1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 35 */ 36 37 #include <sys/cdefs.h> 38 /* 39 * vnode op calls for Sun NFS version 2, 3 and 4 40 */ 41 42 #include "opt_inet.h" 43 44 #include <sys/param.h> 45 #include <sys/kernel.h> 46 #include <sys/systm.h> 47 #include <sys/resourcevar.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/bio.h> 51 #include <sys/buf.h> 52 #include <sys/extattr.h> 53 #include <sys/filio.h> 54 #include <sys/jail.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/socket.h> 59 #include <sys/vnode.h> 60 #include <sys/dirent.h> 61 #include <sys/fcntl.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/sysctl.h> 65 #include <sys/signalvar.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_object.h> 70 #include <vm/vnode_pager.h> 71 72 #include <fs/nfs/nfsport.h> 73 #include <fs/nfsclient/nfsnode.h> 74 #include <fs/nfsclient/nfsmount.h> 75 #include <fs/nfsclient/nfs.h> 76 #include <fs/nfsclient/nfs_kdtrace.h> 77 78 #include <net/if.h> 79 #include <netinet/in.h> 80 #include <netinet/in_var.h> 81 82 #include <nfs/nfs_lock.h> 83 84 #ifdef KDTRACE_HOOKS 85 #include <sys/dtrace_bsd.h> 86 87 dtrace_nfsclient_accesscache_flush_probe_func_t 88 dtrace_nfscl_accesscache_flush_done_probe; 89 uint32_t nfscl_accesscache_flush_done_id; 90 91 dtrace_nfsclient_accesscache_get_probe_func_t 92 dtrace_nfscl_accesscache_get_hit_probe, 93 dtrace_nfscl_accesscache_get_miss_probe; 94 uint32_t nfscl_accesscache_get_hit_id; 95 uint32_t nfscl_accesscache_get_miss_id; 96 97 dtrace_nfsclient_accesscache_load_probe_func_t 98 dtrace_nfscl_accesscache_load_done_probe; 99 uint32_t nfscl_accesscache_load_done_id; 100 #endif /* !KDTRACE_HOOKS */ 101 102 /* Defs */ 103 #define TRUE 1 104 #define FALSE 0 105 106 extern struct nfsstatsv1 nfsstatsv1; 107 extern int nfsrv_useacl; 108 extern int nfscl_debuglevel; 109 NFSCLSTATEMUTEX; 110 MALLOC_DECLARE(M_NEWNFSREQ); 111 112 static vop_read_t nfsfifo_read; 113 static vop_write_t nfsfifo_write; 114 static vop_close_t nfsfifo_close; 115 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 116 struct thread *); 117 static int nfs_get_namedattrdir(struct vnode *, struct componentname *, 118 struct vnode **); 119 static vop_lookup_t nfs_lookup; 120 static vop_create_t nfs_create; 121 static vop_mknod_t nfs_mknod; 122 static vop_open_t nfs_open; 123 static vop_pathconf_t nfs_pathconf; 124 static vop_close_t nfs_close; 125 static vop_access_t nfs_access; 126 static vop_getattr_t nfs_getattr; 127 static vop_setattr_t nfs_setattr; 128 static vop_read_t nfs_read; 129 static vop_fsync_t nfs_fsync; 130 static vop_remove_t nfs_remove; 131 static vop_link_t nfs_link; 132 static vop_rename_t nfs_rename; 133 static vop_mkdir_t nfs_mkdir; 134 static vop_rmdir_t nfs_rmdir; 135 static vop_symlink_t nfs_symlink; 136 static vop_readdir_t nfs_readdir; 137 static vop_strategy_t nfs_strategy; 138 static int nfs_lookitup(struct vnode *, char *, int, 139 struct ucred *, struct thread *, struct nfsnode **); 140 static int nfs_sillyrename(struct vnode *, struct vnode *, 141 struct componentname *); 142 static vop_access_t nfsspec_access; 143 static vop_readlink_t nfs_readlink; 144 static vop_print_t nfs_print; 145 static vop_advlock_t nfs_advlock; 146 static vop_advlockasync_t nfs_advlockasync; 147 static vop_getacl_t nfs_getacl; 148 static vop_setacl_t nfs_setacl; 149 static vop_advise_t nfs_advise; 150 static vop_allocate_t nfs_allocate; 151 static vop_deallocate_t nfs_deallocate; 152 static vop_copy_file_range_t nfs_copy_file_range; 153 static vop_ioctl_t nfs_ioctl; 154 static vop_getextattr_t nfs_getextattr; 155 static vop_setextattr_t nfs_setextattr; 156 static vop_listextattr_t nfs_listextattr; 157 static vop_deleteextattr_t nfs_deleteextattr; 158 static vop_lock1_t nfs_lock; 159 160 /* 161 * Global vfs data structures for nfs 162 */ 163 164 static struct vop_vector newnfs_vnodeops_nosig = { 165 .vop_default = &default_vnodeops, 166 .vop_access = nfs_access, 167 .vop_advlock = nfs_advlock, 168 .vop_advlockasync = nfs_advlockasync, 169 .vop_close = nfs_close, 170 .vop_create = nfs_create, 171 .vop_fsync = nfs_fsync, 172 .vop_getattr = nfs_getattr, 173 .vop_getpages = ncl_getpages, 174 .vop_putpages = ncl_putpages, 175 .vop_inactive = ncl_inactive, 176 .vop_link = nfs_link, 177 .vop_lock1 = nfs_lock, 178 .vop_lookup = nfs_lookup, 179 .vop_mkdir = nfs_mkdir, 180 .vop_mknod = nfs_mknod, 181 .vop_open = nfs_open, 182 .vop_pathconf = nfs_pathconf, 183 .vop_print = nfs_print, 184 .vop_read = nfs_read, 185 .vop_readdir = nfs_readdir, 186 .vop_readlink = nfs_readlink, 187 .vop_reclaim = ncl_reclaim, 188 .vop_remove = nfs_remove, 189 .vop_rename = nfs_rename, 190 .vop_rmdir = nfs_rmdir, 191 .vop_setattr = nfs_setattr, 192 .vop_strategy = nfs_strategy, 193 .vop_symlink = nfs_symlink, 194 .vop_write = ncl_write, 195 .vop_getacl = nfs_getacl, 196 .vop_setacl = nfs_setacl, 197 .vop_advise = nfs_advise, 198 .vop_allocate = nfs_allocate, 199 .vop_deallocate = nfs_deallocate, 200 .vop_copy_file_range = nfs_copy_file_range, 201 .vop_ioctl = nfs_ioctl, 202 .vop_getextattr = nfs_getextattr, 203 .vop_setextattr = nfs_setextattr, 204 .vop_listextattr = nfs_listextattr, 205 .vop_deleteextattr = nfs_deleteextattr, 206 }; 207 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig); 208 209 static int 210 nfs_vnodeops_bypass(struct vop_generic_args *a) 211 { 212 213 return (vop_sigdefer(&newnfs_vnodeops_nosig, a)); 214 } 215 216 struct vop_vector newnfs_vnodeops = { 217 .vop_default = &default_vnodeops, 218 .vop_bypass = nfs_vnodeops_bypass, 219 }; 220 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops); 221 222 static struct vop_vector newnfs_fifoops_nosig = { 223 .vop_default = &fifo_specops, 224 .vop_access = nfsspec_access, 225 .vop_close = nfsfifo_close, 226 .vop_fsync = nfs_fsync, 227 .vop_getattr = nfs_getattr, 228 .vop_inactive = ncl_inactive, 229 .vop_pathconf = nfs_pathconf, 230 .vop_print = nfs_print, 231 .vop_read = nfsfifo_read, 232 .vop_reclaim = ncl_reclaim, 233 .vop_setattr = nfs_setattr, 234 .vop_write = nfsfifo_write, 235 }; 236 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig); 237 238 static int 239 nfs_fifoops_bypass(struct vop_generic_args *a) 240 { 241 242 return (vop_sigdefer(&newnfs_fifoops_nosig, a)); 243 } 244 245 struct vop_vector newnfs_fifoops = { 246 .vop_default = &default_vnodeops, 247 .vop_bypass = nfs_fifoops_bypass, 248 }; 249 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops); 250 251 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 252 struct componentname *cnp, struct vattr *vap); 253 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 254 int namelen, struct ucred *cred, struct thread *td, bool silly); 255 static void nfs_removestatus(struct vnode *vp, nfsremove_status file_status, 256 bool silly, struct thread *td); 257 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 258 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 259 char *tnameptr, int tnamelen, bool silly, struct ucred *cred, 260 struct thread *td); 261 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 262 struct componentname *scnp, struct sillyrename *sp); 263 264 /* 265 * Global variables 266 */ 267 SYSCTL_DECL(_vfs_nfs); 268 269 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 270 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 271 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 272 273 static int nfs_prime_access_cache = 0; 274 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 275 &nfs_prime_access_cache, 0, 276 "Prime NFS ACCESS cache when fetching attributes"); 277 278 static int newnfs_commit_on_close = 0; 279 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 280 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 281 282 static int nfs_clean_pages_on_close = 1; 283 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 284 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 285 286 int newnfs_directio_enable = 0; 287 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 288 &newnfs_directio_enable, 0, "Enable NFS directio"); 289 290 int nfs_keep_dirty_on_error; 291 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 292 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 293 294 /* 295 * This sysctl allows other processes to mmap a file that has been opened 296 * O_DIRECT by a process. In general, having processes mmap the file while 297 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 298 * this by default to prevent DoS attacks - to prevent a malicious user from 299 * opening up files O_DIRECT preventing other users from mmap'ing these 300 * files. "Protected" environments where stricter consistency guarantees are 301 * required can disable this knob. The process that opened the file O_DIRECT 302 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 303 * meaningful. 304 */ 305 int newnfs_directio_allow_mmap = 1; 306 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 307 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 308 309 static uint64_t nfs_maxalloclen = 64 * 1024 * 1024; 310 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW, 311 &nfs_maxalloclen, 0, "NFS max allocate/deallocate length"); 312 313 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 314 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 315 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 316 317 /* 318 * SMP Locking Note : 319 * The list of locks after the description of the lock is the ordering 320 * of other locks acquired with the lock held. 321 * np->n_mtx : Protects the fields in the nfsnode. 322 VM Object Lock 323 VI_MTX (acquired indirectly) 324 * nmp->nm_mtx : Protects the fields in the nfsmount. 325 rep->r_mtx 326 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 327 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 328 nmp->nm_mtx 329 rep->r_mtx 330 * rep->r_mtx : Protects the fields in an nfsreq. 331 */ 332 333 static int 334 nfs_lock(struct vop_lock1_args *ap) 335 { 336 struct vnode *vp; 337 struct nfsnode *np; 338 u_quad_t nsize; 339 int error, lktype; 340 bool onfault; 341 342 vp = ap->a_vp; 343 lktype = ap->a_flags & LK_TYPE_MASK; 344 error = VOP_LOCK1_APV(&default_vnodeops, ap); 345 if (error != 0 || vp->v_op != &newnfs_vnodeops) 346 return (error); 347 np = VTONFS(vp); 348 if (np == NULL) 349 return (0); 350 NFSLOCKNODE(np); 351 if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED && 352 lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE && 353 lktype != LK_TRYUPGRADE)) { 354 NFSUNLOCKNODE(np); 355 return (0); 356 } 357 onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT && 358 (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE && 359 (lktype == LK_SHARED || lktype == LK_EXCLUSIVE); 360 if (onfault && vp->v_vnlock->lk_recurse == 0) { 361 /* 362 * Force retry in vm_fault(), to make the lock request 363 * sleepable, which allows us to piggy-back the 364 * sleepable call to vnode_pager_setsize(). 365 */ 366 NFSUNLOCKNODE(np); 367 VOP_UNLOCK(vp); 368 return (EBUSY); 369 } 370 if ((ap->a_flags & LK_NOWAIT) != 0 || 371 (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) { 372 NFSUNLOCKNODE(np); 373 return (0); 374 } 375 if (lktype == LK_SHARED) { 376 NFSUNLOCKNODE(np); 377 VOP_UNLOCK(vp); 378 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 379 ap->a_flags |= LK_EXCLUSIVE; 380 error = VOP_LOCK1_APV(&default_vnodeops, ap); 381 if (error != 0 || vp->v_op != &newnfs_vnodeops) 382 return (error); 383 if (vp->v_data == NULL) 384 goto downgrade; 385 MPASS(vp->v_data == np); 386 NFSLOCKNODE(np); 387 if ((np->n_flag & NVNSETSZSKIP) == 0) { 388 NFSUNLOCKNODE(np); 389 goto downgrade; 390 } 391 } 392 np->n_flag &= ~NVNSETSZSKIP; 393 nsize = np->n_size; 394 NFSUNLOCKNODE(np); 395 vnode_pager_setsize(vp, nsize); 396 downgrade: 397 if (lktype == LK_SHARED) { 398 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 399 ap->a_flags |= LK_DOWNGRADE; 400 (void)VOP_LOCK1_APV(&default_vnodeops, ap); 401 } 402 return (0); 403 } 404 405 static int 406 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 407 struct ucred *cred, u_int32_t *retmode) 408 { 409 int error = 0, attrflag, i, lrupos; 410 u_int32_t rmode; 411 struct nfsnode *np = VTONFS(vp); 412 struct nfsvattr nfsva; 413 414 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 415 &rmode); 416 if (attrflag) 417 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 418 if (!error) { 419 lrupos = 0; 420 NFSLOCKNODE(np); 421 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 422 if (np->n_accesscache[i].uid == cred->cr_uid) { 423 np->n_accesscache[i].mode = rmode; 424 np->n_accesscache[i].stamp = time_second; 425 break; 426 } 427 if (i > 0 && np->n_accesscache[i].stamp < 428 np->n_accesscache[lrupos].stamp) 429 lrupos = i; 430 } 431 if (i == NFS_ACCESSCACHESIZE) { 432 np->n_accesscache[lrupos].uid = cred->cr_uid; 433 np->n_accesscache[lrupos].mode = rmode; 434 np->n_accesscache[lrupos].stamp = time_second; 435 } 436 NFSUNLOCKNODE(np); 437 if (retmode != NULL) 438 *retmode = rmode; 439 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 440 } else if (NFS_ISV4(vp)) { 441 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 442 } 443 #ifdef KDTRACE_HOOKS 444 if (error != 0) 445 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 446 error); 447 #endif 448 return (error); 449 } 450 451 /* 452 * nfs access vnode op. 453 * For nfs version 2, just return ok. File accesses may fail later. 454 * For nfs version 3, use the access rpc to check accessibility. If file modes 455 * are changed on the server, accesses might still fail later. 456 */ 457 static int 458 nfs_access(struct vop_access_args *ap) 459 { 460 struct vnode *vp = ap->a_vp; 461 int error = 0, i, gotahit; 462 u_int32_t mode, wmode, rmode; 463 int v34 = NFS_ISV34(vp); 464 struct nfsnode *np = VTONFS(vp); 465 466 /* 467 * Disallow write attempts on filesystems mounted read-only; 468 * unless the file is a socket, fifo, or a block or character 469 * device resident on the filesystem. 470 */ 471 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 472 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 473 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 474 switch (vp->v_type) { 475 case VREG: 476 case VDIR: 477 case VLNK: 478 return (EROFS); 479 default: 480 break; 481 } 482 } 483 484 /* 485 * For NFSv4, check for a delegation with an Allow ACE, to see 486 * if that permits access. 487 */ 488 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) != 0) { 489 error = nfscl_delegacecheck(vp, ap->a_accmode, ap->a_cred); 490 if (error == 0) 491 return (error); 492 error = 0; 493 } 494 495 /* 496 * For nfs v3 or v4, check to see if we have done this recently, and if 497 * so return our cached result instead of making an ACCESS call. 498 * If not, do an access rpc, otherwise you are stuck emulating 499 * ufs_access() locally using the vattr. This may not be correct, 500 * since the server may apply other access criteria such as 501 * client uid-->server uid mapping that we do not know about. 502 */ 503 if (v34) { 504 if (ap->a_accmode & VREAD) 505 mode = NFSACCESS_READ; 506 else 507 mode = 0; 508 if (vp->v_type != VDIR) { 509 if (ap->a_accmode & VWRITE) 510 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 511 if (ap->a_accmode & VAPPEND) 512 mode |= NFSACCESS_EXTEND; 513 if (ap->a_accmode & VEXEC) 514 mode |= NFSACCESS_EXECUTE; 515 if (ap->a_accmode & VDELETE) 516 mode |= NFSACCESS_DELETE; 517 } else { 518 if (ap->a_accmode & VWRITE) 519 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 520 if (ap->a_accmode & VAPPEND) 521 mode |= NFSACCESS_EXTEND; 522 if (ap->a_accmode & VEXEC) 523 mode |= NFSACCESS_LOOKUP; 524 if (ap->a_accmode & VDELETE) 525 mode |= NFSACCESS_DELETE; 526 if (ap->a_accmode & VDELETE_CHILD) 527 mode |= NFSACCESS_MODIFY; 528 } 529 /* XXX safety belt, only make blanket request if caching */ 530 if (nfsaccess_cache_timeout > 0) { 531 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 532 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 533 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 534 } else { 535 wmode = mode; 536 } 537 538 /* 539 * Does our cached result allow us to give a definite yes to 540 * this request? 541 */ 542 gotahit = 0; 543 NFSLOCKNODE(np); 544 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 545 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 546 if (time_second < (np->n_accesscache[i].stamp 547 + nfsaccess_cache_timeout) && 548 (np->n_accesscache[i].mode & mode) == mode) { 549 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 550 gotahit = 1; 551 } 552 break; 553 } 554 } 555 NFSUNLOCKNODE(np); 556 #ifdef KDTRACE_HOOKS 557 if (gotahit != 0) 558 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 559 ap->a_cred->cr_uid, mode); 560 else 561 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 562 ap->a_cred->cr_uid, mode); 563 #endif 564 if (gotahit == 0) { 565 /* 566 * Either a no, or a don't know. Go to the wire. 567 */ 568 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 569 error = nfs34_access_otw(vp, wmode, ap->a_td, 570 ap->a_cred, &rmode); 571 if (!error && 572 (rmode & mode) != mode) 573 error = EACCES; 574 } 575 return (error); 576 } else { 577 if ((error = nfsspec_access(ap)) != 0) { 578 return (error); 579 } 580 /* 581 * Attempt to prevent a mapped root from accessing a file 582 * which it shouldn't. We try to read a byte from the file 583 * if the user is root and the file is not zero length. 584 * After calling nfsspec_access, we should have the correct 585 * file size cached. 586 */ 587 NFSLOCKNODE(np); 588 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 589 && VTONFS(vp)->n_size > 0) { 590 struct iovec aiov; 591 struct uio auio; 592 char buf[1]; 593 594 NFSUNLOCKNODE(np); 595 aiov.iov_base = buf; 596 aiov.iov_len = 1; 597 auio.uio_iov = &aiov; 598 auio.uio_iovcnt = 1; 599 auio.uio_offset = 0; 600 auio.uio_resid = 1; 601 auio.uio_segflg = UIO_SYSSPACE; 602 auio.uio_rw = UIO_READ; 603 auio.uio_td = ap->a_td; 604 605 if (vp->v_type == VREG) 606 error = ncl_readrpc(vp, &auio, ap->a_cred); 607 else if (vp->v_type == VDIR) { 608 char* bp; 609 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 610 aiov.iov_base = bp; 611 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 612 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 613 ap->a_td); 614 free(bp, M_TEMP); 615 } else if (vp->v_type == VLNK) 616 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 617 else 618 error = EACCES; 619 } else 620 NFSUNLOCKNODE(np); 621 return (error); 622 } 623 } 624 625 /* 626 * nfs open vnode op 627 * Check to see if the type is ok 628 * and that deletion is not in progress. 629 * For paged in text files, you will need to flush the page cache 630 * if consistency is lost. 631 */ 632 /* ARGSUSED */ 633 static int 634 nfs_open(struct vop_open_args *ap) 635 { 636 struct vnode *vp = ap->a_vp; 637 struct nfsnode *np = VTONFS(vp); 638 struct vattr vattr; 639 int error; 640 int fmode = ap->a_mode; 641 struct ucred *cred; 642 vm_object_t obj; 643 644 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 645 return (EOPNOTSUPP); 646 647 /* 648 * For NFSv4, we need to do the Open Op before cache validation, 649 * so that we conform to RFC3530 Sec. 9.3.1. 650 */ 651 if (NFS_ISV4(vp)) { 652 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 653 if (error) { 654 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 655 (gid_t)0); 656 return (error); 657 } 658 } 659 660 /* 661 * Now, if this Open will be doing reading, re-validate/flush the 662 * cache, so that Close/Open coherency is maintained. 663 */ 664 NFSLOCKNODE(np); 665 if (np->n_flag & NMODIFIED) { 666 NFSUNLOCKNODE(np); 667 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 668 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 669 if (VN_IS_DOOMED(vp)) 670 return (EBADF); 671 } 672 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 673 if (error == EINTR || error == EIO) { 674 if (NFS_ISV4(vp)) 675 (void) nfsrpc_close(vp, 0, ap->a_td); 676 return (error); 677 } 678 NFSLOCKNODE(np); 679 np->n_attrstamp = 0; 680 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 681 if (vp->v_type == VDIR) 682 np->n_direofoffset = 0; 683 NFSUNLOCKNODE(np); 684 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 685 if (error) { 686 if (NFS_ISV4(vp)) 687 (void) nfsrpc_close(vp, 0, ap->a_td); 688 return (error); 689 } 690 NFSLOCKNODE(np); 691 np->n_mtime = vattr.va_mtime; 692 if (NFS_ISV4(vp)) 693 np->n_change = vattr.va_filerev; 694 } else { 695 NFSUNLOCKNODE(np); 696 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 697 if (error) { 698 if (NFS_ISV4(vp)) 699 (void) nfsrpc_close(vp, 0, ap->a_td); 700 return (error); 701 } 702 NFSLOCKNODE(np); 703 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 704 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 705 if (vp->v_type == VDIR) 706 np->n_direofoffset = 0; 707 NFSUNLOCKNODE(np); 708 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 709 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 710 if (VN_IS_DOOMED(vp)) 711 return (EBADF); 712 } 713 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 714 if (error == EINTR || error == EIO) { 715 if (NFS_ISV4(vp)) 716 (void) nfsrpc_close(vp, 0, ap->a_td); 717 return (error); 718 } 719 NFSLOCKNODE(np); 720 np->n_mtime = vattr.va_mtime; 721 if (NFS_ISV4(vp)) 722 np->n_change = vattr.va_filerev; 723 } 724 } 725 726 /* 727 * If the object has >= 1 O_DIRECT active opens, we disable caching. 728 */ 729 if (newnfs_directio_enable && (fmode & O_DIRECT) && 730 (vp->v_type == VREG)) { 731 if (np->n_directio_opens == 0) { 732 NFSUNLOCKNODE(np); 733 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 734 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 735 if (VN_IS_DOOMED(vp)) 736 return (EBADF); 737 } 738 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 739 if (error) { 740 if (NFS_ISV4(vp)) 741 (void) nfsrpc_close(vp, 0, ap->a_td); 742 return (error); 743 } 744 NFSLOCKNODE(np); 745 np->n_flag |= NNONCACHE; 746 } 747 np->n_directio_opens++; 748 } 749 750 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 751 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 752 np->n_flag |= NWRITEOPENED; 753 754 /* 755 * If this is an open for writing, capture a reference to the 756 * credentials, so they can be used by ncl_putpages(). Using 757 * these write credentials is preferable to the credentials of 758 * whatever thread happens to be doing the VOP_PUTPAGES() since 759 * the write RPCs are less likely to fail with EACCES. 760 */ 761 if ((fmode & FWRITE) != 0) { 762 cred = np->n_writecred; 763 np->n_writecred = crhold(ap->a_cred); 764 } else 765 cred = NULL; 766 NFSUNLOCKNODE(np); 767 768 if (cred != NULL) 769 crfree(cred); 770 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 771 772 /* 773 * If the text file has been mmap'd, flush any dirty pages to the 774 * buffer cache and then... 775 * Make sure all writes are pushed to the NFS server. If this is not 776 * done, the modify time of the file can change while the text 777 * file is being executed. This will cause the process that is 778 * executing the text file to be terminated. 779 */ 780 if (vp->v_writecount <= -1) { 781 if ((obj = vp->v_object) != NULL && 782 vm_object_mightbedirty(obj)) { 783 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 784 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 785 if (VN_IS_DOOMED(vp)) 786 return (EBADF); 787 } 788 vnode_pager_clean_sync(vp); 789 } 790 791 /* Now, flush the buffer cache. */ 792 ncl_flush(vp, MNT_WAIT, curthread, 0, 0); 793 794 /* And, finally, make sure that n_mtime is up to date. */ 795 np = VTONFS(vp); 796 NFSLOCKNODE(np); 797 np->n_mtime = np->n_vattr.na_mtime; 798 NFSUNLOCKNODE(np); 799 } 800 return (0); 801 } 802 803 /* 804 * nfs close vnode op 805 * What an NFS client should do upon close after writing is a debatable issue. 806 * Most NFS clients push delayed writes to the server upon close, basically for 807 * two reasons: 808 * 1 - So that any write errors may be reported back to the client process 809 * doing the close system call. By far the two most likely errors are 810 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 811 * 2 - To put a worst case upper bound on cache inconsistency between 812 * multiple clients for the file. 813 * There is also a consistency problem for Version 2 of the protocol w.r.t. 814 * not being able to tell if other clients are writing a file concurrently, 815 * since there is no way of knowing if the changed modify time in the reply 816 * is only due to the write for this client. 817 * (NFS Version 3 provides weak cache consistency data in the reply that 818 * should be sufficient to detect and handle this case.) 819 * 820 * The current code does the following: 821 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 822 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 823 * or commit them (this satisfies 1 and 2 except for the 824 * case where the server crashes after this close but 825 * before the commit RPC, which is felt to be "good 826 * enough". Changing the last argument to ncl_flush() to 827 * a 1 would force a commit operation, if it is felt a 828 * commit is necessary now. 829 * for NFS Version 4 - flush the dirty buffers and commit them, if 830 * nfscl_mustflush() says this is necessary. 831 * It is necessary if there is no write delegation held, 832 * in order to satisfy open/close coherency. 833 * If the file isn't cached on local stable storage, 834 * it may be necessary in order to detect "out of space" 835 * errors from the server, if the write delegation 836 * issued by the server doesn't allow the file to grow. 837 */ 838 /* ARGSUSED */ 839 static int 840 nfs_close(struct vop_close_args *ap) 841 { 842 struct vnode *vp = ap->a_vp; 843 struct nfsnode *np = VTONFS(vp); 844 struct nfsvattr nfsva; 845 struct ucred *cred; 846 int error = 0, ret, localcred = 0; 847 int fmode = ap->a_fflag; 848 struct nfsmount *nmp; 849 850 if (NFSCL_FORCEDISM(vp->v_mount)) 851 return (0); 852 nmp = VFSTONFS(vp->v_mount); 853 /* 854 * During shutdown, a_cred isn't valid, so just use root. 855 */ 856 if (ap->a_cred == NOCRED) { 857 cred = newnfs_getcred(); 858 localcred = 1; 859 } else { 860 cred = ap->a_cred; 861 } 862 if (vp->v_type == VREG) { 863 /* 864 * Examine and clean dirty pages, regardless of NMODIFIED. 865 * This closes a major hole in close-to-open consistency. 866 * We want to push out all dirty pages (and buffers) on 867 * close, regardless of whether they were dirtied by 868 * mmap'ed writes or via write(). 869 */ 870 if (nfs_clean_pages_on_close && vp->v_object) { 871 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 872 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 873 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 874 return (EBADF); 875 } 876 vnode_pager_clean_async(vp); 877 } 878 NFSLOCKNODE(np); 879 if (np->n_flag & NMODIFIED) { 880 NFSUNLOCKNODE(np); 881 if (NFS_ISV3(vp)) { 882 /* 883 * Under NFSv3 we have dirty buffers to dispose of. We 884 * must flush them to the NFS server. We have the option 885 * of waiting all the way through the commit rpc or just 886 * waiting for the initial write. The default is to only 887 * wait through the initial write so the data is in the 888 * server's cache, which is roughly similar to the state 889 * a standard disk subsystem leaves the file in on close(). 890 * 891 * We cannot clear the NMODIFIED bit in np->n_flag due to 892 * potential races with other processes, and certainly 893 * cannot clear it if we don't commit. 894 * These races occur when there is no longer the old 895 * traditional vnode locking implemented for Vnode Ops. 896 */ 897 int cm = newnfs_commit_on_close ? 1 : 0; 898 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 899 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 900 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 901 return (EBADF); 902 } 903 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); 904 /* np->n_flag &= ~NMODIFIED; */ 905 } else if (NFS_ISV4(vp)) { 906 if (!NFSHASNFSV4N(nmp) || 907 (nmp->nm_flag & NFSMNT_NOCTO) == 0 || 908 nfscl_mustflush(vp) != 0) { 909 int cm = newnfs_commit_on_close ? 1 : 0; 910 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 911 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 912 if (VN_IS_DOOMED(vp) && ap->a_fflag != 913 FNONBLOCK) 914 return (EBADF); 915 } 916 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 917 cm, 0); 918 /* 919 * as above w.r.t races when clearing 920 * NMODIFIED. 921 * np->n_flag &= ~NMODIFIED; 922 */ 923 } 924 } else { 925 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 926 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 927 if (VN_IS_DOOMED(vp) && ap->a_fflag != 928 FNONBLOCK) 929 return (EBADF); 930 } 931 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 932 } 933 NFSLOCKNODE(np); 934 } 935 /* 936 * Invalidate the attribute cache in all cases. 937 * An open is going to fetch fresh attrs any way, other procs 938 * on this node that have file open will be forced to do an 939 * otw attr fetch, but this is safe. 940 * --> A user found that their RPC count dropped by 20% when 941 * this was commented out and I can't see any requirement 942 * for it, so I've disabled it when negative lookups are 943 * enabled. (What does this have to do with negative lookup 944 * caching? Well nothing, except it was reported by the 945 * same user that needed negative lookup caching and I wanted 946 * there to be a way to disable it to see if it 947 * is the cause of some caching/coherency issue that might 948 * crop up.) 949 */ 950 if (nmp->nm_negnametimeo == 0) { 951 np->n_attrstamp = 0; 952 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 953 } 954 if (np->n_flag & NWRITEERR) { 955 np->n_flag &= ~NWRITEERR; 956 error = np->n_error; 957 } 958 NFSUNLOCKNODE(np); 959 } 960 961 if (NFS_ISV4(vp)) { 962 /* 963 * Get attributes so "change" is up to date. 964 */ 965 if (error == 0 && nfscl_nodeleg(vp, 0) != 0 && 966 vp->v_type == VREG && 967 (nmp->nm_flag & NFSMNT_NOCTO) == 0) { 968 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva); 969 if (!ret) { 970 np->n_change = nfsva.na_filerev; 971 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 972 0, 0); 973 } 974 } 975 976 /* 977 * and do the close. 978 */ 979 ret = nfsrpc_close(vp, 0, ap->a_td); 980 if (!error && ret) 981 error = ret; 982 if (error) 983 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 984 (gid_t)0); 985 } 986 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 987 NFSLOCKNODE(np); 988 KASSERT((np->n_directio_opens > 0), 989 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 990 np->n_directio_opens--; 991 if (np->n_directio_opens == 0) 992 np->n_flag &= ~NNONCACHE; 993 NFSUNLOCKNODE(np); 994 } 995 if (localcred) 996 NFSFREECRED(cred); 997 return (error); 998 } 999 1000 /* 1001 * nfs getattr call from vfs. 1002 */ 1003 static int 1004 nfs_getattr(struct vop_getattr_args *ap) 1005 { 1006 struct vnode *vp = ap->a_vp; 1007 struct thread *td = curthread; /* XXX */ 1008 struct nfsnode *np = VTONFS(vp); 1009 int error = 0; 1010 struct nfsvattr nfsva; 1011 struct vattr *vap = ap->a_vap; 1012 struct vattr vattr; 1013 struct nfsmount *nmp; 1014 1015 nmp = VFSTONFS(vp->v_mount); 1016 /* 1017 * Update local times for special files. 1018 */ 1019 NFSLOCKNODE(np); 1020 if (np->n_flag & (NACC | NUPD)) 1021 np->n_flag |= NCHG; 1022 NFSUNLOCKNODE(np); 1023 /* 1024 * First look in the cache. 1025 * For "syskrb5" mounts, nm_fhsize might still be zero and 1026 * cached attributes should be ignored. 1027 */ 1028 if (nmp->nm_fhsize > 0 && ncl_getattrcache(vp, &vattr) == 0) { 1029 ncl_copy_vattr(vap, &vattr); 1030 1031 /* 1032 * Get the local modify time for the case of a write 1033 * delegation. 1034 */ 1035 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1036 return (0); 1037 } 1038 1039 if (NFS_ISV34(vp) && nfs_prime_access_cache && 1040 nfsaccess_cache_timeout > 0) { 1041 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 1042 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 1043 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 1044 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 1045 return (0); 1046 } 1047 } 1048 1049 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva); 1050 if (error == 0) 1051 error = nfscl_loadattrcache(&vp, &nfsva, vap, 0, 0); 1052 if (!error) { 1053 /* 1054 * Get the local modify time for the case of a write 1055 * delegation. 1056 */ 1057 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1058 } else if (NFS_ISV4(vp)) { 1059 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1060 } 1061 return (error); 1062 } 1063 1064 /* 1065 * nfs setattr call. 1066 */ 1067 static int 1068 nfs_setattr(struct vop_setattr_args *ap) 1069 { 1070 struct vnode *vp = ap->a_vp; 1071 struct nfsnode *np = VTONFS(vp); 1072 struct thread *td = curthread; /* XXX */ 1073 struct vattr *vap = ap->a_vap; 1074 int error = 0; 1075 u_quad_t tsize; 1076 struct timespec ts; 1077 struct nfsmount *nmp; 1078 1079 #ifndef nolint 1080 tsize = (u_quad_t)0; 1081 #endif 1082 1083 /* 1084 * Only setting of UF_HIDDEN and UF_SYSTEM are supported and 1085 * only for NFSv4 servers that support them. 1086 */ 1087 nmp = VFSTONFS(vp->v_mount); 1088 if (vap->va_flags != VNOVAL && (!NFSHASNFSV4(nmp) || 1089 (vap->va_flags & ~(UF_HIDDEN | UF_SYSTEM)) != 0 || 1090 ((vap->va_flags & UF_HIDDEN) != 0 && 1091 !NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, NFSATTRBIT_HIDDEN)) || 1092 ((vap->va_flags & UF_SYSTEM) != 0 && 1093 !NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, NFSATTRBIT_SYSTEM)))) 1094 return (EOPNOTSUPP); 1095 1096 /* 1097 * Disallow write attempts if the filesystem is mounted read-only. 1098 */ 1099 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 1100 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 1101 vap->va_mtime.tv_sec != VNOVAL || 1102 vap->va_birthtime.tv_sec != VNOVAL || 1103 vap->va_mode != (mode_t)VNOVAL || 1104 vap->va_flags != (u_long)VNOVAL) && 1105 (vp->v_mount->mnt_flag & MNT_RDONLY)) 1106 return (EROFS); 1107 if (vap->va_size != VNOVAL) { 1108 switch (vp->v_type) { 1109 case VDIR: 1110 return (EISDIR); 1111 case VCHR: 1112 case VBLK: 1113 case VSOCK: 1114 case VFIFO: 1115 if (vap->va_mtime.tv_sec == VNOVAL && 1116 vap->va_atime.tv_sec == VNOVAL && 1117 vap->va_birthtime.tv_sec == VNOVAL && 1118 vap->va_mode == (mode_t)VNOVAL && 1119 vap->va_uid == (uid_t)VNOVAL && 1120 vap->va_gid == (gid_t)VNOVAL) 1121 return (0); 1122 vap->va_size = VNOVAL; 1123 break; 1124 default: 1125 /* 1126 * Disallow write attempts if the filesystem is 1127 * mounted read-only. 1128 */ 1129 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1130 return (EROFS); 1131 /* 1132 * We run vnode_pager_setsize() early (why?), 1133 * we must set np->n_size now to avoid vinvalbuf 1134 * V_SAVE races that might setsize a lower 1135 * value. 1136 */ 1137 NFSLOCKNODE(np); 1138 tsize = np->n_size; 1139 NFSUNLOCKNODE(np); 1140 error = ncl_meta_setsize(vp, td, vap->va_size); 1141 NFSLOCKNODE(np); 1142 if (np->n_flag & NMODIFIED) { 1143 tsize = np->n_size; 1144 NFSUNLOCKNODE(np); 1145 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 1146 0 : V_SAVE, td, 1); 1147 if (error != 0) { 1148 vnode_pager_setsize(vp, tsize); 1149 return (error); 1150 } 1151 /* 1152 * Call nfscl_delegmodtime() to set the modify time 1153 * locally, as required. 1154 */ 1155 nfscl_delegmodtime(vp, NULL); 1156 } else 1157 NFSUNLOCKNODE(np); 1158 /* 1159 * np->n_size has already been set to vap->va_size 1160 * in ncl_meta_setsize(). We must set it again since 1161 * nfs_loadattrcache() could be called through 1162 * ncl_meta_setsize() and could modify np->n_size. 1163 */ 1164 NFSLOCKNODE(np); 1165 np->n_vattr.na_size = np->n_size = vap->va_size; 1166 NFSUNLOCKNODE(np); 1167 } 1168 } else { 1169 NFSLOCKNODE(np); 1170 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 1171 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 1172 NFSUNLOCKNODE(np); 1173 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 1174 if (error == EINTR || error == EIO) 1175 return (error); 1176 } else 1177 NFSUNLOCKNODE(np); 1178 } 1179 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 1180 if (vap->va_size != VNOVAL) { 1181 if (error == 0) { 1182 nanouptime(&ts); 1183 NFSLOCKNODE(np); 1184 np->n_localmodtime = ts; 1185 NFSUNLOCKNODE(np); 1186 } else { 1187 NFSLOCKNODE(np); 1188 np->n_size = np->n_vattr.na_size = tsize; 1189 vnode_pager_setsize(vp, tsize); 1190 NFSUNLOCKNODE(np); 1191 } 1192 } 1193 if (vap->va_mtime.tv_sec != VNOVAL && error == 0) 1194 nfscl_delegmodtime(vp, &vap->va_mtime); 1195 return (error); 1196 } 1197 1198 /* 1199 * Do an nfs setattr rpc. 1200 */ 1201 static int 1202 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1203 struct thread *td) 1204 { 1205 struct nfsnode *np = VTONFS(vp); 1206 int error, ret, attrflag, i; 1207 struct nfsvattr nfsva; 1208 1209 if (NFS_ISV34(vp)) { 1210 NFSLOCKNODE(np); 1211 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1212 np->n_accesscache[i].stamp = 0; 1213 np->n_flag |= NDELEGMOD; 1214 NFSUNLOCKNODE(np); 1215 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1216 } 1217 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag); 1218 if (attrflag) { 1219 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1220 if (ret && !error) 1221 error = ret; 1222 } 1223 if (error && NFS_ISV4(vp)) 1224 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1225 return (error); 1226 } 1227 1228 /* 1229 * Get a named attribute directory for the vnode. 1230 */ 1231 static int 1232 nfs_get_namedattrdir(struct vnode *vp, struct componentname *cnp, 1233 struct vnode **vpp) 1234 { 1235 struct nfsfh *nfhp; 1236 struct nfsnode *np; 1237 struct vnode *newvp; 1238 struct nfsvattr nfsva; 1239 int attrflag, error; 1240 1241 attrflag = 0; 1242 *vpp = NULL; 1243 np = VTONFS(vp); 1244 error = nfsrpc_openattr(VFSTONFS(vp->v_mount), vp, np->n_fhp->nfh_fh, 1245 np->n_fhp->nfh_len, (cnp->cn_flags & CREATENAMED), 1246 cnp->cn_cred, curthread, &nfsva, &nfhp, &attrflag); 1247 if (error == NFSERR_NOTSUPP) 1248 error = ENOATTR; 1249 if (error == 0) 1250 error = nfscl_nget(vp->v_mount, vp, nfhp, cnp, curthread, &np, 1251 cnp->cn_lkflags); 1252 if (error != 0) 1253 return (error); 1254 newvp = NFSTOV(np); 1255 vn_irflag_set_cond(newvp, VIRF_NAMEDDIR); 1256 if (attrflag != 0) 1257 (void)nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1258 *vpp = newvp; 1259 return (0); 1260 } 1261 1262 /* 1263 * nfs lookup call, one step at a time... 1264 * First look in cache 1265 * If not found, unlock the directory nfsnode and do the rpc 1266 */ 1267 static int 1268 nfs_lookup(struct vop_lookup_args *ap) 1269 { 1270 struct componentname *cnp = ap->a_cnp; 1271 struct vnode *dvp = ap->a_dvp; 1272 struct vnode **vpp = ap->a_vpp; 1273 struct mount *mp = dvp->v_mount; 1274 uint64_t flags = cnp->cn_flags; 1275 struct vnode *newvp; 1276 struct nfsmount *nmp; 1277 struct nfsnode *np, *newnp; 1278 int error = 0, attrflag, dattrflag, ltype, ncticks; 1279 struct thread *td = curthread; 1280 struct nfsfh *nfhp; 1281 struct nfsvattr dnfsva, nfsva; 1282 struct vattr vattr; 1283 struct timespec nctime, ts; 1284 uint32_t openmode; 1285 bool is_nameddir, needs_nameddir, opennamed; 1286 1287 dattrflag = 0; 1288 *vpp = NULLVP; 1289 nmp = VFSTONFS(mp); 1290 opennamed = (flags & (OPENNAMED | ISLASTCN)) == (OPENNAMED | ISLASTCN); 1291 if (opennamed && (!NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp))) 1292 return (ENOATTR); 1293 is_nameddir = (vn_irflag_read(dvp) & VIRF_NAMEDDIR) != 0; 1294 if ((is_nameddir && (flags & ISLASTCN) == 0 && (cnp->cn_namelen > 1 || 1295 *cnp->cn_nameptr != '.')) || 1296 (opennamed && !is_nameddir && (flags & ISDOTDOT) != 0)) 1297 return (ENOATTR); 1298 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1299 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1300 return (EROFS); 1301 np = VTONFS(dvp); 1302 1303 needs_nameddir = false; 1304 if (opennamed || is_nameddir) { 1305 cnp->cn_flags &= ~MAKEENTRY; 1306 if (!is_nameddir) 1307 needs_nameddir = true; 1308 } 1309 1310 /* 1311 * If the named attribute directory is needed, acquire it now. 1312 */ 1313 newvp = NULLVP; 1314 if (needs_nameddir) { 1315 KASSERT(np->n_v4 == NULL, ("nfs_lookup: O_NAMEDATTR when" 1316 " n_v4 not NULL")); 1317 error = nfs_get_namedattrdir(dvp, cnp, &newvp); 1318 if (error != 0) 1319 goto handle_error; 1320 if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') { 1321 *vpp = newvp; 1322 return (0); 1323 } 1324 dvp = newvp; 1325 np = VTONFS(dvp); 1326 newvp = NULLVP; 1327 } else if (opennamed && cnp->cn_namelen == 1 && 1328 *cnp->cn_nameptr == '.') { 1329 VREF(dvp); 1330 *vpp = dvp; 1331 return (0); 1332 } 1333 1334 if (dvp->v_type != VDIR) 1335 return (ENOTDIR); 1336 1337 /* For NFSv4, wait until any remove is done. */ 1338 NFSLOCKNODE(np); 1339 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1340 np->n_flag |= NREMOVEWANT; 1341 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1342 } 1343 NFSUNLOCKNODE(np); 1344 1345 error = vn_dir_check_exec(dvp, cnp); 1346 if (error != 0) 1347 return (error); 1348 1349 if (!opennamed && !is_nameddir) { 1350 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1351 if (error > 0 && error != ENOENT) 1352 return (error); 1353 if (error == -1) { 1354 /* 1355 * Lookups of "." are special and always return the 1356 * current directory. cache_lookup() already handles 1357 * associated locking bookkeeping, etc. 1358 */ 1359 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1360 return (0); 1361 } 1362 1363 /* 1364 * We only accept a positive hit in the cache if the 1365 * change time of the file matches our cached copy. 1366 * Otherwise, we discard the cache entry and fallback 1367 * to doing a lookup RPC. We also only trust cache 1368 * entries for less than nm_nametimeo seconds. 1369 * 1370 * To better handle stale file handles and attributes, 1371 * clear the attribute cache of this node if it is a 1372 * leaf component, part of an open() call, and not 1373 * locally modified before fetching the attributes. 1374 * This should allow stale file handles to be detected 1375 * here where we can fall back to a LOOKUP RPC to 1376 * recover rather than having nfs_open() detect the 1377 * stale file handle and failing open(2) with ESTALE. 1378 */ 1379 newvp = *vpp; 1380 newnp = VTONFS(newvp); 1381 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1382 (flags & (ISLASTCN | ISOPEN)) == 1383 (ISLASTCN | ISOPEN) && 1384 !(newnp->n_flag & NMODIFIED)) { 1385 NFSLOCKNODE(newnp); 1386 newnp->n_attrstamp = 0; 1387 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1388 NFSUNLOCKNODE(newnp); 1389 } 1390 if (nfscl_nodeleg(newvp, 0) == 0 || 1391 ((u_int)(ticks - ncticks) < 1392 (nmp->nm_nametimeo * hz) && 1393 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1394 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1395 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1396 return (0); 1397 } 1398 cache_purge(newvp); 1399 if (dvp != newvp) 1400 vput(newvp); 1401 else 1402 vrele(newvp); 1403 *vpp = NULLVP; 1404 } else if (error == ENOENT) { 1405 if (VN_IS_DOOMED(dvp)) 1406 return (ENOENT); 1407 /* 1408 * We only accept a negative hit in the cache if the 1409 * modification time of the parent directory matches 1410 * the cached copy in the name cache entry. 1411 * Otherwise, we discard all of the negative cache 1412 * entries for this directory. We also only trust 1413 * negative cache entries for up to nm_negnametimeo 1414 * seconds. 1415 */ 1416 if ((u_int)(ticks - ncticks) < 1417 (nmp->nm_negnametimeo * hz) && 1418 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1419 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1420 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1421 return (ENOENT); 1422 } 1423 cache_purge_negative(dvp); 1424 } 1425 } 1426 1427 openmode = 0; 1428 #if 0 1429 /* 1430 * The use of LookupOpen breaks some builds. It is disabled 1431 * until that is fixed. 1432 */ 1433 /* 1434 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount 1435 * option, it is possible to do the Open operation in the same 1436 * compound as Lookup, so long as delegations are not being 1437 * issued. This saves doing a separate RPC for Open. 1438 * For pnfs, do not do this, since the Open+LayoutGet will 1439 * be needed as a separate RPC. 1440 */ 1441 NFSLOCKMNT(nmp); 1442 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) && 1443 (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 && 1444 (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) && 1445 (flags & (ISLASTCN | ISOPEN | OPENNAMED))) == (ISLASTCN | ISOPEN)) { 1446 if ((flags & OPENREAD) != 0) 1447 openmode |= NFSV4OPEN_ACCESSREAD; 1448 if ((flags & OPENWRITE) != 0) 1449 openmode |= NFSV4OPEN_ACCESSWRITE; 1450 } 1451 NFSUNLOCKMNT(nmp); 1452 #endif 1453 1454 newvp = NULLVP; 1455 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1456 nanouptime(&ts); 1457 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1458 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1459 openmode); 1460 if (dattrflag) 1461 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1462 if (needs_nameddir) { 1463 vput(dvp); 1464 dvp = ap->a_dvp; 1465 } 1466 handle_error: 1467 if (error) { 1468 if (newvp != NULLVP) { 1469 vput(newvp); 1470 *vpp = NULLVP; 1471 } 1472 1473 if (error != ENOENT) { 1474 if (NFS_ISV4(dvp)) 1475 error = nfscl_maperr(td, error, 1476 (uid_t)0, (gid_t)0); 1477 return (error); 1478 } 1479 1480 /* The requested file was not found. */ 1481 if ((cnp->cn_nameiop == CREATE || 1482 cnp->cn_nameiop == RENAME) && 1483 (flags & ISLASTCN)) { 1484 /* 1485 * XXX: UFS does a full VOP_ACCESS(dvp, 1486 * VWRITE) here instead of just checking 1487 * MNT_RDONLY. 1488 */ 1489 if (mp->mnt_flag & MNT_RDONLY) 1490 return (EROFS); 1491 return (EJUSTRETURN); 1492 } 1493 1494 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1495 /* 1496 * Cache the modification time of the parent 1497 * directory from the post-op attributes in 1498 * the name cache entry. The negative cache 1499 * entry will be ignored once the directory 1500 * has changed. Don't bother adding the entry 1501 * if the directory has already changed. 1502 */ 1503 NFSLOCKNODE(np); 1504 if (timespeccmp(&np->n_vattr.na_mtime, 1505 &dnfsva.na_mtime, ==)) { 1506 NFSUNLOCKNODE(np); 1507 cache_enter_time(dvp, NULL, cnp, 1508 &dnfsva.na_mtime, NULL); 1509 } else 1510 NFSUNLOCKNODE(np); 1511 } 1512 return (ENOENT); 1513 } 1514 1515 /* 1516 * Handle RENAME case... 1517 */ 1518 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1519 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1520 free(nfhp, M_NFSFH); 1521 return (EISDIR); 1522 } 1523 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1524 LK_EXCLUSIVE); 1525 if (error) 1526 return (error); 1527 newvp = NFSTOV(np); 1528 /* 1529 * If n_localmodtime >= time before RPC, then 1530 * a file modification operation, such as 1531 * VOP_SETATTR() of size, has occurred while 1532 * the Lookup RPC and acquisition of the vnode 1533 * happened. As such, the attributes might 1534 * be stale, with possibly an incorrect size. 1535 */ 1536 NFSLOCKNODE(np); 1537 if (timespecisset(&np->n_localmodtime) && 1538 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1539 NFSCL_DEBUG(4, "nfs_lookup: rename localmod " 1540 "stale attributes\n"); 1541 attrflag = 0; 1542 } 1543 NFSUNLOCKNODE(np); 1544 if (attrflag) 1545 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1546 0, 1); 1547 *vpp = newvp; 1548 return (0); 1549 } 1550 1551 if (flags & ISDOTDOT) { 1552 ltype = NFSVOPISLOCKED(dvp); 1553 error = vfs_busy(mp, MBF_NOWAIT); 1554 if (error != 0) { 1555 vfs_ref(mp); 1556 NFSVOPUNLOCK(dvp); 1557 error = vfs_busy(mp, 0); 1558 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1559 vfs_rel(mp); 1560 if (error == 0 && VN_IS_DOOMED(dvp)) { 1561 vfs_unbusy(mp); 1562 error = ENOENT; 1563 } 1564 if (error != 0) 1565 return (error); 1566 } 1567 NFSVOPUNLOCK(dvp); 1568 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1569 cnp->cn_lkflags); 1570 if (error == 0) 1571 newvp = NFSTOV(np); 1572 vfs_unbusy(mp); 1573 if (newvp != dvp) 1574 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1575 if (VN_IS_DOOMED(dvp)) { 1576 if (error == 0) { 1577 if (newvp == dvp) 1578 vrele(newvp); 1579 else 1580 vput(newvp); 1581 } 1582 error = ENOENT; 1583 } 1584 if (error != 0) 1585 return (error); 1586 if (attrflag) 1587 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1588 0, 1); 1589 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1590 free(nfhp, M_NFSFH); 1591 VREF(dvp); 1592 newvp = dvp; 1593 if (attrflag) 1594 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1595 0, 1); 1596 } else { 1597 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1598 cnp->cn_lkflags); 1599 if (error) 1600 return (error); 1601 newvp = NFSTOV(np); 1602 if (opennamed) 1603 vn_irflag_set_cond(newvp, VIRF_NAMEDATTR); 1604 /* 1605 * If n_localmodtime >= time before RPC, then 1606 * a file modification operation, such as 1607 * VOP_SETATTR() of size, has occurred while 1608 * the Lookup RPC and acquisition of the vnode 1609 * happened. As such, the attributes might 1610 * be stale, with possibly an incorrect size. 1611 */ 1612 NFSLOCKNODE(np); 1613 if (timespecisset(&np->n_localmodtime) && 1614 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1615 NFSCL_DEBUG(4, "nfs_lookup: localmod " 1616 "stale attributes\n"); 1617 attrflag = 0; 1618 } 1619 NFSUNLOCKNODE(np); 1620 if (attrflag) 1621 (void)nfscl_loadattrcache(&newvp, &nfsva, NULL, 1622 0, 1); 1623 else if ((flags & (ISLASTCN | ISOPEN)) == 1624 (ISLASTCN | ISOPEN) && 1625 !(np->n_flag & NMODIFIED)) { 1626 /* 1627 * Flush the attribute cache when opening a 1628 * leaf node to ensure that fresh attributes 1629 * are fetched in nfs_open() since we did not 1630 * fetch attributes from the LOOKUP reply. 1631 */ 1632 NFSLOCKNODE(np); 1633 np->n_attrstamp = 0; 1634 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1635 NFSUNLOCKNODE(np); 1636 } 1637 } 1638 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp && 1639 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1640 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1641 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1642 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1643 *vpp = newvp; 1644 return (0); 1645 } 1646 1647 /* 1648 * nfs read call. 1649 * Just call ncl_bioread() to do the work. 1650 */ 1651 static int 1652 nfs_read(struct vop_read_args *ap) 1653 { 1654 struct vnode *vp = ap->a_vp; 1655 1656 switch (vp->v_type) { 1657 case VREG: 1658 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1659 case VDIR: 1660 return (EISDIR); 1661 default: 1662 return (EOPNOTSUPP); 1663 } 1664 } 1665 1666 /* 1667 * nfs readlink call 1668 */ 1669 static int 1670 nfs_readlink(struct vop_readlink_args *ap) 1671 { 1672 struct vnode *vp = ap->a_vp; 1673 1674 if (vp->v_type != VLNK) 1675 return (EINVAL); 1676 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1677 } 1678 1679 /* 1680 * Do a readlink rpc. 1681 * Called by ncl_doio() from below the buffer cache. 1682 */ 1683 int 1684 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1685 { 1686 int error, ret, attrflag; 1687 struct nfsvattr nfsva; 1688 1689 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1690 &attrflag); 1691 if (attrflag) { 1692 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1693 if (ret && !error) 1694 error = ret; 1695 } 1696 if (error && NFS_ISV4(vp)) 1697 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1698 return (error); 1699 } 1700 1701 /* 1702 * nfs read rpc call 1703 * Ditto above 1704 */ 1705 int 1706 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1707 { 1708 int error, ret, attrflag; 1709 struct nfsvattr nfsva; 1710 struct nfsmount *nmp; 1711 1712 nmp = VFSTONFS(vp->v_mount); 1713 error = EIO; 1714 attrflag = 0; 1715 if (NFSHASPNFS(nmp)) 1716 error = nfscl_doiods(vp, uiop, NULL, NULL, 1717 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1718 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1719 if (error != 0 && error != EFAULT) 1720 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1721 &attrflag); 1722 if (attrflag) { 1723 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1724 if (ret && !error) 1725 error = ret; 1726 } 1727 if (error && NFS_ISV4(vp)) 1728 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1729 return (error); 1730 } 1731 1732 /* 1733 * nfs write call 1734 */ 1735 int 1736 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1737 int *iomode, int *must_commit, int called_from_strategy, int ioflag) 1738 { 1739 struct nfsvattr nfsva; 1740 int error, attrflag, ret; 1741 struct nfsmount *nmp; 1742 1743 nmp = VFSTONFS(vp->v_mount); 1744 error = EIO; 1745 attrflag = 0; 1746 if (NFSHASPNFS(nmp)) 1747 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1748 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1749 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1750 if (error != 0 && error != EFAULT) 1751 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1752 uiop->uio_td, &nfsva, &attrflag, called_from_strategy, 1753 ioflag); 1754 if (attrflag) { 1755 if (VTONFS(vp)->n_flag & ND_NFSV4) 1756 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1); 1757 else 1758 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1759 if (ret && !error) 1760 error = ret; 1761 } 1762 if (DOINGASYNC(vp)) 1763 *iomode = NFSWRITE_FILESYNC; 1764 if (error && NFS_ISV4(vp)) 1765 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1766 return (error); 1767 } 1768 1769 /* 1770 * nfs mknod rpc 1771 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1772 * mode set to specify the file type and the size field for rdev. 1773 */ 1774 static int 1775 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1776 struct vattr *vap) 1777 { 1778 struct nfsvattr nfsva, dnfsva; 1779 struct vnode *newvp = NULL; 1780 struct nfsnode *np = NULL, *dnp; 1781 struct nfsfh *nfhp; 1782 struct vattr vattr; 1783 int error = 0, attrflag, dattrflag; 1784 u_int32_t rdev; 1785 1786 if (vap->va_type == VCHR || vap->va_type == VBLK) 1787 rdev = vap->va_rdev; 1788 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1789 rdev = 0xffffffff; 1790 else 1791 return (EOPNOTSUPP); 1792 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1793 return (error); 1794 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1795 rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva, 1796 &nfsva, &nfhp, &attrflag, &dattrflag); 1797 if (!error) { 1798 if (!nfhp) 1799 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1800 cnp->cn_namelen, cnp->cn_cred, curthread, 1801 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0); 1802 if (nfhp) 1803 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1804 curthread, &np, LK_EXCLUSIVE); 1805 } 1806 if (dattrflag) 1807 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1808 if (!error) { 1809 newvp = NFSTOV(np); 1810 if (attrflag != 0) { 1811 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1812 if (error != 0) 1813 vput(newvp); 1814 } 1815 } 1816 if (!error) { 1817 *vpp = newvp; 1818 } else if (NFS_ISV4(dvp)) { 1819 error = nfscl_maperr(curthread, error, vap->va_uid, 1820 vap->va_gid); 1821 } 1822 dnp = VTONFS(dvp); 1823 NFSLOCKNODE(dnp); 1824 dnp->n_flag |= NMODIFIED; 1825 if (!dattrflag) { 1826 dnp->n_attrstamp = 0; 1827 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1828 } 1829 NFSUNLOCKNODE(dnp); 1830 return (error); 1831 } 1832 1833 /* 1834 * nfs mknod vop 1835 * just call nfs_mknodrpc() to do the work. 1836 */ 1837 /* ARGSUSED */ 1838 static int 1839 nfs_mknod(struct vop_mknod_args *ap) 1840 { 1841 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1842 } 1843 1844 static struct mtx nfs_cverf_mtx; 1845 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1846 MTX_DEF); 1847 1848 static nfsquad_t 1849 nfs_get_cverf(void) 1850 { 1851 static nfsquad_t cverf; 1852 nfsquad_t ret; 1853 static int cverf_initialized = 0; 1854 1855 mtx_lock(&nfs_cverf_mtx); 1856 if (cverf_initialized == 0) { 1857 cverf.lval[0] = arc4random(); 1858 cverf.lval[1] = arc4random(); 1859 cverf_initialized = 1; 1860 } else 1861 cverf.qval++; 1862 ret = cverf; 1863 mtx_unlock(&nfs_cverf_mtx); 1864 1865 return (ret); 1866 } 1867 1868 /* 1869 * nfs file create call 1870 */ 1871 static int 1872 nfs_create(struct vop_create_args *ap) 1873 { 1874 struct vnode *dvp = ap->a_dvp; 1875 struct vattr *vap = ap->a_vap; 1876 struct componentname *cnp = ap->a_cnp; 1877 struct nfsnode *np = NULL, *dnp; 1878 struct vnode *newvp = NULL; 1879 struct nfsmount *nmp; 1880 struct nfsvattr dnfsva, nfsva; 1881 struct nfsfh *nfhp; 1882 nfsquad_t cverf; 1883 int error = 0, attrflag, dattrflag, fmode = 0; 1884 struct vattr vattr; 1885 bool is_nameddir, needs_nameddir, opennamed; 1886 1887 /* 1888 * Oops, not for me.. 1889 */ 1890 if (vap->va_type == VSOCK) 1891 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1892 1893 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1894 return (error); 1895 if (vap->va_vaflags & VA_EXCLUSIVE) 1896 fmode |= O_EXCL; 1897 dnp = VTONFS(dvp); 1898 nmp = VFSTONFS(dvp->v_mount); 1899 needs_nameddir = false; 1900 if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp)) { 1901 opennamed = (cnp->cn_flags & (OPENNAMED | ISLASTCN)) == 1902 (OPENNAMED | ISLASTCN); 1903 is_nameddir = (vn_irflag_read(dvp) & VIRF_NAMEDDIR) != 0; 1904 if (opennamed || is_nameddir) { 1905 cnp->cn_flags &= ~MAKEENTRY; 1906 if (!is_nameddir) 1907 needs_nameddir = true; 1908 } 1909 } 1910 1911 /* 1912 * If the named attribute directory is needed, acquire it now. 1913 */ 1914 if (needs_nameddir) { 1915 KASSERT(dnp->n_v4 == NULL, ("nfs_create: O_NAMEDATTR when" 1916 " n_v4 not NULL")); 1917 error = nfs_get_namedattrdir(dvp, cnp, &newvp); 1918 if (error != 0) 1919 return (error); 1920 dvp = newvp; 1921 dnp = VTONFS(dvp); 1922 newvp = NULL; 1923 } 1924 1925 again: 1926 /* For NFSv4, wait until any remove is done. */ 1927 NFSLOCKNODE(dnp); 1928 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1929 dnp->n_flag |= NREMOVEWANT; 1930 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1931 } 1932 NFSUNLOCKNODE(dnp); 1933 1934 cverf = nfs_get_cverf(); 1935 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1936 vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva, 1937 &nfhp, &attrflag, &dattrflag); 1938 if (!error) { 1939 if (nfhp == NULL) 1940 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1941 cnp->cn_namelen, cnp->cn_cred, curthread, 1942 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0); 1943 if (nfhp != NULL) 1944 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1945 curthread, &np, LK_EXCLUSIVE); 1946 } 1947 if (dattrflag) 1948 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1949 if (!error) { 1950 newvp = NFSTOV(np); 1951 if (attrflag == 0) 1952 error = nfsrpc_getattr(newvp, cnp->cn_cred, curthread, 1953 &nfsva); 1954 if (error == 0) 1955 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1956 } 1957 if (error) { 1958 if (newvp != NULL) { 1959 vput(newvp); 1960 newvp = NULL; 1961 } 1962 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1963 error == NFSERR_NOTSUPP) { 1964 fmode &= ~O_EXCL; 1965 goto again; 1966 } 1967 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1968 if (nfscl_checksattr(vap, &nfsva)) { 1969 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1970 curthread, &nfsva, &attrflag); 1971 if (error && (vap->va_uid != (uid_t)VNOVAL || 1972 vap->va_gid != (gid_t)VNOVAL)) { 1973 /* try again without setting uid/gid */ 1974 vap->va_uid = (uid_t)VNOVAL; 1975 vap->va_gid = (uid_t)VNOVAL; 1976 error = nfsrpc_setattr(newvp, vap, NULL, 1977 cnp->cn_cred, curthread, &nfsva, &attrflag); 1978 } 1979 if (attrflag) 1980 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1981 0, 1); 1982 if (error != 0) 1983 vput(newvp); 1984 } 1985 } 1986 if (!error) { 1987 if ((cnp->cn_flags & MAKEENTRY) && attrflag) { 1988 if (dvp != newvp) 1989 cache_enter_time(dvp, newvp, cnp, 1990 &nfsva.na_ctime, NULL); 1991 else 1992 printf("nfs_create: bogus NFS server returned " 1993 "the directory as the new file object\n"); 1994 } 1995 *ap->a_vpp = newvp; 1996 } else if (NFS_ISV4(dvp)) { 1997 error = nfscl_maperr(curthread, error, vap->va_uid, 1998 vap->va_gid); 1999 } 2000 NFSLOCKNODE(dnp); 2001 dnp->n_flag |= NMODIFIED; 2002 if (!dattrflag) { 2003 dnp->n_attrstamp = 0; 2004 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2005 } 2006 NFSUNLOCKNODE(dnp); 2007 if (needs_nameddir) 2008 vput(dvp); 2009 return (error); 2010 } 2011 2012 /* 2013 * nfs file remove call 2014 * To try and make nfs semantics closer to ufs semantics, a file that has 2015 * other processes using the vnode is renamed instead of removed and then 2016 * removed later on the last close. 2017 * - If v_usecount > 1 2018 * If a rename is not already in the works 2019 * call nfs_sillyrename() to set it up 2020 * else 2021 * do the remove rpc 2022 */ 2023 static int 2024 nfs_remove(struct vop_remove_args *ap) 2025 { 2026 struct vnode *vp = ap->a_vp; 2027 struct vnode *dvp = ap->a_dvp; 2028 struct componentname *cnp = ap->a_cnp; 2029 struct nfsnode *np = VTONFS(vp); 2030 int error = 0; 2031 struct vattr vattr; 2032 struct nfsmount *nmp; 2033 2034 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 2035 if (vp->v_type == VDIR) 2036 error = EPERM; 2037 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 2038 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 2039 vattr.va_nlink > 1)) { 2040 nmp = VFSTONFS(vp->v_mount); 2041 /* 2042 * Purge the name cache so that the chance of a lookup for 2043 * the name succeeding while the remove is in progress is 2044 * minimized. Without node locking it can still happen, such 2045 * that an I/O op returns ESTALE, but since you get this if 2046 * another host removes the file.. 2047 */ 2048 cache_purge(vp); 2049 /* 2050 * throw away biocache buffers, mainly to avoid 2051 * unnecessary delayed writes later. 2052 * Flushing here would be more correct for the case 2053 * where nfs_close() did not do a flush. However, it 2054 * could be a large performance hit for some servers 2055 * and only matters when the file name being removed is 2056 * one of multiple hard links. 2057 */ 2058 if (!NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp) || 2059 (nmp->nm_flag & NFSMNT_NOCTO) == 0) 2060 error = ncl_vinvalbuf(vp, 0, curthread, 1); 2061 if (error != EINTR && error != EIO) 2062 /* Do the rpc */ 2063 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 2064 cnp->cn_namelen, cnp->cn_cred, curthread, false); 2065 /* 2066 * Kludge City: If the first reply to the remove rpc is lost.. 2067 * the reply to the retransmitted request will be ENOENT 2068 * since the file was in fact removed 2069 * Therefore, we cheat and return success. 2070 */ 2071 if (error == ENOENT) 2072 error = 0; 2073 } else if (!np->n_sillyrename) 2074 error = nfs_sillyrename(dvp, vp, cnp); 2075 NFSLOCKNODE(np); 2076 np->n_attrstamp = 0; 2077 NFSUNLOCKNODE(np); 2078 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2079 return (error); 2080 } 2081 2082 /* 2083 * nfs file remove rpc called from nfs_inactive 2084 */ 2085 int 2086 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 2087 { 2088 /* 2089 * Make sure that the directory vnode is still valid. 2090 * XXX we should lock sp->s_dvp here. 2091 */ 2092 if (sp->s_dvp->v_type == VBAD) 2093 return (0); 2094 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 2095 sp->s_cred, NULL, true)); 2096 } 2097 2098 /* 2099 * Handle the nfsremove_status reply from the RPC function. 2100 */ 2101 static void 2102 nfs_removestatus(struct vnode *vp, nfsremove_status file_status, 2103 bool silly, struct thread *td) 2104 { 2105 2106 switch (file_status) { 2107 case NLINK_ZERO: 2108 /* Get rid of any delegation. */ 2109 nfscl_delegreturnvp(vp, false, td); 2110 /* FALLTHROUGH */ 2111 case DELETED: 2112 /* Throw away buffer cache blocks. */ 2113 (void)ncl_vinvalbuf(vp, 0, td, 1); 2114 break; 2115 case VALID: 2116 /* Nothing to do, delegation is still ok. */ 2117 break; 2118 default: 2119 break; 2120 } 2121 } 2122 2123 /* 2124 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 2125 */ 2126 static int 2127 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 2128 int namelen, struct ucred *cred, struct thread *td, bool silly) 2129 { 2130 struct nfsvattr dnfsva, nfsva; 2131 struct nfsnode *dnp = VTONFS(dvp); 2132 struct nfsmount *nmp; 2133 int attrflag, error = 0, dattrflag; 2134 nfsremove_status file_status; 2135 2136 nmp = VFSTONFS(dvp->v_mount); 2137 NFSLOCKNODE(dnp); 2138 dnp->n_flag |= NREMOVEINPROG; 2139 NFSUNLOCKNODE(dnp); 2140 error = nfsrpc_remove(dvp, name, namelen, vp, &nfsva, &attrflag, 2141 &file_status, &dnfsva, &dattrflag, cred, td); 2142 NFSLOCKNODE(dnp); 2143 if ((dnp->n_flag & NREMOVEWANT)) { 2144 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 2145 NFSUNLOCKNODE(dnp); 2146 wakeup((caddr_t)dnp); 2147 } else { 2148 dnp->n_flag &= ~NREMOVEINPROG; 2149 NFSUNLOCKNODE(dnp); 2150 } 2151 2152 if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp)) { 2153 if (file_status != DELETED && attrflag != 0) 2154 (void)nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2155 if ((nmp->nm_flag & NFSMNT_NOCTO) != 0) 2156 nfs_removestatus(vp, file_status, silly, td); 2157 } 2158 2159 if (dattrflag != 0) 2160 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2161 NFSLOCKNODE(dnp); 2162 dnp->n_flag |= NMODIFIED; 2163 if (dattrflag == 0) { 2164 dnp->n_attrstamp = 0; 2165 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2166 } 2167 NFSUNLOCKNODE(dnp); 2168 if (error && NFS_ISV4(dvp)) 2169 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2170 return (error); 2171 } 2172 2173 /* 2174 * nfs file rename call 2175 */ 2176 static int 2177 nfs_rename(struct vop_rename_args *ap) 2178 { 2179 struct vnode *fvp = ap->a_fvp; 2180 struct vnode *tvp = ap->a_tvp; 2181 struct vnode *fdvp = ap->a_fdvp; 2182 struct vnode *tdvp = ap->a_tdvp; 2183 struct componentname *tcnp = ap->a_tcnp; 2184 struct componentname *fcnp = ap->a_fcnp; 2185 struct nfsnode *fnp = VTONFS(ap->a_fvp); 2186 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 2187 struct nfsv4node *newv4 = NULL; 2188 struct nfsmount *nmp; 2189 int error; 2190 2191 /* Check for cross-device rename */ 2192 if ((fvp->v_mount != tdvp->v_mount) || 2193 (tvp && (fvp->v_mount != tvp->v_mount))) { 2194 error = EXDEV; 2195 goto out; 2196 } 2197 nmp = VFSTONFS(fvp->v_mount); 2198 2199 if (fvp == tvp) { 2200 printf("nfs_rename: fvp == tvp (can't happen)\n"); 2201 error = 0; 2202 goto out; 2203 } 2204 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 2205 goto out; 2206 2207 /* 2208 * We have to flush B_DELWRI data prior to renaming 2209 * the file. If we don't, the delayed-write buffers 2210 * can be flushed out later after the file has gone stale 2211 * under NFSV3. NFSV2 does not have this problem because 2212 * ( as far as I can tell ) it flushes dirty buffers more 2213 * often. 2214 * 2215 * Skip the rename operation if the fsync fails, this can happen 2216 * due to the server's volume being full, when we pushed out data 2217 * that was written back to our cache earlier. Not checking for 2218 * this condition can result in potential (silent) data loss. 2219 */ 2220 if ((nmp->nm_flag & NFSMNT_NOCTO) == 0 || !NFSHASNFSV4(nmp) || 2221 !NFSHASNFSV4N(nmp) || nfscl_mustflush(fvp) != 0) 2222 error = VOP_FSYNC(fvp, MNT_WAIT, curthread); 2223 NFSVOPUNLOCK(fvp); 2224 if (error == 0 && tvp != NULL && ((nmp->nm_flag & NFSMNT_NOCTO) == 0 || 2225 !NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp) || 2226 nfscl_mustflush(tvp) != 0)) 2227 error = VOP_FSYNC(tvp, MNT_WAIT, curthread); 2228 if (error != 0) 2229 goto out; 2230 2231 /* 2232 * If the tvp exists and is in use, sillyrename it before doing the 2233 * rename of the new file over it. 2234 * XXX Can't sillyrename a directory. 2235 */ 2236 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 2237 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 2238 vput(tvp); 2239 tvp = NULL; 2240 } 2241 2242 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 2243 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, false, tcnp->cn_cred, 2244 curthread); 2245 2246 if (error == 0 && NFS_ISV4(tdvp)) { 2247 /* 2248 * For NFSv4, check to see if it is the same name and 2249 * replace the name, if it is different. 2250 */ 2251 newv4 = malloc( 2252 sizeof (struct nfsv4node) + 2253 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 2254 M_NFSV4NODE, M_WAITOK); 2255 NFSLOCKNODE(tdnp); 2256 NFSLOCKNODE(fnp); 2257 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 2258 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 2259 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 2260 tcnp->cn_namelen) || 2261 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 2262 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2263 tdnp->n_fhp->nfh_len))) { 2264 free(fnp->n_v4, M_NFSV4NODE); 2265 fnp->n_v4 = newv4; 2266 newv4 = NULL; 2267 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 2268 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 2269 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2270 tdnp->n_fhp->nfh_len); 2271 NFSBCOPY(tcnp->cn_nameptr, 2272 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 2273 } 2274 NFSUNLOCKNODE(tdnp); 2275 NFSUNLOCKNODE(fnp); 2276 if (newv4 != NULL) 2277 free(newv4, M_NFSV4NODE); 2278 } 2279 2280 if (fvp->v_type == VDIR) { 2281 if (tvp != NULL && tvp->v_type == VDIR) 2282 cache_purge(tdvp); 2283 cache_purge(fdvp); 2284 } 2285 2286 out: 2287 if (tdvp == tvp) 2288 vrele(tdvp); 2289 else 2290 vput(tdvp); 2291 if (tvp) 2292 vput(tvp); 2293 vrele(fdvp); 2294 vrele(fvp); 2295 /* 2296 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2297 */ 2298 if (error == ENOENT) 2299 error = 0; 2300 return (error); 2301 } 2302 2303 /* 2304 * nfs file rename rpc called from nfs_remove() above 2305 */ 2306 static int 2307 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 2308 struct sillyrename *sp) 2309 { 2310 2311 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 2312 sdvp, NULL, sp->s_name, sp->s_namlen, true, scnp->cn_cred, 2313 curthread)); 2314 } 2315 2316 /* 2317 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2318 */ 2319 static int 2320 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 2321 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 2322 int tnamelen, bool silly, struct ucred *cred, struct thread *td) 2323 { 2324 struct nfsvattr fnfsva, tnfsva, tvpnfsva; 2325 struct nfsnode *fdnp = VTONFS(fdvp); 2326 struct nfsnode *tdnp = VTONFS(tdvp); 2327 struct nfsmount *nmp; 2328 int error = 0, fattrflag, tattrflag, tvpattrflag; 2329 nfsremove_status tvp_status; 2330 2331 nmp = VFSTONFS(fdvp->v_mount); 2332 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 2333 tnameptr, tnamelen, &tvp_status, &fnfsva, &tnfsva, &fattrflag, 2334 &tattrflag, &tvpnfsva, &tvpattrflag, cred, td); 2335 NFSLOCKNODE(fdnp); 2336 fdnp->n_flag |= NMODIFIED; 2337 if (fattrflag != 0) { 2338 NFSUNLOCKNODE(fdnp); 2339 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1); 2340 } else { 2341 fdnp->n_attrstamp = 0; 2342 NFSUNLOCKNODE(fdnp); 2343 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 2344 } 2345 NFSLOCKNODE(tdnp); 2346 tdnp->n_flag |= NMODIFIED; 2347 if (tattrflag != 0) { 2348 NFSUNLOCKNODE(tdnp); 2349 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1); 2350 } else { 2351 tdnp->n_attrstamp = 0; 2352 NFSUNLOCKNODE(tdnp); 2353 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2354 } 2355 2356 if (tvp != NULL) { 2357 if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp) && 2358 (nmp->nm_flag & NFSMNT_NOCTO) != 0) 2359 nfs_removestatus(tvp, tvp_status, silly, td); 2360 if (!silly && tvpattrflag != 0) 2361 (void)nfscl_loadattrcache(&tvp, &tvpnfsva, NULL, 0, 1); 2362 } 2363 2364 if (error && NFS_ISV4(fdvp)) 2365 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2366 return (error); 2367 } 2368 2369 /* 2370 * nfs hard link create call 2371 */ 2372 static int 2373 nfs_link(struct vop_link_args *ap) 2374 { 2375 struct vnode *vp = ap->a_vp; 2376 struct vnode *tdvp = ap->a_tdvp; 2377 struct componentname *cnp = ap->a_cnp; 2378 struct nfsnode *np, *tdnp; 2379 struct nfsvattr nfsva, dnfsva; 2380 int error = 0, attrflag, dattrflag; 2381 2382 /* 2383 * Push all writes to the server, so that the attribute cache 2384 * doesn't get "out of sync" with the server. 2385 * XXX There should be a better way! 2386 */ 2387 #ifdef notnow 2388 VOP_FSYNC(vp, MNT_WAIT, curthread); 2389 #endif 2390 2391 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2392 cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, &dattrflag); 2393 tdnp = VTONFS(tdvp); 2394 NFSLOCKNODE(tdnp); 2395 tdnp->n_flag |= NMODIFIED; 2396 if (dattrflag != 0) { 2397 NFSUNLOCKNODE(tdnp); 2398 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1); 2399 } else { 2400 tdnp->n_attrstamp = 0; 2401 NFSUNLOCKNODE(tdnp); 2402 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2403 } 2404 if (attrflag) 2405 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2406 else { 2407 np = VTONFS(vp); 2408 NFSLOCKNODE(np); 2409 np->n_attrstamp = 0; 2410 NFSUNLOCKNODE(np); 2411 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2412 } 2413 /* 2414 * If negative lookup caching is enabled, I might as well 2415 * add an entry for this node. Not necessary for correctness, 2416 * but if negative caching is enabled, then the system 2417 * must care about lookup caching hit rate, so... 2418 */ 2419 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2420 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2421 if (tdvp != vp) 2422 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2423 else 2424 printf("nfs_link: bogus NFS server returned " 2425 "the directory as the new link\n"); 2426 } 2427 if (error && NFS_ISV4(vp)) 2428 error = nfscl_maperr(curthread, error, (uid_t)0, 2429 (gid_t)0); 2430 return (error); 2431 } 2432 2433 /* 2434 * nfs symbolic link create call 2435 */ 2436 static int 2437 nfs_symlink(struct vop_symlink_args *ap) 2438 { 2439 struct vnode *dvp = ap->a_dvp; 2440 struct vattr *vap = ap->a_vap; 2441 struct componentname *cnp = ap->a_cnp; 2442 struct nfsvattr nfsva, dnfsva; 2443 struct nfsfh *nfhp; 2444 struct nfsnode *np = NULL, *dnp; 2445 struct vnode *newvp = NULL; 2446 int error = 0, attrflag, dattrflag, ret; 2447 2448 vap->va_type = VLNK; 2449 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2450 ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva, 2451 &nfsva, &nfhp, &attrflag, &dattrflag); 2452 if (nfhp) { 2453 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2454 &np, LK_EXCLUSIVE); 2455 if (!ret) 2456 newvp = NFSTOV(np); 2457 else if (!error) 2458 error = ret; 2459 } 2460 if (newvp != NULL) { 2461 if (attrflag) 2462 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2463 } else if (!error) { 2464 /* 2465 * If we do not have an error and we could not extract the 2466 * newvp from the response due to the request being NFSv2, we 2467 * have to do a lookup in order to obtain a newvp to return. 2468 */ 2469 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2470 cnp->cn_cred, curthread, &np); 2471 if (!error) 2472 newvp = NFSTOV(np); 2473 } 2474 if (error) { 2475 if (newvp) 2476 vput(newvp); 2477 if (NFS_ISV4(dvp)) 2478 error = nfscl_maperr(curthread, error, 2479 vap->va_uid, vap->va_gid); 2480 } else { 2481 *ap->a_vpp = newvp; 2482 } 2483 2484 dnp = VTONFS(dvp); 2485 NFSLOCKNODE(dnp); 2486 dnp->n_flag |= NMODIFIED; 2487 if (dattrflag != 0) { 2488 NFSUNLOCKNODE(dnp); 2489 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2490 } else { 2491 dnp->n_attrstamp = 0; 2492 NFSUNLOCKNODE(dnp); 2493 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2494 } 2495 /* 2496 * If negative lookup caching is enabled, I might as well 2497 * add an entry for this node. Not necessary for correctness, 2498 * but if negative caching is enabled, then the system 2499 * must care about lookup caching hit rate, so... 2500 */ 2501 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2502 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2503 if (dvp != newvp) 2504 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2505 NULL); 2506 else 2507 printf("nfs_symlink: bogus NFS server returned " 2508 "the directory as the new file object\n"); 2509 } 2510 return (error); 2511 } 2512 2513 /* 2514 * nfs make dir call 2515 */ 2516 static int 2517 nfs_mkdir(struct vop_mkdir_args *ap) 2518 { 2519 struct vnode *dvp = ap->a_dvp; 2520 struct vattr *vap = ap->a_vap; 2521 struct componentname *cnp = ap->a_cnp; 2522 struct nfsnode *np = NULL, *dnp; 2523 struct vnode *newvp = NULL; 2524 struct vattr vattr; 2525 struct nfsfh *nfhp; 2526 struct nfsvattr nfsva, dnfsva; 2527 int error = 0, attrflag, dattrflag, ret; 2528 2529 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2530 return (error); 2531 vap->va_type = VDIR; 2532 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2533 vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, 2534 &attrflag, &dattrflag); 2535 dnp = VTONFS(dvp); 2536 NFSLOCKNODE(dnp); 2537 dnp->n_flag |= NMODIFIED; 2538 if (dattrflag != 0) { 2539 NFSUNLOCKNODE(dnp); 2540 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2541 } else { 2542 dnp->n_attrstamp = 0; 2543 NFSUNLOCKNODE(dnp); 2544 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2545 } 2546 if (nfhp) { 2547 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2548 &np, LK_EXCLUSIVE); 2549 if (!ret) { 2550 newvp = NFSTOV(np); 2551 if (attrflag) 2552 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2553 0, 1); 2554 } else if (!error) 2555 error = ret; 2556 } 2557 if (!error && newvp == NULL) { 2558 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2559 cnp->cn_cred, curthread, &np); 2560 if (!error) { 2561 newvp = NFSTOV(np); 2562 if (newvp->v_type != VDIR) 2563 error = EEXIST; 2564 } 2565 } 2566 if (error) { 2567 if (newvp) 2568 vput(newvp); 2569 if (NFS_ISV4(dvp)) 2570 error = nfscl_maperr(curthread, error, 2571 vap->va_uid, vap->va_gid); 2572 } else { 2573 /* 2574 * If negative lookup caching is enabled, I might as well 2575 * add an entry for this node. Not necessary for correctness, 2576 * but if negative caching is enabled, then the system 2577 * must care about lookup caching hit rate, so... 2578 */ 2579 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2580 (cnp->cn_flags & MAKEENTRY) && 2581 attrflag != 0 && dattrflag != 0) { 2582 if (dvp != newvp) 2583 cache_enter_time(dvp, newvp, cnp, 2584 &nfsva.na_ctime, &dnfsva.na_ctime); 2585 else 2586 printf("nfs_mkdir: bogus NFS server returned " 2587 "the directory that the directory was " 2588 "created in as the new file object\n"); 2589 } 2590 *ap->a_vpp = newvp; 2591 } 2592 return (error); 2593 } 2594 2595 /* 2596 * nfs remove directory call 2597 */ 2598 static int 2599 nfs_rmdir(struct vop_rmdir_args *ap) 2600 { 2601 struct vnode *vp = ap->a_vp; 2602 struct vnode *dvp = ap->a_dvp; 2603 struct componentname *cnp = ap->a_cnp; 2604 struct nfsnode *dnp; 2605 struct nfsvattr dnfsva; 2606 int error, dattrflag; 2607 2608 if (dvp == vp) 2609 return (EINVAL); 2610 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2611 cnp->cn_cred, curthread, &dnfsva, &dattrflag); 2612 dnp = VTONFS(dvp); 2613 NFSLOCKNODE(dnp); 2614 dnp->n_flag |= NMODIFIED; 2615 if (dattrflag != 0) { 2616 NFSUNLOCKNODE(dnp); 2617 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2618 } else { 2619 dnp->n_attrstamp = 0; 2620 NFSUNLOCKNODE(dnp); 2621 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2622 } 2623 2624 cache_purge(dvp); 2625 cache_purge(vp); 2626 if (error && NFS_ISV4(dvp)) 2627 error = nfscl_maperr(curthread, error, (uid_t)0, 2628 (gid_t)0); 2629 /* 2630 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2631 */ 2632 if (error == ENOENT) 2633 error = 0; 2634 return (error); 2635 } 2636 2637 /* 2638 * nfs readdir call 2639 */ 2640 static int 2641 nfs_readdir(struct vop_readdir_args *ap) 2642 { 2643 struct vnode *vp = ap->a_vp; 2644 struct nfsnode *np = VTONFS(vp); 2645 struct uio *uio = ap->a_uio; 2646 ssize_t tresid, left; 2647 int error = 0; 2648 struct vattr vattr; 2649 2650 if (ap->a_eofflag != NULL) 2651 *ap->a_eofflag = 0; 2652 if (vp->v_type != VDIR) 2653 return(EPERM); 2654 2655 /* 2656 * First, check for hit on the EOF offset cache 2657 */ 2658 NFSLOCKNODE(np); 2659 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2660 (np->n_flag & NMODIFIED) == 0) { 2661 NFSUNLOCKNODE(np); 2662 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2663 NFSLOCKNODE(np); 2664 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2665 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2666 NFSUNLOCKNODE(np); 2667 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2668 if (ap->a_eofflag != NULL) 2669 *ap->a_eofflag = 1; 2670 return (0); 2671 } else 2672 NFSUNLOCKNODE(np); 2673 } 2674 } else 2675 NFSUNLOCKNODE(np); 2676 2677 /* 2678 * NFS always guarantees that directory entries don't straddle 2679 * DIRBLKSIZ boundaries. As such, we need to limit the size 2680 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2681 * directory entry. 2682 */ 2683 left = uio->uio_resid % DIRBLKSIZ; 2684 if (left == uio->uio_resid) 2685 return (EINVAL); 2686 uio->uio_resid -= left; 2687 2688 /* 2689 * For readdirplus, if starting to read the directory, 2690 * purge the name cache, since it will be reloaded by 2691 * this directory read. 2692 * This removes potentially stale name cache entries. 2693 */ 2694 if (uio->uio_offset == 0 && 2695 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_RDIRPLUS) != 0) 2696 cache_purge(vp); 2697 2698 /* 2699 * Call ncl_bioread() to do the real work. 2700 */ 2701 tresid = uio->uio_resid; 2702 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2703 2704 if (!error && uio->uio_resid == tresid) { 2705 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2706 if (ap->a_eofflag != NULL) 2707 *ap->a_eofflag = 1; 2708 } 2709 2710 /* Add the partial DIRBLKSIZ (left) back in. */ 2711 uio->uio_resid += left; 2712 return (error); 2713 } 2714 2715 /* 2716 * Readdir rpc call. 2717 * Called from below the buffer cache by ncl_doio(). 2718 */ 2719 int 2720 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2721 struct thread *td) 2722 { 2723 struct nfsvattr nfsva; 2724 nfsuint64 *cookiep, cookie; 2725 struct nfsnode *dnp = VTONFS(vp); 2726 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2727 int error = 0, eof, attrflag; 2728 2729 KASSERT(uiop->uio_iovcnt == 1 && 2730 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2731 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2732 ("nfs readdirrpc bad uio")); 2733 2734 /* 2735 * If there is no cookie, assume directory was stale. 2736 */ 2737 ncl_dircookie_lock(dnp); 2738 NFSUNLOCKNODE(dnp); 2739 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2740 if (cookiep) { 2741 cookie = *cookiep; 2742 ncl_dircookie_unlock(dnp); 2743 } else { 2744 ncl_dircookie_unlock(dnp); 2745 return (NFSERR_BAD_COOKIE); 2746 } 2747 2748 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2749 (void)ncl_fsinfo(nmp, vp, cred, td); 2750 2751 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2752 &attrflag, &eof); 2753 if (attrflag) 2754 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2755 2756 if (!error) { 2757 /* 2758 * We are now either at the end of the directory or have filled 2759 * the block. 2760 */ 2761 if (eof) { 2762 NFSLOCKNODE(dnp); 2763 dnp->n_direofoffset = uiop->uio_offset; 2764 NFSUNLOCKNODE(dnp); 2765 } else { 2766 if (uiop->uio_resid > 0) 2767 printf("EEK! readdirrpc resid > 0\n"); 2768 ncl_dircookie_lock(dnp); 2769 NFSUNLOCKNODE(dnp); 2770 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2771 *cookiep = cookie; 2772 ncl_dircookie_unlock(dnp); 2773 } 2774 } else if (NFS_ISV4(vp)) { 2775 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2776 } 2777 return (error); 2778 } 2779 2780 /* 2781 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2782 */ 2783 int 2784 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2785 struct thread *td) 2786 { 2787 struct nfsvattr nfsva; 2788 nfsuint64 *cookiep, cookie; 2789 struct nfsnode *dnp = VTONFS(vp); 2790 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2791 int error = 0, attrflag, eof; 2792 2793 KASSERT(uiop->uio_iovcnt == 1 && 2794 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2795 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2796 ("nfs readdirplusrpc bad uio")); 2797 2798 /* 2799 * If there is no cookie, assume directory was stale. 2800 */ 2801 ncl_dircookie_lock(dnp); 2802 NFSUNLOCKNODE(dnp); 2803 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2804 if (cookiep) { 2805 cookie = *cookiep; 2806 ncl_dircookie_unlock(dnp); 2807 } else { 2808 ncl_dircookie_unlock(dnp); 2809 return (NFSERR_BAD_COOKIE); 2810 } 2811 2812 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2813 (void)ncl_fsinfo(nmp, vp, cred, td); 2814 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2815 &attrflag, &eof); 2816 if (attrflag) 2817 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2818 2819 if (!error) { 2820 /* 2821 * We are now either at end of the directory or have filled the 2822 * the block. 2823 */ 2824 if (eof) { 2825 NFSLOCKNODE(dnp); 2826 dnp->n_direofoffset = uiop->uio_offset; 2827 NFSUNLOCKNODE(dnp); 2828 } else { 2829 if (uiop->uio_resid > 0) 2830 printf("EEK! readdirplusrpc resid > 0\n"); 2831 ncl_dircookie_lock(dnp); 2832 NFSUNLOCKNODE(dnp); 2833 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2834 *cookiep = cookie; 2835 ncl_dircookie_unlock(dnp); 2836 } 2837 } else if (NFS_ISV4(vp)) { 2838 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2839 } 2840 return (error); 2841 } 2842 2843 /* 2844 * Silly rename. To make the NFS filesystem that is stateless look a little 2845 * more like the "ufs" a remove of an active vnode is translated to a rename 2846 * to a funny looking filename that is removed by nfs_inactive on the 2847 * nfsnode. There is the potential for another process on a different client 2848 * to create the same funny name between the nfs_lookitup() fails and the 2849 * nfs_rename() completes, but... 2850 */ 2851 static int 2852 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2853 { 2854 struct sillyrename *sp; 2855 struct nfsnode *np; 2856 int error; 2857 short pid; 2858 unsigned int lticks; 2859 2860 cache_purge(dvp); 2861 np = VTONFS(vp); 2862 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2863 sp = malloc(sizeof (struct sillyrename), 2864 M_NEWNFSREQ, M_WAITOK); 2865 sp->s_cred = crhold(cnp->cn_cred); 2866 sp->s_dvp = dvp; 2867 VREF(dvp); 2868 2869 /* 2870 * Fudge together a funny name. 2871 * Changing the format of the funny name to accommodate more 2872 * sillynames per directory. 2873 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2874 * CPU ticks since boot. 2875 */ 2876 pid = curthread->td_proc->p_pid; 2877 lticks = (unsigned int)ticks; 2878 for ( ; ; ) { 2879 sp->s_namlen = sprintf(sp->s_name, 2880 ".nfs.%08x.%04x4.4", lticks, 2881 pid); 2882 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2883 curthread, NULL)) 2884 break; 2885 lticks++; 2886 } 2887 error = nfs_renameit(dvp, vp, cnp, sp); 2888 if (error) 2889 goto bad; 2890 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2891 curthread, &np); 2892 np->n_sillyrename = sp; 2893 return (0); 2894 bad: 2895 vrele(sp->s_dvp); 2896 crfree(sp->s_cred); 2897 free(sp, M_NEWNFSREQ); 2898 return (error); 2899 } 2900 2901 /* 2902 * Look up a file name and optionally either update the file handle or 2903 * allocate an nfsnode, depending on the value of npp. 2904 * npp == NULL --> just do the lookup 2905 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2906 * handled too 2907 * *npp != NULL --> update the file handle in the vnode 2908 */ 2909 static int 2910 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2911 struct thread *td, struct nfsnode **npp) 2912 { 2913 struct vnode *newvp = NULL, *vp; 2914 struct nfsnode *np, *dnp = VTONFS(dvp); 2915 struct nfsfh *nfhp, *onfhp; 2916 struct nfsvattr nfsva, dnfsva; 2917 struct componentname cn; 2918 int error = 0, attrflag, dattrflag; 2919 u_int hash; 2920 struct timespec ts; 2921 2922 nanouptime(&ts); 2923 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2924 &nfhp, &attrflag, &dattrflag, 0); 2925 if (dattrflag) 2926 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2927 if (npp && !error) { 2928 if (*npp != NULL) { 2929 np = *npp; 2930 vp = NFSTOV(np); 2931 /* 2932 * For NFSv4, check to see if it is the same name and 2933 * replace the name, if it is different. 2934 */ 2935 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2936 (np->n_v4->n4_namelen != len || 2937 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2938 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2939 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2940 dnp->n_fhp->nfh_len))) { 2941 free(np->n_v4, M_NFSV4NODE); 2942 np->n_v4 = malloc( 2943 sizeof (struct nfsv4node) + 2944 dnp->n_fhp->nfh_len + len - 1, 2945 M_NFSV4NODE, M_WAITOK); 2946 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2947 np->n_v4->n4_namelen = len; 2948 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2949 dnp->n_fhp->nfh_len); 2950 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2951 } 2952 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2953 FNV1_32_INIT); 2954 onfhp = np->n_fhp; 2955 /* 2956 * Rehash node for new file handle. 2957 */ 2958 vfs_hash_rehash(vp, hash); 2959 np->n_fhp = nfhp; 2960 if (onfhp != NULL) 2961 free(onfhp, M_NFSFH); 2962 newvp = NFSTOV(np); 2963 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2964 free(nfhp, M_NFSFH); 2965 VREF(dvp); 2966 newvp = dvp; 2967 } else { 2968 cn.cn_nameptr = name; 2969 cn.cn_namelen = len; 2970 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2971 &np, LK_EXCLUSIVE); 2972 if (error) 2973 return (error); 2974 newvp = NFSTOV(np); 2975 /* 2976 * If n_localmodtime >= time before RPC, then 2977 * a file modification operation, such as 2978 * VOP_SETATTR() of size, has occurred while 2979 * the Lookup RPC and acquisition of the vnode 2980 * happened. As such, the attributes might 2981 * be stale, with possibly an incorrect size. 2982 */ 2983 NFSLOCKNODE(np); 2984 if (timespecisset(&np->n_localmodtime) && 2985 timespeccmp(&np->n_localmodtime, &ts, >=)) { 2986 NFSCL_DEBUG(4, "nfs_lookitup: localmod " 2987 "stale attributes\n"); 2988 attrflag = 0; 2989 } 2990 NFSUNLOCKNODE(np); 2991 } 2992 if (!attrflag && *npp == NULL) { 2993 if (newvp == dvp) 2994 vrele(newvp); 2995 else 2996 vput(newvp); 2997 return (ENOENT); 2998 } 2999 if (attrflag) 3000 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 3001 } 3002 if (npp && *npp == NULL) { 3003 if (error) { 3004 if (newvp) { 3005 if (newvp == dvp) 3006 vrele(newvp); 3007 else 3008 vput(newvp); 3009 } 3010 } else 3011 *npp = np; 3012 } 3013 if (error && NFS_ISV4(dvp)) 3014 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3015 return (error); 3016 } 3017 3018 /* 3019 * Nfs Version 3 and 4 commit rpc 3020 */ 3021 int 3022 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 3023 struct thread *td) 3024 { 3025 struct nfsvattr nfsva; 3026 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3027 struct nfsnode *np; 3028 struct uio uio; 3029 int error, attrflag; 3030 3031 np = VTONFS(vp); 3032 error = EIO; 3033 attrflag = 0; 3034 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 3035 uio.uio_offset = offset; 3036 uio.uio_resid = cnt; 3037 error = nfscl_doiods(vp, &uio, NULL, NULL, 3038 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 3039 if (error != 0) { 3040 NFSLOCKNODE(np); 3041 np->n_flag &= ~NDSCOMMIT; 3042 NFSUNLOCKNODE(np); 3043 } 3044 } 3045 if (error != 0) { 3046 mtx_lock(&nmp->nm_mtx); 3047 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 3048 mtx_unlock(&nmp->nm_mtx); 3049 return (0); 3050 } 3051 mtx_unlock(&nmp->nm_mtx); 3052 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 3053 &attrflag); 3054 } 3055 if (attrflag != 0) 3056 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3057 if (error != 0 && NFS_ISV4(vp)) 3058 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3059 return (error); 3060 } 3061 3062 /* 3063 * Strategy routine. 3064 * For async requests when nfsiod(s) are running, queue the request by 3065 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 3066 * request. 3067 */ 3068 static int 3069 nfs_strategy(struct vop_strategy_args *ap) 3070 { 3071 struct buf *bp; 3072 struct vnode *vp; 3073 struct ucred *cr; 3074 3075 bp = ap->a_bp; 3076 vp = ap->a_vp; 3077 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 3078 KASSERT(!(bp->b_flags & B_DONE), 3079 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 3080 3081 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 3082 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 3083 DEV_BSIZE); 3084 if (bp->b_iocmd == BIO_READ) 3085 cr = bp->b_rcred; 3086 else 3087 cr = bp->b_wcred; 3088 3089 /* 3090 * If the op is asynchronous and an i/o daemon is waiting 3091 * queue the request, wake it up and wait for completion 3092 * otherwise just do it ourselves. 3093 */ 3094 if ((bp->b_flags & B_ASYNC) == 0 || 3095 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 3096 (void) ncl_doio(vp, bp, cr, curthread, 1); 3097 return (0); 3098 } 3099 3100 /* 3101 * fsync vnode op. Just call ncl_flush() with commit == 1. 3102 */ 3103 /* ARGSUSED */ 3104 static int 3105 nfs_fsync(struct vop_fsync_args *ap) 3106 { 3107 3108 if (ap->a_vp->v_type != VREG) { 3109 /* 3110 * For NFS, metadata is changed synchronously on the server, 3111 * so there is nothing to flush. Also, ncl_flush() clears 3112 * the NMODIFIED flag and that shouldn't be done here for 3113 * directories. 3114 */ 3115 return (0); 3116 } 3117 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 3118 } 3119 3120 /* 3121 * Flush all the blocks associated with a vnode. 3122 * Walk through the buffer pool and push any dirty pages 3123 * associated with the vnode. 3124 * If the called_from_renewthread argument is TRUE, it has been called 3125 * from the NFSv4 renew thread and, as such, cannot block indefinitely 3126 * waiting for a buffer write to complete. 3127 */ 3128 int 3129 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 3130 int commit, int called_from_renewthread) 3131 { 3132 struct nfsnode *np = VTONFS(vp); 3133 struct buf *bp; 3134 int i; 3135 struct buf *nbp; 3136 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3137 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 3138 int passone = 1, trycnt = 0; 3139 u_quad_t off, endoff, toff; 3140 struct ucred* wcred = NULL; 3141 struct buf **bvec = NULL; 3142 struct bufobj *bo; 3143 #ifndef NFS_COMMITBVECSIZ 3144 #define NFS_COMMITBVECSIZ 20 3145 #endif 3146 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 3147 u_int bvecsize = 0, bveccount; 3148 struct timespec ts; 3149 3150 if (called_from_renewthread != 0) 3151 slptimeo = hz; 3152 if (nmp->nm_flag & NFSMNT_INT) 3153 slpflag = PCATCH; 3154 if (!commit) 3155 passone = 0; 3156 bo = &vp->v_bufobj; 3157 /* 3158 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 3159 * server, but has not been committed to stable storage on the server 3160 * yet. On the first pass, the byte range is worked out and the commit 3161 * rpc is done. On the second pass, bwrite() is called to do the 3162 * job. 3163 */ 3164 again: 3165 off = (u_quad_t)-1; 3166 endoff = 0; 3167 bvecpos = 0; 3168 if (NFS_ISV34(vp) && commit) { 3169 if (bvec != NULL && bvec != bvec_on_stack) 3170 free(bvec, M_TEMP); 3171 /* 3172 * Count up how many buffers waiting for a commit. 3173 */ 3174 bveccount = 0; 3175 BO_LOCK(bo); 3176 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3177 if (!BUF_ISLOCKED(bp) && 3178 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 3179 == (B_DELWRI | B_NEEDCOMMIT)) 3180 bveccount++; 3181 } 3182 /* 3183 * Allocate space to remember the list of bufs to commit. It is 3184 * important to use M_NOWAIT here to avoid a race with nfs_write. 3185 * If we can't get memory (for whatever reason), we will end up 3186 * committing the buffers one-by-one in the loop below. 3187 */ 3188 if (bveccount > NFS_COMMITBVECSIZ) { 3189 /* 3190 * Release the vnode interlock to avoid a lock 3191 * order reversal. 3192 */ 3193 BO_UNLOCK(bo); 3194 bvec = (struct buf **) 3195 malloc(bveccount * sizeof(struct buf *), 3196 M_TEMP, M_NOWAIT); 3197 BO_LOCK(bo); 3198 if (bvec == NULL) { 3199 bvec = bvec_on_stack; 3200 bvecsize = NFS_COMMITBVECSIZ; 3201 } else 3202 bvecsize = bveccount; 3203 } else { 3204 bvec = bvec_on_stack; 3205 bvecsize = NFS_COMMITBVECSIZ; 3206 } 3207 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3208 if (bvecpos >= bvecsize) 3209 break; 3210 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3211 nbp = TAILQ_NEXT(bp, b_bobufs); 3212 continue; 3213 } 3214 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3215 (B_DELWRI | B_NEEDCOMMIT)) { 3216 BUF_UNLOCK(bp); 3217 nbp = TAILQ_NEXT(bp, b_bobufs); 3218 continue; 3219 } 3220 BO_UNLOCK(bo); 3221 bremfree(bp); 3222 /* 3223 * Work out if all buffers are using the same cred 3224 * so we can deal with them all with one commit. 3225 * 3226 * NOTE: we are not clearing B_DONE here, so we have 3227 * to do it later on in this routine if we intend to 3228 * initiate I/O on the bp. 3229 * 3230 * Note: to avoid loopback deadlocks, we do not 3231 * assign b_runningbufspace. 3232 */ 3233 if (wcred == NULL) 3234 wcred = bp->b_wcred; 3235 else if (wcred != bp->b_wcred) 3236 wcred = NOCRED; 3237 vfs_busy_pages(bp, 0); 3238 3239 BO_LOCK(bo); 3240 /* 3241 * bp is protected by being locked, but nbp is not 3242 * and vfs_busy_pages() may sleep. We have to 3243 * recalculate nbp. 3244 */ 3245 nbp = TAILQ_NEXT(bp, b_bobufs); 3246 3247 /* 3248 * A list of these buffers is kept so that the 3249 * second loop knows which buffers have actually 3250 * been committed. This is necessary, since there 3251 * may be a race between the commit rpc and new 3252 * uncommitted writes on the file. 3253 */ 3254 bvec[bvecpos++] = bp; 3255 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3256 bp->b_dirtyoff; 3257 if (toff < off) 3258 off = toff; 3259 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3260 if (toff > endoff) 3261 endoff = toff; 3262 } 3263 BO_UNLOCK(bo); 3264 } 3265 if (bvecpos > 0) { 3266 /* 3267 * Commit data on the server, as required. 3268 * If all bufs are using the same wcred, then use that with 3269 * one call for all of them, otherwise commit each one 3270 * separately. 3271 */ 3272 if (wcred != NOCRED) 3273 retv = ncl_commit(vp, off, (int)(endoff - off), 3274 wcred, td); 3275 else { 3276 retv = 0; 3277 for (i = 0; i < bvecpos; i++) { 3278 off_t off, size; 3279 bp = bvec[i]; 3280 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3281 bp->b_dirtyoff; 3282 size = (u_quad_t)(bp->b_dirtyend 3283 - bp->b_dirtyoff); 3284 retv = ncl_commit(vp, off, (int)size, 3285 bp->b_wcred, td); 3286 if (retv) break; 3287 } 3288 } 3289 3290 if (retv == NFSERR_STALEWRITEVERF) 3291 ncl_clearcommit(vp->v_mount); 3292 3293 /* 3294 * Now, either mark the blocks I/O done or mark the 3295 * blocks dirty, depending on whether the commit 3296 * succeeded. 3297 */ 3298 for (i = 0; i < bvecpos; i++) { 3299 bp = bvec[i]; 3300 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3301 if (!NFSCL_FORCEDISM(vp->v_mount) && retv) { 3302 /* 3303 * Error, leave B_DELWRI intact 3304 */ 3305 vfs_unbusy_pages(bp); 3306 brelse(bp); 3307 } else { 3308 /* 3309 * Success, remove B_DELWRI ( bundirty() ). 3310 * 3311 * b_dirtyoff/b_dirtyend seem to be NFS 3312 * specific. We should probably move that 3313 * into bundirty(). XXX 3314 */ 3315 bufobj_wref(bo); 3316 bp->b_flags |= B_ASYNC; 3317 bundirty(bp); 3318 bp->b_flags &= ~B_DONE; 3319 bp->b_ioflags &= ~BIO_ERROR; 3320 bp->b_dirtyoff = bp->b_dirtyend = 0; 3321 bufdone(bp); 3322 } 3323 } 3324 } 3325 3326 /* 3327 * Start/do any write(s) that are required. 3328 */ 3329 loop: 3330 BO_LOCK(bo); 3331 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3332 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3333 if (waitfor != MNT_WAIT || passone) 3334 continue; 3335 3336 error = BUF_TIMELOCK(bp, 3337 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 3338 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 3339 if (error == 0) { 3340 BUF_UNLOCK(bp); 3341 goto loop; 3342 } 3343 if (error == ENOLCK) { 3344 error = 0; 3345 goto loop; 3346 } 3347 if (called_from_renewthread != 0) { 3348 /* 3349 * Return EIO so the flush will be retried 3350 * later. 3351 */ 3352 error = EIO; 3353 goto done; 3354 } 3355 if (newnfs_sigintr(nmp, td)) { 3356 error = EINTR; 3357 goto done; 3358 } 3359 if (slpflag == PCATCH) { 3360 slpflag = 0; 3361 slptimeo = 2 * hz; 3362 } 3363 goto loop; 3364 } 3365 if ((bp->b_flags & B_DELWRI) == 0) 3366 panic("nfs_fsync: not dirty"); 3367 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3368 BUF_UNLOCK(bp); 3369 continue; 3370 } 3371 BO_UNLOCK(bo); 3372 bremfree(bp); 3373 bp->b_flags |= B_ASYNC; 3374 bwrite(bp); 3375 if (newnfs_sigintr(nmp, td)) { 3376 error = EINTR; 3377 goto done; 3378 } 3379 goto loop; 3380 } 3381 if (passone) { 3382 passone = 0; 3383 BO_UNLOCK(bo); 3384 goto again; 3385 } 3386 if (waitfor == MNT_WAIT) { 3387 while (bo->bo_numoutput) { 3388 error = bufobj_wwait(bo, slpflag, slptimeo); 3389 if (error) { 3390 BO_UNLOCK(bo); 3391 if (called_from_renewthread != 0) { 3392 /* 3393 * Return EIO so that the flush will be 3394 * retried later. 3395 */ 3396 error = EIO; 3397 goto done; 3398 } 3399 error = newnfs_sigintr(nmp, td); 3400 if (error) 3401 goto done; 3402 if (slpflag == PCATCH) { 3403 slpflag = 0; 3404 slptimeo = 2 * hz; 3405 } 3406 BO_LOCK(bo); 3407 } 3408 } 3409 if (bo->bo_dirty.bv_cnt != 0 && commit) { 3410 BO_UNLOCK(bo); 3411 goto loop; 3412 } 3413 /* 3414 * Wait for all the async IO requests to drain 3415 */ 3416 BO_UNLOCK(bo); 3417 } else 3418 BO_UNLOCK(bo); 3419 if (NFSHASPNFS(nmp)) { 3420 nfscl_layoutcommit(vp, td); 3421 /* 3422 * Invalidate the attribute cache, since writes to a DS 3423 * won't update the size attribute. 3424 */ 3425 NFSLOCKNODE(np); 3426 np->n_attrstamp = 0; 3427 } else 3428 NFSLOCKNODE(np); 3429 if (np->n_flag & NWRITEERR) { 3430 error = np->n_error; 3431 np->n_flag &= ~NWRITEERR; 3432 } 3433 if (commit && bo->bo_dirty.bv_cnt == 0 && 3434 bo->bo_numoutput == 0) 3435 np->n_flag &= ~NMODIFIED; 3436 NFSUNLOCKNODE(np); 3437 done: 3438 if (bvec != NULL && bvec != bvec_on_stack) 3439 free(bvec, M_TEMP); 3440 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 3441 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0)) { 3442 if (trycnt++ < 5) { 3443 /* try, try again... */ 3444 passone = 1; 3445 wcred = NULL; 3446 bvec = NULL; 3447 bvecsize = 0; 3448 goto again; 3449 } 3450 vn_printf(vp, "ncl_flush failed"); 3451 error = called_from_renewthread != 0 ? EIO : EBUSY; 3452 } 3453 if (error == 0) { 3454 nanouptime(&ts); 3455 NFSLOCKNODE(np); 3456 np->n_localmodtime = ts; 3457 NFSUNLOCKNODE(np); 3458 } 3459 return (error); 3460 } 3461 3462 /* 3463 * NFS advisory byte-level locks. 3464 */ 3465 static int 3466 nfs_advlock(struct vop_advlock_args *ap) 3467 { 3468 struct vnode *vp = ap->a_vp; 3469 struct ucred *cred; 3470 struct nfsnode *np = VTONFS(ap->a_vp); 3471 struct proc *p = (struct proc *)ap->a_id; 3472 struct thread *td = curthread; /* XXX */ 3473 struct vattr va; 3474 int ret, error; 3475 u_quad_t size; 3476 struct nfsmount *nmp; 3477 3478 error = NFSVOPLOCK(vp, LK_SHARED); 3479 if (error != 0) 3480 return (EBADF); 3481 nmp = VFSTONFS(vp->v_mount); 3482 if (!NFS_ISV4(vp) || (nmp->nm_flag & NFSMNT_NOLOCKD) != 0) { 3483 if ((nmp->nm_flag & NFSMNT_NOLOCKD) != 0) { 3484 size = np->n_size; 3485 NFSVOPUNLOCK(vp); 3486 error = lf_advlock(ap, &(vp->v_lockf), size); 3487 } else { 3488 if (nfs_advlock_p != NULL) 3489 error = nfs_advlock_p(ap); 3490 else { 3491 NFSVOPUNLOCK(vp); 3492 error = ENOLCK; 3493 } 3494 } 3495 if (error == 0 && ap->a_op == F_SETLK) { 3496 error = NFSVOPLOCK(vp, LK_SHARED); 3497 if (error == 0) { 3498 /* Mark that a file lock has been acquired. */ 3499 NFSLOCKNODE(np); 3500 np->n_flag |= NHASBEENLOCKED; 3501 NFSUNLOCKNODE(np); 3502 NFSVOPUNLOCK(vp); 3503 } 3504 } 3505 return (error); 3506 } else if ((ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3507 if (vp->v_type != VREG) { 3508 error = EINVAL; 3509 goto out; 3510 } 3511 if ((ap->a_flags & F_POSIX) != 0) 3512 cred = p->p_ucred; 3513 else 3514 cred = td->td_ucred; 3515 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 3516 if (VN_IS_DOOMED(vp)) { 3517 error = EBADF; 3518 goto out; 3519 } 3520 3521 /* 3522 * If this is unlocking a write locked region, flush and 3523 * commit them before unlocking. This is required by 3524 * RFC3530 Sec. 9.3.2. 3525 */ 3526 if (ap->a_op == F_UNLCK && 3527 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3528 ap->a_flags)) 3529 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3530 3531 /* 3532 * Mark NFS node as might have acquired a lock. 3533 * This is separate from NHASBEENLOCKED, because it must 3534 * be done before the nfsrpc_advlock() call, which might 3535 * add a nfscllock structure to the client state. 3536 * It is used to check for the case where a nfscllock 3537 * state structure cannot exist for the file. 3538 * Only done for "oneopenown" NFSv4.1/4.2 mounts. 3539 */ 3540 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) { 3541 NFSLOCKNODE(np); 3542 np->n_flag |= NMIGHTBELOCKED; 3543 NFSUNLOCKNODE(np); 3544 } 3545 3546 /* 3547 * Loop around doing the lock op, while a blocking lock 3548 * must wait for the lock op to succeed. 3549 */ 3550 do { 3551 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3552 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3553 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3554 ap->a_op == F_SETLK) { 3555 NFSVOPUNLOCK(vp); 3556 error = nfs_catnap(PZERO | PCATCH, ret, 3557 "ncladvl"); 3558 if (error) 3559 return (EINTR); 3560 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3561 if (VN_IS_DOOMED(vp)) { 3562 error = EBADF; 3563 goto out; 3564 } 3565 } 3566 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3567 ap->a_op == F_SETLK); 3568 if (ret == NFSERR_DENIED) { 3569 error = EAGAIN; 3570 goto out; 3571 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3572 error = ret; 3573 goto out; 3574 } else if (ret != 0) { 3575 error = EACCES; 3576 goto out; 3577 } 3578 3579 /* 3580 * Now, if we just got a lock, invalidate data in the buffer 3581 * cache, as required, so that the coherency conforms with 3582 * RFC3530 Sec. 9.3.2. 3583 */ 3584 if (ap->a_op == F_SETLK) { 3585 if ((np->n_flag & NMODIFIED) == 0) { 3586 np->n_attrstamp = 0; 3587 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3588 ret = VOP_GETATTR(vp, &va, cred); 3589 } 3590 if ((np->n_flag & NMODIFIED) || ret || 3591 np->n_change != va.va_filerev) { 3592 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3593 np->n_attrstamp = 0; 3594 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3595 ret = VOP_GETATTR(vp, &va, cred); 3596 if (!ret) { 3597 np->n_mtime = va.va_mtime; 3598 np->n_change = va.va_filerev; 3599 } 3600 } 3601 /* Mark that a file lock has been acquired. */ 3602 NFSLOCKNODE(np); 3603 np->n_flag |= NHASBEENLOCKED; 3604 NFSUNLOCKNODE(np); 3605 } 3606 } else 3607 error = EOPNOTSUPP; 3608 out: 3609 NFSVOPUNLOCK(vp); 3610 return (error); 3611 } 3612 3613 /* 3614 * NFS advisory byte-level locks. 3615 */ 3616 static int 3617 nfs_advlockasync(struct vop_advlockasync_args *ap) 3618 { 3619 struct vnode *vp = ap->a_vp; 3620 u_quad_t size; 3621 int error; 3622 3623 error = NFSVOPLOCK(vp, LK_SHARED); 3624 if (error) 3625 return (error); 3626 if (NFS_ISV4(vp)) { 3627 NFSVOPUNLOCK(vp); 3628 return (EOPNOTSUPP); 3629 } 3630 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3631 size = VTONFS(vp)->n_size; 3632 NFSVOPUNLOCK(vp); 3633 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3634 } else { 3635 NFSVOPUNLOCK(vp); 3636 error = EOPNOTSUPP; 3637 } 3638 return (error); 3639 } 3640 3641 /* 3642 * Print out the contents of an nfsnode. 3643 */ 3644 static int 3645 nfs_print(struct vop_print_args *ap) 3646 { 3647 struct vnode *vp = ap->a_vp; 3648 struct nfsnode *np = VTONFS(vp); 3649 3650 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3651 (uintmax_t)np->n_vattr.na_fsid); 3652 if (vp->v_type == VFIFO) 3653 fifo_printinfo(vp); 3654 printf("\n"); 3655 return (0); 3656 } 3657 3658 /* 3659 * nfs special file access vnode op. 3660 * Essentially just get vattr and then imitate iaccess() since the device is 3661 * local to the client. 3662 */ 3663 static int 3664 nfsspec_access(struct vop_access_args *ap) 3665 { 3666 struct vattr *vap; 3667 struct ucred *cred = ap->a_cred; 3668 struct vnode *vp = ap->a_vp; 3669 accmode_t accmode = ap->a_accmode; 3670 struct vattr vattr; 3671 int error; 3672 3673 /* 3674 * Disallow write attempts on filesystems mounted read-only; 3675 * unless the file is a socket, fifo, or a block or character 3676 * device resident on the filesystem. 3677 */ 3678 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3679 switch (vp->v_type) { 3680 case VREG: 3681 case VDIR: 3682 case VLNK: 3683 return (EROFS); 3684 default: 3685 break; 3686 } 3687 } 3688 vap = &vattr; 3689 error = VOP_GETATTR(vp, vap, cred); 3690 if (error) 3691 goto out; 3692 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3693 accmode, cred); 3694 out: 3695 return error; 3696 } 3697 3698 /* 3699 * Read wrapper for fifos. 3700 */ 3701 static int 3702 nfsfifo_read(struct vop_read_args *ap) 3703 { 3704 struct nfsnode *np = VTONFS(ap->a_vp); 3705 int error; 3706 3707 /* 3708 * Set access flag. 3709 */ 3710 NFSLOCKNODE(np); 3711 np->n_flag |= NACC; 3712 vfs_timestamp(&np->n_atim); 3713 NFSUNLOCKNODE(np); 3714 error = fifo_specops.vop_read(ap); 3715 return error; 3716 } 3717 3718 /* 3719 * Write wrapper for fifos. 3720 */ 3721 static int 3722 nfsfifo_write(struct vop_write_args *ap) 3723 { 3724 struct nfsnode *np = VTONFS(ap->a_vp); 3725 3726 /* 3727 * Set update flag. 3728 */ 3729 NFSLOCKNODE(np); 3730 np->n_flag |= NUPD; 3731 vfs_timestamp(&np->n_mtim); 3732 NFSUNLOCKNODE(np); 3733 return(fifo_specops.vop_write(ap)); 3734 } 3735 3736 /* 3737 * Close wrapper for fifos. 3738 * 3739 * Update the times on the nfsnode then do fifo close. 3740 */ 3741 static int 3742 nfsfifo_close(struct vop_close_args *ap) 3743 { 3744 struct vnode *vp = ap->a_vp; 3745 struct nfsnode *np = VTONFS(vp); 3746 struct vattr vattr; 3747 struct timespec ts; 3748 3749 NFSLOCKNODE(np); 3750 if (np->n_flag & (NACC | NUPD)) { 3751 vfs_timestamp(&ts); 3752 if (np->n_flag & NACC) 3753 np->n_atim = ts; 3754 if (np->n_flag & NUPD) 3755 np->n_mtim = ts; 3756 np->n_flag |= NCHG; 3757 if (vrefcnt(vp) == 1 && 3758 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3759 VATTR_NULL(&vattr); 3760 if (np->n_flag & NACC) 3761 vattr.va_atime = np->n_atim; 3762 if (np->n_flag & NUPD) 3763 vattr.va_mtime = np->n_mtim; 3764 NFSUNLOCKNODE(np); 3765 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3766 goto out; 3767 } 3768 } 3769 NFSUNLOCKNODE(np); 3770 out: 3771 return (fifo_specops.vop_close(ap)); 3772 } 3773 3774 static int 3775 nfs_getacl(struct vop_getacl_args *ap) 3776 { 3777 int error; 3778 3779 if (ap->a_type != ACL_TYPE_NFS4) 3780 return (EOPNOTSUPP); 3781 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp); 3782 if (error > NFSERR_STALE) { 3783 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3784 error = EPERM; 3785 } 3786 return (error); 3787 } 3788 3789 static int 3790 nfs_setacl(struct vop_setacl_args *ap) 3791 { 3792 int error; 3793 3794 if (ap->a_type != ACL_TYPE_NFS4) 3795 return (EOPNOTSUPP); 3796 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp); 3797 if (error > NFSERR_STALE) { 3798 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3799 error = EPERM; 3800 } 3801 return (error); 3802 } 3803 3804 /* 3805 * VOP_ADVISE for NFS. 3806 * Just return 0 for any errors, since it is just a hint. 3807 */ 3808 static int 3809 nfs_advise(struct vop_advise_args *ap) 3810 { 3811 struct thread *td = curthread; 3812 struct nfsmount *nmp; 3813 uint64_t len; 3814 int error; 3815 3816 /* 3817 * First do vop_stdadvise() to handle the buffer cache. 3818 */ 3819 error = vop_stdadvise(ap); 3820 if (error != 0) 3821 return (error); 3822 if (ap->a_start < 0 || ap->a_end < 0) 3823 return (0); 3824 if (ap->a_end == OFF_MAX) 3825 len = 0; 3826 else if (ap->a_end < ap->a_start) 3827 return (0); 3828 else 3829 len = ap->a_end - ap->a_start + 1; 3830 nmp = VFSTONFS(ap->a_vp->v_mount); 3831 mtx_lock(&nmp->nm_mtx); 3832 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3833 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) == 3834 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) { 3835 mtx_unlock(&nmp->nm_mtx); 3836 return (0); 3837 } 3838 mtx_unlock(&nmp->nm_mtx); 3839 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice, 3840 td->td_ucred, td); 3841 if (error == NFSERR_NOTSUPP) { 3842 mtx_lock(&nmp->nm_mtx); 3843 nmp->nm_privflag |= NFSMNTP_NOADVISE; 3844 mtx_unlock(&nmp->nm_mtx); 3845 } 3846 return (0); 3847 } 3848 3849 /* 3850 * nfs allocate call 3851 */ 3852 static int 3853 nfs_allocate(struct vop_allocate_args *ap) 3854 { 3855 struct vnode *vp = ap->a_vp; 3856 struct thread *td = curthread; 3857 struct nfsvattr nfsva; 3858 struct nfsmount *nmp; 3859 struct nfsnode *np; 3860 off_t alen; 3861 int attrflag, error, ret; 3862 struct timespec ts; 3863 struct uio io; 3864 3865 attrflag = 0; 3866 nmp = VFSTONFS(vp->v_mount); 3867 np = VTONFS(vp); 3868 mtx_lock(&nmp->nm_mtx); 3869 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3870 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) { 3871 mtx_unlock(&nmp->nm_mtx); 3872 alen = *ap->a_len; 3873 if ((uint64_t)alen > nfs_maxalloclen) 3874 alen = nfs_maxalloclen; 3875 3876 /* Check the file size limit. */ 3877 io.uio_offset = *ap->a_offset; 3878 io.uio_resid = alen; 3879 error = vn_rlimit_fsize(vp, &io, td); 3880 3881 /* 3882 * Flush first to ensure that the allocate adds to the 3883 * file's allocation on the server. 3884 */ 3885 if (error == 0) { 3886 vnode_pager_clean_sync(vp); 3887 error = ncl_flush(vp, MNT_WAIT, td, 1, 0); 3888 } 3889 if (error == 0) 3890 error = nfsrpc_allocate(vp, *ap->a_offset, alen, 3891 &nfsva, &attrflag, ap->a_cred, td); 3892 if (error == 0) { 3893 *ap->a_offset += alen; 3894 *ap->a_len -= alen; 3895 nanouptime(&ts); 3896 NFSLOCKNODE(np); 3897 np->n_localmodtime = ts; 3898 NFSUNLOCKNODE(np); 3899 } else if (error == NFSERR_NOTSUPP) { 3900 mtx_lock(&nmp->nm_mtx); 3901 nmp->nm_privflag |= NFSMNTP_NOALLOCATE; 3902 mtx_unlock(&nmp->nm_mtx); 3903 error = EINVAL; 3904 } 3905 } else { 3906 mtx_unlock(&nmp->nm_mtx); 3907 error = EINVAL; 3908 } 3909 if (attrflag != 0) { 3910 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3911 if (error == 0 && ret != 0) 3912 error = ret; 3913 } 3914 if (error != 0) 3915 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3916 return (error); 3917 } 3918 3919 /* 3920 * nfs deallocate call 3921 */ 3922 static int 3923 nfs_deallocate(struct vop_deallocate_args *ap) 3924 { 3925 struct vnode *vp = ap->a_vp; 3926 struct thread *td = curthread; 3927 struct nfsvattr nfsva; 3928 struct nfsmount *nmp; 3929 struct nfsnode *np; 3930 off_t tlen, mlen; 3931 int attrflag, error, ret; 3932 bool clipped; 3933 struct timespec ts; 3934 3935 error = 0; 3936 attrflag = 0; 3937 nmp = VFSTONFS(vp->v_mount); 3938 np = VTONFS(vp); 3939 mtx_lock(&nmp->nm_mtx); 3940 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3941 (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) { 3942 mtx_unlock(&nmp->nm_mtx); 3943 tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len); 3944 NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n", 3945 (intmax_t)*ap->a_offset, (intmax_t)tlen, 3946 (uintmax_t)nmp->nm_maxfilesize); 3947 if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) { 3948 /* Avoid EFBIG error return from the NFSv4.2 server. */ 3949 *ap->a_len = 0; 3950 return (0); 3951 } 3952 clipped = false; 3953 if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize) 3954 tlen = nmp->nm_maxfilesize - *ap->a_offset; 3955 if ((uint64_t)*ap->a_offset < np->n_size) { 3956 /* Limit the len to nfs_maxalloclen before EOF. */ 3957 mlen = omin((off_t)np->n_size - *ap->a_offset, tlen); 3958 if ((uint64_t)mlen > nfs_maxalloclen) { 3959 NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n"); 3960 tlen = nfs_maxalloclen; 3961 clipped = true; 3962 } 3963 } 3964 if (error == 0) 3965 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 3966 if (error == 0) { 3967 vnode_pager_purge_range(vp, *ap->a_offset, 3968 *ap->a_offset + tlen); 3969 error = nfsrpc_deallocate(vp, *ap->a_offset, tlen, 3970 &nfsva, &attrflag, ap->a_cred, td); 3971 NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error); 3972 } 3973 if (error == 0) { 3974 NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n", 3975 attrflag, (uintmax_t)nfsva.na_size); 3976 nanouptime(&ts); 3977 NFSLOCKNODE(np); 3978 np->n_localmodtime = ts; 3979 NFSUNLOCKNODE(np); 3980 if (attrflag != 0) { 3981 if ((uint64_t)*ap->a_offset < nfsva.na_size) 3982 *ap->a_offset += omin((off_t) 3983 nfsva.na_size - *ap->a_offset, 3984 tlen); 3985 } 3986 if (clipped && tlen < *ap->a_len) 3987 *ap->a_len -= tlen; 3988 else 3989 *ap->a_len = 0; 3990 } else if (error == NFSERR_NOTSUPP) { 3991 mtx_lock(&nmp->nm_mtx); 3992 nmp->nm_privflag |= NFSMNTP_NODEALLOCATE; 3993 mtx_unlock(&nmp->nm_mtx); 3994 } 3995 } else { 3996 mtx_unlock(&nmp->nm_mtx); 3997 error = EIO; 3998 } 3999 /* 4000 * If the NFS server cannot perform the Deallocate operation, just call 4001 * vop_stddeallocate() to perform it. 4002 */ 4003 if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) { 4004 error = vop_stddeallocate(ap); 4005 NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error); 4006 } 4007 if (attrflag != 0) { 4008 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4009 if (error == 0 && ret != 0) 4010 error = ret; 4011 } 4012 if (error != 0) 4013 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 4014 return (error); 4015 } 4016 4017 /* 4018 * nfs copy_file_range call 4019 */ 4020 static int 4021 nfs_copy_file_range(struct vop_copy_file_range_args *ap) 4022 { 4023 struct vnode *invp = ap->a_invp; 4024 struct vnode *outvp = ap->a_outvp; 4025 struct mount *mp; 4026 vm_object_t invp_obj; 4027 struct nfsvattr innfsva, outnfsva; 4028 struct vattr va, *vap; 4029 struct uio io; 4030 struct nfsmount *nmp; 4031 size_t len, len2; 4032 ssize_t r; 4033 int error, inattrflag, outattrflag, ret, ret2, invp_lock; 4034 off_t inoff, outoff; 4035 bool consecutive, must_commit, tryoutcred; 4036 4037 /* 4038 * NFSv4.2 Copy is not permitted for infile == outfile. 4039 * TODO: copy_file_range() between multiple NFS mountpoints 4040 */ 4041 if (invp == outvp || invp->v_mount != outvp->v_mount) { 4042 generic_copy: 4043 return (ENOSYS); 4044 } 4045 4046 invp_lock = LK_SHARED; 4047 relock: 4048 4049 /* Lock both vnodes, avoiding risk of deadlock. */ 4050 do { 4051 mp = NULL; 4052 error = vn_start_write(outvp, &mp, V_WAIT); 4053 if (error == 0) { 4054 error = vn_lock(outvp, LK_EXCLUSIVE); 4055 if (error == 0) { 4056 error = vn_lock(invp, invp_lock | LK_NOWAIT); 4057 if (error == 0) 4058 break; 4059 VOP_UNLOCK(outvp); 4060 if (mp != NULL) 4061 vn_finished_write(mp); 4062 mp = NULL; 4063 error = vn_lock(invp, invp_lock); 4064 if (error == 0) 4065 VOP_UNLOCK(invp); 4066 } 4067 } 4068 if (mp != NULL) 4069 vn_finished_write(mp); 4070 } while (error == 0); 4071 if (error != 0) 4072 return (error); 4073 4074 /* 4075 * More reasons to avoid nfs copy: not NFSv4.2, or explicitly 4076 * disabled. 4077 */ 4078 nmp = VFSTONFS(invp->v_mount); 4079 mtx_lock(&nmp->nm_mtx); 4080 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4081 (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0) { 4082 mtx_unlock(&nmp->nm_mtx); 4083 VOP_UNLOCK(invp); 4084 VOP_UNLOCK(outvp); 4085 if (mp != NULL) 4086 vn_finished_write(mp); 4087 goto generic_copy; 4088 } 4089 mtx_unlock(&nmp->nm_mtx); 4090 4091 /* 4092 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer? 4093 */ 4094 io.uio_offset = *ap->a_outoffp; 4095 io.uio_resid = *ap->a_lenp; 4096 error = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd); 4097 *ap->a_lenp = io.uio_resid; 4098 /* 4099 * No need to call vn_rlimit_fsizex_res before return, since the uio is 4100 * local. 4101 */ 4102 4103 /* 4104 * Flush the input file so that the data is up to date before 4105 * the copy. Flush writes for the output file so that they 4106 * do not overwrite the data copied to the output file by the Copy. 4107 * Set the commit argument for both flushes so that the data is on 4108 * stable storage before the Copy RPC. This is done in case the 4109 * server reboots during the Copy and needs to be redone. 4110 */ 4111 if (error == 0) { 4112 invp_obj = invp->v_object; 4113 if (invp_obj != NULL && vm_object_mightbedirty(invp_obj)) { 4114 if (invp_lock != LK_EXCLUSIVE) { 4115 invp_lock = LK_EXCLUSIVE; 4116 VOP_UNLOCK(invp); 4117 VOP_UNLOCK(outvp); 4118 if (mp != NULL) 4119 vn_finished_write(mp); 4120 goto relock; 4121 } 4122 vnode_pager_clean_sync(invp); 4123 } 4124 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0); 4125 } 4126 if (error == 0) 4127 error = ncl_vinvalbuf(outvp, V_SAVE, curthread, 0); 4128 4129 /* Do the actual NFSv4.2 RPC. */ 4130 ret = ret2 = 0; 4131 len = *ap->a_lenp; 4132 mtx_lock(&nmp->nm_mtx); 4133 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0) 4134 consecutive = true; 4135 else 4136 consecutive = false; 4137 mtx_unlock(&nmp->nm_mtx); 4138 inoff = *ap->a_inoffp; 4139 outoff = *ap->a_outoffp; 4140 tryoutcred = true; 4141 must_commit = false; 4142 if (error == 0) { 4143 vap = &VTONFS(invp)->n_vattr.na_vattr; 4144 error = VOP_GETATTR(invp, vap, ap->a_incred); 4145 if (error == 0) { 4146 /* 4147 * Clip "len" at va_size so that RFC compliant servers 4148 * will not reply NFSERR_INVAL. 4149 * Setting "len == 0" for the RPC would be preferred, 4150 * but some Linux servers do not support that. 4151 * If the len is being set to 0, do a Setattr RPC to 4152 * set the server's atime. This behaviour was the 4153 * preferred one for the FreeBSD "collective". 4154 */ 4155 if (inoff >= vap->va_size) { 4156 *ap->a_lenp = len = 0; 4157 if ((nmp->nm_mountp->mnt_flag & MNT_NOATIME) == 4158 0) { 4159 VATTR_NULL(&va); 4160 va.va_atime.tv_sec = 0; 4161 va.va_atime.tv_nsec = 0; 4162 va.va_vaflags = VA_UTIMES_NULL; 4163 inattrflag = 0; 4164 error = nfsrpc_setattr(invp, &va, NULL, 4165 ap->a_incred, curthread, &innfsva, 4166 &inattrflag); 4167 if (inattrflag != 0) 4168 ret = nfscl_loadattrcache(&invp, 4169 &innfsva, NULL, 0, 1); 4170 if (error == 0 && ret != 0) 4171 error = ret; 4172 } 4173 } else if (inoff + len > vap->va_size) 4174 *ap->a_lenp = len = vap->va_size - inoff; 4175 } else 4176 error = 0; 4177 } 4178 4179 /* 4180 * len will be set to 0 upon a successful Copy RPC. 4181 * As such, this only loops when the Copy RPC needs to be retried. 4182 */ 4183 while (len > 0 && error == 0) { 4184 inattrflag = outattrflag = 0; 4185 len2 = len; 4186 if (tryoutcred) 4187 error = nfsrpc_copy_file_range(invp, ap->a_inoffp, 4188 outvp, ap->a_outoffp, &len2, ap->a_flags, 4189 &inattrflag, &innfsva, &outattrflag, &outnfsva, 4190 ap->a_outcred, consecutive, &must_commit); 4191 else 4192 error = nfsrpc_copy_file_range(invp, ap->a_inoffp, 4193 outvp, ap->a_outoffp, &len2, ap->a_flags, 4194 &inattrflag, &innfsva, &outattrflag, &outnfsva, 4195 ap->a_incred, consecutive, &must_commit); 4196 if (inattrflag != 0) 4197 ret = nfscl_loadattrcache(&invp, &innfsva, NULL, 0, 1); 4198 if (outattrflag != 0) 4199 ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL, 4200 1, 1); 4201 if (error == 0) { 4202 if (consecutive == false) { 4203 if (len2 == len) { 4204 mtx_lock(&nmp->nm_mtx); 4205 nmp->nm_privflag |= 4206 NFSMNTP_NOCONSECUTIVE; 4207 mtx_unlock(&nmp->nm_mtx); 4208 } else 4209 error = NFSERR_OFFLOADNOREQS; 4210 } 4211 *ap->a_lenp = len2; 4212 len = 0; 4213 if (len2 > 0 && must_commit && error == 0) 4214 error = ncl_commit(outvp, outoff, *ap->a_lenp, 4215 ap->a_outcred, curthread); 4216 if (error == 0 && ret != 0) 4217 error = ret; 4218 if (error == 0 && ret2 != 0) 4219 error = ret2; 4220 } else if (error == NFSERR_OFFLOADNOREQS && consecutive) { 4221 /* 4222 * Try consecutive == false, which is ok only if all 4223 * bytes are copied. 4224 * If only some bytes were copied when consecutive 4225 * is false, there is no way to know which bytes 4226 * still need to be written. 4227 */ 4228 consecutive = false; 4229 error = 0; 4230 } else if (error == NFSERR_ACCES && tryoutcred) { 4231 /* Try again with incred. */ 4232 tryoutcred = false; 4233 error = 0; 4234 } 4235 if (error == NFSERR_STALEWRITEVERF) { 4236 /* 4237 * Server rebooted, so do it all again. 4238 */ 4239 *ap->a_inoffp = inoff; 4240 *ap->a_outoffp = outoff; 4241 len = *ap->a_lenp; 4242 must_commit = false; 4243 error = 0; 4244 } 4245 } 4246 VOP_UNLOCK(invp); 4247 VOP_UNLOCK(outvp); 4248 if (mp != NULL) 4249 vn_finished_write(mp); 4250 if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS || 4251 error == NFSERR_ACCES) { 4252 /* 4253 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can 4254 * use a_incred for the read and a_outcred for the write, so 4255 * try this for NFSERR_ACCES failures for the Copy. 4256 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can 4257 * never succeed, so disable it. 4258 */ 4259 if (error != NFSERR_ACCES) { 4260 /* Can never do Copy on this mount. */ 4261 mtx_lock(&nmp->nm_mtx); 4262 nmp->nm_privflag |= NFSMNTP_NOCOPY; 4263 mtx_unlock(&nmp->nm_mtx); 4264 } 4265 *ap->a_inoffp = inoff; 4266 *ap->a_outoffp = outoff; 4267 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 4268 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 4269 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 4270 } else if (error != 0) 4271 *ap->a_lenp = 0; 4272 4273 if (error != 0) 4274 error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0); 4275 return (error); 4276 } 4277 4278 /* 4279 * nfs ioctl call 4280 */ 4281 static int 4282 nfs_ioctl(struct vop_ioctl_args *ap) 4283 { 4284 struct vnode *vp = ap->a_vp; 4285 struct nfsvattr nfsva; 4286 struct nfsmount *nmp; 4287 int attrflag, content, error, ret; 4288 bool eof = false; /* shut up compiler. */ 4289 4290 /* Do the actual NFSv4.2 RPC. */ 4291 switch (ap->a_command) { 4292 case FIOSEEKDATA: 4293 content = NFSV4CONTENT_DATA; 4294 break; 4295 case FIOSEEKHOLE: 4296 content = NFSV4CONTENT_HOLE; 4297 break; 4298 default: 4299 return (ENOTTY); 4300 } 4301 4302 error = vn_lock(vp, LK_EXCLUSIVE); 4303 if (error != 0) 4304 return (EBADF); 4305 4306 if (vp->v_type != VREG) { 4307 VOP_UNLOCK(vp); 4308 return (ENOTTY); 4309 } 4310 nmp = VFSTONFS(vp->v_mount); 4311 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) { 4312 VOP_UNLOCK(vp); 4313 error = vop_stdioctl(ap); 4314 return (error); 4315 } 4316 4317 attrflag = 0; 4318 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size) 4319 error = ENXIO; 4320 else { 4321 /* 4322 * Flush all writes, so that the server is up to date. 4323 * Although a Commit is not required, the commit argument 4324 * is set so that, for a pNFS File/Flexible File Layout 4325 * server, the LayoutCommit will be done to ensure the file 4326 * size is up to date on the Metadata Server. 4327 */ 4328 4329 vnode_pager_clean_sync(vp); 4330 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0); 4331 if (error == 0) 4332 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof, 4333 content, ap->a_cred, &nfsva, &attrflag); 4334 /* If at eof for FIOSEEKDATA, return ENXIO. */ 4335 if (eof && error == 0 && content == NFSV4CONTENT_DATA) 4336 error = ENXIO; 4337 } 4338 if (attrflag != 0) { 4339 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4340 if (error == 0 && ret != 0) 4341 error = ret; 4342 } 4343 NFSVOPUNLOCK(vp); 4344 4345 if (error != 0) 4346 error = ENXIO; 4347 return (error); 4348 } 4349 4350 /* 4351 * nfs getextattr call 4352 */ 4353 static int 4354 nfs_getextattr(struct vop_getextattr_args *ap) 4355 { 4356 struct vnode *vp = ap->a_vp; 4357 struct nfsmount *nmp; 4358 struct ucred *cred; 4359 struct thread *td = ap->a_td; 4360 struct nfsvattr nfsva; 4361 ssize_t len; 4362 int attrflag, error, ret; 4363 4364 nmp = VFSTONFS(vp->v_mount); 4365 mtx_lock(&nmp->nm_mtx); 4366 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4367 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4368 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4369 mtx_unlock(&nmp->nm_mtx); 4370 return (EOPNOTSUPP); 4371 } 4372 mtx_unlock(&nmp->nm_mtx); 4373 4374 cred = ap->a_cred; 4375 if (cred == NULL) 4376 cred = td->td_ucred; 4377 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4378 attrflag = 0; 4379 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva, 4380 &attrflag, cred, td); 4381 if (attrflag != 0) { 4382 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4383 if (error == 0 && ret != 0) 4384 error = ret; 4385 } 4386 if (error == 0 && ap->a_size != NULL) 4387 *ap->a_size = len; 4388 4389 switch (error) { 4390 case NFSERR_NOTSUPP: 4391 case NFSERR_OPILLEGAL: 4392 mtx_lock(&nmp->nm_mtx); 4393 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4394 mtx_unlock(&nmp->nm_mtx); 4395 error = EOPNOTSUPP; 4396 break; 4397 case NFSERR_NOXATTR: 4398 case NFSERR_XATTR2BIG: 4399 error = ENOATTR; 4400 break; 4401 default: 4402 error = nfscl_maperr(td, error, 0, 0); 4403 break; 4404 } 4405 return (error); 4406 } 4407 4408 /* 4409 * nfs setextattr call 4410 */ 4411 static int 4412 nfs_setextattr(struct vop_setextattr_args *ap) 4413 { 4414 struct vnode *vp = ap->a_vp; 4415 struct nfsmount *nmp; 4416 struct ucred *cred; 4417 struct thread *td = ap->a_td; 4418 struct nfsvattr nfsva; 4419 int attrflag, error, ret; 4420 4421 nmp = VFSTONFS(vp->v_mount); 4422 mtx_lock(&nmp->nm_mtx); 4423 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4424 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4425 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4426 mtx_unlock(&nmp->nm_mtx); 4427 return (EOPNOTSUPP); 4428 } 4429 mtx_unlock(&nmp->nm_mtx); 4430 4431 if (ap->a_uio->uio_resid < 0) 4432 return (EINVAL); 4433 cred = ap->a_cred; 4434 if (cred == NULL) 4435 cred = td->td_ucred; 4436 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4437 attrflag = 0; 4438 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva, 4439 &attrflag, cred, td); 4440 if (attrflag != 0) { 4441 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4442 if (error == 0 && ret != 0) 4443 error = ret; 4444 } 4445 4446 switch (error) { 4447 case NFSERR_NOTSUPP: 4448 case NFSERR_OPILLEGAL: 4449 mtx_lock(&nmp->nm_mtx); 4450 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4451 mtx_unlock(&nmp->nm_mtx); 4452 error = EOPNOTSUPP; 4453 break; 4454 case NFSERR_NOXATTR: 4455 case NFSERR_XATTR2BIG: 4456 error = ENOATTR; 4457 break; 4458 default: 4459 error = nfscl_maperr(td, error, 0, 0); 4460 break; 4461 } 4462 return (error); 4463 } 4464 4465 /* 4466 * nfs listextattr call 4467 */ 4468 static int 4469 nfs_listextattr(struct vop_listextattr_args *ap) 4470 { 4471 struct vnode *vp = ap->a_vp; 4472 struct nfsmount *nmp; 4473 struct ucred *cred; 4474 struct thread *td = ap->a_td; 4475 struct nfsvattr nfsva; 4476 size_t len, len2; 4477 uint64_t cookie; 4478 int attrflag, error, ret; 4479 bool eof; 4480 4481 nmp = VFSTONFS(vp->v_mount); 4482 mtx_lock(&nmp->nm_mtx); 4483 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4484 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4485 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4486 mtx_unlock(&nmp->nm_mtx); 4487 return (EOPNOTSUPP); 4488 } 4489 mtx_unlock(&nmp->nm_mtx); 4490 4491 cred = ap->a_cred; 4492 if (cred == NULL) 4493 cred = td->td_ucred; 4494 4495 /* Loop around doing List Extended Attribute RPCs. */ 4496 eof = false; 4497 cookie = 0; 4498 len2 = 0; 4499 error = 0; 4500 while (!eof && error == 0) { 4501 len = nmp->nm_rsize; 4502 attrflag = 0; 4503 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof, 4504 &nfsva, &attrflag, cred, td); 4505 if (attrflag != 0) { 4506 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4507 if (error == 0 && ret != 0) 4508 error = ret; 4509 } 4510 if (error == 0) { 4511 len2 += len; 4512 if (len2 > SSIZE_MAX) 4513 error = ENOATTR; 4514 } 4515 } 4516 if (error == 0 && ap->a_size != NULL) 4517 *ap->a_size = len2; 4518 4519 switch (error) { 4520 case NFSERR_NOTSUPP: 4521 case NFSERR_OPILLEGAL: 4522 mtx_lock(&nmp->nm_mtx); 4523 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4524 mtx_unlock(&nmp->nm_mtx); 4525 error = EOPNOTSUPP; 4526 break; 4527 case NFSERR_NOXATTR: 4528 case NFSERR_XATTR2BIG: 4529 error = ENOATTR; 4530 break; 4531 default: 4532 error = nfscl_maperr(td, error, 0, 0); 4533 break; 4534 } 4535 return (error); 4536 } 4537 4538 /* 4539 * nfs setextattr call 4540 */ 4541 static int 4542 nfs_deleteextattr(struct vop_deleteextattr_args *ap) 4543 { 4544 struct vnode *vp = ap->a_vp; 4545 struct nfsmount *nmp; 4546 struct nfsvattr nfsva; 4547 int attrflag, error, ret; 4548 4549 nmp = VFSTONFS(vp->v_mount); 4550 mtx_lock(&nmp->nm_mtx); 4551 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4552 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4553 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4554 mtx_unlock(&nmp->nm_mtx); 4555 return (EOPNOTSUPP); 4556 } 4557 mtx_unlock(&nmp->nm_mtx); 4558 4559 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4560 attrflag = 0; 4561 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred, 4562 ap->a_td); 4563 if (attrflag != 0) { 4564 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4565 if (error == 0 && ret != 0) 4566 error = ret; 4567 } 4568 4569 switch (error) { 4570 case NFSERR_NOTSUPP: 4571 case NFSERR_OPILLEGAL: 4572 mtx_lock(&nmp->nm_mtx); 4573 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4574 mtx_unlock(&nmp->nm_mtx); 4575 error = EOPNOTSUPP; 4576 break; 4577 case NFSERR_NOXATTR: 4578 case NFSERR_XATTR2BIG: 4579 error = ENOATTR; 4580 break; 4581 default: 4582 error = nfscl_maperr(ap->a_td, error, 0, 0); 4583 break; 4584 } 4585 return (error); 4586 } 4587 4588 /* 4589 * Return POSIX pathconf information applicable to nfs filesystems. 4590 */ 4591 static int 4592 nfs_pathconf(struct vop_pathconf_args *ap) 4593 { 4594 struct nfsv3_pathconf pc; 4595 struct nfsvattr nfsva; 4596 struct vnode *vp = ap->a_vp; 4597 struct nfsmount *nmp; 4598 struct thread *td = curthread; 4599 off_t off; 4600 bool eof, has_namedattr, named_enabled; 4601 int attrflag, error; 4602 struct nfsnode *np; 4603 4604 nmp = VFSTONFS(vp->v_mount); 4605 np = VTONFS(vp); 4606 named_enabled = false; 4607 has_namedattr = false; 4608 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 4609 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 4610 ap->a_name == _PC_NO_TRUNC)) || 4611 (NFS_ISV4(vp) && (ap->a_name == _PC_ACL_NFS4 || 4612 ap->a_name == _PC_HAS_NAMEDATTR))) { 4613 /* 4614 * Since only the above 4 a_names are returned by the NFSv3 4615 * Pathconf RPC, there is no point in doing it for others. 4616 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 4617 * be used for _PC_ACL_NFS4 and _PC_HAS_NAMEDATTR as well. 4618 */ 4619 error = nfsrpc_pathconf(vp, &pc, &has_namedattr, td->td_ucred, 4620 td, &nfsva, &attrflag); 4621 if (attrflag != 0) 4622 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4623 if (error != 0) 4624 return (error); 4625 } else if (NFS_ISV4(vp) && ap->a_name == _PC_NAMEDATTR_ENABLED && 4626 (np->n_flag & NNAMEDNOTSUPP) == 0) { 4627 struct nfsfh *nfhp; 4628 4629 error = nfsrpc_openattr(nmp, vp, np->n_fhp->nfh_fh, 4630 np->n_fhp->nfh_len, false, td->td_ucred, td, &nfsva, &nfhp, 4631 &attrflag); 4632 named_enabled = true; 4633 if (error == 0) { 4634 free(nfhp, M_NFSFH); 4635 } else if (error == NFSERR_NOTSUPP) { 4636 named_enabled = false; 4637 NFSLOCKNODE(np); 4638 np->n_flag |= NNAMEDNOTSUPP; 4639 NFSUNLOCKNODE(np); 4640 } 4641 error = 0; 4642 } else { 4643 /* 4644 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 4645 * just fake them. 4646 */ 4647 pc.pc_linkmax = NFS_LINK_MAX; 4648 pc.pc_namemax = NFS_MAXNAMLEN; 4649 pc.pc_notrunc = 1; 4650 pc.pc_chownrestricted = 1; 4651 pc.pc_caseinsensitive = 0; 4652 pc.pc_casepreserving = 1; 4653 error = 0; 4654 } 4655 switch (ap->a_name) { 4656 case _PC_LINK_MAX: 4657 #ifdef _LP64 4658 *ap->a_retval = pc.pc_linkmax; 4659 #else 4660 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 4661 #endif 4662 break; 4663 case _PC_NAME_MAX: 4664 *ap->a_retval = pc.pc_namemax; 4665 break; 4666 case _PC_PIPE_BUF: 4667 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 4668 *ap->a_retval = PIPE_BUF; 4669 else 4670 error = EINVAL; 4671 break; 4672 case _PC_CHOWN_RESTRICTED: 4673 *ap->a_retval = pc.pc_chownrestricted; 4674 break; 4675 case _PC_NO_TRUNC: 4676 *ap->a_retval = pc.pc_notrunc; 4677 break; 4678 case _PC_ACL_NFS4: 4679 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 4680 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 4681 *ap->a_retval = 1; 4682 else 4683 *ap->a_retval = 0; 4684 break; 4685 case _PC_ACL_PATH_MAX: 4686 if (NFS_ISV4(vp)) 4687 *ap->a_retval = ACL_MAX_ENTRIES; 4688 else 4689 *ap->a_retval = 3; 4690 break; 4691 case _PC_PRIO_IO: 4692 *ap->a_retval = 0; 4693 break; 4694 case _PC_SYNC_IO: 4695 *ap->a_retval = 0; 4696 break; 4697 case _PC_ALLOC_SIZE_MIN: 4698 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 4699 break; 4700 case _PC_FILESIZEBITS: 4701 if (NFS_ISV34(vp)) 4702 *ap->a_retval = 64; 4703 else 4704 *ap->a_retval = 32; 4705 break; 4706 case _PC_REC_INCR_XFER_SIZE: 4707 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4708 break; 4709 case _PC_REC_MAX_XFER_SIZE: 4710 *ap->a_retval = -1; /* means ``unlimited'' */ 4711 break; 4712 case _PC_REC_MIN_XFER_SIZE: 4713 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4714 break; 4715 case _PC_REC_XFER_ALIGN: 4716 *ap->a_retval = PAGE_SIZE; 4717 break; 4718 case _PC_SYMLINK_MAX: 4719 *ap->a_retval = NFS_MAXPATHLEN; 4720 break; 4721 case _PC_MIN_HOLE_SIZE: 4722 /* Only some NFSv4.2 servers support Seek for Holes. */ 4723 *ap->a_retval = 0; 4724 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) { 4725 /* 4726 * NFSv4.2 doesn't have an attribute for hole size, 4727 * so all we can do is see if the Seek operation is 4728 * supported and then use f_iosize as a "best guess". 4729 */ 4730 mtx_lock(&nmp->nm_mtx); 4731 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) { 4732 mtx_unlock(&nmp->nm_mtx); 4733 off = 0; 4734 attrflag = 0; 4735 error = nfsrpc_seek(vp, &off, &eof, 4736 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva, 4737 &attrflag); 4738 if (attrflag != 0) 4739 (void) nfscl_loadattrcache(&vp, &nfsva, 4740 NULL, 0, 1); 4741 mtx_lock(&nmp->nm_mtx); 4742 if (error == NFSERR_NOTSUPP) 4743 nmp->nm_privflag |= NFSMNTP_SEEKTESTED; 4744 else 4745 nmp->nm_privflag |= NFSMNTP_SEEKTESTED | 4746 NFSMNTP_SEEK; 4747 error = 0; 4748 } 4749 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0) 4750 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4751 mtx_unlock(&nmp->nm_mtx); 4752 } 4753 break; 4754 case _PC_NAMEDATTR_ENABLED: 4755 if (named_enabled) 4756 *ap->a_retval = 1; 4757 else 4758 *ap->a_retval = 0; 4759 break; 4760 case _PC_HAS_NAMEDATTR: 4761 if (has_namedattr) 4762 *ap->a_retval = 1; 4763 else 4764 *ap->a_retval = 0; 4765 break; 4766 case _PC_HAS_HIDDENSYSTEM: 4767 if (NFS_ISV4(vp) && NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, 4768 NFSATTRBIT_HIDDEN) && 4769 NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, 4770 NFSATTRBIT_SYSTEM)) 4771 *ap->a_retval = 1; 4772 else 4773 *ap->a_retval = 0; 4774 break; 4775 4776 default: 4777 error = vop_stdpathconf(ap); 4778 break; 4779 } 4780 return (error); 4781 } 4782