xref: /titanic_41/usr/src/uts/common/fs/ufs/ufs_vnops.c (revision fd9cb95cbb2f626355a60efb9d02c5f0a33c10e6)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
28 /*	  All Rights Reserved  	*/
29 
30 /*
31  * Portions of this source code were derived from Berkeley 4.3 BSD
32  * under license from the Regents of the University of California.
33  */
34 
35 #pragma ident	"%Z%%M%	%I%	%E% SMI"
36 
37 #include <sys/types.h>
38 #include <sys/t_lock.h>
39 #include <sys/ksynch.h>
40 #include <sys/param.h>
41 #include <sys/time.h>
42 #include <sys/systm.h>
43 #include <sys/sysmacros.h>
44 #include <sys/resource.h>
45 #include <sys/signal.h>
46 #include <sys/cred.h>
47 #include <sys/user.h>
48 #include <sys/buf.h>
49 #include <sys/vfs.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 /*
167  * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
168  *
169  * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
170  */
171 struct vnodeops *ufs_vnodeops;
172 
173 const fs_operation_def_t ufs_vnodeops_template[] = {
174 	VOPNAME_OPEN, ufs_open,	/* will not be blocked by lockfs */
175 	VOPNAME_CLOSE, ufs_close,	/* will not be blocked by lockfs */
176 	VOPNAME_READ, ufs_read,
177 	VOPNAME_WRITE, ufs_write,
178 	VOPNAME_IOCTL, ufs_ioctl,
179 	VOPNAME_GETATTR, ufs_getattr,
180 	VOPNAME_SETATTR, ufs_setattr,
181 	VOPNAME_ACCESS, ufs_access,
182 	VOPNAME_LOOKUP, ufs_lookup,
183 	VOPNAME_CREATE, ufs_create,
184 	VOPNAME_REMOVE, ufs_remove,
185 	VOPNAME_LINK, ufs_link,
186 	VOPNAME_RENAME, ufs_rename,
187 	VOPNAME_MKDIR, ufs_mkdir,
188 	VOPNAME_RMDIR, ufs_rmdir,
189 	VOPNAME_READDIR, ufs_readdir,
190 	VOPNAME_SYMLINK, ufs_symlink,
191 	VOPNAME_READLINK, ufs_readlink,
192 	VOPNAME_FSYNC, ufs_fsync,
193 	VOPNAME_INACTIVE, (fs_generic_func_p) ufs_inactive,  /* not blocked */
194 	VOPNAME_FID, ufs_fid,
195 	VOPNAME_RWLOCK, ufs_rwlock, /* not blocked */
196 	VOPNAME_RWUNLOCK, (fs_generic_func_p) ufs_rwunlock,  /* not blocked */
197 	VOPNAME_SEEK, ufs_seek,
198 	VOPNAME_FRLOCK, ufs_frlock,
199 	VOPNAME_SPACE, ufs_space,
200 	VOPNAME_GETPAGE, ufs_getpage,
201 	VOPNAME_PUTPAGE, ufs_putpage,
202 	VOPNAME_MAP, (fs_generic_func_p) ufs_map,
203 	VOPNAME_ADDMAP, (fs_generic_func_p) ufs_addmap,	/* not blocked */
204 	VOPNAME_DELMAP, ufs_delmap,	/* will not be blocked by lockfs */
205 	VOPNAME_POLL, (fs_generic_func_p) ufs_poll,	/* not blocked */
206 	VOPNAME_DUMP, ufs_dump,
207 	VOPNAME_PATHCONF, ufs_l_pathconf,
208 	VOPNAME_PAGEIO, ufs_pageio,
209 	VOPNAME_DUMPCTL, ufs_dumpctl,
210 	VOPNAME_GETSECATTR, ufs_getsecattr,
211 	VOPNAME_SETSECATTR, ufs_setsecattr,
212 	VOPNAME_VNEVENT, fs_vnevent_support,
213 	NULL, NULL
214 };
215 
216 #define	MAX_BACKFILE_COUNT	9999
217 
218 /*
219  * Created by ufs_dumpctl() to store a file's disk block info into memory.
220  * Used by ufs_dump() to dump data to disk directly.
221  */
222 struct dump {
223 	struct inode	*ip;		/* the file we contain */
224 	daddr_t		fsbs;		/* number of blocks stored */
225 	struct timeval32 time;		/* time stamp for the struct */
226 	daddr32_t 	dblk[1];	/* place holder for block info */
227 };
228 
229 static struct dump *dump_info = NULL;
230 
231 /*
232  * Previously there was no special action required for ordinary files.
233  * (Devices are handled through the device file system.)
234  * Now we support Large Files and Large File API requires open to
235  * fail if file is large.
236  * We could take care to prevent data corruption
237  * by doing an atomic check of size and truncate if file is opened with
238  * FTRUNC flag set but traditionally this is being done by the vfs/vnode
239  * layers. So taking care of truncation here is a change in the existing
240  * semantics of VOP_OPEN and therefore we chose not to implement any thing
241  * here. The check for the size of the file > 2GB is being done at the
242  * vfs layer in routine vn_open().
243  */
244 
245 /* ARGSUSED */
246 static int
247 ufs_open(struct vnode **vpp, int flag, struct cred *cr)
248 {
249 	TRACE_1(TR_FAC_UFS, TR_UFS_OPEN, "ufs_open:vpp %p", vpp);
250 	return (0);
251 }
252 
253 /*ARGSUSED*/
254 static int
255 ufs_close(struct vnode *vp, int flag, int count, offset_t offset,
256 	struct cred *cr)
257 {
258 	TRACE_1(TR_FAC_UFS, TR_UFS_CLOSE, "ufs_close:vp %p", vp);
259 
260 	cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
261 	cleanshares(vp, ttoproc(curthread)->p_pid);
262 
263 	/*
264 	 * Push partially filled cluster at last close.
265 	 * ``last close'' is approximated because the dnlc
266 	 * may have a hold on the vnode.
267 	 * Checking for VBAD here will also act as a forced umount check.
268 	 */
269 	if (vp->v_count <= 2 && vp->v_type != VBAD) {
270 		struct inode *ip = VTOI(vp);
271 		if (ip->i_delaylen) {
272 			ins.in_poc.value.ul++;
273 			(void) ufs_putpages(vp, ip->i_delayoff, ip->i_delaylen,
274 					B_ASYNC | B_FREE, cr);
275 			ip->i_delaylen = 0;
276 		}
277 	}
278 
279 	return (0);
280 }
281 
282 /*ARGSUSED*/
283 static int
284 ufs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cr,
285 	struct caller_context *ct)
286 {
287 	struct inode *ip = VTOI(vp);
288 	struct ufsvfs *ufsvfsp;
289 	struct ulockfs *ulp = NULL;
290 	int error = 0;
291 	int intrans = 0;
292 
293 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
294 	TRACE_3(TR_FAC_UFS, TR_UFS_READ_START,
295 		"ufs_read_start:vp %p uiop %p ioflag %x",
296 		vp, uiop, ioflag);
297 
298 	/*
299 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
300 	 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
301 	 */
302 	if (MANDLOCK(vp, ip->i_mode)) {
303 		/*
304 		 * ufs_getattr ends up being called by chklock
305 		 */
306 		error = chklock(vp, FREAD, uiop->uio_loffset,
307 				uiop->uio_resid, uiop->uio_fmode, ct);
308 		if (error)
309 			goto out;
310 	}
311 
312 	ufsvfsp = ip->i_ufsvfs;
313 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READ_MASK);
314 	if (error)
315 		goto out;
316 
317 	/*
318 	 * In the case that a directory is opened for reading as a file
319 	 * (eg "cat .") with the  O_RSYNC, O_SYNC and O_DSYNC flags set.
320 	 * The locking order had to be changed to avoid a deadlock with
321 	 * an update taking place on that directory at the same time.
322 	 */
323 	if ((ip->i_mode & IFMT) == IFDIR) {
324 
325 		rw_enter(&ip->i_contents, RW_READER);
326 		error = rdip(ip, uiop, ioflag, cr);
327 		rw_exit(&ip->i_contents);
328 
329 		if (error) {
330 			if (ulp)
331 				ufs_lockfs_end(ulp);
332 			goto out;
333 		}
334 
335 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
336 		    TRANS_ISTRANS(ufsvfsp)) {
337 			rw_exit(&ip->i_rwlock);
338 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
339 			    error);
340 			ASSERT(!error);
341 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
342 			    TOP_READ_SIZE);
343 			rw_enter(&ip->i_rwlock, RW_READER);
344 		}
345 	} else {
346 		/*
347 		 * Only transact reads to files opened for sync-read and
348 		 * sync-write on a file system that is not write locked.
349 		 *
350 		 * The ``not write locked'' check prevents problems with
351 		 * enabling/disabling logging on a busy file system.  E.g.,
352 		 * logging exists at the beginning of the read but does not
353 		 * at the end.
354 		 *
355 		 */
356 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
357 		    TRANS_ISTRANS(ufsvfsp)) {
358 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
359 			    error);
360 			ASSERT(!error);
361 			intrans = 1;
362 		}
363 
364 		rw_enter(&ip->i_contents, RW_READER);
365 		error = rdip(ip, uiop, ioflag, cr);
366 		rw_exit(&ip->i_contents);
367 
368 		if (intrans) {
369 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
370 			    TOP_READ_SIZE);
371 		}
372 	}
373 
374 	if (ulp) {
375 		ufs_lockfs_end(ulp);
376 	}
377 out:
378 
379 	TRACE_2(TR_FAC_UFS, TR_UFS_READ_END,
380 		"ufs_read_end:vp %p error %d", vp, error);
381 	return (error);
382 }
383 
384 extern	int	ufs_HW;		/* high water mark */
385 extern	int	ufs_LW;		/* low water mark */
386 int	ufs_WRITES = 1;		/* XXX - enable/disable */
387 int	ufs_throttles = 0;	/* throttling count */
388 int	ufs_allow_shared_writes = 1;	/* directio shared writes */
389 
390 static int
391 ufs_check_rewrite(struct inode *ip, struct uio *uiop, int ioflag)
392 {
393 	int	shared_write;
394 
395 	/*
396 	 * If the FDSYNC flag is set then ignore the global
397 	 * ufs_allow_shared_writes in this case.
398 	 */
399 	shared_write = (ioflag & FDSYNC) | ufs_allow_shared_writes;
400 
401 	/*
402 	 * Filter to determine if this request is suitable as a
403 	 * concurrent rewrite. This write must not allocate blocks
404 	 * by extending the file or filling in holes. No use trying
405 	 * through FSYNC descriptors as the inode will be synchronously
406 	 * updated after the write. The uio structure has not yet been
407 	 * checked for sanity, so assume nothing.
408 	 */
409 	return (((ip->i_mode & IFMT) == IFREG) && !(ioflag & FAPPEND) &&
410 		(uiop->uio_loffset >= (offset_t)0) &&
411 		(uiop->uio_loffset < ip->i_size) && (uiop->uio_resid > 0) &&
412 		((ip->i_size - uiop->uio_loffset) >= uiop->uio_resid) &&
413 		!(ioflag & FSYNC) && !bmap_has_holes(ip) &&
414 		shared_write);
415 }
416 
417 /*ARGSUSED*/
418 static int
419 ufs_write(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cr,
420 	caller_context_t *ct)
421 {
422 	struct inode *ip = VTOI(vp);
423 	struct ufsvfs *ufsvfsp;
424 	struct ulockfs *ulp;
425 	int retry = 1;
426 	int error, resv, resid = 0;
427 	int directio_status;
428 	int exclusive;
429 	int rewriteflg;
430 	long start_resid = uiop->uio_resid;
431 
432 	TRACE_3(TR_FAC_UFS, TR_UFS_WRITE_START,
433 		"ufs_write_start:vp %p uiop %p ioflag %x",
434 		vp, uiop, ioflag);
435 
436 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
437 
438 retry_mandlock:
439 	/*
440 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
441 	 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
442 	 * Check for forced unmounts normally done in ufs_lockfs_begin().
443 	 */
444 	if ((ufsvfsp = ip->i_ufsvfs) == NULL) {
445 		error = EIO;
446 		goto out;
447 	}
448 	if (MANDLOCK(vp, ip->i_mode)) {
449 
450 		ASSERT(RW_WRITE_HELD(&ip->i_rwlock));
451 
452 		/*
453 		 * ufs_getattr ends up being called by chklock
454 		 */
455 		error = chklock(vp, FWRITE, uiop->uio_loffset,
456 				uiop->uio_resid, uiop->uio_fmode, ct);
457 		if (error)
458 			goto out;
459 	}
460 
461 	/* i_rwlock can change in chklock */
462 	exclusive = rw_write_held(&ip->i_rwlock);
463 	rewriteflg = ufs_check_rewrite(ip, uiop, ioflag);
464 
465 	/*
466 	 * Check for fast-path special case of directio re-writes.
467 	 */
468 	if ((ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) &&
469 	    !exclusive && rewriteflg) {
470 
471 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
472 		if (error)
473 			goto out;
474 
475 		rw_enter(&ip->i_contents, RW_READER);
476 		error = ufs_directio_write(ip, uiop, ioflag, 1, cr,
477 			&directio_status);
478 		if (directio_status == DIRECTIO_SUCCESS) {
479 			uint_t i_flag_save;
480 
481 			if (start_resid != uiop->uio_resid)
482 				error = 0;
483 			/*
484 			 * Special treatment of access times for re-writes.
485 			 * If IMOD is not already set, then convert it
486 			 * to IMODACC for this operation. This defers
487 			 * entering a delta into the log until the inode
488 			 * is flushed. This mimics what is done for read
489 			 * operations and inode access time.
490 			 */
491 			mutex_enter(&ip->i_tlock);
492 			i_flag_save = ip->i_flag;
493 			ip->i_flag |= IUPD | ICHG;
494 			ip->i_seq++;
495 			ITIMES_NOLOCK(ip);
496 			if ((i_flag_save & IMOD) == 0) {
497 				ip->i_flag &= ~IMOD;
498 				ip->i_flag |= IMODACC;
499 			}
500 			mutex_exit(&ip->i_tlock);
501 			rw_exit(&ip->i_contents);
502 			if (ulp)
503 				ufs_lockfs_end(ulp);
504 			goto out;
505 		}
506 		rw_exit(&ip->i_contents);
507 		if (ulp)
508 			ufs_lockfs_end(ulp);
509 	}
510 
511 	if (!exclusive && !rw_tryupgrade(&ip->i_rwlock)) {
512 		rw_exit(&ip->i_rwlock);
513 		rw_enter(&ip->i_rwlock, RW_WRITER);
514 		/*
515 		 * Mandatory locking could have been enabled
516 		 * after dropping the i_rwlock.
517 		 */
518 		if (MANDLOCK(vp, ip->i_mode))
519 			goto retry_mandlock;
520 	}
521 
522 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
523 	if (error)
524 		goto out;
525 
526 	/*
527 	 * Amount of log space needed for this write
528 	 */
529 	if (!rewriteflg || !(ioflag & FDSYNC))
530 		TRANS_WRITE_RESV(ip, uiop, ulp, &resv, &resid);
531 
532 	/*
533 	 * Throttle writes.
534 	 */
535 	if (ufs_WRITES && (ip->i_writes > ufs_HW)) {
536 		mutex_enter(&ip->i_tlock);
537 		while (ip->i_writes > ufs_HW) {
538 			ufs_throttles++;
539 			cv_wait(&ip->i_wrcv, &ip->i_tlock);
540 		}
541 		mutex_exit(&ip->i_tlock);
542 	}
543 
544 	/*
545 	 * Enter Transaction
546 	 *
547 	 * If the write is a rewrite there is no need to open a transaction
548 	 * if the FDSYNC flag is set and not the FSYNC.  In this case just
549 	 * set the IMODACC flag to modify do the update at a later time
550 	 * thus avoiding the overhead of the logging transaction that is
551 	 * not required.
552 	 */
553 	if (ioflag & (FSYNC|FDSYNC)) {
554 		if (ulp) {
555 			if (rewriteflg) {
556 				uint_t i_flag_save;
557 
558 				rw_enter(&ip->i_contents, RW_READER);
559 				mutex_enter(&ip->i_tlock);
560 				i_flag_save = ip->i_flag;
561 				ip->i_flag |= IUPD | ICHG;
562 				ip->i_seq++;
563 				ITIMES_NOLOCK(ip);
564 				if ((i_flag_save & IMOD) == 0) {
565 					ip->i_flag &= ~IMOD;
566 					ip->i_flag |= IMODACC;
567 				}
568 				mutex_exit(&ip->i_tlock);
569 				rw_exit(&ip->i_contents);
570 			} else {
571 				int terr = 0;
572 				TRANS_BEGIN_SYNC(ufsvfsp, TOP_WRITE_SYNC, resv,
573 				    terr);
574 				ASSERT(!terr);
575 			}
576 		}
577 	} else {
578 		if (ulp)
579 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_WRITE, resv);
580 	}
581 
582 	/*
583 	 * Write the file
584 	 */
585 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
586 	rw_enter(&ip->i_contents, RW_WRITER);
587 	if ((ioflag & FAPPEND) != 0 && (ip->i_mode & IFMT) == IFREG) {
588 		/*
589 		 * In append mode start at end of file.
590 		 */
591 		uiop->uio_loffset = ip->i_size;
592 	}
593 
594 	/*
595 	 * Mild optimisation, don't call ufs_trans_write() unless we have to
596 	 * Also, suppress file system full messages if we will retry.
597 	 */
598 	if (retry)
599 		ip->i_flag |= IQUIET;
600 	if (resid) {
601 		TRANS_WRITE(ip, uiop, ioflag, error, ulp, cr, resv, resid);
602 	} else {
603 		error = wrip(ip, uiop, ioflag, cr);
604 	}
605 	ip->i_flag &= ~IQUIET;
606 
607 	rw_exit(&ip->i_contents);
608 	rw_exit(&ufsvfsp->vfs_dqrwlock);
609 
610 	/*
611 	 * Leave Transaction
612 	 */
613 	if (ulp) {
614 		if (ioflag & (FSYNC|FDSYNC)) {
615 			if (!rewriteflg) {
616 				int terr = 0;
617 
618 				TRANS_END_SYNC(ufsvfsp, terr, TOP_WRITE_SYNC,
619 					resv);
620 				if (error == 0)
621 					error = terr;
622 			}
623 		} else {
624 			TRANS_END_ASYNC(ufsvfsp, TOP_WRITE, resv);
625 		}
626 		ufs_lockfs_end(ulp);
627 	}
628 out:
629 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
630 		/*
631 		 * Any blocks tied up in pending deletes?
632 		 */
633 		ufs_delete_drain_wait(ufsvfsp, 1);
634 		retry = 0;
635 		goto retry_mandlock;
636 	}
637 
638 	if (error == ENOSPC && (start_resid != uiop->uio_resid))
639 		error = 0;
640 
641 	TRACE_2(TR_FAC_UFS, TR_UFS_WRITE_END,
642 		"ufs_write_end:vp %p error %d", vp, error);
643 	return (error);
644 }
645 
646 /*
647  * Don't cache write blocks to files with the sticky bit set.
648  * Used to keep swap files from blowing the page cache on a server.
649  */
650 int stickyhack = 1;
651 
652 /*
653  * Free behind hacks.  The pager is busted.
654  * XXX - need to pass the information down to writedone() in a flag like B_SEQ
655  * or B_FREE_IF_TIGHT_ON_MEMORY.
656  */
657 int	freebehind = 1;
658 int	smallfile = 0;
659 u_offset_t smallfile64 = 32 * 1024;
660 
661 /*
662  * While we should, in most cases, cache the pages for write, we
663  * may also want to cache the pages for read as long as they are
664  * frequently re-usable.
665  *
666  * If cache_read_ahead = 1, the pages for read will go to the tail
667  * of the cache list when they are released, otherwise go to the head.
668  */
669 int	cache_read_ahead = 0;
670 
671 /*
672  * Freebehind exists  so that as we read  large files  sequentially we
673  * don't consume most of memory with pages  from a few files. It takes
674  * longer to re-read from disk multiple small files as it does reading
675  * one large one sequentially.  As system  memory grows customers need
676  * to retain bigger chunks   of files in  memory.   The advent of  the
677  * cachelist opens up of the possibility freeing pages  to the head or
678  * tail of the list.
679  *
680  * Not freeing a page is a bet that the page will be read again before
681  * it's segmap slot is needed for something else. If we loose the bet,
682  * it means some  other thread is  burdened with the  page free we did
683  * not do. If we win we save a free and reclaim.
684  *
685  * Freeing it at the tail  vs the head of cachelist  is a bet that the
686  * page will survive until the next  read.  It's also saying that this
687  * page is more likely to  be re-used than a  page freed some time ago
688  * and never reclaimed.
689  *
690  * Freebehind maintains a  range of  file offset [smallfile1; smallfile2]
691  *
692  *            0 < offset < smallfile1 : pages are not freed.
693  *   smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
694  *   smallfile2 < offset              : pages freed to head of cachelist.
695  *
696  * The range  is  computed  at most  once  per second  and  depends on
697  * freemem  and  ncpus_online.  Both parameters  are   bounded to be
698  * >= smallfile && >= smallfile64.
699  *
700  * smallfile1 = (free memory / ncpu) / 1000
701  * smallfile2 = (free memory / ncpu) / 10
702  *
703  * A few examples values:
704  *
705  *       Free Mem (in Bytes) [smallfile1; smallfile2]  [smallfile1; smallfile2]
706  *                                 ncpus_online = 4          ncpus_online = 64
707  *       ------------------  -----------------------   -----------------------
708  *             1G                   [256K;  25M]               [32K; 1.5M]
709  *            10G                   [2.5M; 250M]              [156K; 15M]
710  *           100G                    [25M; 2.5G]              [1.5M; 150M]
711  *
712  */
713 
714 #define	SMALLFILE1_D 1000
715 #define	SMALLFILE2_D 10
716 static u_offset_t smallfile1 = 32 * 1024;
717 static u_offset_t smallfile2 = 32 * 1024;
718 static clock_t smallfile_update = 0; /* lbolt value of when to recompute */
719 uint_t smallfile1_d = SMALLFILE1_D;
720 uint_t smallfile2_d = SMALLFILE2_D;
721 
722 /*
723  * wrip does the real work of write requests for ufs.
724  */
725 int
726 wrip(struct inode *ip, struct uio *uio, int ioflag, struct cred *cr)
727 {
728 	rlim64_t limit = uio->uio_llimit;
729 	u_offset_t off;
730 	u_offset_t old_i_size;
731 	struct fs *fs;
732 	struct vnode *vp;
733 	struct ufsvfs *ufsvfsp;
734 	caddr_t base;
735 	long start_resid = uio->uio_resid;	/* save starting resid */
736 	long premove_resid;			/* resid before uiomove() */
737 	uint_t flags;
738 	int newpage;
739 	int iupdat_flag, directio_status;
740 	int n, on, mapon;
741 	int error, pagecreate;
742 	int do_dqrwlock;		/* drop/reacquire vfs_dqrwlock */
743 	int32_t	iblocks;
744 	int	new_iblocks;
745 
746 	/*
747 	 * ip->i_size is incremented before the uiomove
748 	 * is done on a write.  If the move fails (bad user
749 	 * address) reset ip->i_size.
750 	 * The better way would be to increment ip->i_size
751 	 * only if the uiomove succeeds.
752 	 */
753 	int i_size_changed = 0;
754 	o_mode_t type;
755 	int i_seq_needed = 0;
756 
757 	vp = ITOV(ip);
758 
759 	/*
760 	 * check for forced unmount - should not happen as
761 	 * the request passed the lockfs checks.
762 	 */
763 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
764 		return (EIO);
765 
766 	fs = ip->i_fs;
767 
768 	TRACE_1(TR_FAC_UFS, TR_UFS_RWIP_START,
769 		"ufs_wrip_start:vp %p", vp);
770 
771 	ASSERT(RW_WRITE_HELD(&ip->i_contents));
772 
773 	/* check for valid filetype */
774 	type = ip->i_mode & IFMT;
775 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
776 	    (type != IFLNK) && (type != IFSHAD)) {
777 		return (EIO);
778 	}
779 
780 	/*
781 	 * the actual limit of UFS file size
782 	 * is UFS_MAXOFFSET_T
783 	 */
784 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
785 		limit = MAXOFFSET_T;
786 
787 	if (uio->uio_loffset >= limit) {
788 		proc_t *p = ttoproc(curthread);
789 
790 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
791 			"ufs_wrip_end:vp %p error %d", vp, EINVAL);
792 
793 		mutex_enter(&p->p_lock);
794 		(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
795 		    p, RCA_UNSAFE_SIGINFO);
796 		mutex_exit(&p->p_lock);
797 		return (EFBIG);
798 	}
799 
800 	/*
801 	 * if largefiles are disallowed, the limit is
802 	 * the pre-largefiles value of 2GB
803 	 */
804 	if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
805 		limit = MIN(UFS_MAXOFFSET_T, limit);
806 	else
807 		limit = MIN(MAXOFF32_T, limit);
808 
809 	if (uio->uio_loffset < (offset_t)0) {
810 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
811 			"ufs_wrip_end:vp %p error %d", vp, EINVAL);
812 		return (EINVAL);
813 	}
814 	if (uio->uio_resid == 0) {
815 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
816 			"ufs_wrip_end:vp %p error %d", vp, 0);
817 		return (0);
818 	}
819 
820 	if (uio->uio_loffset >= limit)
821 		return (EFBIG);
822 
823 	ip->i_flag |= INOACC;	/* don't update ref time in getpage */
824 
825 	if (ioflag & (FSYNC|FDSYNC)) {
826 		ip->i_flag |= ISYNC;
827 		iupdat_flag = 1;
828 	}
829 	/*
830 	 * Try to go direct
831 	 */
832 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
833 		uio->uio_llimit = limit;
834 		error = ufs_directio_write(ip, uio, ioflag, 0, cr,
835 			&directio_status);
836 		/*
837 		 * If ufs_directio wrote to the file or set the flags,
838 		 * we need to update i_seq, but it may be deferred.
839 		 */
840 		if (start_resid != uio->uio_resid ||
841 					(ip->i_flag & (ICHG|IUPD))) {
842 			i_seq_needed = 1;
843 			ip->i_flag |= ISEQ;
844 		}
845 		if (directio_status == DIRECTIO_SUCCESS)
846 			goto out;
847 	}
848 
849 	/*
850 	 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
851 	 *
852 	 * o shadow inodes: vfs_dqrwlock is not held at all
853 	 * o quota updates: vfs_dqrwlock is read or write held
854 	 * o other updates: vfs_dqrwlock is read held
855 	 *
856 	 * The first case is the only one where we do not hold
857 	 * vfs_dqrwlock at all while entering wrip().
858 	 * We must make sure not to downgrade/drop vfs_dqrwlock if we
859 	 * have it as writer, i.e. if we are updating the quota inode.
860 	 * There is no potential deadlock scenario in this case as
861 	 * ufs_getpage() takes care of this and avoids reacquiring
862 	 * vfs_dqrwlock in that case.
863 	 *
864 	 * This check is done here since the above conditions do not change
865 	 * and we possibly loop below, so save a few cycles.
866 	 */
867 	if ((type == IFSHAD) ||
868 		(rw_owner(&ufsvfsp->vfs_dqrwlock) == curthread)) {
869 			do_dqrwlock = 0;
870 	} else {
871 		do_dqrwlock = 1;
872 	}
873 
874 	/*
875 	 * Large Files: We cast MAXBMASK to offset_t
876 	 * inorder to mask out the higher bits. Since offset_t
877 	 * is a signed value, the high order bit set in MAXBMASK
878 	 * value makes it do the right thing by having all bits 1
879 	 * in the higher word. May be removed for _SOLARIS64_.
880 	 */
881 
882 	fs = ip->i_fs;
883 	do {
884 		u_offset_t uoff = uio->uio_loffset;
885 		off = uoff & (offset_t)MAXBMASK;
886 		mapon = (int)(uoff & (offset_t)MAXBOFFSET);
887 		on = (int)blkoff(fs, uoff);
888 		n = (int)MIN(fs->fs_bsize - on, uio->uio_resid);
889 		new_iblocks = 1;
890 
891 		if (type == IFREG && uoff + n >= limit) {
892 			if (uoff >= limit) {
893 				error = EFBIG;
894 				goto out;
895 			}
896 			/*
897 			 * since uoff + n >= limit,
898 			 * therefore n >= limit - uoff, and n is an int
899 			 * so it is safe to cast it to an int
900 			 */
901 			n = (int)(limit - (rlim64_t)uoff);
902 		}
903 		if (uoff + n > ip->i_size) {
904 			/*
905 			 * We are extending the length of the file.
906 			 * bmap is used so that we are sure that
907 			 * if we need to allocate new blocks, that it
908 			 * is done here before we up the file size.
909 			 */
910 			error = bmap_write(ip, uoff, (int)(on + n),
911 							mapon == 0, cr);
912 			/*
913 			 * bmap_write never drops i_contents so if
914 			 * the flags are set it changed the file.
915 			 */
916 			if (ip->i_flag & (ICHG|IUPD)) {
917 				i_seq_needed = 1;
918 				ip->i_flag |= ISEQ;
919 			}
920 			if (error)
921 				break;
922 			/*
923 			 * There is a window of vulnerability here.
924 			 * The sequence of operations: allocate file
925 			 * system blocks, uiomove the data into pages,
926 			 * and then update the size of the file in the
927 			 * inode, must happen atomically.  However, due
928 			 * to current locking constraints, this can not
929 			 * be done.
930 			 */
931 			ASSERT(ip->i_writer == NULL);
932 			ip->i_writer = curthread;
933 			i_size_changed = 1;
934 			/*
935 			 * If we are writing from the beginning of
936 			 * the mapping, we can just create the
937 			 * pages without having to read them.
938 			 */
939 			pagecreate = (mapon == 0);
940 		} else if (n == MAXBSIZE) {
941 			/*
942 			 * Going to do a whole mappings worth,
943 			 * so we can just create the pages w/o
944 			 * having to read them in.  But before
945 			 * we do that, we need to make sure any
946 			 * needed blocks are allocated first.
947 			 */
948 			iblocks = ip->i_blocks;
949 			error = bmap_write(ip, uoff, (int)(on + n), 1, cr);
950 			/*
951 			 * bmap_write never drops i_contents so if
952 			 * the flags are set it changed the file.
953 			 */
954 			if (ip->i_flag & (ICHG|IUPD)) {
955 				i_seq_needed = 1;
956 				ip->i_flag |= ISEQ;
957 			}
958 			if (error)
959 				break;
960 			pagecreate = 1;
961 			/*
962 			 * check if the new created page needed the
963 			 * allocation of new disk blocks.
964 			 */
965 			if (iblocks == ip->i_blocks)
966 				new_iblocks = 0; /* no new blocks allocated */
967 		} else {
968 			pagecreate = 0;
969 			/*
970 			 * In sync mode flush the indirect blocks which
971 			 * may have been allocated and not written on
972 			 * disk. In above cases bmap_write will allocate
973 			 * in sync mode.
974 			 */
975 			if (ioflag & (FSYNC|FDSYNC)) {
976 				error = ufs_indirblk_sync(ip, uoff);
977 				if (error)
978 					break;
979 			}
980 		}
981 
982 		/*
983 		 * At this point we can enter ufs_getpage() in one
984 		 * of two ways:
985 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
986 		 *    forcefault parameter is true (pagecreate == 0)
987 		 * 2) uiomove() causes a page fault.
988 		 *
989 		 * We have to drop the contents lock to prevent the VM
990 		 * system from trying to reaquire it in ufs_getpage()
991 		 * should the uiomove cause a pagefault.
992 		 *
993 		 * We have to drop the reader vfs_dqrwlock here as well.
994 		 */
995 		rw_exit(&ip->i_contents);
996 		if (do_dqrwlock) {
997 			ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock));
998 			ASSERT(!(RW_WRITE_HELD(&ufsvfsp->vfs_dqrwlock)));
999 			rw_exit(&ufsvfsp->vfs_dqrwlock);
1000 		}
1001 
1002 		base = segmap_getmapflt(segkmap, vp, (off + mapon),
1003 					(uint_t)n, !pagecreate, S_WRITE);
1004 
1005 		/*
1006 		 * segmap_pagecreate() returns 1 if it calls
1007 		 * page_create_va() to allocate any pages.
1008 		 */
1009 		newpage = 0;
1010 
1011 		if (pagecreate)
1012 			newpage = segmap_pagecreate(segkmap, base,
1013 			    (size_t)n, 0);
1014 
1015 		premove_resid = uio->uio_resid;
1016 		error = uiomove(base + mapon, (long)n, UIO_WRITE, uio);
1017 
1018 		/*
1019 		 * If "newpage" is set, then a new page was created and it
1020 		 * does not contain valid data, so it needs to be initialized
1021 		 * at this point.
1022 		 * Otherwise the page contains old data, which was overwritten
1023 		 * partially or as a whole in uiomove.
1024 		 * If there is only one iovec structure within uio, then
1025 		 * on error uiomove will not be able to update uio->uio_loffset
1026 		 * and we would zero the whole page here!
1027 		 *
1028 		 * If uiomove fails because of an error, the old valid data
1029 		 * is kept instead of filling the rest of the page with zero's.
1030 		 */
1031 		if (newpage &&
1032 		    uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) {
1033 			/*
1034 			 * We created pages w/o initializing them completely,
1035 			 * thus we need to zero the part that wasn't set up.
1036 			 * This happens on most EOF write cases and if
1037 			 * we had some sort of error during the uiomove.
1038 			 */
1039 			int nzero, nmoved;
1040 
1041 			nmoved = (int)(uio->uio_loffset - (off + mapon));
1042 			ASSERT(nmoved >= 0 && nmoved <= n);
1043 			nzero = roundup(on + n, PAGESIZE) - nmoved;
1044 			ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE);
1045 			(void) kzero(base + mapon + nmoved, (uint_t)nzero);
1046 		}
1047 
1048 		/*
1049 		 * Unlock the pages allocated by page_create_va()
1050 		 * in segmap_pagecreate()
1051 		 */
1052 		if (newpage)
1053 			segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE);
1054 
1055 		/*
1056 		 * If the size of the file changed, then update the
1057 		 * size field in the inode now.  This can't be done
1058 		 * before the call to segmap_pageunlock or there is
1059 		 * a potential deadlock with callers to ufs_putpage().
1060 		 * They will be holding i_contents and trying to lock
1061 		 * a page, while this thread is holding a page locked
1062 		 * and trying to acquire i_contents.
1063 		 */
1064 		if (i_size_changed) {
1065 			rw_enter(&ip->i_contents, RW_WRITER);
1066 			old_i_size = ip->i_size;
1067 			UFS_SET_ISIZE(uoff + n, ip);
1068 			TRANS_INODE(ufsvfsp, ip);
1069 			/*
1070 			 * file has grown larger than 2GB. Set flag
1071 			 * in superblock to indicate this, if it
1072 			 * is not already set.
1073 			 */
1074 			if ((ip->i_size > MAXOFF32_T) &&
1075 			    !(fs->fs_flags & FSLARGEFILES)) {
1076 				ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES);
1077 				mutex_enter(&ufsvfsp->vfs_lock);
1078 				fs->fs_flags |= FSLARGEFILES;
1079 				ufs_sbwrite(ufsvfsp);
1080 				mutex_exit(&ufsvfsp->vfs_lock);
1081 			}
1082 			mutex_enter(&ip->i_tlock);
1083 			ip->i_writer = NULL;
1084 			cv_broadcast(&ip->i_wrcv);
1085 			mutex_exit(&ip->i_tlock);
1086 			rw_exit(&ip->i_contents);
1087 		}
1088 
1089 		if (error) {
1090 			/*
1091 			 * If we failed on a write, we may have already
1092 			 * allocated file blocks as well as pages.  It's
1093 			 * hard to undo the block allocation, but we must
1094 			 * be sure to invalidate any pages that may have
1095 			 * been allocated.
1096 			 *
1097 			 * If the page was created without initialization
1098 			 * then we must check if it should be possible
1099 			 * to destroy the new page and to keep the old data
1100 			 * on the disk.
1101 			 *
1102 			 * It is possible to destroy the page without
1103 			 * having to write back its contents only when
1104 			 * - the size of the file keeps unchanged
1105 			 * - bmap_write() did not allocate new disk blocks
1106 			 *   it is possible to create big files using "seek" and
1107 			 *   write to the end of the file. A "write" to a
1108 			 *   position before the end of the file would not
1109 			 *   change the size of the file but it would allocate
1110 			 *   new disk blocks.
1111 			 * - uiomove intended to overwrite the whole page.
1112 			 * - a new page was created (newpage == 1).
1113 			 */
1114 
1115 			if (i_size_changed == 0 && new_iblocks == 0 &&
1116 			    newpage) {
1117 
1118 				/* unwind what uiomove eventually last did */
1119 				uio->uio_resid = premove_resid;
1120 
1121 				/*
1122 				 * destroy the page, do not write ambiguous
1123 				 * data to the disk.
1124 				 */
1125 				flags = SM_DESTROY;
1126 			} else {
1127 				/*
1128 				 * write the page back to the disk, if dirty,
1129 				 * and remove the page from the cache.
1130 				 */
1131 				flags = SM_INVAL;
1132 			}
1133 			(void) segmap_release(segkmap, base, flags);
1134 		} else {
1135 			flags = 0;
1136 			/*
1137 			 * Force write back for synchronous write cases.
1138 			 */
1139 			if ((ioflag & (FSYNC|FDSYNC)) || type == IFDIR) {
1140 				/*
1141 				 * If the sticky bit is set but the
1142 				 * execute bit is not set, we do a
1143 				 * synchronous write back and free
1144 				 * the page when done.  We set up swap
1145 				 * files to be handled this way to
1146 				 * prevent servers from keeping around
1147 				 * the client's swap pages too long.
1148 				 * XXX - there ought to be a better way.
1149 				 */
1150 				if (IS_SWAPVP(vp)) {
1151 					flags = SM_WRITE | SM_FREE |
1152 					    SM_DONTNEED;
1153 					iupdat_flag = 0;
1154 				} else {
1155 					flags = SM_WRITE;
1156 				}
1157 			} else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
1158 				/*
1159 				 * Have written a whole block.
1160 				 * Start an asynchronous write and
1161 				 * mark the buffer to indicate that
1162 				 * it won't be needed again soon.
1163 				 */
1164 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
1165 			}
1166 			error = segmap_release(segkmap, base, flags);
1167 			/*
1168 			 * If the operation failed and is synchronous,
1169 			 * then we need to unwind what uiomove() last
1170 			 * did so we can potentially return an error to
1171 			 * the caller.  If this write operation was
1172 			 * done in two pieces and the first succeeded,
1173 			 * then we won't return an error for the second
1174 			 * piece that failed.  However, we only want to
1175 			 * return a resid value that reflects what was
1176 			 * really done.
1177 			 *
1178 			 * Failures for non-synchronous operations can
1179 			 * be ignored since the page subsystem will
1180 			 * retry the operation until it succeeds or the
1181 			 * file system is unmounted.
1182 			 */
1183 			if (error) {
1184 				if ((ioflag & (FSYNC | FDSYNC)) ||
1185 				    type == IFDIR) {
1186 					uio->uio_resid = premove_resid;
1187 				} else {
1188 					error = 0;
1189 				}
1190 			}
1191 		}
1192 
1193 		/*
1194 		 * Re-acquire contents lock.
1195 		 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1196 		 */
1197 		if (do_dqrwlock)
1198 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1199 		rw_enter(&ip->i_contents, RW_WRITER);
1200 
1201 		/*
1202 		 * If the uiomove() failed or if a synchronous
1203 		 * page push failed, fix up i_size.
1204 		 */
1205 		if (error) {
1206 			if (i_size_changed) {
1207 				/*
1208 				 * The uiomove failed, and we
1209 				 * allocated blocks,so get rid
1210 				 * of them.
1211 				 */
1212 				(void) ufs_itrunc(ip, old_i_size, 0, cr);
1213 			}
1214 		} else {
1215 			/*
1216 			 * XXX - Can this be out of the loop?
1217 			 */
1218 			ip->i_flag |= IUPD | ICHG;
1219 			/*
1220 			 * Only do one increase of i_seq for multiple
1221 			 * pieces.  Because we drop locks, record
1222 			 * the fact that we changed the timestamp and
1223 			 * are deferring the increase in case another thread
1224 			 * pushes our timestamp update.
1225 			 */
1226 			i_seq_needed = 1;
1227 			ip->i_flag |= ISEQ;
1228 			if (i_size_changed)
1229 				ip->i_flag |= IATTCHG;
1230 			if ((ip->i_mode & (IEXEC | (IEXEC >> 3) |
1231 			    (IEXEC >> 6))) != 0 &&
1232 			    (ip->i_mode & (ISUID | ISGID)) != 0 &&
1233 			    secpolicy_vnode_setid_retain(cr,
1234 			    (ip->i_mode & ISUID) != 0 && ip->i_uid == 0) != 0) {
1235 				/*
1236 				 * Clear Set-UID & Set-GID bits on
1237 				 * successful write if not privileged
1238 				 * and at least one of the execute bits
1239 				 * is set.  If we always clear Set-GID,
1240 				 * mandatory file and record locking is
1241 				 * unuseable.
1242 				 */
1243 				ip->i_mode &= ~(ISUID | ISGID);
1244 			}
1245 		}
1246 		/*
1247 		 * In the case the FDSYNC flag is set and this is a
1248 		 * "rewrite" we won't log a delta.
1249 		 * The FSYNC flag overrides all cases.
1250 		 */
1251 		if (!ufs_check_rewrite(ip, uio, ioflag) || !(ioflag & FDSYNC)) {
1252 			TRANS_INODE(ufsvfsp, ip);
1253 		}
1254 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1255 
1256 out:
1257 	/*
1258 	 * Make sure i_seq is increased at least once per write
1259 	 */
1260 	if (i_seq_needed) {
1261 		ip->i_seq++;
1262 		ip->i_flag &= ~ISEQ;	/* no longer deferred */
1263 	}
1264 
1265 	/*
1266 	 * Inode is updated according to this table -
1267 	 *
1268 	 *   FSYNC	  FDSYNC(posix.4)
1269 	 *   --------------------------
1270 	 *   always@	  IATTCHG|IBDWRITE
1271 	 *
1272 	 * @ - 	If we are doing synchronous write the only time we should
1273 	 *	not be sync'ing the ip here is if we have the stickyhack
1274 	 *	activated, the file is marked with the sticky bit and
1275 	 *	no exec bit, the file length has not been changed and
1276 	 *	no new blocks have been allocated during this write.
1277 	 */
1278 
1279 	if ((ip->i_flag & ISYNC) != 0) {
1280 		/*
1281 		 * we have eliminated nosync
1282 		 */
1283 		if ((ip->i_flag & (IATTCHG|IBDWRITE)) ||
1284 			((ioflag & FSYNC) && iupdat_flag)) {
1285 			ufs_iupdat(ip, 1);
1286 		}
1287 	}
1288 
1289 	/*
1290 	 * If we've already done a partial-write, terminate
1291 	 * the write but return no error unless the error is ENOSPC
1292 	 * because the caller can detect this and free resources and
1293 	 * try again.
1294 	 */
1295 	if ((start_resid != uio->uio_resid) && (error != ENOSPC))
1296 		error = 0;
1297 
1298 	ip->i_flag &= ~(INOACC | ISYNC);
1299 	ITIMES_NOLOCK(ip);
1300 	TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1301 		"ufs_wrip_end:vp %p error %d", vp, error);
1302 	return (error);
1303 }
1304 
1305 /*
1306  * rdip does the real work of read requests for ufs.
1307  */
1308 int
1309 rdip(struct inode *ip, struct uio *uio, int ioflag, cred_t *cr)
1310 {
1311 	u_offset_t off;
1312 	caddr_t base;
1313 	struct fs *fs;
1314 	struct ufsvfs *ufsvfsp;
1315 	struct vnode *vp;
1316 	long oresid = uio->uio_resid;
1317 	u_offset_t n, on, mapon;
1318 	int error = 0;
1319 	int doupdate = 1;
1320 	uint_t flags;
1321 	int dofree, directio_status;
1322 	krw_t rwtype;
1323 	o_mode_t type;
1324 
1325 	vp = ITOV(ip);
1326 
1327 	TRACE_1(TR_FAC_UFS, TR_UFS_RWIP_START,
1328 		"ufs_rdip_start:vp %p", vp);
1329 
1330 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
1331 
1332 	ufsvfsp = ip->i_ufsvfs;
1333 
1334 	if (ufsvfsp == NULL)
1335 		return (EIO);
1336 
1337 	fs = ufsvfsp->vfs_fs;
1338 
1339 	/* check for valid filetype */
1340 	type = ip->i_mode & IFMT;
1341 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
1342 	    (type != IFLNK) && (type != IFSHAD)) {
1343 		return (EIO);
1344 	}
1345 
1346 	if (uio->uio_loffset > UFS_MAXOFFSET_T) {
1347 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1348 			"ufs_rdip_end:vp %p error %d", vp, EINVAL);
1349 		error = 0;
1350 		goto out;
1351 	}
1352 	if (uio->uio_loffset < (offset_t)0) {
1353 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1354 			"ufs_rdip_end:vp %p error %d", vp, EINVAL);
1355 		return (EINVAL);
1356 	}
1357 	if (uio->uio_resid == 0) {
1358 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1359 			"ufs_rdip_end:vp %p error %d", vp, 0);
1360 		return (0);
1361 	}
1362 
1363 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (fs->fs_ronly == 0) &&
1364 		(!ufsvfsp->vfs_noatime)) {
1365 		mutex_enter(&ip->i_tlock);
1366 		ip->i_flag |= IACC;
1367 		mutex_exit(&ip->i_tlock);
1368 	}
1369 	/*
1370 	 * Try to go direct
1371 	 */
1372 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
1373 		error = ufs_directio_read(ip, uio, cr, &directio_status);
1374 		if (directio_status == DIRECTIO_SUCCESS)
1375 			goto out;
1376 	}
1377 
1378 	rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER);
1379 
1380 	do {
1381 		offset_t diff;
1382 		u_offset_t uoff = uio->uio_loffset;
1383 		off = uoff & (offset_t)MAXBMASK;
1384 		mapon = (u_offset_t)(uoff & (offset_t)MAXBOFFSET);
1385 		on = (u_offset_t)blkoff(fs, uoff);
1386 		n = MIN((u_offset_t)fs->fs_bsize - on,
1387 			(u_offset_t)uio->uio_resid);
1388 
1389 		diff = ip->i_size - uoff;
1390 
1391 		if (diff <= (offset_t)0) {
1392 			error = 0;
1393 			goto out;
1394 		}
1395 		if (diff < (offset_t)n)
1396 			n = (int)diff;
1397 
1398 		/*
1399 		 * We update smallfile2 and smallfile1 at most every second.
1400 		 */
1401 		if (lbolt >= smallfile_update) {
1402 			uint64_t percpufreeb;
1403 			if (smallfile1_d == 0) smallfile1_d = SMALLFILE1_D;
1404 			if (smallfile2_d == 0) smallfile2_d = SMALLFILE2_D;
1405 			percpufreeb = ptob((uint64_t)freemem) / ncpus_online;
1406 			smallfile1 = percpufreeb / smallfile1_d;
1407 			smallfile2 = percpufreeb / smallfile2_d;
1408 			smallfile1 = MAX(smallfile1, smallfile);
1409 			smallfile1 = MAX(smallfile1, smallfile64);
1410 			smallfile2 = MAX(smallfile1, smallfile2);
1411 			smallfile_update = lbolt + hz;
1412 		}
1413 
1414 		dofree = freebehind &&
1415 		    ip->i_nextr == (off & PAGEMASK) && off > smallfile1;
1416 
1417 		/*
1418 		 * At this point we can enter ufs_getpage() in one of two
1419 		 * ways:
1420 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
1421 		 *    forcefault parameter is true (value of 1 is passed)
1422 		 * 2) uiomove() causes a page fault.
1423 		 *
1424 		 * We cannot hold onto an i_contents reader lock without
1425 		 * risking deadlock in ufs_getpage() so drop a reader lock.
1426 		 * The ufs_getpage() dolock logic already allows for a
1427 		 * thread holding i_contents as writer to work properly
1428 		 * so we keep a writer lock.
1429 		 */
1430 		if (rwtype == RW_READER)
1431 			rw_exit(&ip->i_contents);
1432 		base = segmap_getmapflt(segkmap, vp, (off + mapon),
1433 					(uint_t)n, 1, S_READ);
1434 
1435 		error = uiomove(base + mapon, (long)n, UIO_READ, uio);
1436 
1437 		flags = 0;
1438 		if (!error) {
1439 			/*
1440 			 * If  reading sequential  we won't need  this
1441 			 * buffer again  soon.  For  offsets in  range
1442 			 * [smallfile1,  smallfile2] release the pages
1443 			 * at   the  tail  of the   cache list, larger
1444 			 * offsets are released at the head.
1445 			 */
1446 			if (dofree) {
1447 				flags = SM_FREE | SM_ASYNC;
1448 				if ((cache_read_ahead == 0) &&
1449 					(off > smallfile2))
1450 					flags |=  SM_DONTNEED;
1451 			}
1452 			/*
1453 			 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1454 			 * we want to make sure that the page which has
1455 			 * been read, is written on disk if it is dirty.
1456 			 * And corresponding indirect blocks should also
1457 			 * be flushed out.
1458 			 */
1459 			if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) {
1460 				flags &= ~SM_ASYNC;
1461 				flags |= SM_WRITE;
1462 			}
1463 			error = segmap_release(segkmap, base, flags);
1464 		} else
1465 			(void) segmap_release(segkmap, base, flags);
1466 
1467 		if (rwtype == RW_READER)
1468 			rw_enter(&ip->i_contents, rwtype);
1469 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1470 out:
1471 	/*
1472 	 * Inode is updated according to this table if FRSYNC is set.
1473 	 *
1474 	 *   FSYNC	  FDSYNC(posix.4)
1475 	 *   --------------------------
1476 	 *   always	  IATTCHG|IBDWRITE
1477 	 */
1478 	/*
1479 	 * The inode is not updated if we're logging and the inode is a
1480 	 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1481 	 */
1482 	if (ioflag & FRSYNC) {
1483 		if (TRANS_ISTRANS(ufsvfsp) && ((ip->i_mode & IFMT) == IFDIR)) {
1484 				doupdate = 0;
1485 		}
1486 		if (doupdate) {
1487 			if ((ioflag & FSYNC) ||
1488 			    ((ioflag & FDSYNC) &&
1489 			    (ip->i_flag & (IATTCHG|IBDWRITE)))) {
1490 				ufs_iupdat(ip, 1);
1491 			}
1492 		}
1493 	}
1494 	/*
1495 	 * If we've already done a partial read, terminate
1496 	 * the read but return no error.
1497 	 */
1498 	if (oresid != uio->uio_resid)
1499 		error = 0;
1500 	ITIMES(ip);
1501 
1502 	TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1503 		"ufs_rdip_end:vp %p error %d", vp, error);
1504 	return (error);
1505 }
1506 
1507 /* ARGSUSED */
1508 static int
1509 ufs_ioctl(
1510 	struct vnode	*vp,
1511 	int		cmd,
1512 	intptr_t	arg,
1513 	int		flag,
1514 	struct cred	*cr,
1515 	int		*rvalp)
1516 {
1517 	struct lockfs	lockfs, lockfs_out;
1518 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
1519 	char		*comment, *original_comment;
1520 	struct fs	*fs;
1521 	struct ulockfs	*ulp;
1522 	offset_t	off;
1523 	extern int	maxphys;
1524 	int		error;
1525 	int		issync;
1526 	int		trans_size;
1527 
1528 
1529 	/*
1530 	 * forcibly unmounted
1531 	 */
1532 	if (ufsvfsp == NULL) {
1533 		return (EIO);
1534 	}
1535 
1536 	fs = ufsvfsp->vfs_fs;
1537 
1538 	if (cmd == Q_QUOTACTL) {
1539 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_QUOTA_MASK);
1540 		if (error)
1541 			return (error);
1542 
1543 		if (ulp) {
1544 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_QUOTA,
1545 						TOP_SETQUOTA_SIZE(fs));
1546 		}
1547 
1548 		error = quotactl(vp, arg, flag, cr);
1549 
1550 		if (ulp) {
1551 			TRANS_END_ASYNC(ufsvfsp, TOP_QUOTA,
1552 						TOP_SETQUOTA_SIZE(fs));
1553 			ufs_lockfs_end(ulp);
1554 		}
1555 		return (error);
1556 	}
1557 
1558 	switch (cmd) {
1559 		case _FIOLFS:
1560 			/*
1561 			 * file system locking
1562 			 */
1563 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1564 				return (EPERM);
1565 
1566 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1567 				if (copyin((caddr_t)arg, &lockfs,
1568 						sizeof (struct lockfs)))
1569 					return (EFAULT);
1570 			}
1571 #ifdef _SYSCALL32_IMPL
1572 			else {
1573 				struct lockfs32	lockfs32;
1574 				/* Translate ILP32 lockfs to LP64 lockfs */
1575 				if (copyin((caddr_t)arg, &lockfs32,
1576 				    sizeof (struct lockfs32)))
1577 					return (EFAULT);
1578 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1579 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1580 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1581 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1582 				lockfs.lf_comment =
1583 					(caddr_t)(uintptr_t)lockfs32.lf_comment;
1584 			}
1585 #endif /* _SYSCALL32_IMPL */
1586 
1587 			if (lockfs.lf_comlen) {
1588 				if (lockfs.lf_comlen > LOCKFS_MAXCOMMENTLEN)
1589 					return (ENAMETOOLONG);
1590 				comment = kmem_alloc(lockfs.lf_comlen,
1591 						KM_SLEEP);
1592 				if (copyin(lockfs.lf_comment, comment,
1593 					lockfs.lf_comlen)) {
1594 					kmem_free(comment, lockfs.lf_comlen);
1595 					return (EFAULT);
1596 				}
1597 				original_comment = lockfs.lf_comment;
1598 				lockfs.lf_comment = comment;
1599 			}
1600 			if ((error = ufs_fiolfs(vp, &lockfs, 0)) == 0) {
1601 				lockfs.lf_comment = original_comment;
1602 
1603 				if ((flag & DATAMODEL_MASK) ==
1604 				    DATAMODEL_NATIVE) {
1605 					(void) copyout(&lockfs, (caddr_t)arg,
1606 					    sizeof (struct lockfs));
1607 				}
1608 #ifdef _SYSCALL32_IMPL
1609 				else {
1610 					struct lockfs32	lockfs32;
1611 					/* Translate LP64 to ILP32 lockfs */
1612 					lockfs32.lf_lock =
1613 					    (uint32_t)lockfs.lf_lock;
1614 					lockfs32.lf_flags =
1615 					    (uint32_t)lockfs.lf_flags;
1616 					lockfs32.lf_key =
1617 					    (uint32_t)lockfs.lf_key;
1618 					lockfs32.lf_comlen =
1619 					    (uint32_t)lockfs.lf_comlen;
1620 					lockfs32.lf_comment =
1621 					(uint32_t)(uintptr_t)lockfs.lf_comment;
1622 					(void) copyout(&lockfs32, (caddr_t)arg,
1623 					    sizeof (struct lockfs32));
1624 				}
1625 #endif /* _SYSCALL32_IMPL */
1626 
1627 			} else {
1628 				if (lockfs.lf_comlen)
1629 					kmem_free(comment, lockfs.lf_comlen);
1630 			}
1631 			return (error);
1632 
1633 		case _FIOLFSS:
1634 			/*
1635 			 * get file system locking status
1636 			 */
1637 
1638 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1639 				if (copyin((caddr_t)arg, &lockfs,
1640 						sizeof (struct lockfs)))
1641 					return (EFAULT);
1642 			}
1643 #ifdef _SYSCALL32_IMPL
1644 			else {
1645 				struct lockfs32	lockfs32;
1646 				/* Translate ILP32 lockfs to LP64 lockfs */
1647 				if (copyin((caddr_t)arg, &lockfs32,
1648 						sizeof (struct lockfs32)))
1649 					return (EFAULT);
1650 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1651 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1652 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1653 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1654 				lockfs.lf_comment =
1655 					(caddr_t)(uintptr_t)lockfs32.lf_comment;
1656 			}
1657 #endif /* _SYSCALL32_IMPL */
1658 
1659 			if (error =  ufs_fiolfss(vp, &lockfs_out))
1660 				return (error);
1661 			lockfs.lf_lock = lockfs_out.lf_lock;
1662 			lockfs.lf_key = lockfs_out.lf_key;
1663 			lockfs.lf_flags = lockfs_out.lf_flags;
1664 			lockfs.lf_comlen = MIN(lockfs.lf_comlen,
1665 				lockfs_out.lf_comlen);
1666 
1667 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1668 				if (copyout(&lockfs, (caddr_t)arg,
1669 						sizeof (struct lockfs)))
1670 					return (EFAULT);
1671 			}
1672 #ifdef _SYSCALL32_IMPL
1673 			else {
1674 				/* Translate LP64 to ILP32 lockfs */
1675 				struct lockfs32	lockfs32;
1676 				lockfs32.lf_lock = (uint32_t)lockfs.lf_lock;
1677 				lockfs32.lf_flags = (uint32_t)lockfs.lf_flags;
1678 				lockfs32.lf_key = (uint32_t)lockfs.lf_key;
1679 				lockfs32.lf_comlen = (uint32_t)lockfs.lf_comlen;
1680 				lockfs32.lf_comment =
1681 					(uint32_t)(uintptr_t)lockfs.lf_comment;
1682 				if (copyout(&lockfs32, (caddr_t)arg,
1683 					    sizeof (struct lockfs32)))
1684 					return (EFAULT);
1685 			}
1686 #endif /* _SYSCALL32_IMPL */
1687 
1688 			if (lockfs.lf_comlen &&
1689 			    lockfs.lf_comment && lockfs_out.lf_comment)
1690 				if (copyout(lockfs_out.lf_comment,
1691 					lockfs.lf_comment,
1692 					lockfs.lf_comlen))
1693 					return (EFAULT);
1694 			return (0);
1695 
1696 		case _FIOSATIME:
1697 			/*
1698 			 * set access time
1699 			 */
1700 
1701 			/*
1702 			 * if mounted w/o atime, return quietly.
1703 			 * I briefly thought about returning ENOSYS, but
1704 			 * figured that most apps would consider this fatal
1705 			 * but the idea is to make this as seamless as poss.
1706 			 */
1707 			if (ufsvfsp->vfs_noatime)
1708 				return (0);
1709 
1710 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1711 					ULOCKFS_SETATTR_MASK);
1712 			if (error)
1713 				return (error);
1714 
1715 			if (ulp) {
1716 				trans_size = (int)TOP_SETATTR_SIZE(VTOI(vp));
1717 				TRANS_BEGIN_CSYNC(ufsvfsp, issync,
1718 						TOP_SETATTR, trans_size);
1719 			}
1720 
1721 			error = ufs_fiosatime(vp, (struct timeval *)arg,
1722 					flag, cr);
1723 
1724 			if (ulp) {
1725 				TRANS_END_CSYNC(ufsvfsp, error, issync,
1726 						TOP_SETATTR, trans_size);
1727 				ufs_lockfs_end(ulp);
1728 			}
1729 			return (error);
1730 
1731 		case _FIOSDIO:
1732 			/*
1733 			 * set delayed-io
1734 			 */
1735 			return (ufs_fiosdio(vp, (uint_t *)arg, flag, cr));
1736 
1737 		case _FIOGDIO:
1738 			/*
1739 			 * get delayed-io
1740 			 */
1741 			return (ufs_fiogdio(vp, (uint_t *)arg, flag, cr));
1742 
1743 		case _FIOIO:
1744 			/*
1745 			 * inode open
1746 			 */
1747 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1748 					ULOCKFS_VGET_MASK);
1749 			if (error)
1750 				return (error);
1751 
1752 			error = ufs_fioio(vp, (struct fioio *)arg, flag, cr);
1753 
1754 			if (ulp) {
1755 				ufs_lockfs_end(ulp);
1756 			}
1757 			return (error);
1758 
1759 		case _FIOFFS:
1760 			/*
1761 			 * file system flush (push w/invalidate)
1762 			 */
1763 			if ((caddr_t)arg != NULL)
1764 				return (EINVAL);
1765 			return (ufs_fioffs(vp, NULL, cr));
1766 
1767 		case _FIOISBUSY:
1768 			/*
1769 			 * Contract-private interface for Legato
1770 			 * Purge this vnode from the DNLC and decide
1771 			 * if this vnode is busy (*arg == 1) or not
1772 			 * (*arg == 0)
1773 			 */
1774 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1775 				return (EPERM);
1776 			error = ufs_fioisbusy(vp, (int *)arg, cr);
1777 			return (error);
1778 
1779 		case _FIODIRECTIO:
1780 			return (ufs_fiodirectio(vp, (int)arg, cr));
1781 
1782 		case _FIOTUNE:
1783 			/*
1784 			 * Tune the file system (aka setting fs attributes)
1785 			 */
1786 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1787 					ULOCKFS_SETATTR_MASK);
1788 			if (error)
1789 				return (error);
1790 
1791 			error = ufs_fiotune(vp, (struct fiotune *)arg, cr);
1792 
1793 			if (ulp)
1794 				ufs_lockfs_end(ulp);
1795 			return (error);
1796 
1797 		case _FIOLOGENABLE:
1798 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1799 				return (EPERM);
1800 			return (ufs_fiologenable(vp, (void *)arg, cr, flag));
1801 
1802 		case _FIOLOGDISABLE:
1803 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1804 				return (EPERM);
1805 			return (ufs_fiologdisable(vp, (void *)arg, cr, flag));
1806 
1807 		case _FIOISLOG:
1808 			return (ufs_fioislog(vp, (void *)arg, cr, flag));
1809 
1810 		case _FIOSNAPSHOTCREATE_MULTI:
1811 		{
1812 			struct fiosnapcreate_multi	fc, *fcp;
1813 			size_t	fcm_size;
1814 
1815 			if (copyin((void *)arg, &fc, sizeof (fc)))
1816 				return (EFAULT);
1817 			if (fc.backfilecount > MAX_BACKFILE_COUNT)
1818 				return (EINVAL);
1819 			fcm_size = sizeof (struct fiosnapcreate_multi) +
1820 			    (fc.backfilecount - 1) * sizeof (int);
1821 			fcp = (struct fiosnapcreate_multi *)
1822 			    kmem_alloc(fcm_size, KM_SLEEP);
1823 			if (copyin((void *)arg, fcp, fcm_size)) {
1824 				kmem_free(fcp, fcm_size);
1825 				return (EFAULT);
1826 			}
1827 			error = ufs_snap_create(vp, fcp, cr);
1828 			if (!error && copyout(fcp, (void *)arg, fcm_size))
1829 				error = EFAULT;
1830 			kmem_free(fcp, fcm_size);
1831 			return (error);
1832 		}
1833 
1834 		case _FIOSNAPSHOTDELETE:
1835 		{
1836 			struct fiosnapdelete	fc;
1837 
1838 			if (copyin((void *)arg, &fc, sizeof (fc)))
1839 				return (EFAULT);
1840 			error = ufs_snap_delete(vp, &fc, cr);
1841 			if (!error && copyout(&fc, (void *)arg, sizeof (fc)))
1842 				error = EFAULT;
1843 			return (error);
1844 		}
1845 
1846 		case _FIOGETSUPERBLOCK:
1847 			if (copyout(fs, (void *)arg, SBSIZE))
1848 				return (EFAULT);
1849 			return (0);
1850 
1851 		case _FIOGETMAXPHYS:
1852 			if (copyout(&maxphys, (void *)arg, sizeof (maxphys)))
1853 				return (EFAULT);
1854 			return (0);
1855 
1856 		/*
1857 		 * The following 3 ioctls are for TSufs support
1858 		 * although could potentially be used elsewhere
1859 		 */
1860 		case _FIO_SET_LUFS_DEBUG:
1861 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1862 				return (EPERM);
1863 			lufs_debug = (uint32_t)arg;
1864 			return (0);
1865 
1866 		case _FIO_SET_LUFS_ERROR:
1867 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1868 				return (EPERM);
1869 			TRANS_SETERROR(ufsvfsp);
1870 			return (0);
1871 
1872 		case _FIO_GET_TOP_STATS:
1873 		{
1874 			fio_lufs_stats_t *ls;
1875 			ml_unit_t *ul = ufsvfsp->vfs_log;
1876 
1877 			ls = kmem_zalloc(sizeof (*ls), KM_SLEEP);
1878 			ls->ls_debug = ul->un_debug; /* return debug value */
1879 			/* Copy stucture if statistics are being kept */
1880 			if (ul->un_logmap->mtm_tops) {
1881 				ls->ls_topstats = *(ul->un_logmap->mtm_tops);
1882 			}
1883 			error = 0;
1884 			if (copyout(ls, (void *)arg, sizeof (*ls)))
1885 				error = EFAULT;
1886 			kmem_free(ls, sizeof (*ls));
1887 			return (error);
1888 		}
1889 
1890 		case _FIO_SEEK_DATA:
1891 		case _FIO_SEEK_HOLE:
1892 			if (ddi_copyin((void *)arg, &off, sizeof (off), flag))
1893 				return (EFAULT);
1894 			/* offset paramater is in/out */
1895 			error = ufs_fio_holey(vp, cmd, &off);
1896 			if (error)
1897 				return (error);
1898 			if (ddi_copyout(&off, (void *)arg, sizeof (off), flag))
1899 				return (EFAULT);
1900 			return (0);
1901 
1902 		default:
1903 			return (ENOTTY);
1904 	}
1905 }
1906 
1907 /* ARGSUSED */
1908 static int
1909 ufs_getattr(struct vnode *vp, struct vattr *vap, int flags,
1910 	struct cred *cr)
1911 {
1912 	struct inode *ip = VTOI(vp);
1913 	struct ufsvfs *ufsvfsp;
1914 	int err;
1915 
1916 	TRACE_2(TR_FAC_UFS, TR_UFS_GETATTR_START,
1917 		"ufs_getattr_start:vp %p flags %x", vp, flags);
1918 
1919 	if (vap->va_mask == AT_SIZE) {
1920 		/*
1921 		 * for performance, if only the size is requested don't bother
1922 		 * with anything else.
1923 		 */
1924 		UFS_GET_ISIZE(&vap->va_size, ip);
1925 		TRACE_1(TR_FAC_UFS, TR_UFS_GETATTR_END,
1926 			"ufs_getattr_end:vp %p", vp);
1927 		return (0);
1928 	}
1929 
1930 	/*
1931 	 * inlined lockfs checks
1932 	 */
1933 	ufsvfsp = ip->i_ufsvfs;
1934 	if ((ufsvfsp == NULL) || ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs)) {
1935 		err = EIO;
1936 		goto out;
1937 	}
1938 
1939 	rw_enter(&ip->i_contents, RW_READER);
1940 	/*
1941 	 * Return all the attributes.  This should be refined so
1942 	 * that it only returns what's asked for.
1943 	 */
1944 
1945 	/*
1946 	 * Copy from inode table.
1947 	 */
1948 	vap->va_type = vp->v_type;
1949 	vap->va_mode = ip->i_mode & MODEMASK;
1950 	/*
1951 	 * If there is an ACL and there is a mask entry, then do the
1952 	 * extra work that completes the equivalent of an acltomode(3)
1953 	 * call.  According to POSIX P1003.1e, the acl mask should be
1954 	 * returned in the group permissions field.
1955 	 *
1956 	 * - start with the original permission and mode bits (from above)
1957 	 * - clear the group owner bits
1958 	 * - add in the mask bits.
1959 	 */
1960 	if (ip->i_ufs_acl && ip->i_ufs_acl->aclass.acl_ismask) {
1961 		vap->va_mode &= ~((VREAD | VWRITE | VEXEC) >> 3);
1962 		vap->va_mode |=
1963 		    (ip->i_ufs_acl->aclass.acl_maskbits & PERMMASK) << 3;
1964 	}
1965 	vap->va_uid = ip->i_uid;
1966 	vap->va_gid = ip->i_gid;
1967 	vap->va_fsid = ip->i_dev;
1968 	vap->va_nodeid = (ino64_t)ip->i_number;
1969 	vap->va_nlink = ip->i_nlink;
1970 	vap->va_size = ip->i_size;
1971 	if (vp->v_type == VCHR || vp->v_type == VBLK)
1972 		vap->va_rdev = ip->i_rdev;
1973 	else
1974 		vap->va_rdev = 0;	/* not a b/c spec. */
1975 	mutex_enter(&ip->i_tlock);
1976 	ITIMES_NOLOCK(ip);	/* mark correct time in inode */
1977 	vap->va_seq = ip->i_seq;
1978 	vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec;
1979 	vap->va_atime.tv_nsec = ip->i_atime.tv_usec*1000;
1980 	vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec;
1981 	vap->va_mtime.tv_nsec = ip->i_mtime.tv_usec*1000;
1982 	vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec;
1983 	vap->va_ctime.tv_nsec = ip->i_ctime.tv_usec*1000;
1984 	mutex_exit(&ip->i_tlock);
1985 
1986 	switch (ip->i_mode & IFMT) {
1987 
1988 	case IFBLK:
1989 		vap->va_blksize = MAXBSIZE;		/* was BLKDEV_IOSIZE */
1990 		break;
1991 
1992 	case IFCHR:
1993 		vap->va_blksize = MAXBSIZE;
1994 		break;
1995 
1996 	default:
1997 		vap->va_blksize = ip->i_fs->fs_bsize;
1998 		break;
1999 	}
2000 	vap->va_nblocks = (fsblkcnt64_t)ip->i_blocks;
2001 	rw_exit(&ip->i_contents);
2002 	err = 0;
2003 
2004 out:
2005 	TRACE_1(TR_FAC_UFS, TR_UFS_GETATTR_END, "ufs_getattr_end:vp %p", vp);
2006 
2007 	return (err);
2008 }
2009 
2010 /*ARGSUSED4*/
2011 static int
2012 ufs_setattr(
2013 	struct vnode *vp,
2014 	struct vattr *vap,
2015 	int flags,
2016 	struct cred *cr,
2017 	caller_context_t *ct)
2018 {
2019 	struct inode *ip = VTOI(vp);
2020 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2021 	struct fs *fs;
2022 	struct ulockfs *ulp;
2023 	char *errmsg1;
2024 	char *errmsg2;
2025 	long blocks;
2026 	long int mask = vap->va_mask;
2027 	size_t len1, len2;
2028 	int issync;
2029 	int trans_size;
2030 	int dotrans;
2031 	int dorwlock;
2032 	int error;
2033 	int owner_change;
2034 	int dodqlock;
2035 	timestruc_t now;
2036 	vattr_t oldva;
2037 	int retry = 1;
2038 
2039 	TRACE_2(TR_FAC_UFS, TR_UFS_SETATTR_START,
2040 		"ufs_setattr_start:vp %p flags %x", vp, flags);
2041 
2042 	/*
2043 	 * Cannot set these attributes.
2044 	 */
2045 	if (mask & AT_NOSET) {
2046 		error = EINVAL;
2047 		goto out;
2048 	}
2049 
2050 	/*
2051 	 * check for forced unmount
2052 	 */
2053 	if (ufsvfsp == NULL)
2054 		return (EIO);
2055 
2056 	fs = ufsvfsp->vfs_fs;
2057 	if (fs->fs_ronly != 0)
2058 		return (EROFS);
2059 
2060 again:
2061 	errmsg1 = NULL;
2062 	errmsg2 = NULL;
2063 	dotrans = 0;
2064 	dorwlock = 0;
2065 	dodqlock = 0;
2066 
2067 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK);
2068 	if (error)
2069 		goto out;
2070 
2071 	/*
2072 	 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2073 	 * This follows the protocol for read()/write().
2074 	 */
2075 	if (vp->v_type != VDIR) {
2076 		rw_enter(&ip->i_rwlock, RW_WRITER);
2077 		dorwlock = 1;
2078 	}
2079 
2080 	/*
2081 	 * Truncate file.  Must have write permission and not be a directory.
2082 	 */
2083 	if (mask & AT_SIZE) {
2084 		rw_enter(&ip->i_contents, RW_WRITER);
2085 		if (vp->v_type == VDIR) {
2086 			error = EISDIR;
2087 			goto update_inode;
2088 		}
2089 		if (error = ufs_iaccess(ip, IWRITE, cr))
2090 			goto update_inode;
2091 
2092 		rw_exit(&ip->i_contents);
2093 		error = TRANS_ITRUNC(ip, vap->va_size, 0, cr);
2094 		if (error) {
2095 			rw_enter(&ip->i_contents, RW_WRITER);
2096 			goto update_inode;
2097 		}
2098 	}
2099 
2100 	if (ulp) {
2101 		trans_size = (int)TOP_SETATTR_SIZE(ip);
2102 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size);
2103 		++dotrans;
2104 	}
2105 
2106 	/*
2107 	 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2108 	 * This follows the protocol established by
2109 	 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2110 	 */
2111 	if (vp->v_type == VDIR) {
2112 		rw_enter(&ip->i_rwlock, RW_WRITER);
2113 		dorwlock = 1;
2114 	}
2115 
2116 	/*
2117 	 * Grab quota lock if we are changing the file's owner.
2118 	 */
2119 	if (mask & AT_UID) {
2120 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2121 		dodqlock = 1;
2122 	}
2123 	rw_enter(&ip->i_contents, RW_WRITER);
2124 
2125 	oldva.va_mode = ip->i_mode;
2126 	oldva.va_uid = ip->i_uid;
2127 	oldva.va_gid = ip->i_gid;
2128 
2129 	vap->va_mask &= ~AT_SIZE;
2130 	/*
2131 	 * ufs_iaccess is "close enough"; that's because it doesn't
2132 	 * map the defines.
2133 	 */
2134 	error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2135 				ufs_iaccess, ip);
2136 	if (error)
2137 		goto update_inode;
2138 
2139 	mask = vap->va_mask;
2140 
2141 	/*
2142 	 * Change file access modes.
2143 	 */
2144 	if (mask & AT_MODE) {
2145 		ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT);
2146 		TRANS_INODE(ufsvfsp, ip);
2147 		ip->i_flag |= ICHG;
2148 		if (stickyhack) {
2149 			mutex_enter(&vp->v_lock);
2150 			if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX)
2151 				vp->v_flag |= VSWAPLIKE;
2152 			else
2153 				vp->v_flag &= ~VSWAPLIKE;
2154 			mutex_exit(&vp->v_lock);
2155 		}
2156 	}
2157 	if (mask & (AT_UID|AT_GID)) {
2158 		if (mask & AT_UID) {
2159 			/*
2160 			 * Don't change ownership of the quota inode.
2161 			 */
2162 			if (ufsvfsp->vfs_qinod == ip) {
2163 				ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED);
2164 				error = EINVAL;
2165 				goto update_inode;
2166 			}
2167 
2168 			/*
2169 			 * No real ownership change.
2170 			 */
2171 			if (ip->i_uid == vap->va_uid) {
2172 				blocks = 0;
2173 				owner_change = 0;
2174 			}
2175 			/*
2176 			 * Remove the blocks and the file, from the old user's
2177 			 * quota.
2178 			 */
2179 			else {
2180 				blocks = ip->i_blocks;
2181 				owner_change = 1;
2182 
2183 				(void) chkdq(ip, -blocks, /* force */ 1, cr,
2184 						(char **)NULL, (size_t *)NULL);
2185 				(void) chkiq(ufsvfsp, /* change */ -1, ip,
2186 						(uid_t)ip->i_uid,
2187 						/* force */ 1, cr,
2188 						(char **)NULL, (size_t *)NULL);
2189 				dqrele(ip->i_dquot);
2190 			}
2191 
2192 			ip->i_uid = vap->va_uid;
2193 
2194 			/*
2195 			 * There is a real ownership change.
2196 			 */
2197 			if (owner_change) {
2198 				/*
2199 				 * Add the blocks and the file to the new
2200 				 * user's quota.
2201 				 */
2202 				ip->i_dquot = getinoquota(ip);
2203 				(void) chkdq(ip, blocks, /* force */ 1, cr,
2204 						&errmsg1, &len1);
2205 				(void) chkiq(ufsvfsp, /* change */ 1,
2206 						(struct inode *)NULL,
2207 						(uid_t)ip->i_uid,
2208 						/* force */ 1, cr,
2209 						&errmsg2, &len2);
2210 			}
2211 		}
2212 		if (mask & AT_GID) {
2213 			ip->i_gid = vap->va_gid;
2214 		}
2215 		TRANS_INODE(ufsvfsp, ip);
2216 		ip->i_flag |= ICHG;
2217 	}
2218 	/*
2219 	 * Change file access or modified times.
2220 	 */
2221 	if (mask & (AT_ATIME|AT_MTIME)) {
2222 		/* Check that the time value is within ufs range */
2223 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2224 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2225 			error = EOVERFLOW;
2226 			goto update_inode;
2227 		}
2228 
2229 		/*
2230 		 * if the "noaccess" mount option is set and only atime
2231 		 * update is requested, do nothing. No error is returned.
2232 		 */
2233 		if ((ufsvfsp->vfs_noatime) &&
2234 		    ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME))
2235 			goto skip_atime;
2236 
2237 		if (mask & AT_ATIME) {
2238 			ip->i_atime.tv_sec = vap->va_atime.tv_sec;
2239 			ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000;
2240 			ip->i_flag &= ~IACC;
2241 		}
2242 		if (mask & AT_MTIME) {
2243 			ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
2244 			ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000;
2245 			gethrestime(&now);
2246 			if (now.tv_sec > TIME32_MAX) {
2247 				/*
2248 				 * In 2038, ctime sticks forever..
2249 				 */
2250 				ip->i_ctime.tv_sec = TIME32_MAX;
2251 				ip->i_ctime.tv_usec = 0;
2252 			} else {
2253 				ip->i_ctime.tv_sec = now.tv_sec;
2254 				ip->i_ctime.tv_usec = now.tv_nsec / 1000;
2255 			}
2256 			ip->i_flag &= ~(IUPD|ICHG);
2257 			ip->i_flag |= IMODTIME;
2258 		}
2259 		TRANS_INODE(ufsvfsp, ip);
2260 		ip->i_flag |= IMOD;
2261 	}
2262 
2263 skip_atime:
2264 	/*
2265 	 * The presence of a shadow inode may indicate an ACL, but does
2266 	 * not imply an ACL.  Future FSD types should be handled here too
2267 	 * and check for the presence of the attribute-specific data
2268 	 * before referencing it.
2269 	 */
2270 	if (ip->i_shadow) {
2271 		/*
2272 		 * XXX if ufs_iupdat is changed to sandbagged write fix
2273 		 * ufs_acl_setattr to push ip to keep acls consistent
2274 		 *
2275 		 * Suppress out of inodes messages if we will retry.
2276 		 */
2277 		if (retry)
2278 			ip->i_flag |= IQUIET;
2279 		error = ufs_acl_setattr(ip, vap, cr);
2280 		ip->i_flag &= ~IQUIET;
2281 	}
2282 
2283 update_inode:
2284 	/*
2285 	 * Setattr always increases the sequence number
2286 	 */
2287 	ip->i_seq++;
2288 
2289 	/*
2290 	 * if nfsd and not logging; push synchronously
2291 	 */
2292 	if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) {
2293 		ufs_iupdat(ip, 1);
2294 	} else {
2295 		ITIMES_NOLOCK(ip);
2296 	}
2297 
2298 	rw_exit(&ip->i_contents);
2299 	if (dodqlock) {
2300 		rw_exit(&ufsvfsp->vfs_dqrwlock);
2301 	}
2302 	if (dorwlock)
2303 		rw_exit(&ip->i_rwlock);
2304 
2305 	if (ulp) {
2306 		if (dotrans) {
2307 			int terr = 0;
2308 			TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR,
2309 			    trans_size);
2310 			if (error == 0)
2311 				error = terr;
2312 		}
2313 		ufs_lockfs_end(ulp);
2314 	}
2315 out:
2316 	/*
2317 	 * If out of inodes or blocks, see if we can free something
2318 	 * up from the delete queue.
2319 	 */
2320 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
2321 		ufs_delete_drain_wait(ufsvfsp, 1);
2322 		retry = 0;
2323 		if (errmsg1 != NULL)
2324 			kmem_free(errmsg1, len1);
2325 		if (errmsg2 != NULL)
2326 			kmem_free(errmsg2, len2);
2327 		goto again;
2328 	}
2329 	TRACE_2(TR_FAC_UFS, TR_UFS_SETATTR_END,
2330 		"ufs_setattr_end:vp %p error %d", vp, error);
2331 	if (errmsg1 != NULL) {
2332 		uprintf(errmsg1);
2333 		kmem_free(errmsg1, len1);
2334 	}
2335 	if (errmsg2 != NULL) {
2336 		uprintf(errmsg2);
2337 		kmem_free(errmsg2, len2);
2338 	}
2339 	return (error);
2340 }
2341 
2342 /*ARGSUSED*/
2343 static int
2344 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr)
2345 {
2346 	struct inode *ip = VTOI(vp);
2347 	int error;
2348 
2349 	TRACE_3(TR_FAC_UFS, TR_UFS_ACCESS_START,
2350 		"ufs_access_start:vp %p mode %x flags %x", vp, mode, flags);
2351 
2352 	if (ip->i_ufsvfs == NULL)
2353 		return (EIO);
2354 
2355 	rw_enter(&ip->i_contents, RW_READER);
2356 
2357 	/*
2358 	 * The ufs_iaccess function wants to be called with
2359 	 * mode bits expressed as "ufs specific" bits.
2360 	 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2361 	 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2362 	 * But since they're the same we just pass the vnode mode
2363 	 * bit but just verify that assumption at compile time.
2364 	 */
2365 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2366 #error "ufs_access needs to map Vmodes to Imodes"
2367 #endif
2368 	error = ufs_iaccess(ip, mode, cr);
2369 
2370 	rw_exit(&ip->i_contents);
2371 
2372 	TRACE_2(TR_FAC_UFS, TR_UFS_ACCESS_END,
2373 		"ufs_access_end:vp %p error %d", vp, error);
2374 	return (error);
2375 }
2376 
2377 /* ARGSUSED */
2378 static int
2379 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr)
2380 {
2381 	struct inode *ip = VTOI(vp);
2382 	struct ufsvfs *ufsvfsp;
2383 	struct ulockfs *ulp;
2384 	int error;
2385 	int fastsymlink;
2386 
2387 	TRACE_2(TR_FAC_UFS, TR_UFS_READLINK_START,
2388 		"ufs_readlink_start:vp %p uiop %p", uiop, vp);
2389 
2390 	if (vp->v_type != VLNK) {
2391 		error = EINVAL;
2392 		goto nolockout;
2393 	}
2394 
2395 	/*
2396 	 * If the symbolic link is empty there is nothing to read.
2397 	 * Fast-track these empty symbolic links
2398 	 */
2399 	if (ip->i_size == 0) {
2400 		error = 0;
2401 		goto nolockout;
2402 	}
2403 
2404 	ufsvfsp = ip->i_ufsvfs;
2405 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK);
2406 	if (error)
2407 		goto nolockout;
2408 	/*
2409 	 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2410 	 */
2411 again:
2412 	fastsymlink = 0;
2413 	if (ip->i_flag & IFASTSYMLNK) {
2414 		rw_enter(&ip->i_rwlock, RW_READER);
2415 		rw_enter(&ip->i_contents, RW_READER);
2416 		if (ip->i_flag & IFASTSYMLNK) {
2417 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
2418 			    (ip->i_fs->fs_ronly == 0) &&
2419 			    (!ufsvfsp->vfs_noatime)) {
2420 				mutex_enter(&ip->i_tlock);
2421 				ip->i_flag |= IACC;
2422 				mutex_exit(&ip->i_tlock);
2423 			}
2424 			error = uiomove((caddr_t)&ip->i_db[1],
2425 				MIN(ip->i_size, uiop->uio_resid),
2426 				UIO_READ, uiop);
2427 			ITIMES(ip);
2428 			++fastsymlink;
2429 		}
2430 		rw_exit(&ip->i_contents);
2431 		rw_exit(&ip->i_rwlock);
2432 	}
2433 	if (!fastsymlink) {
2434 		ssize_t size;	/* number of bytes read  */
2435 		caddr_t basep;	/* pointer to input data */
2436 		ino_t ino;
2437 		long  igen;
2438 		struct uio tuio;	/* temp uio struct */
2439 		struct uio *tuiop;
2440 		iovec_t tiov;		/* temp iovec struct */
2441 		char kbuf[FSL_SIZE];	/* buffer to hold fast symlink */
2442 		int tflag = 0;		/* flag to indicate temp vars used */
2443 
2444 		ino = ip->i_number;
2445 		igen = ip->i_gen;
2446 		size = uiop->uio_resid;
2447 		basep = uiop->uio_iov->iov_base;
2448 		tuiop = uiop;
2449 
2450 		rw_enter(&ip->i_rwlock, RW_WRITER);
2451 		rw_enter(&ip->i_contents, RW_WRITER);
2452 		if (ip->i_flag & IFASTSYMLNK) {
2453 			rw_exit(&ip->i_contents);
2454 			rw_exit(&ip->i_rwlock);
2455 			goto again;
2456 		}
2457 
2458 		/* can this be a fast symlink and is it a user buffer? */
2459 		if (ip->i_size <= FSL_SIZE &&
2460 		    (uiop->uio_segflg == UIO_USERSPACE ||
2461 		    uiop->uio_segflg == UIO_USERISPACE)) {
2462 
2463 			bzero(&tuio, sizeof (struct uio));
2464 			/*
2465 			 * setup a kernel buffer to read link into.  this
2466 			 * is to fix a race condition where the user buffer
2467 			 * got corrupted before copying it into the inode.
2468 			 */
2469 			size = ip->i_size;
2470 			tiov.iov_len = size;
2471 			tiov.iov_base = kbuf;
2472 			tuio.uio_iov = &tiov;
2473 			tuio.uio_iovcnt = 1;
2474 			tuio.uio_offset = uiop->uio_offset;
2475 			tuio.uio_segflg = UIO_SYSSPACE;
2476 			tuio.uio_fmode = uiop->uio_fmode;
2477 			tuio.uio_extflg = uiop->uio_extflg;
2478 			tuio.uio_limit = uiop->uio_limit;
2479 			tuio.uio_resid = size;
2480 
2481 			basep = tuio.uio_iov->iov_base;
2482 			tuiop = &tuio;
2483 			tflag = 1;
2484 		}
2485 
2486 		error = rdip(ip, tuiop, 0, cr);
2487 		if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) {
2488 			rw_exit(&ip->i_contents);
2489 			rw_exit(&ip->i_rwlock);
2490 			goto out;
2491 		}
2492 
2493 		if (tflag == 0)
2494 			size -= uiop->uio_resid;
2495 
2496 		if ((tflag == 0 && ip->i_size <= FSL_SIZE &&
2497 		    ip->i_size == size) || (tflag == 1 &&
2498 		    tuio.uio_resid == 0)) {
2499 			error = kcopy(basep, &ip->i_db[1], ip->i_size);
2500 			if (error == 0) {
2501 				ip->i_flag |= IFASTSYMLNK;
2502 				/*
2503 				 * free page
2504 				 */
2505 				(void) VOP_PUTPAGE(ITOV(ip),
2506 				    (offset_t)0, PAGESIZE,
2507 				    (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC),
2508 				    cr);
2509 			} else {
2510 				int i;
2511 				/* error, clear garbage left behind */
2512 				for (i = 1; i < NDADDR; i++)
2513 					ip->i_db[i] = 0;
2514 				for (i = 0; i < NIADDR; i++)
2515 					ip->i_ib[i] = 0;
2516 			}
2517 		}
2518 		if (tflag == 1) {
2519 			/* now, copy it into the user buffer */
2520 			error = uiomove((caddr_t)kbuf,
2521 				MIN(size, uiop->uio_resid),
2522 				UIO_READ, uiop);
2523 		}
2524 		rw_exit(&ip->i_contents);
2525 		rw_exit(&ip->i_rwlock);
2526 	}
2527 out:
2528 	if (ulp) {
2529 		ufs_lockfs_end(ulp);
2530 	}
2531 nolockout:
2532 	TRACE_2(TR_FAC_UFS, TR_UFS_READLINK_END,
2533 		"ufs_readlink_end:vp %p error %d", vp, error);
2534 
2535 	return (error);
2536 }
2537 
2538 /* ARGSUSED */
2539 static int
2540 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr)
2541 {
2542 	struct inode *ip = VTOI(vp);
2543 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2544 	struct ulockfs *ulp;
2545 	int error;
2546 
2547 	TRACE_1(TR_FAC_UFS, TR_UFS_FSYNC_START,
2548 		"ufs_fsync_start:vp %p", vp);
2549 
2550 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK);
2551 	if (error)
2552 		return (error);
2553 
2554 	if (TRANS_ISTRANS(ufsvfsp)) {
2555 		/*
2556 		 * First push out any data pages
2557 		 */
2558 		if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2559 		    (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) {
2560 			error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0,
2561 			    0, CRED());
2562 			if (error)
2563 				goto out;
2564 		}
2565 
2566 		/*
2567 		 * Delta any delayed inode times updates
2568 		 * and push inode to log.
2569 		 * All other inode deltas will have already been delta'd
2570 		 * and will be pushed during the commit.
2571 		 */
2572 		if (!(syncflag & FDSYNC) &&
2573 		    ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) {
2574 			if (ulp) {
2575 				TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC,
2576 				    TOP_SYNCIP_SIZE);
2577 			}
2578 			rw_enter(&ip->i_contents, RW_READER);
2579 			mutex_enter(&ip->i_tlock);
2580 			ip->i_flag &= ~IMODTIME;
2581 			mutex_exit(&ip->i_tlock);
2582 			ufs_iupdat(ip, I_SYNC);
2583 			rw_exit(&ip->i_contents);
2584 			if (ulp) {
2585 				TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC,
2586 				    TOP_SYNCIP_SIZE);
2587 			}
2588 		}
2589 
2590 		/*
2591 		 * Commit the Moby transaction
2592 		 *
2593 		 * Deltas have already been made so we just need to
2594 		 * commit them with a synchronous transaction.
2595 		 * TRANS_BEGIN_SYNC() will return an error
2596 		 * if there are no deltas to commit, for an
2597 		 * empty transaction.
2598 		 */
2599 		if (ulp) {
2600 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE,
2601 			    error);
2602 			if (error) {
2603 				error = 0; /* commit wasn't needed */
2604 				goto out;
2605 			}
2606 			TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC,
2607 			    TOP_COMMIT_SIZE);
2608 		}
2609 	} else {	/* not logging */
2610 		if (!(IS_SWAPVP(vp)))
2611 			if (syncflag & FNODSYNC) {
2612 				/* Just update the inode only */
2613 				TRANS_IUPDAT(ip, 1);
2614 				error = 0;
2615 			} else if (syncflag & FDSYNC)
2616 				/* Do data-synchronous writes */
2617 				error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC);
2618 			else
2619 				/* Do synchronous writes */
2620 				error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC);
2621 
2622 		rw_enter(&ip->i_contents, RW_WRITER);
2623 		if (!error)
2624 			error = ufs_sync_indir(ip);
2625 		rw_exit(&ip->i_contents);
2626 	}
2627 out:
2628 	if (ulp) {
2629 		ufs_lockfs_end(ulp);
2630 	}
2631 	TRACE_2(TR_FAC_UFS, TR_UFS_FSYNC_END,
2632 		"ufs_fsync_end:vp %p error %d", vp, error);
2633 	return (error);
2634 }
2635 
2636 /*ARGSUSED*/
2637 static void
2638 ufs_inactive(struct vnode *vp, struct cred *cr)
2639 {
2640 	ufs_iinactive(VTOI(vp));
2641 }
2642 
2643 /*
2644  * Unix file system operations having to do with directory manipulation.
2645  */
2646 int ufs_lookup_idle_count = 2;	/* Number of inodes to idle each time */
2647 /* ARGSUSED */
2648 static int
2649 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
2650 	struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr)
2651 {
2652 	struct inode *ip;
2653 	struct inode *sip;
2654 	struct inode *xip;
2655 	struct ufsvfs *ufsvfsp;
2656 	struct ulockfs *ulp;
2657 	struct vnode *vp;
2658 	int error;
2659 
2660 	TRACE_2(TR_FAC_UFS, TR_UFS_LOOKUP_START,
2661 		"ufs_lookup_start:dvp %p name %s", dvp, nm);
2662 
2663 
2664 	/*
2665 	 * Check flags for type of lookup (regular file or attribute file)
2666 	 */
2667 
2668 	ip = VTOI(dvp);
2669 
2670 	if (flags & LOOKUP_XATTR) {
2671 
2672 		/*
2673 		 * We don't allow recursive attributes...
2674 		 * Maybe someday we will.
2675 		 */
2676 		if ((ip->i_cflags & IXATTR)) {
2677 			return (EINVAL);
2678 		}
2679 
2680 		if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) {
2681 			error = ufs_xattr_getattrdir(dvp, &sip, flags, cr);
2682 			if (error) {
2683 				*vpp = NULL;
2684 				goto out;
2685 			}
2686 
2687 			vp = ITOV(sip);
2688 			dnlc_update(dvp, XATTR_DIR_NAME, vp);
2689 		}
2690 
2691 		/*
2692 		 * Check accessibility of directory.
2693 		 */
2694 		if (vp == DNLC_NO_VNODE) {
2695 			VN_RELE(vp);
2696 			error = ENOENT;
2697 			goto out;
2698 		}
2699 		if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr)) != 0) {
2700 			VN_RELE(vp);
2701 			goto out;
2702 		}
2703 
2704 		*vpp = vp;
2705 		return (0);
2706 	}
2707 
2708 	/*
2709 	 * Check for a null component, which we should treat as
2710 	 * looking at dvp from within it's parent, so we don't
2711 	 * need a call to ufs_iaccess(), as it has already been
2712 	 * done.
2713 	 */
2714 	if (nm[0] == 0) {
2715 		VN_HOLD(dvp);
2716 		error = 0;
2717 		*vpp = dvp;
2718 		goto out;
2719 	}
2720 
2721 	/*
2722 	 * Check for "." ie itself. this is a quick check and
2723 	 * avoids adding "." into the dnlc (which have been seen
2724 	 * to occupy >10% of the cache).
2725 	 */
2726 	if ((nm[0] == '.') && (nm[1] == 0)) {
2727 		/*
2728 		 * Don't return without checking accessibility
2729 		 * of the directory. We only need the lock if
2730 		 * we are going to return it.
2731 		 */
2732 		if ((error = ufs_iaccess(ip, IEXEC, cr)) == 0) {
2733 			VN_HOLD(dvp);
2734 			*vpp = dvp;
2735 		}
2736 		goto out;
2737 	}
2738 
2739 	/*
2740 	 * Fast path: Check the directory name lookup cache.
2741 	 */
2742 	if (vp = dnlc_lookup(dvp, nm)) {
2743 		/*
2744 		 * Check accessibility of directory.
2745 		 */
2746 		if ((error = ufs_iaccess(ip, IEXEC, cr)) != 0) {
2747 			VN_RELE(vp);
2748 			goto out;
2749 		}
2750 		if (vp == DNLC_NO_VNODE) {
2751 			VN_RELE(vp);
2752 			error = ENOENT;
2753 			goto out;
2754 		}
2755 		xip = VTOI(vp);
2756 		ulp = NULL;
2757 		goto fastpath;
2758 	}
2759 
2760 	/*
2761 	 * Keep the idle queue from getting too long by
2762 	 * idling two inodes before attempting to allocate another.
2763 	 *    This operation must be performed before entering
2764 	 *    lockfs or a transaction.
2765 	 */
2766 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2767 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2768 			ins.in_lidles.value.ul += ufs_lookup_idle_count;
2769 			ufs_idle_some(ufs_lookup_idle_count);
2770 		}
2771 
2772 	ufsvfsp = ip->i_ufsvfs;
2773 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK);
2774 	if (error)
2775 		goto out;
2776 
2777 	error = ufs_dirlook(ip, nm, &xip, cr, 1);
2778 
2779 fastpath:
2780 	if (error == 0) {
2781 		ip = xip;
2782 		*vpp = ITOV(ip);
2783 
2784 		/*
2785 		 * If vnode is a device return special vnode instead.
2786 		 */
2787 		if (IS_DEVVP(*vpp)) {
2788 			struct vnode *newvp;
2789 
2790 			newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
2791 			    cr);
2792 			VN_RELE(*vpp);
2793 			if (newvp == NULL)
2794 				error = ENOSYS;
2795 			else
2796 				*vpp = newvp;
2797 		}
2798 	}
2799 	if (ulp) {
2800 		ufs_lockfs_end(ulp);
2801 	}
2802 
2803 out:
2804 	TRACE_3(TR_FAC_UFS, TR_UFS_LOOKUP_END,
2805 		"ufs_lookup_end:dvp %p name %s error %d", vpp, nm, error);
2806 	return (error);
2807 }
2808 
2809 static int
2810 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl,
2811 	int mode, struct vnode **vpp, struct cred *cr, int flag)
2812 {
2813 	struct inode *ip;
2814 	struct inode *xip;
2815 	struct inode *dip;
2816 	struct vnode *xvp;
2817 	struct ufsvfs *ufsvfsp;
2818 	struct ulockfs *ulp;
2819 	int error;
2820 	int issync;
2821 	int truncflag;
2822 	int trans_size;
2823 	int noentry;
2824 	int defer_dip_seq_update = 0;	/* need to defer update of dip->i_seq */
2825 	int retry = 1;
2826 
2827 	TRACE_1(TR_FAC_UFS, TR_UFS_CREATE_START,
2828 		"ufs_create_start:dvp %p", dvp);
2829 
2830 again:
2831 	ip = VTOI(dvp);
2832 	ufsvfsp = ip->i_ufsvfs;
2833 	truncflag = 0;
2834 
2835 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK);
2836 	if (error)
2837 		goto out;
2838 
2839 	if (ulp) {
2840 		trans_size = (int)TOP_CREATE_SIZE(ip);
2841 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size);
2842 	}
2843 
2844 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
2845 		vap->va_mode &= ~VSVTX;
2846 
2847 	if (*name == '\0') {
2848 		/*
2849 		 * Null component name refers to the directory itself.
2850 		 */
2851 		VN_HOLD(dvp);
2852 		/*
2853 		 * Even though this is an error case, we need to grab the
2854 		 * quota lock since the error handling code below is common.
2855 		 */
2856 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2857 		rw_enter(&ip->i_contents, RW_WRITER);
2858 		error = EEXIST;
2859 	} else {
2860 		xip = NULL;
2861 		noentry = 0;
2862 		rw_enter(&ip->i_rwlock, RW_WRITER);
2863 		xvp = dnlc_lookup(dvp, name);
2864 		if (xvp == DNLC_NO_VNODE) {
2865 			noentry = 1;
2866 			VN_RELE(xvp);
2867 			xvp = NULL;
2868 		}
2869 		if (xvp) {
2870 			rw_exit(&ip->i_rwlock);
2871 			if (error = ufs_iaccess(ip, IEXEC, cr)) {
2872 				VN_RELE(xvp);
2873 			} else {
2874 				error = EEXIST;
2875 				xip = VTOI(xvp);
2876 			}
2877 		} else {
2878 			/*
2879 			 * Suppress file system full message if we will retry
2880 			 */
2881 			error = ufs_direnter_cm(ip, name, DE_CREATE,
2882 				vap, &xip, cr,
2883 				(noentry | (retry ? IQUIET : 0)));
2884 			rw_exit(&ip->i_rwlock);
2885 		}
2886 		ip = xip;
2887 		if (ip != NULL) {
2888 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2889 			rw_enter(&ip->i_contents, RW_WRITER);
2890 		}
2891 	}
2892 
2893 	/*
2894 	 * If the file already exists and this is a non-exclusive create,
2895 	 * check permissions and allow access for non-directories.
2896 	 * Read-only create of an existing directory is also allowed.
2897 	 * We fail an exclusive create of anything which already exists.
2898 	 */
2899 	if (error == EEXIST) {
2900 		dip = VTOI(dvp);
2901 		if (excl == NONEXCL) {
2902 			if ((((ip->i_mode & IFMT) == IFDIR) ||
2903 			    ((ip->i_mode & IFMT) == IFATTRDIR)) &&
2904 			    (mode & IWRITE))
2905 				error = EISDIR;
2906 			else if (mode)
2907 				error = ufs_iaccess(ip, mode, cr);
2908 			else
2909 				error = 0;
2910 		}
2911 		if (error) {
2912 			rw_exit(&ip->i_contents);
2913 			rw_exit(&ufsvfsp->vfs_dqrwlock);
2914 			VN_RELE(ITOV(ip));
2915 			goto unlock;
2916 		}
2917 		/*
2918 		 * If the error EEXIST was set, then i_seq can not
2919 		 * have been updated. The sequence number interface
2920 		 * is defined such that a non-error VOP_CREATE must
2921 		 * increase the dir va_seq it by at least one. If we
2922 		 * have cleared the error, increase i_seq. Note that
2923 		 * we are increasing the dir i_seq and in rare cases
2924 		 * ip may actually be from the dvp, so we already have
2925 		 * the locks and it will not be subject to truncation.
2926 		 * In case we have to update i_seq of the parent
2927 		 * directory dip, we have to defer it till we have
2928 		 * released our locks on ip due to lock ordering requirements.
2929 		 */
2930 		if (ip != dip)
2931 			defer_dip_seq_update = 1;
2932 		else
2933 			ip->i_seq++;
2934 
2935 		if (((ip->i_mode & IFMT) == IFREG) &&
2936 		    (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
2937 			/*
2938 			 * Truncate regular files, if requested by caller.
2939 			 * Grab i_rwlock to make sure no one else is
2940 			 * currently writing to the file (we promised
2941 			 * bmap we would do this).
2942 			 * Must get the locks in the correct order.
2943 			 */
2944 			if (ip->i_size == 0) {
2945 				ip->i_flag |= ICHG | IUPD;
2946 				ip->i_seq++;
2947 				TRANS_INODE(ufsvfsp, ip);
2948 			} else {
2949 				/*
2950 				 * Large Files: Why this check here?
2951 				 * Though we do it in vn_create() we really
2952 				 * want to guarantee that we do not destroy
2953 				 * Large file data by atomically checking
2954 				 * the size while holding the contents
2955 				 * lock.
2956 				 */
2957 				if (flag && !(flag & FOFFMAX) &&
2958 				    ((ip->i_mode & IFMT) == IFREG) &&
2959 				    (ip->i_size > (offset_t)MAXOFF32_T)) {
2960 					rw_exit(&ip->i_contents);
2961 					rw_exit(&ufsvfsp->vfs_dqrwlock);
2962 					error = EOVERFLOW;
2963 					goto unlock;
2964 				}
2965 				if (TRANS_ISTRANS(ufsvfsp))
2966 					truncflag++;
2967 				else {
2968 					rw_exit(&ip->i_contents);
2969 					rw_exit(&ufsvfsp->vfs_dqrwlock);
2970 					rw_enter(&ip->i_rwlock, RW_WRITER);
2971 					rw_enter(&ufsvfsp->vfs_dqrwlock,
2972 							RW_READER);
2973 					rw_enter(&ip->i_contents, RW_WRITER);
2974 					(void) ufs_itrunc(ip, (u_offset_t)0, 0,
2975 								cr);
2976 					rw_exit(&ip->i_rwlock);
2977 				}
2978 			}
2979 		}
2980 	}
2981 
2982 	if (error) {
2983 		if (ip != NULL) {
2984 			rw_exit(&ufsvfsp->vfs_dqrwlock);
2985 			rw_exit(&ip->i_contents);
2986 		}
2987 		goto unlock;
2988 	}
2989 
2990 	*vpp = ITOV(ip);
2991 	ITIMES(ip);
2992 	rw_exit(&ip->i_contents);
2993 	rw_exit(&ufsvfsp->vfs_dqrwlock);
2994 
2995 	/*
2996 	 * If vnode is a device return special vnode instead.
2997 	 */
2998 	if (!error && IS_DEVVP(*vpp)) {
2999 		struct vnode *newvp;
3000 
3001 		newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
3002 		VN_RELE(*vpp);
3003 		if (newvp == NULL) {
3004 			error = ENOSYS;
3005 			goto unlock;
3006 		}
3007 		truncflag = 0;
3008 		*vpp = newvp;
3009 	}
3010 unlock:
3011 
3012 	/*
3013 	 * Do the deferred update of the parent directory's sequence
3014 	 * number now.
3015 	 */
3016 	if (defer_dip_seq_update == 1) {
3017 		rw_enter(&dip->i_contents, RW_READER);
3018 		mutex_enter(&dip->i_tlock);
3019 		dip->i_seq++;
3020 		mutex_exit(&dip->i_tlock);
3021 		rw_exit(&dip->i_contents);
3022 	}
3023 
3024 	if (ulp) {
3025 		int terr = 0;
3026 
3027 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE,
3028 		    trans_size);
3029 
3030 		/*
3031 		 * If we haven't had a more interesting failure
3032 		 * already, then anything that might've happened
3033 		 * here should be reported.
3034 		 */
3035 		if (error == 0)
3036 			error = terr;
3037 	}
3038 
3039 	if (!error && truncflag) {
3040 		rw_enter(&ip->i_rwlock, RW_WRITER);
3041 		(void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr);
3042 		rw_exit(&ip->i_rwlock);
3043 	}
3044 
3045 	if (ulp)
3046 		ufs_lockfs_end(ulp);
3047 
3048 	/*
3049 	 * If no inodes available, try to free one up out of the
3050 	 * pending delete queue.
3051 	 */
3052 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3053 		ufs_delete_drain_wait(ufsvfsp, 1);
3054 		retry = 0;
3055 		goto again;
3056 	}
3057 
3058 out:
3059 	TRACE_3(TR_FAC_UFS, TR_UFS_CREATE_END,
3060 		"ufs_create_end:dvp %p name %s error %d", vpp, name, error);
3061 	return (error);
3062 }
3063 
3064 extern int ufs_idle_max;
3065 /*ARGSUSED*/
3066 static int
3067 ufs_remove(struct vnode *vp, char *nm, struct cred *cr)
3068 {
3069 	struct inode *ip = VTOI(vp);
3070 	struct ufsvfs *ufsvfsp	= ip->i_ufsvfs;
3071 	struct ulockfs *ulp;
3072 	vnode_t *rmvp = NULL;	/* Vnode corresponding to name being removed */
3073 	int error;
3074 	int issync;
3075 	int trans_size;
3076 
3077 	TRACE_1(TR_FAC_UFS, TR_UFS_REMOVE_START,
3078 		"ufs_remove_start:vp %p", vp);
3079 
3080 	/*
3081 	 * don't let the delete queue get too long
3082 	 */
3083 	if (ufsvfsp == NULL) {
3084 		error = EIO;
3085 		goto out;
3086 	}
3087 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3088 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3089 
3090 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK);
3091 	if (error)
3092 		goto out;
3093 
3094 	if (ulp)
3095 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE,
3096 		    trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp)));
3097 
3098 	rw_enter(&ip->i_rwlock, RW_WRITER);
3099 	error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0,
3100 	    DR_REMOVE, cr, &rmvp);
3101 	rw_exit(&ip->i_rwlock);
3102 
3103 	if (ulp) {
3104 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size);
3105 		ufs_lockfs_end(ulp);
3106 	}
3107 
3108 	/*
3109 	 * This must be called after the remove transaction is closed.
3110 	 */
3111 	if (rmvp != NULL) {
3112 		/* Only send the event if there were no errors */
3113 		if (error == 0)
3114 			vnevent_remove(rmvp);
3115 		VN_RELE(rmvp);
3116 	}
3117 out:
3118 	TRACE_3(TR_FAC_UFS, TR_UFS_REMOVE_END,
3119 		"ufs_remove_end:vp %p name %s error %d", vp, nm, error);
3120 	return (error);
3121 }
3122 
3123 /*
3124  * Link a file or a directory.  Only privileged processes are allowed to
3125  * make links to directories.
3126  */
3127 static int
3128 ufs_link(struct vnode *tdvp, struct vnode *svp, char *tnm, struct cred *cr)
3129 {
3130 	struct inode *sip;
3131 	struct inode *tdp = VTOI(tdvp);
3132 	struct ufsvfs *ufsvfsp = tdp->i_ufsvfs;
3133 	struct ulockfs *ulp;
3134 	struct vnode *realvp;
3135 	int error;
3136 	int issync;
3137 	int trans_size;
3138 	int isdev;
3139 
3140 	TRACE_1(TR_FAC_UFS, TR_UFS_LINK_START,
3141 		"ufs_link_start:tdvp %p", tdvp);
3142 
3143 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LINK_MASK);
3144 	if (error)
3145 		goto out;
3146 
3147 	if (ulp)
3148 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_LINK,
3149 		    trans_size = (int)TOP_LINK_SIZE(VTOI(tdvp)));
3150 
3151 	if (VOP_REALVP(svp, &realvp) == 0)
3152 		svp = realvp;
3153 
3154 	/*
3155 	 * Make sure link for extended attributes is valid
3156 	 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3157 	 *
3158 	 * Make certain we don't attempt to look at a device node as
3159 	 * a ufs inode.
3160 	 */
3161 
3162 	isdev = IS_DEVVP(svp);
3163 	if (((isdev == 0) && ((VTOI(svp)->i_cflags & IXATTR) == 0) &&
3164 	    ((tdp->i_mode & IFMT) == IFATTRDIR)) ||
3165 	    ((isdev == 0) && (VTOI(svp)->i_cflags & IXATTR) &&
3166 	    ((tdp->i_mode & IFMT) == IFDIR))) {
3167 		error = EINVAL;
3168 		goto unlock;
3169 	}
3170 
3171 	sip = VTOI(svp);
3172 	if ((svp->v_type == VDIR &&
3173 	    secpolicy_fs_linkdir(cr, ufsvfsp->vfs_vfs) != 0) ||
3174 	    (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)) {
3175 		error = EPERM;
3176 		goto unlock;
3177 	}
3178 	rw_enter(&tdp->i_rwlock, RW_WRITER);
3179 	error = ufs_direnter_lr(tdp, tnm, DE_LINK, (struct inode *)0,
3180 	    sip, cr, NULL);
3181 	rw_exit(&tdp->i_rwlock);
3182 
3183 unlock:
3184 	if (ulp) {
3185 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_LINK, trans_size);
3186 		ufs_lockfs_end(ulp);
3187 	}
3188 out:
3189 	TRACE_2(TR_FAC_UFS, TR_UFS_LINK_END,
3190 		"ufs_link_end:tdvp %p error %d", tdvp, error);
3191 	return (error);
3192 }
3193 
3194 uint64_t ufs_rename_retry_cnt;
3195 uint64_t ufs_rename_upgrade_retry_cnt;
3196 uint64_t ufs_rename_dircheck_retry_cnt;
3197 clock_t	 ufs_rename_backoff_delay = 1;
3198 
3199 /*
3200  * Rename a file or directory.
3201  * We are given the vnode and entry string of the source and the
3202  * vnode and entry string of the place we want to move the source
3203  * to (the target). The essential operation is:
3204  *	unlink(target);
3205  *	link(source, target);
3206  *	unlink(source);
3207  * but "atomically".  Can't do full commit without saving state in
3208  * the inode on disk, which isn't feasible at this time.  Best we
3209  * can do is always guarantee that the TARGET exists.
3210  */
3211 
3212 /*ARGSUSED*/
3213 static int
3214 ufs_rename(
3215 	struct vnode *sdvp,		/* old (source) parent vnode */
3216 	char *snm,			/* old (source) entry name */
3217 	struct vnode *tdvp,		/* new (target) parent vnode */
3218 	char *tnm,			/* new (target) entry name */
3219 	struct cred *cr)
3220 {
3221 	struct inode *sip = NULL;	/* source inode */
3222 	struct inode *ip = NULL;	/* check inode */
3223 	struct inode *sdp;		/* old (source) parent inode */
3224 	struct inode *tdp;		/* new (target) parent inode */
3225 	struct vnode *tvp = NULL;	/* target vnode, if it exists */
3226 	struct vnode *realvp;
3227 	struct ufsvfs *ufsvfsp;
3228 	struct ulockfs *ulp;
3229 	struct slot slot;
3230 	timestruc_t now;
3231 	int error;
3232 	int issync;
3233 	int trans_size;
3234 
3235 	TRACE_1(TR_FAC_UFS, TR_UFS_RENAME_START,
3236 		"ufs_rename_start:sdvp %p", sdvp);
3237 
3238 
3239 	sdp = VTOI(sdvp);
3240 	slot.fbp = NULL;
3241 	ufsvfsp = sdp->i_ufsvfs;
3242 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
3243 	if (error)
3244 		goto out;
3245 
3246 	if (ulp)
3247 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
3248 		    trans_size = (int)TOP_RENAME_SIZE(sdp));
3249 
3250 	if (VOP_REALVP(tdvp, &realvp) == 0)
3251 		tdvp = realvp;
3252 
3253 	tdp = VTOI(tdvp);
3254 
3255 
3256 	/*
3257 	 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3258 	 */
3259 	if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) {
3260 		error = EINVAL;
3261 		goto unlock;
3262 	}
3263 
3264 	/*
3265 	 * Look up inode of file we're supposed to rename.
3266 	 */
3267 	gethrestime(&now);
3268 	if (error = ufs_dirlook(sdp, snm, &sip, cr, 0)) {
3269 		goto unlock;
3270 	}
3271 
3272 	/*
3273 	 * Lock both the source and target directories (they may be
3274 	 * the same) to provide the atomicity semantics that was
3275 	 * previously provided by the per file system vfs_rename_lock
3276 	 *
3277 	 * with vfs_rename_lock removed to allow simultaneous renames
3278 	 * within a file system, ufs_dircheckpath can deadlock while
3279 	 * traversing back to ensure that source is not a parent directory
3280 	 * of target parent directory. This is because we get into
3281 	 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3282 	 * If the tdp and sdp of the simultaneous renames happen to be
3283 	 * in the path of each other, it can lead to a deadlock. This
3284 	 * can be avoided by getting the locks as RW_READER here and then
3285 	 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3286 	 */
3287 retry:
3288 	rw_enter(&tdp->i_rwlock, RW_READER);
3289 	if (tdp != sdp) {
3290 		/*
3291 		 * We're locking 2 peer level locks, so must use tryenter
3292 		 * on the 2nd to avoid deadlocks that would occur
3293 		 * if we renamed a->b and b->a concurrently.
3294 		 */
3295 		if (!rw_tryenter(&sdp->i_rwlock, RW_READER)) {
3296 			/*
3297 			 * Reverse the lock grabs in case we have heavy
3298 			 * contention on the 2nd lock.
3299 			 */
3300 			rw_exit(&tdp->i_rwlock);
3301 			rw_enter(&sdp->i_rwlock, RW_READER);
3302 			if (!rw_tryenter(&tdp->i_rwlock, RW_READER)) {
3303 				ufs_rename_retry_cnt++;
3304 				rw_exit(&sdp->i_rwlock);
3305 				goto retry;
3306 			}
3307 		}
3308 	}
3309 
3310 	if (sip == tdp) {
3311 		error = EINVAL;
3312 		goto errout;
3313 	}
3314 	/*
3315 	 * Make sure we can delete the source entry.  This requires
3316 	 * write permission on the containing directory.
3317 	 * Check for sticky directories.
3318 	 */
3319 	rw_enter(&sdp->i_contents, RW_READER);
3320 	rw_enter(&sip->i_contents, RW_READER);
3321 	if ((error = ufs_iaccess(sdp, IWRITE, cr)) != 0 ||
3322 	    (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) {
3323 		rw_exit(&sip->i_contents);
3324 		rw_exit(&sdp->i_contents);
3325 		goto errout;
3326 	}
3327 
3328 	/*
3329 	 * If this is a rename of a directory and the parent is
3330 	 * different (".." must be changed), then the source
3331 	 * directory must not be in the directory hierarchy
3332 	 * above the target, as this would orphan everything
3333 	 * below the source directory.  Also the user must have
3334 	 * write permission in the source so as to be able to
3335 	 * change "..".
3336 	 */
3337 	if ((((sip->i_mode & IFMT) == IFDIR) ||
3338 	    ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) {
3339 		ino_t	inum;
3340 
3341 		if ((error = ufs_iaccess(sip, IWRITE, cr))) {
3342 			rw_exit(&sip->i_contents);
3343 			rw_exit(&sdp->i_contents);
3344 			goto errout;
3345 		}
3346 		inum = sip->i_number;
3347 		rw_exit(&sip->i_contents);
3348 		rw_exit(&sdp->i_contents);
3349 		if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) {
3350 			/*
3351 			 * If we got EAGAIN ufs_dircheckpath detected a
3352 			 * potential deadlock and backed out. We need
3353 			 * to retry the operation since sdp and tdp have
3354 			 * to be released to avoid the deadlock.
3355 			 */
3356 			if (error == EAGAIN) {
3357 				rw_exit(&tdp->i_rwlock);
3358 				if (tdp != sdp)
3359 					rw_exit(&sdp->i_rwlock);
3360 				delay(ufs_rename_backoff_delay);
3361 				ufs_rename_dircheck_retry_cnt++;
3362 				goto retry;
3363 			}
3364 			goto errout;
3365 		}
3366 	} else {
3367 		rw_exit(&sip->i_contents);
3368 		rw_exit(&sdp->i_contents);
3369 	}
3370 
3371 
3372 	/*
3373 	 * Check for renaming '.' or '..' or alias of '.'
3374 	 */
3375 	if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) {
3376 		error = EINVAL;
3377 		goto errout;
3378 	}
3379 
3380 	/*
3381 	 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3382 	 * tries to traverse back the file tree with both tdp and sdp held
3383 	 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3384 	 * as RW_READERS  till ufs_dircheckpath is done.
3385 	 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3386 	 * to RW_WRITER.
3387 	 */
3388 	if (!rw_tryupgrade(&tdp->i_rwlock)) {
3389 		/*
3390 		 * The upgrade failed. We got to give away the lock
3391 		 * as to avoid deadlocking with someone else who is
3392 		 * waiting for writer lock. With the lock gone, we
3393 		 * cannot be sure the checks done above will hold
3394 		 * good when we eventually get them back as writer.
3395 		 * So if we can't upgrade we drop the locks and retry
3396 		 * everything again.
3397 		 */
3398 		rw_exit(&tdp->i_rwlock);
3399 		if (tdp != sdp)
3400 			rw_exit(&sdp->i_rwlock);
3401 		delay(ufs_rename_backoff_delay);
3402 		ufs_rename_upgrade_retry_cnt++;
3403 		goto retry;
3404 	}
3405 	if (tdp != sdp) {
3406 		if (!rw_tryupgrade(&sdp->i_rwlock)) {
3407 			/*
3408 			 * The upgrade failed. We got to give away the lock
3409 			 * as to avoid deadlocking with someone else who is
3410 			 * waiting for writer lock. With the lock gone, we
3411 			 * cannot be sure the checks done above will hold
3412 			 * good when we eventually get them back as writer.
3413 			 * So if we can't upgrade we drop the locks and retry
3414 			 * everything again.
3415 			 */
3416 			rw_exit(&tdp->i_rwlock);
3417 			rw_exit(&sdp->i_rwlock);
3418 			delay(ufs_rename_backoff_delay);
3419 			ufs_rename_upgrade_retry_cnt++;
3420 			goto retry;
3421 		}
3422 	}
3423 
3424 	/*
3425 	 * Now that all the locks are held check to make sure another thread
3426 	 * didn't slip in and take out the sip.
3427 	 */
3428 	slot.status = NONE;
3429 	if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec ||
3430 	    sip->i_ctime.tv_sec > now.tv_sec) {
3431 		rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER);
3432 		rw_enter(&sdp->i_contents, RW_WRITER);
3433 		error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot,
3434 		    &ip, cr, 0);
3435 		rw_exit(&sdp->i_contents);
3436 		rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock);
3437 		if (error) {
3438 			goto errout;
3439 		}
3440 		if (ip == NULL) {
3441 			error = ENOENT;
3442 			goto errout;
3443 		} else {
3444 			/*
3445 			 * If the inode was found need to drop the v_count
3446 			 * so as not to keep the filesystem from being
3447 			 * unmounted at a later time.
3448 			 */
3449 			VN_RELE(ITOV(ip));
3450 		}
3451 
3452 		/*
3453 		 * Release the slot.fbp that has the page mapped and
3454 		 * locked SE_SHARED, and could be used in in
3455 		 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3456 		 * on said page.
3457 		 */
3458 		if (slot.fbp) {
3459 			fbrelse(slot.fbp, S_OTHER);
3460 			slot.fbp = NULL;
3461 		}
3462 	}
3463 
3464 	/*
3465 	 * Link source to the target.  If a target exists, return its
3466 	 * vnode pointer in tvp.  We'll release it after sending the
3467 	 * vnevent.
3468 	 */
3469 	if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr, &tvp)) {
3470 		/*
3471 		 * ESAME isn't really an error; it indicates that the
3472 		 * operation should not be done because the source and target
3473 		 * are the same file, but that no error should be reported.
3474 		 */
3475 		if (error == ESAME)
3476 			error = 0;
3477 		goto errout;
3478 	}
3479 
3480 	/*
3481 	 * Unlink the source.
3482 	 * Remove the source entry.  ufs_dirremove() checks that the entry
3483 	 * still reflects sip, and returns an error if it doesn't.
3484 	 * If the entry has changed just forget about it.  Release
3485 	 * the source inode.
3486 	 */
3487 	if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0,
3488 	    DR_RENAME, cr, NULL)) == ENOENT)
3489 		error = 0;
3490 
3491 errout:
3492 	if (slot.fbp)
3493 		fbrelse(slot.fbp, S_OTHER);
3494 
3495 	rw_exit(&tdp->i_rwlock);
3496 	if (sdp != tdp) {
3497 		rw_exit(&sdp->i_rwlock);
3498 	}
3499 
3500 unlock:
3501 	if (ulp) {
3502 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
3503 		ufs_lockfs_end(ulp);
3504 	}
3505 
3506 	/*
3507 	 * If no errors, send the appropriate events on the source
3508 	 * and destination (a.k.a, target) vnodes, if they exist.
3509 	 * This has to be done after the rename transaction has closed.
3510 	 */
3511 	if (error == 0) {
3512 		if (tvp != NULL)
3513 			vnevent_rename_dest(tvp);
3514 		/*
3515 		 * Note that if ufs_direnter_lr() returned ESAME then
3516 		 * this event will still be sent.  This isn't expected
3517 		 * to be a problem for anticipated usage by consumers.
3518 		 */
3519 		if (sip != NULL)
3520 			vnevent_rename_src(ITOV(sip));
3521 	}
3522 
3523 	if (tvp != NULL)
3524 		VN_RELE(tvp);
3525 
3526 	if (sip != NULL)
3527 		VN_RELE(ITOV(sip));
3528 
3529 out:
3530 	TRACE_5(TR_FAC_UFS, TR_UFS_RENAME_END,
3531 		"ufs_rename_end:sdvp %p snm %s tdvp %p tnm %s error %d",
3532 			sdvp, snm, tdvp, tnm, error);
3533 	return (error);
3534 }
3535 
3536 /*ARGSUSED*/
3537 static int
3538 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap,
3539 	struct vnode **vpp, struct cred *cr)
3540 {
3541 	struct inode *ip;
3542 	struct inode *xip;
3543 	struct ufsvfs *ufsvfsp;
3544 	struct ulockfs *ulp;
3545 	int error;
3546 	int issync;
3547 	int trans_size;
3548 	int retry = 1;
3549 
3550 	ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
3551 
3552 	TRACE_1(TR_FAC_UFS, TR_UFS_MKDIR_START,
3553 		"ufs_mkdir_start:dvp %p", dvp);
3554 
3555 	/*
3556 	 * Can't make directory in attr hidden dir
3557 	 */
3558 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3559 		return (EINVAL);
3560 
3561 again:
3562 	ip = VTOI(dvp);
3563 	ufsvfsp = ip->i_ufsvfs;
3564 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK);
3565 	if (error)
3566 		goto out;
3567 	if (ulp)
3568 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR,
3569 		    trans_size = (int)TOP_MKDIR_SIZE(ip));
3570 
3571 	rw_enter(&ip->i_rwlock, RW_WRITER);
3572 
3573 	error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr,
3574 		(retry ? IQUIET : 0));
3575 
3576 	rw_exit(&ip->i_rwlock);
3577 	if (error == 0) {
3578 		ip = xip;
3579 		*vpp = ITOV(ip);
3580 	} else if (error == EEXIST)
3581 		VN_RELE(ITOV(xip));
3582 
3583 	if (ulp) {
3584 		int terr = 0;
3585 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size);
3586 		ufs_lockfs_end(ulp);
3587 		if (error == 0)
3588 			error = terr;
3589 	}
3590 out:
3591 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3592 		ufs_delete_drain_wait(ufsvfsp, 1);
3593 		retry = 0;
3594 		goto again;
3595 	}
3596 
3597 	TRACE_2(TR_FAC_UFS, TR_UFS_MKDIR_END,
3598 		"ufs_mkdir_end:dvp %p error %d", dvp, error);
3599 	return (error);
3600 }
3601 
3602 /*ARGSUSED*/
3603 static int
3604 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr)
3605 {
3606 	struct inode *ip = VTOI(vp);
3607 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3608 	struct ulockfs *ulp;
3609 	vnode_t *rmvp = NULL;	/* Vnode of removed directory */
3610 	int error;
3611 	int issync;
3612 
3613 	TRACE_1(TR_FAC_UFS, TR_UFS_RMDIR_START,
3614 		"ufs_rmdir_start:vp %p", vp);
3615 
3616 	/*
3617 	 * don't let the delete queue get too long
3618 	 */
3619 	if (ufsvfsp == NULL) {
3620 		error = EIO;
3621 		goto out;
3622 	}
3623 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3624 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3625 
3626 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK);
3627 	if (error)
3628 		goto out;
3629 
3630 	if (ulp)
3631 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR, TOP_RMDIR_SIZE);
3632 
3633 	rw_enter(&ip->i_rwlock, RW_WRITER);
3634 	error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr,
3635 									&rmvp);
3636 	rw_exit(&ip->i_rwlock);
3637 
3638 	if (ulp) {
3639 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR,
3640 				TOP_RMDIR_SIZE);
3641 		ufs_lockfs_end(ulp);
3642 	}
3643 
3644 	/*
3645 	 * This must be done AFTER the rmdir transaction has closed.
3646 	 */
3647 	if (rmvp != NULL) {
3648 		/* Only send the event if there were no errors */
3649 		if (error == 0)
3650 			vnevent_rmdir(rmvp);
3651 		VN_RELE(rmvp);
3652 	}
3653 out:
3654 	TRACE_2(TR_FAC_UFS, TR_UFS_RMDIR_END,
3655 		"ufs_rmdir_end:vp %p error %d", vp, error);
3656 
3657 	return (error);
3658 }
3659 
3660 /* ARGSUSED */
3661 static int
3662 ufs_readdir(
3663 	struct vnode *vp,
3664 	struct uio *uiop,
3665 	struct cred *cr,
3666 	int *eofp)
3667 {
3668 	struct iovec *iovp;
3669 	struct inode *ip;
3670 	struct direct *idp;
3671 	struct dirent64 *odp;
3672 	struct fbuf *fbp;
3673 	struct ufsvfs *ufsvfsp;
3674 	struct ulockfs *ulp;
3675 	caddr_t outbuf;
3676 	size_t bufsize;
3677 	uint_t offset;
3678 	uint_t bytes_wanted, total_bytes_wanted;
3679 	int incount = 0;
3680 	int outcount = 0;
3681 	int error;
3682 
3683 	ip = VTOI(vp);
3684 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
3685 
3686 	TRACE_2(TR_FAC_UFS, TR_UFS_READDIR_START,
3687 		"ufs_readdir_start:vp %p uiop %p", vp, uiop);
3688 
3689 	if (uiop->uio_loffset >= MAXOFF32_T) {
3690 		if (eofp)
3691 			*eofp = 1;
3692 		return (0);
3693 	}
3694 
3695 	/*
3696 	 * Check if we have been called with a valid iov_len
3697 	 * and bail out if not, otherwise we may potentially loop
3698 	 * forever further down.
3699 	 */
3700 	if (uiop->uio_iov->iov_len <= 0) {
3701 		error = EINVAL;
3702 		goto out;
3703 	}
3704 
3705 	/*
3706 	 * Large Files: When we come here we are guaranteed that
3707 	 * uio_offset can be used safely. The high word is zero.
3708 	 */
3709 
3710 	ufsvfsp = ip->i_ufsvfs;
3711 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READDIR_MASK);
3712 	if (error)
3713 		goto out;
3714 
3715 	iovp = uiop->uio_iov;
3716 	total_bytes_wanted = iovp->iov_len;
3717 
3718 	/* Large Files: directory files should not be "large" */
3719 
3720 	ASSERT(ip->i_size <= MAXOFF32_T);
3721 
3722 	/* Force offset to be valid (to guard against bogus lseek() values) */
3723 	offset = (uint_t)uiop->uio_offset & ~(DIRBLKSIZ - 1);
3724 
3725 	/* Quit if at end of file or link count of zero (posix) */
3726 	if (offset >= (uint_t)ip->i_size || ip->i_nlink <= 0) {
3727 		if (eofp)
3728 			*eofp = 1;
3729 		error = 0;
3730 		goto unlock;
3731 	}
3732 
3733 	/*
3734 	 * Get space to change directory entries into fs independent format.
3735 	 * Do fast alloc for the most commonly used-request size (filesystem
3736 	 * block size).
3737 	 */
3738 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) {
3739 		bufsize = total_bytes_wanted;
3740 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
3741 		odp = (struct dirent64 *)outbuf;
3742 	} else {
3743 		bufsize = total_bytes_wanted;
3744 		odp = (struct dirent64 *)iovp->iov_base;
3745 	}
3746 
3747 nextblk:
3748 	bytes_wanted = total_bytes_wanted;
3749 
3750 	/* Truncate request to file size */
3751 	if (offset + bytes_wanted > (int)ip->i_size)
3752 		bytes_wanted = (int)(ip->i_size - offset);
3753 
3754 	/* Comply with MAXBSIZE boundary restrictions of fbread() */
3755 	if ((offset & MAXBOFFSET) + bytes_wanted > MAXBSIZE)
3756 		bytes_wanted = MAXBSIZE - (offset & MAXBOFFSET);
3757 
3758 	/*
3759 	 * Read in the next chunk.
3760 	 * We are still holding the i_rwlock.
3761 	 */
3762 	error = fbread(vp, (offset_t)offset, bytes_wanted, S_OTHER, &fbp);
3763 
3764 	if (error)
3765 		goto update_inode;
3766 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (ip->i_fs->fs_ronly == 0) &&
3767 	    (!ufsvfsp->vfs_noatime)) {
3768 		ip->i_flag |= IACC;
3769 	}
3770 	incount = 0;
3771 	idp = (struct direct *)fbp->fb_addr;
3772 	if (idp->d_ino == 0 && idp->d_reclen == 0 &&
3773 		idp->d_namlen == 0) {
3774 		cmn_err(CE_WARN, "ufs_readdir: bad dir, inumber = %llu, "
3775 			"fs = %s\n",
3776 			(u_longlong_t)ip->i_number, ufsvfsp->vfs_fs->fs_fsmnt);
3777 		fbrelse(fbp, S_OTHER);
3778 		error = ENXIO;
3779 		goto update_inode;
3780 	}
3781 	/* Transform to file-system independent format */
3782 	while (incount < bytes_wanted) {
3783 		/*
3784 		 * If the current directory entry is mangled, then skip
3785 		 * to the next block.  It would be nice to set the FSBAD
3786 		 * flag in the super-block so that a fsck is forced on
3787 		 * next reboot, but locking is a problem.
3788 		 */
3789 		if (idp->d_reclen & 0x3) {
3790 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
3791 			break;
3792 		}
3793 
3794 		/* Skip to requested offset and skip empty entries */
3795 		if (idp->d_ino != 0 && offset >= (uint_t)uiop->uio_offset) {
3796 			ushort_t this_reclen =
3797 			    DIRENT64_RECLEN(idp->d_namlen);
3798 			/* Buffer too small for any entries */
3799 			if (!outcount && this_reclen > bufsize) {
3800 				fbrelse(fbp, S_OTHER);
3801 				error = EINVAL;
3802 				goto update_inode;
3803 			}
3804 			/* If would overrun the buffer, quit */
3805 			if (outcount + this_reclen > bufsize) {
3806 				break;
3807 			}
3808 			/* Take this entry */
3809 			odp->d_ino = (ino64_t)idp->d_ino;
3810 			odp->d_reclen = (ushort_t)this_reclen;
3811 			odp->d_off = (offset_t)(offset + idp->d_reclen);
3812 
3813 			/* use strncpy(9f) to zero out uninitialized bytes */
3814 
3815 			ASSERT(strlen(idp->d_name) + 1 <=
3816 			    DIRENT64_NAMELEN(this_reclen));
3817 			(void) strncpy(odp->d_name, idp->d_name,
3818 			    DIRENT64_NAMELEN(this_reclen));
3819 			outcount += odp->d_reclen;
3820 			odp = (struct dirent64 *)((intptr_t)odp +
3821 				    odp->d_reclen);
3822 			ASSERT(outcount <= bufsize);
3823 		}
3824 		if (idp->d_reclen) {
3825 			incount += idp->d_reclen;
3826 			offset += idp->d_reclen;
3827 			idp = (struct direct *)((intptr_t)idp + idp->d_reclen);
3828 		} else {
3829 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
3830 			break;
3831 		}
3832 	}
3833 	/* Release the chunk */
3834 	fbrelse(fbp, S_OTHER);
3835 
3836 	/* Read whole block, but got no entries, read another if not eof */
3837 
3838 	/*
3839 	 * Large Files: casting i_size to int here is not a problem
3840 	 * because directory sizes are always less than MAXOFF32_T.
3841 	 * See assertion above.
3842 	 */
3843 
3844 	if (offset < (int)ip->i_size && !outcount)
3845 		goto nextblk;
3846 
3847 	/* Copy out the entry data */
3848 	if (uiop->uio_segflg == UIO_SYSSPACE && uiop->uio_iovcnt == 1) {
3849 		iovp->iov_base += outcount;
3850 		iovp->iov_len -= outcount;
3851 		uiop->uio_resid -= outcount;
3852 		uiop->uio_offset = offset;
3853 	} else if ((error = uiomove(outbuf, (long)outcount, UIO_READ,
3854 				    uiop)) == 0)
3855 		uiop->uio_offset = offset;
3856 update_inode:
3857 	ITIMES(ip);
3858 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1)
3859 		kmem_free(outbuf, bufsize);
3860 
3861 	if (eofp && error == 0)
3862 		*eofp = (uiop->uio_offset >= (int)ip->i_size);
3863 unlock:
3864 	if (ulp) {
3865 		ufs_lockfs_end(ulp);
3866 	}
3867 out:
3868 	TRACE_2(TR_FAC_UFS, TR_UFS_READDIR_END,
3869 		"ufs_readdir_end:vp %p error %d", vp, error);
3870 	return (error);
3871 }
3872 
3873 /*ARGSUSED*/
3874 static int
3875 ufs_symlink(
3876 	struct vnode *dvp,		/* ptr to parent dir vnode */
3877 	char *linkname,			/* name of symbolic link */
3878 	struct vattr *vap,		/* attributes */
3879 	char *target,			/* target path */
3880 	struct cred *cr)		/* user credentials */
3881 {
3882 	struct inode *ip, *dip = VTOI(dvp);
3883 	struct ufsvfs *ufsvfsp = dip->i_ufsvfs;
3884 	struct ulockfs *ulp;
3885 	int error;
3886 	int issync;
3887 	int trans_size;
3888 	int residual;
3889 	int ioflag;
3890 	int retry = 1;
3891 
3892 	TRACE_1(TR_FAC_UFS, TR_UFS_SYMLINK_START,
3893 		"ufs_symlink_start:dvp %p", dvp);
3894 
3895 	/*
3896 	 * No symlinks in attrdirs at this time
3897 	 */
3898 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3899 		return (EINVAL);
3900 
3901 again:
3902 	ip = (struct inode *)NULL;
3903 	vap->va_type = VLNK;
3904 	vap->va_rdev = 0;
3905 
3906 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK);
3907 	if (error)
3908 		goto out;
3909 
3910 	if (ulp)
3911 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK,
3912 		    trans_size = (int)TOP_SYMLINK_SIZE(dip));
3913 
3914 	/*
3915 	 * We must create the inode before the directory entry, to avoid
3916 	 * racing with readlink().  ufs_dirmakeinode requires that we
3917 	 * hold the quota lock as reader, and directory locks as writer.
3918 	 */
3919 
3920 	rw_enter(&dip->i_rwlock, RW_WRITER);
3921 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
3922 	rw_enter(&dip->i_contents, RW_WRITER);
3923 
3924 	/*
3925 	 * Suppress any out of inodes messages if we will retry on
3926 	 * ENOSP
3927 	 */
3928 	if (retry)
3929 		dip->i_flag |= IQUIET;
3930 
3931 	error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr);
3932 
3933 	dip->i_flag &= ~IQUIET;
3934 
3935 	rw_exit(&dip->i_contents);
3936 	rw_exit(&ufsvfsp->vfs_dqrwlock);
3937 	rw_exit(&dip->i_rwlock);
3938 
3939 	if (error)
3940 		goto unlock;
3941 
3942 	/*
3943 	 * OK.  The inode has been created.  Write out the data of the
3944 	 * symbolic link.  Since symbolic links are metadata, and should
3945 	 * remain consistent across a system crash, we need to force the
3946 	 * data out synchronously.
3947 	 *
3948 	 * (This is a change from the semantics in earlier releases, which
3949 	 * only created symbolic links synchronously if the semi-documented
3950 	 * 'syncdir' option was set, or if we were being invoked by the NFS
3951 	 * server, which requires symbolic links to be created synchronously.)
3952 	 *
3953 	 * We need to pass in a pointer for the residual length; otherwise
3954 	 * ufs_rdwri() will always return EIO if it can't write the data,
3955 	 * even if the error was really ENOSPC or EDQUOT.
3956 	 */
3957 
3958 	ioflag = FWRITE | FDSYNC;
3959 	residual = 0;
3960 
3961 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
3962 	rw_enter(&ip->i_contents, RW_WRITER);
3963 
3964 	/*
3965 	 * Suppress file system full messages if we will retry
3966 	 */
3967 	if (retry)
3968 		ip->i_flag |= IQUIET;
3969 
3970 	error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target),
3971 	    (offset_t)0, UIO_SYSSPACE, &residual, cr);
3972 
3973 	ip->i_flag &= ~IQUIET;
3974 
3975 	if (error) {
3976 		rw_exit(&ip->i_contents);
3977 		rw_exit(&ufsvfsp->vfs_dqrwlock);
3978 		goto remove;
3979 	}
3980 
3981 	/*
3982 	 * If the link's data is small enough, we can cache it in the inode.
3983 	 * This is a "fast symbolic link".  We don't use the first direct
3984 	 * block because that's actually used to point at the symbolic link's
3985 	 * contents on disk; but we know that none of the other direct or
3986 	 * indirect blocks can be used because symbolic links are restricted
3987 	 * to be smaller than a file system block.
3988 	 */
3989 
3990 	ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip)));
3991 
3992 	if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) {
3993 		if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) {
3994 			ip->i_flag |= IFASTSYMLNK;
3995 		} else {
3996 			int i;
3997 			/* error, clear garbage left behind */
3998 			for (i = 1; i < NDADDR; i++)
3999 				ip->i_db[i] = 0;
4000 			for (i = 0; i < NIADDR; i++)
4001 				ip->i_ib[i] = 0;
4002 		}
4003 	}
4004 
4005 	rw_exit(&ip->i_contents);
4006 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4007 
4008 	/*
4009 	 * OK.  We've successfully created the symbolic link.  All that
4010 	 * remains is to insert it into the appropriate directory.
4011 	 */
4012 
4013 	rw_enter(&dip->i_rwlock, RW_WRITER);
4014 	error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr, NULL);
4015 	rw_exit(&dip->i_rwlock);
4016 
4017 	/*
4018 	 * Fall through into remove-on-error code.  We're either done, or we
4019 	 * need to remove the inode (if we couldn't insert it).
4020 	 */
4021 
4022 remove:
4023 	if (error && (ip != NULL)) {
4024 		rw_enter(&ip->i_contents, RW_WRITER);
4025 		ip->i_nlink--;
4026 		ip->i_flag |= ICHG;
4027 		ip->i_seq++;
4028 		ufs_setreclaim(ip);
4029 		rw_exit(&ip->i_contents);
4030 	}
4031 
4032 unlock:
4033 	if (ip != NULL)
4034 		VN_RELE(ITOV(ip));
4035 
4036 	if (ulp) {
4037 		int terr = 0;
4038 
4039 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK,
4040 				trans_size);
4041 		ufs_lockfs_end(ulp);
4042 		if (error == 0)
4043 			error = terr;
4044 	}
4045 
4046 	/*
4047 	 * We may have failed due to lack of an inode or of a block to
4048 	 * store the target in.  Try flushing the delete queue to free
4049 	 * logically-available things up and try again.
4050 	 */
4051 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
4052 		ufs_delete_drain_wait(ufsvfsp, 1);
4053 		retry = 0;
4054 		goto again;
4055 	}
4056 
4057 out:
4058 	TRACE_2(TR_FAC_UFS, TR_UFS_SYMLINK_END,
4059 		"ufs_symlink_end:dvp %p error %d", dvp, error);
4060 	return (error);
4061 }
4062 
4063 /*
4064  * Ufs specific routine used to do ufs io.
4065  */
4066 int
4067 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base,
4068 	ssize_t len, offset_t offset, enum uio_seg seg, int *aresid,
4069 	struct cred *cr)
4070 {
4071 	struct uio auio;
4072 	struct iovec aiov;
4073 	int error;
4074 
4075 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
4076 
4077 	bzero((caddr_t)&auio, sizeof (uio_t));
4078 	bzero((caddr_t)&aiov, sizeof (iovec_t));
4079 
4080 	aiov.iov_base = base;
4081 	aiov.iov_len = len;
4082 	auio.uio_iov = &aiov;
4083 	auio.uio_iovcnt = 1;
4084 	auio.uio_loffset = offset;
4085 	auio.uio_segflg = (short)seg;
4086 	auio.uio_resid = len;
4087 
4088 	if (rw == UIO_WRITE) {
4089 		auio.uio_fmode = FWRITE;
4090 		auio.uio_extflg = UIO_COPY_DEFAULT;
4091 		auio.uio_llimit = curproc->p_fsz_ctl;
4092 		error = wrip(ip, &auio, ioflag, cr);
4093 	} else {
4094 		auio.uio_fmode = FREAD;
4095 		auio.uio_extflg = UIO_COPY_CACHED;
4096 		auio.uio_llimit = MAXOFFSET_T;
4097 		error = rdip(ip, &auio, ioflag, cr);
4098 	}
4099 
4100 	if (aresid) {
4101 		*aresid = auio.uio_resid;
4102 	} else if (auio.uio_resid) {
4103 		error = EIO;
4104 	}
4105 	return (error);
4106 }
4107 
4108 static int
4109 ufs_fid(vp, fidp)
4110 	struct vnode *vp;
4111 	struct fid *fidp;
4112 {
4113 	struct ufid *ufid;
4114 	struct inode *ip = VTOI(vp);
4115 
4116 	if (ip->i_ufsvfs == NULL)
4117 		return (EIO);
4118 
4119 	if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) {
4120 		fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t);
4121 		return (ENOSPC);
4122 	}
4123 
4124 	ufid = (struct ufid *)fidp;
4125 	bzero((char *)ufid, sizeof (struct ufid));
4126 	ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t);
4127 	ufid->ufid_ino = ip->i_number;
4128 	ufid->ufid_gen = ip->i_gen;
4129 
4130 	return (0);
4131 }
4132 
4133 /* ARGSUSED2 */
4134 static int
4135 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4136 {
4137 	struct inode	*ip = VTOI(vp);
4138 	struct ufsvfs	*ufsvfsp;
4139 	int		forcedirectio;
4140 
4141 	/*
4142 	 * Read case is easy.
4143 	 */
4144 	if (!write_lock) {
4145 		rw_enter(&ip->i_rwlock, RW_READER);
4146 		return (V_WRITELOCK_FALSE);
4147 	}
4148 
4149 	/*
4150 	 * Caller has requested a writer lock, but that inhibits any
4151 	 * concurrency in the VOPs that follow. Acquire the lock shared
4152 	 * and defer exclusive access until it is known to be needed in
4153 	 * other VOP handlers. Some cases can be determined here.
4154 	 */
4155 
4156 	/*
4157 	 * If directio is not set, there is no chance of concurrency,
4158 	 * so just acquire the lock exclusive. Beware of a forced
4159 	 * unmount before looking at the mount option.
4160 	 */
4161 	ufsvfsp = ip->i_ufsvfs;
4162 	forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0;
4163 	if (!(ip->i_flag & IDIRECTIO || forcedirectio) ||
4164 	    !ufs_allow_shared_writes) {
4165 		rw_enter(&ip->i_rwlock, RW_WRITER);
4166 		return (V_WRITELOCK_TRUE);
4167 	}
4168 
4169 	/*
4170 	 * Mandatory locking forces acquiring i_rwlock exclusive.
4171 	 */
4172 	if (MANDLOCK(vp, ip->i_mode)) {
4173 		rw_enter(&ip->i_rwlock, RW_WRITER);
4174 		return (V_WRITELOCK_TRUE);
4175 	}
4176 
4177 	/*
4178 	 * Acquire the lock shared in case a concurrent write follows.
4179 	 * Mandatory locking could have become enabled before the lock
4180 	 * was acquired. Re-check and upgrade if needed.
4181 	 */
4182 	rw_enter(&ip->i_rwlock, RW_READER);
4183 	if (MANDLOCK(vp, ip->i_mode)) {
4184 		rw_exit(&ip->i_rwlock);
4185 		rw_enter(&ip->i_rwlock, RW_WRITER);
4186 		return (V_WRITELOCK_TRUE);
4187 	}
4188 	return (V_WRITELOCK_FALSE);
4189 }
4190 
4191 /*ARGSUSED*/
4192 static void
4193 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4194 {
4195 	struct inode	*ip = VTOI(vp);
4196 
4197 	rw_exit(&ip->i_rwlock);
4198 }
4199 
4200 /* ARGSUSED */
4201 static int
4202 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp)
4203 {
4204 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4205 }
4206 
4207 /* ARGSUSED */
4208 static int
4209 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4210 	offset_t offset, struct flk_callback *flk_cbp, struct cred *cr)
4211 {
4212 	struct inode *ip = VTOI(vp);
4213 
4214 	if (ip->i_ufsvfs == NULL)
4215 		return (EIO);
4216 
4217 	/*
4218 	 * If file is being mapped, disallow frlock.
4219 	 * XXX I am not holding tlock while checking i_mapcnt because the
4220 	 * current locking strategy drops all locks before calling fs_frlock.
4221 	 * So, mapcnt could change before we enter fs_frlock making is
4222 	 * meaningless to have held tlock in the first place.
4223 	 */
4224 	if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode))
4225 		return (EAGAIN);
4226 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr));
4227 }
4228 
4229 /* ARGSUSED */
4230 static int
4231 ufs_space(
4232 	struct vnode *vp,
4233 	int cmd,
4234 	struct flock64 *bfp,
4235 	int flag,
4236 	offset_t offset,
4237 	cred_t *cr,
4238 	caller_context_t *ct)
4239 {
4240 	struct ufsvfs *ufsvfsp	= VTOI(vp)->i_ufsvfs;
4241 	struct ulockfs *ulp;
4242 	int error;
4243 
4244 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SPACE_MASK);
4245 	if (error)
4246 		return (error);
4247 
4248 
4249 	if (cmd != F_FREESP)
4250 		error =  EINVAL;
4251 	else if ((error = convoff(vp, bfp, 0, offset)) == 0)
4252 		error = ufs_freesp(vp, bfp, flag, cr);
4253 
4254 	if (ulp)
4255 		ufs_lockfs_end(ulp);
4256 	return (error);
4257 }
4258 
4259 /*
4260  * Used to determine if read ahead should be done. Also used to
4261  * to determine when write back occurs.
4262  */
4263 #define	CLUSTSZ(ip)		((ip)->i_ufsvfs->vfs_ioclustsz)
4264 
4265 /*
4266  * A faster version of ufs_getpage.
4267  *
4268  * We optimize by inlining the pvn_getpages iterator, eliminating
4269  * calls to bmap_read if file doesn't have UFS holes, and avoiding
4270  * the overhead of page_exists().
4271  *
4272  * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4273  * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4274  * victimizes performance when a file with UFS holes is faulted
4275  * first in the S_READ mode, and then in the S_WRITE mode. We will get
4276  * two MMU faults in this case.
4277  *
4278  * XXX - the inode fields which control the sequential mode are not
4279  *	 protected by any mutex. The read ahead will act wild if
4280  *	 multiple processes will access the file concurrently and
4281  *	 some of them in sequential mode. One particulary bad case
4282  *	 is if another thread will change the value of i_nextrio between
4283  *	 the time this thread tests the i_nextrio value and then reads it
4284  *	 again to use it as the offset for the read ahead.
4285  */
4286 static int
4287 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
4288 	page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr,
4289 	enum seg_rw rw, struct cred *cr)
4290 {
4291 	u_offset_t	uoff = (u_offset_t)off; /* type conversion */
4292 	u_offset_t	pgoff;
4293 	u_offset_t	eoff;
4294 	struct inode 	*ip = VTOI(vp);
4295 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
4296 	struct fs 	*fs;
4297 	struct ulockfs	*ulp;
4298 	page_t		**pl;
4299 	caddr_t		pgaddr;
4300 	krw_t		rwtype;
4301 	int 		err;
4302 	int		has_holes;
4303 	int		beyond_eof;
4304 	int		seqmode;
4305 	int		pgsize = PAGESIZE;
4306 	int		dolock;
4307 	int		do_qlock;
4308 	int		trans_size;
4309 
4310 	TRACE_1(TR_FAC_UFS, TR_UFS_GETPAGE_START,
4311 		"ufs_getpage_start:vp %p", vp);
4312 
4313 	ASSERT((uoff & PAGEOFFSET) == 0);
4314 
4315 	if (protp)
4316 		*protp = PROT_ALL;
4317 
4318 	/*
4319 	 * Obey the lockfs protocol
4320 	 */
4321 	err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg,
4322 			rw == S_READ || rw == S_EXEC, protp);
4323 	if (err)
4324 		goto out;
4325 
4326 	fs = ufsvfsp->vfs_fs;
4327 
4328 	if (ulp && (rw == S_CREATE || rw == S_WRITE) &&
4329 	    !(vp->v_flag & VISSWAP)) {
4330 		/*
4331 		 * Try to start a transaction, will return if blocking is
4332 		 * expected to occur and the address space is not the
4333 		 * kernel address space.
4334 		 */
4335 		trans_size = TOP_GETPAGE_SIZE(ip);
4336 		if (seg->s_as != &kas) {
4337 			TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE,
4338 				trans_size, err)
4339 			if (err == EWOULDBLOCK) {
4340 				/*
4341 				 * Use EDEADLK here because the VM code
4342 				 * can normally never see this error.
4343 				 */
4344 				err = EDEADLK;
4345 				ufs_lockfs_end(ulp);
4346 				goto out;
4347 			}
4348 		} else {
4349 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4350 		}
4351 	}
4352 
4353 	if (vp->v_flag & VNOMAP) {
4354 		err = ENOSYS;
4355 		goto unlock;
4356 	}
4357 
4358 	seqmode = ip->i_nextr == uoff && rw != S_CREATE;
4359 
4360 	rwtype = RW_READER;		/* start as a reader */
4361 	dolock = (rw_owner(&ip->i_contents) != curthread);
4362 	/*
4363 	 * If this thread owns the lock, i.e., this thread grabbed it
4364 	 * as writer somewhere above, then we don't need to grab the
4365 	 * lock as reader in this routine.
4366 	 */
4367 	do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread);
4368 
4369 retrylock:
4370 	if (dolock) {
4371 		/*
4372 		 * Grab the quota lock if we need to call
4373 		 * bmap_write() below (with i_contents as writer).
4374 		 */
4375 		if (do_qlock && rwtype == RW_WRITER)
4376 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4377 		rw_enter(&ip->i_contents, rwtype);
4378 	}
4379 
4380 	/*
4381 	 * We may be getting called as a side effect of a bmap using
4382 	 * fbread() when the blocks might be being allocated and the
4383 	 * size has not yet been up'ed.  In this case we want to be
4384 	 * able to return zero pages if we get back UFS_HOLE from
4385 	 * calling bmap for a non write case here.  We also might have
4386 	 * to read some frags from the disk into a page if we are
4387 	 * extending the number of frags for a given lbn in bmap().
4388 	 * Large Files: The read of i_size here is atomic because
4389 	 * i_contents is held here. If dolock is zero, the lock
4390 	 * is held in bmap routines.
4391 	 */
4392 	beyond_eof = uoff + len > ip->i_size + PAGEOFFSET;
4393 	if (beyond_eof && seg != segkmap) {
4394 		if (dolock) {
4395 			rw_exit(&ip->i_contents);
4396 			if (do_qlock && rwtype == RW_WRITER)
4397 				rw_exit(&ufsvfsp->vfs_dqrwlock);
4398 		}
4399 		err = EFAULT;
4400 		goto unlock;
4401 	}
4402 
4403 	/*
4404 	 * Must hold i_contents lock throughout the call to pvn_getpages
4405 	 * since locked pages are returned from each call to ufs_getapage.
4406 	 * Must *not* return locked pages and then try for contents lock
4407 	 * due to lock ordering requirements (inode > page)
4408 	 */
4409 
4410 	has_holes = bmap_has_holes(ip);
4411 
4412 	if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) {
4413 		int	blk_size;
4414 		u_offset_t offset;
4415 
4416 		/*
4417 		 * We must acquire the RW_WRITER lock in order to
4418 		 * call bmap_write().
4419 		 */
4420 		if (dolock && rwtype == RW_READER) {
4421 			rwtype = RW_WRITER;
4422 
4423 			/*
4424 			 * Grab the quota lock before
4425 			 * upgrading i_contents, but if we can't grab it
4426 			 * don't wait here due to lock order:
4427 			 * vfs_dqrwlock > i_contents.
4428 			 */
4429 			if (do_qlock && rw_tryenter(&ufsvfsp->vfs_dqrwlock,
4430 							RW_READER) == 0) {
4431 				rw_exit(&ip->i_contents);
4432 				goto retrylock;
4433 			}
4434 			if (!rw_tryupgrade(&ip->i_contents)) {
4435 				rw_exit(&ip->i_contents);
4436 				if (do_qlock)
4437 					rw_exit(&ufsvfsp->vfs_dqrwlock);
4438 				goto retrylock;
4439 			}
4440 		}
4441 
4442 		/*
4443 		 * May be allocating disk blocks for holes here as
4444 		 * a result of mmap faults. write(2) does the bmap_write
4445 		 * in rdip/wrip, not here. We are not dealing with frags
4446 		 * in this case.
4447 		 */
4448 		/*
4449 		 * Large Files: We cast fs_bmask field to offset_t
4450 		 * just as we do for MAXBMASK because uoff is a 64-bit
4451 		 * data type. fs_bmask will still be a 32-bit type
4452 		 * as we cannot change any ondisk data structures.
4453 		 */
4454 
4455 		offset = uoff & (offset_t)fs->fs_bmask;
4456 		while (offset < uoff + len) {
4457 			blk_size = (int)blksize(fs, ip, lblkno(fs, offset));
4458 			err = bmap_write(ip, offset, blk_size, 0, cr);
4459 			if (ip->i_flag & (ICHG|IUPD))
4460 				ip->i_seq++;
4461 			if (err)
4462 				goto update_inode;
4463 			offset += blk_size; /* XXX - make this contig */
4464 		}
4465 	}
4466 
4467 	/*
4468 	 * Can be a reader from now on.
4469 	 */
4470 	if (dolock && rwtype == RW_WRITER) {
4471 		rw_downgrade(&ip->i_contents);
4472 		/*
4473 		 * We can release vfs_dqrwlock early so do it, but make
4474 		 * sure we don't try to release it again at the bottom.
4475 		 */
4476 		if (do_qlock) {
4477 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4478 			do_qlock = 0;
4479 		}
4480 	}
4481 
4482 	/*
4483 	 * We remove PROT_WRITE in cases when the file has UFS holes
4484 	 * because we don't  want to call bmap_read() to check each
4485 	 * page if it is backed with a disk block.
4486 	 */
4487 	if (protp && has_holes && rw != S_WRITE && rw != S_CREATE)
4488 		*protp &= ~PROT_WRITE;
4489 
4490 	err = 0;
4491 
4492 	/*
4493 	 * The loop looks up pages in the range [off, off + len).
4494 	 * For each page, we first check if we should initiate an asynchronous
4495 	 * read ahead before we call page_lookup (we may sleep in page_lookup
4496 	 * for a previously initiated disk read).
4497 	 */
4498 	eoff = (uoff + len);
4499 	for (pgoff = uoff, pgaddr = addr, pl = plarr;
4500 	    pgoff < eoff; /* empty */) {
4501 		page_t	*pp;
4502 		u_offset_t	nextrio;
4503 		se_t	se;
4504 		int retval;
4505 
4506 		se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED);
4507 
4508 		/* Handle async getpage (faultahead) */
4509 		if (plarr == NULL) {
4510 			ip->i_nextrio = pgoff;
4511 			(void) ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4512 			pgoff += pgsize;
4513 			pgaddr += pgsize;
4514 			continue;
4515 		}
4516 		/*
4517 		 * Check if we should initiate read ahead of next cluster.
4518 		 * We call page_exists only when we need to confirm that
4519 		 * we have the current page before we initiate the read ahead.
4520 		 */
4521 		nextrio = ip->i_nextrio;
4522 		if (seqmode &&
4523 		    pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
4524 		    nextrio < ip->i_size && page_exists(vp, pgoff)) {
4525 			retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4526 			/*
4527 			 * We always read ahead the next cluster of data
4528 			 * starting from i_nextrio. If the page (vp,nextrio)
4529 			 * is actually in core at this point, the routine
4530 			 * ufs_getpage_ra() will stop pre-fetching data
4531 			 * until we read that page in a synchronized manner
4532 			 * through ufs_getpage_miss(). So, we should increase
4533 			 * i_nextrio if the page (vp, nextrio) exists.
4534 			 */
4535 			if ((retval == 0) && page_exists(vp, nextrio)) {
4536 				ip->i_nextrio = nextrio + pgsize;
4537 			}
4538 		}
4539 
4540 		if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
4541 			/*
4542 			 * We found the page in the page cache.
4543 			 */
4544 			*pl++ = pp;
4545 			pgoff += pgsize;
4546 			pgaddr += pgsize;
4547 			len -= pgsize;
4548 			plsz -= pgsize;
4549 		} else  {
4550 			/*
4551 			 * We have to create the page, or read it from disk.
4552 			 */
4553 			if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr,
4554 			    pl, plsz, rw, seqmode))
4555 				goto error;
4556 
4557 			while (*pl != NULL) {
4558 				pl++;
4559 				pgoff += pgsize;
4560 				pgaddr += pgsize;
4561 				len -= pgsize;
4562 				plsz -= pgsize;
4563 			}
4564 		}
4565 	}
4566 
4567 	/*
4568 	 * Return pages up to plsz if they are in the page cache.
4569 	 * We cannot return pages if there is a chance that they are
4570 	 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4571 	 */
4572 	if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
4573 
4574 		ASSERT((protp == NULL) ||
4575 			!(has_holes && (*protp & PROT_WRITE)));
4576 
4577 		eoff = pgoff + plsz;
4578 		while (pgoff < eoff) {
4579 			page_t		*pp;
4580 
4581 			if ((pp = page_lookup_nowait(vp, pgoff,
4582 			    SE_SHARED)) == NULL)
4583 				break;
4584 
4585 			*pl++ = pp;
4586 			pgoff += pgsize;
4587 			plsz -= pgsize;
4588 		}
4589 	}
4590 
4591 	if (plarr)
4592 		*pl = NULL;			/* Terminate page list */
4593 	ip->i_nextr = pgoff;
4594 
4595 error:
4596 	if (err && plarr) {
4597 		/*
4598 		 * Release any pages we have locked.
4599 		 */
4600 		while (pl > &plarr[0])
4601 			page_unlock(*--pl);
4602 
4603 		plarr[0] = NULL;
4604 	}
4605 
4606 update_inode:
4607 	/*
4608 	 * If the inode is not already marked for IACC (in rdip() for read)
4609 	 * and the inode is not marked for no access time update (in wrip()
4610 	 * for write) then update the inode access time and mod time now.
4611 	 */
4612 	if ((ip->i_flag & (IACC | INOACC)) == 0) {
4613 		if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) {
4614 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
4615 			    (fs->fs_ronly == 0) &&
4616 			    (!ufsvfsp->vfs_noatime)) {
4617 				mutex_enter(&ip->i_tlock);
4618 				ip->i_flag |= IACC;
4619 				ITIMES_NOLOCK(ip);
4620 				mutex_exit(&ip->i_tlock);
4621 			}
4622 		}
4623 	}
4624 
4625 	if (dolock) {
4626 		rw_exit(&ip->i_contents);
4627 		if (do_qlock && rwtype == RW_WRITER)
4628 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4629 	}
4630 
4631 unlock:
4632 	if (ulp) {
4633 		if ((rw == S_CREATE || rw == S_WRITE) &&
4634 		    !(vp->v_flag & VISSWAP)) {
4635 			TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4636 		}
4637 		ufs_lockfs_end(ulp);
4638 	}
4639 out:
4640 	TRACE_2(TR_FAC_UFS, TR_UFS_GETPAGE_END,
4641 		"ufs_getpage_end:vp %p error %d", vp, err);
4642 	return (err);
4643 }
4644 
4645 /*
4646  * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4647  * cache. The page is either read from the disk, or it's created.
4648  * A page is created (without disk read) if rw == S_CREATE, or if
4649  * the page is not backed with a real disk block (UFS hole).
4650  */
4651 /* ARGSUSED */
4652 static int
4653 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg,
4654 	caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq)
4655 {
4656 	struct inode	*ip = VTOI(vp);
4657 	page_t		*pp;
4658 	daddr_t		bn;
4659 	size_t		io_len;
4660 	int		crpage;
4661 	int		err;
4662 	int		contig;
4663 	int		bsize = ip->i_fs->fs_bsize;
4664 
4665 	/*
4666 	 * Figure out whether the page can be created, or must be
4667 	 * must be read from the disk.
4668 	 */
4669 	if (rw == S_CREATE)
4670 		crpage = 1;
4671 	else {
4672 		contig = 0;
4673 		if (err = bmap_read(ip, off, &bn, &contig))
4674 			return (err);
4675 		crpage = (bn == UFS_HOLE);
4676 	}
4677 
4678 	if (crpage) {
4679 		if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg,
4680 		    addr)) == NULL) {
4681 			return (ufs_fault(vp,
4682 				    "ufs_getpage_miss: page_create == NULL"));
4683 		}
4684 
4685 		if (rw != S_CREATE)
4686 			pagezero(pp, 0, PAGESIZE);
4687 		io_len = PAGESIZE;
4688 	} else {
4689 		u_offset_t	io_off;
4690 		uint_t	xlen;
4691 		struct buf	*bp;
4692 		ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
4693 
4694 		/*
4695 		 * If access is not in sequential order, we read from disk
4696 		 * in bsize units.
4697 		 *
4698 		 * We limit the size of the transfer to bsize if we are reading
4699 		 * from the beginning of the file. Note in this situation we
4700 		 * will hedge our bets and initiate an async read ahead of
4701 		 * the second block.
4702 		 */
4703 		if (!seq || off == 0)
4704 			contig = MIN(contig, bsize);
4705 
4706 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4707 		    &io_len, off, contig, 0);
4708 
4709 		/*
4710 		 * Some other thread has entered the page.
4711 		 * ufs_getpage will retry page_lookup.
4712 		 */
4713 		if (pp == NULL) {
4714 			pl[0] = NULL;
4715 			return (0);
4716 		}
4717 
4718 		/*
4719 		 * Zero part of the page which we are not
4720 		 * going to read from the disk.
4721 		 */
4722 		xlen = io_len & PAGEOFFSET;
4723 		if (xlen != 0)
4724 			pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4725 
4726 		bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ);
4727 		bp->b_edev = ip->i_dev;
4728 		bp->b_dev = cmpdev(ip->i_dev);
4729 		bp->b_blkno = bn;
4730 		bp->b_un.b_addr = (caddr_t)0;
4731 		bp->b_file = ip->i_vnode;
4732 		bp->b_offset = off;
4733 
4734 		if (ufsvfsp->vfs_log) {
4735 			lufs_read_strategy(ufsvfsp->vfs_log, bp);
4736 		} else if (ufsvfsp->vfs_snapshot) {
4737 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4738 		} else {
4739 			ufsvfsp->vfs_iotstamp = lbolt;
4740 			ub.ub_getpages.value.ul++;
4741 			(void) bdev_strategy(bp);
4742 			lwp_stat_update(LWP_STAT_INBLK, 1);
4743 		}
4744 
4745 		ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK);
4746 
4747 		/*
4748 		 * If the file access is sequential, initiate read ahead
4749 		 * of the next cluster.
4750 		 */
4751 		if (seq && ip->i_nextrio < ip->i_size)
4752 			(void) ufs_getpage_ra(vp, off, seg, addr);
4753 		err = biowait(bp);
4754 		pageio_done(bp);
4755 
4756 		if (err) {
4757 			pvn_read_done(pp, B_ERROR);
4758 			return (err);
4759 		}
4760 	}
4761 
4762 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4763 	return (0);
4764 }
4765 
4766 /*
4767  * Read ahead a cluster from the disk. Returns the length in bytes.
4768  */
4769 static int
4770 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr)
4771 {
4772 	struct inode	*ip = VTOI(vp);
4773 	page_t		*pp;
4774 	u_offset_t	io_off = ip->i_nextrio;
4775 	ufsvfs_t	*ufsvfsp;
4776 	caddr_t		addr2 = addr + (io_off - off);
4777 	struct buf	*bp;
4778 	daddr_t		bn;
4779 	size_t		io_len;
4780 	int		contig;
4781 	int		xlen;
4782 	int		bsize = ip->i_fs->fs_bsize;
4783 
4784 	/*
4785 	 * If the directio advisory is in effect on this file,
4786 	 * then do not do buffered read ahead. Read ahead makes
4787 	 * it more difficult on threads using directio as they
4788 	 * will be forced to flush the pages from this vnode.
4789 	 */
4790 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
4791 		return (0);
4792 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio)
4793 		return (0);
4794 
4795 	/*
4796 	 * Is this test needed?
4797 	 */
4798 	if (addr2 >= seg->s_base + seg->s_size)
4799 		return (0);
4800 
4801 	contig = 0;
4802 	if (bmap_read(ip, io_off, &bn, &contig) != 0 || bn == UFS_HOLE)
4803 		return (0);
4804 
4805 	/*
4806 	 * Limit the transfer size to bsize if this is the 2nd block.
4807 	 */
4808 	if (io_off == (u_offset_t)bsize)
4809 		contig = MIN(contig, bsize);
4810 
4811 	if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off,
4812 	    &io_len, io_off, contig, 1)) == NULL)
4813 		return (0);
4814 
4815 	/*
4816 	 * Zero part of page which we are not going to read from disk
4817 	 */
4818 	if ((xlen = (io_len & PAGEOFFSET)) > 0)
4819 		pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4820 
4821 	ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK;
4822 
4823 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC);
4824 	bp->b_edev = ip->i_dev;
4825 	bp->b_dev = cmpdev(ip->i_dev);
4826 	bp->b_blkno = bn;
4827 	bp->b_un.b_addr = (caddr_t)0;
4828 	bp->b_file = ip->i_vnode;
4829 	bp->b_offset = off;
4830 
4831 	if (ufsvfsp->vfs_log) {
4832 		lufs_read_strategy(ufsvfsp->vfs_log, bp);
4833 	} else if (ufsvfsp->vfs_snapshot) {
4834 		fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4835 	} else {
4836 		ufsvfsp->vfs_iotstamp = lbolt;
4837 		ub.ub_getras.value.ul++;
4838 		(void) bdev_strategy(bp);
4839 		lwp_stat_update(LWP_STAT_INBLK, 1);
4840 	}
4841 
4842 	return (io_len);
4843 }
4844 
4845 int	ufs_delay = 1;
4846 /*
4847  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
4848  *
4849  * LMXXX - the inode really ought to contain a pointer to one of these
4850  * async args.  Stuff gunk in there and just hand the whole mess off.
4851  * This would replace i_delaylen, i_delayoff.
4852  */
4853 /*ARGSUSED*/
4854 static int
4855 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
4856 	struct cred *cr)
4857 {
4858 	struct inode *ip = VTOI(vp);
4859 	int err = 0;
4860 
4861 	if (vp->v_count == 0) {
4862 		return (ufs_fault(vp, "ufs_putpage: bad v_count == 0"));
4863 	}
4864 
4865 	TRACE_1(TR_FAC_UFS, TR_UFS_PUTPAGE_START,
4866 		"ufs_putpage_start:vp %p", vp);
4867 
4868 	/*
4869 	 * XXX - Why should this check be made here?
4870 	 */
4871 	if (vp->v_flag & VNOMAP) {
4872 		err = ENOSYS;
4873 		goto errout;
4874 	}
4875 
4876 	if (ip->i_ufsvfs == NULL) {
4877 		err = EIO;
4878 		goto errout;
4879 	}
4880 
4881 	if (flags & B_ASYNC) {
4882 		if (ufs_delay && len &&
4883 		    (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
4884 			mutex_enter(&ip->i_tlock);
4885 			/*
4886 			 * If nobody stalled, start a new cluster.
4887 			 */
4888 			if (ip->i_delaylen == 0) {
4889 				ip->i_delayoff = off;
4890 				ip->i_delaylen = len;
4891 				mutex_exit(&ip->i_tlock);
4892 				goto errout;
4893 			}
4894 			/*
4895 			 * If we have a full cluster or they are not contig,
4896 			 * then push last cluster and start over.
4897 			 */
4898 			if (ip->i_delaylen >= CLUSTSZ(ip) ||
4899 			    ip->i_delayoff + ip->i_delaylen != off) {
4900 				u_offset_t doff;
4901 				size_t dlen;
4902 
4903 				doff = ip->i_delayoff;
4904 				dlen = ip->i_delaylen;
4905 				ip->i_delayoff = off;
4906 				ip->i_delaylen = len;
4907 				mutex_exit(&ip->i_tlock);
4908 				err = ufs_putpages(vp, doff, dlen,
4909 				    flags, cr);
4910 				/* LMXXX - flags are new val, not old */
4911 				goto errout;
4912 			}
4913 			/*
4914 			 * There is something there, it's not full, and
4915 			 * it is contig.
4916 			 */
4917 			ip->i_delaylen += len;
4918 			mutex_exit(&ip->i_tlock);
4919 			goto errout;
4920 		}
4921 		/*
4922 		 * Must have weird flags or we are not clustering.
4923 		 */
4924 	}
4925 
4926 	err = ufs_putpages(vp, off, len, flags, cr);
4927 
4928 errout:
4929 	TRACE_2(TR_FAC_UFS, TR_UFS_PUTPAGE_END,
4930 		"ufs_putpage_end:vp %p error %d", vp, err);
4931 	return (err);
4932 }
4933 
4934 /*
4935  * If len == 0, do from off to EOF.
4936  *
4937  * The normal cases should be len == 0 & off == 0 (entire vp list),
4938  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
4939  * (from pageout).
4940  */
4941 /*ARGSUSED*/
4942 static int
4943 ufs_putpages(
4944 	struct vnode *vp,
4945 	offset_t off,
4946 	size_t len,
4947 	int flags,
4948 	struct cred *cr)
4949 {
4950 	u_offset_t io_off;
4951 	u_offset_t eoff;
4952 	struct inode *ip = VTOI(vp);
4953 	page_t *pp;
4954 	size_t io_len;
4955 	int err = 0;
4956 	int dolock;
4957 
4958 	if (vp->v_count == 0)
4959 		return (ufs_fault(vp, "ufs_putpages: v_count == 0"));
4960 	/*
4961 	 * Acquire the readers/write inode lock before locking
4962 	 * any pages in this inode.
4963 	 * The inode lock is held during i/o.
4964 	 */
4965 	if (len == 0) {
4966 		mutex_enter(&ip->i_tlock);
4967 		ip->i_delayoff = ip->i_delaylen = 0;
4968 		mutex_exit(&ip->i_tlock);
4969 	}
4970 	dolock = (rw_owner(&ip->i_contents) != curthread);
4971 	if (dolock) {
4972 		/*
4973 		 * Must synchronize this thread and any possible thread
4974 		 * operating in the window of vulnerability in wrip().
4975 		 * It is dangerous to allow both a thread doing a putpage
4976 		 * and a thread writing, so serialize them.  The exception
4977 		 * is when the thread in wrip() does something which causes
4978 		 * a putpage operation.  Then, the thread must be allowed
4979 		 * to continue.  It may encounter a bmap_read problem in
4980 		 * ufs_putapage, but that is handled in ufs_putapage.
4981 		 * Allow async writers to proceed, we don't want to block
4982 		 * the pageout daemon.
4983 		 */
4984 		if (ip->i_writer == curthread)
4985 			rw_enter(&ip->i_contents, RW_READER);
4986 		else {
4987 			for (;;) {
4988 				rw_enter(&ip->i_contents, RW_READER);
4989 				mutex_enter(&ip->i_tlock);
4990 				/*
4991 				 * If there is no thread in the critical
4992 				 * section of wrip(), then proceed.
4993 				 * Otherwise, wait until there isn't one.
4994 				 */
4995 				if (ip->i_writer == NULL) {
4996 					mutex_exit(&ip->i_tlock);
4997 					break;
4998 				}
4999 				rw_exit(&ip->i_contents);
5000 				/*
5001 				 * Bounce async writers when we have a writer
5002 				 * working on this file so we don't deadlock
5003 				 * the pageout daemon.
5004 				 */
5005 				if (flags & B_ASYNC) {
5006 					mutex_exit(&ip->i_tlock);
5007 					return (0);
5008 				}
5009 				cv_wait(&ip->i_wrcv, &ip->i_tlock);
5010 				mutex_exit(&ip->i_tlock);
5011 			}
5012 		}
5013 	}
5014 
5015 	if (!vn_has_cached_data(vp)) {
5016 		if (dolock)
5017 			rw_exit(&ip->i_contents);
5018 		return (0);
5019 	}
5020 
5021 	if (len == 0) {
5022 		/*
5023 		 * Search the entire vp list for pages >= off.
5024 		 */
5025 		err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage,
5026 					flags, cr);
5027 	} else {
5028 		/*
5029 		 * Loop over all offsets in the range looking for
5030 		 * pages to deal with.
5031 		 */
5032 		if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0)
5033 			eoff = MIN(off + len, eoff);
5034 		else
5035 			eoff = off + len;
5036 
5037 		for (io_off = off; io_off < eoff; io_off += io_len) {
5038 			/*
5039 			 * If we are not invalidating, synchronously
5040 			 * freeing or writing pages, use the routine
5041 			 * page_lookup_nowait() to prevent reclaiming
5042 			 * them from the free list.
5043 			 */
5044 			if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
5045 				pp = page_lookup(vp, io_off,
5046 					(flags & (B_INVAL | B_FREE)) ?
5047 					    SE_EXCL : SE_SHARED);
5048 			} else {
5049 				pp = page_lookup_nowait(vp, io_off,
5050 					(flags & B_FREE) ? SE_EXCL : SE_SHARED);
5051 			}
5052 
5053 			if (pp == NULL || pvn_getdirty(pp, flags) == 0)
5054 				io_len = PAGESIZE;
5055 			else {
5056 				u_offset_t *io_offp = &io_off;
5057 
5058 				err = ufs_putapage(vp, pp, io_offp, &io_len,
5059 				    flags, cr);
5060 				if (err != 0)
5061 					break;
5062 				/*
5063 				 * "io_off" and "io_len" are returned as
5064 				 * the range of pages we actually wrote.
5065 				 * This allows us to skip ahead more quickly
5066 				 * since several pages may've been dealt
5067 				 * with by this iteration of the loop.
5068 				 */
5069 			}
5070 		}
5071 	}
5072 	if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
5073 		/*
5074 		 * We have just sync'ed back all the pages on
5075 		 * the inode, turn off the IMODTIME flag.
5076 		 */
5077 		mutex_enter(&ip->i_tlock);
5078 		ip->i_flag &= ~IMODTIME;
5079 		mutex_exit(&ip->i_tlock);
5080 	}
5081 	if (dolock)
5082 		rw_exit(&ip->i_contents);
5083 	return (err);
5084 }
5085 
5086 static void
5087 ufs_iodone(buf_t *bp)
5088 {
5089 	struct inode *ip;
5090 
5091 	ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
5092 
5093 	bp->b_iodone = NULL;
5094 
5095 	ip = VTOI(bp->b_pages->p_vnode);
5096 
5097 	mutex_enter(&ip->i_tlock);
5098 	if (ip->i_writes >= ufs_LW) {
5099 		if ((ip->i_writes -= bp->b_bcount) <= ufs_LW)
5100 			if (ufs_WRITES)
5101 				cv_broadcast(&ip->i_wrcv); /* wake all up */
5102 	} else {
5103 		ip->i_writes -= bp->b_bcount;
5104 	}
5105 
5106 	mutex_exit(&ip->i_tlock);
5107 	iodone(bp);
5108 }
5109 
5110 /*
5111  * Write out a single page, possibly klustering adjacent
5112  * dirty pages.  The inode lock must be held.
5113  *
5114  * LMXXX - bsize < pagesize not done.
5115  */
5116 /*ARGSUSED*/
5117 int
5118 ufs_putapage(
5119 	struct vnode *vp,
5120 	page_t *pp,
5121 	u_offset_t *offp,
5122 	size_t *lenp,		/* return values */
5123 	int flags,
5124 	struct cred *cr)
5125 {
5126 	u_offset_t io_off;
5127 	u_offset_t off;
5128 	struct inode *ip = VTOI(vp);
5129 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
5130 	struct fs *fs;
5131 	struct buf *bp;
5132 	size_t io_len;
5133 	daddr_t bn;
5134 	int err;
5135 	int contig;
5136 
5137 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
5138 
5139 	TRACE_1(TR_FAC_UFS, TR_UFS_PUTAPAGE_START,
5140 		"ufs_putapage_start:vp %p", vp);
5141 
5142 	if (ufsvfsp == NULL) {
5143 		err = EIO;
5144 		goto out_trace;
5145 	}
5146 
5147 	fs = ip->i_fs;
5148 	ASSERT(fs->fs_ronly == 0);
5149 
5150 	/*
5151 	 * If the modified time on the inode has not already been
5152 	 * set elsewhere (e.g. for write/setattr) we set the time now.
5153 	 * This gives us approximate modified times for mmap'ed files
5154 	 * which are modified via stores in the user address space.
5155 	 */
5156 	if ((ip->i_flag & IMODTIME) == 0) {
5157 		mutex_enter(&ip->i_tlock);
5158 		ip->i_flag |= IUPD;
5159 		ip->i_seq++;
5160 		ITIMES_NOLOCK(ip);
5161 		mutex_exit(&ip->i_tlock);
5162 	}
5163 
5164 	/*
5165 	 * Align the request to a block boundry (for old file systems),
5166 	 * and go ask bmap() how contiguous things are for this file.
5167 	 */
5168 	off = pp->p_offset & (offset_t)fs->fs_bmask;	/* block align it */
5169 	contig = 0;
5170 	err = bmap_read(ip, off, &bn, &contig);
5171 	if (err)
5172 		goto out;
5173 	if (bn == UFS_HOLE) {			/* putpage never allocates */
5174 		/*
5175 		 * logging device is in error mode; simply return EIO
5176 		 */
5177 		if (TRANS_ISERROR(ufsvfsp)) {
5178 			err = EIO;
5179 			goto out;
5180 		}
5181 		/*
5182 		 * Oops, the thread in the window in wrip() did some
5183 		 * sort of operation which caused a putpage in the bad
5184 		 * range.  In this case, just return an error which will
5185 		 * cause the software modified bit on the page to set
5186 		 * and the page will get written out again later.
5187 		 */
5188 		if (ip->i_writer == curthread) {
5189 			err = EIO;
5190 			goto out;
5191 		}
5192 		/*
5193 		 * If the pager is trying to push a page in the bad range
5194 		 * just tell him to try again later when things are better.
5195 		 */
5196 		if (flags & B_ASYNC) {
5197 			err = EAGAIN;
5198 			goto out;
5199 		}
5200 		err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE");
5201 		goto out;
5202 	}
5203 
5204 	/*
5205 	 * Take the length (of contiguous bytes) passed back from bmap()
5206 	 * and _try_ and get a set of pages covering that extent.
5207 	 */
5208 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags);
5209 
5210 	/*
5211 	 * May have run out of memory and not clustered backwards.
5212 	 * off		p_offset
5213 	 * [  pp - 1  ][   pp   ]
5214 	 * [	block		]
5215 	 * We told bmap off, so we have to adjust the bn accordingly.
5216 	 */
5217 	if (io_off > off) {
5218 		bn += btod(io_off - off);
5219 		contig -= (io_off - off);
5220 	}
5221 
5222 	/*
5223 	 * bmap was carefull to tell us the right size so use that.
5224 	 * There might be unallocated frags at the end.
5225 	 * LMXXX - bzero the end of the page?  We must be writing after EOF.
5226 	 */
5227 	if (io_len > contig) {
5228 		ASSERT(io_len - contig < fs->fs_bsize);
5229 		io_len -= (io_len - contig);
5230 	}
5231 
5232 	/*
5233 	 * Handle the case where we are writing the last page after EOF.
5234 	 *
5235 	 * XXX - just a patch for i-mt3.
5236 	 */
5237 	if (io_len == 0) {
5238 		ASSERT(pp->p_offset >= (u_offset_t)(roundup(ip->i_size,
5239 							    PAGESIZE)));
5240 		io_len = PAGESIZE;
5241 	}
5242 
5243 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags);
5244 
5245 	ULOCKFS_SET_MOD(ITOUL(ip));
5246 
5247 	bp->b_edev = ip->i_dev;
5248 	bp->b_dev = cmpdev(ip->i_dev);
5249 	bp->b_blkno = bn;
5250 	bp->b_un.b_addr = (caddr_t)0;
5251 	bp->b_file = ip->i_vnode;
5252 
5253 	if (TRANS_ISTRANS(ufsvfsp)) {
5254 		if ((ip->i_mode & IFMT) == IFSHAD) {
5255 			TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD);
5256 		} else if (ufsvfsp->vfs_qinod == ip) {
5257 			TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR,
5258 			    0, 0);
5259 		}
5260 	}
5261 
5262 	/* write throttle */
5263 
5264 	ASSERT(bp->b_iodone == NULL);
5265 	bp->b_iodone = (int (*)())ufs_iodone;
5266 	mutex_enter(&ip->i_tlock);
5267 	ip->i_writes += bp->b_bcount;
5268 	mutex_exit(&ip->i_tlock);
5269 
5270 	if (bp->b_flags & B_ASYNC) {
5271 		if (ufsvfsp->vfs_log) {
5272 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5273 		} else if (ufsvfsp->vfs_snapshot) {
5274 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5275 		} else {
5276 			ufsvfsp->vfs_iotstamp = lbolt;
5277 			ub.ub_putasyncs.value.ul++;
5278 			(void) bdev_strategy(bp);
5279 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5280 		}
5281 	} else {
5282 		if (ufsvfsp->vfs_log) {
5283 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5284 		} else if (ufsvfsp->vfs_snapshot) {
5285 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5286 		} else {
5287 			ufsvfsp->vfs_iotstamp = lbolt;
5288 			ub.ub_putsyncs.value.ul++;
5289 			(void) bdev_strategy(bp);
5290 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5291 		}
5292 		err = biowait(bp);
5293 		pageio_done(bp);
5294 		pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
5295 	}
5296 
5297 	pp = NULL;
5298 
5299 out:
5300 	if (err != 0 && pp != NULL)
5301 		pvn_write_done(pp, B_ERROR | B_WRITE | flags);
5302 
5303 	if (offp)
5304 		*offp = io_off;
5305 	if (lenp)
5306 		*lenp = io_len;
5307 out_trace:
5308 	TRACE_2(TR_FAC_UFS, TR_UFS_PUTAPAGE_END,
5309 		"ufs_putapage_end:vp %p error %d", vp, err);
5310 	return (err);
5311 }
5312 
5313 /* ARGSUSED */
5314 static int
5315 ufs_map(struct vnode *vp,
5316 	offset_t off,
5317 	struct as *as,
5318 	caddr_t *addrp,
5319 	size_t len,
5320 	uchar_t prot,
5321 	uchar_t maxprot,
5322 	uint_t flags,
5323 	struct cred *cr)
5324 {
5325 	struct segvn_crargs vn_a;
5326 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5327 	struct ulockfs *ulp;
5328 	int error;
5329 
5330 	TRACE_1(TR_FAC_UFS, TR_UFS_MAP_START,
5331 		"ufs_map_start:vp %p", vp);
5332 
5333 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK);
5334 	if (error)
5335 		goto out;
5336 
5337 	if (vp->v_flag & VNOMAP) {
5338 		error = ENOSYS;
5339 		goto unlock;
5340 	}
5341 
5342 	if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) {
5343 		error = ENXIO;
5344 		goto unlock;
5345 	}
5346 
5347 	if (vp->v_type != VREG) {
5348 		error = ENODEV;
5349 		goto unlock;
5350 	}
5351 
5352 	/*
5353 	 * If file is being locked, disallow mapping.
5354 	 */
5355 	if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) {
5356 		error = EAGAIN;
5357 		goto unlock;
5358 	}
5359 
5360 	as_rangelock(as);
5361 	if ((flags & MAP_FIXED) == 0) {
5362 		map_addr(addrp, len, off, 1, flags);
5363 		if (*addrp == NULL) {
5364 			as_rangeunlock(as);
5365 			error = ENOMEM;
5366 			goto unlock;
5367 		}
5368 	} else {
5369 		/*
5370 		 * User specified address - blow away any previous mappings
5371 		 */
5372 		(void) as_unmap(as, *addrp, len);
5373 	}
5374 
5375 	vn_a.vp = vp;
5376 	vn_a.offset = (u_offset_t)off;
5377 	vn_a.type = flags & MAP_TYPE;
5378 	vn_a.prot = prot;
5379 	vn_a.maxprot = maxprot;
5380 	vn_a.cred = cr;
5381 	vn_a.amp = NULL;
5382 	vn_a.flags = flags & ~MAP_TYPE;
5383 	vn_a.szc = 0;
5384 	vn_a.lgrp_mem_policy_flags = 0;
5385 
5386 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
5387 	as_rangeunlock(as);
5388 
5389 unlock:
5390 	if (ulp) {
5391 		ufs_lockfs_end(ulp);
5392 	}
5393 out:
5394 	TRACE_2(TR_FAC_UFS, TR_UFS_MAP_END,
5395 		"ufs_map_end:vp %p error %d", vp, error);
5396 	return (error);
5397 }
5398 
5399 /* ARGSUSED */
5400 static int
5401 ufs_addmap(struct vnode *vp,
5402 	offset_t off,
5403 	struct as *as,
5404 	caddr_t addr,
5405 	size_t	len,
5406 	uchar_t  prot,
5407 	uchar_t  maxprot,
5408 	uint_t    flags,
5409 	struct cred *cr)
5410 {
5411 	struct inode *ip = VTOI(vp);
5412 
5413 	if (vp->v_flag & VNOMAP) {
5414 		return (ENOSYS);
5415 	}
5416 
5417 	mutex_enter(&ip->i_tlock);
5418 	ip->i_mapcnt += btopr(len);
5419 	mutex_exit(&ip->i_tlock);
5420 	return (0);
5421 }
5422 
5423 /*ARGSUSED*/
5424 static int
5425 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5426 	size_t len, uint_t prot,  uint_t maxprot,  uint_t flags,
5427 	struct cred *cr)
5428 {
5429 	struct inode *ip = VTOI(vp);
5430 
5431 	if (vp->v_flag & VNOMAP) {
5432 		return (ENOSYS);
5433 	}
5434 
5435 	mutex_enter(&ip->i_tlock);
5436 	ip->i_mapcnt -= btopr(len); 	/* Count released mappings */
5437 	ASSERT(ip->i_mapcnt >= 0);
5438 	mutex_exit(&ip->i_tlock);
5439 	return (0);
5440 }
5441 /*
5442  * Return the answer requested to poll() for non-device files
5443  */
5444 struct pollhead ufs_pollhd;
5445 
5446 /* ARGSUSED */
5447 int
5448 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp)
5449 {
5450 	struct ufsvfs	*ufsvfsp;
5451 
5452 	*revp = 0;
5453 	ufsvfsp = VTOI(vp)->i_ufsvfs;
5454 
5455 	if (!ufsvfsp) {
5456 		*revp = POLLHUP;
5457 		goto out;
5458 	}
5459 
5460 	if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) ||
5461 	    ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) {
5462 		*revp |= POLLERR;
5463 
5464 	} else {
5465 		if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly &&
5466 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5467 			*revp |= POLLOUT;
5468 
5469 		if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly &&
5470 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5471 			*revp |= POLLWRBAND;
5472 
5473 		if (ev & POLLIN)
5474 			*revp |= POLLIN;
5475 
5476 		if (ev & POLLRDNORM)
5477 			*revp |= POLLRDNORM;
5478 
5479 		if (ev & POLLRDBAND)
5480 			*revp |= POLLRDBAND;
5481 	}
5482 
5483 	if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP)))
5484 		*revp |= POLLPRI;
5485 out:
5486 	*phpp = !any && !*revp ? &ufs_pollhd : (struct pollhead *)NULL;
5487 
5488 	return (0);
5489 }
5490 
5491 /* ARGSUSED */
5492 static int
5493 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr)
5494 {
5495 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
5496 	struct ulockfs	*ulp = NULL;
5497 	struct inode 	*sip = NULL;
5498 	int		error;
5499 	struct inode 	*ip = VTOI(vp);
5500 	int		issync;
5501 
5502 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK);
5503 	if (error)
5504 		return (error);
5505 
5506 	switch (cmd) {
5507 		/*
5508 		 * Have to handle _PC_NAME_MAX here, because the normal way
5509 		 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5510 		 * results in a lock ordering reversal between
5511 		 * ufs_lockfs_{begin,end}() and
5512 		 * ufs_thread_{suspend,continue}().
5513 		 *
5514 		 * Keep in sync with ufs_statvfs().
5515 		 */
5516 	case _PC_NAME_MAX:
5517 		*valp = MAXNAMLEN;
5518 		break;
5519 
5520 	case _PC_FILESIZEBITS:
5521 		if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
5522 			*valp = UFS_FILESIZE_BITS;
5523 		else
5524 			*valp = 32;
5525 		break;
5526 
5527 	case _PC_XATTR_EXISTS:
5528 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
5529 
5530 			error = ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR,
5531 							cr);
5532 			if (error ==  0 && sip != NULL) {
5533 				/* Start transaction */
5534 				if (ulp) {
5535 					TRANS_BEGIN_CSYNC(ufsvfsp, issync,
5536 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5537 				}
5538 				/*
5539 				 * Is directory empty
5540 				 */
5541 				rw_enter(&sip->i_rwlock, RW_WRITER);
5542 				rw_enter(&sip->i_contents, RW_WRITER);
5543 				if (ufs_xattrdirempty(sip,
5544 						sip->i_number, CRED())) {
5545 					rw_enter(&ip->i_contents, RW_WRITER);
5546 					ufs_unhook_shadow(ip, sip);
5547 					rw_exit(&ip->i_contents);
5548 
5549 					*valp = 0;
5550 
5551 				} else
5552 					*valp = 1;
5553 				rw_exit(&sip->i_contents);
5554 				rw_exit(&sip->i_rwlock);
5555 				if (ulp) {
5556 					TRANS_END_CSYNC(ufsvfsp, error, issync,
5557 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5558 				}
5559 				VN_RELE(ITOV(sip));
5560 			} else if (error == ENOENT) {
5561 				*valp = 0;
5562 				error = 0;
5563 			}
5564 		} else {
5565 			error = fs_pathconf(vp, cmd, valp, cr);
5566 		}
5567 		break;
5568 
5569 	case _PC_ACL_ENABLED:
5570 		*valp = _ACL_ACLENT_ENABLED;
5571 		break;
5572 
5573 	case _PC_MIN_HOLE_SIZE:
5574 		*valp = (ulong_t)ip->i_fs->fs_bsize;
5575 		break;
5576 
5577 	default:
5578 		error = fs_pathconf(vp, cmd, valp, cr);
5579 	}
5580 
5581 	if (ulp != NULL) {
5582 		ufs_lockfs_end(ulp);
5583 	}
5584 	return (error);
5585 }
5586 
5587 int ufs_pageio_writes, ufs_pageio_reads;
5588 
5589 /*ARGSUSED*/
5590 static int
5591 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len,
5592 	int flags, struct cred *cr)
5593 {
5594 	struct inode *ip = VTOI(vp);
5595 	struct ufsvfs *ufsvfsp;
5596 	page_t *npp = NULL, *opp = NULL, *cpp = pp;
5597 	struct buf *bp;
5598 	daddr_t bn;
5599 	size_t done_len = 0, cur_len = 0;
5600 	int err = 0;
5601 	int contig = 0;
5602 	int dolock;
5603 	int vmpss = 0;
5604 	struct ulockfs *ulp;
5605 
5606 	if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp &&
5607 	    vp->v_mpssdata != NULL) {
5608 		vmpss = 1;
5609 	}
5610 
5611 	dolock = (rw_owner(&ip->i_contents) != curthread);
5612 	/*
5613 	 * We need a better check.  Ideally, we would use another
5614 	 * vnodeops so that hlocked and forcibly unmounted file
5615 	 * systems would return EIO where appropriate and w/o the
5616 	 * need for these checks.
5617 	 */
5618 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5619 		return (EIO);
5620 
5621 	/*
5622 	 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5623 	 * ul_lock must be taken before locking pages so we can't use it here
5624 	 * if pp is non NULL because segvn already locked pages
5625 	 * SE_EXCL. Instead we rely on the fact that a forced umount or
5626 	 * applying a filesystem lock via ufs_fiolfs() will block in the
5627 	 * implicit call to ufs_flush() until we unlock the pages after the
5628 	 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5629 	 * above 0 until they are done. We have to be careful not to increment
5630 	 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5631 	 *
5632 	 * If pp is NULL use ul_lock to make sure we don't increment
5633 	 * ul_vnops_cnt after forceful unmount hlocks the file system.
5634 	 */
5635 	if (vmpss || pp == NULL) {
5636 		ulp = &ufsvfsp->vfs_ulockfs;
5637 		if (pp == NULL)
5638 			mutex_enter(&ulp->ul_lock);
5639 		if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) {
5640 			if (pp == NULL) {
5641 				mutex_exit(&ulp->ul_lock);
5642 			}
5643 			return (vmpss ? EIO : EINVAL);
5644 		}
5645 		atomic_add_long(&ulp->ul_vnops_cnt, 1);
5646 		if (pp == NULL)
5647 			mutex_exit(&ulp->ul_lock);
5648 		if (ufs_quiesce_pend) {
5649 			if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5650 				cv_broadcast(&ulp->ul_cv);
5651 			return (vmpss ? EIO : EINVAL);
5652 		}
5653 	}
5654 
5655 	if (dolock) {
5656 		/*
5657 		 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
5658 		 * handle a fault against a segment that maps vnode pages with
5659 		 * large mappings.  Segvn creates pages and holds them locked
5660 		 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
5661 		 * use rw_tryenter() to avoid a potential deadlock since in
5662 		 * lock order i_contents needs to be taken first.
5663 		 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
5664 		 */
5665 		if (!vmpss) {
5666 			rw_enter(&ip->i_contents, RW_READER);
5667 		} else if (!rw_tryenter(&ip->i_contents, RW_READER)) {
5668 			if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5669 				cv_broadcast(&ulp->ul_cv);
5670 			return (EDEADLK);
5671 		}
5672 	}
5673 
5674 	/*
5675 	 * Return an error to segvn because the pagefault request is beyond
5676 	 * PAGESIZE rounded EOF.
5677 	 */
5678 	if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) {
5679 		if (dolock)
5680 			rw_exit(&ip->i_contents);
5681 		if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5682 			cv_broadcast(&ulp->ul_cv);
5683 		return (EFAULT);
5684 	}
5685 
5686 	if (pp == NULL) {
5687 		if (bmap_has_holes(ip)) {
5688 			err = ENOSYS;
5689 		} else {
5690 			err = EINVAL;
5691 		}
5692 		if (dolock)
5693 			rw_exit(&ip->i_contents);
5694 		if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5695 			cv_broadcast(&ulp->ul_cv);
5696 		return (err);
5697 	}
5698 
5699 	/*
5700 	 * Break the io request into chunks, one for each contiguous
5701 	 * stretch of disk blocks in the target file.
5702 	 */
5703 	while (done_len < io_len) {
5704 		ASSERT(cpp);
5705 		contig = 0;
5706 		if (err = bmap_read(ip, (u_offset_t)(io_off + done_len),
5707 				    &bn, &contig))
5708 			break;
5709 
5710 		if (bn == UFS_HOLE) {	/* No holey swapfiles */
5711 			if (vmpss) {
5712 				err = EFAULT;
5713 				break;
5714 			}
5715 			err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE");
5716 			break;
5717 		}
5718 
5719 		cur_len = MIN(io_len - done_len, contig);
5720 		/*
5721 		 * Zero out a page beyond EOF, when the last block of
5722 		 * a file is a UFS fragment so that ufs_pageio() can be used
5723 		 * instead of ufs_getpage() to handle faults against
5724 		 * segvn segments that use large pages.
5725 		 */
5726 		page_list_break(&cpp, &npp, btopr(cur_len));
5727 		if ((flags & B_READ) && (cur_len & PAGEOFFSET)) {
5728 			size_t xlen = cur_len & PAGEOFFSET;
5729 			pagezero(cpp->p_prev, xlen, PAGESIZE - xlen);
5730 		}
5731 
5732 		bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
5733 		ASSERT(bp != NULL);
5734 
5735 		bp->b_edev = ip->i_dev;
5736 		bp->b_dev = cmpdev(ip->i_dev);
5737 		bp->b_blkno = bn;
5738 		bp->b_un.b_addr = (caddr_t)0;
5739 		bp->b_file = ip->i_vnode;
5740 
5741 		ufsvfsp->vfs_iotstamp = lbolt;
5742 		ub.ub_pageios.value.ul++;
5743 		if (ufsvfsp->vfs_snapshot)
5744 			fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp);
5745 		else
5746 			(void) bdev_strategy(bp);
5747 
5748 		if (flags & B_READ)
5749 			ufs_pageio_reads++;
5750 		else
5751 			ufs_pageio_writes++;
5752 		if (flags & B_READ)
5753 			lwp_stat_update(LWP_STAT_INBLK, 1);
5754 		else
5755 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5756 		/*
5757 		 * If the request is not B_ASYNC, wait for i/o to complete
5758 		 * and re-assemble the page list to return to the caller.
5759 		 * If it is B_ASYNC we leave the page list in pieces and
5760 		 * cleanup() will dispose of them.
5761 		 */
5762 		if ((flags & B_ASYNC) == 0) {
5763 			err = biowait(bp);
5764 			pageio_done(bp);
5765 			if (err)
5766 				break;
5767 			page_list_concat(&opp, &cpp);
5768 		}
5769 		cpp = npp;
5770 		npp = NULL;
5771 		if (flags & B_READ)
5772 			cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t);
5773 		done_len += cur_len;
5774 	}
5775 	ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len));
5776 	if (err) {
5777 		if (flags & B_ASYNC) {
5778 			/* Cleanup unprocessed parts of list */
5779 			page_list_concat(&cpp, &npp);
5780 			if (flags & B_READ)
5781 				pvn_read_done(cpp, B_ERROR);
5782 			else
5783 				pvn_write_done(cpp, B_ERROR);
5784 		} else {
5785 			/* Re-assemble list and let caller clean up */
5786 			page_list_concat(&opp, &cpp);
5787 			page_list_concat(&opp, &npp);
5788 		}
5789 	}
5790 
5791 	if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) &&
5792 	    ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) {
5793 		mutex_enter(&ip->i_tlock);
5794 		ip->i_flag |= IACC;
5795 		ITIMES_NOLOCK(ip);
5796 		mutex_exit(&ip->i_tlock);
5797 	}
5798 
5799 	if (dolock)
5800 		rw_exit(&ip->i_contents);
5801 	if (vmpss && !atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5802 		cv_broadcast(&ulp->ul_cv);
5803 	return (err);
5804 }
5805 
5806 /*
5807  * Called when the kernel is in a frozen state to dump data
5808  * directly to the device. It uses a private dump data structure,
5809  * set up by dump_ctl, to locate the correct disk block to which to dump.
5810  */
5811 static int
5812 ufs_dump(vnode_t *vp, caddr_t addr, int ldbn, int dblks)
5813 {
5814 	u_offset_t	file_size;
5815 	struct inode    *ip = VTOI(vp);
5816 	struct fs	*fs = ip->i_fs;
5817 	daddr_t		dbn, lfsbn;
5818 	int		disk_blks = fs->fs_bsize >> DEV_BSHIFT;
5819 	int		error = 0;
5820 	int		ndbs, nfsbs;
5821 
5822 	/*
5823 	 * forced unmount case
5824 	 */
5825 	if (ip->i_ufsvfs == NULL)
5826 		return (EIO);
5827 	/*
5828 	 * Validate the inode that it has not been modified since
5829 	 * the dump structure is allocated.
5830 	 */
5831 	mutex_enter(&ip->i_tlock);
5832 	if ((dump_info == NULL) ||
5833 	    (dump_info->ip != ip) ||
5834 	    (dump_info->time.tv_sec != ip->i_mtime.tv_sec) ||
5835 	    (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) {
5836 		mutex_exit(&ip->i_tlock);
5837 		return (-1);
5838 	}
5839 	mutex_exit(&ip->i_tlock);
5840 
5841 	/*
5842 	 * See that the file has room for this write
5843 	 */
5844 	UFS_GET_ISIZE(&file_size, ip);
5845 
5846 	if (ldbtob((offset_t)(ldbn + dblks)) > file_size)
5847 		return (ENOSPC);
5848 
5849 	/*
5850 	 * Find the physical disk block numbers from the dump
5851 	 * private data structure directly and write out the data
5852 	 * in contiguous block lumps
5853 	 */
5854 	while (dblks > 0 && !error) {
5855 		lfsbn = (daddr_t)lblkno(fs, ldbtob((offset_t)ldbn));
5856 		dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks;
5857 		nfsbs = 1;
5858 		ndbs = disk_blks - ldbn % disk_blks;
5859 		while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn +
5860 		    nfsbs]) == dbn + ndbs) {
5861 			nfsbs++;
5862 			ndbs += disk_blks;
5863 		}
5864 		if (ndbs > dblks)
5865 			ndbs = dblks;
5866 		error = bdev_dump(ip->i_dev, addr, dbn, ndbs);
5867 		addr += ldbtob((offset_t)ndbs);
5868 		dblks -= ndbs;
5869 		ldbn += ndbs;
5870 	}
5871 	return (error);
5872 
5873 }
5874 
5875 /*
5876  * Prepare the file system before and after the dump operation.
5877  *
5878  * action = DUMP_ALLOC:
5879  * Preparation before dump, allocate dump private data structure
5880  * to hold all the direct and indirect block info for dump.
5881  *
5882  * action = DUMP_FREE:
5883  * Clean up after dump, deallocate the dump private data structure.
5884  *
5885  * action = DUMP_SCAN:
5886  * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
5887  * if found, the starting file-relative DEV_BSIZE lbn is written
5888  * to *bklp; that lbn is intended for use with VOP_DUMP()
5889  */
5890 static int
5891 ufs_dumpctl(vnode_t *vp, int action, int *blkp)
5892 {
5893 	struct inode	*ip = VTOI(vp);
5894 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
5895 	struct fs	*fs;
5896 	daddr32_t	*dblk, *storeblk;
5897 	daddr32_t	*nextblk, *endblk;
5898 	struct buf	*bp;
5899 	int		i, entry, entries;
5900 	int		n, ncontig;
5901 
5902 	/*
5903 	 * check for forced unmount
5904 	 */
5905 	if (ufsvfsp == NULL)
5906 		return (EIO);
5907 
5908 	if (action == DUMP_ALLOC) {
5909 		/*
5910 		 * alloc and record dump_info
5911 		 */
5912 		if (dump_info != NULL)
5913 			return (EINVAL);
5914 
5915 		ASSERT(vp->v_type == VREG);
5916 		fs = ufsvfsp->vfs_fs;
5917 
5918 		rw_enter(&ip->i_contents, RW_READER);
5919 
5920 		if (bmap_has_holes(ip)) {
5921 			rw_exit(&ip->i_contents);
5922 			return (EFAULT);
5923 		}
5924 
5925 		/*
5926 		 * calculate and allocate space needed according to i_size
5927 		 */
5928 		entries = (int)lblkno(fs, blkroundup(fs, ip->i_size));
5929 		if ((dump_info = (struct dump *)
5930 		    kmem_alloc(sizeof (struct dump) +
5931 		    (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP)) == NULL) {
5932 			    rw_exit(&ip->i_contents);
5933 			    return (ENOMEM);
5934 		}
5935 
5936 		/* Start saving the info */
5937 		dump_info->fsbs = entries;
5938 		dump_info->ip = ip;
5939 		storeblk = &dump_info->dblk[0];
5940 
5941 		/* Direct Blocks */
5942 		for (entry = 0; entry < NDADDR && entry < entries; entry++)
5943 			*storeblk++ = ip->i_db[entry];
5944 
5945 		/* Indirect Blocks */
5946 		for (i = 0; i < NIADDR; i++) {
5947 			int error = 0;
5948 
5949 			bp = UFS_BREAD(ufsvfsp,
5950 				ip->i_dev, fsbtodb(fs, ip->i_ib[i]),
5951 				fs->fs_bsize);
5952 			if (bp->b_flags & B_ERROR)
5953 				error = EIO;
5954 			else {
5955 				dblk = bp->b_un.b_daddr;
5956 				if ((storeblk = save_dblks(ip, ufsvfsp,
5957 				    storeblk, dblk, i, entries)) == NULL)
5958 					error = EIO;
5959 			}
5960 
5961 			brelse(bp);
5962 
5963 			if (error != 0) {
5964 				kmem_free(dump_info, sizeof (struct dump) +
5965 				    (entries - 1) * sizeof (daddr32_t));
5966 				rw_exit(&ip->i_contents);
5967 				dump_info = NULL;
5968 				return (error);
5969 			}
5970 		}
5971 		/* and time stamp the information */
5972 		mutex_enter(&ip->i_tlock);
5973 		dump_info->time = ip->i_mtime;
5974 		mutex_exit(&ip->i_tlock);
5975 
5976 		rw_exit(&ip->i_contents);
5977 	} else if (action == DUMP_FREE) {
5978 		/*
5979 		 * free dump_info
5980 		 */
5981 		if (dump_info == NULL)
5982 			return (EINVAL);
5983 		entries = dump_info->fsbs - 1;
5984 		kmem_free(dump_info, sizeof (struct dump) +
5985 		    entries * sizeof (daddr32_t));
5986 		dump_info = NULL;
5987 	} else if (action == DUMP_SCAN) {
5988 		/*
5989 		 * scan dump_info
5990 		 */
5991 		if (dump_info == NULL)
5992 			return (EINVAL);
5993 
5994 		dblk = dump_info->dblk;
5995 		nextblk = dblk + 1;
5996 		endblk = dblk + dump_info->fsbs - 1;
5997 		fs = ufsvfsp->vfs_fs;
5998 		ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT);
5999 
6000 		/*
6001 		 * scan dblk[] entries; contig fs space is found when:
6002 		 * ((current blkno + frags per block) == next blkno)
6003 		 */
6004 		n = 0;
6005 		while (n < ncontig && dblk < endblk) {
6006 			if ((*dblk + fs->fs_frag) == *nextblk)
6007 				n++;
6008 			else
6009 				n = 0;
6010 			dblk++;
6011 			nextblk++;
6012 		}
6013 
6014 		/*
6015 		 * index is where size bytes of contig space begins;
6016 		 * conversion from index to the file's DEV_BSIZE lbn
6017 		 * is equivalent to:  (index * fs_bsize) / DEV_BSIZE
6018 		 */
6019 		if (n == ncontig) {
6020 			i = (dblk - dump_info->dblk) - ncontig;
6021 			*blkp = i << (fs->fs_bshift - DEV_BSHIFT);
6022 		} else
6023 			return (EFAULT);
6024 	}
6025 	return (0);
6026 }
6027 
6028 /*
6029  * Recursive helper function for ufs_dumpctl().  It follows the indirect file
6030  * system  blocks until it reaches the the disk block addresses, which are
6031  * then stored into the given buffer, storeblk.
6032  */
6033 static daddr32_t *
6034 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp,  daddr32_t *storeblk,
6035     daddr32_t *dblk, int level, int entries)
6036 {
6037 	struct fs	*fs = ufsvfsp->vfs_fs;
6038 	struct buf	*bp;
6039 	int		i;
6040 
6041 	if (level == 0) {
6042 		for (i = 0; i < NINDIR(fs); i++) {
6043 			if (storeblk - dump_info->dblk >= entries)
6044 				break;
6045 			*storeblk++ = dblk[i];
6046 		}
6047 		return (storeblk);
6048 	}
6049 	for (i = 0; i < NINDIR(fs); i++) {
6050 		if (storeblk - dump_info->dblk >= entries)
6051 			break;
6052 		bp = UFS_BREAD(ufsvfsp,
6053 				ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize);
6054 		if (bp->b_flags & B_ERROR) {
6055 			brelse(bp);
6056 			return (NULL);
6057 		}
6058 		storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr,
6059 		    level - 1, entries);
6060 		brelse(bp);
6061 
6062 		if (storeblk == NULL)
6063 			return (NULL);
6064 	}
6065 	return (storeblk);
6066 }
6067 
6068 /* ARGSUSED */
6069 static int
6070 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag,
6071 	struct cred *cr)
6072 {
6073 	struct inode	*ip = VTOI(vp);
6074 	struct ulockfs	*ulp;
6075 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
6076 	ulong_t		vsa_mask = vsap->vsa_mask;
6077 	int		err = EINVAL;
6078 
6079 	TRACE_3(TR_FAC_UFS, TR_UFS_GETSECATTR_START,
6080 	    "ufs_getsecattr_start:vp %p, vsap %p, flags %x", vp, vsap, flag);
6081 
6082 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6083 
6084 	/*
6085 	 * Only grab locks if needed - they're not needed to check vsa_mask
6086 	 * or if the mask contains no acl flags.
6087 	 */
6088 	if (vsa_mask != 0) {
6089 		if (err = ufs_lockfs_begin(ufsvfsp, &ulp,
6090 		    ULOCKFS_GETATTR_MASK))
6091 			return (err);
6092 
6093 		rw_enter(&ip->i_contents, RW_READER);
6094 		err = ufs_acl_get(ip, vsap, flag, cr);
6095 		rw_exit(&ip->i_contents);
6096 
6097 		if (ulp)
6098 			ufs_lockfs_end(ulp);
6099 	}
6100 	TRACE_1(TR_FAC_UFS, TR_UFS_GETSECATTR_END,
6101 	    "ufs_getsecattr_end:vp %p", vp);
6102 	return (err);
6103 }
6104 
6105 /* ARGSUSED */
6106 static int
6107 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr)
6108 {
6109 	struct inode	*ip = VTOI(vp);
6110 	struct ulockfs	*ulp = NULL;
6111 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
6112 	ulong_t		vsa_mask = vsap->vsa_mask;
6113 	int		err;
6114 	int		haverwlock = 1;
6115 	int		trans_size;
6116 	int		donetrans = 0;
6117 	int		retry = 1;
6118 
6119 
6120 	TRACE_3(TR_FAC_UFS, TR_UFS_SETSECATTR_START,
6121 	    "ufs_setsecattr_start:vp %p, vsap %p, flags %x", vp, vsap, flag);
6122 
6123 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
6124 
6125 	/* Abort now if the request is either empty or invalid. */
6126 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6127 	if ((vsa_mask == 0) ||
6128 	    ((vsap->vsa_aclentp == NULL) &&
6129 	    (vsap->vsa_dfaclentp == NULL))) {
6130 		err = EINVAL;
6131 		goto out;
6132 	}
6133 
6134 	/*
6135 	 * Following convention, if this is a directory then we acquire the
6136 	 * inode's i_rwlock after starting a UFS logging transaction;
6137 	 * otherwise, we acquire it beforehand. Since we were called (and
6138 	 * must therefore return) with the lock held, we will have to drop it,
6139 	 * and later reacquire it, if operating on a directory.
6140 	 */
6141 	if (vp->v_type == VDIR) {
6142 		rw_exit(&ip->i_rwlock);
6143 		haverwlock = 0;
6144 	} else {
6145 		/* Upgrade the lock if required. */
6146 		if (!rw_write_held(&ip->i_rwlock)) {
6147 			rw_exit(&ip->i_rwlock);
6148 			rw_enter(&ip->i_rwlock, RW_WRITER);
6149 		}
6150 	}
6151 
6152 again:
6153 	ASSERT(!(vp->v_type == VDIR && haverwlock));
6154 	if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) {
6155 		ulp = NULL;
6156 		retry = 0;
6157 		goto out;
6158 	}
6159 
6160 	/*
6161 	 * Check that the file system supports this operation. Note that
6162 	 * ufs_lockfs_begin() will have checked that the file system had
6163 	 * not been forcibly unmounted.
6164 	 */
6165 	if (ufsvfsp->vfs_fs->fs_ronly) {
6166 		err = EROFS;
6167 		goto out;
6168 	}
6169 	if (ufsvfsp->vfs_nosetsec) {
6170 		err = ENOSYS;
6171 		goto out;
6172 	}
6173 
6174 	if (ulp) {
6175 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR,
6176 			trans_size = TOP_SETSECATTR_SIZE(VTOI(vp)));
6177 		donetrans = 1;
6178 	}
6179 
6180 	if (vp->v_type == VDIR) {
6181 		rw_enter(&ip->i_rwlock, RW_WRITER);
6182 		haverwlock = 1;
6183 	}
6184 
6185 	ASSERT(haverwlock);
6186 
6187 	/* Do the actual work. */
6188 	rw_enter(&ip->i_contents, RW_WRITER);
6189 	/*
6190 	 * Suppress out of inodes messages if we will retry.
6191 	 */
6192 	if (retry)
6193 		ip->i_flag |= IQUIET;
6194 	err = ufs_acl_set(ip, vsap, flag, cr);
6195 	ip->i_flag &= ~IQUIET;
6196 	rw_exit(&ip->i_contents);
6197 
6198 out:
6199 	if (ulp) {
6200 		if (donetrans) {
6201 			/*
6202 			 * top_end_async() can eventually call
6203 			 * top_end_sync(), which can block. We must
6204 			 * therefore observe the lock-ordering protocol
6205 			 * here as well.
6206 			 */
6207 			if (vp->v_type == VDIR) {
6208 				rw_exit(&ip->i_rwlock);
6209 				haverwlock = 0;
6210 			}
6211 			TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size);
6212 		}
6213 		ufs_lockfs_end(ulp);
6214 	}
6215 	/*
6216 	 * If no inodes available, try scaring a logically-
6217 	 * free one out of the delete queue to someplace
6218 	 * that we can find it.
6219 	 */
6220 	if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
6221 		ufs_delete_drain_wait(ufsvfsp, 1);
6222 		retry = 0;
6223 		if (vp->v_type == VDIR && haverwlock) {
6224 			rw_exit(&ip->i_rwlock);
6225 			haverwlock = 0;
6226 		}
6227 		goto again;
6228 	}
6229 	/*
6230 	 * If we need to reacquire the lock then it is safe to do so
6231 	 * as a reader. This is because ufs_rwunlock(), which will be
6232 	 * called by our caller after we return, does not differentiate
6233 	 * between shared and exclusive locks.
6234 	 */
6235 	if (!haverwlock) {
6236 		ASSERT(vp->v_type == VDIR);
6237 		rw_enter(&ip->i_rwlock, RW_READER);
6238 	}
6239 
6240 	TRACE_1(TR_FAC_UFS, TR_UFS_SETSECATTR_END,
6241 	    "ufs_setsecattr_end:vp %p", vp);
6242 	return (err);
6243 }
6244