1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 25 * Copyright 2016 Nexenta Systems, Inc. All rights reserved. 26 */ 27 28 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */ 29 /* All Rights Reserved */ 30 31 /* 32 * University Copyright- Copyright (c) 1982, 1986, 1988 33 * The Regents of the University of California 34 * All Rights Reserved 35 * 36 * University Acknowledgment- Portions of this document are derived from 37 * software developed by the University of California, Berkeley, and its 38 * contributors. 39 */ 40 41 #include <sys/types.h> 42 #include <sys/param.h> 43 #include <sys/t_lock.h> 44 #include <sys/errno.h> 45 #include <sys/cred.h> 46 #include <sys/user.h> 47 #include <sys/uio.h> 48 #include <sys/file.h> 49 #include <sys/pathname.h> 50 #include <sys/vfs.h> 51 #include <sys/vfs_opreg.h> 52 #include <sys/vnode.h> 53 #include <sys/rwstlock.h> 54 #include <sys/fem.h> 55 #include <sys/stat.h> 56 #include <sys/mode.h> 57 #include <sys/conf.h> 58 #include <sys/sysmacros.h> 59 #include <sys/cmn_err.h> 60 #include <sys/systm.h> 61 #include <sys/kmem.h> 62 #include <sys/debug.h> 63 #include <c2/audit.h> 64 #include <sys/acl.h> 65 #include <sys/nbmlock.h> 66 #include <sys/fcntl.h> 67 #include <fs/fs_subr.h> 68 #include <sys/taskq.h> 69 #include <fs/fs_reparse.h> 70 71 /* Determine if this vnode is a file that is read-only */ 72 #define ISROFILE(vp) \ 73 ((vp)->v_type != VCHR && (vp)->v_type != VBLK && \ 74 (vp)->v_type != VFIFO && vn_is_readonly(vp)) 75 76 /* Tunable via /etc/system; used only by admin/install */ 77 int nfs_global_client_only; 78 79 /* 80 * Array of vopstats_t for per-FS-type vopstats. This array has the same 81 * number of entries as and parallel to the vfssw table. (Arguably, it could 82 * be part of the vfssw table.) Once it's initialized, it's accessed using 83 * the same fstype index that is used to index into the vfssw table. 84 */ 85 vopstats_t **vopstats_fstype; 86 87 /* vopstats initialization template used for fast initialization via bcopy() */ 88 static vopstats_t *vs_templatep; 89 90 /* Kmem cache handle for vsk_anchor_t allocations */ 91 kmem_cache_t *vsk_anchor_cache; 92 93 /* file events cleanup routine */ 94 extern void free_fopdata(vnode_t *); 95 96 /* 97 * Root of AVL tree for the kstats associated with vopstats. Lock protects 98 * updates to vsktat_tree. 99 */ 100 avl_tree_t vskstat_tree; 101 kmutex_t vskstat_tree_lock; 102 103 /* Global variable which enables/disables the vopstats collection */ 104 int vopstats_enabled = 1; 105 106 /* 107 * forward declarations for internal vnode specific data (vsd) 108 */ 109 static void *vsd_realloc(void *, size_t, size_t); 110 111 /* 112 * forward declarations for reparse point functions 113 */ 114 static int fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr); 115 116 /* 117 * VSD -- VNODE SPECIFIC DATA 118 * The v_data pointer is typically used by a file system to store a 119 * pointer to the file system's private node (e.g. ufs inode, nfs rnode). 120 * However, there are times when additional project private data needs 121 * to be stored separately from the data (node) pointed to by v_data. 122 * This additional data could be stored by the file system itself or 123 * by a completely different kernel entity. VSD provides a way for 124 * callers to obtain a key and store a pointer to private data associated 125 * with a vnode. 126 * 127 * Callers are responsible for protecting the vsd by holding v_vsd_lock 128 * for calls to vsd_set() and vsd_get(). 129 */ 130 131 /* 132 * vsd_lock protects: 133 * vsd_nkeys - creation and deletion of vsd keys 134 * vsd_list - insertion and deletion of vsd_node in the vsd_list 135 * vsd_destructor - adding and removing destructors to the list 136 */ 137 static kmutex_t vsd_lock; 138 static uint_t vsd_nkeys; /* size of destructor array */ 139 /* list of vsd_node's */ 140 static list_t *vsd_list = NULL; 141 /* per-key destructor funcs */ 142 static void (**vsd_destructor)(void *); 143 144 /* 145 * The following is the common set of actions needed to update the 146 * vopstats structure from a vnode op. Both VOPSTATS_UPDATE() and 147 * VOPSTATS_UPDATE_IO() do almost the same thing, except for the 148 * recording of the bytes transferred. Since the code is similar 149 * but small, it is nearly a duplicate. Consequently any changes 150 * to one may need to be reflected in the other. 151 * Rundown of the variables: 152 * vp - Pointer to the vnode 153 * counter - Partial name structure member to update in vopstats for counts 154 * bytecounter - Partial name structure member to update in vopstats for bytes 155 * bytesval - Value to update in vopstats for bytes 156 * fstype - Index into vsanchor_fstype[], same as index into vfssw[] 157 * vsp - Pointer to vopstats structure (either in vfs or vsanchor_fstype[i]) 158 */ 159 160 #define VOPSTATS_UPDATE(vp, counter) { \ 161 vfs_t *vfsp = (vp)->v_vfsp; \ 162 if (vfsp && vfsp->vfs_implp && \ 163 (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) { \ 164 vopstats_t *vsp = &vfsp->vfs_vopstats; \ 165 uint64_t *stataddr = &(vsp->n##counter.value.ui64); \ 166 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \ 167 size_t, uint64_t *); \ 168 __dtrace_probe___fsinfo_##counter(vp, 0, stataddr); \ 169 (*stataddr)++; \ 170 if ((vsp = vfsp->vfs_fstypevsp) != NULL) { \ 171 vsp->n##counter.value.ui64++; \ 172 } \ 173 } \ 174 } 175 176 #define VOPSTATS_UPDATE_IO(vp, counter, bytecounter, bytesval) { \ 177 vfs_t *vfsp = (vp)->v_vfsp; \ 178 if (vfsp && vfsp->vfs_implp && \ 179 (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) { \ 180 vopstats_t *vsp = &vfsp->vfs_vopstats; \ 181 uint64_t *stataddr = &(vsp->n##counter.value.ui64); \ 182 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \ 183 size_t, uint64_t *); \ 184 __dtrace_probe___fsinfo_##counter(vp, bytesval, stataddr); \ 185 (*stataddr)++; \ 186 vsp->bytecounter.value.ui64 += bytesval; \ 187 if ((vsp = vfsp->vfs_fstypevsp) != NULL) { \ 188 vsp->n##counter.value.ui64++; \ 189 vsp->bytecounter.value.ui64 += bytesval; \ 190 } \ 191 } \ 192 } 193 194 /* 195 * If the filesystem does not support XIDs map credential 196 * If the vfsp is NULL, perhaps we should also map? 197 */ 198 #define VOPXID_MAP_CR(vp, cr) { \ 199 vfs_t *vfsp = (vp)->v_vfsp; \ 200 if (vfsp != NULL && (vfsp->vfs_flag & VFS_XID) == 0) \ 201 cr = crgetmapped(cr); \ 202 } 203 204 /* 205 * Convert stat(2) formats to vnode types and vice versa. (Knows about 206 * numerical order of S_IFMT and vnode types.) 207 */ 208 enum vtype iftovt_tab[] = { 209 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 210 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON 211 }; 212 213 ushort_t vttoif_tab[] = { 214 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFIFO, 215 S_IFDOOR, 0, S_IFSOCK, S_IFPORT, 0 216 }; 217 218 /* 219 * The system vnode cache. 220 */ 221 222 kmem_cache_t *vn_cache; 223 224 225 /* 226 * Vnode operations vector. 227 */ 228 229 static const fs_operation_trans_def_t vn_ops_table[] = { 230 VOPNAME_OPEN, offsetof(struct vnodeops, vop_open), 231 fs_nosys, fs_nosys, 232 233 VOPNAME_CLOSE, offsetof(struct vnodeops, vop_close), 234 fs_nosys, fs_nosys, 235 236 VOPNAME_READ, offsetof(struct vnodeops, vop_read), 237 fs_nosys, fs_nosys, 238 239 VOPNAME_WRITE, offsetof(struct vnodeops, vop_write), 240 fs_nosys, fs_nosys, 241 242 VOPNAME_IOCTL, offsetof(struct vnodeops, vop_ioctl), 243 fs_nosys, fs_nosys, 244 245 VOPNAME_SETFL, offsetof(struct vnodeops, vop_setfl), 246 fs_setfl, fs_nosys, 247 248 VOPNAME_GETATTR, offsetof(struct vnodeops, vop_getattr), 249 fs_nosys, fs_nosys, 250 251 VOPNAME_SETATTR, offsetof(struct vnodeops, vop_setattr), 252 fs_nosys, fs_nosys, 253 254 VOPNAME_ACCESS, offsetof(struct vnodeops, vop_access), 255 fs_nosys, fs_nosys, 256 257 VOPNAME_LOOKUP, offsetof(struct vnodeops, vop_lookup), 258 fs_nosys, fs_nosys, 259 260 VOPNAME_CREATE, offsetof(struct vnodeops, vop_create), 261 fs_nosys, fs_nosys, 262 263 VOPNAME_REMOVE, offsetof(struct vnodeops, vop_remove), 264 fs_nosys, fs_nosys, 265 266 VOPNAME_LINK, offsetof(struct vnodeops, vop_link), 267 fs_nosys, fs_nosys, 268 269 VOPNAME_RENAME, offsetof(struct vnodeops, vop_rename), 270 fs_nosys, fs_nosys, 271 272 VOPNAME_MKDIR, offsetof(struct vnodeops, vop_mkdir), 273 fs_nosys, fs_nosys, 274 275 VOPNAME_RMDIR, offsetof(struct vnodeops, vop_rmdir), 276 fs_nosys, fs_nosys, 277 278 VOPNAME_READDIR, offsetof(struct vnodeops, vop_readdir), 279 fs_nosys, fs_nosys, 280 281 VOPNAME_SYMLINK, offsetof(struct vnodeops, vop_symlink), 282 fs_nosys, fs_nosys, 283 284 VOPNAME_READLINK, offsetof(struct vnodeops, vop_readlink), 285 fs_nosys, fs_nosys, 286 287 VOPNAME_FSYNC, offsetof(struct vnodeops, vop_fsync), 288 fs_nosys, fs_nosys, 289 290 VOPNAME_INACTIVE, offsetof(struct vnodeops, vop_inactive), 291 fs_nosys, fs_nosys, 292 293 VOPNAME_FID, offsetof(struct vnodeops, vop_fid), 294 fs_nosys, fs_nosys, 295 296 VOPNAME_RWLOCK, offsetof(struct vnodeops, vop_rwlock), 297 fs_rwlock, fs_rwlock, 298 299 VOPNAME_RWUNLOCK, offsetof(struct vnodeops, vop_rwunlock), 300 (fs_generic_func_p) fs_rwunlock, 301 (fs_generic_func_p) fs_rwunlock, /* no errors allowed */ 302 303 VOPNAME_SEEK, offsetof(struct vnodeops, vop_seek), 304 fs_nosys, fs_nosys, 305 306 VOPNAME_CMP, offsetof(struct vnodeops, vop_cmp), 307 fs_cmp, fs_cmp, /* no errors allowed */ 308 309 VOPNAME_FRLOCK, offsetof(struct vnodeops, vop_frlock), 310 fs_frlock, fs_nosys, 311 312 VOPNAME_SPACE, offsetof(struct vnodeops, vop_space), 313 fs_nosys, fs_nosys, 314 315 VOPNAME_REALVP, offsetof(struct vnodeops, vop_realvp), 316 fs_nosys, fs_nosys, 317 318 VOPNAME_GETPAGE, offsetof(struct vnodeops, vop_getpage), 319 fs_nosys, fs_nosys, 320 321 VOPNAME_PUTPAGE, offsetof(struct vnodeops, vop_putpage), 322 fs_nosys, fs_nosys, 323 324 VOPNAME_MAP, offsetof(struct vnodeops, vop_map), 325 (fs_generic_func_p) fs_nosys_map, 326 (fs_generic_func_p) fs_nosys_map, 327 328 VOPNAME_ADDMAP, offsetof(struct vnodeops, vop_addmap), 329 (fs_generic_func_p) fs_nosys_addmap, 330 (fs_generic_func_p) fs_nosys_addmap, 331 332 VOPNAME_DELMAP, offsetof(struct vnodeops, vop_delmap), 333 fs_nosys, fs_nosys, 334 335 VOPNAME_POLL, offsetof(struct vnodeops, vop_poll), 336 (fs_generic_func_p) fs_poll, (fs_generic_func_p) fs_nosys_poll, 337 338 VOPNAME_DUMP, offsetof(struct vnodeops, vop_dump), 339 fs_nosys, fs_nosys, 340 341 VOPNAME_PATHCONF, offsetof(struct vnodeops, vop_pathconf), 342 fs_pathconf, fs_nosys, 343 344 VOPNAME_PAGEIO, offsetof(struct vnodeops, vop_pageio), 345 fs_nosys, fs_nosys, 346 347 VOPNAME_DUMPCTL, offsetof(struct vnodeops, vop_dumpctl), 348 fs_nosys, fs_nosys, 349 350 VOPNAME_DISPOSE, offsetof(struct vnodeops, vop_dispose), 351 (fs_generic_func_p) fs_dispose, 352 (fs_generic_func_p) fs_nodispose, 353 354 VOPNAME_SETSECATTR, offsetof(struct vnodeops, vop_setsecattr), 355 fs_nosys, fs_nosys, 356 357 VOPNAME_GETSECATTR, offsetof(struct vnodeops, vop_getsecattr), 358 fs_fab_acl, fs_nosys, 359 360 VOPNAME_SHRLOCK, offsetof(struct vnodeops, vop_shrlock), 361 fs_shrlock, fs_nosys, 362 363 VOPNAME_VNEVENT, offsetof(struct vnodeops, vop_vnevent), 364 (fs_generic_func_p) fs_vnevent_nosupport, 365 (fs_generic_func_p) fs_vnevent_nosupport, 366 367 VOPNAME_REQZCBUF, offsetof(struct vnodeops, vop_reqzcbuf), 368 fs_nosys, fs_nosys, 369 370 VOPNAME_RETZCBUF, offsetof(struct vnodeops, vop_retzcbuf), 371 fs_nosys, fs_nosys, 372 373 NULL, 0, NULL, NULL 374 }; 375 376 /* Extensible attribute (xva) routines. */ 377 378 /* 379 * Zero out the structure, set the size of the requested/returned bitmaps, 380 * set AT_XVATTR in the embedded vattr_t's va_mask, and set up the pointer 381 * to the returned attributes array. 382 */ 383 void 384 xva_init(xvattr_t *xvap) 385 { 386 bzero(xvap, sizeof (xvattr_t)); 387 xvap->xva_mapsize = XVA_MAPSIZE; 388 xvap->xva_magic = XVA_MAGIC; 389 xvap->xva_vattr.va_mask = AT_XVATTR; 390 xvap->xva_rtnattrmapp = &(xvap->xva_rtnattrmap)[0]; 391 } 392 393 /* 394 * If AT_XVATTR is set, returns a pointer to the embedded xoptattr_t 395 * structure. Otherwise, returns NULL. 396 */ 397 xoptattr_t * 398 xva_getxoptattr(xvattr_t *xvap) 399 { 400 xoptattr_t *xoap = NULL; 401 if (xvap->xva_vattr.va_mask & AT_XVATTR) 402 xoap = &xvap->xva_xoptattrs; 403 return (xoap); 404 } 405 406 /* 407 * Used by the AVL routines to compare two vsk_anchor_t structures in the tree. 408 * We use the f_fsid reported by VFS_STATVFS() since we use that for the 409 * kstat name. 410 */ 411 static int 412 vska_compar(const void *n1, const void *n2) 413 { 414 int ret; 415 ulong_t p1 = ((vsk_anchor_t *)n1)->vsk_fsid; 416 ulong_t p2 = ((vsk_anchor_t *)n2)->vsk_fsid; 417 418 if (p1 < p2) { 419 ret = -1; 420 } else if (p1 > p2) { 421 ret = 1; 422 } else { 423 ret = 0; 424 } 425 426 return (ret); 427 } 428 429 /* 430 * Used to create a single template which will be bcopy()ed to a newly 431 * allocated vsanchor_combo_t structure in new_vsanchor(), below. 432 */ 433 static vopstats_t * 434 create_vopstats_template() 435 { 436 vopstats_t *vsp; 437 438 vsp = kmem_alloc(sizeof (vopstats_t), KM_SLEEP); 439 bzero(vsp, sizeof (*vsp)); /* Start fresh */ 440 441 /* VOP_OPEN */ 442 kstat_named_init(&vsp->nopen, "nopen", KSTAT_DATA_UINT64); 443 /* VOP_CLOSE */ 444 kstat_named_init(&vsp->nclose, "nclose", KSTAT_DATA_UINT64); 445 /* VOP_READ I/O */ 446 kstat_named_init(&vsp->nread, "nread", KSTAT_DATA_UINT64); 447 kstat_named_init(&vsp->read_bytes, "read_bytes", KSTAT_DATA_UINT64); 448 /* VOP_WRITE I/O */ 449 kstat_named_init(&vsp->nwrite, "nwrite", KSTAT_DATA_UINT64); 450 kstat_named_init(&vsp->write_bytes, "write_bytes", KSTAT_DATA_UINT64); 451 /* VOP_IOCTL */ 452 kstat_named_init(&vsp->nioctl, "nioctl", KSTAT_DATA_UINT64); 453 /* VOP_SETFL */ 454 kstat_named_init(&vsp->nsetfl, "nsetfl", KSTAT_DATA_UINT64); 455 /* VOP_GETATTR */ 456 kstat_named_init(&vsp->ngetattr, "ngetattr", KSTAT_DATA_UINT64); 457 /* VOP_SETATTR */ 458 kstat_named_init(&vsp->nsetattr, "nsetattr", KSTAT_DATA_UINT64); 459 /* VOP_ACCESS */ 460 kstat_named_init(&vsp->naccess, "naccess", KSTAT_DATA_UINT64); 461 /* VOP_LOOKUP */ 462 kstat_named_init(&vsp->nlookup, "nlookup", KSTAT_DATA_UINT64); 463 /* VOP_CREATE */ 464 kstat_named_init(&vsp->ncreate, "ncreate", KSTAT_DATA_UINT64); 465 /* VOP_REMOVE */ 466 kstat_named_init(&vsp->nremove, "nremove", KSTAT_DATA_UINT64); 467 /* VOP_LINK */ 468 kstat_named_init(&vsp->nlink, "nlink", KSTAT_DATA_UINT64); 469 /* VOP_RENAME */ 470 kstat_named_init(&vsp->nrename, "nrename", KSTAT_DATA_UINT64); 471 /* VOP_MKDIR */ 472 kstat_named_init(&vsp->nmkdir, "nmkdir", KSTAT_DATA_UINT64); 473 /* VOP_RMDIR */ 474 kstat_named_init(&vsp->nrmdir, "nrmdir", KSTAT_DATA_UINT64); 475 /* VOP_READDIR I/O */ 476 kstat_named_init(&vsp->nreaddir, "nreaddir", KSTAT_DATA_UINT64); 477 kstat_named_init(&vsp->readdir_bytes, "readdir_bytes", 478 KSTAT_DATA_UINT64); 479 /* VOP_SYMLINK */ 480 kstat_named_init(&vsp->nsymlink, "nsymlink", KSTAT_DATA_UINT64); 481 /* VOP_READLINK */ 482 kstat_named_init(&vsp->nreadlink, "nreadlink", KSTAT_DATA_UINT64); 483 /* VOP_FSYNC */ 484 kstat_named_init(&vsp->nfsync, "nfsync", KSTAT_DATA_UINT64); 485 /* VOP_INACTIVE */ 486 kstat_named_init(&vsp->ninactive, "ninactive", KSTAT_DATA_UINT64); 487 /* VOP_FID */ 488 kstat_named_init(&vsp->nfid, "nfid", KSTAT_DATA_UINT64); 489 /* VOP_RWLOCK */ 490 kstat_named_init(&vsp->nrwlock, "nrwlock", KSTAT_DATA_UINT64); 491 /* VOP_RWUNLOCK */ 492 kstat_named_init(&vsp->nrwunlock, "nrwunlock", KSTAT_DATA_UINT64); 493 /* VOP_SEEK */ 494 kstat_named_init(&vsp->nseek, "nseek", KSTAT_DATA_UINT64); 495 /* VOP_CMP */ 496 kstat_named_init(&vsp->ncmp, "ncmp", KSTAT_DATA_UINT64); 497 /* VOP_FRLOCK */ 498 kstat_named_init(&vsp->nfrlock, "nfrlock", KSTAT_DATA_UINT64); 499 /* VOP_SPACE */ 500 kstat_named_init(&vsp->nspace, "nspace", KSTAT_DATA_UINT64); 501 /* VOP_REALVP */ 502 kstat_named_init(&vsp->nrealvp, "nrealvp", KSTAT_DATA_UINT64); 503 /* VOP_GETPAGE */ 504 kstat_named_init(&vsp->ngetpage, "ngetpage", KSTAT_DATA_UINT64); 505 /* VOP_PUTPAGE */ 506 kstat_named_init(&vsp->nputpage, "nputpage", KSTAT_DATA_UINT64); 507 /* VOP_MAP */ 508 kstat_named_init(&vsp->nmap, "nmap", KSTAT_DATA_UINT64); 509 /* VOP_ADDMAP */ 510 kstat_named_init(&vsp->naddmap, "naddmap", KSTAT_DATA_UINT64); 511 /* VOP_DELMAP */ 512 kstat_named_init(&vsp->ndelmap, "ndelmap", KSTAT_DATA_UINT64); 513 /* VOP_POLL */ 514 kstat_named_init(&vsp->npoll, "npoll", KSTAT_DATA_UINT64); 515 /* VOP_DUMP */ 516 kstat_named_init(&vsp->ndump, "ndump", KSTAT_DATA_UINT64); 517 /* VOP_PATHCONF */ 518 kstat_named_init(&vsp->npathconf, "npathconf", KSTAT_DATA_UINT64); 519 /* VOP_PAGEIO */ 520 kstat_named_init(&vsp->npageio, "npageio", KSTAT_DATA_UINT64); 521 /* VOP_DUMPCTL */ 522 kstat_named_init(&vsp->ndumpctl, "ndumpctl", KSTAT_DATA_UINT64); 523 /* VOP_DISPOSE */ 524 kstat_named_init(&vsp->ndispose, "ndispose", KSTAT_DATA_UINT64); 525 /* VOP_SETSECATTR */ 526 kstat_named_init(&vsp->nsetsecattr, "nsetsecattr", KSTAT_DATA_UINT64); 527 /* VOP_GETSECATTR */ 528 kstat_named_init(&vsp->ngetsecattr, "ngetsecattr", KSTAT_DATA_UINT64); 529 /* VOP_SHRLOCK */ 530 kstat_named_init(&vsp->nshrlock, "nshrlock", KSTAT_DATA_UINT64); 531 /* VOP_VNEVENT */ 532 kstat_named_init(&vsp->nvnevent, "nvnevent", KSTAT_DATA_UINT64); 533 /* VOP_REQZCBUF */ 534 kstat_named_init(&vsp->nreqzcbuf, "nreqzcbuf", KSTAT_DATA_UINT64); 535 /* VOP_RETZCBUF */ 536 kstat_named_init(&vsp->nretzcbuf, "nretzcbuf", KSTAT_DATA_UINT64); 537 538 return (vsp); 539 } 540 541 /* 542 * Creates a kstat structure associated with a vopstats structure. 543 */ 544 kstat_t * 545 new_vskstat(char *ksname, vopstats_t *vsp) 546 { 547 kstat_t *ksp; 548 549 if (!vopstats_enabled) { 550 return (NULL); 551 } 552 553 ksp = kstat_create("unix", 0, ksname, "misc", KSTAT_TYPE_NAMED, 554 sizeof (vopstats_t)/sizeof (kstat_named_t), 555 KSTAT_FLAG_VIRTUAL|KSTAT_FLAG_WRITABLE); 556 if (ksp) { 557 ksp->ks_data = vsp; 558 kstat_install(ksp); 559 } 560 561 return (ksp); 562 } 563 564 /* 565 * Called from vfsinit() to initialize the support mechanisms for vopstats 566 */ 567 void 568 vopstats_startup() 569 { 570 if (!vopstats_enabled) 571 return; 572 573 /* 574 * Creates the AVL tree which holds per-vfs vopstat anchors. This 575 * is necessary since we need to check if a kstat exists before we 576 * attempt to create it. Also, initialize its lock. 577 */ 578 avl_create(&vskstat_tree, vska_compar, sizeof (vsk_anchor_t), 579 offsetof(vsk_anchor_t, vsk_node)); 580 mutex_init(&vskstat_tree_lock, NULL, MUTEX_DEFAULT, NULL); 581 582 vsk_anchor_cache = kmem_cache_create("vsk_anchor_cache", 583 sizeof (vsk_anchor_t), sizeof (uintptr_t), NULL, NULL, NULL, 584 NULL, NULL, 0); 585 586 /* 587 * Set up the array of pointers for the vopstats-by-FS-type. 588 * The entries will be allocated/initialized as each file system 589 * goes through modload/mod_installfs. 590 */ 591 vopstats_fstype = (vopstats_t **)kmem_zalloc( 592 (sizeof (vopstats_t *) * nfstype), KM_SLEEP); 593 594 /* Set up the global vopstats initialization template */ 595 vs_templatep = create_vopstats_template(); 596 } 597 598 /* 599 * We need to have the all of the counters zeroed. 600 * The initialization of the vopstats_t includes on the order of 601 * 50 calls to kstat_named_init(). Rather that do that on every call, 602 * we do it once in a template (vs_templatep) then bcopy it over. 603 */ 604 void 605 initialize_vopstats(vopstats_t *vsp) 606 { 607 if (vsp == NULL) 608 return; 609 610 bcopy(vs_templatep, vsp, sizeof (vopstats_t)); 611 } 612 613 /* 614 * If possible, determine which vopstats by fstype to use and 615 * return a pointer to the caller. 616 */ 617 vopstats_t * 618 get_fstype_vopstats(vfs_t *vfsp, struct vfssw *vswp) 619 { 620 int fstype = 0; /* Index into vfssw[] */ 621 vopstats_t *vsp = NULL; 622 623 if (vfsp == NULL || (vfsp->vfs_flag & VFS_STATS) == 0 || 624 !vopstats_enabled) 625 return (NULL); 626 /* 627 * Set up the fstype. We go to so much trouble because all versions 628 * of NFS use the same fstype in their vfs even though they have 629 * distinct entries in the vfssw[] table. 630 * NOTE: A special vfs (e.g., EIO_vfs) may not have an entry. 631 */ 632 if (vswp) { 633 fstype = vswp - vfssw; /* Gets us the index */ 634 } else { 635 fstype = vfsp->vfs_fstype; 636 } 637 638 /* 639 * Point to the per-fstype vopstats. The only valid values are 640 * non-zero positive values less than the number of vfssw[] table 641 * entries. 642 */ 643 if (fstype > 0 && fstype < nfstype) { 644 vsp = vopstats_fstype[fstype]; 645 } 646 647 return (vsp); 648 } 649 650 /* 651 * Generate a kstat name, create the kstat structure, and allocate a 652 * vsk_anchor_t to hold it together. Return the pointer to the vsk_anchor_t 653 * to the caller. This must only be called from a mount. 654 */ 655 vsk_anchor_t * 656 get_vskstat_anchor(vfs_t *vfsp) 657 { 658 char kstatstr[KSTAT_STRLEN]; /* kstat name for vopstats */ 659 statvfs64_t statvfsbuf; /* Needed to find f_fsid */ 660 vsk_anchor_t *vskp = NULL; /* vfs <--> kstat anchor */ 661 kstat_t *ksp; /* Ptr to new kstat */ 662 avl_index_t where; /* Location in the AVL tree */ 663 664 if (vfsp == NULL || vfsp->vfs_implp == NULL || 665 (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled) 666 return (NULL); 667 668 /* Need to get the fsid to build a kstat name */ 669 if (VFS_STATVFS(vfsp, &statvfsbuf) == 0) { 670 /* Create a name for our kstats based on fsid */ 671 (void) snprintf(kstatstr, KSTAT_STRLEN, "%s%lx", 672 VOPSTATS_STR, statvfsbuf.f_fsid); 673 674 /* Allocate and initialize the vsk_anchor_t */ 675 vskp = kmem_cache_alloc(vsk_anchor_cache, KM_SLEEP); 676 bzero(vskp, sizeof (*vskp)); 677 vskp->vsk_fsid = statvfsbuf.f_fsid; 678 679 mutex_enter(&vskstat_tree_lock); 680 if (avl_find(&vskstat_tree, vskp, &where) == NULL) { 681 avl_insert(&vskstat_tree, vskp, where); 682 mutex_exit(&vskstat_tree_lock); 683 684 /* 685 * Now that we've got the anchor in the AVL 686 * tree, we can create the kstat. 687 */ 688 ksp = new_vskstat(kstatstr, &vfsp->vfs_vopstats); 689 if (ksp) { 690 vskp->vsk_ksp = ksp; 691 } 692 } else { 693 /* Oops, found one! Release memory and lock. */ 694 mutex_exit(&vskstat_tree_lock); 695 kmem_cache_free(vsk_anchor_cache, vskp); 696 vskp = NULL; 697 } 698 } 699 return (vskp); 700 } 701 702 /* 703 * We're in the process of tearing down the vfs and need to cleanup 704 * the data structures associated with the vopstats. Must only be called 705 * from dounmount(). 706 */ 707 void 708 teardown_vopstats(vfs_t *vfsp) 709 { 710 vsk_anchor_t *vskap; 711 avl_index_t where; 712 713 if (vfsp == NULL || vfsp->vfs_implp == NULL || 714 (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled) 715 return; 716 717 /* This is a safe check since VFS_STATS must be set (see above) */ 718 if ((vskap = vfsp->vfs_vskap) == NULL) 719 return; 720 721 /* Whack the pointer right away */ 722 vfsp->vfs_vskap = NULL; 723 724 /* Lock the tree, remove the node, and delete the kstat */ 725 mutex_enter(&vskstat_tree_lock); 726 if (avl_find(&vskstat_tree, vskap, &where)) { 727 avl_remove(&vskstat_tree, vskap); 728 } 729 730 if (vskap->vsk_ksp) { 731 kstat_delete(vskap->vsk_ksp); 732 } 733 mutex_exit(&vskstat_tree_lock); 734 735 kmem_cache_free(vsk_anchor_cache, vskap); 736 } 737 738 /* 739 * Read or write a vnode. Called from kernel code. 740 */ 741 int 742 vn_rdwr( 743 enum uio_rw rw, 744 struct vnode *vp, 745 caddr_t base, 746 ssize_t len, 747 offset_t offset, 748 enum uio_seg seg, 749 int ioflag, 750 rlim64_t ulimit, /* meaningful only if rw is UIO_WRITE */ 751 cred_t *cr, 752 ssize_t *residp) 753 { 754 struct uio uio; 755 struct iovec iov; 756 int error; 757 int in_crit = 0; 758 759 if (rw == UIO_WRITE && ISROFILE(vp)) 760 return (EROFS); 761 762 if (len < 0) 763 return (EIO); 764 765 VOPXID_MAP_CR(vp, cr); 766 767 iov.iov_base = base; 768 iov.iov_len = len; 769 uio.uio_iov = &iov; 770 uio.uio_iovcnt = 1; 771 uio.uio_loffset = offset; 772 uio.uio_segflg = (short)seg; 773 uio.uio_resid = len; 774 uio.uio_llimit = ulimit; 775 776 /* 777 * We have to enter the critical region before calling VOP_RWLOCK 778 * to avoid a deadlock with ufs. 779 */ 780 if (nbl_need_check(vp)) { 781 int svmand; 782 783 nbl_start_crit(vp, RW_READER); 784 in_crit = 1; 785 error = nbl_svmand(vp, cr, &svmand); 786 if (error != 0) 787 goto done; 788 if (nbl_conflict(vp, rw == UIO_WRITE ? NBL_WRITE : NBL_READ, 789 uio.uio_offset, uio.uio_resid, svmand, NULL)) { 790 error = EACCES; 791 goto done; 792 } 793 } 794 795 (void) VOP_RWLOCK(vp, 796 rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL); 797 if (rw == UIO_WRITE) { 798 uio.uio_fmode = FWRITE; 799 uio.uio_extflg = UIO_COPY_DEFAULT; 800 error = VOP_WRITE(vp, &uio, ioflag, cr, NULL); 801 } else { 802 uio.uio_fmode = FREAD; 803 uio.uio_extflg = UIO_COPY_CACHED; 804 error = VOP_READ(vp, &uio, ioflag, cr, NULL); 805 } 806 VOP_RWUNLOCK(vp, 807 rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL); 808 if (residp) 809 *residp = uio.uio_resid; 810 else if (uio.uio_resid) 811 error = EIO; 812 813 done: 814 if (in_crit) 815 nbl_end_crit(vp); 816 return (error); 817 } 818 819 /* 820 * Release a vnode. Call VOP_INACTIVE on last reference or 821 * decrement reference count. 822 * 823 * To avoid race conditions, the v_count is left at 1 for 824 * the call to VOP_INACTIVE. This prevents another thread 825 * from reclaiming and releasing the vnode *before* the 826 * VOP_INACTIVE routine has a chance to destroy the vnode. 827 * We can't have more than 1 thread calling VOP_INACTIVE 828 * on a vnode. 829 */ 830 void 831 vn_rele(vnode_t *vp) 832 { 833 VERIFY(vp->v_count > 0); 834 mutex_enter(&vp->v_lock); 835 if (vp->v_count == 1) { 836 mutex_exit(&vp->v_lock); 837 VOP_INACTIVE(vp, CRED(), NULL); 838 return; 839 } 840 vp->v_count--; 841 mutex_exit(&vp->v_lock); 842 } 843 844 /* 845 * Release a vnode referenced by the DNLC. Multiple DNLC references are treated 846 * as a single reference, so v_count is not decremented until the last DNLC hold 847 * is released. This makes it possible to distinguish vnodes that are referenced 848 * only by the DNLC. 849 */ 850 void 851 vn_rele_dnlc(vnode_t *vp) 852 { 853 VERIFY((vp->v_count > 0) && (vp->v_count_dnlc > 0)); 854 mutex_enter(&vp->v_lock); 855 if (--vp->v_count_dnlc == 0) { 856 if (vp->v_count == 1) { 857 mutex_exit(&vp->v_lock); 858 VOP_INACTIVE(vp, CRED(), NULL); 859 return; 860 } 861 vp->v_count--; 862 } 863 mutex_exit(&vp->v_lock); 864 } 865 866 /* 867 * Like vn_rele() except that it clears v_stream under v_lock. 868 * This is used by sockfs when it dismantels the association between 869 * the sockfs node and the vnode in the underlaying file system. 870 * v_lock has to be held to prevent a thread coming through the lookupname 871 * path from accessing a stream head that is going away. 872 */ 873 void 874 vn_rele_stream(vnode_t *vp) 875 { 876 VERIFY(vp->v_count > 0); 877 mutex_enter(&vp->v_lock); 878 vp->v_stream = NULL; 879 if (vp->v_count == 1) { 880 mutex_exit(&vp->v_lock); 881 VOP_INACTIVE(vp, CRED(), NULL); 882 return; 883 } 884 vp->v_count--; 885 mutex_exit(&vp->v_lock); 886 } 887 888 static void 889 vn_rele_inactive(vnode_t *vp) 890 { 891 VOP_INACTIVE(vp, CRED(), NULL); 892 } 893 894 /* 895 * Like vn_rele() except if we are going to call VOP_INACTIVE() then do it 896 * asynchronously using a taskq. This can avoid deadlocks caused by re-entering 897 * the file system as a result of releasing the vnode. Note, file systems 898 * already have to handle the race where the vnode is incremented before the 899 * inactive routine is called and does its locking. 900 * 901 * Warning: Excessive use of this routine can lead to performance problems. 902 * This is because taskqs throttle back allocation if too many are created. 903 */ 904 void 905 vn_rele_async(vnode_t *vp, taskq_t *taskq) 906 { 907 VERIFY(vp->v_count > 0); 908 mutex_enter(&vp->v_lock); 909 if (vp->v_count == 1) { 910 mutex_exit(&vp->v_lock); 911 VERIFY(taskq_dispatch(taskq, (task_func_t *)vn_rele_inactive, 912 vp, TQ_SLEEP) != NULL); 913 return; 914 } 915 vp->v_count--; 916 mutex_exit(&vp->v_lock); 917 } 918 919 int 920 vn_open( 921 char *pnamep, 922 enum uio_seg seg, 923 int filemode, 924 int createmode, 925 struct vnode **vpp, 926 enum create crwhy, 927 mode_t umask) 928 { 929 return (vn_openat(pnamep, seg, filemode, createmode, vpp, crwhy, 930 umask, NULL, -1)); 931 } 932 933 934 /* 935 * Open/create a vnode. 936 * This may be callable by the kernel, the only known use 937 * of user context being that the current user credentials 938 * are used for permissions. crwhy is defined iff filemode & FCREAT. 939 */ 940 int 941 vn_openat( 942 char *pnamep, 943 enum uio_seg seg, 944 int filemode, 945 int createmode, 946 struct vnode **vpp, 947 enum create crwhy, 948 mode_t umask, 949 struct vnode *startvp, 950 int fd) 951 { 952 struct vnode *vp; 953 int mode; 954 int accessflags; 955 int error; 956 int in_crit = 0; 957 int open_done = 0; 958 int shrlock_done = 0; 959 struct vattr vattr; 960 enum symfollow follow; 961 int estale_retry = 0; 962 struct shrlock shr; 963 struct shr_locowner shr_own; 964 965 mode = 0; 966 accessflags = 0; 967 if (filemode & FREAD) 968 mode |= VREAD; 969 if (filemode & (FWRITE|FTRUNC)) 970 mode |= VWRITE; 971 if (filemode & (FSEARCH|FEXEC|FXATTRDIROPEN)) 972 mode |= VEXEC; 973 974 /* symlink interpretation */ 975 if (filemode & FNOFOLLOW) 976 follow = NO_FOLLOW; 977 else 978 follow = FOLLOW; 979 980 if (filemode & FAPPEND) 981 accessflags |= V_APPEND; 982 983 top: 984 if (filemode & FCREAT) { 985 enum vcexcl excl; 986 987 /* 988 * Wish to create a file. 989 */ 990 vattr.va_type = VREG; 991 vattr.va_mode = createmode; 992 vattr.va_mask = AT_TYPE|AT_MODE; 993 if (filemode & FTRUNC) { 994 vattr.va_size = 0; 995 vattr.va_mask |= AT_SIZE; 996 } 997 if (filemode & FEXCL) 998 excl = EXCL; 999 else 1000 excl = NONEXCL; 1001 1002 if (error = 1003 vn_createat(pnamep, seg, &vattr, excl, mode, &vp, crwhy, 1004 (filemode & ~(FTRUNC|FEXCL)), umask, startvp)) 1005 return (error); 1006 } else { 1007 /* 1008 * Wish to open a file. Just look it up. 1009 */ 1010 if (error = lookupnameat(pnamep, seg, follow, 1011 NULLVPP, &vp, startvp)) { 1012 if ((error == ESTALE) && 1013 fs_need_estale_retry(estale_retry++)) 1014 goto top; 1015 return (error); 1016 } 1017 1018 /* 1019 * Get the attributes to check whether file is large. 1020 * We do this only if the FOFFMAX flag is not set and 1021 * only for regular files. 1022 */ 1023 1024 if (!(filemode & FOFFMAX) && (vp->v_type == VREG)) { 1025 vattr.va_mask = AT_SIZE; 1026 if ((error = VOP_GETATTR(vp, &vattr, 0, 1027 CRED(), NULL))) { 1028 goto out; 1029 } 1030 if (vattr.va_size > (u_offset_t)MAXOFF32_T) { 1031 /* 1032 * Large File API - regular open fails 1033 * if FOFFMAX flag is set in file mode 1034 */ 1035 error = EOVERFLOW; 1036 goto out; 1037 } 1038 } 1039 /* 1040 * Can't write directories, active texts, or 1041 * read-only filesystems. Can't truncate files 1042 * on which mandatory locking is in effect. 1043 */ 1044 if (filemode & (FWRITE|FTRUNC)) { 1045 /* 1046 * Allow writable directory if VDIROPEN flag is set. 1047 */ 1048 if (vp->v_type == VDIR && !(vp->v_flag & VDIROPEN)) { 1049 error = EISDIR; 1050 goto out; 1051 } 1052 if (ISROFILE(vp)) { 1053 error = EROFS; 1054 goto out; 1055 } 1056 /* 1057 * Can't truncate files on which 1058 * sysv mandatory locking is in effect. 1059 */ 1060 if (filemode & FTRUNC) { 1061 vnode_t *rvp; 1062 1063 if (VOP_REALVP(vp, &rvp, NULL) != 0) 1064 rvp = vp; 1065 if (rvp->v_filocks != NULL) { 1066 vattr.va_mask = AT_MODE; 1067 if ((error = VOP_GETATTR(vp, 1068 &vattr, 0, CRED(), NULL)) == 0 && 1069 MANDLOCK(vp, vattr.va_mode)) 1070 error = EAGAIN; 1071 } 1072 } 1073 if (error) 1074 goto out; 1075 } 1076 /* 1077 * Check permissions. 1078 */ 1079 if (error = VOP_ACCESS(vp, mode, accessflags, CRED(), NULL)) 1080 goto out; 1081 /* 1082 * Require FSEARCH to return a directory. 1083 * Require FEXEC to return a regular file. 1084 */ 1085 if ((filemode & FSEARCH) && vp->v_type != VDIR) { 1086 error = ENOTDIR; 1087 goto out; 1088 } 1089 if ((filemode & FEXEC) && vp->v_type != VREG) { 1090 error = ENOEXEC; /* XXX: error code? */ 1091 goto out; 1092 } 1093 } 1094 1095 /* 1096 * Do remaining checks for FNOFOLLOW and FNOLINKS. 1097 */ 1098 if ((filemode & FNOFOLLOW) && vp->v_type == VLNK) { 1099 error = ELOOP; 1100 goto out; 1101 } 1102 if (filemode & FNOLINKS) { 1103 vattr.va_mask = AT_NLINK; 1104 if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))) { 1105 goto out; 1106 } 1107 if (vattr.va_nlink != 1) { 1108 error = EMLINK; 1109 goto out; 1110 } 1111 } 1112 1113 /* 1114 * Opening a socket corresponding to the AF_UNIX pathname 1115 * in the filesystem name space is not supported. 1116 * However, VSOCK nodes in namefs are supported in order 1117 * to make fattach work for sockets. 1118 * 1119 * XXX This uses VOP_REALVP to distinguish between 1120 * an unopened namefs node (where VOP_REALVP returns a 1121 * different VSOCK vnode) and a VSOCK created by vn_create 1122 * in some file system (where VOP_REALVP would never return 1123 * a different vnode). 1124 */ 1125 if (vp->v_type == VSOCK) { 1126 struct vnode *nvp; 1127 1128 error = VOP_REALVP(vp, &nvp, NULL); 1129 if (error != 0 || nvp == NULL || nvp == vp || 1130 nvp->v_type != VSOCK) { 1131 error = EOPNOTSUPP; 1132 goto out; 1133 } 1134 } 1135 1136 if ((vp->v_type == VREG) && nbl_need_check(vp)) { 1137 /* get share reservation */ 1138 shr.s_access = 0; 1139 if (filemode & FWRITE) 1140 shr.s_access |= F_WRACC; 1141 if (filemode & FREAD) 1142 shr.s_access |= F_RDACC; 1143 shr.s_deny = 0; 1144 shr.s_sysid = 0; 1145 shr.s_pid = ttoproc(curthread)->p_pid; 1146 shr_own.sl_pid = shr.s_pid; 1147 shr_own.sl_id = fd; 1148 shr.s_own_len = sizeof (shr_own); 1149 shr.s_owner = (caddr_t)&shr_own; 1150 error = VOP_SHRLOCK(vp, F_SHARE_NBMAND, &shr, filemode, CRED(), 1151 NULL); 1152 if (error) 1153 goto out; 1154 shrlock_done = 1; 1155 1156 /* nbmand conflict check if truncating file */ 1157 if ((filemode & FTRUNC) && !(filemode & FCREAT)) { 1158 nbl_start_crit(vp, RW_READER); 1159 in_crit = 1; 1160 1161 vattr.va_mask = AT_SIZE; 1162 if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) 1163 goto out; 1164 if (nbl_conflict(vp, NBL_WRITE, 0, vattr.va_size, 0, 1165 NULL)) { 1166 error = EACCES; 1167 goto out; 1168 } 1169 } 1170 } 1171 1172 /* 1173 * Do opening protocol. 1174 */ 1175 error = VOP_OPEN(&vp, filemode, CRED(), NULL); 1176 if (error) 1177 goto out; 1178 open_done = 1; 1179 1180 /* 1181 * Truncate if required. 1182 */ 1183 if ((filemode & FTRUNC) && !(filemode & FCREAT)) { 1184 vattr.va_size = 0; 1185 vattr.va_mask = AT_SIZE; 1186 if ((error = VOP_SETATTR(vp, &vattr, 0, CRED(), NULL)) != 0) 1187 goto out; 1188 } 1189 out: 1190 ASSERT(vp->v_count > 0); 1191 1192 if (in_crit) { 1193 nbl_end_crit(vp); 1194 in_crit = 0; 1195 } 1196 if (error) { 1197 if (open_done) { 1198 (void) VOP_CLOSE(vp, filemode, 1, (offset_t)0, CRED(), 1199 NULL); 1200 open_done = 0; 1201 shrlock_done = 0; 1202 } 1203 if (shrlock_done) { 1204 (void) VOP_SHRLOCK(vp, F_UNSHARE, &shr, 0, CRED(), 1205 NULL); 1206 shrlock_done = 0; 1207 } 1208 1209 /* 1210 * The following clause was added to handle a problem 1211 * with NFS consistency. It is possible that a lookup 1212 * of the file to be opened succeeded, but the file 1213 * itself doesn't actually exist on the server. This 1214 * is chiefly due to the DNLC containing an entry for 1215 * the file which has been removed on the server. In 1216 * this case, we just start over. If there was some 1217 * other cause for the ESTALE error, then the lookup 1218 * of the file will fail and the error will be returned 1219 * above instead of looping around from here. 1220 */ 1221 VN_RELE(vp); 1222 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1223 goto top; 1224 } else 1225 *vpp = vp; 1226 return (error); 1227 } 1228 1229 /* 1230 * The following two accessor functions are for the NFSv4 server. Since there 1231 * is no VOP_OPEN_UP/DOWNGRADE we need a way for the NFS server to keep the 1232 * vnode open counts correct when a client "upgrades" an open or does an 1233 * open_downgrade. In NFS, an upgrade or downgrade can not only change the 1234 * open mode (add or subtract read or write), but also change the share/deny 1235 * modes. However, share reservations are not integrated with OPEN, yet, so 1236 * we need to handle each separately. These functions are cleaner than having 1237 * the NFS server manipulate the counts directly, however, nobody else should 1238 * use these functions. 1239 */ 1240 void 1241 vn_open_upgrade( 1242 vnode_t *vp, 1243 int filemode) 1244 { 1245 ASSERT(vp->v_type == VREG); 1246 1247 if (filemode & FREAD) 1248 atomic_inc_32(&vp->v_rdcnt); 1249 if (filemode & FWRITE) 1250 atomic_inc_32(&vp->v_wrcnt); 1251 1252 } 1253 1254 void 1255 vn_open_downgrade( 1256 vnode_t *vp, 1257 int filemode) 1258 { 1259 ASSERT(vp->v_type == VREG); 1260 1261 if (filemode & FREAD) { 1262 ASSERT(vp->v_rdcnt > 0); 1263 atomic_dec_32(&vp->v_rdcnt); 1264 } 1265 if (filemode & FWRITE) { 1266 ASSERT(vp->v_wrcnt > 0); 1267 atomic_dec_32(&vp->v_wrcnt); 1268 } 1269 1270 } 1271 1272 int 1273 vn_create( 1274 char *pnamep, 1275 enum uio_seg seg, 1276 struct vattr *vap, 1277 enum vcexcl excl, 1278 int mode, 1279 struct vnode **vpp, 1280 enum create why, 1281 int flag, 1282 mode_t umask) 1283 { 1284 return (vn_createat(pnamep, seg, vap, excl, mode, vpp, why, flag, 1285 umask, NULL)); 1286 } 1287 1288 /* 1289 * Create a vnode (makenode). 1290 */ 1291 int 1292 vn_createat( 1293 char *pnamep, 1294 enum uio_seg seg, 1295 struct vattr *vap, 1296 enum vcexcl excl, 1297 int mode, 1298 struct vnode **vpp, 1299 enum create why, 1300 int flag, 1301 mode_t umask, 1302 struct vnode *startvp) 1303 { 1304 struct vnode *dvp; /* ptr to parent dir vnode */ 1305 struct vnode *vp = NULL; 1306 struct pathname pn; 1307 int error; 1308 int in_crit = 0; 1309 struct vattr vattr; 1310 enum symfollow follow; 1311 int estale_retry = 0; 1312 uint32_t auditing = AU_AUDITING(); 1313 1314 ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 1315 1316 /* symlink interpretation */ 1317 if ((flag & FNOFOLLOW) || excl == EXCL) 1318 follow = NO_FOLLOW; 1319 else 1320 follow = FOLLOW; 1321 flag &= ~(FNOFOLLOW|FNOLINKS); 1322 1323 top: 1324 /* 1325 * Lookup directory. 1326 * If new object is a file, call lower level to create it. 1327 * Note that it is up to the lower level to enforce exclusive 1328 * creation, if the file is already there. 1329 * This allows the lower level to do whatever 1330 * locking or protocol that is needed to prevent races. 1331 * If the new object is directory call lower level to make 1332 * the new directory, with "." and "..". 1333 */ 1334 if (error = pn_get(pnamep, seg, &pn)) 1335 return (error); 1336 if (auditing) 1337 audit_vncreate_start(); 1338 dvp = NULL; 1339 *vpp = NULL; 1340 /* 1341 * lookup will find the parent directory for the vnode. 1342 * When it is done the pn holds the name of the entry 1343 * in the directory. 1344 * If this is a non-exclusive create we also find the node itself. 1345 */ 1346 error = lookuppnat(&pn, NULL, follow, &dvp, 1347 (excl == EXCL) ? NULLVPP : vpp, startvp); 1348 if (error) { 1349 pn_free(&pn); 1350 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1351 goto top; 1352 if (why == CRMKDIR && error == EINVAL) 1353 error = EEXIST; /* SVID */ 1354 return (error); 1355 } 1356 1357 if (why != CRMKNOD) 1358 vap->va_mode &= ~VSVTX; 1359 1360 /* 1361 * If default ACLs are defined for the directory don't apply the 1362 * umask if umask is passed. 1363 */ 1364 1365 if (umask) { 1366 1367 vsecattr_t vsec; 1368 1369 vsec.vsa_aclcnt = 0; 1370 vsec.vsa_aclentp = NULL; 1371 vsec.vsa_dfaclcnt = 0; 1372 vsec.vsa_dfaclentp = NULL; 1373 vsec.vsa_mask = VSA_DFACLCNT; 1374 error = VOP_GETSECATTR(dvp, &vsec, 0, CRED(), NULL); 1375 /* 1376 * If error is ENOSYS then treat it as no error 1377 * Don't want to force all file systems to support 1378 * aclent_t style of ACL's. 1379 */ 1380 if (error == ENOSYS) 1381 error = 0; 1382 if (error) { 1383 if (*vpp != NULL) 1384 VN_RELE(*vpp); 1385 goto out; 1386 } else { 1387 /* 1388 * Apply the umask if no default ACLs. 1389 */ 1390 if (vsec.vsa_dfaclcnt == 0) 1391 vap->va_mode &= ~umask; 1392 1393 /* 1394 * VOP_GETSECATTR() may have allocated memory for 1395 * ACLs we didn't request, so double-check and 1396 * free it if necessary. 1397 */ 1398 if (vsec.vsa_aclcnt && vsec.vsa_aclentp != NULL) 1399 kmem_free((caddr_t)vsec.vsa_aclentp, 1400 vsec.vsa_aclcnt * sizeof (aclent_t)); 1401 if (vsec.vsa_dfaclcnt && vsec.vsa_dfaclentp != NULL) 1402 kmem_free((caddr_t)vsec.vsa_dfaclentp, 1403 vsec.vsa_dfaclcnt * sizeof (aclent_t)); 1404 } 1405 } 1406 1407 /* 1408 * In general we want to generate EROFS if the file system is 1409 * readonly. However, POSIX (IEEE Std. 1003.1) section 5.3.1 1410 * documents the open system call, and it says that O_CREAT has no 1411 * effect if the file already exists. Bug 1119649 states 1412 * that open(path, O_CREAT, ...) fails when attempting to open an 1413 * existing file on a read only file system. Thus, the first part 1414 * of the following if statement has 3 checks: 1415 * if the file exists && 1416 * it is being open with write access && 1417 * the file system is read only 1418 * then generate EROFS 1419 */ 1420 if ((*vpp != NULL && (mode & VWRITE) && ISROFILE(*vpp)) || 1421 (*vpp == NULL && dvp->v_vfsp->vfs_flag & VFS_RDONLY)) { 1422 if (*vpp) 1423 VN_RELE(*vpp); 1424 error = EROFS; 1425 } else if (excl == NONEXCL && *vpp != NULL) { 1426 vnode_t *rvp; 1427 1428 /* 1429 * File already exists. If a mandatory lock has been 1430 * applied, return error. 1431 */ 1432 vp = *vpp; 1433 if (VOP_REALVP(vp, &rvp, NULL) != 0) 1434 rvp = vp; 1435 if ((vap->va_mask & AT_SIZE) && nbl_need_check(vp)) { 1436 nbl_start_crit(vp, RW_READER); 1437 in_crit = 1; 1438 } 1439 if (rvp->v_filocks != NULL || rvp->v_shrlocks != NULL) { 1440 vattr.va_mask = AT_MODE|AT_SIZE; 1441 if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) { 1442 goto out; 1443 } 1444 if (MANDLOCK(vp, vattr.va_mode)) { 1445 error = EAGAIN; 1446 goto out; 1447 } 1448 /* 1449 * File cannot be truncated if non-blocking mandatory 1450 * locks are currently on the file. 1451 */ 1452 if ((vap->va_mask & AT_SIZE) && in_crit) { 1453 u_offset_t offset; 1454 ssize_t length; 1455 1456 offset = vap->va_size > vattr.va_size ? 1457 vattr.va_size : vap->va_size; 1458 length = vap->va_size > vattr.va_size ? 1459 vap->va_size - vattr.va_size : 1460 vattr.va_size - vap->va_size; 1461 if (nbl_conflict(vp, NBL_WRITE, offset, 1462 length, 0, NULL)) { 1463 error = EACCES; 1464 goto out; 1465 } 1466 } 1467 } 1468 1469 /* 1470 * If the file is the root of a VFS, we've crossed a 1471 * mount point and the "containing" directory that we 1472 * acquired above (dvp) is irrelevant because it's in 1473 * a different file system. We apply VOP_CREATE to the 1474 * target itself instead of to the containing directory 1475 * and supply a null path name to indicate (conventionally) 1476 * the node itself as the "component" of interest. 1477 * 1478 * The call to VOP_CREATE() is necessary to ensure 1479 * that the appropriate permission checks are made, 1480 * i.e. EISDIR, EACCES, etc. We already know that vpp 1481 * exists since we are in the else condition where this 1482 * was checked. 1483 */ 1484 if (vp->v_flag & VROOT) { 1485 ASSERT(why != CRMKDIR); 1486 error = VOP_CREATE(vp, "", vap, excl, mode, vpp, 1487 CRED(), flag, NULL, NULL); 1488 /* 1489 * If the create succeeded, it will have created a 1490 * new reference on a new vnode (*vpp) in the child 1491 * file system, so we want to drop our reference on 1492 * the old (vp) upon exit. 1493 */ 1494 goto out; 1495 } 1496 1497 /* 1498 * Large File API - non-large open (FOFFMAX flag not set) 1499 * of regular file fails if the file size exceeds MAXOFF32_T. 1500 */ 1501 if (why != CRMKDIR && 1502 !(flag & FOFFMAX) && 1503 (vp->v_type == VREG)) { 1504 vattr.va_mask = AT_SIZE; 1505 if ((error = VOP_GETATTR(vp, &vattr, 0, 1506 CRED(), NULL))) { 1507 goto out; 1508 } 1509 if ((vattr.va_size > (u_offset_t)MAXOFF32_T)) { 1510 error = EOVERFLOW; 1511 goto out; 1512 } 1513 } 1514 } 1515 1516 if (error == 0) { 1517 /* 1518 * Call mkdir() if specified, otherwise create(). 1519 */ 1520 int must_be_dir = pn_fixslash(&pn); /* trailing '/'? */ 1521 1522 if (why == CRMKDIR) 1523 /* 1524 * N.B., if vn_createat() ever requests 1525 * case-insensitive behavior then it will need 1526 * to be passed to VOP_MKDIR(). VOP_CREATE() 1527 * will already get it via "flag" 1528 */ 1529 error = VOP_MKDIR(dvp, pn.pn_path, vap, vpp, CRED(), 1530 NULL, 0, NULL); 1531 else if (!must_be_dir) 1532 error = VOP_CREATE(dvp, pn.pn_path, vap, 1533 excl, mode, vpp, CRED(), flag, NULL, NULL); 1534 else 1535 error = ENOTDIR; 1536 } 1537 1538 out: 1539 1540 if (auditing) 1541 audit_vncreate_finish(*vpp, error); 1542 if (in_crit) { 1543 nbl_end_crit(vp); 1544 in_crit = 0; 1545 } 1546 if (vp != NULL) { 1547 VN_RELE(vp); 1548 vp = NULL; 1549 } 1550 pn_free(&pn); 1551 VN_RELE(dvp); 1552 /* 1553 * The following clause was added to handle a problem 1554 * with NFS consistency. It is possible that a lookup 1555 * of the file to be created succeeded, but the file 1556 * itself doesn't actually exist on the server. This 1557 * is chiefly due to the DNLC containing an entry for 1558 * the file which has been removed on the server. In 1559 * this case, we just start over. If there was some 1560 * other cause for the ESTALE error, then the lookup 1561 * of the file will fail and the error will be returned 1562 * above instead of looping around from here. 1563 */ 1564 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1565 goto top; 1566 return (error); 1567 } 1568 1569 int 1570 vn_link(char *from, char *to, enum uio_seg seg) 1571 { 1572 return (vn_linkat(NULL, from, NO_FOLLOW, NULL, to, seg)); 1573 } 1574 1575 int 1576 vn_linkat(vnode_t *fstartvp, char *from, enum symfollow follow, 1577 vnode_t *tstartvp, char *to, enum uio_seg seg) 1578 { 1579 struct vnode *fvp; /* from vnode ptr */ 1580 struct vnode *tdvp; /* to directory vnode ptr */ 1581 struct pathname pn; 1582 int error; 1583 struct vattr vattr; 1584 dev_t fsid; 1585 int estale_retry = 0; 1586 uint32_t auditing = AU_AUDITING(); 1587 1588 top: 1589 fvp = tdvp = NULL; 1590 if (error = pn_get(to, seg, &pn)) 1591 return (error); 1592 if (auditing && fstartvp != NULL) 1593 audit_setfsat_path(1); 1594 if (error = lookupnameat(from, seg, follow, NULLVPP, &fvp, fstartvp)) 1595 goto out; 1596 if (auditing && tstartvp != NULL) 1597 audit_setfsat_path(3); 1598 if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &tdvp, NULLVPP, tstartvp)) 1599 goto out; 1600 /* 1601 * Make sure both source vnode and target directory vnode are 1602 * in the same vfs and that it is writeable. 1603 */ 1604 vattr.va_mask = AT_FSID; 1605 if (error = VOP_GETATTR(fvp, &vattr, 0, CRED(), NULL)) 1606 goto out; 1607 fsid = vattr.va_fsid; 1608 vattr.va_mask = AT_FSID; 1609 if (error = VOP_GETATTR(tdvp, &vattr, 0, CRED(), NULL)) 1610 goto out; 1611 if (fsid != vattr.va_fsid) { 1612 error = EXDEV; 1613 goto out; 1614 } 1615 if (tdvp->v_vfsp->vfs_flag & VFS_RDONLY) { 1616 error = EROFS; 1617 goto out; 1618 } 1619 /* 1620 * Do the link. 1621 */ 1622 (void) pn_fixslash(&pn); 1623 error = VOP_LINK(tdvp, fvp, pn.pn_path, CRED(), NULL, 0); 1624 out: 1625 pn_free(&pn); 1626 if (fvp) 1627 VN_RELE(fvp); 1628 if (tdvp) 1629 VN_RELE(tdvp); 1630 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1631 goto top; 1632 return (error); 1633 } 1634 1635 int 1636 vn_rename(char *from, char *to, enum uio_seg seg) 1637 { 1638 return (vn_renameat(NULL, from, NULL, to, seg)); 1639 } 1640 1641 int 1642 vn_renameat(vnode_t *fdvp, char *fname, vnode_t *tdvp, 1643 char *tname, enum uio_seg seg) 1644 { 1645 int error; 1646 struct vattr vattr; 1647 struct pathname fpn; /* from pathname */ 1648 struct pathname tpn; /* to pathname */ 1649 dev_t fsid; 1650 int in_crit_src, in_crit_targ; 1651 vnode_t *fromvp, *fvp; 1652 vnode_t *tovp, *targvp; 1653 int estale_retry = 0; 1654 uint32_t auditing = AU_AUDITING(); 1655 1656 top: 1657 fvp = fromvp = tovp = targvp = NULL; 1658 in_crit_src = in_crit_targ = 0; 1659 /* 1660 * Get to and from pathnames. 1661 */ 1662 if (error = pn_get(fname, seg, &fpn)) 1663 return (error); 1664 if (error = pn_get(tname, seg, &tpn)) { 1665 pn_free(&fpn); 1666 return (error); 1667 } 1668 1669 /* 1670 * First we need to resolve the correct directories 1671 * The passed in directories may only be a starting point, 1672 * but we need the real directories the file(s) live in. 1673 * For example the fname may be something like usr/lib/sparc 1674 * and we were passed in the / directory, but we need to 1675 * use the lib directory for the rename. 1676 */ 1677 1678 if (auditing && fdvp != NULL) 1679 audit_setfsat_path(1); 1680 /* 1681 * Lookup to and from directories. 1682 */ 1683 if (error = lookuppnat(&fpn, NULL, NO_FOLLOW, &fromvp, &fvp, fdvp)) { 1684 goto out; 1685 } 1686 1687 /* 1688 * Make sure there is an entry. 1689 */ 1690 if (fvp == NULL) { 1691 error = ENOENT; 1692 goto out; 1693 } 1694 1695 if (auditing && tdvp != NULL) 1696 audit_setfsat_path(3); 1697 if (error = lookuppnat(&tpn, NULL, NO_FOLLOW, &tovp, &targvp, tdvp)) { 1698 goto out; 1699 } 1700 1701 /* 1702 * Make sure both the from vnode directory and the to directory 1703 * are in the same vfs and the to directory is writable. 1704 * We check fsid's, not vfs pointers, so loopback fs works. 1705 */ 1706 if (fromvp != tovp) { 1707 vattr.va_mask = AT_FSID; 1708 if (error = VOP_GETATTR(fromvp, &vattr, 0, CRED(), NULL)) 1709 goto out; 1710 fsid = vattr.va_fsid; 1711 vattr.va_mask = AT_FSID; 1712 if (error = VOP_GETATTR(tovp, &vattr, 0, CRED(), NULL)) 1713 goto out; 1714 if (fsid != vattr.va_fsid) { 1715 error = EXDEV; 1716 goto out; 1717 } 1718 } 1719 1720 if (tovp->v_vfsp->vfs_flag & VFS_RDONLY) { 1721 error = EROFS; 1722 goto out; 1723 } 1724 1725 if (targvp && (fvp != targvp)) { 1726 nbl_start_crit(targvp, RW_READER); 1727 in_crit_targ = 1; 1728 if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) { 1729 error = EACCES; 1730 goto out; 1731 } 1732 } 1733 1734 if (nbl_need_check(fvp)) { 1735 nbl_start_crit(fvp, RW_READER); 1736 in_crit_src = 1; 1737 if (nbl_conflict(fvp, NBL_RENAME, 0, 0, 0, NULL)) { 1738 error = EACCES; 1739 goto out; 1740 } 1741 } 1742 1743 /* 1744 * Do the rename. 1745 */ 1746 (void) pn_fixslash(&tpn); 1747 error = VOP_RENAME(fromvp, fpn.pn_path, tovp, tpn.pn_path, CRED(), 1748 NULL, 0); 1749 1750 out: 1751 pn_free(&fpn); 1752 pn_free(&tpn); 1753 if (in_crit_src) 1754 nbl_end_crit(fvp); 1755 if (in_crit_targ) 1756 nbl_end_crit(targvp); 1757 if (fromvp) 1758 VN_RELE(fromvp); 1759 if (tovp) 1760 VN_RELE(tovp); 1761 if (targvp) 1762 VN_RELE(targvp); 1763 if (fvp) 1764 VN_RELE(fvp); 1765 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1766 goto top; 1767 return (error); 1768 } 1769 1770 /* 1771 * Remove a file or directory. 1772 */ 1773 int 1774 vn_remove(char *fnamep, enum uio_seg seg, enum rm dirflag) 1775 { 1776 return (vn_removeat(NULL, fnamep, seg, dirflag)); 1777 } 1778 1779 int 1780 vn_removeat(vnode_t *startvp, char *fnamep, enum uio_seg seg, enum rm dirflag) 1781 { 1782 struct vnode *vp; /* entry vnode */ 1783 struct vnode *dvp; /* ptr to parent dir vnode */ 1784 struct vnode *coveredvp; 1785 struct pathname pn; /* name of entry */ 1786 enum vtype vtype; 1787 int error; 1788 struct vfs *vfsp; 1789 struct vfs *dvfsp; /* ptr to parent dir vfs */ 1790 int in_crit = 0; 1791 int estale_retry = 0; 1792 1793 top: 1794 if (error = pn_get(fnamep, seg, &pn)) 1795 return (error); 1796 dvp = vp = NULL; 1797 if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &dvp, &vp, startvp)) { 1798 pn_free(&pn); 1799 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1800 goto top; 1801 return (error); 1802 } 1803 1804 /* 1805 * Make sure there is an entry. 1806 */ 1807 if (vp == NULL) { 1808 error = ENOENT; 1809 goto out; 1810 } 1811 1812 vfsp = vp->v_vfsp; 1813 dvfsp = dvp->v_vfsp; 1814 1815 /* 1816 * If the named file is the root of a mounted filesystem, fail, 1817 * unless it's marked unlinkable. In that case, unmount the 1818 * filesystem and proceed to unlink the covered vnode. (If the 1819 * covered vnode is a directory, use rmdir instead of unlink, 1820 * to avoid file system corruption.) 1821 */ 1822 if (vp->v_flag & VROOT) { 1823 if ((vfsp->vfs_flag & VFS_UNLINKABLE) == 0) { 1824 error = EBUSY; 1825 goto out; 1826 } 1827 1828 /* 1829 * Namefs specific code starts here. 1830 */ 1831 1832 if (dirflag == RMDIRECTORY) { 1833 /* 1834 * User called rmdir(2) on a file that has 1835 * been namefs mounted on top of. Since 1836 * namefs doesn't allow directories to 1837 * be mounted on other files we know 1838 * vp is not of type VDIR so fail to operation. 1839 */ 1840 error = ENOTDIR; 1841 goto out; 1842 } 1843 1844 /* 1845 * If VROOT is still set after grabbing vp->v_lock, 1846 * noone has finished nm_unmount so far and coveredvp 1847 * is valid. 1848 * If we manage to grab vn_vfswlock(coveredvp) before releasing 1849 * vp->v_lock, any race window is eliminated. 1850 */ 1851 1852 mutex_enter(&vp->v_lock); 1853 if ((vp->v_flag & VROOT) == 0) { 1854 /* Someone beat us to the unmount */ 1855 mutex_exit(&vp->v_lock); 1856 error = EBUSY; 1857 goto out; 1858 } 1859 vfsp = vp->v_vfsp; 1860 coveredvp = vfsp->vfs_vnodecovered; 1861 ASSERT(coveredvp); 1862 /* 1863 * Note: Implementation of vn_vfswlock shows that ordering of 1864 * v_lock / vn_vfswlock is not an issue here. 1865 */ 1866 error = vn_vfswlock(coveredvp); 1867 mutex_exit(&vp->v_lock); 1868 1869 if (error) 1870 goto out; 1871 1872 VN_HOLD(coveredvp); 1873 VN_RELE(vp); 1874 error = dounmount(vfsp, 0, CRED()); 1875 1876 /* 1877 * Unmounted the namefs file system; now get 1878 * the object it was mounted over. 1879 */ 1880 vp = coveredvp; 1881 /* 1882 * If namefs was mounted over a directory, then 1883 * we want to use rmdir() instead of unlink(). 1884 */ 1885 if (vp->v_type == VDIR) 1886 dirflag = RMDIRECTORY; 1887 1888 if (error) 1889 goto out; 1890 } 1891 1892 /* 1893 * Make sure filesystem is writeable. 1894 * We check the parent directory's vfs in case this is an lofs vnode. 1895 */ 1896 if (dvfsp && dvfsp->vfs_flag & VFS_RDONLY) { 1897 error = EROFS; 1898 goto out; 1899 } 1900 1901 vtype = vp->v_type; 1902 1903 /* 1904 * If there is the possibility of an nbmand share reservation, make 1905 * sure it's okay to remove the file. Keep a reference to the 1906 * vnode, so that we can exit the nbl critical region after 1907 * calling VOP_REMOVE. 1908 * If there is no possibility of an nbmand share reservation, 1909 * release the vnode reference now. Filesystems like NFS may 1910 * behave differently if there is an extra reference, so get rid of 1911 * this one. Fortunately, we can't have nbmand mounts on NFS 1912 * filesystems. 1913 */ 1914 if (nbl_need_check(vp)) { 1915 nbl_start_crit(vp, RW_READER); 1916 in_crit = 1; 1917 if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) { 1918 error = EACCES; 1919 goto out; 1920 } 1921 } else { 1922 VN_RELE(vp); 1923 vp = NULL; 1924 } 1925 1926 if (dirflag == RMDIRECTORY) { 1927 /* 1928 * Caller is using rmdir(2), which can only be applied to 1929 * directories. 1930 */ 1931 if (vtype != VDIR) { 1932 error = ENOTDIR; 1933 } else { 1934 vnode_t *cwd; 1935 proc_t *pp = curproc; 1936 1937 mutex_enter(&pp->p_lock); 1938 cwd = PTOU(pp)->u_cdir; 1939 VN_HOLD(cwd); 1940 mutex_exit(&pp->p_lock); 1941 error = VOP_RMDIR(dvp, pn.pn_path, cwd, CRED(), 1942 NULL, 0); 1943 VN_RELE(cwd); 1944 } 1945 } else { 1946 /* 1947 * Unlink(2) can be applied to anything. 1948 */ 1949 error = VOP_REMOVE(dvp, pn.pn_path, CRED(), NULL, 0); 1950 } 1951 1952 out: 1953 pn_free(&pn); 1954 if (in_crit) { 1955 nbl_end_crit(vp); 1956 in_crit = 0; 1957 } 1958 if (vp != NULL) 1959 VN_RELE(vp); 1960 if (dvp != NULL) 1961 VN_RELE(dvp); 1962 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++)) 1963 goto top; 1964 return (error); 1965 } 1966 1967 /* 1968 * Utility function to compare equality of vnodes. 1969 * Compare the underlying real vnodes, if there are underlying vnodes. 1970 * This is a more thorough comparison than the VN_CMP() macro provides. 1971 */ 1972 int 1973 vn_compare(vnode_t *vp1, vnode_t *vp2) 1974 { 1975 vnode_t *realvp; 1976 1977 if (vp1 != NULL && VOP_REALVP(vp1, &realvp, NULL) == 0) 1978 vp1 = realvp; 1979 if (vp2 != NULL && VOP_REALVP(vp2, &realvp, NULL) == 0) 1980 vp2 = realvp; 1981 return (VN_CMP(vp1, vp2)); 1982 } 1983 1984 /* 1985 * The number of locks to hash into. This value must be a power 1986 * of 2 minus 1 and should probably also be prime. 1987 */ 1988 #define NUM_BUCKETS 1023 1989 1990 struct vn_vfslocks_bucket { 1991 kmutex_t vb_lock; 1992 vn_vfslocks_entry_t *vb_list; 1993 char pad[64 - sizeof (kmutex_t) - sizeof (void *)]; 1994 }; 1995 1996 /* 1997 * Total number of buckets will be NUM_BUCKETS + 1 . 1998 */ 1999 2000 #pragma align 64(vn_vfslocks_buckets) 2001 static struct vn_vfslocks_bucket vn_vfslocks_buckets[NUM_BUCKETS + 1]; 2002 2003 #define VN_VFSLOCKS_SHIFT 9 2004 2005 #define VN_VFSLOCKS_HASH(vfsvpptr) \ 2006 ((((intptr_t)(vfsvpptr)) >> VN_VFSLOCKS_SHIFT) & NUM_BUCKETS) 2007 2008 /* 2009 * vn_vfslocks_getlock() uses an HASH scheme to generate 2010 * rwstlock using vfs/vnode pointer passed to it. 2011 * 2012 * vn_vfslocks_rele() releases a reference in the 2013 * HASH table which allows the entry allocated by 2014 * vn_vfslocks_getlock() to be freed at a later 2015 * stage when the refcount drops to zero. 2016 */ 2017 2018 vn_vfslocks_entry_t * 2019 vn_vfslocks_getlock(void *vfsvpptr) 2020 { 2021 struct vn_vfslocks_bucket *bp; 2022 vn_vfslocks_entry_t *vep; 2023 vn_vfslocks_entry_t *tvep; 2024 2025 ASSERT(vfsvpptr != NULL); 2026 bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vfsvpptr)]; 2027 2028 mutex_enter(&bp->vb_lock); 2029 for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) { 2030 if (vep->ve_vpvfs == vfsvpptr) { 2031 vep->ve_refcnt++; 2032 mutex_exit(&bp->vb_lock); 2033 return (vep); 2034 } 2035 } 2036 mutex_exit(&bp->vb_lock); 2037 vep = kmem_alloc(sizeof (*vep), KM_SLEEP); 2038 rwst_init(&vep->ve_lock, NULL, RW_DEFAULT, NULL); 2039 vep->ve_vpvfs = (char *)vfsvpptr; 2040 vep->ve_refcnt = 1; 2041 mutex_enter(&bp->vb_lock); 2042 for (tvep = bp->vb_list; tvep != NULL; tvep = tvep->ve_next) { 2043 if (tvep->ve_vpvfs == vfsvpptr) { 2044 tvep->ve_refcnt++; 2045 mutex_exit(&bp->vb_lock); 2046 2047 /* 2048 * There is already an entry in the hash 2049 * destroy what we just allocated. 2050 */ 2051 rwst_destroy(&vep->ve_lock); 2052 kmem_free(vep, sizeof (*vep)); 2053 return (tvep); 2054 } 2055 } 2056 vep->ve_next = bp->vb_list; 2057 bp->vb_list = vep; 2058 mutex_exit(&bp->vb_lock); 2059 return (vep); 2060 } 2061 2062 void 2063 vn_vfslocks_rele(vn_vfslocks_entry_t *vepent) 2064 { 2065 struct vn_vfslocks_bucket *bp; 2066 vn_vfslocks_entry_t *vep; 2067 vn_vfslocks_entry_t *pvep; 2068 2069 ASSERT(vepent != NULL); 2070 ASSERT(vepent->ve_vpvfs != NULL); 2071 2072 bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vepent->ve_vpvfs)]; 2073 2074 mutex_enter(&bp->vb_lock); 2075 vepent->ve_refcnt--; 2076 2077 if ((int32_t)vepent->ve_refcnt < 0) 2078 cmn_err(CE_PANIC, "vn_vfslocks_rele: refcount negative"); 2079 2080 if (vepent->ve_refcnt == 0) { 2081 for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) { 2082 if (vep->ve_vpvfs == vepent->ve_vpvfs) { 2083 if (bp->vb_list == vep) 2084 bp->vb_list = vep->ve_next; 2085 else { 2086 /* LINTED */ 2087 pvep->ve_next = vep->ve_next; 2088 } 2089 mutex_exit(&bp->vb_lock); 2090 rwst_destroy(&vep->ve_lock); 2091 kmem_free(vep, sizeof (*vep)); 2092 return; 2093 } 2094 pvep = vep; 2095 } 2096 cmn_err(CE_PANIC, "vn_vfslocks_rele: vp/vfs not found"); 2097 } 2098 mutex_exit(&bp->vb_lock); 2099 } 2100 2101 /* 2102 * vn_vfswlock_wait is used to implement a lock which is logically a writers 2103 * lock protecting the v_vfsmountedhere field. 2104 * vn_vfswlock_wait has been modified to be similar to vn_vfswlock, 2105 * except that it blocks to acquire the lock VVFSLOCK. 2106 * 2107 * traverse() and routines re-implementing part of traverse (e.g. autofs) 2108 * need to hold this lock. mount(), vn_rename(), vn_remove() and so on 2109 * need the non-blocking version of the writers lock i.e. vn_vfswlock 2110 */ 2111 int 2112 vn_vfswlock_wait(vnode_t *vp) 2113 { 2114 int retval; 2115 vn_vfslocks_entry_t *vpvfsentry; 2116 ASSERT(vp != NULL); 2117 2118 vpvfsentry = vn_vfslocks_getlock(vp); 2119 retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_WRITER); 2120 2121 if (retval == EINTR) { 2122 vn_vfslocks_rele(vpvfsentry); 2123 return (EINTR); 2124 } 2125 return (retval); 2126 } 2127 2128 int 2129 vn_vfsrlock_wait(vnode_t *vp) 2130 { 2131 int retval; 2132 vn_vfslocks_entry_t *vpvfsentry; 2133 ASSERT(vp != NULL); 2134 2135 vpvfsentry = vn_vfslocks_getlock(vp); 2136 retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_READER); 2137 2138 if (retval == EINTR) { 2139 vn_vfslocks_rele(vpvfsentry); 2140 return (EINTR); 2141 } 2142 2143 return (retval); 2144 } 2145 2146 2147 /* 2148 * vn_vfswlock is used to implement a lock which is logically a writers lock 2149 * protecting the v_vfsmountedhere field. 2150 */ 2151 int 2152 vn_vfswlock(vnode_t *vp) 2153 { 2154 vn_vfslocks_entry_t *vpvfsentry; 2155 2156 /* 2157 * If vp is NULL then somebody is trying to lock the covered vnode 2158 * of /. (vfs_vnodecovered is NULL for /). This situation will 2159 * only happen when unmounting /. Since that operation will fail 2160 * anyway, return EBUSY here instead of in VFS_UNMOUNT. 2161 */ 2162 if (vp == NULL) 2163 return (EBUSY); 2164 2165 vpvfsentry = vn_vfslocks_getlock(vp); 2166 2167 if (rwst_tryenter(&vpvfsentry->ve_lock, RW_WRITER)) 2168 return (0); 2169 2170 vn_vfslocks_rele(vpvfsentry); 2171 return (EBUSY); 2172 } 2173 2174 int 2175 vn_vfsrlock(vnode_t *vp) 2176 { 2177 vn_vfslocks_entry_t *vpvfsentry; 2178 2179 /* 2180 * If vp is NULL then somebody is trying to lock the covered vnode 2181 * of /. (vfs_vnodecovered is NULL for /). This situation will 2182 * only happen when unmounting /. Since that operation will fail 2183 * anyway, return EBUSY here instead of in VFS_UNMOUNT. 2184 */ 2185 if (vp == NULL) 2186 return (EBUSY); 2187 2188 vpvfsentry = vn_vfslocks_getlock(vp); 2189 2190 if (rwst_tryenter(&vpvfsentry->ve_lock, RW_READER)) 2191 return (0); 2192 2193 vn_vfslocks_rele(vpvfsentry); 2194 return (EBUSY); 2195 } 2196 2197 void 2198 vn_vfsunlock(vnode_t *vp) 2199 { 2200 vn_vfslocks_entry_t *vpvfsentry; 2201 2202 /* 2203 * ve_refcnt needs to be decremented twice. 2204 * 1. To release refernce after a call to vn_vfslocks_getlock() 2205 * 2. To release the reference from the locking routines like 2206 * vn_vfsrlock/vn_vfswlock etc,. 2207 */ 2208 vpvfsentry = vn_vfslocks_getlock(vp); 2209 vn_vfslocks_rele(vpvfsentry); 2210 2211 rwst_exit(&vpvfsentry->ve_lock); 2212 vn_vfslocks_rele(vpvfsentry); 2213 } 2214 2215 int 2216 vn_vfswlock_held(vnode_t *vp) 2217 { 2218 int held; 2219 vn_vfslocks_entry_t *vpvfsentry; 2220 2221 ASSERT(vp != NULL); 2222 2223 vpvfsentry = vn_vfslocks_getlock(vp); 2224 held = rwst_lock_held(&vpvfsentry->ve_lock, RW_WRITER); 2225 2226 vn_vfslocks_rele(vpvfsentry); 2227 return (held); 2228 } 2229 2230 2231 int 2232 vn_make_ops( 2233 const char *name, /* Name of file system */ 2234 const fs_operation_def_t *templ, /* Operation specification */ 2235 vnodeops_t **actual) /* Return the vnodeops */ 2236 { 2237 int unused_ops; 2238 int error; 2239 2240 *actual = (vnodeops_t *)kmem_alloc(sizeof (vnodeops_t), KM_SLEEP); 2241 2242 (*actual)->vnop_name = name; 2243 2244 error = fs_build_vector(*actual, &unused_ops, vn_ops_table, templ); 2245 if (error) { 2246 kmem_free(*actual, sizeof (vnodeops_t)); 2247 } 2248 2249 #if DEBUG 2250 if (unused_ops != 0) 2251 cmn_err(CE_WARN, "vn_make_ops: %s: %d operations supplied " 2252 "but not used", name, unused_ops); 2253 #endif 2254 2255 return (error); 2256 } 2257 2258 /* 2259 * Free the vnodeops created as a result of vn_make_ops() 2260 */ 2261 void 2262 vn_freevnodeops(vnodeops_t *vnops) 2263 { 2264 kmem_free(vnops, sizeof (vnodeops_t)); 2265 } 2266 2267 /* 2268 * Vnode cache. 2269 */ 2270 2271 /* ARGSUSED */ 2272 static int 2273 vn_cache_constructor(void *buf, void *cdrarg, int kmflags) 2274 { 2275 struct vnode *vp; 2276 2277 vp = buf; 2278 2279 mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL); 2280 mutex_init(&vp->v_vsd_lock, NULL, MUTEX_DEFAULT, NULL); 2281 cv_init(&vp->v_cv, NULL, CV_DEFAULT, NULL); 2282 rw_init(&vp->v_nbllock, NULL, RW_DEFAULT, NULL); 2283 vp->v_femhead = NULL; /* Must be done before vn_reinit() */ 2284 vp->v_path = NULL; 2285 vp->v_mpssdata = NULL; 2286 vp->v_vsd = NULL; 2287 vp->v_fopdata = NULL; 2288 2289 return (0); 2290 } 2291 2292 /* ARGSUSED */ 2293 static void 2294 vn_cache_destructor(void *buf, void *cdrarg) 2295 { 2296 struct vnode *vp; 2297 2298 vp = buf; 2299 2300 rw_destroy(&vp->v_nbllock); 2301 cv_destroy(&vp->v_cv); 2302 mutex_destroy(&vp->v_vsd_lock); 2303 mutex_destroy(&vp->v_lock); 2304 } 2305 2306 void 2307 vn_create_cache(void) 2308 { 2309 /* LINTED */ 2310 ASSERT((1 << VNODE_ALIGN_LOG2) == 2311 P2ROUNDUP(sizeof (struct vnode), VNODE_ALIGN)); 2312 vn_cache = kmem_cache_create("vn_cache", sizeof (struct vnode), 2313 VNODE_ALIGN, vn_cache_constructor, vn_cache_destructor, NULL, NULL, 2314 NULL, 0); 2315 } 2316 2317 void 2318 vn_destroy_cache(void) 2319 { 2320 kmem_cache_destroy(vn_cache); 2321 } 2322 2323 /* 2324 * Used by file systems when fs-specific nodes (e.g., ufs inodes) are 2325 * cached by the file system and vnodes remain associated. 2326 */ 2327 void 2328 vn_recycle(vnode_t *vp) 2329 { 2330 ASSERT(vp->v_pages == NULL); 2331 2332 /* 2333 * XXX - This really belongs in vn_reinit(), but we have some issues 2334 * with the counts. Best to have it here for clean initialization. 2335 */ 2336 vp->v_rdcnt = 0; 2337 vp->v_wrcnt = 0; 2338 vp->v_mmap_read = 0; 2339 vp->v_mmap_write = 0; 2340 2341 /* 2342 * If FEM was in use, make sure everything gets cleaned up 2343 * NOTE: vp->v_femhead is initialized to NULL in the vnode 2344 * constructor. 2345 */ 2346 if (vp->v_femhead) { 2347 /* XXX - There should be a free_femhead() that does all this */ 2348 ASSERT(vp->v_femhead->femh_list == NULL); 2349 mutex_destroy(&vp->v_femhead->femh_lock); 2350 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead))); 2351 vp->v_femhead = NULL; 2352 } 2353 if (vp->v_path) { 2354 kmem_free(vp->v_path, strlen(vp->v_path) + 1); 2355 vp->v_path = NULL; 2356 } 2357 2358 if (vp->v_fopdata != NULL) { 2359 free_fopdata(vp); 2360 } 2361 vp->v_mpssdata = NULL; 2362 vsd_free(vp); 2363 } 2364 2365 /* 2366 * Used to reset the vnode fields including those that are directly accessible 2367 * as well as those which require an accessor function. 2368 * 2369 * Does not initialize: 2370 * synchronization objects: v_lock, v_vsd_lock, v_nbllock, v_cv 2371 * v_data (since FS-nodes and vnodes point to each other and should 2372 * be updated simultaneously) 2373 * v_op (in case someone needs to make a VOP call on this object) 2374 */ 2375 void 2376 vn_reinit(vnode_t *vp) 2377 { 2378 vp->v_count = 1; 2379 vp->v_count_dnlc = 0; 2380 vp->v_vfsp = NULL; 2381 vp->v_stream = NULL; 2382 vp->v_vfsmountedhere = NULL; 2383 vp->v_flag = 0; 2384 vp->v_type = VNON; 2385 vp->v_rdev = NODEV; 2386 2387 vp->v_filocks = NULL; 2388 vp->v_shrlocks = NULL; 2389 vp->v_pages = NULL; 2390 2391 vp->v_locality = NULL; 2392 vp->v_xattrdir = NULL; 2393 2394 /* Handles v_femhead, v_path, and the r/w/map counts */ 2395 vn_recycle(vp); 2396 } 2397 2398 vnode_t * 2399 vn_alloc(int kmflag) 2400 { 2401 vnode_t *vp; 2402 2403 vp = kmem_cache_alloc(vn_cache, kmflag); 2404 2405 if (vp != NULL) { 2406 vp->v_femhead = NULL; /* Must be done before vn_reinit() */ 2407 vp->v_fopdata = NULL; 2408 vn_reinit(vp); 2409 } 2410 2411 return (vp); 2412 } 2413 2414 void 2415 vn_free(vnode_t *vp) 2416 { 2417 ASSERT(vp->v_shrlocks == NULL); 2418 ASSERT(vp->v_filocks == NULL); 2419 2420 /* 2421 * Some file systems call vn_free() with v_count of zero, 2422 * some with v_count of 1. In any case, the value should 2423 * never be anything else. 2424 */ 2425 ASSERT((vp->v_count == 0) || (vp->v_count == 1)); 2426 ASSERT(vp->v_count_dnlc == 0); 2427 if (vp->v_path != NULL) { 2428 kmem_free(vp->v_path, strlen(vp->v_path) + 1); 2429 vp->v_path = NULL; 2430 } 2431 2432 /* If FEM was in use, make sure everything gets cleaned up */ 2433 if (vp->v_femhead) { 2434 /* XXX - There should be a free_femhead() that does all this */ 2435 ASSERT(vp->v_femhead->femh_list == NULL); 2436 mutex_destroy(&vp->v_femhead->femh_lock); 2437 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead))); 2438 vp->v_femhead = NULL; 2439 } 2440 2441 if (vp->v_fopdata != NULL) { 2442 free_fopdata(vp); 2443 } 2444 vp->v_mpssdata = NULL; 2445 vsd_free(vp); 2446 kmem_cache_free(vn_cache, vp); 2447 } 2448 2449 /* 2450 * vnode status changes, should define better states than 1, 0. 2451 */ 2452 void 2453 vn_reclaim(vnode_t *vp) 2454 { 2455 vfs_t *vfsp = vp->v_vfsp; 2456 2457 if (vfsp == NULL || 2458 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) { 2459 return; 2460 } 2461 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_RECLAIMED); 2462 } 2463 2464 void 2465 vn_idle(vnode_t *vp) 2466 { 2467 vfs_t *vfsp = vp->v_vfsp; 2468 2469 if (vfsp == NULL || 2470 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) { 2471 return; 2472 } 2473 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_IDLED); 2474 } 2475 void 2476 vn_exists(vnode_t *vp) 2477 { 2478 vfs_t *vfsp = vp->v_vfsp; 2479 2480 if (vfsp == NULL || 2481 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) { 2482 return; 2483 } 2484 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_EXISTS); 2485 } 2486 2487 void 2488 vn_invalid(vnode_t *vp) 2489 { 2490 vfs_t *vfsp = vp->v_vfsp; 2491 2492 if (vfsp == NULL || 2493 vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) { 2494 return; 2495 } 2496 (void) VFS_VNSTATE(vfsp, vp, VNTRANS_DESTROYED); 2497 } 2498 2499 /* Vnode event notification */ 2500 2501 int 2502 vnevent_support(vnode_t *vp, caller_context_t *ct) 2503 { 2504 if (vp == NULL) 2505 return (EINVAL); 2506 2507 return (VOP_VNEVENT(vp, VE_SUPPORT, NULL, NULL, ct)); 2508 } 2509 2510 void 2511 vnevent_rename_src(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct) 2512 { 2513 if (vp == NULL || vp->v_femhead == NULL) { 2514 return; 2515 } 2516 (void) VOP_VNEVENT(vp, VE_RENAME_SRC, dvp, name, ct); 2517 } 2518 2519 void 2520 vnevent_rename_dest(vnode_t *vp, vnode_t *dvp, char *name, 2521 caller_context_t *ct) 2522 { 2523 if (vp == NULL || vp->v_femhead == NULL) { 2524 return; 2525 } 2526 (void) VOP_VNEVENT(vp, VE_RENAME_DEST, dvp, name, ct); 2527 } 2528 2529 void 2530 vnevent_rename_dest_dir(vnode_t *vp, caller_context_t *ct) 2531 { 2532 if (vp == NULL || vp->v_femhead == NULL) { 2533 return; 2534 } 2535 (void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, NULL, NULL, ct); 2536 } 2537 2538 void 2539 vnevent_remove(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct) 2540 { 2541 if (vp == NULL || vp->v_femhead == NULL) { 2542 return; 2543 } 2544 (void) VOP_VNEVENT(vp, VE_REMOVE, dvp, name, ct); 2545 } 2546 2547 void 2548 vnevent_rmdir(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct) 2549 { 2550 if (vp == NULL || vp->v_femhead == NULL) { 2551 return; 2552 } 2553 (void) VOP_VNEVENT(vp, VE_RMDIR, dvp, name, ct); 2554 } 2555 2556 void 2557 vnevent_pre_rename_src(vnode_t *vp, vnode_t *dvp, char *name, 2558 caller_context_t *ct) 2559 { 2560 if (vp == NULL || vp->v_femhead == NULL) { 2561 return; 2562 } 2563 (void) VOP_VNEVENT(vp, VE_PRE_RENAME_SRC, dvp, name, ct); 2564 } 2565 2566 void 2567 vnevent_pre_rename_dest(vnode_t *vp, vnode_t *dvp, char *name, 2568 caller_context_t *ct) 2569 { 2570 if (vp == NULL || vp->v_femhead == NULL) { 2571 return; 2572 } 2573 (void) VOP_VNEVENT(vp, VE_PRE_RENAME_DEST, dvp, name, ct); 2574 } 2575 2576 void 2577 vnevent_pre_rename_dest_dir(vnode_t *vp, vnode_t *nvp, char *name, 2578 caller_context_t *ct) 2579 { 2580 if (vp == NULL || vp->v_femhead == NULL) { 2581 return; 2582 } 2583 (void) VOP_VNEVENT(vp, VE_PRE_RENAME_DEST_DIR, nvp, name, ct); 2584 } 2585 2586 void 2587 vnevent_create(vnode_t *vp, caller_context_t *ct) 2588 { 2589 if (vp == NULL || vp->v_femhead == NULL) { 2590 return; 2591 } 2592 (void) VOP_VNEVENT(vp, VE_CREATE, NULL, NULL, ct); 2593 } 2594 2595 void 2596 vnevent_link(vnode_t *vp, caller_context_t *ct) 2597 { 2598 if (vp == NULL || vp->v_femhead == NULL) { 2599 return; 2600 } 2601 (void) VOP_VNEVENT(vp, VE_LINK, NULL, NULL, ct); 2602 } 2603 2604 void 2605 vnevent_mountedover(vnode_t *vp, caller_context_t *ct) 2606 { 2607 if (vp == NULL || vp->v_femhead == NULL) { 2608 return; 2609 } 2610 (void) VOP_VNEVENT(vp, VE_MOUNTEDOVER, NULL, NULL, ct); 2611 } 2612 2613 void 2614 vnevent_truncate(vnode_t *vp, caller_context_t *ct) 2615 { 2616 if (vp == NULL || vp->v_femhead == NULL) { 2617 return; 2618 } 2619 (void) VOP_VNEVENT(vp, VE_TRUNCATE, NULL, NULL, ct); 2620 } 2621 2622 /* 2623 * Vnode accessors. 2624 */ 2625 2626 int 2627 vn_is_readonly(vnode_t *vp) 2628 { 2629 return (vp->v_vfsp->vfs_flag & VFS_RDONLY); 2630 } 2631 2632 int 2633 vn_has_flocks(vnode_t *vp) 2634 { 2635 return (vp->v_filocks != NULL); 2636 } 2637 2638 int 2639 vn_has_mandatory_locks(vnode_t *vp, int mode) 2640 { 2641 return ((vp->v_filocks != NULL) && (MANDLOCK(vp, mode))); 2642 } 2643 2644 int 2645 vn_has_cached_data(vnode_t *vp) 2646 { 2647 return (vp->v_pages != NULL); 2648 } 2649 2650 /* 2651 * Return 0 if the vnode in question shouldn't be permitted into a zone via 2652 * zone_enter(2). 2653 */ 2654 int 2655 vn_can_change_zones(vnode_t *vp) 2656 { 2657 struct vfssw *vswp; 2658 int allow = 1; 2659 vnode_t *rvp; 2660 2661 if (nfs_global_client_only != 0) 2662 return (1); 2663 2664 /* 2665 * We always want to look at the underlying vnode if there is one. 2666 */ 2667 if (VOP_REALVP(vp, &rvp, NULL) != 0) 2668 rvp = vp; 2669 /* 2670 * Some pseudo filesystems (including doorfs) don't actually register 2671 * their vfsops_t, so the following may return NULL; we happily let 2672 * such vnodes switch zones. 2673 */ 2674 vswp = vfs_getvfsswbyvfsops(vfs_getops(rvp->v_vfsp)); 2675 if (vswp != NULL) { 2676 if (vswp->vsw_flag & VSW_NOTZONESAFE) 2677 allow = 0; 2678 vfs_unrefvfssw(vswp); 2679 } 2680 return (allow); 2681 } 2682 2683 /* 2684 * Return nonzero if the vnode is a mount point, zero if not. 2685 */ 2686 int 2687 vn_ismntpt(vnode_t *vp) 2688 { 2689 return (vp->v_vfsmountedhere != NULL); 2690 } 2691 2692 /* Retrieve the vfs (if any) mounted on this vnode */ 2693 vfs_t * 2694 vn_mountedvfs(vnode_t *vp) 2695 { 2696 return (vp->v_vfsmountedhere); 2697 } 2698 2699 /* 2700 * Return nonzero if the vnode is referenced by the dnlc, zero if not. 2701 */ 2702 int 2703 vn_in_dnlc(vnode_t *vp) 2704 { 2705 return (vp->v_count_dnlc > 0); 2706 } 2707 2708 /* 2709 * vn_has_other_opens() checks whether a particular file is opened by more than 2710 * just the caller and whether the open is for read and/or write. 2711 * This routine is for calling after the caller has already called VOP_OPEN() 2712 * and the caller wishes to know if they are the only one with it open for 2713 * the mode(s) specified. 2714 * 2715 * Vnode counts are only kept on regular files (v_type=VREG). 2716 */ 2717 int 2718 vn_has_other_opens( 2719 vnode_t *vp, 2720 v_mode_t mode) 2721 { 2722 2723 ASSERT(vp != NULL); 2724 2725 switch (mode) { 2726 case V_WRITE: 2727 if (vp->v_wrcnt > 1) 2728 return (V_TRUE); 2729 break; 2730 case V_RDORWR: 2731 if ((vp->v_rdcnt > 1) || (vp->v_wrcnt > 1)) 2732 return (V_TRUE); 2733 break; 2734 case V_RDANDWR: 2735 if ((vp->v_rdcnt > 1) && (vp->v_wrcnt > 1)) 2736 return (V_TRUE); 2737 break; 2738 case V_READ: 2739 if (vp->v_rdcnt > 1) 2740 return (V_TRUE); 2741 break; 2742 } 2743 2744 return (V_FALSE); 2745 } 2746 2747 /* 2748 * vn_is_opened() checks whether a particular file is opened and 2749 * whether the open is for read and/or write. 2750 * 2751 * Vnode counts are only kept on regular files (v_type=VREG). 2752 */ 2753 int 2754 vn_is_opened( 2755 vnode_t *vp, 2756 v_mode_t mode) 2757 { 2758 2759 ASSERT(vp != NULL); 2760 2761 switch (mode) { 2762 case V_WRITE: 2763 if (vp->v_wrcnt) 2764 return (V_TRUE); 2765 break; 2766 case V_RDANDWR: 2767 if (vp->v_rdcnt && vp->v_wrcnt) 2768 return (V_TRUE); 2769 break; 2770 case V_RDORWR: 2771 if (vp->v_rdcnt || vp->v_wrcnt) 2772 return (V_TRUE); 2773 break; 2774 case V_READ: 2775 if (vp->v_rdcnt) 2776 return (V_TRUE); 2777 break; 2778 } 2779 2780 return (V_FALSE); 2781 } 2782 2783 /* 2784 * vn_is_mapped() checks whether a particular file is mapped and whether 2785 * the file is mapped read and/or write. 2786 */ 2787 int 2788 vn_is_mapped( 2789 vnode_t *vp, 2790 v_mode_t mode) 2791 { 2792 2793 ASSERT(vp != NULL); 2794 2795 #if !defined(_LP64) 2796 switch (mode) { 2797 /* 2798 * The atomic_add_64_nv functions force atomicity in the 2799 * case of 32 bit architectures. Otherwise the 64 bit values 2800 * require two fetches. The value of the fields may be 2801 * (potentially) changed between the first fetch and the 2802 * second 2803 */ 2804 case V_WRITE: 2805 if (atomic_add_64_nv((&(vp->v_mmap_write)), 0)) 2806 return (V_TRUE); 2807 break; 2808 case V_RDANDWR: 2809 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) && 2810 (atomic_add_64_nv((&(vp->v_mmap_write)), 0))) 2811 return (V_TRUE); 2812 break; 2813 case V_RDORWR: 2814 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) || 2815 (atomic_add_64_nv((&(vp->v_mmap_write)), 0))) 2816 return (V_TRUE); 2817 break; 2818 case V_READ: 2819 if (atomic_add_64_nv((&(vp->v_mmap_read)), 0)) 2820 return (V_TRUE); 2821 break; 2822 } 2823 #else 2824 switch (mode) { 2825 case V_WRITE: 2826 if (vp->v_mmap_write) 2827 return (V_TRUE); 2828 break; 2829 case V_RDANDWR: 2830 if (vp->v_mmap_read && vp->v_mmap_write) 2831 return (V_TRUE); 2832 break; 2833 case V_RDORWR: 2834 if (vp->v_mmap_read || vp->v_mmap_write) 2835 return (V_TRUE); 2836 break; 2837 case V_READ: 2838 if (vp->v_mmap_read) 2839 return (V_TRUE); 2840 break; 2841 } 2842 #endif 2843 2844 return (V_FALSE); 2845 } 2846 2847 /* 2848 * Set the operations vector for a vnode. 2849 * 2850 * FEM ensures that the v_femhead pointer is filled in before the 2851 * v_op pointer is changed. This means that if the v_femhead pointer 2852 * is NULL, and the v_op field hasn't changed since before which checked 2853 * the v_femhead pointer; then our update is ok - we are not racing with 2854 * FEM. 2855 */ 2856 void 2857 vn_setops(vnode_t *vp, vnodeops_t *vnodeops) 2858 { 2859 vnodeops_t *op; 2860 2861 ASSERT(vp != NULL); 2862 ASSERT(vnodeops != NULL); 2863 2864 op = vp->v_op; 2865 membar_consumer(); 2866 /* 2867 * If vp->v_femhead == NULL, then we'll call atomic_cas_ptr() to do 2868 * the compare-and-swap on vp->v_op. If either fails, then FEM is 2869 * in effect on the vnode and we need to have FEM deal with it. 2870 */ 2871 if (vp->v_femhead != NULL || atomic_cas_ptr(&vp->v_op, op, vnodeops) != 2872 op) { 2873 fem_setvnops(vp, vnodeops); 2874 } 2875 } 2876 2877 /* 2878 * Retrieve the operations vector for a vnode 2879 * As with vn_setops(above); make sure we aren't racing with FEM. 2880 * FEM sets the v_op to a special, internal, vnodeops that wouldn't 2881 * make sense to the callers of this routine. 2882 */ 2883 vnodeops_t * 2884 vn_getops(vnode_t *vp) 2885 { 2886 vnodeops_t *op; 2887 2888 ASSERT(vp != NULL); 2889 2890 op = vp->v_op; 2891 membar_consumer(); 2892 if (vp->v_femhead == NULL && op == vp->v_op) { 2893 return (op); 2894 } else { 2895 return (fem_getvnops(vp)); 2896 } 2897 } 2898 2899 /* 2900 * Returns non-zero (1) if the vnodeops matches that of the vnode. 2901 * Returns zero (0) if not. 2902 */ 2903 int 2904 vn_matchops(vnode_t *vp, vnodeops_t *vnodeops) 2905 { 2906 return (vn_getops(vp) == vnodeops); 2907 } 2908 2909 /* 2910 * Returns non-zero (1) if the specified operation matches the 2911 * corresponding operation for that the vnode. 2912 * Returns zero (0) if not. 2913 */ 2914 2915 #define MATCHNAME(n1, n2) (((n1)[0] == (n2)[0]) && (strcmp((n1), (n2)) == 0)) 2916 2917 int 2918 vn_matchopval(vnode_t *vp, char *vopname, fs_generic_func_p funcp) 2919 { 2920 const fs_operation_trans_def_t *otdp; 2921 fs_generic_func_p *loc = NULL; 2922 vnodeops_t *vop = vn_getops(vp); 2923 2924 ASSERT(vopname != NULL); 2925 2926 for (otdp = vn_ops_table; otdp->name != NULL; otdp++) { 2927 if (MATCHNAME(otdp->name, vopname)) { 2928 loc = (fs_generic_func_p *) 2929 ((char *)(vop) + otdp->offset); 2930 break; 2931 } 2932 } 2933 2934 return ((loc != NULL) && (*loc == funcp)); 2935 } 2936 2937 /* 2938 * fs_new_caller_id() needs to return a unique ID on a given local system. 2939 * The IDs do not need to survive across reboots. These are primarily 2940 * used so that (FEM) monitors can detect particular callers (such as 2941 * the NFS server) to a given vnode/vfs operation. 2942 */ 2943 u_longlong_t 2944 fs_new_caller_id() 2945 { 2946 static uint64_t next_caller_id = 0LL; /* First call returns 1 */ 2947 2948 return ((u_longlong_t)atomic_inc_64_nv(&next_caller_id)); 2949 } 2950 2951 /* 2952 * Given a starting vnode and a path, updates the path in the target vnode in 2953 * a safe manner. If the vnode already has path information embedded, then the 2954 * cached path is left untouched. 2955 */ 2956 2957 size_t max_vnode_path = 4 * MAXPATHLEN; 2958 2959 void 2960 vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp, 2961 const char *path, size_t plen) 2962 { 2963 char *rpath; 2964 vnode_t *base; 2965 size_t rpathlen, rpathalloc; 2966 int doslash = 1; 2967 2968 if (*path == '/') { 2969 base = rootvp; 2970 path++; 2971 plen--; 2972 } else { 2973 base = startvp; 2974 } 2975 2976 /* 2977 * We cannot grab base->v_lock while we hold vp->v_lock because of 2978 * the potential for deadlock. 2979 */ 2980 mutex_enter(&base->v_lock); 2981 if (base->v_path == NULL) { 2982 mutex_exit(&base->v_lock); 2983 return; 2984 } 2985 2986 rpathlen = strlen(base->v_path); 2987 rpathalloc = rpathlen + plen + 1; 2988 /* Avoid adding a slash if there's already one there */ 2989 if (base->v_path[rpathlen-1] == '/') 2990 doslash = 0; 2991 else 2992 rpathalloc++; 2993 2994 /* 2995 * We don't want to call kmem_alloc(KM_SLEEP) with kernel locks held, 2996 * so we must do this dance. If, by chance, something changes the path, 2997 * just give up since there is no real harm. 2998 */ 2999 mutex_exit(&base->v_lock); 3000 3001 /* Paths should stay within reason */ 3002 if (rpathalloc > max_vnode_path) 3003 return; 3004 3005 rpath = kmem_alloc(rpathalloc, KM_SLEEP); 3006 3007 mutex_enter(&base->v_lock); 3008 if (base->v_path == NULL || strlen(base->v_path) != rpathlen) { 3009 mutex_exit(&base->v_lock); 3010 kmem_free(rpath, rpathalloc); 3011 return; 3012 } 3013 bcopy(base->v_path, rpath, rpathlen); 3014 mutex_exit(&base->v_lock); 3015 3016 if (doslash) 3017 rpath[rpathlen++] = '/'; 3018 bcopy(path, rpath + rpathlen, plen); 3019 rpath[rpathlen + plen] = '\0'; 3020 3021 mutex_enter(&vp->v_lock); 3022 if (vp->v_path != NULL) { 3023 mutex_exit(&vp->v_lock); 3024 kmem_free(rpath, rpathalloc); 3025 } else { 3026 vp->v_path = rpath; 3027 mutex_exit(&vp->v_lock); 3028 } 3029 } 3030 3031 /* 3032 * Sets the path to the vnode to be the given string, regardless of current 3033 * context. The string must be a complete path from rootdir. This is only used 3034 * by fsop_root() for setting the path based on the mountpoint. 3035 */ 3036 void 3037 vn_setpath_str(struct vnode *vp, const char *str, size_t len) 3038 { 3039 char *buf = kmem_alloc(len + 1, KM_SLEEP); 3040 3041 mutex_enter(&vp->v_lock); 3042 if (vp->v_path != NULL) { 3043 mutex_exit(&vp->v_lock); 3044 kmem_free(buf, len + 1); 3045 return; 3046 } 3047 3048 vp->v_path = buf; 3049 bcopy(str, vp->v_path, len); 3050 vp->v_path[len] = '\0'; 3051 3052 mutex_exit(&vp->v_lock); 3053 } 3054 3055 /* 3056 * Called from within filesystem's vop_rename() to handle renames once the 3057 * target vnode is available. 3058 */ 3059 void 3060 vn_renamepath(vnode_t *dvp, vnode_t *vp, const char *nm, size_t len) 3061 { 3062 char *tmp; 3063 3064 mutex_enter(&vp->v_lock); 3065 tmp = vp->v_path; 3066 vp->v_path = NULL; 3067 mutex_exit(&vp->v_lock); 3068 vn_setpath(rootdir, dvp, vp, nm, len); 3069 if (tmp != NULL) 3070 kmem_free(tmp, strlen(tmp) + 1); 3071 } 3072 3073 /* 3074 * Similar to vn_setpath_str(), this function sets the path of the destination 3075 * vnode to the be the same as the source vnode. 3076 */ 3077 void 3078 vn_copypath(struct vnode *src, struct vnode *dst) 3079 { 3080 char *buf; 3081 int alloc; 3082 3083 mutex_enter(&src->v_lock); 3084 if (src->v_path == NULL) { 3085 mutex_exit(&src->v_lock); 3086 return; 3087 } 3088 alloc = strlen(src->v_path) + 1; 3089 3090 /* avoid kmem_alloc() with lock held */ 3091 mutex_exit(&src->v_lock); 3092 buf = kmem_alloc(alloc, KM_SLEEP); 3093 mutex_enter(&src->v_lock); 3094 if (src->v_path == NULL || strlen(src->v_path) + 1 != alloc) { 3095 mutex_exit(&src->v_lock); 3096 kmem_free(buf, alloc); 3097 return; 3098 } 3099 bcopy(src->v_path, buf, alloc); 3100 mutex_exit(&src->v_lock); 3101 3102 mutex_enter(&dst->v_lock); 3103 if (dst->v_path != NULL) { 3104 mutex_exit(&dst->v_lock); 3105 kmem_free(buf, alloc); 3106 return; 3107 } 3108 dst->v_path = buf; 3109 mutex_exit(&dst->v_lock); 3110 } 3111 3112 /* 3113 * XXX Private interface for segvn routines that handle vnode 3114 * large page segments. 3115 * 3116 * return 1 if vp's file system VOP_PAGEIO() implementation 3117 * can be safely used instead of VOP_GETPAGE() for handling 3118 * pagefaults against regular non swap files. VOP_PAGEIO() 3119 * interface is considered safe here if its implementation 3120 * is very close to VOP_GETPAGE() implementation. 3121 * e.g. It zero's out the part of the page beyond EOF. Doesn't 3122 * panic if there're file holes but instead returns an error. 3123 * Doesn't assume file won't be changed by user writes, etc. 3124 * 3125 * return 0 otherwise. 3126 * 3127 * For now allow segvn to only use VOP_PAGEIO() with ufs and nfs. 3128 */ 3129 int 3130 vn_vmpss_usepageio(vnode_t *vp) 3131 { 3132 vfs_t *vfsp = vp->v_vfsp; 3133 char *fsname = vfssw[vfsp->vfs_fstype].vsw_name; 3134 char *pageio_ok_fss[] = {"ufs", "nfs", NULL}; 3135 char **fsok = pageio_ok_fss; 3136 3137 if (fsname == NULL) { 3138 return (0); 3139 } 3140 3141 for (; *fsok; fsok++) { 3142 if (strcmp(*fsok, fsname) == 0) { 3143 return (1); 3144 } 3145 } 3146 return (0); 3147 } 3148 3149 /* VOP_XXX() macros call the corresponding fop_xxx() function */ 3150 3151 int 3152 fop_open( 3153 vnode_t **vpp, 3154 int mode, 3155 cred_t *cr, 3156 caller_context_t *ct) 3157 { 3158 int ret; 3159 vnode_t *vp = *vpp; 3160 3161 VN_HOLD(vp); 3162 /* 3163 * Adding to the vnode counts before calling open 3164 * avoids the need for a mutex. It circumvents a race 3165 * condition where a query made on the vnode counts results in a 3166 * false negative. The inquirer goes away believing the file is 3167 * not open when there is an open on the file already under way. 3168 * 3169 * The counts are meant to prevent NFS from granting a delegation 3170 * when it would be dangerous to do so. 3171 * 3172 * The vnode counts are only kept on regular files 3173 */ 3174 if ((*vpp)->v_type == VREG) { 3175 if (mode & FREAD) 3176 atomic_inc_32(&(*vpp)->v_rdcnt); 3177 if (mode & FWRITE) 3178 atomic_inc_32(&(*vpp)->v_wrcnt); 3179 } 3180 3181 VOPXID_MAP_CR(vp, cr); 3182 3183 ret = (*(*(vpp))->v_op->vop_open)(vpp, mode, cr, ct); 3184 3185 if (ret) { 3186 /* 3187 * Use the saved vp just in case the vnode ptr got trashed 3188 * by the error. 3189 */ 3190 VOPSTATS_UPDATE(vp, open); 3191 if ((vp->v_type == VREG) && (mode & FREAD)) 3192 atomic_dec_32(&vp->v_rdcnt); 3193 if ((vp->v_type == VREG) && (mode & FWRITE)) 3194 atomic_dec_32(&vp->v_wrcnt); 3195 } else { 3196 /* 3197 * Some filesystems will return a different vnode, 3198 * but the same path was still used to open it. 3199 * So if we do change the vnode and need to 3200 * copy over the path, do so here, rather than special 3201 * casing each filesystem. Adjust the vnode counts to 3202 * reflect the vnode switch. 3203 */ 3204 VOPSTATS_UPDATE(*vpp, open); 3205 if (*vpp != vp && *vpp != NULL) { 3206 vn_copypath(vp, *vpp); 3207 if (((*vpp)->v_type == VREG) && (mode & FREAD)) 3208 atomic_inc_32(&(*vpp)->v_rdcnt); 3209 if ((vp->v_type == VREG) && (mode & FREAD)) 3210 atomic_dec_32(&vp->v_rdcnt); 3211 if (((*vpp)->v_type == VREG) && (mode & FWRITE)) 3212 atomic_inc_32(&(*vpp)->v_wrcnt); 3213 if ((vp->v_type == VREG) && (mode & FWRITE)) 3214 atomic_dec_32(&vp->v_wrcnt); 3215 } 3216 } 3217 VN_RELE(vp); 3218 return (ret); 3219 } 3220 3221 int 3222 fop_close( 3223 vnode_t *vp, 3224 int flag, 3225 int count, 3226 offset_t offset, 3227 cred_t *cr, 3228 caller_context_t *ct) 3229 { 3230 int err; 3231 3232 VOPXID_MAP_CR(vp, cr); 3233 3234 err = (*(vp)->v_op->vop_close)(vp, flag, count, offset, cr, ct); 3235 VOPSTATS_UPDATE(vp, close); 3236 /* 3237 * Check passed in count to handle possible dups. Vnode counts are only 3238 * kept on regular files 3239 */ 3240 if ((vp->v_type == VREG) && (count == 1)) { 3241 if (flag & FREAD) { 3242 ASSERT(vp->v_rdcnt > 0); 3243 atomic_dec_32(&vp->v_rdcnt); 3244 } 3245 if (flag & FWRITE) { 3246 ASSERT(vp->v_wrcnt > 0); 3247 atomic_dec_32(&vp->v_wrcnt); 3248 } 3249 } 3250 return (err); 3251 } 3252 3253 int 3254 fop_read( 3255 vnode_t *vp, 3256 uio_t *uiop, 3257 int ioflag, 3258 cred_t *cr, 3259 caller_context_t *ct) 3260 { 3261 int err; 3262 ssize_t resid_start = uiop->uio_resid; 3263 3264 VOPXID_MAP_CR(vp, cr); 3265 3266 err = (*(vp)->v_op->vop_read)(vp, uiop, ioflag, cr, ct); 3267 VOPSTATS_UPDATE_IO(vp, read, 3268 read_bytes, (resid_start - uiop->uio_resid)); 3269 return (err); 3270 } 3271 3272 int 3273 fop_write( 3274 vnode_t *vp, 3275 uio_t *uiop, 3276 int ioflag, 3277 cred_t *cr, 3278 caller_context_t *ct) 3279 { 3280 int err; 3281 ssize_t resid_start = uiop->uio_resid; 3282 3283 VOPXID_MAP_CR(vp, cr); 3284 3285 err = (*(vp)->v_op->vop_write)(vp, uiop, ioflag, cr, ct); 3286 VOPSTATS_UPDATE_IO(vp, write, 3287 write_bytes, (resid_start - uiop->uio_resid)); 3288 return (err); 3289 } 3290 3291 int 3292 fop_ioctl( 3293 vnode_t *vp, 3294 int cmd, 3295 intptr_t arg, 3296 int flag, 3297 cred_t *cr, 3298 int *rvalp, 3299 caller_context_t *ct) 3300 { 3301 int err; 3302 3303 VOPXID_MAP_CR(vp, cr); 3304 3305 err = (*(vp)->v_op->vop_ioctl)(vp, cmd, arg, flag, cr, rvalp, ct); 3306 VOPSTATS_UPDATE(vp, ioctl); 3307 return (err); 3308 } 3309 3310 int 3311 fop_setfl( 3312 vnode_t *vp, 3313 int oflags, 3314 int nflags, 3315 cred_t *cr, 3316 caller_context_t *ct) 3317 { 3318 int err; 3319 3320 VOPXID_MAP_CR(vp, cr); 3321 3322 err = (*(vp)->v_op->vop_setfl)(vp, oflags, nflags, cr, ct); 3323 VOPSTATS_UPDATE(vp, setfl); 3324 return (err); 3325 } 3326 3327 int 3328 fop_getattr( 3329 vnode_t *vp, 3330 vattr_t *vap, 3331 int flags, 3332 cred_t *cr, 3333 caller_context_t *ct) 3334 { 3335 int err; 3336 3337 VOPXID_MAP_CR(vp, cr); 3338 3339 /* 3340 * If this file system doesn't understand the xvattr extensions 3341 * then turn off the xvattr bit. 3342 */ 3343 if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) { 3344 vap->va_mask &= ~AT_XVATTR; 3345 } 3346 3347 /* 3348 * We're only allowed to skip the ACL check iff we used a 32 bit 3349 * ACE mask with VOP_ACCESS() to determine permissions. 3350 */ 3351 if ((flags & ATTR_NOACLCHECK) && 3352 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 3353 return (EINVAL); 3354 } 3355 err = (*(vp)->v_op->vop_getattr)(vp, vap, flags, cr, ct); 3356 VOPSTATS_UPDATE(vp, getattr); 3357 return (err); 3358 } 3359 3360 int 3361 fop_setattr( 3362 vnode_t *vp, 3363 vattr_t *vap, 3364 int flags, 3365 cred_t *cr, 3366 caller_context_t *ct) 3367 { 3368 int err; 3369 3370 VOPXID_MAP_CR(vp, cr); 3371 3372 /* 3373 * If this file system doesn't understand the xvattr extensions 3374 * then turn off the xvattr bit. 3375 */ 3376 if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) { 3377 vap->va_mask &= ~AT_XVATTR; 3378 } 3379 3380 /* 3381 * We're only allowed to skip the ACL check iff we used a 32 bit 3382 * ACE mask with VOP_ACCESS() to determine permissions. 3383 */ 3384 if ((flags & ATTR_NOACLCHECK) && 3385 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 3386 return (EINVAL); 3387 } 3388 err = (*(vp)->v_op->vop_setattr)(vp, vap, flags, cr, ct); 3389 VOPSTATS_UPDATE(vp, setattr); 3390 return (err); 3391 } 3392 3393 int 3394 fop_access( 3395 vnode_t *vp, 3396 int mode, 3397 int flags, 3398 cred_t *cr, 3399 caller_context_t *ct) 3400 { 3401 int err; 3402 3403 if ((flags & V_ACE_MASK) && 3404 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 3405 return (EINVAL); 3406 } 3407 3408 VOPXID_MAP_CR(vp, cr); 3409 3410 err = (*(vp)->v_op->vop_access)(vp, mode, flags, cr, ct); 3411 VOPSTATS_UPDATE(vp, access); 3412 return (err); 3413 } 3414 3415 int 3416 fop_lookup( 3417 vnode_t *dvp, 3418 char *nm, 3419 vnode_t **vpp, 3420 pathname_t *pnp, 3421 int flags, 3422 vnode_t *rdir, 3423 cred_t *cr, 3424 caller_context_t *ct, 3425 int *deflags, /* Returned per-dirent flags */ 3426 pathname_t *ppnp) /* Returned case-preserved name in directory */ 3427 { 3428 int ret; 3429 3430 /* 3431 * If this file system doesn't support case-insensitive access 3432 * and said access is requested, fail quickly. It is required 3433 * that if the vfs supports case-insensitive lookup, it also 3434 * supports extended dirent flags. 3435 */ 3436 if (flags & FIGNORECASE && 3437 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3438 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3439 return (EINVAL); 3440 3441 VOPXID_MAP_CR(dvp, cr); 3442 3443 if ((flags & LOOKUP_XATTR) && (flags & LOOKUP_HAVE_SYSATTR_DIR) == 0) { 3444 ret = xattr_dir_lookup(dvp, vpp, flags, cr); 3445 } else { 3446 ret = (*(dvp)->v_op->vop_lookup) 3447 (dvp, nm, vpp, pnp, flags, rdir, cr, ct, deflags, ppnp); 3448 } 3449 if (ret == 0 && *vpp) { 3450 VOPSTATS_UPDATE(*vpp, lookup); 3451 if ((*vpp)->v_path == NULL) { 3452 vn_setpath(rootdir, dvp, *vpp, nm, strlen(nm)); 3453 } 3454 } 3455 3456 return (ret); 3457 } 3458 3459 int 3460 fop_create( 3461 vnode_t *dvp, 3462 char *name, 3463 vattr_t *vap, 3464 vcexcl_t excl, 3465 int mode, 3466 vnode_t **vpp, 3467 cred_t *cr, 3468 int flags, 3469 caller_context_t *ct, 3470 vsecattr_t *vsecp) /* ACL to set during create */ 3471 { 3472 int ret; 3473 3474 if (vsecp != NULL && 3475 vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) { 3476 return (EINVAL); 3477 } 3478 /* 3479 * If this file system doesn't support case-insensitive access 3480 * and said access is requested, fail quickly. 3481 */ 3482 if (flags & FIGNORECASE && 3483 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3484 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3485 return (EINVAL); 3486 3487 VOPXID_MAP_CR(dvp, cr); 3488 3489 ret = (*(dvp)->v_op->vop_create) 3490 (dvp, name, vap, excl, mode, vpp, cr, flags, ct, vsecp); 3491 if (ret == 0 && *vpp) { 3492 VOPSTATS_UPDATE(*vpp, create); 3493 if ((*vpp)->v_path == NULL) { 3494 vn_setpath(rootdir, dvp, *vpp, name, strlen(name)); 3495 } 3496 } 3497 3498 return (ret); 3499 } 3500 3501 int 3502 fop_remove( 3503 vnode_t *dvp, 3504 char *nm, 3505 cred_t *cr, 3506 caller_context_t *ct, 3507 int flags) 3508 { 3509 int err; 3510 3511 /* 3512 * If this file system doesn't support case-insensitive access 3513 * and said access is requested, fail quickly. 3514 */ 3515 if (flags & FIGNORECASE && 3516 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3517 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3518 return (EINVAL); 3519 3520 VOPXID_MAP_CR(dvp, cr); 3521 3522 err = (*(dvp)->v_op->vop_remove)(dvp, nm, cr, ct, flags); 3523 VOPSTATS_UPDATE(dvp, remove); 3524 return (err); 3525 } 3526 3527 int 3528 fop_link( 3529 vnode_t *tdvp, 3530 vnode_t *svp, 3531 char *tnm, 3532 cred_t *cr, 3533 caller_context_t *ct, 3534 int flags) 3535 { 3536 int err; 3537 3538 /* 3539 * If the target file system doesn't support case-insensitive access 3540 * and said access is requested, fail quickly. 3541 */ 3542 if (flags & FIGNORECASE && 3543 (vfs_has_feature(tdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3544 vfs_has_feature(tdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3545 return (EINVAL); 3546 3547 VOPXID_MAP_CR(tdvp, cr); 3548 3549 err = (*(tdvp)->v_op->vop_link)(tdvp, svp, tnm, cr, ct, flags); 3550 VOPSTATS_UPDATE(tdvp, link); 3551 return (err); 3552 } 3553 3554 int 3555 fop_rename( 3556 vnode_t *sdvp, 3557 char *snm, 3558 vnode_t *tdvp, 3559 char *tnm, 3560 cred_t *cr, 3561 caller_context_t *ct, 3562 int flags) 3563 { 3564 int err; 3565 3566 /* 3567 * If the file system involved does not support 3568 * case-insensitive access and said access is requested, fail 3569 * quickly. 3570 */ 3571 if (flags & FIGNORECASE && 3572 ((vfs_has_feature(sdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3573 vfs_has_feature(sdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))) 3574 return (EINVAL); 3575 3576 VOPXID_MAP_CR(tdvp, cr); 3577 3578 err = (*(sdvp)->v_op->vop_rename)(sdvp, snm, tdvp, tnm, cr, ct, flags); 3579 VOPSTATS_UPDATE(sdvp, rename); 3580 return (err); 3581 } 3582 3583 int 3584 fop_mkdir( 3585 vnode_t *dvp, 3586 char *dirname, 3587 vattr_t *vap, 3588 vnode_t **vpp, 3589 cred_t *cr, 3590 caller_context_t *ct, 3591 int flags, 3592 vsecattr_t *vsecp) /* ACL to set during create */ 3593 { 3594 int ret; 3595 3596 if (vsecp != NULL && 3597 vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) { 3598 return (EINVAL); 3599 } 3600 /* 3601 * If this file system doesn't support case-insensitive access 3602 * and said access is requested, fail quickly. 3603 */ 3604 if (flags & FIGNORECASE && 3605 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3606 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3607 return (EINVAL); 3608 3609 VOPXID_MAP_CR(dvp, cr); 3610 3611 ret = (*(dvp)->v_op->vop_mkdir) 3612 (dvp, dirname, vap, vpp, cr, ct, flags, vsecp); 3613 if (ret == 0 && *vpp) { 3614 VOPSTATS_UPDATE(*vpp, mkdir); 3615 if ((*vpp)->v_path == NULL) { 3616 vn_setpath(rootdir, dvp, *vpp, dirname, 3617 strlen(dirname)); 3618 } 3619 } 3620 3621 return (ret); 3622 } 3623 3624 int 3625 fop_rmdir( 3626 vnode_t *dvp, 3627 char *nm, 3628 vnode_t *cdir, 3629 cred_t *cr, 3630 caller_context_t *ct, 3631 int flags) 3632 { 3633 int err; 3634 3635 /* 3636 * If this file system doesn't support case-insensitive access 3637 * and said access is requested, fail quickly. 3638 */ 3639 if (flags & FIGNORECASE && 3640 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3641 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3642 return (EINVAL); 3643 3644 VOPXID_MAP_CR(dvp, cr); 3645 3646 err = (*(dvp)->v_op->vop_rmdir)(dvp, nm, cdir, cr, ct, flags); 3647 VOPSTATS_UPDATE(dvp, rmdir); 3648 return (err); 3649 } 3650 3651 int 3652 fop_readdir( 3653 vnode_t *vp, 3654 uio_t *uiop, 3655 cred_t *cr, 3656 int *eofp, 3657 caller_context_t *ct, 3658 int flags) 3659 { 3660 int err; 3661 ssize_t resid_start = uiop->uio_resid; 3662 3663 /* 3664 * If this file system doesn't support retrieving directory 3665 * entry flags and said access is requested, fail quickly. 3666 */ 3667 if (flags & V_RDDIR_ENTFLAGS && 3668 vfs_has_feature(vp->v_vfsp, VFSFT_DIRENTFLAGS) == 0) 3669 return (EINVAL); 3670 3671 VOPXID_MAP_CR(vp, cr); 3672 3673 err = (*(vp)->v_op->vop_readdir)(vp, uiop, cr, eofp, ct, flags); 3674 VOPSTATS_UPDATE_IO(vp, readdir, 3675 readdir_bytes, (resid_start - uiop->uio_resid)); 3676 return (err); 3677 } 3678 3679 int 3680 fop_symlink( 3681 vnode_t *dvp, 3682 char *linkname, 3683 vattr_t *vap, 3684 char *target, 3685 cred_t *cr, 3686 caller_context_t *ct, 3687 int flags) 3688 { 3689 int err; 3690 xvattr_t xvattr; 3691 3692 /* 3693 * If this file system doesn't support case-insensitive access 3694 * and said access is requested, fail quickly. 3695 */ 3696 if (flags & FIGNORECASE && 3697 (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 && 3698 vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)) 3699 return (EINVAL); 3700 3701 VOPXID_MAP_CR(dvp, cr); 3702 3703 /* check for reparse point */ 3704 if ((vfs_has_feature(dvp->v_vfsp, VFSFT_REPARSE)) && 3705 (strncmp(target, FS_REPARSE_TAG_STR, 3706 strlen(FS_REPARSE_TAG_STR)) == 0)) { 3707 if (!fs_reparse_mark(target, vap, &xvattr)) 3708 vap = (vattr_t *)&xvattr; 3709 } 3710 3711 err = (*(dvp)->v_op->vop_symlink) 3712 (dvp, linkname, vap, target, cr, ct, flags); 3713 VOPSTATS_UPDATE(dvp, symlink); 3714 return (err); 3715 } 3716 3717 int 3718 fop_readlink( 3719 vnode_t *vp, 3720 uio_t *uiop, 3721 cred_t *cr, 3722 caller_context_t *ct) 3723 { 3724 int err; 3725 3726 VOPXID_MAP_CR(vp, cr); 3727 3728 err = (*(vp)->v_op->vop_readlink)(vp, uiop, cr, ct); 3729 VOPSTATS_UPDATE(vp, readlink); 3730 return (err); 3731 } 3732 3733 int 3734 fop_fsync( 3735 vnode_t *vp, 3736 int syncflag, 3737 cred_t *cr, 3738 caller_context_t *ct) 3739 { 3740 int err; 3741 3742 VOPXID_MAP_CR(vp, cr); 3743 3744 err = (*(vp)->v_op->vop_fsync)(vp, syncflag, cr, ct); 3745 VOPSTATS_UPDATE(vp, fsync); 3746 return (err); 3747 } 3748 3749 void 3750 fop_inactive( 3751 vnode_t *vp, 3752 cred_t *cr, 3753 caller_context_t *ct) 3754 { 3755 /* Need to update stats before vop call since we may lose the vnode */ 3756 VOPSTATS_UPDATE(vp, inactive); 3757 3758 VOPXID_MAP_CR(vp, cr); 3759 3760 (*(vp)->v_op->vop_inactive)(vp, cr, ct); 3761 } 3762 3763 int 3764 fop_fid( 3765 vnode_t *vp, 3766 fid_t *fidp, 3767 caller_context_t *ct) 3768 { 3769 int err; 3770 3771 err = (*(vp)->v_op->vop_fid)(vp, fidp, ct); 3772 VOPSTATS_UPDATE(vp, fid); 3773 return (err); 3774 } 3775 3776 int 3777 fop_rwlock( 3778 vnode_t *vp, 3779 int write_lock, 3780 caller_context_t *ct) 3781 { 3782 int ret; 3783 3784 ret = ((*(vp)->v_op->vop_rwlock)(vp, write_lock, ct)); 3785 VOPSTATS_UPDATE(vp, rwlock); 3786 return (ret); 3787 } 3788 3789 void 3790 fop_rwunlock( 3791 vnode_t *vp, 3792 int write_lock, 3793 caller_context_t *ct) 3794 { 3795 (*(vp)->v_op->vop_rwunlock)(vp, write_lock, ct); 3796 VOPSTATS_UPDATE(vp, rwunlock); 3797 } 3798 3799 int 3800 fop_seek( 3801 vnode_t *vp, 3802 offset_t ooff, 3803 offset_t *noffp, 3804 caller_context_t *ct) 3805 { 3806 int err; 3807 3808 err = (*(vp)->v_op->vop_seek)(vp, ooff, noffp, ct); 3809 VOPSTATS_UPDATE(vp, seek); 3810 return (err); 3811 } 3812 3813 int 3814 fop_cmp( 3815 vnode_t *vp1, 3816 vnode_t *vp2, 3817 caller_context_t *ct) 3818 { 3819 int err; 3820 3821 err = (*(vp1)->v_op->vop_cmp)(vp1, vp2, ct); 3822 VOPSTATS_UPDATE(vp1, cmp); 3823 return (err); 3824 } 3825 3826 int 3827 fop_frlock( 3828 vnode_t *vp, 3829 int cmd, 3830 flock64_t *bfp, 3831 int flag, 3832 offset_t offset, 3833 struct flk_callback *flk_cbp, 3834 cred_t *cr, 3835 caller_context_t *ct) 3836 { 3837 int err; 3838 3839 VOPXID_MAP_CR(vp, cr); 3840 3841 err = (*(vp)->v_op->vop_frlock) 3842 (vp, cmd, bfp, flag, offset, flk_cbp, cr, ct); 3843 VOPSTATS_UPDATE(vp, frlock); 3844 return (err); 3845 } 3846 3847 int 3848 fop_space( 3849 vnode_t *vp, 3850 int cmd, 3851 flock64_t *bfp, 3852 int flag, 3853 offset_t offset, 3854 cred_t *cr, 3855 caller_context_t *ct) 3856 { 3857 int err; 3858 3859 VOPXID_MAP_CR(vp, cr); 3860 3861 err = (*(vp)->v_op->vop_space)(vp, cmd, bfp, flag, offset, cr, ct); 3862 VOPSTATS_UPDATE(vp, space); 3863 return (err); 3864 } 3865 3866 int 3867 fop_realvp( 3868 vnode_t *vp, 3869 vnode_t **vpp, 3870 caller_context_t *ct) 3871 { 3872 int err; 3873 3874 err = (*(vp)->v_op->vop_realvp)(vp, vpp, ct); 3875 VOPSTATS_UPDATE(vp, realvp); 3876 return (err); 3877 } 3878 3879 int 3880 fop_getpage( 3881 vnode_t *vp, 3882 offset_t off, 3883 size_t len, 3884 uint_t *protp, 3885 page_t **plarr, 3886 size_t plsz, 3887 struct seg *seg, 3888 caddr_t addr, 3889 enum seg_rw rw, 3890 cred_t *cr, 3891 caller_context_t *ct) 3892 { 3893 int err; 3894 3895 VOPXID_MAP_CR(vp, cr); 3896 3897 err = (*(vp)->v_op->vop_getpage) 3898 (vp, off, len, protp, plarr, plsz, seg, addr, rw, cr, ct); 3899 VOPSTATS_UPDATE(vp, getpage); 3900 return (err); 3901 } 3902 3903 int 3904 fop_putpage( 3905 vnode_t *vp, 3906 offset_t off, 3907 size_t len, 3908 int flags, 3909 cred_t *cr, 3910 caller_context_t *ct) 3911 { 3912 int err; 3913 3914 VOPXID_MAP_CR(vp, cr); 3915 3916 err = (*(vp)->v_op->vop_putpage)(vp, off, len, flags, cr, ct); 3917 VOPSTATS_UPDATE(vp, putpage); 3918 return (err); 3919 } 3920 3921 int 3922 fop_map( 3923 vnode_t *vp, 3924 offset_t off, 3925 struct as *as, 3926 caddr_t *addrp, 3927 size_t len, 3928 uchar_t prot, 3929 uchar_t maxprot, 3930 uint_t flags, 3931 cred_t *cr, 3932 caller_context_t *ct) 3933 { 3934 int err; 3935 3936 VOPXID_MAP_CR(vp, cr); 3937 3938 err = (*(vp)->v_op->vop_map) 3939 (vp, off, as, addrp, len, prot, maxprot, flags, cr, ct); 3940 VOPSTATS_UPDATE(vp, map); 3941 return (err); 3942 } 3943 3944 int 3945 fop_addmap( 3946 vnode_t *vp, 3947 offset_t off, 3948 struct as *as, 3949 caddr_t addr, 3950 size_t len, 3951 uchar_t prot, 3952 uchar_t maxprot, 3953 uint_t flags, 3954 cred_t *cr, 3955 caller_context_t *ct) 3956 { 3957 int error; 3958 u_longlong_t delta; 3959 3960 VOPXID_MAP_CR(vp, cr); 3961 3962 error = (*(vp)->v_op->vop_addmap) 3963 (vp, off, as, addr, len, prot, maxprot, flags, cr, ct); 3964 3965 if ((!error) && (vp->v_type == VREG)) { 3966 delta = (u_longlong_t)btopr(len); 3967 /* 3968 * If file is declared MAP_PRIVATE, it can't be written back 3969 * even if open for write. Handle as read. 3970 */ 3971 if (flags & MAP_PRIVATE) { 3972 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3973 (int64_t)delta); 3974 } else { 3975 /* 3976 * atomic_add_64 forces the fetch of a 64 bit value to 3977 * be atomic on 32 bit machines 3978 */ 3979 if (maxprot & PROT_WRITE) 3980 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)), 3981 (int64_t)delta); 3982 if (maxprot & PROT_READ) 3983 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3984 (int64_t)delta); 3985 if (maxprot & PROT_EXEC) 3986 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 3987 (int64_t)delta); 3988 } 3989 } 3990 VOPSTATS_UPDATE(vp, addmap); 3991 return (error); 3992 } 3993 3994 int 3995 fop_delmap( 3996 vnode_t *vp, 3997 offset_t off, 3998 struct as *as, 3999 caddr_t addr, 4000 size_t len, 4001 uint_t prot, 4002 uint_t maxprot, 4003 uint_t flags, 4004 cred_t *cr, 4005 caller_context_t *ct) 4006 { 4007 int error; 4008 u_longlong_t delta; 4009 4010 VOPXID_MAP_CR(vp, cr); 4011 4012 error = (*(vp)->v_op->vop_delmap) 4013 (vp, off, as, addr, len, prot, maxprot, flags, cr, ct); 4014 4015 /* 4016 * NFS calls into delmap twice, the first time 4017 * it simply establishes a callback mechanism and returns EAGAIN 4018 * while the real work is being done upon the second invocation. 4019 * We have to detect this here and only decrement the counts upon 4020 * the second delmap request. 4021 */ 4022 if ((error != EAGAIN) && (vp->v_type == VREG)) { 4023 4024 delta = (u_longlong_t)btopr(len); 4025 4026 if (flags & MAP_PRIVATE) { 4027 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 4028 (int64_t)(-delta)); 4029 } else { 4030 /* 4031 * atomic_add_64 forces the fetch of a 64 bit value 4032 * to be atomic on 32 bit machines 4033 */ 4034 if (maxprot & PROT_WRITE) 4035 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)), 4036 (int64_t)(-delta)); 4037 if (maxprot & PROT_READ) 4038 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 4039 (int64_t)(-delta)); 4040 if (maxprot & PROT_EXEC) 4041 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)), 4042 (int64_t)(-delta)); 4043 } 4044 } 4045 VOPSTATS_UPDATE(vp, delmap); 4046 return (error); 4047 } 4048 4049 4050 int 4051 fop_poll( 4052 vnode_t *vp, 4053 short events, 4054 int anyyet, 4055 short *reventsp, 4056 struct pollhead **phpp, 4057 caller_context_t *ct) 4058 { 4059 int err; 4060 4061 err = (*(vp)->v_op->vop_poll)(vp, events, anyyet, reventsp, phpp, ct); 4062 VOPSTATS_UPDATE(vp, poll); 4063 return (err); 4064 } 4065 4066 int 4067 fop_dump( 4068 vnode_t *vp, 4069 caddr_t addr, 4070 offset_t lbdn, 4071 offset_t dblks, 4072 caller_context_t *ct) 4073 { 4074 int err; 4075 4076 /* ensure lbdn and dblks can be passed safely to bdev_dump */ 4077 if ((lbdn != (daddr_t)lbdn) || (dblks != (int)dblks)) 4078 return (EIO); 4079 4080 err = (*(vp)->v_op->vop_dump)(vp, addr, lbdn, dblks, ct); 4081 VOPSTATS_UPDATE(vp, dump); 4082 return (err); 4083 } 4084 4085 int 4086 fop_pathconf( 4087 vnode_t *vp, 4088 int cmd, 4089 ulong_t *valp, 4090 cred_t *cr, 4091 caller_context_t *ct) 4092 { 4093 int err; 4094 4095 VOPXID_MAP_CR(vp, cr); 4096 4097 err = (*(vp)->v_op->vop_pathconf)(vp, cmd, valp, cr, ct); 4098 VOPSTATS_UPDATE(vp, pathconf); 4099 return (err); 4100 } 4101 4102 int 4103 fop_pageio( 4104 vnode_t *vp, 4105 struct page *pp, 4106 u_offset_t io_off, 4107 size_t io_len, 4108 int flags, 4109 cred_t *cr, 4110 caller_context_t *ct) 4111 { 4112 int err; 4113 4114 VOPXID_MAP_CR(vp, cr); 4115 4116 err = (*(vp)->v_op->vop_pageio)(vp, pp, io_off, io_len, flags, cr, ct); 4117 VOPSTATS_UPDATE(vp, pageio); 4118 return (err); 4119 } 4120 4121 int 4122 fop_dumpctl( 4123 vnode_t *vp, 4124 int action, 4125 offset_t *blkp, 4126 caller_context_t *ct) 4127 { 4128 int err; 4129 err = (*(vp)->v_op->vop_dumpctl)(vp, action, blkp, ct); 4130 VOPSTATS_UPDATE(vp, dumpctl); 4131 return (err); 4132 } 4133 4134 void 4135 fop_dispose( 4136 vnode_t *vp, 4137 page_t *pp, 4138 int flag, 4139 int dn, 4140 cred_t *cr, 4141 caller_context_t *ct) 4142 { 4143 /* Must do stats first since it's possible to lose the vnode */ 4144 VOPSTATS_UPDATE(vp, dispose); 4145 4146 VOPXID_MAP_CR(vp, cr); 4147 4148 (*(vp)->v_op->vop_dispose)(vp, pp, flag, dn, cr, ct); 4149 } 4150 4151 int 4152 fop_setsecattr( 4153 vnode_t *vp, 4154 vsecattr_t *vsap, 4155 int flag, 4156 cred_t *cr, 4157 caller_context_t *ct) 4158 { 4159 int err; 4160 4161 VOPXID_MAP_CR(vp, cr); 4162 4163 /* 4164 * We're only allowed to skip the ACL check iff we used a 32 bit 4165 * ACE mask with VOP_ACCESS() to determine permissions. 4166 */ 4167 if ((flag & ATTR_NOACLCHECK) && 4168 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 4169 return (EINVAL); 4170 } 4171 err = (*(vp)->v_op->vop_setsecattr) (vp, vsap, flag, cr, ct); 4172 VOPSTATS_UPDATE(vp, setsecattr); 4173 return (err); 4174 } 4175 4176 int 4177 fop_getsecattr( 4178 vnode_t *vp, 4179 vsecattr_t *vsap, 4180 int flag, 4181 cred_t *cr, 4182 caller_context_t *ct) 4183 { 4184 int err; 4185 4186 /* 4187 * We're only allowed to skip the ACL check iff we used a 32 bit 4188 * ACE mask with VOP_ACCESS() to determine permissions. 4189 */ 4190 if ((flag & ATTR_NOACLCHECK) && 4191 vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) { 4192 return (EINVAL); 4193 } 4194 4195 VOPXID_MAP_CR(vp, cr); 4196 4197 err = (*(vp)->v_op->vop_getsecattr) (vp, vsap, flag, cr, ct); 4198 VOPSTATS_UPDATE(vp, getsecattr); 4199 return (err); 4200 } 4201 4202 int 4203 fop_shrlock( 4204 vnode_t *vp, 4205 int cmd, 4206 struct shrlock *shr, 4207 int flag, 4208 cred_t *cr, 4209 caller_context_t *ct) 4210 { 4211 int err; 4212 4213 VOPXID_MAP_CR(vp, cr); 4214 4215 err = (*(vp)->v_op->vop_shrlock)(vp, cmd, shr, flag, cr, ct); 4216 VOPSTATS_UPDATE(vp, shrlock); 4217 return (err); 4218 } 4219 4220 int 4221 fop_vnevent(vnode_t *vp, vnevent_t vnevent, vnode_t *dvp, char *fnm, 4222 caller_context_t *ct) 4223 { 4224 int err; 4225 4226 err = (*(vp)->v_op->vop_vnevent)(vp, vnevent, dvp, fnm, ct); 4227 VOPSTATS_UPDATE(vp, vnevent); 4228 return (err); 4229 } 4230 4231 int 4232 fop_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *uiop, cred_t *cr, 4233 caller_context_t *ct) 4234 { 4235 int err; 4236 4237 if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0) 4238 return (ENOTSUP); 4239 err = (*(vp)->v_op->vop_reqzcbuf)(vp, ioflag, uiop, cr, ct); 4240 VOPSTATS_UPDATE(vp, reqzcbuf); 4241 return (err); 4242 } 4243 4244 int 4245 fop_retzcbuf(vnode_t *vp, xuio_t *uiop, cred_t *cr, caller_context_t *ct) 4246 { 4247 int err; 4248 4249 if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0) 4250 return (ENOTSUP); 4251 err = (*(vp)->v_op->vop_retzcbuf)(vp, uiop, cr, ct); 4252 VOPSTATS_UPDATE(vp, retzcbuf); 4253 return (err); 4254 } 4255 4256 /* 4257 * Default destructor 4258 * Needed because NULL destructor means that the key is unused 4259 */ 4260 /* ARGSUSED */ 4261 void 4262 vsd_defaultdestructor(void *value) 4263 {} 4264 4265 /* 4266 * Create a key (index into per vnode array) 4267 * Locks out vsd_create, vsd_destroy, and vsd_free 4268 * May allocate memory with lock held 4269 */ 4270 void 4271 vsd_create(uint_t *keyp, void (*destructor)(void *)) 4272 { 4273 int i; 4274 uint_t nkeys; 4275 4276 /* 4277 * if key is allocated, do nothing 4278 */ 4279 mutex_enter(&vsd_lock); 4280 if (*keyp) { 4281 mutex_exit(&vsd_lock); 4282 return; 4283 } 4284 /* 4285 * find an unused key 4286 */ 4287 if (destructor == NULL) 4288 destructor = vsd_defaultdestructor; 4289 4290 for (i = 0; i < vsd_nkeys; ++i) 4291 if (vsd_destructor[i] == NULL) 4292 break; 4293 4294 /* 4295 * if no unused keys, increase the size of the destructor array 4296 */ 4297 if (i == vsd_nkeys) { 4298 if ((nkeys = (vsd_nkeys << 1)) == 0) 4299 nkeys = 1; 4300 vsd_destructor = 4301 (void (**)(void *))vsd_realloc((void *)vsd_destructor, 4302 (size_t)(vsd_nkeys * sizeof (void (*)(void *))), 4303 (size_t)(nkeys * sizeof (void (*)(void *)))); 4304 vsd_nkeys = nkeys; 4305 } 4306 4307 /* 4308 * allocate the next available unused key 4309 */ 4310 vsd_destructor[i] = destructor; 4311 *keyp = i + 1; 4312 4313 /* create vsd_list, if it doesn't exist */ 4314 if (vsd_list == NULL) { 4315 vsd_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 4316 list_create(vsd_list, sizeof (struct vsd_node), 4317 offsetof(struct vsd_node, vs_nodes)); 4318 } 4319 4320 mutex_exit(&vsd_lock); 4321 } 4322 4323 /* 4324 * Destroy a key 4325 * 4326 * Assumes that the caller is preventing vsd_set and vsd_get 4327 * Locks out vsd_create, vsd_destroy, and vsd_free 4328 * May free memory with lock held 4329 */ 4330 void 4331 vsd_destroy(uint_t *keyp) 4332 { 4333 uint_t key; 4334 struct vsd_node *vsd; 4335 4336 /* 4337 * protect the key namespace and our destructor lists 4338 */ 4339 mutex_enter(&vsd_lock); 4340 key = *keyp; 4341 *keyp = 0; 4342 4343 ASSERT(key <= vsd_nkeys); 4344 4345 /* 4346 * if the key is valid 4347 */ 4348 if (key != 0) { 4349 uint_t k = key - 1; 4350 /* 4351 * for every vnode with VSD, call key's destructor 4352 */ 4353 for (vsd = list_head(vsd_list); vsd != NULL; 4354 vsd = list_next(vsd_list, vsd)) { 4355 /* 4356 * no VSD for key in this vnode 4357 */ 4358 if (key > vsd->vs_nkeys) 4359 continue; 4360 /* 4361 * call destructor for key 4362 */ 4363 if (vsd->vs_value[k] && vsd_destructor[k]) 4364 (*vsd_destructor[k])(vsd->vs_value[k]); 4365 /* 4366 * reset value for key 4367 */ 4368 vsd->vs_value[k] = NULL; 4369 } 4370 /* 4371 * actually free the key (NULL destructor == unused) 4372 */ 4373 vsd_destructor[k] = NULL; 4374 } 4375 4376 mutex_exit(&vsd_lock); 4377 } 4378 4379 /* 4380 * Quickly return the per vnode value that was stored with the specified key 4381 * Assumes the caller is protecting key from vsd_create and vsd_destroy 4382 * Assumes the caller is holding v_vsd_lock to protect the vsd. 4383 */ 4384 void * 4385 vsd_get(vnode_t *vp, uint_t key) 4386 { 4387 struct vsd_node *vsd; 4388 4389 ASSERT(vp != NULL); 4390 ASSERT(mutex_owned(&vp->v_vsd_lock)); 4391 4392 vsd = vp->v_vsd; 4393 4394 if (key && vsd != NULL && key <= vsd->vs_nkeys) 4395 return (vsd->vs_value[key - 1]); 4396 return (NULL); 4397 } 4398 4399 /* 4400 * Set a per vnode value indexed with the specified key 4401 * Assumes the caller is holding v_vsd_lock to protect the vsd. 4402 */ 4403 int 4404 vsd_set(vnode_t *vp, uint_t key, void *value) 4405 { 4406 struct vsd_node *vsd; 4407 4408 ASSERT(vp != NULL); 4409 ASSERT(mutex_owned(&vp->v_vsd_lock)); 4410 4411 if (key == 0) 4412 return (EINVAL); 4413 4414 vsd = vp->v_vsd; 4415 if (vsd == NULL) 4416 vsd = vp->v_vsd = kmem_zalloc(sizeof (*vsd), KM_SLEEP); 4417 4418 /* 4419 * If the vsd was just allocated, vs_nkeys will be 0, so the following 4420 * code won't happen and we will continue down and allocate space for 4421 * the vs_value array. 4422 * If the caller is replacing one value with another, then it is up 4423 * to the caller to free/rele/destroy the previous value (if needed). 4424 */ 4425 if (key <= vsd->vs_nkeys) { 4426 vsd->vs_value[key - 1] = value; 4427 return (0); 4428 } 4429 4430 ASSERT(key <= vsd_nkeys); 4431 4432 if (vsd->vs_nkeys == 0) { 4433 mutex_enter(&vsd_lock); /* lock out vsd_destroy() */ 4434 /* 4435 * Link onto list of all VSD nodes. 4436 */ 4437 list_insert_head(vsd_list, vsd); 4438 mutex_exit(&vsd_lock); 4439 } 4440 4441 /* 4442 * Allocate vnode local storage and set the value for key 4443 */ 4444 vsd->vs_value = vsd_realloc(vsd->vs_value, 4445 vsd->vs_nkeys * sizeof (void *), 4446 key * sizeof (void *)); 4447 vsd->vs_nkeys = key; 4448 vsd->vs_value[key - 1] = value; 4449 4450 return (0); 4451 } 4452 4453 /* 4454 * Called from vn_free() to run the destructor function for each vsd 4455 * Locks out vsd_create and vsd_destroy 4456 * Assumes that the destructor *DOES NOT* use vsd 4457 */ 4458 void 4459 vsd_free(vnode_t *vp) 4460 { 4461 int i; 4462 struct vsd_node *vsd = vp->v_vsd; 4463 4464 if (vsd == NULL) 4465 return; 4466 4467 if (vsd->vs_nkeys == 0) { 4468 kmem_free(vsd, sizeof (*vsd)); 4469 vp->v_vsd = NULL; 4470 return; 4471 } 4472 4473 /* 4474 * lock out vsd_create and vsd_destroy, call 4475 * the destructor, and mark the value as destroyed. 4476 */ 4477 mutex_enter(&vsd_lock); 4478 4479 for (i = 0; i < vsd->vs_nkeys; i++) { 4480 if (vsd->vs_value[i] && vsd_destructor[i]) 4481 (*vsd_destructor[i])(vsd->vs_value[i]); 4482 vsd->vs_value[i] = NULL; 4483 } 4484 4485 /* 4486 * remove from linked list of VSD nodes 4487 */ 4488 list_remove(vsd_list, vsd); 4489 4490 mutex_exit(&vsd_lock); 4491 4492 /* 4493 * free up the VSD 4494 */ 4495 kmem_free(vsd->vs_value, vsd->vs_nkeys * sizeof (void *)); 4496 kmem_free(vsd, sizeof (struct vsd_node)); 4497 vp->v_vsd = NULL; 4498 } 4499 4500 /* 4501 * realloc 4502 */ 4503 static void * 4504 vsd_realloc(void *old, size_t osize, size_t nsize) 4505 { 4506 void *new; 4507 4508 new = kmem_zalloc(nsize, KM_SLEEP); 4509 if (old) { 4510 bcopy(old, new, osize); 4511 kmem_free(old, osize); 4512 } 4513 return (new); 4514 } 4515 4516 /* 4517 * Setup the extensible system attribute for creating a reparse point. 4518 * The symlink data 'target' is validated for proper format of a reparse 4519 * string and a check also made to make sure the symlink data does not 4520 * point to an existing file. 4521 * 4522 * return 0 if ok else -1. 4523 */ 4524 static int 4525 fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr) 4526 { 4527 xoptattr_t *xoap; 4528 4529 if ((!target) || (!vap) || (!xvattr)) 4530 return (-1); 4531 4532 /* validate reparse string */ 4533 if (reparse_validate((const char *)target)) 4534 return (-1); 4535 4536 xva_init(xvattr); 4537 xvattr->xva_vattr = *vap; 4538 xvattr->xva_vattr.va_mask |= AT_XVATTR; 4539 xoap = xva_getxoptattr(xvattr); 4540 ASSERT(xoap); 4541 XVA_SET_REQ(xvattr, XAT_REPARSE); 4542 xoap->xoa_reparse = 1; 4543 4544 return (0); 4545 } 4546 4547 /* 4548 * Function to check whether a symlink is a reparse point. 4549 * Return B_TRUE if it is a reparse point, else return B_FALSE 4550 */ 4551 boolean_t 4552 vn_is_reparse(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4553 { 4554 xvattr_t xvattr; 4555 xoptattr_t *xoap; 4556 4557 if ((vp->v_type != VLNK) || 4558 !(vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR))) 4559 return (B_FALSE); 4560 4561 xva_init(&xvattr); 4562 xoap = xva_getxoptattr(&xvattr); 4563 ASSERT(xoap); 4564 XVA_SET_REQ(&xvattr, XAT_REPARSE); 4565 4566 if (VOP_GETATTR(vp, &xvattr.xva_vattr, 0, cr, ct)) 4567 return (B_FALSE); 4568 4569 if ((!(xvattr.xva_vattr.va_mask & AT_XVATTR)) || 4570 (!(XVA_ISSET_RTN(&xvattr, XAT_REPARSE)))) 4571 return (B_FALSE); 4572 4573 return (xoap->xoa_reparse ? B_TRUE : B_FALSE); 4574 } 4575