1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */ 27 /* All Rights Reserved */ 28 29 /* 30 * University Copyright- Copyright (c) 1982, 1986, 1988 31 * The Regents of the University of California 32 * All Rights Reserved 33 * 34 * University Acknowledgment- Portions of this document are derived from 35 * software developed by the University of California, Berkeley, and its 36 * contributors. 37 */ 38 39 40 #pragma ident "@(#)vnode.c 1.107 07/12/26 SMI" 41 42 #include <sys/types.h> 43 #include <sys/param.h> 44 #include <sys/t_lock.h> 45 #include <sys/errno.h> 46 #include <sys/cred.h> 47 #include <sys/user.h> 48 #include <sys/uio.h> 49 #include <sys/file.h> 50 #include <sys/pathname.h> 51 #include <sys/vfs.h> 52 #include <sys/vfs_opreg.h> 53 #include <sys/vnode.h> 54 #include <sys/rwstlock.h> 55 #include <sys/fem.h> 56 #include <sys/stat.h> 57 #include <sys/mode.h> 58 #include <sys/conf.h> 59 #include <sys/sysmacros.h> 60 #include <sys/cmn_err.h> 61 #include <sys/systm.h> 62 #include <sys/kmem.h> 63 #include <sys/debug.h> 64 #include <c2/audit.h> 65 #include <sys/acl.h> 66 #include <sys/nbmlock.h> 67 #include <sys/fcntl.h> 68 #include <fs/fs_subr.h> 69 70 /* Determine if this vnode is a file that is read-only */ 71 #define ISROFILE(vp) \ 72 ((vp)->v_type != VCHR && (vp)->v_type != VBLK && \ 73 (vp)->v_type != VFIFO && vn_is_readonly(vp)) 74 75 /* Tunable via /etc/system; used only by admin/install */ 76 int nfs_global_client_only; 77 78 /* 79 * Array of vopstats_t for per-FS-type vopstats. This array has the same 80 * number of entries as and parallel to the vfssw table. (Arguably, it could 81 * be part of the vfssw table.) Once it's initialized, it's accessed using 82 * the same fstype index that is used to index into the vfssw table. 83 */ 84 vopstats_t **vopstats_fstype; 85 86 /* vopstats initialization template used for fast initialization via bcopy() */ 87 static vopstats_t *vs_templatep; 88 89 /* Kmem cache handle for vsk_anchor_t allocations */ 90 kmem_cache_t *vsk_anchor_cache; 91 92 /* file events cleanup routine */ 93 extern void free_fopdata(vnode_t *); 94 95 /* 96 * Root of AVL tree for the kstats associated with vopstats. Lock protects 97 * updates to vsktat_tree. 98 */ 99 avl_tree_t vskstat_tree; 100 kmutex_t vskstat_tree_lock; 101 102 /* Global variable which enables/disables the vopstats collection */ 103 int vopstats_enabled = 1; 104 105 /* 106 * forward declarations for internal vnode specific data (vsd) 107 */ 108 static void *vsd_realloc(void *, size_t, size_t); 109 110 /* 111 * VSD -- VNODE SPECIFIC DATA 112 * The v_data pointer is typically used by a file system to store a 113 * pointer to the file system's private node (e.g. ufs inode, nfs rnode). 114 * However, there are times when additional project private data needs 115 * to be stored separately from the data (node) pointed to by v_data. 116 * This additional data could be stored by the file system itself or 117 * by a completely different kernel entity. VSD provides a way for 118 * callers to obtain a key and store a pointer to private data associated 119 * with a vnode. 120 * 121 * Callers are responsible for protecting the vsd by holding v_lock 122 * for calls to vsd_set() and vsd_get(). 123 */ 124 125 /* 126 * vsd_lock protects: 127 * vsd_nkeys - creation and deletion of vsd keys 128 * vsd_list - insertion and deletion of vsd_node in the vsd_list 129 * vsd_destructor - adding and removing destructors to the list 130 */ 131 static kmutex_t vsd_lock; 132 static uint_t vsd_nkeys; /* size of destructor array */ 133 /* list of vsd_node's */ 134 static list_t *vsd_list = NULL; 135 /* per-key destructor funcs */ 136 static void (**vsd_destructor)(void *); 137 138 /* 139 * The following is the common set of actions needed to update the 140 * vopstats structure from a vnode op. Both VOPSTATS_UPDATE() and 141 * VOPSTATS_UPDATE_IO() do almost the same thing, except for the 142 * recording of the bytes transferred. Since the code is similar 143 * but small, it is nearly a duplicate. Consequently any changes 144 * to one may need to be reflected in the other. 145 * Rundown of the variables: 146 * vp - Pointer to the vnode 147 * counter - Partial name structure member to update in vopstats for counts 148 * bytecounter - Partial name structure member to update in vopstats for bytes 149 * bytesval - Value to update in vopstats for bytes 150 * fstype - Index into vsanchor_fstype[], same as index into vfssw[] 151 * vsp - Pointer to vopstats structure (either in vfs or vsanchor_fstype[i]) 152 */ 153 154 #define VOPSTATS_UPDATE(vp, counter) { \ 155 vfs_t *vfsp = (vp)->v_vfsp; \ 156 if (vfsp && vfsp->vfs_implp && \ 157 (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) { \ 158 vopstats_t *vsp = &vfsp->vfs_vopstats; \ 159 uint64_t *stataddr = &(vsp->n##counter.value.ui64); \ 160 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \ 161 size_t, uint64_t *); \ 162 __dtrace_probe___fsinfo_##counter(vp, 0, stataddr); \ 163 (*stataddr)++; \ 164 if ((vsp = vfsp->vfs_fstypevsp) != NULL) { \ 165 vsp->n##counter.value.ui64++; \ 166 } \ 167 } \ 168 } 169 170 #define VOPSTATS_UPDATE_IO(vp, counter, bytecounter, bytesval) { \ 171 vfs_t *vfsp = (vp)->v_vfsp; \ 172 if (vfsp && vfsp->vfs_implp && \ 173 (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) { \ 174 vopstats_t *vsp = &vfsp->vfs_vopstats; \ 175 uint64_t *stataddr = &(vsp->n##counter.value.ui64); \ 176 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \ 177 size_t, uint64_t *); \ 178 __dtrace_probe___fsinfo_##counter(vp, bytesval, stataddr); \ 179 (*stataddr)++; \ 180 vsp->bytecounter.value.ui64 += bytesval; \ 181 if ((vsp = vfsp->vfs_fstypevsp) != NULL) { \ 182 vsp->n##counter.value.ui64++; \ 183 vsp->bytecounter.value.ui64 += bytesval; \ 184 } \ 185 } \ 186 } 187 188 /* 189 * If the filesystem does not support XIDs map credential 190 * If the vfsp is NULL, perhaps we should also map? 191 */ 192 #define VOPXID_MAP_CR(vp, cr) { \ 193 vfs_t *vfsp = (vp)->v_vfsp; \ 194 if (vfsp != NULL && (vfsp->vfs_flag & VFS_XID) == 0) \ 195 cr = crgetmapped(cr); \ 196 } 197 198 /* 199 * Convert stat(2) formats to vnode types and vice versa. (Knows about 200 * numerical order of S_IFMT and vnode types.) 201 */ 202 enum vtype iftovt_tab[] = { 203 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 204 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON 205 }; 206 207 ushort_t vttoif_tab[] = { 208 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFIFO, 209 S_IFDOOR, 0, S_IFSOCK, S_IFPORT, 0 210 }; 211 212 /* 213 * The system vnode cache. 214 */ 215 216 kmem_cache_t *vn_cache; 217 218 219 /* 220 * Vnode operations vector. 221 */ 222 223 static const fs_operation_trans_def_t vn_ops_table[] = { 224 VOPNAME_OPEN, offsetof(struct vnodeops, vop_open), 225 fs_nosys, fs_nosys, 226 227 VOPNAME_CLOSE, offsetof(struct vnodeops, vop_close), 228 fs_nosys, fs_nosys, 229 230 VOPNAME_READ, offsetof(struct vnodeops, vop_read), 231 fs_nosys, fs_nosys, 232 233 VOPNAME_WRITE, offsetof(struct vnodeops, vop_write), 234 fs_nosys, fs_nosys, 235 236 VOPNAME_IOCTL, offsetof(struct vnodeops, vop_ioctl), 237 fs_nosys, fs_nosys, 238 239 VOPNAME_SETFL, offsetof(struct vnodeops, vop_setfl), 240 fs_setfl, fs_nosys, 241 242 VOPNAME_GETATTR, offsetof(struct vnodeops, vop_getattr), 243 fs_nosys, fs_nosys, 244 245 VOPNAME_SETATTR, offsetof(struct vnodeops, vop_setattr), 246 fs_nosys, fs_nosys, 247 248 VOPNAME_ACCESS, offsetof(struct vnodeops, vop_access), 249 fs_nosys, fs_nosys, 250 251 VOPNAME_LOOKUP, offsetof(struct vnodeops, vop_lookup), 252 fs_nosys, fs_nosys, 253 254 VOPNAME_CREATE, offsetof(struct vnodeops, vop_create), 255 fs_nosys, fs_nosys, 256 257 VOPNAME_REMOVE, offsetof(struct vnodeops, vop_remove), 258 fs_nosys, fs_nosys, 259 260 VOPNAME_LINK, offsetof(struct vnodeops, vop_link), 261 fs_nosys, fs_nosys, 262 263 VOPNAME_RENAME, offsetof(struct vnodeops, vop_rename), 264 fs_nosys, fs_nosys, 265 266 VOPNAME_MKDIR, offsetof(struct vnodeops, vop_mkdir), 267 fs_nosys, fs_nosys, 268 269 VOPNAME_RMDIR, offsetof(struct vnodeops, vop_rmdir), 270 fs_nosys, fs_nosys, 271 272 VOPNAME_READDIR, offsetof(struct vnodeops, vop_readdir), 273 fs_nosys, fs_nosys, 274 275 VOPNAME_SYMLINK, offsetof(struct vnodeops, vop_symlink), 276 fs_nosys, fs_nosys, 277 278 VOPNAME_READLINK, offsetof(struct vnodeops, vop_readlink), 279 fs_nosys, fs_nosys, 280 281 VOPNAME_FSYNC, offsetof(struct vnodeops, vop_fsync), 282 fs_nosys, fs_nosys, 283 284 VOPNAME_INACTIVE, offsetof(struct vnodeops, vop_inactive), 285 fs_nosys, fs_nosys, 286 287 VOPNAME_FID, offsetof(struct vnodeops, vop_fid), 288 fs_nosys, fs_nosys, 289 290 VOPNAME_RWLOCK, offsetof(struct vnodeops, vop_rwlock), 291 fs_rwlock, fs_rwlock, 292 293 VOPNAME_RWUNLOCK, offsetof(struct vnodeops, vop_rwunlock), 294 (fs_generic_func_p) fs_rwunlock, 295 (fs_generic_func_p) fs_rwunlock, /* no errors allowed */ 296 297 VOPNAME_SEEK, offsetof(struct vnodeops, vop_seek), 298 fs_nosys, fs_nosys, 299 300 VOPNAME_CMP, offsetof(struct vnodeops, vop_cmp), 301 fs_cmp, fs_cmp, /* no errors allowed */ 302 303 VOPNAME_FRLOCK, offsetof(struct vnodeops, vop_frlock), 304 fs_frlock, fs_nosys, 305 306 VOPNAME_SPACE, offsetof(struct vnodeops, vop_space), 307 fs_nosys, fs_nosys, 308 309 VOPNAME_REALVP, offsetof(struct vnodeops, vop_realvp), 310 fs_nosys, fs_nosys, 311 312 VOPNAME_GETPAGE, offsetof(struct vnodeops, vop_getpage), 313 fs_nosys, fs_nosys, 314 315 VOPNAME_PUTPAGE, offsetof(struct vnodeops, vop_putpage), 316 fs_nosys, fs_nosys, 317 318 VOPNAME_MAP, offsetof(struct vnodeops, vop_map), 319 (fs_generic_func_p) fs_nosys_map, 320 (fs_generic_func_p) fs_nosys_map, 321 322 VOPNAME_ADDMAP, offsetof(struct vnodeops, vop_addmap), 323 (fs_generic_func_p) fs_nosys_addmap, 324 (fs_generic_func_p) fs_nosys_addmap, 325 326 VOPNAME_DELMAP, offsetof(struct vnodeops, vop_delmap), 327 fs_nosys, fs_nosys, 328 329 VOPNAME_POLL, offsetof(struct vnodeops, vop_poll), 330 (fs_generic_func_p) fs_poll, (fs_generic_func_p) fs_nosys_poll, 331 332 VOPNAME_DUMP, offsetof(struct vnodeops, vop_dump), 333 fs_nosys, fs_nosys, 334 335 VOPNAME_PATHCONF, offsetof(struct vnodeops, vop_pathconf), 336 fs_pathconf, fs_nosys, 337 338 VOPNAME_PAGEIO, offsetof(struct vnodeops, vop_pageio), 339 fs_nosys, fs_nosys, 340 341 VOPNAME_DUMPCTL, offsetof(struct vnodeops, vop_dumpctl), 342 fs_nosys, fs_nosys, 343 344 VOPNAME_DISPOSE, offsetof(struct vnodeops, vop_dispose), 345 (fs_generic_func_p) fs_dispose, 346 (fs_generic_func_p) fs_nodispose, 347 348 VOPNAME_SETSECATTR, offsetof(struct vnodeops, vop_setsecattr), 349 fs_nosys, fs_nosys, 350 351 VOPNAME_GETSECATTR, offsetof(struct vnodeops, vop_getsecattr), 352 fs_fab_acl, fs_nosys, 353 354 VOPNAME_SHRLOCK, offsetof(struct vnodeops, vop_shrlock), 355 fs_shrlock, fs_nosys, 356 357 VOPNAME_VNEVENT, offsetof(struct vnodeops, vop_vnevent), 358 (fs_generic_func_p) fs_vnevent_nosupport, 359 (fs_generic_func_p) fs_vnevent_nosupport, 360 361 NULL, 0, NULL, NULL 362 }; 363 364 /* Extensible attribute (xva) routines. */ 365 366 /* 367 * Zero out the structure, set the size of the requested/returned bitmaps, 368 * set AT_XVATTR in the embedded vattr_t's va_mask, and set up the pointer 369 * to the returned attributes array. 370 */ 371 void 372 xva_init(xvattr_t *xvap) 373 { 374 bzero(xvap, sizeof (xvattr_t)); 375 xvap->xva_mapsize = XVA_MAPSIZE; 376 xvap->xva_magic = XVA_MAGIC; 377 xvap->xva_vattr.va_mask = AT_XVATTR; 378 xvap->xva_rtnattrmapp = &(xvap->xva_rtnattrmap)[0]; 379 } 380 381 /* 382 * If AT_XVATTR is set, returns a pointer to the embedded xoptattr_t 383 * structure. Otherwise, returns NULL. 384 */ 385 xoptattr_t * 386 xva_getxoptattr(xvattr_t *xvap) 387 { 388 xoptattr_t *xoap = NULL; 389 if (xvap->xva_vattr.va_mask & AT_XVATTR) 390 xoap = &xvap->xva_xoptattrs; 391 return (xoap); 392 } 393 394 /* 395 * Used by the AVL routines to compare two vsk_anchor_t structures in the tree. 396 * We use the f_fsid reported by VFS_STATVFS() since we use that for the 397 * kstat name. 398 */ 399 static int 400 vska_compar(const void *n1, const void *n2) 401 { 402 int ret; 403 ulong_t p1 = ((vsk_anchor_t *)n1)->vsk_fsid; 404 ulong_t p2 = ((vsk_anchor_t *)n2)->vsk_fsid; 405 406 if (p1 < p2) { 407 ret = -1; 408 } else if (p1 > p2) { 409 ret = 1; 410 } else { 411 ret = 0; 412 } 413 414 return (ret); 415 } 416 417 /* 418 * Used to create a single template which will be bcopy()ed to a newly 419 * allocated vsanchor_combo_t structure in new_vsanchor(), below. 420 */ 421 static vopstats_t * 422 create_vopstats_template() 423 { 424 vopstats_t *vsp; 425 426 vsp = kmem_alloc(sizeof (vopstats_t), KM_SLEEP); 427 bzero(vsp, sizeof (*vsp)); /* Start fresh */ 428 429 /* VOP_OPEN */ 430 kstat_named_init(&vsp->nopen, "nopen", KSTAT_DATA_UINT64); 431 /* VOP_CLOSE */ 432 kstat_named_init(&vsp->nclose, "nclose", KSTAT_DATA_UINT64); 433 /* VOP_READ I/O */ 434 kstat_named_init(&vsp->nread, "nread", KSTAT_DATA_UINT64); 435 kstat_named_init(&vsp->read_bytes, "read_bytes", KSTAT_DATA_UINT64); 436 /* VOP_WRITE I/O */ 437 kstat_named_init(&vsp->nwrite, "nwrite", KSTAT_DATA_UINT64); 438 kstat_named_init(&vsp->write_bytes, "write_bytes", KSTAT_DATA_UINT64); 439 /* VOP_IOCTL */ 440 kstat_named_init(&vsp->nioctl, "nioctl", KSTAT_DATA_UINT64); 441 /* VOP_SETFL */ 442 kstat_named_init(&vsp->nsetfl, "nsetfl", KSTAT_DATA_UINT64); 443 /* VOP_GETATTR */ 444 kstat_named_init(&vsp->ngetattr, "ngetattr", KSTAT_DATA_UINT64); 445 /* VOP_SETATTR */ 446 kstat_named_init(&vsp->nsetattr, "nsetattr", KSTAT_DATA_UINT64); 447 /* VOP_ACCESS */ 448 kstat_named_init(&vsp->naccess, "naccess", KSTAT_DATA_UINT64); 449 /* VOP_LOOKUP */ 450 kstat_named_init(&vsp->nlookup, "nlookup", KSTAT_DATA_UINT64); 451 /* VOP_CREATE */ 452 kstat_named_init(&vsp->ncreate, "ncreate", KSTAT_DATA_UINT64); 453 /* VOP_REMOVE */ 454 kstat_named_init(&vsp->nremove, "nremove", KSTAT_DATA_UINT64); 455 /* VOP_LINK */ 456 kstat_named_init(&vsp->nlink, "nlink", KSTAT_DATA_UINT64); 457 /* VOP_RENAME */ 458 kstat_named_init(&vsp->nrename, "nrename", KSTAT_DATA_UINT64); 459 /* VOP_MKDIR */ 460 kstat_named_init(&vsp->nmkdir, "nmkdir", KSTAT_DATA_UINT64); 461 /* VOP_RMDIR */ 462 kstat_named_init(&vsp->nrmdir, "nrmdir", KSTAT_DATA_UINT64); 463 /* VOP_READDIR I/O */ 464 kstat_named_init(&vsp->nreaddir, "nreaddir", KSTAT_DATA_UINT64); 465 kstat_named_init(&vsp->readdir_bytes, "readdir_bytes", 466 KSTAT_DATA_UINT64); 467 /* VOP_SYMLINK */ 468 kstat_named_init(&vsp->nsymlink, "nsymlink", KSTAT_DATA_UINT64); 469 /* VOP_READLINK */ 470 kstat_named_init(&vsp->nreadlink, "nreadlink", KSTAT_DATA_UINT64); 471 /* VOP_FSYNC */ 472 kstat_named_init(&vsp->nfsync, "nfsync", KSTAT_DATA_UINT64); 473 /* VOP_INACTIVE */ 474 kstat_named_init(&vsp->ninactive, "ninactive", KSTAT_DATA_UINT64); 475 /* VOP_FID */ 476 kstat_named_init(&vsp->nfid, "nfid", KSTAT_DATA_UINT64); 477 /* VOP_RWLOCK */ 478 kstat_named_init(&vsp->nrwlock, "nrwlock", KSTAT_DATA_UINT64); 479 /* VOP_RWUNLOCK */ 480 kstat_named_init(&vsp->nrwunlock, "nrwunlock", KSTAT_DATA_UINT64); 481 /* VOP_SEEK */ 482 kstat_named_init(&vsp->nseek, "nseek", KSTAT_DATA_UINT64); 483 /* VOP_CMP */ 484 kstat_named_init(&vsp->ncmp, "ncmp", KSTAT_DATA_UINT64); 485 /* VOP_FRLOCK */ 486 kstat_named_init(&vsp->nfrlock, "nfrlock", KSTAT_DATA_UINT64); 487 /* VOP_SPACE */ 488 kstat_named_init(&vsp->nspace, "nspace", KSTAT_DATA_UINT64); 489 /* VOP_REALVP */ 490 kstat_named_init(&vsp->nrealvp, "nrealvp", KSTAT_DATA_UINT64); 491 /* VOP_GETPAGE */ 492 kstat_named_init(&vsp->ngetpage, "ngetpage", KSTAT_DATA_UINT64); 493 /* VOP_PUTPAGE */ 494 kstat_named_init(&vsp->nputpage, "nputpage", KSTAT_DATA_UINT64); 495 /* VOP_MAP */ 496 kstat_named_init(&vsp->nmap, "nmap", KSTAT_DATA_UINT64); 497 /* VOP_ADDMAP */ 498 kstat_named_init(&vsp->naddmap, "naddmap", KSTAT_DATA_UINT64); 499 /* VOP_DELMAP */ 500 kstat_named_init(&vsp->ndelmap, "ndelmap", KSTAT_DATA_UINT64); 501 /* VOP_POLL */ 502 kstat_named_init(&vsp->npoll, "npoll", KSTAT_DATA_UINT64); 503 /* VOP_DUMP */ 504 kstat_named_init(&vsp->ndump, "ndump", KSTAT_DATA_UINT64); 505 /* VOP_PATHCONF */ 506 kstat_named_init(&vsp->npathconf, "npathconf", KSTAT_DATA_UINT64); 507 /* VOP_PAGEIO */ 508 kstat_named_init(&vsp->npageio, "npageio", KSTAT_DATA_UINT64); 509 /* VOP_DUMPCTL */ 510 kstat_named_init(&vsp->ndumpctl, "ndumpctl", KSTAT_DATA_UINT64); 511 /* VOP_DISPOSE */ 512 kstat_named_init(&vsp->ndispose, "ndispose", KSTAT_DATA_UINT64); 513 /* VOP_SETSECATTR */ 514 kstat_named_init(&vsp->nsetsecattr, "nsetsecattr", KSTAT_DATA_UINT64); 515 /* VOP_GETSECATTR */ 516 kstat_named_init(&vsp->ngetsecattr, "ngetsecattr", KSTAT_DATA_UINT64); 517 /* VOP_SHRLOCK */ 518 kstat_named_init(&vsp->nshrlock, "nshrlock", KSTAT_DATA_UINT64); 519 /* VOP_VNEVENT */ 520 kstat_named_init(&vsp->nvnevent, "nvnevent", KSTAT_DATA_UINT64); 521 522 return (vsp); 523 } 524 525 /* 526 * Creates a kstat structure associated with a vopstats structure. 527 */ 528 kstat_t * 529 new_vskstat(char *ksname, vopstats_t *vsp) 530 { 531 kstat_t *ksp; 532 533 if (!vopstats_enabled) { 534 return (NULL); 535 } 536 537 ksp = kstat_create("unix", 0, ksname, "misc", KSTAT_TYPE_NAMED, 538 sizeof (vopstats_t)/sizeof (kstat_named_t), 539 KSTAT_FLAG_VIRTUAL|KSTAT_FLAG_WRITABLE); 540 if (ksp) { 541 ksp->ks_data = vsp; 542 kstat_install(ksp); 543 } 544 545 return (ksp); 546 } 547 548 /* 549 * Called from vfsinit() to initialize the support mechanisms for vopstats 550 */ 551 void 552 vopstats_startup() 553 { 554 if (!vopstats_enabled) 555 return; 556 557 /* 558 * Creates the AVL tree which holds per-vfs vopstat anchors. This 559 * is necessary since we need to check if a kstat exists before we 560 * attempt to create it. Also, initialize its lock. 561 */ 562 avl_create(&vskstat_tree, vska_compar, sizeof (vsk_anchor_t), 563 offsetof(vsk_anchor_t, vsk_node)); 564 mutex_init(&vskstat_tree_lock, NULL, MUTEX_DEFAULT, NULL); 565 566 vsk_anchor_cache = kmem_cache_create("vsk_anchor_cache", 567 sizeof (vsk_anchor_t), sizeof (uintptr_t), NULL, NULL, NULL, 568 NULL, NULL, 0); 569 570 /* 571 * Set up the array of pointers for the vopstats-by-FS-type. 572 * The entries will be allocated/initialized as each file system 573 * goes through modload/mod_installfs. 574 */ 575 vopstats_fstype = (vopstats_t **)kmem_zalloc( 576 (sizeof (vopstats_t *) * nfstype), KM_SLEEP); 577 578 /* Set up the global vopstats initialization template */ 579 vs_templatep = create_vopstats_template(); 580 } 581 582 /* 583 * We need to have the all of the counters zeroed. 584 * The initialization of the vopstats_t includes on the order of 585 * 50 calls to kstat_named_init(). Rather that do that on every call, 586 * we do it once in a template (vs_templatep) then bcopy it over. 587 */ 588 void 589 initialize_vopstats(vopstats_t *vsp) 590 { 591 if (vsp == NULL) 592 return; 593 594 bcopy(vs_templatep, vsp, sizeof (vopstats_t)); 595 } 596 597 /* 598 * If possible, determine which vopstats by fstype to use and 599 * return a pointer to the caller. 600 */ 601 vopstats_t * 602 get_fstype_vopstats(vfs_t *vfsp, struct vfssw *vswp) 603 { 604 int fstype = 0; /* Index into vfssw[] */ 605 vopstats_t *vsp = NULL; 606 607 if (vfsp == NULL || (vfsp->vfs_flag & VFS_STATS) == 0 || 608 !vopstats_enabled) 609 return (NULL); 610 /* 611 * Set up the fstype. We go to so much trouble because all versions 612 * of NFS use the same fstype in their vfs even though they have 613 * distinct entries in the vfssw[] table. 614 * NOTE: A special vfs (e.g., EIO_vfs) may not have an entry. 615 */ 616 if (vswp) { 617 fstype = vswp - vfssw; /* Gets us the index */ 618 } else { 619 fstype = vfsp->vfs_fstype; 620 } 621 622 /* 623 * Point to the per-fstype vopstats. The only valid values are 624 * non-zero positive values less than the number of vfssw[] table 625 * entries. 626 */ 627 if (fstype > 0 && fstype < nfstype) { 628 vsp = vopstats_fstype[fstype]; 629 } 630 631 return (vsp); 632 } 633 634 /* 635 * Generate a kstat name, create the kstat structure, and allocate a 636 * vsk_anchor_t to hold it together. Return the pointer to the vsk_anchor_t 637 * to the caller. This must only be called from a mount. 638 */ 639 vsk_anchor_t * 640 get_vskstat_anchor(vfs_t *vfsp) 641 { 642 char kstatstr[KSTAT_STRLEN]; /* kstat name for vopstats */ 643 statvfs64_t statvfsbuf; /* Needed to find f_fsid */ 644 vsk_anchor_t *vskp = NULL; /* vfs <--> kstat anchor */ 645 kstat_t *ksp; /* Ptr to new kstat */ 646 avl_index_t where; /* Location in the AVL tree */ 647 648 if (vfsp == NULL || vfsp->vfs_implp == NULL || 649 (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled) 650 return (NULL); 651 652 /* Need to get the fsid to build a kstat name */ 653 if (VFS_STATVFS(vfsp, &statvfsbuf) == 0) { 654 /* Create a name for our kstats based on fsid */ 655 (void) snprintf(kstatstr, KSTAT_STRLEN, "%s%lx", 656 VOPSTATS_STR, statvfsbuf.f_fsid); 657 658 /* Allocate and initialize the vsk_anchor_t */ 659 vskp = kmem_cache_alloc(vsk_anchor_cache, KM_SLEEP); 660 bzero(vskp, sizeof (*vskp)); 661 vskp->vsk_fsid = statvfsbuf.f_fsid; 662 663 mutex_enter(&vskstat_tree_lock); 664 if (avl_find(&vskstat_tree, vskp, &where) == NULL) { 665 avl_insert(&vskstat_tree, vskp, where); 666 mutex_exit(&vskstat_tree_lock); 667 668 /* 669 * Now that we've got the anchor in the AVL 670 * tree, we can create the kstat. 671 */ 672 ksp = new_vskstat(kstatstr, &vfsp->vfs_vopstats); 673 if (ksp) { 674 vskp->vsk_ksp = ksp; 675 } 676 } else { 677 /* Oops, found one! Release memory and lock. */ 678 mutex_exit(&vskstat_tree_lock); 679 kmem_cache_free(vsk_anchor_cache, vskp); 680 vskp = NULL; 681 } 682 } 683 return (vskp); 684 } 685 686 /* 687 * We're in the process of tearing down the vfs and need to cleanup 688 * the data structures associated with the vopstats. Must only be called 689 * from dounmount(). 690 */ 691 void 692 teardown_vopstats(vfs_t *vfsp) 693 { 694 vsk_anchor_t *vskap; 695 avl_index_t where; 696 697 if (vfsp == NULL || vfsp->vfs_implp == NULL || 698 (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled) 699 return; 700 701 /* This is a safe check since VFS_STATS must be set (see above) */ 702 if ((vskap = vfsp->vfs_vskap) == NULL) 703 return; 704 705 /* Whack the pointer right away */ 706 vfsp->vfs_vskap = NULL; 707 708 /* Lock the tree, remove the node, and delete the kstat */ 709 mutex_enter(&vskstat_tree_lock); 710 if (avl_find(&vskstat_tree, vskap, &where)) { 711 avl_remove(&vskstat_tree, vskap); 712 } 713 714 if (vskap->vsk_ksp) { 715 kstat_delete(vskap->vsk_ksp); 716 } 717 mutex_exit(&vskstat_tree_lock); 718 719 kmem_cache_free(vsk_anchor_cache, vskap); 720 } 721 722 /* 723 * Read or write a vnode. Called from kernel code. 724 */ 725 int 726 vn_rdwr( 727 enum uio_rw rw, 728 struct vnode *vp, 729 caddr_t base, 730 ssize_t len, 731 offset_t offset, 732 enum uio_seg seg, 733 int ioflag, 734 rlim64_t ulimit, /* meaningful only if rw is UIO_WRITE */ 735 cred_t *cr, 736 ssize_t *residp) 737 { 738 struct uio uio; 739 struct iovec iov; 740 int error; 741 int in_crit = 0; 742 743 if (rw == UIO_WRITE && ISROFILE(vp)) 744 return (EROFS); 745 746 if (len < 0) 747 return (EIO); 748 749 VOPXID_MAP_CR(vp, cr); 750 751 iov.iov_base = base; 752 iov.iov_len = len; 753 uio.uio_iov = &iov; 754 uio.uio_iovcnt = 1; 755 uio.uio_loffset = offset; 756 uio.uio_segflg = (short)seg; 757 uio.uio_resid = len; 758 uio.uio_llimit = ulimit; 759 760 /* 761 * We have to enter the critical region before calling VOP_RWLOCK 762 * to avoid a deadlock with ufs. 763 */ 764 if (nbl_need_check(vp)) { 765 int svmand; 766 767 nbl_start_crit(vp, RW_READER); 768 in_crit = 1; 769 error = nbl_svmand(vp, cr, &svmand); 770 if (error != 0) 771 goto done; 772 if (nbl_conflict(vp, rw == UIO_WRITE ? NBL_WRITE : NBL_READ, 773 uio.uio_offset, uio.uio_resid, svmand, NULL)) { 774 error = EACCES; 775 goto done; 776 } 777 } 778 779 (void) VOP_RWLOCK(vp, 780 rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL); 781 if (rw == UIO_WRITE) { 782 uio.uio_fmode = FWRITE; 783 uio.uio_extflg = UIO_COPY_DEFAULT; 784 error = VOP_WRITE(vp, &uio, ioflag, cr, NULL); 785 } else { 786 uio.uio_fmode = FREAD; 787 uio.uio_extflg = UIO_COPY_CACHED; 788 error = VOP_READ(vp, &uio, ioflag, cr, NULL); 789 } 790 VOP_RWUNLOCK(vp, 791 rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL); 792 if (residp) 793 *residp = uio.uio_resid; 794 else if (uio.uio_resid) 795 error = EIO; 796 797 done: 798 if (in_crit) 799 nbl_end_crit(vp); 800 return (error); 801 } 802 803 /* 804 * Release a vnode. Call VOP_INACTIVE on last reference or 805 * decrement reference count. 806 * 807 * To avoid race conditions, the v_count is left at 1 for 808 * the call to VOP_INACTIVE. This prevents another thread 809 * from reclaiming and releasing the vnode *before* the 810 * VOP_INACTIVE routine has a chance to destroy the vnode. 811 * We can't have more than 1 thread calling VOP_INACTIVE 812 * on a vnode. 813 */ 814 void 815 vn_rele(vnode_t *vp) 816 { 817 if (vp->v_count == 0) 818 cmn_err(CE_PANIC, "vn_rele: vnode ref count 0"); 819 mutex_enter(&vp->v_lock); 820 if (vp->v_count == 1) { 821 mutex_exit(&vp->v_lock); 822 VOP_INACTIVE(vp, CRED(), NULL); 823 } else { 824 vp->v_count--; 825 mutex_exit(&vp->v_lock); 826 } 827 } 828 829 /* 830 * Like vn_rele() except that it clears v_stream under v_lock. 831 * This is used by sockfs when it dismantels the association between 832 * the sockfs node and the vnode in the underlaying file system. 833 * v_lock has to be held to prevent a thread coming through the lookupname 834 * path from accessing a stream head that is going away. 835 */ 836 void 837 vn_rele_stream(vnode_t *vp) 838 { 839 if (vp->v_count == 0) 840 cmn_err(CE_PANIC, "vn_rele: vnode ref count 0"); 841 mutex_enter(&vp->v_lock); 842 vp->v_stream = NULL; 843 if (vp->v_count == 1) { 844 mutex_exit(&vp->v_lock); 845 VOP_INACTIVE(vp, CRED(), NULL); 846 } else { 847 vp->v_count--; 848 mutex_exit(&vp->v_lock); 849 } 850 } 851 852 int 853 vn_open( 854 char *pnamep, 855 enum uio_seg seg, 856 int filemode, 857 int createmode, 858 struct vnode **vpp, 859 enum create crwhy, 860 mode_t umask) 861 { 862 return (vn_openat(pnamep, seg, filemode, createmode, vpp, crwhy, 863 umask, NULL, -1)); 864 } 865 866 867 /* 868 * Open/create a vnode. 869 * This may be callable by the kernel, the only known use 870 * of user context being that the current user credentials 871 * are used for permissions. crwhy is defined iff filemode & FCREAT. 872 */ 873 int 874 vn_openat( 875 char *pnamep, 876 enum uio_seg seg, 877 int filemode, 878 int createmode, 879 struct vnode **vpp, 880 enum create crwhy, 881 mode_t umask, 882 struct vnode *startvp, 883 int fd) 884 { 885 struct vnode *vp; 886 int mode; 887 int accessflags; 888 int error; 889 int in_crit = 0; 890 int open_done = 0; 891 int shrlock_done = 0; 892 struct vattr vattr; 893 enum symfollow follow; 894 int estale_retry = 0; 895 struct shrlock shr; 896 struct shr_locowner shr_own; 897 898 mode = 0; 899 accessflags = 0; 900 if (filemode & FREAD) 901 mode |= VREAD; 902 if (filemode & (FWRITE|FTRUNC)) 903 mode |= VWRITE; 904 if (filemode & FXATTRDIROPEN) 905 mode |= VEXEC; 906 907 /* symlink interpretation */ 908 if (filemode & FNOFOLLOW) 909 follow = NO_FOLLOW; 910 else 911 follow = FOLLOW; 912 913 if (filemode & FAPPEND) 914 accessflags |= V_APPEND; 915 916 top: 917 if (filemode & FCREAT) { 918 enum vcexcl excl; 919 920 /* 921 * Wish to create a file. 922 */ 923 vattr.va_type = VREG; 924 vattr.va_mode = createmode; 925 vattr.va_mask = AT_TYPE|AT_MODE; 926 if (filemode & FTRUNC) { 927 vattr.va_size = 0; 928 vattr.va_mask |= AT_SIZE; 929 } 930 if (filemode & FEXCL) 931 excl = EXCL; 932 else 933 excl = NONEXCL; 934 935 if (error = 936 vn_createat(pnamep, seg, &vattr, excl, mode, &vp, crwhy, 937 (filemode & ~(FTRUNC|FEXCL)), umask, startvp)) 938 return (error); 939 } else { 940 /* 941 * Wish to open a file. Just look it up. 942 */ 943 if (error = lookupnameat(pnamep, seg, follow, 944 NULLVPP, &vp, startvp)) { 945 if ((error == ESTALE) && 946 fs_need_estale_retry(estale_retry++)) 947 goto top; 948 return (error); 949 } 950 951 /* 952 * Get the attributes to check whether file is large. 953 * We do this only if the FOFFMAX flag is not set and 954 * only for regular files. 955 */ 956 957 if (!(filemode & FOFFMAX) && (vp->v_type == VREG)) { 958 vattr.va_mask = AT_SIZE; 959 if ((error = VOP_GETATTR(vp, &vattr, 0, 960 CRED(), NULL))) { 961 goto out; 962 } 963 if (vattr.va_size > (u_offset_t)MAXOFF32_T) { 964 /* 965 * Large File API - regular open fails 966 * if FOFFMAX flag is set in file mode 967 */ 968 error = EOVERFLOW; 969 goto out; 970 } 971 } 972 /* 973 * Can't write directories, active texts, or 974 * read-only filesystems. Can't truncate files 975 * on which mandatory locking is in effect. 976 */ 977 if (filemode & (FWRITE|FTRUNC)) { 978 /* 979 * Allow writable directory if VDIROPEN flag is set. 980 */ 981 if (vp->v_type == VDIR && !(vp->v_flag & VDIROPEN)) { 982 error = EISDIR; 983 goto out; 984 } 985 if (ISROFILE(vp)) { 986 error = EROFS; 987 goto out; 988 } 989 /* 990 * Can't truncate files on which 991 * sysv mandatory locking is in effect. 992 */ 993 if (filemode & FTRUNC) { 994 vnode_t *rvp; 995 996 if (VOP_REALVP(vp, &rvp, NULL) != 0) 997 rvp = vp; 998 if (rvp->v_filocks != NULL) { 999 vattr.va_mask = AT_MODE; 1000 if ((error = VOP_GETATTR(vp, 1001 &vattr, 0, CRED(), NULL)) == 0 && 1002 MANDLOCK(vp, vattr.va_mode)) 1003 error = EAGAIN; 1004 } 1005 } 1006 if (error) 1007 goto out; 1008 } 1009 /* 1010 * Check permissions. 1011 */ 1012 if (error = VOP_ACCESS(vp, mode, accessflags, CRED(), NULL)) 1013 goto out; 1014 } 1015 1016 /* 1017 * Do remaining checks for FNOFOLLOW and FNOLINKS. 1018 */ 1019 if ((filemode & FNOFOLLOW) && vp->v_type == VLNK) { 1020 error = ELOOP; 1021 goto out; 1022 } 1023 if (filemode & FNOLINKS) { 1024 vattr.va_mask = AT_NLINK; 1025 if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))) { 1026 goto out; 1027 } 1028 if (vattr.va_nlink != 1) { 1029 error = EMLINK; 1030 goto out; 1031 } 1032 } 1033 1034 /* 1035 * Opening a socket corresponding to the AF_UNIX pathname 1036 * in the filesystem name space is not supported. 1037 * However, VSOCK nodes in namefs are supported in order 1038 * to make fattach work for sockets. 1039 * 1040 * XXX This uses VOP_REALVP to distinguish between 1041 * an unopened namefs node (where VOP_REALVP returns a 1042 * different VSOCK vnode) and a VSOCK created by vn_create 1043 * in some file system (where VOP_REALVP would never return 1044 * a different vnode). 1045 */ 1046 if (vp->v_type == VSOCK) { 1047 struct vnode *nvp; 1048 1049 error = VOP_REALVP(vp, &nvp, NULL); 1050 if (error != 0 || nvp == NULL || nvp == vp || 1051 nvp->v_type != VSOCK) { 1052 error = EOPNOTSUPP; 1053 goto out; 1054 } 1055 } 1056 1057 if ((vp->v_type == VREG) && nbl_need_check(vp)) { 1058 /* get share reservation */ 1059 shr.s_access = 0; 1060 if (filemode & FWRITE) 1061 shr.s_access |= F_WRACC; 1062 if (filemode & FREAD) 1063 shr.s_access |= F_RDACC; 1064 shr.s_deny = 0; 1065 shr.s_sysid = 0; 1066 shr.s_pid = ttoproc(curthread)->p_pid; 1067 shr_own.sl_pid = shr.s_pid; 1068 shr_own.sl_id = fd; 1069 shr.s_own_len = sizeof (shr_own); 1070 shr.s_owner = (caddr_t)&shr_own; 1071 error = VOP_SHRLOCK(vp, F_SHARE_NBMAND, &shr, filemode, CRED(), 1072 NULL); 1073 if (error) 1074 goto out; 1075 shrlock_done = 1; 1076 1077 /* nbmand conflict check if truncating file */ 1078 if ((filemode & FTRUNC) && !(filemode & FCREAT)) { 1079 nbl_start_crit(vp, RW_READER); 1080 in_crit = 1; 1081 1082 vattr.va_mask = AT_SIZE; 1083 if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) 1084 goto out; 1085 if (nbl_conflict(vp, NBL_WRITE, 0, vattr.va_size, 0, 1086 NULL)) { 1087 error = EACCES; 1088 goto out; 1089 } 1090 } 1091 } 1092 1093 /* 1094 * Do opening protocol. 1095 */ 1096 error = VOP_OPEN(&vp, filemode, CRED(), NULL); 1097 if (error) 1098 goto out; 1099 open_done = 1; 1100 1101 /* 1102 * Truncate if required. 1103 */ 1104 if ((filemode & FTRUNC) && !(filemode & FCREAT)) { 1105 vattr.va_size = 0; 1106 vattr.va_mask = AT_SIZE; 1107 if ((error = VOP_SETATTR(vp, &vattr, 0, CRED(), NULL)) != 0) 1108 goto out; 1109 } 1110 out: 1111 ASSERT(vp->v_count > 0); 1112 1113 if (in_crit) { 1114 nbl_end_crit(vp); 1115 in_crit = 0; 1116 } 1117 if (error) { 1118 if (open_done) { 1119 (void) VOP_CLOSE(vp, filemode, 1, (offset_t)0, CRED(), 1120 NULL); 1121 open_done = 0; 1122 shrlock_done = 0; 1123 } 1124 if (shrlock_done) { 1125 (void) VOP_SHRLOCK(vp, F_UNSHARE, &shr, 0, CRED(), 1126 NULL); 1127 shrlock_done = 0; 1128 } 1129 1130 /* 1131 * The following clause was added to handle a problem 1132 * with NFS consistency. It is possible that a lookup 1133 * of the file to be opened succeeded, but the file 1134 * itself doesn't actually exist on the server. This 1135 * is chiefly due to the DNLC containing an entry for 1136 * the file which has been removed on the server. In 1137 * this case, we just start over. If there was some 1138 * other cause for the ESTALE error, then the lookup 1139 * of the file will fail and the error will be returned 1140 * above instead of looping around from here. 1141 */ 1142 VN_RELE(vp); 1143 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1144 goto top; 1145 } else 1146 *vpp = vp; 1147 return (error); 1148 } 1149 1150 /* 1151 * The following two accessor functions are for the NFSv4 server. Since there 1152 * is no VOP_OPEN_UP/DOWNGRADE we need a way for the NFS server to keep the 1153 * vnode open counts correct when a client "upgrades" an open or does an 1154 * open_downgrade. In NFS, an upgrade or downgrade can not only change the 1155 * open mode (add or subtract read or write), but also change the share/deny 1156 * modes. However, share reservations are not integrated with OPEN, yet, so 1157 * we need to handle each separately. These functions are cleaner than having 1158 * the NFS server manipulate the counts directly, however, nobody else should 1159 * use these functions. 1160 */ 1161 void 1162 vn_open_upgrade( 1163 vnode_t *vp, 1164 int filemode) 1165 { 1166 ASSERT(vp->v_type == VREG); 1167 1168 if (filemode & FREAD) 1169 atomic_add_32(&(vp->v_rdcnt), 1); 1170 if (filemode & FWRITE) 1171 atomic_add_32(&(vp->v_wrcnt), 1); 1172 1173 } 1174 1175 void 1176 vn_open_downgrade( 1177 vnode_t *vp, 1178 int filemode) 1179 { 1180 ASSERT(vp->v_type == VREG); 1181 1182 if (filemode & FREAD) { 1183 ASSERT(vp->v_rdcnt > 0); 1184 atomic_add_32(&(vp->v_rdcnt), -1); 1185 } 1186 if (filemode & FWRITE) { 1187 ASSERT(vp->v_wrcnt > 0); 1188 atomic_add_32(&(vp->v_wrcnt), -1); 1189 } 1190 1191 } 1192 1193 int 1194 vn_create( 1195 char *pnamep, 1196 enum uio_seg seg, 1197 struct vattr *vap, 1198 enum vcexcl excl, 1199 int mode, 1200 struct vnode **vpp, 1201 enum create why, 1202 int flag, 1203 mode_t umask) 1204 { 1205 return (vn_createat(pnamep, seg, vap, excl, mode, vpp, why, flag, 1206 umask, NULL)); 1207 } 1208 1209 /* 1210 * Create a vnode (makenode). 1211 */ 1212 int 1213 vn_createat( 1214 char *pnamep, 1215 enum uio_seg seg, 1216 struct vattr *vap, 1217 enum vcexcl excl, 1218 int mode, 1219 struct vnode **vpp, 1220 enum create why, 1221 int flag, 1222 mode_t umask, 1223 struct vnode *startvp) 1224 { 1225 struct vnode *dvp; /* ptr to parent dir vnode */ 1226 struct vnode *vp = NULL; 1227 struct pathname pn; 1228 int error; 1229 int in_crit = 0; 1230 struct vattr vattr; 1231 enum symfollow follow; 1232 int estale_retry = 0; 1233 1234 ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 1235 1236 /* symlink interpretation */ 1237 if ((flag & FNOFOLLOW) || excl == EXCL) 1238 follow = NO_FOLLOW; 1239 else 1240 follow = FOLLOW; 1241 flag &= ~(FNOFOLLOW|FNOLINKS); 1242 1243 top: 1244 /* 1245 * Lookup directory. 1246 * If new object is a file, call lower level to create it. 1247 * Note that it is up to the lower level to enforce exclusive 1248 * creation, if the file is already there. 1249 * This allows the lower level to do whatever 1250 * locking or protocol that is needed to prevent races. 1251 * If the new object is directory call lower level to make 1252 * the new directory, with "." and "..". 1253 */ 1254 if (error = pn_get(pnamep, seg, &pn)) 1255 return (error); 1256 if (audit_active) 1257 audit_vncreate_start(); 1258 dvp = NULL; 1259 *vpp = NULL; 1260 /* 1261 * lookup will find the parent directory for the vnode. 1262 * When it is done the pn holds the name of the entry 1263 * in the directory. 1264 * If this is a non-exclusive create we also find the node itself. 1265 */ 1266 error = lookuppnat(&pn, NULL, follow, &dvp, 1267 (excl == EXCL) ? NULLVPP : vpp, startvp); 1268 if (error) { 1269 pn_free(&pn); 1270 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1271 goto top; 1272 if (why == CRMKDIR && error == EINVAL) 1273 error = EEXIST; /* SVID */ 1274 return (error); 1275 } 1276 1277 if (why != CRMKNOD) 1278 vap->va_mode &= ~VSVTX; 1279 1280 /* 1281 * If default ACLs are defined for the directory don't apply the 1282 * umask if umask is passed. 1283 */ 1284 1285 if (umask) { 1286 1287 vsecattr_t vsec; 1288 1289 vsec.vsa_aclcnt = 0; 1290 vsec.vsa_aclentp = NULL; 1291 vsec.vsa_dfaclcnt = 0; 1292 vsec.vsa_dfaclentp = NULL; 1293 vsec.vsa_mask = VSA_DFACLCNT; 1294 error = VOP_GETSECATTR(dvp, &vsec, 0, CRED(), NULL); 1295 /* 1296 * If error is ENOSYS then treat it as no error 1297 * Don't want to force all file systems to support 1298 * aclent_t style of ACL's. 1299 */ 1300 if (error == ENOSYS) 1301 error = 0; 1302 if (error) { 1303 if (*vpp != NULL) 1304 VN_RELE(*vpp); 1305 goto out; 1306 } else { 1307 /* 1308 * Apply the umask if no default ACLs. 1309 */ 1310 if (vsec.vsa_dfaclcnt == 0) 1311 vap->va_mode &= ~umask; 1312 1313 /* 1314 * VOP_GETSECATTR() may have allocated memory for 1315 * ACLs we didn't request, so double-check and 1316 * free it if necessary. 1317 */ 1318 if (vsec.vsa_aclcnt && vsec.vsa_aclentp != NULL) 1319 kmem_free((caddr_t)vsec.vsa_aclentp, 1320 vsec.vsa_aclcnt * sizeof (aclent_t)); 1321 if (vsec.vsa_dfaclcnt && vsec.vsa_dfaclentp != NULL) 1322 kmem_free((caddr_t)vsec.vsa_dfaclentp, 1323 vsec.vsa_dfaclcnt * sizeof (aclent_t)); 1324 } 1325 } 1326 1327 /* 1328 * In general we want to generate EROFS if the file system is 1329 * readonly. However, POSIX (IEEE Std. 1003.1) section 5.3.1 1330 * documents the open system call, and it says that O_CREAT has no 1331 * effect if the file already exists. Bug 1119649 states 1332 * that open(path, O_CREAT, ...) fails when attempting to open an 1333 * existing file on a read only file system. Thus, the first part 1334 * of the following if statement has 3 checks: 1335 * if the file exists && 1336 * it is being open with write access && 1337 * the file system is read only 1338 * then generate EROFS 1339 */ 1340 if ((*vpp != NULL && (mode & VWRITE) && ISROFILE(*vpp)) || 1341 (*vpp == NULL && dvp->v_vfsp->vfs_flag & VFS_RDONLY)) { 1342 if (*vpp) 1343 VN_RELE(*vpp); 1344 error = EROFS; 1345 } else if (excl == NONEXCL && *vpp != NULL) { 1346 vnode_t *rvp; 1347 1348 /* 1349 * File already exists. If a mandatory lock has been 1350 * applied, return error. 1351 */ 1352 vp = *vpp; 1353 if (VOP_REALVP(vp, &rvp, NULL) != 0) 1354 rvp = vp; 1355 if ((vap->va_mask & AT_SIZE) && nbl_need_check(vp)) { 1356 nbl_start_crit(vp, RW_READER); 1357 in_crit = 1; 1358 } 1359 if (rvp->v_filocks != NULL || rvp->v_shrlocks != NULL) { 1360 vattr.va_mask = AT_MODE|AT_SIZE; 1361 if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) { 1362 goto out; 1363 } 1364 if (MANDLOCK(vp, vattr.va_mode)) { 1365 error = EAGAIN; 1366 goto out; 1367 } 1368 /* 1369 * File cannot be truncated if non-blocking mandatory 1370 * locks are currently on the file. 1371 */ 1372 if ((vap->va_mask & AT_SIZE) && in_crit) { 1373 u_offset_t offset; 1374 ssize_t length; 1375 1376 offset = vap->va_size > vattr.va_size ? 1377 vattr.va_size : vap->va_size; 1378 length = vap->va_size > vattr.va_size ? 1379 vap->va_size - vattr.va_size : 1380 vattr.va_size - vap->va_size; 1381 if (nbl_conflict(vp, NBL_WRITE, offset, 1382 length, 0, NULL)) { 1383 error = EACCES; 1384 goto out; 1385 } 1386 } 1387 } 1388 1389 /* 1390 * If the file is the root of a VFS, we've crossed a 1391 * mount point and the "containing" directory that we 1392 * acquired above (dvp) is irrelevant because it's in 1393 * a different file system. We apply VOP_CREATE to the 1394 * target itself instead of to the containing directory 1395 * and supply a null path name to indicate (conventionally) 1396 * the node itself as the "component" of interest. 1397 * 1398 * The intercession of the file system is necessary to 1399 * ensure that the appropriate permission checks are 1400 * done. 1401 */ 1402 if (vp->v_flag & VROOT) { 1403 ASSERT(why != CRMKDIR); 1404 error = VOP_CREATE(vp, "", vap, excl, mode, vpp, 1405 CRED(), flag, NULL, NULL); 1406 /* 1407 * If the create succeeded, it will have created 1408 * a new reference to the vnode. Give up the 1409 * original reference. The assertion should not 1410 * get triggered because NBMAND locks only apply to 1411 * VREG files. And if in_crit is non-zero for some 1412 * reason, detect that here, rather than when we 1413 * deference a null vp. 1414 */ 1415 ASSERT(in_crit == 0); 1416 VN_RELE(vp); 1417 vp = NULL; 1418 goto out; 1419 } 1420 1421 /* 1422 * Large File API - non-large open (FOFFMAX flag not set) 1423 * of regular file fails if the file size exceeds MAXOFF32_T. 1424 */ 1425 if (why != CRMKDIR && 1426 !(flag & FOFFMAX) && 1427 (vp->v_type == VREG)) { 1428 vattr.va_mask = AT_SIZE; 1429 if ((error = VOP_GETATTR(vp, &vattr, 0, 1430 CRED(), NULL))) { 1431 goto out; 1432 } 1433 if ((vattr.va_size > (u_offset_t)MAXOFF32_T)) { 1434 error = EOVERFLOW; 1435 goto out; 1436 } 1437 } 1438 } 1439 1440 if (error == 0) { 1441 /* 1442 * Call mkdir() if specified, otherwise create(). 1443 */ 1444 int must_be_dir = pn_fixslash(&pn); /* trailing '/'? */ 1445 1446 if (why == CRMKDIR) 1447 /* 1448 * N.B., if vn_createat() ever requests 1449 * case-insensitive behavior then it will need 1450 * to be passed to VOP_MKDIR(). VOP_CREATE() 1451 * will already get it via "flag" 1452 */ 1453 error = VOP_MKDIR(dvp, pn.pn_path, vap, vpp, CRED(), 1454 NULL, 0, NULL); 1455 else if (!must_be_dir) 1456 error = VOP_CREATE(dvp, pn.pn_path, vap, 1457 excl, mode, vpp, CRED(), flag, NULL, NULL); 1458 else 1459 error = ENOTDIR; 1460 } 1461 1462 out: 1463 1464 if (audit_active) 1465 audit_vncreate_finish(*vpp, error); 1466 if (in_crit) { 1467 nbl_end_crit(vp); 1468 in_crit = 0; 1469 } 1470 if (vp != NULL) { 1471 VN_RELE(vp); 1472 vp = NULL; 1473 } 1474 pn_free(&pn); 1475 VN_RELE(dvp); 1476 /* 1477 * The following clause was added to handle a problem 1478 * with NFS consistency. It is possible that a lookup 1479 * of the file to be created succeeded, but the file 1480 * itself doesn't actually exist on the server. This 1481 * is chiefly due to the DNLC containing an entry for 1482 * the file which has been removed on the server. In 1483 * this case, we just start over. If there was some 1484 * other cause for the ESTALE error, then the lookup 1485 * of the file will fail and the error will be returned 1486 * above instead of looping around from here. 1487 */ 1488 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1489 goto top; 1490 return (error); 1491 } 1492 1493 int 1494 vn_link(char *from, char *to, enum uio_seg seg) 1495 { 1496 struct vnode *fvp; /* from vnode ptr */ 1497 struct vnode *tdvp; /* to directory vnode ptr */ 1498 struct pathname pn; 1499 int error; 1500 struct vattr vattr; 1501 dev_t fsid; 1502 int estale_retry = 0; 1503 1504 top: 1505 fvp = tdvp = NULL; 1506 if (error = pn_get(to, seg, &pn)) 1507 return (error); 1508 if (error = lookupname(from, seg, NO_FOLLOW, NULLVPP, &fvp)) 1509 goto out; 1510 if (error = lookuppn(&pn, NULL, NO_FOLLOW, &tdvp, NULLVPP)) 1511 goto out; 1512 /* 1513 * Make sure both source vnode and target directory vnode are 1514 * in the same vfs and that it is writeable. 1515 */ 1516 vattr.va_mask = AT_FSID; 1517 if (error = VOP_GETATTR(fvp, &vattr, 0, CRED(), NULL)) 1518 goto out; 1519 fsid = vattr.va_fsid; 1520 vattr.va_mask = AT_FSID; 1521 if (error = VOP_GETATTR(tdvp, &vattr, 0, CRED(), NULL)) 1522 goto out; 1523 if (fsid != vattr.va_fsid) { 1524 error = EXDEV; 1525 goto out; 1526 } 1527 if (tdvp->v_vfsp->vfs_flag & VFS_RDONLY) { 1528 error = EROFS; 1529 goto out; 1530 } 1531 /* 1532 * Do the link. 1533 */ 1534 (void) pn_fixslash(&pn); 1535 error = VOP_LINK(tdvp, fvp, pn.pn_path, CRED(), NULL, 0); 1536 out: 1537 pn_free(&pn); 1538 if (fvp) 1539 VN_RELE(fvp); 1540 if (tdvp) 1541 VN_RELE(tdvp); 1542 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1543 goto top; 1544 return (error); 1545 } 1546 1547 int 1548 vn_rename(char *from, char *to, enum uio_seg seg) 1549 { 1550 return (vn_renameat(NULL, from, NULL, to, seg)); 1551 } 1552 1553 int 1554 vn_renameat(vnode_t *fdvp, char *fname, vnode_t *tdvp, 1555 char *tname, enum uio_seg seg) 1556 { 1557 int error; 1558 struct vattr vattr; 1559 struct pathname fpn; /* from pathname */ 1560 struct pathname tpn; /* to pathname */ 1561 dev_t fsid; 1562 int in_crit_src, in_crit_targ; 1563 vnode_t *fromvp, *fvp; 1564 vnode_t *tovp, *targvp; 1565 int estale_retry = 0; 1566 1567 top: 1568 fvp = fromvp = tovp = targvp = NULL; 1569 in_crit_src = in_crit_targ = 0; 1570 /* 1571 * Get to and from pathnames. 1572 */ 1573 if (error = pn_get(fname, seg, &fpn)) 1574 return (error); 1575 if (error = pn_get(tname, seg, &tpn)) { 1576 pn_free(&fpn); 1577 return (error); 1578 } 1579 1580 /* 1581 * First we need to resolve the correct directories 1582 * The passed in directories may only be a starting point, 1583 * but we need the real directories the file(s) live in. 1584 * For example the fname may be something like usr/lib/sparc 1585 * and we were passed in the / directory, but we need to 1586 * use the lib directory for the rename. 1587 */ 1588 1589 if (audit_active) 1590 audit_setfsat_path(1); 1591 /* 1592 * Lookup to and from directories. 1593 */ 1594 if (error = lookuppnat(&fpn, NULL, NO_FOLLOW, &fromvp, &fvp, fdvp)) { 1595 goto out; 1596 } 1597 1598 /* 1599 * Make sure there is an entry. 1600 */ 1601 if (fvp == NULL) { 1602 error = ENOENT; 1603 goto out; 1604 } 1605 1606 if (audit_active) 1607 audit_setfsat_path(3); 1608 if (error = lookuppnat(&tpn, NULL, NO_FOLLOW, &tovp, &targvp, tdvp)) { 1609 goto out; 1610 } 1611 1612 /* 1613 * Make sure both the from vnode directory and the to directory 1614 * are in the same vfs and the to directory is writable. 1615 * We check fsid's, not vfs pointers, so loopback fs works. 1616 */ 1617 if (fromvp != tovp) { 1618 vattr.va_mask = AT_FSID; 1619 if (error = VOP_GETATTR(fromvp, &vattr, 0, CRED(), NULL)) 1620 goto out; 1621 fsid = vattr.va_fsid; 1622 vattr.va_mask = AT_FSID; 1623 if (error = VOP_GETATTR(tovp, &vattr, 0, CRED(), NULL)) 1624 goto out; 1625 if (fsid != vattr.va_fsid) { 1626 error = EXDEV; 1627 goto out; 1628 } 1629 } 1630 1631 if (tovp->v_vfsp->vfs_flag & VFS_RDONLY) { 1632 error = EROFS; 1633 goto out; 1634 } 1635 1636 if (targvp && (fvp != targvp)) { 1637 nbl_start_crit(targvp, RW_READER); 1638 in_crit_targ = 1; 1639 if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) { 1640 error = EACCES; 1641 goto out; 1642 } 1643 } 1644 1645 if (nbl_need_check(fvp)) { 1646 nbl_start_crit(fvp, RW_READER); 1647 in_crit_src = 1; 1648 if (nbl_conflict(fvp, NBL_RENAME, 0, 0, 0, NULL)) { 1649 error = EACCES; 1650 goto out; 1651 } 1652 } 1653 1654 /* 1655 * Do the rename. 1656 */ 1657 (void) pn_fixslash(&tpn); 1658 error = VOP_RENAME(fromvp, fpn.pn_path, tovp, tpn.pn_path, CRED(), 1659 NULL, 0); 1660 1661 out: 1662 pn_free(&fpn); 1663 pn_free(&tpn); 1664 if (in_crit_src) 1665 nbl_end_crit(fvp); 1666 if (in_crit_targ) 1667 nbl_end_crit(targvp); 1668 if (fromvp) 1669 VN_RELE(fromvp); 1670 if (tovp) 1671 VN_RELE(tovp); 1672 if (targvp) 1673 VN_RELE(targvp); 1674 if (fvp) 1675 VN_RELE(fvp); 1676 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1677 goto top; 1678 return (error); 1679 } 1680 1681 /* 1682 * Remove a file or directory. 1683 */ 1684 int 1685 vn_remove(char *fnamep, enum uio_seg seg, enum rm dirflag) 1686 { 1687 return (vn_removeat(NULL, fnamep, seg, dirflag)); 1688 } 1689 1690 int 1691 vn_removeat(vnode_t *startvp, char *fnamep, enum uio_seg seg, enum rm dirflag) 1692 { 1693 struct vnode *vp; /* entry vnode */ 1694 struct vnode *dvp; /* ptr to parent dir vnode */ 1695 struct vnode *coveredvp; 1696 struct pathname pn; /* name of entry */ 1697 enum vtype vtype; 1698 int error; 1699 struct vfs *vfsp; 1700 struct vfs *dvfsp; /* ptr to parent dir vfs */ 1701 int in_crit = 0; 1702 int estale_retry = 0; 1703 1704 top: 1705 if (error = pn_get(fnamep, seg, &pn)) 1706 return (error); 1707 dvp = vp = NULL; 1708 if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &dvp, &vp, startvp)) { 1709 pn_free(&pn); 1710 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1711 goto top; 1712 return (error); 1713 } 1714 1715 /* 1716 * Make sure there is an entry. 1717 */ 1718 if (vp == NULL) { 1719 error = ENOENT; 1720 goto out; 1721 } 1722 1723 vfsp = vp->v_vfsp; 1724 dvfsp = dvp->v_vfsp; 1725 1726 /* 1727 * If the named file is the root of a mounted filesystem, fail, 1728 * unless it's marked unlinkable. In that case, unmount the 1729 * filesystem and proceed to unlink the covered vnode. (If the 1730 * covered vnode is a directory, use rmdir instead of unlink, 1731 * to avoid file system corruption.) 1732 */ 1733 if (vp->v_flag & VROOT) { 1734 if ((vfsp->vfs_flag & VFS_UNLINKABLE) == 0) { 1735 error = EBUSY; 1736 goto out; 1737 } 1738 1739 /* 1740 * Namefs specific code starts here. 1741 */ 1742 1743 if (dirflag == RMDIRECTORY) { 1744 /* 1745 * User called rmdir(2) on a file that has 1746 * been namefs mounted on top of. Since 1747 * namefs doesn't allow directories to 1748 * be mounted on other files we know 1749 * vp is not of type VDIR so fail to operation. 1750 */ 1751 error = ENOTDIR; 1752 goto out; 1753 } 1754 1755 /* 1756 * If VROOT is still set after grabbing vp->v_lock, 1757 * noone has finished nm_unmount so far and coveredvp 1758 * is valid. 1759 * If we manage to grab vn_vfswlock(coveredvp) before releasing 1760 * vp->v_lock, any race window is eliminated. 1761 */ 1762 1763 mutex_enter(&vp->v_lock); 1764 if ((vp->v_flag & VROOT) == 0) { 1765 /* Someone beat us to the unmount */ 1766 mutex_exit(&vp->v_lock); 1767 error = EBUSY; 1768 goto out; 1769 } 1770 vfsp = vp->v_vfsp; 1771 coveredvp = vfsp->vfs_vnodecovered; 1772 ASSERT(coveredvp); 1773 /* 1774 * Note: Implementation of vn_vfswlock shows that ordering of 1775 * v_lock / vn_vfswlock is not an issue here. 1776 */ 1777 error = vn_vfswlock(coveredvp); 1778 mutex_exit(&vp->v_lock); 1779 1780 if (error) 1781 goto out; 1782 1783 VN_HOLD(coveredvp); 1784 VN_RELE(vp); 1785 error = dounmount(vfsp, 0, CRED()); 1786 1787 /* 1788 * Unmounted the namefs file system; now get 1789 * the object it was mounted over. 1790 */ 1791 vp = coveredvp; 1792 /* 1793 * If namefs was mounted over a directory, then 1794 * we want to use rmdir() instead of unlink(). 1795 */ 1796 if (vp->v_type == VDIR) 1797 dirflag = RMDIRECTORY; 1798 1799 if (error) 1800 goto out; 1801 } 1802 1803 /* 1804 * Make sure filesystem is writeable. 1805 * We check the parent directory's vfs in case this is an lofs vnode. 1806 */ 1807 if (dvfsp && dvfsp->vfs_flag & VFS_RDONLY) { 1808 error = EROFS; 1809 goto out; 1810 } 1811 1812 vtype = vp->v_type; 1813 1814 /* 1815 * If there is the possibility of an nbmand share reservation, make 1816 * sure it's okay to remove the file. Keep a reference to the 1817 * vnode, so that we can exit the nbl critical region after 1818 * calling VOP_REMOVE. 1819 * If there is no possibility of an nbmand share reservation, 1820 * release the vnode reference now. Filesystems like NFS may 1821 * behave differently if there is an extra reference, so get rid of 1822 * this one. Fortunately, we can't have nbmand mounts on NFS 1823 * filesystems. 1824 */ 1825 if (nbl_need_check(vp)) { 1826 nbl_start_crit(vp, RW_READER); 1827 in_crit = 1; 1828 if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) { 1829 error = EACCES; 1830 goto out; 1831 } 1832 } else { 1833 VN_RELE(vp); 1834 vp = NULL; 1835 } 1836 1837 if (dirflag == RMDIRECTORY) { 1838 /* 1839 * Caller is using rmdir(2), which can only be applied to 1840 * directories. 1841 */ 1842 if (vtype != VDIR) { 1843 error = ENOTDIR; 1844 } else { 1845 vnode_t *cwd; 1846 proc_t *pp = curproc; 1847 1848 mutex_enter(&pp->p_lock); 1849 cwd = PTOU(pp)->u_cdir; 1850 VN_HOLD(cwd); 1851 mutex_exit(&pp->p_lock); 1852 error = VOP_RMDIR(dvp, pn.pn_path, cwd, CRED(), 1853 NULL, 0); 1854 VN_RELE(cwd); 1855 } 1856 } else { 1857 /* 1858 * Unlink(2) can be applied to anything. 1859 */ 1860 error = VOP_REMOVE(dvp, pn.pn_path, CRED(), NULL, 0); 1861 } 1862 1863 out: 1864 pn_free(&pn); 1865 if (in_crit) { 1866 nbl_end_crit(vp); 1867 in_crit = 0; 1868 } 1869 if (vp != NULL) 1870 VN_RELE(vp); 1871 if (dvp != NULL) 1872 VN_RELE(dvp); 1873 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1874 goto top; 1875 return (error); 1876 } 1877 1878 /* 1879 * Utility function to compare equality of vnodes. 1880 * Compare the underlying real vnodes, if there are underlying vnodes. 1881 * This is a more thorough comparison than the VN_CMP() macro provides. 1882 */ 1883 int 1884 vn_compare(vnode_t *vp1, vnode_t *vp2) 1885 { 1886 vnode_t *realvp; 1887 1888 if (vp1 != NULL && VOP_REALVP(vp1, &realvp, NULL) == 0) 1889 vp1 = realvp; 1890 if (vp2 != NULL && VOP_REALVP(vp2, &realvp, NULL) == 0) 1891 vp2 = realvp; 1892 return (VN_CMP(vp1, vp2)); 1893 } 1894 1895 /* 1896 * The number of locks to hash into. This value must be a power 1897 * of 2 minus 1 and should probably also be prime. 1898 */ 1899 #define NUM_BUCKETS 1023 1900 1901 struct vn_vfslocks_bucket { 1902 kmutex_t vb_lock; 1903 vn_vfslocks_entry_t *vb_list; 1904 char pad[64 - sizeof (kmutex_t) - sizeof (void *)]; 1905 }; 1906 1907 /* 1908 * Total number of buckets will be NUM_BUCKETS + 1 . 1909 */ 1910 1911 #pragma align 64(vn_vfslocks_buckets) 1912 static struct vn_vfslocks_bucket vn_vfslocks_buckets[NUM_BUCKETS + 1]; 1913 1914 #define VN_VFSLOCKS_SHIFT 9 1915 1916 #define VN_VFSLOCKS_HASH(vfsvpptr) \ 1917 ((((intptr_t)(vfsvpptr)) >> VN_VFSLOCKS_SHIFT) & NUM_BUCKETS) 1918 1919 /* 1920 * vn_vfslocks_getlock() uses an HASH scheme to generate 1921 * rwstlock using vfs/vnode pointer passed to it. 1922 * 1923 * vn_vfslocks_rele() releases a reference in the 1924 * HASH table which allows the entry allocated by 1925 * vn_vfslocks_getlock() to be freed at a later 1926 * stage when the refcount drops to zero. 1927 */ 1928 1929 vn_vfslocks_entry_t * 1930 vn_vfslocks_getlock(void *vfsvpptr) 1931 { 1932 struct vn_vfslocks_bucket *bp; 1933 vn_vfslocks_entry_t *vep; 1934 vn_vfslocks_entry_t *tvep; 1935 1936 ASSERT(vfsvpptr != NULL); 1937 bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vfsvpptr)]; 1938 1939 mutex_enter(&bp->vb_lock); 1940 for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) { 1941 if (vep->ve_vpvfs == vfsvpptr) { 1942 vep->ve_refcnt++; 1943 mutex_exit(&bp->vb_lock); 1944 return (vep); 1945 } 1946 } 1947 mutex_exit(&bp->vb_lock); 1948 vep = kmem_alloc(sizeof (*vep), KM_SLEEP); 1949 rwst_init(&vep->ve_lock, NULL, RW_DEFAULT, NULL); 1950 vep->ve_vpvfs = (char *)vfsvpptr; 1951 vep->ve_refcnt = 1; 1952 mutex_enter(&bp->vb_lock); 1953 for (tvep = bp->vb_list; tvep != NULL; tvep = tvep->ve_next) { 1954 if (tvep->ve_vpvfs == vfsvpptr) { 1955 tvep->ve_refcnt++; 1956 mutex_exit(&bp->vb_lock); 1957 1958 /* 1959 * There is already an entry in the hash 1960 * destroy what we just allocated. 1961 */ 1962 rwst_destroy(&vep->ve_lock); 1963 kmem_free(vep, sizeof (*vep)); 1964 return (tvep); 1965 } 1966 } 1967 vep->ve_next = bp->vb_list; 1968 bp->vb_list = vep; 1969 mutex_exit(&bp->vb_lock); 1970 return (vep); 1971 } 1972 1973 void 1974 vn_vfslocks_rele(vn_vfslocks_entry_t *vepent) 1975 { 1976 struct vn_vfslocks_bucket *bp; 1977 vn_vfslocks_entry_t *vep; 1978 vn_vfslocks_entry_t *pvep; 1979 1980 ASSERT(vepent != NULL); 1981 ASSERT(vepent->ve_vpvfs != NULL); 1982 1983 bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vepent->ve_vpvfs)]; 1984 1985 mutex_enter(&bp->vb_lock); 1986 vepent->ve_refcnt--; 1987 1988 if ((int32_t)vepent->ve_refcnt < 0) 1989 cmn_err(CE_PANIC, "vn_vfslocks_rele: refcount negative"); 1990 1991 if (vepent->ve_refcnt == 0) { 1992 for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) { 1993 if (vep->ve_vpvfs == vepent->ve_vpvfs) { 1994 if (bp->vb_list == vep) 1995 bp->vb_list = vep->ve_next; 1996 else { 1997 /* LINTED */ 1998 pvep->ve_next = vep->ve_next; 1999 } 2000 mutex_exit(&bp->vb_lock); 2001 rwst_destroy(&vep->ve_lock); 2002 kmem_free(vep, sizeof (*vep)); 2003 return; 2004 } 2005 pvep = vep; 2006 } 2007 cmn_err(CE_PANIC, "vn_vfslocks_rele: vp/vfs not found"); 2008 } 2009 mutex_exit(&bp->vb_lock); 2010 } 2011 2012 /* 2013 * vn_vfswlock_wait is used to implement a lock which is logically a writers 2014 * lock protecting the v_vfsmountedhere field. 2015 * vn_vfswlock_wait has been modified to be similar to vn_vfswlock, 2016 * except that it blocks to acquire the lock VVFSLOCK. 2017 * 2018 * traverse() and routines re-implementing part of traverse (e.g. autofs) 2019 * need to hold this lock. mount(), vn_rename(), vn_remove() and so on 2020 * need the non-blocking version of the writers lock i.e. vn_vfswlock 2021 */ 2022 int 2023 vn_vfswlock_wait(vnode_t *vp) 2024 { 2025 int retval; 2026 vn_vfslocks_entry_t *vpvfsentry; 2027 ASSERT(vp != NULL); 2028 2029 vpvfsentry = vn_vfslocks_getlock(vp); 2030 retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_WRITER); 2031 2032 if (retval == EINTR) { 2033 vn_vfslocks_rele(vpvfsentry); 2034 return (EINTR); 2035 } 2036 return (retval); 2037 } 2038 2039 int 2040 vn_vfsrlock_wait(vnode_t *vp) 2041 { 2042 int retval; 2043 vn_vfslocks_entry_t *vpvfsentry; 2044 ASSERT(vp != NULL); 2045 2046 vpvfsentry = vn_vfslocks_getlock(vp); 2047 retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_READER); 2048 2049 if (retval == EINTR) { 2050 vn_vfslocks_rele(vpvfsentry); 2051 return (EINTR); 2052 } 2053 2054 return (retval); 2055 } 2056 2057 2058 /* 2059 * vn_vfswlock is used to implement a lock which is logically a writers lock 2060 * protecting the v_vfsmountedhere field. 2061 */ 2062 int 2063 vn_vfswlock(vnode_t *vp) 2064 { 2065 vn_vfslocks_entry_t *vpvfsentry; 2066 2067 /* 2068 * If vp is NULL then somebody is trying to lock the covered vnode 2069 * of /. (vfs_vnodecovered is NULL for /). This situation will 2070 * only happen when unmounting /. Since that operation will fail 2071 * anyway, return EBUSY here instead of in VFS_UNMOUNT. 2072 */ 2073 if (vp == NULL) 2074 return (EBUSY); 2075 2076 vpvfsentry = vn_vfslocks_getlock(vp); 2077 2078 if (rwst_tryenter(&vpvfsentry->ve_lock, RW_WRITER)) 2079 return (0); 2080 2081 vn_vfslocks_rele(vpvfsentry); 2082 return (EBUSY); 2083 } 2084 2085 int 2086 vn_vfsrlock(vnode_t *vp) 2087 { 2088 vn_vfslocks_entry_t *vpvfsentry; 2089 2090 /* 2091 * If vp is NULL then somebody is trying to lock the covered vnode 2092 * of /. (vfs_vnodecovered is NULL for /). This situation will 2093 * only happen when unmounting /. Since that operation will fail 2094 * anyway, return EBUSY here instead of in VFS_UNMOUNT. 2095 */ 2096 if (vp == NULL) 2097 return (EBUSY); 2098 2099 vpvfsentry = vn_vfslocks_getlock(vp); 2100 2101 if (rwst_tryenter(&vpvfsentry->ve_lock, RW_READER)) 2102 return (0); 2103 2104 vn_vfslocks_rele(vpvfsentry); 2105 return (EBUSY); 2106 } 2107 2108 void 2109 vn_vfsunlock(vnode_t *vp) 2110 { 2111 vn_vfslocks_entry_t *vpvfsentry; 2112 2113 /* 2114 * ve_refcnt needs to be decremented twice. 2115 * 1. To release refernce after a call to vn_vfslocks_getlock() 2116 * 2. To release the reference from the locking routines like 2117 * vn_vfsrlock/vn_vfswlock etc,. 2118 */ 2119 vpvfsentry = vn_vfslocks_getlock(vp); 2120 vn_vfslocks_rele(vpvfsentry); 2121 2122 rwst_exit(&vpvfsentry->ve_lock); 2123 vn_vfslocks_rele(vpvfsentry); 2124 } 2125 2126 int 2127 vn_vfswlock_held(vnode_t *vp) 2128 { 2129 int held; 2130 vn_vfslocks_entry_t *vpvfsentry; 2131 2132 ASSERT(vp != NULL); 2133 2134 vpvfsentry = vn_vfslocks_getlock(vp); 2135 held = rwst_lock_held(&vpvfsentry->ve_lock, RW_WRITER); 2136 2137 vn_vfslocks_rele(vpvfsentry); 2138 return (held); 2139 } 2140 2141 2142 int 2143 vn_make_ops( 2144 const char *name, /* Name of file system */ 2145 const fs_operation_def_t *templ, /* Operation specification */ 2146 vnodeops_t **actual) /* Return the vnodeops */ 2147 { 2148 int unused_ops; 2149 int error; 2150 2151 *actual = (vnodeops_t *)kmem_alloc(sizeof (vnodeops_t), KM_SLEEP); 2152 2153 (*actual)->vnop_name = name; 2154 2155 error = fs_build_vector(*actual, &unused_ops, vn_ops_table, templ); 2156 if (error) { 2157 kmem_free(*actual, sizeof (vnodeops_t)); 2158 } 2159 2160 #if DEBUG 2161 if (unused_ops != 0) 2162 cmn_err(CE_WARN, "vn_make_ops: %s: %d operations supplied " 2163 "but not used", name, unused_ops); 2164 #endif 2165 2166 return (error); 2167 } 2168 2169 /* 2170 * Free the vnodeops created as a result of vn_make_ops() 2171 */ 2172 void 2173 vn_freevnodeops(vnodeops_t *vnops) 2174 { 2175 kmem_free(vnops, sizeof (vnodeops_t)); 2176 } 2177 2178 /* 2179 * Vnode cache. 2180 */ 2181 2182 /* ARGSUSED */ 2183 static int 2184 vn_cache_constructor(void *buf, void *cdrarg, int kmflags) 2185 { 2186 struct vnode *vp; 2187 2188 vp = buf; 2189 2190 mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL); 2191 cv_init(&vp->v_cv, NULL, CV_DEFAULT, NULL); 2192 rw_init(&vp->v_nbllock, NULL, RW_DEFAULT, NULL); 2193 rw_init(&vp->v_mslock, NULL, RW_DEFAULT, NULL); 2194 vp->v_femhead = NULL; /* Must be done before vn_reinit() */ 2195 vp->v_path = NULL; 2196 vp->v_mpssdata = NULL; 2197 vp->v_vsd = NULL; 2198 vp->v_fopdata = NULL; 2199 2200 return (0); 2201 } 2202 2203 /* ARGSUSED */ 2204 static void 2205 vn_cache_destructor(void *buf, void *cdrarg) 2206 { 2207 struct vnode *vp; 2208 2209 vp = buf; 2210 2211 rw_destroy(&vp->v_mslock); 2212 rw_destroy(&vp->v_nbllock); 2213 cv_destroy(&vp->v_cv); 2214 mutex_destroy(&vp->v_lock); 2215 } 2216 2217 void 2218 vn_create_cache(void) 2219 { 2220 vn_cache = kmem_cache_create("vn_cache", sizeof (struct vnode), 64, 2221 vn_cache_constructor, vn_cache_destructor, NULL, NULL, 2222 NULL, 0); 2223 } 2224 2225 void 2226 vn_destroy_cache(void) 2227 { 2228 kmem_cache_destroy(vn_cache); 2229 } 2230 2231 /* 2232 * Used by file systems when fs-specific nodes (e.g., ufs inodes) are 2233 * cached by the file system and vnodes remain associated. 2234 */ 2235 void 2236 vn_recycle(vnode_t *vp) 2237 { 2238 ASSERT(vp->v_pages == NULL); 2239 2240 /* 2241 * XXX - This really belongs in vn_reinit(), but we have some issues 2242 * with the counts. Best to have it here for clean initialization. 2243 */ 2244 vp->v_rdcnt = 0; 2245 vp->v_wrcnt = 0; 2246 vp->v_mmap_read = 0; 2247 vp->v_mmap_write = 0; 2248 2249 /* 2250 * If FEM was in use, make sure everything gets cleaned up 2251 * NOTE: vp->v_femhead is initialized to NULL in the vnode 2252 * constructor. 2253 */ 2254 if (vp->v_femhead) { 2255 /* XXX - There should be a free_femhead() that does all this */ 2256 ASSERT(vp->v_femhead->femh_list == NULL); 2257 mutex_destroy(&vp->v_femhead->femh_lock); 2258 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead))); 2259 vp->v_femhead = NULL; 2260 } 2261 if (vp->v_path) { 2262 kmem_free(vp->v_path, strlen(vp->v_path) + 1); 2263 vp->v_path = NULL; 2264 } 2265 2266 if (vp->v_fopdata != NULL) { 2267 free_fopdata(vp); 2268 } 2269 vp->v_mpssdata = NULL; 2270 vsd_free(vp); 2271 } 2272 2273 /* 2274 * Used to reset the vnode fields including those that are directly accessible 2275 * as well as those which require an accessor function. 2276 * 2277 * Does not initialize: 2278 * synchronization objects: v_lock, v_nbllock, v_cv 2279 * v_data (since FS-nodes and vnodes point to each other and should 2280 * be updated simultaneously) 2281 * v_op (in case someone needs to make a VOP call on this object) 2282 */ 2283 void 2284 vn_reinit(vnode_t *vp) 2285 { 2286 vp->v_count = 1; 2287 vp->v_vfsp = NULL; 2288 vp->v_stream = NULL; 2289 vp->v_vfsmountedhere = NULL; 2290 vp->v_flag = 0; 2291 vp->v_type = VNON; 2292 vp->v_rdev = NODEV; 2293 2294 vp->v_filocks = NULL; 2295 vp->v_shrlocks = NULL; 2296 vp->v_pages = NULL; 2297 vp->v_npages = 0; 2298 vp->v_msnpages = 0; 2299 vp->v_scanfront = NULL; 2300 vp->v_scanback = NULL; 2301 2302 vp->v_locality = NULL; 2303 vp->v_scantime = 0; 2304 vp->v_mset = 0; 2305 vp->v_msflags = 0; 2306 vp->v_msnext = NULL; 2307 vp->v_msprev = NULL; 2308 vp->v_xattrdir = NULL; 2309 2310 /* Handles v_femhead, v_path, and the r/w/map counts */ 2311 vn_recycle(vp); 2312 } 2313 2314 vnode_t * 2315 vn_alloc(int kmflag) 2316 { 2317 vnode_t *vp; 2318 2319 vp = kmem_cache_alloc(vn_cache, kmflag); 2320 2321 if (vp != NULL) { 2322 vp->v_femhead = NULL; /* Must be done before vn_reinit() */ 2323 vp->v_fopdata = NULL; 2324 vn_reinit(vp); 2325 } 2326 2327 return (vp); 2328 } 2329 2330 void 2331 vn_free(vnode_t *vp) 2332 { 2333 ASSERT(vp->v_shrlocks == NULL); 2334 ASSERT(vp->v_filocks == NULL); 2335 2336 /* 2337 * Some file systems call vn_free() with v_count of zero, 2338 * some with v_count of 1. In any case, the value should 2339 * never be anything else. 2340 */ 2341 ASSERT((vp->v_count == 0) || (vp->v_count == 1)); 2342 if (vp->v_path != NULL) { 2343 kmem_free(vp->v_path, strlen(vp->v_path) + 1); 2344 vp->v_path = NULL; 2345 } 2346 2347 /* If FEM was in use, make sure everything gets cleaned up */ 2348 if (vp->v_femhead) { 2349 /* XXX - There should be a free_femhead() that does all this */ 2350 ASSERT(vp->v_femhead->femh_list == NULL); 2351 mutex_destroy(&vp->v_femhead->femh_lock); 2352 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead))); 2353 vp->v_femhead = NULL; 2354 } 2355 2356 if (vp->v_fopdata != NULL) { 2357 free_fopdata(vp); 2358 } 2359 vp->v_mpssdata = NULL; 2360 vsd_free(vp); 2361 kmem_cache_free(vn_cache, vp); 2362 } 2363 2364 /* 2365 * vnode status changes, should define better states than 1, 0. 2366 */ 2367 void 2368 vn_reclaim(vnode_t *vp) 2369 { 2370 vfs_t *vfsp = vp->v_vfsp; 2371 2372 if (vfsp == NULL || 2373 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) { 2374 return; 2375 } 2376 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_RECLAIMED); 2377 } 2378 2379 void 2380 vn_idle(vnode_t *vp) 2381 { 2382 vfs_t *vfsp = vp->v_vfsp; 2383 2384 if (vfsp == NULL || 2385 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) { 2386 return; 2387 } 2388 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_IDLED); 2389 } 2390 void 2391 vn_exists(vnode_t *vp) 2392 { 2393 vfs_t *vfsp = vp->v_vfsp; 2394 2395 if (vfsp == NULL || 2396 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) { 2397 return; 2398 } 2399 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_EXISTS); 2400 } 2401 2402 void 2403 vn_invalid(vnode_t *vp) 2404 { 2405 vfs_t *vfsp = vp->v_vfsp; 2406 2407 if (vfsp == NULL || 2408 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) { 2409 return; 2410 } 2411 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_DESTROYED); 2412 } 2413 2414 /* Vnode event notification */ 2415 2416 int 2417 vnevent_support(vnode_t *vp, caller_context_t *ct) 2418 { 2419 if (vp == NULL) 2420 return (EINVAL); 2421 2422 return (VOP_VNEVENT(vp, VE_SUPPORT, NULL, NULL, ct)); 2423 } 2424 2425 void 2426 vnevent_rename_src(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct) 2427 { 2428 if (vp == NULL || vp->v_femhead == NULL) { 2429 return; 2430 } 2431 (void) VOP_VNEVENT(vp, VE_RENAME_SRC, dvp, name, ct); 2432 } 2433 2434 void 2435 vnevent_rename_dest(vnode_t *vp, vnode_t *dvp, char *name, 2436 caller_context_t *ct) 2437 { 2438 if (vp == NULL || vp->v_femhead == NULL) { 2439 return; 2440 } 2441 (void) VOP_VNEVENT(vp, VE_RENAME_DEST, dvp, name, ct); 2442 } 2443 2444 void 2445 vnevent_rename_dest_dir(vnode_t *vp, caller_context_t *ct) 2446 { 2447 if (vp == NULL || vp->v_femhead == NULL) { 2448 return; 2449 } 2450 (void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, NULL, NULL, ct); 2451 } 2452 2453 void 2454 vnevent_remove(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct) 2455 { 2456 if (vp == NULL || vp->v_femhead == NULL) { 2457 return; 2458 } 2459 (void) VOP_VNEVENT(vp, VE_REMOVE, dvp, name, ct); 2460 } 2461 2462 void 2463 vnevent_rmdir(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct) 2464 { 2465 if (vp == NULL || vp->v_femhead == NULL) { 2466 return; 2467 } 2468 (void) VOP_VNEVENT(vp, VE_RMDIR, dvp, name, ct); 2469 } 2470 2471 void 2472 vnevent_create(vnode_t *vp, caller_context_t *ct) 2473 { 2474 if (vp == NULL || vp->v_femhead == NULL) { 2475 return; 2476 } 2477 (void) VOP_VNEVENT(vp, VE_CREATE, NULL, NULL, ct); 2478 } 2479 2480 void 2481 vnevent_link(vnode_t *vp, caller_context_t *ct) 2482 { 2483 if (vp == NULL || vp->v_femhead == NULL) { 2484 return; 2485 } 2486 (void) VOP_VNEVENT(vp, VE_LINK, NULL, NULL, ct); 2487 } 2488 2489 void 2490 vnevent_mountedover(vnode_t *vp, caller_context_t *ct) 2491 { 2492 if (vp == NULL || vp->v_femhead == NULL) { 2493 return; 2494 } 2495 (void) VOP_VNEVENT(vp, VE_MOUNTEDOVER, NULL, NULL, ct); 2496 } 2497 2498 /* 2499 * Vnode accessors. 2500 */ 2501 2502 int 2503 vn_is_readonly(vnode_t *vp) 2504 { 2505 return (vp->v_vfsp->vfs_flag & VFS_RDONLY); 2506 } 2507 2508 int 2509 vn_has_flocks(vnode_t *vp) 2510 { 2511 return (vp->v_filocks != NULL); 2512 } 2513 2514 int 2515 vn_has_mandatory_locks(vnode_t *vp, int mode) 2516 { 2517 return ((vp->v_filocks != NULL) && (MANDLOCK(vp, mode))); 2518 } 2519 2520 int 2521 vn_has_cached_data(vnode_t *vp) 2522 { 2523 return (vp->v_pages != NULL); 2524 } 2525 2526 /* 2527 * Return 0 if the vnode in question shouldn't be permitted into a zone via 2528 * zone_enter(2). 2529 */ 2530 int 2531 vn_can_change_zones(vnode_t *vp) 2532 { 2533 struct vfssw *vswp; 2534 int allow = 1; 2535 vnode_t *rvp; 2536 2537 if (nfs_global_client_only != 0) 2538 return (1); 2539 2540 /* 2541 * We always want to look at the underlying vnode if there is one. 2542 */ 2543 if (VOP_REALVP(vp, &rvp, NULL) != 0) 2544 rvp = vp; 2545 /* 2546 * Some pseudo filesystems (including doorfs) don't actually register 2547 * their vfsops_t, so the following may return NULL; we happily let 2548 * such vnodes switch zones. 2549 */ 2550 vswp = vfs_getvfsswbyvfsops(vfs_getops(rvp->v_vfsp)); 2551 if (vswp != NULL) { 2552 if (vswp->vsw_flag & VSW_NOTZONESAFE) 2553 allow = 0; 2554 vfs_unrefvfssw(vswp); 2555 } 2556 return (allow); 2557 } 2558 2559 /* 2560 * Return nonzero if the vnode is a mount point, zero if not. 2561 */ 2562 int 2563 vn_ismntpt(vnode_t *vp) 2564 { 2565 return (vp->v_vfsmountedhere != NULL); 2566 } 2567 2568 /* Retrieve the vfs (if any) mounted on this vnode */ 2569 vfs_t * 2570 vn_mountedvfs(vnode_t *vp) 2571 { 2572 return (vp->v_vfsmountedhere); 2573 } 2574 2575 /* 2576 * vn_has_other_opens() checks whether a particular file is opened by more than 2577 * just the caller and whether the open is for read and/or write. 2578 * This routine is for calling after the caller has already called VOP_OPEN() 2579 * and the caller wishes to know if they are the only one with it open for 2580 * the mode(s) specified. 2581 * 2582 * Vnode counts are only kept on regular files (v_type=VREG). 2583 */ 2584 int 2585 vn_has_other_opens( 2586 vnode_t *vp, 2587 v_mode_t mode) 2588 { 2589 2590 ASSERT(vp != NULL); 2591 2592 switch (mode) { 2593 case V_WRITE: 2594 if (vp->v_wrcnt > 1) 2595 return (V_TRUE); 2596 break; 2597 case V_RDORWR: 2598 if ((vp->v_rdcnt > 1) || (vp->v_wrcnt > 1)) 2599 return (V_TRUE); 2600 break; 2601 case V_RDANDWR: 2602 if ((vp->v_rdcnt > 1) && (vp->v_wrcnt > 1)) 2603 return (V_TRUE); 2604 break; 2605 case V_READ: 2606 if (vp->v_rdcnt > 1) 2607 return (V_TRUE); 2608 break; 2609 } 2610 2611 return (V_FALSE); 2612 } 2613 2614 /* 2615 * vn_is_opened() checks whether a particular file is opened and 2616 * whether the open is for read and/or write. 2617 * 2618 * Vnode counts are only kept on regular files (v_type=VREG). 2619 */ 2620 int 2621 vn_is_opened( 2622 vnode_t *vp, 2623 v_mode_t mode) 2624 { 2625 2626 ASSERT(vp != NULL); 2627 2628 switch (mode) { 2629 case V_WRITE: 2630 if (vp->v_wrcnt) 2631 return (V_TRUE); 2632 break; 2633 case V_RDANDWR: 2634 if (vp->v_rdcnt && vp->v_wrcnt) 2635 return (V_TRUE); 2636 break; 2637 case V_RDORWR: 2638 if (vp->v_rdcnt || vp->v_wrcnt) 2639 return (V_TRUE); 2640 break; 2641 case V_READ: 2642 if (vp->v_rdcnt) 2643 return (V_TRUE); 2644 break; 2645 } 2646 2647 return (V_FALSE); 2648 } 2649 2650 /* 2651 * vn_is_mapped() checks whether a particular file is mapped and whether 2652 * the file is mapped read and/or write. 2653 */ 2654 int 2655 vn_is_mapped( 2656 vnode_t *vp, 2657 v_mode_t mode) 2658 { 2659 2660 ASSERT(vp != NULL); 2661 2662 #if !defined(_LP64) 2663 switch (mode) { 2664 /* 2665 * The atomic_add_64_nv functions force atomicity in the 2666 * case of 32 bit architectures. Otherwise the 64 bit values 2667 * require two fetches. The value of the fields may be 2668 * (potentially) changed between the first fetch and the 2669 * second 2670 */ 2671 case V_WRITE: 2672 if (atomic_add_64_nv((&(vp->v_mmap_write)), 0)) 2673 return (V_TRUE); 2674 break; 2675 case V_RDANDWR: 2676 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) && 2677 (atomic_add_64_nv((&(vp->v_mmap_write)), 0))) 2678 return (V_TRUE); 2679 break; 2680 case V_RDORWR: 2681 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) || 2682 (atomic_add_64_nv((&(vp->v_mmap_write)), 0))) 2683 return (V_TRUE); 2684 break; 2685 case V_READ: 2686 if (atomic_add_64_nv((&(vp->v_mmap_read)), 0)) 2687 return (V_TRUE); 2688 break; 2689 } 2690 #else 2691 switch (mode) { 2692 case V_WRITE: 2693 if (vp->v_mmap_write) 2694 return (V_TRUE); 2695 break; 2696 case V_RDANDWR: 2697 if (vp->v_mmap_read && vp->v_mmap_write) 2698 return (V_TRUE); 2699 break; 2700 case V_RDORWR: 2701 if (vp->v_mmap_read || vp->v_mmap_write) 2702 return (V_TRUE); 2703 break; 2704 case V_READ: 2705 if (vp->v_mmap_read) 2706 return (V_TRUE); 2707 break; 2708 } 2709 #endif 2710 2711 return (V_FALSE); 2712 } 2713 2714 /* 2715 * Set the operations vector for a vnode. 2716 * 2717 * FEM ensures that the v_femhead pointer is filled in before the 2718 * v_op pointer is changed. This means that if the v_femhead pointer 2719 * is NULL, and the v_op field hasn't changed since before which checked 2720 * the v_femhead pointer; then our update is ok - we are not racing with 2721 * FEM. 2722 */ 2723 void 2724 vn_setops(vnode_t *vp, vnodeops_t *vnodeops) 2725 { 2726 vnodeops_t *op; 2727 2728 ASSERT(vp != NULL); 2729 ASSERT(vnodeops != NULL); 2730 2731 op = vp->v_op; 2732 membar_consumer(); 2733 /* 2734 * If vp->v_femhead == NULL, then we'll call casptr() to do the 2735 * compare-and-swap on vp->v_op. If either fails, then FEM is 2736 * in effect on the vnode and we need to have FEM deal with it. 2737 */ 2738 if (vp->v_femhead != NULL || casptr(&vp->v_op, op, vnodeops) != op) { 2739 fem_setvnops(vp, vnodeops); 2740 } 2741 } 2742 2743 /* 2744 * Retrieve the operations vector for a vnode 2745 * As with vn_setops(above); make sure we aren't racing with FEM. 2746 * FEM sets the v_op to a special, internal, vnodeops that wouldn't 2747 * make sense to the callers of this routine. 2748 */ 2749 vnodeops_t * 2750 vn_getops(vnode_t *vp) 2751 { 2752 vnodeops_t *op; 2753 2754 ASSERT(vp != NULL); 2755 2756 op = vp->v_op; 2757 membar_consumer(); 2758 if (vp->v_femhead == NULL && op == vp->v_op) { 2759 return (op); 2760 } else { 2761 return (fem_getvnops(vp)); 2762 } 2763 } 2764 2765 /* 2766 * Returns non-zero (1) if the vnodeops matches that of the vnode. 2767 * Returns zero (0) if not. 2768 */ 2769 int 2770 vn_matchops(vnode_t *vp, vnodeops_t *vnodeops) 2771 { 2772 return (vn_getops(vp) == vnodeops); 2773 } 2774 2775 /* 2776 * Returns non-zero (1) if the specified operation matches the 2777 * corresponding operation for that the vnode. 2778 * Returns zero (0) if not. 2779 */ 2780 2781 #define MATCHNAME(n1, n2) (((n1)[0] == (n2)[0]) && (strcmp((n1), (n2)) == 0)) 2782 2783 int 2784 vn_matchopval(vnode_t *vp, char *vopname, fs_generic_func_p funcp) 2785 { 2786 const fs_operation_trans_def_t *otdp; 2787 fs_generic_func_p *loc = NULL; 2788 vnodeops_t *vop = vn_getops(vp); 2789 2790 ASSERT(vopname != NULL); 2791 2792 for (otdp = vn_ops_table; otdp->name != NULL; otdp++) { 2793 if (MATCHNAME(otdp->name, vopname)) { 2794 loc = (fs_generic_func_p *) 2795 ((char *)(vop) + otdp->offset); 2796 break; 2797 } 2798 } 2799 2800 return ((loc != NULL) && (*loc == funcp)); 2801 } 2802 2803 /* 2804 * fs_new_caller_id() needs to return a unique ID on a given local system. 2805 * The IDs do not need to survive across reboots. These are primarily 2806 * used so that (FEM) monitors can detect particular callers (such as 2807 * the NFS server) to a given vnode/vfs operation. 2808 */ 2809 u_longlong_t 2810 fs_new_caller_id() 2811 { 2812 static uint64_t next_caller_id = 0LL; /* First call returns 1 */ 2813 2814 return ((u_longlong_t)atomic_add_64_nv(&next_caller_id, 1)); 2815 } 2816 2817 /* 2818 * Given a starting vnode and a path, updates the path in the target vnode in 2819 * a safe manner. If the vnode already has path information embedded, then the 2820 * cached path is left untouched. 2821 */ 2822 2823 size_t max_vnode_path = 4 * MAXPATHLEN; 2824 2825 void 2826 vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp, 2827 const char *path, size_t plen) 2828 { 2829 char *rpath; 2830 vnode_t *base; 2831 size_t rpathlen, rpathalloc; 2832 int doslash = 1; 2833 2834 if (*path == '/') { 2835 base = rootvp; 2836 path++; 2837 plen--; 2838 } else { 2839 base = startvp; 2840 } 2841 2842 /* 2843 * We cannot grab base->v_lock while we hold vp->v_lock because of 2844 * the potential for deadlock. 2845 */ 2846 mutex_enter(&base->v_lock); 2847 if (base->v_path == NULL) { 2848 mutex_exit(&base->v_lock); 2849 return; 2850 } 2851 2852 rpathlen = strlen(base->v_path); 2853 rpathalloc = rpathlen + plen + 1; 2854 /* Avoid adding a slash if there's already one there */ 2855 if (base->v_path[rpathlen-1] == '/') 2856 doslash = 0; 2857 else 2858 rpathalloc++; 2859 2860 /* 2861 * We don't want to call kmem_alloc(KM_SLEEP) with kernel locks held, 2862 * so we must do this dance. If, by chance, something changes the path, 2863 * just give up since there is no real harm. 2864 */ 2865 mutex_exit(&base->v_lock); 2866 2867 /* Paths should stay within reason */ 2868 if (rpathalloc > max_vnode_path) 2869 return; 2870 2871 rpath = kmem_alloc(rpathalloc, KM_SLEEP); 2872 2873 mutex_enter(&base->v_lock); 2874 if (base->v_path == NULL || strlen(base->v_path) != rpathlen) { 2875 mutex_exit(&base->v_lock); 2876 kmem_free(rpath, rpathalloc); 2877 return; 2878 } 2879 bcopy(base->v_path, rpath, rpathlen); 2880 mutex_exit(&base->v_lock); 2881 2882 if (doslash) 2883 rpath[rpathlen++] = '/'; 2884 bcopy(path, rpath + rpathlen, plen); 2885 rpath[rpathlen + plen] = '\0'; 2886 2887 mutex_enter(&vp->v_lock); 2888 if (vp->v_path != NULL) { 2889 mutex_exit(&vp->v_lock); 2890 kmem_free(rpath, rpathalloc); 2891 } else { 2892 vp->v_path = rpath; 2893 mutex_exit(&vp->v_lock); 2894 } 2895 } 2896 2897 /* 2898 * Sets the path to the vnode to be the given string, regardless of current 2899 * context. The string must be a complete path from rootdir. This is only used 2900 * by fsop_root() for setting the path based on the mountpoint. 2901 */ 2902 void 2903 vn_setpath_str(struct vnode *vp, const char *str, size_t len) 2904 { 2905 char *buf = kmem_alloc(len + 1, KM_SLEEP); 2906 2907 mutex_enter(&vp->v_lock); 2908 if (vp->v_path != NULL) { 2909 mutex_exit(&vp->v_lock); 2910 kmem_free(buf, len + 1); 2911 return; 2912 } 2913 2914 vp->v_path = buf; 2915 bcopy(str, vp->v_path, len); 2916 vp->v_path[len] = '\0'; 2917 2918 mutex_exit(&vp->v_lock); 2919 } 2920 2921 /* 2922 * Similar to vn_setpath_str(), this function sets the path of the destination 2923 * vnode to the be the same as the source vnode. 2924 */ 2925 void 2926 vn_copypath(struct vnode *src, struct vnode *dst) 2927 { 2928 char *buf; 2929 int alloc; 2930 2931 mutex_enter(&src->v_lock); 2932 if (src->v_path == NULL) { 2933 mutex_exit(&src->v_lock); 2934 return; 2935 } 2936 alloc = strlen(src->v_path) + 1; 2937 2938 /* avoid kmem_alloc() with lock held */ 2939 mutex_exit(&src->v_lock); 2940 buf = kmem_alloc(alloc, KM_SLEEP); 2941 mutex_enter(&src->v_lock); 2942 if (src->v_path == NULL || strlen(src->v_path) + 1 != alloc) { 2943 mutex_exit(&src->v_lock); 2944 kmem_free(buf, alloc); 2945 return; 2946 } 2947 bcopy(src->v_path, buf, alloc); 2948 mutex_exit(&src->v_lock); 2949 2950 mutex_enter(&dst->v_lock); 2951 if (dst->v_path != NULL) { 2952 mutex_exit(&dst->v_lock); 2953 kmem_free(buf, alloc); 2954 return; 2955 } 2956 dst->v_path = buf; 2957 mutex_exit(&dst->v_lock); 2958 } 2959 2960 /* 2961 * XXX Private interface for segvn routines that handle vnode 2962 * large page segments. 2963 * 2964 * return 1 if vp's file system VOP_PAGEIO() implementation 2965 * can be safely used instead of VOP_GETPAGE() for handling 2966 * pagefaults against regular non swap files. VOP_PAGEIO() 2967 * interface is considered safe here if its implementation 2968 * is very close to VOP_GETPAGE() implementation. 2969 * e.g. It zero's out the part of the page beyond EOF. Doesn't 2970 * panic if there're file holes but instead returns an error. 2971 * Doesn't assume file won't be changed by user writes, etc. 2972 * 2973 * return 0 otherwise. 2974 * 2975 * For now allow segvn to only use VOP_PAGEIO() with ufs and nfs. 2976 */ 2977 int 2978 vn_vmpss_usepageio(vnode_t *vp) 2979 { 2980 vfs_t *vfsp = vp->v_vfsp; 2981 char *fsname = vfssw[vfsp->vfs_fstype].vsw_name; 2982 char *pageio_ok_fss[] = {"ufs", "nfs", NULL}; 2983 char **fsok = pageio_ok_fss; 2984 2985 if (fsname == NULL) { 2986 return (0); 2987 } 2988 2989 for (; *fsok; fsok++) { 2990 if (strcmp(*fsok, fsname) == 0) { 2991 return (1); 2992 } 2993 } 2994 return (0); 2995 } 2996 2997 /* VOP_XXX() macros call the corresponding fop_xxx() function */ 2998 2999 int 3000 fop_open( 3001 vnode_t **vpp, 3002 int mode, 3003 cred_t *cr, 3004 caller_context_t *ct) 3005 { 3006 int ret; 3007 vnode_t *vp = *vpp; 3008 3009 VN_HOLD(vp); 3010 /* 3011 * Adding to the vnode counts before calling open 3012 * avoids the need for a mutex. It circumvents a race 3013 * condition where a query made on the vnode counts results in a 3014 * false negative. The inquirer goes away believing the file is 3015 * not open when there is an open on the file already under way. 3016 * 3017 * The counts are meant to prevent NFS from granting a delegation 3018 * when it would be dangerous to do so. 3019 * 3020 * The vnode counts are only kept on regular files 3021 */ 3022 if ((*vpp)->v_type == VREG) { 3023 if (mode & FREAD) 3024 atomic_add_32(&((*vpp)->v_rdcnt), 1); 3025 if (mode & FWRITE) 3026 atomic_add_32(&((*vpp)->v_wrcnt), 1); 3027 } 3028 3029 VOPXID_MAP_CR(vp, cr); 3030 3031 ret = (*(*(vpp))->v_op->vop_open)(vpp, mode, cr, ct); 3032 3033 if (ret) { 3034 /* 3035 * Use the saved vp just in case the vnode ptr got trashed 3036 * by the error. 3037 */ 3038 VOPSTATS_UPDATE(vp, open); 3039 if ((vp->v_type == VREG) && (mode & FREAD)) 3040 atomic_add_32(&(vp->v_rdcnt), -1); 3041 if ((vp->v_type == VREG) && (mode & FWRITE)) 3042 atomic_add_32(&(vp->v_wrcnt), -1); 3043 } else { 3044 /* 3045 * Some filesystems will return a different vnode, 3046 * but the same path was still used to open it. 3047 * So if we do change the vnode and need to 3048 * copy over the path, do so here, rather than special 3049 * casing each filesystem. Adjust the vnode counts to 3050 * reflect the vnode switch. 3051 */ 3052 VOPSTATS_UPDATE(*vpp, open); 3053 if (*vpp != vp && *vpp != NULL) { 3054 vn_copypath(vp, *vpp); 3055 if (((*vpp)->v_type == VREG) && (mode & FREAD)) 3056 atomic_add_32(&((*vpp)->v_rdcnt), 1); 3057 if ((vp->v_type == VREG) && (mode & FREAD)) 3058 atomic_add_32(&(vp->v_rdcnt), -1); 3059 if (((*vpp)->v_type == VREG) && (mode & FWRITE)) 3060 atomic_add_32(&((*vpp)->v_wrcnt), 1); 3061 if ((vp->v_type == VREG) && (mode & FWRITE)) 3062 atomic_add_32(&(vp->v_wrcnt), -1); 3063 } 3064 } 3065 VN_RELE(vp); 3066 return (ret); 3067 } 3068 3069 int 3070 fop_close( 3071 vnode_t *vp, 3072 int flag, 3073 int count, 3074 offset_t offset, 3075 cred_t *cr, 3076 caller_context_t *ct) 3077 { 3078 int err; 3079 3080 VOPXID_MAP_CR(vp, cr); 3081 3082 err = (*(vp)->v_op->vop_close)(vp, flag, count, offset, cr, ct); 3083 VOPSTATS_UPDATE(vp, close); 3084 /* 3085 * Check passed in count to handle possible dups. Vnode counts are only 3086 * kept on regular files 3087 */ 3088 if ((vp->v_type == VREG) && (count == 1)) { 3089 if (flag & FREAD) { 3090 ASSERT(vp->v_rdcnt > 0); 3091 atomic_add_32(&(vp->v_rdcnt), -1); 3092 } 3093 if (flag & FWRITE) { 3094 ASSERT(vp->v_wrcnt > 0); 3095 atomic_add_32(&(vp->v_wrcnt), -1); 3096 } 3097 } 3098 return (err); 3099 } 3100 3101 int 3102 fop_read( 3103 vnode_t *vp, 3104 uio_t *uiop, 3105 int ioflag, 3106 cred_t *cr, 3107 caller_context_t *ct) 3108 { 3109 int err; 3110 ssize_t resid_start = uiop->uio_resid; 3111 3112 VOPXID_MAP_CR(vp, cr); 3113 3114 err = (*(vp)->v_op->vop_read)(vp, uiop, ioflag, cr, ct); 3115 VOPSTATS_UPDATE_IO(vp, read, 3116 read_bytes, (resid_start - uiop->uio_resid)); 3117 return (err); 3118 } 3119 3120 int 3121 fop_write( 3122 vnode_t *vp, 3123 uio_t *uiop, 3124 int ioflag, 3125 cred_t *cr, 3126 caller_context_t *ct) 3127 { 3128 int err; 3129 ssize_t resid_start = uiop->uio_resid; 3130 3131 VOPXID_MAP_CR(vp, cr); 3132 3133 err = (*(vp)->v_op->vop_write)(vp, uiop, ioflag, cr, ct); 3134 VOPSTATS_UPDATE_IO(vp, write, 3135 write_bytes, (resid_start - uiop->uio_resid)); 3136 return (err); 3137 } 3138 3139 int 3140 fop_ioctl( 3141 vnode_t *vp, 3142 int cmd, 3143 intptr_t arg, 3144 int flag, 3145 cred_t *cr, 3146 int *rvalp, 3147 caller_context_t *ct) 3148 { 3149 int err; 3150 3151 VOPXID_MAP_CR(vp, cr); 3152 3153 err = (*(vp)->v_op->vop_ioctl)(vp, cmd, arg, flag, cr, rvalp, ct); 3154 VOPSTATS_UPDATE(vp, ioctl); 3155 return (err); 3156 } 3157 3158 int 3159 fop_setfl( 3160 vnode_t *vp, 3161 int oflags, 3162 int nflags, 3163 cred_t *cr, 3164 caller_context_t *ct) 3165 { 3166 int err; 3167 3168 VOPXID_MAP_CR(vp, cr); 3169 3170 err = (*(vp)->v_op->vop_setfl)(vp, oflags, nflags, cr, ct); 3171 VOPSTATS_UPDATE(vp, setfl); 3172 return (err); 3173 } 3174 3175 int 3176 fop_getattr( 3177 vnode_t *vp, 3178 vattr_t *vap, 3179 int flags, 3180 cred_t *cr, 3181 caller_context_t *ct) 3182 { 3183 int err; 3184 3185 VOPXID_MAP_CR(vp, cr); 3186 3187 /* 3188 * If this file system doesn't understand the xvattr extensions 3189 * then turn off the xvattr bit. 3190 */ 3191 if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) { 3192 vap->va_mask &= ~AT_XVATTR; 3193 } 3194 3195 /* 3196 * We're only allowed to skip the ACL check iff we used a 32 bit 3197 * ACE mask with VOP_ACCESS() to determine permissions. 3198 */ 3199 if ((flags & ATTR_NOACLCHECK) && 3200 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 3201 return (EINVAL); 3202 } 3203 err = (*(vp)->v_op->vop_getattr)(vp, vap, flags, cr, ct); 3204 VOPSTATS_UPDATE(vp, getattr); 3205 return (err); 3206 } 3207 3208 int 3209 fop_setattr( 3210 vnode_t *vp, 3211 vattr_t *vap, 3212 int flags, 3213 cred_t *cr, 3214 caller_context_t *ct) 3215 { 3216 int err; 3217 3218 VOPXID_MAP_CR(vp, cr); 3219 3220 /* 3221 * If this file system doesn't understand the xvattr extensions 3222 * then turn off the xvattr bit. 3223 */ 3224 if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) { 3225 vap->va_mask &= ~AT_XVATTR; 3226 } 3227 3228 /* 3229 * We're only allowed to skip the ACL check iff we used a 32 bit 3230 * ACE mask with VOP_ACCESS() to determine permissions. 3231 */ 3232 if ((flags & ATTR_NOACLCHECK) && 3233 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 3234 return (EINVAL); 3235 } 3236 err = (*(vp)->v_op->vop_setattr)(vp, vap, flags, cr, ct); 3237 VOPSTATS_UPDATE(vp, setattr); 3238 return (err); 3239 } 3240 3241 int 3242 fop_access( 3243 vnode_t *vp, 3244 int mode, 3245 int flags, 3246 cred_t *cr, 3247 caller_context_t *ct) 3248 { 3249 int err; 3250 3251 if ((flags & V_ACE_MASK) && 3252 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 3253 return (EINVAL); 3254 } 3255 3256 VOPXID_MAP_CR(vp, cr); 3257 3258 err = (*(vp)->v_op->vop_access)(vp, mode, flags, cr, ct); 3259 VOPSTATS_UPDATE(vp, access); 3260 return (err); 3261 } 3262 3263 int 3264 fop_lookup( 3265 vnode_t *dvp, 3266 char *nm, 3267 vnode_t **vpp, 3268 pathname_t *pnp, 3269 int flags, 3270 vnode_t *rdir, 3271 cred_t *cr, 3272 caller_context_t *ct, 3273 int *deflags, /* Returned per-dirent flags */ 3274 pathname_t *ppnp) /* Returned case-preserved name in directory */ 3275 { 3276 int ret; 3277 3278 /* 3279 * If this file system doesn't support case-insensitive access 3280 * and said access is requested, fail quickly. It is required 3281 * that if the vfs supports case-insensitive lookup, it also 3282 * supports extended dirent flags. 3283 */ 3284 if (flags & FIGNORECASE && 3285 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3286 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3287 return (EINVAL); 3288 3289 VOPXID_MAP_CR(dvp, cr); 3290 3291 if ((flags & LOOKUP_XATTR) && (flags & LOOKUP_HAVE_SYSATTR_DIR) == 0) { 3292 ret = xattr_dir_lookup(dvp, vpp, flags, cr); 3293 } else { 3294 ret = (*(dvp)->v_op->vop_lookup) 3295 (dvp, nm, vpp, pnp, flags, rdir, cr, ct, deflags, ppnp); 3296 } 3297 if (ret == 0 && *vpp) { 3298 VOPSTATS_UPDATE(*vpp, lookup); 3299 if ((*vpp)->v_path == NULL) { 3300 vn_setpath(rootdir, dvp, *vpp, nm, strlen(nm)); 3301 } 3302 } 3303 3304 return (ret); 3305 } 3306 3307 int 3308 fop_create( 3309 vnode_t *dvp, 3310 char *name, 3311 vattr_t *vap, 3312 vcexcl_t excl, 3313 int mode, 3314 vnode_t **vpp, 3315 cred_t *cr, 3316 int flags, 3317 caller_context_t *ct, 3318 vsecattr_t *vsecp) /* ACL to set during create */ 3319 { 3320 int ret; 3321 3322 if (vsecp != NULL && 3323 vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) { 3324 return (EINVAL); 3325 } 3326 /* 3327 * If this file system doesn't support case-insensitive access 3328 * and said access is requested, fail quickly. 3329 */ 3330 if (flags & FIGNORECASE && 3331 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3332 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3333 return (EINVAL); 3334 3335 VOPXID_MAP_CR(dvp, cr); 3336 3337 ret = (*(dvp)->v_op->vop_create) 3338 (dvp, name, vap, excl, mode, vpp, cr, flags, ct, vsecp); 3339 if (ret == 0 && *vpp) { 3340 VOPSTATS_UPDATE(*vpp, create); 3341 if ((*vpp)->v_path == NULL) { 3342 vn_setpath(rootdir, dvp, *vpp, name, strlen(name)); 3343 } 3344 } 3345 3346 return (ret); 3347 } 3348 3349 int 3350 fop_remove( 3351 vnode_t *dvp, 3352 char *nm, 3353 cred_t *cr, 3354 caller_context_t *ct, 3355 int flags) 3356 { 3357 int err; 3358 3359 /* 3360 * If this file system doesn't support case-insensitive access 3361 * and said access is requested, fail quickly. 3362 */ 3363 if (flags & FIGNORECASE && 3364 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3365 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3366 return (EINVAL); 3367 3368 VOPXID_MAP_CR(dvp, cr); 3369 3370 err = (*(dvp)->v_op->vop_remove)(dvp, nm, cr, ct, flags); 3371 VOPSTATS_UPDATE(dvp, remove); 3372 return (err); 3373 } 3374 3375 int 3376 fop_link( 3377 vnode_t *tdvp, 3378 vnode_t *svp, 3379 char *tnm, 3380 cred_t *cr, 3381 caller_context_t *ct, 3382 int flags) 3383 { 3384 int err; 3385 3386 /* 3387 * If the target file system doesn't support case-insensitive access 3388 * and said access is requested, fail quickly. 3389 */ 3390 if (flags & FIGNORECASE && 3391 (vfs_has_feature(tdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3392 vfs_has_feature(tdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3393 return (EINVAL); 3394 3395 VOPXID_MAP_CR(tdvp, cr); 3396 3397 err = (*(tdvp)->v_op->vop_link)(tdvp, svp, tnm, cr, ct, flags); 3398 VOPSTATS_UPDATE(tdvp, link); 3399 return (err); 3400 } 3401 3402 int 3403 fop_rename( 3404 vnode_t *sdvp, 3405 char *snm, 3406 vnode_t *tdvp, 3407 char *tnm, 3408 cred_t *cr, 3409 caller_context_t *ct, 3410 int flags) 3411 { 3412 int err; 3413 3414 /* 3415 * If the file system involved does not support 3416 * case-insensitive access and said access is requested, fail 3417 * quickly. 3418 */ 3419 if (flags & FIGNORECASE && 3420 ((vfs_has_feature(sdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3421 vfs_has_feature(sdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))) 3422 return (EINVAL); 3423 3424 VOPXID_MAP_CR(tdvp, cr); 3425 3426 err = (*(sdvp)->v_op->vop_rename)(sdvp, snm, tdvp, tnm, cr, ct, flags); 3427 VOPSTATS_UPDATE(sdvp, rename); 3428 return (err); 3429 } 3430 3431 int 3432 fop_mkdir( 3433 vnode_t *dvp, 3434 char *dirname, 3435 vattr_t *vap, 3436 vnode_t **vpp, 3437 cred_t *cr, 3438 caller_context_t *ct, 3439 int flags, 3440 vsecattr_t *vsecp) /* ACL to set during create */ 3441 { 3442 int ret; 3443 3444 if (vsecp != NULL && 3445 vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) { 3446 return (EINVAL); 3447 } 3448 /* 3449 * If this file system doesn't support case-insensitive access 3450 * and said access is requested, fail quickly. 3451 */ 3452 if (flags & FIGNORECASE && 3453 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3454 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3455 return (EINVAL); 3456 3457 VOPXID_MAP_CR(dvp, cr); 3458 3459 ret = (*(dvp)->v_op->vop_mkdir) 3460 (dvp, dirname, vap, vpp, cr, ct, flags, vsecp); 3461 if (ret == 0 && *vpp) { 3462 VOPSTATS_UPDATE(*vpp, mkdir); 3463 if ((*vpp)->v_path == NULL) { 3464 vn_setpath(rootdir, dvp, *vpp, dirname, 3465 strlen(dirname)); 3466 } 3467 } 3468 3469 return (ret); 3470 } 3471 3472 int 3473 fop_rmdir( 3474 vnode_t *dvp, 3475 char *nm, 3476 vnode_t *cdir, 3477 cred_t *cr, 3478 caller_context_t *ct, 3479 int flags) 3480 { 3481 int err; 3482 3483 /* 3484 * If this file system doesn't support case-insensitive access 3485 * and said access is requested, fail quickly. 3486 */ 3487 if (flags & FIGNORECASE && 3488 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3489 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3490 return (EINVAL); 3491 3492 VOPXID_MAP_CR(dvp, cr); 3493 3494 err = (*(dvp)->v_op->vop_rmdir)(dvp, nm, cdir, cr, ct, flags); 3495 VOPSTATS_UPDATE(dvp, rmdir); 3496 return (err); 3497 } 3498 3499 int 3500 fop_readdir( 3501 vnode_t *vp, 3502 uio_t *uiop, 3503 cred_t *cr, 3504 int *eofp, 3505 caller_context_t *ct, 3506 int flags) 3507 { 3508 int err; 3509 ssize_t resid_start = uiop->uio_resid; 3510 3511 /* 3512 * If this file system doesn't support retrieving directory 3513 * entry flags and said access is requested, fail quickly. 3514 */ 3515 if (flags & V_RDDIR_ENTFLAGS && 3516 vfs_has_feature(vp->v_vfsp, VFSFT_DIRENTFLAGS) == 0) 3517 return (EINVAL); 3518 3519 VOPXID_MAP_CR(vp, cr); 3520 3521 err = (*(vp)->v_op->vop_readdir)(vp, uiop, cr, eofp, ct, flags); 3522 VOPSTATS_UPDATE_IO(vp, readdir, 3523 readdir_bytes, (resid_start - uiop->uio_resid)); 3524 return (err); 3525 } 3526 3527 int 3528 fop_symlink( 3529 vnode_t *dvp, 3530 char *linkname, 3531 vattr_t *vap, 3532 char *target, 3533 cred_t *cr, 3534 caller_context_t *ct, 3535 int flags) 3536 { 3537 int err; 3538 3539 /* 3540 * If this file system doesn't support case-insensitive access 3541 * and said access is requested, fail quickly. 3542 */ 3543 if (flags & FIGNORECASE && 3544 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3545 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3546 return (EINVAL); 3547 3548 VOPXID_MAP_CR(dvp, cr); 3549 3550 err = (*(dvp)->v_op->vop_symlink) 3551 (dvp, linkname, vap, target, cr, ct, flags); 3552 VOPSTATS_UPDATE(dvp, symlink); 3553 return (err); 3554 } 3555 3556 int 3557 fop_readlink( 3558 vnode_t *vp, 3559 uio_t *uiop, 3560 cred_t *cr, 3561 caller_context_t *ct) 3562 { 3563 int err; 3564 3565 VOPXID_MAP_CR(vp, cr); 3566 3567 err = (*(vp)->v_op->vop_readlink)(vp, uiop, cr, ct); 3568 VOPSTATS_UPDATE(vp, readlink); 3569 return (err); 3570 } 3571 3572 int 3573 fop_fsync( 3574 vnode_t *vp, 3575 int syncflag, 3576 cred_t *cr, 3577 caller_context_t *ct) 3578 { 3579 int err; 3580 3581 VOPXID_MAP_CR(vp, cr); 3582 3583 err = (*(vp)->v_op->vop_fsync)(vp, syncflag, cr, ct); 3584 VOPSTATS_UPDATE(vp, fsync); 3585 return (err); 3586 } 3587 3588 void 3589 fop_inactive( 3590 vnode_t *vp, 3591 cred_t *cr, 3592 caller_context_t *ct) 3593 { 3594 /* Need to update stats before vop call since we may lose the vnode */ 3595 VOPSTATS_UPDATE(vp, inactive); 3596 3597 VOPXID_MAP_CR(vp, cr); 3598 3599 (*(vp)->v_op->vop_inactive)(vp, cr, ct); 3600 } 3601 3602 int 3603 fop_fid( 3604 vnode_t *vp, 3605 fid_t *fidp, 3606 caller_context_t *ct) 3607 { 3608 int err; 3609 3610 err = (*(vp)->v_op->vop_fid)(vp, fidp, ct); 3611 VOPSTATS_UPDATE(vp, fid); 3612 return (err); 3613 } 3614 3615 int 3616 fop_rwlock( 3617 vnode_t *vp, 3618 int write_lock, 3619 caller_context_t *ct) 3620 { 3621 int ret; 3622 3623 ret = ((*(vp)->v_op->vop_rwlock)(vp, write_lock, ct)); 3624 VOPSTATS_UPDATE(vp, rwlock); 3625 return (ret); 3626 } 3627 3628 void 3629 fop_rwunlock( 3630 vnode_t *vp, 3631 int write_lock, 3632 caller_context_t *ct) 3633 { 3634 (*(vp)->v_op->vop_rwunlock)(vp, write_lock, ct); 3635 VOPSTATS_UPDATE(vp, rwunlock); 3636 } 3637 3638 int 3639 fop_seek( 3640 vnode_t *vp, 3641 offset_t ooff, 3642 offset_t *noffp, 3643 caller_context_t *ct) 3644 { 3645 int err; 3646 3647 err = (*(vp)->v_op->vop_seek)(vp, ooff, noffp, ct); 3648 VOPSTATS_UPDATE(vp, seek); 3649 return (err); 3650 } 3651 3652 int 3653 fop_cmp( 3654 vnode_t *vp1, 3655 vnode_t *vp2, 3656 caller_context_t *ct) 3657 { 3658 int err; 3659 3660 err = (*(vp1)->v_op->vop_cmp)(vp1, vp2, ct); 3661 VOPSTATS_UPDATE(vp1, cmp); 3662 return (err); 3663 } 3664 3665 int 3666 fop_frlock( 3667 vnode_t *vp, 3668 int cmd, 3669 flock64_t *bfp, 3670 int flag, 3671 offset_t offset, 3672 struct flk_callback *flk_cbp, 3673 cred_t *cr, 3674 caller_context_t *ct) 3675 { 3676 int err; 3677 3678 VOPXID_MAP_CR(vp, cr); 3679 3680 err = (*(vp)->v_op->vop_frlock) 3681 (vp, cmd, bfp, flag, offset, flk_cbp, cr, ct); 3682 VOPSTATS_UPDATE(vp, frlock); 3683 return (err); 3684 } 3685 3686 int 3687 fop_space( 3688 vnode_t *vp, 3689 int cmd, 3690 flock64_t *bfp, 3691 int flag, 3692 offset_t offset, 3693 cred_t *cr, 3694 caller_context_t *ct) 3695 { 3696 int err; 3697 3698 VOPXID_MAP_CR(vp, cr); 3699 3700 err = (*(vp)->v_op->vop_space)(vp, cmd, bfp, flag, offset, cr, ct); 3701 VOPSTATS_UPDATE(vp, space); 3702 return (err); 3703 } 3704 3705 int 3706 fop_realvp( 3707 vnode_t *vp, 3708 vnode_t **vpp, 3709 caller_context_t *ct) 3710 { 3711 int err; 3712 3713 err = (*(vp)->v_op->vop_realvp)(vp, vpp, ct); 3714 VOPSTATS_UPDATE(vp, realvp); 3715 return (err); 3716 } 3717 3718 int 3719 fop_getpage( 3720 vnode_t *vp, 3721 offset_t off, 3722 size_t len, 3723 uint_t *protp, 3724 page_t **plarr, 3725 size_t plsz, 3726 struct seg *seg, 3727 caddr_t addr, 3728 enum seg_rw rw, 3729 cred_t *cr, 3730 caller_context_t *ct) 3731 { 3732 int err; 3733 3734 VOPXID_MAP_CR(vp, cr); 3735 3736 err = (*(vp)->v_op->vop_getpage) 3737 (vp, off, len, protp, plarr, plsz, seg, addr, rw, cr, ct); 3738 VOPSTATS_UPDATE(vp, getpage); 3739 return (err); 3740 } 3741 3742 int 3743 fop_putpage( 3744 vnode_t *vp, 3745 offset_t off, 3746 size_t len, 3747 int flags, 3748 cred_t *cr, 3749 caller_context_t *ct) 3750 { 3751 int err; 3752 3753 VOPXID_MAP_CR(vp, cr); 3754 3755 err = (*(vp)->v_op->vop_putpage)(vp, off, len, flags, cr, ct); 3756 VOPSTATS_UPDATE(vp, putpage); 3757 return (err); 3758 } 3759 3760 int 3761 fop_map( 3762 vnode_t *vp, 3763 offset_t off, 3764 struct as *as, 3765 caddr_t *addrp, 3766 size_t len, 3767 uchar_t prot, 3768 uchar_t maxprot, 3769 uint_t flags, 3770 cred_t *cr, 3771 caller_context_t *ct) 3772 { 3773 int err; 3774 3775 VOPXID_MAP_CR(vp, cr); 3776 3777 err = (*(vp)->v_op->vop_map) 3778 (vp, off, as, addrp, len, prot, maxprot, flags, cr, ct); 3779 VOPSTATS_UPDATE(vp, map); 3780 return (err); 3781 } 3782 3783 int 3784 fop_addmap( 3785 vnode_t *vp, 3786 offset_t off, 3787 struct as *as, 3788 caddr_t addr, 3789 size_t len, 3790 uchar_t prot, 3791 uchar_t maxprot, 3792 uint_t flags, 3793 cred_t *cr, 3794 caller_context_t *ct) 3795 { 3796 int error; 3797 u_longlong_t delta; 3798 3799 VOPXID_MAP_CR(vp, cr); 3800 3801 error = (*(vp)->v_op->vop_addmap) 3802 (vp, off, as, addr, len, prot, maxprot, flags, cr, ct); 3803 3804 if ((!error) && (vp->v_type == VREG)) { 3805 delta = (u_longlong_t)btopr(len); 3806 /* 3807 * If file is declared MAP_PRIVATE, it can't be written back 3808 * even if open for write. Handle as read. 3809 */ 3810 if (flags & MAP_PRIVATE) { 3811 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3812 (int64_t)delta); 3813 } else { 3814 /* 3815 * atomic_add_64 forces the fetch of a 64 bit value to 3816 * be atomic on 32 bit machines 3817 */ 3818 if (maxprot & PROT_WRITE) 3819 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)), 3820 (int64_t)delta); 3821 if (maxprot & PROT_READ) 3822 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3823 (int64_t)delta); 3824 if (maxprot & PROT_EXEC) 3825 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3826 (int64_t)delta); 3827 } 3828 } 3829 VOPSTATS_UPDATE(vp, addmap); 3830 return (error); 3831 } 3832 3833 int 3834 fop_delmap( 3835 vnode_t *vp, 3836 offset_t off, 3837 struct as *as, 3838 caddr_t addr, 3839 size_t len, 3840 uint_t prot, 3841 uint_t maxprot, 3842 uint_t flags, 3843 cred_t *cr, 3844 caller_context_t *ct) 3845 { 3846 int error; 3847 u_longlong_t delta; 3848 3849 VOPXID_MAP_CR(vp, cr); 3850 3851 error = (*(vp)->v_op->vop_delmap) 3852 (vp, off, as, addr, len, prot, maxprot, flags, cr, ct); 3853 3854 /* 3855 * NFS calls into delmap twice, the first time 3856 * it simply establishes a callback mechanism and returns EAGAIN 3857 * while the real work is being done upon the second invocation. 3858 * We have to detect this here and only decrement the counts upon 3859 * the second delmap request. 3860 */ 3861 if ((error != EAGAIN) && (vp->v_type == VREG)) { 3862 3863 delta = (u_longlong_t)btopr(len); 3864 3865 if (flags & MAP_PRIVATE) { 3866 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3867 (int64_t)(-delta)); 3868 } else { 3869 /* 3870 * atomic_add_64 forces the fetch of a 64 bit value 3871 * to be atomic on 32 bit machines 3872 */ 3873 if (maxprot & PROT_WRITE) 3874 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)), 3875 (int64_t)(-delta)); 3876 if (maxprot & PROT_READ) 3877 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3878 (int64_t)(-delta)); 3879 if (maxprot & PROT_EXEC) 3880 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3881 (int64_t)(-delta)); 3882 } 3883 } 3884 VOPSTATS_UPDATE(vp, delmap); 3885 return (error); 3886 } 3887 3888 3889 int 3890 fop_poll( 3891 vnode_t *vp, 3892 short events, 3893 int anyyet, 3894 short *reventsp, 3895 struct pollhead **phpp, 3896 caller_context_t *ct) 3897 { 3898 int err; 3899 3900 err = (*(vp)->v_op->vop_poll)(vp, events, anyyet, reventsp, phpp, ct); 3901 VOPSTATS_UPDATE(vp, poll); 3902 return (err); 3903 } 3904 3905 int 3906 fop_dump( 3907 vnode_t *vp, 3908 caddr_t addr, 3909 int lbdn, 3910 int dblks, 3911 caller_context_t *ct) 3912 { 3913 int err; 3914 3915 err = (*(vp)->v_op->vop_dump)(vp, addr, lbdn, dblks, ct); 3916 VOPSTATS_UPDATE(vp, dump); 3917 return (err); 3918 } 3919 3920 int 3921 fop_pathconf( 3922 vnode_t *vp, 3923 int cmd, 3924 ulong_t *valp, 3925 cred_t *cr, 3926 caller_context_t *ct) 3927 { 3928 int err; 3929 3930 VOPXID_MAP_CR(vp, cr); 3931 3932 err = (*(vp)->v_op->vop_pathconf)(vp, cmd, valp, cr, ct); 3933 VOPSTATS_UPDATE(vp, pathconf); 3934 return (err); 3935 } 3936 3937 int 3938 fop_pageio( 3939 vnode_t *vp, 3940 struct page *pp, 3941 u_offset_t io_off, 3942 size_t io_len, 3943 int flags, 3944 cred_t *cr, 3945 caller_context_t *ct) 3946 { 3947 int err; 3948 3949 VOPXID_MAP_CR(vp, cr); 3950 3951 err = (*(vp)->v_op->vop_pageio)(vp, pp, io_off, io_len, flags, cr, ct); 3952 VOPSTATS_UPDATE(vp, pageio); 3953 return (err); 3954 } 3955 3956 int 3957 fop_dumpctl( 3958 vnode_t *vp, 3959 int action, 3960 int *blkp, 3961 caller_context_t *ct) 3962 { 3963 int err; 3964 err = (*(vp)->v_op->vop_dumpctl)(vp, action, blkp, ct); 3965 VOPSTATS_UPDATE(vp, dumpctl); 3966 return (err); 3967 } 3968 3969 void 3970 fop_dispose( 3971 vnode_t *vp, 3972 page_t *pp, 3973 int flag, 3974 int dn, 3975 cred_t *cr, 3976 caller_context_t *ct) 3977 { 3978 /* Must do stats first since it's possible to lose the vnode */ 3979 VOPSTATS_UPDATE(vp, dispose); 3980 3981 VOPXID_MAP_CR(vp, cr); 3982 3983 (*(vp)->v_op->vop_dispose)(vp, pp, flag, dn, cr, ct); 3984 } 3985 3986 int 3987 fop_setsecattr( 3988 vnode_t *vp, 3989 vsecattr_t *vsap, 3990 int flag, 3991 cred_t *cr, 3992 caller_context_t *ct) 3993 { 3994 int err; 3995 3996 VOPXID_MAP_CR(vp, cr); 3997 3998 /* 3999 * We're only allowed to skip the ACL check iff we used a 32 bit 4000 * ACE mask with VOP_ACCESS() to determine permissions. 4001 */ 4002 if ((flag & ATTR_NOACLCHECK) && 4003 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 4004 return (EINVAL); 4005 } 4006 err = (*(vp)->v_op->vop_setsecattr) (vp, vsap, flag, cr, ct); 4007 VOPSTATS_UPDATE(vp, setsecattr); 4008 return (err); 4009 } 4010 4011 int 4012 fop_getsecattr( 4013 vnode_t *vp, 4014 vsecattr_t *vsap, 4015 int flag, 4016 cred_t *cr, 4017 caller_context_t *ct) 4018 { 4019 int err; 4020 4021 /* 4022 * We're only allowed to skip the ACL check iff we used a 32 bit 4023 * ACE mask with VOP_ACCESS() to determine permissions. 4024 */ 4025 if ((flag & ATTR_NOACLCHECK) && 4026 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 4027 return (EINVAL); 4028 } 4029 4030 VOPXID_MAP_CR(vp, cr); 4031 4032 err = (*(vp)->v_op->vop_getsecattr) (vp, vsap, flag, cr, ct); 4033 VOPSTATS_UPDATE(vp, getsecattr); 4034 return (err); 4035 } 4036 4037 int 4038 fop_shrlock( 4039 vnode_t *vp, 4040 int cmd, 4041 struct shrlock *shr, 4042 int flag, 4043 cred_t *cr, 4044 caller_context_t *ct) 4045 { 4046 int err; 4047 4048 VOPXID_MAP_CR(vp, cr); 4049 4050 err = (*(vp)->v_op->vop_shrlock)(vp, cmd, shr, flag, cr, ct); 4051 VOPSTATS_UPDATE(vp, shrlock); 4052 return (err); 4053 } 4054 4055 int 4056 fop_vnevent(vnode_t *vp, vnevent_t vnevent, vnode_t *dvp, char *fnm, 4057 caller_context_t *ct) 4058 { 4059 int err; 4060 4061 err = (*(vp)->v_op->vop_vnevent)(vp, vnevent, dvp, fnm, ct); 4062 VOPSTATS_UPDATE(vp, vnevent); 4063 return (err); 4064 } 4065 4066 /* 4067 * Default destructor 4068 * Needed because NULL destructor means that the key is unused 4069 */ 4070 /* ARGSUSED */ 4071 void 4072 vsd_defaultdestructor(void *value) 4073 {} 4074 4075 /* 4076 * Create a key (index into per vnode array) 4077 * Locks out vsd_create, vsd_destroy, and vsd_free 4078 * May allocate memory with lock held 4079 */ 4080 void 4081 vsd_create(uint_t *keyp, void (*destructor)(void *)) 4082 { 4083 int i; 4084 uint_t nkeys; 4085 4086 /* 4087 * if key is allocated, do nothing 4088 */ 4089 mutex_enter(&vsd_lock); 4090 if (*keyp) { 4091 mutex_exit(&vsd_lock); 4092 return; 4093 } 4094 /* 4095 * find an unused key 4096 */ 4097 if (destructor == NULL) 4098 destructor = vsd_defaultdestructor; 4099 4100 for (i = 0; i < vsd_nkeys; ++i) 4101 if (vsd_destructor[i] == NULL) 4102 break; 4103 4104 /* 4105 * if no unused keys, increase the size of the destructor array 4106 */ 4107 if (i == vsd_nkeys) { 4108 if ((nkeys = (vsd_nkeys << 1)) == 0) 4109 nkeys = 1; 4110 vsd_destructor = 4111 (void (**)(void *))vsd_realloc((void *)vsd_destructor, 4112 (size_t)(vsd_nkeys * sizeof (void (*)(void *))), 4113 (size_t)(nkeys * sizeof (void (*)(void *)))); 4114 vsd_nkeys = nkeys; 4115 } 4116 4117 /* 4118 * allocate the next available unused key 4119 */ 4120 vsd_destructor[i] = destructor; 4121 *keyp = i + 1; 4122 4123 /* create vsd_list, if it doesn't exist */ 4124 if (vsd_list == NULL) { 4125 vsd_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 4126 list_create(vsd_list, sizeof (struct vsd_node), 4127 offsetof(struct vsd_node, vs_nodes)); 4128 } 4129 4130 mutex_exit(&vsd_lock); 4131 } 4132 4133 /* 4134 * Destroy a key 4135 * 4136 * Assumes that the caller is preventing vsd_set and vsd_get 4137 * Locks out vsd_create, vsd_destroy, and vsd_free 4138 * May free memory with lock held 4139 */ 4140 void 4141 vsd_destroy(uint_t *keyp) 4142 { 4143 uint_t key; 4144 struct vsd_node *vsd; 4145 4146 /* 4147 * protect the key namespace and our destructor lists 4148 */ 4149 mutex_enter(&vsd_lock); 4150 key = *keyp; 4151 *keyp = 0; 4152 4153 ASSERT(key <= vsd_nkeys); 4154 4155 /* 4156 * if the key is valid 4157 */ 4158 if (key != 0) { 4159 uint_t k = key - 1; 4160 /* 4161 * for every vnode with VSD, call key's destructor 4162 */ 4163 for (vsd = list_head(vsd_list); vsd != NULL; 4164 vsd = list_next(vsd_list, vsd)) { 4165 /* 4166 * no VSD for key in this vnode 4167 */ 4168 if (key > vsd->vs_nkeys) 4169 continue; 4170 /* 4171 * call destructor for key 4172 */ 4173 if (vsd->vs_value[k] && vsd_destructor[k]) 4174 (*vsd_destructor[k])(vsd->vs_value[k]); 4175 /* 4176 * reset value for key 4177 */ 4178 vsd->vs_value[k] = NULL; 4179 } 4180 /* 4181 * actually free the key (NULL destructor == unused) 4182 */ 4183 vsd_destructor[k] = NULL; 4184 } 4185 4186 mutex_exit(&vsd_lock); 4187 } 4188 4189 /* 4190 * Quickly return the per vnode value that was stored with the specified key 4191 * Assumes the caller is protecting key from vsd_create and vsd_destroy 4192 * Assumes the caller is holding v_lock to protect the vsd. 4193 */ 4194 void * 4195 vsd_get(vnode_t *vp, uint_t key) 4196 { 4197 struct vsd_node *vsd; 4198 4199 /* 4200 * The caller needs to pass a valid vnode. 4201 */ 4202 ASSERT(vp != NULL); 4203 if (vp == NULL) 4204 return (NULL); 4205 4206 vsd = vp->v_vsd; 4207 4208 if (key && vsd != NULL && key <= vsd->vs_nkeys) 4209 return (vsd->vs_value[key - 1]); 4210 return (NULL); 4211 } 4212 4213 /* 4214 * Set a per vnode value indexed with the specified key 4215 * Assumes the caller is holding v_lock to protect the vsd. 4216 */ 4217 int 4218 vsd_set(vnode_t *vp, uint_t key, void *value) 4219 { 4220 struct vsd_node *vsd = vp->v_vsd; 4221 4222 if (key == 0) 4223 return (EINVAL); 4224 if (vsd == NULL) 4225 vsd = vp->v_vsd = kmem_zalloc(sizeof (*vsd), KM_SLEEP); 4226 4227 /* 4228 * If the vsd was just allocated, vs_nkeys will be 0, so the following 4229 * code won't happen and we will continue down and allocate space for 4230 * the vs_value array. 4231 * If the caller is replacing one value with another, then it is up 4232 * to the caller to free/rele/destroy the previous value (if needed). 4233 */ 4234 if (key <= vsd->vs_nkeys) { 4235 vsd->vs_value[key - 1] = value; 4236 return (0); 4237 } 4238 4239 ASSERT(key <= vsd_nkeys); 4240 4241 if (vsd->vs_nkeys == 0) { 4242 mutex_enter(&vsd_lock); /* lock out vsd_destroy() */ 4243 /* 4244 * Link onto list of all VSD nodes. 4245 */ 4246 list_insert_head(vsd_list, vsd); 4247 mutex_exit(&vsd_lock); 4248 } 4249 4250 /* 4251 * Allocate vnode local storage and set the value for key 4252 */ 4253 vsd->vs_value = vsd_realloc(vsd->vs_value, 4254 vsd->vs_nkeys * sizeof (void *), 4255 key * sizeof (void *)); 4256 vsd->vs_nkeys = key; 4257 vsd->vs_value[key - 1] = value; 4258 4259 return (0); 4260 } 4261 4262 /* 4263 * Called from vn_free() to run the destructor function for each vsd 4264 * Locks out vsd_create and vsd_destroy 4265 * Assumes that the destructor *DOES NOT* use vsd 4266 */ 4267 void 4268 vsd_free(vnode_t *vp) 4269 { 4270 int i; 4271 struct vsd_node *vsd = vp->v_vsd; 4272 4273 if (vsd == NULL) 4274 return; 4275 4276 if (vsd->vs_nkeys == 0) { 4277 kmem_free(vsd, sizeof (*vsd)); 4278 vp->v_vsd = NULL; 4279 return; 4280 } 4281 4282 /* 4283 * lock out vsd_create and vsd_destroy, call 4284 * the destructor, and mark the value as destroyed. 4285 */ 4286 mutex_enter(&vsd_lock); 4287 4288 for (i = 0; i < vsd->vs_nkeys; i++) { 4289 if (vsd->vs_value[i] && vsd_destructor[i]) 4290 (*vsd_destructor[i])(vsd->vs_value[i]); 4291 vsd->vs_value[i] = NULL; 4292 } 4293 4294 /* 4295 * remove from linked list of VSD nodes 4296 */ 4297 list_remove(vsd_list, vsd); 4298 4299 mutex_exit(&vsd_lock); 4300 4301 /* 4302 * free up the VSD 4303 */ 4304 kmem_free(vsd->vs_value, vsd->vs_nkeys * sizeof (void *)); 4305 kmem_free(vsd, sizeof (struct vsd_node)); 4306 vp->v_vsd = NULL; 4307 } 4308 4309 /* 4310 * realloc 4311 */ 4312 static void * 4313 vsd_realloc(void *old, size_t osize, size_t nsize) 4314 { 4315 void *new; 4316 4317 new = kmem_zalloc(nsize, KM_SLEEP); 4318 if (old) { 4319 bcopy(old, new, osize); 4320 kmem_free(old, osize); 4321 } 4322 return (new); 4323 } 4324
Indexes created Tue Nov 24 11:10:09 UTC 2009
Terms of Use |
Privacy |
Trademarks |
Copyright Policy |
Site Guidelines |
Help
Your use of this web site or any of its content or software indicates your agreement to be bound by these Terms of Use.
Copyright © 1995-2008 Sun Microsystems, Inc.