1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 35 */ 36 37 #include <sys/cdefs.h> 38 /* 39 * vnode op calls for Sun NFS version 2, 3 and 4 40 */ 41 42 #include "opt_inet.h" 43 44 #include <sys/param.h> 45 #include <sys/kernel.h> 46 #include <sys/systm.h> 47 #include <sys/resourcevar.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/bio.h> 51 #include <sys/buf.h> 52 #include <sys/extattr.h> 53 #include <sys/filio.h> 54 #include <sys/jail.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/socket.h> 59 #include <sys/vnode.h> 60 #include <sys/dirent.h> 61 #include <sys/fcntl.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/sysctl.h> 65 #include <sys/signalvar.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_object.h> 70 71 #include <fs/nfs/nfsport.h> 72 #include <fs/nfsclient/nfsnode.h> 73 #include <fs/nfsclient/nfsmount.h> 74 #include <fs/nfsclient/nfs.h> 75 #include <fs/nfsclient/nfs_kdtrace.h> 76 77 #include <net/if.h> 78 #include <netinet/in.h> 79 #include <netinet/in_var.h> 80 81 #include <nfs/nfs_lock.h> 82 83 #ifdef KDTRACE_HOOKS 84 #include <sys/dtrace_bsd.h> 85 86 dtrace_nfsclient_accesscache_flush_probe_func_t 87 dtrace_nfscl_accesscache_flush_done_probe; 88 uint32_t nfscl_accesscache_flush_done_id; 89 90 dtrace_nfsclient_accesscache_get_probe_func_t 91 dtrace_nfscl_accesscache_get_hit_probe, 92 dtrace_nfscl_accesscache_get_miss_probe; 93 uint32_t nfscl_accesscache_get_hit_id; 94 uint32_t nfscl_accesscache_get_miss_id; 95 96 dtrace_nfsclient_accesscache_load_probe_func_t 97 dtrace_nfscl_accesscache_load_done_probe; 98 uint32_t nfscl_accesscache_load_done_id; 99 #endif /* !KDTRACE_HOOKS */ 100 101 /* Defs */ 102 #define TRUE 1 103 #define FALSE 0 104 105 extern struct nfsstatsv1 nfsstatsv1; 106 extern int nfsrv_useacl; 107 extern int nfscl_debuglevel; 108 MALLOC_DECLARE(M_NEWNFSREQ); 109 110 static vop_read_t nfsfifo_read; 111 static vop_write_t nfsfifo_write; 112 static vop_close_t nfsfifo_close; 113 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 114 struct thread *); 115 static vop_lookup_t nfs_lookup; 116 static vop_create_t nfs_create; 117 static vop_mknod_t nfs_mknod; 118 static vop_open_t nfs_open; 119 static vop_pathconf_t nfs_pathconf; 120 static vop_close_t nfs_close; 121 static vop_access_t nfs_access; 122 static vop_getattr_t nfs_getattr; 123 static vop_setattr_t nfs_setattr; 124 static vop_read_t nfs_read; 125 static vop_fsync_t nfs_fsync; 126 static vop_remove_t nfs_remove; 127 static vop_link_t nfs_link; 128 static vop_rename_t nfs_rename; 129 static vop_mkdir_t nfs_mkdir; 130 static vop_rmdir_t nfs_rmdir; 131 static vop_symlink_t nfs_symlink; 132 static vop_readdir_t nfs_readdir; 133 static vop_strategy_t nfs_strategy; 134 static int nfs_lookitup(struct vnode *, char *, int, 135 struct ucred *, struct thread *, struct nfsnode **); 136 static int nfs_sillyrename(struct vnode *, struct vnode *, 137 struct componentname *); 138 static vop_access_t nfsspec_access; 139 static vop_readlink_t nfs_readlink; 140 static vop_print_t nfs_print; 141 static vop_advlock_t nfs_advlock; 142 static vop_advlockasync_t nfs_advlockasync; 143 static vop_getacl_t nfs_getacl; 144 static vop_setacl_t nfs_setacl; 145 static vop_advise_t nfs_advise; 146 static vop_allocate_t nfs_allocate; 147 static vop_deallocate_t nfs_deallocate; 148 static vop_copy_file_range_t nfs_copy_file_range; 149 static vop_ioctl_t nfs_ioctl; 150 static vop_getextattr_t nfs_getextattr; 151 static vop_setextattr_t nfs_setextattr; 152 static vop_listextattr_t nfs_listextattr; 153 static vop_deleteextattr_t nfs_deleteextattr; 154 static vop_lock1_t nfs_lock; 155 156 /* 157 * Global vfs data structures for nfs 158 */ 159 160 static struct vop_vector newnfs_vnodeops_nosig = { 161 .vop_default = &default_vnodeops, 162 .vop_access = nfs_access, 163 .vop_advlock = nfs_advlock, 164 .vop_advlockasync = nfs_advlockasync, 165 .vop_close = nfs_close, 166 .vop_create = nfs_create, 167 .vop_fsync = nfs_fsync, 168 .vop_getattr = nfs_getattr, 169 .vop_getpages = ncl_getpages, 170 .vop_putpages = ncl_putpages, 171 .vop_inactive = ncl_inactive, 172 .vop_link = nfs_link, 173 .vop_lock1 = nfs_lock, 174 .vop_lookup = nfs_lookup, 175 .vop_mkdir = nfs_mkdir, 176 .vop_mknod = nfs_mknod, 177 .vop_open = nfs_open, 178 .vop_pathconf = nfs_pathconf, 179 .vop_print = nfs_print, 180 .vop_read = nfs_read, 181 .vop_readdir = nfs_readdir, 182 .vop_readlink = nfs_readlink, 183 .vop_reclaim = ncl_reclaim, 184 .vop_remove = nfs_remove, 185 .vop_rename = nfs_rename, 186 .vop_rmdir = nfs_rmdir, 187 .vop_setattr = nfs_setattr, 188 .vop_strategy = nfs_strategy, 189 .vop_symlink = nfs_symlink, 190 .vop_write = ncl_write, 191 .vop_getacl = nfs_getacl, 192 .vop_setacl = nfs_setacl, 193 .vop_advise = nfs_advise, 194 .vop_allocate = nfs_allocate, 195 .vop_deallocate = nfs_deallocate, 196 .vop_copy_file_range = nfs_copy_file_range, 197 .vop_ioctl = nfs_ioctl, 198 .vop_getextattr = nfs_getextattr, 199 .vop_setextattr = nfs_setextattr, 200 .vop_listextattr = nfs_listextattr, 201 .vop_deleteextattr = nfs_deleteextattr, 202 }; 203 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig); 204 205 static int 206 nfs_vnodeops_bypass(struct vop_generic_args *a) 207 { 208 209 return (vop_sigdefer(&newnfs_vnodeops_nosig, a)); 210 } 211 212 struct vop_vector newnfs_vnodeops = { 213 .vop_default = &default_vnodeops, 214 .vop_bypass = nfs_vnodeops_bypass, 215 }; 216 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops); 217 218 static struct vop_vector newnfs_fifoops_nosig = { 219 .vop_default = &fifo_specops, 220 .vop_access = nfsspec_access, 221 .vop_close = nfsfifo_close, 222 .vop_fsync = nfs_fsync, 223 .vop_getattr = nfs_getattr, 224 .vop_inactive = ncl_inactive, 225 .vop_pathconf = nfs_pathconf, 226 .vop_print = nfs_print, 227 .vop_read = nfsfifo_read, 228 .vop_reclaim = ncl_reclaim, 229 .vop_setattr = nfs_setattr, 230 .vop_write = nfsfifo_write, 231 }; 232 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig); 233 234 static int 235 nfs_fifoops_bypass(struct vop_generic_args *a) 236 { 237 238 return (vop_sigdefer(&newnfs_fifoops_nosig, a)); 239 } 240 241 struct vop_vector newnfs_fifoops = { 242 .vop_default = &default_vnodeops, 243 .vop_bypass = nfs_fifoops_bypass, 244 }; 245 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops); 246 247 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 248 struct componentname *cnp, struct vattr *vap); 249 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 250 int namelen, struct ucred *cred, struct thread *td); 251 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 252 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 253 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 254 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 255 struct componentname *scnp, struct sillyrename *sp); 256 257 /* 258 * Global variables 259 */ 260 SYSCTL_DECL(_vfs_nfs); 261 262 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 264 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 265 266 static int nfs_prime_access_cache = 0; 267 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 268 &nfs_prime_access_cache, 0, 269 "Prime NFS ACCESS cache when fetching attributes"); 270 271 static int newnfs_commit_on_close = 0; 272 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 273 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 274 275 static int nfs_clean_pages_on_close = 1; 276 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 277 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 278 279 int newnfs_directio_enable = 0; 280 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 281 &newnfs_directio_enable, 0, "Enable NFS directio"); 282 283 int nfs_keep_dirty_on_error; 284 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 285 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 286 287 /* 288 * This sysctl allows other processes to mmap a file that has been opened 289 * O_DIRECT by a process. In general, having processes mmap the file while 290 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 291 * this by default to prevent DoS attacks - to prevent a malicious user from 292 * opening up files O_DIRECT preventing other users from mmap'ing these 293 * files. "Protected" environments where stricter consistency guarantees are 294 * required can disable this knob. The process that opened the file O_DIRECT 295 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 296 * meaningful. 297 */ 298 int newnfs_directio_allow_mmap = 1; 299 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 300 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 301 302 static uint64_t nfs_maxalloclen = 64 * 1024 * 1024; 303 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW, 304 &nfs_maxalloclen, 0, "NFS max allocate/deallocate length"); 305 306 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 307 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 308 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 309 310 /* 311 * SMP Locking Note : 312 * The list of locks after the description of the lock is the ordering 313 * of other locks acquired with the lock held. 314 * np->n_mtx : Protects the fields in the nfsnode. 315 VM Object Lock 316 VI_MTX (acquired indirectly) 317 * nmp->nm_mtx : Protects the fields in the nfsmount. 318 rep->r_mtx 319 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 320 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 321 nmp->nm_mtx 322 rep->r_mtx 323 * rep->r_mtx : Protects the fields in an nfsreq. 324 */ 325 326 static int 327 nfs_lock(struct vop_lock1_args *ap) 328 { 329 struct vnode *vp; 330 struct nfsnode *np; 331 u_quad_t nsize; 332 int error, lktype; 333 bool onfault; 334 335 vp = ap->a_vp; 336 lktype = ap->a_flags & LK_TYPE_MASK; 337 error = VOP_LOCK1_APV(&default_vnodeops, ap); 338 if (error != 0 || vp->v_op != &newnfs_vnodeops) 339 return (error); 340 np = VTONFS(vp); 341 if (np == NULL) 342 return (0); 343 NFSLOCKNODE(np); 344 if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED && 345 lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE && 346 lktype != LK_TRYUPGRADE)) { 347 NFSUNLOCKNODE(np); 348 return (0); 349 } 350 onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT && 351 (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE && 352 (lktype == LK_SHARED || lktype == LK_EXCLUSIVE); 353 if (onfault && vp->v_vnlock->lk_recurse == 0) { 354 /* 355 * Force retry in vm_fault(), to make the lock request 356 * sleepable, which allows us to piggy-back the 357 * sleepable call to vnode_pager_setsize(). 358 */ 359 NFSUNLOCKNODE(np); 360 VOP_UNLOCK(vp); 361 return (EBUSY); 362 } 363 if ((ap->a_flags & LK_NOWAIT) != 0 || 364 (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) { 365 NFSUNLOCKNODE(np); 366 return (0); 367 } 368 if (lktype == LK_SHARED) { 369 NFSUNLOCKNODE(np); 370 VOP_UNLOCK(vp); 371 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 372 ap->a_flags |= LK_EXCLUSIVE; 373 error = VOP_LOCK1_APV(&default_vnodeops, ap); 374 if (error != 0 || vp->v_op != &newnfs_vnodeops) 375 return (error); 376 if (vp->v_data == NULL) 377 goto downgrade; 378 MPASS(vp->v_data == np); 379 NFSLOCKNODE(np); 380 if ((np->n_flag & NVNSETSZSKIP) == 0) { 381 NFSUNLOCKNODE(np); 382 goto downgrade; 383 } 384 } 385 np->n_flag &= ~NVNSETSZSKIP; 386 nsize = np->n_size; 387 NFSUNLOCKNODE(np); 388 vnode_pager_setsize(vp, nsize); 389 downgrade: 390 if (lktype == LK_SHARED) { 391 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 392 ap->a_flags |= LK_DOWNGRADE; 393 (void)VOP_LOCK1_APV(&default_vnodeops, ap); 394 } 395 return (0); 396 } 397 398 static int 399 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 400 struct ucred *cred, u_int32_t *retmode) 401 { 402 int error = 0, attrflag, i, lrupos; 403 u_int32_t rmode; 404 struct nfsnode *np = VTONFS(vp); 405 struct nfsvattr nfsva; 406 407 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 408 &rmode); 409 if (attrflag) 410 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 411 if (!error) { 412 lrupos = 0; 413 NFSLOCKNODE(np); 414 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 415 if (np->n_accesscache[i].uid == cred->cr_uid) { 416 np->n_accesscache[i].mode = rmode; 417 np->n_accesscache[i].stamp = time_second; 418 break; 419 } 420 if (i > 0 && np->n_accesscache[i].stamp < 421 np->n_accesscache[lrupos].stamp) 422 lrupos = i; 423 } 424 if (i == NFS_ACCESSCACHESIZE) { 425 np->n_accesscache[lrupos].uid = cred->cr_uid; 426 np->n_accesscache[lrupos].mode = rmode; 427 np->n_accesscache[lrupos].stamp = time_second; 428 } 429 NFSUNLOCKNODE(np); 430 if (retmode != NULL) 431 *retmode = rmode; 432 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 433 } else if (NFS_ISV4(vp)) { 434 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 435 } 436 #ifdef KDTRACE_HOOKS 437 if (error != 0) 438 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 439 error); 440 #endif 441 return (error); 442 } 443 444 /* 445 * nfs access vnode op. 446 * For nfs version 2, just return ok. File accesses may fail later. 447 * For nfs version 3, use the access rpc to check accessibility. If file modes 448 * are changed on the server, accesses might still fail later. 449 */ 450 static int 451 nfs_access(struct vop_access_args *ap) 452 { 453 struct vnode *vp = ap->a_vp; 454 int error = 0, i, gotahit; 455 u_int32_t mode, wmode, rmode; 456 int v34 = NFS_ISV34(vp); 457 struct nfsnode *np = VTONFS(vp); 458 459 /* 460 * Disallow write attempts on filesystems mounted read-only; 461 * unless the file is a socket, fifo, or a block or character 462 * device resident on the filesystem. 463 */ 464 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 465 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 466 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 467 switch (vp->v_type) { 468 case VREG: 469 case VDIR: 470 case VLNK: 471 return (EROFS); 472 default: 473 break; 474 } 475 } 476 /* 477 * For nfs v3 or v4, check to see if we have done this recently, and if 478 * so return our cached result instead of making an ACCESS call. 479 * If not, do an access rpc, otherwise you are stuck emulating 480 * ufs_access() locally using the vattr. This may not be correct, 481 * since the server may apply other access criteria such as 482 * client uid-->server uid mapping that we do not know about. 483 */ 484 if (v34) { 485 if (ap->a_accmode & VREAD) 486 mode = NFSACCESS_READ; 487 else 488 mode = 0; 489 if (vp->v_type != VDIR) { 490 if (ap->a_accmode & VWRITE) 491 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 492 if (ap->a_accmode & VAPPEND) 493 mode |= NFSACCESS_EXTEND; 494 if (ap->a_accmode & VEXEC) 495 mode |= NFSACCESS_EXECUTE; 496 if (ap->a_accmode & VDELETE) 497 mode |= NFSACCESS_DELETE; 498 } else { 499 if (ap->a_accmode & VWRITE) 500 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 501 if (ap->a_accmode & VAPPEND) 502 mode |= NFSACCESS_EXTEND; 503 if (ap->a_accmode & VEXEC) 504 mode |= NFSACCESS_LOOKUP; 505 if (ap->a_accmode & VDELETE) 506 mode |= NFSACCESS_DELETE; 507 if (ap->a_accmode & VDELETE_CHILD) 508 mode |= NFSACCESS_MODIFY; 509 } 510 /* XXX safety belt, only make blanket request if caching */ 511 if (nfsaccess_cache_timeout > 0) { 512 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 513 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 514 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 515 } else { 516 wmode = mode; 517 } 518 519 /* 520 * Does our cached result allow us to give a definite yes to 521 * this request? 522 */ 523 gotahit = 0; 524 NFSLOCKNODE(np); 525 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 526 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 527 if (time_second < (np->n_accesscache[i].stamp 528 + nfsaccess_cache_timeout) && 529 (np->n_accesscache[i].mode & mode) == mode) { 530 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 531 gotahit = 1; 532 } 533 break; 534 } 535 } 536 NFSUNLOCKNODE(np); 537 #ifdef KDTRACE_HOOKS 538 if (gotahit != 0) 539 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 540 ap->a_cred->cr_uid, mode); 541 else 542 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 543 ap->a_cred->cr_uid, mode); 544 #endif 545 if (gotahit == 0) { 546 /* 547 * Either a no, or a don't know. Go to the wire. 548 */ 549 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 550 error = nfs34_access_otw(vp, wmode, ap->a_td, 551 ap->a_cred, &rmode); 552 if (!error && 553 (rmode & mode) != mode) 554 error = EACCES; 555 } 556 return (error); 557 } else { 558 if ((error = nfsspec_access(ap)) != 0) { 559 return (error); 560 } 561 /* 562 * Attempt to prevent a mapped root from accessing a file 563 * which it shouldn't. We try to read a byte from the file 564 * if the user is root and the file is not zero length. 565 * After calling nfsspec_access, we should have the correct 566 * file size cached. 567 */ 568 NFSLOCKNODE(np); 569 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 570 && VTONFS(vp)->n_size > 0) { 571 struct iovec aiov; 572 struct uio auio; 573 char buf[1]; 574 575 NFSUNLOCKNODE(np); 576 aiov.iov_base = buf; 577 aiov.iov_len = 1; 578 auio.uio_iov = &aiov; 579 auio.uio_iovcnt = 1; 580 auio.uio_offset = 0; 581 auio.uio_resid = 1; 582 auio.uio_segflg = UIO_SYSSPACE; 583 auio.uio_rw = UIO_READ; 584 auio.uio_td = ap->a_td; 585 586 if (vp->v_type == VREG) 587 error = ncl_readrpc(vp, &auio, ap->a_cred); 588 else if (vp->v_type == VDIR) { 589 char* bp; 590 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 591 aiov.iov_base = bp; 592 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 593 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 594 ap->a_td); 595 free(bp, M_TEMP); 596 } else if (vp->v_type == VLNK) 597 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 598 else 599 error = EACCES; 600 } else 601 NFSUNLOCKNODE(np); 602 return (error); 603 } 604 } 605 606 /* 607 * nfs open vnode op 608 * Check to see if the type is ok 609 * and that deletion is not in progress. 610 * For paged in text files, you will need to flush the page cache 611 * if consistency is lost. 612 */ 613 /* ARGSUSED */ 614 static int 615 nfs_open(struct vop_open_args *ap) 616 { 617 struct vnode *vp = ap->a_vp; 618 struct nfsnode *np = VTONFS(vp); 619 struct vattr vattr; 620 int error; 621 int fmode = ap->a_mode; 622 struct ucred *cred; 623 vm_object_t obj; 624 625 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 626 return (EOPNOTSUPP); 627 628 /* 629 * For NFSv4, we need to do the Open Op before cache validation, 630 * so that we conform to RFC3530 Sec. 9.3.1. 631 */ 632 if (NFS_ISV4(vp)) { 633 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 634 if (error) { 635 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 636 (gid_t)0); 637 return (error); 638 } 639 } 640 641 /* 642 * Now, if this Open will be doing reading, re-validate/flush the 643 * cache, so that Close/Open coherency is maintained. 644 */ 645 NFSLOCKNODE(np); 646 if (np->n_flag & NMODIFIED) { 647 NFSUNLOCKNODE(np); 648 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 649 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 650 if (VN_IS_DOOMED(vp)) 651 return (EBADF); 652 } 653 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 654 if (error == EINTR || error == EIO) { 655 if (NFS_ISV4(vp)) 656 (void) nfsrpc_close(vp, 0, ap->a_td); 657 return (error); 658 } 659 NFSLOCKNODE(np); 660 np->n_attrstamp = 0; 661 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 662 if (vp->v_type == VDIR) 663 np->n_direofoffset = 0; 664 NFSUNLOCKNODE(np); 665 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 666 if (error) { 667 if (NFS_ISV4(vp)) 668 (void) nfsrpc_close(vp, 0, ap->a_td); 669 return (error); 670 } 671 NFSLOCKNODE(np); 672 np->n_mtime = vattr.va_mtime; 673 if (NFS_ISV4(vp)) 674 np->n_change = vattr.va_filerev; 675 } else { 676 NFSUNLOCKNODE(np); 677 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 678 if (error) { 679 if (NFS_ISV4(vp)) 680 (void) nfsrpc_close(vp, 0, ap->a_td); 681 return (error); 682 } 683 NFSLOCKNODE(np); 684 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 685 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 686 if (vp->v_type == VDIR) 687 np->n_direofoffset = 0; 688 NFSUNLOCKNODE(np); 689 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 690 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 691 if (VN_IS_DOOMED(vp)) 692 return (EBADF); 693 } 694 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 695 if (error == EINTR || error == EIO) { 696 if (NFS_ISV4(vp)) 697 (void) nfsrpc_close(vp, 0, ap->a_td); 698 return (error); 699 } 700 NFSLOCKNODE(np); 701 np->n_mtime = vattr.va_mtime; 702 if (NFS_ISV4(vp)) 703 np->n_change = vattr.va_filerev; 704 } 705 } 706 707 /* 708 * If the object has >= 1 O_DIRECT active opens, we disable caching. 709 */ 710 if (newnfs_directio_enable && (fmode & O_DIRECT) && 711 (vp->v_type == VREG)) { 712 if (np->n_directio_opens == 0) { 713 NFSUNLOCKNODE(np); 714 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 715 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 716 if (VN_IS_DOOMED(vp)) 717 return (EBADF); 718 } 719 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 720 if (error) { 721 if (NFS_ISV4(vp)) 722 (void) nfsrpc_close(vp, 0, ap->a_td); 723 return (error); 724 } 725 NFSLOCKNODE(np); 726 np->n_flag |= NNONCACHE; 727 } 728 np->n_directio_opens++; 729 } 730 731 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 732 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 733 np->n_flag |= NWRITEOPENED; 734 735 /* 736 * If this is an open for writing, capture a reference to the 737 * credentials, so they can be used by ncl_putpages(). Using 738 * these write credentials is preferable to the credentials of 739 * whatever thread happens to be doing the VOP_PUTPAGES() since 740 * the write RPCs are less likely to fail with EACCES. 741 */ 742 if ((fmode & FWRITE) != 0) { 743 cred = np->n_writecred; 744 np->n_writecred = crhold(ap->a_cred); 745 } else 746 cred = NULL; 747 NFSUNLOCKNODE(np); 748 749 if (cred != NULL) 750 crfree(cred); 751 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 752 753 /* 754 * If the text file has been mmap'd, flush any dirty pages to the 755 * buffer cache and then... 756 * Make sure all writes are pushed to the NFS server. If this is not 757 * done, the modify time of the file can change while the text 758 * file is being executed. This will cause the process that is 759 * executing the text file to be terminated. 760 */ 761 if (vp->v_writecount <= -1) { 762 if ((obj = vp->v_object) != NULL && 763 vm_object_mightbedirty(obj)) { 764 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 765 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 766 if (VN_IS_DOOMED(vp)) 767 return (EBADF); 768 } 769 VM_OBJECT_WLOCK(obj); 770 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); 771 VM_OBJECT_WUNLOCK(obj); 772 } 773 774 /* Now, flush the buffer cache. */ 775 ncl_flush(vp, MNT_WAIT, curthread, 0, 0); 776 777 /* And, finally, make sure that n_mtime is up to date. */ 778 np = VTONFS(vp); 779 NFSLOCKNODE(np); 780 np->n_mtime = np->n_vattr.na_mtime; 781 NFSUNLOCKNODE(np); 782 } 783 return (0); 784 } 785 786 /* 787 * nfs close vnode op 788 * What an NFS client should do upon close after writing is a debatable issue. 789 * Most NFS clients push delayed writes to the server upon close, basically for 790 * two reasons: 791 * 1 - So that any write errors may be reported back to the client process 792 * doing the close system call. By far the two most likely errors are 793 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 794 * 2 - To put a worst case upper bound on cache inconsistency between 795 * multiple clients for the file. 796 * There is also a consistency problem for Version 2 of the protocol w.r.t. 797 * not being able to tell if other clients are writing a file concurrently, 798 * since there is no way of knowing if the changed modify time in the reply 799 * is only due to the write for this client. 800 * (NFS Version 3 provides weak cache consistency data in the reply that 801 * should be sufficient to detect and handle this case.) 802 * 803 * The current code does the following: 804 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 805 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 806 * or commit them (this satisfies 1 and 2 except for the 807 * case where the server crashes after this close but 808 * before the commit RPC, which is felt to be "good 809 * enough". Changing the last argument to ncl_flush() to 810 * a 1 would force a commit operation, if it is felt a 811 * commit is necessary now. 812 * for NFS Version 4 - flush the dirty buffers and commit them, if 813 * nfscl_mustflush() says this is necessary. 814 * It is necessary if there is no write delegation held, 815 * in order to satisfy open/close coherency. 816 * If the file isn't cached on local stable storage, 817 * it may be necessary in order to detect "out of space" 818 * errors from the server, if the write delegation 819 * issued by the server doesn't allow the file to grow. 820 */ 821 /* ARGSUSED */ 822 static int 823 nfs_close(struct vop_close_args *ap) 824 { 825 struct vnode *vp = ap->a_vp; 826 struct nfsnode *np = VTONFS(vp); 827 struct nfsvattr nfsva; 828 struct ucred *cred; 829 int error = 0, ret, localcred = 0; 830 int fmode = ap->a_fflag; 831 832 if (NFSCL_FORCEDISM(vp->v_mount)) 833 return (0); 834 /* 835 * During shutdown, a_cred isn't valid, so just use root. 836 */ 837 if (ap->a_cred == NOCRED) { 838 cred = newnfs_getcred(); 839 localcred = 1; 840 } else { 841 cred = ap->a_cred; 842 } 843 if (vp->v_type == VREG) { 844 /* 845 * Examine and clean dirty pages, regardless of NMODIFIED. 846 * This closes a major hole in close-to-open consistency. 847 * We want to push out all dirty pages (and buffers) on 848 * close, regardless of whether they were dirtied by 849 * mmap'ed writes or via write(). 850 */ 851 if (nfs_clean_pages_on_close && vp->v_object) { 852 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 853 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 854 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 855 return (EBADF); 856 } 857 VM_OBJECT_WLOCK(vp->v_object); 858 vm_object_page_clean(vp->v_object, 0, 0, 0); 859 VM_OBJECT_WUNLOCK(vp->v_object); 860 } 861 NFSLOCKNODE(np); 862 if (np->n_flag & NMODIFIED) { 863 NFSUNLOCKNODE(np); 864 if (NFS_ISV3(vp)) { 865 /* 866 * Under NFSv3 we have dirty buffers to dispose of. We 867 * must flush them to the NFS server. We have the option 868 * of waiting all the way through the commit rpc or just 869 * waiting for the initial write. The default is to only 870 * wait through the initial write so the data is in the 871 * server's cache, which is roughly similar to the state 872 * a standard disk subsystem leaves the file in on close(). 873 * 874 * We cannot clear the NMODIFIED bit in np->n_flag due to 875 * potential races with other processes, and certainly 876 * cannot clear it if we don't commit. 877 * These races occur when there is no longer the old 878 * traditional vnode locking implemented for Vnode Ops. 879 */ 880 int cm = newnfs_commit_on_close ? 1 : 0; 881 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 882 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 883 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 884 return (EBADF); 885 } 886 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); 887 /* np->n_flag &= ~NMODIFIED; */ 888 } else if (NFS_ISV4(vp)) { 889 if (nfscl_mustflush(vp) != 0) { 890 int cm = newnfs_commit_on_close ? 1 : 0; 891 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 892 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 893 if (VN_IS_DOOMED(vp) && ap->a_fflag != 894 FNONBLOCK) 895 return (EBADF); 896 } 897 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 898 cm, 0); 899 /* 900 * as above w.r.t races when clearing 901 * NMODIFIED. 902 * np->n_flag &= ~NMODIFIED; 903 */ 904 } 905 } else { 906 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 907 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 908 if (VN_IS_DOOMED(vp) && ap->a_fflag != 909 FNONBLOCK) 910 return (EBADF); 911 } 912 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 913 } 914 NFSLOCKNODE(np); 915 } 916 /* 917 * Invalidate the attribute cache in all cases. 918 * An open is going to fetch fresh attrs any way, other procs 919 * on this node that have file open will be forced to do an 920 * otw attr fetch, but this is safe. 921 * --> A user found that their RPC count dropped by 20% when 922 * this was commented out and I can't see any requirement 923 * for it, so I've disabled it when negative lookups are 924 * enabled. (What does this have to do with negative lookup 925 * caching? Well nothing, except it was reported by the 926 * same user that needed negative lookup caching and I wanted 927 * there to be a way to disable it to see if it 928 * is the cause of some caching/coherency issue that might 929 * crop up.) 930 */ 931 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 932 np->n_attrstamp = 0; 933 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 934 } 935 if (np->n_flag & NWRITEERR) { 936 np->n_flag &= ~NWRITEERR; 937 error = np->n_error; 938 } 939 NFSUNLOCKNODE(np); 940 } 941 942 if (NFS_ISV4(vp)) { 943 /* 944 * Get attributes so "change" is up to date. 945 */ 946 if (error == 0 && nfscl_mustflush(vp) != 0 && 947 vp->v_type == VREG && 948 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 949 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva); 950 if (!ret) { 951 np->n_change = nfsva.na_filerev; 952 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 953 0, 0); 954 } 955 } 956 957 /* 958 * and do the close. 959 */ 960 ret = nfsrpc_close(vp, 0, ap->a_td); 961 if (!error && ret) 962 error = ret; 963 if (error) 964 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 965 (gid_t)0); 966 } 967 if (newnfs_directio_enable) 968 KASSERT((np->n_directio_asyncwr == 0), 969 ("nfs_close: dirty unflushed (%d) directio buffers\n", 970 np->n_directio_asyncwr)); 971 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 972 NFSLOCKNODE(np); 973 KASSERT((np->n_directio_opens > 0), 974 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 975 np->n_directio_opens--; 976 if (np->n_directio_opens == 0) 977 np->n_flag &= ~NNONCACHE; 978 NFSUNLOCKNODE(np); 979 } 980 if (localcred) 981 NFSFREECRED(cred); 982 return (error); 983 } 984 985 /* 986 * nfs getattr call from vfs. 987 */ 988 static int 989 nfs_getattr(struct vop_getattr_args *ap) 990 { 991 struct vnode *vp = ap->a_vp; 992 struct thread *td = curthread; /* XXX */ 993 struct nfsnode *np = VTONFS(vp); 994 int error = 0; 995 struct nfsvattr nfsva; 996 struct vattr *vap = ap->a_vap; 997 struct vattr vattr; 998 struct nfsmount *nmp; 999 1000 nmp = VFSTONFS(vp->v_mount); 1001 /* 1002 * Update local times for special files. 1003 */ 1004 NFSLOCKNODE(np); 1005 if (np->n_flag & (NACC | NUPD)) 1006 np->n_flag |= NCHG; 1007 NFSUNLOCKNODE(np); 1008 /* 1009 * First look in the cache. 1010 * For "syskrb5" mounts, nm_fhsize might still be zero and 1011 * cached attributes should be ignored. 1012 */ 1013 if (nmp->nm_fhsize > 0 && ncl_getattrcache(vp, &vattr) == 0) { 1014 ncl_copy_vattr(vap, &vattr); 1015 1016 /* 1017 * Get the local modify time for the case of a write 1018 * delegation. 1019 */ 1020 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1021 return (0); 1022 } 1023 1024 if (NFS_ISV34(vp) && nfs_prime_access_cache && 1025 nfsaccess_cache_timeout > 0) { 1026 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 1027 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 1028 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 1029 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 1030 return (0); 1031 } 1032 } 1033 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva); 1034 if (!error) 1035 error = nfscl_loadattrcache(&vp, &nfsva, vap, 0, 0); 1036 if (!error) { 1037 /* 1038 * Get the local modify time for the case of a write 1039 * delegation. 1040 */ 1041 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1042 } else if (NFS_ISV4(vp)) { 1043 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1044 } 1045 return (error); 1046 } 1047 1048 /* 1049 * nfs setattr call. 1050 */ 1051 static int 1052 nfs_setattr(struct vop_setattr_args *ap) 1053 { 1054 struct vnode *vp = ap->a_vp; 1055 struct nfsnode *np = VTONFS(vp); 1056 struct thread *td = curthread; /* XXX */ 1057 struct vattr *vap = ap->a_vap; 1058 int error = 0; 1059 u_quad_t tsize; 1060 struct timespec ts; 1061 1062 #ifndef nolint 1063 tsize = (u_quad_t)0; 1064 #endif 1065 1066 /* 1067 * Setting of flags and marking of atimes are not supported. 1068 */ 1069 if (vap->va_flags != VNOVAL) 1070 return (EOPNOTSUPP); 1071 1072 /* 1073 * Disallow write attempts if the filesystem is mounted read-only. 1074 */ 1075 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 1076 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 1077 vap->va_mtime.tv_sec != VNOVAL || 1078 vap->va_birthtime.tv_sec != VNOVAL || 1079 vap->va_mode != (mode_t)VNOVAL) && 1080 (vp->v_mount->mnt_flag & MNT_RDONLY)) 1081 return (EROFS); 1082 if (vap->va_size != VNOVAL) { 1083 switch (vp->v_type) { 1084 case VDIR: 1085 return (EISDIR); 1086 case VCHR: 1087 case VBLK: 1088 case VSOCK: 1089 case VFIFO: 1090 if (vap->va_mtime.tv_sec == VNOVAL && 1091 vap->va_atime.tv_sec == VNOVAL && 1092 vap->va_birthtime.tv_sec == VNOVAL && 1093 vap->va_mode == (mode_t)VNOVAL && 1094 vap->va_uid == (uid_t)VNOVAL && 1095 vap->va_gid == (gid_t)VNOVAL) 1096 return (0); 1097 vap->va_size = VNOVAL; 1098 break; 1099 default: 1100 /* 1101 * Disallow write attempts if the filesystem is 1102 * mounted read-only. 1103 */ 1104 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1105 return (EROFS); 1106 /* 1107 * We run vnode_pager_setsize() early (why?), 1108 * we must set np->n_size now to avoid vinvalbuf 1109 * V_SAVE races that might setsize a lower 1110 * value. 1111 */ 1112 NFSLOCKNODE(np); 1113 tsize = np->n_size; 1114 NFSUNLOCKNODE(np); 1115 error = ncl_meta_setsize(vp, td, vap->va_size); 1116 NFSLOCKNODE(np); 1117 if (np->n_flag & NMODIFIED) { 1118 tsize = np->n_size; 1119 NFSUNLOCKNODE(np); 1120 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 1121 0 : V_SAVE, td, 1); 1122 if (error != 0) { 1123 vnode_pager_setsize(vp, tsize); 1124 return (error); 1125 } 1126 /* 1127 * Call nfscl_delegmodtime() to set the modify time 1128 * locally, as required. 1129 */ 1130 nfscl_delegmodtime(vp); 1131 } else 1132 NFSUNLOCKNODE(np); 1133 /* 1134 * np->n_size has already been set to vap->va_size 1135 * in ncl_meta_setsize(). We must set it again since 1136 * nfs_loadattrcache() could be called through 1137 * ncl_meta_setsize() and could modify np->n_size. 1138 */ 1139 NFSLOCKNODE(np); 1140 np->n_vattr.na_size = np->n_size = vap->va_size; 1141 NFSUNLOCKNODE(np); 1142 } 1143 } else { 1144 NFSLOCKNODE(np); 1145 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 1146 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 1147 NFSUNLOCKNODE(np); 1148 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 1149 if (error == EINTR || error == EIO) 1150 return (error); 1151 } else 1152 NFSUNLOCKNODE(np); 1153 } 1154 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 1155 if (vap->va_size != VNOVAL) { 1156 if (error == 0) { 1157 nanouptime(&ts); 1158 NFSLOCKNODE(np); 1159 np->n_localmodtime = ts; 1160 NFSUNLOCKNODE(np); 1161 } else { 1162 NFSLOCKNODE(np); 1163 np->n_size = np->n_vattr.na_size = tsize; 1164 vnode_pager_setsize(vp, tsize); 1165 NFSUNLOCKNODE(np); 1166 } 1167 } 1168 return (error); 1169 } 1170 1171 /* 1172 * Do an nfs setattr rpc. 1173 */ 1174 static int 1175 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1176 struct thread *td) 1177 { 1178 struct nfsnode *np = VTONFS(vp); 1179 int error, ret, attrflag, i; 1180 struct nfsvattr nfsva; 1181 1182 if (NFS_ISV34(vp)) { 1183 NFSLOCKNODE(np); 1184 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1185 np->n_accesscache[i].stamp = 0; 1186 np->n_flag |= NDELEGMOD; 1187 NFSUNLOCKNODE(np); 1188 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1189 } 1190 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag); 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 return (0); 1256 } 1257 1258 /* 1259 * We only accept a positive hit in the cache if the 1260 * change time of the file matches our cached copy. 1261 * Otherwise, we discard the cache entry and fallback 1262 * to doing a lookup RPC. We also only trust cache 1263 * entries for less than nm_nametimeo seconds. 1264 * 1265 * To better handle stale file handles and attributes, 1266 * clear the attribute cache of this node if it is a 1267 * leaf component, part of an open() call, and not 1268 * locally modified before fetching the attributes. 1269 * This should allow stale file handles to be detected 1270 * here where we can fall back to a LOOKUP RPC to 1271 * recover rather than having nfs_open() detect the 1272 * stale file handle and failing open(2) with ESTALE. 1273 */ 1274 newvp = *vpp; 1275 newnp = VTONFS(newvp); 1276 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1277 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1278 !(newnp->n_flag & NMODIFIED)) { 1279 NFSLOCKNODE(newnp); 1280 newnp->n_attrstamp = 0; 1281 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1282 NFSUNLOCKNODE(newnp); 1283 } 1284 if (nfscl_nodeleg(newvp, 0) == 0 || 1285 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1286 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1287 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1288 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1289 return (0); 1290 } 1291 cache_purge(newvp); 1292 if (dvp != newvp) 1293 vput(newvp); 1294 else 1295 vrele(newvp); 1296 *vpp = NULLVP; 1297 } else if (error == ENOENT) { 1298 if (VN_IS_DOOMED(dvp)) 1299 return (ENOENT); 1300 /* 1301 * We only accept a negative hit in the cache if the 1302 * modification time of the parent directory matches 1303 * the cached copy in the name cache entry. 1304 * Otherwise, we discard all of the negative cache 1305 * entries for this directory. We also only trust 1306 * negative cache entries for up to nm_negnametimeo 1307 * seconds. 1308 */ 1309 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1310 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1311 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1312 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1313 return (ENOENT); 1314 } 1315 cache_purge_negative(dvp); 1316 } 1317 1318 openmode = 0; 1319 /* 1320 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount 1321 * option, it is possible to do the Open operation in the same 1322 * compound as Lookup, so long as delegations are not being 1323 * issued. This saves doing a separate RPC for Open. 1324 * For pnfs, do not do this, since the Open+LayoutGet will 1325 * be needed as a separate RPC. 1326 */ 1327 NFSLOCKMNT(nmp); 1328 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) && 1329 (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 && 1330 (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) && 1331 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) { 1332 if ((flags & OPENREAD) != 0) 1333 openmode |= NFSV4OPEN_ACCESSREAD; 1334 if ((flags & OPENWRITE) != 0) 1335 openmode |= NFSV4OPEN_ACCESSWRITE; 1336 } 1337 NFSUNLOCKMNT(nmp); 1338 1339 newvp = NULLVP; 1340 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1341 nanouptime(&ts); 1342 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1343 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1344 openmode); 1345 if (dattrflag) 1346 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1347 if (error) { 1348 if (newvp != NULLVP) { 1349 vput(newvp); 1350 *vpp = NULLVP; 1351 } 1352 1353 if (error != ENOENT) { 1354 if (NFS_ISV4(dvp)) 1355 error = nfscl_maperr(td, error, (uid_t)0, 1356 (gid_t)0); 1357 return (error); 1358 } 1359 1360 /* The requested file was not found. */ 1361 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1362 (flags & ISLASTCN)) { 1363 /* 1364 * XXX: UFS does a full VOP_ACCESS(dvp, 1365 * VWRITE) here instead of just checking 1366 * MNT_RDONLY. 1367 */ 1368 if (mp->mnt_flag & MNT_RDONLY) 1369 return (EROFS); 1370 return (EJUSTRETURN); 1371 } 1372 1373 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1374 /* 1375 * Cache the modification time of the parent 1376 * directory from the post-op attributes in 1377 * the name cache entry. The negative cache 1378 * entry will be ignored once the directory 1379 * has changed. Don't bother adding the entry 1380 * if the directory has already changed. 1381 */ 1382 NFSLOCKNODE(np); 1383 if (timespeccmp(&np->n_vattr.na_mtime, 1384 &dnfsva.na_mtime, ==)) { 1385 NFSUNLOCKNODE(np); 1386 cache_enter_time(dvp, NULL, cnp, 1387 &dnfsva.na_mtime, NULL); 1388 } else 1389 NFSUNLOCKNODE(np); 1390 } 1391 return (ENOENT); 1392 } 1393 1394 /* 1395 * Handle RENAME case... 1396 */ 1397 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1398 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1399 free(nfhp, M_NFSFH); 1400 return (EISDIR); 1401 } 1402 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, LK_EXCLUSIVE); 1403 if (error) 1404 return (error); 1405 newvp = NFSTOV(np); 1406 /* 1407 * If n_localmodtime >= time before RPC, then 1408 * a file modification operation, such as 1409 * VOP_SETATTR() of size, has occurred while 1410 * the Lookup RPC and acquisition of the vnode 1411 * happened. As such, the attributes might 1412 * be stale, with possibly an incorrect size. 1413 */ 1414 NFSLOCKNODE(np); 1415 if (timespecisset(&np->n_localmodtime) && 1416 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1417 NFSCL_DEBUG(4, "nfs_lookup: rename localmod " 1418 "stale attributes\n"); 1419 attrflag = 0; 1420 } 1421 NFSUNLOCKNODE(np); 1422 if (attrflag) 1423 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1424 *vpp = newvp; 1425 return (0); 1426 } 1427 1428 if (flags & ISDOTDOT) { 1429 ltype = NFSVOPISLOCKED(dvp); 1430 error = vfs_busy(mp, MBF_NOWAIT); 1431 if (error != 0) { 1432 vfs_ref(mp); 1433 NFSVOPUNLOCK(dvp); 1434 error = vfs_busy(mp, 0); 1435 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1436 vfs_rel(mp); 1437 if (error == 0 && VN_IS_DOOMED(dvp)) { 1438 vfs_unbusy(mp); 1439 error = ENOENT; 1440 } 1441 if (error != 0) 1442 return (error); 1443 } 1444 NFSVOPUNLOCK(dvp); 1445 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1446 cnp->cn_lkflags); 1447 if (error == 0) 1448 newvp = NFSTOV(np); 1449 vfs_unbusy(mp); 1450 if (newvp != dvp) 1451 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1452 if (VN_IS_DOOMED(dvp)) { 1453 if (error == 0) { 1454 if (newvp == dvp) 1455 vrele(newvp); 1456 else 1457 vput(newvp); 1458 } 1459 error = ENOENT; 1460 } 1461 if (error != 0) 1462 return (error); 1463 if (attrflag) 1464 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1465 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1466 free(nfhp, M_NFSFH); 1467 VREF(dvp); 1468 newvp = dvp; 1469 if (attrflag) 1470 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1471 } else { 1472 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1473 cnp->cn_lkflags); 1474 if (error) 1475 return (error); 1476 newvp = NFSTOV(np); 1477 /* 1478 * If n_localmodtime >= time before RPC, then 1479 * a file modification operation, such as 1480 * VOP_SETATTR() of size, has occurred while 1481 * the Lookup RPC and acquisition of the vnode 1482 * happened. As such, the attributes might 1483 * be stale, with possibly an incorrect size. 1484 */ 1485 NFSLOCKNODE(np); 1486 if (timespecisset(&np->n_localmodtime) && 1487 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1488 NFSCL_DEBUG(4, "nfs_lookup: localmod " 1489 "stale attributes\n"); 1490 attrflag = 0; 1491 } 1492 NFSUNLOCKNODE(np); 1493 if (attrflag) 1494 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1495 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1496 !(np->n_flag & NMODIFIED)) { 1497 /* 1498 * Flush the attribute cache when opening a 1499 * leaf node to ensure that fresh attributes 1500 * are fetched in nfs_open() since we did not 1501 * fetch attributes from the LOOKUP reply. 1502 */ 1503 NFSLOCKNODE(np); 1504 np->n_attrstamp = 0; 1505 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1506 NFSUNLOCKNODE(np); 1507 } 1508 } 1509 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp && 1510 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1511 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1512 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1513 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1514 *vpp = newvp; 1515 return (0); 1516 } 1517 1518 /* 1519 * nfs read call. 1520 * Just call ncl_bioread() to do the work. 1521 */ 1522 static int 1523 nfs_read(struct vop_read_args *ap) 1524 { 1525 struct vnode *vp = ap->a_vp; 1526 1527 switch (vp->v_type) { 1528 case VREG: 1529 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1530 case VDIR: 1531 return (EISDIR); 1532 default: 1533 return (EOPNOTSUPP); 1534 } 1535 } 1536 1537 /* 1538 * nfs readlink call 1539 */ 1540 static int 1541 nfs_readlink(struct vop_readlink_args *ap) 1542 { 1543 struct vnode *vp = ap->a_vp; 1544 1545 if (vp->v_type != VLNK) 1546 return (EINVAL); 1547 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1548 } 1549 1550 /* 1551 * Do a readlink rpc. 1552 * Called by ncl_doio() from below the buffer cache. 1553 */ 1554 int 1555 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1556 { 1557 int error, ret, attrflag; 1558 struct nfsvattr nfsva; 1559 1560 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1561 &attrflag); 1562 if (attrflag) { 1563 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1564 if (ret && !error) 1565 error = ret; 1566 } 1567 if (error && NFS_ISV4(vp)) 1568 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1569 return (error); 1570 } 1571 1572 /* 1573 * nfs read rpc call 1574 * Ditto above 1575 */ 1576 int 1577 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1578 { 1579 int error, ret, attrflag; 1580 struct nfsvattr nfsva; 1581 struct nfsmount *nmp; 1582 1583 nmp = VFSTONFS(vp->v_mount); 1584 error = EIO; 1585 attrflag = 0; 1586 if (NFSHASPNFS(nmp)) 1587 error = nfscl_doiods(vp, uiop, NULL, NULL, 1588 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1589 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1590 if (error != 0 && error != EFAULT) 1591 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1592 &attrflag); 1593 if (attrflag) { 1594 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1595 if (ret && !error) 1596 error = ret; 1597 } 1598 if (error && NFS_ISV4(vp)) 1599 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1600 return (error); 1601 } 1602 1603 /* 1604 * nfs write call 1605 */ 1606 int 1607 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1608 int *iomode, int *must_commit, int called_from_strategy, int ioflag) 1609 { 1610 struct nfsvattr nfsva; 1611 int error, attrflag, ret; 1612 struct nfsmount *nmp; 1613 1614 nmp = VFSTONFS(vp->v_mount); 1615 error = EIO; 1616 attrflag = 0; 1617 if (NFSHASPNFS(nmp)) 1618 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1619 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1620 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1621 if (error != 0 && error != EFAULT) 1622 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1623 uiop->uio_td, &nfsva, &attrflag, called_from_strategy, 1624 ioflag); 1625 if (attrflag) { 1626 if (VTONFS(vp)->n_flag & ND_NFSV4) 1627 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1); 1628 else 1629 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1630 if (ret && !error) 1631 error = ret; 1632 } 1633 if (DOINGASYNC(vp)) 1634 *iomode = NFSWRITE_FILESYNC; 1635 if (error && NFS_ISV4(vp)) 1636 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1637 return (error); 1638 } 1639 1640 /* 1641 * nfs mknod rpc 1642 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1643 * mode set to specify the file type and the size field for rdev. 1644 */ 1645 static int 1646 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1647 struct vattr *vap) 1648 { 1649 struct nfsvattr nfsva, dnfsva; 1650 struct vnode *newvp = NULL; 1651 struct nfsnode *np = NULL, *dnp; 1652 struct nfsfh *nfhp; 1653 struct vattr vattr; 1654 int error = 0, attrflag, dattrflag; 1655 u_int32_t rdev; 1656 1657 if (vap->va_type == VCHR || vap->va_type == VBLK) 1658 rdev = vap->va_rdev; 1659 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1660 rdev = 0xffffffff; 1661 else 1662 return (EOPNOTSUPP); 1663 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1664 return (error); 1665 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1666 rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva, 1667 &nfsva, &nfhp, &attrflag, &dattrflag); 1668 if (!error) { 1669 if (!nfhp) 1670 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1671 cnp->cn_namelen, cnp->cn_cred, curthread, 1672 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0); 1673 if (nfhp) 1674 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1675 curthread, &np, LK_EXCLUSIVE); 1676 } 1677 if (dattrflag) 1678 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1679 if (!error) { 1680 newvp = NFSTOV(np); 1681 if (attrflag != 0) { 1682 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1683 if (error != 0) 1684 vput(newvp); 1685 } 1686 } 1687 if (!error) { 1688 *vpp = newvp; 1689 } else if (NFS_ISV4(dvp)) { 1690 error = nfscl_maperr(curthread, error, vap->va_uid, 1691 vap->va_gid); 1692 } 1693 dnp = VTONFS(dvp); 1694 NFSLOCKNODE(dnp); 1695 dnp->n_flag |= NMODIFIED; 1696 if (!dattrflag) { 1697 dnp->n_attrstamp = 0; 1698 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1699 } 1700 NFSUNLOCKNODE(dnp); 1701 return (error); 1702 } 1703 1704 /* 1705 * nfs mknod vop 1706 * just call nfs_mknodrpc() to do the work. 1707 */ 1708 /* ARGSUSED */ 1709 static int 1710 nfs_mknod(struct vop_mknod_args *ap) 1711 { 1712 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1713 } 1714 1715 static struct mtx nfs_cverf_mtx; 1716 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1717 MTX_DEF); 1718 1719 static nfsquad_t 1720 nfs_get_cverf(void) 1721 { 1722 static nfsquad_t cverf; 1723 nfsquad_t ret; 1724 static int cverf_initialized = 0; 1725 1726 mtx_lock(&nfs_cverf_mtx); 1727 if (cverf_initialized == 0) { 1728 cverf.lval[0] = arc4random(); 1729 cverf.lval[1] = arc4random(); 1730 cverf_initialized = 1; 1731 } else 1732 cverf.qval++; 1733 ret = cverf; 1734 mtx_unlock(&nfs_cverf_mtx); 1735 1736 return (ret); 1737 } 1738 1739 /* 1740 * nfs file create call 1741 */ 1742 static int 1743 nfs_create(struct vop_create_args *ap) 1744 { 1745 struct vnode *dvp = ap->a_dvp; 1746 struct vattr *vap = ap->a_vap; 1747 struct componentname *cnp = ap->a_cnp; 1748 struct nfsnode *np = NULL, *dnp; 1749 struct vnode *newvp = NULL; 1750 struct nfsmount *nmp; 1751 struct nfsvattr dnfsva, nfsva; 1752 struct nfsfh *nfhp; 1753 nfsquad_t cverf; 1754 int error = 0, attrflag, dattrflag, fmode = 0; 1755 struct vattr vattr; 1756 1757 /* 1758 * Oops, not for me.. 1759 */ 1760 if (vap->va_type == VSOCK) 1761 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1762 1763 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1764 return (error); 1765 if (vap->va_vaflags & VA_EXCLUSIVE) 1766 fmode |= O_EXCL; 1767 dnp = VTONFS(dvp); 1768 nmp = VFSTONFS(dvp->v_mount); 1769 again: 1770 /* For NFSv4, wait until any remove is done. */ 1771 NFSLOCKNODE(dnp); 1772 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1773 dnp->n_flag |= NREMOVEWANT; 1774 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1775 } 1776 NFSUNLOCKNODE(dnp); 1777 1778 cverf = nfs_get_cverf(); 1779 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1780 vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva, 1781 &nfhp, &attrflag, &dattrflag); 1782 if (!error) { 1783 if (nfhp == NULL) 1784 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1785 cnp->cn_namelen, cnp->cn_cred, curthread, 1786 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0); 1787 if (nfhp != NULL) 1788 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1789 curthread, &np, LK_EXCLUSIVE); 1790 } 1791 if (dattrflag) 1792 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1793 if (!error) { 1794 newvp = NFSTOV(np); 1795 if (attrflag == 0) 1796 error = nfsrpc_getattr(newvp, cnp->cn_cred, curthread, 1797 &nfsva); 1798 if (error == 0) 1799 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1800 } 1801 if (error) { 1802 if (newvp != NULL) { 1803 vput(newvp); 1804 newvp = NULL; 1805 } 1806 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1807 error == NFSERR_NOTSUPP) { 1808 fmode &= ~O_EXCL; 1809 goto again; 1810 } 1811 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1812 if (nfscl_checksattr(vap, &nfsva)) { 1813 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1814 curthread, &nfsva, &attrflag); 1815 if (error && (vap->va_uid != (uid_t)VNOVAL || 1816 vap->va_gid != (gid_t)VNOVAL)) { 1817 /* try again without setting uid/gid */ 1818 vap->va_uid = (uid_t)VNOVAL; 1819 vap->va_gid = (uid_t)VNOVAL; 1820 error = nfsrpc_setattr(newvp, vap, NULL, 1821 cnp->cn_cred, curthread, &nfsva, &attrflag); 1822 } 1823 if (attrflag) 1824 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1825 0, 1); 1826 if (error != 0) 1827 vput(newvp); 1828 } 1829 } 1830 if (!error) { 1831 if ((cnp->cn_flags & MAKEENTRY) && attrflag) { 1832 if (dvp != newvp) 1833 cache_enter_time(dvp, newvp, cnp, 1834 &nfsva.na_ctime, NULL); 1835 else 1836 printf("nfs_create: bogus NFS server returned " 1837 "the directory as the new file object\n"); 1838 } 1839 *ap->a_vpp = newvp; 1840 } else if (NFS_ISV4(dvp)) { 1841 error = nfscl_maperr(curthread, error, vap->va_uid, 1842 vap->va_gid); 1843 } 1844 NFSLOCKNODE(dnp); 1845 dnp->n_flag |= NMODIFIED; 1846 if (!dattrflag) { 1847 dnp->n_attrstamp = 0; 1848 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1849 } 1850 NFSUNLOCKNODE(dnp); 1851 return (error); 1852 } 1853 1854 /* 1855 * nfs file remove call 1856 * To try and make nfs semantics closer to ufs semantics, a file that has 1857 * other processes using the vnode is renamed instead of removed and then 1858 * removed later on the last close. 1859 * - If v_usecount > 1 1860 * If a rename is not already in the works 1861 * call nfs_sillyrename() to set it up 1862 * else 1863 * do the remove rpc 1864 */ 1865 static int 1866 nfs_remove(struct vop_remove_args *ap) 1867 { 1868 struct vnode *vp = ap->a_vp; 1869 struct vnode *dvp = ap->a_dvp; 1870 struct componentname *cnp = ap->a_cnp; 1871 struct nfsnode *np = VTONFS(vp); 1872 int error = 0; 1873 struct vattr vattr; 1874 1875 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1876 if (vp->v_type == VDIR) 1877 error = EPERM; 1878 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1879 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1880 vattr.va_nlink > 1)) { 1881 /* 1882 * Purge the name cache so that the chance of a lookup for 1883 * the name succeeding while the remove is in progress is 1884 * minimized. Without node locking it can still happen, such 1885 * that an I/O op returns ESTALE, but since you get this if 1886 * another host removes the file.. 1887 */ 1888 cache_purge(vp); 1889 /* 1890 * throw away biocache buffers, mainly to avoid 1891 * unnecessary delayed writes later. 1892 */ 1893 error = ncl_vinvalbuf(vp, 0, curthread, 1); 1894 if (error != EINTR && error != EIO) 1895 /* Do the rpc */ 1896 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1897 cnp->cn_namelen, cnp->cn_cred, curthread); 1898 /* 1899 * Kludge City: If the first reply to the remove rpc is lost.. 1900 * the reply to the retransmitted request will be ENOENT 1901 * since the file was in fact removed 1902 * Therefore, we cheat and return success. 1903 */ 1904 if (error == ENOENT) 1905 error = 0; 1906 } else if (!np->n_sillyrename) 1907 error = nfs_sillyrename(dvp, vp, cnp); 1908 NFSLOCKNODE(np); 1909 np->n_attrstamp = 0; 1910 NFSUNLOCKNODE(np); 1911 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1912 return (error); 1913 } 1914 1915 /* 1916 * nfs file remove rpc called from nfs_inactive 1917 */ 1918 int 1919 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1920 { 1921 /* 1922 * Make sure that the directory vnode is still valid. 1923 * XXX we should lock sp->s_dvp here. 1924 */ 1925 if (sp->s_dvp->v_type == VBAD) 1926 return (0); 1927 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1928 sp->s_cred, NULL)); 1929 } 1930 1931 /* 1932 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1933 */ 1934 static int 1935 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1936 int namelen, struct ucred *cred, struct thread *td) 1937 { 1938 struct nfsvattr dnfsva; 1939 struct nfsnode *dnp = VTONFS(dvp); 1940 int error = 0, dattrflag; 1941 1942 NFSLOCKNODE(dnp); 1943 dnp->n_flag |= NREMOVEINPROG; 1944 NFSUNLOCKNODE(dnp); 1945 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1946 &dattrflag); 1947 NFSLOCKNODE(dnp); 1948 if ((dnp->n_flag & NREMOVEWANT)) { 1949 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1950 NFSUNLOCKNODE(dnp); 1951 wakeup((caddr_t)dnp); 1952 } else { 1953 dnp->n_flag &= ~NREMOVEINPROG; 1954 NFSUNLOCKNODE(dnp); 1955 } 1956 if (dattrflag) 1957 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1958 NFSLOCKNODE(dnp); 1959 dnp->n_flag |= NMODIFIED; 1960 if (!dattrflag) { 1961 dnp->n_attrstamp = 0; 1962 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1963 } 1964 NFSUNLOCKNODE(dnp); 1965 if (error && NFS_ISV4(dvp)) 1966 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1967 return (error); 1968 } 1969 1970 /* 1971 * nfs file rename call 1972 */ 1973 static int 1974 nfs_rename(struct vop_rename_args *ap) 1975 { 1976 struct vnode *fvp = ap->a_fvp; 1977 struct vnode *tvp = ap->a_tvp; 1978 struct vnode *fdvp = ap->a_fdvp; 1979 struct vnode *tdvp = ap->a_tdvp; 1980 struct componentname *tcnp = ap->a_tcnp; 1981 struct componentname *fcnp = ap->a_fcnp; 1982 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1983 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1984 struct nfsv4node *newv4 = NULL; 1985 int error; 1986 1987 /* Check for cross-device rename */ 1988 if ((fvp->v_mount != tdvp->v_mount) || 1989 (tvp && (fvp->v_mount != tvp->v_mount))) { 1990 error = EXDEV; 1991 goto out; 1992 } 1993 1994 if (fvp == tvp) { 1995 printf("nfs_rename: fvp == tvp (can't happen)\n"); 1996 error = 0; 1997 goto out; 1998 } 1999 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 2000 goto out; 2001 2002 /* 2003 * We have to flush B_DELWRI data prior to renaming 2004 * the file. If we don't, the delayed-write buffers 2005 * can be flushed out later after the file has gone stale 2006 * under NFSV3. NFSV2 does not have this problem because 2007 * ( as far as I can tell ) it flushes dirty buffers more 2008 * often. 2009 * 2010 * Skip the rename operation if the fsync fails, this can happen 2011 * due to the server's volume being full, when we pushed out data 2012 * that was written back to our cache earlier. Not checking for 2013 * this condition can result in potential (silent) data loss. 2014 */ 2015 error = VOP_FSYNC(fvp, MNT_WAIT, curthread); 2016 NFSVOPUNLOCK(fvp); 2017 if (!error && tvp) 2018 error = VOP_FSYNC(tvp, MNT_WAIT, curthread); 2019 if (error) 2020 goto out; 2021 2022 /* 2023 * If the tvp exists and is in use, sillyrename it before doing the 2024 * rename of the new file over it. 2025 * XXX Can't sillyrename a directory. 2026 */ 2027 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 2028 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 2029 vput(tvp); 2030 tvp = NULL; 2031 } 2032 2033 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 2034 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 2035 curthread); 2036 2037 if (error == 0 && NFS_ISV4(tdvp)) { 2038 /* 2039 * For NFSv4, check to see if it is the same name and 2040 * replace the name, if it is different. 2041 */ 2042 newv4 = malloc( 2043 sizeof (struct nfsv4node) + 2044 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 2045 M_NFSV4NODE, M_WAITOK); 2046 NFSLOCKNODE(tdnp); 2047 NFSLOCKNODE(fnp); 2048 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 2049 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 2050 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 2051 tcnp->cn_namelen) || 2052 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 2053 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2054 tdnp->n_fhp->nfh_len))) { 2055 free(fnp->n_v4, M_NFSV4NODE); 2056 fnp->n_v4 = newv4; 2057 newv4 = NULL; 2058 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 2059 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 2060 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2061 tdnp->n_fhp->nfh_len); 2062 NFSBCOPY(tcnp->cn_nameptr, 2063 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 2064 } 2065 NFSUNLOCKNODE(tdnp); 2066 NFSUNLOCKNODE(fnp); 2067 if (newv4 != NULL) 2068 free(newv4, M_NFSV4NODE); 2069 } 2070 2071 if (fvp->v_type == VDIR) { 2072 if (tvp != NULL && tvp->v_type == VDIR) 2073 cache_purge(tdvp); 2074 cache_purge(fdvp); 2075 } 2076 2077 out: 2078 if (tdvp == tvp) 2079 vrele(tdvp); 2080 else 2081 vput(tdvp); 2082 if (tvp) 2083 vput(tvp); 2084 vrele(fdvp); 2085 vrele(fvp); 2086 /* 2087 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2088 */ 2089 if (error == ENOENT) 2090 error = 0; 2091 return (error); 2092 } 2093 2094 /* 2095 * nfs file rename rpc called from nfs_remove() above 2096 */ 2097 static int 2098 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 2099 struct sillyrename *sp) 2100 { 2101 2102 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 2103 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 2104 curthread)); 2105 } 2106 2107 /* 2108 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2109 */ 2110 static int 2111 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 2112 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 2113 int tnamelen, struct ucred *cred, struct thread *td) 2114 { 2115 struct nfsvattr fnfsva, tnfsva; 2116 struct nfsnode *fdnp = VTONFS(fdvp); 2117 struct nfsnode *tdnp = VTONFS(tdvp); 2118 int error = 0, fattrflag, tattrflag; 2119 2120 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 2121 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 2122 &tattrflag); 2123 NFSLOCKNODE(fdnp); 2124 fdnp->n_flag |= NMODIFIED; 2125 if (fattrflag != 0) { 2126 NFSUNLOCKNODE(fdnp); 2127 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1); 2128 } else { 2129 fdnp->n_attrstamp = 0; 2130 NFSUNLOCKNODE(fdnp); 2131 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 2132 } 2133 NFSLOCKNODE(tdnp); 2134 tdnp->n_flag |= NMODIFIED; 2135 if (tattrflag != 0) { 2136 NFSUNLOCKNODE(tdnp); 2137 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1); 2138 } else { 2139 tdnp->n_attrstamp = 0; 2140 NFSUNLOCKNODE(tdnp); 2141 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2142 } 2143 if (error && NFS_ISV4(fdvp)) 2144 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2145 return (error); 2146 } 2147 2148 /* 2149 * nfs hard link create call 2150 */ 2151 static int 2152 nfs_link(struct vop_link_args *ap) 2153 { 2154 struct vnode *vp = ap->a_vp; 2155 struct vnode *tdvp = ap->a_tdvp; 2156 struct componentname *cnp = ap->a_cnp; 2157 struct nfsnode *np, *tdnp; 2158 struct nfsvattr nfsva, dnfsva; 2159 int error = 0, attrflag, dattrflag; 2160 2161 /* 2162 * Push all writes to the server, so that the attribute cache 2163 * doesn't get "out of sync" with the server. 2164 * XXX There should be a better way! 2165 */ 2166 VOP_FSYNC(vp, MNT_WAIT, curthread); 2167 2168 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2169 cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, &dattrflag); 2170 tdnp = VTONFS(tdvp); 2171 NFSLOCKNODE(tdnp); 2172 tdnp->n_flag |= NMODIFIED; 2173 if (dattrflag != 0) { 2174 NFSUNLOCKNODE(tdnp); 2175 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1); 2176 } else { 2177 tdnp->n_attrstamp = 0; 2178 NFSUNLOCKNODE(tdnp); 2179 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2180 } 2181 if (attrflag) 2182 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2183 else { 2184 np = VTONFS(vp); 2185 NFSLOCKNODE(np); 2186 np->n_attrstamp = 0; 2187 NFSUNLOCKNODE(np); 2188 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2189 } 2190 /* 2191 * If negative lookup caching is enabled, I might as well 2192 * add an entry for this node. Not necessary for correctness, 2193 * but if negative caching is enabled, then the system 2194 * must care about lookup caching hit rate, so... 2195 */ 2196 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2197 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2198 if (tdvp != vp) 2199 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2200 else 2201 printf("nfs_link: bogus NFS server returned " 2202 "the directory as the new link\n"); 2203 } 2204 if (error && NFS_ISV4(vp)) 2205 error = nfscl_maperr(curthread, error, (uid_t)0, 2206 (gid_t)0); 2207 return (error); 2208 } 2209 2210 /* 2211 * nfs symbolic link create call 2212 */ 2213 static int 2214 nfs_symlink(struct vop_symlink_args *ap) 2215 { 2216 struct vnode *dvp = ap->a_dvp; 2217 struct vattr *vap = ap->a_vap; 2218 struct componentname *cnp = ap->a_cnp; 2219 struct nfsvattr nfsva, dnfsva; 2220 struct nfsfh *nfhp; 2221 struct nfsnode *np = NULL, *dnp; 2222 struct vnode *newvp = NULL; 2223 int error = 0, attrflag, dattrflag, ret; 2224 2225 vap->va_type = VLNK; 2226 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2227 ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva, 2228 &nfsva, &nfhp, &attrflag, &dattrflag); 2229 if (nfhp) { 2230 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2231 &np, LK_EXCLUSIVE); 2232 if (!ret) 2233 newvp = NFSTOV(np); 2234 else if (!error) 2235 error = ret; 2236 } 2237 if (newvp != NULL) { 2238 if (attrflag) 2239 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2240 } else if (!error) { 2241 /* 2242 * If we do not have an error and we could not extract the 2243 * newvp from the response due to the request being NFSv2, we 2244 * have to do a lookup in order to obtain a newvp to return. 2245 */ 2246 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2247 cnp->cn_cred, curthread, &np); 2248 if (!error) 2249 newvp = NFSTOV(np); 2250 } 2251 if (error) { 2252 if (newvp) 2253 vput(newvp); 2254 if (NFS_ISV4(dvp)) 2255 error = nfscl_maperr(curthread, error, 2256 vap->va_uid, vap->va_gid); 2257 } else { 2258 *ap->a_vpp = newvp; 2259 } 2260 2261 dnp = VTONFS(dvp); 2262 NFSLOCKNODE(dnp); 2263 dnp->n_flag |= NMODIFIED; 2264 if (dattrflag != 0) { 2265 NFSUNLOCKNODE(dnp); 2266 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2267 } else { 2268 dnp->n_attrstamp = 0; 2269 NFSUNLOCKNODE(dnp); 2270 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2271 } 2272 /* 2273 * If negative lookup caching is enabled, I might as well 2274 * add an entry for this node. Not necessary for correctness, 2275 * but if negative caching is enabled, then the system 2276 * must care about lookup caching hit rate, so... 2277 */ 2278 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2279 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2280 if (dvp != newvp) 2281 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2282 NULL); 2283 else 2284 printf("nfs_symlink: bogus NFS server returned " 2285 "the directory as the new file object\n"); 2286 } 2287 return (error); 2288 } 2289 2290 /* 2291 * nfs make dir call 2292 */ 2293 static int 2294 nfs_mkdir(struct vop_mkdir_args *ap) 2295 { 2296 struct vnode *dvp = ap->a_dvp; 2297 struct vattr *vap = ap->a_vap; 2298 struct componentname *cnp = ap->a_cnp; 2299 struct nfsnode *np = NULL, *dnp; 2300 struct vnode *newvp = NULL; 2301 struct vattr vattr; 2302 struct nfsfh *nfhp; 2303 struct nfsvattr nfsva, dnfsva; 2304 int error = 0, attrflag, dattrflag, ret; 2305 2306 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2307 return (error); 2308 vap->va_type = VDIR; 2309 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2310 vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, 2311 &attrflag, &dattrflag); 2312 dnp = VTONFS(dvp); 2313 NFSLOCKNODE(dnp); 2314 dnp->n_flag |= NMODIFIED; 2315 if (dattrflag != 0) { 2316 NFSUNLOCKNODE(dnp); 2317 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2318 } else { 2319 dnp->n_attrstamp = 0; 2320 NFSUNLOCKNODE(dnp); 2321 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2322 } 2323 if (nfhp) { 2324 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2325 &np, LK_EXCLUSIVE); 2326 if (!ret) { 2327 newvp = NFSTOV(np); 2328 if (attrflag) 2329 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2330 0, 1); 2331 } else if (!error) 2332 error = ret; 2333 } 2334 if (!error && newvp == NULL) { 2335 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2336 cnp->cn_cred, curthread, &np); 2337 if (!error) { 2338 newvp = NFSTOV(np); 2339 if (newvp->v_type != VDIR) 2340 error = EEXIST; 2341 } 2342 } 2343 if (error) { 2344 if (newvp) 2345 vput(newvp); 2346 if (NFS_ISV4(dvp)) 2347 error = nfscl_maperr(curthread, error, 2348 vap->va_uid, vap->va_gid); 2349 } else { 2350 /* 2351 * If negative lookup caching is enabled, I might as well 2352 * add an entry for this node. Not necessary for correctness, 2353 * but if negative caching is enabled, then the system 2354 * must care about lookup caching hit rate, so... 2355 */ 2356 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2357 (cnp->cn_flags & MAKEENTRY) && 2358 attrflag != 0 && dattrflag != 0) { 2359 if (dvp != newvp) 2360 cache_enter_time(dvp, newvp, cnp, 2361 &nfsva.na_ctime, &dnfsva.na_ctime); 2362 else 2363 printf("nfs_mkdir: bogus NFS server returned " 2364 "the directory that the directory was " 2365 "created in as the new file object\n"); 2366 } 2367 *ap->a_vpp = newvp; 2368 } 2369 return (error); 2370 } 2371 2372 /* 2373 * nfs remove directory call 2374 */ 2375 static int 2376 nfs_rmdir(struct vop_rmdir_args *ap) 2377 { 2378 struct vnode *vp = ap->a_vp; 2379 struct vnode *dvp = ap->a_dvp; 2380 struct componentname *cnp = ap->a_cnp; 2381 struct nfsnode *dnp; 2382 struct nfsvattr dnfsva; 2383 int error, dattrflag; 2384 2385 if (dvp == vp) 2386 return (EINVAL); 2387 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2388 cnp->cn_cred, curthread, &dnfsva, &dattrflag); 2389 dnp = VTONFS(dvp); 2390 NFSLOCKNODE(dnp); 2391 dnp->n_flag |= NMODIFIED; 2392 if (dattrflag != 0) { 2393 NFSUNLOCKNODE(dnp); 2394 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2395 } else { 2396 dnp->n_attrstamp = 0; 2397 NFSUNLOCKNODE(dnp); 2398 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2399 } 2400 2401 cache_purge(dvp); 2402 cache_purge(vp); 2403 if (error && NFS_ISV4(dvp)) 2404 error = nfscl_maperr(curthread, error, (uid_t)0, 2405 (gid_t)0); 2406 /* 2407 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2408 */ 2409 if (error == ENOENT) 2410 error = 0; 2411 return (error); 2412 } 2413 2414 /* 2415 * nfs readdir call 2416 */ 2417 static int 2418 nfs_readdir(struct vop_readdir_args *ap) 2419 { 2420 struct vnode *vp = ap->a_vp; 2421 struct nfsnode *np = VTONFS(vp); 2422 struct uio *uio = ap->a_uio; 2423 ssize_t tresid, left; 2424 int error = 0; 2425 struct vattr vattr; 2426 2427 if (ap->a_eofflag != NULL) 2428 *ap->a_eofflag = 0; 2429 if (vp->v_type != VDIR) 2430 return(EPERM); 2431 2432 /* 2433 * First, check for hit on the EOF offset cache 2434 */ 2435 NFSLOCKNODE(np); 2436 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2437 (np->n_flag & NMODIFIED) == 0) { 2438 NFSUNLOCKNODE(np); 2439 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2440 NFSLOCKNODE(np); 2441 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2442 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2443 NFSUNLOCKNODE(np); 2444 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2445 if (ap->a_eofflag != NULL) 2446 *ap->a_eofflag = 1; 2447 return (0); 2448 } else 2449 NFSUNLOCKNODE(np); 2450 } 2451 } else 2452 NFSUNLOCKNODE(np); 2453 2454 /* 2455 * NFS always guarantees that directory entries don't straddle 2456 * DIRBLKSIZ boundaries. As such, we need to limit the size 2457 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2458 * directory entry. 2459 */ 2460 left = uio->uio_resid % DIRBLKSIZ; 2461 if (left == uio->uio_resid) 2462 return (EINVAL); 2463 uio->uio_resid -= left; 2464 2465 /* 2466 * Call ncl_bioread() to do the real work. 2467 */ 2468 tresid = uio->uio_resid; 2469 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2470 2471 if (!error && uio->uio_resid == tresid) { 2472 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2473 if (ap->a_eofflag != NULL) 2474 *ap->a_eofflag = 1; 2475 } 2476 2477 /* Add the partial DIRBLKSIZ (left) back in. */ 2478 uio->uio_resid += left; 2479 return (error); 2480 } 2481 2482 /* 2483 * Readdir rpc call. 2484 * Called from below the buffer cache by ncl_doio(). 2485 */ 2486 int 2487 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2488 struct thread *td) 2489 { 2490 struct nfsvattr nfsva; 2491 nfsuint64 *cookiep, cookie; 2492 struct nfsnode *dnp = VTONFS(vp); 2493 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2494 int error = 0, eof, attrflag; 2495 2496 KASSERT(uiop->uio_iovcnt == 1 && 2497 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2498 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2499 ("nfs readdirrpc bad uio")); 2500 2501 /* 2502 * If there is no cookie, assume directory was stale. 2503 */ 2504 ncl_dircookie_lock(dnp); 2505 NFSUNLOCKNODE(dnp); 2506 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2507 if (cookiep) { 2508 cookie = *cookiep; 2509 ncl_dircookie_unlock(dnp); 2510 } else { 2511 ncl_dircookie_unlock(dnp); 2512 return (NFSERR_BAD_COOKIE); 2513 } 2514 2515 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2516 (void)ncl_fsinfo(nmp, vp, cred, td); 2517 2518 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2519 &attrflag, &eof); 2520 if (attrflag) 2521 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2522 2523 if (!error) { 2524 /* 2525 * We are now either at the end of the directory or have filled 2526 * the block. 2527 */ 2528 if (eof) { 2529 NFSLOCKNODE(dnp); 2530 dnp->n_direofoffset = uiop->uio_offset; 2531 NFSUNLOCKNODE(dnp); 2532 } else { 2533 if (uiop->uio_resid > 0) 2534 printf("EEK! readdirrpc resid > 0\n"); 2535 ncl_dircookie_lock(dnp); 2536 NFSUNLOCKNODE(dnp); 2537 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2538 *cookiep = cookie; 2539 ncl_dircookie_unlock(dnp); 2540 } 2541 } else if (NFS_ISV4(vp)) { 2542 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2543 } 2544 return (error); 2545 } 2546 2547 /* 2548 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2549 */ 2550 int 2551 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2552 struct thread *td) 2553 { 2554 struct nfsvattr nfsva; 2555 nfsuint64 *cookiep, cookie; 2556 struct nfsnode *dnp = VTONFS(vp); 2557 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2558 int error = 0, attrflag, eof; 2559 2560 KASSERT(uiop->uio_iovcnt == 1 && 2561 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2562 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2563 ("nfs readdirplusrpc bad uio")); 2564 2565 /* 2566 * If there is no cookie, assume directory was stale. 2567 */ 2568 ncl_dircookie_lock(dnp); 2569 NFSUNLOCKNODE(dnp); 2570 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2571 if (cookiep) { 2572 cookie = *cookiep; 2573 ncl_dircookie_unlock(dnp); 2574 } else { 2575 ncl_dircookie_unlock(dnp); 2576 return (NFSERR_BAD_COOKIE); 2577 } 2578 2579 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2580 (void)ncl_fsinfo(nmp, vp, cred, td); 2581 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2582 &attrflag, &eof); 2583 if (attrflag) 2584 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2585 2586 if (!error) { 2587 /* 2588 * We are now either at end of the directory or have filled the 2589 * the block. 2590 */ 2591 if (eof) { 2592 NFSLOCKNODE(dnp); 2593 dnp->n_direofoffset = uiop->uio_offset; 2594 NFSUNLOCKNODE(dnp); 2595 } else { 2596 if (uiop->uio_resid > 0) 2597 printf("EEK! readdirplusrpc resid > 0\n"); 2598 ncl_dircookie_lock(dnp); 2599 NFSUNLOCKNODE(dnp); 2600 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2601 *cookiep = cookie; 2602 ncl_dircookie_unlock(dnp); 2603 } 2604 } else if (NFS_ISV4(vp)) { 2605 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2606 } 2607 return (error); 2608 } 2609 2610 /* 2611 * Silly rename. To make the NFS filesystem that is stateless look a little 2612 * more like the "ufs" a remove of an active vnode is translated to a rename 2613 * to a funny looking filename that is removed by nfs_inactive on the 2614 * nfsnode. There is the potential for another process on a different client 2615 * to create the same funny name between the nfs_lookitup() fails and the 2616 * nfs_rename() completes, but... 2617 */ 2618 static int 2619 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2620 { 2621 struct sillyrename *sp; 2622 struct nfsnode *np; 2623 int error; 2624 short pid; 2625 unsigned int lticks; 2626 2627 cache_purge(dvp); 2628 np = VTONFS(vp); 2629 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2630 sp = malloc(sizeof (struct sillyrename), 2631 M_NEWNFSREQ, M_WAITOK); 2632 sp->s_cred = crhold(cnp->cn_cred); 2633 sp->s_dvp = dvp; 2634 VREF(dvp); 2635 2636 /* 2637 * Fudge together a funny name. 2638 * Changing the format of the funny name to accommodate more 2639 * sillynames per directory. 2640 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2641 * CPU ticks since boot. 2642 */ 2643 pid = curthread->td_proc->p_pid; 2644 lticks = (unsigned int)ticks; 2645 for ( ; ; ) { 2646 sp->s_namlen = sprintf(sp->s_name, 2647 ".nfs.%08x.%04x4.4", lticks, 2648 pid); 2649 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2650 curthread, NULL)) 2651 break; 2652 lticks++; 2653 } 2654 error = nfs_renameit(dvp, vp, cnp, sp); 2655 if (error) 2656 goto bad; 2657 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2658 curthread, &np); 2659 np->n_sillyrename = sp; 2660 return (0); 2661 bad: 2662 vrele(sp->s_dvp); 2663 crfree(sp->s_cred); 2664 free(sp, M_NEWNFSREQ); 2665 return (error); 2666 } 2667 2668 /* 2669 * Look up a file name and optionally either update the file handle or 2670 * allocate an nfsnode, depending on the value of npp. 2671 * npp == NULL --> just do the lookup 2672 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2673 * handled too 2674 * *npp != NULL --> update the file handle in the vnode 2675 */ 2676 static int 2677 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2678 struct thread *td, struct nfsnode **npp) 2679 { 2680 struct vnode *newvp = NULL, *vp; 2681 struct nfsnode *np, *dnp = VTONFS(dvp); 2682 struct nfsfh *nfhp, *onfhp; 2683 struct nfsvattr nfsva, dnfsva; 2684 struct componentname cn; 2685 int error = 0, attrflag, dattrflag; 2686 u_int hash; 2687 struct timespec ts; 2688 2689 nanouptime(&ts); 2690 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2691 &nfhp, &attrflag, &dattrflag, 0); 2692 if (dattrflag) 2693 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2694 if (npp && !error) { 2695 if (*npp != NULL) { 2696 np = *npp; 2697 vp = NFSTOV(np); 2698 /* 2699 * For NFSv4, check to see if it is the same name and 2700 * replace the name, if it is different. 2701 */ 2702 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2703 (np->n_v4->n4_namelen != len || 2704 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2705 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2706 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2707 dnp->n_fhp->nfh_len))) { 2708 free(np->n_v4, M_NFSV4NODE); 2709 np->n_v4 = malloc( 2710 sizeof (struct nfsv4node) + 2711 dnp->n_fhp->nfh_len + len - 1, 2712 M_NFSV4NODE, M_WAITOK); 2713 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2714 np->n_v4->n4_namelen = len; 2715 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2716 dnp->n_fhp->nfh_len); 2717 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2718 } 2719 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2720 FNV1_32_INIT); 2721 onfhp = np->n_fhp; 2722 /* 2723 * Rehash node for new file handle. 2724 */ 2725 vfs_hash_rehash(vp, hash); 2726 np->n_fhp = nfhp; 2727 if (onfhp != NULL) 2728 free(onfhp, M_NFSFH); 2729 newvp = NFSTOV(np); 2730 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2731 free(nfhp, M_NFSFH); 2732 VREF(dvp); 2733 newvp = dvp; 2734 } else { 2735 cn.cn_nameptr = name; 2736 cn.cn_namelen = len; 2737 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2738 &np, LK_EXCLUSIVE); 2739 if (error) 2740 return (error); 2741 newvp = NFSTOV(np); 2742 /* 2743 * If n_localmodtime >= time before RPC, then 2744 * a file modification operation, such as 2745 * VOP_SETATTR() of size, has occurred while 2746 * the Lookup RPC and acquisition of the vnode 2747 * happened. As such, the attributes might 2748 * be stale, with possibly an incorrect size. 2749 */ 2750 NFSLOCKNODE(np); 2751 if (timespecisset(&np->n_localmodtime) && 2752 timespeccmp(&np->n_localmodtime, &ts, >=)) { 2753 NFSCL_DEBUG(4, "nfs_lookitup: localmod " 2754 "stale attributes\n"); 2755 attrflag = 0; 2756 } 2757 NFSUNLOCKNODE(np); 2758 } 2759 if (!attrflag && *npp == NULL) { 2760 if (newvp == dvp) 2761 vrele(newvp); 2762 else 2763 vput(newvp); 2764 return (ENOENT); 2765 } 2766 if (attrflag) 2767 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2768 } 2769 if (npp && *npp == NULL) { 2770 if (error) { 2771 if (newvp) { 2772 if (newvp == dvp) 2773 vrele(newvp); 2774 else 2775 vput(newvp); 2776 } 2777 } else 2778 *npp = np; 2779 } 2780 if (error && NFS_ISV4(dvp)) 2781 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2782 return (error); 2783 } 2784 2785 /* 2786 * Nfs Version 3 and 4 commit rpc 2787 */ 2788 int 2789 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2790 struct thread *td) 2791 { 2792 struct nfsvattr nfsva; 2793 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2794 struct nfsnode *np; 2795 struct uio uio; 2796 int error, attrflag; 2797 2798 np = VTONFS(vp); 2799 error = EIO; 2800 attrflag = 0; 2801 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 2802 uio.uio_offset = offset; 2803 uio.uio_resid = cnt; 2804 error = nfscl_doiods(vp, &uio, NULL, NULL, 2805 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 2806 if (error != 0) { 2807 NFSLOCKNODE(np); 2808 np->n_flag &= ~NDSCOMMIT; 2809 NFSUNLOCKNODE(np); 2810 } 2811 } 2812 if (error != 0) { 2813 mtx_lock(&nmp->nm_mtx); 2814 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2815 mtx_unlock(&nmp->nm_mtx); 2816 return (0); 2817 } 2818 mtx_unlock(&nmp->nm_mtx); 2819 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2820 &attrflag); 2821 } 2822 if (attrflag != 0) 2823 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2824 if (error != 0 && NFS_ISV4(vp)) 2825 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2826 return (error); 2827 } 2828 2829 /* 2830 * Strategy routine. 2831 * For async requests when nfsiod(s) are running, queue the request by 2832 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2833 * request. 2834 */ 2835 static int 2836 nfs_strategy(struct vop_strategy_args *ap) 2837 { 2838 struct buf *bp; 2839 struct vnode *vp; 2840 struct ucred *cr; 2841 2842 bp = ap->a_bp; 2843 vp = ap->a_vp; 2844 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 2845 KASSERT(!(bp->b_flags & B_DONE), 2846 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2847 2848 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 2849 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 2850 DEV_BSIZE); 2851 if (bp->b_iocmd == BIO_READ) 2852 cr = bp->b_rcred; 2853 else 2854 cr = bp->b_wcred; 2855 2856 /* 2857 * If the op is asynchronous and an i/o daemon is waiting 2858 * queue the request, wake it up and wait for completion 2859 * otherwise just do it ourselves. 2860 */ 2861 if ((bp->b_flags & B_ASYNC) == 0 || 2862 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 2863 (void) ncl_doio(vp, bp, cr, curthread, 1); 2864 return (0); 2865 } 2866 2867 /* 2868 * fsync vnode op. Just call ncl_flush() with commit == 1. 2869 */ 2870 /* ARGSUSED */ 2871 static int 2872 nfs_fsync(struct vop_fsync_args *ap) 2873 { 2874 2875 if (ap->a_vp->v_type != VREG) { 2876 /* 2877 * For NFS, metadata is changed synchronously on the server, 2878 * so there is nothing to flush. Also, ncl_flush() clears 2879 * the NMODIFIED flag and that shouldn't be done here for 2880 * directories. 2881 */ 2882 return (0); 2883 } 2884 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 2885 } 2886 2887 /* 2888 * Flush all the blocks associated with a vnode. 2889 * Walk through the buffer pool and push any dirty pages 2890 * associated with the vnode. 2891 * If the called_from_renewthread argument is TRUE, it has been called 2892 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2893 * waiting for a buffer write to complete. 2894 */ 2895 int 2896 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 2897 int commit, int called_from_renewthread) 2898 { 2899 struct nfsnode *np = VTONFS(vp); 2900 struct buf *bp; 2901 int i; 2902 struct buf *nbp; 2903 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2904 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2905 int passone = 1, trycnt = 0; 2906 u_quad_t off, endoff, toff; 2907 struct ucred* wcred = NULL; 2908 struct buf **bvec = NULL; 2909 struct bufobj *bo; 2910 #ifndef NFS_COMMITBVECSIZ 2911 #define NFS_COMMITBVECSIZ 20 2912 #endif 2913 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2914 u_int bvecsize = 0, bveccount; 2915 struct timespec ts; 2916 2917 if (called_from_renewthread != 0) 2918 slptimeo = hz; 2919 if (nmp->nm_flag & NFSMNT_INT) 2920 slpflag = PCATCH; 2921 if (!commit) 2922 passone = 0; 2923 bo = &vp->v_bufobj; 2924 /* 2925 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2926 * server, but has not been committed to stable storage on the server 2927 * yet. On the first pass, the byte range is worked out and the commit 2928 * rpc is done. On the second pass, ncl_writebp() is called to do the 2929 * job. 2930 */ 2931 again: 2932 off = (u_quad_t)-1; 2933 endoff = 0; 2934 bvecpos = 0; 2935 if (NFS_ISV34(vp) && commit) { 2936 if (bvec != NULL && bvec != bvec_on_stack) 2937 free(bvec, M_TEMP); 2938 /* 2939 * Count up how many buffers waiting for a commit. 2940 */ 2941 bveccount = 0; 2942 BO_LOCK(bo); 2943 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2944 if (!BUF_ISLOCKED(bp) && 2945 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2946 == (B_DELWRI | B_NEEDCOMMIT)) 2947 bveccount++; 2948 } 2949 /* 2950 * Allocate space to remember the list of bufs to commit. It is 2951 * important to use M_NOWAIT here to avoid a race with nfs_write. 2952 * If we can't get memory (for whatever reason), we will end up 2953 * committing the buffers one-by-one in the loop below. 2954 */ 2955 if (bveccount > NFS_COMMITBVECSIZ) { 2956 /* 2957 * Release the vnode interlock to avoid a lock 2958 * order reversal. 2959 */ 2960 BO_UNLOCK(bo); 2961 bvec = (struct buf **) 2962 malloc(bveccount * sizeof(struct buf *), 2963 M_TEMP, M_NOWAIT); 2964 BO_LOCK(bo); 2965 if (bvec == NULL) { 2966 bvec = bvec_on_stack; 2967 bvecsize = NFS_COMMITBVECSIZ; 2968 } else 2969 bvecsize = bveccount; 2970 } else { 2971 bvec = bvec_on_stack; 2972 bvecsize = NFS_COMMITBVECSIZ; 2973 } 2974 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2975 if (bvecpos >= bvecsize) 2976 break; 2977 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2978 nbp = TAILQ_NEXT(bp, b_bobufs); 2979 continue; 2980 } 2981 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2982 (B_DELWRI | B_NEEDCOMMIT)) { 2983 BUF_UNLOCK(bp); 2984 nbp = TAILQ_NEXT(bp, b_bobufs); 2985 continue; 2986 } 2987 BO_UNLOCK(bo); 2988 bremfree(bp); 2989 /* 2990 * Work out if all buffers are using the same cred 2991 * so we can deal with them all with one commit. 2992 * 2993 * NOTE: we are not clearing B_DONE here, so we have 2994 * to do it later on in this routine if we intend to 2995 * initiate I/O on the bp. 2996 * 2997 * Note: to avoid loopback deadlocks, we do not 2998 * assign b_runningbufspace. 2999 */ 3000 if (wcred == NULL) 3001 wcred = bp->b_wcred; 3002 else if (wcred != bp->b_wcred) 3003 wcred = NOCRED; 3004 vfs_busy_pages(bp, 0); 3005 3006 BO_LOCK(bo); 3007 /* 3008 * bp is protected by being locked, but nbp is not 3009 * and vfs_busy_pages() may sleep. We have to 3010 * recalculate nbp. 3011 */ 3012 nbp = TAILQ_NEXT(bp, b_bobufs); 3013 3014 /* 3015 * A list of these buffers is kept so that the 3016 * second loop knows which buffers have actually 3017 * been committed. This is necessary, since there 3018 * may be a race between the commit rpc and new 3019 * uncommitted writes on the file. 3020 */ 3021 bvec[bvecpos++] = bp; 3022 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3023 bp->b_dirtyoff; 3024 if (toff < off) 3025 off = toff; 3026 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3027 if (toff > endoff) 3028 endoff = toff; 3029 } 3030 BO_UNLOCK(bo); 3031 } 3032 if (bvecpos > 0) { 3033 /* 3034 * Commit data on the server, as required. 3035 * If all bufs are using the same wcred, then use that with 3036 * one call for all of them, otherwise commit each one 3037 * separately. 3038 */ 3039 if (wcred != NOCRED) 3040 retv = ncl_commit(vp, off, (int)(endoff - off), 3041 wcred, td); 3042 else { 3043 retv = 0; 3044 for (i = 0; i < bvecpos; i++) { 3045 off_t off, size; 3046 bp = bvec[i]; 3047 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3048 bp->b_dirtyoff; 3049 size = (u_quad_t)(bp->b_dirtyend 3050 - bp->b_dirtyoff); 3051 retv = ncl_commit(vp, off, (int)size, 3052 bp->b_wcred, td); 3053 if (retv) break; 3054 } 3055 } 3056 3057 if (retv == NFSERR_STALEWRITEVERF) 3058 ncl_clearcommit(vp->v_mount); 3059 3060 /* 3061 * Now, either mark the blocks I/O done or mark the 3062 * blocks dirty, depending on whether the commit 3063 * succeeded. 3064 */ 3065 for (i = 0; i < bvecpos; i++) { 3066 bp = bvec[i]; 3067 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3068 if (!NFSCL_FORCEDISM(vp->v_mount) && retv) { 3069 /* 3070 * Error, leave B_DELWRI intact 3071 */ 3072 vfs_unbusy_pages(bp); 3073 brelse(bp); 3074 } else { 3075 /* 3076 * Success, remove B_DELWRI ( bundirty() ). 3077 * 3078 * b_dirtyoff/b_dirtyend seem to be NFS 3079 * specific. We should probably move that 3080 * into bundirty(). XXX 3081 */ 3082 bufobj_wref(bo); 3083 bp->b_flags |= B_ASYNC; 3084 bundirty(bp); 3085 bp->b_flags &= ~B_DONE; 3086 bp->b_ioflags &= ~BIO_ERROR; 3087 bp->b_dirtyoff = bp->b_dirtyend = 0; 3088 bufdone(bp); 3089 } 3090 } 3091 } 3092 3093 /* 3094 * Start/do any write(s) that are required. 3095 */ 3096 loop: 3097 BO_LOCK(bo); 3098 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3099 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3100 if (waitfor != MNT_WAIT || passone) 3101 continue; 3102 3103 error = BUF_TIMELOCK(bp, 3104 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 3105 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 3106 if (error == 0) { 3107 BUF_UNLOCK(bp); 3108 goto loop; 3109 } 3110 if (error == ENOLCK) { 3111 error = 0; 3112 goto loop; 3113 } 3114 if (called_from_renewthread != 0) { 3115 /* 3116 * Return EIO so the flush will be retried 3117 * later. 3118 */ 3119 error = EIO; 3120 goto done; 3121 } 3122 if (newnfs_sigintr(nmp, td)) { 3123 error = EINTR; 3124 goto done; 3125 } 3126 if (slpflag == PCATCH) { 3127 slpflag = 0; 3128 slptimeo = 2 * hz; 3129 } 3130 goto loop; 3131 } 3132 if ((bp->b_flags & B_DELWRI) == 0) 3133 panic("nfs_fsync: not dirty"); 3134 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3135 BUF_UNLOCK(bp); 3136 continue; 3137 } 3138 BO_UNLOCK(bo); 3139 bremfree(bp); 3140 bp->b_flags |= B_ASYNC; 3141 bwrite(bp); 3142 if (newnfs_sigintr(nmp, td)) { 3143 error = EINTR; 3144 goto done; 3145 } 3146 goto loop; 3147 } 3148 if (passone) { 3149 passone = 0; 3150 BO_UNLOCK(bo); 3151 goto again; 3152 } 3153 if (waitfor == MNT_WAIT) { 3154 while (bo->bo_numoutput) { 3155 error = bufobj_wwait(bo, slpflag, slptimeo); 3156 if (error) { 3157 BO_UNLOCK(bo); 3158 if (called_from_renewthread != 0) { 3159 /* 3160 * Return EIO so that the flush will be 3161 * retried later. 3162 */ 3163 error = EIO; 3164 goto done; 3165 } 3166 error = newnfs_sigintr(nmp, td); 3167 if (error) 3168 goto done; 3169 if (slpflag == PCATCH) { 3170 slpflag = 0; 3171 slptimeo = 2 * hz; 3172 } 3173 BO_LOCK(bo); 3174 } 3175 } 3176 if (bo->bo_dirty.bv_cnt != 0 && commit) { 3177 BO_UNLOCK(bo); 3178 goto loop; 3179 } 3180 /* 3181 * Wait for all the async IO requests to drain 3182 */ 3183 BO_UNLOCK(bo); 3184 NFSLOCKNODE(np); 3185 while (np->n_directio_asyncwr > 0) { 3186 np->n_flag |= NFSYNCWAIT; 3187 error = newnfs_msleep(td, &np->n_directio_asyncwr, 3188 &np->n_mtx, slpflag | (PRIBIO + 1), 3189 "nfsfsync", 0); 3190 if (error) { 3191 if (newnfs_sigintr(nmp, td)) { 3192 NFSUNLOCKNODE(np); 3193 error = EINTR; 3194 goto done; 3195 } 3196 } 3197 } 3198 NFSUNLOCKNODE(np); 3199 } else 3200 BO_UNLOCK(bo); 3201 if (NFSHASPNFS(nmp)) { 3202 nfscl_layoutcommit(vp, td); 3203 /* 3204 * Invalidate the attribute cache, since writes to a DS 3205 * won't update the size attribute. 3206 */ 3207 NFSLOCKNODE(np); 3208 np->n_attrstamp = 0; 3209 } else 3210 NFSLOCKNODE(np); 3211 if (np->n_flag & NWRITEERR) { 3212 error = np->n_error; 3213 np->n_flag &= ~NWRITEERR; 3214 } 3215 if (commit && bo->bo_dirty.bv_cnt == 0 && 3216 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 3217 np->n_flag &= ~NMODIFIED; 3218 NFSUNLOCKNODE(np); 3219 done: 3220 if (bvec != NULL && bvec != bvec_on_stack) 3221 free(bvec, M_TEMP); 3222 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 3223 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 3224 np->n_directio_asyncwr != 0)) { 3225 if (trycnt++ < 5) { 3226 /* try, try again... */ 3227 passone = 1; 3228 wcred = NULL; 3229 bvec = NULL; 3230 bvecsize = 0; 3231 goto again; 3232 } 3233 vn_printf(vp, "ncl_flush failed"); 3234 error = called_from_renewthread != 0 ? EIO : EBUSY; 3235 } 3236 if (error == 0) { 3237 nanouptime(&ts); 3238 NFSLOCKNODE(np); 3239 np->n_localmodtime = ts; 3240 NFSUNLOCKNODE(np); 3241 } 3242 return (error); 3243 } 3244 3245 /* 3246 * NFS advisory byte-level locks. 3247 */ 3248 static int 3249 nfs_advlock(struct vop_advlock_args *ap) 3250 { 3251 struct vnode *vp = ap->a_vp; 3252 struct ucred *cred; 3253 struct nfsnode *np = VTONFS(ap->a_vp); 3254 struct proc *p = (struct proc *)ap->a_id; 3255 struct thread *td = curthread; /* XXX */ 3256 struct vattr va; 3257 int ret, error; 3258 u_quad_t size; 3259 struct nfsmount *nmp; 3260 3261 error = NFSVOPLOCK(vp, LK_SHARED); 3262 if (error != 0) 3263 return (EBADF); 3264 nmp = VFSTONFS(vp->v_mount); 3265 if (!NFS_ISV4(vp) || (nmp->nm_flag & NFSMNT_NOLOCKD) != 0) { 3266 if ((nmp->nm_flag & NFSMNT_NOLOCKD) != 0) { 3267 size = np->n_size; 3268 NFSVOPUNLOCK(vp); 3269 error = lf_advlock(ap, &(vp->v_lockf), size); 3270 } else { 3271 if (nfs_advlock_p != NULL) 3272 error = nfs_advlock_p(ap); 3273 else { 3274 NFSVOPUNLOCK(vp); 3275 error = ENOLCK; 3276 } 3277 } 3278 if (error == 0 && ap->a_op == F_SETLK) { 3279 error = NFSVOPLOCK(vp, LK_SHARED); 3280 if (error == 0) { 3281 /* Mark that a file lock has been acquired. */ 3282 NFSLOCKNODE(np); 3283 np->n_flag |= NHASBEENLOCKED; 3284 NFSUNLOCKNODE(np); 3285 NFSVOPUNLOCK(vp); 3286 } 3287 } 3288 return (error); 3289 } else if ((ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3290 if (vp->v_type != VREG) { 3291 error = EINVAL; 3292 goto out; 3293 } 3294 if ((ap->a_flags & F_POSIX) != 0) 3295 cred = p->p_ucred; 3296 else 3297 cred = td->td_ucred; 3298 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 3299 if (VN_IS_DOOMED(vp)) { 3300 error = EBADF; 3301 goto out; 3302 } 3303 3304 /* 3305 * If this is unlocking a write locked region, flush and 3306 * commit them before unlocking. This is required by 3307 * RFC3530 Sec. 9.3.2. 3308 */ 3309 if (ap->a_op == F_UNLCK && 3310 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3311 ap->a_flags)) 3312 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3313 3314 /* 3315 * Mark NFS node as might have acquired a lock. 3316 * This is separate from NHASBEENLOCKED, because it must 3317 * be done before the nfsrpc_advlock() call, which might 3318 * add a nfscllock structure to the client state. 3319 * It is used to check for the case where a nfscllock 3320 * state structure cannot exist for the file. 3321 * Only done for "oneopenown" NFSv4.1/4.2 mounts. 3322 */ 3323 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) { 3324 NFSLOCKNODE(np); 3325 np->n_flag |= NMIGHTBELOCKED; 3326 NFSUNLOCKNODE(np); 3327 } 3328 3329 /* 3330 * Loop around doing the lock op, while a blocking lock 3331 * must wait for the lock op to succeed. 3332 */ 3333 do { 3334 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3335 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3336 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3337 ap->a_op == F_SETLK) { 3338 NFSVOPUNLOCK(vp); 3339 error = nfs_catnap(PZERO | PCATCH, ret, 3340 "ncladvl"); 3341 if (error) 3342 return (EINTR); 3343 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3344 if (VN_IS_DOOMED(vp)) { 3345 error = EBADF; 3346 goto out; 3347 } 3348 } 3349 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3350 ap->a_op == F_SETLK); 3351 if (ret == NFSERR_DENIED) { 3352 error = EAGAIN; 3353 goto out; 3354 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3355 error = ret; 3356 goto out; 3357 } else if (ret != 0) { 3358 error = EACCES; 3359 goto out; 3360 } 3361 3362 /* 3363 * Now, if we just got a lock, invalidate data in the buffer 3364 * cache, as required, so that the coherency conforms with 3365 * RFC3530 Sec. 9.3.2. 3366 */ 3367 if (ap->a_op == F_SETLK) { 3368 if ((np->n_flag & NMODIFIED) == 0) { 3369 np->n_attrstamp = 0; 3370 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3371 ret = VOP_GETATTR(vp, &va, cred); 3372 } 3373 if ((np->n_flag & NMODIFIED) || ret || 3374 np->n_change != va.va_filerev) { 3375 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3376 np->n_attrstamp = 0; 3377 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3378 ret = VOP_GETATTR(vp, &va, cred); 3379 if (!ret) { 3380 np->n_mtime = va.va_mtime; 3381 np->n_change = va.va_filerev; 3382 } 3383 } 3384 /* Mark that a file lock has been acquired. */ 3385 NFSLOCKNODE(np); 3386 np->n_flag |= NHASBEENLOCKED; 3387 NFSUNLOCKNODE(np); 3388 } 3389 } else 3390 error = EOPNOTSUPP; 3391 out: 3392 NFSVOPUNLOCK(vp); 3393 return (error); 3394 } 3395 3396 /* 3397 * NFS advisory byte-level locks. 3398 */ 3399 static int 3400 nfs_advlockasync(struct vop_advlockasync_args *ap) 3401 { 3402 struct vnode *vp = ap->a_vp; 3403 u_quad_t size; 3404 int error; 3405 3406 error = NFSVOPLOCK(vp, LK_SHARED); 3407 if (error) 3408 return (error); 3409 if (NFS_ISV4(vp)) { 3410 NFSVOPUNLOCK(vp); 3411 return (EOPNOTSUPP); 3412 } 3413 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3414 size = VTONFS(vp)->n_size; 3415 NFSVOPUNLOCK(vp); 3416 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3417 } else { 3418 NFSVOPUNLOCK(vp); 3419 error = EOPNOTSUPP; 3420 } 3421 return (error); 3422 } 3423 3424 /* 3425 * Print out the contents of an nfsnode. 3426 */ 3427 static int 3428 nfs_print(struct vop_print_args *ap) 3429 { 3430 struct vnode *vp = ap->a_vp; 3431 struct nfsnode *np = VTONFS(vp); 3432 3433 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3434 (uintmax_t)np->n_vattr.na_fsid); 3435 if (vp->v_type == VFIFO) 3436 fifo_printinfo(vp); 3437 printf("\n"); 3438 return (0); 3439 } 3440 3441 /* 3442 * This is the "real" nfs::bwrite(struct buf*). 3443 * We set B_CACHE if this is a VMIO buffer. 3444 */ 3445 int 3446 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3447 { 3448 int oldflags, rtval; 3449 3450 if (bp->b_flags & B_INVAL) { 3451 brelse(bp); 3452 return (0); 3453 } 3454 3455 oldflags = bp->b_flags; 3456 bp->b_flags |= B_CACHE; 3457 3458 /* 3459 * Undirty the bp. We will redirty it later if the I/O fails. 3460 */ 3461 bundirty(bp); 3462 bp->b_flags &= ~B_DONE; 3463 bp->b_ioflags &= ~BIO_ERROR; 3464 bp->b_iocmd = BIO_WRITE; 3465 3466 bufobj_wref(bp->b_bufobj); 3467 curthread->td_ru.ru_oublock++; 3468 3469 /* 3470 * Note: to avoid loopback deadlocks, we do not 3471 * assign b_runningbufspace. 3472 */ 3473 vfs_busy_pages(bp, 1); 3474 3475 BUF_KERNPROC(bp); 3476 bp->b_iooffset = dbtob(bp->b_blkno); 3477 bstrategy(bp); 3478 3479 if ((oldflags & B_ASYNC) != 0) 3480 return (0); 3481 3482 rtval = bufwait(bp); 3483 if (oldflags & B_DELWRI) 3484 reassignbuf(bp); 3485 brelse(bp); 3486 return (rtval); 3487 } 3488 3489 /* 3490 * nfs special file access vnode op. 3491 * Essentially just get vattr and then imitate iaccess() since the device is 3492 * local to the client. 3493 */ 3494 static int 3495 nfsspec_access(struct vop_access_args *ap) 3496 { 3497 struct vattr *vap; 3498 struct ucred *cred = ap->a_cred; 3499 struct vnode *vp = ap->a_vp; 3500 accmode_t accmode = ap->a_accmode; 3501 struct vattr vattr; 3502 int error; 3503 3504 /* 3505 * Disallow write attempts on filesystems mounted read-only; 3506 * unless the file is a socket, fifo, or a block or character 3507 * device resident on the filesystem. 3508 */ 3509 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3510 switch (vp->v_type) { 3511 case VREG: 3512 case VDIR: 3513 case VLNK: 3514 return (EROFS); 3515 default: 3516 break; 3517 } 3518 } 3519 vap = &vattr; 3520 error = VOP_GETATTR(vp, vap, cred); 3521 if (error) 3522 goto out; 3523 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3524 accmode, cred); 3525 out: 3526 return error; 3527 } 3528 3529 /* 3530 * Read wrapper for fifos. 3531 */ 3532 static int 3533 nfsfifo_read(struct vop_read_args *ap) 3534 { 3535 struct nfsnode *np = VTONFS(ap->a_vp); 3536 int error; 3537 3538 /* 3539 * Set access flag. 3540 */ 3541 NFSLOCKNODE(np); 3542 np->n_flag |= NACC; 3543 vfs_timestamp(&np->n_atim); 3544 NFSUNLOCKNODE(np); 3545 error = fifo_specops.vop_read(ap); 3546 return error; 3547 } 3548 3549 /* 3550 * Write wrapper for fifos. 3551 */ 3552 static int 3553 nfsfifo_write(struct vop_write_args *ap) 3554 { 3555 struct nfsnode *np = VTONFS(ap->a_vp); 3556 3557 /* 3558 * Set update flag. 3559 */ 3560 NFSLOCKNODE(np); 3561 np->n_flag |= NUPD; 3562 vfs_timestamp(&np->n_mtim); 3563 NFSUNLOCKNODE(np); 3564 return(fifo_specops.vop_write(ap)); 3565 } 3566 3567 /* 3568 * Close wrapper for fifos. 3569 * 3570 * Update the times on the nfsnode then do fifo close. 3571 */ 3572 static int 3573 nfsfifo_close(struct vop_close_args *ap) 3574 { 3575 struct vnode *vp = ap->a_vp; 3576 struct nfsnode *np = VTONFS(vp); 3577 struct vattr vattr; 3578 struct timespec ts; 3579 3580 NFSLOCKNODE(np); 3581 if (np->n_flag & (NACC | NUPD)) { 3582 vfs_timestamp(&ts); 3583 if (np->n_flag & NACC) 3584 np->n_atim = ts; 3585 if (np->n_flag & NUPD) 3586 np->n_mtim = ts; 3587 np->n_flag |= NCHG; 3588 if (vrefcnt(vp) == 1 && 3589 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3590 VATTR_NULL(&vattr); 3591 if (np->n_flag & NACC) 3592 vattr.va_atime = np->n_atim; 3593 if (np->n_flag & NUPD) 3594 vattr.va_mtime = np->n_mtim; 3595 NFSUNLOCKNODE(np); 3596 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3597 goto out; 3598 } 3599 } 3600 NFSUNLOCKNODE(np); 3601 out: 3602 return (fifo_specops.vop_close(ap)); 3603 } 3604 3605 /* 3606 * Just call ncl_writebp() with the force argument set to 1. 3607 * 3608 * NOTE: B_DONE may or may not be set in a_bp on call. 3609 */ 3610 static int 3611 nfs_bwrite(struct buf *bp) 3612 { 3613 3614 return (ncl_writebp(bp, 1, curthread)); 3615 } 3616 3617 struct buf_ops buf_ops_newnfs = { 3618 .bop_name = "buf_ops_nfs", 3619 .bop_write = nfs_bwrite, 3620 .bop_strategy = bufstrategy, 3621 .bop_sync = bufsync, 3622 .bop_bdflush = bufbdflush, 3623 }; 3624 3625 static int 3626 nfs_getacl(struct vop_getacl_args *ap) 3627 { 3628 int error; 3629 3630 if (ap->a_type != ACL_TYPE_NFS4) 3631 return (EOPNOTSUPP); 3632 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp); 3633 if (error > NFSERR_STALE) { 3634 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3635 error = EPERM; 3636 } 3637 return (error); 3638 } 3639 3640 static int 3641 nfs_setacl(struct vop_setacl_args *ap) 3642 { 3643 int error; 3644 3645 if (ap->a_type != ACL_TYPE_NFS4) 3646 return (EOPNOTSUPP); 3647 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp); 3648 if (error > NFSERR_STALE) { 3649 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3650 error = EPERM; 3651 } 3652 return (error); 3653 } 3654 3655 /* 3656 * VOP_ADVISE for NFS. 3657 * Just return 0 for any errors, since it is just a hint. 3658 */ 3659 static int 3660 nfs_advise(struct vop_advise_args *ap) 3661 { 3662 struct thread *td = curthread; 3663 struct nfsmount *nmp; 3664 uint64_t len; 3665 int error; 3666 3667 /* 3668 * First do vop_stdadvise() to handle the buffer cache. 3669 */ 3670 error = vop_stdadvise(ap); 3671 if (error != 0) 3672 return (error); 3673 if (ap->a_start < 0 || ap->a_end < 0) 3674 return (0); 3675 if (ap->a_end == OFF_MAX) 3676 len = 0; 3677 else if (ap->a_end < ap->a_start) 3678 return (0); 3679 else 3680 len = ap->a_end - ap->a_start + 1; 3681 nmp = VFSTONFS(ap->a_vp->v_mount); 3682 mtx_lock(&nmp->nm_mtx); 3683 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3684 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) == 3685 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) { 3686 mtx_unlock(&nmp->nm_mtx); 3687 return (0); 3688 } 3689 mtx_unlock(&nmp->nm_mtx); 3690 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice, 3691 td->td_ucred, td); 3692 if (error == NFSERR_NOTSUPP) { 3693 mtx_lock(&nmp->nm_mtx); 3694 nmp->nm_privflag |= NFSMNTP_NOADVISE; 3695 mtx_unlock(&nmp->nm_mtx); 3696 } 3697 return (0); 3698 } 3699 3700 /* 3701 * nfs allocate call 3702 */ 3703 static int 3704 nfs_allocate(struct vop_allocate_args *ap) 3705 { 3706 struct vnode *vp = ap->a_vp; 3707 struct thread *td = curthread; 3708 struct nfsvattr nfsva; 3709 struct nfsmount *nmp; 3710 struct nfsnode *np; 3711 off_t alen; 3712 int attrflag, error, ret; 3713 struct timespec ts; 3714 struct uio io; 3715 3716 attrflag = 0; 3717 nmp = VFSTONFS(vp->v_mount); 3718 np = VTONFS(vp); 3719 mtx_lock(&nmp->nm_mtx); 3720 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3721 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) { 3722 mtx_unlock(&nmp->nm_mtx); 3723 alen = *ap->a_len; 3724 if ((uint64_t)alen > nfs_maxalloclen) 3725 alen = nfs_maxalloclen; 3726 3727 /* Check the file size limit. */ 3728 io.uio_offset = *ap->a_offset; 3729 io.uio_resid = alen; 3730 error = vn_rlimit_fsize(vp, &io, td); 3731 3732 /* 3733 * Flush first to ensure that the allocate adds to the 3734 * file's allocation on the server. 3735 */ 3736 if (error == 0) 3737 error = ncl_flush(vp, MNT_WAIT, td, 1, 0); 3738 if (error == 0) 3739 error = nfsrpc_allocate(vp, *ap->a_offset, alen, 3740 &nfsva, &attrflag, ap->a_cred, td); 3741 if (error == 0) { 3742 *ap->a_offset += alen; 3743 *ap->a_len -= alen; 3744 nanouptime(&ts); 3745 NFSLOCKNODE(np); 3746 np->n_localmodtime = ts; 3747 NFSUNLOCKNODE(np); 3748 } else if (error == NFSERR_NOTSUPP) { 3749 mtx_lock(&nmp->nm_mtx); 3750 nmp->nm_privflag |= NFSMNTP_NOALLOCATE; 3751 mtx_unlock(&nmp->nm_mtx); 3752 error = EINVAL; 3753 } 3754 } else { 3755 mtx_unlock(&nmp->nm_mtx); 3756 error = EINVAL; 3757 } 3758 if (attrflag != 0) { 3759 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3760 if (error == 0 && ret != 0) 3761 error = ret; 3762 } 3763 if (error != 0) 3764 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3765 return (error); 3766 } 3767 3768 /* 3769 * nfs deallocate call 3770 */ 3771 static int 3772 nfs_deallocate(struct vop_deallocate_args *ap) 3773 { 3774 struct vnode *vp = ap->a_vp; 3775 struct thread *td = curthread; 3776 struct nfsvattr nfsva; 3777 struct nfsmount *nmp; 3778 struct nfsnode *np; 3779 off_t tlen, mlen; 3780 int attrflag, error, ret; 3781 bool clipped; 3782 struct timespec ts; 3783 3784 error = 0; 3785 attrflag = 0; 3786 nmp = VFSTONFS(vp->v_mount); 3787 np = VTONFS(vp); 3788 mtx_lock(&nmp->nm_mtx); 3789 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3790 (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) { 3791 mtx_unlock(&nmp->nm_mtx); 3792 tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len); 3793 NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n", 3794 (intmax_t)*ap->a_offset, (intmax_t)tlen, 3795 (uintmax_t)nmp->nm_maxfilesize); 3796 if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) { 3797 /* Avoid EFBIG error return from the NFSv4.2 server. */ 3798 *ap->a_len = 0; 3799 return (0); 3800 } 3801 clipped = false; 3802 if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize) 3803 tlen = nmp->nm_maxfilesize - *ap->a_offset; 3804 if ((uint64_t)*ap->a_offset < np->n_size) { 3805 /* Limit the len to nfs_maxalloclen before EOF. */ 3806 mlen = omin((off_t)np->n_size - *ap->a_offset, tlen); 3807 if ((uint64_t)mlen > nfs_maxalloclen) { 3808 NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n"); 3809 tlen = nfs_maxalloclen; 3810 clipped = true; 3811 } 3812 } 3813 if (error == 0) 3814 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 3815 if (error == 0) { 3816 vnode_pager_purge_range(vp, *ap->a_offset, 3817 *ap->a_offset + tlen); 3818 error = nfsrpc_deallocate(vp, *ap->a_offset, tlen, 3819 &nfsva, &attrflag, ap->a_cred, td); 3820 NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error); 3821 } 3822 if (error == 0) { 3823 NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n", 3824 attrflag, (uintmax_t)nfsva.na_size); 3825 nanouptime(&ts); 3826 NFSLOCKNODE(np); 3827 np->n_localmodtime = ts; 3828 NFSUNLOCKNODE(np); 3829 if (attrflag != 0) { 3830 if ((uint64_t)*ap->a_offset < nfsva.na_size) 3831 *ap->a_offset += omin((off_t) 3832 nfsva.na_size - *ap->a_offset, 3833 tlen); 3834 } 3835 if (clipped && tlen < *ap->a_len) 3836 *ap->a_len -= tlen; 3837 else 3838 *ap->a_len = 0; 3839 } else if (error == NFSERR_NOTSUPP) { 3840 mtx_lock(&nmp->nm_mtx); 3841 nmp->nm_privflag |= NFSMNTP_NODEALLOCATE; 3842 mtx_unlock(&nmp->nm_mtx); 3843 } 3844 } else { 3845 mtx_unlock(&nmp->nm_mtx); 3846 error = EIO; 3847 } 3848 /* 3849 * If the NFS server cannot perform the Deallocate operation, just call 3850 * vop_stddeallocate() to perform it. 3851 */ 3852 if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) { 3853 error = vop_stddeallocate(ap); 3854 NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error); 3855 } 3856 if (attrflag != 0) { 3857 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3858 if (error == 0 && ret != 0) 3859 error = ret; 3860 } 3861 if (error != 0) 3862 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3863 return (error); 3864 } 3865 3866 /* 3867 * nfs copy_file_range call 3868 */ 3869 static int 3870 nfs_copy_file_range(struct vop_copy_file_range_args *ap) 3871 { 3872 struct vnode *invp = ap->a_invp; 3873 struct vnode *outvp = ap->a_outvp; 3874 struct mount *mp; 3875 struct nfsvattr innfsva, outnfsva; 3876 struct vattr va, *vap; 3877 struct uio io; 3878 struct nfsmount *nmp; 3879 size_t len, len2; 3880 ssize_t r; 3881 int error, inattrflag, outattrflag, ret, ret2; 3882 off_t inoff, outoff; 3883 bool consecutive, must_commit, tryoutcred; 3884 3885 /* 3886 * NFSv4.2 Copy is not permitted for infile == outfile. 3887 * TODO: copy_file_range() between multiple NFS mountpoints 3888 */ 3889 if (invp == outvp || invp->v_mount != outvp->v_mount) { 3890 generic_copy: 3891 return (ENOSYS); 3892 } 3893 3894 /* Lock both vnodes, avoiding risk of deadlock. */ 3895 do { 3896 mp = NULL; 3897 error = vn_start_write(outvp, &mp, V_WAIT); 3898 if (error == 0) { 3899 error = vn_lock(outvp, LK_EXCLUSIVE); 3900 if (error == 0) { 3901 error = vn_lock(invp, LK_SHARED | LK_NOWAIT); 3902 if (error == 0) 3903 break; 3904 VOP_UNLOCK(outvp); 3905 if (mp != NULL) 3906 vn_finished_write(mp); 3907 mp = NULL; 3908 error = vn_lock(invp, LK_SHARED); 3909 if (error == 0) 3910 VOP_UNLOCK(invp); 3911 } 3912 } 3913 if (mp != NULL) 3914 vn_finished_write(mp); 3915 } while (error == 0); 3916 if (error != 0) 3917 return (error); 3918 3919 /* 3920 * More reasons to avoid nfs copy: not NFSv4.2, or explicitly 3921 * disabled. 3922 */ 3923 nmp = VFSTONFS(invp->v_mount); 3924 mtx_lock(&nmp->nm_mtx); 3925 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3926 (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0) { 3927 mtx_unlock(&nmp->nm_mtx); 3928 VOP_UNLOCK(invp); 3929 VOP_UNLOCK(outvp); 3930 if (mp != NULL) 3931 vn_finished_write(mp); 3932 goto generic_copy; 3933 } 3934 mtx_unlock(&nmp->nm_mtx); 3935 3936 /* 3937 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer? 3938 */ 3939 io.uio_offset = *ap->a_outoffp; 3940 io.uio_resid = *ap->a_lenp; 3941 error = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd); 3942 *ap->a_lenp = io.uio_resid; 3943 /* 3944 * No need to call vn_rlimit_fsizex_res before return, since the uio is 3945 * local. 3946 */ 3947 3948 /* 3949 * Flush the input file so that the data is up to date before 3950 * the copy. Flush writes for the output file so that they 3951 * do not overwrite the data copied to the output file by the Copy. 3952 * Set the commit argument for both flushes so that the data is on 3953 * stable storage before the Copy RPC. This is done in case the 3954 * server reboots during the Copy and needs to be redone. 3955 */ 3956 if (error == 0) 3957 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0); 3958 if (error == 0) 3959 error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0); 3960 3961 /* Do the actual NFSv4.2 RPC. */ 3962 ret = ret2 = 0; 3963 len = *ap->a_lenp; 3964 mtx_lock(&nmp->nm_mtx); 3965 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0) 3966 consecutive = true; 3967 else 3968 consecutive = false; 3969 mtx_unlock(&nmp->nm_mtx); 3970 inoff = *ap->a_inoffp; 3971 outoff = *ap->a_outoffp; 3972 tryoutcred = true; 3973 must_commit = false; 3974 if (error == 0) { 3975 vap = &VTONFS(invp)->n_vattr.na_vattr; 3976 error = VOP_GETATTR(invp, vap, ap->a_incred); 3977 if (error == 0) { 3978 /* 3979 * Clip "len" at va_size so that RFC compliant servers 3980 * will not reply NFSERR_INVAL. 3981 * Setting "len == 0" for the RPC would be preferred, 3982 * but some Linux servers do not support that. 3983 * If the len is being set to 0, do a Setattr RPC to 3984 * set the server's atime. This behaviour was the 3985 * preferred one for the FreeBSD "collective". 3986 */ 3987 if (inoff >= vap->va_size) { 3988 *ap->a_lenp = len = 0; 3989 VATTR_NULL(&va); 3990 va.va_atime.tv_sec = va.va_atime.tv_nsec = 0; 3991 va.va_vaflags = VA_UTIMES_NULL; 3992 inattrflag = 0; 3993 error = nfsrpc_setattr(invp, &va, NULL, 3994 ap->a_incred, curthread, &innfsva, 3995 &inattrflag); 3996 if (inattrflag != 0) 3997 ret = nfscl_loadattrcache(&invp, 3998 &innfsva, NULL, 0, 1); 3999 if (error == 0 && ret != 0) 4000 error = ret; 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 /* Do the actual NFSv4.2 RPC. */ 4119 switch (ap->a_command) { 4120 case FIOSEEKDATA: 4121 content = NFSV4CONTENT_DATA; 4122 break; 4123 case FIOSEEKHOLE: 4124 content = NFSV4CONTENT_HOLE; 4125 break; 4126 default: 4127 return (ENOTTY); 4128 } 4129 4130 error = vn_lock(vp, LK_SHARED); 4131 if (error != 0) 4132 return (EBADF); 4133 4134 if (vp->v_type != VREG) { 4135 VOP_UNLOCK(vp); 4136 return (ENOTTY); 4137 } 4138 nmp = VFSTONFS(vp->v_mount); 4139 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) { 4140 VOP_UNLOCK(vp); 4141 error = vop_stdioctl(ap); 4142 return (error); 4143 } 4144 4145 attrflag = 0; 4146 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size) 4147 error = ENXIO; 4148 else { 4149 /* 4150 * Flush all writes, so that the server is up to date. 4151 * Although a Commit is not required, the commit argument 4152 * is set so that, for a pNFS File/Flexible File Layout 4153 * server, the LayoutCommit will be done to ensure the file 4154 * size is up to date on the Metadata Server. 4155 */ 4156 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0); 4157 if (error == 0) 4158 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof, 4159 content, ap->a_cred, &nfsva, &attrflag); 4160 /* If at eof for FIOSEEKDATA, return ENXIO. */ 4161 if (eof && error == 0 && content == NFSV4CONTENT_DATA) 4162 error = ENXIO; 4163 } 4164 if (attrflag != 0) { 4165 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4166 if (error == 0 && ret != 0) 4167 error = ret; 4168 } 4169 NFSVOPUNLOCK(vp); 4170 4171 if (error != 0) 4172 error = ENXIO; 4173 return (error); 4174 } 4175 4176 /* 4177 * nfs getextattr call 4178 */ 4179 static int 4180 nfs_getextattr(struct vop_getextattr_args *ap) 4181 { 4182 struct vnode *vp = ap->a_vp; 4183 struct nfsmount *nmp; 4184 struct ucred *cred; 4185 struct thread *td = ap->a_td; 4186 struct nfsvattr nfsva; 4187 ssize_t len; 4188 int attrflag, error, ret; 4189 4190 nmp = VFSTONFS(vp->v_mount); 4191 mtx_lock(&nmp->nm_mtx); 4192 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4193 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4194 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4195 mtx_unlock(&nmp->nm_mtx); 4196 return (EOPNOTSUPP); 4197 } 4198 mtx_unlock(&nmp->nm_mtx); 4199 4200 cred = ap->a_cred; 4201 if (cred == NULL) 4202 cred = td->td_ucred; 4203 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4204 attrflag = 0; 4205 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva, 4206 &attrflag, cred, td); 4207 if (attrflag != 0) { 4208 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4209 if (error == 0 && ret != 0) 4210 error = ret; 4211 } 4212 if (error == 0 && ap->a_size != NULL) 4213 *ap->a_size = len; 4214 4215 switch (error) { 4216 case NFSERR_NOTSUPP: 4217 case NFSERR_OPILLEGAL: 4218 mtx_lock(&nmp->nm_mtx); 4219 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4220 mtx_unlock(&nmp->nm_mtx); 4221 error = EOPNOTSUPP; 4222 break; 4223 case NFSERR_NOXATTR: 4224 case NFSERR_XATTR2BIG: 4225 error = ENOATTR; 4226 break; 4227 default: 4228 error = nfscl_maperr(td, error, 0, 0); 4229 break; 4230 } 4231 return (error); 4232 } 4233 4234 /* 4235 * nfs setextattr call 4236 */ 4237 static int 4238 nfs_setextattr(struct vop_setextattr_args *ap) 4239 { 4240 struct vnode *vp = ap->a_vp; 4241 struct nfsmount *nmp; 4242 struct ucred *cred; 4243 struct thread *td = ap->a_td; 4244 struct nfsvattr nfsva; 4245 int attrflag, error, ret; 4246 4247 nmp = VFSTONFS(vp->v_mount); 4248 mtx_lock(&nmp->nm_mtx); 4249 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4250 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4251 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4252 mtx_unlock(&nmp->nm_mtx); 4253 return (EOPNOTSUPP); 4254 } 4255 mtx_unlock(&nmp->nm_mtx); 4256 4257 if (ap->a_uio->uio_resid < 0) 4258 return (EINVAL); 4259 cred = ap->a_cred; 4260 if (cred == NULL) 4261 cred = td->td_ucred; 4262 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4263 attrflag = 0; 4264 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva, 4265 &attrflag, cred, td); 4266 if (attrflag != 0) { 4267 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4268 if (error == 0 && ret != 0) 4269 error = ret; 4270 } 4271 4272 switch (error) { 4273 case NFSERR_NOTSUPP: 4274 case NFSERR_OPILLEGAL: 4275 mtx_lock(&nmp->nm_mtx); 4276 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4277 mtx_unlock(&nmp->nm_mtx); 4278 error = EOPNOTSUPP; 4279 break; 4280 case NFSERR_NOXATTR: 4281 case NFSERR_XATTR2BIG: 4282 error = ENOATTR; 4283 break; 4284 default: 4285 error = nfscl_maperr(td, error, 0, 0); 4286 break; 4287 } 4288 return (error); 4289 } 4290 4291 /* 4292 * nfs listextattr call 4293 */ 4294 static int 4295 nfs_listextattr(struct vop_listextattr_args *ap) 4296 { 4297 struct vnode *vp = ap->a_vp; 4298 struct nfsmount *nmp; 4299 struct ucred *cred; 4300 struct thread *td = ap->a_td; 4301 struct nfsvattr nfsva; 4302 size_t len, len2; 4303 uint64_t cookie; 4304 int attrflag, error, ret; 4305 bool eof; 4306 4307 nmp = VFSTONFS(vp->v_mount); 4308 mtx_lock(&nmp->nm_mtx); 4309 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4310 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4311 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4312 mtx_unlock(&nmp->nm_mtx); 4313 return (EOPNOTSUPP); 4314 } 4315 mtx_unlock(&nmp->nm_mtx); 4316 4317 cred = ap->a_cred; 4318 if (cred == NULL) 4319 cred = td->td_ucred; 4320 4321 /* Loop around doing List Extended Attribute RPCs. */ 4322 eof = false; 4323 cookie = 0; 4324 len2 = 0; 4325 error = 0; 4326 while (!eof && error == 0) { 4327 len = nmp->nm_rsize; 4328 attrflag = 0; 4329 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof, 4330 &nfsva, &attrflag, cred, td); 4331 if (attrflag != 0) { 4332 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4333 if (error == 0 && ret != 0) 4334 error = ret; 4335 } 4336 if (error == 0) { 4337 len2 += len; 4338 if (len2 > SSIZE_MAX) 4339 error = ENOATTR; 4340 } 4341 } 4342 if (error == 0 && ap->a_size != NULL) 4343 *ap->a_size = len2; 4344 4345 switch (error) { 4346 case NFSERR_NOTSUPP: 4347 case NFSERR_OPILLEGAL: 4348 mtx_lock(&nmp->nm_mtx); 4349 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4350 mtx_unlock(&nmp->nm_mtx); 4351 error = EOPNOTSUPP; 4352 break; 4353 case NFSERR_NOXATTR: 4354 case NFSERR_XATTR2BIG: 4355 error = ENOATTR; 4356 break; 4357 default: 4358 error = nfscl_maperr(td, error, 0, 0); 4359 break; 4360 } 4361 return (error); 4362 } 4363 4364 /* 4365 * nfs setextattr call 4366 */ 4367 static int 4368 nfs_deleteextattr(struct vop_deleteextattr_args *ap) 4369 { 4370 struct vnode *vp = ap->a_vp; 4371 struct nfsmount *nmp; 4372 struct nfsvattr nfsva; 4373 int attrflag, error, ret; 4374 4375 nmp = VFSTONFS(vp->v_mount); 4376 mtx_lock(&nmp->nm_mtx); 4377 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4378 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4379 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4380 mtx_unlock(&nmp->nm_mtx); 4381 return (EOPNOTSUPP); 4382 } 4383 mtx_unlock(&nmp->nm_mtx); 4384 4385 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4386 attrflag = 0; 4387 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred, 4388 ap->a_td); 4389 if (attrflag != 0) { 4390 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4391 if (error == 0 && ret != 0) 4392 error = ret; 4393 } 4394 4395 switch (error) { 4396 case NFSERR_NOTSUPP: 4397 case NFSERR_OPILLEGAL: 4398 mtx_lock(&nmp->nm_mtx); 4399 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4400 mtx_unlock(&nmp->nm_mtx); 4401 error = EOPNOTSUPP; 4402 break; 4403 case NFSERR_NOXATTR: 4404 case NFSERR_XATTR2BIG: 4405 error = ENOATTR; 4406 break; 4407 default: 4408 error = nfscl_maperr(ap->a_td, error, 0, 0); 4409 break; 4410 } 4411 return (error); 4412 } 4413 4414 /* 4415 * Return POSIX pathconf information applicable to nfs filesystems. 4416 */ 4417 static int 4418 nfs_pathconf(struct vop_pathconf_args *ap) 4419 { 4420 struct nfsv3_pathconf pc; 4421 struct nfsvattr nfsva; 4422 struct vnode *vp = ap->a_vp; 4423 struct nfsmount *nmp; 4424 struct thread *td = curthread; 4425 off_t off; 4426 bool eof; 4427 int attrflag, error; 4428 4429 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 4430 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 4431 ap->a_name == _PC_NO_TRUNC)) || 4432 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 4433 /* 4434 * Since only the above 4 a_names are returned by the NFSv3 4435 * Pathconf RPC, there is no point in doing it for others. 4436 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 4437 * be used for _PC_NFS4_ACL as well. 4438 */ 4439 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 4440 &attrflag); 4441 if (attrflag != 0) 4442 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4443 if (error != 0) 4444 return (error); 4445 } else { 4446 /* 4447 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 4448 * just fake them. 4449 */ 4450 pc.pc_linkmax = NFS_LINK_MAX; 4451 pc.pc_namemax = NFS_MAXNAMLEN; 4452 pc.pc_notrunc = 1; 4453 pc.pc_chownrestricted = 1; 4454 pc.pc_caseinsensitive = 0; 4455 pc.pc_casepreserving = 1; 4456 error = 0; 4457 } 4458 switch (ap->a_name) { 4459 case _PC_LINK_MAX: 4460 #ifdef _LP64 4461 *ap->a_retval = pc.pc_linkmax; 4462 #else 4463 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 4464 #endif 4465 break; 4466 case _PC_NAME_MAX: 4467 *ap->a_retval = pc.pc_namemax; 4468 break; 4469 case _PC_PIPE_BUF: 4470 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 4471 *ap->a_retval = PIPE_BUF; 4472 else 4473 error = EINVAL; 4474 break; 4475 case _PC_CHOWN_RESTRICTED: 4476 *ap->a_retval = pc.pc_chownrestricted; 4477 break; 4478 case _PC_NO_TRUNC: 4479 *ap->a_retval = pc.pc_notrunc; 4480 break; 4481 case _PC_ACL_NFS4: 4482 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 4483 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 4484 *ap->a_retval = 1; 4485 else 4486 *ap->a_retval = 0; 4487 break; 4488 case _PC_ACL_PATH_MAX: 4489 if (NFS_ISV4(vp)) 4490 *ap->a_retval = ACL_MAX_ENTRIES; 4491 else 4492 *ap->a_retval = 3; 4493 break; 4494 case _PC_PRIO_IO: 4495 *ap->a_retval = 0; 4496 break; 4497 case _PC_SYNC_IO: 4498 *ap->a_retval = 0; 4499 break; 4500 case _PC_ALLOC_SIZE_MIN: 4501 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 4502 break; 4503 case _PC_FILESIZEBITS: 4504 if (NFS_ISV34(vp)) 4505 *ap->a_retval = 64; 4506 else 4507 *ap->a_retval = 32; 4508 break; 4509 case _PC_REC_INCR_XFER_SIZE: 4510 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4511 break; 4512 case _PC_REC_MAX_XFER_SIZE: 4513 *ap->a_retval = -1; /* means ``unlimited'' */ 4514 break; 4515 case _PC_REC_MIN_XFER_SIZE: 4516 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4517 break; 4518 case _PC_REC_XFER_ALIGN: 4519 *ap->a_retval = PAGE_SIZE; 4520 break; 4521 case _PC_SYMLINK_MAX: 4522 *ap->a_retval = NFS_MAXPATHLEN; 4523 break; 4524 case _PC_MIN_HOLE_SIZE: 4525 /* Only some NFSv4.2 servers support Seek for Holes. */ 4526 *ap->a_retval = 0; 4527 nmp = VFSTONFS(vp->v_mount); 4528 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) { 4529 /* 4530 * NFSv4.2 doesn't have an attribute for hole size, 4531 * so all we can do is see if the Seek operation is 4532 * supported and then use f_iosize as a "best guess". 4533 */ 4534 mtx_lock(&nmp->nm_mtx); 4535 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) { 4536 mtx_unlock(&nmp->nm_mtx); 4537 off = 0; 4538 attrflag = 0; 4539 error = nfsrpc_seek(vp, &off, &eof, 4540 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva, 4541 &attrflag); 4542 if (attrflag != 0) 4543 (void) nfscl_loadattrcache(&vp, &nfsva, 4544 NULL, 0, 1); 4545 mtx_lock(&nmp->nm_mtx); 4546 if (error == NFSERR_NOTSUPP) 4547 nmp->nm_privflag |= NFSMNTP_SEEKTESTED; 4548 else 4549 nmp->nm_privflag |= NFSMNTP_SEEKTESTED | 4550 NFSMNTP_SEEK; 4551 error = 0; 4552 } 4553 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0) 4554 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4555 mtx_unlock(&nmp->nm_mtx); 4556 } 4557 break; 4558 4559 default: 4560 error = vop_stdpathconf(ap); 4561 break; 4562 } 4563 return (error); 4564 } 4565