xref: /titanic_41/usr/src/uts/common/fs/ufs/ufs_vnops.c (revision 6f45ec7b0b964c3be967c4880e8867ac1e7763a5)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * Portions of this source code were derived from Berkeley 4.3 BSD
31  * under license from the Regents of the University of California.
32  */
33 
34 #pragma ident	"%Z%%M%	%I%	%E% SMI"
35 
36 #include <sys/types.h>
37 #include <sys/t_lock.h>
38 #include <sys/ksynch.h>
39 #include <sys/param.h>
40 #include <sys/time.h>
41 #include <sys/systm.h>
42 #include <sys/sysmacros.h>
43 #include <sys/resource.h>
44 #include <sys/signal.h>
45 #include <sys/cred.h>
46 #include <sys/user.h>
47 #include <sys/buf.h>
48 #include <sys/vfs.h>
49 #include <sys/vnode.h>
50 #include <sys/proc.h>
51 #include <sys/disp.h>
52 #include <sys/file.h>
53 #include <sys/fcntl.h>
54 #include <sys/flock.h>
55 #include <sys/atomic.h>
56 #include <sys/kmem.h>
57 #include <sys/uio.h>
58 #include <sys/dnlc.h>
59 #include <sys/conf.h>
60 #include <sys/mman.h>
61 #include <sys/pathname.h>
62 #include <sys/debug.h>
63 #include <sys/vmsystm.h>
64 #include <sys/cmn_err.h>
65 #include <sys/vtrace.h>
66 #include <sys/filio.h>
67 #include <sys/policy.h>
68 
69 #include <sys/fs/ufs_fs.h>
70 #include <sys/fs/ufs_lockfs.h>
71 #include <sys/fs/ufs_filio.h>
72 #include <sys/fs/ufs_inode.h>
73 #include <sys/fs/ufs_fsdir.h>
74 #include <sys/fs/ufs_quota.h>
75 #include <sys/fs/ufs_log.h>
76 #include <sys/fs/ufs_snap.h>
77 #include <sys/fs/ufs_trans.h>
78 #include <sys/fs/ufs_panic.h>
79 #include <sys/fs/ufs_bio.h>
80 #include <sys/dirent.h>		/* must be AFTER <sys/fs/fsdir.h>! */
81 #include <sys/errno.h>
82 #include <sys/fssnap_if.h>
83 #include <sys/unistd.h>
84 #include <sys/sunddi.h>
85 
86 #include <sys/filio.h>		/* _FIOIO */
87 
88 #include <vm/hat.h>
89 #include <vm/page.h>
90 #include <vm/pvn.h>
91 #include <vm/as.h>
92 #include <vm/seg.h>
93 #include <vm/seg_map.h>
94 #include <vm/seg_vn.h>
95 #include <vm/seg_kmem.h>
96 #include <vm/rm.h>
97 #include <sys/swap.h>
98 
99 #include <fs/fs_subr.h>
100 
101 static struct instats ins;
102 
103 static 	int ufs_getpage_ra(struct vnode *, u_offset_t, struct seg *, caddr_t);
104 static	int ufs_getpage_miss(struct vnode *, u_offset_t, size_t, struct seg *,
105 		caddr_t, struct page **, size_t, enum seg_rw, int);
106 static	int ufs_open(struct vnode **, int, struct cred *);
107 static	int ufs_close(struct vnode *, int, int, offset_t, struct cred *);
108 static	int ufs_read(struct vnode *, struct uio *, int, struct cred *,
109 			struct caller_context *);
110 static	int ufs_write(struct vnode *, struct uio *, int, struct cred *,
111 			struct caller_context *);
112 static	int ufs_ioctl(struct vnode *, int, intptr_t, int, struct cred *, int *);
113 static	int ufs_getattr(struct vnode *, struct vattr *, int, struct cred *);
114 static	int ufs_setattr(struct vnode *, struct vattr *, int, struct cred *,
115 			caller_context_t *);
116 static	int ufs_access(struct vnode *, int, int, struct cred *);
117 static	int ufs_lookup(struct vnode *, char *, struct vnode **,
118 		struct pathname *, int, struct vnode *, struct cred *);
119 static	int ufs_create(struct vnode *, char *, struct vattr *, enum vcexcl,
120 			int, struct vnode **, struct cred *, int);
121 static	int ufs_remove(struct vnode *, char *, struct cred *);
122 static	int ufs_link(struct vnode *, struct vnode *, char *, struct cred *);
123 static	int ufs_rename(struct vnode *, char *, struct vnode *, char *,
124 			struct cred *);
125 static	int ufs_mkdir(struct vnode *, char *, struct vattr *, struct vnode **,
126 			struct cred *);
127 static	int ufs_rmdir(struct vnode *, char *, struct vnode *, struct cred *);
128 static	int ufs_readdir(struct vnode *, struct uio *, struct cred *, int *);
129 static	int ufs_symlink(struct vnode *, char *, struct vattr *, char *,
130 			struct cred *);
131 static	int ufs_readlink(struct vnode *, struct uio *, struct cred *);
132 static	int ufs_fsync(struct vnode *, int, struct cred *);
133 static	void ufs_inactive(struct vnode *, struct cred *);
134 static	int ufs_fid(struct vnode *, struct fid *);
135 static	int ufs_rwlock(struct vnode *, int, caller_context_t *);
136 static	void ufs_rwunlock(struct vnode *, int, caller_context_t *);
137 static	int ufs_seek(struct vnode *, offset_t, offset_t *);
138 static	int ufs_frlock(struct vnode *, int, struct flock64 *, int, offset_t,
139 			struct flk_callback *, struct cred *);
140 static  int ufs_space(struct vnode *, int, struct flock64 *, int, offset_t,
141 		cred_t *, caller_context_t *);
142 static	int ufs_getpage(struct vnode *, offset_t, size_t, uint_t *,
143 		struct page **, size_t, struct seg *, caddr_t,
144 		enum seg_rw, struct cred *);
145 static	int ufs_putpage(struct vnode *, offset_t, size_t, int, struct cred *);
146 static	int ufs_putpages(struct vnode *, offset_t, size_t, int, struct cred *);
147 static	int ufs_map(struct vnode *, offset_t, struct as *, caddr_t *, size_t,
148 			uchar_t, uchar_t, uint_t, struct cred *);
149 static	int ufs_addmap(struct vnode *, offset_t, struct as *, caddr_t,  size_t,
150 			uchar_t, uchar_t, uint_t, struct cred *);
151 static	int ufs_delmap(struct vnode *, offset_t, struct as *, caddr_t,  size_t,
152 			uint_t, uint_t, uint_t, struct cred *);
153 static	int ufs_poll(vnode_t *, short, int, short *, struct pollhead **);
154 static	int ufs_dump(vnode_t *, caddr_t, int, int);
155 static	int ufs_l_pathconf(struct vnode *, int, ulong_t *, struct cred *);
156 static	int ufs_pageio(struct vnode *, struct page *, u_offset_t, size_t, int,
157 			struct cred *);
158 static	int ufs_dump(vnode_t *, caddr_t, int, int);
159 static	int ufs_dumpctl(vnode_t *, int, int *);
160 static	daddr32_t *save_dblks(struct inode *, struct ufsvfs *, daddr32_t *,
161 			daddr32_t *, int, int);
162 static	int ufs_getsecattr(struct vnode *, vsecattr_t *, int, struct cred *);
163 static	int ufs_setsecattr(struct vnode *, vsecattr_t *, int, struct cred *);
164 
165 extern int as_map_locked(struct as *, caddr_t, size_t, int ((*)()), void *);
166 
167 /*
168  * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
169  *
170  * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
171  */
172 struct vnodeops *ufs_vnodeops;
173 
174 const fs_operation_def_t ufs_vnodeops_template[] = {
175 	VOPNAME_OPEN, ufs_open,	/* will not be blocked by lockfs */
176 	VOPNAME_CLOSE, ufs_close,	/* will not be blocked by lockfs */
177 	VOPNAME_READ, ufs_read,
178 	VOPNAME_WRITE, ufs_write,
179 	VOPNAME_IOCTL, ufs_ioctl,
180 	VOPNAME_GETATTR, ufs_getattr,
181 	VOPNAME_SETATTR, ufs_setattr,
182 	VOPNAME_ACCESS, ufs_access,
183 	VOPNAME_LOOKUP, ufs_lookup,
184 	VOPNAME_CREATE, ufs_create,
185 	VOPNAME_REMOVE, ufs_remove,
186 	VOPNAME_LINK, ufs_link,
187 	VOPNAME_RENAME, ufs_rename,
188 	VOPNAME_MKDIR, ufs_mkdir,
189 	VOPNAME_RMDIR, ufs_rmdir,
190 	VOPNAME_READDIR, ufs_readdir,
191 	VOPNAME_SYMLINK, ufs_symlink,
192 	VOPNAME_READLINK, ufs_readlink,
193 	VOPNAME_FSYNC, ufs_fsync,
194 	VOPNAME_INACTIVE, (fs_generic_func_p) ufs_inactive,  /* not blocked */
195 	VOPNAME_FID, ufs_fid,
196 	VOPNAME_RWLOCK, ufs_rwlock, /* not blocked */
197 	VOPNAME_RWUNLOCK, (fs_generic_func_p) ufs_rwunlock,  /* not blocked */
198 	VOPNAME_SEEK, ufs_seek,
199 	VOPNAME_FRLOCK, ufs_frlock,
200 	VOPNAME_SPACE, ufs_space,
201 	VOPNAME_GETPAGE, ufs_getpage,
202 	VOPNAME_PUTPAGE, ufs_putpage,
203 	VOPNAME_MAP, (fs_generic_func_p) ufs_map,
204 	VOPNAME_ADDMAP, (fs_generic_func_p) ufs_addmap,	/* not blocked */
205 	VOPNAME_DELMAP, ufs_delmap,	/* will not be blocked by lockfs */
206 	VOPNAME_POLL, (fs_generic_func_p) ufs_poll,	/* not blocked */
207 	VOPNAME_DUMP, ufs_dump,
208 	VOPNAME_PATHCONF, ufs_l_pathconf,
209 	VOPNAME_PAGEIO, ufs_pageio,
210 	VOPNAME_DUMPCTL, ufs_dumpctl,
211 	VOPNAME_GETSECATTR, ufs_getsecattr,
212 	VOPNAME_SETSECATTR, ufs_setsecattr,
213 	VOPNAME_VNEVENT, fs_vnevent_support,
214 	NULL, NULL
215 };
216 
217 #define	MAX_BACKFILE_COUNT	9999
218 
219 /*
220  * Created by ufs_dumpctl() to store a file's disk block info into memory.
221  * Used by ufs_dump() to dump data to disk directly.
222  */
223 struct dump {
224 	struct inode	*ip;		/* the file we contain */
225 	daddr_t		fsbs;		/* number of blocks stored */
226 	struct timeval32 time;		/* time stamp for the struct */
227 	daddr32_t 	dblk[1];	/* place holder for block info */
228 };
229 
230 static struct dump *dump_info = NULL;
231 
232 /*
233  * Previously there was no special action required for ordinary files.
234  * (Devices are handled through the device file system.)
235  * Now we support Large Files and Large File API requires open to
236  * fail if file is large.
237  * We could take care to prevent data corruption
238  * by doing an atomic check of size and truncate if file is opened with
239  * FTRUNC flag set but traditionally this is being done by the vfs/vnode
240  * layers. So taking care of truncation here is a change in the existing
241  * semantics of VOP_OPEN and therefore we chose not to implement any thing
242  * here. The check for the size of the file > 2GB is being done at the
243  * vfs layer in routine vn_open().
244  */
245 
246 /* ARGSUSED */
247 static int
248 ufs_open(struct vnode **vpp, int flag, struct cred *cr)
249 {
250 	TRACE_1(TR_FAC_UFS, TR_UFS_OPEN, "ufs_open:vpp %p", vpp);
251 	return (0);
252 }
253 
254 /*ARGSUSED*/
255 static int
256 ufs_close(struct vnode *vp, int flag, int count, offset_t offset,
257 	struct cred *cr)
258 {
259 	TRACE_1(TR_FAC_UFS, TR_UFS_CLOSE, "ufs_close:vp %p", vp);
260 
261 	cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
262 	cleanshares(vp, ttoproc(curthread)->p_pid);
263 
264 	/*
265 	 * Push partially filled cluster at last close.
266 	 * ``last close'' is approximated because the dnlc
267 	 * may have a hold on the vnode.
268 	 * Checking for VBAD here will also act as a forced umount check.
269 	 */
270 	if (vp->v_count <= 2 && vp->v_type != VBAD) {
271 		struct inode *ip = VTOI(vp);
272 		if (ip->i_delaylen) {
273 			ins.in_poc.value.ul++;
274 			(void) ufs_putpages(vp, ip->i_delayoff, ip->i_delaylen,
275 					B_ASYNC | B_FREE, cr);
276 			ip->i_delaylen = 0;
277 		}
278 	}
279 
280 	return (0);
281 }
282 
283 /*ARGSUSED*/
284 static int
285 ufs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cr,
286 	struct caller_context *ct)
287 {
288 	struct inode *ip = VTOI(vp);
289 	struct ufsvfs *ufsvfsp;
290 	struct ulockfs *ulp = NULL;
291 	int error = 0;
292 	int intrans = 0;
293 
294 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
295 	TRACE_3(TR_FAC_UFS, TR_UFS_READ_START,
296 		"ufs_read_start:vp %p uiop %p ioflag %x",
297 		vp, uiop, ioflag);
298 
299 	/*
300 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
301 	 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
302 	 */
303 	if (MANDLOCK(vp, ip->i_mode)) {
304 		/*
305 		 * ufs_getattr ends up being called by chklock
306 		 */
307 		error = chklock(vp, FREAD, uiop->uio_loffset,
308 				uiop->uio_resid, uiop->uio_fmode, ct);
309 		if (error)
310 			goto out;
311 	}
312 
313 	ufsvfsp = ip->i_ufsvfs;
314 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READ_MASK);
315 	if (error)
316 		goto out;
317 
318 	/*
319 	 * In the case that a directory is opened for reading as a file
320 	 * (eg "cat .") with the  O_RSYNC, O_SYNC and O_DSYNC flags set.
321 	 * The locking order had to be changed to avoid a deadlock with
322 	 * an update taking place on that directory at the same time.
323 	 */
324 	if ((ip->i_mode & IFMT) == IFDIR) {
325 
326 		rw_enter(&ip->i_contents, RW_READER);
327 		error = rdip(ip, uiop, ioflag, cr);
328 		rw_exit(&ip->i_contents);
329 
330 		if (error) {
331 			if (ulp)
332 				ufs_lockfs_end(ulp);
333 			goto out;
334 		}
335 
336 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
337 		    TRANS_ISTRANS(ufsvfsp)) {
338 			rw_exit(&ip->i_rwlock);
339 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
340 			    error);
341 			ASSERT(!error);
342 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
343 			    TOP_READ_SIZE);
344 			rw_enter(&ip->i_rwlock, RW_READER);
345 		}
346 	} else {
347 		/*
348 		 * Only transact reads to files opened for sync-read and
349 		 * sync-write on a file system that is not write locked.
350 		 *
351 		 * The ``not write locked'' check prevents problems with
352 		 * enabling/disabling logging on a busy file system.  E.g.,
353 		 * logging exists at the beginning of the read but does not
354 		 * at the end.
355 		 *
356 		 */
357 		if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
358 		    TRANS_ISTRANS(ufsvfsp)) {
359 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
360 			    error);
361 			ASSERT(!error);
362 			intrans = 1;
363 		}
364 
365 		rw_enter(&ip->i_contents, RW_READER);
366 		error = rdip(ip, uiop, ioflag, cr);
367 		rw_exit(&ip->i_contents);
368 
369 		if (intrans) {
370 			TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
371 			    TOP_READ_SIZE);
372 		}
373 	}
374 
375 	if (ulp) {
376 		ufs_lockfs_end(ulp);
377 	}
378 out:
379 
380 	TRACE_2(TR_FAC_UFS, TR_UFS_READ_END,
381 		"ufs_read_end:vp %p error %d", vp, error);
382 	return (error);
383 }
384 
385 extern	int	ufs_HW;		/* high water mark */
386 extern	int	ufs_LW;		/* low water mark */
387 int	ufs_WRITES = 1;		/* XXX - enable/disable */
388 int	ufs_throttles = 0;	/* throttling count */
389 int	ufs_allow_shared_writes = 1;	/* directio shared writes */
390 
391 static int
392 ufs_check_rewrite(struct inode *ip, struct uio *uiop, int ioflag)
393 {
394 	int	shared_write;
395 
396 	/*
397 	 * If the FDSYNC flag is set then ignore the global
398 	 * ufs_allow_shared_writes in this case.
399 	 */
400 	shared_write = (ioflag & FDSYNC) | ufs_allow_shared_writes;
401 
402 	/*
403 	 * Filter to determine if this request is suitable as a
404 	 * concurrent rewrite. This write must not allocate blocks
405 	 * by extending the file or filling in holes. No use trying
406 	 * through FSYNC descriptors as the inode will be synchronously
407 	 * updated after the write. The uio structure has not yet been
408 	 * checked for sanity, so assume nothing.
409 	 */
410 	return (((ip->i_mode & IFMT) == IFREG) && !(ioflag & FAPPEND) &&
411 		(uiop->uio_loffset >= (offset_t)0) &&
412 		(uiop->uio_loffset < ip->i_size) && (uiop->uio_resid > 0) &&
413 		((ip->i_size - uiop->uio_loffset) >= uiop->uio_resid) &&
414 		!(ioflag & FSYNC) && !bmap_has_holes(ip) &&
415 		shared_write);
416 }
417 
418 /*ARGSUSED*/
419 static int
420 ufs_write(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cr,
421 	caller_context_t *ct)
422 {
423 	struct inode *ip = VTOI(vp);
424 	struct ufsvfs *ufsvfsp;
425 	struct ulockfs *ulp;
426 	int retry = 1;
427 	int error, resv, resid = 0;
428 	int directio_status;
429 	int exclusive;
430 	int rewriteflg;
431 	long start_resid = uiop->uio_resid;
432 
433 	TRACE_3(TR_FAC_UFS, TR_UFS_WRITE_START,
434 		"ufs_write_start:vp %p uiop %p ioflag %x",
435 		vp, uiop, ioflag);
436 
437 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
438 
439 retry_mandlock:
440 	/*
441 	 * Mandatory locking needs to be done before ufs_lockfs_begin()
442 	 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
443 	 * Check for forced unmounts normally done in ufs_lockfs_begin().
444 	 */
445 	if ((ufsvfsp = ip->i_ufsvfs) == NULL) {
446 		error = EIO;
447 		goto out;
448 	}
449 	if (MANDLOCK(vp, ip->i_mode)) {
450 
451 		ASSERT(RW_WRITE_HELD(&ip->i_rwlock));
452 
453 		/*
454 		 * ufs_getattr ends up being called by chklock
455 		 */
456 		error = chklock(vp, FWRITE, uiop->uio_loffset,
457 				uiop->uio_resid, uiop->uio_fmode, ct);
458 		if (error)
459 			goto out;
460 	}
461 
462 	/* i_rwlock can change in chklock */
463 	exclusive = rw_write_held(&ip->i_rwlock);
464 	rewriteflg = ufs_check_rewrite(ip, uiop, ioflag);
465 
466 	/*
467 	 * Check for fast-path special case of directio re-writes.
468 	 */
469 	if ((ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) &&
470 	    !exclusive && rewriteflg) {
471 
472 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
473 		if (error)
474 			goto out;
475 
476 		rw_enter(&ip->i_contents, RW_READER);
477 		error = ufs_directio_write(ip, uiop, ioflag, 1, cr,
478 			&directio_status);
479 		if (directio_status == DIRECTIO_SUCCESS) {
480 			uint_t i_flag_save;
481 
482 			if (start_resid != uiop->uio_resid)
483 				error = 0;
484 			/*
485 			 * Special treatment of access times for re-writes.
486 			 * If IMOD is not already set, then convert it
487 			 * to IMODACC for this operation. This defers
488 			 * entering a delta into the log until the inode
489 			 * is flushed. This mimics what is done for read
490 			 * operations and inode access time.
491 			 */
492 			mutex_enter(&ip->i_tlock);
493 			i_flag_save = ip->i_flag;
494 			ip->i_flag |= IUPD | ICHG;
495 			ip->i_seq++;
496 			ITIMES_NOLOCK(ip);
497 			if ((i_flag_save & IMOD) == 0) {
498 				ip->i_flag &= ~IMOD;
499 				ip->i_flag |= IMODACC;
500 			}
501 			mutex_exit(&ip->i_tlock);
502 			rw_exit(&ip->i_contents);
503 			if (ulp)
504 				ufs_lockfs_end(ulp);
505 			goto out;
506 		}
507 		rw_exit(&ip->i_contents);
508 		if (ulp)
509 			ufs_lockfs_end(ulp);
510 	}
511 
512 	if (!exclusive && !rw_tryupgrade(&ip->i_rwlock)) {
513 		rw_exit(&ip->i_rwlock);
514 		rw_enter(&ip->i_rwlock, RW_WRITER);
515 		/*
516 		 * Mandatory locking could have been enabled
517 		 * after dropping the i_rwlock.
518 		 */
519 		if (MANDLOCK(vp, ip->i_mode))
520 			goto retry_mandlock;
521 	}
522 
523 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
524 	if (error)
525 		goto out;
526 
527 	/*
528 	 * Amount of log space needed for this write
529 	 */
530 	if (!rewriteflg || !(ioflag & FDSYNC))
531 		TRANS_WRITE_RESV(ip, uiop, ulp, &resv, &resid);
532 
533 	/*
534 	 * Throttle writes.
535 	 */
536 	if (ufs_WRITES && (ip->i_writes > ufs_HW)) {
537 		mutex_enter(&ip->i_tlock);
538 		while (ip->i_writes > ufs_HW) {
539 			ufs_throttles++;
540 			cv_wait(&ip->i_wrcv, &ip->i_tlock);
541 		}
542 		mutex_exit(&ip->i_tlock);
543 	}
544 
545 	/*
546 	 * Enter Transaction
547 	 *
548 	 * If the write is a rewrite there is no need to open a transaction
549 	 * if the FDSYNC flag is set and not the FSYNC.  In this case just
550 	 * set the IMODACC flag to modify do the update at a later time
551 	 * thus avoiding the overhead of the logging transaction that is
552 	 * not required.
553 	 */
554 	if (ioflag & (FSYNC|FDSYNC)) {
555 		if (ulp) {
556 			if (rewriteflg) {
557 				uint_t i_flag_save;
558 
559 				rw_enter(&ip->i_contents, RW_READER);
560 				mutex_enter(&ip->i_tlock);
561 				i_flag_save = ip->i_flag;
562 				ip->i_flag |= IUPD | ICHG;
563 				ip->i_seq++;
564 				ITIMES_NOLOCK(ip);
565 				if ((i_flag_save & IMOD) == 0) {
566 					ip->i_flag &= ~IMOD;
567 					ip->i_flag |= IMODACC;
568 				}
569 				mutex_exit(&ip->i_tlock);
570 				rw_exit(&ip->i_contents);
571 			} else {
572 				int terr = 0;
573 				TRANS_BEGIN_SYNC(ufsvfsp, TOP_WRITE_SYNC, resv,
574 				    terr);
575 				ASSERT(!terr);
576 			}
577 		}
578 	} else {
579 		if (ulp)
580 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_WRITE, resv);
581 	}
582 
583 	/*
584 	 * Write the file
585 	 */
586 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
587 	rw_enter(&ip->i_contents, RW_WRITER);
588 	if ((ioflag & FAPPEND) != 0 && (ip->i_mode & IFMT) == IFREG) {
589 		/*
590 		 * In append mode start at end of file.
591 		 */
592 		uiop->uio_loffset = ip->i_size;
593 	}
594 
595 	/*
596 	 * Mild optimisation, don't call ufs_trans_write() unless we have to
597 	 * Also, suppress file system full messages if we will retry.
598 	 */
599 	if (retry)
600 		ip->i_flag |= IQUIET;
601 	if (resid) {
602 		TRANS_WRITE(ip, uiop, ioflag, error, ulp, cr, resv, resid);
603 	} else {
604 		error = wrip(ip, uiop, ioflag, cr);
605 	}
606 	ip->i_flag &= ~IQUIET;
607 
608 	rw_exit(&ip->i_contents);
609 	rw_exit(&ufsvfsp->vfs_dqrwlock);
610 
611 	/*
612 	 * Leave Transaction
613 	 */
614 	if (ulp) {
615 		if (ioflag & (FSYNC|FDSYNC)) {
616 			if (!rewriteflg) {
617 				int terr = 0;
618 
619 				TRANS_END_SYNC(ufsvfsp, terr, TOP_WRITE_SYNC,
620 					resv);
621 				if (error == 0)
622 					error = terr;
623 			}
624 		} else {
625 			TRANS_END_ASYNC(ufsvfsp, TOP_WRITE, resv);
626 		}
627 		ufs_lockfs_end(ulp);
628 	}
629 out:
630 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
631 		/*
632 		 * Any blocks tied up in pending deletes?
633 		 */
634 		ufs_delete_drain_wait(ufsvfsp, 1);
635 		retry = 0;
636 		goto retry_mandlock;
637 	}
638 
639 	if (error == ENOSPC && (start_resid != uiop->uio_resid))
640 		error = 0;
641 
642 	TRACE_2(TR_FAC_UFS, TR_UFS_WRITE_END,
643 		"ufs_write_end:vp %p error %d", vp, error);
644 	return (error);
645 }
646 
647 /*
648  * Don't cache write blocks to files with the sticky bit set.
649  * Used to keep swap files from blowing the page cache on a server.
650  */
651 int stickyhack = 1;
652 
653 /*
654  * Free behind hacks.  The pager is busted.
655  * XXX - need to pass the information down to writedone() in a flag like B_SEQ
656  * or B_FREE_IF_TIGHT_ON_MEMORY.
657  */
658 int	freebehind = 1;
659 int	smallfile = 0;
660 u_offset_t smallfile64 = 32 * 1024;
661 
662 /*
663  * While we should, in most cases, cache the pages for write, we
664  * may also want to cache the pages for read as long as they are
665  * frequently re-usable.
666  *
667  * If cache_read_ahead = 1, the pages for read will go to the tail
668  * of the cache list when they are released, otherwise go to the head.
669  */
670 int	cache_read_ahead = 0;
671 
672 /*
673  * Freebehind exists  so that as we read  large files  sequentially we
674  * don't consume most of memory with pages  from a few files. It takes
675  * longer to re-read from disk multiple small files as it does reading
676  * one large one sequentially.  As system  memory grows customers need
677  * to retain bigger chunks   of files in  memory.   The advent of  the
678  * cachelist opens up of the possibility freeing pages  to the head or
679  * tail of the list.
680  *
681  * Not freeing a page is a bet that the page will be read again before
682  * it's segmap slot is needed for something else. If we loose the bet,
683  * it means some  other thread is  burdened with the  page free we did
684  * not do. If we win we save a free and reclaim.
685  *
686  * Freeing it at the tail  vs the head of cachelist  is a bet that the
687  * page will survive until the next  read.  It's also saying that this
688  * page is more likely to  be re-used than a  page freed some time ago
689  * and never reclaimed.
690  *
691  * Freebehind maintains a  range of  file offset [smallfile1; smallfile2]
692  *
693  *            0 < offset < smallfile1 : pages are not freed.
694  *   smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
695  *   smallfile2 < offset              : pages freed to head of cachelist.
696  *
697  * The range  is  computed  at most  once  per second  and  depends on
698  * freemem  and  ncpus_online.  Both parameters  are   bounded to be
699  * >= smallfile && >= smallfile64.
700  *
701  * smallfile1 = (free memory / ncpu) / 1000
702  * smallfile2 = (free memory / ncpu) / 10
703  *
704  * A few examples values:
705  *
706  *       Free Mem (in Bytes) [smallfile1; smallfile2]  [smallfile1; smallfile2]
707  *                                 ncpus_online = 4          ncpus_online = 64
708  *       ------------------  -----------------------   -----------------------
709  *             1G                   [256K;  25M]               [32K; 1.5M]
710  *            10G                   [2.5M; 250M]              [156K; 15M]
711  *           100G                    [25M; 2.5G]              [1.5M; 150M]
712  *
713  */
714 
715 #define	SMALLFILE1_D 1000
716 #define	SMALLFILE2_D 10
717 static u_offset_t smallfile1 = 32 * 1024;
718 static u_offset_t smallfile2 = 32 * 1024;
719 static clock_t smallfile_update = 0; /* lbolt value of when to recompute */
720 uint_t smallfile1_d = SMALLFILE1_D;
721 uint_t smallfile2_d = SMALLFILE2_D;
722 
723 /*
724  * wrip does the real work of write requests for ufs.
725  */
726 int
727 wrip(struct inode *ip, struct uio *uio, int ioflag, struct cred *cr)
728 {
729 	rlim64_t limit = uio->uio_llimit;
730 	u_offset_t off;
731 	u_offset_t old_i_size;
732 	struct fs *fs;
733 	struct vnode *vp;
734 	struct ufsvfs *ufsvfsp;
735 	caddr_t base;
736 	long start_resid = uio->uio_resid;	/* save starting resid */
737 	long premove_resid;			/* resid before uiomove() */
738 	uint_t flags;
739 	int newpage;
740 	int iupdat_flag, directio_status;
741 	int n, on, mapon;
742 	int error, pagecreate;
743 	int do_dqrwlock;		/* drop/reacquire vfs_dqrwlock */
744 	int32_t	iblocks;
745 	int	new_iblocks;
746 
747 	/*
748 	 * ip->i_size is incremented before the uiomove
749 	 * is done on a write.  If the move fails (bad user
750 	 * address) reset ip->i_size.
751 	 * The better way would be to increment ip->i_size
752 	 * only if the uiomove succeeds.
753 	 */
754 	int i_size_changed = 0;
755 	o_mode_t type;
756 	int i_seq_needed = 0;
757 
758 	vp = ITOV(ip);
759 
760 	/*
761 	 * check for forced unmount - should not happen as
762 	 * the request passed the lockfs checks.
763 	 */
764 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
765 		return (EIO);
766 
767 	fs = ip->i_fs;
768 
769 	TRACE_1(TR_FAC_UFS, TR_UFS_RWIP_START,
770 		"ufs_wrip_start:vp %p", vp);
771 
772 	ASSERT(RW_WRITE_HELD(&ip->i_contents));
773 
774 	/* check for valid filetype */
775 	type = ip->i_mode & IFMT;
776 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
777 	    (type != IFLNK) && (type != IFSHAD)) {
778 		return (EIO);
779 	}
780 
781 	/*
782 	 * the actual limit of UFS file size
783 	 * is UFS_MAXOFFSET_T
784 	 */
785 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
786 		limit = MAXOFFSET_T;
787 
788 	if (uio->uio_loffset >= limit) {
789 		proc_t *p = ttoproc(curthread);
790 
791 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
792 			"ufs_wrip_end:vp %p error %d", vp, EINVAL);
793 
794 		mutex_enter(&p->p_lock);
795 		(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
796 		    p, RCA_UNSAFE_SIGINFO);
797 		mutex_exit(&p->p_lock);
798 		return (EFBIG);
799 	}
800 
801 	/*
802 	 * if largefiles are disallowed, the limit is
803 	 * the pre-largefiles value of 2GB
804 	 */
805 	if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
806 		limit = MIN(UFS_MAXOFFSET_T, limit);
807 	else
808 		limit = MIN(MAXOFF32_T, limit);
809 
810 	if (uio->uio_loffset < (offset_t)0) {
811 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
812 			"ufs_wrip_end:vp %p error %d", vp, EINVAL);
813 		return (EINVAL);
814 	}
815 	if (uio->uio_resid == 0) {
816 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
817 			"ufs_wrip_end:vp %p error %d", vp, 0);
818 		return (0);
819 	}
820 
821 	if (uio->uio_loffset >= limit)
822 		return (EFBIG);
823 
824 	ip->i_flag |= INOACC;	/* don't update ref time in getpage */
825 
826 	if (ioflag & (FSYNC|FDSYNC)) {
827 		ip->i_flag |= ISYNC;
828 		iupdat_flag = 1;
829 	}
830 	/*
831 	 * Try to go direct
832 	 */
833 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
834 		uio->uio_llimit = limit;
835 		error = ufs_directio_write(ip, uio, ioflag, 0, cr,
836 			&directio_status);
837 		/*
838 		 * If ufs_directio wrote to the file or set the flags,
839 		 * we need to update i_seq, but it may be deferred.
840 		 */
841 		if (start_resid != uio->uio_resid ||
842 					(ip->i_flag & (ICHG|IUPD))) {
843 			i_seq_needed = 1;
844 			ip->i_flag |= ISEQ;
845 		}
846 		if (directio_status == DIRECTIO_SUCCESS)
847 			goto out;
848 	}
849 
850 	/*
851 	 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
852 	 *
853 	 * o shadow inodes: vfs_dqrwlock is not held at all
854 	 * o quota updates: vfs_dqrwlock is read or write held
855 	 * o other updates: vfs_dqrwlock is read held
856 	 *
857 	 * The first case is the only one where we do not hold
858 	 * vfs_dqrwlock at all while entering wrip().
859 	 * We must make sure not to downgrade/drop vfs_dqrwlock if we
860 	 * have it as writer, i.e. if we are updating the quota inode.
861 	 * There is no potential deadlock scenario in this case as
862 	 * ufs_getpage() takes care of this and avoids reacquiring
863 	 * vfs_dqrwlock in that case.
864 	 *
865 	 * This check is done here since the above conditions do not change
866 	 * and we possibly loop below, so save a few cycles.
867 	 */
868 	if ((type == IFSHAD) ||
869 		(rw_owner(&ufsvfsp->vfs_dqrwlock) == curthread)) {
870 			do_dqrwlock = 0;
871 	} else {
872 		do_dqrwlock = 1;
873 	}
874 
875 	/*
876 	 * Large Files: We cast MAXBMASK to offset_t
877 	 * inorder to mask out the higher bits. Since offset_t
878 	 * is a signed value, the high order bit set in MAXBMASK
879 	 * value makes it do the right thing by having all bits 1
880 	 * in the higher word. May be removed for _SOLARIS64_.
881 	 */
882 
883 	fs = ip->i_fs;
884 	do {
885 		u_offset_t uoff = uio->uio_loffset;
886 		off = uoff & (offset_t)MAXBMASK;
887 		mapon = (int)(uoff & (offset_t)MAXBOFFSET);
888 		on = (int)blkoff(fs, uoff);
889 		n = (int)MIN(fs->fs_bsize - on, uio->uio_resid);
890 		new_iblocks = 1;
891 
892 		if (type == IFREG && uoff + n >= limit) {
893 			if (uoff >= limit) {
894 				error = EFBIG;
895 				goto out;
896 			}
897 			/*
898 			 * since uoff + n >= limit,
899 			 * therefore n >= limit - uoff, and n is an int
900 			 * so it is safe to cast it to an int
901 			 */
902 			n = (int)(limit - (rlim64_t)uoff);
903 		}
904 		if (uoff + n > ip->i_size) {
905 			/*
906 			 * We are extending the length of the file.
907 			 * bmap is used so that we are sure that
908 			 * if we need to allocate new blocks, that it
909 			 * is done here before we up the file size.
910 			 */
911 			error = bmap_write(ip, uoff, (int)(on + n),
912 			    mapon == 0, NULL, cr);
913 			/*
914 			 * bmap_write never drops i_contents so if
915 			 * the flags are set it changed the file.
916 			 */
917 			if (ip->i_flag & (ICHG|IUPD)) {
918 				i_seq_needed = 1;
919 				ip->i_flag |= ISEQ;
920 			}
921 			if (error)
922 				break;
923 			/*
924 			 * There is a window of vulnerability here.
925 			 * The sequence of operations: allocate file
926 			 * system blocks, uiomove the data into pages,
927 			 * and then update the size of the file in the
928 			 * inode, must happen atomically.  However, due
929 			 * to current locking constraints, this can not
930 			 * be done.
931 			 */
932 			ASSERT(ip->i_writer == NULL);
933 			ip->i_writer = curthread;
934 			i_size_changed = 1;
935 			/*
936 			 * If we are writing from the beginning of
937 			 * the mapping, we can just create the
938 			 * pages without having to read them.
939 			 */
940 			pagecreate = (mapon == 0);
941 		} else if (n == MAXBSIZE) {
942 			/*
943 			 * Going to do a whole mappings worth,
944 			 * so we can just create the pages w/o
945 			 * having to read them in.  But before
946 			 * we do that, we need to make sure any
947 			 * needed blocks are allocated first.
948 			 */
949 			iblocks = ip->i_blocks;
950 			error = bmap_write(ip, uoff, (int)(on + n),
951 			    BI_ALLOC_ONLY, NULL, cr);
952 			/*
953 			 * bmap_write never drops i_contents so if
954 			 * the flags are set it changed the file.
955 			 */
956 			if (ip->i_flag & (ICHG|IUPD)) {
957 				i_seq_needed = 1;
958 				ip->i_flag |= ISEQ;
959 			}
960 			if (error)
961 				break;
962 			pagecreate = 1;
963 			/*
964 			 * check if the new created page needed the
965 			 * allocation of new disk blocks.
966 			 */
967 			if (iblocks == ip->i_blocks)
968 				new_iblocks = 0; /* no new blocks allocated */
969 		} else {
970 			pagecreate = 0;
971 			/*
972 			 * In sync mode flush the indirect blocks which
973 			 * may have been allocated and not written on
974 			 * disk. In above cases bmap_write will allocate
975 			 * in sync mode.
976 			 */
977 			if (ioflag & (FSYNC|FDSYNC)) {
978 				error = ufs_indirblk_sync(ip, uoff);
979 				if (error)
980 					break;
981 			}
982 		}
983 
984 		/*
985 		 * At this point we can enter ufs_getpage() in one
986 		 * of two ways:
987 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
988 		 *    forcefault parameter is true (pagecreate == 0)
989 		 * 2) uiomove() causes a page fault.
990 		 *
991 		 * We have to drop the contents lock to prevent the VM
992 		 * system from trying to reaquire it in ufs_getpage()
993 		 * should the uiomove cause a pagefault.
994 		 *
995 		 * We have to drop the reader vfs_dqrwlock here as well.
996 		 */
997 		rw_exit(&ip->i_contents);
998 		if (do_dqrwlock) {
999 			ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock));
1000 			ASSERT(!(RW_WRITE_HELD(&ufsvfsp->vfs_dqrwlock)));
1001 			rw_exit(&ufsvfsp->vfs_dqrwlock);
1002 		}
1003 
1004 		newpage = 0;
1005 		premove_resid = uio->uio_resid;
1006 		if (vpm_enable) {
1007 			/*
1008 			 * Copy data. If new pages are created, part of
1009 			 * the page that is not written will be initizliazed
1010 			 * with zeros.
1011 			 */
1012 			error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1013 				uio, !pagecreate, &newpage, 0, S_WRITE);
1014 		} else {
1015 
1016 			base = segmap_getmapflt(segkmap, vp, (off + mapon),
1017 					(uint_t)n, !pagecreate, S_WRITE);
1018 
1019 			/*
1020 			 * segmap_pagecreate() returns 1 if it calls
1021 			 * page_create_va() to allocate any pages.
1022 			 */
1023 
1024 			if (pagecreate)
1025 				newpage = segmap_pagecreate(segkmap, base,
1026 				    (size_t)n, 0);
1027 
1028 			error = uiomove(base + mapon, (long)n, UIO_WRITE, uio);
1029 		}
1030 
1031 		/*
1032 		 * If "newpage" is set, then a new page was created and it
1033 		 * does not contain valid data, so it needs to be initialized
1034 		 * at this point.
1035 		 * Otherwise the page contains old data, which was overwritten
1036 		 * partially or as a whole in uiomove.
1037 		 * If there is only one iovec structure within uio, then
1038 		 * on error uiomove will not be able to update uio->uio_loffset
1039 		 * and we would zero the whole page here!
1040 		 *
1041 		 * If uiomove fails because of an error, the old valid data
1042 		 * is kept instead of filling the rest of the page with zero's.
1043 		 */
1044 		if (!vpm_enable && newpage &&
1045 		    uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) {
1046 			/*
1047 			 * We created pages w/o initializing them completely,
1048 			 * thus we need to zero the part that wasn't set up.
1049 			 * This happens on most EOF write cases and if
1050 			 * we had some sort of error during the uiomove.
1051 			 */
1052 			int nzero, nmoved;
1053 
1054 			nmoved = (int)(uio->uio_loffset - (off + mapon));
1055 			ASSERT(nmoved >= 0 && nmoved <= n);
1056 			nzero = roundup(on + n, PAGESIZE) - nmoved;
1057 			ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE);
1058 			(void) kzero(base + mapon + nmoved, (uint_t)nzero);
1059 		}
1060 
1061 		/*
1062 		 * Unlock the pages allocated by page_create_va()
1063 		 * in segmap_pagecreate()
1064 		 */
1065 		if (!vpm_enable && newpage)
1066 			segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE);
1067 
1068 		/*
1069 		 * If the size of the file changed, then update the
1070 		 * size field in the inode now.  This can't be done
1071 		 * before the call to segmap_pageunlock or there is
1072 		 * a potential deadlock with callers to ufs_putpage().
1073 		 * They will be holding i_contents and trying to lock
1074 		 * a page, while this thread is holding a page locked
1075 		 * and trying to acquire i_contents.
1076 		 */
1077 		if (i_size_changed) {
1078 			rw_enter(&ip->i_contents, RW_WRITER);
1079 			old_i_size = ip->i_size;
1080 			UFS_SET_ISIZE(uoff + n, ip);
1081 			TRANS_INODE(ufsvfsp, ip);
1082 			/*
1083 			 * file has grown larger than 2GB. Set flag
1084 			 * in superblock to indicate this, if it
1085 			 * is not already set.
1086 			 */
1087 			if ((ip->i_size > MAXOFF32_T) &&
1088 			    !(fs->fs_flags & FSLARGEFILES)) {
1089 				ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES);
1090 				mutex_enter(&ufsvfsp->vfs_lock);
1091 				fs->fs_flags |= FSLARGEFILES;
1092 				ufs_sbwrite(ufsvfsp);
1093 				mutex_exit(&ufsvfsp->vfs_lock);
1094 			}
1095 			mutex_enter(&ip->i_tlock);
1096 			ip->i_writer = NULL;
1097 			cv_broadcast(&ip->i_wrcv);
1098 			mutex_exit(&ip->i_tlock);
1099 			rw_exit(&ip->i_contents);
1100 		}
1101 
1102 		if (error) {
1103 			/*
1104 			 * If we failed on a write, we may have already
1105 			 * allocated file blocks as well as pages.  It's
1106 			 * hard to undo the block allocation, but we must
1107 			 * be sure to invalidate any pages that may have
1108 			 * been allocated.
1109 			 *
1110 			 * If the page was created without initialization
1111 			 * then we must check if it should be possible
1112 			 * to destroy the new page and to keep the old data
1113 			 * on the disk.
1114 			 *
1115 			 * It is possible to destroy the page without
1116 			 * having to write back its contents only when
1117 			 * - the size of the file keeps unchanged
1118 			 * - bmap_write() did not allocate new disk blocks
1119 			 *   it is possible to create big files using "seek" and
1120 			 *   write to the end of the file. A "write" to a
1121 			 *   position before the end of the file would not
1122 			 *   change the size of the file but it would allocate
1123 			 *   new disk blocks.
1124 			 * - uiomove intended to overwrite the whole page.
1125 			 * - a new page was created (newpage == 1).
1126 			 */
1127 
1128 			if (i_size_changed == 0 && new_iblocks == 0 &&
1129 			    newpage) {
1130 
1131 				/* unwind what uiomove eventually last did */
1132 				uio->uio_resid = premove_resid;
1133 
1134 				/*
1135 				 * destroy the page, do not write ambiguous
1136 				 * data to the disk.
1137 				 */
1138 				flags = SM_DESTROY;
1139 			} else {
1140 				/*
1141 				 * write the page back to the disk, if dirty,
1142 				 * and remove the page from the cache.
1143 				 */
1144 				flags = SM_INVAL;
1145 			}
1146 
1147 			if (vpm_enable) {
1148 				/*
1149 				 *  Flush pages.
1150 				 */
1151 				(void) vpm_sync_pages(vp, off, n, flags);
1152 			} else {
1153 				(void) segmap_release(segkmap, base, flags);
1154 			}
1155 		} else {
1156 			flags = 0;
1157 			/*
1158 			 * Force write back for synchronous write cases.
1159 			 */
1160 			if ((ioflag & (FSYNC|FDSYNC)) || type == IFDIR) {
1161 				/*
1162 				 * If the sticky bit is set but the
1163 				 * execute bit is not set, we do a
1164 				 * synchronous write back and free
1165 				 * the page when done.  We set up swap
1166 				 * files to be handled this way to
1167 				 * prevent servers from keeping around
1168 				 * the client's swap pages too long.
1169 				 * XXX - there ought to be a better way.
1170 				 */
1171 				if (IS_SWAPVP(vp)) {
1172 					flags = SM_WRITE | SM_FREE |
1173 					    SM_DONTNEED;
1174 					iupdat_flag = 0;
1175 				} else {
1176 					flags = SM_WRITE;
1177 				}
1178 			} else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
1179 				/*
1180 				 * Have written a whole block.
1181 				 * Start an asynchronous write and
1182 				 * mark the buffer to indicate that
1183 				 * it won't be needed again soon.
1184 				 */
1185 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
1186 			}
1187 			if (vpm_enable) {
1188 				/*
1189 				 * Flush pages.
1190 				 */
1191 				error = vpm_sync_pages(vp, off, n, flags);
1192 			} else {
1193 				error = segmap_release(segkmap, base, flags);
1194 			}
1195 			/*
1196 			 * If the operation failed and is synchronous,
1197 			 * then we need to unwind what uiomove() last
1198 			 * did so we can potentially return an error to
1199 			 * the caller.  If this write operation was
1200 			 * done in two pieces and the first succeeded,
1201 			 * then we won't return an error for the second
1202 			 * piece that failed.  However, we only want to
1203 			 * return a resid value that reflects what was
1204 			 * really done.
1205 			 *
1206 			 * Failures for non-synchronous operations can
1207 			 * be ignored since the page subsystem will
1208 			 * retry the operation until it succeeds or the
1209 			 * file system is unmounted.
1210 			 */
1211 			if (error) {
1212 				if ((ioflag & (FSYNC | FDSYNC)) ||
1213 				    type == IFDIR) {
1214 					uio->uio_resid = premove_resid;
1215 				} else {
1216 					error = 0;
1217 				}
1218 			}
1219 		}
1220 
1221 		/*
1222 		 * Re-acquire contents lock.
1223 		 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1224 		 */
1225 		if (do_dqrwlock)
1226 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1227 		rw_enter(&ip->i_contents, RW_WRITER);
1228 
1229 		/*
1230 		 * If the uiomove() failed or if a synchronous
1231 		 * page push failed, fix up i_size.
1232 		 */
1233 		if (error) {
1234 			if (i_size_changed) {
1235 				/*
1236 				 * The uiomove failed, and we
1237 				 * allocated blocks,so get rid
1238 				 * of them.
1239 				 */
1240 				(void) ufs_itrunc(ip, old_i_size, 0, cr);
1241 			}
1242 		} else {
1243 			/*
1244 			 * XXX - Can this be out of the loop?
1245 			 */
1246 			ip->i_flag |= IUPD | ICHG;
1247 			/*
1248 			 * Only do one increase of i_seq for multiple
1249 			 * pieces.  Because we drop locks, record
1250 			 * the fact that we changed the timestamp and
1251 			 * are deferring the increase in case another thread
1252 			 * pushes our timestamp update.
1253 			 */
1254 			i_seq_needed = 1;
1255 			ip->i_flag |= ISEQ;
1256 			if (i_size_changed)
1257 				ip->i_flag |= IATTCHG;
1258 			if ((ip->i_mode & (IEXEC | (IEXEC >> 3) |
1259 			    (IEXEC >> 6))) != 0 &&
1260 			    (ip->i_mode & (ISUID | ISGID)) != 0 &&
1261 			    secpolicy_vnode_setid_retain(cr,
1262 			    (ip->i_mode & ISUID) != 0 && ip->i_uid == 0) != 0) {
1263 				/*
1264 				 * Clear Set-UID & Set-GID bits on
1265 				 * successful write if not privileged
1266 				 * and at least one of the execute bits
1267 				 * is set.  If we always clear Set-GID,
1268 				 * mandatory file and record locking is
1269 				 * unuseable.
1270 				 */
1271 				ip->i_mode &= ~(ISUID | ISGID);
1272 			}
1273 		}
1274 		/*
1275 		 * In the case the FDSYNC flag is set and this is a
1276 		 * "rewrite" we won't log a delta.
1277 		 * The FSYNC flag overrides all cases.
1278 		 */
1279 		if (!ufs_check_rewrite(ip, uio, ioflag) || !(ioflag & FDSYNC)) {
1280 			TRANS_INODE(ufsvfsp, ip);
1281 		}
1282 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1283 
1284 out:
1285 	/*
1286 	 * Make sure i_seq is increased at least once per write
1287 	 */
1288 	if (i_seq_needed) {
1289 		ip->i_seq++;
1290 		ip->i_flag &= ~ISEQ;	/* no longer deferred */
1291 	}
1292 
1293 	/*
1294 	 * Inode is updated according to this table -
1295 	 *
1296 	 *   FSYNC	  FDSYNC(posix.4)
1297 	 *   --------------------------
1298 	 *   always@	  IATTCHG|IBDWRITE
1299 	 *
1300 	 * @ - 	If we are doing synchronous write the only time we should
1301 	 *	not be sync'ing the ip here is if we have the stickyhack
1302 	 *	activated, the file is marked with the sticky bit and
1303 	 *	no exec bit, the file length has not been changed and
1304 	 *	no new blocks have been allocated during this write.
1305 	 */
1306 
1307 	if ((ip->i_flag & ISYNC) != 0) {
1308 		/*
1309 		 * we have eliminated nosync
1310 		 */
1311 		if ((ip->i_flag & (IATTCHG|IBDWRITE)) ||
1312 			((ioflag & FSYNC) && iupdat_flag)) {
1313 			ufs_iupdat(ip, 1);
1314 		}
1315 	}
1316 
1317 	/*
1318 	 * If we've already done a partial-write, terminate
1319 	 * the write but return no error unless the error is ENOSPC
1320 	 * because the caller can detect this and free resources and
1321 	 * try again.
1322 	 */
1323 	if ((start_resid != uio->uio_resid) && (error != ENOSPC))
1324 		error = 0;
1325 
1326 	ip->i_flag &= ~(INOACC | ISYNC);
1327 	ITIMES_NOLOCK(ip);
1328 	TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1329 		"ufs_wrip_end:vp %p error %d", vp, error);
1330 	return (error);
1331 }
1332 
1333 /*
1334  * rdip does the real work of read requests for ufs.
1335  */
1336 int
1337 rdip(struct inode *ip, struct uio *uio, int ioflag, cred_t *cr)
1338 {
1339 	u_offset_t off;
1340 	caddr_t base;
1341 	struct fs *fs;
1342 	struct ufsvfs *ufsvfsp;
1343 	struct vnode *vp;
1344 	long oresid = uio->uio_resid;
1345 	u_offset_t n, on, mapon;
1346 	int error = 0;
1347 	int doupdate = 1;
1348 	uint_t flags;
1349 	int dofree, directio_status;
1350 	krw_t rwtype;
1351 	o_mode_t type;
1352 
1353 	vp = ITOV(ip);
1354 
1355 	TRACE_1(TR_FAC_UFS, TR_UFS_RWIP_START,
1356 		"ufs_rdip_start:vp %p", vp);
1357 
1358 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
1359 
1360 	ufsvfsp = ip->i_ufsvfs;
1361 
1362 	if (ufsvfsp == NULL)
1363 		return (EIO);
1364 
1365 	fs = ufsvfsp->vfs_fs;
1366 
1367 	/* check for valid filetype */
1368 	type = ip->i_mode & IFMT;
1369 	if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
1370 	    (type != IFLNK) && (type != IFSHAD)) {
1371 		return (EIO);
1372 	}
1373 
1374 	if (uio->uio_loffset > UFS_MAXOFFSET_T) {
1375 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1376 			"ufs_rdip_end:vp %p error %d", vp, EINVAL);
1377 		error = 0;
1378 		goto out;
1379 	}
1380 	if (uio->uio_loffset < (offset_t)0) {
1381 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1382 			"ufs_rdip_end:vp %p error %d", vp, EINVAL);
1383 		return (EINVAL);
1384 	}
1385 	if (uio->uio_resid == 0) {
1386 		TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1387 			"ufs_rdip_end:vp %p error %d", vp, 0);
1388 		return (0);
1389 	}
1390 
1391 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (fs->fs_ronly == 0) &&
1392 		(!ufsvfsp->vfs_noatime)) {
1393 		mutex_enter(&ip->i_tlock);
1394 		ip->i_flag |= IACC;
1395 		mutex_exit(&ip->i_tlock);
1396 	}
1397 	/*
1398 	 * Try to go direct
1399 	 */
1400 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
1401 		error = ufs_directio_read(ip, uio, cr, &directio_status);
1402 		if (directio_status == DIRECTIO_SUCCESS)
1403 			goto out;
1404 	}
1405 
1406 	rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER);
1407 
1408 	do {
1409 		offset_t diff;
1410 		u_offset_t uoff = uio->uio_loffset;
1411 		off = uoff & (offset_t)MAXBMASK;
1412 		mapon = (u_offset_t)(uoff & (offset_t)MAXBOFFSET);
1413 		on = (u_offset_t)blkoff(fs, uoff);
1414 		n = MIN((u_offset_t)fs->fs_bsize - on,
1415 			(u_offset_t)uio->uio_resid);
1416 
1417 		diff = ip->i_size - uoff;
1418 
1419 		if (diff <= (offset_t)0) {
1420 			error = 0;
1421 			goto out;
1422 		}
1423 		if (diff < (offset_t)n)
1424 			n = (int)diff;
1425 
1426 		/*
1427 		 * We update smallfile2 and smallfile1 at most every second.
1428 		 */
1429 		if (lbolt >= smallfile_update) {
1430 			uint64_t percpufreeb;
1431 			if (smallfile1_d == 0) smallfile1_d = SMALLFILE1_D;
1432 			if (smallfile2_d == 0) smallfile2_d = SMALLFILE2_D;
1433 			percpufreeb = ptob((uint64_t)freemem) / ncpus_online;
1434 			smallfile1 = percpufreeb / smallfile1_d;
1435 			smallfile2 = percpufreeb / smallfile2_d;
1436 			smallfile1 = MAX(smallfile1, smallfile);
1437 			smallfile1 = MAX(smallfile1, smallfile64);
1438 			smallfile2 = MAX(smallfile1, smallfile2);
1439 			smallfile_update = lbolt + hz;
1440 		}
1441 
1442 		dofree = freebehind &&
1443 		    ip->i_nextr == (off & PAGEMASK) && off > smallfile1;
1444 
1445 		/*
1446 		 * At this point we can enter ufs_getpage() in one of two
1447 		 * ways:
1448 		 * 1) segmap_getmapflt() calls ufs_getpage() when the
1449 		 *    forcefault parameter is true (value of 1 is passed)
1450 		 * 2) uiomove() causes a page fault.
1451 		 *
1452 		 * We cannot hold onto an i_contents reader lock without
1453 		 * risking deadlock in ufs_getpage() so drop a reader lock.
1454 		 * The ufs_getpage() dolock logic already allows for a
1455 		 * thread holding i_contents as writer to work properly
1456 		 * so we keep a writer lock.
1457 		 */
1458 		if (rwtype == RW_READER)
1459 			rw_exit(&ip->i_contents);
1460 
1461 		if (vpm_enable) {
1462 			/*
1463 			 * Copy data.
1464 			 */
1465 			error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1466 				uio, 1, NULL, 0, S_READ);
1467 		} else {
1468 			base = segmap_getmapflt(segkmap, vp, (off + mapon),
1469 					(uint_t)n, 1, S_READ);
1470 			error = uiomove(base + mapon, (long)n, UIO_READ, uio);
1471 		}
1472 
1473 		flags = 0;
1474 		if (!error) {
1475 			/*
1476 			 * If  reading sequential  we won't need  this
1477 			 * buffer again  soon.  For  offsets in  range
1478 			 * [smallfile1,  smallfile2] release the pages
1479 			 * at   the  tail  of the   cache list, larger
1480 			 * offsets are released at the head.
1481 			 */
1482 			if (dofree) {
1483 				flags = SM_FREE | SM_ASYNC;
1484 				if ((cache_read_ahead == 0) &&
1485 					(off > smallfile2))
1486 					flags |=  SM_DONTNEED;
1487 			}
1488 			/*
1489 			 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1490 			 * we want to make sure that the page which has
1491 			 * been read, is written on disk if it is dirty.
1492 			 * And corresponding indirect blocks should also
1493 			 * be flushed out.
1494 			 */
1495 			if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) {
1496 				flags &= ~SM_ASYNC;
1497 				flags |= SM_WRITE;
1498 			}
1499 			if (vpm_enable) {
1500 				error = vpm_sync_pages(vp, off, n, flags);
1501 			} else {
1502 				error = segmap_release(segkmap, base, flags);
1503 			}
1504 		} else {
1505 			if (vpm_enable) {
1506 				(void) vpm_sync_pages(vp, off, n, flags);
1507 			} else {
1508 				(void) segmap_release(segkmap, base, flags);
1509 			}
1510 		}
1511 
1512 		if (rwtype == RW_READER)
1513 			rw_enter(&ip->i_contents, rwtype);
1514 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
1515 out:
1516 	/*
1517 	 * Inode is updated according to this table if FRSYNC is set.
1518 	 *
1519 	 *   FSYNC	  FDSYNC(posix.4)
1520 	 *   --------------------------
1521 	 *   always	  IATTCHG|IBDWRITE
1522 	 */
1523 	/*
1524 	 * The inode is not updated if we're logging and the inode is a
1525 	 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1526 	 */
1527 	if (ioflag & FRSYNC) {
1528 		if (TRANS_ISTRANS(ufsvfsp) && ((ip->i_mode & IFMT) == IFDIR)) {
1529 				doupdate = 0;
1530 		}
1531 		if (doupdate) {
1532 			if ((ioflag & FSYNC) ||
1533 			    ((ioflag & FDSYNC) &&
1534 			    (ip->i_flag & (IATTCHG|IBDWRITE)))) {
1535 				ufs_iupdat(ip, 1);
1536 			}
1537 		}
1538 	}
1539 	/*
1540 	 * If we've already done a partial read, terminate
1541 	 * the read but return no error.
1542 	 */
1543 	if (oresid != uio->uio_resid)
1544 		error = 0;
1545 	ITIMES(ip);
1546 
1547 	TRACE_2(TR_FAC_UFS, TR_UFS_RWIP_END,
1548 		"ufs_rdip_end:vp %p error %d", vp, error);
1549 	return (error);
1550 }
1551 
1552 /* ARGSUSED */
1553 static int
1554 ufs_ioctl(
1555 	struct vnode	*vp,
1556 	int		cmd,
1557 	intptr_t	arg,
1558 	int		flag,
1559 	struct cred	*cr,
1560 	int		*rvalp)
1561 {
1562 	struct lockfs	lockfs, lockfs_out;
1563 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
1564 	char		*comment, *original_comment;
1565 	struct fs	*fs;
1566 	struct ulockfs	*ulp;
1567 	offset_t	off;
1568 	extern int	maxphys;
1569 	int		error;
1570 	int		issync;
1571 	int		trans_size;
1572 
1573 
1574 	/*
1575 	 * forcibly unmounted
1576 	 */
1577 	if (ufsvfsp == NULL) {
1578 		return (EIO);
1579 	}
1580 
1581 	fs = ufsvfsp->vfs_fs;
1582 
1583 	if (cmd == Q_QUOTACTL) {
1584 		error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_QUOTA_MASK);
1585 		if (error)
1586 			return (error);
1587 
1588 		if (ulp) {
1589 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_QUOTA,
1590 						TOP_SETQUOTA_SIZE(fs));
1591 		}
1592 
1593 		error = quotactl(vp, arg, flag, cr);
1594 
1595 		if (ulp) {
1596 			TRANS_END_ASYNC(ufsvfsp, TOP_QUOTA,
1597 						TOP_SETQUOTA_SIZE(fs));
1598 			ufs_lockfs_end(ulp);
1599 		}
1600 		return (error);
1601 	}
1602 
1603 	switch (cmd) {
1604 		case _FIOLFS:
1605 			/*
1606 			 * file system locking
1607 			 */
1608 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1609 				return (EPERM);
1610 
1611 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1612 				if (copyin((caddr_t)arg, &lockfs,
1613 						sizeof (struct lockfs)))
1614 					return (EFAULT);
1615 			}
1616 #ifdef _SYSCALL32_IMPL
1617 			else {
1618 				struct lockfs32	lockfs32;
1619 				/* Translate ILP32 lockfs to LP64 lockfs */
1620 				if (copyin((caddr_t)arg, &lockfs32,
1621 				    sizeof (struct lockfs32)))
1622 					return (EFAULT);
1623 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1624 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1625 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1626 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1627 				lockfs.lf_comment =
1628 					(caddr_t)(uintptr_t)lockfs32.lf_comment;
1629 			}
1630 #endif /* _SYSCALL32_IMPL */
1631 
1632 			if (lockfs.lf_comlen) {
1633 				if (lockfs.lf_comlen > LOCKFS_MAXCOMMENTLEN)
1634 					return (ENAMETOOLONG);
1635 				comment = kmem_alloc(lockfs.lf_comlen,
1636 						KM_SLEEP);
1637 				if (copyin(lockfs.lf_comment, comment,
1638 					lockfs.lf_comlen)) {
1639 					kmem_free(comment, lockfs.lf_comlen);
1640 					return (EFAULT);
1641 				}
1642 				original_comment = lockfs.lf_comment;
1643 				lockfs.lf_comment = comment;
1644 			}
1645 			if ((error = ufs_fiolfs(vp, &lockfs, 0)) == 0) {
1646 				lockfs.lf_comment = original_comment;
1647 
1648 				if ((flag & DATAMODEL_MASK) ==
1649 				    DATAMODEL_NATIVE) {
1650 					(void) copyout(&lockfs, (caddr_t)arg,
1651 					    sizeof (struct lockfs));
1652 				}
1653 #ifdef _SYSCALL32_IMPL
1654 				else {
1655 					struct lockfs32	lockfs32;
1656 					/* Translate LP64 to ILP32 lockfs */
1657 					lockfs32.lf_lock =
1658 					    (uint32_t)lockfs.lf_lock;
1659 					lockfs32.lf_flags =
1660 					    (uint32_t)lockfs.lf_flags;
1661 					lockfs32.lf_key =
1662 					    (uint32_t)lockfs.lf_key;
1663 					lockfs32.lf_comlen =
1664 					    (uint32_t)lockfs.lf_comlen;
1665 					lockfs32.lf_comment =
1666 					(uint32_t)(uintptr_t)lockfs.lf_comment;
1667 					(void) copyout(&lockfs32, (caddr_t)arg,
1668 					    sizeof (struct lockfs32));
1669 				}
1670 #endif /* _SYSCALL32_IMPL */
1671 
1672 			} else {
1673 				if (lockfs.lf_comlen)
1674 					kmem_free(comment, lockfs.lf_comlen);
1675 			}
1676 			return (error);
1677 
1678 		case _FIOLFSS:
1679 			/*
1680 			 * get file system locking status
1681 			 */
1682 
1683 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1684 				if (copyin((caddr_t)arg, &lockfs,
1685 						sizeof (struct lockfs)))
1686 					return (EFAULT);
1687 			}
1688 #ifdef _SYSCALL32_IMPL
1689 			else {
1690 				struct lockfs32	lockfs32;
1691 				/* Translate ILP32 lockfs to LP64 lockfs */
1692 				if (copyin((caddr_t)arg, &lockfs32,
1693 						sizeof (struct lockfs32)))
1694 					return (EFAULT);
1695 				lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1696 				lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1697 				lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1698 				lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1699 				lockfs.lf_comment =
1700 					(caddr_t)(uintptr_t)lockfs32.lf_comment;
1701 			}
1702 #endif /* _SYSCALL32_IMPL */
1703 
1704 			if (error =  ufs_fiolfss(vp, &lockfs_out))
1705 				return (error);
1706 			lockfs.lf_lock = lockfs_out.lf_lock;
1707 			lockfs.lf_key = lockfs_out.lf_key;
1708 			lockfs.lf_flags = lockfs_out.lf_flags;
1709 			lockfs.lf_comlen = MIN(lockfs.lf_comlen,
1710 				lockfs_out.lf_comlen);
1711 
1712 			if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1713 				if (copyout(&lockfs, (caddr_t)arg,
1714 						sizeof (struct lockfs)))
1715 					return (EFAULT);
1716 			}
1717 #ifdef _SYSCALL32_IMPL
1718 			else {
1719 				/* Translate LP64 to ILP32 lockfs */
1720 				struct lockfs32	lockfs32;
1721 				lockfs32.lf_lock = (uint32_t)lockfs.lf_lock;
1722 				lockfs32.lf_flags = (uint32_t)lockfs.lf_flags;
1723 				lockfs32.lf_key = (uint32_t)lockfs.lf_key;
1724 				lockfs32.lf_comlen = (uint32_t)lockfs.lf_comlen;
1725 				lockfs32.lf_comment =
1726 					(uint32_t)(uintptr_t)lockfs.lf_comment;
1727 				if (copyout(&lockfs32, (caddr_t)arg,
1728 					    sizeof (struct lockfs32)))
1729 					return (EFAULT);
1730 			}
1731 #endif /* _SYSCALL32_IMPL */
1732 
1733 			if (lockfs.lf_comlen &&
1734 			    lockfs.lf_comment && lockfs_out.lf_comment)
1735 				if (copyout(lockfs_out.lf_comment,
1736 					lockfs.lf_comment,
1737 					lockfs.lf_comlen))
1738 					return (EFAULT);
1739 			return (0);
1740 
1741 		case _FIOSATIME:
1742 			/*
1743 			 * set access time
1744 			 */
1745 
1746 			/*
1747 			 * if mounted w/o atime, return quietly.
1748 			 * I briefly thought about returning ENOSYS, but
1749 			 * figured that most apps would consider this fatal
1750 			 * but the idea is to make this as seamless as poss.
1751 			 */
1752 			if (ufsvfsp->vfs_noatime)
1753 				return (0);
1754 
1755 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1756 					ULOCKFS_SETATTR_MASK);
1757 			if (error)
1758 				return (error);
1759 
1760 			if (ulp) {
1761 				trans_size = (int)TOP_SETATTR_SIZE(VTOI(vp));
1762 				TRANS_BEGIN_CSYNC(ufsvfsp, issync,
1763 						TOP_SETATTR, trans_size);
1764 			}
1765 
1766 			error = ufs_fiosatime(vp, (struct timeval *)arg,
1767 					flag, cr);
1768 
1769 			if (ulp) {
1770 				TRANS_END_CSYNC(ufsvfsp, error, issync,
1771 						TOP_SETATTR, trans_size);
1772 				ufs_lockfs_end(ulp);
1773 			}
1774 			return (error);
1775 
1776 		case _FIOSDIO:
1777 			/*
1778 			 * set delayed-io
1779 			 */
1780 			return (ufs_fiosdio(vp, (uint_t *)arg, flag, cr));
1781 
1782 		case _FIOGDIO:
1783 			/*
1784 			 * get delayed-io
1785 			 */
1786 			return (ufs_fiogdio(vp, (uint_t *)arg, flag, cr));
1787 
1788 		case _FIOIO:
1789 			/*
1790 			 * inode open
1791 			 */
1792 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1793 					ULOCKFS_VGET_MASK);
1794 			if (error)
1795 				return (error);
1796 
1797 			error = ufs_fioio(vp, (struct fioio *)arg, flag, cr);
1798 
1799 			if (ulp) {
1800 				ufs_lockfs_end(ulp);
1801 			}
1802 			return (error);
1803 
1804 		case _FIOFFS:
1805 			/*
1806 			 * file system flush (push w/invalidate)
1807 			 */
1808 			if ((caddr_t)arg != NULL)
1809 				return (EINVAL);
1810 			return (ufs_fioffs(vp, NULL, cr));
1811 
1812 		case _FIOISBUSY:
1813 			/*
1814 			 * Contract-private interface for Legato
1815 			 * Purge this vnode from the DNLC and decide
1816 			 * if this vnode is busy (*arg == 1) or not
1817 			 * (*arg == 0)
1818 			 */
1819 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1820 				return (EPERM);
1821 			error = ufs_fioisbusy(vp, (int *)arg, cr);
1822 			return (error);
1823 
1824 		case _FIODIRECTIO:
1825 			return (ufs_fiodirectio(vp, (int)arg, cr));
1826 
1827 		case _FIOTUNE:
1828 			/*
1829 			 * Tune the file system (aka setting fs attributes)
1830 			 */
1831 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
1832 					ULOCKFS_SETATTR_MASK);
1833 			if (error)
1834 				return (error);
1835 
1836 			error = ufs_fiotune(vp, (struct fiotune *)arg, cr);
1837 
1838 			if (ulp)
1839 				ufs_lockfs_end(ulp);
1840 			return (error);
1841 
1842 		case _FIOLOGENABLE:
1843 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1844 				return (EPERM);
1845 			return (ufs_fiologenable(vp, (void *)arg, cr, flag));
1846 
1847 		case _FIOLOGDISABLE:
1848 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1849 				return (EPERM);
1850 			return (ufs_fiologdisable(vp, (void *)arg, cr, flag));
1851 
1852 		case _FIOISLOG:
1853 			return (ufs_fioislog(vp, (void *)arg, cr, flag));
1854 
1855 		case _FIOSNAPSHOTCREATE_MULTI:
1856 		{
1857 			struct fiosnapcreate_multi	fc, *fcp;
1858 			size_t	fcm_size;
1859 
1860 			if (copyin((void *)arg, &fc, sizeof (fc)))
1861 				return (EFAULT);
1862 			if (fc.backfilecount > MAX_BACKFILE_COUNT)
1863 				return (EINVAL);
1864 			fcm_size = sizeof (struct fiosnapcreate_multi) +
1865 			    (fc.backfilecount - 1) * sizeof (int);
1866 			fcp = (struct fiosnapcreate_multi *)
1867 			    kmem_alloc(fcm_size, KM_SLEEP);
1868 			if (copyin((void *)arg, fcp, fcm_size)) {
1869 				kmem_free(fcp, fcm_size);
1870 				return (EFAULT);
1871 			}
1872 			error = ufs_snap_create(vp, fcp, cr);
1873 			if (!error && copyout(fcp, (void *)arg, fcm_size))
1874 				error = EFAULT;
1875 			kmem_free(fcp, fcm_size);
1876 			return (error);
1877 		}
1878 
1879 		case _FIOSNAPSHOTDELETE:
1880 		{
1881 			struct fiosnapdelete	fc;
1882 
1883 			if (copyin((void *)arg, &fc, sizeof (fc)))
1884 				return (EFAULT);
1885 			error = ufs_snap_delete(vp, &fc, cr);
1886 			if (!error && copyout(&fc, (void *)arg, sizeof (fc)))
1887 				error = EFAULT;
1888 			return (error);
1889 		}
1890 
1891 		case _FIOGETSUPERBLOCK:
1892 			if (copyout(fs, (void *)arg, SBSIZE))
1893 				return (EFAULT);
1894 			return (0);
1895 
1896 		case _FIOGETMAXPHYS:
1897 			if (copyout(&maxphys, (void *)arg, sizeof (maxphys)))
1898 				return (EFAULT);
1899 			return (0);
1900 
1901 		/*
1902 		 * The following 3 ioctls are for TSufs support
1903 		 * although could potentially be used elsewhere
1904 		 */
1905 		case _FIO_SET_LUFS_DEBUG:
1906 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1907 				return (EPERM);
1908 			lufs_debug = (uint32_t)arg;
1909 			return (0);
1910 
1911 		case _FIO_SET_LUFS_ERROR:
1912 			if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1913 				return (EPERM);
1914 			TRANS_SETERROR(ufsvfsp);
1915 			return (0);
1916 
1917 		case _FIO_GET_TOP_STATS:
1918 		{
1919 			fio_lufs_stats_t *ls;
1920 			ml_unit_t *ul = ufsvfsp->vfs_log;
1921 
1922 			ls = kmem_zalloc(sizeof (*ls), KM_SLEEP);
1923 			ls->ls_debug = ul->un_debug; /* return debug value */
1924 			/* Copy stucture if statistics are being kept */
1925 			if (ul->un_logmap->mtm_tops) {
1926 				ls->ls_topstats = *(ul->un_logmap->mtm_tops);
1927 			}
1928 			error = 0;
1929 			if (copyout(ls, (void *)arg, sizeof (*ls)))
1930 				error = EFAULT;
1931 			kmem_free(ls, sizeof (*ls));
1932 			return (error);
1933 		}
1934 
1935 		case _FIO_SEEK_DATA:
1936 		case _FIO_SEEK_HOLE:
1937 			if (ddi_copyin((void *)arg, &off, sizeof (off), flag))
1938 				return (EFAULT);
1939 			/* offset paramater is in/out */
1940 			error = ufs_fio_holey(vp, cmd, &off);
1941 			if (error)
1942 				return (error);
1943 			if (ddi_copyout(&off, (void *)arg, sizeof (off), flag))
1944 				return (EFAULT);
1945 			return (0);
1946 
1947 		default:
1948 			return (ENOTTY);
1949 	}
1950 }
1951 
1952 /* ARGSUSED */
1953 static int
1954 ufs_getattr(struct vnode *vp, struct vattr *vap, int flags,
1955 	struct cred *cr)
1956 {
1957 	struct inode *ip = VTOI(vp);
1958 	struct ufsvfs *ufsvfsp;
1959 	int err;
1960 
1961 	TRACE_2(TR_FAC_UFS, TR_UFS_GETATTR_START,
1962 		"ufs_getattr_start:vp %p flags %x", vp, flags);
1963 
1964 	if (vap->va_mask == AT_SIZE) {
1965 		/*
1966 		 * for performance, if only the size is requested don't bother
1967 		 * with anything else.
1968 		 */
1969 		UFS_GET_ISIZE(&vap->va_size, ip);
1970 		TRACE_1(TR_FAC_UFS, TR_UFS_GETATTR_END,
1971 			"ufs_getattr_end:vp %p", vp);
1972 		return (0);
1973 	}
1974 
1975 	/*
1976 	 * inlined lockfs checks
1977 	 */
1978 	ufsvfsp = ip->i_ufsvfs;
1979 	if ((ufsvfsp == NULL) || ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs)) {
1980 		err = EIO;
1981 		goto out;
1982 	}
1983 
1984 	rw_enter(&ip->i_contents, RW_READER);
1985 	/*
1986 	 * Return all the attributes.  This should be refined so
1987 	 * that it only returns what's asked for.
1988 	 */
1989 
1990 	/*
1991 	 * Copy from inode table.
1992 	 */
1993 	vap->va_type = vp->v_type;
1994 	vap->va_mode = ip->i_mode & MODEMASK;
1995 	/*
1996 	 * If there is an ACL and there is a mask entry, then do the
1997 	 * extra work that completes the equivalent of an acltomode(3)
1998 	 * call.  According to POSIX P1003.1e, the acl mask should be
1999 	 * returned in the group permissions field.
2000 	 *
2001 	 * - start with the original permission and mode bits (from above)
2002 	 * - clear the group owner bits
2003 	 * - add in the mask bits.
2004 	 */
2005 	if (ip->i_ufs_acl && ip->i_ufs_acl->aclass.acl_ismask) {
2006 		vap->va_mode &= ~((VREAD | VWRITE | VEXEC) >> 3);
2007 		vap->va_mode |=
2008 		    (ip->i_ufs_acl->aclass.acl_maskbits & PERMMASK) << 3;
2009 	}
2010 	vap->va_uid = ip->i_uid;
2011 	vap->va_gid = ip->i_gid;
2012 	vap->va_fsid = ip->i_dev;
2013 	vap->va_nodeid = (ino64_t)ip->i_number;
2014 	vap->va_nlink = ip->i_nlink;
2015 	vap->va_size = ip->i_size;
2016 	if (vp->v_type == VCHR || vp->v_type == VBLK)
2017 		vap->va_rdev = ip->i_rdev;
2018 	else
2019 		vap->va_rdev = 0;	/* not a b/c spec. */
2020 	mutex_enter(&ip->i_tlock);
2021 	ITIMES_NOLOCK(ip);	/* mark correct time in inode */
2022 	vap->va_seq = ip->i_seq;
2023 	vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec;
2024 	vap->va_atime.tv_nsec = ip->i_atime.tv_usec*1000;
2025 	vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec;
2026 	vap->va_mtime.tv_nsec = ip->i_mtime.tv_usec*1000;
2027 	vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec;
2028 	vap->va_ctime.tv_nsec = ip->i_ctime.tv_usec*1000;
2029 	mutex_exit(&ip->i_tlock);
2030 
2031 	switch (ip->i_mode & IFMT) {
2032 
2033 	case IFBLK:
2034 		vap->va_blksize = MAXBSIZE;		/* was BLKDEV_IOSIZE */
2035 		break;
2036 
2037 	case IFCHR:
2038 		vap->va_blksize = MAXBSIZE;
2039 		break;
2040 
2041 	default:
2042 		vap->va_blksize = ip->i_fs->fs_bsize;
2043 		break;
2044 	}
2045 	vap->va_nblocks = (fsblkcnt64_t)ip->i_blocks;
2046 	rw_exit(&ip->i_contents);
2047 	err = 0;
2048 
2049 out:
2050 	TRACE_1(TR_FAC_UFS, TR_UFS_GETATTR_END, "ufs_getattr_end:vp %p", vp);
2051 
2052 	return (err);
2053 }
2054 
2055 /*ARGSUSED4*/
2056 static int
2057 ufs_setattr(
2058 	struct vnode *vp,
2059 	struct vattr *vap,
2060 	int flags,
2061 	struct cred *cr,
2062 	caller_context_t *ct)
2063 {
2064 	struct inode *ip = VTOI(vp);
2065 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2066 	struct fs *fs;
2067 	struct ulockfs *ulp;
2068 	char *errmsg1;
2069 	char *errmsg2;
2070 	long blocks;
2071 	long int mask = vap->va_mask;
2072 	size_t len1, len2;
2073 	int issync;
2074 	int trans_size;
2075 	int dotrans;
2076 	int dorwlock;
2077 	int error;
2078 	int owner_change;
2079 	int dodqlock;
2080 	timestruc_t now;
2081 	vattr_t oldva;
2082 	int retry = 1;
2083 	int indeadlock;
2084 
2085 	TRACE_2(TR_FAC_UFS, TR_UFS_SETATTR_START,
2086 		"ufs_setattr_start:vp %p flags %x", vp, flags);
2087 
2088 	/*
2089 	 * Cannot set these attributes.
2090 	 */
2091 	if (mask & AT_NOSET) {
2092 		error = EINVAL;
2093 		goto out;
2094 	}
2095 
2096 	/*
2097 	 * check for forced unmount
2098 	 */
2099 	if (ufsvfsp == NULL)
2100 		return (EIO);
2101 
2102 	fs = ufsvfsp->vfs_fs;
2103 	if (fs->fs_ronly != 0)
2104 		return (EROFS);
2105 
2106 again:
2107 	errmsg1 = NULL;
2108 	errmsg2 = NULL;
2109 	dotrans = 0;
2110 	dorwlock = 0;
2111 	dodqlock = 0;
2112 
2113 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK);
2114 	if (error)
2115 		goto out;
2116 
2117 	/*
2118 	 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2119 	 * This follows the protocol for read()/write().
2120 	 */
2121 	if (vp->v_type != VDIR) {
2122 		/*
2123 		 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2124 		 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2125 		 * possible, retries the operation.
2126 		 */
2127 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_file);
2128 		if (indeadlock) {
2129 			if (ulp)
2130 				ufs_lockfs_end(ulp);
2131 			goto again;
2132 		}
2133 		dorwlock = 1;
2134 	}
2135 
2136 	/*
2137 	 * Truncate file.  Must have write permission and not be a directory.
2138 	 */
2139 	if (mask & AT_SIZE) {
2140 		rw_enter(&ip->i_contents, RW_WRITER);
2141 		if (vp->v_type == VDIR) {
2142 			error = EISDIR;
2143 			goto update_inode;
2144 		}
2145 		if (error = ufs_iaccess(ip, IWRITE, cr))
2146 			goto update_inode;
2147 
2148 		rw_exit(&ip->i_contents);
2149 		error = TRANS_ITRUNC(ip, vap->va_size, 0, cr);
2150 		if (error) {
2151 			rw_enter(&ip->i_contents, RW_WRITER);
2152 			goto update_inode;
2153 		}
2154 	}
2155 
2156 	if (ulp) {
2157 		trans_size = (int)TOP_SETATTR_SIZE(ip);
2158 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size);
2159 		++dotrans;
2160 	}
2161 
2162 	/*
2163 	 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2164 	 * This follows the protocol established by
2165 	 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2166 	 */
2167 	if (vp->v_type == VDIR) {
2168 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_SETATTR,
2169 					retry_dir);
2170 		if (indeadlock)
2171 			goto again;
2172 		dorwlock = 1;
2173 	}
2174 
2175 	/*
2176 	 * Grab quota lock if we are changing the file's owner.
2177 	 */
2178 	if (mask & AT_UID) {
2179 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2180 		dodqlock = 1;
2181 	}
2182 	rw_enter(&ip->i_contents, RW_WRITER);
2183 
2184 	oldva.va_mode = ip->i_mode;
2185 	oldva.va_uid = ip->i_uid;
2186 	oldva.va_gid = ip->i_gid;
2187 
2188 	vap->va_mask &= ~AT_SIZE;
2189 	/*
2190 	 * ufs_iaccess is "close enough"; that's because it doesn't
2191 	 * map the defines.
2192 	 */
2193 	error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2194 				ufs_iaccess, ip);
2195 	if (error)
2196 		goto update_inode;
2197 
2198 	mask = vap->va_mask;
2199 
2200 	/*
2201 	 * Change file access modes.
2202 	 */
2203 	if (mask & AT_MODE) {
2204 		ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT);
2205 		TRANS_INODE(ufsvfsp, ip);
2206 		ip->i_flag |= ICHG;
2207 		if (stickyhack) {
2208 			mutex_enter(&vp->v_lock);
2209 			if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX)
2210 				vp->v_flag |= VSWAPLIKE;
2211 			else
2212 				vp->v_flag &= ~VSWAPLIKE;
2213 			mutex_exit(&vp->v_lock);
2214 		}
2215 	}
2216 	if (mask & (AT_UID|AT_GID)) {
2217 		if (mask & AT_UID) {
2218 			/*
2219 			 * Don't change ownership of the quota inode.
2220 			 */
2221 			if (ufsvfsp->vfs_qinod == ip) {
2222 				ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED);
2223 				error = EINVAL;
2224 				goto update_inode;
2225 			}
2226 
2227 			/*
2228 			 * No real ownership change.
2229 			 */
2230 			if (ip->i_uid == vap->va_uid) {
2231 				blocks = 0;
2232 				owner_change = 0;
2233 			}
2234 			/*
2235 			 * Remove the blocks and the file, from the old user's
2236 			 * quota.
2237 			 */
2238 			else {
2239 				blocks = ip->i_blocks;
2240 				owner_change = 1;
2241 
2242 				(void) chkdq(ip, -blocks, /* force */ 1, cr,
2243 						(char **)NULL, (size_t *)NULL);
2244 				(void) chkiq(ufsvfsp, /* change */ -1, ip,
2245 						(uid_t)ip->i_uid,
2246 						/* force */ 1, cr,
2247 						(char **)NULL, (size_t *)NULL);
2248 				dqrele(ip->i_dquot);
2249 			}
2250 
2251 			ip->i_uid = vap->va_uid;
2252 
2253 			/*
2254 			 * There is a real ownership change.
2255 			 */
2256 			if (owner_change) {
2257 				/*
2258 				 * Add the blocks and the file to the new
2259 				 * user's quota.
2260 				 */
2261 				ip->i_dquot = getinoquota(ip);
2262 				(void) chkdq(ip, blocks, /* force */ 1, cr,
2263 						&errmsg1, &len1);
2264 				(void) chkiq(ufsvfsp, /* change */ 1,
2265 						(struct inode *)NULL,
2266 						(uid_t)ip->i_uid,
2267 						/* force */ 1, cr,
2268 						&errmsg2, &len2);
2269 			}
2270 		}
2271 		if (mask & AT_GID) {
2272 			ip->i_gid = vap->va_gid;
2273 		}
2274 		TRANS_INODE(ufsvfsp, ip);
2275 		ip->i_flag |= ICHG;
2276 	}
2277 	/*
2278 	 * Change file access or modified times.
2279 	 */
2280 	if (mask & (AT_ATIME|AT_MTIME)) {
2281 		/* Check that the time value is within ufs range */
2282 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2283 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2284 			error = EOVERFLOW;
2285 			goto update_inode;
2286 		}
2287 
2288 		/*
2289 		 * if the "noaccess" mount option is set and only atime
2290 		 * update is requested, do nothing. No error is returned.
2291 		 */
2292 		if ((ufsvfsp->vfs_noatime) &&
2293 		    ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME))
2294 			goto skip_atime;
2295 
2296 		if (mask & AT_ATIME) {
2297 			ip->i_atime.tv_sec = vap->va_atime.tv_sec;
2298 			ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000;
2299 			ip->i_flag &= ~IACC;
2300 		}
2301 		if (mask & AT_MTIME) {
2302 			ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
2303 			ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000;
2304 			gethrestime(&now);
2305 			if (now.tv_sec > TIME32_MAX) {
2306 				/*
2307 				 * In 2038, ctime sticks forever..
2308 				 */
2309 				ip->i_ctime.tv_sec = TIME32_MAX;
2310 				ip->i_ctime.tv_usec = 0;
2311 			} else {
2312 				ip->i_ctime.tv_sec = now.tv_sec;
2313 				ip->i_ctime.tv_usec = now.tv_nsec / 1000;
2314 			}
2315 			ip->i_flag &= ~(IUPD|ICHG);
2316 			ip->i_flag |= IMODTIME;
2317 		}
2318 		TRANS_INODE(ufsvfsp, ip);
2319 		ip->i_flag |= IMOD;
2320 	}
2321 
2322 skip_atime:
2323 	/*
2324 	 * The presence of a shadow inode may indicate an ACL, but does
2325 	 * not imply an ACL.  Future FSD types should be handled here too
2326 	 * and check for the presence of the attribute-specific data
2327 	 * before referencing it.
2328 	 */
2329 	if (ip->i_shadow) {
2330 		/*
2331 		 * XXX if ufs_iupdat is changed to sandbagged write fix
2332 		 * ufs_acl_setattr to push ip to keep acls consistent
2333 		 *
2334 		 * Suppress out of inodes messages if we will retry.
2335 		 */
2336 		if (retry)
2337 			ip->i_flag |= IQUIET;
2338 		error = ufs_acl_setattr(ip, vap, cr);
2339 		ip->i_flag &= ~IQUIET;
2340 	}
2341 
2342 update_inode:
2343 	/*
2344 	 * Setattr always increases the sequence number
2345 	 */
2346 	ip->i_seq++;
2347 
2348 	/*
2349 	 * if nfsd and not logging; push synchronously
2350 	 */
2351 	if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) {
2352 		ufs_iupdat(ip, 1);
2353 	} else {
2354 		ITIMES_NOLOCK(ip);
2355 	}
2356 
2357 	rw_exit(&ip->i_contents);
2358 	if (dodqlock) {
2359 		rw_exit(&ufsvfsp->vfs_dqrwlock);
2360 	}
2361 	if (dorwlock)
2362 		rw_exit(&ip->i_rwlock);
2363 
2364 	if (ulp) {
2365 		if (dotrans) {
2366 			int terr = 0;
2367 			TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR,
2368 			    trans_size);
2369 			if (error == 0)
2370 				error = terr;
2371 		}
2372 		ufs_lockfs_end(ulp);
2373 	}
2374 out:
2375 	/*
2376 	 * If out of inodes or blocks, see if we can free something
2377 	 * up from the delete queue.
2378 	 */
2379 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
2380 		ufs_delete_drain_wait(ufsvfsp, 1);
2381 		retry = 0;
2382 		if (errmsg1 != NULL)
2383 			kmem_free(errmsg1, len1);
2384 		if (errmsg2 != NULL)
2385 			kmem_free(errmsg2, len2);
2386 		goto again;
2387 	}
2388 	TRACE_2(TR_FAC_UFS, TR_UFS_SETATTR_END,
2389 		"ufs_setattr_end:vp %p error %d", vp, error);
2390 	if (errmsg1 != NULL) {
2391 		uprintf(errmsg1);
2392 		kmem_free(errmsg1, len1);
2393 	}
2394 	if (errmsg2 != NULL) {
2395 		uprintf(errmsg2);
2396 		kmem_free(errmsg2, len2);
2397 	}
2398 	return (error);
2399 }
2400 
2401 /*ARGSUSED*/
2402 static int
2403 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr)
2404 {
2405 	struct inode *ip = VTOI(vp);
2406 	int error;
2407 
2408 	TRACE_3(TR_FAC_UFS, TR_UFS_ACCESS_START,
2409 		"ufs_access_start:vp %p mode %x flags %x", vp, mode, flags);
2410 
2411 	if (ip->i_ufsvfs == NULL)
2412 		return (EIO);
2413 
2414 	rw_enter(&ip->i_contents, RW_READER);
2415 
2416 	/*
2417 	 * The ufs_iaccess function wants to be called with
2418 	 * mode bits expressed as "ufs specific" bits.
2419 	 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2420 	 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2421 	 * But since they're the same we just pass the vnode mode
2422 	 * bit but just verify that assumption at compile time.
2423 	 */
2424 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2425 #error "ufs_access needs to map Vmodes to Imodes"
2426 #endif
2427 	error = ufs_iaccess(ip, mode, cr);
2428 
2429 	rw_exit(&ip->i_contents);
2430 
2431 	TRACE_2(TR_FAC_UFS, TR_UFS_ACCESS_END,
2432 		"ufs_access_end:vp %p error %d", vp, error);
2433 	return (error);
2434 }
2435 
2436 /* ARGSUSED */
2437 static int
2438 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr)
2439 {
2440 	struct inode *ip = VTOI(vp);
2441 	struct ufsvfs *ufsvfsp;
2442 	struct ulockfs *ulp;
2443 	int error;
2444 	int fastsymlink;
2445 
2446 	TRACE_2(TR_FAC_UFS, TR_UFS_READLINK_START,
2447 		"ufs_readlink_start:vp %p uiop %p", uiop, vp);
2448 
2449 	if (vp->v_type != VLNK) {
2450 		error = EINVAL;
2451 		goto nolockout;
2452 	}
2453 
2454 	/*
2455 	 * If the symbolic link is empty there is nothing to read.
2456 	 * Fast-track these empty symbolic links
2457 	 */
2458 	if (ip->i_size == 0) {
2459 		error = 0;
2460 		goto nolockout;
2461 	}
2462 
2463 	ufsvfsp = ip->i_ufsvfs;
2464 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK);
2465 	if (error)
2466 		goto nolockout;
2467 	/*
2468 	 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2469 	 */
2470 again:
2471 	fastsymlink = 0;
2472 	if (ip->i_flag & IFASTSYMLNK) {
2473 		rw_enter(&ip->i_rwlock, RW_READER);
2474 		rw_enter(&ip->i_contents, RW_READER);
2475 		if (ip->i_flag & IFASTSYMLNK) {
2476 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
2477 			    (ip->i_fs->fs_ronly == 0) &&
2478 			    (!ufsvfsp->vfs_noatime)) {
2479 				mutex_enter(&ip->i_tlock);
2480 				ip->i_flag |= IACC;
2481 				mutex_exit(&ip->i_tlock);
2482 			}
2483 			error = uiomove((caddr_t)&ip->i_db[1],
2484 				MIN(ip->i_size, uiop->uio_resid),
2485 				UIO_READ, uiop);
2486 			ITIMES(ip);
2487 			++fastsymlink;
2488 		}
2489 		rw_exit(&ip->i_contents);
2490 		rw_exit(&ip->i_rwlock);
2491 	}
2492 	if (!fastsymlink) {
2493 		ssize_t size;	/* number of bytes read  */
2494 		caddr_t basep;	/* pointer to input data */
2495 		ino_t ino;
2496 		long  igen;
2497 		struct uio tuio;	/* temp uio struct */
2498 		struct uio *tuiop;
2499 		iovec_t tiov;		/* temp iovec struct */
2500 		char kbuf[FSL_SIZE];	/* buffer to hold fast symlink */
2501 		int tflag = 0;		/* flag to indicate temp vars used */
2502 
2503 		ino = ip->i_number;
2504 		igen = ip->i_gen;
2505 		size = uiop->uio_resid;
2506 		basep = uiop->uio_iov->iov_base;
2507 		tuiop = uiop;
2508 
2509 		rw_enter(&ip->i_rwlock, RW_WRITER);
2510 		rw_enter(&ip->i_contents, RW_WRITER);
2511 		if (ip->i_flag & IFASTSYMLNK) {
2512 			rw_exit(&ip->i_contents);
2513 			rw_exit(&ip->i_rwlock);
2514 			goto again;
2515 		}
2516 
2517 		/* can this be a fast symlink and is it a user buffer? */
2518 		if (ip->i_size <= FSL_SIZE &&
2519 		    (uiop->uio_segflg == UIO_USERSPACE ||
2520 		    uiop->uio_segflg == UIO_USERISPACE)) {
2521 
2522 			bzero(&tuio, sizeof (struct uio));
2523 			/*
2524 			 * setup a kernel buffer to read link into.  this
2525 			 * is to fix a race condition where the user buffer
2526 			 * got corrupted before copying it into the inode.
2527 			 */
2528 			size = ip->i_size;
2529 			tiov.iov_len = size;
2530 			tiov.iov_base = kbuf;
2531 			tuio.uio_iov = &tiov;
2532 			tuio.uio_iovcnt = 1;
2533 			tuio.uio_offset = uiop->uio_offset;
2534 			tuio.uio_segflg = UIO_SYSSPACE;
2535 			tuio.uio_fmode = uiop->uio_fmode;
2536 			tuio.uio_extflg = uiop->uio_extflg;
2537 			tuio.uio_limit = uiop->uio_limit;
2538 			tuio.uio_resid = size;
2539 
2540 			basep = tuio.uio_iov->iov_base;
2541 			tuiop = &tuio;
2542 			tflag = 1;
2543 		}
2544 
2545 		error = rdip(ip, tuiop, 0, cr);
2546 		if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) {
2547 			rw_exit(&ip->i_contents);
2548 			rw_exit(&ip->i_rwlock);
2549 			goto out;
2550 		}
2551 
2552 		if (tflag == 0)
2553 			size -= uiop->uio_resid;
2554 
2555 		if ((tflag == 0 && ip->i_size <= FSL_SIZE &&
2556 		    ip->i_size == size) || (tflag == 1 &&
2557 		    tuio.uio_resid == 0)) {
2558 			error = kcopy(basep, &ip->i_db[1], ip->i_size);
2559 			if (error == 0) {
2560 				ip->i_flag |= IFASTSYMLNK;
2561 				/*
2562 				 * free page
2563 				 */
2564 				(void) VOP_PUTPAGE(ITOV(ip),
2565 				    (offset_t)0, PAGESIZE,
2566 				    (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC),
2567 				    cr);
2568 			} else {
2569 				int i;
2570 				/* error, clear garbage left behind */
2571 				for (i = 1; i < NDADDR; i++)
2572 					ip->i_db[i] = 0;
2573 				for (i = 0; i < NIADDR; i++)
2574 					ip->i_ib[i] = 0;
2575 			}
2576 		}
2577 		if (tflag == 1) {
2578 			/* now, copy it into the user buffer */
2579 			error = uiomove((caddr_t)kbuf,
2580 				MIN(size, uiop->uio_resid),
2581 				UIO_READ, uiop);
2582 		}
2583 		rw_exit(&ip->i_contents);
2584 		rw_exit(&ip->i_rwlock);
2585 	}
2586 out:
2587 	if (ulp) {
2588 		ufs_lockfs_end(ulp);
2589 	}
2590 nolockout:
2591 	TRACE_2(TR_FAC_UFS, TR_UFS_READLINK_END,
2592 		"ufs_readlink_end:vp %p error %d", vp, error);
2593 
2594 	return (error);
2595 }
2596 
2597 /* ARGSUSED */
2598 static int
2599 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr)
2600 {
2601 	struct inode *ip = VTOI(vp);
2602 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2603 	struct ulockfs *ulp;
2604 	int error;
2605 
2606 	TRACE_1(TR_FAC_UFS, TR_UFS_FSYNC_START,
2607 		"ufs_fsync_start:vp %p", vp);
2608 
2609 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK);
2610 	if (error)
2611 		return (error);
2612 
2613 	if (TRANS_ISTRANS(ufsvfsp)) {
2614 		/*
2615 		 * First push out any data pages
2616 		 */
2617 		if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2618 		    (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) {
2619 			error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0,
2620 			    0, CRED());
2621 			if (error)
2622 				goto out;
2623 		}
2624 
2625 		/*
2626 		 * Delta any delayed inode times updates
2627 		 * and push inode to log.
2628 		 * All other inode deltas will have already been delta'd
2629 		 * and will be pushed during the commit.
2630 		 */
2631 		if (!(syncflag & FDSYNC) &&
2632 		    ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) {
2633 			if (ulp) {
2634 				TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC,
2635 				    TOP_SYNCIP_SIZE);
2636 			}
2637 			rw_enter(&ip->i_contents, RW_READER);
2638 			mutex_enter(&ip->i_tlock);
2639 			ip->i_flag &= ~IMODTIME;
2640 			mutex_exit(&ip->i_tlock);
2641 			ufs_iupdat(ip, I_SYNC);
2642 			rw_exit(&ip->i_contents);
2643 			if (ulp) {
2644 				TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC,
2645 				    TOP_SYNCIP_SIZE);
2646 			}
2647 		}
2648 
2649 		/*
2650 		 * Commit the Moby transaction
2651 		 *
2652 		 * Deltas have already been made so we just need to
2653 		 * commit them with a synchronous transaction.
2654 		 * TRANS_BEGIN_SYNC() will return an error
2655 		 * if there are no deltas to commit, for an
2656 		 * empty transaction.
2657 		 */
2658 		if (ulp) {
2659 			TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE,
2660 			    error);
2661 			if (error) {
2662 				error = 0; /* commit wasn't needed */
2663 				goto out;
2664 			}
2665 			TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC,
2666 			    TOP_COMMIT_SIZE);
2667 		}
2668 	} else {	/* not logging */
2669 		if (!(IS_SWAPVP(vp)))
2670 			if (syncflag & FNODSYNC) {
2671 				/* Just update the inode only */
2672 				TRANS_IUPDAT(ip, 1);
2673 				error = 0;
2674 			} else if (syncflag & FDSYNC)
2675 				/* Do data-synchronous writes */
2676 				error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC);
2677 			else
2678 				/* Do synchronous writes */
2679 				error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC);
2680 
2681 		rw_enter(&ip->i_contents, RW_WRITER);
2682 		if (!error)
2683 			error = ufs_sync_indir(ip);
2684 		rw_exit(&ip->i_contents);
2685 	}
2686 out:
2687 	if (ulp) {
2688 		ufs_lockfs_end(ulp);
2689 	}
2690 	TRACE_2(TR_FAC_UFS, TR_UFS_FSYNC_END,
2691 		"ufs_fsync_end:vp %p error %d", vp, error);
2692 	return (error);
2693 }
2694 
2695 /*ARGSUSED*/
2696 static void
2697 ufs_inactive(struct vnode *vp, struct cred *cr)
2698 {
2699 	ufs_iinactive(VTOI(vp));
2700 }
2701 
2702 /*
2703  * Unix file system operations having to do with directory manipulation.
2704  */
2705 int ufs_lookup_idle_count = 2;	/* Number of inodes to idle each time */
2706 /* ARGSUSED */
2707 static int
2708 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
2709 	struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr)
2710 {
2711 	struct inode *ip;
2712 	struct inode *sip;
2713 	struct inode *xip;
2714 	struct ufsvfs *ufsvfsp;
2715 	struct ulockfs *ulp;
2716 	struct vnode *vp;
2717 	int error;
2718 
2719 	TRACE_2(TR_FAC_UFS, TR_UFS_LOOKUP_START,
2720 		"ufs_lookup_start:dvp %p name %s", dvp, nm);
2721 
2722 
2723 	/*
2724 	 * Check flags for type of lookup (regular file or attribute file)
2725 	 */
2726 
2727 	ip = VTOI(dvp);
2728 
2729 	if (flags & LOOKUP_XATTR) {
2730 
2731 		/*
2732 		 * We don't allow recursive attributes...
2733 		 * Maybe someday we will.
2734 		 */
2735 		if ((ip->i_cflags & IXATTR)) {
2736 			return (EINVAL);
2737 		}
2738 
2739 		if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) {
2740 			error = ufs_xattr_getattrdir(dvp, &sip, flags, cr);
2741 			if (error) {
2742 				*vpp = NULL;
2743 				goto out;
2744 			}
2745 
2746 			vp = ITOV(sip);
2747 			dnlc_update(dvp, XATTR_DIR_NAME, vp);
2748 		}
2749 
2750 		/*
2751 		 * Check accessibility of directory.
2752 		 */
2753 		if (vp == DNLC_NO_VNODE) {
2754 			VN_RELE(vp);
2755 			error = ENOENT;
2756 			goto out;
2757 		}
2758 		if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr)) != 0) {
2759 			VN_RELE(vp);
2760 			goto out;
2761 		}
2762 
2763 		*vpp = vp;
2764 		return (0);
2765 	}
2766 
2767 	/*
2768 	 * Check for a null component, which we should treat as
2769 	 * looking at dvp from within it's parent, so we don't
2770 	 * need a call to ufs_iaccess(), as it has already been
2771 	 * done.
2772 	 */
2773 	if (nm[0] == 0) {
2774 		VN_HOLD(dvp);
2775 		error = 0;
2776 		*vpp = dvp;
2777 		goto out;
2778 	}
2779 
2780 	/*
2781 	 * Check for "." ie itself. this is a quick check and
2782 	 * avoids adding "." into the dnlc (which have been seen
2783 	 * to occupy >10% of the cache).
2784 	 */
2785 	if ((nm[0] == '.') && (nm[1] == 0)) {
2786 		/*
2787 		 * Don't return without checking accessibility
2788 		 * of the directory. We only need the lock if
2789 		 * we are going to return it.
2790 		 */
2791 		if ((error = ufs_iaccess(ip, IEXEC, cr)) == 0) {
2792 			VN_HOLD(dvp);
2793 			*vpp = dvp;
2794 		}
2795 		goto out;
2796 	}
2797 
2798 	/*
2799 	 * Fast path: Check the directory name lookup cache.
2800 	 */
2801 	if (vp = dnlc_lookup(dvp, nm)) {
2802 		/*
2803 		 * Check accessibility of directory.
2804 		 */
2805 		if ((error = ufs_iaccess(ip, IEXEC, cr)) != 0) {
2806 			VN_RELE(vp);
2807 			goto out;
2808 		}
2809 		if (vp == DNLC_NO_VNODE) {
2810 			VN_RELE(vp);
2811 			error = ENOENT;
2812 			goto out;
2813 		}
2814 		xip = VTOI(vp);
2815 		ulp = NULL;
2816 		goto fastpath;
2817 	}
2818 
2819 	/*
2820 	 * Keep the idle queue from getting too long by
2821 	 * idling two inodes before attempting to allocate another.
2822 	 *    This operation must be performed before entering
2823 	 *    lockfs or a transaction.
2824 	 */
2825 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2826 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2827 			ins.in_lidles.value.ul += ufs_lookup_idle_count;
2828 			ufs_idle_some(ufs_lookup_idle_count);
2829 		}
2830 
2831 retry_lookup:
2832 	ufsvfsp = ip->i_ufsvfs;
2833 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK);
2834 	if (error)
2835 		goto out;
2836 
2837 	error = ufs_dirlook(ip, nm, &xip, cr, 1);
2838 
2839 fastpath:
2840 	if (error == 0) {
2841 		ip = xip;
2842 		*vpp = ITOV(ip);
2843 
2844 		/*
2845 		 * If vnode is a device return special vnode instead.
2846 		 */
2847 		if (IS_DEVVP(*vpp)) {
2848 			struct vnode *newvp;
2849 
2850 			newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
2851 			    cr);
2852 			VN_RELE(*vpp);
2853 			if (newvp == NULL)
2854 				error = ENOSYS;
2855 			else
2856 				*vpp = newvp;
2857 		}
2858 	}
2859 	if (ulp) {
2860 		ufs_lockfs_end(ulp);
2861 	}
2862 
2863 	if (error == EAGAIN)
2864 		goto retry_lookup;
2865 
2866 out:
2867 	TRACE_3(TR_FAC_UFS, TR_UFS_LOOKUP_END,
2868 		"ufs_lookup_end:dvp %p name %s error %d", vpp, nm, error);
2869 	return (error);
2870 }
2871 
2872 static int
2873 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl,
2874 	int mode, struct vnode **vpp, struct cred *cr, int flag)
2875 {
2876 	struct inode *ip;
2877 	struct inode *xip;
2878 	struct inode *dip;
2879 	struct vnode *xvp;
2880 	struct ufsvfs *ufsvfsp;
2881 	struct ulockfs *ulp;
2882 	int error;
2883 	int issync;
2884 	int truncflag;
2885 	int trans_size;
2886 	int noentry;
2887 	int defer_dip_seq_update = 0;	/* need to defer update of dip->i_seq */
2888 	int retry = 1;
2889 	int indeadlock;
2890 
2891 	TRACE_1(TR_FAC_UFS, TR_UFS_CREATE_START,
2892 		"ufs_create_start:dvp %p", dvp);
2893 
2894 again:
2895 	ip = VTOI(dvp);
2896 	ufsvfsp = ip->i_ufsvfs;
2897 	truncflag = 0;
2898 
2899 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK);
2900 	if (error)
2901 		goto out;
2902 
2903 	if (ulp) {
2904 		trans_size = (int)TOP_CREATE_SIZE(ip);
2905 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size);
2906 	}
2907 
2908 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
2909 		vap->va_mode &= ~VSVTX;
2910 
2911 	if (*name == '\0') {
2912 		/*
2913 		 * Null component name refers to the directory itself.
2914 		 */
2915 		VN_HOLD(dvp);
2916 		/*
2917 		 * Even though this is an error case, we need to grab the
2918 		 * quota lock since the error handling code below is common.
2919 		 */
2920 		rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2921 		rw_enter(&ip->i_contents, RW_WRITER);
2922 		error = EEXIST;
2923 	} else {
2924 		xip = NULL;
2925 		noentry = 0;
2926 		/*
2927 		 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2928 		 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2929 		 * possible, retries the operation.
2930 		 */
2931 		ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_CREATE,
2932 					retry_dir);
2933 		if (indeadlock)
2934 			goto again;
2935 
2936 		xvp = dnlc_lookup(dvp, name);
2937 		if (xvp == DNLC_NO_VNODE) {
2938 			noentry = 1;
2939 			VN_RELE(xvp);
2940 			xvp = NULL;
2941 		}
2942 		if (xvp) {
2943 			rw_exit(&ip->i_rwlock);
2944 			if (error = ufs_iaccess(ip, IEXEC, cr)) {
2945 				VN_RELE(xvp);
2946 			} else {
2947 				error = EEXIST;
2948 				xip = VTOI(xvp);
2949 			}
2950 		} else {
2951 			/*
2952 			 * Suppress file system full message if we will retry
2953 			 */
2954 			error = ufs_direnter_cm(ip, name, DE_CREATE,
2955 				vap, &xip, cr,
2956 				(noentry | (retry ? IQUIET : 0)));
2957 			if (error == EAGAIN) {
2958 				if (ulp) {
2959 					TRANS_END_CSYNC(ufsvfsp, error, issync,
2960 						TOP_CREATE, trans_size);
2961 					ufs_lockfs_end(ulp);
2962 				}
2963 				goto again;
2964 			}
2965 			rw_exit(&ip->i_rwlock);
2966 		}
2967 		ip = xip;
2968 		if (ip != NULL) {
2969 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2970 			rw_enter(&ip->i_contents, RW_WRITER);
2971 		}
2972 	}
2973 
2974 	/*
2975 	 * If the file already exists and this is a non-exclusive create,
2976 	 * check permissions and allow access for non-directories.
2977 	 * Read-only create of an existing directory is also allowed.
2978 	 * We fail an exclusive create of anything which already exists.
2979 	 */
2980 	if (error == EEXIST) {
2981 		dip = VTOI(dvp);
2982 		if (excl == NONEXCL) {
2983 			if ((((ip->i_mode & IFMT) == IFDIR) ||
2984 			    ((ip->i_mode & IFMT) == IFATTRDIR)) &&
2985 			    (mode & IWRITE))
2986 				error = EISDIR;
2987 			else if (mode)
2988 				error = ufs_iaccess(ip, mode, cr);
2989 			else
2990 				error = 0;
2991 		}
2992 		if (error) {
2993 			rw_exit(&ip->i_contents);
2994 			rw_exit(&ufsvfsp->vfs_dqrwlock);
2995 			VN_RELE(ITOV(ip));
2996 			goto unlock;
2997 		}
2998 		/*
2999 		 * If the error EEXIST was set, then i_seq can not
3000 		 * have been updated. The sequence number interface
3001 		 * is defined such that a non-error VOP_CREATE must
3002 		 * increase the dir va_seq it by at least one. If we
3003 		 * have cleared the error, increase i_seq. Note that
3004 		 * we are increasing the dir i_seq and in rare cases
3005 		 * ip may actually be from the dvp, so we already have
3006 		 * the locks and it will not be subject to truncation.
3007 		 * In case we have to update i_seq of the parent
3008 		 * directory dip, we have to defer it till we have
3009 		 * released our locks on ip due to lock ordering requirements.
3010 		 */
3011 		if (ip != dip)
3012 			defer_dip_seq_update = 1;
3013 		else
3014 			ip->i_seq++;
3015 
3016 		if (((ip->i_mode & IFMT) == IFREG) &&
3017 		    (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
3018 			/*
3019 			 * Truncate regular files, if requested by caller.
3020 			 * Grab i_rwlock to make sure no one else is
3021 			 * currently writing to the file (we promised
3022 			 * bmap we would do this).
3023 			 * Must get the locks in the correct order.
3024 			 */
3025 			if (ip->i_size == 0) {
3026 				ip->i_flag |= ICHG | IUPD;
3027 				ip->i_seq++;
3028 				TRANS_INODE(ufsvfsp, ip);
3029 			} else {
3030 				/*
3031 				 * Large Files: Why this check here?
3032 				 * Though we do it in vn_create() we really
3033 				 * want to guarantee that we do not destroy
3034 				 * Large file data by atomically checking
3035 				 * the size while holding the contents
3036 				 * lock.
3037 				 */
3038 				if (flag && !(flag & FOFFMAX) &&
3039 				    ((ip->i_mode & IFMT) == IFREG) &&
3040 				    (ip->i_size > (offset_t)MAXOFF32_T)) {
3041 					rw_exit(&ip->i_contents);
3042 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3043 					error = EOVERFLOW;
3044 					goto unlock;
3045 				}
3046 				if (TRANS_ISTRANS(ufsvfsp))
3047 					truncflag++;
3048 				else {
3049 					rw_exit(&ip->i_contents);
3050 					rw_exit(&ufsvfsp->vfs_dqrwlock);
3051 					ufs_tryirwlock_trans(&ip->i_rwlock,
3052 							RW_WRITER, TOP_CREATE,
3053 							retry_file);
3054 					if (indeadlock) {
3055 						VN_RELE(ITOV(ip));
3056 						goto again;
3057 					}
3058 					rw_enter(&ufsvfsp->vfs_dqrwlock,
3059 							RW_READER);
3060 					rw_enter(&ip->i_contents, RW_WRITER);
3061 					(void) ufs_itrunc(ip, (u_offset_t)0, 0,
3062 								cr);
3063 					rw_exit(&ip->i_rwlock);
3064 				}
3065 			}
3066 		}
3067 	}
3068 
3069 	if (error) {
3070 		if (ip != NULL) {
3071 			rw_exit(&ufsvfsp->vfs_dqrwlock);
3072 			rw_exit(&ip->i_contents);
3073 		}
3074 		goto unlock;
3075 	}
3076 
3077 	*vpp = ITOV(ip);
3078 	ITIMES(ip);
3079 	rw_exit(&ip->i_contents);
3080 	rw_exit(&ufsvfsp->vfs_dqrwlock);
3081 
3082 	/*
3083 	 * If vnode is a device return special vnode instead.
3084 	 */
3085 	if (!error && IS_DEVVP(*vpp)) {
3086 		struct vnode *newvp;
3087 
3088 		newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
3089 		VN_RELE(*vpp);
3090 		if (newvp == NULL) {
3091 			error = ENOSYS;
3092 			goto unlock;
3093 		}
3094 		truncflag = 0;
3095 		*vpp = newvp;
3096 	}
3097 unlock:
3098 
3099 	/*
3100 	 * Do the deferred update of the parent directory's sequence
3101 	 * number now.
3102 	 */
3103 	if (defer_dip_seq_update == 1) {
3104 		rw_enter(&dip->i_contents, RW_READER);
3105 		mutex_enter(&dip->i_tlock);
3106 		dip->i_seq++;
3107 		mutex_exit(&dip->i_tlock);
3108 		rw_exit(&dip->i_contents);
3109 	}
3110 
3111 	if (ulp) {
3112 		int terr = 0;
3113 
3114 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE,
3115 		    trans_size);
3116 
3117 		/*
3118 		 * If we haven't had a more interesting failure
3119 		 * already, then anything that might've happened
3120 		 * here should be reported.
3121 		 */
3122 		if (error == 0)
3123 			error = terr;
3124 	}
3125 
3126 	if (!error && truncflag) {
3127 		ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_trunc);
3128 		if (indeadlock) {
3129 			if (ulp)
3130 				ufs_lockfs_end(ulp);
3131 			VN_RELE(ITOV(ip));
3132 			goto again;
3133 		}
3134 		(void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr);
3135 		rw_exit(&ip->i_rwlock);
3136 	}
3137 
3138 	if (ulp)
3139 		ufs_lockfs_end(ulp);
3140 
3141 	/*
3142 	 * If no inodes available, try to free one up out of the
3143 	 * pending delete queue.
3144 	 */
3145 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3146 		ufs_delete_drain_wait(ufsvfsp, 1);
3147 		retry = 0;
3148 		goto again;
3149 	}
3150 
3151 out:
3152 	TRACE_3(TR_FAC_UFS, TR_UFS_CREATE_END,
3153 		"ufs_create_end:dvp %p name %s error %d", vpp, name, error);
3154 	return (error);
3155 }
3156 
3157 extern int ufs_idle_max;
3158 /*ARGSUSED*/
3159 static int
3160 ufs_remove(struct vnode *vp, char *nm, struct cred *cr)
3161 {
3162 	struct inode *ip = VTOI(vp);
3163 	struct ufsvfs *ufsvfsp	= ip->i_ufsvfs;
3164 	struct ulockfs *ulp;
3165 	vnode_t *rmvp = NULL;	/* Vnode corresponding to name being removed */
3166 	int indeadlock;
3167 	int error;
3168 	int issync;
3169 	int trans_size;
3170 
3171 	TRACE_1(TR_FAC_UFS, TR_UFS_REMOVE_START,
3172 		"ufs_remove_start:vp %p", vp);
3173 
3174 	/*
3175 	 * don't let the delete queue get too long
3176 	 */
3177 	if (ufsvfsp == NULL) {
3178 		error = EIO;
3179 		goto out;
3180 	}
3181 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3182 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3183 
3184 retry_remove:
3185 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK);
3186 	if (error)
3187 		goto out;
3188 
3189 	if (ulp)
3190 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE,
3191 		    trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp)));
3192 
3193 	/*
3194 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3195 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3196 	 * possible, retries the operation.
3197 	 */
3198 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_REMOVE, retry);
3199 	if (indeadlock)
3200 		goto retry_remove;
3201 	error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0,
3202 	    DR_REMOVE, cr, &rmvp);
3203 	rw_exit(&ip->i_rwlock);
3204 
3205 	if (ulp) {
3206 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size);
3207 		ufs_lockfs_end(ulp);
3208 	}
3209 
3210 	/*
3211 	 * This must be called after the remove transaction is closed.
3212 	 */
3213 	if (rmvp != NULL) {
3214 		/* Only send the event if there were no errors */
3215 		if (error == 0)
3216 			vnevent_remove(rmvp);
3217 		VN_RELE(rmvp);
3218 	}
3219 out:
3220 	TRACE_3(TR_FAC_UFS, TR_UFS_REMOVE_END,
3221 		"ufs_remove_end:vp %p name %s error %d", vp, nm, error);
3222 	return (error);
3223 }
3224 
3225 /*
3226  * Link a file or a directory.  Only privileged processes are allowed to
3227  * make links to directories.
3228  */
3229 static int
3230 ufs_link(struct vnode *tdvp, struct vnode *svp, char *tnm, struct cred *cr)
3231 {
3232 	struct inode *sip;
3233 	struct inode *tdp = VTOI(tdvp);
3234 	struct ufsvfs *ufsvfsp = tdp->i_ufsvfs;
3235 	struct ulockfs *ulp;
3236 	struct vnode *realvp;
3237 	int error;
3238 	int issync;
3239 	int trans_size;
3240 	int isdev;
3241 	int indeadlock;
3242 
3243 	TRACE_1(TR_FAC_UFS, TR_UFS_LINK_START,
3244 		"ufs_link_start:tdvp %p", tdvp);
3245 
3246 retry_link:
3247 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LINK_MASK);
3248 	if (error)
3249 		goto out;
3250 
3251 	if (ulp)
3252 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_LINK,
3253 		    trans_size = (int)TOP_LINK_SIZE(VTOI(tdvp)));
3254 
3255 	if (VOP_REALVP(svp, &realvp) == 0)
3256 		svp = realvp;
3257 
3258 	/*
3259 	 * Make sure link for extended attributes is valid
3260 	 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3261 	 *
3262 	 * Make certain we don't attempt to look at a device node as
3263 	 * a ufs inode.
3264 	 */
3265 
3266 	isdev = IS_DEVVP(svp);
3267 	if (((isdev == 0) && ((VTOI(svp)->i_cflags & IXATTR) == 0) &&
3268 	    ((tdp->i_mode & IFMT) == IFATTRDIR)) ||
3269 	    ((isdev == 0) && (VTOI(svp)->i_cflags & IXATTR) &&
3270 	    ((tdp->i_mode & IFMT) == IFDIR))) {
3271 		error = EINVAL;
3272 		goto unlock;
3273 	}
3274 
3275 	sip = VTOI(svp);
3276 	if ((svp->v_type == VDIR &&
3277 	    secpolicy_fs_linkdir(cr, ufsvfsp->vfs_vfs) != 0) ||
3278 	    (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)) {
3279 		error = EPERM;
3280 		goto unlock;
3281 	}
3282 
3283 	/*
3284 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3285 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3286 	 * possible, retries the operation.
3287 	 */
3288 	ufs_tryirwlock_trans(&tdp->i_rwlock, RW_WRITER, TOP_LINK, retry);
3289 	if (indeadlock)
3290 		goto retry_link;
3291 	error = ufs_direnter_lr(tdp, tnm, DE_LINK, (struct inode *)0,
3292 	    sip, cr, NULL);
3293 	rw_exit(&tdp->i_rwlock);
3294 
3295 unlock:
3296 	if (ulp) {
3297 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_LINK, trans_size);
3298 		ufs_lockfs_end(ulp);
3299 	}
3300 out:
3301 	TRACE_2(TR_FAC_UFS, TR_UFS_LINK_END,
3302 		"ufs_link_end:tdvp %p error %d", tdvp, error);
3303 	return (error);
3304 }
3305 
3306 uint64_t ufs_rename_retry_cnt;
3307 uint64_t ufs_rename_upgrade_retry_cnt;
3308 uint64_t ufs_rename_dircheck_retry_cnt;
3309 clock_t	 ufs_rename_backoff_delay = 1;
3310 
3311 /*
3312  * Rename a file or directory.
3313  * We are given the vnode and entry string of the source and the
3314  * vnode and entry string of the place we want to move the source
3315  * to (the target). The essential operation is:
3316  *	unlink(target);
3317  *	link(source, target);
3318  *	unlink(source);
3319  * but "atomically".  Can't do full commit without saving state in
3320  * the inode on disk, which isn't feasible at this time.  Best we
3321  * can do is always guarantee that the TARGET exists.
3322  */
3323 
3324 /*ARGSUSED*/
3325 static int
3326 ufs_rename(
3327 	struct vnode *sdvp,		/* old (source) parent vnode */
3328 	char *snm,			/* old (source) entry name */
3329 	struct vnode *tdvp,		/* new (target) parent vnode */
3330 	char *tnm,			/* new (target) entry name */
3331 	struct cred *cr)
3332 {
3333 	struct inode *sip = NULL;	/* source inode */
3334 	struct inode *ip = NULL;	/* check inode */
3335 	struct inode *sdp;		/* old (source) parent inode */
3336 	struct inode *tdp;		/* new (target) parent inode */
3337 	struct vnode *tvp = NULL;	/* target vnode, if it exists */
3338 	struct vnode *realvp;
3339 	struct ufsvfs *ufsvfsp;
3340 	struct ulockfs *ulp;
3341 	struct ufs_slot slot;
3342 	timestruc_t now;
3343 	int error;
3344 	int issync;
3345 	int trans_size;
3346 	krwlock_t *first_lock;
3347 	krwlock_t *second_lock;
3348 	krwlock_t *reverse_lock;
3349 
3350 	TRACE_1(TR_FAC_UFS, TR_UFS_RENAME_START,
3351 		"ufs_rename_start:sdvp %p", sdvp);
3352 
3353 
3354 	sdp = VTOI(sdvp);
3355 	slot.fbp = NULL;
3356 	ufsvfsp = sdp->i_ufsvfs;
3357 retry_rename:
3358 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
3359 	if (error)
3360 		goto out;
3361 
3362 	if (ulp)
3363 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
3364 		    trans_size = (int)TOP_RENAME_SIZE(sdp));
3365 
3366 	if (VOP_REALVP(tdvp, &realvp) == 0)
3367 		tdvp = realvp;
3368 
3369 	tdp = VTOI(tdvp);
3370 
3371 
3372 	/*
3373 	 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3374 	 */
3375 	if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) {
3376 		error = EINVAL;
3377 		goto unlock;
3378 	}
3379 
3380 	/*
3381 	 * Look up inode of file we're supposed to rename.
3382 	 */
3383 	gethrestime(&now);
3384 	if (error = ufs_dirlook(sdp, snm, &sip, cr, 0)) {
3385 		if (error == EAGAIN) {
3386 			if (ulp) {
3387 				TRANS_END_CSYNC(ufsvfsp, error, issync,
3388 					TOP_RENAME, trans_size);
3389 				ufs_lockfs_end(ulp);
3390 			}
3391 			goto retry_rename;
3392 		}
3393 
3394 		goto unlock;
3395 	}
3396 
3397 	/*
3398 	 * Lock both the source and target directories (they may be
3399 	 * the same) to provide the atomicity semantics that was
3400 	 * previously provided by the per file system vfs_rename_lock
3401 	 *
3402 	 * with vfs_rename_lock removed to allow simultaneous renames
3403 	 * within a file system, ufs_dircheckpath can deadlock while
3404 	 * traversing back to ensure that source is not a parent directory
3405 	 * of target parent directory. This is because we get into
3406 	 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3407 	 * If the tdp and sdp of the simultaneous renames happen to be
3408 	 * in the path of each other, it can lead to a deadlock. This
3409 	 * can be avoided by getting the locks as RW_READER here and then
3410 	 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3411 	 *
3412 	 * We hold the target directory's i_rwlock after calling
3413 	 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3414 	 * VOP_RWLOCK is explicitly called by the filesystem independent code
3415 	 * before calling the file system operation. In these cases the order
3416 	 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3417 	 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3418 	 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3419 	 * synchronizing object which might lead to a deadlock. So we use
3420 	 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3421 	 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3422 	 * operation.
3423 	 */
3424 retry:
3425 	first_lock = &tdp->i_rwlock;
3426 	second_lock = &sdp->i_rwlock;
3427 retry_firstlock:
3428 	if (!rw_tryenter(first_lock, RW_READER)) {
3429 		/*
3430 		 * We didn't get the lock. Check if the SLOCK is set in the
3431 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3432 		 * and wait for SLOCK to be cleared.
3433 		 */
3434 
3435 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3436 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3437 					trans_size);
3438 			ufs_lockfs_end(ulp);
3439 			goto retry_rename;
3440 
3441 		} else {
3442 			/*
3443 			 * SLOCK isn't set so this is a genuine synchronization
3444 			 * case. Let's try again after giving them a breather.
3445 			 */
3446 			delay(RETRY_LOCK_DELAY);
3447 			goto  retry_firstlock;
3448 		}
3449 	}
3450 	/*
3451 	 * Need to check if the tdp and sdp are same !!!
3452 	 */
3453 	if ((tdp != sdp) && (!rw_tryenter(second_lock, RW_READER))) {
3454 		/*
3455 		 * We didn't get the lock. Check if the SLOCK is set in the
3456 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3457 		 * and wait for SLOCK to be cleared.
3458 		 */
3459 
3460 		rw_exit(first_lock);
3461 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3462 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3463 					trans_size);
3464 			ufs_lockfs_end(ulp);
3465 			goto retry_rename;
3466 
3467 		} else {
3468 			/*
3469 			 * So we couldn't get the second level peer lock *and*
3470 			 * the SLOCK bit isn't set. Too bad we can be
3471 			 * contentding with someone wanting these locks otherway
3472 			 * round. Reverse the locks in case there is a heavy
3473 			 * contention for the second level lock.
3474 			 */
3475 			reverse_lock = first_lock;
3476 			first_lock = second_lock;
3477 			second_lock = reverse_lock;
3478 			ufs_rename_retry_cnt++;
3479 			goto  retry_firstlock;
3480 		}
3481 	}
3482 
3483 	if (sip == tdp) {
3484 		error = EINVAL;
3485 		goto errout;
3486 	}
3487 	/*
3488 	 * Make sure we can delete the source entry.  This requires
3489 	 * write permission on the containing directory.
3490 	 * Check for sticky directories.
3491 	 */
3492 	rw_enter(&sdp->i_contents, RW_READER);
3493 	rw_enter(&sip->i_contents, RW_READER);
3494 	if ((error = ufs_iaccess(sdp, IWRITE, cr)) != 0 ||
3495 	    (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) {
3496 		rw_exit(&sip->i_contents);
3497 		rw_exit(&sdp->i_contents);
3498 		goto errout;
3499 	}
3500 
3501 	/*
3502 	 * If this is a rename of a directory and the parent is
3503 	 * different (".." must be changed), then the source
3504 	 * directory must not be in the directory hierarchy
3505 	 * above the target, as this would orphan everything
3506 	 * below the source directory.  Also the user must have
3507 	 * write permission in the source so as to be able to
3508 	 * change "..".
3509 	 */
3510 	if ((((sip->i_mode & IFMT) == IFDIR) ||
3511 	    ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) {
3512 		ino_t	inum;
3513 
3514 		if ((error = ufs_iaccess(sip, IWRITE, cr))) {
3515 			rw_exit(&sip->i_contents);
3516 			rw_exit(&sdp->i_contents);
3517 			goto errout;
3518 		}
3519 		inum = sip->i_number;
3520 		rw_exit(&sip->i_contents);
3521 		rw_exit(&sdp->i_contents);
3522 		if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) {
3523 			/*
3524 			 * If we got EAGAIN ufs_dircheckpath detected a
3525 			 * potential deadlock and backed out. We need
3526 			 * to retry the operation since sdp and tdp have
3527 			 * to be released to avoid the deadlock.
3528 			 */
3529 			if (error == EAGAIN) {
3530 				rw_exit(&tdp->i_rwlock);
3531 				if (tdp != sdp)
3532 					rw_exit(&sdp->i_rwlock);
3533 				delay(ufs_rename_backoff_delay);
3534 				ufs_rename_dircheck_retry_cnt++;
3535 				goto retry;
3536 			}
3537 			goto errout;
3538 		}
3539 	} else {
3540 		rw_exit(&sip->i_contents);
3541 		rw_exit(&sdp->i_contents);
3542 	}
3543 
3544 
3545 	/*
3546 	 * Check for renaming '.' or '..' or alias of '.'
3547 	 */
3548 	if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) {
3549 		error = EINVAL;
3550 		goto errout;
3551 	}
3552 
3553 	/*
3554 	 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3555 	 * tries to traverse back the file tree with both tdp and sdp held
3556 	 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3557 	 * as RW_READERS  till ufs_dircheckpath is done.
3558 	 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3559 	 * to RW_WRITER.
3560 	 */
3561 	if (!rw_tryupgrade(&tdp->i_rwlock)) {
3562 		/*
3563 		 * The upgrade failed. We got to give away the lock
3564 		 * as to avoid deadlocking with someone else who is
3565 		 * waiting for writer lock. With the lock gone, we
3566 		 * cannot be sure the checks done above will hold
3567 		 * good when we eventually get them back as writer.
3568 		 * So if we can't upgrade we drop the locks and retry
3569 		 * everything again.
3570 		 */
3571 		rw_exit(&tdp->i_rwlock);
3572 		if (tdp != sdp)
3573 			rw_exit(&sdp->i_rwlock);
3574 		delay(ufs_rename_backoff_delay);
3575 		ufs_rename_upgrade_retry_cnt++;
3576 		goto retry;
3577 	}
3578 	if (tdp != sdp) {
3579 		if (!rw_tryupgrade(&sdp->i_rwlock)) {
3580 			/*
3581 			 * The upgrade failed. We got to give away the lock
3582 			 * as to avoid deadlocking with someone else who is
3583 			 * waiting for writer lock. With the lock gone, we
3584 			 * cannot be sure the checks done above will hold
3585 			 * good when we eventually get them back as writer.
3586 			 * So if we can't upgrade we drop the locks and retry
3587 			 * everything again.
3588 			 */
3589 			rw_exit(&tdp->i_rwlock);
3590 			rw_exit(&sdp->i_rwlock);
3591 			delay(ufs_rename_backoff_delay);
3592 			ufs_rename_upgrade_retry_cnt++;
3593 			goto retry;
3594 		}
3595 	}
3596 
3597 	/*
3598 	 * Now that all the locks are held check to make sure another thread
3599 	 * didn't slip in and take out the sip.
3600 	 */
3601 	slot.status = NONE;
3602 	if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec ||
3603 	    sip->i_ctime.tv_sec > now.tv_sec) {
3604 		rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER);
3605 		rw_enter(&sdp->i_contents, RW_WRITER);
3606 		error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot,
3607 		    &ip, cr, 0);
3608 		rw_exit(&sdp->i_contents);
3609 		rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock);
3610 		if (error) {
3611 			goto errout;
3612 		}
3613 		if (ip == NULL) {
3614 			error = ENOENT;
3615 			goto errout;
3616 		} else {
3617 			/*
3618 			 * If the inode was found need to drop the v_count
3619 			 * so as not to keep the filesystem from being
3620 			 * unmounted at a later time.
3621 			 */
3622 			VN_RELE(ITOV(ip));
3623 		}
3624 
3625 		/*
3626 		 * Release the slot.fbp that has the page mapped and
3627 		 * locked SE_SHARED, and could be used in in
3628 		 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3629 		 * on said page.
3630 		 */
3631 		if (slot.fbp) {
3632 			fbrelse(slot.fbp, S_OTHER);
3633 			slot.fbp = NULL;
3634 		}
3635 	}
3636 
3637 	/*
3638 	 * Link source to the target.  If a target exists, return its
3639 	 * vnode pointer in tvp.  We'll release it after sending the
3640 	 * vnevent.
3641 	 */
3642 	if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr, &tvp)) {
3643 		/*
3644 		 * ESAME isn't really an error; it indicates that the
3645 		 * operation should not be done because the source and target
3646 		 * are the same file, but that no error should be reported.
3647 		 */
3648 		if (error == ESAME)
3649 			error = 0;
3650 		goto errout;
3651 	}
3652 
3653 	/*
3654 	 * Unlink the source.
3655 	 * Remove the source entry.  ufs_dirremove() checks that the entry
3656 	 * still reflects sip, and returns an error if it doesn't.
3657 	 * If the entry has changed just forget about it.  Release
3658 	 * the source inode.
3659 	 */
3660 	if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0,
3661 	    DR_RENAME, cr, NULL)) == ENOENT)
3662 		error = 0;
3663 
3664 errout:
3665 	if (slot.fbp)
3666 		fbrelse(slot.fbp, S_OTHER);
3667 
3668 	rw_exit(&tdp->i_rwlock);
3669 	if (sdp != tdp) {
3670 		rw_exit(&sdp->i_rwlock);
3671 	}
3672 
3673 unlock:
3674 	if (ulp) {
3675 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
3676 		ufs_lockfs_end(ulp);
3677 	}
3678 
3679 	/*
3680 	 * If no errors, send the appropriate events on the source
3681 	 * and destination (a.k.a, target) vnodes, if they exist.
3682 	 * This has to be done after the rename transaction has closed.
3683 	 */
3684 	if (error == 0) {
3685 		if (tvp != NULL)
3686 			vnevent_rename_dest(tvp);
3687 		/*
3688 		 * Note that if ufs_direnter_lr() returned ESAME then
3689 		 * this event will still be sent.  This isn't expected
3690 		 * to be a problem for anticipated usage by consumers.
3691 		 */
3692 		if (sip != NULL)
3693 			vnevent_rename_src(ITOV(sip));
3694 	}
3695 
3696 	if (tvp != NULL)
3697 		VN_RELE(tvp);
3698 
3699 	if (sip != NULL)
3700 		VN_RELE(ITOV(sip));
3701 
3702 out:
3703 	TRACE_5(TR_FAC_UFS, TR_UFS_RENAME_END,
3704 		"ufs_rename_end:sdvp %p snm %s tdvp %p tnm %s error %d",
3705 			sdvp, snm, tdvp, tnm, error);
3706 	return (error);
3707 }
3708 
3709 /*ARGSUSED*/
3710 static int
3711 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap,
3712 	struct vnode **vpp, struct cred *cr)
3713 {
3714 	struct inode *ip;
3715 	struct inode *xip;
3716 	struct ufsvfs *ufsvfsp;
3717 	struct ulockfs *ulp;
3718 	int error;
3719 	int issync;
3720 	int trans_size;
3721 	int indeadlock;
3722 	int retry = 1;
3723 
3724 	ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
3725 
3726 	TRACE_1(TR_FAC_UFS, TR_UFS_MKDIR_START,
3727 		"ufs_mkdir_start:dvp %p", dvp);
3728 
3729 	/*
3730 	 * Can't make directory in attr hidden dir
3731 	 */
3732 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3733 		return (EINVAL);
3734 
3735 again:
3736 	ip = VTOI(dvp);
3737 	ufsvfsp = ip->i_ufsvfs;
3738 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK);
3739 	if (error)
3740 		goto out;
3741 	if (ulp)
3742 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR,
3743 		    trans_size = (int)TOP_MKDIR_SIZE(ip));
3744 
3745 	/*
3746 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3747 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3748 	 * possible, retries the operation.
3749 	 */
3750 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_MKDIR, retry);
3751 	if (indeadlock)
3752 		goto again;
3753 
3754 	error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr,
3755 		(retry ? IQUIET : 0));
3756 	if (error == EAGAIN) {
3757 		if (ulp) {
3758 			TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_MKDIR,
3759 					trans_size);
3760 			ufs_lockfs_end(ulp);
3761 		}
3762 		goto again;
3763 	}
3764 
3765 	rw_exit(&ip->i_rwlock);
3766 	if (error == 0) {
3767 		ip = xip;
3768 		*vpp = ITOV(ip);
3769 	} else if (error == EEXIST)
3770 		VN_RELE(ITOV(xip));
3771 
3772 	if (ulp) {
3773 		int terr = 0;
3774 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size);
3775 		ufs_lockfs_end(ulp);
3776 		if (error == 0)
3777 			error = terr;
3778 	}
3779 out:
3780 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3781 		ufs_delete_drain_wait(ufsvfsp, 1);
3782 		retry = 0;
3783 		goto again;
3784 	}
3785 
3786 	TRACE_2(TR_FAC_UFS, TR_UFS_MKDIR_END,
3787 		"ufs_mkdir_end:dvp %p error %d", dvp, error);
3788 	return (error);
3789 }
3790 
3791 /*ARGSUSED*/
3792 static int
3793 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr)
3794 {
3795 	struct inode *ip = VTOI(vp);
3796 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3797 	struct ulockfs *ulp;
3798 	vnode_t *rmvp = NULL;	/* Vnode of removed directory */
3799 	int error;
3800 	int issync;
3801 	int trans_size;
3802 	int indeadlock;
3803 
3804 	TRACE_1(TR_FAC_UFS, TR_UFS_RMDIR_START,
3805 		"ufs_rmdir_start:vp %p", vp);
3806 
3807 	/*
3808 	 * don't let the delete queue get too long
3809 	 */
3810 	if (ufsvfsp == NULL) {
3811 		error = EIO;
3812 		goto out;
3813 	}
3814 	if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3815 		ufs_delete_drain(vp->v_vfsp, 1, 1);
3816 
3817 retry_rmdir:
3818 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK);
3819 	if (error)
3820 		goto out;
3821 
3822 	if (ulp)
3823 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR,
3824 					trans_size = TOP_RMDIR_SIZE);
3825 
3826 	/*
3827 	 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3828 	 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3829 	 * possible, retries the operation.
3830 	 */
3831 	ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_RMDIR, retry);
3832 	if (indeadlock)
3833 		goto retry_rmdir;
3834 	error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr,
3835 									&rmvp);
3836 	rw_exit(&ip->i_rwlock);
3837 
3838 	if (ulp) {
3839 		TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR,
3840 				trans_size);
3841 		ufs_lockfs_end(ulp);
3842 	}
3843 
3844 	/*
3845 	 * This must be done AFTER the rmdir transaction has closed.
3846 	 */
3847 	if (rmvp != NULL) {
3848 		/* Only send the event if there were no errors */
3849 		if (error == 0)
3850 			vnevent_rmdir(rmvp);
3851 		VN_RELE(rmvp);
3852 	}
3853 out:
3854 	TRACE_2(TR_FAC_UFS, TR_UFS_RMDIR_END,
3855 		"ufs_rmdir_end:vp %p error %d", vp, error);
3856 
3857 	return (error);
3858 }
3859 
3860 /* ARGSUSED */
3861 static int
3862 ufs_readdir(
3863 	struct vnode *vp,
3864 	struct uio *uiop,
3865 	struct cred *cr,
3866 	int *eofp)
3867 {
3868 	struct iovec *iovp;
3869 	struct inode *ip;
3870 	struct direct *idp;
3871 	struct dirent64 *odp;
3872 	struct fbuf *fbp;
3873 	struct ufsvfs *ufsvfsp;
3874 	struct ulockfs *ulp;
3875 	caddr_t outbuf;
3876 	size_t bufsize;
3877 	uint_t offset;
3878 	uint_t bytes_wanted, total_bytes_wanted;
3879 	int incount = 0;
3880 	int outcount = 0;
3881 	int error;
3882 
3883 	ip = VTOI(vp);
3884 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
3885 
3886 	TRACE_2(TR_FAC_UFS, TR_UFS_READDIR_START,
3887 		"ufs_readdir_start:vp %p uiop %p", vp, uiop);
3888 
3889 	if (uiop->uio_loffset >= MAXOFF32_T) {
3890 		if (eofp)
3891 			*eofp = 1;
3892 		return (0);
3893 	}
3894 
3895 	/*
3896 	 * Check if we have been called with a valid iov_len
3897 	 * and bail out if not, otherwise we may potentially loop
3898 	 * forever further down.
3899 	 */
3900 	if (uiop->uio_iov->iov_len <= 0) {
3901 		error = EINVAL;
3902 		goto out;
3903 	}
3904 
3905 	/*
3906 	 * Large Files: When we come here we are guaranteed that
3907 	 * uio_offset can be used safely. The high word is zero.
3908 	 */
3909 
3910 	ufsvfsp = ip->i_ufsvfs;
3911 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READDIR_MASK);
3912 	if (error)
3913 		goto out;
3914 
3915 	iovp = uiop->uio_iov;
3916 	total_bytes_wanted = iovp->iov_len;
3917 
3918 	/* Large Files: directory files should not be "large" */
3919 
3920 	ASSERT(ip->i_size <= MAXOFF32_T);
3921 
3922 	/* Force offset to be valid (to guard against bogus lseek() values) */
3923 	offset = (uint_t)uiop->uio_offset & ~(DIRBLKSIZ - 1);
3924 
3925 	/* Quit if at end of file or link count of zero (posix) */
3926 	if (offset >= (uint_t)ip->i_size || ip->i_nlink <= 0) {
3927 		if (eofp)
3928 			*eofp = 1;
3929 		error = 0;
3930 		goto unlock;
3931 	}
3932 
3933 	/*
3934 	 * Get space to change directory entries into fs independent format.
3935 	 * Do fast alloc for the most commonly used-request size (filesystem
3936 	 * block size).
3937 	 */
3938 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) {
3939 		bufsize = total_bytes_wanted;
3940 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
3941 		odp = (struct dirent64 *)outbuf;
3942 	} else {
3943 		bufsize = total_bytes_wanted;
3944 		odp = (struct dirent64 *)iovp->iov_base;
3945 	}
3946 
3947 nextblk:
3948 	bytes_wanted = total_bytes_wanted;
3949 
3950 	/* Truncate request to file size */
3951 	if (offset + bytes_wanted > (int)ip->i_size)
3952 		bytes_wanted = (int)(ip->i_size - offset);
3953 
3954 	/* Comply with MAXBSIZE boundary restrictions of fbread() */
3955 	if ((offset & MAXBOFFSET) + bytes_wanted > MAXBSIZE)
3956 		bytes_wanted = MAXBSIZE - (offset & MAXBOFFSET);
3957 
3958 	/*
3959 	 * Read in the next chunk.
3960 	 * We are still holding the i_rwlock.
3961 	 */
3962 	error = fbread(vp, (offset_t)offset, bytes_wanted, S_OTHER, &fbp);
3963 
3964 	if (error)
3965 		goto update_inode;
3966 	if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (ip->i_fs->fs_ronly == 0) &&
3967 	    (!ufsvfsp->vfs_noatime)) {
3968 		ip->i_flag |= IACC;
3969 	}
3970 	incount = 0;
3971 	idp = (struct direct *)fbp->fb_addr;
3972 	if (idp->d_ino == 0 && idp->d_reclen == 0 &&
3973 		idp->d_namlen == 0) {
3974 		cmn_err(CE_WARN, "ufs_readdir: bad dir, inumber = %llu, "
3975 			"fs = %s\n",
3976 			(u_longlong_t)ip->i_number, ufsvfsp->vfs_fs->fs_fsmnt);
3977 		fbrelse(fbp, S_OTHER);
3978 		error = ENXIO;
3979 		goto update_inode;
3980 	}
3981 	/* Transform to file-system independent format */
3982 	while (incount < bytes_wanted) {
3983 		/*
3984 		 * If the current directory entry is mangled, then skip
3985 		 * to the next block.  It would be nice to set the FSBAD
3986 		 * flag in the super-block so that a fsck is forced on
3987 		 * next reboot, but locking is a problem.
3988 		 */
3989 		if (idp->d_reclen & 0x3) {
3990 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
3991 			break;
3992 		}
3993 
3994 		/* Skip to requested offset and skip empty entries */
3995 		if (idp->d_ino != 0 && offset >= (uint_t)uiop->uio_offset) {
3996 			ushort_t this_reclen =
3997 			    DIRENT64_RECLEN(idp->d_namlen);
3998 			/* Buffer too small for any entries */
3999 			if (!outcount && this_reclen > bufsize) {
4000 				fbrelse(fbp, S_OTHER);
4001 				error = EINVAL;
4002 				goto update_inode;
4003 			}
4004 			/* If would overrun the buffer, quit */
4005 			if (outcount + this_reclen > bufsize) {
4006 				break;
4007 			}
4008 			/* Take this entry */
4009 			odp->d_ino = (ino64_t)idp->d_ino;
4010 			odp->d_reclen = (ushort_t)this_reclen;
4011 			odp->d_off = (offset_t)(offset + idp->d_reclen);
4012 
4013 			/* use strncpy(9f) to zero out uninitialized bytes */
4014 
4015 			ASSERT(strlen(idp->d_name) + 1 <=
4016 			    DIRENT64_NAMELEN(this_reclen));
4017 			(void) strncpy(odp->d_name, idp->d_name,
4018 			    DIRENT64_NAMELEN(this_reclen));
4019 			outcount += odp->d_reclen;
4020 			odp = (struct dirent64 *)((intptr_t)odp +
4021 				    odp->d_reclen);
4022 			ASSERT(outcount <= bufsize);
4023 		}
4024 		if (idp->d_reclen) {
4025 			incount += idp->d_reclen;
4026 			offset += idp->d_reclen;
4027 			idp = (struct direct *)((intptr_t)idp + idp->d_reclen);
4028 		} else {
4029 			offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4030 			break;
4031 		}
4032 	}
4033 	/* Release the chunk */
4034 	fbrelse(fbp, S_OTHER);
4035 
4036 	/* Read whole block, but got no entries, read another if not eof */
4037 
4038 	/*
4039 	 * Large Files: casting i_size to int here is not a problem
4040 	 * because directory sizes are always less than MAXOFF32_T.
4041 	 * See assertion above.
4042 	 */
4043 
4044 	if (offset < (int)ip->i_size && !outcount)
4045 		goto nextblk;
4046 
4047 	/* Copy out the entry data */
4048 	if (uiop->uio_segflg == UIO_SYSSPACE && uiop->uio_iovcnt == 1) {
4049 		iovp->iov_base += outcount;
4050 		iovp->iov_len -= outcount;
4051 		uiop->uio_resid -= outcount;
4052 		uiop->uio_offset = offset;
4053 	} else if ((error = uiomove(outbuf, (long)outcount, UIO_READ,
4054 				    uiop)) == 0)
4055 		uiop->uio_offset = offset;
4056 update_inode:
4057 	ITIMES(ip);
4058 	if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1)
4059 		kmem_free(outbuf, bufsize);
4060 
4061 	if (eofp && error == 0)
4062 		*eofp = (uiop->uio_offset >= (int)ip->i_size);
4063 unlock:
4064 	if (ulp) {
4065 		ufs_lockfs_end(ulp);
4066 	}
4067 out:
4068 	TRACE_2(TR_FAC_UFS, TR_UFS_READDIR_END,
4069 		"ufs_readdir_end:vp %p error %d", vp, error);
4070 	return (error);
4071 }
4072 
4073 /*ARGSUSED*/
4074 static int
4075 ufs_symlink(
4076 	struct vnode *dvp,		/* ptr to parent dir vnode */
4077 	char *linkname,			/* name of symbolic link */
4078 	struct vattr *vap,		/* attributes */
4079 	char *target,			/* target path */
4080 	struct cred *cr)		/* user credentials */
4081 {
4082 	struct inode *ip, *dip = VTOI(dvp);
4083 	struct ufsvfs *ufsvfsp = dip->i_ufsvfs;
4084 	struct ulockfs *ulp;
4085 	int error;
4086 	int issync;
4087 	int trans_size;
4088 	int residual;
4089 	int ioflag;
4090 	int retry = 1;
4091 
4092 	TRACE_1(TR_FAC_UFS, TR_UFS_SYMLINK_START,
4093 		"ufs_symlink_start:dvp %p", dvp);
4094 
4095 	/*
4096 	 * No symlinks in attrdirs at this time
4097 	 */
4098 	if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
4099 		return (EINVAL);
4100 
4101 again:
4102 	ip = (struct inode *)NULL;
4103 	vap->va_type = VLNK;
4104 	vap->va_rdev = 0;
4105 
4106 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK);
4107 	if (error)
4108 		goto out;
4109 
4110 	if (ulp)
4111 		TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK,
4112 		    trans_size = (int)TOP_SYMLINK_SIZE(dip));
4113 
4114 	/*
4115 	 * We must create the inode before the directory entry, to avoid
4116 	 * racing with readlink().  ufs_dirmakeinode requires that we
4117 	 * hold the quota lock as reader, and directory locks as writer.
4118 	 */
4119 
4120 	rw_enter(&dip->i_rwlock, RW_WRITER);
4121 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4122 	rw_enter(&dip->i_contents, RW_WRITER);
4123 
4124 	/*
4125 	 * Suppress any out of inodes messages if we will retry on
4126 	 * ENOSP
4127 	 */
4128 	if (retry)
4129 		dip->i_flag |= IQUIET;
4130 
4131 	error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr);
4132 
4133 	dip->i_flag &= ~IQUIET;
4134 
4135 	rw_exit(&dip->i_contents);
4136 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4137 	rw_exit(&dip->i_rwlock);
4138 
4139 	if (error)
4140 		goto unlock;
4141 
4142 	/*
4143 	 * OK.  The inode has been created.  Write out the data of the
4144 	 * symbolic link.  Since symbolic links are metadata, and should
4145 	 * remain consistent across a system crash, we need to force the
4146 	 * data out synchronously.
4147 	 *
4148 	 * (This is a change from the semantics in earlier releases, which
4149 	 * only created symbolic links synchronously if the semi-documented
4150 	 * 'syncdir' option was set, or if we were being invoked by the NFS
4151 	 * server, which requires symbolic links to be created synchronously.)
4152 	 *
4153 	 * We need to pass in a pointer for the residual length; otherwise
4154 	 * ufs_rdwri() will always return EIO if it can't write the data,
4155 	 * even if the error was really ENOSPC or EDQUOT.
4156 	 */
4157 
4158 	ioflag = FWRITE | FDSYNC;
4159 	residual = 0;
4160 
4161 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4162 	rw_enter(&ip->i_contents, RW_WRITER);
4163 
4164 	/*
4165 	 * Suppress file system full messages if we will retry
4166 	 */
4167 	if (retry)
4168 		ip->i_flag |= IQUIET;
4169 
4170 	error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target),
4171 	    (offset_t)0, UIO_SYSSPACE, &residual, cr);
4172 
4173 	ip->i_flag &= ~IQUIET;
4174 
4175 	if (error) {
4176 		rw_exit(&ip->i_contents);
4177 		rw_exit(&ufsvfsp->vfs_dqrwlock);
4178 		goto remove;
4179 	}
4180 
4181 	/*
4182 	 * If the link's data is small enough, we can cache it in the inode.
4183 	 * This is a "fast symbolic link".  We don't use the first direct
4184 	 * block because that's actually used to point at the symbolic link's
4185 	 * contents on disk; but we know that none of the other direct or
4186 	 * indirect blocks can be used because symbolic links are restricted
4187 	 * to be smaller than a file system block.
4188 	 */
4189 
4190 	ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip)));
4191 
4192 	if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) {
4193 		if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) {
4194 			ip->i_flag |= IFASTSYMLNK;
4195 		} else {
4196 			int i;
4197 			/* error, clear garbage left behind */
4198 			for (i = 1; i < NDADDR; i++)
4199 				ip->i_db[i] = 0;
4200 			for (i = 0; i < NIADDR; i++)
4201 				ip->i_ib[i] = 0;
4202 		}
4203 	}
4204 
4205 	rw_exit(&ip->i_contents);
4206 	rw_exit(&ufsvfsp->vfs_dqrwlock);
4207 
4208 	/*
4209 	 * OK.  We've successfully created the symbolic link.  All that
4210 	 * remains is to insert it into the appropriate directory.
4211 	 */
4212 
4213 	rw_enter(&dip->i_rwlock, RW_WRITER);
4214 	error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr, NULL);
4215 	rw_exit(&dip->i_rwlock);
4216 
4217 	/*
4218 	 * Fall through into remove-on-error code.  We're either done, or we
4219 	 * need to remove the inode (if we couldn't insert it).
4220 	 */
4221 
4222 remove:
4223 	if (error && (ip != NULL)) {
4224 		rw_enter(&ip->i_contents, RW_WRITER);
4225 		ip->i_nlink--;
4226 		ip->i_flag |= ICHG;
4227 		ip->i_seq++;
4228 		ufs_setreclaim(ip);
4229 		rw_exit(&ip->i_contents);
4230 	}
4231 
4232 unlock:
4233 	if (ip != NULL)
4234 		VN_RELE(ITOV(ip));
4235 
4236 	if (ulp) {
4237 		int terr = 0;
4238 
4239 		TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK,
4240 				trans_size);
4241 		ufs_lockfs_end(ulp);
4242 		if (error == 0)
4243 			error = terr;
4244 	}
4245 
4246 	/*
4247 	 * We may have failed due to lack of an inode or of a block to
4248 	 * store the target in.  Try flushing the delete queue to free
4249 	 * logically-available things up and try again.
4250 	 */
4251 	if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
4252 		ufs_delete_drain_wait(ufsvfsp, 1);
4253 		retry = 0;
4254 		goto again;
4255 	}
4256 
4257 out:
4258 	TRACE_2(TR_FAC_UFS, TR_UFS_SYMLINK_END,
4259 		"ufs_symlink_end:dvp %p error %d", dvp, error);
4260 	return (error);
4261 }
4262 
4263 /*
4264  * Ufs specific routine used to do ufs io.
4265  */
4266 int
4267 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base,
4268 	ssize_t len, offset_t offset, enum uio_seg seg, int *aresid,
4269 	struct cred *cr)
4270 {
4271 	struct uio auio;
4272 	struct iovec aiov;
4273 	int error;
4274 
4275 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
4276 
4277 	bzero((caddr_t)&auio, sizeof (uio_t));
4278 	bzero((caddr_t)&aiov, sizeof (iovec_t));
4279 
4280 	aiov.iov_base = base;
4281 	aiov.iov_len = len;
4282 	auio.uio_iov = &aiov;
4283 	auio.uio_iovcnt = 1;
4284 	auio.uio_loffset = offset;
4285 	auio.uio_segflg = (short)seg;
4286 	auio.uio_resid = len;
4287 
4288 	if (rw == UIO_WRITE) {
4289 		auio.uio_fmode = FWRITE;
4290 		auio.uio_extflg = UIO_COPY_DEFAULT;
4291 		auio.uio_llimit = curproc->p_fsz_ctl;
4292 		error = wrip(ip, &auio, ioflag, cr);
4293 	} else {
4294 		auio.uio_fmode = FREAD;
4295 		auio.uio_extflg = UIO_COPY_CACHED;
4296 		auio.uio_llimit = MAXOFFSET_T;
4297 		error = rdip(ip, &auio, ioflag, cr);
4298 	}
4299 
4300 	if (aresid) {
4301 		*aresid = auio.uio_resid;
4302 	} else if (auio.uio_resid) {
4303 		error = EIO;
4304 	}
4305 	return (error);
4306 }
4307 
4308 static int
4309 ufs_fid(vp, fidp)
4310 	struct vnode *vp;
4311 	struct fid *fidp;
4312 {
4313 	struct ufid *ufid;
4314 	struct inode *ip = VTOI(vp);
4315 
4316 	if (ip->i_ufsvfs == NULL)
4317 		return (EIO);
4318 
4319 	if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) {
4320 		fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t);
4321 		return (ENOSPC);
4322 	}
4323 
4324 	ufid = (struct ufid *)fidp;
4325 	bzero((char *)ufid, sizeof (struct ufid));
4326 	ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t);
4327 	ufid->ufid_ino = ip->i_number;
4328 	ufid->ufid_gen = ip->i_gen;
4329 
4330 	return (0);
4331 }
4332 
4333 /* ARGSUSED2 */
4334 static int
4335 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4336 {
4337 	struct inode	*ip = VTOI(vp);
4338 	struct ufsvfs	*ufsvfsp;
4339 	int		forcedirectio;
4340 
4341 	/*
4342 	 * Read case is easy.
4343 	 */
4344 	if (!write_lock) {
4345 		rw_enter(&ip->i_rwlock, RW_READER);
4346 		return (V_WRITELOCK_FALSE);
4347 	}
4348 
4349 	/*
4350 	 * Caller has requested a writer lock, but that inhibits any
4351 	 * concurrency in the VOPs that follow. Acquire the lock shared
4352 	 * and defer exclusive access until it is known to be needed in
4353 	 * other VOP handlers. Some cases can be determined here.
4354 	 */
4355 
4356 	/*
4357 	 * If directio is not set, there is no chance of concurrency,
4358 	 * so just acquire the lock exclusive. Beware of a forced
4359 	 * unmount before looking at the mount option.
4360 	 */
4361 	ufsvfsp = ip->i_ufsvfs;
4362 	forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0;
4363 	if (!(ip->i_flag & IDIRECTIO || forcedirectio) ||
4364 	    !ufs_allow_shared_writes) {
4365 		rw_enter(&ip->i_rwlock, RW_WRITER);
4366 		return (V_WRITELOCK_TRUE);
4367 	}
4368 
4369 	/*
4370 	 * Mandatory locking forces acquiring i_rwlock exclusive.
4371 	 */
4372 	if (MANDLOCK(vp, ip->i_mode)) {
4373 		rw_enter(&ip->i_rwlock, RW_WRITER);
4374 		return (V_WRITELOCK_TRUE);
4375 	}
4376 
4377 	/*
4378 	 * Acquire the lock shared in case a concurrent write follows.
4379 	 * Mandatory locking could have become enabled before the lock
4380 	 * was acquired. Re-check and upgrade if needed.
4381 	 */
4382 	rw_enter(&ip->i_rwlock, RW_READER);
4383 	if (MANDLOCK(vp, ip->i_mode)) {
4384 		rw_exit(&ip->i_rwlock);
4385 		rw_enter(&ip->i_rwlock, RW_WRITER);
4386 		return (V_WRITELOCK_TRUE);
4387 	}
4388 	return (V_WRITELOCK_FALSE);
4389 }
4390 
4391 /*ARGSUSED*/
4392 static void
4393 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4394 {
4395 	struct inode	*ip = VTOI(vp);
4396 
4397 	rw_exit(&ip->i_rwlock);
4398 }
4399 
4400 /* ARGSUSED */
4401 static int
4402 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp)
4403 {
4404 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4405 }
4406 
4407 /* ARGSUSED */
4408 static int
4409 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4410 	offset_t offset, struct flk_callback *flk_cbp, struct cred *cr)
4411 {
4412 	struct inode *ip = VTOI(vp);
4413 
4414 	if (ip->i_ufsvfs == NULL)
4415 		return (EIO);
4416 
4417 	/*
4418 	 * If file is being mapped, disallow frlock.
4419 	 * XXX I am not holding tlock while checking i_mapcnt because the
4420 	 * current locking strategy drops all locks before calling fs_frlock.
4421 	 * So, mapcnt could change before we enter fs_frlock making is
4422 	 * meaningless to have held tlock in the first place.
4423 	 */
4424 	if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode))
4425 		return (EAGAIN);
4426 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr));
4427 }
4428 
4429 /* ARGSUSED */
4430 static int
4431 ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4432 	offset_t offset, cred_t *cr, caller_context_t *ct)
4433 {
4434 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
4435 	struct ulockfs *ulp;
4436 	int error;
4437 
4438 	if ((error = convoff(vp, bfp, 0, offset)) == 0) {
4439 		if (cmd == F_FREESP) {
4440 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4441 			    ULOCKFS_SPACE_MASK);
4442 			if (error)
4443 				return (error);
4444 			error = ufs_freesp(vp, bfp, flag, cr);
4445 		} else if (cmd == F_ALLOCSP) {
4446 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
4447 			    ULOCKFS_FALLOCATE_MASK);
4448 			if (error)
4449 				return (error);
4450 			error = ufs_allocsp(vp, bfp, cr);
4451 		} else
4452 			return (EINVAL); /* Command not handled here */
4453 
4454 		if (ulp)
4455 			ufs_lockfs_end(ulp);
4456 
4457 	}
4458 	return (error);
4459 }
4460 
4461 /*
4462  * Used to determine if read ahead should be done. Also used to
4463  * to determine when write back occurs.
4464  */
4465 #define	CLUSTSZ(ip)		((ip)->i_ufsvfs->vfs_ioclustsz)
4466 
4467 /*
4468  * A faster version of ufs_getpage.
4469  *
4470  * We optimize by inlining the pvn_getpages iterator, eliminating
4471  * calls to bmap_read if file doesn't have UFS holes, and avoiding
4472  * the overhead of page_exists().
4473  *
4474  * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4475  * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4476  * victimizes performance when a file with UFS holes is faulted
4477  * first in the S_READ mode, and then in the S_WRITE mode. We will get
4478  * two MMU faults in this case.
4479  *
4480  * XXX - the inode fields which control the sequential mode are not
4481  *	 protected by any mutex. The read ahead will act wild if
4482  *	 multiple processes will access the file concurrently and
4483  *	 some of them in sequential mode. One particulary bad case
4484  *	 is if another thread will change the value of i_nextrio between
4485  *	 the time this thread tests the i_nextrio value and then reads it
4486  *	 again to use it as the offset for the read ahead.
4487  */
4488 static int
4489 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
4490 	page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr,
4491 	enum seg_rw rw, struct cred *cr)
4492 {
4493 	u_offset_t	uoff = (u_offset_t)off; /* type conversion */
4494 	u_offset_t	pgoff;
4495 	u_offset_t	eoff;
4496 	struct inode 	*ip = VTOI(vp);
4497 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
4498 	struct fs 	*fs;
4499 	struct ulockfs	*ulp;
4500 	page_t		**pl;
4501 	caddr_t		pgaddr;
4502 	krw_t		rwtype;
4503 	int 		err;
4504 	int		has_holes;
4505 	int		beyond_eof;
4506 	int		seqmode;
4507 	int		pgsize = PAGESIZE;
4508 	int		dolock;
4509 	int		do_qlock;
4510 	int		trans_size;
4511 
4512 	TRACE_1(TR_FAC_UFS, TR_UFS_GETPAGE_START,
4513 		"ufs_getpage_start:vp %p", vp);
4514 
4515 	ASSERT((uoff & PAGEOFFSET) == 0);
4516 
4517 	if (protp)
4518 		*protp = PROT_ALL;
4519 
4520 	/*
4521 	 * Obey the lockfs protocol
4522 	 */
4523 	err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg,
4524 			rw == S_READ || rw == S_EXEC, protp);
4525 	if (err)
4526 		goto out;
4527 
4528 	fs = ufsvfsp->vfs_fs;
4529 
4530 	if (ulp && (rw == S_CREATE || rw == S_WRITE) &&
4531 	    !(vp->v_flag & VISSWAP)) {
4532 		/*
4533 		 * Try to start a transaction, will return if blocking is
4534 		 * expected to occur and the address space is not the
4535 		 * kernel address space.
4536 		 */
4537 		trans_size = TOP_GETPAGE_SIZE(ip);
4538 		if (seg->s_as != &kas) {
4539 			TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE,
4540 				trans_size, err)
4541 			if (err == EWOULDBLOCK) {
4542 				/*
4543 				 * Use EDEADLK here because the VM code
4544 				 * can normally never see this error.
4545 				 */
4546 				err = EDEADLK;
4547 				ufs_lockfs_end(ulp);
4548 				goto out;
4549 			}
4550 		} else {
4551 			TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4552 		}
4553 	}
4554 
4555 	if (vp->v_flag & VNOMAP) {
4556 		err = ENOSYS;
4557 		goto unlock;
4558 	}
4559 
4560 	seqmode = ip->i_nextr == uoff && rw != S_CREATE;
4561 
4562 	rwtype = RW_READER;		/* start as a reader */
4563 	dolock = (rw_owner(&ip->i_contents) != curthread);
4564 	/*
4565 	 * If this thread owns the lock, i.e., this thread grabbed it
4566 	 * as writer somewhere above, then we don't need to grab the
4567 	 * lock as reader in this routine.
4568 	 */
4569 	do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread);
4570 
4571 retrylock:
4572 	if (dolock) {
4573 		/*
4574 		 * Grab the quota lock if we need to call
4575 		 * bmap_write() below (with i_contents as writer).
4576 		 */
4577 		if (do_qlock && rwtype == RW_WRITER)
4578 			rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4579 		rw_enter(&ip->i_contents, rwtype);
4580 	}
4581 
4582 	/*
4583 	 * We may be getting called as a side effect of a bmap using
4584 	 * fbread() when the blocks might be being allocated and the
4585 	 * size has not yet been up'ed.  In this case we want to be
4586 	 * able to return zero pages if we get back UFS_HOLE from
4587 	 * calling bmap for a non write case here.  We also might have
4588 	 * to read some frags from the disk into a page if we are
4589 	 * extending the number of frags for a given lbn in bmap().
4590 	 * Large Files: The read of i_size here is atomic because
4591 	 * i_contents is held here. If dolock is zero, the lock
4592 	 * is held in bmap routines.
4593 	 */
4594 	beyond_eof = uoff + len > ip->i_size + PAGEOFFSET;
4595 	if (beyond_eof && seg != segkmap) {
4596 		if (dolock) {
4597 			rw_exit(&ip->i_contents);
4598 			if (do_qlock && rwtype == RW_WRITER)
4599 				rw_exit(&ufsvfsp->vfs_dqrwlock);
4600 		}
4601 		err = EFAULT;
4602 		goto unlock;
4603 	}
4604 
4605 	/*
4606 	 * Must hold i_contents lock throughout the call to pvn_getpages
4607 	 * since locked pages are returned from each call to ufs_getapage.
4608 	 * Must *not* return locked pages and then try for contents lock
4609 	 * due to lock ordering requirements (inode > page)
4610 	 */
4611 
4612 	has_holes = bmap_has_holes(ip);
4613 
4614 	if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) {
4615 		int	blk_size;
4616 		u_offset_t offset;
4617 
4618 		/*
4619 		 * We must acquire the RW_WRITER lock in order to
4620 		 * call bmap_write().
4621 		 */
4622 		if (dolock && rwtype == RW_READER) {
4623 			rwtype = RW_WRITER;
4624 
4625 			/*
4626 			 * Grab the quota lock before
4627 			 * upgrading i_contents, but if we can't grab it
4628 			 * don't wait here due to lock order:
4629 			 * vfs_dqrwlock > i_contents.
4630 			 */
4631 			if (do_qlock && rw_tryenter(&ufsvfsp->vfs_dqrwlock,
4632 							RW_READER) == 0) {
4633 				rw_exit(&ip->i_contents);
4634 				goto retrylock;
4635 			}
4636 			if (!rw_tryupgrade(&ip->i_contents)) {
4637 				rw_exit(&ip->i_contents);
4638 				if (do_qlock)
4639 					rw_exit(&ufsvfsp->vfs_dqrwlock);
4640 				goto retrylock;
4641 			}
4642 		}
4643 
4644 		/*
4645 		 * May be allocating disk blocks for holes here as
4646 		 * a result of mmap faults. write(2) does the bmap_write
4647 		 * in rdip/wrip, not here. We are not dealing with frags
4648 		 * in this case.
4649 		 */
4650 		/*
4651 		 * Large Files: We cast fs_bmask field to offset_t
4652 		 * just as we do for MAXBMASK because uoff is a 64-bit
4653 		 * data type. fs_bmask will still be a 32-bit type
4654 		 * as we cannot change any ondisk data structures.
4655 		 */
4656 
4657 		offset = uoff & (offset_t)fs->fs_bmask;
4658 		while (offset < uoff + len) {
4659 			blk_size = (int)blksize(fs, ip, lblkno(fs, offset));
4660 			err = bmap_write(ip, offset, blk_size,
4661 			    BI_NORMAL, NULL, cr);
4662 			if (ip->i_flag & (ICHG|IUPD))
4663 				ip->i_seq++;
4664 			if (err)
4665 				goto update_inode;
4666 			offset += blk_size; /* XXX - make this contig */
4667 		}
4668 	}
4669 
4670 	/*
4671 	 * Can be a reader from now on.
4672 	 */
4673 	if (dolock && rwtype == RW_WRITER) {
4674 		rw_downgrade(&ip->i_contents);
4675 		/*
4676 		 * We can release vfs_dqrwlock early so do it, but make
4677 		 * sure we don't try to release it again at the bottom.
4678 		 */
4679 		if (do_qlock) {
4680 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4681 			do_qlock = 0;
4682 		}
4683 	}
4684 
4685 	/*
4686 	 * We remove PROT_WRITE in cases when the file has UFS holes
4687 	 * because we don't  want to call bmap_read() to check each
4688 	 * page if it is backed with a disk block.
4689 	 */
4690 	if (protp && has_holes && rw != S_WRITE && rw != S_CREATE)
4691 		*protp &= ~PROT_WRITE;
4692 
4693 	err = 0;
4694 
4695 	/*
4696 	 * The loop looks up pages in the range [off, off + len).
4697 	 * For each page, we first check if we should initiate an asynchronous
4698 	 * read ahead before we call page_lookup (we may sleep in page_lookup
4699 	 * for a previously initiated disk read).
4700 	 */
4701 	eoff = (uoff + len);
4702 	for (pgoff = uoff, pgaddr = addr, pl = plarr;
4703 	    pgoff < eoff; /* empty */) {
4704 		page_t	*pp;
4705 		u_offset_t	nextrio;
4706 		se_t	se;
4707 		int retval;
4708 
4709 		se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED);
4710 
4711 		/* Handle async getpage (faultahead) */
4712 		if (plarr == NULL) {
4713 			ip->i_nextrio = pgoff;
4714 			(void) ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4715 			pgoff += pgsize;
4716 			pgaddr += pgsize;
4717 			continue;
4718 		}
4719 		/*
4720 		 * Check if we should initiate read ahead of next cluster.
4721 		 * We call page_exists only when we need to confirm that
4722 		 * we have the current page before we initiate the read ahead.
4723 		 */
4724 		nextrio = ip->i_nextrio;
4725 		if (seqmode &&
4726 		    pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
4727 		    nextrio < ip->i_size && page_exists(vp, pgoff)) {
4728 			retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4729 			/*
4730 			 * We always read ahead the next cluster of data
4731 			 * starting from i_nextrio. If the page (vp,nextrio)
4732 			 * is actually in core at this point, the routine
4733 			 * ufs_getpage_ra() will stop pre-fetching data
4734 			 * until we read that page in a synchronized manner
4735 			 * through ufs_getpage_miss(). So, we should increase
4736 			 * i_nextrio if the page (vp, nextrio) exists.
4737 			 */
4738 			if ((retval == 0) && page_exists(vp, nextrio)) {
4739 				ip->i_nextrio = nextrio + pgsize;
4740 			}
4741 		}
4742 
4743 		if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
4744 			/*
4745 			 * We found the page in the page cache.
4746 			 */
4747 			*pl++ = pp;
4748 			pgoff += pgsize;
4749 			pgaddr += pgsize;
4750 			len -= pgsize;
4751 			plsz -= pgsize;
4752 		} else  {
4753 			/*
4754 			 * We have to create the page, or read it from disk.
4755 			 */
4756 			if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr,
4757 			    pl, plsz, rw, seqmode))
4758 				goto error;
4759 
4760 			while (*pl != NULL) {
4761 				pl++;
4762 				pgoff += pgsize;
4763 				pgaddr += pgsize;
4764 				len -= pgsize;
4765 				plsz -= pgsize;
4766 			}
4767 		}
4768 	}
4769 
4770 	/*
4771 	 * Return pages up to plsz if they are in the page cache.
4772 	 * We cannot return pages if there is a chance that they are
4773 	 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4774 	 */
4775 	if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
4776 
4777 		ASSERT((protp == NULL) ||
4778 			!(has_holes && (*protp & PROT_WRITE)));
4779 
4780 		eoff = pgoff + plsz;
4781 		while (pgoff < eoff) {
4782 			page_t		*pp;
4783 
4784 			if ((pp = page_lookup_nowait(vp, pgoff,
4785 			    SE_SHARED)) == NULL)
4786 				break;
4787 
4788 			*pl++ = pp;
4789 			pgoff += pgsize;
4790 			plsz -= pgsize;
4791 		}
4792 	}
4793 
4794 	if (plarr)
4795 		*pl = NULL;			/* Terminate page list */
4796 	ip->i_nextr = pgoff;
4797 
4798 error:
4799 	if (err && plarr) {
4800 		/*
4801 		 * Release any pages we have locked.
4802 		 */
4803 		while (pl > &plarr[0])
4804 			page_unlock(*--pl);
4805 
4806 		plarr[0] = NULL;
4807 	}
4808 
4809 update_inode:
4810 	/*
4811 	 * If the inode is not already marked for IACC (in rdip() for read)
4812 	 * and the inode is not marked for no access time update (in wrip()
4813 	 * for write) then update the inode access time and mod time now.
4814 	 */
4815 	if ((ip->i_flag & (IACC | INOACC)) == 0) {
4816 		if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) {
4817 			if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
4818 			    (fs->fs_ronly == 0) &&
4819 			    (!ufsvfsp->vfs_noatime)) {
4820 				mutex_enter(&ip->i_tlock);
4821 				ip->i_flag |= IACC;
4822 				ITIMES_NOLOCK(ip);
4823 				mutex_exit(&ip->i_tlock);
4824 			}
4825 		}
4826 	}
4827 
4828 	if (dolock) {
4829 		rw_exit(&ip->i_contents);
4830 		if (do_qlock && rwtype == RW_WRITER)
4831 			rw_exit(&ufsvfsp->vfs_dqrwlock);
4832 	}
4833 
4834 unlock:
4835 	if (ulp) {
4836 		if ((rw == S_CREATE || rw == S_WRITE) &&
4837 		    !(vp->v_flag & VISSWAP)) {
4838 			TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4839 		}
4840 		ufs_lockfs_end(ulp);
4841 	}
4842 out:
4843 	TRACE_2(TR_FAC_UFS, TR_UFS_GETPAGE_END,
4844 		"ufs_getpage_end:vp %p error %d", vp, err);
4845 	return (err);
4846 }
4847 
4848 /*
4849  * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4850  * cache. The page is either read from the disk, or it's created.
4851  * A page is created (without disk read) if rw == S_CREATE, or if
4852  * the page is not backed with a real disk block (UFS hole).
4853  */
4854 /* ARGSUSED */
4855 static int
4856 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg,
4857 	caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq)
4858 {
4859 	struct inode	*ip = VTOI(vp);
4860 	page_t		*pp;
4861 	daddr_t		bn;
4862 	size_t		io_len;
4863 	int		crpage = 0;
4864 	int		err;
4865 	int		contig;
4866 	int		bsize = ip->i_fs->fs_bsize;
4867 
4868 	/*
4869 	 * Figure out whether the page can be created, or must be
4870 	 * must be read from the disk.
4871 	 */
4872 	if (rw == S_CREATE)
4873 		crpage = 1;
4874 	else {
4875 		contig = 0;
4876 		if (err = bmap_read(ip, off, &bn, &contig))
4877 			return (err);
4878 
4879 		crpage = (bn == UFS_HOLE);
4880 
4881 		/*
4882 		 * If its also a fallocated block that hasn't been written to
4883 		 * yet, we will treat it just like a UFS_HOLE and create
4884 		 * a zero page for it
4885 		 */
4886 		if (ISFALLOCBLK(ip, bn))
4887 			crpage = 1;
4888 	}
4889 
4890 	if (crpage) {
4891 		if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg,
4892 		    addr)) == NULL) {
4893 			return (ufs_fault(vp,
4894 				    "ufs_getpage_miss: page_create == NULL"));
4895 		}
4896 
4897 		if (rw != S_CREATE)
4898 			pagezero(pp, 0, PAGESIZE);
4899 
4900 		io_len = PAGESIZE;
4901 	} else {
4902 		u_offset_t	io_off;
4903 		uint_t	xlen;
4904 		struct buf	*bp;
4905 		ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
4906 
4907 		/*
4908 		 * If access is not in sequential order, we read from disk
4909 		 * in bsize units.
4910 		 *
4911 		 * We limit the size of the transfer to bsize if we are reading
4912 		 * from the beginning of the file. Note in this situation we
4913 		 * will hedge our bets and initiate an async read ahead of
4914 		 * the second block.
4915 		 */
4916 		if (!seq || off == 0)
4917 			contig = MIN(contig, bsize);
4918 
4919 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4920 		    &io_len, off, contig, 0);
4921 
4922 		/*
4923 		 * Some other thread has entered the page.
4924 		 * ufs_getpage will retry page_lookup.
4925 		 */
4926 		if (pp == NULL) {
4927 			pl[0] = NULL;
4928 			return (0);
4929 		}
4930 
4931 		/*
4932 		 * Zero part of the page which we are not
4933 		 * going to read from the disk.
4934 		 */
4935 		xlen = io_len & PAGEOFFSET;
4936 		if (xlen != 0)
4937 			pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4938 
4939 		bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ);
4940 		bp->b_edev = ip->i_dev;
4941 		bp->b_dev = cmpdev(ip->i_dev);
4942 		bp->b_blkno = bn;
4943 		bp->b_un.b_addr = (caddr_t)0;
4944 		bp->b_file = ip->i_vnode;
4945 		bp->b_offset = off;
4946 
4947 		if (ufsvfsp->vfs_log) {
4948 			lufs_read_strategy(ufsvfsp->vfs_log, bp);
4949 		} else if (ufsvfsp->vfs_snapshot) {
4950 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4951 		} else {
4952 			ufsvfsp->vfs_iotstamp = lbolt;
4953 			ub.ub_getpages.value.ul++;
4954 			(void) bdev_strategy(bp);
4955 			lwp_stat_update(LWP_STAT_INBLK, 1);
4956 		}
4957 
4958 		ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK);
4959 
4960 		/*
4961 		 * If the file access is sequential, initiate read ahead
4962 		 * of the next cluster.
4963 		 */
4964 		if (seq && ip->i_nextrio < ip->i_size)
4965 			(void) ufs_getpage_ra(vp, off, seg, addr);
4966 		err = biowait(bp);
4967 		pageio_done(bp);
4968 
4969 		if (err) {
4970 			pvn_read_done(pp, B_ERROR);
4971 			return (err);
4972 		}
4973 	}
4974 
4975 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4976 	return (0);
4977 }
4978 
4979 /*
4980  * Read ahead a cluster from the disk. Returns the length in bytes.
4981  */
4982 static int
4983 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr)
4984 {
4985 	struct inode	*ip = VTOI(vp);
4986 	page_t		*pp;
4987 	u_offset_t	io_off = ip->i_nextrio;
4988 	ufsvfs_t	*ufsvfsp;
4989 	caddr_t		addr2 = addr + (io_off - off);
4990 	struct buf	*bp;
4991 	daddr_t		bn;
4992 	size_t		io_len;
4993 	int		err;
4994 	int		contig;
4995 	int		xlen;
4996 	int		bsize = ip->i_fs->fs_bsize;
4997 
4998 	/*
4999 	 * If the directio advisory is in effect on this file,
5000 	 * then do not do buffered read ahead. Read ahead makes
5001 	 * it more difficult on threads using directio as they
5002 	 * will be forced to flush the pages from this vnode.
5003 	 */
5004 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5005 		return (0);
5006 	if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio)
5007 		return (0);
5008 
5009 	/*
5010 	 * Is this test needed?
5011 	 */
5012 	if (addr2 >= seg->s_base + seg->s_size)
5013 		return (0);
5014 
5015 	contig = 0;
5016 	err = bmap_read(ip, io_off, &bn, &contig);
5017 	/*
5018 	 * If its a UFS_HOLE or a fallocated block, do not perform
5019 	 * any read ahead's since there probably is nothing to read ahead
5020 	 */
5021 	if (err || bn == UFS_HOLE || ISFALLOCBLK(ip, bn))
5022 		return (0);
5023 
5024 	/*
5025 	 * Limit the transfer size to bsize if this is the 2nd block.
5026 	 */
5027 	if (io_off == (u_offset_t)bsize)
5028 		contig = MIN(contig, bsize);
5029 
5030 	if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off,
5031 	    &io_len, io_off, contig, 1)) == NULL)
5032 		return (0);
5033 
5034 	/*
5035 	 * Zero part of page which we are not going to read from disk
5036 	 */
5037 	if ((xlen = (io_len & PAGEOFFSET)) > 0)
5038 		pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
5039 
5040 	ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK;
5041 
5042 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC);
5043 	bp->b_edev = ip->i_dev;
5044 	bp->b_dev = cmpdev(ip->i_dev);
5045 	bp->b_blkno = bn;
5046 	bp->b_un.b_addr = (caddr_t)0;
5047 	bp->b_file = ip->i_vnode;
5048 	bp->b_offset = off;
5049 
5050 	if (ufsvfsp->vfs_log) {
5051 		lufs_read_strategy(ufsvfsp->vfs_log, bp);
5052 	} else if (ufsvfsp->vfs_snapshot) {
5053 		fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5054 	} else {
5055 		ufsvfsp->vfs_iotstamp = lbolt;
5056 		ub.ub_getras.value.ul++;
5057 		(void) bdev_strategy(bp);
5058 		lwp_stat_update(LWP_STAT_INBLK, 1);
5059 	}
5060 
5061 	return (io_len);
5062 }
5063 
5064 int	ufs_delay = 1;
5065 /*
5066  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5067  *
5068  * LMXXX - the inode really ought to contain a pointer to one of these
5069  * async args.  Stuff gunk in there and just hand the whole mess off.
5070  * This would replace i_delaylen, i_delayoff.
5071  */
5072 /*ARGSUSED*/
5073 static int
5074 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
5075 	struct cred *cr)
5076 {
5077 	struct inode *ip = VTOI(vp);
5078 	int err = 0;
5079 
5080 	if (vp->v_count == 0) {
5081 		return (ufs_fault(vp, "ufs_putpage: bad v_count == 0"));
5082 	}
5083 
5084 	TRACE_1(TR_FAC_UFS, TR_UFS_PUTPAGE_START,
5085 		"ufs_putpage_start:vp %p", vp);
5086 
5087 	/*
5088 	 * XXX - Why should this check be made here?
5089 	 */
5090 	if (vp->v_flag & VNOMAP) {
5091 		err = ENOSYS;
5092 		goto errout;
5093 	}
5094 
5095 	if (ip->i_ufsvfs == NULL) {
5096 		err = EIO;
5097 		goto errout;
5098 	}
5099 
5100 	if (flags & B_ASYNC) {
5101 		if (ufs_delay && len &&
5102 		    (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
5103 			mutex_enter(&ip->i_tlock);
5104 			/*
5105 			 * If nobody stalled, start a new cluster.
5106 			 */
5107 			if (ip->i_delaylen == 0) {
5108 				ip->i_delayoff = off;
5109 				ip->i_delaylen = len;
5110 				mutex_exit(&ip->i_tlock);
5111 				goto errout;
5112 			}
5113 			/*
5114 			 * If we have a full cluster or they are not contig,
5115 			 * then push last cluster and start over.
5116 			 */
5117 			if (ip->i_delaylen >= CLUSTSZ(ip) ||
5118 			    ip->i_delayoff + ip->i_delaylen != off) {
5119 				u_offset_t doff;
5120 				size_t dlen;
5121 
5122 				doff = ip->i_delayoff;
5123 				dlen = ip->i_delaylen;
5124 				ip->i_delayoff = off;
5125 				ip->i_delaylen = len;
5126 				mutex_exit(&ip->i_tlock);
5127 				err = ufs_putpages(vp, doff, dlen,
5128 				    flags, cr);
5129 				/* LMXXX - flags are new val, not old */
5130 				goto errout;
5131 			}
5132 			/*
5133 			 * There is something there, it's not full, and
5134 			 * it is contig.
5135 			 */
5136 			ip->i_delaylen += len;
5137 			mutex_exit(&ip->i_tlock);
5138 			goto errout;
5139 		}
5140 		/*
5141 		 * Must have weird flags or we are not clustering.
5142 		 */
5143 	}
5144 
5145 	err = ufs_putpages(vp, off, len, flags, cr);
5146 
5147 errout:
5148 	TRACE_2(TR_FAC_UFS, TR_UFS_PUTPAGE_END,
5149 		"ufs_putpage_end:vp %p error %d", vp, err);
5150 	return (err);
5151 }
5152 
5153 /*
5154  * If len == 0, do from off to EOF.
5155  *
5156  * The normal cases should be len == 0 & off == 0 (entire vp list),
5157  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5158  * (from pageout).
5159  */
5160 /*ARGSUSED*/
5161 static int
5162 ufs_putpages(
5163 	struct vnode *vp,
5164 	offset_t off,
5165 	size_t len,
5166 	int flags,
5167 	struct cred *cr)
5168 {
5169 	u_offset_t io_off;
5170 	u_offset_t eoff;
5171 	struct inode *ip = VTOI(vp);
5172 	page_t *pp;
5173 	size_t io_len;
5174 	int err = 0;
5175 	int dolock;
5176 
5177 	if (vp->v_count == 0)
5178 		return (ufs_fault(vp, "ufs_putpages: v_count == 0"));
5179 	/*
5180 	 * Acquire the readers/write inode lock before locking
5181 	 * any pages in this inode.
5182 	 * The inode lock is held during i/o.
5183 	 */
5184 	if (len == 0) {
5185 		mutex_enter(&ip->i_tlock);
5186 		ip->i_delayoff = ip->i_delaylen = 0;
5187 		mutex_exit(&ip->i_tlock);
5188 	}
5189 	dolock = (rw_owner(&ip->i_contents) != curthread);
5190 	if (dolock) {
5191 		/*
5192 		 * Must synchronize this thread and any possible thread
5193 		 * operating in the window of vulnerability in wrip().
5194 		 * It is dangerous to allow both a thread doing a putpage
5195 		 * and a thread writing, so serialize them.  The exception
5196 		 * is when the thread in wrip() does something which causes
5197 		 * a putpage operation.  Then, the thread must be allowed
5198 		 * to continue.  It may encounter a bmap_read problem in
5199 		 * ufs_putapage, but that is handled in ufs_putapage.
5200 		 * Allow async writers to proceed, we don't want to block
5201 		 * the pageout daemon.
5202 		 */
5203 		if (ip->i_writer == curthread)
5204 			rw_enter(&ip->i_contents, RW_READER);
5205 		else {
5206 			for (;;) {
5207 				rw_enter(&ip->i_contents, RW_READER);
5208 				mutex_enter(&ip->i_tlock);
5209 				/*
5210 				 * If there is no thread in the critical
5211 				 * section of wrip(), then proceed.
5212 				 * Otherwise, wait until there isn't one.
5213 				 */
5214 				if (ip->i_writer == NULL) {
5215 					mutex_exit(&ip->i_tlock);
5216 					break;
5217 				}
5218 				rw_exit(&ip->i_contents);
5219 				/*
5220 				 * Bounce async writers when we have a writer
5221 				 * working on this file so we don't deadlock
5222 				 * the pageout daemon.
5223 				 */
5224 				if (flags & B_ASYNC) {
5225 					mutex_exit(&ip->i_tlock);
5226 					return (0);
5227 				}
5228 				cv_wait(&ip->i_wrcv, &ip->i_tlock);
5229 				mutex_exit(&ip->i_tlock);
5230 			}
5231 		}
5232 	}
5233 
5234 	if (!vn_has_cached_data(vp)) {
5235 		if (dolock)
5236 			rw_exit(&ip->i_contents);
5237 		return (0);
5238 	}
5239 
5240 	if (len == 0) {
5241 		/*
5242 		 * Search the entire vp list for pages >= off.
5243 		 */
5244 		err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage,
5245 					flags, cr);
5246 	} else {
5247 		/*
5248 		 * Loop over all offsets in the range looking for
5249 		 * pages to deal with.
5250 		 */
5251 		if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0)
5252 			eoff = MIN(off + len, eoff);
5253 		else
5254 			eoff = off + len;
5255 
5256 		for (io_off = off; io_off < eoff; io_off += io_len) {
5257 			/*
5258 			 * If we are not invalidating, synchronously
5259 			 * freeing or writing pages, use the routine
5260 			 * page_lookup_nowait() to prevent reclaiming
5261 			 * them from the free list.
5262 			 */
5263 			if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
5264 				pp = page_lookup(vp, io_off,
5265 					(flags & (B_INVAL | B_FREE)) ?
5266 					    SE_EXCL : SE_SHARED);
5267 			} else {
5268 				pp = page_lookup_nowait(vp, io_off,
5269 					(flags & B_FREE) ? SE_EXCL : SE_SHARED);
5270 			}
5271 
5272 			if (pp == NULL || pvn_getdirty(pp, flags) == 0)
5273 				io_len = PAGESIZE;
5274 			else {
5275 				u_offset_t *io_offp = &io_off;
5276 
5277 				err = ufs_putapage(vp, pp, io_offp, &io_len,
5278 				    flags, cr);
5279 				if (err != 0)
5280 					break;
5281 				/*
5282 				 * "io_off" and "io_len" are returned as
5283 				 * the range of pages we actually wrote.
5284 				 * This allows us to skip ahead more quickly
5285 				 * since several pages may've been dealt
5286 				 * with by this iteration of the loop.
5287 				 */
5288 			}
5289 		}
5290 	}
5291 	if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
5292 		/*
5293 		 * We have just sync'ed back all the pages on
5294 		 * the inode, turn off the IMODTIME flag.
5295 		 */
5296 		mutex_enter(&ip->i_tlock);
5297 		ip->i_flag &= ~IMODTIME;
5298 		mutex_exit(&ip->i_tlock);
5299 	}
5300 	if (dolock)
5301 		rw_exit(&ip->i_contents);
5302 	return (err);
5303 }
5304 
5305 static void
5306 ufs_iodone(buf_t *bp)
5307 {
5308 	struct inode *ip;
5309 
5310 	ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
5311 
5312 	bp->b_iodone = NULL;
5313 
5314 	ip = VTOI(bp->b_pages->p_vnode);
5315 
5316 	mutex_enter(&ip->i_tlock);
5317 	if (ip->i_writes >= ufs_LW) {
5318 		if ((ip->i_writes -= bp->b_bcount) <= ufs_LW)
5319 			if (ufs_WRITES)
5320 				cv_broadcast(&ip->i_wrcv); /* wake all up */
5321 	} else {
5322 		ip->i_writes -= bp->b_bcount;
5323 	}
5324 
5325 	mutex_exit(&ip->i_tlock);
5326 	iodone(bp);
5327 }
5328 
5329 /*
5330  * Write out a single page, possibly klustering adjacent
5331  * dirty pages.  The inode lock must be held.
5332  *
5333  * LMXXX - bsize < pagesize not done.
5334  */
5335 /*ARGSUSED*/
5336 int
5337 ufs_putapage(
5338 	struct vnode *vp,
5339 	page_t *pp,
5340 	u_offset_t *offp,
5341 	size_t *lenp,		/* return values */
5342 	int flags,
5343 	struct cred *cr)
5344 {
5345 	u_offset_t io_off;
5346 	u_offset_t off;
5347 	struct inode *ip = VTOI(vp);
5348 	struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
5349 	struct fs *fs;
5350 	struct buf *bp;
5351 	size_t io_len;
5352 	daddr_t bn;
5353 	int err;
5354 	int contig;
5355 
5356 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
5357 
5358 	TRACE_1(TR_FAC_UFS, TR_UFS_PUTAPAGE_START,
5359 		"ufs_putapage_start:vp %p", vp);
5360 
5361 	if (ufsvfsp == NULL) {
5362 		err = EIO;
5363 		goto out_trace;
5364 	}
5365 
5366 	fs = ip->i_fs;
5367 	ASSERT(fs->fs_ronly == 0);
5368 
5369 	/*
5370 	 * If the modified time on the inode has not already been
5371 	 * set elsewhere (e.g. for write/setattr) we set the time now.
5372 	 * This gives us approximate modified times for mmap'ed files
5373 	 * which are modified via stores in the user address space.
5374 	 */
5375 	if ((ip->i_flag & IMODTIME) == 0) {
5376 		mutex_enter(&ip->i_tlock);
5377 		ip->i_flag |= IUPD;
5378 		ip->i_seq++;
5379 		ITIMES_NOLOCK(ip);
5380 		mutex_exit(&ip->i_tlock);
5381 	}
5382 
5383 	/*
5384 	 * Align the request to a block boundry (for old file systems),
5385 	 * and go ask bmap() how contiguous things are for this file.
5386 	 */
5387 	off = pp->p_offset & (offset_t)fs->fs_bmask;	/* block align it */
5388 	contig = 0;
5389 	err = bmap_read(ip, off, &bn, &contig);
5390 	if (err)
5391 		goto out;
5392 	if (bn == UFS_HOLE) {			/* putpage never allocates */
5393 		/*
5394 		 * logging device is in error mode; simply return EIO
5395 		 */
5396 		if (TRANS_ISERROR(ufsvfsp)) {
5397 			err = EIO;
5398 			goto out;
5399 		}
5400 		/*
5401 		 * Oops, the thread in the window in wrip() did some
5402 		 * sort of operation which caused a putpage in the bad
5403 		 * range.  In this case, just return an error which will
5404 		 * cause the software modified bit on the page to set
5405 		 * and the page will get written out again later.
5406 		 */
5407 		if (ip->i_writer == curthread) {
5408 			err = EIO;
5409 			goto out;
5410 		}
5411 		/*
5412 		 * If the pager is trying to push a page in the bad range
5413 		 * just tell him to try again later when things are better.
5414 		 */
5415 		if (flags & B_ASYNC) {
5416 			err = EAGAIN;
5417 			goto out;
5418 		}
5419 		err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE");
5420 		goto out;
5421 	}
5422 
5423 	/*
5424 	 * If it is an fallocate'd block, reverse the negativity since
5425 	 * we are now writing to it
5426 	 */
5427 	if (ISFALLOCBLK(ip, bn)) {
5428 		err = bmap_set_bn(vp, off, dbtofsb(fs, -bn));
5429 		if (err)
5430 			goto out;
5431 
5432 		bn = -bn;
5433 	}
5434 
5435 	/*
5436 	 * Take the length (of contiguous bytes) passed back from bmap()
5437 	 * and _try_ and get a set of pages covering that extent.
5438 	 */
5439 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags);
5440 
5441 	/*
5442 	 * May have run out of memory and not clustered backwards.
5443 	 * off		p_offset
5444 	 * [  pp - 1  ][   pp   ]
5445 	 * [	block		]
5446 	 * We told bmap off, so we have to adjust the bn accordingly.
5447 	 */
5448 	if (io_off > off) {
5449 		bn += btod(io_off - off);
5450 		contig -= (io_off - off);
5451 	}
5452 
5453 	/*
5454 	 * bmap was carefull to tell us the right size so use that.
5455 	 * There might be unallocated frags at the end.
5456 	 * LMXXX - bzero the end of the page?  We must be writing after EOF.
5457 	 */
5458 	if (io_len > contig) {
5459 		ASSERT(io_len - contig < fs->fs_bsize);
5460 		io_len -= (io_len - contig);
5461 	}
5462 
5463 	/*
5464 	 * Handle the case where we are writing the last page after EOF.
5465 	 *
5466 	 * XXX - just a patch for i-mt3.
5467 	 */
5468 	if (io_len == 0) {
5469 		ASSERT(pp->p_offset >= (u_offset_t)(roundup(ip->i_size,
5470 							    PAGESIZE)));
5471 		io_len = PAGESIZE;
5472 	}
5473 
5474 	bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags);
5475 
5476 	ULOCKFS_SET_MOD(ITOUL(ip));
5477 
5478 	bp->b_edev = ip->i_dev;
5479 	bp->b_dev = cmpdev(ip->i_dev);
5480 	bp->b_blkno = bn;
5481 	bp->b_un.b_addr = (caddr_t)0;
5482 	bp->b_file = ip->i_vnode;
5483 
5484 	if (TRANS_ISTRANS(ufsvfsp)) {
5485 		if ((ip->i_mode & IFMT) == IFSHAD) {
5486 			TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD);
5487 		} else if (ufsvfsp->vfs_qinod == ip) {
5488 			TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR,
5489 			    0, 0);
5490 		}
5491 	}
5492 
5493 	/* write throttle */
5494 
5495 	ASSERT(bp->b_iodone == NULL);
5496 	bp->b_iodone = (int (*)())ufs_iodone;
5497 	mutex_enter(&ip->i_tlock);
5498 	ip->i_writes += bp->b_bcount;
5499 	mutex_exit(&ip->i_tlock);
5500 
5501 	if (bp->b_flags & B_ASYNC) {
5502 		if (ufsvfsp->vfs_log) {
5503 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5504 		} else if (ufsvfsp->vfs_snapshot) {
5505 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5506 		} else {
5507 			ufsvfsp->vfs_iotstamp = lbolt;
5508 			ub.ub_putasyncs.value.ul++;
5509 			(void) bdev_strategy(bp);
5510 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5511 		}
5512 	} else {
5513 		if (ufsvfsp->vfs_log) {
5514 			lufs_write_strategy(ufsvfsp->vfs_log, bp);
5515 		} else if (ufsvfsp->vfs_snapshot) {
5516 			fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5517 		} else {
5518 			ufsvfsp->vfs_iotstamp = lbolt;
5519 			ub.ub_putsyncs.value.ul++;
5520 			(void) bdev_strategy(bp);
5521 			lwp_stat_update(LWP_STAT_OUBLK, 1);
5522 		}
5523 		err = biowait(bp);
5524 		pageio_done(bp);
5525 		pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
5526 	}
5527 
5528 	pp = NULL;
5529 
5530 out:
5531 	if (err != 0 && pp != NULL)
5532 		pvn_write_done(pp, B_ERROR | B_WRITE | flags);
5533 
5534 	if (offp)
5535 		*offp = io_off;
5536 	if (lenp)
5537 		*lenp = io_len;
5538 out_trace:
5539 	TRACE_2(TR_FAC_UFS, TR_UFS_PUTAPAGE_END,
5540 		"ufs_putapage_end:vp %p error %d", vp, err);
5541 	return (err);
5542 }
5543 
5544 /* ARGSUSED */
5545 static int
5546 ufs_map(struct vnode *vp,
5547 	offset_t off,
5548 	struct as *as,
5549 	caddr_t *addrp,
5550 	size_t len,
5551 	uchar_t prot,
5552 	uchar_t maxprot,
5553 	uint_t flags,
5554 	struct cred *cr)
5555 {
5556 	struct segvn_crargs vn_a;
5557 	struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5558 	struct ulockfs *ulp;
5559 	int error;
5560 
5561 	TRACE_1(TR_FAC_UFS, TR_UFS_MAP_START,
5562 		"ufs_map_start:vp %p", vp);
5563 
5564 retry_map:
5565 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK);
5566 	if (error)
5567 		goto out;
5568 
5569 	if (vp->v_flag & VNOMAP) {
5570 		error = ENOSYS;
5571 		goto unlock;
5572 	}
5573 
5574 	if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) {
5575 		error = ENXIO;
5576 		goto unlock;
5577 	}
5578 
5579 	if (vp->v_type != VREG) {
5580 		error = ENODEV;
5581 		goto unlock;
5582 	}
5583 
5584 	/*
5585 	 * If file is being locked, disallow mapping.
5586 	 */
5587 	if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) {
5588 		error = EAGAIN;
5589 		goto unlock;
5590 	}
5591 
5592 	as_rangelock(as);
5593 	if ((flags & MAP_FIXED) == 0) {
5594 		map_addr(addrp, len, off, 1, flags);
5595 		if (*addrp == NULL) {
5596 			as_rangeunlock(as);
5597 			error = ENOMEM;
5598 			goto unlock;
5599 		}
5600 	} else {
5601 		/*
5602 		 * User specified address - blow away any previous mappings
5603 		 */
5604 		(void) as_unmap(as, *addrp, len);
5605 	}
5606 
5607 	vn_a.vp = vp;
5608 	vn_a.offset = (u_offset_t)off;
5609 	vn_a.type = flags & MAP_TYPE;
5610 	vn_a.prot = prot;
5611 	vn_a.maxprot = maxprot;
5612 	vn_a.cred = cr;
5613 	vn_a.amp = NULL;
5614 	vn_a.flags = flags & ~MAP_TYPE;
5615 	vn_a.szc = 0;
5616 	vn_a.lgrp_mem_policy_flags = 0;
5617 
5618 retry_lock:
5619 	if (!AS_LOCK_TRYENTER(ias, &as->a_lock, RW_WRITER)) {
5620 		/*
5621 		 * We didn't get the lock. Check if the SLOCK is set in the
5622 		 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
5623 		 * and wait for SLOCK to be cleared.
5624 		 */
5625 
5626 		if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
5627 			as_rangeunlock(as);
5628 			ufs_lockfs_end(ulp);
5629 			goto retry_map;
5630 		} else {
5631 			/*
5632 			 * SLOCK isn't set so this is a genuine synchronization
5633 			 * case. Let's try again after giving them a breather.
5634 			 */
5635 			delay(RETRY_LOCK_DELAY);
5636 			goto  retry_lock;
5637 		}
5638 	}
5639 	error = as_map_locked(as, *addrp, len, segvn_create, &vn_a);
5640 	as_rangeunlock(as);
5641 
5642 unlock:
5643 	if (ulp) {
5644 		ufs_lockfs_end(ulp);
5645 	}
5646 out:
5647 	TRACE_2(TR_FAC_UFS, TR_UFS_MAP_END,
5648 		"ufs_map_end:vp %p error %d", vp, error);
5649 	return (error);
5650 }
5651 
5652 /* ARGSUSED */
5653 static int
5654 ufs_addmap(struct vnode *vp,
5655 	offset_t off,
5656 	struct as *as,
5657 	caddr_t addr,
5658 	size_t	len,
5659 	uchar_t  prot,
5660 	uchar_t  maxprot,
5661 	uint_t    flags,
5662 	struct cred *cr)
5663 {
5664 	struct inode *ip = VTOI(vp);
5665 
5666 	if (vp->v_flag & VNOMAP) {
5667 		return (ENOSYS);
5668 	}
5669 
5670 	mutex_enter(&ip->i_tlock);
5671 	ip->i_mapcnt += btopr(len);
5672 	mutex_exit(&ip->i_tlock);
5673 	return (0);
5674 }
5675 
5676 /*ARGSUSED*/
5677 static int
5678 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5679 	size_t len, uint_t prot,  uint_t maxprot,  uint_t flags,
5680 	struct cred *cr)
5681 {
5682 	struct inode *ip = VTOI(vp);
5683 
5684 	if (vp->v_flag & VNOMAP) {
5685 		return (ENOSYS);
5686 	}
5687 
5688 	mutex_enter(&ip->i_tlock);
5689 	ip->i_mapcnt -= btopr(len); 	/* Count released mappings */
5690 	ASSERT(ip->i_mapcnt >= 0);
5691 	mutex_exit(&ip->i_tlock);
5692 	return (0);
5693 }
5694 /*
5695  * Return the answer requested to poll() for non-device files
5696  */
5697 struct pollhead ufs_pollhd;
5698 
5699 /* ARGSUSED */
5700 int
5701 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp)
5702 {
5703 	struct ufsvfs	*ufsvfsp;
5704 
5705 	*revp = 0;
5706 	ufsvfsp = VTOI(vp)->i_ufsvfs;
5707 
5708 	if (!ufsvfsp) {
5709 		*revp = POLLHUP;
5710 		goto out;
5711 	}
5712 
5713 	if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) ||
5714 	    ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) {
5715 		*revp |= POLLERR;
5716 
5717 	} else {
5718 		if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly &&
5719 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5720 			*revp |= POLLOUT;
5721 
5722 		if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly &&
5723 		    !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5724 			*revp |= POLLWRBAND;
5725 
5726 		if (ev & POLLIN)
5727 			*revp |= POLLIN;
5728 
5729 		if (ev & POLLRDNORM)
5730 			*revp |= POLLRDNORM;
5731 
5732 		if (ev & POLLRDBAND)
5733 			*revp |= POLLRDBAND;
5734 	}
5735 
5736 	if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP)))
5737 		*revp |= POLLPRI;
5738 out:
5739 	*phpp = !any && !*revp ? &ufs_pollhd : (struct pollhead *)NULL;
5740 
5741 	return (0);
5742 }
5743 
5744 /* ARGSUSED */
5745 static int
5746 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr)
5747 {
5748 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
5749 	struct ulockfs	*ulp = NULL;
5750 	struct inode 	*sip = NULL;
5751 	int		error;
5752 	struct inode 	*ip = VTOI(vp);
5753 	int		issync;
5754 
5755 	error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK);
5756 	if (error)
5757 		return (error);
5758 
5759 	switch (cmd) {
5760 		/*
5761 		 * Have to handle _PC_NAME_MAX here, because the normal way
5762 		 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5763 		 * results in a lock ordering reversal between
5764 		 * ufs_lockfs_{begin,end}() and
5765 		 * ufs_thread_{suspend,continue}().
5766 		 *
5767 		 * Keep in sync with ufs_statvfs().
5768 		 */
5769 	case _PC_NAME_MAX:
5770 		*valp = MAXNAMLEN;
5771 		break;
5772 
5773 	case _PC_FILESIZEBITS:
5774 		if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
5775 			*valp = UFS_FILESIZE_BITS;
5776 		else
5777 			*valp = 32;
5778 		break;
5779 
5780 	case _PC_XATTR_EXISTS:
5781 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
5782 
5783 			error = ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR,
5784 							cr);
5785 			if (error ==  0 && sip != NULL) {
5786 				/* Start transaction */
5787 				if (ulp) {
5788 					TRANS_BEGIN_CSYNC(ufsvfsp, issync,
5789 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5790 				}
5791 				/*
5792 				 * Is directory empty
5793 				 */
5794 				rw_enter(&sip->i_rwlock, RW_WRITER);
5795 				rw_enter(&sip->i_contents, RW_WRITER);
5796 				if (ufs_xattrdirempty(sip,
5797 						sip->i_number, CRED())) {
5798 					rw_enter(&ip->i_contents, RW_WRITER);
5799 					ufs_unhook_shadow(ip, sip);
5800 					rw_exit(&ip->i_contents);
5801 
5802 					*valp = 0;
5803 
5804 				} else
5805 					*valp = 1;
5806 				rw_exit(&sip->i_contents);
5807 				rw_exit(&sip->i_rwlock);
5808 				if (ulp) {
5809 					TRANS_END_CSYNC(ufsvfsp, error, issync,
5810 					    TOP_RMDIR, TOP_RMDIR_SIZE);
5811 				}
5812 				VN_RELE(ITOV(sip));
5813 			} else if (error == ENOENT) {
5814 				*valp = 0;
5815 				error = 0;
5816 			}
5817 		} else {
5818 			error = fs_pathconf(vp, cmd, valp, cr);
5819 		}
5820 		break;
5821 
5822 	case _PC_ACL_ENABLED:
5823 		*valp = _ACL_ACLENT_ENABLED;
5824 		break;
5825 
5826 	case _PC_MIN_HOLE_SIZE:
5827 		*valp = (ulong_t)ip->i_fs->fs_bsize;
5828 		break;
5829 
5830 	default:
5831 		error = fs_pathconf(vp, cmd, valp, cr);
5832 	}
5833 
5834 	if (ulp != NULL) {
5835 		ufs_lockfs_end(ulp);
5836 	}
5837 	return (error);
5838 }
5839 
5840 int ufs_pageio_writes, ufs_pageio_reads;
5841 
5842 /*ARGSUSED*/
5843 static int
5844 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len,
5845 	int flags, struct cred *cr)
5846 {
5847 	struct inode *ip = VTOI(vp);
5848 	struct ufsvfs *ufsvfsp;
5849 	page_t *npp = NULL, *opp = NULL, *cpp = pp;
5850 	struct buf *bp;
5851 	daddr_t bn;
5852 	size_t done_len = 0, cur_len = 0;
5853 	int err = 0;
5854 	int contig = 0;
5855 	int dolock;
5856 	int vmpss = 0;
5857 	struct ulockfs *ulp;
5858 
5859 	if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp &&
5860 	    vp->v_mpssdata != NULL) {
5861 		vmpss = 1;
5862 	}
5863 
5864 	dolock = (rw_owner(&ip->i_contents) != curthread);
5865 	/*
5866 	 * We need a better check.  Ideally, we would use another
5867 	 * vnodeops so that hlocked and forcibly unmounted file
5868 	 * systems would return EIO where appropriate and w/o the
5869 	 * need for these checks.
5870 	 */
5871 	if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5872 		return (EIO);
5873 
5874 	/*
5875 	 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5876 	 * ul_lock must be taken before locking pages so we can't use it here
5877 	 * if pp is non NULL because segvn already locked pages
5878 	 * SE_EXCL. Instead we rely on the fact that a forced umount or
5879 	 * applying a filesystem lock via ufs_fiolfs() will block in the
5880 	 * implicit call to ufs_flush() until we unlock the pages after the
5881 	 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5882 	 * above 0 until they are done. We have to be careful not to increment
5883 	 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5884 	 *
5885 	 * If pp is NULL use ul_lock to make sure we don't increment
5886 	 * ul_vnops_cnt after forceful unmount hlocks the file system.
5887 	 */
5888 	if (vmpss || pp == NULL) {
5889 		ulp = &ufsvfsp->vfs_ulockfs;
5890 		if (pp == NULL)
5891 			mutex_enter(&ulp->ul_lock);
5892 		if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) {
5893 			if (pp == NULL) {
5894 				mutex_exit(&ulp->ul_lock);
5895 			}
5896 			return (vmpss ? EIO : EINVAL);
5897 		}
5898 		atomic_add_long(&ulp->ul_vnops_cnt, 1);
5899 		if (pp == NULL)
5900 			mutex_exit(&ulp->ul_lock);
5901 		if (ufs_quiesce_pend) {
5902 			if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5903 				cv_broadcast(&ulp->ul_cv);
5904 			return (vmpss ? EIO : EINVAL);
5905 		}
5906 	}
5907 
5908 	if (dolock) {
5909 		/*
5910 		 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
5911 		 * handle a fault against a segment that maps vnode pages with
5912 		 * large mappings.  Segvn creates pages and holds them locked
5913 		 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
5914 		 * use rw_tryenter() to avoid a potential deadlock since in
5915 		 * lock order i_contents needs to be taken first.
5916 		 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
5917 		 */
5918 		if (!vmpss) {
5919 			rw_enter(&ip->i_contents, RW_READER);
5920 		} else if (!rw_tryenter(&ip->i_contents, RW_READER)) {
5921 			if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5922 				cv_broadcast(&ulp->ul_cv);
5923 			return (EDEADLK);
5924 		}
5925 	}
5926 
5927 	/*
5928 	 * Return an error to segvn because the pagefault request is beyond
5929 	 * PAGESIZE rounded EOF.
5930 	 */
5931 	if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) {
5932 		if (dolock)
5933 			rw_exit(&ip->i_contents);
5934 		if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5935 			cv_broadcast(&ulp->ul_cv);
5936 		return (EFAULT);
5937 	}
5938 
5939 	if (pp == NULL) {
5940 		if (bmap_has_holes(ip)) {
5941 			err = ENOSYS;
5942 		} else {
5943 			err = EINVAL;
5944 		}
5945 		if (dolock)
5946 			rw_exit(&ip->i_contents);
5947 		if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
5948 			cv_broadcast(&ulp->ul_cv);
5949 		return (err);
5950 	}
5951 
5952 	/*
5953 	 * Break the io request into chunks, one for each contiguous
5954 	 * stretch of disk blocks in the target file.
5955 	 */
5956 	while (done_len < io_len) {
5957 		ASSERT(cpp);
5958 		contig = 0;
5959 		if (err = bmap_read(ip, (u_offset_t)(io_off + done_len),
5960 				    &bn, &contig))
5961 			break;
5962 
5963 		if (bn == UFS_HOLE) {	/* No holey swapfiles */
5964 			if (vmpss) {
5965 				err = EFAULT;
5966 				break;
5967 			}
5968 			err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE");
5969 			break;
5970 		}
5971 
5972 		cur_len = MIN(io_len - done_len, contig);
5973 		/*
5974 		 * Zero out a page beyond EOF, when the last block of
5975 		 * a file is a UFS fragment so that ufs_pageio() can be used
5976 		 * instead of ufs_getpage() to handle faults against
5977 		 * segvn segments that use large pages.
5978 		 */
5979 		page_list_break(&cpp, &npp, btopr(cur_len));
5980 		if ((flags & B_READ) && (cur_len & PAGEOFFSET)) {
5981 			size_t xlen = cur_len & PAGEOFFSET;
5982 			pagezero(cpp->p_prev, xlen, PAGESIZE - xlen);
5983 		}
5984 
5985 		bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
5986 		ASSERT(bp != NULL);
5987 
5988 		bp->b_edev = ip->i_dev;
5989 		bp->b_dev = cmpdev(ip->i_dev);
5990 		bp->b_blkno = bn;
5991 		bp->b_un.b_addr = (caddr_t)0;
5992 		bp->b_file = ip->i_vnode;
5993 
5994 		ufsvfsp->vfs_iotstamp = lbolt;
5995 		ub.ub_pageios.value.ul++;
5996 		if (ufsvfsp->vfs_snapshot)
5997 			fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp);
5998 		else
5999 			(void) bdev_strategy(bp);
6000 
6001 		if (flags & B_READ)
6002 			ufs_pageio_reads++;
6003 		else
6004 			ufs_pageio_writes++;
6005 		if (flags & B_READ)
6006 			lwp_stat_update(LWP_STAT_INBLK, 1);
6007 		else
6008 			lwp_stat_update(LWP_STAT_OUBLK, 1);
6009 		/*
6010 		 * If the request is not B_ASYNC, wait for i/o to complete
6011 		 * and re-assemble the page list to return to the caller.
6012 		 * If it is B_ASYNC we leave the page list in pieces and
6013 		 * cleanup() will dispose of them.
6014 		 */
6015 		if ((flags & B_ASYNC) == 0) {
6016 			err = biowait(bp);
6017 			pageio_done(bp);
6018 			if (err)
6019 				break;
6020 			page_list_concat(&opp, &cpp);
6021 		}
6022 		cpp = npp;
6023 		npp = NULL;
6024 		if (flags & B_READ)
6025 			cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t);
6026 		done_len += cur_len;
6027 	}
6028 	ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len));
6029 	if (err) {
6030 		if (flags & B_ASYNC) {
6031 			/* Cleanup unprocessed parts of list */
6032 			page_list_concat(&cpp, &npp);
6033 			if (flags & B_READ)
6034 				pvn_read_done(cpp, B_ERROR);
6035 			else
6036 				pvn_write_done(cpp, B_ERROR);
6037 		} else {
6038 			/* Re-assemble list and let caller clean up */
6039 			page_list_concat(&opp, &cpp);
6040 			page_list_concat(&opp, &npp);
6041 		}
6042 	}
6043 
6044 	if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) &&
6045 	    ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) {
6046 		mutex_enter(&ip->i_tlock);
6047 		ip->i_flag |= IACC;
6048 		ITIMES_NOLOCK(ip);
6049 		mutex_exit(&ip->i_tlock);
6050 	}
6051 
6052 	if (dolock)
6053 		rw_exit(&ip->i_contents);
6054 	if (vmpss && !atomic_add_long_nv(&ulp->ul_vnops_cnt, -1))
6055 		cv_broadcast(&ulp->ul_cv);
6056 	return (err);
6057 }
6058 
6059 /*
6060  * Called when the kernel is in a frozen state to dump data
6061  * directly to the device. It uses a private dump data structure,
6062  * set up by dump_ctl, to locate the correct disk block to which to dump.
6063  */
6064 static int
6065 ufs_dump(vnode_t *vp, caddr_t addr, int ldbn, int dblks)
6066 {
6067 	u_offset_t	file_size;
6068 	struct inode    *ip = VTOI(vp);
6069 	struct fs	*fs = ip->i_fs;
6070 	daddr_t		dbn, lfsbn;
6071 	int		disk_blks = fs->fs_bsize >> DEV_BSHIFT;
6072 	int		error = 0;
6073 	int		ndbs, nfsbs;
6074 
6075 	/*
6076 	 * forced unmount case
6077 	 */
6078 	if (ip->i_ufsvfs == NULL)
6079 		return (EIO);
6080 	/*
6081 	 * Validate the inode that it has not been modified since
6082 	 * the dump structure is allocated.
6083 	 */
6084 	mutex_enter(&ip->i_tlock);
6085 	if ((dump_info == NULL) ||
6086 	    (dump_info->ip != ip) ||
6087 	    (dump_info->time.tv_sec != ip->i_mtime.tv_sec) ||
6088 	    (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) {
6089 		mutex_exit(&ip->i_tlock);
6090 		return (-1);
6091 	}
6092 	mutex_exit(&ip->i_tlock);
6093 
6094 	/*
6095 	 * See that the file has room for this write
6096 	 */
6097 	UFS_GET_ISIZE(&file_size, ip);
6098 
6099 	if (ldbtob((offset_t)(ldbn + dblks)) > file_size)
6100 		return (ENOSPC);
6101 
6102 	/*
6103 	 * Find the physical disk block numbers from the dump
6104 	 * private data structure directly and write out the data
6105 	 * in contiguous block lumps
6106 	 */
6107 	while (dblks > 0 && !error) {
6108 		lfsbn = (daddr_t)lblkno(fs, ldbtob((offset_t)ldbn));
6109 		dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks;
6110 		nfsbs = 1;
6111 		ndbs = disk_blks - ldbn % disk_blks;
6112 		while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn +
6113 		    nfsbs]) == dbn + ndbs) {
6114 			nfsbs++;
6115 			ndbs += disk_blks;
6116 		}
6117 		if (ndbs > dblks)
6118 			ndbs = dblks;
6119 		error = bdev_dump(ip->i_dev, addr, dbn, ndbs);
6120 		addr += ldbtob((offset_t)ndbs);
6121 		dblks -= ndbs;
6122 		ldbn += ndbs;
6123 	}
6124 	return (error);
6125 
6126 }
6127 
6128 /*
6129  * Prepare the file system before and after the dump operation.
6130  *
6131  * action = DUMP_ALLOC:
6132  * Preparation before dump, allocate dump private data structure
6133  * to hold all the direct and indirect block info for dump.
6134  *
6135  * action = DUMP_FREE:
6136  * Clean up after dump, deallocate the dump private data structure.
6137  *
6138  * action = DUMP_SCAN:
6139  * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6140  * if found, the starting file-relative DEV_BSIZE lbn is written
6141  * to *bklp; that lbn is intended for use with VOP_DUMP()
6142  */
6143 static int
6144 ufs_dumpctl(vnode_t *vp, int action, int *blkp)
6145 {
6146 	struct inode	*ip = VTOI(vp);
6147 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
6148 	struct fs	*fs;
6149 	daddr32_t	*dblk, *storeblk;
6150 	daddr32_t	*nextblk, *endblk;
6151 	struct buf	*bp;
6152 	int		i, entry, entries;
6153 	int		n, ncontig;
6154 
6155 	/*
6156 	 * check for forced unmount
6157 	 */
6158 	if (ufsvfsp == NULL)
6159 		return (EIO);
6160 
6161 	if (action == DUMP_ALLOC) {
6162 		/*
6163 		 * alloc and record dump_info
6164 		 */
6165 		if (dump_info != NULL)
6166 			return (EINVAL);
6167 
6168 		ASSERT(vp->v_type == VREG);
6169 		fs = ufsvfsp->vfs_fs;
6170 
6171 		rw_enter(&ip->i_contents, RW_READER);
6172 
6173 		if (bmap_has_holes(ip)) {
6174 			rw_exit(&ip->i_contents);
6175 			return (EFAULT);
6176 		}
6177 
6178 		/*
6179 		 * calculate and allocate space needed according to i_size
6180 		 */
6181 		entries = (int)lblkno(fs, blkroundup(fs, ip->i_size));
6182 		if ((dump_info = (struct dump *)
6183 		    kmem_alloc(sizeof (struct dump) +
6184 		    (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP)) == NULL) {
6185 			    rw_exit(&ip->i_contents);
6186 			    return (ENOMEM);
6187 		}
6188 
6189 		/* Start saving the info */
6190 		dump_info->fsbs = entries;
6191 		dump_info->ip = ip;
6192 		storeblk = &dump_info->dblk[0];
6193 
6194 		/* Direct Blocks */
6195 		for (entry = 0; entry < NDADDR && entry < entries; entry++)
6196 			*storeblk++ = ip->i_db[entry];
6197 
6198 		/* Indirect Blocks */
6199 		for (i = 0; i < NIADDR; i++) {
6200 			int error = 0;
6201 
6202 			bp = UFS_BREAD(ufsvfsp,
6203 				ip->i_dev, fsbtodb(fs, ip->i_ib[i]),
6204 				fs->fs_bsize);
6205 			if (bp->b_flags & B_ERROR)
6206 				error = EIO;
6207 			else {
6208 				dblk = bp->b_un.b_daddr;
6209 				if ((storeblk = save_dblks(ip, ufsvfsp,
6210 				    storeblk, dblk, i, entries)) == NULL)
6211 					error = EIO;
6212 			}
6213 
6214 			brelse(bp);
6215 
6216 			if (error != 0) {
6217 				kmem_free(dump_info, sizeof (struct dump) +
6218 				    (entries - 1) * sizeof (daddr32_t));
6219 				rw_exit(&ip->i_contents);
6220 				dump_info = NULL;
6221 				return (error);
6222 			}
6223 		}
6224 		/* and time stamp the information */
6225 		mutex_enter(&ip->i_tlock);
6226 		dump_info->time = ip->i_mtime;
6227 		mutex_exit(&ip->i_tlock);
6228 
6229 		rw_exit(&ip->i_contents);
6230 	} else if (action == DUMP_FREE) {
6231 		/*
6232 		 * free dump_info
6233 		 */
6234 		if (dump_info == NULL)
6235 			return (EINVAL);
6236 		entries = dump_info->fsbs - 1;
6237 		kmem_free(dump_info, sizeof (struct dump) +
6238 		    entries * sizeof (daddr32_t));
6239 		dump_info = NULL;
6240 	} else if (action == DUMP_SCAN) {
6241 		/*
6242 		 * scan dump_info
6243 		 */
6244 		if (dump_info == NULL)
6245 			return (EINVAL);
6246 
6247 		dblk = dump_info->dblk;
6248 		nextblk = dblk + 1;
6249 		endblk = dblk + dump_info->fsbs - 1;
6250 		fs = ufsvfsp->vfs_fs;
6251 		ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT);
6252 
6253 		/*
6254 		 * scan dblk[] entries; contig fs space is found when:
6255 		 * ((current blkno + frags per block) == next blkno)
6256 		 */
6257 		n = 0;
6258 		while (n < ncontig && dblk < endblk) {
6259 			if ((*dblk + fs->fs_frag) == *nextblk)
6260 				n++;
6261 			else
6262 				n = 0;
6263 			dblk++;
6264 			nextblk++;
6265 		}
6266 
6267 		/*
6268 		 * index is where size bytes of contig space begins;
6269 		 * conversion from index to the file's DEV_BSIZE lbn
6270 		 * is equivalent to:  (index * fs_bsize) / DEV_BSIZE
6271 		 */
6272 		if (n == ncontig) {
6273 			i = (dblk - dump_info->dblk) - ncontig;
6274 			*blkp = i << (fs->fs_bshift - DEV_BSHIFT);
6275 		} else
6276 			return (EFAULT);
6277 	}
6278 	return (0);
6279 }
6280 
6281 /*
6282  * Recursive helper function for ufs_dumpctl().  It follows the indirect file
6283  * system  blocks until it reaches the the disk block addresses, which are
6284  * then stored into the given buffer, storeblk.
6285  */
6286 static daddr32_t *
6287 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp,  daddr32_t *storeblk,
6288     daddr32_t *dblk, int level, int entries)
6289 {
6290 	struct fs	*fs = ufsvfsp->vfs_fs;
6291 	struct buf	*bp;
6292 	int		i;
6293 
6294 	if (level == 0) {
6295 		for (i = 0; i < NINDIR(fs); i++) {
6296 			if (storeblk - dump_info->dblk >= entries)
6297 				break;
6298 			*storeblk++ = dblk[i];
6299 		}
6300 		return (storeblk);
6301 	}
6302 	for (i = 0; i < NINDIR(fs); i++) {
6303 		if (storeblk - dump_info->dblk >= entries)
6304 			break;
6305 		bp = UFS_BREAD(ufsvfsp,
6306 				ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize);
6307 		if (bp->b_flags & B_ERROR) {
6308 			brelse(bp);
6309 			return (NULL);
6310 		}
6311 		storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr,
6312 		    level - 1, entries);
6313 		brelse(bp);
6314 
6315 		if (storeblk == NULL)
6316 			return (NULL);
6317 	}
6318 	return (storeblk);
6319 }
6320 
6321 /* ARGSUSED */
6322 static int
6323 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag,
6324 	struct cred *cr)
6325 {
6326 	struct inode	*ip = VTOI(vp);
6327 	struct ulockfs	*ulp;
6328 	struct ufsvfs	*ufsvfsp = ip->i_ufsvfs;
6329 	ulong_t		vsa_mask = vsap->vsa_mask;
6330 	int		err = EINVAL;
6331 
6332 	TRACE_3(TR_FAC_UFS, TR_UFS_GETSECATTR_START,
6333 	    "ufs_getsecattr_start:vp %p, vsap %p, flags %x", vp, vsap, flag);
6334 
6335 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6336 
6337 	/*
6338 	 * Only grab locks if needed - they're not needed to check vsa_mask
6339 	 * or if the mask contains no acl flags.
6340 	 */
6341 	if (vsa_mask != 0) {
6342 		if (err = ufs_lockfs_begin(ufsvfsp, &ulp,
6343 		    ULOCKFS_GETATTR_MASK))
6344 			return (err);
6345 
6346 		rw_enter(&ip->i_contents, RW_READER);
6347 		err = ufs_acl_get(ip, vsap, flag, cr);
6348 		rw_exit(&ip->i_contents);
6349 
6350 		if (ulp)
6351 			ufs_lockfs_end(ulp);
6352 	}
6353 	TRACE_1(TR_FAC_UFS, TR_UFS_GETSECATTR_END,
6354 	    "ufs_getsecattr_end:vp %p", vp);
6355 	return (err);
6356 }
6357 
6358 /* ARGSUSED */
6359 static int
6360 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr)
6361 {
6362 	struct inode	*ip = VTOI(vp);
6363 	struct ulockfs	*ulp = NULL;
6364 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
6365 	ulong_t		vsa_mask = vsap->vsa_mask;
6366 	int		err;
6367 	int		haverwlock = 1;
6368 	int		trans_size;
6369 	int		donetrans = 0;
6370 	int		retry = 1;
6371 
6372 
6373 	TRACE_3(TR_FAC_UFS, TR_UFS_SETSECATTR_START,
6374 	    "ufs_setsecattr_start:vp %p, vsap %p, flags %x", vp, vsap, flag);
6375 
6376 	ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
6377 
6378 	/* Abort now if the request is either empty or invalid. */
6379 	vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6380 	if ((vsa_mask == 0) ||
6381 	    ((vsap->vsa_aclentp == NULL) &&
6382 	    (vsap->vsa_dfaclentp == NULL))) {
6383 		err = EINVAL;
6384 		goto out;
6385 	}
6386 
6387 	/*
6388 	 * Following convention, if this is a directory then we acquire the
6389 	 * inode's i_rwlock after starting a UFS logging transaction;
6390 	 * otherwise, we acquire it beforehand. Since we were called (and
6391 	 * must therefore return) with the lock held, we will have to drop it,
6392 	 * and later reacquire it, if operating on a directory.
6393 	 */
6394 	if (vp->v_type == VDIR) {
6395 		rw_exit(&ip->i_rwlock);
6396 		haverwlock = 0;
6397 	} else {
6398 		/* Upgrade the lock if required. */
6399 		if (!rw_write_held(&ip->i_rwlock)) {
6400 			rw_exit(&ip->i_rwlock);
6401 			rw_enter(&ip->i_rwlock, RW_WRITER);
6402 		}
6403 	}
6404 
6405 again:
6406 	ASSERT(!(vp->v_type == VDIR && haverwlock));
6407 	if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) {
6408 		ulp = NULL;
6409 		retry = 0;
6410 		goto out;
6411 	}
6412 
6413 	/*
6414 	 * Check that the file system supports this operation. Note that
6415 	 * ufs_lockfs_begin() will have checked that the file system had
6416 	 * not been forcibly unmounted.
6417 	 */
6418 	if (ufsvfsp->vfs_fs->fs_ronly) {
6419 		err = EROFS;
6420 		goto out;
6421 	}
6422 	if (ufsvfsp->vfs_nosetsec) {
6423 		err = ENOSYS;
6424 		goto out;
6425 	}
6426 
6427 	if (ulp) {
6428 		TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR,
6429 			trans_size = TOP_SETSECATTR_SIZE(VTOI(vp)));
6430 		donetrans = 1;
6431 	}
6432 
6433 	if (vp->v_type == VDIR) {
6434 		rw_enter(&ip->i_rwlock, RW_WRITER);
6435 		haverwlock = 1;
6436 	}
6437 
6438 	ASSERT(haverwlock);
6439 
6440 	/* Do the actual work. */
6441 	rw_enter(&ip->i_contents, RW_WRITER);
6442 	/*
6443 	 * Suppress out of inodes messages if we will retry.
6444 	 */
6445 	if (retry)
6446 		ip->i_flag |= IQUIET;
6447 	err = ufs_acl_set(ip, vsap, flag, cr);
6448 	ip->i_flag &= ~IQUIET;
6449 	rw_exit(&ip->i_contents);
6450 
6451 out:
6452 	if (ulp) {
6453 		if (donetrans) {
6454 			/*
6455 			 * top_end_async() can eventually call
6456 			 * top_end_sync(), which can block. We must
6457 			 * therefore observe the lock-ordering protocol
6458 			 * here as well.
6459 			 */
6460 			if (vp->v_type == VDIR) {
6461 				rw_exit(&ip->i_rwlock);
6462 				haverwlock = 0;
6463 			}
6464 			TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size);
6465 		}
6466 		ufs_lockfs_end(ulp);
6467 	}
6468 	/*
6469 	 * If no inodes available, try scaring a logically-
6470 	 * free one out of the delete queue to someplace
6471 	 * that we can find it.
6472 	 */
6473 	if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
6474 		ufs_delete_drain_wait(ufsvfsp, 1);
6475 		retry = 0;
6476 		if (vp->v_type == VDIR && haverwlock) {
6477 			rw_exit(&ip->i_rwlock);
6478 			haverwlock = 0;
6479 		}
6480 		goto again;
6481 	}
6482 	/*
6483 	 * If we need to reacquire the lock then it is safe to do so
6484 	 * as a reader. This is because ufs_rwunlock(), which will be
6485 	 * called by our caller after we return, does not differentiate
6486 	 * between shared and exclusive locks.
6487 	 */
6488 	if (!haverwlock) {
6489 		ASSERT(vp->v_type == VDIR);
6490 		rw_enter(&ip->i_rwlock, RW_READER);
6491 	}
6492 
6493 	TRACE_1(TR_FAC_UFS, TR_UFS_SETSECATTR_END,
6494 	    "ufs_setsecattr_end:vp %p", vp);
6495 	return (err);
6496 }
6497