Print this page
OS-5148 ftruncate at offset should emit proper events
Reviewed by: Bryan Cantrill <bryan@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
OS-5291 lxbrand inotify02 LTP regression
Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
OS-3294 add support for inotify
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Robert Mustacchi <rm@joyent.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/tmpfs/tmp_vnops.c
+++ new/usr/src/uts/common/fs/tmpfs/tmp_vnops.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
|
↓ open down ↓ |
17 lines elided |
↑ open up ↑ |
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
24 24 * Use is subject to license terms.
25 25 */
26 26
27 27 /*
28 - * Copyright (c) 2015, Joyent, Inc. All rights reserved.
28 + * Copyright 2016, Joyent, Inc.
29 29 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
30 30 * Copyright 2016 RackTop Systems.
31 31 */
32 32
33 33 #include <sys/types.h>
34 34 #include <sys/param.h>
35 35 #include <sys/t_lock.h>
36 36 #include <sys/systm.h>
37 37 #include <sys/sysmacros.h>
38 38 #include <sys/user.h>
39 39 #include <sys/time.h>
40 40 #include <sys/vfs.h>
41 41 #include <sys/vfs_opreg.h>
42 42 #include <sys/vnode.h>
43 43 #include <sys/file.h>
44 44 #include <sys/fcntl.h>
45 45 #include <sys/flock.h>
46 46 #include <sys/kmem.h>
47 47 #include <sys/uio.h>
48 48 #include <sys/errno.h>
49 49 #include <sys/stat.h>
50 50 #include <sys/cred.h>
51 51 #include <sys/dirent.h>
52 52 #include <sys/pathname.h>
53 53 #include <sys/vmsystm.h>
54 54 #include <sys/fs/tmp.h>
55 55 #include <sys/fs/tmpnode.h>
56 56 #include <sys/mman.h>
57 57 #include <vm/hat.h>
58 58 #include <vm/seg_vn.h>
59 59 #include <vm/seg_map.h>
60 60 #include <vm/seg.h>
61 61 #include <vm/anon.h>
62 62 #include <vm/as.h>
63 63 #include <vm/page.h>
64 64 #include <vm/pvn.h>
65 65 #include <sys/cmn_err.h>
66 66 #include <sys/debug.h>
67 67 #include <sys/swap.h>
68 68 #include <sys/buf.h>
69 69 #include <sys/vm.h>
70 70 #include <sys/vtrace.h>
71 71 #include <sys/policy.h>
72 72 #include <fs/fs_subr.h>
73 73
74 74 static int tmp_getapage(struct vnode *, u_offset_t, size_t, uint_t *,
75 75 page_t **, size_t, struct seg *, caddr_t, enum seg_rw, struct cred *);
76 76 static int tmp_putapage(struct vnode *, page_t *, u_offset_t *, size_t *,
77 77 int, struct cred *);
78 78
79 79 /* ARGSUSED1 */
80 80 static int
81 81 tmp_open(struct vnode **vpp, int flag, struct cred *cred, caller_context_t *ct)
82 82 {
83 83 /*
84 84 * swapon to a tmpfs file is not supported so access
85 85 * is denied on open if VISSWAP is set.
86 86 */
87 87 if ((*vpp)->v_flag & VISSWAP)
88 88 return (EINVAL);
89 89 return (0);
90 90 }
91 91
92 92 /* ARGSUSED1 */
93 93 static int
94 94 tmp_close(
95 95 struct vnode *vp,
96 96 int flag,
97 97 int count,
98 98 offset_t offset,
99 99 struct cred *cred,
100 100 caller_context_t *ct)
101 101 {
102 102 cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
103 103 cleanshares(vp, ttoproc(curthread)->p_pid);
104 104 return (0);
105 105 }
106 106
107 107 /*
108 108 * wrtmp does the real work of write requests for tmpfs.
109 109 */
110 110 static int
111 111 wrtmp(
112 112 struct tmount *tm,
113 113 struct tmpnode *tp,
114 114 struct uio *uio,
115 115 struct cred *cr,
116 116 struct caller_context *ct)
117 117 {
118 118 pgcnt_t pageoffset; /* offset in pages */
119 119 ulong_t segmap_offset; /* pagesize byte offset into segmap */
120 120 caddr_t base; /* base of segmap */
121 121 ssize_t bytes; /* bytes to uiomove */
122 122 pfn_t pagenumber; /* offset in pages into tmp file */
123 123 struct vnode *vp;
124 124 int error = 0;
125 125 int pagecreate; /* == 1 if we allocated a page */
126 126 int newpage;
127 127 rlim64_t limit = uio->uio_llimit;
128 128 long oresid = uio->uio_resid;
129 129 timestruc_t now;
130 130
131 131 long tn_size_changed = 0;
132 132 long old_tn_size;
133 133 long new_tn_size;
134 134
135 135 vp = TNTOV(tp);
136 136 ASSERT(vp->v_type == VREG);
137 137
138 138 TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START,
139 139 "tmp_wrtmp_start:vp %p", vp);
140 140
141 141 ASSERT(RW_WRITE_HELD(&tp->tn_contents));
142 142 ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
143 143
144 144 if (MANDLOCK(vp, tp->tn_mode)) {
145 145 rw_exit(&tp->tn_contents);
146 146 /*
147 147 * tmp_getattr ends up being called by chklock
148 148 */
149 149 error = chklock(vp, FWRITE, uio->uio_loffset, uio->uio_resid,
150 150 uio->uio_fmode, ct);
151 151 rw_enter(&tp->tn_contents, RW_WRITER);
152 152 if (error != 0) {
153 153 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
154 154 "tmp_wrtmp_end:vp %p error %d", vp, error);
155 155 return (error);
156 156 }
157 157 }
158 158
159 159 if (uio->uio_loffset < 0)
160 160 return (EINVAL);
161 161
162 162 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
163 163 limit = MAXOFFSET_T;
164 164
165 165 if (uio->uio_loffset >= limit) {
166 166 proc_t *p = ttoproc(curthread);
167 167
168 168 mutex_enter(&p->p_lock);
169 169 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
170 170 p, RCA_UNSAFE_SIGINFO);
171 171 mutex_exit(&p->p_lock);
172 172 return (EFBIG);
173 173 }
174 174
175 175 if (uio->uio_loffset >= MAXOFF_T) {
176 176 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
177 177 "tmp_wrtmp_end:vp %p error %d", vp, EINVAL);
178 178 return (EFBIG);
179 179 }
180 180
181 181 if (uio->uio_resid == 0) {
182 182 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
183 183 "tmp_wrtmp_end:vp %p error %d", vp, 0);
184 184 return (0);
185 185 }
186 186
187 187 if (limit > MAXOFF_T)
188 188 limit = MAXOFF_T;
189 189
190 190 do {
191 191 long offset;
192 192 long delta;
193 193
194 194 offset = (long)uio->uio_offset;
195 195 pageoffset = offset & PAGEOFFSET;
196 196 /*
197 197 * A maximum of PAGESIZE bytes of data is transferred
198 198 * each pass through this loop
199 199 */
200 200 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
201 201
202 202 if (offset + bytes >= limit) {
203 203 if (offset >= limit) {
204 204 error = EFBIG;
205 205 goto out;
206 206 }
207 207 bytes = limit - offset;
208 208 }
209 209 pagenumber = btop(offset);
210 210
211 211 /*
212 212 * delta is the amount of anonymous memory
213 213 * to reserve for the file.
214 214 * We always reserve in pagesize increments so
215 215 * unless we're extending the file into a new page,
216 216 * we don't need to call tmp_resv.
217 217 */
218 218 delta = offset + bytes -
219 219 P2ROUNDUP_TYPED(tp->tn_size, PAGESIZE, u_offset_t);
220 220 if (delta > 0) {
221 221 pagecreate = 1;
222 222 if (tmp_resv(tm, tp, delta, pagecreate)) {
223 223 /*
224 224 * Log file system full in the zone that owns
225 225 * the tmpfs mount, as well as in the global
226 226 * zone if necessary.
227 227 */
228 228 zcmn_err(tm->tm_vfsp->vfs_zone->zone_id,
229 229 CE_WARN, "%s: File system full, "
230 230 "swap space limit exceeded",
231 231 tm->tm_mntpath);
232 232
233 233 if (tm->tm_vfsp->vfs_zone->zone_id !=
234 234 GLOBAL_ZONEID) {
235 235
236 236 vfs_t *vfs = tm->tm_vfsp;
237 237
238 238 zcmn_err(GLOBAL_ZONEID,
239 239 CE_WARN, "%s: File system full, "
240 240 "swap space limit exceeded",
241 241 vfs->vfs_vnodecovered->v_path);
242 242 }
243 243 error = ENOSPC;
244 244 break;
245 245 }
246 246 tmpnode_growmap(tp, (ulong_t)offset + bytes);
247 247 }
248 248 /* grow the file to the new length */
249 249 if (offset + bytes > tp->tn_size) {
250 250 tn_size_changed = 1;
251 251 old_tn_size = tp->tn_size;
252 252 /*
253 253 * Postpone updating tp->tn_size until uiomove() is
254 254 * done.
255 255 */
256 256 new_tn_size = offset + bytes;
257 257 }
258 258 if (bytes == PAGESIZE) {
259 259 /*
260 260 * Writing whole page so reading from disk
261 261 * is a waste
262 262 */
263 263 pagecreate = 1;
264 264 } else {
265 265 pagecreate = 0;
266 266 }
267 267 /*
268 268 * If writing past EOF or filling in a hole
269 269 * we need to allocate an anon slot.
270 270 */
271 271 if (anon_get_ptr(tp->tn_anon, pagenumber) == NULL) {
272 272 (void) anon_set_ptr(tp->tn_anon, pagenumber,
273 273 anon_alloc(vp, ptob(pagenumber)), ANON_SLEEP);
274 274 pagecreate = 1;
275 275 tp->tn_nblocks++;
276 276 }
277 277
278 278 /*
279 279 * We have to drop the contents lock to allow the VM
280 280 * system to reacquire it in tmp_getpage()
281 281 */
282 282 rw_exit(&tp->tn_contents);
283 283
284 284 /*
285 285 * Touch the page and fault it in if it is not in core
286 286 * before segmap_getmapflt or vpm_data_copy can lock it.
287 287 * This is to avoid the deadlock if the buffer is mapped
288 288 * to the same file through mmap which we want to write.
289 289 */
290 290 uio_prefaultpages((long)bytes, uio);
291 291
292 292 newpage = 0;
293 293 if (vpm_enable) {
294 294 /*
295 295 * Copy data. If new pages are created, part of
296 296 * the page that is not written will be initizliazed
297 297 * with zeros.
298 298 */
299 299 error = vpm_data_copy(vp, offset, bytes, uio,
300 300 !pagecreate, &newpage, 1, S_WRITE);
301 301 } else {
302 302 /* Get offset within the segmap mapping */
303 303 segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
304 304 base = segmap_getmapflt(segkmap, vp,
305 305 (offset & MAXBMASK), PAGESIZE, !pagecreate,
306 306 S_WRITE);
307 307 }
308 308
309 309
310 310 if (!vpm_enable && pagecreate) {
311 311 /*
312 312 * segmap_pagecreate() returns 1 if it calls
313 313 * page_create_va() to allocate any pages.
314 314 */
315 315 newpage = segmap_pagecreate(segkmap,
316 316 base + segmap_offset, (size_t)PAGESIZE, 0);
317 317 /*
318 318 * Clear from the beginning of the page to the starting
319 319 * offset of the data.
320 320 */
321 321 if (pageoffset != 0)
322 322 (void) kzero(base + segmap_offset,
323 323 (size_t)pageoffset);
324 324 }
325 325
326 326 if (!vpm_enable) {
327 327 error = uiomove(base + segmap_offset + pageoffset,
328 328 (long)bytes, UIO_WRITE, uio);
329 329 }
330 330
331 331 if (!vpm_enable && pagecreate &&
332 332 uio->uio_offset < P2ROUNDUP(offset + bytes, PAGESIZE)) {
333 333 long zoffset; /* zero from offset into page */
334 334 /*
335 335 * We created pages w/o initializing them completely,
336 336 * thus we need to zero the part that wasn't set up.
337 337 * This happens on most EOF write cases and if
338 338 * we had some sort of error during the uiomove.
339 339 */
340 340 long nmoved;
341 341
342 342 nmoved = uio->uio_offset - offset;
343 343 ASSERT((nmoved + pageoffset) <= PAGESIZE);
344 344
345 345 /*
346 346 * Zero from the end of data in the page to the
347 347 * end of the page.
348 348 */
349 349 if ((zoffset = pageoffset + nmoved) < PAGESIZE)
350 350 (void) kzero(base + segmap_offset + zoffset,
351 351 (size_t)PAGESIZE - zoffset);
352 352 }
353 353
354 354 /*
355 355 * Unlock the pages which have been allocated by
356 356 * page_create_va() in segmap_pagecreate()
357 357 */
358 358 if (!vpm_enable && newpage) {
359 359 segmap_pageunlock(segkmap, base + segmap_offset,
360 360 (size_t)PAGESIZE, S_WRITE);
361 361 }
362 362
363 363 if (error) {
364 364 /*
365 365 * If we failed on a write, we must
366 366 * be sure to invalidate any pages that may have
367 367 * been allocated.
368 368 */
369 369 if (vpm_enable) {
370 370 (void) vpm_sync_pages(vp, offset, PAGESIZE,
371 371 SM_INVAL);
372 372 } else {
373 373 (void) segmap_release(segkmap, base, SM_INVAL);
374 374 }
375 375 } else {
376 376 if (vpm_enable) {
377 377 error = vpm_sync_pages(vp, offset, PAGESIZE,
378 378 0);
379 379 } else {
380 380 error = segmap_release(segkmap, base, 0);
381 381 }
382 382 }
383 383
384 384 /*
385 385 * Re-acquire contents lock.
386 386 */
387 387 rw_enter(&tp->tn_contents, RW_WRITER);
388 388
389 389 /*
390 390 * Update tn_size.
391 391 */
392 392 if (tn_size_changed)
393 393 tp->tn_size = new_tn_size;
394 394
395 395 /*
396 396 * If the uiomove failed, fix up tn_size.
397 397 */
398 398 if (error) {
399 399 if (tn_size_changed) {
400 400 /*
401 401 * The uiomove failed, and we
402 402 * allocated blocks,so get rid
403 403 * of them.
404 404 */
405 405 (void) tmpnode_trunc(tm, tp,
406 406 (ulong_t)old_tn_size);
407 407 }
408 408 } else {
409 409 /*
410 410 * XXX - Can this be out of the loop?
411 411 */
412 412 if ((tp->tn_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) &&
413 413 (tp->tn_mode & (S_ISUID | S_ISGID)) &&
414 414 secpolicy_vnode_setid_retain(cr,
415 415 (tp->tn_mode & S_ISUID) != 0 && tp->tn_uid == 0)) {
416 416 /*
417 417 * Clear Set-UID & Set-GID bits on
418 418 * successful write if not privileged
419 419 * and at least one of the execute bits
420 420 * is set. If we always clear Set-GID,
421 421 * mandatory file and record locking is
422 422 * unuseable.
423 423 */
424 424 tp->tn_mode &= ~(S_ISUID | S_ISGID);
425 425 }
426 426 gethrestime(&now);
427 427 tp->tn_mtime = now;
428 428 tp->tn_ctime = now;
429 429 }
430 430 } while (error == 0 && uio->uio_resid > 0 && bytes != 0);
431 431
432 432 out:
433 433 /*
434 434 * If we've already done a partial-write, terminate
435 435 * the write but return no error.
436 436 */
437 437 if (oresid != uio->uio_resid)
438 438 error = 0;
439 439 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
440 440 "tmp_wrtmp_end:vp %p error %d", vp, error);
441 441 return (error);
442 442 }
443 443
444 444 /*
445 445 * rdtmp does the real work of read requests for tmpfs.
446 446 */
447 447 static int
448 448 rdtmp(
449 449 struct tmount *tm,
450 450 struct tmpnode *tp,
451 451 struct uio *uio,
452 452 struct caller_context *ct)
453 453 {
454 454 ulong_t pageoffset; /* offset in tmpfs file (uio_offset) */
455 455 ulong_t segmap_offset; /* pagesize byte offset into segmap */
456 456 caddr_t base; /* base of segmap */
457 457 ssize_t bytes; /* bytes to uiomove */
458 458 struct vnode *vp;
459 459 int error;
460 460 long oresid = uio->uio_resid;
461 461
462 462 #if defined(lint)
463 463 tm = tm;
464 464 #endif
465 465 vp = TNTOV(tp);
466 466
467 467 TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START, "tmp_rdtmp_start:vp %p",
468 468 vp);
469 469
470 470 ASSERT(RW_LOCK_HELD(&tp->tn_contents));
471 471
472 472 if (MANDLOCK(vp, tp->tn_mode)) {
473 473 rw_exit(&tp->tn_contents);
474 474 /*
475 475 * tmp_getattr ends up being called by chklock
476 476 */
477 477 error = chklock(vp, FREAD, uio->uio_loffset, uio->uio_resid,
478 478 uio->uio_fmode, ct);
479 479 rw_enter(&tp->tn_contents, RW_READER);
480 480 if (error != 0) {
481 481 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
482 482 "tmp_rdtmp_end:vp %p error %d", vp, error);
483 483 return (error);
484 484 }
485 485 }
486 486 ASSERT(tp->tn_type == VREG);
487 487
488 488 if (uio->uio_loffset >= MAXOFF_T) {
489 489 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
490 490 "tmp_rdtmp_end:vp %p error %d", vp, EINVAL);
491 491 return (0);
492 492 }
493 493 if (uio->uio_loffset < 0)
494 494 return (EINVAL);
495 495 if (uio->uio_resid == 0) {
496 496 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
497 497 "tmp_rdtmp_end:vp %p error %d", vp, 0);
498 498 return (0);
499 499 }
500 500
501 501 vp = TNTOV(tp);
502 502
503 503 do {
504 504 long diff;
505 505 long offset;
506 506
507 507 offset = uio->uio_offset;
508 508 pageoffset = offset & PAGEOFFSET;
509 509 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
510 510
511 511 diff = tp->tn_size - offset;
512 512
513 513 if (diff <= 0) {
514 514 error = 0;
515 515 goto out;
516 516 }
517 517 if (diff < bytes)
518 518 bytes = diff;
519 519
520 520 /*
521 521 * We have to drop the contents lock to allow the VM system
522 522 * to reacquire it in tmp_getpage() should the uiomove cause a
523 523 * pagefault.
524 524 */
525 525 rw_exit(&tp->tn_contents);
526 526
527 527 if (vpm_enable) {
528 528 /*
529 529 * Copy data.
530 530 */
531 531 error = vpm_data_copy(vp, offset, bytes, uio, 1, NULL,
532 532 0, S_READ);
533 533 } else {
534 534 segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
535 535 base = segmap_getmapflt(segkmap, vp, offset & MAXBMASK,
536 536 bytes, 1, S_READ);
537 537
538 538 error = uiomove(base + segmap_offset + pageoffset,
539 539 (long)bytes, UIO_READ, uio);
540 540 }
541 541
542 542 if (error) {
543 543 if (vpm_enable) {
544 544 (void) vpm_sync_pages(vp, offset, PAGESIZE, 0);
545 545 } else {
546 546 (void) segmap_release(segkmap, base, 0);
547 547 }
548 548 } else {
549 549 if (vpm_enable) {
550 550 error = vpm_sync_pages(vp, offset, PAGESIZE,
551 551 0);
552 552 } else {
553 553 error = segmap_release(segkmap, base, 0);
554 554 }
555 555 }
556 556
557 557 /*
558 558 * Re-acquire contents lock.
559 559 */
560 560 rw_enter(&tp->tn_contents, RW_READER);
561 561
562 562 } while (error == 0 && uio->uio_resid > 0);
563 563
564 564 out:
565 565 gethrestime(&tp->tn_atime);
566 566
567 567 /*
568 568 * If we've already done a partial read, terminate
569 569 * the read but return no error.
570 570 */
571 571 if (oresid != uio->uio_resid)
572 572 error = 0;
573 573
574 574 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
575 575 "tmp_rdtmp_end:vp %x error %d", vp, error);
576 576 return (error);
577 577 }
578 578
579 579 /* ARGSUSED2 */
580 580 static int
581 581 tmp_read(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cred,
582 582 struct caller_context *ct)
583 583 {
584 584 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
585 585 struct tmount *tm = (struct tmount *)VTOTM(vp);
586 586 int error;
587 587
588 588 /*
589 589 * We don't currently support reading non-regular files
590 590 */
591 591 if (vp->v_type == VDIR)
592 592 return (EISDIR);
593 593 if (vp->v_type != VREG)
594 594 return (EINVAL);
595 595 /*
596 596 * tmp_rwlock should have already been called from layers above
597 597 */
598 598 ASSERT(RW_READ_HELD(&tp->tn_rwlock));
599 599
600 600 rw_enter(&tp->tn_contents, RW_READER);
601 601
602 602 error = rdtmp(tm, tp, uiop, ct);
603 603
604 604 rw_exit(&tp->tn_contents);
605 605
606 606 return (error);
607 607 }
608 608
609 609 static int
610 610 tmp_write(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cred,
611 611 struct caller_context *ct)
612 612 {
613 613 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
614 614 struct tmount *tm = (struct tmount *)VTOTM(vp);
615 615 int error;
616 616
617 617 /*
618 618 * We don't currently support writing to non-regular files
619 619 */
620 620 if (vp->v_type != VREG)
621 621 return (EINVAL); /* XXX EISDIR? */
622 622
623 623 /*
624 624 * tmp_rwlock should have already been called from layers above
625 625 */
626 626 ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
627 627
628 628 rw_enter(&tp->tn_contents, RW_WRITER);
629 629
630 630 if (ioflag & FAPPEND) {
631 631 /*
632 632 * In append mode start at end of file.
633 633 */
634 634 uiop->uio_loffset = tp->tn_size;
635 635 }
636 636
637 637 error = wrtmp(tm, tp, uiop, cred, ct);
638 638
639 639 rw_exit(&tp->tn_contents);
640 640
641 641 return (error);
642 642 }
643 643
644 644 /* ARGSUSED */
645 645 static int
646 646 tmp_ioctl(
647 647 struct vnode *vp,
648 648 int com,
649 649 intptr_t data,
650 650 int flag,
651 651 struct cred *cred,
652 652 int *rvalp,
653 653 caller_context_t *ct)
654 654 {
655 655 return (ENOTTY);
656 656 }
657 657
658 658 /* ARGSUSED2 */
659 659 static int
660 660 tmp_getattr(
661 661 struct vnode *vp,
662 662 struct vattr *vap,
663 663 int flags,
664 664 struct cred *cred,
665 665 caller_context_t *ct)
666 666 {
667 667 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
668 668 struct vnode *mvp;
669 669 struct vattr va;
670 670 int attrs = 1;
671 671
672 672 /*
673 673 * A special case to handle the root tnode on a diskless nfs
674 674 * client who may have had its uid and gid inherited
675 675 * from an nfs vnode with nobody ownership. Likely the
676 676 * root filesystem. After nfs is fully functional the uid/gid
677 677 * may be mapable so ask again.
678 678 * vfsp can't get unmounted because we hold vp.
679 679 */
680 680 if (vp->v_flag & VROOT &&
681 681 (mvp = vp->v_vfsp->vfs_vnodecovered) != NULL) {
682 682 mutex_enter(&tp->tn_tlock);
683 683 if (tp->tn_uid == UID_NOBODY || tp->tn_gid == GID_NOBODY) {
684 684 mutex_exit(&tp->tn_tlock);
685 685 bzero(&va, sizeof (struct vattr));
686 686 va.va_mask = AT_UID|AT_GID;
687 687 attrs = VOP_GETATTR(mvp, &va, 0, cred, ct);
688 688 } else {
689 689 mutex_exit(&tp->tn_tlock);
690 690 }
691 691 }
692 692 mutex_enter(&tp->tn_tlock);
693 693 if (attrs == 0) {
694 694 tp->tn_uid = va.va_uid;
695 695 tp->tn_gid = va.va_gid;
696 696 }
697 697 vap->va_type = vp->v_type;
698 698 vap->va_mode = tp->tn_mode & MODEMASK;
699 699 vap->va_uid = tp->tn_uid;
700 700 vap->va_gid = tp->tn_gid;
701 701 vap->va_fsid = tp->tn_fsid;
702 702 vap->va_nodeid = (ino64_t)tp->tn_nodeid;
703 703 vap->va_nlink = tp->tn_nlink;
704 704 vap->va_size = (u_offset_t)tp->tn_size;
705 705 vap->va_atime = tp->tn_atime;
706 706 vap->va_mtime = tp->tn_mtime;
707 707 vap->va_ctime = tp->tn_ctime;
708 708 vap->va_blksize = PAGESIZE;
709 709 vap->va_rdev = tp->tn_rdev;
710 710 vap->va_seq = tp->tn_seq;
711 711
712 712 /*
713 713 * XXX Holes are not taken into account. We could take the time to
714 714 * run through the anon array looking for allocated slots...
715 715 */
716 716 vap->va_nblocks = (fsblkcnt64_t)btodb(ptob(btopr(vap->va_size)));
717 717 mutex_exit(&tp->tn_tlock);
718 718 return (0);
719 719 }
720 720
721 721 /*ARGSUSED4*/
722 722 static int
723 723 tmp_setattr(
724 724 struct vnode *vp,
725 725 struct vattr *vap,
726 726 int flags,
727 727 struct cred *cred,
728 728 caller_context_t *ct)
729 729 {
730 730 struct tmount *tm = (struct tmount *)VTOTM(vp);
731 731 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
732 732 int error = 0;
733 733 struct vattr *get;
734 734 long mask;
735 735
736 736 /*
737 737 * Cannot set these attributes
738 738 */
739 739 if ((vap->va_mask & AT_NOSET) || (vap->va_mask & AT_XVATTR))
740 740 return (EINVAL);
741 741
742 742 mutex_enter(&tp->tn_tlock);
743 743
744 744 get = &tp->tn_attr;
745 745 /*
746 746 * Change file access modes. Must be owner or have sufficient
747 747 * privileges.
748 748 */
749 749 error = secpolicy_vnode_setattr(cred, vp, vap, get, flags, tmp_taccess,
750 750 tp);
751 751
752 752 if (error)
753 753 goto out;
754 754
755 755 mask = vap->va_mask;
756 756
757 757 if (mask & AT_MODE) {
758 758 get->va_mode &= S_IFMT;
759 759 get->va_mode |= vap->va_mode & ~S_IFMT;
760 760 }
761 761
762 762 if (mask & AT_UID)
763 763 get->va_uid = vap->va_uid;
764 764 if (mask & AT_GID)
765 765 get->va_gid = vap->va_gid;
766 766 if (mask & AT_ATIME)
767 767 get->va_atime = vap->va_atime;
768 768 if (mask & AT_MTIME)
769 769 get->va_mtime = vap->va_mtime;
770 770
771 771 if (mask & (AT_UID | AT_GID | AT_MODE | AT_MTIME))
772 772 gethrestime(&tp->tn_ctime);
773 773
774 774 if (mask & AT_SIZE) {
775 775 ASSERT(vp->v_type != VDIR);
776 776
777 777 /* Don't support large files. */
778 778 if (vap->va_size > MAXOFF_T) {
779 779 error = EFBIG;
|
↓ open down ↓ |
741 lines elided |
↑ open up ↑ |
780 780 goto out;
781 781 }
782 782 mutex_exit(&tp->tn_tlock);
783 783
784 784 rw_enter(&tp->tn_rwlock, RW_WRITER);
785 785 rw_enter(&tp->tn_contents, RW_WRITER);
786 786 error = tmpnode_trunc(tm, tp, (ulong_t)vap->va_size);
787 787 rw_exit(&tp->tn_contents);
788 788 rw_exit(&tp->tn_rwlock);
789 789
790 - if (error == 0 && vap->va_size == 0)
791 - vnevent_truncate(vp, ct);
790 + if (error == 0) {
791 + if (vap->va_size == 0) {
792 + vnevent_truncate(vp, ct);
793 + } else {
794 + vnevent_resize(vp, ct);
795 + }
796 + }
792 797
793 798 goto out1;
794 799 }
795 800 out:
796 801 mutex_exit(&tp->tn_tlock);
797 802 out1:
798 803 return (error);
799 804 }
800 805
801 806 /* ARGSUSED2 */
802 807 static int
803 808 tmp_access(
804 809 struct vnode *vp,
805 810 int mode,
806 811 int flags,
807 812 struct cred *cred,
808 813 caller_context_t *ct)
809 814 {
810 815 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
811 816 int error;
812 817
813 818 mutex_enter(&tp->tn_tlock);
814 819 error = tmp_taccess(tp, mode, cred);
815 820 mutex_exit(&tp->tn_tlock);
816 821 return (error);
817 822 }
818 823
819 824 /* ARGSUSED3 */
820 825 static int
821 826 tmp_lookup(
822 827 struct vnode *dvp,
823 828 char *nm,
824 829 struct vnode **vpp,
825 830 struct pathname *pnp,
826 831 int flags,
827 832 struct vnode *rdir,
828 833 struct cred *cred,
829 834 caller_context_t *ct,
830 835 int *direntflags,
831 836 pathname_t *realpnp)
832 837 {
833 838 struct tmpnode *tp = (struct tmpnode *)VTOTN(dvp);
834 839 struct tmpnode *ntp = NULL;
835 840 int error;
836 841
837 842
838 843 /* allow cd into @ dir */
839 844 if (flags & LOOKUP_XATTR) {
840 845 struct tmpnode *xdp;
841 846 struct tmount *tm;
842 847
843 848 /*
844 849 * don't allow attributes if not mounted XATTR support
845 850 */
846 851 if (!(dvp->v_vfsp->vfs_flag & VFS_XATTR))
847 852 return (EINVAL);
848 853
849 854 if (tp->tn_flags & ISXATTR)
850 855 /* No attributes on attributes */
851 856 return (EINVAL);
852 857
853 858 rw_enter(&tp->tn_rwlock, RW_WRITER);
854 859 if (tp->tn_xattrdp == NULL) {
855 860 if (!(flags & CREATE_XATTR_DIR)) {
856 861 rw_exit(&tp->tn_rwlock);
857 862 return (ENOENT);
858 863 }
859 864
860 865 /*
861 866 * No attribute directory exists for this
862 867 * node - create the attr dir as a side effect
863 868 * of this lookup.
864 869 */
865 870
866 871 /*
867 872 * Make sure we have adequate permission...
868 873 */
869 874
870 875 if ((error = tmp_taccess(tp, VWRITE, cred)) != 0) {
871 876 rw_exit(&tp->tn_rwlock);
872 877 return (error);
873 878 }
874 879
875 880 xdp = tmp_memalloc(sizeof (struct tmpnode),
876 881 TMP_MUSTHAVE);
877 882 tm = VTOTM(dvp);
878 883 tmpnode_init(tm, xdp, &tp->tn_attr, NULL);
879 884 /*
880 885 * Fix-up fields unique to attribute directories.
881 886 */
882 887 xdp->tn_flags = ISXATTR;
883 888 xdp->tn_type = VDIR;
884 889 if (tp->tn_type == VDIR) {
885 890 xdp->tn_mode = tp->tn_attr.va_mode;
886 891 } else {
887 892 xdp->tn_mode = 0700;
888 893 if (tp->tn_attr.va_mode & 0040)
889 894 xdp->tn_mode |= 0750;
890 895 if (tp->tn_attr.va_mode & 0004)
891 896 xdp->tn_mode |= 0705;
892 897 }
893 898 xdp->tn_vnode->v_type = VDIR;
894 899 xdp->tn_vnode->v_flag |= V_XATTRDIR;
895 900 tdirinit(tp, xdp);
896 901 tp->tn_xattrdp = xdp;
897 902 } else {
898 903 VN_HOLD(tp->tn_xattrdp->tn_vnode);
899 904 }
900 905 *vpp = TNTOV(tp->tn_xattrdp);
901 906 rw_exit(&tp->tn_rwlock);
902 907 return (0);
903 908 }
904 909
905 910 /*
906 911 * Null component name is a synonym for directory being searched.
907 912 */
908 913 if (*nm == '\0') {
909 914 VN_HOLD(dvp);
910 915 *vpp = dvp;
911 916 return (0);
912 917 }
913 918 ASSERT(tp);
914 919
915 920 error = tdirlookup(tp, nm, &ntp, cred);
916 921
917 922 if (error == 0) {
918 923 ASSERT(ntp);
919 924 *vpp = TNTOV(ntp);
920 925 /*
921 926 * If vnode is a device return special vnode instead
922 927 */
923 928 if (IS_DEVVP(*vpp)) {
924 929 struct vnode *newvp;
925 930
926 931 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
927 932 cred);
928 933 VN_RELE(*vpp);
929 934 *vpp = newvp;
930 935 }
931 936 }
932 937 TRACE_4(TR_FAC_TMPFS, TR_TMPFS_LOOKUP,
933 938 "tmpfs lookup:vp %p name %s vpp %p error %d",
934 939 dvp, nm, vpp, error);
935 940 return (error);
936 941 }
937 942
938 943 /*ARGSUSED7*/
939 944 static int
940 945 tmp_create(
941 946 struct vnode *dvp,
942 947 char *nm,
943 948 struct vattr *vap,
944 949 enum vcexcl exclusive,
945 950 int mode,
946 951 struct vnode **vpp,
947 952 struct cred *cred,
948 953 int flag,
949 954 caller_context_t *ct,
950 955 vsecattr_t *vsecp)
951 956 {
952 957 struct tmpnode *parent;
953 958 struct tmount *tm;
954 959 struct tmpnode *self;
955 960 int error;
956 961 struct tmpnode *oldtp;
957 962
958 963 again:
959 964 parent = (struct tmpnode *)VTOTN(dvp);
960 965 tm = (struct tmount *)VTOTM(dvp);
961 966 self = NULL;
962 967 error = 0;
963 968 oldtp = NULL;
964 969
965 970 /* device files not allowed in ext. attr dirs */
966 971 if ((parent->tn_flags & ISXATTR) &&
967 972 (vap->va_type == VBLK || vap->va_type == VCHR ||
968 973 vap->va_type == VFIFO || vap->va_type == VDOOR ||
969 974 vap->va_type == VSOCK || vap->va_type == VPORT))
970 975 return (EINVAL);
971 976
972 977 if (vap->va_type == VREG && (vap->va_mode & VSVTX)) {
973 978 /* Must be privileged to set sticky bit */
974 979 if (secpolicy_vnode_stky_modify(cred))
975 980 vap->va_mode &= ~VSVTX;
976 981 } else if (vap->va_type == VNON) {
977 982 return (EINVAL);
978 983 }
979 984
980 985 /*
981 986 * Null component name is a synonym for directory being searched.
982 987 */
983 988 if (*nm == '\0') {
984 989 VN_HOLD(dvp);
985 990 oldtp = parent;
986 991 } else {
987 992 error = tdirlookup(parent, nm, &oldtp, cred);
988 993 }
989 994
990 995 if (error == 0) { /* name found */
991 996 boolean_t trunc = B_FALSE;
992 997
993 998 ASSERT(oldtp);
994 999
995 1000 rw_enter(&oldtp->tn_rwlock, RW_WRITER);
996 1001
997 1002 /*
998 1003 * if create/read-only an existing
999 1004 * directory, allow it
1000 1005 */
1001 1006 if (exclusive == EXCL)
1002 1007 error = EEXIST;
1003 1008 else if ((oldtp->tn_type == VDIR) && (mode & VWRITE))
1004 1009 error = EISDIR;
1005 1010 else {
1006 1011 error = tmp_taccess(oldtp, mode, cred);
1007 1012 }
1008 1013
1009 1014 if (error) {
1010 1015 rw_exit(&oldtp->tn_rwlock);
1011 1016 tmpnode_rele(oldtp);
1012 1017 return (error);
1013 1018 }
1014 1019 *vpp = TNTOV(oldtp);
1015 1020 if ((*vpp)->v_type == VREG && (vap->va_mask & AT_SIZE) &&
1016 1021 vap->va_size == 0) {
1017 1022 rw_enter(&oldtp->tn_contents, RW_WRITER);
1018 1023 (void) tmpnode_trunc(tm, oldtp, 0);
1019 1024 rw_exit(&oldtp->tn_contents);
1020 1025 trunc = B_TRUE;
1021 1026 }
1022 1027 rw_exit(&oldtp->tn_rwlock);
1023 1028 if (IS_DEVVP(*vpp)) {
1024 1029 struct vnode *newvp;
1025 1030
1026 1031 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
1027 1032 cred);
1028 1033 VN_RELE(*vpp);
1029 1034 if (newvp == NULL) {
1030 1035 return (ENOSYS);
1031 1036 }
1032 1037 *vpp = newvp;
1033 1038 }
1034 1039
1035 1040 if (trunc)
1036 1041 vnevent_create(*vpp, ct);
1037 1042
1038 1043 return (0);
1039 1044 }
1040 1045
1041 1046 if (error != ENOENT)
1042 1047 return (error);
1043 1048
1044 1049 rw_enter(&parent->tn_rwlock, RW_WRITER);
1045 1050 error = tdirenter(tm, parent, nm, DE_CREATE,
1046 1051 (struct tmpnode *)NULL, (struct tmpnode *)NULL,
1047 1052 vap, &self, cred, ct);
1048 1053 rw_exit(&parent->tn_rwlock);
1049 1054
1050 1055 if (error) {
1051 1056 if (self)
1052 1057 tmpnode_rele(self);
1053 1058
1054 1059 if (error == EEXIST) {
1055 1060 /*
1056 1061 * This means that the file was created sometime
1057 1062 * after we checked and did not find it and when
1058 1063 * we went to create it.
1059 1064 * Since creat() is supposed to truncate a file
1060 1065 * that already exits go back to the begining
1061 1066 * of the function. This time we will find it
1062 1067 * and go down the tmp_trunc() path
1063 1068 */
1064 1069 goto again;
1065 1070 }
1066 1071 return (error);
1067 1072 }
1068 1073
1069 1074 *vpp = TNTOV(self);
1070 1075
1071 1076 if (!error && IS_DEVVP(*vpp)) {
1072 1077 struct vnode *newvp;
1073 1078
1074 1079 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cred);
1075 1080 VN_RELE(*vpp);
1076 1081 if (newvp == NULL)
1077 1082 return (ENOSYS);
1078 1083 *vpp = newvp;
1079 1084 }
1080 1085 TRACE_3(TR_FAC_TMPFS, TR_TMPFS_CREATE,
1081 1086 "tmpfs create:dvp %p nm %s vpp %p", dvp, nm, vpp);
1082 1087 return (0);
1083 1088 }
1084 1089
1085 1090 /* ARGSUSED3 */
1086 1091 static int
1087 1092 tmp_remove(
1088 1093 struct vnode *dvp,
1089 1094 char *nm,
1090 1095 struct cred *cred,
1091 1096 caller_context_t *ct,
1092 1097 int flags)
1093 1098 {
1094 1099 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1095 1100 int error;
1096 1101 struct tmpnode *tp = NULL;
1097 1102
1098 1103 error = tdirlookup(parent, nm, &tp, cred);
1099 1104 if (error)
1100 1105 return (error);
1101 1106
1102 1107 ASSERT(tp);
1103 1108 rw_enter(&parent->tn_rwlock, RW_WRITER);
1104 1109 rw_enter(&tp->tn_rwlock, RW_WRITER);
1105 1110
1106 1111 if (tp->tn_type != VDIR ||
1107 1112 (error = secpolicy_fs_linkdir(cred, dvp->v_vfsp)) == 0)
1108 1113 error = tdirdelete(parent, tp, nm, tp->tn_type == VDIR ?
1109 1114 DR_RMDIR : DR_REMOVE, cred);
1110 1115
1111 1116 rw_exit(&tp->tn_rwlock);
1112 1117 rw_exit(&parent->tn_rwlock);
1113 1118 vnevent_remove(TNTOV(tp), dvp, nm, ct);
1114 1119 tmpnode_rele(tp);
1115 1120
1116 1121 TRACE_3(TR_FAC_TMPFS, TR_TMPFS_REMOVE,
1117 1122 "tmpfs remove:dvp %p nm %s error %d", dvp, nm, error);
1118 1123 return (error);
1119 1124 }
1120 1125
1121 1126 /* ARGSUSED4 */
1122 1127 static int
1123 1128 tmp_link(
1124 1129 struct vnode *dvp,
1125 1130 struct vnode *srcvp,
1126 1131 char *tnm,
1127 1132 struct cred *cred,
1128 1133 caller_context_t *ct,
1129 1134 int flags)
1130 1135 {
1131 1136 struct tmpnode *parent;
1132 1137 struct tmpnode *from;
1133 1138 struct tmount *tm = (struct tmount *)VTOTM(dvp);
1134 1139 int error;
1135 1140 struct tmpnode *found = NULL;
1136 1141 struct vnode *realvp;
1137 1142
1138 1143 if (VOP_REALVP(srcvp, &realvp, ct) == 0)
1139 1144 srcvp = realvp;
1140 1145
1141 1146 parent = (struct tmpnode *)VTOTN(dvp);
1142 1147 from = (struct tmpnode *)VTOTN(srcvp);
1143 1148
1144 1149 if ((srcvp->v_type == VDIR &&
1145 1150 secpolicy_fs_linkdir(cred, dvp->v_vfsp)) ||
1146 1151 (from->tn_uid != crgetuid(cred) && secpolicy_basic_link(cred)))
1147 1152 return (EPERM);
1148 1153
1149 1154 /*
1150 1155 * Make sure link for extended attributes is valid
1151 1156 * We only support hard linking of xattr's in xattrdir to an xattrdir
1152 1157 */
1153 1158 if ((from->tn_flags & ISXATTR) != (parent->tn_flags & ISXATTR))
1154 1159 return (EINVAL);
1155 1160
1156 1161 error = tdirlookup(parent, tnm, &found, cred);
1157 1162 if (error == 0) {
1158 1163 ASSERT(found);
1159 1164 tmpnode_rele(found);
1160 1165 return (EEXIST);
1161 1166 }
1162 1167
1163 1168 if (error != ENOENT)
1164 1169 return (error);
1165 1170
1166 1171 rw_enter(&parent->tn_rwlock, RW_WRITER);
1167 1172 error = tdirenter(tm, parent, tnm, DE_LINK, (struct tmpnode *)NULL,
1168 1173 from, NULL, (struct tmpnode **)NULL, cred, ct);
1169 1174 rw_exit(&parent->tn_rwlock);
1170 1175 if (error == 0) {
1171 1176 vnevent_link(srcvp, ct);
1172 1177 }
1173 1178 return (error);
1174 1179 }
1175 1180
1176 1181 /* ARGSUSED5 */
1177 1182 static int
1178 1183 tmp_rename(
1179 1184 struct vnode *odvp, /* source parent vnode */
1180 1185 char *onm, /* source name */
1181 1186 struct vnode *ndvp, /* destination parent vnode */
1182 1187 char *nnm, /* destination name */
1183 1188 struct cred *cred,
1184 1189 caller_context_t *ct,
1185 1190 int flags)
1186 1191 {
1187 1192 struct tmpnode *fromparent;
1188 1193 struct tmpnode *toparent;
1189 1194 struct tmpnode *fromtp = NULL; /* source tmpnode */
1190 1195 struct tmpnode *totp; /* target tmpnode */
1191 1196 struct tmount *tm = (struct tmount *)VTOTM(odvp);
1192 1197 int error;
1193 1198 int samedir = 0; /* set if odvp == ndvp */
1194 1199 struct vnode *realvp;
1195 1200
1196 1201 if (VOP_REALVP(ndvp, &realvp, ct) == 0)
1197 1202 ndvp = realvp;
1198 1203
1199 1204 fromparent = (struct tmpnode *)VTOTN(odvp);
1200 1205 toparent = (struct tmpnode *)VTOTN(ndvp);
1201 1206
1202 1207 if ((fromparent->tn_flags & ISXATTR) != (toparent->tn_flags & ISXATTR))
1203 1208 return (EINVAL);
1204 1209
1205 1210 mutex_enter(&tm->tm_renamelck);
1206 1211
1207 1212 /*
1208 1213 * Look up tmpnode of file we're supposed to rename.
1209 1214 */
1210 1215 error = tdirlookup(fromparent, onm, &fromtp, cred);
1211 1216 if (error) {
1212 1217 mutex_exit(&tm->tm_renamelck);
1213 1218 return (error);
1214 1219 }
1215 1220
1216 1221 /*
1217 1222 * Make sure we can delete the old (source) entry. This
1218 1223 * requires write permission on the containing directory. If
1219 1224 * that directory is "sticky" it requires further checks.
1220 1225 */
1221 1226 if (((error = tmp_taccess(fromparent, VWRITE, cred)) != 0) ||
1222 1227 (error = tmp_sticky_remove_access(fromparent, fromtp, cred)) != 0)
1223 1228 goto done;
1224 1229
1225 1230 /*
1226 1231 * Check for renaming to or from '.' or '..' or that
1227 1232 * fromtp == fromparent
1228 1233 */
1229 1234 if ((onm[0] == '.' &&
1230 1235 (onm[1] == '\0' || (onm[1] == '.' && onm[2] == '\0'))) ||
1231 1236 (nnm[0] == '.' &&
1232 1237 (nnm[1] == '\0' || (nnm[1] == '.' && nnm[2] == '\0'))) ||
1233 1238 (fromparent == fromtp)) {
1234 1239 error = EINVAL;
1235 1240 goto done;
1236 1241 }
1237 1242
1238 1243 samedir = (fromparent == toparent);
1239 1244 /*
1240 1245 * Make sure we can search and rename into the new
1241 1246 * (destination) directory.
1242 1247 */
1243 1248 if (!samedir) {
1244 1249 error = tmp_taccess(toparent, VEXEC|VWRITE, cred);
1245 1250 if (error)
1246 1251 goto done;
1247 1252 }
1248 1253
1249 1254 if (tdirlookup(toparent, nnm, &totp, cred) == 0) {
1250 1255 vnevent_pre_rename_dest(TNTOV(totp), ndvp, nnm, ct);
1251 1256 tmpnode_rele(totp);
1252 1257 }
1253 1258
1254 1259 /* Notify the target dir. if not the same as the source dir. */
1255 1260 if (ndvp != odvp) {
1256 1261 vnevent_pre_rename_dest_dir(ndvp, TNTOV(fromtp), nnm, ct);
1257 1262 }
1258 1263
1259 1264 vnevent_pre_rename_src(TNTOV(fromtp), odvp, onm, ct);
1260 1265
1261 1266 /*
1262 1267 * Link source to new target
1263 1268 */
1264 1269 rw_enter(&toparent->tn_rwlock, RW_WRITER);
1265 1270 error = tdirenter(tm, toparent, nnm, DE_RENAME,
1266 1271 fromparent, fromtp, (struct vattr *)NULL,
1267 1272 (struct tmpnode **)NULL, cred, ct);
1268 1273 rw_exit(&toparent->tn_rwlock);
1269 1274
1270 1275 if (error) {
1271 1276 /*
1272 1277 * ESAME isn't really an error; it indicates that the
1273 1278 * operation should not be done because the source and target
1274 1279 * are the same file, but that no error should be reported.
1275 1280 */
1276 1281 if (error == ESAME)
1277 1282 error = 0;
1278 1283 goto done;
1279 1284 }
1280 1285
1281 1286 /*
1282 1287 * Unlink from source.
1283 1288 */
1284 1289 rw_enter(&fromparent->tn_rwlock, RW_WRITER);
1285 1290 rw_enter(&fromtp->tn_rwlock, RW_WRITER);
1286 1291
1287 1292 error = tdirdelete(fromparent, fromtp, onm, DR_RENAME, cred);
1288 1293
1289 1294 /*
1290 1295 * The following handles the case where our source tmpnode was
1291 1296 * removed before we got to it.
1292 1297 *
1293 1298 * XXX We should also cleanup properly in the case where tdirdelete
1294 1299 * fails for some other reason. Currently this case shouldn't happen.
1295 1300 * (see 1184991).
1296 1301 */
|
↓ open down ↓ |
495 lines elided |
↑ open up ↑ |
1297 1302 if (error == ENOENT)
1298 1303 error = 0;
1299 1304
1300 1305 rw_exit(&fromtp->tn_rwlock);
1301 1306 rw_exit(&fromparent->tn_rwlock);
1302 1307
1303 1308 if (error == 0) {
1304 1309 vnevent_rename_src(TNTOV(fromtp), odvp, onm, ct);
1305 1310 /*
1306 1311 * vnevent_rename_dest is called in tdirenter().
1307 - * Notify the target dir if not same as source dir.
1308 1312 */
1309 - if (ndvp != odvp)
1310 - vnevent_rename_dest_dir(ndvp, ct);
1313 + vnevent_rename_dest_dir(ndvp, TNTOV(fromtp), nnm, ct);
1311 1314 }
1312 1315
1313 1316 done:
1314 1317 tmpnode_rele(fromtp);
1315 1318 mutex_exit(&tm->tm_renamelck);
1316 1319
1317 1320 TRACE_5(TR_FAC_TMPFS, TR_TMPFS_RENAME,
1318 1321 "tmpfs rename:ovp %p onm %s nvp %p nnm %s error %d", odvp, onm,
1319 1322 ndvp, nnm, error);
1320 1323 return (error);
1321 1324 }
1322 1325
1323 1326 /* ARGSUSED5 */
1324 1327 static int
1325 1328 tmp_mkdir(
1326 1329 struct vnode *dvp,
1327 1330 char *nm,
1328 1331 struct vattr *va,
1329 1332 struct vnode **vpp,
1330 1333 struct cred *cred,
1331 1334 caller_context_t *ct,
1332 1335 int flags,
1333 1336 vsecattr_t *vsecp)
1334 1337 {
1335 1338 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1336 1339 struct tmpnode *self = NULL;
1337 1340 struct tmount *tm = (struct tmount *)VTOTM(dvp);
1338 1341 int error;
1339 1342
1340 1343 /* no new dirs allowed in xattr dirs */
1341 1344 if (parent->tn_flags & ISXATTR)
1342 1345 return (EINVAL);
1343 1346
1344 1347 /*
1345 1348 * Might be dangling directory. Catch it here,
1346 1349 * because a ENOENT return from tdirlookup() is
1347 1350 * an "o.k. return".
1348 1351 */
1349 1352 if (parent->tn_nlink == 0)
1350 1353 return (ENOENT);
1351 1354
1352 1355 error = tdirlookup(parent, nm, &self, cred);
1353 1356 if (error == 0) {
1354 1357 ASSERT(self);
1355 1358 tmpnode_rele(self);
1356 1359 return (EEXIST);
1357 1360 }
1358 1361 if (error != ENOENT)
1359 1362 return (error);
1360 1363
1361 1364 rw_enter(&parent->tn_rwlock, RW_WRITER);
1362 1365 error = tdirenter(tm, parent, nm, DE_MKDIR, (struct tmpnode *)NULL,
1363 1366 (struct tmpnode *)NULL, va, &self, cred, ct);
1364 1367 if (error) {
1365 1368 rw_exit(&parent->tn_rwlock);
1366 1369 if (self)
1367 1370 tmpnode_rele(self);
1368 1371 return (error);
1369 1372 }
1370 1373 rw_exit(&parent->tn_rwlock);
1371 1374 *vpp = TNTOV(self);
1372 1375 return (0);
1373 1376 }
1374 1377
1375 1378 /* ARGSUSED4 */
1376 1379 static int
1377 1380 tmp_rmdir(
1378 1381 struct vnode *dvp,
1379 1382 char *nm,
1380 1383 struct vnode *cdir,
1381 1384 struct cred *cred,
1382 1385 caller_context_t *ct,
1383 1386 int flags)
1384 1387 {
1385 1388 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1386 1389 struct tmpnode *self = NULL;
1387 1390 struct vnode *vp;
1388 1391 int error = 0;
1389 1392
1390 1393 /*
1391 1394 * Return error when removing . and ..
1392 1395 */
1393 1396 if (strcmp(nm, ".") == 0)
1394 1397 return (EINVAL);
1395 1398 if (strcmp(nm, "..") == 0)
1396 1399 return (EEXIST); /* Should be ENOTEMPTY */
1397 1400 error = tdirlookup(parent, nm, &self, cred);
1398 1401 if (error)
1399 1402 return (error);
1400 1403
1401 1404 rw_enter(&parent->tn_rwlock, RW_WRITER);
1402 1405 rw_enter(&self->tn_rwlock, RW_WRITER);
1403 1406
1404 1407 vp = TNTOV(self);
1405 1408 if (vp == dvp || vp == cdir) {
1406 1409 error = EINVAL;
1407 1410 goto done1;
1408 1411 }
1409 1412 if (self->tn_type != VDIR) {
1410 1413 error = ENOTDIR;
1411 1414 goto done1;
1412 1415 }
1413 1416
1414 1417 mutex_enter(&self->tn_tlock);
1415 1418 if (self->tn_nlink > 2) {
1416 1419 mutex_exit(&self->tn_tlock);
1417 1420 error = EEXIST;
1418 1421 goto done1;
1419 1422 }
1420 1423 mutex_exit(&self->tn_tlock);
1421 1424
1422 1425 if (vn_vfswlock(vp)) {
1423 1426 error = EBUSY;
1424 1427 goto done1;
1425 1428 }
1426 1429 if (vn_mountedvfs(vp) != NULL) {
1427 1430 error = EBUSY;
1428 1431 goto done;
1429 1432 }
1430 1433
1431 1434 /*
1432 1435 * Check for an empty directory
1433 1436 * i.e. only includes entries for "." and ".."
1434 1437 */
1435 1438 if (self->tn_dirents > 2) {
1436 1439 error = EEXIST; /* SIGH should be ENOTEMPTY */
1437 1440 /*
1438 1441 * Update atime because checking tn_dirents is logically
1439 1442 * equivalent to reading the directory
1440 1443 */
1441 1444 gethrestime(&self->tn_atime);
1442 1445 goto done;
1443 1446 }
1444 1447
1445 1448 error = tdirdelete(parent, self, nm, DR_RMDIR, cred);
1446 1449 done:
1447 1450 vn_vfsunlock(vp);
1448 1451 done1:
1449 1452 rw_exit(&self->tn_rwlock);
1450 1453 rw_exit(&parent->tn_rwlock);
1451 1454 vnevent_rmdir(TNTOV(self), dvp, nm, ct);
1452 1455 tmpnode_rele(self);
1453 1456
1454 1457 return (error);
1455 1458 }
1456 1459
1457 1460 /* ARGSUSED2 */
1458 1461 static int
1459 1462 tmp_readdir(
1460 1463 struct vnode *vp,
1461 1464 struct uio *uiop,
1462 1465 struct cred *cred,
1463 1466 int *eofp,
1464 1467 caller_context_t *ct,
1465 1468 int flags)
1466 1469 {
1467 1470 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1468 1471 struct tdirent *tdp;
1469 1472 int error = 0;
1470 1473 size_t namelen;
1471 1474 struct dirent64 *dp;
1472 1475 ulong_t offset;
1473 1476 ulong_t total_bytes_wanted;
1474 1477 long outcount = 0;
1475 1478 long bufsize;
1476 1479 int reclen;
1477 1480 caddr_t outbuf;
1478 1481
1479 1482 if (uiop->uio_loffset >= MAXOFF_T) {
1480 1483 if (eofp)
1481 1484 *eofp = 1;
1482 1485 return (0);
1483 1486 }
1484 1487 /*
1485 1488 * assuming system call has already called tmp_rwlock
1486 1489 */
1487 1490 ASSERT(RW_READ_HELD(&tp->tn_rwlock));
1488 1491
1489 1492 if (uiop->uio_iovcnt != 1)
1490 1493 return (EINVAL);
1491 1494
1492 1495 if (vp->v_type != VDIR)
1493 1496 return (ENOTDIR);
1494 1497
1495 1498 /*
1496 1499 * There's a window here where someone could have removed
1497 1500 * all the entries in the directory after we put a hold on the
1498 1501 * vnode but before we grabbed the rwlock. Just return.
1499 1502 */
1500 1503 if (tp->tn_dir == NULL) {
1501 1504 if (tp->tn_nlink) {
1502 1505 panic("empty directory 0x%p", (void *)tp);
1503 1506 /*NOTREACHED*/
1504 1507 }
1505 1508 return (0);
1506 1509 }
1507 1510
1508 1511 /*
1509 1512 * Get space for multiple directory entries
1510 1513 */
1511 1514 total_bytes_wanted = uiop->uio_iov->iov_len;
1512 1515 bufsize = total_bytes_wanted + sizeof (struct dirent64);
1513 1516 outbuf = kmem_alloc(bufsize, KM_SLEEP);
1514 1517
1515 1518 dp = (struct dirent64 *)outbuf;
1516 1519
1517 1520
1518 1521 offset = 0;
1519 1522 tdp = tp->tn_dir;
1520 1523 while (tdp) {
1521 1524 namelen = strlen(tdp->td_name); /* no +1 needed */
1522 1525 offset = tdp->td_offset;
1523 1526 if (offset >= uiop->uio_offset) {
1524 1527 reclen = (int)DIRENT64_RECLEN(namelen);
1525 1528 if (outcount + reclen > total_bytes_wanted) {
1526 1529 if (!outcount)
1527 1530 /*
1528 1531 * Buffer too small for any entries.
1529 1532 */
1530 1533 error = EINVAL;
1531 1534 break;
1532 1535 }
1533 1536 ASSERT(tdp->td_tmpnode != NULL);
1534 1537
1535 1538 /* use strncpy(9f) to zero out uninitialized bytes */
1536 1539
1537 1540 (void) strncpy(dp->d_name, tdp->td_name,
1538 1541 DIRENT64_NAMELEN(reclen));
1539 1542 dp->d_reclen = (ushort_t)reclen;
1540 1543 dp->d_ino = (ino64_t)tdp->td_tmpnode->tn_nodeid;
1541 1544 dp->d_off = (offset_t)tdp->td_offset + 1;
1542 1545 dp = (struct dirent64 *)
1543 1546 ((uintptr_t)dp + dp->d_reclen);
1544 1547 outcount += reclen;
1545 1548 ASSERT(outcount <= bufsize);
1546 1549 }
1547 1550 tdp = tdp->td_next;
1548 1551 }
1549 1552
1550 1553 if (!error)
1551 1554 error = uiomove(outbuf, outcount, UIO_READ, uiop);
1552 1555
1553 1556 if (!error) {
1554 1557 /* If we reached the end of the list our offset */
1555 1558 /* should now be just past the end. */
1556 1559 if (!tdp) {
1557 1560 offset += 1;
1558 1561 if (eofp)
1559 1562 *eofp = 1;
1560 1563 } else if (eofp)
1561 1564 *eofp = 0;
1562 1565 uiop->uio_offset = offset;
1563 1566 }
1564 1567 gethrestime(&tp->tn_atime);
1565 1568 kmem_free(outbuf, bufsize);
1566 1569 return (error);
1567 1570 }
1568 1571
1569 1572 /* ARGSUSED5 */
1570 1573 static int
1571 1574 tmp_symlink(
1572 1575 struct vnode *dvp,
1573 1576 char *lnm,
1574 1577 struct vattr *tva,
1575 1578 char *tnm,
1576 1579 struct cred *cred,
1577 1580 caller_context_t *ct,
1578 1581 int flags)
1579 1582 {
1580 1583 struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1581 1584 struct tmpnode *self = (struct tmpnode *)NULL;
1582 1585 struct tmount *tm = (struct tmount *)VTOTM(dvp);
1583 1586 char *cp = NULL;
1584 1587 int error;
1585 1588 size_t len;
1586 1589
1587 1590 /* no symlinks allowed to files in xattr dirs */
1588 1591 if (parent->tn_flags & ISXATTR)
1589 1592 return (EINVAL);
1590 1593
1591 1594 error = tdirlookup(parent, lnm, &self, cred);
1592 1595 if (error == 0) {
1593 1596 /*
1594 1597 * The entry already exists
1595 1598 */
1596 1599 tmpnode_rele(self);
1597 1600 return (EEXIST); /* was 0 */
1598 1601 }
1599 1602
1600 1603 if (error != ENOENT) {
1601 1604 if (self != NULL)
1602 1605 tmpnode_rele(self);
1603 1606 return (error);
1604 1607 }
1605 1608
1606 1609 rw_enter(&parent->tn_rwlock, RW_WRITER);
1607 1610 error = tdirenter(tm, parent, lnm, DE_CREATE, (struct tmpnode *)NULL,
1608 1611 (struct tmpnode *)NULL, tva, &self, cred, ct);
1609 1612 rw_exit(&parent->tn_rwlock);
1610 1613
1611 1614 if (error) {
1612 1615 if (self)
1613 1616 tmpnode_rele(self);
1614 1617 return (error);
1615 1618 }
1616 1619 len = strlen(tnm) + 1;
1617 1620 cp = tmp_memalloc(len, 0);
1618 1621 if (cp == NULL) {
1619 1622 tmpnode_rele(self);
1620 1623 return (ENOSPC);
1621 1624 }
1622 1625 (void) strcpy(cp, tnm);
1623 1626
1624 1627 self->tn_symlink = cp;
1625 1628 self->tn_size = len - 1;
1626 1629 tmpnode_rele(self);
1627 1630 return (error);
1628 1631 }
1629 1632
1630 1633 /* ARGSUSED2 */
1631 1634 static int
1632 1635 tmp_readlink(
1633 1636 struct vnode *vp,
1634 1637 struct uio *uiop,
1635 1638 struct cred *cred,
1636 1639 caller_context_t *ct)
1637 1640 {
1638 1641 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1639 1642 int error = 0;
1640 1643
1641 1644 if (vp->v_type != VLNK)
1642 1645 return (EINVAL);
1643 1646
1644 1647 rw_enter(&tp->tn_rwlock, RW_READER);
1645 1648 rw_enter(&tp->tn_contents, RW_READER);
1646 1649 error = uiomove(tp->tn_symlink, tp->tn_size, UIO_READ, uiop);
1647 1650 gethrestime(&tp->tn_atime);
1648 1651 rw_exit(&tp->tn_contents);
1649 1652 rw_exit(&tp->tn_rwlock);
1650 1653 return (error);
1651 1654 }
1652 1655
1653 1656 /* ARGSUSED */
1654 1657 static int
1655 1658 tmp_fsync(
1656 1659 struct vnode *vp,
1657 1660 int syncflag,
1658 1661 struct cred *cred,
1659 1662 caller_context_t *ct)
1660 1663 {
1661 1664 return (0);
1662 1665 }
1663 1666
1664 1667 /* ARGSUSED */
1665 1668 static void
1666 1669 tmp_inactive(struct vnode *vp, struct cred *cred, caller_context_t *ct)
1667 1670 {
1668 1671 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1669 1672 struct tmount *tm = (struct tmount *)VFSTOTM(vp->v_vfsp);
1670 1673
1671 1674 rw_enter(&tp->tn_rwlock, RW_WRITER);
1672 1675 top:
1673 1676 mutex_enter(&tp->tn_tlock);
1674 1677 mutex_enter(&vp->v_lock);
1675 1678 ASSERT(vp->v_count >= 1);
1676 1679
1677 1680 /*
1678 1681 * If we don't have the last hold or the link count is non-zero,
1679 1682 * there's little to do -- just drop our hold.
1680 1683 */
1681 1684 if (vp->v_count > 1 || tp->tn_nlink != 0) {
1682 1685 vp->v_count--;
1683 1686 mutex_exit(&vp->v_lock);
1684 1687 mutex_exit(&tp->tn_tlock);
1685 1688 rw_exit(&tp->tn_rwlock);
1686 1689 return;
1687 1690 }
1688 1691
1689 1692 /*
1690 1693 * We have the last hold *and* the link count is zero, so this
1691 1694 * tmpnode is dead from the filesystem's viewpoint. However,
1692 1695 * if the tmpnode has any pages associated with it (i.e. if it's
1693 1696 * a normal file with non-zero size), the tmpnode can still be
1694 1697 * discovered by pageout or fsflush via the page vnode pointers.
1695 1698 * In this case we must drop all our locks, truncate the tmpnode,
1696 1699 * and try the whole dance again.
1697 1700 */
1698 1701 if (tp->tn_size != 0) {
1699 1702 if (tp->tn_type == VREG) {
1700 1703 mutex_exit(&vp->v_lock);
1701 1704 mutex_exit(&tp->tn_tlock);
1702 1705 rw_enter(&tp->tn_contents, RW_WRITER);
1703 1706 (void) tmpnode_trunc(tm, tp, 0);
1704 1707 rw_exit(&tp->tn_contents);
1705 1708 ASSERT(tp->tn_size == 0);
1706 1709 ASSERT(tp->tn_nblocks == 0);
1707 1710 goto top;
1708 1711 }
1709 1712 if (tp->tn_type == VLNK)
1710 1713 tmp_memfree(tp->tn_symlink, tp->tn_size + 1);
1711 1714 }
1712 1715
1713 1716 /*
1714 1717 * Remove normal file/dir's xattr dir and xattrs.
1715 1718 */
1716 1719 if (tp->tn_xattrdp) {
1717 1720 struct tmpnode *xtp = tp->tn_xattrdp;
1718 1721
1719 1722 ASSERT(xtp->tn_flags & ISXATTR);
1720 1723 tmpnode_hold(xtp);
1721 1724 rw_enter(&xtp->tn_rwlock, RW_WRITER);
1722 1725 tdirtrunc(xtp);
1723 1726 DECR_COUNT(&xtp->tn_nlink, &xtp->tn_tlock);
1724 1727 tp->tn_xattrdp = NULL;
1725 1728 rw_exit(&xtp->tn_rwlock);
1726 1729 tmpnode_rele(xtp);
1727 1730 }
1728 1731
1729 1732 mutex_exit(&vp->v_lock);
1730 1733 mutex_exit(&tp->tn_tlock);
1731 1734 /* Here's our chance to send invalid event while we're between locks */
1732 1735 vn_invalid(TNTOV(tp));
1733 1736 mutex_enter(&tm->tm_contents);
1734 1737 if (tp->tn_forw == NULL)
1735 1738 tm->tm_rootnode->tn_back = tp->tn_back;
1736 1739 else
1737 1740 tp->tn_forw->tn_back = tp->tn_back;
1738 1741 tp->tn_back->tn_forw = tp->tn_forw;
1739 1742 mutex_exit(&tm->tm_contents);
1740 1743 rw_exit(&tp->tn_rwlock);
1741 1744 rw_destroy(&tp->tn_rwlock);
1742 1745 mutex_destroy(&tp->tn_tlock);
1743 1746 vn_free(TNTOV(tp));
1744 1747 tmp_memfree(tp, sizeof (struct tmpnode));
1745 1748 }
1746 1749
1747 1750 /* ARGSUSED2 */
1748 1751 static int
1749 1752 tmp_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
1750 1753 {
1751 1754 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1752 1755 struct tfid *tfid;
1753 1756
1754 1757 if (fidp->fid_len < (sizeof (struct tfid) - sizeof (ushort_t))) {
1755 1758 fidp->fid_len = sizeof (struct tfid) - sizeof (ushort_t);
1756 1759 return (ENOSPC);
1757 1760 }
1758 1761
1759 1762 tfid = (struct tfid *)fidp;
1760 1763 bzero(tfid, sizeof (struct tfid));
1761 1764 tfid->tfid_len = (int)sizeof (struct tfid) - sizeof (ushort_t);
1762 1765
1763 1766 tfid->tfid_ino = tp->tn_nodeid;
1764 1767 tfid->tfid_gen = tp->tn_gen;
1765 1768
1766 1769 return (0);
1767 1770 }
1768 1771
1769 1772
1770 1773 /*
1771 1774 * Return all the pages from [off..off+len] in given file
1772 1775 */
1773 1776 /* ARGSUSED */
1774 1777 static int
1775 1778 tmp_getpage(
1776 1779 struct vnode *vp,
1777 1780 offset_t off,
1778 1781 size_t len,
1779 1782 uint_t *protp,
1780 1783 page_t *pl[],
1781 1784 size_t plsz,
1782 1785 struct seg *seg,
1783 1786 caddr_t addr,
1784 1787 enum seg_rw rw,
1785 1788 struct cred *cr,
1786 1789 caller_context_t *ct)
1787 1790 {
1788 1791 int err = 0;
1789 1792 struct tmpnode *tp = VTOTN(vp);
1790 1793 anoff_t toff = (anoff_t)off;
1791 1794 size_t tlen = len;
1792 1795 u_offset_t tmpoff;
1793 1796 timestruc_t now;
1794 1797
1795 1798 rw_enter(&tp->tn_contents, RW_READER);
1796 1799
1797 1800 if (off + len > tp->tn_size + PAGEOFFSET) {
1798 1801 err = EFAULT;
1799 1802 goto out;
1800 1803 }
1801 1804 /*
1802 1805 * Look for holes (no anon slot) in faulting range. If there are
1803 1806 * holes we have to switch to a write lock and fill them in. Swap
1804 1807 * space for holes was already reserved when the file was grown.
1805 1808 */
1806 1809 tmpoff = toff;
1807 1810 if (non_anon(tp->tn_anon, btop(off), &tmpoff, &tlen)) {
1808 1811 if (!rw_tryupgrade(&tp->tn_contents)) {
1809 1812 rw_exit(&tp->tn_contents);
1810 1813 rw_enter(&tp->tn_contents, RW_WRITER);
1811 1814 /* Size may have changed when lock was dropped */
1812 1815 if (off + len > tp->tn_size + PAGEOFFSET) {
1813 1816 err = EFAULT;
1814 1817 goto out;
1815 1818 }
1816 1819 }
1817 1820 for (toff = (anoff_t)off; toff < (anoff_t)off + len;
1818 1821 toff += PAGESIZE) {
1819 1822 if (anon_get_ptr(tp->tn_anon, btop(toff)) == NULL) {
1820 1823 /* XXX - may allocate mem w. write lock held */
1821 1824 (void) anon_set_ptr(tp->tn_anon, btop(toff),
1822 1825 anon_alloc(vp, toff), ANON_SLEEP);
1823 1826 tp->tn_nblocks++;
1824 1827 }
1825 1828 }
1826 1829 rw_downgrade(&tp->tn_contents);
1827 1830 }
1828 1831
1829 1832
1830 1833 err = pvn_getpages(tmp_getapage, vp, (u_offset_t)off, len, protp,
1831 1834 pl, plsz, seg, addr, rw, cr);
1832 1835
1833 1836 gethrestime(&now);
1834 1837 tp->tn_atime = now;
1835 1838 if (rw == S_WRITE)
1836 1839 tp->tn_mtime = now;
1837 1840
1838 1841 out:
1839 1842 rw_exit(&tp->tn_contents);
1840 1843 return (err);
1841 1844 }
1842 1845
1843 1846 /*
1844 1847 * Called from pvn_getpages to get a particular page.
1845 1848 */
1846 1849 /*ARGSUSED*/
1847 1850 static int
1848 1851 tmp_getapage(
1849 1852 struct vnode *vp,
1850 1853 u_offset_t off,
1851 1854 size_t len,
1852 1855 uint_t *protp,
1853 1856 page_t *pl[],
1854 1857 size_t plsz,
1855 1858 struct seg *seg,
1856 1859 caddr_t addr,
1857 1860 enum seg_rw rw,
1858 1861 struct cred *cr)
1859 1862 {
1860 1863 struct page *pp;
1861 1864 int flags;
1862 1865 int err = 0;
1863 1866 struct vnode *pvp;
1864 1867 u_offset_t poff;
1865 1868
1866 1869 if (protp != NULL)
1867 1870 *protp = PROT_ALL;
1868 1871 again:
1869 1872 if (pp = page_lookup(vp, off, rw == S_CREATE ? SE_EXCL : SE_SHARED)) {
1870 1873 if (pl) {
1871 1874 pl[0] = pp;
1872 1875 pl[1] = NULL;
1873 1876 } else {
1874 1877 page_unlock(pp);
1875 1878 }
1876 1879 } else {
1877 1880 pp = page_create_va(vp, off, PAGESIZE,
1878 1881 PG_WAIT | PG_EXCL, seg, addr);
1879 1882 /*
1880 1883 * Someone raced in and created the page after we did the
1881 1884 * lookup but before we did the create, so go back and
1882 1885 * try to look it up again.
1883 1886 */
1884 1887 if (pp == NULL)
1885 1888 goto again;
1886 1889 /*
1887 1890 * Fill page from backing store, if any. If none, then
1888 1891 * either this is a newly filled hole or page must have
1889 1892 * been unmodified and freed so just zero it out.
1890 1893 */
1891 1894 err = swap_getphysname(vp, off, &pvp, &poff);
1892 1895 if (err) {
1893 1896 panic("tmp_getapage: no anon slot vp %p "
1894 1897 "off %llx pp %p\n", (void *)vp, off, (void *)pp);
1895 1898 }
1896 1899 if (pvp) {
1897 1900 flags = (pl == NULL ? B_ASYNC|B_READ : B_READ);
1898 1901 err = VOP_PAGEIO(pvp, pp, (u_offset_t)poff, PAGESIZE,
1899 1902 flags, cr, NULL);
1900 1903 if (flags & B_ASYNC)
1901 1904 pp = NULL;
1902 1905 } else if (rw != S_CREATE) {
1903 1906 pagezero(pp, 0, PAGESIZE);
1904 1907 }
1905 1908 if (err && pp)
1906 1909 pvn_read_done(pp, B_ERROR);
1907 1910 if (err == 0) {
1908 1911 if (pl)
1909 1912 pvn_plist_init(pp, pl, plsz, off, PAGESIZE, rw);
1910 1913 else
1911 1914 pvn_io_done(pp);
1912 1915 }
1913 1916 }
1914 1917 return (err);
1915 1918 }
1916 1919
1917 1920
1918 1921 /*
1919 1922 * Flags are composed of {B_INVAL, B_DIRTY B_FREE, B_DONTNEED}.
1920 1923 * If len == 0, do from off to EOF.
1921 1924 */
1922 1925 static int tmp_nopage = 0; /* Don't do tmp_putpage's if set */
1923 1926
1924 1927 /* ARGSUSED */
1925 1928 int
1926 1929 tmp_putpage(
1927 1930 register struct vnode *vp,
1928 1931 offset_t off,
1929 1932 size_t len,
1930 1933 int flags,
1931 1934 struct cred *cr,
1932 1935 caller_context_t *ct)
1933 1936 {
1934 1937 register page_t *pp;
1935 1938 u_offset_t io_off;
1936 1939 size_t io_len = 0;
1937 1940 int err = 0;
1938 1941 struct tmpnode *tp = VTOTN(vp);
1939 1942 int dolock;
1940 1943
1941 1944 if (tmp_nopage)
1942 1945 return (0);
1943 1946
1944 1947 ASSERT(vp->v_count != 0);
1945 1948
1946 1949 if (vp->v_flag & VNOMAP)
1947 1950 return (ENOSYS);
1948 1951
1949 1952 /*
1950 1953 * This being tmpfs, we don't ever do i/o unless we really
1951 1954 * have to (when we're low on memory and pageout calls us
1952 1955 * with B_ASYNC | B_FREE or the user explicitly asks for it with
1953 1956 * B_DONTNEED).
1954 1957 * XXX to approximately track the mod time like ufs we should
1955 1958 * update the times here. The problem is, once someone does a
1956 1959 * store we never clear the mod bit and do i/o, thus fsflush
1957 1960 * will keep calling us every 30 seconds to do the i/o and we'll
1958 1961 * continually update the mod time. At least we update the mod
1959 1962 * time on the first store because this results in a call to getpage.
1960 1963 */
1961 1964 if (flags != (B_ASYNC | B_FREE) && (flags & B_INVAL) == 0 &&
1962 1965 (flags & B_DONTNEED) == 0)
1963 1966 return (0);
1964 1967 /*
1965 1968 * If this thread owns the lock, i.e., this thread grabbed it
1966 1969 * as writer somewhere above, then we don't need to grab the
1967 1970 * lock as reader in this routine.
1968 1971 */
1969 1972 dolock = (rw_owner(&tp->tn_contents) != curthread);
1970 1973
1971 1974 /*
1972 1975 * If this is pageout don't block on the lock as you could deadlock
1973 1976 * when freemem == 0 (another thread has the read lock and is blocked
1974 1977 * creating a page, and a third thread is waiting to get the writers
1975 1978 * lock - waiting writers priority blocks us from getting the read
1976 1979 * lock). Of course, if the only freeable pages are on this tmpnode
1977 1980 * we're hosed anyways. A better solution might be a new lock type.
1978 1981 * Note: ufs has the same problem.
1979 1982 */
1980 1983 if (curproc == proc_pageout) {
1981 1984 if (!rw_tryenter(&tp->tn_contents, RW_READER))
1982 1985 return (ENOMEM);
1983 1986 } else if (dolock)
1984 1987 rw_enter(&tp->tn_contents, RW_READER);
1985 1988
1986 1989 if (!vn_has_cached_data(vp))
1987 1990 goto out;
1988 1991
1989 1992 if (len == 0) {
1990 1993 if (curproc == proc_pageout) {
1991 1994 panic("tmp: pageout can't block");
1992 1995 /*NOTREACHED*/
1993 1996 }
1994 1997
1995 1998 /* Search the entire vp list for pages >= off. */
1996 1999 err = pvn_vplist_dirty(vp, (u_offset_t)off, tmp_putapage,
1997 2000 flags, cr);
1998 2001 } else {
1999 2002 u_offset_t eoff;
2000 2003
2001 2004 /*
2002 2005 * Loop over all offsets in the range [off...off + len]
2003 2006 * looking for pages to deal with.
2004 2007 */
2005 2008 eoff = MIN(off + len, tp->tn_size);
2006 2009 for (io_off = off; io_off < eoff; io_off += io_len) {
2007 2010 /*
2008 2011 * If we are not invalidating, synchronously
2009 2012 * freeing or writing pages use the routine
2010 2013 * page_lookup_nowait() to prevent reclaiming
2011 2014 * them from the free list.
2012 2015 */
2013 2016 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2014 2017 pp = page_lookup(vp, io_off,
2015 2018 (flags & (B_INVAL | B_FREE)) ?
2016 2019 SE_EXCL : SE_SHARED);
2017 2020 } else {
2018 2021 pp = page_lookup_nowait(vp, io_off,
2019 2022 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2020 2023 }
2021 2024
2022 2025 if (pp == NULL || pvn_getdirty(pp, flags) == 0)
2023 2026 io_len = PAGESIZE;
2024 2027 else {
2025 2028 err = tmp_putapage(vp, pp, &io_off, &io_len,
2026 2029 flags, cr);
2027 2030 if (err != 0)
2028 2031 break;
2029 2032 }
2030 2033 }
2031 2034 }
2032 2035 /* If invalidating, verify all pages on vnode list are gone. */
2033 2036 if (err == 0 && off == 0 && len == 0 &&
2034 2037 (flags & B_INVAL) && vn_has_cached_data(vp)) {
2035 2038 panic("tmp_putpage: B_INVAL, pages not gone");
2036 2039 /*NOTREACHED*/
2037 2040 }
2038 2041 out:
2039 2042 if ((curproc == proc_pageout) || dolock)
2040 2043 rw_exit(&tp->tn_contents);
2041 2044 /*
2042 2045 * Only reason putapage is going to give us SE_NOSWAP as error
2043 2046 * is when we ask a page to be written to physical backing store
2044 2047 * and there is none. Ignore this because we might be dealing
2045 2048 * with a swap page which does not have any backing store
2046 2049 * on disk. In any other case we won't get this error over here.
2047 2050 */
2048 2051 if (err == SE_NOSWAP)
2049 2052 err = 0;
2050 2053 return (err);
2051 2054 }
2052 2055
2053 2056 long tmp_putpagecnt, tmp_pagespushed;
2054 2057
2055 2058 /*
2056 2059 * Write out a single page.
2057 2060 * For tmpfs this means choose a physical swap slot and write the page
2058 2061 * out using VOP_PAGEIO. For performance, we attempt to kluster; i.e.,
2059 2062 * we try to find a bunch of other dirty pages adjacent in the file
2060 2063 * and a bunch of contiguous swap slots, and then write all the pages
2061 2064 * out in a single i/o.
2062 2065 */
2063 2066 /*ARGSUSED*/
2064 2067 static int
2065 2068 tmp_putapage(
2066 2069 struct vnode *vp,
2067 2070 page_t *pp,
2068 2071 u_offset_t *offp,
2069 2072 size_t *lenp,
2070 2073 int flags,
2071 2074 struct cred *cr)
2072 2075 {
2073 2076 int err;
2074 2077 ulong_t klstart, kllen;
2075 2078 page_t *pplist, *npplist;
2076 2079 extern int klustsize;
2077 2080 long tmp_klustsize;
2078 2081 struct tmpnode *tp;
2079 2082 size_t pp_off, pp_len;
2080 2083 u_offset_t io_off;
2081 2084 size_t io_len;
2082 2085 struct vnode *pvp;
2083 2086 u_offset_t pstart;
2084 2087 u_offset_t offset;
2085 2088 u_offset_t tmpoff;
2086 2089
2087 2090 ASSERT(PAGE_LOCKED(pp));
2088 2091
2089 2092 /* Kluster in tmp_klustsize chunks */
2090 2093 tp = VTOTN(vp);
2091 2094 tmp_klustsize = klustsize;
2092 2095 offset = pp->p_offset;
2093 2096 klstart = (offset / tmp_klustsize) * tmp_klustsize;
2094 2097 kllen = MIN(tmp_klustsize, tp->tn_size - klstart);
2095 2098
2096 2099 /* Get a kluster of pages */
2097 2100 pplist =
2098 2101 pvn_write_kluster(vp, pp, &tmpoff, &pp_len, klstart, kllen, flags);
2099 2102
2100 2103 pp_off = (size_t)tmpoff;
2101 2104
2102 2105 /*
2103 2106 * Get a cluster of physical offsets for the pages; the amount we
2104 2107 * get may be some subrange of what we ask for (io_off, io_len).
2105 2108 */
2106 2109 io_off = pp_off;
2107 2110 io_len = pp_len;
2108 2111 err = swap_newphysname(vp, offset, &io_off, &io_len, &pvp, &pstart);
2109 2112 ASSERT(err != SE_NOANON); /* anon slot must have been filled */
2110 2113 if (err) {
2111 2114 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2112 2115 /*
2113 2116 * If this routine is called as a result of segvn_sync
2114 2117 * operation and we have no physical swap then we can get an
2115 2118 * error here. In such case we would return SE_NOSWAP as error.
2116 2119 * At this point, we expect only SE_NOSWAP.
2117 2120 */
2118 2121 ASSERT(err == SE_NOSWAP);
2119 2122 if (flags & B_INVAL)
2120 2123 err = ENOMEM;
2121 2124 goto out;
2122 2125 }
2123 2126 ASSERT(pp_off <= io_off && io_off + io_len <= pp_off + pp_len);
2124 2127 ASSERT(io_off <= offset && offset < io_off + io_len);
2125 2128
2126 2129 /* Toss pages at front/rear that we couldn't get physical backing for */
2127 2130 if (io_off != pp_off) {
2128 2131 npplist = NULL;
2129 2132 page_list_break(&pplist, &npplist, btop(io_off - pp_off));
2130 2133 ASSERT(pplist->p_offset == pp_off);
2131 2134 ASSERT(pplist->p_prev->p_offset == io_off - PAGESIZE);
2132 2135 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2133 2136 pplist = npplist;
2134 2137 }
2135 2138 if (io_off + io_len < pp_off + pp_len) {
2136 2139 npplist = NULL;
2137 2140 page_list_break(&pplist, &npplist, btop(io_len));
2138 2141 ASSERT(npplist->p_offset == io_off + io_len);
2139 2142 ASSERT(npplist->p_prev->p_offset == pp_off + pp_len - PAGESIZE);
2140 2143 pvn_write_done(npplist, B_ERROR | B_WRITE | flags);
2141 2144 }
2142 2145
2143 2146 ASSERT(pplist->p_offset == io_off);
2144 2147 ASSERT(pplist->p_prev->p_offset == io_off + io_len - PAGESIZE);
2145 2148 ASSERT(btopr(io_len) <= btopr(kllen));
2146 2149
2147 2150 /* Do i/o on the remaining kluster */
2148 2151 err = VOP_PAGEIO(pvp, pplist, (u_offset_t)pstart, io_len,
2149 2152 B_WRITE | flags, cr, NULL);
2150 2153
2151 2154 if ((flags & B_ASYNC) == 0) {
2152 2155 pvn_write_done(pplist, ((err) ? B_ERROR : 0) | B_WRITE | flags);
2153 2156 }
2154 2157 out:
2155 2158 if (!err) {
2156 2159 if (offp)
2157 2160 *offp = io_off;
2158 2161 if (lenp)
2159 2162 *lenp = io_len;
2160 2163 tmp_putpagecnt++;
2161 2164 tmp_pagespushed += btop(io_len);
2162 2165 }
2163 2166 if (err && err != ENOMEM && err != SE_NOSWAP)
2164 2167 cmn_err(CE_WARN, "tmp_putapage: err %d\n", err);
2165 2168 return (err);
2166 2169 }
2167 2170
2168 2171 /* ARGSUSED */
2169 2172 static int
2170 2173 tmp_map(
2171 2174 struct vnode *vp,
2172 2175 offset_t off,
2173 2176 struct as *as,
2174 2177 caddr_t *addrp,
2175 2178 size_t len,
2176 2179 uchar_t prot,
2177 2180 uchar_t maxprot,
2178 2181 uint_t flags,
2179 2182 struct cred *cred,
2180 2183 caller_context_t *ct)
2181 2184 {
2182 2185 struct segvn_crargs vn_a;
2183 2186 struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
2184 2187 int error;
2185 2188
2186 2189 #ifdef _ILP32
2187 2190 if (len > MAXOFF_T)
2188 2191 return (ENOMEM);
2189 2192 #endif
2190 2193
2191 2194 if (vp->v_flag & VNOMAP)
2192 2195 return (ENOSYS);
2193 2196
2194 2197 if (off < 0 || (offset_t)(off + len) < 0 ||
2195 2198 off > MAXOFF_T || (off + len) > MAXOFF_T)
2196 2199 return (ENXIO);
2197 2200
2198 2201 if (vp->v_type != VREG)
2199 2202 return (ENODEV);
2200 2203
2201 2204 /*
2202 2205 * Don't allow mapping to locked file
2203 2206 */
2204 2207 if (vn_has_mandatory_locks(vp, tp->tn_mode)) {
2205 2208 return (EAGAIN);
2206 2209 }
2207 2210
2208 2211 as_rangelock(as);
2209 2212 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
2210 2213 if (error != 0) {
2211 2214 as_rangeunlock(as);
2212 2215 return (error);
2213 2216 }
2214 2217
2215 2218 vn_a.vp = vp;
2216 2219 vn_a.offset = (u_offset_t)off;
2217 2220 vn_a.type = flags & MAP_TYPE;
2218 2221 vn_a.prot = prot;
2219 2222 vn_a.maxprot = maxprot;
2220 2223 vn_a.flags = flags & ~MAP_TYPE;
2221 2224 vn_a.cred = cred;
2222 2225 vn_a.amp = NULL;
2223 2226 vn_a.szc = 0;
2224 2227 vn_a.lgrp_mem_policy_flags = 0;
2225 2228
2226 2229 error = as_map(as, *addrp, len, segvn_create, &vn_a);
2227 2230 as_rangeunlock(as);
2228 2231 return (error);
2229 2232 }
2230 2233
2231 2234 /*
2232 2235 * tmp_addmap and tmp_delmap can't be called since the vp
2233 2236 * maintained in the segvn mapping is NULL.
2234 2237 */
2235 2238 /* ARGSUSED */
2236 2239 static int
2237 2240 tmp_addmap(
2238 2241 struct vnode *vp,
2239 2242 offset_t off,
2240 2243 struct as *as,
2241 2244 caddr_t addr,
2242 2245 size_t len,
2243 2246 uchar_t prot,
2244 2247 uchar_t maxprot,
2245 2248 uint_t flags,
2246 2249 struct cred *cred,
2247 2250 caller_context_t *ct)
2248 2251 {
2249 2252 return (0);
2250 2253 }
2251 2254
2252 2255 /* ARGSUSED */
2253 2256 static int
2254 2257 tmp_delmap(
2255 2258 struct vnode *vp,
2256 2259 offset_t off,
2257 2260 struct as *as,
2258 2261 caddr_t addr,
2259 2262 size_t len,
2260 2263 uint_t prot,
2261 2264 uint_t maxprot,
2262 2265 uint_t flags,
2263 2266 struct cred *cred,
2264 2267 caller_context_t *ct)
2265 2268 {
2266 2269 return (0);
2267 2270 }
2268 2271
2269 2272 static int
2270 2273 tmp_freesp(struct vnode *vp, struct flock64 *lp, int flag)
2271 2274 {
2272 2275 register int i;
2273 2276 register struct tmpnode *tp = VTOTN(vp);
2274 2277 int error;
2275 2278
2276 2279 ASSERT(vp->v_type == VREG);
2277 2280 ASSERT(lp->l_start >= 0);
2278 2281
2279 2282 if (lp->l_len != 0)
2280 2283 return (EINVAL);
2281 2284
2282 2285 rw_enter(&tp->tn_rwlock, RW_WRITER);
2283 2286 if (tp->tn_size == lp->l_start) {
2284 2287 rw_exit(&tp->tn_rwlock);
2285 2288 return (0);
2286 2289 }
2287 2290
2288 2291 /*
2289 2292 * Check for any mandatory locks on the range
2290 2293 */
2291 2294 if (MANDLOCK(vp, tp->tn_mode)) {
2292 2295 long save_start;
2293 2296
2294 2297 save_start = lp->l_start;
2295 2298
2296 2299 if (tp->tn_size < lp->l_start) {
2297 2300 /*
2298 2301 * "Truncate up" case: need to make sure there
2299 2302 * is no lock beyond current end-of-file. To
2300 2303 * do so, we need to set l_start to the size
2301 2304 * of the file temporarily.
2302 2305 */
2303 2306 lp->l_start = tp->tn_size;
2304 2307 }
2305 2308 lp->l_type = F_WRLCK;
2306 2309 lp->l_sysid = 0;
2307 2310 lp->l_pid = ttoproc(curthread)->p_pid;
2308 2311 i = (flag & (FNDELAY|FNONBLOCK)) ? 0 : SLPFLCK;
2309 2312 if ((i = reclock(vp, lp, i, 0, lp->l_start, NULL)) != 0 ||
2310 2313 lp->l_type != F_UNLCK) {
2311 2314 rw_exit(&tp->tn_rwlock);
2312 2315 return (i ? i : EAGAIN);
2313 2316 }
2314 2317
2315 2318 lp->l_start = save_start;
2316 2319 }
2317 2320 VFSTOTM(vp->v_vfsp);
2318 2321
2319 2322 rw_enter(&tp->tn_contents, RW_WRITER);
2320 2323 error = tmpnode_trunc((struct tmount *)VFSTOTM(vp->v_vfsp),
2321 2324 tp, (ulong_t)lp->l_start);
2322 2325 rw_exit(&tp->tn_contents);
2323 2326 rw_exit(&tp->tn_rwlock);
2324 2327 return (error);
2325 2328 }
2326 2329
2327 2330 /* ARGSUSED */
2328 2331 static int
2329 2332 tmp_space(
2330 2333 struct vnode *vp,
2331 2334 int cmd,
2332 2335 struct flock64 *bfp,
2333 2336 int flag,
2334 2337 offset_t offset,
2335 2338 cred_t *cred,
2336 2339 caller_context_t *ct)
|
↓ open down ↓ |
1016 lines elided |
↑ open up ↑ |
2337 2340 {
2338 2341 int error;
2339 2342
2340 2343 if (cmd != F_FREESP)
2341 2344 return (EINVAL);
2342 2345 if ((error = convoff(vp, bfp, 0, (offset_t)offset)) == 0) {
2343 2346 if ((bfp->l_start > MAXOFF_T) || (bfp->l_len > MAXOFF_T))
2344 2347 return (EFBIG);
2345 2348 error = tmp_freesp(vp, bfp, flag);
2346 2349
2347 - if (error == 0 && bfp->l_start == 0)
2348 - vnevent_truncate(vp, ct);
2350 + if (error == 0) {
2351 + if (bfp->l_start == 0) {
2352 + vnevent_truncate(vp, ct);
2353 + } else {
2354 + vnevent_resize(vp, ct);
2355 + }
2356 + }
2349 2357 }
2350 2358 return (error);
2351 2359 }
2352 2360
2353 2361 /* ARGSUSED */
2354 2362 static int
2355 2363 tmp_seek(
2356 2364 struct vnode *vp,
2357 2365 offset_t ooff,
2358 2366 offset_t *noffp,
2359 2367 caller_context_t *ct)
2360 2368 {
2361 2369 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
2362 2370 }
2363 2371
2364 2372 /* ARGSUSED2 */
2365 2373 static int
2366 2374 tmp_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2367 2375 {
2368 2376 struct tmpnode *tp = VTOTN(vp);
2369 2377
2370 2378 if (write_lock) {
2371 2379 rw_enter(&tp->tn_rwlock, RW_WRITER);
2372 2380 } else {
2373 2381 rw_enter(&tp->tn_rwlock, RW_READER);
2374 2382 }
2375 2383 return (write_lock);
2376 2384 }
2377 2385
2378 2386 /* ARGSUSED1 */
2379 2387 static void
2380 2388 tmp_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2381 2389 {
2382 2390 struct tmpnode *tp = VTOTN(vp);
2383 2391
2384 2392 rw_exit(&tp->tn_rwlock);
2385 2393 }
2386 2394
2387 2395 static int
2388 2396 tmp_pathconf(
2389 2397 struct vnode *vp,
2390 2398 int cmd,
2391 2399 ulong_t *valp,
2392 2400 cred_t *cr,
2393 2401 caller_context_t *ct)
2394 2402 {
2395 2403 struct tmpnode *tp = NULL;
2396 2404 int error;
2397 2405
2398 2406 switch (cmd) {
2399 2407 case _PC_XATTR_EXISTS:
2400 2408 if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
2401 2409 *valp = 0; /* assume no attributes */
2402 2410 error = 0; /* okay to ask */
2403 2411 tp = VTOTN(vp);
2404 2412 rw_enter(&tp->tn_rwlock, RW_READER);
2405 2413 if (tp->tn_xattrdp) {
2406 2414 rw_enter(&tp->tn_xattrdp->tn_rwlock, RW_READER);
2407 2415 /* do not count "." and ".." */
2408 2416 if (tp->tn_xattrdp->tn_dirents > 2)
2409 2417 *valp = 1;
2410 2418 rw_exit(&tp->tn_xattrdp->tn_rwlock);
2411 2419 }
2412 2420 rw_exit(&tp->tn_rwlock);
2413 2421 } else {
2414 2422 error = EINVAL;
2415 2423 }
2416 2424 break;
2417 2425 case _PC_SATTR_ENABLED:
2418 2426 case _PC_SATTR_EXISTS:
2419 2427 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2420 2428 (vp->v_type == VREG || vp->v_type == VDIR);
2421 2429 error = 0;
2422 2430 break;
2423 2431 case _PC_TIMESTAMP_RESOLUTION:
2424 2432 /* nanosecond timestamp resolution */
2425 2433 *valp = 1L;
2426 2434 error = 0;
2427 2435 break;
2428 2436 default:
2429 2437 error = fs_pathconf(vp, cmd, valp, cr, ct);
2430 2438 }
2431 2439 return (error);
2432 2440 }
2433 2441
2434 2442
2435 2443 struct vnodeops *tmp_vnodeops;
2436 2444
2437 2445 const fs_operation_def_t tmp_vnodeops_template[] = {
2438 2446 VOPNAME_OPEN, { .vop_open = tmp_open },
2439 2447 VOPNAME_CLOSE, { .vop_close = tmp_close },
2440 2448 VOPNAME_READ, { .vop_read = tmp_read },
2441 2449 VOPNAME_WRITE, { .vop_write = tmp_write },
2442 2450 VOPNAME_IOCTL, { .vop_ioctl = tmp_ioctl },
2443 2451 VOPNAME_GETATTR, { .vop_getattr = tmp_getattr },
2444 2452 VOPNAME_SETATTR, { .vop_setattr = tmp_setattr },
2445 2453 VOPNAME_ACCESS, { .vop_access = tmp_access },
2446 2454 VOPNAME_LOOKUP, { .vop_lookup = tmp_lookup },
2447 2455 VOPNAME_CREATE, { .vop_create = tmp_create },
2448 2456 VOPNAME_REMOVE, { .vop_remove = tmp_remove },
2449 2457 VOPNAME_LINK, { .vop_link = tmp_link },
2450 2458 VOPNAME_RENAME, { .vop_rename = tmp_rename },
2451 2459 VOPNAME_MKDIR, { .vop_mkdir = tmp_mkdir },
2452 2460 VOPNAME_RMDIR, { .vop_rmdir = tmp_rmdir },
2453 2461 VOPNAME_READDIR, { .vop_readdir = tmp_readdir },
2454 2462 VOPNAME_SYMLINK, { .vop_symlink = tmp_symlink },
2455 2463 VOPNAME_READLINK, { .vop_readlink = tmp_readlink },
2456 2464 VOPNAME_FSYNC, { .vop_fsync = tmp_fsync },
2457 2465 VOPNAME_INACTIVE, { .vop_inactive = tmp_inactive },
2458 2466 VOPNAME_FID, { .vop_fid = tmp_fid },
2459 2467 VOPNAME_RWLOCK, { .vop_rwlock = tmp_rwlock },
2460 2468 VOPNAME_RWUNLOCK, { .vop_rwunlock = tmp_rwunlock },
2461 2469 VOPNAME_SEEK, { .vop_seek = tmp_seek },
2462 2470 VOPNAME_SPACE, { .vop_space = tmp_space },
2463 2471 VOPNAME_GETPAGE, { .vop_getpage = tmp_getpage },
2464 2472 VOPNAME_PUTPAGE, { .vop_putpage = tmp_putpage },
2465 2473 VOPNAME_MAP, { .vop_map = tmp_map },
2466 2474 VOPNAME_ADDMAP, { .vop_addmap = tmp_addmap },
2467 2475 VOPNAME_DELMAP, { .vop_delmap = tmp_delmap },
2468 2476 VOPNAME_PATHCONF, { .vop_pathconf = tmp_pathconf },
2469 2477 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
2470 2478 NULL, NULL
2471 2479 };
|
↓ open down ↓ |
113 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX