xref: /illumos-gate/usr/src/uts/common/fs/udfs/udf_vnops.c (revision 2bbdd445a21f9d61f4a0ca0faf05d5ceb2bd91f3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 /*
27  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
28  */
29 
30 #include <sys/types.h>
31 #include <sys/t_lock.h>
32 #include <sys/param.h>
33 #include <sys/time.h>
34 #include <sys/systm.h>
35 #include <sys/sysmacros.h>
36 #include <sys/resource.h>
37 #include <sys/signal.h>
38 #include <sys/cred.h>
39 #include <sys/user.h>
40 #include <sys/buf.h>
41 #include <sys/vfs.h>
42 #include <sys/vfs_opreg.h>
43 #include <sys/stat.h>
44 #include <sys/vnode.h>
45 #include <sys/mode.h>
46 #include <sys/proc.h>
47 #include <sys/disp.h>
48 #include <sys/file.h>
49 #include <sys/fcntl.h>
50 #include <sys/flock.h>
51 #include <sys/kmem.h>
52 #include <sys/uio.h>
53 #include <sys/dnlc.h>
54 #include <sys/conf.h>
55 #include <sys/errno.h>
56 #include <sys/mman.h>
57 #include <sys/fbuf.h>
58 #include <sys/pathname.h>
59 #include <sys/debug.h>
60 #include <sys/vmsystm.h>
61 #include <sys/cmn_err.h>
62 #include <sys/dirent.h>
63 #include <sys/errno.h>
64 #include <sys/modctl.h>
65 #include <sys/statvfs.h>
66 #include <sys/mount.h>
67 #include <sys/sunddi.h>
68 #include <sys/bootconf.h>
69 #include <sys/policy.h>
70 
71 #include <vm/hat.h>
72 #include <vm/page.h>
73 #include <vm/pvn.h>
74 #include <vm/as.h>
75 #include <vm/seg.h>
76 #include <vm/seg_map.h>
77 #include <vm/seg_kmem.h>
78 #include <vm/seg_vn.h>
79 #include <vm/rm.h>
80 #include <vm/page.h>
81 #include <sys/swap.h>
82 
83 #include <fs/fs_subr.h>
84 
85 #include <sys/fs/udf_volume.h>
86 #include <sys/fs/udf_inode.h>
87 
88 static int32_t udf_open(struct vnode **,
89 	int32_t, struct cred *, caller_context_t *);
90 static int32_t udf_close(struct vnode *,
91 	int32_t, int32_t, offset_t, struct cred *, caller_context_t *);
92 static int32_t udf_read(struct vnode *,
93 	struct uio *, int32_t, struct cred *, caller_context_t *);
94 static int32_t udf_write(struct vnode *,
95 	struct uio *, int32_t, struct cred *, caller_context_t *);
96 static int32_t udf_ioctl(struct vnode *,
97 	int32_t, intptr_t, int32_t, struct cred *, int32_t *,
98 	caller_context_t *);
99 static int32_t udf_getattr(struct vnode *,
100 	struct vattr *, int32_t, struct cred *, caller_context_t *);
101 static int32_t udf_setattr(struct vnode *,
102 	struct vattr *, int32_t, struct cred *, caller_context_t *);
103 static int32_t udf_access(struct vnode *,
104 	int32_t, int32_t, struct cred *, caller_context_t *);
105 static int32_t udf_lookup(struct vnode *,
106 	char *, struct vnode **, struct pathname *,
107 	int32_t, struct vnode *, struct cred *,
108 	caller_context_t *, int *, pathname_t *);
109 static int32_t udf_create(struct vnode *,
110 	char *, struct vattr *, enum vcexcl,
111 	int32_t, struct vnode **, struct cred *, int32_t,
112 	caller_context_t *, vsecattr_t *);
113 static int32_t udf_remove(struct vnode *,
114 	char *, struct cred *, caller_context_t *, int);
115 static int32_t udf_link(struct vnode *,
116 	struct vnode *, char *, struct cred *, caller_context_t *, int);
117 static int32_t udf_rename(struct vnode *,
118 	char *, struct vnode *, char *, struct cred *, caller_context_t *, int);
119 static int32_t udf_mkdir(struct vnode *,
120 	char *, struct vattr *, struct vnode **, struct cred *,
121 	caller_context_t *, int, vsecattr_t *);
122 static int32_t udf_rmdir(struct vnode *,
123 	char *, struct vnode *, struct cred *, caller_context_t *, int);
124 static int32_t udf_readdir(struct vnode *,
125 	struct uio *, struct cred *, int32_t *, caller_context_t *, int);
126 static int32_t udf_symlink(struct vnode *,
127 	char *, struct vattr *, char *, struct cred *, caller_context_t *, int);
128 static int32_t udf_readlink(struct vnode *,
129 	struct uio *, struct cred *, caller_context_t *);
130 static int32_t udf_fsync(struct vnode *,
131 	int32_t, struct cred *, caller_context_t *);
132 static void udf_inactive(struct vnode *,
133 	struct cred *, caller_context_t *);
134 static int32_t udf_fid(struct vnode *, struct fid *, caller_context_t *);
135 static int udf_rwlock(struct vnode *, int32_t, caller_context_t *);
136 static void udf_rwunlock(struct vnode *, int32_t, caller_context_t *);
137 static int32_t udf_seek(struct vnode *, offset_t, offset_t *,
138 	caller_context_t *);
139 static int32_t udf_frlock(struct vnode *, int32_t,
140 	struct flock64 *, int32_t, offset_t, struct flk_callback *, cred_t *,
141 	caller_context_t *);
142 static int32_t udf_space(struct vnode *, int32_t,
143 	struct flock64 *, int32_t, offset_t, cred_t *, caller_context_t *);
144 static int32_t udf_getpage(struct vnode *, offset_t,
145 	size_t, uint32_t *, struct page **, size_t,
146 	struct seg *, caddr_t, enum seg_rw, struct cred *, caller_context_t *);
147 static int32_t udf_putpage(struct vnode *, offset_t,
148 	size_t, int32_t, struct cred *, caller_context_t *);
149 static int32_t udf_map(struct vnode *, offset_t, struct as *,
150 	caddr_t *, size_t, uint8_t, uint8_t, uint32_t, struct cred *,
151 	caller_context_t *);
152 static int32_t udf_addmap(struct vnode *, offset_t, struct as *,
153 	caddr_t, size_t, uint8_t, uint8_t, uint32_t, struct cred *,
154 	caller_context_t *);
155 static int32_t udf_delmap(struct vnode *, offset_t, struct as *,
156 	caddr_t, size_t, uint32_t, uint32_t, uint32_t, struct cred *,
157 	caller_context_t *);
158 static int32_t udf_l_pathconf(struct vnode *, int32_t,
159 	ulong_t *, struct cred *, caller_context_t *);
160 static int32_t udf_pageio(struct vnode *, struct page *,
161 	u_offset_t, size_t, int32_t, struct cred *, caller_context_t *);
162 
163 int32_t ud_getpage_miss(struct vnode *, u_offset_t,
164 	size_t, struct seg *, caddr_t, page_t *pl[],
165 	size_t, enum seg_rw, int32_t);
166 void ud_getpage_ra(struct vnode *, u_offset_t, struct seg *, caddr_t);
167 int32_t ud_putpages(struct vnode *, offset_t, size_t, int32_t, struct cred *);
168 int32_t ud_page_fill(struct ud_inode *, page_t *,
169 	u_offset_t, uint32_t, u_offset_t *);
170 int32_t ud_iodone(struct buf *);
171 int32_t ud_rdip(struct ud_inode *, struct uio *, int32_t, cred_t *);
172 int32_t ud_wrip(struct ud_inode *, struct uio *, int32_t, cred_t *);
173 int32_t ud_multi_strat(struct ud_inode *, page_t *, struct buf *, u_offset_t);
174 int32_t ud_slave_done(struct buf *);
175 
176 /*
177  * Structures to control multiple IO operations to get or put pages
178  * that are backed by discontiguous blocks. The master struct is
179  * a dummy that holds the original bp from pageio_setup. The
180  * slave struct holds the working bp's to do the actual IO. Once
181  * all the slave IOs complete. The master is processed as if a single
182  * IO op has completed.
183  */
184 uint32_t master_index = 0;
185 typedef struct mio_master {
186 	kmutex_t	mm_mutex;	/* protect the fields below */
187 	int32_t		mm_size;
188 	buf_t		*mm_bp;		/* original bp */
189 	int32_t		mm_resid;	/* bytes remaining to transfer */
190 	int32_t		mm_error;	/* accumulated error from slaves */
191 	int32_t		mm_index;	/* XXX debugging */
192 } mio_master_t;
193 
194 typedef struct mio_slave {
195 	buf_t		ms_buf;		/* working buffer for this IO chunk */
196 	mio_master_t	*ms_ptr;	/* pointer to master */
197 } mio_slave_t;
198 
199 struct vnodeops *udf_vnodeops;
200 
201 const fs_operation_def_t udf_vnodeops_template[] = {
202 	VOPNAME_OPEN,		{ .vop_open = udf_open },
203 	VOPNAME_CLOSE,		{ .vop_close = udf_close },
204 	VOPNAME_READ,		{ .vop_read = udf_read },
205 	VOPNAME_WRITE,		{ .vop_write = udf_write },
206 	VOPNAME_IOCTL,		{ .vop_ioctl = udf_ioctl },
207 	VOPNAME_GETATTR,	{ .vop_getattr = udf_getattr },
208 	VOPNAME_SETATTR,	{ .vop_setattr = udf_setattr },
209 	VOPNAME_ACCESS,		{ .vop_access = udf_access },
210 	VOPNAME_LOOKUP,		{ .vop_lookup = udf_lookup },
211 	VOPNAME_CREATE,		{ .vop_create = udf_create },
212 	VOPNAME_REMOVE,		{ .vop_remove = udf_remove },
213 	VOPNAME_LINK,		{ .vop_link = udf_link },
214 	VOPNAME_RENAME,		{ .vop_rename = udf_rename },
215 	VOPNAME_MKDIR,		{ .vop_mkdir = udf_mkdir },
216 	VOPNAME_RMDIR,		{ .vop_rmdir = udf_rmdir },
217 	VOPNAME_READDIR,	{ .vop_readdir = udf_readdir },
218 	VOPNAME_SYMLINK,	{ .vop_symlink = udf_symlink },
219 	VOPNAME_READLINK,	{ .vop_readlink = udf_readlink },
220 	VOPNAME_FSYNC,		{ .vop_fsync = udf_fsync },
221 	VOPNAME_INACTIVE,	{ .vop_inactive = udf_inactive },
222 	VOPNAME_FID,		{ .vop_fid = udf_fid },
223 	VOPNAME_RWLOCK,		{ .vop_rwlock = udf_rwlock },
224 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = udf_rwunlock },
225 	VOPNAME_SEEK,		{ .vop_seek = udf_seek },
226 	VOPNAME_FRLOCK,		{ .vop_frlock = udf_frlock },
227 	VOPNAME_SPACE,		{ .vop_space = udf_space },
228 	VOPNAME_GETPAGE,	{ .vop_getpage = udf_getpage },
229 	VOPNAME_PUTPAGE,	{ .vop_putpage = udf_putpage },
230 	VOPNAME_MAP,		{ .vop_map = udf_map },
231 	VOPNAME_ADDMAP,		{ .vop_addmap = udf_addmap },
232 	VOPNAME_DELMAP,		{ .vop_delmap = udf_delmap },
233 	VOPNAME_PATHCONF,	{ .vop_pathconf = udf_l_pathconf },
234 	VOPNAME_PAGEIO,		{ .vop_pageio = udf_pageio },
235 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
236 	NULL,			NULL
237 };
238 
239 /* ARGSUSED */
240 static int32_t
241 udf_open(
242 	struct vnode **vpp,
243 	int32_t flag,
244 	struct cred *cr,
245 	caller_context_t *ct)
246 {
247 	ud_printf("udf_open\n");
248 
249 	return (0);
250 }
251 
252 /* ARGSUSED */
253 static int32_t
254 udf_close(
255 	struct vnode *vp,
256 	int32_t flag,
257 	int32_t count,
258 	offset_t offset,
259 	struct cred *cr,
260 	caller_context_t *ct)
261 {
262 	struct ud_inode *ip = VTOI(vp);
263 
264 	ud_printf("udf_close\n");
265 
266 	ITIMES(ip);
267 
268 	cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
269 	cleanshares(vp, ttoproc(curthread)->p_pid);
270 
271 	/*
272 	 * Push partially filled cluster at last close.
273 	 * ``last close'' is approximated because the dnlc
274 	 * may have a hold on the vnode.
275 	 */
276 	if (vp->v_count <= 2 && vp->v_type != VBAD) {
277 		struct ud_inode *ip = VTOI(vp);
278 		if (ip->i_delaylen) {
279 			(void) ud_putpages(vp, ip->i_delayoff, ip->i_delaylen,
280 			    B_ASYNC | B_FREE, cr);
281 			ip->i_delaylen = 0;
282 		}
283 	}
284 
285 	return (0);
286 }
287 
288 /* ARGSUSED */
289 static int32_t
290 udf_read(
291 	struct vnode *vp,
292 	struct uio *uiop,
293 	int32_t ioflag,
294 	struct cred *cr,
295 	caller_context_t *ct)
296 {
297 	struct ud_inode *ip = VTOI(vp);
298 	int32_t error;
299 
300 	ud_printf("udf_read\n");
301 
302 #ifdef	__lock_lint
303 	rw_enter(&ip->i_rwlock, RW_READER);
304 #endif
305 
306 	ASSERT(RW_READ_HELD(&ip->i_rwlock));
307 
308 	if (MANDLOCK(vp, ip->i_char)) {
309 		/*
310 		 * udf_getattr ends up being called by chklock
311 		 */
312 		error = chklock(vp, FREAD, uiop->uio_loffset,
313 		    uiop->uio_resid, uiop->uio_fmode, ct);
314 		if (error) {
315 			goto end;
316 		}
317 	}
318 
319 	rw_enter(&ip->i_contents, RW_READER);
320 	error = ud_rdip(ip, uiop, ioflag, cr);
321 	rw_exit(&ip->i_contents);
322 
323 end:
324 #ifdef	__lock_lint
325 	rw_exit(&ip->i_rwlock);
326 #endif
327 
328 	return (error);
329 }
330 
331 
332 int32_t ud_WRITES = 1;
333 int32_t ud_HW = 96 * 1024;
334 int32_t ud_LW = 64 * 1024;
335 int32_t ud_throttles = 0;
336 
337 /* ARGSUSED */
338 static int32_t
339 udf_write(
340 	struct vnode *vp,
341 	struct uio *uiop,
342 	int32_t ioflag,
343 	struct cred *cr,
344 	caller_context_t *ct)
345 {
346 	struct ud_inode *ip = VTOI(vp);
347 	int32_t error = 0;
348 
349 	ud_printf("udf_write\n");
350 
351 #ifdef	__lock_lint
352 	rw_enter(&ip->i_rwlock, RW_WRITER);
353 #endif
354 
355 	ASSERT(RW_WRITE_HELD(&ip->i_rwlock));
356 
357 	if (MANDLOCK(vp, ip->i_char)) {
358 		/*
359 		 * ud_getattr ends up being called by chklock
360 		 */
361 		error = chklock(vp, FWRITE, uiop->uio_loffset,
362 		    uiop->uio_resid, uiop->uio_fmode, ct);
363 		if (error) {
364 			goto end;
365 		}
366 	}
367 	/*
368 	 * Throttle writes.
369 	 */
370 	mutex_enter(&ip->i_tlock);
371 	if (ud_WRITES && (ip->i_writes > ud_HW)) {
372 		while (ip->i_writes > ud_HW) {
373 			ud_throttles++;
374 			cv_wait(&ip->i_wrcv, &ip->i_tlock);
375 		}
376 	}
377 	mutex_exit(&ip->i_tlock);
378 
379 	/*
380 	 * Write to the file
381 	 */
382 	rw_enter(&ip->i_contents, RW_WRITER);
383 	if ((ioflag & FAPPEND) != 0 && (ip->i_type == VREG)) {
384 		/*
385 		 * In append mode start at end of file.
386 		 */
387 		uiop->uio_loffset = ip->i_size;
388 	}
389 	error = ud_wrip(ip, uiop, ioflag, cr);
390 	rw_exit(&ip->i_contents);
391 
392 end:
393 #ifdef	__lock_lint
394 	rw_exit(&ip->i_rwlock);
395 #endif
396 
397 	return (error);
398 }
399 
400 /* ARGSUSED */
401 static int32_t
402 udf_ioctl(
403 	struct vnode *vp,
404 	int32_t cmd,
405 	intptr_t arg,
406 	int32_t flag,
407 	struct cred *cr,
408 	int32_t *rvalp,
409 	caller_context_t *ct)
410 {
411 	return (ENOTTY);
412 }
413 
414 /* ARGSUSED */
415 static int32_t
416 udf_getattr(
417 	struct vnode *vp,
418 	struct vattr *vap,
419 	int32_t flags,
420 	struct cred *cr,
421 	caller_context_t *ct)
422 {
423 	struct ud_inode *ip = VTOI(vp);
424 
425 	ud_printf("udf_getattr\n");
426 
427 	if (vap->va_mask == AT_SIZE) {
428 		/*
429 		 * for performance, if only the size is requested don't bother
430 		 * with anything else.
431 		 */
432 		vap->va_size = ip->i_size;
433 		return (0);
434 	}
435 
436 	rw_enter(&ip->i_contents, RW_READER);
437 
438 	vap->va_type = vp->v_type;
439 	vap->va_mode = UD2VA_PERM(ip->i_perm) | ip->i_char;
440 
441 	vap->va_uid = ip->i_uid;
442 	vap->va_gid = ip->i_gid;
443 	vap->va_fsid = ip->i_dev;
444 	vap->va_nodeid = ip->i_icb_lbano;
445 	vap->va_nlink = ip->i_nlink;
446 	vap->va_size = ip->i_size;
447 	vap->va_seq = ip->i_seq;
448 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
449 		vap->va_rdev = ip->i_rdev;
450 	} else {
451 		vap->va_rdev = 0;
452 	}
453 
454 	mutex_enter(&ip->i_tlock);
455 	ITIMES_NOLOCK(ip);	/* mark correct time in inode */
456 	vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec;
457 	vap->va_atime.tv_nsec = ip->i_atime.tv_nsec;
458 	vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec;
459 	vap->va_mtime.tv_nsec = ip->i_mtime.tv_nsec;
460 	vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec;
461 	vap->va_ctime.tv_nsec = ip->i_ctime.tv_nsec;
462 	mutex_exit(&ip->i_tlock);
463 
464 	switch (ip->i_type) {
465 		case VBLK:
466 			vap->va_blksize = MAXBSIZE;
467 			break;
468 		case VCHR:
469 			vap->va_blksize = MAXBSIZE;
470 			break;
471 		default:
472 			vap->va_blksize = ip->i_udf->udf_lbsize;
473 			break;
474 	}
475 	vap->va_nblocks = ip->i_lbr << ip->i_udf->udf_l2d_shift;
476 
477 	rw_exit(&ip->i_contents);
478 
479 	return (0);
480 }
481 
482 static int
483 ud_iaccess_vmode(void *ip, int mode, struct cred *cr)
484 {
485 	return (ud_iaccess(ip, UD_UPERM2DPERM(mode), cr, 0));
486 }
487 
488 /*ARGSUSED4*/
489 static int32_t
490 udf_setattr(
491 	struct vnode *vp,
492 	struct vattr *vap,
493 	int32_t flags,
494 	struct cred *cr,
495 	caller_context_t *ct)
496 {
497 	int32_t error = 0;
498 	uint32_t mask = vap->va_mask;
499 	struct ud_inode *ip;
500 	timestruc_t now;
501 	struct vattr ovap;
502 
503 	ud_printf("udf_setattr\n");
504 
505 	ip = VTOI(vp);
506 
507 	/*
508 	 * not updates allowed to 4096 files
509 	 */
510 	if (ip->i_astrat == STRAT_TYPE4096) {
511 		return (EINVAL);
512 	}
513 
514 	/*
515 	 * Cannot set these attributes
516 	 */
517 	if (mask & AT_NOSET) {
518 		return (EINVAL);
519 	}
520 
521 	rw_enter(&ip->i_rwlock, RW_WRITER);
522 	rw_enter(&ip->i_contents, RW_WRITER);
523 
524 	ovap.va_uid = ip->i_uid;
525 	ovap.va_mode = UD2VA_PERM(ip->i_perm) | ip->i_char;
526 	error = secpolicy_vnode_setattr(cr, vp, vap, &ovap, flags,
527 	    ud_iaccess_vmode, ip);
528 	if (error)
529 		goto update_inode;
530 
531 	mask = vap->va_mask;
532 	/*
533 	 * Change file access modes.
534 	 */
535 	if (mask & AT_MODE) {
536 		ip->i_perm = VA2UD_PERM(vap->va_mode);
537 		ip->i_char = vap->va_mode & (VSUID | VSGID | VSVTX);
538 		mutex_enter(&ip->i_tlock);
539 		ip->i_flag |= ICHG;
540 		mutex_exit(&ip->i_tlock);
541 	}
542 	if (mask & (AT_UID|AT_GID)) {
543 		if (mask & AT_UID) {
544 			ip->i_uid = vap->va_uid;
545 		}
546 		if (mask & AT_GID) {
547 			ip->i_gid = vap->va_gid;
548 		}
549 		mutex_enter(&ip->i_tlock);
550 		ip->i_flag |= ICHG;
551 		mutex_exit(&ip->i_tlock);
552 	}
553 	/*
554 	 * Truncate file.  Must have write permission and not be a directory.
555 	 */
556 	if (mask & AT_SIZE) {
557 		if (vp->v_type == VDIR) {
558 			error = EISDIR;
559 			goto update_inode;
560 		}
561 		if (error = ud_iaccess(ip, IWRITE, cr, 0)) {
562 			goto update_inode;
563 		}
564 		if (vap->va_size > MAXOFFSET_T) {
565 			error = EFBIG;
566 			goto update_inode;
567 		}
568 		if (error = ud_itrunc(ip, vap->va_size, 0, cr)) {
569 			goto update_inode;
570 		}
571 
572 		if (vap->va_size == 0)
573 			vnevent_truncate(vp, ct);
574 	}
575 	/*
576 	 * Change file access or modified times.
577 	 */
578 	if (mask & (AT_ATIME|AT_MTIME)) {
579 		mutex_enter(&ip->i_tlock);
580 		if (mask & AT_ATIME) {
581 			ip->i_atime.tv_sec = vap->va_atime.tv_sec;
582 			ip->i_atime.tv_nsec = vap->va_atime.tv_nsec;
583 			ip->i_flag &= ~IACC;
584 		}
585 		if (mask & AT_MTIME) {
586 			ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
587 			ip->i_mtime.tv_nsec = vap->va_mtime.tv_nsec;
588 			gethrestime(&now);
589 			ip->i_ctime.tv_sec = now.tv_sec;
590 			ip->i_ctime.tv_nsec = now.tv_nsec;
591 			ip->i_flag &= ~(IUPD|ICHG);
592 			ip->i_flag |= IMODTIME;
593 		}
594 		ip->i_flag |= IMOD;
595 		mutex_exit(&ip->i_tlock);
596 	}
597 
598 update_inode:
599 	if (curthread->t_flag & T_DONTPEND) {
600 		ud_iupdat(ip, 1);
601 	} else {
602 		ITIMES_NOLOCK(ip);
603 	}
604 	rw_exit(&ip->i_contents);
605 	rw_exit(&ip->i_rwlock);
606 
607 	return (error);
608 }
609 
610 /* ARGSUSED */
611 static int32_t
612 udf_access(
613 	struct vnode *vp,
614 	int32_t mode,
615 	int32_t flags,
616 	struct cred *cr,
617 	caller_context_t *ct)
618 {
619 	struct ud_inode *ip = VTOI(vp);
620 
621 	ud_printf("udf_access\n");
622 
623 	if (ip->i_udf == NULL) {
624 		return (EIO);
625 	}
626 
627 	return (ud_iaccess(ip, UD_UPERM2DPERM(mode), cr, 1));
628 }
629 
630 int32_t udfs_stickyhack = 1;
631 
632 /* ARGSUSED */
633 static int32_t
634 udf_lookup(
635 	struct vnode *dvp,
636 	char *nm,
637 	struct vnode **vpp,
638 	struct pathname *pnp,
639 	int32_t flags,
640 	struct vnode *rdir,
641 	struct cred *cr,
642 	caller_context_t *ct,
643 	int *direntflags,
644 	pathname_t *realpnp)
645 {
646 	int32_t error;
647 	struct vnode *vp;
648 	struct ud_inode *ip, *xip;
649 
650 	ud_printf("udf_lookup\n");
651 	/*
652 	 * Null component name is a synonym for directory being searched.
653 	 */
654 	if (*nm == '\0') {
655 		VN_HOLD(dvp);
656 		*vpp = dvp;
657 		error = 0;
658 		goto out;
659 	}
660 
661 	/*
662 	 * Fast path: Check the directory name lookup cache.
663 	 */
664 	ip = VTOI(dvp);
665 	if (vp = dnlc_lookup(dvp, nm)) {
666 		/*
667 		 * Check accessibility of directory.
668 		 */
669 		if ((error = ud_iaccess(ip, IEXEC, cr, 1)) != 0) {
670 			VN_RELE(vp);
671 		}
672 		xip = VTOI(vp);
673 	} else {
674 		error = ud_dirlook(ip, nm, &xip, cr, 1);
675 		ITIMES(ip);
676 	}
677 
678 	if (error == 0) {
679 		ip = xip;
680 		*vpp = ITOV(ip);
681 		if ((ip->i_type != VDIR) &&
682 		    (ip->i_char & ISVTX) &&
683 		    ((ip->i_perm & IEXEC) == 0) &&
684 		    udfs_stickyhack) {
685 			mutex_enter(&(*vpp)->v_lock);
686 			(*vpp)->v_flag |= VISSWAP;
687 			mutex_exit(&(*vpp)->v_lock);
688 		}
689 		ITIMES(ip);
690 		/*
691 		 * If vnode is a device return special vnode instead.
692 		 */
693 		if (IS_DEVVP(*vpp)) {
694 			struct vnode *newvp;
695 			newvp = specvp(*vpp, (*vpp)->v_rdev,
696 			    (*vpp)->v_type, cr);
697 			VN_RELE(*vpp);
698 			if (newvp == NULL) {
699 				error = ENOSYS;
700 			} else {
701 				*vpp = newvp;
702 			}
703 		}
704 	}
705 out:
706 	return (error);
707 }
708 
709 /* ARGSUSED */
710 static int32_t
711 udf_create(
712 	struct vnode *dvp,
713 	char *name,
714 	struct vattr *vap,
715 	enum vcexcl excl,
716 	int32_t mode,
717 	struct vnode **vpp,
718 	struct cred *cr,
719 	int32_t flag,
720 	caller_context_t *ct,
721 	vsecattr_t *vsecp)
722 {
723 	int32_t error;
724 	struct ud_inode *ip = VTOI(dvp), *xip;
725 
726 	ud_printf("udf_create\n");
727 
728 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
729 		vap->va_mode &= ~VSVTX;
730 
731 	if (*name == '\0') {
732 		/*
733 		 * Null component name refers to the directory itself.
734 		 */
735 		VN_HOLD(dvp);
736 		ITIMES(ip);
737 		error = EEXIST;
738 	} else {
739 		xip = NULL;
740 		rw_enter(&ip->i_rwlock, RW_WRITER);
741 		error = ud_direnter(ip, name, DE_CREATE,
742 		    (struct ud_inode *)0, (struct ud_inode *)0,
743 		    vap, &xip, cr, ct);
744 		rw_exit(&ip->i_rwlock);
745 		ITIMES(ip);
746 		ip = xip;
747 	}
748 #ifdef	__lock_lint
749 	rw_enter(&ip->i_contents, RW_WRITER);
750 #else
751 	if (ip != NULL) {
752 		rw_enter(&ip->i_contents, RW_WRITER);
753 	}
754 #endif
755 
756 	/*
757 	 * If the file already exists and this is a non-exclusive create,
758 	 * check permissions and allow access for non-directories.
759 	 * Read-only create of an existing directory is also allowed.
760 	 * We fail an exclusive create of anything which already exists.
761 	 */
762 	if (error == EEXIST) {
763 		if (excl == NONEXCL) {
764 			if ((ip->i_type == VDIR) && (mode & VWRITE)) {
765 				error = EISDIR;
766 			} else if (mode) {
767 				error = ud_iaccess(ip,
768 				    UD_UPERM2DPERM(mode), cr, 0);
769 			} else {
770 				error = 0;
771 			}
772 		}
773 		if (error) {
774 			rw_exit(&ip->i_contents);
775 			VN_RELE(ITOV(ip));
776 			goto out;
777 		} else if ((ip->i_type == VREG) &&
778 		    (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
779 			/*
780 			 * Truncate regular files, if requested by caller.
781 			 * Grab i_rwlock to make sure no one else is
782 			 * currently writing to the file (we promised
783 			 * bmap we would do this).
784 			 * Must get the locks in the correct order.
785 			 */
786 			if (ip->i_size == 0) {
787 				ip->i_flag |= ICHG | IUPD;
788 			} else {
789 				rw_exit(&ip->i_contents);
790 				rw_enter(&ip->i_rwlock, RW_WRITER);
791 				rw_enter(&ip->i_contents, RW_WRITER);
792 				(void) ud_itrunc(ip, 0, 0, cr);
793 				rw_exit(&ip->i_rwlock);
794 			}
795 			vnevent_create(ITOV(ip), ct);
796 		}
797 	}
798 
799 	if (error == 0) {
800 		*vpp = ITOV(ip);
801 		ITIMES(ip);
802 	}
803 #ifdef	__lock_lint
804 	rw_exit(&ip->i_contents);
805 #else
806 	if (ip != NULL) {
807 		rw_exit(&ip->i_contents);
808 	}
809 #endif
810 	if (error) {
811 		goto out;
812 	}
813 
814 	/*
815 	 * If vnode is a device return special vnode instead.
816 	 */
817 	if (!error && IS_DEVVP(*vpp)) {
818 		struct vnode *newvp;
819 
820 		newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
821 		VN_RELE(*vpp);
822 		if (newvp == NULL) {
823 			error = ENOSYS;
824 			goto out;
825 		}
826 		*vpp = newvp;
827 	}
828 out:
829 	return (error);
830 }
831 
832 /* ARGSUSED */
833 static int32_t
834 udf_remove(
835 	struct vnode *vp,
836 	char *nm,
837 	struct cred *cr,
838 	caller_context_t *ct,
839 	int flags)
840 {
841 	int32_t error;
842 	struct ud_inode *ip = VTOI(vp);
843 
844 	ud_printf("udf_remove\n");
845 
846 	rw_enter(&ip->i_rwlock, RW_WRITER);
847 	error = ud_dirremove(ip, nm,
848 	    (struct ud_inode *)0, (struct vnode *)0, DR_REMOVE, cr, ct);
849 	rw_exit(&ip->i_rwlock);
850 	ITIMES(ip);
851 
852 	return (error);
853 }
854 
855 /* ARGSUSED */
856 static int32_t
857 udf_link(
858 	struct vnode *tdvp,
859 	struct vnode *svp,
860 	char *tnm,
861 	struct cred *cr,
862 	caller_context_t *ct,
863 	int flags)
864 {
865 	int32_t error;
866 	struct vnode *realvp;
867 	struct ud_inode *sip;
868 	struct ud_inode *tdp;
869 
870 	ud_printf("udf_link\n");
871 	if (VOP_REALVP(svp, &realvp, ct) == 0) {
872 		svp = realvp;
873 	}
874 
875 	/*
876 	 * Do not allow links to directories
877 	 */
878 	if (svp->v_type == VDIR) {
879 		return (EPERM);
880 	}
881 
882 	sip = VTOI(svp);
883 
884 	if (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)
885 		return (EPERM);
886 
887 	tdp = VTOI(tdvp);
888 
889 	rw_enter(&tdp->i_rwlock, RW_WRITER);
890 	error = ud_direnter(tdp, tnm, DE_LINK, (struct ud_inode *)0,
891 	    sip, (struct vattr *)0, (struct ud_inode **)0, cr, ct);
892 	rw_exit(&tdp->i_rwlock);
893 	ITIMES(sip);
894 	ITIMES(tdp);
895 
896 	if (error == 0) {
897 		vnevent_link(svp, ct);
898 	}
899 
900 	return (error);
901 }
902 
903 /* ARGSUSED */
904 static int32_t
905 udf_rename(
906 	struct vnode *sdvp,
907 	char *snm,
908 	struct vnode *tdvp,
909 	char *tnm,
910 	struct cred *cr,
911 	caller_context_t *ct,
912 	int flags)
913 {
914 	int32_t error = 0;
915 	struct udf_vfs *udf_vfsp;
916 	struct ud_inode *sip;		/* source inode */
917 	struct ud_inode *sdp, *tdp;	/* source and target parent inode */
918 	struct vnode *realvp;
919 
920 	ud_printf("udf_rename\n");
921 
922 	if (VOP_REALVP(tdvp, &realvp, ct) == 0) {
923 		tdvp = realvp;
924 	}
925 
926 	sdp = VTOI(sdvp);
927 	tdp = VTOI(tdvp);
928 
929 	udf_vfsp = sdp->i_udf;
930 
931 	mutex_enter(&udf_vfsp->udf_rename_lck);
932 	/*
933 	 * Look up inode of file we're supposed to rename.
934 	 */
935 	if (error = ud_dirlook(sdp, snm, &sip, cr, 0)) {
936 		mutex_exit(&udf_vfsp->udf_rename_lck);
937 		return (error);
938 	}
939 	/*
940 	 * be sure this is not a directory with another file system mounted
941 	 * over it.  If it is just give up the locks, and return with
942 	 * EBUSY
943 	 */
944 	if (vn_mountedvfs(ITOV(sip)) != NULL) {
945 		error = EBUSY;
946 		goto errout;
947 	}
948 	/*
949 	 * Make sure we can delete the source entry.  This requires
950 	 * write permission on the containing directory.  If that
951 	 * directory is "sticky" it further requires (except for
952 	 * privileged users) that the user own the directory or the
953 	 * source entry, or else have permission to write the source
954 	 * entry.
955 	 */
956 	rw_enter(&sdp->i_contents, RW_READER);
957 	rw_enter(&sip->i_contents, RW_READER);
958 	if ((error = ud_iaccess(sdp, IWRITE, cr, 0)) != 0 ||
959 	    (error = ud_sticky_remove_access(sdp, sip, cr)) != 0) {
960 		rw_exit(&sip->i_contents);
961 		rw_exit(&sdp->i_contents);
962 		ITIMES(sip);
963 		goto errout;
964 	}
965 
966 	/*
967 	 * Check for renaming '.' or '..' or alias of '.'
968 	 */
969 	if ((strcmp(snm, ".") == 0) ||
970 	    (strcmp(snm, "..") == 0) ||
971 	    (sdp == sip)) {
972 		error = EINVAL;
973 		rw_exit(&sip->i_contents);
974 		rw_exit(&sdp->i_contents);
975 		goto errout;
976 	}
977 	rw_exit(&sip->i_contents);
978 	rw_exit(&sdp->i_contents);
979 
980 
981 	/*
982 	 * Link source to the target.
983 	 */
984 	rw_enter(&tdp->i_rwlock, RW_WRITER);
985 	if (error = ud_direnter(tdp, tnm, DE_RENAME, sdp, sip,
986 	    (struct vattr *)0, (struct ud_inode **)0, cr, ct)) {
987 		/*
988 		 * ESAME isn't really an error; it indicates that the
989 		 * operation should not be done because the source and target
990 		 * are the same file, but that no error should be reported.
991 		 */
992 		if (error == ESAME) {
993 			error = 0;
994 		}
995 		rw_exit(&tdp->i_rwlock);
996 		goto errout;
997 	}
998 	vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
999 	rw_exit(&tdp->i_rwlock);
1000 
1001 	rw_enter(&sdp->i_rwlock, RW_WRITER);
1002 	/*
1003 	 * Unlink the source.
1004 	 * Remove the source entry.  ud_dirremove() checks that the entry
1005 	 * still reflects sip, and returns an error if it doesn't.
1006 	 * If the entry has changed just forget about it.  Release
1007 	 * the source inode.
1008 	 */
1009 	if ((error = ud_dirremove(sdp, snm, sip, (struct vnode *)0,
1010 	    DR_RENAME, cr, ct)) == ENOENT) {
1011 		error = 0;
1012 	}
1013 	rw_exit(&sdp->i_rwlock);
1014 errout:
1015 	ITIMES(sdp);
1016 	ITIMES(tdp);
1017 	VN_RELE(ITOV(sip));
1018 	mutex_exit(&udf_vfsp->udf_rename_lck);
1019 
1020 	return (error);
1021 }
1022 
1023 /* ARGSUSED */
1024 static int32_t
1025 udf_mkdir(
1026 	struct vnode *dvp,
1027 	char *dirname,
1028 	struct vattr *vap,
1029 	struct vnode **vpp,
1030 	struct cred *cr,
1031 	caller_context_t *ct,
1032 	int flags,
1033 	vsecattr_t *vsecp)
1034 {
1035 	int32_t error;
1036 	struct ud_inode *ip;
1037 	struct ud_inode *xip;
1038 
1039 	ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
1040 
1041 	ud_printf("udf_mkdir\n");
1042 
1043 	ip = VTOI(dvp);
1044 	rw_enter(&ip->i_rwlock, RW_WRITER);
1045 	error = ud_direnter(ip, dirname, DE_MKDIR,
1046 	    (struct ud_inode *)0, (struct ud_inode *)0, vap, &xip, cr, ct);
1047 	rw_exit(&ip->i_rwlock);
1048 	ITIMES(ip);
1049 	if (error == 0) {
1050 		ip = xip;
1051 		*vpp = ITOV(ip);
1052 		ITIMES(ip);
1053 	} else if (error == EEXIST) {
1054 		ITIMES(xip);
1055 		VN_RELE(ITOV(xip));
1056 	}
1057 
1058 	return (error);
1059 }
1060 
1061 /* ARGSUSED */
1062 static int32_t
1063 udf_rmdir(
1064 	struct vnode *vp,
1065 	char *nm,
1066 	struct vnode *cdir,
1067 	struct cred *cr,
1068 	caller_context_t *ct,
1069 	int flags)
1070 {
1071 	int32_t error;
1072 	struct ud_inode *ip = VTOI(vp);
1073 
1074 	ud_printf("udf_rmdir\n");
1075 
1076 	rw_enter(&ip->i_rwlock, RW_WRITER);
1077 	error = ud_dirremove(ip, nm, (struct ud_inode *)0, cdir, DR_RMDIR,
1078 	    cr, ct);
1079 	rw_exit(&ip->i_rwlock);
1080 	ITIMES(ip);
1081 
1082 	return (error);
1083 }
1084 
1085 /* ARGSUSED */
1086 static int32_t
1087 udf_readdir(
1088 	struct vnode *vp,
1089 	struct uio *uiop,
1090 	struct cred *cr,
1091 	int32_t *eofp,
1092 	caller_context_t *ct,
1093 	int flags)
1094 {
1095 	struct ud_inode *ip;
1096 	struct dirent64 *nd;
1097 	struct udf_vfs *udf_vfsp;
1098 	int32_t error = 0, len, outcount = 0;
1099 	uint32_t dirsiz, offset;
1100 	uint32_t bufsize, ndlen, dummy;
1101 	caddr_t outbuf;
1102 	caddr_t outb, end_outb;
1103 	struct iovec *iovp;
1104 
1105 	uint8_t *dname;
1106 	int32_t length;
1107 
1108 	uint8_t *buf = NULL;
1109 
1110 	struct fbuf *fbp = NULL;
1111 	struct file_id *fid;
1112 	uint8_t *name;
1113 
1114 
1115 	ud_printf("udf_readdir\n");
1116 
1117 	ip = VTOI(vp);
1118 	udf_vfsp = ip->i_udf;
1119 
1120 	dirsiz = ip->i_size;
1121 	if ((uiop->uio_offset >= dirsiz) ||
1122 	    (ip->i_nlink <= 0)) {
1123 		if (eofp) {
1124 			*eofp = 1;
1125 		}
1126 		return (0);
1127 	}
1128 
1129 	offset = uiop->uio_offset;
1130 	iovp = uiop->uio_iov;
1131 	bufsize = iovp->iov_len;
1132 
1133 	outb = outbuf = (char *)kmem_alloc((uint32_t)bufsize, KM_SLEEP);
1134 	end_outb = outb + bufsize;
1135 	nd = (struct dirent64 *)outbuf;
1136 
1137 	dname = (uint8_t *)kmem_zalloc(1024, KM_SLEEP);
1138 	buf = (uint8_t *)kmem_zalloc(udf_vfsp->udf_lbsize, KM_SLEEP);
1139 
1140 	if (offset == 0) {
1141 		len = DIRENT64_RECLEN(1);
1142 		if (((caddr_t)nd + len) >= end_outb) {
1143 			error = EINVAL;
1144 			goto end;
1145 		}
1146 		nd->d_ino = ip->i_icb_lbano;
1147 		nd->d_reclen = (uint16_t)len;
1148 		nd->d_off = 0x10;
1149 		nd->d_name[0] = '.';
1150 		bzero(&nd->d_name[1], DIRENT64_NAMELEN(len) - 1);
1151 		nd = (struct dirent64 *)((char *)nd + nd->d_reclen);
1152 		outcount++;
1153 	} else if (offset == 0x10) {
1154 		offset = 0;
1155 	}
1156 
1157 	while (offset < dirsiz) {
1158 		error = ud_get_next_fid(ip, &fbp,
1159 		    offset, &fid, &name, buf);
1160 		if (error != 0) {
1161 			break;
1162 		}
1163 
1164 		if ((fid->fid_flags & FID_DELETED) == 0) {
1165 			if (fid->fid_flags & FID_PARENT) {
1166 
1167 				len = DIRENT64_RECLEN(2);
1168 				if (((caddr_t)nd + len) >= end_outb) {
1169 					error = EINVAL;
1170 					break;
1171 				}
1172 
1173 				nd->d_ino = ip->i_icb_lbano;
1174 				nd->d_reclen = (uint16_t)len;
1175 				nd->d_off = offset + FID_LEN(fid);
1176 				nd->d_name[0] = '.';
1177 				nd->d_name[1] = '.';
1178 				bzero(&nd->d_name[2],
1179 				    DIRENT64_NAMELEN(len) - 2);
1180 				nd = (struct dirent64 *)
1181 				    ((char *)nd + nd->d_reclen);
1182 			} else {
1183 				if ((error = ud_uncompress(fid->fid_idlen,
1184 				    &length, name, dname)) != 0) {
1185 					break;
1186 				}
1187 				if (length == 0) {
1188 					offset += FID_LEN(fid);
1189 					continue;
1190 				}
1191 				len = DIRENT64_RECLEN(length);
1192 				if (((caddr_t)nd + len) >= end_outb) {
1193 					if (!outcount) {
1194 						error = EINVAL;
1195 					}
1196 					break;
1197 				}
1198 				(void) strncpy(nd->d_name,
1199 				    (caddr_t)dname, length);
1200 				bzero(&nd->d_name[length],
1201 				    DIRENT64_NAMELEN(len) - length);
1202 				nd->d_ino = ud_xlate_to_daddr(udf_vfsp,
1203 				    SWAP_16(fid->fid_icb.lad_ext_prn),
1204 				    SWAP_32(fid->fid_icb.lad_ext_loc), 1,
1205 				    &dummy);
1206 				nd->d_reclen = (uint16_t)len;
1207 				nd->d_off = offset + FID_LEN(fid);
1208 				nd = (struct dirent64 *)
1209 				    ((char *)nd + nd->d_reclen);
1210 			}
1211 			outcount++;
1212 		}
1213 
1214 		offset += FID_LEN(fid);
1215 	}
1216 
1217 end:
1218 	if (fbp != NULL) {
1219 		fbrelse(fbp, S_OTHER);
1220 	}
1221 	ndlen = ((char *)nd - outbuf);
1222 	/*
1223 	 * In case of error do not call uiomove.
1224 	 * Return the error to the caller.
1225 	 */
1226 	if ((error == 0) && (ndlen != 0)) {
1227 		error = uiomove(outbuf, (long)ndlen, UIO_READ, uiop);
1228 		uiop->uio_offset = offset;
1229 	}
1230 	kmem_free((caddr_t)buf, udf_vfsp->udf_lbsize);
1231 	kmem_free((caddr_t)dname, 1024);
1232 	kmem_free(outbuf, (uint32_t)bufsize);
1233 	if (eofp && error == 0) {
1234 		*eofp = (uiop->uio_offset >= dirsiz);
1235 	}
1236 	return (error);
1237 }
1238 
1239 /* ARGSUSED */
1240 static int32_t
1241 udf_symlink(
1242 	struct vnode *dvp,
1243 	char *linkname,
1244 	struct vattr *vap,
1245 	char *target,
1246 	struct cred *cr,
1247 	caller_context_t *ct,
1248 	int flags)
1249 {
1250 	int32_t error = 0, outlen;
1251 	uint32_t ioflag = 0;
1252 	struct ud_inode *ip, *dip = VTOI(dvp);
1253 
1254 	struct path_comp *pc;
1255 	int8_t *dname = NULL, *uname = NULL, *sp;
1256 
1257 	ud_printf("udf_symlink\n");
1258 
1259 	ip = (struct ud_inode *)0;
1260 	vap->va_type = VLNK;
1261 	vap->va_rdev = 0;
1262 
1263 	rw_enter(&dip->i_rwlock, RW_WRITER);
1264 	error = ud_direnter(dip, linkname, DE_CREATE,
1265 	    (struct ud_inode *)0, (struct ud_inode *)0, vap, &ip, cr, ct);
1266 	rw_exit(&dip->i_rwlock);
1267 	if (error == 0) {
1268 		dname = kmem_zalloc(1024, KM_SLEEP);
1269 		uname = kmem_zalloc(PAGESIZE, KM_SLEEP);
1270 
1271 		pc = (struct path_comp *)uname;
1272 		/*
1273 		 * If the first character in target is "/"
1274 		 * then skip it and create entry for it
1275 		 */
1276 		if (*target == '/') {
1277 			pc->pc_type = 2;
1278 			pc->pc_len = 0;
1279 			pc = (struct path_comp *)(((char *)pc) + 4);
1280 			while (*target == '/') {
1281 				target++;
1282 			}
1283 		}
1284 
1285 		while (*target != NULL) {
1286 			sp = target;
1287 			while ((*target != '/') && (*target != '\0')) {
1288 				target ++;
1289 			}
1290 			/*
1291 			 * We got the next component of the
1292 			 * path name. Create path_comp of
1293 			 * appropriate type
1294 			 */
1295 			if (((target - sp) == 1) && (*sp == '.')) {
1296 				/*
1297 				 * Dot entry.
1298 				 */
1299 				pc->pc_type = 4;
1300 				pc = (struct path_comp *)(((char *)pc) + 4);
1301 			} else if (((target - sp) == 2) &&
1302 			    (*sp == '.') && ((*(sp + 1)) == '.')) {
1303 				/*
1304 				 * DotDot entry.
1305 				 */
1306 				pc->pc_type = 3;
1307 				pc = (struct path_comp *)(((char *)pc) + 4);
1308 			} else {
1309 				/*
1310 				 * convert the user given name
1311 				 * into appropriate form to be put
1312 				 * on the media
1313 				 */
1314 				outlen = 1024;	/* set to size of dname */
1315 				if (error = ud_compress(target - sp, &outlen,
1316 				    (uint8_t *)sp, (uint8_t *)dname)) {
1317 					break;
1318 				}
1319 				pc->pc_type = 5;
1320 				/* LINTED */
1321 				pc->pc_len = outlen;
1322 				dname[outlen] = '\0';
1323 				(void) strcpy((char *)pc->pc_id, dname);
1324 				pc = (struct path_comp *)
1325 				    (((char *)pc) + 4 + outlen);
1326 			}
1327 			while (*target == '/') {
1328 				target++;
1329 			}
1330 			if (*target == NULL) {
1331 				break;
1332 			}
1333 		}
1334 
1335 		rw_enter(&ip->i_contents, RW_WRITER);
1336 		if (error == 0) {
1337 			ioflag = FWRITE;
1338 			if (curthread->t_flag & T_DONTPEND) {
1339 				ioflag |= FDSYNC;
1340 			}
1341 			error = ud_rdwri(UIO_WRITE, ioflag, ip,
1342 			    uname, ((int8_t *)pc) - uname,
1343 			    (offset_t)0, UIO_SYSSPACE, (int32_t *)0, cr);
1344 		}
1345 		if (error) {
1346 			ud_idrop(ip);
1347 			rw_exit(&ip->i_contents);
1348 			rw_enter(&dip->i_rwlock, RW_WRITER);
1349 			(void) ud_dirremove(dip, linkname, (struct ud_inode *)0,
1350 			    (struct vnode *)0, DR_REMOVE, cr, ct);
1351 			rw_exit(&dip->i_rwlock);
1352 			goto update_inode;
1353 		}
1354 		rw_exit(&ip->i_contents);
1355 	}
1356 
1357 	if ((error == 0) || (error == EEXIST)) {
1358 		VN_RELE(ITOV(ip));
1359 	}
1360 
1361 update_inode:
1362 	ITIMES(VTOI(dvp));
1363 	if (uname != NULL) {
1364 		kmem_free(uname, PAGESIZE);
1365 	}
1366 	if (dname != NULL) {
1367 		kmem_free(dname, 1024);
1368 	}
1369 
1370 	return (error);
1371 }
1372 
1373 /* ARGSUSED */
1374 static int32_t
1375 udf_readlink(
1376 	struct vnode *vp,
1377 	struct uio *uiop,
1378 	struct cred *cr,
1379 	caller_context_t *ct)
1380 {
1381 	int32_t error = 0, off, id_len, size, len;
1382 	int8_t *dname = NULL, *uname = NULL;
1383 	struct ud_inode *ip;
1384 	struct fbuf *fbp = NULL;
1385 	struct path_comp *pc;
1386 
1387 	ud_printf("udf_readlink\n");
1388 
1389 	if (vp->v_type != VLNK) {
1390 		return (EINVAL);
1391 	}
1392 
1393 	ip = VTOI(vp);
1394 	size = ip->i_size;
1395 	if (size > PAGESIZE) {
1396 		return (EIO);
1397 	}
1398 
1399 	if (size == 0) {
1400 		return (0);
1401 	}
1402 
1403 	dname = kmem_zalloc(1024, KM_SLEEP);
1404 	uname = kmem_zalloc(PAGESIZE, KM_SLEEP);
1405 
1406 	rw_enter(&ip->i_contents, RW_READER);
1407 
1408 	if ((error = fbread(vp, 0, size, S_READ, &fbp)) != 0) {
1409 		goto end;
1410 	}
1411 
1412 	off = 0;
1413 
1414 	while (off < size) {
1415 		pc = (struct path_comp *)(fbp->fb_addr + off);
1416 		switch (pc->pc_type) {
1417 			case 1 :
1418 				(void) strcpy(uname, ip->i_udf->udf_fsmnt);
1419 				(void) strcat(uname, "/");
1420 				break;
1421 			case 2 :
1422 				if (pc->pc_len != 0) {
1423 					goto end;
1424 				}
1425 				uname[0] = '/';
1426 				uname[1] = '\0';
1427 				break;
1428 			case 3 :
1429 				(void) strcat(uname, "../");
1430 				break;
1431 			case 4 :
1432 				(void) strcat(uname, "./");
1433 				break;
1434 			case 5 :
1435 				if ((error = ud_uncompress(pc->pc_len, &id_len,
1436 				    pc->pc_id, (uint8_t *)dname)) != 0) {
1437 					break;
1438 				}
1439 				dname[id_len] = '\0';
1440 				(void) strcat(uname, dname);
1441 				(void) strcat(uname, "/");
1442 				break;
1443 			default :
1444 				error = EINVAL;
1445 				goto end;
1446 		}
1447 		off += 4 + pc->pc_len;
1448 	}
1449 	len = strlen(uname) - 1;
1450 	if (uname[len] == '/') {
1451 		if (len == 0) {
1452 			/*
1453 			 * special case link to /
1454 			 */
1455 			len = 1;
1456 		} else {
1457 			uname[len] = '\0';
1458 		}
1459 	}
1460 
1461 	error = uiomove(uname, len, UIO_READ, uiop);
1462 
1463 	ITIMES(ip);
1464 
1465 end:
1466 	if (fbp != NULL) {
1467 		fbrelse(fbp, S_OTHER);
1468 	}
1469 	rw_exit(&ip->i_contents);
1470 	if (uname != NULL) {
1471 		kmem_free(uname, PAGESIZE);
1472 	}
1473 	if (dname != NULL) {
1474 		kmem_free(dname, 1024);
1475 	}
1476 	return (error);
1477 }
1478 
1479 /* ARGSUSED */
1480 static int32_t
1481 udf_fsync(
1482 	struct vnode *vp,
1483 	int32_t syncflag,
1484 	struct cred *cr,
1485 	caller_context_t *ct)
1486 {
1487 	int32_t error = 0;
1488 	struct ud_inode *ip = VTOI(vp);
1489 
1490 	ud_printf("udf_fsync\n");
1491 
1492 	rw_enter(&ip->i_contents, RW_WRITER);
1493 	if (!(IS_SWAPVP(vp))) {
1494 		error = ud_syncip(ip, 0, I_SYNC); /* Do synchronous writes */
1495 	}
1496 	if (error == 0) {
1497 		error = ud_sync_indir(ip);
1498 	}
1499 	ITIMES(ip);		/* XXX: is this necessary ??? */
1500 	rw_exit(&ip->i_contents);
1501 
1502 	return (error);
1503 }
1504 
1505 /* ARGSUSED */
1506 static void
1507 udf_inactive(struct vnode *vp, struct cred *cr, caller_context_t *ct)
1508 {
1509 	ud_printf("udf_iinactive\n");
1510 
1511 	ud_iinactive(VTOI(vp), cr);
1512 }
1513 
1514 /* ARGSUSED */
1515 static int32_t
1516 udf_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
1517 {
1518 	struct udf_fid *udfidp;
1519 	struct ud_inode *ip = VTOI(vp);
1520 
1521 	ud_printf("udf_fid\n");
1522 
1523 	if (fidp->fid_len < (sizeof (struct udf_fid) - sizeof (uint16_t))) {
1524 		fidp->fid_len = sizeof (struct udf_fid) - sizeof (uint16_t);
1525 		return (ENOSPC);
1526 	}
1527 
1528 	udfidp = (struct udf_fid *)fidp;
1529 	bzero((char *)udfidp, sizeof (struct udf_fid));
1530 	rw_enter(&ip->i_contents, RW_READER);
1531 	udfidp->udfid_len = sizeof (struct udf_fid) - sizeof (uint16_t);
1532 	udfidp->udfid_uinq_lo = ip->i_uniqid & 0xffffffff;
1533 	udfidp->udfid_prn = ip->i_icb_prn;
1534 	udfidp->udfid_icb_lbn = ip->i_icb_block;
1535 	rw_exit(&ip->i_contents);
1536 
1537 	return (0);
1538 }
1539 
1540 /* ARGSUSED2 */
1541 static int
1542 udf_rwlock(struct vnode *vp, int32_t write_lock, caller_context_t *ctp)
1543 {
1544 	struct ud_inode *ip = VTOI(vp);
1545 
1546 	ud_printf("udf_rwlock\n");
1547 
1548 	if (write_lock) {
1549 		rw_enter(&ip->i_rwlock, RW_WRITER);
1550 	} else {
1551 		rw_enter(&ip->i_rwlock, RW_READER);
1552 	}
1553 #ifdef	__lock_lint
1554 	rw_exit(&ip->i_rwlock);
1555 #endif
1556 	return (write_lock);
1557 }
1558 
1559 /* ARGSUSED */
1560 static void
1561 udf_rwunlock(struct vnode *vp, int32_t write_lock, caller_context_t *ctp)
1562 {
1563 	struct ud_inode *ip = VTOI(vp);
1564 
1565 	ud_printf("udf_rwunlock\n");
1566 
1567 #ifdef	__lock_lint
1568 	rw_enter(&ip->i_rwlock, RW_WRITER);
1569 #endif
1570 
1571 	rw_exit(&ip->i_rwlock);
1572 
1573 }
1574 
1575 /* ARGSUSED */
1576 static int32_t
1577 udf_seek(struct vnode *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
1578 {
1579 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
1580 }
1581 
1582 static int32_t
1583 udf_frlock(
1584 	struct vnode *vp,
1585 	int32_t cmd,
1586 	struct flock64 *bfp,
1587 	int32_t flag,
1588 	offset_t offset,
1589 	struct flk_callback *flk_cbp,
1590 	cred_t *cr,
1591 	caller_context_t *ct)
1592 {
1593 	struct ud_inode *ip = VTOI(vp);
1594 
1595 	ud_printf("udf_frlock\n");
1596 
1597 	/*
1598 	 * If file is being mapped, disallow frlock.
1599 	 * XXX I am not holding tlock while checking i_mapcnt because the
1600 	 * current locking strategy drops all locks before calling fs_frlock.
1601 	 * So, mapcnt could change before we enter fs_frlock making is
1602 	 * meaningless to have held tlock in the first place.
1603 	 */
1604 	if ((ip->i_mapcnt > 0) &&
1605 	    (MANDLOCK(vp, ip->i_char))) {
1606 		return (EAGAIN);
1607 	}
1608 
1609 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
1610 }
1611 
1612 /*ARGSUSED6*/
1613 static int32_t
1614 udf_space(
1615 	struct vnode *vp,
1616 	int32_t cmd,
1617 	struct flock64 *bfp,
1618 	int32_t flag,
1619 	offset_t offset,
1620 	cred_t *cr,
1621 	caller_context_t *ct)
1622 {
1623 	int32_t error = 0;
1624 
1625 	ud_printf("udf_space\n");
1626 
1627 	if (cmd != F_FREESP) {
1628 		error =  EINVAL;
1629 	} else if ((error = convoff(vp, bfp, 0, offset)) == 0) {
1630 		error = ud_freesp(vp, bfp, flag, cr);
1631 
1632 		if (error == 0 && bfp->l_start == 0)
1633 			vnevent_truncate(vp, ct);
1634 	}
1635 
1636 	return (error);
1637 }
1638 
1639 /* ARGSUSED */
1640 static int32_t
1641 udf_getpage(
1642 	struct vnode *vp,
1643 	offset_t off,
1644 	size_t len,
1645 	uint32_t *protp,
1646 	struct page **plarr,
1647 	size_t plsz,
1648 	struct seg *seg,
1649 	caddr_t addr,
1650 	enum seg_rw rw,
1651 	struct cred *cr,
1652 	caller_context_t *ct)
1653 {
1654 	struct ud_inode *ip = VTOI(vp);
1655 	int32_t error, has_holes, beyond_eof, seqmode, dolock;
1656 	int32_t pgsize = PAGESIZE;
1657 	struct udf_vfs *udf_vfsp = ip->i_udf;
1658 	page_t **pl;
1659 	u_offset_t pgoff, eoff, uoff;
1660 	krw_t rwtype;
1661 	caddr_t pgaddr;
1662 
1663 	ud_printf("udf_getpage\n");
1664 
1665 	uoff = (u_offset_t)off; /* type conversion */
1666 	if (protp) {
1667 		*protp = PROT_ALL;
1668 	}
1669 	if (vp->v_flag & VNOMAP) {
1670 		return (ENOSYS);
1671 	}
1672 	seqmode = ip->i_nextr == uoff && rw != S_CREATE;
1673 
1674 	rwtype = RW_READER;
1675 	dolock = (rw_owner(&ip->i_contents) != curthread);
1676 retrylock:
1677 #ifdef	__lock_lint
1678 	rw_enter(&ip->i_contents, rwtype);
1679 #else
1680 	if (dolock) {
1681 		rw_enter(&ip->i_contents, rwtype);
1682 	}
1683 #endif
1684 
1685 	/*
1686 	 * We may be getting called as a side effect of a bmap using
1687 	 * fbread() when the blocks might be being allocated and the
1688 	 * size has not yet been up'ed.  In this case we want to be
1689 	 * able to return zero pages if we get back UDF_HOLE from
1690 	 * calling bmap for a non write case here.  We also might have
1691 	 * to read some frags from the disk into a page if we are
1692 	 * extending the number of frags for a given lbn in bmap().
1693 	 */
1694 	beyond_eof = uoff + len > ip->i_size + PAGEOFFSET;
1695 	if (beyond_eof && seg != segkmap) {
1696 #ifdef	__lock_lint
1697 		rw_exit(&ip->i_contents);
1698 #else
1699 		if (dolock) {
1700 			rw_exit(&ip->i_contents);
1701 		}
1702 #endif
1703 		return (EFAULT);
1704 	}
1705 
1706 	/*
1707 	 * Must hold i_contents lock throughout the call to pvn_getpages
1708 	 * since locked pages are returned from each call to ud_getapage.
1709 	 * Must *not* return locked pages and then try for contents lock
1710 	 * due to lock ordering requirements (inode > page)
1711 	 */
1712 
1713 	has_holes = ud_bmap_has_holes(ip);
1714 
1715 	if ((rw == S_WRITE || rw == S_CREATE) && (has_holes || beyond_eof)) {
1716 		int32_t	blk_size, count;
1717 		u_offset_t offset;
1718 
1719 		/*
1720 		 * We must acquire the RW_WRITER lock in order to
1721 		 * call bmap_write().
1722 		 */
1723 		if (dolock && rwtype == RW_READER) {
1724 			rwtype = RW_WRITER;
1725 
1726 			if (!rw_tryupgrade(&ip->i_contents)) {
1727 
1728 				rw_exit(&ip->i_contents);
1729 
1730 				goto retrylock;
1731 			}
1732 		}
1733 
1734 		/*
1735 		 * May be allocating disk blocks for holes here as
1736 		 * a result of mmap faults. write(2) does the bmap_write
1737 		 * in rdip/wrip, not here. We are not dealing with frags
1738 		 * in this case.
1739 		 */
1740 		offset = uoff;
1741 		while ((offset < uoff + len) &&
1742 		    (offset < ip->i_size)) {
1743 			/*
1744 			 * the variable "bnp" is to simplify the expression for
1745 			 * the compiler; * just passing in &bn to bmap_write
1746 			 * causes a compiler "loop"
1747 			 */
1748 
1749 			blk_size = udf_vfsp->udf_lbsize;
1750 			if ((offset + blk_size) > ip->i_size) {
1751 				count = ip->i_size - offset;
1752 			} else {
1753 				count = blk_size;
1754 			}
1755 			error = ud_bmap_write(ip, offset, count, 0, cr);
1756 			if (error) {
1757 				goto update_inode;
1758 			}
1759 			offset += count; /* XXX - make this contig */
1760 		}
1761 	}
1762 
1763 	/*
1764 	 * Can be a reader from now on.
1765 	 */
1766 #ifdef	__lock_lint
1767 	if (rwtype == RW_WRITER) {
1768 		rw_downgrade(&ip->i_contents);
1769 	}
1770 #else
1771 	if (dolock && rwtype == RW_WRITER) {
1772 		rw_downgrade(&ip->i_contents);
1773 	}
1774 #endif
1775 
1776 	/*
1777 	 * We remove PROT_WRITE in cases when the file has UDF holes
1778 	 * because we don't  want to call bmap_read() to check each
1779 	 * page if it is backed with a disk block.
1780 	 */
1781 	if (protp && has_holes && rw != S_WRITE && rw != S_CREATE) {
1782 		*protp &= ~PROT_WRITE;
1783 	}
1784 
1785 	error = 0;
1786 
1787 	/*
1788 	 * The loop looks up pages in the range <off, off + len).
1789 	 * For each page, we first check if we should initiate an asynchronous
1790 	 * read ahead before we call page_lookup (we may sleep in page_lookup
1791 	 * for a previously initiated disk read).
1792 	 */
1793 	eoff = (uoff + len);
1794 	for (pgoff = uoff, pgaddr = addr, pl = plarr;
1795 	    pgoff < eoff; /* empty */) {
1796 		page_t	*pp;
1797 		u_offset_t	nextrio;
1798 		se_t	se;
1799 
1800 		se = ((rw == S_CREATE) ? SE_EXCL : SE_SHARED);
1801 
1802 		/*
1803 		 * Handle async getpage (faultahead)
1804 		 */
1805 		if (plarr == NULL) {
1806 			ip->i_nextrio = pgoff;
1807 			ud_getpage_ra(vp, pgoff, seg, pgaddr);
1808 			pgoff += pgsize;
1809 			pgaddr += pgsize;
1810 			continue;
1811 		}
1812 
1813 		/*
1814 		 * Check if we should initiate read ahead of next cluster.
1815 		 * We call page_exists only when we need to confirm that
1816 		 * we have the current page before we initiate the read ahead.
1817 		 */
1818 		nextrio = ip->i_nextrio;
1819 		if (seqmode &&
1820 		    pgoff + RD_CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
1821 		    nextrio < ip->i_size && page_exists(vp, pgoff))
1822 			ud_getpage_ra(vp, pgoff, seg, pgaddr);
1823 
1824 		if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
1825 
1826 			/*
1827 			 * We found the page in the page cache.
1828 			 */
1829 			*pl++ = pp;
1830 			pgoff += pgsize;
1831 			pgaddr += pgsize;
1832 			len -= pgsize;
1833 			plsz -= pgsize;
1834 		} else  {
1835 
1836 			/*
1837 			 * We have to create the page, or read it from disk.
1838 			 */
1839 			if (error = ud_getpage_miss(vp, pgoff, len,
1840 			    seg, pgaddr, pl, plsz, rw, seqmode)) {
1841 				goto error_out;
1842 			}
1843 
1844 			while (*pl != NULL) {
1845 				pl++;
1846 				pgoff += pgsize;
1847 				pgaddr += pgsize;
1848 				len -= pgsize;
1849 				plsz -= pgsize;
1850 			}
1851 		}
1852 	}
1853 
1854 	/*
1855 	 * Return pages up to plsz if they are in the page cache.
1856 	 * We cannot return pages if there is a chance that they are
1857 	 * backed with a UDF hole and rw is S_WRITE or S_CREATE.
1858 	 */
1859 	if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
1860 
1861 		ASSERT((protp == NULL) ||
1862 		    !(has_holes && (*protp & PROT_WRITE)));
1863 
1864 		eoff = pgoff + plsz;
1865 		while (pgoff < eoff) {
1866 			page_t		*pp;
1867 
1868 			if ((pp = page_lookup_nowait(vp, pgoff,
1869 			    SE_SHARED)) == NULL)
1870 				break;
1871 
1872 			*pl++ = pp;
1873 			pgoff += pgsize;
1874 			plsz -= pgsize;
1875 		}
1876 	}
1877 
1878 	if (plarr)
1879 		*pl = NULL;			/* Terminate page list */
1880 	ip->i_nextr = pgoff;
1881 
1882 error_out:
1883 	if (error && plarr) {
1884 		/*
1885 		 * Release any pages we have locked.
1886 		 */
1887 		while (pl > &plarr[0])
1888 			page_unlock(*--pl);
1889 
1890 		plarr[0] = NULL;
1891 	}
1892 
1893 update_inode:
1894 #ifdef	__lock_lint
1895 	rw_exit(&ip->i_contents);
1896 #else
1897 	if (dolock) {
1898 		rw_exit(&ip->i_contents);
1899 	}
1900 #endif
1901 
1902 	/*
1903 	 * If the inode is not already marked for IACC (in rwip() for read)
1904 	 * and the inode is not marked for no access time update (in rwip()
1905 	 * for write) then update the inode access time and mod time now.
1906 	 */
1907 	mutex_enter(&ip->i_tlock);
1908 	if ((ip->i_flag & (IACC | INOACC)) == 0) {
1909 		if ((rw != S_OTHER) && (ip->i_type != VDIR)) {
1910 			ip->i_flag |= IACC;
1911 		}
1912 		if (rw == S_WRITE) {
1913 			ip->i_flag |= IUPD;
1914 		}
1915 		ITIMES_NOLOCK(ip);
1916 	}
1917 	mutex_exit(&ip->i_tlock);
1918 
1919 	return (error);
1920 }
1921 
1922 int32_t ud_delay = 1;
1923 
1924 /* ARGSUSED */
1925 static int32_t
1926 udf_putpage(
1927 	struct vnode *vp,
1928 	offset_t off,
1929 	size_t len,
1930 	int32_t flags,
1931 	struct cred *cr,
1932 	caller_context_t *ct)
1933 {
1934 	struct ud_inode *ip;
1935 	int32_t error = 0;
1936 
1937 	ud_printf("udf_putpage\n");
1938 
1939 	ip = VTOI(vp);
1940 #ifdef	__lock_lint
1941 	rw_enter(&ip->i_contents, RW_WRITER);
1942 #endif
1943 
1944 	if (vp->v_count == 0) {
1945 		cmn_err(CE_WARN, "ud_putpage : bad v_count");
1946 		error = EINVAL;
1947 		goto out;
1948 	}
1949 
1950 	if (vp->v_flag & VNOMAP) {
1951 		error = ENOSYS;
1952 		goto out;
1953 	}
1954 
1955 	if (flags & B_ASYNC) {
1956 		if (ud_delay && len &&
1957 		    (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
1958 			mutex_enter(&ip->i_tlock);
1959 
1960 			/*
1961 			 * If nobody stalled, start a new cluster.
1962 			 */
1963 			if (ip->i_delaylen == 0) {
1964 				ip->i_delayoff = off;
1965 				ip->i_delaylen = len;
1966 				mutex_exit(&ip->i_tlock);
1967 				goto out;
1968 			}
1969 
1970 			/*
1971 			 * If we have a full cluster or they are not contig,
1972 			 * then push last cluster and start over.
1973 			 */
1974 			if (ip->i_delaylen >= WR_CLUSTSZ(ip) ||
1975 			    ip->i_delayoff + ip->i_delaylen != off) {
1976 				u_offset_t doff;
1977 				size_t dlen;
1978 
1979 				doff = ip->i_delayoff;
1980 				dlen = ip->i_delaylen;
1981 				ip->i_delayoff = off;
1982 				ip->i_delaylen = len;
1983 				mutex_exit(&ip->i_tlock);
1984 				error = ud_putpages(vp, doff, dlen, flags, cr);
1985 				/* LMXXX - flags are new val, not old */
1986 				goto out;
1987 			}
1988 
1989 			/*
1990 			 * There is something there, it's not full, and
1991 			 * it is contig.
1992 			 */
1993 			ip->i_delaylen += len;
1994 			mutex_exit(&ip->i_tlock);
1995 			goto out;
1996 		}
1997 
1998 		/*
1999 		 * Must have weird flags or we are not clustering.
2000 		 */
2001 	}
2002 
2003 	error = ud_putpages(vp, off, len, flags, cr);
2004 
2005 out:
2006 #ifdef	__lock_lint
2007 	rw_exit(&ip->i_contents);
2008 #endif
2009 	return (error);
2010 }
2011 
2012 /* ARGSUSED */
2013 static int32_t
2014 udf_map(
2015 	struct vnode *vp,
2016 	offset_t off,
2017 	struct as *as,
2018 	caddr_t *addrp,
2019 	size_t len,
2020 	uint8_t prot,
2021 	uint8_t maxprot,
2022 	uint32_t flags,
2023 	struct cred *cr,
2024 	caller_context_t *ct)
2025 {
2026 	struct segvn_crargs vn_a;
2027 	int32_t error = 0;
2028 
2029 	ud_printf("udf_map\n");
2030 
2031 	if (vp->v_flag & VNOMAP) {
2032 		error = ENOSYS;
2033 		goto end;
2034 	}
2035 
2036 	if ((off < (offset_t)0) ||
2037 	    ((off + len) < (offset_t)0)) {
2038 		error = EINVAL;
2039 		goto end;
2040 	}
2041 
2042 	if (vp->v_type != VREG) {
2043 		error = ENODEV;
2044 		goto end;
2045 	}
2046 
2047 	/*
2048 	 * If file is being locked, disallow mapping.
2049 	 */
2050 	if (vn_has_mandatory_locks(vp, VTOI(vp)->i_char)) {
2051 		error = EAGAIN;
2052 		goto end;
2053 	}
2054 
2055 	as_rangelock(as);
2056 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
2057 	if (error != 0) {
2058 		as_rangeunlock(as);
2059 		goto end;
2060 	}
2061 
2062 	vn_a.vp = vp;
2063 	vn_a.offset = off;
2064 	vn_a.type = flags & MAP_TYPE;
2065 	vn_a.prot = prot;
2066 	vn_a.maxprot = maxprot;
2067 	vn_a.cred = cr;
2068 	vn_a.amp = NULL;
2069 	vn_a.flags = flags & ~MAP_TYPE;
2070 	vn_a.szc = 0;
2071 	vn_a.lgrp_mem_policy_flags = 0;
2072 
2073 	error = as_map(as, *addrp, len, segvn_create, (caddr_t)&vn_a);
2074 	as_rangeunlock(as);
2075 
2076 end:
2077 	return (error);
2078 }
2079 
2080 /* ARGSUSED */
2081 static int32_t
2082 udf_addmap(struct vnode *vp,
2083 	offset_t off,
2084 	struct as *as,
2085 	caddr_t addr,
2086 	size_t len,
2087 	uint8_t prot,
2088 	uint8_t maxprot,
2089 	uint32_t flags,
2090 	struct cred *cr,
2091 	caller_context_t *ct)
2092 {
2093 	struct ud_inode *ip = VTOI(vp);
2094 
2095 	ud_printf("udf_addmap\n");
2096 
2097 	if (vp->v_flag & VNOMAP) {
2098 		return (ENOSYS);
2099 	}
2100 
2101 	mutex_enter(&ip->i_tlock);
2102 	ip->i_mapcnt += btopr(len);
2103 	mutex_exit(&ip->i_tlock);
2104 
2105 	return (0);
2106 }
2107 
2108 /* ARGSUSED */
2109 static int32_t
2110 udf_delmap(
2111 	struct vnode *vp, offset_t off,
2112 	struct as *as,
2113 	caddr_t addr,
2114 	size_t len,
2115 	uint32_t prot,
2116 	uint32_t maxprot,
2117 	uint32_t flags,
2118 	struct cred *cr,
2119 	caller_context_t *ct)
2120 {
2121 	struct ud_inode *ip = VTOI(vp);
2122 
2123 	ud_printf("udf_delmap\n");
2124 
2125 	if (vp->v_flag & VNOMAP) {
2126 		return (ENOSYS);
2127 	}
2128 
2129 	mutex_enter(&ip->i_tlock);
2130 	ip->i_mapcnt -= btopr(len); 	/* Count released mappings */
2131 	ASSERT(ip->i_mapcnt >= 0);
2132 	mutex_exit(&ip->i_tlock);
2133 
2134 	return (0);
2135 }
2136 
2137 /* ARGSUSED */
2138 static int32_t
2139 udf_l_pathconf(
2140 	struct vnode *vp,
2141 	int32_t cmd,
2142 	ulong_t *valp,
2143 	struct cred *cr,
2144 	caller_context_t *ct)
2145 {
2146 	int32_t error = 0;
2147 
2148 	ud_printf("udf_l_pathconf\n");
2149 
2150 	if (cmd == _PC_FILESIZEBITS) {
2151 		/*
2152 		 * udf supports 64 bits as file size
2153 		 * but there are several other restrictions
2154 		 * it only supports 32-bit block numbers and
2155 		 * daddr32_t is only and int32_t so taking these
2156 		 * into account we can stay just as where ufs is
2157 		 */
2158 		*valp = 41;
2159 	} else if (cmd == _PC_TIMESTAMP_RESOLUTION) {
2160 		/* nanosecond timestamp resolution */
2161 		*valp = 1L;
2162 	} else {
2163 		error = fs_pathconf(vp, cmd, valp, cr, ct);
2164 	}
2165 
2166 	return (error);
2167 }
2168 
2169 uint32_t ud_pageio_reads = 0, ud_pageio_writes = 0;
2170 #ifndef	__lint
2171 _NOTE(SCHEME_PROTECTS_DATA("safe sharing", ud_pageio_reads))
2172 _NOTE(SCHEME_PROTECTS_DATA("safe sharing", ud_pageio_writes))
2173 #endif
2174 /*
2175  * Assumption is that there will not be a pageio request
2176  * to a enbedded file
2177  */
2178 /* ARGSUSED */
2179 static int32_t
2180 udf_pageio(
2181 	struct vnode *vp,
2182 	struct page *pp,
2183 	u_offset_t io_off,
2184 	size_t io_len,
2185 	int32_t flags,
2186 	struct cred *cr,
2187 	caller_context_t *ct)
2188 {
2189 	daddr_t bn;
2190 	struct buf *bp;
2191 	struct ud_inode *ip = VTOI(vp);
2192 	int32_t dolock, error = 0, contig, multi_io;
2193 	size_t done_len = 0, cur_len = 0;
2194 	page_t *npp = NULL, *opp = NULL, *cpp = pp;
2195 
2196 	if (pp == NULL) {
2197 		return (EINVAL);
2198 	}
2199 
2200 	dolock = (rw_owner(&ip->i_contents) != curthread);
2201 
2202 	/*
2203 	 * We need a better check.  Ideally, we would use another
2204 	 * vnodeops so that hlocked and forcibly unmounted file
2205 	 * systems would return EIO where appropriate and w/o the
2206 	 * need for these checks.
2207 	 */
2208 	if (ip->i_udf == NULL) {
2209 		return (EIO);
2210 	}
2211 
2212 #ifdef	__lock_lint
2213 	rw_enter(&ip->i_contents, RW_READER);
2214 #else
2215 	if (dolock) {
2216 		rw_enter(&ip->i_contents, RW_READER);
2217 	}
2218 #endif
2219 
2220 	/*
2221 	 * Break the io request into chunks, one for each contiguous
2222 	 * stretch of disk blocks in the target file.
2223 	 */
2224 	while (done_len < io_len) {
2225 		ASSERT(cpp);
2226 		bp = NULL;
2227 		contig = 0;
2228 		if (error = ud_bmap_read(ip, (u_offset_t)(io_off + done_len),
2229 		    &bn, &contig)) {
2230 			break;
2231 		}
2232 
2233 		if (bn == UDF_HOLE) {   /* No holey swapfiles */
2234 			cmn_err(CE_WARN, "SWAP file has HOLES");
2235 			error = EINVAL;
2236 			break;
2237 		}
2238 
2239 		cur_len = MIN(io_len - done_len, contig);
2240 
2241 		/*
2242 		 * Check if more than one I/O is
2243 		 * required to complete the given
2244 		 * I/O operation
2245 		 */
2246 		if (ip->i_udf->udf_lbsize < PAGESIZE) {
2247 			if (cur_len >= PAGESIZE) {
2248 				multi_io = 0;
2249 				cur_len &= PAGEMASK;
2250 			} else {
2251 				multi_io = 1;
2252 				cur_len = MIN(io_len - done_len, PAGESIZE);
2253 			}
2254 		}
2255 		page_list_break(&cpp, &npp, btop(cur_len));
2256 
2257 		bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
2258 		ASSERT(bp != NULL);
2259 
2260 		bp->b_edev = ip->i_dev;
2261 		bp->b_dev = cmpdev(ip->i_dev);
2262 		bp->b_blkno = bn;
2263 		bp->b_un.b_addr = (caddr_t)0;
2264 		bp->b_file = vp;
2265 		bp->b_offset = (offset_t)(io_off + done_len);
2266 
2267 /*
2268  *		ub.ub_pageios.value.ul++;
2269  */
2270 		if (multi_io == 0) {
2271 			(void) bdev_strategy(bp);
2272 		} else {
2273 			error = ud_multi_strat(ip, cpp, bp,
2274 			    (u_offset_t)(io_off + done_len));
2275 			if (error != 0) {
2276 				pageio_done(bp);
2277 				break;
2278 			}
2279 		}
2280 		if (flags & B_READ) {
2281 			ud_pageio_reads++;
2282 		} else {
2283 			ud_pageio_writes++;
2284 		}
2285 
2286 		/*
2287 		 * If the request is not B_ASYNC, wait for i/o to complete
2288 		 * and re-assemble the page list to return to the caller.
2289 		 * If it is B_ASYNC we leave the page list in pieces and
2290 		 * cleanup() will dispose of them.
2291 		 */
2292 		if ((flags & B_ASYNC) == 0) {
2293 			error = biowait(bp);
2294 			pageio_done(bp);
2295 			if (error) {
2296 				break;
2297 			}
2298 			page_list_concat(&opp, &cpp);
2299 		}
2300 		cpp = npp;
2301 		npp = NULL;
2302 		done_len += cur_len;
2303 	}
2304 
2305 	ASSERT(error || (cpp == NULL && npp == NULL && done_len == io_len));
2306 	if (error) {
2307 		if (flags & B_ASYNC) {
2308 			/* Cleanup unprocessed parts of list */
2309 			page_list_concat(&cpp, &npp);
2310 			if (flags & B_READ) {
2311 				pvn_read_done(cpp, B_ERROR);
2312 			} else {
2313 				pvn_write_done(cpp, B_ERROR);
2314 			}
2315 		} else {
2316 			/* Re-assemble list and let caller clean up */
2317 			page_list_concat(&opp, &cpp);
2318 			page_list_concat(&opp, &npp);
2319 		}
2320 	}
2321 
2322 #ifdef	__lock_lint
2323 	rw_exit(&ip->i_contents);
2324 #else
2325 	if (dolock) {
2326 		rw_exit(&ip->i_contents);
2327 	}
2328 #endif
2329 	return (error);
2330 }
2331 
2332 
2333 
2334 
2335 /* -------------------- local functions --------------------------- */
2336 
2337 
2338 
2339 int32_t
2340 ud_rdwri(enum uio_rw rw, int32_t ioflag,
2341 	struct ud_inode *ip, caddr_t base, int32_t len,
2342 	offset_t offset, enum uio_seg seg, int32_t *aresid, struct cred *cr)
2343 {
2344 	int32_t error;
2345 	struct uio auio;
2346 	struct iovec aiov;
2347 
2348 	ud_printf("ud_rdwri\n");
2349 
2350 	bzero((caddr_t)&auio, sizeof (uio_t));
2351 	bzero((caddr_t)&aiov, sizeof (iovec_t));
2352 
2353 	aiov.iov_base = base;
2354 	aiov.iov_len = len;
2355 	auio.uio_iov = &aiov;
2356 	auio.uio_iovcnt = 1;
2357 	auio.uio_loffset = offset;
2358 	auio.uio_segflg = (int16_t)seg;
2359 	auio.uio_resid = len;
2360 
2361 	if (rw == UIO_WRITE) {
2362 		auio.uio_fmode = FWRITE;
2363 		auio.uio_extflg = UIO_COPY_DEFAULT;
2364 		auio.uio_llimit = curproc->p_fsz_ctl;
2365 		error = ud_wrip(ip, &auio, ioflag, cr);
2366 	} else {
2367 		auio.uio_fmode = FREAD;
2368 		auio.uio_extflg = UIO_COPY_CACHED;
2369 		auio.uio_llimit = MAXOFFSET_T;
2370 		error = ud_rdip(ip, &auio, ioflag, cr);
2371 	}
2372 
2373 	if (aresid) {
2374 		*aresid = auio.uio_resid;
2375 	} else if (auio.uio_resid) {
2376 		error = EIO;
2377 	}
2378 	return (error);
2379 }
2380 
2381 /*
2382  * Free behind hacks.  The pager is busted.
2383  * XXX - need to pass the information down to writedone() in a flag like B_SEQ
2384  * or B_FREE_IF_TIGHT_ON_MEMORY.
2385  */
2386 int32_t ud_freebehind = 1;
2387 int32_t ud_smallfile = 32 * 1024;
2388 
2389 /* ARGSUSED */
2390 int32_t
2391 ud_getpage_miss(struct vnode *vp, u_offset_t off,
2392 	size_t len, struct seg *seg, caddr_t addr, page_t *pl[],
2393 	size_t plsz, enum seg_rw rw, int32_t seq)
2394 {
2395 	struct ud_inode *ip = VTOI(vp);
2396 	int32_t err = 0;
2397 	size_t io_len;
2398 	u_offset_t io_off;
2399 	u_offset_t pgoff;
2400 	page_t *pp;
2401 
2402 	pl[0] = NULL;
2403 
2404 	/*
2405 	 * Figure out whether the page can be created, or must be
2406 	 * read from the disk
2407 	 */
2408 	if (rw == S_CREATE) {
2409 		if ((pp = page_create_va(vp, off,
2410 		    PAGESIZE, PG_WAIT, seg, addr)) == NULL) {
2411 			cmn_err(CE_WARN, "ud_getpage_miss: page_create");
2412 			return (EINVAL);
2413 		}
2414 		io_len = PAGESIZE;
2415 	} else {
2416 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
2417 		    &io_len, off, PAGESIZE, 0);
2418 
2419 		/*
2420 		 * Some other thread has entered the page.
2421 		 * ud_getpage will retry page_lookup.
2422 		 */
2423 		if (pp == NULL) {
2424 			return (0);
2425 		}
2426 
2427 		/*
2428 		 * Fill the page with as much data as we can from the file.
2429 		 */
2430 		err = ud_page_fill(ip, pp, off, B_READ, &pgoff);
2431 		if (err) {
2432 			pvn_read_done(pp, B_ERROR);
2433 			return (err);
2434 		}
2435 
2436 		/*
2437 		 * XXX ??? ufs has io_len instead of pgoff below
2438 		 */
2439 		ip->i_nextrio = off + ((pgoff + PAGESIZE - 1) & PAGEMASK);
2440 
2441 		/*
2442 		 * If the file access is sequential, initiate read ahead
2443 		 * of the next cluster.
2444 		 */
2445 		if (seq && ip->i_nextrio < ip->i_size) {
2446 			ud_getpage_ra(vp, off, seg, addr);
2447 		}
2448 	}
2449 
2450 outmiss:
2451 	pvn_plist_init(pp, pl, plsz, (offset_t)off, io_len, rw);
2452 	return (err);
2453 }
2454 
2455 /* ARGSUSED */
2456 void
2457 ud_getpage_ra(struct vnode *vp,
2458 	u_offset_t off, struct seg *seg, caddr_t addr)
2459 {
2460 	page_t *pp;
2461 	size_t io_len;
2462 	struct ud_inode *ip = VTOI(vp);
2463 	u_offset_t io_off = ip->i_nextrio, pgoff;
2464 	caddr_t addr2 = addr + (io_off - off);
2465 	daddr_t bn;
2466 	int32_t contig = 0;
2467 
2468 	/*
2469 	 * Is this test needed?
2470 	 */
2471 
2472 	if (addr2 >= seg->s_base + seg->s_size) {
2473 		return;
2474 	}
2475 
2476 	contig = 0;
2477 	if (ud_bmap_read(ip, io_off, &bn, &contig) != 0 || bn == UDF_HOLE) {
2478 		return;
2479 	}
2480 
2481 	pp = pvn_read_kluster(vp, io_off, seg, addr2,
2482 	    &io_off, &io_len, io_off, PAGESIZE, 1);
2483 
2484 	/*
2485 	 * Some other thread has entered the page.
2486 	 * So no read head done here (ie we will have to and wait
2487 	 * for the read when needed).
2488 	 */
2489 
2490 	if (pp == NULL) {
2491 		return;
2492 	}
2493 
2494 	(void) ud_page_fill(ip, pp, io_off, (B_READ|B_ASYNC), &pgoff);
2495 	ip->i_nextrio =  io_off + ((pgoff + PAGESIZE - 1) & PAGEMASK);
2496 }
2497 
2498 int
2499 ud_page_fill(struct ud_inode *ip, page_t *pp, u_offset_t off,
2500 	uint32_t bflgs, u_offset_t *pg_off)
2501 {
2502 	daddr_t bn;
2503 	struct buf *bp;
2504 	caddr_t kaddr, caddr;
2505 	int32_t error = 0, contig = 0, multi_io = 0;
2506 	int32_t lbsize = ip->i_udf->udf_lbsize;
2507 	int32_t lbmask = ip->i_udf->udf_lbmask;
2508 	uint64_t isize;
2509 
2510 	isize = (ip->i_size + lbmask) & (~lbmask);
2511 	if (ip->i_desc_type == ICB_FLAG_ONE_AD) {
2512 
2513 		/*
2514 		 * Embedded file read file_entry
2515 		 * from buffer cache and copy the required
2516 		 * portions
2517 		 */
2518 		bp = ud_bread(ip->i_dev,
2519 		    ip->i_icb_lbano << ip->i_udf->udf_l2d_shift, lbsize);
2520 		if ((bp->b_error == 0) &&
2521 		    (bp->b_resid == 0)) {
2522 
2523 			caddr = bp->b_un.b_addr + ip->i_data_off;
2524 
2525 			/*
2526 			 * mapin to kvm
2527 			 */
2528 			kaddr = (caddr_t)ppmapin(pp,
2529 			    PROT_READ | PROT_WRITE, (caddr_t)-1);
2530 			(void) kcopy(caddr, kaddr, ip->i_size);
2531 
2532 			/*
2533 			 * mapout of kvm
2534 			 */
2535 			ppmapout(kaddr);
2536 		}
2537 		brelse(bp);
2538 		contig = ip->i_size;
2539 	} else {
2540 
2541 		/*
2542 		 * Get the continuous size and block number
2543 		 * at offset "off"
2544 		 */
2545 		if (error = ud_bmap_read(ip, off, &bn, &contig))
2546 			goto out;
2547 		contig = MIN(contig, PAGESIZE);
2548 		contig = (contig + lbmask) & (~lbmask);
2549 
2550 		/*
2551 		 * Zero part of the page which we are not
2552 		 * going to read from the disk.
2553 		 */
2554 
2555 		if (bn == UDF_HOLE) {
2556 
2557 			/*
2558 			 * This is a HOLE. Just zero out
2559 			 * the page
2560 			 */
2561 			if (((off + contig) == isize) ||
2562 			    (contig == PAGESIZE)) {
2563 				pagezero(pp->p_prev, 0, PAGESIZE);
2564 				goto out;
2565 			}
2566 		}
2567 
2568 		if (contig < PAGESIZE) {
2569 			uint64_t count;
2570 
2571 			count = isize - off;
2572 			if (contig != count) {
2573 				multi_io = 1;
2574 				contig = (int32_t)(MIN(count, PAGESIZE));
2575 			} else {
2576 				pagezero(pp->p_prev, contig, PAGESIZE - contig);
2577 			}
2578 		}
2579 
2580 		/*
2581 		 * Get a bp and initialize it
2582 		 */
2583 		bp = pageio_setup(pp, contig, ip->i_devvp, bflgs);
2584 		ASSERT(bp != NULL);
2585 
2586 		bp->b_edev = ip->i_dev;
2587 		bp->b_dev = cmpdev(ip->i_dev);
2588 		bp->b_blkno = bn;
2589 		bp->b_un.b_addr = 0;
2590 		bp->b_file = ip->i_vnode;
2591 
2592 		/*
2593 		 * Start I/O
2594 		 */
2595 		if (multi_io == 0) {
2596 
2597 			/*
2598 			 * Single I/O is sufficient for this page
2599 			 */
2600 			(void) bdev_strategy(bp);
2601 		} else {
2602 
2603 			/*
2604 			 * We need to do the I/O in
2605 			 * piece's
2606 			 */
2607 			error = ud_multi_strat(ip, pp, bp, off);
2608 			if (error != 0) {
2609 				goto out;
2610 			}
2611 		}
2612 		if ((bflgs & B_ASYNC) == 0) {
2613 
2614 			/*
2615 			 * Wait for i/o to complete.
2616 			 */
2617 
2618 			error = biowait(bp);
2619 			pageio_done(bp);
2620 			if (error) {
2621 				goto out;
2622 			}
2623 		}
2624 	}
2625 	if ((off + contig) >= ip->i_size) {
2626 		contig = ip->i_size - off;
2627 	}
2628 
2629 out:
2630 	*pg_off = contig;
2631 	return (error);
2632 }
2633 
2634 int32_t
2635 ud_putpages(struct vnode *vp, offset_t off,
2636 	size_t len, int32_t flags, struct cred *cr)
2637 {
2638 	struct ud_inode *ip;
2639 	page_t *pp;
2640 	u_offset_t io_off;
2641 	size_t io_len;
2642 	u_offset_t eoff;
2643 	int32_t err = 0;
2644 	int32_t dolock;
2645 
2646 	ud_printf("ud_putpages\n");
2647 
2648 	if (vp->v_count == 0) {
2649 		cmn_err(CE_WARN, "ud_putpages: bad v_count");
2650 		return (EINVAL);
2651 	}
2652 
2653 	ip = VTOI(vp);
2654 
2655 	/*
2656 	 * Acquire the readers/write inode lock before locking
2657 	 * any pages in this inode.
2658 	 * The inode lock is held during i/o.
2659 	 */
2660 	if (len == 0) {
2661 		mutex_enter(&ip->i_tlock);
2662 		ip->i_delayoff = ip->i_delaylen = 0;
2663 		mutex_exit(&ip->i_tlock);
2664 	}
2665 #ifdef	__lock_lint
2666 	rw_enter(&ip->i_contents, RW_READER);
2667 #else
2668 	dolock = (rw_owner(&ip->i_contents) != curthread);
2669 	if (dolock) {
2670 		rw_enter(&ip->i_contents, RW_READER);
2671 	}
2672 #endif
2673 
2674 	if (!vn_has_cached_data(vp)) {
2675 #ifdef	__lock_lint
2676 		rw_exit(&ip->i_contents);
2677 #else
2678 		if (dolock) {
2679 			rw_exit(&ip->i_contents);
2680 		}
2681 #endif
2682 		return (0);
2683 	}
2684 
2685 	if (len == 0) {
2686 		/*
2687 		 * Search the entire vp list for pages >= off.
2688 		 */
2689 		err = pvn_vplist_dirty(vp, (u_offset_t)off, ud_putapage,
2690 		    flags, cr);
2691 	} else {
2692 		/*
2693 		 * Loop over all offsets in the range looking for
2694 		 * pages to deal with.
2695 		 */
2696 		if ((eoff = blkroundup(ip->i_udf, ip->i_size)) != 0) {
2697 			eoff = MIN(off + len, eoff);
2698 		} else {
2699 			eoff = off + len;
2700 		}
2701 
2702 		for (io_off = off; io_off < eoff; io_off += io_len) {
2703 			/*
2704 			 * If we are not invalidating, synchronously
2705 			 * freeing or writing pages, use the routine
2706 			 * page_lookup_nowait() to prevent reclaiming
2707 			 * them from the free list.
2708 			 */
2709 			if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2710 				pp = page_lookup(vp, io_off,
2711 				    (flags & (B_INVAL | B_FREE)) ?
2712 				    SE_EXCL : SE_SHARED);
2713 			} else {
2714 				pp = page_lookup_nowait(vp, io_off,
2715 				    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2716 			}
2717 
2718 			if (pp == NULL || pvn_getdirty(pp, flags) == 0) {
2719 				io_len = PAGESIZE;
2720 			} else {
2721 
2722 				err = ud_putapage(vp, pp,
2723 				    &io_off, &io_len, flags, cr);
2724 				if (err != 0) {
2725 					break;
2726 				}
2727 				/*
2728 				 * "io_off" and "io_len" are returned as
2729 				 * the range of pages we actually wrote.
2730 				 * This allows us to skip ahead more quickly
2731 				 * since several pages may've been dealt
2732 				 * with by this iteration of the loop.
2733 				 */
2734 			}
2735 		}
2736 	}
2737 	if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
2738 		/*
2739 		 * We have just sync'ed back all the pages on
2740 		 * the inode, turn off the IMODTIME flag.
2741 		 */
2742 		mutex_enter(&ip->i_tlock);
2743 		ip->i_flag &= ~IMODTIME;
2744 		mutex_exit(&ip->i_tlock);
2745 	}
2746 #ifdef	__lock_lint
2747 	rw_exit(&ip->i_contents);
2748 #else
2749 	if (dolock) {
2750 		rw_exit(&ip->i_contents);
2751 	}
2752 #endif
2753 	return (err);
2754 }
2755 
2756 /* ARGSUSED */
2757 int32_t
2758 ud_putapage(struct vnode *vp,
2759 	page_t *pp, u_offset_t *offp,
2760 	size_t *lenp, int32_t flags, struct cred *cr)
2761 {
2762 	daddr_t bn;
2763 	size_t io_len;
2764 	struct ud_inode *ip;
2765 	int32_t error = 0, contig, multi_io = 0;
2766 	struct udf_vfs *udf_vfsp;
2767 	u_offset_t off, io_off;
2768 	caddr_t kaddr, caddr;
2769 	struct buf *bp = NULL;
2770 	int32_t lbmask;
2771 	uint64_t isize;
2772 	uint16_t crc_len;
2773 	struct file_entry *fe;
2774 
2775 	ud_printf("ud_putapage\n");
2776 
2777 	ip = VTOI(vp);
2778 	ASSERT(ip);
2779 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
2780 	lbmask = ip->i_udf->udf_lbmask;
2781 	isize = (ip->i_size + lbmask) & (~lbmask);
2782 
2783 	udf_vfsp = ip->i_udf;
2784 	ASSERT(udf_vfsp->udf_flags & UDF_FL_RW);
2785 
2786 	/*
2787 	 * If the modified time on the inode has not already been
2788 	 * set elsewhere (e.g. for write/setattr) we set the time now.
2789 	 * This gives us approximate modified times for mmap'ed files
2790 	 * which are modified via stores in the user address space.
2791 	 */
2792 	if (((ip->i_flag & IMODTIME) == 0) || (flags & B_FORCE)) {
2793 		mutex_enter(&ip->i_tlock);
2794 		ip->i_flag |= IUPD;
2795 		ITIMES_NOLOCK(ip);
2796 		mutex_exit(&ip->i_tlock);
2797 	}
2798 
2799 
2800 	/*
2801 	 * Align the request to a block boundry (for old file systems),
2802 	 * and go ask bmap() how contiguous things are for this file.
2803 	 */
2804 	off = pp->p_offset & ~(offset_t)lbmask;
2805 				/* block align it */
2806 
2807 
2808 	if (ip->i_desc_type == ICB_FLAG_ONE_AD) {
2809 		ASSERT(ip->i_size <= ip->i_max_emb);
2810 
2811 		pp = pvn_write_kluster(vp, pp, &io_off,
2812 		    &io_len, off, PAGESIZE, flags);
2813 		if (io_len == 0) {
2814 			io_len = PAGESIZE;
2815 		}
2816 
2817 		bp = ud_bread(ip->i_dev,
2818 		    ip->i_icb_lbano << udf_vfsp->udf_l2d_shift,
2819 		    udf_vfsp->udf_lbsize);
2820 		fe = (struct file_entry *)bp->b_un.b_addr;
2821 		if ((bp->b_flags & B_ERROR) ||
2822 		    (ud_verify_tag_and_desc(&fe->fe_tag, UD_FILE_ENTRY,
2823 		    ip->i_icb_block,
2824 		    1, udf_vfsp->udf_lbsize) != 0)) {
2825 			if (pp != NULL)
2826 				pvn_write_done(pp, B_ERROR | B_WRITE | flags);
2827 			if (bp->b_flags & B_ERROR) {
2828 				error = EIO;
2829 			} else {
2830 				error = EINVAL;
2831 			}
2832 			brelse(bp);
2833 			return (error);
2834 		}
2835 		if ((bp->b_error == 0) &&
2836 		    (bp->b_resid == 0)) {
2837 
2838 			caddr = bp->b_un.b_addr + ip->i_data_off;
2839 			kaddr = (caddr_t)ppmapin(pp,
2840 			    PROT_READ | PROT_WRITE, (caddr_t)-1);
2841 			(void) kcopy(kaddr, caddr, ip->i_size);
2842 			ppmapout(kaddr);
2843 		}
2844 		crc_len = offsetof(struct file_entry, fe_spec) +
2845 		    SWAP_32(fe->fe_len_ear);
2846 		crc_len += ip->i_size;
2847 		ud_make_tag(ip->i_udf, &fe->fe_tag,
2848 		    UD_FILE_ENTRY, ip->i_icb_block, crc_len);
2849 
2850 		bwrite(bp);
2851 
2852 		if (flags & B_ASYNC) {
2853 			pvn_write_done(pp, flags);
2854 		}
2855 		contig = ip->i_size;
2856 	} else {
2857 
2858 		if (error = ud_bmap_read(ip, off, &bn, &contig)) {
2859 			goto out;
2860 		}
2861 		contig = MIN(contig, PAGESIZE);
2862 		contig = (contig + lbmask) & (~lbmask);
2863 
2864 		if (contig < PAGESIZE) {
2865 			uint64_t count;
2866 
2867 			count = isize - off;
2868 			if (contig != count) {
2869 				multi_io = 1;
2870 				contig = (int32_t)(MIN(count, PAGESIZE));
2871 			}
2872 		}
2873 
2874 		if ((off + contig) > isize) {
2875 			contig = isize - off;
2876 		}
2877 
2878 		if (contig > PAGESIZE) {
2879 			if (contig & PAGEOFFSET) {
2880 				contig &= PAGEMASK;
2881 			}
2882 		}
2883 
2884 		pp = pvn_write_kluster(vp, pp, &io_off,
2885 		    &io_len, off, contig, flags);
2886 		if (io_len == 0) {
2887 			io_len = PAGESIZE;
2888 		}
2889 
2890 		bp = pageio_setup(pp, contig, ip->i_devvp, B_WRITE | flags);
2891 		ASSERT(bp != NULL);
2892 
2893 		bp->b_edev = ip->i_dev;
2894 		bp->b_dev = cmpdev(ip->i_dev);
2895 		bp->b_blkno = bn;
2896 		bp->b_un.b_addr = 0;
2897 		bp->b_file = vp;
2898 		bp->b_offset = (offset_t)off;
2899 
2900 
2901 		/*
2902 		 * write throttle
2903 		 */
2904 		ASSERT(bp->b_iodone == NULL);
2905 		bp->b_iodone = ud_iodone;
2906 		mutex_enter(&ip->i_tlock);
2907 		ip->i_writes += bp->b_bcount;
2908 		mutex_exit(&ip->i_tlock);
2909 
2910 		if (multi_io == 0) {
2911 
2912 			(void) bdev_strategy(bp);
2913 		} else {
2914 			error = ud_multi_strat(ip, pp, bp, off);
2915 			if (error != 0) {
2916 				goto out;
2917 			}
2918 		}
2919 
2920 		if ((flags & B_ASYNC) == 0) {
2921 			/*
2922 			 * Wait for i/o to complete.
2923 			 */
2924 			error = biowait(bp);
2925 			pageio_done(bp);
2926 		}
2927 	}
2928 
2929 	if ((flags & B_ASYNC) == 0) {
2930 		pvn_write_done(pp, ((error) ? B_ERROR : 0) | B_WRITE | flags);
2931 	}
2932 
2933 	pp = NULL;
2934 
2935 out:
2936 	if (error != 0 && pp != NULL) {
2937 		pvn_write_done(pp, B_ERROR | B_WRITE | flags);
2938 	}
2939 
2940 	if (offp) {
2941 		*offp = io_off;
2942 	}
2943 	if (lenp) {
2944 		*lenp = io_len;
2945 	}
2946 
2947 	return (error);
2948 }
2949 
2950 
2951 int32_t
2952 ud_iodone(struct buf *bp)
2953 {
2954 	struct ud_inode *ip;
2955 
2956 	ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
2957 
2958 	bp->b_iodone = NULL;
2959 
2960 	ip = VTOI(bp->b_pages->p_vnode);
2961 
2962 	mutex_enter(&ip->i_tlock);
2963 	if (ip->i_writes >= ud_LW) {
2964 		if ((ip->i_writes -= bp->b_bcount) <= ud_LW) {
2965 			if (ud_WRITES) {
2966 				cv_broadcast(&ip->i_wrcv); /* wake all up */
2967 			}
2968 		}
2969 	} else {
2970 		ip->i_writes -= bp->b_bcount;
2971 	}
2972 	mutex_exit(&ip->i_tlock);
2973 	iodone(bp);
2974 	return (0);
2975 }
2976 
2977 /* ARGSUSED3 */
2978 int32_t
2979 ud_rdip(struct ud_inode *ip, struct uio *uio, int32_t ioflag, cred_t *cr)
2980 {
2981 	struct vnode *vp;
2982 	struct udf_vfs *udf_vfsp;
2983 	krw_t rwtype;
2984 	caddr_t base;
2985 	uint32_t flags;
2986 	int32_t error, n, on, mapon, dofree;
2987 	u_offset_t off;
2988 	long oresid = uio->uio_resid;
2989 
2990 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
2991 	if ((ip->i_type != VREG) &&
2992 	    (ip->i_type != VDIR) &&
2993 	    (ip->i_type != VLNK)) {
2994 		return (EIO);
2995 	}
2996 
2997 	if (uio->uio_loffset > MAXOFFSET_T) {
2998 		return (0);
2999 	}
3000 
3001 	if ((uio->uio_loffset < (offset_t)0) ||
3002 	    ((uio->uio_loffset + uio->uio_resid) < 0)) {
3003 		return (EINVAL);
3004 	}
3005 	if (uio->uio_resid == 0) {
3006 		return (0);
3007 	}
3008 
3009 	vp = ITOV(ip);
3010 	udf_vfsp = ip->i_udf;
3011 	mutex_enter(&ip->i_tlock);
3012 	ip->i_flag |= IACC;
3013 	mutex_exit(&ip->i_tlock);
3014 
3015 	rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER);
3016 
3017 	do {
3018 		offset_t diff;
3019 		u_offset_t uoff = uio->uio_loffset;
3020 		off = uoff & (offset_t)MAXBMASK;
3021 		mapon = (int)(uoff & (offset_t)MAXBOFFSET);
3022 		on = (int)blkoff(udf_vfsp, uoff);
3023 		n = (int)MIN(udf_vfsp->udf_lbsize - on, uio->uio_resid);
3024 
3025 		diff = ip->i_size - uoff;
3026 
3027 		if (diff <= (offset_t)0) {
3028 			error = 0;
3029 			goto out;
3030 		}
3031 		if (diff < (offset_t)n) {
3032 			n = (int)diff;
3033 		}
3034 		dofree = ud_freebehind &&
3035 		    ip->i_nextr == (off & PAGEMASK) &&
3036 		    off > ud_smallfile;
3037 
3038 #ifndef	__lock_lint
3039 		if (rwtype == RW_READER) {
3040 			rw_exit(&ip->i_contents);
3041 		}
3042 #endif
3043 
3044 		base = segmap_getmapflt(segkmap, vp, (off + mapon),
3045 		    (uint32_t)n, 1, S_READ);
3046 		error = uiomove(base + mapon, (long)n, UIO_READ, uio);
3047 
3048 		flags = 0;
3049 		if (!error) {
3050 			/*
3051 			 * If read a whole block, or read to eof,
3052 			 * won't need this buffer again soon.
3053 			 */
3054 			if (n + on == MAXBSIZE && ud_freebehind && dofree &&
3055 			    freemem < lotsfree + pages_before_pager) {
3056 				flags = SM_FREE | SM_DONTNEED |SM_ASYNC;
3057 			}
3058 			/*
3059 			 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
3060 			 * we want to make sure that the page which has
3061 			 * been read, is written on disk if it is dirty.
3062 			 * And corresponding indirect blocks should also
3063 			 * be flushed out.
3064 			 */
3065 			if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) {
3066 				flags &= ~SM_ASYNC;
3067 				flags |= SM_WRITE;
3068 			}
3069 			error = segmap_release(segkmap, base, flags);
3070 		} else    {
3071 			(void) segmap_release(segkmap, base, flags);
3072 		}
3073 
3074 #ifndef __lock_lint
3075 		if (rwtype == RW_READER) {
3076 			rw_enter(&ip->i_contents, rwtype);
3077 		}
3078 #endif
3079 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
3080 out:
3081 	/*
3082 	 * Inode is updated according to this table if FRSYNC is set.
3083 	 *
3084 	 *	FSYNC	FDSYNC(posix.4)
3085 	 *	--------------------------
3086 	 *	always	IATTCHG|IBDWRITE
3087 	 */
3088 	if (ioflag & FRSYNC) {
3089 		if ((ioflag & FSYNC) ||
3090 		    ((ioflag & FDSYNC) &&
3091 		    (ip->i_flag & (IATTCHG|IBDWRITE)))) {
3092 		rw_exit(&ip->i_contents);
3093 		rw_enter(&ip->i_contents, RW_WRITER);
3094 		ud_iupdat(ip, 1);
3095 		}
3096 	}
3097 	/*
3098 	 * If we've already done a partial read, terminate
3099 	 * the read but return no error.
3100 	 */
3101 	if (oresid != uio->uio_resid) {
3102 		error = 0;
3103 	}
3104 	ITIMES(ip);
3105 
3106 	return (error);
3107 }
3108 
3109 int32_t
3110 ud_wrip(struct ud_inode *ip, struct uio *uio, int ioflag, struct cred *cr)
3111 {
3112 	caddr_t base;
3113 	struct vnode *vp;
3114 	struct udf_vfs *udf_vfsp;
3115 	uint32_t flags;
3116 	int32_t error = 0, iupdat_flag, n, on, mapon, i_size_changed = 0;
3117 	int32_t pagecreate, newpage;
3118 	uint64_t old_i_size;
3119 	u_offset_t off;
3120 	long start_resid = uio->uio_resid, premove_resid;
3121 	rlim64_t limit = uio->uio_limit;
3122 
3123 
3124 	ASSERT(RW_WRITE_HELD(&ip->i_contents));
3125 	if ((ip->i_type != VREG) &&
3126 	    (ip->i_type != VDIR) &&
3127 	    (ip->i_type != VLNK)) {
3128 		return (EIO);
3129 	}
3130 
3131 	if (uio->uio_loffset >= MAXOFFSET_T) {
3132 		return (EFBIG);
3133 	}
3134 	/*
3135 	 * see udf_l_pathconf
3136 	 */
3137 	if (limit > (((uint64_t)1 << 40) - 1)) {
3138 		limit = ((uint64_t)1 << 40) - 1;
3139 	}
3140 	if (uio->uio_loffset >= limit) {
3141 		proc_t *p = ttoproc(curthread);
3142 
3143 		mutex_enter(&p->p_lock);
3144 		(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
3145 		    p, RCA_UNSAFE_SIGINFO);
3146 		mutex_exit(&p->p_lock);
3147 		return (EFBIG);
3148 	}
3149 	if ((uio->uio_loffset < (offset_t)0) ||
3150 	    ((uio->uio_loffset + uio->uio_resid) < 0)) {
3151 		return (EINVAL);
3152 	}
3153 	if (uio->uio_resid == 0) {
3154 		return (0);
3155 	}
3156 
3157 	mutex_enter(&ip->i_tlock);
3158 	ip->i_flag |= INOACC;
3159 
3160 	if (ioflag & (FSYNC | FDSYNC)) {
3161 		ip->i_flag |= ISYNC;
3162 		iupdat_flag = 1;
3163 	}
3164 	mutex_exit(&ip->i_tlock);
3165 
3166 	udf_vfsp = ip->i_udf;
3167 	vp = ITOV(ip);
3168 
3169 	do {
3170 		u_offset_t uoff = uio->uio_loffset;
3171 		off = uoff & (offset_t)MAXBMASK;
3172 		mapon = (int)(uoff & (offset_t)MAXBOFFSET);
3173 		on = (int)blkoff(udf_vfsp, uoff);
3174 		n = (int)MIN(udf_vfsp->udf_lbsize - on, uio->uio_resid);
3175 
3176 		if (ip->i_type == VREG && uoff + n >= limit) {
3177 			if (uoff >= limit) {
3178 				error = EFBIG;
3179 				goto out;
3180 			}
3181 			n = (int)(limit - (rlim64_t)uoff);
3182 		}
3183 		if (uoff + n > ip->i_size) {
3184 			/*
3185 			 * We are extending the length of the file.
3186 			 * bmap is used so that we are sure that
3187 			 * if we need to allocate new blocks, that it
3188 			 * is done here before we up the file size.
3189 			 */
3190 			error = ud_bmap_write(ip, uoff,
3191 			    (int)(on + n), mapon == 0, cr);
3192 			if (error) {
3193 				break;
3194 			}
3195 			i_size_changed = 1;
3196 			old_i_size = ip->i_size;
3197 			ip->i_size = uoff + n;
3198 			/*
3199 			 * If we are writing from the beginning of
3200 			 * the mapping, we can just create the
3201 			 * pages without having to read them.
3202 			 */
3203 			pagecreate = (mapon == 0);
3204 		} else if (n == MAXBSIZE) {
3205 			/*
3206 			 * Going to do a whole mappings worth,
3207 			 * so we can just create the pages w/o
3208 			 * having to read them in.  But before
3209 			 * we do that, we need to make sure any
3210 			 * needed blocks are allocated first.
3211 			 */
3212 			error = ud_bmap_write(ip, uoff,
3213 			    (int)(on + n), 1, cr);
3214 			if (error) {
3215 				break;
3216 			}
3217 			pagecreate = 1;
3218 		} else {
3219 			pagecreate = 0;
3220 		}
3221 
3222 		rw_exit(&ip->i_contents);
3223 
3224 		/*
3225 		 * Touch the page and fault it in if it is not in
3226 		 * core before segmap_getmapflt can lock it. This
3227 		 * is to avoid the deadlock if the buffer is mapped
3228 		 * to the same file through mmap which we want to
3229 		 * write to.
3230 		 */
3231 		uio_prefaultpages((long)n, uio);
3232 
3233 		base = segmap_getmapflt(segkmap, vp, (off + mapon),
3234 		    (uint32_t)n, !pagecreate, S_WRITE);
3235 
3236 		/*
3237 		 * segmap_pagecreate() returns 1 if it calls
3238 		 * page_create_va() to allocate any pages.
3239 		 */
3240 		newpage = 0;
3241 		if (pagecreate) {
3242 			newpage = segmap_pagecreate(segkmap, base,
3243 			    (size_t)n, 0);
3244 		}
3245 
3246 		premove_resid = uio->uio_resid;
3247 		error = uiomove(base + mapon, (long)n, UIO_WRITE, uio);
3248 
3249 		if (pagecreate &&
3250 		    uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) {
3251 			/*
3252 			 * We created pages w/o initializing them completely,
3253 			 * thus we need to zero the part that wasn't set up.
3254 			 * This happens on most EOF write cases and if
3255 			 * we had some sort of error during the uiomove.
3256 			 */
3257 			int nzero, nmoved;
3258 
3259 			nmoved = (int)(uio->uio_loffset - (off + mapon));
3260 			ASSERT(nmoved >= 0 && nmoved <= n);
3261 			nzero = roundup(on + n, PAGESIZE) - nmoved;
3262 			ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE);
3263 			(void) kzero(base + mapon + nmoved, (uint32_t)nzero);
3264 		}
3265 
3266 		/*
3267 		 * Unlock the pages allocated by page_create_va()
3268 		 * in segmap_pagecreate()
3269 		 */
3270 		if (newpage) {
3271 			segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE);
3272 		}
3273 
3274 		if (error) {
3275 			/*
3276 			 * If we failed on a write, we may have already
3277 			 * allocated file blocks as well as pages.  It's
3278 			 * hard to undo the block allocation, but we must
3279 			 * be sure to invalidate any pages that may have
3280 			 * been allocated.
3281 			 */
3282 			(void) segmap_release(segkmap, base, SM_INVAL);
3283 		} else {
3284 			flags = 0;
3285 			/*
3286 			 * Force write back for synchronous write cases.
3287 			 */
3288 			if ((ioflag & (FSYNC|FDSYNC)) || ip->i_type == VDIR) {
3289 				/*
3290 				 * If the sticky bit is set but the
3291 				 * execute bit is not set, we do a
3292 				 * synchronous write back and free
3293 				 * the page when done.  We set up swap
3294 				 * files to be handled this way to
3295 				 * prevent servers from keeping around
3296 				 * the client's swap pages too long.
3297 				 * XXX - there ought to be a better way.
3298 				 */
3299 				if (IS_SWAPVP(vp)) {
3300 					flags = SM_WRITE | SM_FREE |
3301 					    SM_DONTNEED;
3302 					iupdat_flag = 0;
3303 				} else {
3304 					flags = SM_WRITE;
3305 				}
3306 			} else if (((mapon + n) == MAXBSIZE) ||
3307 			    IS_SWAPVP(vp)) {
3308 				/*
3309 				 * Have written a whole block.
3310 				 * Start an asynchronous write and
3311 				 * mark the buffer to indicate that
3312 				 * it won't be needed again soon.
3313 				 */
3314 				flags = SM_WRITE |SM_ASYNC | SM_DONTNEED;
3315 			}
3316 			error = segmap_release(segkmap, base, flags);
3317 
3318 			/*
3319 			 * If the operation failed and is synchronous,
3320 			 * then we need to unwind what uiomove() last
3321 			 * did so we can potentially return an error to
3322 			 * the caller.  If this write operation was
3323 			 * done in two pieces and the first succeeded,
3324 			 * then we won't return an error for the second
3325 			 * piece that failed.  However, we only want to
3326 			 * return a resid value that reflects what was
3327 			 * really done.
3328 			 *
3329 			 * Failures for non-synchronous operations can
3330 			 * be ignored since the page subsystem will
3331 			 * retry the operation until it succeeds or the
3332 			 * file system is unmounted.
3333 			 */
3334 			if (error) {
3335 				if ((ioflag & (FSYNC | FDSYNC)) ||
3336 				    ip->i_type == VDIR) {
3337 					uio->uio_resid = premove_resid;
3338 				} else {
3339 					error = 0;
3340 				}
3341 			}
3342 		}
3343 
3344 		/*
3345 		 * Re-acquire contents lock.
3346 		 */
3347 		rw_enter(&ip->i_contents, RW_WRITER);
3348 		/*
3349 		 * If the uiomove() failed or if a synchronous
3350 		 * page push failed, fix up i_size.
3351 		 */
3352 		if (error) {
3353 			if (i_size_changed) {
3354 				/*
3355 				 * The uiomove failed, and we
3356 				 * allocated blocks,so get rid
3357 				 * of them.
3358 				 */
3359 				(void) ud_itrunc(ip, old_i_size, 0, cr);
3360 			}
3361 		} else {
3362 			/*
3363 			 * XXX - Can this be out of the loop?
3364 			 */
3365 			ip->i_flag |= IUPD | ICHG;
3366 			if (i_size_changed) {
3367 				ip->i_flag |= IATTCHG;
3368 			}
3369 			if ((ip->i_perm & (IEXEC | (IEXEC >> 5) |
3370 			    (IEXEC >> 10))) != 0 &&
3371 			    (ip->i_char & (ISUID | ISGID)) != 0 &&
3372 			    secpolicy_vnode_setid_retain(cr,
3373 			    (ip->i_char & ISUID) != 0 && ip->i_uid == 0) != 0) {
3374 				/*
3375 				 * Clear Set-UID & Set-GID bits on
3376 				 * successful write if not privileged
3377 				 * and at least one of the execute bits
3378 				 * is set.  If we always clear Set-GID,
3379 				 * mandatory file and record locking is
3380 				 * unuseable.
3381 				 */
3382 				ip->i_char &= ~(ISUID | ISGID);
3383 			}
3384 		}
3385 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
3386 
3387 out:
3388 	/*
3389 	 * Inode is updated according to this table -
3390 	 *
3391 	 *	FSYNC	FDSYNC(posix.4)
3392 	 *	--------------------------
3393 	 *	always@	IATTCHG|IBDWRITE
3394 	 *
3395 	 * @ -  If we are doing synchronous write the only time we should
3396 	 *	not be sync'ing the ip here is if we have the stickyhack
3397 	 *	activated, the file is marked with the sticky bit and
3398 	 *	no exec bit, the file length has not been changed and
3399 	 *	no new blocks have been allocated during this write.
3400 	 */
3401 	if ((ip->i_flag & ISYNC) != 0) {
3402 		/*
3403 		 * we have eliminated nosync
3404 		 */
3405 		if ((ip->i_flag & (IATTCHG|IBDWRITE)) ||
3406 		    ((ioflag & FSYNC) && iupdat_flag)) {
3407 			ud_iupdat(ip, 1);
3408 		}
3409 	}
3410 
3411 	/*
3412 	 * If we've already done a partial-write, terminate
3413 	 * the write but return no error.
3414 	 */
3415 	if (start_resid != uio->uio_resid) {
3416 		error = 0;
3417 	}
3418 	ip->i_flag &= ~(INOACC | ISYNC);
3419 	ITIMES_NOLOCK(ip);
3420 
3421 	return (error);
3422 }
3423 
3424 int32_t
3425 ud_multi_strat(struct ud_inode *ip,
3426 	page_t *pp, struct buf *bp, u_offset_t start)
3427 {
3428 	daddr_t bn;
3429 	int32_t error = 0, io_count, contig, alloc_sz, i;
3430 	uint32_t io_off;
3431 	mio_master_t *mm = NULL;
3432 	mio_slave_t *ms = NULL;
3433 	struct buf *rbp;
3434 
3435 	ASSERT(!(start & PAGEOFFSET));
3436 
3437 	/*
3438 	 * Figure out how many buffers to allocate
3439 	 */
3440 	io_count = 0;
3441 	for (io_off = 0; io_off < bp->b_bcount; io_off += contig) {
3442 		contig = 0;
3443 		if (error = ud_bmap_read(ip, (u_offset_t)(start + io_off),
3444 		    &bn, &contig)) {
3445 			goto end;
3446 		}
3447 		if (contig == 0) {
3448 			goto end;
3449 		}
3450 		contig = MIN(contig, PAGESIZE - io_off);
3451 		if (bn != UDF_HOLE) {
3452 			io_count ++;
3453 		} else {
3454 			/*
3455 			 * HOLE
3456 			 */
3457 			if (bp->b_flags & B_READ) {
3458 
3459 				/*
3460 				 * This is a hole and is read
3461 				 * it should be filled with 0's
3462 				 */
3463 				pagezero(pp, io_off, contig);
3464 			}
3465 		}
3466 	}
3467 
3468 
3469 	if (io_count != 0) {
3470 
3471 		/*
3472 		 * Allocate memory for all the
3473 		 * required number of buffers
3474 		 */
3475 		alloc_sz = sizeof (mio_master_t) +
3476 		    (sizeof (mio_slave_t) * io_count);
3477 		mm = (mio_master_t *)kmem_zalloc(alloc_sz, KM_SLEEP);
3478 		if (mm == NULL) {
3479 			error = ENOMEM;
3480 			goto end;
3481 		}
3482 
3483 		/*
3484 		 * initialize master
3485 		 */
3486 		mutex_init(&mm->mm_mutex, NULL, MUTEX_DEFAULT, NULL);
3487 		mm->mm_size = alloc_sz;
3488 		mm->mm_bp = bp;
3489 		mm->mm_resid = 0;
3490 		mm->mm_error = 0;
3491 		mm->mm_index = master_index++;
3492 
3493 		ms = (mio_slave_t *)(((caddr_t)mm) + sizeof (mio_master_t));
3494 
3495 		/*
3496 		 * Initialize buffers
3497 		 */
3498 		io_count = 0;
3499 		for (io_off = 0; io_off < bp->b_bcount; io_off += contig) {
3500 			contig = 0;
3501 			if (error = ud_bmap_read(ip,
3502 			    (u_offset_t)(start + io_off),
3503 			    &bn, &contig)) {
3504 				goto end;
3505 			}
3506 			ASSERT(contig);
3507 			if ((io_off + contig) > bp->b_bcount) {
3508 				contig = bp->b_bcount - io_off;
3509 			}
3510 			if (bn != UDF_HOLE) {
3511 				/*
3512 				 * Clone the buffer
3513 				 * and prepare to start I/O
3514 				 */
3515 				ms->ms_ptr = mm;
3516 				bioinit(&ms->ms_buf);
3517 				rbp = bioclone(bp, io_off, (size_t)contig,
3518 				    bp->b_edev, bn, ud_slave_done,
3519 				    &ms->ms_buf, KM_NOSLEEP);
3520 				ASSERT(rbp == &ms->ms_buf);
3521 				mm->mm_resid += contig;
3522 				io_count++;
3523 				ms ++;
3524 			}
3525 		}
3526 
3527 		/*
3528 		 * Start I/O's
3529 		 */
3530 		ms = (mio_slave_t *)(((caddr_t)mm) + sizeof (mio_master_t));
3531 		for (i = 0; i < io_count; i++) {
3532 			(void) bdev_strategy(&ms->ms_buf);
3533 			ms ++;
3534 		}
3535 	}
3536 
3537 end:
3538 	if (error != 0) {
3539 		bp->b_flags |= B_ERROR;
3540 		bp->b_error = error;
3541 		if (mm != NULL) {
3542 			mutex_destroy(&mm->mm_mutex);
3543 			kmem_free(mm, mm->mm_size);
3544 		}
3545 	}
3546 	return (error);
3547 }
3548 
3549 int32_t
3550 ud_slave_done(struct buf *bp)
3551 {
3552 	mio_master_t *mm;
3553 	int32_t resid;
3554 
3555 	ASSERT(SEMA_HELD(&bp->b_sem));
3556 	ASSERT((bp->b_flags & B_DONE) == 0);
3557 
3558 	mm = ((mio_slave_t *)bp)->ms_ptr;
3559 
3560 	/*
3561 	 * Propagate error and byte count info from slave struct to
3562 	 * the master struct
3563 	 */
3564 	mutex_enter(&mm->mm_mutex);
3565 	if (bp->b_flags & B_ERROR) {
3566 
3567 		/*
3568 		 * If multiple slave buffers get
3569 		 * error we forget the old errors
3570 		 * this is ok because we any way
3571 		 * cannot return multiple errors
3572 		 */
3573 		mm->mm_error = bp->b_error;
3574 	}
3575 	mm->mm_resid -= bp->b_bcount;
3576 	resid = mm->mm_resid;
3577 	mutex_exit(&mm->mm_mutex);
3578 
3579 	/*
3580 	 * free up the resources allocated to cloned buffers.
3581 	 */
3582 	bp_mapout(bp);
3583 	biofini(bp);
3584 
3585 	if (resid == 0) {
3586 
3587 		/*
3588 		 * This is the last I/O operation
3589 		 * clean up and return the original buffer
3590 		 */
3591 		if (mm->mm_error) {
3592 			mm->mm_bp->b_flags |= B_ERROR;
3593 			mm->mm_bp->b_error = mm->mm_error;
3594 		}
3595 		biodone(mm->mm_bp);
3596 		mutex_destroy(&mm->mm_mutex);
3597 		kmem_free(mm, mm->mm_size);
3598 	}
3599 	return (0);
3600 }
3601