1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 /*
28 * Copyright (c) 2015, Joyent, Inc. All rights reserved.
29 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
30 * Copyright 2016 RackTop Systems.
31 */
32
33 #include <sys/types.h>
34 #include <sys/param.h>
35 #include <sys/t_lock.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/user.h>
39 #include <sys/time.h>
40 #include <sys/vfs.h>
41 #include <sys/vfs_opreg.h>
42 #include <sys/vnode.h>
43 #include <sys/file.h>
44 #include <sys/fcntl.h>
45 #include <sys/flock.h>
46 #include <sys/kmem.h>
47 #include <sys/uio.h>
48 #include <sys/errno.h>
49 #include <sys/stat.h>
50 #include <sys/cred.h>
51 #include <sys/dirent.h>
52 #include <sys/pathname.h>
53 #include <sys/vmsystm.h>
54 #include <sys/fs/tmp.h>
55 #include <sys/fs/tmpnode.h>
56 #include <sys/mman.h>
57 #include <vm/hat.h>
58 #include <vm/seg_vn.h>
59 #include <vm/seg_map.h>
60 #include <vm/seg.h>
61 #include <vm/anon.h>
62 #include <vm/as.h>
63 #include <vm/page.h>
64 #include <vm/pvn.h>
65 #include <sys/cmn_err.h>
66 #include <sys/debug.h>
67 #include <sys/swap.h>
68 #include <sys/buf.h>
69 #include <sys/vm.h>
70 #include <sys/vtrace.h>
71 #include <sys/policy.h>
72 #include <fs/fs_subr.h>
73
74 static int tmp_getapage(struct vnode *, u_offset_t, size_t, uint_t *,
75 page_t **, size_t, struct seg *, caddr_t, enum seg_rw, struct cred *);
76 static int tmp_putapage(struct vnode *, page_t *, u_offset_t *, size_t *,
77 int, struct cred *);
78
79 /* ARGSUSED1 */
80 static int
81 tmp_open(struct vnode **vpp, int flag, struct cred *cred, caller_context_t *ct)
82 {
83 /*
84 * swapon to a tmpfs file is not supported so access
85 * is denied on open if VISSWAP is set.
86 */
87 if ((*vpp)->v_flag & VISSWAP)
88 return (EINVAL);
89 return (0);
90 }
91
92 /* ARGSUSED1 */
93 static int
94 tmp_close(
95 struct vnode *vp,
96 int flag,
97 int count,
98 offset_t offset,
99 struct cred *cred,
100 caller_context_t *ct)
101 {
102 cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
103 cleanshares(vp, ttoproc(curthread)->p_pid);
104 return (0);
105 }
106
107 /*
108 * wrtmp does the real work of write requests for tmpfs.
109 */
110 static int
111 wrtmp(
112 struct tmount *tm,
113 struct tmpnode *tp,
114 struct uio *uio,
115 struct cred *cr,
116 struct caller_context *ct)
117 {
118 pgcnt_t pageoffset; /* offset in pages */
119 ulong_t segmap_offset; /* pagesize byte offset into segmap */
120 caddr_t base; /* base of segmap */
121 ssize_t bytes; /* bytes to uiomove */
122 pfn_t pagenumber; /* offset in pages into tmp file */
123 struct vnode *vp;
124 int error = 0;
125 int pagecreate; /* == 1 if we allocated a page */
126 int newpage;
127 rlim64_t limit = uio->uio_llimit;
128 long oresid = uio->uio_resid;
129 timestruc_t now;
130
131 long tn_size_changed = 0;
132 long old_tn_size;
133 long new_tn_size;
134
135 vp = TNTOV(tp);
136 ASSERT(vp->v_type == VREG);
137
138 TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START,
139 "tmp_wrtmp_start:vp %p", vp);
140
141 ASSERT(RW_WRITE_HELD(&tp->tn_contents));
142 ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
143
144 if (MANDLOCK(vp, tp->tn_mode)) {
145 rw_exit(&tp->tn_contents);
146 /*
147 * tmp_getattr ends up being called by chklock
148 */
149 error = chklock(vp, FWRITE, uio->uio_loffset, uio->uio_resid,
150 uio->uio_fmode, ct);
151 rw_enter(&tp->tn_contents, RW_WRITER);
152 if (error != 0) {
153 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
154 "tmp_wrtmp_end:vp %p error %d", vp, error);
155 return (error);
156 }
157 }
158
159 if (uio->uio_loffset < 0)
160 return (EINVAL);
161
162 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
163 limit = MAXOFFSET_T;
164
165 if (uio->uio_loffset >= limit) {
166 proc_t *p = ttoproc(curthread);
167
168 mutex_enter(&p->p_lock);
169 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
170 p, RCA_UNSAFE_SIGINFO);
171 mutex_exit(&p->p_lock);
172 return (EFBIG);
173 }
174
175 if (uio->uio_loffset >= MAXOFF_T) {
176 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
177 "tmp_wrtmp_end:vp %p error %d", vp, EINVAL);
178 return (EFBIG);
179 }
180
181 if (uio->uio_resid == 0) {
182 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
183 "tmp_wrtmp_end:vp %p error %d", vp, 0);
184 return (0);
185 }
186
187 if (limit > MAXOFF_T)
188 limit = MAXOFF_T;
189
190 do {
191 long offset;
192 long delta;
193
194 offset = (long)uio->uio_offset;
195 pageoffset = offset & PAGEOFFSET;
196 /*
197 * A maximum of PAGESIZE bytes of data is transferred
198 * each pass through this loop
199 */
200 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
201
202 if (offset + bytes >= limit) {
203 if (offset >= limit) {
204 error = EFBIG;
205 goto out;
206 }
207 bytes = limit - offset;
208 }
209 pagenumber = btop(offset);
210
211 /*
212 * delta is the amount of anonymous memory
213 * to reserve for the file.
214 * We always reserve in pagesize increments so
215 * unless we're extending the file into a new page,
216 * we don't need to call tmp_resv.
217 */
218 delta = offset + bytes -
219 P2ROUNDUP_TYPED(tp->tn_size, PAGESIZE, u_offset_t);
220 if (delta > 0) {
221 pagecreate = 1;
222 if (tmp_resv(tm, tp, delta, pagecreate)) {
223 /*
224 * Log file system full in the zone that owns
225 * the tmpfs mount, as well as in the global
226 * zone if necessary.
227 */
228 zcmn_err(tm->tm_vfsp->vfs_zone->zone_id,
229 CE_WARN, "%s: File system full, "
230 "swap space limit exceeded",
231 tm->tm_mntpath);
232
233 if (tm->tm_vfsp->vfs_zone->zone_id !=
234 GLOBAL_ZONEID) {
235
236 vfs_t *vfs = tm->tm_vfsp;
237
238 zcmn_err(GLOBAL_ZONEID,
239 CE_WARN, "%s: File system full, "
240 "swap space limit exceeded",
241 vfs->vfs_vnodecovered->v_path);
242 }
243 error = ENOSPC;
244 break;
245 }
246 tmpnode_growmap(tp, (ulong_t)offset + bytes);
247 }
248 /* grow the file to the new length */
249 if (offset + bytes > tp->tn_size) {
250 tn_size_changed = 1;
251 old_tn_size = tp->tn_size;
252 /*
253 * Postpone updating tp->tn_size until uiomove() is
254 * done.
255 */
256 new_tn_size = offset + bytes;
257 }
258 if (bytes == PAGESIZE) {
259 /*
260 * Writing whole page so reading from disk
261 * is a waste
262 */
263 pagecreate = 1;
264 } else {
265 pagecreate = 0;
266 }
267 /*
268 * If writing past EOF or filling in a hole
269 * we need to allocate an anon slot.
270 */
271 if (anon_get_ptr(tp->tn_anon, pagenumber) == NULL) {
272 (void) anon_set_ptr(tp->tn_anon, pagenumber,
273 anon_alloc(vp, ptob(pagenumber)), ANON_SLEEP);
274 pagecreate = 1;
275 tp->tn_nblocks++;
276 }
277
278 /*
279 * We have to drop the contents lock to allow the VM
280 * system to reacquire it in tmp_getpage()
281 */
282 rw_exit(&tp->tn_contents);
283
284 /*
285 * Touch the page and fault it in if it is not in core
286 * before segmap_getmapflt or vpm_data_copy can lock it.
287 * This is to avoid the deadlock if the buffer is mapped
288 * to the same file through mmap which we want to write.
289 */
290 uio_prefaultpages((long)bytes, uio);
291
292 newpage = 0;
293 if (vpm_enable) {
294 /*
295 * Copy data. If new pages are created, part of
296 * the page that is not written will be initizliazed
297 * with zeros.
298 */
299 error = vpm_data_copy(vp, offset, bytes, uio,
300 !pagecreate, &newpage, 1, S_WRITE);
301 } else {
302 /* Get offset within the segmap mapping */
303 segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
304 base = segmap_getmapflt(segkmap, vp,
305 (offset & MAXBMASK), PAGESIZE, !pagecreate,
306 S_WRITE);
307 }
308
309
310 if (!vpm_enable && pagecreate) {
311 /*
312 * segmap_pagecreate() returns 1 if it calls
313 * page_create_va() to allocate any pages.
314 */
315 newpage = segmap_pagecreate(segkmap,
316 base + segmap_offset, (size_t)PAGESIZE, 0);
317 /*
318 * Clear from the beginning of the page to the starting
319 * offset of the data.
320 */
321 if (pageoffset != 0)
322 (void) kzero(base + segmap_offset,
323 (size_t)pageoffset);
324 }
325
326 if (!vpm_enable) {
327 error = uiomove(base + segmap_offset + pageoffset,
328 (long)bytes, UIO_WRITE, uio);
329 }
330
331 if (!vpm_enable && pagecreate &&
332 uio->uio_offset < P2ROUNDUP(offset + bytes, PAGESIZE)) {
333 long zoffset; /* zero from offset into page */
334 /*
335 * We created pages w/o initializing them completely,
336 * thus we need to zero the part that wasn't set up.
337 * This happens on most EOF write cases and if
338 * we had some sort of error during the uiomove.
339 */
340 long nmoved;
341
342 nmoved = uio->uio_offset - offset;
343 ASSERT((nmoved + pageoffset) <= PAGESIZE);
344
345 /*
346 * Zero from the end of data in the page to the
347 * end of the page.
348 */
349 if ((zoffset = pageoffset + nmoved) < PAGESIZE)
350 (void) kzero(base + segmap_offset + zoffset,
351 (size_t)PAGESIZE - zoffset);
352 }
353
354 /*
355 * Unlock the pages which have been allocated by
356 * page_create_va() in segmap_pagecreate()
357 */
358 if (!vpm_enable && newpage) {
359 segmap_pageunlock(segkmap, base + segmap_offset,
360 (size_t)PAGESIZE, S_WRITE);
361 }
362
363 if (error) {
364 /*
365 * If we failed on a write, we must
366 * be sure to invalidate any pages that may have
367 * been allocated.
368 */
369 if (vpm_enable) {
370 (void) vpm_sync_pages(vp, offset, PAGESIZE,
371 SM_INVAL);
372 } else {
373 (void) segmap_release(segkmap, base, SM_INVAL);
374 }
375 } else {
376 if (vpm_enable) {
377 error = vpm_sync_pages(vp, offset, PAGESIZE,
378 0);
379 } else {
380 error = segmap_release(segkmap, base, 0);
381 }
382 }
383
384 /*
385 * Re-acquire contents lock.
386 */
387 rw_enter(&tp->tn_contents, RW_WRITER);
388
389 /*
390 * Update tn_size.
391 */
392 if (tn_size_changed)
393 tp->tn_size = new_tn_size;
394
395 /*
396 * If the uiomove failed, fix up tn_size.
397 */
398 if (error) {
399 if (tn_size_changed) {
400 /*
401 * The uiomove failed, and we
402 * allocated blocks,so get rid
403 * of them.
404 */
405 (void) tmpnode_trunc(tm, tp,
406 (ulong_t)old_tn_size);
407 }
408 } else {
409 /*
410 * XXX - Can this be out of the loop?
411 */
412 if ((tp->tn_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) &&
413 (tp->tn_mode & (S_ISUID | S_ISGID)) &&
414 secpolicy_vnode_setid_retain(cr,
415 (tp->tn_mode & S_ISUID) != 0 && tp->tn_uid == 0)) {
416 /*
417 * Clear Set-UID & Set-GID bits on
418 * successful write if not privileged
419 * and at least one of the execute bits
420 * is set. If we always clear Set-GID,
421 * mandatory file and record locking is
422 * unuseable.
423 */
424 tp->tn_mode &= ~(S_ISUID | S_ISGID);
425 }
426 gethrestime(&now);
427 tp->tn_mtime = now;
428 tp->tn_ctime = now;
429 }
430 } while (error == 0 && uio->uio_resid > 0 && bytes != 0);
431
432 out:
433 /*
434 * If we've already done a partial-write, terminate
435 * the write but return no error.
436 */
437 if (oresid != uio->uio_resid)
438 error = 0;
439 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
440 "tmp_wrtmp_end:vp %p error %d", vp, error);
441 return (error);
442 }
443
444 /*
445 * rdtmp does the real work of read requests for tmpfs.
446 */
447 static int
448 rdtmp(
449 struct tmount *tm,
450 struct tmpnode *tp,
451 struct uio *uio,
452 struct caller_context *ct)
453 {
454 ulong_t pageoffset; /* offset in tmpfs file (uio_offset) */
455 ulong_t segmap_offset; /* pagesize byte offset into segmap */
456 caddr_t base; /* base of segmap */
457 ssize_t bytes; /* bytes to uiomove */
458 struct vnode *vp;
459 int error;
460 long oresid = uio->uio_resid;
461
462 #if defined(lint)
463 tm = tm;
464 #endif
465 vp = TNTOV(tp);
466
467 TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START, "tmp_rdtmp_start:vp %p",
468 vp);
469
470 ASSERT(RW_LOCK_HELD(&tp->tn_contents));
471
472 if (MANDLOCK(vp, tp->tn_mode)) {
473 rw_exit(&tp->tn_contents);
474 /*
475 * tmp_getattr ends up being called by chklock
476 */
477 error = chklock(vp, FREAD, uio->uio_loffset, uio->uio_resid,
478 uio->uio_fmode, ct);
479 rw_enter(&tp->tn_contents, RW_READER);
480 if (error != 0) {
481 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
482 "tmp_rdtmp_end:vp %p error %d", vp, error);
483 return (error);
484 }
485 }
486 ASSERT(tp->tn_type == VREG);
487
488 if (uio->uio_loffset >= MAXOFF_T) {
489 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
490 "tmp_rdtmp_end:vp %p error %d", vp, EINVAL);
491 return (0);
492 }
493 if (uio->uio_loffset < 0)
494 return (EINVAL);
495 if (uio->uio_resid == 0) {
496 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
497 "tmp_rdtmp_end:vp %p error %d", vp, 0);
498 return (0);
499 }
500
501 vp = TNTOV(tp);
502
503 do {
504 long diff;
505 long offset;
506
507 offset = uio->uio_offset;
508 pageoffset = offset & PAGEOFFSET;
509 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
510
511 diff = tp->tn_size - offset;
512
513 if (diff <= 0) {
514 error = 0;
515 goto out;
516 }
517 if (diff < bytes)
518 bytes = diff;
519
520 /*
521 * We have to drop the contents lock to allow the VM system
522 * to reacquire it in tmp_getpage() should the uiomove cause a
523 * pagefault.
524 */
525 rw_exit(&tp->tn_contents);
526
527 if (vpm_enable) {
528 /*
529 * Copy data.
530 */
531 error = vpm_data_copy(vp, offset, bytes, uio, 1, NULL,
532 0, S_READ);
533 } else {
534 segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
535 base = segmap_getmapflt(segkmap, vp, offset & MAXBMASK,
536 bytes, 1, S_READ);
537
538 error = uiomove(base + segmap_offset + pageoffset,
539 (long)bytes, UIO_READ, uio);
540 }
541
542 if (error) {
543 if (vpm_enable) {
544 (void) vpm_sync_pages(vp, offset, PAGESIZE, 0);
545 } else {
546 (void) segmap_release(segkmap, base, 0);
547 }
548 } else {
549 if (vpm_enable) {
550 error = vpm_sync_pages(vp, offset, PAGESIZE,
551 0);
552 } else {
553 error = segmap_release(segkmap, base, 0);
554 }
555 }
556
557 /*
558 * Re-acquire contents lock.
559 */
560 rw_enter(&tp->tn_contents, RW_READER);
561
562 } while (error == 0 && uio->uio_resid > 0);
563
564 out:
565 gethrestime(&tp->tn_atime);
566
567 /*
568 * If we've already done a partial read, terminate
569 * the read but return no error.
570 */
571 if (oresid != uio->uio_resid)
572 error = 0;
573
574 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
575 "tmp_rdtmp_end:vp %x error %d", vp, error);
576 return (error);
577 }
578
579 /* ARGSUSED2 */
580 static int
581 tmp_read(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cred,
582 struct caller_context *ct)
583 {
584 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
585 struct tmount *tm = (struct tmount *)VTOTM(vp);
586 int error;
587
588 /*
589 * We don't currently support reading non-regular files
590 */
591 if (vp->v_type == VDIR)
592 return (EISDIR);
593 if (vp->v_type != VREG)
594 return (EINVAL);
595 /*
596 * tmp_rwlock should have already been called from layers above
597 */
598 ASSERT(RW_READ_HELD(&tp->tn_rwlock));
599
600 rw_enter(&tp->tn_contents, RW_READER);
601
602 error = rdtmp(tm, tp, uiop, ct);
603
604 rw_exit(&tp->tn_contents);
605
606 return (error);
607 }
608
609 static int
610 tmp_write(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cred,
611 struct caller_context *ct)
612 {
613 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
614 struct tmount *tm = (struct tmount *)VTOTM(vp);
615 int error;
616
617 /*
618 * We don't currently support writing to non-regular files
619 */
620 if (vp->v_type != VREG)
621 return (EINVAL); /* XXX EISDIR? */
622
623 /*
624 * tmp_rwlock should have already been called from layers above
625 */
626 ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
627
628 rw_enter(&tp->tn_contents, RW_WRITER);
629
630 if (ioflag & FAPPEND) {
631 /*
632 * In append mode start at end of file.
633 */
634 uiop->uio_loffset = tp->tn_size;
635 }
636
637 error = wrtmp(tm, tp, uiop, cred, ct);
638
639 rw_exit(&tp->tn_contents);
640
641 return (error);
642 }
643
644 /* ARGSUSED */
645 static int
646 tmp_ioctl(
647 struct vnode *vp,
648 int com,
649 intptr_t data,
650 int flag,
651 struct cred *cred,
652 int *rvalp,
653 caller_context_t *ct)
654 {
655 return (ENOTTY);
656 }
657
658 /* ARGSUSED2 */
659 static int
660 tmp_getattr(
661 struct vnode *vp,
662 struct vattr *vap,
663 int flags,
664 struct cred *cred,
665 caller_context_t *ct)
666 {
667 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
668 struct vnode *mvp;
669 struct vattr va;
670 int attrs = 1;
671
672 /*
673 * A special case to handle the root tnode on a diskless nfs
674 * client who may have had its uid and gid inherited
675 * from an nfs vnode with nobody ownership. Likely the
676 * root filesystem. After nfs is fully functional the uid/gid
677 * may be mapable so ask again.
678 * vfsp can't get unmounted because we hold vp.
679 */
680 if (vp->v_flag & VROOT &&
681 (mvp = vp->v_vfsp->vfs_vnodecovered) != NULL) {
682 mutex_enter(&tp->tn_tlock);
683 if (tp->tn_uid == UID_NOBODY || tp->tn_gid == GID_NOBODY) {
684 mutex_exit(&tp->tn_tlock);
685 bzero(&va, sizeof (struct vattr));
686 va.va_mask = AT_UID|AT_GID;
687 attrs = VOP_GETATTR(mvp, &va, 0, cred, ct);
688 } else {
689 mutex_exit(&tp->tn_tlock);
690 }
691 }
692 mutex_enter(&tp->tn_tlock);
693 if (attrs == 0) {
694 tp->tn_uid = va.va_uid;
695 tp->tn_gid = va.va_gid;
696 }
697 vap->va_type = vp->v_type;
698 vap->va_mode = tp->tn_mode & MODEMASK;
699 vap->va_uid = tp->tn_uid;
700 vap->va_gid = tp->tn_gid;
701 vap->va_fsid = tp->tn_fsid;
702 vap->va_nodeid = (ino64_t)tp->tn_nodeid;
703 vap->va_nlink = tp->tn_nlink;
704 vap->va_size = (u_offset_t)tp->tn_size;
705 vap->va_atime = tp->tn_atime;
706 vap->va_mtime = tp->tn_mtime;
707 vap->va_ctime = tp->tn_ctime;
708 vap->va_blksize = PAGESIZE;
709 vap->va_rdev = tp->tn_rdev;
710 vap->va_seq = tp->tn_seq;
711
712 /*
713 * XXX Holes are not taken into account. We could take the time to
714 * run through the anon array looking for allocated slots...
715 */
716 vap->va_nblocks = (fsblkcnt64_t)btodb(ptob(btopr(vap->va_size)));
717 mutex_exit(&tp->tn_tlock);
718 return (0);
719 }
720
721 /*ARGSUSED4*/
722 static int
723 tmp_setattr(
724 struct vnode *vp,
725 struct vattr *vap,
726 int flags,
727 struct cred *cred,
728 caller_context_t *ct)
729 {
730 struct tmount *tm = (struct tmount *)VTOTM(vp);
731 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
732 int error = 0;
733 struct vattr *get;
734 long mask;
735
736 /*
737 * Cannot set these attributes
738 */
739 if ((vap->va_mask & AT_NOSET) || (vap->va_mask & AT_XVATTR))
740 return (EINVAL);
741
742 mutex_enter(&tp->tn_tlock);
743
744 get = &tp->tn_attr;
745 /*
746 * Change file access modes. Must be owner or have sufficient
747 * privileges.
748 */
749 error = secpolicy_vnode_setattr(cred, vp, vap, get, flags, tmp_taccess,
750 tp);
751
752 if (error)
753 goto out;
754
755 mask = vap->va_mask;
756
757 if (mask & AT_MODE) {
758 get->va_mode &= S_IFMT;
759 get->va_mode |= vap->va_mode & ~S_IFMT;
760 }
761
762 if (mask & AT_UID)
763 get->va_uid = vap->va_uid;
764 if (mask & AT_GID)
765 get->va_gid = vap->va_gid;
766 if (mask & AT_ATIME)
767 get->va_atime = vap->va_atime;
768 if (mask & AT_MTIME)
769 get->va_mtime = vap->va_mtime;
770
771 if (mask & (AT_UID | AT_GID | AT_MODE | AT_MTIME))
772 gethrestime(&tp->tn_ctime);
773
774 if (mask & AT_SIZE) {
775 ASSERT(vp->v_type != VDIR);
776
777 /* Don't support large files. */
778 if (vap->va_size > MAXOFF_T) {
779 error = EFBIG;
780 goto out;
781 }
782 mutex_exit(&tp->tn_tlock);
783
784 rw_enter(&tp->tn_rwlock, RW_WRITER);
785 rw_enter(&tp->tn_contents, RW_WRITER);
786 error = tmpnode_trunc(tm, tp, (ulong_t)vap->va_size);
787 rw_exit(&tp->tn_contents);
788 rw_exit(&tp->tn_rwlock);
789
790 if (error == 0 && vap->va_size == 0)
791 vnevent_truncate(vp, ct);
792
793 goto out1;
794 }
795 out:
796 mutex_exit(&tp->tn_tlock);
797 out1:
798 return (error);
799 }
800
801 /* ARGSUSED2 */
802 static int
803 tmp_access(
804 struct vnode *vp,
805 int mode,
806 int flags,
807 struct cred *cred,
808 caller_context_t *ct)
809 {
810 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
811 int error;
812
813 mutex_enter(&tp->tn_tlock);
814 error = tmp_taccess(tp, mode, cred);
815 mutex_exit(&tp->tn_tlock);
816 return (error);
817 }
818
819 /* ARGSUSED3 */
820 static int
821 tmp_lookup(
822 struct vnode *dvp,
823 char *nm,
824 struct vnode **vpp,
825 struct pathname *pnp,
826 int flags,
827 struct vnode *rdir,
828 struct cred *cred,
829 caller_context_t *ct,
830 int *direntflags,
831 pathname_t *realpnp)
832 {
833 struct tmpnode *tp = (struct tmpnode *)VTOTN(dvp);
834 struct tmpnode *ntp = NULL;
835 int error;
836
837
838 /* allow cd into @ dir */
839 if (flags & LOOKUP_XATTR) {
840 struct tmpnode *xdp;
841 struct tmount *tm;
842
843 /*
844 * don't allow attributes if not mounted XATTR support
845 */
846 if (!(dvp->v_vfsp->vfs_flag & VFS_XATTR))
847 return (EINVAL);
848
849 if (tp->tn_flags & ISXATTR)
850 /* No attributes on attributes */
851 return (EINVAL);
852
853 rw_enter(&tp->tn_rwlock, RW_WRITER);
854 if (tp->tn_xattrdp == NULL) {
855 if (!(flags & CREATE_XATTR_DIR)) {
856 rw_exit(&tp->tn_rwlock);
857 return (ENOENT);
858 }
859
860 /*
861 * No attribute directory exists for this
862 * node - create the attr dir as a side effect
863 * of this lookup.
864 */
865
866 /*
867 * Make sure we have adequate permission...
868 */
869
870 if ((error = tmp_taccess(tp, VWRITE, cred)) != 0) {
871 rw_exit(&tp->tn_rwlock);
872 return (error);
873 }
874
875 xdp = tmp_memalloc(sizeof (struct tmpnode),
876 TMP_MUSTHAVE);
877 tm = VTOTM(dvp);
878 tmpnode_init(tm, xdp, &tp->tn_attr, NULL);
879 /*
880 * Fix-up fields unique to attribute directories.
881 */
882 xdp->tn_flags = ISXATTR;
883 xdp->tn_type = VDIR;
884 if (tp->tn_type == VDIR) {
885 xdp->tn_mode = tp->tn_attr.va_mode;
886 } else {
887 xdp->tn_mode = 0700;
888 if (tp->tn_attr.va_mode & 0040)
889 xdp->tn_mode |= 0750;
890 if (tp->tn_attr.va_mode & 0004)
891 xdp->tn_mode |= 0705;
892 }
893 xdp->tn_vnode->v_type = VDIR;
894 xdp->tn_vnode->v_flag |= V_XATTRDIR;
895 tdirinit(tp, xdp);
896 tp->tn_xattrdp = xdp;
897 } else {
898 VN_HOLD(tp->tn_xattrdp->tn_vnode);
899 }
900 *vpp = TNTOV(tp->tn_xattrdp);
901 rw_exit(&tp->tn_rwlock);
902 return (0);
903 }
904
905 /*
906 * Null component name is a synonym for directory being searched.
907 */
908 if (*nm == '\0') {
909 VN_HOLD(dvp);
910 *vpp = dvp;
911 return (0);
912 }
913 ASSERT(tp);
914
915 error = tdirlookup(tp, nm, &ntp, cred);
916
917 if (error == 0) {
918 ASSERT(ntp);
919 *vpp = TNTOV(ntp);
920 /*
921 * If vnode is a device return special vnode instead
922 */
923 if (IS_DEVVP(*vpp)) {
924 struct vnode *newvp;
925
926 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
927 cred);
928 VN_RELE(*vpp);
929 *vpp = newvp;
930 }
931 }
932 TRACE_4(TR_FAC_TMPFS, TR_TMPFS_LOOKUP,
933 "tmpfs lookup:vp %p name %s vpp %p error %d",
934 dvp, nm, vpp, error);
935 return (error);
936 }
937
938 /*ARGSUSED7*/
939 static int
940 tmp_create(
941 struct vnode *dvp,
942 char *nm,
943 struct vattr *vap,
944 enum vcexcl exclusive,
945 int mode,
946 struct vnode **vpp,
947 struct cred *cred,
948 int flag,
949 caller_context_t *ct,
950 vsecattr_t *vsecp)
951 {
952 struct tmpnode *parent;
953 struct tmount *tm;
954 struct tmpnode *self;
955 int error;
956 struct tmpnode *oldtp;
957
958 again:
959 parent = (struct tmpnode *)VTOTN(dvp);
960 tm = (struct tmount *)VTOTM(dvp);
961 self = NULL;
962 error = 0;
963 oldtp = NULL;
964
965 /* device files not allowed in ext. attr dirs */
966 if ((parent->tn_flags & ISXATTR) &&
967 (vap->va_type == VBLK || vap->va_type == VCHR ||
968 vap->va_type == VFIFO || vap->va_type == VDOOR ||
969 vap->va_type == VSOCK || vap->va_type == VPORT))
970 return (EINVAL);
971
972 if (vap->va_type == VREG && (vap->va_mode & VSVTX)) {
973 /* Must be privileged to set sticky bit */
974 if (secpolicy_vnode_stky_modify(cred))
975 vap->va_mode &= ~VSVTX;
976 } else if (vap->va_type == VNON) {
977 return (EINVAL);
978 }
979
980 /*
981 * Null component name is a synonym for directory being searched.
982 */
983 if (*nm == '\0') {
984 VN_HOLD(dvp);
985 oldtp = parent;
986 } else {
987 error = tdirlookup(parent, nm, &oldtp, cred);
988 }
989
990 if (error == 0) { /* name found */
991 boolean_t trunc = B_FALSE;
992
993 ASSERT(oldtp);
994
995 rw_enter(&oldtp->tn_rwlock, RW_WRITER);
996
997 /*
998 * if create/read-only an existing
999 * directory, allow it
1000 */
1001 if (exclusive == EXCL)
1002 error = EEXIST;
1003 else if ((oldtp->tn_type == VDIR) && (mode & VWRITE))
1004 error = EISDIR;
1005 else {
1006 error = tmp_taccess(oldtp, mode, cred);
1007 }
1008
1009 if (error) {
1010 rw_exit(&oldtp->tn_rwlock);
1011 tmpnode_rele(oldtp);
1012 return (error);
1013 }
1014 *vpp = TNTOV(oldtp);
1015 if ((*vpp)->v_type == VREG && (vap->va_mask & AT_SIZE) &&
1016 vap->va_size == 0) {
1017 rw_enter(&oldtp->tn_contents, RW_WRITER);
1018 (void) tmpnode_trunc(tm, oldtp, 0);
1019 rw_exit(&oldtp->tn_contents);
1020 trunc = B_TRUE;
1021 }
1022 rw_exit(&oldtp->tn_rwlock);
1023 if (IS_DEVVP(*vpp)) {
1024 struct vnode *newvp;
1025
1026 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
1027 cred);
1028 VN_RELE(*vpp);
1029 if (newvp == NULL) {
1030 return (ENOSYS);
1031 }
1032 *vpp = newvp;
1033 }
1034
1035 if (trunc)
1036 vnevent_create(*vpp, ct);
1037
1038 return (0);
1039 }
1040
1041 if (error != ENOENT)
1042 return (error);
1043
1044 rw_enter(&parent->tn_rwlock, RW_WRITER);
1045 error = tdirenter(tm, parent, nm, DE_CREATE,
1046 (struct tmpnode *)NULL, (struct tmpnode *)NULL,
1047 vap, &self, cred, ct);
1048 rw_exit(&parent->tn_rwlock);
1049
1050 if (error) {
1051 if (self)
1052 tmpnode_rele(self);
1053
1054 if (error == EEXIST) {
1055 /*
1056 * This means that the file was created sometime
1057 * after we checked and did not find it and when
1058 * we went to create it.
1059 * Since creat() is supposed to truncate a file
1060 * that already exits go back to the begining
1061 * of the function. This time we will find it
1062 * and go down the tmp_trunc() path
1063 */
1064 goto again;
1065 }
1066 return (error);
1067 }
1068
1069 *vpp = TNTOV(self);
1070
1071 if (!error && IS_DEVVP(*vpp)) {
1072 struct vnode *newvp;
1073
1074 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cred);
1075 VN_RELE(*vpp);
1076 if (newvp == NULL)
1077 return (ENOSYS);
1078 *vpp = newvp;
1079 }
1080 TRACE_3(TR_FAC_TMPFS, TR_TMPFS_CREATE,
1081 "tmpfs create:dvp %p nm %s vpp %p", dvp, nm, vpp);
1082 return (0);
1083 }
1084
1085 /* ARGSUSED3 */
1086 static int
1087 tmp_remove(
1088 struct vnode *dvp,
1089 char *nm,
1090 struct cred *cred,
1091 caller_context_t *ct,
1092 int flags)
1093 {
1094 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1095 int error;
1096 struct tmpnode *tp = NULL;
1097
1098 error = tdirlookup(parent, nm, &tp, cred);
1099 if (error)
1100 return (error);
1101
1102 ASSERT(tp);
1103 rw_enter(&parent->tn_rwlock, RW_WRITER);
1104 rw_enter(&tp->tn_rwlock, RW_WRITER);
1105
1106 error = tp->tn_type == VDIR ? EPERM :
1107 tdirdelete(parent, tp, nm, DR_REMOVE, cred);
1108
1109 rw_exit(&tp->tn_rwlock);
1110 rw_exit(&parent->tn_rwlock);
1111 vnevent_remove(TNTOV(tp), dvp, nm, ct);
1112 tmpnode_rele(tp);
1113
1114 TRACE_3(TR_FAC_TMPFS, TR_TMPFS_REMOVE,
1115 "tmpfs remove:dvp %p nm %s error %d", dvp, nm, error);
1116 return (error);
1117 }
1118
1119 /* ARGSUSED4 */
1120 static int
1121 tmp_link(
1122 struct vnode *dvp,
1123 struct vnode *srcvp,
1124 char *tnm,
1125 struct cred *cred,
1126 caller_context_t *ct,
1127 int flags)
1128 {
1129 struct tmpnode *parent;
1130 struct tmpnode *from;
1131 struct tmount *tm = (struct tmount *)VTOTM(dvp);
1132 int error;
1133 struct tmpnode *found = NULL;
1134 struct vnode *realvp;
1135
1136 if (VOP_REALVP(srcvp, &realvp, ct) == 0)
1137 srcvp = realvp;
1138
1139 parent = (struct tmpnode *)VTOTN(dvp);
1140 from = (struct tmpnode *)VTOTN(srcvp);
1141
1142 if (srcvp->v_type == VDIR ||
1143 (from->tn_uid != crgetuid(cred) && secpolicy_basic_link(cred)))
1144 return (EPERM);
1145
1146 /*
1147 * Make sure link for extended attributes is valid
1148 * We only support hard linking of xattr's in xattrdir to an xattrdir
1149 */
1150 if ((from->tn_flags & ISXATTR) != (parent->tn_flags & ISXATTR))
1151 return (EINVAL);
1152
1153 error = tdirlookup(parent, tnm, &found, cred);
1154 if (error == 0) {
1155 ASSERT(found);
1156 tmpnode_rele(found);
1157 return (EEXIST);
1158 }
1159
1160 if (error != ENOENT)
1161 return (error);
1162
1163 rw_enter(&parent->tn_rwlock, RW_WRITER);
1164 error = tdirenter(tm, parent, tnm, DE_LINK, (struct tmpnode *)NULL,
1165 from, NULL, (struct tmpnode **)NULL, cred, ct);
1166 rw_exit(&parent->tn_rwlock);
1167 if (error == 0) {
1168 vnevent_link(srcvp, ct);
1169 }
1170 return (error);
1171 }
1172
1173 /* ARGSUSED5 */
1174 static int
1175 tmp_rename(
1176 struct vnode *odvp, /* source parent vnode */
1177 char *onm, /* source name */
1178 struct vnode *ndvp, /* destination parent vnode */
1179 char *nnm, /* destination name */
1180 struct cred *cred,
1181 caller_context_t *ct,
1182 int flags)
1183 {
1184 struct tmpnode *fromparent;
1185 struct tmpnode *toparent;
1186 struct tmpnode *fromtp = NULL; /* source tmpnode */
1187 struct tmpnode *totp; /* target tmpnode */
1188 struct tmount *tm = (struct tmount *)VTOTM(odvp);
1189 int error;
1190 int samedir = 0; /* set if odvp == ndvp */
1191 struct vnode *realvp;
1192
1193 if (VOP_REALVP(ndvp, &realvp, ct) == 0)
1194 ndvp = realvp;
1195
1196 fromparent = (struct tmpnode *)VTOTN(odvp);
1197 toparent = (struct tmpnode *)VTOTN(ndvp);
1198
1199 if ((fromparent->tn_flags & ISXATTR) != (toparent->tn_flags & ISXATTR))
1200 return (EINVAL);
1201
1202 mutex_enter(&tm->tm_renamelck);
1203
1204 /*
1205 * Look up tmpnode of file we're supposed to rename.
1206 */
1207 error = tdirlookup(fromparent, onm, &fromtp, cred);
1208 if (error) {
1209 mutex_exit(&tm->tm_renamelck);
1210 return (error);
1211 }
1212
1213 /*
1214 * Make sure we can delete the old (source) entry. This
1215 * requires write permission on the containing directory. If
1216 * that directory is "sticky" it requires further checks.
1217 */
1218 if (((error = tmp_taccess(fromparent, VWRITE, cred)) != 0) ||
1219 (error = tmp_sticky_remove_access(fromparent, fromtp, cred)) != 0)
1220 goto done;
1221
1222 /*
1223 * Check for renaming to or from '.' or '..' or that
1224 * fromtp == fromparent
1225 */
1226 if ((onm[0] == '.' &&
1227 (onm[1] == '\0' || (onm[1] == '.' && onm[2] == '\0'))) ||
1228 (nnm[0] == '.' &&
1229 (nnm[1] == '\0' || (nnm[1] == '.' && nnm[2] == '\0'))) ||
1230 (fromparent == fromtp)) {
1231 error = EINVAL;
1232 goto done;
1233 }
1234
1235 samedir = (fromparent == toparent);
1236 /*
1237 * Make sure we can search and rename into the new
1238 * (destination) directory.
1239 */
1240 if (!samedir) {
1241 error = tmp_taccess(toparent, VEXEC|VWRITE, cred);
1242 if (error)
1243 goto done;
1244 }
1245
1246 if (tdirlookup(toparent, nnm, &totp, cred) == 0) {
1247 vnevent_pre_rename_dest(TNTOV(totp), ndvp, nnm, ct);
1248 tmpnode_rele(totp);
1249 }
1250
1251 /* Notify the target dir. if not the same as the source dir. */
1252 if (ndvp != odvp) {
1253 vnevent_pre_rename_dest_dir(ndvp, TNTOV(fromtp), nnm, ct);
1254 }
1255
1256 vnevent_pre_rename_src(TNTOV(fromtp), odvp, onm, ct);
1257
1258 /*
1259 * Link source to new target
1260 */
1261 rw_enter(&toparent->tn_rwlock, RW_WRITER);
1262 error = tdirenter(tm, toparent, nnm, DE_RENAME,
1263 fromparent, fromtp, (struct vattr *)NULL,
1264 (struct tmpnode **)NULL, cred, ct);
1265 rw_exit(&toparent->tn_rwlock);
1266
1267 if (error) {
1268 /*
1269 * ESAME isn't really an error; it indicates that the
1270 * operation should not be done because the source and target
1271 * are the same file, but that no error should be reported.
1272 */
1273 if (error == ESAME)
1274 error = 0;
1275 goto done;
1276 }
1277
1278 /*
1279 * Unlink from source.
1280 */
1281 rw_enter(&fromparent->tn_rwlock, RW_WRITER);
1282 rw_enter(&fromtp->tn_rwlock, RW_WRITER);
1283
1284 error = tdirdelete(fromparent, fromtp, onm, DR_RENAME, cred);
1285
1286 /*
1287 * The following handles the case where our source tmpnode was
1288 * removed before we got to it.
1289 *
1290 * XXX We should also cleanup properly in the case where tdirdelete
1291 * fails for some other reason. Currently this case shouldn't happen.
1292 * (see 1184991).
1293 */
1294 if (error == ENOENT)
1295 error = 0;
1296
1297 rw_exit(&fromtp->tn_rwlock);
1298 rw_exit(&fromparent->tn_rwlock);
1299
1300 if (error == 0) {
1301 vnevent_rename_src(TNTOV(fromtp), odvp, onm, ct);
1302 /*
1303 * vnevent_rename_dest is called in tdirenter().
1304 * Notify the target dir if not same as source dir.
1305 */
1306 if (ndvp != odvp)
1307 vnevent_rename_dest_dir(ndvp, ct);
1308 }
1309
1310 done:
1311 tmpnode_rele(fromtp);
1312 mutex_exit(&tm->tm_renamelck);
1313
1314 TRACE_5(TR_FAC_TMPFS, TR_TMPFS_RENAME,
1315 "tmpfs rename:ovp %p onm %s nvp %p nnm %s error %d", odvp, onm,
1316 ndvp, nnm, error);
1317 return (error);
1318 }
1319
1320 /* ARGSUSED5 */
1321 static int
1322 tmp_mkdir(
1323 struct vnode *dvp,
1324 char *nm,
1325 struct vattr *va,
1326 struct vnode **vpp,
1327 struct cred *cred,
1328 caller_context_t *ct,
1329 int flags,
1330 vsecattr_t *vsecp)
1331 {
1332 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1333 struct tmpnode *self = NULL;
1334 struct tmount *tm = (struct tmount *)VTOTM(dvp);
1335 int error;
1336
1337 /* no new dirs allowed in xattr dirs */
1338 if (parent->tn_flags & ISXATTR)
1339 return (EINVAL);
1340
1341 /*
1342 * Might be dangling directory. Catch it here,
1343 * because a ENOENT return from tdirlookup() is
1344 * an "o.k. return".
1345 */
1346 if (parent->tn_nlink == 0)
1347 return (ENOENT);
1348
1349 error = tdirlookup(parent, nm, &self, cred);
1350 if (error == 0) {
1351 ASSERT(self);
1352 tmpnode_rele(self);
1353 return (EEXIST);
1354 }
1355 if (error != ENOENT)
1356 return (error);
1357
1358 rw_enter(&parent->tn_rwlock, RW_WRITER);
1359 error = tdirenter(tm, parent, nm, DE_MKDIR, (struct tmpnode *)NULL,
1360 (struct tmpnode *)NULL, va, &self, cred, ct);
1361 if (error) {
1362 rw_exit(&parent->tn_rwlock);
1363 if (self)
1364 tmpnode_rele(self);
1365 return (error);
1366 }
1367 rw_exit(&parent->tn_rwlock);
1368 *vpp = TNTOV(self);
1369 return (0);
1370 }
1371
1372 /* ARGSUSED4 */
1373 static int
1374 tmp_rmdir(
1375 struct vnode *dvp,
1376 char *nm,
1377 struct vnode *cdir,
1378 struct cred *cred,
1379 caller_context_t *ct,
1380 int flags)
1381 {
1382 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1383 struct tmpnode *self = NULL;
1384 struct vnode *vp;
1385 int error = 0;
1386
1387 /*
1388 * Return error when removing . and ..
1389 */
1390 if (strcmp(nm, ".") == 0)
1391 return (EINVAL);
1392 if (strcmp(nm, "..") == 0)
1393 return (EEXIST); /* Should be ENOTEMPTY */
1394 error = tdirlookup(parent, nm, &self, cred);
1395 if (error)
1396 return (error);
1397
1398 rw_enter(&parent->tn_rwlock, RW_WRITER);
1399 rw_enter(&self->tn_rwlock, RW_WRITER);
1400
1401 vp = TNTOV(self);
1402 if (vp == dvp || vp == cdir) {
1403 error = EINVAL;
1404 goto done1;
1405 }
1406 if (self->tn_type != VDIR) {
1407 error = ENOTDIR;
1408 goto done1;
1409 }
1410
1411 mutex_enter(&self->tn_tlock);
1412 if (self->tn_nlink > 2) {
1413 mutex_exit(&self->tn_tlock);
1414 error = EEXIST;
1415 goto done1;
1416 }
1417 mutex_exit(&self->tn_tlock);
1418
1419 if (vn_vfswlock(vp)) {
1420 error = EBUSY;
1421 goto done1;
1422 }
1423 if (vn_mountedvfs(vp) != NULL) {
1424 error = EBUSY;
1425 goto done;
1426 }
1427
1428 /*
1429 * Check for an empty directory
1430 * i.e. only includes entries for "." and ".."
1431 */
1432 if (self->tn_dirents > 2) {
1433 error = EEXIST; /* SIGH should be ENOTEMPTY */
1434 /*
1435 * Update atime because checking tn_dirents is logically
1436 * equivalent to reading the directory
1437 */
1438 gethrestime(&self->tn_atime);
1439 goto done;
1440 }
1441
1442 error = tdirdelete(parent, self, nm, DR_RMDIR, cred);
1443 done:
1444 vn_vfsunlock(vp);
1445 done1:
1446 rw_exit(&self->tn_rwlock);
1447 rw_exit(&parent->tn_rwlock);
1448 vnevent_rmdir(TNTOV(self), dvp, nm, ct);
1449 tmpnode_rele(self);
1450
1451 return (error);
1452 }
1453
1454 /* ARGSUSED2 */
1455 static int
1456 tmp_readdir(
1457 struct vnode *vp,
1458 struct uio *uiop,
1459 struct cred *cred,
1460 int *eofp,
1461 caller_context_t *ct,
1462 int flags)
1463 {
1464 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1465 struct tdirent *tdp;
1466 int error = 0;
1467 size_t namelen;
1468 struct dirent64 *dp;
1469 ulong_t offset;
1470 ulong_t total_bytes_wanted;
1471 long outcount = 0;
1472 long bufsize;
1473 int reclen;
1474 caddr_t outbuf;
1475
1476 if (uiop->uio_loffset >= MAXOFF_T) {
1477 if (eofp)
1478 *eofp = 1;
1479 return (0);
1480 }
1481 /*
1482 * assuming system call has already called tmp_rwlock
1483 */
1484 ASSERT(RW_READ_HELD(&tp->tn_rwlock));
1485
1486 if (uiop->uio_iovcnt != 1)
1487 return (EINVAL);
1488
1489 if (vp->v_type != VDIR)
1490 return (ENOTDIR);
1491
1492 /*
1493 * There's a window here where someone could have removed
1494 * all the entries in the directory after we put a hold on the
1495 * vnode but before we grabbed the rwlock. Just return.
1496 */
1497 if (tp->tn_dir == NULL) {
1498 if (tp->tn_nlink) {
1499 panic("empty directory 0x%p", (void *)tp);
1500 /*NOTREACHED*/
1501 }
1502 return (0);
1503 }
1504
1505 /*
1506 * Get space for multiple directory entries
1507 */
1508 total_bytes_wanted = uiop->uio_iov->iov_len;
1509 bufsize = total_bytes_wanted + sizeof (struct dirent64);
1510 outbuf = kmem_alloc(bufsize, KM_SLEEP);
1511
1512 dp = (struct dirent64 *)outbuf;
1513
1514
1515 offset = 0;
1516 tdp = tp->tn_dir;
1517 while (tdp) {
1518 namelen = strlen(tdp->td_name); /* no +1 needed */
1519 offset = tdp->td_offset;
1520 if (offset >= uiop->uio_offset) {
1521 reclen = (int)DIRENT64_RECLEN(namelen);
1522 if (outcount + reclen > total_bytes_wanted) {
1523 if (!outcount)
1524 /*
1525 * Buffer too small for any entries.
1526 */
1527 error = EINVAL;
1528 break;
1529 }
1530 ASSERT(tdp->td_tmpnode != NULL);
1531
1532 /* use strncpy(9f) to zero out uninitialized bytes */
1533
1534 (void) strncpy(dp->d_name, tdp->td_name,
1535 DIRENT64_NAMELEN(reclen));
1536 dp->d_reclen = (ushort_t)reclen;
1537 dp->d_ino = (ino64_t)tdp->td_tmpnode->tn_nodeid;
1538 dp->d_off = (offset_t)tdp->td_offset + 1;
1539 dp = (struct dirent64 *)
1540 ((uintptr_t)dp + dp->d_reclen);
1541 outcount += reclen;
1542 ASSERT(outcount <= bufsize);
1543 }
1544 tdp = tdp->td_next;
1545 }
1546
1547 if (!error)
1548 error = uiomove(outbuf, outcount, UIO_READ, uiop);
1549
1550 if (!error) {
1551 /* If we reached the end of the list our offset */
1552 /* should now be just past the end. */
1553 if (!tdp) {
1554 offset += 1;
1555 if (eofp)
1556 *eofp = 1;
1557 } else if (eofp)
1558 *eofp = 0;
1559 uiop->uio_offset = offset;
1560 }
1561 gethrestime(&tp->tn_atime);
1562 kmem_free(outbuf, bufsize);
1563 return (error);
1564 }
1565
1566 /* ARGSUSED5 */
1567 static int
1568 tmp_symlink(
1569 struct vnode *dvp,
1570 char *lnm,
1571 struct vattr *tva,
1572 char *tnm,
1573 struct cred *cred,
1574 caller_context_t *ct,
1575 int flags)
1576 {
1577 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1578 struct tmpnode *self = (struct tmpnode *)NULL;
1579 struct tmount *tm = (struct tmount *)VTOTM(dvp);
1580 char *cp = NULL;
1581 int error;
1582 size_t len;
1583
1584 /* no symlinks allowed to files in xattr dirs */
1585 if (parent->tn_flags & ISXATTR)
1586 return (EINVAL);
1587
1588 error = tdirlookup(parent, lnm, &self, cred);
1589 if (error == 0) {
1590 /*
1591 * The entry already exists
1592 */
1593 tmpnode_rele(self);
1594 return (EEXIST); /* was 0 */
1595 }
1596
1597 if (error != ENOENT) {
1598 if (self != NULL)
1599 tmpnode_rele(self);
1600 return (error);
1601 }
1602
1603 rw_enter(&parent->tn_rwlock, RW_WRITER);
1604 error = tdirenter(tm, parent, lnm, DE_CREATE, (struct tmpnode *)NULL,
1605 (struct tmpnode *)NULL, tva, &self, cred, ct);
1606 rw_exit(&parent->tn_rwlock);
1607
1608 if (error) {
1609 if (self)
1610 tmpnode_rele(self);
1611 return (error);
1612 }
1613 len = strlen(tnm) + 1;
1614 cp = tmp_memalloc(len, 0);
1615 if (cp == NULL) {
1616 tmpnode_rele(self);
1617 return (ENOSPC);
1618 }
1619 (void) strcpy(cp, tnm);
1620
1621 self->tn_symlink = cp;
1622 self->tn_size = len - 1;
1623 tmpnode_rele(self);
1624 return (error);
1625 }
1626
1627 /* ARGSUSED2 */
1628 static int
1629 tmp_readlink(
1630 struct vnode *vp,
1631 struct uio *uiop,
1632 struct cred *cred,
1633 caller_context_t *ct)
1634 {
1635 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1636 int error = 0;
1637
1638 if (vp->v_type != VLNK)
1639 return (EINVAL);
1640
1641 rw_enter(&tp->tn_rwlock, RW_READER);
1642 rw_enter(&tp->tn_contents, RW_READER);
1643 error = uiomove(tp->tn_symlink, tp->tn_size, UIO_READ, uiop);
1644 gethrestime(&tp->tn_atime);
1645 rw_exit(&tp->tn_contents);
1646 rw_exit(&tp->tn_rwlock);
1647 return (error);
1648 }
1649
1650 /* ARGSUSED */
1651 static int
1652 tmp_fsync(
1653 struct vnode *vp,
1654 int syncflag,
1655 struct cred *cred,
1656 caller_context_t *ct)
1657 {
1658 return (0);
1659 }
1660
1661 /* ARGSUSED */
1662 static void
1663 tmp_inactive(struct vnode *vp, struct cred *cred, caller_context_t *ct)
1664 {
1665 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1666 struct tmount *tm = (struct tmount *)VFSTOTM(vp->v_vfsp);
1667
1668 rw_enter(&tp->tn_rwlock, RW_WRITER);
1669 top:
1670 mutex_enter(&tp->tn_tlock);
1671 mutex_enter(&vp->v_lock);
1672 ASSERT(vp->v_count >= 1);
1673
1674 /*
1675 * If we don't have the last hold or the link count is non-zero,
1676 * there's little to do -- just drop our hold.
1677 */
1678 if (vp->v_count > 1 || tp->tn_nlink != 0) {
1679 vp->v_count--;
1680 mutex_exit(&vp->v_lock);
1681 mutex_exit(&tp->tn_tlock);
1682 rw_exit(&tp->tn_rwlock);
1683 return;
1684 }
1685
1686 /*
1687 * We have the last hold *and* the link count is zero, so this
1688 * tmpnode is dead from the filesystem's viewpoint. However,
1689 * if the tmpnode has any pages associated with it (i.e. if it's
1690 * a normal file with non-zero size), the tmpnode can still be
1691 * discovered by pageout or fsflush via the page vnode pointers.
1692 * In this case we must drop all our locks, truncate the tmpnode,
1693 * and try the whole dance again.
1694 */
1695 if (tp->tn_size != 0) {
1696 if (tp->tn_type == VREG) {
1697 mutex_exit(&vp->v_lock);
1698 mutex_exit(&tp->tn_tlock);
1699 rw_enter(&tp->tn_contents, RW_WRITER);
1700 (void) tmpnode_trunc(tm, tp, 0);
1701 rw_exit(&tp->tn_contents);
1702 ASSERT(tp->tn_size == 0);
1703 ASSERT(tp->tn_nblocks == 0);
1704 goto top;
1705 }
1706 if (tp->tn_type == VLNK)
1707 tmp_memfree(tp->tn_symlink, tp->tn_size + 1);
1708 }
1709
1710 /*
1711 * Remove normal file/dir's xattr dir and xattrs.
1712 */
1713 if (tp->tn_xattrdp) {
1714 struct tmpnode *xtp = tp->tn_xattrdp;
1715
1716 ASSERT(xtp->tn_flags & ISXATTR);
1717 tmpnode_hold(xtp);
1718 rw_enter(&xtp->tn_rwlock, RW_WRITER);
1719 tdirtrunc(xtp);
1720 DECR_COUNT(&xtp->tn_nlink, &xtp->tn_tlock);
1721 tp->tn_xattrdp = NULL;
1722 rw_exit(&xtp->tn_rwlock);
1723 tmpnode_rele(xtp);
1724 }
1725
1726 mutex_exit(&vp->v_lock);
1727 mutex_exit(&tp->tn_tlock);
1728 /* Here's our chance to send invalid event while we're between locks */
1729 vn_invalid(TNTOV(tp));
1730 mutex_enter(&tm->tm_contents);
1731 if (tp->tn_forw == NULL)
1732 tm->tm_rootnode->tn_back = tp->tn_back;
1733 else
1734 tp->tn_forw->tn_back = tp->tn_back;
1735 tp->tn_back->tn_forw = tp->tn_forw;
1736 mutex_exit(&tm->tm_contents);
1737 rw_exit(&tp->tn_rwlock);
1738 rw_destroy(&tp->tn_rwlock);
1739 mutex_destroy(&tp->tn_tlock);
1740 vn_free(TNTOV(tp));
1741 tmp_memfree(tp, sizeof (struct tmpnode));
1742 }
1743
1744 /* ARGSUSED2 */
1745 static int
1746 tmp_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
1747 {
1748 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1749 struct tfid *tfid;
1750
1751 if (fidp->fid_len < (sizeof (struct tfid) - sizeof (ushort_t))) {
1752 fidp->fid_len = sizeof (struct tfid) - sizeof (ushort_t);
1753 return (ENOSPC);
1754 }
1755
1756 tfid = (struct tfid *)fidp;
1757 bzero(tfid, sizeof (struct tfid));
1758 tfid->tfid_len = (int)sizeof (struct tfid) - sizeof (ushort_t);
1759
1760 tfid->tfid_ino = tp->tn_nodeid;
1761 tfid->tfid_gen = tp->tn_gen;
1762
1763 return (0);
1764 }
1765
1766
1767 /*
1768 * Return all the pages from [off..off+len] in given file
1769 */
1770 /* ARGSUSED */
1771 static int
1772 tmp_getpage(
1773 struct vnode *vp,
1774 offset_t off,
1775 size_t len,
1776 uint_t *protp,
1777 page_t *pl[],
1778 size_t plsz,
1779 struct seg *seg,
1780 caddr_t addr,
1781 enum seg_rw rw,
1782 struct cred *cr,
1783 caller_context_t *ct)
1784 {
1785 int err = 0;
1786 struct tmpnode *tp = VTOTN(vp);
1787 anoff_t toff = (anoff_t)off;
1788 size_t tlen = len;
1789 u_offset_t tmpoff;
1790 timestruc_t now;
1791
1792 rw_enter(&tp->tn_contents, RW_READER);
1793
1794 if (off + len > tp->tn_size + PAGEOFFSET) {
1795 err = EFAULT;
1796 goto out;
1797 }
1798 /*
1799 * Look for holes (no anon slot) in faulting range. If there are
1800 * holes we have to switch to a write lock and fill them in. Swap
1801 * space for holes was already reserved when the file was grown.
1802 */
1803 tmpoff = toff;
1804 if (non_anon(tp->tn_anon, btop(off), &tmpoff, &tlen)) {
1805 if (!rw_tryupgrade(&tp->tn_contents)) {
1806 rw_exit(&tp->tn_contents);
1807 rw_enter(&tp->tn_contents, RW_WRITER);
1808 /* Size may have changed when lock was dropped */
1809 if (off + len > tp->tn_size + PAGEOFFSET) {
1810 err = EFAULT;
1811 goto out;
1812 }
1813 }
1814 for (toff = (anoff_t)off; toff < (anoff_t)off + len;
1815 toff += PAGESIZE) {
1816 if (anon_get_ptr(tp->tn_anon, btop(toff)) == NULL) {
1817 /* XXX - may allocate mem w. write lock held */
1818 (void) anon_set_ptr(tp->tn_anon, btop(toff),
1819 anon_alloc(vp, toff), ANON_SLEEP);
1820 tp->tn_nblocks++;
1821 }
1822 }
1823 rw_downgrade(&tp->tn_contents);
1824 }
1825
1826
1827 err = pvn_getpages(tmp_getapage, vp, (u_offset_t)off, len, protp,
1828 pl, plsz, seg, addr, rw, cr);
1829
1830 gethrestime(&now);
1831 tp->tn_atime = now;
1832 if (rw == S_WRITE)
1833 tp->tn_mtime = now;
1834
1835 out:
1836 rw_exit(&tp->tn_contents);
1837 return (err);
1838 }
1839
1840 /*
1841 * Called from pvn_getpages to get a particular page.
1842 */
1843 /*ARGSUSED*/
1844 static int
1845 tmp_getapage(
1846 struct vnode *vp,
1847 u_offset_t off,
1848 size_t len,
1849 uint_t *protp,
1850 page_t *pl[],
1851 size_t plsz,
1852 struct seg *seg,
1853 caddr_t addr,
1854 enum seg_rw rw,
1855 struct cred *cr)
1856 {
1857 struct page *pp;
1858 int flags;
1859 int err = 0;
1860 struct vnode *pvp;
1861 u_offset_t poff;
1862
1863 if (protp != NULL)
1864 *protp = PROT_ALL;
1865 again:
1866 if (pp = page_lookup(vp, off, rw == S_CREATE ? SE_EXCL : SE_SHARED)) {
1867 if (pl) {
1868 pl[0] = pp;
1869 pl[1] = NULL;
1870 } else {
1871 page_unlock(pp);
1872 }
1873 } else {
1874 pp = page_create_va(vp, off, PAGESIZE,
1875 PG_WAIT | PG_EXCL, seg, addr);
1876 /*
1877 * Someone raced in and created the page after we did the
1878 * lookup but before we did the create, so go back and
1879 * try to look it up again.
1880 */
1881 if (pp == NULL)
1882 goto again;
1883 /*
1884 * Fill page from backing store, if any. If none, then
1885 * either this is a newly filled hole or page must have
1886 * been unmodified and freed so just zero it out.
1887 */
1888 err = swap_getphysname(vp, off, &pvp, &poff);
1889 if (err) {
1890 panic("tmp_getapage: no anon slot vp %p "
1891 "off %llx pp %p\n", (void *)vp, off, (void *)pp);
1892 }
1893 if (pvp) {
1894 flags = (pl == NULL ? B_ASYNC|B_READ : B_READ);
1895 err = VOP_PAGEIO(pvp, pp, (u_offset_t)poff, PAGESIZE,
1896 flags, cr, NULL);
1897 if (flags & B_ASYNC)
1898 pp = NULL;
1899 } else if (rw != S_CREATE) {
1900 pagezero(pp, 0, PAGESIZE);
1901 }
1902 if (err && pp)
1903 pvn_read_done(pp, B_ERROR);
1904 if (err == 0) {
1905 if (pl)
1906 pvn_plist_init(pp, pl, plsz, off, PAGESIZE, rw);
1907 else
1908 pvn_io_done(pp);
1909 }
1910 }
1911 return (err);
1912 }
1913
1914
1915 /*
1916 * Flags are composed of {B_INVAL, B_DIRTY B_FREE, B_DONTNEED}.
1917 * If len == 0, do from off to EOF.
1918 */
1919 static int tmp_nopage = 0; /* Don't do tmp_putpage's if set */
1920
1921 /* ARGSUSED */
1922 int
1923 tmp_putpage(
1924 register struct vnode *vp,
1925 offset_t off,
1926 size_t len,
1927 int flags,
1928 struct cred *cr,
1929 caller_context_t *ct)
1930 {
1931 register page_t *pp;
1932 u_offset_t io_off;
1933 size_t io_len = 0;
1934 int err = 0;
1935 struct tmpnode *tp = VTOTN(vp);
1936 int dolock;
1937
1938 if (tmp_nopage)
1939 return (0);
1940
1941 ASSERT(vp->v_count != 0);
1942
1943 if (vp->v_flag & VNOMAP)
1944 return (ENOSYS);
1945
1946 /*
1947 * This being tmpfs, we don't ever do i/o unless we really
1948 * have to (when we're low on memory and pageout calls us
1949 * with B_ASYNC | B_FREE or the user explicitly asks for it with
1950 * B_DONTNEED).
1951 * XXX to approximately track the mod time like ufs we should
1952 * update the times here. The problem is, once someone does a
1953 * store we never clear the mod bit and do i/o, thus fsflush
1954 * will keep calling us every 30 seconds to do the i/o and we'll
1955 * continually update the mod time. At least we update the mod
1956 * time on the first store because this results in a call to getpage.
1957 */
1958 if (flags != (B_ASYNC | B_FREE) && (flags & B_INVAL) == 0 &&
1959 (flags & B_DONTNEED) == 0)
1960 return (0);
1961 /*
1962 * If this thread owns the lock, i.e., this thread grabbed it
1963 * as writer somewhere above, then we don't need to grab the
1964 * lock as reader in this routine.
1965 */
1966 dolock = (rw_owner(&tp->tn_contents) != curthread);
1967
1968 /*
1969 * If this is pageout don't block on the lock as you could deadlock
1970 * when freemem == 0 (another thread has the read lock and is blocked
1971 * creating a page, and a third thread is waiting to get the writers
1972 * lock - waiting writers priority blocks us from getting the read
1973 * lock). Of course, if the only freeable pages are on this tmpnode
1974 * we're hosed anyways. A better solution might be a new lock type.
1975 * Note: ufs has the same problem.
1976 */
1977 if (curproc == proc_pageout) {
1978 if (!rw_tryenter(&tp->tn_contents, RW_READER))
1979 return (ENOMEM);
1980 } else if (dolock)
1981 rw_enter(&tp->tn_contents, RW_READER);
1982
1983 if (!vn_has_cached_data(vp))
1984 goto out;
1985
1986 if (len == 0) {
1987 if (curproc == proc_pageout) {
1988 panic("tmp: pageout can't block");
1989 /*NOTREACHED*/
1990 }
1991
1992 /* Search the entire vp list for pages >= off. */
1993 err = pvn_vplist_dirty(vp, (u_offset_t)off, tmp_putapage,
1994 flags, cr);
1995 } else {
1996 u_offset_t eoff;
1997
1998 /*
1999 * Loop over all offsets in the range [off...off + len]
2000 * looking for pages to deal with.
2001 */
2002 eoff = MIN(off + len, tp->tn_size);
2003 for (io_off = off; io_off < eoff; io_off += io_len) {
2004 /*
2005 * If we are not invalidating, synchronously
2006 * freeing or writing pages use the routine
2007 * page_lookup_nowait() to prevent reclaiming
2008 * them from the free list.
2009 */
2010 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2011 pp = page_lookup(vp, io_off,
2012 (flags & (B_INVAL | B_FREE)) ?
2013 SE_EXCL : SE_SHARED);
2014 } else {
2015 pp = page_lookup_nowait(vp, io_off,
2016 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2017 }
2018
2019 if (pp == NULL || pvn_getdirty(pp, flags) == 0)
2020 io_len = PAGESIZE;
2021 else {
2022 err = tmp_putapage(vp, pp, &io_off, &io_len,
2023 flags, cr);
2024 if (err != 0)
2025 break;
2026 }
2027 }
2028 }
2029 /* If invalidating, verify all pages on vnode list are gone. */
2030 if (err == 0 && off == 0 && len == 0 &&
2031 (flags & B_INVAL) && vn_has_cached_data(vp)) {
2032 panic("tmp_putpage: B_INVAL, pages not gone");
2033 /*NOTREACHED*/
2034 }
2035 out:
2036 if ((curproc == proc_pageout) || dolock)
2037 rw_exit(&tp->tn_contents);
2038 /*
2039 * Only reason putapage is going to give us SE_NOSWAP as error
2040 * is when we ask a page to be written to physical backing store
2041 * and there is none. Ignore this because we might be dealing
2042 * with a swap page which does not have any backing store
2043 * on disk. In any other case we won't get this error over here.
2044 */
2045 if (err == SE_NOSWAP)
2046 err = 0;
2047 return (err);
2048 }
2049
2050 long tmp_putpagecnt, tmp_pagespushed;
2051
2052 /*
2053 * Write out a single page.
2054 * For tmpfs this means choose a physical swap slot and write the page
2055 * out using VOP_PAGEIO. For performance, we attempt to kluster; i.e.,
2056 * we try to find a bunch of other dirty pages adjacent in the file
2057 * and a bunch of contiguous swap slots, and then write all the pages
2058 * out in a single i/o.
2059 */
2060 /*ARGSUSED*/
2061 static int
2062 tmp_putapage(
2063 struct vnode *vp,
2064 page_t *pp,
2065 u_offset_t *offp,
2066 size_t *lenp,
2067 int flags,
2068 struct cred *cr)
2069 {
2070 int err;
2071 ulong_t klstart, kllen;
2072 page_t *pplist, *npplist;
2073 extern int klustsize;
2074 long tmp_klustsize;
2075 struct tmpnode *tp;
2076 size_t pp_off, pp_len;
2077 u_offset_t io_off;
2078 size_t io_len;
2079 struct vnode *pvp;
2080 u_offset_t pstart;
2081 u_offset_t offset;
2082 u_offset_t tmpoff;
2083
2084 ASSERT(PAGE_LOCKED(pp));
2085
2086 /* Kluster in tmp_klustsize chunks */
2087 tp = VTOTN(vp);
2088 tmp_klustsize = klustsize;
2089 offset = pp->p_offset;
2090 klstart = (offset / tmp_klustsize) * tmp_klustsize;
2091 kllen = MIN(tmp_klustsize, tp->tn_size - klstart);
2092
2093 /* Get a kluster of pages */
2094 pplist =
2095 pvn_write_kluster(vp, pp, &tmpoff, &pp_len, klstart, kllen, flags);
2096
2097 pp_off = (size_t)tmpoff;
2098
2099 /*
2100 * Get a cluster of physical offsets for the pages; the amount we
2101 * get may be some subrange of what we ask for (io_off, io_len).
2102 */
2103 io_off = pp_off;
2104 io_len = pp_len;
2105 err = swap_newphysname(vp, offset, &io_off, &io_len, &pvp, &pstart);
2106 ASSERT(err != SE_NOANON); /* anon slot must have been filled */
2107 if (err) {
2108 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2109 /*
2110 * If this routine is called as a result of segvn_sync
2111 * operation and we have no physical swap then we can get an
2112 * error here. In such case we would return SE_NOSWAP as error.
2113 * At this point, we expect only SE_NOSWAP.
2114 */
2115 ASSERT(err == SE_NOSWAP);
2116 if (flags & B_INVAL)
2117 err = ENOMEM;
2118 goto out;
2119 }
2120 ASSERT(pp_off <= io_off && io_off + io_len <= pp_off + pp_len);
2121 ASSERT(io_off <= offset && offset < io_off + io_len);
2122
2123 /* Toss pages at front/rear that we couldn't get physical backing for */
2124 if (io_off != pp_off) {
2125 npplist = NULL;
2126 page_list_break(&pplist, &npplist, btop(io_off - pp_off));
2127 ASSERT(pplist->p_offset == pp_off);
2128 ASSERT(pplist->p_prev->p_offset == io_off - PAGESIZE);
2129 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2130 pplist = npplist;
2131 }
2132 if (io_off + io_len < pp_off + pp_len) {
2133 npplist = NULL;
2134 page_list_break(&pplist, &npplist, btop(io_len));
2135 ASSERT(npplist->p_offset == io_off + io_len);
2136 ASSERT(npplist->p_prev->p_offset == pp_off + pp_len - PAGESIZE);
2137 pvn_write_done(npplist, B_ERROR | B_WRITE | flags);
2138 }
2139
2140 ASSERT(pplist->p_offset == io_off);
2141 ASSERT(pplist->p_prev->p_offset == io_off + io_len - PAGESIZE);
2142 ASSERT(btopr(io_len) <= btopr(kllen));
2143
2144 /* Do i/o on the remaining kluster */
2145 err = VOP_PAGEIO(pvp, pplist, (u_offset_t)pstart, io_len,
2146 B_WRITE | flags, cr, NULL);
2147
2148 if ((flags & B_ASYNC) == 0) {
2149 pvn_write_done(pplist, ((err) ? B_ERROR : 0) | B_WRITE | flags);
2150 }
2151 out:
2152 if (!err) {
2153 if (offp)
2154 *offp = io_off;
2155 if (lenp)
2156 *lenp = io_len;
2157 tmp_putpagecnt++;
2158 tmp_pagespushed += btop(io_len);
2159 }
2160 if (err && err != ENOMEM && err != SE_NOSWAP)
2161 cmn_err(CE_WARN, "tmp_putapage: err %d\n", err);
2162 return (err);
2163 }
2164
2165 /* ARGSUSED */
2166 static int
2167 tmp_map(
2168 struct vnode *vp,
2169 offset_t off,
2170 struct as *as,
2171 caddr_t *addrp,
2172 size_t len,
2173 uchar_t prot,
2174 uchar_t maxprot,
2175 uint_t flags,
2176 struct cred *cred,
2177 caller_context_t *ct)
2178 {
2179 struct segvn_crargs vn_a;
2180 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
2181 int error;
2182
2183 #ifdef _ILP32
2184 if (len > MAXOFF_T)
2185 return (ENOMEM);
2186 #endif
2187
2188 if (vp->v_flag & VNOMAP)
2189 return (ENOSYS);
2190
2191 if (off < 0 || (offset_t)(off + len) < 0 ||
2192 off > MAXOFF_T || (off + len) > MAXOFF_T)
2193 return (ENXIO);
2194
2195 if (vp->v_type != VREG)
2196 return (ENODEV);
2197
2198 /*
2199 * Don't allow mapping to locked file
2200 */
2201 if (vn_has_mandatory_locks(vp, tp->tn_mode)) {
2202 return (EAGAIN);
2203 }
2204
2205 as_rangelock(as);
2206 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
2207 if (error != 0) {
2208 as_rangeunlock(as);
2209 return (error);
2210 }
2211
2212 vn_a.vp = vp;
2213 vn_a.offset = (u_offset_t)off;
2214 vn_a.type = flags & MAP_TYPE;
2215 vn_a.prot = prot;
2216 vn_a.maxprot = maxprot;
2217 vn_a.flags = flags & ~MAP_TYPE;
2218 vn_a.cred = cred;
2219 vn_a.amp = NULL;
2220 vn_a.szc = 0;
2221 vn_a.lgrp_mem_policy_flags = 0;
2222
2223 error = as_map(as, *addrp, len, segvn_create, &vn_a);
2224 as_rangeunlock(as);
2225 return (error);
2226 }
2227
2228 /*
2229 * tmp_addmap and tmp_delmap can't be called since the vp
2230 * maintained in the segvn mapping is NULL.
2231 */
2232 /* ARGSUSED */
2233 static int
2234 tmp_addmap(
2235 struct vnode *vp,
2236 offset_t off,
2237 struct as *as,
2238 caddr_t addr,
2239 size_t len,
2240 uchar_t prot,
2241 uchar_t maxprot,
2242 uint_t flags,
2243 struct cred *cred,
2244 caller_context_t *ct)
2245 {
2246 return (0);
2247 }
2248
2249 /* ARGSUSED */
2250 static int
2251 tmp_delmap(
2252 struct vnode *vp,
2253 offset_t off,
2254 struct as *as,
2255 caddr_t addr,
2256 size_t len,
2257 uint_t prot,
2258 uint_t maxprot,
2259 uint_t flags,
2260 struct cred *cred,
2261 caller_context_t *ct)
2262 {
2263 return (0);
2264 }
2265
2266 static int
2267 tmp_freesp(struct vnode *vp, struct flock64 *lp, int flag)
2268 {
2269 register int i;
2270 register struct tmpnode *tp = VTOTN(vp);
2271 int error;
2272
2273 ASSERT(vp->v_type == VREG);
2274 ASSERT(lp->l_start >= 0);
2275
2276 if (lp->l_len != 0)
2277 return (EINVAL);
2278
2279 rw_enter(&tp->tn_rwlock, RW_WRITER);
2280 if (tp->tn_size == lp->l_start) {
2281 rw_exit(&tp->tn_rwlock);
2282 return (0);
2283 }
2284
2285 /*
2286 * Check for any mandatory locks on the range
2287 */
2288 if (MANDLOCK(vp, tp->tn_mode)) {
2289 long save_start;
2290
2291 save_start = lp->l_start;
2292
2293 if (tp->tn_size < lp->l_start) {
2294 /*
2295 * "Truncate up" case: need to make sure there
2296 * is no lock beyond current end-of-file. To
2297 * do so, we need to set l_start to the size
2298 * of the file temporarily.
2299 */
2300 lp->l_start = tp->tn_size;
2301 }
2302 lp->l_type = F_WRLCK;
2303 lp->l_sysid = 0;
2304 lp->l_pid = ttoproc(curthread)->p_pid;
2305 i = (flag & (FNDELAY|FNONBLOCK)) ? 0 : SLPFLCK;
2306 if ((i = reclock(vp, lp, i, 0, lp->l_start, NULL)) != 0 ||
2307 lp->l_type != F_UNLCK) {
2308 rw_exit(&tp->tn_rwlock);
2309 return (i ? i : EAGAIN);
2310 }
2311
2312 lp->l_start = save_start;
2313 }
2314 VFSTOTM(vp->v_vfsp);
2315
2316 rw_enter(&tp->tn_contents, RW_WRITER);
2317 error = tmpnode_trunc((struct tmount *)VFSTOTM(vp->v_vfsp),
2318 tp, (ulong_t)lp->l_start);
2319 rw_exit(&tp->tn_contents);
2320 rw_exit(&tp->tn_rwlock);
2321 return (error);
2322 }
2323
2324 /* ARGSUSED */
2325 static int
2326 tmp_space(
2327 struct vnode *vp,
2328 int cmd,
2329 struct flock64 *bfp,
2330 int flag,
2331 offset_t offset,
2332 cred_t *cred,
2333 caller_context_t *ct)
2334 {
2335 int error;
2336
2337 if (cmd != F_FREESP)
2338 return (EINVAL);
2339 if ((error = convoff(vp, bfp, 0, (offset_t)offset)) == 0) {
2340 if ((bfp->l_start > MAXOFF_T) || (bfp->l_len > MAXOFF_T))
2341 return (EFBIG);
2342 error = tmp_freesp(vp, bfp, flag);
2343
2344 if (error == 0 && bfp->l_start == 0)
2345 vnevent_truncate(vp, ct);
2346 }
2347 return (error);
2348 }
2349
2350 /* ARGSUSED */
2351 static int
2352 tmp_seek(
2353 struct vnode *vp,
2354 offset_t ooff,
2355 offset_t *noffp,
2356 caller_context_t *ct)
2357 {
2358 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
2359 }
2360
2361 /* ARGSUSED2 */
2362 static int
2363 tmp_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2364 {
2365 struct tmpnode *tp = VTOTN(vp);
2366
2367 if (write_lock) {
2368 rw_enter(&tp->tn_rwlock, RW_WRITER);
2369 } else {
2370 rw_enter(&tp->tn_rwlock, RW_READER);
2371 }
2372 return (write_lock);
2373 }
2374
2375 /* ARGSUSED1 */
2376 static void
2377 tmp_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2378 {
2379 struct tmpnode *tp = VTOTN(vp);
2380
2381 rw_exit(&tp->tn_rwlock);
2382 }
2383
2384 static int
2385 tmp_pathconf(
2386 struct vnode *vp,
2387 int cmd,
2388 ulong_t *valp,
2389 cred_t *cr,
2390 caller_context_t *ct)
2391 {
2392 struct tmpnode *tp = NULL;
2393 int error;
2394
2395 switch (cmd) {
2396 case _PC_XATTR_EXISTS:
2397 if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
2398 *valp = 0; /* assume no attributes */
2399 error = 0; /* okay to ask */
2400 tp = VTOTN(vp);
2401 rw_enter(&tp->tn_rwlock, RW_READER);
2402 if (tp->tn_xattrdp) {
2403 rw_enter(&tp->tn_xattrdp->tn_rwlock, RW_READER);
2404 /* do not count "." and ".." */
2405 if (tp->tn_xattrdp->tn_dirents > 2)
2406 *valp = 1;
2407 rw_exit(&tp->tn_xattrdp->tn_rwlock);
2408 }
2409 rw_exit(&tp->tn_rwlock);
2410 } else {
2411 error = EINVAL;
2412 }
2413 break;
2414 case _PC_SATTR_ENABLED:
2415 case _PC_SATTR_EXISTS:
2416 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2417 (vp->v_type == VREG || vp->v_type == VDIR);
2418 error = 0;
2419 break;
2420 case _PC_TIMESTAMP_RESOLUTION:
2421 /* nanosecond timestamp resolution */
2422 *valp = 1L;
2423 error = 0;
2424 break;
2425 default:
2426 error = fs_pathconf(vp, cmd, valp, cr, ct);
2427 }
2428 return (error);
2429 }
2430
2431
2432 struct vnodeops *tmp_vnodeops;
2433
2434 const fs_operation_def_t tmp_vnodeops_template[] = {
2435 VOPNAME_OPEN, { .vop_open = tmp_open },
2436 VOPNAME_CLOSE, { .vop_close = tmp_close },
2437 VOPNAME_READ, { .vop_read = tmp_read },
2438 VOPNAME_WRITE, { .vop_write = tmp_write },
2439 VOPNAME_IOCTL, { .vop_ioctl = tmp_ioctl },
2440 VOPNAME_GETATTR, { .vop_getattr = tmp_getattr },
2441 VOPNAME_SETATTR, { .vop_setattr = tmp_setattr },
2442 VOPNAME_ACCESS, { .vop_access = tmp_access },
2443 VOPNAME_LOOKUP, { .vop_lookup = tmp_lookup },
2444 VOPNAME_CREATE, { .vop_create = tmp_create },
2445 VOPNAME_REMOVE, { .vop_remove = tmp_remove },
2446 VOPNAME_LINK, { .vop_link = tmp_link },
2447 VOPNAME_RENAME, { .vop_rename = tmp_rename },
2448 VOPNAME_MKDIR, { .vop_mkdir = tmp_mkdir },
2449 VOPNAME_RMDIR, { .vop_rmdir = tmp_rmdir },
2450 VOPNAME_READDIR, { .vop_readdir = tmp_readdir },
2451 VOPNAME_SYMLINK, { .vop_symlink = tmp_symlink },
2452 VOPNAME_READLINK, { .vop_readlink = tmp_readlink },
2453 VOPNAME_FSYNC, { .vop_fsync = tmp_fsync },
2454 VOPNAME_INACTIVE, { .vop_inactive = tmp_inactive },
2455 VOPNAME_FID, { .vop_fid = tmp_fid },
2456 VOPNAME_RWLOCK, { .vop_rwlock = tmp_rwlock },
2457 VOPNAME_RWUNLOCK, { .vop_rwunlock = tmp_rwunlock },
2458 VOPNAME_SEEK, { .vop_seek = tmp_seek },
2459 VOPNAME_SPACE, { .vop_space = tmp_space },
2460 VOPNAME_GETPAGE, { .vop_getpage = tmp_getpage },
2461 VOPNAME_PUTPAGE, { .vop_putpage = tmp_putpage },
2462 VOPNAME_MAP, { .vop_map = tmp_map },
2463 VOPNAME_ADDMAP, { .vop_addmap = tmp_addmap },
2464 VOPNAME_DELMAP, { .vop_delmap = tmp_delmap },
2465 VOPNAME_PATHCONF, { .vop_pathconf = tmp_pathconf },
2466 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
2467 NULL, NULL
2468 };