xref: /illumos-gate/usr/src/uts/common/fs/smbclnt/smbfs/smbfs_vnops.c (revision 0153d828c132fdb1a17c11b99386a3d1b87994cf)
1 /*
2  * Copyright (c) 2000-2001 Boris Popov
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *    This product includes software developed by Boris Popov.
16  * 4. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * $Id: smbfs_vnops.c,v 1.128.36.1 2005/05/27 02:35:28 lindak Exp $
33  */
34 
35 /*
36  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
37  * Copyright 2021 Tintri by DDN, Inc.  All rights reserved.
38  */
39 
40 /*
41  * Vnode operations
42  *
43  * This file is similar to nfs3_vnops.c
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/cred.h>
49 #include <sys/vnode.h>
50 #include <sys/vfs.h>
51 #include <sys/filio.h>
52 #include <sys/uio.h>
53 #include <sys/dirent.h>
54 #include <sys/errno.h>
55 #include <sys/sunddi.h>
56 #include <sys/sysmacros.h>
57 #include <sys/kmem.h>
58 #include <sys/cmn_err.h>
59 #include <sys/vfs_opreg.h>
60 #include <sys/policy.h>
61 #include <sys/sdt.h>
62 #include <sys/taskq_impl.h>
63 #include <sys/zone.h>
64 
65 #ifdef	_KERNEL
66 #include <sys/vmsystm.h>	// for desfree
67 #include <vm/hat.h>
68 #include <vm/as.h>
69 #include <vm/page.h>
70 #include <vm/pvn.h>
71 #include <vm/seg.h>
72 #include <vm/seg_map.h>
73 #include <vm/seg_kpm.h>
74 #include <vm/seg_vn.h>
75 #endif	// _KERNEL
76 
77 #include <netsmb/smb_osdep.h>
78 #include <netsmb/smb.h>
79 #include <netsmb/smb_conn.h>
80 #include <netsmb/smb_subr.h>
81 
82 #include <smbfs/smbfs.h>
83 #include <smbfs/smbfs_node.h>
84 #include <smbfs/smbfs_subr.h>
85 
86 #include <sys/fs/smbfs_ioctl.h>
87 #include <fs/fs_subr.h>
88 
89 #ifndef	MAXOFF32_T
90 #define	MAXOFF32_T	0x7fffffff
91 #endif
92 
93 /*
94  * We assign directory offsets like the NFS client, where the
95  * offset increments by _one_ after each directory entry.
96  * Further, the entries "." and ".." are always at offsets
97  * zero and one (respectively) and the "real" entries from
98  * the server appear at offsets starting with two.  This
99  * macro is used to initialize the n_dirofs field after
100  * setting n_dirseq with a _findopen call.
101  */
102 #define	FIRST_DIROFS	2
103 
104 /*
105  * These characters are illegal in NTFS file names.
106  * ref: http://support.microsoft.com/kb/147438
107  *
108  * Careful!  The check in the XATTR case skips the
109  * first character to allow colon in XATTR names.
110  */
111 static const char illegal_chars[] = {
112 	':',	/* colon - keep this first! */
113 	'\\',	/* back slash */
114 	'/',	/* slash */
115 	'*',	/* asterisk */
116 	'?',	/* question mark */
117 	'"',	/* double quote */
118 	'<',	/* less than sign */
119 	'>',	/* greater than sign */
120 	'|',	/* vertical bar */
121 	0
122 };
123 
124 /*
125  * Turning this on causes nodes to be created in the cache
126  * during directory listings, normally avoiding a second
127  * OtW attribute fetch just after a readdir.
128  */
129 int smbfs_fastlookup = 1;
130 
131 struct vnodeops *smbfs_vnodeops = NULL;
132 
133 /* local static function defines */
134 
135 static int	smbfslookup_cache(vnode_t *, char *, int, vnode_t **,
136 			cred_t *);
137 static int	smbfslookup(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr,
138 			int cache_ok, caller_context_t *);
139 static int	smbfsremove(vnode_t *dvp, vnode_t *vp, struct smb_cred *scred,
140 			int flags);
141 static int	smbfsrename(vnode_t *odvp, vnode_t *ovp, vnode_t *ndvp,
142 			char *nnm, struct smb_cred *scred, int flags);
143 static int	smbfssetattr(vnode_t *, struct vattr *, int, cred_t *);
144 static int	smbfs_accessx(void *, int, cred_t *);
145 static int	smbfs_readvdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
146 			caller_context_t *);
147 static int	smbfsflush(smbnode_t *, struct smb_cred *);
148 static void	smbfs_rele_fid(smbnode_t *, struct smb_cred *);
149 static uint32_t xvattr_to_dosattr(smbnode_t *, struct vattr *);
150 
151 static int	smbfs_fsync(vnode_t *, int, cred_t *, caller_context_t *);
152 
153 static int	smbfs_putpage(vnode_t *, offset_t, size_t, int, cred_t *,
154 			caller_context_t *);
155 #ifdef	_KERNEL
156 static int	smbfs_getapage(vnode_t *, u_offset_t, size_t, uint_t *,
157 			page_t *[], size_t, struct seg *, caddr_t,
158 			enum seg_rw, cred_t *);
159 static int	smbfs_putapage(vnode_t *, page_t *, u_offset_t *, size_t *,
160 			int, cred_t *);
161 static void	smbfs_delmap_async(void *);
162 
163 static int	smbfs_rdwrlbn(vnode_t *, page_t *, u_offset_t, size_t, int,
164 			cred_t *);
165 static int	smbfs_bio(struct buf *, int, cred_t *);
166 static int	smbfs_writenp(smbnode_t *np, caddr_t base, int tcount,
167 			struct uio *uiop, int pgcreated);
168 #endif	// _KERNEL
169 
170 /*
171  * Error flags used to pass information about certain special errors
172  * which need to be handled specially.
173  */
174 #define	SMBFS_EOF			-98
175 
176 /* When implementing OtW locks, make this a real function. */
177 #define	smbfs_lm_has_sleep(vp) 0
178 
179 /*
180  * These are the vnode ops routines which implement the vnode interface to
181  * the networked file system.  These routines just take their parameters,
182  * make them look networkish by putting the right info into interface structs,
183  * and then calling the appropriate remote routine(s) to do the work.
184  *
185  * Note on directory name lookup cacheing:  If we detect a stale fhandle,
186  * we purge the directory cache relative to that vnode.  This way, the
187  * user won't get burned by the cache repeatedly.  See <smbfs/smbnode.h> for
188  * more details on smbnode locking.
189  */
190 
191 
192 /*
193  * XXX
194  * When new and relevant functionality is enabled, we should be
195  * calling vfs_set_feature() to inform callers that pieces of
196  * functionality are available, per PSARC 2007/227.
197  */
198 /* ARGSUSED */
199 static int
200 smbfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
201 {
202 	smbnode_t	*np;
203 	vnode_t		*vp;
204 	smbfattr_t	fa;
205 	smb_fh_t	*fid = NULL;
206 	smb_fh_t	*oldfid;
207 	uint32_t	rights;
208 	struct smb_cred scred;
209 	smbmntinfo_t	*smi;
210 	smb_share_t	*ssp;
211 	cred_t		*oldcr;
212 	int		error = 0;
213 
214 	vp = *vpp;
215 	np = VTOSMB(vp);
216 	smi = VTOSMI(vp);
217 	ssp = smi->smi_share;
218 
219 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
220 		return (EIO);
221 
222 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
223 		return (EIO);
224 
225 	if (vp->v_type != VREG && vp->v_type != VDIR) { /* XXX VLNK? */
226 		SMBVDEBUG("open eacces vtype=%d\n", vp->v_type);
227 		return (EACCES);
228 	}
229 
230 	/*
231 	 * Get exclusive access to n_fid and related stuff.
232 	 * No returns after this until out.
233 	 */
234 	if (smbfs_rw_enter_sig(&np->r_lkserlock, RW_WRITER, SMBINTR(vp)))
235 		return (EINTR);
236 	smb_credinit(&scred, cr);
237 
238 	/*
239 	 * Keep track of the vnode type at first open.
240 	 * It may change later, and we need close to do
241 	 * cleanup for the type we opened.  Also deny
242 	 * open of new types until old type is closed.
243 	 */
244 	if (np->n_ovtype == VNON) {
245 		ASSERT(np->n_dirrefs == 0);
246 		ASSERT(np->n_fidrefs == 0);
247 	} else if (np->n_ovtype != vp->v_type) {
248 		SMBVDEBUG("open n_ovtype=%d v_type=%d\n",
249 		    np->n_ovtype, vp->v_type);
250 		error = EACCES;
251 		goto out;
252 	}
253 
254 	/*
255 	 * Directory open.  See smbfs_readvdir()
256 	 */
257 	if (vp->v_type == VDIR) {
258 		if (np->n_dirseq == NULL) {
259 			/* first open */
260 			error = smbfs_smb_findopen(np, "*", 1,
261 			    SMB_FA_SYSTEM | SMB_FA_HIDDEN | SMB_FA_DIR,
262 			    &scred, &np->n_dirseq);
263 			if (error != 0)
264 				goto out;
265 		}
266 		np->n_dirofs = FIRST_DIROFS;
267 		np->n_dirrefs++;
268 		goto have_fid;
269 	}
270 
271 	/*
272 	 * If caller specified O_TRUNC/FTRUNC, then be sure to set
273 	 * FWRITE (to drive successful setattr(size=0) after open)
274 	 */
275 	if (flag & FTRUNC)
276 		flag |= FWRITE;
277 
278 	/*
279 	 * If we already have it open, and the FID is still valid,
280 	 * check whether the rights are sufficient for FID reuse.
281 	 */
282 	if (np->n_fidrefs > 0 &&
283 	    (fid = np->n_fid) != NULL &&
284 	    fid->fh_vcgenid == ssp->ss_vcgenid) {
285 		int upgrade = 0;
286 
287 		if ((flag & FWRITE) &&
288 		    !(fid->fh_rights & SA_RIGHT_FILE_WRITE_DATA))
289 			upgrade = 1;
290 		if ((flag & FREAD) &&
291 		    !(fid->fh_rights & SA_RIGHT_FILE_READ_DATA))
292 			upgrade = 1;
293 		if (!upgrade) {
294 			/*
295 			 *  the existing open is good enough
296 			 */
297 			np->n_fidrefs++;
298 			goto have_fid;
299 		}
300 		fid = NULL;
301 	}
302 	rights = (fid != NULL) ? fid->fh_rights : 0;
303 
304 	/*
305 	 * we always ask for READ_CONTROL so we can always get the
306 	 * owner/group IDs to satisfy a stat.  Ditto attributes.
307 	 */
308 	rights |= (STD_RIGHT_READ_CONTROL_ACCESS |
309 	    SA_RIGHT_FILE_READ_ATTRIBUTES);
310 	if ((flag & FREAD))
311 		rights |= SA_RIGHT_FILE_READ_DATA;
312 	if ((flag & FWRITE))
313 		rights |= SA_RIGHT_FILE_WRITE_DATA |
314 		    SA_RIGHT_FILE_APPEND_DATA |
315 		    SA_RIGHT_FILE_WRITE_ATTRIBUTES;
316 
317 	bzero(&fa, sizeof (fa));
318 	error = smbfs_smb_open(np,
319 	    NULL, 0, 0, /* name nmlen xattr */
320 	    rights, &scred,
321 	    &fid, &fa);
322 	if (error)
323 		goto out;
324 	smbfs_attrcache_fa(vp, &fa);
325 
326 	/*
327 	 * We have a new FID and access rights.
328 	 */
329 	VERIFY(fid != NULL);
330 	oldfid = np->n_fid;
331 	np->n_fid = fid;
332 	np->n_fidrefs++;
333 	if (oldfid != NULL)
334 		smb_fh_rele(oldfid);
335 
336 	/*
337 	 * This thread did the open.
338 	 * Save our credentials too.
339 	 */
340 	mutex_enter(&np->r_statelock);
341 	oldcr = np->r_cred;
342 	np->r_cred = cr;
343 	crhold(cr);
344 	if (oldcr)
345 		crfree(oldcr);
346 	mutex_exit(&np->r_statelock);
347 
348 have_fid:
349 	/*
350 	 * Keep track of the vnode type at first open.
351 	 * (see comments above)
352 	 */
353 	if (np->n_ovtype == VNON)
354 		np->n_ovtype = vp->v_type;
355 
356 out:
357 	smb_credrele(&scred);
358 	smbfs_rw_exit(&np->r_lkserlock);
359 	return (error);
360 }
361 
362 /*ARGSUSED*/
363 static int
364 smbfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
365 	caller_context_t *ct)
366 {
367 	smbnode_t	*np;
368 	smbmntinfo_t	*smi;
369 	struct smb_cred scred;
370 	int error = 0;
371 
372 	np = VTOSMB(vp);
373 	smi = VTOSMI(vp);
374 
375 	/*
376 	 * Don't "bail out" for VFS_UNMOUNTED here,
377 	 * as we want to do cleanup, etc.
378 	 */
379 
380 	/*
381 	 * zone_enter(2) prevents processes from changing zones with SMBFS files
382 	 * open; if we happen to get here from the wrong zone we can't do
383 	 * anything over the wire.
384 	 */
385 	if (smi->smi_zone_ref.zref_zone != curproc->p_zone) {
386 		/*
387 		 * We could attempt to clean up locks, except we're sure
388 		 * that the current process didn't acquire any locks on
389 		 * the file: any attempt to lock a file belong to another zone
390 		 * will fail, and one can't lock an SMBFS file and then change
391 		 * zones, as that fails too.
392 		 *
393 		 * Returning an error here is the sane thing to do.  A
394 		 * subsequent call to VN_RELE() which translates to a
395 		 * smbfs_inactive() will clean up state: if the zone of the
396 		 * vnode's origin is still alive and kicking, an async worker
397 		 * thread will handle the request (from the correct zone), and
398 		 * everything (minus the final smbfs_getattr_otw() call) should
399 		 * be OK. If the zone is going away smbfs_async_inactive() will
400 		 * throw away cached pages inline.
401 		 */
402 		return (EIO);
403 	}
404 
405 	/*
406 	 * If we are using local locking for this filesystem, then
407 	 * release all of the SYSV style record locks.  Otherwise,
408 	 * we are doing network locking and we need to release all
409 	 * of the network locks.  All of the locks held by this
410 	 * process on this file are released no matter what the
411 	 * incoming reference count is.
412 	 */
413 	if (smi->smi_flags & SMI_LLOCK) {
414 		pid_t pid = ddi_get_pid();
415 		cleanlocks(vp, pid, 0);
416 		cleanshares(vp, pid);
417 	}
418 	/*
419 	 * else doing OtW locking.  SMB servers drop all locks
420 	 * on the file ID we close here, so no _lockrelease()
421 	 */
422 
423 	/*
424 	 * This (passed in) count is the ref. count from the
425 	 * user's file_t before the closef call (fio.c).
426 	 * The rest happens only on last close.
427 	 */
428 	if (count > 1)
429 		return (0);
430 
431 	/* NFS has DNLC purge here. */
432 
433 	/*
434 	 * If the file was open for write and there are pages,
435 	 * then make sure dirty pages written back.
436 	 *
437 	 * NFS does this async when "close-to-open" is off
438 	 * (MI_NOCTO flag is set) to avoid blocking the caller.
439 	 * For now, always do this synchronously (no B_ASYNC).
440 	 */
441 	if ((flag & FWRITE) && vn_has_cached_data(vp)) {
442 		error = smbfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
443 		if (error == EAGAIN)
444 			error = 0;
445 	}
446 	if (error == 0) {
447 		mutex_enter(&np->r_statelock);
448 		np->r_flags &= ~RSTALE;
449 		np->r_error = 0;
450 		mutex_exit(&np->r_statelock);
451 	}
452 
453 	/*
454 	 * Decrement the reference count for the FID
455 	 * and possibly do the OtW close.
456 	 *
457 	 * Exclusive lock for modifying n_fid stuff.
458 	 * Don't want this one ever interruptible.
459 	 */
460 	(void) smbfs_rw_enter_sig(&np->r_lkserlock, RW_WRITER, 0);
461 	smb_credinit(&scred, cr);
462 
463 	smbfs_rele_fid(np, &scred);
464 
465 	smb_credrele(&scred);
466 	smbfs_rw_exit(&np->r_lkserlock);
467 
468 	return (0);
469 }
470 
471 /*
472  * Helper for smbfs_close.  Decrement the reference count
473  * for an SMB-level file or directory ID, and when the last
474  * reference for the fid goes away, do the OtW close.
475  * Also called in smbfs_inactive (defensive cleanup).
476  */
477 static void
478 smbfs_rele_fid(smbnode_t *np, struct smb_cred *scred)
479 {
480 	cred_t		*oldcr;
481 	struct smbfs_fctx *fctx;
482 	int		error;
483 	smb_fh_t	*ofid;
484 
485 	error = 0;
486 
487 	/* Make sure we serialize for n_dirseq use. */
488 	ASSERT(smbfs_rw_lock_held(&np->r_lkserlock, RW_WRITER));
489 
490 	/*
491 	 * Note that vp->v_type may change if a remote node
492 	 * is deleted and recreated as a different type, and
493 	 * our getattr may change v_type accordingly.
494 	 * Now use n_ovtype to keep track of the v_type
495 	 * we had during open (see comments above).
496 	 */
497 	switch (np->n_ovtype) {
498 	case VDIR:
499 		ASSERT(np->n_dirrefs > 0);
500 		if (--np->n_dirrefs)
501 			return;
502 		if ((fctx = np->n_dirseq) != NULL) {
503 			np->n_dirseq = NULL;
504 			np->n_dirofs = 0;
505 			error = smbfs_smb_findclose(fctx, scred);
506 		}
507 		break;
508 
509 	case VREG:
510 		ASSERT(np->n_fidrefs > 0);
511 		if (--np->n_fidrefs)
512 			return;
513 		if ((ofid = np->n_fid) != NULL) {
514 			np->n_fid = NULL;
515 			smb_fh_rele(ofid);
516 		}
517 		break;
518 
519 	default:
520 		SMBVDEBUG("bad n_ovtype %d\n", np->n_ovtype);
521 		break;
522 	}
523 	if (error) {
524 		SMBVDEBUG("error %d closing %s\n",
525 		    error, np->n_rpath);
526 	}
527 
528 	/* Allow next open to use any v_type. */
529 	np->n_ovtype = VNON;
530 
531 	/*
532 	 * Other "last close" stuff.
533 	 */
534 	mutex_enter(&np->r_statelock);
535 	if (np->n_flag & NATTRCHANGED)
536 		smbfs_attrcache_rm_locked(np);
537 	oldcr = np->r_cred;
538 	np->r_cred = NULL;
539 	mutex_exit(&np->r_statelock);
540 	if (oldcr != NULL)
541 		crfree(oldcr);
542 }
543 
544 /* ARGSUSED */
545 static int
546 smbfs_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
547 	caller_context_t *ct)
548 {
549 	struct smb_cred scred;
550 	struct vattr	va;
551 	smbnode_t	*np;
552 	smbmntinfo_t	*smi;
553 	offset_t	endoff;
554 	ssize_t		past_eof;
555 	int		error;
556 
557 	np = VTOSMB(vp);
558 	smi = VTOSMI(vp);
559 
560 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
561 		return (EIO);
562 
563 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
564 		return (EIO);
565 
566 	/* Sanity check: should have a valid open */
567 	if (np->n_fid == NULL)
568 		return (EIO);
569 
570 	ASSERT(smbfs_rw_lock_held(&np->r_rwlock, RW_READER));
571 
572 	if (vp->v_type != VREG)
573 		return (EISDIR);
574 
575 	if (uiop->uio_resid == 0)
576 		return (0);
577 
578 	/*
579 	 * Like NFS3, just check for 63-bit overflow.
580 	 * Our SMB layer takes care to return EFBIG
581 	 * when it has to fallback to a 32-bit call.
582 	 */
583 	endoff = uiop->uio_loffset + uiop->uio_resid;
584 	if (uiop->uio_loffset < 0 || endoff < 0)
585 		return (EINVAL);
586 
587 	/* get vnode attributes from server */
588 	va.va_mask = AT_SIZE | AT_MTIME;
589 	if (error = smbfsgetattr(vp, &va, cr))
590 		return (error);
591 
592 	/* Update mtime with mtime from server here? */
593 
594 	/* if offset is beyond EOF, read nothing */
595 	if (uiop->uio_loffset >= va.va_size)
596 		return (0);
597 
598 	/*
599 	 * Limit the read to the remaining file size.
600 	 * Do this by temporarily reducing uio_resid
601 	 * by the amount the lies beyoned the EOF.
602 	 */
603 	if (endoff > va.va_size) {
604 		past_eof = (ssize_t)(endoff - va.va_size);
605 		uiop->uio_resid -= past_eof;
606 	} else
607 		past_eof = 0;
608 
609 	/*
610 	 * Bypass VM if caching has been disabled (e.g., locking) or if
611 	 * using client-side direct I/O and the file is not mmap'd and
612 	 * there are no cached pages.
613 	 */
614 	if ((vp->v_flag & VNOCACHE) ||
615 	    (((np->r_flags & RDIRECTIO) || (smi->smi_flags & SMI_DIRECTIO)) &&
616 	    np->r_mapcnt == 0 && np->r_inmap == 0 &&
617 	    !vn_has_cached_data(vp))) {
618 
619 		/* Shared lock for n_fid use in smb_rwuio */
620 		if (smbfs_rw_enter_sig(&np->r_lkserlock, RW_READER, SMBINTR(vp)))
621 			return (EINTR);
622 		smb_credinit(&scred, cr);
623 
624 		error = smb_rwuio(np->n_fid, UIO_READ,
625 		    uiop, &scred, smb_timo_read);
626 
627 		smb_credrele(&scred);
628 		smbfs_rw_exit(&np->r_lkserlock);
629 
630 		/* undo adjustment of resid */
631 		uiop->uio_resid += past_eof;
632 
633 		return (error);
634 	}
635 
636 #ifdef	_KERNEL
637 	/* (else) Do I/O through segmap. */
638 	do {
639 		caddr_t		base;
640 		u_offset_t	off;
641 		size_t		n;
642 		int		on;
643 		uint_t		flags;
644 
645 		off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
646 		on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
647 		n = MIN(MAXBSIZE - on, uiop->uio_resid);
648 
649 		error = smbfs_validate_caches(vp, cr);
650 		if (error)
651 			break;
652 
653 		/* NFS waits for RINCACHEPURGE here. */
654 
655 		if (vpm_enable) {
656 			/*
657 			 * Copy data.
658 			 */
659 			error = vpm_data_copy(vp, off + on, n, uiop,
660 			    1, NULL, 0, S_READ);
661 		} else {
662 			base = segmap_getmapflt(segkmap, vp, off + on, n, 1,
663 			    S_READ);
664 
665 			error = uiomove(base + on, n, UIO_READ, uiop);
666 		}
667 
668 		if (!error) {
669 			/*
670 			 * If read a whole block or read to eof,
671 			 * won't need this buffer again soon.
672 			 */
673 			mutex_enter(&np->r_statelock);
674 			if (n + on == MAXBSIZE ||
675 			    uiop->uio_loffset == np->r_size)
676 				flags = SM_DONTNEED;
677 			else
678 				flags = 0;
679 			mutex_exit(&np->r_statelock);
680 			if (vpm_enable) {
681 				error = vpm_sync_pages(vp, off, n, flags);
682 			} else {
683 				error = segmap_release(segkmap, base, flags);
684 			}
685 		} else {
686 			if (vpm_enable) {
687 				(void) vpm_sync_pages(vp, off, n, 0);
688 			} else {
689 				(void) segmap_release(segkmap, base, 0);
690 			}
691 		}
692 	} while (!error && uiop->uio_resid > 0);
693 #else	// _KERNEL
694 	error = ENOSYS;
695 #endif	// _KERNEL
696 
697 	/* undo adjustment of resid */
698 	uiop->uio_resid += past_eof;
699 
700 	return (error);
701 }
702 
703 
704 /* ARGSUSED */
705 static int
706 smbfs_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
707 	caller_context_t *ct)
708 {
709 	struct smb_cred scred;
710 	struct vattr    va;
711 	smbnode_t	*np;
712 	smbmntinfo_t	*smi;
713 	offset_t	endoff, limit;
714 	ssize_t		past_limit;
715 	int		error, timo;
716 	u_offset_t	last_off;
717 	size_t		last_resid;
718 #ifdef	_KERNEL
719 	uint_t		bsize;
720 #endif
721 
722 	np = VTOSMB(vp);
723 	smi = VTOSMI(vp);
724 
725 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
726 		return (EIO);
727 
728 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
729 		return (EIO);
730 
731 	/* Sanity check: should have a valid open */
732 	if (np->n_fid == NULL)
733 		return (EIO);
734 
735 	ASSERT(smbfs_rw_lock_held(&np->r_rwlock, RW_WRITER));
736 
737 	if (vp->v_type != VREG)
738 		return (EISDIR);
739 
740 	if (uiop->uio_resid == 0)
741 		return (0);
742 
743 	/*
744 	 * Handle ioflag bits: (FAPPEND|FSYNC|FDSYNC)
745 	 */
746 	if (ioflag & (FAPPEND | FSYNC)) {
747 		if (np->n_flag & NMODIFIED) {
748 			smbfs_attrcache_remove(np);
749 		}
750 	}
751 	if (ioflag & FAPPEND) {
752 		/*
753 		 * File size can be changed by another client
754 		 *
755 		 * Todo: Consider redesigning this to use a
756 		 * handle opened for append instead.
757 		 */
758 		va.va_mask = AT_SIZE;
759 		if (error = smbfsgetattr(vp, &va, cr))
760 			return (error);
761 		uiop->uio_loffset = va.va_size;
762 	}
763 
764 	/*
765 	 * Like NFS3, just check for 63-bit overflow.
766 	 */
767 	endoff = uiop->uio_loffset + uiop->uio_resid;
768 	if (uiop->uio_loffset < 0 || endoff < 0)
769 		return (EINVAL);
770 
771 	/*
772 	 * Check to make sure that the process will not exceed
773 	 * its limit on file size.  It is okay to write up to
774 	 * the limit, but not beyond.  Thus, the write which
775 	 * reaches the limit will be short and the next write
776 	 * will return an error.
777 	 *
778 	 * So if we're starting at or beyond the limit, EFBIG.
779 	 * Otherwise, temporarily reduce resid to the amount
780 	 * that is after the limit.
781 	 */
782 	limit = uiop->uio_llimit;
783 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
784 		limit = MAXOFFSET_T;
785 	if (uiop->uio_loffset >= limit) {
786 #ifdef	_KERNEL
787 		proc_t *p = ttoproc(curthread);
788 
789 		mutex_enter(&p->p_lock);
790 		(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE],
791 		    p->p_rctls, p, RCA_UNSAFE_SIGINFO);
792 		mutex_exit(&p->p_lock);
793 #endif	// _KERNEL
794 		return (EFBIG);
795 	}
796 	if (endoff > limit) {
797 		past_limit = (ssize_t)(endoff - limit);
798 		uiop->uio_resid -= past_limit;
799 	} else
800 		past_limit = 0;
801 
802 	/*
803 	 * Bypass VM if caching has been disabled (e.g., locking) or if
804 	 * using client-side direct I/O and the file is not mmap'd and
805 	 * there are no cached pages.
806 	 */
807 	if ((vp->v_flag & VNOCACHE) ||
808 	    (((np->r_flags & RDIRECTIO) || (smi->smi_flags & SMI_DIRECTIO)) &&
809 	    np->r_mapcnt == 0 && np->r_inmap == 0 &&
810 	    !vn_has_cached_data(vp))) {
811 
812 #ifdef	_KERNEL
813 smbfs_fwrite:
814 #endif	// _KERNEL
815 		if (np->r_flags & RSTALE) {
816 			last_resid = uiop->uio_resid;
817 			last_off = uiop->uio_loffset;
818 			error = np->r_error;
819 			/*
820 			 * A close may have cleared r_error, if so,
821 			 * propagate ESTALE error return properly
822 			 */
823 			if (error == 0)
824 				error = ESTALE;
825 			goto bottom;
826 		}
827 
828 		/* Timeout: longer for append. */
829 		timo = smb_timo_write;
830 		if (endoff > np->r_size)
831 			timo = smb_timo_append;
832 
833 		/* Shared lock for n_fid use in smb_rwuio */
834 		if (smbfs_rw_enter_sig(&np->r_lkserlock, RW_READER, SMBINTR(vp)))
835 			return (EINTR);
836 		smb_credinit(&scred, cr);
837 
838 		error = smb_rwuio(np->n_fid, UIO_WRITE,
839 		    uiop, &scred, timo);
840 
841 		if (error == 0) {
842 			mutex_enter(&np->r_statelock);
843 			np->n_flag |= (NFLUSHWIRE | NATTRCHANGED);
844 			if (uiop->uio_loffset > (offset_t)np->r_size)
845 				np->r_size = (len_t)uiop->uio_loffset;
846 			mutex_exit(&np->r_statelock);
847 			if (ioflag & (FSYNC | FDSYNC)) {
848 				/* Don't error the I/O if this fails. */
849 				(void) smbfsflush(np, &scred);
850 			}
851 		}
852 
853 		smb_credrele(&scred);
854 		smbfs_rw_exit(&np->r_lkserlock);
855 
856 		/* undo adjustment of resid */
857 		uiop->uio_resid += past_limit;
858 
859 		return (error);
860 	}
861 
862 #ifdef	_KERNEL
863 	/* (else) Do I/O through segmap. */
864 	bsize = vp->v_vfsp->vfs_bsize;
865 
866 	do {
867 		caddr_t		base;
868 		u_offset_t	off;
869 		size_t		n;
870 		int		on;
871 		uint_t		flags;
872 
873 		off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
874 		on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
875 		n = MIN(MAXBSIZE - on, uiop->uio_resid);
876 
877 		last_resid = uiop->uio_resid;
878 		last_off = uiop->uio_loffset;
879 
880 		if (np->r_flags & RSTALE) {
881 			error = np->r_error;
882 			/*
883 			 * A close may have cleared r_error, if so,
884 			 * propagate ESTALE error return properly
885 			 */
886 			if (error == 0)
887 				error = ESTALE;
888 			break;
889 		}
890 
891 		/*
892 		 * From NFS: Don't create dirty pages faster than they
893 		 * can be cleaned.
894 		 *
895 		 * Here NFS also checks for async writes (np->r_awcount)
896 		 */
897 		mutex_enter(&np->r_statelock);
898 		while (np->r_gcount > 0) {
899 			if (SMBINTR(vp)) {
900 				klwp_t *lwp = ttolwp(curthread);
901 
902 				if (lwp != NULL)
903 					lwp->lwp_nostop++;
904 				if (!cv_wait_sig(&np->r_cv, &np->r_statelock)) {
905 					mutex_exit(&np->r_statelock);
906 					if (lwp != NULL)
907 						lwp->lwp_nostop--;
908 					error = EINTR;
909 					goto bottom;
910 				}
911 				if (lwp != NULL)
912 					lwp->lwp_nostop--;
913 			} else
914 				cv_wait(&np->r_cv, &np->r_statelock);
915 		}
916 		mutex_exit(&np->r_statelock);
917 
918 		/*
919 		 * Touch the page and fault it in if it is not in core
920 		 * before segmap_getmapflt or vpm_data_copy can lock it.
921 		 * This is to avoid the deadlock if the buffer is mapped
922 		 * to the same file through mmap which we want to write.
923 		 */
924 		uio_prefaultpages((long)n, uiop);
925 
926 		if (vpm_enable) {
927 			/*
928 			 * It will use kpm mappings, so no need to
929 			 * pass an address.
930 			 */
931 			error = smbfs_writenp(np, NULL, n, uiop, 0);
932 		} else {
933 			if (segmap_kpm) {
934 				int pon = uiop->uio_loffset & PAGEOFFSET;
935 				size_t pn = MIN(PAGESIZE - pon,
936 				    uiop->uio_resid);
937 				int pagecreate;
938 
939 				mutex_enter(&np->r_statelock);
940 				pagecreate = (pon == 0) && (pn == PAGESIZE ||
941 				    uiop->uio_loffset + pn >= np->r_size);
942 				mutex_exit(&np->r_statelock);
943 
944 				base = segmap_getmapflt(segkmap, vp, off + on,
945 				    pn, !pagecreate, S_WRITE);
946 
947 				error = smbfs_writenp(np, base + pon, n, uiop,
948 				    pagecreate);
949 
950 			} else {
951 				base = segmap_getmapflt(segkmap, vp, off + on,
952 				    n, 0, S_READ);
953 				error = smbfs_writenp(np, base + on, n, uiop, 0);
954 			}
955 		}
956 
957 		if (!error) {
958 			if (smi->smi_flags & SMI_NOAC)
959 				flags = SM_WRITE;
960 			else if ((uiop->uio_loffset % bsize) == 0 ||
961 			    IS_SWAPVP(vp)) {
962 				/*
963 				 * Have written a whole block.
964 				 * Start an asynchronous write
965 				 * and mark the buffer to
966 				 * indicate that it won't be
967 				 * needed again soon.
968 				 */
969 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
970 			} else
971 				flags = 0;
972 			if ((ioflag & (FSYNC|FDSYNC)) ||
973 			    (np->r_flags & ROUTOFSPACE)) {
974 				flags &= ~SM_ASYNC;
975 				flags |= SM_WRITE;
976 			}
977 			if (vpm_enable) {
978 				error = vpm_sync_pages(vp, off, n, flags);
979 			} else {
980 				error = segmap_release(segkmap, base, flags);
981 			}
982 		} else {
983 			if (vpm_enable) {
984 				(void) vpm_sync_pages(vp, off, n, 0);
985 			} else {
986 				(void) segmap_release(segkmap, base, 0);
987 			}
988 			/*
989 			 * In the event that we got an access error while
990 			 * faulting in a page for a write-only file just
991 			 * force a write.
992 			 */
993 			if (error == EACCES)
994 				goto smbfs_fwrite;
995 		}
996 	} while (!error && uiop->uio_resid > 0);
997 #else	// _KERNEL
998 	last_resid = uiop->uio_resid;
999 	last_off = uiop->uio_loffset;
1000 	error = ENOSYS;
1001 #endif	// _KERNEL
1002 
1003 bottom:
1004 	/* undo adjustment of resid */
1005 	if (error) {
1006 		uiop->uio_resid = last_resid + past_limit;
1007 		uiop->uio_loffset = last_off;
1008 	} else {
1009 		uiop->uio_resid += past_limit;
1010 	}
1011 
1012 	return (error);
1013 }
1014 
1015 #ifdef	_KERNEL
1016 
1017 /*
1018  * Like nfs_client.c: writerp()
1019  *
1020  * Write by creating pages and uiomove data onto them.
1021  */
1022 
1023 int
1024 smbfs_writenp(smbnode_t *np, caddr_t base, int tcount, struct uio *uio,
1025     int pgcreated)
1026 {
1027 	int		pagecreate;
1028 	int		n;
1029 	int		saved_n;
1030 	caddr_t		saved_base;
1031 	u_offset_t	offset;
1032 	int		error;
1033 	int		sm_error;
1034 	vnode_t		*vp = SMBTOV(np);
1035 
1036 	ASSERT(tcount <= MAXBSIZE && tcount <= uio->uio_resid);
1037 	ASSERT(smbfs_rw_lock_held(&np->r_rwlock, RW_WRITER));
1038 	if (!vpm_enable) {
1039 		ASSERT(((uintptr_t)base & MAXBOFFSET) + tcount <= MAXBSIZE);
1040 	}
1041 
1042 	/*
1043 	 * Move bytes in at most PAGESIZE chunks. We must avoid
1044 	 * spanning pages in uiomove() because page faults may cause
1045 	 * the cache to be invalidated out from under us. The r_size is not
1046 	 * updated until after the uiomove. If we push the last page of a
1047 	 * file before r_size is correct, we will lose the data written past
1048 	 * the current (and invalid) r_size.
1049 	 */
1050 	do {
1051 		offset = uio->uio_loffset;
1052 		pagecreate = 0;
1053 
1054 		/*
1055 		 * n is the number of bytes required to satisfy the request
1056 		 *   or the number of bytes to fill out the page.
1057 		 */
1058 		n = (int)MIN((PAGESIZE - (offset & PAGEOFFSET)), tcount);
1059 
1060 		/*
1061 		 * Check to see if we can skip reading in the page
1062 		 * and just allocate the memory.  We can do this
1063 		 * if we are going to rewrite the entire mapping
1064 		 * or if we are going to write to or beyond the current
1065 		 * end of file from the beginning of the mapping.
1066 		 *
1067 		 * The read of r_size is now protected by r_statelock.
1068 		 */
1069 		mutex_enter(&np->r_statelock);
1070 		/*
1071 		 * When pgcreated is nonzero the caller has already done
1072 		 * a segmap_getmapflt with forcefault 0 and S_WRITE. With
1073 		 * segkpm this means we already have at least one page
1074 		 * created and mapped at base.
1075 		 */
1076 		pagecreate = pgcreated ||
1077 		    ((offset & PAGEOFFSET) == 0 &&
1078 		    (n == PAGESIZE || ((offset + n) >= np->r_size)));
1079 
1080 		mutex_exit(&np->r_statelock);
1081 		if (!vpm_enable && pagecreate) {
1082 			/*
1083 			 * The last argument tells segmap_pagecreate() to
1084 			 * always lock the page, as opposed to sometimes
1085 			 * returning with the page locked. This way we avoid a
1086 			 * fault on the ensuing uiomove(), but also
1087 			 * more importantly (to fix bug 1094402) we can
1088 			 * call segmap_fault() to unlock the page in all
1089 			 * cases. An alternative would be to modify
1090 			 * segmap_pagecreate() to tell us when it is
1091 			 * locking a page, but that's a fairly major
1092 			 * interface change.
1093 			 */
1094 			if (pgcreated == 0)
1095 				(void) segmap_pagecreate(segkmap, base,
1096 				    (uint_t)n, 1);
1097 			saved_base = base;
1098 			saved_n = n;
1099 		}
1100 
1101 		/*
1102 		 * The number of bytes of data in the last page can not
1103 		 * be accurately be determined while page is being
1104 		 * uiomove'd to and the size of the file being updated.
1105 		 * Thus, inform threads which need to know accurately
1106 		 * how much data is in the last page of the file.  They
1107 		 * will not do the i/o immediately, but will arrange for
1108 		 * the i/o to happen later when this modify operation
1109 		 * will have finished.
1110 		 */
1111 		ASSERT(!(np->r_flags & RMODINPROGRESS));
1112 		mutex_enter(&np->r_statelock);
1113 		np->r_flags |= RMODINPROGRESS;
1114 		np->r_modaddr = (offset & MAXBMASK);
1115 		mutex_exit(&np->r_statelock);
1116 
1117 		if (vpm_enable) {
1118 			/*
1119 			 * Copy data. If new pages are created, part of
1120 			 * the page that is not written will be initizliazed
1121 			 * with zeros.
1122 			 */
1123 			error = vpm_data_copy(vp, offset, n, uio,
1124 			    !pagecreate, NULL, 0, S_WRITE);
1125 		} else {
1126 			error = uiomove(base, n, UIO_WRITE, uio);
1127 		}
1128 
1129 		/*
1130 		 * r_size is the maximum number of
1131 		 * bytes known to be in the file.
1132 		 * Make sure it is at least as high as the
1133 		 * first unwritten byte pointed to by uio_loffset.
1134 		 */
1135 		mutex_enter(&np->r_statelock);
1136 		if (np->r_size < uio->uio_loffset)
1137 			np->r_size = uio->uio_loffset;
1138 		np->r_flags &= ~RMODINPROGRESS;
1139 		np->r_flags |= RDIRTY;
1140 		mutex_exit(&np->r_statelock);
1141 
1142 		/* n = # of bytes written */
1143 		n = (int)(uio->uio_loffset - offset);
1144 
1145 		if (!vpm_enable) {
1146 			base += n;
1147 		}
1148 		tcount -= n;
1149 		/*
1150 		 * If we created pages w/o initializing them completely,
1151 		 * we need to zero the part that wasn't set up.
1152 		 * This happens on a most EOF write cases and if
1153 		 * we had some sort of error during the uiomove.
1154 		 */
1155 		if (!vpm_enable && pagecreate) {
1156 			if ((uio->uio_loffset & PAGEOFFSET) || n == 0)
1157 				(void) kzero(base, PAGESIZE - n);
1158 
1159 			if (pgcreated) {
1160 				/*
1161 				 * Caller is responsible for this page,
1162 				 * it was not created in this loop.
1163 				 */
1164 				pgcreated = 0;
1165 			} else {
1166 				/*
1167 				 * For bug 1094402: segmap_pagecreate locks
1168 				 * page. Unlock it. This also unlocks the
1169 				 * pages allocated by page_create_va() in
1170 				 * segmap_pagecreate().
1171 				 */
1172 				sm_error = segmap_fault(kas.a_hat, segkmap,
1173 				    saved_base, saved_n,
1174 				    F_SOFTUNLOCK, S_WRITE);
1175 				if (error == 0)
1176 					error = sm_error;
1177 			}
1178 		}
1179 	} while (tcount > 0 && error == 0);
1180 
1181 	return (error);
1182 }
1183 
1184 /*
1185  * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED}
1186  * Like nfs3_rdwrlbn()
1187  */
1188 static int
1189 smbfs_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len,
1190 	int flags, cred_t *cr)
1191 {
1192 	smbmntinfo_t	*smi = VTOSMI(vp);
1193 	struct buf *bp;
1194 	int error;
1195 	int sync;
1196 
1197 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
1198 		return (EIO);
1199 
1200 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1201 		return (EIO);
1202 
1203 	bp = pageio_setup(pp, len, vp, flags);
1204 	ASSERT(bp != NULL);
1205 
1206 	/*
1207 	 * pageio_setup should have set b_addr to 0.  This
1208 	 * is correct since we want to do I/O on a page
1209 	 * boundary.  bp_mapin will use this addr to calculate
1210 	 * an offset, and then set b_addr to the kernel virtual
1211 	 * address it allocated for us.
1212 	 */
1213 	ASSERT(bp->b_un.b_addr == 0);
1214 
1215 	bp->b_edev = 0;
1216 	bp->b_dev = 0;
1217 	bp->b_lblkno = lbtodb(off);
1218 	bp->b_file = vp;
1219 	bp->b_offset = (offset_t)off;
1220 	bp_mapin(bp);
1221 
1222 	/*
1223 	 * Calculate the desired level of stability to write data.
1224 	 */
1225 	if ((flags & (B_WRITE|B_ASYNC)) == (B_WRITE|B_ASYNC) &&
1226 	    freemem > desfree) {
1227 		sync = 0;
1228 	} else {
1229 		sync = 1;
1230 	}
1231 
1232 	error = smbfs_bio(bp, sync, cr);
1233 
1234 	bp_mapout(bp);
1235 	pageio_done(bp);
1236 
1237 	return (error);
1238 }
1239 
1240 
1241 /*
1242  * Corresponds to nfs3_vnopc.c : nfs3_bio(), though the NFS code
1243  * uses nfs3read()/nfs3write() where we use smb_rwuio().  Also,
1244  * NFS has this later in the file.  Move it up here closer to
1245  * the one call site just above.
1246  */
1247 
1248 static int
1249 smbfs_bio(struct buf *bp, int sync, cred_t *cr)
1250 {
1251 	struct iovec aiov[1];
1252 	struct uio  auio;
1253 	struct smb_cred scred;
1254 	smbnode_t *np = VTOSMB(bp->b_vp);
1255 	smbmntinfo_t *smi = np->n_mount;
1256 	offset_t offset;
1257 	offset_t endoff;
1258 	size_t count;
1259 	size_t past_eof;
1260 	int error;
1261 
1262 	ASSERT(curproc->p_zone == smi->smi_zone_ref.zref_zone);
1263 
1264 	offset = ldbtob(bp->b_lblkno);
1265 	count = bp->b_bcount;
1266 	endoff = offset + count;
1267 	if (offset < 0 || endoff < 0)
1268 		return (EINVAL);
1269 
1270 	/*
1271 	 * Limit file I/O to the remaining file size, but see
1272 	 * the notes in smbfs_getpage about SMBFS_EOF.
1273 	 */
1274 	mutex_enter(&np->r_statelock);
1275 	if (offset >= np->r_size) {
1276 		mutex_exit(&np->r_statelock);
1277 		if (bp->b_flags & B_READ) {
1278 			return (SMBFS_EOF);
1279 		} else {
1280 			return (EINVAL);
1281 		}
1282 	}
1283 	if (endoff > np->r_size) {
1284 		past_eof = (size_t)(endoff - np->r_size);
1285 		count -= past_eof;
1286 	} else
1287 		past_eof = 0;
1288 	mutex_exit(&np->r_statelock);
1289 	ASSERT(count > 0);
1290 
1291 	/* Caller did bpmapin().  Mapped address is... */
1292 	aiov[0].iov_base = bp->b_un.b_addr;
1293 	aiov[0].iov_len = count;
1294 	auio.uio_iov = aiov;
1295 	auio.uio_iovcnt = 1;
1296 	auio.uio_loffset = offset;
1297 	auio.uio_segflg = UIO_SYSSPACE;
1298 	auio.uio_fmode = 0;
1299 	auio.uio_resid = count;
1300 
1301 	/* Shared lock for n_fid use in smb_rwuio */
1302 	if (smbfs_rw_enter_sig(&np->r_lkserlock, RW_READER,
1303 	    smi->smi_flags & SMI_INT))
1304 		return (EINTR);
1305 	smb_credinit(&scred, cr);
1306 
1307 	DTRACE_IO1(start, struct buf *, bp);
1308 
1309 	if (bp->b_flags & B_READ) {
1310 
1311 		error = smb_rwuio(np->n_fid, UIO_READ,
1312 		    &auio, &scred, smb_timo_read);
1313 
1314 		/* Like NFS, only set b_error here. */
1315 		bp->b_error = error;
1316 		bp->b_resid = auio.uio_resid;
1317 
1318 		if (!error && auio.uio_resid != 0)
1319 			error = EIO;
1320 		if (!error && past_eof != 0) {
1321 			/* Zero the memory beyond EOF. */
1322 			bzero(bp->b_un.b_addr + count, past_eof);
1323 		}
1324 	} else {
1325 
1326 		error = smb_rwuio(np->n_fid, UIO_WRITE,
1327 		    &auio, &scred, smb_timo_write);
1328 
1329 		/* Like NFS, only set b_error here. */
1330 		bp->b_error = error;
1331 		bp->b_resid = auio.uio_resid;
1332 
1333 		if (!error && auio.uio_resid != 0)
1334 			error = EIO;
1335 		if (!error && sync) {
1336 			(void) smbfsflush(np, &scred);
1337 		}
1338 	}
1339 
1340 	/*
1341 	 * This comes from nfs3_commit()
1342 	 */
1343 	if (error != 0) {
1344 		mutex_enter(&np->r_statelock);
1345 		if (error == ESTALE)
1346 			np->r_flags |= RSTALE;
1347 		if (!np->r_error)
1348 			np->r_error = error;
1349 		mutex_exit(&np->r_statelock);
1350 		bp->b_flags |= B_ERROR;
1351 	}
1352 
1353 	DTRACE_IO1(done, struct buf *, bp);
1354 
1355 	smb_credrele(&scred);
1356 	smbfs_rw_exit(&np->r_lkserlock);
1357 
1358 	if (error == ESTALE)
1359 		smbfs_attrcache_remove(np);
1360 
1361 	return (error);
1362 }
1363 #endif	// _KERNEL
1364 
1365 /*
1366  * Here NFS has: nfs3write, nfs3read
1367  * We use smb_rwuio instead.
1368  */
1369 
1370 /* ARGSUSED */
1371 static int
1372 smbfs_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag,
1373 	cred_t *cr, int *rvalp,	caller_context_t *ct)
1374 {
1375 	int		error;
1376 	smbmntinfo_t	*smi;
1377 
1378 	smi = VTOSMI(vp);
1379 
1380 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
1381 		return (EIO);
1382 
1383 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1384 		return (EIO);
1385 
1386 	switch (cmd) {
1387 
1388 	case _FIOFFS:
1389 		error = smbfs_fsync(vp, 0, cr, ct);
1390 		break;
1391 
1392 		/*
1393 		 * The following two ioctls are used by bfu.
1394 		 * Silently ignore to avoid bfu errors.
1395 		 */
1396 	case _FIOGDIO:
1397 	case _FIOSDIO:
1398 		error = 0;
1399 		break;
1400 
1401 #if 0	/* Todo - SMB ioctl query regions */
1402 	case _FIO_SEEK_DATA:
1403 	case _FIO_SEEK_HOLE:
1404 #endif
1405 
1406 	case _FIODIRECTIO:
1407 		error = smbfs_directio(vp, (int)arg, cr);
1408 		break;
1409 
1410 		/*
1411 		 * Allow get/set with "raw" security descriptor (SD) data.
1412 		 * Useful for testing, diagnosing idmap problems, etc.
1413 		 */
1414 	case SMBFSIO_GETSD:
1415 		error = smbfs_acl_iocget(vp, arg, flag, cr);
1416 		break;
1417 
1418 	case SMBFSIO_SETSD:
1419 		error = smbfs_acl_iocset(vp, arg, flag, cr);
1420 		break;
1421 
1422 	default:
1423 		error = ENOTTY;
1424 		break;
1425 	}
1426 
1427 	return (error);
1428 }
1429 
1430 
1431 /*
1432  * Return either cached or remote attributes. If get remote attr
1433  * use them to check and invalidate caches, then cache the new attributes.
1434  */
1435 /* ARGSUSED */
1436 static int
1437 smbfs_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1438 	caller_context_t *ct)
1439 {
1440 	smbnode_t *np;
1441 	smbmntinfo_t *smi;
1442 	int error;
1443 
1444 	smi = VTOSMI(vp);
1445 
1446 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
1447 		return (EIO);
1448 
1449 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1450 		return (EIO);
1451 
1452 	/*
1453 	 * If it has been specified that the return value will
1454 	 * just be used as a hint, and we are only being asked
1455 	 * for size, fsid or rdevid, then return the client's
1456 	 * notion of these values without checking to make sure
1457 	 * that the attribute cache is up to date.
1458 	 * The whole point is to avoid an over the wire GETATTR
1459 	 * call.
1460 	 */
1461 	np = VTOSMB(vp);
1462 	if (flags & ATTR_HINT) {
1463 		if (vap->va_mask ==
1464 		    (vap->va_mask & (AT_SIZE | AT_FSID | AT_RDEV))) {
1465 			mutex_enter(&np->r_statelock);
1466 			if (vap->va_mask | AT_SIZE)
1467 				vap->va_size = np->r_size;
1468 			if (vap->va_mask | AT_FSID)
1469 				vap->va_fsid = vp->v_vfsp->vfs_dev;
1470 			if (vap->va_mask | AT_RDEV)
1471 				vap->va_rdev = vp->v_rdev;
1472 			mutex_exit(&np->r_statelock);
1473 			return (0);
1474 		}
1475 	}
1476 
1477 	/*
1478 	 * Only need to flush pages if asking for the mtime
1479 	 * and if there any dirty pages.
1480 	 *
1481 	 * Here NFS also checks for async writes (np->r_awcount)
1482 	 */
1483 	if (vap->va_mask & AT_MTIME) {
1484 		if (vn_has_cached_data(vp) &&
1485 		    ((np->r_flags & RDIRTY) != 0)) {
1486 			mutex_enter(&np->r_statelock);
1487 			np->r_gcount++;
1488 			mutex_exit(&np->r_statelock);
1489 			error = smbfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1490 			mutex_enter(&np->r_statelock);
1491 			if (error && (error == ENOSPC || error == EDQUOT)) {
1492 				if (!np->r_error)
1493 					np->r_error = error;
1494 			}
1495 			if (--np->r_gcount == 0)
1496 				cv_broadcast(&np->r_cv);
1497 			mutex_exit(&np->r_statelock);
1498 		}
1499 	}
1500 
1501 	return (smbfsgetattr(vp, vap, cr));
1502 }
1503 
1504 /* smbfsgetattr() in smbfs_client.c */
1505 
1506 /*ARGSUSED4*/
1507 static int
1508 smbfs_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1509 		caller_context_t *ct)
1510 {
1511 	vfs_t		*vfsp;
1512 	smbmntinfo_t	*smi;
1513 	int		error;
1514 	uint_t		mask;
1515 	struct vattr	oldva;
1516 
1517 	vfsp = vp->v_vfsp;
1518 	smi = VFTOSMI(vfsp);
1519 
1520 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
1521 		return (EIO);
1522 
1523 	if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
1524 		return (EIO);
1525 
1526 	mask = vap->va_mask;
1527 	if (mask & AT_NOSET)
1528 		return (EINVAL);
1529 
1530 	if (vfsp->vfs_flag & VFS_RDONLY)
1531 		return (EROFS);
1532 
1533 	/*
1534 	 * This is a _local_ access check so that only the owner of
1535 	 * this mount can set attributes.  With ACLs enabled, the
1536 	 * file owner can be different from the mount owner, and we
1537 	 * need to check the _mount_ owner here.  See _access_rwx
1538 	 */
1539 	bzero(&oldva, sizeof (oldva));
1540 	oldva.va_mask = AT_TYPE | AT_MODE;
1541 	error = smbfsgetattr(vp, &oldva, cr);
1542 	if (error)
1543 		return (error);
1544 	oldva.va_mask |= AT_UID | AT_GID;
1545 	oldva.va_uid = smi->smi_uid;
1546 	oldva.va_gid = smi->smi_gid;
1547 
1548 	error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
1549 	    smbfs_accessx, vp);
1550 	if (error)
1551 		return (error);
1552 
1553 	if (mask & (AT_UID | AT_GID)) {
1554 		if (smi->smi_flags & SMI_ACL)
1555 			error = smbfs_acl_setids(vp, vap, cr);
1556 		else
1557 			error = ENOSYS;
1558 		if (error != 0) {
1559 			SMBVDEBUG("error %d seting UID/GID on %s",
1560 			    error, VTOSMB(vp)->n_rpath);
1561 			/*
1562 			 * It might be more correct to return the
1563 			 * error here, but that causes complaints
1564 			 * when root extracts a cpio archive, etc.
1565 			 * So ignore this error, and go ahead with
1566 			 * the rest of the setattr work.
1567 			 */
1568 		}
1569 	}
1570 
1571 	error = smbfssetattr(vp, vap, flags, cr);
1572 
1573 #ifdef	SMBFS_VNEVENT
1574 	if (error == 0 && (vap->va_mask & AT_SIZE) && vap->va_size == 0)
1575 		vnevent_truncate(vp, ct);
1576 #endif
1577 
1578 	return (error);
1579 }
1580 
1581 /*
1582  * Mostly from Darwin smbfs_setattr()
1583  * but then modified a lot.
1584  */
1585 /* ARGSUSED */
1586 static int
1587 smbfssetattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr)
1588 {
1589 	int		error = 0;
1590 	smbnode_t	*np = VTOSMB(vp);
1591 	smbmntinfo_t	*smi = np->n_mount;
1592 	uint_t		mask = vap->va_mask;
1593 	struct timespec	*mtime, *atime;
1594 	struct smb_cred	scred;
1595 	int		modified = 0;
1596 	smb_fh_t	*fid = NULL;
1597 	uint32_t rights = 0;
1598 	uint32_t dosattr = 0;
1599 
1600 	ASSERT(curproc->p_zone == VTOSMI(vp)->smi_zone_ref.zref_zone);
1601 
1602 	/*
1603 	 * There are no settable attributes on the XATTR dir,
1604 	 * so just silently ignore these.  On XATTR files,
1605 	 * you can set the size but nothing else.
1606 	 */
1607 	if (vp->v_flag & V_XATTRDIR)
1608 		return (0);
1609 	if (np->n_flag & N_XATTR) {
1610 		if (mask & AT_TIMES)
1611 			SMBVDEBUG("ignore set time on xattr\n");
1612 		mask &= AT_SIZE;
1613 	}
1614 
1615 	/*
1616 	 * Only need to flush pages if there are any pages and
1617 	 * if the file is marked as dirty in some fashion.  The
1618 	 * file must be flushed so that we can accurately
1619 	 * determine the size of the file and the cached data
1620 	 * after the SETATTR returns.  A file is considered to
1621 	 * be dirty if it is either marked with RDIRTY, has
1622 	 * outstanding i/o's active, or is mmap'd.  In this
1623 	 * last case, we can't tell whether there are dirty
1624 	 * pages, so we flush just to be sure.
1625 	 */
1626 	if (vn_has_cached_data(vp) &&
1627 	    ((np->r_flags & RDIRTY) ||
1628 	    np->r_count > 0 ||
1629 	    np->r_mapcnt > 0)) {
1630 		ASSERT(vp->v_type != VCHR);
1631 		error = smbfs_putpage(vp, (offset_t)0, 0, 0, cr, NULL);
1632 		if (error && (error == ENOSPC || error == EDQUOT)) {
1633 			mutex_enter(&np->r_statelock);
1634 			if (!np->r_error)
1635 				np->r_error = error;
1636 			mutex_exit(&np->r_statelock);
1637 		}
1638 	}
1639 
1640 	/*
1641 	 * If our caller is trying to set multiple attributes, they
1642 	 * can make no assumption about what order they are done in.
1643 	 * Here we try to do them in order of decreasing likelihood
1644 	 * of failure, just to minimize the chance we'll wind up
1645 	 * with a partially complete request.
1646 	 */
1647 
1648 	smb_credinit(&scred, cr);
1649 
1650 	/*
1651 	 * If the caller has provided extensible attributes,
1652 	 * map those into DOS attributes supported by SMB.
1653 	 * Note: zero means "no change".
1654 	 */
1655 	if (mask & AT_XVATTR)
1656 		dosattr = xvattr_to_dosattr(np, vap);
1657 
1658 	/*
1659 	 * Will we need an open handle for this setattr?
1660 	 * If so, what rights will we need?
1661 	 */
1662 	if (dosattr || (mask & (AT_ATIME | AT_MTIME))) {
1663 		rights |=
1664 		    SA_RIGHT_FILE_WRITE_ATTRIBUTES;
1665 	}
1666 	if (mask & AT_SIZE) {
1667 		rights |=
1668 		    SA_RIGHT_FILE_WRITE_DATA |
1669 		    SA_RIGHT_FILE_APPEND_DATA;
1670 	}
1671 
1672 	/*
1673 	 * Only SIZE really requires a handle, but it's
1674 	 * simpler and more reliable to set via a handle.
1675 	 * Some servers like NT4 won't set times by path.
1676 	 * Also, we're usually setting everything anyway.
1677 	 */
1678 	if (rights != 0) {
1679 		error = smbfs_smb_tmpopen(np, rights, &scred, &fid);
1680 		if (error) {
1681 			SMBVDEBUG("error %d opening %s\n",
1682 			    error, np->n_rpath);
1683 			goto out;
1684 		}
1685 		ASSERT(fid != NULL);
1686 	}
1687 
1688 	/*
1689 	 * If the server supports the UNIX extensions, right here is where
1690 	 * we'd support changes to uid, gid, mode, and possibly va_flags.
1691 	 * For now we claim to have made any such changes.
1692 	 */
1693 
1694 	if (mask & AT_SIZE) {
1695 		/*
1696 		 * If the new file size is less than what the client sees as
1697 		 * the file size, then just change the size and invalidate
1698 		 * the pages.
1699 		 */
1700 
1701 		/*
1702 		 * Set the file size to vap->va_size.
1703 		 */
1704 		ASSERT(fid != NULL);
1705 		error = smbfs_smb_setfsize(smi->smi_share, fid,
1706 		    vap->va_size, &scred);
1707 		if (error) {
1708 			SMBVDEBUG("setsize error %d file %s\n",
1709 			    error, np->n_rpath);
1710 		} else {
1711 			/*
1712 			 * Darwin had code here to zero-extend.
1713 			 * Tests indicate the server will zero-fill,
1714 			 * so looks like we don't need to do that.
1715 			 */
1716 			mutex_enter(&np->r_statelock);
1717 			np->r_size = vap->va_size;
1718 			np->n_flag |= (NFLUSHWIRE | NATTRCHANGED);
1719 			mutex_exit(&np->r_statelock);
1720 			modified = 1;
1721 		}
1722 	}
1723 
1724 	/*
1725 	 * Todo: Implement setting create_time (which is
1726 	 * different from ctime).
1727 	 */
1728 	mtime = ((mask & AT_MTIME) ? &vap->va_mtime : 0);
1729 	atime = ((mask & AT_ATIME) ? &vap->va_atime : 0);
1730 
1731 	if (dosattr || mtime || atime) {
1732 		/*
1733 		 * Always use the handle-based set attr call now.
1734 		 */
1735 		ASSERT(fid != NULL);
1736 		error = smbfs_smb_setfattr(smi->smi_share, fid,
1737 		    dosattr, mtime, atime, &scred);
1738 		if (error) {
1739 			SMBVDEBUG("set times error %d file %s\n",
1740 			    error, np->n_rpath);
1741 		} else {
1742 			modified = 1;
1743 		}
1744 	}
1745 
1746 out:
1747 	if (fid != NULL)
1748 		smbfs_smb_tmpclose(np, fid);
1749 
1750 	smb_credrele(&scred);
1751 
1752 	if (modified) {
1753 		/*
1754 		 * Invalidate attribute cache in case the server
1755 		 * doesn't set exactly the attributes we asked.
1756 		 */
1757 		smbfs_attrcache_remove(np);
1758 
1759 		/*
1760 		 * If changing the size of the file, invalidate
1761 		 * any local cached data which is no longer part
1762 		 * of the file.  We also possibly invalidate the
1763 		 * last page in the file.  We could use
1764 		 * pvn_vpzero(), but this would mark the page as
1765 		 * modified and require it to be written back to
1766 		 * the server for no particularly good reason.
1767 		 * This way, if we access it, then we bring it
1768 		 * back in.  A read should be cheaper than a
1769 		 * write.
1770 		 */
1771 		if (mask & AT_SIZE) {
1772 			smbfs_invalidate_pages(vp,
1773 			    (vap->va_size & PAGEMASK), cr);
1774 		}
1775 	}
1776 
1777 	return (error);
1778 }
1779 
1780 /*
1781  * Helper function for extensible system attributes (PSARC 2007/315)
1782  * Compute the DOS attribute word to pass to _setfattr (see above).
1783  * This returns zero IFF no change is being made to attributes.
1784  * Otherwise return the new attributes or SMB_EFA_NORMAL.
1785  */
1786 static uint32_t
1787 xvattr_to_dosattr(smbnode_t *np, struct vattr *vap)
1788 {
1789 	xvattr_t *xvap = (xvattr_t *)vap;
1790 	xoptattr_t *xoap = NULL;
1791 	uint32_t attr = np->r_attr.fa_attr;
1792 	boolean_t anyset = B_FALSE;
1793 
1794 	if ((xoap = xva_getxoptattr(xvap)) == NULL)
1795 		return (0);
1796 
1797 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1798 		if (xoap->xoa_archive)
1799 			attr |= SMB_FA_ARCHIVE;
1800 		else
1801 			attr &= ~SMB_FA_ARCHIVE;
1802 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
1803 		anyset = B_TRUE;
1804 	}
1805 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1806 		if (xoap->xoa_system)
1807 			attr |= SMB_FA_SYSTEM;
1808 		else
1809 			attr &= ~SMB_FA_SYSTEM;
1810 		XVA_SET_RTN(xvap, XAT_SYSTEM);
1811 		anyset = B_TRUE;
1812 	}
1813 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1814 		if (xoap->xoa_readonly)
1815 			attr |= SMB_FA_RDONLY;
1816 		else
1817 			attr &= ~SMB_FA_RDONLY;
1818 		XVA_SET_RTN(xvap, XAT_READONLY);
1819 		anyset = B_TRUE;
1820 	}
1821 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1822 		if (xoap->xoa_hidden)
1823 			attr |= SMB_FA_HIDDEN;
1824 		else
1825 			attr &= ~SMB_FA_HIDDEN;
1826 		XVA_SET_RTN(xvap, XAT_HIDDEN);
1827 		anyset = B_TRUE;
1828 	}
1829 
1830 	if (anyset == B_FALSE)
1831 		return (0);	/* no change */
1832 	if (attr == 0)
1833 		attr = SMB_EFA_NORMAL;
1834 
1835 	return (attr);
1836 }
1837 
1838 /*
1839  * smbfs_access_rwx()
1840  * Common function for smbfs_access, etc.
1841  *
1842  * The security model implemented by the FS is unusual
1843  * due to the current "single user mounts" restriction:
1844  * All access under a given mount point uses the CIFS
1845  * credentials established by the owner of the mount.
1846  *
1847  * Most access checking is handled by the CIFS server,
1848  * but we need sufficient Unix access checks here to
1849  * prevent other local Unix users from having access
1850  * to objects under this mount that the uid/gid/mode
1851  * settings in the mount would not allow.
1852  *
1853  * With this model, there is a case where we need the
1854  * ability to do an access check before we have the
1855  * vnode for an object.  This function takes advantage
1856  * of the fact that the uid/gid/mode is per mount, and
1857  * avoids the need for a vnode.
1858  *
1859  * We still (sort of) need a vnode when we call
1860  * secpolicy_vnode_access, but that only uses
1861  * the vtype field, so we can use a pair of fake
1862  * vnodes that have only v_type filled in.
1863  */
1864 static int
1865 smbfs_access_rwx(vfs_t *vfsp, int vtype, int mode, cred_t *cr)
1866 {
1867 	/* See the secpolicy call below. */
1868 	static const vnode_t tmpl_vdir = { .v_type = VDIR };
1869 	static const vnode_t tmpl_vreg = { .v_type = VREG };
1870 	vattr_t		va;
1871 	vnode_t		*tvp;
1872 	struct smbmntinfo *smi = VFTOSMI(vfsp);
1873 	int shift = 0;
1874 
1875 	/*
1876 	 * Build our (fabricated) vnode attributes.
1877 	 */
1878 	bzero(&va, sizeof (va));
1879 	va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
1880 	va.va_type = vtype;
1881 	va.va_mode = (vtype == VDIR) ?
1882 	    smi->smi_dmode : smi->smi_fmode;
1883 	va.va_uid = smi->smi_uid;
1884 	va.va_gid = smi->smi_gid;
1885 
1886 	/*
1887 	 * Disallow write attempts on read-only file systems,
1888 	 * unless the file is a device or fifo node.  Note:
1889 	 * Inline vn_is_readonly and IS_DEVVP here because
1890 	 * we may not have a vnode ptr.  Original expr. was:
1891 	 * (mode & VWRITE) && vn_is_readonly(vp) && !IS_DEVVP(vp))
1892 	 */
1893 	if ((mode & VWRITE) &&
1894 	    (vfsp->vfs_flag & VFS_RDONLY) &&
1895 	    !(vtype == VCHR || vtype == VBLK || vtype == VFIFO))
1896 		return (EROFS);
1897 
1898 	/*
1899 	 * Disallow attempts to access mandatory lock files.
1900 	 * Similarly, expand MANDLOCK here.
1901 	 */
1902 	if ((mode & (VWRITE | VREAD | VEXEC)) &&
1903 	    va.va_type == VREG && MANDMODE(va.va_mode))
1904 		return (EACCES);
1905 
1906 	/*
1907 	 * Access check is based on only
1908 	 * one of owner, group, public.
1909 	 * If not owner, then check group.
1910 	 * If not a member of the group,
1911 	 * then check public access.
1912 	 */
1913 	if (crgetuid(cr) != va.va_uid) {
1914 		shift += 3;
1915 		if (!groupmember(va.va_gid, cr))
1916 			shift += 3;
1917 	}
1918 
1919 	/*
1920 	 * We need a vnode for secpolicy_vnode_access,
1921 	 * but the only thing it looks at is v_type,
1922 	 * so pass one of the templates above.
1923 	 */
1924 	tvp = (va.va_type == VDIR) ?
1925 	    (vnode_t *)&tmpl_vdir :
1926 	    (vnode_t *)&tmpl_vreg;
1927 
1928 	return (secpolicy_vnode_access2(cr, tvp, va.va_uid,
1929 	    va.va_mode << shift, mode));
1930 }
1931 
1932 /*
1933  * See smbfs_setattr
1934  */
1935 static int
1936 smbfs_accessx(void *arg, int mode, cred_t *cr)
1937 {
1938 	vnode_t *vp = arg;
1939 	/*
1940 	 * Note: The caller has checked the current zone,
1941 	 * the SMI_DEAD and VFS_UNMOUNTED flags, etc.
1942 	 */
1943 	return (smbfs_access_rwx(vp->v_vfsp, vp->v_type, mode, cr));
1944 }
1945 
1946 /*
1947  * XXX
1948  * This op should support PSARC 2007/403, Modified Access Checks for CIFS
1949  */
1950 /* ARGSUSED */
1951 static int
1952 smbfs_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
1953 {
1954 	vfs_t		*vfsp;
1955 	smbmntinfo_t	*smi;
1956 
1957 	vfsp = vp->v_vfsp;
1958 	smi = VFTOSMI(vfsp);
1959 
1960 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
1961 		return (EIO);
1962 
1963 	if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
1964 		return (EIO);
1965 
1966 	return (smbfs_access_rwx(vfsp, vp->v_type, mode, cr));
1967 }
1968 
1969 
1970 /* ARGSUSED */
1971 static int
1972 smbfs_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct)
1973 {
1974 	/* Not yet... */
1975 	return (ENOSYS);
1976 }
1977 
1978 
1979 /*
1980  * Flush local dirty pages to stable storage on the server.
1981  *
1982  * If FNODSYNC is specified, then there is nothing to do because
1983  * metadata changes are not cached on the client before being
1984  * sent to the server.
1985  */
1986 /* ARGSUSED */
1987 static int
1988 smbfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
1989 {
1990 	int		error = 0;
1991 	smbmntinfo_t	*smi;
1992 	smbnode_t	*np;
1993 	struct smb_cred scred;
1994 
1995 	np = VTOSMB(vp);
1996 	smi = VTOSMI(vp);
1997 
1998 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
1999 		return (EIO);
2000 
2001 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
2002 		return (EIO);
2003 
2004 	if ((syncflag & FNODSYNC) || IS_SWAPVP(vp))
2005 		return (0);
2006 
2007 	if ((syncflag & (FSYNC|FDSYNC)) == 0)
2008 		return (0);
2009 
2010 	error = smbfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
2011 	if (error)
2012 		return (error);
2013 
2014 	/* Shared lock for n_fid use in _flush */
2015 	if (smbfs_rw_enter_sig(&np->r_lkserlock, RW_READER, SMBINTR(vp)))
2016 		return (EINTR);
2017 	smb_credinit(&scred, cr);
2018 
2019 	error = smbfsflush(np, &scred);
2020 
2021 	smb_credrele(&scred);
2022 	smbfs_rw_exit(&np->r_lkserlock);
2023 
2024 	return (error);
2025 }
2026 
2027 static int
2028 smbfsflush(smbnode_t *np, struct smb_cred *scrp)
2029 {
2030 	struct smb_share *ssp = np->n_mount->smi_share;
2031 	smb_fh_t *fhp;
2032 	int error;
2033 
2034 	/* Shared lock for n_fid use below. */
2035 	ASSERT(smbfs_rw_lock_held(&np->r_lkserlock, RW_READER));
2036 
2037 	if (!(np->n_flag & NFLUSHWIRE))
2038 		return (0);
2039 	if (np->n_fidrefs == 0)
2040 		return (0); /* not open */
2041 	if ((fhp = np->n_fid) == NULL)
2042 		return (0);
2043 
2044 	/* After reconnect, n_fid is invalid */
2045 	if (fhp->fh_vcgenid != ssp->ss_vcgenid)
2046 		return (ESTALE);
2047 
2048 	error = smbfs_smb_flush(ssp, fhp, scrp);
2049 
2050 	if (!error) {
2051 		mutex_enter(&np->r_statelock);
2052 		np->n_flag &= ~NFLUSHWIRE;
2053 		mutex_exit(&np->r_statelock);
2054 	}
2055 	return (error);
2056 }
2057 
2058 /*
2059  * Last reference to vnode went away.
2060  */
2061 /* ARGSUSED */
2062 static void
2063 smbfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
2064 {
2065 	struct smb_cred scred;
2066 	smbnode_t	*np = VTOSMB(vp);
2067 	int error;
2068 
2069 	/*
2070 	 * Don't "bail out" for VFS_UNMOUNTED here,
2071 	 * as we want to do cleanup, etc.
2072 	 * See also pcfs_inactive
2073 	 */
2074 
2075 	/*
2076 	 * If this is coming from the wrong zone, we let someone in the right
2077 	 * zone take care of it asynchronously.  We can get here due to
2078 	 * VN_RELE() being called from pageout() or fsflush().  This call may
2079 	 * potentially turn into an expensive no-op if, for instance, v_count
2080 	 * gets incremented in the meantime, but it's still correct.
2081 	 */
2082 
2083 	/*
2084 	 * From NFS:rinactive()
2085 	 *
2086 	 * Before freeing anything, wait until all asynchronous
2087 	 * activity is done on this rnode.  This will allow all
2088 	 * asynchronous read ahead and write behind i/o's to
2089 	 * finish.
2090 	 */
2091 	mutex_enter(&np->r_statelock);
2092 	while (np->r_count > 0)
2093 		cv_wait(&np->r_cv, &np->r_statelock);
2094 	mutex_exit(&np->r_statelock);
2095 
2096 	/*
2097 	 * Flush and invalidate all pages associated with the vnode.
2098 	 */
2099 	if (vn_has_cached_data(vp)) {
2100 		if ((np->r_flags & RDIRTY) && !np->r_error) {
2101 			error = smbfs_putpage(vp, (u_offset_t)0, 0, 0, cr, ct);
2102 			if (error && (error == ENOSPC || error == EDQUOT)) {
2103 				mutex_enter(&np->r_statelock);
2104 				if (!np->r_error)
2105 					np->r_error = error;
2106 				mutex_exit(&np->r_statelock);
2107 			}
2108 		}
2109 		smbfs_invalidate_pages(vp, (u_offset_t)0, cr);
2110 	}
2111 	/*
2112 	 * This vnode should have lost all cached data.
2113 	 */
2114 	ASSERT(vn_has_cached_data(vp) == 0);
2115 
2116 	/*
2117 	 * Defend against the possibility that higher-level callers
2118 	 * might not correctly balance open and close calls.  If we
2119 	 * get here with open references remaining, it means there
2120 	 * was a missing VOP_CLOSE somewhere.  If that happens, do
2121 	 * the close here so we don't "leak" FIDs on the server.
2122 	 *
2123 	 * Exclusive lock for modifying n_fid stuff.
2124 	 * Don't want this one ever interruptible.
2125 	 */
2126 	(void) smbfs_rw_enter_sig(&np->r_lkserlock, RW_WRITER, 0);
2127 	smb_credinit(&scred, cr);
2128 
2129 	switch (np->n_ovtype) {
2130 	case VNON:
2131 		/* not open (OK) */
2132 		break;
2133 
2134 	case VDIR:
2135 		if (np->n_dirrefs == 0)
2136 			break;
2137 		SMBVDEBUG("open dir: refs %d path %s\n",
2138 		    np->n_dirrefs, np->n_rpath);
2139 		/* Force last close. */
2140 		np->n_dirrefs = 1;
2141 		smbfs_rele_fid(np, &scred);
2142 		break;
2143 
2144 	case VREG:
2145 		if (np->n_fidrefs == 0)
2146 			break;
2147 		SMBVDEBUG("open file: refs %d path %s\n",
2148 		    np->n_fidrefs, np->n_rpath);
2149 		/* Force last close. */
2150 		np->n_fidrefs = 1;
2151 		smbfs_rele_fid(np, &scred);
2152 		break;
2153 
2154 	default:
2155 		SMBVDEBUG("bad n_ovtype %d\n", np->n_ovtype);
2156 		np->n_ovtype = VNON;
2157 		break;
2158 	}
2159 
2160 	smb_credrele(&scred);
2161 	smbfs_rw_exit(&np->r_lkserlock);
2162 
2163 	/*
2164 	 * XATTR directories (and the files under them) have
2165 	 * little value for reclaim, so just remove them from
2166 	 * the "hash" (AVL) as soon as they go inactive.
2167 	 * Note that the node may already have been removed
2168 	 * from the hash by smbfsremove.
2169 	 */
2170 	if ((np->n_flag & N_XATTR) != 0 &&
2171 	    (np->r_flags & RHASHED) != 0)
2172 		smbfs_rmhash(np);
2173 
2174 	smbfs_addfree(np);
2175 }
2176 
2177 /*
2178  * Remote file system operations having to do with directory manipulation.
2179  */
2180 /* ARGSUSED */
2181 static int
2182 smbfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
2183 	int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
2184 	int *direntflags, pathname_t *realpnp)
2185 {
2186 	vfs_t		*vfs;
2187 	smbmntinfo_t	*smi;
2188 	smbnode_t	*dnp;
2189 	int		error;
2190 
2191 	vfs = dvp->v_vfsp;
2192 	smi = VFTOSMI(vfs);
2193 
2194 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
2195 		return (EPERM);
2196 
2197 	if (smi->smi_flags & SMI_DEAD || vfs->vfs_flag & VFS_UNMOUNTED)
2198 		return (EIO);
2199 
2200 	dnp = VTOSMB(dvp);
2201 
2202 	/*
2203 	 * Are we looking up extended attributes?  If so, "dvp" is
2204 	 * the file or directory for which we want attributes, and
2205 	 * we need a lookup of the (faked up) attribute directory
2206 	 * before we lookup the rest of the path.
2207 	 */
2208 	if (flags & LOOKUP_XATTR) {
2209 		/*
2210 		 * Require the xattr mount option.
2211 		 */
2212 		if ((vfs->vfs_flag & VFS_XATTR) == 0)
2213 			return (EINVAL);
2214 
2215 		error = smbfs_get_xattrdir(dvp, vpp, cr, flags);
2216 		return (error);
2217 	}
2218 
2219 	if (smbfs_rw_enter_sig(&dnp->r_rwlock, RW_READER, SMBINTR(dvp)))
2220 		return (EINTR);
2221 
2222 	error = smbfslookup(dvp, nm, vpp, cr, 1, ct);
2223 
2224 	smbfs_rw_exit(&dnp->r_rwlock);
2225 
2226 	/*
2227 	 * If the caller passes an invalid name here, we'll have
2228 	 * error == EINVAL but want to return ENOENT.  This is
2229 	 * common with things like "ls foo*" with no matches.
2230 	 */
2231 	if (error == EINVAL)
2232 		error = ENOENT;
2233 
2234 	return (error);
2235 }
2236 
2237 /* ARGSUSED */
2238 static int
2239 smbfslookup(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr,
2240 	int cache_ok, caller_context_t *ct)
2241 {
2242 	int		error;
2243 	int		supplen; /* supported length */
2244 	vnode_t		*vp;
2245 	smbnode_t	*np;
2246 	smbnode_t	*dnp;
2247 	smbmntinfo_t	*smi;
2248 	/* struct smb_vc	*vcp; */
2249 	const char	*ill;
2250 	const char	*name = (const char *)nm;
2251 	int		nmlen = strlen(nm);
2252 	int		rplen;
2253 	struct smb_cred scred;
2254 	struct smbfattr fa;
2255 
2256 	smi = VTOSMI(dvp);
2257 	dnp = VTOSMB(dvp);
2258 
2259 	ASSERT(curproc->p_zone == smi->smi_zone_ref.zref_zone);
2260 
2261 	supplen = 255;
2262 
2263 	/*
2264 	 * RWlock must be held, either reader or writer.
2265 	 */
2266 	ASSERT(dnp->r_rwlock.count != 0);
2267 
2268 	/*
2269 	 * If lookup is for "", just return dvp.
2270 	 * No need to perform any access checks.
2271 	 */
2272 	if (nmlen == 0) {
2273 		VN_HOLD(dvp);
2274 		*vpp = dvp;
2275 		return (0);
2276 	}
2277 
2278 	/*
2279 	 * Can't do lookups in non-directories.
2280 	 */
2281 	if (dvp->v_type != VDIR)
2282 		return (ENOTDIR);
2283 
2284 	/*
2285 	 * Need search permission in the directory.
2286 	 */
2287 	error = smbfs_access(dvp, VEXEC, 0, cr, ct);
2288 	if (error)
2289 		return (error);
2290 
2291 	/*
2292 	 * If lookup is for ".", just return dvp.
2293 	 * Access check was done above.
2294 	 */
2295 	if (nmlen == 1 && name[0] == '.') {
2296 		VN_HOLD(dvp);
2297 		*vpp = dvp;
2298 		return (0);
2299 	}
2300 
2301 	/*
2302 	 * Now some sanity checks on the name.
2303 	 * First check the length.
2304 	 */
2305 	if (nmlen > supplen)
2306 		return (ENAMETOOLONG);
2307 
2308 	/*
2309 	 * Avoid surprises with characters that are
2310 	 * illegal in Windows file names.
2311 	 * Todo: CATIA mappings?
2312 	 */
2313 	ill = illegal_chars;
2314 	if (dnp->n_flag & N_XATTR)
2315 		ill++; /* allow colon */
2316 	if (strpbrk(nm, ill))
2317 		return (EINVAL);
2318 
2319 	/*
2320 	 * Special handling for lookup of ".."
2321 	 *
2322 	 * We keep full pathnames (as seen on the server)
2323 	 * so we can just trim off the last component to
2324 	 * get the full pathname of the parent.  Note:
2325 	 * We don't actually copy and modify, but just
2326 	 * compute the trimmed length and pass that with
2327 	 * the current dir path (not null terminated).
2328 	 *
2329 	 * We don't go over-the-wire to get attributes
2330 	 * for ".." because we know it's a directory,
2331 	 * and we can just leave the rest "stale"
2332 	 * until someone does a getattr.
2333 	 */
2334 	if (nmlen == 2 && name[0] == '.' && name[1] == '.') {
2335 		if (dvp->v_flag & VROOT) {
2336 			/*
2337 			 * Already at the root.  This can happen
2338 			 * with directory listings at the root,
2339 			 * which lookup "." and ".." to get the
2340 			 * inode numbers.  Let ".." be the same
2341 			 * as "." in the FS root.
2342 			 */
2343 			VN_HOLD(dvp);
2344 			*vpp = dvp;
2345 			return (0);
2346 		}
2347 
2348 		/*
2349 		 * Special case for XATTR directory
2350 		 */
2351 		if (dvp->v_flag & V_XATTRDIR) {
2352 			error = smbfs_xa_parent(dvp, vpp);
2353 			return (error);
2354 		}
2355 
2356 		/*
2357 		 * Find the parent path length.
2358 		 */
2359 		rplen = dnp->n_rplen;
2360 		ASSERT(rplen > 0);
2361 		while (--rplen >= 0) {
2362 			if (dnp->n_rpath[rplen] == '\\')
2363 				break;
2364 		}
2365 		if (rplen <= 0) {
2366 			/* Found our way to the root. */
2367 			vp = SMBTOV(smi->smi_root);
2368 			VN_HOLD(vp);
2369 			*vpp = vp;
2370 			return (0);
2371 		}
2372 		np = smbfs_node_findcreate(smi,
2373 		    dnp->n_rpath, rplen, NULL, 0, 0,
2374 		    &smbfs_fattr0); /* force create */
2375 		ASSERT(np != NULL);
2376 		vp = SMBTOV(np);
2377 		vp->v_type = VDIR;
2378 
2379 		/* Success! */
2380 		*vpp = vp;
2381 		return (0);
2382 	}
2383 
2384 	/*
2385 	 * Normal lookup of a name under this directory.
2386 	 * Note we handled "", ".", ".." above.
2387 	 */
2388 	if (cache_ok) {
2389 		/*
2390 		 * The caller indicated that it's OK to use a
2391 		 * cached result for this lookup, so try to
2392 		 * reclaim a node from the smbfs node cache.
2393 		 */
2394 		error = smbfslookup_cache(dvp, nm, nmlen, &vp, cr);
2395 		if (error)
2396 			return (error);
2397 		if (vp != NULL) {
2398 			/* hold taken in lookup_cache */
2399 			*vpp = vp;
2400 			return (0);
2401 		}
2402 	}
2403 
2404 	/*
2405 	 * OK, go over-the-wire to get the attributes,
2406 	 * then create the node.
2407 	 */
2408 	smb_credinit(&scred, cr);
2409 	/* Note: this can allocate a new "name" */
2410 	error = smbfs_smb_lookup(dnp, &name, &nmlen, &fa, &scred);
2411 	smb_credrele(&scred);
2412 	if (error == ENOTDIR) {
2413 		/*
2414 		 * Lookup failed because this directory was
2415 		 * removed or renamed by another client.
2416 		 * Remove any cached attributes under it.
2417 		 */
2418 		smbfs_attrcache_remove(dnp);
2419 		smbfs_attrcache_prune(dnp);
2420 	}
2421 	if (error)
2422 		goto out;
2423 
2424 	error = smbfs_nget(dvp, name, nmlen, &fa, &vp);
2425 	if (error)
2426 		goto out;
2427 
2428 	/* Success! */
2429 	*vpp = vp;
2430 
2431 out:
2432 	/* smbfs_smb_lookup may have allocated name. */
2433 	if (name != nm)
2434 		smbfs_name_free(name, nmlen);
2435 
2436 	return (error);
2437 }
2438 
2439 /*
2440  * smbfslookup_cache
2441  *
2442  * Try to reclaim a node from the smbfs node cache.
2443  * Some statistics for DEBUG.
2444  *
2445  * This mechanism lets us avoid many of the five (or more)
2446  * OtW lookup calls per file seen with "ls -l" if we search
2447  * the smbfs node cache for recently inactive(ated) nodes.
2448  */
2449 #ifdef DEBUG
2450 int smbfs_lookup_cache_calls = 0;
2451 int smbfs_lookup_cache_error = 0;
2452 int smbfs_lookup_cache_miss = 0;
2453 int smbfs_lookup_cache_stale = 0;
2454 int smbfs_lookup_cache_hits = 0;
2455 #endif /* DEBUG */
2456 
2457 /* ARGSUSED */
2458 static int
2459 smbfslookup_cache(vnode_t *dvp, char *nm, int nmlen,
2460 	vnode_t **vpp, cred_t *cr)
2461 {
2462 	struct vattr va;
2463 	smbnode_t *dnp;
2464 	smbnode_t *np;
2465 	vnode_t *vp;
2466 	int error;
2467 	char sep;
2468 
2469 	dnp = VTOSMB(dvp);
2470 	*vpp = NULL;
2471 
2472 #ifdef DEBUG
2473 	smbfs_lookup_cache_calls++;
2474 #endif
2475 
2476 	/*
2477 	 * First make sure we can get attributes for the
2478 	 * directory.  Cached attributes are OK here.
2479 	 * If we removed or renamed the directory, this
2480 	 * will return ENOENT.  If someone else removed
2481 	 * this directory or file, we'll find out when we
2482 	 * try to open or get attributes.
2483 	 */
2484 	va.va_mask = AT_TYPE | AT_MODE;
2485 	error = smbfsgetattr(dvp, &va, cr);
2486 	if (error) {
2487 #ifdef DEBUG
2488 		smbfs_lookup_cache_error++;
2489 #endif
2490 		return (error);
2491 	}
2492 
2493 	/*
2494 	 * Passing NULL smbfattr here so we will
2495 	 * just look, not create.
2496 	 */
2497 	sep = SMBFS_DNP_SEP(dnp);
2498 	np = smbfs_node_findcreate(dnp->n_mount,
2499 	    dnp->n_rpath, dnp->n_rplen,
2500 	    nm, nmlen, sep, NULL);
2501 	if (np == NULL) {
2502 #ifdef DEBUG
2503 		smbfs_lookup_cache_miss++;
2504 #endif
2505 		return (0);
2506 	}
2507 
2508 	/*
2509 	 * Found it.  Attributes still valid?
2510 	 */
2511 	vp = SMBTOV(np);
2512 	if (np->r_attrtime <= gethrtime()) {
2513 		/* stale */
2514 #ifdef DEBUG
2515 		smbfs_lookup_cache_stale++;
2516 #endif
2517 		VN_RELE(vp);
2518 		return (0);
2519 	}
2520 
2521 	/*
2522 	 * Success!
2523 	 * Caller gets hold from smbfs_node_findcreate
2524 	 */
2525 #ifdef DEBUG
2526 	smbfs_lookup_cache_hits++;
2527 #endif
2528 	*vpp = vp;
2529 	return (0);
2530 }
2531 
2532 
2533 /*
2534  * XXX
2535  * vsecattr_t is new to build 77, and we need to eventually support
2536  * it in order to create an ACL when an object is created.
2537  *
2538  * This op should support the new FIGNORECASE flag for case-insensitive
2539  * lookups, per PSARC 2007/244.
2540  */
2541 /* ARGSUSED */
2542 static int
2543 smbfs_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive,
2544 	int mode, vnode_t **vpp, cred_t *cr, int lfaware, caller_context_t *ct,
2545 	vsecattr_t *vsecp)
2546 {
2547 	int		error;
2548 	vfs_t		*vfsp;
2549 	vnode_t		*vp;
2550 	smbnode_t	*np;
2551 	smbnode_t	*dnp;
2552 	smbmntinfo_t	*smi;
2553 	struct vattr	vattr;
2554 	struct smbfattr	fattr;
2555 	struct smb_cred	scred;
2556 	const char *name = (const char *)nm;
2557 	int		nmlen = strlen(nm);
2558 	uint32_t	disp;
2559 	smb_fh_t	*fid = NULL;
2560 	int		xattr;
2561 
2562 	vfsp = dvp->v_vfsp;
2563 	smi = VFTOSMI(vfsp);
2564 	dnp = VTOSMB(dvp);
2565 	vp = NULL;
2566 
2567 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
2568 		return (EPERM);
2569 
2570 	if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
2571 		return (EIO);
2572 
2573 	/*
2574 	 * Note: this may break mknod(2) calls to create a directory,
2575 	 * but that's obscure use.  Some other filesystems do this.
2576 	 * Todo: redirect VDIR type here to _mkdir.
2577 	 */
2578 	if (va->va_type != VREG)
2579 		return (EINVAL);
2580 
2581 	/*
2582 	 * If the pathname is "", just use dvp, no checks.
2583 	 * Do this outside of the rwlock (like zfs).
2584 	 */
2585 	if (nmlen == 0) {
2586 		VN_HOLD(dvp);
2587 		*vpp = dvp;
2588 		return (0);
2589 	}
2590 
2591 	/* Don't allow "." or ".." through here. */
2592 	if ((nmlen == 1 && name[0] == '.') ||
2593 	    (nmlen == 2 && name[0] == '.' && name[1] == '.'))
2594 		return (EISDIR);
2595 
2596 	/*
2597 	 * We make a copy of the attributes because the caller does not
2598 	 * expect us to change what va points to.
2599 	 */
2600 	vattr = *va;
2601 
2602 	if (smbfs_rw_enter_sig(&dnp->r_rwlock, RW_WRITER, SMBINTR(dvp)))
2603 		return (EINTR);
2604 	smb_credinit(&scred, cr);
2605 
2606 	/*
2607 	 * NFS needs to go over the wire, just to be sure whether the
2608 	 * file exists or not.  Using a cached result is dangerous in
2609 	 * this case when making a decision regarding existence.
2610 	 *
2611 	 * The SMB protocol does NOT really need to go OTW here
2612 	 * thanks to the expressive NTCREATE disposition values.
2613 	 * Unfortunately, to do Unix access checks correctly,
2614 	 * we need to know if the object already exists.
2615 	 * When the object does not exist, we need VWRITE on
2616 	 * the directory.  Note: smbfslookup() checks VEXEC.
2617 	 */
2618 	error = smbfslookup(dvp, nm, &vp, cr, 0, ct);
2619 	if (error == 0) {
2620 		/*
2621 		 * The file already exists.  Error?
2622 		 * NB: have a hold from smbfslookup
2623 		 */
2624 		if (exclusive == EXCL) {
2625 			error = EEXIST;
2626 			VN_RELE(vp);
2627 			goto out;
2628 		}
2629 		/*
2630 		 * Verify requested access.
2631 		 */
2632 		error = smbfs_access(vp, mode, 0, cr, ct);
2633 		if (error) {
2634 			VN_RELE(vp);
2635 			goto out;
2636 		}
2637 
2638 		/*
2639 		 * Truncate (if requested).
2640 		 */
2641 		if ((vattr.va_mask & AT_SIZE) && vp->v_type == VREG) {
2642 			np = VTOSMB(vp);
2643 			/*
2644 			 * Check here for large file truncation by
2645 			 * LF-unaware process, like ufs_create().
2646 			 */
2647 			if (!(lfaware & FOFFMAX)) {
2648 				mutex_enter(&np->r_statelock);
2649 				if (np->r_size > MAXOFF32_T)
2650 					error = EOVERFLOW;
2651 				mutex_exit(&np->r_statelock);
2652 			}
2653 			if (error) {
2654 				VN_RELE(vp);
2655 				goto out;
2656 			}
2657 			vattr.va_mask = AT_SIZE;
2658 			error = smbfssetattr(vp, &vattr, 0, cr);
2659 			if (error) {
2660 				VN_RELE(vp);
2661 				goto out;
2662 			}
2663 #ifdef	SMBFS_VNEVENT
2664 			/* Existing file was truncated */
2665 			vnevent_create(vp, ct);
2666 #endif
2667 			/* invalidate pages done in smbfssetattr() */
2668 		}
2669 		/* Success! */
2670 		*vpp = vp;
2671 		goto out;
2672 	}
2673 
2674 	/*
2675 	 * The file did not exist.  Need VWRITE in the directory.
2676 	 */
2677 	error = smbfs_access(dvp, VWRITE, 0, cr, ct);
2678 	if (error)
2679 		goto out;
2680 
2681 	/*
2682 	 * Now things get tricky.  We also need to check the
2683 	 * requested open mode against the file we may create.
2684 	 * See comments at smbfs_access_rwx
2685 	 */
2686 	error = smbfs_access_rwx(vfsp, VREG, mode, cr);
2687 	if (error)
2688 		goto out;
2689 
2690 	/*
2691 	 * Now the code derived from Darwin,
2692 	 * but with greater use of NT_CREATE
2693 	 * disposition options.  Much changed.
2694 	 *
2695 	 * Create (or open) a new child node.
2696 	 * Note we handled "." and ".." above.
2697 	 */
2698 
2699 	if (exclusive == EXCL)
2700 		disp = NTCREATEX_DISP_CREATE;
2701 	else {
2702 		/* Truncate regular files if requested. */
2703 		if ((va->va_type == VREG) &&
2704 		    (va->va_mask & AT_SIZE) &&
2705 		    (va->va_size == 0))
2706 			disp = NTCREATEX_DISP_OVERWRITE_IF;
2707 		else
2708 			disp = NTCREATEX_DISP_OPEN_IF;
2709 	}
2710 	xattr = (dnp->n_flag & N_XATTR) ? 1 : 0;
2711 	error = smbfs_smb_create(dnp,
2712 	    name, nmlen, xattr,
2713 	    disp, &scred, &fid);
2714 	if (error)
2715 		goto out;
2716 
2717 	/*
2718 	 * Should use the fid to get/set the size
2719 	 * while we have it opened here.  See above.
2720 	 */
2721 	smbfs_smb_close(fid);
2722 
2723 	/*
2724 	 * In the open case, the name may differ a little
2725 	 * from what we passed to create (case, etc.)
2726 	 * so call lookup to get the (opened) name.
2727 	 *
2728 	 * XXX: Could avoid this extra lookup if the
2729 	 * "createact" result from NT_CREATE says we
2730 	 * created the object.
2731 	 */
2732 	error = smbfs_smb_lookup(dnp, &name, &nmlen, &fattr, &scred);
2733 	if (error)
2734 		goto out;
2735 
2736 	/* update attr and directory cache */
2737 	smbfs_attr_touchdir(dnp);
2738 
2739 	error = smbfs_nget(dvp, name, nmlen, &fattr, &vp);
2740 	if (error)
2741 		goto out;
2742 
2743 	/* Success! */
2744 	*vpp = vp;
2745 	error = 0;
2746 
2747 out:
2748 	smb_credrele(&scred);
2749 	smbfs_rw_exit(&dnp->r_rwlock);
2750 	if (name != nm)
2751 		smbfs_name_free(name, nmlen);
2752 	return (error);
2753 }
2754 
2755 /*
2756  * XXX
2757  * This op should support the new FIGNORECASE flag for case-insensitive
2758  * lookups, per PSARC 2007/244.
2759  */
2760 /* ARGSUSED */
2761 static int
2762 smbfs_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct,
2763 	int flags)
2764 {
2765 	struct smb_cred	scred;
2766 	vnode_t		*vp = NULL;
2767 	smbnode_t	*dnp = VTOSMB(dvp);
2768 	smbmntinfo_t	*smi = VTOSMI(dvp);
2769 	int		error;
2770 
2771 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
2772 		return (EPERM);
2773 
2774 	if (smi->smi_flags & SMI_DEAD || dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
2775 		return (EIO);
2776 
2777 	/*
2778 	 * Verify access to the dirctory.
2779 	 */
2780 	error = smbfs_access(dvp, VWRITE|VEXEC, 0, cr, ct);
2781 	if (error)
2782 		return (error);
2783 
2784 	if (smbfs_rw_enter_sig(&dnp->r_rwlock, RW_WRITER, SMBINTR(dvp)))
2785 		return (EINTR);
2786 	smb_credinit(&scred, cr);
2787 
2788 	/* Lookup the file to remove. */
2789 	error = smbfslookup(dvp, nm, &vp, cr, 0, ct);
2790 	if (error != 0)
2791 		goto out;
2792 
2793 	/* Don't allow unlink of a directory. */
2794 	if (vp->v_type == VDIR) {
2795 		error = EPERM;
2796 		goto out;
2797 	}
2798 
2799 	/*
2800 	 * Do the real remove work
2801 	 */
2802 	error = smbfsremove(dvp, vp, &scred, flags);
2803 	if (error != 0)
2804 		goto out;
2805 
2806 #ifdef	SMBFS_VNEVENT
2807 	vnevent_remove(vp, dvp, nm, ct);
2808 #endif
2809 
2810 out:
2811 	if (vp != NULL)
2812 		VN_RELE(vp);
2813 
2814 	smb_credrele(&scred);
2815 	smbfs_rw_exit(&dnp->r_rwlock);
2816 
2817 	return (error);
2818 }
2819 
2820 /*
2821  * smbfsremove does the real work of removing in SMBFS
2822  * Caller has done dir access checks etc.
2823  *
2824  * The normal way to delete a file over SMB is open it (with DELETE access),
2825  * set the "delete-on-close" flag, and close the file.  The problem for Unix
2826  * applications is that they expect the file name to be gone once the unlink
2827  * completes, and the SMB server does not actually delete the file until ALL
2828  * opens of that file are closed.  We can't assume our open handles are the
2829  * only open handles on a file we're deleting, so to be safe we'll try to
2830  * rename the file to a temporary name and then set delete-on-close.  If we
2831  * fail to set delete-on-close (i.e. because other opens prevent it) then
2832  * undo the changes we made and give up with EBUSY.  Note that we might have
2833  * permission to delete a file but lack permission to rename, so we want to
2834  * continue in cases where rename fails.  As an optimization, only do the
2835  * rename when we have the file open.
2836  *
2837  * This is similar to what NFS does when deleting a file that has local opens,
2838  * but thanks to SMB delete-on-close, we don't need to keep track of when the
2839  * last local open goes away and send a delete.  The server does that for us.
2840  */
2841 /* ARGSUSED */
2842 static int
2843 smbfsremove(vnode_t *dvp, vnode_t *vp, struct smb_cred *scred,
2844     int flags)
2845 {
2846 	smbnode_t	*dnp = VTOSMB(dvp);
2847 	smbnode_t	*np = VTOSMB(vp);
2848 	smbmntinfo_t	*smi = np->n_mount;
2849 	char		*tmpname = NULL;
2850 	int		tnlen;
2851 	int		error;
2852 	smb_fh_t	*fid = NULL;
2853 	boolean_t	renamed = B_FALSE;
2854 
2855 	/*
2856 	 * The dvp RWlock must be held as writer.
2857 	 */
2858 	ASSERT(dnp->r_rwlock.owner == curthread);
2859 
2860 	/*
2861 	 * We need to flush any dirty pages which happen to
2862 	 * be hanging around before removing the file.  This
2863 	 * shouldn't happen very often and mostly on file
2864 	 * systems mounted "nocto".
2865 	 */
2866 	if (vn_has_cached_data(vp) &&
2867 	    ((np->r_flags & RDIRTY) || np->r_count > 0)) {
2868 		error = smbfs_putpage(vp, (offset_t)0, 0, 0,
2869 		    scred->scr_cred, NULL);
2870 		if (error && (error == ENOSPC || error == EDQUOT)) {
2871 			mutex_enter(&np->r_statelock);
2872 			if (!np->r_error)
2873 				np->r_error = error;
2874 			mutex_exit(&np->r_statelock);
2875 		}
2876 	}
2877 
2878 	/*
2879 	 * Get a file handle with delete access.
2880 	 * Close this FID before return.
2881 	 */
2882 	error = smbfs_smb_tmpopen(np, STD_RIGHT_DELETE_ACCESS,
2883 	    scred, &fid);
2884 	if (error) {
2885 		SMBVDEBUG("error %d opening %s\n",
2886 		    error, np->n_rpath);
2887 		goto out;
2888 	}
2889 	ASSERT(fid != NULL);
2890 
2891 	/*
2892 	 * If we have the file open, try to rename it to a temporary name.
2893 	 * If we can't rename, continue on and try setting DoC anyway.
2894 	 * Unnecessary for directories.
2895 	 */
2896 	if (vp->v_type != VDIR && vp->v_count > 1 && np->n_fidrefs > 0) {
2897 		tmpname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2898 		tnlen = smbfs_newname(tmpname, MAXNAMELEN);
2899 		error = smbfs_smb_rename(dnp, np, dnp, tmpname, tnlen,
2900 		    fid, scred);
2901 		if (error != 0) {
2902 			SMBVDEBUG("error %d renaming %s -> %s\n",
2903 			    error, np->n_rpath, tmpname);
2904 			/* Keep going without the rename. */
2905 		} else {
2906 			renamed = B_TRUE;
2907 		}
2908 	}
2909 
2910 	/*
2911 	 * Mark the file as delete-on-close.  If we can't,
2912 	 * undo what we did and err out.
2913 	 */
2914 	error = smbfs_smb_setdisp(smi->smi_share, fid, 1, scred);
2915 	if (error != 0) {
2916 		SMBVDEBUG("error %d setting DoC on %s\n",
2917 		    error, np->n_rpath);
2918 		/*
2919 		 * Failed to set DoC. If we renamed, undo that.
2920 		 * Need np->n_rpath relative to parent (dnp).
2921 		 * Use parent path name length plus one for
2922 		 * the separator ('/' or ':')
2923 		 */
2924 		if (renamed) {
2925 			char *oldname;
2926 			int oldnlen;
2927 			int err2;
2928 
2929 			oldname = np->n_rpath + (dnp->n_rplen + 1);
2930 			oldnlen = np->n_rplen - (dnp->n_rplen + 1);
2931 			err2 = smbfs_smb_rename(dnp, np, dnp, oldname, oldnlen,
2932 			    fid, scred);
2933 			SMBVDEBUG("error %d un-renaming %s -> %s\n",
2934 			    err2, tmpname, np->n_rpath);
2935 		}
2936 		error = EBUSY;
2937 		goto out;
2938 	}
2939 	/* Done! */
2940 	smbfs_attrcache_remove(np);
2941 	smbfs_attrcache_prune(np);
2942 
2943 out:
2944 	if (tmpname != NULL)
2945 		kmem_free(tmpname, MAXNAMELEN);
2946 	if (fid != NULL)
2947 		smbfs_smb_tmpclose(np, fid);
2948 
2949 	if (error == 0) {
2950 		/* Keep lookup from finding this node anymore. */
2951 		smbfs_rmhash(np);
2952 	}
2953 
2954 	return (error);
2955 }
2956 
2957 
2958 /* ARGSUSED */
2959 static int
2960 smbfs_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr,
2961 	caller_context_t *ct, int flags)
2962 {
2963 	/* Not yet... */
2964 	return (ENOSYS);
2965 }
2966 
2967 
2968 /*
2969  * XXX
2970  * This op should support the new FIGNORECASE flag for case-insensitive
2971  * lookups, per PSARC 2007/244.
2972  */
2973 /* ARGSUSED */
2974 static int
2975 smbfs_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2976 	caller_context_t *ct, int flags)
2977 {
2978 	struct smb_cred	scred;
2979 	smbnode_t	*odnp = VTOSMB(odvp);
2980 	smbnode_t	*ndnp = VTOSMB(ndvp);
2981 	vnode_t		*ovp;
2982 	int error;
2983 
2984 	if (curproc->p_zone != VTOSMI(odvp)->smi_zone_ref.zref_zone ||
2985 	    curproc->p_zone != VTOSMI(ndvp)->smi_zone_ref.zref_zone)
2986 		return (EPERM);
2987 
2988 	if (VTOSMI(odvp)->smi_flags & SMI_DEAD ||
2989 	    VTOSMI(ndvp)->smi_flags & SMI_DEAD ||
2990 	    odvp->v_vfsp->vfs_flag & VFS_UNMOUNTED ||
2991 	    ndvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
2992 		return (EIO);
2993 
2994 	if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 ||
2995 	    strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0)
2996 		return (EINVAL);
2997 
2998 	/*
2999 	 * Check that everything is on the same filesystem.
3000 	 * vn_rename checks the fsid's, but in case we don't
3001 	 * fill those in correctly, check here too.
3002 	 */
3003 	if (odvp->v_vfsp != ndvp->v_vfsp)
3004 		return (EXDEV);
3005 
3006 	/*
3007 	 * Need write access on source and target.
3008 	 * Server takes care of most checks.
3009 	 */
3010 	error = smbfs_access(odvp, VWRITE|VEXEC, 0, cr, ct);
3011 	if (error)
3012 		return (error);
3013 	if (odvp != ndvp) {
3014 		error = smbfs_access(ndvp, VWRITE, 0, cr, ct);
3015 		if (error)
3016 			return (error);
3017 	}
3018 
3019 	/*
3020 	 * Need to lock both old/new dirs as writer.
3021 	 *
3022 	 * Avoid deadlock here on old vs new directory nodes
3023 	 * by always taking the locks in order of address.
3024 	 * The order is arbitrary, but must be consistent.
3025 	 */
3026 	if (odnp < ndnp) {
3027 		if (smbfs_rw_enter_sig(&odnp->r_rwlock, RW_WRITER,
3028 		    SMBINTR(odvp)))
3029 			return (EINTR);
3030 		if (smbfs_rw_enter_sig(&ndnp->r_rwlock, RW_WRITER,
3031 		    SMBINTR(ndvp))) {
3032 			smbfs_rw_exit(&odnp->r_rwlock);
3033 			return (EINTR);
3034 		}
3035 	} else {
3036 		if (smbfs_rw_enter_sig(&ndnp->r_rwlock, RW_WRITER,
3037 		    SMBINTR(ndvp)))
3038 			return (EINTR);
3039 		if (smbfs_rw_enter_sig(&odnp->r_rwlock, RW_WRITER,
3040 		    SMBINTR(odvp))) {
3041 			smbfs_rw_exit(&ndnp->r_rwlock);
3042 			return (EINTR);
3043 		}
3044 	}
3045 	smb_credinit(&scred, cr);
3046 
3047 	/* Lookup the "old" name */
3048 	error = smbfslookup(odvp, onm, &ovp, cr, 0, ct);
3049 	if (error == 0) {
3050 		/*
3051 		 * Do the real rename work
3052 		 */
3053 		error = smbfsrename(odvp, ovp, ndvp, nnm, &scred, flags);
3054 		VN_RELE(ovp);
3055 	}
3056 
3057 	smb_credrele(&scred);
3058 	smbfs_rw_exit(&odnp->r_rwlock);
3059 	smbfs_rw_exit(&ndnp->r_rwlock);
3060 
3061 	return (error);
3062 }
3063 
3064 /*
3065  * smbfsrename does the real work of renaming in SMBFS
3066  * Caller has done dir access checks etc.
3067  */
3068 /* ARGSUSED */
3069 static int
3070 smbfsrename(vnode_t *odvp, vnode_t *ovp, vnode_t *ndvp, char *nnm,
3071     struct smb_cred *scred, int flags)
3072 {
3073 	smbnode_t	*odnp = VTOSMB(odvp);
3074 	smbnode_t	*onp = VTOSMB(ovp);
3075 	smbnode_t	*ndnp = VTOSMB(ndvp);
3076 	vnode_t		*nvp = NULL;
3077 	int		error;
3078 	int		nvp_locked = 0;
3079 	smb_fh_t	*fid = NULL;
3080 
3081 	/* Things our caller should have checked. */
3082 	ASSERT(curproc->p_zone == VTOSMI(odvp)->smi_zone_ref.zref_zone);
3083 	ASSERT(odvp->v_vfsp == ndvp->v_vfsp);
3084 	ASSERT(odnp->r_rwlock.owner == curthread);
3085 	ASSERT(ndnp->r_rwlock.owner == curthread);
3086 
3087 	/*
3088 	 * Lookup the target file.  If it exists, it needs to be
3089 	 * checked to see whether it is a mount point and whether
3090 	 * it is active (open).
3091 	 */
3092 	error = smbfslookup(ndvp, nnm, &nvp, scred->scr_cred, 0, NULL);
3093 	if (!error) {
3094 		/*
3095 		 * Target (nvp) already exists.  Check that it
3096 		 * has the same type as the source.  The server
3097 		 * will check this also, (and more reliably) but
3098 		 * this lets us return the correct error codes.
3099 		 */
3100 		if (ovp->v_type == VDIR) {
3101 			if (nvp->v_type != VDIR) {
3102 				error = ENOTDIR;
3103 				goto out;
3104 			}
3105 		} else {
3106 			if (nvp->v_type == VDIR) {
3107 				error = EISDIR;
3108 				goto out;
3109 			}
3110 		}
3111 
3112 		/*
3113 		 * POSIX dictates that when the source and target
3114 		 * entries refer to the same file object, rename
3115 		 * must do nothing and exit without error.
3116 		 */
3117 		if (ovp == nvp) {
3118 			error = 0;
3119 			goto out;
3120 		}
3121 
3122 		/*
3123 		 * Also must ensure the target is not a mount point,
3124 		 * and keep mount/umount away until we're done.
3125 		 */
3126 		if (vn_vfsrlock(nvp)) {
3127 			error = EBUSY;
3128 			goto out;
3129 		}
3130 		nvp_locked = 1;
3131 		if (vn_mountedvfs(nvp) != NULL) {
3132 			error = EBUSY;
3133 			goto out;
3134 		}
3135 
3136 		/*
3137 		 * CIFS may give a SHARING_VIOLATION error when
3138 		 * trying to rename onto an exising object,
3139 		 * so try to remove the target first.
3140 		 * (Only for files, not directories.)
3141 		 */
3142 		if (nvp->v_type == VDIR) {
3143 			error = EEXIST;
3144 			goto out;
3145 		}
3146 		error = smbfsremove(ndvp, nvp, scred, flags);
3147 		if (error != 0)
3148 			goto out;
3149 
3150 		/*
3151 		 * OK, removed the target file.  Continue as if
3152 		 * lookup target had failed (nvp == NULL).
3153 		 */
3154 		vn_vfsunlock(nvp);
3155 		nvp_locked = 0;
3156 		VN_RELE(nvp);
3157 		nvp = NULL;
3158 	} /* nvp */
3159 
3160 	/*
3161 	 * Get a file handle with delete access.
3162 	 * Close this FID before return.
3163 	 */
3164 	error = smbfs_smb_tmpopen(onp, STD_RIGHT_DELETE_ACCESS,
3165 	    scred, &fid);
3166 	if (error) {
3167 		SMBVDEBUG("error %d opening %s\n",
3168 		    error, onp->n_rpath);
3169 		goto out;
3170 	}
3171 
3172 	smbfs_attrcache_remove(onp);
3173 	error = smbfs_smb_rename(odnp, onp, ndnp, nnm, strlen(nnm),
3174 	    fid, scred);
3175 
3176 	smbfs_smb_tmpclose(onp, fid);
3177 
3178 	/*
3179 	 * If the old name should no longer exist,
3180 	 * discard any cached attributes under it.
3181 	 */
3182 	if (error == 0) {
3183 		smbfs_attrcache_prune(onp);
3184 		/* SMBFS_VNEVENT... */
3185 	}
3186 
3187 out:
3188 	if (nvp) {
3189 		if (nvp_locked)
3190 			vn_vfsunlock(nvp);
3191 		VN_RELE(nvp);
3192 	}
3193 
3194 	return (error);
3195 }
3196 
3197 /*
3198  * XXX
3199  * vsecattr_t is new to build 77, and we need to eventually support
3200  * it in order to create an ACL when an object is created.
3201  *
3202  * This op should support the new FIGNORECASE flag for case-insensitive
3203  * lookups, per PSARC 2007/244.
3204  */
3205 /* ARGSUSED */
3206 static int
3207 smbfs_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp,
3208 	cred_t *cr, caller_context_t *ct, int flags, vsecattr_t *vsecp)
3209 {
3210 	vnode_t		*vp;
3211 	struct smbnode	*dnp = VTOSMB(dvp);
3212 	struct smbmntinfo *smi = VTOSMI(dvp);
3213 	struct smb_cred	scred;
3214 	struct smbfattr	fattr;
3215 	const char		*name = (const char *) nm;
3216 	int		nmlen = strlen(name);
3217 	int		error;
3218 
3219 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
3220 		return (EPERM);
3221 
3222 	if (smi->smi_flags & SMI_DEAD || dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
3223 		return (EIO);
3224 
3225 	if ((nmlen == 1 && name[0] == '.') ||
3226 	    (nmlen == 2 && name[0] == '.' && name[1] == '.'))
3227 		return (EEXIST);
3228 
3229 	/* Only plain files are allowed in V_XATTRDIR. */
3230 	if (dvp->v_flag & V_XATTRDIR)
3231 		return (EINVAL);
3232 
3233 	if (smbfs_rw_enter_sig(&dnp->r_rwlock, RW_WRITER, SMBINTR(dvp)))
3234 		return (EINTR);
3235 	smb_credinit(&scred, cr);
3236 
3237 	/*
3238 	 * Require write access in the containing directory.
3239 	 */
3240 	error = smbfs_access(dvp, VWRITE, 0, cr, ct);
3241 	if (error)
3242 		goto out;
3243 
3244 	error = smbfs_smb_mkdir(dnp, name, nmlen, &scred);
3245 	if (error)
3246 		goto out;
3247 
3248 	error = smbfs_smb_lookup(dnp, &name, &nmlen, &fattr, &scred);
3249 	if (error)
3250 		goto out;
3251 
3252 	smbfs_attr_touchdir(dnp);
3253 
3254 	error = smbfs_nget(dvp, name, nmlen, &fattr, &vp);
3255 	if (error)
3256 		goto out;
3257 
3258 	/* Success! */
3259 	*vpp = vp;
3260 	error = 0;
3261 out:
3262 	smb_credrele(&scred);
3263 	smbfs_rw_exit(&dnp->r_rwlock);
3264 
3265 	if (name != nm)
3266 		smbfs_name_free(name, nmlen);
3267 
3268 	return (error);
3269 }
3270 
3271 /*
3272  * XXX
3273  * This op should support the new FIGNORECASE flag for case-insensitive
3274  * lookups, per PSARC 2007/244.
3275  */
3276 /* ARGSUSED */
3277 static int
3278 smbfs_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr,
3279 	caller_context_t *ct, int flags)
3280 {
3281 	struct smb_cred	scred;
3282 	vnode_t		*vp = NULL;
3283 	int		vp_locked = 0;
3284 	struct smbmntinfo *smi = VTOSMI(dvp);
3285 	struct smbnode	*dnp = VTOSMB(dvp);
3286 	struct smbnode	*np;
3287 	int		error;
3288 
3289 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
3290 		return (EPERM);
3291 
3292 	if (smi->smi_flags & SMI_DEAD || dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
3293 		return (EIO);
3294 
3295 	/*
3296 	 * Verify access to the dirctory.
3297 	 */
3298 	error = smbfs_access(dvp, VWRITE|VEXEC, 0, cr, ct);
3299 	if (error)
3300 		return (error);
3301 
3302 	if (smbfs_rw_enter_sig(&dnp->r_rwlock, RW_WRITER, SMBINTR(dvp)))
3303 		return (EINTR);
3304 	smb_credinit(&scred, cr);
3305 
3306 	/*
3307 	 * First lookup the entry to be removed.
3308 	 */
3309 	error = smbfslookup(dvp, nm, &vp, cr, 0, ct);
3310 	if (error)
3311 		goto out;
3312 	np = VTOSMB(vp);
3313 
3314 	/*
3315 	 * Disallow rmdir of "." or current dir, or the FS root.
3316 	 * Also make sure it's a directory, not a mount point,
3317 	 * and lock to keep mount/umount away until we're done.
3318 	 */
3319 	if ((vp == dvp) || (vp == cdir) || (vp->v_flag & VROOT)) {
3320 		error = EINVAL;
3321 		goto out;
3322 	}
3323 	if (vp->v_type != VDIR) {
3324 		error = ENOTDIR;
3325 		goto out;
3326 	}
3327 	if (vn_vfsrlock(vp)) {
3328 		error = EBUSY;
3329 		goto out;
3330 	}
3331 	vp_locked = 1;
3332 	if (vn_mountedvfs(vp) != NULL) {
3333 		error = EBUSY;
3334 		goto out;
3335 	}
3336 
3337 	/*
3338 	 * Do the real rmdir work
3339 	 */
3340 	error = smbfsremove(dvp, vp, &scred, flags);
3341 	if (error)
3342 		goto out;
3343 
3344 #ifdef	SMBFS_VNEVENT
3345 	vnevent_rmdir(vp, dvp, nm, ct);
3346 #endif
3347 
3348 	mutex_enter(&np->r_statelock);
3349 	dnp->n_flag |= NMODIFIED;
3350 	mutex_exit(&np->r_statelock);
3351 	smbfs_attr_touchdir(dnp);
3352 	smbfs_rmhash(np);
3353 
3354 out:
3355 	if (vp) {
3356 		if (vp_locked)
3357 			vn_vfsunlock(vp);
3358 		VN_RELE(vp);
3359 	}
3360 	smb_credrele(&scred);
3361 	smbfs_rw_exit(&dnp->r_rwlock);
3362 
3363 	return (error);
3364 }
3365 
3366 
3367 /* ARGSUSED */
3368 static int
3369 smbfs_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr,
3370 	caller_context_t *ct, int flags)
3371 {
3372 	/* Not yet... */
3373 	return (ENOSYS);
3374 }
3375 
3376 
3377 /* ARGSUSED */
3378 static int
3379 smbfs_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp,
3380 	caller_context_t *ct, int flags)
3381 {
3382 	struct smbnode	*np = VTOSMB(vp);
3383 	int		error = 0;
3384 	smbmntinfo_t	*smi;
3385 
3386 	smi = VTOSMI(vp);
3387 
3388 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
3389 		return (EIO);
3390 
3391 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
3392 		return (EIO);
3393 
3394 	/*
3395 	 * Require read access in the directory.
3396 	 */
3397 	error = smbfs_access(vp, VREAD, 0, cr, ct);
3398 	if (error)
3399 		return (error);
3400 
3401 	ASSERT(smbfs_rw_lock_held(&np->r_rwlock, RW_READER));
3402 
3403 	/*
3404 	 * Todo readdir cache here
3405 	 *
3406 	 * I am serializing the entire readdir opreation
3407 	 * now since we have not yet implemented readdir
3408 	 * cache. This fix needs to be revisited once
3409 	 * we implement readdir cache.
3410 	 */
3411 	if (smbfs_rw_enter_sig(&np->r_lkserlock, RW_WRITER, SMBINTR(vp)))
3412 		return (EINTR);
3413 
3414 	error = smbfs_readvdir(vp, uiop, cr, eofp, ct);
3415 
3416 	smbfs_rw_exit(&np->r_lkserlock);
3417 
3418 	return (error);
3419 }
3420 
3421 /* ARGSUSED */
3422 static int
3423 smbfs_readvdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
3424 	caller_context_t *ct)
3425 {
3426 	/*
3427 	 * Note: "limit" tells the SMB-level FindFirst/FindNext
3428 	 * functions how many directory entries to request in
3429 	 * each OtW call.  It needs to be large enough so that
3430 	 * we don't make lots of tiny OtW requests, but there's
3431 	 * no point making it larger than the maximum number of
3432 	 * OtW entries that would fit in a maximum sized trans2
3433 	 * response (64k / 48).  Beyond that, it's just tuning.
3434 	 * WinNT used 512, Win2k used 1366.  We use 1000.
3435 	 */
3436 	static const int limit = 1000;
3437 	/* Largest possible dirent size. */
3438 	static const size_t dbufsiz = DIRENT64_RECLEN(SMB_MAXFNAMELEN);
3439 	struct smb_cred scred;
3440 	vnode_t		*newvp;
3441 	struct smbnode	*np = VTOSMB(vp);
3442 	struct smbfs_fctx *ctx;
3443 	struct dirent64 *dp;
3444 	ssize_t		save_resid;
3445 	offset_t	save_offset; /* 64 bits */
3446 	int		offset; /* yes, 32 bits */
3447 	int		nmlen, error;
3448 	ushort_t	reclen;
3449 
3450 	ASSERT(curproc->p_zone == VTOSMI(vp)->smi_zone_ref.zref_zone);
3451 
3452 	/* Make sure we serialize for n_dirseq use. */
3453 	ASSERT(smbfs_rw_lock_held(&np->r_lkserlock, RW_WRITER));
3454 
3455 	/*
3456 	 * Make sure smbfs_open filled in n_dirseq
3457 	 */
3458 	if (np->n_dirseq == NULL)
3459 		return (EBADF);
3460 
3461 	/* Check for overflow of (32-bit) directory offset. */
3462 	if (uio->uio_loffset < 0 || uio->uio_loffset > INT32_MAX ||
3463 	    (uio->uio_loffset + uio->uio_resid) > INT32_MAX)
3464 		return (EINVAL);
3465 
3466 	/* Require space for at least one dirent. */
3467 	if (uio->uio_resid < dbufsiz)
3468 		return (EINVAL);
3469 
3470 	SMBVDEBUG("dirname='%s'\n", np->n_rpath);
3471 	smb_credinit(&scred, cr);
3472 	dp = kmem_alloc(dbufsiz, KM_SLEEP);
3473 
3474 	save_resid = uio->uio_resid;
3475 	save_offset = uio->uio_loffset;
3476 	offset = uio->uio_offset;
3477 	SMBVDEBUG("in: offset=%d, resid=%d\n",
3478 	    (int)uio->uio_offset, (int)uio->uio_resid);
3479 	error = 0;
3480 
3481 	/*
3482 	 * Generate the "." and ".." entries here so we can
3483 	 * (1) make sure they appear (but only once), and
3484 	 * (2) deal with getting their I numbers which the
3485 	 * findnext below does only for normal names.
3486 	 */
3487 	while (offset < FIRST_DIROFS) {
3488 		/*
3489 		 * Tricky bit filling in the first two:
3490 		 * offset 0 is ".", offset 1 is ".."
3491 		 * so strlen of these is offset+1.
3492 		 */
3493 		reclen = DIRENT64_RECLEN(offset + 1);
3494 		if (uio->uio_resid < reclen)
3495 			goto out;
3496 		bzero(dp, reclen);
3497 		dp->d_reclen = reclen;
3498 		dp->d_name[0] = '.';
3499 		dp->d_name[1] = '.';
3500 		dp->d_name[offset + 1] = '\0';
3501 		/*
3502 		 * Want the real I-numbers for the "." and ".."
3503 		 * entries.  For these two names, we know that
3504 		 * smbfslookup can get the nodes efficiently.
3505 		 */
3506 		error = smbfslookup(vp, dp->d_name, &newvp, cr, 1, ct);
3507 		if (error) {
3508 			dp->d_ino = np->n_ino + offset; /* fiction */
3509 		} else {
3510 			dp->d_ino = VTOSMB(newvp)->n_ino;
3511 			VN_RELE(newvp);
3512 		}
3513 		/*
3514 		 * Note: d_off is the offset that a user-level program
3515 		 * should seek to for reading the NEXT directory entry.
3516 		 * See libc: readdir, telldir, seekdir
3517 		 */
3518 		dp->d_off = offset + 1;
3519 		error = uiomove(dp, reclen, UIO_READ, uio);
3520 		if (error)
3521 			goto out;
3522 		/*
3523 		 * Note: uiomove updates uio->uio_offset,
3524 		 * but we want it to be our "cookie" value,
3525 		 * which just counts dirents ignoring size.
3526 		 */
3527 		uio->uio_offset = ++offset;
3528 	}
3529 
3530 	/*
3531 	 * If there was a backward seek, we have to reopen.
3532 	 */
3533 	if (offset < np->n_dirofs) {
3534 		SMBVDEBUG("Reopening search %d:%d\n",
3535 		    offset, np->n_dirofs);
3536 		error = smbfs_smb_findopen(np, "*", 1,
3537 		    SMB_FA_SYSTEM | SMB_FA_HIDDEN | SMB_FA_DIR,
3538 		    &scred, &ctx);
3539 		if (error) {
3540 			SMBVDEBUG("can not open search, error = %d", error);
3541 			goto out;
3542 		}
3543 		/* free the old one */
3544 		(void) smbfs_smb_findclose(np->n_dirseq, &scred);
3545 		/* save the new one */
3546 		np->n_dirseq = ctx;
3547 		np->n_dirofs = FIRST_DIROFS;
3548 	} else {
3549 		ctx = np->n_dirseq;
3550 	}
3551 
3552 	/*
3553 	 * Skip entries before the requested offset.
3554 	 */
3555 	while (np->n_dirofs < offset) {
3556 		error = smbfs_smb_findnext(ctx, limit, &scred);
3557 		if (error != 0)
3558 			goto out;
3559 		np->n_dirofs++;
3560 	}
3561 
3562 	/*
3563 	 * While there's room in the caller's buffer:
3564 	 *	get a directory entry from SMB,
3565 	 *	convert to a dirent, copyout.
3566 	 * We stop when there is no longer room for a
3567 	 * maximum sized dirent because we must decide
3568 	 * before we know anything about the next entry.
3569 	 */
3570 	while (uio->uio_resid >= dbufsiz) {
3571 		error = smbfs_smb_findnext(ctx, limit, &scred);
3572 		if (error != 0)
3573 			goto out;
3574 		np->n_dirofs++;
3575 
3576 		/* Sanity check the name length. */
3577 		nmlen = ctx->f_nmlen;
3578 		if (nmlen > SMB_MAXFNAMELEN) {
3579 			nmlen = SMB_MAXFNAMELEN;
3580 			SMBVDEBUG("Truncating name: %s\n", ctx->f_name);
3581 		}
3582 		if (smbfs_fastlookup) {
3583 			/* See comment at smbfs_fastlookup above. */
3584 			if (smbfs_nget(vp, ctx->f_name, nmlen,
3585 			    &ctx->f_attr, &newvp) == 0)
3586 				VN_RELE(newvp);
3587 		}
3588 
3589 		reclen = DIRENT64_RECLEN(nmlen);
3590 		bzero(dp, reclen);
3591 		dp->d_reclen = reclen;
3592 		bcopy(ctx->f_name, dp->d_name, nmlen);
3593 		dp->d_name[nmlen] = '\0';
3594 		dp->d_ino = ctx->f_inum;
3595 		dp->d_off = offset + 1;	/* See d_off comment above */
3596 		error = uiomove(dp, reclen, UIO_READ, uio);
3597 		if (error)
3598 			goto out;
3599 		/* See comment re. uio_offset above. */
3600 		uio->uio_offset = ++offset;
3601 	}
3602 
3603 out:
3604 	/*
3605 	 * When we come to the end of a directory, the
3606 	 * SMB-level functions return ENOENT, but the
3607 	 * caller is not expecting an error return.
3608 	 *
3609 	 * Also note that we must delay the call to
3610 	 * smbfs_smb_findclose(np->n_dirseq, ...)
3611 	 * until smbfs_close so that all reads at the
3612 	 * end of the directory will return no data.
3613 	 */
3614 	if (error == ENOENT) {
3615 		error = 0;
3616 		if (eofp)
3617 			*eofp = 1;
3618 	}
3619 	/*
3620 	 * If we encountered an error (i.e. "access denied")
3621 	 * from the FindFirst call, we will have copied out
3622 	 * the "." and ".." entries leaving offset == 2.
3623 	 * In that case, restore the original offset/resid
3624 	 * so the caller gets no data with the error.
3625 	 */
3626 	if (error != 0 && offset == FIRST_DIROFS) {
3627 		uio->uio_loffset = save_offset;
3628 		uio->uio_resid = save_resid;
3629 	}
3630 	SMBVDEBUG("out: offset=%d, resid=%d\n",
3631 	    (int)uio->uio_offset, (int)uio->uio_resid);
3632 
3633 	kmem_free(dp, dbufsiz);
3634 	smb_credrele(&scred);
3635 	return (error);
3636 }
3637 
3638 /*
3639  * Here NFS has: nfs3_bio
3640  * See smbfs_bio above.
3641  */
3642 
3643 /* ARGSUSED */
3644 static int
3645 smbfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
3646 {
3647 	return (ENOSYS);
3648 }
3649 
3650 
3651 /*
3652  * The pair of functions VOP_RWLOCK, VOP_RWUNLOCK
3653  * are optional functions that are called by:
3654  *    getdents, before/after VOP_READDIR
3655  *    pread, before/after ... VOP_READ
3656  *    pwrite, before/after ... VOP_WRITE
3657  *    (other places)
3658  *
3659  * Careful here: None of the above check for any
3660  * error returns from VOP_RWLOCK / VOP_RWUNLOCK!
3661  * In fact, the return value from _rwlock is NOT
3662  * an error code, but V_WRITELOCK_TRUE / _FALSE.
3663  *
3664  * Therefore, it's up to _this_ code to make sure
3665  * the lock state remains balanced, which means
3666  * we can't "bail out" on interrupts, etc.
3667  */
3668 
3669 /* ARGSUSED2 */
3670 static int
3671 smbfs_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3672 {
3673 	smbnode_t	*np = VTOSMB(vp);
3674 
3675 	if (!write_lock) {
3676 		(void) smbfs_rw_enter_sig(&np->r_rwlock, RW_READER, FALSE);
3677 		return (V_WRITELOCK_FALSE);
3678 	}
3679 
3680 
3681 	(void) smbfs_rw_enter_sig(&np->r_rwlock, RW_WRITER, FALSE);
3682 	return (V_WRITELOCK_TRUE);
3683 }
3684 
3685 /* ARGSUSED */
3686 static void
3687 smbfs_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3688 {
3689 	smbnode_t	*np = VTOSMB(vp);
3690 
3691 	smbfs_rw_exit(&np->r_rwlock);
3692 }
3693 
3694 
3695 /* ARGSUSED */
3696 static int
3697 smbfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
3698 {
3699 	smbmntinfo_t	*smi;
3700 
3701 	smi = VTOSMI(vp);
3702 
3703 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
3704 		return (EPERM);
3705 
3706 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
3707 		return (EIO);
3708 
3709 	/*
3710 	 * Because we stuff the readdir cookie into the offset field
3711 	 * someone may attempt to do an lseek with the cookie which
3712 	 * we want to succeed.
3713 	 */
3714 	if (vp->v_type == VDIR)
3715 		return (0);
3716 
3717 	/* Like NFS3, just check for 63-bit overflow. */
3718 	if (*noffp < 0)
3719 		return (EINVAL);
3720 
3721 	return (0);
3722 }
3723 
3724 /* mmap support ******************************************************** */
3725 
3726 #ifdef	_KERNEL
3727 
3728 #ifdef DEBUG
3729 static int smbfs_lostpage = 0;	/* number of times we lost original page */
3730 #endif
3731 
3732 /*
3733  * Return all the pages from [off..off+len) in file
3734  * Like nfs3_getpage
3735  */
3736 /* ARGSUSED */
3737 static int
3738 smbfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3739 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3740 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3741 {
3742 	smbnode_t	*np;
3743 	smbmntinfo_t	*smi;
3744 	int		error;
3745 
3746 	np = VTOSMB(vp);
3747 	smi = VTOSMI(vp);
3748 
3749 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
3750 		return (EIO);
3751 
3752 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
3753 		return (EIO);
3754 
3755 	if (vp->v_flag & VNOMAP)
3756 		return (ENOSYS);
3757 
3758 	if (protp != NULL)
3759 		*protp = PROT_ALL;
3760 
3761 	/*
3762 	 * Now valididate that the caches are up to date.
3763 	 */
3764 	error = smbfs_validate_caches(vp, cr);
3765 	if (error)
3766 		return (error);
3767 
3768 retry:
3769 	mutex_enter(&np->r_statelock);
3770 
3771 	/*
3772 	 * Don't create dirty pages faster than they
3773 	 * can be cleaned ... (etc. see nfs)
3774 	 *
3775 	 * Here NFS also tests:
3776 	 *  (mi->mi_max_threads != 0 &&
3777 	 *  rp->r_awcount > 2 * mi->mi_max_threads)
3778 	 */
3779 	if (rw == S_CREATE) {
3780 		while (np->r_gcount > 0)
3781 			cv_wait(&np->r_cv, &np->r_statelock);
3782 	}
3783 
3784 	/*
3785 	 * If we are getting called as a side effect of a write
3786 	 * operation the local file size might not be extended yet.
3787 	 * In this case we want to be able to return pages of zeroes.
3788 	 */
3789 	if (off + len > np->r_size + PAGEOFFSET && seg != segkmap) {
3790 		mutex_exit(&np->r_statelock);
3791 		return (EFAULT);		/* beyond EOF */
3792 	}
3793 
3794 	mutex_exit(&np->r_statelock);
3795 
3796 	error = pvn_getpages(smbfs_getapage, vp, off, len, protp,
3797 	    pl, plsz, seg, addr, rw, cr);
3798 
3799 	switch (error) {
3800 	case SMBFS_EOF:
3801 		smbfs_purge_caches(vp, cr);
3802 		goto retry;
3803 	case ESTALE:
3804 		/*
3805 		 * Here NFS has: PURGE_STALE_FH(error, vp, cr);
3806 		 * In-line here as we only use it once.
3807 		 */
3808 		mutex_enter(&np->r_statelock);
3809 		np->r_flags |= RSTALE;
3810 		if (!np->r_error)
3811 			np->r_error = (error);
3812 		mutex_exit(&np->r_statelock);
3813 		if (vn_has_cached_data(vp))
3814 			smbfs_invalidate_pages(vp, (u_offset_t)0, cr);
3815 		smbfs_purge_caches(vp, cr);
3816 		break;
3817 	default:
3818 		break;
3819 	}
3820 
3821 	return (error);
3822 }
3823 
3824 /*
3825  * Called from pvn_getpages to get a particular page.
3826  * Like nfs3_getapage
3827  */
3828 /* ARGSUSED */
3829 static int
3830 smbfs_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp,
3831 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3832 	enum seg_rw rw, cred_t *cr)
3833 {
3834 	smbnode_t	*np;
3835 	smbmntinfo_t   *smi;
3836 
3837 	uint_t		bsize;
3838 	struct buf	*bp;
3839 	page_t		*pp;
3840 	u_offset_t	lbn;
3841 	u_offset_t	io_off;
3842 	u_offset_t	blkoff;
3843 	size_t		io_len;
3844 	uint_t blksize;
3845 	int error;
3846 	/* int readahead; */
3847 	int readahead_issued = 0;
3848 	/* int ra_window; * readahead window */
3849 	page_t *pagefound;
3850 
3851 	np = VTOSMB(vp);
3852 	smi = VTOSMI(vp);
3853 
3854 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
3855 		return (EIO);
3856 
3857 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
3858 		return (EIO);
3859 
3860 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3861 
3862 reread:
3863 	bp = NULL;
3864 	pp = NULL;
3865 	pagefound = NULL;
3866 
3867 	if (pl != NULL)
3868 		pl[0] = NULL;
3869 
3870 	error = 0;
3871 	lbn = off / bsize;
3872 	blkoff = lbn * bsize;
3873 
3874 	/*
3875 	 * NFS queues up readahead work here.
3876 	 */
3877 
3878 again:
3879 	if ((pagefound = page_exists(vp, off)) == NULL) {
3880 		if (pl == NULL) {
3881 			(void) 0; /* Todo: smbfs_async_readahead(); */
3882 		} else if (rw == S_CREATE) {
3883 			/*
3884 			 * Block for this page is not allocated, or the offset
3885 			 * is beyond the current allocation size, or we're
3886 			 * allocating a swap slot and the page was not found,
3887 			 * so allocate it and return a zero page.
3888 			 */
3889 			if ((pp = page_create_va(vp, off,
3890 			    PAGESIZE, PG_WAIT, seg, addr)) == NULL)
3891 				cmn_err(CE_PANIC, "smbfs_getapage: page_create");
3892 			io_len = PAGESIZE;
3893 			mutex_enter(&np->r_statelock);
3894 			np->r_nextr = off + PAGESIZE;
3895 			mutex_exit(&np->r_statelock);
3896 		} else {
3897 			/*
3898 			 * Need to go to server to get a BLOCK, exception to
3899 			 * that being while reading at offset = 0 or doing
3900 			 * random i/o, in that case read only a PAGE.
3901 			 */
3902 			mutex_enter(&np->r_statelock);
3903 			if (blkoff < np->r_size &&
3904 			    blkoff + bsize >= np->r_size) {
3905 				/*
3906 				 * If only a block or less is left in
3907 				 * the file, read all that is remaining.
3908 				 */
3909 				if (np->r_size <= off) {
3910 					/*
3911 					 * Trying to access beyond EOF,
3912 					 * set up to get at least one page.
3913 					 */
3914 					blksize = off + PAGESIZE - blkoff;
3915 				} else
3916 					blksize = np->r_size - blkoff;
3917 			} else if ((off == 0) ||
3918 			    (off != np->r_nextr && !readahead_issued)) {
3919 				blksize = PAGESIZE;
3920 				blkoff = off; /* block = page here */
3921 			} else
3922 				blksize = bsize;
3923 			mutex_exit(&np->r_statelock);
3924 
3925 			pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3926 			    &io_len, blkoff, blksize, 0);
3927 
3928 			/*
3929 			 * Some other thread has entered the page,
3930 			 * so just use it.
3931 			 */
3932 			if (pp == NULL)
3933 				goto again;
3934 
3935 			/*
3936 			 * Now round the request size up to page boundaries.
3937 			 * This ensures that the entire page will be
3938 			 * initialized to zeroes if EOF is encountered.
3939 			 */
3940 			io_len = ptob(btopr(io_len));
3941 
3942 			bp = pageio_setup(pp, io_len, vp, B_READ);
3943 			ASSERT(bp != NULL);
3944 
3945 			/*
3946 			 * pageio_setup should have set b_addr to 0.  This
3947 			 * is correct since we want to do I/O on a page
3948 			 * boundary.  bp_mapin will use this addr to calculate
3949 			 * an offset, and then set b_addr to the kernel virtual
3950 			 * address it allocated for us.
3951 			 */
3952 			ASSERT(bp->b_un.b_addr == 0);
3953 
3954 			bp->b_edev = 0;
3955 			bp->b_dev = 0;
3956 			bp->b_lblkno = lbtodb(io_off);
3957 			bp->b_file = vp;
3958 			bp->b_offset = (offset_t)off;
3959 			bp_mapin(bp);
3960 
3961 			/*
3962 			 * If doing a write beyond what we believe is EOF,
3963 			 * don't bother trying to read the pages from the
3964 			 * server, we'll just zero the pages here.  We
3965 			 * don't check that the rw flag is S_WRITE here
3966 			 * because some implementations may attempt a
3967 			 * read access to the buffer before copying data.
3968 			 */
3969 			mutex_enter(&np->r_statelock);
3970 			if (io_off >= np->r_size && seg == segkmap) {
3971 				mutex_exit(&np->r_statelock);
3972 				bzero(bp->b_un.b_addr, io_len);
3973 			} else {
3974 				mutex_exit(&np->r_statelock);
3975 				error = smbfs_bio(bp, 0, cr);
3976 			}
3977 
3978 			/*
3979 			 * Unmap the buffer before freeing it.
3980 			 */
3981 			bp_mapout(bp);
3982 			pageio_done(bp);
3983 
3984 			/* Here NFS3 updates all pp->p_fsdata */
3985 
3986 			if (error == SMBFS_EOF) {
3987 				/*
3988 				 * If doing a write system call just return
3989 				 * zeroed pages, else user tried to get pages
3990 				 * beyond EOF, return error.  We don't check
3991 				 * that the rw flag is S_WRITE here because
3992 				 * some implementations may attempt a read
3993 				 * access to the buffer before copying data.
3994 				 */
3995 				if (seg == segkmap)
3996 					error = 0;
3997 				else
3998 					error = EFAULT;
3999 			}
4000 
4001 			if (!readahead_issued && !error) {
4002 				mutex_enter(&np->r_statelock);
4003 				np->r_nextr = io_off + io_len;
4004 				mutex_exit(&np->r_statelock);
4005 			}
4006 		}
4007 	}
4008 
4009 	if (pl == NULL)
4010 		return (error);
4011 
4012 	if (error) {
4013 		if (pp != NULL)
4014 			pvn_read_done(pp, B_ERROR);
4015 		return (error);
4016 	}
4017 
4018 	if (pagefound) {
4019 		se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED);
4020 
4021 		/*
4022 		 * Page exists in the cache, acquire the appropriate lock.
4023 		 * If this fails, start all over again.
4024 		 */
4025 		if ((pp = page_lookup(vp, off, se)) == NULL) {
4026 #ifdef DEBUG
4027 			smbfs_lostpage++;
4028 #endif
4029 			goto reread;
4030 		}
4031 		pl[0] = pp;
4032 		pl[1] = NULL;
4033 		return (0);
4034 	}
4035 
4036 	if (pp != NULL)
4037 		pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4038 
4039 	return (error);
4040 }
4041 
4042 /*
4043  * Here NFS has: nfs3_readahead
4044  * No read-ahead in smbfs yet.
4045  */
4046 
4047 #endif	// _KERNEL
4048 
4049 /*
4050  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE}
4051  * If len == 0, do from off to EOF.
4052  *
4053  * The normal cases should be len == 0 && off == 0 (entire vp list),
4054  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
4055  * (from pageout).
4056  *
4057  * Like nfs3_putpage + nfs_putpages
4058  */
4059 /* ARGSUSED */
4060 static int
4061 smbfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4062 	caller_context_t *ct)
4063 {
4064 #ifdef	_KERNEL
4065 	smbnode_t *np;
4066 	smbmntinfo_t *smi;
4067 	page_t *pp;
4068 	u_offset_t eoff;
4069 	u_offset_t io_off;
4070 	size_t io_len;
4071 	int error;
4072 	int rdirty;
4073 	int err;
4074 
4075 	np = VTOSMB(vp);
4076 	smi = VTOSMI(vp);
4077 
4078 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
4079 		return (EIO);
4080 
4081 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
4082 		return (EIO);
4083 
4084 	if (vp->v_flag & VNOMAP)
4085 		return (ENOSYS);
4086 
4087 	/* Here NFS does rp->r_count (++/--) stuff. */
4088 
4089 	/* Beginning of code from nfs_putpages. */
4090 
4091 	if (!vn_has_cached_data(vp))
4092 		return (0);
4093 
4094 	/*
4095 	 * If ROUTOFSPACE is set, then all writes turn into B_INVAL
4096 	 * writes.  B_FORCE is set to force the VM system to actually
4097 	 * invalidate the pages, even if the i/o failed.  The pages
4098 	 * need to get invalidated because they can't be written out
4099 	 * because there isn't any space left on either the server's
4100 	 * file system or in the user's disk quota.  The B_FREE bit
4101 	 * is cleared to avoid confusion as to whether this is a
4102 	 * request to place the page on the freelist or to destroy
4103 	 * it.
4104 	 */
4105 	if ((np->r_flags & ROUTOFSPACE) ||
4106 	    (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED))
4107 		flags = (flags & ~B_FREE) | B_INVAL | B_FORCE;
4108 
4109 	if (len == 0) {
4110 		/*
4111 		 * If doing a full file synchronous operation, then clear
4112 		 * the RDIRTY bit.  If a page gets dirtied while the flush
4113 		 * is happening, then RDIRTY will get set again.  The
4114 		 * RDIRTY bit must get cleared before the flush so that
4115 		 * we don't lose this information.
4116 		 *
4117 		 * NFS has B_ASYNC vs sync stuff here.
4118 		 */
4119 		if (off == (u_offset_t)0 &&
4120 		    (np->r_flags & RDIRTY)) {
4121 			mutex_enter(&np->r_statelock);
4122 			rdirty = (np->r_flags & RDIRTY);
4123 			np->r_flags &= ~RDIRTY;
4124 			mutex_exit(&np->r_statelock);
4125 		} else
4126 			rdirty = 0;
4127 
4128 		/*
4129 		 * Search the entire vp list for pages >= off, and flush
4130 		 * the dirty pages.
4131 		 */
4132 		error = pvn_vplist_dirty(vp, off, smbfs_putapage,
4133 		    flags, cr);
4134 
4135 		/*
4136 		 * If an error occurred and the file was marked as dirty
4137 		 * before and we aren't forcibly invalidating pages, then
4138 		 * reset the RDIRTY flag.
4139 		 */
4140 		if (error && rdirty &&
4141 		    (flags & (B_INVAL | B_FORCE)) != (B_INVAL | B_FORCE)) {
4142 			mutex_enter(&np->r_statelock);
4143 			np->r_flags |= RDIRTY;
4144 			mutex_exit(&np->r_statelock);
4145 		}
4146 	} else {
4147 		/*
4148 		 * Do a range from [off...off + len) looking for pages
4149 		 * to deal with.
4150 		 */
4151 		error = 0;
4152 		io_len = 1; /* quiet warnings */
4153 		eoff = off + len;
4154 
4155 		for (io_off = off; io_off < eoff; io_off += io_len) {
4156 			mutex_enter(&np->r_statelock);
4157 			if (io_off >= np->r_size) {
4158 				mutex_exit(&np->r_statelock);
4159 				break;
4160 			}
4161 			mutex_exit(&np->r_statelock);
4162 			/*
4163 			 * If we are not invalidating, synchronously
4164 			 * freeing or writing pages use the routine
4165 			 * page_lookup_nowait() to prevent reclaiming
4166 			 * them from the free list.
4167 			 */
4168 			if ((flags & B_INVAL) || !(flags & B_ASYNC)) {
4169 				pp = page_lookup(vp, io_off,
4170 				    (flags & (B_INVAL | B_FREE)) ?
4171 				    SE_EXCL : SE_SHARED);
4172 			} else {
4173 				pp = page_lookup_nowait(vp, io_off,
4174 				    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4175 			}
4176 
4177 			if (pp == NULL || !pvn_getdirty(pp, flags))
4178 				io_len = PAGESIZE;
4179 			else {
4180 				err = smbfs_putapage(vp, pp, &io_off,
4181 				    &io_len, flags, cr);
4182 				if (!error)
4183 					error = err;
4184 				/*
4185 				 * "io_off" and "io_len" are returned as
4186 				 * the range of pages we actually wrote.
4187 				 * This allows us to skip ahead more quickly
4188 				 * since several pages may've been dealt
4189 				 * with by this iteration of the loop.
4190 				 */
4191 			}
4192 		}
4193 	}
4194 
4195 	return (error);
4196 
4197 #else	// _KERNEL
4198 	return (ENOSYS);
4199 #endif	// _KERNEL
4200 }
4201 
4202 #ifdef	_KERNEL
4203 
4204 /*
4205  * Write out a single page, possibly klustering adjacent dirty pages.
4206  *
4207  * Like nfs3_putapage / nfs3_sync_putapage
4208  */
4209 static int
4210 smbfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
4211 	int flags, cred_t *cr)
4212 {
4213 	smbnode_t *np;
4214 	u_offset_t io_off;
4215 	u_offset_t lbn_off;
4216 	u_offset_t lbn;
4217 	size_t io_len;
4218 	uint_t bsize;
4219 	int error;
4220 
4221 	np = VTOSMB(vp);
4222 
4223 	ASSERT(!vn_is_readonly(vp));
4224 
4225 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
4226 	lbn = pp->p_offset / bsize;
4227 	lbn_off = lbn * bsize;
4228 
4229 	/*
4230 	 * Find a kluster that fits in one block, or in
4231 	 * one page if pages are bigger than blocks.  If
4232 	 * there is less file space allocated than a whole
4233 	 * page, we'll shorten the i/o request below.
4234 	 */
4235 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off,
4236 	    roundup(bsize, PAGESIZE), flags);
4237 
4238 	/*
4239 	 * pvn_write_kluster shouldn't have returned a page with offset
4240 	 * behind the original page we were given.  Verify that.
4241 	 */
4242 	ASSERT((pp->p_offset / bsize) >= lbn);
4243 
4244 	/*
4245 	 * Now pp will have the list of kept dirty pages marked for
4246 	 * write back.  It will also handle invalidation and freeing
4247 	 * of pages that are not dirty.  Check for page length rounding
4248 	 * problems.
4249 	 */
4250 	if (io_off + io_len > lbn_off + bsize) {
4251 		ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE);
4252 		io_len = lbn_off + bsize - io_off;
4253 	}
4254 	/*
4255 	 * The RMODINPROGRESS flag makes sure that smbfs_bio() sees a
4256 	 * consistent value of r_size. RMODINPROGRESS is set in writerp().
4257 	 * When RMODINPROGRESS is set it indicates that a uiomove() is in
4258 	 * progress and the r_size has not been made consistent with the
4259 	 * new size of the file. When the uiomove() completes the r_size is
4260 	 * updated and the RMODINPROGRESS flag is cleared.
4261 	 *
4262 	 * The RMODINPROGRESS flag makes sure that smbfs_bio() sees a
4263 	 * consistent value of r_size. Without this handshaking, it is
4264 	 * possible that smbfs_bio() picks  up the old value of r_size
4265 	 * before the uiomove() in writerp() completes. This will result
4266 	 * in the write through smbfs_bio() being dropped.
4267 	 *
4268 	 * More precisely, there is a window between the time the uiomove()
4269 	 * completes and the time the r_size is updated. If a VOP_PUTPAGE()
4270 	 * operation intervenes in this window, the page will be picked up,
4271 	 * because it is dirty (it will be unlocked, unless it was
4272 	 * pagecreate'd). When the page is picked up as dirty, the dirty
4273 	 * bit is reset (pvn_getdirty()). In smbfs_write(), r_size is
4274 	 * checked. This will still be the old size. Therefore the page will
4275 	 * not be written out. When segmap_release() calls VOP_PUTPAGE(),
4276 	 * the page will be found to be clean and the write will be dropped.
4277 	 */
4278 	if (np->r_flags & RMODINPROGRESS) {
4279 		mutex_enter(&np->r_statelock);
4280 		if ((np->r_flags & RMODINPROGRESS) &&
4281 		    np->r_modaddr + MAXBSIZE > io_off &&
4282 		    np->r_modaddr < io_off + io_len) {
4283 			page_t *plist;
4284 			/*
4285 			 * A write is in progress for this region of the file.
4286 			 * If we did not detect RMODINPROGRESS here then this
4287 			 * path through smbfs_putapage() would eventually go to
4288 			 * smbfs_bio() and may not write out all of the data
4289 			 * in the pages. We end up losing data. So we decide
4290 			 * to set the modified bit on each page in the page
4291 			 * list and mark the rnode with RDIRTY. This write
4292 			 * will be restarted at some later time.
4293 			 */
4294 			plist = pp;
4295 			while (plist != NULL) {
4296 				pp = plist;
4297 				page_sub(&plist, pp);
4298 				hat_setmod(pp);
4299 				page_io_unlock(pp);
4300 				page_unlock(pp);
4301 			}
4302 			np->r_flags |= RDIRTY;
4303 			mutex_exit(&np->r_statelock);
4304 			if (offp)
4305 				*offp = io_off;
4306 			if (lenp)
4307 				*lenp = io_len;
4308 			return (0);
4309 		}
4310 		mutex_exit(&np->r_statelock);
4311 	}
4312 
4313 	/*
4314 	 * NFS handles (flags & B_ASYNC) here...
4315 	 * (See nfs_async_putapage())
4316 	 *
4317 	 * This code section from: nfs3_sync_putapage()
4318 	 */
4319 
4320 	flags |= B_WRITE;
4321 
4322 	error = smbfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4323 
4324 	if ((error == ENOSPC || error == EDQUOT || error == EFBIG ||
4325 	    error == EACCES) &&
4326 	    (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) {
4327 		if (!(np->r_flags & ROUTOFSPACE)) {
4328 			mutex_enter(&np->r_statelock);
4329 			np->r_flags |= ROUTOFSPACE;
4330 			mutex_exit(&np->r_statelock);
4331 		}
4332 		flags |= B_ERROR;
4333 		pvn_write_done(pp, flags);
4334 		/*
4335 		 * If this was not an async thread, then try again to
4336 		 * write out the pages, but this time, also destroy
4337 		 * them whether or not the write is successful.  This
4338 		 * will prevent memory from filling up with these
4339 		 * pages and destroying them is the only alternative
4340 		 * if they can't be written out.
4341 		 *
4342 		 * Don't do this if this is an async thread because
4343 		 * when the pages are unlocked in pvn_write_done,
4344 		 * some other thread could have come along, locked
4345 		 * them, and queued for an async thread.  It would be
4346 		 * possible for all of the async threads to be tied
4347 		 * up waiting to lock the pages again and they would
4348 		 * all already be locked and waiting for an async
4349 		 * thread to handle them.  Deadlock.
4350 		 */
4351 		if (!(flags & B_ASYNC)) {
4352 			error = smbfs_putpage(vp, io_off, io_len,
4353 			    B_INVAL | B_FORCE, cr, NULL);
4354 		}
4355 	} else {
4356 		if (error)
4357 			flags |= B_ERROR;
4358 		else if (np->r_flags & ROUTOFSPACE) {
4359 			mutex_enter(&np->r_statelock);
4360 			np->r_flags &= ~ROUTOFSPACE;
4361 			mutex_exit(&np->r_statelock);
4362 		}
4363 		pvn_write_done(pp, flags);
4364 	}
4365 
4366 	/* Now more code from: nfs3_putapage */
4367 
4368 	if (offp)
4369 		*offp = io_off;
4370 	if (lenp)
4371 		*lenp = io_len;
4372 
4373 	return (error);
4374 }
4375 
4376 #endif	// _KERNEL
4377 
4378 
4379 /*
4380  * NFS has this in nfs_client.c (shared by v2,v3,...)
4381  * We have it here so smbfs_putapage can be file scope.
4382  */
4383 void
4384 smbfs_invalidate_pages(vnode_t *vp, u_offset_t off, cred_t *cr)
4385 {
4386 	smbnode_t *np;
4387 
4388 	np = VTOSMB(vp);
4389 
4390 	mutex_enter(&np->r_statelock);
4391 	while (np->r_flags & RTRUNCATE)
4392 		cv_wait(&np->r_cv, &np->r_statelock);
4393 	np->r_flags |= RTRUNCATE;
4394 
4395 	if (off == (u_offset_t)0) {
4396 		np->r_flags &= ~RDIRTY;
4397 		if (!(np->r_flags & RSTALE))
4398 			np->r_error = 0;
4399 	}
4400 	/* Here NFSv3 has np->r_truncaddr = off; */
4401 	mutex_exit(&np->r_statelock);
4402 
4403 #ifdef	_KERNEL
4404 	(void) pvn_vplist_dirty(vp, off, smbfs_putapage,
4405 	    B_INVAL | B_TRUNC, cr);
4406 #endif	// _KERNEL
4407 
4408 	mutex_enter(&np->r_statelock);
4409 	np->r_flags &= ~RTRUNCATE;
4410 	cv_broadcast(&np->r_cv);
4411 	mutex_exit(&np->r_statelock);
4412 }
4413 
4414 #ifdef	_KERNEL
4415 
4416 /* Like nfs3_map */
4417 
4418 /* ARGSUSED */
4419 static int
4420 smbfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4421 	size_t len, uchar_t prot, uchar_t maxprot, uint_t flags,
4422 	cred_t *cr, caller_context_t *ct)
4423 {
4424 	segvn_crargs_t	vn_a;
4425 	struct vattr	va;
4426 	smbnode_t	*np;
4427 	smbmntinfo_t	*smi;
4428 	int		error;
4429 
4430 	np = VTOSMB(vp);
4431 	smi = VTOSMI(vp);
4432 
4433 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
4434 		return (EIO);
4435 
4436 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
4437 		return (EIO);
4438 
4439 	/* Sanity check: should have a valid open */
4440 	if (np->n_fid == NULL)
4441 		return (EIO);
4442 
4443 	if (vp->v_flag & VNOMAP)
4444 		return (ENOSYS);
4445 
4446 	if (off < 0 || off + (ssize_t)len < 0)
4447 		return (ENXIO);
4448 
4449 	if (vp->v_type != VREG)
4450 		return (ENODEV);
4451 
4452 	/*
4453 	 * NFS does close-to-open consistency stuff here.
4454 	 * Just get (possibly cached) attributes.
4455 	 */
4456 	va.va_mask = AT_ALL;
4457 	if ((error = smbfsgetattr(vp, &va, cr)) != 0)
4458 		return (error);
4459 
4460 	/*
4461 	 * Check to see if the vnode is currently marked as not cachable.
4462 	 * This means portions of the file are locked (through VOP_FRLOCK).
4463 	 * In this case the map request must be refused.  We use
4464 	 * rp->r_lkserlock to avoid a race with concurrent lock requests.
4465 	 */
4466 	/*
4467 	 * Atomically increment r_inmap after acquiring r_rwlock. The
4468 	 * idea here is to acquire r_rwlock to block read/write and
4469 	 * not to protect r_inmap. r_inmap will inform smbfs_read/write()
4470 	 * that we are in smbfs_map(). Now, r_rwlock is acquired in order
4471 	 * and we can prevent the deadlock that would have occurred
4472 	 * when smbfs_addmap() would have acquired it out of order.
4473 	 *
4474 	 * Since we are not protecting r_inmap by any lock, we do not
4475 	 * hold any lock when we decrement it. We atomically decrement
4476 	 * r_inmap after we release r_lkserlock.  Note that rwlock is
4477 	 * re-entered as writer in smbfs_addmap (called via as_map).
4478 	 */
4479 
4480 	if (smbfs_rw_enter_sig(&np->r_rwlock, RW_WRITER, SMBINTR(vp)))
4481 		return (EINTR);
4482 	atomic_inc_uint(&np->r_inmap);
4483 	smbfs_rw_exit(&np->r_rwlock);
4484 
4485 	if (smbfs_rw_enter_sig(&np->r_lkserlock, RW_WRITER, SMBINTR(vp))) {
4486 		atomic_dec_uint(&np->r_inmap);
4487 		return (EINTR);
4488 	}
4489 
4490 	if (vp->v_flag & VNOCACHE) {
4491 		error = EAGAIN;
4492 		goto done;
4493 	}
4494 
4495 	/*
4496 	 * Don't allow concurrent locks and mapping if mandatory locking is
4497 	 * enabled.
4498 	 */
4499 	if ((flk_has_remote_locks(vp) || smbfs_lm_has_sleep(vp)) &&
4500 	    MANDLOCK(vp, va.va_mode)) {
4501 		error = EAGAIN;
4502 		goto done;
4503 	}
4504 
4505 	as_rangelock(as);
4506 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4507 	if (error != 0) {
4508 		as_rangeunlock(as);
4509 		goto done;
4510 	}
4511 
4512 	vn_a.vp = vp;
4513 	vn_a.offset = off;
4514 	vn_a.type = (flags & MAP_TYPE);
4515 	vn_a.prot = (uchar_t)prot;
4516 	vn_a.maxprot = (uchar_t)maxprot;
4517 	vn_a.flags = (flags & ~MAP_TYPE);
4518 	vn_a.cred = cr;
4519 	vn_a.amp = NULL;
4520 	vn_a.szc = 0;
4521 	vn_a.lgrp_mem_policy_flags = 0;
4522 
4523 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4524 	as_rangeunlock(as);
4525 
4526 done:
4527 	smbfs_rw_exit(&np->r_lkserlock);
4528 	atomic_dec_uint(&np->r_inmap);
4529 	return (error);
4530 }
4531 
4532 /*
4533  * This uses addmap/delmap functions to hold the SMB FID open as long as
4534  * there are pages mapped in this as/seg.  Increment the FID refs. when
4535  * the maping count goes from zero to non-zero, and release the FID ref
4536  * when the maping count goes from non-zero to zero.
4537  */
4538 
4539 /* ARGSUSED */
4540 static int
4541 smbfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4542 	size_t len, uchar_t prot, uchar_t maxprot, uint_t flags,
4543 	cred_t *cr, caller_context_t *ct)
4544 {
4545 	smbnode_t *np = VTOSMB(vp);
4546 	boolean_t inc_fidrefs = B_FALSE;
4547 
4548 	/*
4549 	 * When r_mapcnt goes from zero to non-zero,
4550 	 * increment n_fidrefs
4551 	 */
4552 	mutex_enter(&np->r_statelock);
4553 	if (np->r_mapcnt == 0)
4554 		inc_fidrefs = B_TRUE;
4555 	np->r_mapcnt += btopr(len);
4556 	mutex_exit(&np->r_statelock);
4557 
4558 	if (inc_fidrefs) {
4559 		(void) smbfs_rw_enter_sig(&np->r_lkserlock, RW_WRITER, 0);
4560 		np->n_fidrefs++;
4561 		smbfs_rw_exit(&np->r_lkserlock);
4562 	}
4563 
4564 	return (0);
4565 }
4566 
4567 /*
4568  * Args passed to smbfs_delmap_async
4569  */
4570 typedef struct smbfs_delmap_args {
4571 	taskq_ent_t		dm_tqent;
4572 	cred_t			*dm_cr;
4573 	vnode_t			*dm_vp;
4574 	offset_t		dm_off;
4575 	caddr_t			dm_addr;
4576 	size_t			dm_len;
4577 	uint_t			dm_prot;
4578 	uint_t			dm_maxprot;
4579 	uint_t			dm_flags;
4580 	boolean_t		dm_rele_fid;
4581 } smbfs_delmap_args_t;
4582 
4583 /*
4584  * Using delmap not only to release the SMB FID (as described above)
4585  * but to flush dirty pages as needed.  Both of those do the actual
4586  * work in an async taskq job to avoid interfering with locks held
4587  * in the VM layer when this is called.
4588  */
4589 
4590 /* ARGSUSED */
4591 static int
4592 smbfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4593 	size_t len, uint_t prot, uint_t maxprot, uint_t flags,
4594 	cred_t *cr, caller_context_t *ct)
4595 {
4596 	smbnode_t		*np = VTOSMB(vp);
4597 	smbmntinfo_t		*smi = VTOSMI(vp);
4598 	smbfs_delmap_args_t	*dmapp;
4599 
4600 	dmapp = kmem_zalloc(sizeof (*dmapp), KM_SLEEP);
4601 
4602 	/*
4603 	 * The VM layer may segvn_free the seg holding this vnode
4604 	 * before our callback has a chance run, so take a hold on
4605 	 * the vnode here and release it in the callback.
4606 	 * (same for the cred)
4607 	 */
4608 	crhold(cr);
4609 	VN_HOLD(vp);
4610 
4611 	dmapp->dm_vp = vp;
4612 	dmapp->dm_cr = cr;
4613 	dmapp->dm_off = off;
4614 	dmapp->dm_addr = addr;
4615 	dmapp->dm_len = len;
4616 	dmapp->dm_prot = prot;
4617 	dmapp->dm_maxprot = maxprot;
4618 	dmapp->dm_flags = flags;
4619 	dmapp->dm_rele_fid = B_FALSE;
4620 
4621 	/*
4622 	 * Go ahead and decrement r_mapcount now, which is
4623 	 * the primary purpose of this function.
4624 	 *
4625 	 * When r_mapcnt goes to zero, we need to call
4626 	 * smbfs_rele_fid, but can't do that here, so
4627 	 * set a flag telling the async task to do it.
4628 	 */
4629 	mutex_enter(&np->r_statelock);
4630 	np->r_mapcnt -= btopr(len);
4631 	ASSERT(np->r_mapcnt >= 0);
4632 	if (np->r_mapcnt == 0)
4633 		dmapp->dm_rele_fid = B_TRUE;
4634 	mutex_exit(&np->r_statelock);
4635 
4636 	taskq_dispatch_ent(smi->smi_taskq, smbfs_delmap_async, dmapp, 0,
4637 	    &dmapp->dm_tqent);
4638 
4639 	return (0);
4640 }
4641 
4642 /*
4643  * Remove some pages from an mmap'd vnode.  Flush any
4644  * dirty pages in the unmapped range.
4645  */
4646 /* ARGSUSED */
4647 static void
4648 smbfs_delmap_async(void *varg)
4649 {
4650 	smbfs_delmap_args_t	*dmapp = varg;
4651 	cred_t			*cr;
4652 	vnode_t			*vp;
4653 	smbnode_t		*np;
4654 	smbmntinfo_t		*smi;
4655 
4656 	cr = dmapp->dm_cr;
4657 	vp = dmapp->dm_vp;
4658 	np = VTOSMB(vp);
4659 	smi = VTOSMI(vp);
4660 
4661 	/* Decremented r_mapcnt in smbfs_delmap */
4662 
4663 	/*
4664 	 * Initiate a page flush and potential commit if there are
4665 	 * pages, the file system was not mounted readonly, the segment
4666 	 * was mapped shared, and the pages themselves were writeable.
4667 	 *
4668 	 * mark RDIRTY here, will be used to check if a file is dirty when
4669 	 * unmount smbfs
4670 	 */
4671 	if (vn_has_cached_data(vp) && !vn_is_readonly(vp) &&
4672 	    dmapp->dm_flags == MAP_SHARED &&
4673 	    (dmapp->dm_maxprot & PROT_WRITE) != 0) {
4674 		mutex_enter(&np->r_statelock);
4675 		np->r_flags |= RDIRTY;
4676 		mutex_exit(&np->r_statelock);
4677 
4678 		/*
4679 		 * Need to finish the putpage before we
4680 		 * close the OtW FID needed for I/O.
4681 		 */
4682 		(void) smbfs_putpage(vp, dmapp->dm_off, dmapp->dm_len, 0,
4683 		    dmapp->dm_cr, NULL);
4684 	}
4685 
4686 	if ((np->r_flags & RDIRECTIO) || (smi->smi_flags & SMI_DIRECTIO))
4687 		(void) smbfs_putpage(vp, dmapp->dm_off, dmapp->dm_len,
4688 		    B_INVAL, dmapp->dm_cr, NULL);
4689 
4690 	/*
4691 	 * If r_mapcnt went to zero, drop our FID ref now.
4692 	 * On the last fidref, this does an OtW close.
4693 	 */
4694 	if (dmapp->dm_rele_fid) {
4695 		struct smb_cred scred;
4696 
4697 		(void) smbfs_rw_enter_sig(&np->r_lkserlock, RW_WRITER, 0);
4698 		smb_credinit(&scred, dmapp->dm_cr);
4699 
4700 		smbfs_rele_fid(np, &scred);
4701 
4702 		smb_credrele(&scred);
4703 		smbfs_rw_exit(&np->r_lkserlock);
4704 	}
4705 
4706 	/* Release holds taken in smbfs_delmap */
4707 	VN_RELE(vp);
4708 	crfree(cr);
4709 
4710 	kmem_free(dmapp, sizeof (*dmapp));
4711 }
4712 
4713 /* No smbfs_pageio() or smbfs_dispose() ops. */
4714 
4715 #endif	// _KERNEL
4716 
4717 /* misc. ******************************************************** */
4718 
4719 
4720 /*
4721  * XXX
4722  * This op may need to support PSARC 2007/440, nbmand changes for CIFS Service.
4723  */
4724 static int
4725 smbfs_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
4726 	offset_t offset, struct flk_callback *flk_cbp, cred_t *cr,
4727 	caller_context_t *ct)
4728 {
4729 	if (curproc->p_zone != VTOSMI(vp)->smi_zone_ref.zref_zone)
4730 		return (EIO);
4731 
4732 	if (VTOSMI(vp)->smi_flags & SMI_LLOCK)
4733 		return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4734 	else
4735 		return (ENOSYS);
4736 }
4737 
4738 /*
4739  * Free storage space associated with the specified vnode.  The portion
4740  * to be freed is specified by bfp->l_start and bfp->l_len (already
4741  * normalized to a "whence" of 0).
4742  *
4743  * Called by fcntl(fd, F_FREESP, lkp) for libc:ftruncate, etc.
4744  */
4745 /* ARGSUSED */
4746 static int
4747 smbfs_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
4748 	offset_t offset, cred_t *cr, caller_context_t *ct)
4749 {
4750 	int		error;
4751 	smbmntinfo_t	*smi;
4752 
4753 	smi = VTOSMI(vp);
4754 
4755 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
4756 		return (EIO);
4757 
4758 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
4759 		return (EIO);
4760 
4761 	/* Caller (fcntl) has checked v_type */
4762 	ASSERT(vp->v_type == VREG);
4763 	if (cmd != F_FREESP)
4764 		return (EINVAL);
4765 
4766 	/*
4767 	 * Like NFS3, no 32-bit offset checks here.
4768 	 * Our SMB layer takes care to return EFBIG
4769 	 * when it has to fallback to a 32-bit call.
4770 	 */
4771 
4772 	error = convoff(vp, bfp, 0, offset);
4773 	if (!error) {
4774 		ASSERT(bfp->l_start >= 0);
4775 		if (bfp->l_len == 0) {
4776 			struct vattr va;
4777 
4778 			/*
4779 			 * ftruncate should not change the ctime and
4780 			 * mtime if we truncate the file to its
4781 			 * previous size.
4782 			 */
4783 			va.va_mask = AT_SIZE;
4784 			error = smbfsgetattr(vp, &va, cr);
4785 			if (error || va.va_size == bfp->l_start)
4786 				return (error);
4787 			va.va_mask = AT_SIZE;
4788 			va.va_size = bfp->l_start;
4789 			error = smbfssetattr(vp, &va, 0, cr);
4790 			/* SMBFS_VNEVENT... */
4791 		} else
4792 			error = EINVAL;
4793 	}
4794 
4795 	return (error);
4796 }
4797 
4798 
4799 /* ARGSUSED */
4800 static int
4801 smbfs_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
4802 {
4803 
4804 	return (ENOSYS);
4805 }
4806 
4807 
4808 /* ARGSUSED */
4809 static int
4810 smbfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4811 	caller_context_t *ct)
4812 {
4813 	vfs_t *vfs;
4814 	smbmntinfo_t *smi;
4815 	struct smb_share *ssp;
4816 
4817 	vfs = vp->v_vfsp;
4818 	smi = VFTOSMI(vfs);
4819 
4820 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
4821 		return (EIO);
4822 
4823 	if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
4824 		return (EIO);
4825 
4826 	switch (cmd) {
4827 	case _PC_FILESIZEBITS:
4828 		ssp = smi->smi_share;
4829 		if (SSTOVC(ssp)->vc_sopt.sv_caps & SMB_CAP_LARGE_FILES)
4830 			*valp = 64;
4831 		else
4832 			*valp = 32;
4833 		break;
4834 
4835 	case _PC_LINK_MAX:
4836 		/* We only ever report one link to an object */
4837 		*valp = 1;
4838 		break;
4839 
4840 	case _PC_ACL_ENABLED:
4841 		/*
4842 		 * Always indicate that ACLs are enabled and
4843 		 * that we support ACE_T format, otherwise
4844 		 * libsec will ask for ACLENT_T format data
4845 		 * which we don't support.
4846 		 */
4847 		*valp = _ACL_ACE_ENABLED;
4848 		break;
4849 
4850 	case _PC_SYMLINK_MAX:	/* No symlinks until we do Unix extensions */
4851 		*valp = 0;
4852 		break;
4853 
4854 	case _PC_XATTR_EXISTS:
4855 		if (vfs->vfs_flag & VFS_XATTR) {
4856 			*valp = smbfs_xa_exists(vp, cr);
4857 			break;
4858 		}
4859 		return (EINVAL);
4860 
4861 	case _PC_SATTR_ENABLED:
4862 	case _PC_SATTR_EXISTS:
4863 		*valp = 1;
4864 		break;
4865 
4866 	case _PC_TIMESTAMP_RESOLUTION:
4867 		/*
4868 		 * Windows times are tenths of microseconds
4869 		 * (multiples of 100 nanoseconds).
4870 		 */
4871 		*valp = 100L;
4872 		break;
4873 
4874 	default:
4875 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4876 	}
4877 	return (0);
4878 }
4879 
4880 /* ARGSUSED */
4881 static int
4882 smbfs_getsecattr(vnode_t *vp, vsecattr_t *vsa, int flag, cred_t *cr,
4883 	caller_context_t *ct)
4884 {
4885 	vfs_t *vfsp;
4886 	smbmntinfo_t *smi;
4887 	int	error;
4888 	uint_t	mask;
4889 
4890 	vfsp = vp->v_vfsp;
4891 	smi = VFTOSMI(vfsp);
4892 
4893 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
4894 		return (EIO);
4895 
4896 	if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
4897 		return (EIO);
4898 
4899 	/*
4900 	 * Our _pathconf indicates _ACL_ACE_ENABLED,
4901 	 * so we should only see VSA_ACE, etc here.
4902 	 * Note: vn_create asks for VSA_DFACLCNT,
4903 	 * and it expects ENOSYS and empty data.
4904 	 */
4905 	mask = vsa->vsa_mask & (VSA_ACE | VSA_ACECNT |
4906 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
4907 	if (mask == 0)
4908 		return (ENOSYS);
4909 
4910 	if (smi->smi_flags & SMI_ACL)
4911 		error = smbfs_acl_getvsa(vp, vsa, flag, cr);
4912 	else
4913 		error = ENOSYS;
4914 
4915 	if (error == ENOSYS)
4916 		error = fs_fab_acl(vp, vsa, flag, cr, ct);
4917 
4918 	return (error);
4919 }
4920 
4921 /* ARGSUSED */
4922 static int
4923 smbfs_setsecattr(vnode_t *vp, vsecattr_t *vsa, int flag, cred_t *cr,
4924 	caller_context_t *ct)
4925 {
4926 	vfs_t *vfsp;
4927 	smbmntinfo_t *smi;
4928 	int	error;
4929 	uint_t	mask;
4930 
4931 	vfsp = vp->v_vfsp;
4932 	smi = VFTOSMI(vfsp);
4933 
4934 	if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
4935 		return (EIO);
4936 
4937 	if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
4938 		return (EIO);
4939 
4940 	/*
4941 	 * Our _pathconf indicates _ACL_ACE_ENABLED,
4942 	 * so we should only see VSA_ACE, etc here.
4943 	 */
4944 	mask = vsa->vsa_mask & (VSA_ACE | VSA_ACECNT);
4945 	if (mask == 0)
4946 		return (ENOSYS);
4947 
4948 	if (vfsp->vfs_flag & VFS_RDONLY)
4949 		return (EROFS);
4950 
4951 	/*
4952 	 * Allow only the mount owner to do this.
4953 	 * See comments at smbfs_access_rwx.
4954 	 */
4955 	error = secpolicy_vnode_setdac(cr, smi->smi_uid);
4956 	if (error != 0)
4957 		return (error);
4958 
4959 	if (smi->smi_flags & SMI_ACL)
4960 		error = smbfs_acl_setvsa(vp, vsa, flag, cr);
4961 	else
4962 		error = ENOSYS;
4963 
4964 	return (error);
4965 }
4966 
4967 
4968 /*
4969  * XXX
4970  * This op should eventually support PSARC 2007/268.
4971  */
4972 static int
4973 smbfs_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr,
4974 	caller_context_t *ct)
4975 {
4976 	if (curproc->p_zone != VTOSMI(vp)->smi_zone_ref.zref_zone)
4977 		return (EIO);
4978 
4979 	if (VTOSMI(vp)->smi_flags & SMI_LLOCK)
4980 		return (fs_shrlock(vp, cmd, shr, flag, cr, ct));
4981 	else
4982 		return (ENOSYS);
4983 }
4984 
4985 
4986 /*
4987  * Most unimplemented ops will return ENOSYS because of fs_nosys().
4988  * The only ops where that won't work are ACCESS (due to open(2)
4989  * failures) and ... (anything else left?)
4990  */
4991 const fs_operation_def_t smbfs_vnodeops_template[] = {
4992 	VOPNAME_OPEN,		{ .vop_open = smbfs_open },
4993 	VOPNAME_CLOSE,		{ .vop_close = smbfs_close },
4994 	VOPNAME_READ,		{ .vop_read = smbfs_read },
4995 	VOPNAME_WRITE,		{ .vop_write = smbfs_write },
4996 	VOPNAME_IOCTL,		{ .vop_ioctl = smbfs_ioctl },
4997 	VOPNAME_GETATTR,	{ .vop_getattr = smbfs_getattr },
4998 	VOPNAME_SETATTR,	{ .vop_setattr = smbfs_setattr },
4999 	VOPNAME_ACCESS,		{ .vop_access = smbfs_access },
5000 	VOPNAME_LOOKUP,		{ .vop_lookup = smbfs_lookup },
5001 	VOPNAME_CREATE,		{ .vop_create = smbfs_create },
5002 	VOPNAME_REMOVE,		{ .vop_remove = smbfs_remove },
5003 	VOPNAME_LINK,		{ .vop_link = smbfs_link },
5004 	VOPNAME_RENAME,		{ .vop_rename = smbfs_rename },
5005 	VOPNAME_MKDIR,		{ .vop_mkdir = smbfs_mkdir },
5006 	VOPNAME_RMDIR,		{ .vop_rmdir = smbfs_rmdir },
5007 	VOPNAME_READDIR,	{ .vop_readdir = smbfs_readdir },
5008 	VOPNAME_SYMLINK,	{ .vop_symlink = smbfs_symlink },
5009 	VOPNAME_READLINK,	{ .vop_readlink = smbfs_readlink },
5010 	VOPNAME_FSYNC,		{ .vop_fsync = smbfs_fsync },
5011 	VOPNAME_INACTIVE,	{ .vop_inactive = smbfs_inactive },
5012 	VOPNAME_FID,		{ .vop_fid = smbfs_fid },
5013 	VOPNAME_RWLOCK,		{ .vop_rwlock = smbfs_rwlock },
5014 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = smbfs_rwunlock },
5015 	VOPNAME_SEEK,		{ .vop_seek = smbfs_seek },
5016 	VOPNAME_FRLOCK,		{ .vop_frlock = smbfs_frlock },
5017 	VOPNAME_SPACE,		{ .vop_space = smbfs_space },
5018 	VOPNAME_REALVP,		{ .vop_realvp = smbfs_realvp },
5019 #ifdef	_KERNEL
5020 	VOPNAME_GETPAGE,	{ .vop_getpage = smbfs_getpage },
5021 	VOPNAME_PUTPAGE,	{ .vop_putpage = smbfs_putpage },
5022 	VOPNAME_MAP,		{ .vop_map = smbfs_map },
5023 	VOPNAME_ADDMAP,		{ .vop_addmap = smbfs_addmap },
5024 	VOPNAME_DELMAP,		{ .vop_delmap = smbfs_delmap },
5025 #endif	// _KERNEL
5026 	VOPNAME_PATHCONF,	{ .vop_pathconf = smbfs_pathconf },
5027 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = smbfs_setsecattr },
5028 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = smbfs_getsecattr },
5029 	VOPNAME_SHRLOCK,	{ .vop_shrlock = smbfs_shrlock },
5030 #ifdef	SMBFS_VNEVENT
5031 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5032 #endif
5033 	{ NULL, NULL }
5034 };
5035