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(vp, 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_ARCHIVE, 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_ARCHIVE | UF_HIDDEN | UF_SYSTEM)) != 0 || 1090 ((vap->va_flags & UF_ARCHIVE) != 0 && 1091 !NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, NFSATTRBIT_ARCHIVE)) || 1092 ((vap->va_flags & UF_HIDDEN) != 0 && 1093 !NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, NFSATTRBIT_HIDDEN)) || 1094 ((vap->va_flags & UF_SYSTEM) != 0 && 1095 !NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, NFSATTRBIT_SYSTEM)))) 1096 return (EOPNOTSUPP); 1097 1098 /* 1099 * Disallow write attempts if the filesystem is mounted read-only. 1100 */ 1101 if ((vap->va_flags != (u_long)VNOVAL || vap->va_uid != (uid_t)VNOVAL || 1102 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 1103 vap->va_mtime.tv_sec != VNOVAL || 1104 vap->va_birthtime.tv_sec != VNOVAL || 1105 vap->va_mode != (mode_t)VNOVAL) && 1106 (vp->v_mount->mnt_flag & MNT_RDONLY)) 1107 return (EROFS); 1108 if (vap->va_size != VNOVAL) { 1109 switch (vp->v_type) { 1110 case VDIR: 1111 return (EISDIR); 1112 case VCHR: 1113 case VBLK: 1114 case VSOCK: 1115 case VFIFO: 1116 if (vap->va_mtime.tv_sec == VNOVAL && 1117 vap->va_atime.tv_sec == VNOVAL && 1118 vap->va_birthtime.tv_sec == VNOVAL && 1119 vap->va_mode == (mode_t)VNOVAL && 1120 vap->va_uid == (uid_t)VNOVAL && 1121 vap->va_gid == (gid_t)VNOVAL) 1122 return (0); 1123 vap->va_size = VNOVAL; 1124 break; 1125 default: 1126 /* 1127 * Disallow write attempts if the filesystem is 1128 * mounted read-only. 1129 */ 1130 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1131 return (EROFS); 1132 /* 1133 * We run vnode_pager_setsize() early (why?), 1134 * we must set np->n_size now to avoid vinvalbuf 1135 * V_SAVE races that might setsize a lower 1136 * value. 1137 */ 1138 NFSLOCKNODE(np); 1139 tsize = np->n_size; 1140 NFSUNLOCKNODE(np); 1141 error = ncl_meta_setsize(vp, td, vap->va_size); 1142 NFSLOCKNODE(np); 1143 if (np->n_flag & NMODIFIED) { 1144 tsize = np->n_size; 1145 NFSUNLOCKNODE(np); 1146 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 1147 0 : V_SAVE, td, 1); 1148 if (error != 0) { 1149 vnode_pager_setsize(vp, tsize); 1150 return (error); 1151 } 1152 /* 1153 * Call nfscl_delegmodtime() to set the modify time 1154 * locally, as required. 1155 */ 1156 nfscl_delegmodtime(vp, NULL); 1157 } else 1158 NFSUNLOCKNODE(np); 1159 /* 1160 * np->n_size has already been set to vap->va_size 1161 * in ncl_meta_setsize(). We must set it again since 1162 * nfs_loadattrcache() could be called through 1163 * ncl_meta_setsize() and could modify np->n_size. 1164 */ 1165 NFSLOCKNODE(np); 1166 np->n_vattr.na_size = np->n_size = vap->va_size; 1167 NFSUNLOCKNODE(np); 1168 } 1169 } else { 1170 NFSLOCKNODE(np); 1171 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 1172 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 1173 NFSUNLOCKNODE(np); 1174 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 1175 if (error == EINTR || error == EIO) 1176 return (error); 1177 } else 1178 NFSUNLOCKNODE(np); 1179 } 1180 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 1181 if (vap->va_size != VNOVAL) { 1182 if (error == 0) { 1183 nanouptime(&ts); 1184 NFSLOCKNODE(np); 1185 np->n_localmodtime = ts; 1186 NFSUNLOCKNODE(np); 1187 } else { 1188 NFSLOCKNODE(np); 1189 np->n_size = np->n_vattr.na_size = tsize; 1190 vnode_pager_setsize(vp, tsize); 1191 NFSUNLOCKNODE(np); 1192 } 1193 } 1194 if (vap->va_mtime.tv_sec != VNOVAL && error == 0) 1195 nfscl_delegmodtime(vp, &vap->va_mtime); 1196 return (error); 1197 } 1198 1199 /* 1200 * Do an nfs setattr rpc. 1201 */ 1202 static int 1203 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1204 struct thread *td) 1205 { 1206 struct nfsnode *np = VTONFS(vp); 1207 int error, ret, attrflag, i; 1208 struct nfsvattr nfsva; 1209 1210 if (NFS_ISV34(vp)) { 1211 NFSLOCKNODE(np); 1212 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1213 np->n_accesscache[i].stamp = 0; 1214 np->n_flag |= NDELEGMOD; 1215 NFSUNLOCKNODE(np); 1216 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1217 } 1218 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag); 1219 if (attrflag) { 1220 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1221 if (ret && !error) 1222 error = ret; 1223 } 1224 if (error && NFS_ISV4(vp)) 1225 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1226 return (error); 1227 } 1228 1229 /* 1230 * Get a named attribute directory for the vnode. 1231 */ 1232 static int 1233 nfs_get_namedattrdir(struct vnode *vp, struct componentname *cnp, 1234 struct vnode **vpp) 1235 { 1236 struct nfsfh *nfhp; 1237 struct nfsnode *np; 1238 struct vnode *newvp; 1239 struct nfsvattr nfsva; 1240 int attrflag, error; 1241 1242 attrflag = 0; 1243 *vpp = NULL; 1244 np = VTONFS(vp); 1245 error = nfsrpc_openattr(VFSTONFS(vp->v_mount), vp, np->n_fhp->nfh_fh, 1246 np->n_fhp->nfh_len, (cnp->cn_flags & CREATENAMED), 1247 cnp->cn_cred, curthread, &nfsva, &nfhp, &attrflag); 1248 if (error == NFSERR_NOTSUPP) 1249 error = ENOATTR; 1250 if (error == 0) 1251 error = nfscl_nget(vp->v_mount, vp, nfhp, cnp, curthread, &np, 1252 cnp->cn_lkflags); 1253 if (error != 0) 1254 return (error); 1255 newvp = NFSTOV(np); 1256 vn_irflag_set_cond(newvp, VIRF_NAMEDDIR); 1257 if (attrflag != 0) 1258 (void)nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1259 *vpp = newvp; 1260 return (0); 1261 } 1262 1263 /* 1264 * nfs lookup call, one step at a time... 1265 * First look in cache 1266 * If not found, unlock the directory nfsnode and do the rpc 1267 */ 1268 static int 1269 nfs_lookup(struct vop_lookup_args *ap) 1270 { 1271 struct componentname *cnp = ap->a_cnp; 1272 struct vnode *dvp = ap->a_dvp; 1273 struct vnode **vpp = ap->a_vpp; 1274 struct mount *mp = dvp->v_mount; 1275 uint64_t flags = cnp->cn_flags; 1276 struct vnode *newvp; 1277 struct nfsmount *nmp; 1278 struct nfsnode *np, *newnp; 1279 int error = 0, attrflag, dattrflag, ltype, ncticks; 1280 struct thread *td = curthread; 1281 struct nfsfh *nfhp; 1282 struct nfsvattr dnfsva, nfsva; 1283 struct vattr vattr; 1284 struct timespec nctime, ts; 1285 uint32_t openmode; 1286 bool is_nameddir, needs_nameddir, opennamed; 1287 1288 dattrflag = 0; 1289 *vpp = NULL; 1290 nmp = VFSTONFS(mp); 1291 opennamed = (flags & (OPENNAMED | ISLASTCN)) == (OPENNAMED | ISLASTCN); 1292 if (opennamed && (!NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp))) 1293 return (ENOATTR); 1294 is_nameddir = (vn_irflag_read(dvp) & VIRF_NAMEDDIR) != 0; 1295 if ((is_nameddir && (flags & ISLASTCN) == 0 && (cnp->cn_namelen > 1 || 1296 *cnp->cn_nameptr != '.')) || 1297 (opennamed && !is_nameddir && (flags & ISDOTDOT) != 0)) 1298 return (ENOATTR); 1299 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1300 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1301 return (EROFS); 1302 np = VTONFS(dvp); 1303 1304 needs_nameddir = false; 1305 if (opennamed || is_nameddir) { 1306 cnp->cn_flags &= ~MAKEENTRY; 1307 if (!is_nameddir) 1308 needs_nameddir = true; 1309 } 1310 1311 /* 1312 * If the named attribute directory is needed, acquire it now. 1313 */ 1314 newvp = NULL; 1315 if (needs_nameddir) { 1316 KASSERT(np->n_v4 == NULL, ("nfs_lookup: O_NAMEDATTR when" 1317 " n_v4 not NULL")); 1318 error = nfs_get_namedattrdir(dvp, cnp, &newvp); 1319 if (error != 0) 1320 goto handle_error; 1321 if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') { 1322 *vpp = newvp; 1323 return (0); 1324 } 1325 dvp = newvp; 1326 np = VTONFS(dvp); 1327 newvp = NULL; 1328 } else if (opennamed && cnp->cn_namelen == 1 && 1329 *cnp->cn_nameptr == '.') { 1330 vref(dvp); 1331 *vpp = dvp; 1332 return (0); 1333 } 1334 1335 if (dvp->v_type != VDIR) 1336 return (ENOTDIR); 1337 1338 /* For NFSv4, wait until any remove is done. */ 1339 NFSLOCKNODE(np); 1340 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1341 np->n_flag |= NREMOVEWANT; 1342 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1343 } 1344 NFSUNLOCKNODE(np); 1345 1346 error = vn_dir_check_exec(dvp, cnp); 1347 if (error != 0) 1348 return (error); 1349 1350 if (!opennamed && !is_nameddir) { 1351 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1352 if (error > 0 && error != ENOENT) 1353 return (error); 1354 if (error == -1) { 1355 /* 1356 * Lookups of "." are special and always return the 1357 * current directory. cache_lookup() already handles 1358 * associated locking bookkeeping, etc. 1359 */ 1360 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1361 return (0); 1362 } 1363 1364 /* 1365 * We only accept a positive hit in the cache if the 1366 * change time of the file matches our cached copy. 1367 * Otherwise, we discard the cache entry and fallback 1368 * to doing a lookup RPC. We also only trust cache 1369 * entries for less than nm_nametimeo seconds. 1370 * 1371 * To better handle stale file handles and attributes, 1372 * clear the attribute cache of this node if it is a 1373 * leaf component, part of an open() call, and not 1374 * locally modified before fetching the attributes. 1375 * This should allow stale file handles to be detected 1376 * here where we can fall back to a LOOKUP RPC to 1377 * recover rather than having nfs_open() detect the 1378 * stale file handle and failing open(2) with ESTALE. 1379 */ 1380 newvp = *vpp; 1381 newnp = VTONFS(newvp); 1382 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1383 (flags & (ISLASTCN | ISOPEN)) == 1384 (ISLASTCN | ISOPEN) && 1385 !(newnp->n_flag & NMODIFIED)) { 1386 NFSLOCKNODE(newnp); 1387 newnp->n_attrstamp = 0; 1388 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1389 NFSUNLOCKNODE(newnp); 1390 } 1391 if (nfscl_nodeleg(newvp, 0) == 0 || 1392 ((u_int)(ticks - ncticks) < 1393 (nmp->nm_nametimeo * hz) && 1394 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1395 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1396 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1397 return (0); 1398 } 1399 cache_purge(newvp); 1400 if (dvp != newvp) 1401 vput(newvp); 1402 else 1403 vrele(newvp); 1404 *vpp = NULL; 1405 } else if (error == ENOENT) { 1406 if (VN_IS_DOOMED(dvp)) 1407 return (ENOENT); 1408 /* 1409 * We only accept a negative hit in the cache if the 1410 * modification time of the parent directory matches 1411 * the cached copy in the name cache entry. 1412 * Otherwise, we discard all of the negative cache 1413 * entries for this directory. We also only trust 1414 * negative cache entries for up to nm_negnametimeo 1415 * seconds. 1416 */ 1417 if ((u_int)(ticks - ncticks) < 1418 (nmp->nm_negnametimeo * hz) && 1419 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1420 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1421 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1422 return (ENOENT); 1423 } 1424 cache_purge_negative(dvp); 1425 } 1426 } 1427 1428 openmode = 0; 1429 #if 0 1430 /* 1431 * The use of LookupOpen breaks some builds. It is disabled 1432 * until that is fixed. 1433 */ 1434 /* 1435 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount 1436 * option, it is possible to do the Open operation in the same 1437 * compound as Lookup, so long as delegations are not being 1438 * issued. This saves doing a separate RPC for Open. 1439 * For pnfs, do not do this, since the Open+LayoutGet will 1440 * be needed as a separate RPC. 1441 */ 1442 NFSLOCKMNT(nmp); 1443 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) && 1444 (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 && 1445 (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) && 1446 (flags & (ISLASTCN | ISOPEN | OPENNAMED))) == (ISLASTCN | ISOPEN)) { 1447 if ((flags & OPENREAD) != 0) 1448 openmode |= NFSV4OPEN_ACCESSREAD; 1449 if ((flags & OPENWRITE) != 0) 1450 openmode |= NFSV4OPEN_ACCESSWRITE; 1451 } 1452 NFSUNLOCKMNT(nmp); 1453 #endif 1454 1455 newvp = NULL; 1456 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1457 nanouptime(&ts); 1458 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1459 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1460 openmode); 1461 if (dattrflag) 1462 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1463 if (needs_nameddir) { 1464 vput(dvp); 1465 dvp = ap->a_dvp; 1466 } 1467 handle_error: 1468 if (error) { 1469 if (newvp != NULL) { 1470 vput(newvp); 1471 *vpp = NULL; 1472 } 1473 1474 if (error != ENOENT) { 1475 if (NFS_ISV4(dvp)) 1476 error = nfscl_maperr(td, error, 1477 (uid_t)0, (gid_t)0); 1478 return (error); 1479 } 1480 1481 /* The requested file was not found. */ 1482 if ((cnp->cn_nameiop == CREATE || 1483 cnp->cn_nameiop == RENAME) && 1484 (flags & ISLASTCN)) { 1485 /* 1486 * XXX: UFS does a full VOP_ACCESS(dvp, 1487 * VWRITE) here instead of just checking 1488 * MNT_RDONLY. 1489 */ 1490 if (mp->mnt_flag & MNT_RDONLY) 1491 return (EROFS); 1492 return (EJUSTRETURN); 1493 } 1494 1495 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1496 /* 1497 * Cache the modification time of the parent 1498 * directory from the post-op attributes in 1499 * the name cache entry. The negative cache 1500 * entry will be ignored once the directory 1501 * has changed. Don't bother adding the entry 1502 * if the directory has already changed. 1503 */ 1504 NFSLOCKNODE(np); 1505 if (timespeccmp(&np->n_vattr.na_mtime, 1506 &dnfsva.na_mtime, ==)) { 1507 NFSUNLOCKNODE(np); 1508 cache_enter_time(dvp, NULL, cnp, 1509 &dnfsva.na_mtime, NULL); 1510 } else 1511 NFSUNLOCKNODE(np); 1512 } 1513 return (ENOENT); 1514 } 1515 1516 /* 1517 * Handle RENAME case... 1518 */ 1519 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1520 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1521 free(nfhp, M_NFSFH); 1522 return (EISDIR); 1523 } 1524 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1525 LK_EXCLUSIVE); 1526 if (error) 1527 return (error); 1528 newvp = NFSTOV(np); 1529 /* 1530 * If n_localmodtime >= time before RPC, then 1531 * a file modification operation, such as 1532 * VOP_SETATTR() of size, has occurred while 1533 * the Lookup RPC and acquisition of the vnode 1534 * happened. As such, the attributes might 1535 * be stale, with possibly an incorrect size. 1536 */ 1537 NFSLOCKNODE(np); 1538 if (timespecisset(&np->n_localmodtime) && 1539 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1540 NFSCL_DEBUG(4, "nfs_lookup: rename localmod " 1541 "stale attributes\n"); 1542 attrflag = 0; 1543 } 1544 NFSUNLOCKNODE(np); 1545 if (attrflag) 1546 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1547 0, 1); 1548 *vpp = newvp; 1549 return (0); 1550 } 1551 1552 if (flags & ISDOTDOT) { 1553 ltype = NFSVOPISLOCKED(dvp); 1554 error = vfs_busy(mp, MBF_NOWAIT); 1555 if (error != 0) { 1556 vfs_ref(mp); 1557 NFSVOPUNLOCK(dvp); 1558 error = vfs_busy(mp, 0); 1559 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1560 vfs_rel(mp); 1561 if (error == 0 && VN_IS_DOOMED(dvp)) { 1562 vfs_unbusy(mp); 1563 error = ENOENT; 1564 } 1565 if (error != 0) 1566 return (error); 1567 } 1568 NFSVOPUNLOCK(dvp); 1569 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1570 cnp->cn_lkflags); 1571 if (error == 0) 1572 newvp = NFSTOV(np); 1573 vfs_unbusy(mp); 1574 if (newvp != dvp) 1575 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1576 if (VN_IS_DOOMED(dvp)) { 1577 if (error == 0) { 1578 if (newvp == dvp) 1579 vrele(newvp); 1580 else 1581 vput(newvp); 1582 } 1583 error = ENOENT; 1584 } 1585 if (error != 0) 1586 return (error); 1587 if (attrflag) 1588 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1589 0, 1); 1590 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1591 free(nfhp, M_NFSFH); 1592 vref(dvp); 1593 newvp = dvp; 1594 if (attrflag) 1595 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1596 0, 1); 1597 } else { 1598 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1599 cnp->cn_lkflags); 1600 if (error) 1601 return (error); 1602 newvp = NFSTOV(np); 1603 if (opennamed) 1604 vn_irflag_set_cond(newvp, VIRF_NAMEDATTR); 1605 /* 1606 * If n_localmodtime >= time before RPC, then 1607 * a file modification operation, such as 1608 * VOP_SETATTR() of size, has occurred while 1609 * the Lookup RPC and acquisition of the vnode 1610 * happened. As such, the attributes might 1611 * be stale, with possibly an incorrect size. 1612 */ 1613 NFSLOCKNODE(np); 1614 if (timespecisset(&np->n_localmodtime) && 1615 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1616 NFSCL_DEBUG(4, "nfs_lookup: localmod " 1617 "stale attributes\n"); 1618 attrflag = 0; 1619 } 1620 NFSUNLOCKNODE(np); 1621 if (attrflag) 1622 (void)nfscl_loadattrcache(&newvp, &nfsva, NULL, 1623 0, 1); 1624 else if ((flags & (ISLASTCN | ISOPEN)) == 1625 (ISLASTCN | ISOPEN) && 1626 !(np->n_flag & NMODIFIED)) { 1627 /* 1628 * Flush the attribute cache when opening a 1629 * leaf node to ensure that fresh attributes 1630 * are fetched in nfs_open() since we did not 1631 * fetch attributes from the LOOKUP reply. 1632 */ 1633 NFSLOCKNODE(np); 1634 np->n_attrstamp = 0; 1635 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1636 NFSUNLOCKNODE(np); 1637 } 1638 } 1639 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp && 1640 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1641 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1642 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1643 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1644 *vpp = newvp; 1645 return (0); 1646 } 1647 1648 /* 1649 * nfs read call. 1650 * Just call ncl_bioread() to do the work. 1651 */ 1652 static int 1653 nfs_read(struct vop_read_args *ap) 1654 { 1655 struct vnode *vp = ap->a_vp; 1656 1657 switch (vp->v_type) { 1658 case VREG: 1659 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1660 case VDIR: 1661 return (EISDIR); 1662 default: 1663 return (EOPNOTSUPP); 1664 } 1665 } 1666 1667 /* 1668 * nfs readlink call 1669 */ 1670 static int 1671 nfs_readlink(struct vop_readlink_args *ap) 1672 { 1673 struct vnode *vp = ap->a_vp; 1674 1675 if (vp->v_type != VLNK) 1676 return (EINVAL); 1677 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1678 } 1679 1680 /* 1681 * Do a readlink rpc. 1682 * Called by ncl_doio() from below the buffer cache. 1683 */ 1684 int 1685 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1686 { 1687 int error, ret, attrflag; 1688 struct nfsvattr nfsva; 1689 1690 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1691 &attrflag); 1692 if (attrflag) { 1693 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1694 if (ret && !error) 1695 error = ret; 1696 } 1697 if (error && NFS_ISV4(vp)) 1698 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1699 return (error); 1700 } 1701 1702 /* 1703 * nfs read rpc call 1704 * Ditto above 1705 */ 1706 int 1707 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1708 { 1709 int error, ret, attrflag; 1710 struct nfsvattr nfsva; 1711 struct nfsmount *nmp; 1712 1713 nmp = VFSTONFS(vp->v_mount); 1714 error = EIO; 1715 attrflag = 0; 1716 if (NFSHASPNFS(nmp)) 1717 error = nfscl_doiods(vp, uiop, NULL, NULL, 1718 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1719 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1720 if (error != 0 && error != EFAULT) 1721 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1722 &attrflag); 1723 if (attrflag) { 1724 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1725 if (ret && !error) 1726 error = ret; 1727 } 1728 if (error && NFS_ISV4(vp)) 1729 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1730 return (error); 1731 } 1732 1733 /* 1734 * nfs write call 1735 */ 1736 int 1737 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1738 int *iomode, int *must_commit, int called_from_strategy, int ioflag) 1739 { 1740 struct nfsvattr nfsva; 1741 int error, attrflag, ret; 1742 struct nfsmount *nmp; 1743 1744 nmp = VFSTONFS(vp->v_mount); 1745 error = EIO; 1746 attrflag = 0; 1747 if (NFSHASPNFS(nmp)) 1748 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1749 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1750 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1751 if (error != 0 && error != EFAULT) 1752 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1753 uiop->uio_td, &nfsva, &attrflag, called_from_strategy, 1754 ioflag); 1755 if (attrflag) { 1756 if (VTONFS(vp)->n_flag & ND_NFSV4) 1757 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1); 1758 else 1759 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1760 if (ret && !error) 1761 error = ret; 1762 } 1763 if (DOINGASYNC(vp)) 1764 *iomode = NFSWRITE_FILESYNC; 1765 if (error && NFS_ISV4(vp)) 1766 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1767 return (error); 1768 } 1769 1770 /* 1771 * nfs mknod rpc 1772 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1773 * mode set to specify the file type and the size field for rdev. 1774 */ 1775 static int 1776 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1777 struct vattr *vap) 1778 { 1779 struct nfsvattr nfsva, dnfsva; 1780 struct vnode *newvp = NULL; 1781 struct nfsnode *np = NULL, *dnp; 1782 struct nfsfh *nfhp; 1783 struct vattr vattr; 1784 int error = 0, attrflag, dattrflag; 1785 u_int32_t rdev; 1786 1787 if (VATTR_ISDEV(vap)) 1788 rdev = vap->va_rdev; 1789 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1790 rdev = 0xffffffff; 1791 else 1792 return (EOPNOTSUPP); 1793 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1794 return (error); 1795 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1796 rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva, 1797 &nfsva, &nfhp, &attrflag, &dattrflag); 1798 if (!error) { 1799 if (!nfhp) 1800 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1801 cnp->cn_namelen, cnp->cn_cred, curthread, 1802 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0); 1803 if (nfhp) 1804 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1805 curthread, &np, LK_EXCLUSIVE); 1806 } 1807 if (dattrflag) 1808 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1809 if (!error) { 1810 newvp = NFSTOV(np); 1811 if (attrflag != 0) { 1812 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1813 if (error != 0) 1814 vput(newvp); 1815 } 1816 } 1817 if (!error) { 1818 *vpp = newvp; 1819 } else if (NFS_ISV4(dvp)) { 1820 error = nfscl_maperr(curthread, error, vap->va_uid, 1821 vap->va_gid); 1822 } 1823 dnp = VTONFS(dvp); 1824 NFSLOCKNODE(dnp); 1825 dnp->n_flag |= NMODIFIED; 1826 if (!dattrflag) { 1827 dnp->n_attrstamp = 0; 1828 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1829 } 1830 NFSUNLOCKNODE(dnp); 1831 return (error); 1832 } 1833 1834 /* 1835 * nfs mknod vop 1836 * just call nfs_mknodrpc() to do the work. 1837 */ 1838 /* ARGSUSED */ 1839 static int 1840 nfs_mknod(struct vop_mknod_args *ap) 1841 { 1842 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1843 } 1844 1845 static struct mtx nfs_cverf_mtx; 1846 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1847 MTX_DEF); 1848 1849 static nfsquad_t 1850 nfs_get_cverf(void) 1851 { 1852 static nfsquad_t cverf; 1853 nfsquad_t ret; 1854 static int cverf_initialized = 0; 1855 1856 mtx_lock(&nfs_cverf_mtx); 1857 if (cverf_initialized == 0) { 1858 cverf.lval[0] = arc4random(); 1859 cverf.lval[1] = arc4random(); 1860 cverf_initialized = 1; 1861 } else 1862 cverf.qval++; 1863 ret = cverf; 1864 mtx_unlock(&nfs_cverf_mtx); 1865 1866 return (ret); 1867 } 1868 1869 /* 1870 * nfs file create call 1871 */ 1872 static int 1873 nfs_create(struct vop_create_args *ap) 1874 { 1875 struct vnode *dvp = ap->a_dvp; 1876 struct vattr *vap = ap->a_vap; 1877 struct componentname *cnp = ap->a_cnp; 1878 struct nfsnode *np = NULL, *dnp; 1879 struct vnode *newvp = NULL; 1880 struct nfsmount *nmp; 1881 struct nfsvattr dnfsva, nfsva; 1882 struct nfsfh *nfhp; 1883 nfsquad_t cverf; 1884 int error = 0, attrflag, dattrflag, fmode = 0; 1885 struct vattr vattr; 1886 bool is_nameddir, needs_nameddir, opennamed; 1887 1888 /* 1889 * Oops, not for me.. 1890 */ 1891 if (vap->va_type == VSOCK) 1892 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1893 1894 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1895 return (error); 1896 if (vap->va_vaflags & VA_EXCLUSIVE) 1897 fmode |= O_EXCL; 1898 dnp = VTONFS(dvp); 1899 nmp = VFSTONFS(dvp->v_mount); 1900 needs_nameddir = false; 1901 if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp)) { 1902 opennamed = (cnp->cn_flags & (OPENNAMED | ISLASTCN)) == 1903 (OPENNAMED | ISLASTCN); 1904 is_nameddir = (vn_irflag_read(dvp) & VIRF_NAMEDDIR) != 0; 1905 if (opennamed || is_nameddir) { 1906 cnp->cn_flags &= ~MAKEENTRY; 1907 if (!is_nameddir) 1908 needs_nameddir = true; 1909 } 1910 } 1911 1912 /* 1913 * If the named attribute directory is needed, acquire it now. 1914 */ 1915 if (needs_nameddir) { 1916 KASSERT(dnp->n_v4 == NULL, ("nfs_create: O_NAMEDATTR when" 1917 " n_v4 not NULL")); 1918 error = nfs_get_namedattrdir(dvp, cnp, &newvp); 1919 if (error != 0) 1920 return (error); 1921 dvp = newvp; 1922 dnp = VTONFS(dvp); 1923 newvp = NULL; 1924 } 1925 1926 again: 1927 /* For NFSv4, wait until any remove is done. */ 1928 NFSLOCKNODE(dnp); 1929 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1930 dnp->n_flag |= NREMOVEWANT; 1931 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1932 } 1933 NFSUNLOCKNODE(dnp); 1934 1935 cverf = nfs_get_cverf(); 1936 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1937 vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva, 1938 &nfhp, &attrflag, &dattrflag); 1939 if (!error) { 1940 if (nfhp == NULL) 1941 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1942 cnp->cn_namelen, cnp->cn_cred, curthread, 1943 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0); 1944 if (nfhp != NULL) 1945 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1946 curthread, &np, LK_EXCLUSIVE); 1947 } 1948 if (dattrflag) 1949 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1950 if (!error) { 1951 newvp = NFSTOV(np); 1952 if (attrflag == 0) 1953 error = nfsrpc_getattr(newvp, cnp->cn_cred, curthread, 1954 &nfsva); 1955 if (error == 0) 1956 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1957 } 1958 if (error) { 1959 if (newvp != NULL) { 1960 vput(newvp); 1961 newvp = NULL; 1962 } 1963 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1964 error == NFSERR_NOTSUPP) { 1965 fmode &= ~O_EXCL; 1966 goto again; 1967 } 1968 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1969 if (nfscl_checksattr(vap, &nfsva)) { 1970 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1971 curthread, &nfsva, &attrflag); 1972 if (error && (vap->va_uid != (uid_t)VNOVAL || 1973 vap->va_gid != (gid_t)VNOVAL)) { 1974 /* try again without setting uid/gid */ 1975 vap->va_uid = (uid_t)VNOVAL; 1976 vap->va_gid = (uid_t)VNOVAL; 1977 error = nfsrpc_setattr(newvp, vap, NULL, 1978 cnp->cn_cred, curthread, &nfsva, &attrflag); 1979 } 1980 if (attrflag) 1981 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1982 0, 1); 1983 if (error != 0) 1984 vput(newvp); 1985 } 1986 } 1987 if (!error) { 1988 if ((cnp->cn_flags & MAKEENTRY) && attrflag) { 1989 if (dvp != newvp) 1990 cache_enter_time(dvp, newvp, cnp, 1991 &nfsva.na_ctime, NULL); 1992 else 1993 printf("nfs_create: bogus NFS server returned " 1994 "the directory as the new file object\n"); 1995 } 1996 *ap->a_vpp = newvp; 1997 } else if (NFS_ISV4(dvp)) { 1998 error = nfscl_maperr(curthread, error, vap->va_uid, 1999 vap->va_gid); 2000 } 2001 NFSLOCKNODE(dnp); 2002 dnp->n_flag |= NMODIFIED; 2003 if (!dattrflag) { 2004 dnp->n_attrstamp = 0; 2005 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2006 } 2007 NFSUNLOCKNODE(dnp); 2008 if (needs_nameddir) 2009 vput(dvp); 2010 return (error); 2011 } 2012 2013 /* 2014 * nfs file remove call 2015 * To try and make nfs semantics closer to ufs semantics, a file that has 2016 * other processes using the vnode is renamed instead of removed and then 2017 * removed later on the last close. 2018 * - If v_usecount > 1 2019 * If a rename is not already in the works 2020 * call nfs_sillyrename() to set it up 2021 * else 2022 * do the remove rpc 2023 */ 2024 static int 2025 nfs_remove(struct vop_remove_args *ap) 2026 { 2027 struct vnode *vp = ap->a_vp; 2028 struct vnode *dvp = ap->a_dvp; 2029 struct componentname *cnp = ap->a_cnp; 2030 struct nfsnode *np = VTONFS(vp); 2031 int error = 0; 2032 struct vattr vattr; 2033 struct nfsmount *nmp; 2034 2035 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 2036 if (vp->v_type == VDIR) 2037 error = EPERM; 2038 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 2039 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 2040 vattr.va_nlink > 1)) { 2041 nmp = VFSTONFS(vp->v_mount); 2042 /* 2043 * Purge the name cache so that the chance of a lookup for 2044 * the name succeeding while the remove is in progress is 2045 * minimized. Without node locking it can still happen, such 2046 * that an I/O op returns ESTALE, but since you get this if 2047 * another host removes the file.. 2048 */ 2049 cache_purge(vp); 2050 /* 2051 * throw away biocache buffers, mainly to avoid 2052 * unnecessary delayed writes later. 2053 * Flushing here would be more correct for the case 2054 * where nfs_close() did not do a flush. However, it 2055 * could be a large performance hit for some servers 2056 * and only matters when the file name being removed is 2057 * one of multiple hard links. 2058 */ 2059 if (!NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp) || 2060 (nmp->nm_flag & NFSMNT_NOCTO) == 0) 2061 error = ncl_vinvalbuf(vp, 0, curthread, 1); 2062 if (error != EINTR && error != EIO) 2063 /* Do the rpc */ 2064 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 2065 cnp->cn_namelen, cnp->cn_cred, curthread, false); 2066 /* 2067 * Kludge City: If the first reply to the remove rpc is lost.. 2068 * the reply to the retransmitted request will be ENOENT 2069 * since the file was in fact removed 2070 * Therefore, we cheat and return success. 2071 */ 2072 if (error == ENOENT) 2073 error = 0; 2074 } else if (!np->n_sillyrename) 2075 error = nfs_sillyrename(dvp, vp, cnp); 2076 NFSLOCKNODE(np); 2077 np->n_attrstamp = 0; 2078 NFSUNLOCKNODE(np); 2079 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2080 return (error); 2081 } 2082 2083 /* 2084 * nfs file remove rpc called from nfs_inactive 2085 */ 2086 int 2087 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 2088 { 2089 /* 2090 * Make sure that the directory vnode is still valid. 2091 * XXX we should lock sp->s_dvp here. 2092 */ 2093 if (sp->s_dvp->v_type == VBAD) 2094 return (0); 2095 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 2096 sp->s_cred, NULL, true)); 2097 } 2098 2099 /* 2100 * Handle the nfsremove_status reply from the RPC function. 2101 */ 2102 static void 2103 nfs_removestatus(struct vnode *vp, nfsremove_status file_status, 2104 bool silly, struct thread *td) 2105 { 2106 2107 switch (file_status) { 2108 case NLINK_ZERO: 2109 /* Get rid of any delegation. */ 2110 nfscl_delegreturnvp(vp, false, td); 2111 /* FALLTHROUGH */ 2112 case DELETED: 2113 /* Throw away buffer cache blocks. */ 2114 (void)ncl_vinvalbuf(vp, 0, td, 1); 2115 break; 2116 case VALID: 2117 /* Nothing to do, delegation is still ok. */ 2118 break; 2119 default: 2120 break; 2121 } 2122 } 2123 2124 /* 2125 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 2126 */ 2127 static int 2128 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 2129 int namelen, struct ucred *cred, struct thread *td, bool silly) 2130 { 2131 struct nfsvattr dnfsva, nfsva; 2132 struct nfsnode *dnp = VTONFS(dvp); 2133 struct nfsmount *nmp; 2134 int attrflag, error = 0, dattrflag; 2135 nfsremove_status file_status; 2136 2137 nmp = VFSTONFS(dvp->v_mount); 2138 NFSLOCKNODE(dnp); 2139 dnp->n_flag |= NREMOVEINPROG; 2140 NFSUNLOCKNODE(dnp); 2141 error = nfsrpc_remove(dvp, name, namelen, vp, &nfsva, &attrflag, 2142 &file_status, &dnfsva, &dattrflag, cred, td); 2143 NFSLOCKNODE(dnp); 2144 if ((dnp->n_flag & NREMOVEWANT)) { 2145 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 2146 NFSUNLOCKNODE(dnp); 2147 wakeup((caddr_t)dnp); 2148 } else { 2149 dnp->n_flag &= ~NREMOVEINPROG; 2150 NFSUNLOCKNODE(dnp); 2151 } 2152 2153 if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp)) { 2154 if (file_status != DELETED && attrflag != 0) 2155 (void)nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2156 if ((nmp->nm_flag & NFSMNT_NOCTO) != 0) 2157 nfs_removestatus(vp, file_status, silly, td); 2158 } 2159 2160 if (dattrflag != 0) 2161 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2162 NFSLOCKNODE(dnp); 2163 dnp->n_flag |= NMODIFIED; 2164 if (dattrflag == 0) { 2165 dnp->n_attrstamp = 0; 2166 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2167 } 2168 NFSUNLOCKNODE(dnp); 2169 if (error && NFS_ISV4(dvp)) 2170 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2171 return (error); 2172 } 2173 2174 /* 2175 * nfs file rename call 2176 */ 2177 static int 2178 nfs_rename(struct vop_rename_args *ap) 2179 { 2180 struct vnode *fvp = ap->a_fvp; 2181 struct vnode *tvp = ap->a_tvp; 2182 struct vnode *fdvp = ap->a_fdvp; 2183 struct vnode *tdvp = ap->a_tdvp; 2184 struct componentname *tcnp = ap->a_tcnp; 2185 struct componentname *fcnp = ap->a_fcnp; 2186 struct nfsnode *fnp = VTONFS(ap->a_fvp); 2187 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 2188 struct nfsv4node *newv4 = NULL; 2189 struct nfsmount *nmp; 2190 int error; 2191 2192 /* Check for cross-device rename */ 2193 if ((fvp->v_mount != tdvp->v_mount) || 2194 (tvp && (fvp->v_mount != tvp->v_mount))) { 2195 error = EXDEV; 2196 goto out; 2197 } 2198 nmp = VFSTONFS(fvp->v_mount); 2199 2200 if (fvp == tvp) { 2201 printf("nfs_rename: fvp == tvp (can't happen)\n"); 2202 error = 0; 2203 goto out; 2204 } 2205 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 2206 goto out; 2207 2208 /* 2209 * We have to flush B_DELWRI data prior to renaming 2210 * the file. If we don't, the delayed-write buffers 2211 * can be flushed out later after the file has gone stale 2212 * under NFSV3. NFSV2 does not have this problem because 2213 * ( as far as I can tell ) it flushes dirty buffers more 2214 * often. 2215 * 2216 * Skip the rename operation if the fsync fails, this can happen 2217 * due to the server's volume being full, when we pushed out data 2218 * that was written back to our cache earlier. Not checking for 2219 * this condition can result in potential (silent) data loss. 2220 */ 2221 if ((nmp->nm_flag & NFSMNT_NOCTO) == 0 || !NFSHASNFSV4(nmp) || 2222 !NFSHASNFSV4N(nmp) || nfscl_mustflush(fvp) != 0) 2223 error = VOP_FSYNC(fvp, MNT_WAIT, curthread); 2224 NFSVOPUNLOCK(fvp); 2225 if (error == 0 && tvp != NULL && ((nmp->nm_flag & NFSMNT_NOCTO) == 0 || 2226 !NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp) || 2227 nfscl_mustflush(tvp) != 0)) 2228 error = VOP_FSYNC(tvp, MNT_WAIT, curthread); 2229 if (error != 0) 2230 goto out; 2231 2232 /* 2233 * If the tvp exists and is in use, sillyrename it before doing the 2234 * rename of the new file over it. 2235 * XXX Can't sillyrename a directory. 2236 */ 2237 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 2238 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 2239 vput(tvp); 2240 tvp = NULL; 2241 } 2242 2243 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 2244 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, false, tcnp->cn_cred, 2245 curthread); 2246 2247 if (error == 0 && NFS_ISV4(tdvp)) { 2248 /* 2249 * For NFSv4, check to see if it is the same name and 2250 * replace the name, if it is different. 2251 */ 2252 newv4 = malloc( 2253 sizeof (struct nfsv4node) + 2254 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 2255 M_NFSV4NODE, M_WAITOK); 2256 NFSLOCKNODE(tdnp); 2257 NFSLOCKNODE(fnp); 2258 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 2259 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 2260 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 2261 tcnp->cn_namelen) || 2262 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 2263 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2264 tdnp->n_fhp->nfh_len))) { 2265 free(fnp->n_v4, M_NFSV4NODE); 2266 fnp->n_v4 = newv4; 2267 newv4 = NULL; 2268 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 2269 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 2270 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2271 tdnp->n_fhp->nfh_len); 2272 NFSBCOPY(tcnp->cn_nameptr, 2273 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 2274 } 2275 NFSUNLOCKNODE(tdnp); 2276 NFSUNLOCKNODE(fnp); 2277 if (newv4 != NULL) 2278 free(newv4, M_NFSV4NODE); 2279 } 2280 2281 if (fvp->v_type == VDIR) { 2282 if (tvp != NULL && tvp->v_type == VDIR) 2283 cache_purge(tdvp); 2284 cache_purge(fdvp); 2285 } 2286 2287 out: 2288 if (tdvp == tvp) 2289 vrele(tdvp); 2290 else 2291 vput(tdvp); 2292 if (tvp) 2293 vput(tvp); 2294 vrele(fdvp); 2295 vrele(fvp); 2296 /* 2297 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2298 */ 2299 if (error == ENOENT) 2300 error = 0; 2301 return (error); 2302 } 2303 2304 /* 2305 * nfs file rename rpc called from nfs_remove() above 2306 */ 2307 static int 2308 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 2309 struct sillyrename *sp) 2310 { 2311 2312 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 2313 sdvp, NULL, sp->s_name, sp->s_namlen, true, scnp->cn_cred, 2314 curthread)); 2315 } 2316 2317 /* 2318 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2319 */ 2320 static int 2321 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 2322 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 2323 int tnamelen, bool silly, struct ucred *cred, struct thread *td) 2324 { 2325 struct nfsvattr fnfsva, tnfsva, tvpnfsva; 2326 struct nfsnode *fdnp = VTONFS(fdvp); 2327 struct nfsnode *tdnp = VTONFS(tdvp); 2328 struct nfsmount *nmp; 2329 int error = 0, fattrflag, tattrflag, tvpattrflag; 2330 nfsremove_status tvp_status; 2331 2332 nmp = VFSTONFS(fdvp->v_mount); 2333 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 2334 tnameptr, tnamelen, &tvp_status, &fnfsva, &tnfsva, &fattrflag, 2335 &tattrflag, &tvpnfsva, &tvpattrflag, cred, td); 2336 NFSLOCKNODE(fdnp); 2337 fdnp->n_flag |= NMODIFIED; 2338 if (fattrflag != 0) { 2339 NFSUNLOCKNODE(fdnp); 2340 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1); 2341 } else { 2342 fdnp->n_attrstamp = 0; 2343 NFSUNLOCKNODE(fdnp); 2344 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 2345 } 2346 NFSLOCKNODE(tdnp); 2347 tdnp->n_flag |= NMODIFIED; 2348 if (tattrflag != 0) { 2349 NFSUNLOCKNODE(tdnp); 2350 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1); 2351 } else { 2352 tdnp->n_attrstamp = 0; 2353 NFSUNLOCKNODE(tdnp); 2354 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2355 } 2356 2357 if (tvp != NULL) { 2358 if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp) && 2359 (nmp->nm_flag & NFSMNT_NOCTO) != 0) 2360 nfs_removestatus(tvp, tvp_status, silly, td); 2361 if (!silly && tvpattrflag != 0) 2362 (void)nfscl_loadattrcache(&tvp, &tvpnfsva, NULL, 0, 1); 2363 } 2364 2365 if (error && NFS_ISV4(fdvp)) 2366 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2367 return (error); 2368 } 2369 2370 /* 2371 * nfs hard link create call 2372 */ 2373 static int 2374 nfs_link(struct vop_link_args *ap) 2375 { 2376 struct vnode *vp = ap->a_vp; 2377 struct vnode *tdvp = ap->a_tdvp; 2378 struct componentname *cnp = ap->a_cnp; 2379 struct nfsnode *np, *tdnp; 2380 struct nfsvattr nfsva, dnfsva; 2381 int error = 0, attrflag, dattrflag; 2382 2383 /* 2384 * Push all writes to the server, so that the attribute cache 2385 * doesn't get "out of sync" with the server. 2386 * XXX There should be a better way! 2387 */ 2388 #ifdef notnow 2389 VOP_FSYNC(vp, MNT_WAIT, curthread); 2390 #endif 2391 2392 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2393 cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, &dattrflag); 2394 tdnp = VTONFS(tdvp); 2395 NFSLOCKNODE(tdnp); 2396 tdnp->n_flag |= NMODIFIED; 2397 if (dattrflag != 0) { 2398 NFSUNLOCKNODE(tdnp); 2399 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1); 2400 } else { 2401 tdnp->n_attrstamp = 0; 2402 NFSUNLOCKNODE(tdnp); 2403 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2404 } 2405 if (attrflag) 2406 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2407 else { 2408 np = VTONFS(vp); 2409 NFSLOCKNODE(np); 2410 np->n_attrstamp = 0; 2411 NFSUNLOCKNODE(np); 2412 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2413 } 2414 /* 2415 * If negative lookup caching is enabled, I might as well 2416 * add an entry for this node. Not necessary for correctness, 2417 * but if negative caching is enabled, then the system 2418 * must care about lookup caching hit rate, so... 2419 */ 2420 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2421 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2422 if (tdvp != vp) 2423 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2424 else 2425 printf("nfs_link: bogus NFS server returned " 2426 "the directory as the new link\n"); 2427 } 2428 if (error && NFS_ISV4(vp)) 2429 error = nfscl_maperr(curthread, error, (uid_t)0, 2430 (gid_t)0); 2431 return (error); 2432 } 2433 2434 /* 2435 * nfs symbolic link create call 2436 */ 2437 static int 2438 nfs_symlink(struct vop_symlink_args *ap) 2439 { 2440 struct vnode *dvp = ap->a_dvp; 2441 struct vattr *vap = ap->a_vap; 2442 struct componentname *cnp = ap->a_cnp; 2443 struct nfsvattr nfsva, dnfsva; 2444 struct nfsfh *nfhp; 2445 struct nfsnode *np = NULL, *dnp; 2446 struct vnode *newvp = NULL; 2447 int error = 0, attrflag, dattrflag, ret; 2448 2449 vap->va_type = VLNK; 2450 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2451 ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva, 2452 &nfsva, &nfhp, &attrflag, &dattrflag); 2453 if (nfhp) { 2454 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2455 &np, LK_EXCLUSIVE); 2456 if (!ret) 2457 newvp = NFSTOV(np); 2458 else if (!error) 2459 error = ret; 2460 } 2461 if (newvp != NULL) { 2462 if (attrflag) 2463 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2464 } else if (!error) { 2465 /* 2466 * If we do not have an error and we could not extract the 2467 * newvp from the response due to the request being NFSv2, we 2468 * have to do a lookup in order to obtain a newvp to return. 2469 */ 2470 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2471 cnp->cn_cred, curthread, &np); 2472 if (!error) 2473 newvp = NFSTOV(np); 2474 } 2475 if (error) { 2476 if (newvp) 2477 vput(newvp); 2478 if (NFS_ISV4(dvp)) 2479 error = nfscl_maperr(curthread, error, 2480 vap->va_uid, vap->va_gid); 2481 } else { 2482 *ap->a_vpp = newvp; 2483 } 2484 2485 dnp = VTONFS(dvp); 2486 NFSLOCKNODE(dnp); 2487 dnp->n_flag |= NMODIFIED; 2488 if (dattrflag != 0) { 2489 NFSUNLOCKNODE(dnp); 2490 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2491 } else { 2492 dnp->n_attrstamp = 0; 2493 NFSUNLOCKNODE(dnp); 2494 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2495 } 2496 /* 2497 * If negative lookup caching is enabled, I might as well 2498 * add an entry for this node. Not necessary for correctness, 2499 * but if negative caching is enabled, then the system 2500 * must care about lookup caching hit rate, so... 2501 */ 2502 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2503 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2504 if (dvp != newvp) 2505 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2506 NULL); 2507 else 2508 printf("nfs_symlink: bogus NFS server returned " 2509 "the directory as the new file object\n"); 2510 } 2511 return (error); 2512 } 2513 2514 /* 2515 * nfs make dir call 2516 */ 2517 static int 2518 nfs_mkdir(struct vop_mkdir_args *ap) 2519 { 2520 struct vnode *dvp = ap->a_dvp; 2521 struct vattr *vap = ap->a_vap; 2522 struct componentname *cnp = ap->a_cnp; 2523 struct nfsnode *np = NULL, *dnp; 2524 struct vnode *newvp = NULL; 2525 struct vattr vattr; 2526 struct nfsfh *nfhp; 2527 struct nfsvattr nfsva, dnfsva; 2528 int error = 0, attrflag, dattrflag, ret; 2529 2530 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2531 return (error); 2532 vap->va_type = VDIR; 2533 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2534 vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, 2535 &attrflag, &dattrflag); 2536 dnp = VTONFS(dvp); 2537 NFSLOCKNODE(dnp); 2538 dnp->n_flag |= NMODIFIED; 2539 if (dattrflag != 0) { 2540 NFSUNLOCKNODE(dnp); 2541 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2542 } else { 2543 dnp->n_attrstamp = 0; 2544 NFSUNLOCKNODE(dnp); 2545 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2546 } 2547 if (nfhp) { 2548 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2549 &np, LK_EXCLUSIVE); 2550 if (!ret) { 2551 newvp = NFSTOV(np); 2552 if (attrflag) 2553 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2554 0, 1); 2555 } else if (!error) 2556 error = ret; 2557 } 2558 if (!error && newvp == NULL) { 2559 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2560 cnp->cn_cred, curthread, &np); 2561 if (!error) { 2562 newvp = NFSTOV(np); 2563 if (newvp->v_type != VDIR) 2564 error = EEXIST; 2565 } 2566 } 2567 if (error) { 2568 if (newvp) 2569 vput(newvp); 2570 if (NFS_ISV4(dvp)) 2571 error = nfscl_maperr(curthread, error, 2572 vap->va_uid, vap->va_gid); 2573 } else { 2574 /* 2575 * If negative lookup caching is enabled, I might as well 2576 * add an entry for this node. Not necessary for correctness, 2577 * but if negative caching is enabled, then the system 2578 * must care about lookup caching hit rate, so... 2579 */ 2580 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2581 (cnp->cn_flags & MAKEENTRY) && 2582 attrflag != 0 && dattrflag != 0) { 2583 if (dvp != newvp) 2584 cache_enter_time(dvp, newvp, cnp, 2585 &nfsva.na_ctime, &dnfsva.na_ctime); 2586 else 2587 printf("nfs_mkdir: bogus NFS server returned " 2588 "the directory that the directory was " 2589 "created in as the new file object\n"); 2590 } 2591 *ap->a_vpp = newvp; 2592 } 2593 return (error); 2594 } 2595 2596 /* 2597 * nfs remove directory call 2598 */ 2599 static int 2600 nfs_rmdir(struct vop_rmdir_args *ap) 2601 { 2602 struct vnode *vp = ap->a_vp; 2603 struct vnode *dvp = ap->a_dvp; 2604 struct componentname *cnp = ap->a_cnp; 2605 struct nfsnode *dnp; 2606 struct nfsvattr dnfsva; 2607 int error, dattrflag; 2608 2609 if (dvp == vp) 2610 return (EINVAL); 2611 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2612 cnp->cn_cred, curthread, &dnfsva, &dattrflag); 2613 dnp = VTONFS(dvp); 2614 NFSLOCKNODE(dnp); 2615 dnp->n_flag |= NMODIFIED; 2616 if (dattrflag != 0) { 2617 NFSUNLOCKNODE(dnp); 2618 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2619 } else { 2620 dnp->n_attrstamp = 0; 2621 NFSUNLOCKNODE(dnp); 2622 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2623 } 2624 2625 cache_purge(dvp); 2626 cache_purge(vp); 2627 if (error && NFS_ISV4(dvp)) 2628 error = nfscl_maperr(curthread, error, (uid_t)0, 2629 (gid_t)0); 2630 /* 2631 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2632 */ 2633 if (error == ENOENT) 2634 error = 0; 2635 return (error); 2636 } 2637 2638 /* 2639 * nfs readdir call 2640 */ 2641 static int 2642 nfs_readdir(struct vop_readdir_args *ap) 2643 { 2644 struct vnode *vp = ap->a_vp; 2645 struct nfsnode *np = VTONFS(vp); 2646 struct uio *uio = ap->a_uio; 2647 ssize_t tresid, left; 2648 int error = 0; 2649 struct vattr vattr; 2650 2651 if (ap->a_eofflag != NULL) 2652 *ap->a_eofflag = 0; 2653 if (vp->v_type != VDIR) 2654 return(EPERM); 2655 2656 /* 2657 * First, check for hit on the EOF offset cache 2658 */ 2659 NFSLOCKNODE(np); 2660 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2661 (np->n_flag & NMODIFIED) == 0) { 2662 NFSUNLOCKNODE(np); 2663 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2664 NFSLOCKNODE(np); 2665 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2666 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2667 NFSUNLOCKNODE(np); 2668 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2669 if (ap->a_eofflag != NULL) 2670 *ap->a_eofflag = 1; 2671 return (0); 2672 } else 2673 NFSUNLOCKNODE(np); 2674 } 2675 } else 2676 NFSUNLOCKNODE(np); 2677 2678 /* 2679 * NFS always guarantees that directory entries don't straddle 2680 * DIRBLKSIZ boundaries. As such, we need to limit the size 2681 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2682 * directory entry. 2683 */ 2684 left = uio->uio_resid % DIRBLKSIZ; 2685 if (left == uio->uio_resid) 2686 return (EINVAL); 2687 uio->uio_resid -= left; 2688 2689 /* 2690 * For readdirplus, if starting to read the directory, 2691 * purge the name cache, since it will be reloaded by 2692 * this directory read. 2693 * This removes potentially stale name cache entries. 2694 */ 2695 if (uio->uio_offset == 0 && 2696 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_RDIRPLUS) != 0) 2697 cache_purge(vp); 2698 2699 /* 2700 * Call ncl_bioread() to do the real work. 2701 */ 2702 tresid = uio->uio_resid; 2703 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2704 2705 if (!error && uio->uio_resid == tresid) { 2706 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2707 if (ap->a_eofflag != NULL) 2708 *ap->a_eofflag = 1; 2709 } 2710 2711 /* Add the partial DIRBLKSIZ (left) back in. */ 2712 uio->uio_resid += left; 2713 return (error); 2714 } 2715 2716 /* 2717 * Readdir rpc call. 2718 * Called from below the buffer cache by ncl_doio(). 2719 */ 2720 int 2721 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2722 struct thread *td) 2723 { 2724 struct nfsvattr nfsva; 2725 nfsuint64 *cookiep, cookie; 2726 struct nfsnode *dnp = VTONFS(vp); 2727 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2728 int error = 0, eof, attrflag; 2729 2730 KASSERT(uiop->uio_iovcnt == 1 && 2731 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2732 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2733 ("nfs readdirrpc bad uio")); 2734 2735 /* 2736 * If there is no cookie, assume directory was stale. 2737 */ 2738 ncl_dircookie_lock(dnp); 2739 NFSUNLOCKNODE(dnp); 2740 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2741 if (cookiep) { 2742 cookie = *cookiep; 2743 ncl_dircookie_unlock(dnp); 2744 } else { 2745 ncl_dircookie_unlock(dnp); 2746 return (NFSERR_BAD_COOKIE); 2747 } 2748 2749 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2750 (void)ncl_fsinfo(nmp, vp, cred, td); 2751 2752 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2753 &attrflag, &eof); 2754 if (attrflag) 2755 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2756 2757 if (!error) { 2758 /* 2759 * We are now either at the end of the directory or have filled 2760 * the block. 2761 */ 2762 if (eof) { 2763 NFSLOCKNODE(dnp); 2764 dnp->n_direofoffset = uiop->uio_offset; 2765 NFSUNLOCKNODE(dnp); 2766 } else { 2767 if (uiop->uio_resid > 0) 2768 printf("EEK! readdirrpc resid > 0\n"); 2769 ncl_dircookie_lock(dnp); 2770 NFSUNLOCKNODE(dnp); 2771 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2772 *cookiep = cookie; 2773 ncl_dircookie_unlock(dnp); 2774 } 2775 } else if (NFS_ISV4(vp)) { 2776 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2777 } 2778 return (error); 2779 } 2780 2781 /* 2782 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2783 */ 2784 int 2785 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2786 struct thread *td) 2787 { 2788 struct nfsvattr nfsva; 2789 nfsuint64 *cookiep, cookie; 2790 struct nfsnode *dnp = VTONFS(vp); 2791 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2792 int error = 0, attrflag, eof; 2793 2794 KASSERT(uiop->uio_iovcnt == 1 && 2795 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2796 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2797 ("nfs readdirplusrpc bad uio")); 2798 2799 /* 2800 * If there is no cookie, assume directory was stale. 2801 */ 2802 ncl_dircookie_lock(dnp); 2803 NFSUNLOCKNODE(dnp); 2804 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2805 if (cookiep) { 2806 cookie = *cookiep; 2807 ncl_dircookie_unlock(dnp); 2808 } else { 2809 ncl_dircookie_unlock(dnp); 2810 return (NFSERR_BAD_COOKIE); 2811 } 2812 2813 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2814 (void)ncl_fsinfo(nmp, vp, cred, td); 2815 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2816 &attrflag, &eof); 2817 if (attrflag) 2818 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2819 2820 if (!error) { 2821 /* 2822 * We are now either at end of the directory or have filled the 2823 * the block. 2824 */ 2825 if (eof) { 2826 NFSLOCKNODE(dnp); 2827 dnp->n_direofoffset = uiop->uio_offset; 2828 NFSUNLOCKNODE(dnp); 2829 } else { 2830 if (uiop->uio_resid > 0) 2831 printf("EEK! readdirplusrpc resid > 0\n"); 2832 ncl_dircookie_lock(dnp); 2833 NFSUNLOCKNODE(dnp); 2834 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2835 *cookiep = cookie; 2836 ncl_dircookie_unlock(dnp); 2837 } 2838 } else if (NFS_ISV4(vp)) { 2839 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2840 } 2841 return (error); 2842 } 2843 2844 /* 2845 * Silly rename. To make the NFS filesystem that is stateless look a little 2846 * more like the "ufs" a remove of an active vnode is translated to a rename 2847 * to a funny looking filename that is removed by nfs_inactive on the 2848 * nfsnode. There is the potential for another process on a different client 2849 * to create the same funny name between the nfs_lookitup() fails and the 2850 * nfs_rename() completes, but... 2851 */ 2852 static int 2853 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2854 { 2855 struct sillyrename *sp; 2856 struct nfsnode *np; 2857 int error; 2858 short pid; 2859 unsigned int lticks; 2860 2861 cache_purge(dvp); 2862 np = VTONFS(vp); 2863 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2864 sp = malloc(sizeof (struct sillyrename), 2865 M_NEWNFSREQ, M_WAITOK); 2866 sp->s_cred = crhold(cnp->cn_cred); 2867 sp->s_dvp = dvp; 2868 vref(dvp); 2869 2870 /* 2871 * Fudge together a funny name. 2872 * Changing the format of the funny name to accommodate more 2873 * sillynames per directory. 2874 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2875 * CPU ticks since boot. 2876 */ 2877 pid = curthread->td_proc->p_pid; 2878 lticks = (unsigned int)ticks; 2879 for ( ; ; ) { 2880 sp->s_namlen = sprintf(sp->s_name, 2881 ".nfs.%08x.%04x4.4", lticks, 2882 pid); 2883 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2884 curthread, NULL)) 2885 break; 2886 lticks++; 2887 } 2888 error = nfs_renameit(dvp, vp, cnp, sp); 2889 if (error) 2890 goto bad; 2891 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2892 curthread, &np); 2893 np->n_sillyrename = sp; 2894 return (0); 2895 bad: 2896 vrele(sp->s_dvp); 2897 crfree(sp->s_cred); 2898 free(sp, M_NEWNFSREQ); 2899 return (error); 2900 } 2901 2902 /* 2903 * Look up a file name and optionally either update the file handle or 2904 * allocate an nfsnode, depending on the value of npp. 2905 * npp == NULL --> just do the lookup 2906 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2907 * handled too 2908 * *npp != NULL --> update the file handle in the vnode 2909 */ 2910 static int 2911 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2912 struct thread *td, struct nfsnode **npp) 2913 { 2914 struct vnode *newvp = NULL, *vp; 2915 struct nfsnode *np, *dnp = VTONFS(dvp); 2916 struct nfsfh *nfhp, *onfhp; 2917 struct nfsvattr nfsva, dnfsva; 2918 struct componentname cn; 2919 int error = 0, attrflag, dattrflag; 2920 u_int hash; 2921 struct timespec ts; 2922 2923 nanouptime(&ts); 2924 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2925 &nfhp, &attrflag, &dattrflag, 0); 2926 if (dattrflag) 2927 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2928 if (npp && !error) { 2929 if (*npp != NULL) { 2930 np = *npp; 2931 vp = NFSTOV(np); 2932 /* 2933 * For NFSv4, check to see if it is the same name and 2934 * replace the name, if it is different. 2935 */ 2936 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2937 (np->n_v4->n4_namelen != len || 2938 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2939 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2940 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2941 dnp->n_fhp->nfh_len))) { 2942 free(np->n_v4, M_NFSV4NODE); 2943 np->n_v4 = malloc( 2944 sizeof (struct nfsv4node) + 2945 dnp->n_fhp->nfh_len + len - 1, 2946 M_NFSV4NODE, M_WAITOK); 2947 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2948 np->n_v4->n4_namelen = len; 2949 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2950 dnp->n_fhp->nfh_len); 2951 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2952 } 2953 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2954 FNV1_32_INIT); 2955 onfhp = np->n_fhp; 2956 /* 2957 * Rehash node for new file handle. 2958 */ 2959 vfs_hash_rehash(vp, hash); 2960 np->n_fhp = nfhp; 2961 if (onfhp != NULL) 2962 free(onfhp, M_NFSFH); 2963 newvp = NFSTOV(np); 2964 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2965 free(nfhp, M_NFSFH); 2966 vref(dvp); 2967 newvp = dvp; 2968 } else { 2969 cn.cn_nameptr = name; 2970 cn.cn_namelen = len; 2971 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2972 &np, LK_EXCLUSIVE); 2973 if (error) 2974 return (error); 2975 newvp = NFSTOV(np); 2976 /* 2977 * If n_localmodtime >= time before RPC, then 2978 * a file modification operation, such as 2979 * VOP_SETATTR() of size, has occurred while 2980 * the Lookup RPC and acquisition of the vnode 2981 * happened. As such, the attributes might 2982 * be stale, with possibly an incorrect size. 2983 */ 2984 NFSLOCKNODE(np); 2985 if (timespecisset(&np->n_localmodtime) && 2986 timespeccmp(&np->n_localmodtime, &ts, >=)) { 2987 NFSCL_DEBUG(4, "nfs_lookitup: localmod " 2988 "stale attributes\n"); 2989 attrflag = 0; 2990 } 2991 NFSUNLOCKNODE(np); 2992 } 2993 if (!attrflag && *npp == NULL) { 2994 if (newvp == dvp) 2995 vrele(newvp); 2996 else 2997 vput(newvp); 2998 return (ENOENT); 2999 } 3000 if (attrflag) 3001 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 3002 } 3003 if (npp && *npp == NULL) { 3004 if (error) { 3005 if (newvp) { 3006 if (newvp == dvp) 3007 vrele(newvp); 3008 else 3009 vput(newvp); 3010 } 3011 } else 3012 *npp = np; 3013 } 3014 if (error && NFS_ISV4(dvp)) 3015 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3016 return (error); 3017 } 3018 3019 /* 3020 * Nfs Version 3 and 4 commit rpc 3021 */ 3022 int 3023 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 3024 struct thread *td) 3025 { 3026 struct nfsvattr nfsva; 3027 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3028 struct nfsnode *np; 3029 struct uio uio; 3030 int error, attrflag; 3031 3032 np = VTONFS(vp); 3033 error = EIO; 3034 attrflag = 0; 3035 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 3036 uio.uio_offset = offset; 3037 uio.uio_resid = cnt; 3038 error = nfscl_doiods(vp, &uio, NULL, NULL, 3039 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 3040 if (error != 0) { 3041 NFSLOCKNODE(np); 3042 np->n_flag &= ~NDSCOMMIT; 3043 NFSUNLOCKNODE(np); 3044 } 3045 } 3046 if (error != 0) { 3047 mtx_lock(&nmp->nm_mtx); 3048 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 3049 mtx_unlock(&nmp->nm_mtx); 3050 return (0); 3051 } 3052 mtx_unlock(&nmp->nm_mtx); 3053 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 3054 &attrflag); 3055 } 3056 if (attrflag != 0) 3057 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3058 if (error != 0 && NFS_ISV4(vp)) 3059 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3060 return (error); 3061 } 3062 3063 /* 3064 * Strategy routine. 3065 * For async requests when nfsiod(s) are running, queue the request by 3066 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 3067 * request. 3068 */ 3069 static int 3070 nfs_strategy(struct vop_strategy_args *ap) 3071 { 3072 struct buf *bp; 3073 struct vnode *vp; 3074 struct ucred *cr; 3075 3076 bp = ap->a_bp; 3077 vp = ap->a_vp; 3078 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 3079 KASSERT(!(bp->b_flags & B_DONE), 3080 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 3081 3082 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 3083 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 3084 DEV_BSIZE); 3085 if (bp->b_iocmd == BIO_READ) 3086 cr = bp->b_rcred; 3087 else 3088 cr = bp->b_wcred; 3089 3090 /* 3091 * If the op is asynchronous and an i/o daemon is waiting 3092 * queue the request, wake it up and wait for completion 3093 * otherwise just do it ourselves. 3094 */ 3095 if ((bp->b_flags & B_ASYNC) == 0 || 3096 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 3097 (void) ncl_doio(vp, bp, cr, curthread, 1); 3098 return (0); 3099 } 3100 3101 /* 3102 * fsync vnode op. Just call ncl_flush() with commit == 1. 3103 */ 3104 /* ARGSUSED */ 3105 static int 3106 nfs_fsync(struct vop_fsync_args *ap) 3107 { 3108 3109 if (ap->a_vp->v_type != VREG) { 3110 /* 3111 * For NFS, metadata is changed synchronously on the server, 3112 * so there is nothing to flush. Also, ncl_flush() clears 3113 * the NMODIFIED flag and that shouldn't be done here for 3114 * directories. 3115 */ 3116 return (0); 3117 } 3118 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 3119 } 3120 3121 /* 3122 * Flush all the blocks associated with a vnode. 3123 * Walk through the buffer pool and push any dirty pages 3124 * associated with the vnode. 3125 * If the called_from_renewthread argument is TRUE, it has been called 3126 * from the NFSv4 renew thread and, as such, cannot block indefinitely 3127 * waiting for a buffer write to complete. 3128 */ 3129 int 3130 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 3131 int commit, int called_from_renewthread) 3132 { 3133 struct nfsnode *np = VTONFS(vp); 3134 struct buf *bp; 3135 int i; 3136 struct buf *nbp; 3137 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3138 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 3139 int passone = 1, trycnt = 0; 3140 u_quad_t off, endoff, toff; 3141 struct ucred* wcred = NULL; 3142 struct buf **bvec = NULL; 3143 struct bufobj *bo; 3144 #ifndef NFS_COMMITBVECSIZ 3145 #define NFS_COMMITBVECSIZ 20 3146 #endif 3147 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 3148 u_int bvecsize = 0, bveccount; 3149 struct timespec ts; 3150 3151 if (called_from_renewthread != 0) 3152 slptimeo = hz; 3153 if (nmp->nm_flag & NFSMNT_INT) 3154 slpflag = PCATCH; 3155 if (!commit) 3156 passone = 0; 3157 bo = &vp->v_bufobj; 3158 /* 3159 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 3160 * server, but has not been committed to stable storage on the server 3161 * yet. On the first pass, the byte range is worked out and the commit 3162 * rpc is done. On the second pass, bwrite() is called to do the 3163 * job. 3164 */ 3165 again: 3166 off = (u_quad_t)-1; 3167 endoff = 0; 3168 bvecpos = 0; 3169 if (NFS_ISV34(vp) && commit) { 3170 if (bvec != NULL && bvec != bvec_on_stack) 3171 free(bvec, M_TEMP); 3172 /* 3173 * Count up how many buffers waiting for a commit. 3174 */ 3175 bveccount = 0; 3176 BO_LOCK(bo); 3177 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3178 if (!BUF_ISLOCKED(bp) && 3179 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 3180 == (B_DELWRI | B_NEEDCOMMIT)) 3181 bveccount++; 3182 } 3183 /* 3184 * Allocate space to remember the list of bufs to commit. It is 3185 * important to use M_NOWAIT here to avoid a race with nfs_write. 3186 * If we can't get memory (for whatever reason), we will end up 3187 * committing the buffers one-by-one in the loop below. 3188 */ 3189 if (bveccount > NFS_COMMITBVECSIZ) { 3190 /* 3191 * Release the vnode interlock to avoid a lock 3192 * order reversal. 3193 */ 3194 BO_UNLOCK(bo); 3195 bvec = (struct buf **) 3196 malloc(bveccount * sizeof(struct buf *), 3197 M_TEMP, M_NOWAIT); 3198 BO_LOCK(bo); 3199 if (bvec == NULL) { 3200 bvec = bvec_on_stack; 3201 bvecsize = NFS_COMMITBVECSIZ; 3202 } else 3203 bvecsize = bveccount; 3204 } else { 3205 bvec = bvec_on_stack; 3206 bvecsize = NFS_COMMITBVECSIZ; 3207 } 3208 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3209 if (bvecpos >= bvecsize) 3210 break; 3211 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3212 nbp = TAILQ_NEXT(bp, b_bobufs); 3213 continue; 3214 } 3215 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3216 (B_DELWRI | B_NEEDCOMMIT)) { 3217 BUF_UNLOCK(bp); 3218 nbp = TAILQ_NEXT(bp, b_bobufs); 3219 continue; 3220 } 3221 BO_UNLOCK(bo); 3222 bremfree(bp); 3223 /* 3224 * Work out if all buffers are using the same cred 3225 * so we can deal with them all with one commit. 3226 * 3227 * NOTE: we are not clearing B_DONE here, so we have 3228 * to do it later on in this routine if we intend to 3229 * initiate I/O on the bp. 3230 * 3231 * Note: to avoid loopback deadlocks, we do not 3232 * assign b_runningbufspace. 3233 */ 3234 if (wcred == NULL) 3235 wcred = bp->b_wcred; 3236 else if (wcred != bp->b_wcred) 3237 wcred = NOCRED; 3238 vfs_busy_pages(bp, 0); 3239 3240 BO_LOCK(bo); 3241 /* 3242 * bp is protected by being locked, but nbp is not 3243 * and vfs_busy_pages() may sleep. We have to 3244 * recalculate nbp. 3245 */ 3246 nbp = TAILQ_NEXT(bp, b_bobufs); 3247 3248 /* 3249 * A list of these buffers is kept so that the 3250 * second loop knows which buffers have actually 3251 * been committed. This is necessary, since there 3252 * may be a race between the commit rpc and new 3253 * uncommitted writes on the file. 3254 */ 3255 bvec[bvecpos++] = bp; 3256 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3257 bp->b_dirtyoff; 3258 if (toff < off) 3259 off = toff; 3260 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3261 if (toff > endoff) 3262 endoff = toff; 3263 } 3264 BO_UNLOCK(bo); 3265 } 3266 if (bvecpos > 0) { 3267 /* 3268 * Commit data on the server, as required. 3269 * If all bufs are using the same wcred, then use that with 3270 * one call for all of them, otherwise commit each one 3271 * separately. 3272 */ 3273 if (wcred != NOCRED) 3274 retv = ncl_commit(vp, off, (int)(endoff - off), 3275 wcred, td); 3276 else { 3277 retv = 0; 3278 for (i = 0; i < bvecpos; i++) { 3279 off_t off, size; 3280 bp = bvec[i]; 3281 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3282 bp->b_dirtyoff; 3283 size = (u_quad_t)(bp->b_dirtyend 3284 - bp->b_dirtyoff); 3285 retv = ncl_commit(vp, off, (int)size, 3286 bp->b_wcred, td); 3287 if (retv) break; 3288 } 3289 } 3290 3291 if (retv == NFSERR_STALEWRITEVERF) 3292 ncl_clearcommit(vp->v_mount); 3293 3294 /* 3295 * Now, either mark the blocks I/O done or mark the 3296 * blocks dirty, depending on whether the commit 3297 * succeeded. 3298 */ 3299 for (i = 0; i < bvecpos; i++) { 3300 bp = bvec[i]; 3301 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3302 if (!NFSCL_FORCEDISM(vp->v_mount) && retv) { 3303 /* 3304 * Error, leave B_DELWRI intact 3305 */ 3306 vfs_unbusy_pages(bp); 3307 brelse(bp); 3308 } else { 3309 /* 3310 * Success, remove B_DELWRI ( bundirty() ). 3311 * 3312 * b_dirtyoff/b_dirtyend seem to be NFS 3313 * specific. We should probably move that 3314 * into bundirty(). XXX 3315 */ 3316 bufobj_wref(bo); 3317 bp->b_flags |= B_ASYNC; 3318 bundirty(bp); 3319 bp->b_flags &= ~B_DONE; 3320 bp->b_ioflags &= ~BIO_ERROR; 3321 bp->b_dirtyoff = bp->b_dirtyend = 0; 3322 bufdone(bp); 3323 } 3324 } 3325 } 3326 3327 /* 3328 * Start/do any write(s) that are required. 3329 */ 3330 loop: 3331 BO_LOCK(bo); 3332 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3333 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3334 if (waitfor != MNT_WAIT || passone) 3335 continue; 3336 3337 error = BUF_TIMELOCK(bp, 3338 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 3339 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 3340 if (error == 0) { 3341 BUF_UNLOCK(bp); 3342 goto loop; 3343 } 3344 if (error == ENOLCK) { 3345 error = 0; 3346 goto loop; 3347 } 3348 if (called_from_renewthread != 0) { 3349 /* 3350 * Return EIO so the flush will be retried 3351 * later. 3352 */ 3353 error = EIO; 3354 goto done; 3355 } 3356 if (newnfs_sigintr(nmp, td)) { 3357 error = EINTR; 3358 goto done; 3359 } 3360 if (slpflag == PCATCH) { 3361 slpflag = 0; 3362 slptimeo = 2 * hz; 3363 } 3364 goto loop; 3365 } 3366 if ((bp->b_flags & B_DELWRI) == 0) 3367 panic("nfs_fsync: not dirty"); 3368 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3369 BUF_UNLOCK(bp); 3370 continue; 3371 } 3372 BO_UNLOCK(bo); 3373 bremfree(bp); 3374 bp->b_flags |= B_ASYNC; 3375 bwrite(bp); 3376 if (newnfs_sigintr(nmp, td)) { 3377 error = EINTR; 3378 goto done; 3379 } 3380 goto loop; 3381 } 3382 if (passone) { 3383 passone = 0; 3384 BO_UNLOCK(bo); 3385 goto again; 3386 } 3387 if (waitfor == MNT_WAIT) { 3388 while (bo->bo_numoutput) { 3389 error = bufobj_wwait(bo, slpflag, slptimeo); 3390 if (error) { 3391 BO_UNLOCK(bo); 3392 if (called_from_renewthread != 0) { 3393 /* 3394 * Return EIO so that the flush will be 3395 * retried later. 3396 */ 3397 error = EIO; 3398 goto done; 3399 } 3400 error = newnfs_sigintr(nmp, td); 3401 if (error) 3402 goto done; 3403 if (slpflag == PCATCH) { 3404 slpflag = 0; 3405 slptimeo = 2 * hz; 3406 } 3407 BO_LOCK(bo); 3408 } 3409 } 3410 if (bo->bo_dirty.bv_cnt != 0 && commit) { 3411 BO_UNLOCK(bo); 3412 goto loop; 3413 } 3414 /* 3415 * Wait for all the async IO requests to drain 3416 */ 3417 BO_UNLOCK(bo); 3418 } else 3419 BO_UNLOCK(bo); 3420 if (NFSHASPNFS(nmp)) { 3421 nfscl_layoutcommit(vp, td); 3422 /* 3423 * Invalidate the attribute cache, since writes to a DS 3424 * won't update the size attribute. 3425 */ 3426 NFSLOCKNODE(np); 3427 np->n_attrstamp = 0; 3428 } else 3429 NFSLOCKNODE(np); 3430 if (np->n_flag & NWRITEERR) { 3431 error = np->n_error; 3432 np->n_flag &= ~NWRITEERR; 3433 } 3434 if (commit && bo->bo_dirty.bv_cnt == 0 && 3435 bo->bo_numoutput == 0) 3436 np->n_flag &= ~NMODIFIED; 3437 NFSUNLOCKNODE(np); 3438 done: 3439 if (bvec != NULL && bvec != bvec_on_stack) 3440 free(bvec, M_TEMP); 3441 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 3442 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0)) { 3443 if (trycnt++ < 5) { 3444 /* try, try again... */ 3445 passone = 1; 3446 wcred = NULL; 3447 bvec = NULL; 3448 bvecsize = 0; 3449 goto again; 3450 } 3451 vn_printf(vp, "ncl_flush failed"); 3452 error = called_from_renewthread != 0 ? EIO : EBUSY; 3453 } 3454 if (error == 0) { 3455 nanouptime(&ts); 3456 NFSLOCKNODE(np); 3457 np->n_localmodtime = ts; 3458 NFSUNLOCKNODE(np); 3459 } 3460 return (error); 3461 } 3462 3463 /* 3464 * NFS advisory byte-level locks. 3465 */ 3466 static int 3467 nfs_advlock(struct vop_advlock_args *ap) 3468 { 3469 struct vnode *vp = ap->a_vp; 3470 struct ucred *cred; 3471 struct nfsnode *np = VTONFS(ap->a_vp); 3472 struct proc *p = (struct proc *)ap->a_id; 3473 struct thread *td = curthread; /* XXX */ 3474 struct vattr va; 3475 int ret, error; 3476 u_quad_t size; 3477 struct nfsmount *nmp; 3478 3479 error = NFSVOPLOCK(vp, LK_EXCLUSIVE); 3480 if (error != 0) 3481 return (EBADF); 3482 nmp = VFSTONFS(vp->v_mount); 3483 if (!NFS_ISV4(vp) || (nmp->nm_flag & NFSMNT_NOLOCKD) != 0) { 3484 if ((nmp->nm_flag & NFSMNT_NOLOCKD) != 0) { 3485 size = np->n_size; 3486 NFSVOPUNLOCK(vp); 3487 error = lf_advlock(ap, &(vp->v_lockf), size); 3488 } else { 3489 if (nfs_advlock_p != NULL) 3490 error = nfs_advlock_p(ap); 3491 else { 3492 NFSVOPUNLOCK(vp); 3493 error = ENOLCK; 3494 } 3495 } 3496 if (error == 0 && ap->a_op == F_SETLK) { 3497 error = NFSVOPLOCK(vp, LK_SHARED); 3498 if (error == 0) { 3499 /* Mark that a file lock has been acquired. */ 3500 NFSLOCKNODE(np); 3501 np->n_flag |= NHASBEENLOCKED; 3502 NFSUNLOCKNODE(np); 3503 NFSVOPUNLOCK(vp); 3504 } 3505 } 3506 return (error); 3507 } else if ((ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3508 if (vp->v_type != VREG) { 3509 error = EINVAL; 3510 goto out; 3511 } 3512 if ((ap->a_flags & F_POSIX) != 0) 3513 cred = p->p_ucred; 3514 else 3515 cred = td->td_ucred; 3516 3517 /* 3518 * If this is unlocking a write locked region, flush and 3519 * commit them before unlocking. This is required by 3520 * RFC3530 Sec. 9.3.2. 3521 */ 3522 if (ap->a_op == F_UNLCK && 3523 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3524 ap->a_flags)) 3525 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3526 3527 /* 3528 * Mark NFS node as might have acquired a lock. 3529 * This is separate from NHASBEENLOCKED, because it must 3530 * be done before the nfsrpc_advlock() call, which might 3531 * add a nfscllock structure to the client state. 3532 * It is used to check for the case where a nfscllock 3533 * state structure cannot exist for the file. 3534 * Only done for "oneopenown" NFSv4.1/4.2 mounts. 3535 */ 3536 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) { 3537 NFSLOCKNODE(np); 3538 np->n_flag |= NMIGHTBELOCKED; 3539 NFSUNLOCKNODE(np); 3540 } 3541 3542 /* 3543 * Loop around doing the lock op, while a blocking lock 3544 * must wait for the lock op to succeed. 3545 */ 3546 do { 3547 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3548 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3549 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3550 ap->a_op == F_SETLK) { 3551 NFSVOPUNLOCK(vp); 3552 error = nfs_catnap(PZERO | PCATCH, ret, 3553 "ncladvl"); 3554 if (error) 3555 return (EINTR); 3556 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3557 if (VN_IS_DOOMED(vp)) { 3558 error = EBADF; 3559 goto out; 3560 } 3561 } 3562 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3563 ap->a_op == F_SETLK); 3564 if (ret == NFSERR_DENIED) { 3565 error = EAGAIN; 3566 goto out; 3567 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3568 error = ret; 3569 goto out; 3570 } else if (ret != 0) { 3571 error = EACCES; 3572 goto out; 3573 } 3574 3575 /* 3576 * Now, if we just got a lock, invalidate data in the buffer 3577 * cache, as required, so that the coherency conforms with 3578 * RFC3530 Sec. 9.3.2. 3579 */ 3580 if (ap->a_op == F_SETLK) { 3581 if ((np->n_flag & NMODIFIED) == 0) { 3582 np->n_attrstamp = 0; 3583 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3584 ret = VOP_GETATTR(vp, &va, cred); 3585 } 3586 if ((np->n_flag & NMODIFIED) || ret || 3587 np->n_change != va.va_filerev) { 3588 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3589 np->n_attrstamp = 0; 3590 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3591 ret = VOP_GETATTR(vp, &va, cred); 3592 if (!ret) { 3593 np->n_mtime = va.va_mtime; 3594 np->n_change = va.va_filerev; 3595 } 3596 } 3597 /* Mark that a file lock has been acquired. */ 3598 NFSLOCKNODE(np); 3599 np->n_flag |= NHASBEENLOCKED; 3600 NFSUNLOCKNODE(np); 3601 } 3602 } else 3603 error = EOPNOTSUPP; 3604 out: 3605 NFSVOPUNLOCK(vp); 3606 return (error); 3607 } 3608 3609 /* 3610 * NFS advisory byte-level locks. 3611 */ 3612 static int 3613 nfs_advlockasync(struct vop_advlockasync_args *ap) 3614 { 3615 struct vnode *vp = ap->a_vp; 3616 u_quad_t size; 3617 int error; 3618 3619 error = NFSVOPLOCK(vp, LK_SHARED); 3620 if (error) 3621 return (error); 3622 if (NFS_ISV4(vp)) { 3623 NFSVOPUNLOCK(vp); 3624 return (EOPNOTSUPP); 3625 } 3626 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3627 size = VTONFS(vp)->n_size; 3628 NFSVOPUNLOCK(vp); 3629 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3630 } else { 3631 NFSVOPUNLOCK(vp); 3632 error = EOPNOTSUPP; 3633 } 3634 return (error); 3635 } 3636 3637 /* 3638 * Print out the contents of an nfsnode. 3639 */ 3640 static int 3641 nfs_print(struct vop_print_args *ap) 3642 { 3643 struct vnode *vp = ap->a_vp; 3644 struct nfsnode *np = VTONFS(vp); 3645 3646 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3647 (uintmax_t)np->n_vattr.na_fsid); 3648 if (vp->v_type == VFIFO) 3649 fifo_printinfo(vp); 3650 printf("\n"); 3651 return (0); 3652 } 3653 3654 /* 3655 * nfs special file access vnode op. 3656 * Essentially just get vattr and then imitate iaccess() since the device is 3657 * local to the client. 3658 */ 3659 static int 3660 nfsspec_access(struct vop_access_args *ap) 3661 { 3662 struct vattr *vap; 3663 struct ucred *cred = ap->a_cred; 3664 struct vnode *vp = ap->a_vp; 3665 accmode_t accmode = ap->a_accmode; 3666 struct vattr vattr; 3667 int error; 3668 3669 /* 3670 * Disallow write attempts on filesystems mounted read-only; 3671 * unless the file is a socket, fifo, or a block or character 3672 * device resident on the filesystem. 3673 */ 3674 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3675 switch (vp->v_type) { 3676 case VREG: 3677 case VDIR: 3678 case VLNK: 3679 return (EROFS); 3680 default: 3681 break; 3682 } 3683 } 3684 vap = &vattr; 3685 error = VOP_GETATTR(vp, vap, cred); 3686 if (error) 3687 goto out; 3688 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3689 accmode, cred); 3690 out: 3691 return error; 3692 } 3693 3694 /* 3695 * Read wrapper for fifos. 3696 */ 3697 static int 3698 nfsfifo_read(struct vop_read_args *ap) 3699 { 3700 struct nfsnode *np = VTONFS(ap->a_vp); 3701 int error; 3702 3703 /* 3704 * Set access flag. 3705 */ 3706 NFSLOCKNODE(np); 3707 np->n_flag |= NACC; 3708 vfs_timestamp(&np->n_atim); 3709 NFSUNLOCKNODE(np); 3710 error = fifo_specops.vop_read(ap); 3711 return error; 3712 } 3713 3714 /* 3715 * Write wrapper for fifos. 3716 */ 3717 static int 3718 nfsfifo_write(struct vop_write_args *ap) 3719 { 3720 struct nfsnode *np = VTONFS(ap->a_vp); 3721 3722 /* 3723 * Set update flag. 3724 */ 3725 NFSLOCKNODE(np); 3726 np->n_flag |= NUPD; 3727 vfs_timestamp(&np->n_mtim); 3728 NFSUNLOCKNODE(np); 3729 return(fifo_specops.vop_write(ap)); 3730 } 3731 3732 /* 3733 * Close wrapper for fifos. 3734 * 3735 * Update the times on the nfsnode then do fifo close. 3736 */ 3737 static int 3738 nfsfifo_close(struct vop_close_args *ap) 3739 { 3740 struct vnode *vp = ap->a_vp; 3741 struct nfsnode *np = VTONFS(vp); 3742 struct vattr vattr; 3743 struct timespec ts; 3744 3745 NFSLOCKNODE(np); 3746 if (np->n_flag & (NACC | NUPD)) { 3747 vfs_timestamp(&ts); 3748 if (np->n_flag & NACC) 3749 np->n_atim = ts; 3750 if (np->n_flag & NUPD) 3751 np->n_mtim = ts; 3752 np->n_flag |= NCHG; 3753 if (vrefcnt(vp) == 1 && 3754 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3755 VATTR_NULL(&vattr); 3756 if (np->n_flag & NACC) 3757 vattr.va_atime = np->n_atim; 3758 if (np->n_flag & NUPD) 3759 vattr.va_mtime = np->n_mtim; 3760 NFSUNLOCKNODE(np); 3761 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3762 goto out; 3763 } 3764 } 3765 NFSUNLOCKNODE(np); 3766 out: 3767 return (fifo_specops.vop_close(ap)); 3768 } 3769 3770 static int 3771 nfs_getacl(struct vop_getacl_args *ap) 3772 { 3773 int error; 3774 3775 if (ap->a_type != ACL_TYPE_NFS4) 3776 return (EOPNOTSUPP); 3777 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp); 3778 if (error > NFSERR_STALE) { 3779 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3780 error = EPERM; 3781 } 3782 return (error); 3783 } 3784 3785 static int 3786 nfs_setacl(struct vop_setacl_args *ap) 3787 { 3788 int error; 3789 3790 if (ap->a_type != ACL_TYPE_NFS4) 3791 return (EOPNOTSUPP); 3792 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp); 3793 if (error > NFSERR_STALE) { 3794 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3795 error = EPERM; 3796 } 3797 return (error); 3798 } 3799 3800 /* 3801 * VOP_ADVISE for NFS. 3802 * Just return 0 for any errors, since it is just a hint. 3803 */ 3804 static int 3805 nfs_advise(struct vop_advise_args *ap) 3806 { 3807 struct thread *td = curthread; 3808 struct nfsmount *nmp; 3809 uint64_t len; 3810 int error; 3811 3812 /* 3813 * First do vop_stdadvise() to handle the buffer cache. 3814 */ 3815 error = vop_stdadvise(ap); 3816 if (error != 0) 3817 return (error); 3818 if (ap->a_start < 0 || ap->a_end < 0) 3819 return (0); 3820 if (ap->a_end == OFF_MAX) 3821 len = 0; 3822 else if (ap->a_end < ap->a_start) 3823 return (0); 3824 else 3825 len = ap->a_end - ap->a_start + 1; 3826 nmp = VFSTONFS(ap->a_vp->v_mount); 3827 mtx_lock(&nmp->nm_mtx); 3828 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3829 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) == 3830 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) { 3831 mtx_unlock(&nmp->nm_mtx); 3832 return (0); 3833 } 3834 mtx_unlock(&nmp->nm_mtx); 3835 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice, 3836 td->td_ucred, td); 3837 if (error == NFSERR_NOTSUPP) { 3838 mtx_lock(&nmp->nm_mtx); 3839 nmp->nm_privflag |= NFSMNTP_NOADVISE; 3840 mtx_unlock(&nmp->nm_mtx); 3841 } 3842 return (0); 3843 } 3844 3845 /* 3846 * nfs allocate call 3847 */ 3848 static int 3849 nfs_allocate(struct vop_allocate_args *ap) 3850 { 3851 struct vnode *vp = ap->a_vp; 3852 struct thread *td = curthread; 3853 struct nfsvattr nfsva; 3854 struct nfsmount *nmp; 3855 struct nfsnode *np; 3856 off_t alen; 3857 int attrflag, error, ret; 3858 struct timespec ts; 3859 struct uio io; 3860 3861 attrflag = 0; 3862 nmp = VFSTONFS(vp->v_mount); 3863 np = VTONFS(vp); 3864 mtx_lock(&nmp->nm_mtx); 3865 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3866 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) { 3867 mtx_unlock(&nmp->nm_mtx); 3868 alen = *ap->a_len; 3869 if ((uint64_t)alen > nfs_maxalloclen) 3870 alen = nfs_maxalloclen; 3871 3872 /* Check the file size limit. */ 3873 io.uio_offset = *ap->a_offset; 3874 io.uio_resid = alen; 3875 error = vn_rlimit_fsize(vp, &io, td); 3876 3877 /* 3878 * Flush first to ensure that the allocate adds to the 3879 * file's allocation on the server. 3880 */ 3881 if (error == 0) { 3882 vnode_pager_clean_sync(vp); 3883 error = ncl_flush(vp, MNT_WAIT, td, 1, 0); 3884 } 3885 if (error == 0) 3886 error = nfsrpc_allocate(vp, *ap->a_offset, alen, 3887 &nfsva, &attrflag, ap->a_cred, td); 3888 if (error == 0) { 3889 *ap->a_offset += alen; 3890 *ap->a_len -= alen; 3891 nanouptime(&ts); 3892 NFSLOCKNODE(np); 3893 np->n_localmodtime = ts; 3894 NFSUNLOCKNODE(np); 3895 } else if (error == NFSERR_NOTSUPP) { 3896 mtx_lock(&nmp->nm_mtx); 3897 nmp->nm_privflag |= NFSMNTP_NOALLOCATE; 3898 mtx_unlock(&nmp->nm_mtx); 3899 error = EINVAL; 3900 } 3901 } else { 3902 mtx_unlock(&nmp->nm_mtx); 3903 error = EINVAL; 3904 } 3905 if (attrflag != 0) { 3906 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3907 if (error == 0 && ret != 0) 3908 error = ret; 3909 } 3910 if (error != 0) 3911 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3912 return (error); 3913 } 3914 3915 /* 3916 * nfs deallocate call 3917 */ 3918 static int 3919 nfs_deallocate(struct vop_deallocate_args *ap) 3920 { 3921 struct vnode *vp = ap->a_vp; 3922 struct thread *td = curthread; 3923 struct nfsvattr nfsva; 3924 struct nfsmount *nmp; 3925 struct nfsnode *np; 3926 off_t tlen, mlen; 3927 int attrflag, error, ret; 3928 bool clipped; 3929 struct timespec ts; 3930 3931 error = 0; 3932 attrflag = 0; 3933 nmp = VFSTONFS(vp->v_mount); 3934 np = VTONFS(vp); 3935 mtx_lock(&nmp->nm_mtx); 3936 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3937 (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) { 3938 mtx_unlock(&nmp->nm_mtx); 3939 tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len); 3940 NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n", 3941 (intmax_t)*ap->a_offset, (intmax_t)tlen, 3942 (uintmax_t)nmp->nm_maxfilesize); 3943 if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) { 3944 /* Avoid EFBIG error return from the NFSv4.2 server. */ 3945 *ap->a_len = 0; 3946 return (0); 3947 } 3948 clipped = false; 3949 if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize) 3950 tlen = nmp->nm_maxfilesize - *ap->a_offset; 3951 if ((uint64_t)*ap->a_offset < np->n_size) { 3952 /* Limit the len to nfs_maxalloclen before EOF. */ 3953 mlen = omin((off_t)np->n_size - *ap->a_offset, tlen); 3954 if ((uint64_t)mlen > nfs_maxalloclen) { 3955 NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n"); 3956 tlen = nfs_maxalloclen; 3957 clipped = true; 3958 } 3959 } 3960 if (error == 0) 3961 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 3962 if (error == 0) { 3963 vnode_pager_purge_range(vp, *ap->a_offset, 3964 *ap->a_offset + tlen); 3965 error = nfsrpc_deallocate(vp, *ap->a_offset, tlen, 3966 &nfsva, &attrflag, ap->a_cred, td); 3967 NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error); 3968 } 3969 if (error == 0) { 3970 NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n", 3971 attrflag, (uintmax_t)nfsva.na_size); 3972 nanouptime(&ts); 3973 NFSLOCKNODE(np); 3974 np->n_localmodtime = ts; 3975 NFSUNLOCKNODE(np); 3976 if (attrflag != 0) { 3977 if ((uint64_t)*ap->a_offset < nfsva.na_size) 3978 *ap->a_offset += omin((off_t) 3979 nfsva.na_size - *ap->a_offset, 3980 tlen); 3981 } 3982 if (clipped && tlen < *ap->a_len) 3983 *ap->a_len -= tlen; 3984 else 3985 *ap->a_len = 0; 3986 } else if (error == NFSERR_NOTSUPP) { 3987 mtx_lock(&nmp->nm_mtx); 3988 nmp->nm_privflag |= NFSMNTP_NODEALLOCATE; 3989 mtx_unlock(&nmp->nm_mtx); 3990 } 3991 } else { 3992 mtx_unlock(&nmp->nm_mtx); 3993 error = EIO; 3994 } 3995 /* 3996 * If the NFS server cannot perform the Deallocate operation, just call 3997 * vop_stddeallocate() to perform it. 3998 */ 3999 if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) { 4000 error = vop_stddeallocate(ap); 4001 NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error); 4002 } 4003 if (attrflag != 0) { 4004 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4005 if (error == 0 && ret != 0) 4006 error = ret; 4007 } 4008 if (error != 0) 4009 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 4010 return (error); 4011 } 4012 4013 /* 4014 * nfs copy_file_range call 4015 */ 4016 static int 4017 nfs_copy_file_range(struct vop_copy_file_range_args *ap) 4018 { 4019 struct vnode *invp = ap->a_invp; 4020 struct vnode *outvp = ap->a_outvp; 4021 struct mount *mp; 4022 vm_object_t invp_obj; 4023 struct nfsvattr innfsva, outnfsva; 4024 struct vattr va, *vap; 4025 struct uio io; 4026 struct nfsmount *nmp; 4027 struct nfsnode *np; 4028 size_t len, len2; 4029 ssize_t r; 4030 int error, inattrflag, outattrflag, ret, ret2, invp_lock; 4031 off_t inoff, outoff; 4032 bool consecutive, must_commit, onevp, toeof, tryclone, tryoutcred; 4033 bool mustclone; 4034 4035 /* 4036 * NFSv4.2 Copy is not permitted for infile == outfile. 4037 * The NFSv4.2 Clone operation does work on non-overlapping 4038 * byte ranges in the same file, but only if offsets 4039 * (and len if not to EOF) are aligned properly. 4040 * TODO: copy_file_range() between multiple NFS mountpoints 4041 * --> This is not possible now, since each mount appears to 4042 * the NFSv4.n server as a separate client. 4043 */ 4044 if ((invp == outvp && (ap->a_flags & COPY_FILE_RANGE_CLONE) == 0) || 4045 (invp != outvp && invp->v_mount != outvp->v_mount)) { 4046 generic_copy: 4047 return (ENOSYS); 4048 } 4049 if (invp == outvp) { 4050 onevp = true; 4051 invp_lock = LK_EXCLUSIVE; 4052 } else { 4053 onevp = false; 4054 invp_lock = LK_SHARED; 4055 } 4056 mustclone = false; 4057 if (onevp || (ap->a_flags & COPY_FILE_RANGE_CLONE) != 0) 4058 mustclone = true; 4059 relock: 4060 inoff = *ap->a_inoffp; 4061 outoff = *ap->a_outoffp; 4062 4063 /* Lock vnode(s), avoiding risk of deadlock. */ 4064 do { 4065 mp = NULL; 4066 error = vn_start_write(outvp, &mp, V_WAIT); 4067 if (error == 0) { 4068 error = vn_lock(outvp, LK_EXCLUSIVE); 4069 if (error == 0) { 4070 if (onevp) 4071 break; 4072 error = vn_lock(invp, invp_lock | LK_NOWAIT); 4073 if (error == 0) 4074 break; 4075 VOP_UNLOCK(outvp); 4076 if (mp != NULL) 4077 vn_finished_write(mp); 4078 mp = NULL; 4079 error = vn_lock(invp, invp_lock); 4080 if (error == 0) 4081 VOP_UNLOCK(invp); 4082 } 4083 } 4084 if (mp != NULL) 4085 vn_finished_write(mp); 4086 } while (error == 0); 4087 if (error != 0) 4088 return (error); 4089 4090 /* 4091 * More reasons to avoid nfs copy/clone: not NFSv4.2, explicitly 4092 * disabled or requires cloning and unable to clone. 4093 * Only clone if the clone_blksize attribute is supported 4094 * and the clone_blksize is greater than 0. 4095 * Alignment of offsets and length will be checked later. 4096 */ 4097 nmp = VFSTONFS(invp->v_mount); 4098 np = VTONFS(invp); 4099 mtx_lock(&nmp->nm_mtx); 4100 if ((nmp->nm_privflag & NFSMNTP_NOCOPY) != 0) 4101 mustclone = true; 4102 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4103 (mustclone && (!NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, 4104 NFSATTRBIT_CLONEBLKSIZE) || nmp->nm_cloneblksize == 0))) { 4105 mtx_unlock(&nmp->nm_mtx); 4106 VOP_UNLOCK(invp); 4107 if (!onevp) 4108 VOP_UNLOCK(outvp); /* For onevp, same as invp. */ 4109 if (mp != NULL) 4110 vn_finished_write(mp); 4111 goto generic_copy; 4112 } 4113 mtx_unlock(&nmp->nm_mtx); 4114 4115 /* 4116 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer? 4117 */ 4118 io.uio_offset = *ap->a_outoffp; 4119 io.uio_resid = *ap->a_lenp; 4120 error = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd); 4121 *ap->a_lenp = io.uio_resid; 4122 /* 4123 * No need to call vn_rlimit_fsizex_res before return, since the uio is 4124 * local. 4125 */ 4126 4127 /* 4128 * Flush the input file so that the data is up to date before 4129 * the copy. Flush writes for the output file so that they 4130 * do not overwrite the data copied to the output file by the Copy. 4131 * Set the commit argument for both flushes so that the data is on 4132 * stable storage before the Copy RPC. This is done in case the 4133 * server reboots during the Copy and needs to be redone. 4134 */ 4135 if (error == 0) { 4136 invp_obj = invp->v_object; 4137 if (invp_obj != NULL && vm_object_mightbedirty(invp_obj)) { 4138 if (invp_lock != LK_EXCLUSIVE) { 4139 KASSERT(!onevp, ("nfs_copy_file_range: " 4140 "invp_lock LK_SHARED for onevp")); 4141 invp_lock = LK_EXCLUSIVE; 4142 VOP_UNLOCK(invp); 4143 VOP_UNLOCK(outvp); 4144 if (mp != NULL) 4145 vn_finished_write(mp); 4146 goto relock; 4147 } 4148 vnode_pager_clean_sync(invp); 4149 } 4150 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0); 4151 } 4152 if (error == 0) 4153 error = ncl_vinvalbuf(outvp, V_SAVE, curthread, 0); 4154 4155 /* Do the actual NFSv4.2 RPC. */ 4156 ret = ret2 = 0; 4157 len = *ap->a_lenp; 4158 mtx_lock(&nmp->nm_mtx); 4159 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0) 4160 consecutive = true; 4161 else 4162 consecutive = false; 4163 mtx_unlock(&nmp->nm_mtx); 4164 tryoutcred = true; 4165 must_commit = false; 4166 toeof = false; 4167 4168 if (error == 0) { 4169 vap = &VTONFS(invp)->n_vattr.na_vattr; 4170 error = VOP_GETATTR(invp, vap, ap->a_incred); 4171 if (error == 0) { 4172 /* 4173 * Clip "len" at va_size so that RFC compliant servers 4174 * will not reply NFSERR_INVAL. 4175 * Setting "len == 0" for the RPC would be preferred, 4176 * but some Linux servers do not support that. 4177 * If the len is being set to 0, do a Setattr RPC to 4178 * set the server's atime. This behaviour was the 4179 * preferred one for the FreeBSD "collective". 4180 */ 4181 if (inoff >= vap->va_size) { 4182 *ap->a_lenp = len = 0; 4183 if ((nmp->nm_mountp->mnt_flag & MNT_NOATIME) == 4184 0) { 4185 VATTR_NULL(&va); 4186 va.va_atime.tv_sec = 0; 4187 va.va_atime.tv_nsec = 0; 4188 va.va_vaflags = VA_UTIMES_NULL; 4189 inattrflag = 0; 4190 error = nfsrpc_setattr(invp, &va, NULL, 4191 ap->a_incred, curthread, &innfsva, 4192 &inattrflag); 4193 if (inattrflag != 0) 4194 ret = nfscl_loadattrcache(&invp, 4195 &innfsva, NULL, 0, 1); 4196 if (error == 0 && ret != 0) 4197 error = ret; 4198 } 4199 } else if (inoff + len >= vap->va_size) { 4200 toeof = true; 4201 *ap->a_lenp = len = vap->va_size - inoff; 4202 } 4203 } else 4204 error = 0; 4205 } 4206 4207 /* 4208 * For cloning, the offsets must be clone blksize aligned and 4209 * the len must be blksize aligned unless it goes to EOF on 4210 * the input file. 4211 */ 4212 tryclone = false; 4213 if (len > 0) { 4214 if (error == 0 && NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, 4215 NFSATTRBIT_CLONEBLKSIZE) && nmp->nm_cloneblksize != 0 && 4216 (inoff % nmp->nm_cloneblksize) == 0 && 4217 (outoff % nmp->nm_cloneblksize) == 0 && 4218 (toeof || (len % nmp->nm_cloneblksize) == 0)) 4219 tryclone = true; 4220 else if (mustclone) 4221 error = ENOSYS; 4222 } 4223 4224 /* 4225 * len will be set to 0 upon a successful Copy RPC. 4226 * As such, this only loops when the Copy/Clone RPC needs to be retried. 4227 */ 4228 while (len > 0 && error == 0) { 4229 inattrflag = outattrflag = 0; 4230 len2 = len; 4231 if (tryclone) { 4232 if (tryoutcred) 4233 error = nfsrpc_clone(invp, ap->a_inoffp, outvp, 4234 ap->a_outoffp, &len2, toeof, &inattrflag, 4235 &innfsva, &outattrflag, &outnfsva, 4236 ap->a_outcred); 4237 else 4238 error = nfsrpc_clone(invp, ap->a_inoffp, outvp, 4239 ap->a_outoffp, &len2, toeof, &inattrflag, 4240 &innfsva, &outattrflag, &outnfsva, 4241 ap->a_incred); 4242 } else { 4243 if (tryoutcred) 4244 error = nfsrpc_copy_file_range(invp, 4245 ap->a_inoffp, outvp, ap->a_outoffp, &len2, 4246 ap->a_flags, &inattrflag, &innfsva, 4247 &outattrflag, &outnfsva, 4248 ap->a_outcred, consecutive, &must_commit); 4249 else 4250 error = nfsrpc_copy_file_range(invp, 4251 ap->a_inoffp, outvp, ap->a_outoffp, &len2, 4252 ap->a_flags, &inattrflag, &innfsva, 4253 &outattrflag, &outnfsva, 4254 ap->a_incred, consecutive, &must_commit); 4255 } 4256 if (inattrflag != 0) 4257 ret = nfscl_loadattrcache(&invp, &innfsva, NULL, 0, 1); 4258 if (outattrflag != 0) 4259 ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL, 4260 1, 1); 4261 if (error == 0) { 4262 if (consecutive == false) { 4263 if (len2 == len) { 4264 mtx_lock(&nmp->nm_mtx); 4265 nmp->nm_privflag |= 4266 NFSMNTP_NOCONSECUTIVE; 4267 mtx_unlock(&nmp->nm_mtx); 4268 } else 4269 error = NFSERR_OFFLOADNOREQS; 4270 } 4271 *ap->a_lenp = len2; 4272 len = 0; 4273 if (len2 > 0 && must_commit && error == 0) 4274 error = ncl_commit(outvp, outoff, *ap->a_lenp, 4275 ap->a_outcred, curthread); 4276 if (error == 0 && ret != 0) 4277 error = ret; 4278 if (error == 0 && ret2 != 0) 4279 error = ret2; 4280 } else if (error == NFSERR_OFFLOADNOREQS && consecutive) { 4281 /* 4282 * Try consecutive == false, which is ok only if all 4283 * bytes are copied. 4284 * If only some bytes were copied when consecutive 4285 * is false, there is no way to know which bytes 4286 * still need to be written. 4287 */ 4288 consecutive = false; 4289 error = 0; 4290 } else if (error == NFSERR_ACCES && tryoutcred) { 4291 /* Try again with incred. */ 4292 tryoutcred = false; 4293 error = 0; 4294 } else if (tryclone && error != 0) { 4295 if (mustclone) { 4296 error = ENOSYS; 4297 } else { 4298 tryclone = false; 4299 error = 0; 4300 } 4301 } 4302 if (error == NFSERR_STALEWRITEVERF) { 4303 /* 4304 * Server rebooted, so do it all again. 4305 */ 4306 *ap->a_inoffp = inoff; 4307 *ap->a_outoffp = outoff; 4308 len = *ap->a_lenp; 4309 must_commit = false; 4310 error = 0; 4311 } 4312 } 4313 VOP_UNLOCK(invp); 4314 if (!onevp) 4315 VOP_UNLOCK(outvp); /* For onevp, same as invp. */ 4316 if (mp != NULL) 4317 vn_finished_write(mp); 4318 if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS || 4319 error == NFSERR_ACCES || error == ENOSYS) { 4320 /* 4321 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can 4322 * use a_incred for the read and a_outcred for the write, so 4323 * try this for NFSERR_ACCES failures for the Copy. 4324 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can 4325 * never succeed, so disable it. 4326 */ 4327 if (error != NFSERR_ACCES && error != ENOSYS) { 4328 /* Can never do Copy on this mount. */ 4329 mtx_lock(&nmp->nm_mtx); 4330 nmp->nm_privflag |= NFSMNTP_NOCOPY; 4331 mtx_unlock(&nmp->nm_mtx); 4332 } 4333 *ap->a_inoffp = inoff; 4334 *ap->a_outoffp = outoff; 4335 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 4336 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 4337 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 4338 } else if (error != 0) 4339 *ap->a_lenp = 0; 4340 4341 if (error != 0) 4342 error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0); 4343 return (error); 4344 } 4345 4346 /* 4347 * nfs ioctl call 4348 */ 4349 static int 4350 nfs_ioctl(struct vop_ioctl_args *ap) 4351 { 4352 struct vnode *vp = ap->a_vp; 4353 struct nfsvattr nfsva; 4354 struct nfsmount *nmp; 4355 int attrflag, content, error, ret; 4356 bool eof = false; /* shut up compiler. */ 4357 4358 /* Do the actual NFSv4.2 RPC. */ 4359 switch (ap->a_command) { 4360 case FIOSEEKDATA: 4361 content = NFSV4CONTENT_DATA; 4362 break; 4363 case FIOSEEKHOLE: 4364 content = NFSV4CONTENT_HOLE; 4365 break; 4366 default: 4367 return (ENOTTY); 4368 } 4369 4370 error = vn_lock(vp, LK_EXCLUSIVE); 4371 if (error != 0) 4372 return (EBADF); 4373 4374 if (vp->v_type != VREG) { 4375 VOP_UNLOCK(vp); 4376 return (ENOTTY); 4377 } 4378 nmp = VFSTONFS(vp->v_mount); 4379 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) { 4380 VOP_UNLOCK(vp); 4381 error = vop_stdioctl(ap); 4382 return (error); 4383 } 4384 4385 attrflag = 0; 4386 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size) 4387 error = ENXIO; 4388 else { 4389 /* 4390 * Flush all writes, so that the server is up to date. 4391 * Although a Commit is not required, the commit argument 4392 * is set so that, for a pNFS File/Flexible File Layout 4393 * server, the LayoutCommit will be done to ensure the file 4394 * size is up to date on the Metadata Server. 4395 */ 4396 4397 vnode_pager_clean_sync(vp); 4398 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0); 4399 if (error == 0) 4400 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof, 4401 content, ap->a_cred, &nfsva, &attrflag); 4402 /* If at eof for FIOSEEKDATA, return ENXIO. */ 4403 if (eof && error == 0 && content == NFSV4CONTENT_DATA) 4404 error = ENXIO; 4405 } 4406 if (attrflag != 0) { 4407 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4408 if (error == 0 && ret != 0) 4409 error = ret; 4410 } 4411 NFSVOPUNLOCK(vp); 4412 4413 if (error != 0) 4414 error = ENXIO; 4415 return (error); 4416 } 4417 4418 /* 4419 * nfs getextattr call 4420 */ 4421 static int 4422 nfs_getextattr(struct vop_getextattr_args *ap) 4423 { 4424 struct vnode *vp = ap->a_vp; 4425 struct nfsmount *nmp; 4426 struct ucred *cred; 4427 struct thread *td = ap->a_td; 4428 struct nfsvattr nfsva; 4429 ssize_t len; 4430 int attrflag, error, ret; 4431 4432 nmp = VFSTONFS(vp->v_mount); 4433 mtx_lock(&nmp->nm_mtx); 4434 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4435 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4436 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4437 mtx_unlock(&nmp->nm_mtx); 4438 return (EOPNOTSUPP); 4439 } 4440 mtx_unlock(&nmp->nm_mtx); 4441 4442 cred = ap->a_cred; 4443 if (cred == NULL) 4444 cred = td->td_ucred; 4445 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4446 attrflag = 0; 4447 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva, 4448 &attrflag, cred, td); 4449 if (attrflag != 0) { 4450 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4451 if (error == 0 && ret != 0) 4452 error = ret; 4453 } 4454 if (error == 0 && ap->a_size != NULL) 4455 *ap->a_size = len; 4456 4457 switch (error) { 4458 case NFSERR_NOTSUPP: 4459 case NFSERR_OPILLEGAL: 4460 mtx_lock(&nmp->nm_mtx); 4461 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4462 mtx_unlock(&nmp->nm_mtx); 4463 error = EOPNOTSUPP; 4464 break; 4465 case NFSERR_NOXATTR: 4466 case NFSERR_XATTR2BIG: 4467 error = ENOATTR; 4468 break; 4469 default: 4470 error = nfscl_maperr(td, error, 0, 0); 4471 break; 4472 } 4473 return (error); 4474 } 4475 4476 /* 4477 * nfs setextattr call 4478 */ 4479 static int 4480 nfs_setextattr(struct vop_setextattr_args *ap) 4481 { 4482 struct vnode *vp = ap->a_vp; 4483 struct nfsmount *nmp; 4484 struct ucred *cred; 4485 struct thread *td = ap->a_td; 4486 struct nfsvattr nfsva; 4487 int attrflag, error, ret; 4488 4489 nmp = VFSTONFS(vp->v_mount); 4490 mtx_lock(&nmp->nm_mtx); 4491 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4492 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4493 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4494 mtx_unlock(&nmp->nm_mtx); 4495 return (EOPNOTSUPP); 4496 } 4497 mtx_unlock(&nmp->nm_mtx); 4498 4499 if (ap->a_uio->uio_resid < 0) 4500 return (EINVAL); 4501 cred = ap->a_cred; 4502 if (cred == NULL) 4503 cred = td->td_ucred; 4504 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4505 attrflag = 0; 4506 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva, 4507 &attrflag, cred, td); 4508 if (attrflag != 0) { 4509 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4510 if (error == 0 && ret != 0) 4511 error = ret; 4512 } 4513 4514 switch (error) { 4515 case NFSERR_NOTSUPP: 4516 case NFSERR_OPILLEGAL: 4517 mtx_lock(&nmp->nm_mtx); 4518 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4519 mtx_unlock(&nmp->nm_mtx); 4520 error = EOPNOTSUPP; 4521 break; 4522 case NFSERR_NOXATTR: 4523 case NFSERR_XATTR2BIG: 4524 error = ENOATTR; 4525 break; 4526 default: 4527 error = nfscl_maperr(td, error, 0, 0); 4528 break; 4529 } 4530 return (error); 4531 } 4532 4533 /* 4534 * nfs listextattr call 4535 */ 4536 static int 4537 nfs_listextattr(struct vop_listextattr_args *ap) 4538 { 4539 struct vnode *vp = ap->a_vp; 4540 struct nfsmount *nmp; 4541 struct ucred *cred; 4542 struct thread *td = ap->a_td; 4543 struct nfsvattr nfsva; 4544 size_t len, len2; 4545 uint64_t cookie; 4546 int attrflag, error, ret; 4547 bool eof; 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 cred = ap->a_cred; 4560 if (cred == NULL) 4561 cred = td->td_ucred; 4562 4563 /* Loop around doing List Extended Attribute RPCs. */ 4564 eof = false; 4565 cookie = 0; 4566 len2 = 0; 4567 error = 0; 4568 while (!eof && error == 0) { 4569 len = nmp->nm_rsize; 4570 attrflag = 0; 4571 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof, 4572 &nfsva, &attrflag, cred, td); 4573 if (attrflag != 0) { 4574 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4575 if (error == 0 && ret != 0) 4576 error = ret; 4577 } 4578 if (error == 0) { 4579 len2 += len; 4580 if (len2 > SSIZE_MAX) 4581 error = ENOATTR; 4582 } 4583 } 4584 if (error == 0 && ap->a_size != NULL) 4585 *ap->a_size = len2; 4586 4587 switch (error) { 4588 case NFSERR_NOTSUPP: 4589 case NFSERR_OPILLEGAL: 4590 mtx_lock(&nmp->nm_mtx); 4591 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4592 mtx_unlock(&nmp->nm_mtx); 4593 error = EOPNOTSUPP; 4594 break; 4595 case NFSERR_NOXATTR: 4596 case NFSERR_XATTR2BIG: 4597 error = ENOATTR; 4598 break; 4599 default: 4600 error = nfscl_maperr(td, error, 0, 0); 4601 break; 4602 } 4603 return (error); 4604 } 4605 4606 /* 4607 * nfs setextattr call 4608 */ 4609 static int 4610 nfs_deleteextattr(struct vop_deleteextattr_args *ap) 4611 { 4612 struct vnode *vp = ap->a_vp; 4613 struct nfsmount *nmp; 4614 struct nfsvattr nfsva; 4615 int attrflag, error, ret; 4616 4617 nmp = VFSTONFS(vp->v_mount); 4618 mtx_lock(&nmp->nm_mtx); 4619 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4620 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4621 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4622 mtx_unlock(&nmp->nm_mtx); 4623 return (EOPNOTSUPP); 4624 } 4625 mtx_unlock(&nmp->nm_mtx); 4626 4627 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4628 attrflag = 0; 4629 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred, 4630 ap->a_td); 4631 if (attrflag != 0) { 4632 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4633 if (error == 0 && ret != 0) 4634 error = ret; 4635 } 4636 4637 switch (error) { 4638 case NFSERR_NOTSUPP: 4639 case NFSERR_OPILLEGAL: 4640 mtx_lock(&nmp->nm_mtx); 4641 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4642 mtx_unlock(&nmp->nm_mtx); 4643 error = EOPNOTSUPP; 4644 break; 4645 case NFSERR_NOXATTR: 4646 case NFSERR_XATTR2BIG: 4647 error = ENOATTR; 4648 break; 4649 default: 4650 error = nfscl_maperr(ap->a_td, error, 0, 0); 4651 break; 4652 } 4653 return (error); 4654 } 4655 4656 /* 4657 * Return POSIX pathconf information applicable to nfs filesystems. 4658 */ 4659 static int 4660 nfs_pathconf(struct vop_pathconf_args *ap) 4661 { 4662 struct nfsv3_pathconf pc; 4663 struct nfsvattr nfsva; 4664 struct vnode *vp = ap->a_vp; 4665 struct nfsmount *nmp; 4666 struct thread *td = curthread; 4667 off_t off; 4668 uint32_t clone_blksize; 4669 bool eof, has_namedattr, named_enabled; 4670 int attrflag, error; 4671 struct nfsnode *np; 4672 4673 nmp = VFSTONFS(vp->v_mount); 4674 np = VTONFS(vp); 4675 named_enabled = false; 4676 has_namedattr = false; 4677 clone_blksize = 0; 4678 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 4679 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 4680 ap->a_name == _PC_NO_TRUNC)) || 4681 (NFS_ISV4(vp) && (ap->a_name == _PC_ACL_NFS4 || 4682 ap->a_name == _PC_HAS_NAMEDATTR || 4683 ap->a_name == _PC_CLONE_BLKSIZE))) { 4684 /* 4685 * Since only the above 4 a_names are returned by the NFSv3 4686 * Pathconf RPC, there is no point in doing it for others. 4687 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 4688 * be used for _PC_ACL_NFS4, _PC_HAS_NAMEDATTR and 4689 * _PC_CLONE_BLKSIZE as well. 4690 */ 4691 error = nfsrpc_pathconf(vp, &pc, &has_namedattr, &clone_blksize, 4692 td->td_ucred, td, &nfsva, &attrflag); 4693 if (attrflag != 0) 4694 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4695 if (error != 0) 4696 return (error); 4697 } else if (NFS_ISV4(vp) && ap->a_name == _PC_NAMEDATTR_ENABLED && 4698 (np->n_flag & NNAMEDNOTSUPP) == 0) { 4699 struct nfsfh *nfhp; 4700 4701 error = nfsrpc_openattr(nmp, vp, np->n_fhp->nfh_fh, 4702 np->n_fhp->nfh_len, false, td->td_ucred, td, &nfsva, &nfhp, 4703 &attrflag); 4704 named_enabled = true; 4705 if (error == 0) { 4706 free(nfhp, M_NFSFH); 4707 } else if (error == NFSERR_NOTSUPP) { 4708 named_enabled = false; 4709 NFSLOCKNODE(np); 4710 np->n_flag |= NNAMEDNOTSUPP; 4711 NFSUNLOCKNODE(np); 4712 } 4713 error = 0; 4714 } else { 4715 /* 4716 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 4717 * just fake them. 4718 */ 4719 pc.pc_linkmax = NFS_LINK_MAX; 4720 pc.pc_namemax = NFS_MAXNAMLEN; 4721 pc.pc_notrunc = 1; 4722 pc.pc_chownrestricted = 1; 4723 pc.pc_caseinsensitive = 0; 4724 pc.pc_casepreserving = 1; 4725 error = 0; 4726 } 4727 switch (ap->a_name) { 4728 case _PC_LINK_MAX: 4729 #ifdef _LP64 4730 *ap->a_retval = pc.pc_linkmax; 4731 #else 4732 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 4733 #endif 4734 break; 4735 case _PC_NAME_MAX: 4736 *ap->a_retval = pc.pc_namemax; 4737 break; 4738 case _PC_PIPE_BUF: 4739 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 4740 *ap->a_retval = PIPE_BUF; 4741 else 4742 error = EINVAL; 4743 break; 4744 case _PC_CHOWN_RESTRICTED: 4745 *ap->a_retval = pc.pc_chownrestricted; 4746 break; 4747 case _PC_NO_TRUNC: 4748 *ap->a_retval = pc.pc_notrunc; 4749 break; 4750 case _PC_ACL_NFS4: 4751 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 4752 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 4753 *ap->a_retval = 1; 4754 else 4755 *ap->a_retval = 0; 4756 break; 4757 case _PC_ACL_PATH_MAX: 4758 if (NFS_ISV4(vp)) 4759 *ap->a_retval = ACL_MAX_ENTRIES; 4760 else 4761 *ap->a_retval = 3; 4762 break; 4763 case _PC_PRIO_IO: 4764 *ap->a_retval = 0; 4765 break; 4766 case _PC_SYNC_IO: 4767 *ap->a_retval = 0; 4768 break; 4769 case _PC_ALLOC_SIZE_MIN: 4770 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 4771 break; 4772 case _PC_FILESIZEBITS: 4773 if (NFS_ISV34(vp)) 4774 *ap->a_retval = 64; 4775 else 4776 *ap->a_retval = 32; 4777 break; 4778 case _PC_REC_INCR_XFER_SIZE: 4779 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4780 break; 4781 case _PC_REC_MAX_XFER_SIZE: 4782 *ap->a_retval = -1; /* means ``unlimited'' */ 4783 break; 4784 case _PC_REC_MIN_XFER_SIZE: 4785 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4786 break; 4787 case _PC_REC_XFER_ALIGN: 4788 *ap->a_retval = PAGE_SIZE; 4789 break; 4790 case _PC_SYMLINK_MAX: 4791 *ap->a_retval = NFS_MAXPATHLEN; 4792 break; 4793 case _PC_MIN_HOLE_SIZE: 4794 /* Only some NFSv4.2 servers support Seek for Holes. */ 4795 *ap->a_retval = 0; 4796 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) { 4797 /* 4798 * NFSv4.2 doesn't have an attribute for hole size, 4799 * so all we can do is see if the Seek operation is 4800 * supported and then use f_iosize as a "best guess". 4801 */ 4802 mtx_lock(&nmp->nm_mtx); 4803 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) { 4804 mtx_unlock(&nmp->nm_mtx); 4805 off = 0; 4806 attrflag = 0; 4807 error = nfsrpc_seek(vp, &off, &eof, 4808 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva, 4809 &attrflag); 4810 if (attrflag != 0) 4811 (void) nfscl_loadattrcache(&vp, &nfsva, 4812 NULL, 0, 1); 4813 mtx_lock(&nmp->nm_mtx); 4814 if (error == NFSERR_NOTSUPP) 4815 nmp->nm_privflag |= NFSMNTP_SEEKTESTED; 4816 else 4817 nmp->nm_privflag |= NFSMNTP_SEEKTESTED | 4818 NFSMNTP_SEEK; 4819 error = 0; 4820 } 4821 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0) 4822 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4823 mtx_unlock(&nmp->nm_mtx); 4824 } 4825 break; 4826 case _PC_NAMEDATTR_ENABLED: 4827 if (named_enabled) 4828 *ap->a_retval = 1; 4829 else 4830 *ap->a_retval = 0; 4831 break; 4832 case _PC_HAS_NAMEDATTR: 4833 if (has_namedattr) 4834 *ap->a_retval = 1; 4835 else 4836 *ap->a_retval = 0; 4837 break; 4838 case _PC_HAS_HIDDENSYSTEM: 4839 if (NFS_ISV4(vp) && NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, 4840 NFSATTRBIT_ARCHIVE) && 4841 NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, 4842 NFSATTRBIT_HIDDEN) && 4843 NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, 4844 NFSATTRBIT_SYSTEM)) 4845 *ap->a_retval = 1; 4846 else 4847 *ap->a_retval = 0; 4848 break; 4849 case _PC_CLONE_BLKSIZE: 4850 *ap->a_retval = clone_blksize; 4851 break; 4852 4853 default: 4854 error = vop_stdpathconf(ap); 4855 break; 4856 } 4857 return (error); 4858 } 4859