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 __FBSDID("$FreeBSD$"); 39 40 /* 41 * vnode op calls for Sun NFS version 2, 3 and 4 42 */ 43 44 #include "opt_inet.h" 45 46 #include <sys/param.h> 47 #include <sys/kernel.h> 48 #include <sys/systm.h> 49 #include <sys/resourcevar.h> 50 #include <sys/proc.h> 51 #include <sys/mount.h> 52 #include <sys/bio.h> 53 #include <sys/buf.h> 54 #include <sys/extattr.h> 55 #include <sys/filio.h> 56 #include <sys/jail.h> 57 #include <sys/malloc.h> 58 #include <sys/mbuf.h> 59 #include <sys/namei.h> 60 #include <sys/socket.h> 61 #include <sys/vnode.h> 62 #include <sys/dirent.h> 63 #include <sys/fcntl.h> 64 #include <sys/lockf.h> 65 #include <sys/stat.h> 66 #include <sys/sysctl.h> 67 #include <sys/signalvar.h> 68 69 #include <vm/vm.h> 70 #include <vm/vm_extern.h> 71 #include <vm/vm_object.h> 72 73 #include <fs/nfs/nfsport.h> 74 #include <fs/nfsclient/nfsnode.h> 75 #include <fs/nfsclient/nfsmount.h> 76 #include <fs/nfsclient/nfs.h> 77 #include <fs/nfsclient/nfs_kdtrace.h> 78 79 #include <net/if.h> 80 #include <netinet/in.h> 81 #include <netinet/in_var.h> 82 83 #include <nfs/nfs_lock.h> 84 85 #ifdef KDTRACE_HOOKS 86 #include <sys/dtrace_bsd.h> 87 88 dtrace_nfsclient_accesscache_flush_probe_func_t 89 dtrace_nfscl_accesscache_flush_done_probe; 90 uint32_t nfscl_accesscache_flush_done_id; 91 92 dtrace_nfsclient_accesscache_get_probe_func_t 93 dtrace_nfscl_accesscache_get_hit_probe, 94 dtrace_nfscl_accesscache_get_miss_probe; 95 uint32_t nfscl_accesscache_get_hit_id; 96 uint32_t nfscl_accesscache_get_miss_id; 97 98 dtrace_nfsclient_accesscache_load_probe_func_t 99 dtrace_nfscl_accesscache_load_done_probe; 100 uint32_t nfscl_accesscache_load_done_id; 101 #endif /* !KDTRACE_HOOKS */ 102 103 /* Defs */ 104 #define TRUE 1 105 #define FALSE 0 106 107 extern struct nfsstatsv1 nfsstatsv1; 108 extern int nfsrv_useacl; 109 extern int nfscl_debuglevel; 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 vop_lookup_t nfs_lookup; 118 static vop_create_t nfs_create; 119 static vop_mknod_t nfs_mknod; 120 static vop_open_t nfs_open; 121 static vop_pathconf_t nfs_pathconf; 122 static vop_close_t nfs_close; 123 static vop_access_t nfs_access; 124 static vop_getattr_t nfs_getattr; 125 static vop_setattr_t nfs_setattr; 126 static vop_read_t nfs_read; 127 static vop_fsync_t nfs_fsync; 128 static vop_remove_t nfs_remove; 129 static vop_link_t nfs_link; 130 static vop_rename_t nfs_rename; 131 static vop_mkdir_t nfs_mkdir; 132 static vop_rmdir_t nfs_rmdir; 133 static vop_symlink_t nfs_symlink; 134 static vop_readdir_t nfs_readdir; 135 static vop_strategy_t nfs_strategy; 136 static int nfs_lookitup(struct vnode *, char *, int, 137 struct ucred *, struct thread *, struct nfsnode **); 138 static int nfs_sillyrename(struct vnode *, struct vnode *, 139 struct componentname *); 140 static vop_access_t nfsspec_access; 141 static vop_readlink_t nfs_readlink; 142 static vop_print_t nfs_print; 143 static vop_advlock_t nfs_advlock; 144 static vop_advlockasync_t nfs_advlockasync; 145 static vop_getacl_t nfs_getacl; 146 static vop_setacl_t nfs_setacl; 147 static vop_advise_t nfs_advise; 148 static vop_allocate_t nfs_allocate; 149 static vop_deallocate_t nfs_deallocate; 150 static vop_copy_file_range_t nfs_copy_file_range; 151 static vop_ioctl_t nfs_ioctl; 152 static vop_getextattr_t nfs_getextattr; 153 static vop_setextattr_t nfs_setextattr; 154 static vop_listextattr_t nfs_listextattr; 155 static vop_deleteextattr_t nfs_deleteextattr; 156 static vop_lock1_t nfs_lock; 157 158 /* 159 * Global vfs data structures for nfs 160 */ 161 162 static struct vop_vector newnfs_vnodeops_nosig = { 163 .vop_default = &default_vnodeops, 164 .vop_access = nfs_access, 165 .vop_advlock = nfs_advlock, 166 .vop_advlockasync = nfs_advlockasync, 167 .vop_close = nfs_close, 168 .vop_create = nfs_create, 169 .vop_fsync = nfs_fsync, 170 .vop_getattr = nfs_getattr, 171 .vop_getpages = ncl_getpages, 172 .vop_putpages = ncl_putpages, 173 .vop_inactive = ncl_inactive, 174 .vop_link = nfs_link, 175 .vop_lock1 = nfs_lock, 176 .vop_lookup = nfs_lookup, 177 .vop_mkdir = nfs_mkdir, 178 .vop_mknod = nfs_mknod, 179 .vop_open = nfs_open, 180 .vop_pathconf = nfs_pathconf, 181 .vop_print = nfs_print, 182 .vop_read = nfs_read, 183 .vop_readdir = nfs_readdir, 184 .vop_readlink = nfs_readlink, 185 .vop_reclaim = ncl_reclaim, 186 .vop_remove = nfs_remove, 187 .vop_rename = nfs_rename, 188 .vop_rmdir = nfs_rmdir, 189 .vop_setattr = nfs_setattr, 190 .vop_strategy = nfs_strategy, 191 .vop_symlink = nfs_symlink, 192 .vop_write = ncl_write, 193 .vop_getacl = nfs_getacl, 194 .vop_setacl = nfs_setacl, 195 .vop_advise = nfs_advise, 196 .vop_allocate = nfs_allocate, 197 .vop_deallocate = nfs_deallocate, 198 .vop_copy_file_range = nfs_copy_file_range, 199 .vop_ioctl = nfs_ioctl, 200 .vop_getextattr = nfs_getextattr, 201 .vop_setextattr = nfs_setextattr, 202 .vop_listextattr = nfs_listextattr, 203 .vop_deleteextattr = nfs_deleteextattr, 204 }; 205 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig); 206 207 static int 208 nfs_vnodeops_bypass(struct vop_generic_args *a) 209 { 210 211 return (vop_sigdefer(&newnfs_vnodeops_nosig, a)); 212 } 213 214 struct vop_vector newnfs_vnodeops = { 215 .vop_default = &default_vnodeops, 216 .vop_bypass = nfs_vnodeops_bypass, 217 }; 218 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops); 219 220 static struct vop_vector newnfs_fifoops_nosig = { 221 .vop_default = &fifo_specops, 222 .vop_access = nfsspec_access, 223 .vop_close = nfsfifo_close, 224 .vop_fsync = nfs_fsync, 225 .vop_getattr = nfs_getattr, 226 .vop_inactive = ncl_inactive, 227 .vop_pathconf = nfs_pathconf, 228 .vop_print = nfs_print, 229 .vop_read = nfsfifo_read, 230 .vop_reclaim = ncl_reclaim, 231 .vop_setattr = nfs_setattr, 232 .vop_write = nfsfifo_write, 233 }; 234 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig); 235 236 static int 237 nfs_fifoops_bypass(struct vop_generic_args *a) 238 { 239 240 return (vop_sigdefer(&newnfs_fifoops_nosig, a)); 241 } 242 243 struct vop_vector newnfs_fifoops = { 244 .vop_default = &default_vnodeops, 245 .vop_bypass = nfs_fifoops_bypass, 246 }; 247 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops); 248 249 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 250 struct componentname *cnp, struct vattr *vap); 251 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 252 int namelen, struct ucred *cred, struct thread *td); 253 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 254 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 255 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 256 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 257 struct componentname *scnp, struct sillyrename *sp); 258 259 /* 260 * Global variables 261 */ 262 SYSCTL_DECL(_vfs_nfs); 263 264 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 265 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 266 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 267 268 static int nfs_prime_access_cache = 0; 269 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 270 &nfs_prime_access_cache, 0, 271 "Prime NFS ACCESS cache when fetching attributes"); 272 273 static int newnfs_commit_on_close = 0; 274 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 275 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 276 277 static int nfs_clean_pages_on_close = 1; 278 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 279 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 280 281 int newnfs_directio_enable = 0; 282 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 283 &newnfs_directio_enable, 0, "Enable NFS directio"); 284 285 int nfs_keep_dirty_on_error; 286 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 287 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 288 289 /* 290 * This sysctl allows other processes to mmap a file that has been opened 291 * O_DIRECT by a process. In general, having processes mmap the file while 292 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 293 * this by default to prevent DoS attacks - to prevent a malicious user from 294 * opening up files O_DIRECT preventing other users from mmap'ing these 295 * files. "Protected" environments where stricter consistency guarantees are 296 * required can disable this knob. The process that opened the file O_DIRECT 297 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 298 * meaningful. 299 */ 300 int newnfs_directio_allow_mmap = 1; 301 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 302 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 303 304 static uint64_t nfs_maxalloclen = 64 * 1024 * 1024; 305 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW, 306 &nfs_maxalloclen, 0, "NFS max allocate/deallocate length"); 307 308 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 309 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 310 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 311 312 /* 313 * SMP Locking Note : 314 * The list of locks after the description of the lock is the ordering 315 * of other locks acquired with the lock held. 316 * np->n_mtx : Protects the fields in the nfsnode. 317 VM Object Lock 318 VI_MTX (acquired indirectly) 319 * nmp->nm_mtx : Protects the fields in the nfsmount. 320 rep->r_mtx 321 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 322 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 323 nmp->nm_mtx 324 rep->r_mtx 325 * rep->r_mtx : Protects the fields in an nfsreq. 326 */ 327 328 static int 329 nfs_lock(struct vop_lock1_args *ap) 330 { 331 struct vnode *vp; 332 struct nfsnode *np; 333 u_quad_t nsize; 334 int error, lktype; 335 bool onfault; 336 337 vp = ap->a_vp; 338 lktype = ap->a_flags & LK_TYPE_MASK; 339 error = VOP_LOCK1_APV(&default_vnodeops, ap); 340 if (error != 0 || vp->v_op != &newnfs_vnodeops) 341 return (error); 342 np = VTONFS(vp); 343 if (np == NULL) 344 return (0); 345 NFSLOCKNODE(np); 346 if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED && 347 lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE && 348 lktype != LK_TRYUPGRADE)) { 349 NFSUNLOCKNODE(np); 350 return (0); 351 } 352 onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT && 353 (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE && 354 (lktype == LK_SHARED || lktype == LK_EXCLUSIVE); 355 if (onfault && vp->v_vnlock->lk_recurse == 0) { 356 /* 357 * Force retry in vm_fault(), to make the lock request 358 * sleepable, which allows us to piggy-back the 359 * sleepable call to vnode_pager_setsize(). 360 */ 361 NFSUNLOCKNODE(np); 362 VOP_UNLOCK(vp); 363 return (EBUSY); 364 } 365 if ((ap->a_flags & LK_NOWAIT) != 0 || 366 (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) { 367 NFSUNLOCKNODE(np); 368 return (0); 369 } 370 if (lktype == LK_SHARED) { 371 NFSUNLOCKNODE(np); 372 VOP_UNLOCK(vp); 373 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 374 ap->a_flags |= LK_EXCLUSIVE; 375 error = VOP_LOCK1_APV(&default_vnodeops, ap); 376 if (error != 0 || vp->v_op != &newnfs_vnodeops) 377 return (error); 378 if (vp->v_data == NULL) 379 goto downgrade; 380 MPASS(vp->v_data == np); 381 NFSLOCKNODE(np); 382 if ((np->n_flag & NVNSETSZSKIP) == 0) { 383 NFSUNLOCKNODE(np); 384 goto downgrade; 385 } 386 } 387 np->n_flag &= ~NVNSETSZSKIP; 388 nsize = np->n_size; 389 NFSUNLOCKNODE(np); 390 vnode_pager_setsize(vp, nsize); 391 downgrade: 392 if (lktype == LK_SHARED) { 393 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 394 ap->a_flags |= LK_DOWNGRADE; 395 (void)VOP_LOCK1_APV(&default_vnodeops, ap); 396 } 397 return (0); 398 } 399 400 static int 401 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 402 struct ucred *cred, u_int32_t *retmode) 403 { 404 int error = 0, attrflag, i, lrupos; 405 u_int32_t rmode; 406 struct nfsnode *np = VTONFS(vp); 407 struct nfsvattr nfsva; 408 409 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 410 &rmode, NULL); 411 if (attrflag) 412 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 413 if (!error) { 414 lrupos = 0; 415 NFSLOCKNODE(np); 416 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 417 if (np->n_accesscache[i].uid == cred->cr_uid) { 418 np->n_accesscache[i].mode = rmode; 419 np->n_accesscache[i].stamp = time_second; 420 break; 421 } 422 if (i > 0 && np->n_accesscache[i].stamp < 423 np->n_accesscache[lrupos].stamp) 424 lrupos = i; 425 } 426 if (i == NFS_ACCESSCACHESIZE) { 427 np->n_accesscache[lrupos].uid = cred->cr_uid; 428 np->n_accesscache[lrupos].mode = rmode; 429 np->n_accesscache[lrupos].stamp = time_second; 430 } 431 NFSUNLOCKNODE(np); 432 if (retmode != NULL) 433 *retmode = rmode; 434 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 435 } else if (NFS_ISV4(vp)) { 436 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 437 } 438 #ifdef KDTRACE_HOOKS 439 if (error != 0) 440 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 441 error); 442 #endif 443 return (error); 444 } 445 446 /* 447 * nfs access vnode op. 448 * For nfs version 2, just return ok. File accesses may fail later. 449 * For nfs version 3, use the access rpc to check accessibility. If file modes 450 * are changed on the server, accesses might still fail later. 451 */ 452 static int 453 nfs_access(struct vop_access_args *ap) 454 { 455 struct vnode *vp = ap->a_vp; 456 int error = 0, i, gotahit; 457 u_int32_t mode, wmode, rmode; 458 int v34 = NFS_ISV34(vp); 459 struct nfsnode *np = VTONFS(vp); 460 461 /* 462 * Disallow write attempts on filesystems mounted read-only; 463 * unless the file is a socket, fifo, or a block or character 464 * device resident on the filesystem. 465 */ 466 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 467 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 468 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 469 switch (vp->v_type) { 470 case VREG: 471 case VDIR: 472 case VLNK: 473 return (EROFS); 474 default: 475 break; 476 } 477 } 478 /* 479 * For nfs v3 or v4, check to see if we have done this recently, and if 480 * so return our cached result instead of making an ACCESS call. 481 * If not, do an access rpc, otherwise you are stuck emulating 482 * ufs_access() locally using the vattr. This may not be correct, 483 * since the server may apply other access criteria such as 484 * client uid-->server uid mapping that we do not know about. 485 */ 486 if (v34) { 487 if (ap->a_accmode & VREAD) 488 mode = NFSACCESS_READ; 489 else 490 mode = 0; 491 if (vp->v_type != VDIR) { 492 if (ap->a_accmode & VWRITE) 493 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 494 if (ap->a_accmode & VAPPEND) 495 mode |= NFSACCESS_EXTEND; 496 if (ap->a_accmode & VEXEC) 497 mode |= NFSACCESS_EXECUTE; 498 if (ap->a_accmode & VDELETE) 499 mode |= NFSACCESS_DELETE; 500 } else { 501 if (ap->a_accmode & VWRITE) 502 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 503 if (ap->a_accmode & VAPPEND) 504 mode |= NFSACCESS_EXTEND; 505 if (ap->a_accmode & VEXEC) 506 mode |= NFSACCESS_LOOKUP; 507 if (ap->a_accmode & VDELETE) 508 mode |= NFSACCESS_DELETE; 509 if (ap->a_accmode & VDELETE_CHILD) 510 mode |= NFSACCESS_MODIFY; 511 } 512 /* XXX safety belt, only make blanket request if caching */ 513 if (nfsaccess_cache_timeout > 0) { 514 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 515 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 516 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 517 } else { 518 wmode = mode; 519 } 520 521 /* 522 * Does our cached result allow us to give a definite yes to 523 * this request? 524 */ 525 gotahit = 0; 526 NFSLOCKNODE(np); 527 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 528 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 529 if (time_second < (np->n_accesscache[i].stamp 530 + nfsaccess_cache_timeout) && 531 (np->n_accesscache[i].mode & mode) == mode) { 532 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 533 gotahit = 1; 534 } 535 break; 536 } 537 } 538 NFSUNLOCKNODE(np); 539 #ifdef KDTRACE_HOOKS 540 if (gotahit != 0) 541 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 542 ap->a_cred->cr_uid, mode); 543 else 544 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 545 ap->a_cred->cr_uid, mode); 546 #endif 547 if (gotahit == 0) { 548 /* 549 * Either a no, or a don't know. Go to the wire. 550 */ 551 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 552 error = nfs34_access_otw(vp, wmode, ap->a_td, 553 ap->a_cred, &rmode); 554 if (!error && 555 (rmode & mode) != mode) 556 error = EACCES; 557 } 558 return (error); 559 } else { 560 if ((error = nfsspec_access(ap)) != 0) { 561 return (error); 562 } 563 /* 564 * Attempt to prevent a mapped root from accessing a file 565 * which it shouldn't. We try to read a byte from the file 566 * if the user is root and the file is not zero length. 567 * After calling nfsspec_access, we should have the correct 568 * file size cached. 569 */ 570 NFSLOCKNODE(np); 571 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 572 && VTONFS(vp)->n_size > 0) { 573 struct iovec aiov; 574 struct uio auio; 575 char buf[1]; 576 577 NFSUNLOCKNODE(np); 578 aiov.iov_base = buf; 579 aiov.iov_len = 1; 580 auio.uio_iov = &aiov; 581 auio.uio_iovcnt = 1; 582 auio.uio_offset = 0; 583 auio.uio_resid = 1; 584 auio.uio_segflg = UIO_SYSSPACE; 585 auio.uio_rw = UIO_READ; 586 auio.uio_td = ap->a_td; 587 588 if (vp->v_type == VREG) 589 error = ncl_readrpc(vp, &auio, ap->a_cred); 590 else if (vp->v_type == VDIR) { 591 char* bp; 592 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 593 aiov.iov_base = bp; 594 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 595 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 596 ap->a_td); 597 free(bp, M_TEMP); 598 } else if (vp->v_type == VLNK) 599 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 600 else 601 error = EACCES; 602 } else 603 NFSUNLOCKNODE(np); 604 return (error); 605 } 606 } 607 608 /* 609 * nfs open vnode op 610 * Check to see if the type is ok 611 * and that deletion is not in progress. 612 * For paged in text files, you will need to flush the page cache 613 * if consistency is lost. 614 */ 615 /* ARGSUSED */ 616 static int 617 nfs_open(struct vop_open_args *ap) 618 { 619 struct vnode *vp = ap->a_vp; 620 struct nfsnode *np = VTONFS(vp); 621 struct vattr vattr; 622 int error; 623 int fmode = ap->a_mode; 624 struct ucred *cred; 625 vm_object_t obj; 626 627 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 628 return (EOPNOTSUPP); 629 630 /* 631 * For NFSv4, we need to do the Open Op before cache validation, 632 * so that we conform to RFC3530 Sec. 9.3.1. 633 */ 634 if (NFS_ISV4(vp)) { 635 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 636 if (error) { 637 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 638 (gid_t)0); 639 return (error); 640 } 641 } 642 643 /* 644 * Now, if this Open will be doing reading, re-validate/flush the 645 * cache, so that Close/Open coherency is maintained. 646 */ 647 NFSLOCKNODE(np); 648 if (np->n_flag & NMODIFIED) { 649 NFSUNLOCKNODE(np); 650 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 651 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 652 if (VN_IS_DOOMED(vp)) 653 return (EBADF); 654 } 655 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 656 if (error == EINTR || error == EIO) { 657 if (NFS_ISV4(vp)) 658 (void) nfsrpc_close(vp, 0, ap->a_td); 659 return (error); 660 } 661 NFSLOCKNODE(np); 662 np->n_attrstamp = 0; 663 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 664 if (vp->v_type == VDIR) 665 np->n_direofoffset = 0; 666 NFSUNLOCKNODE(np); 667 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 668 if (error) { 669 if (NFS_ISV4(vp)) 670 (void) nfsrpc_close(vp, 0, ap->a_td); 671 return (error); 672 } 673 NFSLOCKNODE(np); 674 np->n_mtime = vattr.va_mtime; 675 if (NFS_ISV4(vp)) 676 np->n_change = vattr.va_filerev; 677 } else { 678 NFSUNLOCKNODE(np); 679 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 680 if (error) { 681 if (NFS_ISV4(vp)) 682 (void) nfsrpc_close(vp, 0, ap->a_td); 683 return (error); 684 } 685 NFSLOCKNODE(np); 686 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 687 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 688 if (vp->v_type == VDIR) 689 np->n_direofoffset = 0; 690 NFSUNLOCKNODE(np); 691 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 692 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 693 if (VN_IS_DOOMED(vp)) 694 return (EBADF); 695 } 696 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 697 if (error == EINTR || error == EIO) { 698 if (NFS_ISV4(vp)) 699 (void) nfsrpc_close(vp, 0, ap->a_td); 700 return (error); 701 } 702 NFSLOCKNODE(np); 703 np->n_mtime = vattr.va_mtime; 704 if (NFS_ISV4(vp)) 705 np->n_change = vattr.va_filerev; 706 } 707 } 708 709 /* 710 * If the object has >= 1 O_DIRECT active opens, we disable caching. 711 */ 712 if (newnfs_directio_enable && (fmode & O_DIRECT) && 713 (vp->v_type == VREG)) { 714 if (np->n_directio_opens == 0) { 715 NFSUNLOCKNODE(np); 716 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 717 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 718 if (VN_IS_DOOMED(vp)) 719 return (EBADF); 720 } 721 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 722 if (error) { 723 if (NFS_ISV4(vp)) 724 (void) nfsrpc_close(vp, 0, ap->a_td); 725 return (error); 726 } 727 NFSLOCKNODE(np); 728 np->n_flag |= NNONCACHE; 729 } 730 np->n_directio_opens++; 731 } 732 733 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 734 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 735 np->n_flag |= NWRITEOPENED; 736 737 /* 738 * If this is an open for writing, capture a reference to the 739 * credentials, so they can be used by ncl_putpages(). Using 740 * these write credentials is preferable to the credentials of 741 * whatever thread happens to be doing the VOP_PUTPAGES() since 742 * the write RPCs are less likely to fail with EACCES. 743 */ 744 if ((fmode & FWRITE) != 0) { 745 cred = np->n_writecred; 746 np->n_writecred = crhold(ap->a_cred); 747 } else 748 cred = NULL; 749 NFSUNLOCKNODE(np); 750 751 if (cred != NULL) 752 crfree(cred); 753 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 754 755 /* 756 * If the text file has been mmap'd, flush any dirty pages to the 757 * buffer cache and then... 758 * Make sure all writes are pushed to the NFS server. If this is not 759 * done, the modify time of the file can change while the text 760 * file is being executed. This will cause the process that is 761 * executing the text file to be terminated. 762 */ 763 if (vp->v_writecount <= -1) { 764 if ((obj = vp->v_object) != NULL && 765 vm_object_mightbedirty(obj)) { 766 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 767 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 768 if (VN_IS_DOOMED(vp)) 769 return (EBADF); 770 } 771 VM_OBJECT_WLOCK(obj); 772 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); 773 VM_OBJECT_WUNLOCK(obj); 774 } 775 776 /* Now, flush the buffer cache. */ 777 ncl_flush(vp, MNT_WAIT, curthread, 0, 0); 778 779 /* And, finally, make sure that n_mtime is up to date. */ 780 np = VTONFS(vp); 781 NFSLOCKNODE(np); 782 np->n_mtime = np->n_vattr.na_mtime; 783 NFSUNLOCKNODE(np); 784 } 785 return (0); 786 } 787 788 /* 789 * nfs close vnode op 790 * What an NFS client should do upon close after writing is a debatable issue. 791 * Most NFS clients push delayed writes to the server upon close, basically for 792 * two reasons: 793 * 1 - So that any write errors may be reported back to the client process 794 * doing the close system call. By far the two most likely errors are 795 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 796 * 2 - To put a worst case upper bound on cache inconsistency between 797 * multiple clients for the file. 798 * There is also a consistency problem for Version 2 of the protocol w.r.t. 799 * not being able to tell if other clients are writing a file concurrently, 800 * since there is no way of knowing if the changed modify time in the reply 801 * is only due to the write for this client. 802 * (NFS Version 3 provides weak cache consistency data in the reply that 803 * should be sufficient to detect and handle this case.) 804 * 805 * The current code does the following: 806 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 807 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 808 * or commit them (this satisfies 1 and 2 except for the 809 * case where the server crashes after this close but 810 * before the commit RPC, which is felt to be "good 811 * enough". Changing the last argument to ncl_flush() to 812 * a 1 would force a commit operation, if it is felt a 813 * commit is necessary now. 814 * for NFS Version 4 - flush the dirty buffers and commit them, if 815 * nfscl_mustflush() says this is necessary. 816 * It is necessary if there is no write delegation held, 817 * in order to satisfy open/close coherency. 818 * If the file isn't cached on local stable storage, 819 * it may be necessary in order to detect "out of space" 820 * errors from the server, if the write delegation 821 * issued by the server doesn't allow the file to grow. 822 */ 823 /* ARGSUSED */ 824 static int 825 nfs_close(struct vop_close_args *ap) 826 { 827 struct vnode *vp = ap->a_vp; 828 struct nfsnode *np = VTONFS(vp); 829 struct nfsvattr nfsva; 830 struct ucred *cred; 831 int error = 0, ret, localcred = 0; 832 int fmode = ap->a_fflag; 833 834 if (NFSCL_FORCEDISM(vp->v_mount)) 835 return (0); 836 /* 837 * During shutdown, a_cred isn't valid, so just use root. 838 */ 839 if (ap->a_cred == NOCRED) { 840 cred = newnfs_getcred(); 841 localcred = 1; 842 } else { 843 cred = ap->a_cred; 844 } 845 if (vp->v_type == VREG) { 846 /* 847 * Examine and clean dirty pages, regardless of NMODIFIED. 848 * This closes a major hole in close-to-open consistency. 849 * We want to push out all dirty pages (and buffers) on 850 * close, regardless of whether they were dirtied by 851 * mmap'ed writes or via write(). 852 */ 853 if (nfs_clean_pages_on_close && vp->v_object) { 854 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 855 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 856 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 857 return (EBADF); 858 } 859 VM_OBJECT_WLOCK(vp->v_object); 860 vm_object_page_clean(vp->v_object, 0, 0, 0); 861 VM_OBJECT_WUNLOCK(vp->v_object); 862 } 863 NFSLOCKNODE(np); 864 if (np->n_flag & NMODIFIED) { 865 NFSUNLOCKNODE(np); 866 if (NFS_ISV3(vp)) { 867 /* 868 * Under NFSv3 we have dirty buffers to dispose of. We 869 * must flush them to the NFS server. We have the option 870 * of waiting all the way through the commit rpc or just 871 * waiting for the initial write. The default is to only 872 * wait through the initial write so the data is in the 873 * server's cache, which is roughly similar to the state 874 * a standard disk subsystem leaves the file in on close(). 875 * 876 * We cannot clear the NMODIFIED bit in np->n_flag due to 877 * potential races with other processes, and certainly 878 * cannot clear it if we don't commit. 879 * These races occur when there is no longer the old 880 * traditional vnode locking implemented for Vnode Ops. 881 */ 882 int cm = newnfs_commit_on_close ? 1 : 0; 883 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 884 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 885 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 886 return (EBADF); 887 } 888 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); 889 /* np->n_flag &= ~NMODIFIED; */ 890 } else if (NFS_ISV4(vp)) { 891 if (nfscl_mustflush(vp) != 0) { 892 int cm = newnfs_commit_on_close ? 1 : 0; 893 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 894 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 895 if (VN_IS_DOOMED(vp) && ap->a_fflag != 896 FNONBLOCK) 897 return (EBADF); 898 } 899 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 900 cm, 0); 901 /* 902 * as above w.r.t races when clearing 903 * NMODIFIED. 904 * np->n_flag &= ~NMODIFIED; 905 */ 906 } 907 } else { 908 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 909 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 910 if (VN_IS_DOOMED(vp) && ap->a_fflag != 911 FNONBLOCK) 912 return (EBADF); 913 } 914 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 915 } 916 NFSLOCKNODE(np); 917 } 918 /* 919 * Invalidate the attribute cache in all cases. 920 * An open is going to fetch fresh attrs any way, other procs 921 * on this node that have file open will be forced to do an 922 * otw attr fetch, but this is safe. 923 * --> A user found that their RPC count dropped by 20% when 924 * this was commented out and I can't see any requirement 925 * for it, so I've disabled it when negative lookups are 926 * enabled. (What does this have to do with negative lookup 927 * caching? Well nothing, except it was reported by the 928 * same user that needed negative lookup caching and I wanted 929 * there to be a way to disable it to see if it 930 * is the cause of some caching/coherency issue that might 931 * crop up.) 932 */ 933 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 934 np->n_attrstamp = 0; 935 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 936 } 937 if (np->n_flag & NWRITEERR) { 938 np->n_flag &= ~NWRITEERR; 939 error = np->n_error; 940 } 941 NFSUNLOCKNODE(np); 942 } 943 944 if (NFS_ISV4(vp)) { 945 /* 946 * Get attributes so "change" is up to date. 947 */ 948 if (error == 0 && nfscl_mustflush(vp) != 0 && 949 vp->v_type == VREG && 950 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 951 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 952 NULL); 953 if (!ret) { 954 np->n_change = nfsva.na_filerev; 955 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 956 0, 0); 957 } 958 } 959 960 /* 961 * and do the close. 962 */ 963 ret = nfsrpc_close(vp, 0, ap->a_td); 964 if (!error && ret) 965 error = ret; 966 if (error) 967 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 968 (gid_t)0); 969 } 970 if (newnfs_directio_enable) 971 KASSERT((np->n_directio_asyncwr == 0), 972 ("nfs_close: dirty unflushed (%d) directio buffers\n", 973 np->n_directio_asyncwr)); 974 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 975 NFSLOCKNODE(np); 976 KASSERT((np->n_directio_opens > 0), 977 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 978 np->n_directio_opens--; 979 if (np->n_directio_opens == 0) 980 np->n_flag &= ~NNONCACHE; 981 NFSUNLOCKNODE(np); 982 } 983 if (localcred) 984 NFSFREECRED(cred); 985 return (error); 986 } 987 988 /* 989 * nfs getattr call from vfs. 990 */ 991 static int 992 nfs_getattr(struct vop_getattr_args *ap) 993 { 994 struct vnode *vp = ap->a_vp; 995 struct thread *td = curthread; /* XXX */ 996 struct nfsnode *np = VTONFS(vp); 997 int error = 0; 998 struct nfsvattr nfsva; 999 struct vattr *vap = ap->a_vap; 1000 struct vattr vattr; 1001 1002 /* 1003 * Update local times for special files. 1004 */ 1005 NFSLOCKNODE(np); 1006 if (np->n_flag & (NACC | NUPD)) 1007 np->n_flag |= NCHG; 1008 NFSUNLOCKNODE(np); 1009 /* 1010 * First look in the cache. 1011 */ 1012 if (ncl_getattrcache(vp, &vattr) == 0) { 1013 ncl_copy_vattr(vap, &vattr); 1014 1015 /* 1016 * Get the local modify time for the case of a write 1017 * delegation. 1018 */ 1019 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1020 return (0); 1021 } 1022 1023 if (NFS_ISV34(vp) && nfs_prime_access_cache && 1024 nfsaccess_cache_timeout > 0) { 1025 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 1026 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 1027 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 1028 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 1029 return (0); 1030 } 1031 } 1032 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 1033 if (!error) 1034 error = nfscl_loadattrcache(&vp, &nfsva, vap, 0, 0); 1035 if (!error) { 1036 /* 1037 * Get the local modify time for the case of a write 1038 * delegation. 1039 */ 1040 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1041 } else if (NFS_ISV4(vp)) { 1042 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1043 } 1044 return (error); 1045 } 1046 1047 /* 1048 * nfs setattr call. 1049 */ 1050 static int 1051 nfs_setattr(struct vop_setattr_args *ap) 1052 { 1053 struct vnode *vp = ap->a_vp; 1054 struct nfsnode *np = VTONFS(vp); 1055 struct thread *td = curthread; /* XXX */ 1056 struct vattr *vap = ap->a_vap; 1057 int error = 0; 1058 u_quad_t tsize; 1059 struct timespec ts; 1060 1061 #ifndef nolint 1062 tsize = (u_quad_t)0; 1063 #endif 1064 1065 /* 1066 * Setting of flags and marking of atimes are not supported. 1067 */ 1068 if (vap->va_flags != VNOVAL) 1069 return (EOPNOTSUPP); 1070 1071 /* 1072 * Disallow write attempts if the filesystem is mounted read-only. 1073 */ 1074 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 1075 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 1076 vap->va_mtime.tv_sec != VNOVAL || 1077 vap->va_birthtime.tv_sec != VNOVAL || 1078 vap->va_mode != (mode_t)VNOVAL) && 1079 (vp->v_mount->mnt_flag & MNT_RDONLY)) 1080 return (EROFS); 1081 if (vap->va_size != VNOVAL) { 1082 switch (vp->v_type) { 1083 case VDIR: 1084 return (EISDIR); 1085 case VCHR: 1086 case VBLK: 1087 case VSOCK: 1088 case VFIFO: 1089 if (vap->va_mtime.tv_sec == VNOVAL && 1090 vap->va_atime.tv_sec == VNOVAL && 1091 vap->va_birthtime.tv_sec == VNOVAL && 1092 vap->va_mode == (mode_t)VNOVAL && 1093 vap->va_uid == (uid_t)VNOVAL && 1094 vap->va_gid == (gid_t)VNOVAL) 1095 return (0); 1096 vap->va_size = VNOVAL; 1097 break; 1098 default: 1099 /* 1100 * Disallow write attempts if the filesystem is 1101 * mounted read-only. 1102 */ 1103 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1104 return (EROFS); 1105 /* 1106 * We run vnode_pager_setsize() early (why?), 1107 * we must set np->n_size now to avoid vinvalbuf 1108 * V_SAVE races that might setsize a lower 1109 * value. 1110 */ 1111 NFSLOCKNODE(np); 1112 tsize = np->n_size; 1113 NFSUNLOCKNODE(np); 1114 error = ncl_meta_setsize(vp, td, vap->va_size); 1115 NFSLOCKNODE(np); 1116 if (np->n_flag & NMODIFIED) { 1117 tsize = np->n_size; 1118 NFSUNLOCKNODE(np); 1119 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 1120 0 : V_SAVE, td, 1); 1121 if (error != 0) { 1122 vnode_pager_setsize(vp, tsize); 1123 return (error); 1124 } 1125 /* 1126 * Call nfscl_delegmodtime() to set the modify time 1127 * locally, as required. 1128 */ 1129 nfscl_delegmodtime(vp); 1130 } else 1131 NFSUNLOCKNODE(np); 1132 /* 1133 * np->n_size has already been set to vap->va_size 1134 * in ncl_meta_setsize(). We must set it again since 1135 * nfs_loadattrcache() could be called through 1136 * ncl_meta_setsize() and could modify np->n_size. 1137 */ 1138 NFSLOCKNODE(np); 1139 np->n_vattr.na_size = np->n_size = vap->va_size; 1140 NFSUNLOCKNODE(np); 1141 } 1142 } else { 1143 NFSLOCKNODE(np); 1144 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 1145 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 1146 NFSUNLOCKNODE(np); 1147 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 1148 if (error == EINTR || error == EIO) 1149 return (error); 1150 } else 1151 NFSUNLOCKNODE(np); 1152 } 1153 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 1154 if (vap->va_size != VNOVAL) { 1155 if (error == 0) { 1156 nanouptime(&ts); 1157 NFSLOCKNODE(np); 1158 np->n_localmodtime = ts; 1159 NFSUNLOCKNODE(np); 1160 } else { 1161 NFSLOCKNODE(np); 1162 np->n_size = np->n_vattr.na_size = tsize; 1163 vnode_pager_setsize(vp, tsize); 1164 NFSUNLOCKNODE(np); 1165 } 1166 } 1167 return (error); 1168 } 1169 1170 /* 1171 * Do an nfs setattr rpc. 1172 */ 1173 static int 1174 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1175 struct thread *td) 1176 { 1177 struct nfsnode *np = VTONFS(vp); 1178 int error, ret, attrflag, i; 1179 struct nfsvattr nfsva; 1180 1181 if (NFS_ISV34(vp)) { 1182 NFSLOCKNODE(np); 1183 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1184 np->n_accesscache[i].stamp = 0; 1185 np->n_flag |= NDELEGMOD; 1186 NFSUNLOCKNODE(np); 1187 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1188 } 1189 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1190 NULL); 1191 if (attrflag) { 1192 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1193 if (ret && !error) 1194 error = ret; 1195 } 1196 if (error && NFS_ISV4(vp)) 1197 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1198 return (error); 1199 } 1200 1201 /* 1202 * nfs lookup call, one step at a time... 1203 * First look in cache 1204 * If not found, unlock the directory nfsnode and do the rpc 1205 */ 1206 static int 1207 nfs_lookup(struct vop_lookup_args *ap) 1208 { 1209 struct componentname *cnp = ap->a_cnp; 1210 struct vnode *dvp = ap->a_dvp; 1211 struct vnode **vpp = ap->a_vpp; 1212 struct mount *mp = dvp->v_mount; 1213 int flags = cnp->cn_flags; 1214 struct vnode *newvp; 1215 struct nfsmount *nmp; 1216 struct nfsnode *np, *newnp; 1217 int error = 0, attrflag, dattrflag, ltype, ncticks; 1218 struct thread *td = curthread; 1219 struct nfsfh *nfhp; 1220 struct nfsvattr dnfsva, nfsva; 1221 struct vattr vattr; 1222 struct timespec nctime, ts; 1223 uint32_t openmode; 1224 1225 *vpp = NULLVP; 1226 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1227 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1228 return (EROFS); 1229 if (dvp->v_type != VDIR) 1230 return (ENOTDIR); 1231 nmp = VFSTONFS(mp); 1232 np = VTONFS(dvp); 1233 1234 /* For NFSv4, wait until any remove is done. */ 1235 NFSLOCKNODE(np); 1236 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1237 np->n_flag |= NREMOVEWANT; 1238 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1239 } 1240 NFSUNLOCKNODE(np); 1241 1242 error = vn_dir_check_exec(dvp, cnp); 1243 if (error != 0) 1244 return (error); 1245 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1246 if (error > 0 && error != ENOENT) 1247 return (error); 1248 if (error == -1) { 1249 /* 1250 * Lookups of "." are special and always return the 1251 * current directory. cache_lookup() already handles 1252 * associated locking bookkeeping, etc. 1253 */ 1254 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1255 /* XXX: Is this really correct? */ 1256 if (cnp->cn_nameiop != LOOKUP && 1257 (flags & ISLASTCN)) 1258 cnp->cn_flags |= SAVENAME; 1259 return (0); 1260 } 1261 1262 /* 1263 * We only accept a positive hit in the cache if the 1264 * change time of the file matches our cached copy. 1265 * Otherwise, we discard the cache entry and fallback 1266 * to doing a lookup RPC. We also only trust cache 1267 * entries for less than nm_nametimeo seconds. 1268 * 1269 * To better handle stale file handles and attributes, 1270 * clear the attribute cache of this node if it is a 1271 * leaf component, part of an open() call, and not 1272 * locally modified before fetching the attributes. 1273 * This should allow stale file handles to be detected 1274 * here where we can fall back to a LOOKUP RPC to 1275 * recover rather than having nfs_open() detect the 1276 * stale file handle and failing open(2) with ESTALE. 1277 */ 1278 newvp = *vpp; 1279 newnp = VTONFS(newvp); 1280 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1281 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1282 !(newnp->n_flag & NMODIFIED)) { 1283 NFSLOCKNODE(newnp); 1284 newnp->n_attrstamp = 0; 1285 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1286 NFSUNLOCKNODE(newnp); 1287 } 1288 if (nfscl_nodeleg(newvp, 0) == 0 || 1289 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1290 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1291 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1292 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1293 if (cnp->cn_nameiop != LOOKUP && 1294 (flags & ISLASTCN)) 1295 cnp->cn_flags |= SAVENAME; 1296 return (0); 1297 } 1298 cache_purge(newvp); 1299 if (dvp != newvp) 1300 vput(newvp); 1301 else 1302 vrele(newvp); 1303 *vpp = NULLVP; 1304 } else if (error == ENOENT) { 1305 if (VN_IS_DOOMED(dvp)) 1306 return (ENOENT); 1307 /* 1308 * We only accept a negative hit in the cache if the 1309 * modification time of the parent directory matches 1310 * the cached copy in the name cache entry. 1311 * Otherwise, we discard all of the negative cache 1312 * entries for this directory. We also only trust 1313 * negative cache entries for up to nm_negnametimeo 1314 * seconds. 1315 */ 1316 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1317 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1318 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1319 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1320 return (ENOENT); 1321 } 1322 cache_purge_negative(dvp); 1323 } 1324 1325 openmode = 0; 1326 /* 1327 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount 1328 * option, it is possible to do the Open operation in the same 1329 * compound as Lookup, so long as delegations are not being 1330 * issued. This saves doing a separate RPC for Open. 1331 * For pnfs, do not do this, since the Open+LayoutGet will 1332 * be needed as a separate RPC. 1333 */ 1334 NFSLOCKMNT(nmp); 1335 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) && 1336 (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 && 1337 (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) && 1338 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) { 1339 if ((flags & OPENREAD) != 0) 1340 openmode |= NFSV4OPEN_ACCESSREAD; 1341 if ((flags & OPENWRITE) != 0) 1342 openmode |= NFSV4OPEN_ACCESSWRITE; 1343 } 1344 NFSUNLOCKMNT(nmp); 1345 1346 newvp = NULLVP; 1347 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1348 nanouptime(&ts); 1349 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1350 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1351 NULL, openmode); 1352 if (dattrflag) 1353 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1354 if (error) { 1355 if (newvp != NULLVP) { 1356 vput(newvp); 1357 *vpp = NULLVP; 1358 } 1359 1360 if (error != ENOENT) { 1361 if (NFS_ISV4(dvp)) 1362 error = nfscl_maperr(td, error, (uid_t)0, 1363 (gid_t)0); 1364 return (error); 1365 } 1366 1367 /* The requested file was not found. */ 1368 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1369 (flags & ISLASTCN)) { 1370 /* 1371 * XXX: UFS does a full VOP_ACCESS(dvp, 1372 * VWRITE) here instead of just checking 1373 * MNT_RDONLY. 1374 */ 1375 if (mp->mnt_flag & MNT_RDONLY) 1376 return (EROFS); 1377 cnp->cn_flags |= SAVENAME; 1378 return (EJUSTRETURN); 1379 } 1380 1381 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1382 /* 1383 * Cache the modification time of the parent 1384 * directory from the post-op attributes in 1385 * the name cache entry. The negative cache 1386 * entry will be ignored once the directory 1387 * has changed. Don't bother adding the entry 1388 * if the directory has already changed. 1389 */ 1390 NFSLOCKNODE(np); 1391 if (timespeccmp(&np->n_vattr.na_mtime, 1392 &dnfsva.na_mtime, ==)) { 1393 NFSUNLOCKNODE(np); 1394 cache_enter_time(dvp, NULL, cnp, 1395 &dnfsva.na_mtime, NULL); 1396 } else 1397 NFSUNLOCKNODE(np); 1398 } 1399 return (ENOENT); 1400 } 1401 1402 /* 1403 * Handle RENAME case... 1404 */ 1405 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1406 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1407 free(nfhp, M_NFSFH); 1408 return (EISDIR); 1409 } 1410 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, LK_EXCLUSIVE); 1411 if (error) 1412 return (error); 1413 newvp = NFSTOV(np); 1414 /* 1415 * If n_localmodtime >= time before RPC, then 1416 * a file modification operation, such as 1417 * VOP_SETATTR() of size, has occurred while 1418 * the Lookup RPC and acquisition of the vnode 1419 * happened. As such, the attributes might 1420 * be stale, with possibly an incorrect size. 1421 */ 1422 NFSLOCKNODE(np); 1423 if (timespecisset(&np->n_localmodtime) && 1424 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1425 NFSCL_DEBUG(4, "nfs_lookup: rename localmod " 1426 "stale attributes\n"); 1427 attrflag = 0; 1428 } 1429 NFSUNLOCKNODE(np); 1430 if (attrflag) 1431 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1432 *vpp = newvp; 1433 cnp->cn_flags |= SAVENAME; 1434 return (0); 1435 } 1436 1437 if (flags & ISDOTDOT) { 1438 ltype = NFSVOPISLOCKED(dvp); 1439 error = vfs_busy(mp, MBF_NOWAIT); 1440 if (error != 0) { 1441 vfs_ref(mp); 1442 NFSVOPUNLOCK(dvp); 1443 error = vfs_busy(mp, 0); 1444 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1445 vfs_rel(mp); 1446 if (error == 0 && VN_IS_DOOMED(dvp)) { 1447 vfs_unbusy(mp); 1448 error = ENOENT; 1449 } 1450 if (error != 0) 1451 return (error); 1452 } 1453 NFSVOPUNLOCK(dvp); 1454 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1455 cnp->cn_lkflags); 1456 if (error == 0) 1457 newvp = NFSTOV(np); 1458 vfs_unbusy(mp); 1459 if (newvp != dvp) 1460 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1461 if (VN_IS_DOOMED(dvp)) { 1462 if (error == 0) { 1463 if (newvp == dvp) 1464 vrele(newvp); 1465 else 1466 vput(newvp); 1467 } 1468 error = ENOENT; 1469 } 1470 if (error != 0) 1471 return (error); 1472 if (attrflag) 1473 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1474 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1475 free(nfhp, M_NFSFH); 1476 VREF(dvp); 1477 newvp = dvp; 1478 if (attrflag) 1479 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1480 } else { 1481 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1482 cnp->cn_lkflags); 1483 if (error) 1484 return (error); 1485 newvp = NFSTOV(np); 1486 /* 1487 * If n_localmodtime >= time before RPC, then 1488 * a file modification operation, such as 1489 * VOP_SETATTR() of size, has occurred while 1490 * the Lookup RPC and acquisition of the vnode 1491 * happened. As such, the attributes might 1492 * be stale, with possibly an incorrect size. 1493 */ 1494 NFSLOCKNODE(np); 1495 if (timespecisset(&np->n_localmodtime) && 1496 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1497 NFSCL_DEBUG(4, "nfs_lookup: localmod " 1498 "stale attributes\n"); 1499 attrflag = 0; 1500 } 1501 NFSUNLOCKNODE(np); 1502 if (attrflag) 1503 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1504 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1505 !(np->n_flag & NMODIFIED)) { 1506 /* 1507 * Flush the attribute cache when opening a 1508 * leaf node to ensure that fresh attributes 1509 * are fetched in nfs_open() since we did not 1510 * fetch attributes from the LOOKUP reply. 1511 */ 1512 NFSLOCKNODE(np); 1513 np->n_attrstamp = 0; 1514 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1515 NFSUNLOCKNODE(np); 1516 } 1517 } 1518 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1519 cnp->cn_flags |= SAVENAME; 1520 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp && 1521 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1522 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1523 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1524 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1525 *vpp = newvp; 1526 return (0); 1527 } 1528 1529 /* 1530 * nfs read call. 1531 * Just call ncl_bioread() to do the work. 1532 */ 1533 static int 1534 nfs_read(struct vop_read_args *ap) 1535 { 1536 struct vnode *vp = ap->a_vp; 1537 1538 switch (vp->v_type) { 1539 case VREG: 1540 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1541 case VDIR: 1542 return (EISDIR); 1543 default: 1544 return (EOPNOTSUPP); 1545 } 1546 } 1547 1548 /* 1549 * nfs readlink call 1550 */ 1551 static int 1552 nfs_readlink(struct vop_readlink_args *ap) 1553 { 1554 struct vnode *vp = ap->a_vp; 1555 1556 if (vp->v_type != VLNK) 1557 return (EINVAL); 1558 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1559 } 1560 1561 /* 1562 * Do a readlink rpc. 1563 * Called by ncl_doio() from below the buffer cache. 1564 */ 1565 int 1566 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1567 { 1568 int error, ret, attrflag; 1569 struct nfsvattr nfsva; 1570 1571 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1572 &attrflag, NULL); 1573 if (attrflag) { 1574 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1575 if (ret && !error) 1576 error = ret; 1577 } 1578 if (error && NFS_ISV4(vp)) 1579 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1580 return (error); 1581 } 1582 1583 /* 1584 * nfs read rpc call 1585 * Ditto above 1586 */ 1587 int 1588 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1589 { 1590 int error, ret, attrflag; 1591 struct nfsvattr nfsva; 1592 struct nfsmount *nmp; 1593 1594 nmp = VFSTONFS(vp->v_mount); 1595 error = EIO; 1596 attrflag = 0; 1597 if (NFSHASPNFS(nmp)) 1598 error = nfscl_doiods(vp, uiop, NULL, NULL, 1599 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1600 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1601 if (error != 0) 1602 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1603 &attrflag, NULL); 1604 if (attrflag) { 1605 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1606 if (ret && !error) 1607 error = ret; 1608 } 1609 if (error && NFS_ISV4(vp)) 1610 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1611 return (error); 1612 } 1613 1614 /* 1615 * nfs write call 1616 */ 1617 int 1618 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1619 int *iomode, int *must_commit, int called_from_strategy, int ioflag) 1620 { 1621 struct nfsvattr nfsva; 1622 int error, attrflag, ret; 1623 struct nfsmount *nmp; 1624 1625 nmp = VFSTONFS(vp->v_mount); 1626 error = EIO; 1627 attrflag = 0; 1628 if (NFSHASPNFS(nmp)) 1629 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1630 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1631 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1632 if (error != 0) 1633 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1634 uiop->uio_td, &nfsva, &attrflag, called_from_strategy, 1635 ioflag); 1636 if (attrflag) { 1637 if (VTONFS(vp)->n_flag & ND_NFSV4) 1638 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1); 1639 else 1640 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1641 if (ret && !error) 1642 error = ret; 1643 } 1644 if (DOINGASYNC(vp)) 1645 *iomode = NFSWRITE_FILESYNC; 1646 if (error && NFS_ISV4(vp)) 1647 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1648 return (error); 1649 } 1650 1651 /* 1652 * nfs mknod rpc 1653 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1654 * mode set to specify the file type and the size field for rdev. 1655 */ 1656 static int 1657 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1658 struct vattr *vap) 1659 { 1660 struct nfsvattr nfsva, dnfsva; 1661 struct vnode *newvp = NULL; 1662 struct nfsnode *np = NULL, *dnp; 1663 struct nfsfh *nfhp; 1664 struct vattr vattr; 1665 int error = 0, attrflag, dattrflag; 1666 u_int32_t rdev; 1667 1668 if (vap->va_type == VCHR || vap->va_type == VBLK) 1669 rdev = vap->va_rdev; 1670 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1671 rdev = 0xffffffff; 1672 else 1673 return (EOPNOTSUPP); 1674 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1675 return (error); 1676 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1677 rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva, 1678 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1679 if (!error) { 1680 if (!nfhp) 1681 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1682 cnp->cn_namelen, cnp->cn_cred, curthread, 1683 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1684 NULL, 0); 1685 if (nfhp) 1686 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1687 curthread, &np, LK_EXCLUSIVE); 1688 } 1689 if (dattrflag) 1690 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1691 if (!error) { 1692 newvp = NFSTOV(np); 1693 if (attrflag != 0) { 1694 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1695 if (error != 0) 1696 vput(newvp); 1697 } 1698 } 1699 if (!error) { 1700 *vpp = newvp; 1701 } else if (NFS_ISV4(dvp)) { 1702 error = nfscl_maperr(curthread, error, vap->va_uid, 1703 vap->va_gid); 1704 } 1705 dnp = VTONFS(dvp); 1706 NFSLOCKNODE(dnp); 1707 dnp->n_flag |= NMODIFIED; 1708 if (!dattrflag) { 1709 dnp->n_attrstamp = 0; 1710 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1711 } 1712 NFSUNLOCKNODE(dnp); 1713 return (error); 1714 } 1715 1716 /* 1717 * nfs mknod vop 1718 * just call nfs_mknodrpc() to do the work. 1719 */ 1720 /* ARGSUSED */ 1721 static int 1722 nfs_mknod(struct vop_mknod_args *ap) 1723 { 1724 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1725 } 1726 1727 static struct mtx nfs_cverf_mtx; 1728 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1729 MTX_DEF); 1730 1731 static nfsquad_t 1732 nfs_get_cverf(void) 1733 { 1734 static nfsquad_t cverf; 1735 nfsquad_t ret; 1736 static int cverf_initialized = 0; 1737 1738 mtx_lock(&nfs_cverf_mtx); 1739 if (cverf_initialized == 0) { 1740 cverf.lval[0] = arc4random(); 1741 cverf.lval[1] = arc4random(); 1742 cverf_initialized = 1; 1743 } else 1744 cverf.qval++; 1745 ret = cverf; 1746 mtx_unlock(&nfs_cverf_mtx); 1747 1748 return (ret); 1749 } 1750 1751 /* 1752 * nfs file create call 1753 */ 1754 static int 1755 nfs_create(struct vop_create_args *ap) 1756 { 1757 struct vnode *dvp = ap->a_dvp; 1758 struct vattr *vap = ap->a_vap; 1759 struct componentname *cnp = ap->a_cnp; 1760 struct nfsnode *np = NULL, *dnp; 1761 struct vnode *newvp = NULL; 1762 struct nfsmount *nmp; 1763 struct nfsvattr dnfsva, nfsva; 1764 struct nfsfh *nfhp; 1765 nfsquad_t cverf; 1766 int error = 0, attrflag, dattrflag, fmode = 0; 1767 struct vattr vattr; 1768 1769 /* 1770 * Oops, not for me.. 1771 */ 1772 if (vap->va_type == VSOCK) 1773 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1774 1775 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1776 return (error); 1777 if (vap->va_vaflags & VA_EXCLUSIVE) 1778 fmode |= O_EXCL; 1779 dnp = VTONFS(dvp); 1780 nmp = VFSTONFS(dvp->v_mount); 1781 again: 1782 /* For NFSv4, wait until any remove is done. */ 1783 NFSLOCKNODE(dnp); 1784 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1785 dnp->n_flag |= NREMOVEWANT; 1786 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1787 } 1788 NFSUNLOCKNODE(dnp); 1789 1790 cverf = nfs_get_cverf(); 1791 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1792 vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva, 1793 &nfhp, &attrflag, &dattrflag, NULL); 1794 if (!error) { 1795 if (nfhp == NULL) 1796 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1797 cnp->cn_namelen, cnp->cn_cred, curthread, 1798 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1799 NULL, 0); 1800 if (nfhp != NULL) 1801 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1802 curthread, &np, LK_EXCLUSIVE); 1803 } 1804 if (dattrflag) 1805 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1806 if (!error) { 1807 newvp = NFSTOV(np); 1808 if (attrflag == 0) 1809 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1810 curthread, &nfsva, NULL); 1811 if (error == 0) 1812 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1813 } 1814 if (error) { 1815 if (newvp != NULL) { 1816 vput(newvp); 1817 newvp = NULL; 1818 } 1819 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1820 error == NFSERR_NOTSUPP) { 1821 fmode &= ~O_EXCL; 1822 goto again; 1823 } 1824 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1825 if (nfscl_checksattr(vap, &nfsva)) { 1826 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1827 curthread, &nfsva, &attrflag, NULL); 1828 if (error && (vap->va_uid != (uid_t)VNOVAL || 1829 vap->va_gid != (gid_t)VNOVAL)) { 1830 /* try again without setting uid/gid */ 1831 vap->va_uid = (uid_t)VNOVAL; 1832 vap->va_gid = (uid_t)VNOVAL; 1833 error = nfsrpc_setattr(newvp, vap, NULL, 1834 cnp->cn_cred, curthread, &nfsva, 1835 &attrflag, NULL); 1836 } 1837 if (attrflag) 1838 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1839 0, 1); 1840 if (error != 0) 1841 vput(newvp); 1842 } 1843 } 1844 if (!error) { 1845 if ((cnp->cn_flags & MAKEENTRY) && attrflag) { 1846 if (dvp != newvp) 1847 cache_enter_time(dvp, newvp, cnp, 1848 &nfsva.na_ctime, NULL); 1849 else 1850 printf("nfs_create: bogus NFS server returned " 1851 "the directory as the new file object\n"); 1852 } 1853 *ap->a_vpp = newvp; 1854 } else if (NFS_ISV4(dvp)) { 1855 error = nfscl_maperr(curthread, error, vap->va_uid, 1856 vap->va_gid); 1857 } 1858 NFSLOCKNODE(dnp); 1859 dnp->n_flag |= NMODIFIED; 1860 if (!dattrflag) { 1861 dnp->n_attrstamp = 0; 1862 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1863 } 1864 NFSUNLOCKNODE(dnp); 1865 return (error); 1866 } 1867 1868 /* 1869 * nfs file remove call 1870 * To try and make nfs semantics closer to ufs semantics, a file that has 1871 * other processes using the vnode is renamed instead of removed and then 1872 * removed later on the last close. 1873 * - If v_usecount > 1 1874 * If a rename is not already in the works 1875 * call nfs_sillyrename() to set it up 1876 * else 1877 * do the remove rpc 1878 */ 1879 static int 1880 nfs_remove(struct vop_remove_args *ap) 1881 { 1882 struct vnode *vp = ap->a_vp; 1883 struct vnode *dvp = ap->a_dvp; 1884 struct componentname *cnp = ap->a_cnp; 1885 struct nfsnode *np = VTONFS(vp); 1886 int error = 0; 1887 struct vattr vattr; 1888 1889 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1890 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1891 if (vp->v_type == VDIR) 1892 error = EPERM; 1893 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1894 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1895 vattr.va_nlink > 1)) { 1896 /* 1897 * Purge the name cache so that the chance of a lookup for 1898 * the name succeeding while the remove is in progress is 1899 * minimized. Without node locking it can still happen, such 1900 * that an I/O op returns ESTALE, but since you get this if 1901 * another host removes the file.. 1902 */ 1903 cache_purge(vp); 1904 /* 1905 * throw away biocache buffers, mainly to avoid 1906 * unnecessary delayed writes later. 1907 */ 1908 error = ncl_vinvalbuf(vp, 0, curthread, 1); 1909 if (error != EINTR && error != EIO) 1910 /* Do the rpc */ 1911 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1912 cnp->cn_namelen, cnp->cn_cred, curthread); 1913 /* 1914 * Kludge City: If the first reply to the remove rpc is lost.. 1915 * the reply to the retransmitted request will be ENOENT 1916 * since the file was in fact removed 1917 * Therefore, we cheat and return success. 1918 */ 1919 if (error == ENOENT) 1920 error = 0; 1921 } else if (!np->n_sillyrename) 1922 error = nfs_sillyrename(dvp, vp, cnp); 1923 NFSLOCKNODE(np); 1924 np->n_attrstamp = 0; 1925 NFSUNLOCKNODE(np); 1926 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1927 return (error); 1928 } 1929 1930 /* 1931 * nfs file remove rpc called from nfs_inactive 1932 */ 1933 int 1934 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1935 { 1936 /* 1937 * Make sure that the directory vnode is still valid. 1938 * XXX we should lock sp->s_dvp here. 1939 */ 1940 if (sp->s_dvp->v_type == VBAD) 1941 return (0); 1942 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1943 sp->s_cred, NULL)); 1944 } 1945 1946 /* 1947 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1948 */ 1949 static int 1950 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1951 int namelen, struct ucred *cred, struct thread *td) 1952 { 1953 struct nfsvattr dnfsva; 1954 struct nfsnode *dnp = VTONFS(dvp); 1955 int error = 0, dattrflag; 1956 1957 NFSLOCKNODE(dnp); 1958 dnp->n_flag |= NREMOVEINPROG; 1959 NFSUNLOCKNODE(dnp); 1960 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1961 &dattrflag, NULL); 1962 NFSLOCKNODE(dnp); 1963 if ((dnp->n_flag & NREMOVEWANT)) { 1964 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1965 NFSUNLOCKNODE(dnp); 1966 wakeup((caddr_t)dnp); 1967 } else { 1968 dnp->n_flag &= ~NREMOVEINPROG; 1969 NFSUNLOCKNODE(dnp); 1970 } 1971 if (dattrflag) 1972 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1973 NFSLOCKNODE(dnp); 1974 dnp->n_flag |= NMODIFIED; 1975 if (!dattrflag) { 1976 dnp->n_attrstamp = 0; 1977 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1978 } 1979 NFSUNLOCKNODE(dnp); 1980 if (error && NFS_ISV4(dvp)) 1981 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1982 return (error); 1983 } 1984 1985 /* 1986 * nfs file rename call 1987 */ 1988 static int 1989 nfs_rename(struct vop_rename_args *ap) 1990 { 1991 struct vnode *fvp = ap->a_fvp; 1992 struct vnode *tvp = ap->a_tvp; 1993 struct vnode *fdvp = ap->a_fdvp; 1994 struct vnode *tdvp = ap->a_tdvp; 1995 struct componentname *tcnp = ap->a_tcnp; 1996 struct componentname *fcnp = ap->a_fcnp; 1997 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1998 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1999 struct nfsv4node *newv4 = NULL; 2000 int error; 2001 2002 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 2003 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 2004 /* Check for cross-device rename */ 2005 if ((fvp->v_mount != tdvp->v_mount) || 2006 (tvp && (fvp->v_mount != tvp->v_mount))) { 2007 error = EXDEV; 2008 goto out; 2009 } 2010 2011 if (fvp == tvp) { 2012 printf("nfs_rename: fvp == tvp (can't happen)\n"); 2013 error = 0; 2014 goto out; 2015 } 2016 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 2017 goto out; 2018 2019 /* 2020 * We have to flush B_DELWRI data prior to renaming 2021 * the file. If we don't, the delayed-write buffers 2022 * can be flushed out later after the file has gone stale 2023 * under NFSV3. NFSV2 does not have this problem because 2024 * ( as far as I can tell ) it flushes dirty buffers more 2025 * often. 2026 * 2027 * Skip the rename operation if the fsync fails, this can happen 2028 * due to the server's volume being full, when we pushed out data 2029 * that was written back to our cache earlier. Not checking for 2030 * this condition can result in potential (silent) data loss. 2031 */ 2032 error = VOP_FSYNC(fvp, MNT_WAIT, curthread); 2033 NFSVOPUNLOCK(fvp); 2034 if (!error && tvp) 2035 error = VOP_FSYNC(tvp, MNT_WAIT, curthread); 2036 if (error) 2037 goto out; 2038 2039 /* 2040 * If the tvp exists and is in use, sillyrename it before doing the 2041 * rename of the new file over it. 2042 * XXX Can't sillyrename a directory. 2043 */ 2044 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 2045 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 2046 vput(tvp); 2047 tvp = NULL; 2048 } 2049 2050 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 2051 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 2052 curthread); 2053 2054 if (error == 0 && NFS_ISV4(tdvp)) { 2055 /* 2056 * For NFSv4, check to see if it is the same name and 2057 * replace the name, if it is different. 2058 */ 2059 newv4 = malloc( 2060 sizeof (struct nfsv4node) + 2061 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 2062 M_NFSV4NODE, M_WAITOK); 2063 NFSLOCKNODE(tdnp); 2064 NFSLOCKNODE(fnp); 2065 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 2066 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 2067 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 2068 tcnp->cn_namelen) || 2069 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 2070 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2071 tdnp->n_fhp->nfh_len))) { 2072 #ifdef notdef 2073 { char nnn[100]; int nnnl; 2074 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 2075 bcopy(tcnp->cn_nameptr, nnn, nnnl); 2076 nnn[nnnl] = '\0'; 2077 printf("ren replace=%s\n",nnn); 2078 } 2079 #endif 2080 free(fnp->n_v4, M_NFSV4NODE); 2081 fnp->n_v4 = newv4; 2082 newv4 = NULL; 2083 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 2084 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 2085 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2086 tdnp->n_fhp->nfh_len); 2087 NFSBCOPY(tcnp->cn_nameptr, 2088 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 2089 } 2090 NFSUNLOCKNODE(tdnp); 2091 NFSUNLOCKNODE(fnp); 2092 if (newv4 != NULL) 2093 free(newv4, M_NFSV4NODE); 2094 } 2095 2096 if (fvp->v_type == VDIR) { 2097 if (tvp != NULL && tvp->v_type == VDIR) 2098 cache_purge(tdvp); 2099 cache_purge(fdvp); 2100 } 2101 2102 out: 2103 if (tdvp == tvp) 2104 vrele(tdvp); 2105 else 2106 vput(tdvp); 2107 if (tvp) 2108 vput(tvp); 2109 vrele(fdvp); 2110 vrele(fvp); 2111 /* 2112 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2113 */ 2114 if (error == ENOENT) 2115 error = 0; 2116 return (error); 2117 } 2118 2119 /* 2120 * nfs file rename rpc called from nfs_remove() above 2121 */ 2122 static int 2123 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 2124 struct sillyrename *sp) 2125 { 2126 2127 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 2128 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 2129 curthread)); 2130 } 2131 2132 /* 2133 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2134 */ 2135 static int 2136 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 2137 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 2138 int tnamelen, struct ucred *cred, struct thread *td) 2139 { 2140 struct nfsvattr fnfsva, tnfsva; 2141 struct nfsnode *fdnp = VTONFS(fdvp); 2142 struct nfsnode *tdnp = VTONFS(tdvp); 2143 int error = 0, fattrflag, tattrflag; 2144 2145 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 2146 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 2147 &tattrflag, NULL, NULL); 2148 NFSLOCKNODE(fdnp); 2149 fdnp->n_flag |= NMODIFIED; 2150 if (fattrflag != 0) { 2151 NFSUNLOCKNODE(fdnp); 2152 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1); 2153 } else { 2154 fdnp->n_attrstamp = 0; 2155 NFSUNLOCKNODE(fdnp); 2156 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 2157 } 2158 NFSLOCKNODE(tdnp); 2159 tdnp->n_flag |= NMODIFIED; 2160 if (tattrflag != 0) { 2161 NFSUNLOCKNODE(tdnp); 2162 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1); 2163 } else { 2164 tdnp->n_attrstamp = 0; 2165 NFSUNLOCKNODE(tdnp); 2166 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2167 } 2168 if (error && NFS_ISV4(fdvp)) 2169 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2170 return (error); 2171 } 2172 2173 /* 2174 * nfs hard link create call 2175 */ 2176 static int 2177 nfs_link(struct vop_link_args *ap) 2178 { 2179 struct vnode *vp = ap->a_vp; 2180 struct vnode *tdvp = ap->a_tdvp; 2181 struct componentname *cnp = ap->a_cnp; 2182 struct nfsnode *np, *tdnp; 2183 struct nfsvattr nfsva, dnfsva; 2184 int error = 0, attrflag, dattrflag; 2185 2186 /* 2187 * Push all writes to the server, so that the attribute cache 2188 * doesn't get "out of sync" with the server. 2189 * XXX There should be a better way! 2190 */ 2191 VOP_FSYNC(vp, MNT_WAIT, curthread); 2192 2193 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2194 cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, 2195 &dattrflag, NULL); 2196 tdnp = VTONFS(tdvp); 2197 NFSLOCKNODE(tdnp); 2198 tdnp->n_flag |= NMODIFIED; 2199 if (dattrflag != 0) { 2200 NFSUNLOCKNODE(tdnp); 2201 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1); 2202 } else { 2203 tdnp->n_attrstamp = 0; 2204 NFSUNLOCKNODE(tdnp); 2205 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2206 } 2207 if (attrflag) 2208 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2209 else { 2210 np = VTONFS(vp); 2211 NFSLOCKNODE(np); 2212 np->n_attrstamp = 0; 2213 NFSUNLOCKNODE(np); 2214 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2215 } 2216 /* 2217 * If negative lookup caching is enabled, I might as well 2218 * add an entry for this node. Not necessary for correctness, 2219 * but if negative caching is enabled, then the system 2220 * must care about lookup caching hit rate, so... 2221 */ 2222 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2223 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2224 if (tdvp != vp) 2225 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2226 else 2227 printf("nfs_link: bogus NFS server returned " 2228 "the directory as the new link\n"); 2229 } 2230 if (error && NFS_ISV4(vp)) 2231 error = nfscl_maperr(curthread, error, (uid_t)0, 2232 (gid_t)0); 2233 return (error); 2234 } 2235 2236 /* 2237 * nfs symbolic link create call 2238 */ 2239 static int 2240 nfs_symlink(struct vop_symlink_args *ap) 2241 { 2242 struct vnode *dvp = ap->a_dvp; 2243 struct vattr *vap = ap->a_vap; 2244 struct componentname *cnp = ap->a_cnp; 2245 struct nfsvattr nfsva, dnfsva; 2246 struct nfsfh *nfhp; 2247 struct nfsnode *np = NULL, *dnp; 2248 struct vnode *newvp = NULL; 2249 int error = 0, attrflag, dattrflag, ret; 2250 2251 vap->va_type = VLNK; 2252 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2253 ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva, 2254 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2255 if (nfhp) { 2256 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2257 &np, LK_EXCLUSIVE); 2258 if (!ret) 2259 newvp = NFSTOV(np); 2260 else if (!error) 2261 error = ret; 2262 } 2263 if (newvp != NULL) { 2264 if (attrflag) 2265 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2266 } else if (!error) { 2267 /* 2268 * If we do not have an error and we could not extract the 2269 * newvp from the response due to the request being NFSv2, we 2270 * have to do a lookup in order to obtain a newvp to return. 2271 */ 2272 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2273 cnp->cn_cred, curthread, &np); 2274 if (!error) 2275 newvp = NFSTOV(np); 2276 } 2277 if (error) { 2278 if (newvp) 2279 vput(newvp); 2280 if (NFS_ISV4(dvp)) 2281 error = nfscl_maperr(curthread, error, 2282 vap->va_uid, vap->va_gid); 2283 } else { 2284 *ap->a_vpp = newvp; 2285 } 2286 2287 dnp = VTONFS(dvp); 2288 NFSLOCKNODE(dnp); 2289 dnp->n_flag |= NMODIFIED; 2290 if (dattrflag != 0) { 2291 NFSUNLOCKNODE(dnp); 2292 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2293 } else { 2294 dnp->n_attrstamp = 0; 2295 NFSUNLOCKNODE(dnp); 2296 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2297 } 2298 /* 2299 * If negative lookup caching is enabled, I might as well 2300 * add an entry for this node. Not necessary for correctness, 2301 * but if negative caching is enabled, then the system 2302 * must care about lookup caching hit rate, so... 2303 */ 2304 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2305 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2306 if (dvp != newvp) 2307 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2308 NULL); 2309 else 2310 printf("nfs_symlink: bogus NFS server returned " 2311 "the directory as the new file object\n"); 2312 } 2313 return (error); 2314 } 2315 2316 /* 2317 * nfs make dir call 2318 */ 2319 static int 2320 nfs_mkdir(struct vop_mkdir_args *ap) 2321 { 2322 struct vnode *dvp = ap->a_dvp; 2323 struct vattr *vap = ap->a_vap; 2324 struct componentname *cnp = ap->a_cnp; 2325 struct nfsnode *np = NULL, *dnp; 2326 struct vnode *newvp = NULL; 2327 struct vattr vattr; 2328 struct nfsfh *nfhp; 2329 struct nfsvattr nfsva, dnfsva; 2330 int error = 0, attrflag, dattrflag, ret; 2331 2332 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2333 return (error); 2334 vap->va_type = VDIR; 2335 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2336 vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, 2337 &attrflag, &dattrflag, NULL); 2338 dnp = VTONFS(dvp); 2339 NFSLOCKNODE(dnp); 2340 dnp->n_flag |= NMODIFIED; 2341 if (dattrflag != 0) { 2342 NFSUNLOCKNODE(dnp); 2343 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2344 } else { 2345 dnp->n_attrstamp = 0; 2346 NFSUNLOCKNODE(dnp); 2347 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2348 } 2349 if (nfhp) { 2350 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2351 &np, LK_EXCLUSIVE); 2352 if (!ret) { 2353 newvp = NFSTOV(np); 2354 if (attrflag) 2355 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2356 0, 1); 2357 } else if (!error) 2358 error = ret; 2359 } 2360 if (!error && newvp == NULL) { 2361 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2362 cnp->cn_cred, curthread, &np); 2363 if (!error) { 2364 newvp = NFSTOV(np); 2365 if (newvp->v_type != VDIR) 2366 error = EEXIST; 2367 } 2368 } 2369 if (error) { 2370 if (newvp) 2371 vput(newvp); 2372 if (NFS_ISV4(dvp)) 2373 error = nfscl_maperr(curthread, error, 2374 vap->va_uid, vap->va_gid); 2375 } else { 2376 /* 2377 * If negative lookup caching is enabled, I might as well 2378 * add an entry for this node. Not necessary for correctness, 2379 * but if negative caching is enabled, then the system 2380 * must care about lookup caching hit rate, so... 2381 */ 2382 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2383 (cnp->cn_flags & MAKEENTRY) && 2384 attrflag != 0 && dattrflag != 0) { 2385 if (dvp != newvp) 2386 cache_enter_time(dvp, newvp, cnp, 2387 &nfsva.na_ctime, &dnfsva.na_ctime); 2388 else 2389 printf("nfs_mkdir: bogus NFS server returned " 2390 "the directory that the directory was " 2391 "created in as the new file object\n"); 2392 } 2393 *ap->a_vpp = newvp; 2394 } 2395 return (error); 2396 } 2397 2398 /* 2399 * nfs remove directory call 2400 */ 2401 static int 2402 nfs_rmdir(struct vop_rmdir_args *ap) 2403 { 2404 struct vnode *vp = ap->a_vp; 2405 struct vnode *dvp = ap->a_dvp; 2406 struct componentname *cnp = ap->a_cnp; 2407 struct nfsnode *dnp; 2408 struct nfsvattr dnfsva; 2409 int error, dattrflag; 2410 2411 if (dvp == vp) 2412 return (EINVAL); 2413 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2414 cnp->cn_cred, curthread, &dnfsva, &dattrflag, NULL); 2415 dnp = VTONFS(dvp); 2416 NFSLOCKNODE(dnp); 2417 dnp->n_flag |= NMODIFIED; 2418 if (dattrflag != 0) { 2419 NFSUNLOCKNODE(dnp); 2420 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2421 } else { 2422 dnp->n_attrstamp = 0; 2423 NFSUNLOCKNODE(dnp); 2424 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2425 } 2426 2427 cache_purge(dvp); 2428 cache_purge(vp); 2429 if (error && NFS_ISV4(dvp)) 2430 error = nfscl_maperr(curthread, error, (uid_t)0, 2431 (gid_t)0); 2432 /* 2433 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2434 */ 2435 if (error == ENOENT) 2436 error = 0; 2437 return (error); 2438 } 2439 2440 /* 2441 * nfs readdir call 2442 */ 2443 static int 2444 nfs_readdir(struct vop_readdir_args *ap) 2445 { 2446 struct vnode *vp = ap->a_vp; 2447 struct nfsnode *np = VTONFS(vp); 2448 struct uio *uio = ap->a_uio; 2449 ssize_t tresid, left; 2450 int error = 0; 2451 struct vattr vattr; 2452 2453 if (ap->a_eofflag != NULL) 2454 *ap->a_eofflag = 0; 2455 if (vp->v_type != VDIR) 2456 return(EPERM); 2457 2458 /* 2459 * First, check for hit on the EOF offset cache 2460 */ 2461 NFSLOCKNODE(np); 2462 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2463 (np->n_flag & NMODIFIED) == 0) { 2464 NFSUNLOCKNODE(np); 2465 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2466 NFSLOCKNODE(np); 2467 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2468 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2469 NFSUNLOCKNODE(np); 2470 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2471 if (ap->a_eofflag != NULL) 2472 *ap->a_eofflag = 1; 2473 return (0); 2474 } else 2475 NFSUNLOCKNODE(np); 2476 } 2477 } else 2478 NFSUNLOCKNODE(np); 2479 2480 /* 2481 * NFS always guarantees that directory entries don't straddle 2482 * DIRBLKSIZ boundaries. As such, we need to limit the size 2483 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2484 * directory entry. 2485 */ 2486 left = uio->uio_resid % DIRBLKSIZ; 2487 if (left == uio->uio_resid) 2488 return (EINVAL); 2489 uio->uio_resid -= left; 2490 2491 /* 2492 * Call ncl_bioread() to do the real work. 2493 */ 2494 tresid = uio->uio_resid; 2495 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2496 2497 if (!error && uio->uio_resid == tresid) { 2498 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2499 if (ap->a_eofflag != NULL) 2500 *ap->a_eofflag = 1; 2501 } 2502 2503 /* Add the partial DIRBLKSIZ (left) back in. */ 2504 uio->uio_resid += left; 2505 return (error); 2506 } 2507 2508 /* 2509 * Readdir rpc call. 2510 * Called from below the buffer cache by ncl_doio(). 2511 */ 2512 int 2513 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2514 struct thread *td) 2515 { 2516 struct nfsvattr nfsva; 2517 nfsuint64 *cookiep, cookie; 2518 struct nfsnode *dnp = VTONFS(vp); 2519 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2520 int error = 0, eof, attrflag; 2521 2522 KASSERT(uiop->uio_iovcnt == 1 && 2523 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2524 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2525 ("nfs readdirrpc bad uio")); 2526 2527 /* 2528 * If there is no cookie, assume directory was stale. 2529 */ 2530 ncl_dircookie_lock(dnp); 2531 NFSUNLOCKNODE(dnp); 2532 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2533 if (cookiep) { 2534 cookie = *cookiep; 2535 ncl_dircookie_unlock(dnp); 2536 } else { 2537 ncl_dircookie_unlock(dnp); 2538 return (NFSERR_BAD_COOKIE); 2539 } 2540 2541 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2542 (void)ncl_fsinfo(nmp, vp, cred, td); 2543 2544 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2545 &attrflag, &eof, NULL); 2546 if (attrflag) 2547 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2548 2549 if (!error) { 2550 /* 2551 * We are now either at the end of the directory or have filled 2552 * the block. 2553 */ 2554 if (eof) { 2555 NFSLOCKNODE(dnp); 2556 dnp->n_direofoffset = uiop->uio_offset; 2557 NFSUNLOCKNODE(dnp); 2558 } else { 2559 if (uiop->uio_resid > 0) 2560 printf("EEK! readdirrpc resid > 0\n"); 2561 ncl_dircookie_lock(dnp); 2562 NFSUNLOCKNODE(dnp); 2563 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2564 *cookiep = cookie; 2565 ncl_dircookie_unlock(dnp); 2566 } 2567 } else if (NFS_ISV4(vp)) { 2568 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2569 } 2570 return (error); 2571 } 2572 2573 /* 2574 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2575 */ 2576 int 2577 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2578 struct thread *td) 2579 { 2580 struct nfsvattr nfsva; 2581 nfsuint64 *cookiep, cookie; 2582 struct nfsnode *dnp = VTONFS(vp); 2583 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2584 int error = 0, attrflag, eof; 2585 2586 KASSERT(uiop->uio_iovcnt == 1 && 2587 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2588 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2589 ("nfs readdirplusrpc bad uio")); 2590 2591 /* 2592 * If there is no cookie, assume directory was stale. 2593 */ 2594 ncl_dircookie_lock(dnp); 2595 NFSUNLOCKNODE(dnp); 2596 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2597 if (cookiep) { 2598 cookie = *cookiep; 2599 ncl_dircookie_unlock(dnp); 2600 } else { 2601 ncl_dircookie_unlock(dnp); 2602 return (NFSERR_BAD_COOKIE); 2603 } 2604 2605 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2606 (void)ncl_fsinfo(nmp, vp, cred, td); 2607 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2608 &attrflag, &eof, NULL); 2609 if (attrflag) 2610 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2611 2612 if (!error) { 2613 /* 2614 * We are now either at end of the directory or have filled the 2615 * the block. 2616 */ 2617 if (eof) { 2618 NFSLOCKNODE(dnp); 2619 dnp->n_direofoffset = uiop->uio_offset; 2620 NFSUNLOCKNODE(dnp); 2621 } else { 2622 if (uiop->uio_resid > 0) 2623 printf("EEK! readdirplusrpc resid > 0\n"); 2624 ncl_dircookie_lock(dnp); 2625 NFSUNLOCKNODE(dnp); 2626 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2627 *cookiep = cookie; 2628 ncl_dircookie_unlock(dnp); 2629 } 2630 } else if (NFS_ISV4(vp)) { 2631 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2632 } 2633 return (error); 2634 } 2635 2636 /* 2637 * Silly rename. To make the NFS filesystem that is stateless look a little 2638 * more like the "ufs" a remove of an active vnode is translated to a rename 2639 * to a funny looking filename that is removed by nfs_inactive on the 2640 * nfsnode. There is the potential for another process on a different client 2641 * to create the same funny name between the nfs_lookitup() fails and the 2642 * nfs_rename() completes, but... 2643 */ 2644 static int 2645 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2646 { 2647 struct sillyrename *sp; 2648 struct nfsnode *np; 2649 int error; 2650 short pid; 2651 unsigned int lticks; 2652 2653 cache_purge(dvp); 2654 np = VTONFS(vp); 2655 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2656 sp = malloc(sizeof (struct sillyrename), 2657 M_NEWNFSREQ, M_WAITOK); 2658 sp->s_cred = crhold(cnp->cn_cred); 2659 sp->s_dvp = dvp; 2660 VREF(dvp); 2661 2662 /* 2663 * Fudge together a funny name. 2664 * Changing the format of the funny name to accommodate more 2665 * sillynames per directory. 2666 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2667 * CPU ticks since boot. 2668 */ 2669 pid = curthread->td_proc->p_pid; 2670 lticks = (unsigned int)ticks; 2671 for ( ; ; ) { 2672 sp->s_namlen = sprintf(sp->s_name, 2673 ".nfs.%08x.%04x4.4", lticks, 2674 pid); 2675 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2676 curthread, NULL)) 2677 break; 2678 lticks++; 2679 } 2680 error = nfs_renameit(dvp, vp, cnp, sp); 2681 if (error) 2682 goto bad; 2683 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2684 curthread, &np); 2685 np->n_sillyrename = sp; 2686 return (0); 2687 bad: 2688 vrele(sp->s_dvp); 2689 crfree(sp->s_cred); 2690 free(sp, M_NEWNFSREQ); 2691 return (error); 2692 } 2693 2694 /* 2695 * Look up a file name and optionally either update the file handle or 2696 * allocate an nfsnode, depending on the value of npp. 2697 * npp == NULL --> just do the lookup 2698 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2699 * handled too 2700 * *npp != NULL --> update the file handle in the vnode 2701 */ 2702 static int 2703 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2704 struct thread *td, struct nfsnode **npp) 2705 { 2706 struct vnode *newvp = NULL, *vp; 2707 struct nfsnode *np, *dnp = VTONFS(dvp); 2708 struct nfsfh *nfhp, *onfhp; 2709 struct nfsvattr nfsva, dnfsva; 2710 struct componentname cn; 2711 int error = 0, attrflag, dattrflag; 2712 u_int hash; 2713 struct timespec ts; 2714 2715 nanouptime(&ts); 2716 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2717 &nfhp, &attrflag, &dattrflag, NULL, 0); 2718 if (dattrflag) 2719 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2720 if (npp && !error) { 2721 if (*npp != NULL) { 2722 np = *npp; 2723 vp = NFSTOV(np); 2724 /* 2725 * For NFSv4, check to see if it is the same name and 2726 * replace the name, if it is different. 2727 */ 2728 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2729 (np->n_v4->n4_namelen != len || 2730 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2731 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2732 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2733 dnp->n_fhp->nfh_len))) { 2734 #ifdef notdef 2735 { char nnn[100]; int nnnl; 2736 nnnl = (len < 100) ? len : 99; 2737 bcopy(name, nnn, nnnl); 2738 nnn[nnnl] = '\0'; 2739 printf("replace=%s\n",nnn); 2740 } 2741 #endif 2742 free(np->n_v4, M_NFSV4NODE); 2743 np->n_v4 = malloc( 2744 sizeof (struct nfsv4node) + 2745 dnp->n_fhp->nfh_len + len - 1, 2746 M_NFSV4NODE, M_WAITOK); 2747 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2748 np->n_v4->n4_namelen = len; 2749 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2750 dnp->n_fhp->nfh_len); 2751 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2752 } 2753 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2754 FNV1_32_INIT); 2755 onfhp = np->n_fhp; 2756 /* 2757 * Rehash node for new file handle. 2758 */ 2759 vfs_hash_rehash(vp, hash); 2760 np->n_fhp = nfhp; 2761 if (onfhp != NULL) 2762 free(onfhp, M_NFSFH); 2763 newvp = NFSTOV(np); 2764 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2765 free(nfhp, M_NFSFH); 2766 VREF(dvp); 2767 newvp = dvp; 2768 } else { 2769 cn.cn_nameptr = name; 2770 cn.cn_namelen = len; 2771 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2772 &np, LK_EXCLUSIVE); 2773 if (error) 2774 return (error); 2775 newvp = NFSTOV(np); 2776 /* 2777 * If n_localmodtime >= time before RPC, then 2778 * a file modification operation, such as 2779 * VOP_SETATTR() of size, has occurred while 2780 * the Lookup RPC and acquisition of the vnode 2781 * happened. As such, the attributes might 2782 * be stale, with possibly an incorrect size. 2783 */ 2784 NFSLOCKNODE(np); 2785 if (timespecisset(&np->n_localmodtime) && 2786 timespeccmp(&np->n_localmodtime, &ts, >=)) { 2787 NFSCL_DEBUG(4, "nfs_lookitup: localmod " 2788 "stale attributes\n"); 2789 attrflag = 0; 2790 } 2791 NFSUNLOCKNODE(np); 2792 } 2793 if (!attrflag && *npp == NULL) { 2794 if (newvp == dvp) 2795 vrele(newvp); 2796 else 2797 vput(newvp); 2798 return (ENOENT); 2799 } 2800 if (attrflag) 2801 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2802 } 2803 if (npp && *npp == NULL) { 2804 if (error) { 2805 if (newvp) { 2806 if (newvp == dvp) 2807 vrele(newvp); 2808 else 2809 vput(newvp); 2810 } 2811 } else 2812 *npp = np; 2813 } 2814 if (error && NFS_ISV4(dvp)) 2815 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2816 return (error); 2817 } 2818 2819 /* 2820 * Nfs Version 3 and 4 commit rpc 2821 */ 2822 int 2823 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2824 struct thread *td) 2825 { 2826 struct nfsvattr nfsva; 2827 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2828 struct nfsnode *np; 2829 struct uio uio; 2830 int error, attrflag; 2831 2832 np = VTONFS(vp); 2833 error = EIO; 2834 attrflag = 0; 2835 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 2836 uio.uio_offset = offset; 2837 uio.uio_resid = cnt; 2838 error = nfscl_doiods(vp, &uio, NULL, NULL, 2839 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 2840 if (error != 0) { 2841 NFSLOCKNODE(np); 2842 np->n_flag &= ~NDSCOMMIT; 2843 NFSUNLOCKNODE(np); 2844 } 2845 } 2846 if (error != 0) { 2847 mtx_lock(&nmp->nm_mtx); 2848 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2849 mtx_unlock(&nmp->nm_mtx); 2850 return (0); 2851 } 2852 mtx_unlock(&nmp->nm_mtx); 2853 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2854 &attrflag, NULL); 2855 } 2856 if (attrflag != 0) 2857 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2858 if (error != 0 && NFS_ISV4(vp)) 2859 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2860 return (error); 2861 } 2862 2863 /* 2864 * Strategy routine. 2865 * For async requests when nfsiod(s) are running, queue the request by 2866 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2867 * request. 2868 */ 2869 static int 2870 nfs_strategy(struct vop_strategy_args *ap) 2871 { 2872 struct buf *bp; 2873 struct vnode *vp; 2874 struct ucred *cr; 2875 2876 bp = ap->a_bp; 2877 vp = ap->a_vp; 2878 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 2879 KASSERT(!(bp->b_flags & B_DONE), 2880 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2881 2882 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 2883 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 2884 DEV_BSIZE); 2885 if (bp->b_iocmd == BIO_READ) 2886 cr = bp->b_rcred; 2887 else 2888 cr = bp->b_wcred; 2889 2890 /* 2891 * If the op is asynchronous and an i/o daemon is waiting 2892 * queue the request, wake it up and wait for completion 2893 * otherwise just do it ourselves. 2894 */ 2895 if ((bp->b_flags & B_ASYNC) == 0 || 2896 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 2897 (void) ncl_doio(vp, bp, cr, curthread, 1); 2898 return (0); 2899 } 2900 2901 /* 2902 * fsync vnode op. Just call ncl_flush() with commit == 1. 2903 */ 2904 /* ARGSUSED */ 2905 static int 2906 nfs_fsync(struct vop_fsync_args *ap) 2907 { 2908 2909 if (ap->a_vp->v_type != VREG) { 2910 /* 2911 * For NFS, metadata is changed synchronously on the server, 2912 * so there is nothing to flush. Also, ncl_flush() clears 2913 * the NMODIFIED flag and that shouldn't be done here for 2914 * directories. 2915 */ 2916 return (0); 2917 } 2918 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 2919 } 2920 2921 /* 2922 * Flush all the blocks associated with a vnode. 2923 * Walk through the buffer pool and push any dirty pages 2924 * associated with the vnode. 2925 * If the called_from_renewthread argument is TRUE, it has been called 2926 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2927 * waiting for a buffer write to complete. 2928 */ 2929 int 2930 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 2931 int commit, int called_from_renewthread) 2932 { 2933 struct nfsnode *np = VTONFS(vp); 2934 struct buf *bp; 2935 int i; 2936 struct buf *nbp; 2937 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2938 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2939 int passone = 1, trycnt = 0; 2940 u_quad_t off, endoff, toff; 2941 struct ucred* wcred = NULL; 2942 struct buf **bvec = NULL; 2943 struct bufobj *bo; 2944 #ifndef NFS_COMMITBVECSIZ 2945 #define NFS_COMMITBVECSIZ 20 2946 #endif 2947 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2948 u_int bvecsize = 0, bveccount; 2949 struct timespec ts; 2950 2951 if (called_from_renewthread != 0) 2952 slptimeo = hz; 2953 if (nmp->nm_flag & NFSMNT_INT) 2954 slpflag = PCATCH; 2955 if (!commit) 2956 passone = 0; 2957 bo = &vp->v_bufobj; 2958 /* 2959 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2960 * server, but has not been committed to stable storage on the server 2961 * yet. On the first pass, the byte range is worked out and the commit 2962 * rpc is done. On the second pass, ncl_writebp() is called to do the 2963 * job. 2964 */ 2965 again: 2966 off = (u_quad_t)-1; 2967 endoff = 0; 2968 bvecpos = 0; 2969 if (NFS_ISV34(vp) && commit) { 2970 if (bvec != NULL && bvec != bvec_on_stack) 2971 free(bvec, M_TEMP); 2972 /* 2973 * Count up how many buffers waiting for a commit. 2974 */ 2975 bveccount = 0; 2976 BO_LOCK(bo); 2977 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2978 if (!BUF_ISLOCKED(bp) && 2979 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2980 == (B_DELWRI | B_NEEDCOMMIT)) 2981 bveccount++; 2982 } 2983 /* 2984 * Allocate space to remember the list of bufs to commit. It is 2985 * important to use M_NOWAIT here to avoid a race with nfs_write. 2986 * If we can't get memory (for whatever reason), we will end up 2987 * committing the buffers one-by-one in the loop below. 2988 */ 2989 if (bveccount > NFS_COMMITBVECSIZ) { 2990 /* 2991 * Release the vnode interlock to avoid a lock 2992 * order reversal. 2993 */ 2994 BO_UNLOCK(bo); 2995 bvec = (struct buf **) 2996 malloc(bveccount * sizeof(struct buf *), 2997 M_TEMP, M_NOWAIT); 2998 BO_LOCK(bo); 2999 if (bvec == NULL) { 3000 bvec = bvec_on_stack; 3001 bvecsize = NFS_COMMITBVECSIZ; 3002 } else 3003 bvecsize = bveccount; 3004 } else { 3005 bvec = bvec_on_stack; 3006 bvecsize = NFS_COMMITBVECSIZ; 3007 } 3008 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3009 if (bvecpos >= bvecsize) 3010 break; 3011 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3012 nbp = TAILQ_NEXT(bp, b_bobufs); 3013 continue; 3014 } 3015 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3016 (B_DELWRI | B_NEEDCOMMIT)) { 3017 BUF_UNLOCK(bp); 3018 nbp = TAILQ_NEXT(bp, b_bobufs); 3019 continue; 3020 } 3021 BO_UNLOCK(bo); 3022 bremfree(bp); 3023 /* 3024 * Work out if all buffers are using the same cred 3025 * so we can deal with them all with one commit. 3026 * 3027 * NOTE: we are not clearing B_DONE here, so we have 3028 * to do it later on in this routine if we intend to 3029 * initiate I/O on the bp. 3030 * 3031 * Note: to avoid loopback deadlocks, we do not 3032 * assign b_runningbufspace. 3033 */ 3034 if (wcred == NULL) 3035 wcred = bp->b_wcred; 3036 else if (wcred != bp->b_wcred) 3037 wcred = NOCRED; 3038 vfs_busy_pages(bp, 1); 3039 3040 BO_LOCK(bo); 3041 /* 3042 * bp is protected by being locked, but nbp is not 3043 * and vfs_busy_pages() may sleep. We have to 3044 * recalculate nbp. 3045 */ 3046 nbp = TAILQ_NEXT(bp, b_bobufs); 3047 3048 /* 3049 * A list of these buffers is kept so that the 3050 * second loop knows which buffers have actually 3051 * been committed. This is necessary, since there 3052 * may be a race between the commit rpc and new 3053 * uncommitted writes on the file. 3054 */ 3055 bvec[bvecpos++] = bp; 3056 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3057 bp->b_dirtyoff; 3058 if (toff < off) 3059 off = toff; 3060 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3061 if (toff > endoff) 3062 endoff = toff; 3063 } 3064 BO_UNLOCK(bo); 3065 } 3066 if (bvecpos > 0) { 3067 /* 3068 * Commit data on the server, as required. 3069 * If all bufs are using the same wcred, then use that with 3070 * one call for all of them, otherwise commit each one 3071 * separately. 3072 */ 3073 if (wcred != NOCRED) 3074 retv = ncl_commit(vp, off, (int)(endoff - off), 3075 wcred, td); 3076 else { 3077 retv = 0; 3078 for (i = 0; i < bvecpos; i++) { 3079 off_t off, size; 3080 bp = bvec[i]; 3081 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3082 bp->b_dirtyoff; 3083 size = (u_quad_t)(bp->b_dirtyend 3084 - bp->b_dirtyoff); 3085 retv = ncl_commit(vp, off, (int)size, 3086 bp->b_wcred, td); 3087 if (retv) break; 3088 } 3089 } 3090 3091 if (retv == NFSERR_STALEWRITEVERF) 3092 ncl_clearcommit(vp->v_mount); 3093 3094 /* 3095 * Now, either mark the blocks I/O done or mark the 3096 * blocks dirty, depending on whether the commit 3097 * succeeded. 3098 */ 3099 for (i = 0; i < bvecpos; i++) { 3100 bp = bvec[i]; 3101 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3102 if (!NFSCL_FORCEDISM(vp->v_mount) && retv) { 3103 /* 3104 * Error, leave B_DELWRI intact 3105 */ 3106 vfs_unbusy_pages(bp); 3107 brelse(bp); 3108 } else { 3109 /* 3110 * Success, remove B_DELWRI ( bundirty() ). 3111 * 3112 * b_dirtyoff/b_dirtyend seem to be NFS 3113 * specific. We should probably move that 3114 * into bundirty(). XXX 3115 */ 3116 bufobj_wref(bo); 3117 bp->b_flags |= B_ASYNC; 3118 bundirty(bp); 3119 bp->b_flags &= ~B_DONE; 3120 bp->b_ioflags &= ~BIO_ERROR; 3121 bp->b_dirtyoff = bp->b_dirtyend = 0; 3122 bufdone(bp); 3123 } 3124 } 3125 } 3126 3127 /* 3128 * Start/do any write(s) that are required. 3129 */ 3130 loop: 3131 BO_LOCK(bo); 3132 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3133 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3134 if (waitfor != MNT_WAIT || passone) 3135 continue; 3136 3137 error = BUF_TIMELOCK(bp, 3138 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 3139 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 3140 if (error == 0) { 3141 BUF_UNLOCK(bp); 3142 goto loop; 3143 } 3144 if (error == ENOLCK) { 3145 error = 0; 3146 goto loop; 3147 } 3148 if (called_from_renewthread != 0) { 3149 /* 3150 * Return EIO so the flush will be retried 3151 * later. 3152 */ 3153 error = EIO; 3154 goto done; 3155 } 3156 if (newnfs_sigintr(nmp, td)) { 3157 error = EINTR; 3158 goto done; 3159 } 3160 if (slpflag == PCATCH) { 3161 slpflag = 0; 3162 slptimeo = 2 * hz; 3163 } 3164 goto loop; 3165 } 3166 if ((bp->b_flags & B_DELWRI) == 0) 3167 panic("nfs_fsync: not dirty"); 3168 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3169 BUF_UNLOCK(bp); 3170 continue; 3171 } 3172 BO_UNLOCK(bo); 3173 bremfree(bp); 3174 bp->b_flags |= B_ASYNC; 3175 bwrite(bp); 3176 if (newnfs_sigintr(nmp, td)) { 3177 error = EINTR; 3178 goto done; 3179 } 3180 goto loop; 3181 } 3182 if (passone) { 3183 passone = 0; 3184 BO_UNLOCK(bo); 3185 goto again; 3186 } 3187 if (waitfor == MNT_WAIT) { 3188 while (bo->bo_numoutput) { 3189 error = bufobj_wwait(bo, slpflag, slptimeo); 3190 if (error) { 3191 BO_UNLOCK(bo); 3192 if (called_from_renewthread != 0) { 3193 /* 3194 * Return EIO so that the flush will be 3195 * retried later. 3196 */ 3197 error = EIO; 3198 goto done; 3199 } 3200 error = newnfs_sigintr(nmp, td); 3201 if (error) 3202 goto done; 3203 if (slpflag == PCATCH) { 3204 slpflag = 0; 3205 slptimeo = 2 * hz; 3206 } 3207 BO_LOCK(bo); 3208 } 3209 } 3210 if (bo->bo_dirty.bv_cnt != 0 && commit) { 3211 BO_UNLOCK(bo); 3212 goto loop; 3213 } 3214 /* 3215 * Wait for all the async IO requests to drain 3216 */ 3217 BO_UNLOCK(bo); 3218 NFSLOCKNODE(np); 3219 while (np->n_directio_asyncwr > 0) { 3220 np->n_flag |= NFSYNCWAIT; 3221 error = newnfs_msleep(td, &np->n_directio_asyncwr, 3222 &np->n_mtx, slpflag | (PRIBIO + 1), 3223 "nfsfsync", 0); 3224 if (error) { 3225 if (newnfs_sigintr(nmp, td)) { 3226 NFSUNLOCKNODE(np); 3227 error = EINTR; 3228 goto done; 3229 } 3230 } 3231 } 3232 NFSUNLOCKNODE(np); 3233 } else 3234 BO_UNLOCK(bo); 3235 if (NFSHASPNFS(nmp)) { 3236 nfscl_layoutcommit(vp, td); 3237 /* 3238 * Invalidate the attribute cache, since writes to a DS 3239 * won't update the size attribute. 3240 */ 3241 NFSLOCKNODE(np); 3242 np->n_attrstamp = 0; 3243 } else 3244 NFSLOCKNODE(np); 3245 if (np->n_flag & NWRITEERR) { 3246 error = np->n_error; 3247 np->n_flag &= ~NWRITEERR; 3248 } 3249 if (commit && bo->bo_dirty.bv_cnt == 0 && 3250 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 3251 np->n_flag &= ~NMODIFIED; 3252 NFSUNLOCKNODE(np); 3253 done: 3254 if (bvec != NULL && bvec != bvec_on_stack) 3255 free(bvec, M_TEMP); 3256 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 3257 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 3258 np->n_directio_asyncwr != 0)) { 3259 if (trycnt++ < 5) { 3260 /* try, try again... */ 3261 passone = 1; 3262 wcred = NULL; 3263 bvec = NULL; 3264 bvecsize = 0; 3265 goto again; 3266 } 3267 vn_printf(vp, "ncl_flush failed"); 3268 error = called_from_renewthread != 0 ? EIO : EBUSY; 3269 } 3270 if (error == 0) { 3271 nanouptime(&ts); 3272 NFSLOCKNODE(np); 3273 np->n_localmodtime = ts; 3274 NFSUNLOCKNODE(np); 3275 } 3276 return (error); 3277 } 3278 3279 /* 3280 * NFS advisory byte-level locks. 3281 */ 3282 static int 3283 nfs_advlock(struct vop_advlock_args *ap) 3284 { 3285 struct vnode *vp = ap->a_vp; 3286 struct ucred *cred; 3287 struct nfsnode *np = VTONFS(ap->a_vp); 3288 struct proc *p = (struct proc *)ap->a_id; 3289 struct thread *td = curthread; /* XXX */ 3290 struct vattr va; 3291 int ret, error; 3292 u_quad_t size; 3293 struct nfsmount *nmp; 3294 3295 error = NFSVOPLOCK(vp, LK_SHARED); 3296 if (error != 0) 3297 return (EBADF); 3298 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3299 if (vp->v_type != VREG) { 3300 error = EINVAL; 3301 goto out; 3302 } 3303 if ((ap->a_flags & F_POSIX) != 0) 3304 cred = p->p_ucred; 3305 else 3306 cred = td->td_ucred; 3307 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 3308 if (VN_IS_DOOMED(vp)) { 3309 error = EBADF; 3310 goto out; 3311 } 3312 3313 /* 3314 * If this is unlocking a write locked region, flush and 3315 * commit them before unlocking. This is required by 3316 * RFC3530 Sec. 9.3.2. 3317 */ 3318 if (ap->a_op == F_UNLCK && 3319 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3320 ap->a_flags)) 3321 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3322 3323 /* 3324 * Mark NFS node as might have acquired a lock. 3325 * This is separate from NHASBEENLOCKED, because it must 3326 * be done before the nfsrpc_advlock() call, which might 3327 * add a nfscllock structure to the client state. 3328 * It is used to check for the case where a nfscllock 3329 * state structure cannot exist for the file. 3330 * Only done for "oneopenown" NFSv4.1/4.2 mounts. 3331 */ 3332 nmp = VFSTONFS(vp->v_mount); 3333 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) { 3334 NFSLOCKNODE(np); 3335 np->n_flag |= NMIGHTBELOCKED; 3336 NFSUNLOCKNODE(np); 3337 } 3338 3339 /* 3340 * Loop around doing the lock op, while a blocking lock 3341 * must wait for the lock op to succeed. 3342 */ 3343 do { 3344 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3345 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3346 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3347 ap->a_op == F_SETLK) { 3348 NFSVOPUNLOCK(vp); 3349 error = nfs_catnap(PZERO | PCATCH, ret, 3350 "ncladvl"); 3351 if (error) 3352 return (EINTR); 3353 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3354 if (VN_IS_DOOMED(vp)) { 3355 error = EBADF; 3356 goto out; 3357 } 3358 } 3359 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3360 ap->a_op == F_SETLK); 3361 if (ret == NFSERR_DENIED) { 3362 error = EAGAIN; 3363 goto out; 3364 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3365 error = ret; 3366 goto out; 3367 } else if (ret != 0) { 3368 error = EACCES; 3369 goto out; 3370 } 3371 3372 /* 3373 * Now, if we just got a lock, invalidate data in the buffer 3374 * cache, as required, so that the coherency conforms with 3375 * RFC3530 Sec. 9.3.2. 3376 */ 3377 if (ap->a_op == F_SETLK) { 3378 if ((np->n_flag & NMODIFIED) == 0) { 3379 np->n_attrstamp = 0; 3380 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3381 ret = VOP_GETATTR(vp, &va, cred); 3382 } 3383 if ((np->n_flag & NMODIFIED) || ret || 3384 np->n_change != va.va_filerev) { 3385 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3386 np->n_attrstamp = 0; 3387 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3388 ret = VOP_GETATTR(vp, &va, cred); 3389 if (!ret) { 3390 np->n_mtime = va.va_mtime; 3391 np->n_change = va.va_filerev; 3392 } 3393 } 3394 /* Mark that a file lock has been acquired. */ 3395 NFSLOCKNODE(np); 3396 np->n_flag |= NHASBEENLOCKED; 3397 NFSUNLOCKNODE(np); 3398 } 3399 } else if (!NFS_ISV4(vp)) { 3400 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3401 size = VTONFS(vp)->n_size; 3402 NFSVOPUNLOCK(vp); 3403 error = lf_advlock(ap, &(vp->v_lockf), size); 3404 } else { 3405 if (nfs_advlock_p != NULL) 3406 error = nfs_advlock_p(ap); 3407 else { 3408 NFSVOPUNLOCK(vp); 3409 error = ENOLCK; 3410 } 3411 } 3412 if (error == 0 && ap->a_op == F_SETLK) { 3413 error = NFSVOPLOCK(vp, LK_SHARED); 3414 if (error == 0) { 3415 /* Mark that a file lock has been acquired. */ 3416 NFSLOCKNODE(np); 3417 np->n_flag |= NHASBEENLOCKED; 3418 NFSUNLOCKNODE(np); 3419 NFSVOPUNLOCK(vp); 3420 } 3421 } 3422 return (error); 3423 } else 3424 error = EOPNOTSUPP; 3425 out: 3426 NFSVOPUNLOCK(vp); 3427 return (error); 3428 } 3429 3430 /* 3431 * NFS advisory byte-level locks. 3432 */ 3433 static int 3434 nfs_advlockasync(struct vop_advlockasync_args *ap) 3435 { 3436 struct vnode *vp = ap->a_vp; 3437 u_quad_t size; 3438 int error; 3439 3440 if (NFS_ISV4(vp)) 3441 return (EOPNOTSUPP); 3442 error = NFSVOPLOCK(vp, LK_SHARED); 3443 if (error) 3444 return (error); 3445 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3446 size = VTONFS(vp)->n_size; 3447 NFSVOPUNLOCK(vp); 3448 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3449 } else { 3450 NFSVOPUNLOCK(vp); 3451 error = EOPNOTSUPP; 3452 } 3453 return (error); 3454 } 3455 3456 /* 3457 * Print out the contents of an nfsnode. 3458 */ 3459 static int 3460 nfs_print(struct vop_print_args *ap) 3461 { 3462 struct vnode *vp = ap->a_vp; 3463 struct nfsnode *np = VTONFS(vp); 3464 3465 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3466 (uintmax_t)np->n_vattr.na_fsid); 3467 if (vp->v_type == VFIFO) 3468 fifo_printinfo(vp); 3469 printf("\n"); 3470 return (0); 3471 } 3472 3473 /* 3474 * This is the "real" nfs::bwrite(struct buf*). 3475 * We set B_CACHE if this is a VMIO buffer. 3476 */ 3477 int 3478 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3479 { 3480 int oldflags, rtval; 3481 3482 if (bp->b_flags & B_INVAL) { 3483 brelse(bp); 3484 return (0); 3485 } 3486 3487 oldflags = bp->b_flags; 3488 bp->b_flags |= B_CACHE; 3489 3490 /* 3491 * Undirty the bp. We will redirty it later if the I/O fails. 3492 */ 3493 bundirty(bp); 3494 bp->b_flags &= ~B_DONE; 3495 bp->b_ioflags &= ~BIO_ERROR; 3496 bp->b_iocmd = BIO_WRITE; 3497 3498 bufobj_wref(bp->b_bufobj); 3499 curthread->td_ru.ru_oublock++; 3500 3501 /* 3502 * Note: to avoid loopback deadlocks, we do not 3503 * assign b_runningbufspace. 3504 */ 3505 vfs_busy_pages(bp, 1); 3506 3507 BUF_KERNPROC(bp); 3508 bp->b_iooffset = dbtob(bp->b_blkno); 3509 bstrategy(bp); 3510 3511 if ((oldflags & B_ASYNC) != 0) 3512 return (0); 3513 3514 rtval = bufwait(bp); 3515 if (oldflags & B_DELWRI) 3516 reassignbuf(bp); 3517 brelse(bp); 3518 return (rtval); 3519 } 3520 3521 /* 3522 * nfs special file access vnode op. 3523 * Essentially just get vattr and then imitate iaccess() since the device is 3524 * local to the client. 3525 */ 3526 static int 3527 nfsspec_access(struct vop_access_args *ap) 3528 { 3529 struct vattr *vap; 3530 struct ucred *cred = ap->a_cred; 3531 struct vnode *vp = ap->a_vp; 3532 accmode_t accmode = ap->a_accmode; 3533 struct vattr vattr; 3534 int error; 3535 3536 /* 3537 * Disallow write attempts on filesystems mounted read-only; 3538 * unless the file is a socket, fifo, or a block or character 3539 * device resident on the filesystem. 3540 */ 3541 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3542 switch (vp->v_type) { 3543 case VREG: 3544 case VDIR: 3545 case VLNK: 3546 return (EROFS); 3547 default: 3548 break; 3549 } 3550 } 3551 vap = &vattr; 3552 error = VOP_GETATTR(vp, vap, cred); 3553 if (error) 3554 goto out; 3555 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3556 accmode, cred); 3557 out: 3558 return error; 3559 } 3560 3561 /* 3562 * Read wrapper for fifos. 3563 */ 3564 static int 3565 nfsfifo_read(struct vop_read_args *ap) 3566 { 3567 struct nfsnode *np = VTONFS(ap->a_vp); 3568 int error; 3569 3570 /* 3571 * Set access flag. 3572 */ 3573 NFSLOCKNODE(np); 3574 np->n_flag |= NACC; 3575 vfs_timestamp(&np->n_atim); 3576 NFSUNLOCKNODE(np); 3577 error = fifo_specops.vop_read(ap); 3578 return error; 3579 } 3580 3581 /* 3582 * Write wrapper for fifos. 3583 */ 3584 static int 3585 nfsfifo_write(struct vop_write_args *ap) 3586 { 3587 struct nfsnode *np = VTONFS(ap->a_vp); 3588 3589 /* 3590 * Set update flag. 3591 */ 3592 NFSLOCKNODE(np); 3593 np->n_flag |= NUPD; 3594 vfs_timestamp(&np->n_mtim); 3595 NFSUNLOCKNODE(np); 3596 return(fifo_specops.vop_write(ap)); 3597 } 3598 3599 /* 3600 * Close wrapper for fifos. 3601 * 3602 * Update the times on the nfsnode then do fifo close. 3603 */ 3604 static int 3605 nfsfifo_close(struct vop_close_args *ap) 3606 { 3607 struct vnode *vp = ap->a_vp; 3608 struct nfsnode *np = VTONFS(vp); 3609 struct vattr vattr; 3610 struct timespec ts; 3611 3612 NFSLOCKNODE(np); 3613 if (np->n_flag & (NACC | NUPD)) { 3614 vfs_timestamp(&ts); 3615 if (np->n_flag & NACC) 3616 np->n_atim = ts; 3617 if (np->n_flag & NUPD) 3618 np->n_mtim = ts; 3619 np->n_flag |= NCHG; 3620 if (vrefcnt(vp) == 1 && 3621 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3622 VATTR_NULL(&vattr); 3623 if (np->n_flag & NACC) 3624 vattr.va_atime = np->n_atim; 3625 if (np->n_flag & NUPD) 3626 vattr.va_mtime = np->n_mtim; 3627 NFSUNLOCKNODE(np); 3628 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3629 goto out; 3630 } 3631 } 3632 NFSUNLOCKNODE(np); 3633 out: 3634 return (fifo_specops.vop_close(ap)); 3635 } 3636 3637 /* 3638 * Just call ncl_writebp() with the force argument set to 1. 3639 * 3640 * NOTE: B_DONE may or may not be set in a_bp on call. 3641 */ 3642 static int 3643 nfs_bwrite(struct buf *bp) 3644 { 3645 3646 return (ncl_writebp(bp, 1, curthread)); 3647 } 3648 3649 struct buf_ops buf_ops_newnfs = { 3650 .bop_name = "buf_ops_nfs", 3651 .bop_write = nfs_bwrite, 3652 .bop_strategy = bufstrategy, 3653 .bop_sync = bufsync, 3654 .bop_bdflush = bufbdflush, 3655 }; 3656 3657 static int 3658 nfs_getacl(struct vop_getacl_args *ap) 3659 { 3660 int error; 3661 3662 if (ap->a_type != ACL_TYPE_NFS4) 3663 return (EOPNOTSUPP); 3664 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3665 NULL); 3666 if (error > NFSERR_STALE) { 3667 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3668 error = EPERM; 3669 } 3670 return (error); 3671 } 3672 3673 static int 3674 nfs_setacl(struct vop_setacl_args *ap) 3675 { 3676 int error; 3677 3678 if (ap->a_type != ACL_TYPE_NFS4) 3679 return (EOPNOTSUPP); 3680 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3681 NULL); 3682 if (error > NFSERR_STALE) { 3683 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3684 error = EPERM; 3685 } 3686 return (error); 3687 } 3688 3689 /* 3690 * VOP_ADVISE for NFS. 3691 * Just return 0 for any errors, since it is just a hint. 3692 */ 3693 static int 3694 nfs_advise(struct vop_advise_args *ap) 3695 { 3696 struct thread *td = curthread; 3697 struct nfsmount *nmp; 3698 uint64_t len; 3699 int error; 3700 3701 /* 3702 * First do vop_stdadvise() to handle the buffer cache. 3703 */ 3704 error = vop_stdadvise(ap); 3705 if (error != 0) 3706 return (error); 3707 if (ap->a_start < 0 || ap->a_end < 0) 3708 return (0); 3709 if (ap->a_end == OFF_MAX) 3710 len = 0; 3711 else if (ap->a_end < ap->a_start) 3712 return (0); 3713 else 3714 len = ap->a_end - ap->a_start + 1; 3715 nmp = VFSTONFS(ap->a_vp->v_mount); 3716 mtx_lock(&nmp->nm_mtx); 3717 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3718 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) == 3719 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) { 3720 mtx_unlock(&nmp->nm_mtx); 3721 return (0); 3722 } 3723 mtx_unlock(&nmp->nm_mtx); 3724 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice, 3725 td->td_ucred, td); 3726 if (error == NFSERR_NOTSUPP) { 3727 mtx_lock(&nmp->nm_mtx); 3728 nmp->nm_privflag |= NFSMNTP_NOADVISE; 3729 mtx_unlock(&nmp->nm_mtx); 3730 } 3731 return (0); 3732 } 3733 3734 /* 3735 * nfs allocate call 3736 */ 3737 static int 3738 nfs_allocate(struct vop_allocate_args *ap) 3739 { 3740 struct vnode *vp = ap->a_vp; 3741 struct thread *td = curthread; 3742 struct nfsvattr nfsva; 3743 struct nfsmount *nmp; 3744 struct nfsnode *np; 3745 off_t alen; 3746 int attrflag, error, ret; 3747 struct timespec ts; 3748 struct uio io; 3749 3750 attrflag = 0; 3751 nmp = VFSTONFS(vp->v_mount); 3752 np = VTONFS(vp); 3753 mtx_lock(&nmp->nm_mtx); 3754 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3755 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) { 3756 mtx_unlock(&nmp->nm_mtx); 3757 alen = *ap->a_len; 3758 if ((uint64_t)alen > nfs_maxalloclen) 3759 alen = nfs_maxalloclen; 3760 3761 /* Check the file size limit. */ 3762 io.uio_offset = *ap->a_offset; 3763 io.uio_resid = alen; 3764 error = vn_rlimit_fsize(vp, &io, td); 3765 3766 /* 3767 * Flush first to ensure that the allocate adds to the 3768 * file's allocation on the server. 3769 */ 3770 if (error == 0) 3771 error = ncl_flush(vp, MNT_WAIT, td, 1, 0); 3772 if (error == 0) 3773 error = nfsrpc_allocate(vp, *ap->a_offset, alen, 3774 &nfsva, &attrflag, ap->a_cred, td, NULL); 3775 if (error == 0) { 3776 *ap->a_offset += alen; 3777 *ap->a_len -= alen; 3778 nanouptime(&ts); 3779 NFSLOCKNODE(np); 3780 np->n_localmodtime = ts; 3781 NFSUNLOCKNODE(np); 3782 } else if (error == NFSERR_NOTSUPP) { 3783 mtx_lock(&nmp->nm_mtx); 3784 nmp->nm_privflag |= NFSMNTP_NOALLOCATE; 3785 mtx_unlock(&nmp->nm_mtx); 3786 error = EINVAL; 3787 } 3788 } else { 3789 mtx_unlock(&nmp->nm_mtx); 3790 error = EINVAL; 3791 } 3792 if (attrflag != 0) { 3793 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3794 if (error == 0 && ret != 0) 3795 error = ret; 3796 } 3797 if (error != 0) 3798 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3799 return (error); 3800 } 3801 3802 /* 3803 * nfs deallocate call 3804 */ 3805 static int 3806 nfs_deallocate(struct vop_deallocate_args *ap) 3807 { 3808 struct vnode *vp = ap->a_vp; 3809 struct thread *td = curthread; 3810 struct nfsvattr nfsva; 3811 struct nfsmount *nmp; 3812 struct nfsnode *np; 3813 off_t tlen, mlen; 3814 int attrflag, error, ret; 3815 bool clipped; 3816 struct timespec ts; 3817 3818 error = 0; 3819 attrflag = 0; 3820 nmp = VFSTONFS(vp->v_mount); 3821 np = VTONFS(vp); 3822 mtx_lock(&nmp->nm_mtx); 3823 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3824 (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) { 3825 mtx_unlock(&nmp->nm_mtx); 3826 tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len); 3827 NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n", 3828 (intmax_t)*ap->a_offset, (intmax_t)tlen, 3829 (uintmax_t)nmp->nm_maxfilesize); 3830 if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) { 3831 /* Avoid EFBIG error return from the NFSv4.2 server. */ 3832 *ap->a_len = 0; 3833 return (0); 3834 } 3835 clipped = false; 3836 if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize) 3837 tlen = nmp->nm_maxfilesize - *ap->a_offset; 3838 if ((uint64_t)*ap->a_offset < np->n_size) { 3839 /* Limit the len to nfs_maxalloclen before EOF. */ 3840 mlen = omin((off_t)np->n_size - *ap->a_offset, tlen); 3841 if ((uint64_t)mlen > nfs_maxalloclen) { 3842 NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n"); 3843 tlen = nfs_maxalloclen; 3844 clipped = true; 3845 } 3846 } 3847 if (error == 0) 3848 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 3849 if (error == 0) { 3850 vnode_pager_purge_range(vp, *ap->a_offset, 3851 *ap->a_offset + tlen); 3852 error = nfsrpc_deallocate(vp, *ap->a_offset, tlen, 3853 &nfsva, &attrflag, ap->a_cred, td, NULL); 3854 NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error); 3855 } 3856 if (error == 0) { 3857 NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n", 3858 attrflag, (uintmax_t)nfsva.na_size); 3859 nanouptime(&ts); 3860 NFSLOCKNODE(np); 3861 np->n_localmodtime = ts; 3862 NFSUNLOCKNODE(np); 3863 if (attrflag != 0) { 3864 if ((uint64_t)*ap->a_offset < nfsva.na_size) 3865 *ap->a_offset += omin((off_t) 3866 nfsva.na_size - *ap->a_offset, 3867 tlen); 3868 } 3869 if (clipped && tlen < *ap->a_len) 3870 *ap->a_len -= tlen; 3871 else 3872 *ap->a_len = 0; 3873 } else if (error == NFSERR_NOTSUPP) { 3874 mtx_lock(&nmp->nm_mtx); 3875 nmp->nm_privflag |= NFSMNTP_NODEALLOCATE; 3876 mtx_unlock(&nmp->nm_mtx); 3877 } 3878 } else { 3879 mtx_unlock(&nmp->nm_mtx); 3880 error = EIO; 3881 } 3882 /* 3883 * If the NFS server cannot perform the Deallocate operation, just call 3884 * vop_stddeallocate() to perform it. 3885 */ 3886 if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) { 3887 error = vop_stddeallocate(ap); 3888 NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error); 3889 } 3890 if (attrflag != 0) { 3891 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3892 if (error == 0 && ret != 0) 3893 error = ret; 3894 } 3895 if (error != 0) 3896 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3897 return (error); 3898 } 3899 3900 /* 3901 * nfs copy_file_range call 3902 */ 3903 static int 3904 nfs_copy_file_range(struct vop_copy_file_range_args *ap) 3905 { 3906 struct vnode *invp = ap->a_invp; 3907 struct vnode *outvp = ap->a_outvp; 3908 struct mount *mp; 3909 struct nfsvattr innfsva, outnfsva; 3910 struct vattr *vap; 3911 struct uio io; 3912 struct nfsmount *nmp; 3913 size_t len, len2; 3914 int error, inattrflag, outattrflag, ret, ret2; 3915 off_t inoff, outoff; 3916 bool consecutive, must_commit, tryoutcred; 3917 3918 ret = ret2 = 0; 3919 nmp = VFSTONFS(invp->v_mount); 3920 mtx_lock(&nmp->nm_mtx); 3921 /* NFSv4.2 Copy is not permitted for infile == outfile. */ 3922 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3923 (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) { 3924 mtx_unlock(&nmp->nm_mtx); 3925 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 3926 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 3927 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 3928 return (error); 3929 } 3930 mtx_unlock(&nmp->nm_mtx); 3931 3932 /* Lock both vnodes, avoiding risk of deadlock. */ 3933 do { 3934 mp = NULL; 3935 error = vn_start_write(outvp, &mp, V_WAIT); 3936 if (error == 0) { 3937 error = vn_lock(outvp, LK_EXCLUSIVE); 3938 if (error == 0) { 3939 error = vn_lock(invp, LK_SHARED | LK_NOWAIT); 3940 if (error == 0) 3941 break; 3942 VOP_UNLOCK(outvp); 3943 if (mp != NULL) 3944 vn_finished_write(mp); 3945 mp = NULL; 3946 error = vn_lock(invp, LK_SHARED); 3947 if (error == 0) 3948 VOP_UNLOCK(invp); 3949 } 3950 } 3951 if (mp != NULL) 3952 vn_finished_write(mp); 3953 } while (error == 0); 3954 if (error != 0) 3955 return (error); 3956 3957 /* 3958 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer? 3959 */ 3960 io.uio_offset = *ap->a_outoffp; 3961 io.uio_resid = *ap->a_lenp; 3962 error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd); 3963 3964 /* 3965 * Flush the input file so that the data is up to date before 3966 * the copy. Flush writes for the output file so that they 3967 * do not overwrite the data copied to the output file by the Copy. 3968 * Set the commit argument for both flushes so that the data is on 3969 * stable storage before the Copy RPC. This is done in case the 3970 * server reboots during the Copy and needs to be redone. 3971 */ 3972 if (error == 0) 3973 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0); 3974 if (error == 0) 3975 error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0); 3976 3977 /* Do the actual NFSv4.2 RPC. */ 3978 len = *ap->a_lenp; 3979 mtx_lock(&nmp->nm_mtx); 3980 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0) 3981 consecutive = true; 3982 else 3983 consecutive = false; 3984 mtx_unlock(&nmp->nm_mtx); 3985 inoff = *ap->a_inoffp; 3986 outoff = *ap->a_outoffp; 3987 tryoutcred = true; 3988 must_commit = false; 3989 if (error == 0) { 3990 vap = &VTONFS(invp)->n_vattr.na_vattr; 3991 error = VOP_GETATTR(invp, vap, ap->a_incred); 3992 if (error == 0) { 3993 /* 3994 * Clip "len" at va_size so that RFC compliant servers 3995 * will not reply NFSERR_INVAL. 3996 * Setting "len == 0" for the RPC would be preferred, 3997 * but some Linux servers do not support that. 3998 */ 3999 if (inoff >= vap->va_size) 4000 *ap->a_lenp = len = 0; 4001 else if (inoff + len > vap->va_size) 4002 *ap->a_lenp = len = vap->va_size - inoff; 4003 } else 4004 error = 0; 4005 } 4006 4007 /* 4008 * len will be set to 0 upon a successful Copy RPC. 4009 * As such, this only loops when the Copy RPC needs to be retried. 4010 */ 4011 while (len > 0 && error == 0) { 4012 inattrflag = outattrflag = 0; 4013 len2 = len; 4014 if (tryoutcred) 4015 error = nfsrpc_copy_file_range(invp, ap->a_inoffp, 4016 outvp, ap->a_outoffp, &len2, ap->a_flags, 4017 &inattrflag, &innfsva, &outattrflag, &outnfsva, 4018 ap->a_outcred, consecutive, &must_commit); 4019 else 4020 error = nfsrpc_copy_file_range(invp, ap->a_inoffp, 4021 outvp, ap->a_outoffp, &len2, ap->a_flags, 4022 &inattrflag, &innfsva, &outattrflag, &outnfsva, 4023 ap->a_incred, consecutive, &must_commit); 4024 if (inattrflag != 0) 4025 ret = nfscl_loadattrcache(&invp, &innfsva, NULL, 0, 1); 4026 if (outattrflag != 0) 4027 ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL, 4028 1, 1); 4029 if (error == 0) { 4030 if (consecutive == false) { 4031 if (len2 == len) { 4032 mtx_lock(&nmp->nm_mtx); 4033 nmp->nm_privflag |= 4034 NFSMNTP_NOCONSECUTIVE; 4035 mtx_unlock(&nmp->nm_mtx); 4036 } else 4037 error = NFSERR_OFFLOADNOREQS; 4038 } 4039 *ap->a_lenp = len2; 4040 len = 0; 4041 if (len2 > 0 && must_commit && error == 0) 4042 error = ncl_commit(outvp, outoff, *ap->a_lenp, 4043 ap->a_outcred, curthread); 4044 if (error == 0 && ret != 0) 4045 error = ret; 4046 if (error == 0 && ret2 != 0) 4047 error = ret2; 4048 } else if (error == NFSERR_OFFLOADNOREQS && consecutive) { 4049 /* 4050 * Try consecutive == false, which is ok only if all 4051 * bytes are copied. 4052 * If only some bytes were copied when consecutive 4053 * is false, there is no way to know which bytes 4054 * still need to be written. 4055 */ 4056 consecutive = false; 4057 error = 0; 4058 } else if (error == NFSERR_ACCES && tryoutcred) { 4059 /* Try again with incred. */ 4060 tryoutcred = false; 4061 error = 0; 4062 } 4063 if (error == NFSERR_STALEWRITEVERF) { 4064 /* 4065 * Server rebooted, so do it all again. 4066 */ 4067 *ap->a_inoffp = inoff; 4068 *ap->a_outoffp = outoff; 4069 len = *ap->a_lenp; 4070 must_commit = false; 4071 error = 0; 4072 } 4073 } 4074 VOP_UNLOCK(invp); 4075 VOP_UNLOCK(outvp); 4076 if (mp != NULL) 4077 vn_finished_write(mp); 4078 if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS || 4079 error == NFSERR_ACCES) { 4080 /* 4081 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can 4082 * use a_incred for the read and a_outcred for the write, so 4083 * try this for NFSERR_ACCES failures for the Copy. 4084 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can 4085 * never succeed, so disable it. 4086 */ 4087 if (error != NFSERR_ACCES) { 4088 /* Can never do Copy on this mount. */ 4089 mtx_lock(&nmp->nm_mtx); 4090 nmp->nm_privflag |= NFSMNTP_NOCOPY; 4091 mtx_unlock(&nmp->nm_mtx); 4092 } 4093 *ap->a_inoffp = inoff; 4094 *ap->a_outoffp = outoff; 4095 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 4096 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 4097 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 4098 } else if (error != 0) 4099 *ap->a_lenp = 0; 4100 4101 if (error != 0) 4102 error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0); 4103 return (error); 4104 } 4105 4106 /* 4107 * nfs ioctl call 4108 */ 4109 static int 4110 nfs_ioctl(struct vop_ioctl_args *ap) 4111 { 4112 struct vnode *vp = ap->a_vp; 4113 struct nfsvattr nfsva; 4114 struct nfsmount *nmp; 4115 int attrflag, content, error, ret; 4116 bool eof = false; /* shut up compiler. */ 4117 4118 if (vp->v_type != VREG) 4119 return (ENOTTY); 4120 nmp = VFSTONFS(vp->v_mount); 4121 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) { 4122 error = vop_stdioctl(ap); 4123 return (error); 4124 } 4125 4126 /* Do the actual NFSv4.2 RPC. */ 4127 switch (ap->a_command) { 4128 case FIOSEEKDATA: 4129 content = NFSV4CONTENT_DATA; 4130 break; 4131 case FIOSEEKHOLE: 4132 content = NFSV4CONTENT_HOLE; 4133 break; 4134 default: 4135 return (ENOTTY); 4136 } 4137 4138 error = vn_lock(vp, LK_SHARED); 4139 if (error != 0) 4140 return (EBADF); 4141 attrflag = 0; 4142 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size) 4143 error = ENXIO; 4144 else { 4145 /* 4146 * Flush all writes, so that the server is up to date. 4147 * Although a Commit is not required, the commit argument 4148 * is set so that, for a pNFS File/Flexible File Layout 4149 * server, the LayoutCommit will be done to ensure the file 4150 * size is up to date on the Metadata Server. 4151 */ 4152 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0); 4153 if (error == 0) 4154 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof, 4155 content, ap->a_cred, &nfsva, &attrflag); 4156 /* If at eof for FIOSEEKDATA, return ENXIO. */ 4157 if (eof && error == 0 && content == NFSV4CONTENT_DATA) 4158 error = ENXIO; 4159 } 4160 if (attrflag != 0) { 4161 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4162 if (error == 0 && ret != 0) 4163 error = ret; 4164 } 4165 NFSVOPUNLOCK(vp); 4166 4167 if (error != 0) 4168 error = ENXIO; 4169 return (error); 4170 } 4171 4172 /* 4173 * nfs getextattr call 4174 */ 4175 static int 4176 nfs_getextattr(struct vop_getextattr_args *ap) 4177 { 4178 struct vnode *vp = ap->a_vp; 4179 struct nfsmount *nmp; 4180 struct ucred *cred; 4181 struct thread *td = ap->a_td; 4182 struct nfsvattr nfsva; 4183 ssize_t len; 4184 int attrflag, error, ret; 4185 4186 nmp = VFSTONFS(vp->v_mount); 4187 mtx_lock(&nmp->nm_mtx); 4188 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4189 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4190 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4191 mtx_unlock(&nmp->nm_mtx); 4192 return (EOPNOTSUPP); 4193 } 4194 mtx_unlock(&nmp->nm_mtx); 4195 4196 cred = ap->a_cred; 4197 if (cred == NULL) 4198 cred = td->td_ucred; 4199 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4200 attrflag = 0; 4201 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva, 4202 &attrflag, cred, td); 4203 if (attrflag != 0) { 4204 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4205 if (error == 0 && ret != 0) 4206 error = ret; 4207 } 4208 if (error == 0 && ap->a_size != NULL) 4209 *ap->a_size = len; 4210 4211 switch (error) { 4212 case NFSERR_NOTSUPP: 4213 case NFSERR_OPILLEGAL: 4214 mtx_lock(&nmp->nm_mtx); 4215 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4216 mtx_unlock(&nmp->nm_mtx); 4217 error = EOPNOTSUPP; 4218 break; 4219 case NFSERR_NOXATTR: 4220 case NFSERR_XATTR2BIG: 4221 error = ENOATTR; 4222 break; 4223 default: 4224 error = nfscl_maperr(td, error, 0, 0); 4225 break; 4226 } 4227 return (error); 4228 } 4229 4230 /* 4231 * nfs setextattr call 4232 */ 4233 static int 4234 nfs_setextattr(struct vop_setextattr_args *ap) 4235 { 4236 struct vnode *vp = ap->a_vp; 4237 struct nfsmount *nmp; 4238 struct ucred *cred; 4239 struct thread *td = ap->a_td; 4240 struct nfsvattr nfsva; 4241 int attrflag, error, ret; 4242 4243 nmp = VFSTONFS(vp->v_mount); 4244 mtx_lock(&nmp->nm_mtx); 4245 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4246 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4247 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4248 mtx_unlock(&nmp->nm_mtx); 4249 return (EOPNOTSUPP); 4250 } 4251 mtx_unlock(&nmp->nm_mtx); 4252 4253 if (ap->a_uio->uio_resid < 0) 4254 return (EINVAL); 4255 cred = ap->a_cred; 4256 if (cred == NULL) 4257 cred = td->td_ucred; 4258 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4259 attrflag = 0; 4260 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva, 4261 &attrflag, cred, td); 4262 if (attrflag != 0) { 4263 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4264 if (error == 0 && ret != 0) 4265 error = ret; 4266 } 4267 4268 switch (error) { 4269 case NFSERR_NOTSUPP: 4270 case NFSERR_OPILLEGAL: 4271 mtx_lock(&nmp->nm_mtx); 4272 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4273 mtx_unlock(&nmp->nm_mtx); 4274 error = EOPNOTSUPP; 4275 break; 4276 case NFSERR_NOXATTR: 4277 case NFSERR_XATTR2BIG: 4278 error = ENOATTR; 4279 break; 4280 default: 4281 error = nfscl_maperr(td, error, 0, 0); 4282 break; 4283 } 4284 return (error); 4285 } 4286 4287 /* 4288 * nfs listextattr call 4289 */ 4290 static int 4291 nfs_listextattr(struct vop_listextattr_args *ap) 4292 { 4293 struct vnode *vp = ap->a_vp; 4294 struct nfsmount *nmp; 4295 struct ucred *cred; 4296 struct thread *td = ap->a_td; 4297 struct nfsvattr nfsva; 4298 size_t len, len2; 4299 uint64_t cookie; 4300 int attrflag, error, ret; 4301 bool eof; 4302 4303 nmp = VFSTONFS(vp->v_mount); 4304 mtx_lock(&nmp->nm_mtx); 4305 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4306 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4307 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4308 mtx_unlock(&nmp->nm_mtx); 4309 return (EOPNOTSUPP); 4310 } 4311 mtx_unlock(&nmp->nm_mtx); 4312 4313 cred = ap->a_cred; 4314 if (cred == NULL) 4315 cred = td->td_ucred; 4316 4317 /* Loop around doing List Extended Attribute RPCs. */ 4318 eof = false; 4319 cookie = 0; 4320 len2 = 0; 4321 error = 0; 4322 while (!eof && error == 0) { 4323 len = nmp->nm_rsize; 4324 attrflag = 0; 4325 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof, 4326 &nfsva, &attrflag, cred, td); 4327 if (attrflag != 0) { 4328 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4329 if (error == 0 && ret != 0) 4330 error = ret; 4331 } 4332 if (error == 0) { 4333 len2 += len; 4334 if (len2 > SSIZE_MAX) 4335 error = ENOATTR; 4336 } 4337 } 4338 if (error == 0 && ap->a_size != NULL) 4339 *ap->a_size = len2; 4340 4341 switch (error) { 4342 case NFSERR_NOTSUPP: 4343 case NFSERR_OPILLEGAL: 4344 mtx_lock(&nmp->nm_mtx); 4345 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4346 mtx_unlock(&nmp->nm_mtx); 4347 error = EOPNOTSUPP; 4348 break; 4349 case NFSERR_NOXATTR: 4350 case NFSERR_XATTR2BIG: 4351 error = ENOATTR; 4352 break; 4353 default: 4354 error = nfscl_maperr(td, error, 0, 0); 4355 break; 4356 } 4357 return (error); 4358 } 4359 4360 /* 4361 * nfs setextattr call 4362 */ 4363 static int 4364 nfs_deleteextattr(struct vop_deleteextattr_args *ap) 4365 { 4366 struct vnode *vp = ap->a_vp; 4367 struct nfsmount *nmp; 4368 struct nfsvattr nfsva; 4369 int attrflag, error, ret; 4370 4371 nmp = VFSTONFS(vp->v_mount); 4372 mtx_lock(&nmp->nm_mtx); 4373 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4374 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4375 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4376 mtx_unlock(&nmp->nm_mtx); 4377 return (EOPNOTSUPP); 4378 } 4379 mtx_unlock(&nmp->nm_mtx); 4380 4381 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4382 attrflag = 0; 4383 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred, 4384 ap->a_td); 4385 if (attrflag != 0) { 4386 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4387 if (error == 0 && ret != 0) 4388 error = ret; 4389 } 4390 4391 switch (error) { 4392 case NFSERR_NOTSUPP: 4393 case NFSERR_OPILLEGAL: 4394 mtx_lock(&nmp->nm_mtx); 4395 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4396 mtx_unlock(&nmp->nm_mtx); 4397 error = EOPNOTSUPP; 4398 break; 4399 case NFSERR_NOXATTR: 4400 case NFSERR_XATTR2BIG: 4401 error = ENOATTR; 4402 break; 4403 default: 4404 error = nfscl_maperr(ap->a_td, error, 0, 0); 4405 break; 4406 } 4407 return (error); 4408 } 4409 4410 /* 4411 * Return POSIX pathconf information applicable to nfs filesystems. 4412 */ 4413 static int 4414 nfs_pathconf(struct vop_pathconf_args *ap) 4415 { 4416 struct nfsv3_pathconf pc; 4417 struct nfsvattr nfsva; 4418 struct vnode *vp = ap->a_vp; 4419 struct nfsmount *nmp; 4420 struct thread *td = curthread; 4421 off_t off; 4422 bool eof; 4423 int attrflag, error; 4424 4425 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 4426 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 4427 ap->a_name == _PC_NO_TRUNC)) || 4428 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 4429 /* 4430 * Since only the above 4 a_names are returned by the NFSv3 4431 * Pathconf RPC, there is no point in doing it for others. 4432 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 4433 * be used for _PC_NFS4_ACL as well. 4434 */ 4435 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 4436 &attrflag, NULL); 4437 if (attrflag != 0) 4438 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4439 if (error != 0) 4440 return (error); 4441 } else { 4442 /* 4443 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 4444 * just fake them. 4445 */ 4446 pc.pc_linkmax = NFS_LINK_MAX; 4447 pc.pc_namemax = NFS_MAXNAMLEN; 4448 pc.pc_notrunc = 1; 4449 pc.pc_chownrestricted = 1; 4450 pc.pc_caseinsensitive = 0; 4451 pc.pc_casepreserving = 1; 4452 error = 0; 4453 } 4454 switch (ap->a_name) { 4455 case _PC_LINK_MAX: 4456 #ifdef _LP64 4457 *ap->a_retval = pc.pc_linkmax; 4458 #else 4459 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 4460 #endif 4461 break; 4462 case _PC_NAME_MAX: 4463 *ap->a_retval = pc.pc_namemax; 4464 break; 4465 case _PC_PIPE_BUF: 4466 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 4467 *ap->a_retval = PIPE_BUF; 4468 else 4469 error = EINVAL; 4470 break; 4471 case _PC_CHOWN_RESTRICTED: 4472 *ap->a_retval = pc.pc_chownrestricted; 4473 break; 4474 case _PC_NO_TRUNC: 4475 *ap->a_retval = pc.pc_notrunc; 4476 break; 4477 case _PC_ACL_NFS4: 4478 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 4479 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 4480 *ap->a_retval = 1; 4481 else 4482 *ap->a_retval = 0; 4483 break; 4484 case _PC_ACL_PATH_MAX: 4485 if (NFS_ISV4(vp)) 4486 *ap->a_retval = ACL_MAX_ENTRIES; 4487 else 4488 *ap->a_retval = 3; 4489 break; 4490 case _PC_PRIO_IO: 4491 *ap->a_retval = 0; 4492 break; 4493 case _PC_SYNC_IO: 4494 *ap->a_retval = 0; 4495 break; 4496 case _PC_ALLOC_SIZE_MIN: 4497 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 4498 break; 4499 case _PC_FILESIZEBITS: 4500 if (NFS_ISV34(vp)) 4501 *ap->a_retval = 64; 4502 else 4503 *ap->a_retval = 32; 4504 break; 4505 case _PC_REC_INCR_XFER_SIZE: 4506 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4507 break; 4508 case _PC_REC_MAX_XFER_SIZE: 4509 *ap->a_retval = -1; /* means ``unlimited'' */ 4510 break; 4511 case _PC_REC_MIN_XFER_SIZE: 4512 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4513 break; 4514 case _PC_REC_XFER_ALIGN: 4515 *ap->a_retval = PAGE_SIZE; 4516 break; 4517 case _PC_SYMLINK_MAX: 4518 *ap->a_retval = NFS_MAXPATHLEN; 4519 break; 4520 case _PC_MIN_HOLE_SIZE: 4521 /* Only some NFSv4.2 servers support Seek for Holes. */ 4522 *ap->a_retval = 0; 4523 nmp = VFSTONFS(vp->v_mount); 4524 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) { 4525 /* 4526 * NFSv4.2 doesn't have an attribute for hole size, 4527 * so all we can do is see if the Seek operation is 4528 * supported and then use f_iosize as a "best guess". 4529 */ 4530 mtx_lock(&nmp->nm_mtx); 4531 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) { 4532 mtx_unlock(&nmp->nm_mtx); 4533 off = 0; 4534 attrflag = 0; 4535 error = nfsrpc_seek(vp, &off, &eof, 4536 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva, 4537 &attrflag); 4538 if (attrflag != 0) 4539 nfscl_loadattrcache(&vp, &nfsva, 4540 NULL, 0, 1); 4541 mtx_lock(&nmp->nm_mtx); 4542 if (error == NFSERR_NOTSUPP) 4543 nmp->nm_privflag |= NFSMNTP_SEEKTESTED; 4544 else 4545 nmp->nm_privflag |= NFSMNTP_SEEKTESTED | 4546 NFSMNTP_SEEK; 4547 error = 0; 4548 } 4549 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0) 4550 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4551 mtx_unlock(&nmp->nm_mtx); 4552 } 4553 break; 4554 4555 default: 4556 error = vop_stdpathconf(ap); 4557 break; 4558 } 4559 return (error); 4560 } 4561