136 			 * it is contig.
5137 			 */
5138 			ip->i_delaylen += len;
5139 			mutex_exit(&ip->i_tlock);
5140 			goto errout;
5141 		}
5142 		/*
5143 		 * Must have weird flags or we are not clustering.
5144 		 */
5145 	}
5146 
5147 	err = ufs_putpages(vp, off, len, flags, cr);
5148 
5149 errout:
5150 	TRACE_2(TR_FAC_UFS, TR_UFS_PUTPAGE_END,
5151 		"ufs_putpage_end:vp %p error %d", vp, err);
5152 	return (err);
5153 }
5154 
5155 /*
5156  * If len == 0, do from off to EOF.
5157  *
5158  * The normal cases should be len == 0 & off == 0 (entire vp list),
5159  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5160  * (from pageout).
5161  */
5162 /*ARGSUSED*/
5163 static int
5164 ufs_putpages(
5165 	struct vnode *vp,
5166 	offset_t off,
5167 	size_t len,
5168 	int flags,
5169 	struct cred *cr)
5170 {
5171 	u_offset_t io_off;
5172 	u_offset_t eoff;
5173 	struct inode *ip = VTOI(vp);
5174 	page_t *pp;
5175 	size_t io_len;
5176 	int err = 0;
5177 	int dolock;
5178 
5179 	if (vp->v_count == 0)
5180 		return (ufs_fault(vp, "ufs_putpages: v_count == 0"));
5181 	/*
5182 	 * Acquire the readers/write inode lock before locking
5183 	 * any pages in this inode.
5184 	 * The inode lock is held during i/o.
5185 	 */
5186 	if (len == 0) {
5187 		mutex_enter(&ip->i_tlock);
5188 		ip->i_delayoff = ip->i_delaylen = 0;
5189 		mutex_exit(&ip->i_tlock);
5190 	}
5191 	dolock = (rw_owner(&ip->i_contents) != curthread);
5192 	if (dolock) {
5193 		/*
5194 		 * Must synchronize this thread and any possible thread
5195 		 * operating in the window of vulnerability in wrip().
5196 		 * It is dangerous to allow both a thread doing a putpage
5197 		 * and a thread writing, so serialize them.  The exception
5198 		 * is when the thread in wrip() does something which causes
5199 		 * a putpage operation.  Then, the thread must be allowed
5200 		 * to continue.  It may encounter a bmap_read problem in
5201 		 * ufs_putapage, but that is handled in ufs_putapage.
5202 		 * Allow async writers to proceed, we don't want to block
5203 		 * the pageout daemon.
5204 		 */
5205 		if (ip->i_writer == curthread)
5206 			rw_enter(&ip->i_contents, RW_READER);
5207 		else {
5208 			for (;;) {
5209 				rw_enter(&ip->i_contents, RW_READER);
5210 				mutex_enter(&ip->i_tlock);
5211 				/*
5212 				 * If there is no thread in the critical
5213 				 * section of wrip(), then proceed.
5214 				 * Otherwise, wait until there isn't one.
5215 				 */
5216 				if (ip->i_writer == NULL) {
5217 					mutex_exit(&ip->i_tlock);
5218 					break;
5219 				}
5220 				rw_exit(&ip->i_contents);
5221 				/*
5222 				 * Bounce async writers when we have a writer
5223 				 * working on this file so we don't deadlock
5224 				 * the pageout daemon.
5225 				 */
5226 				if (flags & B_ASYNC) {
5227 					mutex_exit(&ip->i_tlock);
5228 					return (0);
5229 				}
5230 				cv_wait(&ip->i_wrcv, &ip->i_tlock);
5231 				mutex_exit(&ip->i_tlock);
5232 			}
5233 		}
5234 	}
5235 
5236 	if (!vn_has_cached_data(vp)) {
5237 		if (dolock)
5238 			rw_exit(&ip->i_contents);
5239 		return (0);
5240 	}
5241 
5242 	if (len == 0) {
5243 		/*
5244 		 * Search the entire vp list for pages >= off.
5245 		 */
5246 		err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage,
5247 					flags, cr);
5248 	} else {
5249 		/*
5250 		 * Loop over all offsets in the range looking for
5251 		 * pages to deal with.
5252 		 */
5253 		if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0)
5254 			eoff = MIN(off + len, eoff);
5255 		else
5256 			eoff = off + len;
5257 
5258 		for (io_off = off; io_off < eoff; io_off += io_len) {
5259 			/*
5260 			 * If we are not invalidating, synchronously
5261 			 * freeing or writing pages, use the routine
5262 			 * page_lookup_nowait() to prevent reclaiming
5263 			 * them from the free list.
5264 			 */
5265 			if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
5266 				pp = page_lookup(vp, io_off,
5267 					(flags & (B_INVAL | B_FREE)) ?
5268 					    SE_EXCL : SE_SHARED);
5269 			} else {
5270 				pp = page_lookup_nowait(vp, io_off,
5271 					(flags & B_FREE) ? SE_EXCL : SE_SHARED);
5272 			}
5273 
5274 			if (pp == NULL || pvn_getdirty(pp, flags) == 0)
5275 				io_len = PAGESIZE;
5276 			else {
5277 				u_offset_t *io_offp = &io_off;
5278 
5279 				err = ufs_putapage(vp, pp, io_offp, &io_len,
5280 				    flags, cr);
5281 				if (err != 0)
5282 					break;
5283 				/*
5284 				 * "io_off" and "io_len" are returned as
5285 				 * the range of pages we actually wrote.
5286 				 * This allows us to skip ahead more quickly
5287 				 * since several pages may've been dealt
5288 				 * with by this iteration of the loop.
5289 				 */
5290 			}
5291 		}
5292 	}
5293 	if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
5294 		/*
5295 		 * We have just sync'ed back all the pages on
5296 		 * the inode, turn off the IMODTIME flag.
5297 		 */
5298 		mutex_enter(&ip->i_tlock);
5299 		ip->i_flag &= ~IMODTIME;
5300 		mutex_exit(&ip->i_tlock);
5301 	}
5302 	if (dolock)
5303 		rw_exit(&ip->i_contents);
5304 	return (err);
5305 }
5306 
5307 static void
5308 ufs_iodone(buf_t *bp)
5309 {
5310 	struct inode *ip;
5311 
5312 	ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
5313 
5314 	bp->b_iodone = NULL;
5315 
5316 	ip = VTOI(bp->b_pages->p_vnode);
5317 
5318 	mutex_enter(&ip->i_tlock);
5319 	if (ip->i_writes >= ufs_LW) {
5320 		if ((ip->i_writes -= bp->b_bcount) <= ufs_LW)
5321 			if (ufs_WRITES)
5322 				cv_broadcast(&ip->i_wrcv); /* wake all up */
5323 	} else {
5324 		ip->i_writes -= bp->b_bcount;
5325 	}
5326 
5327 	mutex_exit(&ip->i_tlock);
5328 	iodone(bp);
5329 }
5330 
5331 /*
5332  * Write out a single page, possibly klustering adjacent
5333  * dirty pages.  The inode lock must be held.
5334  *
5335  * LMXXX - bsize < pagesize not done.
5336  */
5337 /*ARGSUSED*/
5338 int
5339 ufs_putapage(
5340 	struct vnode *vp,
5341 	page_t *pp,
5342 	u_offset_t *offp,
5343 	size_t *lenp,		/* return values */
5344 	int flags,
5345 	struct cred *cr)
5346 {
5347 	u_offset_t io_off;
5348 	u_offset_t off;
5349 	struct inode *ip = VTOI(vp);
5350 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
5351 	struct fs *fs;
5352 	struct buf *bp;
5353 	size_t io_len;
5354 	daddr_t bn;
5355 	int err;
5356 	int contig;
5357 
5358 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
5359 
5360 	TRACE_1(TR_FAC_UFS, TR_UFS_PUTAPAGE_START,
5361 		"ufs_putapage_start:vp %p", vp);
5362 
5363 	if (ufsvfsp == NULL) {
5364 		err = EIO;
5365 		goto out_trace;
5366 	}
5367 
5368 	fs = ip->i_fs;
5369 	ASSERT(fs->fs_ronly == 0);
5370 
5371 	/*
5372 	 * If the modified time on the inode has not already been
5373 	 * set elsewhere (e.g. for write/setattr) we set the time now.
5374 	 * This gives us approximate modified times for mmap'ed files
5375 	 * which are modified via stores in the user address space.
5376 	 */
5377 	if ((ip->i_flag & IMODTIME) == 0) {
5378 		mutex_enter(&ip->i_tlock);
5379 		ip->i_flag |= IUPD;
5380 		ip->i_seq++;
5381 		ITIMES_NOLOCK(ip);
5382 		mutex_exit(&ip->i_tlock);
5383 	}
5384 
5385 	/*
5386 	 * Align the request to a block boundry (for old file systems),
5387 	 * and go ask bmap() how contiguous things are for this file.
5388 	 */
5389 	off = pp->p_offset & (offset_t)fs->fs_bmask;	/* block align it */
5390 	contig = 0;
5391 	err = bmap_read(ip, off, &bn, &contig);
5392 	if (err)
5393 		goto out;
5394 	if (bn == UFS_HOLE) {			/* putpage never allocates */
5395 		/*
5396 		 * logging device is in error mode; simply return EIO
5397 		 */
5398 		if (TRANS_ISERROR(ufsvfsp)) {
5399 			err = EIO;
5400 			goto out;
5401 		}
5402 		/*
5403 		 * Oops, the thread in the window in wrip() did some
5404 		 * sort of operation which caused a putpage in the bad
5405 		 * range.  In this case, just return an error which will
5406 		 * cause the software modified bit on the page to set
5407 		 * and the page will get written out again later.
5408 		 */
5409 		if (ip->i_writer == curthread) {
5410 			err = EIO;
5411 			goto out;
5412 		}
5413 		/*
5414 		 * If the pager is trying to push a page in the bad range
5415 		 * just tell him to try again later when things are better.
5416 		 */
5417 		if (flags & B_ASYNC) {
5418 			err = EAGAIN;
5419 			goto out;
5420 		}
5421 		err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE");
5422 		goto out;
5423 	}
5424 
5425 	/*
5426 	 * If it is an fallocate'd block, reverse the negativity since
5427 	 * we are now writing to it
5428 	 */
5429 	if (ISFALLOCBLK(ip, bn)) {
5430 		err = bmap_set_bn(vp, off, dbtofsb(fs, -bn));
5431 		if (err)
5432 			goto out;
5433 
5434 		bn = -bn;
5435 	}
5436 
5437 	/*
5438 	 * Take the length (of contiguous bytes) passed back from bmap()
5439 	 * and _try_ and get a set of pages covering that extent.
5440 	 */
5441 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags);
5442 
5443 	/*
5444 	 * May have run out of memory and not clustered backwards.
5445 	 * off		p_offset
5446 	 * [  pp - 1  ][   pp   ]
5447 	 * [	block		]
5448 	 * We told bmap off, so we have to adjust the bn accordingly.
5449 	 */
5450 	if (io_off > off) {
5451 		bn += btod(io_off - off);
5452 		contig -= (io_off - off);
5453 	}
5454 
5455 	/*
5456 	 * bmap was carefull to tell us the right size so use that.
5457 	 * There might be unallocated frags at the end.
5458 	 * LMXXX - bzero the end of the page?  We must be writing after EOF.
5459 	 */
5460 	if (io_len > contig) {
5461 		ASSERT(io_len - contig < fs->fs_bsize);
5462 		io_len -= (io_len - contig);
5463 	}
5464 
5465 	/*
5466 	 * Handle the case where we are writing the last page after EOF.
5467 	 *
5468 	 * XXX - just a patch for i-mt3.
5469 	 */
5470 	if (io_len == 0) {
5471 		ASSERT(pp->p_offset >= (u_offset_t)(roundup(ip->i_size,
5472 							    PAGESIZE)));
5473 		io_len = PAGESIZE;
5474 	}
5475 
5476 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags);
5477 
5478 	ULOCKFS_SET_MOD(ITOUL(ip));
5479 
5480 	bp->b_edev = ip->i_dev;
5481 	bp->b_dev = cmpdev(ip->i_dev);
5482 	bp->b_blkno = bn;
5483 	bp->b_un.b_addr = (caddr_t)0;
5484 	bp->b_file = ip->i_vnode;
5485 
5486 	if (TRANS_ISTRANS(ufsvfsp)) {
5487 		if ((ip->i_mode & IFMT) == IFSHAD) {
5488 			TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD);
5489 		} else if (ufsvfsp->vfs_qinod == ip) {
5490 			TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR,
5491 			    0, 0);
5492 		}
5493 	}
5494 
5495 	/* write throttle */
5496 
5497 	ASSERT(bp->b_iodone == NULL);
5498 	bp->b_iodone = (int (*)())ufs_iodone;
5499 	mutex_enter(&ip->i_tlock);
5500 	ip->i_writes += bp->b_bcount;
5501 	mutex_exit(&ip->i_tlock);
5502 
5503 	if (bp->b_flags & B_ASYNC) {
5504 		if (ufsvfsp->vfs_log) {
5505 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5506 		} else if (ufsvfsp->vfs_snapshot) {
5507 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5508 		} else {
5509 			ufsvfsp->vfs_iotstamp = lbolt;
5510 			ub.ub_putasyncs.value.ul++;
5511 			(void) bdev_strategy(bp);
5512 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5513 		}
5514 	} else {
5515 		if (ufsvfsp->vfs_log) {
5516 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5517 		} else if (ufsvfsp->vfs_snapshot) {
5518 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5519 		} else {
5520 			ufsvfsp->vfs_iotstamp = lbolt;
5521 			ub.ub_putsyncs.value.ul++;
5522 			(void) bdev_strategy(bp);
5523 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5524 		}
5525 		err = biowait(bp);
5526 		pageio_done(bp);
5527 		pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
5528 	}
5529 
5530 	pp = NULL;
5531 
5532 out:
5533 	if (err != 0 && pp != NULL)
5534 		pvn_write_done(pp, B_ERROR | B_WRITE | flags);
5535 
5536 	if (offp)
5537 		*offp = io_off;
5538 	if (lenp)
5539 		*lenp = io_len;
5540 out_trace:
5541 	TRACE_2(TR_FAC_UFS, TR_UFS_PUTAPAGE_END,
5542 		"ufs_putapage_end:vp %p error %d", vp, err);
5543 	return (err);
5544 }
5545 
5546 /* ARGSUSED */
5547 static int
5548 ufs_map(struct vnode *vp,
5549 	offset_t off,
5550 	struct as *as,
5551 	caddr_t *addrp,
5552 	size_t len,
5553 	uchar_t prot,
5554 	uchar_t maxprot,
5555 	uint_t flags,
5556 	struct cred *cr)
5557 {
5558 	struct segvn_crargs vn_a;
5559 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5560 	struct ulockfs *ulp;
5561 	int error;
5562 
5563 	TRACE_1(TR_FAC_UFS, TR_UFS_MAP_START,
5564 		"ufs_map_start:vp %p", vp);
5565 
5566 retry_map:
5567 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK);
5568 	if (error)
5569 		goto out;
5570 
5571 	if (vp->v_flag & VNOMAP) {
5572 		error = ENOSYS;
5573 		goto unlock;
5574 	}
5575 
5576 	if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) {
5577 		error = ENXIO;
5578 		goto unlock;
5579 	}
5580 
5581 	if (vp->v_type != VREG) {
5582 		error = ENODEV;
5583 		goto unlock;
5584 	}
5585 
5586 	/*
5587 	 * If file is being locked, disallow mapping.
5588 	 */
5589 	if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) {
5590 		error = EAGAIN;
5591 		goto unlock;
5592 	}
5593 
5594 	as_rangelock(as);
5595 	if ((flags & MAP_FIXED) == 0) {
5596 		map_addr(addrp, len, off, 1, flags);
5597 		if (*addrp == NULL) {
5598 			as_rangeunlock(as);
5599 			error = ENOMEM;
5600 			goto unlock;
5601 		}
5602 	} else {
5603 		/*
5604 		 * User specified address - blow away any previous mappings
5605 		 */
5606 		(void) as_unmap(as, *addrp, len);
5607 	}
5608 
5609 	vn_a.vp = vp;
5610 	vn_a.offset = (u_offset_t)off;
5611 	vn_a.type = flags & MAP_TYPE;
5612 	vn_a.prot = prot;
5613 	vn_a.maxprot = maxprot;
5614 	vn_a.cred = cr;
5615 	vn_a.amp = NULL;
5616 	vn_a.flags = flags & ~MAP_TYPE;
5617 	vn_a.szc = 0;
5618 	vn_a.lgrp_mem_policy_flags = 0;
5619 
5620 retry_lock:
5621 	if (!AS_LOCK_TRYENTER(ias, &as->a_lock, RW_WRITER)) {
5622 		/*
5623 		 * We didn't get the lock. Check if the SLOCK is set in the
5624 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
5625 		 * and wait for SLOCK to be cleared.
5626 		 */
5627 
5628 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
5629 			as_rangeunlock(as);
5630 			ufs_lockfs_end(ulp);
5631 			goto retry_map;
5632 		} else {
5633 			/*
5634 			 * SLOCK isn't set so this is a genuine synchronization
5635 			 * case. Let's try again after giving them a breather.
5636 			 */
5637 			delay(RETRY_LOCK_DELAY);
5638 			goto  retry_lock;
5639 		}
5640 	}
5641 	error = as_map_locked(as, *addrp, len, segvn_create, &vn_a);
5642 	as_rangeunlock(as);
5643 
5644 unlock:
5645 	if (ulp) {
5646 		ufs_lockfs_end(ulp);
5647 	}
5648 out:
5649 	TRACE_2(TR_FAC_UFS, TR_UFS_MAP_END,
5650 		"ufs_map_end:vp %p error %d", vp, error);
5651 	return (error);
5652 }
5653 
5654 /* ARGSUSED */
5655 static int
5656 ufs_addmap(struct vnode *vp,
5657 	offset_t off,
5658 	struct as *as,
5659 	caddr_t addr,
5660 	size_t	len,
5661 	uchar_t  prot,
5662 	uchar_t  maxprot,
5663 	uint_t    flags,
5664 	struct cred *cr)
5665 {
5666 	struct inode *ip = VTOI(vp);
5667 
5668 	if (vp->v_flag & VNOMAP) {
5669 		return (ENOSYS);
5670 	}
5671 
5672 	mutex_enter(&ip->i_tlock);
5673 	ip->i_mapcnt += btopr(len);
5674 	mutex_exit(&ip->i_tlock);
5675 	return (0);
5676 }
5677 
5678 /*ARGSUSED*/
5679 static int
5680 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5681 	size_t len, uint_t prot,  uint_t maxprot,  uint_t flags,
5682 	struct cred *cr)
5683 {
5684 	struct inode *ip = VTOI(vp);
5685 
5686 	if (vp->v_flag & VNOMAP) {
5687 		return (ENOSYS);
5688 	}
5689 
5690 	mutex_enter(&ip->i_tlock);
5691 	ip->i_mapcnt -= btopr(len); 	/* Count released mappings */
5692 	ASSERT(ip->i_mapcnt >= 0);
5693 	mutex_exit(&ip->i_tlock);
5694 	return (0);
5695 }
5696 /*
5697  * Return the answer requested to poll() for non-device files
5698  */
5699 struct pollhead ufs_pollhd;
5700 
5701 /* ARGSUSED */
5702 int
5703 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp)
5704 {
5705 	struct ufsvfs	*ufsvfsp;
5706 
5707 	*revp = 0;
5708 	ufsvfsp = VTOI(vp)->i_ufsvfs;
5709 
5710 	if (!ufsvfsp) {
5711 		*revp = POLLHUP;
5712 		goto out;
5713 	}
5714 
5715 	if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) ||
5716 	    ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) {
5717 		*revp |= POLLERR;
5718 
5719 	} else {
5720 		if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly &&
5721 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5722 			*revp |= POLLOUT;
5723 
5724 		if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly &&
5725 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5726 			*revp |= POLLWRBAND;
5727 
5728 		if (ev & POLLIN)
5729 			*revp |= POLLIN;
5730 
5731 		if (ev & POLLRDNORM)
5732 			*revp |= POLLRDNORM;
5733 
5734 		if (ev & POLLRDBAND)
5735 			*revp |= POLLRDBAND;
5736 	}
5737 
5738 	if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP)))
5739 		*revp |= POLLPRI;
5740 out:
5741 	*phpp = !any && !*revp ? &ufs_pollhd : (struct pollhead *)NULL;
5742 
5743 	return (0);
5744 }
5745 
5746 /* ARGSUSED */
5747 static int
5748 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr)
5749 {
5750 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
5751 	struct ulockfs	*ulp = NULL;
5752 	struct inode 	*sip = NULL;
5753 	int		error;
5754 	struct inode 	*ip = VTOI(vp);
5755 	int		issync;
5756 
5757 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK);
5758 	if (error)
5759 		return (error);
5760 
5761 	switch (cmd) {
5762 		/*
5763 		 * Have to handle _PC_NAME_MAX here, because the normal way
5764 		 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5765 		 * results in a lock ordering reversal between
5766 		 * ufs_lockfs_{begin,end}() and
5767 		 * ufs_thread_{suspend,continue}().
5768 		 *
5769 		 * Keep in sync with ufs_statvfs().
5770 		 */
5771 	case _PC_NAME_MAX:
5772 		*valp = MAXNAMLEN;
5773 		break;
5774 
5775 	case _PC_FILESIZEBITS:
5776 		if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
5777 			*valp = UFS_FILESIZE_BITS;
5778 		else
5779 			*valp = 32;
5780 		break;
5781 
5782 	case _PC_XATTR_EXISTS:
5783 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
5784 
5785 			error = ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR,
5786 							cr);
5787 			if (error ==  0 && sip != NULL) {
5788 				/* Start transaction */
5789 				if (ulp) {
5790 					TRANS_BEGIN_CSYNC(ufsvfsp, issync,
5791 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5792 				}
5793 				/*
5794 				 * Is directory empty
5795 				 */
5796 				rw_enter(&sip->i_rwlock, RW_WRITER);
5797 				rw_enter(&sip->i_contents, RW_WRITER);
5798 				if (ufs_xattrdirempty(sip,
5799 						sip->i_number, CRED())) {
5800 					rw_enter(&ip->i_contents, RW_WRITER);
5801 					ufs_unhook_shadow(ip, sip);
5802 					rw_exit(&ip->i_contents);
5803 
5804 					*valp = 0;
5805 
5806 				} else
5807 					*valp = 1;
5808 				rw_exit(&sip->i_contents);
5809 				rw_exit(&sip->i_rwlock);
5810 				if (ulp) {
5811 					TRANS_END_CSYNC(ufsvfsp, error, issync,
5812 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5813 				}
5814 				VN_RELE(ITOV(sip));
5815 			} else if (error == ENOENT) {
5816 				*valp = 0;
5817 				error = 0;
5818 			}
5819 		} else {
5820 			error = fs_pathconf(vp, cmd, valp, cr);
5821 		}
5822 		break;
5823 
5824 	case _PC_ACL_ENABLED:
5825 		*valp = _ACL_ACLENT_ENABLED;
5826 		break;
5827 
5828 	case _PC_MIN_HOLE_SIZE:
5829 		*valp = (ulong_t)ip->i_fs->fs_bsize;
5830 		break;
5831 
5832 	default:
5833 		error = fs_pathconf(vp, cmd, valp, cr);
5834 	}
5835 
5836 	if (ulp != NULL) {
5837 		ufs_lockfs_end(ulp);
5838 	}
5839 	return (error);
5840 }
5841 
5842 int ufs_pageio_writes, ufs_pageio_reads;
5843 
5844 /*ARGSUSED*/
5845 static int
5846 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len,
5847 	int flags, struct cred *cr)
5848 {
5849 	struct inode *ip = VTOI(vp);
5850 	struct ufsvfs *ufsvfsp;
5851 	page_t *npp = NULL, *opp = NULL, *cpp = pp;
5852 	struct buf *bp;
5853 	daddr_t bn;
5854 	size_t done_len = 0, cur_len = 0;
5855 	int err = 0;
5856 	int contig = 0;
5857 	int dolock;
5858 	int vmpss = 0;
5859 	struct ulockfs *ulp;
5860 
5861 	if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp &&
5862 	    vp->v_mpssdata != NULL) {
5863 		vmpss = 1;
5864 	}
5865 
5866 	dolock = (rw_owner(&ip->i_contents) != curthread);
5867 	/*
5868 	 * We need a better check.  Ideally, we would use another
5869 	 * vnodeops so that hlocked and forcibly unmounted file
5870 	 * systems would return EIO where appropriate and w/o the
5871 	 * need for these checks.
5872 	 */
5873 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5874 		return (EIO);
5875 
5876 	/*
5877 	 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5878 	 * ul_lock must be taken before locking pages so we can't use it here
5879 	 * if pp is non NULL because segvn already locked pages
5880 	 * SE_EXCL. Instead we rely on the fact that a forced umount or
5881 	 * applying a filesystem lock via ufs_fiolfs() will block in the
5882 	 * implicit call to ufs_flush() until we unlock the pages after the
5883 	 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5884 	 * above 0 until they are done. We have to be careful not to increment
5885 	 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5886 	 *
5887 	 * If pp is NULL use ul_lock to make sure we don't increment
5888 	 * ul_vnops_cnt after forceful unmount hlocks the file system.
5889 	 */
5890 	if (vmpss || pp == NULL) {
5891 		ulp = &ufsvfsp->vfs_ulockfs;
5892 		if (pp == NULL)
5893 			mutex_enter(&ulp->ul_lock);
5894 		if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) {
5895 			if (pp == NULL) {
5896 				mutex_exit(&ulp->ul_lock);
5897 			}
5898 			return (vmpss ? EIO : EINVAL);
5899 		}
5900 		atomic_add_long(&ulp->ul_vnops_cnt, 1);
5901 		if (pp == NULL)
5902 			mutex_exit(&ulp->ul_lock);
5903 		if (ufs_quiesce_pend) {
5904 			if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5905 				cv_broadcast(&ulp->ul_cv);
5906 			return (vmpss ? EIO : EINVAL);
5907 		}
5908 	}
5909 
5910 	if (dolock) {
5911 		/*
5912 		 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
5913 		 * handle a fault against a segment that maps vnode pages with
5914 		 * large mappings.  Segvn creates pages and holds them locked
5915 		 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
5916 		 * use rw_tryenter() to avoid a potential deadlock since in
5917 		 * lock order i_contents needs to be taken first.
5918 		 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
5919 		 */
5920 		if (!vmpss) {
5921 			rw_enter(&ip->i_contents, RW_READER);
5922 		} else if (!rw_tryenter(&ip->i_contents, RW_READER)) {
5923 			if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5924 				cv_broadcast(&ulp->ul_cv);
5925 			return (EDEADLK);
5926 		}
5927 	}
5928 
5929 	/*
5930 	 * Return an error to segvn because the pagefault request is beyond
5931 	 * PAGESIZE rounded EOF.
5932 	 */
5933 	if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) {
5934 		if (dolock)
5935 			rw_exit(&ip->i_contents);
5936 		if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5937 			cv_broadcast(&ulp->ul_cv);
5938 		return (EFAULT);
5939 	}
5940 
5941 	if (pp == NULL) {
5942 		if (bmap_has_holes(ip)) {
5943 			err = ENOSYS;
5944 		} else {
5945 			err = EINVAL;
5946 		}
5947 		if (dolock)
5948 			rw_exit(&ip->i_contents);
5949 		if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5950 			cv_broadcast(&ulp->ul_cv);
5951 		return (err);
5952 	}
5953 
5954 	/*
5955 	 * Break the io request into chunks, one for each contiguous
5956 	 * stretch of disk blocks in the target file.
5957 	 */
5958 	while (done_len < io_len) {
5959 		ASSERT(cpp);
5960 		contig = 0;
5961 		if (err = bmap_read(ip, (u_offset_t)(io_off + done_len),
5962 				    &bn, &contig))
5963 			break;
5964 
5965 		if (bn == UFS_HOLE) {	/* No holey swapfiles */
5966 			if (vmpss) {
5967 				err = EFAULT;
5968 				break;
5969 			}
5970 			err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE");
5971 			break;
5972 		}
5973 
5974 		cur_len = MIN(io_len - done_len, contig);
5975 		/*
5976 		 * Zero out a page beyond EOF, when the last block of
5977 		 * a file is a UFS fragment so that ufs_pageio() can be used
5978 		 * instead of ufs_getpage() to handle faults against
5979 		 * segvn segments that use large pages.
5980 		 */
5981 		page_list_break(&cpp, &npp, btopr(cur_len));
5982 		if ((flags & B_READ) && (cur_len & PAGEOFFSET)) {
5983 			size_t xlen = cur_len & PAGEOFFSET;
5984 			pagezero(cpp->p_prev, xlen, PAGESIZE - xlen);
5985 		}
5986 
5987 		bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
5988 		ASSERT(bp != NULL);
5989 
5990 		bp->b_edev = ip->i_dev;
5991 		bp->b_dev = cmpdev(ip->i_dev);
5992 		bp->b_blkno = bn;
5993 		bp->b_un.b_addr = (caddr_t)0;
5994 		bp->b_file = ip->i_vnode;
5995 
5996 		ufsvfsp->vfs_iotstamp = lbolt;
5997 		ub.ub_pageios.value.ul++;
5998 		if (ufsvfsp->vfs_snapshot)
5999 			fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp);
6000 		else
6001 			(void) bdev_strategy(bp);
6002 
6003 		if (flags & B_READ)
6004 			ufs_pageio_reads++;
6005 		else
6006 			ufs_pageio_writes++;
6007 		if (flags & B_READ)
6008 			lwp_stat_update(LWP_STAT_INBLK, 1);
6009 		else
6010 			lwp_stat_update(LWP_STAT_OUBLK, 1);
6011 		/*
6012 		 * If the request is not B_ASYNC, wait for i/o to complete
6013 		 * and re-assemble the page list to return to the caller.
6014 		 * If it is B_ASYNC we leave the page list in pieces and
6015 		 * cleanup() will dispose of them.
6016 		 */
6017 		if ((flags & B_ASYNC) == 0) {
6018 			err = biowait(bp);
6019 			pageio_done(bp);
6020 			if (err)
6021 				break;
6022 			page_list_concat(&opp, &cpp);
6023 		}
6024 		cpp = npp;
6025 		npp = NULL;
6026 		if (flags & B_READ)
6027 			cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t);
6028 		done_len += cur_len;
6029 	}
6030 	ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len));
6031 	if (err) {
6032 		if (flags & B_ASYNC) {
6033 			/* Cleanup unprocessed parts of list */
6034 			page_list_concat(&cpp, &npp);
6035 			if (flags & B_READ)
6036 				pvn_read_done(cpp, B_ERROR);
6037 			else
6038 				pvn_write_done(cpp, B_ERROR);
6039 		} else {
6040 			/* Re-assemble list and let caller clean up */
6041 			page_list_concat(&opp, &cpp);
6042 			page_list_concat(&opp, &npp);
6043 		}
6044 	}
6045 
6046 	if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) &&
6047 	    ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) {
6048 		mutex_enter(&ip->i_tlock);
6049 		ip->i_flag |= IACC;
6050 		ITIMES_NOLOCK(ip);
6051 		mutex_exit(&ip->i_tlock);
6052 	}
6053 
6054 	if (dolock)
6055 		rw_exit(&ip->i_contents);
6056 	if (vmpss && !atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
6057 		cv_broadcast(&ulp->ul_cv);
6058 	return (err);
6059 }
6060 
6061 /*
6062  * Called when the kernel is in a frozen state to dump data
6063  * directly to the device. It uses a private dump data structure,
6064  * set up by dump_ctl, to locate the correct disk block to which to dump.
6065  */
6066 static int
6067 ufs_dump(vnode_t *vp, caddr_t addr, int ldbn, int dblks)
6068 {
6069 	u_offset_t	file_size;
6070 	struct inode    *ip = VTOI(vp);
6071 	struct fs	*fs = ip->i_fs;
6072 	daddr_t		dbn, lfsbn;
6073 	int		disk_blks = fs->fs_bsize >> DEV_BSHIFT;
6074 	int		error = 0;
6075 	int		ndbs, nfsbs;
6076 
6077 	/*
6078 	 * forced unmount case
6079 	 */
6080 	if (ip->i_ufsvfs == NULL)
6081 		return (EIO);
6082 	/*
6083 	 * Validate the inode that it has not been modified since
6084 	 * the dump structure is allocated.
6085 	 */
6086 	mutex_enter(&ip->i_tlock);
6087 	if ((dump_info == NULL) ||
6088 	    (dump_info->ip != ip) ||
6089 	    (dump_info->time.tv_sec != ip->i_mtime.tv_sec) ||
6090 	    (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) {
6091 		mutex_exit(&ip->i_tlock);
6092 		return (-1);
6093 	}
6094 	mutex_exit(&ip->i_tlock);
6095 
6096 	/*
6097 	 * See that the file has room for this write
6098 	 */
6099 	UFS_GET_ISIZE(&file_size, ip);
6100 
6101 	if (ldbtob((offset_t)(ldbn + dblks)) > file_size)
6102 		return (ENOSPC);
6103 
6104 	/*
6105 	 * Find the physical disk block numbers from the dump
6106 	 * private data structure directly and write out the data
6107 	 * in contiguous block lumps
6108 	 */
6109 	while (dblks > 0 && !error) {
6110 		lfsbn = (daddr_t)lblkno(fs, ldbtob((offset_t)ldbn));
6111 		dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks;
6112 		nfsbs = 1;
6113 		ndbs = disk_blks - ldbn % disk_blks;
6114 		while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn +
6115 		    nfsbs]) == dbn + ndbs) {
6116 			nfsbs++;
6117 			ndbs += disk_blks;
6118 		}
6119 		if (ndbs > dblks)
6120 			ndbs = dblks;
6121 		error = bdev_dump(ip->i_dev, addr, dbn, ndbs);
6122 		addr += ldbtob((offset_t)ndbs);
6123 		dblks -= ndbs;
6124 		ldbn += ndbs;
6125 	}
6126 	return (error);
6127 
6128 }
6129 
6130 /*
6131  * Prepare the file system before and after the dump operation.
6132  *
6133  * action = DUMP_ALLOC:
6134  * Preparation before dump, allocate dump private data structure
6135  * to hold all the direct and indirect block info for dump.
6136  *
6137  * action = DUMP_FREE:
6138  * Clean up after dump, deallocate the dump private data structure.
6139  *
6140  * action = DUMP_SCAN:
6141  * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6142  * if found, the starting file-relative DEV_BSIZE lbn is written
6143  * to *bklp; that lbn is intended for use with VOP_DUMP()
6144  */
6145 static int
6146 ufs_dumpctl(vnode_t *vp, int action, int *blkp)
6147 {
6148 	struct inode	*ip = VTOI(vp);
6149 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
6150 	struct fs	*fs;
6151 	daddr32_t	*dblk, *storeblk;
6152 	daddr32_t	*nextblk, *endblk;
6153 	struct buf	*bp;
6154 	int		i, entry, entries;
6155 	int		n, ncontig;
6156 
6157 	/*
6158 	 * check for forced unmount
6159 	 */
6160 	if (ufsvfsp == NULL)
6161 		return (EIO);
6162 
6163 	if (action == DUMP_ALLOC) {
6164 		/*
6165 		 * alloc and record dump_info
6166 		 */
6167 		if (dump_info != NULL)
6168 			return (EINVAL);
6169 
6170 		ASSERT(vp->v_type == VREG);
6171 		fs = ufsvfsp->vfs_fs;
6172 
6173 		rw_enter(&ip->i_contents, RW_READER);
6174 
6175 		if (bmap_has_holes(ip)) {
6176 			rw_exit(&ip->i_contents);
6177 			return (EFAULT);
6178 		}
6179 
6180 		/*
6181 		 * calculate and allocate space needed according to i_size
6182 		 */
6183 		entries = (int)lblkno(fs, blkroundup(fs, ip->i_size));
6184 		if ((dump_info = (struct dump *)
6185 		    kmem_alloc(sizeof (struct dump) +
6186 		    (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP)) == NULL) {
6187 			    rw_exit(&ip->i_contents);
6188 			    return (ENOMEM);
6189 		}
6190 
6191 		/* Start saving the info */
6192 		dump_info->fsbs = entries;
6193 		dump_info->ip = ip;
6194 		storeblk = &dump_info->dblk[0];
6195 
6196 		/* Direct Blocks */
6197 		for (entry = 0; entry < NDADDR && entry < entries; entry++)
6198 			*storeblk++ = ip->i_db[entry];
6199 
6200 		/* Indirect Blocks */
6201 		for (i = 0; i < NIADDR; i++) {
6202 			int error = 0;
6203 
6204 			bp = UFS_BREAD(ufsvfsp,
6205 				ip->i_dev, fsbtodb(fs, ip->i_ib[i]),
6206 				fs->fs_bsize);
6207 			if (bp->b_flags & B_ERROR)
6208 				error = EIO;
6209 			else {
6210 				dblk = bp->b_un.b_daddr;
6211 				if ((storeblk = save_dblks(ip, ufsvfsp,
6212 				    storeblk, dblk, i, entries)) == NULL)
6213 					error = EIO;
6214 			}
6215 
6216 			brelse(bp);
6217 
6218 			if (error != 0) {
6219 				kmem_free(dump_info, sizeof (struct dump) +
6220 				    (entries - 1) * sizeof (daddr32_t));
6221 				rw_exit(&ip->i_contents);
6222 				dump_info = NULL;
6223 				return (error);
6224 			}
6225 		}
6226 		/* and time stamp the information */
6227 		mutex_enter(&ip->i_tlock);
6228 		dump_info->time = ip->i_mtime;
6229 		mutex_exit(&ip->i_tlock);
6230 
6231 		rw_exit(&ip->i_contents);
6232 	} else if (action == DUMP_FREE) {
6233 		/*
6234 		 * free dump_info
6235 		 */
6236 		if (dump_info == NULL)
6237 			return (EINVAL);
6238 		entries = dump_info->fsbs - 1;
6239 		kmem_free(dump_info, sizeof (struct dump) +
6240 		    entries * sizeof (daddr32_t));
6241 		dump_info = NULL;
6242 	} else if (action == DUMP_SCAN) {
6243 		/*
6244 		 * scan dump_info
6245 		 */
6246 		if (dump_info == NULL)
6247 			return (EINVAL);
6248 
6249 		dblk = dump_info->dblk;
6250 		nextblk = dblk + 1;
6251 		endblk = dblk + dump_info->fsbs - 1;
6252 		fs = ufsvfsp->vfs_fs;
6253 		ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT);
6254 
6255 		/*
6256 		 * scan dblk[] entries; contig fs space is found when:
6257 		 * ((current blkno + frags per block) == next blkno)
6258 		 */
6259 		n = 0;
6260 		while (n < ncontig && dblk < endblk) {
6261 			if ((*dblk + fs->fs_frag) == *nextblk)
6262 				n++;
6263 			else
6264 				n = 0;
6265 			dblk++;
6266 			nextblk++;
6267 		}
6268 
6269 		/*
6270 		 * index is where size bytes of contig space begins;
6271 		 * conversion from index to the file's DEV_BSIZE lbn
6272 		 * is equivalent to:  (index * fs_bsize) / DEV_BSIZE
6273 		 */
6274 		if (n == ncontig) {
6275 			i = (dblk - dump_info->dblk) - ncontig;
6276 			*blkp = i << (fs->fs_bshift - DEV_BSHIFT);
6277 		} else
6278 			return (EFAULT);
6279 	}
6280 	return (0);
6281 }
6282 
6283 /*
6284  * Recursive helper function for ufs_dumpctl().  It follows the indirect file
6285  * system  blocks until it reaches the the disk block addresses, which are
6286  * then stored into the given buffer, storeblk.
6287  */
6288 static daddr32_t *
6289 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp,  daddr32_t *storeblk,
6290     daddr32_t *dblk, int level, int entries)
6291 {
6292 	struct fs	*fs = ufsvfsp->vfs_fs;
6293 	struct buf	*bp;
6294 	int		i;
6295 
6296 	if (level == 0) {
6297 		for (i = 0; i < NINDIR(fs); i++) {
6298 			if (storeblk - dump_info->dblk >= entries)
6299 				break;
6300 			*storeblk++ = dblk[i];
6301 		}
6302 		return (storeblk);
6303 	}
6304 	for (i = 0; i < NINDIR(fs); i++) {
6305 		if (storeblk - dump_info->dblk >= entries)
6306 			break;
6307 		bp = UFS_BREAD(ufsvfsp,
6308 				ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize);
6309 		if (bp->b_flags & B_ERROR) {
6310 			brelse(bp);
6311 			return (NULL);
6312 		}
6313 		storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr,
6314 		    level - 1, entries);
6315 		brelse(bp);
6316 
6317 		if (storeblk == NULL)
6318 			return (NULL);
6319 	}
6320 	return (storeblk);
6321 }
6322 
6323 /* ARGSUSED */
6324 static int
6325 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag,
6326 	struct cred *cr)
6327 {
6328 	struct inode	*ip = VTOI(vp);
6329 	struct ulockfs	*ulp;
6330 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
6331 	ulong_t		vsa_mask = vsap->vsa_mask;
6332 	int		err = EINVAL;
6333 
6334 	TRACE_3(TR_FAC_UFS, TR_UFS_GETSECATTR_START,
6335 	    "ufs_getsecattr_start:vp %p, vsap %p, flags %x", vp, vsap, flag);
6336 
6337 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6338 
6339 	/*
6340 	 * Only grab locks if needed - they're not needed to check vsa_mask
6341 	 * or if the mask contains no acl flags.
6342 	 */
6343 	if (vsa_mask != 0) {
6344 		if (err = ufs_lockfs_begin(ufsvfsp, &ulp,
6345 		    ULOCKFS_GETATTR_MASK))
6346 			return (err);
6347 
6348 		rw_enter(&ip->i_contents, RW_READER);
6349 		err = ufs_acl_get(ip, vsap, flag, cr);
6350 		rw_exit(&ip->i_contents);
6351 
6352 		if (ulp)
6353 			ufs_lockfs_end(ulp);
6354 	}
6355 	TRACE_1(TR_FAC_UFS, TR_UFS_GETSECATTR_END,
6356 	    "ufs_getsecattr_end:vp %p", vp);
6357 	return (err);
6358 }
6359 
6360 /* ARGSUSED */
6361 static int
6362 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr)
6363 {
6364 	struct inode	*ip = VTOI(vp);
6365 	struct ulockfs	*ulp = NULL;
6366 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
6367 	ulong_t		vsa_mask = vsap->vsa_mask;
6368 	int		err;
6369 	int		haverwlock = 1;
6370 	int		trans_size;
6371 	int		donetrans = 0;
6372 	int		retry = 1;
6373 
6374 
6375 	TRACE_3(TR_FAC_UFS, TR_UFS_SETSECATTR_START,
6376 	    "ufs_setsecattr_start:vp %p, vsap %p, flags %x", vp, vsap, flag);
6377 
6378 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
6379 
6380 	/* Abort now if the request is either empty or invalid. */
6381 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6382 	if ((vsa_mask == 0) ||
6383 	    ((vsap->vsa_aclentp == NULL) &&
6384 	    (vsap->vsa_dfaclentp == NULL))) {
6385 		err = EINVAL;
6386 		goto out;
6387 	}
6388 
6389 	/*
6390 	 * Following convention, if this is a directory then we acquire the
6391 	 * inode's i_rwlock after starting a UFS logging transaction;
6392 	 * otherwise, we acquire it beforehand. Since we were called (and
6393 	 * must therefore return) with the lock held, we will have to drop it,
6394 	 * and later reacquire it, if operating on a directory.
6395 	 */
6396 	if (vp->v_type == VDIR) {
6397 		rw_exit(&ip->i_rwlock);
6398 		haverwlock = 0;
6399 	} else {
6400 		/* Upgrade the lock if required. */
6401 		if (!rw_write_held(&ip->i_rwlock)) {
6402 			rw_exit(&ip->i_rwlock);
6403 			rw_enter(&ip->i_rwlock, RW_WRITER);
6404 		}
6405 	}
6406 
6407 again:
6408 	ASSERT(!(vp->v_type == VDIR && haverwlock));
6409 	if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) {
6410 		ulp = NULL;
6411 		retry = 0;
6412 		goto out;
6413 	}
6414 
6415 	/*
6416 	 * Check that the file system supports this operation. Note that
6417 	 * ufs_lockfs_begin() will have checked that the file system had
6418 	 * not been forcibly unmounted.
6419 	 */
6420 	if (ufsvfsp->vfs_fs->fs_ronly) {
6421 		err = EROFS;
6422 		goto out;
6423 	}
6424 	if (ufsvfsp->vfs_nosetsec) {
6425 		err = ENOSYS;
6426 		goto out;
6427 	}
6428 
6429 	if (ulp) {
6430 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR,
6431 			trans_size = TOP_SETSECATTR_SIZE(VTOI(vp)));
6432 		donetrans = 1;
6433 	}
6434 
6435 	if (vp->v_type == VDIR) {
6436 		rw_enter(&ip->i_rwlock, RW_WRITER);
6437 		haverwlock = 1;
6438 	}
6439 
6440 	ASSERT(haverwlock);
6441 
6442 	/* Do the actual work. */
6443 	rw_enter(&ip->i_contents, RW_WRITER);
6444 	/*
6445 	 * Suppress out of inodes messages if we will retry.
6446 	 */
6447 	if (retry)
6448 		ip->i_flag |= IQUIET;
6449 	err = ufs_acl_set(ip, vsap, flag, cr);
6450 	ip->i_flag &= ~IQUIET;
6451 	rw_exit(&ip->i_contents);
6452 
6453 out:
6454 	if (ulp) {
6455 		if (donetrans) {
6456 			/*
6457 			 * top_end_async() can eventually call
6458 			 * top_end_sync(), which can block. We must
6459 			 * therefore observe the lock-ordering protocol
6460 			 * here as well.
6461 			 */
6462 			if (vp->v_type == VDIR) {
6463 				rw_exit(&ip->i_rwlock);
6464 				haverwlock = 0;
6465 			}
6466 			TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size);
6467 		}
6468 		ufs_lockfs_end(ulp);
6469 	}
6470 	/*
6471 	 * If no inodes available, try scaring a logically-
6472 	 * free one out of the delete queue to someplace
6473 	 * that we can find it.
6474 	 */
6475 	if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
6476 		ufs_delete_drain_wait(ufsvfsp, 1);
6477 		retry = 0;
6478 		if (vp->v_type == VDIR && haverwlock) {
6479 			rw_exit(&ip->i_rwlock);
6480 			haverwlock = 0;
6481 		}
6482 		goto again;
6483 	}
6484 	/*
6485 	 * If we need to reacquire the lock then it is safe to do so
6486 	 * as a reader. This is because ufs_rwunlock(), which will be
6487 	 * called by our caller after we return, does not differentiate
6488 	 * between shared and exclusive locks.
6489 	 */
6490 	if (!haverwlock) {
6491 		ASSERT(vp->v_type == VDIR);
6492 		rw_enter(&ip->i_rwlock, RW_READER);
6493 	}
6494 
6495 	TRACE_1(TR_FAC_UFS, TR_UFS_SETSECATTR_END,
6496 	    "ufs_setsecattr_end:vp %p", vp);
6497 	return (err);
6498 }
6499