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 */
31
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/t_lock.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/user.h>
38 #include <sys/time.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/fcntl.h>
44 #include <sys/flock.h>
45 #include <sys/kmem.h>
46 #include <sys/uio.h>
47 #include <sys/errno.h>
48 #include <sys/stat.h>
49 #include <sys/cred.h>
50 #include <sys/dirent.h>
51 #include <sys/pathname.h>
52 #include <sys/vmsystm.h>
53 #include <sys/fs/tmp.h>
54 #include <sys/fs/tmpnode.h>
55 #include <sys/mman.h>
56 #include <vm/hat.h>
57 #include <vm/seg_vn.h>
58 #include <vm/seg_map.h>
59 #include <vm/seg.h>
60 #include <vm/anon.h>
61 #include <vm/as.h>
62 #include <vm/page.h>
63 #include <vm/pvn.h>
64 #include <sys/cmn_err.h>
65 #include <sys/debug.h>
66 #include <sys/swap.h>
67 #include <sys/buf.h>
68 #include <sys/vm.h>
69 #include <sys/vtrace.h>
70 #include <sys/policy.h>
71 #include <fs/fs_subr.h>
72
73 static int tmp_getapage(struct vnode *, u_offset_t, size_t, uint_t *,
74 page_t **, size_t, struct seg *, caddr_t, enum seg_rw, struct cred *);
75 static int tmp_putapage(struct vnode *, page_t *, u_offset_t *, size_t *,
76 int, struct cred *);
77
78 /* ARGSUSED1 */
79 static int
80 tmp_open(struct vnode **vpp, int flag, struct cred *cred, caller_context_t *ct)
81 {
82 /*
83 * swapon to a tmpfs file is not supported so access
84 * is denied on open if VISSWAP is set.
85 */
86 if ((*vpp)->v_flag & VISSWAP)
87 return (EINVAL);
88 return (0);
89 }
90
91 /* ARGSUSED1 */
92 static int
93 tmp_close(
94 struct vnode *vp,
95 int flag,
96 int count,
97 offset_t offset,
98 struct cred *cred,
99 caller_context_t *ct)
100 {
101 cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
102 cleanshares(vp, ttoproc(curthread)->p_pid);
103 return (0);
104 }
105
106 /*
107 * wrtmp does the real work of write requests for tmpfs.
108 */
109 static int
110 wrtmp(
111 struct tmount *tm,
112 struct tmpnode *tp,
113 struct uio *uio,
114 struct cred *cr,
115 struct caller_context *ct)
116 {
117 pgcnt_t pageoffset; /* offset in pages */
118 ulong_t segmap_offset; /* pagesize byte offset into segmap */
119 caddr_t base; /* base of segmap */
120 ssize_t bytes; /* bytes to uiomove */
121 pfn_t pagenumber; /* offset in pages into tmp file */
122 struct vnode *vp;
123 int error = 0;
124 int pagecreate; /* == 1 if we allocated a page */
125 int newpage;
126 rlim64_t limit = uio->uio_llimit;
127 long oresid = uio->uio_resid;
128 timestruc_t now;
129
130 long tn_size_changed = 0;
131 long old_tn_size;
132 long new_tn_size;
133
134 vp = TNTOV(tp);
135 ASSERT(vp->v_type == VREG);
136
137 TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START,
138 "tmp_wrtmp_start:vp %p", vp);
139
140 ASSERT(RW_WRITE_HELD(&tp->tn_contents));
141 ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
142
143 if (MANDLOCK(vp, tp->tn_mode)) {
144 rw_exit(&tp->tn_contents);
145 /*
146 * tmp_getattr ends up being called by chklock
147 */
148 error = chklock(vp, FWRITE, uio->uio_loffset, uio->uio_resid,
149 uio->uio_fmode, ct);
150 rw_enter(&tp->tn_contents, RW_WRITER);
151 if (error != 0) {
152 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
153 "tmp_wrtmp_end:vp %p error %d", vp, error);
154 return (error);
155 }
156 }
157
158 if (uio->uio_loffset < 0)
159 return (EINVAL);
160
161 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
162 limit = MAXOFFSET_T;
163
164 if (uio->uio_loffset >= limit) {
165 proc_t *p = ttoproc(curthread);
166
167 mutex_enter(&p->p_lock);
168 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
169 p, RCA_UNSAFE_SIGINFO);
170 mutex_exit(&p->p_lock);
171 return (EFBIG);
172 }
173
174 if (uio->uio_loffset >= MAXOFF_T) {
175 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
176 "tmp_wrtmp_end:vp %p error %d", vp, EINVAL);
177 return (EFBIG);
178 }
179
180 if (uio->uio_resid == 0) {
181 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
182 "tmp_wrtmp_end:vp %p error %d", vp, 0);
183 return (0);
184 }
185
186 if (limit > MAXOFF_T)
187 limit = MAXOFF_T;
188
189 do {
190 long offset;
191 long delta;
192
193 offset = (long)uio->uio_offset;
194 pageoffset = offset & PAGEOFFSET;
195 /*
196 * A maximum of PAGESIZE bytes of data is transferred
197 * each pass through this loop
198 */
199 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
200
201 if (offset + bytes >= limit) {
202 if (offset >= limit) {
203 error = EFBIG;
204 goto out;
205 }
206 bytes = limit - offset;
207 }
208 pagenumber = btop(offset);
209
210 /*
211 * delta is the amount of anonymous memory
212 * to reserve for the file.
213 * We always reserve in pagesize increments so
214 * unless we're extending the file into a new page,
215 * we don't need to call tmp_resv.
216 */
217 delta = offset + bytes -
218 P2ROUNDUP_TYPED(tp->tn_size, PAGESIZE, u_offset_t);
219 if (delta > 0) {
220 pagecreate = 1;
221 if (tmp_resv(tm, tp, delta, pagecreate)) {
222 /*
223 * Log file system full in the zone that owns
224 * the tmpfs mount, as well as in the global
225 * zone if necessary.
226 */
227 zcmn_err(tm->tm_vfsp->vfs_zone->zone_id,
228 CE_WARN, "%s: File system full, "
229 "swap space limit exceeded",
230 tm->tm_mntpath);
231
232 if (tm->tm_vfsp->vfs_zone->zone_id !=
233 GLOBAL_ZONEID) {
234
235 vfs_t *vfs = tm->tm_vfsp;
236
237 zcmn_err(GLOBAL_ZONEID,
238 CE_WARN, "%s: File system full, "
239 "swap space limit exceeded",
240 vfs->vfs_vnodecovered->v_path);
241 }
242 error = ENOSPC;
243 break;
244 }
245 tmpnode_growmap(tp, (ulong_t)offset + bytes);
246 }
247 /* grow the file to the new length */
248 if (offset + bytes > tp->tn_size) {
249 tn_size_changed = 1;
250 old_tn_size = tp->tn_size;
251 /*
252 * Postpone updating tp->tn_size until uiomove() is
253 * done.
254 */
255 new_tn_size = offset + bytes;
256 }
257 if (bytes == PAGESIZE) {
258 /*
259 * Writing whole page so reading from disk
260 * is a waste
261 */
262 pagecreate = 1;
263 } else {
264 pagecreate = 0;
265 }
266 /*
267 * If writing past EOF or filling in a hole
268 * we need to allocate an anon slot.
269 */
270 if (anon_get_ptr(tp->tn_anon, pagenumber) == NULL) {
271 (void) anon_set_ptr(tp->tn_anon, pagenumber,
272 anon_alloc(vp, ptob(pagenumber)), ANON_SLEEP);
273 pagecreate = 1;
274 tp->tn_nblocks++;
275 }
276
277 /*
278 * We have to drop the contents lock to allow the VM
279 * system to reacquire it in tmp_getpage()
280 */
281 rw_exit(&tp->tn_contents);
282
283 /*
284 * Touch the page and fault it in if it is not in core
285 * before segmap_getmapflt or vpm_data_copy can lock it.
286 * This is to avoid the deadlock if the buffer is mapped
287 * to the same file through mmap which we want to write.
288 */
289 uio_prefaultpages((long)bytes, uio);
290
291 newpage = 0;
292 if (vpm_enable) {
293 /*
294 * Copy data. If new pages are created, part of
295 * the page that is not written will be initizliazed
296 * with zeros.
297 */
298 error = vpm_data_copy(vp, offset, bytes, uio,
299 !pagecreate, &newpage, 1, S_WRITE);
300 } else {
301 /* Get offset within the segmap mapping */
302 segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
303 base = segmap_getmapflt(segkmap, vp,
304 (offset & MAXBMASK), PAGESIZE, !pagecreate,
305 S_WRITE);
306 }
307
308
309 if (!vpm_enable && pagecreate) {
310 /*
311 * segmap_pagecreate() returns 1 if it calls
312 * page_create_va() to allocate any pages.
313 */
314 newpage = segmap_pagecreate(segkmap,
315 base + segmap_offset, (size_t)PAGESIZE, 0);
316 /*
317 * Clear from the beginning of the page to the starting
318 * offset of the data.
319 */
320 if (pageoffset != 0)
321 (void) kzero(base + segmap_offset,
322 (size_t)pageoffset);
323 }
324
325 if (!vpm_enable) {
326 error = uiomove(base + segmap_offset + pageoffset,
327 (long)bytes, UIO_WRITE, uio);
328 }
329
330 if (!vpm_enable && pagecreate &&
331 uio->uio_offset < P2ROUNDUP(offset + bytes, PAGESIZE)) {
332 long zoffset; /* zero from offset into page */
333 /*
334 * We created pages w/o initializing them completely,
335 * thus we need to zero the part that wasn't set up.
336 * This happens on most EOF write cases and if
337 * we had some sort of error during the uiomove.
338 */
339 long nmoved;
340
341 nmoved = uio->uio_offset - offset;
342 ASSERT((nmoved + pageoffset) <= PAGESIZE);
343
344 /*
345 * Zero from the end of data in the page to the
346 * end of the page.
347 */
348 if ((zoffset = pageoffset + nmoved) < PAGESIZE)
349 (void) kzero(base + segmap_offset + zoffset,
350 (size_t)PAGESIZE - zoffset);
351 }
352
353 /*
354 * Unlock the pages which have been allocated by
355 * page_create_va() in segmap_pagecreate()
356 */
357 if (!vpm_enable && newpage) {
358 segmap_pageunlock(segkmap, base + segmap_offset,
359 (size_t)PAGESIZE, S_WRITE);
360 }
361
362 if (error) {
363 /*
364 * If we failed on a write, we must
365 * be sure to invalidate any pages that may have
366 * been allocated.
367 */
368 if (vpm_enable) {
369 (void) vpm_sync_pages(vp, offset, PAGESIZE,
370 SM_INVAL);
371 } else {
372 (void) segmap_release(segkmap, base, SM_INVAL);
373 }
374 } else {
375 if (vpm_enable) {
376 error = vpm_sync_pages(vp, offset, PAGESIZE,
377 0);
378 } else {
379 error = segmap_release(segkmap, base, 0);
380 }
381 }
382
383 /*
384 * Re-acquire contents lock.
385 */
386 rw_enter(&tp->tn_contents, RW_WRITER);
387
388 /*
389 * Update tn_size.
390 */
391 if (tn_size_changed)
392 tp->tn_size = new_tn_size;
393
394 /*
395 * If the uiomove failed, fix up tn_size.
396 */
397 if (error) {
398 if (tn_size_changed) {
399 /*
400 * The uiomove failed, and we
401 * allocated blocks,so get rid
402 * of them.
403 */
404 (void) tmpnode_trunc(tm, tp,
405 (ulong_t)old_tn_size);
406 }
407 } else {
408 /*
409 * XXX - Can this be out of the loop?
410 */
411 if ((tp->tn_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) &&
412 (tp->tn_mode & (S_ISUID | S_ISGID)) &&
413 secpolicy_vnode_setid_retain(cr,
414 (tp->tn_mode & S_ISUID) != 0 && tp->tn_uid == 0)) {
415 /*
416 * Clear Set-UID & Set-GID bits on
417 * successful write if not privileged
418 * and at least one of the execute bits
419 * is set. If we always clear Set-GID,
420 * mandatory file and record locking is
421 * unuseable.
422 */
423 tp->tn_mode &= ~(S_ISUID | S_ISGID);
424 }
425 gethrestime(&now);
426 tp->tn_mtime = now;
427 tp->tn_ctime = now;
428 }
429 } while (error == 0 && uio->uio_resid > 0 && bytes != 0);
430
431 out:
432 /*
433 * If we've already done a partial-write, terminate
434 * the write but return no error.
435 */
436 if (oresid != uio->uio_resid)
437 error = 0;
438 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
439 "tmp_wrtmp_end:vp %p error %d", vp, error);
440 return (error);
441 }
442
443 /*
444 * rdtmp does the real work of read requests for tmpfs.
445 */
446 static int
447 rdtmp(
448 struct tmount *tm,
449 struct tmpnode *tp,
450 struct uio *uio,
451 struct caller_context *ct)
452 {
453 ulong_t pageoffset; /* offset in tmpfs file (uio_offset) */
454 ulong_t segmap_offset; /* pagesize byte offset into segmap */
455 caddr_t base; /* base of segmap */
456 ssize_t bytes; /* bytes to uiomove */
457 struct vnode *vp;
458 int error;
459 long oresid = uio->uio_resid;
460
461 #if defined(lint)
462 tm = tm;
463 #endif
464 vp = TNTOV(tp);
465
466 TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START, "tmp_rdtmp_start:vp %p",
467 vp);
468
469 ASSERT(RW_LOCK_HELD(&tp->tn_contents));
470
471 if (MANDLOCK(vp, tp->tn_mode)) {
472 rw_exit(&tp->tn_contents);
473 /*
474 * tmp_getattr ends up being called by chklock
475 */
476 error = chklock(vp, FREAD, uio->uio_loffset, uio->uio_resid,
477 uio->uio_fmode, ct);
478 rw_enter(&tp->tn_contents, RW_READER);
479 if (error != 0) {
480 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
481 "tmp_rdtmp_end:vp %p error %d", vp, error);
482 return (error);
483 }
484 }
485 ASSERT(tp->tn_type == VREG);
486
487 if (uio->uio_loffset >= MAXOFF_T) {
488 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
489 "tmp_rdtmp_end:vp %p error %d", vp, EINVAL);
490 return (0);
491 }
492 if (uio->uio_loffset < 0)
493 return (EINVAL);
494 if (uio->uio_resid == 0) {
495 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
496 "tmp_rdtmp_end:vp %p error %d", vp, 0);
497 return (0);
498 }
499
500 vp = TNTOV(tp);
501
502 do {
503 long diff;
504 long offset;
505
506 offset = uio->uio_offset;
507 pageoffset = offset & PAGEOFFSET;
508 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
509
510 diff = tp->tn_size - offset;
511
512 if (diff <= 0) {
513 error = 0;
514 goto out;
515 }
516 if (diff < bytes)
517 bytes = diff;
518
519 /*
520 * We have to drop the contents lock to allow the VM system
521 * to reacquire it in tmp_getpage() should the uiomove cause a
522 * pagefault.
523 */
524 rw_exit(&tp->tn_contents);
525
526 if (vpm_enable) {
527 /*
528 * Copy data.
529 */
530 error = vpm_data_copy(vp, offset, bytes, uio, 1, NULL,
531 0, S_READ);
532 } else {
533 segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
534 base = segmap_getmapflt(segkmap, vp, offset & MAXBMASK,
535 bytes, 1, S_READ);
536
537 error = uiomove(base + segmap_offset + pageoffset,
538 (long)bytes, UIO_READ, uio);
539 }
540
541 if (error) {
542 if (vpm_enable) {
543 (void) vpm_sync_pages(vp, offset, PAGESIZE, 0);
544 } else {
545 (void) segmap_release(segkmap, base, 0);
546 }
547 } else {
548 if (vpm_enable) {
549 error = vpm_sync_pages(vp, offset, PAGESIZE,
550 0);
551 } else {
552 error = segmap_release(segkmap, base, 0);
553 }
554 }
555
556 /*
557 * Re-acquire contents lock.
558 */
559 rw_enter(&tp->tn_contents, RW_READER);
560
561 } while (error == 0 && uio->uio_resid > 0);
562
563 out:
564 gethrestime(&tp->tn_atime);
565
566 /*
567 * If we've already done a partial read, terminate
568 * the read but return no error.
569 */
570 if (oresid != uio->uio_resid)
571 error = 0;
572
573 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
574 "tmp_rdtmp_end:vp %x error %d", vp, error);
575 return (error);
576 }
577
578 /* ARGSUSED2 */
579 static int
580 tmp_read(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cred,
581 struct caller_context *ct)
582 {
583 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
584 struct tmount *tm = (struct tmount *)VTOTM(vp);
585 int error;
586
587 /*
588 * We don't currently support reading non-regular files
589 */
590 if (vp->v_type == VDIR)
591 return (EISDIR);
592 if (vp->v_type != VREG)
593 return (EINVAL);
594 /*
595 * tmp_rwlock should have already been called from layers above
596 */
597 ASSERT(RW_READ_HELD(&tp->tn_rwlock));
598
599 rw_enter(&tp->tn_contents, RW_READER);
600
601 error = rdtmp(tm, tp, uiop, ct);
602
603 rw_exit(&tp->tn_contents);
604
605 return (error);
606 }
607
608 static int
609 tmp_write(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cred,
610 struct caller_context *ct)
611 {
612 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
613 struct tmount *tm = (struct tmount *)VTOTM(vp);
614 int error;
615
616 /*
617 * We don't currently support writing to non-regular files
618 */
619 if (vp->v_type != VREG)
620 return (EINVAL); /* XXX EISDIR? */
621
622 /*
623 * tmp_rwlock should have already been called from layers above
624 */
625 ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
626
627 rw_enter(&tp->tn_contents, RW_WRITER);
628
629 if (ioflag & FAPPEND) {
630 /*
631 * In append mode start at end of file.
632 */
633 uiop->uio_loffset = tp->tn_size;
634 }
635
636 error = wrtmp(tm, tp, uiop, cred, ct);
637
638 rw_exit(&tp->tn_contents);
639
640 return (error);
641 }
642
643 /* ARGSUSED */
644 static int
645 tmp_ioctl(
646 struct vnode *vp,
647 int com,
648 intptr_t data,
649 int flag,
650 struct cred *cred,
651 int *rvalp,
652 caller_context_t *ct)
653 {
654 return (ENOTTY);
655 }
656
657 /* ARGSUSED2 */
658 static int
659 tmp_getattr(
660 struct vnode *vp,
661 struct vattr *vap,
662 int flags,
663 struct cred *cred,
664 caller_context_t *ct)
665 {
666 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
667 struct vnode *mvp;
668 struct vattr va;
669 int attrs = 1;
670
671 /*
672 * A special case to handle the root tnode on a diskless nfs
673 * client who may have had its uid and gid inherited
674 * from an nfs vnode with nobody ownership. Likely the
675 * root filesystem. After nfs is fully functional the uid/gid
676 * may be mapable so ask again.
677 * vfsp can't get unmounted because we hold vp.
678 */
679 if (vp->v_flag & VROOT &&
680 (mvp = vp->v_vfsp->vfs_vnodecovered) != NULL) {
681 mutex_enter(&tp->tn_tlock);
682 if (tp->tn_uid == UID_NOBODY || tp->tn_gid == GID_NOBODY) {
683 mutex_exit(&tp->tn_tlock);
684 bzero(&va, sizeof (struct vattr));
685 va.va_mask = AT_UID|AT_GID;
686 attrs = VOP_GETATTR(mvp, &va, 0, cred, ct);
687 } else {
688 mutex_exit(&tp->tn_tlock);
689 }
690 }
691 mutex_enter(&tp->tn_tlock);
692 if (attrs == 0) {
693 tp->tn_uid = va.va_uid;
694 tp->tn_gid = va.va_gid;
695 }
696 vap->va_type = vp->v_type;
697 vap->va_mode = tp->tn_mode & MODEMASK;
698 vap->va_uid = tp->tn_uid;
699 vap->va_gid = tp->tn_gid;
700 vap->va_fsid = tp->tn_fsid;
701 vap->va_nodeid = (ino64_t)tp->tn_nodeid;
702 vap->va_nlink = tp->tn_nlink;
703 vap->va_size = (u_offset_t)tp->tn_size;
704 vap->va_atime = tp->tn_atime;
705 vap->va_mtime = tp->tn_mtime;
706 vap->va_ctime = tp->tn_ctime;
707 vap->va_blksize = PAGESIZE;
708 vap->va_rdev = tp->tn_rdev;
709 vap->va_seq = tp->tn_seq;
710
711 /*
712 * XXX Holes are not taken into account. We could take the time to
713 * run through the anon array looking for allocated slots...
714 */
715 vap->va_nblocks = (fsblkcnt64_t)btodb(ptob(btopr(vap->va_size)));
716 mutex_exit(&tp->tn_tlock);
717 return (0);
718 }
719
720 /*ARGSUSED4*/
721 static int
722 tmp_setattr(
723 struct vnode *vp,
724 struct vattr *vap,
725 int flags,
726 struct cred *cred,
727 caller_context_t *ct)
728 {
729 struct tmount *tm = (struct tmount *)VTOTM(vp);
730 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
731 int error = 0;
732 struct vattr *get;
733 long mask;
734
735 /*
736 * Cannot set these attributes
737 */
738 if ((vap->va_mask & AT_NOSET) || (vap->va_mask & AT_XVATTR))
739 return (EINVAL);
740
741 mutex_enter(&tp->tn_tlock);
742
743 get = &tp->tn_attr;
744 /*
745 * Change file access modes. Must be owner or have sufficient
746 * privileges.
747 */
748 error = secpolicy_vnode_setattr(cred, vp, vap, get, flags, tmp_taccess,
749 tp);
750
751 if (error)
752 goto out;
753
754 mask = vap->va_mask;
755
756 if (mask & AT_MODE) {
757 get->va_mode &= S_IFMT;
758 get->va_mode |= vap->va_mode & ~S_IFMT;
759 }
760
761 if (mask & AT_UID)
762 get->va_uid = vap->va_uid;
763 if (mask & AT_GID)
764 get->va_gid = vap->va_gid;
765 if (mask & AT_ATIME)
766 get->va_atime = vap->va_atime;
767 if (mask & AT_MTIME)
768 get->va_mtime = vap->va_mtime;
769
770 if (mask & (AT_UID | AT_GID | AT_MODE | AT_MTIME))
771 gethrestime(&tp->tn_ctime);
772
773 if (mask & AT_SIZE) {
774 ASSERT(vp->v_type != VDIR);
775
776 /* Don't support large files. */
777 if (vap->va_size > MAXOFF_T) {
778 error = EFBIG;
779 goto out;
780 }
781 mutex_exit(&tp->tn_tlock);
782
783 rw_enter(&tp->tn_rwlock, RW_WRITER);
784 rw_enter(&tp->tn_contents, RW_WRITER);
785 error = tmpnode_trunc(tm, tp, (ulong_t)vap->va_size);
786 rw_exit(&tp->tn_contents);
787 rw_exit(&tp->tn_rwlock);
788
789 if (error == 0 && vap->va_size == 0)
790 vnevent_truncate(vp, ct);
791
792 goto out1;
793 }
794 out:
795 mutex_exit(&tp->tn_tlock);
796 out1:
797 return (error);
798 }
799
800 /* ARGSUSED2 */
801 static int
802 tmp_access(
803 struct vnode *vp,
804 int mode,
805 int flags,
806 struct cred *cred,
807 caller_context_t *ct)
808 {
809 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
810 int error;
811
812 mutex_enter(&tp->tn_tlock);
813 error = tmp_taccess(tp, mode, cred);
814 mutex_exit(&tp->tn_tlock);
815 return (error);
816 }
817
818 /* ARGSUSED3 */
819 static int
820 tmp_lookup(
821 struct vnode *dvp,
822 char *nm,
823 struct vnode **vpp,
824 struct pathname *pnp,
825 int flags,
826 struct vnode *rdir,
827 struct cred *cred,
828 caller_context_t *ct,
829 int *direntflags,
830 pathname_t *realpnp)
831 {
832 struct tmpnode *tp = (struct tmpnode *)VTOTN(dvp);
833 struct tmpnode *ntp = NULL;
834 int error;
835
836
837 /* allow cd into @ dir */
838 if (flags & LOOKUP_XATTR) {
839 struct tmpnode *xdp;
840 struct tmount *tm;
841
842 /*
843 * don't allow attributes if not mounted XATTR support
844 */
845 if (!(dvp->v_vfsp->vfs_flag & VFS_XATTR))
846 return (EINVAL);
847
848 if (tp->tn_flags & ISXATTR)
849 /* No attributes on attributes */
850 return (EINVAL);
851
852 rw_enter(&tp->tn_rwlock, RW_WRITER);
853 if (tp->tn_xattrdp == NULL) {
854 if (!(flags & CREATE_XATTR_DIR)) {
855 rw_exit(&tp->tn_rwlock);
856 return (ENOENT);
857 }
858
859 /*
860 * No attribute directory exists for this
861 * node - create the attr dir as a side effect
862 * of this lookup.
863 */
864
865 /*
866 * Make sure we have adequate permission...
867 */
868
869 if ((error = tmp_taccess(tp, VWRITE, cred)) != 0) {
870 rw_exit(&tp->tn_rwlock);
871 return (error);
872 }
873
874 xdp = tmp_memalloc(sizeof (struct tmpnode),
875 TMP_MUSTHAVE);
876 tm = VTOTM(dvp);
877 tmpnode_init(tm, xdp, &tp->tn_attr, NULL);
878 /*
879 * Fix-up fields unique to attribute directories.
880 */
881 xdp->tn_flags = ISXATTR;
882 xdp->tn_type = VDIR;
883 if (tp->tn_type == VDIR) {
884 xdp->tn_mode = tp->tn_attr.va_mode;
885 } else {
886 xdp->tn_mode = 0700;
887 if (tp->tn_attr.va_mode & 0040)
888 xdp->tn_mode |= 0750;
889 if (tp->tn_attr.va_mode & 0004)
890 xdp->tn_mode |= 0705;
891 }
892 xdp->tn_vnode->v_type = VDIR;
893 xdp->tn_vnode->v_flag |= V_XATTRDIR;
894 tdirinit(tp, xdp);
895 tp->tn_xattrdp = xdp;
896 } else {
897 VN_HOLD(tp->tn_xattrdp->tn_vnode);
898 }
899 *vpp = TNTOV(tp->tn_xattrdp);
900 rw_exit(&tp->tn_rwlock);
901 return (0);
902 }
903
904 /*
905 * Null component name is a synonym for directory being searched.
906 */
907 if (*nm == '\0') {
908 VN_HOLD(dvp);
909 *vpp = dvp;
910 return (0);
911 }
912 ASSERT(tp);
913
914 error = tdirlookup(tp, nm, &ntp, cred);
915
916 if (error == 0) {
917 ASSERT(ntp);
918 *vpp = TNTOV(ntp);
919 /*
920 * If vnode is a device return special vnode instead
921 */
922 if (IS_DEVVP(*vpp)) {
923 struct vnode *newvp;
924
925 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
926 cred);
927 VN_RELE(*vpp);
928 *vpp = newvp;
929 }
930 }
931 TRACE_4(TR_FAC_TMPFS, TR_TMPFS_LOOKUP,
932 "tmpfs lookup:vp %p name %s vpp %p error %d",
933 dvp, nm, vpp, error);
934 return (error);
935 }
936
937 /*ARGSUSED7*/
938 static int
939 tmp_create(
940 struct vnode *dvp,
941 char *nm,
942 struct vattr *vap,
943 enum vcexcl exclusive,
944 int mode,
945 struct vnode **vpp,
946 struct cred *cred,
947 int flag,
948 caller_context_t *ct,
949 vsecattr_t *vsecp)
950 {
951 struct tmpnode *parent;
952 struct tmount *tm;
953 struct tmpnode *self;
954 int error;
955 struct tmpnode *oldtp;
956
957 again:
958 parent = (struct tmpnode *)VTOTN(dvp);
959 tm = (struct tmount *)VTOTM(dvp);
960 self = NULL;
961 error = 0;
962 oldtp = NULL;
963
964 /* device files not allowed in ext. attr dirs */
965 if ((parent->tn_flags & ISXATTR) &&
966 (vap->va_type == VBLK || vap->va_type == VCHR ||
967 vap->va_type == VFIFO || vap->va_type == VDOOR ||
968 vap->va_type == VSOCK || vap->va_type == VPORT))
969 return (EINVAL);
970
971 if (vap->va_type == VREG && (vap->va_mode & VSVTX)) {
972 /* Must be privileged to set sticky bit */
973 if (secpolicy_vnode_stky_modify(cred))
974 vap->va_mode &= ~VSVTX;
975 } else if (vap->va_type == VNON) {
976 return (EINVAL);
977 }
978
979 /*
980 * Null component name is a synonym for directory being searched.
981 */
982 if (*nm == '\0') {
983 VN_HOLD(dvp);
984 oldtp = parent;
985 } else {
986 error = tdirlookup(parent, nm, &oldtp, cred);
987 }
988
989 if (error == 0) { /* name found */
990 boolean_t trunc = B_FALSE;
991
992 ASSERT(oldtp);
993
994 rw_enter(&oldtp->tn_rwlock, RW_WRITER);
995
996 /*
997 * if create/read-only an existing
998 * directory, allow it
999 */
1000 if (exclusive == EXCL)
1001 error = EEXIST;
1002 else if ((oldtp->tn_type == VDIR) && (mode & VWRITE))
1003 error = EISDIR;
1004 else {
1005 error = tmp_taccess(oldtp, mode, cred);
1006 }
1007
1008 if (error) {
1009 rw_exit(&oldtp->tn_rwlock);
1010 tmpnode_rele(oldtp);
1011 return (error);
1012 }
1013 *vpp = TNTOV(oldtp);
1014 if ((*vpp)->v_type == VREG && (vap->va_mask & AT_SIZE) &&
1015 vap->va_size == 0) {
1016 rw_enter(&oldtp->tn_contents, RW_WRITER);
1017 (void) tmpnode_trunc(tm, oldtp, 0);
1018 rw_exit(&oldtp->tn_contents);
1019 trunc = B_TRUE;
1020 }
1021 rw_exit(&oldtp->tn_rwlock);
1022 if (IS_DEVVP(*vpp)) {
1023 struct vnode *newvp;
1024
1025 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
1026 cred);
1027 VN_RELE(*vpp);
1028 if (newvp == NULL) {
1029 return (ENOSYS);
1030 }
1031 *vpp = newvp;
1032 }
1033
1034 if (trunc)
1035 vnevent_create(*vpp, ct);
1036
1037 return (0);
1038 }
1039
1040 if (error != ENOENT)
1041 return (error);
1042
1043 rw_enter(&parent->tn_rwlock, RW_WRITER);
1044 error = tdirenter(tm, parent, nm, DE_CREATE,
1045 (struct tmpnode *)NULL, (struct tmpnode *)NULL,
1046 vap, &self, cred, ct);
1047 rw_exit(&parent->tn_rwlock);
1048
1049 if (error) {
1050 if (self)
1051 tmpnode_rele(self);
1052
1053 if (error == EEXIST) {
1054 /*
1055 * This means that the file was created sometime
1056 * after we checked and did not find it and when
1057 * we went to create it.
1058 * Since creat() is supposed to truncate a file
1059 * that already exits go back to the begining
1060 * of the function. This time we will find it
1061 * and go down the tmp_trunc() path
1062 */
1063 goto again;
1064 }
1065 return (error);
1066 }
1067
1068 *vpp = TNTOV(self);
1069
1070 if (!error && IS_DEVVP(*vpp)) {
1071 struct vnode *newvp;
1072
1073 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cred);
1074 VN_RELE(*vpp);
1075 if (newvp == NULL)
1076 return (ENOSYS);
1077 *vpp = newvp;
1078 }
1079 TRACE_3(TR_FAC_TMPFS, TR_TMPFS_CREATE,
1080 "tmpfs create:dvp %p nm %s vpp %p", dvp, nm, vpp);
1081 return (0);
1082 }
1083
1084 /* ARGSUSED3 */
1085 static int
1086 tmp_remove(
1087 struct vnode *dvp,
1088 char *nm,
1089 struct cred *cred,
1090 caller_context_t *ct,
1091 int flags)
1092 {
1093 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1094 int error;
1095 struct tmpnode *tp = NULL;
1096
1097 error = tdirlookup(parent, nm, &tp, cred);
1098 if (error)
1099 return (error);
1100
1101 ASSERT(tp);
1102 rw_enter(&parent->tn_rwlock, RW_WRITER);
1103 rw_enter(&tp->tn_rwlock, RW_WRITER);
1104
1105 if (tp->tn_type != VDIR ||
1106 (error = secpolicy_fs_linkdir(cred, dvp->v_vfsp)) == 0)
1107 error = 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 secpolicy_fs_linkdir(cred, dvp->v_vfsp)) ||
1144 (from->tn_uid != crgetuid(cred) && secpolicy_basic_link(cred)))
1145 return (EPERM);
1146
1147 /*
1148 * Make sure link for extended attributes is valid
1149 * We only support hard linking of xattr's in xattrdir to an xattrdir
1150 */
1151 if ((from->tn_flags & ISXATTR) != (parent->tn_flags & ISXATTR))
1152 return (EINVAL);
1153
1154 error = tdirlookup(parent, tnm, &found, cred);
1155 if (error == 0) {
1156 ASSERT(found);
1157 tmpnode_rele(found);
1158 return (EEXIST);
1159 }
1160
1161 if (error != ENOENT)
1162 return (error);
1163
1164 rw_enter(&parent->tn_rwlock, RW_WRITER);
1165 error = tdirenter(tm, parent, tnm, DE_LINK, (struct tmpnode *)NULL,
1166 from, NULL, (struct tmpnode **)NULL, cred, ct);
1167 rw_exit(&parent->tn_rwlock);
1168 if (error == 0) {
1169 vnevent_link(srcvp, ct);
1170 }
1171 return (error);
1172 }
1173
1174 /* ARGSUSED5 */
1175 static int
1176 tmp_rename(
1177 struct vnode *odvp, /* source parent vnode */
1178 char *onm, /* source name */
1179 struct vnode *ndvp, /* destination parent vnode */
1180 char *nnm, /* destination name */
1181 struct cred *cred,
1182 caller_context_t *ct,
1183 int flags)
1184 {
1185 struct tmpnode *fromparent;
1186 struct tmpnode *toparent;
1187 struct tmpnode *fromtp = NULL; /* source tmpnode */
1188 struct tmpnode *totp; /* target tmpnode */
1189 struct tmount *tm = (struct tmount *)VTOTM(odvp);
1190 int error;
1191 int samedir = 0; /* set if odvp == ndvp */
1192 struct vnode *realvp;
1193
1194 if (VOP_REALVP(ndvp, &realvp, ct) == 0)
1195 ndvp = realvp;
1196
1197 fromparent = (struct tmpnode *)VTOTN(odvp);
1198 toparent = (struct tmpnode *)VTOTN(ndvp);
1199
1200 if ((fromparent->tn_flags & ISXATTR) != (toparent->tn_flags & ISXATTR))
1201 return (EINVAL);
1202
1203 mutex_enter(&tm->tm_renamelck);
1204
1205 /*
1206 * Look up tmpnode of file we're supposed to rename.
1207 */
1208 error = tdirlookup(fromparent, onm, &fromtp, cred);
1209 if (error) {
1210 mutex_exit(&tm->tm_renamelck);
1211 return (error);
1212 }
1213
1214 /*
1215 * Make sure we can delete the old (source) entry. This
1216 * requires write permission on the containing directory. If
1217 * that directory is "sticky" it requires further checks.
1218 */
1219 if (((error = tmp_taccess(fromparent, VWRITE, cred)) != 0) ||
1220 (error = tmp_sticky_remove_access(fromparent, fromtp, cred)) != 0)
1221 goto done;
1222
1223 /*
1224 * Check for renaming to or from '.' or '..' or that
1225 * fromtp == fromparent
1226 */
1227 if ((onm[0] == '.' &&
1228 (onm[1] == '\0' || (onm[1] == '.' && onm[2] == '\0'))) ||
1229 (nnm[0] == '.' &&
1230 (nnm[1] == '\0' || (nnm[1] == '.' && nnm[2] == '\0'))) ||
1231 (fromparent == fromtp)) {
1232 error = EINVAL;
1233 goto done;
1234 }
1235
1236 samedir = (fromparent == toparent);
1237 /*
1238 * Make sure we can search and rename into the new
1239 * (destination) directory.
1240 */
1241 if (!samedir) {
1242 error = tmp_taccess(toparent, VEXEC|VWRITE, cred);
1243 if (error)
1244 goto done;
1245 }
1246
1247 if (tdirlookup(toparent, nnm, &totp, cred) == 0) {
1248 vnevent_pre_rename_dest(TNTOV(totp), ndvp, nnm, ct);
1249 tmpnode_rele(totp);
1250 }
1251
1252 /* Notify the target dir. if not the same as the source dir. */
1253 if (ndvp != odvp) {
1254 vnevent_pre_rename_dest_dir(ndvp, TNTOV(fromtp), nnm, ct);
1255 }
1256
1257 vnevent_pre_rename_src(TNTOV(fromtp), odvp, onm, ct);
1258
1259 /*
1260 * Link source to new target
1261 */
1262 rw_enter(&toparent->tn_rwlock, RW_WRITER);
1263 error = tdirenter(tm, toparent, nnm, DE_RENAME,
1264 fromparent, fromtp, (struct vattr *)NULL,
1265 (struct tmpnode **)NULL, cred, ct);
1266 rw_exit(&toparent->tn_rwlock);
1267
1268 if (error) {
1269 /*
1270 * ESAME isn't really an error; it indicates that the
1271 * operation should not be done because the source and target
1272 * are the same file, but that no error should be reported.
1273 */
1274 if (error == ESAME)
1275 error = 0;
1276 goto done;
1277 }
1278
1279 /*
1280 * Unlink from source.
1281 */
1282 rw_enter(&fromparent->tn_rwlock, RW_WRITER);
1283 rw_enter(&fromtp->tn_rwlock, RW_WRITER);
1284
1285 error = tdirdelete(fromparent, fromtp, onm, DR_RENAME, cred);
1286
1287 /*
1288 * The following handles the case where our source tmpnode was
1289 * removed before we got to it.
1290 *
1291 * XXX We should also cleanup properly in the case where tdirdelete
1292 * fails for some other reason. Currently this case shouldn't happen.
1293 * (see 1184991).
1294 */
1295 if (error == ENOENT)
1296 error = 0;
1297
1298 rw_exit(&fromtp->tn_rwlock);
1299 rw_exit(&fromparent->tn_rwlock);
1300
1301 if (error == 0) {
1302 vnevent_rename_src(TNTOV(fromtp), odvp, onm, ct);
1303 /*
1304 * vnevent_rename_dest is called in tdirenter().
1305 * Notify the target dir if not same as source dir.
1306 */
1307 if (ndvp != odvp)
1308 vnevent_rename_dest_dir(ndvp, ct);
1309 }
1310
1311 done:
1312 tmpnode_rele(fromtp);
1313 mutex_exit(&tm->tm_renamelck);
1314
1315 TRACE_5(TR_FAC_TMPFS, TR_TMPFS_RENAME,
1316 "tmpfs rename:ovp %p onm %s nvp %p nnm %s error %d", odvp, onm,
1317 ndvp, nnm, error);
1318 return (error);
1319 }
1320
1321 /* ARGSUSED5 */
1322 static int
1323 tmp_mkdir(
1324 struct vnode *dvp,
1325 char *nm,
1326 struct vattr *va,
1327 struct vnode **vpp,
1328 struct cred *cred,
1329 caller_context_t *ct,
1330 int flags,
1331 vsecattr_t *vsecp)
1332 {
1333 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1334 struct tmpnode *self = NULL;
1335 struct tmount *tm = (struct tmount *)VTOTM(dvp);
1336 int error;
1337
1338 /* no new dirs allowed in xattr dirs */
1339 if (parent->tn_flags & ISXATTR)
1340 return (EINVAL);
1341
1342 /*
1343 * Might be dangling directory. Catch it here,
1344 * because a ENOENT return from tdirlookup() is
1345 * an "o.k. return".
1346 */
1347 if (parent->tn_nlink == 0)
1348 return (ENOENT);
1349
1350 error = tdirlookup(parent, nm, &self, cred);
1351 if (error == 0) {
1352 ASSERT(self);
1353 tmpnode_rele(self);
1354 return (EEXIST);
1355 }
1356 if (error != ENOENT)
1357 return (error);
1358
1359 rw_enter(&parent->tn_rwlock, RW_WRITER);
1360 error = tdirenter(tm, parent, nm, DE_MKDIR, (struct tmpnode *)NULL,
1361 (struct tmpnode *)NULL, va, &self, cred, ct);
1362 if (error) {
1363 rw_exit(&parent->tn_rwlock);
1364 if (self)
1365 tmpnode_rele(self);
1366 return (error);
1367 }
1368 rw_exit(&parent->tn_rwlock);
1369 *vpp = TNTOV(self);
1370 return (0);
1371 }
1372
1373 /* ARGSUSED4 */
1374 static int
1375 tmp_rmdir(
1376 struct vnode *dvp,
1377 char *nm,
1378 struct vnode *cdir,
1379 struct cred *cred,
1380 caller_context_t *ct,
1381 int flags)
1382 {
1383 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1384 struct tmpnode *self = NULL;
1385 struct vnode *vp;
1386 int error = 0;
1387
1388 /*
1389 * Return error when removing . and ..
1390 */
1391 if (strcmp(nm, ".") == 0)
1392 return (EINVAL);
1393 if (strcmp(nm, "..") == 0)
1394 return (EEXIST); /* Should be ENOTEMPTY */
1395 error = tdirlookup(parent, nm, &self, cred);
1396 if (error)
1397 return (error);
1398
1399 rw_enter(&parent->tn_rwlock, RW_WRITER);
1400 rw_enter(&self->tn_rwlock, RW_WRITER);
1401
1402 vp = TNTOV(self);
1403 if (vp == dvp || vp == cdir) {
1404 error = EINVAL;
1405 goto done1;
1406 }
1407 if (self->tn_type != VDIR) {
1408 error = ENOTDIR;
1409 goto done1;
1410 }
1411
1412 mutex_enter(&self->tn_tlock);
1413 if (self->tn_nlink > 2) {
1414 mutex_exit(&self->tn_tlock);
1415 error = EEXIST;
1416 goto done1;
1417 }
1418 mutex_exit(&self->tn_tlock);
1419
1420 if (vn_vfswlock(vp)) {
1421 error = EBUSY;
1422 goto done1;
1423 }
1424 if (vn_mountedvfs(vp) != NULL) {
1425 error = EBUSY;
1426 goto done;
1427 }
1428
1429 /*
1430 * Check for an empty directory
1431 * i.e. only includes entries for "." and ".."
1432 */
1433 if (self->tn_dirents > 2) {
1434 error = EEXIST; /* SIGH should be ENOTEMPTY */
1435 /*
1436 * Update atime because checking tn_dirents is logically
1437 * equivalent to reading the directory
1438 */
1439 gethrestime(&self->tn_atime);
1440 goto done;
1441 }
1442
1443 error = tdirdelete(parent, self, nm, DR_RMDIR, cred);
1444 done:
1445 vn_vfsunlock(vp);
1446 done1:
1447 rw_exit(&self->tn_rwlock);
1448 rw_exit(&parent->tn_rwlock);
1449 vnevent_rmdir(TNTOV(self), dvp, nm, ct);
1450 tmpnode_rele(self);
1451
1452 return (error);
1453 }
1454
1455 /* ARGSUSED2 */
1456 static int
1457 tmp_readdir(
1458 struct vnode *vp,
1459 struct uio *uiop,
1460 struct cred *cred,
1461 int *eofp,
1462 caller_context_t *ct,
1463 int flags)
1464 {
1465 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1466 struct tdirent *tdp;
1467 int error = 0;
1468 size_t namelen;
1469 struct dirent64 *dp;
1470 ulong_t offset;
1471 ulong_t total_bytes_wanted;
1472 long outcount = 0;
1473 long bufsize;
1474 int reclen;
1475 caddr_t outbuf;
1476
1477 if (uiop->uio_loffset >= MAXOFF_T) {
1478 if (eofp)
1479 *eofp = 1;
1480 return (0);
1481 }
1482 /*
1483 * assuming system call has already called tmp_rwlock
1484 */
1485 ASSERT(RW_READ_HELD(&tp->tn_rwlock));
1486
1487 if (uiop->uio_iovcnt != 1)
1488 return (EINVAL);
1489
1490 if (vp->v_type != VDIR)
1491 return (ENOTDIR);
1492
1493 /*
1494 * There's a window here where someone could have removed
1495 * all the entries in the directory after we put a hold on the
1496 * vnode but before we grabbed the rwlock. Just return.
1497 */
1498 if (tp->tn_dir == NULL) {
1499 if (tp->tn_nlink) {
1500 panic("empty directory 0x%p", (void *)tp);
1501 /*NOTREACHED*/
1502 }
1503 return (0);
1504 }
1505
1506 /*
1507 * Get space for multiple directory entries
1508 */
1509 total_bytes_wanted = uiop->uio_iov->iov_len;
1510 bufsize = total_bytes_wanted + sizeof (struct dirent64);
1511 outbuf = kmem_alloc(bufsize, KM_SLEEP);
1512
1513 dp = (struct dirent64 *)outbuf;
1514
1515
1516 offset = 0;
1517 tdp = tp->tn_dir;
1518 while (tdp) {
1519 namelen = strlen(tdp->td_name); /* no +1 needed */
1520 offset = tdp->td_offset;
1521 if (offset >= uiop->uio_offset) {
1522 reclen = (int)DIRENT64_RECLEN(namelen);
1523 if (outcount + reclen > total_bytes_wanted) {
1524 if (!outcount)
1525 /*
1526 * Buffer too small for any entries.
1527 */
1528 error = EINVAL;
1529 break;
1530 }
1531 ASSERT(tdp->td_tmpnode != NULL);
1532
1533 /* use strncpy(9f) to zero out uninitialized bytes */
1534
1535 (void) strncpy(dp->d_name, tdp->td_name,
1536 DIRENT64_NAMELEN(reclen));
1537 dp->d_reclen = (ushort_t)reclen;
1538 dp->d_ino = (ino64_t)tdp->td_tmpnode->tn_nodeid;
1539 dp->d_off = (offset_t)tdp->td_offset + 1;
1540 dp = (struct dirent64 *)
1541 ((uintptr_t)dp + dp->d_reclen);
1542 outcount += reclen;
1543 ASSERT(outcount <= bufsize);
1544 }
1545 tdp = tdp->td_next;
1546 }
1547
1548 if (!error)
1549 error = uiomove(outbuf, outcount, UIO_READ, uiop);
1550
1551 if (!error) {
1552 /* If we reached the end of the list our offset */
1553 /* should now be just past the end. */
1554 if (!tdp) {
1555 offset += 1;
1556 if (eofp)
1557 *eofp = 1;
1558 } else if (eofp)
1559 *eofp = 0;
1560 uiop->uio_offset = offset;
1561 }
1562 gethrestime(&tp->tn_atime);
1563 kmem_free(outbuf, bufsize);
1564 return (error);
1565 }
1566
1567 /* ARGSUSED5 */
1568 static int
1569 tmp_symlink(
1570 struct vnode *dvp,
1571 char *lnm,
1572 struct vattr *tva,
1573 char *tnm,
1574 struct cred *cred,
1575 caller_context_t *ct,
1576 int flags)
1577 {
1578 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1579 struct tmpnode *self = (struct tmpnode *)NULL;
1580 struct tmount *tm = (struct tmount *)VTOTM(dvp);
1581 char *cp = NULL;
1582 int error;
1583 size_t len;
1584
1585 /* no symlinks allowed to files in xattr dirs */
1586 if (parent->tn_flags & ISXATTR)
1587 return (EINVAL);
1588
1589 error = tdirlookup(parent, lnm, &self, cred);
1590 if (error == 0) {
1591 /*
1592 * The entry already exists
1593 */
1594 tmpnode_rele(self);
1595 return (EEXIST); /* was 0 */
1596 }
1597
1598 if (error != ENOENT) {
1599 if (self != NULL)
1600 tmpnode_rele(self);
1601 return (error);
1602 }
1603
1604 rw_enter(&parent->tn_rwlock, RW_WRITER);
1605 error = tdirenter(tm, parent, lnm, DE_CREATE, (struct tmpnode *)NULL,
1606 (struct tmpnode *)NULL, tva, &self, cred, ct);
1607 rw_exit(&parent->tn_rwlock);
1608
1609 if (error) {
1610 if (self)
1611 tmpnode_rele(self);
1612 return (error);
1613 }
1614 len = strlen(tnm) + 1;
1615 cp = tmp_memalloc(len, 0);
1616 if (cp == NULL) {
1617 tmpnode_rele(self);
1618 return (ENOSPC);
1619 }
1620 (void) strcpy(cp, tnm);
1621
1622 self->tn_symlink = cp;
1623 self->tn_size = len - 1;
1624 tmpnode_rele(self);
1625 return (error);
1626 }
1627
1628 /* ARGSUSED2 */
1629 static int
1630 tmp_readlink(
1631 struct vnode *vp,
1632 struct uio *uiop,
1633 struct cred *cred,
1634 caller_context_t *ct)
1635 {
1636 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1637 int error = 0;
1638
1639 if (vp->v_type != VLNK)
1640 return (EINVAL);
1641
1642 rw_enter(&tp->tn_rwlock, RW_READER);
1643 rw_enter(&tp->tn_contents, RW_READER);
1644 error = uiomove(tp->tn_symlink, tp->tn_size, UIO_READ, uiop);
1645 gethrestime(&tp->tn_atime);
1646 rw_exit(&tp->tn_contents);
1647 rw_exit(&tp->tn_rwlock);
1648 return (error);
1649 }
1650
1651 /* ARGSUSED */
1652 static int
1653 tmp_fsync(
1654 struct vnode *vp,
1655 int syncflag,
1656 struct cred *cred,
1657 caller_context_t *ct)
1658 {
1659 return (0);
1660 }
1661
1662 /* ARGSUSED */
1663 static void
1664 tmp_inactive(struct vnode *vp, struct cred *cred, caller_context_t *ct)
1665 {
1666 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1667 struct tmount *tm = (struct tmount *)VFSTOTM(vp->v_vfsp);
1668
1669 rw_enter(&tp->tn_rwlock, RW_WRITER);
1670 top:
1671 mutex_enter(&tp->tn_tlock);
1672 mutex_enter(&vp->v_lock);
1673 ASSERT(vp->v_count >= 1);
1674
1675 /*
1676 * If we don't have the last hold or the link count is non-zero,
1677 * there's little to do -- just drop our hold.
1678 */
1679 if (vp->v_count > 1 || tp->tn_nlink != 0) {
1680 vp->v_count--;
1681 mutex_exit(&vp->v_lock);
1682 mutex_exit(&tp->tn_tlock);
1683 rw_exit(&tp->tn_rwlock);
1684 return;
1685 }
1686
1687 /*
1688 * We have the last hold *and* the link count is zero, so this
1689 * tmpnode is dead from the filesystem's viewpoint. However,
1690 * if the tmpnode has any pages associated with it (i.e. if it's
1691 * a normal file with non-zero size), the tmpnode can still be
1692 * discovered by pageout or fsflush via the page vnode pointers.
1693 * In this case we must drop all our locks, truncate the tmpnode,
1694 * and try the whole dance again.
1695 */
1696 if (tp->tn_size != 0) {
1697 if (tp->tn_type == VREG) {
1698 mutex_exit(&vp->v_lock);
1699 mutex_exit(&tp->tn_tlock);
1700 rw_enter(&tp->tn_contents, RW_WRITER);
1701 (void) tmpnode_trunc(tm, tp, 0);
1702 rw_exit(&tp->tn_contents);
1703 ASSERT(tp->tn_size == 0);
1704 ASSERT(tp->tn_nblocks == 0);
1705 goto top;
1706 }
1707 if (tp->tn_type == VLNK)
1708 tmp_memfree(tp->tn_symlink, tp->tn_size + 1);
1709 }
1710
1711 /*
1712 * Remove normal file/dir's xattr dir and xattrs.
1713 */
1714 if (tp->tn_xattrdp) {
1715 struct tmpnode *xtp = tp->tn_xattrdp;
1716
1717 ASSERT(xtp->tn_flags & ISXATTR);
1718 tmpnode_hold(xtp);
1719 rw_enter(&xtp->tn_rwlock, RW_WRITER);
1720 tdirtrunc(xtp);
1721 DECR_COUNT(&xtp->tn_nlink, &xtp->tn_tlock);
1722 tp->tn_xattrdp = NULL;
1723 rw_exit(&xtp->tn_rwlock);
1724 tmpnode_rele(xtp);
1725 }
1726
1727 mutex_exit(&vp->v_lock);
1728 mutex_exit(&tp->tn_tlock);
1729 /* Here's our chance to send invalid event while we're between locks */
1730 vn_invalid(TNTOV(tp));
1731 mutex_enter(&tm->tm_contents);
1732 if (tp->tn_forw == NULL)
1733 tm->tm_rootnode->tn_back = tp->tn_back;
1734 else
1735 tp->tn_forw->tn_back = tp->tn_back;
1736 tp->tn_back->tn_forw = tp->tn_forw;
1737 mutex_exit(&tm->tm_contents);
1738 rw_exit(&tp->tn_rwlock);
1739 rw_destroy(&tp->tn_rwlock);
1740 mutex_destroy(&tp->tn_tlock);
1741 vn_free(TNTOV(tp));
1742 tmp_memfree(tp, sizeof (struct tmpnode));
1743 }
1744
1745 /* ARGSUSED2 */
1746 static int
1747 tmp_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
1748 {
1749 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1750 struct tfid *tfid;
1751
1752 if (fidp->fid_len < (sizeof (struct tfid) - sizeof (ushort_t))) {
1753 fidp->fid_len = sizeof (struct tfid) - sizeof (ushort_t);
1754 return (ENOSPC);
1755 }
1756
1757 tfid = (struct tfid *)fidp;
1758 bzero(tfid, sizeof (struct tfid));
1759 tfid->tfid_len = (int)sizeof (struct tfid) - sizeof (ushort_t);
1760
1761 tfid->tfid_ino = tp->tn_nodeid;
1762 tfid->tfid_gen = tp->tn_gen;
1763
1764 return (0);
1765 }
1766
1767
1768 /*
1769 * Return all the pages from [off..off+len] in given file
1770 */
1771 /* ARGSUSED */
1772 static int
1773 tmp_getpage(
1774 struct vnode *vp,
1775 offset_t off,
1776 size_t len,
1777 uint_t *protp,
1778 page_t *pl[],
1779 size_t plsz,
1780 struct seg *seg,
1781 caddr_t addr,
1782 enum seg_rw rw,
1783 struct cred *cr,
1784 caller_context_t *ct)
1785 {
1786 int err = 0;
1787 struct tmpnode *tp = VTOTN(vp);
1788 anoff_t toff = (anoff_t)off;
1789 size_t tlen = len;
1790 u_offset_t tmpoff;
1791 timestruc_t now;
1792
1793 rw_enter(&tp->tn_contents, RW_READER);
1794
1795 if (off + len > tp->tn_size + PAGEOFFSET) {
1796 err = EFAULT;
1797 goto out;
1798 }
1799 /*
1800 * Look for holes (no anon slot) in faulting range. If there are
1801 * holes we have to switch to a write lock and fill them in. Swap
1802 * space for holes was already reserved when the file was grown.
1803 */
1804 tmpoff = toff;
1805 if (non_anon(tp->tn_anon, btop(off), &tmpoff, &tlen)) {
1806 if (!rw_tryupgrade(&tp->tn_contents)) {
1807 rw_exit(&tp->tn_contents);
1808 rw_enter(&tp->tn_contents, RW_WRITER);
1809 /* Size may have changed when lock was dropped */
1810 if (off + len > tp->tn_size + PAGEOFFSET) {
1811 err = EFAULT;
1812 goto out;
1813 }
1814 }
1815 for (toff = (anoff_t)off; toff < (anoff_t)off + len;
1816 toff += PAGESIZE) {
1817 if (anon_get_ptr(tp->tn_anon, btop(toff)) == NULL) {
1818 /* XXX - may allocate mem w. write lock held */
1819 (void) anon_set_ptr(tp->tn_anon, btop(toff),
1820 anon_alloc(vp, toff), ANON_SLEEP);
1821 tp->tn_nblocks++;
1822 }
1823 }
1824 rw_downgrade(&tp->tn_contents);
1825 }
1826
1827
1828 err = pvn_getpages(tmp_getapage, vp, (u_offset_t)off, len, protp,
1829 pl, plsz, seg, addr, rw, cr);
1830
1831 gethrestime(&now);
1832 tp->tn_atime = now;
1833 if (rw == S_WRITE)
1834 tp->tn_mtime = now;
1835
1836 out:
1837 rw_exit(&tp->tn_contents);
1838 return (err);
1839 }
1840
1841 /*
1842 * Called from pvn_getpages to get a particular page.
1843 */
1844 /*ARGSUSED*/
1845 static int
1846 tmp_getapage(
1847 struct vnode *vp,
1848 u_offset_t off,
1849 size_t len,
1850 uint_t *protp,
1851 page_t *pl[],
1852 size_t plsz,
1853 struct seg *seg,
1854 caddr_t addr,
1855 enum seg_rw rw,
1856 struct cred *cr)
1857 {
1858 struct page *pp;
1859 int flags;
1860 int err = 0;
1861 struct vnode *pvp;
1862 u_offset_t poff;
1863
1864 if (protp != NULL)
1865 *protp = PROT_ALL;
1866 again:
1867 if (pp = page_lookup(vp, off, rw == S_CREATE ? SE_EXCL : SE_SHARED)) {
1868 if (pl) {
1869 pl[0] = pp;
1870 pl[1] = NULL;
1871 } else {
1872 page_unlock(pp);
1873 }
1874 } else {
1875 pp = page_create_va(vp, off, PAGESIZE,
1876 PG_WAIT | PG_EXCL, seg, addr);
1877 /*
1878 * Someone raced in and created the page after we did the
1879 * lookup but before we did the create, so go back and
1880 * try to look it up again.
1881 */
1882 if (pp == NULL)
1883 goto again;
1884 /*
1885 * Fill page from backing store, if any. If none, then
1886 * either this is a newly filled hole or page must have
1887 * been unmodified and freed so just zero it out.
1888 */
1889 err = swap_getphysname(vp, off, &pvp, &poff);
1890 if (err) {
1891 panic("tmp_getapage: no anon slot vp %p "
1892 "off %llx pp %p\n", (void *)vp, off, (void *)pp);
1893 }
1894 if (pvp) {
1895 flags = (pl == NULL ? B_ASYNC|B_READ : B_READ);
1896 err = VOP_PAGEIO(pvp, pp, (u_offset_t)poff, PAGESIZE,
1897 flags, cr, NULL);
1898 if (flags & B_ASYNC)
1899 pp = NULL;
1900 } else if (rw != S_CREATE) {
1901 pagezero(pp, 0, PAGESIZE);
1902 }
1903 if (err && pp)
1904 pvn_read_done(pp, B_ERROR);
1905 if (err == 0) {
1906 if (pl)
1907 pvn_plist_init(pp, pl, plsz, off, PAGESIZE, rw);
1908 else
1909 pvn_io_done(pp);
1910 }
1911 }
1912 return (err);
1913 }
1914
1915
1916 /*
1917 * Flags are composed of {B_INVAL, B_DIRTY B_FREE, B_DONTNEED}.
1918 * If len == 0, do from off to EOF.
1919 */
1920 static int tmp_nopage = 0; /* Don't do tmp_putpage's if set */
1921
1922 /* ARGSUSED */
1923 int
1924 tmp_putpage(
1925 register struct vnode *vp,
1926 offset_t off,
1927 size_t len,
1928 int flags,
1929 struct cred *cr,
1930 caller_context_t *ct)
1931 {
1932 register page_t *pp;
1933 u_offset_t io_off;
1934 size_t io_len = 0;
1935 int err = 0;
1936 struct tmpnode *tp = VTOTN(vp);
1937 int dolock;
1938
1939 if (tmp_nopage)
1940 return (0);
1941
1942 ASSERT(vp->v_count != 0);
1943
1944 if (vp->v_flag & VNOMAP)
1945 return (ENOSYS);
1946
1947 /*
1948 * This being tmpfs, we don't ever do i/o unless we really
1949 * have to (when we're low on memory and pageout calls us
1950 * with B_ASYNC | B_FREE or the user explicitly asks for it with
1951 * B_DONTNEED).
1952 * XXX to approximately track the mod time like ufs we should
1953 * update the times here. The problem is, once someone does a
1954 * store we never clear the mod bit and do i/o, thus fsflush
1955 * will keep calling us every 30 seconds to do the i/o and we'll
1956 * continually update the mod time. At least we update the mod
1957 * time on the first store because this results in a call to getpage.
1958 */
1959 if (flags != (B_ASYNC | B_FREE) && (flags & B_INVAL) == 0 &&
1960 (flags & B_DONTNEED) == 0)
1961 return (0);
1962 /*
1963 * If this thread owns the lock, i.e., this thread grabbed it
1964 * as writer somewhere above, then we don't need to grab the
1965 * lock as reader in this routine.
1966 */
1967 dolock = (rw_owner(&tp->tn_contents) != curthread);
1968
1969 /*
1970 * If this is pageout don't block on the lock as you could deadlock
1971 * when freemem == 0 (another thread has the read lock and is blocked
1972 * creating a page, and a third thread is waiting to get the writers
1973 * lock - waiting writers priority blocks us from getting the read
1974 * lock). Of course, if the only freeable pages are on this tmpnode
1975 * we're hosed anyways. A better solution might be a new lock type.
1976 * Note: ufs has the same problem.
1977 */
1978 if (curproc == proc_pageout) {
1979 if (!rw_tryenter(&tp->tn_contents, RW_READER))
1980 return (ENOMEM);
1981 } else if (dolock)
1982 rw_enter(&tp->tn_contents, RW_READER);
1983
1984 if (!vn_has_cached_data(vp))
1985 goto out;
1986
1987 if (len == 0) {
1988 if (curproc == proc_pageout) {
1989 panic("tmp: pageout can't block");
1990 /*NOTREACHED*/
1991 }
1992
1993 /* Search the entire vp list for pages >= off. */
1994 err = pvn_vplist_dirty(vp, (u_offset_t)off, tmp_putapage,
1995 flags, cr);
1996 } else {
1997 u_offset_t eoff;
1998
1999 /*
2000 * Loop over all offsets in the range [off...off + len]
2001 * looking for pages to deal with.
2002 */
2003 eoff = MIN(off + len, tp->tn_size);
2004 for (io_off = off; io_off < eoff; io_off += io_len) {
2005 /*
2006 * If we are not invalidating, synchronously
2007 * freeing or writing pages use the routine
2008 * page_lookup_nowait() to prevent reclaiming
2009 * them from the free list.
2010 */
2011 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2012 pp = page_lookup(vp, io_off,
2013 (flags & (B_INVAL | B_FREE)) ?
2014 SE_EXCL : SE_SHARED);
2015 } else {
2016 pp = page_lookup_nowait(vp, io_off,
2017 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2018 }
2019
2020 if (pp == NULL || pvn_getdirty(pp, flags) == 0)
2021 io_len = PAGESIZE;
2022 else {
2023 err = tmp_putapage(vp, pp, &io_off, &io_len,
2024 flags, cr);
2025 if (err != 0)
2026 break;
2027 }
2028 }
2029 }
2030 /* If invalidating, verify all pages on vnode list are gone. */
2031 if (err == 0 && off == 0 && len == 0 &&
2032 (flags & B_INVAL) && vn_has_cached_data(vp)) {
2033 panic("tmp_putpage: B_INVAL, pages not gone");
2034 /*NOTREACHED*/
2035 }
2036 out:
2037 if ((curproc == proc_pageout) || dolock)
2038 rw_exit(&tp->tn_contents);
2039 /*
2040 * Only reason putapage is going to give us SE_NOSWAP as error
2041 * is when we ask a page to be written to physical backing store
2042 * and there is none. Ignore this because we might be dealing
2043 * with a swap page which does not have any backing store
2044 * on disk. In any other case we won't get this error over here.
2045 */
2046 if (err == SE_NOSWAP)
2047 err = 0;
2048 return (err);
2049 }
2050
2051 long tmp_putpagecnt, tmp_pagespushed;
2052
2053 /*
2054 * Write out a single page.
2055 * For tmpfs this means choose a physical swap slot and write the page
2056 * out using VOP_PAGEIO. For performance, we attempt to kluster; i.e.,
2057 * we try to find a bunch of other dirty pages adjacent in the file
2058 * and a bunch of contiguous swap slots, and then write all the pages
2059 * out in a single i/o.
2060 */
2061 /*ARGSUSED*/
2062 static int
2063 tmp_putapage(
2064 struct vnode *vp,
2065 page_t *pp,
2066 u_offset_t *offp,
2067 size_t *lenp,
2068 int flags,
2069 struct cred *cr)
2070 {
2071 int err;
2072 ulong_t klstart, kllen;
2073 page_t *pplist, *npplist;
2074 extern int klustsize;
2075 long tmp_klustsize;
2076 struct tmpnode *tp;
2077 size_t pp_off, pp_len;
2078 u_offset_t io_off;
2079 size_t io_len;
2080 struct vnode *pvp;
2081 u_offset_t pstart;
2082 u_offset_t offset;
2083 u_offset_t tmpoff;
2084
2085 ASSERT(PAGE_LOCKED(pp));
2086
2087 /* Kluster in tmp_klustsize chunks */
2088 tp = VTOTN(vp);
2089 tmp_klustsize = klustsize;
2090 offset = pp->p_offset;
2091 klstart = (offset / tmp_klustsize) * tmp_klustsize;
2092 kllen = MIN(tmp_klustsize, tp->tn_size - klstart);
2093
2094 /* Get a kluster of pages */
2095 pplist =
2096 pvn_write_kluster(vp, pp, &tmpoff, &pp_len, klstart, kllen, flags);
2097
2098 pp_off = (size_t)tmpoff;
2099
2100 /*
2101 * Get a cluster of physical offsets for the pages; the amount we
2102 * get may be some subrange of what we ask for (io_off, io_len).
2103 */
2104 io_off = pp_off;
2105 io_len = pp_len;
2106 err = swap_newphysname(vp, offset, &io_off, &io_len, &pvp, &pstart);
2107 ASSERT(err != SE_NOANON); /* anon slot must have been filled */
2108 if (err) {
2109 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2110 /*
2111 * If this routine is called as a result of segvn_sync
2112 * operation and we have no physical swap then we can get an
2113 * error here. In such case we would return SE_NOSWAP as error.
2114 * At this point, we expect only SE_NOSWAP.
2115 */
2116 ASSERT(err == SE_NOSWAP);
2117 if (flags & B_INVAL)
2118 err = ENOMEM;
2119 goto out;
2120 }
2121 ASSERT(pp_off <= io_off && io_off + io_len <= pp_off + pp_len);
2122 ASSERT(io_off <= offset && offset < io_off + io_len);
2123
2124 /* Toss pages at front/rear that we couldn't get physical backing for */
2125 if (io_off != pp_off) {
2126 npplist = NULL;
2127 page_list_break(&pplist, &npplist, btop(io_off - pp_off));
2128 ASSERT(pplist->p_offset == pp_off);
2129 ASSERT(pplist->p_prev->p_offset == io_off - PAGESIZE);
2130 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2131 pplist = npplist;
2132 }
2133 if (io_off + io_len < pp_off + pp_len) {
2134 npplist = NULL;
2135 page_list_break(&pplist, &npplist, btop(io_len));
2136 ASSERT(npplist->p_offset == io_off + io_len);
2137 ASSERT(npplist->p_prev->p_offset == pp_off + pp_len - PAGESIZE);
2138 pvn_write_done(npplist, B_ERROR | B_WRITE | flags);
2139 }
2140
2141 ASSERT(pplist->p_offset == io_off);
2142 ASSERT(pplist->p_prev->p_offset == io_off + io_len - PAGESIZE);
2143 ASSERT(btopr(io_len) <= btopr(kllen));
2144
2145 /* Do i/o on the remaining kluster */
2146 err = VOP_PAGEIO(pvp, pplist, (u_offset_t)pstart, io_len,
2147 B_WRITE | flags, cr, NULL);
2148
2149 if ((flags & B_ASYNC) == 0) {
2150 pvn_write_done(pplist, ((err) ? B_ERROR : 0) | B_WRITE | flags);
2151 }
2152 out:
2153 if (!err) {
2154 if (offp)
2155 *offp = io_off;
2156 if (lenp)
2157 *lenp = io_len;
2158 tmp_putpagecnt++;
2159 tmp_pagespushed += btop(io_len);
2160 }
2161 if (err && err != ENOMEM && err != SE_NOSWAP)
2162 cmn_err(CE_WARN, "tmp_putapage: err %d\n", err);
2163 return (err);
2164 }
2165
2166 /* ARGSUSED */
2167 static int
2168 tmp_map(
2169 struct vnode *vp,
2170 offset_t off,
2171 struct as *as,
2172 caddr_t *addrp,
2173 size_t len,
2174 uchar_t prot,
2175 uchar_t maxprot,
2176 uint_t flags,
2177 struct cred *cred,
2178 caller_context_t *ct)
2179 {
2180 struct segvn_crargs vn_a;
2181 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
2182 int error;
2183
2184 #ifdef _ILP32
2185 if (len > MAXOFF_T)
2186 return (ENOMEM);
2187 #endif
2188
2189 if (vp->v_flag & VNOMAP)
2190 return (ENOSYS);
2191
2192 if (off < 0 || (offset_t)(off + len) < 0 ||
2193 off > MAXOFF_T || (off + len) > MAXOFF_T)
2194 return (ENXIO);
2195
2196 if (vp->v_type != VREG)
2197 return (ENODEV);
2198
2199 /*
2200 * Don't allow mapping to locked file
2201 */
2202 if (vn_has_mandatory_locks(vp, tp->tn_mode)) {
2203 return (EAGAIN);
2204 }
2205
2206 as_rangelock(as);
2207 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
2208 if (error != 0) {
2209 as_rangeunlock(as);
2210 return (error);
2211 }
2212
2213 vn_a.vp = vp;
2214 vn_a.offset = (u_offset_t)off;
2215 vn_a.type = flags & MAP_TYPE;
2216 vn_a.prot = prot;
2217 vn_a.maxprot = maxprot;
2218 vn_a.flags = flags & ~MAP_TYPE;
2219 vn_a.cred = cred;
2220 vn_a.amp = NULL;
2221 vn_a.szc = 0;
2222 vn_a.lgrp_mem_policy_flags = 0;
2223
2224 error = as_map(as, *addrp, len, segvn_create, &vn_a);
2225 as_rangeunlock(as);
2226 return (error);
2227 }
2228
2229 /*
2230 * tmp_addmap and tmp_delmap can't be called since the vp
2231 * maintained in the segvn mapping is NULL.
2232 */
2233 /* ARGSUSED */
2234 static int
2235 tmp_addmap(
2236 struct vnode *vp,
2237 offset_t off,
2238 struct as *as,
2239 caddr_t addr,
2240 size_t len,
2241 uchar_t prot,
2242 uchar_t maxprot,
2243 uint_t flags,
2244 struct cred *cred,
2245 caller_context_t *ct)
2246 {
2247 return (0);
2248 }
2249
2250 /* ARGSUSED */
2251 static int
2252 tmp_delmap(
2253 struct vnode *vp,
2254 offset_t off,
2255 struct as *as,
2256 caddr_t addr,
2257 size_t len,
2258 uint_t prot,
2259 uint_t maxprot,
2260 uint_t flags,
2261 struct cred *cred,
2262 caller_context_t *ct)
2263 {
2264 return (0);
2265 }
2266
2267 static int
2268 tmp_freesp(struct vnode *vp, struct flock64 *lp, int flag)
2269 {
2270 register int i;
2271 register struct tmpnode *tp = VTOTN(vp);
2272 int error;
2273
2274 ASSERT(vp->v_type == VREG);
2275 ASSERT(lp->l_start >= 0);
2276
2277 if (lp->l_len != 0)
2278 return (EINVAL);
2279
2280 rw_enter(&tp->tn_rwlock, RW_WRITER);
2281 if (tp->tn_size == lp->l_start) {
2282 rw_exit(&tp->tn_rwlock);
2283 return (0);
2284 }
2285
2286 /*
2287 * Check for any mandatory locks on the range
2288 */
2289 if (MANDLOCK(vp, tp->tn_mode)) {
2290 long save_start;
2291
2292 save_start = lp->l_start;
2293
2294 if (tp->tn_size < lp->l_start) {
2295 /*
2296 * "Truncate up" case: need to make sure there
2297 * is no lock beyond current end-of-file. To
2298 * do so, we need to set l_start to the size
2299 * of the file temporarily.
2300 */
2301 lp->l_start = tp->tn_size;
2302 }
2303 lp->l_type = F_WRLCK;
2304 lp->l_sysid = 0;
2305 lp->l_pid = ttoproc(curthread)->p_pid;
2306 i = (flag & (FNDELAY|FNONBLOCK)) ? 0 : SLPFLCK;
2307 if ((i = reclock(vp, lp, i, 0, lp->l_start, NULL)) != 0 ||
2308 lp->l_type != F_UNLCK) {
2309 rw_exit(&tp->tn_rwlock);
2310 return (i ? i : EAGAIN);
2311 }
2312
2313 lp->l_start = save_start;
2314 }
2315 VFSTOTM(vp->v_vfsp);
2316
2317 rw_enter(&tp->tn_contents, RW_WRITER);
2318 error = tmpnode_trunc((struct tmount *)VFSTOTM(vp->v_vfsp),
2319 tp, (ulong_t)lp->l_start);
2320 rw_exit(&tp->tn_contents);
2321 rw_exit(&tp->tn_rwlock);
2322 return (error);
2323 }
2324
2325 /* ARGSUSED */
2326 static int
2327 tmp_space(
2328 struct vnode *vp,
2329 int cmd,
2330 struct flock64 *bfp,
2331 int flag,
2332 offset_t offset,
2333 cred_t *cred,
2334 caller_context_t *ct)
2335 {
2336 int error;
2337
2338 if (cmd != F_FREESP)
2339 return (EINVAL);
2340 if ((error = convoff(vp, bfp, 0, (offset_t)offset)) == 0) {
2341 if ((bfp->l_start > MAXOFF_T) || (bfp->l_len > MAXOFF_T))
2342 return (EFBIG);
2343 error = tmp_freesp(vp, bfp, flag);
2344
2345 if (error == 0 && bfp->l_start == 0)
2346 vnevent_truncate(vp, ct);
2347 }
2348 return (error);
2349 }
2350
2351 /* ARGSUSED */
2352 static int
2353 tmp_seek(
2354 struct vnode *vp,
2355 offset_t ooff,
2356 offset_t *noffp,
2357 caller_context_t *ct)
2358 {
2359 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
2360 }
2361
2362 /* ARGSUSED2 */
2363 static int
2364 tmp_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2365 {
2366 struct tmpnode *tp = VTOTN(vp);
2367
2368 if (write_lock) {
2369 rw_enter(&tp->tn_rwlock, RW_WRITER);
2370 } else {
2371 rw_enter(&tp->tn_rwlock, RW_READER);
2372 }
2373 return (write_lock);
2374 }
2375
2376 /* ARGSUSED1 */
2377 static void
2378 tmp_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2379 {
2380 struct tmpnode *tp = VTOTN(vp);
2381
2382 rw_exit(&tp->tn_rwlock);
2383 }
2384
2385 static int
2386 tmp_pathconf(
2387 struct vnode *vp,
2388 int cmd,
2389 ulong_t *valp,
2390 cred_t *cr,
2391 caller_context_t *ct)
2392 {
2393 struct tmpnode *tp = NULL;
2394 int error;
2395
2396 switch (cmd) {
2397 case _PC_XATTR_EXISTS:
2398 if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
2399 *valp = 0; /* assume no attributes */
2400 error = 0; /* okay to ask */
2401 tp = VTOTN(vp);
2402 rw_enter(&tp->tn_rwlock, RW_READER);
2403 if (tp->tn_xattrdp) {
2404 rw_enter(&tp->tn_xattrdp->tn_rwlock, RW_READER);
2405 /* do not count "." and ".." */
2406 if (tp->tn_xattrdp->tn_dirents > 2)
2407 *valp = 1;
2408 rw_exit(&tp->tn_xattrdp->tn_rwlock);
2409 }
2410 rw_exit(&tp->tn_rwlock);
2411 } else {
2412 error = EINVAL;
2413 }
2414 break;
2415 case _PC_SATTR_ENABLED:
2416 case _PC_SATTR_EXISTS:
2417 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2418 (vp->v_type == VREG || vp->v_type == VDIR);
2419 error = 0;
2420 break;
2421 case _PC_TIMESTAMP_RESOLUTION:
2422 /* nanosecond timestamp resolution */
2423 *valp = 1L;
2424 error = 0;
2425 break;
2426 default:
2427 error = fs_pathconf(vp, cmd, valp, cr, ct);
2428 }
2429 return (error);
2430 }
2431
2432
2433 struct vnodeops *tmp_vnodeops;
2434
2435 const fs_operation_def_t tmp_vnodeops_template[] = {
2436 VOPNAME_OPEN, { .vop_open = tmp_open },
2437 VOPNAME_CLOSE, { .vop_close = tmp_close },
2438 VOPNAME_READ, { .vop_read = tmp_read },
2439 VOPNAME_WRITE, { .vop_write = tmp_write },
2440 VOPNAME_IOCTL, { .vop_ioctl = tmp_ioctl },
2441 VOPNAME_GETATTR, { .vop_getattr = tmp_getattr },
2442 VOPNAME_SETATTR, { .vop_setattr = tmp_setattr },
2443 VOPNAME_ACCESS, { .vop_access = tmp_access },
2444 VOPNAME_LOOKUP, { .vop_lookup = tmp_lookup },
2445 VOPNAME_CREATE, { .vop_create = tmp_create },
2446 VOPNAME_REMOVE, { .vop_remove = tmp_remove },
2447 VOPNAME_LINK, { .vop_link = tmp_link },
2448 VOPNAME_RENAME, { .vop_rename = tmp_rename },
2449 VOPNAME_MKDIR, { .vop_mkdir = tmp_mkdir },
2450 VOPNAME_RMDIR, { .vop_rmdir = tmp_rmdir },
2451 VOPNAME_READDIR, { .vop_readdir = tmp_readdir },
2452 VOPNAME_SYMLINK, { .vop_symlink = tmp_symlink },
2453 VOPNAME_READLINK, { .vop_readlink = tmp_readlink },
2454 VOPNAME_FSYNC, { .vop_fsync = tmp_fsync },
2455 VOPNAME_INACTIVE, { .vop_inactive = tmp_inactive },
2456 VOPNAME_FID, { .vop_fid = tmp_fid },
2457 VOPNAME_RWLOCK, { .vop_rwlock = tmp_rwlock },
2458 VOPNAME_RWUNLOCK, { .vop_rwunlock = tmp_rwunlock },
2459 VOPNAME_SEEK, { .vop_seek = tmp_seek },
2460 VOPNAME_SPACE, { .vop_space = tmp_space },
2461 VOPNAME_GETPAGE, { .vop_getpage = tmp_getpage },
2462 VOPNAME_PUTPAGE, { .vop_putpage = tmp_putpage },
2463 VOPNAME_MAP, { .vop_map = tmp_map },
2464 VOPNAME_ADDMAP, { .vop_addmap = tmp_addmap },
2465 VOPNAME_DELMAP, { .vop_delmap = tmp_delmap },
2466 VOPNAME_PATHCONF, { .vop_pathconf = tmp_pathconf },
2467 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
2468 NULL, NULL
2469 };