Print this page
5042 stop using deprecated atomic functions
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/vnode.c
+++ new/usr/src/uts/common/fs/vnode.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]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
25 25 */
26 26
27 27 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
28 28 /* All Rights Reserved */
29 29
30 30 /*
31 31 * University Copyright- Copyright (c) 1982, 1986, 1988
32 32 * The Regents of the University of California
33 33 * All Rights Reserved
34 34 *
35 35 * University Acknowledgment- Portions of this document are derived from
36 36 * software developed by the University of California, Berkeley, and its
37 37 * contributors.
38 38 */
39 39
40 40 #include <sys/types.h>
41 41 #include <sys/param.h>
42 42 #include <sys/t_lock.h>
43 43 #include <sys/errno.h>
44 44 #include <sys/cred.h>
45 45 #include <sys/user.h>
46 46 #include <sys/uio.h>
47 47 #include <sys/file.h>
48 48 #include <sys/pathname.h>
49 49 #include <sys/vfs.h>
50 50 #include <sys/vfs_opreg.h>
51 51 #include <sys/vnode.h>
52 52 #include <sys/rwstlock.h>
53 53 #include <sys/fem.h>
54 54 #include <sys/stat.h>
55 55 #include <sys/mode.h>
56 56 #include <sys/conf.h>
57 57 #include <sys/sysmacros.h>
58 58 #include <sys/cmn_err.h>
59 59 #include <sys/systm.h>
60 60 #include <sys/kmem.h>
61 61 #include <sys/debug.h>
62 62 #include <c2/audit.h>
63 63 #include <sys/acl.h>
64 64 #include <sys/nbmlock.h>
65 65 #include <sys/fcntl.h>
66 66 #include <fs/fs_subr.h>
67 67 #include <sys/taskq.h>
68 68 #include <fs/fs_reparse.h>
69 69
70 70 /* Determine if this vnode is a file that is read-only */
71 71 #define ISROFILE(vp) \
72 72 ((vp)->v_type != VCHR && (vp)->v_type != VBLK && \
73 73 (vp)->v_type != VFIFO && vn_is_readonly(vp))
74 74
75 75 /* Tunable via /etc/system; used only by admin/install */
76 76 int nfs_global_client_only;
77 77
78 78 /*
79 79 * Array of vopstats_t for per-FS-type vopstats. This array has the same
80 80 * number of entries as and parallel to the vfssw table. (Arguably, it could
81 81 * be part of the vfssw table.) Once it's initialized, it's accessed using
82 82 * the same fstype index that is used to index into the vfssw table.
83 83 */
84 84 vopstats_t **vopstats_fstype;
85 85
86 86 /* vopstats initialization template used for fast initialization via bcopy() */
87 87 static vopstats_t *vs_templatep;
88 88
89 89 /* Kmem cache handle for vsk_anchor_t allocations */
90 90 kmem_cache_t *vsk_anchor_cache;
91 91
92 92 /* file events cleanup routine */
93 93 extern void free_fopdata(vnode_t *);
94 94
95 95 /*
96 96 * Root of AVL tree for the kstats associated with vopstats. Lock protects
97 97 * updates to vsktat_tree.
98 98 */
99 99 avl_tree_t vskstat_tree;
100 100 kmutex_t vskstat_tree_lock;
101 101
102 102 /* Global variable which enables/disables the vopstats collection */
103 103 int vopstats_enabled = 1;
104 104
105 105 /*
106 106 * forward declarations for internal vnode specific data (vsd)
107 107 */
108 108 static void *vsd_realloc(void *, size_t, size_t);
109 109
110 110 /*
111 111 * forward declarations for reparse point functions
112 112 */
113 113 static int fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr);
114 114
115 115 /*
116 116 * VSD -- VNODE SPECIFIC DATA
117 117 * The v_data pointer is typically used by a file system to store a
118 118 * pointer to the file system's private node (e.g. ufs inode, nfs rnode).
119 119 * However, there are times when additional project private data needs
120 120 * to be stored separately from the data (node) pointed to by v_data.
121 121 * This additional data could be stored by the file system itself or
122 122 * by a completely different kernel entity. VSD provides a way for
123 123 * callers to obtain a key and store a pointer to private data associated
124 124 * with a vnode.
125 125 *
126 126 * Callers are responsible for protecting the vsd by holding v_vsd_lock
127 127 * for calls to vsd_set() and vsd_get().
128 128 */
129 129
130 130 /*
131 131 * vsd_lock protects:
132 132 * vsd_nkeys - creation and deletion of vsd keys
133 133 * vsd_list - insertion and deletion of vsd_node in the vsd_list
134 134 * vsd_destructor - adding and removing destructors to the list
135 135 */
136 136 static kmutex_t vsd_lock;
137 137 static uint_t vsd_nkeys; /* size of destructor array */
138 138 /* list of vsd_node's */
139 139 static list_t *vsd_list = NULL;
140 140 /* per-key destructor funcs */
141 141 static void (**vsd_destructor)(void *);
142 142
143 143 /*
144 144 * The following is the common set of actions needed to update the
145 145 * vopstats structure from a vnode op. Both VOPSTATS_UPDATE() and
146 146 * VOPSTATS_UPDATE_IO() do almost the same thing, except for the
147 147 * recording of the bytes transferred. Since the code is similar
148 148 * but small, it is nearly a duplicate. Consequently any changes
149 149 * to one may need to be reflected in the other.
150 150 * Rundown of the variables:
151 151 * vp - Pointer to the vnode
152 152 * counter - Partial name structure member to update in vopstats for counts
153 153 * bytecounter - Partial name structure member to update in vopstats for bytes
154 154 * bytesval - Value to update in vopstats for bytes
155 155 * fstype - Index into vsanchor_fstype[], same as index into vfssw[]
156 156 * vsp - Pointer to vopstats structure (either in vfs or vsanchor_fstype[i])
157 157 */
158 158
159 159 #define VOPSTATS_UPDATE(vp, counter) { \
160 160 vfs_t *vfsp = (vp)->v_vfsp; \
161 161 if (vfsp && vfsp->vfs_implp && \
162 162 (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) { \
163 163 vopstats_t *vsp = &vfsp->vfs_vopstats; \
164 164 uint64_t *stataddr = &(vsp->n##counter.value.ui64); \
165 165 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
166 166 size_t, uint64_t *); \
167 167 __dtrace_probe___fsinfo_##counter(vp, 0, stataddr); \
168 168 (*stataddr)++; \
169 169 if ((vsp = vfsp->vfs_fstypevsp) != NULL) { \
170 170 vsp->n##counter.value.ui64++; \
171 171 } \
172 172 } \
173 173 }
174 174
175 175 #define VOPSTATS_UPDATE_IO(vp, counter, bytecounter, bytesval) { \
176 176 vfs_t *vfsp = (vp)->v_vfsp; \
177 177 if (vfsp && vfsp->vfs_implp && \
178 178 (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) { \
179 179 vopstats_t *vsp = &vfsp->vfs_vopstats; \
180 180 uint64_t *stataddr = &(vsp->n##counter.value.ui64); \
181 181 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
182 182 size_t, uint64_t *); \
183 183 __dtrace_probe___fsinfo_##counter(vp, bytesval, stataddr); \
184 184 (*stataddr)++; \
185 185 vsp->bytecounter.value.ui64 += bytesval; \
186 186 if ((vsp = vfsp->vfs_fstypevsp) != NULL) { \
187 187 vsp->n##counter.value.ui64++; \
188 188 vsp->bytecounter.value.ui64 += bytesval; \
189 189 } \
190 190 } \
191 191 }
192 192
193 193 /*
194 194 * If the filesystem does not support XIDs map credential
195 195 * If the vfsp is NULL, perhaps we should also map?
196 196 */
197 197 #define VOPXID_MAP_CR(vp, cr) { \
198 198 vfs_t *vfsp = (vp)->v_vfsp; \
199 199 if (vfsp != NULL && (vfsp->vfs_flag & VFS_XID) == 0) \
200 200 cr = crgetmapped(cr); \
201 201 }
202 202
203 203 /*
204 204 * Convert stat(2) formats to vnode types and vice versa. (Knows about
205 205 * numerical order of S_IFMT and vnode types.)
206 206 */
207 207 enum vtype iftovt_tab[] = {
208 208 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
209 209 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON
210 210 };
211 211
212 212 ushort_t vttoif_tab[] = {
213 213 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFIFO,
214 214 S_IFDOOR, 0, S_IFSOCK, S_IFPORT, 0
215 215 };
216 216
217 217 /*
218 218 * The system vnode cache.
219 219 */
220 220
221 221 kmem_cache_t *vn_cache;
222 222
223 223
224 224 /*
225 225 * Vnode operations vector.
226 226 */
227 227
228 228 static const fs_operation_trans_def_t vn_ops_table[] = {
229 229 VOPNAME_OPEN, offsetof(struct vnodeops, vop_open),
230 230 fs_nosys, fs_nosys,
231 231
232 232 VOPNAME_CLOSE, offsetof(struct vnodeops, vop_close),
233 233 fs_nosys, fs_nosys,
234 234
235 235 VOPNAME_READ, offsetof(struct vnodeops, vop_read),
236 236 fs_nosys, fs_nosys,
237 237
238 238 VOPNAME_WRITE, offsetof(struct vnodeops, vop_write),
239 239 fs_nosys, fs_nosys,
240 240
241 241 VOPNAME_IOCTL, offsetof(struct vnodeops, vop_ioctl),
242 242 fs_nosys, fs_nosys,
243 243
244 244 VOPNAME_SETFL, offsetof(struct vnodeops, vop_setfl),
245 245 fs_setfl, fs_nosys,
246 246
247 247 VOPNAME_GETATTR, offsetof(struct vnodeops, vop_getattr),
248 248 fs_nosys, fs_nosys,
249 249
250 250 VOPNAME_SETATTR, offsetof(struct vnodeops, vop_setattr),
251 251 fs_nosys, fs_nosys,
252 252
253 253 VOPNAME_ACCESS, offsetof(struct vnodeops, vop_access),
254 254 fs_nosys, fs_nosys,
255 255
256 256 VOPNAME_LOOKUP, offsetof(struct vnodeops, vop_lookup),
257 257 fs_nosys, fs_nosys,
258 258
259 259 VOPNAME_CREATE, offsetof(struct vnodeops, vop_create),
260 260 fs_nosys, fs_nosys,
261 261
262 262 VOPNAME_REMOVE, offsetof(struct vnodeops, vop_remove),
263 263 fs_nosys, fs_nosys,
264 264
265 265 VOPNAME_LINK, offsetof(struct vnodeops, vop_link),
266 266 fs_nosys, fs_nosys,
267 267
268 268 VOPNAME_RENAME, offsetof(struct vnodeops, vop_rename),
269 269 fs_nosys, fs_nosys,
270 270
271 271 VOPNAME_MKDIR, offsetof(struct vnodeops, vop_mkdir),
272 272 fs_nosys, fs_nosys,
273 273
274 274 VOPNAME_RMDIR, offsetof(struct vnodeops, vop_rmdir),
275 275 fs_nosys, fs_nosys,
276 276
277 277 VOPNAME_READDIR, offsetof(struct vnodeops, vop_readdir),
278 278 fs_nosys, fs_nosys,
279 279
280 280 VOPNAME_SYMLINK, offsetof(struct vnodeops, vop_symlink),
281 281 fs_nosys, fs_nosys,
282 282
283 283 VOPNAME_READLINK, offsetof(struct vnodeops, vop_readlink),
284 284 fs_nosys, fs_nosys,
285 285
286 286 VOPNAME_FSYNC, offsetof(struct vnodeops, vop_fsync),
287 287 fs_nosys, fs_nosys,
288 288
289 289 VOPNAME_INACTIVE, offsetof(struct vnodeops, vop_inactive),
290 290 fs_nosys, fs_nosys,
291 291
292 292 VOPNAME_FID, offsetof(struct vnodeops, vop_fid),
293 293 fs_nosys, fs_nosys,
294 294
295 295 VOPNAME_RWLOCK, offsetof(struct vnodeops, vop_rwlock),
296 296 fs_rwlock, fs_rwlock,
297 297
298 298 VOPNAME_RWUNLOCK, offsetof(struct vnodeops, vop_rwunlock),
299 299 (fs_generic_func_p) fs_rwunlock,
300 300 (fs_generic_func_p) fs_rwunlock, /* no errors allowed */
301 301
302 302 VOPNAME_SEEK, offsetof(struct vnodeops, vop_seek),
303 303 fs_nosys, fs_nosys,
304 304
305 305 VOPNAME_CMP, offsetof(struct vnodeops, vop_cmp),
306 306 fs_cmp, fs_cmp, /* no errors allowed */
307 307
308 308 VOPNAME_FRLOCK, offsetof(struct vnodeops, vop_frlock),
309 309 fs_frlock, fs_nosys,
310 310
311 311 VOPNAME_SPACE, offsetof(struct vnodeops, vop_space),
312 312 fs_nosys, fs_nosys,
313 313
314 314 VOPNAME_REALVP, offsetof(struct vnodeops, vop_realvp),
315 315 fs_nosys, fs_nosys,
316 316
317 317 VOPNAME_GETPAGE, offsetof(struct vnodeops, vop_getpage),
318 318 fs_nosys, fs_nosys,
319 319
320 320 VOPNAME_PUTPAGE, offsetof(struct vnodeops, vop_putpage),
321 321 fs_nosys, fs_nosys,
322 322
323 323 VOPNAME_MAP, offsetof(struct vnodeops, vop_map),
324 324 (fs_generic_func_p) fs_nosys_map,
325 325 (fs_generic_func_p) fs_nosys_map,
326 326
327 327 VOPNAME_ADDMAP, offsetof(struct vnodeops, vop_addmap),
328 328 (fs_generic_func_p) fs_nosys_addmap,
329 329 (fs_generic_func_p) fs_nosys_addmap,
330 330
331 331 VOPNAME_DELMAP, offsetof(struct vnodeops, vop_delmap),
332 332 fs_nosys, fs_nosys,
333 333
334 334 VOPNAME_POLL, offsetof(struct vnodeops, vop_poll),
335 335 (fs_generic_func_p) fs_poll, (fs_generic_func_p) fs_nosys_poll,
336 336
337 337 VOPNAME_DUMP, offsetof(struct vnodeops, vop_dump),
338 338 fs_nosys, fs_nosys,
339 339
340 340 VOPNAME_PATHCONF, offsetof(struct vnodeops, vop_pathconf),
341 341 fs_pathconf, fs_nosys,
342 342
343 343 VOPNAME_PAGEIO, offsetof(struct vnodeops, vop_pageio),
344 344 fs_nosys, fs_nosys,
345 345
346 346 VOPNAME_DUMPCTL, offsetof(struct vnodeops, vop_dumpctl),
347 347 fs_nosys, fs_nosys,
348 348
349 349 VOPNAME_DISPOSE, offsetof(struct vnodeops, vop_dispose),
350 350 (fs_generic_func_p) fs_dispose,
351 351 (fs_generic_func_p) fs_nodispose,
352 352
353 353 VOPNAME_SETSECATTR, offsetof(struct vnodeops, vop_setsecattr),
354 354 fs_nosys, fs_nosys,
355 355
356 356 VOPNAME_GETSECATTR, offsetof(struct vnodeops, vop_getsecattr),
357 357 fs_fab_acl, fs_nosys,
358 358
359 359 VOPNAME_SHRLOCK, offsetof(struct vnodeops, vop_shrlock),
360 360 fs_shrlock, fs_nosys,
361 361
362 362 VOPNAME_VNEVENT, offsetof(struct vnodeops, vop_vnevent),
363 363 (fs_generic_func_p) fs_vnevent_nosupport,
364 364 (fs_generic_func_p) fs_vnevent_nosupport,
365 365
366 366 VOPNAME_REQZCBUF, offsetof(struct vnodeops, vop_reqzcbuf),
367 367 fs_nosys, fs_nosys,
368 368
369 369 VOPNAME_RETZCBUF, offsetof(struct vnodeops, vop_retzcbuf),
370 370 fs_nosys, fs_nosys,
371 371
372 372 NULL, 0, NULL, NULL
373 373 };
374 374
375 375 /* Extensible attribute (xva) routines. */
376 376
377 377 /*
378 378 * Zero out the structure, set the size of the requested/returned bitmaps,
379 379 * set AT_XVATTR in the embedded vattr_t's va_mask, and set up the pointer
380 380 * to the returned attributes array.
381 381 */
382 382 void
383 383 xva_init(xvattr_t *xvap)
384 384 {
385 385 bzero(xvap, sizeof (xvattr_t));
386 386 xvap->xva_mapsize = XVA_MAPSIZE;
387 387 xvap->xva_magic = XVA_MAGIC;
388 388 xvap->xva_vattr.va_mask = AT_XVATTR;
389 389 xvap->xva_rtnattrmapp = &(xvap->xva_rtnattrmap)[0];
390 390 }
391 391
392 392 /*
393 393 * If AT_XVATTR is set, returns a pointer to the embedded xoptattr_t
394 394 * structure. Otherwise, returns NULL.
395 395 */
396 396 xoptattr_t *
397 397 xva_getxoptattr(xvattr_t *xvap)
398 398 {
399 399 xoptattr_t *xoap = NULL;
400 400 if (xvap->xva_vattr.va_mask & AT_XVATTR)
401 401 xoap = &xvap->xva_xoptattrs;
402 402 return (xoap);
403 403 }
404 404
405 405 /*
406 406 * Used by the AVL routines to compare two vsk_anchor_t structures in the tree.
407 407 * We use the f_fsid reported by VFS_STATVFS() since we use that for the
408 408 * kstat name.
409 409 */
410 410 static int
411 411 vska_compar(const void *n1, const void *n2)
412 412 {
413 413 int ret;
414 414 ulong_t p1 = ((vsk_anchor_t *)n1)->vsk_fsid;
415 415 ulong_t p2 = ((vsk_anchor_t *)n2)->vsk_fsid;
416 416
417 417 if (p1 < p2) {
418 418 ret = -1;
419 419 } else if (p1 > p2) {
420 420 ret = 1;
421 421 } else {
422 422 ret = 0;
423 423 }
424 424
425 425 return (ret);
426 426 }
427 427
428 428 /*
429 429 * Used to create a single template which will be bcopy()ed to a newly
430 430 * allocated vsanchor_combo_t structure in new_vsanchor(), below.
431 431 */
432 432 static vopstats_t *
433 433 create_vopstats_template()
434 434 {
435 435 vopstats_t *vsp;
436 436
437 437 vsp = kmem_alloc(sizeof (vopstats_t), KM_SLEEP);
438 438 bzero(vsp, sizeof (*vsp)); /* Start fresh */
439 439
440 440 /* VOP_OPEN */
441 441 kstat_named_init(&vsp->nopen, "nopen", KSTAT_DATA_UINT64);
442 442 /* VOP_CLOSE */
443 443 kstat_named_init(&vsp->nclose, "nclose", KSTAT_DATA_UINT64);
444 444 /* VOP_READ I/O */
445 445 kstat_named_init(&vsp->nread, "nread", KSTAT_DATA_UINT64);
446 446 kstat_named_init(&vsp->read_bytes, "read_bytes", KSTAT_DATA_UINT64);
447 447 /* VOP_WRITE I/O */
448 448 kstat_named_init(&vsp->nwrite, "nwrite", KSTAT_DATA_UINT64);
449 449 kstat_named_init(&vsp->write_bytes, "write_bytes", KSTAT_DATA_UINT64);
450 450 /* VOP_IOCTL */
451 451 kstat_named_init(&vsp->nioctl, "nioctl", KSTAT_DATA_UINT64);
452 452 /* VOP_SETFL */
453 453 kstat_named_init(&vsp->nsetfl, "nsetfl", KSTAT_DATA_UINT64);
454 454 /* VOP_GETATTR */
455 455 kstat_named_init(&vsp->ngetattr, "ngetattr", KSTAT_DATA_UINT64);
456 456 /* VOP_SETATTR */
457 457 kstat_named_init(&vsp->nsetattr, "nsetattr", KSTAT_DATA_UINT64);
458 458 /* VOP_ACCESS */
459 459 kstat_named_init(&vsp->naccess, "naccess", KSTAT_DATA_UINT64);
460 460 /* VOP_LOOKUP */
461 461 kstat_named_init(&vsp->nlookup, "nlookup", KSTAT_DATA_UINT64);
462 462 /* VOP_CREATE */
463 463 kstat_named_init(&vsp->ncreate, "ncreate", KSTAT_DATA_UINT64);
464 464 /* VOP_REMOVE */
465 465 kstat_named_init(&vsp->nremove, "nremove", KSTAT_DATA_UINT64);
466 466 /* VOP_LINK */
467 467 kstat_named_init(&vsp->nlink, "nlink", KSTAT_DATA_UINT64);
468 468 /* VOP_RENAME */
469 469 kstat_named_init(&vsp->nrename, "nrename", KSTAT_DATA_UINT64);
470 470 /* VOP_MKDIR */
471 471 kstat_named_init(&vsp->nmkdir, "nmkdir", KSTAT_DATA_UINT64);
472 472 /* VOP_RMDIR */
473 473 kstat_named_init(&vsp->nrmdir, "nrmdir", KSTAT_DATA_UINT64);
474 474 /* VOP_READDIR I/O */
475 475 kstat_named_init(&vsp->nreaddir, "nreaddir", KSTAT_DATA_UINT64);
476 476 kstat_named_init(&vsp->readdir_bytes, "readdir_bytes",
477 477 KSTAT_DATA_UINT64);
478 478 /* VOP_SYMLINK */
479 479 kstat_named_init(&vsp->nsymlink, "nsymlink", KSTAT_DATA_UINT64);
480 480 /* VOP_READLINK */
481 481 kstat_named_init(&vsp->nreadlink, "nreadlink", KSTAT_DATA_UINT64);
482 482 /* VOP_FSYNC */
483 483 kstat_named_init(&vsp->nfsync, "nfsync", KSTAT_DATA_UINT64);
484 484 /* VOP_INACTIVE */
485 485 kstat_named_init(&vsp->ninactive, "ninactive", KSTAT_DATA_UINT64);
486 486 /* VOP_FID */
487 487 kstat_named_init(&vsp->nfid, "nfid", KSTAT_DATA_UINT64);
488 488 /* VOP_RWLOCK */
489 489 kstat_named_init(&vsp->nrwlock, "nrwlock", KSTAT_DATA_UINT64);
490 490 /* VOP_RWUNLOCK */
491 491 kstat_named_init(&vsp->nrwunlock, "nrwunlock", KSTAT_DATA_UINT64);
492 492 /* VOP_SEEK */
493 493 kstat_named_init(&vsp->nseek, "nseek", KSTAT_DATA_UINT64);
494 494 /* VOP_CMP */
495 495 kstat_named_init(&vsp->ncmp, "ncmp", KSTAT_DATA_UINT64);
496 496 /* VOP_FRLOCK */
497 497 kstat_named_init(&vsp->nfrlock, "nfrlock", KSTAT_DATA_UINT64);
498 498 /* VOP_SPACE */
499 499 kstat_named_init(&vsp->nspace, "nspace", KSTAT_DATA_UINT64);
500 500 /* VOP_REALVP */
501 501 kstat_named_init(&vsp->nrealvp, "nrealvp", KSTAT_DATA_UINT64);
502 502 /* VOP_GETPAGE */
503 503 kstat_named_init(&vsp->ngetpage, "ngetpage", KSTAT_DATA_UINT64);
504 504 /* VOP_PUTPAGE */
505 505 kstat_named_init(&vsp->nputpage, "nputpage", KSTAT_DATA_UINT64);
506 506 /* VOP_MAP */
507 507 kstat_named_init(&vsp->nmap, "nmap", KSTAT_DATA_UINT64);
508 508 /* VOP_ADDMAP */
509 509 kstat_named_init(&vsp->naddmap, "naddmap", KSTAT_DATA_UINT64);
510 510 /* VOP_DELMAP */
511 511 kstat_named_init(&vsp->ndelmap, "ndelmap", KSTAT_DATA_UINT64);
512 512 /* VOP_POLL */
513 513 kstat_named_init(&vsp->npoll, "npoll", KSTAT_DATA_UINT64);
514 514 /* VOP_DUMP */
515 515 kstat_named_init(&vsp->ndump, "ndump", KSTAT_DATA_UINT64);
516 516 /* VOP_PATHCONF */
517 517 kstat_named_init(&vsp->npathconf, "npathconf", KSTAT_DATA_UINT64);
518 518 /* VOP_PAGEIO */
519 519 kstat_named_init(&vsp->npageio, "npageio", KSTAT_DATA_UINT64);
520 520 /* VOP_DUMPCTL */
521 521 kstat_named_init(&vsp->ndumpctl, "ndumpctl", KSTAT_DATA_UINT64);
522 522 /* VOP_DISPOSE */
523 523 kstat_named_init(&vsp->ndispose, "ndispose", KSTAT_DATA_UINT64);
524 524 /* VOP_SETSECATTR */
525 525 kstat_named_init(&vsp->nsetsecattr, "nsetsecattr", KSTAT_DATA_UINT64);
526 526 /* VOP_GETSECATTR */
527 527 kstat_named_init(&vsp->ngetsecattr, "ngetsecattr", KSTAT_DATA_UINT64);
528 528 /* VOP_SHRLOCK */
529 529 kstat_named_init(&vsp->nshrlock, "nshrlock", KSTAT_DATA_UINT64);
530 530 /* VOP_VNEVENT */
531 531 kstat_named_init(&vsp->nvnevent, "nvnevent", KSTAT_DATA_UINT64);
532 532 /* VOP_REQZCBUF */
533 533 kstat_named_init(&vsp->nreqzcbuf, "nreqzcbuf", KSTAT_DATA_UINT64);
534 534 /* VOP_RETZCBUF */
535 535 kstat_named_init(&vsp->nretzcbuf, "nretzcbuf", KSTAT_DATA_UINT64);
536 536
537 537 return (vsp);
538 538 }
539 539
540 540 /*
541 541 * Creates a kstat structure associated with a vopstats structure.
542 542 */
543 543 kstat_t *
544 544 new_vskstat(char *ksname, vopstats_t *vsp)
545 545 {
546 546 kstat_t *ksp;
547 547
548 548 if (!vopstats_enabled) {
549 549 return (NULL);
550 550 }
551 551
552 552 ksp = kstat_create("unix", 0, ksname, "misc", KSTAT_TYPE_NAMED,
553 553 sizeof (vopstats_t)/sizeof (kstat_named_t),
554 554 KSTAT_FLAG_VIRTUAL|KSTAT_FLAG_WRITABLE);
555 555 if (ksp) {
556 556 ksp->ks_data = vsp;
557 557 kstat_install(ksp);
558 558 }
559 559
560 560 return (ksp);
561 561 }
562 562
563 563 /*
564 564 * Called from vfsinit() to initialize the support mechanisms for vopstats
565 565 */
566 566 void
567 567 vopstats_startup()
568 568 {
569 569 if (!vopstats_enabled)
570 570 return;
571 571
572 572 /*
573 573 * Creates the AVL tree which holds per-vfs vopstat anchors. This
574 574 * is necessary since we need to check if a kstat exists before we
575 575 * attempt to create it. Also, initialize its lock.
576 576 */
577 577 avl_create(&vskstat_tree, vska_compar, sizeof (vsk_anchor_t),
578 578 offsetof(vsk_anchor_t, vsk_node));
579 579 mutex_init(&vskstat_tree_lock, NULL, MUTEX_DEFAULT, NULL);
580 580
581 581 vsk_anchor_cache = kmem_cache_create("vsk_anchor_cache",
582 582 sizeof (vsk_anchor_t), sizeof (uintptr_t), NULL, NULL, NULL,
583 583 NULL, NULL, 0);
584 584
585 585 /*
586 586 * Set up the array of pointers for the vopstats-by-FS-type.
587 587 * The entries will be allocated/initialized as each file system
588 588 * goes through modload/mod_installfs.
589 589 */
590 590 vopstats_fstype = (vopstats_t **)kmem_zalloc(
591 591 (sizeof (vopstats_t *) * nfstype), KM_SLEEP);
592 592
593 593 /* Set up the global vopstats initialization template */
594 594 vs_templatep = create_vopstats_template();
595 595 }
596 596
597 597 /*
598 598 * We need to have the all of the counters zeroed.
599 599 * The initialization of the vopstats_t includes on the order of
600 600 * 50 calls to kstat_named_init(). Rather that do that on every call,
601 601 * we do it once in a template (vs_templatep) then bcopy it over.
602 602 */
603 603 void
604 604 initialize_vopstats(vopstats_t *vsp)
605 605 {
606 606 if (vsp == NULL)
607 607 return;
608 608
609 609 bcopy(vs_templatep, vsp, sizeof (vopstats_t));
610 610 }
611 611
612 612 /*
613 613 * If possible, determine which vopstats by fstype to use and
614 614 * return a pointer to the caller.
615 615 */
616 616 vopstats_t *
617 617 get_fstype_vopstats(vfs_t *vfsp, struct vfssw *vswp)
618 618 {
619 619 int fstype = 0; /* Index into vfssw[] */
620 620 vopstats_t *vsp = NULL;
621 621
622 622 if (vfsp == NULL || (vfsp->vfs_flag & VFS_STATS) == 0 ||
623 623 !vopstats_enabled)
624 624 return (NULL);
625 625 /*
626 626 * Set up the fstype. We go to so much trouble because all versions
627 627 * of NFS use the same fstype in their vfs even though they have
628 628 * distinct entries in the vfssw[] table.
629 629 * NOTE: A special vfs (e.g., EIO_vfs) may not have an entry.
630 630 */
631 631 if (vswp) {
632 632 fstype = vswp - vfssw; /* Gets us the index */
633 633 } else {
634 634 fstype = vfsp->vfs_fstype;
635 635 }
636 636
637 637 /*
638 638 * Point to the per-fstype vopstats. The only valid values are
639 639 * non-zero positive values less than the number of vfssw[] table
640 640 * entries.
641 641 */
642 642 if (fstype > 0 && fstype < nfstype) {
643 643 vsp = vopstats_fstype[fstype];
644 644 }
645 645
646 646 return (vsp);
647 647 }
648 648
649 649 /*
650 650 * Generate a kstat name, create the kstat structure, and allocate a
651 651 * vsk_anchor_t to hold it together. Return the pointer to the vsk_anchor_t
652 652 * to the caller. This must only be called from a mount.
653 653 */
654 654 vsk_anchor_t *
655 655 get_vskstat_anchor(vfs_t *vfsp)
656 656 {
657 657 char kstatstr[KSTAT_STRLEN]; /* kstat name for vopstats */
658 658 statvfs64_t statvfsbuf; /* Needed to find f_fsid */
659 659 vsk_anchor_t *vskp = NULL; /* vfs <--> kstat anchor */
660 660 kstat_t *ksp; /* Ptr to new kstat */
661 661 avl_index_t where; /* Location in the AVL tree */
662 662
663 663 if (vfsp == NULL || vfsp->vfs_implp == NULL ||
664 664 (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled)
665 665 return (NULL);
666 666
667 667 /* Need to get the fsid to build a kstat name */
668 668 if (VFS_STATVFS(vfsp, &statvfsbuf) == 0) {
669 669 /* Create a name for our kstats based on fsid */
670 670 (void) snprintf(kstatstr, KSTAT_STRLEN, "%s%lx",
671 671 VOPSTATS_STR, statvfsbuf.f_fsid);
672 672
673 673 /* Allocate and initialize the vsk_anchor_t */
674 674 vskp = kmem_cache_alloc(vsk_anchor_cache, KM_SLEEP);
675 675 bzero(vskp, sizeof (*vskp));
676 676 vskp->vsk_fsid = statvfsbuf.f_fsid;
677 677
678 678 mutex_enter(&vskstat_tree_lock);
679 679 if (avl_find(&vskstat_tree, vskp, &where) == NULL) {
680 680 avl_insert(&vskstat_tree, vskp, where);
681 681 mutex_exit(&vskstat_tree_lock);
682 682
683 683 /*
684 684 * Now that we've got the anchor in the AVL
685 685 * tree, we can create the kstat.
686 686 */
687 687 ksp = new_vskstat(kstatstr, &vfsp->vfs_vopstats);
688 688 if (ksp) {
689 689 vskp->vsk_ksp = ksp;
690 690 }
691 691 } else {
692 692 /* Oops, found one! Release memory and lock. */
693 693 mutex_exit(&vskstat_tree_lock);
694 694 kmem_cache_free(vsk_anchor_cache, vskp);
695 695 vskp = NULL;
696 696 }
697 697 }
698 698 return (vskp);
699 699 }
700 700
701 701 /*
702 702 * We're in the process of tearing down the vfs and need to cleanup
703 703 * the data structures associated with the vopstats. Must only be called
704 704 * from dounmount().
705 705 */
706 706 void
707 707 teardown_vopstats(vfs_t *vfsp)
708 708 {
709 709 vsk_anchor_t *vskap;
710 710 avl_index_t where;
711 711
712 712 if (vfsp == NULL || vfsp->vfs_implp == NULL ||
713 713 (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled)
714 714 return;
715 715
716 716 /* This is a safe check since VFS_STATS must be set (see above) */
717 717 if ((vskap = vfsp->vfs_vskap) == NULL)
718 718 return;
719 719
720 720 /* Whack the pointer right away */
721 721 vfsp->vfs_vskap = NULL;
722 722
723 723 /* Lock the tree, remove the node, and delete the kstat */
724 724 mutex_enter(&vskstat_tree_lock);
725 725 if (avl_find(&vskstat_tree, vskap, &where)) {
726 726 avl_remove(&vskstat_tree, vskap);
727 727 }
728 728
729 729 if (vskap->vsk_ksp) {
730 730 kstat_delete(vskap->vsk_ksp);
731 731 }
732 732 mutex_exit(&vskstat_tree_lock);
733 733
734 734 kmem_cache_free(vsk_anchor_cache, vskap);
735 735 }
736 736
737 737 /*
738 738 * Read or write a vnode. Called from kernel code.
739 739 */
740 740 int
741 741 vn_rdwr(
742 742 enum uio_rw rw,
743 743 struct vnode *vp,
744 744 caddr_t base,
745 745 ssize_t len,
746 746 offset_t offset,
747 747 enum uio_seg seg,
748 748 int ioflag,
749 749 rlim64_t ulimit, /* meaningful only if rw is UIO_WRITE */
750 750 cred_t *cr,
751 751 ssize_t *residp)
752 752 {
753 753 struct uio uio;
754 754 struct iovec iov;
755 755 int error;
756 756 int in_crit = 0;
757 757
758 758 if (rw == UIO_WRITE && ISROFILE(vp))
759 759 return (EROFS);
760 760
761 761 if (len < 0)
762 762 return (EIO);
763 763
764 764 VOPXID_MAP_CR(vp, cr);
765 765
766 766 iov.iov_base = base;
767 767 iov.iov_len = len;
768 768 uio.uio_iov = &iov;
769 769 uio.uio_iovcnt = 1;
770 770 uio.uio_loffset = offset;
771 771 uio.uio_segflg = (short)seg;
772 772 uio.uio_resid = len;
773 773 uio.uio_llimit = ulimit;
774 774
775 775 /*
776 776 * We have to enter the critical region before calling VOP_RWLOCK
777 777 * to avoid a deadlock with ufs.
778 778 */
779 779 if (nbl_need_check(vp)) {
780 780 int svmand;
781 781
782 782 nbl_start_crit(vp, RW_READER);
783 783 in_crit = 1;
784 784 error = nbl_svmand(vp, cr, &svmand);
785 785 if (error != 0)
786 786 goto done;
787 787 if (nbl_conflict(vp, rw == UIO_WRITE ? NBL_WRITE : NBL_READ,
788 788 uio.uio_offset, uio.uio_resid, svmand, NULL)) {
789 789 error = EACCES;
790 790 goto done;
791 791 }
792 792 }
793 793
794 794 (void) VOP_RWLOCK(vp,
795 795 rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL);
796 796 if (rw == UIO_WRITE) {
797 797 uio.uio_fmode = FWRITE;
798 798 uio.uio_extflg = UIO_COPY_DEFAULT;
799 799 error = VOP_WRITE(vp, &uio, ioflag, cr, NULL);
800 800 } else {
801 801 uio.uio_fmode = FREAD;
802 802 uio.uio_extflg = UIO_COPY_CACHED;
803 803 error = VOP_READ(vp, &uio, ioflag, cr, NULL);
804 804 }
805 805 VOP_RWUNLOCK(vp,
806 806 rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL);
807 807 if (residp)
808 808 *residp = uio.uio_resid;
809 809 else if (uio.uio_resid)
810 810 error = EIO;
811 811
812 812 done:
813 813 if (in_crit)
814 814 nbl_end_crit(vp);
815 815 return (error);
816 816 }
817 817
818 818 /*
819 819 * Release a vnode. Call VOP_INACTIVE on last reference or
820 820 * decrement reference count.
821 821 *
822 822 * To avoid race conditions, the v_count is left at 1 for
823 823 * the call to VOP_INACTIVE. This prevents another thread
824 824 * from reclaiming and releasing the vnode *before* the
825 825 * VOP_INACTIVE routine has a chance to destroy the vnode.
826 826 * We can't have more than 1 thread calling VOP_INACTIVE
827 827 * on a vnode.
828 828 */
829 829 void
830 830 vn_rele(vnode_t *vp)
831 831 {
832 832 VERIFY(vp->v_count > 0);
833 833 mutex_enter(&vp->v_lock);
834 834 if (vp->v_count == 1) {
835 835 mutex_exit(&vp->v_lock);
836 836 VOP_INACTIVE(vp, CRED(), NULL);
837 837 return;
838 838 }
839 839 vp->v_count--;
840 840 mutex_exit(&vp->v_lock);
841 841 }
842 842
843 843 /*
844 844 * Release a vnode referenced by the DNLC. Multiple DNLC references are treated
845 845 * as a single reference, so v_count is not decremented until the last DNLC hold
846 846 * is released. This makes it possible to distinguish vnodes that are referenced
847 847 * only by the DNLC.
848 848 */
849 849 void
850 850 vn_rele_dnlc(vnode_t *vp)
851 851 {
852 852 VERIFY((vp->v_count > 0) && (vp->v_count_dnlc > 0));
853 853 mutex_enter(&vp->v_lock);
854 854 if (--vp->v_count_dnlc == 0) {
855 855 if (vp->v_count == 1) {
856 856 mutex_exit(&vp->v_lock);
857 857 VOP_INACTIVE(vp, CRED(), NULL);
858 858 return;
859 859 }
860 860 vp->v_count--;
861 861 }
862 862 mutex_exit(&vp->v_lock);
863 863 }
864 864
865 865 /*
866 866 * Like vn_rele() except that it clears v_stream under v_lock.
867 867 * This is used by sockfs when it dismantels the association between
868 868 * the sockfs node and the vnode in the underlaying file system.
869 869 * v_lock has to be held to prevent a thread coming through the lookupname
870 870 * path from accessing a stream head that is going away.
871 871 */
872 872 void
873 873 vn_rele_stream(vnode_t *vp)
874 874 {
875 875 VERIFY(vp->v_count > 0);
876 876 mutex_enter(&vp->v_lock);
877 877 vp->v_stream = NULL;
878 878 if (vp->v_count == 1) {
879 879 mutex_exit(&vp->v_lock);
880 880 VOP_INACTIVE(vp, CRED(), NULL);
881 881 return;
882 882 }
883 883 vp->v_count--;
884 884 mutex_exit(&vp->v_lock);
885 885 }
886 886
887 887 static void
888 888 vn_rele_inactive(vnode_t *vp)
889 889 {
890 890 VOP_INACTIVE(vp, CRED(), NULL);
891 891 }
892 892
893 893 /*
894 894 * Like vn_rele() except if we are going to call VOP_INACTIVE() then do it
895 895 * asynchronously using a taskq. This can avoid deadlocks caused by re-entering
896 896 * the file system as a result of releasing the vnode. Note, file systems
897 897 * already have to handle the race where the vnode is incremented before the
898 898 * inactive routine is called and does its locking.
899 899 *
900 900 * Warning: Excessive use of this routine can lead to performance problems.
901 901 * This is because taskqs throttle back allocation if too many are created.
902 902 */
903 903 void
904 904 vn_rele_async(vnode_t *vp, taskq_t *taskq)
905 905 {
906 906 VERIFY(vp->v_count > 0);
907 907 mutex_enter(&vp->v_lock);
908 908 if (vp->v_count == 1) {
909 909 mutex_exit(&vp->v_lock);
910 910 VERIFY(taskq_dispatch(taskq, (task_func_t *)vn_rele_inactive,
911 911 vp, TQ_SLEEP) != NULL);
912 912 return;
913 913 }
914 914 vp->v_count--;
915 915 mutex_exit(&vp->v_lock);
916 916 }
917 917
918 918 int
919 919 vn_open(
920 920 char *pnamep,
921 921 enum uio_seg seg,
922 922 int filemode,
923 923 int createmode,
924 924 struct vnode **vpp,
925 925 enum create crwhy,
926 926 mode_t umask)
927 927 {
928 928 return (vn_openat(pnamep, seg, filemode, createmode, vpp, crwhy,
929 929 umask, NULL, -1));
930 930 }
931 931
932 932
933 933 /*
934 934 * Open/create a vnode.
935 935 * This may be callable by the kernel, the only known use
936 936 * of user context being that the current user credentials
937 937 * are used for permissions. crwhy is defined iff filemode & FCREAT.
938 938 */
939 939 int
940 940 vn_openat(
941 941 char *pnamep,
942 942 enum uio_seg seg,
943 943 int filemode,
944 944 int createmode,
945 945 struct vnode **vpp,
946 946 enum create crwhy,
947 947 mode_t umask,
948 948 struct vnode *startvp,
949 949 int fd)
950 950 {
951 951 struct vnode *vp;
952 952 int mode;
953 953 int accessflags;
954 954 int error;
955 955 int in_crit = 0;
956 956 int open_done = 0;
957 957 int shrlock_done = 0;
958 958 struct vattr vattr;
959 959 enum symfollow follow;
960 960 int estale_retry = 0;
961 961 struct shrlock shr;
962 962 struct shr_locowner shr_own;
963 963
964 964 mode = 0;
965 965 accessflags = 0;
966 966 if (filemode & FREAD)
967 967 mode |= VREAD;
968 968 if (filemode & (FWRITE|FTRUNC))
969 969 mode |= VWRITE;
970 970 if (filemode & (FSEARCH|FEXEC|FXATTRDIROPEN))
971 971 mode |= VEXEC;
972 972
973 973 /* symlink interpretation */
974 974 if (filemode & FNOFOLLOW)
975 975 follow = NO_FOLLOW;
976 976 else
977 977 follow = FOLLOW;
978 978
979 979 if (filemode & FAPPEND)
980 980 accessflags |= V_APPEND;
981 981
982 982 top:
983 983 if (filemode & FCREAT) {
984 984 enum vcexcl excl;
985 985
986 986 /*
987 987 * Wish to create a file.
988 988 */
989 989 vattr.va_type = VREG;
990 990 vattr.va_mode = createmode;
991 991 vattr.va_mask = AT_TYPE|AT_MODE;
992 992 if (filemode & FTRUNC) {
993 993 vattr.va_size = 0;
994 994 vattr.va_mask |= AT_SIZE;
995 995 }
996 996 if (filemode & FEXCL)
997 997 excl = EXCL;
998 998 else
999 999 excl = NONEXCL;
1000 1000
1001 1001 if (error =
1002 1002 vn_createat(pnamep, seg, &vattr, excl, mode, &vp, crwhy,
1003 1003 (filemode & ~(FTRUNC|FEXCL)), umask, startvp))
1004 1004 return (error);
1005 1005 } else {
1006 1006 /*
1007 1007 * Wish to open a file. Just look it up.
1008 1008 */
1009 1009 if (error = lookupnameat(pnamep, seg, follow,
1010 1010 NULLVPP, &vp, startvp)) {
1011 1011 if ((error == ESTALE) &&
1012 1012 fs_need_estale_retry(estale_retry++))
1013 1013 goto top;
1014 1014 return (error);
1015 1015 }
1016 1016
1017 1017 /*
1018 1018 * Get the attributes to check whether file is large.
1019 1019 * We do this only if the FOFFMAX flag is not set and
1020 1020 * only for regular files.
1021 1021 */
1022 1022
1023 1023 if (!(filemode & FOFFMAX) && (vp->v_type == VREG)) {
1024 1024 vattr.va_mask = AT_SIZE;
1025 1025 if ((error = VOP_GETATTR(vp, &vattr, 0,
1026 1026 CRED(), NULL))) {
1027 1027 goto out;
1028 1028 }
1029 1029 if (vattr.va_size > (u_offset_t)MAXOFF32_T) {
1030 1030 /*
1031 1031 * Large File API - regular open fails
1032 1032 * if FOFFMAX flag is set in file mode
1033 1033 */
1034 1034 error = EOVERFLOW;
1035 1035 goto out;
1036 1036 }
1037 1037 }
1038 1038 /*
1039 1039 * Can't write directories, active texts, or
1040 1040 * read-only filesystems. Can't truncate files
1041 1041 * on which mandatory locking is in effect.
1042 1042 */
1043 1043 if (filemode & (FWRITE|FTRUNC)) {
1044 1044 /*
1045 1045 * Allow writable directory if VDIROPEN flag is set.
1046 1046 */
1047 1047 if (vp->v_type == VDIR && !(vp->v_flag & VDIROPEN)) {
1048 1048 error = EISDIR;
1049 1049 goto out;
1050 1050 }
1051 1051 if (ISROFILE(vp)) {
1052 1052 error = EROFS;
1053 1053 goto out;
1054 1054 }
1055 1055 /*
1056 1056 * Can't truncate files on which
1057 1057 * sysv mandatory locking is in effect.
1058 1058 */
1059 1059 if (filemode & FTRUNC) {
1060 1060 vnode_t *rvp;
1061 1061
1062 1062 if (VOP_REALVP(vp, &rvp, NULL) != 0)
1063 1063 rvp = vp;
1064 1064 if (rvp->v_filocks != NULL) {
1065 1065 vattr.va_mask = AT_MODE;
1066 1066 if ((error = VOP_GETATTR(vp,
1067 1067 &vattr, 0, CRED(), NULL)) == 0 &&
1068 1068 MANDLOCK(vp, vattr.va_mode))
1069 1069 error = EAGAIN;
1070 1070 }
1071 1071 }
1072 1072 if (error)
1073 1073 goto out;
1074 1074 }
1075 1075 /*
1076 1076 * Check permissions.
1077 1077 */
1078 1078 if (error = VOP_ACCESS(vp, mode, accessflags, CRED(), NULL))
1079 1079 goto out;
1080 1080 /*
1081 1081 * Require FSEARCH to return a directory.
1082 1082 * Require FEXEC to return a regular file.
1083 1083 */
1084 1084 if ((filemode & FSEARCH) && vp->v_type != VDIR) {
1085 1085 error = ENOTDIR;
1086 1086 goto out;
1087 1087 }
1088 1088 if ((filemode & FEXEC) && vp->v_type != VREG) {
1089 1089 error = ENOEXEC; /* XXX: error code? */
1090 1090 goto out;
1091 1091 }
1092 1092 }
1093 1093
1094 1094 /*
1095 1095 * Do remaining checks for FNOFOLLOW and FNOLINKS.
1096 1096 */
1097 1097 if ((filemode & FNOFOLLOW) && vp->v_type == VLNK) {
1098 1098 error = ELOOP;
1099 1099 goto out;
1100 1100 }
1101 1101 if (filemode & FNOLINKS) {
1102 1102 vattr.va_mask = AT_NLINK;
1103 1103 if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))) {
1104 1104 goto out;
1105 1105 }
1106 1106 if (vattr.va_nlink != 1) {
1107 1107 error = EMLINK;
1108 1108 goto out;
1109 1109 }
1110 1110 }
1111 1111
1112 1112 /*
1113 1113 * Opening a socket corresponding to the AF_UNIX pathname
1114 1114 * in the filesystem name space is not supported.
1115 1115 * However, VSOCK nodes in namefs are supported in order
1116 1116 * to make fattach work for sockets.
1117 1117 *
1118 1118 * XXX This uses VOP_REALVP to distinguish between
1119 1119 * an unopened namefs node (where VOP_REALVP returns a
1120 1120 * different VSOCK vnode) and a VSOCK created by vn_create
1121 1121 * in some file system (where VOP_REALVP would never return
1122 1122 * a different vnode).
1123 1123 */
1124 1124 if (vp->v_type == VSOCK) {
1125 1125 struct vnode *nvp;
1126 1126
1127 1127 error = VOP_REALVP(vp, &nvp, NULL);
1128 1128 if (error != 0 || nvp == NULL || nvp == vp ||
1129 1129 nvp->v_type != VSOCK) {
1130 1130 error = EOPNOTSUPP;
1131 1131 goto out;
1132 1132 }
1133 1133 }
1134 1134
1135 1135 if ((vp->v_type == VREG) && nbl_need_check(vp)) {
1136 1136 /* get share reservation */
1137 1137 shr.s_access = 0;
1138 1138 if (filemode & FWRITE)
1139 1139 shr.s_access |= F_WRACC;
1140 1140 if (filemode & FREAD)
1141 1141 shr.s_access |= F_RDACC;
1142 1142 shr.s_deny = 0;
1143 1143 shr.s_sysid = 0;
1144 1144 shr.s_pid = ttoproc(curthread)->p_pid;
1145 1145 shr_own.sl_pid = shr.s_pid;
1146 1146 shr_own.sl_id = fd;
1147 1147 shr.s_own_len = sizeof (shr_own);
1148 1148 shr.s_owner = (caddr_t)&shr_own;
1149 1149 error = VOP_SHRLOCK(vp, F_SHARE_NBMAND, &shr, filemode, CRED(),
1150 1150 NULL);
1151 1151 if (error)
1152 1152 goto out;
1153 1153 shrlock_done = 1;
1154 1154
1155 1155 /* nbmand conflict check if truncating file */
1156 1156 if ((filemode & FTRUNC) && !(filemode & FCREAT)) {
1157 1157 nbl_start_crit(vp, RW_READER);
1158 1158 in_crit = 1;
1159 1159
1160 1160 vattr.va_mask = AT_SIZE;
1161 1161 if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))
1162 1162 goto out;
1163 1163 if (nbl_conflict(vp, NBL_WRITE, 0, vattr.va_size, 0,
1164 1164 NULL)) {
1165 1165 error = EACCES;
1166 1166 goto out;
1167 1167 }
1168 1168 }
1169 1169 }
1170 1170
1171 1171 /*
1172 1172 * Do opening protocol.
1173 1173 */
1174 1174 error = VOP_OPEN(&vp, filemode, CRED(), NULL);
1175 1175 if (error)
1176 1176 goto out;
1177 1177 open_done = 1;
1178 1178
1179 1179 /*
1180 1180 * Truncate if required.
1181 1181 */
1182 1182 if ((filemode & FTRUNC) && !(filemode & FCREAT)) {
1183 1183 vattr.va_size = 0;
1184 1184 vattr.va_mask = AT_SIZE;
1185 1185 if ((error = VOP_SETATTR(vp, &vattr, 0, CRED(), NULL)) != 0)
1186 1186 goto out;
1187 1187 }
1188 1188 out:
1189 1189 ASSERT(vp->v_count > 0);
1190 1190
1191 1191 if (in_crit) {
1192 1192 nbl_end_crit(vp);
1193 1193 in_crit = 0;
1194 1194 }
1195 1195 if (error) {
1196 1196 if (open_done) {
1197 1197 (void) VOP_CLOSE(vp, filemode, 1, (offset_t)0, CRED(),
1198 1198 NULL);
1199 1199 open_done = 0;
1200 1200 shrlock_done = 0;
1201 1201 }
1202 1202 if (shrlock_done) {
1203 1203 (void) VOP_SHRLOCK(vp, F_UNSHARE, &shr, 0, CRED(),
1204 1204 NULL);
1205 1205 shrlock_done = 0;
1206 1206 }
1207 1207
1208 1208 /*
1209 1209 * The following clause was added to handle a problem
1210 1210 * with NFS consistency. It is possible that a lookup
1211 1211 * of the file to be opened succeeded, but the file
1212 1212 * itself doesn't actually exist on the server. This
1213 1213 * is chiefly due to the DNLC containing an entry for
1214 1214 * the file which has been removed on the server. In
1215 1215 * this case, we just start over. If there was some
1216 1216 * other cause for the ESTALE error, then the lookup
1217 1217 * of the file will fail and the error will be returned
1218 1218 * above instead of looping around from here.
1219 1219 */
1220 1220 VN_RELE(vp);
1221 1221 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1222 1222 goto top;
1223 1223 } else
1224 1224 *vpp = vp;
1225 1225 return (error);
1226 1226 }
1227 1227
1228 1228 /*
1229 1229 * The following two accessor functions are for the NFSv4 server. Since there
1230 1230 * is no VOP_OPEN_UP/DOWNGRADE we need a way for the NFS server to keep the
1231 1231 * vnode open counts correct when a client "upgrades" an open or does an
1232 1232 * open_downgrade. In NFS, an upgrade or downgrade can not only change the
1233 1233 * open mode (add or subtract read or write), but also change the share/deny
1234 1234 * modes. However, share reservations are not integrated with OPEN, yet, so
1235 1235 * we need to handle each separately. These functions are cleaner than having
1236 1236 * the NFS server manipulate the counts directly, however, nobody else should
1237 1237 * use these functions.
1238 1238 */
1239 1239 void
1240 1240 vn_open_upgrade(
1241 1241 vnode_t *vp,
1242 1242 int filemode)
1243 1243 {
1244 1244 ASSERT(vp->v_type == VREG);
1245 1245
1246 1246 if (filemode & FREAD)
1247 1247 atomic_add_32(&(vp->v_rdcnt), 1);
1248 1248 if (filemode & FWRITE)
1249 1249 atomic_add_32(&(vp->v_wrcnt), 1);
1250 1250
1251 1251 }
1252 1252
1253 1253 void
1254 1254 vn_open_downgrade(
1255 1255 vnode_t *vp,
1256 1256 int filemode)
1257 1257 {
1258 1258 ASSERT(vp->v_type == VREG);
1259 1259
1260 1260 if (filemode & FREAD) {
1261 1261 ASSERT(vp->v_rdcnt > 0);
1262 1262 atomic_add_32(&(vp->v_rdcnt), -1);
1263 1263 }
1264 1264 if (filemode & FWRITE) {
1265 1265 ASSERT(vp->v_wrcnt > 0);
1266 1266 atomic_add_32(&(vp->v_wrcnt), -1);
1267 1267 }
1268 1268
1269 1269 }
1270 1270
1271 1271 int
1272 1272 vn_create(
1273 1273 char *pnamep,
1274 1274 enum uio_seg seg,
1275 1275 struct vattr *vap,
1276 1276 enum vcexcl excl,
1277 1277 int mode,
1278 1278 struct vnode **vpp,
1279 1279 enum create why,
1280 1280 int flag,
1281 1281 mode_t umask)
1282 1282 {
1283 1283 return (vn_createat(pnamep, seg, vap, excl, mode, vpp, why, flag,
1284 1284 umask, NULL));
1285 1285 }
1286 1286
1287 1287 /*
1288 1288 * Create a vnode (makenode).
1289 1289 */
1290 1290 int
1291 1291 vn_createat(
1292 1292 char *pnamep,
1293 1293 enum uio_seg seg,
1294 1294 struct vattr *vap,
1295 1295 enum vcexcl excl,
1296 1296 int mode,
1297 1297 struct vnode **vpp,
1298 1298 enum create why,
1299 1299 int flag,
1300 1300 mode_t umask,
1301 1301 struct vnode *startvp)
1302 1302 {
1303 1303 struct vnode *dvp; /* ptr to parent dir vnode */
1304 1304 struct vnode *vp = NULL;
1305 1305 struct pathname pn;
1306 1306 int error;
1307 1307 int in_crit = 0;
1308 1308 struct vattr vattr;
1309 1309 enum symfollow follow;
1310 1310 int estale_retry = 0;
1311 1311 uint32_t auditing = AU_AUDITING();
1312 1312
1313 1313 ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
1314 1314
1315 1315 /* symlink interpretation */
1316 1316 if ((flag & FNOFOLLOW) || excl == EXCL)
1317 1317 follow = NO_FOLLOW;
1318 1318 else
1319 1319 follow = FOLLOW;
1320 1320 flag &= ~(FNOFOLLOW|FNOLINKS);
1321 1321
1322 1322 top:
1323 1323 /*
1324 1324 * Lookup directory.
1325 1325 * If new object is a file, call lower level to create it.
1326 1326 * Note that it is up to the lower level to enforce exclusive
1327 1327 * creation, if the file is already there.
1328 1328 * This allows the lower level to do whatever
1329 1329 * locking or protocol that is needed to prevent races.
1330 1330 * If the new object is directory call lower level to make
1331 1331 * the new directory, with "." and "..".
1332 1332 */
1333 1333 if (error = pn_get(pnamep, seg, &pn))
1334 1334 return (error);
1335 1335 if (auditing)
1336 1336 audit_vncreate_start();
1337 1337 dvp = NULL;
1338 1338 *vpp = NULL;
1339 1339 /*
1340 1340 * lookup will find the parent directory for the vnode.
1341 1341 * When it is done the pn holds the name of the entry
1342 1342 * in the directory.
1343 1343 * If this is a non-exclusive create we also find the node itself.
1344 1344 */
1345 1345 error = lookuppnat(&pn, NULL, follow, &dvp,
1346 1346 (excl == EXCL) ? NULLVPP : vpp, startvp);
1347 1347 if (error) {
1348 1348 pn_free(&pn);
1349 1349 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1350 1350 goto top;
1351 1351 if (why == CRMKDIR && error == EINVAL)
1352 1352 error = EEXIST; /* SVID */
1353 1353 return (error);
1354 1354 }
1355 1355
1356 1356 if (why != CRMKNOD)
1357 1357 vap->va_mode &= ~VSVTX;
1358 1358
1359 1359 /*
1360 1360 * If default ACLs are defined for the directory don't apply the
1361 1361 * umask if umask is passed.
1362 1362 */
1363 1363
1364 1364 if (umask) {
1365 1365
1366 1366 vsecattr_t vsec;
1367 1367
1368 1368 vsec.vsa_aclcnt = 0;
1369 1369 vsec.vsa_aclentp = NULL;
1370 1370 vsec.vsa_dfaclcnt = 0;
1371 1371 vsec.vsa_dfaclentp = NULL;
1372 1372 vsec.vsa_mask = VSA_DFACLCNT;
1373 1373 error = VOP_GETSECATTR(dvp, &vsec, 0, CRED(), NULL);
1374 1374 /*
1375 1375 * If error is ENOSYS then treat it as no error
1376 1376 * Don't want to force all file systems to support
1377 1377 * aclent_t style of ACL's.
1378 1378 */
1379 1379 if (error == ENOSYS)
1380 1380 error = 0;
1381 1381 if (error) {
1382 1382 if (*vpp != NULL)
1383 1383 VN_RELE(*vpp);
1384 1384 goto out;
1385 1385 } else {
1386 1386 /*
1387 1387 * Apply the umask if no default ACLs.
1388 1388 */
1389 1389 if (vsec.vsa_dfaclcnt == 0)
1390 1390 vap->va_mode &= ~umask;
1391 1391
1392 1392 /*
1393 1393 * VOP_GETSECATTR() may have allocated memory for
1394 1394 * ACLs we didn't request, so double-check and
1395 1395 * free it if necessary.
1396 1396 */
1397 1397 if (vsec.vsa_aclcnt && vsec.vsa_aclentp != NULL)
1398 1398 kmem_free((caddr_t)vsec.vsa_aclentp,
1399 1399 vsec.vsa_aclcnt * sizeof (aclent_t));
1400 1400 if (vsec.vsa_dfaclcnt && vsec.vsa_dfaclentp != NULL)
1401 1401 kmem_free((caddr_t)vsec.vsa_dfaclentp,
1402 1402 vsec.vsa_dfaclcnt * sizeof (aclent_t));
1403 1403 }
1404 1404 }
1405 1405
1406 1406 /*
1407 1407 * In general we want to generate EROFS if the file system is
1408 1408 * readonly. However, POSIX (IEEE Std. 1003.1) section 5.3.1
1409 1409 * documents the open system call, and it says that O_CREAT has no
1410 1410 * effect if the file already exists. Bug 1119649 states
1411 1411 * that open(path, O_CREAT, ...) fails when attempting to open an
1412 1412 * existing file on a read only file system. Thus, the first part
1413 1413 * of the following if statement has 3 checks:
1414 1414 * if the file exists &&
1415 1415 * it is being open with write access &&
1416 1416 * the file system is read only
1417 1417 * then generate EROFS
1418 1418 */
1419 1419 if ((*vpp != NULL && (mode & VWRITE) && ISROFILE(*vpp)) ||
1420 1420 (*vpp == NULL && dvp->v_vfsp->vfs_flag & VFS_RDONLY)) {
1421 1421 if (*vpp)
1422 1422 VN_RELE(*vpp);
1423 1423 error = EROFS;
1424 1424 } else if (excl == NONEXCL && *vpp != NULL) {
1425 1425 vnode_t *rvp;
1426 1426
1427 1427 /*
1428 1428 * File already exists. If a mandatory lock has been
1429 1429 * applied, return error.
1430 1430 */
1431 1431 vp = *vpp;
1432 1432 if (VOP_REALVP(vp, &rvp, NULL) != 0)
1433 1433 rvp = vp;
1434 1434 if ((vap->va_mask & AT_SIZE) && nbl_need_check(vp)) {
1435 1435 nbl_start_crit(vp, RW_READER);
1436 1436 in_crit = 1;
1437 1437 }
1438 1438 if (rvp->v_filocks != NULL || rvp->v_shrlocks != NULL) {
1439 1439 vattr.va_mask = AT_MODE|AT_SIZE;
1440 1440 if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) {
1441 1441 goto out;
1442 1442 }
1443 1443 if (MANDLOCK(vp, vattr.va_mode)) {
1444 1444 error = EAGAIN;
1445 1445 goto out;
1446 1446 }
1447 1447 /*
1448 1448 * File cannot be truncated if non-blocking mandatory
1449 1449 * locks are currently on the file.
1450 1450 */
1451 1451 if ((vap->va_mask & AT_SIZE) && in_crit) {
1452 1452 u_offset_t offset;
1453 1453 ssize_t length;
1454 1454
1455 1455 offset = vap->va_size > vattr.va_size ?
1456 1456 vattr.va_size : vap->va_size;
1457 1457 length = vap->va_size > vattr.va_size ?
1458 1458 vap->va_size - vattr.va_size :
1459 1459 vattr.va_size - vap->va_size;
1460 1460 if (nbl_conflict(vp, NBL_WRITE, offset,
1461 1461 length, 0, NULL)) {
1462 1462 error = EACCES;
1463 1463 goto out;
1464 1464 }
1465 1465 }
1466 1466 }
1467 1467
1468 1468 /*
1469 1469 * If the file is the root of a VFS, we've crossed a
1470 1470 * mount point and the "containing" directory that we
1471 1471 * acquired above (dvp) is irrelevant because it's in
1472 1472 * a different file system. We apply VOP_CREATE to the
1473 1473 * target itself instead of to the containing directory
1474 1474 * and supply a null path name to indicate (conventionally)
1475 1475 * the node itself as the "component" of interest.
1476 1476 *
1477 1477 * The intercession of the file system is necessary to
1478 1478 * ensure that the appropriate permission checks are
1479 1479 * done.
1480 1480 */
1481 1481 if (vp->v_flag & VROOT) {
1482 1482 ASSERT(why != CRMKDIR);
1483 1483 error = VOP_CREATE(vp, "", vap, excl, mode, vpp,
1484 1484 CRED(), flag, NULL, NULL);
1485 1485 /*
1486 1486 * If the create succeeded, it will have created
1487 1487 * a new reference to the vnode. Give up the
1488 1488 * original reference. The assertion should not
1489 1489 * get triggered because NBMAND locks only apply to
1490 1490 * VREG files. And if in_crit is non-zero for some
1491 1491 * reason, detect that here, rather than when we
1492 1492 * deference a null vp.
1493 1493 */
1494 1494 ASSERT(in_crit == 0);
1495 1495 VN_RELE(vp);
1496 1496 vp = NULL;
1497 1497 goto out;
1498 1498 }
1499 1499
1500 1500 /*
1501 1501 * Large File API - non-large open (FOFFMAX flag not set)
1502 1502 * of regular file fails if the file size exceeds MAXOFF32_T.
1503 1503 */
1504 1504 if (why != CRMKDIR &&
1505 1505 !(flag & FOFFMAX) &&
1506 1506 (vp->v_type == VREG)) {
1507 1507 vattr.va_mask = AT_SIZE;
1508 1508 if ((error = VOP_GETATTR(vp, &vattr, 0,
1509 1509 CRED(), NULL))) {
1510 1510 goto out;
1511 1511 }
1512 1512 if ((vattr.va_size > (u_offset_t)MAXOFF32_T)) {
1513 1513 error = EOVERFLOW;
1514 1514 goto out;
1515 1515 }
1516 1516 }
1517 1517 }
1518 1518
1519 1519 if (error == 0) {
1520 1520 /*
1521 1521 * Call mkdir() if specified, otherwise create().
1522 1522 */
1523 1523 int must_be_dir = pn_fixslash(&pn); /* trailing '/'? */
1524 1524
1525 1525 if (why == CRMKDIR)
1526 1526 /*
1527 1527 * N.B., if vn_createat() ever requests
1528 1528 * case-insensitive behavior then it will need
1529 1529 * to be passed to VOP_MKDIR(). VOP_CREATE()
1530 1530 * will already get it via "flag"
1531 1531 */
1532 1532 error = VOP_MKDIR(dvp, pn.pn_path, vap, vpp, CRED(),
1533 1533 NULL, 0, NULL);
1534 1534 else if (!must_be_dir)
1535 1535 error = VOP_CREATE(dvp, pn.pn_path, vap,
1536 1536 excl, mode, vpp, CRED(), flag, NULL, NULL);
1537 1537 else
1538 1538 error = ENOTDIR;
1539 1539 }
1540 1540
1541 1541 out:
1542 1542
1543 1543 if (auditing)
1544 1544 audit_vncreate_finish(*vpp, error);
1545 1545 if (in_crit) {
1546 1546 nbl_end_crit(vp);
1547 1547 in_crit = 0;
1548 1548 }
1549 1549 if (vp != NULL) {
1550 1550 VN_RELE(vp);
1551 1551 vp = NULL;
1552 1552 }
1553 1553 pn_free(&pn);
1554 1554 VN_RELE(dvp);
1555 1555 /*
1556 1556 * The following clause was added to handle a problem
1557 1557 * with NFS consistency. It is possible that a lookup
1558 1558 * of the file to be created succeeded, but the file
1559 1559 * itself doesn't actually exist on the server. This
1560 1560 * is chiefly due to the DNLC containing an entry for
1561 1561 * the file which has been removed on the server. In
1562 1562 * this case, we just start over. If there was some
1563 1563 * other cause for the ESTALE error, then the lookup
1564 1564 * of the file will fail and the error will be returned
1565 1565 * above instead of looping around from here.
1566 1566 */
1567 1567 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1568 1568 goto top;
1569 1569 return (error);
1570 1570 }
1571 1571
1572 1572 int
1573 1573 vn_link(char *from, char *to, enum uio_seg seg)
1574 1574 {
1575 1575 return (vn_linkat(NULL, from, NO_FOLLOW, NULL, to, seg));
1576 1576 }
1577 1577
1578 1578 int
1579 1579 vn_linkat(vnode_t *fstartvp, char *from, enum symfollow follow,
1580 1580 vnode_t *tstartvp, char *to, enum uio_seg seg)
1581 1581 {
1582 1582 struct vnode *fvp; /* from vnode ptr */
1583 1583 struct vnode *tdvp; /* to directory vnode ptr */
1584 1584 struct pathname pn;
1585 1585 int error;
1586 1586 struct vattr vattr;
1587 1587 dev_t fsid;
1588 1588 int estale_retry = 0;
1589 1589 uint32_t auditing = AU_AUDITING();
1590 1590
1591 1591 top:
1592 1592 fvp = tdvp = NULL;
1593 1593 if (error = pn_get(to, seg, &pn))
1594 1594 return (error);
1595 1595 if (auditing && fstartvp != NULL)
1596 1596 audit_setfsat_path(1);
1597 1597 if (error = lookupnameat(from, seg, follow, NULLVPP, &fvp, fstartvp))
1598 1598 goto out;
1599 1599 if (auditing && tstartvp != NULL)
1600 1600 audit_setfsat_path(3);
1601 1601 if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &tdvp, NULLVPP, tstartvp))
1602 1602 goto out;
1603 1603 /*
1604 1604 * Make sure both source vnode and target directory vnode are
1605 1605 * in the same vfs and that it is writeable.
1606 1606 */
1607 1607 vattr.va_mask = AT_FSID;
1608 1608 if (error = VOP_GETATTR(fvp, &vattr, 0, CRED(), NULL))
1609 1609 goto out;
1610 1610 fsid = vattr.va_fsid;
1611 1611 vattr.va_mask = AT_FSID;
1612 1612 if (error = VOP_GETATTR(tdvp, &vattr, 0, CRED(), NULL))
1613 1613 goto out;
1614 1614 if (fsid != vattr.va_fsid) {
1615 1615 error = EXDEV;
1616 1616 goto out;
1617 1617 }
1618 1618 if (tdvp->v_vfsp->vfs_flag & VFS_RDONLY) {
1619 1619 error = EROFS;
1620 1620 goto out;
1621 1621 }
1622 1622 /*
1623 1623 * Do the link.
1624 1624 */
1625 1625 (void) pn_fixslash(&pn);
1626 1626 error = VOP_LINK(tdvp, fvp, pn.pn_path, CRED(), NULL, 0);
1627 1627 out:
1628 1628 pn_free(&pn);
1629 1629 if (fvp)
1630 1630 VN_RELE(fvp);
1631 1631 if (tdvp)
1632 1632 VN_RELE(tdvp);
1633 1633 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1634 1634 goto top;
1635 1635 return (error);
1636 1636 }
1637 1637
1638 1638 int
1639 1639 vn_rename(char *from, char *to, enum uio_seg seg)
1640 1640 {
1641 1641 return (vn_renameat(NULL, from, NULL, to, seg));
1642 1642 }
1643 1643
1644 1644 int
1645 1645 vn_renameat(vnode_t *fdvp, char *fname, vnode_t *tdvp,
1646 1646 char *tname, enum uio_seg seg)
1647 1647 {
1648 1648 int error;
1649 1649 struct vattr vattr;
1650 1650 struct pathname fpn; /* from pathname */
1651 1651 struct pathname tpn; /* to pathname */
1652 1652 dev_t fsid;
1653 1653 int in_crit_src, in_crit_targ;
1654 1654 vnode_t *fromvp, *fvp;
1655 1655 vnode_t *tovp, *targvp;
1656 1656 int estale_retry = 0;
1657 1657 uint32_t auditing = AU_AUDITING();
1658 1658
1659 1659 top:
1660 1660 fvp = fromvp = tovp = targvp = NULL;
1661 1661 in_crit_src = in_crit_targ = 0;
1662 1662 /*
1663 1663 * Get to and from pathnames.
1664 1664 */
1665 1665 if (error = pn_get(fname, seg, &fpn))
1666 1666 return (error);
1667 1667 if (error = pn_get(tname, seg, &tpn)) {
1668 1668 pn_free(&fpn);
1669 1669 return (error);
1670 1670 }
1671 1671
1672 1672 /*
1673 1673 * First we need to resolve the correct directories
1674 1674 * The passed in directories may only be a starting point,
1675 1675 * but we need the real directories the file(s) live in.
1676 1676 * For example the fname may be something like usr/lib/sparc
1677 1677 * and we were passed in the / directory, but we need to
1678 1678 * use the lib directory for the rename.
1679 1679 */
1680 1680
1681 1681 if (auditing && fdvp != NULL)
1682 1682 audit_setfsat_path(1);
1683 1683 /*
1684 1684 * Lookup to and from directories.
1685 1685 */
1686 1686 if (error = lookuppnat(&fpn, NULL, NO_FOLLOW, &fromvp, &fvp, fdvp)) {
1687 1687 goto out;
1688 1688 }
1689 1689
1690 1690 /*
1691 1691 * Make sure there is an entry.
1692 1692 */
1693 1693 if (fvp == NULL) {
1694 1694 error = ENOENT;
1695 1695 goto out;
1696 1696 }
1697 1697
1698 1698 if (auditing && tdvp != NULL)
1699 1699 audit_setfsat_path(3);
1700 1700 if (error = lookuppnat(&tpn, NULL, NO_FOLLOW, &tovp, &targvp, tdvp)) {
1701 1701 goto out;
1702 1702 }
1703 1703
1704 1704 /*
1705 1705 * Make sure both the from vnode directory and the to directory
1706 1706 * are in the same vfs and the to directory is writable.
1707 1707 * We check fsid's, not vfs pointers, so loopback fs works.
1708 1708 */
1709 1709 if (fromvp != tovp) {
1710 1710 vattr.va_mask = AT_FSID;
1711 1711 if (error = VOP_GETATTR(fromvp, &vattr, 0, CRED(), NULL))
1712 1712 goto out;
1713 1713 fsid = vattr.va_fsid;
1714 1714 vattr.va_mask = AT_FSID;
1715 1715 if (error = VOP_GETATTR(tovp, &vattr, 0, CRED(), NULL))
1716 1716 goto out;
1717 1717 if (fsid != vattr.va_fsid) {
1718 1718 error = EXDEV;
1719 1719 goto out;
1720 1720 }
1721 1721 }
1722 1722
1723 1723 if (tovp->v_vfsp->vfs_flag & VFS_RDONLY) {
1724 1724 error = EROFS;
1725 1725 goto out;
1726 1726 }
1727 1727
1728 1728 if (targvp && (fvp != targvp)) {
1729 1729 nbl_start_crit(targvp, RW_READER);
1730 1730 in_crit_targ = 1;
1731 1731 if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
1732 1732 error = EACCES;
1733 1733 goto out;
1734 1734 }
1735 1735 }
1736 1736
1737 1737 if (nbl_need_check(fvp)) {
1738 1738 nbl_start_crit(fvp, RW_READER);
1739 1739 in_crit_src = 1;
1740 1740 if (nbl_conflict(fvp, NBL_RENAME, 0, 0, 0, NULL)) {
1741 1741 error = EACCES;
1742 1742 goto out;
1743 1743 }
1744 1744 }
1745 1745
1746 1746 /*
1747 1747 * Do the rename.
1748 1748 */
1749 1749 (void) pn_fixslash(&tpn);
1750 1750 error = VOP_RENAME(fromvp, fpn.pn_path, tovp, tpn.pn_path, CRED(),
1751 1751 NULL, 0);
1752 1752
1753 1753 out:
1754 1754 pn_free(&fpn);
1755 1755 pn_free(&tpn);
1756 1756 if (in_crit_src)
1757 1757 nbl_end_crit(fvp);
1758 1758 if (in_crit_targ)
1759 1759 nbl_end_crit(targvp);
1760 1760 if (fromvp)
1761 1761 VN_RELE(fromvp);
1762 1762 if (tovp)
1763 1763 VN_RELE(tovp);
1764 1764 if (targvp)
1765 1765 VN_RELE(targvp);
1766 1766 if (fvp)
1767 1767 VN_RELE(fvp);
1768 1768 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1769 1769 goto top;
1770 1770 return (error);
1771 1771 }
1772 1772
1773 1773 /*
1774 1774 * Remove a file or directory.
1775 1775 */
1776 1776 int
1777 1777 vn_remove(char *fnamep, enum uio_seg seg, enum rm dirflag)
1778 1778 {
1779 1779 return (vn_removeat(NULL, fnamep, seg, dirflag));
1780 1780 }
1781 1781
1782 1782 int
1783 1783 vn_removeat(vnode_t *startvp, char *fnamep, enum uio_seg seg, enum rm dirflag)
1784 1784 {
1785 1785 struct vnode *vp; /* entry vnode */
1786 1786 struct vnode *dvp; /* ptr to parent dir vnode */
1787 1787 struct vnode *coveredvp;
1788 1788 struct pathname pn; /* name of entry */
1789 1789 enum vtype vtype;
1790 1790 int error;
1791 1791 struct vfs *vfsp;
1792 1792 struct vfs *dvfsp; /* ptr to parent dir vfs */
1793 1793 int in_crit = 0;
1794 1794 int estale_retry = 0;
1795 1795
1796 1796 top:
1797 1797 if (error = pn_get(fnamep, seg, &pn))
1798 1798 return (error);
1799 1799 dvp = vp = NULL;
1800 1800 if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &dvp, &vp, startvp)) {
1801 1801 pn_free(&pn);
1802 1802 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1803 1803 goto top;
1804 1804 return (error);
1805 1805 }
1806 1806
1807 1807 /*
1808 1808 * Make sure there is an entry.
1809 1809 */
1810 1810 if (vp == NULL) {
1811 1811 error = ENOENT;
1812 1812 goto out;
1813 1813 }
1814 1814
1815 1815 vfsp = vp->v_vfsp;
1816 1816 dvfsp = dvp->v_vfsp;
1817 1817
1818 1818 /*
1819 1819 * If the named file is the root of a mounted filesystem, fail,
1820 1820 * unless it's marked unlinkable. In that case, unmount the
1821 1821 * filesystem and proceed to unlink the covered vnode. (If the
1822 1822 * covered vnode is a directory, use rmdir instead of unlink,
1823 1823 * to avoid file system corruption.)
1824 1824 */
1825 1825 if (vp->v_flag & VROOT) {
1826 1826 if ((vfsp->vfs_flag & VFS_UNLINKABLE) == 0) {
1827 1827 error = EBUSY;
1828 1828 goto out;
1829 1829 }
1830 1830
1831 1831 /*
1832 1832 * Namefs specific code starts here.
1833 1833 */
1834 1834
1835 1835 if (dirflag == RMDIRECTORY) {
1836 1836 /*
1837 1837 * User called rmdir(2) on a file that has
1838 1838 * been namefs mounted on top of. Since
1839 1839 * namefs doesn't allow directories to
1840 1840 * be mounted on other files we know
1841 1841 * vp is not of type VDIR so fail to operation.
1842 1842 */
1843 1843 error = ENOTDIR;
1844 1844 goto out;
1845 1845 }
1846 1846
1847 1847 /*
1848 1848 * If VROOT is still set after grabbing vp->v_lock,
1849 1849 * noone has finished nm_unmount so far and coveredvp
1850 1850 * is valid.
1851 1851 * If we manage to grab vn_vfswlock(coveredvp) before releasing
1852 1852 * vp->v_lock, any race window is eliminated.
1853 1853 */
1854 1854
1855 1855 mutex_enter(&vp->v_lock);
1856 1856 if ((vp->v_flag & VROOT) == 0) {
1857 1857 /* Someone beat us to the unmount */
1858 1858 mutex_exit(&vp->v_lock);
1859 1859 error = EBUSY;
1860 1860 goto out;
1861 1861 }
1862 1862 vfsp = vp->v_vfsp;
1863 1863 coveredvp = vfsp->vfs_vnodecovered;
1864 1864 ASSERT(coveredvp);
1865 1865 /*
1866 1866 * Note: Implementation of vn_vfswlock shows that ordering of
1867 1867 * v_lock / vn_vfswlock is not an issue here.
1868 1868 */
1869 1869 error = vn_vfswlock(coveredvp);
1870 1870 mutex_exit(&vp->v_lock);
1871 1871
1872 1872 if (error)
1873 1873 goto out;
1874 1874
1875 1875 VN_HOLD(coveredvp);
1876 1876 VN_RELE(vp);
1877 1877 error = dounmount(vfsp, 0, CRED());
1878 1878
1879 1879 /*
1880 1880 * Unmounted the namefs file system; now get
1881 1881 * the object it was mounted over.
1882 1882 */
1883 1883 vp = coveredvp;
1884 1884 /*
1885 1885 * If namefs was mounted over a directory, then
1886 1886 * we want to use rmdir() instead of unlink().
1887 1887 */
1888 1888 if (vp->v_type == VDIR)
1889 1889 dirflag = RMDIRECTORY;
1890 1890
1891 1891 if (error)
1892 1892 goto out;
1893 1893 }
1894 1894
1895 1895 /*
1896 1896 * Make sure filesystem is writeable.
1897 1897 * We check the parent directory's vfs in case this is an lofs vnode.
1898 1898 */
1899 1899 if (dvfsp && dvfsp->vfs_flag & VFS_RDONLY) {
1900 1900 error = EROFS;
1901 1901 goto out;
1902 1902 }
1903 1903
1904 1904 vtype = vp->v_type;
1905 1905
1906 1906 /*
1907 1907 * If there is the possibility of an nbmand share reservation, make
1908 1908 * sure it's okay to remove the file. Keep a reference to the
1909 1909 * vnode, so that we can exit the nbl critical region after
1910 1910 * calling VOP_REMOVE.
1911 1911 * If there is no possibility of an nbmand share reservation,
1912 1912 * release the vnode reference now. Filesystems like NFS may
1913 1913 * behave differently if there is an extra reference, so get rid of
1914 1914 * this one. Fortunately, we can't have nbmand mounts on NFS
1915 1915 * filesystems.
1916 1916 */
1917 1917 if (nbl_need_check(vp)) {
1918 1918 nbl_start_crit(vp, RW_READER);
1919 1919 in_crit = 1;
1920 1920 if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
1921 1921 error = EACCES;
1922 1922 goto out;
1923 1923 }
1924 1924 } else {
1925 1925 VN_RELE(vp);
1926 1926 vp = NULL;
1927 1927 }
1928 1928
1929 1929 if (dirflag == RMDIRECTORY) {
1930 1930 /*
1931 1931 * Caller is using rmdir(2), which can only be applied to
1932 1932 * directories.
1933 1933 */
1934 1934 if (vtype != VDIR) {
1935 1935 error = ENOTDIR;
1936 1936 } else {
1937 1937 vnode_t *cwd;
1938 1938 proc_t *pp = curproc;
1939 1939
1940 1940 mutex_enter(&pp->p_lock);
1941 1941 cwd = PTOU(pp)->u_cdir;
1942 1942 VN_HOLD(cwd);
1943 1943 mutex_exit(&pp->p_lock);
1944 1944 error = VOP_RMDIR(dvp, pn.pn_path, cwd, CRED(),
1945 1945 NULL, 0);
1946 1946 VN_RELE(cwd);
1947 1947 }
1948 1948 } else {
1949 1949 /*
1950 1950 * Unlink(2) can be applied to anything.
1951 1951 */
1952 1952 error = VOP_REMOVE(dvp, pn.pn_path, CRED(), NULL, 0);
1953 1953 }
1954 1954
1955 1955 out:
1956 1956 pn_free(&pn);
1957 1957 if (in_crit) {
1958 1958 nbl_end_crit(vp);
1959 1959 in_crit = 0;
1960 1960 }
1961 1961 if (vp != NULL)
1962 1962 VN_RELE(vp);
1963 1963 if (dvp != NULL)
1964 1964 VN_RELE(dvp);
1965 1965 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1966 1966 goto top;
1967 1967 return (error);
1968 1968 }
1969 1969
1970 1970 /*
1971 1971 * Utility function to compare equality of vnodes.
1972 1972 * Compare the underlying real vnodes, if there are underlying vnodes.
1973 1973 * This is a more thorough comparison than the VN_CMP() macro provides.
1974 1974 */
1975 1975 int
1976 1976 vn_compare(vnode_t *vp1, vnode_t *vp2)
1977 1977 {
1978 1978 vnode_t *realvp;
1979 1979
1980 1980 if (vp1 != NULL && VOP_REALVP(vp1, &realvp, NULL) == 0)
1981 1981 vp1 = realvp;
1982 1982 if (vp2 != NULL && VOP_REALVP(vp2, &realvp, NULL) == 0)
1983 1983 vp2 = realvp;
1984 1984 return (VN_CMP(vp1, vp2));
1985 1985 }
1986 1986
1987 1987 /*
1988 1988 * The number of locks to hash into. This value must be a power
1989 1989 * of 2 minus 1 and should probably also be prime.
1990 1990 */
1991 1991 #define NUM_BUCKETS 1023
1992 1992
1993 1993 struct vn_vfslocks_bucket {
1994 1994 kmutex_t vb_lock;
1995 1995 vn_vfslocks_entry_t *vb_list;
1996 1996 char pad[64 - sizeof (kmutex_t) - sizeof (void *)];
1997 1997 };
1998 1998
1999 1999 /*
2000 2000 * Total number of buckets will be NUM_BUCKETS + 1 .
2001 2001 */
2002 2002
2003 2003 #pragma align 64(vn_vfslocks_buckets)
2004 2004 static struct vn_vfslocks_bucket vn_vfslocks_buckets[NUM_BUCKETS + 1];
2005 2005
2006 2006 #define VN_VFSLOCKS_SHIFT 9
2007 2007
2008 2008 #define VN_VFSLOCKS_HASH(vfsvpptr) \
2009 2009 ((((intptr_t)(vfsvpptr)) >> VN_VFSLOCKS_SHIFT) & NUM_BUCKETS)
2010 2010
2011 2011 /*
2012 2012 * vn_vfslocks_getlock() uses an HASH scheme to generate
2013 2013 * rwstlock using vfs/vnode pointer passed to it.
2014 2014 *
2015 2015 * vn_vfslocks_rele() releases a reference in the
2016 2016 * HASH table which allows the entry allocated by
2017 2017 * vn_vfslocks_getlock() to be freed at a later
2018 2018 * stage when the refcount drops to zero.
2019 2019 */
2020 2020
2021 2021 vn_vfslocks_entry_t *
2022 2022 vn_vfslocks_getlock(void *vfsvpptr)
2023 2023 {
2024 2024 struct vn_vfslocks_bucket *bp;
2025 2025 vn_vfslocks_entry_t *vep;
2026 2026 vn_vfslocks_entry_t *tvep;
2027 2027
2028 2028 ASSERT(vfsvpptr != NULL);
2029 2029 bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vfsvpptr)];
2030 2030
2031 2031 mutex_enter(&bp->vb_lock);
2032 2032 for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) {
2033 2033 if (vep->ve_vpvfs == vfsvpptr) {
2034 2034 vep->ve_refcnt++;
2035 2035 mutex_exit(&bp->vb_lock);
2036 2036 return (vep);
2037 2037 }
2038 2038 }
2039 2039 mutex_exit(&bp->vb_lock);
2040 2040 vep = kmem_alloc(sizeof (*vep), KM_SLEEP);
2041 2041 rwst_init(&vep->ve_lock, NULL, RW_DEFAULT, NULL);
2042 2042 vep->ve_vpvfs = (char *)vfsvpptr;
2043 2043 vep->ve_refcnt = 1;
2044 2044 mutex_enter(&bp->vb_lock);
2045 2045 for (tvep = bp->vb_list; tvep != NULL; tvep = tvep->ve_next) {
2046 2046 if (tvep->ve_vpvfs == vfsvpptr) {
2047 2047 tvep->ve_refcnt++;
2048 2048 mutex_exit(&bp->vb_lock);
2049 2049
2050 2050 /*
2051 2051 * There is already an entry in the hash
2052 2052 * destroy what we just allocated.
2053 2053 */
2054 2054 rwst_destroy(&vep->ve_lock);
2055 2055 kmem_free(vep, sizeof (*vep));
2056 2056 return (tvep);
2057 2057 }
2058 2058 }
2059 2059 vep->ve_next = bp->vb_list;
2060 2060 bp->vb_list = vep;
2061 2061 mutex_exit(&bp->vb_lock);
2062 2062 return (vep);
2063 2063 }
2064 2064
2065 2065 void
2066 2066 vn_vfslocks_rele(vn_vfslocks_entry_t *vepent)
2067 2067 {
2068 2068 struct vn_vfslocks_bucket *bp;
2069 2069 vn_vfslocks_entry_t *vep;
2070 2070 vn_vfslocks_entry_t *pvep;
2071 2071
2072 2072 ASSERT(vepent != NULL);
2073 2073 ASSERT(vepent->ve_vpvfs != NULL);
2074 2074
2075 2075 bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vepent->ve_vpvfs)];
2076 2076
2077 2077 mutex_enter(&bp->vb_lock);
2078 2078 vepent->ve_refcnt--;
2079 2079
2080 2080 if ((int32_t)vepent->ve_refcnt < 0)
2081 2081 cmn_err(CE_PANIC, "vn_vfslocks_rele: refcount negative");
2082 2082
2083 2083 if (vepent->ve_refcnt == 0) {
2084 2084 for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) {
2085 2085 if (vep->ve_vpvfs == vepent->ve_vpvfs) {
2086 2086 if (bp->vb_list == vep)
2087 2087 bp->vb_list = vep->ve_next;
2088 2088 else {
2089 2089 /* LINTED */
2090 2090 pvep->ve_next = vep->ve_next;
2091 2091 }
2092 2092 mutex_exit(&bp->vb_lock);
2093 2093 rwst_destroy(&vep->ve_lock);
2094 2094 kmem_free(vep, sizeof (*vep));
2095 2095 return;
2096 2096 }
2097 2097 pvep = vep;
2098 2098 }
2099 2099 cmn_err(CE_PANIC, "vn_vfslocks_rele: vp/vfs not found");
2100 2100 }
2101 2101 mutex_exit(&bp->vb_lock);
2102 2102 }
2103 2103
2104 2104 /*
2105 2105 * vn_vfswlock_wait is used to implement a lock which is logically a writers
2106 2106 * lock protecting the v_vfsmountedhere field.
2107 2107 * vn_vfswlock_wait has been modified to be similar to vn_vfswlock,
2108 2108 * except that it blocks to acquire the lock VVFSLOCK.
2109 2109 *
2110 2110 * traverse() and routines re-implementing part of traverse (e.g. autofs)
2111 2111 * need to hold this lock. mount(), vn_rename(), vn_remove() and so on
2112 2112 * need the non-blocking version of the writers lock i.e. vn_vfswlock
2113 2113 */
2114 2114 int
2115 2115 vn_vfswlock_wait(vnode_t *vp)
2116 2116 {
2117 2117 int retval;
2118 2118 vn_vfslocks_entry_t *vpvfsentry;
2119 2119 ASSERT(vp != NULL);
2120 2120
2121 2121 vpvfsentry = vn_vfslocks_getlock(vp);
2122 2122 retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_WRITER);
2123 2123
2124 2124 if (retval == EINTR) {
2125 2125 vn_vfslocks_rele(vpvfsentry);
2126 2126 return (EINTR);
2127 2127 }
2128 2128 return (retval);
2129 2129 }
2130 2130
2131 2131 int
2132 2132 vn_vfsrlock_wait(vnode_t *vp)
2133 2133 {
2134 2134 int retval;
2135 2135 vn_vfslocks_entry_t *vpvfsentry;
2136 2136 ASSERT(vp != NULL);
2137 2137
2138 2138 vpvfsentry = vn_vfslocks_getlock(vp);
2139 2139 retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_READER);
2140 2140
2141 2141 if (retval == EINTR) {
2142 2142 vn_vfslocks_rele(vpvfsentry);
2143 2143 return (EINTR);
2144 2144 }
2145 2145
2146 2146 return (retval);
2147 2147 }
2148 2148
2149 2149
2150 2150 /*
2151 2151 * vn_vfswlock is used to implement a lock which is logically a writers lock
2152 2152 * protecting the v_vfsmountedhere field.
2153 2153 */
2154 2154 int
2155 2155 vn_vfswlock(vnode_t *vp)
2156 2156 {
2157 2157 vn_vfslocks_entry_t *vpvfsentry;
2158 2158
2159 2159 /*
2160 2160 * If vp is NULL then somebody is trying to lock the covered vnode
2161 2161 * of /. (vfs_vnodecovered is NULL for /). This situation will
2162 2162 * only happen when unmounting /. Since that operation will fail
2163 2163 * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2164 2164 */
2165 2165 if (vp == NULL)
2166 2166 return (EBUSY);
2167 2167
2168 2168 vpvfsentry = vn_vfslocks_getlock(vp);
2169 2169
2170 2170 if (rwst_tryenter(&vpvfsentry->ve_lock, RW_WRITER))
2171 2171 return (0);
2172 2172
2173 2173 vn_vfslocks_rele(vpvfsentry);
2174 2174 return (EBUSY);
2175 2175 }
2176 2176
2177 2177 int
2178 2178 vn_vfsrlock(vnode_t *vp)
2179 2179 {
2180 2180 vn_vfslocks_entry_t *vpvfsentry;
2181 2181
2182 2182 /*
2183 2183 * If vp is NULL then somebody is trying to lock the covered vnode
2184 2184 * of /. (vfs_vnodecovered is NULL for /). This situation will
2185 2185 * only happen when unmounting /. Since that operation will fail
2186 2186 * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2187 2187 */
2188 2188 if (vp == NULL)
2189 2189 return (EBUSY);
2190 2190
2191 2191 vpvfsentry = vn_vfslocks_getlock(vp);
2192 2192
2193 2193 if (rwst_tryenter(&vpvfsentry->ve_lock, RW_READER))
2194 2194 return (0);
2195 2195
2196 2196 vn_vfslocks_rele(vpvfsentry);
2197 2197 return (EBUSY);
2198 2198 }
2199 2199
2200 2200 void
2201 2201 vn_vfsunlock(vnode_t *vp)
2202 2202 {
2203 2203 vn_vfslocks_entry_t *vpvfsentry;
2204 2204
2205 2205 /*
2206 2206 * ve_refcnt needs to be decremented twice.
2207 2207 * 1. To release refernce after a call to vn_vfslocks_getlock()
2208 2208 * 2. To release the reference from the locking routines like
2209 2209 * vn_vfsrlock/vn_vfswlock etc,.
2210 2210 */
2211 2211 vpvfsentry = vn_vfslocks_getlock(vp);
2212 2212 vn_vfslocks_rele(vpvfsentry);
2213 2213
2214 2214 rwst_exit(&vpvfsentry->ve_lock);
2215 2215 vn_vfslocks_rele(vpvfsentry);
2216 2216 }
2217 2217
2218 2218 int
2219 2219 vn_vfswlock_held(vnode_t *vp)
2220 2220 {
2221 2221 int held;
2222 2222 vn_vfslocks_entry_t *vpvfsentry;
2223 2223
2224 2224 ASSERT(vp != NULL);
2225 2225
2226 2226 vpvfsentry = vn_vfslocks_getlock(vp);
2227 2227 held = rwst_lock_held(&vpvfsentry->ve_lock, RW_WRITER);
2228 2228
2229 2229 vn_vfslocks_rele(vpvfsentry);
2230 2230 return (held);
2231 2231 }
2232 2232
2233 2233
2234 2234 int
2235 2235 vn_make_ops(
2236 2236 const char *name, /* Name of file system */
2237 2237 const fs_operation_def_t *templ, /* Operation specification */
2238 2238 vnodeops_t **actual) /* Return the vnodeops */
2239 2239 {
2240 2240 int unused_ops;
2241 2241 int error;
2242 2242
2243 2243 *actual = (vnodeops_t *)kmem_alloc(sizeof (vnodeops_t), KM_SLEEP);
2244 2244
2245 2245 (*actual)->vnop_name = name;
2246 2246
2247 2247 error = fs_build_vector(*actual, &unused_ops, vn_ops_table, templ);
2248 2248 if (error) {
2249 2249 kmem_free(*actual, sizeof (vnodeops_t));
2250 2250 }
2251 2251
2252 2252 #if DEBUG
2253 2253 if (unused_ops != 0)
2254 2254 cmn_err(CE_WARN, "vn_make_ops: %s: %d operations supplied "
2255 2255 "but not used", name, unused_ops);
2256 2256 #endif
2257 2257
2258 2258 return (error);
2259 2259 }
2260 2260
2261 2261 /*
2262 2262 * Free the vnodeops created as a result of vn_make_ops()
2263 2263 */
2264 2264 void
2265 2265 vn_freevnodeops(vnodeops_t *vnops)
2266 2266 {
2267 2267 kmem_free(vnops, sizeof (vnodeops_t));
2268 2268 }
2269 2269
2270 2270 /*
2271 2271 * Vnode cache.
2272 2272 */
2273 2273
2274 2274 /* ARGSUSED */
2275 2275 static int
2276 2276 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
2277 2277 {
2278 2278 struct vnode *vp;
2279 2279
2280 2280 vp = buf;
2281 2281
2282 2282 mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
2283 2283 mutex_init(&vp->v_vsd_lock, NULL, MUTEX_DEFAULT, NULL);
2284 2284 cv_init(&vp->v_cv, NULL, CV_DEFAULT, NULL);
2285 2285 rw_init(&vp->v_nbllock, NULL, RW_DEFAULT, NULL);
2286 2286 vp->v_femhead = NULL; /* Must be done before vn_reinit() */
2287 2287 vp->v_path = NULL;
2288 2288 vp->v_mpssdata = NULL;
2289 2289 vp->v_vsd = NULL;
2290 2290 vp->v_fopdata = NULL;
2291 2291
2292 2292 return (0);
2293 2293 }
2294 2294
2295 2295 /* ARGSUSED */
2296 2296 static void
2297 2297 vn_cache_destructor(void *buf, void *cdrarg)
2298 2298 {
2299 2299 struct vnode *vp;
2300 2300
2301 2301 vp = buf;
2302 2302
2303 2303 rw_destroy(&vp->v_nbllock);
2304 2304 cv_destroy(&vp->v_cv);
2305 2305 mutex_destroy(&vp->v_vsd_lock);
2306 2306 mutex_destroy(&vp->v_lock);
2307 2307 }
2308 2308
2309 2309 void
2310 2310 vn_create_cache(void)
2311 2311 {
2312 2312 /* LINTED */
2313 2313 ASSERT((1 << VNODE_ALIGN_LOG2) ==
2314 2314 P2ROUNDUP(sizeof (struct vnode), VNODE_ALIGN));
2315 2315 vn_cache = kmem_cache_create("vn_cache", sizeof (struct vnode),
2316 2316 VNODE_ALIGN, vn_cache_constructor, vn_cache_destructor, NULL, NULL,
2317 2317 NULL, 0);
2318 2318 }
2319 2319
2320 2320 void
2321 2321 vn_destroy_cache(void)
2322 2322 {
2323 2323 kmem_cache_destroy(vn_cache);
2324 2324 }
2325 2325
2326 2326 /*
2327 2327 * Used by file systems when fs-specific nodes (e.g., ufs inodes) are
2328 2328 * cached by the file system and vnodes remain associated.
2329 2329 */
2330 2330 void
2331 2331 vn_recycle(vnode_t *vp)
2332 2332 {
2333 2333 ASSERT(vp->v_pages == NULL);
2334 2334
2335 2335 /*
2336 2336 * XXX - This really belongs in vn_reinit(), but we have some issues
2337 2337 * with the counts. Best to have it here for clean initialization.
2338 2338 */
2339 2339 vp->v_rdcnt = 0;
2340 2340 vp->v_wrcnt = 0;
2341 2341 vp->v_mmap_read = 0;
2342 2342 vp->v_mmap_write = 0;
2343 2343
2344 2344 /*
2345 2345 * If FEM was in use, make sure everything gets cleaned up
2346 2346 * NOTE: vp->v_femhead is initialized to NULL in the vnode
2347 2347 * constructor.
2348 2348 */
2349 2349 if (vp->v_femhead) {
2350 2350 /* XXX - There should be a free_femhead() that does all this */
2351 2351 ASSERT(vp->v_femhead->femh_list == NULL);
2352 2352 mutex_destroy(&vp->v_femhead->femh_lock);
2353 2353 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
2354 2354 vp->v_femhead = NULL;
2355 2355 }
2356 2356 if (vp->v_path) {
2357 2357 kmem_free(vp->v_path, strlen(vp->v_path) + 1);
2358 2358 vp->v_path = NULL;
2359 2359 }
2360 2360
2361 2361 if (vp->v_fopdata != NULL) {
2362 2362 free_fopdata(vp);
2363 2363 }
2364 2364 vp->v_mpssdata = NULL;
2365 2365 vsd_free(vp);
2366 2366 }
2367 2367
2368 2368 /*
2369 2369 * Used to reset the vnode fields including those that are directly accessible
2370 2370 * as well as those which require an accessor function.
2371 2371 *
2372 2372 * Does not initialize:
2373 2373 * synchronization objects: v_lock, v_vsd_lock, v_nbllock, v_cv
2374 2374 * v_data (since FS-nodes and vnodes point to each other and should
2375 2375 * be updated simultaneously)
2376 2376 * v_op (in case someone needs to make a VOP call on this object)
2377 2377 */
2378 2378 void
2379 2379 vn_reinit(vnode_t *vp)
2380 2380 {
2381 2381 vp->v_count = 1;
2382 2382 vp->v_count_dnlc = 0;
2383 2383 vp->v_vfsp = NULL;
2384 2384 vp->v_stream = NULL;
2385 2385 vp->v_vfsmountedhere = NULL;
2386 2386 vp->v_flag = 0;
2387 2387 vp->v_type = VNON;
2388 2388 vp->v_rdev = NODEV;
2389 2389
2390 2390 vp->v_filocks = NULL;
2391 2391 vp->v_shrlocks = NULL;
2392 2392 vp->v_pages = NULL;
2393 2393
2394 2394 vp->v_locality = NULL;
2395 2395 vp->v_xattrdir = NULL;
2396 2396
2397 2397 /* Handles v_femhead, v_path, and the r/w/map counts */
2398 2398 vn_recycle(vp);
2399 2399 }
2400 2400
2401 2401 vnode_t *
2402 2402 vn_alloc(int kmflag)
2403 2403 {
2404 2404 vnode_t *vp;
2405 2405
2406 2406 vp = kmem_cache_alloc(vn_cache, kmflag);
2407 2407
2408 2408 if (vp != NULL) {
2409 2409 vp->v_femhead = NULL; /* Must be done before vn_reinit() */
2410 2410 vp->v_fopdata = NULL;
2411 2411 vn_reinit(vp);
2412 2412 }
2413 2413
2414 2414 return (vp);
2415 2415 }
2416 2416
2417 2417 void
2418 2418 vn_free(vnode_t *vp)
2419 2419 {
2420 2420 ASSERT(vp->v_shrlocks == NULL);
2421 2421 ASSERT(vp->v_filocks == NULL);
2422 2422
2423 2423 /*
2424 2424 * Some file systems call vn_free() with v_count of zero,
2425 2425 * some with v_count of 1. In any case, the value should
2426 2426 * never be anything else.
2427 2427 */
2428 2428 ASSERT((vp->v_count == 0) || (vp->v_count == 1));
2429 2429 ASSERT(vp->v_count_dnlc == 0);
2430 2430 if (vp->v_path != NULL) {
2431 2431 kmem_free(vp->v_path, strlen(vp->v_path) + 1);
2432 2432 vp->v_path = NULL;
2433 2433 }
2434 2434
2435 2435 /* If FEM was in use, make sure everything gets cleaned up */
2436 2436 if (vp->v_femhead) {
2437 2437 /* XXX - There should be a free_femhead() that does all this */
2438 2438 ASSERT(vp->v_femhead->femh_list == NULL);
2439 2439 mutex_destroy(&vp->v_femhead->femh_lock);
2440 2440 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
2441 2441 vp->v_femhead = NULL;
2442 2442 }
2443 2443
2444 2444 if (vp->v_fopdata != NULL) {
2445 2445 free_fopdata(vp);
2446 2446 }
2447 2447 vp->v_mpssdata = NULL;
2448 2448 vsd_free(vp);
2449 2449 kmem_cache_free(vn_cache, vp);
2450 2450 }
2451 2451
2452 2452 /*
2453 2453 * vnode status changes, should define better states than 1, 0.
2454 2454 */
2455 2455 void
2456 2456 vn_reclaim(vnode_t *vp)
2457 2457 {
2458 2458 vfs_t *vfsp = vp->v_vfsp;
2459 2459
2460 2460 if (vfsp == NULL ||
2461 2461 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2462 2462 return;
2463 2463 }
2464 2464 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_RECLAIMED);
2465 2465 }
2466 2466
2467 2467 void
2468 2468 vn_idle(vnode_t *vp)
2469 2469 {
2470 2470 vfs_t *vfsp = vp->v_vfsp;
2471 2471
2472 2472 if (vfsp == NULL ||
2473 2473 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2474 2474 return;
2475 2475 }
2476 2476 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_IDLED);
2477 2477 }
2478 2478 void
2479 2479 vn_exists(vnode_t *vp)
2480 2480 {
2481 2481 vfs_t *vfsp = vp->v_vfsp;
2482 2482
2483 2483 if (vfsp == NULL ||
2484 2484 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2485 2485 return;
2486 2486 }
2487 2487 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_EXISTS);
2488 2488 }
2489 2489
2490 2490 void
2491 2491 vn_invalid(vnode_t *vp)
2492 2492 {
2493 2493 vfs_t *vfsp = vp->v_vfsp;
2494 2494
2495 2495 if (vfsp == NULL ||
2496 2496 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2497 2497 return;
2498 2498 }
2499 2499 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_DESTROYED);
2500 2500 }
2501 2501
2502 2502 /* Vnode event notification */
2503 2503
2504 2504 int
2505 2505 vnevent_support(vnode_t *vp, caller_context_t *ct)
2506 2506 {
2507 2507 if (vp == NULL)
2508 2508 return (EINVAL);
2509 2509
2510 2510 return (VOP_VNEVENT(vp, VE_SUPPORT, NULL, NULL, ct));
2511 2511 }
2512 2512
2513 2513 void
2514 2514 vnevent_rename_src(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2515 2515 {
2516 2516 if (vp == NULL || vp->v_femhead == NULL) {
2517 2517 return;
2518 2518 }
2519 2519 (void) VOP_VNEVENT(vp, VE_RENAME_SRC, dvp, name, ct);
2520 2520 }
2521 2521
2522 2522 void
2523 2523 vnevent_rename_dest(vnode_t *vp, vnode_t *dvp, char *name,
2524 2524 caller_context_t *ct)
2525 2525 {
2526 2526 if (vp == NULL || vp->v_femhead == NULL) {
2527 2527 return;
2528 2528 }
2529 2529 (void) VOP_VNEVENT(vp, VE_RENAME_DEST, dvp, name, ct);
2530 2530 }
2531 2531
2532 2532 void
2533 2533 vnevent_rename_dest_dir(vnode_t *vp, caller_context_t *ct)
2534 2534 {
2535 2535 if (vp == NULL || vp->v_femhead == NULL) {
2536 2536 return;
2537 2537 }
2538 2538 (void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, NULL, NULL, ct);
2539 2539 }
2540 2540
2541 2541 void
2542 2542 vnevent_remove(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2543 2543 {
2544 2544 if (vp == NULL || vp->v_femhead == NULL) {
2545 2545 return;
2546 2546 }
2547 2547 (void) VOP_VNEVENT(vp, VE_REMOVE, dvp, name, ct);
2548 2548 }
2549 2549
2550 2550 void
2551 2551 vnevent_rmdir(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2552 2552 {
2553 2553 if (vp == NULL || vp->v_femhead == NULL) {
2554 2554 return;
2555 2555 }
2556 2556 (void) VOP_VNEVENT(vp, VE_RMDIR, dvp, name, ct);
2557 2557 }
2558 2558
2559 2559 void
2560 2560 vnevent_create(vnode_t *vp, caller_context_t *ct)
2561 2561 {
2562 2562 if (vp == NULL || vp->v_femhead == NULL) {
2563 2563 return;
2564 2564 }
2565 2565 (void) VOP_VNEVENT(vp, VE_CREATE, NULL, NULL, ct);
2566 2566 }
2567 2567
2568 2568 void
2569 2569 vnevent_link(vnode_t *vp, caller_context_t *ct)
2570 2570 {
2571 2571 if (vp == NULL || vp->v_femhead == NULL) {
2572 2572 return;
2573 2573 }
2574 2574 (void) VOP_VNEVENT(vp, VE_LINK, NULL, NULL, ct);
2575 2575 }
2576 2576
2577 2577 void
2578 2578 vnevent_mountedover(vnode_t *vp, caller_context_t *ct)
2579 2579 {
2580 2580 if (vp == NULL || vp->v_femhead == NULL) {
2581 2581 return;
2582 2582 }
2583 2583 (void) VOP_VNEVENT(vp, VE_MOUNTEDOVER, NULL, NULL, ct);
2584 2584 }
2585 2585
2586 2586 void
2587 2587 vnevent_truncate(vnode_t *vp, caller_context_t *ct)
2588 2588 {
2589 2589 if (vp == NULL || vp->v_femhead == NULL) {
2590 2590 return;
2591 2591 }
2592 2592 (void) VOP_VNEVENT(vp, VE_TRUNCATE, NULL, NULL, ct);
2593 2593 }
2594 2594
2595 2595 /*
2596 2596 * Vnode accessors.
2597 2597 */
2598 2598
2599 2599 int
2600 2600 vn_is_readonly(vnode_t *vp)
2601 2601 {
2602 2602 return (vp->v_vfsp->vfs_flag & VFS_RDONLY);
2603 2603 }
2604 2604
2605 2605 int
2606 2606 vn_has_flocks(vnode_t *vp)
2607 2607 {
2608 2608 return (vp->v_filocks != NULL);
2609 2609 }
2610 2610
2611 2611 int
2612 2612 vn_has_mandatory_locks(vnode_t *vp, int mode)
2613 2613 {
2614 2614 return ((vp->v_filocks != NULL) && (MANDLOCK(vp, mode)));
2615 2615 }
2616 2616
2617 2617 int
2618 2618 vn_has_cached_data(vnode_t *vp)
2619 2619 {
2620 2620 return (vp->v_pages != NULL);
2621 2621 }
2622 2622
2623 2623 /*
2624 2624 * Return 0 if the vnode in question shouldn't be permitted into a zone via
2625 2625 * zone_enter(2).
2626 2626 */
2627 2627 int
2628 2628 vn_can_change_zones(vnode_t *vp)
2629 2629 {
2630 2630 struct vfssw *vswp;
2631 2631 int allow = 1;
2632 2632 vnode_t *rvp;
2633 2633
2634 2634 if (nfs_global_client_only != 0)
2635 2635 return (1);
2636 2636
2637 2637 /*
2638 2638 * We always want to look at the underlying vnode if there is one.
2639 2639 */
2640 2640 if (VOP_REALVP(vp, &rvp, NULL) != 0)
2641 2641 rvp = vp;
2642 2642 /*
2643 2643 * Some pseudo filesystems (including doorfs) don't actually register
2644 2644 * their vfsops_t, so the following may return NULL; we happily let
2645 2645 * such vnodes switch zones.
2646 2646 */
2647 2647 vswp = vfs_getvfsswbyvfsops(vfs_getops(rvp->v_vfsp));
2648 2648 if (vswp != NULL) {
2649 2649 if (vswp->vsw_flag & VSW_NOTZONESAFE)
2650 2650 allow = 0;
2651 2651 vfs_unrefvfssw(vswp);
2652 2652 }
2653 2653 return (allow);
2654 2654 }
2655 2655
2656 2656 /*
2657 2657 * Return nonzero if the vnode is a mount point, zero if not.
2658 2658 */
2659 2659 int
2660 2660 vn_ismntpt(vnode_t *vp)
2661 2661 {
2662 2662 return (vp->v_vfsmountedhere != NULL);
2663 2663 }
2664 2664
2665 2665 /* Retrieve the vfs (if any) mounted on this vnode */
2666 2666 vfs_t *
2667 2667 vn_mountedvfs(vnode_t *vp)
2668 2668 {
2669 2669 return (vp->v_vfsmountedhere);
2670 2670 }
2671 2671
2672 2672 /*
2673 2673 * Return nonzero if the vnode is referenced by the dnlc, zero if not.
2674 2674 */
2675 2675 int
2676 2676 vn_in_dnlc(vnode_t *vp)
2677 2677 {
2678 2678 return (vp->v_count_dnlc > 0);
2679 2679 }
2680 2680
2681 2681 /*
2682 2682 * vn_has_other_opens() checks whether a particular file is opened by more than
2683 2683 * just the caller and whether the open is for read and/or write.
2684 2684 * This routine is for calling after the caller has already called VOP_OPEN()
2685 2685 * and the caller wishes to know if they are the only one with it open for
2686 2686 * the mode(s) specified.
2687 2687 *
2688 2688 * Vnode counts are only kept on regular files (v_type=VREG).
2689 2689 */
2690 2690 int
2691 2691 vn_has_other_opens(
2692 2692 vnode_t *vp,
2693 2693 v_mode_t mode)
2694 2694 {
2695 2695
2696 2696 ASSERT(vp != NULL);
2697 2697
2698 2698 switch (mode) {
2699 2699 case V_WRITE:
2700 2700 if (vp->v_wrcnt > 1)
2701 2701 return (V_TRUE);
2702 2702 break;
2703 2703 case V_RDORWR:
2704 2704 if ((vp->v_rdcnt > 1) || (vp->v_wrcnt > 1))
2705 2705 return (V_TRUE);
2706 2706 break;
2707 2707 case V_RDANDWR:
2708 2708 if ((vp->v_rdcnt > 1) && (vp->v_wrcnt > 1))
2709 2709 return (V_TRUE);
2710 2710 break;
2711 2711 case V_READ:
2712 2712 if (vp->v_rdcnt > 1)
2713 2713 return (V_TRUE);
2714 2714 break;
2715 2715 }
2716 2716
2717 2717 return (V_FALSE);
2718 2718 }
2719 2719
2720 2720 /*
2721 2721 * vn_is_opened() checks whether a particular file is opened and
2722 2722 * whether the open is for read and/or write.
2723 2723 *
2724 2724 * Vnode counts are only kept on regular files (v_type=VREG).
2725 2725 */
2726 2726 int
2727 2727 vn_is_opened(
2728 2728 vnode_t *vp,
2729 2729 v_mode_t mode)
2730 2730 {
2731 2731
2732 2732 ASSERT(vp != NULL);
2733 2733
2734 2734 switch (mode) {
2735 2735 case V_WRITE:
2736 2736 if (vp->v_wrcnt)
2737 2737 return (V_TRUE);
2738 2738 break;
2739 2739 case V_RDANDWR:
2740 2740 if (vp->v_rdcnt && vp->v_wrcnt)
2741 2741 return (V_TRUE);
2742 2742 break;
2743 2743 case V_RDORWR:
2744 2744 if (vp->v_rdcnt || vp->v_wrcnt)
2745 2745 return (V_TRUE);
2746 2746 break;
2747 2747 case V_READ:
2748 2748 if (vp->v_rdcnt)
2749 2749 return (V_TRUE);
2750 2750 break;
2751 2751 }
2752 2752
2753 2753 return (V_FALSE);
2754 2754 }
2755 2755
2756 2756 /*
2757 2757 * vn_is_mapped() checks whether a particular file is mapped and whether
2758 2758 * the file is mapped read and/or write.
2759 2759 */
2760 2760 int
2761 2761 vn_is_mapped(
2762 2762 vnode_t *vp,
2763 2763 v_mode_t mode)
2764 2764 {
2765 2765
2766 2766 ASSERT(vp != NULL);
2767 2767
2768 2768 #if !defined(_LP64)
2769 2769 switch (mode) {
2770 2770 /*
2771 2771 * The atomic_add_64_nv functions force atomicity in the
2772 2772 * case of 32 bit architectures. Otherwise the 64 bit values
2773 2773 * require two fetches. The value of the fields may be
2774 2774 * (potentially) changed between the first fetch and the
2775 2775 * second
2776 2776 */
2777 2777 case V_WRITE:
2778 2778 if (atomic_add_64_nv((&(vp->v_mmap_write)), 0))
2779 2779 return (V_TRUE);
2780 2780 break;
2781 2781 case V_RDANDWR:
2782 2782 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) &&
2783 2783 (atomic_add_64_nv((&(vp->v_mmap_write)), 0)))
2784 2784 return (V_TRUE);
2785 2785 break;
2786 2786 case V_RDORWR:
2787 2787 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) ||
2788 2788 (atomic_add_64_nv((&(vp->v_mmap_write)), 0)))
2789 2789 return (V_TRUE);
2790 2790 break;
2791 2791 case V_READ:
2792 2792 if (atomic_add_64_nv((&(vp->v_mmap_read)), 0))
2793 2793 return (V_TRUE);
2794 2794 break;
2795 2795 }
2796 2796 #else
2797 2797 switch (mode) {
2798 2798 case V_WRITE:
2799 2799 if (vp->v_mmap_write)
2800 2800 return (V_TRUE);
2801 2801 break;
2802 2802 case V_RDANDWR:
2803 2803 if (vp->v_mmap_read && vp->v_mmap_write)
2804 2804 return (V_TRUE);
2805 2805 break;
2806 2806 case V_RDORWR:
2807 2807 if (vp->v_mmap_read || vp->v_mmap_write)
2808 2808 return (V_TRUE);
2809 2809 break;
2810 2810 case V_READ:
2811 2811 if (vp->v_mmap_read)
2812 2812 return (V_TRUE);
2813 2813 break;
2814 2814 }
2815 2815 #endif
2816 2816
2817 2817 return (V_FALSE);
2818 2818 }
2819 2819
2820 2820 /*
2821 2821 * Set the operations vector for a vnode.
2822 2822 *
2823 2823 * FEM ensures that the v_femhead pointer is filled in before the
2824 2824 * v_op pointer is changed. This means that if the v_femhead pointer
2825 2825 * is NULL, and the v_op field hasn't changed since before which checked
2826 2826 * the v_femhead pointer; then our update is ok - we are not racing with
2827 2827 * FEM.
2828 2828 */
2829 2829 void
|
↓ open down ↓ |
2829 lines elided |
↑ open up ↑ |
2830 2830 vn_setops(vnode_t *vp, vnodeops_t *vnodeops)
2831 2831 {
2832 2832 vnodeops_t *op;
2833 2833
2834 2834 ASSERT(vp != NULL);
2835 2835 ASSERT(vnodeops != NULL);
2836 2836
2837 2837 op = vp->v_op;
2838 2838 membar_consumer();
2839 2839 /*
2840 - * If vp->v_femhead == NULL, then we'll call casptr() to do the
2841 - * compare-and-swap on vp->v_op. If either fails, then FEM is
2840 + * If vp->v_femhead == NULL, then we'll call atomic_cas_ptr() to do
2841 + * the compare-and-swap on vp->v_op. If either fails, then FEM is
2842 2842 * in effect on the vnode and we need to have FEM deal with it.
2843 2843 */
2844 - if (vp->v_femhead != NULL || casptr(&vp->v_op, op, vnodeops) != op) {
2844 + if (vp->v_femhead != NULL || atomic_cas_ptr(&vp->v_op, op, vnodeops) !=
2845 + op) {
2845 2846 fem_setvnops(vp, vnodeops);
2846 2847 }
2847 2848 }
2848 2849
2849 2850 /*
2850 2851 * Retrieve the operations vector for a vnode
2851 2852 * As with vn_setops(above); make sure we aren't racing with FEM.
2852 2853 * FEM sets the v_op to a special, internal, vnodeops that wouldn't
2853 2854 * make sense to the callers of this routine.
2854 2855 */
2855 2856 vnodeops_t *
2856 2857 vn_getops(vnode_t *vp)
2857 2858 {
2858 2859 vnodeops_t *op;
2859 2860
2860 2861 ASSERT(vp != NULL);
2861 2862
2862 2863 op = vp->v_op;
2863 2864 membar_consumer();
2864 2865 if (vp->v_femhead == NULL && op == vp->v_op) {
2865 2866 return (op);
2866 2867 } else {
2867 2868 return (fem_getvnops(vp));
2868 2869 }
2869 2870 }
2870 2871
2871 2872 /*
2872 2873 * Returns non-zero (1) if the vnodeops matches that of the vnode.
2873 2874 * Returns zero (0) if not.
2874 2875 */
2875 2876 int
2876 2877 vn_matchops(vnode_t *vp, vnodeops_t *vnodeops)
2877 2878 {
2878 2879 return (vn_getops(vp) == vnodeops);
2879 2880 }
2880 2881
2881 2882 /*
2882 2883 * Returns non-zero (1) if the specified operation matches the
2883 2884 * corresponding operation for that the vnode.
2884 2885 * Returns zero (0) if not.
2885 2886 */
2886 2887
2887 2888 #define MATCHNAME(n1, n2) (((n1)[0] == (n2)[0]) && (strcmp((n1), (n2)) == 0))
2888 2889
2889 2890 int
2890 2891 vn_matchopval(vnode_t *vp, char *vopname, fs_generic_func_p funcp)
2891 2892 {
2892 2893 const fs_operation_trans_def_t *otdp;
2893 2894 fs_generic_func_p *loc = NULL;
2894 2895 vnodeops_t *vop = vn_getops(vp);
2895 2896
2896 2897 ASSERT(vopname != NULL);
2897 2898
2898 2899 for (otdp = vn_ops_table; otdp->name != NULL; otdp++) {
2899 2900 if (MATCHNAME(otdp->name, vopname)) {
2900 2901 loc = (fs_generic_func_p *)
2901 2902 ((char *)(vop) + otdp->offset);
2902 2903 break;
2903 2904 }
2904 2905 }
2905 2906
2906 2907 return ((loc != NULL) && (*loc == funcp));
2907 2908 }
2908 2909
2909 2910 /*
2910 2911 * fs_new_caller_id() needs to return a unique ID on a given local system.
2911 2912 * The IDs do not need to survive across reboots. These are primarily
2912 2913 * used so that (FEM) monitors can detect particular callers (such as
2913 2914 * the NFS server) to a given vnode/vfs operation.
2914 2915 */
2915 2916 u_longlong_t
2916 2917 fs_new_caller_id()
2917 2918 {
2918 2919 static uint64_t next_caller_id = 0LL; /* First call returns 1 */
2919 2920
2920 2921 return ((u_longlong_t)atomic_add_64_nv(&next_caller_id, 1));
2921 2922 }
2922 2923
2923 2924 /*
2924 2925 * Given a starting vnode and a path, updates the path in the target vnode in
2925 2926 * a safe manner. If the vnode already has path information embedded, then the
2926 2927 * cached path is left untouched.
2927 2928 */
2928 2929
2929 2930 size_t max_vnode_path = 4 * MAXPATHLEN;
2930 2931
2931 2932 void
2932 2933 vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
2933 2934 const char *path, size_t plen)
2934 2935 {
2935 2936 char *rpath;
2936 2937 vnode_t *base;
2937 2938 size_t rpathlen, rpathalloc;
2938 2939 int doslash = 1;
2939 2940
2940 2941 if (*path == '/') {
2941 2942 base = rootvp;
2942 2943 path++;
2943 2944 plen--;
2944 2945 } else {
2945 2946 base = startvp;
2946 2947 }
2947 2948
2948 2949 /*
2949 2950 * We cannot grab base->v_lock while we hold vp->v_lock because of
2950 2951 * the potential for deadlock.
2951 2952 */
2952 2953 mutex_enter(&base->v_lock);
2953 2954 if (base->v_path == NULL) {
2954 2955 mutex_exit(&base->v_lock);
2955 2956 return;
2956 2957 }
2957 2958
2958 2959 rpathlen = strlen(base->v_path);
2959 2960 rpathalloc = rpathlen + plen + 1;
2960 2961 /* Avoid adding a slash if there's already one there */
2961 2962 if (base->v_path[rpathlen-1] == '/')
2962 2963 doslash = 0;
2963 2964 else
2964 2965 rpathalloc++;
2965 2966
2966 2967 /*
2967 2968 * We don't want to call kmem_alloc(KM_SLEEP) with kernel locks held,
2968 2969 * so we must do this dance. If, by chance, something changes the path,
2969 2970 * just give up since there is no real harm.
2970 2971 */
2971 2972 mutex_exit(&base->v_lock);
2972 2973
2973 2974 /* Paths should stay within reason */
2974 2975 if (rpathalloc > max_vnode_path)
2975 2976 return;
2976 2977
2977 2978 rpath = kmem_alloc(rpathalloc, KM_SLEEP);
2978 2979
2979 2980 mutex_enter(&base->v_lock);
2980 2981 if (base->v_path == NULL || strlen(base->v_path) != rpathlen) {
2981 2982 mutex_exit(&base->v_lock);
2982 2983 kmem_free(rpath, rpathalloc);
2983 2984 return;
2984 2985 }
2985 2986 bcopy(base->v_path, rpath, rpathlen);
2986 2987 mutex_exit(&base->v_lock);
2987 2988
2988 2989 if (doslash)
2989 2990 rpath[rpathlen++] = '/';
2990 2991 bcopy(path, rpath + rpathlen, plen);
2991 2992 rpath[rpathlen + plen] = '\0';
2992 2993
2993 2994 mutex_enter(&vp->v_lock);
2994 2995 if (vp->v_path != NULL) {
2995 2996 mutex_exit(&vp->v_lock);
2996 2997 kmem_free(rpath, rpathalloc);
2997 2998 } else {
2998 2999 vp->v_path = rpath;
2999 3000 mutex_exit(&vp->v_lock);
3000 3001 }
3001 3002 }
3002 3003
3003 3004 /*
3004 3005 * Sets the path to the vnode to be the given string, regardless of current
3005 3006 * context. The string must be a complete path from rootdir. This is only used
3006 3007 * by fsop_root() for setting the path based on the mountpoint.
3007 3008 */
3008 3009 void
3009 3010 vn_setpath_str(struct vnode *vp, const char *str, size_t len)
3010 3011 {
3011 3012 char *buf = kmem_alloc(len + 1, KM_SLEEP);
3012 3013
3013 3014 mutex_enter(&vp->v_lock);
3014 3015 if (vp->v_path != NULL) {
3015 3016 mutex_exit(&vp->v_lock);
3016 3017 kmem_free(buf, len + 1);
3017 3018 return;
3018 3019 }
3019 3020
3020 3021 vp->v_path = buf;
3021 3022 bcopy(str, vp->v_path, len);
3022 3023 vp->v_path[len] = '\0';
3023 3024
3024 3025 mutex_exit(&vp->v_lock);
3025 3026 }
3026 3027
3027 3028 /*
3028 3029 * Called from within filesystem's vop_rename() to handle renames once the
3029 3030 * target vnode is available.
3030 3031 */
3031 3032 void
3032 3033 vn_renamepath(vnode_t *dvp, vnode_t *vp, const char *nm, size_t len)
3033 3034 {
3034 3035 char *tmp;
3035 3036
3036 3037 mutex_enter(&vp->v_lock);
3037 3038 tmp = vp->v_path;
3038 3039 vp->v_path = NULL;
3039 3040 mutex_exit(&vp->v_lock);
3040 3041 vn_setpath(rootdir, dvp, vp, nm, len);
3041 3042 if (tmp != NULL)
3042 3043 kmem_free(tmp, strlen(tmp) + 1);
3043 3044 }
3044 3045
3045 3046 /*
3046 3047 * Similar to vn_setpath_str(), this function sets the path of the destination
3047 3048 * vnode to the be the same as the source vnode.
3048 3049 */
3049 3050 void
3050 3051 vn_copypath(struct vnode *src, struct vnode *dst)
3051 3052 {
3052 3053 char *buf;
3053 3054 int alloc;
3054 3055
3055 3056 mutex_enter(&src->v_lock);
3056 3057 if (src->v_path == NULL) {
3057 3058 mutex_exit(&src->v_lock);
3058 3059 return;
3059 3060 }
3060 3061 alloc = strlen(src->v_path) + 1;
3061 3062
3062 3063 /* avoid kmem_alloc() with lock held */
3063 3064 mutex_exit(&src->v_lock);
3064 3065 buf = kmem_alloc(alloc, KM_SLEEP);
3065 3066 mutex_enter(&src->v_lock);
3066 3067 if (src->v_path == NULL || strlen(src->v_path) + 1 != alloc) {
3067 3068 mutex_exit(&src->v_lock);
3068 3069 kmem_free(buf, alloc);
3069 3070 return;
3070 3071 }
3071 3072 bcopy(src->v_path, buf, alloc);
3072 3073 mutex_exit(&src->v_lock);
3073 3074
3074 3075 mutex_enter(&dst->v_lock);
3075 3076 if (dst->v_path != NULL) {
3076 3077 mutex_exit(&dst->v_lock);
3077 3078 kmem_free(buf, alloc);
3078 3079 return;
3079 3080 }
3080 3081 dst->v_path = buf;
3081 3082 mutex_exit(&dst->v_lock);
3082 3083 }
3083 3084
3084 3085 /*
3085 3086 * XXX Private interface for segvn routines that handle vnode
3086 3087 * large page segments.
3087 3088 *
3088 3089 * return 1 if vp's file system VOP_PAGEIO() implementation
3089 3090 * can be safely used instead of VOP_GETPAGE() for handling
3090 3091 * pagefaults against regular non swap files. VOP_PAGEIO()
3091 3092 * interface is considered safe here if its implementation
3092 3093 * is very close to VOP_GETPAGE() implementation.
3093 3094 * e.g. It zero's out the part of the page beyond EOF. Doesn't
3094 3095 * panic if there're file holes but instead returns an error.
3095 3096 * Doesn't assume file won't be changed by user writes, etc.
3096 3097 *
3097 3098 * return 0 otherwise.
3098 3099 *
3099 3100 * For now allow segvn to only use VOP_PAGEIO() with ufs and nfs.
3100 3101 */
3101 3102 int
3102 3103 vn_vmpss_usepageio(vnode_t *vp)
3103 3104 {
3104 3105 vfs_t *vfsp = vp->v_vfsp;
3105 3106 char *fsname = vfssw[vfsp->vfs_fstype].vsw_name;
3106 3107 char *pageio_ok_fss[] = {"ufs", "nfs", NULL};
3107 3108 char **fsok = pageio_ok_fss;
3108 3109
3109 3110 if (fsname == NULL) {
3110 3111 return (0);
3111 3112 }
3112 3113
3113 3114 for (; *fsok; fsok++) {
3114 3115 if (strcmp(*fsok, fsname) == 0) {
3115 3116 return (1);
3116 3117 }
3117 3118 }
3118 3119 return (0);
3119 3120 }
3120 3121
3121 3122 /* VOP_XXX() macros call the corresponding fop_xxx() function */
3122 3123
3123 3124 int
3124 3125 fop_open(
3125 3126 vnode_t **vpp,
3126 3127 int mode,
3127 3128 cred_t *cr,
3128 3129 caller_context_t *ct)
3129 3130 {
3130 3131 int ret;
3131 3132 vnode_t *vp = *vpp;
3132 3133
3133 3134 VN_HOLD(vp);
3134 3135 /*
3135 3136 * Adding to the vnode counts before calling open
3136 3137 * avoids the need for a mutex. It circumvents a race
3137 3138 * condition where a query made on the vnode counts results in a
3138 3139 * false negative. The inquirer goes away believing the file is
3139 3140 * not open when there is an open on the file already under way.
3140 3141 *
3141 3142 * The counts are meant to prevent NFS from granting a delegation
3142 3143 * when it would be dangerous to do so.
3143 3144 *
3144 3145 * The vnode counts are only kept on regular files
3145 3146 */
3146 3147 if ((*vpp)->v_type == VREG) {
3147 3148 if (mode & FREAD)
3148 3149 atomic_add_32(&((*vpp)->v_rdcnt), 1);
3149 3150 if (mode & FWRITE)
3150 3151 atomic_add_32(&((*vpp)->v_wrcnt), 1);
3151 3152 }
3152 3153
3153 3154 VOPXID_MAP_CR(vp, cr);
3154 3155
3155 3156 ret = (*(*(vpp))->v_op->vop_open)(vpp, mode, cr, ct);
3156 3157
3157 3158 if (ret) {
3158 3159 /*
3159 3160 * Use the saved vp just in case the vnode ptr got trashed
3160 3161 * by the error.
3161 3162 */
3162 3163 VOPSTATS_UPDATE(vp, open);
3163 3164 if ((vp->v_type == VREG) && (mode & FREAD))
3164 3165 atomic_add_32(&(vp->v_rdcnt), -1);
3165 3166 if ((vp->v_type == VREG) && (mode & FWRITE))
3166 3167 atomic_add_32(&(vp->v_wrcnt), -1);
3167 3168 } else {
3168 3169 /*
3169 3170 * Some filesystems will return a different vnode,
3170 3171 * but the same path was still used to open it.
3171 3172 * So if we do change the vnode and need to
3172 3173 * copy over the path, do so here, rather than special
3173 3174 * casing each filesystem. Adjust the vnode counts to
3174 3175 * reflect the vnode switch.
3175 3176 */
3176 3177 VOPSTATS_UPDATE(*vpp, open);
3177 3178 if (*vpp != vp && *vpp != NULL) {
3178 3179 vn_copypath(vp, *vpp);
3179 3180 if (((*vpp)->v_type == VREG) && (mode & FREAD))
3180 3181 atomic_add_32(&((*vpp)->v_rdcnt), 1);
3181 3182 if ((vp->v_type == VREG) && (mode & FREAD))
3182 3183 atomic_add_32(&(vp->v_rdcnt), -1);
3183 3184 if (((*vpp)->v_type == VREG) && (mode & FWRITE))
3184 3185 atomic_add_32(&((*vpp)->v_wrcnt), 1);
3185 3186 if ((vp->v_type == VREG) && (mode & FWRITE))
3186 3187 atomic_add_32(&(vp->v_wrcnt), -1);
3187 3188 }
3188 3189 }
3189 3190 VN_RELE(vp);
3190 3191 return (ret);
3191 3192 }
3192 3193
3193 3194 int
3194 3195 fop_close(
3195 3196 vnode_t *vp,
3196 3197 int flag,
3197 3198 int count,
3198 3199 offset_t offset,
3199 3200 cred_t *cr,
3200 3201 caller_context_t *ct)
3201 3202 {
3202 3203 int err;
3203 3204
3204 3205 VOPXID_MAP_CR(vp, cr);
3205 3206
3206 3207 err = (*(vp)->v_op->vop_close)(vp, flag, count, offset, cr, ct);
3207 3208 VOPSTATS_UPDATE(vp, close);
3208 3209 /*
3209 3210 * Check passed in count to handle possible dups. Vnode counts are only
3210 3211 * kept on regular files
3211 3212 */
3212 3213 if ((vp->v_type == VREG) && (count == 1)) {
3213 3214 if (flag & FREAD) {
3214 3215 ASSERT(vp->v_rdcnt > 0);
3215 3216 atomic_add_32(&(vp->v_rdcnt), -1);
3216 3217 }
3217 3218 if (flag & FWRITE) {
3218 3219 ASSERT(vp->v_wrcnt > 0);
3219 3220 atomic_add_32(&(vp->v_wrcnt), -1);
3220 3221 }
3221 3222 }
3222 3223 return (err);
3223 3224 }
3224 3225
3225 3226 int
3226 3227 fop_read(
3227 3228 vnode_t *vp,
3228 3229 uio_t *uiop,
3229 3230 int ioflag,
3230 3231 cred_t *cr,
3231 3232 caller_context_t *ct)
3232 3233 {
3233 3234 int err;
3234 3235 ssize_t resid_start = uiop->uio_resid;
3235 3236
3236 3237 VOPXID_MAP_CR(vp, cr);
3237 3238
3238 3239 err = (*(vp)->v_op->vop_read)(vp, uiop, ioflag, cr, ct);
3239 3240 VOPSTATS_UPDATE_IO(vp, read,
3240 3241 read_bytes, (resid_start - uiop->uio_resid));
3241 3242 return (err);
3242 3243 }
3243 3244
3244 3245 int
3245 3246 fop_write(
3246 3247 vnode_t *vp,
3247 3248 uio_t *uiop,
3248 3249 int ioflag,
3249 3250 cred_t *cr,
3250 3251 caller_context_t *ct)
3251 3252 {
3252 3253 int err;
3253 3254 ssize_t resid_start = uiop->uio_resid;
3254 3255
3255 3256 VOPXID_MAP_CR(vp, cr);
3256 3257
3257 3258 err = (*(vp)->v_op->vop_write)(vp, uiop, ioflag, cr, ct);
3258 3259 VOPSTATS_UPDATE_IO(vp, write,
3259 3260 write_bytes, (resid_start - uiop->uio_resid));
3260 3261 return (err);
3261 3262 }
3262 3263
3263 3264 int
3264 3265 fop_ioctl(
3265 3266 vnode_t *vp,
3266 3267 int cmd,
3267 3268 intptr_t arg,
3268 3269 int flag,
3269 3270 cred_t *cr,
3270 3271 int *rvalp,
3271 3272 caller_context_t *ct)
3272 3273 {
3273 3274 int err;
3274 3275
3275 3276 VOPXID_MAP_CR(vp, cr);
3276 3277
3277 3278 err = (*(vp)->v_op->vop_ioctl)(vp, cmd, arg, flag, cr, rvalp, ct);
3278 3279 VOPSTATS_UPDATE(vp, ioctl);
3279 3280 return (err);
3280 3281 }
3281 3282
3282 3283 int
3283 3284 fop_setfl(
3284 3285 vnode_t *vp,
3285 3286 int oflags,
3286 3287 int nflags,
3287 3288 cred_t *cr,
3288 3289 caller_context_t *ct)
3289 3290 {
3290 3291 int err;
3291 3292
3292 3293 VOPXID_MAP_CR(vp, cr);
3293 3294
3294 3295 err = (*(vp)->v_op->vop_setfl)(vp, oflags, nflags, cr, ct);
3295 3296 VOPSTATS_UPDATE(vp, setfl);
3296 3297 return (err);
3297 3298 }
3298 3299
3299 3300 int
3300 3301 fop_getattr(
3301 3302 vnode_t *vp,
3302 3303 vattr_t *vap,
3303 3304 int flags,
3304 3305 cred_t *cr,
3305 3306 caller_context_t *ct)
3306 3307 {
3307 3308 int err;
3308 3309
3309 3310 VOPXID_MAP_CR(vp, cr);
3310 3311
3311 3312 /*
3312 3313 * If this file system doesn't understand the xvattr extensions
3313 3314 * then turn off the xvattr bit.
3314 3315 */
3315 3316 if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) {
3316 3317 vap->va_mask &= ~AT_XVATTR;
3317 3318 }
3318 3319
3319 3320 /*
3320 3321 * We're only allowed to skip the ACL check iff we used a 32 bit
3321 3322 * ACE mask with VOP_ACCESS() to determine permissions.
3322 3323 */
3323 3324 if ((flags & ATTR_NOACLCHECK) &&
3324 3325 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3325 3326 return (EINVAL);
3326 3327 }
3327 3328 err = (*(vp)->v_op->vop_getattr)(vp, vap, flags, cr, ct);
3328 3329 VOPSTATS_UPDATE(vp, getattr);
3329 3330 return (err);
3330 3331 }
3331 3332
3332 3333 int
3333 3334 fop_setattr(
3334 3335 vnode_t *vp,
3335 3336 vattr_t *vap,
3336 3337 int flags,
3337 3338 cred_t *cr,
3338 3339 caller_context_t *ct)
3339 3340 {
3340 3341 int err;
3341 3342
3342 3343 VOPXID_MAP_CR(vp, cr);
3343 3344
3344 3345 /*
3345 3346 * If this file system doesn't understand the xvattr extensions
3346 3347 * then turn off the xvattr bit.
3347 3348 */
3348 3349 if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) {
3349 3350 vap->va_mask &= ~AT_XVATTR;
3350 3351 }
3351 3352
3352 3353 /*
3353 3354 * We're only allowed to skip the ACL check iff we used a 32 bit
3354 3355 * ACE mask with VOP_ACCESS() to determine permissions.
3355 3356 */
3356 3357 if ((flags & ATTR_NOACLCHECK) &&
3357 3358 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3358 3359 return (EINVAL);
3359 3360 }
3360 3361 err = (*(vp)->v_op->vop_setattr)(vp, vap, flags, cr, ct);
3361 3362 VOPSTATS_UPDATE(vp, setattr);
3362 3363 return (err);
3363 3364 }
3364 3365
3365 3366 int
3366 3367 fop_access(
3367 3368 vnode_t *vp,
3368 3369 int mode,
3369 3370 int flags,
3370 3371 cred_t *cr,
3371 3372 caller_context_t *ct)
3372 3373 {
3373 3374 int err;
3374 3375
3375 3376 if ((flags & V_ACE_MASK) &&
3376 3377 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3377 3378 return (EINVAL);
3378 3379 }
3379 3380
3380 3381 VOPXID_MAP_CR(vp, cr);
3381 3382
3382 3383 err = (*(vp)->v_op->vop_access)(vp, mode, flags, cr, ct);
3383 3384 VOPSTATS_UPDATE(vp, access);
3384 3385 return (err);
3385 3386 }
3386 3387
3387 3388 int
3388 3389 fop_lookup(
3389 3390 vnode_t *dvp,
3390 3391 char *nm,
3391 3392 vnode_t **vpp,
3392 3393 pathname_t *pnp,
3393 3394 int flags,
3394 3395 vnode_t *rdir,
3395 3396 cred_t *cr,
3396 3397 caller_context_t *ct,
3397 3398 int *deflags, /* Returned per-dirent flags */
3398 3399 pathname_t *ppnp) /* Returned case-preserved name in directory */
3399 3400 {
3400 3401 int ret;
3401 3402
3402 3403 /*
3403 3404 * If this file system doesn't support case-insensitive access
3404 3405 * and said access is requested, fail quickly. It is required
3405 3406 * that if the vfs supports case-insensitive lookup, it also
3406 3407 * supports extended dirent flags.
3407 3408 */
3408 3409 if (flags & FIGNORECASE &&
3409 3410 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3410 3411 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3411 3412 return (EINVAL);
3412 3413
3413 3414 VOPXID_MAP_CR(dvp, cr);
3414 3415
3415 3416 if ((flags & LOOKUP_XATTR) && (flags & LOOKUP_HAVE_SYSATTR_DIR) == 0) {
3416 3417 ret = xattr_dir_lookup(dvp, vpp, flags, cr);
3417 3418 } else {
3418 3419 ret = (*(dvp)->v_op->vop_lookup)
3419 3420 (dvp, nm, vpp, pnp, flags, rdir, cr, ct, deflags, ppnp);
3420 3421 }
3421 3422 if (ret == 0 && *vpp) {
3422 3423 VOPSTATS_UPDATE(*vpp, lookup);
3423 3424 if ((*vpp)->v_path == NULL) {
3424 3425 vn_setpath(rootdir, dvp, *vpp, nm, strlen(nm));
3425 3426 }
3426 3427 }
3427 3428
3428 3429 return (ret);
3429 3430 }
3430 3431
3431 3432 int
3432 3433 fop_create(
3433 3434 vnode_t *dvp,
3434 3435 char *name,
3435 3436 vattr_t *vap,
3436 3437 vcexcl_t excl,
3437 3438 int mode,
3438 3439 vnode_t **vpp,
3439 3440 cred_t *cr,
3440 3441 int flags,
3441 3442 caller_context_t *ct,
3442 3443 vsecattr_t *vsecp) /* ACL to set during create */
3443 3444 {
3444 3445 int ret;
3445 3446
3446 3447 if (vsecp != NULL &&
3447 3448 vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) {
3448 3449 return (EINVAL);
3449 3450 }
3450 3451 /*
3451 3452 * If this file system doesn't support case-insensitive access
3452 3453 * and said access is requested, fail quickly.
3453 3454 */
3454 3455 if (flags & FIGNORECASE &&
3455 3456 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3456 3457 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3457 3458 return (EINVAL);
3458 3459
3459 3460 VOPXID_MAP_CR(dvp, cr);
3460 3461
3461 3462 ret = (*(dvp)->v_op->vop_create)
3462 3463 (dvp, name, vap, excl, mode, vpp, cr, flags, ct, vsecp);
3463 3464 if (ret == 0 && *vpp) {
3464 3465 VOPSTATS_UPDATE(*vpp, create);
3465 3466 if ((*vpp)->v_path == NULL) {
3466 3467 vn_setpath(rootdir, dvp, *vpp, name, strlen(name));
3467 3468 }
3468 3469 }
3469 3470
3470 3471 return (ret);
3471 3472 }
3472 3473
3473 3474 int
3474 3475 fop_remove(
3475 3476 vnode_t *dvp,
3476 3477 char *nm,
3477 3478 cred_t *cr,
3478 3479 caller_context_t *ct,
3479 3480 int flags)
3480 3481 {
3481 3482 int err;
3482 3483
3483 3484 /*
3484 3485 * If this file system doesn't support case-insensitive access
3485 3486 * and said access is requested, fail quickly.
3486 3487 */
3487 3488 if (flags & FIGNORECASE &&
3488 3489 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3489 3490 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3490 3491 return (EINVAL);
3491 3492
3492 3493 VOPXID_MAP_CR(dvp, cr);
3493 3494
3494 3495 err = (*(dvp)->v_op->vop_remove)(dvp, nm, cr, ct, flags);
3495 3496 VOPSTATS_UPDATE(dvp, remove);
3496 3497 return (err);
3497 3498 }
3498 3499
3499 3500 int
3500 3501 fop_link(
3501 3502 vnode_t *tdvp,
3502 3503 vnode_t *svp,
3503 3504 char *tnm,
3504 3505 cred_t *cr,
3505 3506 caller_context_t *ct,
3506 3507 int flags)
3507 3508 {
3508 3509 int err;
3509 3510
3510 3511 /*
3511 3512 * If the target file system doesn't support case-insensitive access
3512 3513 * and said access is requested, fail quickly.
3513 3514 */
3514 3515 if (flags & FIGNORECASE &&
3515 3516 (vfs_has_feature(tdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3516 3517 vfs_has_feature(tdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3517 3518 return (EINVAL);
3518 3519
3519 3520 VOPXID_MAP_CR(tdvp, cr);
3520 3521
3521 3522 err = (*(tdvp)->v_op->vop_link)(tdvp, svp, tnm, cr, ct, flags);
3522 3523 VOPSTATS_UPDATE(tdvp, link);
3523 3524 return (err);
3524 3525 }
3525 3526
3526 3527 int
3527 3528 fop_rename(
3528 3529 vnode_t *sdvp,
3529 3530 char *snm,
3530 3531 vnode_t *tdvp,
3531 3532 char *tnm,
3532 3533 cred_t *cr,
3533 3534 caller_context_t *ct,
3534 3535 int flags)
3535 3536 {
3536 3537 int err;
3537 3538
3538 3539 /*
3539 3540 * If the file system involved does not support
3540 3541 * case-insensitive access and said access is requested, fail
3541 3542 * quickly.
3542 3543 */
3543 3544 if (flags & FIGNORECASE &&
3544 3545 ((vfs_has_feature(sdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3545 3546 vfs_has_feature(sdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)))
3546 3547 return (EINVAL);
3547 3548
3548 3549 VOPXID_MAP_CR(tdvp, cr);
3549 3550
3550 3551 err = (*(sdvp)->v_op->vop_rename)(sdvp, snm, tdvp, tnm, cr, ct, flags);
3551 3552 VOPSTATS_UPDATE(sdvp, rename);
3552 3553 return (err);
3553 3554 }
3554 3555
3555 3556 int
3556 3557 fop_mkdir(
3557 3558 vnode_t *dvp,
3558 3559 char *dirname,
3559 3560 vattr_t *vap,
3560 3561 vnode_t **vpp,
3561 3562 cred_t *cr,
3562 3563 caller_context_t *ct,
3563 3564 int flags,
3564 3565 vsecattr_t *vsecp) /* ACL to set during create */
3565 3566 {
3566 3567 int ret;
3567 3568
3568 3569 if (vsecp != NULL &&
3569 3570 vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) {
3570 3571 return (EINVAL);
3571 3572 }
3572 3573 /*
3573 3574 * If this file system doesn't support case-insensitive access
3574 3575 * and said access is requested, fail quickly.
3575 3576 */
3576 3577 if (flags & FIGNORECASE &&
3577 3578 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3578 3579 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3579 3580 return (EINVAL);
3580 3581
3581 3582 VOPXID_MAP_CR(dvp, cr);
3582 3583
3583 3584 ret = (*(dvp)->v_op->vop_mkdir)
3584 3585 (dvp, dirname, vap, vpp, cr, ct, flags, vsecp);
3585 3586 if (ret == 0 && *vpp) {
3586 3587 VOPSTATS_UPDATE(*vpp, mkdir);
3587 3588 if ((*vpp)->v_path == NULL) {
3588 3589 vn_setpath(rootdir, dvp, *vpp, dirname,
3589 3590 strlen(dirname));
3590 3591 }
3591 3592 }
3592 3593
3593 3594 return (ret);
3594 3595 }
3595 3596
3596 3597 int
3597 3598 fop_rmdir(
3598 3599 vnode_t *dvp,
3599 3600 char *nm,
3600 3601 vnode_t *cdir,
3601 3602 cred_t *cr,
3602 3603 caller_context_t *ct,
3603 3604 int flags)
3604 3605 {
3605 3606 int err;
3606 3607
3607 3608 /*
3608 3609 * If this file system doesn't support case-insensitive access
3609 3610 * and said access is requested, fail quickly.
3610 3611 */
3611 3612 if (flags & FIGNORECASE &&
3612 3613 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3613 3614 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3614 3615 return (EINVAL);
3615 3616
3616 3617 VOPXID_MAP_CR(dvp, cr);
3617 3618
3618 3619 err = (*(dvp)->v_op->vop_rmdir)(dvp, nm, cdir, cr, ct, flags);
3619 3620 VOPSTATS_UPDATE(dvp, rmdir);
3620 3621 return (err);
3621 3622 }
3622 3623
3623 3624 int
3624 3625 fop_readdir(
3625 3626 vnode_t *vp,
3626 3627 uio_t *uiop,
3627 3628 cred_t *cr,
3628 3629 int *eofp,
3629 3630 caller_context_t *ct,
3630 3631 int flags)
3631 3632 {
3632 3633 int err;
3633 3634 ssize_t resid_start = uiop->uio_resid;
3634 3635
3635 3636 /*
3636 3637 * If this file system doesn't support retrieving directory
3637 3638 * entry flags and said access is requested, fail quickly.
3638 3639 */
3639 3640 if (flags & V_RDDIR_ENTFLAGS &&
3640 3641 vfs_has_feature(vp->v_vfsp, VFSFT_DIRENTFLAGS) == 0)
3641 3642 return (EINVAL);
3642 3643
3643 3644 VOPXID_MAP_CR(vp, cr);
3644 3645
3645 3646 err = (*(vp)->v_op->vop_readdir)(vp, uiop, cr, eofp, ct, flags);
3646 3647 VOPSTATS_UPDATE_IO(vp, readdir,
3647 3648 readdir_bytes, (resid_start - uiop->uio_resid));
3648 3649 return (err);
3649 3650 }
3650 3651
3651 3652 int
3652 3653 fop_symlink(
3653 3654 vnode_t *dvp,
3654 3655 char *linkname,
3655 3656 vattr_t *vap,
3656 3657 char *target,
3657 3658 cred_t *cr,
3658 3659 caller_context_t *ct,
3659 3660 int flags)
3660 3661 {
3661 3662 int err;
3662 3663 xvattr_t xvattr;
3663 3664
3664 3665 /*
3665 3666 * If this file system doesn't support case-insensitive access
3666 3667 * and said access is requested, fail quickly.
3667 3668 */
3668 3669 if (flags & FIGNORECASE &&
3669 3670 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3670 3671 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3671 3672 return (EINVAL);
3672 3673
3673 3674 VOPXID_MAP_CR(dvp, cr);
3674 3675
3675 3676 /* check for reparse point */
3676 3677 if ((vfs_has_feature(dvp->v_vfsp, VFSFT_REPARSE)) &&
3677 3678 (strncmp(target, FS_REPARSE_TAG_STR,
3678 3679 strlen(FS_REPARSE_TAG_STR)) == 0)) {
3679 3680 if (!fs_reparse_mark(target, vap, &xvattr))
3680 3681 vap = (vattr_t *)&xvattr;
3681 3682 }
3682 3683
3683 3684 err = (*(dvp)->v_op->vop_symlink)
3684 3685 (dvp, linkname, vap, target, cr, ct, flags);
3685 3686 VOPSTATS_UPDATE(dvp, symlink);
3686 3687 return (err);
3687 3688 }
3688 3689
3689 3690 int
3690 3691 fop_readlink(
3691 3692 vnode_t *vp,
3692 3693 uio_t *uiop,
3693 3694 cred_t *cr,
3694 3695 caller_context_t *ct)
3695 3696 {
3696 3697 int err;
3697 3698
3698 3699 VOPXID_MAP_CR(vp, cr);
3699 3700
3700 3701 err = (*(vp)->v_op->vop_readlink)(vp, uiop, cr, ct);
3701 3702 VOPSTATS_UPDATE(vp, readlink);
3702 3703 return (err);
3703 3704 }
3704 3705
3705 3706 int
3706 3707 fop_fsync(
3707 3708 vnode_t *vp,
3708 3709 int syncflag,
3709 3710 cred_t *cr,
3710 3711 caller_context_t *ct)
3711 3712 {
3712 3713 int err;
3713 3714
3714 3715 VOPXID_MAP_CR(vp, cr);
3715 3716
3716 3717 err = (*(vp)->v_op->vop_fsync)(vp, syncflag, cr, ct);
3717 3718 VOPSTATS_UPDATE(vp, fsync);
3718 3719 return (err);
3719 3720 }
3720 3721
3721 3722 void
3722 3723 fop_inactive(
3723 3724 vnode_t *vp,
3724 3725 cred_t *cr,
3725 3726 caller_context_t *ct)
3726 3727 {
3727 3728 /* Need to update stats before vop call since we may lose the vnode */
3728 3729 VOPSTATS_UPDATE(vp, inactive);
3729 3730
3730 3731 VOPXID_MAP_CR(vp, cr);
3731 3732
3732 3733 (*(vp)->v_op->vop_inactive)(vp, cr, ct);
3733 3734 }
3734 3735
3735 3736 int
3736 3737 fop_fid(
3737 3738 vnode_t *vp,
3738 3739 fid_t *fidp,
3739 3740 caller_context_t *ct)
3740 3741 {
3741 3742 int err;
3742 3743
3743 3744 err = (*(vp)->v_op->vop_fid)(vp, fidp, ct);
3744 3745 VOPSTATS_UPDATE(vp, fid);
3745 3746 return (err);
3746 3747 }
3747 3748
3748 3749 int
3749 3750 fop_rwlock(
3750 3751 vnode_t *vp,
3751 3752 int write_lock,
3752 3753 caller_context_t *ct)
3753 3754 {
3754 3755 int ret;
3755 3756
3756 3757 ret = ((*(vp)->v_op->vop_rwlock)(vp, write_lock, ct));
3757 3758 VOPSTATS_UPDATE(vp, rwlock);
3758 3759 return (ret);
3759 3760 }
3760 3761
3761 3762 void
3762 3763 fop_rwunlock(
3763 3764 vnode_t *vp,
3764 3765 int write_lock,
3765 3766 caller_context_t *ct)
3766 3767 {
3767 3768 (*(vp)->v_op->vop_rwunlock)(vp, write_lock, ct);
3768 3769 VOPSTATS_UPDATE(vp, rwunlock);
3769 3770 }
3770 3771
3771 3772 int
3772 3773 fop_seek(
3773 3774 vnode_t *vp,
3774 3775 offset_t ooff,
3775 3776 offset_t *noffp,
3776 3777 caller_context_t *ct)
3777 3778 {
3778 3779 int err;
3779 3780
3780 3781 err = (*(vp)->v_op->vop_seek)(vp, ooff, noffp, ct);
3781 3782 VOPSTATS_UPDATE(vp, seek);
3782 3783 return (err);
3783 3784 }
3784 3785
3785 3786 int
3786 3787 fop_cmp(
3787 3788 vnode_t *vp1,
3788 3789 vnode_t *vp2,
3789 3790 caller_context_t *ct)
3790 3791 {
3791 3792 int err;
3792 3793
3793 3794 err = (*(vp1)->v_op->vop_cmp)(vp1, vp2, ct);
3794 3795 VOPSTATS_UPDATE(vp1, cmp);
3795 3796 return (err);
3796 3797 }
3797 3798
3798 3799 int
3799 3800 fop_frlock(
3800 3801 vnode_t *vp,
3801 3802 int cmd,
3802 3803 flock64_t *bfp,
3803 3804 int flag,
3804 3805 offset_t offset,
3805 3806 struct flk_callback *flk_cbp,
3806 3807 cred_t *cr,
3807 3808 caller_context_t *ct)
3808 3809 {
3809 3810 int err;
3810 3811
3811 3812 VOPXID_MAP_CR(vp, cr);
3812 3813
3813 3814 err = (*(vp)->v_op->vop_frlock)
3814 3815 (vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3815 3816 VOPSTATS_UPDATE(vp, frlock);
3816 3817 return (err);
3817 3818 }
3818 3819
3819 3820 int
3820 3821 fop_space(
3821 3822 vnode_t *vp,
3822 3823 int cmd,
3823 3824 flock64_t *bfp,
3824 3825 int flag,
3825 3826 offset_t offset,
3826 3827 cred_t *cr,
3827 3828 caller_context_t *ct)
3828 3829 {
3829 3830 int err;
3830 3831
3831 3832 VOPXID_MAP_CR(vp, cr);
3832 3833
3833 3834 err = (*(vp)->v_op->vop_space)(vp, cmd, bfp, flag, offset, cr, ct);
3834 3835 VOPSTATS_UPDATE(vp, space);
3835 3836 return (err);
3836 3837 }
3837 3838
3838 3839 int
3839 3840 fop_realvp(
3840 3841 vnode_t *vp,
3841 3842 vnode_t **vpp,
3842 3843 caller_context_t *ct)
3843 3844 {
3844 3845 int err;
3845 3846
3846 3847 err = (*(vp)->v_op->vop_realvp)(vp, vpp, ct);
3847 3848 VOPSTATS_UPDATE(vp, realvp);
3848 3849 return (err);
3849 3850 }
3850 3851
3851 3852 int
3852 3853 fop_getpage(
3853 3854 vnode_t *vp,
3854 3855 offset_t off,
3855 3856 size_t len,
3856 3857 uint_t *protp,
3857 3858 page_t **plarr,
3858 3859 size_t plsz,
3859 3860 struct seg *seg,
3860 3861 caddr_t addr,
3861 3862 enum seg_rw rw,
3862 3863 cred_t *cr,
3863 3864 caller_context_t *ct)
3864 3865 {
3865 3866 int err;
3866 3867
3867 3868 VOPXID_MAP_CR(vp, cr);
3868 3869
3869 3870 err = (*(vp)->v_op->vop_getpage)
3870 3871 (vp, off, len, protp, plarr, plsz, seg, addr, rw, cr, ct);
3871 3872 VOPSTATS_UPDATE(vp, getpage);
3872 3873 return (err);
3873 3874 }
3874 3875
3875 3876 int
3876 3877 fop_putpage(
3877 3878 vnode_t *vp,
3878 3879 offset_t off,
3879 3880 size_t len,
3880 3881 int flags,
3881 3882 cred_t *cr,
3882 3883 caller_context_t *ct)
3883 3884 {
3884 3885 int err;
3885 3886
3886 3887 VOPXID_MAP_CR(vp, cr);
3887 3888
3888 3889 err = (*(vp)->v_op->vop_putpage)(vp, off, len, flags, cr, ct);
3889 3890 VOPSTATS_UPDATE(vp, putpage);
3890 3891 return (err);
3891 3892 }
3892 3893
3893 3894 int
3894 3895 fop_map(
3895 3896 vnode_t *vp,
3896 3897 offset_t off,
3897 3898 struct as *as,
3898 3899 caddr_t *addrp,
3899 3900 size_t len,
3900 3901 uchar_t prot,
3901 3902 uchar_t maxprot,
3902 3903 uint_t flags,
3903 3904 cred_t *cr,
3904 3905 caller_context_t *ct)
3905 3906 {
3906 3907 int err;
3907 3908
3908 3909 VOPXID_MAP_CR(vp, cr);
3909 3910
3910 3911 err = (*(vp)->v_op->vop_map)
3911 3912 (vp, off, as, addrp, len, prot, maxprot, flags, cr, ct);
3912 3913 VOPSTATS_UPDATE(vp, map);
3913 3914 return (err);
3914 3915 }
3915 3916
3916 3917 int
3917 3918 fop_addmap(
3918 3919 vnode_t *vp,
3919 3920 offset_t off,
3920 3921 struct as *as,
3921 3922 caddr_t addr,
3922 3923 size_t len,
3923 3924 uchar_t prot,
3924 3925 uchar_t maxprot,
3925 3926 uint_t flags,
3926 3927 cred_t *cr,
3927 3928 caller_context_t *ct)
3928 3929 {
3929 3930 int error;
3930 3931 u_longlong_t delta;
3931 3932
3932 3933 VOPXID_MAP_CR(vp, cr);
3933 3934
3934 3935 error = (*(vp)->v_op->vop_addmap)
3935 3936 (vp, off, as, addr, len, prot, maxprot, flags, cr, ct);
3936 3937
3937 3938 if ((!error) && (vp->v_type == VREG)) {
3938 3939 delta = (u_longlong_t)btopr(len);
3939 3940 /*
3940 3941 * If file is declared MAP_PRIVATE, it can't be written back
3941 3942 * even if open for write. Handle as read.
3942 3943 */
3943 3944 if (flags & MAP_PRIVATE) {
3944 3945 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3945 3946 (int64_t)delta);
3946 3947 } else {
3947 3948 /*
3948 3949 * atomic_add_64 forces the fetch of a 64 bit value to
3949 3950 * be atomic on 32 bit machines
3950 3951 */
3951 3952 if (maxprot & PROT_WRITE)
3952 3953 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)),
3953 3954 (int64_t)delta);
3954 3955 if (maxprot & PROT_READ)
3955 3956 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3956 3957 (int64_t)delta);
3957 3958 if (maxprot & PROT_EXEC)
3958 3959 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3959 3960 (int64_t)delta);
3960 3961 }
3961 3962 }
3962 3963 VOPSTATS_UPDATE(vp, addmap);
3963 3964 return (error);
3964 3965 }
3965 3966
3966 3967 int
3967 3968 fop_delmap(
3968 3969 vnode_t *vp,
3969 3970 offset_t off,
3970 3971 struct as *as,
3971 3972 caddr_t addr,
3972 3973 size_t len,
3973 3974 uint_t prot,
3974 3975 uint_t maxprot,
3975 3976 uint_t flags,
3976 3977 cred_t *cr,
3977 3978 caller_context_t *ct)
3978 3979 {
3979 3980 int error;
3980 3981 u_longlong_t delta;
3981 3982
3982 3983 VOPXID_MAP_CR(vp, cr);
3983 3984
3984 3985 error = (*(vp)->v_op->vop_delmap)
3985 3986 (vp, off, as, addr, len, prot, maxprot, flags, cr, ct);
3986 3987
3987 3988 /*
3988 3989 * NFS calls into delmap twice, the first time
3989 3990 * it simply establishes a callback mechanism and returns EAGAIN
3990 3991 * while the real work is being done upon the second invocation.
3991 3992 * We have to detect this here and only decrement the counts upon
3992 3993 * the second delmap request.
3993 3994 */
3994 3995 if ((error != EAGAIN) && (vp->v_type == VREG)) {
3995 3996
3996 3997 delta = (u_longlong_t)btopr(len);
3997 3998
3998 3999 if (flags & MAP_PRIVATE) {
3999 4000 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4000 4001 (int64_t)(-delta));
4001 4002 } else {
4002 4003 /*
4003 4004 * atomic_add_64 forces the fetch of a 64 bit value
4004 4005 * to be atomic on 32 bit machines
4005 4006 */
4006 4007 if (maxprot & PROT_WRITE)
4007 4008 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)),
4008 4009 (int64_t)(-delta));
4009 4010 if (maxprot & PROT_READ)
4010 4011 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4011 4012 (int64_t)(-delta));
4012 4013 if (maxprot & PROT_EXEC)
4013 4014 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4014 4015 (int64_t)(-delta));
4015 4016 }
4016 4017 }
4017 4018 VOPSTATS_UPDATE(vp, delmap);
4018 4019 return (error);
4019 4020 }
4020 4021
4021 4022
4022 4023 int
4023 4024 fop_poll(
4024 4025 vnode_t *vp,
4025 4026 short events,
4026 4027 int anyyet,
4027 4028 short *reventsp,
4028 4029 struct pollhead **phpp,
4029 4030 caller_context_t *ct)
4030 4031 {
4031 4032 int err;
4032 4033
4033 4034 err = (*(vp)->v_op->vop_poll)(vp, events, anyyet, reventsp, phpp, ct);
4034 4035 VOPSTATS_UPDATE(vp, poll);
4035 4036 return (err);
4036 4037 }
4037 4038
4038 4039 int
4039 4040 fop_dump(
4040 4041 vnode_t *vp,
4041 4042 caddr_t addr,
4042 4043 offset_t lbdn,
4043 4044 offset_t dblks,
4044 4045 caller_context_t *ct)
4045 4046 {
4046 4047 int err;
4047 4048
4048 4049 /* ensure lbdn and dblks can be passed safely to bdev_dump */
4049 4050 if ((lbdn != (daddr_t)lbdn) || (dblks != (int)dblks))
4050 4051 return (EIO);
4051 4052
4052 4053 err = (*(vp)->v_op->vop_dump)(vp, addr, lbdn, dblks, ct);
4053 4054 VOPSTATS_UPDATE(vp, dump);
4054 4055 return (err);
4055 4056 }
4056 4057
4057 4058 int
4058 4059 fop_pathconf(
4059 4060 vnode_t *vp,
4060 4061 int cmd,
4061 4062 ulong_t *valp,
4062 4063 cred_t *cr,
4063 4064 caller_context_t *ct)
4064 4065 {
4065 4066 int err;
4066 4067
4067 4068 VOPXID_MAP_CR(vp, cr);
4068 4069
4069 4070 err = (*(vp)->v_op->vop_pathconf)(vp, cmd, valp, cr, ct);
4070 4071 VOPSTATS_UPDATE(vp, pathconf);
4071 4072 return (err);
4072 4073 }
4073 4074
4074 4075 int
4075 4076 fop_pageio(
4076 4077 vnode_t *vp,
4077 4078 struct page *pp,
4078 4079 u_offset_t io_off,
4079 4080 size_t io_len,
4080 4081 int flags,
4081 4082 cred_t *cr,
4082 4083 caller_context_t *ct)
4083 4084 {
4084 4085 int err;
4085 4086
4086 4087 VOPXID_MAP_CR(vp, cr);
4087 4088
4088 4089 err = (*(vp)->v_op->vop_pageio)(vp, pp, io_off, io_len, flags, cr, ct);
4089 4090 VOPSTATS_UPDATE(vp, pageio);
4090 4091 return (err);
4091 4092 }
4092 4093
4093 4094 int
4094 4095 fop_dumpctl(
4095 4096 vnode_t *vp,
4096 4097 int action,
4097 4098 offset_t *blkp,
4098 4099 caller_context_t *ct)
4099 4100 {
4100 4101 int err;
4101 4102 err = (*(vp)->v_op->vop_dumpctl)(vp, action, blkp, ct);
4102 4103 VOPSTATS_UPDATE(vp, dumpctl);
4103 4104 return (err);
4104 4105 }
4105 4106
4106 4107 void
4107 4108 fop_dispose(
4108 4109 vnode_t *vp,
4109 4110 page_t *pp,
4110 4111 int flag,
4111 4112 int dn,
4112 4113 cred_t *cr,
4113 4114 caller_context_t *ct)
4114 4115 {
4115 4116 /* Must do stats first since it's possible to lose the vnode */
4116 4117 VOPSTATS_UPDATE(vp, dispose);
4117 4118
4118 4119 VOPXID_MAP_CR(vp, cr);
4119 4120
4120 4121 (*(vp)->v_op->vop_dispose)(vp, pp, flag, dn, cr, ct);
4121 4122 }
4122 4123
4123 4124 int
4124 4125 fop_setsecattr(
4125 4126 vnode_t *vp,
4126 4127 vsecattr_t *vsap,
4127 4128 int flag,
4128 4129 cred_t *cr,
4129 4130 caller_context_t *ct)
4130 4131 {
4131 4132 int err;
4132 4133
4133 4134 VOPXID_MAP_CR(vp, cr);
4134 4135
4135 4136 /*
4136 4137 * We're only allowed to skip the ACL check iff we used a 32 bit
4137 4138 * ACE mask with VOP_ACCESS() to determine permissions.
4138 4139 */
4139 4140 if ((flag & ATTR_NOACLCHECK) &&
4140 4141 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
4141 4142 return (EINVAL);
4142 4143 }
4143 4144 err = (*(vp)->v_op->vop_setsecattr) (vp, vsap, flag, cr, ct);
4144 4145 VOPSTATS_UPDATE(vp, setsecattr);
4145 4146 return (err);
4146 4147 }
4147 4148
4148 4149 int
4149 4150 fop_getsecattr(
4150 4151 vnode_t *vp,
4151 4152 vsecattr_t *vsap,
4152 4153 int flag,
4153 4154 cred_t *cr,
4154 4155 caller_context_t *ct)
4155 4156 {
4156 4157 int err;
4157 4158
4158 4159 /*
4159 4160 * We're only allowed to skip the ACL check iff we used a 32 bit
4160 4161 * ACE mask with VOP_ACCESS() to determine permissions.
4161 4162 */
4162 4163 if ((flag & ATTR_NOACLCHECK) &&
4163 4164 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
4164 4165 return (EINVAL);
4165 4166 }
4166 4167
4167 4168 VOPXID_MAP_CR(vp, cr);
4168 4169
4169 4170 err = (*(vp)->v_op->vop_getsecattr) (vp, vsap, flag, cr, ct);
4170 4171 VOPSTATS_UPDATE(vp, getsecattr);
4171 4172 return (err);
4172 4173 }
4173 4174
4174 4175 int
4175 4176 fop_shrlock(
4176 4177 vnode_t *vp,
4177 4178 int cmd,
4178 4179 struct shrlock *shr,
4179 4180 int flag,
4180 4181 cred_t *cr,
4181 4182 caller_context_t *ct)
4182 4183 {
4183 4184 int err;
4184 4185
4185 4186 VOPXID_MAP_CR(vp, cr);
4186 4187
4187 4188 err = (*(vp)->v_op->vop_shrlock)(vp, cmd, shr, flag, cr, ct);
4188 4189 VOPSTATS_UPDATE(vp, shrlock);
4189 4190 return (err);
4190 4191 }
4191 4192
4192 4193 int
4193 4194 fop_vnevent(vnode_t *vp, vnevent_t vnevent, vnode_t *dvp, char *fnm,
4194 4195 caller_context_t *ct)
4195 4196 {
4196 4197 int err;
4197 4198
4198 4199 err = (*(vp)->v_op->vop_vnevent)(vp, vnevent, dvp, fnm, ct);
4199 4200 VOPSTATS_UPDATE(vp, vnevent);
4200 4201 return (err);
4201 4202 }
4202 4203
4203 4204 int
4204 4205 fop_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *uiop, cred_t *cr,
4205 4206 caller_context_t *ct)
4206 4207 {
4207 4208 int err;
4208 4209
4209 4210 if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0)
4210 4211 return (ENOTSUP);
4211 4212 err = (*(vp)->v_op->vop_reqzcbuf)(vp, ioflag, uiop, cr, ct);
4212 4213 VOPSTATS_UPDATE(vp, reqzcbuf);
4213 4214 return (err);
4214 4215 }
4215 4216
4216 4217 int
4217 4218 fop_retzcbuf(vnode_t *vp, xuio_t *uiop, cred_t *cr, caller_context_t *ct)
4218 4219 {
4219 4220 int err;
4220 4221
4221 4222 if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0)
4222 4223 return (ENOTSUP);
4223 4224 err = (*(vp)->v_op->vop_retzcbuf)(vp, uiop, cr, ct);
4224 4225 VOPSTATS_UPDATE(vp, retzcbuf);
4225 4226 return (err);
4226 4227 }
4227 4228
4228 4229 /*
4229 4230 * Default destructor
4230 4231 * Needed because NULL destructor means that the key is unused
4231 4232 */
4232 4233 /* ARGSUSED */
4233 4234 void
4234 4235 vsd_defaultdestructor(void *value)
4235 4236 {}
4236 4237
4237 4238 /*
4238 4239 * Create a key (index into per vnode array)
4239 4240 * Locks out vsd_create, vsd_destroy, and vsd_free
4240 4241 * May allocate memory with lock held
4241 4242 */
4242 4243 void
4243 4244 vsd_create(uint_t *keyp, void (*destructor)(void *))
4244 4245 {
4245 4246 int i;
4246 4247 uint_t nkeys;
4247 4248
4248 4249 /*
4249 4250 * if key is allocated, do nothing
4250 4251 */
4251 4252 mutex_enter(&vsd_lock);
4252 4253 if (*keyp) {
4253 4254 mutex_exit(&vsd_lock);
4254 4255 return;
4255 4256 }
4256 4257 /*
4257 4258 * find an unused key
4258 4259 */
4259 4260 if (destructor == NULL)
4260 4261 destructor = vsd_defaultdestructor;
4261 4262
4262 4263 for (i = 0; i < vsd_nkeys; ++i)
4263 4264 if (vsd_destructor[i] == NULL)
4264 4265 break;
4265 4266
4266 4267 /*
4267 4268 * if no unused keys, increase the size of the destructor array
4268 4269 */
4269 4270 if (i == vsd_nkeys) {
4270 4271 if ((nkeys = (vsd_nkeys << 1)) == 0)
4271 4272 nkeys = 1;
4272 4273 vsd_destructor =
4273 4274 (void (**)(void *))vsd_realloc((void *)vsd_destructor,
4274 4275 (size_t)(vsd_nkeys * sizeof (void (*)(void *))),
4275 4276 (size_t)(nkeys * sizeof (void (*)(void *))));
4276 4277 vsd_nkeys = nkeys;
4277 4278 }
4278 4279
4279 4280 /*
4280 4281 * allocate the next available unused key
4281 4282 */
4282 4283 vsd_destructor[i] = destructor;
4283 4284 *keyp = i + 1;
4284 4285
4285 4286 /* create vsd_list, if it doesn't exist */
4286 4287 if (vsd_list == NULL) {
4287 4288 vsd_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
4288 4289 list_create(vsd_list, sizeof (struct vsd_node),
4289 4290 offsetof(struct vsd_node, vs_nodes));
4290 4291 }
4291 4292
4292 4293 mutex_exit(&vsd_lock);
4293 4294 }
4294 4295
4295 4296 /*
4296 4297 * Destroy a key
4297 4298 *
4298 4299 * Assumes that the caller is preventing vsd_set and vsd_get
4299 4300 * Locks out vsd_create, vsd_destroy, and vsd_free
4300 4301 * May free memory with lock held
4301 4302 */
4302 4303 void
4303 4304 vsd_destroy(uint_t *keyp)
4304 4305 {
4305 4306 uint_t key;
4306 4307 struct vsd_node *vsd;
4307 4308
4308 4309 /*
4309 4310 * protect the key namespace and our destructor lists
4310 4311 */
4311 4312 mutex_enter(&vsd_lock);
4312 4313 key = *keyp;
4313 4314 *keyp = 0;
4314 4315
4315 4316 ASSERT(key <= vsd_nkeys);
4316 4317
4317 4318 /*
4318 4319 * if the key is valid
4319 4320 */
4320 4321 if (key != 0) {
4321 4322 uint_t k = key - 1;
4322 4323 /*
4323 4324 * for every vnode with VSD, call key's destructor
4324 4325 */
4325 4326 for (vsd = list_head(vsd_list); vsd != NULL;
4326 4327 vsd = list_next(vsd_list, vsd)) {
4327 4328 /*
4328 4329 * no VSD for key in this vnode
4329 4330 */
4330 4331 if (key > vsd->vs_nkeys)
4331 4332 continue;
4332 4333 /*
4333 4334 * call destructor for key
4334 4335 */
4335 4336 if (vsd->vs_value[k] && vsd_destructor[k])
4336 4337 (*vsd_destructor[k])(vsd->vs_value[k]);
4337 4338 /*
4338 4339 * reset value for key
4339 4340 */
4340 4341 vsd->vs_value[k] = NULL;
4341 4342 }
4342 4343 /*
4343 4344 * actually free the key (NULL destructor == unused)
4344 4345 */
4345 4346 vsd_destructor[k] = NULL;
4346 4347 }
4347 4348
4348 4349 mutex_exit(&vsd_lock);
4349 4350 }
4350 4351
4351 4352 /*
4352 4353 * Quickly return the per vnode value that was stored with the specified key
4353 4354 * Assumes the caller is protecting key from vsd_create and vsd_destroy
4354 4355 * Assumes the caller is holding v_vsd_lock to protect the vsd.
4355 4356 */
4356 4357 void *
4357 4358 vsd_get(vnode_t *vp, uint_t key)
4358 4359 {
4359 4360 struct vsd_node *vsd;
4360 4361
4361 4362 ASSERT(vp != NULL);
4362 4363 ASSERT(mutex_owned(&vp->v_vsd_lock));
4363 4364
4364 4365 vsd = vp->v_vsd;
4365 4366
4366 4367 if (key && vsd != NULL && key <= vsd->vs_nkeys)
4367 4368 return (vsd->vs_value[key - 1]);
4368 4369 return (NULL);
4369 4370 }
4370 4371
4371 4372 /*
4372 4373 * Set a per vnode value indexed with the specified key
4373 4374 * Assumes the caller is holding v_vsd_lock to protect the vsd.
4374 4375 */
4375 4376 int
4376 4377 vsd_set(vnode_t *vp, uint_t key, void *value)
4377 4378 {
4378 4379 struct vsd_node *vsd;
4379 4380
4380 4381 ASSERT(vp != NULL);
4381 4382 ASSERT(mutex_owned(&vp->v_vsd_lock));
4382 4383
4383 4384 if (key == 0)
4384 4385 return (EINVAL);
4385 4386
4386 4387 vsd = vp->v_vsd;
4387 4388 if (vsd == NULL)
4388 4389 vsd = vp->v_vsd = kmem_zalloc(sizeof (*vsd), KM_SLEEP);
4389 4390
4390 4391 /*
4391 4392 * If the vsd was just allocated, vs_nkeys will be 0, so the following
4392 4393 * code won't happen and we will continue down and allocate space for
4393 4394 * the vs_value array.
4394 4395 * If the caller is replacing one value with another, then it is up
4395 4396 * to the caller to free/rele/destroy the previous value (if needed).
4396 4397 */
4397 4398 if (key <= vsd->vs_nkeys) {
4398 4399 vsd->vs_value[key - 1] = value;
4399 4400 return (0);
4400 4401 }
4401 4402
4402 4403 ASSERT(key <= vsd_nkeys);
4403 4404
4404 4405 if (vsd->vs_nkeys == 0) {
4405 4406 mutex_enter(&vsd_lock); /* lock out vsd_destroy() */
4406 4407 /*
4407 4408 * Link onto list of all VSD nodes.
4408 4409 */
4409 4410 list_insert_head(vsd_list, vsd);
4410 4411 mutex_exit(&vsd_lock);
4411 4412 }
4412 4413
4413 4414 /*
4414 4415 * Allocate vnode local storage and set the value for key
4415 4416 */
4416 4417 vsd->vs_value = vsd_realloc(vsd->vs_value,
4417 4418 vsd->vs_nkeys * sizeof (void *),
4418 4419 key * sizeof (void *));
4419 4420 vsd->vs_nkeys = key;
4420 4421 vsd->vs_value[key - 1] = value;
4421 4422
4422 4423 return (0);
4423 4424 }
4424 4425
4425 4426 /*
4426 4427 * Called from vn_free() to run the destructor function for each vsd
4427 4428 * Locks out vsd_create and vsd_destroy
4428 4429 * Assumes that the destructor *DOES NOT* use vsd
4429 4430 */
4430 4431 void
4431 4432 vsd_free(vnode_t *vp)
4432 4433 {
4433 4434 int i;
4434 4435 struct vsd_node *vsd = vp->v_vsd;
4435 4436
4436 4437 if (vsd == NULL)
4437 4438 return;
4438 4439
4439 4440 if (vsd->vs_nkeys == 0) {
4440 4441 kmem_free(vsd, sizeof (*vsd));
4441 4442 vp->v_vsd = NULL;
4442 4443 return;
4443 4444 }
4444 4445
4445 4446 /*
4446 4447 * lock out vsd_create and vsd_destroy, call
4447 4448 * the destructor, and mark the value as destroyed.
4448 4449 */
4449 4450 mutex_enter(&vsd_lock);
4450 4451
4451 4452 for (i = 0; i < vsd->vs_nkeys; i++) {
4452 4453 if (vsd->vs_value[i] && vsd_destructor[i])
4453 4454 (*vsd_destructor[i])(vsd->vs_value[i]);
4454 4455 vsd->vs_value[i] = NULL;
4455 4456 }
4456 4457
4457 4458 /*
4458 4459 * remove from linked list of VSD nodes
4459 4460 */
4460 4461 list_remove(vsd_list, vsd);
4461 4462
4462 4463 mutex_exit(&vsd_lock);
4463 4464
4464 4465 /*
4465 4466 * free up the VSD
4466 4467 */
4467 4468 kmem_free(vsd->vs_value, vsd->vs_nkeys * sizeof (void *));
4468 4469 kmem_free(vsd, sizeof (struct vsd_node));
4469 4470 vp->v_vsd = NULL;
4470 4471 }
4471 4472
4472 4473 /*
4473 4474 * realloc
4474 4475 */
4475 4476 static void *
4476 4477 vsd_realloc(void *old, size_t osize, size_t nsize)
4477 4478 {
4478 4479 void *new;
4479 4480
4480 4481 new = kmem_zalloc(nsize, KM_SLEEP);
4481 4482 if (old) {
4482 4483 bcopy(old, new, osize);
4483 4484 kmem_free(old, osize);
4484 4485 }
4485 4486 return (new);
4486 4487 }
4487 4488
4488 4489 /*
4489 4490 * Setup the extensible system attribute for creating a reparse point.
4490 4491 * The symlink data 'target' is validated for proper format of a reparse
4491 4492 * string and a check also made to make sure the symlink data does not
4492 4493 * point to an existing file.
4493 4494 *
4494 4495 * return 0 if ok else -1.
4495 4496 */
4496 4497 static int
4497 4498 fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr)
4498 4499 {
4499 4500 xoptattr_t *xoap;
4500 4501
4501 4502 if ((!target) || (!vap) || (!xvattr))
4502 4503 return (-1);
4503 4504
4504 4505 /* validate reparse string */
4505 4506 if (reparse_validate((const char *)target))
4506 4507 return (-1);
4507 4508
4508 4509 xva_init(xvattr);
4509 4510 xvattr->xva_vattr = *vap;
4510 4511 xvattr->xva_vattr.va_mask |= AT_XVATTR;
4511 4512 xoap = xva_getxoptattr(xvattr);
4512 4513 ASSERT(xoap);
4513 4514 XVA_SET_REQ(xvattr, XAT_REPARSE);
4514 4515 xoap->xoa_reparse = 1;
4515 4516
4516 4517 return (0);
4517 4518 }
4518 4519
4519 4520 /*
4520 4521 * Function to check whether a symlink is a reparse point.
4521 4522 * Return B_TRUE if it is a reparse point, else return B_FALSE
4522 4523 */
4523 4524 boolean_t
4524 4525 vn_is_reparse(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4525 4526 {
4526 4527 xvattr_t xvattr;
4527 4528 xoptattr_t *xoap;
4528 4529
4529 4530 if ((vp->v_type != VLNK) ||
4530 4531 !(vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR)))
4531 4532 return (B_FALSE);
4532 4533
4533 4534 xva_init(&xvattr);
4534 4535 xoap = xva_getxoptattr(&xvattr);
4535 4536 ASSERT(xoap);
4536 4537 XVA_SET_REQ(&xvattr, XAT_REPARSE);
4537 4538
4538 4539 if (VOP_GETATTR(vp, &xvattr.xva_vattr, 0, cr, ct))
4539 4540 return (B_FALSE);
4540 4541
4541 4542 if ((!(xvattr.xva_vattr.va_mask & AT_XVATTR)) ||
4542 4543 (!(XVA_ISSET_RTN(&xvattr, XAT_REPARSE))))
4543 4544 return (B_FALSE);
4544 4545
4545 4546 return (xoap->xoa_reparse ? B_TRUE : B_FALSE);
4546 4547 }
|
↓ open down ↓ |
1692 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX